F.

GANTMACHER

Lectures
in Analytical
Mechanics

TRANSLATED FROM THE RUSSIAN

BY GEORGE YANKOVSKY

MIR PUBLISHERS MOSCOW

ki3nATEJIbCTBO (tHAYKA*
MOCKBA
 Contents
! Preface . ... . . . ....
\chapter 1 . THE' D I F F E R E ~ T I A LE Q U ~ ~ O N
O SF ~ . ~ O T IOOF ~AN
ARBITRARY SYSTEM O F PARTICLES . . .. . . . .
1. Free and Constrained Systems. Constraints and Their Classi-
fication . . . . . . . . . . . . . . . . . . . . . .
2. Possible and Virtual Displacements. Ideal Constraints . . .
3. The General Equation of Dynamics. Lagrange's Equations
F i r s t published 1970 of the First Kind . . . . . . . . . . . . . . . . . .
Second printing 1975 4. The Principle of Virtual ~ i s ~ j a c e m e n tD'Alernbertls
s. Princi-
ple . . . . . . . . . . . . . . . . . . . . . . . . .
5. Holonomic Svstems. Independent ~ o o r a i n a t e s . Generalized
Forces . . . . . .
. . . . . . . . . . . . . . . . .
6. Lagrange's Equations of the Second Kind in Independent
Coordinates .. . . . . . . . . . . . . . . . . . . . .
T h i s book h a s been recommended for 7. Investigating Lagrange's Equations . . . . . . . . . . .
publication by Indian s p e c i a l i s t s under the 8. Theorem on Variation of Total Energy. Potential, Gyroscopic
and Dissipative 1'orces
' . . . . .. . . . . . . . . . .
programme of the J o i n t Indo-Soviet Board 9. Electromechanica! Analogies . . . . . . . . . . . . .
to make the b e s t Soviet textbooks 10. Appell's Equations for ~ o n h o l o n o m i cSystems. ~seudocoordi-
available for Indian s t u d e n t s nates ... . . .
. . . .l . . . . . . . . . . . . . . .
'\Chapter 2. T H E EQUATIONS O F MOTION IN A POTENTIAL FIELD
11. Lagrange's Equations for Potential Forces. The Generalized
Potential. Nonnatural Systems . . . . . . . . . . . . .
12. Canonical Equations of Hamilton . . . . . . . . . . . .
13. Routh's Equations . . . . . . . . . . . . . . . . . . .
14. Cyclic Coordinates . . . . . . . . . . . . . . . . . . .
15. The Poisson Bracket . . . . . . . . . . . . . . . . . .
Chapter 3. VARIATIONAL PRINCIPLES AND INTEGRAL INVARI-
ANTS . . . . . . . . . . . . . . . . . . . . . . . .
16. Hamilton's Principle . . . . . . . . . . . . . . . . .
17. Second Forni of Hamilton's Principle . . . . . . . . . .
18. The Basic Integral Invariant of Mechanics (PoincarB-Cartan
Integral Invariant) . . . . . . . . . . . . . . . . .
19. A Hydrodynamical Interpretation o£ the Basic Integral Inva-
riant. The Theorems of Thomson and Helmholtz on Circula-
tion and Vortices . . . . . . . . . . . . . . . . . .
20. Generalized Conservative Systems. whittaker's Equations.
Jacobi's Equations. The Maupertuis-Lagrange Principle of
Least Action . . . . . . . . . . . . . . . . . . .
21. Inertial Motion. R'elation to Geodesic Lines in the Arbitrary
Motion of a Conservative System . . . . . . . . . . . .
22. The Universal Integral Invariant of Poincar6. Lee Hwa-
Chung's Theorem .... .. ........ ....
23. Invariance of Volume in the phase Space. Liouiville's Theorem
\.Chapter 4. CANONICAL TRANSFORMATIONS AND T H E HAM1 L-
TON-JACOBI EQUATION . . . . . . . . . . . . . . .
24. Canonical Transformations . . . . . . . . . . . . . . .
25. Free Canonical Transformations . . . . . . . . . . . . .
26. The Hamilton-Jacobi Equation . . . . . . . . . . . .
Ha anznulicKoM ~ 3 b t ~ e 27. Method of Separation of Variables. ~ x a m ~ i r .s . . . . . .

@ English translation ,Mir Publishers, 1975

6 Contents
A plying Canonical Transformations t o Perturbation Theory
TTesting
R Structure
~ the Canonical
of an Arbitrary Canonical Transformation
Character of a Transformation. The'
Lagrange Brackets . . . . . . . . . . . . . . . . . .
The Simplicial Nat*ure of the ~ a c d b i a nMatrix of a Canonical
Transformation . . . . . . . . . . . . . . . . . . .
Invariance of the Poisson Brackek in a Canonical ~ r a n s f o r -
mation . . . . . . . . . . . . . . . . . . . . . . . .
Chapter 5 . STABILITY O F EQUILIBHIUM AND T H E MOTIONS OF
A SYSTEM . . . . . . . . . . . . . . . . . . . . . .
Lagrange's Theorem on the Stability of an Equilibrium
Asymptotic Stabil
ve-Systems . . . . . . . . . . . . . . . .
Conditional ~tabilit;. General Statement of the Problem. Sta-
bility of Motion or of an Arbitrary Process. Lyapunov's
Theorem . . . . . . . . . . . . . . . . . . . . .
Stability of ine ear Systems . . . . . . . . . . . . . .
Stability in Linear Approximation . . . . . . . . . . .
Criteria of Asymptotic Stability of Linear Systems . . . .
Chapter 6 . SMALL OSCILLATIONS . . . . . . . . . . - . . . . .
, 40. Small Oscillations of a Conservative System . . . . . . .
,41. Normal Coordinates . . . . . . . . . . . . . . . . .
42. The Effect of Periodic External Forces on the ~ s c i l l a t i o ~of~ s
a Conservative System . . . . . . . . . . . . . . . .
Extremal Propertips of Frequencies of a Consrrvative system.
Rayleigh's Theorem on Frequency Variation with Change in
Inertia and Rigidity of the System. Superimposition of Con-
straints . . . . . . . . . . . . . . . . . . . . . . . .
Small Oscillations of Elastic Systems . . . . . . . . .
Small Oscillations of a Scleronomic System under the Action
of Forces Not Explicitly Dependent on the Time . . . .
Rayleigh's Dissipative Function, The Effect of ~ r n a l iDissi-
pative Forces on the Oscillations of a Conservative System
The Effect of a Time-Dependent External Force on Small Osci-
llations of a Scleronomic System. The Amplitude-Phase Cha-
racteristic . . . . . . . . . . . . . . . . . . . . . . .
Chapter 7. SYSTEMS WITH CYCLIC COORDINATES . . . . . . .
48. Reduced System. The Routh Potential. Ridden Motions.
Hertz' Conception of the Kinetic Origin of Potential Energy
49. Stability of Stationary Motions . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . .
Name Index . . . . . . . . . . . . . . . . . . . . . . . . . .
Subject Index . . . . . . . . . . . . . . . . . . . . . .
Preface
In the literature on mechanics there is no single generally accepted
iilterprotation of tlie t,ernl "analytical mechanics". Some writers
identify analytical mechanics with theoretical mechanics.* Others
maintain that an exposition ir, generalized coordinates constitutes
the distinguishin@ feature of analytical mechanics. A third poir~t
of view - the one adhered to by the .author in calling this book "Le-
ctures in Analytical Mec,h2nics''-consists in the fact that analytical
mechanics is characterized both by a specific system of presentation
and also by a definite range of problems investigated.
The characteristic feature of the system of presentation of analy-
t,ical rneclianics is that general principles (differential or integral)
serve as the foundation; then the basic differential equations of moti-
on are derived from those principles analytically. The basic content
of analytical mechanics consists in describing the general principles
of rnechanics, deriving froill them the fundamental differential equa-
t,ions of motion, and investigating the equations obt.ained and
methods of integrating them.
Analytical mechanics constitutes a part of the course of theore-
tical mechanics that conles wit,hin the syllabus of mechanico-mathe-
inatical, physical and physical-engineering departments of ur~i.
versities and pedagogical institutes. Yet the general syllabus in
lheoretical mechanics of techi~ological institutes either does not
contain analytical mechanics a t all or preserves only elements of
i t , though modern technology poses problenis that lie beyond the
fundamental course of theoretical mechanics as given i n the tradi-
I.ional divisions of "statics", "kinematics", and "the dynamics of
a particle and a system". Research engineers working in diverse areas
ol modern technology must also master the general methods of
259
: ~ ~ ~ ~ l ~rnechallics
t i c a l which yield a universal analytical apparatus
for illrestigating comples problems that refer not only to purely
260 Inechaliical, but also t o electrical and electromechanical phenomena.
261
For example, t h e w&known courses of theoretical mechanics of G. K . Sus-
I,,v ;,,l(\Ch, de l a Vall&-Poussin were called courses of analytical mechanics
I)y l llcmthors.
i
l
8 Preface
The present text does not claim to cover fully the material of
analytical mechanics. It developed out of a course of lectures given
by the author over a period of six years in the Moscow Physico-
Technical Institute. This circumstance determined the choice of
material and the manner of its presentation.
The course of analytical mechanics is a foundation supporting
such divisions of theoretical physics as quantum mechanics, the
special and general theories of relativity, and so forth. For this
reason, a detailed presentation is give11 of variational principles
and the integral invariants of mechanics, canonical transformati-
ons, the Hamilton-Jacobi equation, and systems with cyclic (igno-
rable) coordinates (Chapters 2, 3, 4 and 7). Following the ideas of
Poincar6 and Cartan, the author takes the integral invariants of
mechanics as the basis of presentation. Here they do not represent
an embellishment of the theory but its actual workaday machinery.
The technical applications are associated with a consideration of
constrained systems, which are studied in detail in Chapter 1.
In a special section of that chapter, which is devoted to electrome-
chanical analogies, the possibility is investigated of extending the
analytical methods of mechanics to electrical and electromechanical
systems. In Chapters 5 and 6 are given applications of analytical
mechanics to Lyapunov's theory of stability and the theory of
oscillations. Elements of modern frequency methods are given along
with the classical problems in the theory of linear oscillations.
Problems in the dynamics of rigid bodies are taken up in individual
examples.
It is aisumed the reader is acquainted with the general fundamen-
tals of theoretical mechanics and higher mathematics. The text is
designed for undergraduate and graduate students of mechanico-
mathematical, physical and physical-engineering departments of
universities, and also for research engineers and other specialists
who feel a need to extend and deepen their knowledge in the field
of mechanics.
CHAPTER 1
The Differential Equations of Motion
of an Arbitrary Sys em of Particles'
1. Free and Constrained Systems.
Constraints and Their Classification
The motion is studied of a system of particles P, (v = 1, . . ., N)
relative to some inertial (Galilean) system of coordinates. Imposed
on the positions and velocities of the particles of the system are
restrictions of a geometrical or kinematical nature, called constraints.
Systems with such constraints are known as constrained systems
in contradistinction to free systems, in which such constraints are
absent .
Analytically a constraint is expressed by the equation
where the left-hand side includes the time t, the radius vectors r , and
the velocities a , = P , of all points P, of the system (v = 1, . . ., N).
In the special case when the velocities r, do not enter into the con-
straint equation (l), the constraint is termed finite or geometric.
Analytically, it is written as
f ( t , ,l*,) = 0 (2)
In the general case, the constraint (1) is called diferentinl or kine-
matical. From now on we shall confine our consideration solely t o
such differeutial constraints whose equations contain the veloci-
ties of the particles in linear form:
N .
2 Zvrv+D=O
v= i
(3)
* A dot over a letter denotes differentiation of the given quantity with
respect to time. All radius vectors are constructed from a single pole that i s
.
stationary in the given system of coordinates. Also, f (t, r,, r,) is an abridged
.
notation for the function f (t, ril . . ., rlv, pi, . . ., rN).Such abbreviated
notation will be used throughout this text. If X,, y,, and z , are the Cartesian
coordinates of a point P, in the given system of coordinates (v = 1, . . ., N )
.. -
then the function f. may be regarded as a function of 6N f 1 scalar arguments t,
X,, y,, z,, X,, y,, and Z, (v = 1, . . ., N ) . The function f and all other functions
that will be encountered 'in the sequel, are assumed (unless otherwise stated)
to be continuous together with those of their derivatives which occur in appro-
priate parts of the text.
 10 Equations of Motion of a n A r b i t m r y S y s t e m of Pnrticles [Ch. 1 Sec. 11 Free and Constrained Susterns 11
,
where Z,r, is the scalar product of the vectors I , and r,, and the The finite constraint (2) or (2') is termed stationary if t is not
vectors Zv and the scalar D are specified functions of t and of all expressed explicitly in the constraint equation, i.e. if - a/ == 0.
.
l * , , ( & v = 1 , . ., N). I t is assumed here that the ~ e c t o r sI, cannot
at
In this case, the left-hand side of the equation of differential cons-
all vanish a t the same time. traint (4) is linear and homogeneous in the velocities. By analogy,
Given a h i t e constraint of type (2), a system cannot occupy an the differential constraint (3) or (3') is termed stationary if D = 0
arbitrary position in space at every given instant of time. Finite and if the vectors I, in equation (3) [and, respectively, the coef-
constraints impose restrictions to possible positions of the system ficients AV,B,, and C, in equation (3')l are not explicit functions of t .
a t time t. But, with a differential constraint alone, the system may A system of particles is called holonomic if t,he particles of the
occupy any arbitrary position in space a t any tilne t . However, in system are not subjected to differential nonintegrable constraints.
this position the velocities of the particles of the system cannot any Thus, a holonomic system is any free system of particles and also
longer be arbitrary, since the differential constraint imposes restric- any constrained system with finite or differential but integrable
tions on these velocities. constraints. All constraints in a holonomic system may be written
Each finite constraint of type (2) implies, as a consequence, in closed form.
a differential constraint whose equation is obtained bp termwise A system is called nonholonomic* if there are differential noainte-
differentiation of equation (2): grable constraints.
N
A system is termed scleronomic if only stationary constraints are
imposed, otherwise it is rheonomic.
Example 1. A particle is constrained to move over a surlace. Let the
where a?., af (v I . . . , N . But such a differential
-= p a d v f
equation of this surface he given in the form
f (?')=0 (6)
constraint is not equivalent to the finite constraint (2). I i is equi- or
valent to the finite constraint f ( X , y. q = O . (6')
This is a finite stationary constraint.
f ( t , ,vv)- -- c (5) If the surface is moving or undergoing deformation. the time t enters
t h e equation of the surface cxplicitly:
where c is an arbitrary constant. For this reason, the finite constraint
f (t, r)=O (7)
(4) is called integrable.
I t will be noted that in rectangular Cartesian coordinates, the or
constraint equations (1) to (4) are written as: f ( t , 2, y , z)--o. (7')
I n this case, the constraint is finite but nonstationary.
Example 2. Two particles are connected by a rod of constant length I.
Ilere the constraint equation is of the form
(gal - rs3)?- P =0 (S]
or
(xi--x2)+(y1- y2)2+(~i-22)2--13=0. (8')
This is a holonomic sc!eronomic system.
Let us note that a rigid body may be regarded as a system of particles equi-
distant from one another, that is, subjected to constraints of type (8). From
this point of view, a free rigid body is a special case of a constrained holonomic
scleronomic system of particles.
Exam Be 3. Two particles are connected bp a rod of variable length 1 =
* If F, -: zvl 4- +
z V k , where i , j , k are mutually ortl~ogonalunit - f ( t ) .T\e constraint equation is written thus:
vectors of the coordinate axes, then

* Nonintegrable differential constraints are themselves frequently called

**Av, DV, and CV(v=1, . .., N) are scalar funct.ious of I, ~ 1 yi.
, 21, . .. nonholonomic. Somctilnes integrable differential constraints are termed serni-
- ,. XN, YNI Z ~ v
I~olonomic,
 12 Equations of Motion of an Arbitrary System of Particles [Ch. I Sec. 21 Possible and Virtual D isp lacements 13

or We replace the finite constraints by the differential constraints

( ~ , - ~ 2 ) 2 + ( ~ 1 - ~ 2 ) ~ + ( ~ i - z 2 ) ~ - f(t)=O.
~ (9') t h a t follow from them:
This is a holonomic rheonomic system. N
Example 4. Two particles in a plane are connected by a rod of constant
length I and are constrained to move in such a manner that the velocity of the
middle of the rod is in the direction of the rod (the motion of a skate on a plane).
The constraint equations are written as follows: The system of vectors v, will be called possible velocities for a certain
q=O, z2=0 instant of time t and for a certain possible (at that instant) position of
(~i-~2)2+(~i-~2)2-~2=o +
the system if the vectors v, satisfy d g linear equations (2) and (3).
. . . .
xi+x2 Yl+Y2
(10) Thus, possible velocities are velocities permitted by the con-
-- straints. For every possible position of the system a t time t there
Xi--% Yi-Y2 exists an infinity of systems of possible velocities. One of these
This is a nonholonomic system because the last equation in (10) defines a non- systems of velocities is realized in the actual motion of the system
integrable differential constraint. at time t.
In addition t o constraints of type ( l ) , which are called bilateral We shall call the system of infinitesimal displacements
constraints, mechanics also studies unilateral constraints, which are
written in the form of a n inequality: where v, (v = l , . . ., N ) are the possible velocities, possible infi-
nitesimal displacements or, for brevity, simply possible displacements.
Multiplying equations (2) and (3) termwise by dt, we get equations
An example is the case of two particles connected by a thread of t h a t determine the possible displacements:
length l. This constraint is expressed by the inequality
l2 - ( ~ ~ - >r o~ ) ~ (12)
If in condition (11) we have an equal sign, i t is said that thecon-
straint is taut.
The motion of a system on which a unilateral constraint is impo-
sed may be divided into portions so that in certain portions the
constraint is taut and the motion occurs as if the constraint were Let us take two systems of possible displacenlents a t one and the
bilateral, and in other portions the constraint is not taut and the same instant of time and for one and the same position of the system:
motion occurs as if there were no such constraint. In other words, dyv = v, dt, and d l r v = 29; d t (v = 1, . . . , N).
in certain portions a unilateral constraint is either replaced by
'
a bilateral constraint or is eliminated altogether. For this reason Both d r , and d l r v satisfy the equations (5), while the differences
we shall henceforth consider only bilateral constraints. 6 r v = d 1 ~ r v - d r , ( v = l , ..., N ) (6)
satisly the homogeneous relations
2. Possible and Virtual Displacements. N
Ideal Constraints
Let there be imposed, on a material system, d finite constraints
f a ( t , r v ) = O ( a - l , ..., d) (1) 2
v= i
z * v 6 r v = o @ = l , ..., g)
and g differential constraints*
N The dilferences 61sv = d l r v - d r , will be called virtual displace-
ment~.Any system of vectors 6 r v satisfying equations (7) is a system
of virtual displacements. The equations (7) for virtual displace-
* In equations of differential constraints, in place of r v wc write v,. rno~rl,adifler from equations (5), which define possible displacements.
I4 Equations of Motion of an Arbitrary Systeni of Particles
--
[ch. I
in the absence of the terms g
--
dt and Do dt. I t is said for this reason
that virtual displacements coincide with possible displacements in the
case of 'frozen' constraints.
Indeed, in 'freezing', the time t that enters into the equations of
finite constraints becomes fixed, that is the constraint congeals, as
i t were, in the configuration that i t had a t time t. Then the terms
at dt do not appear when differentiating the functions f,, and
the first d equations (5) coincide with the corresponding equations
( 7 ) . For a differential constraint, 'freezing' signifies imposing on i t
a stationary character, i.e. the elimination of D p on the left-hand
side of the constraint. equation and the fixing oi t which enters expli-
citly in the coefficients lPV.Then the last g equations (5) will also
coincide with the corresponding equations (7).
We can say that virtual displacements are displacements of
points of a system from one possible position of the system a t time
t to another infinitely close possible (for the same instant of time t)
position of the system.
In the case of stationary constraints, virtual displacements coin-
cide with possible displacements.
Example i . A particle is in motion on a fixed surface (Fig. 1).
In this case, any vector v constructed from the point P and tangent to the
surface at this point will constitute a possible velocity. The corresponding pos-
sible displacement d r = t* d t likewise lies in the plane tangent to the surface
Fig. 2 . Fig. 2.
at the point P. The difference O r = d'r - d r of thr two tangent vectors is ill
turn a vector tangent to the surface at this point. Thus, any vedtor constructed
from P and lying in the tangent planc may be regarded as a certain d r and a s
a certain 8 r . Here thr constraint is ~tationaryand the virtual displacements
coincide with the possible displacements.
Example 2. The constraint is a surface S which is itself in motion (as a
rigid body) with a certain velocity zc relativc to the original system of coordi-
nates (Fig. 2).
Sec. 21 Possible and Virtual Displacements 15
In this case, the possible velocity v is obtained from an arbitrary vector
ai that is tangent to the surface by adding to it the velocity U :
w=?'~+w
Therefore,
d r = LT dt = 2.1 d t $ ~d t .
Similarly, for another possible displacement
a'?-= L'; d t + U dt
the virtual displacement
6 r = . d 1 r - d r =. (c;-vl) d t
is (unlike d ~ a ) vector lying in a plane tangent to the surface at the point P
(Fig. 3). The vector 8 r is a possible displacement for the 'stopped' surface S.
Fig. 3.
In Cartesian coordinates, the vector 6 r v is characterized by three
projections on the axes Cjx,, 6yv, 62, ( v = l , . . . , N ) , and the
equations ( 7 ) which define the virtual displacements may be written
in the following form:
If these d +
g equations are independent, then among 3N virtual
increments of the coordinates 6xv, ayv, 62, there will be n = 3 N -
- d - g independent virtual increments, where n is called t h e
number of degrees of freedom of the given system of particles.
Let the corresponding forces E', (v = 1 , . . ., N) be impressed
a t the points P, of the system.* If the constraints were absent, then
by Newton's second law we would have the relations m,w, =
= Fv(v = 1 , ..
., N) between the masses m,, the accelerations
W, and the forces P,. Given constraints, the accelerations W, =
-
-- - l F, (at a given instant of time t, in a given position of t h e
mv
* By F, is understood the resultant of all forces applied directly to the
.,
particle P, (v = 4, . . N ) .
16 Equations of Motion of an Arbitrary System of Particles [Ch. I
particles of the system v,, and for given velocities v,) may prove
incompatible with the constraints. Indeed, differentiating the equa-
tions (3) and (2) termwise with respect to the time, we get an analy-
tical expression for the restrictions imposed by the constraints on
t h e accelerations W , of the particles of the system:*
N AT . ..
1 to the important class of ideal constraints.
The accelerations W, = -F, may not satisfy these relations.
'nv
Then the materially effected constraints will act on the particles
of the system P, with certain supplementary forces R, (v = 1 , . ..
. .
. , N); these forces are called the reaction forces of the constraints.**
The reactions that arise are such that t.he accelerations determined
from the equations
are already permitted by the constraints.
.
Unlike the reactions R, (v = 1 , . ., N), the preassigned for-
ces P, (v = 1, . . ., N) are termed eflectiue forces. Effective forces
are ordinarily specified as known functions of the time, position
and velocities of the particles of the system:***
F v = F v ( t r vav, V,) ( v = l , .*., N). (10)
The basic problem of the dynamics of a constrained system con-
sists in the following.
Given the effective forces P, = IfT, (t, T , , v,) and the initial
positions TO, and the initial velocities 230, of the particles of the system
(v = 1 , ..., N)-both are compatible with constraints-it is requi-
red to determine the motion of the system and the reactions of the
constraints B, (v = 1 , ..
., N).****
* The left sides in relations (8)are linearly dependent on the accelerations
U),. As will readily be seen after differentiation, these left-hand sides are also
dependent on t, P,, v, (v = 1. . . ., N ) .
+
** In the case of several constraints (d g > 1), R, is the resultant of
all the reactions of constraints for the point P, (v = 1, . . ., N ) .
*** In the general case. the rieht-hand sides of (10)
. , deoend on t and also
on all r, anJ v, (v = 1; . . ., X ) .
**** In the case of a free system, there is no problem of determinin reactions
and there only remains the problem of determining the motion of t i e system.
Sec. 21 Possible and Virtual Displacements 17
If nothing is known about the nature of the constraints except the
defining equations (1) and (2) and, consequently, nothing is known
about the reactions R, produced by these constraints, then the
above problem is indeterminate, since the number of scalar quan-
tities X,, y,, z,, R,,, R,,, R,, that have to be determined is greater
.. ..
than the number of available scalar relations. i.e. the e~7:ations
mvxv = F v x + RVI, mvyv = F,, +
Rvy, ~ V Z V= F v z
the constraint equations (1) and (2) [6N > 3 N d
RV, and
+ +
g].
+
For the basic problem of dynamics t o become determinate, i t
is necessary t o have some kind of additional 6 N - (3N d + +
g) =
= 3 N - d - g = n independent relations between the sought-for
quantities. These relations can be obtained if we confine ourselves
Constraints are termed ideal if the sum of the works of the
reactions of these constraints on any virtual displacements is equal
to zero, i.e. if
2 KvGrav= 0.
v= 1
(11)
This equation may be rewritten i n expanded form:
Among the 3 N quantities 8xv, ay,, 82,, there are n independent
ones (n = 3 N - d - g is the number of degrees of freedom of the
given system). I t is therefore possible in (11') t o express 3 N - n
dependent increments dx,, ay,, dz, in terms of n independent incre-
ments and equate to zero the coefficients of these independent incre-
ments. We then ~ b t a i nthe n relations still lacking and needed t o make
determinate the basic problem of the dynamics of a constrained
system.
After considering the following examples of ideal constraivts i t
will be clear how natural and important practically is the class of
constraints that we have isolated.
E x a m ~ l e1. A oarticle is constrained to move on a stationarv smooth sur-
face ( ~ i 4).
~ :
In this case, any possible displacement dr, like any virtual displacement 6r,
lies in a olane that is taneent to the surface at the ooint P . and the reaction
of the smooth surface is directed normally to the surface at this point. For this
reason, we always have
R d r = O or R6r=0
Example 2. A particle is constrained to move on a mobile or deforming
smooth surface (Fig. 5).
In this case, the possiblevelocity of the article and, hence, the infinitesimal
~Iisplacementd r = r dt no longer lies in t i e tangential plane (see Example 2
18 Equations of Motion of an Arbitrary System of Particles [Ch. 1 Sec. 21 Possible and virtual Displacef~fents 19

on page 14). The virtual displacement 6 r , which is an infinitesimal possible Further

displacement for a 'stopped' or 'frozen' surface, lies in the tangential plane.
Inasmuch as the reaction, in the case of a mobile or deformin smooth surface
is directed along the normal to the surface. it follows that R %r= 0 (whereas Let R2=c(r2-pi), then
R d r # 0).
Thus, a smooth surface, whether fixed or mobile or deforming, constitutes ~ ~ 6 r , + ~ ~ 6(p2-pi)
r ~ = dc (r2-ri)=;d (rz-ri)2=0
an ideal constraidt.
since (r2-r i ) 2 =const.
It may be taken that an absoldtely rigid body is a system of particles in
which a constraint of the type considered is imposed on any two particles.
For this reason, a rigid body may be regarded as a system of particles subject
to ideal constraints. In the absence of other constraints (other than those that
effect the rigid connection of the particles of the body) a rigid body is called
free.
Example 4. Two rigid bodies are connected by a hingc a t the point A
(Fig. 7). Disregarding the mass and the dimensions of the hinge, i t may be
asserted (as in the previous example) that R i + R2 = 0. Then

Example 5. Two rigid bodies i n motion have their ideally smooth sur-
faces in contact (we disregard frictionl) (Fig. 8). In this case too. we have
Fig. 4. Fig. 5. Fig. 6.

Example 2 explains vividly why, when determining nonstationary

ideal constraints, it is necessary to equate to zero the work of the
reaction forces on arbitrary virtual displacements and not on possible
displacements.
In the examples that follow we will encounter only stationary
constraints.*
Example 3. Two particles are connected by a rod of invariable length and W
of negligibly small mass (Fig. 6). Fig. 7. Fig. 8.
Let us denote by R , and R 2 the reaction forces of constraints impressed on
particles P, and P z . Then by Newton's third law, the rod is acted upon by the
forces Ni= -9,and NZ= - R z Denoting by m and W the mass of the rod R I + &=O. Here, R i and R 2 are directed along the common normal to the
and the acceleration of its centre of inertia, and by I and E the central moment surfaces. On the other hand, the relative velocity of these bodies at the
of inertia and the angular acceleration, we will have point of contact is v2-v,, and hence the difference of the possible displa-
cements dr2-dri= (v2-v,) dt lies in the common tangent plane. Therefore

where L is the total moment of the forces N iand N Zabout the centre of inertia. Example 6. Two rigid bodies are i n motion with their ideally rough
+
But it is given that m = 0 and I = 0. Hence, N i N Z = 0 and L = O.**
From these equations it follows that the forces Ni and N Z and wnsequrntly
(toothed) surfaces in contact. In this case the relative velocity of sliding
is ,u2-vi=O. Consequently, we also have dr2- dri=(c2--23,) dt=O. The-
R , and R z are in direct opposition, i.e. are directed along the rod. refore, here too
+
R i b i R26rZ= R 2 ((Er2--
dri) =0
* From the definition of ideal constraints it follows that a nonstationary
A complex mechanism may be regarded as a system of rigid bodies
constraint is ideal if all its configurations at different instants (regarded as sta-
tionary constraints) are ideal. which are connected in pairs either rigidly or by hinges, or the surfa-
d -
** If the motion of the rod is not plane-parallel, the scalar equation I e = L
-
is replaced by the vector equation - (Io)= L, where I is the tensor of inertia
ces of which are in contact. If we consider all rigid links absolutely
rigid, all articulations as ideal and all contacting planes as ideally
dt smooth or ideally rough, then any complex mechanism may be inter-
and o is the angular velocity; from I^ = 0 it again follows that L = 0. preted as a system of particles that obeys ideal constraints.

20 Equations of Motion of a n A r b r t m w Sustent of Particles
I t will be noted that in many cases this idealization is not permi-
ssible. For example, neglect of the forces of friction may on occasion
substantially distort the physical picture of a phenomenon. In
such a case, the condition of ideal constraints should be rejected and
other conditions should be taken in its place that follow from the
nature of the constraints ahd the laws of friction.
111 this interpretation, the concept of ideal constraints becomes
almost universally applicable.
From now on it will be assumed that all constraints imposed on
a system are ideal.
3. The General Equation of Dynamics.
Lagrange's Equations of the First Kind
The following equations hold for particles of a constrained
system:
m,w, = E', +
R, (V= 1, , N).. . (1)
where m, is the mass of the v t h particle, W , is its acceleration,
and F,,and R, are, respectively, the resultant of the effective
forces and the resultant of the forces of reaction operating on this
particle (v = l , . . . , N). Since the constraints are ideal, for any
position of the system under any virtual displacements, we have
Substituting, here, in place of the reaction forces B, their expres-
sions from equations (1) and multiplying both sides of the resul-
ting equation by - 1, we get
5
v= i
( F , - m,w,) 6 r v = Q.
This is known as the general equation of dynamics. I t states that,
given a system in motion, a t any instant of time the sum of the works
of the effective forces and the forces of inertia on any virtual displacernents
is zero.
Thus, the general equation of dynamics always holds for any
motion that is compatible with constraints and that corresponds
to the specified effective forces F, (v = 1 , . . , N ) . . = 3 N - d -- g increments. Here, the determinant J, which is made
Conversely, let there now be given a certain motion of a system
that is compatible with constraints and for which motion the general
equation of dynamics (3) holds. Then, assuming
Sec. 31 T h e General E q u a t i o n of ~ ~ n a r n i c s 21
[Ch. 1
we will have the equations (1) and (2). Thus, a t any instant of time
i t is possible to choose reaction forces R, such that, by virtue of (2),
would be admissible for the given constraints aiid for which the equa-
tions (1) obtained from Newton's second law would hold. We consi-
der that these reaction forces R, are actually realized ("the hypothe-
sis of the realization of admissible reaction forces') and that con-
sequently the motion under consideration corresponds to the given
(v = 1 , . . ., X ) . Thus, the general
effective forces P, ( t , T , , tab,)
equation of dynanzics expresses a necessary and sufficient condition
for motion compatible with constraints to correspond to agiven system
of elgective forces F , (v = 1, . . ., N).*
Let us find expressions for the reaction forces R, by means of the
so-called undetermined multipliers of Lagrange. We write out the
relations defining the virtual displacements of particles of a system
(see Sec. 2):
iv
5
v= l
16,6~,= Q (p= 1, . . ., g ) . (5)
Multiplyi-ng the equations (4) and (5) termwise by arbitrary scalar
m.ultipliers --A, and - yB and adding termwise the resulting
equations to equation (2), we get
N d g
This relatiori may be writtell in expanded form as
N d P
(3)
Here, we used {y), and {z), to denote in abbreviated fashion the
expressions that differ from the coefficient of 6zv, given in formu-
la (6'), by replacing the letters X , A with y , B or z and C, respe-
ctively.
The relations (7'), Sec. 2, permit expressing d g out of the +
3 N virtual increments 6zv, 6 y v , 62, in terms of the remaining n =
* Here, i t should be kept in mind that the general equation of dynamics (3\
is actually not one cquation, but a set of equations because arbitrary virtual
displacements may be substituted into equation (3) i n place of 6r, (v =
== 1, . . ., N) for any instant of timc 1 .
22 Equations of Motion of an Arbitrary System of Particles [Ch. 1 Sec. 3) The General Equation of Dynamics 23

up of the coefficients of the '6dependent" increments in the equations a non-holonomic system. In these equations the number of unknown
( 7 7 , Sec. 2, is different from zero. scalar quantities (and, hence, the number of equations) is 3N - d ,
We choose d + g multipliers h, and p6 so that in equation (6') i.e. 2d +
g units less than in the set of equations (8) and (9).
the coefficients of d 3-g "iridependent" increments vanish. This may Exam le. Two ponderable particles M , and M 2 of identical mass m = 1
be done, and done uniquely, since the determinant J of the coeffi- are j o i n e t by a rod of invariable length 1 and negligibly small mass. The system
cients of the quantities h,, .pg being determined is not equal to zero. is constrained to move in the vertical plane and only in such manner that the
Then the only thing remainmg in (6') are the terms with the indepen- velocity of the rdidpoint of the rod is directed along it. Determine the motion
of the particles M , and M,.
dent increments 8xv, ay,, 82,. But then the coefficients of these Let X,, y, and X,,. y , be the coordinates of the particles Mi and M,. We write
independent increments must also be zero. In other words, the the constra~ntequat~ons:
undetermined multipliers A, and p* may be chosen so that all the
scalar coefficients in (6') and, hence, all the vector coefficients in
(6) vanish. But then
d g
B,= ~ h , s +Z p R l s v ( v = l , ..., N). (7) The Lagrange equations with undetermined multipliers h and p are of the
a= 1 p= i
form
..
We have obtained a general expression for the reaction forces xi= -a (xz-xi)-p (YZ-Y~) (11)

-
of ideal constraints i n terms of the undetermined multipliers
of Lagrange h,, p~ ( a = l , . . ., d; P l , . . . , g).
Putting the expressions (7) for R V into equation ( l ) , we get
the so-called Lagrange equations of the first kind:* .~.z = - - g r h ( ~ 2 - ~ i ) + p ( ~ ~ - ~ i )
We determine h and p from the equations (11) taking into account the first
equation of (10):

T o these equations one should add the constraint equations:

By replacing each vector equation by three scalar equations, we .. .. .. ..

I t will be noted that equations (12) are obtained from (11) if in the latter
may consider that equations (8) and (9) constitute a set of 3N d + + we replace h by - h and X,, y i by xz, yz. Thus, determining h and y from
+ g scalar equations in 3N + +
d g unknown scalar quantities t h e equations (IZ), we find
X y , 2 , A , p Integrating this set we get the final equations of
motion and, a t t e same time, from equations (7) we get the magni-
tudes of the reaction forces of constraints. However, integration of
such a set of equations is ordinarily very involved due to the large
number of equations. That is why the Lagrange equations of the Equating the appropriate expressions for p and h in the formulae (13) and (14)
first kind find little use in actual practice. we get (after simple manipulations)
In Secs. 6 and 10 we will derive the Lagrange equations of the
second kind for a holonomic system and the Appell equations for

* These equations were obtained by the French mathematician and me-

chanician J . L. Lagrange in his celebrated treatise Mecanique Analytique (1788), We introduce the following abbreviated notation:
in which the fundamentals of analytical mechanics were presented for the first
timc.
24 Equations of Motion of an Arbitrary System of Particles [Ch. 1 Sec. 41 Princinle of Virtual Diswlacements 25

Then equations (10) and (15) will be written as follows: 4. The Principle of Virtual Displacements.
D'Alembert's Principle
An equilibrium position is a position of a system such that the *
system will all the time reside in i t if a t the initial instant of time
~u+i)v+2gu=~ (19)
Equations (17) show that i n a U , v-plane a particle with coordinates U, u
moves in .a circle with radius l and with centre at the origin; its accelera-
tion will all the time be directed towards the centre. The motion of this
-
i t was in that position and the velocities of all its particles were zero.
The position of a system 1.; ( v = 1 , . . ., N ) will be an equilibri-
um position if and only if the "motion" T , ( t ) r$ ( v = 1 , . . ., LV)
satisfies the general equation of dynamics, that is, when
N
'

particle will then be uniform. For this reason,

Fv8rv=0 (1)
u = l cos (P, u = = lsin cp, cp=a=const, rp=at+B (20) v= i

According to (18) we can put in this position of the system.

Equation (1) expresses the principle of virtual displacernents.
P = - Uf , f
Q=-V For some position (compatible with constraints) of a system to be
1 l
a n equilibrium position, it is necessary and sufficient that i n th,isposition
Substituting these expressions into (19) and taking into account (17) and (20), the sum of the works of effective forces on any virtual displacements of
we find
the system be zero.
2g .-
f +-i-~-O, that is, f=-2gsincp Ordinarily, the principle of virtual displacements is applied
to stationary constraints. If the constraints are stationary, then the.
Then term LLcompatiblewith constraints" signifies t h a t the position of the
system satisfies finite constraints. Now since differential constraints
are linear and homogeneous with respect to velocities, they are
Consequently, by virtue of (20) and (21), we have automatically satisfied since we assume v, = 0 ( v = 1 , . . ., N ) .
If the constraints are nonstationary, then the term "compatible
with constraints" signifies that they are satisfied ,for any t if in them
we put r, = and U, = 0 ( v = 1 , . . ., N). Note that in this case
Integrating, we find .
the virtual displacements 6rv ( v = 1 , . ., N ) may also be diffe-
xi+z1= j P df g
rent for different t.
In the general case, the forces F, depend on t , r,, U, ( p =. 1 , . ..
a

Y ~ + Y ~ = - - -2~
acos
a s i n ( ~ + ~ s i n c p c o s c p +a &

(P-B
a2
COS2 (P+2E
2 q+26
- .
. . ., N ) , that is F, F, ( t , r , v,) (v = 1 , . . , N). I t is then
assumed that the equation ( 1 ) h o b s for any value of t if in theexpre-
ssion for F, we put all r, = r; and all v, = 0.
From eqr~ations(16), (20) and (23) we finally obtain I n the most elementary special cases, the principle of virtual
x i = - aYs i n ( ~ + ~ 2s ai 2n c p c o s c p + ~ r p - - - c ol s c p + 6
2a2 2
1 displacements (or, as it is sometimes called when applied to sclero-
nom.ic systems, the principle of possible displacements) was known at
the time of Galileo under the name of the "golden rule of mechanics' *.
yi=-- Y
a
COS (P- -g COS^ (P--sin
2a2
l
2
(P+&
Let forces F , and P, act at the ends of a weightless lever in a
state of equilibrium. Then, denoting by F; and F6 the tangential

i
x 2 = 2 Si n r p g+ ~ s i n p c o2a2
s p + ~21 ~cosrp+6
- (to possible trajectories) components of these forces, and by 61,
and 61, the rnagnitudes of the corresponding elementary possible
Yo s c p - ~ ~2a2
Y z = - - ca o s ~ c p + - - s i2In ( p + E displacements, we will [by virtue of equation ( l ) ]have, except for
(P=at+B J * Galileo attributed substantiation of the "golden rule of mechanics" to
Aristotle. In i t s general statement, the principle of virtual displacenlents was
(whore a, p, y , S, and E are arbitrary constants). first given by J . Bernoulli in 1717.
1
l - 2ti Eqrrations of Motion of a n Arbitrary S y s t e m of Particles [Ch. 1 Sec. 41 Principle o f V i r t u a l D isplacements

sinn.
- . Denote by M the sum of the masses of all bodies and by z, the vertical coordinate
of the centre of gravity of the system of bodies (we take the z-axis to be directed
that is, vertically downwards). Then, according to equation (i), we get
S A =M g s z , =0
and, consequently, the conditions of equilibrium of t h e system have the
( a gain i n force is compensated for by a loss i n displacement and vice aspect
versa-the "golden rule of mechanics'). Sz, =-- 0 (4)
The principle of virtual displacements is the most general prin- Thus, the equilibrium positions of a system of rigid bodies are such posi-
ciple of analytical statics. I t is possible to obtain from i t the con- tions in which the centre of gravity occupies the lowest, the highest or any other
ditions of equilibrium of any actual mechanical system. ~ ~ s t a t i o n a rposition
y~~ in the vertical (~~Torricelli's
principle").

Example l. Let us derive from equation (1) the conditions of equilibrium

of a free rigid body, which in courses of mechanics are ordinarily obtained
throu h reasoning in geometrical statics. Denoting by 8-0 the velocity of some
yartic!e of the rigid body, by m the angular velocity of the body, by P and
the principal vector and the principal moment about the pole 0 for the system
of external forces operating on the rigid body, we equate to zero the expression*
for the elementary work of forces impressed on the rigid body in an arbitrary
infinitesimal displacement of the body:

Since the vectors 2.0 and m are arbitrary, equation (2) holds when and
only when
F=O, and Lo=O (3)
These equations are the necessary and sufficient conditions for equilibrium o f a
free body. Pig. 9.
In similar fashion we obtain the conditions of equilibrium of a constrained
rigid body. For example, let point 0 be fixed. Then v. = 0 and e uation (2)
is of the form d A = Lomdt = 0 , whence, by virtue of the vector m %sing arbi- Example 3. Find the form of equilibrium of a heavy homogeneous chain
trary we get the desired equilibrium condition, L. = 0 . attached at two points. Regarding a heavy homogeneous chain as a system of
If the body can only rotate about a fixed axis u (with unit vector c ) , the rigid bodies (links) i t is possible to write relation (4). But (see Fig. 9, where Oxz
equations (2) take on the form S A = Lame d t = 0 , whence, by virtue of the is the vertical plane and z is the vertical)
arbitrary nature of o the condition of equilibrium follows: L, 70 ; here, L, =
= L,e is the principal moment of external forces about the axis U. zc=-
S zds
Example 2. Let us derive the equilibrium conditions of an arbitrary con-
strained system of rigid bodies experiencing the action of the force of gravity. S
and, since the length of a homogeneous chain does not change during displa-
*The equation 6 A = ( F u o f Low) d t may be obtained in the following man- cements, condition (4) takes the form
ner. We denote by P ithe forces acting on particles of the rigid body, by r i and
V , the radii vectors (drawn from point 0 of t h e body) and the velocities of the
points of application of the forces P i ( i = 1 , 2, . . .), respectively. Then, using
the sign X to indicate vector multiplication, we find
6 S zds=O
This relation may also be written as

21
As is established i n the calculus of variations, i n the class of curves X = f (z)
(substitution of 6ri by d r i .= v i d t is legitimate due to the fact that the rigid passing through two specified points, the curve that imparts to t,he integral
body is a scleronomic system; see page 11). By virtue of Newton's third law,
the principal vector and the principal moment of the internal forces in the rigid
I,ody are zero. For this reason, X
i
P i= F , and 2i
r i X Fi= L,.
28 Equations of Motion o,f a n Arbitrary System of Particles [Ch. l
an ertremal (more precisely, stationary) value, for which
r
6 F dz=0,
21
must satisfy the differential equation*
I n our case F = z
1+ ($1 . TI~erefore, equation (6) assume, the
form
Froin this we have
and
where c is an arbitrary constant. Integrating, we obtain the equation
of a catenary:
where the values of the arbitrary constants c and U are determined from the
conditions of fixing the ends. Thus, the shape of a homogeneous heavy chain
in equilibrium is a catenary.**
Example 4. An invariable plane figure can slide on two of its points A and B
along stationary curves lying in the same plane. Let us find out under the action
of what force F the figure can be in equilibrium (Fig. 10).
Ucsides the effective force F , the figure is acted upon by two reaction forces
directed along normals to the curves, and the lines of action of these three forces
have to intersect in a single point. In other words, the line of action of the force P
has to pass through the point of intersection of the normals to the curves a t the
points A and B, i.e. the line of action of the force F has to pass through the
instantaneous centre of possible velocities C of the figure.***
* This equation was obtained by Euler. For its derivation see page 89
and the note on page 91.
** Galileo thought the parabola to be such a form of equilibrium. Galileols
error was corrected by Huyghens.
*** Hcrc, the magnitude and direction of the force F may be arbitrary.
Sec. 41 Principle of Virtual Displacements 29
We can arrive a t this conclusion by proceeding from the principle of possible
displacements. Indeed, denote by 0 some point on the line of action of the
force F . Then from the condition b A = F v o dt = 0 we conclude that u o l F ,
whence i t follows that the instantaneous centre of possible velocities of thc
figure is located on the line of action of the
f&e P.
Example 5. Some geometrical applications.
We begin with a preliminary remark. Let there
be given, in a plane, a certain curve C and
a point P (the curve C can degenerate into a
point in a special case). Draw through P a nor-
mal to the curve C and denote by r the distance
along the normal from the curve C to the
point P . Thus, r = P O P (Fig. 11). Apply to
point P a certain force F directed along the
normal P O Pand consider F > O if the direction of
the force F coincides with the direction from PO
to P , and F < U otherwise. Thc elementary work
+
of the force F is equal to 6 A = F ( d r o d r ) . But
d r is composed of two elementary displacements:
the displacement along the straight line P O P
(the magnitude of this displaccment is d r ) and the
displacement of the point P caused by rotation
of the straight line P O P . This displacement, like
d r o is perpendicular to the line P O P ,i.e. to the
line of action of the force F . Therefore,*
Fig. 11.
Let there be n curves C,, C z , . . ., C, and a point P in one and the same
plane. Denote by ri, r z , . . ., rn the distances (along the normals) from P to
thesc curves** (Fig. 1 2 corresponds to the case of n = 2). In the same plane
we consider the curve D given by the equation***
f ( r i , r27 rn)=O (11)
* When the curve C degenerates into a point, formula (10) yields an
expression for the work of a central force.
** Some (or all) of the curvcs C,, C z , . . ., C, can degenerate into points.
*** Each of the quantities ri, rz,
. . ., r, is a function of two Cartesian
coordinates of the point P . For this reason, equation (11) is the equation of
a certain curve in the plane.
30 Equations of Motion of an Arbitrary System of Particles [Ch. I Sec. 41 Principle of Virtual D isplacements 31

We shall now show how, using equation ( I l ) , to construct a normal to the D'Alembert7sprinciple. A n y position of a system in motion may be
curve D at the point P .
For any infinitesimal displacement of point P along the curve D we get regarded as an equilibrium position if to the effective forces acting on
the system in that position we add the fictitious forces of inertia.

Now apply to point P the forces F i = - af directed along the normals ri

ar;
(i=1, ...,
n). Then (12) will be written

Now this, in accord with the preliminary remark we made, signifies that the
sum of the works of the forces F , , F z , . . ., 3; is zero in the case of arbitrary
displacement of point P along the curve D . But then any constrafned point
that can move along the smooth curve D will
be in equilibrium under the action of the Fig. 13. Fig. 14.
forces F i , F z , . . . , F , . Therefore, the re-
sultant of the forces F i , F z , . . ., ,F, is D'Alembert's principle permits extending the techniques and
2-35 directed along the normal to the curve D . methods of solution of static problems to problems of dynamics.
We thus have a very simple method for
the geometrical construction of a normal to In particular, it permits determining dynamic reaction forces by
curve D given by equation (11).
Let us consider some special cases:
(a) D is an ellipse. In this case, C l and C z
are points (the foci of the ellipse), equation
(11) is of the form ri + rz - 2a = 0, F , =
I , Fz = 1, and the normal to the ellipse is
the bisector of the angle between the focal
radii-vectors (Fig. 13):
(b) D is a hyperbola. The equation or
the hyperbola is ri - rz - 2a = 0, F! = 1,
Fz = -1, and from the construction ( F I ~24) .
it is easv to see that the tangent to the
Fig. 12. hyperboG is the bisector of the aGgle between
the focal radii vectors (while the normal Fig. 15.
is the bisector of the adjacent angle).
(c) D is a parabola (Fig. 15). C l is a straight line (the directrix), C 2 is a statical methods. Indeed, in an equilibrium position the reaction
point (focus). The equation of the parabola is r, - r2 = 0. As in the ca.w of forces R, differ from F, - m v w v in direction only:
the hyperbola, i t Iollows from the construction that the tangent to the parabola
is the bisector of the angle between the focal radius vector r, and the perpencli- F , - m V w v = - R v ( v = ! , . . ., N ) .
cular r2 dropped on the directrix. R U L then
For the principle of virtual displacements, equation (1) is a special .
mvwv = F + R, (v = l , . . , N ) ,
case of the general equation of dynamics [see equation (3) on page i.e. the reaction forces R , determined by means of the d'Alembert prin-
201. However, the general equation of dynamics may be regarded as ciple are the desired dynamical reaction forces. And so we can add the
an equation expressing the principle of virtual displscements and following proposition to the above formulated d'dlembert principle:
characterizing the position of equilibrium of the system that, results When regarding inertial forces as additional effective forces applied to
if to the effective forces P,,we add the fictitious inertial forces the points of a system, we replace the given dynamical problem by a new
.
- m v w v ( v = 1 , . . , N). We have thus arrived at d'dlembert's sioiical problem. In the new problem, the static reaction forces coincide
principle. with the sought-for reaction forces in the initial dynamical problem.
32 Equations of Motion of an Arbitrary System of Particles
[Ch. I
The following examples will serve to illustrate the use of statical
methods i n the solution of problems in dynamics.
Example 1. A locomotive tender filled with water is in motion with an
acceleration W. I t is required to determine the shape and position of the water
surface.
In the absence of acceleration the surface of the water is horizontal. The
given plane is perpendicular, a t each point, t o the direction of the body forces
Fig. 16. Fig. 17. Fig. 18.
of gravity applied to the water. This static osition is also applicable to the
case of the accelerated motion of the tender i?to each element of mass d m there
is applied an additional fictitious force of inertia d J = - d m w . The surface
of water will be a plane perpendicular to the resultant of the two body forces:
the vertical force of the gravity d m g and the horizontal force of inertia -dmio
(Fig. 16). The water surface will be inclined to the horizon a t an angle cp, where
W
tan q, = - .
g
Example 2. Let us write the differential equation of rotation of a rigid
body about a fixed axis u (Fig. 17). To each element of mass d m we apply a
fictitious inertial force - d m w . Calculate the principal moment of inertia forces
about the axis of rotation
.. ..
- S drnpcpp = -v
"S p2 d m = -lye
J
1
where I, = p2 d m is the moment of inertia of the body about the rotation
axis U. Denote by L, the principal moment of the external forces ap lied to the
body about the axis u.* Then, according to d9Alembert's p r i n c i p i the body
..
will be inequilibrium under the action of the total moment L , - I,cp. Therefore,
(see page 26) this total moment must equal zero. We get
Example 3. A horizontal homogeneous shaft is in uniform rotation with
angular velocity o . A homogeneous disc is mounted eccentrically on the shaft
perpendicular to the axis of the shaft a t equal distances from the bearings. I t
* The principal moment of the internal forces is zero.
Sec. 51 Holonomic Systems 33
is required to determine the pressure on the bearings during rotation of the
shaft.
We consider the inertial forces d m o 2 r t h a t correspond to the separate elements
d m of the disc (Fig. 18). These are converging forces directed away from the
axis of the shaft. The resultant of these forces is equal to J = 02 r d m =
J
= M i o 2 r c , where M i is the mass of the disc, and rc = OC ( 0 is the point of
intersection of the plane of the disc with the axis of the shaft, and C is the geo-
metric centre of the disc). Apply the d'Alembert principle and determine the
static pressure on the bearings, assuming t h a t three forces are applied to the
axis of the shaft:* (1) the force of gravity of the shaft M g ; (2) the force of
gravity of the disc M i g ,and (3) the force J = M i o 2 r C .
The pressure N on each bearing is determined from the formula
The force N has a maximum value
in the position of the disc when the geometric centre of the disc C is located
under the point 0.
5. Holonomic Systems. Independent Coordinates.
Generalized Forces
Let there be given a holonomic system of N .particles P, with
radii vectors T , = x,i+ y , j + z v k (v = 1, . . ., N)** subject to the
finite coristraints
f a ( t , r , ) = = O ( a = 1 , ..., d) (1)
or (in equivalent notation)
We shall assume that d functions f a of 3N arguments X,, y,, z,
(v = 1, . . ., N ) are independent ***; here t is regarded as a para-
meter. We can therefore express d coordinates of equations (l') as
the functions of 3N - d of the remaining ones and the time t and
regard these 3N - d coordinates as independent quantities that
define the position of the system a t time t.
* The resultant of the elementary forces of inertia for a shaft is zero and
is therefore ignored.
** C, j , k are the unit vectors of the X - , y-, and z-axes of the inertial system
of coordinates.
*** Otherwise, if we have, for example, a relation like
one of the constraints (in the given case, fd = 0) would either contradict the
others [for Q (0, . . ., 0 , t ) f 01, or would be a consequence cif the others
[for Q ( 0 , . . ., 0, t ) = O l .
34 Eouotions of Motion of an Arbitraru Sustem of Particles [Ch. 1
However, one need not take Cartesian coordinates for such inde-
pendent coordinates. All the 3 N Cartesian coordinates may be
expressed in the form of functions of n = 3N - d independent.
parameters q,, . . ., q, and of t:
When these functions are put in the constraint equations (l'),
the latter become identities. We will assume that any position of
the system that is compatible with constraints a t the given instant
of time may be obtained from the equations (2) for certain values
.
of the quantities qi, . ., (in.
The equations (2) are equivalent to the vector equations
The scalar functions (2) and consequently the vector functions (2')
are assumed continuous and differentiable.
The minimal number of quantities qi with the aid of which formulas
(2) can embrace all possible positions of a holononiic system coin-
cides with the number of degrees of freedom of the system n = 3 N -
- d (see page 15).
The quantities qi, . . ., q, in formulas (2) or (2') (n is the number
of degrees of freedom) are called the independent generalized coor-
dinates of the system.
For each instant of time t, a one-to-one correspondence is estab-
lished between the possible states of the system and the points of a cer-
tain region in the n-dimensional coordinate space (qi, . . ., g,).
To each position of the system a t time t there corresponds a point
in the space (q,, . . ., q,) that describes this position of the system.
The motion of a point in the coordinate space (q,, . . ., q,) corre-
sponds to the motion of the system.
If all constraints are stationary (a scleronomic system!), then the
tinie t does not appear explicitly in equations (1'). I t is then always
possible to choose coordinates qi, . . ., q, such that the time t does
not enter the equations (2) either. From now on i t is assumed that
for a scleronomic system the independent coordinates qi, . . ., qn
are chosen in precisely that way. Then, for a scleronomic system, the
formulas (2) and (2') take on the form
Sec. 51 Holonomic Systems 35
Example 1. A double pendulum (Fig. 19) moving in a plane has two degrees
of freedom. For the independent coordinates q, and q2 we can take the angles
cp and $.
Example 2. A free rigid body has six degrees of freedom. For the indepen-
dent coordinates we can take the three coordinates XA, I/A, ZA of some point A
of-the body and the three Eulerian angles $, 0 and cp that define the rotation
of a system of axes AEq5 (that is invariably fixed to the body) about a stationary
system of coordinate axes Oxyz.
Euler's angles are determined as follows (Fig. 20). Draw through point A
the axes Axl, Ay,, Az, parallel with the axes Ox, Oy, Oz and in the same dire-
ctions. The line A N that intersects the planes Ax,y, and Agq is called the nodal
Fig. 19. Fig. 20.
line.* Then 0, or the angle of nutation, is thc angle between the axes Az, and
A5; $, or the angle of precession, is the angle between the axes Ax, and AN;
and cp is the angle of pure rotation formed by the axes A N and AE.
By means of three parallel displacements-along the X-, y- and z-axis-by
X*, yA4, ZA the coordinate trihedral Ozyz is displaced to the position Ax,y,zi.
By means oE three successive rotations-through the angle $7 about the axis Az,,
through the angle 0 about the axis A N , and through the angle cp about the axis
Al-the trihedral Axiy,zi is displaced to the position AEqC.
Thus, the quantities xA, y ~ ZA, , $, 0, cp determine the position of the coor-
dinate trihedral AEqc relative to the trihedral Oxyz, i.e. they determine the
position of the given rigid body relative to the initial system of coordinate axes.
Take an arbitrary point of a rigid body. I t is drtermined by specifying i t s
coordinates E, q, 5. Then the coordinates X, y, z of this point may be represented
as functions of the quantities XA, yA, ZA, $ 0, cp. Thus, for example, from
Fig. 20 i t will readily be seen that z = zA -( g sin cp sin 0
t t cos 0.
+ q cos cp sin 0 f
Similar, though somewhat more involved formulas are obtained for X and y
[241. These formulas are a special case of the formulas (2). They do not contain 1
explicitly. A free rigid body is a scleronomic system.
I t will be noted that in the motion of a rigid body the quantities ZA, y ~ ,
z,, $, 0, cp vary, and the decomposition given abovc of the transition from
* Direct the axis A N so that rotation about this axis from the axis Az,
to A5 through the smallest angle is counterclockwise.
3*
 36 Squations oj Motion oj an Arbitrary System of Particles [Ch. I
Holonomic Systems
Sec. 51 37
Oxyz to AEqS into three parallel displacements and three rotations gives thc
impression of arbitrary motion of the rigid body in the form of complex (com- Rut the virtual differentials, i.e. the differentials for fixed "frozen" t ,
pound) motion consisting of six simple motions: three translational nlotiom of the function r , ( t , qi)* are the virtual displacements dr,:
(along the X-, y-. and z-axis) and three purely rotational motions (about the
axes A z , , AN, and A t ) . Since in complex motion the angular velocity is equal
to the vector sum of thc component angular velocities, i t follows that 6rv= 2 --
849
arv
6qt (v = 4, . .. , N ) (8)
W=@,,,$wJ+w, (4) i=i

where a,,,,me, cow. are directed along the axes Azi, AN, A t , respectivclg; Substitute the expressions ( 8 ) into the right-hand side of for-
here, a@==$, Q=B; and qr=q mula (7) and express the elementary work of the effective forces
Example 3. A free particle M has three degrees of freedom. For thc indc-
pendent coordinates WC can take Cartesian coordinates or some other coordinates
of the point. For the case when cylindrical coordinates r , 4, x are taken for

Fig. 23.
Fig. 21. Fig. 22.
on the virtual displacements in terms of arbitrary elementary
q,, qz, 43, the formulas (2) will appear as (Fig. 21): increments 6qi of t h e independent coordinates qi ( i = l , . . ., n):
N n n N n

In the case of the spherical coordinates r, cp, $ (Fig. 22), we will have,
i n place of formulas (5),
x=rcos$sincp, y=rsin$sincp, z=rcoscp (6) where the coefficients of 6qi, or the generalized forces Q i , are
Example 4. A constrained pzrticle M is located on a mobile sphere determined by the equations
N
(~=at)2+(~-bt)2+(z-~t)2=r~
Then n = 2 , and the longitude and latitude on the spherc (Fig. 23) may
be utilized as the independent coordinates:
x = a t - + r c o s q i ~ ~ ~ yq=~b,t f r s i n q i c o s q 2 , * Indeed, when the functions rv (t, qi) (v = 1, . . ., N) are substituted
z=ct+rsinqz
.
into the constraint equations fa (t, P,) = 0 (a= 1 , . ., d), the latter be-
come identities. Let us differentiate termwise the identities obtained (for fixed t).
We then find
To every coordinate g i there corresponds a generalized force
Q i(i = 1, . . ., n). The generalized forces are determined as follows.
N

Consider the elementary work of effective forces on virtual displa-

cements
N where Sr, (v = 1, . . ., N ) are the virtual differentials. But the equations (*I
6A = 2 Fv6rv
v-l
coincide with the first d equatitms (7) on page 13 that determine the virtual
displacements of a holonomic system. Consequently, the virtual differentials
of the radii vectors are virtual displacements of points of a holonomic system.
38 Equations of Motion of an Arbitrary System of Particles [Ch. 1
It will be noted that for practical purposes formula (10) is by far
not always used to find the quantity Qi. Instead, the system is given
a n elementary virtual displacement such that only the ith coordinate
q i receives a certain incremenbwhile the remaining independent coor-
dinates do not change. After that the work of the effective forces
6Ai is calculated on just such a specially chosen displacement. Then,
6Ai = Q i 6 q i , and Qi =-6Ai
6qi
Example 5. A rigid body is constrained to move translat ionally along
the x-axis. Then n = 1 and the abscissa X of some point of the body A may
be taken as the independent coordinate. Here.
6 A = X6x (11)
where X is t h e sum of the projections, on the x-axis, of all effective forces
acting on the body. Obviously X is the generalized force for the coordinate X:
Q=X (12)
Example 6. A rigid body is constrained t o rotate about a certain fixed
axis U. The angle of rotation rp may be taken as the independent coordinate.
Then
where L, is the total moment of all effectiveforces about the axis of rotation, and
Q = L, (14)
Example 7. A free rigid body. For the independent coordinates let us take
the three coordinates XA,y,, 2, of some point A of the body and the three Eule-
rian angles $, 0, cp (see Example 2 on pages 35-36). Then, according to equa-
tion (g),
To determine Q, we impart to the body an elementary displacement along the
x-axis. Then 6yA = 62, = 0, and 6$ = 60 = 6cp = 0. Therefore, 6A =
= Q,6xA. A comparison with (11) yields
Qx - X
Similarly, Qy = Y, Q, =. Z. Here, X , Y, Z are the projections on the stationary
axes X, y, z of the principal vector of all effective forces acting on the body.
We now impart to our body an elementary displacement such that only
the angle 11, changes, while the quantities XA, yA, zA, 0 and rp remain invariable.
Then
6A =Q@$
On t h e other hand, the elementary dispIacement of the body is a rotation
about the axis Az,. Therefore, in accord with formula (13),
QQ =L*
where Lg is the total moment of all effective forces about the axis Azi, about
which a rotation through the angle is performed.
I n quite analogous fashion, Qe = Le and Q, = L,, where Le and L, are
the total moments of the effective forces about the axes A N and Ag.
.YPc. 51 Holonomic Sustems 39
We can arrive at the same expressions for the generalized forces if we take
advantage of the expression for the elementary work of the effective forces
applied to a rigid body (see page 26):*
6A= R 6 r A + L A w dt (16)
where R and L, are the principal vector and the principal moment of the
system of forces about the pole A. Inasmuch as [see formula ( 4 ) j m = q +
+me+ m,, where mu,=$, me= 0 , m,= rp, and t h e projections of the vector
L, on the directions of the vectors wg, we, w, are equal, respectively,
to L,p Le, L,, from formula (16) we find
A comparison of the expressions (17) and (15) gives us expressions for the gene-
ralized forces.
Now let a certain position of the system be a position of equili-
brium. According to the principle of virtual displacements this is
possible if and only if
n
6A = 2 Qi6qi =. 0 (18)
i= i
But the increments 6qi in the independent coordinates qi may be
quite arbitrary. Therefore the equation (18) is equivalent to the
system of equations
Thus, the position of a holonomic system is a n equilibrium position
i f and only if all the generalized forces i n this position are zero.
Example 8. I n accordance with the equations (19), the conditions of equi-
librium of a free rigid body may be written as
(see preceding example). Here, X, Y, Z are the projections, on the coordinate
axes, of the principal vector R of the external forces acting on the body, and
Lu,, Le, L, are the projections of the principal moment LA of these forces on
three non-coplanar directions. For this reason, the scalar equations (20) are
equivalent to two vector equations:
These are the necessary and sufficient conditions of equilibrium of a free rigid
body t h a t had already been estabIished on page 26.
* Since we have to do here with a scleronomic system, i n place of 6
we can write d, and vice versa. Therefore, dr, = Sr, and 811,= d$= $ dt,
6 0 = O dt and 6cp= rp dt.
40 Equations of Motion of an Arbitrary System of Particles [Ch. 1 Sec. 61 L n n r a n ~ eEquations of Second K i n d 41

6. Lagrange's Equations of the Second Kind in On the other hand, from the same equation (6) we obtain
Independent Coordinates
In deriving the differential equations of motion of a holonomic
system in independent coordinates qi, . . ., g,, we shall proceed
from the general equation of dynamics
Therefore, expression (5) for Z i may also be written as follows:

Let us recall the expression, obtained in Sec. 5, for the ele-

mentary work of effective forces:
N n
where T is the kinetic energy of the system:
W here
N
v= 1
The general equation of dynamics (1) gives us
6A+6AJ=0
Quite analogously i t is possible to represent the elementary work
of the inertial forces - mVwv (v = 1, . . ., N): or, by virtue of the equations (2) and (4),

Since the qi are independent coordinates and, for this reason, the
where, by analogy with expression (3), Bqi are absolutely arbitrary increments in the coordinates (i = 1 , . . .
. . ., n), i t follows that (12) can hold when and only when all the
coefficients of 6qi in equation (12) are equal to zero. Therefore, the
general equation of dynamics (12) is equivalent to the set of equations
Z i = Q i ( i = I , ..., n) (13)
which, according to t.he relations (g), may also be written in the
following form:
aT
But the velocity ----=Qi
a ~ i
(i=1, ,.., n ) .
Equations (14) are called the Lagrange equations of the second kind
or the Lagrange equations in independent coordinates.
The quantities qi (i = 1 , . . ., n) are called generalized velocities.
is linearly dependent on qk (k = l , . .. , n). From this formula we
find that The velocities of points of the system v, = r vare expressed in terms
of the generalized velocities (and also in terms of independent coor-
Or, arv . . ., N ) dinates aqd the time) by means of the formulas (6). The quantities
--- (i=1, *..,n; v= 4, ..
ayi qr (i = 1 , . . ., n) are called generalized accelerations.
 42 Equations of Motion of an Arbitrary System of Particles [Ch. 1 Sec. 61 Lagrange Equations of Second Kind 43

After performing the operation =, the

d
left-hand sides of the
The Lagrange equation

Lagrange equations (14) contain the time t, the generalized coordi-

nates qi, the generalized velocities i i , and the generalized accelera-
.. after the substitution
.
tions qi ( i = 1 , . ., n). The generalized forces Qi (i = 1 , . . ., n)
on the right-hand sides of the Lagrange equations are ordinarily

I
qk
specified * as functions of t, q ~ , (k = 1 , . . ., n): assumes the form
..
Iucp = L,
This is the differential equation of the rotation of a rigid body about
The Lagrange equations (14) form a set of n ordinary differential a stationary axis.
Example 2. A double simple pendulum is i n motion i n a plane (Fig. 24).
equations of the second order in n unknown functions qi of the inde- We form a n expression for the elementary work:
pendent variable t. The order of this system is 2n. Note that the set
of differential equations determining the motion of a holonomic
6A = nzigbi mzg& +
system with n degrees of freedom cannot be of order less than 272, where zi = li cos rpi, z2= li COS rpi + l2 COS 9 2 .
since by virtue of the arbitrariness of the initial values of the quan-
tities qi and qi (i = 1 , . . ., n), the solution of the system must
contain a t least 2n arbitrary constants. Thus, the set of Lagrange
11 equations in independent coordinates has the lowest possible order.
In the case of a constrained system, the reaction forces R,
.
(v = 1 , . ., N) have likewise to be determined. The reaction forces
do not enter into Lagrange's equations. This is an essential advan-
tage of Lagrange's equations. After Lagrange's equations have been
integrated and the functions q, (t) (i = l , . . ., n) found, r, =
= r, (t) are determined [by substitution of these functions into for-
..
mulas (2') on page 341 and, consequently, v, = T , , W , = r, and
.
F, (t, r,, r,) (v = 1, . ., N). After that the unknown reaction
forces are determined from the formulas

Fig. 24.
In the case of a free system of particles, Lagrange's equations are
a compact notation of the equations of motion in an arbitrary system Calculating 62, and 6zz, we find
of coordinates.
6 A = -(mi +
,712) gli s i n cpi6cpi -m2glz sin cpzhp2
Example 1. A rigid body is i n rotation about a stationary axis U . For and
the independent coordinate we take the angle of rotation cp. The appropriate
generalized force Q (see Example 6 on page 38) is equal to the moment Q i = -(rn+m2)glisinrpi, Q2= -m2g12sincp2
1 On the other hand,
of rotation L,. On the other hand, T = - 1,qP where I, is the moment
2
of inertia of the body about the axis of -rotation.

* See formulas (3) and (6) of this section and also formula (10) on page 16
and formula (2') on page 34.
 44 Eouations of Motion of an Arbitraru Sustem of Particles [Ch. 1 Sec. 71 Investieatine Laeranrre's Eouations 45

I The first Lagrange equation

7. Investigating Lagrange's Equations
In order to form the Lagrange equations i t is necessary first to find
the expression for the kinetic energy as a function of the time t ,
n is of the form the generalized coordinates q i and the generalized velocities qi (i =
.
= 1 , . ., n). Lzt us do this in the general form:
N

It is left to the reader to form the second equation that corresponds to the
coordinate cpz.
Example 3. I t is required to determine the differential equations of motion
of a free particle in spherical coordinates (see Example 3 on page 36, and Fig. 22).
The velocity of the particle is equal to the vector sum of: (a) radial velocity;
(b) rotatory velocity due to the rotation of the radius in the plane of the meri- Here, the coefficients a i k , a i , a. are functions of t , g,, . . . , 411
dian, and (c) rotatory velocity due to rotation of the plane of meridian. The
velocity components are pairwise orthogonal and therefore defined by the equations
N

= Zm,-'--ar
aqi
dr,
agh
(t,k=1, . . . , n)*
To find the generalized force Q,, we displace the particle along the radius.
Then 6A, = F&, where F, is the projection of the applied force F on the dire-
ction of the radius. Whence Q, = F,.
Now let us impart to the particle an elementary displacement along the
meridian. Then F, r8cp, where F, is the projection of the force F on the
tangent to the merldlan.* Therefore
Q, =F,r
Analogously,
Q$ = F$r sin cp Formula ( l ) show s that the kinetic energy of a holonomic sys-
tem is a function (polynomial) of the second degree in the gene-
where F* is the projection of the force F on the tangent to the parallel. ralizcd velocities:
The Lagrange equation of the coordinate r
T=Tz+Ti+To
where

takes on the form

For the coordinates cp and $ we find As was explained in Sec. 1, in the case of a scleronomic system,
t h e time does not explicitly enter into the relation between v,
and qi, and for this reason
.. +2r cos qv$)
m (r sin cp'$+ 2 sin cprq
. . =F*
We have obtained three differential equations of the motion of a free
particle i n spherical coordinates. But then, according to equations (3) and (4),

* We send the tangents (to the meridian) and the parallels in the direction " From formulas (2) it will bc seen that aik = ski ( i , lr = 1, . . ., n).
of appropriate increasing coordinates cp and $.
-
 Equations of Motion of an Arbitrary System of Particles [Ch. I Sec. 71 Investigating Lagrange's Equations 47
46

and The equations (9') show that i n the Jacobian functional matrix
n

Thus, the kinetic energy of a scleronomic system appears i n the

form of a homogeneous function of the second degree (quadratic form)
of the generalized velocities.
It will be noted that in an arbitrary (scleronomic or rheonomic)
holonomic system, the form T, is always nondegenerate, i.e. a de-
terminant made up of i t s coefficients is different from zero:

Indeed, let the columns are linearly dependent, i.e. the rank p of this functional
det (aik)? 0 matrix is less than n. Then among the 3N functions XI, y,, zi, . ., .
X,, y,, z, of the n arguments qt . . ., qn ( t is regarded as. a para-
Then the system of homogeneous linear equations meter) there ar e p independent quantities in terms of which all the
remaining Cartesian coordinates of the points of the system may be
expressed. This is a contradiction, since the minimal number of indep-
endent coordinates of the system is equal to the number of degrees of
freedom n, while p < n. The inequality (7) is thus established.*
The property of coefficients of quadratic form T2 expressed by the
has a real nonzero solution. inequality (7) is very essential and we shall repeatedly make use
Multiplying the set of equations (8) termwise by Xi, then sum- of i t in the sequel. Note that since always T2 >, 0 (T2 is the kinetic
ming with respect to i from 1 to n and utilizing formulas (2), energy in the case of ~lfrozen''constraints!), it follows from inequality
we get
n n N (7) that the quadratic form T2 =
1
2
n
..
aikqiqkis positive definite,
i, h=i
i.e. T2 >, 0, and T, = 0 only when all qi (i = 1 , . . ., n) are equal
to zero. Therefore, for the coefficients aik we have the determinantal
inequalities of Sylvester :
v= l
whence
n
Putting the expression (1) for kinetic energy into the Lagrange
equations

These N vector equations may be replaced by 3N scalar equations: -.

n n n * The rank of the functional matrix (10) may be less than n at individual
(singular) points. At these points the equality det (ai,)? = 0 is possibl~.
In the sequel we shall not consider such special positions of the system.
45 Equations o f M o t i o n o f a n Arbitrary S y s t e m of Particles [Ch. 1 Sec. S ] Theorem on Variation of Total Energy 49

we get
Here, the symbol (**) indicates the sum of the terms not involving
second derivatives of the coordinates with respect to the time. The
Let us now consider the general case when in addition to the
potential forces determined by the potential II, the system is acted
upon also by nonpotential forces
d

Q i = ~ i ( t q, j , ij)( i = I , ..., n ) . (4)

Then
right-hand sides likewise do not contain second derivatives since
in the general case they are functions of the quantities t, q j , q j
( j = I , . . ., n ) .
and the Lagrange equations assume the form
Since det (ai,):, , 3 = i # 0 , it follows that the equations (13) may
be solved for the second derivatives and represented in the form
..
qi -Gi ( t , qh, qh) ( i = l , . . . , 71). (44)
We now consider the total energy E, which is equal to the sum
But t,hen, as we know from the theory of differential equations, for of the kinetic energy and potential energy
certain assumptions relative to the right-hand sides G i , which in
mechanics are always assumed to be satisfied*, there is one and only
one solution of the Lagrange equations for arbitrary preassigned ini-
tial qq, q? with t = t o ( i = 1 , . . ., n ) . Thus, the motion of a holo-
and compute the derivative dE =.TO do this, we first find
nomic system is uniquely determined by specifying the initial
position (q!) and the initial velocities (d).
8. Theorem on Variation of Total Energy.
POten tial, Gyroscopic and Dissipative
Forces
If the generalized forces do not depend on the generalized velocities
No tirig that T = T2 + Ti+ To and utilizing the Lagrange equa-
tions (6),we obtain*
Q i = Q i ( t , ~ i ,-.., Q,) ( i = l , ..., n ) (1)
and there exists a function II ( t , q,, . . . , q,) such that

then the forces Qi are called potential, and the function U is the
potential of the forces or the potential energy. Equations ( 2 ) , which
determine the potential II may be written as**

6 A=
n
2 Qi6qi = -6II
* The Euler formula f g zi=nf holds for a homogeneous function
(3)
i= l f (X,, . i= i
. ., X,) of t h e mth degree. Applying this formula to the linear form
f l and t o the quadratic form T2, we find
* For instance. if the functions G;. (.i = 1, . . ., n) have continuous first-
order partial deri~ratives.
** The time t is first fixed when computing the virtual differential 6n.
n
Therefore 6n-
an 6 q i .
2-
%i The validity of these identities also follows directly from the expressions
Li for T2 and T i given on page 45.
 50 Eouations of Motion o f an Arbitraru Sustem o f Particles [Ch. I Sec. 81 Theorem on Variation of Total Energy 51

From this, taking into account equation (7), we obtain The total energy of a conservative system does not change when
the system is in motion.
Equation (16), which does not involve b u t involves the arbi-
i= i trary constant 12, determines the first integral of the equations of
The expression on the right, motion. Equation (16) is called the energy integral.
n Nonpotential forces are called gyroscopic forces if their power is
zero:
n

where SA-~S the elementary work of the nonpotential forces Q , is

and dissipatiue forces if their power* is negative or zero:
ths power of %he nonpotential forces ( i = l , . . . , n). The term
on the right-hand side,
d
+
d t (T, 2To) -
-
dT
(12)
If the potential energy is not explicitly dependent on t , then from
is different from zero only for a rheonomic system (for a scleronomic dE
dT equations (14) and (17) i t follows that z==O, and thus for a
system, Ti = T, = 0, and -- at
= 0). The latter term is nonzero
scleronomic system with gyroscopic forces we also have the energy
only when the potential energy-II is explicitly dependent on the time. integral
Formula (10) determines the change in the total energy of an arbi-
trary holonomic system in motion. Let us consider some special E = cons t
cases. * In the case of a scleronomic system
(a) A scleronomic system. Then

whence after termwise division by d t we find

(b) A scleronornic system where the potential energy is not
explicitly dependent on the time. Then
For this reason equation (17) expresses the condition of gyroscopicity
N
2 Fvvv=O
v= i
Fcr such a system the derivative of the total energy with respect t o
nnd equation (18) the condition of dissipativity
the time is equal to the power of the nonpotential forces.
(c) A conservative system, i.e. (1) a scleronomic system, (2) a sys-
tem where all forces are potential, and (3) where the potential energy
II is not explicitly dependent on the time. According to equation f n the case of a rheonomic system, equation (*) may not hold. Then
(IO), for a conservative system N n
fir,=--dr,,-
a~
-2 dt and from the equation 2 F&,= 2 Qi6qi it follows
at

i.e. for any motion of the system

E = const =h
52 Equations of Motion of an Arbitrary System of Particles [Ch. I Sec. 81 Theorem on Variation of Total Enernv 53

If the rigid body possesses dynamical symmetry, I is the moment of inertia

But if such a system is acted upon by dissipative forces, then about the axisof symmetry, o2is the angular velocity of <'purerotation" directed
along the axis of symmetry, and oiis the angular velocity of precessional motion,
Td ,E< o then the moment L. = I (miX a2)is called gyroscopic. Thus, the forces t h a t
create the gyroscopic moment are gyroscopic.
when the system is in motion; i.e. the total energy diminishes during 2". Let
the motion.* In that case we call the system a dissipative system. n .
In the relations (17) and ( 1 8 ) ,the generalized forces Q iin the gene- Q i = - X bikqk ( i = 1, . . ., n) (21)
k=i
ral case depend on the generalized velocities. Consider some impor-
tant special cases in which this dependence is linear and homoge- where the matrix of coefficients bik is symmetrical,
neous.
1". Let n ..
and let t h e quadratic form bikqiqk be positive; i.e.
i, k = i

and let the matrix of the coefficients yik be skew-symmetrice*:

Then the forces (19) are gyroscopic. Then for the scleronomic system the power of the forces is equal to
Indeed, in this case

and the forces oi

are dissipative.
I n this case the quadratic form
This equation shows that the skew-symmetry of the matrix of coeffi-
cients y i k is not only a sufficient but also a necessary condition for 1 ..
the forces (19) applied to the scleronomic system t o be gyroscopic. R = -2 2 bikqiqk (24)
i , k=i
Example i . For a scleronomic system, the Coriolis forces of inertia are is called Rayleigh's dissipative function. I t is easy to see that t h e
gyroscopic forces. Indeed, the Coriolis force of inertia applied to a particle
p, of a system is determined from the formula generalized forces (21) are obtained from Rayleigh's dissipative
F,=-2mv (ox.v,) function by means of the formulas
where, m , is the mass of the particle P , , v, i s i t s velocity i n the noniner-
t i a l system of coordinate axes'under consideration, and o is the angular velo-
city of rotation of this system relative to some inertial system of coordinates
.
( v = 1, . ., N). Then
N
I f the system is scleronomic and the potential energy does not
depend explicitly on the time, then by virtue of the equalities
X
v= 1
Fvvv=0 ( 1 4 ) , ( 2 3 ) , and (25)
Example 2. Let a rigid body with a stationary point 0 be acted upon by
forces with principal moment L0 = I (o,X oz)where I is a scalar, and let
+
= of o2 be the angular velocity of the body. Then the forces applied
to the body are gyroscopic since their power is equal to zero: This formula points t o the physical meaning of Rayleigh's function:
L0o=O the doubled Rayleigh function is equal to the rate of decrease in
* Energy is dissipated in the case of dis;ipative forces; hence the term.
total energy.
** In the case of a skew-symmetric matrix 11 y i k 11, always y i i = 0 ( i = If the Rayleigh function (24) is a positive definite quadratic form
-
- 1, . . ., n). of the generalized velocities, then one speaks of the total dissipation
54 Equations o f Motion of an Arbitraru Sustem of Particles [Ch. 1 Sec, 91 Electromechanical Analogies 55

of energy. In this case, we will call the system definite dissipative. to the sum of the voltages across the separate elements, and we have
According to formula (26), the total energy of such a system strictly
diminishes.
By way of illustration, let us consider the forces of resistance of the medium
applied to points of the system, the forces being proportional to the first power
of the velocities of the points:
F..,= -pu, (v=I, ..., N) (27) This equation is an analogue of the equation of mechanical
In this case oscillations:
N d2q dq.
a - at2
+bx$cq=Q(t) (4)

where Here to the inductance L there corresponds the inertial coefficient

(generalized mass) a , to the ohmic resistance R the dissipative coef-

9. Electromechanical
In this sec.tion we will show how the equations of analytical
mechanics may be applied not only to mechanical systems but to
electrical and electromechanical systems as well.
Analogies
Fig. 26
1
c
ficient b, t o the coefficient , where C is the capacitance, comes-
sponds the reduced coefficient of elastic force c; the charge q corre-
sponds t o the generalized coordinate q, and the electromotive force
e (t) to the generalized force Q (t).
On the other hand, in the circuit shown in Fig. 26 the currents
passing through the inductor, the resistor and the capacitor are
additive, and so

Fig. 25
Diflerentiating termwise, we get
Consider a circuit with inductance L, resistance R and capacitance
C connected in series (Fig. 25). For these elements, the relation bet-
ween the voltage U (the difference between the potentials a t the ends Here we have another system of analogies in which the voltage U
corresponds to the coordinate q, the mechanical coefficients a , b, c
of the element) and the current i i = -
will, respectively, be equal to
dq where q is the charge
(
dt 1 are replaced by C,
I
, 1 ; and the quantity di corresponds to the
generalized force Q (t).
Two electric systems having the same (up to notations) equations are
two different electric analogues of one and the same mechanical system.
If in addition the circuit has an external source of electiomo- To the kinetical and potential energies, the Rayleigh function,
tive force e(t). then we write that the electromotive force is equal and the generalized force of a mechanical system with one degree
56 Equations of Motion of an Arbitrary System of Particles [Ch. I Sec. 101 Appell's Equations for Nonholonomic Systems 57

of freedom Also, e2=e3=0. P u t

T=- 1 - 1 '
aqa, R = T b q 2 1
1
II=gcq2, Q=Q(t) ei = A sin Qt
Now write down the Lagrange equations:
in the first system of analogies there correspond the quantities
1 . - 1 . . 1
.. 1
Liqi+Riqi+-qi--qz-
1
AsinQt
T=-Lq2, R=-R 2 c12 c12
2 P l n==q2, e=e(tj
and in the second, the quantities

Thus, the systems of electromechanical analogies are determined These equations ,will be the equations of the electric circuit
by the following table: depicted in Fig. 27.

Mechanical: q lol l bl c p I T = ~ a ; 2 1 ~ = ~ b ; n l n = ~ ~ 10. Appell's Equations for Nonholonomic Systems.

Pseu docoordinates
In this section we shall derive the Appell equations that deter-
S . . , mine the motion of a nonholonomic system. Let d finite and g
2nd electrical: differential constraints be imposed on a nonholonomic system (see
U Sec. 1). First utilizing only d finite constraints, we express the
radii vectors of the points of the system in terms of m = 3 N - d
Let us consider the electric circuit shown in Fig. 27 as a n example independent coordinates qi, . . ., q,, and the time t:
of a more complex illustration. r v = r(vt , qil . . ., qm) ( v = 1, . . .,,N) (1)
From this

T
:
:c3 and

l-

Fig. 27
R3
.
However, r , and r , (v = l , . . . , N) also satisfv the differential
constraints*
We form the Lagrange equations adhering to the first system of N
analogies; first we compute
. . 2 Z,&+D~=O (B=I. ..., g) (3)
1 ' 1 v= 1
T = 7 L i d -t-2.L23 (q2 - ~ 3 ) '
where Zsv and Da are functions of f and r , (v= 1 , . . ., N).
R = 1 Riq:' + 1 R 2'4 +1T R3&
' p-

* The functions (l), when substituted into the equations of finite constra-
ints, convert the latter into identities. Therefore, when using the representation
(1) it suffices to consider only the differential constraints.
58 Equations of Motion of an Arbitrary System of Particles Sec. 101 Appell's Equations for Nonholonomic Systems 59
[Ch. 1

Substituting expressions (1) and (2) for r , and r , into the con- by solving the set of linear equations (4) and (5). I n this way we get
straint equations (3) we represent these equations in the form

where his and h i are functions of t and qi, ..

., 4nz.
The quantities n,, which are linear forms of the generalized velo-
where the coefficients A R i of qi and the absolute terms A, are func- cities, will be called pseudouelocities, and the symbols n,, pseudoco-
tions of t and q,, ..
., q,,.
ordinates (S = 1 , . . ., n). In particular, ns can coincide with certain
Thus for a nonholonomic system the coordinates qi, . . , frln . generalized velocities. But in the general case of m +
n quantities,
can take on arbitrary values, but then the generalized velocjties
.
qi, . ., gm can no longer be arbitrary; they are connected by the isand ii
are connected by the relations (5) and (6).
relations (4). Considering the g constraints of (4) independent, we In order to find the restrictions imposed by differential constraints
can express g generalized velocities from equations (4), for example on the virtual displacements 8qi i t is necessary (see Sec. 2) in equati-
.
qn+i, . ., q, in terms of the remaining qi, . . ., qn ( n = nt - ons (4) to discard the absolute terms A, and replace the qi by 6qi
.
(i = 1 , . ., n).
- g = 3 N - d - g is the number of degrees of freedom of the
We then obtain
.
system; see page 15). The velocities qi, . ., q,, may be given arbi- m
trary values and then the values of the remaining velocities will 2 A p i 6 q i ~ 0( p = l , . . ., g)
i=!
(4')
be determined.
However, we will take the more general path and for the indepen-
dent quantities will take not n (n is the number of degrees of freedom) I n accord with equations (5) we introduce the notation*
of the generalized velocities, but some n independent linear combi-
nations of these velocities*

By assumption, forms (4') and (5') are linearly independent. The-

refore 6n, can assume arbitrary values while the corresponding
6qi will be determined from the set of equations (4') and (5'):
where fSi are functions of t and q,, . . ., q,.
Only one condition need be imposed on the linear forms (5):
these n linear forms together with the g linear forms
The expression for the work of elementary forces in virtual
displacements may be given as
m

must constitute a complete system made up of m = n +

g linearly
independent forms; i.e. the determinant composed of the coefficients
of these m forms must be other than zero. Then the quantities where, a s for a holonomic system,
ns (S = 1 , . . ., n) can take on arbitrary values, since for any values
of these quantities we will find the corresponding qi (i = 1 , . . . , m)
v= i

* I t will be convenient t o denote the linear combinations (5) b y A,, although * In the case of a scleronomic system 6qi=dqi = q i dt and therefore, in
thc symbol n, may be meaningless since the right-hand side of (5) may not be
a total derivative. accordance with formulas (5) and (S'), Gns=zc, dt.
 l 60 Equations of Motion of an Arbitrary System of Particles [Ch. I Sec. 101 Appell's Equations for Nonholonomic Systems 61

l Now, substituting into (7) the expressions (6') in place of 6qi,

as follows
we get
m n n m

n Since 6n, are quite arbitrary multipliers, it follows that

I
6A = 2l IT,Gn,
S=
i
E where
I1 m N
Let us now bring into the discussion the "energy of accelera-
i= 1 i=l v=l
a.
t ions"
We shall call the quantities TT, generalized forces that correspond
N
.. . ..
to the pseudocoordinates n, (S = l , . ., n). . U=2-
1
2 mv~:= U (t, qi, nsc ns) (16)
v= l
i On the other hand, substituting into (2) the expressions (6)
Noting that on the basis of formulas (12)

where e,, and e v (v = l , . . ., N; S = l , . . ., n) are certain vector

we can write equations (15) as follows:
functions of t and qi, . . ., qm.
From the equations (10) we find* au
.
. -ns ( ~ = l ,..., n).
--
8%

The equations (18) were first derived by Appell and are called
and Appell's equations.
.. 2 evsn,+
rv=
..
. . . (v = 1 , . . . , N )
These n = 3N - d-g differential equations, together with the g
(12) constraint equations
S= i
where on
involving (S
..
the right-hand sides of (12) we take only
the terms
pseudoaccelerations n, = l , .. . , n).
By means of the equations (8) and (11) we write down the
general equation of dynamics and the n differential relatiocs

*
. .
The quantities P,, n, and qi are connected by the relations (2), (4) and (5).
form the system of differential equations that determi: e the motion
of a nonholonomic system.
Eliminating bi from these relations we find the formulas (10). The quantities Let us write the Appell equation in expanded form. To do this,
..
6 r , , 6ns and 6qi satisfy the homogeneous relations (2'), (4') and (5'), which differ
from (2), (4) and (5) solely i n the absence of absolute terms. For this reason, put expressions (12) in formula (15) inslead of r,. Then we get
also the formulas (11), which are the result of eliminating 6 q i from (2'), (4')
and (5'), are obtained from (10) by replacing P, by 6rv, n, by 6n, and by re-
jecting the absolute terms e,.
 62 Equations of Motion of an Arbitrary System of Particles [Ch. I Sec. 101 Appell's Equations for Nonholonomic Systems 63

the Appell equations take the form

.===Qs
a7 * (s=1, ..., n) (26)
@S

(p, s = l , ..., n) (22) Note 2. In particular the Appell equations may also be applied
to a holonomic system. In this case, all velocities qi are indepen-
..
The symbol (**) i n equations (21) denotes the terms that do not
contain pseudoaccelerations n, (S= l , . . . , n).
.
dent, Qi= Q$ (i == 1, . . , n!, and, equations (26) are simply another
notation for the Lagrange equations of the second kind.*
I t may be proved that a determinant made up of the coeffici-
ents u p s is not identically equal to zero: Example 1. By means of the Appell equations we determine the motion
of the system described in the example in Sec. 3 (see page 23). This will enable
the reader to compare two methods of finding the motion of a nonholonomic

Then equations (21) may be solved for the pseudoaccelerations

On the other hand, the relations (19) and (20) can also be represented

j
.
in a form solved for qi (i = 1, . ., m) [see formulas (6)l.
I Thus, the motion of a nonholonomic system is determined by
a system of n $- m differential equations of the first order in the
unknown functions g,, . . ., q,, n,, . . ., n,, and these equations
are solved for the derivatives. Then specification of the initial
. .
data, q!, . ., g;, n:, . ., nk uniquely determines the motion of Fig. 28
the system. But with the aid of these initial data, formulas (1)
and (6) are used to specify the arbitrary initial velocities and arbitra- system-by means of Lagrange multipliers and with the aid of Appell's equa-
ry initial position that are compatible with constraints. For this tions-and to be convinced of the advantages of the latter. For the independent
reason, specification of the initial position of the system and the coordinates, we introduce the coordinates X , y of the centre of the rod and t h e
initial velocities that do not contradict the finite and differential angle cp formed by the segment MiMB with the horizontal axis X (Fig. 28).
Then
constraints uniquely determines the motion of a nonholonomic system.
Note 1. If in a special case, n independent generalized velocities,
.
for example, qi, . ., q,, are taken for the pseudovelocities, then
in order t o determine the appropriate generalized forces Q;, . ., Q; .
i t is necessary in (7) t o express 6qn+i, . . ., 69, in terms of 6q1, ., ..
.. ., 6q,: The equation of differential constraint i n the new coordinates becomes
m n
--sinY cp
X
--
coscp
It may readily be verified t h ; ~ tthe energy of accelerations, U, can be ex-
I n this case the energy of accelerations, U, may be given i n t h e
. .. .. pressed in the following m a n ~ ~ o r :
form of the function B (t, q,, . . . , g,, q,, . . ., q,,, qi, . . ., qn) and

* This determinant may be zero a t certain points. These singular points

are not considered here. Inequality ( 2 3 ) is substantiated in a similar manner * However, as applied t o a holonomic system, Appell's equations in pseu-
t o the inequality det (ai,)?, +
0 on pp. 46-47. docoordinates yield quite different forms of the equations of motion.
64 Equations of Motion of an Arbitrary System of Particles [Ch. I Sec. 101 A p p e l l ' s Equations for Nonholonomis Systems 65

(see pp. 35-36)s. We can therefore take p, q, r for the three pseudoveloci-
We introduce the pseudovelocity n, assuming ties. Let us calculate the energy of accelerations:**

then
zu= S w2drn=
S (exr+oxv)2dm=

where the nonwritten terms do not involve accelerations. We now determine

the generalized forces. To do this, write
6A=II6n+@Gcp= -2g6y= -2gsin (p6n
From this
U=-2gsincp, @=O
It will be noted that
dw
8=-=-+W
6w
X W==.
em a
Here, - and -
fi
Now form the Appell equations dt dt dt dt
..
denote, respectively, the differentiation in a stationary system of axes
and in a system of axes in\ariably fixed i n a body. Thercfore p, q and r are
the projections of the angular acceleration E on the axes OE, Oq, 05.
I n t h i s case, these equations do not contain the coordinates X, y and have Then, by analogy with the expression for kinetic encrgy,
the form
..
n=-gsincp,
..
cp=O
Integrating, we obtain ( A , B and C are the moments of inertia about the principal axes of inertia
cp=at-f $ OE, 011, OL) we can write

We find X and y: On the other hand, the kinetic moment G = r X v drn has the con~ponents
J
Ap, Bq, C r . For this reason, we finally obtain the following expression
for 2U:

dcp
1 .
dy
-=-
a
1 . .
Y=-nSlncp=-
a g cos c p ~ i n c p a
a 2 +Lsincp 273 -~ i +a +
B& & +2 [(C- B) & f (A- C ) r p ; +
( B - A) ..
On the other hand, for the elementary work of external forces we have
+.
From this we have
+ +
6A= L o o dt = L E pd t Lqq dt Lgr d t
Thcrelore, Appell's equations yield Euler's equations directly:
y=-
(La +2a2L cos cp) cor (P+E

Substituting a t + $ for cp, we get the final equations of motion that contain
five arbitrary constants, a , $, y , 6 and E:

*We get expressions for p, q, r by projecting termwise on the coordi-

Example 2. We will show how i t is possible, from the Appell equations,
to obtain the dynamical equations of Euler for a rigid body with fixed
point 0.
Let p , q , r be the projections of the angular velocity o on the principal
. . .
axes of inertia 05, Oq, Ob. They are known to be linear combinations of the
generalized velocities $, 8, cp, where $, 8, rp are the Euler angles
nate axes the vector equation o = o g + w o + o Q where m+=$, we=O,
o,=cp.
** Here n e take advantage of the familiar identity
~ ~ ( W X U ) + ~ X ( ~ X T ) + V X ( ~ X ~ ) = @
tlio last term on the lelt is zero since c = o X r. The nonaritten terms
in formula (27) do not contain the angular acceleration E.
 Sec. 111 Lagrange's Equations for Potential Forces 67
CHAPTER 2
Note t h a t when the effective forces E: = X,i -+ -+
Y v j Z,k
(v = 1, . . . , N) acting on the particles have a potential I l (t, X,,
The Equations of Motion in a y,, z,) in Cartesian coordinates X,, y,, z, ( v = l , . . . , N), i.e.
POten tial Field
these forces have a potential even in the independent coordinates
11. Lagrange's Equations for POten tial Forces. q ( i = I . . . , n) (the converse assertion i n the general case is
The Generalized Potential. Nonnatural Systems untrue !), and this potential i s the same potential 11 only expressed
in terms of the coordinates g,, . . . , q, and the time t. Indeed,
Let the generalized forces Qi be potential, i.e. let there exist
a force potential (potential energy) 11 =II (t, qi) (see Sec. 8) and
QI. -
- --an ( i = l , ...,
n)
aqi (1)
l'hen the Lagrange equations whence equations (1) follow.
Let us now consider the case when i n place of the ordinary
potential 11 (t, qk) there exists a generalized potential V (t, qk, i k )
i n terms of which the generalized forces Qi are expressed by
are written i n the form* means of the formulas

where Then the Lagrange equations

L=T-II
The function L is called the Lagrangian function or the kinetic
potential. will again be written in the form (2) where we now have
The kinetic ~ o t e n t i a lL, like the kinetic energy
- - T, is a quadra- L=T-V
tic function of- the generalized velocities: (7)
From formulas (6) i t follows that
where

where (**) denotes the sum of the terms that do not contain -genera-
Here, the coefficients cik, C i , CO are functions of the c0ordinat.e~
..
lized accelerations q, (k = 1 , . . ., n).
g,, . . . , g, and the time t (i,k = l , . . . , n). A comparison of the Inasmuch as in mechanics we consider only the case when the
formula (3) and formula (5) on page 45 yields generalized forces Qi are not explicitly dependent on the general-
L2=T2, L i = T i , L o = T o - n (5) ized accelerations but depend solely on the time, on the coordinates
and on the generalized velocities
* Since the potential energy n doesa Lnot depend
aT
on the generalize& velo-
a~ aT an
cities, and L = T - n , it follows that -=---;- and -= -- -(i =
aqi apt a4i aqi aqi i t then follows according to forn~ulas(8) that all partial second-
=l, ...,n ) . order derivatives of V with respect to the generalized velocities must
5*
 68 The Equations of Motion in a Potential Field [Ch. 2
Sec. 111 Lagrange's Equations for Potential Forces 69
be identically equal to zero, i.e. the generalized potential V depends
linearly op the generalized velocities: vectors E and H are expressed in terms of the scalar potential cp and the
vector potential A by means of the formulas*

where Hi.( i = 1 , . . ., n) and I t are functions of the coordinates We shall find the generalized potential V for the Lorentz force F.
q i , . . ., q, and of the time t . But then, according to (7), L will From the formulas (14) and (15) we have
again be a quadratic function in the velocities qi and i n place of e aA e
F=-egrad cp----+-(VX rot A ) =
equations (5) we will have* c at c
Lz=T2, Li==Ti-Vi, Lo=To-Il (11)
Substituting expression (10) for V into formula (6), we get
= - e grad cp-

where the velocity U i n the expression grad (uA) is considered a vector that
is independent of a field oint**.
n Hence, choosing for t i e independent coordinates Cartesian coordinates
=--+X
an
aqi
an ml,
(A-)
aqh qh
ani
aqi + of the point X , y, z and setting
h= i

The formulas (12) show that when the linear part V, of the
generalized potential does not depend explicitly on the time
(i=l, ..., n ) ] ,the generalized forces Qi are made up
ail
out of potential forces--
aqi
(i= l, . . ., n ) and gyroscopic forces

where * See, for example, L. D. Landau and E. M. Lifshits, Field Theory,

hloscow-Leningrad, 1948, p. 55 (In Russian).
BA aA a A we take
** Here, for the expression (vV) A =v - ax+ v -all
+ vz
The importance of considering t h e generalized potential is evident advantage of the familiar formula of vectbr analy& (CV) A + v X rot A =
from t h e following example. =grad (uA), in which v is regarded as a constant vec.tor.. We are readily
convinced of the validity of this formula by comparin the projections 011
Example. A point electric charge in a n electromagnetic field is acted the X - , y-, and z-axis of the left and right sides of t f e equation. Indeed,
upon by a Lorentz force for the x-axis,

where v is the velocity of the point, e is the charge, c is the speed of light,
and E and H are the intensities of the electric and magnetic fields. The

* The coefficients in the expressions for L and T are interrelated. Indeed,

for the ordinary potential, ci = a i , while for the generalized potential c i =
- ai - lTi (i =1, . . ., n ) . In both cases, cik = aik ( i , k = 1, . . ., n ) , c. =
-
3 a,, - H , and L z = TZ is a positive definite quadratic form. Analogous formulas hold for the projections on the y-axis gnd the z-axis.
70 The Equations of Motion in a Potential Field
Similar formulas are found for F1, and F,. Thus, the generalized poten-
tial of the Lorentz force (14) is determined from the formula (17). For the
Lagrangian function L we have the expression
Classical systems in which the forces have the ordinary potential
Il ( t , qi) or the generalized potential V (t, qi, qi) will be called natu-
ral. For such systems the Lagrangian function L is a quadratic
function of the generalized velocities, i.e. i t is given by the expre-
ssion (4), where L 2 is a positive definite quadratic form in the gene-
ralized velocities.
As an example of a nonnatural system we can examine, i n relativistic
theory, the motion of a particle in the absence of a force field. The motion
of such a particle is determined by the Lagrange equations i n which
. . .
a quadratic function of the velocities s, y, z.
. ..
where u2=x2+y2+zz, and c is the speed of light. Here, L is no longer
i 12. Canonical Equations of Hamilton
If i n the expression for the function L we expand (l-$)T i n a power
series of and reject all terms of second and higher orders i n -fF,i.e. set
l function) L = L (t, qi, qi) is known.
('-2g* 1--- 1 v2 , then we have the "classical" expression for the
2 c2
Lagrangian function for an isolated particle, namely:
L=- '
2
mu2 + const
In this and the following chapters we will confine our discussion
to systems of the general type,* whose motion is determined by
Lagrange's equations (2) with an arbitrary function L = L (t, qi, qi).
We shall only assume that the Hessian of the function L with respect
t o the generalized velocities is not identically equal t o zero;**
* Such propositions as hold only for natural systems will be specially sti-
pulated.
** For natural systems, - a2L - U-
-
..
-aik(i, .,
k=1, . . n) and
aqiaqk aqiaqk
hence, by what was proved i n Sec. 7 (pages 46-47), inequality (19) holds
true.
[Ch. 2 Sec. 121 Canonical Equations of Hamilton 71
In expanded form, the equations (2) may be written as
I
I
..
where (**) denotes the sum of the terms t h a t do not contain the
generalized accelerations qi (i = 1 , . . ., n). Insofar as the deter-
minant of the system of linear (in );; equations (20) is different
from zero [see inequality (19)], it follows that the system (20) may
be solved for the generalized accelerations and written in the form
Therefore the conclusion drawn in Sec. 7 as to the unique determina-
tion of the motion of a system by specification of the initial data
.
49, q! (i = 1 , . ., n) holds not only for natural systems but also
for systems of a more general type that are here under consideration.
Lagrange showed how the differential equations of motion of
a system are written down if the kinetical potential (the Lagrangian
We will call the variables t, qi, ii
(i = 1 , . . ., n), in terms of
which the Lagrangian function is expressed, Lagrangian va~iables.
The system of these variables characterizes the instant of time
and the corresponding state of the system, i.e. the position of the
system and the velocities of its particles. As has already beeii noted
a t the end of the preceding section, specification of the Lagrangian
function and the initial state uniquely determines the motion of the
system.
For the basic variables that characterize the state of a system,
Hamilton proposed the quantities t, qi, p i (i = 1 , . . ., n), where
..
p i (i = 1 , ., n) are the generalized momenta defined by the
equalities
aL
p,--
L- . (i=1, . . ., n) (1)
aqi
We shall call the variables t , q i , p i (i = 1 , . . ., n) Hamiltonian
variables.
Since the Jacobian of the right sides of equations (1) with
respect t o the variables 'qi is the Hessian (different from zero) of the
function L [see the condition (19) on page 701, i t follows that
 72 The Equations of Motion in a Potential Field [Ch. 2 Sec. 121 Canonical Eauations of Hamilton 73
equations (1) may be solved for qi (i = 1 , . . ., n): Example. For a free particle the Cartesian coordinates X, y, z are inde-
endent, and i n the potential field n = n (t, X, y, z) the Lagrangian function
1as the form

Thus, Hamilton's variables may be expressed in terms of the

1 . . .
L = - m (~2+yz+z2)-II (t, X, y, z)
2
Lagrange variables and vice versa, and the state of the system may
To the Cartesian coordinates there correspond the momenta
be described both as a system of values of the Lagrange variables
and as a system of values of Hamiltonian variables.
In the case of a natural system, L is a quadratic function (see
pages 66-68) of the generalized velocities and, according to equa-
tions (l), the generalized momenta are linearly expressed i n terms
. ..
If from this we determine X, y, z and ut the expressions obtained into L,
of the generalized velocities: we will have the associated expression for L:

If i n place of n (t, X, y, z) we have the generalized potential

Solving this set of linear equations for q,*, we again get linear v=n,l+n,l;+n,l+n,
expressions for qi: where n,, n2, U,, and
equation
n are functions of t, X, y, and z, then from the

where bik and bi are functions of t, g,, . . . , 4n. we find

If i n a natural system the forces Qi (i = 1 , . . ., n) have an ordi-
nary potential II (t, qi), i t follows from the equation L = T - II
that**
and the associated expression 2 has the form
'T
p 2. --- .
'qi
(5)

I t will be noted that any function of Lagrangian variables Hamilton introduced the function H (t, qi, pi) defined by the
equation

after substitution of the expressions (2) or (4) into i t in place of

the generalized velocities qi,
is converted into a certain function and demonstrated that with the aid of this function the equations
F (t, qi, p i ) of the Hamiltonian variables. We shall call the of motion may be written in the form of the following system of 2n
function F(t, qi, p i ) the associated expression of the function ordinary differential equations of the first order:

F (t, q i ~qi).
* As was established in Sec. 7, det (aik):, + 0.
** Formulas (5) do not hold for forces that have a generalized potential.
this case [see equations (7) and (10) on pp. 67-68]
These equations are called canonical equations or Hamilton's equa-
tions*. The function H (t, qi, p i ) defined by equation (8) is called
the Hamiltonian function.
1

* These equations were first obtained in general form by W. R. Hamilton

in 1834.
74 T h e Equations of Motion i n a Potential Field [Ch. 2 Sec. 121 Canonical Eouations o f H a m i l t o n 75

The derivation of the canonical equations of Hamilton will rest
on the following mathematical theorem.
Donkin's theorem.* Given a certain function X (X,, ., X,)
the Hessian of which is different from zero:
But, by equations ( l l ) , the two sums on the right-hand side of
this equation cancel and hence, the formulas (12) hold true.
.. .
Now let X contain parameters a i , . ., a, in addition to the
variables X,, . . ., X,. Then these parameters occur in the direct
transformation (11) and, consequently, in the reverse one as well:

Let there also be a transformation of the variables "generated" The function Y is determined by equation (13) in which the xi
by the function X (X,, . . . , X,): . .
are replaced by f i (y,, . . , y,; a , , . . , a,) and so*
m

Then there exists a transformation, the inverse of the transfor-

mation ( I I ) , which likewise generates some function Y (y,, . . . , y,):

Hcre, the generating function Y of the inverse transformation is

connected with the generating function X of the direct transfor- This completes the proof of Donkin's theorem.
mation by the formula*" We utilize Donkin's theorem to make the transition from the
Lagrangian variables to the Hamiltonian variables by replacing
in the theorem the function X by L, the variables X,, . . . , X,
.
by, Q,, . . ., q,, the parameters a,, . . . , a, by q,, . . , q, and t , the
If the function X contains the parameters a,, . . . , a,, i.e.
X = X (xi, . . . , 5,; all . . . , am), then Y also contains these para- . -
variables y,, . .,y, by p,, . . .,p, and finally (taking account of the foot-
71
m

meters, i.e. Y = Y (y,, . . . , yn; ai, . . . , a,), and

h

note 2 on page 74), the function Y = xiyi - X by H =

piqi-L.
i=i i= l
Then, by Donkin's theorem [the Hessian of the function L with
Proof. The Hessian of the function X coincides with the Jaco-
respect to the qi (i- l , ..
., n) is not zero !] i t follows from for-
mulas (1) that
bian of the right-hand sides of equations (11). For this reason,
the condition (10) shows that using equations (11) it is possible
to express the variables X,, . . . , X, in terms of yi, . . . , y,:

Let the function Y (yi, . . . , y,) be-defined by the formula (13),

in which the variables xi are replaced by the expressions (15). The equatious (16) and (17) are identities resulting from the
Then relation (1) between generalized velocities and generalized momenta.
But the Lagrange equatioqs may, by virtue of ( l ) , be written as
ay a 8% a x ark follows:
a ~ i ayi
h= i
xh~k-x) = 2 %yk+
h= l
xi- 2 --
k=
8% a y
i
i

* Philosoph. Trans., 1854. The transition from the variables xi to the

variables yi (i = l, . . ., n ) which is dealt with in Donkin's theorem, is fre- ay
quently called a Lcgendre transformation.
** It is assumed that on the left-hand side of (13) all the xi are expressed in aaj the quantities yi, ..., yn are regarded
* In computing the derivative -
as constants.
terms of yi, that is that Y = Y ( y l , . . ., y,).
 76 The Eouations o f Motion in a Potential Field [Ch. 2 Sec. 121 Canonical Equations of Hamilton 77

Therefore, for a n arbitrary natural system we finally get

I t is these equations together with the equations (16) t h a t lead
us to the canonical equations of Hamilton: H=LP-L.
- (23)
+ +
L e t T = T2 T, To. If t h e forces have the ordinary potential
+
II = IS (t, qi) or the generalized potential V = V, D, then L 2 = T2,
- In addition to the Hamilton equations we obtained the iden-
t i t y (17) which will be utilized in the sequel:
L. = To-- II, and therefore
-
From equations (19) there follows the identity
n
H=Tz-To+n
-
If t h e system is scleronomic, then T T2, T o = 0, and so
(24)

(20)
i= i Thus, for a scleronomic natural system the Hamiltonian function
We shall call a system generalized-conservative if the Hamilto- H is the total energy* expressed in terms of Hamiltonian variables.
nian function H does not explicitly depend on t, i.e. if Now let us consider a conservative system, i.e. a natural sclerono-
mic holonomic system with an ordinary force potential that is
H =H (qi pi) 1
not dependent explicitly on the time.
aH
In this case = 0* and, consequently, by virtue of identity (20). In this case the time t does not enter m t o expression (25) explicitly
dH for the total energy H and so, according to (21), we have the law
-=O,
,
lt
W"
i.e. when the system is in motion of conservation of energy
H (qi, pi) = const = h (21) T+II=h (26)
where h is an arbitrary constant. We shall call the function H A conservative system is a special case of a generalized-conser-
the generalized total energy, and the relation (21) which does not vative system, and in this special case the generalized integral
contain q or p and includes an arbitrary constant h, the genera- of energy becomes a n ordinary integral.
lized integral of energy. If the system is scleronomic and the forces have a generalized
We will explain this terminology by examining a natural system. +
potential V = V, ll in which II does not depend explicitly on the
an = 0 ) , then the function H is again determined by the
Then L is a quadratic function [see equation (4), Sec. 111 time t
L=Lz+L,+Lo fo;mula' (25) and does not depend explicitly on t . For this reason,
and in this case too the system is a generalized-conservative system,
n n and the energy integral (26) holds.*
Example. A horizontal rack is in rotation about a vertical axis, and a
weight of mass m is in motion along the rack. The force acting on the weight
has the potential II (r), where r is the distance of the weight from the axis of
rotation.
Denote by cp the angle of rotation of the rack and by Z = md2 its moment
of inertia about the vertical axis of rotation. Then r and cp are independent
But by Euler's theorem on homogeneous functions,** coordinates of the system and
n I
I
T = - -1Z q -2f -m (r2f r2cp2) =-1 m [r2f
.
(rzf dz) rp2]
2 2 2
From this
pr =-
ar a~
= rnr, p v-- - = m (r2 -1- d2) cp
aL a; @
* From (17) i t also follows that x=O
- for a generalized-conservative
system, i.e. t does not appear explicitly in the Lagrangian function L either. * Given a generalized potential V = Vi f l l , we continue to call the quan-
** See footnote on page 49. t i t y T f II (and not T f V) the total energy.
78 The Equations of Motion i n a Potential Field [Ch. 2 Sec. 131 Routh's Equations 79

Since the system is conservative, it follows -that Then, applying the Donkin theorem t h a t was proved in the pre-
ceding section, we get a transformation that is inverse to t h e
1
H = T + n = - 2m (P:+ L)
+n
r2+d2 (r) transformation (2), namely
The canonical equations in the given case have the form
.
q --
aR
( a = m + l , ..., n), (4)
- a ~ a

where R = R (t, qi, qa, qi, pa) is the Routh function defined by
the equation
m -
and the energy integral is written as

p:+- +2mII (r)= h i (hi= h h ) the signn signifying here t h a t all the 4,
are expressed in terms
of Pa.
The variables t, qi, qv., qi (i = l , . . . , m; a = m f l , . . ., n)
13. Routh's Equations are now regarded as parameters and for this reason, by virtue of
For the basic variables characterizing t h e state of a system a t the same Donkin theorem,
a time t, Routh proposed taking part of the Lagrangian variables
and part of the Hamiltonian variables. The Routh variables are
aR _
--
@i
aL
aqi '
.
_ aR - aL (i = 1, . . ., m),
aqi
the quantities
aR
= - p a L ( a = m + l , .. ., n),
l , qi, qa, qi, Pa ( i = 1 , . . ., m; a = m + I , .. ., n) (1) %a @U
where m is an arbitrary fixed number less than R .
In order to pass from the Lagrangian variables to these variab-
les, we have to express all the qa in terms of p, ( a = m + 1 , . , R), .. Lagrange's equations of t h e coordinates qi may be rewritten, in
utilizing the relations accord with the equations (6), a s

Suppose t h a t the Hessian of the function L of the generalized Lagrange's equations of t h e coordinates qa
velocities (the "small Hessian") is different from zero:*
dpa aL ( a = m H , .. ., n)
-
dt @a
together with the formulas (4) aud (7) yield
* In the general case this inequality does not follow from the inequa-
lity (19) on page 70 but is an auxiliary condition. Now for a natural sys-

tem, the inequality (3) follows from the fact that L z = y

I
2
n
..
arSq,qs is a
r, s=i Equations (9) and (10)
positive definite quadratic form. I n this case, the principal minor compo-
sed of coefficients of the quadratic form:

must be positive.
80 The Equations of Motion i n a Potential Field [Ch. 2
Sec. 141 Cyclic Coordinates 81
form a set of Routh equations. They consist of m second-order
differential equations of the Lagrange type and 2 (n - m) first- From the structure (2) of the function H it follows that equations
order differential equations of the Hamiltonian type; and the (3) form a system of 2m first-order differential equations in 2m
.Routh functions in the first equations play the role of the Lagrangian .
unknown functions qi, p i (i = 1 , . ., m). Integrating this system
function, while those in the latter equations play the role of the we find
Hamiltonian function. * qi = qi ( t , cat ci, C ; ) ,
14. Cyclic Coordinates ( i = l , . . . , m), (4)
The coordinate g, is called cyclic (ignorable) if it does not enter Pi=$i (2, car ci, C;)
aL = 0.
t h e Lagrangian function L explicitly, i.e. if --- .
where ci, c; (i = l , . . , m) are 2m new arbitrary constants. !3y
ao, -- substituting expressions (4) into H we can determine the q, from
When deriving the Hamilton and the Routh equations the follo-
wing equalities were established: the equations

aL , by quadratures
From these equations i t follows that the partial derivatives ----
a9,
- 8R
aH and - can only vanish simultaneously. For this reason,
@a %a
the cyclic coordinate may also be defi.i.4 as a coordinate that Thus, essentially the whole question reduced to integrating the
does not appear in the Hamiltonian function H or the Routh system (3) whose order, 2m, is less than the order of the initial
function R explicitly. All these definitions are equivalent. system, 2n, by 2r units, where r = n - m is the number of cyclic
The generalized momentum that corresponds to the cyclic coor- coordinates, i.e. the presence of r cyclic coordinates made i t possible
dinate g, maintains a constant value during motion. Indeed, from to lower the order of the system by 2r units.
the canonical equations i t follows t h a t The system of equations (3) is Hamiltonian. We shall show that
with the aid of the. Routh equations i t is possible to obtain a n auto-
dpa
-=--=O aB nomous* system of m second-order differential equations of the
tit @a Lagrangian type. Indeed, by replacing the generalized momenta
i.e. p, =const = c,. p,, whjch correspond t o the cyclic coordinates qa, by the constants
Now suppose that we have r = n - m cyclic coordinates q,
(a = m + 1 , . . ., n). The cyclic coordinates q, do not enter into
.
c, (a .= m -11,- . .: n), we can write the Routh function as a function
H explicitly while the momenta p, corresponding to these coordinates .
of t, q;, qi and ca (i = 1 , . ., m ; a = m + .
4 , . ., n):
can be replaced by the constants c,. Then** R = R ( t , qir qi, ca) (7)
H = H ( t , qir P i , ca) (2) Then t h e Routh equations for non-cyclic coordinates
Let us write down the canonical equations for non-cyclic coor-
dinates:

p-
-. form the desired autonomous system, while the cyclic coordinates
*
If we wish to retain the connection betweell the generalized mornenta q, are determined from the appropriate Routh equations (11) of
and the Lagrangian function, we must take i t that in the first equations the preceding section by means of quadratures:
-,
since, according to ( G ) , pi =
%i
aL
-
of (11) the role of the Lagrangian function is played by the function -R,
a(-R)
aqi
.
---;---- ( i = i , . ., m).

** The index i runs through the values 1. . . ... * Here, by an autonomous system is meant a system of differential equations
through m + .
l , . ., n .
m., while the index a runs
that contains no "superfluous" unknown functions which should have been de-
termined prior t o integration of the given system of equations.
 82 The Equations of Motion in a Potential Field [Ch. 2
Here, first all the qi and ii in *
dca
are replaced by the functions of
+ ..
2m 1 arguments t, ci and ci' (i = 1, ., m), which functions are
obtained by integrating the system (8).
Example. In the example considered at the end of Sec. 12,
cp is a cyclic coordinate, and
pQ =m (ra+dz) cp= const = cQ
We form the Routh function
l
Determining the motion reduces to integrating one second-order diffe-
rential equation
which in expanded form reads
It will be noted that this is an equation of the relative motion of a
weight along a rack, since the right-hand side involves the centrifugal
force ,
c2cP
- -m r ~ 2 (CV = PQ)
m (r2+ d2)2-
Cyclic coordinates are sometimes called ignorable or kinosthenic
coordinates. This name stems from the fact that in integrating
the system of eqnations (3) or (8), we ignore the existence of the
cyclic coordinates, regarding the p, as constant parameters.
In the example we have examined, ignoring the cyclic coordinate
led to ignoring the rotational motion of the rack and we obtained
a differential equation for the relative motion along the rack.
The actual name "cyclic coordinate" is associated with the fact
' that in many problems in mechanics the angle cp which describes
motion in closed trajectories (cycles) does not enter the expression
for L explicitly and for this reason is a cyclic coordinate.
Sec. 151 The Poisson Bracket 83
We note a certain analogy between a holonomic system having
a cyclic coordinate and a generalized-conservative system. FOP
aH aH
the former system, - = 0, for the latter, - = 0. For the former,
dqa at
we have the integral p, = const, for the latter, H = const. The
roots of this analogy will be revealed later on when we consider
the principal integral invariant of mechanics.
A more profound investigation of the motion of systems with
cyclic coordinates will be carried out in Chapter 7.
15. The Poisson Bracket
In this section we consider some properties of i n t e g ~ a l sof a Hamil-
tonian system of equations of motion.
A certain function f (t, qi, pi) is called the integral of the equa-
tions of motion
if for any motion of the given system this function retains a con-
stant value c:*
f (t, qi, pi) c (2)
Sometimes the relation (2) itself is called the integral.
For a generalized-conservative system the function H (qi, pi) is
the integral. If the q , are cyclic coordinates, then p, will be the
integral.
Obviously, if the functions f,, ..
., f l are integrals of the equa-
tions of motion, then an arbitrary function of these integrals
will also be an integral. Henceforward we shall therefore he interested
only in independent integrals.
If the "complete" system of integrals consisting of 2n independent
integrals f i , . . ., fzn (n is the number of degrees of freedom of
the system) is known, then, solving the relations
for qi and pi, we get the final equations of motion
which contain 2n arbitrary constants cl, . . ., ten.
* This value of c may change when we pass from one motion of the system
to another.
 Sec. 151 The Poisson Bracket 85
T h e Equations of Motion i n a Potential Field [Ch. 2
.
where X k , Yk (k = l , . . , m) are functions of the variables
Thus, if 2n independent integrals are known, then all the motions X,, . . . , X,. Then the "commutator" Z = X Y - YX will also be a
of the system are known. If we know L independent integrals first-order operator:* m

.
fi, . ., f l , where 1 < 2n, then we have only a partial idea of the
motions of the system, and the greater 1 is, the more exhaustive
our conception is. We are therefore always interested in finding
the greatest possible number of independent integrals.
We shall familiarize ourselves here with the method of finding the Let us return to the Poisson bracket (cpf). I t may be regarded
integrals of the equations of motion proposed by Poisson and Jacobi. as t h e result of applying a linear operator CD of the form (8) to
Let f ( t , qi, p i ) be the integral of the equations (1). Then substitut- the function f of the variables qi, pi (i- 1, . . . , n):
ing any solution of the Hamiltonian system (1) in place of q i , p i
.
(i = 1, . ., n), the function f is converted into a constant c, i.e.
in accord with equations (l)

This first-order differential operator is fully determined by the

Poisson introduced a special term, the Poisson bracket, for the
following expression composed of the partial derivatives of two
arbitrary functions cp (t, qi: pi) and 4 (t, qi, pi):
n
function cp. Analogous operators Y 7 X are determined by the functi-
ons and X.
Let us now undertake directly to establish the Poisson identity 4".
After removing the complex brackets, any term on the left-hand
side of 4" will contain as a factor a second-order partial derivative
01 one of the functions cp, g, X. But ((9%) X) does not contain
second-order partial derivatives of X , and the sum

By means of the Poisson bracket, equation (5) -the necessary

and sufficient condition for the function f (t, qi, pi) t o be the
integral of the equations (l) -is written as follows:
We note the following properties of the Poisson bracket:
is a first-order differential operator relative to X. Thus, the left-
hand side of the Poisson identity does not involve second-order
partial derivatives of X, and hence, by virtue of symmetry, there
are no second-order partial derivatives of cp and $. In other words,
all the terms on the left-hand side of 4" cancel, which is what we
soughs t o prove.

-
For any functions 9 ( t , qi, pi), (t, qi, pi), X (t, qi, pi):
1". (cp*) - ($9);
Let us now prove the basic theorem.
Jacobi-Poisson Theorem. I f f and g are integrals of equations

+ +-
2". (CV*) = c (qg)(c is a constant); of motion, then (fg) is also a n integral of these equations.
3". ((P *X) (W) (4%);++
4". ((949 X) ((*X) 9 ) ( ( ~ 99)) = Q;
Proo). I t is required to prove t h a t for the function (fg) the
relatioil (7) holds:
5O.
a
(94=
acp
(- *) + (P Z)
The identities l", Y, 3", 5" are obtained directly from the defini-
tion (6) of the Poisson bracket. when the same relation holds for each of the functions f , g:
Identity 4', which is called Poisson's identity, is established by
means of special reasoning. Let X and Y be differential operators,
of the first order on the function f (X,, ..
., X,):
* Direct calculations show t h a t on the right-hand side of (9) all terms in-
volving second-order partial derivatives of the function f cancel.
86 The Equations of Motion i n a Potential Field [Ch. 2 87
Sec. 151 The Poisson Bracket

Indeed, by 5" L. ( L x , L,, L,) any of these integrals occurs if the corresponding one of
the quantities X , Y, Z, L,, Ly, L , is equal to zero.
Form the Poisson bracket for the quantities relating to a single particle:

Therefore, using the Poisson identity, we get

(m m )-----p-am, am, am, --spy-yp+=m,
Y - az ap, ap, az

thus completing the proof. I t will be noted that the Poisson bracket, in which one quantity ( p or m )
refers to one particle, and the other to another particle, is allVays zero;
The theorem just proved yields a n automatic rule that permits for this reason,
obtaining a third integral from the two integrals f It, qi, pi),
g It, q i , p i ) by means of algebraic operations and differentiation: (PXP,) = 2 (PXP,) = 09

2 =P,,
( p x M , ) = r) ( p X m , )= PZ (14')

(M,My)= 2 ( m x m y )= 2 m, =M Z
Taking the Poisson bracket of, for example, f and If g), we again Cyclic permutation of the letters s, y, z yields similar relations:
get an integral, and so forth. However, i t should not be forgotten
that the new integral may prove t o be either identically zero or
a function of earlier known integrals. Thus, only by special choice
.
of the independent integrals f,, . ., f l ( I < 2n) can we be sure
that i t is possible to obtain, by means of the Poisson bracket, the
.
integrals f l+,, . ., f,, that are lacking when we want a complete
(P,P,) =0 ,
( P y M z )= P , ,
( M yM,) = M x ,
(PZPX)= 0 ,
(PzMx!= P y ,
( M z M x )= M y 1
The six laws of conservation (13) are not independent. It follows from the
relations (14') and (14") that if we have the integrals
(14")

P , = cir M , = C&,M y = c5 (15)

system.
then we also have the integrals
By way of illustration* let us consider the integrals of momenta and (16)
angular momenta for a system of free particles that is isolated from exter- P y = c 2 , P z = c 3 , &=c6
nal actions:** This is all true, of course, of a potential field of force. In a nonpoten-
Px= 2 P,, P, = 2 Py, P,= 2 P*, tial force field,
(17)
X = O , L x = 0 , L,=O
M,= 2 m, = 2 ( Y P , - ~ P , ) , M y = 2 m y = r ) (~Px-x~zp,),
do not yield the equations
M , = 2 m, = 2 ( x P , - Y P ~ ) , Y = O , Z=O, L,=O*
W here
* The conservation integrals (15) hold only if equations (17) are satisfied.
px=mx, py=rny, pz=mz Similarly, integrals (16) hold only if we have the equations ( 1 8 ) .
The functions P,, P y , P , , M,, M y , M , are integrals, i.e. the "conser-
vation integrals"

hold if the system is isolated. However, in the presence of an external

force field with principal vector R (X, Y , Z) and principal moment
* See Landau, L., Pyatigorsky, L. Mechanics, Moscow-Leningrad, 1940, p. 151
(in Russian).
** In this example the summation is taken over all the particles of the
system.
 Sec. 161 Hamilton's Principle 89
CHAPTER 3
If the system is natural and constrained, then the motions discus-
sed here are subjected to only one restriction: during motion of
the system the constraints imposed on particles of the system must
Variational Principles and Int,egral not he violated. This condition is fulfilled automatically when we
specify the motion i n independent coordinates, assuming qi = qi (i)
Invariants .
(i = 1 , . ., n).
Suppose that among the paths considered there is a so-called
"straight" path, i.e. one along which the system can move for a
16. Hamilton's Principle specified function L (i.e. in a given field of force). For a straight
.
path the functions q, = q, (t) (i = 1 , . ., n) satisfy the Lagrange
We consider an arbitrary holonomic system with independent equations
.
coordinates q,, . ., q, and the Lagrangian function L (t, qi, 4,).
The integral

W=
1
to
Ldt All the other paths passing through the points M O and M, will
be called "circuitous' paths. (In Fig. 29, the straight-line path
is depicted by a solid line, the circuitous paths by dashed lines.)
is called the action (in Hamilton's sense) during a time interval We shall prove that the action W has an extremal (more precisely,
(to, ti) and the expression Ldt is the elementary action*. stationary) value for the straight path as compared with the circui-
Since the function L is of the form L = L ( t , qi, q,), i t is neces- tous paths. Therein lies Hamilton's principle.*
sary, in order to compute the action ( l ) , to specify the functions Let us consider an arbitrary one-parameter family of paths
qi = qi (t) (i = 1 , . . ., n) in the time
,. M, interval t o,< t G ti. In other words, the

/
action W is a functional dependent on
,

@
,'// the motion of the system.
////I If we arbitrarily specify the functions containing, a t a = O , a given straight path; for a # 0 the paths
p ,'/',l
.
qi = 4; (t) (i= 1 , . ., n), then we get are circuitous. Let all these paths have a common origin M O and
a common terminal point M,:
a certain kinematically possible motion
MO (that is, motion admissible by the const-
raints). In the extended (n + l)-dimen-
g2 sional coordinate space, where the quan-
tities qi and the time t are the coordi-
nates, this motion is depicted by some The action W as computed along a path belonging to this family
curve. We shall consider all possible such is a function of the parameter a.
curves, or "paths", passing through two
Fig. 29 specified points of space M O (to, qq) and
M, (t,, q:) (see Fig. 29 for n = 2), i.e. all possible motions t h a t
translate the system from a given initial position (q",, which i t
occupied a t time to, to a given final position (q:), which i t occupies
a t time t,. Here, from the start we fix the initial and terminal * This principle is given in the works of W. Hamilton published in 1834-
instants of time t o and t,, and the initial and terminal positions 1835. Hamilton proceeded from the assumption that the initial system was
scleronomic (he proceeded from the concept of kinetic energy T as a quadratic
of the system. The motions are otherwise arbitrary. form of the generalized vel'ocities). This principle was formulated and substan-
tiated for the general case of nonstationary constraints by M. V. Ostrogradsky
* For natural systems, L = T - II has the dimensions of energy. There- in 1848. For this reason, the principle is sometimes called the Hamilton-Ostro-
[ore the dimensions of action are energy X time = force X length X time. gradsky principle.
90 Variational P r i n c i ~ l e s and Inteeral Tnvariants [Ch. 3
We compute the variation of the action W, i.e. the differential
with respect to a:
tr tr n equations of motion in the Lagrangian form and also by means
Here, we transformed the integral by means of integration by
parts, making use of the commutability of the operation of varia-
tion 6 and the operation of differentiation with respect to the
time -
d
:
dt
a
6qi=6Tqi(t,
a a
a)=--qi(t,
aa d t a)6a=
The straight path and all the circuitons paths pass through a fixed
initial point and a fixed terminal point in extended coordinate
space. Therefore, a t t = to and a t t = t, the variations 6qi = 0
and the integrated part vanishes.
From equation (3) i t is seen that for the straight path, i.e. for
a = 0, the 'expression under the sign of the transformed integral
is zero by virtue of Lagrange equations. And so
for the straight path 6W = 0 (4)
This is the mathematical expression of Hamilton's principle.
The converse proposition is valid as well: if for some path bW = 0,
then the path is straight. Indeed, because of the arbitrari ature
of the variations 6qi (i = 1 , . . ., n) (at the ends they are zero,
while a t the remaining points of the path they are quite arbitrary)
there follow from the condition 6 W = 0 , by virtue of the equation
(3), the equations (2), i.e. the Lagrange equations for a straight path.
Since from the Hamilton principle there follow the Lagrange
equations in independent coordinates (and vice versa), Hamilton's
principle may be placed a t the foundation of the dynamics of
holonomic systems.*
* The statement of the Hamilton principle given above presumes (in the
case of a natural system) the existence of a potential of force. The more general
formulation of the principle, which embraces the case of nonpotential forces
as well, will be given below [formula (9) on page 931.
Sec. 16) Hamilton's Principle 91
The straight-line paths, i.e. the "true" motions for a given func-
tion L , may be characterized both by means of the differential
of the variational principle of Hamilton. However, there is one
fundamental difference between the differential equations of motion
and the variational principles.
The differential equations of motion express a certain relation
between the instant of time t, the position of the system, and the
velocities and accelerations of its points a t that time. If this rela-
tion holds at each point of some path, then the path is a straight
line. The variational principle characterizes the entire straight
path as a whole. I t formulates the extremal (stationary) property
of a certain functional, which property distinguishes the straight-
line path from among other kinematically possible paths. The
variational principle has a more surveyable and compact form and
is frequently used as a foundation for new (nonclassical) domains
of mechanics.
Note. The differential equations (2) are necessary and sufficient
conditions for the first var7ation 6W, where the integral W has the
form ( l ) , to be equal to zero. In the calculus of variations the equa-
tions (2) are called the differential equations of Euler for the varia-
tional problem
tl
Lagrange's equations in independent coordinates were utilized
to substantiate Hamilton's principle. But the equations themselves,
in the case of a natural system, were obtained from the general
equation of dynamics
We shall show how Hamilton's principle can be substantiated directly
by means of the general equation of dynamics (5). Then Lagrange's
equations are obtainable directly from Hamilton's principle.
If in the expressions for r,
~ " , = r , ( t * qf) (v==1, ..., N)
we put in place of qi the functions qi (t, a), then r, will become
a complex function of t and a. We differentiate it with respect
to a, i.e. we compute the variations
n
ar,
6rV=xK6qt (v=l* .... N )
i-1
 92 Variational Principles and Intefral Invariants [Ch. 3
Sec. 161 Hamilton's Principle 93
These formulas coincide with the formulas (8) on page 37, which
determine the virtual displacements of points of a holonomic system. As before, denoting by 6A the elementary work of effective for-
Thus, the variations of the radii vectors for a n y t are virtual ces Fvi n virtual displacements 6 r v ( v = 1, . . . , N), we can, using
displacements of the points of the system. the transformation (7), write down the equation (5) i n the form
It is possible to convince oneself of this without resorting t o N
formula (6) but by proceeding solely from the definition of a varia-
ST+~A--$ rnv;v~~*v=~
tion and of virtual displacement. Indeed, the variation 6 r v = al., 6a
is an infinitesimal displacement of a point of the system P, which
U - v= l
transfers the point of the trajectory produced for a certain fixed Integrate both sides of this equation with respect to t from
value of a (for a straight-line path a = 0), t o the point of an adjacent t = t o to t = ti:

Here [ 1:~:: denotes the difference between the values (for t = t i

and t = to) of the expression in the square brackets.. But when
t = to and t = ti, the radius vector r, does not vary since t h e ini-
tial position and terminal position of the system are fixed:
Fig. 30. r v ( t O ,a ) = & ..
r v ( t , , a ) =rl, (v = 1, ., N )
trajectory which corresponds to the parameter a +6 a (Fig. 30). Therefore 6 r v = 0 for t = to and for t = ti. The second term in (8)
is zero, and this equation takes the form
Here, both point,s are taken for one and the same instant of time t,
since in differentiation with respect t o a the value of t is fixed.
Consequently, 6 r v accomplishes the transition from one possible f(aT+aA)dt =o (9)
position of the point P, a t time t t o another possible position for '0
the same time t, i.e. 6 r v is a virtual displacement of a point P, of
the system. We consider the case when t h e forces have t h e potential 11=
Thus, in the general equation of dynamics (5) we can consider =11 (t, qi)- Then
6 r v to be a variation of the radius vector r,. But then i t is possible 6A = -6II
d where 6II i s the virtual differential (variation) of the function
to change the sequence of the operation and the operation 6
II(t, qi)* and equation (9) is written
of differentiation with respect to a :
a a a
x 6 r v =xaarv(t, a)6u=-rv(t, a)6a=6rv
aa
Therefore 10
whence

Thus, the basic equation of dynamics (5) has led us to Hamilton's

principle 6W = 0, and from this, as was indicated above, we imme-
where 6T is a variation of the kinetic energy T = - I
2
2 mvg:t.
v= l
94 Variational Principles and Integral Invariants [Ch. 3 Sec. 161 Hamilton's Principle 95

diately obtain Lagrange's equations: Thus, the straight-line path amounts t o uniform motion along
an arc of a great circle. Here,
tr

In order t o obtain Lagrange's equations, i n the case of the

S ( ~ - * ) d t $ ~5 ( ~ - ~ , ) ~ d t a fuc'(, u - u ~ ) ~ ~ = ~ ~ ( o ~ ~ ~ - o
t1
nonpotential forces Qi one should proceed from the equation (9) 1
11

=vo
[in place of (4)]. Applying formula (3) to the integral 5 ST dt,
to
10 to to
The length of the arc of a great circle cfSt is less than the length o,i,
i.e. replacing the function L by the function T ( t , qi, qi), and of any other curve on the sphere connecting the same two points M,
using the expression for the elementary work of the effective forces and Mi. For this reason,
n
Wst <Wcir
However, this is valid only when ost = U M OM, < nR. If ust >n R ,
then WSt will no longer be always less than Weir and the least value

.
From this i t follows that since the quantities 8qi (i = 1 , . ., n)
are arbitrary, the expressions in the brackets under the integral
sign (10) must be zero, i.e the Lagrange equations must hold for
the straight-line path:

Let %S find out whether the value of the action for a straight-
line path is the least when compared with circuitous paths. MO
By way of illustration let us consider the motion of a particle Fig. 31
constrained to move on a sphere in the absence of a field of folce
(inertial motion on a sphere). Let the mass of the particle m 1 2. of the action W will be attained on an auxiliary arc of a great circle,
In spherical coordinates (Fig. 31) which in this case will be the shortest distance between M Oand M,.
If we move the point M i along an arc of a great circle, increasing
the arc, then the critical point M,* (prior t o this point, W,, will
be a mininium, and after M l passes through M,* i t will no longer
For the straight-line part be such) is a point diametrically opposite the point M,.
aL The situation is similar in the general case. I t may be proved that
-0 and = const
dt a'P al) if the point M i is chosen sufficiently close t o M,, then one straight-
(where 11, is a cyclic coordinate), i.e. line path passes through M, and Mi. But if the separation between
M i and M , is sufficient, then two straight-line paths can pass
through M O and M1 or even a pencil of straight-line paths.. Such
a position M,$ of the point M l is called the conjugate kinetic focus
Without loss of generality, we may say that the initial velocity of MO. I t has been established that an action directed along the
v,,for a straight path is directed along a meridian ($=const), straight-line path M,Mi is of least value compared with those
i.e. qo=O; then along circuitous pathways if on the arc MoMl there is no kinetic
focus M,* conjugate t o M,.
$ = 0, rp = const and v2 = R2'pa= const

96 Variational Principles and I n t e ~ r a l Invariants
17. Second Form of Hamilton's Principle
Let us examine yet another form of Hamilton's variational prin-
ciple. In place of an (n +l)-dimensional extended coordinate
space we consider a (2n +
l)-dimensional extended phase space
in which the quantities qi, p i (i = 1 , . . ., n) and t are the coordi-
nates of a point. In this space we consider a straight-line path pas-
sing through two points B . (g!, p:, t,) and Bi (g:, pli, ti) and also
Sec. 171 Second Form of Hamilton's Princinle 97
[Ch. 3
With the aid of this function, Hamilton's canonical equations (1)
can be written, as is readily seen, in the Lagrangian form:
Since a straight-line path is characteri.zed by equations (3) of the
Lagrangian type, the straight-line path in extended phase space
stands out among the circuitous pathways, as was earlier established,
by the fact that for i t the integral

has the stationary value

10 l0 i=i

At first glance it might appear that the second form (5) of Hamil-
ton's principle does not differ in any way from the first, 6W = 0
Fig. 32 since, according to formula (8) on page 73, the expression for L*
coincides with the function L. This is not always so, however; it
is true only for motions of a system, i.e. for paths such that qi =
all other possible curves connecting these points (circuitous pathways;
see Fig. 32, n = l ) .
= qi (t), p i = p i (t) (i = l , ..
. n), for which the functions
qi (t) and pi (t) are connected by the relations
.
The functions qi (2) and p i (t) (i = 1, . ., n), which specify
the straight-line path, satisfy Hamilton's equations

However, in the second form of Hamilton's principle (in contrast

to the first!), arbitrary curves of (2n + l)-dimensional extended
phase space passing through the points B , and Bl are admitted for
comparison as circuitous pathways. These paths may not satisfy
. .
We introduce a function of 4n+ l independent variables relations (6) and for this reason, in the general case, for them L* +
L* (t, qi, pi, qi, pi) defined by the equation* +L. But if in formula (5) we confine ourselves solely to those
circuitous paths for which (6) holds, then the second form of Hamil-
ton's principle passes into the first form, 6W = 0.
Note also that unlike the points MO and M, in the first form of
Hamilton's principle, points B , and Bi cannot be chosen in arbitrary
* The p'i actually are not involved in the right-hand side of (2). There- fashion because i t is impossible in the general case to draw a straight-
fore, the function L* is, in this case, independent of these quantities, line path through two arbitrary points of extended phase space.
aL*
and --_=O ( i = l , ..., n).
The points B . and Bi are chosen on the straight-line path for which
a ~ i Hamilton's principle is formulated.
98 Variational Principles and Integral Invariants ICh. 3 Sec. 181 Basic Integral Invariant of Mechanics 99

' 18. The Basic Integral Invariant of Mechanics Substituting the expressions (4) and. (5) for [6qilt,tl and [6qilt=to
(Poincar6-Cartan Integral Invariant) into expression (2) for 6 W , expressing as usual qi i n terms of pi,
Let us derive a formula for the variation of action 6 W in the and noting t h a t
general case when the initial and terminal instants of time, just n -. -
like the initial and terminal coordinates, are not fixed but are
functions of a parameter a:
2 piqi-L
i=1
=H

we obtain the following formula for the variation of action 6 W

i n the general case:
tl
Here, differentiating the integral W = 1 L dt with respect to t h e
to
parameter a and integrating by parts we find

6W =6 '$
to(@
L dt = L& -Lo6to+
where
21 n

In the special case when for any a the appropriate path

.
straight line, i.e. when qi = qi (1, a ) (i = l , . ., n) is a family
of straight-line paths, the integral on the right of (6) is equal to
zero for any a , and the formula for the variation of the action takes
on the following simple form:

On the other hand, for the variations of the terminal coordina- In place of the (n + 2)-dimensional extended coordinate space
tes 42 = q1 [ t , (a), a ] we have the formulas we take the (2n + l)-dimensional extended phase.space in which
.
the quantities q i , p i (i = 1, . ., n) and t will be the coordinates
of the point. In this space we take a n arbitrary closed curve C O
given by the equations
or
6qf = [6qiJl,tl +q:6ti (i = l, .. ., n)
From this
[6qiIt+ =6qi-q:6ti (i = l , . . ., n) (4)
Here, we have one and the same point of the curve CO for a = 0
and a = I . Taking each point on the curve C O as the initial one
I n similar fashion we draw the appropriate straight-line path. Such a path is uniquely
determined (after specifying the initial point) by a system of Hamil-
[6qi]t=to= 694 -q46to (i = l , . . . , n) (5) ton's canonical equations. We obtain a closed tube of straight
100 Variational Principles and Intearal Invariants ICh. 3 Sec. 181 Basic Inte~raE Invariant of Mechanics 101

trajectories (see Fig. 33, n = 1) Integrating t h i s equation termwise with respect to a from a = 0
to a = l , we get
l n
where
Qi (t, 0) qi (tr l), Pi (t, 0) Pi ( t , E)
( i = l , . . ., n)
On this tube we arbitrarily choose a second closed curve C,
around the tube that has only one common point with each gene-

I t i s thus established t h a t the line integral

i= i

Fig. 33
ratrix. The equations of cnrve C, may be wyitten i n the form*
Let us examine the action W along the generatrix of the tube
from the curve CO to the curve C,:
taken along a n arbitrary closed contour does not change its value
in the case of a n arbitrary displacement (with deformation) of the
contour along a tube of straight-line trajectories, i.e. i t is an integral
invariant. W e shall call the integral I the P o i n c a r X a r t a n integral
invariant.
We now prove the converse proposition. Let i t be known that
the straight-line paths are defined by the following system of first-
order differential equations:

W=''rLdt
to(@)
Then for any a, by formula (7), This supposition is natural in t h a t the motion of a system must
n be defined uniquely on the basis of the initially given q!, p: (i = 1, . . .
6W - W f (a) 6 a = [ 2 pi6qi-Hat]:
i= 1
. . ., n). Besides, let i t be given that the Poincarh-Cartan integral
(12) is a n integral invariant with respect to the straight paths
defined by the set of equations (13), i.e. for any tube of these paths ,
* To each value of a of the intereal 0 < a ,< l there corresponds a definite the PoincarB-Cartan integral computed round the closed contour .
"generatrix" of the tube (straight-line path) and on this generatrix there is embracing the tube does not change its magnitude when the points
oint belonging t o the curve C,. Therefore, to each value of a there
correspon
Only One 'As only one point of the curve C,, i.e. the coordinates of the point of
the curve C%are functions of the para~netera.
of the contour are arbitrarily displaced along the generatrices of
the tube. We will now prove t h a t the following relations hold bet-
'
102 Variational Principles and Integra:l Invariants [Ch. 3 Sec. 181 .Basic Integral Invariant of Mechanics

ween the function H and the functions Qi, Pi:
That is to say, we will prove that equations (13) are Hamilton's
canonical equations with the functioil H that enters into the expres-
sion under the sign of the integral I.
To proceed with the proof we introduce an auxiliary variable
(parameter) p , supplementing the system (13) with yet another
equation:
By virtue of invariance,
d I = 0,
where the letter d denotes differentiation with respect to the para-
meter p. Differentiating under the sign of the integral, we find
n
Writing 6 dqi and 6dt in place of d6qi and d6t, and integrating
by parts round the closed contour, we get*

Here, n = n (t, qi, pi) is an arbitrary function of a point in -6pi dqi) - dH6t + 6H dt] =
the extended phase space. Integrating the system (15), we find the
expressions for qi, p , and t in the form of functions of the variable p
and of the arbitrary initially given values, g:, p",( = l , . . ., n),
and to (for p = O ) :
qi = q i (p; P!, to),
pi = $i (p; $7 P?, to) , (i = 1, n) (16)
t = X (p; q!, PP, to) or, by virtue of (15), dividing termwise by
m
We have thus obtained parametric equations for the family of all
straight-line paths. Since we only need those straight-line paths
that form the given tube, we have to choose the initial point
M O(g!, p!, to) on the curve CO,i.e. we have to put

The expression under the integral sign must be a total differential

into equations (16) in place of qj, p! and to. Having done that we with an arbitrary factor 3 t ; but this is only possible when the expres-
will find the parametric equations for the straight-line paths that sion in the braces is zero. Equating this expression to zero, we get
form the given tube:
pi= --aH Qi S-
aH
(izl, ..., n),
aqi ' a ~ i

which completes the proof.**

Here, the value of a isolates a definite straight-line path (the Igene-
ratrix" of the tube), while the value of the parameter p fixes a ' * The operations d and Gmay be interchanged since they represent diffe-
definite point on this path. rentiation with respect to different independent variables p and a. Further,
when integrating by parts, the integrated part is lost (is equal t o zero) since
Assuming p = const, we-,obtain a point on every generatrix, the terminal point of the integration path coincides with the initial point.
and on the tube we get a closed curve. We take i t that in the integral
(12), g,, p i and t are replaced by their expressions (17). Then the
integral I will be a function of the parameter p and for every fixed
Thus, for any two functions
over a closed contour.
U and v ,
(§ uGv = -
4 vGu when integrating

aH
value of p, it will be a line integral along the corresponding closed ** We also obtain the identity --dH
at at
= -, which is a consequence of Hamil-
curve p = const. ton's canonical equations [see equation (20) on page 761.
 104 Variational Principles and Integral Invariants [Ch.3 S e c . 1.91 Theorems of Thomson nnd Helrnholtz on Circulation and Vortices 105

From what has been proved it follows that the invariance of the We can illustrate this in the case of a linear oscillator for which
PoincarB-Cartan integral may be placed a t the foundation of mecha-
nics since from this invariance it follows that the motion of a system
obeys Hamilton's canonical equations.
Form the canonical equations taking q for the independent variable.
Nde. In the proof we introduced an auxiliary variable p and used the cir- To do this, put
cumstance that the integral I does not change its value in passmg from one
curve of the family p = const to another curve of the same family. Because
the function n ( t , q,, p i ) is arbitrary, the family of curves p = const i s actually
an arbitrary family of nonintersecting closed curves embracing the given tube whence
of straight-line paths. If we had not introduced the parameter p, and had taken
the time t for the parameter, then, by the same reasoning, we would have only
partially made use of the invariance of the integral I (only for the curves of the Thus, we 'have
simultaneous states t = const) and we could not have arrived a t the desired -
K = = --I/m1/-22-cq2
result.
Now let us take a more detailed look a t the structure of the The appropriate canonical equations ( 2 2 ) will then read
PoincarB-Cartan integral.
In the PoincarB-Cartan integral (12) the time t enters as the coor-
dinate q i , and the role of the corresponding momentum is played
by the quantity -H, i.e. the energy taken with opposite sign.
This is a far-reaching analogy. From the second equation, z = const = -h. We find t by means
We change the variables in the integral I by introducing a new of a quadrature:
variable z connected with the old variables by the relation
i

Using thik relation let us express p,: /

where o = I/ m, and
C
a is an arbitrary constant, or

Then the basic integral I in the new variables will read wt +a =Z arc sin

Thus, in the new variables the integral I has the aspect of the
PoincarB-Cartan integral, but the role of the time is now played
by the variable qi and in place of the earlier energy H we have 19. A Hydrodynamical Interpretation of the
the momentum pi taken with reversed sign, i.e. K. Thus, by what Basic Integral Invariant. The Theorems of
has been proved, the motion of a system in the new variables should Thomson and Helrnholtz on Circulation and
be described by the following Hamiltonian system of differential Vortices
equations: . For a concrete interpretation of the concept of an integral invariant
let us consider the motion of an ideal fluid under the action of
external forces with a potential IT ( t , X, y , z ) . As we know from
hydrodynamics the equation of motion of a particle of this fluid
has the form
1
Where, g, is the independent variable. a = - grad IT - - gracl p, (1)
P
106 Variational Principles and Integral Invariants [Ch. 3 Sec. 191 Theorems of Thomson and Helmholtz o n Circulation and Vortices 107

where a is the acceleration of the particle, p and p are its density arbitrarily along the paths of motion of the particles; i.e. the integral
and pressure, while the potential IT is referred to unit mass.
We take i t that p and p are connected by the functional relation
C
U (1, X, y, 2) dx+ U (t. 5 7 Y, 2) dy +
p = f (p) (in particular this occurs in a n isothermal development
of the process). Then, putting +W (t, X, y, z) dz--E (t, X, y, z) dt (5)
n=n+ j $ is an integral invariant in extended coordinate space for the motion
of a fluid with a given field of velocities.
we can write equation (1) as If the path of integration consists of simultaneous states
a = - grad 11
- (t = const), then the integral (2) takes the form
(1')
This equation shows that the motion of a particle of the fluid
9 usx +usy +wsz
C
is identical to the motion of a particle with mass m = 1 i n a poten-
a
tial field fi = (t, X, y, z). Therefore, the integral invariant for In hydrodynamics this integral is called the circulation of the
velocity along the contour C. Along with this we obtained Thornson's
the motion of fluid particles will be the Poincark-Cartan integral,
which in the given case is of the form theorem on the conservation of the circulation of velocity: the
magnitude of the circulation (6) does not change if the fluid particles
forming the contour a t time ti are transferred to positions that they
occupg at any other arbitrary instant of time t2.
[f the fluid particles a t some time form a line, then these same
where U, v, and W are the components of velocity of the particle particles will form another line a t another instant of time. We
(in the given case-for m = l-they are the momenta pi), and shall speak of a &'fluidline" that moves and is deformed with time.
E is the energy defined by the formula In sipzilar fashion we define the concept of a ';fluid surface".
The theorem on the conservation of circulation states that to
every closed fluid line there corresponds a definite circulation.
Note that by the Stokes formula the integral (6) is written in the
Thus, the integral (2) taken round an arbitrarily closed contour form of an integral over the surface S bounded by the contour C:
in the seven-dimensional space (t, X, y, z, U , v, W) does not change
its magnitude in an arbitrary displacement of points of the contour
in accordance with the motion of the fluid. This motion obeys
11
S
5 6 ~ l-
6 $~2 6 ~ + @X&. (72
the differential equations, which, by virtue of the formula (l'),
have the form where

are the comppnents of some vector 9 called the curl (rotor) of the
For the given special case, equations (4) are Hamilton's canonical velocity or simply the curl. Integral (7) is ordinarily given i n the
equations. form
If the specific motion of a fluid is given for which the field of
velocities is known, i.e. if we know the functions U ( t , X, y, z),
v (t, X , y, z), and W (t, X, y, z), then the integral (2) may be
regarded as an integral in extended coordinate space, i.e. in the
four-dimensional space t , X , y, z. The value of this integral does where Q, is the projection of the vector 9 on the normal to the
not change if the points of the path of integration are displaced surface, and d S is an element of the surface S. From this it is seen
 108 Variational Principles and Integral Invariants [Ch. 3
that the integral (7) gives the magnitude of the vortex flow through
the surface. The theorem on the conservation of the circulation of
velocity passes into the theorem on the conservation of vortex
flow: t o every bounded fluid surface there corresponds a definite
magnitude of vortex flow through this surface.*
The motion of a fluid with a given field of velocities is determined
by the differential equations
W i t h respect t o the integral curves of the system (g), the inte-
gral (5) is a n integral invariant.
We now pose the question of what other systems of differential
equations of the type
besides the system (g), possess this property; in other words, f o r
what other systems is the integral (5) an integral invariant?
To answer this question, let us introduce a parameter y for the
trajectories of the system (10) and, as was done in the preceding
section, let us equate the relations (10) to the product n (t, X, y,
z) dy, where n is an arbitrary function. Let us consider a tube of
integral curves of the system (10) and a closed contour C around this
tube, for which contour y = const = c. Note that the value of
the integral (5) along the contour C does not depend on the quanti-
t y y =. c.
Denoting by d the differentiation with respect to y and reasoning
as on page 103 we get [using formulas (8)l
where the coefficients of 6y and 62 are obtained from the coefficients
of 6x by a cyclic permutation.
* Frbm this it follows in particular t h a t vortices i n the Quid volume of an
ideal fluid can neither vanish nor appear [if of course the forces have a potential
and the relation p = f ( p ) holds].
Sec. 191 Theorems of Thomson and Helmholtz on Circulation and Vortices 109
Replace dx, dy, dz, dt by the denominators of the fractions in
(10) multiplied by st (t, X, y, z) dt. Since for any choice of the
function n (t, X, y, z) the expression under the integral sign must
be a total differential, i t must be identically zero. And so the
expressions in square brackets are equal to zero (after the differentials
dx, dy, dz, dt in them are replaced by the proportional quantities
P, Q, R , U); i.e. we have the equations
and
The relation (12) is a consequence of the equations (11) if U # 0,
and of the equations (11) and (1') if U = 0.
The equations (11) together with equations (10) determine all
the differential systems with respect to which the integral (5) is
an integral invariant. Among these systems we shall seek those
for which U = 0, i.e. dt = 0. Then from (11) we get
and the system (10) takes the form
The integral curves of the system (13) are called vortex lines.
Thus, the system of differential equations of vortex lines is the only
system with d t = 0, with respect to which the integral (5) is a n integral
invariant. *
From this important proposition follows an important corollary.
In the space (X, y, z, t) take an arbitrary tube of vortex lines
and two contours Ci and C 2 bounding i t (Fig. 34). By virtue of the
invariance of the integral (5) with respect to the vortex lines,
9 u6x + +w6z -EFL 9 u6z +uSy +w6z -Eat
=
C1 c2
* The integral (5) is an integral invariant for other systems as well [for
example for the system (g)] for which d t # 0.
110 Variational P r i n c i ~ l e s and Intearal Invariants
We arbitrarily specify the magnitude a > 0 and transfer any
point of the space (2,y, z, t) to the point (X', y', z', t X) where
X', y', z' are the coordinates a t the time t +
a of the fluid particle
which a t time t had the coordinates X, y, z. I n such a displacement
along the trajectories of the fluid particles the vortex lines will
move to certain new lines which we will call &&displaced lines".
The tube of vortex lines that we took will move to the tube of
Fig. 34
displaced lines, and the contours C , and C z will move into the con--
tours D , and D 2 (see Fig. 34).* Since the displacement was accomp-
lished by the motion of fluid particles, the integral (5) does not
change its value during the displacement:
but then
I t may be taken that D1 and D z are two arbitrary contours bound-
ing the tube of displaced lines. Therefore, the equation (14) expres-
ses the invariance of the integral (5) with respect to the "displaced
lines". Here, dt = 0 along each displaced line, as i t is along a vortex
line. Consequently, displaced lines possess the same properties that
(as was shown earlier) only vortex lines can possess. Hence, the
displaced lines are vortex lines. And the time of displacement .c
is arbitrary. Thus any vortex line always remains a vortex line during
the motion of the fluid particles that comprise it.
We have arrived a t the Helmholtz theorem, which may be formula-
ted as follows: a vortex line is a fluid line.
* The tubes of vortex lines considered here are located in the four-dimen-
sional space ( X , y, z , t ) but the t-axis is not depicted in Pig. 34.
[Ch. 3 Sac. 201 Generalized Conservative Sustems 111
At the same time we found that with each vortex tube there
+ is associated a definite "intensity" defined by the integral
The magnitude of this intensity does not change during motion
of the fluid. If we take the tube of vortex lines for one and the same
instant of time t (at = 0), then the intensity (15) is the circulation
of the velocity round the contour C :
20. Generalized Conservative Systems.
Whit taker's Equations. Jacobi's Equations,
The Maupertuis-Lagrange Principle of Least Action
We consider a generalized conservative system, i.e. a n arbitrary
system (and perhaps a non-natural one as well) for which the func-
tion H is not explicitly dependent on the time. For it we have t h e
generalized integral of energy
H (qi pi) = h (1)
This integral is analogous t o the integral of conservation of
momentum p, = c, which holds when q, is a cyclic coordinate,
8H
i.e. whkn - = 0.
841
Proceeding from the analogy between the variable of time t and
a cyclic coordinate, we may expect to be able, with the aid of the
energy integral ( l ) , t o reduce the order of the set of differential
equations of motion by two units.
With this purpose in mind, we consider an ordinary (i.e. nonex-
tended) 2n-dimensional phase space in which the quantities qi, p i
.
(i = 1 , . ., n) are the coordinates of a point. We confine ourselves
to only those points of the phase space whose coordinates satisfy
the equation (1) with fixed value of the constant h,. I n other words,
we confine ourselves solely to those states of the system to which
the given magnitude of the total energy corresponds:
H = H (q!, pp) = h,
Then the basic integral invariant 1 will appear as
112 Variational Principles and Integral Invariants [Ch. 3 Sec. 201 Generalized Conservative Systems 113

since, by virtue of equality (l),
Now solve equation (1) for one of the momenta, for example pi:
pi = - K (qi, . qn, p2, . pn, ho) (3)
and put the expression obtained into the integral (2) in place
of p,; then
n where q; =
But the integral invariant (4) again has the form of the Poincar6-
Cartan integral if it is assumed that the basic coordinates and
momenta are the quantities qj and p j ( j = 2, . . ., n), and the
variable qi plays the role of the time variable (in place of the func-
tion H we have the function K). For this reason (see Sec. 18), the
motion of a generalized conservative system should satisfy the
following Hamiltonian system of differential equations of order
2 n - 2:
These equations were obtained by Whittaker and are known a s
Whittaker's equations.
Integrating the system of Whittaker's equations we find the qj
and p j ( j = 2, . . ., n) as functions of the variable qi and 2n - 2
arbitrary constants cl, . . ., c ~ ~ Moreover,
- ~ . the integrals of Whit-
taker's equations will contain an arbitrary constant h,, i.e. they
will be of the form
where all the variables in the partial derivative g
are expres-
sed in terms of qi with the aid of the equations (6jr;nd (6').
The Hamiltonian system of Whittaker's equations (5) may be rep-
laced by an equivalent system of equations of the Lagrangian type:
-
dqj
dPi
, and the function P (the analogue of the Lagran-
gian function) is connected with the function K (the analogue
of the Hamiltonian function) by the equation*
Note 'that the system (8) contains not n but n- l second-order
equations. I n the latter part of this relation the momenta p j must
be replaced by their expression in terms of
.which may be obtained from the first n - l equations (5).
We transform the expression for the function P by proceeding
from the equations (3) and (9):
In the case of a natural,. i.e. an ordinary conservative system,
+
L = T - T[ and H = T IT; consequently,
1 2T
P=-;--(L+H)=-;-
Pi Pi
and the kinetic energy T may be written as

All the trajectories in phase space (i.e. the geometrical pattern

of motion) are determined in this way. where
We obtain the dependence of the coordinates on the time t from n

at
--
d P i - aH
the equation -
api
by means of a quadrature:

* See equation (8) on page 73.

8-3420
114 Variational Principles and Integral Invariants [Ch. 3 Sec. 201 Generalized Conservative Systems il5.

From equations (1) and (12) we find and Hamilton action W:

and for the function P we get t h e folloiving expression:

For an ordinary (natural) conservative system, i t is possible to
take advantage of expression (11) and represent the Lagrange action
in the form
The dillerential equations (g), in which the function P is of the
form (15) and which therefore belong to the natural (i.e. ordinary)
conservative systems are referred t o as Jacobi equations.*
Integrating Jacobi's equations, we determine all the trajectories
.
in the coordinate space (g1, . ., g,):
But i t has already been assumed that only those motions of a system
are permitted to "compete", for which the total energy of the system
The relation between the coordinates and the time variable has one and the same value of h.
is established from (14) by a quadrature:

Let us now establish the principle of least action of Maupertuis-

Lagrange. Since the equations (8) are Lagrangian type equations,
there follows the variational principle according t o which for the
straight-line path
6W* = 0 (18)
where

W *=
i P dqi
(19)
is the Lagrange action. Here, all the motions of a generalized-conser-
vative system that transfer the system from a given initial position
g! to a specified terminal position qt (Fig. 35) are permitted to con-
tend. The instants of time t o and t1 are not fixed and may vary
when passing from the straight-line path to circuitous paths. * *
Expressing the function P in the integral (19) with the aid of
equation (10), we find the coiinection between Lagrange actioii W*
* These equations were derived by the Geman niatllematician K.G.J. Ja-
cobi and are found in his classical Vorlesungen iiber Dynamik published in 1886.
In the case of a non-natural generalized-conservative system, the function P
involved in the Jacobi equations is determined by the formula (9).
** Equation (18) states that for a straight-line path, the Lagrange. action
has a fixed value. The question of when the action W* is least for a straight-line
path is solved by invoking kinetic foci in the same way as for Hamilton's prin-
ciple.
Fig. 35
The expression obtained for W* indicates that the Lagrange action
W* is equal to the work of the momentum vectors of points of the
system i n a corresponding displacement of the system.
The variational principle (18) is referred to as the M.aupt?rtuis-
Lagrange principle of least action.*
By way of illustration let us compare the optical principle of Fermat and
the principle of least action of Maupertuis-Lagrange **
* First formulated vaguely by Maupertuis in 1747. In 1760 Lagrange gave
a rigorous formulation and substantiation of the principle.
.** See, for example, H. V. Rose Lectures i n Analytical Mechanics, Part 1.
Leningrad, 1938, pp. 93-94 (in Russian).
8*
 216 Variational Principles and Integral Invariants [Ch. 3 117
Sec. 211 Inertial Motion
According to Fermat's principle, l i ht is propagated in an inhomoge-
l neous medium so that the time of flig&
21. Inertial Motion. Relation to Geodesic Lines
in the Arbitrary Motion of a Conservative
Systern
SO
i s a minimum. Let there be given an arbitrary scleronomic system; its kinetic
energy is
Introducing at each point of the meaium the refractive index n = , G n
i.e. the ratio of the speed of light c in vacuo to the speed v in the given
medium, we transform the formula to

t = Ci ) n d s We introduce a metric in the coordinate space (q,, . . ., qn) by

determining the square of the arc length ds2 with the aid of the
SO
positive-definite quadratic differential form
where n is a function of a point of the medium. Then the Fermat principle
is written as

6 j nds=0 (23)
Then the magnitude of the arc of the curve connecting two points of
SO

On the other hand, for a single particle of mass m the Manpertuis- coordinate space, (q;) and (qi), will be determined by the equation
Lagrange principle yields

a j m v a ~ = v < 6
SO
5
so
~ ~ j d s = o (24)
Comparing formulas (1) and (2) we find that in the motion of the
The trajectory of a light ray coincides with the trajectory of a particle system
. if [see formulas (23) and (24)]

If it is assumed that near the earth's surisce the refractive index n

That is, the kinetic energy of a system [given the metric (2)l always
diminishes as a linear function of the altitude z coincides with the kinetic energy of the representative point in the n-dime-
nsional coordinate space if to this point we assign a mass of m = 1.
Now consider the inertial motion of the system (Il = 0). Then
in that motion all possible trajectories of the image point are
where H is the height of the atmosphere, then, neglecting small terms of referred to as geodesic lines [with respect to the metric -(2)1. From
the order of (+)', we can write the energy integral
T=h
1 nl ( I - 2 ~ 4 ) = c + g z
Irl (27) by formula (4). i t follows that
where ds -
-
dt
=1/2h
c = h - - n 221
kn2
0, g=> H (28)
That is, to inertial motion (and also to any motion with a constant
Formula (27) defines the potential of the force of gravity near the surface value h of the kinetic energy) there corresponds in the coordinate
of the earth with a modified value of g.
Therefore, i f the refractive index n varies with altitude in the manner space (qi, . . ., qn) a uniform motion of the representative point with
indicated, light will be propagated in a parabola with a vertical axis. velocity
128 Variational Princinles and Inteeral Invariants [Ch. 3 Sec. 221 Universal Integral Invariant of Poincari

In accordance with the principle of least action, the geodesic lines Comparing (10) and (8) we conclude that for a conservative system
are extremals* of the variational problem the trajectories of straight-line paths are geodesic lines in a coor-
dinate space with the metric
6W* = o (6) n
where W* is the Lagrange action. However, i n the case under
consideration we have the energy integral T = h with fixed value
of h both for the straight-line path and for a circuitous path.
Therefore, 22. The Universal Integral Invariant of PoincarB.
Lee Hwa-Chung's Theorem
i
W*== 2 ~ d t - = 2 h ( t - t ~ ) = 1 / % s
to
(7)
Now consider the integral 1 = 5 [X
n

pi6qi -HI%] round the

where S = l/% (t - to) is the length of the curve in the coordinate i= 1
-
Space (qi, . ., qn). contour C consisting of simultaneous states of a system. Such
The variational problem (6) takes on the form a contour results if a tube of straight-line paths (see Fig. 33 on

Ss=*
,$
7 \/ r,
i, k=i
aikdqidqk=O (8)

'Thus a geodesic line is distinguished by the fact that the arc

length of the curve has an estremal (stationary, to be more precise)
value as compared with the arcs of other curves having the same
end points as the geodesic line** (see Fig. 35 on page 115).
When for a straight-line path the Lagrange action is a minimum,
the arc length of the geodesic is less than the length of any other
curve connecting the same points as does the arc of the geodesic.
, That is why geodesic lines are also called the shortest lines i n space.
Now consider a conservative system, i.e. a scleronomic system Fig. 36
with potential energy IT = IT (q,, . . ., q,) not explicitly dependent
on t. Then, by (15) and (19) of the preceding section, page 100) is cut by a hyperplane t=const. For such a contour,
6t = O and the basic integral invariant takes the form

Therefore, the molion of a conservative system with a given value

of total energy h is accomplished in a coordillate space along the
extremal of the variational problem (with fixed end points):
n
*That is, curves Ior which W* has a stationary value.
**This always occurs when the ends of the arc of a geodesic line arc suf-
ficiently close together.
This integral was first introduced by Poincare'. Later, Cartan exten-
ded the integral to contours consisting of nonsimultaneous states
by introducing an additional summand --H&.
Poincarh's integral invariant I, does not change its value if the
contour C is displaced along the tube of straight-line paths to the
contour C : which again consists of simultaneous states. I t is conveni-
ent to consider the integral I, in the ordinary (nonextended) 2n-di-
mensional phase space (q,, p,, . . ., Q,,,p,). I n this space, to the con-
tours C and C' (Fig. 33) there correspond the contours D and D'
 120 Variational Principles and Integral Invariants [Ch. 3 Sec. 221 Universal Integral Invariant of Poincare' 222
;,l
bounding the tube of "straight-line" paths (Fig. 36). Here, l l
tem. That is why the integral I , is called the universal integral
l
11 m invariant.
4 The following proposition may likewise be readily proved.
If for some system of differential equations
Note that one of the contours D and D' (say D) may be chosen quite
arbitrarily. It may be taken that the points of the contour D are
different states of the system a t one and the same instant of time t; the integral I , is invariant, then the system (4) is Hamiltonian.
then the corresponding states of the system at time t' will form the Indeed, in this case
contour D'. T,

In place of the phase space we consider n separate phase planes

(qi, pi) (i = 1 , . . ., n). Project an arbitrary closed curve (contour)

9 From this it follows that the expression under the integral sign i s
Fig. 37 a total virtual differential of some function -H (t, qk, pk):*

D situated in the phase space onto these planes (Fig. 37). We get the
contours D i (i = I , . . ., n). Then for any i
that is

where Si is the area of the region bounded by the contour D i in the

plane (qil pi) (i = I , . . ., n). The sense of circulation about the and the proof is complete.
contour D induces the sense of circulation on the projection Di. Note the following terms: the Poincar6-Cartan integral I and
In forn~ula(2) the sign before Si is plus if the circulation of D i the Poincar6 integral I , are called relative integral invariants of the
is clockwise (that is, in the direction of the shortest turn of the axis first order. The term &relative1means that the domain of integration
p i to the axis qi), and it is minus otherwise. is a closed contour and the &firstorder' implies that the differentials
Then enter linearly in the expression under the integral sign. Note that
n n the relative integral invariant of the first order I , may, by means
Ii = 52 i= i
pidqi = 2 & Si
i= i
(3)
I
of Stokes' formula, be represented in the form of an absolute integral
invariant of the second order:

,Thus the contours D and D i vary during the motion of the system,
and the areas Si also vary, but the algebraic sum (3) of these areas
remains constant. That is the geometrical interpretation of the inva-
riance of the Poincar6 integral.
H does not appear in the expression for I , . Consequently, Poin- * Hen, 8 ~ =2 (-aqi
aH
aqi +-aH h p i ) .
a ~ i
car6's integral I , is invariant with respect to any Hamiltonian, sys- i= 1
 122 V a r i a t i o n a l Principles and I n t e g r a l I n v a r i a n t s
where the integral on the right is taken over the surface S bounded
by the closed contour D.
This formula, as applied to the integral I , , yields
i= i
I n 2n-dimensional phase space there are known to exist the fol-
lowing universal relative integral invariants 1zk-, of odd orders
and the absolute integral invariants JZk of even orders,
[Ch. 3
Sec. 221 Universnl I n t e g r a l I n v a r i a n t of Poiizcare' 123
The general solution of this system is of the form
\\*liere qo, p, are the initial values of q and p for t = to.
Let
q=qo(u), p==p,(a) (9)
[O,< a 4 1; qo (0) = cl0 (l), PO(0) = p , (41
be the equations of the closed contour D, in the phase plane (Fig. 38).
The points which at the time t = t, were located on the contour D,
form the contour D at some other arbitrary instant oI time t. The

= Jzn = -. 1
Spilhqil . . hpinhqtn Fig. 38
I11 1947, the Chinese scientist Lee Hwa-Chung proved the unique-
ness of these universal integral invariants. He demonstrated parametric equations of this contour are obtained from the equalities
that any other universal integral invariant differs by a constant factor (8) if q, and p, are replaced by their expressions (9). This yields
I
From one of the enumerated integrals.
We shall need the Lee Hwa-Chung theorem later on for an integral
invariant of the first order; we therefore formulate the theorem for Putting these functions in the integral I' in place of q and p ,
an arbitrary n and will prove it for n = 1. we get I' as a function of the parameter t. From the invariance
Theorem of Lee Hwa-Chung. I f dI'
n of I' i t follows that = 0. Difierentiating under the integral

If= @ 2 [Ai (t, qk, pr) 8qi + B. (L, ~ A pa)

I 6piI sign and integrating by parts" we find
dA dB d d
i= i
~ dt = ~ - ~ ~ B-hp-
dht ~ + ~
is a universal relative integra,l invariant, then
dB ay dP
l If= c l t = @-$hq+-hp dt +Ah-+BA--
dt dt
where c is a constant, and It is the Poincart! integral. drl dq dB dl,
Proof. Let =$-nTaq-a~z++;iT~P-~~- dt =

be a universal integral invariant. The integration is taken round air a.4 an ao

a closed contour in the phase plane (q, p). Further, let there be given =@ ( - ~ ~ + T ) h u + ( - z ~ + F ) ~ ~
some Hamiltonian system of differential equations with the function
H (L, q, p): * See the footnote* on page 103. When making transformations, we
dq
-- aH dp
-=-p
aH aA BA
assume 6 A = -6q+ -6 p and 6 D:-
aB
6q
aB
+-
6 p and then use the equ-
at-=' dt % (7) d9 a~ ay a 2,
ations (7).
124 Variational Principles and Integral Invarial~ts [Ch. 3 Sec. 231 Invariance of Volume i n the Phase Space 125

where 23. Invariance of Volume in the Phase Space.

Liou ville's Theorem
We consider the "total" absolute integral invariant
The last integral is equal to zero for any value of the variable
t considered as a parameter and for an arbitrary path of integration.
Therefore the expression under the sign of the integral must be a to-
tal differential with respect to the variables q and p. From this, The invariance of this integral implies the invariance of the phase
volume in a 2n-dimensional phase space and is established in the
following manner."
Let us write down in the following form the final equations of mo-
tion that result after integration of Hamilton's equations:
or, after elementary transformations,

where qi, p i are the initial values of q,, p, for t = to (k = 1 , . . ., n).

Since the function H may be chosen quite arbitrarily, i t follows In phase space we choose some volume Jo and take each point
that
of this volume as the initial point (for t = to). Then by time t the
transformation (2) will have carried the volume Jo into the volu-
m e J. Here,
that is,

Then
where

and, consequently, there exists a function cD (t, q, p) such that* that is, I is the Jacobian composed of the partial derivatives of
q j , .p j with respect to the initially given q!, p? (i, j = 1 , . . ., n).
Without loss of generality, we can consider that the Jacobian under
th6 integral sign in (3) is positive and we can drop the absolute-
But then value sign.
When t = to, this Jacobian is equal t o 1 since for this value of t
A6q +B6p = cpdq + 6@ a l l the q, = qj: and the p, = p:. When t varies, the Jacobian varies
and therefore continuously without vanishing, since the singular points a t which
this Jacobian might vanish are eliminated from our consideration;
I' =
4 A6q +Bap c $ pdq
= = cl, i . e. i t is assumed that there are no such points i n the volume under
examination. Then the Jacobian is positive in this volume.**
which completes the proof.
For n > l , the idea of proof is retained, though the proof itself * Henceforward (see Sec. 31) the. invariance of the phase volume will be
becomes more involved. established on the basis of the general properlies of inotion of Hamiltonian
systems.
* Here, the time t is taken as a parameter.
** No assumptionsconcerning singular points are required in the proof on
pages 161-162 mentioned i n the preceding footnote.
 126 Variational Principles and Integral Invariants [Ch. 3 Sec. 231 Inuarinnce of Volume i n the Phase Space 127

Differentiating with respect to t under the integral sign we get
Jo
Compute the derivative of the Jacobian I:
2n
where Ii is the determinant obtained from the Jacobian by differen-
tiation of the i t h row. Now taking into account that
2" for any
From the invariance of the phase volume there follows one of the
basic theorems of statistical mechanics, Liouville's theorem.
Imagine a very large number of absolutely identical replicas
of a system differing from one another solely i n their initial states
97, pp ( i = 1 , . . ., n). All these replicas form a statistical ensemble.
An instance of a statistical ensemble is the collection of mole-
cules of a gas in a given volume.
To each volunle element d V of the phase space we can assign a
LLmass" d y that describes the quantity of replicas per given element of
volume d V . By virtue of the proved invariance of volume in phase
space, the magnitude of d V does not change with time. Physical-
l y speaking, the magnitude of d y does not change either, since the
replicas which a t soine time are in the volume d V will be displaced
together with the volunle. That is why the density of the statistical
ensemble remains unchanged during motion:

3" for any

we find
In expancled form, equation (5) may be written thus (see Sec. 15):

and

I I i lt=to = / 1
api
7
api t=tn
for i = n + 1 , ..., 2n
where (pH) is a Poisson bracket. According to (6), the function
p ( t , qi, p i ) is the integral of motion.
We have thus proved the following theorem.
therefore Liouville's theorem. T h e density of a statistical ensemble is always
a n integral of motion.
Thus, for example, for a conservative system any function of the
energy of the system may serve as the density of the statistical
ensemble.

= 0. Since the initial instant to may be chosen quite

arbitrarily, i t follows that for any t

which is to say that the magnitude of the phase volume J does not
change upon a displacement of the points of this volume from
, states occupied a t time to to states occupied a t any other arbitrary
instant of time t .
 Sec. 241 Canonical Transformations 129
CHAPTER 4
Example 3. The transformation
d

qi =pi tan t, pi =qi cot t (i = 1, ..., n)

Canonical Transformations and the will be canonical because it is readily verifiable that from the equations (2)
Hamilton-Jacobi Equation we always get equations (3) for
n
1
l 24. Canonical Transformations
H=-H+sin t cos t 2
i= l
i F i

The transformation of coordinates in a 2n-dimensional phase space To derive the conditions under which the transformation (I)
l
I
(containivag, in the general case, the time variable t as a parameter)
- is canonical, we consider two extended (2n
W W
+
I)-dimensional phase
spaces (gi, pi, t) and (gi, pi, t), one passing into the other under

is called canonical if the transformation carries any Hamiltonian

svstem

1 again into a Hamiltonian system (generally speaking, with another

I Hamiltonian function g):

The importance of studying canonical transformations is due

to the fact that these transformations permit replacing a given
Fig. 39
Hamiltonian system (2) by another Hamiltonian system (3) in which
the function g is of a 'simpler structure than H. the canonical transformation (I), and two tubes of straight-line
If in a phase space we perform two canonical transformations paths of Hamiltonian systems (2) and (3) (Fig. 39).
in succession, the resulting transformation will again be canonical.
Furthermore, a transformation that is inverse to a certain canonical Let'us take two arbitrary closed contours C and ?t that bound
these tubes and correspond to one another by virtue of the transfor-
-
transformation will always be canonical, and the identical transfor-
mation qi = qi, p i = p i (i = 1, . . ., n) is canonical. Therefore,
d
mation (1).Furthermore, cut both tubes with one and the same hyper-
plane t = const. I n the cross-section we get two "plane" contours
all canonical transformations taken together form a group.
Example 1. The transformation
CO and zo.
These contours likewise pass one into the other under
W -
qi=aqi, pi=$pi ( i = l , ..., n; a f 0, $ # 0)
the canonical transformation ( l ) ,since the quantity t remains un-
changed in a canonical transformation. From the invariance of the
as may readily be verified is canonical. It transforms the system (2) into PoincarB-Cartan integral it follows that
the system (3) with
fi=a$El
Example 2. The transformation
qi=aPi, pi=Fqi (i=l, ..., i z ; U # 0, B # O )
C

will be canonical. In this case,

-H = -abH
 Sec. 241 Cartonical Transformations 131
130 Canonical Transformations, Hamilton-Jacobi Equation [Ch. 4

On the other hand, if in the universal integral invariant We shall call the function F the generating function, and the
constant c the valence of the canonical transformation (1) under

$ $S$ we pass to the variables g,, pi ( i = 1, . .., n) by means
i= l
of the canonical transformatio~~(l), then this integral will pass
into a certain universal integral invariant of the first order
in 2n-dimensional phase (qi, pi)-space; by the theorem of Lee
Hwa-Chung, the invariant obtained can differ from
i n the constant factor c. Therefore
n n
From the equations (4) to (6) i t follows t h a t
n n
consideration. The canonical transformation will be called uniua-
lent if for i t c = 1.
A necessary and sufficient condition for the transformation (1) to be
canonical is the existence of a generating function F and some constant
c for which the equation (9) is identically satisfied by virtue of the trans-
n formation (1).
82 - 1.
i=
~ i % i only
IYote. If the transformation (1) is canonical, then there exist
a generating function F and a valence c # 0 such that the equation
(9) holds for any function H and the corresponding function g.
However, if (9) holds for one pair of functions H and 2, then trans-
formation (I) is already canonical."
Indeed, in addition to H take another arbitrary function H,
and define R, from the condition

L.
- Multiplying both parts of this equation by 6t and subtracting
If in the first integral we consider the variables &, , p,> .. . termwise the resulting equation from (g), we get
to be expressed i n terms of the variables q,, . . ., p, (here, the path
of integration E is replaced by the p a t h of integration C), then
the equation (7) may be written as
n
Thus, (9) holds for any function H, and the corresponding func-
tion 5,.
But C is an absolutely arbitrary contour in a (212. 1)-dimensi- + Canonical transformations are sometimes also called contact
transformations.
onal extended phase space. Therefore the expression under t h e
integral sign in (8) must be a tot,al differential of some function In the literature, only ullivalent canonical transformations are
of the 212 +
1 arguments q,, pi, . . ., Q,,, pI1 and t. We will find it frequently considered, and many authors erroneously hold that
these transformations exhaust all the transformations (I) that carry
convenient t o denote this function in terms of - F (t, qi, p i ) ; then*
71. Hamiltonian systems again into Hamiltonian systems. These authors
2 pi6qi --I% = c (i=2 pi6qi -~ 6 t -) 6 F
i=l l
(9) do not notice the arbitrary constant factor c which must figure
in the general formula for an arbitrary canonical transformation.
Note that the constant c in the identity (9) is always different
-
m

from zero, c # 0 , since the expression 2 pi6gi

i=i
C

- g6t is not a total * I t does not follow from this however that a canonical transformation may
be defined as a transformation t h a t carries one given Hamiltonian system into
differential*" a r ~ dtherefore cannot be eqial to -6F. some other Hamiltonian system.
'n Tks,fz example, the arbitrary noncanonical transformation = (gk,
38' aF dF 61. p h ) , p i = pi ( q k , pk) ( i = 1 , . . ., n ) carries the Hamiltonian system with
* Here, 6F = 2 (- dq i 6qi +-
a ~ 6pi)
i

**
i= i --
With respect to the independent,variables qi, p i , I and, hence, with
H = 0 into the Hamiltonian system with i? 0.

respect to the independent variables q i , pi, t ( i = l,. . . , n).

 132 Canonical Transformations, Hamilton-Jacobi E~uation [Ch. 4 Sec. 251 Free Canonical Transformations 133

25. Free Canonical Transformations would pass into the equality

Let us investigate in more detail the so-called free canonical
transformations. These transformations are characterized, in addi- but this is only possible when Q S 0, since the coordinates of a point
tion, by the inequality of phase space qi, p i (i = 1, . . ., n) are independent quantities.
as
From the independence of the derivatives -. (i = 1, . . ., n) con-
m m

that ensures the iadependence of the quantities t, qi, . . ., qn,
- the Jacobian of these functions is not identically zero. Thus, for
q,, . . ., &, which can now be taken as the basic variables.* Indeed,
this inequality enables us, using the first n equations ( l ) , Sec. 24,
to express the generalized momenta p,, . . ., p, in terms of 2n 1 +
quantities t, qi, qi (i = 1 , . . ., n) and, consequently, to represent
any function of the variables t, qi, p i (i = 1, . . ., n) as a function
of the variables t, qi, (i = 1, . . ., n). In this case it may be taken
Ci
that the generating function is represented as a function S of these
variables:
a9,
-.
sidered as fnnctions of the variables G , . . ., &, i t follows that
the generating functions S of a free canonical transformation the
determinant must be different from zero, i. e.:
From the inequality (5) i t follows that the first n equations (3)
.
may be solved for Gi (i = 1, . ., n) and thus all the new phase
variables qi, pi (i = 1, . . . , n) may be expressed in terms of the
W m
old variables qi, pi (i = 1, . . . , n). The transformation of the
m

form (1) thus obtained will be reversible, i.e. for i t a(q1' '."
Then the basic identity (9) of the preceding section will
be written as
? (qt7 -. ~ n #)O
- 9

since by virtue of inequality (5) the last n equations of (3) may

-
be solved for qi and, consequently, all the qi, pi may be expressed
i n terms of qi, pi ( i = 1, . . ., n). Thus, the equations (3) define
W

a free canonical transformation with a given generating function

Equating the coefficients of 6qi, 8q? and 6t on the left and the
right, we get the following formulas
S (t, qi, G)and with a given valence c # 0, while the formulas (4)
establish a simple relation between the Hamiltonian functions H
and g .
By running through the different generating functions S that
satisfy the condition (5), and the different valences c # 0, we can
obtain all the free canonical transformations* with the aid of the
formulas (3).
l
Equations (3) define the canonical transformation under conside- For univalent (c = 1) free canonical transformations, the formu-
ration. We shall show that they can be reduced to the form (1), las (3) and (4) take on a simpler form:
I Sec. 24.
The partial derivatives
as (i = 1, ..., n) on the left-hand sides
I
- 2 I

of the first n equations (3), as functions of the quantities

are independent because by virtue of (3) the relation**
zi, . .., 4n
m
and

The last equation shows that after applying one and the same
In the case of a nonfree canonical transformation, the quantities t , Q,
free univalent canonical transformation to various Hamiltonian
d

qi (i = 1, . . ., n) are connected by certain relations. * For nonfree canonical transformations there exist defining formulas si-
** .
In this relation, the quantities qi, . ., qn are taken as parameters. milar to (3). These formulas will be established in Sec. 29.
134 Canonical Transformations, Hamilton-Jacobi Equation [Ch. 4 Sec. 261 The Hamilton-Jacobi Equation 135

systems, the difference between 5 and H will be one and the same .
For a natural system, the coordinates q,, . ., q, define the
as position of the system, and together with the momenta p,,. . ., p,
at 1
(equal t o - i n all cases. they define the state of the system, that is, the positions and velo-
cities of its points. This specificity of the coordinates is lost in
From (4) it follows that B = CH when and only when as ;i;- = 0,
a general-type canonical transformation. The quantities G,
. . ., 4n
-
""
i.e. when the generating function S does not depend explicitly on t. no longer . define the position of the system, and only together with
In this case, by virtue of (3), the time t will not enter explicitly
into the formulas of the canonical transformation. Under such
the Fi, . . ., p, do they define the state of the system. The variables
canonical transformations the function H does not change essential- .
q,, . ., & will as before define the position of the system only
ly, i t is only multiplied by tlle constant c. For this reason, if we in the particular case of a point canonical transformation for which
desire to obtain a new system with an essentially simpler Hamilto- the functions (t, qk, pk) actually do not contain the momenta:
-.
nian function, we must definitely take a free caiionical transforma-
W

tion containing the time t as a parameter.
Example i . On page 128 we considered three canonical transformations:
d -
- -
( I ) qi=aqi, ~ i = $ ~ i (;2 ) q i = a p i , ~ i = B q i ;
(3) q i = p i t a n t , p i = q i c o t t ( i = l , ..., 12)
The transfornlations (2) and (3) are free. They have generating functions
n m
..
qi =qt ( L , q k ) (i = l , ., n)
Note that subsequently the transformation of an arbitrary Hamil-
tonian system into a system with the function H of simple struc-
ture may be effected with the aid of a free canonical transformation.
But a free canonical transformation is not a point transformation.
Thus, nonpoint canonical transformations play a n essential role
in the theory of Hamiltonian systems.

and valences, respectively, S = -$ qiqi, c= -a$, and S = --cot

qiTi, t
i=i i=1 26. The Hamilton-Jacobi Equation
c=-l. But the transformation (1) is not free. For i-t, c = a $ , F r 0.
Example 2. Consider a n arbitrary affine transformation of the phase The theory of canonical transformations leads us directly to the
plane (q, p) (with @ = 1):
Hamilton-Jacobi equation.
q=acl+Pp, Let there be given a holonomic system whose motion obeys the
P = ais +&P (4%-uiS # 0 )
(8)
canonical equations of Hamilton:
Put the right-hand sides of (8) into the basic identity (9) on page 130.
Since the variable t does not enter into the right-hand sides of ( 8 ) , we shall
also seek F in the form of a function not dependent on the time explicitly:
F=F ( Q , p). Then t h e basic identity will take on the form We shall try to determine a free univalent canonical transfor-
( a i q f flip) ( a h + $6e)-cpth= -6F mation such that i n the transformed Hamiltonian system
or
6 (+a a i q ~1 +$ $~i p a ) + a i P b p + (.S,-4 = -m
The left-hand side of this equation will be a total differential provided that
c=mPi-aiS
the function will be identically zero:
Thus, the transformation (8) is canonical with valence c, equal to the deter-
minant of the transformation, and n i t h the gciierating functio~i
Then the system (2) is integrated directly:
F=- 1 a a i r l q T 1 B h p 2 + a d q ~

This fransformation will be free if += 0.

where ai and pi are 2n arbitrary constants. Knowing the canonical
Example 3. The transforinntion ?-- cos 2p, F= s i n 2p is a free
univalent canonical transformation u it11 the gencmting function transformation, i.e. the relation between qil p i (i = 1,
- .. ., n)
1
- and q i , p i (i = 1 , . . ., n), we can express all the qi and p i as func-
S =- q arc cos q
2 71/; --- 21 -
q v- tions of tlle time t and of the 2n arbitrary constants a,, (k =
136 Canonical Transformations, Hamilton-Jacobi Equation [Ch. 4 S E C .261 The Hamilton-Jacobi Eouation 137

= 1, .. ., n), that is we can find the final equations of motion of
the given holonomic system completely [all the solutions of the
system (l)].
How are we to determine the canonical transformation that
we need? To do this, b y virtue of formula (7) of the preceding
section, i t is necessary and sufficient that the equation
hold for the generating. function S (t, qi, G)
of the desired cano-
nical transformation. This, together with the formulas (6) of the
same section, yields
The resulting partial differential equation (6) is called the Hamil-
W
a holonomic system with the given function H may be written in the
form*
where a i and fii are arbitrary constants (i = 1 , . . ., n).
Thus, a knowledge of the complete integral of the partial differen-
tial equation (6) relieves us of the necessity of integrating the
system of ordinary differential equations (1). The problem of inte-
grating this system is replaced by an equivalent problem of seeking
the complete integral of the partial differential equation of Hamil-
ton- Jacobi.
Note. The general solution of the partial differential equation
depends on several arbitrary functions. For this reason a complete
integral of the Hamilton-Jacobi equation is by no means a general
solution. A complete integral embraces only a small number of so-

ton-Jacobi equation. Thus the generating function S (t, qi, qi) with lutions when compared with a general solution. I t is nevertheless
basic variables t and qi (the qi
are regarded as parameters) satisfies possible, on the basis of a complete integral, to restore the initial
the partial differential equation of Hamilton-Jacobi. Besides the equation (whence the name "complete" integral). Indeed, differen-
Hamilton-Jacobi equation, the condition tiating the complete integral, we get

m
must hold for the generating function S (t, qi, qi).
m
As soon as the generating function S (t, qi, qi) is found, the If the complete integral S (t, qk, ak) is known, then the func-
formulas tions f (t, qk, ak) (i = 0, 1, . . . , n) are also known. From relation
(10) i t is possible to express every a k in terms of the partial
as
derivatives - , t and qi since, by virtue of condition (8),
aqi
will define the sought-for free canonical transformation. If in these
formulas we replace the Gi
by ai and the pi
by pi, we will get the
equations of motion of the given holonomic system in closed form.
Putting the expressions obtained for a k into (11) we get the initial
start the gi in S are replaced by ai (i -
This whole process is more conveniently described if from the very
1, ..
., n).
We introduce a definition. The solution S (t, qi, a i ) of the partial
partial differential equation (6). **
As an instance of a complete integral of the Hamilton-Jacobi
equation, we consider what is called Hamilton's principal function.
differential equation of Hamilton-Jacobi containing n arbitrary To do this, we return to formula (7) on page 99 and to Fig. 33
.
constants a,, . ., a, is called the complete integral of this equation on page 100. We now consider only the special case when to (a) =
if the condition = const ='to, i.e. we assume that the contour CO consists of the

is fulfilled.
Now let us formulate the theorem we have proved.
Jacobi's theorem. If S (t, qi, a i ) is some complete integral of the
Hamilton-Jacobi equation (6), then the final equations of motion o j
* Here, in place of the arbitrary constants -pi we write simply pi. By virtue
ot the condition (g), the last n equations (9) may be solved for qi and we can
express qi, . . ., q, as functions of t and the 2n arbitrary constants a,, pi.
+ ** I t may be considered that the complete integral S also contains an ( n
Z)st additive arbitrary constant a,+~,since only the derivatives of S enter
+
into equation (6) and not the function S itself.
,138 Canonical Transformations, Hamilton-Jacobi Equation [Ch. 4 Sec. 261 The Hamilton-Jacobi Eauation I39

initial states of the system when t = to. Besides, in place of t,,
qt, p;, H, we shall write simply t , qi, p i , H. Then, if W is the
action along a straight-line path (i.e. along the generatrix of the
tube) from the initial point (t = to) to the terminal point correspon-
ding to the given value of t, we get
If we take advantage of the final equations of motion
and in place of qi (t) put their values (14) into the expression
for the action
W=
i
10
L (t, qi, qi) dt
q: and the p:. The result was a vicious circle: t o write the final equa-
tions of motion (17) one needs Hamilton's principal function, and
to construct this function, as has already been demonstrated above,
one needs to know the final equations of motion.
Jacobi's contribution consists in the fact that he continued Hamil-
ton's investigations and broke the vicious circle. He demonstrated
that the final equations of motion might be written in the form (9)
by means of an arbitrary complete integral S (t, qi, a i ) of the Hamil-
t,on-J acobi equation.
Let us return to the identity (13) and compare i t with the identi- .,-
ty (2) on page 132. I t will be seen from this comparison that the
formulas (14) and (15), which are the final equations of motion and
which express the Hamiltonian coordinates qi, p i of the state
.
of the system a t time t in terms of the initial coordinates qy, . ., p i ,
may be regarded as a free univalent canonical transformation from
.
the variables qp, p! (i = 1 , . ., n) to the variables q,, p, (k =
= 1, . . ., n); -W is the generating function of this canonical
transformation, and W is Hamilton's principal function.*
,

then W will be a function of t , q!, p! (i = 1 , . . ., n). Hamilton Thus, a transformation of phase space carried out by means of the
proposed using the final equations of motion (14) to express p i motions of any Hamiltonian system is canonical (and also free and
in terms of t, qP and qi and thus to represent the action in the form univalent).
W = W (t, qi, q9) (18) Example. Form the Harniltonian function for the inertial motion
of a free particle. In this case (assume t o = O )
The action W given in the form (16), i.e. in the form of a function
of the initial coordinates, the terminal coordinates and the terminal x=xo+xot, Y=Yo+yot, z=zo+zot
and so
instant of time t, is called Hamilton's principal function. Considering t
that in (13) W is Hamilton's principal function, we get (on the basis
of this equation)
aw arv +(Y--Yo)~+ (~-20)~l
aqi 'Pi, a@ - -p:
-- ( i = l , ..., n)
I f we proceed from Hamilton's principal function t.hat we found
and nz
aw
--
at - - H ( t , qi, pi)
W = -[(X-xo)2+
2t
(y- yo)2+ (z-zo)2]
(18)
.then the equations of motion are obtainable from the formulas (17), which
From the equations (17) and (18) i t follows t h a t Hamilton's in the given inseance appear as
principal function satisfies the Hamilton-Jacobi equation

and the relations (17) are the final equations of motion containing
2n arbitrary constants qt, pp (i = 1 , . . ., n).
Thus, Hamilton demonstraied how the final equations of motion
are written by means of a complete integral of the equation (19). * For the inverse canonical transformation from the variables qi, pi to the
However, in Ha~nilton'scase this complete integral was not arbit- variables qp, p!, I-Iamilton's principal function M' will be the generating func-
rary; and the arbitrary constants in i t were the initial values of the tion.
140 Canonical Transformations, Hamilton-Jacobi Equation [Ch. 4 Sec. 271 Mefhod o f S e ~ n r a t i o no f Variables. Examnles 14 1

Suppose we have a generalized-conservative system ( $ = O ) . and the complete solution of the Hamilton-Jacobi equation may
be sought in the form
Then the Hamilton-Jacobi equation has the form
S= 2
p=rn+ i a p q p f SOU, qi, . . . , qm, %it . . ., a,) (25)
and its complete solution may be found in the form For the function So we get the equation
S = -ht+V(qi, ..., q,; a,, ..., a,-,, h) (21)
where h and a i , . . ., an-iare arbitrary constants.
Putting this expression for S in equation (20), we get the follo- If in a generalized-conservative system the coordinates g,,,, . . ., 4n
wing equation for determining the function V: are cyclic, then the furiction S is sought in the form

Finding the complete integral of this equation, i.e. the solution

V(qi; a,, . . ., an-i, h) for which the inequality where the function V, is determined from the equation
det
aqiaak i, k=i
p 0 (a. -h)
holds, we obtain with the aid of formulas (9) and (21) the followillg
final equations of motion of a generalized-consertative systenl: 27. Method of Separation of Variables.
Examples
We have shown that integrating a system of canonical equations
av reduces to finding a complete solution of the Hamilton-Jacobi equa-
aaj - p j
-- (j=l, . . ., n - l ) , tion. The interest here is not only theoretical. I t turns out that many
problems of dynamics, including problems of interest in theoretical
physics, may be conveniently solved in this way.
We shall now examine the method of separation of variables
where a j , pi ( j = 1 , . . ., n - l ) , h and y are arbitrary constants. for finding a complete solution t o the Hamilton-Jacobi equation.
By virtue of the condition (23) the coordinates qi, . . ., Qn may This method is applied in cases when the Hamiltonian function
be determined from the equation (24") as functions of t and the 2n H of a generalized-conservative system has a special structure.
arbitrary constants a j, f i j , h and y. Putting the expressions obtained 1". Let
into equations (24'), we get similar expressions for the generalized
momenta p i (i = 1 , . . ., n)." H=G[fi(qi, - . a , fn(q.9 ~n)l (1)
In the case of a . generalized-conservative system we replaced
equation (6) by equation (20), in which the number of independent Here, the variables in the expression for the function H are sepa-
variables is less by one. A similar reduction i n the number of inde- rated, i.e. only one pair of conjugate variables qi, p i enters into
pendent variables in the partial differential equation m a y be also each function fi.
accomplished when one of the coordinates is cyclic (ignorable). Equation (22) of the preceding section will now be written as
Let us consider a t once the general case when several coordinates
.
qm+i, . ., qn are cyclic. Then H = H (t, qi, . . ., qm, pi, . . , p.), .
* Using the first n - 1 equations of ( 2 4 " ) i t is possible to express n - 1 Put
coordinates in terms of the remaining hth coordinate and 2 n - 1 arbitrary con-
stants. We then obtain the equations of trajectories i n coordinate space. The
last equation of (24") connects the coordinates with the time variable t .
 142 Canonical Transformations, H a m i l t o n - J a c o b i E q u a t i o n [Ch. 4 Sec. 271 Method o f S e p d r a t i o n o f Variables. E x a m p l e s ,l 4:3

where a,, .: .
, a, are arbitrary constants. Then, by formula ( l ) , and condition (7) always holds. Therefore formula ( 6 ) defines
the constant h may be expressed in terms of the constants ai, . . . a complete integral of the Hamilton-Jacobi equation.
.. ., an as ~ O ~ ~ O W S : In the given case the final equations of motion
h = G (a,, , a,) ... (4)
Solving (3) for
av
-, we find*
841 will be written as*
av (i= l, .. ., n),
-= F i (qi, a i )
aqi

Thus, the final equations of motion are found by quadratures.

and
Example 1. Consider a n oscillator with one degree of freedom. Here

and, thus,
In this case
=$S -a~~ av
---==O for i#k (i, k = l , ..., 12)
aqi aai aai ' aqi a u k and equation ( 2 ) is of t h e form
I and the basic condition
Put h- a ; then
reduces to the inequality
I n
and from the final equations of motion ( 1 1 ) me find the expression for
the momentum.
p = y2n2a--mcq.2 (13)
Since the relation and the equation of motion itself will be written as
fi (4i pi) =a i
is equivalent to the equation
P i = F i (qi, a i ) * I n (5) and (6) by the integral
S P i ( q i , a i ) dqi we mean the integral
it follows that

* We assume that each function f i ( q i , p i ) actually contains the momen-

f
Yi
F i ( g i , a i ) dqi where the constant y i is fixed and is independent of the

values of the arbitrary constants a k ; then

a f . # 0. In this case equation (3) may be solved for
turn p i , i.e. that -2
I)

a ~ i
pi = -av ; each function F i is a function of the two variables qi and a i ;
aqi
i = 2 , ..., n . We then make use of (10).
l
144 Canonical Transformations, Hamilton-Jacobi Equatzon [Ch. 4 Sec. 271 Method of Separation of Variables. Ezamples 145

where Here

Noting that the integral in (14) is equal to arc sin

of motion in the following form:
z, we get the equation for i <k ( i , k = l , .. .,n)
And so the condition (7) reduces to the inequality
2". Let

which always holds since the equation

gi (ai-i, qi, pi) = ai
We introduce the arbitrary variables a,, . ..,an-, and a,=h is equivalent to the equation
and, successively, put pi = Gi (qit ~ i - i . a i )
and for this reason

For what is to follow we need the expressions for the

f3G.
derivatives 2 which
a ~ i -,
~ are found from equations (20) and (21),
Determining from this the partial derivatives, we find* namely:
av
--
84, - G, (gig a,),
av
-=
892 G2 (92, a,. at),
Putting expression (19) into the final equations of motion (11)
and taking into account the formulas (22) and (23), we finally
W hence get
r dqi -

* We ag. f 0, i.e. that pi actually enters into the func-

assume that 2
a ~ .i
tion gi (Qi,p i ) .
** Here and henceforward in this section, the indefinite integral is under-
stood to mean {as in Item 14 the definite integral with variable upper limit and
fixed constant 'lower limit 'independent of c&, . . ., an.
I ,l46 Canonical Transformations, Hamilton-Jacobi Eauation [Ch. 4 Sec. 271 Method of Separation of Variables. Examples 147

Here, the first n - 1 equations of (24) are equations for a family

of trajectories in coordinate space; these equations contain 2n - 1
arbitrary constants a,, . . ., a,, p,, . . ., on-*.
The last equation
of (24) contains a new arbitrary constarit 8, and connects the coor-
dinates q i with the time variable t.
Alter substituting into equations (25) the functions q i ( t , uk, We have obtained the final equations for Koplerian motion.
In sfudying this motion, i t may be taken, without loss of generality,
P k ) (i = 1, . . ., n ) found from (24), equations (25) determine the' that the initial velocity lies in the plane of tne meridian $=const. Then
inomenta p i ( 6 = 1, . . ., n ) as functions of t and all arbitrary
constants a i , Pi (i = 1, . . ., n). a t the initial Lime, *=o
a0
and, hence, by formula (%a),

Example 2. Consider Keplerian motion in which a particle of mass m

is attracted to a centre with a force inversely proportional to the square and $=P,=const, i.e. the motion is plane motion. Differentiating t h e
of t h e distance from t h e centre. equations (26b) and (%c) termwise, we find t h a t the sector velocity* i s
In this case, i n spherical coordinates,

which means that the motion in thc plane $=const occurs in accordance
with the law of areas.
1.
Finally, t o determine the trajectory we put - = X and from (26b),
taking (27) into consideration, we find

and Lhe equation t h a t determines 7' is of t h e form

where

Put Computing t l ~ cintegral wc get

and, hence,
x=k+~~ccos(8-$)
Then, by forrnulas (?h), me find 1.
Finally, recalling t h a t X= - , we have for the trajectory an equation
of a conic section, in one focus of which is the centre of attraction:

* That is, the time derivative of the area described by a radius vector '
drawn from the centre.
io*
148 Canonical Transformations, Hamilton-Jacobi Eauation [Ch. 4 Sec. 271 Method of Separation of Variables. Examples 149

where the parameter and eccentricity of the conic section are determined Take the solution of (33) in the form
from the equations

V= );J S ~i (qi, ai, h) dqi (37)

Let a point describe a closed orbit (the motion of a planet attracted by i= i
the sun). Then this orbit is an ellipse a t one focus or which is the centre
of attraction (sun). Therefore a complete integral of the Hamilton-Jacobi equation has the form
>
Denoting by F and a, b (a b) the area and the semi-axes of the ellipse,
we and (since i t is known that p = E )that

Theii the final equations of motion (11) are obtained by means of quadra-
Let r be the period (of revolution), r = - 2mF . Then on the basis of (30) tures:
vs

where by Newton's law of gravitation, ,

depends solely on the centre of attra-
ction. We have obtained Kepler's three laws of planetary motion about the sun:
(1) the lanets are in motion with constant sector velocity in plane orbits;
these oriits are ellipses in one focus of which lies the sun; (3) the ratio ol t e
squares of the orbital periods to the cubes of the major axes of the orbits is
P'
the same for all planets.
3". By way of further illustrating the application of the method of srpara-
tion of variables, consider the case when Solving the equation f i (qi, pi)-hgi (gi, p i ) = a i for pi, we get pi =
.
.. , n). It is assumed that the following condition of
=F i (qi, a ; , h) (i = l ,
afi h ag.2 f 0. In this case the derivatives of the
solvability holds: --
a ~ i a ~ i
implicit function F i are
Then the basic differential equation
H (h
av
.-.,~ n ,aqi f ...,-)=
alr
aqn h

may be written as and

n
2
i= l
[fi (Si?
av
=) -hgi (qil %)l
av aFi. ([ a/i ' g i
(92, pi)}
pi=Fi(qi, ai. h )
...,
(i=l, n)

Put In final forni? the equations (39) are

where a i , . . ., a,-i are arbitrary constants, and the constant a,, by virtue
of equation (33) and equalities (3%),is expressed in terms of a i , . . ., % - I * dqn=pj (j=l, ..., n-l),
namely
a, = -ai- . . . -an-i (35)

Solve (34) for the derivatives -

av :
aqi
150 Canonical Transformations, Hamilton-Jacobi Equation [Ch. 4 Sec. 281 Canonical Transformations and Perturbation Theory 152

As a special case we get Liouville's theorem:

If the kinetic and potential energies of a system have the form
with function a = 0 [see formula (13) on page 1381

ancl the Hamiltonian systein (3) into the Hamiltonian system with
funclion g , which will be equal to H,: *
dP! -
a'l:
---=-
at
819,
dpj: ' at
--8Hi
840, (5)
(k=l, ..,,n)
..
where Ai, B i and IIi a r e f u n c t i o n s of one variable qi ( i .= l,. , n ) , then the
Thus, the new variables q! and p! possess the following remarkable
final equations of motion of t h e system may be obtair~ed b!l quad~atures. property; Eor unperlurbed motion, they preserve constant values,
Indeed, f o r a Liouville system which are equal to the initial values; for perturbed motion, they
are innctions of the time and of the initial values

defined as a general solution of the Hainiltonian system (5), in

which the "perturbation energy" H , is the Hamiltonian function.
but this is a special case of formula (32).
q i , p z (i-
The final equations for perturbed motion in the initial coordinates
1 , . . ., 1%) are obtained by substituting the functions
(6) into the forrnulas for unperturbed motion (1) in place of the
28. Applying Canonical Transformations to constants q'j, and p!.
Perturbation Theory We hdW! succeeded, through the use of the theory of canonical
Let the motions of a system with a given function 1%be known, transformations, t o replace integration of the Hamiltonian system
t h a t is, we know the solutions (3) by integration of the Haniiltonian systems (2) and (5). We ob-
tain the general solution of the system (3) from the general solutions
(1) and (6) of lhese systems by means of superposition:

of the system of differential equations

and let i t be required to determine the nlotioii of a "perturbed'. Actually, we have shown that "perturbation in the energy" of
+
system with Hamil tonian function H H,; i.e. let i t be required a systern is equivalent to "perturbation of the initial conditions".
This is illusthated in Fig. 40.
to determine the solutions of the system of differential equations
In the hyperplane t = 0, of an extended phase space, take a fixed
point A& and draw through i t an unperturbed straight-line path.
i.e. a straight-line path (1) for the system (2). In Fig. 40, this is de-
I f in formulas (1) we regard q; and p: (k = 1 , . . ., n) as new
variables, then, as has been elucidated on page 139, formulas (1) * This follows from the reldtion g - (H -/- 11,) -= 0 - I I . The left and
B '5
determine a free univalent canonical transformation. This transfor- right sides of thisequation are equal t o , where S is the generating fili~ction
mation turns the Hamiltonian system (2) into a Hamiltonian system of the free univalent canonical transformation under consideration (see p. 133).
152 Canonical Transformations, Hamilton-Jacobi Equation [Ch. 4 Sec. 291 The Structure of an Arbitrary Canonical Transformation 153

picted as the heavy line The light line M & ' , depicts the i t is possible to take the following quantities as 2n independent
displacement, in the hyperplane t = 0, of the initial point given
by the functions (6). From point Mot of this curve draw an unper-
ones:
qi~
-
q17 Pi+i, - - 3 Pn, qi,
d - -
q7n7 Pm+i, - . - , P n
turbed straight-line path MotNot (it is shown dashed in Fig. 40).
On this path take a point P with a given value of the time coordi- (042, m<n) (3)
iiate t. This will then be the position of the system in perturbed
motion a t time t. In the case of unperturbed motion, the system Theii with the aid of the identities

\ we can write the basic defining equation (9) on page 130 in the
I form m n

Fig. 40
where
occupied the position Q a t time t. Thus, the perturbation appeared
i11 the "shift" QP. The straight-line path in perturbed motion is
shown by the heavy line MOP. Thus, perturbed molion may be re-
garded as "compound" motion in phase space: a point is'i11 motion
along an unperturbed straight-line path but the path itself is dis- Siiice all the 4n quantities of (1) are expressed in terms of the 2n
placed (in the general case i t is deformed) due to the "perturbation" quantities of (3), we may take i t that U is a function of the quan-
of the initial conditions. tities of (3). Then from the identity (4) we a t o w e find

29. The Structure OF an Arbitrary

Canonical Transformation
In this and the subsequent sections of this chapter we shall
give certain additional information about canonical transformations.
For a n arbitrary canonical transformation i t is possible to estab-
lish formulas that define i t with the aid of a generating function and
valence c, as was done in Sec. 25 for a free canonical transformation. 1
For example, of the 4n quantities The formulas (6) are equivalent t o the formulas (2) and determine
the canonical transformation under consideration with the aid
of the valence c and the generating function U of the independent
connected by the canonical transformation i quantities (3).
# - I,
Below we shall prove a mathematical lemma, according to which
i t is always possible to choose 2n independent quantities of the
4i = (Pi (t, qk, ~ k ) Pi
, = $i (t, ~ k pk)
,
- -
4n quantities (1) connected by the transformation (2).in such a way *

that there will not be a single pair of conjugate q i ,pi or qi, p i among
I I
 154 Canonicnl Transformations, Ramilton-Jocobi Equation [Ch. 4 Sec. 291 T h e Structure of a n Arbitrary Canonical Transformation 155

-
the quantities chosen. Then if we appropriately renumber the coor- - -
and qi, p i , i t may be assumed that the chosen ones are the following
.
- qi, qk ( i , k = 1 , . ., n) and do the same with the momenta
dinates rt+ d quantities:
- -
p i , p k (i, k = 1 , . . ., n), the chosen 2n independent quantities 41, . - . , q n , ~ i , a . . , qd (d<n) (7)
may be given in the form (3). Therefore, for an arbitrary canonical
transformation there are formulas that can difFer from the formulas We shall .call the quantities of (7) the maximum basis. I t is
(G) solely in the numbering of the quantities of (l).* obvious that
As was done in Sec. 25 for a free canonical transformation, it may - = f ( ~ t ,. -
qn, 41, ... r ~d)r
also be shown that for an arbitrary canonical transformation a deter- Qj 7

minant of order n composed of mixed second-order derivatives of the

generating function U is different from zero.** For this reason, the
- -
-
(¶i,
Pj=f . G'd) - 3
-
Qn, Qi, - 7
M
(j-dd-1, ..
first n equations (6) inay be solved for the quantities q i l E),,(j =
= 1 , . . ., n2; h = nz 4- 1 , . . ., 12). After substituting tlie expres- where f is the functional symbol. *
sion obtained into the last n equations of (6) we cau represpilt the Without loss of generality, i t may be taken tha t of the quan-
equations of the canonical transformation (6) in the form of (2). tit ies
M

Let us state and prove the lemma which we used to obtain the Pi1 . - - , P n r Pi, Pd
structural fornlulas (G) for an arbitrary canonical transfornlation.
', . . ., Qn
M

Lemma. If there are given 2n independent functions that do not enter into the formulas (B), tlie-- quantities
l -
pi, . . ., p, of 2n independent quantities q,, . . .,
7
? M

- - qn, Pi, ., Pn, Pi, m a S 9 Par Pi, . . t pb ( a d n , b<d) (9)

I then out of 4n quantities qi, p 2 , (li, p i (i -- 1 , . . ., n) it is always
possible to choose 2n independent ones in such a manner that there are functions of the basal quantities (7), and each of the quanti-
will not be a single pair of conjugate ones ( q k , pk) or (L, among t.ies
them. - -
Proof. ~ s s u i the
e converse, that is, that any 2n of the quantities Pa+ir Pn, Pe+i, ...',Pd (10)
under consideration, among which there is not a single pair of con- is independent relative to the basis. Thus,
jugates, are always dependent. Then choose n +
d independent
ones of the given 4n quantities so that the first n do not have the
sign M , and the last d have this sign; choose these n 4-d quanti-
ties so that there are no conjugate ones anlong them and so that
the number d has the greatest of all possible values. By assump- C

tion d < n. We shall now show t h a t the quantities q,+,, . . ., qn, qo+i, ..., qd
which are conjugate to the "independent" quantities of (IO), actu-
Since, without altering the conditions or the statement of t,he ally do not appear on the right-hand sides of (8). Indeed, let
lemma it is possible to interchange the roles of the t,wo conjugate
q~ ( h > a) enter into, say, the expression for a certain ( j > d):
quantities q i and p i or and gi
and also t o perform an arbitrary
permutation of the indices 1 , . . ., n both of Ihe quantities yi, p i
* This proposition is given by Carathbodory in his hook Varratiortsrech,nung
und partielle Differentialgleichungen erstes O r d n u n g , 2 Aufl., Bd. I , 1956,
-
Then the system of quantities q,, . , qh-I, qh+i,
c .. .. . , qn, qi,- . . .
. . . , Q, q j is equivalent to the basis (7):
Sec. 96). However, the lemma on which our proof is based [see below) was estab-
- -
lished by Carathkodory for the special case when the transition from the variab-
(~11. . - qh-ir ., (lnr
M

.
- v

.
les qi, p i t o the variables qi, p i (i = 1, . ., n ) is a canonical transformation. qh+i?
v
qi1 qdc qj)
** This condition may be written as follows: det
71
f Oif m(qi1 , qn, 41, (Id) ( l 2)
by ri, . . ., r,,
- -
r i , ..., rn we deiiot,~,respectively, the quantities (3) i n the * I n different formulas. the same letter f is used t o denote different functio-
order given above [see (3)). nal relations.
 156 Canonical Transformations, Hamilton-Jacobi Equation [Ch. 4 Sec. 291 The Structure o f an Arbitraru Canonical Transformation 157

that is, every one of the quantities that appear in one of the two are independent, which fact contradicts the "maximality" of the
systems is expressed i n terms of the quantities of the other and basis (7).
+
vice versa. * Therefore, the n f d l quantities Thus,
-
.
F

Pil = f (Q,,
phl=f (Q,, . - . , q a , 41,
Qar Qit
F
., B)
-
4b) (16)
are independent, which contradicts the "masimality" of the
basis (7).Thus, the formulas (8) may be written as (il = 1 , .. .
,a,, k, = 1 , .. ., b,)
-
-
Denote as . ..
. ., qa2, qbl+,, . , qb2 those of the quantities qi,
qk ( i > a,, k > b,) which actually occur i n the right members of
formulas (16). Then

.
W

Denote by qi, . . . , qal, qi, . . , (ai < a , b, 4 b) all the quanti-

ties that actually appear a t least in one of the right-hand sides
of the formulas (13); then We shall now show t h a t the quantities qa, (h. > a , p > b)
actually do not occur in the expressions (11) for pi, where
i as, k ,< bz. Indeed, let qa (h > a) for example actually appear
in the expression for piz(a, < i2< a,):

Now we shall show that the quantities qn, cl;, (h >a , p > b) But the quantity qi2 actually occurs in one of the expressions (17),
actually do not appear in those formulas (11) where i,< a,, and for instance in the expression for pi,. Then the quantity qi, actu-
Ic < bi. Assume the contrary. For instance, let q~(h > a ) actually ally occurs in one of the expressions (14), for instance in the
appear in the expression (11) i'or pi,, where i, ,< a t , expression for qj ( j >d):

Rut the quantity yi, actually appears in one of the expressions (14);
for example, let i t appear in the expression for ( j > d): In this case an equivalence is established between the system of
quantities

and the basis (7). For this reason, the quantity p* is independent
of the quantities of (18). Adding pn to the quantities of (18) we
and hence the n S-d + 1 quantities get a basis of n + d + l quantities, which is impossible.
Thus,
Pf?= f ( ~ i , -
-
qa, 41,.
-
, qb)
- - (19)
* If qa actually does appear in sorno functional relation, this relation may
pkZ=f (Q,, -
,7 - . ., qb)
~ a 41

bc solved for qa. (i2=qi+1, ..., a2, k z = b i + l , ..., b2)

158 Canonical Transformations, Hamilton-Jacubi Equation [Ch. 4 Sec. 501 Testing Canonical Character of Transformation 159
-
Denote by qa,+i, . . . , qa3, ~b.-ti, . . . , qb3 those of the quanti- so t h a t the transformation defined by these functions should be
ties qi, ;16 (i > a2, k > b,) which actually enter into the formu- canonical.
las (19). Then (19) will be written as Let the transformation (1) ~, be canonical. We write out the
defining identity for it:
n
L: Fiacli-fi6t
i=i
=C (X
i=i
pi6qi - ~ 6 t )- 6 F (t, qi, pi) (2)
Take an arbitrary fixed value t = F. Then from the identily (2)
we get
his
= b,,,
process may be continued until t h e q u a l i t i e s a,
are attained simultaneously; then
- a,,.,, b, =

4 - But (S) is a defining identity for a transformation that does not

Pi f(qi,...,qarrqlr...,qba) ) contain t h e time explicitly,
-
q i z q i ( t , qk, ph), Pi = $ i (t, qh, PA) ( i = 1, . . ., n) (4)
Hence, formulas (4) define a canonical transformation with valence
c which is independent of the chosen value of t = 2.
On the contrary, let i t now be given that all transforn~ations
111 place of the formulas (14), (17), (20), (21) we can write obtained from the transformation (1) by replacing the variable
- .
t by various fixed values of t a r e canonical and with one and the
same valence c . Then, defining the function g by the equatiori

i= i
we get equation (2) froni (3) and (S), i.e. we find that the transfer-
- let -a, > b,. Then from the formulas (23) we can elimiiiate
Now
ination (1) that depends on the time t is canonical.
qi, . . ., qb, and obtain a relatioii between qi and p i ; but this cont- Thus, for the time-dependent transformation (1) to be canonical
radicts the hypothesis of the lemma. it is necessary and sufficient that all the time-independent trans-
If a, ,< b,, then a, < b, +
1. Then from the formulas (22) and formations obtained from the transformation (1) by replacing t
(24) we can eliminate all the qi and ohtain a relation between with an arbitrary value of t h e canonical and with one and the same
and L; this again contradic,ts the hypothesis. valence c.
Thus, the assumption that there exists a maximal basis (7) in For this reason, when establishing tests for canouical character,
which d < n has led us to a contradiction. The lemma is proved. we can confine ourselves t o canonical transformations that do not
contain the time variable t explicitly:
30. Testing lhe Canonical Character of a
Transformation. The Lagrange Brackets
Let us establish certain tests for the caiionical character of a trans-
formation, i.e. the necessary aud sufficient conditions that must For t h e carionical transformation (fj), the defining identity (2)
he satisfied by 2n independent functions [with respect to q k , is written as
p h ( k = 1 , . . ., n)l
-
q i = q i ( t , q k - p k ) , ~ i ~ $ i ( t , q k , ~ k( i)= i r . . . r n ) (l)
 160 Canonical Transformations, Hamilton-Jacobi Equation [Ch. 4 Sec. 311 Simplicial Nature o f Jacobian Matrix o f Canonical Transformation 161

Here, express 6qk i n terms of 6qi and 6pi using formulas (1). this notation and taking for the cyi, qi the functions qi,
Then (7) will take the form
-
-pi Using
(i l , .... n) defined by the formulas (6), we can write the
conditions (10) as

where
where c is the valence of the canonical transformation.
The equalities (12) express t h e necessary and sufficient conditions
for the transformation (6) to be canonical. In the case of a time-
I t remains to write out the conditions that the left-hand side dependent transformation, the conditions (12) are preserved,
of (8) should be differential and we will have a test for canonical provided that they hold for any value of t.
character in the form of the equalities
aoi ao,k= ayi -ayk 31. The Simplicial Nature of the Jacobian
- Matrix of a Canonical Transformation
aqk aqi l a ~ h api '
Consider the Jacobian matrix of the canonical transformation

Substituting the expressions (8') into these equalities we find

(after elementary transformations)

--
a ~ na ~ n
a . .
a ~ n
aqn apt a ~ n

i= i J
8;
Here - is a Jacobian matrix of order n
an . - . . similarly, we
where is Kronecker's symbol: d i k = 0 for i # k; d i k = 1 for a; a; and -
a; .
i = k ( i , k = l , ..., n). define the n-order Jacobian matrices -, -
The conditions (10) may be written in compact form if we intro-
ap aq a~
We introduce a special matrix of order 2n:
duce what are called Lagrange brackets, which, for the given 2n
functions ( p i , $ i (i = 1 , ..., n) of the two variables q and p are
defined as follows:*
...
J=
0
1 ...
0
0
0
0
. . m

...
. . . . . . . . . . . .
-1
0 =L 0 -E
0 ) (2)
* The reader can compare the Lagrango brackets with the Poisson ,0...1 o... 0
brackets introduced in Sec. 15. There, two functions rp, 9 were given of 2n
variables qi, pi and t h e Poisson brackets equaled the sum of the Jacobians
a(T' )'
where E is a unit matrix of order n. Consider the matrix M and also
. Here, however, 2 n functions of two arguments are given and its transpose M'; form the product M',JM and prove that by virtue
a (qi?~ i j
the Lagrange brackets are equal to the sum of the Jacobians (11). of the relation (12) of the preceding section this product is identical-
 162 Canonical Transfornzations, Hamilton-Jacobt Equation [Ch. 4 Sec. 321 Invariance of Poisson Brackets in Canonical Transformation 163'

l y equal t o c a . We shall call the matrix M that satisfies the relation (3) a gene-
M'JM = c J ralized-simplicial matrix (with valence c).*
where c is the valence of the canonical transformation. Since the relations (12) of the preceding section reduced to the
Indeed,* condition of the generalized-simplicial nature of the Jacobian mat-
rix $1, i t follows that the test for the canonic.al character of a trans-
formation may be stated thus:
- -
For a certain transformation qi = qi (t, qk, pk), Pi = Pi (,t, q k 6
M -
pk) (i = 1 , . . ., n) to be canonical, it is necessary and sufficient
that the Jacobian matrix--M corresponding to this transformation,
should be generalized-simplicial with constant valence c. (In the
case of a univalent transformation, the matrix N is an ordinary
simplicial matrix.) Then, the condition of the simplicial nature
of (3) must hold identically relative to all the variables t , qi, p i (i =
= 1 , . . ., n).
But In Sec. 26 i t was established that the motion of any Halniltonian
system may be regarded as a free univalent canonical transforma-
tion. Consequently, its Jacobian matrix is simplicial and has the
determinant I (see pages 125-126) equal to & l .
-+
Since a t the initial time I (0) = $ 1 , so I = 1 a t all subsequent
Performing similar computations for the remaining three blocks, times, but i t is precisely this determinant that serves as the integrand
we net in the expression for the phase volume in 2n-dimensional space ,
[formula (3), Sec. 231. \,
This remark may be regarded as a proof of the Liouville theorem
that differs from the proof given in Sec. 23.
which is what we sought to prove.
For a univalent canonical transformation, equation (3) is written 32. Invariance of the Poisson Brackets in
as a Canonical Transformation
M ' J M = cf (4) Let us give the condition of canonicity of a transformation written
in the form of equation (3) in the preceding section in a somewhat
The matrices M for which the equation (4) holds are called siinpli- modified form. Multiply both sides of this equation by (M')-l on
cial. Since det J -- 1 , and the determinant of a product of matrices the left and by M-l on the right. We get
is equal to the product of the determinants of the matrices being
multiplied, we find from (4) that
detM=+I
Thus, simplicial matrices are nonsingular. ** Simplicial matrices are characterized by the following properties. In the
n
bilinear form f = (xiyi-yixi), we subject the 2n variables xi: y i and the
*
When the elements of the matrix are matrix blocks, multiplication is i= i
perfolmed according to the same rules as when the elements of the matrix are 2n variables xi, y i to one and the same linear transformation with tile
nllmhnrs.. i.e. the rows of the first matrix factor are multiplied by the columns
a------
simplicial matrix of coefficients M. Then the form f preserves its aspect i n
of the second matrix factor. n
** As may readily be verified, the product of two simplicial matrices, the the new variables xf, yf and X?', g:': f= (xtyr'-yTxTr).
c inverse matrix for any simplicial matrix and the unit matrix are again i- i
simplicial matrices. Therefore, simplicial matrices form al r group,
xl~
a simpli- * All generalized-simplicial matrices (for all c + 0) also f o w a group.
Lial group. (Continued on P. 163.) If M is a generalized-simplicial matrix, then det M = +-cn.

11*
 Sec. 321 Invariance of Poisson Brackets i n Canonical Transformation 165
164 Canonical Transformations, Hamilton-Jacobi Eauation [Ch. 4

Take the inverse matrices of both sides of the equation, noting Let us now consider two functions cp and I$ of the quantities qi,
t h a t J-l= -J : * ..
pi (i = l , . , n) and t. Expressing in these functions the qi, Pi
- -
W M ' = CJ ( i = l , . . ., n) in terms of yk, pk ( k = 1, . . . , n) with the aid of a n
-
The equality M J M ' c J is obtained from the equality ICIrJM=
- c J by replacing the Jacobian matrix 31 by its transpose M'.
(2)
inverse canonical transformation, we can regard these same func-- -
tions as functions of the variables qk, pk (k -- l , . . ., n). Accor-
But this substitution [see formula ( l ) , Sec. 311 reduces to repla- dingly, the Poisson brackets of cp, Il, may he evaluated both with
a ayi aFi aFi
cing the derivatives 2 - - - respectively, by the
-
respect to the variables qi, pi [denoted (v$)] and with respect to
the variables qi, pi [denoted (@)-l.
- - aqk a ~ Lk aqk ' a ~ h Let us prove the validity of the identity
derivatives %,
aqi ap*
aq,
9
api api
in other words, in each deriva-
(W)= c (W)-- (6)
tive the letters and indices above and below** are interchanged.
Thus, if the equation M I J M = = c J was equivalent to the system Proof of this identity rests on a familiar espression of the
Jacobian of a system of composite functions
-
I then the equation (2) will be equivalent to the system of equa-
lities
a (%
a(qj, ~
+)
j )
=
i,
n
a ((p, 9) a (qi,
2k=i [a(;,
i< k
a ( q j , Pj)
-
qk)
' a (v, W a h)
a(&&) a(qj7 pj) l+
[qiqkl* ==O, [pipkl* =O, [qiph]* =6ik (i, k = 1, . . . , n) (4)
where the asterisk (*) indicates that the aforementioned interchange
of derivatives is to be performed within the Lagrange brackets.
But then the Lagrange brackets become Poisson brackets. Indeed,
Summing these identities termwise, we get
n n
(v? 1' (pi&)] +2 a 9+ ?) (;iFk)
i, k=i [ a ,) , a (pi7pk) i, k=i a (qir ~ k )
i< k
From this, using equalities (5), we find
n
6
d

where (gi, qk) are Poisson brackets of the functions and qk w i t h

respect to the independent variables g,, p,, . . . , g,, p,,. Similarly,

The converse assertion also holds. If for any two functions rp

Therefore, the conditions of the canonicity of transformation (4) and I$, the identity (7) is fulfilled for one and the same constant
may be written in the following form with the aid of the Poisson c # 0, then the transition from the 2n variables qi, p i to the 2n
brackets :
- - -- variables qi, p i is accomplished by a canonical transformation
with valence c.
(qiqk)=O, (pipk)"O, (qipk)=c6ik (i,k=l,...,n) (5)
For a univalent canonical transformation c = l , and therefore
* Here we use the following rule: the inverse matrix of a product of matrices
is equal to the product of inverse matrices in reversed order: (BBC)-I =
-
- C-'B-lA-l. Besides,
I n other words, the Poisson brackets are invariant under univalent
canonical transformations. This property of univalent canonical
transformations singles out these transformations from among all
that is, J-1 = -J.
** As before, the sign - remains on top. possible transformations of phase space.
 Sec. 331 Logrange's Theorem on Stabi1:ty of Equilibrium Position ?67
CHAPTER 5
I t is convenient to intekpret inequalities (1) and (2) geometrically
in a 2n-dimensional state space (qi, qi). For the case n = 1 , Fig. 41
Stability of Equilibrium and the depicts, in the plane ( q , 4),
two neighbourhoods of the coordinate
origin 0 that correspond to the inequalities (1) and (2). If the origin
Motions o f a System 0 is a stable state of equilibrium and for a given E > 0 some 6 > 0

33. Lagrange's Theorem on the Stability of a n

Equilibrium Position
We begin by defining a stable position of equilibrium.
First recall* that a position of a system is called a position of
equilibrium if the system, which a t the initial time was in that
position with zero velocities, remains in that position.
l
Let the position of a holono~nicsystem be defined by the indepen-
dent coordinates q,, . . ., q, (where n is the number of degrees of
freedom of the system). As was explained in Sec. 5 , in a position
of equilibrium (and only in that position) all generalized forces
are zero: Q, = 0 (i = 1 , . . ., n).** Without loss of generality, we
may take i t that the position under study of the system lies a t the Fig. 41
.
coordinate origin q, = . . - q, = 0. Then the coordinates of any
other position of the system q,, . . ., q, characterize a deviation is suitably chosen, then any motion starting a t time to within a
of this position from the equilibrium position and for this reason square with centre a t 0 and side 26 will occur all the time within
are called deviations of the system. just such a square with side 28.
.
The position of equilibrium q, = . . = q, = 0 (or the state of
Example 1. A heavy ball can move along a rim having the shape of a circle
equilibrium qi = . . . = q, = 0, q, = . . - .
- q, = 0) is called and being located in the vertical plane. There are two positions of equilibrium:
stable if for sufficiently small initial deviations qi and sufficiently the lowest and the highest point of the circle. The former is the stable position
of equilibrium while the latter is the unstable position of equilibrium.
small initial velocities q: (i = 1 , . . ., n) the system does not go Example 2. Linear oscillator. The position of equilibrium is stable. Indeed,
beyond the limits of an arbitrary small (preassigned!) neighbourhood 1 1
for a linear oscillator T = - mq2, I1 = - cq ( m > 0, c > 0), and the differential
of the position of equilibrium during the entire period of motion ..+ 2
and has arbitrarily small velocities qi ( i = 1, . . ., n); that is equation of motion mq cq = 0 has the general ,solution
to say, if for any E > 0 i t is possible to indicate a 6 = 6 (E) > 0 40
such that for all t >, to the inequalities q=qOcoso(t-to)+-sino(t-to)
0
Therefore

* See Sec. 4.
** If the generalized forccs Q i depend not only on the coordinates qk but
also on the generalized velocities qk (k = 1 , . . ., n), then the equalities Qi = 0
must hold when the coordinates qk appearing in the expression for Q iare rep-
laced by the coordinates of an equilibrium position, and the generalized velo-
cities are assumed equal to zero.
< <
only i f ( qo I 6, and I qo I 6 where, for example, 6 = m i n -,
(Z"o 3.
Example 3. A particle of mass m can move along the x-axis under the
action of two forces: a restoring force proportional to the deviation -cx,
and a resisting force of the medium, which is proportional to the first power
>
of the velocity - 2 j i (c 0, f > 0).
The point x = O will be a stable position of equilibrium. We first consider
the case when the coefficient of t h e force of resistance is small: 0< f < v&.
 168 Stability of Equilibrium and the Motions of a System [Ch. 5 Sec. 331 Lagrange's Theorem on Stability of Equilibrium Position 169
S . .
Then the differential equation of motion mx+ 2fx+ cx-- 0 has the general tions of motion and their investigation are extremely involved i n
solution more complicated (for example, nonlinear) problems. Of interest
therefore are criteria of the stability of an equilibrium position
that do not require preliminary integration of the differential
equations of motion of the system.
As far back as 1644, Torricelli knew that the position of a system
of bodies acted upon by gravity would be stable if the centre of
gravity of the system occupied the lowest of possible positions.
From this, for any value of t , Lagrange generalized this principle of Torricelli to the case of arbitrary
m potential forces and established a criterion for the stability of the
equilibrium position of a conservative system:
and Lagrange's theorem.* If in some position of a conservative system
the potential energy has a strict minimum, then this position is the
position of stable equilibrium of the system.
provided that I xo I < 6, and I ioI < 8, where, for example, Proof. Without loss of generality we may assume that all the
coordinates q,, . . ., qn and the potential energy II (qi, . . ., q,)
are equal to zero in the position under consideration; that is,
Q , = . . . - qn = 0, and (0, . . ., 0) = O.** Since in the given
position of the system the function II has a minimum, the general-
But if f 2 fi,
.. . then the general solution of the differential equation of
motion m z + 2 f x + c x = O has the form
ized forces in this position are zero:

X =ci,-v1(t-to)+ ~~~-V?(t-to)
where i.e. the point q, == . . . = q, = 0 is the equilibrium position of the
system. Further, from the fact that the value of II (0, . . . , 0) = 0
is a strict minimum i t follows that in a certain A-neighbourhood
of the equilibrium position
Whence, for any value of t ,

we have a strict inequality

and (q,, . . ., qn) > (0, . . . , 0) = 0 (4)
provided, however, that all coordinates qi do not vanish a t once.
Form the expression for the total energy of the system:
provided that I so I < 6, 1 lOI < 8, where

In Examples 2 and 3, the stability of the position of equilibrium

was established with the aid of final equations obtained by integrat- --

ing the differential equations of motion of the system. These final
equations of motion gave us the dependence of the deviations of qi
and the generalized velocities qi upon the time t and the initially
.
given qj, qq (i = 1 , . ., n). The determination of these final equa-
* This theorem appeared i n Lagrange's MEcanique Analytique (Ist ed.,
1788), but a rigorous proof of the theorem was first given b y P. G. Lejeune Di-
richlet. For this reason, the theorem is frequently called Dirichlet's theorem.
** The potential energy II is determined up t o an arbitrary additive con-
stant. We choose this constant so t h a t the value of II should be zero in the equi-
librium position.
170 Stability of Equilibrium and the Motions of a System [Ch. 5
From the inequality (4) and from the fact t h a t T > 0, if a t
least one of the generalized velocities qi is not equal to zero *
i t follows that, always,
E>O
when (3) holds, provided that all the 2n quantities qi, di
(i = 1, . . ., n) do not vanish simultaneously; i.e. the total energy
E (qi, qi), while vanishing a t the origin 0 of a 2n-dimensional
state space, has a strict minimum (equal to zero) a t this point.
E~E'
I .* I
Fig. 42
Now select an arbitrary number E subject to the sole restriction
0 < E < A; and consider the values of the total energy E on the
boundary of the E-neighbourhood defined by the inequalities
(Fig. 42). Since this boundary is a closed set of points, the continuous
function E reaches its minimum E * on this boundary. Since all
values of E are positive on the boundary of the E-neighbourhood,
the minimum E* also is positive. Thus, on t h e boundary of the
E-neighbourhood
E)E*>O (7)
On the other hand, since the continuous function E vanishes a t
the origin 0 , there will always be a b-neighbourhood of the point
O** such that in this neighbourhood
E<E* (8)
* This is true if the b-neighbourhood of the coordinate origin 0 in the coor-
dinate space does not contain any singular points (see footnote on page 47).
We assume that the point 0,a t which the function H has a minimum. is not
singular. But then there are no singular points in some h,,-neighbourhood of
point 0. We choose h < ho.
** Since the inequality (8) holds i n the S-neighbourhood, S ,< e.
Sec. 331 Laeronae's Theorem on Stability o f Equilibrium Position 174
Therefore if the initial coordinates and the initial velocities satisfy
the inequalities (2), the initial energy E. < E*. But in the motion
of a conservative system, its total energy retains the initial magni-
tude E. and consequently E < E * during the whole time of motion.
For this reason, when a system is in motion, the point depicting
this motion in state space cannot attain the boundary of the E-neigh-
bourhood on which E > E * and all the time lies inside this neigh-
bourhood.
The theorem is proved.
Note two things with regard to this theorem.
Note 1. The Lagrange theorem remains valid for a nonconserva-
tive system that is obtained from a conservative system by the
addition of gyroscopic and dissipative forces. Indeed, first of all
note that the equilibrium position is preserved if gyroscopic or
dissipative forces Q ~ ( qk) ~ ~are, added t o the system. For these
forces
As before, choose for the equilibrium position the coordinate
origin of the state space and assume that among the functions Qv
there is a t least oneOj
such that oj
(0) # 0. Then, due to continuity,
Q j # 0 in the A-neighbourhood of the coordinate origin as well.
But since q, and qk are independent, their values in this neighbour-
hood may always be chosen so that
But'this contradicts the hypothesis of dissipativity or gyroscopicity
of the forces. Therefore, the supposition about the existence of
(0) # 0 leads to a contradiction, i.e. all & (0) = 0, v = 1 , . . ., n ,
and this indicates that the addition of gyroscopic and dissipative
forces does not violate the equilibrium.
Gyroscopic forces do not upset the law of conservation of total
energy (see Sec. 8) apd for this reason the whole proof of the Lagrange
theorem remains unchanged in the case of gyroscopic forces as well.
I n the case of dissipative forces, the total energy E = 5" +n
diminishes as the system moves, and consequently when the system
is in motion we have the inequality E ,< E. in place of the equality
E = Eo. But from this i t also follows that throughout the motion
E < E*, if E. < E*. Thus the proof of the theorem is preserved
here too with that slight modification.
172 Stability of Equilibrium and the Motions o f a System [Ch. 5
Note 2. The equilibrium position of a conservati,ve system will
also be stable when the potential energy II in this position has
a nonstrict minimum, but in any &-neighbourhood of the equilibrium
position there exists a closed hypersurface
that contains the equilibrium position within it and has the property
that on this hypersurface the values of the potential energy are
strictly greater than the value of II in the equilibrium position.
Indeed, as before, in the equilibrium position let q, = . . . - -
= qn = 0 and II (0, . . ., 0) = 0. Also, let the equation of the
hypersurface (9) be chosen so that for points located inside the
closed hypersurface (9) the following inequality holds:
Then this inequality, together with the inequalities
defines, in the 2n-dimensional state space, a region (a finite "hyper-
cylinder") G located inside the &-neighbourhood (6). On the boundary
of the region G, either f = 0 (then II > 0, T >, 0) or a t least for
one k the equality I 6, I = c k holds (then T > 0 , II >, 0). Therefore,
the strict inequality E = T $ II > 0 always holds on the boundary
of the region G.
In the case under consideration, the minimum of the function E
on the boundary of the &-neighbourhood (6) can be zero. Then when
proving the Lagrange theorem, in place of the &-neighbourhood
one has to take the region G situated inside it. On the boundary
of the region G the minimum of total energy E* > 0. From then
on the remainder of the proof remains without change.
For n = 1 the closed hypersurface (9) degenerates into a collection
of two points on the q-axis located on different sides of the origin 0 ,
while the region G degenerates into a rectangle situated inside
the &-neighbourhood of the point 0 (Fig. 43).
The reasoning given in Note 2 holds also for the case when gyrosco-
pic and dissipative forces (see Note 1) are additionally applied to
the system.
If the points a t which the function II has a minimum II = 0
fill the solid curve emanating from the equilibrium position, then
this equilibrium position can also be unstable. To illustrate, take
a free particle with potential energy that does not contain one of
the coordinates, for example X: II = II (y, z), and II (0, 0) = 0
and II (y, z) > 0 for y2 + 22 > 0. In this case, the minimum points
fill the x-axis. The equilibrium position X = y = z = 0 is unstable,
Sec. 341 Instability Criteria of Equilibrium Position 173
since for an arbitrarily small (in magnitude) initial velocity along
the x-axis, the point will execute uniform motion along the x-axis.
In the Examples 1 and 2 given on page 167, a conservative system
was considered, and in Example 3 (page 167) a dissipative force
acts on a particle as well. The potential energy has a strict minimum
Fig. 43
in Example 1 a t the lowest. point of the neighbourhood, and in
Examples 2 and 3 a t X = 0. For this reason, these eqnilibrium posi-
tions are called stable.
Example 4. A conservative system with one degree of fr~cdomhas a poten-
1
tial energy 11 = q 4 sin2 -[we also define 11 (0) = 01. In accordance with Note 2,
4
the position q = 0 is a stable position of equilibrium.
34. Criteria of Instability of an Equilibrium Position.
Theorems of Lyapunov and Chetayev
As far back as 1892, A. M. Lyapunov, in his celebrated disserta-
tion entitled "The General Problem of Stability o.f Motion", posed
the question of the converse of the Lagrange theorem. The problem
has still not been solved completely. A partial solution is given
by two theorems of Lyapunov and a theorem of Chetayev, in which
certain sufficient conditions are established for instability of an
equilibrium position.
As before, let qi = . . . = qn = 0 and TJ (0, . . ., 0) = 0 in an
equilibrium position. Write an expansion of the potential energy
in a power series of coordinates ("deviations'):
 Stability of Equilibrium and the Motions of a System [CA. 5 Sec. 341 Instabilitu Criteria o f Equilibrium Position 175
174

.
where IIh (gi, . ., g,) is a homogeneous function of the kth degree Consider the auxiliary quadratic form
+
(k = m, m 1 , . . .), and the lowest of the powers of terms actual-
l y appearing in the expansion is m > 2, since all (%) = 0 in the
dqi o
position of equilibrium.
Lyapunov's first theorem. If the potential energy II (gi, ., g,) ..
of a conservative system in a position of equilibrium does not have where ~ k = l for a h > 0 and ~ h = - 1 for a h < 0 ( k = l , ..., n),
a minimum and this circumstance may be seen from the second- and the number p>O.
degree terms 112(g*, ..
., g,) of the expansion (l)*, then, the given I t is verified directly t h a t , by virtue of equations (3) and equa-
position of equilibrium is unstable. lity (4).
Proof. In the expression for the kinetic energy T = 2-
1
n
2 ..
aihqiqh,
n

i, k=i
expand the coefficients a i k (gi, . . . , qn) in a power series of coordi- when a system is in motion.

n
..
-;- -
nates and denote by afk the absolute terms (i.e. the values of t h e
functions aik (g,, . . ., g,) for g, . . . = q , 0). Then, putting
Without loss of generality, we assume that a, < 0, and t h a t
is the largest of the negative numbers Ak. The positive number p
To=
1
2 a;hqiqlz, we will have
is chosen so that the following inequalities hold simultaneously:

From the first inequality i t follows that the sum of the right-
here and henceforward we shall use the symbol (*) to designate hand side of (5) is a positive definite quadratic form. But then,
the sum of the terms having a higher order of smallness relative for €lk and i)k sufficiently small (in absolute value),
to the coordinates and velocities than the terms written out earlier.
Since To is a positive definite quadratic form with constant coef-
ficients, i t is possible, by means of a nonsingular linear transforma- the right-hand side of (5) will always be positive, i.e.
tion of the variables, to simultaneously reduce the two quadratic
forms To and nz to a sum of squares, after which the expansion
for T and II in the new variables €li, . . ., 0 , will take the form*"
whence

or

Since the quadratic form ]IT, assumes certain negative values, a t
least one A,+ <0.
The Lagrange equations may be written in the coordinates 8k
thus
* That is, m = 2, and n2is a quadratic f o m assuming negative values (per-
haps along with positive values).
** The coordinates et, . . ., 0, are called normal or principal coordinates.
They will be considered in more detail in Sec. 41.
Put all the initial values Q;, 0; (k = 1 , . . ., n) equal to zero,
with the exception of O;, which we shall take, in absolute value,
less than A . Then, using expression (4) and the second inequality
of (6), we find V. > 0. But in that case the motion will definitely
go beyond the limits of the neighbourhood of (7), however small
/ 0: 1 may be, since otherwise i t would follow from inequality (8)
that lim V = oo, while the quadratic form V in the neighbourhood
t-rm
of (7) is bounded.
The theorem is proved.
176 Stability of Equilibrium and the Motions of a System [Ch. 5 Sec. 351 Asymptotic Stability of Equilibrium Position 177

For the case when m > 2, in expansion (1) we can use the following are satisfied. In a geometrical interpretation (Fig. 44), this signifies
two theorems that are given without proof.*
Lyapunov's second theorem. If the potential energy ll of a conserua- that in the state space (qi, qi) all trajectories emanating from the
tive system for q, = . . . = q, = 0 has a strict maximum and this do-neighbourhood of the origin 0 asymptotically approach t h e point
circumstance may be determined by proceeding from the .terms of lowest 0 as t - t m .
.
power of H, (q,, . ., q,) (m >,2) in the expansion (l),** then the Of the Examples 1 , 2 and 3 examined on pages 167-168, only in
Example 3 is the stable equilibrium position asymptotically stable.
position q, = . . . = q, = 0 is an unstable position of equilibrium of
the system. We shall consider scleronomic systems under the action of the
~ h e t a ~ e v 'theorem.
s If the potential energy ll of a conservative an
potential forces -- and the nonpotential forces Gi . .
(i = 1, . ,n)
system is a homogeneous function of the deviations g,, . . ., q,, and aqi
in the position of equilibrium qi = . . . - - q, = 0 does not have
a minimum, then this equilibrium position is unstable.
Example 1. Let II=A(I-cosaq); n = 1 . Tllc function II has strict minima
2kn
a t the points qzk=- and strict maxima (k=O, % 1, _t 2, ...) at the points
a
(2k- l ) n
Yzk-I =
- m, . The last circumstance follows from the form of the
lowest-order term in the expansion i n powers of the deviations

Then, according to the theorems of Lagrange and Lyapunov, to the points assuming that the potential energy and the nonpotential for-
qzk there correspond stable equilibrium positions, and to the points qzk-l,
unstable ones. ces Qi do not depend explicitly on the time:
Example 2. II = Aq, . . . g,. From Chetayev's theorem it follows that
the position qi = . . . = q,, = 0 is an unstable position of equilibrium.
In this case the time t does not enter explicitly into the Lagrange
35. Asymptotic Stability of an Equilibrium Position. equations, which m a y be written in the following form (solved
Dissipa ti ve Systems for the generalized accelerations) (see Sec. 7, page 47):
Now introduce the concept of an asymptotically stable position
of equilibrium. An equilibrium position is termed asymptotically
stable if it is stable and if, besides, for sufficiently small (in absolute
value) initial deviations and initial velocities, all deviations and In the case under consideration, the total energy E of the
velocities tend to zero as the time t increases without bound, i.e. scleronomic system does not contain the time explicitly:
if there exists a number 6 0 > 0 such that
lim qi (t) = 0, lim qi (t) = 0
t+w t+m Computing its total derivative with respect to the time when the
(i=l, ..., n) system is in motion, we find
every time the inequalities

* The reader will find the roofs in the books bv" L "v a ~. u n o v1161.
- - Chetavev
.
Thus, a t every point of the state space (gk, qk), not only the
[5], and Malkin [18].
** That is, in a certain neighbourhood of the origin (excluding the origin total energy but also its total derivative with respect t o the time
.
itself) always II, (gi, . ., g,) < 0. This is only possible if m i s even. has a definite value.
4 78 Stability of Equilibrium and the Motions of a System [Ch. 5 Sec. 351 Asymptotic Stability of Eauilibrium Position 179

..
If the forces Qi (i= l, . , n) are dissipative forces (see Sec. 8), neighbourhood in which condition (8) of Sec. 33 (a0,<6) is fulfi-
dE
then =,<0 when the system is in motion, i.e. in the region of lled. We consider one of the motions initiated in the do-neighbo-
-- - 0
urhood. Since, given motion, the energy E diminishes, i t follows*
state space under consideration the fur~ction E' (qk, qA) assumes [.hat
on1y nonpositive values. lim E (t) = E , >/ 0
In the case of a definite dissipative system (see Sec. a),
l-tm

xt
dE
= E' (qi, qi) vanishes only a t those points of the state space
and E (t) 2-Em (t 2 to). First assume that E , # 0, i.e. E, > 0, and
consider the sequence of the values of the time t, -+oo and the
(gi,qi) where all qi = 0, i = l , . .., n. We shall assume that the sequence of values of the phase coordinatss

Since the entire trajectory lies in the &-neighbourhood, the fol-

lowing inequalities hold for all S and i:

By virtue of the Bolzano-Weierstrass lemma, i t is possible, from

the infinite bounded sequences qjs) and qy), t o choose the subse-
quences qjk) and #). Let
Fig. 45

equilibrium position of the system is isolated, i.e. there are no

lim q!"
k-tm '
= qt, lim
k-tm
=,
:; i = 1, . ., n (7)
other equilihrium positions in its neighbourhood. Then we have where
the following:
Theorem o n asymptotic stability. If the potential energy H of
a scleronomic definite dissipative system has a strict minimum in some But then, because E (t) is continuous,
equilibrium position that is isolated, then the equilibrium position
is asymptotically stable.
Proof. Again let
0

q i = q 2 = . . . =qn=O and I I ( 0 ) = 0
in the equilibrium position. As in the proof of the Lagrange theorem,
choose in the state spaLe an &-neighbourhood of the coordinate origin
0 in which the energy E is positive,
a t all points different from 0 and in which there are no equilibrium
states different from 0.
Since according to the Lagrange theorem, the position of equ-
.
ilibrium g, = . . = q, = 0 is stable, for any E >0 i t is possible
to indicate a G(&)> O such that all motions occur inside the
&-neighbourhood of the point 0 if the initial point is chosen in
the G-neighbourhood (Fig. 4 5 ) . For the Go-neighbourhood take the
By hypothesis, the point (qT, q5) does not coincide with the coor-
dinate origin, where E = 0.
..
Let us take the point (qf, g?), i = l , . , n for the initial point
ol motion when t = to. Since this point does not coincide with the
point 0 , i.e. i t is not an equilibrium position, then when the
system is in motion a t least one of the generalized velocities ii
will be different from zero and for this reason - d E (0. Rut then
dl
for some t = t, the inequality E < E , will hold.
Further, let us consider the motion of a system starting a t
t = to from tho point (qlh), g/h)).
By virtue of (7), the values of
* This limit exists by virtue of the fact that E (t) i?:a continuous
monotonic decreasing nonnegative function.

180 Stability of Equilibrium and the Motions of a System [Ch. 5
and for sufficiently large k, will be arbitrarily close to the
values of qr and qr respectively. Consequently, for sufficiently
large k, the values of the phase coordinates for t = t, will also be
arbitrarily close i n the case of motions starting, when t = t o ,
from the initial points (qj", #') and (qf, h)
(the solutions of the
systems of differential equations are continuous functions of initial
data). Therefore, at t = t, the inequality E (t,) <E, will hold for
a motion that started, when t = t o , from the point (qe), qj*) since
the total energy E is a continuous function of the phase coordi-
nates. But by virtue of the uniqueness of the solutions of the
Lagrange equations, the state (at time t = t,) of a system that
a t t = to started from a n initial point (qik), qik)) i = 1, . . ., IZ,
coincides * with the state a t time tk+t, of a system which a t
t = to started from the initial point (qlo), qjo)), i = l , . ., n; for.
+
this reason, the value of the energy E (tk ti) that interests us
must satisfy the inequality
I +
E (tk 4 ) <E m
1 but this is impossible, since E (t)> E, for any t >to.
We have thus arrived a t a contradiction by assuming that
E, # 0; hence,
E, = O and l i m E (t) = O (8)
t+m
Since, by virtue of the inequality ( 6 ) , the equality E = 0 occurs
only a t the point 0, it follows from (8) that as t -+ oo the point
depicting the system in state space tends to the coordinate origin,
which means that the relations (1) hold. The theorem is proved.
1 .
In Example 3 on page 167, E --5- mxz + -X1 czz and
since f > 0. The system is definite dissipative and the equilibrium position is
isolated; this follows from the equation of motion when substituting the solu-
tion X = const. By virtue of the theorem, the position of equilibrium is asymp-
totically stable.
* Since the time t does not appear explicitly in the equation of motion (3).
choice of zero time is immaterial. Therefore, if in place of qin), $2) we take for the
initial state q y ) , i = 1, ..
., n, then the system will subsequently pass
through the same states as in the initial motion, but a t different instants of
time.
Sec. 551 A s u m ~ t o t i cStabilitu o f Eauilibrium Position 181
An investigation of the motion of a scleronomic definite dissipative
system in the neighbourhood of its asymptotically stable equilibrium
position will be given in Sec. 46.
Lyapunov proved a theorem that generalizes the Lagrange theo-
rem. He noticed that in proving the Lagrange theorem 'it is possible,
in place of the energy E , to take any continuous (with continuous
first-order partial derivatives) function V (qk, i k ) which has a strict
minimum in the equilibrium state and which does not increase
under any motion of the system.
Compute the time derivative of the function V by taking advantage
of the equation of motion (3):
One thing that follows from this formula is that the function V'
vanishes a t the coordinate origin 0 of the state space, since the
.
point 0 corresponds to the state of equilibrium in which all the
.
qi = 0 and all the qi = Gi= 0. If the functisn '1 does not increase
dV
in any motion, the-a t = V' (qk, qk) \< 0. In this case, the function
V' has a maximum in the state of equilibrium 0. Now if this maxi-
mum is strict, then in the neighbourhood of the point 0 (with the
exception of the point 0 itself) V' < 0 , and when the system is
in motion within the limits of this neighbourhood, the function V
decreases strictly.
I t is now possible to repeat almost literally the proof of the
Lagrange theorem, using the function V in place oi E . In the case
of asymptotic stability (for example, for a dissipative system), the
proof will even be simpler... if we demand that in the equilibrium
dV
position the derivative have a strict extremum of a type opposite
to the extremum of the function V.*
Note also that i n formulating the stability criterion, the words
LLminimum"and "maximum" can be interchanged, since substituting
the function -V for the function V returns us to the earlier formula-
tion.
Thus the following theorem has been proved:
Theorem. If there is given the equilibrium position of a sclerono-
mic system that is under the action of forces which do not depend
* In the proof of the theorem on the stability of a dissipative system, the
dE
derivative - did not have a strict maximum in the equilibrium position. In
at
this connection we had to stipulate that the equilibrium position was isolated.
182 S t a b i l i t y of E q u i l i b r i u m and the Motions of n S y s t e r ~ ~ [Ch. 5 S e c . 361 Coaditionnl S t a b i l i t y 183

explicitly on the time, and there exists a function V (q,, q,) that
is continuous together with its first partial derivatives and that
has a strict extremum in the given state of equilibrium, whereas
the derivative V' of V with respect to the time (computed by the
equations of motion) has in this same state an extremum of the
opposite type, then the equilibrium position under consideration
i s stable. If the extremum of the derivative is also strict, then the
equilibrium position is asymptotically stable.
The function V (g,, q,) spoken of in the theorem is generally
called the Lyapunov function.
36. Conditional Stability. General Statement
of the Problem. Stability of Motion or of
an Arbitrary Process. Lyapunov's Theorem
Inequalities (l) and (2) (see page 166) that define the stability
of an equilibrium position involved all the deviations qi and all
the generalized velocities qi. However, in many problems we encoun-
ter a conditional stability when the indicated inequalities hold
for certain of the 2n quantities qi, . . ., q,, ql, . . ., q, or, in a
more general statement, for certain functions xi, . ., X,, of these
quantities:
xi=fi(¶kr¶k) ( i = L . . . , m )
I t is also assumed that all the functions (1) vanish for
qk-0 ( k = l , ..., n), i.e. fi(O, ..., 0 ) = 0
and satisfy the autonomous * system of ordinary first-order
differential equations **
Such a solution presupposes that the right-hand sides of equa-
t,ions (2) satisfy t,he condition
From the mathematical point of view it is a question of the stabi-
lity of a zero solution of the system of differential equations (2),
this stability being defined as follows: for any E > 0 there exists
a 6 = 6 (E) > 0 such that
for any t 2 to
I x i ( t ) I < ~ (i==l, ..., m ~ ) (5)
a s long as
Ixi(to)/<6 (i=1, .s.,uc)
For a geometric interpretation of the inequalities (5) and (6),
use is made of the E- and K-neighbourhoods of the coordinate origiii
in the m-diinensional space (xi, . . ., X,,). In the case of asymptotic
stability we also require the existence of a 6, > 0 such that
.
If the stability (not conditional!) of a n equilibrium position
(1) is under study, then for xi, . . ., X,, we can take the quantities
. . . .
81. . ., q., qv . ., L or qi, . ., q., p i , . ., P?. 1" the former
case, the equations (2) are the Lagrange equations wrltten as a system
of 2n first-order differential equations in the unknown functions
.
q!, . ., p., I n the latter case, equations (2) are the canonical equa-
tions of Hamilton:

where Xi (xi, . . . , X,,,,t) (i = l , .. . , m) are continuous functions in

Let us consider two important special cases of the system of
the region
equations (2) which are frequently encountered i11 applications.
Ixil<A, t > t o (i==l, ..., m) (3) l". Stationary case, when t does not appear explicitly on the
(to is a fixed initial instant of time). right-hand sides X i of equations (2), that is, when
To the equilibrium state there corresponds a zero solution
xi = 0 (i = l , . ..
, m) of the system of differential equations (2).
* See footnote on page 81.
2". Periodic case, when the right-hand sides X ihave a period
** In investigating the conditional stability of scleronomic systems, the T
functions X i(i = 1, . . ., m) do not depend explicitly on t . We have put t with respect t o the variable t:
on the right sides as arguments, having in view nonscleronomic systems and
subsequent generalizations.
184 Stability of Equilibrium and the Motions of o S!lstem [Ch. 5
In both cases the stability of the zero solution of the system of diffe-
rential equations (2) is determined by a theorem that is a direct
generalization of the theorem given a t the end of Sec. 35.
Lyapunov's Theorem. If in the stationary or the periodic case
there exists a function V (xi, . . ., X,, t) continuous together with
its first partial derivatives i n the region (3), which function, for any
t considered as a parameter, has a t the point xi = . . - X, = 0 . Note that in proving the Lagrange theorem for a conservative
a strict extremum, whereas again at the same point, and for any t,
n Now let us consider motions or more general processes described
its time derivative V' (xi, . . ., X,, t) 2 aya v Xi av
= has a n extre-
i= i
mum of opposite type, then the zero solution of the system (2) is stable.
Now if the extremum of the derivative V' is also strict, then the zero
solution of the system (2) is asymptotically stable. Here it is assumed
that in the stationary case the function V is not expliciily dependent
on the time t, but i n the periodic case this function is periodic with
respect to t with period T [T is the period of the right-hand sides of
equations (2)l.
I t suffices to prove this theorem for the periddic case since the
stationary case may be regarded as a special case of the periodic
with any T. The proof consists in repeating the reasoning given
earlier in the proof of the Lagrange theorem and the theorem on
asymptotic -stability with the following modifications: in place of
E we now use the difference
and in place of the space (qi, qi) we take the m-dimensional space
(xi, . . ., X,). We regard t, which appears in V, as a parameter.
This parameter, by virtue of its periodicity, may be varied in the
finite interval to ,< t ,< to + T. Because of this circumstance, the
presence of the variable t in V does not give rise to any complica-
tions in the proof of the theorem.
Note that in the general (nonstationary and nonperiodic) case,
more rigorous conditions [see 51 must be imposed on the Lyapunov
function.
We note one special case of the Lyapunov theorem, which is
frequently used as a test for simple (nonasymptotic) stability.
Let the function V (xi, . . ., X,, t) be the integral of the system
of differential equations (2), i.e. the function V becomes a constant
upon substitution into i t of any solution of the system (2). I n this
---
case -
dV
= V' (xi, . . ., X,, t) 0, and it may be taken that the
dt
function V' has a maximum and minimum (nonstrict, of course)
a t the point xi = . . . - - X, = 0 for any t. For this reason, we
have the following corollary to the Lyapunov theorem:
Sec. 361 C o ~ ~ d i t i o n oSl t ~ b i l i t ~ 185
Corollary. If the system of differential equations (2) has an
integral V (xi, . . ., X,,, t) [which is not dependent on t in the
stationary case and is periodic with respect to t with period T in
the periodic case] and this integral has a strict extremum a t the
point xi = . . . -
- X,, = 0 for any fixed t, then the zero solution
of the system (2) is stable.
system, use is made of the energy integral E.
by the system of first-order differential equations
where the right-hand sides are continuous functions in some region
.
of variation of the variables zi, . ., z,,, for t >, to that satisfy
the conditions of existence and uniqueness of solution on the basis
of the initially given zi (to) (i = 1 , . . ., m).
Let (t) (i = 1 , . . ., m) be a solution of the system of equations
6
(10) defining some process. To elucidate the question of thestability
of this process, we introduce new unknown functions or "deviations",
in place of the unknown functions zi, . . ., z,:
Then in the new variables, the system of differeatial equations (10)
will he written in the form of the system (2), where
The zero solution xi = . . . - - X,,, = 0 of the system of differential
equations (2) corresponds to the solution zi = Li(t) (i = 1 , . . ., m)
of the system of differential equations (10) in the new variables.
This circumstance permits the question of the stability of the
process ii(t) (i = 1 , . . ., m) to be reduced to the problem, which
we have already studied, of the stability of the zero solution of the
system of differential equations (2). In other words, the solution
zi = zi (t) (i = 1 , . . ., m) of the system (10) is called stable (or
correspondingly, asymptotically stable) if the zero solution
xi = . . . -
- X, = 0 of the system of equations in deviations (2)
with the right-hand sides defined by the formulas (12), is stable
(or correspondingly, asymptotically stable). All this opens up a
wide field for application of the Lyapunov theorem given in this
section. This theorem may be used to determine not only the stability
of an equilibrium position, but also the stability of motion, i n
general, of any process defined by a system of ordinary differential
equations.
181i Stability of Equilibriam and the Motions of a System [Ch.-
5 Sec. 361 Conditional Stability 187

Example 1. Consider the inertial rotation of a solid having a fixed point 0. (the axis of dynamical symmetry) of the shell is defined by two angles: angle a
Euler's dynamical equations in this case have the form formed by projection exi of the axis CE on the Cxy-plane with axis Cx, and
angle between Cx, and C%(Fig. 46). By successive rotations through the angle
a and through the angle p, the trihedron of axes Cxyz passes into the trihedron
CF,qc. A supplementary rotation through the angle cp about the axis C t carries
!<ere! p , q, r are projections of the angular velocity w on the principal axes of
mertia of the solid, OE, Or),05; and A , B, C are the moments of inertia about
these ases.
Equations (13) allow for the following three particular solutions defining
permanent rotations of the solid about the principal axes:

We confine ourselves to working out the stability of rotation 1°, since 2'
and 3" may be written as l0with the axes labelled differently. Here, the stability
of the solution l0 of Euler's equations (13) will determine the conditional sta-
bility of rotation of 1" with respect to the angular velocity o.*
Form the equations in the deviations, putting xi = p - p o , x2 = q, 2 3 = r :

Fig. 46
Let the major.or minor axis of the ellipsoid of inertia be along the axis 05
of the permanent rotation under study. The fact that A , B, and C are inversely the trihedron CtqG into the trihedron of axes fixed in the shell. Therefore, the
proportional to the squares of the axes of the ellipsoid of inertia means that angular velocity CO of the shell consists of three components:
A < B. C or A >B, C. For the Lyapunov function, let us take the function
v=(Ax:f ~xe+Cx;f 2Apoxi)azk [B (B-A) zg+C (C-A) X!]
where oi = a, oz=P and ~ 3 (P. = The projections of the angular velocity on
whcre the plus sign is taken for A < B, C and the minus sign for A > B, C. the principal axes of inertia Ct, C q , CL are defined by the formulas
The function V vanishes at X, = x2 = xa = 0 and is positive in the neigh-
bourhood of this ~ o i n t .that is. the function V has a strict minimum at this
point. On the other hand, it is readily verified that = 0 by virtue of the
UL q=-p, r=acosp
equations (14), i.e. the function V is an integral of the system of differential
equations (14). For this reason, in accord with the corollary to the Lyapunov Denoting by A the axial and by B the equatorial moment of inertia of the
theorem. vermanent rotation about the maior or minor axis of an elliosoid of shell, we get the following expression for the kinetic energy:
inertia isAstable. 1 4
I t is also possible to show that permanent rotation about the mean axis
of an ellipsoid of inertia is unstable, but this would require use of the instabi-
T = -2[ A ~ ~ + B (qz+rz)] =-2 [A (i+& (fb+&~ cos2 p)]
sin P ) ~ + B
lity test of Chetayev [5]. We shall assume that besides the force of gravity there is applied, a t point D
Examole 2. Bv wav of illustration. consider the oroblem of stabilitv of on the axis of the shell (in the %entre of pressure3?),the force of air drag R,
rotationai motion"of a; artillery shell. ' which is constant* and directed opposite to the velocity w , i.e. in the negative
To simplify the problem, assume that the centre of gravity of the shell C direction of the x-axis. Let 1 = CD, and y be the angle between the axes Cx
is in rectilinear motion along the horizontal x-axis with constant velocity and Ct. Then the moment. of the force R about C is equal to R1 sin y , and the
v=const. Let Czy be the vertical plane of firing. The position of the axis CF, elementary work of the force R will be
8 A =R1 sin y6y = -6 (R1 cos y)
* Stability with respect to o means that a small change in the initial angu-
lar velocity ooinvolves a small change in the vector o throughout the motion.
In other words, this is stability with respect to deviations xi = p - pot z, = * R is a function of v, R =f (v), and so from v=const it follows that R =
= q, 1 3 = r. Naturally Euler's angles steadily increase here.
= const.
188 Stability of Equilibrium and the Motions of a System [Ch. 5 Sec. 371 Stability of Linear Systems l89

therefore, for the force potential we can take the function*

Cancelling out B and rearranging, we get
I I = R l cos y = R l cos a cosj3 Bh2-Aphf Rl<O
Oscillations of the axis of the shell are characterized by change in the angles
a and j3. To determine the stability of rotational motion of the shell we will For the last inequality to hold for r\ certain real h, it is necessary that the
proceed from three integrals of motion: quadratic trinomial on the left-hand side of this inequality have real roots,
i.e. there must be the inequality

The first integral is the energy integral; G, and GE are projections of the
>
A2p2 4BRl
kinetic moment G , on the axes Cx and Ct. The constancv of G,, when the system This is the condition that ensures the stability of rotational motion of the
is in motion, follows from the fact that the moment of the force R about the
Cx-axis is zero. The third integral expresses the constancy of the generalized shell (for "deviations" a , P,&,B) because if this condition is fulfilled it is pos-
momentum Ap, which corresponds to the cyclic coordinate cp. Note that (see
Fig. 46)
. .
sible to select a real value of h for which the integral Wi - kWz will have,
at a = j3 = a = j3 = 0, a strict minimum equal to zero.
+
G, =GE COS ("g) +G,, cos ( q ) +G% cos (X<)= Ap cos (IQ Bq cos (xq) +
+ +
Br cos (xc) = Ap cos a cos $ B ($ s i n a -U cos a cos 3j s i n p)** 37. Stability of Linear Systems
Combining the first two integrals with the third and using the expressions In the preceding section i t was shown that the investigation
and W z that vanish for a =P =a = P =0:
.
obtained for T, II, and G,, we find the following two integrals of motion W,
. of the stability of any process defined by a system of ordinary
differential equations* reduces t o investigation of the stability of
1 .
+
W, =7 B (a2 cos2 $ $2) +R1 (COSa cos j3-1) = const, the zero solution of a system of deviation equations.
Let the differential deviation equations be linear and have constant
+
FV2= B ($ sin a-a cos a cos P sin j3) Ap (cos a cos j3-1) = const coefficients
n
We shall seek the fixed-sign linear combination of integrals of motion
W,--kW2. First, determine in W , and W Z the terms of lowest degree in the
m .
small quantities a , j3, a, p:
We shall seek a. special solution of this system in the form
. .
W,=B(j3a-aj3)-TAp(a2+j32)+
1
...
then
Putting expressions (2) into equations (1) and cancelling out eht,
W , --hW,=
1
-
2
+ (Aph-RI) p]+
[B&+ ~ B X & we get relations that connect the desired quantities ui and A:

So that each expression in the square brackets should be positive definite,

i t is sufficient that the following inequality be fulfilled:
<
B2h8 B (Aph- RI)
* On a sphere with centre at C, the arcs a , P, y form a rectangular spherical
triangle with "sides" a , j3 and an hypotenuse^ y. For such a triangle, the formula where is Kro~iecker's symbol
cosy = cos U cos p holds. The truth of the formula follows from elementary
geometrical reasoning. (tjik=l at i = k , dik=O at i#k)
n
** The arcs a, -+B
2
and (xc) form a rectangular spherical triangle with Since in the sought-for solution (2), a t least one of the constants ui
"sides" a and -+B,
n
therefore
must be different from zero, the determinant of the system of homo-
2
* It is known that such a system may ordinarily be written as a system of
first-order differential equations solved for the derivatives. This notation is
always possible for a system of Lagrange equations (see Sec. 7).
190 Stability of Equilibrium and the Motions o f a System [Ch. 5 Sec. 371 Stability o f Linear Systems 191

geneous equations (4) must be zero: The column U +=

0 that, together with the number h, satisfies
the relation (3') is called the eigenvector of the matrix ,4 correspond-
ing t o the characteristic number h. Thus, in each solution of the
system (l'), having the form (2'), h is the characteristic number
of the matrix A , and u is the corresponding eigenvector.
We first consider the case when the characteristic equation (5)
.
has n different roots A k (k = 1 , . ., n). To each characteristic
Thus, for determining h we obtained an algebraic equation i n h number h k there corresponds an eigenvector u k and a particular
of degree n. solution of the system (1) of the form ukehki. A linear combination
Equation (5) is called the characteristic or secular equation for of these solutions with arbitrary constant coefficients
the matrix of coefficients
11
aft ai2 ain - II
n

1
A ,a2.i a22 . . . a2n

:a: in;:. in: 1

The roots of the characteristic equation are called characteristic
numbers of the matrix ,-L.
will again be a solution of the system (1).
To show that formula (6) embraces all solutions of (1), we shall
first prove that the column vectors u i , u 2 , . . ., U, which correspond
t o distinct characteristic numbers hi, h,, . . ., hn are linearly inde-
Taking for h some characteristic number of the matrix d , we pendent.
will find the constants ui of the system of linear equations (4) that Let
n
correspond to this number. CAU~= 0 (7)
I t will be convenient for what is to follow to introduce matrix k= i
notation both for the initial system (1) and for the systems of the
relations (2) and (3). Multiply both sides of (7) from the left by t h e matrix A . Then,
We introduce the column vectors using the equalities

we find
n

Then, in place of (1) to (3) and (5) we can write* Eliminate the constant c, from (7) and (8):
dx
-- n
dt -Ax,

X = ueht,
A u = h u (u#O), Again multiply this equality from the left by A and use the
equality obtained, together with ( g ) , to eliminate c2 and so forth.
det ( A - h E ) = O We finally get
where, E = j l 6 i ~ l l ~ R , ~is= the
~ unit matrix. .
(hn-hi) (hn -h2) . . (hn-An-i) cnun 0 (10)
* When lnultiylying the squarematrix A by columnx, we nlultiply elements whence c, = 0. Since all the terms i n (7) are equivalent, i t follows that
of the ith row of A by the corresponding clement of column z and add all these
products. The sum thus obtained is the ith element of the column product A z .
dx .
The derivative of the column - is obtained by differentiating each element i.e. there is no relation of the form (7) between the eigenvectors
dt
of the column X. 2 t 1 , u 2 , . . ., U,, and these vectors are linearly independent.
 192 Stabilitu of h'ouilibrium and the Motions of n Sustenz [Ch. 5
Putting t = 0 in (6), we have B a systcm of two first-order differential equations if we put xi = X , x2 = X:
With an arbitrarily given initial vector X,, we can uniquely deter-
mine Ck (k = I , . . .,
n) from (11) by virtue of the linear indepen-
.
dence of the vectors u i , . ., U,,. Thus, (6) embraces the solutions
of the system ( I ) that satisfy any initial conditions X (0) = X,,
that is to say, i t embraces all solutions of the system (I).
In courses on the theory of differential equations i t is proved
t h a t in the case of multiple roots, formula (6) becomes somewhat
involved. So-called secular terms can appear in the formula t h a t
contain a polynomial in t in place of the constant vector u k :
u k + uLt +
. . . . In the general case, an arbitrary solution of the
system of differential equations (I) is determined by t l - - formula
I Important corollaries follow a t once from formulas (6) and (12).
I". If all the characteristic numbers of the matrix d have negative
real parts, i.e.
max ReAk= - a < 0
l<k<n
then lim
has
X (t) = 0, and the zero solution of the system of differential
equations ( I ) is asymptotically stable."
Now let Re 3Lk > 0 a t least for one k. Then ( I ) has a nonzero
solution X = c k u k e h k fwhich tends to infinity as t -+ m. At the
same time, the initial value X, = Ckuk (for t = 0) may be arbitrarily
small, since Ck is an arbitrary constant. In this case, the solution
X = 0 is unstable. We thus have the proposition:
2'. If at least one characteristic number 3Lk of the matrix 9 has
a positive real part (Re 3Lk > 0), then the zero solution of the system
of diflerential equations ( I ) is unstable.**
Example. The position of equilibrium of a linear oscillator in a medium with
a resistance proportional to the first power of the velocity will be asymptoti-
of motion mx
..+ +
cally stable. Indeed (see Example 3 on page 167), the differential equation
2fx cx = 0 ( m , c , f > 0) may be written in the form of
* The number a > 0 i s sometimes called the degree of stability.
** If all Re hh ,< 0 (k = 1, . . ,, n ) and if in a t least one of these relations
there is an equal sign, then the solution X = 0 will be stable, provided t h a t i n
(12) there are no secular terms in all summands where Rehk = 0. Otherwise,
he solu tion X = 0 will be unstable.
Sec. 381 Stabilitrl in Linear Approximation 1'33
The cliaracterislic equation
has roots with negative real part --m , which is what onsures t h e asymptotic
s t a l i l i t y of the equilibrium position.
38. Stability in Linear Approximation
In the system of differential equations (non-linear!)
expalid the right sides in a series of powers of the deviations
where f i is the sum of all terms of the expansion X i beginning
with second-order terms in xi, . . . , X, (i = I , . . . , n).
In the stationary case, aik are constant coefficients, and the
.
functions f i depend on X,, . ., X, and do not depend on t. In the
periodic case, a i k are periodic functions of t with period T, and the
nonlinear terms f i = f i (X,, . . ., X,, t) are also periodic in t with
period T.
Jf in (1') we reject all nonlinear terms f i , we get a linear system
of differential equations that is called a linear approximation for
. the nonlinear system (1).
At the end of last century, i t was established in the investigations
of Poincar6 and Lyapunov that both in the stationary and the perio-
dic case, one can judge about t h e stability of a zero solution of
a nonlinear system (I) from the linear approximation; namely,
from the asymptotic stability of the zero solution of a linear approxi-
mation there follows the asymptotic stability of the zero solution
of a nonlinear system.* This proposition finds broad applications
since the investigation of linear systems is substantially simpler
than that of nonlinear systems.
* I t was assumed hcre that the right-hand sides X i( X , , . . ., X,, t ) are con-
tinuous functions. At the present time it i s known what is t o be understood by
lincar ap roximation for discontinuous right members of Xi, and a n appropriate
criterion g a s been established for stability based on a linear approximation both
i n tlic periodic case and i n certain nonperiodic cases.
194 Stability of Equilibrium and the Motions of a System [Ch. 5
Sec. 381 S t a b i l i t ~ in Linear Approsimation 195

We will confine ourselves to a consideration of the stationary where 3Li, . . . . 3Ln are characteristic numbers of the matrix ,4,
case and here we will write the system (1') in the matrix form in while the moduli of the "nondiagonal" coefficients b , , (i < k) may be
order to prove our assertion: made arbitrarily small by a suitable choice of transformation (5). *
The transformation (5) is applicable to the nonlinear system ( l v ) .
The system (1") will be written in new variables as follows:
Here, A = 11 a i k 1: k = i is a square matrix with constant elements,
and f (X) is a column with elements f i (X,, . . . , X,) (i = 1, .... n).
Since by assumption the zero solution of a linear approximation
is asymptotically stable, i t follows (see Sec. 37) that all the cha-
racteristic numbers A,, .... 3Ln of the matrix A have negative real
parts
max Re3Lk= -a<O
i<k<n

Let us agree to denote the "length" 01 the column vector ;ir wit1h
where the column g ( @ with elements g i ( y ) , .... .gn(!/) is defined
hy the equality
comp~nenLsz,, .... Z,Lby I n~ I:
- t g (y ) = U-lf ( ET!)) (8)
and satisfies the icequali ty**
IY(v)I<TIYI (9)
Since each element of the column f (X) begins with terina of where the number g (like the number E) may he made arbitrarily
second degree, i t follows that small by choosing a sufficiently small neighbourhood I X l < A (and,
If ( x ) I - < E I x I (l9 respectively, 1 y l < A,).
Then, in accordance with equations (7) and inequalities (2) and
where the constant E > 0 may be chosen arbitrarily small if we (g), we find***
confine the variation of the variables XI, ....
r, to a suf5cieritly
sniall neighbourhood I X l < A.
The prool* of: the assertion of Poincarl and Lyapunov inay be
huilt upon the following lemma:
Lemma. Using the linear nonsingular transformation of variables
X = U?) (det U # 0 ) (5)
dr
the linear system of differential equations L=:
dt A x may be reducetl
to the "triangular" form **
* The proof of this lemma is given on pages 196-197.
** If we define the norm of the matrix d = 11 aik 1:) by the equality
n -1
(1 A 11 = ( 2
i, k = i
I aik 1 2 ) ~i t will be easy to verify tl e truth of the inequality
I A ~ l S l l ~ l l l ~ l
Therefore, from inequality (4) and equality (8) i t follows that
* See J . J . Stoker, Nonlinear Vibrations in Mechanical unrl Electrical Sy- I s ( v ) l S l l c-lllIf 0 2 l ) I < & I I U-lIIII VIIIarI
stems, London, Interscience Publishers, 1950. and, thus, i n icequality (9) we can put
** As a result of the transformation (5 , the matrix of coefficients A is rep-
2
laced by the matrix U-lAU, which is o triangular form. q=Il UII l1 u-lll&.
*** By yk we denote a cuinber that is complex conjugate t o yk ( k = l , ..., n).
13*
 Sec. 391 Criteria of Asymptotic Stability of Lineor Systems 197
196 Stabilitu of Eouilibrium and the Motions of a Sustem [Ch. 5
Therefore, the transformed system of differential equations

k= i
has the solution by (14),

Choose positive numbers 6 and q so that the inequalities a - 6 > 0
and <a-6 hold; then from (10) i t follows that
i.e. the solution y = 0 of the system (7) is asymptotically stable.
But the vectors X and ?j are connected by the linear transformation
(5); therefore the solution X = 0 of the system (1) is asymptotically
stable.*
Proof of the lemnza. First we shall show that with the a i d of a transfor-
mation S = U(1) z of the form (5) i t is possible to reduce t h e system of
differential equations
dx
-=As
dt (12)
to a form i n nhich: (a) the first variable zl does not appear on the right
of all equations beginning with the second, and (b) i n the first equation the
coefficient of z, is equal to the characteristic number Ii of the matrix A ;
i.e. to t h e folloning form:
which is only possible when b;,= Ii aild b;, = ... = b;li= 0, i.e. when the
system (16) has the form (13).
Since i n a linear nonsingular transformation. the characteristic equation
of the matrix A does not change,* the matrix 11 bjk 1; has as its characte-
ristic numbers the remaining n- l characteristic numbers ( I z , .... I n ) of
the matrix A.
Applying a similar transformation to the system of the last n-1 equations
(13) and so forth, we can finally, with the aid of a nonsingular linear
transformation, reduce the initial system of differential equations to the form
Finally, we perform the transformation of variables
z k = p k y k (p>O; k = 1 , .... n)
Then the system (17) will be replaced by the system (6) i n which the off-
<
diagonal coefficients bik=pk-ibik (i k) may be made arbitrarily small i n
absolute value if the number p > O is chosen sufficiently small. The lemma
is proved.

39, Criteria of Asymptotic Stability of

Linear Systems
For this purpose, i t is sufficient, for the first column of t h e matrix C ( l ) , In the preceding two sections i t was established that in the
to take the eigenvector u i which corresponds to the characteristic number stationary case, the zero solution of a n arbitrary (noolinear) system
Ii ( A u i = h p i r m 4 # 0) and choose the other columns of the matrix zcz, .... m&,
so that together with zci-they are linearly independent (then det U(l) # 0). of differential equations in deviations is asymptotically stable if
Indeed, the transformation X = z may be also written as follows: all the roots of the characteristic equation composed for the matrix
of coefficients of a linear approximation have negative real parts.
where zi, .... zn are the coordinates of the vector S i n the basis u i , * The characteristic equations of the matrices A and F l r l U (see foot-
US, ..., U,,. The system (22) has the solution note** on page 194) coincide since
U-1 ( A - I E ) U = C - l a c - - h E
s =uie'lt (15)
and therefore
* In accordance w i t h footnote ** on the preceding page, from the equa- det (U-lAZT--LE) = det G-1 det (A --LE) det U = det (A - I B )
lity (5) follows the inequality
( E is the unit matrix).
198 Stability of Equilibrium and the Motions of a System [Ch. 5 Sec. 391 Criteria of Asymptotic Stability of Linear Systems 199
Therefore, of great practical importance become the necessary and
sufficient conditions that all the roots of the algebraic equation The Routh-Hurwitz condition. For all roots of equation (1) to
with real coefficients haoe negative real parts, it is necessary and sufficient that the following
inequalities hold:
Ai>O, & > 0 , An>o
a . . , (5)
have negative real parts.
Denote by Ak (k = 1, . . ., g) the real, and by rj isj + If the coefficients of (1) are given as numbers, then the condilions
(5) are readily verified. But if the coefficients of (1) contain literal
( 7)
j = l , . . ., the complex roots of the equation ( l ) , and parameters. then the computation of the determinants A k for large
assume that in the complex plane all these roots lie to the left k is already complicated.
of the imaginary axis, i.e. that I t is therefore of interest to examine other conditions established
in 1914 by the French mathematicians L i h a r d and Chipart. I n
these conditions, the number of determinantal inequalities is roughly
half that in the conditib~is(5) of Routh and Hurwitz.
Then The 1,iknard-Chipart conditions. For a polynomial f (h) z aohn +
+alAn-l -1 . . . + a,-,? 4 a,, when a. > 0, to have all roots with
negative real parts, i t is necessary and sufficient that
(1) all coefficients of the polynomial f (A) be positive

(2) the determiriautal inequalities

Since, by the inequalities (2), each term in the last part of (3)
has positive coefficients, i t follows that in (1) all the coefficients be valid. (Here, as before, Ak denotes the EIurwitz determir~antof
are positive. The positivity of all coefficients is a necessary (for the k t h order.)
a. > 0) but by no means sufficient condition for all roots of the We shall now get acquainted with the geometric criterion of
equation (1) to be located to the left of the imaginary axis. stability.
In 1875 the English mechanician Routh, with whom the reader 111 the equality*
is already acquainted, produced an algorithm by means of which n
it is possible, using the coefficients of the polynomial f (A), to find f (A)=ao k=fli (A-Ah)
out whether the polynomial is "stable", i.e. whether a l l its roots
have negative real parts. In 1895, the German mathematician replace A by io and let o vary from - m to + m . Compute the
Hurwitz, independently of Routh, established the very same criterion appropriate increment of the angle 8 =arg f (io):
in a modified form with the help of determinants (Hurwitz deter-
minants):
a, a, a, . . . . . .
a. a2 a, . . . . . . Now note that** (Fig. 47)
0 ai a,. . . . ..
0 a. a 2 . . . ...
A?: arg ( i o - Ah) = { A:
if E",*,=",
S . . . . . .

. . . . . .a * Here, the n roots 01 the polynomial f (h) are denoted by h,, . . ., h,.
(Here, a, = 0 for p >n everywhere.) ** We assume here that not a single one of the roots hk lies on the imaginary
axis.
 200 Stability of Equilibrium and the Motions of a System [Ch. 5 Src. 391 Criteria of Asymptotic Stability of Linear Sgsten~s 201

Therefore, denoting by L and r the number of roots lyillg to the where n is the degree of the polynomial f (h) (see Fig. 48 for n = 6).
left and, respectively, to the right of the imaginary axis ( L -+ r = n ) , Note that for a stable polynomial* the argument 0 varies mono-
we will have tonically as o varies from 0 to CO. This follows from the fornlula
+m
A-,%(m) =(L-r)n rL

We consider the curve described by the affix* of the coinplex 0 (m)= h= i arg ( i o- 3Lk)
number f ( h ) as o varies from -CO to +CO. This curve decomposes
into two branches: on one w > 0, on the other 0 < 0. One branch since for such a polynomial each summand on the right-hand side
is a monotonic increasing function of to.

Fig. 47
l
Fig. 48
is obtained from the other by a rnirror mapping about the real axis,
since f ( i o ) and f (-h)are complex conjugate numbers. For this +
Example. Let f (h)= h5+5h4+ IOh3+ l l h z 7h+2. Then f (io)=U (W)-+
+ib' (W) where
reason, denating by A? the increment as o varies from 0 to CO, U (0)=5o4-11o2+2, V(O)=O(W~-1002+7)
we obtain
1 +m To construct the hodograph of f (to) we note that U (0)-2 and V ( a ) vanishes
AF%(o)=-A+,%
2
( ~ o ) = ( I - nr ) ~ (8) < <
at o .= 0 and a t m== o i , o = oz (0 oi wz); the squares of and o$ are
determined from the quadratic equation

-
Whence i t is seen that all the roots will be located to the left
of the imaginary axis ( L n, r = 0 ) when, and only when,
Geometric criterion of stability.** For a polynomial f (h) 1 0 be
stable, i.e. for all its roots to he located to the left of theiinaginary
axis, i t is necessary and sufficient:
(1) that the hodograph of f ( i w ) , ns o varies from O to m, should
o~=5-~?ii,--0.76; @=5+ v18a9.24
It will readily bc seen that Li (m,) < 0, U ( a z ) > 0. Besides, V' (0) = 7 > U.
Thus, for o = 0 the hodograph begins on the positive real axis, goes upwards
a t first, intersects the positive imaginary axis, then the negative real axis (at
o = a , ) , the negative imaginary axis, and, finally, again the positive real axis
(at o = o z ) When o > oz, the hodograph does not intersect the coordinate
axes and goes off to infinity in the fifth quadrant ( U > 0, V > 0), since J L = 5.
Here, tan 0 = - -+ $ CO when o -+
U (0)
+ CO. Hence,

not pass through the zero point*** and

(2) that for this hodograph
i.e. f (h) is a stable polynomial.
We could arrive a t thc same conclusion by proceeding from the LiBnard-
Chipart criterion, since all the coefficients in f (h) are positive and
* The affix of a complex number z is an appropriate point in the complex
z-alane.
' ** This criterion was first applied by A . V. Mikhailov for investigating au-
tomatic control systems. For this reason, in the technical literature, the geo- 1 0 1 10 7 1
metric criterion of stability is frequently called RJikhailov's criterion.
*** Condition ( l ) signifies that f (h) does not have any purely imaginaq roots. * The stable polynomial is also called thc Hurwitz polynomial.
 S e c . 401 S m a l l Oscillations o f a Conseroative Sustem 203
CHAPTER 6
Also expand the potential energy in a power series of the coor-
dinates:
n

Small Oscillations
By hypothesis., IIo=O. Resides, the generalized forces i n the equi-
librium position are zero:

40. Small Oscillations of a Conservative System

If at an initial instant of time the position of a scleronoinicsystem Therefore, introducing the notation
is chosen sufficiently close to a position of stable equilibrium and
the i i ~ i t i a lvelocities are sufficiently small in absolute value, then
throughout all the motion both the deviations from the position of
equilibrium and the generalized velocities will be small in absolute we can also represent the potential energy in the form
value. This circumstance permits retaining, in the differential n

equations of motion, only the linear terms in the deviations and n=1 2 Cikqiqh f (**I t5)
the velocities, and higher-order terms may be rejected. Theii the i , k=l
difl'rrential equations of motion become linear, i.e. the problem is
Dropping terms of the third and higher orders of smallness i n
"linearized". I11 this section we will consider the linearization of
equations of motion for a conservative system. qi and qh in formulas (3) and ( S ) , we can represent the kinetic
The kinetic and the potential energy of a conservative system with and potential energies, as quadratic forms with constant coefficients:
l I
n degrees of freedom are expressed in terms of the independent
coordinates qi i n d the generalized velocities ii
(i = 1 , . . ., n) in
the following ma'nner:
n
.. where aih = ski, cih -- cki (i, k = 1, . ., n). .
T =T
1
2
i, k = i
aih (qi, . q,) qiqh, n = n ((l,, . . . , qn) (1) From the physic.al meaning of the kinetic energy i t is clear that
always T>O. Since we assume that the position of equilibrium
is not a singular point*, then always T > 0 if only not all gene-
As in the preceding chapter, we assume that the coordinate origin ralized velocities are simultaneously equal to zero, i.e. the quad-
qi = . . . = q, = 0 is a position of equilibrium and that in this
3 ..
position IT = 0. Expand the coefficients a i k (qi, . ., q,,) in series . ratic' form
i. k=l
aihqiqk= 2T is positive definite
in powers of the coordinates:

-
where aik aih (0, . . . , 0). (aih = ski; it k = I , . . . , n) are Constants.
Putting these expressions for the coefficients into formula (1) for
Further, in order to ensure stability of the given position of equi-
librium, we demand that (in accord with the Lagrange theorem)
thc kinetic energy, we have the potential energy have a strict minimum in the equilibrium
n
position. Since H, = 0, this means that in a certain neighbourhood
of the origin ,
n n

wliere by (**) we denote the sum of the terms of third and higher
orders in qi and qi (i = l, . ., n) . . * See footnote on page 47.
204 S m a l l Oscillations [Ch. 6
S e c . 401 S m a l l Oscillations of U Conservative S y s l ~ r n 205
But the quadratic form (8) is a quadratic homogeneous functioii
in the coordinates. Hence, inequality (8) holds throughout the space, Expaiiding the determinant, we get on the left-hand side a poly-
with the exception of the origin where this form vanishes. In other nomial of degree n in h. Thus, the square of the frequency h = o2
words, the potential energy is now represented as a positive definite of the desired harmonic solution (10) must satisfy the nth degree
quadratic form in the coordinates.* algebraic equation (13), which is called the secular equation or
Form the Lagrange equations by proceeding from the expressions Jrequency equation.
(6) for T and II: To every root h of (13) there corresponds a particular solution
(10) (for an arbitrary constant a ) of the system of differential equa-
tions (9). In this solution, o = J/h.
Write the above formulas in matrix form. We introduce two
We shall seek a particular solution of this system of linear diffe- symmetric positive definite matrices*
rential equations i n the form aii ... at,' . Cin
-
Cif

qi== uisin (at + a ) (i= l , .. , l n) (10) A = 11 aik 11 = . . . . . , C = 11 cikl] . . . . . (l4)

ant ... ann Cni . Cnn
i.e. in the form of harmonic oscillations with one and the same
frequency o and the same constant a for all coordinates. and the column vectors
Substituting expressions (10) for qi in the differential equations
(9) and setting
h = o2 (11)
Q=
we obtain the followiug system of algebraic equations [after can-
celling out sin(ot'u)] that are linear in the amplitudes ui:**

(U is the amplitude vector). Then the system of diflerential equa-

tions (9) will be written as
Since all the amplitudes ui of the sought-for oscillation must not A~+c'~=o (16)
vanish, the determinant of the system of homogeneous equations
(12) must be equal to zero: The particular solution (10) will look like
p= usin(ot+a) (17)
and the result of substitut.ing the solution (17) into equation (16),
i.e. the system of algebraic equations (12), has the form
(C-hA)u-0 (h=a2) (18)
The frequency equalion will be written as
* Of course, cases are possible when the function n (qi, . . . , g,) prior to det(C-hS)=O (h-=02) (19)
rcjection of tcrms (W) has a strict minimum a t thc origin, and after rejection
of these terms has a nonstrict minimum at the same point (for example, n = So as to figure out whether the roots h of the secular equation (19)
- +
c2 ( q , +
. . . qn)2 jc12 (9: + +
. . . q:), C > 0, d > 0. But we shall con-
sidcr such cases as special and exclude them. In these special cases, rejection
are always real and positive, first consider some properties of
quadratic forms with real coefficients.
of thc terms (**) in the expression for is not justified, since i t can radically
distort the picture of the motion. * The symmetric matrix A =]l a i k 1: is called positive definite if the
** We call ui the amplitude of harmonic oscillation (10) of the coordinate qi, n
though actually the absolute value I u i 1 is the amplitude; the initial phase
(at t = 0) of harmonic oscillations (10) is either a (at u i > 0) or -a (at u i < 0). quadratic form 2 aikqiqk corresponding to i t is positive definite.
i, k = i
 Sec. 401 Small Oscillations oj a Conservative System 207
206 Small Oscillations [Ch.6
then for any vector z*
To each quadratic form 5
i, k=i
ainuiua there corresponds a certain C (U, W)= h A (U, v)
n
,Indeed, in scalar notation, (22) taltes the form
bilinear form 2
i, k=i
aikuivk for which we int,roduce the abbreviated
notation

Multiplying b0t.h sides of the ith equation of (22') by v i and

summing over i , we get
Then the quadrat,ic form will be written as n n

which is the equality (23).

Tho following properties of a bilinear form are readily verifiable:
I". A (U, f ?cz, P ) = -4 (rr,, P ) f A ( a t Z , v).
2". A ( h u , V ) :-hA (U: 7 3 (h is a scalar).
3". A ( u , u ) = A ( u , u ) . *
We shall also show t h a t for any complex vector U :
4". A ( U , G) is a real number.**
+
Indeed, setting zi = v i w ( v and za are real column rectors)
we find, by virtue of 1" to 3":
A G) == A (v 1-i w , P- i w ) = A ( T , v)- iA ( v , W) + iA ( W , V ) +
-
4-A ( W , W ) -4 (U,v) -F A ( W ,U )
The last expression is obviously real.
Froin (20) i t also follows that:
5". 1f A ( 1 4 , 7 0 is a positive definite quadratic form and
is an arbitrary complex vector, then
We shall now show that for a n y two amplitude vectors u and 71,'
corresponding to diflerent roots h and h' (h #h') of the secular equa-
tion the following relation* is ~lalid:
Indeed, according to 6", two equations hold:
C ( U , U')= hA ( U , U'), C (U, U') = h'A ( U ,U') (25)
Rut h # h f . Therefore, from (25) follows relation (24).**
We now prove that from the symmetrical nature of the matrices A
(20)
and C' and from the positive definiteness of the matrix B it follows that
the secular equation (13) [or (19)l has only real roots.
Indeed, let h be a complex root of the secular equation (h #X)
and let a complex vector u # O correspond t o it. Then h is also
U f0 a root of the secular equation with amplitude vector 6 . Since h #h,
i t follows frpm what has been proved that A ( M , G) = 0, but this
contradicts the inequality (21).

Indeed, assuming ?L = -+
itc we have A (v, v) >, 0 , and
A ( w , W ) >, 0. In one of these relations we have the sign >, since
-
1f h is real. then also the amplitude vector 71 # 0 that corresponds
to i t may be chosen real. Now, putting t* 7r in (23) and noting that

from U # 0 follows a t least one of the inequalities .L. # 0, # 0. * If we introduce into n-dimensional space an A metric, i.e. if we take
Then from equality (20) follows the equality (21). the square of the length of the vector zc to be the magnitude of the quadratic
Let us now prove: form
n
6". If h is a root of the secular equation det (C - hA4)= 0, and
u is the amplitude vector corresponding to i t !see (18)l A (m,U ' ) = aikujuk
i, k=i

* Unlike equalities 1' and 2", equality 3" holds only for a bilinear form
with a symmetric matrix of coefficients.
** The bar marks the transition t o complex conjugate quantities. Property 4'
is valid only for a symmetric matrix with real elements.
then A (U,m') is the "scalar product" of the vectors U and U' in this metric.
Therefore, (3)expresses the following property of amplitude vector?: amplitude
vectors corresponding to different roots of the secular equation are always mu-
tually orthogonal in the A metric.
** By virtue of (25),A (U,26') = 0 takes place along with C (U,U') = 0.
 Sec. 40) Small Oscillations of a Conservative System 209
203 S n ~ a l lOscillations [Ch. 6
But A ( u k ru j ) = 0 for k #j.and A ( u k , u j ) ) 0 for k = j. Therefore,
A(u, U) >O, we find from equalities (31) it follows that

i.e. there can be 110 linear relationship between the vectors

13ut in our case tlw quadratic form C ( u , U ) ;- 2
cikuiutt is also 'Ui, .. ., ZCn.
i, k = i Xow in formula (30) we choose the values of the arbitrary
positive delinite. Then not only A ( U , U)> 0, but also C(zc, W ) >O. constants Cj, a j in such manner that the following arbitrary
Hence, h > 0. preassigned initial conditions should hold:
Thus, the secular equation (13) has n positive roots Ai, to which
there correspond real positive frequencies oj= and real arnpli-
.
lude vectors u j ( j l ~, . . , n).
or in malris notation
Let us first cons~der the case when all the roots of the secular
equation are distinct. To each A j there corresponds a particular
solution
Q = u j s i n (ojt$aj) (oj=1/3Lj) (27) From formula (30) we get

with an amplitude vector zcj, the coordinates of which uij, . . ., unj

must satisfy the system of linear equations
n
Qo= ojCj c o s a j u j
j= 1

or in matrix notation By virtue of the linear independence of the vectors uj (j = 1, .. .

Since the system of differential equations (9) [or ( I C i ) ] is linear,
the linear combination with constant coefficients of the solutions (27)
is again a solution of this system. For this reason
.
with arbitrary constants Cj, a j ( j = l , . . , n) is a solution of the
system (9) or (16). We shall show that the formula (30) embraces
all motions of the system.
We shall first prove t h a t n amplitude vectors uj ( j = 1, . ., n) .
are liuearly independent.* Indeed, let
Then for any fixed k (1 ,< k ,<n)
n n
. . .. n ) , we determine uniquely from this the products Cj sin a j
and oljCjcos aj and, consequently, since oj # 0, we uniquely
determine the values of the arbitrary constants Cj and a j
( j = 1, . . ., n).*
Thus, in the absence of multiple roots in the secular equation
the formula (30) embraces all oscillations of the system.**
Now if the frequency equation has multiple roots, then i t may
be asserted. that there will be a t least m solutions of the form
u sin ( @ l + v.), where m is the number of distinct roots h j of the
secular equation.
Lagrange considered that in the case of multiple frequencies,
the general solution of the system (9) is no longer representable
in the form (30) and that on the right-hand side of (30) so-called
secular terms of the form
( U $- u ' t $- u"t2 $ . . .) sin ( o t -F a )
appear.
* The a j are determined u p t o a n additive a multiple of 2n.
** The following relations hold for t h e amplitude vectors uj:

* This was proved in Sec. 37 for the special case when A is a unit matrix.
 210 Small Oscillations [Ch. 6 Sec. 411 Normal Coordinates 211

However, Lagrange was in error. As Weierstrass demonstrated Retaining i n II only t,he quadratic terms, we finally get
later, to each root A of multiplicity p there corresponds exact.1~I;
linearly independent solutions of the system of linear equations
(12), i.e. for each root Lj of multiplicity p it is possible to find p
linearly independent amplitude vectors. Thus, also in the case of
multiple frequencies there exist n linearly independent amplitude where
vectors, and the formula (30) which is formed with their help yields a = mla, c =mgl +?h2
a general solution in this case as well.
b= yh2
The oscillations
Write the equation of frequencies
q = = C 7 j ~ j ~ i n ( w j t + a j( )j = = l , . . ., ?L) (35)
that make up an arbitrary oscillation of a system are called principal
oscillations of the system. and one of two equations for determining the amplitudes i n the principal
A rigorous derivation of the formula (30) for the general case oscillations (these two equations are dependent):
(i.e. given multiple frequencies as well) with the aid of the so-called ui b
"normal" coordinates will be given in the next section. In this (c-ka) ui-Eu2=0, i.e. -=-
u2 C-ka.
derivation, the case of multipie roots of the secular equation is From the frequency equation we get
not specifically isolated.
Example. Coupled pendulums. The points of suspension of two identical
simple pendulums each of mass m and length l are located on one horizontal
straight line. Points of these pendulums, a t a distance h from the points of Accordingly, for the first principal oscillation ui - uz = Ci and
w = C i sin ( o i t + a i ) , c ~ z = C isin ( q t + a i ) (cpi=cp~)
for the second q = - u 2 = C 2 and
'pi =C2 sin (02t+ a2), 9)2= -C2 sin (02t+ a 2 ) (Q = - - v 3 )
I n the first principal oscillation both pendulums are all the time in one phase,
the spring is relaxed and the pendulums do not exert any effect on one another.
In the second principal oscillation the pendulums are all the time in opposite ,
phases.
Arbitrary oscillation is obtained b y superimposition of the two principal
oscillations:
+
pi =Ci sin ( q t a i ) +C2 sin (o2t +ad

Fig. 49
+
cp2= Ci s i n (oit a,)-C2 sin ( ~ ~ t + a ~ )

suspension (0 < h ,< 1) are connected by a spring of rigidity y ; the spring is 41. Normal Coordinates
relaxed when the pendulums occupy the vertical position. I t is required to deter-
mine the oscillation of the system in the vertical plane. Two quadratic forms
For the independent coordinates we take the angles cp, and cp2 formed by
the pendulums and the vertical (Fig. 49). I n the equilibrium position, cpi = cp2 =
= 0. Up to higher-order infinitesimals, the extension of the spring is equal to
h I sin cp2 - sin cp, I = h I cp2 - cpi I. Therefore, in the given case and

of which a t least one, say A (q, q ) , is positive definite, can always,

by means of one and the same (nonsingular) transformation of
14*
21 2 Small Oscillations [Ch. 6
Sec. 411 , Normal Coordinates 213
variables,
n
Since the transformation of coordinates (2) is nonsingular, the
. . . , n; corresponding determinant is nonzero:
q i= 2 uijOj
j= 1
(i = 1, det (uij)?, j=i f. 0) (2)

be reduced to the "sum of squares" i.e. the vectors M , , . . ., t i n are linearly independent.
Utilizing the simple expressions (4) for T and TI in normal
coordinates, form the Lagrange equations i n these coordinates:

Here, all h j > 0, since (see Sec. 40) the form C (p, q ) is also
positive definite. Each of these equations contains only one unknown function. The
Since the generalized velocities ii
and bj are connected by the general solutions of the equations (6) are known to determine the
harmonic oscillations
same relations which connect the qi and Clj:

-
where Cj and aj are arbitrary constants ( j I , . . . , n). Substitu-
ting these expressions into formula (5), we get a general formula
i t follows that i n the first of the equalities (3) we can replace qi for oscillations:
n

and 8 j by qi and 8j, after which we get the following expressions

for the kinetic energy and the potential energy:
I t is thus rigorously established that this formula embraces all
small oscillations of a conservative system in the most general
case.*
Assuming, in formula (8), all arbitrary constants, except Cj and
a j , equal to zero, we get the j t h "principal" harmonic oscillation

(In normal coordinates, this oscillation occurs when all €li = 0

The variables 8,, . . . , -0, are called normal or principal coordi-
nates. The formulas (2) of transformation from arbitrary coordina-
tes to normal coordinates may be given in vector notation as
follo\vs:
12
for i + j, and only the coordinate Clj changes.) In the preceding
section it was established that the square of the frequency hj = o!
satisfies the frequency equation. Since for q there exist no harmonic
oscillations of the type (g), other than those which occur as summands
in the general formula (8), i t follows that hi = o! ( j = 1 , . . ., n)

p= r,i ojaj
j=
(5)
are all the roots of the secular equation. Besides, if some root is
repeated p times, then t o i t there correspond p linearly independent
amplitude vectors z c j determined from the system of linear equations
where (28) or (29) of the preceding section.
We have thus again proved that all the roots hi of the secular
equation are real and positive and we have established that to the
n frequencies oj = I/& there correspond n linearly independent
amplitude vectors w j ( j = 1, . . ., n).

* I n the preceding section, this forruula was established solely for the case
when the secular equation has no multiple roots.
 t
2i4 Small Oscillations [Ch. 6 Sec. 421 Eflect of Periodic External Forces on Oscillations 215
J

Substituting into the formula A ( g , y) the expression (5) for q Hence, equating the coefficients of the independent increments of
i n place of g , we obtain the normal coordinates 60j, we get
n
l

oj = i=-2i uijQi ( j = 1, .... n) (5)

On the other hand,

A (q, q ) =
n
El (33 (11)
"ne~7" ones Bj by means of the matrix U 11 uijll:
qb= -+
~ ~ 2 0 2. + +~lnBn,
-
Thus, if the "old'' coordinates qi are expressed in terms of the

j=
+
qs = uziei ~ ~f .~. . f uznen,3 ~+ (6)
Comparing (10) and ( I I ) , we get the following relations* for the
amplitude vectors uj ( j == l , .... n), with the aid of which, using
. . . . . . . . . . . . . . . . . ( q = o B , det u # 0 )
formula (5), the transformation to normal coordinates is accomplished: 4n = h i e i + ~n2e2 + ... -I- %non

I, i=j, then the "new" generalized forces O j are expressed in terms of

A ( u ~ ,Z L =
~ )hij= (i, j = l , .... n) (12) the "old'' generalized forces Qi by transposing the matrix U':
l
01 = uitQi +
u21Q2 . + +
uniQn9
42. The Effect of Periodic External Forces @ 2 = Y ~ z QS~ u22V2 + -
. GzQn
. . . . . . . . . . . . . . . . . . (0 = o,Q) (7)
on the Oscillations of a Conservative
System uZnQ2 + . - + unnQn

a ~ i
dn
Aside from the potential forces --- let a system be acted
l
@n = uinQi j
Comparing the matrix formulas q -
778 and Q = (U1)-%, we
see that in passing from coordinates to forces the transiormation
upon by certain forces Qi = Qi ( t ) (i = l, .... n). Pass to the nor- matrix Cr is replaced* by the matrix (U1)-l.
mal coordinates using the formulas I This is expressed in words as follows: the gensratized forces are
n transformed contravariantly with respect to the coordinates.**
q j j (i=1, ..., n; det(uij);j=i+0) (1) Aiter learning how to determine the 0, from the given Qi,let
3= 1 us write the Eagrange equations in normal coordinates using for
To the forces Qi in coordinates qi (i = l , ..., n) there correspond T and ll the expressions (4) of the preceding section:
the forces O j in coordinates O j ( j = 1, . . . , n). Proceeding from l .~., + o ; ~ ~ = @ ~ ((j t=)I , ..., n) (8)
the fact that the expressions for the elementary work of the
forces are equal, let us establish a relation between Qi and Oj -
Denote by BJ ( j l , ..., n) an arbitrary
of equation (8). Then the general solution of
particular solutiou
(8) will be
n
X
i=i
QiO= jzi
n

8,601 (2) l
Bj=Cjsin(ojt$aj)+05 ( ] = l , ....
Let O j ( t ) be the 'periodic force and let this
n) (93
force be sinusoidal
Noting that by virtue of formulas (1) with frequency Q:
1 O j (t) = AjsinQt (10)
Then i t will readily be seen that for the 8; we can take

and substituting these expressions for 6qi into (2) we obtain J

* I n other words, the vectors z c j
metric (soe footnotc on page 207).
* :
If L; = (1 u i j l/ is an orthogonal matrix, then (U')-1 = U, and the
forces are transformed exactly like the coordinates.
** In the general case, when the. trznsformation of coordinates is nonlinear,
( j = 1, .... n ) are orthonormal in the a the geueralized forces are transformed contravariantly with respect to the di-
fferentials of coordinates.
716 S m a l l Oscillations [Ch. 6 Sec. 431 Eztremnl Properties of Frequencies of Conseruati~;eSystem 317

.
If Q i (t) (i = l , . . , n), and hence also Oj (t) ( j -- l , . . . , n) are t,hat accomplishes the transition to the normal coordinates
arbitrary periodic forces with period T and frequency Q=- 2.7 g,, . . . , 9, i n which the quadratic forms*
T '
i t follows t h a t Oj (t) may be expanded i n a Fourier series:
00

O j (t) =
m= 0
Ajm sin (m& + qj,)
have the simple ("canonical") form:
( j = I , ..., n) (12)
Then, by virtue of the linearity of the equations (g),
a3 A (g, 9) = ]=i
.fI o!, 2
C (g, q) = j= i ‘O~B?
g: = 2 A" sin(mQt+qjm)
0;-~ ~ 2 s - p
( 1 n ) (13)
m= 0 From now on we shall assume that the principal oscillations a r e
If some mC2 coincides with wj, and the corresponding Aj,, # 0, numbered in such fashion t h a t their frequencies proceed in the
then for the coordinate Bj we have the phenomenon of resonance. following increasing order:
Substituting expressions (9) for 9 j into the formula
n
Q= ejuj Consider the ratio of quadratic forms (3)
j= i
we get
q =ir^+q*
where
n
g= 2i Cjzcjsin (wjt + a j )
j=
for any q # O or, which is the same thing, for ally values
..
Oi, . , Q, t h a t are not simultaneously equal to zero. Replacing
all the wq in the numerator of (5) by a smaller or equal number
are the free oscillations and
W:, we have

are the forced oscillations of the system, and u j is the amplitude

vector with coordinates uij, ugj, . . ., u n j ( j-- l , . . . , n). 011 the other hand, from (5) i t is at once obvious that for 82 = . . .
. . . =0, = 0 the ratio C (" attains the value m:. Hence."
A (97 rl)
43. Extremal Proper ties of Frequencies of a Conservative
System. Rayleigh's Theorem on Frequency Variation
with Change in Inertia and Rigidity of the System.
Superimposition of Constraints
In Sec. 4 1 we considered the linear nonsingular transformation
of coordinates
. .
* C ( q , Q ) i s the doubled potential energy, and A ( q , q ) is obtained from
the expression f o r thc doubled kinctic energy A ! q , g ) by replacii~g q q.
n W: If on a real-number axis we plot the points 02, a :, . .., :W and at
q = ' 2ejuj
j= l
.
these points concentrate masses m i = O f , I M ~ = O;, . ., m,==8$, then, by ( S ) ,
or, i n scalar notation, the quantity C(R.will be the coordinate of thc centre of these masses.
A ( g , Q)
Whence follow imniediately t h e relations ( G ) ancl (7), since the centre of
pi= 2 uijBj
j=l
( i = 1, . . . , n; det (uij)?, j=i # 0) (1') mass is always located between the extreme inasses and coincides with one
of the extreme masses when a l l the other mas,ces are zero.
 218 Small Oscillations [Ch. 6 Sec. 43) Estremal, Properties of Frequencies of Conservative System 2 19

Now let a linear homogeneous constraint* be imposed on the system

91

Thus, of all the constraints of form L=O the quantity

Expressing q,, q2, . . . , q, in terms of the normal coordinates by
means of the transformation ( l ) , we will again have, in normal
coordinates, a linear homogeneous constraint:
reaches its greatest value o: with the constraint 8, =O. Hence,

We abbreviate constraint (8) or (8') to

L=O
In place of one c.onstraint L= 0 , i t i s possible t o impose on t h e
-
I t is always possible to find the values 8, and O2 such t h a t toge-
ther with 8, . . . = 0, = 0 they will satisfy the constraint equa-
tion (S'). According to formula (5), for the corresponding g,
system a number of const,raints L , = 0, . . ., Lh-, = 0 . AS was done
in the special case of one constraint, i t i s possible to show t h a t
= max min
0;:
C(q1
L1=O A (Q' 9)
(h=2, . . ., n)
Therefore, also,"'
min. C o: 'The formulas (7) and (11) express the extremal properties of fre-
L=,, A (Q, 9)
quencies of a conservative system. These properties are sometimes
We shall now vary the constraint L = 0. Then the left side i n c a l l ~ dma.xirninima1 properties.
the inequality (9) will vary remaining all the time smaller than I n place of the formulas (7) and ( I I ) , we can easily obtain
or equal to a;. But for the constraint 8, = 0 (here 1; = 1 , 1; = . . . similar formulas:
. . . ==1; -- 0) the ratio '(Q'' " i s given by the formula o i == max
c (919) (7')
A (Q', 9) A ( Q , Q)

and for this reason [by analogy with formulas (5) and (7)]
* If a nonlinear constraint irl given and the equilibrium position satisfies
the equation of the constraint? then there is no absolute term in power T h e extremal properties of the principal frequencies expressed by
expansion of the left side of the constraint equation:
(7') and (11') are sometimes called rntnimaxirn,al properties.
I n addition to the given system, let us consider yet another
conservative system with kinetic and potential energies
n
Besides, we assume that there are indeed linear terms, i.e. that
i=i
> 0.
Then, rejecting second and higher-order terms, we represent the constraint
equation as (8).
** The symliol on the left of the inequality (9) denotes the mininlwn of the
aiid with principal frequencies
ratio '(9.9) , provided that only vectors g f 0 are considered that satisfy the
* (474)
constraint equation (8).
 Sec. 431 Extremal Properties of Frequencies of Conseruutiue System 221
220 Srnall Oscillations [Ch. 6

For this system

Impose on a system
-L,=O, L-z = 0 , . . . , L,=O
S independent linear constraints

Let the conservative system thus obtained with n - S degrees

of freedom have principal frequei-lcies my of.< . . . <m:-.,. Here,

....
Ls=O
Let the new system have greater rigidity for the same inertia,
i.e. for any q Comparing formula (19) with (7) and (11) (when h -- l = S ) , we get
a, 4 < us+, (20)
or less inertia for the same rigidity In exactly the same way, for any h,< 12-5
c ( c f , 9)
ogZ= max
- min
- A(q,a)
L1=0,
L1=0,. .. .. ,. Lh-l=O
, L,=O
In both cases, for any q#O,
- -
Here the constraints L, -- 0, . . . , L,$..= 0 are fixed, while the const-
-
raints L, = 0, . . . , Lh-, 0 vary. Compari~lg (22) with (11) and
with the formula
But then both the minimums and the maximums of these ratios
will be connected by the same inequality, i.e. from the inequality
(.16), by virtue of (7), (11), (14), and (15), there follows

i n which all S +h -l constraints vary, we will have

Here, a t least one of these relations has the sign <, so long as
t,he identity*
does not hold.
We have arrived a t the Rayleigh theorem:**
The principal frequencies increase when the rigidity of the system is
increased or the inertia of the system is decreased.***
Let us find out how the imposition of constraints affects t h e
magnitudes of the principal frequencies wl ,< o, ,< . . ,< on of a .
conservative system.
* Indeed, according t o ( 5 ) , identity (18) holds in the case of oj = oj ( j =
= 1, . . ., n).
** This theorem was established by the English physicist Rayleigh in 1873
(flayleigh 3 . W. The Theory of Sound. Vol. 1, 2nd ed. Macmillan, London. 1894).
*** How much the principal oscillations increase may be judged i n the
first case by the difference C ( g , 9)-C(q, g), and i n the second case by the
difference z ( q , 9)-A(q, g ) [see 71.
Fornlulas (22) show that in the imposition of S independent const-
raints each of the first n - S principal frequencies increases and in
the process does not exceed the old principal frequency, the number
of which is S units greater than the number of the given frequency.
1. As a n application of t h i s proposition, i t may be shown that the roots
Ai ,< A2 Q . . .<
An of the secular equation A (A) -= det (cik-hih)f: k,i = O
are separated by the roots A: ,< . . .<
A:-i of the equation*
A i (A) = det (cik-Aaik)7,-ii=i = O
i.e.
Al<Y<A2<?+< ... <Az-i<An (23)
Indeed, the equation Ai (A) = 0 is a secular equation for a conservative
system obtained from the initial one by superimposition of one constraint
* A, (h) is the principal minor of order (n - 1) in the determinant A (h).
I t is sometimes said that inequalities (23) express the "separation theorem"
for roots of the secular equation. Inequalities (23) may be used t o find the lower
and upper bounds of the roots of a secular equation (see for example [2]).
222 Small Oscillations [Ch. 6 Sec. 441 Small Oscillations of Elastic Svstems 223

, = 0. Therefore, setting hk = r&

p
. . ., R - l),
we a t
(k = l,. . ., n), h; -
o:z ( j = 1 , . . .
once get the inequalities (23) from the inequalities (22) Then for the elastic forces F f =
an ( i = I, . . ., n) we find
-%
for S = l. n
2. We indicate yet another noteworthy application of the proposition on
variation of frequencies upon the superimposition of constraints. Cracks in
a glass are known to be revealed by tapping one's fingers on it. The reason lies
F?= - 2 cikyk (i = I , . . ., n) (2)
k=l
in the fact that a glass without cracks has supplementary constraints (compared
with a glass with cracks) between its parts. That is why in a glass without Considering the position of equilibrium yi = . . . -- yn = 0 to be
cracks the oscillations must bo higher. stable, we assume that the quadratic form {i) that expresses the
potential energy as a function of deflection is positive definite:
44. Small Oscillations of Elastic Systems
As an important example of small oscillatioas of a conservativo
system, consider n masses mi, mz, ..
., m, concentrated a t n points The kinetic energy of the system has a simple form:
(I), (2), . . ., (n) of an elastic system S (a string, rod, membrane,
plate, and the like) having finite dimensions and in some way attached
a t the ends. We shall assume that the displacements (deflections) 2'" 2-21 miy; '
(4)
y,, y,, . . ., y, of points ( i ) , (2), . . ., (n) of the system S and i=l
In the search for the harmonic oscillation yi = ui sin {wt +a ) ,
I6 IF* we arrive (as we did i n Sec. 40) a t the frequency equation

Pig. 50

.
the forces F i , F?, . ., F,, acting on the masses mi, m,, . ., m, .
Icni Cnz cnn -m&. m l
and a t the algebraic equations for determining the amplitudes:
are parallel to one and the same direction and for this reason are
n
determined by their algebraic magnitudes (Fig. 50). The deflections
yi, yz, . . ., y,, may be regarded as independent coordinates of the 2i (cik-
k=
hm&) uk = o ( i = 1, . . ., n) (6)
system, and the forces F,, F,, . . ., F, as the corresponding gene-
ralized forces, since the elementary work ol these forces is equal to The system has n frequencies

and the corresponding amplitude columns u i , U,, . . ., Un; t h e

free oscillations are determined by the formula
When investigating free oscillations, we take for the forces
F;, F,, ..
., F,, the elastic forces F:, . . ., F; acting on the masses (8)
mi, m,, . . ., m, of the elastic system S, given the deflections
.
yi, yz, . ., y,. The system under consideration of n particles acted where C j and aj (j = 1 , .. ., n) are arbitrary constants determined
upon by elastic forces is a conservative system and has a definite from the initial conditions.
potential energy II (yi, y,, . . ., y,). Expanding the function Let external forces F i , ..
., F, give rise t o static deflection
II (yi, yz, . . ., y,) in a power series and retaining only the quadra- pi, .. ., yn. Then the forces Fi are balanced by the elastic forces
tic terms (see Sec. 40), we get for II the expression F; (F.2 - -F:; i = 1, ..
., n) and for this reason, according to
n equation (2),
"2
2
2 CikYiYk ( ~ i k= ~ k i ,i, k = 1, . . ., n) (1)
n

i, k=l
 224 Small Oscillations ICh. 6
Sec. 441 Small Oscillations of Elastic Systems 225
An important role in the study of elastic systems is played by
l the inatris G = 11 gik ,:(l which is the inverse* of the matrix
2". g i k > O f o r l i - k l < l (i,k= l, ...,
n).
l 3". The determinant det G = ( g i k : 1 > 0 .
C = I[ C i k ! l :
n
Matrices having the properties l", 2" and 3" are called oscillatory
p1 -- c-l matrices.
Using matrix G i t is possible to solve the system of relations I t will be noted that for every positive definite matrix G , the
(9) for y , , . . ., yn and to represent i t in the form property 3" and also the inequalities 1" for the principal minors and
I n the inequalities 2" for the diagonal elements gii (i = 1 , . ., n) .
are valid. However, the nonnegativity of the nonprincipal minors
I
The quantity g i k is equal to the deflection a t the point (i)
caused by a unit external force applied a t the point (k), and is
called the influence coefficient of point (k) on the point (i) (i, k =
= 1 , . . . , n). From the symmetricalness of the matrix C there follows
the synlmetricalness of the inverse matrix G composed of the
influence coefficients*"
gik = gki (i, k = l, . . . , n) (11)
ailcl from the positive definiteness of the forms (3) there follows
the positive definiteness of the quadratic form
Fig. 51

of any order p* and the positivity of the elements g12, . . ., gn-4, n

12 are specific properties of the matrix of influence coefficients of a
since the quadratic form 2
cikyiyk goes into the form (12) in
.
4 . h=! -
linear elastic system.
From the oscillatory nature of the matrix of influence coefficients
the case of transformation of variables (10): follow the basic "oscillatory" properties of elastic oscillations of
a linear system:
1". All frequencies are distinct:
i, k = l i= 1 i, k = i q<o2<. ..<on
Consider a linear elastic system S-a string or a rod with ends 2". All the amplitudes uii, u , ~ ,. . ., unl in the first principal
fixed as usual. I t may be shown that in this case the matrix of the oscillation (with frequency oi) are different from zero and have
coefficients of influence, G , possesses the following properties: the same signs.
I". All minors (and not only the principal ones!) of any order
of the matrix G are nonnegative:
.
3". Among the amplitudes u l j , u g j , . ., u n j in the jth principal
oscillation (with frequency toj) there are exactly j - 1 changes
.
of sign ( j = 1, 2 , . ., n) (Fig. 51).
An investigation of oscillatory matrices and a substantiation of
oscillatory properties of elastic oscillations go beyond the scope
of this book.**
Example. Consider the classical problem of the oscillation of a string of
length 1 with attached ends for the case when the entire mass of the string is
* The matrix C is not singular (det C # 0), since the quadratic form (3) concentrated in n equidistant (from one another and from the endpoints) points,
is positive definite. the concentrated masses being all equal (and equal to m) (Fig. 52).
** Equation (11) expresses the so-called hlaxwell principle of reciprocity: --

"a deflection a t point (i) due t o a unit force applied to point (k) i s equal t o the * I n particular the nonnegativity of the off-diagonal elements of the mat-
deflection at (k) under a unit force applied to point (c)". rix G ( g i h >, 0 when i # k ) .
** This material can be found i n [7].
226 Small Oscillations [Ch. 6 Sec. 441 Small Oscillations of Elastic Systems 227

Tho elongation of the ith segment (between points with deflection p i and Here, each of the 'Lboundary3vconditions (IS) is satisfied automatically, and
rleflection yi+,) will be expressed (to within fourth-order infinitesimals) as the second one yields the condition for determining the desired frequencies:
follows:
sin (n + l ) B= 0 (20)
Hence, Oj=-- in (j= .
l , . . , n) and, by (16),
n+l

Taking the tensioll a of the string to be constant*, we get an expres-

sion for the potential energy:

The kinetic energy is of the simple form

n To find the amplitudes of the i t h principal oscillation uijr u2jr ..., unj
we put O = B j in equations (19):

We choose an indirect path to find the principal frequencies and the cor-
responding amplitude vectors. Write equations (G) for the amplitudes, making An arbitrary free oscillation of the system is determined by the formula
n n --
yk- 2 Cjukjsill (ajt+aj)-- 2 C j sin n-
kin sin 2 sin----- in
+l ( a c . + l , V ~t S a j
/ C

j= 1 j= i
(23)
From (21) and (22) it is a t once evident that the principal oscillations obtain-
Fig. 52 ed possess the oscillatory properties 1' to 3'.
Lagrange demonstrated how i t is possible, using these formulas and a limit
passage, to obtain the free oscillations of a homogeneous string (with fixed ends),
use of the expressions (14) and (25) for TI and T. Divide each of the equations (6) the mass of which is no longer concentrated a t n points b u t is distributed uni-
termwise by c and introduce the abridged notation: formly along the string, which has density p.
In this problem we put m = @ , and then find the discrete analogue of a
llomogeneous string wit,h principal freqaencies
where 0 is an auxiliary quantity. Then equations (6) will take the following
Form for the amplitudes:
k = I , ..., n)
c ~ s O + u ~ + ~ - -( O (17)
where I n the l i m i t , as n + m , we get the following familiar expressions for t h e
frequencies oj of a fixed homogeneous string:
uo=un+i=o (18)
The algebraic equations (6) may be satisfied by putting**

These formulas express the Mersenne law, according to,which a l l frequencies

--
-
..
* This supposition i s justified i n that only small deflections p i , y2, ., y,
a e considered.
are integral multiples of the frequency of the fundamental tone q = -
and each of the frequencies is directly proportional to the square root of the
3
'
:
x* Substituting (19) into (27) we get the trigonometric identities tension and inv~rsely proportional to the length and the square root of the
s i n ( k - - l ) U - 2 s i n k 0 ~ 0 ~ 0 + s i n ( k + 1 ) ~ = 0(k=O, l,..., n) density.
2 5*
 Sec. 451 Small Oscillations of Scleronomic System 229
228 Small Oscillations [Ch. 6

Give the jth harmonic oscillation of a homogeneous string i n the form Since the coordinate origin is a position of equilibrium, all the
generalized forces, for zero coordinates and velocities must be
yI. ( x , t ) = u j ( z ) s i n ( o j t + a,) ( o < x , < I ) (26) zero, i.e.
where uj(x) is the amplitude deflection in this oscillation.
Considering that the amplitude deflection uj(x) may be obtained from Qio=O ( i = I , ..., n) (4)
the quantities (22) by passage to the limit We introduce the following notation:
u j (x)=lim u k j

as n tends to infinity and -k1

- + X , we get from (22)
n+l
Then, rejecting in the expansion (3) all terms of second and hig-
her orders of smallness, we will have
Then the free oscillation of a homogeneous string, which is obtained by
linear superposition of the principal oscillations ( X ) , is expressed by the
formula
W

y (X, t)
where C j and a j ( j = l , 2,
= 2 C j sin -
jnx
1
sin ( o j t f a,) Substituting into the Lagrange equations (1) the expressions (2)
and (6) for the kinetic energy and for the generalized forces, we
j=i
get the linear differential equations of motion for small oscillations
.. .) are arbitrary constants. of a scleronomic system:

45. Small Oscillations of a Scleronomic System under

the Action of Forces Not Explicitly Dependent
on the Time Denote by A , B, C the square matrices*
Write the Lagrange equations for a scleronomic system for the
case when the generalized forces Q idepend solely on the coordinates
and the velocities: .
and by q the column made up of g,, . . , q,. Then in matrix
notation, the system of differential equations (7) will look like
14ir'+~i+~q=~ (8)
Let the coordinate origin be a position of equilibrium. Then We shall seek the solution of (8) of the form
(see Sec. 40) the kinetic energy, to within third-order infinitesimals q = uebt (9
in gi and qi (i = I , : . ., n)? may be represented by a quadratic
form with constant coefficients: where u is a column with constant elements U,, . . , U,, and y .
is a number.
Substituting (9) into the matrix equation (8) and cancelling efit,

where aih =ski (i, k- l,. . . , n).

we get

or, in expanded form,

+
(Ap2 B P + C) U o - (10)

Let us now expand the generalized forces Qi (gk, qk) in a power n

series of qk and qh
2 (aihp2+bi@ + ~ i kUk) = 0
k= l
(i = I , . . ., n) (10')
m

* Note that A is a positive definite symmetric matrix (this fact is not uti-
li zed here).
"0 S m a l l Oscillations [Ch. 6 Sec. 461 R a y leigh's D issipatiue Function 23 1

For the system (10) or (10') to have a nonzero solution U , i t

is necessary and sufficient t h a t the determinant of the system be 46. Raylleigh's Dissipative Function. The Effect of Smallll
equal to zero: Dissipative Forces on the Oscilllations of a Conservative System
A(y) t det(Ap"fp+C)=O (11) Consider a n important special case when the asymptotic stability
of a position of equilibrium is predetermined and there is no necessi-
or in the expanded form, t y to resort to the stability criterion given in Sec. 39.
In the expressions for the generalized forces [see equation (6) on
gage 2291
n . n
Qi= - E bikqk- E cikqk (i = 1 , . . . , n) (1)
k= i k= i

let the matrices :

= 11 bik 11 and C = )I c i k 11 :, which are composed
of the coefficients, be symmetric and positive definite.
Equation (1.1) is called the secular eqr~ationof the given system. Then, by introducing the potential energy I7 and Rayleigh's
This is an algebraic equation in p of degree 2n. dissipative function R (see Sec. 8), which are given by the positive-
We confine ourselves to the basic case when all the roots of the definite quadratic forms
secular equation p i , . . . , pzn are distinct. To each root ph there cor-
.
responds a certain nonzero solution zeh (uih, , ., u , ~ of ~ )t h e system
of homogeneous algebraic equations (10) and, hence, a particular
solution uheWhtof the system of differential equations (8) (h = 1 ,
. . ., 2n). The general solution of this system is obtained as a linear
combination (with arbitrary constant coefficients) of these particu-
lar solutions:
we c,an rewrite formulas (1) as

Of particular importance is the case when all real parts of the

roots ph are negative: Aside from the potential forces -
aqi
(i - 1, . . . , n), the system
is acted upon by the dissipative forces - ,
aH
defined by the Rayleigh
aqi
In this case, the position of equilibrium of the system is asymptoti- function. I n Sec. 8 i t was explained that in this case the system is
cally stable not only for the linearized system (8) but also for the definite dissipative. Since, according to the first of the formulas (2).
initial nonlinear scleronomic system with the differential equati- the potential energy in an equilibrium position has a strict minimum
ons (1) (see Sec. 38). and the equilibrium position is isolated, it follows that (see the theo-
In conclusion, note that for a conservative system, B = rem on page 178) the equilibrium position is asymptotically stable.
= 11 bik 1:) = 0, whereas, A = 11 a ~ [kI: and C = 11 cik 1: are symmetric Thus, the dissipative forces defined by the Rayleigh function not
positive-definite matrices. The secular equation det (Ap2 C) = 0 + only do not disturb the stability of the equilibrium position of a con-
servative system, but even make this position (in certain cases)
passes into the equation det (C- LA) = 0 from Sec. 40, if we put
asymptotically stable.
y = i)/x ( i = 1 / 7 1 ) .But, as was shown in Sec. 40, the equation I n the case under consideration i t is possible to establish simple
det (C- LA) = 0 has only positive and real roots. Therefore, in formulas to evaluate the .roots of the secular equation. Let us again
the case of a conservative system, equation (11) has only imagi- seek a solution of the form uebt. To determine the column U , we gel
nary roots. the equation (see page 229)
232 Small Oscillations [Ch. 6 I Sec. 461 Rayli.igh's Dissipative Function 233

or, in expanded form, If we have two real roots y and p' and if we denote the correspon-
n ding columns by U and U', then by multiplying both sides of the ith
2 (a&'
h= i
+ bikp + CIR)uk =0 (i = I , . . ., n) (40 1
equation (4') by U; (in place of Li), we get the following equation
in place of (5):
Multiplying both parts of the i t h equation of (4') by ui (ill is the
complex conjugate of ui) and summing over i, we find
n n
Interchanging the roles of the vectors U and U', we conclude that
the number p' also satisfies the equation (8). Therefore
or in abridged notation

Let us now see how the principal oscillations of conservative sys-

where A ( U , G) >0 , B (U, G)> 0 and C ( U , G ) >O.* tems change under the action of small dissipative forces [see 271.
Thus, any root p of the secular equation satisfies the quadratic We introduce the normal coordinates g,, . . ., 9,. I n these coordinates
equation (5) with positive coefficients. From this i t follows iminedia-
tely that Re p < 0.
If the secular equation has the complex root p = y iB, then this +
same equation also has the complex conjugate root = y - i6. c
The numbers p and F are roots-of the quadratic equation (5). There-
fore, assuniing U = v +
i w , U = v - i w (v, W are real column
vectors), we can write **
-
2Rep=p+p= - B(1'? V ) ) - - B("> v) t B ( w 7 W ) where oi (i = 1 , . . ., n) are the principal frequencies of the con-
A (U, G) A(c, v)+A(w, W)
servative system, and the coefficients i n the expression (11) for
I P I ~ = P P = - Ac (( Uu ,, G4) - Ac((Vv ,7 u +
U) C(W, W )
)+A(w,w)
the Rayleigh functions pi, Pik (i, k = 1, . . ., n; pih = Phi for
i # k) are small (squares and products of these quantities can be
1
ignored). From the positive definiteness of the quadratic form (11)
To the complex conjugate roots p and F'
there correspond the com- i t follows that pi > 0 (i = 1 , . . ., n).
plex conjugate oscillations ueNt and Lgt. I t is possible to reduce the We form the Lagrange equations
sum of the appropriate terms in the expression for q [see formula (12)
on page 2301 to the real form with complex conjugate values of the
1
arbitrary constants C = - ( F
2
1
+
iG), C = - ( F - iG):
2
-- -
+
C ueht + CuePt = - (F iG) (v ire) e(v +-is)t
1
2
+ + Then, substituting

= He {[Fv -Gzu + i ( F w + Gu)] eyt (cos 6t + i sin 6t)) = and cancelling ept, we get the following system of linear equations:
= eyt [(Fu - Gw) cos 6t - (Fw +Gv) sin 6t]
* See 5' on page 206.
** See formula (20) on page 206.
I
234 Small Oscillations [Ch. G Sec. 471 E f e c t of Time-Dependent Esternal Force on Small Oscillations 235

Equating the determinant of this system to zero, we find the Taking the linear combination of the given oscillation and the
secular equation: complex conjugate oscillation, we get a "principal" oscillation which
corresponds to the roots
- P
p i = ----!-+ioiB2 and p i = - 2 -2 i q ;

Expanding the determinant and neglecting terms containing

products of the smzll coefficients pi, Bik, we can represent the
secular equat.ion in the following form:
To a first approximation, the roots of the secular equation are
of the form
We get similar expressions for t,he other principal oscillations as
weli.
Thus, to a first approximation: (1) small dissipative forces do not
alter the frequencies of a conservative system; (2) under these forces
the oscillations decay as t -+ m ; (3) in the jth principal oscillation
all the coordinates are small compared with the j t h coordinate
and differ from i t in phase by a quarter of a period (j = 1 , . . . , n).

-
Let us find the amplitudes X,, xz, .... xn for p pi = - -P1_ 4-im i '
2
Substituting this value of y into the coefficients of the last n - l
47. The Effect of a Time-Dependent External Force on
Sniall Oscillations of a Scleronornic System. The
equations (14) and dividing the left sides of the equations by p,, Amplitude-Phase Characteristic
we get In addition to the forces spoken of in Sec. 45, let the forces Q i(t)
(i = 1, . . . . n) act on a scleronomic system. Then the Lagrange
equations for small oscillations of the system will differ from the
equations (7) on page 229 solely in the presence of nonzero right-
hand sides:

From t h i s system we determine the ratio -

Xk
xi ( k = 2, ..., n), dis-
.
carding second and higher-order terms i n pi, pih: The general solution of this nonhomogeneous system of differen-
t i a l equations is given in the form
Xk
LE^,
Xi
. eh= --
Pklmi
0"k W;
(k=2, .... n) (19)
Formulas (19) show that are small real quantities (k= 2, ..., n).
To the root p, = - A B + im, there corresponds the "complex oscilla-
2 where the first summand is the general solution of the corresponding
-tionV(we assume X, = Aeia, A > 0): homogeneous system, and Q* is some particular solution of the
system (1).
We assume t h a t the position of the system qi = . . . = q, = 0
is a n asymptotically stable position of equilibrium, that is, that
Reph < 0 (h = 1 , .... 2n). Then the first summand tends t o zero
236 Small Oscillations [Ch. 6
Sec. 471 E f f e c tof Time-Dependent Ezternal Force on Small Oscillations 237
as t tends to infinity*, and for sufficiently large t the general solution
q of the nonhomogeneous system practically coincides with q*.
For this reason, in the future we shall be interested only in "forced -
-
Then the response of the coordinate qk to the external action Q,
AeiQt is obtained by multiplying this action by the frequency
-
oscillations" q*, which we shall denote simply by q . characteristic W i k (iQ):
Since the system of differential equations ( 1 ) is linear, the general q k = W i k(iQ)AeiQt (9)
case for seeking forced oscillations reduces (at the expense of a linear Putting
superposition of particular solutions) to the case when only one of
the generalized forces Qi ( t ) differs from zero. Wik ( i Q ) =Rik ( Q )ei'J',,p)
For instance, let Q, ( t ) # 0 and Q j ( t ) = 0 (j = 2 , ..
., n ) . Also [ R i k ( Q ) > 01 (10)
let us first assume that Q, ( t ) is a harmonic force, i.e. R t k ( Q ) is the amplitude characteristic and 'Pih(Q) is the phase
Qi ( t )= AeiQt (3) characteristic], we write t h e formula (9) as follows: '

Then the differential equations ( 1 ) will be written as iIQf+Y,,(Wl

qk = R4k ( Q )Ae
( k = l , ..., n) (11)
Now let
Q, = AsinQt (12)
i.e.
Qi A (eiPt
=- e-iQt
2i )
We seek the forced oscillations in the form
qk = BkeiPt ( k = l , . . . , n) The corresponding response will be*
(5)
Substituting these expressions for q, (k = 1 , . . ., n) into the diffe-
rential equations ( 4 ) and cancelling eiQt,we get the following system t h a t is,
of algebraic equations for a determination of the quantities B,:
qk = Rlk ( Q ) A sin [Qt 4- Y i k ( Q ) ] (13)
In other words, when passing from a sinusoidal force (12) to
the corresponding response, i.e. t o the sinusoidal forced oscillation

Solving this system, we find for k- l , . . . , n:

Bh -- W i k(iQ)A (7)
where
Aik (in)
W i k(iQ)= -
A (in) (8)
is a regular fractional-rational function of iQ with real coefficients; Fig. 53
the hodograph of this function in the complex plane, and sometimes
the function itself, is termed the frequency characteristic or the ampli- ( 1 3 ) , the amplitude of the force is multiplied by the amplitude
tude-phase characteristic. characteristic R i k ( Q ) , and the phase shift is determined by the
phase characteristic YYlh( Q ) taken for the same value of Q .
* If the secular equation has multiple roots, then the sum on the right side Fig. 53 depicts the amplitude-phase characteristic W i l l (iQ)
+
of (2)can involve secular terms of the form Ch (uh u h t uitz+ +
. . .) e h t . ( 0 < Q < W). If for a given Q the corresponding R,,+ (Q) is very
However, in this case too the sum tends to zero as t tends to infinity if all
He ph < 0. * Since W i k (iQ)and Wik (-iQ) are complex conjugate numbers, it follows
that Rik (-Q) = Rih ( Q )and Yik (-Q) = -Yik (Q).
238 Small Oscillations [Ch. 6 Sec. 471 E#ect of Time-Dependent External Force on Small Oscillations 239

small, then the amplitude of the response is extremely small com- Therefore, proceeding from the principle of the linear superposition
pared with the amplitude of the "excitation" A sin Qt of the given of responses, we find
hequency Q . On the contrary, if for a given Q the corresponding
R I , ( Q ) is very great, then the response amplitude is great in compa-
rison with the amplitude of the generalized force Q i . Thus, selecting
qk = --I
2n
+Sm
-cc
W i h(iQ)F (Q) eiQtdQ

a system with suitable amplitude characteristics, we can dampen ( k = = 1 , . . ., n )

the oscillations in certain frequency ranges and increase the ampli- (19)
tudes of such oscillations on other frequencies. This is the principle that is, the complex spectrum W i h . ( i Q ) F ( Q ) for the coordinate
on which filters are based. qh is obtained by multiplying the complex spectrum of the action
Since W ( i Q ) is a regular fractional-rational function and for this Q1 ( 1 ) by the corresponding frequency characteristic of the system
reason W ( i Q ) -+ 0 as Q -+ oo, any system actually passes only Wik ( W .
a finite range of frequencies. Let the system be a t rest for t < 0, and the motion of the system,
Now let Qi ( t ) be an arbitrary periodic function specified by the given zero initial conditions, be caused solely by the external action
Fourier series
m
+
Q ( t ) 0 when t > 0. Giving the complex spectrum of the response
in the form
Qi ( t ) =
m=O
Amsin (mQt + qm) (14) 1; ( Q )W i k ( i Q )= G (Q) $ iH ( Q ) (20)
Combining responses to separate harmonics of this series, me get we will have, according to the formula ( 1 9 ) ,
CO

qk =
m=O
R i k (mQ)A, sin [mQt fr p , 4-y i k (mQ)l 2n
[G (Q) -1iH (B)][COSQt + i sin Qt] dQ =
-cc
(k=!, . . ., n ) (15)
Now let Qi ( t ) be an arbitrary nonperiodic function f ( t ) of t ,
which may be given in the form of a Fourier integral
- 17
2n
-cc
[G (Q) cos Qt -H (R) sin Qtl d n +

Since qk (tj is a real function, the second summand on the right.

is zero* and for this reason

we will have
In,
Now suppose that
Qi ( t ) =0 ,
The function F,(S2) is called the complex spectrum of the function qk(t)=O (k=1, ..., n ) for t <0
Qi = f ( 4 . then
The action
1
-2Jc
F ( Q ) dQeiQt 0 =l
2% 7 [G (Q) cos Qt - H ( Q ) sin RI] dQ (for t <0 )
will give rise to the response
1 W l k (iQ)F ( Q ) dQeiQt " A real system cannot have a frequency characteristic that does not sa-
2n tisfy this condition.
240 Small Oscillations [Ch. 6
? Sec. 471 Efect of Time-Dependent Ezternal Force on Small 0scillation.s 241
I

Replacing here t by -t, we have i where Qj,, and A j are shown i n Fig. 55, and A j is the area of the
trapezoid. Therefore

0=1
2n
7
-m
[G (Q) cos Qt + H (Q) sin Qt] dQ (for f > 0) (23)
1
i qk F
1
2 sin Qj,,t sin A jt
Aj Q1av
i
Ajt

Adding the equations (21) and (23) termwise, we have I where the summation is taken over all trapezoids obtained in the
l approximation of G (Q) by a polygonal line.
Tables of the functions may be used for such a n approximation

I t may be shown that the expression (24) is the general solution

of a stable system for the coordinate qk, for the conditions (22)

Fig. 54 Fig. 55

and for zero initial conditions *. The function Wik (Q) is constru-
cted directly from the coefficients a i k , bik and cik, which appear in
equations (l), and the function F (Q) is determined by the expre-
ssion (17). After that- the problem of determining the motion
described by equations (1) reduces to a single quadrature with the
help of (24).
For this reason, any method of approximate integration is suitable
for an approximate determination of the motion. For example, it
is possible to construct a graph of the function G (Q), replace i t with
a polygonal line and through the salient points draw horizontal
lines to the axis of ordinates. Then the function G (Q) is approxima-
tely replaced by the sum of the functions g j (Q), the graph of each
of which is a trapezoid (a triangle, in the special case), as shown
in Fig. 54.
The integral (24) may be computed for one such function
g j (Q), as follows:
cc
1 A j sin Qj,,t sin h$
qjk = gj (Q) cos Rt dQ = _
Qjao Ajt
-m

* This follows from the fact that expression (20) defines the Fourier trans-
form of the solution for zero initial conditions, and expression (24) defines the
inverse Fourier transform.
 Sec. 481 Reduced System. R o u t h Potential. H i d d e n Motions 243
CHAPTER 7
Here, baR and yai are functions of noncyclic (or, as they are
sometimes called, position) coordinates qi (i = l , . . . , m). Putting
Systems with Cyclic Coordinates
expressions (6) for h,
into formula ( l ) , we get the expression T^
for the kinetic energy in Routh variables:
m n m n
48. Reduced Systcm. The Routh Potential. Hidden Motions.
Hertz' Conception of the Kinetic Origin of Potential
Energy
In the present chapter the general propositions given in Chapters
A remarkable circumstance (one noticed by Routh) is t h a t -
2 and 5 are used for investigating the motion of a holonomic sclero- in this formula all the a;, = 0, i.e. the e.upressi0.n for T is the sum
nomic system with cyclic coordinates q, ( a = m 4- 1 , . . ., n). of the quadratic form in the position velocities pi, . . . , Q m and
The kinetic energy of such a system is of the form the quadratic form in the generalized momenta p,+i, . . . , prim*
n Indeed.

Let us find an expression for the kinetic energy i11 Routh variables n

q i , qi, p, (i ==l , . . ., m; a = m + l , . . ., n). To do this, express since 2 baBpRis dependent only on the variables qi and pe,
all q, in terms of p, (U = m +
1 , . . ., n), utilizing the initial
P=m+ i
which are regarded as independent with respect to the
relations
m n
qi ( i = 1 , . . . , m; p = m + l , . . ., n).
Now compute the coefficients a&:

Since the determinant D = det (aae);, p,,n+l # O)*, from relations

(2) we get
n m
( a , p = m + l , . ., n).
Similarly, we find** the coefficients aTj

where j[baRlj:+i is the inverse matrix of the matris 11 a , ~\ c f i ,

Il b a Il~= ll a a Il-l
~ (4)
Setting
n
yai= 2 baRaBi ( i = 1 , ..., m ; a = m + 1 , ..., n) (5)
* The transformation of variables used here t h a t corresponds t o a transitior~
B=m+ i
we write (3) in the following form:
. .
from the variables qi, g, to the variables qi, p, i s ordinarily used in the theory
of quadratic forms to reduce (by Lagrange s method) a quadratic form t o a sum
of squares. Indeed, applying such transformations a number of times, we can
represent a quadratic form in n variables in the form of a sum of n quadratic
forms, each of which is equal t o the product of the square of this variable b y
some real coefficient.
* See footnote on page 78. i, i,
** Here and henceforward =. . . . signifies that all the in the expression
in square brackets are replaced b y their expressions (6).
244 Systems with Cyclic Coordinates [Ch. 7 Sec. 481 Reduced Sustem. Routh Potential. Hidden Motions 245

n Consider the function H* (t, qi, p,), which we shall call the
a -- 3 T
[$l.
a2T dq,
Routh potential:*
.. . .
aqi @- j + a=m+i
-7=
aqj 39, aqi n
q=
,
n*==II+ 1
2 bafipaps (13)
a,B=m+ l

Taking advantage of the fact that

Utilizing equations (5) we get Apa
bar3 = -jy
where ARa is the cofactor of the element aOa i n the determinant D,
we can write the expression for the Routh potential also in the
But on the basis of (4), bap= 4, , where AB, is the cofactor following form:
I 0 Pm,, -
Pn I
l
of the element aoa in the determinant D. Taking advantage of
this circumstance, in place of (10) we can write Pm+, arnti, m t i . - . a m t i , n
20

l am+,,j am+,, m t i -- amti, n

(i, j = I , ..., m) (10')
aS=31 . . . . S . . . . . . . . . . We also introduce the notation

Thus, formula (7) has the form

Then, according to (12), the function -41, which for the posi-
tion coordinates plays the role of the Lagrangian function, is
Here, the coefficients a$ and bafi are determined by the equations
(4) and (10). where V is the generalized potential defined by the equation
Let the forces impressed on the scleronomic system have the
potential II = ll (t, qi). Then, L T - IT. Compute the Routh fun-
ction (see Sec. 13):
-
n
Let us now consider some motion of the initial system. In this
motion,
Pa=cc)nst=ca (a=m+1, ., n) ..
(17)
and the variation of the position coordinates qi = qi (t) (i = 1, .. .
. . . , m) may be determined from the differential equations (8) on
page 81, in which p, should everywhere be replaced by c, (a =; m +
+ .
1, . . , n). But these equations are Lagrange equations (with the
Lagrangian function -R = T* - V) for a certain auxiliary natural
scleronomic system with m degrees of freedom having the kinetic

X Routh E . G . , The Advanczd P a r t of a Treatise on the Dynonzics of a S!/-

stern of Rigid Bodies, 5 t h e d . , 1892, Sec. 99.
 S e c . 481 Reduced System. R o u f h Potential. Hidden Motions 247
246 Systems with Cyclic Coordinates [Ch. 7
Eesides, from ( 1 0 ) i t follows that for the gyroscopically uncoupled
energy T*=T
1
m
..
2 a?,qiqj and the generalized force potential
system a 4 = a i j ( i , j = 1 , . . ., m ) , i.e.
m
i , j=i
m n -

i, j=i
V-- - 2( 2:
i=i a=m-ki
,i- ll* (tf yi, C,).
y a i ~ a ); The Routh potential
In this case, the initial system has the kinetic energy
TT* ( t , qi, ca) is the potential energy of this system. From formu-
las ( 1 1 ) to ( 1 3 ) i t follows that T = -1 2 . . +81 2
a 2/9iqj
.. . b a ~ ~ a ~ B
2
i, j = l p=m+i a,
and the potential energy IT = 11 ( t , qi) while the reduced system
We shall call this auxiliary system a reduced system.
Thus, the variation of the position coordinates q , , . . ., q,,, deter-
1
n
..
mines ihe molion of a reduced system (with m degrees cf freedoln) with
has the kinetic energy T* = -
2
2 i, j=m+i ai jqiqj and potential energy
kinetic energy T* and with generalized potential V -- V , +
II*, where n
IT* is the R o ~ ~ potential.
th Given appropriate motions of the initial
and reduced systems, the total energies of these systems are equal.
When the functions q, = qi ( t ) ( i = 1 , . . ., m) which define the

the cyclic coordinates q, -

q, ( t ) ( a m -
motion of a reduced system are found, then the time variation of
1 , . . . , n) may be
determined from the formula (9) on page 81, which can now be (
We see that part o f th,e kinetic energy o f the initial system

2
n

baBc,cB) h a s passed into the potential energy of the

written as a , B=m-ki
m reduced system.
All that has been said about scleronomic systems holds true, in
particular, for conservative systems in which
OL
<E
= 0 i. e. H =
= H (gi). The reduced system for a gyroscopic uncoupled conser-
\.ative system mill again be conservative.
Now let there be given an arbitrary conservative system with
Consider a special case when the expression for the kinetic energy
of the initial system ( 1 ) does not contain products of the position 2
- In degrees of freedom, with kinetic energy, T* = - 2
In
aijx
velocities qi by the cyclic velocities g,, i.e. a case when all a,, 0 .. i , ,=i
( i = 1 , . . ., m; a = m +
1 , . . ., 71). I n this case, the kinetic X (91, . . .. U,,,) qiqj and potenlial energy H* = H* (qi). Consider
a
energy T breaks up into two quadratic forms, of which one contains conservative system in which the number of degrees of freedom is
only the position velocities q i , while the other contains only the
m + 1 and which has m position coordinates g,, . ., qm and one .
cyclic coordinate q,,.
cyclic velocities qa. Such a system is termed gyroscopically u n - In the new system let II = 0 and the kinetic energy have the fornl
coupled. For a gyroscopically uncoupled system, according to
formulas (5), all Lhe y a i = 0, and hence, Vi = 0 and V = H*.
Thus, if the initial system is gyroscopically uncoupled, the reduced Then
system has the ordinary potential H*. *

*
an*
The reduced system is acted on by potential forces --8 4 ; and gyros-
copic forces determined by the potential Vi. In the case of a ~ ~ y r o s c o p i c a l l y
free system V i = O and gyroscopic forces arc absent.
248 Systems with Cyclic Coordinates [Ch. 7
Sec. 481 Reduced System. Routh Potential. Hidden Motions 249

Rut then for p,+,= h t i =1/% angle of pure rotation. Let 1 = OD and A and C be the equatorial and axial
moments of inertia, respectively. Then

n* -
Thus, in the given conservative system, T* is the kinetic, and
II* (qi)is the potential energy, whereas in the new "extended"
system we have, respectively, T = T* + It* (gi) and II == 0.
where p, g, r a r e the projections of the angular velocity o on the principal
axes of inertia 06,011,05. But

T h e potential energy of the given system results from the kinetic energy
of the "extended" system which has a greaier number of degrees of freedom. therefore, we finally have
Motions, under which only cyclic (hidden) coordinates vary are 2~=~(h+@sin20)+~(cp'+$cos0)2,
called hidden motions. *
Above we saw t h a t the potential energy oE a conservative system II= Mgl cos 0 (21)
can always be regarded as the kinetic energy of hidden motions. Find expressions for the generalized momenta p,, p* corresponding t o
This conception about the kinetic origin of potential energy and, t,he cyclic coordinates cp, 9:
hence, about the kinetic origin of forces applied to bodies performing
explicit (nonhidden) motions was broadly developed by Hertz in
1894. **
Example. Let us consider the motion of a rigid body with fixed point 0
in the Lagrangian case when the body is acted upon by the force of gravity My, Solving these relations for 6 and 6,we have
when there is an axis of dynamic symmetry, and the centre of gravity D is locat-
ed on this axis. . Pu-COS 0pq
1
'=
.
cp=
Asin20
--c cos ep, +(A sin2 0 + C cos2 8) p,
AC sin2 0
/
J
(23)

The coefficients of p* and p, o?i the right sides of (23) are the quanti-
ties bU0 The expression for the kinetic energy in the Routh variables 8,
8, p*, p, is obtained by substituting expressions (23) for cp and $ into (21)
for 2T or straightway by formula (11)*:
(A sin2 0 + C cos2 0) p;
2~ = A & + Cp$ -2C cos Op,,,p,AC+sin2 0 (24)

When the system is in motion, the momenta p* and p, retain constant

values:
p ~ = a ,pq= b (25)
Fig. 56 The nutational motion is determined by the reduced system, for which
We shall specify the rigid body by means of the three Euler angles 0, 9, p,, 1 ( n -b cos 0)2 b2
where the angle of nutation 0 is the angle between the vertical axis Oz and the T* ==-A02, 17*=Mgl COS 0+
2 A sin2 0 +C
axis of dynamic symmetry 05 (Fig. 56), $ is the angle of precession, and p, is the
The change in the angle of nutation 0-0 (t) is found from the corresponding
* Systems containing within them gyroscopes are a brilliant example of energy integral
the manifestation of hidden motions. Because of these "hidden" motions of gy-
roscopes, the motion of such systems differs radically from the motion of sy-
stems without gyroscopes.
1
-
2
+
~ l bMgl cos 0 -F( a- b cos 0)2
A sil12 = conat =h
** H. Hertz, Die Prinzipien der Mechanik in neuem Zusammenhange dar-
gestellt, see also W. Thomson and P. Tait, Treatise on Natural Philosophy,
Part 1, Sec. 319 and A. G. Webster, The Dynamics of Particles and of Rigid, * In the given case the rigid body is a gyroscopically uncoul~led
Elastic and Fluid Bodies, 1925. system. so t h a t in formula (11) a ? j = n i j (i, j = i , ..., m).
 It
2 50 Systems with Cyclic Coordinates [Ch. 7 Sec. PS1 Reduced Sustem. Routh Potential. Hidden Motions 251

Introduce the auxiliary variable u=cos 8. IvIultiplying both sides of (27) ,

2 2 But if U, < < u2, then the velocity of precession changes sign when
by - (1-u2)=- s i n z 8 , we find
A A
cos 8" = Q , and the trace of the dynamic axis describes on the sphere a curve
h
as shown in Fig. 59.
where f (U) is a third-degree polynomial i n U: If =u2, then the velocity of procession 11, does not change sign, but vani-
shes on the upper parallel 0 = O2 (Fig. GO).*

Putting U=* 1 and u=uo=cos < 1, we find*

Then the polynomial f (U) has three real roots cos Oi, u2 = cos O2 and
1 -l<ui<uz<l<~L'
U,= U':

and [since f ( + m ) = + m ] the graph of the polynomial f (U) has the form
shown in Fig. 57. , 1
Fig. 58

Analytically, thc change in angle of nutation is found from the formulas

5:
du
1m--
t $ const, 8 = a r c cos u
- (31)

The plus sign is taken when 0 varies from O2 to 8, and the minus sign when
i t varies in the reverse direction. I t is obvious that the variation in time of

i
Fig. 57 l
1
Since when the body is in motion, -1 ,< u=cos 8 ,<+l and f (U)=
1 I
> 0, i t follows that U = cos 8 must vary in the interval ui U 4, <
>
i.e. 8' > 8 02. The angular velocity of precession is determined from the
first formula in (23):

*=
a-bcosO
Asin28 (30)

If the ratio lies outside the interval ui ,< u ,< u 2 , then the velocity nf preces- Fig. 59 Fig. 60
b
the angle (3 will be a periodic function 8 (t) with period
sion 11, retains a constant slgn and precessional motion occurs all the time in one
direction. In this case the trace left by the axis 05 on a fixed sphere n i t h centre U"

a t 0 describes a curve depicted in Fig. 58.

Here we assume that a # & b. The special cases a = + b will he con-

*
sidered below. Besides, the initial value O0 is chosen so that O0 0 and, + l After the variation of the angle of nutation 8 = 0 ( 1 ) is found, the variations
of the angles 11, and cp are determined from the formulas (23).
l~ence, (S), =+ 0.
* I t may be proved that cannot vanish on the lower parallcl.
17*
252 ' Sustems with Cuclic Coordinates [Ch. 7 J
Sec. 491 S t a b i l i t y of Stationary Motions 253

Let us now consider the motion of a rigid body passing through the ~Lsingular*? these quantities are the integrals of motion:
position 0 = 0. At this point the kinetic energy is given by the degenerate quad-
ratic form [see expression (21) for 0 = 01 H ( q i , Pi, pa)=H(P!, PP, P:)
I p a = p k ( a = m + I , . . ., n) (3)
According to formulas (22), the equation p + = p m t h a t is The motion of a system is called stationary if during i t all the
position coordinates retain constant values qi = const = q! (i =
should occur a t the singular point 0=0. Assuming t h a t the arbitrary con- = 1 , . . ., m).
stants n and b are connected by the relation (34), we readily evaluate the In stationarv motion all nosition velocities are zero and therefore,
indeterminate form in the expression (26) for the Routh potential:* , by equations "(l) and eqkality (a),
..-
m n

ii = h=E cik (PP)p a t a=mf

i Cia (g!) p: = o (i = I, . . . , m) (4)
l
Then the nutational motion is again dcterrnined from the energy integral From these relations i t follows,* t h a t in stationary motion all
T* + n* = const.
position 0 = n, then in place
If the motion passes through the L6sing~~lar3.
of (34) we will have thc relation
. .
position monienta are also of constant magnitude: pi =c.onst =P:.
..
Since qi = p i = 0 (i = l , . , m), i t follows from (1) t h a t for sta-
1
tionary motion
and the expression for the Routh potential will take the form
--
l[* Mgl cos 0 - +
a2 l +cos 0
fl l-cos0 Equations (5) are necessary and sufficient conditions which the
In the above considered two "~ingnlar--cases, thc upper (or, respectively, t
initial values g!, pp, p: (i --. l , . . . , m; a.= m -1- l , . . , n) must .
lower) parallel in Figs. 58-60 degenerates into a point. satisfy for t h e motion defined by them to be stationary.
Note also that in stationary motion the cyclic velocities
49. Stability of Stationary Motions Pa ( a - m -tl , . . ., n)
n
also have constant magnitude, since by (1)
aH
Consider a conservative system, the position of which is specified
ga -- a' ~=a c,, (PO)2 = const ( a =m 4 I , . . ., M) (6)
= 1 , . . ., m; cx -
b y m position coordinates qi and n - m cyclic coordinates q, (i =
m + 1 , . . ., n). The inotion of such a system
is defined by the canonical equations
Therefore
S= i

where the total energy of the systein I$ =H (gi, p i , p,) has the form
where the quadratic form on the right side is positive definite **.
When the system is in motion, the function H and the generalized
momenta p, (a = m -G 1 , . . ., n) do not alter their values, i.e.
* I n the expression for II* we discard the constant term
~ , = q k ( t - t ~ ) + & ( a = m + j , . . ., n) (7)
The initial cyclic coordinates q i are not arbitrary constants and do
not enter into the conditions (5).
Thus, in stationary motion the position coordinates retain constant
I value while the cyclic coordinates vary in accordance with a linear law.
Example.' The regular precession of a heavy symmetric gyroscope is a n
instance of stationary motion.
Indeed, regular precession is characterized by the equations
O = c 0 n s t = 0 ~ , $=const, cp=const
where the angle of precession $ and the angle of pure rotation rp are cyclic coordi-
* Relations (4) may be solved f o r the position momenta ph, since the
62 1 determinant det [cik (q!)]? h=i is differeiit from zero, inasmuch as the

:W This quadratic form is the kinetic energy of the system expressed in terms
of generalized rnomenta.
quadistic form 5
r , s=i
sr,prps is positive definite.
 254 Systems with Cyclic Coordinates [Ch. 7 Sec. 491 Stabilitu o f Stationaru Motions 255

l
nates, while the angle of nutation 0-an angle formed by the axis of the gyros- for the Lyapunov function. This function is the integral of motion
I cope and the vertical-is a position coordinate.* and has a strict minimum equal to zero a t gi = 0, q i = 0, q, = 0
l
Note that in accordance with relations (6), a slight change in the ( i = 1 , . . ., m; a - - m + 1 , . ., n). .
i initial quantities q:, p:, pO, yields a small change in the original
cyclic velocities qi (E = m $- 1, . . ., n). However, a small
Let us establish an analogy between the stability of the state of
equilibrium and the stability of a stationary motion. Consider a re-
duced system with m independent coordinates q,, . ., q, and with .
change in the quantities g:, according to formula (7), yields, as time potential energy equal to the Routh potential:
progresses, an arbitrarily large change in the cyclic coordinates
themselves. For this reason, stationary motion cannot be stable with
respect to cyclic coordinates.
Henceforward, by stability of stationary motion we shall mean

coordinates qi (i = 1 , . . ., m: a -
the stability with respect to all momenta p i and p, and position
m -F 1, . . ., n). **
We then have the following criterion of stability of stationary
motion.
Motion with the initial condztions q!, pi, pO, (i = 1 , . ., m;
a - m +
l , . . ., i t ) is stable stationary motion if the function
H (qi, pi, p:) at the point qi = q:, p i = p: (i = 1. . . ., m) has
a strict estremum.
Indeed, the motion under consideration will be stationary inasmuch
as i t follows from the existence of the extremum of the function
H (gi, pi, p:) t h a t the quantities g:, pp (i = 1, . . ., m) together
with the quantities p i ( a = m +
1 , . . ., n) satisfy equations (5).
We introduce the deviations
kxqi-q;, q i = p ~ - ~ ; , qaypa-p:.
Then, utilizing the motion integral H (g: + gi, p: +
q i , p:) as the
Lyapur~ovfunction, we can conclucle (see page 185) that the zero
solution gi = 0, q i = 0 (i.e. the stationary motion) is stable on the
assumption that the cyclic momenta p, do not experience perturba-
tions (i.e. that these quantities have the same values p: for pertur-
bed motion as for unperturbed motion). ***
However, the above-formulated criterion of stability is valid in
the general case as well, when for perturbed motion the quantities
11, = p, - p!' ( a = m + .
1 , . ., n) are different from zero. To
see this, it will suffice to use a corollary to the Lyapunov theorem
(see preceding section.). The reduced system is acted upon by gyro-
an*
scopic forces in addition t o the potential forces - a(l; (i = 1 , . . .
. . ., m). Since to the stationary motion of the i n i t i d system qi =
. - 47, pa = p: (i = 1 , . . ., m; a = m +
1 , . . ., n) there corre-
sponds an equilibrium position of the reduced system, i t follows
that the quantities q?, pO, must satisfy the equations
These are necessary and sufficient conditions for the existence of statio-
nary motion. They are obviously equivalent to the conditions (5)
and are obtained from the latter by elimination of the quantities
.
p i (i = 1 , . ., m). Applying the Lagrange theorem t o the position
of equilibrium qi = q? of the reduced system, we get the criterion
of stability of stationary motion in the following form.
Motion with the initial conditions q:, q: = 0, p i (i = 1 , . . ., m;
a =m +
1 , . . ., n) is a stable stationary motion if the Routh poten-
tial II* (gi, p:) has a strict minimum a t qi = q4 (i = 1 , . . ., m).
By applying the Lagrange theorem to a reduced system, we fixed
the constant values of the cyclic momenta p, = pO, ( a = m 1, .. . +
. . ., n). However, the criterion is still valid for variations of
momenta p: as well. To establish this fact i t is sufficient to take,
for the Lyapunov function, the integral of motion
11.

(page 185) taking the function 7;

I E * ( q 4 t E i , i i , ~ & ) - E * ( q f , 0, pk)I2+
(1 1)
n a=m+i
[H(q!+Ei, pY+?i, pk)-H(q!, P!, pL)I2+ a=m L1 11; (8) where E * = T * +
II * is the total energy of the reduced system
(it coincides with the total energy of the initial system expressed
* From (21) of the preceding section i t is seen that the coordinates cp and
11, do not appear explicitly in thc expressions for the kinetic energy and the in the Routh potentials qi, qi, p, (see Sec. 48), while g i = qi -
potential energy.
** Or, which is the same thing, thc stability with respect t o the quantities - 41, qi, qa = Pa-p: (i = I , . . ., m; a = m 1 , . ., n) are + .
deviations of perturbed m o t i m (from the given stationary motion).
qi,
***
bi, I-)=( i = 1, . . ., m; a = rn +1, . . .; n).
This is precisely the kind of stability that Houth considered. The function (11) has a strict minimum (equal to zero) at gi = 0,
256 Sustems W ith Cuclic Coordinates ICh.. 7
Sec. 491 Stability of Stationary Motions 257
9

q i = O , q a = O ( i = 1, . . ., m ; a = m + 1 , . . . , n). In addition, pe=(A+ Md2 sinz 0) 0.

The stability criterion of stationary motion given here was estab-
lished in somewhat different form by Routh in 1884. Write the expression for the Hamiltonian function in the variables 0, p , = a ,
I ~ ~ pe,= P ~h = Y P, O = ~ :
Example. Determine the stable stationary motions of a nonhomogeneous
ponderable sphere on a smooth horizontal plane if the centre of gravity of the
sphere, D ,is distant d from the geometric centre 0,the massof the sphere is M,

where

is the Houth potential.*

Here, the conditions for the existence of stationary motion (5) are of
m*
the form -=O, pe=O.
ae
The condition for stability-the presence of a strict extremum of the
1 function H at pe=O and a t a certain desired value of 0-will be fulfilled
l if for this value of 0 the function II* has a strict minimum. To find t h i s
value O=Bo, put u=cos 0 and

Fig. 61
the moment df inertia about the OD axis is C, and the two other principal cen-
tral moments of inertia are equal: A = B (Fig. 61).*
For the independent coordinates, take two horizontal coordinates of the
centre of gravity X,, y, and the three Euler angles rp, $, 0. Here, the angle cp
is the angle of '$pure rotation" about the axis of the dynamic symmetry 05
passing through the points D and 0;the direction of the 05 axis coincides with
the direction of the vector DO. Write the expressions for the kinetic and poten-
tial energies:
Suppose that' the equation f' (u)=O has a root U such that l U l l. <
I t is to this value I U I =cos 0 that the stationary motion of a sphere corres-
ponds, for which motion the centre of the sphere is in uniform rectilinear
Here p, .q, r are projections of the angular velocity w on the central axes motion and the angles cp and $ vary by a linear law.
of inertla. But
. . sin2 0, r=cp+$
p2+q2=02+$2
. . cos 0, zD= -d cos 0 >
To find the stability of stationary motion, we shall first prove that
<
f " ( U ) 0 when l U l l. Indeed, if f" (U) were equal to zero for 1 U l l, <
1+3uz
Therefore then from the expression for f " (U) it would follow that p=-. From
+
2T= ( A + Mdz sin2 0) 6 2 +A@ sin2 0 Crz +M
. +y.g )
(X&
> <
this i t is readily seen that I p ( 1 for I U I 1, which is igpo&ible, since
II = -Mgd cos 0 ,,(, <
p=- v 2 ~ 6 . Consequently, f " (U) # 0 when 1 U / 1, i.e. f;' (U) retains its

The coordinates XD, YD, cp and $ are cyclic. During motion, the appro-
sign 'in 'the interval (-1, + l ) . But fn (0)=y2+62 > 0. > 0 for
Hence, f " (U)
priate momenta retain constant values, namely: l u l < l. Since

p*= A$ sin2 0 4 Cr cos 0 = 6 1 * We do not consider the singular values 0 = 0 and 0 = n . Therefore
* The rigid body considered in this problem is a physical pendulum, the sin 0 # 0.
axis of suspension being the axis of dynamic symmetry, and t h e point of sup i ** Here y2+62>O since for y = 6 = 0 the function II*=-Mgdcos0 has
port 0 can slide freely (without friction) on the horizontal plane. a strict minimum at 0=0.
258 Systems with Cyclic Coordinates [Ch. 7
References
i t follows that for the value of 0 under consideration the function II* has
a strict minimum, i.e. the appropriate stationary motion is stable.
The condition for the existence of stationary motion f f ( u ) = O may be

. .
transformed by putting 6 = yu f A$ (1-us).
If we then put y = Cr = C (cpf 9 cos 0) into t h e equation obtained, this
condition will take the final form
..
Ccpll,f (C-A)$2cos 04-Mgd=O (15) 1. A1,~ell P. E. Trait6 de m k a n i q u e rationelle. Vols. I and 11. Gauthier-
~ i f l a r s ,P.aris, 1909.
This is the well-known condition for t_he existence of regular precession 2. Babakov L. M. Theory of Oscillations. Gostekhizdat, Moscow, 1958 (in Rus-
under the action of an external moment Mgd s i n e (the 111ornent of the ver- sian).
tical reaction N = M g about the pole D). 3. Bulgakov B. V. Oscillations. Gostekhizdat, Moscow, 1954 (in Russian).
Consider three separate cases. 4. Cartan E . Legons sur les invariants intkgraux. Ilermann, Paris, 1922.
>
1". If ( M ~ ~ + cl ; $I -4-C l @ then the condition (15) is not fulfil- 5. Chetayev N. G. Stability of Motion, 3rd ed. & & N a ~ kMoscow,
Russian).
a ~ ~ , 1965 (in
led for a n y real value of 0, and there does not exist a n y stationary motion
with such angular velocities. 6. Corden H. C.. Stehle P. Classical Mechanics. Wiley, New York; Chapman,
London, 1950.
<
2". If I Mgdf C;$ I IA-C 1 6 2 and the quantities M g d + c i $ and A-C 7. Gantmacher F. R., Krein M. G. Oscillatory Matrices and Nuclei and Small
have the same sign, then for such angular velocities there exists stationary Oscillations of Mechanical Systems, 2nd ed. Gostekhizdat, Moscow, 1950
>
motion with cos 0 0.
.. (in Russian).
<
3O. But if I Mgdf C@$ I IA-Cl and t h e quantities Mgd+Ccp$ and 8. Goldstein M. Classical Mechanics. Addison-Wesley, Reading,
setts, 1951.
- Massachu-
A-C have different signs, then cos 0 < O in the case of stationary motion.
I n this case there exists stable stationary motion such that the centre of 9. Jacobi K.G. J . Vorlesungen uber Dynamik. Clebsch, 1866.
located above the geometric centre of the sphere. 10. Lagraage J . L. MQcanique analytique. Paris, 1788.
Now consider special cases. 11. Lanczos C. The Variational Principles of Mechanics. University of Toronto
(a) 00=0: Then from formulas (12) i t follows t h a t y =6. Therefore Press, Toronto, 1949.
12. Landan L. D., ~ i f s h i t sE. M. Mechanics. Fizmatgiz, 1958 (in Russian).
t.3. La Salle J., Lefschetz S . Stability by Lyapunov's Direct Method with
Applications. Academic Press, New York and London. 1961.
14. La Vall6e-Poussin Ch. J . de, Legons de mecanique analytique, tome 1-2.
Paris, 1925.
15. Loitsyansky L. G., Lurie A. I. Theoretical Mechanics, Part 111. ONTI.
Moscow, 1934 (in Russian).
16. Lyapunov A. M. The General Problem of Stability of Motion. Gostekhizdat,
Moscow, 1950 (in Russian).
The stationary motion is always stable. 17. MacMillan W. D. Dynamics of Rigid Bodies. McGraw-Hill, New York
(b) 00= n. From the formulas (12) we find y= -6. Therefore and London, 4936.
18. bialkin I. G. Theory of Stability of Motion. Gostekhizdat, Moscow, 1952
(in H ussian) .
19. Merkin D. R. Gyroscopic Systems. Gostekhizdat, Moscow, 1956 (in Russian).
20. Hose N. V. Lectures in Analytical Mechanics, Part I . Leningrad Univer-
sity Press, .Leningrad, 1938 (in Russian).
21. Routh E . T. The Advanced Part of a Treatise on the Dynamics of a System
of Rigid Bodies, 6th ed. Macmillan, London, 1905.
22. Siege1 C. L. Vorlesungen iiber Himmelsmechanik. Springer, Berlin, 1956.
23. Sommerfeld A. Mechanik. 4te neubearb. Aufl. Geest und Portig, Leipzig,
2 1948.
The stationary motion will be stable if the condition L
4
>K is fulfil- 24. Suslov G. K . The Principles of Analytical Mechanics, 2nd ed. Kiev, 1911-
1912; 3rd ed. revised by N. N. Buchholte and V. K. Goltzman, entitled:
led, which in expanded notation comes out like this: Theoretical Mechanics. Gostekhizdat, Moscow, 1944 (in Russian).
>
C2r2 4AMgd 25. Synge J . L. Classical Dynamics. McGraw-Will, New York, 1958.
26. Variational Principles. Collection of articles, ed. b y L. S. Polak. Fizmatgiz.
If inequality (16) occurs, then, although in the case under consideration the Moscow, 1959 (in Russian).
centre of gravity is located over the geometric centre of the sphere, rotation 27. Whittaker E . 'I'A . Treatise on the Analytical Dynamics of Particles and
about the vertical axis will be stable stationary motion. Rigid Bodies. Dover Publications, New York. 1944.
 Name Index Subject Index

Aristotle 25 Malkin, I. G. 176

Maupertuis, P. 115 Accelerations, gericralized 41 Complete integral 136
Bernoulli, J . 25 Mikhailov, A . V. 200 Action, elementary 88 Complex spectrum 238
Hamilton 88, 115 Conditional stability 182 ff.
CarathBodory, C. 154 Lagrange 114, 113 Conjugate kinetic focus 95
Cartan, E. 8 , 119 Ostrogradsky, M. V. 89
Affix (of a complex number) 200 Conservation of energy, law of 77
Chetayev, N. G. 173, 176 Analytical mechanics 7, 8 Constrained systems 9
Courant, R. 219 PoincarQ, J . H. 8, 119, 193
Poisson, S. D. 8$ Angle, of nutation 35 Constraint(s)
bilateral 12 9
Dirichlet, P. G. Lejeune 169 Pyatigorsky, L. 86 of precession 35
of pure rotation 33 classification of 9 ff.
Fischer, E. 219 Rayleigh, J . W. 220 Appell's equations 22, 61 differential 9
Fermat, P. 115 Rose, H. V. 115 for nonl~oloi~omic systems 57 ff. differential stationary 11
Routh, E. 198, 243, 245, 254, 256 Associated expression (of a function) finite 9
Galileo, G. 25, 28 72. 73 finite, integrable 10
Stoker, J . J . 194 Asymptotic stability, of equilibrium finite, stationary I I
Hamilton, W. R. 73, 89 position 176 ff. ~~frozenl'14
Helmholtz, H. L. F. 105 Suslov, G. K. 7, 114
of linear system, criteria of 197 geometric 9
Hertz, H. 248 theorem on 178 ideal 17 ff.
Hurwitz, A. 198 Tait, P. 248
Thomson, W. 105, 248 Autonomous system (of differential kinematical 9
Huyghens, C. 28 equations) 51 nonholonomic 11
Hwa-Chung, Lee 122 Torricelli, E. 169
nonstationary 18
Jacobi, K. G. J . 84, 114 VallQe-Poussin, Ch. J. 7 semiholonomic 11
Bolzano-weierstrass lemma I79 taut 12
Lagrange, J . L. 22, 115, 169, 210 Webster, A. G. 248 unilateral 12
Landau, L. 86 Weierstrass, K. 210 Contact transformations 131
Lyapunov, A. M. 173, 176, 181, 193 Whittaker, E. T. 112 Canonical equations 73 Coordinates, cyclic 80 ff.
of Hamilton 71 ff. ignorable 80 ff.
Canonical transformation(s) 128 ff. kinosthenic 52
free 132 ff., 135 noncyclic 80, 243
nonfree 132, 133 normal 174, 210 ff.
nonpoint 135 position 243
structure of an arbitrary 152 ff. principal 174, 212
univalent 131 Coriolis force 52
valence of 131 Coupled pendulums 210
Canonicity of transformation 163, 164 Curl (of velocity) 107
Characteristic equation 190 Cyclic coordinates 80 ff.
Characteristic numbers I90
Characteristic, amplitude-phase 236,
237 D'Alembert's principle 30, 31
frequency 236, 239 Degree of stability 192
Chetayev's instability test 186 Degrees of freedom 15
Chetayev's theorem 176 Deviations 166
Circulation of velocity 107 Diriclilet's theorem 169
Compatibility with constraints 25 Displaced lines 110
r
2 62 Subject Index
Displacements, possible 13, 14
possible infinitesimal 13
virtual 13, 14
Dissipation of energy, total 53, 54
Dissipative forces 51 f f .
Dissipative system 52
definite 54
Donkin's theorem 74, 75
Dynamical equations of Euler 64, 186
Effective forces 16
Eigenvector 191
Elementary action 88
Energy integral 51
Equation, characteristic 190
frequency 205
secular 190, 205, 230
Equilibrium position 25, 166
asymptotically stable 176
stable 166, 167
unstable 167
Eulerian angles 33
Euler's differential equations 91
Euler's dynamical equations 64, 186
Euler's equations 64, 65
Euler's formula 49
EuIer's theorem on homogeneous fun-
ctions 76
Fermat's principle 215, 116
Filter 238
Fluid line 107
Fluid surface 107
Free body 19
Free systems 9
Frequency equation 205
General equation of dynamics 20 ff.,
-
2n
Generalized accelerations 41
Generalized conscrvative system(s)
56, 111 ff.
Generalized coordinates, independent
34
Generalized forces 60
Generalized integral of cnergy 76
Generalized momenta 71
Generalized potential 67, 68
Generalized total energy 76
Generalized velocities 41, 68
Generating function 131
Geodesic lines 117, 118
Golden rule of nicchanics 25, 26
Group 128
Gyroscopic forces 51
Gyroscopic moment 53
Gyroscopically uncoupled system
246 ff.
Hamilton action 115
Hamiltoiiian function 73, 77, 8 0
Hamiltonian variables 71, 72, 75, 77
Hamilton-Jacobi euuation 135 ff.. 143
Hamilton-ostrogradsky principle 89
Hamilton's canonical equations 102,
104, 106
Hamilton's equations 73
Hamilton's principal function 137-
139
Hamilton's principle 88 ff.
second form of 96 ff.
Helmholtz theorem 110
Hessian 70, 71, 74, 75
small 78
Hidden motions 248
Holonomic rhconomic system 12
Holonomic scleronomic system II
Holonomic system 11, 33 ff.
Hurwitz determinant 198, 199
Hurwitz polynomial 201
Hydrodynamical interpretation of inte-
gral invariant of mechanics 105 ff.
Ilypothesis of realization of admissib-
l e reaction forces 2 1
Ignorabl c coordinates 8 0 ff.
Influelice coefficient 224
Integral of equations of motion 83
Integral invariant($ 8, 101
absolute 122
in extended coordinate space 107
of mechanics, hasic 98 ff.
relative 121
universal 122
universal relative 122
Invariance of Poincarb integral, geo-
metrical interpretation of 120
Invariance of volume in phase
space l 2 5 ff.
Jacobi cquations 114
Jacobi theorem 136
Jacobian 71, 74
Jacohian functional matrix 47
Jacohian matrix, simplicial nature
of 161
Subject Index
of order n 161
Jacobi-Poisson theorem 85
Keplerian motion 146, 147
Kepler's three laws of planetary
motion 148
Kinetic potential 66
Kinosthenic coordinates 82
Kronecker's symbol 160, 189
Lagrange action 114, 115
Lagrange brackets 160
Lagrange's equations, investigations
of 45 ff.
Lagrange's equations for potential
forces 66 f f .
Lagrange's equations in independent
coordinates 41
Lagrange's equations of the first
kind 22
Lagrange's equations of the second
kind 22, 63
in independent coordinates 40 f f .
Lagrange's theorem 169, 171
Lagrange's theorem on stability of
equilibrium position 166 ff.
Lagrangian function 66, 73, 245
Lagrangian variahles 71, 72, 75, 78
Legendre transformation 74
LiBnard-Chipart conditions 199
Lihard-Chipart criterion 201
Linear approximation for nonlinear
system 193
Liouville's theorem 127, 150
Lorentz force 68
L y a p u n ~ vfunction 182, 254
Lyapunov's first theorem 174
Lyapunov's second theo~enl 176
Lyapunov's theorem 184, 185
corollary to 184, 185
Matrix. oscillatory 225
generalized-simplicial 163
positive definite 205
simplicial 162
Maupertuis-Lagrange principle ol
least action 114 ff.
Rlaximinimal properties 2 19
Maxwell's rinciple of reciprocity 224
Mersenne Paw 227
Mikhailov's criterion 200
Minimaximal properties 219
2 63
Natural systems 70
Newton's law of gravitation 148
Nodal line 3 5
Noncyclic coordinates 80, 243
Nonholonomic system 11, 12
Nonnatural system 70
Normal coordinates 174. 210 ff.
Optical principle of Fermat 115, 116
Oscillations, forced 216, 236
free 216
Oscillatory matrix 225
Paths, circuitous 89
straight-line 89
Perturbation theory, application of
canonical transformations to
150 ff.
~ o i n c a k -integral 121
Poincarb-Cartan integral 101, 104,
106, 121, 129
Poincar6-Cartan integral invariant
ff.
- - --.
98
Point transformation 135
Poisson bracket(s) 83 ff., 160
invariance of, i n canonical trans-
formation 163 ff.
Poisson identity 84
Position coordinates 243
Potential energy 48
Potential forces 48
Potential of forces 48
Power of nonpotential forces 50
Principal coordinates 174, 212
Principal oscillations 210
Principle of least action of Mauper-
tuis-Lagrange 114, 115
Principle of possible displaccments
25. 29
~ r L l & p l eof reciprocity, Maxwell's 224
Principle of virtual displacements 25,
26, 30, 39
Pseudoaccelerations 60, 62
Pscudocoordinates 59
I'scudovelocities 59
I l a y l c i g l ~ ' ~dissiputivc function 53,
23 1
Ilaylcig11's iunctions 233
Ilaylaigh's theorem 220
Hc;iction forces of constraints 16
13rducctl system 246
ll csoriarlce 216
 264 Subject I n d e x
Rheonomic system 11
Rotor (of velocity) 107
Routh equations 78 ff., 80, 81
Routh functions 80, 81, 244
Routh potential 245, 246, 252, 257
Routh variables 78, 242, 243
.Routh-Hurwitz condition 199
Scleronomic system 11, 35
Secular equation 190, 205, 230
Secular terms 192, 209
Separation of variables, method of
141 ff.
Separation theorem 221
Shortest lines in space 118
Simplicial matrix 162
Simplicial group 162
Solution. asvmototicallv stable 185
stable '185 "a
Spectrum, complex 238 III
Stability, geometric criterion of 200
, of linear systems 189 ff.
of stationarv motions 252 ff.
Stationary motion, conditions for
existence of 255
criterion of stability of 254, 256
stability of 252 ff.
stable 254, 255
Statistical ensemble 127
Printed in the U n i o n of Soviet Socialist Republics
Theorem of Lee Hwa-Chung 122, 130
Theorem on asymptotic stability 178
Theorem on conservation of cir-
culation of velocity 108
Theorem on conservation of vortex
flow 108
Theorem on stability of a dissipative
system 181
Thomson's theorem 107
Torricelli's principle 27
Total energy 170
Transformation (see canonical point)
testing canonical character of ,
158 ff.
Undetermined multipliers of La-
grange 21, 22
Universal internal invariant 121, 130
of ~ o i n c a r 6 ~ 1 1ff.
9
Valence of canonical transformation
131, 161
Velocities, generalized 41, 68
possible 13 I
Vortex lines 109, 110
Whittaker's equations 112