Piskunov - Differential and Integral Calculus - Volume 2 - Mir

N.
PISKUNOV
differential
and
intégral
calculus
VO i n
MIR PUBLISHERS • M OSCOW

ABOUT THE BOOK
Textbook by the late Prof. Nikolai

Piskunov, D. Sc. (Phys, and Math.),
is devoted to the most important
divisions of higher mathematics.
This édition, revised and enlarged,
is published in two volumes, the
second volume dealing with the
following topics: Differential
Equations, Multiple Intégrais, Line
and Surface Intégrais, Sériés, Fou-
rier Sériés, The Equations of Mathe-
matical Physics, Operational Cal-
culus and Certain of Its Applica
tions, Eléments of the Theory of
Probability and Mathematical Sta-
tistics, Matrices.
There are numerous examples and
problems in each section of the
course; many of them demonstrate
the ties between mathematics and
other sciences, making the book
a useful aid for self-study.
This is a textbook for higher tech-
nical schools that has gone through
several éditions in Russian and
alëo been translated into French
and Spanish.
H. C. I l H C K y H O B
Æ H O G E P E H U H A J Ib H O E H H H T E rP A / Ib H O E
H CM HCJlEHM fl
TOM II
H3ZIATEJlbCTBO «HAYK A» MOCKBA

N. P IS K U N O V
DIFFERENTIAL
AND
INTEGRAL
CALCULUS
VOL. I I
Translatée! from the Russian

by
George Yankovsky
M I R P U B L IS H E R S MOSCOW
First published 1964
Second printing 1966
Third printing 1969
Second édition 1974
Fourth printing 1981
H a OH3AUÜCKOM R3blKê
© Engli^h translation, Mir Publishers, 1974
P r in te d in ih e U n io n o f S o v ie t S o c ia lis t R e p u b lic s
CONTENTS
CHAPTER 1 D1FFERENTIAL EQUATIONS
1.1 Statement of the problem. The équation of motion of a body with

résistance of the medium proportional to the velocity. The équation
of a c a te n a ry ............................................................................................. 11
1.2 D éfinitions................................................................................................. 14
1.3 First-order differential équations (general notions) .......................... 15
1.4 Equations with separated and separable variables. The problem of
the disintegration of radium ............................................................... 20
1.5 Homogeneous first-order é q u a tio n s ........................................................ 24
1.6 Equations reducible to homogeneous é q u a tio n s.................................. 26
1.7 First-order liuear é q u a tio n s ................................................................... 29
1.8 Bernoulli’s é q u a t i o n .............................................................................. 32
1.9 Exact differential éq u atio n s................................................................... 34
1.10 Integrating fa c to r...................................................................................... 37
1.11 The envelope of a familv of c u r v e s .................................................... 39
1.12 Singular solutions of a first-order differential é q u a t i o n ................... 45
1.13 Clairaut’s é q u a tio n ................................................................................... 46
1.14 Lagrange’s équation ............................................................................... 48
1.L5 Orthogonal, and isogonal trajecto ries.................................................... 50
1.16 Higher-order differential équations (fundamentals) .......................... 55
1.17 An équation of the form yW = f ( x ) ..................................................... 56
1.18 Some types of second-order differential équations reducible to first-
order équations. Escape-velocity problem . .......................................... 59
1.19 Graphical method of integrating second-order differential équations 66
1.20 Homogeneous linear équations. Définitions andgeneral properties 68
1.21 Second-order homogeneous linear équations with constant coefficients 75
1.22 Homogeneous linear équations of the nth order with constant coeffi
cients ........................................................................................................ 80
1.23 Nonhomogeneous second-order linear é q u a tio n s .................................. 82
1.24 Nonhomogeneous second-order linear équations with constant coeffi
cients ........................................................................................................ 86
1.25 Higher-order nonhomogeneous linear é q u a tio n s.................................. 93
1.26 The differential équation of mechanical v ib r a tio n s .......................... 97
1.27 Free o sc illa tio n s ...................................................................................... 98
1.28 Forced o s c illa tio n s.................................................................................. 102
1.29 Systems of ordinary differential é q u a tio n s ......................................... 106
1.30 Systems of linear differential équations with constant coefficients 111
1.31 On Lyapunov’s theory of s ta b ility ........................................................ 117
1.32 Euler's method of approximate solution of first-order differential
équations ................................................................................................. 133
1.33 A différence method for approximate solution of differential équa
tions based on Taylor's formula. Adams m e th o d .............................. 135
1.34 An approximate method for integrating Systems of first-order diffe
rential é qua tions...................................................................................... 142
Exercises on Chapter 1 146
6 Contents
CHAPTER 2 MULTIPLE INTEGRALS
2.1 Double in té g r a is ...................................................................... 158

2.2 Calculating double in té g ra is................................................................... 161
2.3 Calculating double intégrais (c o n tin u e d )............................................. 166
2.4 Calculating areas and volumes by means of doubleintégrais . . . 172
2.5 The double intégral in polar coordinates............................................. 175
2.6 Change of variables in a double intégral(general c a s e )..................... 182
2.7 Computing the area of a surface ........................................................ 187
2.8 The density distribution of matter and the doubleintégral . . . . 190
2.9 The moment of inertia of the area of a plane f i g u r e ...................... 191
2.10 The coordinates of the centre of gravity of the area of a plane
f i g u r e ........................................................................................................ 196
2.11 Triple intégrais ...................................................................................... 197
2.12 Evaluating a triple intégral ............................................................... 198
2.13 Change of variables in a triple i n t é g r a l ............................................. 204
2.14 The moment of inertia and the coordinates of the centre of gravity
of a solid . . .......................................................................................... 207
2.15 Computing intégrais dépendent on a p aram eter.................................. 209
CHAPTER 3 LINE INTEGRALS AND SURFACE INTEGRALS
3.1
Line intégrais ......................................................................................... 216
3.2
Evaluating a line in té g ra l....................................................................... 219
3.3
Green’s formula ...................................................................................... 225
3.4
Conditions for a line intégral to be independent of the path of inté
gration .................................................................................................... 227
3.5 Surface in té g r a is ...................................................................................... 232
3.6 Evaluating surface in té g ra is................................................................... 234
3.7 Stokes* formula ...................................................................................... 236
3.8 Ostrogradsky’s f o r m u la ........................................................................... 241
3.9 The Hamiltonian operatorand some ap p lic a tio n s................................ 244
CHAPTER 4 SERIES
4.1 Sériés. Sum of a s é r i é s .......................................................................... 253
4.2 Necessary condition for convergence of a sériés . . . . ! ! ! ! ! ! 256
4.3 Comparing sériés with positive terms ............................. ! ! ! ! ! ! 258
4.4 D’Alembert’s t e s t ...................................................................................... 260
4.5 Cauchy’s t e s t ....................................................................... ! ! . ! ! ! ! 264
4.6 The intégral test for convergence of a s é r i é s ......................................[ 266
4.7 Alternating sériés. Leibniz* th e o re m .....................................................’ 269
4.8 Plus-and-minus sériés. Absolute and conditional convergence . . . . 271
4.9 Functional s é r i é s .............................................................................. 274
4.10 Dcminated s é r i é s ........................................................ ! ! ! ! ! ! ! ! ! 275
4.11 The continuity of the sum of as é r i é s .................................... 277
4.12 Intégration and différentiation of sériés .......................... * 280
4.13 Power sériés. Interval of convergence ............................................ ’ . 283
4.14 Différentiation of power sé rié s ........................................................... 288
4.15 Seriez in powers o f \ —a ................................. . ! . ! . ! . . ! ! ! 289
4.16 Taylor*s sériés and JVu>çlaurin*ssériés ....................................... 290
4.17 Sériés expansion of f u n b tio n s ................................................ 292
4.18 Euler’s formula .................................................... ! . 1 ! ! 294
4.19 The binomial sériés . 295
Contents 7
4.20 Expansion of the function ln (l-fx ) in a power sériés. Computing

lo g a rith m s ................................................................................................. 297
4.21 Sériés évaluation of definite intégrais . .# ......................................... 299
4.22 Integrating differential équations by means of sériés ................... 301
4.23 BesseFs é q u a tio n ...................................................................................... 303
4.24 Sériés with complex te rm s .................. .................................................... 308
4.25 Power sériés in a complex v a r i a b l e .................................................... '309
4.26 The solution of first-order differential équations by the method of
successive approximations (methodof ité ra tio n ).................................. 312
4.27 Proof of the, existence of a solution of a differential equatfon. Error
estimation in approximate s o lu tio n s ...................................................... 313
4.28 The uniqueness theorem of the solution of a differential équation 318
CHAPTER 5 FOU R 1ER SERIES

5.1 Définition. Statement of the p ro b le m ........................................... 327
5.2 Expansions of functions in Fourier sériés ......................................... 331
5.3 A remark on the expansion of a periodic function in aFouriersériés 336
5.4 Fourier sériés for even and odd f u n c tio n s .................................... 338
5.5 The Fourier sériés for a function with period 2 1 ........................ 339
5.6 On the expansion of a nonperiodic function in aFourier sériés . . 341
5.7 Mean approximation of a given function by a trigonométrie poly
nomial ......................................................................................................... 343
5.8 The Dirichlet intégral ........................................................................... 348
5.9 The convergence of a Fourier sériés at a given p o i n t ....................... 351
5.10 Certain sufficient conditions for the convergence of aFouriersériés 352
5.11 Practical harmonie a n a ly s is ............................................................. 355
5.12 The Fourier sériés in complex f o r m .............................................. 356
5.13 Fourier intégral ...................................................................................... 358
5.14 The Fourier intégral in complex f o r m .......................................... 362
5.15 Fourier sériés expansion with respect to an orthogonal System of
fun ctio n s..................................................................................................... 364
5.16 The concept of a linear function space. Expansion of functions in
Fourier sériéscompared with décomposition of vectors . . . . . . 367
CHAPTER 6 EQUATIONS OF MATHEMATICAL PHYSICS

6.1 Basic types of équations of mathematical p h y sics.............................. 374
6.2 Deriving the équation of the vibrating string. Formulating the bo-
undary-value problem. Deriving équations of electric oscillations in
w ires....................................................................... i ................................. 375
6.3 Solution of the équation of the vibrating string by the method of
séparation of variables (the Fourier m e th o d )..................................... 378
6.4 The équation of heat conduction in a rod. Formulation of the bo-
undary-value problem .............................................................................. 382
6.5 Heat transfer in space .......................................................................... 384
6.6 Solution of the first boundary-value problem for the heat-conduction
équation by the method of nnite d ifféren ces................................. . 387
6.7 Heat transfer in an unbounded r o d .................................................... 389
6.8 Problems that reduce to investigating solutions of the Laplace équa
tion. Stating boundary-value problem s................................................. 394
6.9 The Laplace équation in cylindrical coordinates. Solution of the Di
richlet problem for an annulus with constant values of the desired
function on the inner and outer circumferences.................................. 399
8 Contents
6.10 The solution of Dirichlet’s problem for a circle 401
.
6.11 Solution of the Dirichlet problem by the method of finite différences 405
Exercises on Chapter 6 ...................................................................................... 407
CHAPTER 7 OPERATIONAL CALCULAIS AND CERTAIN

OF ITS APPLICATIONS
7.1 The original function and its tr a n s f o r m .................. 411
7.2 Transforms of the functions a0 (f). s in /, cos / .................................... 413
7.3 The transform of a function with changed scale of the independent
variable. Transforms of the functions sin af, cos a t .......................... 414
7.4 The linearity property of a tr a n s f o r m ................................................. 415
7.5 The shift theorem .................................................................................. 416
7.6 Transforms of the functions e~a t, sinh at, cosh a t, e " a / sin a tt
e~ai cos a t ................................................................................................. 416
7.7 Différentiation of transform s................................................................... 417
7.8 The transforms of d é riv a tiv e s ............................................................... 419
7.9 Table of transform s.................................................................................. 420
7.10 An auxiliary équation for a given differential é q u a tio n ................... 422
7.11 Décomposition th e o r e m ........................................................................... 426
7.12 Examples of solutions of differential équations and Systems of diffe
rential équations by the operational m e t h o d ..................................... 428
7.13 The convoiution th e o r e m ....................................................................... 429
7.14 The differential équations of mechanical vibrations. The differential
équations of electric-circuit th e o ry ........................................................ 432
7.15 Solution of the differential équation ofo sc illa tio n s ............................ 433
7.16 Investigating free o sc illa tio n s............................................................... 435
7.17 Investigating mechanical and electrical oscillations in the case of a
periodic external force ........................................................................... 435
7.18 Solving the oscillation équation in thecase of ré so n a n c e ................. 437
7.19 The delay th e o re m .................................................................................. 439
7.20 The delta function and its transform .................................................... 440
Exercises on Chapter 7 ...................... 443
CHAPTER 8 ELEMENTS OF THE THEORY OF PROBABILITY
AND MATHEMATICAL STATISTICS
8.1 Random event. Relative frequency of a random event. The proba-
bility of an event. The subject of probability th e o ry ....................... 445
8.2 The classical définition of probability and the calculation of proba
bilités ......................................................................................................... 447
8.3 The addition of probabilités. Complementary random events . . . 449
8.4 Multiplication of probabilités of independent e v e n t s ....................... 452
8.5 Dépendent events. Conditional probability. Total probability . . . 454
8.6 Probability of causes. Bayes's fo rm u la ................................................. 457
8.7 A discrète random variable. The distribution law of a discrète ran
dom variable .............................................................................................. 460
8.8 Relative frequency and the probability of relative frequency in re-
peated t r i a l s .............................................................................................. 462
8.9 The mathematical expectation of a discrète random variable . . . 466
8.10 Variance. Root-mean-square (standard) déviation. Moments . . . . 471
8.11 Functions of random variables ............................................................ 474
8.12 Continuous random variable. Probability density function of a con-
tinuous random variable. The probability of the random variable
falling in a specified in t e r v a l ................................................................ 475
8.13 The distribution function. Law of uniform d istrib u tio n ................... 479
8.14 Numerical characteristics of a continuous random variable . . . . 482
Contents 9
8.15 Normal distribution. The expectation of a normal distribution . . . 485

8.16 Variance and standard déviation of a normally distributed random
v a r i a b l e .......................................................« .......................................... 487
817 The probability of a value of the random variable falling in a given
interval. The Laplace function. Normal distribution function . . . 488
8.18 Probable e r r o r .......................................................................................... 493
8.19 Expressing the normal distribution in terms of the probable error.
The reduced Laplace fu n c tio n ................................................................ 494
8.20 The three-sigma rule. Error d is trib u tio n ............................................. 496
8.21 Mean arithmetic e r r o r ............................................................................... 497
8.22 Modulus of précision. Relationships between the characteristics of
the distribution of e r r o r s ....................................................................... 498
8.23 Two-dimensional random v a ria b le s............................. 499
8.24 Normal distribution in the plane ......................................................... 502
8.25 The probability of a two-dimensional random variable falling in a
rectangle with si des parai lel to the principal axes of dispersion
under the normal distribution l a w ........................................................ 504
8.26 The probability of a two-dimensional random variable falling in the
ellipse of dispersion................................................................................... 506
8.27 Problems of mathematical statistics. Statistical d a t a ....................... 507
8.28 Statistical sériés. H istogram .................................................................... 508
8.29 Determining a suitable value of a measured q u a n tity ....................... 511
8.30 Determining the parameters of a distribution law. Lyapunov’s theo-
rem. Laplace's th e o r e m ........................................................................... 512
Exercises on Chapter 8 ....................................................................................... 516
CHAPTER 9 MATRICES
9.1 Linear transformations.Matrix n o t a t i o n ............................................... 519
9.2 General définitions involving m a t r i c e s ................................................. 522
9.3 Inverse transformation................................................ 524
9.4 Operations on matrices. Addition of m atrices...................................... 526
9.5 Transforming a vector into another vector by means of a matrix 529
9.6 Inverse m a t r i x .............................................................................. 531
9.7 Malrix in v e r s io n ...................................................................................... 532
9.8 Matrix notation for Systems oflinear équations and solutions of
Systems of linear e a u a tio n s .................................................................... 534
9.9 Solving Systems of linear équations by thematrix m ethod................. 535
9.10 Orthogonal mappings. Orthogonal m a tr ic e s ......................................... 537
9.11 The eigenvector of a linear tran sfo rm atio n......................................... 540
9.12 The matrix of a linear transformation under which the base vectors
are eigenvectors .............................................. * .................................... 543
9.13 Transforming the matrix of a linear transformation when changing
the basis ...................................................................................................... 544
9.14 Quadratic forms and their transformation .......................................... 547
9.15 The rank of a matrix. The existence of solutions of a System of
linear équations ...................................................................................... 549
9.16 Différentiation and intégration of m a tr ic e s ......................................... 550
9.17 Matrix notation for Systems of differential équations and solutions
of Systems of differential équations with consta/it coefficients . . . 552
9.18 Matrix notation for a linear équation oforder n ................................ 557
9.19 Solving a System of linear differential équations with variable co
efficients by the method of successive approximations using matrix
n o t a t i o n ..................................................................................................... 558
Exercises on Chapter 9 ...................................................................................... 563
a p p e n d i x ......................................................................................................... 565
i n d e x ................................................................................................................. 567
CHAPTER 1
D1FFERENTIAL EQUATIONS
1.1 STATEMENT OF THE PROBLEM. THE EQUATION

OF MOTION OF A BODY WITH RESISTANCE
OF THE MEDIUM PROPORTIONAL TO THE VELOCITY.
THE EQUATION OF A CATENARY
Let a function y = f(x) reflect the quantitative aspect of some
phenomenon. Frequently, it is not possible to establish directly
the type of dependence of y on x t but it is possible to give the
relation between the quantities x and y and the dérivatives of y
with respect to x: y \ y \ . . . , yin). That is, we are able to
Write a differential équation.
From the relationship established between the variables jt, y
and the dérivatives it is required to détermine the direct depen
dence of y on x\ that is, to find y = f(x) or, as we say, to inte-
grate the differential équation.
Let us consider two examples.
Example 1. A body of mass m is dropped from some height. It is required
to establish the law according to which the velocity v will vary as the body
falls, if, in addition to the force of gravity, the body is acted upon by the
decelerating force of the air, which is proportional to the velocity (with con
stant of proportionality k)\ in other words, it is required to find v = f( t) .
Solution. By Newton’s second law
where is the accélération of a moving body (the dérivative of the velocity

with respect to time) and F is the force acting on the body in the direction
of motion. This force is the résultant of two forces: the force of gravity mg
and the force of air résistance,— kv, which has the minus sign because it is in
the opposite direction to that of the velocity. And so we hâve
m ^ = m g — kv (1)
This relation connects the unknown function v and its dérivative we hâve
at
a differential équation in the unknown function v. This is the équation of mo
tion of certain types of parachutes. To solve a differential équation means to
find a function v = f( t) such that identically satisfies the given differential
équation. There is an infinitude of such functions. The student can easily verify
that any function of the form
( 2)
12 Ch. 1 Differential Equations
satisfies équation (1) no matter what the constant C is. Which one of ttiese
functions yields the sought-for dependence of v o n /?T o find it we take advantage
of a supplementary condition: when the body was dropped it was imparted an
initial velocity v0 (which may be zéro as a particular case); we assume this
initial velocity to be known. But then the unknown function v = f(t) must be
such that when t = 0 (when motion begins) the condition v = v 0 is fulfilled.
Substituting t = 0, v = u0 into formula (2), we find
whence
-o ,
C -t» —
Thus, the constant C fs found, and the sought-for dependence of v on / is
From this formula it follows that for sufficiently large t the velocity v dé
pends but slightly on u0.
It will be noted that if k = 0 (the air résistance is absent or so small that
we can disregard it), then we hâve a resuit familiar from physics: *
v = v0+ g t (2")
This function satisfies the differential équation (1) and the initial condition:
v = v0 when t = 0.
Example 2. A flexible homogeneous thread is suspended at two ends. Find
the équation of the curve that it describes under its own weight (it is the same
as for any suspended ropes, wires, chains).
Solution. Let Af0 (0, b) be the lowest point
of the thread, and M an arbitrary point
(Fig. 1). Let us consider a part of the thread
M 0M. This part is in equilibrium under the
action of the résultant of tnree forces:
(1) the tension 7\ acting along the tangent
at the point M and forming an angle <p with
the x-axis;
(2) the tension H at M0 acting horizontally;
(3) the weight of the thread ys acting ver
tical 1y downwards, where s is the length of the
arc M0M and y is the linear spécifie weight of
the thread.
Breaking up the tension T into horizontal
and vertical components, we get the équations
of equilibrium:
Tcosqp = //, Tsinq) = YS
Dividing the ternis of the second équation by the cojresponding terms of
the first, we obtain
tan çp= ~t- s (3)
* Formula (2") can be obtained from (2;) by passing to the limit:
A". [('-TK-+Ï]=-+"
l.î Statement of the Problem 13
Now suppose that the équation of the sought-for curve can be written in
the form y — f(x). Here, f (x) is an unknown function that has to be found.
It will be noted that #
tanq> = / ' (■*) = •£"
Hence,
dx a (4)
where the ratio — is denoted by a.
y
Differentiate both sides of (4) with respect to x:
d2y __1_ ds
dx2 a dx (5)
But, as we know (see Sec. 6.1, Vol. I),
£=V^+(af)
Substituting this expression into équation (5), we get the differential équa
tion of the sought-for curve:
S -7 «
It expresses the relationship between the first and second dérivatives of the
unknown function y.
Without going into the methods of solving the équations, we note that any
function of the form
j, = flc o s h ( i - + C 1) + C , (7)
satisfies équation (6) for any values that Cx and C%may assume. This is évident
if we put the first and second dérivatives of the given function into (6). We
shall la ter on indicate (Sec. 1.18), without proof, that these functions (for
different Ci an<j C2) exhaust ail possible solutions of eauation (6).
The graphs of ail the functions thus obtained are called catenaries.
Let us now find out how one should choose the constants Cx and C2 so as
to obtain precisely that catenary whose lowest point M has coordinates (0, b),
Since for x = 0 the point of the catenary occupies the lowest possible position,
the tangent here is horizontal, ~ = 0 . Also, it is given that at this point the
ordinate is equal to b y = b.
From (7) we find
*',=sinh (■ j+ c *)
Putting x = 0 here, we obtain 0 = sinhC!. Hence, Ci = 0. If the ordinate of
the point Af0 is bt then y — b when * = 0 .
From équation (7) we get b = (1 + 1) + Ca, assuming x = 0 and Ci = 0,
whence C2= b — a. Finally we hâve
y = acosh
Equation (7) assumes a very simple form if we take the ordinate of M0 equal
to a. Then the équation of the catenary is
y = a cosh (x/a)
1.2 DEFINITIONS
Définition 1. A differential équation is one which connects the
independent variable x , unknown function y = f(x)t and its deri-
vatives y' , t f , . . . . yM .
Symbolically, a differential équation may be written as follows;
F{x, y, y ', y \ . . . . «/<">) = 0
or
f* y *ny \ _ a
’ dx* ’ • • • ’ dxn J
If the sought-for function y = f{x) is a function of one indepen
dent variable, then the differential équation is called ordinary.
In this chapter we shall deal only with ordinary differential
équations. *
Définition 2. The order of a differential équation is the order
of the highest dérivative which appears.
For example, the équation
y '— 2*ÿa + 5 = 0
is an équation of the first order.
The équation
y"-\-ky'—by—sinx = 0
is an équation of the second order, etc.
The équation considered in the preceding section in Example 1
is an équation of the first order, in Example 2, one of the second
order.
Définition 3. The solution or intégral of a differential équation
is any function y = f(x), which, when put into the équation,
converts it into an identity.
* ln addition to ordinary differential équations, mathematical analysis also
deals with partial differential équations. Such an équation is a relation between
an unknown function z that is dépendent upon twoorseveral variables*, y ........
these variables x, y, . . . . and the partial dérivatives of z: ^ , etc.
The following is an example of a partial differential équation in the unknown
function z(x, y):
dz dz
X Fx= y Ty
It is easy to verify that this équation is satisfied by the function z = x 2y2
(and also by a multitude of other functions).
In this course partial differential équations are discussed in Chapter 6.
1.3 First’Order Differentiai Equations 15
Example 1. Suppose we hâve the équation
The functions y = sin*, y = 2 cos x, t/ = 3 sin x — cos x and, in general,

functions of the form ^ C x s i n x , y = C2 cos x or
y = Ct sin jc + C2 cos x
are solutions of the given équation for any choice of constants Cx and Ca; this
is évident if we put these functions into the équation.
Example 2. Let us consider the équation
y'x — x2— y = 0
Its solutions are ail functions of the form
y = x* + Cx
where C is any constant. Indeed, differentiating the functions y = x®+ Cx, we
find
y'=*2x + C
Putting the expressions for y and y ' into the initial équation, we get the
identity
(2x + C ) x - x 2- x 2- C x = 0
Each of the équations considered in Examples 1 and 2 has an infinitude of
solutions.
1.3 F1RST-ORDER DIFFERENTIAL EQUATIONS

(GENERAL NOTIONS)
1. A differential équation of the first order is of the form
F(x, y, y') = 0 (1)
If this équation can be solved for y', it can be written in the form
y’ = f(x.y) (1')
In this case we say that the differential équation is solved for
the dérivative. For such an équation the following theorem, called
the theorem of existence and uniqueness of solution of a differen
tial équation, holds.
Theorem. If in the équation
y'= f(x, y)
the function f(x, y) and its partial dérivative with respect to y,
■gj, are continuons in some domain D, in an xy-plane, containing
sortie point (x0, y0), then there is a unique solution to this équation,
t/ = <p (x)
which satisfi.es the condition y = y0 at x —x0.
This theorem will be proved in Sec. 4.27.
The géométrie meaning of the theorem consists in the tact that

there exists one and only one such function y = y(x), the graph
of which passes through the point (x0, t/„).
It follows from this theorem that équation (T) has an infinitude
of solutions [for example, a solution the graph of which passes
through (x0, y0); another solution whose graph passes through
(*„, j/i); through (x0, yt), etc., provided these points lie in the
domain £>].
The condition that for x = x0 the function y must be equal to
the given number yQis called the initial condition. It is frequently
written in the form
y |*=*, = I/o
Définition 1. The general solution of a first-order differential
équation is a function
y = (f(x, C) (2)
which dépends on a single arbitrary constant C and satisfies the
following conditions:
(a) it satisfies the differential équation for any spécifie value of
the constant C;
(b) no matter what the initial condition y —y0 for x = x0, that
is, (y)x=x. — ÿo> it is possible to find a value C = C„ such that the
function y — ip(x, C0) satisfies the given initial condition. It is
assumed here that the values x0 and y0 belong to the range of the
variables x and y in which the conditions of the existence and
uniqueness theorem are fulfilled.
2. In searching for the general solution of a differential équation
we ofien arrive at a relation like
<D(v, y, C) = 0 (2')
which is not solved for y. Solving this relationship for y, we get
the general solution. However, it is not always possible to express
y from (2') in terms of elementary functions; in such cases, the
general solution is left in implicit form. An équation of the form
0(;c, y , C) = 0 which gives an implicit general solution is called
the complété intégral of the differential équation.
Définition 2. A particular solution is any function y = <p(x, C0)
which is obtained from the general solution i/ = <p(x, C), if in the
latter we assign to the arbitrary constant C a definite value C = C„.
In this case, the relation <P(x, y, C0) = 0 is called a particular
intégral of the équation.
Example 1. For the first-order équation
dy y
1.3 First-Order Differential Equations 17
the general solution is a family of functions y = — ; this can be checked by

simple substitution in the équation.
Let us find a particular solution that will# satisfy the following initial
condition: y0= l when x0= 2.
C C
Putting these values into the formula y = — , we hâve 1=~2 or ^ = 2.
2
Consequently, the function y = — will be the particular solution we are
seeking.
From the géométrie viewpoint, the general solution (complété

intégral) is a family of curves in a coordinate plane, which family
dépends on a single arbitrary constant C (or, as it is common to
say, on a single parameter C). These curves are called intégral
curves of the given differential équation. A particular intégral is
associated with one curve of the family that passes through a
certain given point of the plane.
Thus, in the latter example, the complété intégral is geometri-
cally depicted by a family of hyperbolas # = y ,w h ile the partic
ular intégral defined by the given initial condition is depicted
by one of these hyperbolas passing through the point Af0(2, 1).
Fig. 2 shows the curves of a family that are associated with
certain values of the parameter: C = ÿ , C = l, C = 2, C = — l,etc.
To make the reasoning still more pictorial, we shall from now
on say that not only the function y = <p(x, C0) that satisfies the
::o82
équation but also the associated intégral curve is a solution of

the équation. We will therefore speak of a solution passing through
the point (*„, */,).
Note. The équation —-j has no solution passipg through a
point lying on the </-axis (see Fig. 2). This is because the right
side of the équation is not defined for jc= 0 and consequently is
not continuous.
To solve (or as we frequently say, to integrate) a differential
équation means:
(a) to find its general solution or complété intégral (if the initial
conditions are not specified) or
(b) to find a pàrticular solution of the équation that will satisfy
the given initial conditions (if such exist).
3. Let us now give a géométrie interprétation of a first-order
differential équation.
Let there be a differential équation solved for the dérivative:
£ = f( x , y)
d (H
and let t/ = <p(x, C) be the general solution of this équation. This

general solution détermines the family of intégral curves in the
xy-plane.
For each point M with coordinates x and y, équation (T)
defines the value of the dérivative , or the slope of the tangent
line to the intégral curve passing through this point. Thus, the
differential équation (T) yields a collection of directions or, as
we say, defines a direct ion-field in the xy-plane.
Consequently, from the géométrie point of view, the problem
of integrating a differential équation consists in finding curves,
the directions of the tangents to which coïncide with the direction
of the field at the corresponding points.
For the differential équation (1) the locus of points at which
the relation ^r-
dx
= C = const holds true is called the isocline of the
given differential équation.
Different isoclines resuit from different values of C. The équation
of the isocline corresponding to the value C will clearly be f (x, y)—C.
If we hâve constructed a family of isoclines, it is possible to make
a rough construction of the family of intégral curves. We say
that a knowledge of the isoclines enables one to détermine quali-
tatively the locations of the intégral curves in a plane.
Fig. 3 depicts a direction-field defined by the differential équation
dy _ y
1.3 First-Order Differential Equations 19
The isoclines of the given differential équation are

— 7 = C, or y = f — Cx
This is a family of straight lines. They are constrycted in Fig. 3.

4. Let us now consider the following problem.
Let there be given a family of functions that dépends on a

single parameter C:
t/ = <p(x, C) (2)
and let only one curve of this family pass through each point of
the plane (or some région in the plane).
For what differential équation is this family of functions a com
plété intégral?
From relation (2), differentiating with respect to x, we find
C> (3 )
Since only one curve of the family passes through each point
of the plane, for every number pair x and y, a unique value of
C is determined from équation (2). Putting this value of C into
(3), we find as a function of x and y. This is what yields
the differential équation that is satisfied by every function of the
family (2).
\
Hence, to establish a relationship between *, y and ^ that is,

to Write a differential équation whose general solution is given by
formula (2), one has to eliminate C from relations (2) and (3).
Example 2. Find the differential
équation of the family of parabolas
y = Cx2 (Fig. 4).
Differentiating the équation of
the family with respect to x t we
get
Putting the value C = — into

this équation from the équation of
the family, we obtain a differential
équation of the given family:
dy_= îy_
dx x
This differential équation is
meaningful when x ^ 0 which is to
say, in any région not containing points on the y-axis.
1.4 EQUATIONS WITH SEPARATED AND SEPARABLE VARIABLES.

THE PROBLEM OF THE DISINTEGRATION OF RADIUM
Let us consider a differential équation of the form
i= fA * )h (y ) (O
where the right side is a product of a function dépendent only

on x by a function dépendent only on y. We transform it in the
following manner assuming that f t (y)¥= 0:
j ^ - ) dy = f 1{x)dx (1')
Considering y a known function of x, équation (T) may be regard-
ed as an equality of two differentials, while the indefinite intégrais
of them will differ by a constant term. Integrating the left side
with respect to y and the right with respect to x, we obtain
<r >
which is a relationship connecting the solution of y, the indepen-

dent variable x, and an arbitrary constant C; we hâve thus obtained
a general solution (complété intégral) of équation (1).
1. A type (T) differential équation
M(x)dx-\-N (y) dy=c 0 (2 )
1.4 Equations with Séparaied and Separable Variables 21
is called an équation with separated variables. Front what has

been proved, its complété intégral is .
\M(x)dx+$N(y)dy = C
Example 1. Given an équation with separated variables:
x d x + y dy = 0
Integrating we get the general solution:
X2 U2
T + T = C‘
Since the left side of this équation is nonnegative, the right side is also
nonnegative. Denoting 2C* by C2, we will hâve
x 2+ y 2 = C2
This is the équation of a family of concentric

circles (Fig. 5) with centre at the coordinate
origin and radius C.
2. An équation of the form

Af! (X) Nx (y) dx + M t (x) Nt (y) dy = 0 ( 3)
is called an équation with separable

variables. It can be reduced * to an
équation with separated variables by
dividing both sides by the expression N1(y)M2(x):
Mt (x) Ni (y) , M t (x) Nt (y) A
N i (y) M t ( x , a x T* Ni (y) M t (x) ü y ü
or
Mi W
dx + ^ d y = 0
A*«W ‘ Ni (y)
that is, to an équation like (2).
Example 2. Given the équation
ày _ y
dx x
Separating variables, we hâve
dy ____dx_
y x
Integrating, we find
I
* These transformations are permissible only in a région where neither N l (y)
nor Afa (x) vanish.
which is
ln | y | = —ln | x | + ln | C | * or ln |* /| = ln
£
x
ç
whence we get the general solution: y = — .

( \ + x ) y dx + (l —y) x d y = 0
Separating variables we hâve
(1 + *) dx-f i ç l d s r - O ,
x
Integrating, we obtain
\ n \ x \ + x + \ n \ y \ — y = C or ln | xy | + x — y = C
This relation is the complété intégral of the given équation.
Example 4. It is known that the decay rate of radium is directly propor-
tional to its quantity at each given instant. Find the law of variation of a
mass of radium as a function of the time if at / = 0 the mass of radium was m0.
The decay rate is determined as follows. Let there be mass m at time and
A /7 2
mass m + Am at time / + A/. During time At mass Am decays. The ratio is
the mean rate of decay. The limit of this ratio as A t —>-0
A m _dm
lim
a/ -►o ~Ât ~dt
is the rate of decay of radium at time t.
It is given that
dm
= — km (4)
HT
where k is the constant of proportionality (k > 0). We use the minus sign
dtn
because the mass of radium diminishes as time increases and therefore -rr < 0.
dt
Equation (4) is an équation with separable variables. Let us separate the
variables:
dm
= -kd t
m
Solving the équation we obtain
l n m = — kt-\-\nC
whence
ln f — «
m = Ce~kt (5)
Since at f = 0 the mass of radium was m0, C must satisfy the relationship
m0 = Ce~k'° = C
* Having in view subséquent transformations, we denoted the arbitrary

constant by ln | C I, which is permissible since ln | C | (when C £ 0) can take on
any value from -*-oo to + o o .
Î A Equations with Séparaied and Separable Variables 23
Putting the value of C into (5) we get the desired mass of radium as a function
of time (Fig. 6): #
m = m0e - * f (6)
The constant k is determined from observations as follows. During time tQ let
a% of the original mass of radium . Hence, the following relationship is
fulfilled:
whence
—kt0= \n ^ 1 Tôô)
or
Thus, it has been determined that for radium k = 0.000436 (the unit of time
measurement is one year).
Putting this value of k into (6) we obtain
m = m0e-°'OOQA36i
Let us find the half-life of radium, which is the interval of time during which
half of the original mass of radium decays. Putting in place of m in the
latter formula, we get an équation for determining the half-life T:
= m0e “ 0-000486r
whence
—0.0004367 = — ln 2
or
Some other problems of physics and chemistry can be reduced to équations

of the form (4).
Note 1. Let the function f 2(y) of (1) hâve a root y = b, i.e.,
ft (b) = 0. Then the function y ~ b is clearly a solution of équation
(1); this can be readily verified by direct substitution. The formula
(T) may not yield the solution y = b. We will not consider this
case here, but will note that the uniqueness condition may fail
to hold on the line y — 0.
By way of example, the équation y’ = 2 Ÿ y has the general
solution y = ( x + c f and the solution y = 0, which does not follow
from the general solution. The uniqueness condition is violated
on the line y = 0.
Note 2. The simplest differential équation with separated vari
ables is one of the form
Tx = f ( x) or dy — f (x) dx
24 Ch. / Differential Equations
Its general solution is of the form

y = $ f(x)dx + C
We dealt with the solution of équations of this kind in Ch. 10,
Vol. I.
1.5 HOMOGENEOUS FIRST-ORDER EQUATIONS

Définition 1. The function f(x, y) is called a homogeneous
function of degree n in the variables x and y, if for any A, the
following identity is true:
f(Xx, Xy) = Xnf (x, y)
Example 1. The function f (x, y) = p ^ x 3 + y3 is a homogeneous function of
degree one, since
f(Xx, X y ) = l / ( X x ) 3 + (Xy)3 = X \ / x 3 + y3 = Xf (x, y)
Example 2. f (x, y) = x y — y2 is a homogeneous function of degree two, since
(A*) (Xy)— (Xy)2= X2 (xy— y2).
.A
Example 3. / (x, y) = is a homogeneous function of degree zéro since
xy
(Xx)2— (Xy)2 _ x 2— y 2
(A x H A y )------ xy ' that is’ ^ ) = /(* . V) or /(A*. Xy) = X»f (x, y).
Définition 2. An équation of the first order
$=/(*. y) (1)
is called homogeneous in x and y if the function f(x, y) is a ho

mogeneous function of degree zéro in x and y.
Solution of a homogeneous équation. It is given that f (Ax, Xy) =
= f ( x >y)- Putting A,= ÿ in this identity, we hâve
/(* . y ) = f ( i > i )
Thus, a homogeneous function of degree zéro is dépendent only

on the ratio of the arguments.
In this case, équation (1) takes the form
Making the substitution

î-/('-î) <>')
u = -j or y = ux
we get
dy , du
5 - “ + æ '£
1.5 Homogeneous First-Order Equations 25
Putting this expression of the dérivative into équation (T), we

obtain •
This is an équation with variables separable

du £ ,, v du dx
x d ii = f ^ or n i;u )-u le
Intégrât ing we find
Putting the ratio — in place of U after Intégration, we get the

intégral of équation (1').
Exemple 4. Given the équation
dy xy
dx xi — yi
On the right is a zero-degree homogeneous function, which means that we hâve
a homogeneous équation. Making the substitution -“ = u> we hâve
du . du
y~ux, âc = u + x Tx
. du u du u9
U+ X d x ~ T = I ? ’ Xd x ~ l- u *
Separating variables, we obtain
<1 U9 X \ U9 U J X
Whence, integrating, we find
“ 2 ^ ~ Ul , “ 1= ,n|jC , + , n | C | OT - ^ ? = l n \ uxC \-
Substituting u = - j , we get the general solution of the original équation:
- £ = l n |C ÿ|
It is impossible here to get y as an explicit function of x in terms of elemen-

tary functions. Incidentally, it is very easy to express x in terms of y:
x = y Y —2 \n\C y \
Note. An équation of the type
■M(x, y)dx + N(x, y)dy = 0
will be homogeneous if, and only if, M(x, y) and N (x, y) are
homogeneous functions of the same degree. This follows from the
26 Ch. I Differential Equations
fact that the ratio of two homogeneous functions of the same de-
gree is a homogeneous function of degree zéro.
ExampJe 5. The équations
(2* + 3y) dx+ (x —2y) dy = 0
(x2+ y2) dx — 2xy dy = 0
are homogeneous.
1.6 EQUATIONS REDUCIBLE TO HOMOGENEOUS EQUATIONS

Equations of the following type are reducible to homogeneous
équations:
dy___ ax + b y + c
dx axx + bxy -f-cx (i)
If c1= c = 0, then équation (1) is obviously homogeneous. Now let

c and c, (or one of them) be different from zéro. Make a change
of variables:
x=Xy+h, y = yi + k
Then
dy = dyi
dx dxx ( 2)
Putting into (2) the expressions for x, y , and , we obtain

dyi axy + byy-{-ah+bk-\-c
dXi OyXy+ t>iy i + ayh -\-byk -\-Cy
Choose h and k so that the following équations are fulfilled:

cth -|- bk -f- c —0 1
üyh -\-byk-\-Cy— 0 / ^
In other words, define h and k as solutions of a System of équa
tions (4). Equation (3) then becomes homogeneous:
dy\_ axj + byy
dXy O yX y + b y ÿ y
Solving this équation and passing once again to x and y by

formulas (2), we obtain the solution of équation (1).
The System (4) has no solution if
a b
a i by
i .e.,aby = ayb. But in this case ^f- = ~ = %, that is, üy = %a, by = Kb,
and, hence, équation (1) may be transformed to
dy (a x + b y ) + c
dx X{ax-\-by)-\-Cy
1.6 Equations Reducible to Homogeneous Equations 27
Then by substitution
z = ax-\-by • ( 6)
the équation is reduced to one with variables separable.

Indeed,
dz
—
dx
= a +ri bl. dy
—
dx
whence
d y_ _ }_ d z a
dx b dx b ( 7)
Putting into (5) expressions (6) and (7), we get

1 dz a _ z-j-c
b dx b ~~ Xz + Ci
which is an équation with variables separable.

The device applied to integrating équation (1) is also applied
to the intégration of the équation
dy = f ( ax + by + c \
dx 1 [ a ^ + b tf+ C iJ
where f is an arbitrary continuous function.
dy x + y —3
dx x —y — 1
To convert it into a homogeneous équation, make the substitution = X i-\-h,
y = yi + k. Then
d y i _ Xi-{-y1-\- h -\-k — 3
dx i *i —yi + h — k — 1
Solving the set of two équations
h + £ —3 = 0, h - k - 1=0
we find
h = 2, k= l
As a resuit we get the homogeneous équation
dy i __ ~t~j/i
dxi x 1— yl
which we solve by substitution:
Ï1 U
*1
then
dyx f du
d F r u + X i dTi
du 1+ «
U -\-X 1
~dxx 1— w
28 Ch. î Differential Equations
and we get an équation with variables separable

du 1+ u2
x' d j r r r = ï ;
Separating the variables, we hâve
l—u dxi
du
1+ ua
arctan u — y ln (1 + aa) = ln | xx | + ln | C|
arctan u == In \C xx Y 1+ u2 |
or
Cxx Y ï + t f = eaTci*nu
P u ttin g — in place of u, we obtain
*i
arctan —
c V x \+ y \= e
Passing to the variables x and y , we finally get
.___________________ arctan u-
-—-1
C /( * - 2 ) * + (y -l)» = e
Example 2. The équation
2x+ y— \
y' = 4 x + 2 y + 5
cannot be solved by the substitution x = x x+ h9 y = y x + k t since in this case

the set of équations that serves to détermine h and k is insolvable (here, the
2 11
1^ 2 °f ^ e coefficients of the variables is equal to zéro).
This équation may be reduced to one with variables separable by the
substitution
2x + y = z
Then y' = z' —2 and the équation is reduced to the form
z—1
z' —2 =
2z+5
or
5z + 9
z =
2z + 5
Solving it we find
|* + l l n |5 z + 9 |= * + C
Since z = 2 x + y , we obtain the final solution of the initial équation in the form
- H - ( 2 * + y ) + ^ ln | 1 0 * + 5 iH -9 |= x + C
or
lOy— 5jc+7 ln | 1 0 * + 5 y + 9 | = CX
that is, as an implicit function y of x.
1.7 First-Order Linear Equations 29
1.7 FIRST-ORDER LINEAR EQUATIONS

Définition. A first-order linear équation is an équation that is
linear in the unknown function and its dérivative. It is of the
form
% + P (x)y= Q (x) (1)
where P (x) and Q(x) are given continuous functions of x (or are
constants).
Solution of linear équation (1). Let us seek the solution of
équation (1) in the form of a product of two functions of x :
y = u(x)v(x) (2)
One of these functions may be arbitrary, while the other will
be determined from équation (1).
Difîerentiating both sides of (2), we find
du dv . du
£ = U Âc + Vàc
,4k into (1), we
Putting the expression obtained of the dérivative^
hâve
(h) . (tti i n
u *c + d ï V + P u V = Q
ÙT
u { f x + P v ) + v Tx =C^ <3)
Let us choose the function v such that
X + p° - 0 (4)
Separating the variables in this difîerential équation with respect
to the function v, we find
^V ----- Pdx
Integrating we obtain
—In| Cx1+ ln 11>| = — J Pdx
or
- f Pdx
v = Cje J
Since it is sufficient for us to hâve some nonzero solution of
équation (4), we take, for the function v(x),
- f Pdx
v(x) = e J (5)
where J Pdx is some antiderivative. Obviously, o (x )^ 0 .
Putting the value of v(x) which we hâve found into (3), we get
^notingthat ^ + P i> = o j:
or
du _ Q (x)
dx v (x)
whence
5
“- U i d x+ c
Substituting into formula (2), we finally get
or
y = v (x) § T $ d x+ c °(x) (6)
Note. It is obvious that expression (6) will not change if in

place of the function u(x) defined by (5) we take some function
ul (x) = Cu(x). Indeed, putting vt (x) in (6) in place of v(x), we
get
y = C v ( x ) ^ ^ d x + CCv(x)
The C in the first term cancel out; in the second term the product
CC is an arbitrary constant, which we shall dénoté by C, and we
again arrive at expression (6). If we dénoté dx = qp(x), then
expression (6) will take the form
y = v (x)y (*) + Co(x) (6')
It is obvious that this is a complété intégral, since C may be
chosen in such a manner that the initial condition will be satisfied:
when x = xu, y = y0
The value of C is determined from the équation
1/q= ï '( * o)<P(* o) + C ü (*„)
Example. Solve the équation
^ ~ F f T ^/=(JC+1),
1.7 First-Or der Linear Equations 31
Solution. Putting
y — uv
we hâve
dy dv . du
dx dx 1 dx
Putting the expression ^^ into

; the original équation, we obtain
dv . du 2 , .
uTx+dxv- J + î uv = (x+]r
f dv 2 \ I du .
HU - F n t,) + t ’3 r =(* + , >* (7)
To détermine v we get the équation
dv 2
-- ------~ t > = 0
dx x + \
that is,
dv 2dx
~ v~ ~ x+ l
whence
l n |o | = 2 1 n |x + l | or o = (x -f l)a
Putting the expression o/ the function v into équation (7), we get the following
équation for u:
or du 7 , ,v
( x + W d£ = ( x + l ) * £ = (* + !)
whence
U -& & + C
Thus, the complété intégral of the given équation will be o! the form
_ (* + l)4 ~ C (x + 1)2
y=
The family obtaihed is the general solution. No matter what the initial con
dition (x0, y0), where x0 ^ — 1, it is alway6 possible to choose C so that the
corresponding particular solution should satisfy the given initial condition. For
example, thehe particular solution that satisfies the condition y0 = 3 when x0= 0
is found
' * as folio
' Mows:
(0 + 1 )4 - C ( 0 + l) a; C=y
Consequently, the desired particular solution is

_ (* + l)4 , 5
y-- ( * + l) 2
However, if the initial condition (x0, y0) is chosen so that *« = — 1, we will

not find the particular solution that satisfies this condition. This is due to the
2
fact that when x0= — 1 the function P (x) = — *s discôntinuous and,hen-
ce, the conditions of the theorem of the existence of a solution are not observed.
Note. In applications, one frequently encounters linear équa

tions with constant coefficients:
Tx + ay = b (8 )
where a and b are constants.

This équation may be solved either by the substitution (2) or
by séparation of variables:
dy = (— ay + b)dx, _ ^ +b=dx, — -M n| — ay + b\ ^ x + Ct
l n |— ay + b | = — (ax + C*), where C* = aCx
— a y + b = <r<"+c*>, y--= — ± e-<«+c*) + l
or, finally,
^where — - e ~ c*= . This is the general solution of équation (8).
1.8 BERNOULLI’S EQUATION
We consider an équation of the form *

% + P(x)y= <H x)y* (i)
where P (x) and Q (x) are continuous functions of x (or constants),

and n=^= 0 and 1 (otherwise we would hâve a linear équation).
This équation is called Bernoulli's équation and reduces to a linear
équation by the following transformation.
Dividing ail terms of the équation by y", we get
y-n% + Py~ n+1 = Q (2)
Making the substitution
2 = y^n+1
we hâve
s-1 — +1> r ‘ l
* This équation results from the problem of the motion of a body provided
the résistance of the medium F dépends on the velocity: F — X1v + \ 2vn.
The équation of motion will then assume the form m or
dv . Xi X.
1.8 Bernoulli*s Equation 33
Substituting into (2), we get

- § + ( - « + l)/>z = ( - n + l ) Q
This is a linear équation.
Finding its complété intégral and substituting the expression
y~n+1 for z, we get the complété intégral of the Bernoulli équation.
ÿx + x y = x3y3 (3)
Solution. Dividing ail terms by y3, we hâve

y - 3y' + x y ~ 2 (4)
Introducing the new function
-y-
we get
t e = - 2 y - * dJL
dx y dx
Substituting tjiese values into équation (4), we obtain the linear équation
dz
-2xz = — 2xr* (5)
dx
Let us find its complété intégral:
dz dv . du
Z — u v \ - ~ = u - -- |—
—j—V
dx dx 1 dx
Put expressions z and —■ into (5):
dv . du _ ^ .
u -7— \-~r v — 2xuv = — 2x*
dx 1 dx
or
u { ê - 2xv) + v Tx= - 2*
Equate to zéro the expression in the brackets:
dv A A dv .
------2xv = 0, — = 2 x d x
dx v
l n |o | = jt2, v = ex*
For u we get the équation
rî du n«
ex ^ - = — 2x*
dx
du = — 2e~x*x3 dx, « = — 2 J e~x*x? d x+ C
Integrating by parts, we find
u==x2e~x* + +
z = uv = x2-t-l-i-Cex*
8 2082
34 Ch. / Differential Equations
Consequently, the complété intégral of the given équation is
y ~ 2 = x2-{-1 -{-Ce*2 or y=
V x 2+1 + Ce*2
Note. Just as was done for linear équations, it may be shown that
the solution of the Bernoulli équation may be sought in the
form of a product of two functions:
y = u(x)v (*)
where u(jc) is some nonzero function that satisfies the équation
v' + Pv —0.
1.9 EXACT DIFFERENTIAL EQUATIONS

Définition. The équation
M(x, y)dx + N (x, y)dy = 0 ( 1)
is called an exact differential équation if M (x , y) and N (x, y) are
continuous différentiable functions for which the following rela-
tionship is fulfilled
dM dN
dy dx ( 2)
, dM . dN ..
and and t- are continuous in some région.
Integrating exact differential équations. We shall prove that if
the left side of équation (1) is an exact differential, then condi
tion (2) is fulfilled, and, conversely, if condition (2) is fulfilled
the left side of équation (1) is an exact differential of some fun
ction u(x, y). That is, équation (1) is an équation of the form
du(x, y) ~ 0 ( 3)
and, consequently, its complété intégral is

u(x, y) = C
Let us first assume that the left side of (1) is an exact diffe
rential of some function u(x, y)-, that is,
M(x, y) dx + N (x, y)dy = d u ^ — dx + ~ d y
then
( 4)
Differentiating the first relation with respect to y, and the

second with respect to x, we obtain
dM â2u dN â2u
1.9 Exact Differential Equations 35
Assuming continuity of the second dérivatives, we hâve

d M _dN
dy dx
that is, (2) is a necessaiy condition for the left side of (1) to be
an exact differential of some function u (x, y). We shall show that
this condition is also sufficient: if (2) is fulfilled then the left
side of (1) is an exact differential of some function u(x, y).
From the relation
we find
X
u = [ M (x , y)dx + <p(y)
*o
where x0 is the abscissa of any point of the domain of existence
of the solution.
When intégrâting with respect to x we consider y constant, and
therefore the arbitrary constant of intégration may be dépendent
on y. Let us choose a function <p(y) so that the second of the
relations (4) is fulfilled. To do this, we difïerentiate* both sides
of the latter équation with respect to y and eauate the resuit to
N (x%y):
| J = l i ü d x + v ' ( y y ^ N (x ' y )
. . . dM dN
but since = , we can write
X
^ d x + ÿ (y) = N that is, N (x, y) | £, + <p' (y) = N {x, y)

X0
or
N (x, y) — N (x0, y) + q>' (y) = N (x, y)
Hence,
<p' (y) = N (x0, y)*3
The intégral J M (x, y)dx is dépendent on y. To find the dérivative of

*0
this intégral with respect to y , difïerentiate the integrand with respect to y:
X X
^ ^ AI (x, y) dx= H d x . This follows from Leibniz’ theorem for differen-

xn JC„
tiating a definite intégral with respect to a parameter (see Sec. 11.10. Vol. I).
3*
or
qp(ÿ)= J N (x0, y)dy + C1
yo
Thus, the function u(x, y) will hâve the form
* y
u= (.v, y)dx+ 5 N (xt , y)dy + Ct
x. y»
Here P (x0, ya) is a point in the neighbourhood of which there
is a solution of the differential équation (1).
Equating this expression to an arbitrary constant C, we get the
complété intégral of équation (1):
x y
l M (x, y)dx + J N (x0, y)dy = C (5)
xo yo
Example. Given ihe équation
2x , . y2 — 3x2 .
y3 X 1 y4 y ~
Let us check to see whether this is an exact differential équation.
Denoting
yJ y4
we hâve
dM 6x M 6*
ày ' 'y * ' d x z ’P
For y £ 0, condition (2) is fulfilled. Hence, the left side of this équation is
an exact differential of some unknown function u (x, y). Let us find this function.
Since ~ = , it follows that
dx y3
« = J •p -< te + « p (ÿ > = p -+ ç (y )
where <p (y) is an as yet undetermined function of y.
Differentiating this relation with respect to y and noting that
~3x2
ày
we find
3x* , , . . y2 — 3x2
— r r f q » (y)=-
y4
hence
9 (y) —— - 4 -
•2
“ (*• y ) = ÿ “ 7 + Cl
Thus the complété intégral of the original équation is
x2 1
1.10 Integrating Factor 37
1.10 INTEGRATING FACTOR
Let the left side of the équation •

M(x, y)dx + N (x, y)dy = Q (1)
not be an exact differential. It is sometimes possible to choose
a function p(x, y) such that after multiplying ali terms of the
équation by it the left side of the équation is converted into an
exact differential. The general solution of the équation thus ob-
tained coïncides with the general solution of the original équation;
the function p(.r, y) is called the integrating factor of équation (1).
In order to find the integrating factor p, do as follows. Mul-
tiply both sides of the given équation by the as yet unknown
integrating factor p:
p M dx + pW dy = 0
For this équation to be an exact differential équation, it is neces-
sary and sufficient that the following relation be fulfilled:
d (pAf ) _d (pJV)
d ÿ ~ ~ dx
that is.
i1 dy ' Vl dy ^ dx ' dx
or
dy dx ^ [d x ây J
After dividing both sides of the latter équation by p, we get

A /fd l n ^ A 7c ) l n j L i dN 0M
M ~dÿ N ~ d x ~ ~ ~ d 7 ~ ~dÿ ( 2)
It is obvious that any function p(x, y) that satisfies this équa

tion is the integrating factor of équation (1). Equation (2) is
a partial differential équation in the unknown function p dépen
dent on the two variables x and y. It can be proved that under
definite conditions it has an infinitude of solutions and that, con-
sequently, équation (1) has an integrating factor. But in the gene
ral case, the problem of finding p (x, y) from équation (2) is har-
der than the original problem of integrating équation (1). Only
in certain particular cases does one manage to find the function
P (x, y)-
For instance, let équation (1) admit an integrating factor dépen
dent only on y . Then
and to find p we obtain an ordinary differential équation,

dN ^dM
d In fi _ dx dy
dy M
from which we détermine (by a single quadrature) ln p., and, hence,
H as well. It is clear that this can be done only if the expression
dN dM
dx dy- is not dépendent on x.
â_N_â_M
Similarly, if the expression dx Ndy is not dépendent on y but
only on x, then it is easy to find an integrating factor that
dépends only on x .
(y 4- xy2) d x— x d y = 0
Solution. Here, M = y-\-xy2t N — — x
dM dM , diï
= 1 + 2 xy.
dy dx “ dy F dx
Thus, the left side of the équation is not an exact differential. Lèt us see
whether this équation allows for an integrating factor dépendent only on y or
not. Noting that
dN dM
dx d y _—1— 1— 2xy 2
M ~ y + xy2 “ y
we conclude that the équation permits of an integrating factor dépendent
only on y. We find it:
d In | L i _ _ _ _ _ _ 2_
dy ~ ~ ~ ÿ
whence
ln (i = — 2 ln y , i.e., =
After multiplying through by the integrating factor p, we obtain the équation
( 7 + x ) d x~ — dy = Q
as an exact differential équation ( ^ - = ^ L = z -----!- j^ . Solving this équation.

\d y dx y‘
we find its complété intégral:
J + T +c =0
or
2x
x2+ 2C
1.11 The Envelope of a Family of Curves 39
1.11 THE ENVELOPE OF A FAMILY OF CURVES

Let there be an équation of the form#
<£(*. y , C) = o ( 1)
where x and y are variable Cartesian coordinates and C is a para-
meter that can take on a variety of fixed values.
For each given value of the pa-
rameter C, équation (1) defines some
curve in the xy-plane. Assigning
to C ail possible values, we obtain
a family of curves dépendent on
-H
Fig. 7 Fig. 8
a single parameter, or using the more common term, a one-para-

meter family of curves. Thus, équation (1) is the équation of a
one-parameter family of curves (because it contains only one arbit-
rary constant).
Définition. A curve L is called the envelope of a one-parameter
family of curves if at each point it touches a curve of the family,
and different curves of the given family touch the curve L at
different points (Fig. 7).
Exampfe 1. Consider the family of curves
(x -C )* + ÿ* = fl*
where R is a constant and C is a parameter.
This is a family of circles of radius R with centres on the x-axis. This
family will obviously hâve as envelopes the straight lines y = R and j/ = — R
(Fig. 8).
Finding the équation of the envelope of a given family. Let
there be given a family of curves,
0 ( x . if, C) = 0 (1)
that dépend on the parameter C.
Let us assume that this family has an envelope whose équation
may be written’ in the form i/ = q>(x), where <p(x) is a continuous
and différentiable function of x. We consider some point M (x, y)
lying on the envelope. This point also lies on some curve of the
family (1). To this curve there corresponds a definite value of fhe
parameter C, which value is determined from équation (1), for
given (x , y): C = C(x, y). Thus, for ali points of the envelope the
following équation holds:
<D[x, y, C (x1, y)\ = 0 (2)
Suppose that C(x, y) is a différentiable function that is not con
stant in any interval of the values of x and y under considération.
From équation (2) of the envelope we find the slope of the tan
gent to the envelope at the point M (x, y). Differentiate (2) with
respect to x assuming that y is a function of x:
, d® a c , p<D . cXD <3C] , _ n
dx + dC dx • [ d y dC dy \ y ~ U
or
v * + v j+ ® h [ § + § y '] = o (3)
The slope of the tangent to the curve of the family (1.) at the
point M (x, y) is found from
—0 (4)
(on this curve, C is constant).
We assume that «D^Ô, otherwise we would consider x as the
function and y as the argument. Since the slope k of the envelope
is equal to the slope k of the curve of the family, from (3) and
(4) we obtain
*[§+£']-•
But since on the envelope C(x, y)=^= const, it follows that
dC . dC , , n
and so for its points the following équation holds:

<t>c(x, y, C) = 0 (5)
Thus, the following two équations serve to détermine the envelope:
® ( x ,y ,C ) = 0 \
(x, y, C) — 0 f w
Conversely, if, by eliminating C from these équations, we get an
équation i/ = <p(a:), where q>(x) is a différentiable function and
C=#= const on this curve, then ÿ = <p(x) is the équation of the
envelope.
Note 1. If for the family (1) a certain function y = (p(x) is the
équation of the locus of singular points, that is, of points where
i = 0 and <1)^= 0, then the coordinates of these points also satisfy
équations (6).
l . î l The Envelope of a Family of Curves 41
Indeed, the coordinates of singular points may be expressed

in terms of the parameter C that en têts into équation (1):
* = MC). y = v (C ) « (7)
If these expressions are substituted in équation (1), we get an
identity in C:
<D[MC), p (C), C] = 0
Differentiating this identity with respect to C, we obtain
Since for any points the equalities <J>i = 0, ^ = 0 are fulfilled,
it follows that for them the equality G>c = 0 is also satisfjed.
We hâve thus proved that the coordinates of singular points
satisfy équations (6).
Summarizing, équations (6) define either the envelope or the
locus of singular points of the curves of the family (1), or a
combination of both. Thus, after obtaining a curve that satisfiés
équations (6), one has further to find out whether it is an envelope
or the locus of singular points.
Example 2. Find the envelope of the family of circles
( * - C ) 2 + £/2- / ? 2 = 0
that are dépendent on the single parameter C.
Solution. Differentiating the équation of the family with respect to C,
we get
2 (x — C) = 0
Eliminating C from these two équations, we obtain the équation
y2 — R2 = 0 or y = ± R
It is clear, by géométrie reasoning, that the pair of straight Unes is the
envelope (and not the locus of singular points, since the circles of a family
do not hâve singular points).
Example 3. Find the e;ivelope of the family of straight lines
xcos a -t y sin a —p = 0 (a)
where a is a parameter.
Solution. Differentiating the given équation of the family with respect
tu oc, we hâve
—x sin a + y cos a = 0 (b)
To eliminate the parameter a from équations (a) and (b), multiply the
terms of the fîrst by cos a, and of the second, by sin a, and then subtract
the second from the first; we then hâve
x = p cos a
Putting this expression into (b), we find
y — P sin a
Squaring the terms of the last two équations and adding termwise, we get
x2 + y 2 = p2
This is a circle. It is the envelope of the family (and not the lccus of singular
points, since straight lines do not hâve singular points) (Fig. 9).
Example 4. Find the envelope of the trajectories of projectiles fired from
a gun with velocity u0 at different angles of inclination of the barrel to the
horizon. We shall consider that the gun is located at the coordinate origin and
that the trajectories of projectiles lie in the jq/-plane (air résistance is disre-
garded).
Solution. First find the équation of the trajectory for the case'when the barrel
makes an angle a with the positive x-axis. In flight, the projectile participâtes
simultaneously in two motions: a uniform motion with velocity v0 in the direc
tion of the barrel and a falling motion due to the force of gravity. Therefore,
at each instant of time / the position of the projectile M (Fig. 10) will be
defined by the équations
x = v0t cos a
at2
y = v0t sin a — — ~
These are parametric équations of the trajectory (the parameter is the time t).
Eliminating t , we get the équation of the trajectory in the form
gxi
y = * ta n a —
2v\ cos2 a
a
Finally, introducing the notation tan a = k t = a, we get
y = k x —-ax2 (1 + /e2) (8)

This équation defines a parabola with vertical axis passing through the origin
and concave down. We obtain a variety of trajectories for the different values of k.
Consequently, équation (8) is the équation of a one-parameter family of para-
bolas, which are the trajectories o! a projectile for different angles a and for
a given initial velocity v0 (Fig. 11).
Let us find the envelope of this family of parabolas. Differentiating both
sides of (8) with respect to k , we hâve
x —2akx2 = 0 (9)
Eliminating k from équations (8) and (9), we get
1.11 The Envelope of a Family of Curves 43
This is the équation of a parabola with vertex at the point (^0, , the axis
of which coïncides with the y-axis. It is not £ locus of singular points [since
parabolas (8) do not hâve singular points]. Thus, the parabola
is the envelope of the family of trajectories. It is called a safety parabola be-

cause no point outside it is in reach of a projectile fired from a given gun with
a given initial velocity v0.
Example 5. Find the envelope of the family of semicubical parabolas
y3 — (x — C)2 = 0
Solution. DifTerentiate the given équation of the family with respect to the
parameter C:
2 (x —C) = 0
Eliminating the parameter C from the two équations, we get
y=o
The x-axis is a locus of singular points —a cusp of the first kind (Fig.. 12).
Indeed, let us find the singular points of the curve
y3— (x—C)2 = 0
for a fixed value of C. Differentiating with respect to x and y, we find
F; = - 2 ( x - C ) = 0
Fy — 3y2 = 0
Solving the three foregoing équations simultaneously, we find the coordinates
of the singular point: x = C, y = 0; thus, each curve of the given family has
a singular point on the x-axis.
For continuous variation of the parameter C, the singular points will fill the
entire x-axis.
Example 6. Find the envelope and locus of singular points of the family
(y ~ C )* -± (x -C )* = 0 ( 10)
Solution. Differentiating both sides of (10) with respect to C, we find
—2 (y _ C ) + y 3 < * - C ) a = 0
or
y - C — (x-~C)*=: 0 00
into the équation of the family, we get

(* _ C )* -y (x --C )a = o
or
(AT-C)S =0
whence we obtain two possible values of C and two solutions of the problem
corresponding to them.
First solution: Second solution.
C —x
C==X~ T
and so from (11) we find and so from (11) we find
or
y —x — (jc-x )2= 0
*-*+i-[*-*+ r=° 4
* “ •*— T
1.12 Singular Solutions 45
2
We hâve obtained two straight Unes: y — x and y - x — -^-. The first is the locus
of singular points, the second is the envelope (Fig. 13).
Note 2. In Sec. 6.7, Vol.I it was prtfved that the normal to a
curve serves as a tangent to its evolute. Hence, the family of
normals to a given curve is at the same time a family of tangents
to its evolute. Thus, the evolute of the curve is the envelope of the
family of normals of this curve (Fig. 14).
This remark enables us to point out another method for finding
evolutes: to obtain the équation of an evolute, first find the family
of ail normals of the given curve and then find the envelope of
this family.
1.12 SINGULAR SOLUTIONS OF A FIRST-ORDER
DIFFERENTIAL EQUATION
Let the differential équation
hâve a complété intégral

<D(*, y, C) = 0 (2)
Let us assume that the family of intégral curves that corresponds
to équation (2) has an envelope. We shall prove that this envelope
is also an intégral curve of the differential équation (1).
Indeed, at each point the envelope touches some curve of the
family; that is, it has a common tangent with it. Thus, at each
common point the envelope and the curve of the family hâve the
same values of x, y, y'.
But for a curve of the family, the numbers x, y, and y' satisfy
équation (1). Consequently, the very same équation is satisfied by
the abscissa, the ordinate and the slope of each point of the
envelope. But this means that the envelope is an intégral curve
and its équation is a solution of the given differential équation.
Since, generally speaking, the envelope is not a curve of the
family, its équation cannot be obtained from the complété inté
gral (2) for any particular value of C. The solution of the differential
équation which is not obtained from the complété intégral for any
value of C and which has as its graph the envelope of the family
of intégral curves entering into the general solution, is called a
singular solution of the differential équation.
Let the complété intégral be known:
<D(x, y ,C ) = 0
éliminât ingC from this équation and from the equationc(x,J/,C)=0,
we get t|i(x, y) — 0. If this function satisfies the differential équation
[and does not belong to the family (2)], then it is a singular intégral.
It should be noted that at least two intégral curves pass through

each point of the curve that describes a singular solution; that is,
uniqueness of solution is violated at each point of a singular
solution.
We note that a point at which uniqueness of solution of a diffe-
rential équation has been violated, that is, a point through which
at least two intégral curves pass, is called a singular point *. Thus,
a singular solution consists of singular points.
Example. Find a singular solution of the équation
y2 ( i+ y f2) = R 2 (*)
Solution. Let us find its complété intégral. We solve the équation for y':
dy V R * -y *
dx y
± V R 1- y 2
Whence, integrating, we find the complété intégral:
( * - C ) 2 + (/2 = fl2
It is easy to see that the family of intégral Unes is a familvof circles of

radius R with centres on the jr-axis. The pair of straight Unes y = ± R will
be îh e envelope of the family of curves.
The functions y = ± R satisfy the differential équation (*). This, conse-
quently, is a singular intégral.
1.13 CLAIRAUT’S EQUATION
Let us consider the so-called Clairaut équation:
y = x%+^{%) (h
It is integrated by introducing an auxiliary parameter. P u t-^ = p;
then équation (1) will take the form
y = x p + * (p ) (V)
Differentiate ail the terms of this équation with respect to x,
bearing in mind that p = is a function of x :
P = X% + P + * ' W %
* The boundary points of the domain of existence of a solution are also

termed singular points. An interior point through which a unique intégral curve
of a differential équation passes is called an ordinary point.
1.13 Clairaut's Equation 47
or
[ x + Ÿ (p)]d£ ~ o
Equating each factor to zéro, we get
and
* + ♦'(/>) = 0 (3)
(1) Integrating (2) we obtain p = C(C — const). Putting this
value of p into (T), we find its complété intégral:
y = xC + y{C) (4)
which, geometrically, is a family of straight lines.
(2) If from (3) we find p as a function of x and put it into
(T), we obtain the function
t/ = x p (x )+ i)> [/?(*)] (1")
which may be readily shown to be the solution of équation (1).

Indeed, by virtue of (3), we hâve
| = p + [ .v + t|) '( p ) ] |, i.e. % = p
And so, by substituting the function (1") into équation (1) we get
the identity
The solution (1") is not obtained from the complété intégral (4)
for any value of C. This is a singular solution; it is obtained by
élimination of the parameter p from the équations
y = xp + Mp(p) ^
x + i|>'(/>) = 0 j
or, which is the same thing, by eliminating C from the équations
y = xC 4- ib (C)
* + tc (Q = 0
Thus, the singular solution of Clairaut’s équation defines the
envelope of a family of straight lines represented by the complété
intégral (4).
Example. Find the general and singular solutions of the équation
dy
a
dy dx
y=x
dx 2
/■+(=)
4a Ch. 1 Differential Equations
Solution. The general solution is obtained by substituting C for

aC
y = xC-
FT +c5
To obtain the singular solution, differentiate the latter équation with res
pect to C:
(1 + C2) 2
The singular solution (the équation of the envelope) is obtained in parametric
form (where the parameter is C):
(1 + C 2) 2
aC3
3
(1 + C2) 2
Eliminating C, we get a direct relationship be-
tween x and y Raising both sides of each equa-
2
tion to the power and adding the résultant
équations termwise, we get the singular solution
in the following form:
2 _2_ 2
x 3 + y, 33 = a '3
This is an astroid. However, the envelope of the
family (and, hence, the singular solution) is not
the entire astroid, but only its left half (since it is évident from the parametric
équations that jt< ;0) (Fig. 15).
1.14 LAGRANGE’S EQUATION
The Lagrange équation is an équation of the form

y = xy W )+ W ) ( 1)
where <p and t|3 are known functions of

This équation is linear in y and x. Clairaut’s équation, which
was considered in the preceding section, is a particular case of
the Lagrange équation when y(y') = y’. The Lagrange équation,
like Clairaut’s, is integrated bymeansof introducing an auxiliary
parameter p. Put
y’ = P
then the original équation is written in the form
y = x<p(p) + y(p) (1')
1.14 Lagrange's Equations 4$
Differentiating with respect to x , we obtain
p = v ( p ) + [ * ? ' ( p) + %
or
p — *P (p )~ N > ' (p) +♦' (p)] % (O
From this équation we can straightway find certain solutions:

namely, it becomes an identity for any constant value p = p„
that satisfies the condition
Po— <P(Po) = 0
Indeed, for a constant value p the dérivative ^ = 0, and both
sides of équation (1*) vanish.
The solution corresponding toeach value p = pt, that is, ^ = P #
is a linear function of x ^since the dérivative ^ is constant only
in the case of linear fonctions^. To find this function it is suf-
ficient to put into (T) the value p = p„ :
y = x<t>(Po) + V(Po)
If it tums out that this solution is not obtainable from the gener
al solution for any value of the arbitrary constant, it will be a
singular solution.
Let us now find the general solution.. Write (T) in the form
dx _ Ÿ (p )
dp * p — <p (p) p —<p (p)
and regard x as a function of p. Then the équation obtained

will be a linear differential équation in the function x of p.
Solving it, we find
x = <ù(p, C) (2)
Eliminating the parameter p from équations (T) and (2), we
get the complété intégral of (1) in the form 0 (x , y, C) —0.
Example. Given the équation
y = x y '* + y 'i (I)
Putting y ’ = p we hâve
ÿ =*p*+p* (F)
Differentiating with respect to x , we get
p= p» + [2 x p + 2 p ]g <n
4 2082
Let us find the singular solutions. Since p = p2 for p0= 0 and px= 1, the
solutions will be linear functions [see (I')]:
y = x -02 + 02, that is, y = 0
and
y= x+ 1
When we find the complété intégral, we will see whether these functions are
particular or singular solutions. To do so, write équation (T) in the form
dx 2 2
dp x \ — p i - p
and regard x as a function of the independent variable p. Integrating this
linear (in x) équation, we find
«“ - ' + 5 ^ <">
Eliminating p from équations (T) and (II), we get the complété intégral
y = (C+V"x+1)2
The singular intégral of the original équation is
y= 0
since this solution is not obtainable from the general solution for any value
of C.
However, the function p = * + 1 is not a singular but a particular solution;
it is obtained from the general solution when C = 0.
1.15 ORTHOGONAL AND ISOGONAL TRAJECTORIES
Suppose we hâve a one-parameter family of curves

<t>(x,y,Q = 0 (1)
Lines intersecting ail the curves of the given family (1) at a
constant angle are called isogonal trajedories. If this angle is
a right angle, they are orthogonal trajedories.
Orthogonal trajectories. Let us find the équation of orthogonal
trajectories. Write the differential équation of the given family of
curves, eliminating the parameter C from the équations
<D(x, y, C) —0
and
, d à y_ Q
dx ‘ dy dx
Let this differential équation be
F { x>y* ï )= 0 ( i f)
Here, -- is the slope of the tangent to some member of the

family at the point M (x, y). Since an orthogonal trajectory pass-
ing through the point Al (x, y) is perpendicular to the correspond-
1.15 Orthogonal and Isogonal Trajectories 51
ing curve of the family, the slope of the tangent to it, — , is

connected with ^ by the relation (Fig. 16)
dy_ 1
dx~~ dÿ£ ( 2)
dx
Putting this expression into équation (T) and dropping the
subscript T, we get a relationship between the coordinates of an
arbitrary point (x , y) and the slope of the orthogonal trajectory
at this point, that is, a differential
équation of orthogonal trajectories:
f (* • ». - - 5 V 0 <3>
The complété intégral of this equa*

tion
«M*. y> C) = 0
yields a family of orthogonal trajec
tories.
A considération of the plane flow
of a fluid involves orthogonal trajec
tories.
Let us suppose that the flow of
fluids in a plane takes place in such
a manner that at each point of the xy-plane the velocity vec-
tor v(x, y) is defined. If this vector dépends solely on the
position of the point in the plane, but is independent of the
time, the motion is called stationary or steady-state. We shall
consider such motion. In addition, we shall assume that there
exists a potential of velocities, that is, a function u(x, y) such
that the projections of the vector v(x, y) on the coordinate axes,
vx (x, y) and vy (x, y), are its partial dérivatives with respect to x
and y:
du
dx vx (4)
The Unes of the family
u(x, y) = C ( 5)
are called equipotential Unes (Unes of equal potential).

The lines, whose tangents at ail points coincide with the
vector v(x, y), are called flow Unes and yield the trajectories of
moving points.
We shall show that the flow lines are the orthogonal trajecto
ries of a family of equipotential lines (Fig. 17).
1
52 Ch. I Différentiat Equations
Let qp be an angle formed by the velocity vector v with the

x-axis. Then, by relation (4),
a“ <£y)==[«>lC0S<P. — y) = |t>lsin
dx
We obtain the slope of the tangent to the equipotential lineby
differentiating relation (5) with respect
to x :
du_ | du dj£_n
dx ‘ dy dx
whence
du
dy_ ÈL
dx~ du (7 )
àÿ
Thus, in magnitude and sign, the slope
of the tangent to the equipotential line is
the inverse of the slope of the tangent to
the flow line. Whence it follows that equipotential lines and flow
Unes are mutually orthogonal.
In the case of an electric or magnetic field, the lines of force
of the field serve as the orthogonal trajectories of the family of
equipotential lines.
Example 1. Find the orthogonal trajectories of the family of parabolas
y — Cx2
Solution. We Write the differential équation of the family:
y' = 2Cx
Eliminating C, we get
y x
Substituting — ^ for y \ we obtain a differential équation of the family of
y
orthogonal trajectories
Î.15 Orthogonal and Isogonal Trajectories 53
lis complété integra! is

£_i
T+2~c •
Hence, the orthogonal trajectories of the given family of parabolas will be
representexl by a certain family of ellipses with semi-axes a = 2C, b = C 1^2
(Fig. 18).
Isogonal trajectories. Let the trajectories eut the curve of a

given family at an angle a, where tana = k.
The slope -^ = tan to the isogonal trajectory are
connected by the relationship
tan — tan a
tan<p = tan(tf—a) 1+ ta n a tan t|?
that is, dyT ~~k
dy_ dx
dx ' ( 2')
Substituting this expression into équation (T) and dropping the

subscript T, we obtain the differential équation of isogonal tra
jectories.
54 Ch. i Differential Equations
Example 2. Find the isogonal trajectories of a family of straight lines,

y = Cx ( 8)
that eut the lines of the given family at an angle a , the tangent of which
equals k.
Solution. Let us Write the differential équation of the given family. Difife-
rentiating équation (8) with respect to ac
we find
On the other hand, from the same équa

tion we hâve
Consequently, the differential équation

of the given family is of the form
dy= ]L
dx x
Utilizing relationship (2') we get the
differential équation of isogonal trajec
tories
Fig. 20 & L -k
dx y
A + 1 *
dx 1
whence, dropping the subscript 7\ we find
dy
dx
i-k ±
Integrating this homogeneous équation, we get the complété intégral:
ln Vx? + y2 = ~ arc tan -~ + ln C (9)
which defines the family of isogonal trajectories. To find out precisely which
curves enter into this family, let us change to polar coordinates:
~ = tan<p, 'ffx 2+ y 2 = p
Substituting these expressions into (9), we obtain
ln p = y (p + lnC
p = Cek
Consequently, the family of isogonal trajectories is a family of logarithmic
spirals (Fig. 20).
1.16 Higher-Order Differential Equations 55
1.16 HIGHER-ORDER DIFFERENTIAL EQUATIONS

(FUNDAMENTALS)
As has already been indicated above (see Sec. 1.2), a differen

tial équation of the nth order may be written symbolically in
the form
F(x, y, y ’, y", y n =0 ( 1)
or, if it can be solved for the nth dérivative,
yw = f(x, y, y', y"......... tf*-1') (1')
ln this chapter we shall consider only such équations of higher
order that may be solved for the highest dérivative. For these
équations we hâve a theorem on the existence and uniqueness of
a solution, similar to the corresponding theorem on the solution
of first-order équations.
Theorem. If in the équation
0 (n) = /(■*. y, y ’, . . . , 0 *"-»)
the function f(x, y, y ', . . . , yln~1)) and its partial dérivatives with
respect to the arguments y, y', . . . . y(n~l) are continuous in some
région containing the values x-— x0, y —y0, y ’ -- y'0, .
yin-i) —yin-D' (fien there is one and only one solution, y —y(x), of
the équation that satisfies the conditions
i/*=jc0= y»
y'x~x,=y«
= y ln- l)
These conditions are called initial conditions. The proof is beyond
the scope of this book.
If we consider a second-order équation y" ~ f (x, y, y'), then the
initial conditions for the solution, when x ^ - x 0, will be
y=- y0’ y ' = y»
where x0, y0, y ’0 are given numbers, these conditions hâve the fol-
lowing géométrie meaning: only one curve passes through a given
point (x0, y0) of the plane with given tangent of the angle of
inclination of the tangent line y'0. From this it follows that if we
want to assign different values of y'0 for constant x0 and y0, we
get an infinitude of intégral curves with different angles of incli
nation passing through the given point.
We now introduce the concept of a general solution of an équa
tion of the nth order.
Définition. The general solution of a differential équation of the

nth order is a function
y— Cj, Cj, •••. Cn)
which is dépendent on n arbitrary constants Clt C%, . . . , C„ and
such that:
(a) it satisfies the équation for any values of the constants
Cj, • • • > Cn,
(b) for specified initial conditions
yx-=xQ= y«
yx—x„~ y<>
the constants Clt C2..........Cn may be chosen so that the func

tion y = y(x, Clt C2, Cn) will satisfy these conditions (on the
assumption that the initial values x0, y0, y'0, . . . , yin~l) belong
to the région where the conditions of the existence of a solution
are fulfilled).
A relationship of the form 0 (x, y, Clt C2, C„) = 0, which
implicitly defines the general solution, is called the complété
intégral of the differential équation.
Any function obtained from the general solution for spécifie
values of the constants C,, C2, Cn is called a particular
solution. The graph of a particular solution is called an intégral
curve of the given differential équation.
To solve (integrate) a differential équation of the nth order
means:
(1) to find its general solution (if the initial conditions are not
given) or
(2) to find a particular solution of the équation that satisfies
the given initial conditions (if there are such).
In the following sections we shall présent methods of solving
various équations of the nth order.
1.17 AN EQUATION OF THE FORM y ™ z = f(x )

The siraplest type of équation of the nth order is of the form
y w = f(x) (1)
Let us find the complété intégral of this équation.
Integrating the left and right sides with respect to x and
taking into account that yln) = (ÿlB-l,) \ we obtain
X
/ ■ - " = S / ( * ) < * * + c»
*•
1.17 An Equation of the Form ÿ(n) = f (x) 57
where x„ is any fixed value of x, and Cl is the constant of

intégration. .
Intégrât ing once more, we get
X / X
jd x ( x — x0) + C s
X, 'X ,
Continuing, we finally get (after n intégrations) the expression of

the complété intégral:
A *
dx- Cx (* -* » )” ->
( n - 2)1 + • • • + Cn
x, x.
In order to find a particular solution satisfying the initial condi
tions
yx=x.=yü, y'x=x,= y'0< ■■• • y(x=xl. ) = y\
it is sufficient to put
C„ = </«, C„^l = y\, . • •• Cl = yn~l
Example 1. Find the complété intégral of the équation
y ”= sin (kx)
and a particular solution satisfying the initial conditions
**-o = °- y'x-o= 1
Solution.
X
/ r . . . , ^ cos k x — i . ~
y = \ sin kxdx + Cx = ------------------
o
y =~ î ( c~—£ — - ) <* * +* * +( : ,
0 0
or
sin kx » x . ~ _
y= — jp— b y + C i x + c ,
This is the complété intégral. To find a particular solution satisfying the

given initial conditions, it is sufficient to détermine the corresponding values
of Q and C2.
From the condition yx=0 = 0 , we find C2= 0.
From the condition y ^ 0= l , we find ^ = 1 .
Thus, the desired particular solution is of the form
sin kx . ( 1 , , \
y ---------
Differential équations of this kind are encountered in the theory of the
bending of girders.
Example 2. Let us consider an elastic prismatic girder bending under the
action of external forces both continuously distributed (weight, load) and con-
centrated. Let the x-axis be horizontal along the axis of the girder in its un-
dèrformed state and let the i/-axis be directed vertically downwards (Fig. 21).
Each force acting on the girder (the load of the girder, and the reaction of
the supports, for instance) has a moment, relative to some cross section of the
girder, equal to the product of the force by the distance of the point of appli
cation of the force from the given cross sec
tion. The sum, M (jc), of the moments of ail
the forces applied to that part of the girder
situated to one side of the given cross section
with abscissa x is called the bending moment
of the girder relative to the given cross sec
tion. In courses of strength of materials,
it is proved that the bending moment of the
girder is
EJ
R
where E is the so-called modulus of elasticity
which dépends on the material of the girder,
J is the moment of inertia of the cross-secti-
onal area of the girder relative to the horizontal line passing through the centre
of gravity of the cross-sectional area, and R is the radius of Curvaturé of the
axis of the bent girder, which radius is expressed by the formula (Sec. 6.6,
Vol. I)
p _ il± Æ *
R~ I y" I
Thus, the differentiaf équation of the bent axis of a girder has thê fortn
y' _ M (x)
( 2)
(l + „'*)V. EJ
If we consider that the deformations are small and that the tangents to
the axis of the girder, when bent, form a small angle with the x-axis, we can
disregard the square of the small quantity y '2 and consider
\_
R
y"
Then the differential équation of the bent girder will hâve the form
M (x)
( 2 ')
EJ
but this équation is of the form of (1).

Example 3. A girder is fixed in place at the extremity O and is subjected
to the action of a concentrated vertical force P applied to the end of the
girder L at a distance / from O (Fig. 21). The weight of the girder is ignored.
We consider a cross section at the point N (x). The bending moment rela
tive to section N is, in the given case, equal to
M(x) = ( l - x ) P
The differential équation (2') has the form
P
1.Î8 Second-Order Equations Reducible to First-Order Equations 59
The initial conditions are: for x = 0 the deflection y is equal to zéro and the
tangent to the bent axis of the girder coïncides with the *-axis; that is,
0 *=o = °- y'x=o ^ °
Integrating the équation, we find
y' = î h î (l- x)dxs=i j ( lx- t )

o
y = 2ËT{lX - T ) (3)
In particular, from formula (3) we déterminé the deflection h at the extre

mity of the girder L:
. P/3
1.18 SOME TYPES OF SECOND-ORDER DIFFERENTIAL

EQUATIONS REDUCIBLE TO FIRST-ORDER EQUATIONS.
ESCAPE-VELOCIT Y PROBLEM
I. An équation of the type

*!L
riyl = If (\ x*' dA
dx J ( 1)
doés not explicitly contain the unknown function y.

Solution. Let us dénoté the dérivative ày
— by p, that is, we set
% = p- Then S = i -
Putting these expressions of the dérivatives into équation (1),
we get a first-order équation,
in the unknown function p of a:. Integrating this équation, we

find its general solution:
p = p(x, C,)
and then from the relation ^ —p w'e get the complété intégral
of équation (1):
0 = $ P(x, C,) dx -f C2
Example 1. Let us consider the differential équation ot a catenary (see
Sec. 1. 1):
d2y _ 1 , / . ; dy V-
dx* a V ‘ \ dx /
60 Ch. J Differential Equations
Set
then
d2y _dp
dx2~~dx
and we get a first-order differential équation in the auxiliary function p of x :
ÿdx= -La VT+p*
dp _dx
whence
in { p + /T + 7 ) = -+ C t
p = s i n h ( 'z r + c ‘ )
But since p = ^ , the latter relation is a differential équation in the sought-
for function y. Integrating it, we obtain the équation of a catenary (see
Sec. 1. 1):
y = a cos h ^ - - 4 - C j^ + Ca
Let us find the particular solution that satisfies the following initial con
ditions:
yx=o =
yx=o = °
The first condition yields Ct = 0, the second, C j= 0 .
We finally obtain
y = ocos h
Note. We can similarly integrate the équation

ÿ<n>= / (x, /" -« )
Setting y in~v = p , we get for a détermination of p the first-
order équation
P)
From here we get p a s a function of x, and from the relation

ÿ(n-x)_p we gn(j y (see §ec 117).
II. A n éq u a tio n o f the t y p e
does n o t con tain the in d e p e n d e n t v a ria b le x e x p l i c i t l y .

1.18 Second-Order Equations Reducible to First-Order Equations 61
To solve it, we again set

dy
dx P ( 3)
but now we shaîl consider p as a function of y (and not of x,

as before). Then
cPy_d p __dp d y __dp
dx2 dx dy dx dy ^
Putting into (2) the expressions ^ and ^ , we get a first-

order équation in the auxiliary function p:
P % - f{ y ,p ) (4)
Integrating it, we find p as a function of y and the arbitrary
constant Ct:
p=p(y> Cx)
Substituting this value in (3), we get a first-order differential
équation for the function y of x :
l- p to .c .)

dy ■dx
p (y. c j)
Integrating this équation, we get the complété intégral of the
initial équation:
<D(x, y, Cx, C4) = 0
Example 2. Find the complété intégral of the équation
3yn = y 3
dy dp
Solution. Put P ~ -fc and consider p as a function of y. Then y" = p j î
and we get a first-order équation for the auxiliary function p:
Integrating this équation, we find
Pa= Ci —y or p = ± K c l - sr ,/»
_dy.
But p = — ; consequently, for a détermination of y we get the équation
dy y 1* dy
. = dx, or = dx
V c .- y 'l -
whence
__ y1/3 dy
x -)- C2
V C tft'-X
To compute the latter intégral we make the substitution
C,
Then
1
dy = 3t (** +
Consequently,
y u dy
V C , / '* - !
Final 1y we get
cî
I 3< (<»+D
t
* + C 2 = ± - L K c . ÿ ‘^ - 1
J F K c l!/,/>-l(C1ÿ*/’ + 2)
^1
( C l ÿ *'> + 2)
w
Example 3. Let a point move along the jc-axis under the action of a force
that dépends solely on the position of the point. The differential équation of
motion will be
d2x
m = F(x)
dt2
At t = 0 let x = x0, — = v0.
Multiplying both sides of the équation by — dt and integrating from 0 to t,

we hâve
x
The first term of this équation is the kinetic energy, the second term, the
potential energy of the moving point. From this équation it follows that the
sum of the kinetic and potential energies remains constant throughout the time
of motion.
The problem of a simple pendulum. Let there be a material point of mass mt
which is in motion (due to the force of gravity) along a circle L lying in the
vertical plane. Let us find the équation ot motion of the point neglecting résis
tance forces (friction, air résistance, etc.).
1.Î8 Second-Order Equations Reducible to First-Order Equations 63
Putting the origin at the lowest point of the circle, we put the *-axis along
the tangent to the circle (Fig. 22).
Dénoté bv / the radius of the circle, by s thç arc length from the origin O
to the variable point M where the mass m is located; this length is taken with
the appropriate sign (s > 0 , if the point M is on
the right of O; s < 0 if M is on the left of O). y
Our problem consists in establishing s as a function
of the time t.
Let us décomposé the force of gravity mg into
tangential and normal components. The former, equal
to —mgsin <p, produces motion, the latter is can-
A
celled by the reaction of the curve along which the
mass m is moving.
Thus, the équation of motion is of the form
d2s \
M (xÀ J>
m d t * ^ ~ mg Sinq> S ,
5 0
Since the angle qp= y f°r a circle, we get the équation *
d2s s
— S sm T Fig. 22
This is a Type II differential équation (since it does not contain the independent
variable t explicitly).
Let us integrate it in the appropriate fashion:
ds d2s
T r ~ p> dS
Hence,
or
dp__
P ds ■f
p dp — —
T*
whence
p2--=2gl COS y + CX
Let us dénoté by s0 the greatest arc length to which the point M swings.
For s = s0 the velocity of the point is zéro:
ds
= P = 0
It s = s0
This enables us to détermine C1:
0 = 2gt c o s ^ l+ C ,
whence
Ct = —2gl COS y -
Therefore,
fd s \* „ , ! s «o
— ' — 2g/ ' ™
64 Ch. î Different ial Equations
or, applying to the last expression the formula for the différence of cosines,
/ d s \ 2 . , . s + s 0 . s0— s
\ d t ) = 4& s in —^21 2 sin-°----
21 (5)
or *
j {= 2 Y g l ^ s i n s- ± j± sir ( 6)
21
This is an équation with variables separable. Separating the variables, we get
* =2 V g ld t (V
S “f - S a . S a -----S
Y s ,n ^ r sm V
We shall assume, for the time being, that s =é s0, then the denominator of the
î/
fraction is different from zéro. If we consider that s = 0 for /=^0, then from (7)
^ye get
* -= 2 Ÿ g ït (8)
s -f sfl . s0— s
sin - n r smJLw -
This is the équation that yields s as a function of t. The intégral on the left
cannot be expressed in terms of elementary functions; neither can the function
s of t. Let us consider this problem approximately. We shall assume that the
angles and y are small. The angles ~^ S° and S° ÿ S will not exceed-—
In (6) let us replace, approximately, the sines of the angles by the angles
—s
^dt= 2 Ÿ~gi î / ’ 21
r 8 V 2T
<6 '>
Separating variables, we get (assuming, for the time being, that s 7^ s0)
FT=T- Yï*‘
Again we consider that s = 0 when / = 0 . Integrating this équation, we get
(8 ')
.2 .2 f l
or
whence
(9)
We take the plus sign in front of the root. From the note at the end of
the solution it follows that there is no need to consider the case with the mi
nus sign.
1.18 Second-Order Equations Reducible ta First-Order Equations 65
Note. When solving, we assumed that s ^ s0» But it is clear,. by direct sub
stitution , that the fonction (9) is the solution of équation (6 ') for any value of t.
Let it be recalled that the solution (9) is an ^pproximate solution of équa
tion (5), since équation (6) was replaced by the approximate équation (6 ').
Equation (9) shows that the point M (which may be regarded as the extre-
mity of the pendulum) performs harmonie oscillations with a period T — 2n
This period is in dépendent of the amplitude s0.
Example 4. Escape-velocity probfem.
Détermine the smallest velocity with whfch a body mnst be thrown verti-
cally upwards so that it wi11 not return to the earth. Air resistar.ee is neglectedi.
Solution. Dénoté the mass of the earth and the mass of the body by M
and m respectively. By Newton’s Iaw of gravitation, the force of attraction f
acting on the body m is
where r is the distance between the centre of the earth and the centre of gra-
vity of the body, and k is the gravitational constant.
The differential équation of motion of this body with mass m will be
d2r M •m
md T '= - k ~ W
or
A*
~ * r2
The minus sign indicates that the accélération is négative. The differential
équation (10) is an équation of type (2). We shall solve it for the following
initial conditions:
for / = 0 r = R, ^=*V
Here, R is the radius of the earth and v0 is the launching velocity. We dénoté
dr_ d2r _dv _dv dr dv
d t ~ V> dt2~~dt~~dr " dt~~Vdr
where v is the velocity of motion. Putting this into (10), we get
dv M
k
dr r*
vdv = —kM —•
Intégrâting this équation, we find
v2 1
-= k M ^ C t («O
From the condition that v — v0 at the earth’s surface (for r = R)f we détermine Cx*
vô
66 Ch. 1 Differential Équations
We put the value of Cx into (11):
or
It is given that the body should move so that the velocity is always positive;
v2 kM
hence, > 0 . Since for a boundless increase of r the quantity — becomes
v2
arbitrarily small, the condition -y > 0 will be fulfilled for any r only for the
case
(13)
or
Hence, the lowest velocity will be .. .a n lined by the équation
(14)
where
k = 6 .66 - 1 0 - 8 cm3/g-sec2
R = 63-107 cm
At the earth’s surface, for r = /?, the accélération of gravity isg(g = 981 cm/sec2).
For this reason, from (10) we obtain
or
Putting this value of Af into (14), we obtain

y0= Y"2-981 -63-107«a 11.2*10* cm /sec= ll.2 km/sec
1.19 GRAPHICAL METHOD OP INTEGRATING

SECOND-ORDER DIFFERENTIAL EQUATIONS
t
Let us find out the géométrie meaning of a second-order differ-

ential équation. Suppose we hâve an équation
y" = f(x, y, y') (i)
Dénoté by <p the angle fprmed by the positive x-axis and the
tangent to the curve; then
( 2)
1.19 Graphical Method of Intégrâting Second-Order Equations 67
To find the géométrie meaning of the second dérivative, recall

the formula that détermines the radius of curvature of a curve at
a given point: * •
n . ( l + ÿ 'f '
y"
whence
y ~ R
But
y ' = tan<p, 1 +y'* = 1 -f-tan*<p = sec*<p,
1
(1 + y nYu = | sec3(p | | COS3 <p I
therefore
y R | cos3 cp | (3 )
Now putting into (1) the expressions obtained for y and y", we
hâve
T T \ l à ^ = f ( x ’ y ' tan<P)
or
| cos8 <p | •/ (*, y, tan<p) ' '
It is thus évident that a second-order differential équation déter
mines the magnitude of the radius of curvature of an intégral
curve if the coordinates of the point and the direction of the
tangent to this point are specified.
From the foregoing there follows a method of approximate con
struction of an integraLcurve by means of a smooth curve com-
posed of arcs of circles. **
To illustrate, let it be required to find the solution of équation
(1) that satisfies the following initial conditions:
yx—x0—yoi yx—xa—ya
Through the point M0(x0, y0) draw a ray M0Ta with slope
y' = tan<p„ = y'9 (Fig. 23). From équation (4) we find the value
R = R0. Lay off a segment MtC0, equal to R0, perpendicular
to Af0T0, and from the point C0 (as centre) strike an arc
* Up till now we hâve always considered the radius of curvature positive;
in this section we shall consider it a number that can take on both positive
and négative values: if the curve is convex (y" < 0 ), we consider the radius
of curvature négative (R < 0); if the curve is concave (y* > 0), it is positive
(R > 0).
** A curve is called smooth if it has tangents at ail points and the angle of
inclination of the tangent is a continuous function of the arc length s.
m Ch. 1 Differential Equations
with radius /?„. It should be noted that if /?„ < 0, then the seg
ment M„C0 must be drawn in that direction so that the arc of
the circle ïs convex upwards, and for R„ > 0, convex down (see
footnote on page 67).
Then let x lt yx be the coordinates of the point M x which lies
on the constructed arc and is sufficiently close to the point Af„
while tan q>j is the slope of the tan
gent M XTX to the circle drawn at M x.
From équation (4) we fînd the value
R — Rx that corresponds to Af1. Draw
the segment M XCX, perpendicular to
MXTX, equal to Rt, and from Cx (as
centre) strike an arc M^M2 with ra
dius R x. Then on this arc take a point
Af, (xt, ÿa) close to Mx and continue
construction as before until we get
a sufficiently large piece of the curve
consisting of the arcs of circles. From
the foregoing it is clearthat this curve
is approximately an intégral curve that
passes through the point M0. Obviously, the smaller the arcs
M^MX, AÇM2, . . . . the doser the constructed curve will be to
the intégral curve.
1.20 HOMOGENEOUS LINEAR EQUATIONS.

DEFINITIONS AND GENERAL PROPERTIES
Définition 1. An nth-order differential équation is called linear
if it is of the first degree in the unknown function y and its
dérivatives y', . . . . yin~li, y{a); that is, if it is of the form
a<ty{n) + a1y{n- 1’>+ . . . + any = f (x) (I)
where a0, au a2, . . . . a„ and f(x) are given functions of x or
constants, and 0 for ail values of x in the domain in which
we consider équation (1). From now on we shall présumé that the
functions a„, at, . . . , a„ and / (x) are continuous for ail values
of x and that the coefficient an= 1 (if it is not equal to 1 we
can divide ail terms of the équation by it). The function f (x) on
the right side of the équation is called the right-hand member of
the équation.
If f (x) ÿà 0, then the équation is called a nonhomogeneous linear
équation. But if /(*)==. 0, then the équation has the form
yM 4- a,tfn- " 4- . . . + a„y = 0 ( 2)
1.20 Homogeneous Linear Equations
and is called a homogeneous linear équation (the left side oî this

équation is a homogeneous function of the first degree in y, y',
Let us détermine some of the basic properties of homogeneous
linear équations, confining our proof to second-order équations.
Theorem 1. If y x and ty2 are two particular solutions of a homo
geneous linear équation of the second order
i f + a ^ ' + a2y ^ 0 (3)
then y, + y2 is also a solution of this équation.
Proof. Since yt and y2 are solutions of the équation, we hâve
yl + <*ty'i + ajti = 0 ]
and / (4)
yl + a1y'î + a2yt = 0 )
Putting into équation (3) the sum y 1-\-yi and taking into account
the identities (4), \ve will hâve
{Ui. + yù”+ ai (ÿi + yù' + a2(ÿj + y2)
= (y\ + + y'x + + + ) + (yl + asy‘2-f a2y2) = 0 + 0 = 0
Thus, y y-h y„ is a solution of the équation.
Theorem 2. If yx is a solution of équation (3) and C is a con
stant, then Cyt is also a solution of (3).
Proof. Substituting into (3) the expression Ci/,, we get
(C</,)" + a, (Ci/,)' + a2(Cy2) = C [y\ + aty[ + a2yt] = C- 0 = 0
and the theorem is thus proved.
D é fin itio n 2. The two solutions of équation (3), ÿ, and y2, are
called linearly independenfjm an interval [a, b] if their ratio on
this interval is not a constant; that is, if
—+=const
9t
Otherwise the solutions are called linearly dépendent. In other
words, two solutions, (/, and y2, are called linearly dépendent on
an interval [a, b] if there exists a c o n s ta n t number K such that
—= k when a ^ x s ^ b . In this case, yy —%ys.
y2
Example 1. Let there be an équation y " — y ~~0. It is easy to verify that the
functions ex , e~x t 3ex , 5e~x are solutions of this équation. Here, the functions
ex
and e~x are linearly independent on any interval because the ratio — - = e 2x
6~ x
does not remain constant as x varies. But the functions ex and 3ex are linearly
3ex
dépendent, since = 3 = const.
Définition 3. If y x and y2 are functions of x, the déterminant

Vi
W (yt, y2) =
y[ y2
is called the Wronskian of the given functions.
Theorem 3. // the functions y1 and y2 are linearly dépendent on
an interval [a, 6], then the Wronskian on this interval is identi-
cally zéro.
Indeed, if y2—^yi where A,= const, then y ^ h y 'i and
y i y* \ y y l-y i yi yi = o
W (ÿi. y t) =
U\ y 2 ! y'i W i y[ y\
Theorem 4. If the Wronskian W (ylt y2), formed for the solutions
yl atyl y2 of the homogeneous linear équation (3), is not zéro for
some value x - x0 on an interval [a, b] where the coefficients of the
équation are continuous, then it does not vanish for any value of x
whatsoever on that interval.
Proof. Since i/1 and y2 are two solutions of équation (3), we
hâve
y l + a^y* + a 2y 2 = 0
yl + a ^ + a ^ ^ O
Multiplying the terms of the first équation by ylt the terms of
the second équation by y2 and subtracting the latter from the
former, we get
(y tâ — é î l + û i W - y'iy2) = o (5)
The différence in the second parenthesis is the Wronskian
W ( y ly y 2), namely, W (</„ y 2) = ^ ( y ty 2— y'iy2). The différence in the
first parenthesis is the dérivative of the Wronskian:
W'x ( y lt y 2) = ( y ^ — y l y A ' = y ^ l — yly*
Hf ice, équation (5) takes the form

W '+ a .W ^ O (6)
LA us find the solution to the last équation that satisfies the
initial condition W\XssXn = W(s. We first find the general solution
to équation (6) on the assumption that W ^=0. Separating variab
les in (6), we get ^ r = —at dx.
Integrating we find
X
ln W = —J at dx+ InC
X.
1.20 Homogeneous Linear Equations 71
or
x
ln w = — r\ al dxm
whence
X
- \ a' dx
W = Ce *° (7)
It will be noted that we can write the function (7) and say
that this function satisfies équation (6), which is clear if we sub-
stitute this function directly into (6). The assumption 0 is
not required.
Formula (7) is called Liouville’s formula.
We define C so that the initial condition is satisfied. Substitut-
ing x-=;t0 into the left and right sides of (7), we get
r 0=c
Consequently, the solution that satisfies the initial conditions
takes the form
-Î-*
W = W0e x* (T)
By hypothesis, W0^=0. But then from (7') it follows that IF=^0
for any value of x, because the exponential function does not va-
nish for any finite value of the argument. The proof is complété.
Note I. If the Wronskian is zéro for some value x —x0, then
it is also zéro for any value x in the interval under considération.
This follows directly from (7): if 1F = 0 when x = x0, then
( r ) , =, . - c = o
consequently, flP=0, no matter what the value of the upper limit

of x in formula (7).
Theorem 5. If the solutions yl and y2 of équation (3) are linear-
ly independent on an interval [a, 6], then the Wronskian W,
formed for these solutions, does not vanish at any point of the given
interval.
Proof. To start with, note the following. The function y = 0 is
a solution of équation (3) on the interval [a, b] that satisfies the
initial conditions
yx=x. — üx—x. ~ ®
where x0 is any point in the interval [a, b]. From the existence
and uniqueness theorem (see Sec. 1.16), which is applicable to
72 Ch. î Different iat Equations
équation (3), it follows that there is no other solution to équa

tion (3) that satisfies the initial conditions
yX—X0= 0, ^JC=A0 ~~0
From this same theorem, it also follows that if the solution to
équation (3) is identically zéro on some closed interval or the
open interval (a, P) belonging to the interval [a, b], then this
solution is identically zéro over the whole interval [a,/;]. Indeed,
at the point x = p (and at the point x ~ a ) the solution satisfies
the initial conditions
ÿx=p = 0, y;_ 3 - 0
Hence, by the uniqueness theorem, it is also zéro on some inter
val p—d < x < P - j- d , where d is determined by the coefficients
of équation (3). Thus, extending the interval each time by the
amount d, where y ^ Q , we can prove that y = 0 over the entire
interval [a, bj.
Let us now begin the proof of Theorem 5. Assume that
W ( y y 2) = 0 at some point of the interval [û, b]. Then, by Theo
rem 3, the Wronskian W (yt, y2) will be zéro at ail points of
[a, b]:
ft7 = 0 or y xy \— y[yt = 0
Suppose that yx¥=0 on the interval [a, b], Then, by the last
équation, we can write
0 or f& Y -O
y\ \yiJ
whence it follows that
—= >„-= const
yi
that is, the solutions yx and y2 are linearly dépendent, which runs
counter to the assumption that they are linearly independent.
Suppose, furthermore, that yx — 0 at the points x,, x2......... xk
in [a, b]. We consider the interval (a, xx). On this interval, y ^ O .
Hence, by what has just been proved, it follows that on the in
terval (a, xx)
— — const or y2= kyx
We consider the function y = y .— kyx. Since y2 and yt are solu
tions of équation (3), then y==yî —ky1 is a solution of (3) and
y = 0 over the interval (a, x,). Thus, by the remark made at the
beginning of the proof it follows that y = y2—kyx^ 0 on the in
terval [a, b] or
1.20 Homogeneous Unear Equations 73
on the interval [a, 6], which is to say y.t and yx are linearly
dépendent. #
But this contradicts the assumption that the solutions y2 and
yY are linearly independent. We hâve thus proved that the Wrons-
kian does not vanish at any point of the interval [a, b\.
Theorem 6. If yx and y, are two linearly independent solutions
of équation (3), then
y ^ C iy i + C2y2 (8)
where Ct and C2 are arbitrary constants, is its general solution.
Proof. From Theorems 1 and 2 it follows that the function
t'ii/i ~t~
is a solution of équation (3) for any values of Cx and C2.
We shall now prove that no matter what the initial conditions
yx=x0 y'x- *0 :r= y'o» it is possible to choose the values of the ar
bitrary constants Cx and C2 so that the corresponding particular
solution C1yl + C2yl, should satisfy the given initial conditions.
Substituting the initial conditions into (8), we hâve
y<) ~ ^ i^/io 4“Cai/io g
I/o “ Cxyï0 -|- C2y2Q j
where we put
(yi)x~x0 ÿ10; {y?)x-r„ ~ y»o» (*/i)*=*0 i/io> (*/2)*=*0 -- i/2o
From .system (9) we can détermine Cx and C2, since the determi-
nant of this system
y io y 20
~~y loi/îio i/ioi/ao
ylo y»o
is the Wronskian for and, hence, is not equal to 0 (by vir-
tue of the linear independence of the solutions yx and y2). The
particular solution obtained from the family (8) for the found
values of C1 and C2 satisfies the given initial conditions. Thus,
the theorem is proved.
Example 2 . The équation
whose coefficients ax= - and = ----- ^ are continuous on any interval that
does not contain the point x —0 , admits the particular solutions
yi = *. y*= -x-
(Ihis is readily verified by substitution). Hence, its general solution is of the
form
y = C1* + C 2 —
Note 2. There are no general methods for finding (in closed

form) the general solution of a linear équation with variable coef
ficients. However, such a method exists for an équation with
constant coefficients. It will be given in the next section. For the
case of équations with variable coefficients, certain devices will
be given in Chapter 4 (Sériés) that will enable us to find appro-
ximate solutions satisfying definite initial conditions.
Here we shall prove a theorem that will enable us to find the
general solution of a second-order differential équation with vari
able coefficients if one of its particular solutions is knowh. Since
it is sometimes possible to find or guess one particular solution
directly, this theorem will prove useful in many cases.
Theorem 7. If we ktiow one particular solution of a second-order
homogeneous linear équation, the finding of the general solution
reduces to intégrât ing the functions.
Proof. Let yl be some known particular solution of the équation
y ' + a\9'+ ad} —o
We find another particular solution of the given équation so that
yx and yx are linearly independent. Then the general solution will
be expressed by the formula y —Cly l + Cjt/2, where Cx and Cs are
arbitrary constants. By virtue of formula (7) (see proof of Theo
rem 4), we can Write
Thus, for a détermination of yt we obtain a first-order linear

équation. Integrate it as follows. Divide ail terms by y\\
or
whence
Since we are seeking a particular solution, we get (by putting

C = 0 and C = 1)
( 10)
1.21 Second-Order Homogeneous Linear Equations 75
It is obvious that yx and y2 are linearly independent solutions

since —=^const.
y\
Thus, the general solution of the original équation is of the form
y = Clyl + C„yl <

\j - dx ( 11)
Example 3. Find the general solution of the équation

( \ — x2)y" — 2xy'+ 2 y = 0
Solution. It is évident, by direct vérification, that this équation has a par-
ticular solution y\ = x. Let us find a second particqlar solution y 2> so that yx
and y2 should be linearly independent.
_2x
Noting that in our case ai= -j— ~ » we hâve, by ( 10),
ç 2x dx
■In II -*«| r dx
■dx
<=x V - i ? - dx=xŸ-
1 1 1
=*H: x2 ' 2 (1 —x ) ‘ 2(l + x )
Consequently, the general solution is of the form
1 . \ +x
y = C1x+ C s[ y x l n
1.21 SECOND-ORDER HOMOGENEOUS LINEAR

EQUATIONS WITH CONSTANT COEFFICIENTS
We hâve a second-order homogeneous linear équation

y"+py' + Qy= o (i)
where p and q are real constants. To find the complété intégral
of this équation, it is sufficient (as has already been proved) to
find two linearly independent particuLar solutions.
Let us look for the particular solutions in the form
y —ekx, where k —const (2)
then
y' =kekx, y" = k2ekx
Substituting the expressions of the dérivatives into équation (1),
we find
ekx (£2 _|_ pk _). q} —o
Since e*x ^ 0 , this means that
k2+ pk + q = 0 (3)
76 Ch. 1 Différentiai Equations
Thus, if k satisfies équation (3), then ekx will be a solution

of (1). Equation (3) is called an auxiliary équation with respect
to équation (1).
The auxiliary équation is a quadratic équation with two roots;
let us dénoté them by k t and k 2. Then
*1 = - } + YT - * ’ k*=-T-Yf-ï
The following cases are possible:
I. and k 2 are real numbers and not equal (k 1^ k î ),
II. fe, and k 2 are complex numbers,
III. k, and k 2 are real and equal numbers ( k ^ k ^ .
Let us consider each case separately.
I. T he ro o ts o f th e a u x ilia ry é q u a tio n are real a n d d is tin c t,
k 1^ k î . Here, the particular solutions are the functions
# ,= £ * '* , y 3 =-<;***
These solutions are linearly independent because
— =
t/L e*.* ^
const
Hence, the complété intégral has the form
y H-
Example 1. Giveri the équation
y" + y' — 2y=r.O
The auxiliary équation is of the form
—2 = 0
We find the roots of the auxiliary équation:
= —y ± Y"4"“^2
^ = 1, 2
The complété intégral is
y = C,ex -j-Cte~2x
I I . T he ro o ts of th e a u x ilia ry é q u a tio n a re com plex. Since com
p le x roots are co n ju g a te in p airs, w e W rite
kl =a-\- ip, kt ■=a —«p
w h ere ______
a = —T* P=Yq~7
The particular solutions may be written in the form
yl = eia +if t x, = (4)
These are complex functions of a real argument that satisfy the
differential équation (1) (see Sec. 7.4, Vol. I).
It is obvious that if some complex function of a real argument

y --u (x ) + iv(x) • (5)
satisfies (1), then this équation is satisfied by the functions u (x)
and v(x).
Indeed, putting expression (5) into (1), we hâve
[u (x) + iv (x)]" + p[u (x) iv (x)]' + q[u (x) + iv (x)] = 0
or
( u * p u ’-\-qu) i (v" -f pv' -f- qv) ss 0
But a complex function is equal to zéro if, and only if, the real
part and the imaginary part are equal to zéro; that is,
u" + pu + qu = 0
v" + pv’ + qv —0
Thus we hâve proved that m(x) and v(x) are solutions of the
équation.
Let us rewrite the complex solutions (4) in the form of a sum
of the real part and the imaginary part:
y t —eP* cos p.x + ieP-x sin $x
y2 = eax cos px— reax sin px
From what has been proved, the particular solutions of (1) are
the real functions
ÿ, —eajt cos p* (6')
y2 —e** sin Px (6")
The functions yx and yt are linearly independent, since
y± eax cos Px
= cot Px const
y* eax sin Px
Consequently, the general solution of équation (1) in the case of
complex roots of the auxiliary équation is of the. form
y —Ciyl + C2ÿa —C,eax cos Px + C2eax sin px
or
y = eax (C, cos px + C2sin Px) (7)
where Cy and C2 are arbitrary constants.
An important particular case of the solution (7) is the case
where the roots of the auxiliary équation are pure imaginary.
This occurs when p = 0 in équation (1) and it has the forai
y"+ qy = 0
The auxiliary équation (3) assumes the form

** + <7 = 0 , <7 > 0
The roots of the auxiliary équation are
* i , 2 = ± iV q = ± t'P, a = 0
The solution (7) becomes
y —C1cos (bc + C2sin p*
Example 2. Qiven the équation
y m+ 2y' + 5y = 0
Find the complété intégral and a particular solution that satisfies the initial
conditions y x =o = Q* y'x= o = l - Construct the graph.
Solution. (1) We Write the auxiliary équation
k* + 2 k + 5 = 0
and find its roots:
= — 1+ 2/, /?2 —— 1 — 2i
Thus, the complété intégral is
y = e~x (Ci cos 2* + C2 sin 2x)
(2) We find a particular solution that satisfies the given initial conditions
and détermine the corresponding values of C1 and Ca.
From the first condition we find
0 = e -° (Cj cos 2 0 + C2 sin 2 0), whence C ^ O
Noting that
y' = e~x 2C2 cos 2x — e~x C2 sin 2x
we obtain from the second condition
• = 2 C2, Le., C2= i
Thus, the desired particular solution is
y = Y e~x sin2*
Its graph is shown in Fig. 24.
y ”+ 9y = 0
Find the general solution and a particular solution satisfying the initial con
ditions
y x —Q— y x=o = 3
Solution. Write the auxiliary équation
*2 + 9 = 0
We find its roots to be
k 1= 3/\ kty = —3t
The general solution is
y = Ci cos 3jc+C 2 sin 3x
Let us find a particular solution. First we find

y' = — 3CXsin 3 * + 3C2 çps 3x
The constants Cx and C2 are defined from the initial conditions
0 = Cx cos 0 + C2 sin 0
3 = — 3CXsin 0 + 3C2 cos 0
They are
C i — 0, C2= 1
The particular solution is
y = sin 3x
III. The roots of the auxiliary équation are real and equal. Here,
kx= k2.
One particular solution, y1= ek*x, is obtained from earlier reason-
ing. We must find a second particular solution, which is linearly
independent of the first (the function ek*x is identically equal to
ek*x and therefore cannot be regarded as a second particular solu
tion).
We shall seek the second particular solution in the form
y2= u (x) ek*x
where u(x) is the unknown function to be determined.
Differentiating, we find
y2= u'ek*x + kxuek*x = ek*x (u! -f kxu)
y2= + 2klu,ek*x + k\uek'x = ek*x (u" + 2kxu' +
Putting the expressions of the dérivatives into (1), we obtain
ektx [u" + (2kt + p)u' + (k\ + pkx+ q) u] = 0
Since kx is a multiple root of the auxiliary équation, we hâve
k\ + pk1+ q = 0
78 Ch. î Dijferential Equations
The auxiliary équation (3) assumes the form

k* + q = 0, q > 0
The roots of the auxiliary équation are
î,2 = ± iV <7= ± *P» a = 0
The solution (7) becomes
y —C, cos p* -(- C2sin px
y” -\-2y' + 5y = o
Find the complété intégral and a particular solution that satisfies the initial
conditions y x - 0 = 0 , (/'* _ o = l. Construct the graph.
Solution. (1) We Write the auxiliary équation
*2 + 2 * + 5 = 0
and find its roots:
ki = — 1+ 2i, t?2 = — 1—2i
Thus, the complété intégral is
y = e~x (C1 cos 2* + C2 sin 2x)
(2) We find a particular solution that satisfies the given initial conditions
and détermine the corresponding values of Cl and Ca.
From the first condition we find
0 = e -° (Cj cos 2 0 + C2 sin 2 0), whence C ^ O
Noting that
y' = e~x 2C2 cos 2x — e~x C2 sin 2x
we obtain from the second condition
1 = 2C2, i. e., C2— —
Thus, the desired particular solution is
y — ~21 e- *x sin
• O2x
Its graph is shown in Fig. 24.

y" + 9y = 0
Find the general solution and a particular solution satisfying the initial con
ditions
yx=o= y x=o ~ 3
A2+ 9 = 0
We find its roots to be
k i = 3i, k2 = —3i
y = Cl cos 3jc + C2 sin 3x
Let us find a particular solution. First we find

y' = — 3CXsin 3* + 3C2 çps 3*
The constants Cx and C2 are defined from the initial conditions
0 = (?! cos 0 + C2 sin 0
3 = — 3Cj sin 0 + 3C2 cos 0
They are
C2- 0, C2 = 1
The particular solution is
y = sin 3*
III. The roots of the auxiliary équation are real and equal. Here,
kx — k2.
One particular solution, yx = ek^x, is obtained from earlier reason-
ing. We must find a second particular solution, which is linearly
independent of the first (the function ek*x is identically equal to
ek*x and therefore cannot be regarded as a second particular solu
tion).
We shall seek the second particular solution in the form
y2= u (x) ek'x
where u(x) is the unknown function to be determined.
Differentiating, we find
y2= u’ek*x + = ek*x (u' + kxu)
y2- + 2klufek^x + k\uek'x = ek*x (u" + 2kxu' + k\u)
Putting the expressions of the dérivatives into (1), we obtain
ek'x [u" + (2kx+ p)u' + (k\ + pkx+ q) u] = 0
Since kl is a multiple root of the auxiliary équation, we hâve
k\ + pk1+ q = Q
Besides, kl = k2= ~ — or 2k1= — p, 2k1+ p ~ 0 .

Hence, in order to find u(x) we must solve the équation
= 0 or u" = 0. Integrating, we get u = Ax + B. In particular,
we can set A = 1 and B =-- 0; then
U — X
Thus, for the second particular solution we can take

y2= xe^x
This solution is linearly independent of the first, since —= .v ^ const.
yi
Therefore, the following function is the general solution:
y = Cxek'x+ C2xefctx = eM (C1+ Ctx)
y" — Au' —0
Write the auxiliary équation kz — 4k-{-4 = 0. Find its roots: kx = k2 — 2.
The complété intégral is then
y^Ctftx + Ctxe2*
1.22 HOMOGENEOUS LINEAR EQUATIONS OF THE NTH ORDER WITH

CONSTANT COEFFICIENTS
Let us consider a homogeneous linear équation of the /ith order:

^ + ^ ^ - * > + . . . + 0^ = 0 ( 1)
We shall assume that a,, a2, . an are constants. Before giving

a method for solving équation (1). we introduce a définition that
will be needed later on.
D é fin itio n I . If for ali x of the interval [a, b] we hâve the
equality
<P„ (*) = î<Px(*)■+ A&t (x) + • ■• + An- l<fn_i (x)
where A t, A2, . . . , An are constants, not ail equal to zéro, then we
say that <p„(.v) is expressed linearly in tenus of the functions
« P iO ). ? * ( * ) . •...
2. n functions q>2(jc), (p2(x), . . . .
D é fin itio n (jp„(v) are
called linearly independent if not one of the functions is expressed
linearly in ternis of the rest.
N o te 1. From the définitions it follows that if the functions
<pt (x), <p.,(x), . . . , <p„(x) are linearly dépendent, there will be con
stants Cy, C2, . . . . Cn, not ail zeroes, such that for ail x in the
interval [a, b] the following identity will hold true:
Ci<Pi (x) + Csqp.j (.«) -F . . . + Cnip„ (x) = 0
1.22 Homogeneous Linear Equations of the ni h Order 81
Examples:
1. The functions y 1= ex , y2 = e2x, y 9= 3ex are linearly dépendent, since for
C1 = l , C2 = 0, Ca= — we hâve the identity
CXex + Cte2x + C93ex ^ 0
2 . The functions yi = l, y 2 = x, y3 — x2 are linearly independent, since the
expression
Cil -\~C2X-\-C9x2
will not be identically zéro for any Clt C2, C3 that are not simultaneously
equal to zéro.
8 . The functions y l =ektXy y2=ek*x, . . . , y n ~ e knx, . . . , where k it k2, . . . , kn, . . .
are different numbers, are linearly independent. (This assertion is given without
p roof.)
Let us now solve équation (1). For this équation» the following
theorem holds.
Theorem. If the functions yx, yt, . . yn are linearly independent
solutions of équation (1), then its general solution is
y = Ci!h + C2y2+ • ■• + CnlJn (2)
where Cly . . Cn are arbitrary constants.
If the coefficients of équation (I) are constant, the general so
lution is found in the same way as in the case of second-order
équations.
(1) Form the auxiliary équation
kn + a1kn~1+ a 2kn- 2+ . . . + an = 0.
(2) Find the roots of the auxiliary équation
• • •» hn
(3) From the character of the roots Write out the particular
linearly independent solutions, taking note of the fact that:
(a) to every real root k of order one there corresponds a parti
cular solution ekx\
(b) to every pair of complex conjugate roots &(1) —a + fp and
k{2)=â— t’P there correspond two particular solutions e^cospx
and ea*sinpx;
(c) to every real root k of multiplicity r there correspond r
linearly independent particular solutions
ekx, xekx, . . . , xr~'ekx\
(d) to each pair of complex conjugate roots &(1) = oc-ftp,
k(1) = a — /p of mult iplicity \i there correspond 2\i particular so
lutions:
eax cos p*, xeax cos p*, . . . , x*- 1eax cos Px,
eax sin Px, xeax sin Px, . . . , x*- leax sin px
82 Ch. 1 Differentiai Equations
The number of these particular solutions is exactly equal to the

degree of the auxiliary équation (that is, to the order of the given
linear difïerential équation). It may be proved that these solutions
are linearly independent.
(4) After finding n linearly independent particular solutions
ylt ÿî. •••. yn we construct the general solution of the given
linear équation:
y — Cil/i + ^2^2 + • • • + C„yn
where Clt C2, . . Cn are arbitrary constants.
y,v - y = 0
Solution. Form the auxiliary équation
tr1—1=0
Find the roots of the auxiliary équation:
Arx = 1. k 2 = — 1, k-3 = i, ft4= — i
Write the complété intégral
y= ^3 cos x "["Cj sin x
where C1( C2, Cs, C4 are arbitrary constants.
Note 2. From the foregoing it follows that the whole difficulty

in solving homogeneous linear differential équations with constant
coefficients lies in the solution of the auxiliary équation.
1.23 NONHOMOGENEOUS SECOND-ORDER LINEAR EQUATIONS
Let there be a nonhomogeneous second-order linear équation

y" + a1y' + aty = f(x) (1)
The structure of the general solution of such an équation is
determined by the following theorem.
Theorem 1. The general solution of the nonhomogeneous équation
(1) is represented as the sum of some particular solution of the
équation y* and the general solution y of the corresponding homo
geneous équation
y"+ aiy' + a2y = 0 (2)
Proof. We need to prove that the sum
y = ÿ+y* (3)
is the general solution of équation (1). Let us first prove that the
function (3) is a solution of (1).
1.23 Nonhomogeneous Second-Order Linear Equations 83
Substituting the sum y + y* into (1) in place of y, we get

(ÿ + y*)"+ «i (y+ y*Y + a2(ÿ+y*) = / (*)
or
(y" + aCy' + ad) + (y*”+ a,y*' + a2y*) = f (x) (4)
Since y is a solution of (2), the expression in the first brackets
is identically zéro. Since y* is a solution of (1), the expression
in the second brackets is equal to f(x). Consequently, (4) is an
identity. Thus, the first part of the theorem is proved.
We shall now prove that expression (3) is the general solution
of équation (1); in other words, we shall prove that the arbitrary
constants that enter into the expression may be chosen so that
the following initial conditions are satisfied:
yx=x„ — t/o ^
v*-* = V. I (5)
no matter what the numbers x0, y0 and y'0 [provided that x0 is
taken from the région where the functions alt a2 and f(x) are
continuous].
Noting that ~ÿ may be given in the form
V = Ciy l + C ji>
where yx and y2 are linearly independent solutions of équation (2),
and Cx and C2 are arbitrary constants, we can rewrite (3) in the
form
y ^ C ^ + C ^ + y* (3')
Then, by the conditions (5), we will hâve*
Cîo + C2y20+ yl = ya
Cii/io + £îf/2o + # î — y'à
From this System of équations we hâve to détermine C4 and Ct .
Rewriting the System in the form
Cii/io = y0 y0 ^
Ciyio~yC2y to — y(> y0 j
we note that the déterminant of this System is the Wronskian
for the functions yt and yt at the point x = x0. Since it is given
that these functions are linearly independent, the Wronskian is
not zéro; consequently, System (6) has a definite solution,
Here, y10, y2». yl. y'io, y’io. yl dénoté the numerical values of the func-
and C2; in other words, there exist values Cx and C2 such that
formula (3) defines the solution of équation (1) which satisfies
the given initial conditions. The theorem is completely proved.
Thus, if we know the general solution y of the homogeneous
équation (2), the basic difficulty, when integrating the nonhomo-
geneous équation (1), lies in finding some particular solution y*.
We shall give a general method for finding the particular so
lutions of a nonhomogeneous équation.
The method of variation of arbitrary constants (paranteters).
We write the general solution of the homogeneous équation (2):
y = Cxyx + C2y2 (7)
We shall seek a particular solution of the nonhomogeneous
équation (1) in the form (7), considering Cx and C2 as some (as
yet) undeterm ined functions of x.
Differentiate (7):
y' = Cxyx+ C2y2'1- C1y1T C2y2
Now choose the needed functions Cx and Ct so that the following
équation holds:
C%yi “HC2t/2= 0 (8)
If we take note of this additional condition, the first dérivative y'
will take the form
y —Cxyx-J- C2y2
Diiferentiating this expression, we find y":
y "— Cxyx + C2y2-f- Cjÿi 4- C2y2
Putting y, y' and y" into (1), we get
Cxyx-f- C2y2-{- Cxyx-f- C2y 2-f- ctl (Cxyx -f- C2y2) -|- ct2(Cxyx-f- C2y2) = f (.v)
or
C x (lJ \ + û iÿi + fl2ÿi) + C 2 (y*. + axy2+ a y ù + Cxyx-f- C„y2— f (x)
2
The expressions in the first two brackets vanish, since yx and y2

are solutions of the homogeneous équation. Hence, the latter équa
tion takes on the form
C[y'x + C’ly 2= f(x) (9)
Thus, the function (7) will be a solution of the nonhomogeneous
équation (1) provided the functions Cx and C2 satisfy the System
of équations (8) and (9); that is, if
C'xyx+ C2y2= 0, C[y[ + C’^ = f (x)
Since the déterminant of this system is the Wronskian for the
linearly independent solutions yx and y2 of équation (2), it is not
1.23 Nonliomogeneous Second-Order Linear Equations 85
eq u a l to zéro. H e n c e , in s o iv in g th e S ystem w e w ill fin d C\ and C'2

as d e f in it e fu n c tio n s of x:
CJ = Ti(*). C; = <?,(*)
Intégrâting, we obtain
Ci =$?!(*)<& + Cn, C ,=
where Ct and Cs are constants of intégration.
Substituting the expressions obtained for C, and C2 into (7),
we find an intégral that is dépendent on the two arbitrary con
stants Cj and Ct; that is, we find the general solution of the
nonhomogeneous équation. *
Example. Find the general solution of the équation
Solution. Let us find the general solution of the homogeneous équation
*y’ — —
X
=0
Since
u” —
1 we hâve ln y' = ln *-|-ln C, y' = Cx
y x
and so
y = Clx2-\-C%
For the latter expression to be a solution of the given équation, we hâve
to define Cx and C2 as functions of x from the System
CiX2-f- C2 • 1 = 0 , 2Cjx -J- C2 •0 = x
Soiving this System, we find
whence, after intégration, we get

C i = — + Clt C2= — g- + Q
Putting the functions obtained into the formula ^ = Cpcï + Ca, we get the
general solution of the nonhomogeneous équation:
— — Jf® x®
i r - c 1*«+ci + Ç - ^
_ — jf3 —
or y — C1x2 + Ca + “5- , where Cx and C2 are arbitrary constants.
ü
When seeking particular solutions, it is useful to take advan-
tage of the results of the following theorem.
* If we put C i= C, = 0 , we get a particular solution of équation ( 1).
86 Ch. 1 Different ial Equations
Theorem 2. The solution y• of the équation

y" + aiU' + = U (x) + fi (x) ( 10)
where the right member is a sum of two functions f1(x) and f ï (x), *
may be represented as the sum y* = y\-\-y\, where y[ and y\ are,
respectively, the solutions of the équations
y ï + adiï+ < m \ = fi(x) (11)
y ï + aly\'+ a2yl = /, (x) ( 12)
Proof. Adding the right and left members of (11) and (12), we
get
(y l+ y d " + a t (yl + y*i)' + û2{y*-\-y\) = /iW + /»W (13)
From this équation it follows that the sum
y l + y ‘i = y
is a solution of équation (10).
Example. Find the particular solution y • of the équation
y n— 4y = x + 3ex
Solution. The particular solution to the équation
yi +4*/i = x
is
The particular solution to the équation
*/2 + 4(/2 = 3e*

is
* 3 *
»*=-.Ke
The particular solution y* of this équation is
. 1 , 3 *
^ = T X+T ^
1.24 NONHOMOGENEOUS SECOND-ORDER LINEAR

EQUATIONS WITH CONSTANT COEFFICIENTS
Suppose we hâve the équation

y"+ py' + y y=f ( x ) (i)
where p and q are real numbers.
* It is clear that the theorem holds true for any number of terms in the
right member of the équation.
1.24 Nonhomogeneous Second-Order Linear Equations 87
A general method for finding the solution of a nonhomogeneous

équation was given in the preceding section. In the case of an
équation with constant coefficients, it is sometimes easier to find
a particular solution without resorting fo intégration. Let usconsi-
der several such possibilities for équation (1).
I. Let the right side of (1) be the product of an exponential
function by a polynomial; that is, of the form
f (*) = P n (x)eaX (2 )
where P„(x) is a polynomial of degree n. Then the following par
ticular cases are possible:
(a) The number a is not a root of the auxiliary équation
ki + pk + q = 0
In this case, the particular solution must be sought for in the
form
y* = (A0xn + A^x"-1+ • • • + An) e** = Qn (x) e** (3)
Indeed, substituting y* into équation (1) and cancelling e** out
of ail terms, we will hâve
Qn (x) + (2a + p) Qn (x) + (a* + pa + q) Qn (x) = Pn (x) (4)
Q„(x) is a polynomial of degree n, Q'n(x) is a polynomial of de
gree n — 1, and Q"n(x) is a polynomial of degree n — 2. Thus,
n-degree polynomials are found on the left and right of the equa-
lity sign. Equating the coefficients of the same degrees of x (the
number of unknown coefficients is n + 1), we get a system of n-(- 1
équations for determining the unknown coefficients A0, Alt
A2, • • •• An.
(b) The number a is a simple (single) root of the auxiliary équa
tion.
If in this case we should seek the particular solution in the
form (3), then on the left side of (4) we would hâve a polyno
mial of degree n— l.since the coefficient of Q„(x), that is,a 2+/?a-|-<7,
is equal to zéro, and the polynomials Q'„(x) and Q"n(x) hâve deg
rees less than n. Hence, (4) would not be an identity, no matter
what A0, Alt . . . , A „ . For this reason, the particular solution
in this case has to be taken in the form of a polynomial of degree
n + l , but without the absolute term (since the absolute term of
this polynomial vanishes upon différentiation):*
y* = xQ„ (x)e«
* We remark that ail the results given above also hold for the case where a
is a complex number (this follows from the rules for differentiating the func
tion emx, where m is any complex number; see Sec. 7.4, Vol. I).
(c) The number a is a double root of the auxiliary équation.

Then, as a resuit of the substitution of the function Qn (xje** into
the differential équation, the degree of the polynomial is diminished
by two units. Indeed, if a is a root of the auxiliary équation,
then at + pa-\-ç = 0; moreover, since a is a double root, it follows
that 2a = —p (since by a familiar theorem of elementary algebra,
the sum of the roots of a reduced quadratic équation is equal to
the coefficient of the unknown to the first power with signrever-
sed). And so 2a + p = 0.
Consequently, on the left side of (4) there remains Q’„(x), that
is, a polynomial of degree n —2. To obtain a polynomial of deg-
ree n as a resuit of substitution, one should seek the particular
solution in the form of a product of e** by the (n + 2)th degree
polynomial. Then the absolute term of this polynomial and the
first-degree term will vanish upon différentiation; for this reason,
they need not be included in the particular solution.
Thus, when a is a double root of the auxiliary équation, the
particular solution may be taken in the form
y = x*Qa{x)e«
Exemple 1. Find the general solution of the équation
y " + 4 y '+ $ y = x
Solution. The general solution of the corresponding homogeneous équation is
~ÿ = C1e - x + C2e~*x
Since the right-hand side of the given nonhomogeneous équation is of the
form xe9x [that is, of the form P i(x)eox], and 0 is not a root of the auxiliary
équation ki -\-4k-\-3 = 0, it follows that we should seek the particular solution
in the form y* = Qi(x)e°x ; in other words, we put
y * = A 0x-Sr A 1
Substituting this expression into the given équation, we will hâve
4Aq-]-3(AqX-\-Ai) = x
Equating the coefficients of identical powers of x, we get
3i4j = 1, 4 AQ- j - 3 Ai — 0
whence
Consequently,
The general solution of y = y+y* will be

y = Cle-* + C8e - » * + l x - i -
Î.24 Nonhomogeneous Second-Order Linear Equations 89

yf,+ 9 y = (x2+\)<?x
Solution. The general solution of the homogenÊous équation is readily found:
y = Ci cos 3jc+ Ca sin 3x
The right side of the given équation (x2+ l ) e 8jf has the form
Pa (x) (**
Since the coefficient 3 in the exponent is not a root of the auxiliary équa
tion, we seek the particular solution in the form
y* = Q 2 (x)e?x or y* = (4*2+ 3 * + C ) e?x
Substituting this expression in the differential équation, we hâve
f9 (Ax2 + 3* + C) + 6 (24* + 3) + 24 + 9 (Àx2 + 3* + C)] é*x = (x2+ \)ë*x
Cancelling out eAx and equating the coefficients of identical powers of x , we
obtain
184 = 1, 124 + 183 = 0, 2 4 + 6 3 + 18C= 1
1 „ 5
whence 4 = yg, 3 = —^ , C = ^y. Consequently the particular solution is
27 81
and the general solution is

y = C1c o s3 * + C , sin 3 * + g j)* * x
Example 3. To solve the équation
y”- 7 y '+ 6 y = (x— 2) ex
Solution. Here, the right side is of the form Px (x) elx and the coefficient 1 in
the exponent is a simple root of the auxiliary polynomial. Hence, we seek the
particular solution in the form y* = xQl (x)ex or
y* = x ( 4 * + 3 ) ex
Putting this expression in the équation, we get
[ ( A x 2 + B x ) + (4A x + 23) + 24 - 7 (4*2 + B x ) —7 (24* + 3)
+ 6 (4*2+ 3*)] ex = (x — 2) ex
or
( — 104*—53 + 24) ex = (*—2) ex
Equating the coefficients of identical powers of *, we get
- 1 0 4 = 1, —53 + 2 4 = - 2
1 9
whence 4 = —yg , 3 = ^ ? . Consequently, the particular solution is
ÿ*==* ( — eX
II. Let the right side hâve the form

f (x) —P (x) e*x cos Px + Q (x) e** s in px (5)
where P (x) and Q (x) are polynomials.
We may handle this case by the technique used in I, if we pass
from trigonométrie functions to exponential functions. Replacing
cospx and sinpx by exponential functions via Euler’s formulas
(see Sec. 7.5, Vol. I), we obtain
gWt —g-W*
f(x) — P (x) eax 2
Q (x) eax 21
or
/ W = [ j P W + i Q (*)] «<*+‘Wx+ e<a" fp' x <6)
Here, the square brackets contain polynomials whose degrees are
equal to the highest degree of the polynomials ^(x) and Q (x).
We hâve thus obtained the right side of the form considered in
Case I.
It can be proved (we omit the proof) that it is possible to find
particular solutions which do not contain complex numbers.
Thus, if the right side of équation (1) is of the form
f (x) — P (x) eax cos px -f- Q (x) eax s in px (7)
where P (x) and Q (x) are polynomials in x, then the form of the
particular solution is determined as follows:
(a) if the number a + /p is not a root of the auxiliary équation,
then the particular solution of équation ( 1) should be sought in
the form
y* = U (x)ea*c.ospx-fV, (x)e0t*sin px (8 )
where U (x) and V (x) are polynomials of degree equal to the highest
degree of the polynomials P (x) and Q (x);
(b) if the number a-J-ip is a root of the auxiliary équation,
we then Write the particular solution in the form
y* —x [U (x) eax cos px + V (x) ea*sin Px] (9)
Here, in order to avoid mistakes we must note that these forms
of particular solutions, (8 ) and (9), are obviously retained when
one of the polynomials P (x) and Q (x) on the right side of équa
tion ( 1) is identically zéro; that is, when the right side is of the
form
P (x) eax cos Px or Q (x) eax s in px
Let us further consider an important spécial case. Let the right
side of a second-order linear équation hâve the form
f(x) = M cos px -(- N sin Px {T)
where M and N are constants.
Î.24 Nonhomoqeneous Second-Order Linear Equations 91
(a) If pi is not a root of the auxiliary équation, the particular

solution should be sought in the form
ÿ* = A cos p* + B sin px (8 ')
(b) If pi is a root of the auxiliary équation, then the particular
solution should be sought in the form
y? = x (A cos Pjc + B sin px) (9')
We remark that the function (7') is a spécial case of the func-
tion (7) [P(x) = M, Q(x) = N, a = 0]; the functions (8 ') and (9')
are spécial cases of the functions (8 ) and (9).
Example 4. Find the general solution of the nonhomogeneous linear équa
tion
y " -\- 2 y ' -{ -5 y = 2 cos x
Solution. The auxiliary équation k2+ 2k + 5 = 0 has roots k1= —l+ 2 i;

k2 —— 1 —2t. Therefore, the general solution of the corresponding homogeneous
équation is
y = e~ x (Cj cos 2 x + C2 sin 2x)
We seek the particular solution of the nonhomogeneous équation in the form

y* = A cos x -\- B sin x
where A and B are constant coefficients to be determined.
Putting y* into the given équation, we hâve
— ^4 cos x —5 sin x + 2 (— -4 sin x + B cos x ) + 5 (A cos x - \ - B sin x) = 2 cos x
Equating the coefficients of cos x and sin jc, we get two équations for deter-
mining A and B :
— 4 + 25 + 54 = 2, — 5 - 2 4 + 55 = 0
2 1 —
whence 4 = -g-, 5 = -g-. The general solution of the given équation is y = y + y*t
that is,
2 1
y = e~* (Cj cos 2jc+ C2 sin 2jc)+-g- cos x + -g- sin x
Example 5. Solve the équation

y" + 4y = cos 2 x
Solution. The auxiliary équation has roots k 1= 2i, k2= —2/; therefore, the
general solution of the homogeneous équation is of the form
y = Cj cos 2 * + C2 sin 2x

y* = x (4 cos 2 x + 5 sin 2jc)
Then
y•' = 2x (— 4 sin 2x + 5 cos 2jc) + (4 cos 2x + 5 sin 2x)
/ " = - 4 x (— A cos 2x — B sin 2x) + 4 (— 4 sin 2* + 5 cos 2x)
92 Ch, 1 Differential Equations
Putting these expressions of the dérivatives into the given équation and
equating the coefficients of cos 2x and sin 2x, we get a System of équations
for determining 4 and B:
4 3 = 1 , —44 = 0
whence 4 = 0 and 3 = -^-. Thus, the general solution of the given équation is
y = C1 cos 2 x + C 2 sin sin 2x

Example %. Solve the équation
y”— y — 3e2x cos x
Solution. The right side of the équation has the form
/ (x) = e2* (M cos x + N sin x)
and M = 3, N = 0. The auxiliary équation fc2— 1 = 0 has roots k 1= \, k2= —1.
The general solution of the homogeneous équation is
ÿ = C1e * + C 2e~*
Since the number a + i‘P = 2 + M is not a root of the auxiliary équation, we
seek the particular solution in the form
y*=ze?x (A cos x + 3 sin x)
Putting this expression into the équation, we get (after collecting like ternis)
(24 + 4B) e2x cos x + (—44 + 2B) e2x sin x = 3e2x cos x
Equating the coefficients of cosx and sinx, we obtain
24 + 4fl = 3, —44 + 23 = 0
3 3
whence 4 = jg and 3 = -g-. Consequently, the particular solution is
y* = e2x ^ c o s x + - | - s i n x ^

y = C1ex + C2e~x + e2x ^ cos x + - |- sinx^
Note. Ail the arguments of this section hold true also for a
linear équation of the first order. To illustrate, let us consider
a first-order équation with constant coefficients (this équation is
frequently encountered in engineering applications):
% + ay = b (io)
where a and b are constants. We find the general solution of the
homogeneous équation
% + ay = 0
Form the auxiliary équation
£ + a = 0 , k = —a
1.25 Hlgher-Order Nonhomogeneous Linear Equations 93

Ce~ax .
We seek a particular solution y* of the nonhomogeneous équation
in the form
!T = B
Substituting into équation (10), we get
aB —b, B = b/a
Th us
y*—b!a
The general solution of (10) will then be
y=--y + ÿ* or y = Ce~ax + b/a ( 11 )
1.25 HIGHER-ORDER NONHOMOGENEOUS LINEAR EQUATIONS

Let us consider the équation
yM + aly<'‘- » + . . . + a ny = f(x) ( 1)
where at, a.......... a,„ f (jc) are continuous functions of x (or con
stants).
Suppose we know the general solution
y = Ciy\ + Cjt/j + . . . + Cnyn (2 )
of the corresponding homogeneous équation
yM + aiy " -» + a2y'n- 2'+ . . . + a ny = Q (3)
As in the case of a second-order équation, the following asser
tion holds for équation ( 1).
Theorem. I f y is the general solution of the homogeneous équa
tion (3) and y* is a particular solution of the nonhomogeneous
équation ( 1), then
Y = y + f
is the general solution of the nonhomogeneous équation.

Thus, the problem of integrating équation (1), as in the case
of a second-order équation, reduces to finding a particular solution
of the nonhomogeneous équation.
As in the case of a second-order équation, the particular solution
of équation ( 1) may be found by the method of variation of para-
meters, considering Cx, C2, . . . , Cn in expression (2 ) as functions
of x.
We form the system of équations (cf. Sec. 1.23):

CiHi + Ciüi + • • • + Cnyri = 0
Cxyx+ C2y2+ . • • + C'nyn = 0
(4)
c ; < - 2>+ « - * > + . . . + « - « = o
C[y[n~" + +... + = /(* ) j
This system of équations in the unknown functions C’x, C2, . . . ,
C'a has very definite solutions. (Thç déterminant of the coef
ficients of C'x, C2......... C'n is the Wronskian formed for the parti-
cular solutions yx, y2, . . . . yn of a homogeneous équation, and
since these particular solutions are, by définition, linearly inde-
pendent, the Wronskian is not zéro.)
Thus, the system (4) may be solved for the functions C’x, C2, . . . ,
C'n. Finding them and integrating, we obtain
C2 = ^ Cxdx -f- C1( C2 = J C2 dx -J- C2, . . . , Cn = J Cn dx -j- Cn
where Cj, C2, . . . , C„ are the constants of intégration.
We shall prove that in such a case the expression
y* = C1yl + Ciy2+ . . . + C nyn (5 )
is the general solution of the nonhomogeneous équation ( 1).
Differentiate expression (5) n times, each time taking into account
équations (4); this yields
ÿ = Ciyx + C2ÿ2 + . . . + C j f n
y*’ ~ ^ xy x + C2i/2 + . . . -\-C ny'n
y in~« = + c 2*/<n- i> + . . . + c ^ r #

y*™ = Cxy r + C 2yïtn) + . . . + C n<> + /(*)
Multiplying the terms of the first, second, . . . and, finally, last
équation by an, a„_1, . . . , alf and 1 respectively, and adding,
we get
y ^ + axy ^ + . . . + a ny = f(x)
since yx, yt, . . . , y„ are particular solutions of the homogeneous
équation; for this reason, the sums of the terms obtained in adding
vertical columns are equal to zéro.
Hence, the function y = Cxyx+ . . . + Cny„ [where Cx......... C„
are functions of x determined from équations (4 )] is a solution of
the nonhomogeneous équation (|), ajid since _this solution dépends
on the n arbitrary constants Cx, C2, . . . . C„, it is the general
solution. This can be proved, as in the case of the second-order
équation.
1.25 Higher-Order Nonhomogeneous Linear Equations 95
The proposition is thus proved.

For the case of a higher-order nonhomogeneous équation with
constant coefficients (cf. Sec. 1.24), the particular solutions are found
more easily, namely:
I. Let there be a function on the right side of the differential
équation: f (x) = P (x) eax, where P (x) is a polynomial in x; then
we hâve to distinguish two cases:
(a) if a is not a root of the auxiliary équation, then the parti
cular solution may be sought in the form
y*= Q (x)ea*
where Q (x) is a polynomial of the same degree as P (x), but with
undetermined coefficients;
(b) if a is a root of multiplicity p. of the auxiliary équation,
then the particular solution of the nonhomogeneous équation may
be sought in the form
y* —x^Q (x) eP-x
where Q (x) is a polynomial of the same degree as P (x).
II. Let the right side of the équation hâve the form
/ (x) —M cos Px-f Wsin Px
where M and N are constants. Then the form of the particular
solution will be determined as follows:
(a) if the number Pt is not a root of the auxiliary équation,
then the particular solution has the form
y* — A cos Px -f- fl sin px
where A and fl are constant undetermined coefficients;
(b) if the number Pt is a root of the auxiliary équation of mul
tiplicity p, then
y* = xv-(A cos px + fl sin px)
III. Let
/ (x) = P (x) eax cos Px + Q (*) ea*sin Px
where P (x) and Q (x) are polynomials in x. Then:
(a) if the number a + Pt is not a root of the auxiliary polyno
mial, then we seek the particular solution in the form
y* — U (x) e®*cos Px -f F (x) eax sin px
where U (x) and V (x) are polynomials of degree equal to the
highest degree of the polynomials P(x) and Q (x);
(b) if the number a + pt is a root of multiplicity p, of the auxi
liary polynomial, then we seek the particular solution in the form
y* = x^ [U (x) eax cos px + V (x) eax sin Px]
where U (x) and V (x) hâve the same meaning as in Case (a).
General remarks on Cases 11 and III. Even when the right side
of the équation contains an expression with only cosp* or only
sin^jt, we must seek the solution in the form indicated, that is,
with sine and cosine. In other words, from the fact that the right
side does not contain cos|fo or smp*, it does not in the least
follow that the particular solution of the équation does not contain
these functions. This was évident when we considered Examples 4,
5, 6 of the preceding section, and also Example 2 of the présent
section.
yiv -.y==x3+ i
Solution. The auxiliary équation k* — 1 = 0 has the roots
fci = 1, fc2 ——1, k3= i, k 4 = — i
We find the general solution of the homogeneous équation (see Example 4,
Sec. 1.22):
y = C1e* + C2e~* + Cs cos* + C4 sin x
y* = Aç>x?+ A xx2+ ,4a*+ A3
Differentiating y* four times and substituting the expressions obtained into
the given équation, we get
— A0x? — A xx2— A 2x — A 3~ x®+ 1
Equating the coefficients of identical powers of x , we hâve
— A 0= \ t — Ai — 0, — A 2= Q, — A3— \
Hence
y* = — x3— 1
Thejjeneral solution of the nonhomogeneous équation is found from the formula
y — y -{-y * '.
y — Ctex -j-C2e -x + C3 cos x + C4 sin x — x3— 1
Example 2. Solve the équation
y iv — y - 5 cos x
Solution. The auxiliary équation k x— 1 = 0 has the roots — 1, k2 — — 1,
k3= i , k 4= — t. Hence, the general solution of the corresponding homogeneous
équation is _
y — Ciex + C tf-* + C3 cos * + C4 sin x
Further, the right side of the given nonhomogeneous équation has the form
f (x) = M cos x + N sin x
where Af = 5 and N — 0.
Since i is a simple root of the auxiliary équation, we seek the particular
solution in the form
y* = x(A cos x + B sin x)
Putting this expression into the équation, we find
4A sin x — 4B cos x = 5 cos x
1.26 The Differential Equation of Mechanical Vibrations 97
whence
4/1 = 0, —4Æ= 5
5
or j4 = 0 , B — . Consequently, the particular solution of the differential
•
équation is
.
u — ---- 5
- x sin*
4
1.26 THE DIFFERENTIAL EQUATION

OF MECHANICAL VIBRATIONS
In this and the following sections we will consider a problem
in applied mechanics and will investigate and solve it by means
of linear differential équations.
Let a load of mass Q be at rest on an elastic spring (Fig. 25).
We dénoté by y the déviation of the load from the equilibrium
position. We shall consider déviation downwards as positive,
upwards as négative. In the equilibrium position, the weight is
balanced by the elasticity of the spring. Let us suppose that
the force that tends to return the load to equilibrium (the so-called
restoring force) is proportional to the deflection, that is, equal
to —ky, where k is some constant for the given spring (the
so-called “spring rigidity”). *
Let us suppose that the motion of the load Q is restricted by
a résistance force operating in a direction opposite to that of
motion and proportional to the velocity of the load relative to
the lower point of the spring; that is, a force — At>= — Aj^,
where A= const > 0 (shock absorber). Write the differential équa
tion of the motion of the load on the spring. By Newton’s second
law we hâve
(1)
(here, k and A are positive numbers). We thus hâve a homoge-

neous linear differential équation of the second order with con
stant coefficients.
This équation may be rewritten as follows:
(H
* Springs whose restoring force is proportional to the deflection are called

springs with a “linear characteristic”.
7—2082
where
Let it further be assumed that the lower point of the spring A

executes vertical motions under the law z = cp(t). This will occur,
for instance, if the lower end of the spring is attached to a rol-
ler, which moves over an uneven spot together with the spring
and the load (Fig. 26).
In this case the restoring force will be equal not to — kyy but
to —k[y + ty(t)]> f°rce °f résistance will be — X[yf + q>' (/)].
and in place of équation ( 1) we will hâve the équation
Q ^ r + *-% + ky = — k<p(t) — W ( t) (2 )
or
- g - + /> -§ -+ W = /( 0 (2')
where
We thus hâve a nonhomogeneous second-order differential

équation.
Equation (1 ')is called an équation of free oscillations, équation (2 ')
is an équation of forced oscillations.
1.27 FREE OSCILLATIONS

Let us first consider the équation of free oscillations
y'' + py' + qy = 0 (p > 0 , q > 0 ; see Sec. 1.26) (1)
We Write the corresponding auxiliary équation
k? + pk + q = 0
1.27 Free Oscillations 99
and find its roots:
*.— 7 + / f 3 ?. *. = - ! • - Y T - i
p2
(1) Let Then the roots kt and k2 are real négative num-
bers. The general solution is expressed in terms of exponential
functions:
y = Cxe*‘f + C2eV (k, < 0 , k2 < 0 ) (2 )
From this formula it follows that the déviation of y for any
initial conditions approaches zéro asymptotically if t —►<». In the
given case, there will be no oscillations, since the forces of résist
ance are great compared to the coefficient of rigidity of the
spring k.
(2 ) Let y = <7; then the roots kt and k2 are equal ând are
also equal to the négative number — ■£-) . Therefore, the general
solution will be
y = C1e ^ t + CJe~^* =(C , + (3)

Here the déviation also approaches zéro as t —>-oo, but not so
rapidly as in the preceding case (due to the factor Cl + C2t).
(3) Let p = 0 (no résistance). Equation (1) is of the form
y”+ q y = o (4)
The auxiliary équation is k2+ q = 0 and its roots are fcx= p/,
k2= — pi, where p = V q- The general solution is
ÿ --C l co sp i-fC 2sinpi (5)
In the latter formula, we replace the arbitrary constants Cx
and C2 with others. We introduce the constants A and <p0, which
are connected with Ct and C2 by the relations
Cx= /lsinq)0, C2 = j4cos<p0
A and <p# are defined as follows in terms of Cx and C2:
A = V C \ + Cl, (p0 = a rc ta n &
Substituting the values of Cx and C2 into formula (5), we get
y = A sin q>„ cos fit + A cos q>0 sin pi
or
«/ = ^4 sin (Pi + <p0) ( 6)
7*
These oscillations are called harmonie. The intégral curves are

sine curVes. The time interval T, during which the argument of
the sine varies by 2 n, is called the period of oscillation; here,
T= . The frequency is the
1Jt Hsin(fit*qp) number of oscillations during
✓ 'T-'V X - % 2x-<p,
p y "
time 2 n; here, the frequency
- is P; A is the greatest dévia
%i VS / T tion from equilibrium and
fi
^ ------------- T-------------- is called the amplitude-, <p0 is
Fig. 27 the initial phase. The graph
of the function (6 ) is shown
in Fig. 27.
In electrical engineering and elsewhere, wide use is made of
complex and vectorial représentations of harmonie oscillations.
Let us take the complex plane xOy and consider a radius vector
A = A{t) of constant length \A | = A = const.
As the parameter t varies (here, t is the time), the terminus
of the vector A describes a circle of radius A centred at the coor-
dinate origin (Fig. 28). Suppose the angle i|) formed by the vector A
and the x-axis is expressed thus: i|j = pf-f q>0. Then P is called the
angular velocity of the vector A. The
projections of A on the y- and x-axes y 1 < <2>
are
(x,y)
y = 4 sin (p/ + q> 0) \
Po) /i \
x = A cos (p/ + <Po) p0 J (7) i / / •» \ \
( y â s l A
The expressions (7) are the solutions I
of équation (4). l 0
X J X
We consider the complex quantity
z = x + iy =
= A cos (P* + <p0) + iA sin (P* + qp0)
or
Fig. 28
z = A [cos (P* + <p0) + i sin (pf + <p0)] (8 )
As was pointed out in Sec. 7.1, Vol. I, the complex quantity z

in (8 ) is represented by the vector A.
Thus, the solutions to the équation of harmonie oscillations (4)
may be regarded as the projections of the vector A on the y- and
x-axes, the vector with initiât phase <p0 rotating with angular ve
locity p.
Using the Euler formula [see (4), Sec. 7.5, Vol. IJ, the expres
sion (8 ) may be rewritten as
z = A e t ®t+và (9)
1.27 Free Oscillations 101
The imaginary and real parts of (9) are solutions of équation (4).
Expression (9) is called the complex solution of (4). Wecanrewrite (9)
as follows:
2 = A ë^éV * ( 10 )
The expression Ae‘f> is called the complex amplitude. We dénoté
it by A*. Then the complex solution (10) can be rewritten as
z = A'eW ( 11 )
(4) Let p ^ 0 and Ç < q.
In this case, the roots of the auxiliary équation are complex

numbers:
&! = <*+ t'P, kt = a —t'P
where
a = — y < 0. P= Y ~ Ï~ T
The general solution has the form

y = ea< (Cj cos pf + Ct sin p/) (12)
or
y = /le01' sin (P* + <p0) (13)
Here, for the amplitude we hâve to consider the quantity Aeal

which dépends on the time. Since a < 0, it approaches zéro as
t —<-oo, which means that here we are dealing with damped oscil
lations. The graph of damped oscillations is shown in Fig. 29.
1.28 FORCED OSCILLATIONS

The équation of forced oscillations has the form
•f + py'+ <iy = f (t) ( p > 0 , <7 > 0; see Sec. 1.26) ( 1)
Let us consider an important practical case when the disturb-
ing extemal force is periodic and varies under the law
/(*) = asin at
then the équation will hâve the form
t f + py' + qy = a$m<ùt (T)
p2
(1) Let us first présumé that p^=0 and ^ < q , that is, the
roots of the auxiliary équation are the complex numbers a ± ip.
In this case [see formulas (12) and (13), Sec. 1.27], the general
solution of the homogeneous équation has the form
ÿ = 4 e a'sin(p/ + <p0) (2 )
We seek a particular solution of the nonhomogeneous équation
in the form
ÿ — M cos ait + N sin <at (3)
Putting this expression of y* into the original differential équa
tion, we find the values of M and N:
—p<ûa (g—(Ù2) g
M (q—û)2)a+ pho2 ’ (q—co2)2 + p2wa
Before putting these values of M and N into (3), let us introduce
the new constants A* and <p*, setting
M = A* sin qp*, N — A* cos «p*
that is
a
a*= Vm î + n î tan(p* = ~
Y (g— coa)2+ p2ü)2
Then the particular solution of the nonhomogeneous équation may

be written in the form
y* = A*sin <p*cos cof -f- A* cos <p*sin a>( = A*sin (co< + qp*)
or, finally,
ÿ , = 7 i ? = S ¥ W si" (“ ( + 'p' )
The general solution of équation ( 1) is y = y + y* or
y = ^4e“'sin (p/ 4- qp0) sin ((ùt - f qp*)
y (g—uY + P 2®2
1.28 Forced Oscillations 103
The first term of the sum on the right side (the solution of the
homogeneous équation) represents damped oscillations; it dimini-
shes with increasing t and, consequently, after some interval of
time the second term (which détermines the forced oscillations)
will dominate. The frequency co of these oscillations is equal to
the frequency of the external force /(/); the amplitude of the forced
vibrations is the greater, the smaller p and the doser w2 is to q.
Let us investigate more closely the dependence of the amplitude
of forced vibrations on the frequency co for various values of p.
For this, we dénoté the amplitude of forced vibrations by D(«):
^ ^ V (q —<û2)2-f-p2o)2
Putting 9 = (for p = 0, p, would be equal to its natural fre
quency), we hâve
a
D( o)) =
V '( P Î - û .2)2 + p2(02
Introducing the notation
where X is the ratio of the frequency of the disturbing force to

the frequency of free oscillations of the System, and the constant y
is independent of the disturbing force, we find that the magnitude
of the amplitude will be expressed by the formula
(4)
D ^ p î y a - x 2)2 + v 2* 2
Let us find the maximum of this function. It will obviously be

for that value of X for which the denominator has a minimum.
But the minimum of the function
1/(1 —X2)2 + Y2X2 (5)
is reached when
X= J / T = !
and is equal to
t / Ï = ?
Hence, the maximum amplitude is equal to
The graphs of the function D (A.) for various values of y are shown
in Fig. 30 (in constructing the graphs we put a= 1, Px= l for
the sake of definiteness). These curves are called résonance curves.
From formula (5) it follows that for small y the maximum
value of amplitude is attained for values of A close to unity, that
is, when the frequency of the external force is close to the fre-
quency of free oscillations. If y = 0 (hence, /? = 0), that is, if there
is no résistance to motion, the amplitude of forced vibrations
increases without bound as A—>1 or as co—
lim D (A) = oo
x»—
►î
(Y=0)
At « 8 = 9 we hâve résonance.
(2) Now let us suppose that p — 0; that is, we consider the
équation of elastic oscillations without résistance but with a
periodic external force:
y" + ?ÿ = asin ait ( 6)
1.28 Forced Oscillations 105

y —Cl cos pt + C2sin p/ (Pa = <7)
•
If p=^co, that is, if the frequency of the external force is not
equal to the natural frequency, then the particular solution of
the nonhomogeneous équation will
hâve the form
y* = M cos (ùt + N sin (ot
Putting this expression into the
original équation, we find
M = 0, N-- 7 ---- fl)2

y = A sin (P* + %) + — ^5 sin tot
Thus, motion results from the super

position of a natural oscillation with
frequency P and a forced vibration
with frequency to.
If P = co, that is, the natural fre
quency coïncides with the frequency
of the external force, then function (3)
is not a solution of équation (6 ). In Fig. 31
this case, in accord with the results
of Sec. 1.24, we hâve to seek the particular solution in the form
y* = t(M cos fl( -f- N sin p/) (7)
Substituting this expression into the équation, we find M and N:
M=~ k ' ^ = 0
Consequently,
y* = — âp t cos p t
The general solution will hâve the form
y = A sin (P* + <P0) —^ t cos p*
The second term on the right side shows that in this case the
amplitude increases without bound with the time t. This pheno-
menon, which occurs when the natural frequency of the System
coïncides with the frequency of the external force, is called
résonance.
The graph of the function y* is shown in Fig. 31.
1.29 SYSTEMS OF ORDINARY DIFFERENTIAL EQUATIONS
In the solution of many problems it is required to find the

functions yl = y1(x)> yi = y î(x)......... yn = ynix)y which satisfy a
system of differential équations containing the argument x, the
unknown functions ylt y2, yn and their dérivatives.
Consider the following system of lîrst-order équations:
TE* = M *. Vu h i ...... y «
)
dx ~ h (x < t/i> ÿa> •••> yn) , /j.
f n(X> ÿl> y^y •

•••Vn)
where ylt y2, . . y„ are unknown functions and x is the argument.

A system of this kind, where the left sides of the équations
contain first-order dérivatives, while the right sides do not contain
dérivatives, is called normal.
To integrate the system means to détermine the functions
yu y* which satisfy the system of équations ( 1) and the
given initial conditions:
{ y ù x= x a ÿ io> (yî)x=x0 ^ 20* •••> ( yn)x=x0 ym (2)
Intégration of a system like (1) is performed as follows. Differen-

tiate the first équation of ( 1) with respect to x:
<Py\ _ à h , àfi dyi . . dfj dyn
dx2 dx ' dyi dx ' ' * * ' dyn dx
Replacing the dérivatives ..., with their expressions

fiy fty • • • . / „ from équations ( 1), we get- the équation
§r=M *. y , ....... yn)
Differentiating this équation and then doing as before, we obtain
dx~i ~ ^ 3 (x > V f y%y • • •> I/n)
Continuing in the same fashion, we finally get the équation
g i = Fn(x, y v . . . . yn)
1.29 Systems of Ordinary Differential Equations___________ 107
We thus get th e fo llo w in g System:

ï = M * . y» • • • » Un)
•
^ r = F A x > y » •• • . y n) ► (3)
d ny \ _f ( y u
fan ^ n \ x 9 y 1» *
From the first « — 1 équations we détermine (if this is possible)

y2, <y.i. • • •. Un and express them in terms of x, yx and the deri-
,. dyl d?yx dn~ 1y i
vat,ves ir > i d ’ d x"-1
y 2 = <p3 (x , y t . y'u ■ • y r( l)) )
*/3=<P3 (*. ÿ l . y[< • • y[n- u ) > (4)
y n = <t>n(x> yi< y ’x. • • y (r u) ,

Putting these expressions into the last of the équations (3), we
get an nth-order équation for determining yY\
d”yi = 0 >(x, y u y ’u . . . . jff-»)
dxn (5)
Solving this équation, we find «/,:

y i = 'f i (x > Ci» •••> c«) (6 )
Differentiating the latter expression n — 1 times, we find the
erivatives , ..., as functions of x, C„ C„ . . . . C„.
Substituting these functions into équations (4), we détermine
yt, y3, ■■■, yn-
Ui —^2 (■*> Clf C2, .
(7)
»« = ♦..(*. c„ c2, . •.,c„),
For this solution to satisfy the given initial conditions (2), it
remains for u's to find [from équations (6 ) and (7)] the appropriate
values of the constants C,. C2, . Cn (like we did in the case
of a single differential équation).
Note 1. If the System (1) is linear in the unknown functions,
then équation (5) is also linear.
Example 1. Integrate the System
dy dz
t e = y + z + x , —= - 4 y - 3 2 + 2 * (a)
dx
with the initial conditions
(y)x-o—i» (z)x—o—o (b)
Solution. (1) Differentiating the first équation with respect to x t we hâve

&v__dy , dz
dx* d x 'd x '
Putting the expressions and ^ from équations (a) into this équation,
we get
j g = ( y + z + x ) + ( - 4 y - 3 z + 2x) + l
or
g = - 3 y - 2 2 + 3x+l (c)
(2) From the first équation of System (a) we find
and put it into the équation just obtained; we get
or
S + 2E + » - 5‘ + ' w
The general solution of this équation is
y = (Cl + Cîx )e -* + 5x—9 (f)
and from (d) we hâve
z = (C2— 2Ci — 2C2x) e - * — 6 jc + 14 (g)
Choosing the constants Cx and C2 so that the initial conditions (b) are
satisfied,
(y)x=o= i » (z)x=o= o
we get, from équations (f) and (g),
1= C i —9, 0 = Ca—2CX+ 14
whence C x=10 and C2 = 6.

Thus, the solution that satisfies the given initial conditions (b) has the form
y = (10 + 6*) e~* + 5x—9, z = (—14— 12*) e~x — 6x-{-14
Note 2. In the foregoing we assumed that from the first n — 1

équations of the System (3) it is possible to détermine the functions
ÿ2, y s* •••» yn• It may happen that the variables y2y . yn are
eliminated not from n, but from a smaller number of équations.
Then to détermine yx we will hâve an équation of order less than n.
Example 2. Integrate the System
1.29 Systems of Ordinary Differential Equations 109
Solution. Differentiating the first équation with respect to t, we find

d1x dy . dz
5 F = '3?+5F=(jc+ 2) + lJt+i')
d2x .
- t.= 2 x + y + z
Eliminating the variables y and z from the équations

dx . d2x n . .
d i= y+ z' a tï= 2x+y+z
we get a second-order équation in x :
dt2 dt
Integrating this équation, we obtain its general solution:
x = C1e~t + C2e2t (a )
Whence we find
^jjf = — C1e - t + 2C2e2t and y = ~ —z = — C1e - t -{-2C»e2t— z (P)

Putting into the third of the given équations the expressions that hâve been
found for x and y t we get an équation for determining z:
— +z = Z C ^
Integrating this équation, we find
z = C3e-<-\-C1e*t (V)
But then, from équation (P), we get
y = —(^î + Cs) e~t -{-C2e2t (6)
Equations (a), (y), and (6) give the general solution of the given System.
The differential équations of a System may contain higher-ordei
dérivatives. This then yields a System of differential équations of
higher order.
For instance, the problem of the motion of a material point under the action
of a force F reduces to a System of three second-order differential équations.
Let Fx, F», Fz be the projections of the force F on the coordinate axes. The
position o r the point at any instant of time t is determined by its coordinates x,
(/, and z. Hence, x, y , z are functions of t. The projections of the velocity veç-
dx dy dz
tor of the point on the axes will
be H t ' dt ’ dt
Suppose that the force F and, hence, its projections FXt Fy, F? dépend on
the time /, the coordinates x, y , z of the point, and on the velocity ot motion
dx du dz
of the point, that is, on .
In this problem the following three functions are the sought-for functions:
x = x(t), y = y ( t) t z = z (t)
These fonctions aredetermined from équations of dynamics (Newton’s law>:

d2x / dx dz \
dy
Xy y t Zy
II
Pi.1
('• Ht 9 dt 9 d tJ
d2y dx dy dz \
*■>
Xy y t Zy
y
( 8)
%
('■ dt 9 dt ’ dt J
dx dy dz \
Xy y y Zy
dt2 ~ F* 1('■ Ht 9 dt * dt J
We thus hâve a System of three second-order differential équations. In the case
of plane motion, that is, motion in which the trajectory is a plane curve (lying,
for example, in the x^-plane), we get a System of two équations for determin-
ing the fonctions x (t) and y (t):
mS = F* { i ’ x > w ) <9>
m^ Fy ( f’ x>y>-W' w ) (,0)
It is possible to solve a System of differential équations of higher order by
reducing it to a System of first-order équations. Using équations (9) and (10) as
examples, we shall show how this is done. We introduce the notation
dy
dt
Then
d2x du cPy__ dv
dt*~~ H T 9 dt2 H t
The system of two second-order équations (9) and (10) in two unknown fonc
tions *(/) and y (t) is replaced by a system of four first-order équations in four
unknown fonctions x y y t u f v:
dy_
v
dt
= x, y, u, t-)
dv
X, y f u, v)
Ht
We remark in conclusion that the general method that we hâve considered of
solving the system may, in certain spécifie cases, be replaced by some artificial
technique that gels the resuit faster.
Example 3. Find the general solution of the following system of differential
équations:
d2z
dx2 = y
1.30 Systems of Linear Differential Equations 111
Solution. Differentiate both sides of the first équation twice with respect
to *:
d4y _ d2z
lW~~dx* m
d?z
But -j-£ = y, and so we get a fourth-order équation:
d4y
dx4 = y
Integrating this équation, we obtain its general solution (see Sec. 1.22,
Example 4):
y = C1ex + C2e~ x -\-C3 cos jc+ C 4 sin x
d2u
Finding — from this équation and putting it into the first équation, we find z:
z = C}ex -\-C2e - x —C3 cos x —C4 sin x
1.30 SYSTEMS OF LINEAR D1FFERENT1AL EQUATIONS

WITH CONSTANT COEFFICIENTS
Suppose we hâve the following System of differential équations:
= aux i + + • • • +alnxn
^7 = 1*1 + a , ^ t + . . . + aïnxn
at > m
-jf = a»A + a»^* + • • • + annxn

j
where the coefficients au are constants. Here, t is the argument,

and xt (t), x2 (t), . . . , xn (t) are the unknown functions. The System ( 1)
is a system of homogeneous linear differential équations with con
stant coefficients.
As was pointed out in the preceding section, this system may
be solved by reducing it to a single équation of order n, which
in the given instance will be linear (this was indicated in Note 1
of the preceding section). But system (1) may be solved in
another way, without reducing it to an équation of the nth order.
This method makes it possible to analyze the character of the
solutions more clearly.
We seek a particular solution of the system in the following
form:
x1= a 1ekt, x2 = a^*f, . . . , x n = aneki (2 )
It is required to détermine the constants a,, a 2, . . . , a„ and k
in such a way that the functions a,?**, a 2e*'..........anekt should
satisfy the system of équations (1). Putting them into System (1),
we get:
Aaxe*f = (a11a 1+ a12a 2+ . . . + a lna „ ) e kt
k a 2ekt = (a2ia i + ü22a2+ . . . + a 2na n) ekt
kdne* = (anla, + an2a 2+ . . . + annan) ekt

Cancel out eki. Transposing ail terms to one side and collecting
coefficients of a 1( a 2, . . . , a„, we get a system of équations:
(“n — A)«i + ûn a 2 + • • • + dlna„ = 0
ati«i + — k) a2+ . . . + a2nan = 0
(3)
ania i + an2a 2+ . . . + (a„n—Æ)an = 0

Choose Oj, a 2, . . . , a„ and k such that will satisfy the system (3).
This is a system of linear algebraic équations in a lt a 2, an.
Let us form the déterminant of system (3):
an k d\1 , . . ûln
ü22 ^ ■• • ^2 n
A(*) = (4)
U ni dn2• • • ann— k
If k is such that the déterminant A is different from zéro, then
system (3) has only trivial solutions ctj = a . = . . . = a„ = 0 and,
hence, formulas (2 ) yield only trivial solutions:
*i (t) = xt (t)= . . . = x „ ( 0 = 0
Thus, we obtain nontrivial solutions (2) only for k such that
the déterminant (4) vanishes. We arrive at an équation of order n
for determining k :
an k a \2 • ^1 n
d22 k .. • @2 n
= 0 (5)
a n\ a n2 •• • ann— k
This équation is called the auxiliary équation of the system ( 1),
and its roots are the roots of the auxiliary équation.
Let us consider a few cases.
I. The roots of the auxiliary équation are real and distinct.
Dénoté by k2, kn the roots of the auxiliary équation. For
each root kt write the system (3) and détermine the coefficients
a 1 » y*2</>» ' » 'An

1.30 Systems of Linear Differential Equations 1.13
It may be shown that one of them is arbitrary; it may be consi

dérée! equal to unity. Thus we obtain:
for the root kt the following solution of the System ( 1)
xJ1’ = a[l)ekit, x(2u = 1= ’e**'
for the root ka the solution of the system ( 1)
n = a[t)ek*‘, jei*) = c4 î)e*«, > x ^ = aj,2)efc*(
for the root kn the solution of the system ( 1)

x[n) = a[n)eknt, x2n) = <x[n)eknt, . . . , xj?’ = a (nn)eknl
By direct substitution into the équations we see that the system of
functions
xl = Cla[l)ek'‘ -f Cja^’c4»' + • • • + Cna[n)ek*t \
x3= C1a(2ue>!'1+ Cjc42)e**' + Cna2n)ek*t I
f (6 )
x„ - + C tainî)ek‘‘ -f . . . -f C„a<1' V »' )
where Clt Cit . . . , C„ are arbitrary constants, is likewise a solution
of the system of differential équations (1). This is the general
solution of system (1). It may readily be shown that one can find
values of the constants such that the solution will satisfy the
given initial conditions.
Exaniple 1. Find the general solution of the System of équations
-— - = 2 x i + 2 x 2 , —j j j = * i + 3*2

2 —k 2
1 3 —k
=0
or k2— 5 * + 4 = 0. Find its roots:
£1 = 1, *2 = 4
Seek the solution of the system in the form
and
v <2)
*i : =«{««"', 4» = oi, , e“
Form the system (3) for the root k x — 1 and détermine a}1’ and c4u :
(2—1) a î l) + 2a*,) = 0 \
l«xl> + (3—1) a21) = 0 j
or
a J1’ + 2aj1>= 0
a ! l)4 - 2 a t,, ) = 0
8—2082
114 Ch, 1 Differential Equations
whence = — j a i(1)- Putting a{1)= l , we getoc2X) = — — Thus, we obtain

the solution of the System:
xi« = e*. 4 u = - - « f
Now form System (3) for the root *a = 4 and détermine a{2) and ctâ2):
- 2 a l 2,+ 2 a ^ = 0
cc{2, -< 4 2, = 0
whence a i2) = a j2) and a i2) = l, a22>= l . We obtain the second solution of the
System:
*x2)= e4f, *j2)=e4t
The general solution of the System will be [see (6)]
X\ =
X t— — — C x e f - |- C 2e 4 f
II. The roots of the auxiliary équation are distinct, but indude
complex roots. Among the roots of the auxiliary équation let there
be two complex conjugate roots:
/j1= a+t'P, k î = a —ip
To these roots will correspond the solutions
x}l) = a '/1 1 ( / = 1 , 2 , . . . . n) (7)
x}2>= aj2) * (/ = 1, 2, . . . , n) (8)
The coefficients a ‘n and aJ-2) are determined from the System of
équations (3).
Just as in Sec. 1.21, it may be shown that the real and imagi-
nary parts of the complex solution are also solutions. We thus
obtain two particular solutions:
x<D — e *t (Aj.‘>cos px + Aÿ2) sin Px) |
(9)
x f — (A.J" sin px -(- A)2>cos px) j
where A)1' , Xÿ\ A.)1’, A'/* are real numbers determined in terms of
and a f ).
Appropriate combinations of functions (9) will enter iitto the
general solution of the system.
Example 2. Find the general solution of the System
dx i
~ d f= :— ^Xl *2
d*2 ___ 9 x ___ C j »

dt - 1 5*2
1.30 Systems of Linear Differential Equations 115

—7 — k 1
—2 —5—
or + 12/5 + 37 = 0 and find its roots:
k j = — 6-J-t, &2==— 6 — i
Substituting k t = — 6 + i into the System (3), we find
ai1’= 1, aà** = 1+ i
We write the solution (7):
4 1>= le(-«H/)*, 4 1) = (l+ /)e (-« + ft* <7')
Putting fc2 = — 6 — i into System (3), we find
a i« = l, «<’> = ! - *
We get a second System of solutions (8):
= 4 2, =(1—«)e(- ,-,)t (8 ')
Rewrite the solution (7'):
= (cos / + / sin /)
jc21) = (1 i) e ~ et (cos / + i sin /)
or
= cos / + fe“ 8* sin /
x™ = e -* f (cos / — sin /) + i> -8* (cos / + sin /)
Rewrite the solution (8'):
xli } = e“9t cos t — i e - 9t sin /
*22) = e~ot (cos / — sin t) — ie~*i (cos / + sin /)
For Systems of particular solutions we can take the real parts and the imagi-
nary parts separately:
x[l) = e“et cos /, x21) = e” 8*(cos / — sin /) 1 ^
xj2) = e - 8f s i n /, *22) = e ~ st (cos / + sin /) f
The general solution of the System is
x1= C1e~6t cos / -j-C2e~et sin /
^2 = C1g“ 6^ (cos / — sin /) + C2e~8* (cos / + sin /)
By a similar method it is possible to find the solution of a System
of linear differential équations of higher order with constant coef
ficients.
For instance, in mechanics and electric-circuit theory a study
is made of the solution of a System of second-order differential
équations:
- ^ 2 = a n x + a12t/
( 10)
d2u
^j/îT = #21^ “f" ^22y
8
116 Ch. 1 Difterential Equations
Again we seek the solution in the form

x = aekt, y = fiekt
Putting these expressions into System (10) and cancelling out ekt,
we get a system of équations for determining a, p and k :
(a»—*s)cc + a12p = 0 1 nn
021®+(ûîs—*2)P = 0 / { ’
Nonzero a and p are determined only when the déterminant of
the system is equal to zéro:
an — k2 a12
( 12)
&21 ^22
This is the auxiliary équation of system (10); it is a fourth-order

équation in k. Let k lf k 2, k 3, and k x be its roots (we assume that
the roots are distinct). For each root kt of system (11) we find
the values of a and p. The general solution, like (6 ), will hâve
the form
x = Cxa (1)e*»*-f- C2a (a)eV + C3a (3)e*3*+ C4a (4)eV
y = Cju'e*'* + C2P<2)eV + C3p<3)e*‘*+ C4P(4)e*‘*
If there are complex roots, then to each pair of complex roots
in the general solution there will correspond expressions of the
form (9).
Example 3. Find the general solution of the following system of differential
équations
d2x
= x — 4y
dt2
d2y
I t 2 = —x+y
Solution. Write the auxiliary équation (12) and find its roots:
k 1= i, k2 = — i, k3= Ÿ " 3, k 4 = — }r 3.
We shall seek the solution in the form

* (l| = a U) e tt< y { l) — p « ) g it
*<2 > = 0 <*>e - H y i 2) = p (2 ) g - , 7 j
J ((S ) = a (8 ) e V ~ 3 \ y W = p (3 ) g Y a t
y < 4 )= = p < 4 )e - ^ F <

j(U > = a< 4>
1.31 On Lyapunov's Theory of Stability 117
From System (11) we find a (7> and P(A*

a(i) = i( pai= i .
«<*> = 1, P<2' = y
a ‘s>=l, p w = - i
« (« = 1, P“' = - y
We Write out the complex solutions:
*(1) = elt = cos / + t sin /, y(1)= y ( c o s i~\~i sin t)
jt(2> = e-/f = cos t — i sin /, i/(2) = -i- (cos t — i sin /)
The real and imaginary parts separately form the solution:
j?1) = co s/, cos/
*(2>= s in /, y(2)= y s i n f
We can now Write the general solution:

x = Cl cos t + C2 sin t + C3eVs 1+ CAe~V
y = Y C lcos < + 4 Ca sin *~ T c * 3‘ ~ y c *e~' 3‘

Note. In this section we did not consider the case of multiple
roots of the auxiliary équation. This question is dealt with in
detail in “Lectures on the Theory of Ordinary Differential Equa
tions” by I. G. Petrovsky.
1.31 ON LYAPUNOV’S THEORY OF STABILITY

Since the solutions of most differential équations and Systems
of équations are not expressible in terms of elementary functions
or quadratures, use is made of approximate methods of intégration
in these cases when solving concrète differential équations. The
éléments of these methods were given in Sec. 1.3; some of these
methods will also be considered in Secs. 1.32-1.34 and in
Chapter 4.
The drawback of these methods lies in the fact that they yield
only one particular solution; to obtain other particular solutions,
one has to carry out ail the calculations again. Knowing one par
ticular solution does not permit us to draw conclusions about the
character of the other solutions.
118 Ch. 1 Différential Equations
In many problems of mechanics and engineering it is sometimes

important to know not the spécifie values of a solution for some
concrète value of the argument, but the type of behaviour for
changes in the argument and, in particular, for a boundless increase
in the argument. For example, it is sometimes important to know
whether the solutions that satisfy the given initial conditions are
periodic, whether they approach some known function asymptoti-
cally, etc. These are the questions with which the qualitative
theory of differential équations deals.
One of the basic problems of the qualitative theory is that of
the stability of the solution or of the stability of motion; this
problem was investigated in detail by the noted Russian mathe-
matician A. M. Lyapunov (1857-1918).
Let there be given a system of differential équations:
-2 T = M *. x, y) )
d y _ flf A (1)
—/t(^« x, y) I
Let x = x(t) and y = y(t) be the solutions of this system that

satisfy the initial conditions
*< = 0 = * 0
} (H
Further, let x —x(t) and y = y(t) be the solutions of équation

( 1) that satisfy the initial conditions
*f =0 —*0 (i" )
ÿt=o= y0
Définition. The solutions x = x(t) and y = y(t) that satisfy the
équations (1) and the initial conditions (T) are called Lyapunov
stable as t —>oo if for every àrbitrarily small e > 0 there is a
6 > 0 such that for ail values t > 0 the following inequalities are
fulfilled:
lf(0 - * ( 0 l< « \
\y(t )— y ( t ) \ < e f
if the initial data satisfy the inequalities
1*0— *0 I< 6 \
(3)
li/o—*/0 l < 6 I
Let us figure out the meaning of this définition. From inequa
lities (2) and (3) it follows that for small variations in the initial
Î.3Î On Lyapunov's Theory of Stability 119
conditions, the corresponding solutions differ but little for ail positive
values of /. If the System of differential équations is a System that
describes some motion, then in the case#of stability of solutions,
the nature of the motions
changes but slightly for
small changes in the initial
data.
Let us analyze an example
of a first-order équation.
Suppose we hâve the diffe-
rential équation:
t = - y +1 w
The general solution of this équation is the function

y = Ce- * + 1 (b)
Find a particular solution that satisfies the initial condition
y f= o = l (c)
It is obvious that the solution y = 1 results when C = 0 (Fig. 32). Then find
the particular solution that satisfies the initial condition
ÿ*=o = ÿo
Find the value of C from équation (b):
yo —C+ 1
whence
C= ÿ0- 1
Putting this value of C into équation (b), we get
ÿ = (P o —
The solution y = \ is obviously stable. Indeed,
y — y = [ ( ÿ o— i = (ÿ0— i ) «- t -»-o
when t oo.
Hence, inequality (3) will be fulfilled for an arbitrary e if the following
inequality holds true:
(i/o — 1) = 6 < e
I! the équations ( 1) describe motion and the argument t, the
time, is being given implicitly, that is, if we hâve a system of
the form
W = f ^ x' ^
7ïf-=M*. y)
then this system is termed autonomous.

120 Ch, î Differential Equations
Let us âlso consider the following system of linear differential

équations:
f = °x+ g y \
. . | (4)
•ar = a x +by )
We assume that the coefficients a> b, c, g are constants; by direct
substitution it is clear that x = 0, y —0 is a solution of the Sys
tem (4). Let us investigate the question of the conditions that
must be satisfied by the coefficients of the system so that the
solution x - ^ 0 y y = 0 is stable. This investigation is done asfollows.
Differentiate the first équation and eliminate y and —- on the
basis of the équations of the system
d2x dx . du dx . , , , s dx . . , /' dx \
w = c w + 8 i r ==c w + g ( a x + by ï = c - d r + z a x + b { w - c x )
or
-g — (6 + c) g —(ag—bc)x = 0 (5)
The auxiliary équation of the differential équation (5) is of the form

l 2— (b + c)K— (ag— bc) = 0 (6 )
This équation may be written in determinantal form:
c— X g
=0 (7)
a b— X
[see équation (4), Sec. 1.30].
We dénoté the roots of the auxiliary équation (7) by and X2.
As we shall see below, the stability or instability of the solutions
of system (4) is determined by the nature of the roots Xx and X2.
Let us consider ail possible cases.
I. The roots of the auxiliary équation are real, négative and
distinct: Xt ^ X 2. From équation (5) we find
x —C^-'1-)- C
Knowing x, we find y from the first équation of (4 ). Thus, the
solution of system (4) is of the form:
x = Cje*** + C£x*1 'l
y = [ C A K - c ) e Kt + C A K - c ) é * t] ± j (g)
Note. If g = 0 and 0, then we form the équation (5) for
the function y. Finding y , we then find x from the second equa-
tion of (4). The structure of the solutions (8) is preserved. But if

g = 0, a = 0, then the solution of the System of équations becomes
x ^ C ^ 0*, y = C2ebi (8')
In this case, the analysis of the character of the solutions is
easier to carry out. Choose Cx and C2 so that the solutions (8)
satisfy the initial conditions
= y \t —0 I/o
The solution satisfying the initial conditions will be
__cx0+ g y n- x „ X 2 „x., , x 0X1 — c x 0 — g y n ^ t
~ "T X,x- \ 2
;= _1_ r c*o + gÿo —X„l2 p , t , XoKi— cxq— gyo
(9)
( K — c) ex*'J
? L ^1 —^-2 1 ^<1
From these équations it follows that for arbitrary e > 0, it is
possible to select |x0| and |i/0| so small that for ail ( > 0 it will
be true that | x (t) | < e, | y (t) | < e since e*»' < 1, eV < 1.
We note that in this case
lim a:(0 = 0
/-► +00
lim y ( 0 = 0
t-*+00
Consider the xy-plane. For the system of differential équations (4)
and for the differential équation (5), this plane is termed the
phase plane. We will consider the solutions (8) and (9) of system
(4) as parametric équations of some curve in the phase plane xOy:
x = ff(t, Clt C2) j
(11)
y = ÿ ( t , Cu C2) J
x = y ( t, x0, ÿ0) 1
( 12)
y =V( t , *o. I/o) j
These curves are the intégral curves ( solution curves) of the diffe
rential équation
dy _ ax + by
dx cx-\-gy (13)
which is obtained from the system (4) by dividing the right and
left members by each other.
The origin, 0(0, 0), is a singular point of the differential équa
tion (13), since this point does not belong to the domain of exis
tence and uniqueness of the solution.
The nature of the solutions (9) and, generally, of the solutions

of the System (4) is illustrated by the arrangement of the intég
ral curves
F(x, y, C )~ 0
which form the complété intégral of the differential équation (13).
The constant C is determined from the initial condition yx=x. = y0-
Substituting the value of C, we obtain the équation of the family
in the form
F(x, y, x0, y0) (14)
In the case of solutions (9), the singular point is called a stable
nodal point. We say that a point moving along the intégral curve
approaches a singular point without bound as / —*+ oo.
It is obvious that the relation (14) may be obtained by eli-
minating the parameter t from the System (12). We will not con
tinue the analysis of the arrangement of intégral curves near a
singular point on the phase plane for ail possible cases of roots
of the auxiliary équation and will confine ourselves to an illu
stration of this fact in elementary instances that do not require
unwieldy computations. We note that for arbitrary coefficients the
behaviour of intégral curves of the équation (13) near the origin
is qualitatively the same as is now to be examined in thefollow-
ing examples.
Example 1. Investigate the stability of the solution x = 0, y = 0 of the System
of équations
Solution. The auxiliary équation is

0 =0
0 —2— X
The roots of ihe auxiliary équation are
*,! = —1, X2= - 2
In this case, the solutions (8 ') will be
x = Cle~t) y = C2e~2t
The solutions (9) witl be
x = xde~t, y = yoé~2t (a)
Clearly, x ( t ) —►0, and y ( t ) — >-0 as t — > + oo. The solution x = 0, y = 0 is
stable. Now let us examine the phase plane. Eliminating the parameter t from
the équations (a), we get an équation of type (14).
(b)
1.31 On Lyapunov*s Theory of Stability 123
This is a family of parabolas (Fig. 33).

The équation of type (13) will, for the given example, be
dy_2y
dx x
Integrating, we get
ln | y [= 2 ln | x | + ln I C | (c)
y=- Cx2
Détermine C from the condition
Üx—Xq—ÜO* r-y o
4
Substituting the value of C thus found into (c), we get the solution (b).
The singular point 0 (0, 0) is a stable nodal
point.
II. The ro o ts of th e a u x ilia ry é q u a
tio n a re r e a l, p o s itiv e and d is tin c t:
^ > 0 , X2> 0 , 'k ijk k 2. In this case,
the solutions are also expressed by the
formulas (8) and (9). But in this case,
for arbitrarily small |x0| and \y0 \ it will
be true that | jc (/) | —►oo, | y (/) | —►oo
as t —► ~J~oo, sincee*^—>oo and e ^ ^ o o
as t —►+ oo. On the phase plane, the sin- Fig. 33
gular point is an unstable nodal point:
as t —► -j—oo the point on the intégral curve recedes from the rest
point x = Q, y = 0.
Example 2. Investigate the stability of the solutions of the System
dx dy
dt X* dt
1-À 0
=0
0 2 -k
lis solutions are
^ i= l, X2= 2
The solution will be
x = xùett y = y0e2i
The solution is unstable, since | * ( / ) | —►<», \ y ( t ) \ —►oo as t —► + «. Elimi-
nating t %we obtain
(Fig. 34). The singular point O (0, 0) is an unstable nodal point

III. The roots of the auxiliary équation are real and of unlike
sign, for example: ^ > 0 , 2ca< 0 . From formulas (9) it follows
that for arbitrarilv small U J and |#n|, if cxa+ gy0— it
124 Ch. 1. Differential Equations
will be true that |jc(/)| —>-00, \y(t)\ —*oo as / - + + 0 0 . The solu

tion is unstable. The singular point on the phase plane is termed
a saddle point.
Example 3. Investigate the stability of the solution of the System
dx dy
d t ~ X’ T t ------ 2y
I 1—X 0
= 0
J 0 —2-X
hence, = 1, X2 = —2. The solution is

x = x t f + t, y = y0e - îi
The solution is unstable. Eliminating the parameter t, we get a family of curves
on the phase plane:
yxi = yl)xl
The singular point O (0 , 0) is a saddle point (Fig. 35).
IV . The ro o ts of th e a u x ilia ry é q u a tio n a re com plex w ith n é

g a tiv e re a l p a r t: ^ = B + Xs = a — ( a < 0 ) . The solution of
System (4) is
x = eat [Cxcos pt + Cj sin p/] >1
y = ~ ^ ‘ [(aC1+ pCî - c C 1)cosp^ + (aC8- p C l -c C i)sin p (] J (15)
C = V C î + CI, sinô = £ , cos ô = -?

1.31 On Lyapunoo’s Theory of Stability 125
we can rewrite équations (15) as

x = Ce!“* sin ({5/ + 6)
y = ^= — [(a—c)sin(P^ + 6) + Pcos(p/ + ô)] ^
® >
where C, and C2 are arbitrary constants that are determined from
the initial conditions: x = x0, y —y0 when / = 0, and
*0= C sinô, ^0= --- |^(a—c) sin ô + 0 cos ôj
whence we find
C, = x „ C ,- ™ - '* - '1 (17)
Note again that if g = 0, then the form of the solution will be

somewhat different, but the nature of the analysis does not change.
It is obvious that for arbitrary e > 0, given sufficiently small
| jc0| and |ÿ0|, the following relations will hold:
J * (0 l< e . |0 ( O I < «
The solution is stable. In this case, as t —► + oo,
x (t) —►O and y ( t) —+0
changing sign an unlimited number of times. The singular point
on the phase plane is called a stable focal point.
Example 4. Investigate the stability of the solution of the System of équations
Solution. Form the auxiliary équation and find its roots:

—1—A, 1
», X2 + 2X+ 2 = 0
— 1 —1 - X ~
X1>2 = — 1 ± lf» a = — 1. P = 1
We find Cx and Ca from formulas (17): C1 = x0t Ca = y0. Substituting into (15),
we get
*= (x0 cos * + y0 sin /) \
y = e - t (y0 c o s t — x0 s\nt) f
It is obvious that for arbitrary values of /,
I x | ^ | x0 | + 1y 0 |, | y I ^ I *0 I+ 1Vo 1
As t —►-J- 00, x ( t ) —►O and y ( t) —►O, and the solution is stable.
Let us analyze the arrangement of the curves on the phase plane in this
case. We transform expressions (A). Let
x0= M cos Ô, y0 = M sin ô
M ^ V xl + yl, tan ô = —
Xq
Then équations (A) become
* = M * -< c o s(P / —ô) \
y=jM e~( sin(p/ —ô) /
Pass to polar coordinates p and 6 in the phase plane and establish the relation-
ship p = / (0). Equations (B) become
p cos 0 = M e- * cos (0 / —ô) 1
p sin 0 = Me-* sin (p /^ ô ) / ^
Squaring the right and left members and adding, we obtain
p * = M *e-2t
or
p = Me-* (D)
How does / dépend on 0? Dividing the terms of the lower équation of (C) by
the corresponding terms of the upper équation, we get
tan 0 = tan (p /—ô)
whence
.0 +0
P
Substituting into (D), we hâve
9+6
p = Me ^
p = Me
6_
Putting Me P = M lf we finally obtain
p = M xe 3 (E)
This is a family of logarithmic spirals. In this case, a point moving along an

intégral curve approaches the origin as t —►«>. The singular point 0 (0 , 0) is
a stable focal point.
V. The roots of the auxiliary équation are complex wtth positive
real part: X,2= a — /p ( a > 0 ) . Here the solution will
also be expressed by the formulas (15), where a > 0. For arbitrary
initial conditions^ and y0 (V xl + yl¥= 0), | a: (/) | and \y{t)\ can
assume arbitrarily large values as t —► + «>• The solution is un-
stable. The singular point in the phase plane is called an unstable
focal point. A point on an intégral curve recedes without bound
from the origin of coordinates.

dx
ar*+ y
dy
d t= ~ X+y
1-X 1
= 0, X2—2\-\-2 = 0
—1 1 -K
~ 1+ *» *2= 1 — i
Taking into account (17), the solution (15) in this case will be
x = ef (x0 cos t + y 0 sin /) y
y = e* (y0 cos t Xy sin t)
In the phase plane, we obtain the curve in
polar coordinates:
p=
x
The singular point is an unstable focal point
(Fig. 36).
VI. The roots of the auxiliary
équation are pure imaginary: A^ss/p,
k2= — /p. The solutions (15) in this Fig. 36
case assume the form
x = C1cos pt + C2sin pt
(18)
i/ = - [(pc2—cC,) cosp/+ (—PC,—cC2) sin §t]
The constants C, and C2 are tound from formulas (17):
C i= * 0. C2= g- ^ ± ^ (19)
Clearly, for arbitrary e > 0 and for ail sufficiently small |x0| and
|i/J it will be true that | jc(/) j < c, |y ( 0 |< ® for arbitrary t.
The solution is stable. Here, x and y are periodic functions of t.
In order to carry out the analysis of the intégral curves on the
phase plane, it is advisable to Write the first équation of (18) as
[see (16)]
x —Csin(p/ + Ô)
( 20)
i/ = ^ c o s (p / + 6) —-s in ( p / + 6)
where C and 6 are arbitrary constants. From the expressions (20)

it follows that x and y are periodic functions of t. Éliminate the
parameter t from (20):

cp - . f , x2 c
« ■ « K ' - p - ï '
Eliminating the radical, we get
(»-7 * ) ' < ? ) ' ( ' - £ ) <*»

This is a family of quadric curves (which are real) that dépend
on an arbitrary constant C. Neither of them recettes to infinitif.
Consequently, this is a family of ellipses
around the origin (for c = 0, the axes of the
ellipses are parallel to the coordinate axes).
The singular point is termed the centre
(Fig. 37).
Example 6 . Investigate the stability of the solu
tion of the System of équations
dx
dt * ’
^dt= - 4 *
Fig. 37
Solution. Form the auxiliary équation and find
its roots:
- k 1
= 0, X2 + 4 = 0, k = ±2i
-4 —X
The solutions (20) will be
x = C sin (2/ + 6),
y = 2C cos (2t + 6 )
Equation (21) will assume the form
& + £ ->
We hâve a System of ellipses on the phase plane, and the singular point is a
centre.
VII. Let X1= 0, X3< 0 . The solution (8) in this case becomes
x = Cj + C4eV
( 22)
y = L [ - C lC+ C2{ K - c ) e ^ ] }
Clearly, for arbitrary e > 0 and for sufficiently small |x0| and |t/0|
it will be true that |x ( f ) |< e , |# ( / ) |< e when t > 0. Hence, the
solution is stable.
Example 7. Investigate the stability of the solution of the System
dx r, du
57=°' 5 7 = -* («)
1.3Î On Lyapunov's Theory of Stability 129
Solution. We find the roots of the auxiliary équation

—k 0
= 0, ^2+A, = 0, ^ = 0, Xa= —1
0 - \- k
Here, g = 0. The solutions are found directly by solving the System and without
using the formulas (22):
x = Clt y — C ^ -* (P) y
The solution satisfying the initial conditions . u
x = x0, y = y0 when t = 0 is u u
x = x 0, y = y0e - t (y) 0 x
The solution is clearly stable. The differential n n n u il n n
équation on the phase plane will be ^ ~ = 0. The
complété intégral will be x = C. The intégral Fig. 38
curves are straight Unes parallel to the (/-axis.
From the équations (y) it follows that the points along the intégral curves
approach the straight line y = 0 (Fig. 38).
V I I I . Let ^ = 0, X2> 0 . From formulas (22) or (8') it follows

that the solution is unstable since | x(t) | + | y (/) | —. 0 0 as t —. + o o .
IX . Let ^ = ^ < 0 . The solution is
x = (Cx+ Ctt) eM t
y = Ls é ^ [ C A K - c ) + C A \+ K i - c t ) ) /I (23)
Since gV—►O and te%^ —>-0 as t —►■■f-oo, then for an arbitrary

e > 0 it is possible to choose Cx and C2 (by selecting x0 and y0)
such that it will be true that |x ( / ) |< e , \y ( t) \< * for arbitrary
/ > 0. The solution is thus stable, and x (/) —►0 and y ( t) —►O as
t — ►+ OO.
Example 8 . Investigate the stability of the solution of the System
dx dy
= —x,
dt dt
Solution. We find the roots of the auxiliary équation:
- \- k 0
(A»-f-l)2 = 0 , A,i = à,2 = — 1
0 — 1— k
Here g = 0. The solution of the System will be of the form (8 '):
x — C1e~ ty y = C2e~t
and x —>-0, y —>- 0 as t —►+<». The solution is stable. The family of curves
on the phase plane will be
— = — = £, that is, y — kx
x c_i
130 Ch. 1 Different iat Equations
This is a family of straight lines passing through the origin. The points along
the intégral curves approach the origin. The singular point O (0 , 0 ) is a nodal
point (Fig. 39).
Note that in the case of ^ = ^ > 0 the form of the solution (22)
is preserved, but when 00
I*(01 —*00 and |y (0 1 *
The solution is unstable.
X. Let ^ = 0. Then
x = Cl -\-Cit
(24)
y = — [—cCj + C2—cCtt\
S
Whence it is quite clear that x —<-oo and y —>-oo as t + 00.
The solution is unstable.
dx dy
=0
di = y' dt
Solution. Find the roots of the auxiliary équation
q ^ =0» î = 0» A,1= X2= 0
We find the solutions to be
y — C2» x = C2t-\-C i
It is quite obvious that x —►oo as t —►+<». The solution is unstable. The
équation on the phase plane is ^“ = 0- Ttie intégral curves y — C are straight
lines parallel to the axis (Fig. 40). The singular point is called a degenerate
saddle point.
To give a general criterion of the stability of solution of tne

System (4), we do as follows.
1.3Î On Lyapunov's Theory of Stability 131
We write the roots of the auxiliary équation in the form of com-

plex numbers:
\ = K + iK*
x2=À*+a***
(in the case of real roots, Xî* = 0 and Xî* = 0).
Let us take the plane of a complex variable %*%** and display
the roots of the auxiliary équation by points in this plane. Then,
on the basis of the cases that hâve been
considered, the condition of stability of
solution of the system (4) may be for-
mulated as follows. IInstable
I f not a single one of the roots X2
of the auxiliary équation (6) lies to the
right of the axis of imaginaries, and at
least one root is nonzero, then the solution
is stable; if at least one root lies to the
right of the axis of imaginaries, or both
roots are equal to zéro, then the solution Fig. 41
is unstable (Fig. 41).
Let us now consider a more general system of équations:
% = cx + gy + P(x, y)
(25)
% = ax + b y + Q (x , y)
But for exceptional cases, the solution of this system is not expres-
sible in terms of elementary functions and quadratures.
To establish whether the solutions of this system are stable or
unstable, they are compared with the solutions of a linear
system. Suppose that for x —*-0 and y —► (), the functions P(x, y)
and Q (x, y) also approach zéro and approach it faster than p,
where p = l/x 2+ ÿ2; in other words,
l i m F (£1jd = 0; iif n Q Jîijd = o
p-»o P p-*-o P
Then it may be proved that, save for the exceptional case, the
solution of the system (25) will be stable when the solution of
the system
ÿ = cx + gy '
Tt= ax + by t
is stable, and unstable when the solution of the system (4) is
unstable. The exception is that case when both roots of the auxil-
iary équation lie on the axis of imaginaries; then the question of
the stability or instability oî solution of the system (25) is consi*

derably more involved.
Lyapunov * investigated the question of the stability of solutions
of systems of équations for rather general assumptions concerning
the form of these équations.
In oscillation theory, one often has to deal with the équation
£ ? -'(* •§ ) <26>
Put
Then we get the system of équations

dx_
d t~
(28)
7t = f ( x ’ v)
The phase plane for this system is the xu-plane. The intég
ral curves on the phase plane geometrically represent the velocity v
as a function of the x-coordinarte and give a pictorial and quali
tative description of the variation of x and v. If the point x = 0,
v — 0 is a singular point, then it détermines a position of equilib-
rium.
Thus, for example, if the singular point of a system of équations
is a centre, that is, the intégral curves on the phase plane are
closed curves circling the origin, then the motions described by
équation (26) are undamped oscillations. If the singular point of
the phase plane is a focal point (and then | x | —>0, |u |—»0 as
t —►oo), then the motions defined by équation (26) are damped
oscillations. If the singular point is a nodal point or a saddle
point (and this is the only singular point), then x —>.±oo as
t —*•<». In this case a moving material point recedes to infinity.
d?x dx
If équation (26) is a linear équation of the form ^ :=a* + ^ >
then the system (28) looks like
cbc dv . i
di = u, -j-t = ax + bu
This is a system of type (4). The point jc = 0, u = 0 is a singular
point, it defines a position of equilibrium. Note that variable x is
not necessarily a mechanical displacement of a point. It may hâve
a variety of physical meanings; for instance, it may represent a
quantity describing electrical oscillations.
• A. M. Lyapunov, The General Problem of Stability of Motion, ONTI,
1935 (in Russian).
1.32 Euler's Method of Solution of Differential Equations 133
1.32 EULER’S METHOD OF APPROXIMATE SOLUTION

OF FIRST-ORDER DIFFERENTIAL EQUATIONS
We shall consider two methods of rfumerical solution of first-

order differential équation. In this section we consider Euler’s
method. Find (approximately) the solution of the équation
(1)
on the interval [*0, b] that satisfies the initial condition at x = x0

y — y„. Divide the interval [*0, b] by the pointsx0, xlt x2, x„ = b
into n equal parts (here x0 < x, < x2 < . . . < x„). Dénoté —x0 =
= x2—x t = . . . = b—x„.1= Ax = h; hence,
Let y*=tp(x) be some approximate solution of équation (1) and

y„ =* ç (*<>). Vx = «P ( * i) . • • •. yn = «p (*■)
Put
Ay2—y \ y2, Ay2 y2 y2, . . Ayn—t yn yn—i
At each of the points x0, xlt . . . , x n in équation (1) we replace
the dérivative with a ratio of finite différences:
(2 )
Ay=--f(x, y) Ax (2')
When x = x0 we hâve
/
—/ (-'‘a» yà)t Aÿ0—/ (x„, y0) Ax
or
«/i—*/»= /(* ». yo)h
In this équation, x0, y0, h are known; thus we find
Vx=y«+ /(*«. yo)h

When x = xlt équation (2') takes the form
= ÿx)h
or
ÿ2— yx = f ( xx>Vx)h
y*= & + / ( * ». yJf*
Here, xlt ylt h are known and y2 is determined.

Similarly, we find
& = * * + /(* * . y*)h
0 a+i = 0 a+ /(**. y * ) h
= + /(*»-!. y n -Jh
We hâve thus found the approximate values of the solution at

the points x0, xt, . . . , xn. Connecting, in a coordinate plane, the
points (x0, i/0), (x„ t/j)........ (xn, yn) by
straight-line segments, we get a polygonal
line—an approximate intégral curve
(Fig. 42). This line is called Euler's poly-
gon.
Note. We dénoté by y = (x) an ap
proximate solution of équation (1), which
corresponds to Euler’s polygon when Ax =
=/t. It may be proved * that if there exists
a unique solution y = cp*(x) of équation
p. 42 (1) thât satisfies the initial conditions and
lg‘ is defined on the interval [x0, b], then
Jim q>A(x)—<p*(x)|=-0 for any x of the interval [x0, b].
h- o
Example. Find the approximate value (for x = 1) of the solution of the
équation
y '= y + *
that satisfies the initial condition y0= 1 for x 0 = 0 .
Solution. Divide the interval [0, 1] into 10 parts by the points x0 = 0 ,
0.1, 0.2...........1.0. Hence, A= 0.1. We seek the values y lt yt .......... yn by for
mula (2 '):
Ay k = (yk + * k )h
or
yn+i=yk+(yk+Xk)h
We thus get
y , = l + ( l + 0)- 0.1 = l + 0 .I = l.l
y, = l . l - f ( 1.1 + 0 . 1). 0.1 = 1.22
Tabulating the results as we solve, we get:
* For the proof see, for example, I. G. Petrovsky’s Lectures on the Theory
of Ordinary Differential Equations.
1.33 A Différence Method for Âpproximate Solution of Differential Equations 135
Xk y.k Vk+Xk &yk=(yk+xk)h

•
x0 = 0 1.000 1.000 0.100
xl = 0.1 1.100 1.200 0.120
x2 = 0.2 1.220 1.420 0.142
*3 = 0 .3 1.362 1.620 0.162
xA = 0 .4 1.524 1.924 0.1924
= 0 .5 1.7164 2.2164 0.2216
*8 = 0.6 1.9380 2.5380 0.2538
x1 = 0 .7 2.1918 2.8918 0.2892
Xg = 0.8 2.4810 3.2810 0.3281
' x9 = 0 .9 2.8091 3.7091 0.3709
XlQ= 1.0 3.1800
We hâve found the approximate value y \ x- \ =3.1800. The exact solution

of this équation that satisfies the indicated initial conditions is
y = 2ex — x — 1
Hence,
y \x=l = 2 ( e - \ ) = 3.4366
0 2566
The absolute error is 0.2566; the relative error is 3^4355 = ®.075 a 8 %.
1.33 A DIFFERENCE METHOD FOR APPROXIMATE
SOLUTION OF DIFFERENTIAL EQUATIONS BASED
ON TAYLOR'S FORMULA. ADAMS METHOD
We once again seek the solution of the équation
y' = f (x, y) . (1)
on the interval [x0, b], which solution satisfies the initial condi
tion y = y0 when x = x0. We introduce notation that will be needed
later on. The approximate values of the solution at the points
•*•<>> ^1* *• **
will be
Vt )> f/l> ü i. • • • > lJn
The first différences, or différences of the first order, are
t y o = y i — yof i j r ÿ ! - ! / . , • • ^ y n - i = y n — y n- i
The second différences, or différences of the second order, are
Aty» =-- A{/t — &y(t = yî — 2y1+ y0
Aaÿx = Ayt —Ay x = y 3—2t/4+ yx
&*yn- t = &yn- i — a«/„_2= yn— tyn-i+ yn-*

Différences of the second différences are called différences of the
thfrd order, and so forth. We dénoté by y'„, y[, . . . , y'„ the approx-
imate values of the dérivatives, and by y"0, y[, . ..,y"n the

approximate values of the second dérivatives, etc. Similarly we
détermine the first différences of the dérivatives:
Ay^—yi ^ i/i= y% i/i> • • •> &ÿn—1 = ÿn y<i-i
the second différences of the dérivatives:
A% - Ay \— Ay’0, Aîy[ = Ayt — Ay\........ A*y’n_t = Ayn. v —Ay'n_t
and so on.
Write Tàylor’s formula for solving an équation in the neigh-
bourhood of the point x = x0 [see (6) Sec. 4.6, Vol. I]:
y = y * + ^ y \ + ^ f - y ' ' « + ••• + + (2>
In this formula y 0 is known, and the values y 0, y l , . . . of the
dérivatives are found from équation (1) as follows. Putting the
initial values x9 and y 0 into the right side of équation (1), we
find y'0:
y ’» = * f ( x a, y u)
Differentiating the terms of (1) with respect to x, we get
y "= % + % y ' (3)

Substituting into the right side the values x0, y#, y„, we find
y ^ { r X + % y ' ) x=x., y = y .,ï= y '0

Once more differentiating (3) with respect to x and substituting
the values x0, y0, y'„, yl, we find y0". Continuing in this fashion,
we can find the values of the dérivatives of any order for x = x9.*
Ail terms are known, except the remainder Rm in the right side
of (2). Thus, neglecting the remainder, we can obtain an approx
imation of the solution for any value of x\ the accuracy will
dépend upon the quantity \x —x0| and the number of terms in
the expansion.
In the method given below, we détermine by formula (2) only
the first few values of y when \x —x0 \ is small. We détermine
the values y, and yt for x1= x0+ h and for xt —x0+ 2h, taking
four terms of the expansion (y0 is known from the initial data):
, h , , h2 » . hz , ,,
yi= y»+ -çy< ,+ j^y*+ $\yo (4)
. 2 / 1 , , (2/i)* . , (2/i)*
yt = y e + j y e + xrtyi> + -âi-y* (4')
* From now on we shall assume that the function / (x, y) is différentiable
w ith respect to x and y as many times as is required by the reasoning.
1.33 A Différence Method for Approximate Solution of Differential Equations 137
We thus consider known three values* of the function: y0,

yv i On the basis of these values and using équation (1),
we find •
ÿ o = f ( ^«»ÿo)> ÿ i = / C ^ i » yù * y * ~ f (•*■!>f t )
Knowing, y y [ , yt, it is possible to détermine Ay„ Ay[, à*y'0.

Tabulate the results of the computations:
X y y’ AV
*0 yo y'»
x i = x0+ A yi y[
Ay [
•** = ■*0+2* y* y*
... ... ... ... ...
* *-* = * o + ( * — 2 )* y* -i y'k -t
a» ;.,
* * - i = * o + ( * — • )* y k - 1 y'k -1
& y 'k~ i
** = •*0 + ** yk y'k
Now suppose that we know the values of the solution

y», y i. y*.......0*
From these values we can compute [using équation (1)] the values
of the dérivatives
y«> y'u y \ ....... y'k
and, hence.
Ai/0, Ay[, . . . . A
* If we were to seek the solution with greater accuracy, we would hâve

to compute more than the first three values of y. This is dealt with in
detail by Ya. S. Bezikovich in Approximate Calculations, Gostekhizdat,
1949 (in Rqssian).
and
^y'<» &2y'u • • •. bVk-z
Let us détermine the value of yk+l from Taylor’s formula setting
a = x„, x = xk+1 = xk + h:
h. hfi . /ifli
yk+ 1 — yk+ ~ { / * + ]7 2 yk + 1.2 .3 y>t + ••• + ~i~
In our case we shall confine ourselves to four terms of the

expansion:
h . h2 h3
yk+i = y/' + - r y k + ï^ y k + Y j^ y k (5)
The unknowns in this formula are yk and yk”, which we shall
try to détermine by using the known first and second différences.
First, represent in Taylor’s formula, putting a = x*.
x — a= —■h:
e>i
1-
(6)
II
T
and yk_k . putting a — xk, x — a = —2h:

yk-2 — yk+ | Uk+ j.2 yk (7)
From (6) we find
. / a , h » h2 ,,,
yk— yk-i - At/t.j = — yk— y^yk (8)
Subtracting the terms of (7) from those of (6), we get
, . a ' h 3h2 ...
yk-1 yk-2 — ty k —2 —"j- yk 2 y* (9)
From (8) and (9) we obtain
&y'k-i — by'k-2 = ^y'k -2 = ^y'k"
or
<1
(10)
II
M
1
Putting the expression yk ' into (8), we get

^ «
<
<
*««1«Ç
T
(11)
1
Thus, yk and yk" hâve been found. Putting expressions (10)

and (11) into the expansion (5), we obtain
yt,+i = y k + j y k + ^ y k - i + ^ ^ y ' k - z 02)
This is the so-called Adams formula with four terms. Formula

(12) enables one to compute yk+1 when yk, yk- l, yk- 2 are known.
Thus, knowing y0, y, and y%we can find y., and, further yt , yit . . . .
1.33 A Différence Method for Approximate Solution of Differenlial Equations 139
Note I. We state without proof that if there exists a unique

solution of équation (1) on the interval [x0, f>], which solution
satisfies the initial conditions, then the error of the approximate
values determined from formula (12) does not exceed, in absolute
value, Mh4, where M is a constant dépendent on the length of
the interval and the form of the function f (x, y) and independent
of the magnitude of h.
Note 2. If we want to obtain greater accuracy in our computa
tions, we must take more terms than in expansion (5), and for
mula (12) will change accordingly. For instance, if in place of
formula (5) we take a formula containing five terms to the right,
that is, if we complété it with a term of order h4, then in place
of formula (12) we, in similar fashion, get the formula
*/*+!= yk + y y'k + y Ay L i + ^ A î-s + j

Here, yk+l is determined by means of the values yk, yk-i,
and yk. 3. Thus, in order to begin computation using this formula
we must know the first four values of the solution: y0, yt, y%, </,.
When calculating these values from formulas of type (4) one
should take five terms of the expansion.
Example 1. Find approximate values of the solution of the équation
y '= y + *
that satisfies the initial condition
y0= 1 when x0 = 0
Détermine the values of the solution for x = 0.1, 0.2, 0.3, 0.4.
Solution. First we find yx and y2 using formulas (4) and (4'). From the
équation and the initial data we get
#0= (y-M)x=o = */o+ *o= 1+0=1
Differentiating the given équation, we hâve
f = y '~H
Hence,
yo = (y’ + l)*=o= 1+ 1 = 2
Differentiating once again, we get
Hence,
/>/ » .
ÿo =ÿo = 2
Substituting into (4) the values yt , y0, yl and h = 0A, we get
0 i = l + + - 1 + ^ . 2 + M . 2=1.1103
1 T h2 123
Similarly, for ft = 0.2 we hâve
. , 0.2 , , (0.2)2 o i (°-2)3 o ,
Knowing yQt y2, we find (on the basis of thé équation)

yo —yo~h xo~ 1
yi = yi -j- Xi = 1.1103 + 0.1 = 1.2103
y2 = y 2 + x 2 = \ .2427 + 0.2=1.4427
A ^ = 0.2103
A y [ = 0.2324
A*yo= 0.0221
Tabulating the values obtained, we hâve
X y V' a y A V
*0 = 0 ÿ o = 1.0000 1/0 = 1
1 Ayô = 0.2103
x1 = 0.1 = 1.1103 y[ = 1.2103 A2yi = 0.0221
Aî = 0.2324
*a= 0.2 y2= 1.2427 y2= 1.4427 bpy’i = 0.0244
Ay2 = 0.2568
*3 = 0.3 ys — 1.3995 y3 = 1.6995
* 4 = 0.4 y4= 1.5833
From formula (12) we find y3:
y» = 1 .2 4 2 7 + - ^ . 1.4427+— - . 0 . 2 3 2 4 + 0 . 0 2 2 1 = 1.3995
We then find the values of y3, Ay2t A2y[. Again using formula (12) we find y4*
yt = 1.3995+— - • 1.6995 + + • 0.2568 + — 0.1 0.0244= 1.5833
The exact expression of the solution of the given équation is

y = 2ex — x — 1
1 32 A Différence Method for Approximate Solution of Differential Equations 141
Hence, yx=n,4= 2e0A ^ 0.4— 1 = 1.58364. The absolute error is 0.0003; the rela-
0 0003
tive error, ~ 0.0002 = 0 .02 % (In Euler’s ipethod, the absolute error
of y 4 is 0.06, the relative error, 0.038 = 3.8%.)
Example 2 . Find approximate values of the solution of the équation
y '= y 2+ x 2
that satisfies the initial condition yo = 0 for *0 = 0. Détermine the values of the
solution for * = 0.1, 0.2, 0.3, 0.4.
Solution. We find
= O2+ O2= 0
yx-o = (2^' + 2*)x=o = 0
y'x- o = (2 y'2 + 2yy'f + 2)x=0 = 2
By formulas (4) and (4') we hâve
ÿi = — •2 = °.0°03. y, = — ^ -2 = 0.0027
From the équation we find

yo = 0 , ^ = 0 .0100, 1/2 = 0*0400
Using these data, we construct the first rows of the table, and then déter
mine the values of y3 and y 4 from formula ( 12).
X y y' Af/' AV
•*
*o = 0
il
yo — Q
o
Ayi = 0.0100
*!= 0.1 = 0.0003 y[ =0.0100 Aîy0 = 0.0200
=0.0300
x2 = 0.2 y, = 0.0027 y\ — 0.0400 A2y| — 0.0201
Ay, = 0.0501
*3 = 0*3 yt = 0.0090 y3 = 0.0901
* 4 = 0.4 y4 = 0.0204
Thusf
y3= 0 . 0 0 2 7 + • 0.0400 + 0.0300 + ^ •0 .1•0.0200 = 0.0090
yt = 0.0090 + 0.0901 + ^1-0.0501 + — 0.1 0.0201 =0.0214
Note that, to four places, y A= 0.0213. (This may be obtained by other, more
précisé, methods with error éstimation.)
1.34 AN APPROXIMATE METHOD FOR INTEGRATING SYSTEMS

OF FIRST-ORDER DIFFERENTIAL EQUATIONS
The methods of approximate intégration of differential équa

tions considered in Secs. 1.32 and 1.33 are also applicable for
solving Systems of first-order differential équations. Here,
we consider the différence method for solving Systems of équa
tions. We will be concerned with Systems of two équations in
two unknown functions.
It is required to find the solutions of a System of équations
% = fi(x,y,z) (1)
% = fi(x,y,z) (2)
that satisfy the initial conditions y = y0, z —z0 when x = x0.
We détermine the values of the functions y and z for values of
the argument *0, x2, . . . , xk, xA+1, . . . . xn. Again, let
xk+i— xk — ^x = h (k —0, 1,2, ...» n — 1) (3)
We dénoté the approximate values of the functions y and z, respec-
tively, by
i/o» y i> • • • * yk< i/*+i> • • •» yn
and
2o> Z,, . . ., Zh, Zfr+j , . . .i z„
Write the récurrence formulas of type (12), Sec. 1.33:
Vk+i = y*+ y#* + y &y'k-i + j2 (4)
2*+, = z* + y z ; + 4 A2*-i + Â ^ _ , (5)

To begin computations using these formulas we must know yu t/2;
z„ z2 in addition to y0 and z0; we find these values from formu
las of type (4) and (4'), Sec. 1.33:
1.34 An Approximate Method for Integrating Systems of Equations 143
, 2 h , x (2/i)2 „ . (2h f ,,
y 2 —Ho+ y i/o + ~2~ yo+"3?" y°
_ . h ,, h2 h**
zo~r j i 2 ' 3! Z°
2/i (2/i)2 w -
^2 —2o+ T z° ”1” 3! Z°
' To apply these formulas, one has to know y0, y"Qi y’0", z'0t zï,
z’0" 9 which we shall now détermine. From (1) and (2) we find
y* = /i (*., y 0, z0)
Zo fî (Xo> I/o» 20)
DifFerentiating (1) and (2) and substituting the values of ■*o> ÿo> 20(
y'0 and z'0, we find
y a —(y )*=*„
2o —(2 )jc=jc0
Differentiating once again, we find and z’0 Knowing j/„ â,

z1( z2, we find from the given équations (1) and (2),
y[, y'i. z'i, z;, At/i, Aÿ;, A2y;, Az;, az;, a 2z;
after which we can fill in the first five rows of the table:
X y y' ! 2 2* A2' A*2'
*0 yo yo z’o
Il * 1
Ay'o AZq
*1 yi y'i A2ÿo j z, *1 A*zi
Ay[ Azî
0
*2 y2 y'2 A*y[ ! z2 *2 A2z|
Aÿâ AZ2
*3 y3 yo *3 2S
1
144 Ch. 1 Difjerential Equations
From formulas (4) and (5) we find y3 and z3, and from equa-
tions (1) and (2) we find y3 and z3. Computing Ay't\ A2y[, Az2,
we find yAt and z4, etc., by applying formulas (4) and (5) once
again.
Example. Approximate the solutions of the System
y' = z, *'= y
wlth initial conditions yo = 0 and z0= l for x = 0. Compute the values of the
solutions for x = 0, 0.1, 0.2, 0.3, 0.4. f
Solution. From the given équations, we find
i/o= 2*=o=l
zo=yx=o= ®
Differentiating the given équations, we find
yo = {y")x=o = (z')x=o = 0
zo = (z )x-o = (y )x=o = f
o=( 2w)x=o==i
yô" = ly '" )x =
^ // = (2/")x =O=(^)x =O= 0
Using formulas of type (4) and (5), we find
(0.1)3•1=0.1002
3!
( 0 . 2)2 ( 0 . 2)3
y ,= o + — i- 1-2 3!
•1=0.2013
(Q.l)2 (O.l)3
*i = 1+ y 'û - 1-2 3!
•0=1.0050
( 0 . 2)2 (0 .2)3
?,= 1 + ^ - 0 - 2! 3!
•0=1.0200
Using the given équations, we find
y x= 1.0050, zx= 0.1002

y 2= 1.0200, Zo —0.2013
Ay'o = 0.0050, Az^= 0.1002
&y‘i = o.oi50, Azi = 0.1011
A2^ = 0.0100, A2zi = 0.0009
and fill in the first five rows of the table (see p. 145).
From formulas (4) and (5) we find
y3 = 0.2013 + ^ - 1.0200+ - - 0 . 0 I 5 0 +-^-0.1-0.0100 = 0.3045
?3= 1.0200 + ^— 0.2013 + — -0.1011 + --0 .1 -0 .0 0 0 9 = 1.0452

X y y' Ay' J AV
0
x 0= 0 y ’o = 1 •
II Ayi = 0.0050
xt = 0.1 1/4 = 0.1002 1/1 = 1.0050 A*yô = 0.0100
A ^ = 0.0150
^2 = 0.2 1/2 = 0.2013 1/ 1 = 1.0200 A*y| =0.0102
Ayl = 0.0252
X3 = 0.3 t/g = 0.3045 * = 1.0452

O
l/4 = 0.4107
H
II
TP
X z z' Az' AV
O
0H
z i= 0
II
Z 0 —1
Azi = 0.1002
*t = 0.1 24 = 1.0050 z[ = 0.1002 A2zi = 0.0009
Azl = 0.1014
*2 = 0.2 z2= 1.0200 zl = 0.2013 A»zl = 0.0021
A22 = 0.1032
x 3 = 0.3 z3= 1.0452 Zs = 0.3045

1
jc4 = 0.4 z4= 1.0809
10—2082
and similarly
y 4 = 0.3045+—j—-1-0452 0.0252 + —-0.1 -0.0102 = 0.4107
zt = 1.0452+-9 p . 0.3045 + - — 0.1032 + ~ 0.1 •0.0021 = 1.0809
lt is obvious that the exact solutions of the System of équations (the solu
tions satislying the initial conditions) will be
y = sinhx, z = cosh*
And so, solutions correct to the fifth décimal place are
y A= sinh 0.4 = 0.41075, z4 = cosh 0.4 = 1.08107
Note. Since higher-order équations and Systems of équations in
many cases reduce to a System of first-order équations, the meth-
od given above is applicable to the solution of such problems.
Exercises on Chapter 1
Show that the indicated functions, which dépend on arbitrary constants,
satisfy the corresponding differential équations:
Functions Differential Equations
du 1
1. y = sin x — 1+ C e~sxnxf ■fc + y cos x = Y s in 2x.
2 . y = Cx + C — C2t
3. y2 = 2Cx + C2,
a2C
4. y* = Cx2-
1+C*
y=Cix+-£-{-c3, | Z à*y
5. dx3 ‘ x dx2
6.# = (C1+ C^)<**+ 7F^ n r , S — 2k% +k2y = e \

7. y = CxeoarcsinJt+ C ïe - ° arcsinjc.
# y | 2 dy
dx2 ' x dx
Integrate the differential équations with variables separable:
9 . y dx—x d y = 0. Ans. y = Cx. 10. (1 -f u) vdu-\-(\ —u) udv = 0. Ans.
In uv + u —v = C. 11. (\ + y)dx — (1—x)dy = 0. Ans. (1 + y ) (1 —x) = C.
12. ( t * - x t t) ^ ~ - + x * + tx i = 0 . Ans. i ^ i + ln — = C. 13. ( y - a ) d x + x2dy=0.
Ans. ( y —a) = Cex . 14. z d t — (<* —at )dz = 0. Ans. zsa = C * ° . 15. —=
= |± £ i . Ans. x = = j t ± ^ . 16. (1 + s * ) d t— Ÿ'Tds — O. Ans. 2 —

— arctan s — C. 17. dp + p tan 0 d0 = O. Ans. p = C cos0. 18. sin0cos<pd0 —

— cos 0 sin <pcf<p= 0. Ans. cos <p= C cos 0. 19. sec2 0tan <pd0 + sec2 <ptan 0dcp=O.
Ans. tan0tan<p = C. 20. sec2 0 tanjpdç + sec2 <ptan0 dd = 0. Ans. sin 2 0 +
+ sin2 <p= C. 21. (î + x 2) dy — J/ l -—y2 dx = 0. Ans. arcsin y — arctanx—Ce
22. Y l - x 2 d y — Y l'= ÿ î dx = 0 Ans. . y V l - ~ x * - x V \ — y2 = C.
23. 3e* tan */dx + (l —e*)sec2 y d y = 0. Ans. tan y = C (1 —ex)3. 24. (x — y 2x)dx+
+ (y —x2y) dy = 0. Ans. x2+ y2 = x2y2-\-C.
Problems in Setting up Differential Equations
25. Prove that a curve having the slope of the tangent to any point propor-
tional to the abscissa of the point of tangency is a parabola. Ans. y = ax2 + C.
26. Find a curve passing through the point (0, —2) such that the slope of
the tangent at any point of it is equal to the ordinate of this point increased by
three units. Ans. y = ex — 3.
27. Find a curve passing through tfoe point (1, 1) so that the slope of the
tangent to the curve at any point is proportional to the square of the ordinate
of this point. Ans. k ( x - \ ) y — y-{~l= 0.
28. Find a curve for which the slope of the tangent at any point is n times
the slope of a straight line connecting this point with the origin. Ans. y = Cxn.
29. Through the point (2, 1) draw a curve for which the tangent at any
point coïncides with the direction of the radius vector drawn from the origin
to the same point. Ans. y = ~ x .
30. In polar coordinates, find the équation of a curve at each point of
which the tangent of the angle between the radius vector and the tangent line
is equal to the reciprocal of the radius vector with sign reversed. Ans.
r(0 + C ) = l.
31. In polar coordinates, find the équation of a curve at each point of which
the tangent of the angle formed by the radius vector and the tangent line is
equal to the square of the radius vector. Ans. r 2 = 2(0 + C).
32. Prove that a curve with the property that ail its normals pass through
a constant point is a circle.
33. Find a curve such that at each point of j t the length of the subtangent
is equal to the doubled abscissa. Ans. y = C Y *•
34. Find a curve for which the radius vector is equal to the length of the
tangent between the point of tangency and the Jt-axis.
Solution. By hypothesis, V 1+ y'* = Y x 2+ y2, whence —
Q
Integrating, we get two families of curves: y = Cx and .
35. By Newton’s law, the rate of cooling of some body in air is proportional
to the différence between the température of the body and the température ol
the air. If the température of the air is 20°C and the body cools for 20 mi
nutes from 100° to 60°C, how long will it take for its température to drop
to 30°C?
Solution. The differential équation of the problem is — = k ( T — 20). Inte
grating we find: T —20 = Cek*; T = 100 when f = 0, T = 60 when / = 20; there-
_i_ t
fore, C = 80, 40 = Ce20*, **==^"<j~ ) 20 * consequently, r = 20 + 80 ^ - - ^ 20 .
Assuming T — 30, we find / = 60 min.
36. During what time T will the water flow out of an opening 0.5 cm2 at
the bottom of a conic funnel 10 cm high with the vertex angle d = 60°?
Solution. In two ways we calculate the volume of water that will flow out
during the time between the instants t and / + A/. Given a constant rate t»,
during 1 sec a cylinder of water with base 0.5 cm2 and altitude h flows out,
and during time At the outflow is the volume of water dv equal to
—dv= — 0 .h v d t = — 0.3 ÿ 'ï g h d f
On the other hand, due to the outflow, the height of the water receives
a négative “incrément” dh, and the differential of the volume of water out
flow is
- d v = nr'-dh= — (h + 0,7)* dft
Thus,
— (h+ 0 .7 )2dh = — 0.3 V 2gh dt
ü
whence
/Ô.OSIStlO11/* —A,/‘) + 0.0732(10,/‘—hv «)+0.078( Y h)
Setting h = Q, we get the time of outflow T = 12.5sec.
37. The retarding action of friction on a disk rotating in a liquid is pro
port ional to the angular velocity of rotation cû. Find the dependence of this
angular velocity on the time if it is known that the disk begins rotating at
100 révolutions per minute and, after the elapse of one minute, rotâtes at
60 révolutions per minute. Ans. co= 100^-|-^ rpm.
38. Suppose that in a vertical column of air the pressure,at each level is
due to the pressure of the above-lying layers. Find the dependence of the pres
sure on the height if it is known that at sea level this pressure is 1 kg per cm2,
while at 500 m above sea level it is 0.92 kg per cm2.
Hint. Take advantage of the Boyle-Mariotte law, by virtue of which the
density of a gas is proportional to the pressure. The differential équation of
the problem is dp — — kpdht whence p = e~0000l7h. Ans. p = e “ 0 000l7ft.
Integrate the following homogeneous differential équations:
39. (y — x) dx + (y + x) dy = 0. Ans. y 2+ 2xy— x2 = C. 40. (x + y ) dx + x dy = 0 .
Ans. *2 + 2xy — C. 41. (x + y) d x + ( y — x) dy — Q. Ans. In (a:2 + y 2)l/*—
— a rc ta n -y = C . 42. x d y — y d x = yrx2 + y2dx. Ans. 1 + 2 C y—C2*2 = 0 .
43. (8y + 10*) dx + (5y + 7*) dy = 0 . Ans. (x + y)2 (2* + y f — C. 44. (2 }^st —s) d t+
l/jL C
+ t ds = 0. Ans. t e v * ==C or s = / l n 2 — . 45. (t — s) di + t ds = 0. Ans.
S t
— r _
t e * — C or s = / l n y . 46. xy2 dy = (x3 + y3) dx. Ans. y = x ^ /3 \n C x .
47. * cos ^ - ( y dx + xdy) — y sin — (x d y — y dx). Ans. xy cos — = C .

X X X
Integrate the differential équations that lead to homogeneous équations:
48. (3y— 7x + 7 )d x— (3*—7y— 3)dy = 0. Ans. (x + y — l )6 (x -y -~ - 1)2 -=C.
49. (x + 2 y + l) d x — (2x + 4y + 3)dy = 0. Ans. \n(4x + 8 y + 5) + 8y— 4x — C.
50. (x + 2 y + l ) d x - ( 2 * - 3 ) d i / = 0 . Ans. l n ( 2 * - 3 ) - ^ t | = C .
* The rate of outflow v of water from an opening a distance h from the

free surface is given by the formula u = 0.6 Vr2gh, where g is the accélération
of gravity.
51. Détermine the curve whose subnormal is the arithmetic mean between
the abscissa and the ordinate. Ans. (x— y)2 (x + 2y) = C.
52. Détermine the curve in which the ratio of the y-intercept of the tangent
to the radins vector is equal to a constant. •
_ x <*y
Solution. By hypothesis, — m, w hence^— ) — *
53. Détermine the curve in which the ratio of the x-intercept of the normal
to the radius vector is equal to a constant.
x+ y
ày_
dx
Solution. It is given that =m, whence x2-f- y2 = m2 (x — C)2.
Y ^T y2
54. Détermine the curve in which the y-intercept of the tangent line is
equal to asec 0 , where 0 is the angle between the radius vector and the
x-axis.
Solution. Since ta n 0 = -|- and by hypothesis
y —x -r~ — o sec 0
9 dx
we obtain
dy V x i + yi
« x ïx = a x
whence
x r T +b - ( — +b
y= 2 \ e — 1e v
55. Détermine the curve for which the y-intercept of the normal drawn to
some point of the curve is equal to the distance of this point from the origin.
x
Solution. The y-intercept of the normal is y - 1— , ; therefore, by hypothesis,
y
we hâve
y+^r=
U
V x2+yt
whence
x2 = C(2y + C)
56. Find the shape of a mirror such that ail rays emerging from a single
point 0 are reflected parallel to the given direction.
Solution. For the x-axis we take the given direction, O is the origin. Let
OM be the incident ray, MP the reflected ray, and MQ the normal to the
desired curve:
a = p, OM = OQ, NM — y
NQ = NO + OQ= — x + V'x*+y* = yco tp = y-£-
whence _____
y d y = ( — x + Y x t + y !‘)dx
Intégrâting, we hâve
y2= C2+ 2Cx
Integrate the following linear differential équations:
57. * ' - - ^ = ( * + 1)». Ans. 2y = (* + l)* + C (* + l)* . 58. y ' - a J L = ï± ± .
i4ns. y = C*a-|~— - — . 59. (x —*»)y' + (2*2— \ ) y — ax? = 0. Ans. y = ax +
+ C x Y 1 —x*. 60. -j^cos <+ ssin / = 1. Ans. s = sin / - f Ccosf. 61. ^ + scos / =
= -^- sin 2t. Ans. s = sin / — 1 + Ce “ Sln *. 62. y' — — y —e*xn. Ans. y —xn (e*-{-C).
& x
1
63. y '- 1---- */ = — . Ans. xny = ax~\-C. 64. y ’ + */ = — . Ans. exy — x + C.
X X / \\ &
55. y ' + ^ r r ^ y — 1 = 0 . Ans. y = *2 ( l + C e J .

Integrate the Bernoulli équations:
66. y '-f xy = x3y9. Ans, y2 (x2 + 1+ Cex*) = 1. 67. (1 —x2) y' — xy — axy2 = 0.
Ans. (C Ÿ 1—x2—a ) y = 1. 68. 3y2y ' — ayz — x — 1 = 0 . Ans. a2y3 = Ceax —
f il ] il
— a ( x + l ) — 1. 69. y' (x2y3 + xy)= 1. Ans. x \(2 — y2) e 2 + C \ = e 2 .
70. (y \n x --2 ) y dx = x d y . Ans. y(C x2 + lnx 2 + l ) = 4. 71. y — y' cos x =
_ .. . . . ta n x + s e c x
= y 2 cos x (\ — sin x). Ans. y = s[nx + c •
Integrate the following exact differential équations:
72. (x2 + y)dx + (x — 2y)dy = Q. Ans. *Ç-\~yx — y2 = C. 73. (y —3x2) dx —
— (4y— x) du = 0. Ans. 2y2 — xy-\-x? = C. 74. (y3— x) y' — y. Ans. y 4 = 4* 1/ + C.
76. 2 (3xy2 + 2X3) dx + 3 (2x2y + y2) dy = 0. Ans. x4 + 3x2y2 + fi = C.
” ln « '+ 7> - Î T ? = c - ™- ( ? + ¥ ) ^ - 2- ^ -
Ans. 7». x^ —
7 u- C . «0- * « + » * -
^ y d x _ x d y _ ' Ang je*-f.y*_ 2 arctan —= C .
x2 + i/2 * y
81. Détermine the curve that has the property that the product of the square
of the distance of any point of it from the origin into the x-intercept oï the
normal at this point is equal to the cube of the abscissa of this point. Ans.
y2 (2x2 + y2) = C.
82. Find the envelope of the following families of lines: (a)y = Cx + C2.
Ans. x2 + 4y = 0. (b) y = — -\-C2. Ans. 27x2 = 4y3. (c) —g^ = 2 . Ans. 27y — x3.
(d) C2x + C y - 1 = 0 . Ans. y2 + 4x = 0. (e) ( x - C )8 + ( y - C )2 = C2. Ans. x = 0 ,
y = 0. (f) (x—C)2 + y2 = 4C. Ans. y2 = 4x + 4. (g) (X- C ) 2+ ( y - C ) 2 = 4. Ans.
(x—y)2= 8. (h) Cx2 + C2y = 1. Ans. *4+ 4y = 0.
83. A straight line is in motion so thaï the sum of the segments it cuts off
on the axes is a constant a. Form the équation of the envelope of ail positions
of the straight line. Ans. x1/* + y,/f = a,/# (parabola).
84. Find the envelope of a family of straight lines on which the coordinate
2_ 2_ 2_
axes eut off a segment of constant length a. Ans. x 3 + y 3 = a 3 .
85. Find the envelope of a family of circles whose diameters are the doubled
ordinates of the parabola y2 = 2px. A n s.y2 — 2p
86. Find ihe envelope of a family of circles whose centres lie on the para-
bola y2 = 2px\ ail the circles of the family pass through the vertex of this
parabola. Ans. The cissoid x? + y2 (x + p ) = 0.
87. Find the envelope of a family of oircles whose diameters are
chords of the ellipse b2x2-\~a2y2 = a?b2 perpendicular to the x-axis. Ans.
—* + 4 = 1.
a2+ b2 ^ b 2
88. Find the evolute of the ellipse b2x2 + a2y2 = cPb2 as the envelope of its
2_
normals. Ans. (,ax)*!* + (b y = (a2— b2) 3 .
Integrate the following équations (Lagrange équations):
2C — p9
89. y = 2xy, + y ,i
Ans• * = ^ - 4 p- 3P
90. y = xy'2+ y '2. Ans. y = ( V^x+ 1 + C )2. Singular solution: y = 0.
91. y = x ( \ + y f) + (y')2. Ans. x = C e~ P -2 p + 2t y = C (p+ 1) e - P - ^ + 2
92. y = yyfZ+ 2xy'. Ans. 4Cx = 4C2— y2. 93. Find a curve with constant nor
mal. Ans. ( x - C ) 2+ y2 = a2. Singular solution: y = ± a .
Integrate the given Clairaut équations:
94. y = xy' y ’ — y ,%. Ans. y = Cx + C — C2. Singular solution: 4*/ = ( x + l) a.
95. y = xy‘ + Y 1 —y'*- Ans. y = Cx-{- V 1—C2. Singular solution: y2—x2= 1.
96. y = xy' + y'. Ans. y = C x+ C . 97. y = x y '+ — . Ans. y = C x + S i n g u l a r
1 J ^
solution: y2 = 4x. 98. y = x y '----- Ans. y = C x - r^ .. Singular solution:
y '2
, 27 a
v3= - 4 -* •
99. The area of a triangle formed by the tangent to the sought-for curve
and the coordinate axes is a constant. Find the curve. Ans. The équilatéral
hyperbola 4x y = ± a2. Also, any straight line of the family y = Cx ± aJ^C .
100. Find a curve such that the segment of its tangent between the coordi
aC
nate axes is of constant length a. Ans. y = Cx± . Singular solution:
1^1+c*
y t a = flV».
101. Find a curve the tangents to which form, on the axes, segments whose
2aC
sum is 2a. Ans. y = Cx — Singular solution: (y — x —2a)2 = Sax.
1-C*
102. Find curves for which the product of the distance of any tangent line
to two given points is constant. Ans. Ellipses and hyperbolas. (Orthogonal and
isogonal trajectories.)
103. Find the orthogonal trajectories of the family of curves y = axn. Ans.
x2+ ny2 = C.
104. Find the orthogonal trajectories of the family of parabolas y2 = 2p (x— a)
x
(a is the parameter of the family). Ans. y = Ce p .

105. Find the orthogonal trajectories of the family of curves x2— y2 = a
Q
(a is the parameter). Ans. y = — .
106. Find the orthogonal trajectories of the family of circles x2-\-y2= 2ax.
Ans. Circles: y = C (x2+ y 2).
107. Find the orthogonal trajectories of equal parabolas tangent at the vertex
ol the given straight line. Ans. If 2p is the parameter of the parabolas and the
given straight line is the y-axis, then the équation of the trajectory will be
108. Find the orthogonal trajectories of the cissoids y2— ^ n*•

(x2+ u2)2 = C (y2+ 2x2).
109. Find the orthogonal trajectories of the lemniscates (x2-\-y2)* = (x2— y2) a2.
Ans. (x2+ y2)2= Cxy.
110. Find the isogonal trajectories of the family of curves: *2= 2a (y — x Y H ) ,
where a is a variable parameter if the constant angle formed by the trajectories
and the Unes of the family is © = 60°.
Solution. We find the differential équation of the family t — Y 3 and
(/' — tan© y '- f â
for y' substitute the expression q = If ©=60°, then <7=
1 + y' tan© ! + / ■ 3</'
and we get the differential équation —-----— V"3. The complété in-
l + if '^ 3 *
tegral y* = C (x —y 3) yields the desired family of trajectories.
111. Find the isogonal trajectories of the family of parabolas y2= 4Cx when
6 , 2 y —x
—— arctan __
© = 45°. Ans. y2— x y + 2jc2 —CeV 7 xV 7.
112. Find the isogonal trajectories of the family of straight lines y = Cx far
2 V 3 arctan —
x2-f- y2 = e x
the case © = 30°, 45°. Ans. The logarithmic spirals u
2 arctan —
X2+ y2= e x.
113. y = C1e* + Ca£-*. Eliminate Ci and C2. Ans. y" —y = 0 .
114. Write the differential équation of ail circles lying in one plane.
Ans. ( l+ y '* ) y '" — 3y'y"t =--0.
115. Write the differential équation of ail central quadric curves whose prin
cipal axes coïncide with the x-and y-axes. Ans. x (y y n y'*) — y'y — 0 .
116. Given the differential équation y '" — 2y" — y'-j-2y = 0 and its general
solution y = C1e*-FC 2e-*-j-C 3e2*.
It is required to: ( 1) verify that the given family of curves is indeed the
general solution; (2) find a particular solution if for* = 0 we hâve y = 1, y' = Ü,
y " = — 1. Ans. y = — (9ex + e - x — 4e2*).
117. Given the differential équation and its general solution
y = ± | - ( * + C I)T + C î .
It is required to: (1) verify that the given family of curves is indeed the
general solution; (2) find the intégral curve passing through the point ( 1, 2) if
the tangent at this point forms with the positive x-axis an angle of 45°. Ans.
Integrate some of the simpler types of differential équations of the second

order that lead to first-order équations.
118. x y "' = 2 . Ans. y — x2 In j c + C ^ + C** + C3; pick out a particular solu
tion that satisfies the following initial conditions: jç = l, y = l , y' = l, y" = 3 .
Exercises on Chapter î 153
m\ xm + n
Ant. y = * * l n * + l . 119.y<">=s«. Ans. y = + c ix" ~ 1+ • • • + c n - i* +
+ C „. 120. y" = a*y. Ant. a* = ln(ay + / a y ^ - C j + C , or (,=£><■* +
121. y ' = 4 - - A nt. (C1je + Ca)*= C 1yi!—a.
r
In Nos. 122-125 pick out a particular solution that satisfies the following
initial conditions: x = 0, y — — 1, y ' = 0.
122. x \f — y' = x2ex . Ans. y — e* (x — 1) + Particular solution:
y = ex (x — 1). 123. yy”— (y')2+ (y')s = 0. Ans. y + C1 ln y = x-\-C2. Particular
solution: y = — \. 124. y”--\-y' tan jt=sin 2jc. Ans. y —C2+ C l s \n x — x — ^-sin2*.
Particular solution: y = 2 s\n x — sinxcosjc—je— 1. 125. (y”)2-\-(yf)2= a2. Ans.
y= C 2—a cos (je-j-Ci). Particular solutions: y = a — 1—a cos je, y= a cos je—(û+1).
(Hint. Parametric form: y”= a c o st, y' = a s\n t.) 126. y”= — r . Ans. y =
= ± |- ( * + C 1)3/*+Ci!. m . y ’" = y ’\A n s . y ^ i C ^ - x ) [ln ( C ^ * ) - 1 ]+ C ^ + C 3.

128. y ’y ’" — 3y”È= 0. Ans. x = C1y2-\-C2y + C3.
Integra te the following linear differential équations with constant coefficients:
129. y* = 9y. Ans. y = C1e?x -\-C2e - 3x. 130. y" + y = 0. Ans. y = A c o sx -\-
+ Bsinje. 131. y ”— y ' = 0. Ans. y — Cl -\-C2ex . 132. 0*+12y = 7y'. Ans.
y = Cy*x + C2e4x. 133. y”- 4 y ' + 4y = 0. Ans. y = (Cl + C2x)e2x. 134. / + 2 y '+
+ 10t/ = 0. Ans. y = e~x (/4cos3jc + Æ sin 3je). 135.y”+ ty'~~ 2</ = 0. Ans.
- 3 + V 17 - 3 - V 17 3x
y= C 1e 2 X+ C 2e 2 *. 130. 4y’ - 12y'+9y=0. Ans. y= (C 1+ C *x)e2 .
137. y" + y ’ + y = 0. Ans. y = e *'• ‘ ^r cosY »3. + S s i n j •

138. Two identical loads are suspended from the end of a spring. Find the
motion imparted to one load if the other breaks loose. Ans. x = a cos ( / ? - ) •
where a is the increase in length of the spring under the action of one load at
rest.
139. A material point of mass m is attracted by each of two centres with a
force proportional to the distance. The factor of proportionality is k. The
distance between the centres is 2c. At the initial instant the point lies on the
line connecting the centres at a distance a from the middle. The initial velocity
is zéro. Find the law of motion of the point. Ans. jc = acos
(/#-)
140. yl v -~ 5 /+ 4 y = 0 . Ans. y= C 1ex+ C 2e - x +C3e2x+ C 4e~2x. 141. y " — 2 ÿ —
—j/'+ 2 j/ = 0. Ans. y = C1e2x+ C2ex + C3e~x . 142. y '" — 3ay* + 3aV —a3y = 0.
Ans. y = (Cj -f- C2x -j- C3x2) eax. 143. y ^ —4y//,==0. A/ts. y = Cj-{-C2Jf-f-C3A»2-f-
-FC4e2*+C 5^ 2*• 144. yIV+2j/’r+9^==0. Ans. y= (C x cos Y%x-\-C2 sin Y 2 x )e ~ x+
+ (C3 cos >^2jt-f-C4sin Ÿ 2 x ) e x . 145. yIV—8y*+16y = 0. Ans. y = C1e2x+
X
IV
+ C2e-*x + C3xe2x + Ctx e -2X. 146. y 1v + y = 0. Ans. y = e^ 2(^CXcos ^ = +
x
+ c * sin y i ) + * ^ 2 ( Ca cos y j + c *sin ï % ) •

147. y IV — a4y ~ 0. Find the general solution and pick out a particular
solution that satisfies the initial conditions for jro = 0, y = 1, y' = 0, y”= —a2.
154 Ch. î. Differential Equations
y " ' = 0. Ans. General solution: y = C1eax + C2e” a* + C3 cos ax-{-CAsin ax. Parti-
cular solution: y« = cosax.
Integrate the following nonhomogeneous linear differential équations (find the
general solution):
148. y"—7y' + \2y = x. Ans. y = Cl(?* + Cae«* + 1 + 7- . 149. s”— a2s = t + 1.
Ans. s = C1eai + C2e - at— t ^ . 150. — 2y = 8 sin 2x. Ans. y — C1ex+
+ C2e“2x — ~ (6 sin 2 * + 2 cos 2x). 151. i/'" —y = 5jc-f-2. i4ns. y = Cxe*+ •

5
et
— 5jc—-2. 152. s”— 2as'-\-a2s = et (a ^ 1). Ans. s = C1eat+ C 2teat-\-
163. y" + 6y’ + 5y = eîx . Ans. y = C1e -* + Cîe ~ * * + ^ é >*. 154. y”+ 9y = 6e3x.
Ans. y = C1 cos 3* + Ca sin 3x+-g-«s*. 156. y"—3 y '= 2 —6x. Ans. y = C 1+ C ïea* +
+ x 2. 156. yn— 2y'+3y ==e~x cos x. Ans. y = ex(A cos Ÿ 2 x + B sin y r2jc)+
e~x
-]———(5 cos x —4 sin x ). 157. y"-\-4y = 2 sin 2*. Ans. y = A sin 2x + B cos 2x—
- y C o s 2 * . 158. y ’”— 4 / + 5 / — 2y = 2* + 3. Ans. y = (Ct + C2x) ex + Cse2x—

—x — 4. 159. y ]W—a*y=5a*eax sinax. Ans. y = (Cl — sm a x)eax + C2e - ax-\-
+ C 8 cos ax+Ct sin ax. 160. y i v +2a2yn+a*y=8 cos ax. Ans. y= (C i+ C 2x) cos ax+
x2
+ (C3 + C4*) sin ax — cos ax.
161. Find the intégral curve of the équation y" + k2y = 0 that passes through
the point M (x0, yo)> and is tangent at the point A4 to the straight line y = ax.
Ans. y = y0 cos k (x— x0) + *—sin k (x— jc0).
162. Find à solution of the équation y*-\-2hy'-\- n2y = 0 that satisfies the
conditions y = at y ’ = C when * = 0. Ans. For h < n y = e~hxâcos V^n2— h2x+
H— — - sin "\fn2—h2x \ ; for h = nt y = e~h x [(C-\-ah)x-\-a]\ for h > n,

Y n2— h2 J
C +a (h+ V h 2 — n2) r- ( h - V h * - n » ) x C + a (h — V h 2— n2) r_ (h -V h * -n * ) x
y 2 Ÿh2—n2 2 V 'ï^ n 2
163. Find a solution of the équation y”+ n2y = h sin px {p ^ n) that satisfies
the conditions: y = a, y' = C forx = 0. Ans. y —a cos nx + ^ ^ s*n nx+ *4
164. Â load weighing 4 kg is suspended from a spring and increases the length
of the spring by 1 cm. Find the law of motion of the load if we assume that
the upper end of the spring performs harmonie oscillations under the law
y = sin V^lOO g t, where y is measured vertically.
Solution. Denoting by x the vertical coordinate of the load reckoned from the
position of rest, we hâve
4 d2x
where / is the length of the spring in the free state and £ = 400, as is évident
from the initial conditions. Whence -j^+100gjc==100g sin Y 100gï-fl00 Ig. We

must seek the particular intégral of this equation# in the form
t (Cx cos Y 100g t + C2 sin Y 100g t) + /
since the first term on the right enters into tht solution of the homogeneous
équation.
165. In Problem 139, the initial velocity is u0 and the direction is perpendi-
cular to the straight line connecting the centres. Find the trajectories.
Solution. If for the origin we take the mid-point between the centres, the
differential équations of motion will be
d*x d 2u
m - ^ 2 = k (C— x) — k (C + *) = —2kx, m -jjÇ -= — 2ky
The initial data for t = 0 are

dx dy__
x = a, — = 0 ,y = 0 ,
di ”
* = « cos( Sin ( l / — ^)
Whence (ellipse).
a mvo
166. A horizontal tube is in rotation about a vertical axis with constant
angular velocity co. A sphere inside the tube slides along it without friction.
Find the law of motion of the sphere if at the initial instant it lies on the axis
of rotation and has velocity v0 (along the tube).
dlr
Hint. The differential équation of motion is - ^ - = o ù2r. The initial data are:
r = 0, ^ - = v Q for t = 0. Integrating, we find
Applying the method of variation of parameters, intégrale the following diffe

rential équations:
167. t f - l y ' + S y - s m x . Ans. y = + Cte** + 5 S‘n C° S* ■
168. y" + y = sec x. Ans. y = CXcos * + C2 sin x -fjtsin jc+cos x ln cos x.
169. y*-f-y= - --------- Ans. y = C1 cosx + C2s\n x — Y cos 2x.
cos 2x y cos 2x
Integrate the following Systems of équations:
170. ' W =zy~ ^lt ~ à t= X o u * Pâr^ cuâr sortions that satisfy
the initial conditions x = —2, y = 0 for / = 0. Ans. y = C1et ~\-C2e - t — l,
x = C1et — C2e~ t — 1. Particular solution: x* = —e - f — 1, y* = e~ f— 1.
171. — = x — 2yt ^ = x — y. Pick out the particular solutions that satisfy
the initial conditions: x = \ , y = 1 for t = 0. Ans. y = Ct cos/ + C2 sin/,
x = (Ci H- C2) cos / + (C8—Cx) sin t. Particular solution; ** = cos *—sin t,
y * = cos t.
156 Ch. 1. Differential Equations
. dx du , _ . . i4ns. Jt = C1e-^ + C2^**s^

4 d r ~ a r + 3x=sm< y = ( 7 ^ - * + 3Ca<?-8*+ cos /.
172.
^ - + y = cos /.
d2y
dt'1
i4/L5. x= C j^ -j- -j- C3 cos / ~|“ sin t ^
173. y^C xet + C tf- t — Ca cos / —C4 sin *•
fx
dt 2 = y.
d2Ac . dy .A
57*+rf7+ '*==e i4rts- * = Ct +C ,< + <V 2 - -g- fi + et,
174. y = Ci - { C l + 2 C , ) t - ± ( C i - l ) t * .
d/ ‘ dt2 ~~ 1* -T V + H
dy i4/is. y = (C1+ C2^)e-^,
£ = * -y
175.
\ T x = - y - 3t-
4/is. ^ C ^ + Crf-2*
Î + - ®
176. z= —2 (C ^ 2*—C2<?-2*).
dz_
+ 4^ = 0.
dx
-f 2 y + z= sin ac i4/is. y = Ci + CaAc+ 2 sin ac,

dx
177. z = — 2Ci — C2 (2ac+ 1) —3 sin ac— 2 cos x.
-dz^ — A
4y — O
2z = cos ac.
dAC
Ans. AC= C1e-* + Cae2*,
Tt = y + t
y = C3e~t + Cîe2tf
z — — (Cj + C3) e~ *+ C*z2*.
178. dy .
dt = x +*
dz
d i= x + y -
i4«s. z = Cgecix,
179.
y = x + c k e- c'x-
Ans. — = Clt z y * - Y *a= Cs.
180.
Exercises un Chapter / 157
Intégrale the following different types of équations:

m . y y " = y ‘t + 1. Ans. y = - ^ r |<>c,u-c1>-|-e -c,< * -cJi|. |82. ^ ~ y *,dx = Q.
2Cl {x—yV
Ans. ~ ^ = C. 183* y = xy'2 y ’1. Ans. y = ( V x *\-1 + C ) 2. Singular solutions:
y = 0, x - j- l= 0 . 184. t/*4-y = sec*. Ans. y = C, cos x + C2 sin *4-x sin*4-
4 - cos x ln cos x. 185. (1 4-x2) y' —**/ —a = 0. Ans. y = ax-\-C |/"ï 4 - jc*.
u du U sin“
186. * cos — -j- — y cos — — x. Ans. xe x = C . 187. yn— ±y = e2x sin 2x. Ans.
x dx x
t/ = C je- 2* 4 -C 2e2* —^ - ( s i n 2* 4 - 2 ,cos 2x). 188. xy' 4 - y — y* In * = 0 . Ans.
(ln * -f 1+ C x ) y = I. 189. (2*4-2^— 1)d*-f-(*4-t/ —2)dy = 0. .4ns. 2*4-^—
— 3 ln ( * 4 - y + 1) = C. 190. 3ex tan y dx + ( \ — ex )sec2 y dy = 0. Ans. tan y =
= C(1—e*)a.
Investigate and détermine whether the solution * = 0, y = 0 is stable for
the following systerns of differential équations:
j % = 2 x-3 y
191. Ans. Unstable.
\ 5 = B* + 6»-
'g = -4 x -1 0 ÿ
192. Ans. Stable.
g=12*+18y
193. Ans. Unstable.
= — Sx— \2y.
dt
194. Approximate the solution of the équation y*—y2+ x that satisfies the
initial condition y = 1 when * = 0 . Find the values of the solution for * equal
to 0.1, 0.2, 0.3, 0.4, 0.5. Ans. y x=0.6 = 2.235.
195. Approximate the value of y XsslmA of a solution of the équation y ’-f-
4 - ^ - y = ex that satisfies the initial conditions y = 1 when * = 1. Compare the
resuit obtained with the exact solution.
196. Find the approximate values of x t=1,4 and y t - lt4 of the solutions of
a System of équations -£ = y —x t ~~ = — x — 3y that satisfy the initial condi
tions * = 0, y = 1 when t = 1. Compare the values obtained with the exact
values.
CHAPTER 2
M U L T IP L E IN T E G R A L S
2.1 DOUBLE INTEGRALS

In an x^-plane we consider a closed* domain D bounded by a
line L.
In D let there be given a continuons function
*=/(*> y)
Using arbitrary lines we divide the domain D into n parts
Asj, As,, As8) . . . . As„
(Fig. 43) which we shall call subdomains. So as not to introduce
new symbols we will dénoté by As1( . . . . As„ both the subdomains
and their areas. In each subdomain As,- (it is immaterial whether
in the interior or on the boundary) take a
point P,-; we will then hâve n points:
Pi* P u • • • > P n
We dénoté by HP,), /(/»,), . . . . /(/>„) the
values of the functions at the chosen points
and then form the sum of the products
f(Pi) As,-:
— f(Pl) ASj + / (Pa) As, + . . . + / (Pn) As„
= 2 / ( / >/)As1- ( 1)
^=1
This is the intégral sum of the function
f (x , y) in the domain D.
If f ^ 0 in D, then each term / (P,) As,- may be represented
geometrically as the volume of a small cylinder with base As,-
and altitude /(P,-).
The sum Va is the sum of the volumes of the indicated ele-
mentary cylinders, that is, the volume of a certain “step-like”
solid (Fig. 44).
* A domain D is called closed if it is bounded by a closed line, and the

points lying on the boundary are considered as belonging to D.
2.1 Double Intégrais 159
Consider an arbitrary sequence of intégral sums formed by

means of the function /(*, y) for the given domain D,
Vn,t ynt t •••» ••• (2)
for different ways of partitioning D into subdomains As,-. We shall
assume that the maximum diameter of the subdomains As,- approach-
es zéro as nk —>oo. Then the following proposition, which we
give without proof, holds true.
Theorem 1. I f a function f (x, y) is continuous in a closed domain

D, then the sequence (2) of intégral sums (1) hns a limit if the
maximum diameter of the subdomains As,- approaches zéro as nk—►oo.
This limit is the same for any sequence of type (2), that is, it is
independent either of the way D is partitioned into subdomains As,-
or of the choice of the point Pt inside a subdomain As,-.
This limit is called the double intégral of the function f(x, y)
over D and is denoted by
$$/(/»)ds or S\ f ( x , y)dxdy,
D D
that is,
1im 2 f (Pi) As,- = 1$ f (x, y) dx dy

diam As { -*> 0 i= 1 p
This domain D is called the domain (région) of . intégration.

If f(x, y ) ^ 0, then the double intégral of f(x, y) over D is
equal to the volume of the solid Q bounded by surface z = f(x, y),
the plane z — 0, and a cylindrical surface whose generators are
parallel to the z-axis, while the directrix is the boundary of the
domain D (Fig. 45).
Now consider the following theorems about the double intégral.
160 Ch. 2 Multiple Intégrais
Theorem 2. The double intégral of a sum of two functions

qp(x, y) + r|3(x, y) over a domain D is equal to the sum of the
double intégrais over D of each of the functions taken separately:
S5[<P(*. ÿ) + i|>(x, y)\ds = y)ds+\)y\>(x, y) ds
D D D
Theorem 3. A constant factor may be taken outside the double

intégral sign:
if a = const, then
$$a<p(x, y) ds = a $ $ <p(x, y)ds
D D
The proof of both theorems is exactly the same as that of the

corresponding theorems for the definite intégral (see. Sec. 11.3, Vol. I).
Theorem 4. If a région D is divided into two domains D1 and

without common interior points, and a function f(x, y) is
continuons at ail points of D, then
y ) d x d y = [ \ f ( x , y ) d x d y + \ \ f ( x , y)dxdy (3)
D D, D,
Proof. The intégral sum over D may be given in the form

(Fig. 46)
D D ,
2 f (P,) As, = 2 î (Pd As,-+ 2 / (Pi) As,-
D,
(4)
where the first sum contains ternis that correspond to the subdo
mains of Dx, the second, those corresponding to the subdomains
of D2. Indeed, since the double intégral does not dépend on the
manner of partition, we divide D so that the common bùundary
of the domains D1 and D2 is a boundary of the subdomains As,-.
Passing to the limit in (4) as As,-—>-0, we get (3). This theorem
is obviously true for any number of terms.
2.2 Calcutating Double Intégrais 161
2.2 CALCULATING DOUBLE INTEGRALS

Let a domain D lying in the xy-plane be such that any straight
line parallel to one of the coordinate axes (for example, the y-axis)
and passing through an interior* poinf of the domain, cuts the
boundary of the domain at two points N 1 and (Fig. 47).
In this case we assume that the domain D is bounded by the
Unes: y ^ ^ ^ x ) , y = <p2(x), x = a, x = b and that
<Pi(x)<<P*(x), a < b
while the functions <px(x) and <p2(x) are continuous on the interval
[a, b], We shall call such a domain regular in the y-direction. The
définition is similar for a domain regular in the x-diredion.
A domain that is regular in both x- and y-directions we shall
simply call a regular domain. In Fig. 47 we hâve a regular
domain D.
Let the function f(x, y) be continuous in D.
Consider the expression
b /<r. (*) \
/ z > = $ ( S f(x, y ) d y j d x
a \<j>, (*)
which we shall call a twofold iterated intégral of /(x, y) over D.
In this expression we first calculate the intégral in the parenthèses
(the intégration is performed with respect to y while x is considered
to be constant). The intégration yields a continuous** function of x:
<p, <*)
<D(x)= J /(x, y)dy
<Pi (*)
Integrating this function with respect to x from a to b,
b
ID=^d> (x) dx
a
we get a certain constant.
Example. Calculate the twofold iterated intégral
l ,x* v
/ / > = $ ( $<*• + **)«& )dx
o\ o /
Solution. First calculate the inner intégral (in brackets):
Œ» (x) = J (x2+ y3) dy = ( x * y =a 2 1I *

*4 + t
* An interior point of a domain is one that does not lie on its boundary.
** We do not prove here that the function <D(x) is continuous.
11—2082
Integrating the fonction obtained from 0 to 1, we find

1
XT_ 26
3-7 21 105
Détermine the domain D. Here, D is tbe domain bounded by the Unes (Fig. 48)
f/ = 0, x-— 0, y = x'2, x—l
It may happen that the domain D is such that one of the func-
tions y = y l (x), */=<p2(x) cannot be represented by one analytic
expression over the entire range of x (from x = a to x = b). For

example, let a < c i (*)"=%(*) on the interval [c, b]
where ^(x) and %(x) are analytically specified functions (Fig. 49).
Then the twofold iterated intégral will be written as follows:
b \
W S f(x,y)dy\dx
a X cpt ( x) J
c /<fi(x) \ b /fp * ( x) \
= S( S f(x, y ) d y ) d x + U $ f(x,y)dyjdx
a ( x) y c \ ^ t (x )
C /<T* ( * ) \ b , y t ( x) \
= S( S f ( x , y ) à y ) d x - \ - K $ f(x,y)dy)dx
a \ i|> ( x) J c \ X( x) y
The first of these équations is written on the basis of a familiar

property of the definite intégral, the second, by virtue of the
fact that on the interval [a, c] we hâve q>j (x) = (x), and on
the interval [c, b] we hâve <pj (x) = %(x).
We would also hâve a similar notation for the twofold iterated
intégral if the function f 2(*) were defined by different analytic
exoressions on different subintervals of the interval [a, b].
2.2 Calcutta! ing Double Intégrais 163
Let us establish some properties of a twofold iterated intégral.

Property 1. If a regular y-direction domain D is divided into
two domains Dx and D2 by a straight0 line parallel to the y-axis
or the x-axis, then the twofold iterated intégral ID over D will be
equal to the sum of such intégrais over Dx and D2; that is,
1d ~ I dx+ ^£>a (0
Proof. (a) Let the straight line x ~ c (a < c (x) dx = ^ (x) dx + ^ ® (x) dx

a \< p , (J t) J a a c
c / <Pf U) \ à / ( ? t (x) • \
= $( S f(x, y ) d y ) d x + U J f(x, y ) d y ] d x = I Di + I
(*) (Pl (*) J
(b) Let the straight line y = h divide the domain D into two
regular «/-direction domains Dl and D2 as shown in Fig. 50.
Dénoté by M t and M2 the points of
intersection of the straight line y — h nlf y=<p*(x)
with the boundary L o i D. Dénoté the
abscissas of these points by a2 and bx.
The domain Dx is bounded by con
tinuons lines:
(1) «/ = <M*);
(2) the curve whose
équation we shall conditionally write
in the form
bt b ce
ÿ = <PÎ (*)
Fig. 50
having in view that qpï (*) = <p2(x)
when a a n d when and that
(f>ï( x) = h when a1^ . x ^.b^,
(3) by the straight lines x — a, x = b.
The domain D2 is bounded by the lines
# = <Pi*W» ÿ = <Pj W , where a1< x
* The fact that a part of the boundary of the domain Dx (and of D3) is a
portion of the vertical straight line does not stop this domain from being reg
ular in the ^-direction: for a domain to be regular, it is only necessary that
any vertical straight line passing through an interior point of the domain
should hâve no more than two common points with the boundary.
11 *
We write the identity by applying to the inner intégral the

theorem on partitioning the interval of intégration:
b /(P. <*) \
id = \{ S f(x, y ) d y j d x
a \< P , ( x ) y
b y <Tl (*) <p, (*) \

-J( S f(x, y ) d y + J f(x, y) dyj dx
«' «
r*(*) q* (x) J
We break up the latter intégral into three intégrais and apply

to the outer intégral the theorem on partitioning the interval of
intégration:
since <pj(jc) = <ps (*) on [a, a j and on [&,, b], it follows that the
first and third intégrais are identically zéro. Therefore,
Here, the first intégral is a twofold iterated intégral over Dt,

the second, over D2. Consequently,
ID — ID, +
The proof will be similar for any position of the cutting straight
line If MtM2 divides D into three or a larger number of
domains, we get a relation similar to (1), in the first part of
which we will hâve the appropriate number of ternis.
Corollary. We can again divide each of the domains obtained
(using a straight line parallel to the ÿ-axis or xr-axis) into regular
ÿ-direction domains, and we can apply to them équation (1).
2.2 Calculaiing Double Intégrais 165
Thus, D may be divided by straight Unes parallel to the coor-

dinate axes into any number of regular domains
Dj, D» . . . . Dt
and the assertion that the twofold iterated intégral over D is
equal to tlie sutn of twofold iterated
intégrais over the subdomains holds;
that is (Fig. 51),
I d = ^D, + ^D, + I d, + •••+ 1Dt (2)
Property 2 (Evaluation of an ite

rated intégral). Let m and M be the
least and greatest values of the func-
tion f(x, y) in the domain D. De-
note by S the area of D. Then we
hâve the relation
b / <J, <*)
mS < W J /(•'•'» y ) d y j d x ^ . M S (3)
<fi (*>
Proof. Evaluate the inner intégral denoting it by O(x):
(fi <*) <Ti(■*)
<D(x)= J f(x, y ) d y ^ $ Mày = M [<p2(-v)-^cpl (jc)J
<Fl (X) (*i(•*)
We then hâve
b /<p, (x) \ b
ID= J ( S f(x, y ) d y ) d x ^ \ M [ < f î (x)— (pl {x)]dx= MS
a Ncpt ( * )
that is,
1d ^ M S (3')
Similarly
Qt (x) <f«<*>
(D(x) =
«p» <*)
b
«Ti (x) ■s ■
J f(x, y) d y ^ $ mdy = m [<p2(.v) —<pt (x)]
1D= 5 O (x) dx > 5 m [<p2(x)—«Pi (*)] dx = mS

that is,
I D> m S (3")
From the inequalities (3') and (3") follows the relation (3):
mS ^ I d ^ . M S
In the next section we will détermine the géométrie meaning of
this theorem.
Property 3 (Mean-value theorem). A twofold iterated intégral

I D of a continuons function f(x, y) over a domain D with area S
is equal to the product of the area S by the value of the function
at some point P in D\ that is,
$( S /(*. y ) dyj dx = f ( P ) S (4)

a \ « p , (je) /
Proof. From (3) we obtain

m ^ — / D< A f
The number 1D lies between the greatest and least values of

f(x, y) in D. Due to the continuity of the function / (x, y), at
some point P of D it takes on a value equal to the number — ID\
that is,.
whence
ID= f ( P) S (5)
2.3 CALCULATING DOUBLE INTEGRALS (CONTINUED)

Theorem. The double intégral of a continuous function f (x, y)
over a regular domain D is equal to the twofold iterated intégral
of this function over D; that is, *
b /<P« (JC) \
\ \ f { x , y ) d x d y ^ J( J f(x,y)dy)dx
D a \< p , ( x) J
Proof. Partition the domain D with straight Unes parallel to the

coordinate axes into n regular (rectangular) subdomains:
As2, • • • » Asn
By Property 1 [formula (2)] of the preceding section we hâve
Ip — + Âs, + •. • + lAsn = 2 Âs/ (0

i= 1
We transform each of the terms on the right by the mean-
value theorem for a twofold iterated intégral:
' as. = /(/>,) As,
* Here, we again assume that the domain D is regular in the y-direction
and bounded by the lines y = cpx (*), y ^ (p 2 (jt), x = a, x = b.
2.3 Calculating Double Intégrais (Continued) 167
Then (1) takes the form

h = f {Pù As, + / (/>,) As, + . . . + / (Pn) As„ = 2 f (Pj) As,- (2)
• i =1
where />,- is some point of the subdomain As,-. On the right is the
intégral sum of the function f(x, y) over D. From the existence
theorem of a double intégral it follows that the limit of this
sum, as n —«■oo and as the gréa test diameter of the subdomains
As, approaches zéro, exists and is equal to the double intégral of
f(x, y) over D. The value of the double intégral ID on the left
side of (2) does not dépend on n. Thus, passing to the limit
in (2), we obtain
JD= lim 2 f (Pi) As,- = J $ f (x, y) dxdy
diam As --►0 q
or
$$/(*. y ) d x d y = I D (3)
D
Writing. out in full the expression of the twofold iterated inté

gral Ip, we finally get
&r<F.(*> n
SS/(*. y ) dxdy = \ J f(x, y ) dy\ dx (4)
D a L<P« W J
Note I. For the case where f(x, y ) ^ 0, formula (4) has a pic-
torial géométrie interprétation. Consider a solid bounded by a
surface z = f(x, y), a plane z = 0, and a cylindrical surface whose
generators are parallel to the z-axis and the directrix of which
is the boundary of the région z-f(x.y)
D (Fig. 52). Calculate the vo
lume of this solid V. It has
already been shown that the
volume of this solid is equal
to the double intégral of the
function f(x, y) over the do
main D:
V - H f ( x , y)dxdy (5)
D
Now let us calculate the vo

lume of this solid using the
results of Sec. 12.4, Vol. I, on
the évaluation of the volume of a solid from the areas of parallel
sections (slices). Draw the plane x = const (a < x < b ) that cutsthe
solid. Calculate the area S(x) of the figure obtained in the sec
tion x = const. This figure is a curvilinear trapezoid bounded by
the Unes z = /(x, y) (x = const), z = 0, l/ = «p1(x), y = qp8(x). Hence,
the area can be expressed by the intégral

<r*(*)
S W = S f(x, y)dy (6)
<Tt<*)
Knowing the areas of parallel sections, it is easÿ to find the
volume of the solid:
V= J s W d r
a
of, substituting expression (6) for the afea S W . we S6*

b'Vti*) \
V = l ( S / w y) dy)dx (7)
a V i (x) J
In formulas (5) and (7) the left sides are equal; and so the right
sides are equal too:
b /(Pt <*) \
S J /W y) dxdy= J ( S /W y)dy)dx
D a \<p, (*) /
It is now easy to figure out the géométrie meaning of theorem
ott evaluating a twofold iterated intégral (Property 2, Sec. 2.2):
the volume V of a solid bounded by the surface z = f(x, y), the
dary of the région D, exceeds the volume of a cvlinder with base

area S and altitude m, but is less than the volume of a cylinder
with base area S and altitude M [where m and M are the least
and greatest values of the function z = f(x, y) in the domain D
(Fig. 53)]. This follows from the fact that the twofold iterated
intégral ID is equal to the volume V of this solid.
2.3 Calculaiing Double Intégrais (Continued) 169
Example 1. Evaluate the double intégral J J (4—x2 — y 2) dx dy if the do
main D is bounded by the straight Unes x = 0, v c = lt y = 0, and y = ~ .

Solution. By the formula
V=
]_ 35
3 8
Example 2. Evaluate the double intégral of the function f( x , y) = \ + x + y

over a région bounded by the Unes y = —x, x = y = 2, z = 0 (Fig. 54).
Solution.
= | [( Ÿ l + y f dy
=\{ V ï + Z
i + y V ~ y + £ \d y
3
2y2 , , 2ÿs
3 "l" 4 i ' 5 =S ^ + T
Note 2. Let a regular ^-direction domain D be bounded by the
Unes
*= *= y = c, y = d
and let (Fig. 55).
In this case, obviously,
<</ ’Mirt \
) \ f ( x , y ) d x d y = U J f {x, i f )dx)dy (8)
D c 'ij), (y) J
To evaluate the double intégral we must represent it as a twofold

iterated intégral. As we hâve already seen, this may be done in
two different ways: either by formula (4) or by formula (8).
Depending upon the type of domain D or the integrand in each
spécifie case, we choose one of the formulas to calculate the
double intégral.
170 Ch. 2 M ultiple Intégrais
Example 3. Change the order of intégration in the intégral
Solution. The domain of intégration is bounded by the straight line y = x

and the parabola y = V~x (Fig. 56).
Every straight line parallel to the x-axis cuts the boundary of the domain
at no more than two points; hence, we can compute the intégral by formula (8),
setting
ti(y)=yi> ÿî(y)=y, 1
then
1/ y
Ç / (x, y) dx
0 \ÿ *
y_
Example 4. Evaluate J J e x ds if the domain D is a triangle bounded by
D
the straight lines y = x, y = 0, and x = l (Fig. 57).
Solution. Replace this double intégral by a twofold iterated intégral using
formula (4). [If we used formula (8), we would hâve to integrate the function
y_
e x with respect to x; but this intégral is not expressible in terms of elemen-
tary functions]:
Note 3. If the domain D is not regular either in the x-direction

or the (/-direction (that is, there exist vertical and horizontal
2.3 Calculating Double Intégrais (Continued) 171
straight Unes which, while passing through interior points of the

domain, eut the boundary of the domain at more than two points),
then we cannot represent the double intégral over this domain in
the form of a twofold itérated intégral. R we manage to partition
the irregular domain D into a finite number of regular x-direction
or «/-direction domains Dv Dt, . . . . £>„, then, by evaluating the
double intégral over each of these subdomains by means of the
twofold iterated intégral and adding the results obtained, we get
the sought-for intégral over D.
Fig. 58 is an example of how an irregular domain D may be

divided into three regular subdomains Dlt Dt and D3,
Example 5. Evaluate the double intégral
over a domain D which lies between two squares with centre at the origin and
with sides parallel to the axes of coordinates, if each side of the inner square
Is equal to 2 and that of the outer square is 4 (Fig. 59).
Solution. D is irregular. However, the straight lines x — — 1 and x = \ di-
vide it into four regular subdomains Dlt Da, Da, D4. Therefore,
J J e*+y ds= JJ e*+y ds+ J J e*+y ds+ J J e * + y d s + J J ex+ yds
D D% D| D, d .
Representing each of these intégrais in the form of a twofold iterated intégral,

we find ^ ^
^ e x+yds= J £ J ex + y J a*+.V d# j d*
+ J £ J ex + y + J J ex + y dx
= (e2— e~2) (e -1— e~2) -f (e2—e) (e — e - 1) + (e~1— e~2) (e—e - 1)

+ (e2—a~2) (e2— e) = (e? — e - 3) (e — e - 1) = 4 sinh 3 sinh 1
Note 4. From now on, when writing the twofold iterated intégral
s
r<Pi (X)
1d — \ f(x, y) dy \dx
a l.<Pi (x)
we will drop the brackets containing the inner intégral and will
Write
b q>, (x )
70 = $ S f(x, y)dydx
a <p, (jc)
Here, just as in the case when we hâve brackets, we will consider

that the first intégration is performed with respect to the variable
whose differential is written first, and then with respect to the
variable whose differential is written second. [We note, however,
that this is not the generally accepted practice; in some books
the reverse is done: intégration is performed first with respect to
the variable whose differential is last.*]
2.4 CALCULATING AREAS AND VOLUMES

BY MEANS OF DOUBLE INTEGRALS
1. Volume. As we saw in Sec. 2.1, the volume V of a solid

bounded by a surface z — f{x, y), where f(x, y) is a nonnegative
function, by a plane z = 0 and by a cylindrical surface whose
directrix is the boundary of the domain D and the generators are
parallel to the z-axis, is equal to the double intégral of the func
tion f (x, y) over D:
V = $$/(*, y)ds
D
Example 1 . Calculate the voîume of a solid bounded by the surfaces x = 0,

</ = 0, x + J/ + Z = l , 2 * ± 0 (Fig* 60).
Solution.
(1 - x - y ) d y d x
D
where D is (in Fig. 60) the hatched triangular région in the Xy*plane bounded
by the straight Unes jt = 0, y = 0, and x-{-y= 1. Putting the limits on the double
* The following notation is also sometimes used:

à f<p*
J f ( x ,y ) d y
a L<Pi a cp,
2.4 Calculâting Areas and Volumes by Means of Double Intégrais 173
intégral, we calculate the volume:
y= j | ( \ - x - y ) d y d K = i\) [ ( \ - x ) y - Ç y t ~Xdx = ^ ( l - x ) * d x = ^
oo o o
Thus, K= — cubic units.
o
Note 1. If a solid, the volume of which is being sought, is
bounded above by the surface z = Oi (x, y ) ~^ 0, and below by the
surface 2 = 0 , (x, y) ^ 0, and the domain D is the projection of
both surfaces on the xÿ-plane, then the volume V of this solid is

equal to the différence between the volumes of the two “cylindri-
cal” bodies; the first of these cylindrical bodies has the domain D
for its lower base, and the surface z = Oî (x, y) for its upper base;
the second body also has D for its lower base, and the surface
z = <t»,(x, y) for its upper base (Fig. 61).
Therefore, the volume V is equal to the différence between the
two double intégrais
F = 5 5 o î (x, y)ds— $$<!>, (x, y)ds
D D
or
v = SS D
(*. y )—®i (*. y)] ds (i)
Further, it is easy to prove that formula (1) holds true not
only for the case where Ô, (x , y) and O, (x, y) are nonnegative, but
also where O, (x, y) and 0 2(x, y) are any continuous functions
that satisfy the relationship
« M * . y ) > ® A x , y)
Note 2. If in the domain D the function /(x, y) changes sign,
then we divide the domain into two parts: (1) the subdomain D,
where f(x, y ) ^ 0; (2) the subdomain D2 where f(x, y ) ^ . 0. Sup

pose the subdomains D, and D%are such that the double intégrais
over them exist. Then the intégral over Z)j will, be positive and
equal to the volume of the solid lying above the j«/-plane. The
intégral over will be négative and equal, in absolute value, to
the volume of the solid lying below the j«/-plane. Thus, the in
tégral over D will be expressed as the différence between the cor-
responding volumes.
2. Calculating the area of a plane région. If we form the inté
gral sum of the function f(x, t/) = l over the domain D, then this
sum will be equal to the area S,
S = 2 1• As,-
i= 1
for any mode of partition. Passing to the limit on the right side
of the équation, we get
S=\\dxdy
If D is regular (see, for instance, Fig. 47), then the area will be
expressed by the iterated intégral
&[>,(*> 1
S= \ :=S S\ ddyy |l dx
o Lfl>i (*) J
Performing the intégration in the brackets, we obviously hâve
b
S = S [T»W—Vi(x)]dx
(cf. Sec. 12.1, Vol. I).
Example 2. Calculate the area of a région bounded by the curves
y = 2 — x2, y = x
Solution. Détermine the points of intersection of the given curves (Fig. 62).
At the point of intersection the ordinates are equal; that is,
x = 2—x2
whence
x2+ x — 2 = 0
x\ = — 2
x2= 1
We get two points of intersection: —2), Af2 (l, 1). Hence, the required
area is
2.5 The Double Intégral in Polar Coordinate* 175
2.5 THE DOUBLE INTEGRAL IN POLAR COOR DINATES
Suppose that in a polar coordinate System 0, p, a domain D is

given such that each ray * passing thrdugh an interior point of
the région cuts the boundary of D at no more than two points.
Suppose that D is bounded by the curves p = <l>, (0), p = 2(0) and
the rays 0 = a and 0 = p, where <!>, (0) ^ O, (0) and a < p (Fig. 63).
Again we shall call such a région a regular domain.
Let there be given in D a continuous function of the coordi-

nates 0 and p:
z = F(Q, p)
We divide D in some way into subdomains Aslt As2I . . . , As„.
Form the (intégral) sum
V „ = É f (P*)Asa (1)
k= 1
where Pk is some point in the subdomain Ask.
From the existence theorem of a double intégral it follows that
as the greatest diameter of the subdomain As* approaches zéro,
there exists a limit V of the intégral sum (1). By définition, this
limit V is the double intégral of the function F(0, p) over the
domain D:
V =SjF(0,p)ds (2)
Let us now evaluate this double intégral.
* A ray is any haîf-line issuing from the coordinate origin, that is, from
the pôle P.
Since the limit of the sum is independent of the manner of par-

titioning D into subdomains As», we can divide the domain in a
way that is most convenient. This most convenient (for purposes
of calculation) manner will be to partition the domain by méans
of the rays 0 = 0O, 0= 0lt 0 = 02, . . . , 0 = 0„ (where 0„ = a, 0„ = p,
0O < 0! < 02 < . • • < 0„) and the concentric circles p = p0, p = plt
. . . , p = pm [where p0 is equal to the least value of the function
<D1(0), and pm, to tne greatest value of the function 0 2(0) in the
interval a < 0 < p, p0 < pj < . . . < p j .
Dénoté by As,* the subdomain bounded by the Unes p = p,_l,
p = p,., 0 = 0 *_lt 0 = 0*.
The subdomains As,* will be of three kinds:
(1) those that are not eut by the boundary and lie in D;
(2) those that are not eut by the boundary and lie outside D;
(3) those that are eut by the boundary of D.
The sum of the terms corresponding to the eut subdomains hâve
zéro as their limit when A0*—>-0 and Ap,-—>-0 and for this reason
these terms will be disregarded. The subdomains As,* that lie
outside D do not interest us since they do not enter into the sum.
Thus, the sum may be written as follows:
Vn--= 2 [ 2 F (p,.*) As,-J

*=1 1 1 J
where P,* is an arbitrary point of the subdomain As,*.
The double summation sign here should be understood as mean-
ing that we first perform the summation with respect to the in
dex i, holding k fast (that is, we pick out ail terms that corres
pond to the subdomains lying between two adjacent rays*). The
outer summation sign signifies that we take together ail the sums
obtained in the first summation (that is, we sum with respect to
the index k).
Let us find the expression of the area of the subdomain As,*
that is not eut by the boundary of the domain. It will be equal
to the différence of the areas of the two sectors:
As,* = - ( p , + Ap,f A 0 * - - P?AO* = ( p , + ^ ) Ap,- A0*

or
As,* = p; Ap, A0*, where p,- < p,- < p,- + Ap,-
* Note that in summing with respect to the index i this index will not run
thrpugh ail values from 1 to m, because not ail of the subdomains lying between
the rays 0 = 0# and 0 = 0ft + l belong to D.
2.5 The Double Intégral in Polar Coordinates 177
Thus, the intégral sum will hâve the form*
Vn- i [ S F ( 0 * , p / ) p ; a p,a0*J
where P(0*, p*) is a point of the subdomain As,*. Now take the
factor A0* outside the sign of the inner sum (this is permissible
since it is a common factor for ail the terms of this sum):
y „ = J ] [ Ç f ( 0 ! ,p ; ) p ; a P/] A0 *
Suppose that Ap,-—»0 and A0* remains constant. Then the ex
pression in the brackets will tend to the
intégral
®. (»*)
S F( q;, P) p dp
(eÂ)
Now, assuming that A0*—<-0, we finally
get **
P /<K,(0) \
V = U S F (0,p)prfp rf0 (3)
a Va»! (G) /
Formula (3) is used to compute double intégrais in polar coor

dinates.
If the first intégration is performed with respect to 0 and the
second one to p, then we get the formula (Fig. 64)
P« /û)t (P) \
v - U S F (0, p) d0 ) p dp (3')
Pl (p) J
* We can consider the intégral sum in this form because the limit of the
sum does not dépend on the position of the point inside the subdomain.
** Our dérivation of formula (3) is not rigorous; in deriving this formula we
first let Api approach zéro, leaving A0* constant, and only then made A0*
approach zéro. This does not exactly correspond to the définition of a double
intégral, which we regard as the limit of an intégral sum as the diameters ofthe
subdomains approach zéro (i.e., in the simultaneous approach to zéro of A0*
and Ap|). However, though the proof lacks rigour, the resuit is true [i.e., for
mula (3) is true]. This formula could be rigorously derived by the method
uscd when considering the double intégral in rectangular coordinates. We also
note that this formula will be derived once again in Sec. 2.6 with different rea-
soning (as a particular case of the more general formula for transforming coor
dinates in a double intégral).
I2 2082
Let it be required to compute the double intégral of a function

f(x, y) over a domain D given in rectangular coordinates:
/(*. y) dxdy
D
If D is regular in the polar coordinates 0, p, then the computation
of the given intégral can be reduced to computing the iterated
intégral in polar coordinates.
Indeed, since
x — p cos 0, y = p sin 0
f(x, y) — f [p cos 0, p sin 0] = f (0, p),
it follows that
P /<X>,(0) \
SS/(*. y ) d x d y = U S / [pcos0, psin 0 ]p d p )d 0 (4)
D a \d>t (0) /
Example 1. Compute the volume V o! a solid bounded by the spherical surface
x2+ y2+ z2 = 4a2
and the cylinder
*2+ y2~ Zay = 0
Solution. For the domain of intégration here we can take the base of the
cylinder x2-\-y2— 2ay = 0, that is, a circle with centre at (0, a) and radius a.
The équation of this circle may be
written in the form x2-\-(y— a)2= a2
(Fig. 65).
Fig. 66
We calculate of the required volume V, namely that part which issituated

in the first octant. Then for the domain of intégration we will hâve to take the
semicircle whose boundaries are defined by the équations
* = <Pi(fd = 0, * = <MéO = Vïay—y1
U=0, y = 2a
The integrand is
* = /(*. y )= F "4a* -* » -ÿ *
2.5 The Double Intégral in Polar Coordinates 179
Conséquent 1y
2a /V 2ay-y9 \
^V = J ( J V 4a2 — x2— y2 d x j dy
0 ' 0 '
Transform the intégral obtained to the polar coordinates 0, p:
jc = p c o s 0 , y=psin0
Détermine the limits of intégration. To do so, Write the équation of the
given circle in polar coordinates; since
x2“f~y 2 = pa
y = p sin 0
it follows that
p2 — 2ap sin 0 = 0
p = 2a sin 0
Hence, in polar coordinates (Fig. 66), the boundaries of the domain are defined
by the équations
p = <D1 (0) = O, p = <D2 (0) = 2a sin 0, a = 0, P=y
and the integrand has the form

F(0, p ) = V 4 a 2- p2
Thus, we hâve
n_
2 ! 2a sin 0
(4a?— 2a s," e
H J • ' « " - p ’ prfp U - j [ dô
0 \ ô
JL 71
2 2
_ _ 1 J [(4ai>-4a*sin*0)*^-(4a!!)* /* jd 0 = ^ ! J ( l - c o s 3 0)d0
= 9- o3 (3n—4)
Example 2. Evaluate the Poisson intégral

+ 00
J e-*dx
Solution. First evaluate the intégral l R = ^ J «-**->*dxdy, where the domain
of intégration D is the circle x2-{~y2= R 2 (Fig. 67).

Passing to the polar coordinates 0, p, we obtain
2ji ,R 2/1 R
= e~ptp dp^dQ = — ~ ^ e~pt j d0 = n (1
12 :
180 Ch. 2 M ultiple Intégrais
Now, if vve increase the radius R without bound (that is, if we expand
without limit the domain of intégration), we get the so-called improper iterated
intégral:
2ji / ®
e~ p* p dp dQ = lim d 0 = lim n (1 —e-R1) = n
S ( S * " p,pdp R->cc
0 \a
We shall show that the intégral J J é~x% y* dxdy approaches the limit jt if
D
a domain D' of arbitrary form expands in such manner that finally any point
of the plane is in D' and remains there (we shall conditionally indicate such
an expansion of D' by the relationship D' —►<»).
Let R x and R2 be the least and greatest distances of the boundary of D9

from the origin (Fig. 68).
Since the function e~x*~y* is everywhere greater than zéro, the following
inequalities hold:
e~x ,~ y‘ d x d y < Ii&
D'
or
Rl ) <; J J e~x*~y* dxdy< z ji ( 1—e R 2)
D'
Since as D' — it is obvious that Rx — and R 2 — it follows that

the extreme parts of the inequality tend to the same limit ji. Hence, the
middle terni also approaches this limit; that is,
lim x% y* dxdy = n (5)
D '-* - oo
As a particular instance, let D' be a square with side 2a and centre at the
origin; then
a a
J J e~xl~yt d x d y = J J e~x*~y%dxdy
a a
= J J e-**e~y*dx dy = **e~ y*d x^ dy

y
-a -a
2.5 The Double intégral in Polar Coordinates 181
Now take the factor e~y* outsidè the sign of the inner intégral (this is per-
missible since e~y% does not dépend on the variable of intégration*). Then
J ^ e ~ x*~y*dx dy — J e~y*( ^ e ~ x t d x jd y
D' -a \~ a J
a
Set J e~xi dx = Ba. This is a constant (dépendent only on a)\ therefore,
-a
a a
S 5 e~X*~ÿt dxdy = J e~y%Ba dy = Ba J e~y*dy
D' -fl -fl
But the latter intégral is likewise equal to Ba ( because Çe~x*dx =

\ -a
= J e~yt dy j ; thus,
J J «-*•-** dxdy = BaBa = B l
We pass to the limit In this équation, by making û âpproach ifîfinity (in the
process, D' expands without limit):
r a "|2 r+ » -i2
lim C [ e~x*~y* d x d y = lim B\ = lim = f £ -* 2d * |
D’-+<a a —►co a -> « *1 I JI
D L-fl J L-® J
But, as has been proved [see (5)],
Æ ss D'
dxdy = n
Hence,
0b
^ e~x*dx =n
or
J e ~x2d x = V n
This intégral is frequently encountered in probability theory and in statistics.

We remark that we would not be able to compute this intégral directly (by
means of an indefinitc intégral) because the antiderivative of e~x* is not expfes-
sible in terms of elementary fonctions.
2.6 CHANGE OF VARIABLES IN A DOUBLE INTEGRAL

(GENERAL CASE)
In the xy-plane let there be a domain D bounded by a line L.

Suppose that the coordin^tes x and y are functions of new vari
ables u and v:
* = <P(U, v), y = y(u, v) (1)
Let the functions <p(u, v) and («, v) be single-valued and con
tinuons, and let them hâve continuous dérivatives in some do
main D', which will be defined later on. Then byformulas (1)
to each pair of values u and v there corresponds a unique pair
of values x and y. Further, suppose that the functions (p and
are such that if we give x and y definite values in D, then by
formulas (1) we will find definite values of u and v.
Consider a rectangular coordinate system Ouv (Fig. 69). From

the foregoing it follows that with each point P [x, y) in the xÿ-
plane (Fig. 70) there is uniquely associated a point P' (u, v) in
the üu-plane with coordinates u, v, which are determined by for
mulas (1). The numbers u and v are called curvilinear coordina
tes of the point P.
If in the xÿ-plane a point describes a closed line L bounding
the domain D, then in the «o-plane a corresponding point will
trace out a closed line L‘ bounding a certain domain D’; and to
each point of D' there will correspond a point of D.
Thus, the formulas (1) establish a one-to-one correspondence
between the points of the domains D and D', or, the mapping, by
formulas (1), of D onto D' is said to be one-to-one.
In the domain D' let us consider a line « = const. By formu
las (1) we find that in the jcÿ-plane there will, generally speak-
ing, be a certain curve corresponding to it. In exactly the same
way, to each straight line u = const of the uu-plane there will cor
respond some line in the x^-plane.
2.6 Change of Variables in a Double Intégral 133
Let us divide £>' (using the straight lines u = const and

u = const) into rectangular subdomains (we shall disregard subdo
mains that overlap the boundary of the région D'). Using suit-
able curves, divide D into certain * curvilinear quadrangles
(Fig. 70).
Consider, in the uu-plane, a rectangular subdomain As' bounded
by the straight lines u = const, u + Au = const, u = const, u + Au=
= const, and consider also the curvilinear subdomain As corres-
ponding to it in the xy-plane. We dénoté the areas of these sub
domains by As' and As, respectively. Then, obviously,
As' = Au Au
Générally speaking, the areas As and As' are different.
Suppose in D we hâve a continuous function
z = f(x, y)
To each value of the function z = f( x , y) in D there corres
ponds the very same value of the function z = F( uy v) in D \
where
F(u, u) = /[<p(u, u), i|)(u, u)]
Consider the intégral sums of the function z over D. It is
obvious that we hâve the following équation:
2 / ( * , y) As = 2 F ( u, u) As (2)
Let us compute As, which is the area of the curvilinear quad-
rangle /,1Pi!P3P4 in the xy-plane (see Fig. 70).
We détermine the coordinates of its vertices:
P l(*l- */i). *i = <P(«. v), y1= ÿ(u, v)
P2(x3, y2), x2=■=<p(u + Am, u), t/2= (« + Au, u)
P3(x3, y3), x3-- <p(u + Au, u-f-Au), y3= ÿ ( u + Au, u -f Au) (3)
Pt(x 4, yt), xt = <f(u, u + Au), y, = ÿ(u, u + Au)
When computing the area of the curvilinear quadrangle P 1P 2P3P4
we shall consider the lines P1PÎ, P2P3, P 3P4, P4PX as parallel in
pairs; we shall also replace the incréments of the functions by
corresponding difïerentials. We shall thus ignore infinitesimals of
order higher than the infinitesimals Au, Au. Then formulas (3)
will hâve the form
*i = <P(«, v), »/x= t(«> u)
x3= <p(u, u) + |Ï A u , y.2= ij>(u, ü)+ ^ Am
x3= cp(u, u) + |jA u + ^ A u , U3= q(u, u J - f ^ A u - f ^ A u (3')
x4-cp(u, u) + ^ A u . t/4 = ÿ(u, u) + ^ A u

With these assumptions, the curvilinear quadrangle P ^tP aP t

may be regarded as a parailelogram. Its area As is approxima-
tely equal to the doubled area of the triangle /^P*/** and is found
by the follawing formula of analytic geometry:
As « | {x3— xx) {y3— y3)— (xa— x3) (y3—y 1) \
d
du dv
di|? dif Au Av
du dv
Here, the outer vertical lines indicate that the absolute value of
the déterminant is taken. We introduce the notation
dtp d(p
du dv
cty d\|) = /
du dv
Thus,
As » | / | As' (4)
The déterminant / is called the futtctional déterminant of the
functions <p(u, v) and ip(u, v). It is also called the Jaçobian after
the German mathematician Jacobi.
Equation (4) is only approximate, beçause in the process
of computing the area of As we neglected infinitesimals of higher
order. However, the smaller the dimensions of the subdomains As
and As', the more exact will this équation be. And it becomes
absolutely exact in the limit, when the diameters of the subdo
mains As and As' approach zéro:
As
|/ |= lim As7
diam As'-»0
Let us now apply the équation obtained to an évaluation of

the double intégral. From (2) we can write
2 / ( * . * /) A s « 2 f («. » )|/|A s '
(the intégral sum on the right is extended over the domain D').
Passing to the limit as diam A s'—>0, we get the exact équation
$$ /(* , y)dxdy= l \ F (u, v)\I | du dv (5)

D £>'
2.6 Change of Variables in a Double Intégral 185
This is the formula for transformation of coordinates in a double

intégral. It permits reducing the évaluation of a double intégral
over a domain D to the computation of #a double intégral over a
domain D', which may simplify the prolîlem. A rigorous proof of
this formula was first given by the noted Russian mathematician
M. V. Ostrogradsky.
N ote. The transformation from rectangular coordinates to polar
coordinates considered in the preceding section is a spécial case
of change of variables in a double intégral. Here, « = 0 , o = p:
ac = p cos 0j # = p s in 0
The curve AB(p = p1) in the Acy-plane (Fig. 71) is transformed
into the straight line A'B' in the 0 p-plane (Fig. 72). The curve
DC (p = p2) in the xi/-plane is transformed into the straight line

D'C' in the 0p-plane.
The straight Unes AD and BC in the ACÿ-plane are transformed
into the straight Unes A'D' and B'C’ in the 0 p-plane. The curves
L, and L2 are transformed into the curves L'x and L2.
Let us calculate the Jacobian of a transformation of the Car-
tesian coordinates x and y into the polar coordinates 0 and p:
dx dx_
d6 dp p sin 0 cos 0
= —p sin2 0 —p cos2 0 = —p
~ ày dy p cos 0 sin 0
dS dp
Hence, | / 1= p and therefore

P/KtO) \
\ \ f ( x , y)dxdy^-- $( J F (0, p) p dp ) dQ
D a \0>, (0) J
This was the formula that we derived in the preceding section.

Example. Let it be required to compute the double intégral
^ (y — x)dxdy
D
over the région D in the xy-plane bounded by the straight Unes
1 7 1
y= x + l, ÿ = x —3, y = —— x + 5
It would be difficult to compute this double intégral directly; however, a

simple change of variables permits reducing this intégral to one over a rect
angle whose sides are parallel to the coordi-
nate axes.
Set
u = y —x, v = y - \ ~ ^ x (6 )
Then the straight Unes y = x - \- 1, y = x — 3

will be transformed, respectively, into the
straight Unes u = 1, u = —3 in the wu-plane;
1 7
and the straight Unes y = —
Jt + 5 will be transformed into

7
the straight lines o = —■, v = 5.
Consequently, the given domain D is
transformed into the rectangular domain D'
shown in Fig. 73. It remains to compute the Jacobian of the transformation.
To do this, express x and y in terms of u and v. Solving the System of équations
(6), we obtain
3 , 3 1 .3
x = — 4- “ + 4- t’* y = t u + t v
Consequently,
dx dx 3 3
du dv 4 4
dy_ dy_ 1 3
du dv 4 4
3
and the absolute value of the Jacobian is 111 = -£ • Therefore,
2.7 Computing the Area of a Surface. 187
2.7 COMPUTING THE AREA OF A SURFACE

Let it be required to compute the area#of a surface bounded by
a curve T (Fig. 74); the surface is defined by the équation
z = f(x, y), where the function f(x, y) is continuous and has con-
tinuous partial dérivatives. Dénoté the projection of T on the
xi/-plane by L. Dénoté by D the domain on the xy-plane bounded
by the curve L.
In arbitrary fashion, divide D into n elementary subdomains

As,, As2, . . . , As„. In each subdomain As,-take a point P, (£,-, r),-).
To the point P, there will correspond, on the surface, a point
M t[h, /(!,-, ri,)]
Through Af,- draw a tangent plane to the surface. Its équation
is of the form
z — *i = fx $ h »!/)(*—ïi) + fy(îh 7]i)(y— r]i) ( 1)
(see Sec. 9.6, Vol. I). In this plane, pick out a subdomain Ao,
which is projected onto the xy-plane in the form of a subdomain
As,-. Consider the sum of ail the subdomains Ao,-:
2 Ao,-
i= î
We shall call the limit o of this sum, when the greatest of the
diameters of the subdomains Ao,- approaches zéro, the area of the
surface; that is, by définition we set
n
a= lim V Ao, ( 2)
diam Aa - ->-0 t-_ j
Now let us calculate the area of the surface. Dénoté by y,- the
angle between the tangent plane and the xy-plane. Using a fami-
liar formula of analytic geometry we can write (Fig. 75)
ASi = Aa, cos Y;
or
As;
A<J/ COS y; (3)
The angle y,- is at the same time the angle between the z-axis
and the perpendicular to the plane ( 1). Therefore, by équation
( 1) and the formula of analytic geometry we hâve
l
COS Y; =
V l + f f i l l . r]i ) + f y ( l i , Tl/)
Hence,
A*; = V \ T]d + fyH ln %) AS;
Putting this expression into formula (2 ), we get
lim S V T + f ? ( h , nù + f y d i , h,) Asf

diam As - -+0 i = 1
Since the limit of the intégral sum on the right side of the last équation
is, by définition, the double intégral J J j / ”1
D
we finally get
a=lf D
V i + {%) + ( 0 ) dxdy (4)
This is the formula used to compute the area of the surface
z = f(x>y)-
If the équation of the surface is given in the form
x = n(y, z) or in the form y = %{x, z)
then the corresponding formulas for calculating the surfacé area
are of the form
°=JT D'
+{m)îdydz
a = if V ^ 1 + (SD + { tz) dxdz (3")
D"
where D' and D" are the domains in the xy-plane and the xz-plane
in which the givep surface is projected.
2.7 Computing the Area of a Surface 189
Example 1. Compute the surface area a of the sphere

x2+ y 2 + z2 = R2
Solution. Compute the surface area of the uppeç half of the sphere:
z= y w - x ^ - ÿ *
ÔZ _____________ X _____________
dx Y R * - x 2- y 2
dz_= _________ y
dy~ Y R i —x2—y 2
Hence,
Y l+ { ^ ) + {P y) ~ Y W - x ï - Ÿ = VT2
The domain of intégration is defined by thç condition
x2-\-y2<sR2
Thus, by formula (4) we will hâve
p [ Vr*~X*
_L o = R
2 I I
-R \ - V Td
Y R1—*1— y2
dx
To compute the double intégral obtained let us make the transformation

t o polar coordinates. In polar coordinates the boundary of the domain of inté
gration is determined by the équation p = R. Hence,
2jl f *
o = 2j l | Y Y — 2 p dp ) dQ= 2R i [— V R 2— p2Jo do
o 'b P ' b
2jt
= 2R ^ R dd = 4jx/?a
Example 2. Find the area of that part of the surface of the cylinder
x2-\-y2= a2
which is eut out by the cylinder
X2+ Z 2 = ü2
Solution. Fig. 77 shows 1/ 8 th of the desired surface. The équation of the
surface has the form y = \ r a2—x2\ therefore,
ày__ x
dx y a2—x2 ’ dz
x2
The domain of intégration is a quarter of the circle, that is, it is determined
by the conditions
x2 + z2< t a2, jcô » 2 ^ 0
Consequently,
a / V a*-x* Va* - x*
J ô
a = 8aa
2.8 THE DENSITY DISTRIBUTION OF MATTER

AND THE DOUBLE INTEGRAL
In a domain D, let a certain substance be distributed in such

manner that there is a definite amount per unit area of D. We
shall henceforward speak of the distribution of mass, although our
reasoning will hold also for the case when speaking of the distri
bution of electric charge, quantity of heat, and so forth.
We consider an arbitrary subdomain As of the domain D. Let
the mass of substance associated with this given subdomain be Am.
A/n
Then the ratio -gj is called the mean surface density of the sub
stance in the subdomain As.
Now let the subdomain As decrease and contract to the point
P (x , y). Consider the limit lim ^ . If this limit exists, then, ge-
AS-.0 as
nerally speaking, it will dépend on the position of the point P,
that is, upon its coordinates x and y, and will be some function
f(P) of the point P. We shall call this limit the surface density
of the substance at the point P:
} s-►
A i mnT
O ^X = f ( P ) = f ( x .y ) (i)
Thus, the surface density is a function f (x, y) of the coordinates
of the point of the domain.
2.8 The Démit y Distribution of Mat ter and the Double Intégral 191
Conversely, in a domain D, let the surface density of some sub

stance be given as a continuous function f(P) = f(x, y) and let it
be required to détermine the total quantity of substance M contained
in D. Divide D into subdomains As,- (t=^l, 2, . . n) and in each
subdomain take a point P,-; then /(P ,) is the surface density at
the point P,-.
To within higher-order infinitesimals, the product f (P,-) As/ gives
us the quantity of substance contained in the subdomain As,-, and
the sum
2 f (Pi) As,
expresses approximately the total quantity of substance distributed

in the domain D. But this is the intégral sum of the function
/(P ) in D. The exact value is obtained in the limit as As,-—»0.
Thus,*
M = lim S f ( P ,) A s ,= S S /( P ) d s = J S /( x t y)rf*dy (2 )
Asf -*0 i= 1 D D
or the total quantity of substance in D is equal to the double

intégral (over D) of the density f(P) = f(x, y) of this substance.
Example. Détermine the mass of a circular lamina of radius R il the surface
density / (x , y) of the material at each point P (x , y) is proportional to the
distance of the point (x , y) from the centre of the circle, that is, if
f ( x f y) = k }r x2 + y2
Solution. By formula (2) we hâve
M = J J k V x 2 + y 2dxdy
D
where the domain of intégration D is the circle *2 + y2 ^ R2.

Passing to polar coordinates, we obtain
R
M= k dQ = k2n = Y knRi
0
2.9 THE MOMENT OF INERTIA OF THE AREA OF A PLANE FIGURE

The moment of inertia I of a material point M of mass m re
lative to some point 0 is the product of the mass m by the square
of its distance r from the point O:
I = mr2
* The relationship As/ —►0 is to be understood in the sense that the dia-
meter of the subdomain As/ approaches zéro.
The moment of inertia of a System of material points m„ mit

. . . . mn relative to 0 is the sum of moments of inertia of the
individual points of the System:
I=
2 tntf
i=1
Let us détermine the moment of inertia of a material plane
figure D.
Let D be located in an jn/-coordinate plane. Let us détermine
the moment of inertia of this figure relative to the origin, assuming
that the surface density is everywhere equal
to unity.
Divide the domain D into elementary
subdomains 1, 2, . . . , n) (Fig. 78).
Im each subdomain take a point P( with
coordinates rj,-. Let us call the pro-
duct of the mass of the subdomain AS,*
by the square of the distance rï = tî + T]î
an elementary moment of inertia A/, of
Fig. 78 the subdomain AS,-:
A//=(g? + »tf) AS,
and let us form the sum of such moments:
J j ( « + n » AS,-
This is the intégral sum of the function f (x, ÿ) = xiJr yi over the
domain D.
We define the moment of inertia of the figure D as the limit
of this sum when the diameter of each elementary subdomain AS,
approaches zéro:
/„ = Hm £ ( a + ttf)A S,
diam AS . - + Q i - 1
But the limit of this sum is the double intégral ^ (x2+ y2)dxdy.
D
Thus, the moment of inertia of the figure D relative to the origin is
/ o = H t xt + tf)àxdV ( 1)
D
where D is a domain which coïncides with the given plane figure.

The intégrais
IxX= \ \ y i dxdy (2 )
hv^W ^dxdy (3)

2.9 The Moment of Inertia of the Area of a Plane Figure 193
are called, respectively, the moments of inertia of the figure D

relative to the x-axis and y-axis.
Example 1. Compute the moment of inertia #of the area of a circle D of
radius R relative to the centre 0.
/ # = S$ (* * + il*)dxdy
D
To evaluate this intégral we change to the polar coordinates 0, p. The équation

of the circle in polar coordinates is p = R. Therefore
Note. If the surface density y is not equal to unity, but is some

function of x and y, i.e., y = v(*, y). then the mass of the sub
domain AS( will, to within infinitesimals of higher order, be equal to
y(|,-, qf) AS; and, for this reason, the moment of inertia of the
plane figure relative to the origin will be
7»= JS Y(x. y)(x2+y*)dxdy (1 ')
D
Example 2. Compute the moment of inertia of a plane material figure D

bounded by the lines y2 — \ — x\ x = 0 , y = 0 relative to the y-axis if the sur
face density at each point is equal to y
(Fig. 79).
Solution.
fVT^x
/W = [( | y x * d y ) d x = { x^ ~ d* = Ÿ f x2 ( l - x ) d x = —
o\ o / o o 0
Ellipse of inertia. Let us détermine the moment of inertia of
the area of a plane figure D relative to some axis OL that passes
I.) -2082
through the point 0 , which we shall take as the coordinate origin.

Dénoté by <p the angle formed by the straight line OL with the
positive x-axis (Fig. 80).
The normal équation of OL is
xsinqp—ycosqp = 0
The distance r of sonie point M (x, y) from this line is
r = | x sin <p—y cos <p|
The moment of inertia / of the area of D relative to OL is
expressed, by définition, by the intégral
/= r*dxdy = (xsinq)—ycos<p)*dxdy
D D
= sin2<p JJ x*dxdy— 2 sin<pcos<p J J xydxdy -f-cos2qp JJ y2dxdy

D D D
Therefore
I = Iyy sin 2 <p— 2 Ixy sin qpcos cp+ Ixx cos2 «P (4)
here, I vv = J J x2 dxdy is the moment of inertia of the figure relative
D
to the y-axis, I * * = $ $ ÿ 2 dxdy is the moment of inertia relative
D
to the x-axis, and Ixy= \ \ xydxdy. Dividing ail terms of the last
D
équation by I, we get
___î y __
1= L / E 2/ / cos f \ _L / /sinjpX2
\ v t ) ^, x v K v r ) { r i ) + vv W t ) (5)
On the straight line OL take a point A (X , Y) such that
1
OA
V T
To the various directions of the OL-axis, that is, to various values

of the angle qp, there correspond different values / and different
points A. Let us find the locus of the points A. Obviously,
^ 1 1
cosqp, sinq)
vr VT
By virtue of (5), the quantities X and Y are connected by the
relation
1 = / mX2 - 2 / ^ + / / ‘ (6 )
Thus, the locus of points A (X , Y) is a second-degree curve (6 ).
We shall prove that this curve is an ellipse.
2.9 The Moment of Inertia of the Area of a Plane Figure 195
The following inequality established by the Russian mathema-

tician Bunyakovsky * holds true:
(yx y d^x d y \ <

$ $ * 2 dxdy'j Ç)j '\j yl dxdy
or
Ixxlyy 1xy 0
Thus, the discriminant of the curve
(6 ) is positive and, consequently, the
curve is an ellipse (Fig. 81). This ellipse
js called the ellipse of inertia. The
notion of an ellipse of inertia is very pig gj
important in meehanics.
We note that the lengths of the axes of the ellipse of inertia
and its position in the plane dépend on the shape of the given
plane figure. Since the distance from the origin to some point A
* To prove Bunyakovsky’s (also spelt Buniakowski) inequality, we con-
sider the following obvions inequality:
SS[/ (.X, y) —À<p (xt y )]2 d x d y ^ z 0

D
where % is a constant. The equality sign is possible only when f ( x t y) —

— X<p(x, t/) = 0; that is, if f (x, y) = h(p(x, y). If we assume that • ^
9c const = X, then the inequality sign will always hold. Thus, removing brackets
under the intégral sign, we obtain
/*(•*. y ) d x d y — 2k JJ f(x, y)<p(x, y ) d x d y + k‘ i JJ y )dxdy > 0
Consider the expression on the left as a fonction of X. This is a second-degree
polynomial that never vanishes; hence, its roots are complex, and this will occur
whën the discriminant formed of the coefficients of the quadratic polynomial
is négative, that is,
f<p dx dyV — f 2 dx dy J J <p2 dx dy < 0
D J D D
or
f 2 dx dy J J cp2 dx dy
D
This is Bunyakovsky*s inequality.

x
In our case, f (xt y) = x t <p(*, y) = y , — ?= const.
Bunyakovsky’s inequality is widely used in various fields of mathematics.
In many textbooks it is called Schwarz’ inequality. Bunyakovsky published it
(among other important inequalities) in 1859. Schwarz published his work 16
years later, in 1875.
of the ellipse is equal to where I is the moment of inertia

VT
of the figure relative to the (M-axis, it follows that, after cons-
tructing the ellipse, we can readily calculate the moment of
inertia of the figure D relative to some straight line passing through
the coordinate origin. In particular, it is easy to see that the
moment of inertia of the figure will be least relative to the major
axis of the ellipse of inertia and greatest relative to the minor
axis of this ellipse.
2.10 THE COORDINATES OF THE CENTRE OF GRAVITY

OF THE AREA OF A PLANE FIGURE
In Sec. 12.8, Vol. I, it was stated that the coordinates of the
centre of gravity of a System of material points Plt P......... . Pn
with masses mlt mt, . . . , m„ are defined by the formulas
2 */«i
" v “— « = - s ----- ( 1)
fTl( ^ TTl£
Let us now détermine the coordinates of the centre of gravity
of a plane figure D. Divide this figure into very smali subdo
mains AS,-. If the surface density is taken equal to unity, then
the mass of a subdomain will be equal to its area. If it is ap-
proximately taken that the entire mass of subdomain AS,- is con-
centrated in some point of it, q,-), the figure D may be
regarded as a System of material points. Then, by formulas (1),
the coordinates of the centre of gravity of this figure will be
approximately determined by the équations
n
2
i=\
É/AS, 2 T),AS,
n * yc
2 A5<- 2 AS<
-
f=i ï =i
In the limit, as AS, —>-0, the intégral sums in the numerators
and denominators of the fractions will pass into double intégrais,
and we will obtain exact formulas for computing the coordinates
of the centre of gravity of a plane figure:
^xdxdy ^ydxdy
xc = T c-------* (2)
SSdxdy
D
SS<fc4y
D
These formulas, which hâve been derived for a plane figure with
2.11 Triple Intégrait 197
surface density 1, obviously hold true also for a figure with any
other density y constant at ail points.
If, however, the surface density is variable,
y = y (* . y) '
then the correspondis formulas will hâve the form
^ y(*> y ) x d x d y JSy(*. y )y dxdy
y __Ç___________ J D ______________
XC ff * yc
y (*. y)**dy SSyc*. v)*1* dy
D
The expressions My = J 5 Y(*» y) x d x d y and Mx = 5 5 Y(*» y) y d x d y

D D
are called static moments of the plane
figure D relative to the y-axis and x-axis.
The intégral 55 YC** ÿ)dxdy expresses
the quantity of mass of the figure in
question.
Example. Détermine the coordinates of the cen
tre of gravity of a quarter of the ellipse (Fig. 82)
+ fr2
assuming that the surface density at ail points is equal to 1.
Solution. By formulas (2) we hâve
2.11 TRIPLE INTEGRALS

Let there be given, in space, a certain domain V boundtd by
a closed surface S. Let some continuous function / (x , y, z), where
x, y, z are the rectangular coordinates of a point of the domain,
be given in V and on its boundary. For clarity, if f(x, y, z)~ ^0,
we can regard this fonction as the density distribution of some

substance in the domain V.
Divide V, in arbitrary fashion, into subdomains Ai»,-; the Sym
bol Au,- will dénoté not only the domain itself, but its volume as
well. Within the limits of each subdomain Ao,-, choose an arbitrary
point Pj and dénoté by f(P t) the value of the function f at this
point. Form a sum of the type
S 0 )
and increase without bound the number of subdomains Av,- so that
the largest diameter of Au, should approach zéro.* If the function
f (x, y , z) is continuous, sums of type (1) will hâve a limit. This
limit is to be understood in the same sense as for the définition
of the double intégral.** It is not dépendent either on the manner
of partitioningthedomainForonthe choiceof points P,-; itisdesignated
by the symbol and is called a triple intégral. Thus,
v
by définition,
lim 2 / (Pt) Au,- = JJ J f(P) dv
diam -> 0 y
or
y< *)dxdydz (2 )
V V
If / (x , y, z) is considered the volume density of distribution of

a substance over the domain V, then the intégral (2 ) yields the
mass of the entire substance contained in V.
2.12 EVALDAT1NO A T R IP L E INTEG RAL

Suppose that a spatial (three-dimensional) domain V bounded
by a closed surface 5 possesses the following properties:
( 1) every straight line parallel to the z-axis and dràwn through
an interior (that is, not lying on the boundary S) point of the
domain V cuts the surface 5 at two points;
(2 ) the entire domain V is projected on the xÿ-plane into a
regular (two-dimensional) domain D;
(3) any part of the domain V eut off by a plane parallel to any
one of the coordinate planes (Oxy, Oxz, Qyz) likewise possesses
Properties 1 and 2.
* The diameter of a subdomain Au,- is the maximum distance between points
lying on the boundary of the subdomain.
** This theorem of the existence of a limit of intégral sums (that is, of
the existence of a triple intégral) for any function continuous in a closed
domain V (including the boundary) is accepted without proof.
2.12 Evaluai ing a Triple Intégral 199
We shall call the domain V that possesses the indicated proper-

ties a regular three-dimensional domain.
To illustrate, an ellipsoid, a rectangular parallelepiped, a tet-
rahedron, and so on are examples of «regular three-dimensional
domains. An instance of an irregular three-dimensional domain is
given in Fig. 83. In this section we will consider only regular
Let the surface bounding V below hâve the équation z = %(x, y),
and the surface bounding this domain above, the équation z = \|>(x, y)
(Fig. 84).
We introduce the concept of a threefold iterated intégral I v,
over the domain F, of a function of three variables f(x, y, z)
defined and continuous in V. Suppose that the domain D is the
projection of the domain V onto the xz/-plane bounded by the curves
y = <Pi(x)> y=--V%(x), x = a, x=b
Then a threefold iterated intégral of the function / (x, y, z) over
V is defined as follows:
>>r<p. (•*) (*• v) 'v1
r= S S S /(*. «/• z ) d z \d y \ dx (D
a L<P. (X) VX (X. y) J J
We note that as a resuit of intégration with respect to z and
substitution of limits in the braces (inner brackets) we get a func
tion of x and y. We then compute the double intégral of this
function over the domain D as has already been done.
The following is an example of the évaluation of a three
fold iterated intégral.
Example 1. Compute the threefold iterated intégral of the function

f( x , y , z) = xyz over the domain V bounded by the planes
* = 0, y = 0, z = 0, x + y + z = 1
Solution. This domain is regular, it is bounded above and below by the
planes 2 = 0 and z — \ —x — y and is projected
on the jcy-plane into a regular plane domain D,
which is a triangle bounded by tne straight lines
x = 0, y — 0, y = l —x (Fig. 85). Therefore, the
threefold iterated intégral f v is computed as fol-
lows:
ri-x -y -1
D L 0
S **** JU
Setting up the limits in the twofold iterated in
tégral over the domain D, we obtaln
1 /I -* r 1-x- y n 1 /!-* z - 1-x-y v
-ÎÜU ... d ,Y
= 1 { Ï - * » ( i - * - » ) i ^ | * = | è ( 1- x)*d*= 735
Let us now consider some of the properties of a threefold iterated

intégral.
Property I. If a domain V is divided into two domains V1 and
V2 by a plane parallel to one of the coordinate planes, then the
threefold iterated intégral over V is equal to the sum of the three
fold iterated intégrais over the domains V, and VV
The proof of this property is exactly the same as that for
twofold iterated intégrais. We shall not repeat it.
Corollary. For any kind of partition of the domain V into a
finite number of subdomains V,, . . . , Vn by planes parallel to the
coordinate planes, we hâve the equality
h = h t+ h, + •••+ Ivn
Property 2 (Theorem on the évaluation of a threefold iterated
Intégral). If m and M are, respectiwly, the smallest and largest
values of the function f (x, y, z) in the domain V, we hâve the
inequality
m V Îv^M V
where V is the volume of the given domain and I Y is a threefold

iterated intégral of the function f(x, y, z) over V.
2.12 Evaluâting a Triple Intégral 201
Proof. Let us first evaluate the inside intégral in the iterated

(■+ <*. u) "J
intégral I v = $ J $ f(x, y, z) dz day
o Lx (*. y)
$ (x , y) (x , y) (x , y) $ ( x, y )
$ /(*, y y z )d z < 5 Mdz = M J dz = Afz

t ( x, y ) % (x , y) x (*. y) X (x , y)
= y)— X(x, y)]

Thus, the inside intégral dœs not exceed the expression M [\|3 (x, y)—
— X(x, {/)]• Therefore, by virtue of the theorem of Sec. 2.1 on
double intégrais, we get (denoting by D the projection of the
domain V on the xy-plane)
[■*<*.»> "J
/y=SS S /(*. y> z) dz Uo < JJM £tp(x, y)— %(x, y)]do
D L X ( x. U) J D
= y )— %(x, y)] do
But the latter iterated intégral is equal to the double intégral of

the function ip ( jc , y)—%(x, y) and, consequently, is equal to the
volume of the domain which lies between the surfaces z = %(x, y)
and z = i|)(jc, y), that is, to the volume of the domain V. There
fore,
Iy^M V
It is similarly proved that I v ~^mV. Property 2 is thus proved.
Property 3 (Mean-value theorem). The threefold iterated intég
ral l v of a continuous function f (x, y, z) over a domain V is equal
to the product of its volume V by the value of the function at some
point P of V; that is,
* r<p« (jc) f ♦ (j. v)
*v = SIS
a Wi W L ï (*. J)
S /(*• ^ z) dz d y \d x = f (P) V 2
( )
The proof of this property is carried out in the same way as that
for a twofold iterated intégral [see Sec. 2.2, Property 3, formula (4)].
We can now prove the theorem for evaluating a triple intégral.
Theorem. The triple intégral of a function f (x, y, z) over a
regular domain V is equal to a threefold iterated intégral over the
same domain; that is,
b (<ft (*) r 'J1(j. y)
SSS/(*. y, *)à» = \ | J J f{x, y, z)dz dy\ dx
V a 1<P, ( x) l X (*• y)
Proof. Divide the domain V by planes parallel to the coordi-

nate planes into n regular subdomains:
Aülf Ao2, . . . . Avn
As done above, dénoté by I v the threeîold iterated intégral of
the function f {x, y , z) over the domain V, and by /a0/ the three-
fold iterated intégral of this function over the subdomain Avt. Then
by the corollary of Property 1 we can write the équation
/y = /au, “I- l A», 4" • • • “1“ IàVa (3)

We transform each of the terms on the right by formula (2 ):
Iv = f(P i)à v 1+ f( P J A v t + . . . + f(P n)Av„ (4)
where Pt is some point of the subdomain Au,-.
On the right side of this équation is an intégral sum. It is
assumed that the function f(x, y, z) is continuous in V\ and for
this reason the limit of this sum, as the largest diameter of Au,-
approaches zéro, exists and is equal to the triple intégral of the
function f(x, y, z) over V. Thus, passing to the limit in (4), as
diam Au,-—>-0, we get
/ f = SSS /(*. y> *)dv
V
or, finally, interchanging the expressions on the right and left,

à ( <P* (<*) r ^ ( x, y ) "| \
SSS f(* ’ y> z)dv= S{ S S f(x, y, z ) d z \ d y \d x

V a v <p, (*) L x (*. y) J )
Thus, the theorem is proved.

Here, z —i(x , y) and z = if(x, y) are the équations of the sur
faces bounding the regular domain V below and above. The Unes
j/ = q>i (x), ÿ = <p2 (x), x — a, x = b bound the domain D, which is
the projection of V onto the xy-plane.
Note. As in the case of the double intégral, we can form a three-
fold iterated intégral with a different order of intégration with
respect to the variables and with other limits, if, of course, the
shape of the domain V permits this.
Computing the volume of a solid by means of a threefold ite
rated intégral. If the integrand f(x, y , z ) = l , then the triple
intégral over the domain V expresses the volume of V:
V= 555 dxdydz (5)

2.12 Evaluating a Triple Intégral 203
Example 2 . Compute the volume of the ellipsoid

fl , fl , il_ ,
a2 ~'~ b2 ' c2 1 #
Solution. The ellipsoid (Fig. 86) is bounded below by the surface z =
/
JC2 |/2
1 —^2 —^2 ’ and above by the
surface z = c - £ - £ • T hepr°-
jection of this ellipsoid on the jct/-plane
y2 *#2
(domain D) is an ellipse, 1.
Hence, reducing the computation of vo
lume to that of a threefold iterated
intégral, we obtain 1
Fig. 86
a bV i - £ / cV
= ç
j
i - £ -
dz dy dx
- a
- 6
V \-y
a r ‘ / - 7
= 2c
I
- a
b2 dy dx
. - 6 / » - ! -
When computing the inside intégral, x is held constant. Make the substitution:
y = b ]/~ ï^ ~ sin t. dy = b ] / l - - ^ - c o s / c f t
The variable y varies from —b Y 1- i . o b 1 —^ ; therefore t varies
from — to - y . Putting new limits in the intégral, we get
-n
2
dx
I / ( ,--à -H ,- 5 ) sinî<b 1~S’c°s1 dt
1
JL
2
je2) djc=
- “ i o - £ ) i “■*'dt
Hence,
V = -|- nabc
If = we get thc volume of the sphere:
v=î rf
2.13 CHANGE OF VARIABLES IN A TRIPLE INTEGRAL

1. Triple intégral in cylindrical coordinates. In the case of cy-
lindrical coordinates, the position of a point P in space is deter-
mined by three numbers 0 , p, z, where 0 and p are polar coordinates
of the projection of the point P on the xy-plane and z is the
Fig. 87 Fig. 88
plane—with the plus sign if the point lies above the xy-plane,
and with the minus sign if below the xy-plane (Fig. 87).
In this case, we divide the given three-dimensional domain V
into elementary volumes by the coordinate surfaces 0 = 0,-, p = P/,
z = zk (half-planes adjoining the z-axis, circular cylinders whose
axis coincides with the z-axis, planes perpendicular to the z-axis).
The curvilinear “prism” shown in Fig. 88 is a volume element.
The base area of this prism is equal, to within infinitesimals of
higher order, to pA0Ap, the altitude is Az (to simplify notation
we drop the indices i, /, k). Thus, Ai>= pA0 ApAz. Hence, the
triple intégral of the function F ( 0 , p, z) over the domain V has
the form
/ = $ $ $ F ( 0 , p, z)pdQdpdz ( 1)
v
The limits of intégration are determined by the shape of the
domain V.
2.13 Change of Variables in a Triple Intégral 205
If a triple intégral of the function f(x, y , z) is given in rectan-

gular coordinates, it can readily be changed to a triple intégral
in cylindrical coordinates. Indeed, notiijg that
x = p cos 0 , */=-psin 0 , z —z
we hâve
$$$!(*, y, z)d xd yd z= $ JJ F (B, p, z)pdQdpdz
v v
where
/(p c o s 0 , p s in 0 , z)--F(b, p, z)
Example. Détermine the mass M of a hemisphere of radius R with centre
at the origin, if the density F of its substance at each point (x, y> z) is pro-
portional to the distance of this point from the base, that is, F ^ k z .
Solution. The équation of the upper part of the hemisphere
z = Ÿ~R2—x2— y 2
in cylindrical coordinates has the form
z= V > -p a
Hence,
2n “R / V R * - P* v “1
M = J J J kzp dS d p d z = ^ S d z j p dp J de
J V ' 0
2n f R va* —p* 2nr R “|
-S Lo
îï 0
P dp
0 L0 J
de
k_ k_R* knR*
2 2 4 4
2. Triple intégral in spherical coordinates. In spherical coor

dinates, the position of a point P in space is determined by three
numbers, 0 , r, <p, where r is the distance of the point from the
origin, the so-called radius vector of the point, <p is the angle
between the radius vector and the z-axis, 0 is the angle between
the projection of the radius vector on the xy-plane and the x-axis
reckoned from this axis in a positive sense (counterclockwise)
(Fig. 89). For any point of space we hâve
0 ^r< o o , 0 ^C<p ^ it; 0 ^ 0 ^ 2it
Divide the domain V into volume éléments At» by the coordi-
nate surfaces r = const (spheres), <p= const (conic surfaces with ver-
tices at origin), 0 = const (half-planes passing through the z-axis).
To within infinitesimals of higher order, the volume element At»
may be considered a parallelepiped with edges of length Ar, rAqj,
t sin drd0 d<p (T)
Jv
The limits of intégration are determined by the shape of the
domain V. From Fig. 89 it is easy to establish the expressions of
Cartesian coordinates in terms of spherical coordinates:

x = r sin q>cos 0
t/ = r sin f sin 0
z = r cos sin 0 , r cosq>] r 2 sin qpdr d0 dq)

v
3. G en eral ch a n g e of v a ria b le s in a trip le in té g r a l. Transforma
tions from Cartesian coordinates to cylindrical and spherical coor
dinates in a triple intégral represent spécial cases of the general
transformation of coordinates in space.
Let the functions
x = <p(u, t, w)
y = ÿ(K, /, w)
z = X(«. t, w)
2.14 Moment of Inertie and Coordinates of Centre of Gravit y of a Solid 207
map, in one-to-one manner, a domain V in Cartesian coordinates

x, y , z onto a domain V' in curvilinear coordinates u, t, w. Let
the volume element Av of V be carried over to the volume element
Av' of V and let
lim = |/ 1
Then
$$$/(*, y . z)dxdydz
V
= t, W), •»!>(«, t, w), x (u, t , w )]\I\d u d td w

V’
As in the case of the double intégral, / is called the Jacobian;

and as in the case of double intégrais, it may be proved that the
Jacobian is numerically equal to a déterminant of order three:
dx dx dx
du dt dw
dy dy dy
du dt dw
dz dz dz
du dt dw
Thus, in the case of cylindrical coordinates we hâve
x= pcos 0 , t/ = p s in 0 , z —z (p = «, 0 = /, z = w)
cos 0 —p s ïn 0 0
I sin 0 pcos 0 0 = P
0 0 1
In the case of spherical coordinates we hâve

x —r sin q>cos 0 , y = r sinq>sin0 , z —r coscos 0 r cos (p cos 0 —r sin 9 sin 0
/= sin q>sin 0 r cos 9 sin 0 r sin 9 cos 0 = r 4 sin 9
cosep —r s in 9 0
2.14 THE MOMENT OF INERTIA AND THE COORDINATES

OF THE CENTRE OF GRAVITY OF A SOLID
1. The moment of inertia of a solid. The moments of inertia

of a point M (x, y, z) of mass m relative to the coordinate axes
Ox, Oy and Oz (Fig. 91) are expressed, respectively, by the formulas
= + z‘) m> = (* * + y2) m

The moments of inertia of a solid are expressed by the corre-

sponding intégrais. For instance, the moment of inertia of a
solid relative to the z-axis is expressed by the intégral I zz —
— $ $ $ (jc* + y2) y (x, y, z)dxdydz, where y(x, y, z) is the density
of the substance.
Example 1. Compute the moment of inertia of a right circular cylinder of
altitude 2h and radius R relative to the diameter of its médian section, consi-
dering the density constant and equal to y0.
Solution. Choose a coordinate System as
follows: direct the z-axis along the axis of
the cylinder, and put the origin of coordi-
nates at its centre of symmetry (Fig. 92).
Fig. 91
Then the problem reduces to computing the moment of inertia of the cy
linder relative to the *-axis:
Ixx —S
j SS (y2+ z 2) Yodxdydz
Changing to cylindrical coordinates, we ôbtain
2ji /R r h -i \
t= Yo S ! S S ^ + e 231"40) ^ p *p
231 (R \
= Yo j ^ j^ ^ - + 2/ip2 sin2 eJ pdp | d0 = Yo ^ j
0 10 o
[2h9R* 2hR*
= Yo[
2. The coordinates of the centre of gravity of a solid. Like what

we had in Sec. 12.8, Vol. I, for plane figures, the coordinates of
2.15 Computing Intégrais Dépendent on a Parameter 209
the centre of gravity of a solid are expressed by the formulas

5 SS y. «) dx dUdz J 5 5 yy c*. y, z) àx dy dz
^ V y _ y___________________
y, z) d x d y d z ^ J J Y (*. ÿ. z)à x d y dz
*v J vJ
SSS y> ^dx dy dz
2C Z
$ $ $ Y (* . y, z)dx dy dz
where y(x, y, z) is the density.

Example 2 . Détermine the coordinates of the centre of gravity of the upper
half of a sphere of radius R with centre at the origin, assuming the density y0
constant.
Solution. The hemisphere is bounded by the surfaces
r= YR % -y*. 2=0
The 2-coordinate of its centre of gravity is given by the formula
*Yod xdydz
2c =
J J $ Yodx d y d z
Changing to spherical coordinates, we get
2n
e\ 1 \ r cos 9 /® sin cp dr \ j dy
d9 2n ^ -l
0 _o \o / 4 2 _ 3
*c= -
" 4 D3 8
CM
-6 nR
*---- S 0
\ ( \ r2 sin 9 dr J d9 de
e
_o \o /
Obviously, by virtue of the symmetry of the hemisphere, x c —y c = 0.
2.15 COMPUTING INTEGRALS DEPENDENT

ON A PARAMETER
Consider an intégral dépendent on the parameter a:

b
1 (a ) = $ /(*. <*)dx
a
(We examined such intégrais in Sec. 11.10, Vol. I.) We state with-
out proof that if a function f(x, a.) is continuons with respect to
14—2082
x over an interval [a, b] and with respect to a over an interval

[alt <x2], then the function
b
I (a) = J / (x, a) dx
a
is a continuons function on [a,, a t]. Consequently, the function
I(a) may be integrated with respect to a over the interval [a,, a 2]:
a, a, r -j
J / (a) da = J J / (x , a) dx da
<Xt ou La J
The expression on the right is an iterated intégral of the function
f(x, a) over a rectangle situated in the plane xOa. We can change
the order of intégration in this intégral:
Q| b b Ctj
^ ^ f (x, a) dx da = J ^ f (x , a) da dx
ctj Lfl J a Lai
This formula shows that for intégration of an intégral dépendent
on a parameter a, it is sufficient to integrate the element of
intégration with respect to the parameter a. This formula is also
useful when computing definite intégrais.
Example. Compute the intégral
7 - a x _p-bx
î ■dx (a > 0 , b > 0 )
This intégral is not expressible in terms of elementary functions. To evatuate

it, we consider another intégral that may be readily computed:
oo
r !dx = — (a > 0)
Integrating this équation between the limits a = a and a = b, we get
fl Lo
a Ln J a
Changing the order of intégration in the first intégral, we rewrite this équation
in the following form:
b
| [ î ‘- “ + = ,4
whence, computing the inner intégral, we get

w
•e-a x — e- bx
oP
d x = ln —
x a
Evaluate the intégrais: '
1 2 2 jc K 3
*■ I §(xi+yi) dxdy - Ans- t • 2- J Ans-Ini i ' 3* J J xydxdy.

0 1 3 1 1 X
2n a a x
Ans. ^ . 4. J fj rdr de. Ans. -i- na». 5. fj § x ^+y * ' AnS' ^ T ~ a arctan7 •
0 fls in O 0 _*
a
a 2y bn/2
6. J* J* x yd x dy . Ans. - . 7. J J* pdGdp. Ans, *^rJi&a.
0 y-a b_ 0
2
Détermine the limits of intégration for the intégral J J /( x , y )dx dy where

P
the domain of intégration is bounded by the lines:
3 5
8. x = 2, jc = 3, £/ = — 1, y = 5. Ans. J J f (xt y)dydx. 9. y—0, y = 1—x®.

2 -1
1 1 -J C * a
i4ns. J J
-1 0
/(* , y)dydx. 10. *2+ yE= a*. i4ns. J J f(x , y)dydx.
- a -Vâï^x*
H. y = ~ \+ i? ’ y = *2- Ans‘ J $ f(x,y)dydx. 12. y = 0, y = a , y = x ,

*r -i *•
a y + 2a
y = x — 2a. Ans. f (x, y)dxdy. J J
o y
Change the order of intégration in the intégrais:
2 4 4 2 \V ~ k
13. $ $ /(* . y)dydx. Ans.

13 3
JJ f ( x t y)dxdy.
1
14. J J f (xt y) dy dx. Ans.
o jc *
1 \ / y a V 2 a y - y* a a
J J f ( x t y)dxdy. 15. J J f(x, y)dxdy. Ans. J J f ( x t y)dydx.
o y• o o 0 a — Va* - j c *
N L
* If the intégral is written as J J /( * . y )dx dy then, as has already been
M K
stated, we can consider that the first intégration is performed with respect to
the variable whose differential occupies the first place; that is,
N L N /L \
$ J / (JC, y ) d x d y = J ( J /(jc , y ) tdx )dy
MK M \K
14 *
1 V\-x* lV\-y•
1«. ^ ^ /(*. y) dy dx. Ans. J J f (x, y)dxdy.
-i o o -VT^ÿi
i l - y o V i - x ‘ 1 l- x
17. J ^ f(x, y)dxdy. i4ns. J J / (x, y ) d y d x + J J f(x,y)dydx.

0 _V \-y * —1 0 0 0
Compute the following intégrais by changing to polar coordinates:
_____ ji
a Va* - x* 2 a
18. J J Vra*— x- — y't dydx. Ans. J J V a 1— p* p dp d Q = j a».
0 0 0 0
n
a Va*-y* 2 a
19. j ^ (x2+ y 2)dx dy. Ans. ^ J p3 dpd0 = ^ - . 20. J ^X%^ yt)dydx.
o o 0 0 0 0
ji 71
2 » 2 a V2ax - x* 2 2a cos 0
Ans. P*pdpd0 = -j-. 21. J J dy dx. Ans. J J p d p d 0 = -^ -.
oo o o o o
Transform the double intégrais by introducing new variables u and v conneo
ted with x and y by the formulas x = u —uvt y = uv:
J
elix U fi i + v c b
22* J ^ ^ ns‘ J J f ( u ’—uv>uv)ududv. 23. JJ/(*, y) dy dx.
o ax a 0 0 0
1+a
b c
b+c 1- v b v
Ans. J J f(u — uvt uv)ududv-\- J J f (u — uv, uv)ududv.
o o b o
b +c
Calculating Areas by Means of Double Intégrais
24. Compute the area of a figure bonnded by the parabola y 2 = 2x and the
2
straight line y = x. Ans. — .
*3
25. Compute the area of a figure bounded by the curves y2 = 4ax, x-f-y = 3at
y — 0. Ans. ~ a 2.
26. Compute the area of a figure bounded by the curves x 1^2 -j- y 1/f2 = a1/2,
x + y = a. Ans.
27. Compute the area of a figure bounded by the curves y = sin x, y = cos x,
x = 0. Ans. Y 2 — 1.
28. Compute the area of a loop of the curve p = asin20. Ans. —— .
JTÛ2
o
29. Compute the entire area bounded by the lemniscate pa = a 3 cos 2<p. Ans. a2.
2xy
30. Compute the area of a loop of the curve F + irH
Hint. Change to new variables x = pacos0 and i/ = p6sin0. Ans.
Calculating Volumes
31. Compute the volumes of solids bounded by the following surfaces:
abc
—+ - £ + —= 1 , j c = 0, y = 0, 2 = 0. Ans
a 1b 1 c "F 32. z = 0, x2-\-y2= 1, x + y -f
+ z = 3. Ans. 3jt. 33. (x— 1)2+ (y — 1)2= 1, xy = z, 2 = 0. Ans. n. 34. x2-f- y2 —
32
— 2a* = 0, 2=0, x2+ y 2 = z2. Ans. — a3. 35. y = x 2, x = y 2, z = 0, z = \ 2 + y — x 2.
A 5 4 9
AnS' Ï44-
36. Compute the volumes of solids bounded by the coordinate planes, the
plane 2x + 3y — 12 = 0 and the cylinder z = ^ y 2. Ans. 16.
37. Compute the volumes of solids bounded by a circular cylinder of radius a,
whose axis coïncides with the z-axis, the coordinate planes and the plane
a 1 a \ 4 3J
38. Compute the volumes of solids bounded by the cylinders x2-\-y2 = a2t
x2-j-z2 = a2. Ans. j û 3. 39. y 2~l~z2 = x, x = y f z = 0. Ans. 40. x2-j-y2 -j-z2=
—a2, x2-\-y2 = R 2, a > R. Ans. [a3 — ( Y a2— R2)3]. 41. az = x2-\- y2,
ü
3
z = 0, x2-{-y2 = 2ax. Ans. — Jia3. 42. p 2 = a2 cos20, x 2+ y 2+ z2 = a2, 2 = 0.
(Compute the volume that is interior with respect to the cylinder.)
Ans. i - a 3 (3Ji + 20— 16 V~2).
Calculating Surface Areas
43. Compute the area of that part of the surface of the cone x2 + y 2~ z 2
which is eut out by the cylinder x2-\-y2 = 2ax. Ans. 2na2 V ' 2.
44. Compute the area of that part of the plane x + y + z = 2a which lies in
the first octant and is bounded by the cylinder x2-{-y2 = a2. Ans. Y 3.
45. Compute the surface area of a spherical segment (minor)if the radius of
the sphere is a and the radius of the base of the segment is b.
Ans. 2ji (û2— a Y a2— b2).
46. Find the area of that part of the surface of the sphere *2 y 2 + *a = û2
x2 y2
which is eut out by the surface of the cylinder = 1 (a > b).
A ns. 4jtû2—8a2 arcsi n —------------.

a
47. Find the surface area of a solid that is the common part of two cylin
ders jc2+ y 2 = û2, y2-\-z2= a2. Ans. 16a2.
48. Compute the area of that part of the surface of the cylinder x2 + y2 = 2ax
which lies between the plane 2 = 0 and the cône x2+ y2— z2. Ans. 8a2.
49. Compute the area of that part of the surface of the cylinder x2-\-y2 = a2
which lies between the plane z = mx and the plane 2 = 0. Ans. 2ma2.
50. Compute the area of that part of the surface of the paraboloid y 2 + z2 = 2ax
which lies between the parabolic cylinder y 2 = ax and the plane x = a.
Ans. y J i a 2 (3 V U — l).
Computing the Mass, the Coordinates of the Centre of Gravitv,

and the Moment of Inertia of Plane Solids
(In Problems 51-62 and 64 we consider the surface density constant and
equal to unity.)
51. Détermine the mass of a lamina in the shape of a circle of radius a if
the density at any point P is inversely proportional to the distance of P from
the axis of the cylinder (the proportionality factor is K). Ans. 2naK-
52. Compute the coordinates of the centre of gravity of an équilatéral triangle
if we take its altitude for the Jt-axis and the vertex of the triangle for the
a
coordinate origin. Ans. x = — g— , y = 0 .
53. Find the coordinates of the centre of gravity of a circular sector of radius a,
taking the bisector of its angle as the *-axis. The angle of spread of the sector
• o a 2a s i n a A
is 2a. Ans. xc = — , 0c = 0-
54. Find the coordinates of the centre of gravity of the upper half of the
circle x2-\-y2 = a2. Ans. x c = 0 , yc = ~ .
55. Find the coordinates of the centre of gravity of the area of one arch of
5a
the cycloid x = a(t — sin t), y = a( 1—ccs t). Ans. xc = an, yc = -§-
56. Find the coordinates of the centre of gravity of the area bounded by a
na j/'~2
loop of the curve pa = a 2 cos20. Ans. x c = ----- 5------, yc = 0-
O
57. Find the coordinates of fhe centre of gravity of the area of the cardioid
5a
p = a (1 + c o s 0 ).
A n s . x c = -çr, yc = 0-
58. Compute the moment of inertia of the area of a rectangle bounded by the
straight lines * = 0, x = a, y = 0 , y = b relative to the origin. Ans.
o
X2 U2
59. Compute the moment of inertia of the ellipse —j - (a) relative
to the y-axis; (b) relative to the origin. Ans. (a) 2 ? ; fl>) ■

—£■<*+6*).
60. Compute the moment of inertia of the area of the circle p = 2acos0
3
relative to the pôle. Ans. -g-na4.
61. Compute the moment of inertia of the area of the cardioid p = a (l —cos 0)
relative to the pôle. Ans. -^1° .
lo
62. Compute the moment of inertia of the area of the circle (x— a )2 + (y — b)2=
= 2a2 relative to the y-axis. Ans. 3na4.
63. The density at any point of a square lamina with side a is proportional
to the distance of this point from one of the vertices of the square. Compute
the moment of inertia of the lamina relative to the side passing through this
vertex. Ans. ^ k a h \ l >^2 + 3 l n ( ] ^ 2 + 1 )], where k is the proportionality
factor.
64. Compute the moment of inertia of the area of a figure, bounded by the
parabola y2= ax and the straight line x = at relative to the straight line y = —a.
Ans. -=- a4,
o
Triple Intégrais
dx dy dz
65. Compute ^ ^ § if the domain of intégration is bounded
(x + y + z + \ f
by the coordinate planes and the plane x-\-y-\-z = 1. Ans.
66. Evaluate dx.

Ans■ 48-
ô Lô \ô
67. Compute the volume of a solid bounded by the sphere *2+ y2+z* = 4
19
and the surface of the paraboloid *a+ ya = 3z. Ans. -g-jx.
68. * Compute the coordinates of the centre of gravity and the moments of
X U Z
inertia of a pyramid bounded by the planes x = 0, y = 0, z = 0 , ---- h-r-H------- 1.
a o c
A a b c . a3bc . b*ac . c?ab . abc y « ,
Ans. x c — 4 , y c — 4 » * c— 4 ; I x — 60 . Jy — 60 » h — 60 » A>— 60 (a +
+ b* + c*).
69. Compute the moment of inertia of a circular right cône relative to its
axis. Ans. where h is the altitude and r is the radius of the base of
the cône.
70. Compute the volume of a solid bounded by a surface with équation
(x2~\-y2+ z2)2 = a*x. Ans. ^-na3.
ü
71. Compute the moment of inertia of a circular cône relative to the diameter
jThr2
of the base. Ans. -gg- (2/i2+ 3r2).
72. Compute the coordinates of the centre of gravity of a solid lying between
a sphere of radius a and a conic surface with angle at the vertex 2a, if the
vertex of the cône coïncides with the centre of the sphere. Ans. xc = 0f yc = 0t
3
z c = “S“û (1 + c o s a) (the z-axis is the axis of the cône, and the vertex lies at
o
the origin).
73. Compute the coordinates of the centre of gravity of a solid bounded by
a sphere of radius a and by two planes passing through the centre of the sphere
9 ji
and forming an angle of 60°. Ans. p = y g a , 0 = 0, <p= y (the line of inter
section of the planes is taken for the z-axis, the centre of the sphere for the
origin; p, 0 , (p are spherical coordinates).
oc
1
74. Using the équation - 7 ^ = - ^
2 f\ e w * d a (a > 0 ), compute the intégrais
y x y ji j
0
OD 00 ____ ____
cos x dx , f sinxdx A -, f -- f n
f
ji
0
~VT and J ~ Y T - Ans■ h '
0
V T-
* In Problems 68 , 69 and 71 to 73 we consider the density constant and
equal to unity.
CHAPTER 3
L I N E IN T E G R A L S A N D S U R FA C E IN T E G R A L S
3.1 LINE INTEGRALS

Let a point P(x, y) be in motion along some plane curve L
from the point Af to the point N. To P is applied a force Fwhich
varies in magnitude and direction with the motion of P; it is thus
some function of the coordinates
of P:
F = F(P)
Let us compute the work A ot
the force F as the point P is tran-
slated from Af to N (Fig. 93). To
do this, we divide the curve MN
into n arbitrary parts by the points
Af=Af0, Af,, M2, . . . . M „ = N in
the direction from Af to N and we
dénoté by As,- the vector Af,Af,-+1.
We dénoté by F,- the magnitude of
the force F at the point Af,-. Then
the scalar product F,-As,- may be
regarded as an approximate expres-
s ion of the work of the F along the arc Af,-Af,-+1:
At « F,-As,-
Let
F = X(x, y)i + Y(x, y ) j
where X (x, y) and Y (x, y) are the projections of the vector F on
the x- and y-axes. Denoting by Ax, and Ai/,- the incréments of the
coordinates xt and yt when changing from the point Af,- to the
point Af,-+1, we get
ASi = A*,-/ + Ay J
Hence,
F,-As,- = X (*,-, (/,) A Xi + Y (.Xi, (/,-) At/,-
The approximate value of the work A of the force F over the
entire curve MN will be
a « 2 FiAs‘ = S (*/» và Axi + Y (xi' üi) Ayî\ ( 1)

i-I <=1
3.1 Line Intégrais 217
Without making any précisé statements, we shall say that if

the expression on the right has a limit as As,- -* 0 (here, obviously,
Axj -+Q and Ay,-+0), then this limit expresses the work of the
force F over the curve L from the point Al to the point N:
A = lim ^ [ X ( x „ yt)Axi + Y(Xi, y,)Ay,] (2 )
-° f=i
ay( -o
The limit* on the right is called the line intégral of X (x, y) and
Y (x, y) a long the curve L and is denoted by
A = l X(x, y)dx + Y(x, y)dy (3)
L
or
m
A — J X(x, y)dx + Y(x, y)dy (3')
<Af)
Limits of sums of type (2 ) frequently occur in mathematics and
mechanics; here, X (x , y) and Y (x, y) are regarded as functions of
two variables in some domain D.
The letters M and N, which take the place of the limits of
intégration, are in brackets to signify that they are not numbers
but symbols of the end points of the curve over which the line
intégral is taken. The direction along the curve L from M to N
is called the sense of intégration.
If the curve Z, is a space curve, then the line intégral of three
functions X (x, y, z), Y (x , y, z), Z (x, y , z) is defined similarly:
J X (x, y, z)dx + Y (x, y, z)dy + Z(x, y, z)dz
L
= lim y»< z*) A** + Y (xk, yk, zk) Ayk + Z(xk, yk, zk) Azk
Ay 0
Az - 0
k
The letter L under the intégral sign indicates that the intégration
is performed along the curve L.
We note two properties of a line intégral.
P ro p e rty 1 . A line intégral is determined by the element of in
tégration, the form of the curve of intégration, and the sense of
intégration.
* Here, the limit of the intégral sum is to be understood in the same sense
as ïn the case of the déftnite intégral, see Sec. 11.2 (Vol. I).
218 Ch. 3 Line Intégrais and Surface Intégrais
A line intégral changes sign when the sense of intégration js

reversed, since in that case the vector As and, hence, its projec
tions Ax and Ay, change sign.
Property 2 . Divide the curve L by the point K into pièces Lx
and La so that MN = M K + K N (Fig. 94). Then, from formula (1)
it follows directly that
(N) (K) (N)
J Xd x + Y dy— J X dx + Yd y + J X d x + Yd y
(M ) (M) (K)
This relation holds for any number of terms.

It will furthèr be noted that the définition of a line intégral holds
true also for the case when the curve L is closed.
In this case, the initial and terminal points of the curve coin-
cide. Therefore, in the case of a closed curve we cannot write
(V )
^ X d x + Ydy, but only ^ X d x + Ydy; and

(M) L
we hâve to indicate the direction of circu
lation (sense of description) over the closed
curve L. The line intégral over a closed con
tour L is frequently denoted also by the
symbol (fiXdx + Ydy.
L
Note. We arrived at the concept of a line intégral while consi-
dering the problem of the work of a force F on a curved path L,.
Here, at ail points of the curve L the force F was given as a
vector function F of the coordinates of the point of application
(*, y)\ the projections of the variable vector F on the coordinate
axes are equal to the scalar (numerical, that is) functions X (x , y)
and Y (x, y). For this reason, a line intégral of the form ^ X d x + Ydy
L
may be regarded as an intégral of the vector function F given by
the projections X and Y.
The intégral of a vector function F along a curve L is denoted
by the symbol
S Fds
L
If the vector F is defined by its projections X, Y, Z then this

intégral is equal to the line intégral
\ X d x + Y d y + Zdz
L
As a particular instance, if the vector F lies in the xy-plane, then

3.2 Evaluai ing a Line Intégral 219
the intégral of this vector is equal to

J Xdx+Ydy .
L
When the line intégral of a vector function F is taken along a

closed curve L, this line intégral is also called a circulation of the
vector F over the closed contour L.
3.2 EVALUATING A LINE INTEGRAL

In this section we shall make more précisé the concept of the
limit of the sum (1) of Sec. 3.1 and in this connection we shall
make more précisé the concept of the line intégral and indi-
cate a method for calculating it.
Let a curve L be represented by équations in parametric form:
*=<p( 0 .
Consider the arc of the curve MN (Fig. 95). Let the points M
and N correspond to the values of the parameter a and p. Divide
the arc MN into subarcs As,- by the
points AM*!, yt), M2(x2, y2), . . . .
Mn (xn, y n), and put *, =<p (/,), y , =
=*(*,)•
Consider the line intégral
$ X(x, y)dx + Y(x, y)dy (i)
L
defined in the preceding section. We
give without proof the existence theo-
rem of a line intégral. I f the fun-
ctions <p(t) and (t) are continuous and hâve continuous dérivatives
cp' (/) and if' (t), and also continuous are the functions X [<p (t), if (/)]
and Y [<p (/), if (/)] as functions of t on the interval [a, p], then the
folloœing limit s exist:
lim V X yf) Ax( = A '
Sxx.i --+
*■0 /=1
—
r i _ _ #— — (2)
lim % Y
iy. 0 t= 1
y ù * yi = B
where *, and y; are the coordinates of some point lying on the arc
ASj. These limits do not dépend on the way the arc L is divided into
subarcs As(-, provided that As,- 0, and do not dépend on the choice
of the point Af,•(*,•, yf) on the subarc As,-; they are called line
intégrais and are denoted as

A = \ x ( x , y)dx
: (2 ')
B = \ Y ( x , y)dy
Note. Froni this theorem it follows that the sums defined in the
preceding section, where the points yt) are the extremities
of the subarc As, and the manner of partition of the arc L into
subarcs Ast is arbitrary, approach the same lim it—the line intégral.
This theorem makes it possible to develop a method for com
puting a line intégral.
Thus, by définition, we hâve
( N) n _
5 X (x, y ) d x = lim V X (xh £/,) Axt (3)
(M) = I
where
A x i = X i— x ^ 1 = (/«-_,)
Transform this latter différence by the Lagrange formula
Axt = <p(/,)—q>(ti- 1) = <p' (x,) ( t i - t i - J = <p' (x,) At,
where x(- is some value of t that lies between the values and t(.
Since the point xit y( on the subarc As,- may be chosen at plea-
sure, we shall choose it so that its coordinates correspond to the
value of the parameter x,-:
*/ = <P(*,). ÿ/ = ^(x,)
Substituting into (3) the values of xit yt and Axt that we hâve
found, we get
(ao n
S X (x, y)dx = lim £X [«p (X,.)i|>(x,)]q>' (x,) Att
(Al) 4<,--0 i= l
On the right is the limit of the intégral sum for the continuous
function of a single variable X[q> (/), i|) (*)] <P’ ( 0 on the interval [a, P].
Hence, this limit is equal to the definite intégral of the function:
(N) fi
$ X (x, y) dx r =^ X [' (0 dt
( M) a
In analogous fashion we get the formula

m P
S Y {x, y)dy = \ Y [cp (/), if (/)] Ÿ (0 dt
(M ) <x
3.2 Evaluaiing a Line Intégral 221
Adding these équations term by terni, we obtain

<A0 ft
SA (X, y) dx + Y (X , y)dy = \ { X ^ (t), « (/)] <p' (t)
(M) a
+ K [ ( 0 ] i p '( 0 } ^ w

This is the desired formula for computing a line intégral.
In similar manner we compute the line intégral
J X dx + Y dy -f- Z dz
over the space curve defined by the équa
tions Jf = q>(0 , y = ÿ(t), z = X(0 -
Example 1. Compute the line intégral of three
functions: je3, 3zy2> —x2y (or, what is the same
thing, of the vector function x9i + 3zy2j —x2yk)
along the straight-line segment from the point
M (3, 2, 1) to the point N (0, 0, 0) (Fig. 96).
Fig. 96
Solution. To find the parametric équations of the line MN, along which the
intégration is to be performed, we Write the équation of the straight line that
passes through the given two points:
3 2 1
and dénoté ail these ratios by a single letter t\ then we get the équations of
the straight line in parametric form:
x = 3t, y = 2 t, 2 — t
Here, obviously, to the origin of the segment MM corresponds the value of tne
parameter t = 1, and to the terminus of the segment, the value / = 0 . The déri
vatives of x , y , z with respect to the parameter t (which will be needed for
cvaluating the line intégral) are easily foundf
= 3» = 2 , z^ = \
Now the desired line intégral may be computed by formula (4):
{N) o
$ je3 d x + 3z(/® d y - x * y d z= J [(3/)3 -3 + 3 / (2/)2 - 2 - ( 3 0 2 -2MM<
(Àf) 1
1
Example 2. Evaluate the line intégral of a pair of fonctions, 6x2y, and lOx^2,
along a plane curve y = jc3 from the point Af (1, 1) tothe point N (2, 8) (Fig. 97).
Solution. To compute the required intégral
(N)
Ç 6x2y dx+lOxy2 dy
(M )
we must hâve the parametric équations of the given curve. However, the exp-
licitly defined équation of the curve y — x 3 is a spécial case of the parametric
équation: here, the abscissa x of the point of the curve serves as the parameter,
and the parametric équations of the curve are
x = x, y = x 3
The parameter x varies from x x = \ to x2 = 2. The dérivatives with respect to
the parameter are readily evaluated:
3*s
Hence.
(AT) 2 2
^ 6x*ydx+l0xyi dy= J [6jc2jc* - 1 -f I0xxa-3x*]dx= Ç (6xs + 30x») dx
(M) 1 1
= [jc*+3 jc10]| = 3132
We now indicate certain applications of a line intégral.

1. The expression of the area of a région bounded by a curve
in terms of a line intégral. In an
xy-plane let there be given a do
main D (bounded by a contour L)
such that any straight line paral-
lel to one of the coordinate axes
and passing through an interior
point of the domain cuts the boun-
dary L of the domain in no more
than two points (which means that
D is regular) (Fig. 98).
Suppose that the domain D is
projected on the x-axis in the
interval [a, 6 ], and it is bound
ed below by the curve (/,):
V = yii*)
3.2 Evaluaiing a Line Intégral 223
and above by the curve (/2):

y=y*(x)
[ÿi W < ÿ i W ] .
Then the area o! the domain D is
b b
S = J yt (x)dx— J yl (x) dx
a a
But the first intégral is a line intégral over the curve l2(MPN),
since y = yt (x) is the équation of this curve; hence,
b
\ y i (x)dx= f ydx
a MPN
The second intégral is a line intégral over the curve lx(MQN),
that is,
b
Jt/i(*)dx= J ydx.
a MQN
By Property 1 of the line intégral we hâve
J ydx = — J ydx
MPN NPM
Hence,
S = — J y d x — J ydx = — \ y d x (5)
NPM MQN L
Here, the curve L is traced in a counterclockwise direction.

If part of the boundary L is the segment M.M, parallel to the
(M)
y-axis, then C ydx = 0, and équation (5) holds true in this case
(M .)
as well (Fig. 99).
Similarly, it may be shown that
S=^xdy (6 )
Adding (5) and (6 ) term by term and dividing by 2, we get

another formula for computing the area S:
S = Y § x d y — ydx (7)
L
Example 3. Compute the area of the ellipse
x — a cos t, y = b s i n t
Solution. By formula (7) we find

271
5= ^ [a cos tb cos t — b sin t (— a sin /)] dt-=nab

o
We note that formula (7) and formulas (5) and (6 ) as well hold
true also for areas whose boundaries are eut in more than two
points by lines parallel to the coordinate axes (Fig. 100).
To prove this, we divide the given domain (Fig. 100) into two
regular domains by the line l*.
Formula (7) holds for each of these domains. Adding the left and
right sides, we get (on the left) the area of the given domain, on
the right, a line intégral taken over the entire boundary with
coefficient 1/ 2 since the line intégral over
the division line /* is taken twice: in the
direct and reverse senses; hence, it is equal
to zéro.
2. C o m p u tin g th e w ork of a v a ria b le
force F on sotne cu rv ed p ath L. As was
shown at the beginning of Sec. 3.1, the
work done by a force F — X(x, y, z)i +
+ Y (x, y, z ) j + Z ( x , y, z)k along a line
L = MN is equal to the line intégral
A = C X ( x , y, z)dx + Y(x, y, z)dy + Z(x, y , z)dz

<M )
Let us consider an instance that shows how to calculate the work

of a force in concrète cases.
Example 4. Détermine the work A of the force of gravity F when a mass m
is translatée! from the point M 1(alt blt cx) to the point M 2 (a2, b2, c2) along
an arbitrary path L (Fig. 101).
Solution. The projections of the force of gravity F on the coordinate axes
are
X = 0, K = 0, Z = — mg
3.3 Green's Formula 225
Hence, the desired work is

( M #)
A= [ Xdx + Ydy + Z (— mg)dz = mg(c1—c2)

( A * i)
Consequently, in this case the line intégral is independent of the path of in
tégration and dépendent only* on the initial and terminal points. More precisely,
the work of the force of gravity is dépendent only on the différence between
the heights of the terminal and initial points of the path.
3.3 GREEN’S FORMULA
Let us establish a connection between a double intégral over

some plane domain D and the line intégral around the boundary L
of that domain.
In an xy-plane, let there be given a domain D, which is regu-
lar both in the direction of the x-axis and the y-axis, bounded by
a closed contour L. Let the domain be bounded below by the
curve y = yt (x), and above by the curve y = y2(x), yx (x) < y2(x)
( a ^ i x ^ i b ) (Fig. 98).
Together, both these curves represent the closed contour L. Let
there be given, in D, continuous functions X (x, y) and Y (x, y)
that hâve continuous partial dérivatives. We consider the intégral
Representing it in the form of a twofold itérated intégral, we find
D a ÿt(x) a y t ( x)
b
=
5 [X{x, y%(x))— X (x, yl (x))] dx (1)
a
We note that the intégral
6
S * (x , »t (x))dx
a
is numerically equal to the line intégral

S X(x,y)dx
MPN
taken along the curve MPN, whose équations, in parametric
form, are
x = x, y = yt (x)
where x is the parameter.
15—2082
Thus
b
J X(x, yl (x))dx= J X(x, y)dx (2)
a MPN
Similarly, the intégral

b
J X (x, y1(x)) dx
a
is numerically equal to the line intégral along the arc MQN :

b
\ x ( x , yAx))dx= ] X (x, y) dx (3)
a MQN
Substituting expressions (2) and (3) into formula (1), \ve obtain
5 S w dxd&= 5 y)dx ~ j X(x,y)dx (4)

D M PN MQN
But
J X (x, y)dx = — J X(x, y) dx
MQN NQM
(see Sec. 3.1, Property 1). And so formula (4) may be written thus:
i ï w
O
dxdy=1 I *(*» y ) d x + J
M PN NQM
x ( x > y ) dx
But the sum of the line intégrais on the right is equal to the line
intégral taken along the entire closed curve L in the clockwise
direction. Hence, the last équation can be reduced to the form
^^-dxdy= J X(x,y)dx (5)
D L (In the clockwise sense)
If part of the boundary is the segment / 3 parallel to the t/-axis,
then ^ X (x, y) dx = 0, and équation (5) holds true in this case as
u
well.
Analogously, we find
^™ -d xd y = - J Y(x,y)dy (6 )
D L (In the clockwise sense)
Subtracting (6 ) from (5), we obtain
N { w ~ % r ) dxdy= S Xd x + Ydy
D L (in the clockwise sense)
3.4 Conditions for a Line Intégral to Be Independent of ihe Pat h 227
If the contour is traversed in the counterclockwise sense, then *
D L •
This is Green's formula, named after the English physicist and

mathematician G. Green (1793-1841). **
We assumed that the domain D is regular. But, as in the area
problem (see Sec. 3.2), it may be shown that this formula holds
true for any domain that may be divided into regular domains.
3.4 CONDITIONS FOR A LINE INTEGRAL TO BE

INDEPENDENT OF THE PATH OF INTEGRATION
Consider the line intégral

(W)
C X(dx) + Ydy
(M)
taken around some plane curve L connecting the points Af and N.
We assume that the functions X (x, y) and Y (x, y) hâve conti-
nuous partial dérivatives in the domain D
under considération. Let us find out under
what conditions the line intégral above is
independent of the shape of the curve L
and is dépendent only on the position of Af P
the initial and terminal points Af and N. Fig. 102
Consider two arbitrary curves MPN and
MQN lying in a given domain D and connecting the points Af
and N (Fig. 102). Let
J Xd x + Y d y = $ Xd x + Ydy (1)
MPN MQN
that is,
J X d x + Y d y — $ Xd x + Ydy = 0
MPN MQN
Then, on the basis of Properties 1 and 2 of line intégrais (Sec. 3.1)
we hâve
J Xdx + Yd y + J Xd x + Yd y = 0
MPN NQM
* If in a line intégral along a closed contour the direction of circulation

is not indicated, it is assumed to be counterclockwise. If the direction of circu
lation is clockwise, this must be specified.
** This formula is a spécial case of a more general formula discovered by
the Russian mathematician M. V. Ostrogradsky.
which is a line intégral around the closed contour L:

J X d x + Ydy = 0 (2)
In this formula, the line intégral is taken around the closed

contour L, which is made up of the curves MPN and NQM. This
contour L may obviously be considered arbitrary.
Thus, from the condition that for any two points M and N the
line intégral is independent of the shape of the curve connecting
them and is dépendent only on the position of these points, it
follows that the line intégral along any closed contour is equal
to zéro.
The converse conclusion is also true: if a line intégral around
any closed contour is equal to zéro, then this line intégral is inde
pendent of the shape of the curve connecting the two points, and
dépends only upon the position of the points. Indeed, équation (1)
follows from équation (2 ).
In Example 4 of Sec. 3 .2 , the line intégral is independent of
the path of intégration; in Example 3 the line intégral dépends
on the path of intégration because there the intégral around the
closed contour is not equal to zéro, but yields an area bounded by
the contour in question; in Examples 1 and 2 the line intégrais are
likewise dépendent on the path of intégration.
The natural question arises: what conditions must the functions
X (x , y) and Y (x , y) satisfy in order that the line intégral
^ X d x + Y dy along any closed contour be equal to zéro. The
answer is given by the following theorem.
Theorem. At ail points of sortie domain D, let the functions
X {x, y), Y (x, y), together with their partial dérivatives ^ and
dY<dx ^ con^ nuous- Then, for the line intégral along any closed
contour L lying in this domain to be zéro, that is, for
$X (*. y ) d x +Y( x , y)dy-= 0 (2')
it is necessary and sufficient that the équation

dX = dY
dy dx (3)
holds at ail points of D.
Proof. Consider an arbitrary closed contour L in D and Write
Green’s formula for it:
n l
3.4 Conditions for a Line Intégral to Be Independent of the Pat h 229
If condition (3) is fulfilled, then the double intégral on the left

îs identically zéro and, hence,
S Xdx + Ydy = 9
This proves the sufficiency of condition (3).

Now we prove the necessity of this condition; that is, we prove
that if (2) is fulfilled for any closed curve L in the domain D,
then condition (3) is also fulfilled at each point of this domain.
Let us assume, on the contrary, that équation (2 ) holds true,
that is,
J
X dx + Y d y ^
L
0
and that condition (3) does not hold,
dY dX_
ày
at least in one point. For example, suppose at some point P(xt , y 0)
we hâve the inequality
dx dy u
Since there is a continuous function on the left, it will be positive
and greater than some number 6 > 0 at ail points of some suffi-
ciently small domain D’ containing the point P (x0, y0). Take the
double intégral of the différence — — ^ over this domain. Itw ill
hâve a positive value. Indeed,
^ { ^ ~ w ) d x d y > ^ 6dxdy = 6 ^ dXdy:=zàD' > 0

D' D■ D'
But by Green’s formula the left side of the last inequality is

equal to a line intégral along the boundary L' of D' , which, by
assumption, is zéro. Hence, the last inequality contradicts condi
tion (2 ) and therefore the assumption that 4^- — is different
from zéro in at least one point is not correct. Whence it follows that
at ail points of the given domain D.

The proof of the theorem is thus complété.
In Sec. 1.9, it was proved that fulfilment of thé condition
dY(x, y) _ d X ( x , y)
dx dy
is tantamount to the fact that the expression X d x + Y d y is an

exact differential of sotne function u (x , y), or
X dx + Y dy =du(x, y)
and
X(x, y) = %, Y{x, y) - p y
But in this case the vector
F = Xi + Y j= p x i + pyJ
is the gradient of the function u (x , y)\ the function u (x, y), the
gradient of which is equal to the vector X l + Yj, is called the
potential of this vector.
( N)
We shall prove that in this case the Une intégral I = f X dx + Y dy
( M)
along any curve L connecting the points M and N is equal to the
différence between the values of the function u at these points:
(AT) (AT)
J X d x + Yd y = J du(x, y) = u ( N ) — u(M)
( M) <M)
Proof. If Xd x + Yd y is the exact differential of the function
u(x, y), then X = p , Y = p and the line intégral takes on the
form
(N)
M (M
ï -) f c + S *
To evaluate this intégral we Write the parametric équations of
the curve L connecting the points M and N:
* = <P(Q. y = W )
We shall assume that to the value of the parameter t = /„
corresponds the point M, and to t = T , the point N. Then the
line intégral reduces to the following definite intégral:
/ = Jf [\ dJdLxdJdi t+^ ddydt

1
J i dl \ d t
\ ac
to
The expression in the brackets is a function of t, and this func
tion is the total dérivative of the function i|>(0 ]with
respect to t. Therefore
T
I = $ p d t = u [ i P (/), * (f)] \l - u [<p (0 , * (0 ]

3.4 Condition* for a Line Intégral to Be Independent of the Path 231
As we see, the line intégral of an exact differential is independent

of the shape of the curve along which the intégration is performed.
We hâve a similar assertion for a lÿie intégral over a space
curve (see below, Sec. 3.7).
Note. It is sometimes necessary to consider line intégrais of
some function X (x , y) along the arc length L:
\ X ( x , y) ds= lim J X( x t, Vt) As, (4)

L As,—° i = l
where ds is the differential arc length. Such intégrais are evaluated

in similar fashion to the line intégrais considered above. Let the
curve L be represented by the parametric équations
*=<p( 0 . y = $ ( t )
where <p(0 . i|»(0 » q>'(0 » Ÿ (t) are continuous functions of t.
Let a and p be values of the parameter t corresponding to the
initial and terminal points of the arc L.
Since
ds = l 'V W ' + 'l’' (0 *dt,
we get a formula for evaluating intégral (4):
$ X (.x, y) ds = J X [<p (t), * (/)] W (0 2 + ^ ' (O2 dt

L a
We can consider the line intégral along the arc of the space curve
* = < P (0 . = z = X (0 :
y, z ) d s = l X [ v ( t ) , ♦(/), x(t)]V'P'(ty + V ( t y + x'(t)t dt

L a
By the use of line intégrais along an arc we can déterminé, for

example, the coordinates of the centre of gravity of lines.
Reasoning as in Sec. 1 2 .8 , Vol. I, we obtain a formula for
evaluating the coordinates of the centre of gravity of a space curve:
J y ds ^ z ds
Xr —‘ Vc (5)
Jds
F ' L L
Example. Find the coordinates of the centre of gravity of one turn of the
hélix
x = ac os t , y = asint, z — bt (0^t<2:i)
if its linear density is constant.
Solution. Applying formula (5), we find

2n
Ç a cos / Vâ2 sin 2 / + 02 côs2 1 -\-b*dt
*c=- 2Jt
J V û2 sin2 t -\-a2 cos2 / + b2dt
o
2*t
s
J a cos /
0
2ji
+ b2 d/
2rt
$ Và1+ b*
Similarly, t/c = 0,
2jx _______________
^ bt Y a2 sin21 + a 2 cos2 t + b2dt
b- 4ji2 -nb
*C 2jiVra2 + *2 2-2:1
Thus, the coordinates of the centre of gravity of one turn of the hélix are
*c = 0, J/c = 0. zc = nb
3.5 SURFACE INTEGRALS
Let a domain V be given in a rectangular xÿz-coordinate S ystem .

Let a surface a bounded by a certain space curve k be given in V.
With respect to the surface a we shall assume that at each
point P of it the positive direction of the normal is determined
by the unit vector n(P), the direction cosines of which are con-
tinuous functions of the coordinates of the surface points.
At each point of the surface let there be defined a vector
F = X ( x , y , z)l + Y{x, y , z ) j +Z ( x , y, z)k
where X, Y, Z are continuous functions of the coordinates.
Divide the surface in some way into subdomains A<r,-. In each
one take an arbitrary point P{ and consider the sum
2 (F(P,)n(P,)) Ao,. ( 1)
where F (P;) is the value of the vector F at the point Pt of the

subdomain Aa;; n (P,) is the unit normal vector at this point and
Fn is the scalar product of these vectors.
The limit of the sum (1) extended over ail subdomains Aa, as
the diameters of ail such subdomains approach zéro is called the
3.5 Surface Intégrais 233
surface intégral and is denoted by the symbol

\\Fndo .
a
Thus, by définition,*
lim 2 F fl, A0 , Fndo (2)
diam AtJj—*0 q
Each term of the sum (1)

FiniAoi —F'Ao, cos (»,-, F,) (3)
may be interpreted mechanically as follows: this product is equal
to the volume of a cylinder with base Acr,- and altitude /r,cos(»l-, Ff).
If the vector F is the rate of flow of a liquid through the sur
face a, then the product (3) is equal to the quantity of liquid
flowing through the subdomain Aat in unit time in the direction
of the vector », (Fig. 103).
The expression ^ F n d a yields the total quantity of liquid

O
flowing in unit time through the surface a in the positive direc
tion if by the vector F we assume the flow-rate vector of the
liquid at the given point. Therefore, the surface intégral (2) is
called the flux of the vector field F through the surface a.
From the définition of a surface intégral it follows that if the
surface a is divided into the parts alt o2, . . . , ok, then
J J Fndo = 55 Fndo + 5S^7/I^(I+ ••• + $$ Fndo
a Oj at ak
* If the surface o is such that at each point of it there exists a tangent

plane that constantly varies as the point P is translated over the surface,
and if the vector function F is continuous on this surface, then this limit
exists (we accept this existence theorem of a surface intégral without proof).
Let us express the unit vector n in terms of its projections on

the coordinate axes:
» = cos(/i, x )/ + cos(/i, y )j+ c a s (n, z)k
Substituting into the intégral (2) the expressions of the vectors
F and n in terms of their projections, we get
JJ Fn d o = J J [Xcos(n, x) + Kcos(rt, y) + Zcos(n, z)]da (2')
a a
The product Aocos(n, z) is the projection of the subdomain Ao on

the xy-plane (Fig. 104); an analogous assertion holds true for the
following products as well:
Aocos (n, x) = Aoyz, Ao cos(n, y) = Aoxz, Ao cos (n, z) — Aoxy (4)
where Aayz, Aoxz, Aoxy are the projections of the subdomain Ao
on the appropriate coordinate planes.
On this basis, intégral (2 ') can also be written in the form
J J Fn do — J J [X cos (n, x) + Y cos (n, y) -f Z cos (n, z)j do
a o
= J ÜX dydz + Y dzdx + Zdxdy (2")
3.6 EVALUATING SURFACE INTEGRALS
Computing the intégral over a curved surface reduces to eva-

luating a double intégral over a plane région.
To illustrate, the following is a method of computing the intégral
J J Z cos (n, z) do
a
Let the surface o be such that any straight line parallel to the
z.-axis cuts it in one point. Then the équation of the surface may
be written in the form
z = f ( x , y)
Denoting by D the projection of the surface o on the xy-plane,

we get (by the définition of a surface intégral)
ÇÇz(x, y, z)cos(n, z)do = lim 2 2(x,, yit z,) cos(nh z) Acr,-

q diam AcFj —
►0 ^ = 1
3.6 Evaluaiing Surface Intégrais 235
Noting, further, the last of formulas (4), Sec. 3.5, we obtain

n
$$Zcos(«, z ) da= lim £ z ( .v * f ( x h y,)) (Aa^),-
0 diam Ao x y —
►o ^= j
n
= ± [[m Z , z (x'< yh f (x« y à) I Ao |,-
diam A<j -► o 1= \
the last expression is the intégral sum for a double intégral of

the function Z(x, y, f(x, y)) over the domain D. Therefore,
$$zcos(n, z)do= ± JSz (x > y* /(*• y))dxdy

a D
The plus sign in front of the double intégral is taken if cos (n, z)^0,
the minus sign, if cos (n, z)^0.
If the surface a does not satisfy the condition indicated at the
beginning of this section, then it is divided into parts that satisfy
this condition, and the intégral is computed over each part
separately.
The following intégrais are computed in similar fashion:
^ Xcos(n, x) do, 5J r cos (n, y)da
a a
The foregoing proof justifies the notation of a surface intégral
in the form of (2"), Sec. 3.5.
Here, the right side of (2") may be regarded as the sum of
double intégrais over the appropriate projections of the région a
and the signs of these double intégrais (or, otherwise stated, the
signs of the products dydz, dxdz, dxdy) are taken in accord with
the foregoing rule.
Example 1. Let a closed surface o be such that any straight line parallel
to the z-axis cuts it in no more than two points.
Consider the intégral
^ ^ z cos (n, z) do
o
We shall call the ou ter normal the positive direction of the normal.
In this case, the surface may be divided into two parts: lower and upper;
their équations are, respectively,
* = f i ( x , y ) and z = f2 (x, y)
Dénoté by D the projection of a on the xy-plane (Fig. 105); then
5 5 zco s(n , z )d o = 5 5 M * JJ y)dxdy

a D D
The minus sign in the second intégral is taken because in a surface intégral
the sign of dxdy on a surface z = fi(x , y) must be taken négative, since for it
cos (n, z) is négative.
But the différence between the intégrais on the right in the last formula
yields a volume bounded by the surface o. This means that the volume of the
solid bounded by the closed surface a is equal to the following intégral over
the surface
MSa z cos (n, z) do
Example 2. A positive electric charge e placed at the coordinate oriîn

créâtes a vector field such that at ench point of space the vector F is defined
by the Coulomb law as
F = k± -r
where r is the distance of the given point from the origin, r is the unit vector
directed along the radius vector of the given point (Fig. 106), and k is a con
stant coefficient. Détermine the vector-field flux through a sphere of radius R
with centre at the origin of coordinates.
Solution. Taking into account that
S S k ^ r nda^ w S S r n d a
o a
But the last intégral is equal to the area of the surface o. Indeed, b y the
définition of an intégral (noting that r n = - l ) , we obtain
« rndo=
Ao* -
lim V /v t* A o * =
"
lim
boh-»o **
VAcf* —o
ke __ ke
Hence, the flux is 4n^* = 4nA;e.
3.7 STOKES’ FORMULA

Let there be a surface a such that any straight line parallel
to the z-axis cuts it in one point. Dénoté by X the boundary of
the surface cr. Take the positive direction of the normal n so
that it forms an acute angle with the positive z-axis (Fig. 107).
3.7 Stokcs' Formula 237
Let the équation of the surface be z = f(x, y). The direction

cosines of the normal are expressed by the formulas (see Sec. 9.6,
Vol. I):
df_
dx
cos (n, x) =
j / i + (ÊL f f V
\ dx + \ < > y i
df_
COS(rt, t/) = ày a)
V (ÊL (ÊL
i + ^ dx '+ \ d y
cos (n, z)-
' + m
We shall assume that the surface a lies entirely in some do*
main V. Let there be a function X (x, y y z) given in V 1hat is
continuous togetherwith its first-order par
tial dérivatives. Consider the line intégral
along the curve X:
KX (x, y, z) dx
i
On the curve X, z = / (x, y), where x , y
are the coord ina tes of the points of the
curve L, which is the projection of X on
the xi/-plane (Fig. 107). Thus, we can write
the équation
J X (x, 11, z) d x ^ ^ X (x, y, f (x, y)) dx (2)
X L
The last intégral is a line intégral along

L. Transform this intégral by Green’s formula, putting
x {X, y, f (x, y)) = X (x, y), 0 = V (x, y) _
Substituting into Green’s formula the expressions of X and F,
we obtain
_ J j d*~(X' % J {X—- ~dxdy = § X ( x , y, f (x, y)) dx (3)
D L
where the domain D is bounded by the curve L. On the basis of

the dérivative of the composite function X (x, y , f(x, y)), where
y enters both directly and in terms of the function z = f (x, y),
we find
dX (x , y, f (x , y)) ___ dX (x . y, z) . dX (x , y, z) df (x , y)
dy ày ' àz ày (4)
Substituting expression (4) into the left side of (3), we obtain
~ ÿ d x i y
= J x (x, y, f (x, y)) dx
Taking into account (2), the last équation may be rewritten as

y, z)dx = - ^ ^ - d x d y — ^ ^ - ^ d x d y (5)
l D D
The last two intégrais can be transformed into surface intégrais.

Indeed, from formula (2"), Sec. 3.5, it follows that if we hâve
some function A (x, y, z), the following équation is true:
J J A (x, y , z) cos (n, z )do= JJ Ad x d y
a D
On the basis of this équation, the intégrais on the right side
of (5) are transformed as follows:
[ S w dxdy = [ S w cos(n’ z)da 1

N w %
D
dxdy = N
a
£ w C0Sint z ) d a (6)
Transform the last intégral using formulas (1) of this section:
dividing the second of these équations by the third termwise,
we find
cos (n, y ) df
cos (n, z) dy
or
|^-cos (n, z) = cos (n, y)
Hence,
S S ^ - § d x d y ^ - S S ^ cos(n’ y)da (7)

D o
Substituting expressions (6 ) and (7) into équation (5), we get
y, z)dx = — Jj-^cos(n, z)da + j j — cos(n, y)do ( )
8
X a a
The direction of circulation of the contour K must agréé with

the chosen direction of the positive normal n. Namely, if an
3.7 Stokes' Formula 239
observer looks from the end of the normal, he sees the circulation
along the curve X as being counterclockwise.
Formula (8 ) holds true for any surface if this surface can be
divided into parts whose équations hâve the form z = f(x, y).
Similarly, we can write the formulas
y, z)dy = J J [ — -^-cos(n, x) + -^ c o s (/i, z)jda (8 ')
k o
§Z(x, y, z)dz = J j [— ^ -c o s (n , y) + ~ c o s ( n , x)]do (8 ")

k o
Adding the left and right sides of (8 ), (8 '), and (8 "), we get the
formula
^ X d x + Ydy + Zdz = J J [ ( | ~ — 1 cos(n, x)
k a
+ ( ï r - S B cos(n’ ^ - ) c o s <rt* z ) ] d a <9 )

This formula is called Stokes’ formula after the English physicist
and mathematician G. H. Stokes (1819-1903). It establishes a re-
lationship between the intégral over the surface o and the line
intégral along the boundary X of this surface, the curve X being
traversed according to the same rule as that given earlier.
The vector B, defined by the projections
R _ dZ dY R _ d X ___ dZ_ R dY dX
° * ~ dy dz ’ ° y ~ dz dx ’ ° z~ dx dy '
is called the curl of the vector function F= Xi + Y j + Z k and is
denoted by the symbol curl F.
Thus, in vector notation, formula (9) will hâve the form
J Fds = J J n curl F do (9')
X o
and Stokes’ theorem is formulated thus:
The circulation of a vector around the contour of sotne surface
Is equal to the flux of the curl through this surface.
Note. If the surface a is a piece of plane parallel to the
xy-plane, then Az = 0, and we get Green’s formula as a spécial
case of Stokes’ formula.
From formula (9) it follows that if
dY dX _ 0 ar _ n dX dZ _
dx dy ' dy dz ~~ ’ dz dx ~~ ( 10)
then the line intégral along any closed space curve X is zéro:
J X dx + Y dy + Zdz = 0 01 )
K
Whence it follows that the line intégral is independent of the

shape of the curve of intégration.
As in the case of a plane curve, it may be shown that the
indicated conditions are not only sufficient but also necessary.
In the fulfilment of these conditions, the expression under the
intégral sign is an exact difîerential of some function u(x, y, z):
X dx + Y dy + Z dz = du (x, y, z)
and, consequently,
<W) ( N)
J X d x + Y d y + Zd z = J du = u(N) — u(M)
( M) ( M)
This is proved exactly like the corresponding formula for a func

tion of two variables (see Sec. 3.4).
Example 1. Write the basic équations of the dynamics of a material point:
dv.
Z
dt
Here, m is the mass of the point, X t Y , Z are the projections of a force,
dx dy dz
acting on the point, onto the coordinate axes; vx =
dt vy-~df v‘= d tare
the projections of velocity v onto the axes.
Multiply the left and right si des of these équations by the expressions
vxdt = dx, Vydt = dyt vzdt = dz
Adding the given équations term by term, we obtain
m (vxdvx + v-ydvy + vzduz) = X dx + Y dy + Z dz
m ^ d ( v l + v l + v\) = X d x + Y d y + Z d z
Since vl+Vy-{-vl = v2f we can Write
d ( ^ m v tyj = X d x - \ - Y d y + Zdz
Take the intégral along the trajectory connecting the points M i and Aft :
(M.)
I 2 1 2 f X d x + Y d y + Zdz
T m v 2 ~ m v l = j
(Mt)
where vx and v2 are the velocities at the points Afj and M a.
This last équation expresses the work-kinetic energy theorem: the increase in
kinetic energy when passing from one point to another is equal to the work of
the force acting on the mass m.
Example 2 . Détermine the work of the force of Newtonian attraction to a
fixed centre of mass m in the translation of unit mass from M l (alt bl9 cx) to
M* c,).
3.8 Ostrogradtky'i Formula 241
Solution. Let the origin be in the fixed centre of attraction. Dénoté by r

the radius vector of the point M (Fig. 108) corresponding to an arbitrary po
sition of unit mass, and by r ° the unit vector directed along the vector r .
km
Then F — ---- ^ - r 0, wêre k is the gravitation constant. The projections of the
force F on the coordinate axes will be
X=-km -L — , K = -4 m — - , Z = -k m -\--
r2 r r2 r r2 r
Then the work of the force F over the path
AfjAij is
<Mg)
x dx + y dy -|- z dz
r8
(Mx)
<Mf) (Mt)
= — km f L^L — km f d
(M.) r (Mj)
(since r2= x* + y2 + z2, it follows that Fig. 108
r dr = x d x + ydy~\~zdz). If we dénoté by r1
and r a the lengths of the radius vectors of the points and Af2» then
A- b " ( - h ~ 7 ; )
Thus, here again the line intégral does not dépend on the shape of the
curve of intégration, but only on the position of the initial and terminal
■* km
points. The fùnction u = — is called the potential of the gravitational field
generated by the mass m. In the given case,
_du v du 7 du
A = u (M2) — u (M i )
~~d~x' dy' dz '
that is, the work done in moving unit mass is equal to the différence between
the values of the potential at the terminal and initial points.
3.8 OSTROGRADSKY’S FORMULA
Let there be given, in space, a regular three-dimensional domain

V bounded by a closed surface a and projected on an xy-plane
into a regular two-dimensional domain D. We shall assume that
the surface a may be divided into three parts a lt a2, and a 3
such that the équations of the first two hâve the form
z —fi (x > y) and z = /,(*, y)
where /, (jc, y) and f 2(x, y) are functions continuons in D and the
third part o, is a cylindrical surface with generator parallel to
the z-axis.
16—2082
Consider the intégral
First perform the intégration with respect to z:
' “ ïDï (V , I(jc. if) /
= $SZ (*> y, f t (x, y))dxdy— \ \ z ( x , y, / X(x, y))dxdy ( 1)
On the normal to the surface, choose a definite direction, name-

ly that which coïncides with the direction of the outer normal
to the surface a. Then cos (n, z) will be positive on the surface
at and négative on the surface on the surface o3 it will be
zéro.
The double intégrais on the right of (1) are equal to the cor-
responding surface intégrais:
$$Z(x, y, f t (x, y ) ) d x d y = \ \ z ( x , y, z) cos (n, z)da (2 ')

D Ot
$$Z(x, y, f ^ x , y ) ) d x d y = \ \ Z ( x , y, z)(— cos(n, z))da

D at
In the last intégral we wrote [ —cos(/i, z)] because the éléments

of surface and <r5 and the element of area As of the domain D
are connected by the relation As = A o[—cos(/i, z)], since the
angle (n, z) is obtuse.
Thus,
SS Z (x, y, f l (x, y)) dx dy =- — J $ Z (x,
D a,
y, f 1(x, y)) cos (n, z) da (2")
Substituting (2') and (2") into (1), we obtain
z)cos(n, z ) d a + ^ Z ( x , y, z)cos(n, z)da

a, o,
For the sake of convenience in subséquent formulas, we shall rewrite

the last équation as follows f adding y, z)cos(n, z)da = 0 ,
\
3.8 Ostrogradsky's Formula 243
since the équation cos(n, z) — 0 is fulfilled on the surface a 3
dZ{Xÿ — dxdy.dz
V
= $$Zcos(n, z)do + J$Zcos(n, z)do + ^ Z c o s(/i, z)da

at (T j (J»
But the sum of intégrais on the right of this équation is an in

tégral over the entire closed surface a; therefore,
l ^ l i ï d x d y d z ^ § § Z (X ’ y ' 2) c o s ("> 2) d a
J V a
Analogously, we can obtain the relations

[ \ { - ^ dx dydz --- (T Y (x , y, z) cos (n, y) do
J J J JJ
^ ^ ^ ■ d x d y d z = ^ X ( x , y, z)cos(n, x)do,
^V xJÜ
Adding together the last three équations term by term, we get

Ostrogradsky’s formula:*
= 5^ (Xcos(n, x) -j- Y cos (n, y)-\-Zcos(n, z))do (2)

a
dX dY dZ
The expression -g^- + -^--l--^-is called the divergence of the vec-
tor (or the divergence of the vector function):
F = X l + Yj+Zk
and is denoted by the symbol div F:
,. „ dX . dY . dZ
dlv F = -W + -dÿ + ~dT
We note that this formula holds true for any domain which
may be divided into subdomains that satisfy the conditions indi-
cated at the beginning of this section.
Let us examine a hydromechanical interprétation of this formula.
Let the vector F —X l + Y j + Z k be the velocity vector of a
liquid flowing through the domain V. Then the surface intégral in
* This formula (sometimes called the Ostrogradsky-Gauss formula) was disco-
vered by the noted Russian mathematician M. V. Ostrogradsky (1801-1861) and
published in 1828 in an article entitled “A Note on the Theory of Heat”.
I(>
formula (2 ) is an intégral of the projection of the vector F on the

outer normal n\ it yields the quantity of liquid flowing out of
domain V through the surface a in unit time (or flowing into V if the
intégral is négative). This quantity is expressed in terms of the
triple intégral of div F.
If div F==0f then the double intégral over any closed surface
is equal to zéro, that is, the quantity of liquid flowing out of
(or into) something through any closed surface a will be zéro (no
sources). More precisely, the quantity of liquid flowing into a domain
is equal to the quantity of liquid flowing out of thedomain.
In vector notation, Ostrogradsky’s formula has the form
S5SdivF<fo=$$f>*ds (i')
V a
and reads: the intégral of the divergence of a vector field F exten-

ded over some volume is equal to the vector flux through the surface
bounding the given volume.
3.9 THE HAMILTONIAN OPERATOR

AND SOME APPLICATIONS
Suppose wehâve a function u = u(x, y, z). At eachpoint of the
domain in which the function u(x, y , z) is definedanddifféren
tiable, the following gradient is determined:
grad« „ = /.d-u+ . , .d_ u +, *- du
_ /tv
(i)
The gradient of the function u(x, y, z) is sometimes denoted as
follows:
_ .d u , .d u , . du ,n .
^ u = l d-x + J r y + k ^ (2)
The symbol V is read “del”.
(1) It is convenient to Write équation (2) symbolicaliy as
v u = { l Tx+J Ty + k h ) u (2')
and to consider the symbol
v ==iêc+ Jdjj + fo- k

as a “symbolic vector”. This symbolic vector is called the Hamil-
tonian operator or del operator (V-opérâtor). From formulas (2 )
and (2 ') it follows that “multiplication” of the symbolic vector v
by the scalar function u gives the gradient of this function:
Vu = grad u (4)
3.9 The Hamiltonian Operator and Some Applications 245
(2) We can form the scalar product oî the symbolic vector y by

the vector F = IX + JY + kZ:
T M , K + 4 + *£)<,Jt+'/K+*z>
- s x + i ; r + dz
d dz
(see Sec. 3.8). Thus,
y F = d iv F (5)
(3) Form the vector product of the symbolic vector y by the vector
F = t X + j Y + kZ:
' xF- ( ' I + 4 + * -s-) x (W + - ^ + t2 )

d_ d_ d_ _d_
i j * A L
d_ d_ d_ dy dz dx dz dx dy
dx dy dz Y Z X Z -\-k X Y
X Y Z
* \ dy dz J J { d x dz ) ' " \ dx dy )
j(d Z dY \ , . f d X à Z \ . u (dY dX \ , «
= i [ r y - ^ ) + J [ ^ - d 7 ) + k { w - - d i ) ^ CUTV F
(see Sec. 3.7). Thus
Vx F — curl F (6 )
From the foregoing it follows that vector operations may be
greatly condensed by the use of the symbolic vector y. Let us
consider several more formulas.
(4) The vector field F(x, y, z) = i X + J Y + kZ is called a poten-
tial vector field if the vector F is the gradient of some scalar
function u(x, y, z):
F = g ra d «
or
t du , jdu , g^du
F = #l dx~
£ + / dy
= +1 * dz
£
In this case the projections of the vector F will be
y _ dtt
Y = du z = fdz-
A ~~dx' dy’
From these équations it follows (see Sec. 8 . 12 , Vol. I) that
âX dY ÔY dZ dX dZ
Hence, for the vector F under considération,

curl F = 0
Thus, we get
curl (grad u) — 0 (7)
Applying the del operator V» we can write (7) as follows [on the
basis of (4) and (6 )]:
(V X V u) = 0 (T)
Taking advantage of the property that for multiplication of a
vector product by a scalar it is sufficient to multiply the scalar
by one of the factors, we write
(VXV)« = 0 (7")
Here, the del operator again has the properties of an ordinary
vector; the vector product of a vector into itself is zéro.
The vector field F(x, y, z), for which curl F = 0, is called irro-
tational. From (7) it follows that every potential field is irrota-
tional.
The converse also holds: if some vector field F is irrotational,
then it is potential. The truth of this statement follows from rea-
soning given at the end of Sec. 3.7.
(5) A vector field F (x , y, z) for which
div F = 0
that is, a vector field in which there are no sources (see Sec. 3.8)
is called solenoidal. We shall prove that
div (curl F) = 0 (8 )
or that the rotational field is source-free.
Indeed, if F = i X + j ’Y + kZ, then
1P .(d Z dY \ . , ( dX d Z \ , u (dY d X \
cn T lF - i { - d i - ^ r ) + J { - d r - ^ r ) + k { ^ - - d i )
and therefore
. . . . » d f dZ âY \ . d I dX dZ \ , d ( dY dX \ n
d.v (curl F) = Tx [- w — w ) + ry [ i » - i > ï ) + - d ï ( i i ï r - w ) = 0
Using the del operator, we can write équation (8 ) as
V (V x F) = 0 (8')
The left side of this équation may be regarded as a vector-scalar
(mixed) product of three vectors: V, V, F, of which two are the
same. This product is obviously equal to zéro.
(6 ) Let there be a scalar field u = u(x, y, z). Détermine the
gradient field:
j .du , .du , . du
grad u = t rx + J Tÿ + k w
Then find
div (grad «) = ! ( ! ) + £ ( % ) + ê { l i ï )
or
dj 1•v(grad
/ i \
„ ) =d2u
_ +. ^d2*/5. +, _d2a /ftV
(9)
The right member of this expression is denoted by
* d2u , d2u , d2u
( 10)
or, symbolically,
A / a2 . a2 . d2 \
A u ~ { dx* + dy2 + dz2 ) “
The symbol
is termed the Laplacian operator.

Hence, (9) may be written as
div (grad u) — Au (H)
Using the del operator V we can Write (11) as
(VV «) = Au, i.e., A = V2 (\V)
We note that the équation
d2u . 3*11 . d2u __
dië + dÿ*"1"H? ~ U (12)
or
Au = 0 ( 12')
is called the Laplace équation. The function that satisfies the Lap-
l.ace équation is called a harmonie function.
Compute the following line intégrais:
1. J y 2 dx + 2xy dy over the circle x — a c o s t %y = a s in t. Ans. 0 .
2 . J y d x —x d y over an arc of the ellipse jt = a c o s /, y = b s in t.
Ans. —2ji ab.
3' Î { x t + V*dX x2+ y2<iy) over a circle vvith centre at the origin. Ans. 0 .
4. J \ '^x2^ yX2 ^^) 0Ver 2 seêmen* °1 straight line y = x from jc=1 to

x = 2. Ans. In 2 .
5. ^ yz dx + xz dy + xy dz over an arc of the hélix x = acos t, y = a s \ n t t
z — ht as t varies from 0 to 2n. Ans. 0 .
6. ^ x d y — ydx over an arch of the astroid x = a c o s 3 / , y = a s i n r t.

Ans Jia2 (the double area of the astroid).
7. ^ x d y — ydx over the loop of the folium of Descartes x = j »

3a/*
Ans. 3a2 (the double area of the région bounded by the indicated
y ~ i +<*
loop).
8. ^ x d y — y d x over the curve x = a (/ — sin /), y = a (l — cos /) (0 < / < 2ji).
Ans.— 6naa (the double area of the région bounded by one arch of a cycloid
and the x-axis).
Prove that:
9 . grad (cq>) = cgrad 9 , where c is a constant.
10. grad (clq>+ c2\|5) = c1 grad <p+ c2 grad tf», where cx and c2 are constants.
11. grad (9 ^ = 9 grad 9 + 9 grad 9 .
12. Find grad r, grad r2, grad grad f (r)t where r = Y x 2+ y 2+ z 2.
Ans. - , 2r . - - J - , /' (r)-£ -.

13. Prove that div (A + B) = div A + div B.
14. Compute div r, where r — x i + y j + z k . Ans. 3.
15. Compute div(A<p), where A is a vector function and 9 is a scalar functkm.
Ans. 9 div A + (grad 9 A).
16. Compute div (r-c) where c is a constant vector. Ans.
(c*/*)7.
17. Compute div B(rA). Ans. A'&
Prove that:
18. rot (fjAi -\-c2A2) = cl rot Ax+ c2 rot A2, where C! and c2 are constants.
19. rot (Ac) = grad A x c , where c is a constant vector.
20. rot rot A = grad div A — A A.
21. A xrot 9 = rot (9 A).
Surface Intégrais
22. Prove that cos (nt z)da = 0 if o is a closed surface and n fs a
normal to it.
23. Find the moment of inertia of the surface of a segment of a sphere with
équation x2+ y2+ z2= R 2 eut off by the plane z=^H relative to the z-axis.
Ans. ^ ( 2 Æ » - 3 W /+ H9).
24. Find the moment of inertia of the surface of the paraboloid of révolution
x2+ y 2= 2cz eut off by the plane z — c relative to the z-axis. Ans. 4jic4 ^ ■.
lo
25. Compute the coordinates of the centre of gravity of a part of the surface
of the cône x2+ y2= j ^ z 2 eut off by the plane z = H . Ans. 0 , 0, - |- //.
26. Compute the coordinates of the centre of gravity of a segment of the
surface of the sphere x2+ y2+ z2= R 2 eut off by the plane z = //.
Ans. (o . 0, •
Exercise* on Chapter S 249
27. Find JJ [x cos (/fx) + y cos (ny) + z cos (nz)\ do, where o is a closed
a
surface. Ans. 3 F , where F is the volume of the solid bounded by the surface o.
28. Find J J z d x d y , where 5 is the external side of the sphere x2 + j /2 +
s
+ z2^=R2. Ans. y ji/? 3.
29. Find J J x2 dy dz + y dz dx + z dx dy,
2 2 where 5 is the external side of
s
the surface of the sphere x2-\-y2+ z2 — R2. Ans. 2jiRa.
30. Find J J Ÿ 'x 3+ y2ds, where 5 is the latéral surface of the cône
s
x* , y2 z2 n n L 2jia 2 V a2-f- b2
0 < Z< 6 . Ans. -------- 3--------- .
31. Using the Stokes formula, transform the intégral J y d x + td y + x d z .
L
(cos a + cos p + cos y) ds.

s
Find the line intégrais, applying the Stokes formula and directly:
J
32. (y + z) dx + (z + x) dy + (x+ y) dz, where L is the circle x2 + y2+ z2 = a2,
L
x + y + z = 0 . Ans. 0 . 33. J x2y3dx-\~dy-\-zdz, where L is the circle x2-\-y2= R2,

L
JlR*
2 = 0. Ans. —
8 *
Applying the Ostrogradsky formula, transform the surface intégrais into
volume intégrais:
34. J J (x cos a + y cos p + z cos y) ds. Ans. J J J 3 dxdydz = 3F.
s v
35* JJ(xa + yf + *a) {dy d z+ d x d z + dx dy). Ans. 2 J J J (x-\~y + z) dx dy dz.
s Jv
IjJ (S+^+Sr)
Using the Ostrogradsky formula compute the following intégrais:
38. J J (x c o sa + f c o s P + z c o s T ) * . where S is the surface of the ellipsofd
s
Î j + I j + ^ " = ^• Ans. 4nabc. 39. (x3 cos a + y3 cos P + 23 cos y) ds, where S
is the surface of the sphere x * + p * + z 2 = /?2. Ans. ^rnR *. 40. x* dy dz -

• 2^
-f y 1 dz dx-\ z2 dx dy, where 5 is the surface of the cône ^ ^ =0
Jlû262 p (*
(0 «^z^<6 ). 4ns. —^— .4 1 . \ \ x dy dz + y dxdz + zd xd y, where S is the
S
surface of the cylinder jc2 + t/2 = a2, —H ^ z ^ H . Ans. 3na2H. 42. Prove the
identity J J d*dy = J g “ ^s, where C is a contour bounding the
D c
domain £>, and is the directional dérivative of the outer normal.
dn
Solution.
H ( ^ + $ j r ) dxdy = f j — y àx + X d y = ^ l —Y cos (s, x) + X sin (s, x)]ds

d c C
where (s, x) is the angle between the tangent line to the contour C and the
x-axis. If we dénoté by (n, x) the angle between the normal and the x-axis,
then sin (s, x) = cos(n, x), cos (s, x) = — sin (n, x). Hence,
m ^ + 5- ) ^ ^ = J l ^ c o s(n. «) + K sin(n, x)]d*

D C
Setting X ; Y == , we get
I I ( £ + 0 ) ^ = 1 [ S cos(n- Xî + T y si n{n’ x>]*

D C
or
The expression called the Laplacian operator.

43. Prove the identity (called Green*s formula)
IIIV l”to -u à v )d x d y d * = ^
a
(o|-« ^do
where u and v are continuous functions with continuous dérivatives up to the

second order in the domain D.
The symbols Au and Av dénoté
A d2u . d2u . d2u A d2v . dao , d2v
A u=—dx* Atf ——dx* + ^4-+â?"
Solution. In the formula
III (^^<^^^)d*dyd*=!![Xcos(n’ cos("’y)+zc°s(n>

V o
zïïda
we put
X = vux —uvx
Y= VUy — UVy
Z= Vu'z — UVg
Then
^ - + g - + ^ - = V (u x x + U y y + u ’z z ) — U (v x x + V yy + v"z z ) = v A ll — U AV
X COS (/I, x) + Y COS (fl, COS (fl, 2)
= u[w*cos(/f, x) + tty cos (/i, y)-\-u'z cos («, 2)]

du dv
— COS (/I, *)+0yCOS(/f, y) + t>zCOS(/I, 2)] =
dn dn
Hence,
§ fâ (v A u -u A v )d x d y d z = tf ^ - u ^ d a
44. Prove the identity
u A d*u , , d2u
where = ,
Solution. In Green’s formula, which was derived in the preceding problem,
put u = l . Then Au = 0, and we get the desired identity.
45. If u(x, y, 2) is a harmonie function in some domain, that is, a function
which at every point of the domain satisfies the Laplace équation
d2u . d2u . ôflu
then
ws*-
a
where o is a closed surface.
Solution. This follows directly from the formula of Problem 44.
46. Let u (x , y, z) be a harmonie function in some domain V and let there
be, in Vf a sphere. a with centre at the point M (xlt y lt zt ) and with radius R .
Prove that
« f e . »x, z1) = ^ ¥ ^ u d a
Solution. Consider the domain Q bounded by two spheres a, a of radii R

and p (p < R) with centres at the point M (jclf y lt 2j). Apply Green’s formula
(found in Problem 43) to this domain taking for u the above-indicated function,
and for the function v,
_ _ 1 _ ___________________ 1_________________
r V ( X — Xi)"- + ( y — y 1)J+ (z— Zi)s
d2v
By direct différentiation and substitution we are convinced that -5- 3-
> , dx*
o+a
1 du
\T fa ~ ~ u
lz)dn • t o - j j i r Ê ï - '
lz
dn
do = 0
On the surfaces a and a the quantity ~ is constant and so can

be taken outside the intégral sign. By virtue of the resuit
It obtained
obt * in
■ Problem 45,
we hâve
à
Hence,
- $“ *+jâ[ u d a - [
but
dn dr
Therefore,
+
S S a ^ da- S S u ^ d a = 0
2 ô
(O
2 Ô
Apply the mean-value theorem to the intégral on the left:
( 2)
where u (g, t|, £) is a point on the surface of a sphere of radius p with centre
at the point M (x lt y u 2 ^ .
We make p approach zéro; then w (g, tj, £) - + u ( x l t y l t ZJ:
ct
Hence, as p -►0, we get
p j j u d o - m ( x t, yu *t) 4n
Further, since the right side of (1) is independent of p, it follows that as

p -►0 we finally get
F ' J i “ da=4îl“ (*1’ Vu *i)
â
Vu
CH A PTE R 4
SERIES
4.1 SERIES. SUM OF A SERIES
Définition 1. Suppose we hâve an infinité sequence of numbers *

Wj, * * *. ...
The expression
Ul + U2 + Ws+ • • • + U »+ • • • ( 1)
is called a numerical sériés. Here, the numbers ult u„ . . . , un, . . .
are called the terms of the sériés.
Définition 2. The sum of a finite number of terms (the first n
terms) of a sériés is called the nth partial sum of the sériés:
sn = u l + u 2 + ••• + u n
Consider the partial sums

Si = u ,
s2 = u1+ ut
s3 = u1+ u3+ u3
sn —u1+ wa + w, + . . . + un
If there exists a finite limit
s = Iim s„
n -►od
it is called the sum of the sériés ( 1) and we say that the sériés
converges.
If Iims„ does not exist (for example, s„—►oo as n —*■<»), then
rt—
►<*
we say that the sériés ( 1) diverges and has no sum.
Example. Consider the sériés
a+ aq + a ç * + . . . + a q " - '+ . . . (2)
This is a géométrie progression with first term a and ratio q (a 0).
* A sequence is considered specified if we know the law by which it is

possible to détermine any term un for a given n.
254 Ch. 4 Sériés
The sum of the first n ternis of the géométrie progression is (when q 7= 1)

a — aqn
sn = -
1 —q
or
a aqn
Sn 1— q ~ \-q
'<7
(1) If \q | < 1, then qn -►0 as n -* 00 and, consequently,
lim sn = lim ( ------- 7^ — ) = = -^ —
n -* o c \ 1 <7 1 <7 / ^ Q
Hence, in the case of | <7 1 < 1, the sériés (2) converges and its sum is
s=
1- q
d—CLQn
(2) If | q | > 1, then | qn | -►00 as n-> 00 and then -j— ^-----►± 00 as n -* 0 0 ,.
that is, lim sn does not exist. Thus, when \q \ > 1, the sériés (2) diverges.
(3) If <7=1, then the sériés (2) has the form
Cl -J - Cl -J - Q -J - . . •
In this case
sn = na, lim sn = 00
and the sériés diverges.

(4) If q = — 1, then the sériés (2) has the form
a — a + a —a +...
In this case
0 when n is even
a when n is odd
Thus, sn has no limit and the sériés diverges.
Thus, a géométrie progression (with first term different from zéro) converges
only when the ratio of the progression is less than unity in absolute value.
Theorem 1. If a sériés obtained from a given sériés (1) by sup

pression of some of its ternis converges, then the given sériés itself
converges. Conversely, if a given sériés converges, then a sériés
obtained from the given sériés by suppression of several terms also
converges.
In other words, the convergence of a sériés is not affected by the
suppression of a finite number of its terms.
Proof. Let s„ be the sum of the first n terms of the sériés (1),
ck, the sum of k suppressed terms (we note that for a sufficiently
large n, ail suppressed terms are contained in the sum s„), and
o„_A, the sum of the terms of the sériés that enter into the sum sn
but do not enter into ck. Then we hâve
Â ® n —k
where ck is a constant that is independent of n.

4.1 Sériés. Sum of a Sériés 255
From this relation it follows that if Iim tr„_A exists, then Iim sn
/!-*• oo n -* &
exists as well; if Iim s„ exists, then lim also exists; which
«-♦■ao n -+ oo
proves the theorem.
We conclude this section with two simple properties of sériés.
Theorem 2 . I f a sériés
ai + a2 + ••• (3)
converges and its sum is s, then the sériés
ca1+ cai + . . . (4)
where c is some fixed number, also converges, and its sum is es.
Proof. Dénoté the nth partial sum of the sériés (3) by s„, and
that of the sériés (4), by o„. Then
o„ = ca1+ . . . +ca„ = c (û , + . . . +a„) = cs„
It is now clear that the limit of the nth partial sum of the
sériés (4) exists, since
Iim o„ = Iim (cs„) = c lim sn = cs
n -* c o n -* oo /i-eoo
Thus, the sériés (4) converges and its sum is equal to es.
Theorem 3. If the sériés
a1+ a2+ . . . (5 )
and
bi + ^2 + • • • (6 )
converge and their sums, respectively, are s and s, then the sériés
(ax + bj) '+ (a2+ b2) + . . . (7)
and
(a1—6 i) + (æ2—&2) + • • • (8 )
also converge and their sums are s + s and s—s, respectively.
Proof. We prove the convergence of the sériés (7). Denoting its
nth partial sum by_a„ and the nth partial sums of the sériés (5)
and (6 ) by sn and s„, respectively, we get
a n = (a i + î) + • • • + (a n + b n)
= (al + . . . -\-an) + (b1+ . . . +bn) = sn-\ sn

Passing to the limit in this équation as n —* oo, we get
lim an = lim (sn + sn) = lim sn + lim sn = s + s
n -*■oo n-*- oo n-*-co n -+ &
Thus, the sériés (7) converges and its sum is s -fs .

256 Ch. 4 Sériés
Similarly we can prove that the sériés (8 ) also converges and

its sum is equal to s —s.
We say that the sériés (7) and (8 ) vvere obtained by means of
termwise addition or, respectively, termwise subtraction of the
sériés (5) and (6 ).
4.2 NECESSARY CONDITION FOR CONVERGENCE

OF A SERIES
One of the basic questions, when investigating sériés, is that of

whether the given sériés converges or diverges. We shall establish
sufficient conditions for one to décidé this question. We shall also
examine the necessary condition for convergence of a sériés; in
other words, we shall establish a condition for which the sériés
will diverge if it is not fulfilled.
Theorem. If a sériés converges, its nth term approaches zéro as n
becomes infinité.
Proof. Let the sériés
U1 + U2 + u s + ••• + u n + •••
converge; that is, let us hâve

Iim sn —s
fl-*-»
where s is the sum of the sériés (a finite fixed number). But then
we also hâve the équation
lim sn_! = s
n -+ oo
since (n — 1) also tends to infinity as n —►oo. Subtracting the

second équation from the first termwise, we obtain
lim s„—Iim sn_j = 0
/!-►00 n->QO
or
lim (s„—s„_1) = 0
But
sn— s„-! = ua
Hence,
lim «„ = 0
fl-*00
which is what was to be proved.
Corollary. If the nth term of a sériés does not tend to zéro as
n —►oo, then the sériés diverges.
4.2 Necessary Condition for Convergence of a Sériés 257
Example. The sériés
T + l + f + - + 2% + -
diverges, since
Bm u„ = lim f
n -* od n -*■ ce \ “•* i ^ ^
We stress the fact that this condition is only a necessary con
dition, but not a sufficient condition; in other words, from the
fact that the nth term approaches zéro, it does not follow that the
sériés converges, for the sériés may diverge.
For example, the so-called harmonie sériés
1+ t + t + t + ---+4'+--*
diverges, although
lim u„= lim —= 0
To prove this, write the harmonie sériés in more detail:
1+ T + ÿ + T + î '+ 'g '+ T + T
+ ¥ + T ô + i7 + Ï2 + B + ü + T 5 + T 6 + l7 + W
V ■ 1■V
We also write the auxiliary sériés
1 ^j 2Lj' 4Lj‘ 4! |‘ 8Lj‘ 8Lj‘ 8Lj' 8 L

16 terms
+ 16"*” 1 6 + 1 6 + 1 6 + 1 6 + 1 6 + 1 6 + 1 6 + 3 2 + ‘ ‘ ' + 3 2 + ' * ' ^

The sériés (2) is constructed as follows: its first term is equal
to unity, its second is */,. its third and fourth are V«. the fifth
to the eighth terms are equal to V». the terms 9 to 16 are equal
to 1/lt, the terms 17 to 32 are equal to 1/,î, etc.
Dénoté by s p the sum of the first n terms of the harmonie sériés (1)
and by stf* the sum of the first n terms of the sériés (2 ).
Since each term of the sériés ( 1) is greater than the corres-
ponding term of the sériés (2 ) or equal to it, then for n > 2
s(n" > & (3)
We compute the partial sums of the sériés (2 ) for values of n
equal to 2 \ 2 2, 2 8, 2 4, 2 S:
S2 = 1 + Y = *+ l "2
17-2082
258 Ch. 4 Sériés
s4= 1 + T + ( t + t ) = 1 +T+ T= 1+ 2 'T

S ,= 1 + t + (t + t ) + (t + T + T + t ) = 1 + 3 ’T
Si.= 1 + 7 + ( t + t ) + ( t + ••• + t ) + ( îÎ + • • • + !? )
4 terms 8 terms
= 1+ 4' ÿ
16 terms
in the same way we find that s2»= 1+ 6 - ÿ , s ^ = l + 7 - - - and,

generally, s2*= 1 -f £ — •
Thus, for sufficiently large k, the partial sums of the sériés (2 )
can be made greater than any positive number; that is,
Iim s},2) = oo
but then from the relation (3) it also follows that
lim s»> = oo
which means that the harmonie sériés ( 1) diverges.
4.3 COMPARING SERIES WITH POSITIVE TERMS
Suppose we hâve two sériés with positive terms:

«l + Wj+ Uj-f ( 1)
Vi + t'2 + i>3+ • • • ••• (2 )
For them the following assertions hold true.
Theorem 1 . If the terms of the sériés (1 ) do not exceed the cor-
responding terms of the sériés (2 ); that is,
un<v„ (n= 1, 2, . . . ) (3)
and the sériés (2 ) converges, then the sériés ( 1) also converges.
Proof. Dénoté by s„ and o„, respectively, the partial sums of
the first and second sériés:
n n
= 2 u,, <*„ = 2 v(
i= 1 i- 1
4.3 Cotnparing Sériés with Positive Ter ms 259
From the condition (3) it follows that

Sn < (4)
Since the sériés (2) converges, its partial sum has a limit a:
lim o„ = a
n-*-ao
From the tact that the terms of the sériés (1) and (2) are posi
tive, it follows that a„ < a, and then, by virtue of (4),
sn < o
We hâve thus proved that the partial sums s„ are bounded. We
note that as n increases, the partial sum s„ increases, and from
the fact that the sequence of partial sums is bounded and increases,
it follows that it has a limit *
lim s„ = s
n -> c o
and it is obvious that

s <;<j
Using Theorem 1, we can judge of the convergence of certain
sériés.
Example 1. The sériés
l + ^ + p ,+ î ï + - - - + ^ + - - -
converges because its terms are smaller than the corresponding terms of the
sériés
1 l± + ± + ± + 4 -1 +
1 r 22 i 23 i 24 ' **‘ ' 2n~ *‘ ’
But the latter sériés converges because its terms, beginning with the second,
form a géométrie progression with common ratio - y . The sum of this sériés
is equal to 1 1 . Hence, by virtue of Theorem 1, the given sériés also conver
ges, and its sum does not exceed 1 1 .
T heorem 2. If the terms of the sériés (1) are not smaller than
the corresponding terms of the sériés (2 ); that is,
u„> vn (5)
and the sériés (2 ) diverges, then the sériés ( 1) also diverges.
* To convince ourselves that the variable s n has a limit, let us recall a con
dition for the existence of a limit of a sequence (see Theorem 7, Sec. 2.5, Vol. I):
“if a variable is bounded and increases, it has a limit.” Here, the sequence of
sums sn is bounded and increases. Hence it has a limit, i.e., the sériés con
verges.
17 :i:
260 Ch. 4 Sériés
Proof. From condition (5) it follows that

S„ > (6 )
Since the terms of the sériés (2 ) are positive, its partial sum aB
increases with increasing n, and since it diverges, it follows that
liin on = oo
But then, by virtue of (6 ),

liin sn = oo
n-* 8
i.e., the sériés ( 1) diverges.
1+ — + — +•••■
Vn
diverges because its ternis (from the second on) are greater than the correspon-
ding terms of the harmonie sériés
'-‘4 + T + - + T + -
which, as we know, diverges.
N o te . Both the conditionsthat we hâve proved (Theorems 1
and 2) hold only for sériés with positive terms. They also hold
true when some of the terms of the first or second sériés are zéro.
But these conditions do not hold if some of the terms of the sériés
are négative numbers.
4.4 D’ALEMBERT’S TEST
T heorem ( d ’A le m b e rt’s te s t) . If in a sériés with positive terms

« l + «2 + « 3 + - - - + « „ + . - - (1)
the ratio of the (n-\-\)th term to the nth term, as n —<-oo, has a
(finite) limit l, that is,
lim -^ -= / (2 )
rt-+a>
then:
( 1) the sériés converges for l < 1
(2 ) the sériés diverges for l > 1 .
(For / = 1, the theorem does not yield the convergence or diver
gence of the sériés.)
P roof. (1) Let / < 1. Consider a number q ( l < . q < 1) (Fig. 109).
From the définition of a limit and relation (2) it follows that
for ail values of n after a certain integer N, that is, for N,
4.4 D*Alembert's Test 261
we will hâve the inequality

< <7 ( 2 ')
Indeed, since the quantity *1 tends to the limit /, the dif

férence between the quantity and the number l may (after
a certain N) be made less (in absolute value) than any positive
number, in particular less than q —/; that is,
ht +1 __^
<q—l
Inequality (2') follows from this last inequality. Writing ine
quality (2') for various values of n, from N onwards, we get
un + x< <J UN
«A+2 < jv*i < <12un
u N+3 < Qu y + a < Q 3u jv
(3)
Now consi der the two sériés

U1 "f" u i “H W3 -f- . . . + M j v + u A,+ l~f“ MA'+2- f” • • • (1)
uN+q u N+ q îuN+ . . . ( 1')
The sériés (1') is a géométrie progression with positive common
ratio q < 1 . Hence, this sériés converges. The terms of the
q-L q-l L-l L-l
l üjnil 1
“a
Fig. 109 Fig. 110
sériés ( 1), after uy+1, are less than the terms of the sériés (T).
By Theorem 1, Sec. 4.3, and Theorem 1, Sec. 4.1, it follows that
the sériés ( 1) converges.
(2) Let / > 1. Then from the équation limî± i = / (where / > 1)
n—
►ao un
it follows that, after a certain N , that is for n ^ N , we will hâve
the inequality
(Fig. 110), or ull+1> u ,t for ail n ^ N . But this means that the
terms of the sériés increase after the term A f+ l, and for this
reason the general term of the sériés does not tend tozero. Hence,
the sériés diverges.
262 Ch. 4 Sériés
Note 1. The sériés will also diverge when lim ^ L±1 = oo. This
n -+ ao un
follows from the fact that if l i m ^ ±i = oo, then after a certain

fi —
►oc 11n
n — N we will hâve the inequality î^2 + l> 1, or un+1 > un.

un
Example 1. Test the followirig sériés for convergence:
i j —!— |— !— L . j --------!--------- 1_ .
Solution. Here,
1 1 _ 1 _ 1
Un~ 1-2 . . . . . n ni W', + 1“ 1. 2 . . . . . / i ( n + l ) (n + l)l
un+i ni l
Un (« + !)! n + l
Hence,
lim f ^ ± i = lim — = 0 < i
a-» » un a-* » n -j- 1
The sériés converges.
Example 2 . Test for convergence the sériés
2 2a 2;J 2n
1+ T + T + - + T + -
Solution. Here,
_2n
un — n » un + 1— T + T ' ZH±1= 2 lim îîa ± l= lim 2 - ^ = 2 > 1
Un n+ 1 n-*-» /l+ 1
The sériés diverges and its general terni un approaches infinity.
Note 2 . D’Alembert’s test tells us whether a given positive

sériés converges; but it does so only when lim exists and is
Fl —
►oc Un
different from 1. But if this limit does not exist or if it does
exist and lim — ^ l , then d’Alembert’s test does not enable us
fl-*oc un
to tell whether the sériés converges or diverges, because in this
case the sériés may prove to be both convergent and divergent.
Some other test is needed to détermine the convergence of such
sériés.
Note 3. If lim - — = l , but the ratio for ail n (after a
n - * oc un Un
certain one) is greater than unity, the sériés diverges. This follows
from the fact that if ^ > 1, then un+l>u„ and the general
I ^
term does not approach zéro as n —>oo.
To illustrate, let us examine some examples.
4.4 D'Alembert's Test 263
Example 3. Test for convergence the sériés

J_[_A_L.A_i_ _|__-__L
Solution. Here,
n+ I
lim “s±I=l«m .. n* + 2 n + \ .
J L tl, lim — — r 1—= 1
n -« n2 + 2/i
n+ 1
In this case the sériés diverges because > 1 for ail n:
nn
un +i _ri1+ 2n -\-1
un n2-\-2n > 1
Example 4. Using the d’Alembert test, examine the harmonie sériés
l + T + ÿ + - - - + 'i ‘+---
We note that wn = — , un +\ ~ and, consequently.
lim lim —y—r = 1

n -*» n-*- oo W-|- 1
Thus, d’Alembert’s test does not allow us to détermine the convergence or
divergence of the given sériés. But we earlier found out by a different expedient-
that a harmonie sériés diverges.
1 -2 + 2 .3 + 3 -4 '* • • • + / i ( n + l ) + - “
Solution. Here,
1 1
“B— n ( n + l ) ’ “" + 1 “ l n + l ) ( n - f 2)
lim ffü ± l= lim »(«+!) lim — = 1
Un
AZ-* « n-*-« (»+!)(« + 2) «-*-» n~r2
D’Alembert’s test does not permit us to infer that the sériés converges;
but by other reasoning we can establish the fact that this sériés converges.
Noting that
1 _ 1 ___ 1_
« ( n |l ) - » n-j- 1
we can Write the given sériés in the form
The partial sum of the first n te rn is , after removing brackets and cancell-
ing, is
1
sn —
n+\
264 Ch. 4 Sériés
Hence,
lim s„ = üm ( = l
« -♦ o o n -t-o o \ n -r 1 /
that is, the sériés converges and its sum is 1.
4.5 CAUCHY’S TEST

Theorem (Cauchy’s test). If for a sériés with positive terms
_ ul + u.i + u3+ . . . + « „ + . . . ( 1)
the quantity \/u„ has a finite limit l as n —*-oo, that is,
1 lim ’{ / un — l
n-*ao
therr. ( 1) for / < 1 , the sériés converges
(2 ) for l > 1 , the sériés diverges.
Proof. (1) Let / < 1. Consider the number q that satisfies the
relation / < q < 1 .
After some n = N we will hâve the relation
\ ÿ ü a- i \ < q - i
whence it follows that _
y ü n< q
or
un <Qn
for ail n ~ ^ N .
Now consider two sériés:
U1+ ut+ u s + •••+ uN~f~UN+14" UjV+t+ • • • O)
qN + q N * l + q/ f + t + . . . ( 1')
The sériés (T) converges, since its terms form a decreasing
géométrie progression. The terms of the sériés (1), after uN, are
less than the terms of the sériés (T). Consequently, the sériés ( 1)
converges,
(2) Let l > 1. Then, after some n = N, we will hâve
Î^ > 1
or
»n> 1
But if ail the terms of this sériés, after uN, exceed 1, then the
sériés diverges, since its general term does not tend to zéro.
Example. Test for convergence the serîes
t + ( t ) + ( t ) + - - - + ( 2^ n ) " + - - -
4.5 Cauchy*s Test 265
Solution. Apply the Cauchy test:

fl 1
lim < 1
ri-*- « n-* » V ( s t t ) ”= .1r . 2 2
The sériés converges.
Note. As in the d ’Alembert test, the case

lim {/«„ = / = 1
n -+ a o
requires further investigation. Among the sériés that satisfy this

condition are convergent and divergent sériés. Thus, for the har
monie sériés (which is known to be divergent)
lim \ / u n — \\m 1/ —= 1
n-+<x> n -+ rc T fl
To be sure, we shall prove that limln y —= 0. Indeed,

/!->Q
O W n
1-
lim iln 1
-S f 1
/ —= 1-lim ------
—In/i
n —> cd * ^ n -+ c c fl
Here, the numerator and dénominator of the fraction approach

infinity. Applying l’Hospital’â rule, we find
_ __ 1_
limln l / " —= lim ~ lnfl= lim —r^- = 0
n-*-cc " n n -r oc n n—
►qd *
Thus, ln ——*-0, but then Ÿ 1 — 1 , i.e.,

lim
/7-*-CC
For the sériés
“1 +
"T— +
^ —
32 +
^ +
^ —
n2 ^+
we also hâve the equality
lim y/ un — lim i X X = lim
n—
X.TT ^ ” /Î-+-QO
but this sériés converges, since if we suppress the first term, the
terms of the rcmaining sériés will be less than the corresponding
terms of the converging sériés
_ L i _L_ j- j ___ !___ l
1-2 ' 2-3 ^ * * * ^ n ( / i + l ) ^ * * ‘
(see Example 5, Sec. 4.4)f
266 Ch. 4 Sériés
4.6 THE INTEGRAL TEST FOR CONVERGENCE

OF A SERIES
Theorem. Let the terms of the sériés
+ + *• *+ ^ /i+ • • • (0
be positive and nonincreasing, that is,
( 1')
and let f (x) be a continuous nonincreasing function such that
f ( 1) = « 1 . / ( 2 ) = « 2.................... /(« )-= « » (2)
7 /im following assertions hold true:
( 1) if the improper intégral
S / (*) ^
converges (see Sec. 11.7, Vol. I), then the sériés (1) converges too;
(2 ) if the given intégral diverges, then the sériés ( 1) diverges
as well.
Proof. Depict the terms of the sériés geometrically by plotting

on the x-axis the terms 1, 2, 3, . . . , n, n + 1, . . . of the sériés,
and on the i/-axis, the corresponding values of the terms of the
sériés «„ u2, . . . . u„, . . . (Fig. 111 ).
In the same coordinate System plot the graph of the continuous
nonincreasing function
y=f(x)
which satisfies condition (2 ).
An examination of Fig. 111 shows that the first of the construct-
ed rectangles has base equal to 1 and altitude / ( 1) = The
area of this rectangle is thus ux. The area of the second one is u2,
and so on; finally, the area of the last (nth) of the constructed
rectangles is un. The sum of the areas of the constructed rectangles
4.6 The Intégral Test for Convergence of a Sériés 267
is equal to the sum s„ of the first n terms of the sériés. On the

other hand, the step-like figure formed by these rectangles embraces
a région bounded by the curve y = f (x\ and the straight lines
jc = 1, x — n + 1 , y - 0 ; the area of this région is equal to
n+1
^ f(x)dx. Hence,
i
n+I
S„ > S / (X) dx (3)
î
Let us now consider Fig. 112. Here the first of the constructed
rectangles on the left has altitude u2; and so its area is u2.
The area of the second rectangle is and so forth. The area
of the last of the constructed rectangles is u,l+1. Hence, the sum
of the areas of ail constructed rectangles is equal to the sum of
ail terms of the sériés beginning with the second to the (n + l)th
or sn+i — «i- On the other hand, it is readily seen that the step-
like figure formed by these rectangles is contained within the
curvilinear figure bounded by the curve y =f(x) and the straight
lines x - 1, x = n + 1 , y = 0 . The area of this curvilinear figure is
n+I
equal to J f(x)dx. Hence,
î
n+ 1
S„+1— «1 < S f(x)dx
i
whence
n+ 1
s„+i < S f(x)dx + ul (4)
Let us now consider both cases.

1. We assume that the intégral J f (x) dx converges, that is,
î
has a finite value.
Since
n+1
5 f{x)dx< ^ f(x)dx
î î
it follows, by virtue of inequality (4), that
00
sn < s„ +1 < \ f ( x ) d x + Ui
1
Thus, the partial sum sn remains bounded for ail values of n.
But it increases with increasing n, since ail the terms un are
268 Ch. 4 Sériés
positive. Consequently, s„ (as n —<-oo) has the finite limit

lims„ = s
n -* c c
and the sériés converges.
2. Assume, further, that ^ f ( x ) dx = oo. This means that
î
+i
^ f(x)dx increases without bound as n increases. But then, by
virtue of inequality (3), sn likewise increases indefinitely with n\
the sériés diverges.
The proof of the theorem is complété.
Note. This theorem remains valid if inequalities (T) are ful-
filled only from some integer N onwards.
Example. Test for convergence the sériés
-+ -+ -+ +— +
Solution. Apply the intégral test, putting
f(x)=Tp
This function satisfies ail the conditions of the theorem. Consider the intégral
1 1
Cdx= x l ~P (N1-P — 1) when p ?= 1
1- P 1- P
J xP
ln x |? = lnyV when p = 1
Allow N to approach infinity and détermine whether the improper intégral
converges in various cases.
It will then be possible to judge about the convergence or divergence of the
sériés for various values of p.
00
For p > 1, ^ — 1— > intégral is finite and, hence, the sériés
l
converges;
ce
P dx
for p < 1, \ — = o o , the intégral is infinité, and the sériés diverges;
1
00
P dx
for p = l , \ —= o o , the intégral is infinité, and the sériés diverges.
1
We note that neither the d’Alembert test nor the Cauchy test, which were
considered earlier, décidé whether the sériés is convergent or not, since
4.7 Alternating Sériés. Leibniz* Theorem 269
4.7 ALTERNATING SERIES. LEIBNIZ’ THEOREM

So far we hâve been considering sériés whose terms are ail
positive. In this section we consider sériés whose terms hâve
alternating signs, that is, sériés of the form
Ui — U2 + U3— U4 + •••
where u19 u2, . . . , un, . . . are positive.
Leibniz* Theorem. If in the alternating sériés
— u2+ u3— ut + . . . ( u „ > 0 ) ( 1)
the terms are such that
«1 > «2 > u3 > . . . (2)
and
limu„ = 0 (3 )
then the sériés ( 1) converges, its sum is positive and does not exceed
the first term.
Proof. Consider the sum of the first n-=2m terms of the sériés (1):
Sîm = («l —«2) + («3 —«4) + • • •
From condition (2) it follows that the expression in each of the
brackets is positive. Hence, the sum s2m is positive,
and increases with increasing m. Now Write this sum as follows:

= — («2 — « 3) — («4 — « . ) — • • • — ( « 2 » - 2 — « 2 * - l ) — « 2 »
By virtue of condition (2), each of the parenthèses is positive.

Therefore, subtracting these parenthèses from ut we get a number
less than ult or
Sim < «x
We hâve thus established that stm increases with increasing m
and is bounded above. Whence it follows that s2m has the limit s:
lim sî(B= s
m-► oo
and
0 < s <u,
However, we hâve not yet proved the convergence of the sériés;
we hâve only proved that a sequence of “even” partial sums has
as its limit the number s. We now prove that “odd” partial sums
also approach the limit s.
Consider the sum of the first n = 2 m + l terms of the sériés (1):
S'im+■^ S2b»"t" utm+1
270 Ch. 4 Sériés
Since, by condition (3), lim u 2m+1 = 0, it follows that

m - + oc
lim s2nj+1 = lim s2(„ + lim «2m+1= lim sîm = s

m->- ao m-*- x m->co m->»oo
We hâve thus proved that lim sn = s both for even n and for
n -+ oo
odd n. Hence, the sériés (1) converges.
Note 1. The Leibniz theorem holds if inequalities (2) hold true
from some N onwards.
Note 2 . The Leibniz theorem may be illustrated geometrically
as follows. Plot the following partial sums on a number line
(Fig. 113)
s2= ul — u2= s1— u21 s3 = s2+ u3, s4 = s3 — w4, S5 = s4 + WB,
etc.
The points corresponding to partial sums will approach a certain
point s, which depicts the sum of the sériés. Here, the points
corresponding to the even
----------------------- u i partial sums lie on the left of
“J s, and those corresponding
--------- to odd sums, on the right
^ ------- U s ------- ** of s.
Note 3. If an alternat-
ing sériés satisfies thestate-
ment of the Leibniz theo
Fig. 113 rem, then it is easy to eva-
luate the error that results
if we replace its sum, s, by the partial sum sn. In this substi
tution we suppress ail terms after un+1. But these numbers form
by themselves an alternating sériés, whose sum (in absolute value)
is less than the first term ofthis sériés (that is, less than un+l).
Thus, theerror obtainedwhenreplacing s by sn does not exceed
(in absolute value) the first of the suppressed terms.
1 _ ±2+-r±3_ ± _
4 L1 .
converges, since (1) 1 > — > — > . . . ; (2) lim un = lim —= 0 .

^ O n-*oo fl-*, oo fl
The sum of the first n terms of this sériés
1 , 1
*»=1 - Y + 3— 4 +••
•+(-l)" +1-
dififers from the sum s of the sériés by a quantity less than
»+l *
21 ^3!
converges by virtue of the Leibniz theorem.
4.8 P lus-and-Minus Sériés. Abzolute and Conditional Convergence 271
4.8 PLUS-AND-MINUS SERIES.

ABSOLUTE AND CONDITIONAL CONVERGENCE
We give the name plus-and-minus serîes to a sériés that has

both positive and négative terms.
Obviously, the alternating sériés considered in Sec. 4.7 is a
spécial case of plus-and-minus sériés. *
We shall consider some properties of alternating sériés.
In contrast to the agreement made in the preceding section we
will now assume that the numbers u19 w2, ...» un . . . can be both
positive and négative.
First, let us give an important sufficient condition for the con
vergence of an alternating sériés.
Theorem 1 . If the alternating sériés
+ u2+ . . + un -f- . . . ( 1)
is such that a sériés made up of the absolute values of its terms
I“ i I+ 1“*I+ • • • + 1“«. I+ • • • (2 )
converges, then the given alternating sériés also converges.
Proof. Let sn and an be the sums of the first n terms of the
sériés ( 1) and (2 ).
Also, let s'n be the sum of ail the positive terms» and s", the
sum of the absolute values of ail the négative terms of the first
n terms of the given sériés; then
8n = 8n Gn = Sn Sn
By hypothesis, on has the limit a; s' and s" are positive in-
creasing quantities less than a. Consequently, they hâve the
limits s' and s". From the relationship sn = s'n— Sn it follows that
sn also has a limit and that this limit is equal to s' —s", which
means that the alternating sériés ( 1) converges.
The above-proved theorem enables one to judge about the con
vergence of certain alternating sériés. In this case, testing the
alternating sériés for convergence reduces to investigating a sériés
with positive terms.
Consider two examples.
sin a . sin 2a , sin 3a . . sin na .
i r + ^ r + - 3r + - + i r + - (3)
where a is any number.
* In this English édition we shall use the terni alternating sériés for both
types.—Tr.
272 Ch. 4 Sériés
Solution. Also consider the sériés

sin a sin 2 a sin 3 a 1 , .
l2 + 22
4-
3’ ! + ••• +
(4)
and
7*+ 55-+7^-+ • ■• + •• • (5)

The sériés (5) converges (see Sec. 4.6), The terms of the sériés (4) do not
exceed the corresponding terms of the sériés (5); hence, the sériés (4) also con
verges. But then, in virtue of the theorem just proved, the given sériés (3)
likewise converges.
n
cos —
3ji cos ■5ji (2n — 1) ji
4
( 6)
32 33 3"
Solution. In addition to this sériés, consider the sériés
T + i + 'F + - - - + 3 * ‘+ --- (7)

This sériés converges because it is a decreasing géométrie progression with
ratio — . But then the given sériés (6 ) converges, since the absolute values
O
of its terms are less than those of the corresponding terms of the sériés (7).
We note that the convergence condition that was proved earlier

is only a su fficien t condition for convergence of an alternating
sériés, but not a necessary condition: there are alternating sériés
which converge, but sériés formed from the absolute values of
their terms diverge. In this connection, it is useful to introduce
the concepts of absolute and conditional convergence of an alter
nating sériés and, on the basis of these concepts, to classify alter
nating sériés.
D é fin itio n . The alternating sériés
Ul + U2 + U3 + •• • + Un + ••• (1 )
is called absolutely convergent if a sériés made up of the absolute

values of its terms converges:
1U 1 I+ IU 2 1+ 1U Z I+ • • . + I U n | -f* . . . (2)
If the alternating sériés (1) converges, while the sériés (2) com-
posed of the absolute values of its terms diverges, then the given
alternating sériés ( 1) is called a conditionally convergent sériés.
Example 3. The alternating sériés
is conditionally convergent, since a sériés composed of the absolute values of

4.8 P lus-and-Minus Sériés. Absolute and Conditional Convergence 273
its lerms is a harmonie sériés,
'+ T + Ÿ + T + - - '
which diverges. The sériés itself converges (this can be readily verified by
Leibniz* test).
i-± + ± -± + ..
21^3! 41^
is absolutely convergent, since a sériés made up of the absolute values of its
terms,
converges, as established in Sec. 4.4.
Theorem 1 is frequently stated (with the help of the concept

of absolute convergence) as follows: every absolutely convergent
sériés is a convergent sériés.
In conclusion, we note (without proof) the following properties
of absolutely convergent and conditionally convergent sériés.
Theorem 2 . If a sériés converges absolutely, it retnains abso
lutely convergent for any rearrangement of its terms. The sum of
the sériés is independent of the order of its terms.
This property does not hold for conditionally convergent sériés.
Theorem 3. I f a sériés converges conditionally, then no matter
what number A is given, the terms of this sériés can be rearranged
in such manner that its sum is exactly equal to A. What is more,
it is possible so to rearrange the terms of a conditionally conver
gent sériés that the sériés resulting after the rearrangement is di
vergent.
The proof of these theorems is outside the scope of this course,
and can be found in more fundamental texts (see, for example,
G. M. Fikhtengolts, “Course of Differential and Intégral Calculus”,
1962, Vol. II, pp. 319-320, in Russian).
To illustrate the fact that the sum of a conditionally convergent
sériés can change upon rearrangement of its terms, consider the
following example.
' ~ T + T ~ T + ‘-- (8)

converges conditionally. Dénoté its sum by s. It is obvious that s > 0. Réarrangé
the terms ofthe sériés(8 ) so that two négative terms follow one positive term:
1 1 1 ! 1 1 1 1
1 2 4 ‘ 3 6 8 ‘ * *2k— 1 4 * -2 4/e+ *' ’ (9)
We shall Drove that the résultantsériés converges, but that its sum s' is
18—2082
274 Ch. 4 Sériés
half the sum of the sériés (8 ): that is, — s. Dénoté by sn and s'n the partial
sums of the sériés (8 ) and (9). Consider the sum of 3k terms of the sériés (9):
^ - ( 1 “ T ~ T j + ( T - T ~ T ) + --- + (2Â~T“ 4F=2- i )
= ( y " i ) + ( ï " ï ) + ''- + ( 4 i T 2 " i )

[ ( 1 “ T ) + ( T ~ T ) + --- + (âT=rr“ i ) ]
4 ( ‘ ~ } + r T + " ' '2k — 1 2k J 2

Consequently,
lim s^k— lim =
k —>■CD k -*■ CD 2 2
Further,
lim s'3k+i= lim ^ S3A
C— 1 >\ 1
|- 2k+\J~~
*->00 k —►CD 2 S
lim sçk+ 2 = lim ( c'^j-----!
k-*> CD V 3k^ 2 k + \ 4/e + 2» )-* ■
And we obtain
lim sn = s’= . L s
n-*>co 2
Thus, in this case the sum of the sériés changed after its terms were
rearranged (it diminished by a factor of 2).
4.9 FUNCTIONAL SERIES

A sériés . ux+ u9+ . . . + un + . . . is called a functional sériés
(or sériés of functions) if its terms are functions of x.
Consider the functional sériés
U1(* ) + U 2 (* ) + W3 ( x ) + • • • + U n ( x ) + • • • (0
Assigning to x definite numerical values, we get different nume-
rical sériés, which may prove to be convergent or divergent.
The set of ail those values of x for which the functional sériés
converges is called the domain of convergence of the sériés.
Obviously, in the domain of convergence of a sériés its sum is
some function of x. Therefore, the sum of a functional sériés is
denoted by s(*).
Example. Consider the functional sériés
1+ X + x2+ . . . + xn + . . .
This sériés converges for ail values of x in the interval (—1, 1), that is,
for ail x that satisfy the condition \x | < 1. For each value of x in the interval
( — 1, 1), the sum of the sériés is equal to -j— - (the sum of a decreasing geo-
4.10 Dominât ed Sériés 275
metric progression with ratio x). Thus, in the interval (—1, 1) the given sériés
defines the function
which is the sum of the sériés; that is,
y-|:^ = l + ^ + Jf2+A:3+ ...

Dénoté by sn (x) the sum of the first n terms of the sériés (1).
If this sériés converges and its sum is equal to s(x), then
s M = s„ (*) + /•„(£)
where rn(x) is the sum of the sériés un+l (*) + «„+a (x) -f- . . . , i.e.,
r n (x) = m „+i (x) + «„+2 ( * ) + • • •
Here, the quantity rn (x) is called the remainder of the sériés (I).
For ail values of x in the domain of convergence of the sériés we
hâve the relation Iims„(Af) = s(x); therefore,
/I —
►OD
lim /•„(*)= lim [s (*)—s„ (x)] = 0

n -> oo n-+<x>
which means that the remainder rn (x) of a convergent sériés ap-

proaches zéro as n —*■oo.
4.10 DOMINATED SERIES
Définition. The functional sériés

« i(x) + ut (x) + u3(x) + . . . + un(x) + . . . (1 )
is called dominated in some range of x if there exists a conver
gent numerical sériés
« i +«» + « * + .• • + « » + • • • (2 )
with positive terms such that for ail values of x from this range
the following relations are fulfilled:
l« iW K « n •••. ••• (3)
In other words, a sériés is called dominated if each of its terms
does not exceed, in absolute value, the corresponding term of some
convergent numerical sériés with positive terms.
For example, the sériés
cos x , cos 2x , cos 3x , , cos nx ,
— r — 22 I 32 r •• •+ ~ ^ i r •• •
is a sériés dominated on the entire x-axis. Indeed, for ail values

18 *
276 Ch. 4 Sériés
of jc, the relation

cos nx I_ 1 , i o \
— \< 7 ? <n = 1 ’ 2* •••)
is fulfilled and the sériés
J—
1 IU J-
22 ‘ 32 T • • »
as we know, converges.
From the définition it follows straightway that a sériés domina-
ted in some range converges absolutely at ail points of this range
(see Sec. 4.8). Also, a dominated sériés has the following important
property.
Theorem. Let the functional sériés
u1(x) + u2(x)+ . . . +u„(x) + . . .
be dominated on the interval [a, b], Let s (a) be the sum of this
sériés and sa(x) the sum of thé first n terms of this sériés. Then
for each arbitrarily small number e > 0 there will be a positive
integer N such that for ail n ^ N the following inequality will be
fulfilled,
\s(x)— sm(x)\ <B
no matter what the x of the interval [a, b].
Proof. Dénoté by a the sum of the sériés (2):
a = a i + a 2+ a 3+ ••• + an+ a n + 1+ •••
then
o = <J„+ e„
where a„ is the sum of the first n terms of the sériés (2 ), and e„
is the sum of the remaining terms of this sériés; that is,
^n +1 “t" + 2 “f“ • • •
Since this sériés converges, it follows that
lim <j„ = a
n -* -c c
and, conséquently,
lim e„ = 0
n —►od
Let us now represent the sum of the functional sériés (1) in the
form
s (x) = sn (x) + rn (x)
where
Sn {x) = U, (x) + . . . + Un (x)
Tn W = « n + l (X) + Un + 1 (X ) + « B + s (X) + . . .
_________ 4.11 The Continuity of the Sum of a Sériés 277
From condition (3) it follows that
and therefore
I rn (a) | < e„
for ail x of the range under considération.
Thus,
| s (a) —s„( a)| < e„
for ail x of the interval [a, b], and e„—>0 as n —* oo.
Note 1 . This resuit may be represented geometrically as follows.
Consider the graph of the function y — s(x). About this curve
construct a band of width 2e„; in other words, construct the curves
t/=s(A) + e„ and i/=s(a) —en
(Fig. 114). Then for any e„ the
graph of the function s„(A)will
lie completely in the band under
considération. The graphs of ail
successive partial sums will
likewise lie within this band.
Note 2 . Not every functional
sériés convergent on the interval
[a, b] has the property indicated
in the foregoing theorem. How-
ever, there are nondominated sé
riés such that possess this pro
perty. A sériés that possesses this property is called a uniformly
convergent sériés on the interval [a, b].
Thus, the functional sériés ut (x) + u2(a) + . . . + u„ (x) + . . . is
called a uniformly convergent sériés on the interval [a, b] if for
any arbitrarily small e > 0 there is an integer N such that for
ail n ^ N the inequality
\ s { x ) — sn ( x ) \ < t
will be fulfilled for any x of the interval [a, b].
From the theorem that has been proved it follows that a domi-
nated sériés is a sériés that uniformly converges.
4.11 THE CONTINUITY OF THE SUM OF A SERIES

Let there be a sériés made up of continuous functions
«i (a) + u, (a) + • • • + u„ (x) + . . .
convergent on some interval [a, b].
In Chapter 2, Vol. 1, we proved a theorem which stated that
the sum of a finite number of continuous functions is a continuous
278 Ch. 4 Sériés
function. This property does not hold for the sum of a sériés con-
sisting of an infinité number of terms. Some functional sériés with
continuons terms hâve for the sum a continuous function, while
in the case of other functional sériés with continuous terms, the
sum is a discontinuous function.
Example. Consider the sériés
The terms of this sériés (each term is bracketed) are continuous functions for
ail values of x. We shall prove that this sériés converges and that its sum
is a discontinuous function.
We find the sum of the first n terms of the sériés:

1
sn = x2n + 1 —X
Find the sum of the sériés:
if x > 0, then
s= lim s„ (* )= lim + 1 —jc) = 1 —^

n -+ oo n -*■ cc
if x < 0, then
5= limsn (x)= lim ( — | x | 2/7+1 —x ) = —1 —j»

n -* - <x> n -+ c c
if * = 0, then sn = 0, and so s = lim s„ = 0. Thus, we hâve

n -►oo
s (x) — — \ — x for x < 0
s (jc) = 0 for x = 0
s {x) = 1 —x for x > 0
4.11 The Continuity of the Sum of a Sériés 279
And so the sum of the given sériés is a discontinuous function. Its graph is
shown in Fig. 115 along with the graphs of the partial sums st (x),
and s3 (x).
The following theorem holds true for dôminated sériés.

Theorem. The sum of a sériés (of continuous functions) dôminated
on some interval [a, b] is a function continuous on that interval.
Proof. Let there be a sériés of continuous functions that is do-
minated on the interval [a, b]:
ul (x) + u2(x) + u3(x)+ . . . ( 1)
Let us represent its sum in the form
s (x) = s„ (x) + r„ (x)
where
sn (x) = «,(*) + . . . + un (x)
and
rn (*) = «n + x (x) + « n + 2 (x) + . . .
On the interval [a, b] take an arbitrary value of the argument
x and give it an incrément Ax such that the point x-fA x also
lies on the interval [a, b].
We introduce the notation
As = s (x + Ax)—s (x)
As„ = s„ (x -f Ax) —sn (x)
then
As = As„ + rn (x + Ax) —rn (x)
from which we hâve
I As | < | As„ | +1 rn (x + Ax) | + 1r„ (x) | (2)
This inequality is true for any integer n.
To prove the continuity of s(x), we hâve to show that for any
preassigned and arbitrarily small e > 0 there will be a number
ô > 0 such that for ail | Ax | < 6 we will hâve | As | < c.
Since the given sériés (1) is dôminated, it follows that for any
preassigned e > 0 there will be found an integer N such that for
ail n ^ N (and as a particular case, n = N) the inequality
IM*)l<y (3)
will be fulfilled for any x of the interval [a, 6 ]. The value x + Ax
lies on the interval [a, b] and therefore the following inequality
is fulfilled:
|rjv(x + A x )|< — (3')
280 Ch. 4 Sériés
Further, for the chosen N the partial sum s,v (a) is a continuous
function (the sum of a finite number of continuous functions) and,
consequently, a positive number ô may be chosen such that for
every Ax that satisfies the condition J Ajc | < ô the following
inequality is fulfilled:
|ASa,| < ! (4)
By inequalities (2), (3), (3'), and (4), we hâve
l*l<T +T +ï = 8
that is,
| As | < e for | Ax | < ô
which means that s(x) is a continuous function at the point *
(and, consequently, at any point of the interval [a, b]).
Note. From this theorem it follows that if the sum of a sériés
is discontinuous on some interval [n, b], then the sériés is not
dominated on this interval. In particular, the sériés given in the
Example is not dominated on any interval containing the point
x — 0 , that is to say, a point of discontinuity of the sum of the
sériés.
We note, finally, that the converse statement is not true: there
are sériés, not dominated on an interval, which, however, converge
on that interval to a continuous function. For instance, every
sériés uniformly convergent on the interval [a, b] (even if it is not
dominated) has a continuous function for its sum (if, of course,
ail terms of the sériés are continuous).
4.12 INTEGRATION AND DIFFERENTIATION OF SERIES

Theorem 1. Let there be a sériés of continuous functions
«1 (X) + U2(x) + . . . + un (x) + . . . (1)
dominated on the interval [a, b] and let s (x) be the sum of the
sériés. Then the intégral of s(x) from a to v, which limits belong
to the interval [<a, b], is equal to the sum of the intégrais of the
terms of the given sériés; that is,
X X X X
$ s (t) dt = J «, (/) dt + J (t) dt + . . . + 5 u„ (t) dt + . . .

a a a a
Proof. The function s ( a:) may be represented in the form
s (x) ^ s„ (x) + r„ (x)
or
s(x) = u, (a ) + (x) + . .. + un (x) + rn (x)
4.12 Intégration and Différentiation of Sériés 281
Then
X X X X X
S S ( 0 dt = S u, (t) dt + $ ut ( 0 dt + . . . +. 5 u„ (t) dt + \ r n (t) dt (2 )

a a a a a
(the intégral of the sum of a finite number of terms is equal to
the sum of the intégrais of these terms).
Since the original sériés (1) is dominated, it follows that for
every x we hâve | r„ (x ) | < e„, where e„ —>- 0 as n —>■oo. Therefore, *
X X X
\ r n{t)dt < ± $ | rn (t) | dt < ± J e„ dt = ± e„ (*—a ) < e„(b—a)

a a a
Since e„—>-0, it follows that
X
lim \ r n(t)d t= 0
n ■* « a
But from équation (2) we hâve
x x r x x n
5r„(/) d/ - aJs(od t —L aS«x(o ^ + •••+S

a a
“n (o d t
J
Hence
lim j ^s ^ hx(t)dt + . . . + J tin(t) dt jj =0

_a
or
lim $ U t(t)dt+ . . . + $ un(t)dt = $ s (/) dt (3)
n -► ac _a a J a
The sum in the brackets is a partial sum of the sériés
X X
)j u1(t)dt+ . . . + J « „ (/)d /+ . . .
< (4)
Since the partial sums of this sériés hâve a limit, this sériés
converges and its sum, by virtue of équation (3), is equal to
X
^ s (t)d t, i.e.,
a
X X X X
S s (i)dt — ^ U1(t) dt + ^ W2 (t)dt -f- . . . + 5 Un (t)dt + • • •

a a a a
This is the équation that had to be proved.
* In the estimâtes given below we take the + sign for a < x and the — sign
for x < a.
282 Ch. 4 Sériés
N ote 1 . If a sériés is not dominated, term-by-term intégration

of it is not always possible. This is to be understood in the sense
X
that the intégral [ s(t)d t of the sum of the sériés ( 1) 1s not always
a
equal to the sum of the intégrais of its terms [that is, to the sum
of the sériés (4)].
T h eo rem 2. If a sériés
«1 (*) + « , W + . . . + H „ W + . . . (5)
made up of functions having continuous dérivatives on the interval
[a, b] converges (on this interval) to the sum s (x) and the sériés
u'1(x) + iït { x ) + ...+ u 'n ( x ) + ... (6 )
made up of the dérivatives of its terms is dominated on the same
interval, then the sum of the sériés of dérivatives is equal to the
dérivative of the sum of the original sériés; that is,
s' (x) = (x) -[- ua (x) -f- u8 (jc) • + u'n (x) + . . .
P ro o f. Dénoté by F (x) the sum of the sériés (6 ):
F (x) = u[ (x) + a» (*) (at) + . . .
and prove that
F(x) = s’ (x)
Since the sériés (6 ) is dominated, it follows, by the preceding
theorem, that
X X X X
5 F (t) dt — J u[ (t) dt + J «â (t) dt + . . . ^ u'n (t) dt + . . .

a a a a
Performing the intégration, we get

X
[ F (t)d t = [u^x)— (a)]

a
+ [ « t (* ) — “ « ( « ) ] + • • • + K (*) — “ n ( a ) ] + . . .
But, by hypothesis,
s (*) = «x(x) + u2( * ) + • • • + «n (* )+ •• •
s (a) —«! (a) + U, (a) + • • • + « „ (« )+ •••
no matter what the numbers x and a on the interval fa, b],
Therefore,
X
J F (t)dt — s (*)—s (a)

4.13 Power Sériés. Interval of Convergence 283
Differentiating both sides of this équation with respect to x,

we obtain
F (x) = s' (x)
We hâve thus proved that when the conditions of the theorem
are fulfilled, the dérivative of the sum of the sériés is equal to
the sum of the dérivatives of the terms of the sériés.
Note 2 . The requirement of dominance (majorization) of a sériés
of dérivatives is extremely essential, and if not fulfilled it can
make term-by-term différentiation of the sériés impossible. This is
illustrated by a dominated sériés that does not admit term-by-term
différentiation.
Consider the sériés
sin l*x . sin 2*x . sin 34* . . sin n*x .
[à I 2* • 3* r • • • H ^2 r • • •
This sériés converges to a continuous function because it is

dominated. Indeed, for every x its terms are (in absolute value)
less than the terms of the numerical convergent sériés with
positive terms
—1 ^+ —
2a ^+ —
3a ^4- ' ‘ ' +^ —
n2+^ ‘
Write a sériés composed of the dérivatives of the terms of the
original sériés:
cos x + 2 a cos 2*x + . . . + n2cos n*x + . . .
This sériés diverges. Thus, for instance, for x = 0 it turns into
the sériés
1 + 2a 4- 32 4- • • • 4- 4- • • •
(It may be shown that it diverges not only for x = 0.)
4.13 POWER SERIES. INTERVAL OP CONVERGENCE

Définition 1. A power sériés is a functional sériés of the form
a0+ alx + a2x2+ ■• ■+ a nxn+ . . . ( 1)
where a0, alt a2, . . . , an, . . . are constants called coefficients of
the sériés.
The domain of convergence of a power sériés is always some
interval, which, in a particular case, can degenerate into a point.
To convince ourselves of this, let us first prove the following theorem,
which is very important for the whole theory of power sériés.
Theorem 1 (Abel’s theorem). (1) If a power sériés converges for
some nonzero value xQ, then it converges absolutely for any value
of x, for which
Ix I I*0 I
284 Ch. 4 Sériés
(2 ) if a sériés diverges for some value x'0t then it diverges for every
x for which
1*1 > K l
Proof. (1) Since, by assumption, the numerical sériés
an+ atx 0 + a2xl + . . . + anx%+ . . . (2 )
converges, it follows that its common term a„x{J—>-0 as n —►oo,
and this means that there exists a positive number M such that
ail the ternis of the sériés are less than M in absolute value.
Rewrite the sériés ( 1) in the form
a0+ ûi*o (^") + ) + •• • + a K ( —) + ••• (3)
and consider a sériés of the absolute values of its tcrms:
K I + K *ol + 1azxl I + •• • + | I + ••• (4 )
The terms of this sériés are less than the corresponding terms
of the sériés
X X
M+M M
*7
‘+ . . . + M
*0
+ ... (5 )
For \ x \ < \ x Q\ the latter sériés is a géométrie progression with
X
ratio < 1 and, consequently, converges. Since the terms of the
*o
sériés (4) are less than the corresponding terms of the sériés (5),
the sériés (4) also converges, and this means that the sériés (3)
or ( 1) converges absolutely.
(2) It is now easy to prove the second part of the theorem: let
the sériés ( 1) diverge at some point x'Q. Then it will diverge at
any point x that satisfies the condition |* |> |* i |. Indeed, if at
some point x that satisfies this condition the sériés converged, then,
by virtue of the first part (just proved) of the theorem, it should
converge at the point x'0 as well, since \x'0\< \x \. But this con-
tradicts the condition that at the point x'0 the sériés diverges.
Hence the sériés diverges at the point x as well. The theorem is
thus completely proved.
Abel’s theorem makes it possible to judge the position of the
points of convergence and divergence of a power sériés. Indeed,
if x0 is a point of convergence, then the entire interval (— 1*0|,
| jc0 |) is filled with points of absolute convergence. If x'0 is a point
of divergence, then the whole infinité half-line to the right of the
point |*o| and the whole half-line to the left of the point — \x'0\
consist of points of divergence.
From this it may be concluded that there exists a number R
such that for |a*| < R we hâve points of absolute convergence and
for | jc | > /?, points of divergence.
We thus hâve the following theorem on the structure of the

domain of convergence of a power sériés:
Theorem 2 . The domain of convergence of a power sériés is an
inierval with centre at the coordinate origin.
Définition 2. The interval of convergence of a power sériés is an
interval from —R to -\-R such that for any point x lying inside
this interval, the sériés converges and converges absolutely, while
Sériés converges
for points x lying outside it, the sériés diverges (Fig. 116). The
number R is called the radius of convergence of the power sériés.
At the end points of the interval (at x = R a n d a tx = — R) the
question of the convergence or divergence of a given sériés is
decided separately for each spécifie sériés.
We note that in some sériés the interval of convergence dégé
nérâtes into a point (Æ= 0 ), while in others it encompasses the
entire x-axis (R = oo).
We give a method for determining the radius of convergence of
a power sériés.
Let there be a sériés
a0+ a1x-\-a2xi + ...+ a „ x n+ . . . ( 1)
Consider a sériés made up of the absolute values of its terms:
I a01+ 1a, 1 1x | + 1a 2 11 a: |2 + 1a 3 11 x |s
+ 1a411x |4+ • • • + 1 11x |" + . . . (6)
To détermine the convergence of this sériés (with positive terms!),
apply the d’Alembert test.
Let us assume that there exists a limit:
a n + ix n + 1 I « an+1
lim 2a±±= Iitn M I ~ AX11x \ = L \x \
I—
* rr> Ptn a»xn \ n ^ , û/i
Then, by the d ’Alembert test, the sériés (6 ) converges, if L \x \ < 1 ;
that is, if |* |< -£ - . and diverges if L \x \ > 1 , that is, if |x |> - ^ - .
Consequently, sériés ( 1) converges absolutely when | * |< X - But
if | x | > - , then lim —|x | L > 1 and sériés (6 ) diverges, and
L n -+*> un
286 Ch. 4 Sériés
its general term does not tend to zéro. * But then neither does
the general term of the given power sériés ( 1) tend to zéro, and
this means that (on the basis of the necessary condition of con
vergence) this power sériés diverges ^when
From the foregoing it follows that the interval ( T ’ r ) is
the interval of convergence of the power sériés ( 1), i.e.,
R = -r-= lim
L n oo an+ l
)
Similarly, to détermine the interval of convergence we can make
use of the Cauchy test, and then
D _ _____*
lim { /K J
Tl —► CO
Bxample 1. Détermine the interval of convergence of the sériés

1 +JC + *2 + Jt3+ . . . + x n + . . .
Solution. Applying d’Alembert’s test directly, we get
xn+ l
lim = X\
Xn
Thus, the sériés converges when \x \ < 1 and diverges when \x \ > 1. At
the extremities of the interval (— 1, 1) it is impossible to investigate the
sériés by means of d’Alembert’s test. However, it is immediately apparent
that when x = — 1 and when x = \ the sériés diverges.
Example 2. Détermine the interval of convergence of the sériés
2x (2x)2 . ( 2 j c ) 3
T 29 nI 3
Solution. We apply the d’Alembert test:
(2*)”* 1
n - \-1
lim
(2x)n = n lim« n + \ I2x I = I 2x I
The sériés converges if \2x\ < 1, that is, if \x \ < when ^ = y the sériés
converges; when x = — the sériés diverges.

Example 3. Détermine the interval of convergence of the sériés
V*2 y3 yTl
JC+2 T + 3 i+ -"+ ^ T + -"

* It will be recalled that in proving d’Alembert’s test (see Sec. 4.4) we
found that if lim — +1 > 1, then the general term of the sériés increases and,
un
consequently, does not tend to zéro.
Solution. Applying the d’Alembert test, we get

W« + 1 — 111mil x n+1n\ --- 1Jm JC
-- 11 — mu
«H n -►x x n (« + !)! n -+ oo n + i
Since the limit is independent of x and is less than unity, the sériés con
verges for ail values of x.
Example 4. The sériés 1 + x -|- (2jc)2+ (3*)3 (nx)n + . . . diverges for
ail values of x except = 0 because (nx)n —►oo as n —►oo no matter what
jc
the je, as long as it is different from zéro.

Theorem 3. The power sériés
a„ + a lx + a ix 2+ . . . + a nxn + ■■• ( 1)
is dominated on any interval [— p, p] that lies completely inside
the interval of convergence.
Inter val of convergence
—Q-
-a
interval of majorUation
Fig. 117
Proof. It is given that p < R (Fig. 117) and therefore the num-
ber sériés (with positive terms)
Ia Q I + Ia \ IP + Iü 2 IP2 + ** *+ Iün IP” (7 )
converges. But when | x | < p , the terms of the sériés ( 1) do not
exceed, in absolute value, the corresponding terms of sériés (7).
Hence, sériés ( 1) is dominated on the interval [— p, p].
-/? oc 0 fi R
Fig. 118
Corollary 1. On every interval lying entirely within the interval

of convergence, the sum of a power sériés is a continuons function.
Indeed, the sériés on this interval is dominated, and its terms are
continuous functions of x. Consequently, on the basis of Theorem 1 ,
Sec. 4.11, the sum of this sériés is a continuous function.
Corollary 2 . If the limit s of intégration a, p lie within the interval of
convergence of a power sériés, then the intégral of the sum of the sériés
is equal to the sum of the intégrais of the terms of the sériés, be
cause the domain of intégration may be taken in the interval
[— P» P]» where the sériés is dominated (Fig. 118) (see Theorem 1,
Sec. 4.12, on the possibility of term-by-term intégration of a domi
nated sériés).
288 Ch. 4 Sériés
4.14 DIFFERENTIATION OF POWER SERIES

Theorem 1. If a power sériés
s (x) = aQ+ aix + a2x 2 + a3x*+ a4^ + • • • + anxTt + • • • 0 )
has an interval of convergence (— R, R)t ihen the sériés
<p(x) = a1Jr 2atx + 3ûgjc2 + • .. + nanxn~x+ • • • (2 )
obtained by termwise différentiation of the sériés ( 1) has the same
interval of convergence (— /?, /?); here,
q) (x) = s' (x) if | * | < /?
i . e i n s i d e the interval of convergence the dérivative of the sum
of the power sériés ( 1) is equal to the sum of the sériés obtained by
termwise différentiation of the sériés ( 1).
Proof. We shall prove that the sériés (2) is dominated on any
interval [— p, p] that lies completely within the interval of con
vergence.
■■*■— - ■ ! M il
-R -p 0 x p £ R xz s,
Fig. 119
Take a point £ such that p < £ < Æ (Fig. 119). The sériés (1)
converges at this point, hence lim = it is therefore possible
n -► o c
to indicate a constant number M such that

| an\ n | < M (n= 1 , 2,...)
If | x | < p , then
p |» -i . M . ,
Ina«jc" - 1 | < | nanp""1 1== n | a^ " - 1
fl <"T^
where
Thus the terms of the sériés (2), when ,v | ^ p, are less, in abso-
lute value, than the terms of a 'positive number sériés with con
stant terms:
—•(1 + 2q + 3q*+ . . . + n q " -'+ . . . )
But the latter sériés converges, as will be évident if we apply
the d’Alembert test:
nqn~ 1
lim • 7 <I
n -+• » (n — \)q n~ 2
4.15 Sériés in Powers of x — a 289
Hence, the sériés (2) is dominated on the interval [— p, p], and by

Theorem 2, Sec. 4.12, its sum is a dérivative o! the sum of the
given sériés on the interval [— p, p], i.e.,
<p(x) = s' (x)
Since every interior point of the interval (— R, R) may be
included in some interval [— p, p], it follows that the sériés (2 )
converges at every interior point of the interval (— R, R).
We shall prove that outside the interval (— R, R) the sériés (2)
diverges. Assume that the sériés (2 ) converges when xt > R .
Integrating it termwise in the interval (0, *,), where R < xt <
we would find that the sériés ( 1) converges at the point x2, but
this contradicts the hypothèses of the theorem. Thus, the interval
(— R, R) is the interval of convergence of sériés (2 )..And the
theorem is proved completely.
Sériés (2) may again be differentiated term by term, and this
may be continued as many times as one pleases. We thus hâve
the conclusion:
Theorem 2 . If a power sériés converges in an interval (— R, R),
its sum is a function which has, inside the interval of convergence,
dérivatives of any order, each of which is the sum of a sériés re
suit ing from term-by-term différentiation of the given sériés an
appropriate number of times; here, the interval of convergence of
each sériés obtained by différentiation is the same interval (— R, R).
4.15 SERIES IN POWERS OF x —a

Also called a power sériés is a functional sériés of the form
û0 + ûi (x— a) + a1 (x— a)*+ . . . +a„ (x— a)n+ ... ( 1)
where the constants a0, alt . . . , a„, . . . are likewise termed coeffi
cients of the sériés. This is a power sériés arranged in powers of
the binomial x —a.
When a = 0, we hâve a power sériés in powers of x, which,
consequently, is a spécial case of sériés ( 11.
To détermine the région of convergence of sériés ( 1), substitute
x —a = X
Sériés (1) then takes on the form
ao aiX + a*A2 + • • • + o.nX n -f-. . . (2)
we thus hâve a power sériés in powers of X.
Let the interval — R < X < R be the interval of convergence
of the sériés (2) (Fig. 120, a). It thus follows that sériés (1) will
converge for values of x that satisfy the inequality — R < x — a < R
or a—/? < * < a + i?. Since sériés (2) diverges for | X | > / ? , the
19— 2082
290 Ch. 4 Sériés
sériés ( 1) will diverge for |jc—a | > / ? , that is, it will diverge
outside the interval a — R < x < a + R (Fig. 120,/?).
And so the interval (a—R, a + R) with centre at the point a
will be the interval of convergence of sériés (1). Ail the properties
of a sériés in powers of x inside the interval of convergence
(— R, + R) are retained completely for a sériés in powers of
x —a inside the interval of convergence (a —R , a + /?). For example,
-R 0 R X
(a) + —I----- 1--+
a - R -- --------. a *R x 0 12 3 x
(b) * ---- 1----a1---- 1---- H--- 1--- 1---- 1------
Fig. 120 Fig. 121
after term-by-term intégration of the power sériés ( 1), if the limits

of intégration lie within the interval of convergence (a—R, a + R),
we get a sériés whose sum is equal to the corresponding intégral
of the sum of the given sériés ( 1). In the case of termwise diffé
rentiation of the power sériés ( 1), for ail x lying inside the inter
val of convergence (a —R, a + R) we obtain a sériés whose sum
is equal to the dérivative of the sum of the given sériés ( 1).
Example. Find the région of convergence of the sériés
(x —2) + (* —2)2 + (* —2)3 + . . . + ( * —2)n + . . .
Solution. Putting x — 2 = X, we get the sériés
X + X2 + * 3+ .
This sériés converges when —1 < X < + !• Hence, the given sériés converges
for ail x that satisfy the inequality — 1 < x — 2 < I, that is, when 1 < x < 3
(Fig. 121).
4.16 TAYLOR’S SERIES AND MACLAURIN’S SERIES
In Sec. 4 .6 , Vol. I, it was shown that for a function f(x) that

has ail dérivatives up to the (n + l)th order inclusive, Taylor’s
formula holds in the neighbourhqod of the point x = a (that is,
in some interval containing the point x = a):
f(x)= J(a) + ^ r ( a )
+ r(a)+... + —
{ ^ r( a ) + Rn(x) (1)
where the so-called remainder term R„ (x) is computed from the

formula
R» = i ^ r n r f(n+" [ <*+0 (*—*)]. o< e< î
4.16 Taylor*s Sériés and Maclaurin's Sériés 291
If the function f(x ) has dérivatives of ail orders in the neigh-

bourhood of the point x —a, then in Taylor’s formula the number
n may be taken as large as we please. Suppose that in the
neighbourhood under considération the remainder term Rn tends
to zéro as n —<-oo:
lim Rn (x) = 0
n -* oc
Then, passing to the limit in formula (1) as n —»-oo, we get an
infinité sériés on the right which is called the Taylor sériés:
/ (x) = / (a) + ^ /' (a) + . . . + /<»>(a) + . . . (2 )
This équation is valid only when R„(x) —>0 as n —»-oo. Then the
sériés on the right converges and its sum is equal to the given
function f(x). Let us prove that this is indeed the case:
f(x) = Pn(x) + Rn (x)
where
P„ (x) = /(<*)+ / » + ■ • • + (- ^ /<«> (a)
Since it is given that lim Rn(x) = 0, we hâve
n -* oo
f(x )= lim Pn(x)
n -*■ cd
But P„(x) is the nth partial sum of the sériés (2); its limit is
equal to the sum of the sériés on the right side of (2). Hence,
(2 ) is true:
/ « - / « o + ^ r («)+ f e ^ 2 f w + . . . + — r : r ” ( a ) + - - -
From the foregoing it follows that the Taylor sériés is a given
function f(x) only when lim Rn(x) = 0. If lim Rn (x) 0, then the sériés
is not the given function, although it mav converge (to a different
function).
If in the Taylor sériés we put a = 0 , we get a spécial case of
this sériés known as Maclaurin's sériés:
/ w = / ( 0) + f r ( 0) + - r ( 0) + . . . + f | r ( 0) + . .. (3)
If for some function we hâve a formally written Taylor’s
sériés, then in order to prove that this sériés is a given function
it is either necessary to prove that the remainder term approaches
zéro, or to be convinced in some way that this sériés converges
to the given function.
We note that for each of the elementary functions defined in
Sec. 1.8, Vol. I, there exists an a and an R such that in the inter-
val (a—R, a + R) it may be expanded into a Taylor’s sériés or
(if a = 0) into a Maclaurin’s sériés.
19
292 Ch. 4 Seriez
4.17 SERIES EXPANSION OF FUNCTIONS
1. E x p a n d in g th e fu n c tio n / ( j r ) = s i n j r in a M a c la u rin ’s sériés.

In Sec. 4.7, Vol. I, \ve obtained the formula
sinx=--x — 57 + 5 T+ • • • + ( — 1)"+1 (2/i-D ! ■/?*.(*)
Since it has been proved that lim Rîn(x) = 0, it follows, by what was
n -#» od
said in the preceding section, that we get an expansion of sinx

in a Maclaurin’s sériés:
X-ln-l
sinx = x— + (—Dn+1 (2n - 1)! ( 1)
Since the remainder term approaches zéro for any x, the given
sériés converges and, for its sum, has the function sinx for
any x.
Fig. 122 shows the graphs of the function sinx and of the
first three partial sums of the sériés ( 1).
This sériés is used to compute the values of sinx for different
values of x.
To illustrate, let us compute sin 10° to the fifth décimal place.
4.17 Sériés Expansion of Functions 293
Since 10o = jg (radians) æ 0.174533, we hâve

• iAo Jt 1/ji\3 i 1 f \ f n V t
sm 10 je 37 ~^5l \Ts J 7l CÏ8 J + • • •
Confining ourselves to the first two terms, we get the following
approximate équation:
ji n 1 / n \
Sln Ï S « i S — fi
ÏS
vis;
here, we are in error by 6 , which in absolute value is less than
the first of the suppressed terms; that is,
< l 2ô (0 -2 )6 < 4 ' 10

If each term in the expression for sin-j^ is computed to six
décimal places, we get
sin -g = 0.173647
We can be sure of the first four décimais.
2. E x p a n d in g th e fu n c tio n f ( x ) = ex in a M a c la u rin ’s sériés.
On the basis of Sec. 4.7, Vol. I, we hâve
y2 y3 y fl
e * = l + * + 2ï + 3, + -- - + ^ i + (2)
since it was proved that Iim Rn (x) = 0 for any x. Hence, the
n -* oo
sériés converges for ail values of x and is the function ex.
Substituting x = — x, in (2 ), we get
y2 y3 y4
o-x _ 1 x 4 - - ____ — 4 - — -
e 1 '* ^ 2 ! 31 ‘ 41 (3)
3. E x p a n d in g th e fu n c tio n / ( jr ) = c o s x in a M a c la u rin ’s sériés.

From Sec. 4.7, Vol. I, we hâve
« X “ . xv .
C O S X - 1 — 21 + 4J — 6 Ï + • • • (4)
for ail values of x the sériés converges and represents the function
COSX.
4. E x p a n sio n , in M a c la u rin ’s sériés, of th e fu n c tio n s
e* — e--*
f (x) = sinh x ^ 2
e*+ e-
/ (x) = cosh x =
These functions are readily expanded by subtracting and adding
the sériés (2) and (3) and dividing by 2.
294 Ch. 4 Sériés
Thus,
JC3 , JC5 , X1
sinh* = x + 3j + 5ï + 7 î + . . . (5)
coshx= l + £ + £ + ^ + . . . (6)
4.18 EULER’S FORMULA
Up till now we hâve considered only sériés with real terms and
hâve not dealt with sériés with complex terms. We shall not give
the complété theory oi sériés with complex terms, for this goes
beyond the scope of this text. We shall consider only one important
example in this field.
In Chapter 7, Vol. I we defined the function e*+i* by the équa
tion
e x+ iy _ gX (C0S y _|_ (• s iu y}
When jc = 0 , we get Euler's formula:

e‘y = cosy + i sin y
If we détermine the exponential function e with imaginary
exponent by means of formula (2), Sec. 4.17, which represents the
function e* in the form of a power sériés, we will get the very
same Euler équation. Indeed, détermine e'y by putting the expres
sion iy in place of x in équation (2), Sec. 4.17:
(1)
Taking into account that i* = —1, i 3 = — t, i'4 = l , i* = t, i' = — 1 ,

and so forth, we transform ( 1) to the form
Separating in this sériés the reals from the imaginaries, we find
The parenthèses contain power sériés whose sums are equal to

cos y and siny, respectively [see formulas (4) and (1) of the pre-
ceding section]. Consequently,
e'y —cos y + i sin y (2)
Thus, we hâve again arrived at Euler's formula.
4.19 The Binomial Sériés 295
4.19 THE BINOMIAL SERIES

1. Let us expand the following function in a Maclaurin’s sériés:
/(*) = ( l +* ) «
where m is an arbitrary constant number.
Here the évaluation of the remainder term présents certain dif-
ficulties and so we shall approach the sériés expansion of this
function somewhat differently.
Noting that the function / (x) = ( 1+ x)“ satisfies the differential
équation
( 1 + * )/'(* ) = «/(*) ( 1)
and the condition
/ (0 ) = 1
we find a power sériés whose sum s(x) satisfies équation ( 1) and
the condition s( 0 ) = 1:
s(x )= l + a 1x + û2xî + . . . + anxn + . . . * (2 )
Putting this sériés into équation ( 1), we get
(1 + *) (a, + 2atx + 3a3x 2 + . . . -\-nanx " - 1+ . . . )
= m ( 1 + axx -f a 2xJ + anxn-f . . . )
Equating the coefficients of identical powers of x in different parts
of the équation, we find
a, = m, at + 2 a 2 = ma1........ na„ + (n + l)a n+l = matt, . . .
Whence for the coefficients of the sériés we get the expressions
. ax (m — 1) m (m — l)
a0 — 1 » ûj z tn j cl2 n 9•
a2 (m — 2) m (m — l)(m —2)
3 — ÜT3
m (m— 1).. .[m—/i + 1)
r r —n ’ *
These are binomial coefficients.
Putting them into formula (2), we obtain
s (x) = 1+ mx + « V * + •••
, m(m— —(n — 1)] „n ,
I 1 .9 m A I
* We took the absolute term equal to unity by virtue of the initial con
dition s( 0 ) = l .
296 Ch. 4 Sériés
If m is a positive integer, then beginning with the term con-

taining x m+1 ail coefficients are equal to zéro, and the sériés is
converted into a polynomial. For m fractional or a négative inte
ger, we hâve an infinité sériés.
Let us détermine the radius of convergence of sériés (3):
m (m — 1) . . .[m — n+ 1] m (m — l)...[m —n + 2] %
n\ X",
[n — 1)1 x
m(m — \) ...( m — n~\-\)(n— \)\
lim
/!-»•» “"+,
“n i| = nlira
-» » m (m — 1) . . . (m —n + 2) ni
= lira \x\ = \x\

n -►oo
Thus, sériés (3) converges for |x | < 1.
In the interval (—1, 1), sériés (3) is a function s(x) that satis-
fies the differential équation ( 1) and the condition
s( 0 ) = I
Since the differential équation (1) and the condition s(0) = 1
are satisfied by a unique function, it follows that the sum of the
sériés (3) is identically equal to the function (1 + * )“ , and we
obtain the expansion
(1 + , ) - = 1 + .,, (3")
For the particular case m = —1, we hâve

l
l + * ' 1 —x-\-xi —x 3 -f-... (4)
For m — ]r we get
1 .. 1 », 1-3 , 1-3-5
V l + x —1+-K
2
-X—577+
2-4
;r;, „-V 2-4-6-8 .V4 + . . .
'2 - 4 - 6 (5)
For m = — g- we hâve
— ~ 1 ------1 ..v ,_ 1-3 , ___
| _ _ v-2 1-3-5 , , 1-3-5-7
x 2 x ^2 -4 * 2 -4-6 1 2-4-6-Sî J f — . . . (6)
2. Let us apply the binomial expansion to the expansion of

other iunctions. Expand the function
f (x) = arcsin x
in a Maclaurin’s sériés. Putting into équation (6 ) the expression —x*
in place of x, we get
1 1- 3-5 1-3-5. ..(2 « — 1)
V i- * 2 1+ "J ** + ^75 x* 2- 4-6 2 -4 -6 .,.2/i
X2n + . . .
4.20 Expansion of ln ( l + x) in a Power Sériés. Computing Logarithms 297
By the theorem on intégrât ing power sériés we hâve, for

MO:
dt , 1 x3 , 1-3 1- 3.5 JC7
I V'l-t2
= arcsin^ = x + T T + s r ï T 2- 4-b 7 + * * *
1 -3*5.. .( 2n — 1) jc2" * 1
+ 2 •4 •6 . . . 2/i 2 /i+ 1 *" *
This sériés converges in the interval (—1, 1). One could prove
that the sériés converges for x ^ ± \ as well as that for these
values the sum of the sériés is likewise equal to arcsinx. Then,
setting * = 1 , we get a formula for computing j i :
• i n , , 1 1 , 1 3 1 13 -5 1
arcsinl = 2 - = l + T . -3+274. 5 2*4*6 * 7 +
4.20 EXPANSION OF THE FUNCTION l n ( l + x )

IN A POWER SERIES. COMPUTING LOGARITHMS
Integrating équation (4), Sec. 4.19, from 0 to x (when | * | < 1),

we obtain
i i
o o
or
y2 y3 v4 y /J
l n ( l + * ) - * - + + £ — - + . . . + (- l.)" +1- + . . . ( 1)
This équation holds true in the interval (—1, 1).
If in this formula x is replaced by —x, then we get the sériés
y2 y3 y4
ln (1 —x) = —x — —— ^ — ... (2 )
which converges in the interval (— 1, 1).
Using the sériés ( 1) and (2 ) we can compute the logarithms of
numbers lying between zéro and two. We note, without proof,
that for x = l the expansion ( 1) also holds true.
We will now dérivé a formula for computing the natural loga
rithms of ail integers.
Since in the term-by-term subtraction of two convergent sériés
we get a convergent sériés (see Sec. 4.1, Theorem 3), then by sub-
tracting équation (2 ) from équation ( 1) term by term, we find
In (l- M)—|n ( l _ , ) = l „ } ± i = 2 [* + £ + £ + . . . ]
Now put 7+ = -^ + -; then * = 3+ 7 - F°r an.v n > 0 we hâve

298 Ch. 4 Sériés
0 < x< 1; therefore
3 (2/t+ !)!*“ 5 (2/1+1)»

whence
ln(«+l) lnn -2 [ 2rt + 1 + 33 ((22/t+
n + ll)»
)»+~ 55( 2 n + l) »
+ ...] (3)
For n = l we then obtain
ln 2 —2 3 + 3.33 + 5 .3 s + • • •j
To compute ln2 to a given degree of accuracy 6 , one has to
compute the partial sum s„, choosing the number p of its terms
such that the sum of the suppressed terms (that is, the error Rp
committed when replacing s by sp) is less than the admissible
error 6 . To do this, let us evaluate the error Rp:
Rp = 2 [(2p + 1) 32J'+l + (2p + 3) 3*P+3■*" (2p + 5)3V+» + • ‘ ]

Since the numbers 2p + 3, 2/> + 5, . . . are greater than 2 p + l , it
follows that by replacing them by 2 p + 1 we increase each fraction.
Therefore,
R p < 2 [(2 p + l)3 2P+l + ( 2 p + 1) 32/'+» + (2p+ 1) 32p+» + • • • ]
or
The sériés in the brackets is a géométrie progression with ratio

g-. Computing the sum of this progression we find
(4)
9
If we now want to compute ln2 to, for example, nine décimal
places, we must choose p such that ^<0.000000001. This can be
done by selecting p so that the right side of inequality (4) is less
than 0.000000001. By direct choice we find that it is sufficient
to take p —8 . To nine-place accuracy we hâve
ln 2 » sg= 2
1-3 ‘ 3-33 ~ 5-3» “ 7-37 ~ 9-3» ~ 11-3U
^ T Î ^ + T S ^ ] =°-69314718°
Thus, ln 2 = 0.693147180, correct to nine places.
4.21 Sériés Evaluation of Ùefinite Intégrais 299
Assuming n — 2 in formula (3), we obtain

In 3 = ln 2 -f 2 [ ^ + 3753"+ 575» + . .. j
= 1.098612288, and so forth.
In this way we obtain the natural logarithms of any integer.
To get the common logarithms of numbers, use the following
relation (see Sec. 2.8, Vol. I):
logV = MlnW
where Af = 0.434294. Then, for example, we get
log 2 = 0.434294 x 0.693147 = 0.30103
4.21 SERIES EVALUATION OF DEFINITE INTEGRALS

In Chapters 10 and 11 (Vol. I) it was noted that there exist
definite intégrais, which, as functions of the upper limit, are not
expressible in terms of elementary functions in closed form. It is
sometimes convenient to compute such intégrais by means of sériés.
Let us consider several examples.
1. Let it be required to compute the intégral
J e~x*dx
0
Here, the antiderivative of e~x% is not an elementary function. To
evaluate this intégral we expand the integrand in a sériés, replac-
ing x by —xi in the expansion of e* [see formula (2), Sec. 4.17]:
• y2 y4 y# y2n
Integrating both sides of this équation from 0 to a, we obtain

x7
1
y ax~ \ i 1. 3 ^ 21-5 31-7 1
0
a a3 . a 6
û7 1
“ 1 l!-3 1 2! 5 31-7 1 *
Using this équation, we can calculate the given intégral to any
degree of accuracy for any a.
2. It is required to evaluate the intégral
°^ d x
0
Expand the integrand in a sériés: from the équation
300 Ch. 4 Sériés
we get
sin x _ . __ x*_ . x*___xs
~ “ 1 — 3! + 5! 7! ' ’ ’ *
the latter sériés converges for ail values of x. Integrating term by
term, we obtain
sin jc ___ a3 , a6 a7 .
~ a x ~ a— 5^ 3 -— 7P 7 + •••
0
The sum of the sériés is readily computed to any degree of accu-
racy for any a.
3. Evaluate the elliptic intégral
n
2
J V \ —Æ*sin 2 qpcfqs (k < 1)
0
Expand the integrand in a binomial sériés, putting m =ÿ,

x = —Æ2 sin2 q) [see formula (5), Sec. 4.19]:
V 1 —/52 sin 2 <p= 1 —-i-£2 sin2ç — — — k* sin4 <p—ÿ-^--|- 6 6 sin<,9 —. . .
This sériés converges for ail values of 9 and admits term-by-term

intégration because it is dominated on any interval. Therefore,
J V 1—£2 sin 2 td i = <p— ^-Æ2 J sin2t d t——-i-64j sin4 1 dt

0 0 0
"“ T T T * ' J s i n ^ ^ — • • •
The intégrais on the right are computed in elementary fashion.

For if = - w e hâve
ü
2
1 -3 ...(2 /i— l) n
J sin8n(pd<p 2 4 ...2 n 2
0
(see Sec. 1 1 .6 , Vol. I) and, hence,
n
2
/ 1-3\s *■»
§ K l —A2 sin2 9 ^ 9 = — [ 1 _ ( y ) * R \_2-4 / 3 {'2-4-6J 5
4.22 Intégrâting Di/Jerential Equations by Means of Sériés_____ 301
4.22 INTEGRATING DIFFERENT 1AL EQUATIONS

BY MEANS OF SERIES
If the intégration of a differential équation does not reduce to
quadratures, one resorts to approximate *methods of integrating
the équation. One of these methods is representing the équation
as a Taylor’s sériés; the sum of a finite number of terms of this
sériés will be approximately equal to the desired particular
solution.
To take an example, let it be required to find the solution of
a second-order differential équation,
y" = F (x, y, y') ( 1)
that satisfies the initial conditions
(y)x=x, — yî ( y ) x = x t — yo (2)
Suppose that the solution y = f(x) exists and may be given in
the form of a Taylor’s sériés (we will not discuss the conditions
under which this occurs):
y = f(x) = / (*»)+— * /' (*,)+ (± ^ r (*,) + . . . (3)
We hâve to find f(x 0), f (x0), /'(*„), . . . . i.e., the values of the
dérivatives of the particular solution when x —x9. But this can
be done by means of équation ( 1) and conditions (2 ).
Indeed, from conditions (2) it follows that
/(*«)= & . r w = i i ;
from équation ( 1) we hâve
f (*o)= (l/’)x=xt = F (x, y0, y0)
Differentiating both sides of (1) with respect to x, we get
y"’ = F'x (x, y, y') + F'y (x, y , y') y' + F'y, (x, i/, y ’) y' (4)
and substituting the value x = x 0 into the right side, we find
r ( * ,) = ( / % « .
Differentiating the relationship (4) once again, we find
f l v (*o) = ( y xv)x=x.
and so on.
We put these values of the dérivatives into (3). For those va
lues of x for which this sériés converges, this sériés represents the
solution of the équation.
Example 1. Find the solution ol the équation
y" = —yx2
which satisfies the initial conditions
(V)jc=o=1. (y')x=o=°
302 Ch. 4 Sériés
Solution. We hâve
/ ( 0) = f t = l . / ' (0) = y'0 = 0
From the given équation we find (y")x - 0 = f" (0) = 0; further,
y '" = —y ’x* — 2xy, (y "')JC=0 = / '" (0 ) = 0 ,
yiV = - y * xt - 4 Xy ' - 2 y , (ÿ,v)x=# = —2
and, generally, di fièrent iating k times both sides of the équation by the Leib
niz formula, we find (Sec. 3.22, Vol. I)
yik + 2) = —yM)x2_ 2kxyi* - l>- k (k - 1 ) y<* -*>
Putting x = 0, we hâve
(fc+2) «, /*, i\ .,<*-2)
yo = —* ( « — i) i/o
or, setting fc+ 2 = /i,
ÿ <") = — ( „ _ 3 ) ( „ —2) y " - »
Whence
J/JV = —1-2, y ,8’ = —5-6yJv = (—1)* (1 -2) (5-6)
yÔ12’ = - 9 - l<tyj8) = (-1)3 (1 -2) (5-6) (9-10)
y o k>= ( - 1 )* ( 1■2) (5 •6) (9 • 10) . . . I(4* - 3) (4* - 2)]
In addition,
yô5> = 0 , y p = 0 .............. ÿ^ +‘>= 0.........
y<*>=0, |, « » = 0 ............ yl‘k+*>= 0..........
</i7, = 0, y<"> = 0............ y(o**+3>= 0, . . .
Thus, only those dérivatives whose order is a multiple of four do not
become zéro.
Putting the values of the dérivatives that we hâve found into a Maclaurin
sériés, we get the solution of the équation
y = 1 - ^ - 1 . 2 + | i (1-2) ( 5 - 6 ) - ^ - ( 1 - 2 ) (5-6) ( 9 - 1 0 ) + . . .
+ ( - 1)*‘( C r (1,2,(5'6)---[(4*“ 3)(4* ~ 2)I+ " -

By means of d’Alembert’s test we can verify that this sériés converges for ail
values of x; hence, it is the solution of the équation.
If the équation is linear, it is more convenient to seek the
coefficients of expansion of the particular solution by the method
of undetermined coefficients. To do this, we put the sériés
ÿ = a 0 + alx + a 2x*+ . . . + anxn + . . .
into the differential équation and equate the coefficients of identi-
cal powers of x on different sides of the équation.
Example 2 . Find the solution of the équation
y" = 2xy' + Ay
4.23 Bessel's Equation 303

( y ) * = o = 0 , (y')*=o = >
Solution. We set
y = a0 + axx + atx* + a,*3 + . . . -f + ••.
On the basis of the initial conditions we find
a0 = 0 , 1
Hence,
y = x + a 2x2+ a^x* + . . . + anxn -f-. . .
y ' = 1 + 2a2* + 3a3*2 + . . . + nanxn - 1 + . . .
y”= 2a 2 + 3• 2a3* + . . . + n (n — 1) anxn~2 + . . .
Putting these expressions into the given équation and equating the coefficients
of identical powers of x t we obtain
2a2 = 0, whence a 2 = 0
3* 2û3 = 2-f- 4, whence a3 = 1
4-3a 4 = 4a2 + 4a2, whence a 4 = 0
n ( n — \)a n = (n — 2) 2an- 2 + 4an_ 2, whence an 2Qn - a

n -\
Consequently,
o 1
2-1 1 2 1 1
a* 4 21 ’ a ,~ 6 3! ’ — 4! ’ ' ‘
2 1
*(fe —1)1 1
**+1 2k ~ k \ .......
o 4 = 0 ; a, = 0 ........... «** = 0 , . . .
Substituting the coefficients which we hâve found, we get the desired solution:
xtk +i
y X+ 1 2 ! +31 k\
The sériés thus obtained converges for ail values of x.
It will be noted that this particular solution may be expressed in terms of
elementary functions: taking x outside the brackets we get (inside the brackets)
an expansion of the function e*1. Hence,
y = xe**
4.23 BESSEL’S EQUATION
Bessel's équation is a differential équation of the form

+ **/' + (*2 —P2)ÿ = 0 (P = const) (1)
The solution of this équation (and also of certain other équations
with variable coefficients) should be sought not in the form of a
power sériés, but in the form of a product of some power of x by
304 Ch. 4 Sériés
a power sériés:
y = xr 2 akx» (2)
k=0
The coefficient a0 may be considered nonzero due to the indefi-
niteness of the exponent r.
We rewrite the expression (2) in the form
y = k2= 0 û**r+*
and find its dérivatives:
y ' - k=0 +
y"= S (r + k)(r + k — l)a kxr+*-*

h=0
Put these expressions into équation (1):
OD
x2 2 (r + k)(r + k — l)a*x ' +* - 2
CD CO
+x 2 (r + fc)a*xr+*-l + (x*—/72) 2 û*xr+* = 0

*=0 *=0
Equating to zéro the coefficients of x to the powers r, r + 1 ,
r + 2, . . . . r + A>, we get a System of équations:
[r(r— l) + r —p 2] a 0 = 0 or (r2—p 2)a 0 = 0'
[ ( r + l ) r + ( r + l ) —p2]ûj = 0 or [(r + l)2—p2] ax = 0
[(r+2) (r + l)+ (r+ 2 )—p2] a2+ a o= 0 or [(r + 2)2- p 2] + + a0= 0
[(r + *)(r + fc— l) + ( r + *)—/>*]a* + a*_, = 0 or

[(r+ £)2— p2K + a*_2= 0
Let us consider the équation

[{r + k f — p 2]a* + a * _ 2 = 0 (3 ')
It may be rewritten as follows:
[(r + k — p)(r + k + p)]ak + ak. t = 0
lt is given that a0+= 0 ; hence,
r 2- p 2 = 0
dherefore, r t = p or r 2 = — p.
î-et us first consider the solution for r, = p > 0 .
4.23 Besset'8 Equation 305
From the system of équations (3) we détermine ail the coeffi

cients av a4, . . . in succession; a0 remains arbitrary. For instance,
put a0 = 1. Then
ak-2 *
a^ = - k ^ f k )
Assigning various values to k, we find
ax= 0 , a, = 0 and, generally, aîm+l = 0 ,
1 _ i
°î — 2 (2p + 2) ’ a « — 2 -4 (2 p + 2 )(2 p + 4) ’ - ‘ (4)
° 2v = (— J)v + 1 2 .4 - 6 ...2v (2p + 2) (2p + 4 )...(2 p + 2v) ’
Putting the coefficients found into (2), we obtain
y i = xp [l '2 ( 2 p + 2 ) " f'2 -4 (2 p + 2 ) (2p + 4)
2-4.6 (2p + 2) (2p + 4) (2p + 6)

...] (5)
Ail the coefficients a4V will be determined, since for every k the
coefficient of ak in (3),
i^ + k r -p *
will be different from zéro.
Thus, y 1 is a particular solution of équation (1).
Let us further establish the conditions under which ail the coef
ficients ak will be determined for the second root r 2 —— p aswell.
This will occur if for any even intégral positive k the following
inequalities are fulfilled:
(rt + k y - p * ^ 0 (6 )
or
r* + k ^ p
But p — r{, hence,
ri + k=£r 1
Thus, condition (6 ) is in this case équivalent to the following
r1— rî ^ k
where k is a positive even integer. But
r ^ P, r.. = — p
hence
r1— rt = 2p
Thus, if p is not equal to an integer, it is possible to Write a
second particular solution that is obtained from expression (5 ) by
306 Ch. 4 Sériés
substituting —p for p:
yt = x~p [ 1 - 2 (— 2p + 2 )T 2 4 ( — 2p + 2) ( - 2p + 4)
xe , 1
2-4-6 ( - 2p + 2) ( - 2p + 4) ( - 2 p + 6 ) (5')
The power sériés (5) and (5') converge for ail values of x\ this
is readily found by d ’Alembert’s test. It is likewise obvious that yx
and t/2 are. linearly independent.*
The solution y x multiplied by a certain constant is called a
Bessel function of the first kind of order p and is designated by
the symbol J p. The solution yz is denoted by the symbol J „p.
Thus, for p not equal to an integer, the general solution of
équation ( 1) has the form
y = C1Jp + CiJ _p
For instance, when p = y the sériés (5) will hâve the form
\ fi x* i x* x4 i1
X L *1 2-3 + 2-4.3-5 2-4-6-3-5-7*‘ •••J
__ i r x* . x 5 *!_< 1
~ Y 7 \x ~ 3 f + 5T- ' • "J
This solution multiplied by the constant factor j/" -^ is called

Bessel’s function / , ; we note that the brackets contain a sériés
T
whose sum is equal to sin*. Hence,
J ± (x )= Y ' à ' sinjc
In exactly the same way, using formula (5'), we obtain
j __l( * ) = y r- ^ c°sx
* The linear independence of functions is verified as follows. Consider the

relation
j/2 2 (— 2p -f~2 ) ‘ 2 •4 (— 2p -f- 2) (— 2p -f 4)

y i “ X 2* X4
1 ~ 2 (2 p + 2 )+ 2 .4 (2 p + 2)(2p + 4)“ - "
This relation is not constant, since it approaches infinity as x —►0. Hence the
functions y, and ya are linearly independent.
4.23 Bessel's Equation 307
The complété intégral of ( 1) for P - = ÿ is

y ~ c kj (x) -t-c2</ (x)
2 2
Now let p be an integer which we shall dénoté by n ( n ^ 0 ) .

The solution (5) will in this case be meaningful and is the first
particular solution of ( 1).
But the solution (5') will not be meaningful because one of the
factors of the denominator will become zéro upon expansion.
For positive intégral p —n the Bessel function J n is determined
by the sériés (5) multiplied into the constant factor (when
b = 0 w e m u l t i p l y b y 1):
J»(x) 2nn\ 2(2n + 2 )n 2.4(2n + 2)(2n + 4)
2 •4 •6 (2n + 2) (2n+4) (2n + 6) •••]

or
, , V (~ l)v (*y+
J n \ x ) — Z u v! (n + v)! \ 2 J (7)
v = 0
It may be shown that the second particular solution should in

this case be sought in the form
CD
Kn (*) = J n (x) ln X + x~n 2 bkxk

k=0
Putting this expression into ( 1), we détermine the coefficients bk.
The function Kn (x), with the coefficients thus determined, mul
tiplied by a certain constant is called Bessel's function of the second
kind of order n.
T h is is the se c o n d so lu tio n of (1), w h ich , together w ith the first
one, form a lin early in d e p e n d en t System .
The general intégral will be of the form
y = C1Jn (x) + CiKn(x) (8 )
We note that
lim K„ {x) = oo
x -+ 0
Hence, if we want to consider the final solutions forx = 0, then
we must put C2 = 0 into formula (8 ).
Example. Find the solution of Bessel’s équation, for p — 0,
y"+^-y'+y=°
20 *
308 Ch. 4 Sériés
that satisfies the initial conditions: for x = 0f

y = 2 , y' = 0
Solution. From (7) we find one particular solution:
X
, ,. v (~ 1)V
(v!)* "2
v=0
Using this solution, we can Write a solution that satisfies the given initial
conditions, namely:
y = 2J0 (x)
Note. If we had to find the general intégral of this given équation we would
seek the second particular solution in the form
Ko (*) = Jo (*) In *+ 2 b*x*

k=o
Without giving ail the computations, we indicate that the second particular
solution, which we dénoté by K0 (x), is of the form
This function multiplied by some constant factor is called Bessel's funeUon of

the second kind of order zéro.
4.24 SERIES WITH COMPL EX TERMS
Let us consider a sequence of complex numbers:

^1» ^2> • • • » Zn, • • •
where
Zn = a n + ibn (n = l , 2 , . . . )
Définition 1. A complex number zt = a + ib is the limit of a se
quence of complex numbers zn = an + ib„ if
lim \zn— z0| = 0 ( 1)
n -► oo
Let us Write condition (1) in expanded form:
—20 = (a„ + ib„) — (a + ib) = (an— a) + 1 (b„— b)
lim | z„—z01= lim V (an— a)* + (bn— b)2= 0 (2 )
n -*■ œ n -*■ oo
From (2) it follows that condition (1) will hold provided the
following conditions are valid:
lim an = a, lim bn = b (3 )
n -►oo n -♦» oo
Let us form a sériés of complex numbers
w ,+ wt + . . . + w n+ . . . (4)
4.24 Sériés with Complex Terms 309
where
w„ = un + iv„ (« = 1 . 2 , . . . )
We consider a sum of n terms of the se»ies (4), which we dénoté
by s„:
sn = wl + wî + . . . + w n (5 )
where s„ is a complex number:
*-U-)+,U*) <6>
Définition 2. If the limit
lim s„ = s = A-\-iB
n oo
exists, then sériés (4) is termed a convergent sériés and s is called
the sum of the sériés:
s = ^ » , = A + ifl (7)
From condition (6 ), on the basis of (3), follow the équations
n n
A = lim 2 “*i B = lim 2 v* (8 )

n -+ cc k = 1 n -►oo k — 1
If there is no lims„, then sériés (4) is a divergent sériés.
An effective theorem for testing the convergence of sériés (4) is
the following.
Theorem 1. If a sériés made up of the absolute values ( moduli)
of the terms of sériés (4) converges,
| “'il + | “’*| + • • • + | ® » | + • • -, where \wn\ = V ul +v% (9)
then sériés (4) converges as well.
Proof. From the convergence of sériés (9) and from the conditions
|«n| < r «n + l £ = \ w n \ , K | < K « n + l'rt = |a'J
follow équations (8 ) (on the basis of the appropriate theorem on
the absolute convergence of sériés with real terms) and, hence,
équation (7).
This theorem enables us, when testing the convergence of a sériés
with complex terms, to apply ail the sufficient criteria of conver
gence of sériés with positive terms.
4.25 POWER SERIES IN A COMPLEX VARIABLE
We now consider power sériés with complex terms.
Définition 1. A sériés
• • • 4 "C„2 n-l-— ( 1)
where z = x + iy is a complex variable, x and y are real numbers.
310 Ch. 4 Sériés
and cn are constant complex or real numbers, is called a power

sériés in a complex variable. The theory of complex power sériés is
similar to the theory of power sériés with real terms.
Définition 2. The collection of values of z in the complex plane
for which the sériés ( 1) converges is called the domain of conver
gence of the power sériés ( 1) (for each concrète value of z, the
sériés (1) becomes a numerical sériés with complex terms of type (4),
Sec. 4.24).
Définition 3. The sériés (1) is absolutely convergent if the sé
riés composed of the moduli of the terms of ( 1) converges:
ICoI+ IC1Z I+ ICï2*| +• ■• • + | cnz" | + • • • (2 )
We give without proof the following theorem.
Theorem 1. The domain of convergence of a power sériés with
complex terms ( 1) is a circle in the complex plane with centre at
the origin. This circle is called the circle of convergence. The sériés
( 1) converges absolutely at points lying inside the circle of convergence.
The radius R of the circle of convergence is termed the radius
of convergence of the power sériés. If R is the radius of conver
gence of the power sériés ( 1), then we write that the sériés con
verges in the domain
U\<R
(The question of the convergence of the sériés at the boundary
points |z | = /? requires a supplementary investigation; this is si
milar to the problem of the convergence of a power sériés in a
real variable at the end points of an interval.) Note that the ra
dius R of convergence can take on any value from R = 0 to R = <x>
depending on the nature of the coefficients c„. In the former case,
the sériés converges only at the point z = 0 , in the latter, it con
verges for any value of z.
We write the équation
/ ( 2) = c0 + c1z + cîz2+ . . . + caza+ . . . (3)
If z assumes distinct values within the circle of convergence, the
function f(z) will assume distinct values. Thus, every power sériés
in a complex variable defines an appropriate function of the com
plex variable within the circle of convergence. This function is
called an analytic function of a complex variable The following
are some examples of functions of a complex variable defined by
power sériés in a complex variable.
1. «*= 1+ Z + | j + | y + . . . + ^ f + . . . (4)
This is the exponential function of a complex variable. If y —0 ,
then formula (4) becomes formula (2), Sec. 4.17. If x = 0, then we
get équation (1), Sec. 4.18.
4.25 Power Sériés in a Complex Variable 311
2. sinz = z - - + - [- j J+ . . . (5)
This is the sine of a complex variable. Fdt y = 0, formula (5) turns
into formula (1) of Sec. 4.17.
3. c o s z = l - ^ + 5 - S + --- (6 )
This is the cosine of a complex variable. If in formula (4) we put
(iz) in place of z on the right and on the left, then (like what
was done in Sec. 4.18) we get
eiz = cos z + i sin z (7 )
This is Euler’s formula for a complex z. If z is a real pumber,
then this formula coïncides with formula (2 ) of Sec. 4.18.
4. sinh z = z + 57 + 3 ] + ••• (8 )
5. coshz= 1 + 2 ! + - 4 ; + gj + • • • (9)
These last two formulas are similar to formulas (5) and (6 ) of
Sec. 4.17 and coincide with the latter when z = x is a real number.
By (4), (5), (6 ), (8 ), and (9) and via addition, subtraction of
the sériés and replacement of z by (iz), we get the following équa
tions:
e* = cosh z + sinh z, ( 10 )
e~z = cosh z — sinh z, (ii)
u e*+ e~z • 1 ez—e~~z
cosh z = —- — , sinh z = — ^— » ( 12)
e-* + e* . e“ z—ez
COS IZ = -------» sin iz = — 2i— (13)
Note that the sériés (4), (5), (6 ), (8 ), (9) converge for ail values
of z. This is clearly seen on the basis of Theorem 1 and of Sec.
4.24. As in the case of power sériés of a real variable, we can
consider sériés of a complex variable in powers of (z—z0), where z0
is a complex number, cn are complex or real constants
Co+ M z —z0) + c,(z—z0)*+ • • • + c„(z—z0)n+ . . . (14)
The sériés (14) is reduced to (1) by the substitution z—z„ = z*.
Ail the properties and theorems that are valid for sériés of type
(1) carry over to sériés of type (14), only the centre of thecircle
of convergence of (14) lies at point z0 instead of at the origin.
If R is the radius of convergence of sériés (14), then we Write:
the sériés converges in the domain
| z - z 0| < f l
312 Ch. 4 Sériés
4.26. THE SOLUTION OF FIRST-ORDER DIFFERENTIAL

EQUATIONS BY THE METHOD OF SUCCESSIVE
APPROXIMATIONS (METHOD OF ITERATION)
In Secs. 1.32, 1.33, and 1.34 we considered the approximate

intégration of differential équations and Systems of differential
équations by différence methods. Here we will consider a different
method of approximate intégration of a differential équation. It
will be noted that this is at the same time proof of the theorem
of the existence of a solution of a differential équation (see Sec. 1.2).
We will need the theory of sériés in this case.
Let it be required to find the solution to the differential équation
!=/(*. y) (i)
that satisfies the initial condition
y = y0 for * = (2 )
Integrating the terms of équation (1) from xa to x and taking into
account that y\x=x, = y<>, we get
X
y = l f ( x , y)dx + y0 (3)
*0
In this équation, the desired function y is under the intégral sign
and so the équation is called an intégral équation.
The function y = y(x) which satisfies ( 1) and the initial condi
tions (2) satisfies équation (3). It is clear that the function y = y (x)
which satisfies (3). satisfies équation (1) and the initial conditions (2).
Let us first consider a method of obtaining approximate solutions
to équation ( 1) for the initial conditions (2 ).
Wewill take y0 for the zeroth approximation of the solution.
Substituting yQinto the integrand on the right of (3) inplace of y y
we get
X
y A x ) = [ f ( x , y0)dx + y0 (4)
This is the first approximation of the differential équation (1)

that satisfies the initial conditions (2 ).
Substituting the first approximation y t (x) into the integrand
in (3), we get
X
y%(*) = S/ t*. yi (*)] dx + y0 (5)

*t
4.27 Proof of the Existence of a Solution of a Differential Equation 313
This is the second approximation. We continue the process:

X •
y* M = $ /[ * - y A x )]d x + y a,
*o
yn (x) = J / [x, t/„_, (*)] dx + yQ,

Xo
We will indicate below what approximation is to be taken so

as to satisfy our requirements of accuracy.
Fxample. Approximate the équation y' = x + y2 which satisfies the initial
conditions
y 0 = l for x= 0
X
lfx = $ (* + l)< te + l= £ -h r+ l
o 2
[ - + ( T + Je+1) ] ^ +,==S + T + 2T + 3T + Jc+ 1

and so fort h.
4.27. PROOF OF THE EXISTENCE OF A SOLUTION

OF A DIFFERENTIAL EQUATION.
ERROR ESTIMATION IN APPROXIMATE SOLUTIONS
We prove the following theorem.
Theorem. Given the differential équation
S=/(** y) (!)
and the initial conditions
y = y0 for x = x0 (2 )
Let f (xf y) and f'y (x, y) be continuons in a ctosed domain D:
D{x0— a ^ x ^ x 0+ a, y<>— b ï ^ y ^ y 0+ b\ (Fig. 123)(3)
Then, in some interval
x0— l < x < x 9+ l (4 )
there exists a solution to équation ( 1) which satisfies the initial
conditions (2); and the solution is unique. The number l will be
defined below.
314 Ch. 4 Sériés
P roof. Note that from the tact that f(x, y) and f'y (x, y) are
continuons in a closed domain D it follows that there exist cons
tants M > 0 and N > 0 such that for ail points of the domain
the following relations hold:
l/(*. y ) \ < M , (5)
\ry (x, y ) \ ^ N (6 )
(This property is similar to the property indicated in Sec. 2.10,
Vol. I.) The number l in (4) is the
(Xo 4 o * b)
smaller of the numbers a and -77
M
- (Xg*a,yg) that is,
l = min 7â ) (7)
Fig. 123 Let us apply the Lagrange theorem

to the function f(x, y) for two arbit-
rary points At (x, yx) and A2(x, y2) belonging to D:
f(x, y*)— Hx, </,) = f'y (x, ri)(yî —y l)
where yt < ti < y2, hence, | fÿ (x, t)) | ^ AL Thus, for any two points
the following inequality is true:*
l/(*. 0 .) —/(*. 0 i ) |< W |y t —ÿ,| (8 )
Let us return to équation (4) of Sec. 4.26. From this équation,
taking into account (5), (4), (7), we get
101— 00 S/(*. ya) dx < 5 Mdx = M \ x — (9)

*0
Thus, the function y = yt (x) defined by (4), Sec. 4.26, on the
interval (4) does not go outside domain D.
Now we take up équation (5) of Sec. 4.26. The arguments of
the function / [x, y, (x)] do not leave D. Hence, we can Write
X
102 — 00 $ /[•*. y A x)\dx M | x —x0 ( 10)

x„
* Note that if for some function F (y) the following condition is satisfied:
IF (y2)—F (yi) | < | y2 —0 i |
where y2 and y x are any points in the domain and K is a constant, then this
condition is called the Lipschitz condition. Thus, having established the rela
tion (8), we demonstrated that if a function f(x, y) has a dérivative
bounded in some domain, then it satisfies the Lipschitz condition in that
domain. The converse may not orove to be true.
4.27 P roof of the Existence of a Solution of a Différent ial Equation 315
By the method of complété induction we can prove that for

arbitrary n
\ y n— y ü \ qo
and the function y(x) satisfies the differential équation ( 1) and
the initial conditions (2 ). To prove this, consider the sériés
#o + ( # i — #o) + (#2 # i ) + • • • + ( # « - 1— + — y n- 1 ) + • • • (1 3 )
with general term un = y„—ÿ„_, with u0= y0. The sum of n + l
ternis of this sériés is clearly equal to
1= 2 ut = yn (14)
i=0
Let us estimate the terms of sériés (13) in absolute value:
|ÿ i— v*\ = S î (x, y 0) d x ^ M Ix —xn (15)

X,
On the basis of (4), (5), Sec. 4.26, and (6 ) we find
\y%— y i l Stf(*» yx) — f(x, y0)]dx = J/y(x, r\1)(yi — y l)dx

x0 x0
X
< ± N $ M Ix — x01dx = N — Ix — x0 \2
(the plus sign is taken if x0 < x, and the minus sign if *0 > x ).
Thus
Iy*— y x K m |x—xj! (16)
Similarly, taking into account (16),
\ y 3— y 3\ = S [/(*. y ù — fix, yt)]dx

X0
:
Sfy{x, tia) (y^— y j d x

X0
< ± N J^ | x—x0 \2dx = M 1x —x 0 (17)

316 Ch. 4 Sériés
Continuing in the same fashion, we hâve

I (18)
Thus, for the interval \ x —x0| < / the sériés (13) of functions is
dominated. The corresponding numerical sériés with positive terms,
which exceed in absolute value the corresponding terms of sériés (13), is
AA I ,
i/o +, AT/ -f- AA-gj-H-Al
Nl 2 l AA H H 3
-gj— . ,
MA A—jjj----
^ n ~ l l n
h. • • • / 1 <Yl
(19)
with general term v„ = M - ” , l" . This sériés converges, as may

readily be seen by applying d’Alembert’s test:
N „ - l [n
M ----- :---
lim Vn — lim _____ 5 *___ = lim ^ . = 0 < 1
n->œ V„—i n~.cc /!-►» n
( « - 1)1
Thus, the sériés (13) is dominated and hence converges. Since its
terms are continuous functions, it converges to a continuous func-
tion y(x). Thus
lim sn_x= lim yn = y (x) (2 0 )
where y(x) is a continuous function. This function satisfies the
initial condition since for ail n
yn(x0) ^ y 0
We will prove that the function y(x) thus obtained satisfies équa
tion (1). We again Write down the last équation of (6 ), Sec. 4.26:
0*= «/«+ $ /(* . yn-i)dx (2 1 )

X,
We will prove that

X X
lim J f [x, (/„_! (x)) dx = J / (x, y) dx (2 2 )

n-*CBX,
where y(x) is defined by équation (2 0 ).
First note the following. Since the sériés (13) is dominated, it
follows from (2 0 ) that for any e > 0 there is an n such that
\ y — yn \ < & (23)
Having regard for (23) over the entire interval (4), we can write
X X X
$ /(* , y) dx — \ f ( x , y„)dx < ± $ |/ ( x , y) — f(x, y„)\dx

Xg Xq X0
X
< ± $ N \ y — y „ \ d x ^ N R \ x — x0\
4.27 Proof of the Exittence of a Solution of a Differential Equation 317
But lime = 0. Hence

n-*cD
X X
lim J f(x, y)dx— 5 f(x, tfn) dx = 0

n-Kx> X0 X,
From this équation follows équation (2 2 ).

Now passing to the limit in both members of (21) as n —*•<»,
we find that y(x), defined by (2 0 ), satisfies the équation
X
y- =y<>+ S f ( x>y)dx (24)

Xt
From this it follows, as was pointed out above, that the func-
tion y(x) thus found satisfies the differential équation ( 1) and the
initial conditions (2 ).
Note I. Using other methods of proof it is possible to assert
that there exists a solution of équation ( 1) that satisfies the condi
tions (2) if the function f(x, y) is continuous in a domain D
(Peano’s theorem).
Note 2. Using a device similar to the one employed to obtain
relation (18), we can demonstrate that the error resulting from
replacing the solution y(x) by its nth approximation yn is given
by the formula
NnM MNn ln+l
I y — y n \ < (n+l)! -*o|n+1 (n + 1)! (25)
Example. Let us apply this estimate with respect to the fifth approximation
y t to the solution of the équation
y '= x + y i
given the initial conditions y0— 1 when x = 0.
Suppose the domain D is
D { — 1 /2 < x < 1 / 2 , —
that is, a = 1/2, 6 = 1. Then M = 3/2, N = 2. We then détermine
/ = min
By formula (25) we get

2*
î
\ y — yt>\< — 960
Note that the estimate (25) is rather rough. In the case at hand, wecouldshow
by other methods that the error is tens of times less.
318 Ch. 4 Sériés
4.28 THE UNIQUENESS THEOREM OF THE SOLUTION

OF A DIFFERENTIAL EQUATION
We will prove the following theorem.
Theorem. If a function f (x, y) is continuous and has a continuons
dérivative ^ (defined in Sec. 4.27) in a domain D then the so
lution of the differential équation
t- f ^ y ) (0
is unique for the initial conditions

y = y0 when x=-xt (2)
That is, through the point (x0, y0) there passes a unique intégral
curve of équation ( 1).
Proof. Assume there are two solutions to équation (1) satisfying
the conditions (2 ), that is, two curves emanating from the point
A (x0, y<,)-y(x) and z(x). Consequently, both functions satisfy équa
tion (24) of Sec. 4.27:
X X
0 = 0 0 + $ /(•* . y)dx, z —y0+ $ f(x, z ) =dx

x0 x.
Consider the différence
X
£(*) — *(*) = S [/(*. y)— f(x, Z)]dx (3)

*0
Transform the différence in the integrand by Lagrange’s formula
taking into account the inequality (6 ) of Sec. 4.27:
f(x, y ) - f ( x , z) = dl ^ ( y - z) (4)
From this équation we get
I f (x, y)— f (x, z) | < IV| y z | (5)
On the basis of (3), with regard for (5), we can write the inequality
X
Iy — z N \ y — z\dx ( 6)
xg
Let us consider a value of x such that |a:—x0|< - ^ - . For the sake
of definiteness, we assume that xü < for the case of x < x0 the
proof is analogous.
Suppose, on the interval x —x0 < \ y —z\ assumes a maximum
value for x = x* and let it be equal to K. Then inequality (6 ) as-
sûmes, for the point x*t the form

X* ,
Xdx = NX(x*- x0) <»NX -jr <X

x*
or
X<X
With the assumption that there exist two distinct solutions we
arrived at a contradiction. Hence, the solution is unique.
Note 1. It may be demonstrated that the solution y" I
is unique for less strict restrictions on the function /
f(x, y). See, for example, I. G. Petrovsky’s “Lee- /y=x3
tures on the Theory of Ordinary Differential Equa- J
tions”. ----- *•
Note 2 . If a function f(x, y) has an unbounded 0 ^ x
partial dérivative -în the domain, then there may be F'g
several solutions satisfying the équation ( 1) and the initial condi
tions (2 ).
Indeed, consider the équation
y' = 3 x 3/ y (7)
with initial conditions
y=0 for x=0 (8 )
H e re ^ = x«/-,/>—<■<» as «/—►O. In this case there are two solutions
to équation (7) that satisfy the initial conditions (8 ):
y = 0, y = x3
Direct substitution of the functions into the équation convinces
us that they are solutions of (7). Two intégral curves pass through
the coordinate origin (Fig. 124).
Write the first few terms of the sériés on the basis of the given general term:
1
t• * 2 u 3 u - (- ^ l
“" “ rt+1 ' 6 n (2n)V
4 u = f_l)» +iî2
„*•
n ( » + l)
u„ = V n3+ 1— y rn2+ 1.
Test the following sériés for convergence:
«• ••• • Ans• Conver&es-

7.
, i , , i = + • • • • Ans. Diverges.
LTô ~ y lô 1^30 Y ion
320 Ch. 4 Sériés
s. 2 + 4 + T + - - - + ^ r " + - - - • Ans• Diverges-

9. . . Ans. Diverges.
V~1 V* V «+6
, 0 - t + ( t ) 4+ ( t ) , + - - + ( ï ï t ï ) " + - - - • Ans- Converges-
“ • T + i+ à + 1 7 + ---F T T + --- • Diverges-
,2* T + T + ^ + 1 7 + ---+ T T ^ + -" • Ans■Converges-
Test for convergence the following sériés with given general terms:
1 1 2
13. un= ^ . Ans. Converges. U. un = Y^==-. Ans. Diverges. 15. un = ê~^jy .
Ans. Diverges. 16. un = ^ ”2- . Ans. Diverges.
17. kw= y i a—t-ô • Converges. 18. un^ 1= -r^— . Ans. Diverges. 19. Prove
n£-\-Zn-\~o n \n n
the inequality
1+ T + T + • • •+ T > '“ ^ + •> > T + T + •• •+ ÎT 1 •

20. Is the Leibniz theorem applicable to the sériés
*_________ !____ [____ !_________ !____ l j ____ !_________ ?... ,-L. ?
Ÿ 2 -1 V ^+ l V 3 -1 r 3 + l ^ - l ^ n + l^ “
Ans. It is not applicable because the terms of the sériés do not decrease mo*
notonically in absolute value. The sériés diverges.
How many first terms must be taken in the sériés so that their sum should
not differ by more than 10- ° from the sum of the corresponding sériés;
21, î ~ 2«+ ' ' ' + ( - ') " +12»+ ••• • Ans- ti = 20.
224 - t + t - t + - (- i) 4 + - - • n= io#-
23--p- — 1)” 7jr+ --- • Ans■n = 10*.
24, T ~ 2 3 ' ^ 2 T F 4 ~ 2 - 3 - 4 - 5 " ^ " ‘ + ^_1)B n T ^ ' " * A n S ‘ n = 1 0 '
Find out which of the following sériés converges absolutely:
25. 1 — p ~ + ~ p — ^ ~ + F + ’ ‘ • + ( - 1)n+1 (2n ! _ |)i + ’ ' • • Ans• Converges

absolutely. 2 6 . + + . Ans. Con-
verges absolutely.
27- T ^ - Ô + n ^ - i Ï Ï 5 + - - + (- ,)n I ^ + -- • ^ ns- Converges conditio-
nally. 28. - I + T ^ - - V r = r + T j= r + . . . + ( — 1)" T7=r + • • • • Ans- Converges

y 2 y 3 y 4 y n
conditionally.
Find the sum of the sériés:

1 , 1 1
. . . . Ans. •
29
1*2*3 ‘ 2*3*4 +' •••
‘ **+' n (n-\-\) (n + 2) 4
For what values of x do the following sériés ccmverge
xn y2 ^
30. Ans. - 2 < x < 2 . 31-
2n
__ __l .. j_ (—nn+i xn + ... Ans. -1 ^ x ^ 1. 32. 3x + 3‘jt* + 3»x» +
4^* } h
+ ... + 3 Ans. \x \ < -s-. 33. 1
lOOx , 10 000*2 . 1000000 **
1-3 1*3*5 1-3-5-7 ■f'"* *
Ans. - o o < * < oo.
34. sin *-|-2sin-^—|-4sin-^— • • + 2" sin • • • Ans. —oo < * < oo.
x , x2 , . xn
35. -------,— -J---------r -4 -... —
J----------,— . Ans. — 1 <tX < ! ..
l+ lT î 2+ V2 n+ V^î
ok 3*
36. * + i r ^ + l r ^ + . . . + _ JCn + . . . . i4ns. —- oo < * < oo.
37. * + | f * , + y ^ + . . . + ^ * » + . . . • Ans. — e < x < e.

38. 0-2-3)* . W)1 xn + . . . . Ans. —4 < x < 4.
~ 6~ * + (2/i)I
39. Find the sum of the sériés x + 2x2 + . . . + n x n + . . . ( |x | < 1). Ans.
Détermine which of the following sériés are dominated on the indicated
intervals:
40. 1 + -—+ ^ 2’i “ “ - + “ â+ -* * ( 0 < JC< 0* Ans. Dominated. 41. 1 + y - f -
, x2 , x3 , . xn . /A_ ^ 1V . KT . . . * j sin * , sin 2x ,
+ Y + T + *• *+ T + • • • (0 < * < O* A n t' Not dominated. 42. ---- - A— (-
. sin 3x . .s in nx . rA n . „
H— p — f-...H — ^ 2— K-*[0» 2îll-
-
Am'
. . .
Dominated.
Expanding Functions in Sériés
43. Expand in powers of * and détermine the interval of convergence.

10-f“X
Ans. The sériés converges for —10 < * < 10.
44. Expand cos * in powers of ^x — . Ans. ^x — —
~ 2 } T ü { X~ ^ ) + 6 7 ï ( * ' " ' t ) +•*• •

X* x*
45. Expand e -* in powers of x. Ans. 1 + ••• •
46. Expand ex in powers of (x— 2). Ans. eP+e2 (x—2) fx—2)a +
+ 3f C 2 )8+ ... .
47. Expand x3 —2x2 + 5 x —7 in powers of (x—1). Ans. —3 + 4 ( x — !) +
+ ( x - l )2 + ( x - l ) 3.
48. Expand the polynomial x10 + 2x9 —3x7 —6x€ + 3x4 + 6xs —x — 2 in a
Taylor’s sériés in powers of (x—1); check to see that this polynomial has the
21—2082
322 Ch. 4 Sériés
number 1 for a triple root. Ans. f (*) = 81 (* — l )3 + 270 (* — l )4 + 405 (* — 1)6+

+ 351 (* — 1)6 + 189 (* — 1)7 + 63 (* — 1)8+ 12 (x— l)e + (x — l)10.
jf2
49. Expand cos(* + a) in powers of x. Ans. cos a —x sin a — cos a +
JC8 X*
+ 3 fs in a + 4 r c o s a _ --- •
50. Expand Inx in powers of (x— 1). Ans. (x —1)— i-(x — 1)2 + -^-(x — l)3—
- + * - ' ) 4+ . . . •
(x -i-2)n I
1 + 2mà -- J *
52. Expand cos2* in a sériés of powers of ^* — .
2/1-1
4"" 1 (* —t )
(_- n» /i------X----^-L
Ans. ( 1*1 < co).
T+ H (2n—1)1
n - 1
53. Expand in a sériés of powers of (x-f- 1). Ans. ^ (n + I)(x + l)B

n= 0
(—2 < x < 0 ).
54. Expand tan x in a sériés of powers of ^ x —-2-^. Ans. 1 + 2 —-2-j-f"
+ ( x- t ) + - •
Write the first four terms of the sériés expansion, in powers of *, of the
following functions:
... A . x3 . 2*6 17*7 .
55. tan x. Ans. x + T + 1 F + - ^ + . . . .
Co sx a ( * x2 i 4*4 31*° \
S6' * • ^ V -2 T + 4 r-7 2 Ô ----j-
57. éarctan * a ii i X2 X3 7x4 .
Ans. 1 + X + -2 — g— 2 4 + - -
58. In (1 +e*). Ans. ln2 + - = - + i — ^ + . . .
59. esin x Ans. 1+x + i — .
60. (1+x)*. Ans. l + x 3 - ^ - + | - x 1- . . . .

y2 5*4
61. sec x. Ans. 1 + - ^ - + - ^ - + . . . .
y2 Y4 yG
62. Incos*. Ans. — ^ ~ • •
2 12 45
(kx)3 (kx)b (kx)7
63. Expand sin kx in powers of *. Ans. k x ----- ^ — 1--------
5! 7! i * ** •
64. Expand sin** in powers of * and détermine the interval of convergent.
9y2 03 y 4 06y6 0 2 /i-ly 2 /j
Ans. ------jj—|—^ — . . . + ( —I ) " - 1— ---- h*** • The sériés converges for
2! 4! (2*)!
ail values of *,
65. Expand in a power sériés of x. Ans. 1 —jc2 + jc4—jc6+ «

\ +x 2
66. Expand arctanjc in a power sériés of x.
f1 dx x^ x®
Hint. Take advantage of the formula arctan x = j ^ ^ . Ans. x — g--|—g—
1
67. Expand in a s é r ié s o f p o w e r s o f x . Ans. 1 — 2 jc + 3 jc2 — 4x3 + ...
0 + * )2
—1 < x < 1).
Using the formulas for expansion of the functions ex y sin x, cosx, ln (l+ ;c )
and (l--fjc)m into power sériés and applying various procedures, expand the
following functions in power sériés and détermine the intervals of convergence:
x? JC®
68. sinh x. Ans. —b • • • ( — « > < * < oo).
69. cosh x. Ans. 1 + ^ f + • • • (— o o < x < o o ).
70. cos* x. Ans. 1 + — ^ - (— oo < x < co).

n-\
71. ( l + * ) l n ( l + x ) . Ans. x + J ^ ( - 1 ) " (ff^ g ( |x |< l ) .

n=2
72. ( l + x ) e - * . Ans. l + £ ( - 1 )" -1 ( _ w < * < « > )•

n =2
QO
<?*—1
73. j - L - , . 4 >is. 2 ç £ l ( l * l < ^ 2 )• 74 ^«s. !+ £ + £ +
«=0
Xn-1
+ ...
n\
. .(—oo < x < oo) 75. . Ans. £ ( « + !)* » (1*1 < 0 -
n=0
„ . . , » , 2JC3 4jc6 , , -AJT- . rtJlJC"
rtJl JC" ,
76. e* sin x. Ans. * + x * + -g j------ gj- + . . . -f- V 2 n sin - j - ^ + . . . (—oo<x<eo).
51
77. In ( * + / ! + ,*). Ans. * - ~ + ^ | £ +
Xn
78. J ln ( l + x) dx. Ans. £ (_ l) » + i J £ ( | * | < 1)
0 n- 1
x oo
ârctanjc j ^ , tv„je2" * 1
79 dx. Ans. ^ (—1)”
(2/1+ 1)* ( - K * < 1).
n= 0
03
80. ^S 2 1 Id x . Ans. C + l n |* | + £ ( —1 ) « o o < x < 0 a n d 0 < x < o e ) .
21 *
324 Ch. 4 Sériés
A
dx * 9 /1 -8
81
I
0
1 —x9
Ans.
n= 1
9n=8'
82. Prove the équations
sin (a -f x) = sin a cos * + cos a sin x
cos (a + *) = cos a cos x — sin a sin x
by expanding the left sides in powers of x.
Utilizing appropriate sériés, compute:
83. cos10° to four décimais. Ans. 0.9848. 84. sin1° to fourdécimais.
Ans. 0.0175. 85. sin 18° to three décimais. Ans. 0.309. 86. sin ^ to four de-
4
cimals. Ans. 0.7071. 87. arctan 4- to four décimais. Ans. 0.1973. 88. In5 to
ü
three décimais. Ans. 1.609. 89. log105 to three décimais. Ans. 0.699.
90. arcsin 1 to within 0.0001. Ans. 1.5708. 91. ÿ e to within 0.0001.
Ans. 1.6487. 92. loge to within 0.00001. Ans. 0.43429. 93. cos 1 to within
0 .00001. Ans. 0.5403. _____
Using a Maclaurin sériés expansion of the function /( * ) = \ / a n ~\-xy com
pute to within 0 .001 : _
94. j/j3 0 . Ans. 3.107. Ans. 4.121. _96. j/iÔÔ. Ans. 7.937.
97. y/250. Ans. 3.017. 98. 84. Ans. 9.165. 99. f / 2 . Ans. 1.2598.
Expanding the integrand in a sériés, compute the intégrais:
9. J* —^
100. dx to five décimal places. Ans. 0*94608. 101. ^ e -* * d x to four
o o
n
4
décimais. Ans. 0.7468. 102. J s\n(x2)dx to four décimais. Ans. 0.1571.
o
1
T 05
arctan x
103. J e^x djctotwo décimais. Ans. 0.81. 104. J dx to three décimal
places. Ans. 0.487. 105. J cos Ÿlc dx to within 0.001. Ans. 0.764.
o
1
T 1
106. J ln ( \ + V x ) d x to within 0.001. i4ns. 0.071. 107. J e 4 dx to within
5
0.0001. Ans. 0.9226. 108. f dx to within
withi 0 .0001. Ans. 0.0214.
J Ÿ 1—x
o
0.5
109. J j to within 0.001. Ans. 0.494. 110. J 1
^ ± ^ d x . Ans. £ .
x 12
Note. When solving this exercise and the two following ones it is well to
bear in mind the équations:
V 1 _Ji2 V ( - 0 " " 1 .y* 1 Jia

Z*
n-
n —11
6 ' Zmà
- 1 n*
n—1
n n=1(2* —
12 * ZLl (2n -I)a 8
which will be established in Sec. 5.2.
l
An,, £ . 112. Çln | - i £ — Ans. ^ r .
J x b J 1— x x 4
Intégrâting Differential Equations

by Means of Sériés
113. Find the solution of the équation y”= xy that satisfies the initial con
ditions for jc = 0 , y = 1, y ’= 0.
JC® JC®
Hint. Look for the solution in the formof a sériés. Ans. 1-'
2-3 ' 2-3-5-G"
JC®*
+ • ‘ ' 2*3*5*6.. .(3k—1)3£ ' ’ •
114. Find the solution of the équation y * + x y ' + y = 0 that satisfies the initial
r® v® (—hn +M n-l
conditions for jc = 0 f y = 0, y ’= 1. Ans. x — 5—|----------- J — ------------------------
3 ' 1*3*5 1 -3-5.. .(2/1 — 1)’
115. Find the general solution of the équation
*y”+ xy'-^-^x* — y= 0
Hlnt. Seek the solution in the form^ y = xP (A0+ A iX + 4*jç*4- . • •)•
1
c ‘‘ ’ [ | - # + l r - 7 f + • • • ] + c■ ' ' , [ ‘ - I r + î f — •]
sin jc cos JC
C*116.VxFind^ Vx
the solution of the équation xy”+ y ’ + x y = 0 that satisfies the
vl vl mC
initial conditions for j c = 0 , y = \ t y ' = 0 . Ans. 1 -
22 ' (1 *2)a 2 4 (l-2-3)a 2®
JC2*
îz+ ----
(*!)a 2a*
Hint. The last two differential équations are particular cases of the Bessel
équation
jc2y 0+ jcy' + (jc®— n*)y = 0
for n = ~ 2 anc* n = 0 .
117. Find the general solution of the équation
4xy”+ 2y’ + y = b
Hint. Seek the solution in the form of a sériés jc** (a0 + Ojjc + ^jc 2 + .. •)•
Ans. Cjcos V x + C t sin V~x /
118. Find the solution of the équation (1 —jc2) y*— x y ’ = 0 that satisfies the
1 jc® 1 3 jc®
initial conditions y = 0, y' = 1 when jc = 0. Ans. x 4“ “5^"^
, 1 3 5 jc7 .
326 Ch. 4 Sériés
119. Find the solution of the équation (1 + x 2) y"-\-2xy' = 0 that satisfies

j^3 j^6
the initial conditions y = 0, y' = l when x = 0. Ans. x — ——— -—| - . . . .
J o 7
120. Find the solution of the équation y" = xyy* that satisfies
J^3 2x ^
the initial conditions y = 1, y ' = 1 when * = 0 . .4ns. 1 + * ^ -f" ~7i—h
O' Tl
-r 51 t ---
121. Find the solution ôf the équation (1—x) y ’ = \ + x — y that satisfies
the initial conditions y = 0 when jc = 0 , and indicate the interval of conver-
x2 j^3 j^4
gence of the sériés obtained. Ans.
122. Find the solution of the équation xy" + y = 0 that satisfies the initial
conditions y = 0 , y ' = 1 when * = 0 and indicate the interval of convergence.
j^4 gtl
Ans. ^ - ( n j î 2 + ( M j ^ “ (3Î)M ",' - - - + ( _ ,) n + 1 [(n — l)!]a
f—00<*<»). 2
123. Find the solution of the équation y '4 -y = 0 that satisfies the
initial conditions y=* 1, y ' = 1 when * = 0 . Ans. — * .

x
124. Find the solution of the équation y” y' + y = 0 that satisfies the
initial conditions y = 1, y ' = 0 when x = 0, and indicate the interval of con-
vergence of the sériés obtained. ,4ns. 1- +
y 2/1
Find the first three terms of the power-series expansion of the solutions ôf
the following differentiâl équations for the given initial conditions:
dy3
125. y '= x 2-\~y2; for x = 0, y = 1. Ans. \ + x + x 2-\—g—f - . . . .
py2 y3
126. y"= éy+ x; for jc = 0 , y = 1. ÿ ' = 0 . /1ns. 1+ ^ - + ^ + . . . .
X2 X2
127. £/' = sin y — sin *; for x = 0, y = 0 . /4/îs. — ----- g— . . . .
Find several terms of the sériés expansion of solutions of differential équa
tions under the indicated initial conditions:
y2 O y3
128. y" = y y '— xi when x = 0, y = 1 and y ' = 1. Ans. 1 + * + " 2f + “3]—H
+ " • • ,29> y ' = y* + * 3 when * = 0 and y — j . Ans.
+ -gr *2+ ^ j *3+ § g *4+ - • • ■ ,3°- y' = xi — y î when x= 0 and y = 0.
A n s . x 3 —^ x 7-f~—f j - ç j x 11 — . . . . 131. y '= x * y î — l when x = 0 and

X2 X^ X2
y = 1. Ans. \ — x - \ ~ —I—5— •••* 132‘ y ' —e y + x y when x =0
a A a . x2 . 2x2 . llx* .
and y = 0. Ans. x + T + ~ r + ^ À+ . . . .
CHAPTER 5
FOURIER SERIES
5.1 DEFINITION. STATEMENT OF THE PROBLEM
A functional sériés of the form

• y + a, cosjc + ftj sinx-j-aacos 2 je + fcasin 2 x + . . .
or, more compactly, a sériés of the form
ao
Y + ] £ (a„ cos nx + b„ sin nx) ( 1)
n= 1
is called a trigonométrie sériés. The constants a„, an and bn

( n = l , 2 , . . . ) are called coefficients of the trigonométrie sériés.
If sériés ( 1) converges, then its sum is a periodic function f(x)
with period 2n, sincé sinnx and cos nx are periodic functions
with period 2n.
Thus,
f(x) = f ( x + 2n)
Let us pose the following problem.
Given a function f (x) which is periodic and has a period 2ji,
Under what conditions for f (x) is it possible to find a trigonomét
rie sériés convergent to the given function?
That is the problem that we shall solve in this chapter.
Determining the coefficients of a sériés from Fourier’s formulas.
Let the periodic function f(x ) with period 2n be such that it may
be represented as a trigonométrie sériés convergent to a given
function jn the interval (—ji, n); i.e., that it is the sum of this
sériés:
ao
/ (*) = y + X («» cos nx + K sin nx) (2 )
n= 1
Suppose that the intégral of the function on the left-hand side

of this équation is equal to the sum of the intégrais of the terms
of the sériés (2). This will be the case, for example, if we assume
that the numerical sériés made up of the coefficients of the given
trigonométrie sériés converges absolutely; that is, that the
328 Ch. 5 Fourier Sériés
following positive number sériés converges:

£o + 1ai I + 1 I+ 1Û21 + 1bt | + . . . + 1a„ | + 1bn | + . . . (3)
2
Then sériés ( 1) is dominated and, consequently, it may be inte-
grated termwise in the interval from —ji to n. Let us take advan-
tage of this for computing the coefficient a0.
Integrate both sides of (2) from —« to
ji 3t qq » jt n i
cos n x d x + ÿ bn s in nx dx)
-Jl -JT / ï = l \ - JT -JT ^
Evaluate separately each intégral on the right side:
i f ' * * - =jiaa
-JT
Jl îl
^ a„ cos nx dx = a„ jj cos nx dx = s^n nx = 0

-JT -Jl
JT JT
J bn sin nx dx = bn J sin nx dx = —bn co^ nx J'"1n = 0
Consequently,
JT
J f (x) dx = jia0
-JT
whence
JT
“«= -■ § f ( x ) d x (4 )
-Jl
To calculate the other coefficients of the sériés we shall need

certain definite intégrais, which we will consider first.
If n and k are integers, then we hâve the following équations:
if n ^ k , then
J cos nx cos kxdx = 0

-JT
JT
JT
J cos nxsinkxdx = 0 ( 1)
- J•JT
l
Jl
JT
J sin nxsin kxdx = 0

5.1 Définition. Statement of the Problem 329
but if ai —k, then

JC
5 cos2 kxdx = n #
-JC
JC
^ s'mkxcoskxdx = 0 (H)
-h
TC
$ sin2kedx —ji
To take an example, evaluate the first intégral of group (I).

Since
cos nx cos kx = — [cos (n + k) x + cos (n— k) x]
it follows that
JC JC JT
J cos nx cos kx dx = y ^ cos (n + k) x dx + J cos(n—k)xdx = 0
- jc - jc -n
The other formulas of (I)* are obtained in similar fashion. The

intégrais of group (II) are computed directly (see Ch. 10, Vol. I).
Now we can compute the coefficients ak and bh of sériés (2).
To find the coefficient ak for some definite value k=£0, mul-
tiply both sides of (2 ) by coskx:
oo
f (x) coskx = ^ -c o s k x + ^ (an cos nx cos kx + bn sin nx coskx) (2 ')
n—1
The resulting sériés on the right is dominated, since its terms do
not exceed (in absolute value) the terms of the convergent positive
sériés (3). We can therefore integrate it termwise on any interval.
Integrate (2') from — n to n:
JC
— J cos kxdx
-JC
0 0 / JX JC v
+ y*. ( 5 cos nx coskx dx + bn J sin nx cos kxdx J

n- 1 \ - jc - jc J
Taking into account formulas (II) and (I), we see that ail the
intégrais on the right are equal to zéro, with the exception of the
* By means of the formulas
cos nx sin = V2 [sin (n-\-k) * —sin (n — k) x]
sin nx sin kx = x/2 [— cos (n + k) * + cos (n — k) x]
intégral with coefficient ak. Hence,

JI Jl
J f (x) coskx dx = ak J cos2 kxdx = akn
whence
ji
= — J f(x) cos kx dx (5)
Multiplying both sides of (2) by sinAuc and again integrating

from — Jt to ji, we find
Jt n
J f (x) s \n kx dx = bk J sin* kxdx = bkn,

whence
Jt
bk = ~ j f(x)sinkxdx (6 )
-Jt
The coefficients determined from formulas (4), (5) and (6 ) are
called Fourier coefficients of the function f(x), and the trigono
métrie sériés (1) with such coefficients is called a Fourier sériés
of the function f(x).
Let us now revert to the question posed at the beginning of
this section: What properties must a function hâve so that the
Fourier sériés constructed for it should con
verge and so that the sum of the const
ructed Fourier sériés should equal the va
lues of the given function at correspon-
ding points? We shall here State a theorem
that will yield sufficient conditions for
representing a function f (x) by a Fourier
sériés.
Définition. A function f(x) is called
pieeewise monotonie on the interval [a, b]
Fig- 125 if this interval may be divided by a finite
number of points xlt x2, . . . , into sub-
intervals (a, xJ, (jq, x2)......... (*n-i> b) such that the function
is monotonie (that is, either nonincreasing or nondecreasing) on
each of the subintervals.
From the définition it follows that if the function f(x) is piece-
•wise monotonie and bounded on the interval [a, 6 ], then it can
hâve only discontinuities of the first kind. Indeed, if x = c is
a point of discontinuity of the function /(x), then, by virtue of
the monotonicity of the function, there exist the limits
lim f(x) = f (c— 0 ), lim / (x) = / (c + 0 )
X —*■ c — o x —►C+ 0
5.2 Expansions of Functions in Fourier Sériés 331
i.e., the point c is a discontinuity of the first kind (Fig. 125)

We now state the following theorem.
Theorem. If a periodic function f{x) wUh period 2n is piecewise
monotonie and bounded on the interval [—n, ji], then the Fourier
sériés constructed for this function converges at ail points. The sum
of the résultant sériés s(x) is equal to the value of f (x) at the
points of its continuity. At the discontinuities of f (x), the
sum of the sériés is equal to the arithmetic mean of the limits of
f(x) on the right and on the left; that is, if x — c is a disconti
nuity of the function f(x), then
From this theorem it follows that the class of funetions that

may be represented by Fourier sériés is rather broad. That is why
Fourier sériés hâve found extensive applications in various divi
sions of mathematics. Particularly effective use is made of Fourier
sériés in mathematical physics and its applications to spécifie
problems of mechanics and physics (see Ch. 6 ).
We give this theorem without proof. In Secs. 5.8-5.10 we will
prove another sufficient condition for the expandability of a func
tion in a Fourier sériés, whiçh condition in a certain sense deals
with a narrower class of functions.
5.2 EXPANSIONS OF FUNCTIONS IN FOURIER SERIES
The following are some instances of the expansion of functions

of Fourier sériés.
Example 1. A periodic function f (x) with period 2n is defined as follows:
f (x) = x, —n <x n
This function is piecewise monotonie and bounded (Fig. 126). Hence, it
admits expansion in a Fourier sériés.
y
Fig. 126
By formula (4), Sec. 5.1, we find

n
332 Ch. 5 Fourler Sériés
Applying formula (5), Sec. 5.1, and integrating by parts, \ve find
a* = ir f * C0S^ ^ = i r [ * !1T - T n- T f »«"**“*j=

By formula (6), Sec. 5.1, we hâve
6*= T f■ = | cosfadrj = (-l)*+ i 4
Thus, we get the sériés

sin 2x sin 3jc +i sin kx
~~2~ ~~k
This équation occurs at ail points except at points of discontinuity. At each
discontinuity, the sum of the sériés is equal to the arithmetical mean of its
limits on the right and left, which is zéro.
Î7ï -*tji I n - ln -n 0 n Zn Sn 4n 5ji
Fig. 127
Example 2 . A periodic function f (x) with period 2ji is defined as follows:

f (x) = — x for — ji ^ x ^ 0
f (x) = x for 0 < x < ji
[or / ( x) = \x\] (Fig. 127). This function is also piecewise monotonie and boun-
ded on the interval — j i ^ x C j i .
Let us détermine its Fourier coefficients:
-n L- * o
1
ji
r? r l
\ (— x) cos kx dx + \ x cos kx dx
1-----3t 0 J
r x sin /fjc |o 1 0 x sin kx

ji ------— L + T ' J s,n**d'+
L -JI
0 for k even
1 r cos kx COS kx 1*1 4
We thus obtain the sériés

* n 4 ["cos x . cos 3x , cos 5* , , c o s (2 p + l)* , 1
,K ) 2 Ji |_ 12 “T 32 52 + — (2p+J)2 r .- .J
This sériés converges at ail points, and its sum is equal to the given fonction.
Example 3. A periodic fonction f (x) with period 2jt is defined as follows:
/(*) = — 1 for — ji < x < 0
f (x)= 1 for 0 C n
This fonction (Fig. 128) is piecewise monotonie and bounded on the interval
— Jl.
y,
-3 71 -2 71 - 71 0 71 271 3n x
-1
F ig .'128
Let us compute its Fourier coefficients:

jt r o ” 1
" • 4 Î ^ ^ 4 J < - » * + $ * =o 4
-Jt •----JT 0 -I
o ji n
f , , f » , I . s i n f o |0 sinfc*|*
ak = — \ (—1) cos kxdx-\- \ coskxdx = — 1
* ji kn I—Jt /jjx [o
-jt o J
o Jt "I
cos&c |0 cos kx
bk = —
* ji
J (—1) sin kxdx + J sin kxdx 1= -^ -£
-jt o J :]
0 for k even
= _ [, _ c°s n *1
=
{l V nk
for k odd
Conséquent!y, for the fonction at hand the Fourier sériés has the form
sinx , sin3x , sin5x . . sin ( 2p + l)x
1 + 2p-f 1
This équation holds at ail points with the exception of discontinuities.
Fig. 129 illustrâtes how the partial sums sn of the sériés represent more and
more accurately the fonction f (x) as n oo.
Example 4. A periodic fonction f (x) with period 2 j i is defined as follows:
f( x ) = x a, — (Fig. 130)
Détermine its Fourier coefficients.
JT.
n 2jt2
- ji 3
n
n
_2_ X cos kx n , 1 f , ■
nk k -n + T J C° SkxdX
~rr for k cven

k*
= ^ 2[JlC0S kn]=
4 - for k odd
Thus, the Fourier sériés of the given function has the form
. ji 2 . / cos x cos 2x , cos 3jc
* = 3 ~ '4H ------V - + — s-----
Since the function is piecewise monotonie, bounded and continuous, this équa
tion holds at ail points.
Putting x = n in the équation obtained, we get
n= 1
Example 5. A periodic function f (x) with period 2jc is defined as follows:

/ (x) == 0 for — jc < x <; 0
f( x ) = x for 0 < * ^ jc (Fig. 131)
0 n
1 ji2 n
x dx
fl#=-sr I f{x)dx= T
-n
S■n 0dx+S
0
n ~2~~2
x sin kx I*
cos kxdx = —
— H * ji k |o
1 cos kx
nk k
~ w * i0Tk odd
0 for k even
jt r « n
c 1 T * u a 1 x cos kx « , 1 P . ,
b k = — \ x sir\ kxdx = — \ ------- — -\—^ \ c o s k x d x \
o L o J
for k odd
= _ _ c°s k n = i
— r- for k even
k
The Fourier sériés will thus hâve the form

. , v ji 2 / cosx , cos 3* . cos 5x , \ , fsinx sin : sin 3jc
f W =-4~n + 2 “ 3
At the discontinuities of the function /(*), the sum of the sériés is equal to
the arithmetic mean of its limits on the right and left în this case, to the
number
Putting x = 0 in the équation obtained, we get
1
(2n - l )2
n =î
5.3 A REMARK ON THE EXPANSION OF A PERIODIC

FUNCTION IN A FOURIER SERIES
We note the following property of a periodic function with
period 2 j i :
Jt X+2JI
$ \|) (x) dx = J if (x) dx
-JI X
no matter what the number X,
Indeed, since
it follows that, putting x = | —2n, we can Write (for ali c and d):
d d+ 2Jt d+2Jt d + 231
5 (x) dx --= 5 2 n) d l = 5 (£) d\ = J 'K-v)dx

c c+ tn c + îj i c+ tn
In particular, taking c = — n, d = X, we get
X X+2Jt
5 t|)(x)dx= ^ (x) dx
-Jt n
therefore,
X+2ji -Jt ji X+2jt
J t|>(x)dx = ^ ÿ ( x ) d x + ^ ÿ ( x ) d x + J 1|i(x)dx
X X -Jt Jt
“ JI Jt X Jt
= 5 yp(x)dx+ J t|>(*)dx + J T|)(x)dx= ^ ÿ(x)dx
X -Jt -J t -J t
This property means that the intégral of a periodic function tip(x)

over any interval whose length is equal to the period always has
the santé value. This tact is readily illustrated geometrically: the
cross-hatched areas in Fig. 132 are equal.
5.3 Expansion of a Periodic Function in a Fourier Sériés 337
From the property that has been proved it follows that when
computing Fourier coefficients we can replace the interval of in
tégration (— n, n) by the interval of intégration (X, À+ 2 n), that
is, we can put •
\+2n X+2n
«0 = - j* f(x)dx, an = J f (x) cos nx dx

X
X+2jt ( 1)
b„ = f (x) sin tixdx
x
where A, is any number.
This follows from the fact that the function f(x) is, by hypothe-
sis, periodic with period 2 n; hence, both the functions f(x)cosnx
and f(x)s\nnx are periodic functions with period 2jt. We now
illustrate how this property simplifies the process of finding coef
ficients in certain cases.
Example. Lct it be required to expand in a Fourier sériés the function

f (x) with period 2 .i, which is given on the interval 0 ^ x ^ 2.ti by the équation
/( x ) =x
The graph of f (x) is shown in Fig. 133. On the interval [— ji, jiJ this func
tion is represented by two formulas: f(x) = x + 2n on the interval [— n, OJ
and f{x) = x on the interval [0, Jt]. Yet, on [0, 2r] it is far more simply
represented by a single formula f(x) = x. Therefore, to expand this function
in a Fourier sériés it is better to make use of formulas (1), setting A,= 0:
2n 2n
a° = iSf<x) dx= T § xdx=2n

0 0
2jt 2jt
1 T c/ v . 1 P . 1 [*sinrtx . cos nx~\
an= — \ f (x) cos nx dx ==— \ x cos nxdx = — -----------------s— =0
" îl J ' JIJ jx ^ 1 /i2 J o
o 0
2jt 2jï
. 1 P £/ x . . 1 T . , 1 T x cos nx . sin nx 12jc 2_
= — \ / ( jc) sin ha: = — \ jt sin nx dx = — --------------------- 5— =
Ji J î t j jx L / i ‘ n2 Jo n
o o
Consequently,
2 2 2 2
/ (x) = ji — 2 sin jc— —sin 2jc — r- sin 3jc — T sin 4jc — =- sin 5jc— . . .
Z o 4 o
This sériés yields the given function at ail points with the exception of points
of discontinuity (i.e., except the points = 0 , 2ji, 4jx, At these points the
jc
sum of the sériés is equal to the half sum of the limiting values of the function
f (x) on the right and on the left (to the number ji, in this case).
5.4 FOURIER SERIES FOR EVEN AND ODD FUNCTIONS

From the définitions of an even and odd functions it follows
that if i|)(jc) is an even function, then
Jt Jt
^ i|) (x) dx = 2 J (x) dx
-Jt 0
Indeed,
Jt 0 Jt Jt Jt
J i|) (x) dx — ^ ij) (x) dx + J (jc) dx = J i|>(— x) dx + J ip (x) dx
-J t -J t 0 0 0
Jt Jt Jt
= ^ i|)(x)dx + J '!>(*)dx = 2 J (x)dx
o o o
since, by the définition of an even function, t|>(— x) —i|j (x).
It may similarly be proved that if <p(x) is an o d d function, then
Jt Jt Jt Jt Jt
Ç <p(x)dx=$<p(— x) dx + ^ q>(a;) dx — — ^ <p(x) dx + J <p(x) dx — 0
-n o o o o
If an o d d function f(x) is expanded in a Fourier sériés, then
the product /(x) coskx is also an odd function, while /(x)sinÆx
is an even function; hence,
Jt
«0 = 4 - J f(x)dx = 0
Jt
aA= — J / (x) cos kx dx = 0 o)
-J t
Jt Jt
bk = J f(x) sin kxdx = — J / (x) sin kx dx
-Jt o
5.5 The Fourier Sériés for a Function with Period 21 339
Thus the Fourier sériés of an odd function contains “only sines”

(see Example 1, Sec. 5.2).
If an even function is expanded in a Fourier sériés, the pro-
duct f(x)s\nkx is an odd function, whfle f(x) coskx is an even
function and, hence,
n
a0:= — ^ f ( x ) d x
0
Jt
ak = J f (x) cos kx dx ( 2)
o
Jt
bk = J f ( x ) s mkx d x = 0
-Jt
Thus, the Fourier sériés of an even function contains aonly cosines”

(see Example 2 , Sec. 5 .2 ).
The formulas obtained permit simplifying computations when
seeking Fourier coefficients in cases when the given function is
even or odd. It is obvious that not every periodic function is
even or odd (see Example 5, Sec. 5.2).
Example. Let it be required to expand in a Fourier sériés the even function
f (x) which has a period of 2ji and on the interval [0 , j i ] is given by the équ
ation
y=x
We hâve already expanded this function in a Fourier sériés in Example 2,
Sec. 5.2 (Fig. 127). Let us again compute the Fourier sériés of this function,
taking advantage of the fact that the given function is even.
By virtue of formulas (2) = 0 for any k\
JI Jt
2 2
^ xd x = ji, Vk = J x cos kx dx
,# = T ji
0 0
_ 2 Tx sin kx , cos k x ' 2 ( 0 for k even
Jl i * + k2 | o : “ ÏÏF
for k odd
We obtained the same coefficients as in Example 2, Sec. 5.2, but this time by
a short eut.
5.5 THE FOURIER SERIES FOR A FUNCTION

WITH PERIOD 21
Let f (x) be a periodic function with period 21, generally
speaking, different from 2n. Expand it in a Fourier sériés.
Make the substitution
22 *
Then the function will be a periodic function of t with

period 2 n.
It may be expanded in a Fourier sériés on the interval — n ^
sC x sC ji :
( 1)
where
n n
Jt
Now let us retum to the original variable x :

x=—
ji
t,’ t — x-/7 -’, dt — ^ -/ d x
We will then hâve
1 t
a 0 = y j f(x)dx, a„ = ÿ j / (x) cos k ^-xdx
~r l - l
l ( 2)
bk = j- J / (x) sin k-j-xdx

-1
Formula ( 1) takes the form
(3)
where the coefficients a0, ak, bk are computed from formulas (2 ).

This is the Fourier sériés for a periodic function with period 21.
We note that ail the theorems that hold for Fourier sériés of
periodic functions with period 2 jï hold also for Fourier sériés of
periodic functions with some other period 21. In particular, the
sufficient condition for expansion of a function in a Fourier sériés
(see end of Sec. 5.1) holds true, as do also the remark on the
possibility of computing coefficients of the sériés by integrating
over any interval whose length is equal to the period (see Sec. 5.3),
and the remark on the possibility of simplifying computation of
coefficients of the sériés if the function is even or odd (Sec. 5.4).
Example. Expand in a Fourier sériés the periodic function f(x) with period
2/ which on the interval [— l, l] is given by the équation /(*) = |x | (Fig. 134).
5.6 Expansion of a Nonperiodic Function in a Fourier Sériés 341
Solution. Since the function at hand is even, it follows tliat

i
= 0 . a0= j ^ x d x T l
0
/ Jl / 0 for k even
2 p knx 21 f I 4/
ak = — \ x cos - j - dx = -^ ô- \ x cos kx dx= < for k odd
o o V j i 2A2
Hence, the expansion is of the form

3 jt (2p+l)jx
—x cos — x cos ' j— x
'
Z 41 f"cos T" *
2 JISL 1 3* (2p + l )2 ]
5.6 ON THE EXPANSION OP A NONPERIODIC FUNCTION
IN A FOURIER SERIES
Let there be given, on some interval [a, b], a piecewise mono
tonie function f (x) (Fig. 135). We shall show that this function
f(x) may be represented in the form of a sum of a Fourier sériés
at the points of its discontinuity. To do this, let us consider an

arbitrary periodic piecewise monotonie function /,(x) with period
2 p. ^ b —a, which coïncides with the function f (x) on the inter-
val [a, b]. [We hâve redefined the function f(x).]
Expand h(x) in a Fourier sériés. At ail points of the interval
[a, b] (with the exception of points of discontinuity) the sum of
this sériés coïncides with the given function f(x)\ in other words,
342 C/i. 5 Fourler Sériés
we hâve expainded the function f(x) in a Fourier sériés on the

interval [a, b].
Let us now consider the following important case. Let a func
tion fyx) be given on the interval [0, /]. Redefining this function
in arbitrarv fashion on the interval [— /, 0 ] (retaining piecewise
monotonicity), we can expand it in a Fourier sériés. In particular,
if we redefine this function so that when — / < 0 , f (x) ==/(—*),
we will get an even function (Fig. 136). [In this case we say that
tains only cosines. Thus, we hâve expanded in cosines the function

f(x) given on the interval [0 , /].
But if we redefine the function f(x) when — l ^ x < 0 as follows:
f {x) = —/ (—x), then we get an odd function which may be expand
ed in sines (Fig. 137). [The function f(x) is “continued in odd
fashion”.) Thus, if on the interval [0, /] there is given some piece
wise monotonie function f(x), it may be expanded in a Fourier
sériés both. in cosines and in sines.
Example 1. Let it be required to expand the function f ( x ) —x in the sine
sériés on the interval [0 , n).
Solution. Continuing this function in odd fashion (Fig. 126), we get the sériés
„ T sin x sin 2x , sin 3x 1
X= 2 [ - ----- 2~+—--- -J
(see Example I, Sec. 5.2).
Example 2. Expand the function f(x) = x in the cosine sériés on the interval
[0 , n].
Solution. Continuing this function in even fashion, we get
/ ( X )= \X \, — Jl < X < t T L
(Fig. 127). Expanding it in a sériés we find

r , , ji 4 f cos x . cos 3x . cos 5x . 1
n x ) = 2 ~ n [— + - P - + — + •••]
(see Example 2, Sec. 5.2). And so on the interval [0, j i ] we hâve the équation
n 4 Ttos x . cos 3x . cos Sx , 1
* = 2 ~ n 1 — i— 1— 3*— •— BS— H - . - I
5.7 Mean Approximation of a Given Function 343
5.7 MEAN APPROXIMATION OF A GIVEN FUNCTION

BY A TRIGONOMETR1C POLYNOMIAL
Representing a function by an infinité sériés (Fourier’s, Taylor’s

and so forth) has the following meaning in practice: the fi ni te
sum obtained in terminating the sériés with the >zth term is an
approximate expression of the function being expanded. This -appro-
ximate expression may be made as accurate as desired by choosing
a sufficiently large value of n. However, the character of the
approximate représentation may
differ.
For instance, the sum of the
first n ter ms sn of a Taylor sé
riés coïncides with the function
at hand at one point, and at
this point has dérivatives up to
the nth order that coincide with
the dérivatives of the function
under considération. An nth
degree Lagrange polynomial (see
Sec. 7.9, Vol. I) coïncides with the function under considération
at n + 1 points.
* Let us see what the character is of an approximate représen
tation of a periodic function f (x) by trigonométrie polynomials
of the form
n
sn (x) = -y + ^ ak cos kx + bk sin kx

k=\
where a0> aly bly a2, &2, . . . , ant bn are Fourier coefficients; that
is, by the sum of the first n terms of a Fourier sériés. We first
make several remarks.
Suppose we regard some function y = f(x) on the interval [a, b]
and want to evaluate the error when replacing this function by
another function <p(*). For the measure of error we can, for in
stance, take max | f (x) — <p(x)| on the interval [a, b], which is
the so-called maximum déviation of cp(jc) from f(x). But it is
sometimes more natural to take for the measure of error the so-called
root-mean-square déviation ô, which is defined by the équation
b
62 = J [/ (*)—9 WJ*dx
a
Fig. 138 illustrâtes the différence between the root-mean-square

déviation and the maximum déviation.
Let the solid line depict the function y = f(x), the dashed Unes
the approximations qpA(jc) and <p2 (x). The maximum déviation of
the curve y = <p1 (jc) is less than that of the curve y = q>2(*), but the
root-mean-square déviation of the first curve is greater than that of
the second because the curve z/ = <p2 (x) is considerably different from
the curve y — f (x) only on a narrow section and for this reasofi
characterizes the curve y = f(x) better than the first.
Now let us return to our problem.
Let there be given a periodic function f(x) with period 2n.
From among ail the trigonométrie polynomials of order n
+ ^ (a*cos/uc + p*sin&t)
*=i
it is required to find (by ckoice of the coefficients a* and p*) that
polynomial for which the root-mean-square déviation defined by the
équation
ôî f(x) — y —^ (a* cos kx -f- p* sin kx) 1 dx
k= î
has the smallest value.
The problem reduces to finding the minimum of the function
of 2 / i + l variables a„, a lt . . . . a„, pit p„, p„.
Expanding the square under the intégral sign and integrating
termwise, we get
= i S j / 2 (x) — 2f(x) [^T + è (a * cos kx + P* sin Aur)j
+ + («*cos^ + P*sin ^ ) j
7t 71 n Jl
= i S /*(*)<**— § - 5 Î W dx~ l c Ÿ ., a * y f(x)coskxdx

-71 -71 k- 1 -JT
— P* J f (x) sin kxdx + j ^ - ^ ft dx + — £ oe? \ cos*kxdx

k=\ -JT —JT h I
cO / v'
+ i Z P*f sinl k x d v
k - 1 -JT
- > t L '** f
k= 1 - :t
“ » L p* f ' in k v <**- i 4 £ X, a *a / J kxro> jx dx

k=\ 1 /=1
We note that
n n
— j f(x)dx = a0, - j f (x) cqskxdx = ük
-J l -J l
ji
j f (x) sin kxdx = b„

-J l
are the Fourier coefficients of the function f(x).
Further, by formulas (I) and (II), Sec. 5.1, we hâve: for k = \
Jl Jl
J cosi kxdx = n, J s'm2kxdx = n
-J l -J l
for arbitrary k and /

Jl
J sin kx cos jx dx = 0
and for k =# j
n jx
J cos kx cos jx dx = 0 . J sin Æx sin jx dx = 0
-n -n
Thus, we obtain
Jl
i I r w d x - z g — ;£ > * < * * + p a )+ f + y i> * + p »

-ji k= i k = î
Adding and subtracting the sum
t + t i > * + 6î>
k=l
we will hâve
I
-J l
è ( û ï+ f c ii) + T ( a o-«o)a
k = 1
(i)
k= I
The first three terms of this sum are independent of the choice
of coefficients a 0, a 1( . . . . a„, plt . . . . p„. The remaining terms
l ( a 0 - a 0) * + | +
k= i
are nonnegative. Their sum reaches the least value (equal to zéro)
if we put a 0 = a0, a , = «„ . . . . a„ = a„, 0 , = ^ .......... P„ = &„.

With this choice of coefficients a0, a lt . . . , a„, pt, . . . , p„ the
trigonométrie polynomial
n
Y + X (“ * cos kx + P* sin kx)

k=\
will least of ail differ from the function f(x) in the sense that in
such a choice of coefficients the square déviation will be the least.
We hâve thus proved the theorem:
Of ail trigonométrie polynomials of order n> that polynomial has
the least root-mean-square déviation from the function f(x)y the
coefficients of which polynomial are the Fourier coefficients of the
function f(x).
The least square déviation is
3t g n
i j P W à x - - ° f - ~ ^ (al + bl) (2 )
- Jl k= 1
Since ô £ ^ 0 , it follows that for any n we hâve

n 2 n
— fj r ( x ) d x > - ^ + - jr ( a t + bgj
- Jt k- 1
Hence, the sériés on the right converges (as n —►oo), and we
can write
f > ( x ) d x ^ f + j ^ ( a % + bl) (3)

-Jt k- 1
This relation is called Bessel’s inequality.
We note without proof that for any bounded and pieeewise
monotonie function the root-mean-square déviation obtained upon
replacing the given function. by the nth partial sum of the Fourier
sériés tends to zéro as n —<-oo, that is, ôjj—>-0 as n —*■«>. But
then from formula (2 ) there follows the équation
4 +
i t (at + w = - k ] f t (x ) d x <3')
fc=i - ji
which is called the Lyapunov-Parseval équation. We note that
this équation has been proved for a broader class of functions
than that which we here eonsider.
From what has been proved it follows that for a function which
satisfies the Lyapunov-Parseval équation (in particular, for any
bounded pieeewise monotonie function), the corresponding Fourier
sériés yields a root-mean-square déviation equal to zéro.
Note. Let us establish a property of Fourier coefficients that

will be needed in the future. We first introduce a définition.
A function f(x) is called piecewise continuous on an interval
(a, b] if it has a definite number of discontinuities of the first
kind on this interval (or is everywhere continuous).
We shall prove the following proposition.
I f a function f (x) is piecewise continuous on the interval [— n, ji],
then its Fourier coefficients approach zéro as n —►oo; that is,
lim = 0, lim £>*<= 0 (4)
n -+• oo n -* oo
P ro o f. If the function / ( a ) is piecewise continuous on the inter-
val [— n, n], then the function / 2 (x) too is piecewise continuous
JT
on this interval. Then J f*(x)dx exists and is a finite number.*

-JT
In this case, from the Bessel inequality (3) it follows that the
oo
sériés 2 (ah+ b\) converges. But if the sériés converges then its
n= 1
general term approaches zéro; in this case, lim (a%+ b%) = 0 .
n -►0
Whence we get équations (4) directly. Thus, the following équations
are valid for a piecewise continuous and bounded function:
JT
lim \ f (x) cosnxdx = 0

:n
JT
lim J f (x) sin nx dx = 0

®-n
If a function f(x) is periodic with period 2n, then the latter
équations may be written as follows (for any a):
a + 2n a -f 2 jt
lim J f (x) cos nxdx = 0 , lim ^ f(x)si nnxdx = 0

n -.® o n -* » o
We note that these équations continue to hold if in the intégrais

we take any arbitrary interval of intégration [n, b], which is to
say that the intégrais
b b
^f (a ) c os nx dx and Jf (a) sin nx dx
a a
* This intégral may be presented as the sum of definite intégrais of con

tinuous functions over the subintervals into which the interval [— jt, jt] is
subdivided.
ap p ro ach zéro w hen n increases w ith o u t b o und if / (x) is a b o u n d

ed and piecew ise c o n tin u o n s fu n c tio n .
In d eed , ta k in g b— a < 2n for definiteness, we co n sid er th e a u x i-
liary fu n ctio n (x ) sin nx dx
a a
Since <p(x) is a b o u n d ed an d piecew ise c o n tin u o u s fu n c tio n , th e

in té g ra is on th e r ig h t a p p ro a c h zéro as n —<•<». H ence, th e in té g
ra is on th e left a p p ro a c h zéro as w e ll. T h u s, th e p ro p o sitio n is
prov ed ; th a t is,
b b
lim [ f (x)cosnxdx = 0, litn { f(x )s ln n x d x = 0 (5)
fi —
►oc a n —
►® g
for an y n u m b e rs a a n d b and an y fu n c tio n /( * ) piecew ise c o n t
in uo u s a n d b o u n d ed on [a, b ] .
5.8 THE DIR1CHLET INTEGRAL

In th is sectio n we sh a ll d ériv é a fo rm u la th a t ex p resses th e n th
p a rtia l su m of a F o u rie r sériés in te rm s of a c e rta in in té g ra l. T h is
fo rm u la w ill be n eeded in th e su b sé q u e n t sections.
C o n sid er th e n th p a r tia l sum of a F o u rie r sériés for th e peri-
o d ic fu n c tio n f (x) w ith p e rio d 2n:
s„ ( * ) = - y + X («* cos kx + bk sin kx)
w here
ji n
a* = - J f ( t ) cos k t dt , b„ = ^ j f (t) sin kt dt

-Jt -Jt
P u ttin g th ese e x p re ssio n s in to th e fo rm u la for sn (x), we o b ta in

Jt
s"(*) = i S f (0 d t
-Jt
Jt Jt
J / (t) cos kt dt + Sin^ x J f (t) sin kt dt

-Jt -Jt
5.8 The Dirichlet Intégral 349
or b rin g in g c o s k x and sin k x u n d e r th e in tég ral sig n (w hich is

p o ssib le sin ce cos fcc and sink*: are in d e p e n d e n t of th e v a ria b le
of in té g ra tio n a n d , hence, can be re g a rd e d as c o n sta n ts), we get
n
-Jt
Jt Jt "I
J cos k x cos + J (/) sin sin

+Î I k=l -Jt
f (t) k td t
-Jt
f kx kt dt
J
N ow ta k in g — o u ts id e th e b ra c k e ts an d re p la c in g th e sum of in tég
ra is by th e in té g ra l of th e su m , we o b ta in
H t)
^ [ / ( 0 cos kx cos k t + f (t) sin kx sin kt] t dt
k=i
or
sn (x) = J /(/) (cos kt cos kx + sin kt sin kx) dt

-jt L k=i J
= ^ J /(0 | ■
7 + ] £ c o s k ( t — x Udt ( 1)
-n L ft=t J
T ran sfo rm th e e x p re ssio n in th e b ra c k e ts. L et
o„ ( 2 ) = 4- + cos z - f cos 2z + . . . + cos nz
th e n
2 <j„ (z ) cos z = cos 2 + 2 cos 2 cos 2 + 2 cos 2 cos 2 2 + . . .
+ 2 cos 2 cos nz = cos 2 + (1 + cos 2 2 ) + (cos 2 + cos 3 2 )
+ (c o s 2 2 + c o s 42) + . . . + [ c o s ( n — l ) z + c o s (n + l ) z ]
= 1 + 2 c o s 2 + 2 c o s 2 2 + . . . + 2 c o s ( n — l ) z + c o srtz + c o s (n + l ) z
or
2cr„ (2 ) cos 2 = 2a„ (z )— cos n z + cos ( n + 1) z
cos nz — cos ( n + \ ) z
On ( 2 ) 2 ( 1— cos z)
But
cos n z — cos (n + l ) z = 2 s i n ( 2 n + l ) - |- s i n - |-
1 — cosz = 2 3 ^ - 2
H ence,
sin (2n + 1)
<*«(*) = ----------—
2sin T
T h u s, é q u a tio n (1) m ay be re w ritte n as
t —x
sin (2/t + 1)
s „ (*) = -£• \ / ( O t —x
dt
2 sin
S in ce th e in te g ra n d is p e rio d ic (w ith p erio d 2 n ), it follow s th a t

th e in té g ra l re ta in s its v a lu e o n an y in te rv a l of in té g ra tio n of
len g th 2 n . W e can th erefo re w rite
t —x
sin (2 n + 1)
M *) = - \ H t) ■dt
n2 sin
■ t——x
t-—
In tro d u c in g a new v a ria b le a , we p u t
t — x = a, t = jc + a
T hen w e g et th e fo rm u la
sin ( 2 n + l ) y
stt(x) = - \ f ( X + a) da (2)
4S 2sinf
T he in té g ra l on th e rig h t is D i r i c h l e t ’ s i n t é g r a l .
In th is fo rm u la p u t / ( x ) = 1 ; th e n a 0 = 2, a k = 0, b k — 0 w hen
k > 0; hence, s „ ( * ) = l for an y n a n d we g et th e id e n tity
sin (2/t+ l)---

_l_
da
Jl
S o • a
2sln 2
(3)
w h ic h w e w ill n eed la te r on.

5.9 The Convergence of a Fourièr Sériés at a Given Point 351
5.9 THE CONVERGENCE OF A FOURIER SERIES AT A GIVEN POINT

Assume that the function f (x) is piecewise continuous on the
interval [—n, n]. •
Multiplying both sides of the identity (3) of the preceding
section by f(x) and bringing f(x) under the intégral sign, we get
the équation
Jt
sin (2 n + l)-j
/w -4 \.f w a da
2sini
-Jt
Subtract the terms of the latter équation from the corresponding

terms of (2 ) of the preceding section; we get
31
sin ( 2 n + l ) y
[/ (x + a)— f(x)] da
sn ( x ) — f ( x ) = ~
S
- Jt
2o SmT
• a
Thus, the convergence of a Fourier sériés to the value of a func

tion f(x) at a given point dépends on whether the intégral on the
right approaches zéro as n —►<».
Let us break up this intégral into two intégrais:
Jt
a
cos ~2
[f (x + a )—/ (*)]------ — sin na da
«»(*)—/(*) = i r
J OC
2 sin y
+ T1T j* [/(* + “)—/(*)] cosnada

-Jt
taking advantage of the fact that sin (2/1 + 1) y = sinnacos —

-f-cosnasinY . Break up the first of the intégrais on the right
of the latter équation into three intégrais:
a
C0S 2"
s„ (x ) [f(x + a)— f(x)] sin na da
2 s in f
-ô
-6 a
i r cos ~2
+ Y \ [/(JC+ a)—/(*)]---- --Sinnada
J 2 sin -pr
n a
+ - f [/ (Jf + a) —/ W ] ------ -- sin na da
i 2sinT
n
+i Ç [f(x+ct) — f (x)]jCOsnada
-3 1
Put <D, (a) = flx + ay -f (*).' Since f (x) is a bounded piecewise con-
tinuous function, it follows that O, (a) is also a bounded and
piecewise continuous periodic function of a. Hence, the latter inté
gral approaches zéro as n —►oo, since it is a Fourier coefficient
of •this function. The function oc
cos T
(«) = [/ (* + « ) - / ( * ) ] -------z—
2sinT
is bounded wnen —j t 2 (a) is piecewise continuous. Hence, by formulas (5) of
Sec. 5.7, the second and third intégrais approach zéro as n —*-oo.
We can thus Write
i r cosf
üm [sn(x) — f(x)] =lim - - [ [ / ( r + a ) - / ( j : ) ] ----- — sin n a d a ( 1)
«-►<» n-+ » 2 s in —
In the expression on the right, the intégration is performed over

the interval —ô ^ a ^ ô ; consequently, the intégral is dépendent
on the values of the function f (x) only in the interval from x — 6
to jc + fi- An important proposition thus îollows from the last
équation: the convergence of a Fourier sériés at a gipen point x
dépends only on the behaviour of the function f (x) in an arbitrarily
small neighbourhood of this point.
Therein lies the so-called principle of localization in the study
of Fourier sériés. If two functions f t (x) and f 2(x) coincide in the
neighbourhood of some point x, then their Fourier sériés simulta-
neously either converge or diverge at this point.
5.10 CERTAIN SUFFICIENT CONDITIONS
FOR THE CONVERGENCE OF A FOURIER SERIES
In the preceding section it was shown that if a function/(x)
is piecewise continuous in the interval [ —ji, ji], then the conver
gence of a Fourier sériés at a given point x0 to a value of the
5.10 Certain Sufficient Conditions for the Convergence of a Fourier Sériés 353
function f(x0) dépends on the behaviour of the function in a certain

arbitrarily small neighbourhood [x0 — 6 , x0 -f-ô] with centre at the
point x0.
Let us now prove that if in the neighbourhood of the point x0
the function f(x) is such
that there exist jinite limits
lim ,/(*»+«)-/(*") = ^ ( 1)
a -+-0 a
lim l ^ ± ± z m . = kt
«->+0 « s
( 2)
while the function is con
tinuons at the very point
x0 (Fig. 139), then the Fourier sériés converges at this point to a
corresponding value of the function f (x) *.
Proof. Let us consider the function <D2 (a) defined in the prece-
ding section:
a
C0ST
<1>.(«) = [/(*« + « ) - /( * « ) ] o • a
2 sin -s-
since the function f (x) is piecewise continuous on the interval

—ji, n] and is continuous at the point *0, it is therefore con-
iinuous in some neighbourhood [x0 —ô, *0 -j-ô] of the point x0.
-or this reason, the function <D2 (a) is continuous at ail points where
a#= 0 and |a|^Ç Ô . When a = 0 the function 0 2 (a) is not defined.
Let us find lim ® 2 (a)and lim 0 2 (a), making use of conditions (1)
Œ -> 0 -0 cl —►O + 0
and (2 ):
a
C0ST
lim <D2 (oc) = Hm [/(*„ + «) —/(*<>)]------â
a-►O- 0 a -►0-0 2 sin —
= lim - J — cqs«
—o-o « si„« 2
a
2
= lim lim lim cos-- = kx- 1 .1 = k x
a -> 0 -0 CL—►O — 0 -
g j jj ^ CL—►O — 0
* If conditions (1) and (2) are fulfilled, then we say that the function f (x)
has, at the point x, a dérivative on the right and a dérivative on the left.
Fig. 139 shows a function where /?l = tancp1, fc2 = tan<p2, kl ^=k2y H kl = k2,
that is, if the dérivatives on the right and left are equal, then the function will
be différentiable at the given point.
2 3 -2 08 2
Thus, if we redefine the function 0 2 (<x) by putting <P2 (0 ) = klt

then it will be continuous on the interval [ —ô, 0 ], and, hence,
bounded as well. Similarly we prove that
Iim <D2 (a) = A2
oc—
►O+ 0
Consequently, the function <D2 (a) is bounded and continuous on
the interval [0, 6 ], Thus, on the interval [ — 6 , 6 ] the function
<D2 (a) is bounded and piecewise continuous. Now let us return to
équation (1), Sec. 5.9 (denoting x by x0),
a
C °s y
lim [s„ (x0) —f (x0)] = lim - f [/ (*« + a ) —/ (*„)]----- — sin na da
rz-ôo n -> <
js «J
2 s in y
or
6
lim [s„(*0) —/ (a:0)] = l i m Ç <D2 (a) sin na da
n -* o o n-y<x> J.
— o
From formulas (5) of Sec. 5.7 we conclude that the limit on the
right is equal to zéro, and therefore
liro [s« (*0)— f fo ) ]=
/2->ao
0
or
lim sn(x0) = f(x0)
n-+co
The theorem is proved.
This theorem difFers from the theorem stated in Sec. 5.1 in that
in the latter case it was required, for convergence of the Fourier
sériés at a point x0 to the value of the function f (x0), that the
point x0 should be a point of continuity on the interval [ —n, jt],
whereas the function should be piecewise monotonie; here, however,
it is required that the function at the p e n t x0 should be a point
of continuity and that the conditions ( 1) and (2 ) be fulfilled, while
throughout the interval [ —n, ji] the function should be piecewise
continuous and bounded. It is obvious that these conditions are
different.
Note 1. If a piecewise continuous function is différentiable at
the point x0, it is obvious that conditions ( 1) and (2 ) are fulfilled.
Here, = &2. Hence, at points where the function f (x) is différen
tiable, the Fourier sériés converges to the value of the function
at the corresponding point.
Note 2 . (a) The function considered in Example 2, Sec. 5.2
(Fig. 127), satisfies conditions (1) and (2) at the points 0, ± 2 n,
± 4n, . . . . At ail the other points it is différentiable. Consequently,
a Fourier sériés constructed for it converges at each point to the
value of this function.
5.11 Practical Harmonie Analysis 355
(b) The function considered in Example 4, Sec. 5.2 (Fig. 130);

satisfies conditions ( 1) and (2) at the points ± j t , ± 3ji, ±5rç. It
is differentiable at ail the other points. It is represented by a
Foürier sériés at each point. •
(c) The function considered in Example 1, Sec. 5.2 (Fig. 126),
is discontinuous at the points ± n , ±3ji, ± 5 n. At ail other points
it is différentiable. Hence, at ail points, with the exception of
points of discontinuity, the Fourier sériés corresponding to it con
verges to the value of the function at the corresponding points.
At the discontinuities, the sum of the Fourier sériés is equal to
the arithmetic mean limit of the function on the right and on the
left (in this case, zéro).
5.11 PRACTICAL HARMONIC ANALYSIS

The theory of expanding functions in Fourier sériés is called
harmonie analysis. We shall now make several remarks about
approximate computation of the coefficients of a Fourier sériés,
that is to say, about practical harmonie analysis.
As we know, the Fourier coefficients of a function f(x) with
period 2n are defined by the formulas
Jt Jt
ao= ^ ^ f ( x) dx> a* = - ^ l\i f(x) cos kxdx,

-Jt -Jt
Jt
bh = J / (x) sin kx dx
-Jt
In many practical cases, the function f{x) is represented either

in tabular form (when the functional relation is obtained by
experiment) or in the form of a curve which is plotted by some
kind of instrument. In these cases the Fourier coefficients are
calculated by means of approximate methods of intégration (see
Sec. 1 1 .8 , Vol. I).
Let us consider the interval — of length 2jt. This
can always be done by proper choice of scale on the x-axis.
Divide the interval [ —j i , j i ] into n equal parts by the points
Jt = Xq, X j, X2, ..., X n — Tt
Then the subinterval will be

a =—
Ax 2n
n
We dénoté the values of the function /(x) at the points x0, xlf
x2, . . . , x„ (respectively) by
Ho* Ui* • • • » Un
23 *
These values are determined either from a table or from the graph
of the given function (by measuring the corresponding ordinates).
Then, taking advantage, for example, of the rectangular formula
[see formula ( 1'), Sec. 1 1 .8 , Vol. I], we détermine the Fourier
coefficients:
n n n
a0= Vi> = T 2 & c o s kxi' b» = | 2 Ut sin kxi

Diagrams hâve been devised that simplify computation of Fourier
coefficients. We cannot deal here with the details but we can note
that there are instruments (Fourier analyzers) which permit appro-
ximating the values of Fourier coefficients from the graph of the
function.
5.12 THE FOURIER SERIES IN COMPLEX FORM

Suppose we hâve a Fourier sériés for a periodic function f(x)
with period 2 ji:
00
/ = a„cosnx + sinnx ( 1)
n- 1
We express cos nx and sinnx in terms of exponential functions

[see formulas (3), Sec. 7.5, Vol. I]:
eiy+, -iy ê * — e ~ iy
cos y- sin y- 2i
Thus,
p i n x A _ p —l n x p in x __ e —i n x p in x p —l n x
cos nx = -----—
2
-------,’ sin nx = ----- 2i
^ -------= — i ------^-----
2
Putting these values of cos nx and sin nx into formula ( 1) and
performing appropriate manipulations, we get
t/ \ a0 , V + eln x — e - i n x
f(x)= ~ ï + Hg-------- tb n------- 2-----
n —1
00
_ a0 j ( an ibn çlnx j an ~l~M>n ç - i n x
/t=l '
a0 _ „ a n —i^n __„ an + „
»•
2 2 ~ Ln> 2 ~ L-n (3)
we see that (2 ) becomes
ao
/ (x) = c0 + £ (cneinx + c -ne~inx)
n= 1
5.12 The Fourier Sériés in Complex Form 367
This équation can be written more compactly as

CD
/(*> = Z CneinX * (4)

n= - oo
that is the complex form of the Fourier sériés.

Let us express the coefficients cn and c_„ in terms of intégrais.
Using formulas (4), (5), and (6 ) of Sec. 5.1, we can Write formu
las (3) as follows:
n n -i
°n 2.i ^ f (jc) cos nx dx — t j f ( x ) s mn xd x I
-31 -JT J
31 3t
2ji
J / (*) (cos nx— i sin nx) dx = — ^ f (x) e~lnxdx
Thus
3t
= f(x)e~ in*dx (5')

-a
Similarly
Tt
C-» = Ü
f(x)einxdx (5")
-Jt
The expression for c0 and formulas (5') and (5") can be combi-
ned into a single formula
JT
c»= i $ (rt = 0> ±1> ±2> ± 3 > •••) (6)

-Jt
cn and c_„ are termed the complex Fourier coefficients of the fun-
ction /(x).
If the function f(x) is periodic with period 21, then the Fourier
sériés of /(x) will be
00
/(x) = ^ + £ a nc o s - ^ x + b,,sin-^-x (7)
n —1
[see formula (3), Sec. 5.5].

In this case, clearly, the Fourier sériés in complex form will te
expressed by the formula
00 i 1FL
/(*) = X C„e ‘ (8 )
/!=- 30
instead of formula (4).
The coefficients cn of the sériés are expressed by the formulas

i A t nn
cn = 2ï ) J ( x ) e ~ 1 Xdx (n = 0, ± 1 , ± 2 , . . . ) (9)
The terminology in electrical and radio engineering is as follows:

i njl
the expressions e 1 are called harmonies, the numbers a„ = —
(n = 0 , ± 1, ± 2 , . . . ) are called the wave numbers of the function
00
/( * ) = E v ' v ( 10 )
flS-OD
The set of wave numbers forms what is known as a spectrum.
Plotting these numbers on a number axis, we get a collection of
separate points. This collection of points is called a discrète collec
tion, and the associated spectrum is termed a discrète spectrum.
The coefficients cn defined by formulas (9) are called the complex
amplitude. In some engineering texts, the collection of moduli of
the amplitudes, | c„ |, is also called the spectrum of the function / (x).
5.13 FOURIER INTEGRAL
Let a function f(x) be defined in an infinité interval (—oo, oo)

and absolutely intégrable over it; that is, there exists an intégral
OO
S \f (x)\dx — Q (1)
—00
Further, let the function f (x) be such that it is expandable into
a Fourier sériés in any interval (—/, + /):
oo
f ( x ) =^- + Y i akc o s ^ x + bksm^Y-x (2)
*=1
where
i i
= f{t) cos ^ - t d t ,
bk = ^ f ( t ) s m ^ - t d t (3 )
-i -i
Putting into sériés (2) the expressions of the coefficients ak and bk
from formulas (3), we can write
( m , i
j/ ( 0 * + r 2 ( ^f(t)cos^tdt)cos^x
5.13 Fourier Intégral 359
J / ( 0 sin ^ p / d /^ sin ^ p x
_1_
2/ j/(/)rf/ + - ^ j/(/) [cos-ï / cos^x + sitiT /sinT x] ^
or
l 00 l
/(*) = i j f P + jZ f/(/)c o s Î2 i^ rf/. (4)
-/ *=* -/
Let us investigate what form expansion (4) will take when pas-
sing to the limit as /->oo,
We introduce the following notation:
2 ji kn
«1 = 7 -. «« = — . «*= T . an d A a t = • (5)
f (x) =
j f w dt + TT2 ( j* f ( 0 cosa» (*—*)d/) Aa* (6 )
*=i \_r, /
As /-►oo, the first term on the right approaches zéro. Indeed,
I l 00
-/ -Z -00
For any fixed /, the expression in the parenthèses is a function
of <xA [see formulas (5)], which takes on values from j to oo. We
will show, without proof, that if the function f (x) is piecewise
monotonie on every finite interval, is bounded on an infinité in
terval and satisfies condition ( 1), then as / -* + oo formula (6 )
takes the form
CO / O
O V
/ (*) = —■j y J f (t) cos a (t—x) d t j da (7)
The expression on the right is known as the Fourier intégral of

the function f(x). Equation (7) occurs for ail points where the
function is continuous. At points of discontinuity we hâve the
équation
00 / CD V
- ^ j / (0cos«V — x ) dx = /(*+Q)+/(*-Q)
2
(T )
Let us traiisform the intégral on the right of (7) by expanding

cos a (t—x):
cos a (t—x) = cos a t cos ax -f sin a t sin ax
Putting this expression into formula (7) and taking cosax and
sin exx outside the intégral signs, where the intégration is perfor-
med with respect to the variable t, we get
CD / 06 \
= § f(t )cosatdt jco sax d a

0 '-00 '
CO / OD \
+ J /(/)s in a /d ^ Jsin ax d a (8)
0 N — CD
Each «1 the intégrais in brackets with respect to t exists, since

the fupetion /(/) is absolutely intégrable in the interval (—oo, oo),
and therefore the functions / (t) cos a / and f(t) sin a / are also ab
solutely intégrable.
Let us consider particular cases of formula (8 ).
1. Let /(x) be even. Then f(t) cos a t is an even function, while
/(/) sin at is odd and we hâve
oo
S/(O cos atdt = 2^ f (t) cos at dt
o
CD
J f{t) sin a tdt = 0
Formula (8 ) in this case takes the form

CD s CD v
/(x) = --- § f ( 0 cosat dt Jcosaxda (9)

''O
0 '
2. Let f(x) be odd. Analyzing the character of the intégrais in
formula (8 ) in this case, we obtain
= M / (O s'm atdt J sin axda (10)
If /(x) is defined only in the interval (0, oo), then tor x > 0 it
may be represented by either formula (9) or (10). In the first case
we redefine it in the interval (—oo, 0 ) in even fashion; in the
latter case, in odd fashion.
Let it be noted once again that at the points of discontinuity
one should write the following expression in place of /(x) in the
5.13 Fourier Intégral 361
left-hand members of (9) and (10):

/(* + Q )+ /(* - 0)
2
Let us return to formula (8 ). The intégrais in brackets are func-
tions of a. We introduce the following notation:
CD
A (a) —— j* / (t) cos at dt

—O
D
oo
B(a) = -i- J f (t) sin at dt
— CD
Then formula (8 ) may be rewritten as follows:

oo
f(x) — J [4 (a) cos ax + £ (a) sin ou] doc ( 11 )
o
We say that formula (11) yields an expansion of the function f(x)
into harmonies with a frequency a that varies continuously from
0 to oo. The law of distribution of amplitudes and initial phases
as dépendent upon the frequency oc is expressed in terms of the
functions A(a) and B (a).
Let us return to formula (9). We set
FW=ÿlc]f(t)<XK*tdt ( 12)
o
then formula (9) takes the form
/— *
f(x)= y j F (a) cos ax da (13)
o
The function F (a) is called the Fourier cosine transform of the
function f(x).
If in (12) we consider F (a) as given and f (t) as the unknown
function, then it is an intégral équation of the function f (t). Formula
(13) gives the solution of this équation.
On the basis of formula (10) we can write the following équations:
/— 00
<S>(a) = y J f (t) sin at dt (14)
o
/----*
f (x) = y J (a) sin ax da (15)
o
The function O (a) is called the Fourier sine transform.
Example. Let
f(x)=e~Q* (P > 0, jc =3:0)
From (12) we détermine the Fourier cosine transform:
From (14) we détermine the Fourier sine transform:

_ 00 __
°<a)= Y ~ r ] e~*tsinatdt= Y rai pP2î+ ?a a
0
From formulas (13) and (15) we find the reciprocal relationships:
oc
5.14 THE FOURIER INTEGRAL IN COMPLEX FORM
In the Fourier intégral [formula (7), Sec. 5.12], the brackets

contain an even function of a; hence, it is defined for négative
values of a as well. On the basis of the foregoing, formula (7) can
be rewritten as follows:
(1)
—oo —oo
Let us now consider the following expression, which is identically
equal to zéro:
/ (/) sin a (t—x) dt )da — 0
The expression on the left is identically equal to zéro because

the function of a in the brackets is an odd function, and the in
tégral of an odd function from — M to + A 1 is equal to zéro. It
is obvious that
or
(2)
5.14 The Fourler Intégral in Cômplex Form 363
Note. It is necessary to point to the following. A convergent

intégral with infinité limits is defined as follows:
J q>(a) da = J <p(a) da -f- J <p(a) da

» «0 —oo c
c M
= lim Ç cp (a) da + lim Ç <p(a) da (*)
M-*>oo - JM J
provided that each of the limits to the right exists (see Sec. 11.7,
Vol. I). But in équation (2 ) we wrote
M
[ tp (a) da = lim C <p(a) da (**)
J M-*•ce-M
«-1_
Obviously, it may happen that the limit (**) exists, while the
limits on the right side of équation (*) do not exist. The expres
sion on the right of (**) is called the principal value of the in
tégral. Thus, in équation (2) we consider the principal value of
the improper (outer) intégral. The subséquent intégrais of this sec
tion will be written in this sense.
Let us multiply the terms of (2) by—^ and add them to the
corresponding terms of ( 1); we then get
oo r » “i
/(*) = — J J / (/) (cos a (f —x) — ts in a ( f —x))dt da
or
/(*) - s r i f î n o e - " " - " <#]«*» (3)

— 00 *---- CD J
The right member in formula (3) is called the Fourier intégral

in complex form of the function f(x).
Let us rewrite formula (3) thus:
/(*) = ] ] f { t ) e - i(ltd t ^ e iaxd a (4)
or, compactly
f(x)= JC (a) ela* d a (5)
'•OO
where
00
c («) = •— J f ( t ) e ~ ,a‘ d t (6)
Formula (5) is similar to formula (10) of Sec. 5.12; a is also

called the wave number, but here it assumes ail values from — oo
to + o o and the spectrum of wave numbers is termed a continuous
spectrum. There are other similarities between (5) and (10), Sec.
5.12, whereas in formula (10) of Sec. 5 .1 2 the wave number an
is associated with the complex amplitude cn, in formula (5) the
wave numbers lying in interval (a,, aj + Aa) are associated with
the complex amplitude C(aj). The function C (a) is called the
spectral density or the spectral function. (The term density is
used here in same meaning as in Sec. 2.8, where we discussed the
density of distribution over a two-dimensional domain.)
Equation (4) is usually written as two équations:
QD
F*(a) = r f = j f (i)e~lal dt (7)
—OO
QO
= §F*(a)eiaxda (8 )
—00
The function F* (a) defined by formula (7) is called the Fourier
transform of the function /(*). The function / (x) defined by for
mula (8 ) is called the inverse Fourier transform of the function
F*(a) (the transforms differ in the sign in front of i). The func
tion F*{ql) differs from the function C (a) by the constant factor
__l__
]/"2n '
From the transformations (7) and (8 ) follow the transformations
(12), (14), (13) and (15), Sec. 5.13 (to within the constant fac
tor 1/2). The transformations ( 12 ) and (14) are obtained if we
substitute into (7)
- côs af —j sjn af ' F* (a) - F (a) — /O (a)
and equate the real and imaginary parts. In similar fashion it is
possible to obtain transformations (13) and (15) from the trans
formation (8 ).
Note that in Chapter 7 (“Operational Calculus and Certain of
Its Applications”) we will make use of transformations that are
similar to the Fourier transformations.
5.15 FOURIER SERIES EXPANSION WITH RESPECT

TO AN ORTHOGONAL SYSTEM OF FUNCTIONS
Définition 1. An infinité sequence (system) of functions

<Pi(*). <P2(*). •••• <P0(*), ••• (1)
is said to be orthogonal on the interval [a, b] if for any n ^ k
5.15 Fourier Sériés Expansion 365
the équation
s<f>„ (*)<P* (x)dx = 0 (2)
holds true. It is assumed here that

b
5 [q>« (X )]* d x^h 0
Example 1. The System of functions

1, cos x t sin , cos 2 , sin 2x, . . . , cos nx, sin nxf . . .
jc jc (3)
is orthogonal on the interval [— jt, :i]. This follows from the équations (I) and
(II) of Sec. 5.1.
i 3T . J t n Jt , _ J t rmx ajijc
1, COS - y X t Sin — Xy COS 2-y X t S in 2-y JC, cos - sin - (3')
is orthogonal on the interval [—/, /]. This can readily be verified.
1, cos jc, cos 2xy cos 3jc, ...» cos riXy . . . (4)
is orthogonal on the interval [0, jt].
sin j c . sin 2 j c ......... sin nxt . . . (5)
is orthogonal on the interval [0, jt].
B e lo w w e g iv e oth er S ystem s of orth o g o n a l fu n ctio n s.
Suppose a function f(x), defined on the interval [a, b\, is such
that it can be represented by a sériés relative to the functions of
the orthogonal System ( 1), which converges to the given function
on [a, b]:
/ (*) = 2 c„<p„ (*) (6 )
n —0
We détermine the coefficients cn. Suppose that the sériés obtained
by multiplying the sériés (6 ) by any function (*) admits term-
wise intégration.
We multiply both members of (6 ) by <p* (x) and integrate from
a to b. Taking into account équation (2), we get
b b
J f (*) (x) dx = ck J (x)dx
a a
whence
b
J l(x)<fk (x)dx
= --------------- (7)
l tp| (X) dx
366 Ch. 5 Fourler Seriez
The coefficients ck computed from formulas (7) are called the Fou
rier coefficients of the function f (x) relative to the system of ortho
gonal functions (1). Sériés (6 ) is called the Fourier sériés relative
to the system of functions ( 1).
Définition 2. The orthogonal system of functions
<Pi(*)» <Pî (* ).............................. • • •
is called a complété system if for any quadratically intégrable

function f(x), that is, such that
b
J / 2(x) d x < o o
the following équation holds true:
lim J \f (x) — 2 c m (x) 1 dx = 0 (8 )

n -* * i L i= û J
By virtue of the définitions of Sec. 5.7, équation (8 ) may also

be interpreted as follows. The root-mean-square déviation of the
n
sum 2 c/<P/(je) from the function f (x) tends to zéro as n —*■<».

i=0
If équation (8 ) holds, then we say that the Fourier sériés (6 )
converges to the function f(x) in the mean.
It is quite obvious that convergence in the mean does not imply
convergence at every point of the interval [a, b],
We note without proof that the trigonométrie Systems indicated
in examples 1 to 4 are complété over the appropriate intervals.
Extensive use is made of the system of Bessel functions
Jn J n (^2X) ’ • • •> J n iîX) > • • •(9)
which were considered in Sec. 4.23. Here, A,, Xît . . . , A,-, . . . are
the roots of the Bessel function, that is, numbers such that satisfy
the relation
J n(h) = 0 (t = 1, 2, ...)
We will now point out, without proof, that the system of func
tions
y/~x J n (Atx), x J n (Aax), . . . , )f x J n (A/x), . . . (10)

5.16 The Concept of a Linear Function Space 367
is orthogonal on the interval [0 , 1]:

î
J XJn fy*kX) Jn (kjX) dX = 0 * # (k /) ( 11 )
0
Systems of orthogonal Legendre polynomials also find applications.
They are defined as follows:
* > .( * ) = U 2ln, dn[(Xi~n m ( r t = l , 2, . . . )
They satisfy the équations

(x*— 1) / + 2xy'— n (n + 1) y = 0
O ther S ystem s of o rth o g o n a l p o ly n o m ia ls are a lso u sed .
5.16 THE CONCEPT OF A LINEAR FUNCTION SPACE.

EXPANSION OF FUNCTIONS IN FOURIER
SERIES COMPARED WITH DECOMPOSITION OF VECTORS
In analytic geometry, a vector in three-dimensional space is

defined as
A = Axi + Ayj + A fi
where /, j, k are unit, mutually perpendicular vectors taken along
the coordinate axes. The vectors /, j, k will from now on be de-
noted by eu eit e3.
In similar fashion we can define a vector in n-dimensional space:
A = 2 A fit
A set of vectors of the form A will be called an n-dimensional

Euclidean space and will be symbolized by En. The vectors A will
be called the éléments or points of the n-dimensional Euclidean
space. (It is also possible to consider vectors in an infinite-dimen-
sional space.) Let us examine the properties of the space En.
Suppose we hâve two vectors in the space En:
n n
A = S A fit and B = 2 Bfit
* If for the functions <p* (*), q)y (x) the relation

b
5 P W n (x) «P/ (*) d x = 0 (/ / k)
a
holds, then we say that the functions <p/ (jc) are orthogonal with weight p (x).
Hence, the functions J n (X,*) (for k 5* /) are orthogonal with weight x.
If C1 and C2 are real numbers, then by analogy with three-dimen-

sional space
ClA + C^B ( 1)
is a vector in En.
The scalar product (also called dot product or inner product) of
two vectors A and B is given by the expression
(AB) = £ AiBi (2 )
The vectors elt e2, . . en lie in the space £„ and formula (2 )

holds true.
Hence we get
(e,ey) = 0 (*=*/) (2 ')
and
(eA ) = l (< = /)
Vectors whose scalar product is equal to zéro are called ortho
gonal vectors. Consequently, the vectors elt et, . . . , en are ortho
gonal.
As in the case of three-dimensional space, we can easily es-
tablish the following properties of a scalar product:
(AB) = (BA) \
(A + B, C) — (AC) + (BC) l (3)
( U , B) = %{AB) I
The length or modulas of a vector A is defined as in three-di
mensional space:
\A \ = V {AA ) = ] / " M (4)
The length of the différence between two vectors is defined na-

turally as follows:
\A -B \ =Ÿ ^(A t-B p (5)

In particular,
\A-A\ = \ 2 ( A - A ) ! =0
r i- 1
The angle <p between two vectors is defined as
(AB)
costp = \ A\ - \ B\ (6)
5.16 The Concept of a Linear Function Space 369
Let us consider the collection of ail piecewise monotonie bounded

functions on the interval [a, 6 ]. * We dénoté this collection by <J>
and call it the space of functions <î>. The functions of this space will
be called éléments or points of the space*. It is possible to es-
tablish operations on the functions of the space 0 that are similar
to the operations we performed on the vectors of the space En.
If Cj and C2 are any real numbers and f 1(x), f t {x) are éléments
of the space , then
C M * ) + C J %{x) (7)
is an element of .
If f (x) and qp(x) are two functions in O, then the scalar product
of the functions f (x) and qp (x) is given by the expression
b
(/. y) = [f(x)-<p(x)dx (8 )
a
This expression is similar to expression (2 ). It is easy to verify

that the scalar product (8 ) has properties similar to the properties (3)
for vectors:
(/. <p) = (<p. /) ]
Ui + f*. <P) = (/i. <p) + (/2. <p) f (9)
(Xf, y)-—X(f, (p) )
Like the définition of the modulus of a vector by formula (4),
we hâve the so-called norm of an element f (x) of the space O:
11/ 11= K ( 7 7 7 ) = ] / ' l l f ( x) ] 2dx ( 10 )

a
The distance between éléments f (x) and qp (x) of space O is defined

by the expression
h/ - « pii= ] /" j t f w - v w r d * (H)

a
which is similar to formula (5).

Expression (11), which gives the distance between éléments of
the space, is called the metric of the space. To within the factor
Ÿ b —a, it coïncides with the root-mean-square déviation 6 defined
in Sec. 5.7.
* A class of functions of this kind is considered in the theorem of Sec. 5.1.

We could also consider a broader class of functions for which ail the assertions
of Sec. 5.1 are preserved.
24—2082
Clearly, if /(*) ==(p (x), that is, / ( jc) and q>(jc) coïncide at ail
points of the interval [a, b], then | | / —qp|| = 0. But if | | / —qp|| = 0,
then / (jc) = q>(jc) at ail, except a finite number, points of [a, b].*
But in this case we also say that the éléments of the space O are
identical.
A space of piecewise monotonie bounded functions, in which are
defined the operations (7), (8 ) and the metric is given by équ
ation ( 11 ), is called a linear function space with quadratic metric.
The éléments of the space are called points of the space or vectors.
Now let us consider a sequence of functions
<Pi(*). •••. <P*W. ••• (12)
lying in the space .
The sequence of functions (12) is said to be orthogonal on the
interval [a, b] if for arbitrary i, / (i j) the équations
b
(<Pi. <Py) = S <P, (x) <P/ (x) d x Ô (13)
a
hold true.
On the basis of équations (I), Sec. 5.1, it follows that, for instance,
the System of functions
1, cos x, sin x, cos 2x, sin 2x, cos 3x, sin 3x, . . .
is orthogonal on the interval [ —n, «].
We will now show that expanding a function in a Fourier sériés
of orthogonal functions is similar to the décomposition of a vector
into orthogonal vectors. Suppose we hâve a vector
A = A & + A,e2+ . . . + Akek + . . . + Anen (14)
We assume that the vectors elt e2, . . . , en are orthogonal, that is,
if i /, then
(efij) = 0 (15)
To détermine the projection Ak, we form the scalar products of
each member of (14) by the vector ek. By properties (2) and (3)
we get
{Aek) = Ax (e1ek) + A2(e2ek) + . . . + A k (ekek) + . . . + A n (enek)
Taking account of (15) we obtain
(Aek) = Ak (ekek)
whence
< * = ' ’ 2 .........») 06)
* There can also be an infinité number of points, where / (jc) ^ < p (jc ).
Exercises on Chapter 5 37!
Further assume that the function f(x) has been expanded relative
to a System of orthogonal functions:
/(*) = k=î
S ûa<Pa(*)* (17)
Forming scalar products of each member of (17) by q>*(x) and
taking into account (9) and (13), we get*
( / . <Pa) = M ) __ a_________ (18)
* (f*. <P*) *
] 1<P* M l2 dx
Formula (18) is similar to formula (16).

Now set
n
= 2 akip* (x) (19)
«» = ||/-s» H "(« = l. 2, ...) ( 20)
th en th e S ystem of o rth o g o n a l fu n ctio n s (12) is complété on th e

in te rv a l [a, b\.
The Fourier sériés (17) converges to the function f(x) in themean.
1. Expand the following function in a Fourier sériés in the interval ( — ji, ji):
f(x) — 2x for
f( x ) = x for — ji < x C 0
- —1t- ji-------
Ans. 2 cos x , cos Sx , cos 5* lin 2x
4 ji l2 32 52
sin Sx
...)
2. Taking advantage of the expansion of the function f(x )= 1 in the interval
(0, ji) in thesines of multiple arcs, calculate the sum of the sériés 1— 5- + 4 ™
o 5
—y + ... . AnS. y .
* We présumé that the sériés obtained in the process converge and that
term-by-term intégration is legitimate.
3. Utilizing the expansion of the function / ( x) = x2 in a Fourier sériés,

1 1 1
compute the sum of the sériés -jy —y + ^ p - —-j2-+ • • • • Ans.
12 ‘
4. Expand the function f(x) = y in a Fourier sériés in the interval
cos 2x . cos 3jc cos 4 x
(-J i, ji). Ans. cos x-
22 ' 32 42 r *’ '
5. Expand the following function in a Fourier sériés in the interval (— j i , ji)
/ (x ) = ---- for —j i < x < 0
/W = -2 (n - x ) for 0^ jc < ji
Ans. sin * + y sin 2* + y sin 3 x + . . .

6. Expand in a Fourier sériés, in the interval (— j i , ji), the function
f (x) = —x for — j i < x ^ 0
f (jc) = 0 for 0 < x < ji
_ji__22 y cos (2m + 1) x

Ans.
1)2 - Ij—
( - l )" -1
4 ji
m= 0 '(2m+1 ■' n=l * ' «
7. Expand in a Fourier sériés, in the interval (— jc, ji), the function
f(x )= 1 for — ji < x ^ 0
/ ( jc) = —2 for 0 < x ^ ji
1 6 v sin (2n + \ )x
2 ji 2 *
2 /i+ l
n= 0
8. Expand the function / ( x) = x2t in the interval (0, ji), in a sériés of sines.
Ans. - X ( - D " +1 { + + [ ( - ! ) ” - 1] | sin nx.

1
n= ' '
9. Expand the function y = cos 2* in a sériés of sines in the interval (0, ji).
- 4 f sin x . 3 sin 3jc , 5 sin 5jc , 1
AnS- - n [ 2 ^ + - 2 * = V + - 2 ï = V + - - - \ -
10. Expand the function y = sinjc in a sériés of cosines in the interval (0, ji).
„ 4 f 1 , cos 2 jc , cos 4x , 1
Am- û l T + T i 2 r + - T = 4 r + - J -
11. Expand the function y — ex in a Fourier sériés in the interval (— /, /).
, . nnx
(—l)71 sin —
el — e~l
Ans. f /(ef—e-*)X
21 n=1 /a+ n 2J i2
/ ijijc
(—l) " -1 n sin —j—
+n (<•'-«-*)X l* + n*n*
12. Expand the function f(x) = 2x in a sériés of sines in the interval (0, 1),
a 4 sin nnx
Ans-- z 2 *
n—1
13. Expand the function f ( x ) = x in a sériés of sines in the interval (0, /).
- . nnx
Ans. (-!)■ ♦ » —
n n - 1
' ' n
14. Expand the function
x for 0 < x 1
/«-x for 1 < x < 2
in the interval (0, 2): (a) in a sériés of sines; (b) in a sériés of cosines.
. (2/i+ 1) n*
sm----- T ~
/tns. ( a + 2 £ ( —l)n
n= 0
(2/1+ 1)»
00
... 1 4 cos (2/i + 1) nx
W (2/1+ I)'2
CHAPTER 6
E Q U A T IO N S O F M ATHEM ATICAL PH Y SIC S
6.1 BASIC TYPES OF EQUATIONS OF MATHEMATICAL PHYSICS
The basic équations of mathematical physics (for the case of

functions of two independent variables) are the following second-
order partial differential équations.
1. W av e é q u a tio n :
d * a _ 2 dht
dt%~ a dx* (1)
This équation is invoked in the study of processes of transversal

vibrations of a string, the longitudinal vibrations of rods, electric
oscillations in wires, the torsional oscillations of shafts, oscillations
in gases and so forth. This équation is the simplest of the class of
équations of hyperbolic type.
II. F o u rie r é q u a tio n fo r h e a t c o n d u c tio n :
du „ d2u
---= Q*---- (2)
dt a dx 2
This équation is invoked in the study of processes of the propa

gation of heat, the filtration of liquids and gases in a porous
medium (for example, the filtration of oil and gas in subterranean
sandstones), some problems in probability theory, etc. This équation
is the simplest of the class of équations of parabolic type.
I I I . L a p la c e é q u a tio n :
^ -L ^ - 0 /3\
dx* ^ dy*~ U W
This équation is invoked in the study of problems dealing with

electric and magnetic fields, stationary thermal state, problems in
hydrodynamics, diffusion, and so on. This équation is the simplest
in the class of équations of elliptic type.
In équations (1), (2), and (3), the unknown function u dépends
on two variables. Also considered are appropriate équations of
functions with a larger number of variables. Thus, the wave
équation in three independent variables is of the form
dt * ~ a Vdx2 ‘ d y 2) 0 ')
6.2 Deriving the Equation of the Vibraiing String 375
the heat-conduction équation in three independent variables is of

the form
the Laplace équation in three independent variables has the form

d2u , dhi d2u __
(3')
6.2 DERIVING THE EQUATION OF THE VIBRATING STRING.

FORMULATING THE BOUNDARY-VALUE PROBLEM.
DERIVING EQUATIONS OF ELECTRIC OSCILLATIONS IN WIRES
In mathematical physics a string is understood to be a flexible

and elastic thread. The tensions that arise in a string at any
instant of time are directed along a tangent to its profile. Let a
string of length Z be, at the initial instant, directed along a seg
ment of the x-axis from 0 to /. Assume
that the ends of the string are fixed
at the points x = 0 and x = Z. If the
string is deflected from its original po
sition and then let loose; or if without
deflecting the string we impart to its
points a certain velocity at the initial time, Fig. 140
or if we deflect the string and impart a
velocity to its points, then the points of the string will perform
certain motions; we say that the string is set into vibration. The
problem is to détermine the shape of the string at any instant
of time and to détermine the law of motion of every point of the
string as a function of time.
Let us consider small deflections of the points of the string from
the initial position. We may suppose that the motion of the points
of the string is perpendicular to the x-axis and in a single plane.
On this assumption, the process of vibration of the string is
described by a single function u(x, Z), which yields the amount
that a point of the string with abscissa x has moved at time t
(Fig. 140).
Since we consider small deflections of the string in the xu-
plane, we shall assume that the length of an element of string
i'M 1Â12 is equal to its projection on the x-axis, that is,* vAf,Afa=
* This assumption is équivalent to neglecting u * as compared with 1.

Indeed,
*• _____ *. X,
V 1 + uxdx = § ( l - f - l u i * — . . . J d x « J dx=x2—x1
X, X, X,
376 Ch. 6 Equations of Mathematical Physics
= xt —xt. We also assume that the tension of the string at ail

points is the same; we dénoté it by T.
Consider an element of the string MM' (Fig. 141). Forces T
act at the ends of this element along tangents to the string. Let
the tangents form with the x-axis
angles qp and qp+ Aqp. Then the
projection on the u-axis of forces
acting on the element MM' vvill
be equal to T sin(qp +Aqp) — T sincp.
Since the angle (p is small, we can
put tancp » sinqp, andwewill hâve
7’ sin(cp + Aq>)— T sinqp
» T tan <» + A r ) - T tan <p- T [**+ »«■ 0 _ * g 0 ]
_ rr d2u (x + Q\x, t) \ ___ rj, d2u (*, t)
“ 1 dx2 ^ 1 dx2
Ax
0 < 0 < 1
(here, we applied the Lagrange theorem to the expression in the

square brackets).
In order to obtain the équation of motion, we must equate to
the force of inertia the external forces applied to the element.
Let p be the linear density of the string. Then the mass of an
element of string will be pAx. The accélération of the element
is . Hence, by d’Alembert’s principle we will hâve
A d2u rr d2u A
pA x-dF = T W*Ax
T
Cancelling out Ax and denoting — = a2, we get the équation of
motion:
d2u _ 2 d2u
~ d F ~ a di? (1)
This is the wave équation, the équation of the vibrating string.

Equation (1) by itself is not sufficient for a complété définition
of the motion of a string. The desired function u (x, t) must also
satisfy boundary conditions that indicate what occurs at the ends
of the string (x — 0 and x = l ) and initial conditions, which describe
the state of the string àt the initial time (t — 0). The boundary
and initial conditions are referred to collectively as boundary-value
conditions.
For example, as we assumed, let the ends of the string at x = 0
and x = l be fixed. Then for any t the following équations must
6.2 Deriving the Equation of the Vibraiing String 377
hold:
u (0 , 0 = 0 (2)
u(l, o = o • ( 2 ')
These équations are the boundary conditions for our problem.

At / = 0 the string has a definite shape, that which we gave it.
Let this shape be defined by a function f(x). We should then
hâve
u(x, 0) = m|<=0 = /(*) (3)
Furthermore, at the initial instant the velocity at each point
of the string must be given; it is defined by the function <p(.c).
Thus, we should hâve
du
W t =0 = <p(*) (3 ')
The conditions (3) and (3') are the initial conditions.
Note. For a spécial case we may hâve /(*) = 0 or <j>(x)==0.
But if f(x) = 0 and <p(x) = 0 , then the string will be in a state
of rest; hence, u(x, t) = 0 .
As has already been pointed out, the problem of electric oscil
lations in wires likewise leads to équation (1). Let us show this
to be the case. The electric current in a wire is characterized by
the quantity i (x, t) and the voltage v(x,t) which are dépendent
on the coordinate x of a point of the wire and on the time t.
Regarding an element of wire Ax, we can write that the voltage
drop on the element Ax is equal to v(x, t)—i>(x + Ax, t) «
« —gÂx. This voltage drop consists of the résistance drop, equal
to iRAx, and the inductive drop, equal to -^-LAx. Thus,
—^ Ax = iRAx-\--jj-LAx (4)
where R and L are the résistance and the inductance reckoned per
unit length of wire. The minus sign indicates that the current
flow is in a direction opposite to the build-up of v. Cancelling
out Ax, we get the équation
1 + iR + l J - O (5)
Further, the différence between the current leaving element Ax
and entering it during time At will be
[i (x, t) — i (x -f- Ax, /)] At « — Ax At
It is taken up in charging the element (this is equal i o C A x ^ A t }
and in leakage through the latéral surface of the wire due to

imperfect insulation, equal to AvAxAt (here A is the leakage
coefficient). Equating these expressions and cancelling out AxAt,
we get the équation
¥ + C¥ + ^ = °
Equations (5) and (6 ) are generally called telegraph équations.
From the System of équations (5) and (6 ) we can obtain an
équation that contains only the desired function i (x , t), and an
équation containing only the desired function v(x, t). Differentiate
the terms of équation (6 ) with respect to x; differentiate the terms
of (5) with respect to t and multiply them by C. Subtracting,
we get
& + A % - c * ê - CL& ~ °
Substituting into this équation the expression ^ from (5), we get
* + A ( - tR - L § ) - C R £ - C L % t ~0
or
* * = C L ^ + (CR + A L ) £ + ARi (7)
Similarly, we obtain an équation for determining v(x, t):
^ = C L ^ + (CR + A L ) ^ - + A R v (8)
If we neglect the leakage through the insulation (/4 = 0) and

the résistance (R = 0), then équations (7) and (8 ) pass into the
wave équations
, d*v d*v
a dx* ~ dt* ’ a dxi ~ d tî
wherea 2 = ^ - . The physical conditions dictate the formulation of

the boundary and initial conditions of the problem.
6.3 SOLUTION OF THE EQUATION OF THE VIBRATING

STRING BY THE METHOD OF SEPARATION OF VARIABLES
(THE FOURIER METHOD)
The method of séparation of variables (or the Fourier method),

which we shall now discuss, is typical of the solution of many
problems in mathematical physics. Let it be required to find the
solution of the équation
dû _ g d2u
~ d F ~ Ü dx* (1)
6.3 Solution of the Equation of the Vibrating String 379
which satisfies the boundary-value conditions

« ( 0 , /) = 0 (2 )
u (Z, Z) = 0 • (3)
u (x , 0 ) = / (x) (4)
(5)
We shall seek a particular solution (not identically equal to zéro)
of équation (1) that satisfies the boundary conditions (2) and (3),
in the form of a product of two functions X (x) and T (t), of
which the former is dépendent only on x, and the latter, only
on t :
u (x, t) = X (x) T (t) (6 )
Substituting into équation (1), we get X (x) T" (t) = a2X" (x) T (t),
and dividing the terms of the équation by a2XT,
The left member of this équation is a function that does not

dépend on x, the right member is a function that does not dépend
on t. Equation (7) is possible only when the left and right mem-
bers are not dépendent either on x or on /, that is, are equal to
a constant number. We dénoté it by —A,, where X > 0 (later on
we will consider the case X < 0). Thus,
From these équations we get two équations:

X u + XX = 0 (8 )
T" + a2XT = 0 (9)
The general solutions of these équations are (see Sec. 1.21):
X (x) = A c o s K ^ + ôsin V Xx (10)
T (t) = C cos aV X t + D sinaVXt ( 11)
where A, B, C and D are arbitrary constants.
Substituting the expressions X (x) and T (t) into (6 ), we get
u (x , t) = (/4cosl/Xx: + Æsin V^Xx) (Ccos aVXt + Dsin aV X t)
Now choose the constants A and B so that the conditions (2) and
(3) are satisfied. Since T ( t ) ^ 0 [otherwise we would hâve u (*, /)= 0 ,
which contradicts the hypothesis], the function X(x) must satisfy
the conditions (2) and (3); that is, we must hâve X (0) = 0, X (Z) = 0.
Putting the values * = 0 and x = l into (10), we obtain, on the
basis of (2) and (3),

0 = y4-l+ B -0
0 = A cos Y k l + B sin Y M
From the first équation we find A = 0. From the second it follows
that
5 s i n |A / = 0
B=£ 0, since otherwise we would hâve X = 0 and u = 0, which
contradicts the hypothesis. Consequently, we must hâve
sin |/ kl = 0
whence
Yk. = ™ (rt= l, 2, ...) ( 12 )
(we do not take the value n = 0, since then we would hâve X = 0
and u = 0 ). And so we hâve
X = B sïn — x (13)
These values of k are called eigenvalues of the given boundary-
value problem. The functions X (x) corresponding to them are
called eigenfunctions.
Note. If in place of —k we took the expression +X = £2, then
équation (8 ) would take the form
X"— k*X = 0
The general solution of this équation is
X = Aekx + Be~kx
A nonzero solution in this form cannot satisfy the boundary con
ditions (2) and (3).
Knowing Y k we can [utilizing ( 11 )] write
T(t) = C c o s ^ t + D s i n ™ t ( n = l,2 , ...) (14)
For each value of n, hence for every k, we put the expressions
(13) and (14) into (6 ) and obtain a solution of équation ( 1) that
satisfies the boundary conditions (2 ) and (3). We dénoté this solu
tion by un (x , t):
i jv • fl JT ( C tflJX » - |- v • Clf l T l * \ / •% ^ \
un (x, t) = sm -j-x lC ncos — t + D„sm — t) (15)
For each value of n we can take the constants C and D and thus
write C„ and Dn (the constant B is included in C„ and D„). Since
équation ( 1 ) is linear and homogeneous, the sum of the solutions
6.3 Solution of the Equation of the Vibrâting String 381
is also a solution, and therefore the function represented by the

sériés
u(x, 0 = 2 u„(x*t)
u(x, 0 = 51 ^Crtc o s ^ / + D „ s i n —^-t^sin — x (16)
will likewise be a solution of the differential équation ( 1), which

will satisfy the boundary conditions (2 ) and (3). Sériés (16) will
obviously be a solution of équation (1) only if the coefficients C„
and Dn are such that this sériés converges and that the sériés
resulting from a double term-by-term différentiation with respect
to x and to t converge as well.
The solution (16) should also satisfy the initial conditions (4)
and (5). We shall try to do this by choosing the constants Cn
and Dn. Substituting into (16) f = 0, we get [see condition (4)]:
00
/(*) = £ C „ s in ^ * (17)
n—1
If the function f (x) is such that in the interval (0, /) it may be
expanded in a Fourier sériés (see Sec. 5.1), the condition (171 will
be fulfilled if we put
i
C„ = l § f ( x ) s i n ^ x d x (18)
6
We then differentiate the terms of (16) with respect to t and
substitute t = 0. From condition (5) we get the équation
00
r\ anit njt
=
- D« — Sin — x
2
n—1
We define the Fourier coefficients of this sériés:
t
~ ami 2 f . v . nn .
Dn - r = T \j y (x )s m — xdx
D" = ï ï n S v W sînT xdx <19)

o
Thus, we hâve proved that the sériés (16), where the coefficients
Cn and Dn are defined by formulas (18) and (19) [if it admits
double termwise différentiation], is a function u (x ,t), which is
tlio solution of équation ( 1) and satisfies the boundary and initial

conditions (2) to (5).
Note. Solving the problem at hand for the wave équation by
a different method, we can prove that the sériés (16) is a solution
cven when it does not admit termwise différentiation. In this case
Ihe function f(x) must be twice différentiable and q>(je) must be
once différentiable.*
6.4 THE EQUATION Of HEAT CONDUCTION IN A ROD.

FORMULATION OF THE BOUNDARY-VALUE PROBLEM
Let us consider a homogeneous rod of length /. We assume

that the latéral surface of the rod is impénétrable to heat transfer
and that the température is the same at ail points of any cross-
sectional area of the rod. Let us study
i i i | the process of propagation of heat in the
0 x t xz i rod.
We place the x-axis so that one end
Fig. 142 of the rod coincides with the point
x = 0 , the other with the point x = l
(Fig. 142). Let u (x ,t) be the température in the cross section
of the rod with abscissa x at time t. Experiment tells us that the
rate of propagation of heat (that is, the quantity of heat passing
through a cross section with abscissa x in unit time) is given by
the formula
= (1)
where S is the cross-sectional area of the rod and k is the coefficient

of thermal conductivity.**
Let us examine an element of rod contained between cross
sections with abscissas xt and x2 (x2— xt = Ax). The quantity of
heat passing through the cross section with abscissa xl during
time A t will be equal to
du
AQl = —k dx X —Xt
SAt (2)
* These conditions are dealt with in detail in Equations of Mathematical

Physics by A. N. Tikhonov and A. A. Samarsky, Gostekhizdat, 1954 (in
Russian).
** The rate of heat transfer, or the rate of heat flux, is determined by
where AQ is the quantity of heat that has passed through a cross section S
during a time At.
6.4 The Equation of Heat Conduction in a Rod 383
and the same for the cross section with abscissa x9

du
AQ2= —k ~dx SAt (3)
x=xg
The influx of heat AQX—AQt into the rod element during time
At will be
[-*& L H - H
kÂ xSA t (4)
(we applied the Lagrange theorem to the différence —

\ x OX x -x t
du
dxX= Xt J
). This influx of heat during time At was spent in
raising the température of the rod element by Au:
AQj — AQ2 = cpAxSAu
or
AQX—AQ2 » cpAx5 Ai (5)
where c is the heat capacity of the substance of the rod and p

is the density of the substance (pAxS is the mass of an element
of rod).
Equating expressions (4) and (5) of one and the same quantity
of heat AQ!—AQs, we get
k AxSAt — cpAxS -gj- At
or
d u __ k_ dî u
dt cp dx2
Denoting Cp
= aa, we finally get
du « d2u
~ d t~ a ü ? (6)
This is the équation for the propagation of heat (the équation of

heat conduction) in a homogeneous rod.
For the solution of équation (6 ) to be definite, the function u (x, t)
must satisfy the boundary-value conditions corresponding to the
physical conditions of the problem. For the solution of équation (6 ),
the boundary-value conditions may differ. The conditions which
correspond to the so-called first boundary-value problem for 0 ^ t ^ T
are as follows:
u(x, 0 ) = (p (x) (7)
u (0 , 0 = f i ( 0 ( 8)
u(l, t) = ^ ( t ) (9)
Physically, condition (7) (the initial condition) corresponds to

the tact that for / = 0 a température is given in various cross
sections of the rod equal to <p(jt). Conditions (8 ) and (9) (the
boundary conditions) correspond to the tact that at the ends of
the rod, x — 0 and x — l, a température is maintained equal to
i|), (t) and ^ t (t), respectively.
It is proved that the équation (6 ) has only one solution in the
région which satisfies the conditions (7 ), (8 ),
and (9).
6.5 HEAT TRANSPER IN SPACE
Let us further consider the process of heat transfer in three-di-

mensional space. Let u(x, y, z, t) be the température at a point
with coordinates (x, y, z) at time t. Experiment states that the
rate of heat passage through an area As, that is, the amount of
heat passing through in unit time is governed by the formula [si-
milar to formula ( 1) of the preceding section]
AQ = - * ^ A s (1)
where k is the coefficient of thermal conductivity of the medium
under considération, which we regard as homogeneous and isotro
pie, n is the unit vector directed normally to the area As in the
direction of flow of the heat. Taking advantage of Sec. 8.14, Vol. I,
we can Write
du du , du n , du
_ = _ cosa + _ cosP + _ c o s Y
where cosa, cosp, cos y are the direction cosines of the vector n, or
du j
■^ = »gradn
Substituting the expression into formula (1), we get

AQ = — ktt grad u As
The amount of heat flowing in time At through the elementary
area As will be
AQ At = — kn grad u At As
Now let us return to the problem posed at the beginning of
the section. In the medium at hand we pick out a small volume V
bounded by the surface S. The amount of heat flowing through
the surface S will be
Q = — At ^ J kn grad u ds
(2 )
6.5 Heat T rans fer in Space 385
where n is the unit vector directed along the extemal normal to

the surface S. It is obvious that formula (2) yields the amount of
heat entering the volume V (or leaving the volume V) during
time At. The amount of heat entering.V' is spent in raising the
température of the substance of this volume.
Let us consider an elementary volume Au. Let its température
rise by Au in time At. Obviously, the amount of heat expended
on raising the température of the element Au will be
cAvp Au » cAvp At
where c is the heat capacity of the substance and p is the density.
The total amount of heat consumed in raising the température in
the volume V during time At will be
But this is the heat that has entered the volume V during the
time At\ it is defined by formula (2). Thus, we hâve the équation
— At kn grad uds = At c p dv
Cancelling out At, we get

—JÇ kn grad u d s= [ § § CP (3)
The surface intégral on the left-hand side of this équation we

transform by the Ostrogradsky formula (see Sec. 3.8), assuming
/ r = A:grad u:
^ J (k grad u) n ds = J J J div (k grad u) dv
s vJ
Replacing the double intégral on the left of (3) by a triple inté
gral, we get
div (k grad u)dv = c p ^-d v
or
I I I [div ^ grad + Cp W"] dv = 0(4)
Applying the mean-value theorem to the triple intégral on the left
(see Sec. 2.12), we get
[div (6 grad u) + c p |r" t- lj x = x „ u = y „
L z=z,
=0 (5)
where the point P (xlt ylt zx) is some point in the volume V.
2 5 — 2082
Since we can pick out an arbitrary volume V in three-dimen-

sional space where heat is being transferred, and since we assume
that the integrand in (4) is continuous, (5) will hold true at each
point of the space. Thus,
cp-^r = — div (Ægrad u) (6 )
But
kgradu = k ^ i + k ^ j + k ^ k
and (see Sec. 3.8)
div(*gradW) = - ( * — ) + | ( Æ | - ) + | ( ^ )
Substituting into (6 ), we obtain
If k is a constant, then
div (k grad u) = k div (grad u) - k
and équation (6 ) then yields
_ co — - k (V—
P àt
4- — -4- — \
àx* ^ dy* ^ dz2 J
or, putting — — — a2,

du 2 ( d‘2u . d2u d*u \ /Qv
w = a ( i >f> + w + m ) (8)
Equation (8 ) is compactly written as
ÔU 2 A
âT = û Au
where =^ + is the Laplacian operator. Equation (8 )
is the équation of heat conduction in space. To find its unique
solution that corresponds to the problem posed here, *it is necessary
to specify the boundary-value conditions.
Let there be a body £2 with a surface a. Iri this body we con-
sider the process of heat transfer. At the initial time the tempe-
rature of the body is specified, which means that the solution is
known for / = 0 (the initial condition):
u (x , y, z, 0) = qp (x, y y z) (9)
In addition to that we must know the température at any point M
of the surface a of the body at any time t (the boundary condi-
6.6 Solution of the First Boundary-Value Problem 387
tiùn):
u (M, t) = \p(M, t) (10)
(Other boundary conditions are possible t®o.)
If the desired function u (x , y, z, t) is independent of z, which
corresponds to the température being independent of z, we obtain
the équation
which is the équation of heat transfer in a plane.

If we consider heat transfer in a flat domain D with boundary C,
then the boundary-value conditions, like (9) and (10), are formu-
lated as follows:
u(x, y, 0 ) = <p(x, y)
u (M, t) = ip(M, t)
where <p and \|) are specified functions and M is a point on the
boundary C.
But if the function u does not dépend either on z or on y, then
we get the équation
du » d2u
dt a dx2
which is the équation of heat transfer in a rod.
6.6 SOLUTION OF THE FIRST BOUNDARY-VALUE

PROBLEM FOR THE HEAT-CONDUCTION EQUATION
BY THE METHOD OF FINITE DIFFERENCES
As in the case of ordinary differential équations, when we solve
partial differential équations by the method of finite différences,
the dérivatives are replaced by appropriate différences (see Fig. 143);
du (x, t) _u (x-\-h, t) — u(x, t)
dx h ( 1)
â2u (x , t) 1
■1 ['f u (jc + ft, t) — u (x, t) u (x, t) — u (x—h, <))
dx2 h L
or
d2u(x, t ) _u(x-j-h, t) — 2u(x, t)-\-u (x — h, t)
di? ^ h2 (2)
similarly,
du (x, t) _ u (x, t + l) — u(x, t)
dt ~ / (3)
The first boundary-value problem for the heat-conduction équation

is stated (see Sec. 6.4) as follows. It is required to find the solution
25
of the équation
du . d*u
-d t= a S * (4)
that satisfies the boundary-value conditions
u (x , 0) = <p(;r), (5)
u (0, 0 = f i ( 0 . o < * < r ( 6)
u (i, /)= ♦ ,(/). o < / < r (7)
that is, we hâve to find the solution u (x, t) in a rectangle bound-
ed by the straight Unes / = 0, x = 0, x = L, t — T, if the values
df the desired function are giv- .
en on thfee' of its sides: / = 0 , £i|.
J
î
1
\(x,t+0 JM î
i
(H.M) ajt>(MM
(x-h,t) (x,t) (x+h,t) 7
I
L x
F ig . 143 Fig. 144
x = 0 ,x = L (Fig. 144). We cover our région with a grid formed by the

straight lines
x = ih, i = 1, 2 , . . .
t = kl, k = 1, 2 , . . .
and approximate the values at the lattice points of the grid, that
is, at the points of intersection of these lines. Introducing the no
tation u (ih, kt) = u/, *, we write [in place of équation (4)] a cor-
résponding différence équation for the point (ih, kl). In accord
with (3) and (2), we get
Ui,k + l --- <H,k . >*l+t, 2«{ * + « / _ ! fe
------ ;------------- p ----------- <8>
We détermine u,-, *+1:
( l —; + a* tJt («/+i.* + (9)
From (9) it follows that if we know three values in the kth row:
“i+i, *> we can détermine the value * +1 in the
(&4 -l)th row. We know ail the values on the straight line / = 0
[see formula (5)]. By formula (9) détermine the values at ail the
interior points of the segment / = /. We know the values at the
6.7 Heat Transfer in an Unbounded Rod 389
end points of this segment by virtue of (6 ) and (7). In this way,

row by row, we détermine the values of the desired solution at ail
lattice points of the grid.
It may be proved that from formula (9) we can obtain an ap-
proximate value of the solution not for an arbitrary relationship
h?
between the steps h and /, but only for / ^ . Formula (9) is
greatly simplified if the step length l along the /-axis is chosen so
that
or
In this case, (9) takes the form

“ /.* +! = y (“/ +!,* + ( 10 )
This formula is particularly convenient for computations (Fig. 145).
This method gives the solution at the lattice
points of the grid. Solutions between these
points may be obtained, for example, by ext: (<,k+1)
rapolation, by drawing a plane through every
three points in the space (x , t, u). Let us de-
W (i+1,k)
note by uh (x , t) a solution obtained by formula
(10) and this extrapolation. It can be proved that Fig. 145
Iim uh (x, t) = u (x , t)
h-*0
where u(x, t) is the solution of our problem. It can also be proved*
that
\uk (x, t)— u{x, /)| < Mh*
where M is a constant independent of h.
6.7 HEAT TRA N SFER IN AN UNBOUNDED ROD

Let the température be given at various sections of an unbound
ed rod at an initial instant of time. It is required to détermine
the température distribution in the rod at subséquent instants cf
time. (Physical problems reduce to that of heat transfer in an un
bounded rod when the rod is so long that the température in the
* This question is dealt with in more detail in D. Yu. Panov’s Handbook
of Numerical Solutions of Partial Differential Equations, Qostekhizdat, 1951;
Lothar Collatz, Numerische Behandlung von Differential g luchungen, 2nd édi
tion, Springer, 1955.
390 Ch. 6 Equations of MaihemcUical Physics
interior points of the rod at the instants of time under considéra

tion is but slightly dépendent on the conditions at the ends
of the rod.)
If the rod coïncides with the x-axis, the problem is stated ma-
thematically as follows. Find the solution to the équation
du , d2u
i r = a -w (i)
in the région — o o < x < o o , 0 < t, which satisfies the initial

condition
u(x, 0 ) = <p (x) (2 )
To find the solution, we apply the method of séparation of
variables (see Sec. 6.3); that is, we shall seek a particular solution
of équation ( 1) in the form of a product of two functions:
u (x, t) = X (x )T (t) (3)
Putting this into équation ( 1) we hâve X (x) T' (t) = cPX" (x) T (t)
or
T ' _ X"
a2T ~ X (4)
Neither of these relations can be dépendent either on x or on t ;
therefore, we equate them to a constant,*—k2. From (4) we get
two équations:
T'+a*k*T = 0 (5)
X'' + k2X = 0 (6 )
Solving them we find
T = Ce-a,*-‘‘
X = A cos kx + B sin kx
Substituting into (3), we obtain
u\(x, t) = e~a‘K,‘ [i4(X)cosA,x + 5(X)sinA,x] (7 )
[the constant C is included in A (X) and in 5(X)].
For each value of k we obtain a solution of the form (7 ). For
each value of k the arbitrary constants A and B hâve definite
values. We can therefore consider A and B functions of k. The sum
of the solutions of form (7) is likewise a solution [since équation (1)
is linear]:
S e-*'»1[A (k) cos k x + B (k) sin kx]
* Since from the meaning of the problem T (t) must be bounded for any t
T1
!f q>(*) is bounded, it follows that y must be négative. And so we Write—A,2.
6.7 Heat Transfer in an Unboundéd Rod 391
Integrating expression (7) with respect to the parameter A between 0

and oo, we also get a solution
co •
u(x, t) — J r a'w [A (A) cosAx + fl (A) sin AxjdA (8 )
o
if A (A) and B (A) are such that this intégral, its dérivative with
respect to t and the second dérivative with respect to x exist and
are obtained by différentiation of the intégral with respect to t
and x. We choose A (A) and B (A) such that the solution u (x, t)
satisfies the condition (2). Putting t = 0 in (8 ), we get [on the
basis of condition (2 )]:
ao
u (x, 0) = <p(x) = ^ [ A (A) cos Ax + B (A) sin Ax] dA (9)
o
Suppose that the function <p (x) is such that it may be represented
by the Fourier intégral (see Sec. 5.13):
<p(x) = -^- U ^ <p (a) cos A(a —x)dajdA

'o \ - » /
00r CO
<p(x) = -i-J Ç J <p(a)cosAad a '\cosAx
+ <p(a) sin Aa da^ sin Axj dA (10)
Comparing the right sides of (9) and ( 10 ), we get

oo
A (A) = ^ <p(a) cos Aa da
—oo
ao ( 11)
B (A) = — <p(a) sin Aa da
Putting the expressions thus found of A (A) and B (A) into (8 ), we

obtain
co r «
u(x, = ^ J <p(a)cosAada cosAx
/ +«
+ ( j «P(a) sin Aada dA
\ - 00
90
J <p(a) (cos Xa cos Xx + sin Xa sin Xx) da dX

-T Î'
— 00
00 ®
= Ç (p(a)cosX(a—x)dajdX
or, interchanging the order of intégration, we finally get

» r /*
«(x, /)l = -^- ]\ |[*<“
= TT q>(et)(
> ($\ 'e-°,x,'cosX(a—a:)jdct (12)
-ao L \o
This is the solution of the problem.
Let us transform formula (12). Compute the intégral in the
parenthèses:
00 ao
j* e~a,KUcos X (a —x) dX — —— J e~z' cos pz dz (13)
The intégral 1s transformed by substitution:
aX Y t = z. (14)
a Y~t
We dénoté
G
O
/C (P) = $e"z’ cos pzdz (15)
o
Differentiating,* we get
oo
K' (P) = — $ z sin pz dz
o
Intégrât ing by parts, we find
oo
K' (p) = y [e-** sin N " — ^ e~**cos Pzdz
*'<P )=~ |*< P)
Integrating this differential équation, we obtain
/C (p) = Ce * . (16)
Différentiation here is easily justified.

6.7 Heat T ransfer in an Unbounded Rod 393
Détermine the constant C. From (15) it follows that

oo
AT(0 ) = j V **dz-
(see Sec. 2.5). Hence, in (16) we must hâve

„ vu
And so
*(P) = 2 e (17)
Put the value (17) of the intégral (15) into (13):
®. - p*
j e-o*x,»/CosX(a—x)dX = — -^Ç—e 4
In place of (J we substitute its expression (14) and finally get the

value of the intégral (13):
r i r ir
J e -0,^ c o s À (a —x)d\ = ± } / j e *** (18)
o
Putting t’his expression of the intégral into the solution (12), we
finally get
P («-*)»
“ (x, t) = y = j (p(a)e da (19)
— CD
This formula, called the Poisson intégral, is the solution to the

problem of heat transfer in an unbounded rod.
Note. It may be proved that the function u(x, t), defined by
intégral (19), is a solution of équation ( 1) and satisfies condition
(2 ) if the function <p(je) is bounded on an infinité interval (—oo, oo).
Let us establish the physical meaning of formula (19). We con-
sider the function
' 0 for — oo < x < x0
<P*W = <p(x) for x0 ^ -f- Ajc ( 20)
0 for x0 + Ax < x < oo
Then the function
î
U*(x, 0 = “ ■< da ( 21)
2a Y"n t
—O
O
is the solution to équation ( 1), which solution takes on the value

<p*(x) when / = 0 . Taking (20) into considération, we can Write
X0 + &X
«*-*)»
u*(x,
‘)=- j n r s - I » (a)e *a,‘ da
Applying the mean-value theorem to the latter intégral, we get
u*(x, t) - 6 l°'‘ ' *o < 5 < *o + A* (2 2 )

Formula (22) gives the value of température at a point in the rod
at any time if for t = 0 the température in the rod is everywhere
u* = 0, with the exception of the interval [x0, x0+ Ajc] where it is
<p(x). The sum of températures of form (22) is what yields the
solution of (19). It will be noted that if p is the linear density
of the rod, c the heat capacity of the material, then the quantity
of heat in the element [x0, x0 + Ajc] for *= 0 will be
AQ »<p (£) Axpc (23)
Let us now consider the function
i (t-*)*
1 _ e~ 4au (24)
2a y n i
Comparing it with the right side of (22) and taking into account
(23), we may say that it yields the température at any point of
the rod at any instant of time t if for t = 0 there was an instan-
taneous heat source with amount of heat Q —cp in the cross section |
(the limiting case as Ax~—*-0).
6.8 PROBLEMS THAT REDUCE TO 1NVESTIOATING

SOLUTIONS OF THE LAPLACE EQUATION.
STATINO BOUNDARY-VALUE PROBLEMS
In this section we shall consider certain problems that reduce to

the solution of the Laplace équation:
d*u | d2u | d2u „
dx* dyî dz2 (1)
As already pointed out, the left side of équation ( 1) is symbolized by

d2u . à2u d2u
dx2 ^ dy2 dz2
: Au
where A is called the Laplacian operator. The functions u which

satisfy the Laplace équation are called harmonie functions.
6.8 Stating Boundary-Value Problems 395
I. A stationary (steady-state) distribution of température in a

homogeneous body. Let there be a homogeneous body Q bounded
by a surface a. In Sec. 6.5 it was showji that the température at
various points of the body satisfies the équation
du 2 / d2u . d2u . d2u \
ü t ~ a V dx* + ~dÿ• + dz* )
If the process is steady-state, that is, if the température is not

dépendent on the time, but only on the coordinates of the
points of the body, then -|~ = 0 and, consequently, the tempéra
ture satisfies the Laplace équation
d*u d*u d*u _ n /n
dx* + dy* + dz2 — 0 l 1'
To détermine the température in the body uniquely from this équa

tion, one has to know the température on the surface a. Thus, for
équation ( 1), the boundary-value problem is formulated as follows.
Find the function u(x, y, z) that satisfies équation (1) inside
the volume Q and that takes on specified values at each point M
of the surface o:
«|o = *(M ) (2 )
This problem is called the Dirichlet problem or the first boundary-
value problem of équation ( 1).
If the température on the surface of the body is not known, but
the heat flux at every point of the surface is, which is proportional
to ^ (see Sec. 6.5), then in place of the boundary-value condi
tion (2 ) on the surface a we will hâve the condition
du
dn
= l|5* (M) (3)
The problem of finding the solution to ( 1) that satisfies the boun

dary-value condition (3) is called the Neumann problem or the se
cond boundary-value problem.
If we consider the température distribution in a two-dimensional
domain D bounded by a contour C, then the function u will dé
pend on two variables x and y and will satisfy the équation
d*u , d*u n
(4)
-§ * + w = 0
which is called the Laplace équation in a plane. The boundary-
value conditions (2) and (3) must be fulfilled on the contour C.
II. The potential flow of a fluid. Equation of continuity. Let there
be a flow of liquid inside a volume Q bounded by a surface a
(in a particular case, Q may also be unbounded). Let p be the

density of the liquid. We dénoté the velocity of the liquid by
v = vxi + vyj + v zk (5)
where vx, vy, vz are the projections of the vector v on the coordi-
nate axes. In the body Q pick out a small volume ©, bounded
by the surface S. The following quantity of liquid will pass through
each element As of the surface S in a time At:
AQ=ç>vn As At
where n is the unit vector directed along the outer normal to the
surface S. The total amount of liquid Q entering the volume ©
(or flowing out of the volume ©) is expressed by the intégral
Q = A t\jp v n d s (6)
(see Secs. 3.5 and 3.6). The amount of liquid in the volume ©
at time t was
(ù
During time At the amount of liquid will change (due to changes
in density) by the amount
Q= jjjA p d » * 4 ( (7)
0 ©
Assuming that there are no sources in the volume <o, we con-
clude that this change is brought about by an influx of liquid to
an amount that is determined by équation (6 ). Equating the right
sides of (6 ) and (7) and cancelling out At, we get
(8)
S 0
We transform the iterated intégral on the left by Ostrogradsky’s
formula (Sec. 3.8). Then (8 ) will assume the form
0 0
or
I W ( - t ~ div(pv)) d£ù=0
0
Since the volume © is arbitrary and the integrand is continuous,
we obtain
div(p®) = 0 (9)
6.8 Stating Bûundary-Vatue Problems 397
or
aT- T x (P°*) •
~ T z^ = 0(Q')
This is the équation of continuity of a compressible liquid.
Note. In certain problems, for instance when consitfering the
movement of oil or gas in a subterranean porous medium to awell,
it may be taken that
®= — -j^-grad p
where p is the pressure and k is the coefficient of permeability and

dp K ËL
dt dt
A,= const. Substituting into the équation of continuity (9), we get
% + div (k grad p) = 0
or
If k is a constant, then this équation takes on the form

dp _ k f à2p . d*p . àïp \ ,, n
dt ~ k \ dx* ^ dy* "T" de* J ^^
and we arrive ât the heat-conduction équation.

Let us return to équation (9). If the liquid is incompressible,
then p = const, —- = 0, and (9) becomes
div w = 0 ( 12 )
If the motion is potential, that is, if the vector v is the gradient
of some function q>:
v = grad _n

dxî f d y t ^ à z * (13)
that is, the potential function of the velocity (p must satisfy the
Laplace équation.
In many problems, as, for example, those dealing with filtration,

we can put
v = — kt grad p
where p is the pressure and kx is a constant; we then get the Lap-
lace équation for determining the pressure:
dx* + dy* + dz* ~ U ' ld ’
The boundary-value conditions for équation (13) or (13') may

be the following:
1. On the surface o are specified the values of the desired
function p, pressure [condition (2)]. This is the Dirichlet problem.
2 . On the surface a are specified the values of the normal dé
rivative ; the Dow through the surface is specified [condition (3)].
This is the Neumann problem.
3. On parts of the surface a are specified the values of the
desired function p , pressure, and on parts of the surface are spe
cified the values of the normal dérivative ^ , the flux through the
surface. This is the Dirichlet-Neumann problem.
If the motion is two-dimensional-parallel — that is, the function
<p(or p) does not dépend on z —then we get the Laplace équation
in a two-dimensional domain D with boundary C:
^ £ 4 _ ^ î = v0
dx*^dy* (14)
Boundary-value conditions of type (2), the Dirichlet problem, or
of type (3), the Neumann problem, are specified on the contour C.
III. The potential of a steady-state electric current. Let a ho-
mogeneous medium fill some volume V, and let an electric current
pass through it whose density at each point is given by the vector
J(x, y, z) = J J + J y J + J zk. Suppose that the current density is
independent of the time t. Further assume that there are no cur
rent sources in the volume under considération. Thus, the flux of
the vector J through any closed surface S lying inside the volume
V will be equal to zéro:
Jnds = 0
s
where » is a unit vector directed along the outer normal to the
surface. From Ostrogradsky’s formula we conclude that
div / = 0 (15)
The electric force E in the conducting medium at hand is, on the
6.9 The Laplace Equation in Cylindrical Coordinates 399
basis of Ohm’s generalized law,
e =T . (l6)
or
J=XE
where X is the conductivity of the medium, which we shall con-
sider constant.
From the general electromagnetic-field équations it follows that
if the process is stationary, then the vector field E is irrotational,
that is, rot E = 0. Then, like the case we had when considering
the velocity field of a liquid, the vector field ispotential (see Sec. 3.9).
There is a function <p such that
£ = g ra d ç (17)
From (16) we get
J=Xgrad<p (18)
From (15) and (18) we hâve
Xdiv(grad <p) = 0
or
d x * ^ ‘ dy*~t dz2 (19)
We thus hâve the Laplace équation.
Solving this équation for appropriate boundary-value conditions,
we find the function qp, and from formulas (18) and (17) we find
the current J and the electric force E.
6.9 THE LAPLACE EQUATION IN CYLINDRICAL

COORDINATES. SOLUTION OF THE DIRICHLET PROBLEM
FOR AN ANNULUS WITH CONSTANT VALUES
OF THE DESIRED FUNCTION ON THE INNER
AND OUTER CIRCUMFERENCES
Let u(x, y, z) be a harmonie function of three variables, Then
*0. + + ^!îi = o m
dx* ' dy'1 ^ dz2 U
We introduce the cylindrical coordinates (r, q>, z):
x = rcosqp, x = rsinqp, z= z
whence
r = V x t + y2, qpârctan^-, z= z (2 )
Replacing the independent variables x, y, and z by r, <p, and z,

we arrive at the function u*:
u(x, y, z) = u*(r, q>, z)
Let us find the équation that will be satisfied by u.*(r, «p, z) as a
function of the arguments r, <p, and z; we hâve
du __ du* dr , du* dtp
dx dr cix' dtp ~dx
d2u d2u * ( d r \ 2 du* d2r . 0 d*u* dr dtp f dy \ 2 , du* d*q>
dx2 “ dr2 [ d x ) + dr dx2 ' ^ dr dtp dx dx 1 a<p* \^dx J 'r dq> dx2 { ;
similarly,
d2u d2u* fd r \ 2 . du* d2r . 0 d2u* dr d<p ,. d2u* ( * £ Y . l * î ! 2 ï /a\
dy 2 dr2 \ d y ) dr dy2 ' ^ dr dtp dy dy '1 dtp2 { dy J ' dtp dy2 ' '
besides,
d2u d2u*
dz2 àt2 (5 )
We find the expressions for

dr dr d2r d2r d<p d<p d*<p d2<p
dx’ d ÿ ’ dx*' Hÿ2 ' ~àx ' W ’ ‘3 **’ ’ dy*
from équations (2 ). Adding the right sides of (3), (4) and (5), and
equating the sum to zéro [since the sum of the left-hand sides of
these équations is zéro by virtue of ( 1)], we get
d2u* 1 du* 1 â2u* â2u*
dr2 r dr + r2 dtp* + âz2 (6)
This is the Laplace équation in cylindrical coordinates.

If the function u is independent of z and is dépendent on x and
y, then the function u*, dépendent only on r and <p, satisfies the
équation
d2u* 1 du* 1 d2u*
dr2 ' r dr ' r2 dtp* (7)
where r and <p are polar coordinates in a plane.

Now let us find the solution to Laplace’s équation in trie domain D
(annulus) bounded by the circles Clt x2+ y2= R2, and C2, x2+ y2= R 2,
with the following boundary values imposed:
«le, = «i (8 )
“ Ic, = «a (9)
where wx and u2 are constants.
We will solve the problem in polar coordinates. Obviously, it
is désirable to seek a solution that is independent of <p. Equation
6.10 The Solution of Dirichlet’s Problem for a Circle 401
(7) in this case takes the form

d2u , 1 d u _
dr2 ' r dr
Integrating this équation we find
u — A ln r + B ( 10)
We détermine A and B from conditions (8 ) and (9):
U-I —A ln -j- B
«, = A ln R 2+ B
Whence we find
ln R i _ut ln Rî — Uj ln R t
4 = B = Uj—- (ua—ut)
Substituting the values of A and B thus found into (10), we

finally get
u = ut («2 —Ul) ~ (H )
Note. We hâve actually solved the following problem. Find the

function u that satisfies the Laplace équation in a région bounded
by the surfaces (in cylindrical coordinates)
r -=/?!, r — R v 2 = 0, z = H
and that satisfies the following boundary conditions:
u\r=R, — u|r=R, = Wa
du du I
0
Hz dz \z =
(the Dirichlet-Neumann problem). It is obvious that the desired
solution does not dépend either on z or on <p and is given by for
mula ( 11 ).
6.10 THE SOLUTION OF DIRICHLET’S PROBLEM

FOR A CIRCLE
In an jcÿ-plane, let there be a circle of radius R with centre at
the origin and let a certain function /(<p), where q> is the
polar angle, be given on its circumference. It is required to
find the function u(r, q>) continuous in the circle (including
the boundary) and Satisfying (inside the circle) the Laplace équa
tion
and, on the circumference, assuming the specified values

«Ir=R = /(q>) (2)
We shall solve the problem in polar coordinates. Rewrite équation
( 1) in these coordinates:
d*u . 1 du . 1 d*u _ ^
d r a_ h r d r + r2 ftp2 - ^
or
<n
We shall seek the solution by the method of séparation of variables,
placing
u = Q(<p)R(r) (3)
Substituting into équation (T), we get
r e0 (q>) R" (r) + r 0 (9 ) R' (r) + 0 " (9 ) R(r) = 0
or
0 *(9 ) r»/?»Q-) + r / ? » ,a
0(9) /? (')
(4)
Since the left side of this équation is independent of r and the
fight is independent of 9 , it follows that they are equal to a
constant which we dénoté by —ka. Thus, équation (4) yields two
équations:
0" (9 ) + &20 (9 ) = 0 (5)
r2Rn(r) + rR, ( r ) - k iR(r) = 0 (5')
The general solution of (5) will be
0 = A cos fop + B sin £ 9 (6 )
We seek the solution of (5') in the form R ( r ) = r m. Substituting
R(r) = rm into (5'), we get
r2m (m— 1)rm~i + rrnrm~1—klrm= 0
or
m2 —£* = 0
Thus, there are two particular linearly independent solutions r* and
r~k. The general solution of équation (5') is
R = C r k + Dr~k (7)
We substitute expressions (6 ) and (7) into (3):
u„ = (Ak cos £ 9 + Bk sin £9 ) (Ckrk + Dkr~k) (8 )
Function (8 ) will be the solution of (T) for any value of k different
6.10 The Solution of Dirichlet's Problem for a Circle 403
from zéro. If k = 0, then équations (5) and (5') take the form
0 >" = 0 , rR"(r) + R'(r) = 0
and, consequently,
«o = (4» + B0V)(Co + £>olnr) (8')
The solution must be a periodic function of <p, since for one and
the same value of r for <p and <p+ 2 n we must hâve the same so*
lution, because one and the same point of the circle is considered.
It is therefore obvious that in formula (8 ') we must hâve Bg— 0.
To continue, we seek a solution that is continuous and finite in
the circle. Hence, in the centre of the circle the solution must be
finite for r = 0 , and for that reason we must hâve £>„ = 0 in (8 ')
and Dk = 0 in (8 ).
Thus, the right side of (8 ') becomes the product AgCt, which
we dénoté by A j 2 . Thus,
We shall form the solution to our problem as a sum of solutions

of the form (8 ), since a sum of solutions is a solution. The sum
must be a periodic function of q>. This will be the case if each
term is a periodic function of <p. For this, k must take on intégral
values. [We note that if we equated the sides of (4) to the number
+ k 2, we would not obtain a periodic solution.] We shall confine
ourselves only to positive values:
/î~l, 2, . . . , n, . . .
because the constants A, B, C, D are arbitrary and therefore the
négative values of k do not yield new particular solutions.
Thus,
u(r, <P) “ y + ^ cos «<P+ B„ sin n<p) rn (9)

n= 1
(the constant C„ is included in A„ and Bn). Let us now choose

arbitrary constants An and Bn so as to satisfy the boundary-value
condition (2). Putting into (9) r — R, we get, from condition (2),
/ (qp) = ^ (^4/*cos 4- sin«<p) ( 10 )

n= 1
For (10) to be valid, it is necessary that the function f(q>) be

expandable in a Fourier sériés in the interval (—jx, jt) and that
AnRn and BnRn be its Fourier coefficients. Hence, An and Bn must
404 Ch. 6 Equations of Mathematical Physlcs
be defined by the formulas

n
A n = ~ ^ ^ ^ cosnt dt
( 11)
-Jt
jt
Bn = - s i n rit dt
Thus, the sériés (9) with coefficients defined by formulas (11) will
be a solution of our problem if it admits termwise itèrated diffé
rentiation with respect to r and <p (but we hâve not proved this).
Let us transform formula (9). Putting, in place of An and B„,
their expressions ( 11) and performing the trigonométrie transfor
mations, we get
Jt ooJt
u(r, <P) = i j + J /(/)c o s /t(# —q>) <*<(-£-)"

—Jt 71= 1 - J t
= / ( 0 | 1 + 2 2 (- --)" co s« (/—f j d / ( 12 )
-n L n=\ J
Let us transform the expression in the square brackets: *
oo ao
1 + 2 ^ (-4 -Y cosn (t — qp) = 1 + £ ( j —YêinV-V + e-lnV-it)]
n=l * ' rt=l ^ K '
= i+ L [ ( i e,'('" ) B + ( i e' <<<" p>) n]
= 1
R R
+
1 __ C_e*(t-<p) i ___
R R
2
- ( t ) /?*— /■*
R2— 2Rr cos (t—<p) + r 2 (13)
l-2 -+ c o s « -< p )+ (+ -)2
Replacing the expression in square brackets in ( 12) by expres

sion ( 13 ), we get
R 2—r 2
“ (r * 2rR cos ( t — <p) + r2
dt (14)
* In the dérivation we détermine the sum of an infinité géométrie prog

ression whose ratio is a complex number, the modulus of which is less than
unity. This formula of the sum of a géométrie progression is derived in the
same way as in the aase of real numbers. It is also necessary to take into
aecount the définition of the limit of a complex function of a real argument.
Here, the argument is n (see Sec. 7.4, Vol. I).
6.11 Solution of Dirichlet's Problem b y Method of Fini te Différences 405
Formula (14) is called Poisson's intégral. By an analysis of this

formula it is possible to prove that if the function / (<p) is con-
tinuous, then the function u (r, <p) defined by the intégral (14)
also satisfies équation (T) and u (r, <p)—►•/((p) as r —►/?. That is,
it is a solution of the Dirichlet problem for a circle.
6.11 SOLUTION OF THE DIRICHLET PROBLEM

BY THE METHOD OF FINITE DIFFERENCES
In an xy-plane, let there be given a domain D bounded by

a contour C. Let there be given a continuous function / on the
contour C. It is required to find an approximate solution to the
Laplace équation
d*u , d*u _
dxl ' dy 2 (i)
that satisfies the boundary condition

u\c = f (2 )
We draw two families of straight Unes:
x = ih and y = kh (3 )
where h is the given number, and i and k assume successive intégral
values. We shall say that D is covered with a grid. We call the
points of intersection of the straight
lines lattice points of the grid. ■ c* c
We dénoté by u: k the approxi r
mate value of the desired function l
/
at the point x = ih, y = kh; that is, f
u(ih, kh) = uitk. We approximate r N
D by the grid domain D*, which /
consists of ail the squares that lie /
completely in D and of some that \ fJ
are crossed by the boundary C \ /
(these may be disregarded). Here, H-,
the contour C is approximated by
the contour C*, which consists of ~ô
segments of straight lines of type pig. 146
(3). In each lattice point lying on
the contour C* we specify a value f*, which is equal to the va
lue of the function / at the closest point of the contour C (Fig. 146).
The values of the desired function will be considered only at
the lattice points of the grid. As has already been pointed out
in Sec. 6 . 6 the dérivatives in this approximate method are replaced
by finite différences:
d*u ui + i , k ~ ^ ui,k~\~ul - \ . k
dx* x = ih , y= kh h3
(Pu «Uti — +
dy 2 x = ih , u-kh /J2
The differential équation ( 1) is replaced by a différence équation
(after cancelling out h2):
u t + 1, *— 2u/, * + «,-!, * + «/, *+i 2u(>A+ «,•, =0
or (Fig, 147)
Ui, ft = "î" (M/+J. + *+l + Ui-l, ft + U(, A-l)(4)
For each lattice point of the grid lying inside D* (and not lying
on the boundary C*), we form an équation (4). If the point (x = ih,
y --=kh) is adjacent to the point of the contour C*, then the right
side of (4) will contain known values of /*. Thus, we obtain a
nonhomogeneous System of N équations
(i.k+1) in N unknowns, where N is the number
y\
of lattice points of the grid lying in
(i-W GJc) ïm ) side D*.
We shall prove that the System (4)
has one, and only one, solution. This
is a System of N linear équations in N
0 X unknowns. It has a unique solution if
Fig. 147 the déterminant of the System is not
zéro. The déterminant of the System is
nonzero if the homogeneous System has only a trivial solution. The
System will be homogeneous if /* = 0 at the lattice points on the
boundary of the contour C*. We shall prove that in this case ail
the values uit k at ail interior lattice points of the grid are equal
to zéro. Inside the domain, let uit k be different from zéro. For the
sake of definiteness, we assume that the greatest of them is posi
tive. Let us designate it by uit k > 0 .
By (4) we write
« / . * = T (« / + ! . * + “ /.* + ! + « , - X ,* + M / , * - l ) ( 4 ')
This équation is possible only if ail the values o f , u on the

right are equal to the greatest uit k. We now hâve five points at
which the values of the desired function are k. If none of these
points is a boundary point, then, taking one ôf them and writing
for it the équation (4), we will prove that at certain other points
the value of the desired function will be equal to k. Continuing
in this fashion, we will reach the boundary and will prove that
at the boundary point the value of the function will be equal to

uj<k, which is contrary to the tact that f* — 0 at boundary points.
Assuming that inside the domain there is the least négative
value, we will prove that on the boundarjf the value of the function
is négative, which contradicts the hypothesis.
And so System (4) has a solution and it is a unique solution.
The values uitk defined front the System (4) are approximate
values of the solution of the Dirichlet problem formulated above.
It was proved that if the solution of the Dirichlet problem for
a given domain D and a given function / exists [we dénoté it by
u (x, y)] and if uit * is a solution of (4), then we hâve the relation
\u(x, y) — ui>k\< Ah2 (5)
where A is a constant independent of h.
Note. It is sometimes justifiable (though this has not been rigo-
rously proved) to use the following procedure for estimating. the
error of the approximate solution. Let uffi be an approximate so
lution for a step 2 h, ule% an approximate solution for a step h,
and let Eh (x, y) be the error of the solution u ^ . Then we hâve
the approximate équation
E„(x, y ) t t ± . ( u M - u f t )
at the common lattice points of the grids. Thus, in order to déter
mine the error of the approximate solution for a step h, it is
necessary to find the solution for a step 2h. One third of the diffé
rence of these approximate solutions is the error estimate of
the solution for a step (mesh-length) of h. This remark also refers
to the solution of the heat-conduction équation by the finite-diffe-
rence method.
1. Dérivé an équation of torsional vibrations of a homogeneous cylindrical rod.
Hint. The torque in a cross section of the rod with abscissa x is determined
by the formula M = G I ^ t where 0 (x, i) is the angle of torque of a cross
section with abscissa x at time /, G is the shear modulus, and / is the polar
moment of inertia of a cross section of the rod.
A ns. 4 J r = a 24 - ir , where a2— ^ - t and k is the moment of inertia of unit
o t2 dx2 k
length of the rod.
d 20 d 20
2. Find a solution of the équation -^Tr= a2 5-5 that satisfies the conditions
^ d t2 dx2
0(0, O = 0> t) = 09 0 (*, 0) = cp(*), ~9"~y -0)~0» where
ç(*) = ^ for 0
<p(x) = ^ - + 2 0 o for - « * < /

408 Ch. 6 Equations of Mathematical Phtisies
Give a mechanical interprétation of the problem.

(-O* (2k + 1) nx (2A? + 1) jtat
Ans- 0(x’ /)= 5 E ( à r r p sin1 / - cos ■
/
k=o
3. Dérivé an équation of longitudinal vibrations of a homogeneous cylindrical
rod.
Hint. If u (x t t) is the translation of a cross section of the rod with abs-
cissa x at time t , then the tensile stress T in a cross section x is defined by
the formula T — E S g j , where E is the elasticity modulus of the material and S
is the cross-sectional area of the rod.
Ans. ^ 0,2 » where a2= — , and p is the density of the rod ma
terial.
4. A homogeneous rod of length 21 was shortened by 2X under the action of
forces applied to its ends. At t = 0 it is free of forces acting externally. Déter
mine the displacement u (x, t) of a cross section of the rod with abscissa x at
time / (the mid-point of the axis of the rod has abscissa * = 0).
(—1)*+1 . ( 2 k + \) n x (2k + 1 ) nat
Ans. ----- ; sin ----- -zr1— cos'
(2A+ 1)*‘ 21 21
k=o
5. One end of a rod of length / is fixed, the other end is acted upon by a
tensile force P. Find the longitudinal vibrations of the rod if the force P dôes
8PI . (2 n -\-\)nx
( - 0 " sin----- (2n + 1) nal
not operate when / = 0 . Ans. -ptt1— cos-
ESn* Z*i(2n + l)2 21 21
n- 0
(E and S as in Problem 3).
d2u d2u
6. Find a solution to the équation = a 2— that satisfies the conditions
u(0 , o = o, u (/, t) = A sin iùt
du (x, 0)
u(x, 0) = 0, !=0
dt
Give a mechanical interprétation of the problem.
A sin — x sinco*
a , 2A(ùa . nnat . nnx
Ans. u (jc, 0 = Z sin —— sin j .
(0 ,
sin — /
l
n= 1 û>2-
a
Hint. Seek the solution in the form of a sum of two solutions:
A sin — jesin cot

a
u = v-\-wt where w =
co .
sin — /
is the solution that satisfies the conditions
0(0, 0 = 0, v(l, 0 = 0
dw (x , 0)
V(x. 0) = —w (x, 0), ^ 1 :
dt
^ It is assumed that sin
7. Find a solution to the équation 4 - - = a 2 f-^ that satisfies the conditions

n dt dxz
u (0, /) = 0, u ( L t) = 0,# t > 0
* when 0 C * C
u( x, 0) = { t Z
l —x when < x < l
Ât oo , ( 2 n + 1)* n*a*t /rt , lx _
Ans u ( x t) — 4 V * *
ji* 2 - ( 2n+l)* c <* ç j n
1’ l n*
n=0
Hint. Solve the problem by the method of séparation of variables.
8. Find a solution to the équation satisfies the conditions
x (l-x )
u( 0 , t) = u (/, / ) = 0 , u(x, 0)
" /2
oo (2/i+ 1)f TPaït . (2n+ 1 ) nx
An*. u(x, “ sin ------— — .
n= 0
9. Find a solution to the équation '^ " ==aa^ that satisfies the conditions
du
= 0, u(l, t) = u0, u ( x , 0) = <p(x)
dx x = o
Point out the physical meaning of the problem.
00
Ans. u (x, 0 = ^ 0 + *cos ^ where
n =0
/
(2/t+l)n,x (—l)n4u^
^n = 7 - jW )C O S X ji(2/i + 1) *
2/
Hint. Seek the solution in the form u = u0-\-v (xt t).

10. Find a solution to the équation -^ - = a 2 that satisfies the conditions
«(0 . 0 - 0. • I - L .,— W + .,. u(x, 0) = <p(x)
Point out the physical meaning of the problem.
Ans. U(x, /) = £ A„ p2+ng /• . P„X

where
sin î—-
_ P ( P + l) + P/i
/
i 4 „ = y ^ <p (*) sin — -c£*, p = Hl, \il9 fi2, . . . , |xn are positive roots of the
équation tan fi = — .
Hint. At the end of the rod (when x = l) a heat exchange occurs with the
environment, which has a température of zéro.
11. Find [by formula (10), Sec. 6.6, putting h = 0.2] an approximate solution
to the équation - ^ - = 2 that satisfies the conditions
u(x, 0 ) = * ^ - |— , « (0 ,< ) = 0, u ( l, <) = + 0 < /< 4 1

d2u , d2u
12. Find a solution to the Laplace équation 0, in a strip
dy2
0 < f/ < oo that satisfies the conditions
«(0, y) = 0, u(a, y ) = 0, « (*, 0) = A ^1 —- j ) , u(x, oo) = 0
- . . 2i4 1
sin
MIX
Ans. u(x, 0 = — > —*
7 ji n
Hint. Use the method of the séparation of variables.

d2u d2u
13. Find a solution to the Laplace équation "g^â' + ' ^ r = :0 ' n the rectangle
0 < x < « , 0 < i / « s 6 that satisfies the conditions
u(x, 0) = 0 , u(x, b) = 0, «(0, y) = Ay(b — y), u(a, y) = 0.
. (2 n + 1) n ( a — x) . (2n + \ ) n y
sin h ---- !—'-r2------- sin --------- - —
.
Ans. / t)
u(x, m= —^ 8Ab*
~ 2V»
u
(2/i+I)* . (2/1+ 1) Jta
sinh----- -T——
o
d2u d2u
14. Find a solution to the équation ïns^ e an annulus boun-
ded by the circles *2+ y2 = Rl, x?-\~y2 = R\ that satisfies the conditions
du
“ r=R . = “»
dr r=Rt + X2JI7?! ’
Give a hydrodynamic interprétation of the problem.
Hint. Solve the problem in polar coordinates.
A Q 1 # 2
Ans. u = w2 —T 2vâî—
,ji 1°—1
r •
15. Prove that the function w(x, i/) = sin x is a solution of the équation
d2u d2u
' =0 in the square 0 ^ a : ^ 1 , 0 ^ y ^ \ that satisfies the conditions
dx2 1 dy2
u (0, y) = 0, u (1, y) = e~y sin 1, u (x, 0) = sin jc, u (x, l) = e- 1 sinx.
16. In Problems 12-15 solve the Laplace équations for given boundary con
ditions by the finite-difference method for /i = 0.25. Compare the approximate
solution with the exact solution.
CHAPTER 7
OPERATIONAL CALCULUS AND

CERTAIN OF ITS APPLICATIONS
Operational calculus is an important branch of mathematical

analysis. The methods of operational calculus are used in physics,
mechanics, electrical engineering and elsewhere. Operational calcu
lus finds especially broad applications in automation and teleme-
chanics. In this chapter we give (on the basis of the foregoing
material of this text) the fundamental concepts of operational cal
culus and operational methods of solving ordinary differential
équations. *
7.1 THE ORIGINAL FUNCTION AND ITS TRANSFORM
Suppose we hâve a function of a real variable t defined for t ^ O

[we shall sometimes consider that the function / (/) is defined en
an infinité interval — oo < t < oo, but f(t) — 0 when t < 0]. We
shall assume that the function }(t) is pieeewise continuous, that is,
such that in any finite interval it has a finite number of discon-
tinuities of the first kind (see Sec. 2.9, Vol. I). To ensure the
existence of certain intégrais in the infinité interval 0 ^ . t < oo we
impose an additional restriction on the function f (t): namely, we
suppose that there exist constant positive numbers M and s0 such
that
\f{t)\<Me*>t ( 1)
for any value t in the interval 0 < ! / < oo.
Let us consider the product of the function f (t) by the complex
function e~pt of a real variable** t , where p = a + ib (a > 0 ) is
some complex number:
e~ptf(t) (2 )
* The following books are recommended for further study of operational

calculus: A. I. Lurie, Operational Calculus and Its Applications to Problems
of Mechanics, Moscow-Leningrad, Gostekhizdat, 1950; V. A. Ditkin and
P. I. Kuznetsov, Handbook of Operational Calculus, Moscow, Leningrad,
Gostekhizdat, 1951; V. A. Ditkin and A. P. Prudnikov, Intégral Transforma
tions and Operational Calculus, Fizmatgiz, 1961; Jan G. Mikusinski, Rachu-
nek opérât oràw, Warszawa, 1953, (Operational Calculus).
** See Sec. 7.4, Vol. I, conceming complex functions of a real variable.
412 Ch. 7 Operational Calcul us and Certain of lis Applications
Function (2 ) is also a complex function o! a real variable t :

e - p ' f (t) = e ~<a + m f (t ) = e ~ aif (t ) e ~ ibt
= e ~ atf (t) cos bt — i e ~ aif (t) sin bt
Let us further consider the improper intégral

oo 00 oo
J e~ptf (t ) d t — J e ~ atf (t) cos bt d t — i J e ~ atf (t) sin b t d t (3)
o o o
We shall show that if the function f (t) satisfied condition (1) and
a > s0, then the intégrais on the right of (3) exist and the con
vergence of the intégrais is absolute. We begin by evaluating the
first of these intégrais:
^ e ~ atf (t) cos bt dt

o
< M Çe-t*-»,>*dt=> —
M § e ~ ates«t d t
o
In similar fashion we evaluate the second intégral. Thus, the in
tégral J e~pif ( t ) d t exist. It defines a certain function of p, which
o
we dénoté* by F (p):
oo
F (p) — \ e ~ptf (t) d t (4)
o
The function F (p) is called the L apl ac e t r a n s f o r m , or the L - t r a n s -
forrn, or simply the t r a n s f o r m of the function f ( t ) . The function
f ( t ) is known as the o r i g i n a l f un c t i o n. If F (p) is the transform
of /(/), then we Write
F(p)+f(t) (5)
f(t ) + F( p) (6)
L{ f ( t ) } = F( p) (7)
As we shall presently see, with the help of transforms it is pos
sible to simplify the solution of many problems, for instance, to
reduce the solution of difîerential équations to simple algebraic
* The function F (p), for p 7= 0, is a function of a complex variable (for
example, see V. I. Smirnov’s Course of Higher Mathematics, Vol. III, Part 2,
in Russian). The transformation (4) is similar to the Fourier transformation consi
dérée! in Sec. 5.14.
7.2 Transforms of the Functions <J0 (t), sin t, cos t 413
operations in finding a transform. Knowing the transform one can

find the original function either from specially prepared “original
function-transform” tables or by methods that will be given below.
Certain natural questions arise. *
Let there be given a certain function F(p). Does there exist a
function f (t) for which F (p) is a transform? If there does, then
is this function the only one? The answer is yes to both ques
tions, given certain definite assumptions with respect to F (p) and
/(/). For example, the following theorem, which we give without
proof, establishes that the transform is unique:
Uniqueness Theorem. If two continuons functions <p (t) and \|) (t)
hâve one and the same L-transform F (p), then these functions are
identically equal.
This theorem will play an important rôle throughout the subsé
quent text. Indeed, if in the solution of some practical problem
we hâve in some way determined the transform of the desired
function, and from the transform the original function, then on
the basis of the foregoing theorem we conclude that the function
we hâve found is the solution of the given problem and that no
other solutions exist.
7.2 TRANSFORMS OF THE FUNCTIONS O0( t ) , S În f, COS*
1. The function f (t), defined as

/ (/) = 1 for 0
f ( t ) = 0 for t < 0
is called the Heaviside unit function and is denoted by o0(t). The
graph of this function is given in Fig. 148.
Let us find the L-transform of the Heavi 1 6o(t)
1s
side function:
-pt 0 f
L{o9(t)} = §e~Ptdt:
p Fig. 148
0
Thus,*
l t - ( 1)
p
or, more precisely,
°o (0 * 7
* In computing the intégral y e - P t d i it is possible to represent it as the

o
sum of intégrais of real functions; the same resuit would be obtained. This also
holds for the next two intégrais.
414 Ch. 7 Operational Calculas and Certain of Ils Applications
In some books on operational calculus the following expression

is calléd the transform of the function f(t):
CO
F*{p) = p \ e - # f ( t ) d t
o
With this définition we hâve a0(/)-f 1 and, consequently, C+rC,
more exactly, Ca0(t)-{rC.
II. Let f(t) = smt \ then
oo
P . i. e~Pt (— p s i n / — cos 0
L {sin/}= \ e pts m t d t = -------—^ - j ------- -
o P2+ l
And so
sin t -*r (2)
P2+ l
III. Let f(t) = cost\ then
L {cos t } = ] e-r< cos t dt = ^ (si" ^ 008 0

o p2 + l
o
Thus,
cos t
P2+ 1
(3)
7.3 THE TRANSFORM OF A FUNCTION WITH CHANGED

SCALE OF THE INDEPENDENT VARIABLE.
TRANSFORMS OF THE FUNCTIONS sln a t , cos a t
Let us consider the transform of the function f{at), where a > 0:

oo
L{f (at)} = J e~ptf (at) dt
o
We change the variable in the intégral, putting z = at\ hence,
dz = adt\ then we get
oo p
L { i ( a t ) ) = ^ e ~ Zf(z)dz
0
or
m m ) - ^ r ( i)
Thus, if
F( p) + Ï (t )
then
-f(f)+/(«o ai
7.4 The Linearity Property of a Transform 415
Example 1. From (2) of Sec. 7.2, by (1), we straightway get

. . .
sin at *-----
1 1
• • (*> ■ *»
or
sin at ■ (2)
• p2+ a a
Example 2. From (3), Sec. 7.2, by (1), we obtain
P_
cos at
. 1 a
' * (-5 )* + .
or
cos at + P
• p2+ a2 (3)
7.4 THE LINEARITY PROPERTY OF A TRANSFORM

Theorem. The transform of a sum of several functions multiplied
by constants is equal to the sum of the transforms of these func
tions multiplied by the corresponding constants, that is, if
f(t)= ic,f,(t) (O
i= I
(C/ are constants) and
F( p)+f(t), F M + f r f )
then
F { p ) = ' k c , F l {p) (!')
i=i
Proof. Multiplying ail the terms of (1) by e~Pf and integrating
with respect to t from 0 to oo (taking the factors C (- outside the
intégral sign), we get (T).
Example 1. Find the transform of the function
/ (/) = 3 sin 4t —2 cos 51
Solution. Applying formulas (2), (3), Sec. 7.3, and (T), we hâve
L If (t)\ —3 ^ - 2 ^ ^ ______
' '' p2+ 1 6 p2+ 2 5 ~ p 2+ 1 6 p2+ 25
Example 2. Find the original function whose transform is expressed by the
formula
5 20p
F (P)
P2+ 4 P2+ 9
Solution. We represent F (p) as
5 2
416 Ch. 7 Operational Calculus and Certain of lis Applications
Hence, by (2), (3), Sec. 7.3, and (T), we hâve

/ ( / ) = A sin 2/ + 2 0 cos 31
From the uniqueness theorem, Sec. 7.1, it follows that this is the on 1y original
function that corresponds to the given F (p).
7.5 THE SHIFT THEOREM
Theorem. If F (p) is the transform of the function f (t), then

F(p-\~a) is the transform of the function e -atf\t), that is,
if F ( p ) + f( t) 1
( 1)
then F ( p + a ) e ~ at f (t) J
[It is assumed here that Re(/? + a) > V ]
Proof. Find the transform of the function e~aif(t):
L{c-“7 (0 }- 5e~pl~atf(t) dt = $ e~(p+a) 7 (t)dt

o o
Thus,
L{e~atf (t)} = F(p + a)
This theorem makes it possible to expand considerably the class

of transforms for which it is easy to find the original functions.
7.6 TRANSFORMS OF THE FUNCTIONS « - “ ' s i n h a * , c o s h a * ,

e~ai s in at, e~at c o s at
From ( 1) (Sec. 7.2, on the basis of ( 1), Sec. 7.5, we straight-
way get
-P+a
+ - •L- e - “< '( 1')
Similarly,
p —a • (H
Subtracting from the terms of (T) the corresponding terms of (1)
and dividing the results by two, we get
or
+ + 2 -^ s in h a / (2 )
Similarly, by adding ( 1) and (T), we obtain
i^ c o s h a / (3)
7.7 Différentiation of Transfor ms 417
From (2), Sec 7.3, by ( 1), Sec. 7.5, we hâve

(p+-a}*+a^ e-al s in * (4)
From (3), Sec. 7.3, by ( 1), Sec. 7.5, we get

p+ a
( p + a ) 3+ a2
■e~a tcos a/ (5)
Example 1. Find the original function whose transform is given by the
formula
f ( p ) = p2+ 1 0 p + 41
Solution. Transform F (p) to the form of the expression on the left-hand
side of (4):
7 7 7 4
p2+ 1Op + 41 (p + 5)2+ 1 6 4 (p+ 5)*+ 4*
Thus
7 4
F(p) =
4 (p + 5)2+ 42
Hence, by formula (4) we will hâve
F (p) sin 4/
Example 2. Find the original function whose transform is given by the
formula
F (p) -- P+ 3
P2+ 2p + 10
Solution. Transform the function F (p):
p+ 3 _ ( p + l ) + 2p + 1
p2 + 2 p + 1 0 ( p + l ) 2+ 9 (p + l)* + 3*
_|_ 2 P+1 | _2_
(p + l)* + 3* ( P + l ) 2+ 32 3 ( p + l ) 2+ 32
Using formulas (4) and (5), we find the original function:
F (p) +-► cos 3/ + - |- e - f sin 3/
o
7.7 DIFFERENTIATION OF TRANSFORMS

Theorem. If F (t), tken
O)
Proof. We first prove that if f(t) satisfies condition (1), Sec. 7.1,
then the intégral
co
[e-P*(-tyf{t)dt ( 2)
exists.
27—2082
418_______ Ch. 7 Operational Calcul us and Certain of lis Applications
By hypothesis, | / (t) | < Mes>‘ , p = a + ib, a > s0; and a > 0,

s0 > 0 . Obviously, there will be an e > 0 such that the inequality
a > s „ + e will be fulfilled. As in Sec. 7.1, it is proved that the
following intégral exists:
oo
^g -(a-e) (t)\dt
0
We then evaluate the intégral (2 )
oo oo
J | e~pttnf (t) | dt = J | e-<P-E>te-** tnf (t) | dt

0 0
Since the function £“c//'z is bounded and, in absolute value, is
less than some number N for any value t > 0, we can write
00 00 CO
5 | e-P*tnf (t) Idt < N J I e-tp-w f (t) \ dt = JV J e -</>-*>' | / (/) | dt <oo
0 0 0
It is thus proved that the intégral (2) exists. But this intégral
may be regarded as an nth-order dérivative with respect to the
parameter * p of the intégral
oo
\e~P*f(t)dt
o
And so, from formula
oo
F{P) — \ e~Ptf (t) dt

o
we get the formula
oo oo
J e-pt ( _ 0 * f {t) dt = U j e-ptf ( 0 dt

0 0
From these two équations we hâve
oo
( - l ) n £-nF(P) = $ e - pttnHt)
0
which is formula ( 1).
Let us use (2) to find the transform of a power function. We
write the formula ( 1), Sec. 7.2:
* Earlier we found a formula for differentiating a definite intégral with res

pect to a real parameter (see Sec. 11.10, Vol. I). Here, the parameter p is a
complex number, but the différentiation formula holds true.
7.8 The Transforms of Dérivatives 419
Using formula ( 1) of this section, from this formula we get
( t
or
Similarly
For any n we hâve

n! .
p/l +1 . tn (3)
Example 1. From the formula [see (2), Sec. 7.3]

00
■2+T"û22 = J^ e~Pt SÎn
P2 dt
0
by difïerentiating the left and right sides with respect to the parameter p,
we get
2pa
• t sin ai
(p2+ a 2)2 • (4)
Example 2. From (3), Sec. 7.3, on the basis of (1), we hâve
*t cos at (5)
(P2+ a2)2
Example 3. From (1), Sec. 7.6, by (1), we hâve
1
*te- a t ( 6)
(p+a )2
7.8 THE TRANSFORMS OF DERIVATIVES
Theorem. If F ^ - ^ f i t ) , then
p F (p )-H O )^f'(t) ( 1)
Proof. From the définition of a transform we can Write
ce
L { r ( t ) } = l e - r ‘f' (t )d t (2 )
o
We shall assume that ail the dérivatives f'(t), f ( t ) .........
which we encounter satisfy the condition ( 1), Sec. 7.1, and, con-
sequently, the intégral (2 ) and similar intégrais for subséquent
420 Ch. 7 OperationaL Calculas and Certain of lis Applications
dérivatives exist. Computing by parts the intégral on the right

of (2 ), we find
ao ao
L \ r (0 } = $ e-P‘r (t) dt = e -# f (t) |0" + p ^ f (t) dt
o o
But by condition ( 1), Sec. 7.1,
Iim e~ptf(t) — 0
t -* 00
and
œ
\ e - p t n t ) d t = F(.p)
o
Therefore
L \ r (t)} = - f ( 0 ) + pF(p)
The theorem is proved. Let us now consider the transforms of
dérivatives of any order. Substituting into ( 1) the expression
pF(p)—f ( 0) in place of F (p) and the expression /' (/) in place
of f(t), we get
p[pF(p)-f(0)]-r(0)-^f"(t)
or, removing brackets,
p* F ( p ) - p f ( 0 ) - r ( 0 ) ^ r ( t ) (3)
The transform for a dérivative of order n will be
pnF (p) - [pn- ( 0 ) + pn~*ff (0) + . . .
+ p/(n- 2) (0) + /(" - 1>(0 )] (t) (4)
Note. Formulas (1), (3), and (4) are simplified if / (0) = /' (0) =
= . . . = / (n~l)(0) = 0. In this case we get
p F ( p ) - ^ r (t)
p« F (p )-f-* r(o
7.9 TABLE OF TRANSFORMS
For convenience, the transforms which we obtained are here

given in the form of a table.
7.9 Table of Transfor ms 421
oc
No. P(P) = $ e - ptHt) dt fit)
0
•
1
1 1
P
a
2 sin at
P%+ a2
3 P cos at
P4 +<**
1
4 e
P+ «
a
5 sinh at
p2— a 2
6
P cosh at
p2— a 2
a
7 e~at sin at
(p + a )4+ a%
P+ a e~at cos at
8 ( p + a ) 2-(-a*
ni t«
9 pn +l
2pa t sin at
10 (p2+ a2)2
p2— a2 t cos at
11
(p2+ a2)2
1 t e - at
12
(P + a )2
1 ^ (sin at— at cos at)
13 (p2 + a2)3
14 (-l)^ f(p ) t nf( t)

t
15 Fi (P) Fi (p) $ /i W h V -r)dx
0
Note. Formulas 13 and 15 of this table will be derived later on.

If for the transform of the function f(t) we take
ao
F* (p) = p \ e ~ P tf(t)dt
o
then in the formulas 1-13 of the table the expressions in the first
column must be multiplied by p (formulas 14 and 15 will change
422 Ch. 7 Opérâtionai Calculus and Certain of lis Applications
more fundamentally). Since F*(p) —pF(p), it follows that by

substituting into the left side of formula 14 the expression ■ ■
in place of F (p) and multiplying by p, we get
14'
Substituting into the left side of formula 15
f U p) c f U p)
FAP) = p ■ F*
■* (P): p
and multiplying this product by p, we hâve
15' j F l (p) Fl (p) (t) /, (t- t) dx
7.10 AN AUXILIARY EQUATION FOR A GIVEN

DIFFERENTIAL EQUATION
Suppose we hâve a linear differential équation of order n with

constant coefficients a0, alt . . . . an- lt an\
dnx . dn~1x , , dx - /,v £, /1X
a°375 + ai jjn- 1+ • • • + ÛB-1 57 + an-*:(0 = / (0 (1)
It is required to find a solution of this équation x = x(t) i o r t ^ O
*(0) = xo, x'(0) = x'„ .... jc<"-»(0) = x<'>-1> (2)
Before, we solved this problem as follows: we found the general
solution of équation ( 1) containing n arbitrary constants; then we
determined the constants so that they should satisfy the initial
conditions (2 ).
Here we give a simpler method of solving this problem using
operational calculus. We seek the L-transform of the solution x(t)
of (1) satisfying the conditions (2 ). We designate this L-transform
by x(p); thus, x(p)-:r*x(t).
Let us suppose that there exist transforms of the solution of (1)
and of its dérivatives up to order n inclusive (after finding the
solution we can test the truth of this assumption). We multiply
ail terras of ( 1) by e~pt, where p = a + ib, and integrate with res
pect to t from 0 to oo:
ao ] e~ptÿ z dt + ai ] e~pti ï £ r d t +
0 0
oo oo
+ . . . + an j e-p,x (t) dt = j e -# f (t) dt (3 )
7.10 An Auxiliary Equation for a Given Differential Equation 423
On the left-hand side of the équation are the L-transforms of the

function x(t) and its dérivatives, on the right, the L-transform of
the function f {(), which we dénoté by F (j>). Hence, équation (3)
may be rewritten as
Û«L { S } + 1a'L { } + • • • + ' anL (0 } = ■

L {/ (/)}
Substituting into this équation the expressions (1), (3), and (4 ),
Sec. 7 .8 , in place of the transforms of the function and of its
dérivatives, we get
a0{pnx(p) — [p"~1x0+ pn- îx'0+ pn~ % + . . . + 4 n- ‘>]}
+ a1{pn~1x{p) — [pn- ix0+ pn- ax'0+ . . .
+ an. 1{px(p)— x(>}-\-anx{p) = F<j}) (4 )

Equation (4) is known as the auxiliary équation, or the transform
équation. The unknown in this équation is the transform T(p),
which is determined_from it. Transform it leaving on the left the
terms that contain x{p):
'x(p)[a0pn+ a 1pn~1+ . . . + a n. 1p + an]
= a 0 [pn~1x0-\-pn- ix'0-\-. . . + 4 "-1>]
+ at [pn- 2x0+ Pn- 3x'0+ ■• • + 4 " -,>]
+ a«-»[pxt>+ *'•]+<*„-i*o + F (p). (4')

The coefficient of x(p) on the left of (4') is an nth-degree poly
nomial in p, which results when in place of the dérivatives we
put the corresponding powers of p into the left-hand member of
équation (1). We dénoté the polynomial by <p„(p):
<P„ (P) = a*Pn + + • • • + fl„- ,P + an (5)
The right-hand side of (4') is formed as follows:
the coefficient an_! is multiplied by x0,
the coefficient a „ _ 2 is multiplied by px0+ x'0,
the coefficient ax is multiplied by pn- 2x0+ p n~3x'0-\-. . . + x 0(n~î)

the coefficient a0 is multiplied by pn~1x„ + pn~2x'0+ . . . -Fxl"^
Ail these products are combined. To this is also added the transform
of the right side of the differential équation F(p). Ail terms of
the right side of (4'), with the exception of F(p), form, after
collecting like terms, a polynomial in p of degree n — 1 with
424 Ch. 7 Opérai ional Calcul us and Certain of Iis Applications
known coefficients. We dénoté it by And so équation (4')

can be written as follows:
x (P) <P„ (P) = t n-i (P) + F (P)
Frora this équation we détermine x(p):
/ n\ _ t n - l ( P ) , f (p)
( 6)
<pn (P) "1~ q>« (p)
Determined in this way, x(p) is the transform of the solution
x(t) of the équation ( 1), which solution satisfies the initial con
ditions (2). If we now find the function x* (t) whose transform is
the function x(p) determined by équation (6 ), then, by the uni-
queness theorem formulated in Sec. 7.1, it will follow that x*(t)
is the solution of équation ( 1) that satisfies the conditions (2 ),
that is,
x*(/) = jc( 0
If we seek the solution of (1) for zéro initial conditions:
x, = *« = *ô = ••• = xl0n~v = 0 , then in (6 ) we will hâve (p)=0
and the équation will take the form
x(p) F (P)
<P» (P)
or
F (p)
*(/>) = - ( 6 ')
a0P”+ûlPB-1+ • ••
Example 1. Find the solution of the équation
dx
w + x=l
ifying
1
* ( p ) { p + *) = 0 + y or *(P ) = -
(P + 1 )P
Decomposing the fraction on the right into partial fractions, we get
-, , 1 1
,w - 7 ~
Using formulas 1 and 4 of the table, we find the solution:
x (t) = 1—e-*
$ ■ + * -'
that satisfies the initial conditions: x b = x 'q= Q for t=Q.
Solution. Write the auxiliary équation (4')
7.10 An Auxiliary Equation for a Given Differential Equation 425
Decomposing this fraction into partial fractions, we get

1 1
*(/?) = 9 p 9
p2 + 9 +7
Using formulas 1 and 3 of the table we find the solution:
x(t) = - - ~ c o s 3 l + ± -
i £ + * w + 2x= t
that satisfies the initial conditions: xo = x'o = 0 for / = 0 .
Solution. Write the auxiliary équation (4')
x (P) (P2 + 3p + 2) =
or
P2 (p2 + 3p + 2) p*(p+ \)(p + 2)

Decomposing this fraction into partial fractions by the method of undetèrmined
coefficients, we obtain
11 3 1 ,1 1 .
* (P)~ 2 pa 4 p + p + l 4(p + 2)
From formulas 9, 1 and 4 of the table we find the solution:
* (0 = Ÿ < - | + e - ' - { « - ît
d2x . _ dx , e . .
~ d F + 2 i r + 5 x = S in t
satisfying the initial conditions *0= 1 , x0 = 2 for / = 0.
Solution. Write the auxiliary équation (4'):
*(p) (p2+ 2 p + 5) = p - l + 2 + 2 1 + L { s in < }
or
x(p)(p2 + 2p + 5) = p + 4 + — -T
whence we find x (p):

x tp)= _ _ £ Ü _ - 1 _________ !_________
KP) p* + 2p + 5 + (p2+ l ) ( p 2+ 2p + 5)
Decomposing the latter fraction on the right into partial fractions, we can write
11 , , 1 , 1
10p + 4 , 10P+ 5
* W ~ pi*+ 2p + 6 + p*+l
426 Ch. 7 Operational Calculus and Certain of Its ApplicationV
or
ü P+l | 29 2 1 p , 1 1
KP) 10 ' ( p + l ) 2 + 22 - r 10.2 * ( p + l ) 2+ 22 10* p * + l ’1" 5 pa+ l
Applying formulas 8, 7, 3, and 2 of the table, we get the solution
, , , 1 1 f 0 , , 29 # 1 , - 1 . ,
* (0 = î ô ^ fcos 2/+2Ôe" sin2/” Jô cos/+ “5 sin*
or, finally,
11 OQ \ 1 1
( ygcos 2/ + 2Q sin 2 t J — jq c o s i + g - s i n /
7.11 DECOMPOSITION THEOREM
From formula (6 ) of the previous section it follows that the

transform of the solution of a linear difïerential équation consists
of two terms: the first term is a proper rational fraction in p, the
second term is a fraction whose numerator is the transform of the
right side of the équation F (p), while the denominator is the
polynomial q>„(p). If F (p) is a rational fraction, then the second
term will also be a rational fraction. It is thus necessary to be
able to find the original function whose transform is the proper
rational fraction. We shall deal with this question in the présent
section. Let the L-transform of some function be a proper rational
fraction in p:
^n-l(P)
<M p )
It is required to find the original function. In Sec. 10.7, Vol. I,
it was shown that any proper rational fraction may be represented
in the form of a sum of partial fractions of four types:
11. — ^ — ;
(p—a)*
III. p i Q2. where the roots in the denominator are complex,

2
that is, --— a2 < 0;
IV. ------— —, where k ^ 2 , the roots of the denominator

(P2+ûiP + a2)*
are complex.
Let us find the original functions for these partial fractions.
For a type I fraction we get (on the basis of formula 4 of the
table)
A
p—a
Aeat
7AÎ Décomposition Theorem 427
For a type II fraction, by formulas 9 and 4 of the table, we hâve
i \ - xeat (i)
(p—a)k ' — O*
Let us now consider a type III fraction. We perform identity

transformations:
Ap+B Ap + B
P3+aiP+aa
( p + T ) 2+ ( l ^ a*—t )
p+
_ ^ iiW !z îL -
( ',+ t ) ’+ ( Ÿ t) ( ,,+ t ) * + ( V * - + )
+(b- ^ V
w m a - )' 4
Denoting the first and second terms by M and N, respectively,

we get (from formulas 8 and 7 of the table)
2
M^+Ae COS t y r “. - 4 -
N ( B - 4)
And, finally,
Ap+B
P2+ ai P + at
-± tr /---
A cos t sinrf Y aa ( 2)
l_
We shall not consider a type IV partial fraction, since it would
involve considérable calculations. We shall consider certain spécial
cases below. For further information see the list of books at the
beginning of this chapter.
«8 Ch. 7 Operational Calculus and Certain of tts Applications
7.12 EXAMPLES OF SOLUTIONS OF DIFFERENTIAL EQUATIONS

AND SYSTEMS OF DIFFERENTIAL EQUATIONS
BY THE OPERATIONAL METHOD
^d“X . . = sin3jc
T+4* . „
that satisfies the initial conditions: x0 = 0, jcq = 0 when t — 0.

x(p) (PJ + 4 )= p2 + 9 , Jc(p)“ (p4+9)(p2+ 4 )

or
_ 3_ 3_
-, 5 . 5 1 3 ,3 2
Xlp)— p* + 9 ~t p » + 4 “ 5 ' p * + 9 MO* p * + 4
whence we get the solution
j = sin2/ — ~ sin 3/
lu o
cPx . n
-dpr+x=0
that satisfies the initial conditions: x0= \ 9 x'0 = 3, *o = 8 when t —Q.
Solution. Forming the auxiliary équation
x (p) (p3+ l) = p2-l + p-3 + 3
we find
P2+ 3p + 8 _ pa + 3p + 8
*(P) =
P3+ 1 ( p + l ) ( p 2- p + l )
Decomposing the rational fraction obtained into partial fractions, we get
p* + 3p + 8 _ 2 , -p + 6
(P + 1 )(P 2- P + 1 ) P+l P2— P + l
1 VJ
=2. 1 P~~2 11 2
P+l
Using Table of Transforms (Sec. 7.9), we write the solution:
*(<) = 2e-t+ e 2 ( - c o s ^ ^ s i n Ç f )
d2x
+ j c = t cos 2 t
dt*
that satisfies the initial conditions: x = 0, xo = 0 when f = 0 .
7.13 The Convoi ut ion Theorem 429
whence, after some manipulâting, we get

5 1 ,5 1 8
* (P )= - 9 p2+ 1 ^ 9 p2+ 4 > 3 (p2+ 4)2
Conséquent ly,
x(t)= — sin t + ~ sin 2/ + — sin 2t — t cos 21
Obviously, the. operational method may also be used to solve Systems of linear
differential équations. The following is an illustration.
Example 4. Find the solution of the set of équations
i r + 4 i n ' + 3!/=0
that satisfies the initial conditions: x = 0, y = 0 when t = 0.
Solution. We dénoté x ( t ) ^ r x( p) t y (t) -«-f- y (p) and write the System ôf
auxiliary équations:
(3p + 2) x ( p) +pÿ ( p) = j
Px(P)-h(*P + 3)ÿ(p) = 0
Solving this System, we find
—, 4p + 3 _ 1 133
X(p> p ( p + 1) (1 lp + 6) 2p 5 ( p + l ) 10(11/7 + 6)
yyfJ’ (1 lp + 6) (p + 1) 5 V P-r * Hp + 6 J
From the transforms we find the original functions, i.e., the sought-for solutions
of the System:
Linear Systems of higher orders are solved in similar fashion.
7.13 THE CONVOLUTION THEOREM
The following convolution theorem is frequently useful when

solving differential équations by the operational method.
Convolution Theorem. If Fx (p) and F2(p) are the transforms of
the functions f1(t) and f 2(t), that is,
F A P ) ^ h (t) and Ft ( p ) ^ f A t )
430 Ch. 7 Operational Calcul us and Certain of Its Applications
then F1(p)- F2(p) is the transform of the function

t
\ h (T) t
0
that is,
*
F1(p)F2( p ) ^ l f l (r)fAt-*)dT ( 1)
0
Proof. We find the transform of the function
S f i (*)/*(t— x)dx
0
from the définition of a transform:
e - p t
{} fi (T)/* (<—T)dtj = 5 S fi W ^ (/—t) dx^dt
The intégral on the right is a twofold iterated

intégral of the form J J 0 ( t , t) dt dx, which
D
is taken over a région bounded by the straight
lines t = 0, x = t (Fig. 149). Changing the
order of intégration in this intégral, we get
|î fi (t)/. u — t) dT| = S[/i w Se ~pt f» v — T) dt j dx

Changing the variable t — x = z in the inner intégral, we obtain
00 00 00
J e-r*ft (t — x)dt — J e~P^+x>f t (z)dz = e~Px J e~Pz f t (z)dz = e~'nF2(p)
t o o
Hence,
L | S fi (t) fi ( t — t ) dxj = J /j (t ) e-p'Fi (p) dx

oo
= Pi (P) S e~pxfi (t) dx = Ft (p) F1(P)
And so
Pi(p)FAp)
0
7:13 The Convoi ut ivn Theorem 43»
This is formula 15 of Table of Transforms (see Sec. 7.9).

t
Note I. The expression \ f i ( x ) f i ( t — x)dx is called the convolu•
o •
tion (Faltung, résultant) of two functions f 1(t) and /2(/). The ope
ration of obtaining it is also known as the convolution of two
functions; here
t t
S fl (t) f 2(t — x) d x = ^ f l (t — x) f t (t) dx
0 0
That this équation is true is évident if we make the change of

variable t — x — z in the right-hand intégral.
Example. Find the solution to the équation
d?x
~£+*=r«)
that satisfies the initial conditions: x0 = *i = 0 for / = 0 .
X(P) (p2+ l) = F (p)
where F (p) is the transform of the function f(/). Hence,
* ( P ) = p21+ [ - f (P), but sin t and F (p) -f» /(<)■
Applying the convolution formula (1) and denoting j = F 2 (p), F (p) = F1(p)t
we get
t
X (t) = 5 f (T) sin (* — T) dt (2)
0
Note 2. On the basis of the convolution theorem it is easy to

find the transform of the intégral of the given function if we
know the transform of this function; namely, if F (p)-r*f(t), then
t
jF (p )^]f(x)d x (3)
o
Indeed, if we dénoté
M 0 = /(0 . M 0 = 1, then Fl (p) = F(p), Ft (p) = j
Putting these functions into (1), we get formula (3).

432 Ch. 7 Operational Calculus and Certain of Iis Applications
7.14 THE DIFFERENTIAL EQUATIONS OF MECHANICAL

VIBRATIONS. THE DIFFERENTIAL EQUATIONS
OF ELECTRIC-CIRCUIT THEORY
F rom m ec h a n ics w e k n o w th a t th e v ib r a tio n s o f a m a te r ia l p o in t

o f m a ss m are d escrib ed b y th e é q u a tio n *
tPX , \ dx . k 1 t ,,, /.v

d F + m d î+ m X ~ m ^ ( /) (0
w h ere x is th e d eflec tio n of th e p o in t from a ce rta in p o sitio n

an d k is th e r ig id ity of th e e la s tic S ystem , for in sta n c e , a sp rin g
(a car sp rin g ), th e force of ré sista n ce to m o tio n is p ro p o rtio n a l
(th e p r o p o r tio n a lity c o n sta n t is X) to th e first
^ L
p ow er of th e v e lo c it y , an d fr (/) is th e o u te r (or
d istu r b in g ) force.
E q u a tio n s of ty p e (1) d escrib e sm a ll v ib r a
t io n s of oth er m ec h a n ica l System s w ith o n e deg-
ree of freed om , for e x a m p le , th e to rsio n a l v ib
l~W r a tio n s of a flyw h eel on an e la s tic sh a ft, if x is
Fig. 150 th e a n g le of r o ta tio n of th e fly w h eel, m is th e
m o m en t o f in er tia of th e fly w h eel, k is th e
to r sio n a l r ig id ity of th e sh a ft, an d (/) is th e m o m en t of th e
o u te r forces r e la tiv e t o th e a x is of r o ta tio n . E q u a tio n s o f ty p e
(1) d escrib e n o t o n ly m ec h a n ica l v ib r a tio n s b u t a lso p h en o m en a
th a t occu r in e le c tr ic c ir c u its.
Suppose we hâve an electric circuit consisting of an inductance L,
a résistance R and a capacitance C, to which is applied an emf E
(Fig.150). We dénoté by i the current in the circuit, by Q the
charge of the capacitor; then, as we know from electrical engi
neering, i and Q satisfy the following équations:
L àt + I*i J r % = E (2)
(3)
From (3) we get
d*Q _ di
dt2 di (3')
Substituting (3) and (3') into (2), we get for Q an équation of

type ( 1):
* See, for example, Sec. 1.26, where such an équation is derived in consi-
dering the vibration of a weight on a car spring.
7.15 Solution of the Differential Equation of Oscillât ions 433
Differentiating both sides of (2) and utilizing (3), we obtain an

équation for determining the current i:
L dP + * dt + C 1 ~ dt ' 5'
Equations (4) and (5) are type ( 1) équations.
7.15 SOLUTION OF THE DIFFERENTIAL EQUATION

OF OSCILLATIONS
Let us write the oscillation équation in the form
where the mechanical and physical meaning of the desired func-

tion x, of the coefficients a,, a2 and of the function f (t) isreadily
established by comparing this équation with équations (1), (4),
(5) of the preceding section. Let us find the solution to équation
( 1) that satisfies the initial conditions x = x0, x' — x'B when ( = 0.
We form the auxiliary équation for équation (1):
x (P ) (P2 + a j -f a2) = x0p+x'0+ axx 0 + F (p) (2)
where F (p) is the transform of the function f(t). From (2) we find
~f i X qP + x +
’o ^ x q F( p )
pi + cup+ Oi “r p2 + a1p+a. (3)
Thus, for a solution Q(t) of équation (4), Sec. 7.14, that satisfies
the initial conditions Q = Q0, Q' = Q'0 when f = 0 , the transform
will hâve the form
q ^p ^ _ L ( Q 0p - | - Q o ) - | - / ? Q o _|________ E(p)
Lp2 + /?p+-i Lp2+/?p+l
The type of solution is significantly dépendent un whether the roots

of the trinomial p1+ axp -f-a2 are complex, or real and distinct,
or real and equal. Let us examine in detail the case where the
roots of the trinomial are complex, that is, when —a2 < 0 .
The other cases are considered in similar fashion.
Since the transform of a sum of two functions is equal to the
sum of their transforms, it follows from formula (2), Sec. 7.11,
that the original function for the first fraction on the right of (3)
28—2082
434 Ch. 7 Operational Calculus and Certain of Its Applications
will hâve the form

r
ûi
*oP~l-*o4~fli*o •
P2+ ai P + ûa
x{) cos t
]Â~T
*°H— 2~
+ sin t
v <*- (4)
Y ^4 4 J
Let us then find the original function corresponding to the fraction
F (p)
p3+ aiP +Oa
Here, we take advantage of the convolution theorem, first noting
that
l :sin
p2+ fliP+<Jï '
Y al
Hence, from ( 1), Sec. 7.13, we get

t
F (P) 1
P 2 + û iP + a a J /(*)* 2 (<’ T)sin(<—t ) ] / rflt_ ^ d T (5)
And so, from (3), taking into account (4) and (5), we get
r J_V i
—t>-%-t
2 , i/ al . 2 . , i /r oî
x (t) =e je, cos/ |/ û2— Y H----- . sin t y a2— j-
lV/ aa*— 4"
■ y- 1 — r f / ( * ) « ' 2 U T,sin(<—T) )/ a2— ( 6)

1/ „ «i J
r —r °
If the external force f(t) = 0, which means that if we hâve free
mechanical or electrical oscillations, then the solution is given by
the first term on the right-hand side of expression (6 ). If the ini
tial data are equal to zéro, i.e., if x0 = xj, = 0 , then the solution
is given by the second term on the right side of (6 ). Let us con-
sider these cases in more detail.
7.16 Investigating Free Oscillations 435
7.16 INVESTIGATING FREE OSCILLATIONS
Let équation (1) of the preceding section describe free oscillations,

that is, f(t) = 0. For convenience in wilting we introduce the
notation al = 2n, aî = k2, k\ — k2—n2. Then it will hâve the form
S - + 2n | + ^ = 0 ( 1)
The solution of this équation xIr that satisfies the initial condi
tions x = x„, x' = x„ for t = 0 is given by the formula (4), Sec. 7.15,
or by the first term of (6 ):
xfr (t) = e~ nt j^JCocos k j + x°n sin k j J (2)
We dénoté x0—a, x- ~ ^ n-=rb. It is obvious that for any a and b

we can select A4 and 6 such that the following equalities will be
fulfilled:
a = A4 sin 6 , b = M cosô
here,
M 2= a2+ b2, tan 6 = -|-
We rewrite formula (2 ) as
xfr —e~nt [M cosk j sin ô + A4 sinÆ^ cos 6 ]
or, in final form, the solution may be written thus:
x/r = V a2 -j-b2e~ntsin (k^+6) (3)
Solution (3) corresponds to damped oscillations.
If 2n = a1— 0, that is, if there is no internai friction, then the
solution will be of the form
xlr — y a2 + 6 2sin ( k j -f-ô)
In this case harmonie oscillations occur. (In Sec. 1.27, Fig. 27
and 29 give graphs of harmonie and damped oscillations.)
7.17 INVESTIGATING MECHANICAL

AND ELECTRICAL OSCILLATIONS IN THE CASE
OF A PERIODIC EXTERNAL FORCE
When studying elastic vibrations of mechanical Systems and, in

particular, when studying electrical oscillations, one has to considér
different types of extemal force f(t). Let us considér in detail the
case of a periodic extemal force. Let équation ( 1), Sec. 7.15, hâve
the form
+ 2n Tt +■k%x = A sin (ù/ O)
To détermine the nature of the motion it is sufficient to consider
the case when xn= x'o= 0. One could obtain the solution of the
équation by formula (6 ), Sec. 7.15, but we obtain the solution by
carrying out ail the intermediate calculations.
Let us write the transform équation
x(p)(p2+ 2np + k2) = A P2+“ (t>2
from which we get
x(p) A(ù
(p* + 2np + k*) (p2+(o2) ( 2)
We consider the case when 2n=F0 (n2 < k2). Décomposé the frac
tion on the right into partial fractions:
Am Np + B Cp+ D
(p2+ 2 n p -f-k2) (p*-|-(o2) p2 2 np -\-k2
' p2+ co2 (3)
We détermine the constants fi, C, Dy N by the method of unde-

termined coefficients. Using formula (2 ), Sec. 7.11, and (2) of this
section, we find the original function:
x(t) = (k 2 —û)2)2+ 4n2(ù^ | (^2—<•>*) sin lot — 2 ncû cos iot

+ e”" ^ ( 2 n 2—k2+ to2) sin k1t+2rm cos kxt^ | (4)
here again, kx = V k2—n2. This is the solution of équation (1) that

satisfies the initial conditions x{) = x'o= 0 when t= 0.
Let us consider a spécial case when 2n = 0. This corresponds to
a mechanical System with no internai résistance (no damper), or to
an electric circuit where R = 0 (no internai résistance in the cir
cuit). Equation ( 1) then takes the form
d2x
dt2
-\-k2x ~ ,4 sin g)/ (5)
and we get the solution of this équation satisfying the conditions

*o = *o=^0 for / = 0 if in (4) we put n = 0:
x (t') = (k2-< ù2) k [ ~ ' sin & + * sin of] (6)

Here we hâve the sum of two harmonie oscillations: natural oscilla
tions with frequency k :
A (ù
X„t — k 2—(ù2 k sin kt
7.18 Solving the Oscillation Equation in the Case of Résonance 437
and forced oscillations with frequency co:
The type of oscillations for the case k^>a> is shown in Fig. 151.
Let us again return to formula (4). If 2n > 0 (which occurs in
the mechanical and electrical Systems un- xai
der considération), then the term containing
the factor e~nt, which represents. damped
natural oscillations, rapidly decreases for I I i I |'
I I I T • .
increasing t. For t sufficiently large, the I1 11I I
character of the oscillations will be deter- I I I h |!
mined by the term that does not contain 11i ! ! ! 1 M !
the factor e~nt; that is, by the term
i ! i ! ! 1 1 1 1l 1i
| 1 | I I ! I
x(0i i 1 1 1 11 M
* <*>= (*«-«,*)* + 4^ {(*»-«>«) sin o,/
— 2 niocoso)/} (7)
We introduce the notation Fig. 151
4(*»-<o2) =M cos 6 , A-2ruù ,, . c
(k1
(V — or8)2+ 4n2(û* (k*- -- rr.,
-to2)2+ 4n2co2
= M Sin 0 ( 8)
where
M-
V (k* —(o2)2+ 4n2(o2
The solution (7) may be rewritten as follows:
x(t) sin (û)f-|-6 ) (9)
From formula (9) it follows that the frequency k of forced oscil

lations does not coincide with the frequency co of the external force.
If the internai résistance, characterized by the number n, is small
and the frequency co is not very different from k, then the amplitude
of oscillations may be made as great as one pleases, since the de-
nominator may be arbitrarily small. For n = 0, (0*= ^*, the solution
is not expressed by formula (9).
7.18 SOLVING THE OSCILLATION EQUATION

IN THE CASE OF RESONANCE
Let us consider the spécial case when al = 2n=0, that is, when
there is no résistance and the frequency of the external force coïn
cides with that of the natural oscillations, Æ= co. The équation
438 Ch. 7 Operational Calculus and Certain of lis Applications
then takes the form

j£-{- kîx = A sin kt ( 1)
We shall seek the solution that satisfies the initial conditions

x0= 0, = 0 for ? = The auxiliary équation will be
x(p)(pt + ki) = A - ^ Ti
whence
Ak
x (p) ~ w + w ( 2)
We hâve a proper rational fraction of type IV, which we hâve not

considered in the general form. To find the original function for
the transform (2 ), we take advantage of the following procedure.
We write the identity (formula 2 of Table of Transforms, Sec. 7.9)
CD
—1 . ^ = ^ e-pt Sin ktdt (3)

o
We differentiate both sides of this équation with respect to k (the
intégral on the right may be represented in the form of a sum of
two intégrais of a real variable, each of which dépends on the
parameter k):
__ !______—— — Çe-pt( cos
(p*+**j* J ‘ tosK tat
o
Utilizing (3) we can rewrite this équation as
00
= [ * C0S ^ “ T s i n k t ] d t
o
Whence it follows directly that
(p2 + ft2)2 T->2ft ( x 31" k t ~ tC0S

(from this formula we obtain formula 13 of the table). Thus, the
Solution of équation ( 1) will be
x(t) = £ ( ± s \ n k t — tc os k t) (4)
Let us study the second term of this équation:
xs (/) = — tcoskt (4 ')
This quantity is not bounded as t increases. The amplitude of
oscillations that correspond to formula (4') increases without bound
7.19 The Delay Theorem 439
as t increases without bound. Hence, the amplitude of oscillations

corresponding to formula (4) also increases without bound. This is
résonance; it occurs when the frequency of the natural oscillations
coïncides with that of the external force (see also Sec. 1.28, Figs.
30, 31).
7.19 THE DELAY THEOREM
Let a function f(t), for * < 0, be identically equal to zéro

(Fig. 152, a). Then the function f (t — /„) will be identically zéro
for t < t0 (Fig. 152,6). We shall prove a theorem which is known
as the delay theorem.
Theorem. If F (p) is the iransform of the function f (t), then e~P*« F (p)
is the transform of the function f( t — /„); that is, if f (t) F (p),
then
f( t — e~ptoF (p) (1)
Proof. By the définition of a transform we hâve
ao
L {f{t-t,)}= \e-P tf { t - Q d t
0
t0 «
= \e~r ‘ f ( t - t j dt + jj e~pt f ( t —10) dt
o <•
The first intégral on the right of the équation iszerosincef (tf —10) = 0
for t < t0. In the last intégral we change the variable, putting
t —t0— z:
oc oo
L{f (t — t0)}= J e~p (*+<o)/ (z) dz = e~p1» § e~P* f (z) dz=e~P,«F (p)
o o
Thus, f( t — t ^ - ^ e - P ^ F (p).
Example. In Sec. 7.2 it was established for the Heaviside unit function that
It follows, from the theorem that has just been proved, that for the function
a0 ( t — h) depicted in Fig. 153, the L-transform is
— e~Ph
that is,
(2)
7.20 THE DELTA FUNCTION AND ITS TRANSFORM
We consider the function

0 for t < 0
J_
h ) ^ T [a0(t) — aa(t — h.)] = h for Os^t < h O)
0 for h ^ t
depicted in Fig. 154.
If this function is interpreted as a force acting over a time
interval from 0 to h and being zéro the rest of the time, then,
clearly, the impulse of this force
will be equal to unity. MW
60(t-h) i
h
h 0 i 1
Fig. 153 Fig. 154
By formulas ( 1), Sec. 7.2, and (2 ) of the preceding section, the

transform of this function is
that is,
( 2)
In mechanics it is convenient to regard forces acting over very

brief time intervals as forces acting instantaneously but having
a finite impulse. For this reason, the function ô(/) is introduced
as the limit of the function ol (t, h) when h->- 0 :
(/) = lim (Jj (t , h)
6 (3)
h-*0
This function is called the unit impulse function, or the delta
function. (It should be noted that ô(/) is not a function in the
7.20 The Delta Function and Its Transform 441
ordinary meaning of the word; physicists often call ô (t) the Dirac
function or the Dirac delta function.)
It is natural to put
oo
S 1 (4)
—oe
One also writes
o
$ ô (/)d / = 1 (5)
o
The function 6(x) finds application not only in mechanics but
in many divisions of mathematics, in particular in the solution of
numerous problems involving the équations of mathematical physics.
Let us consider the action of 6 (t) if it is represented as a force.
We find the solution to the équation
S = ô(0 (6 )
that satisfies the conditions: s = 0, •^■ = 0. when / = 0. From (6 ) we

find, taking account of (5),
t
v= § = ^(T )d x = l (7)
0
for any t , in particular for t — 0 . Hence, by defining 6 (a:) by the
équation (3), we can interpret this function as a force imparting
to a unit mass at time / = 0 a velocity equal to unity.
We define the L-transform of the function 6 (t) as the limit of
the transform of the function ox(/, h) as h —>-0 :
L {6 (jc)} = lim — • -— ■P= 1
* w ' h-o P h p
(here we made use of l’Hospital’s rule for finding a limit). Thus,

6 (0 ^ - 1 (8 )
We then define the function ô(/ — /„), which is interpreted as a
force that instantaneously, at time t — t0, imparts a unit velocity
to a unit mass. Obviously, on the basis of the delay theorem we
will hâve
à V - t J + e -# (9)
As in the case of (5), we can write
to
5 « ( / - / 0)<tt=i ( 10)
to
442 Ch. 7 Operational Calcul us and Certain of fis Applications
From the mechanical interprétation of the delta function it fol-

lows that the presence of the delta function in the right member
of the équation can be replaced by an appropriate change in the
initial conditions. We will illustrate this by a simple example.
Suppose we hâve a differential équation
| r = / ( 0 + ô (0 ( 11)
with initial conditions x0 = 0 , *ô = 0 when

f = 0. The auxiliary équation is
p*x{p) = F(p)+ 1 ( 12 )
whence
^ ) = 7 + ?
Using formulas 9 and 15 of the table, we get
x(t) = \f( T )(t — T)dx + t (13)

We would arrive at the same resuit if we found the solution to
the équation
$ = /( ')
with initial conditions *0 = 0, x’0= l , when ^ = 0. In this case the
auxiliary équation would hâve the form
pix(p )— l= F (p ) (14)
It is équivalent to the auxiliary équation (12) and, hence, the
solution will coïncide with the solution (13).
In conclusion, we note the following important property of the
delta function. On the basis of (4) and (5), we can Write
Ç ô (T )d T = { ^ [°r < *< ° (15)

J» .
— 1v 1 for 0 < t < oo ' '
In other words, this intégral is equal to the Heaviside unit function
MO- Thus, ,
M 0 = —S00 ô(T)dT (16)
Differentiating the right and left sides of the équation with respect
to t, we get the conditional équation
o; (0 = 6(0 (17)
To get at the meaning of (17), let us examine the function
o# ( 0 h) shown in Fig. 155. It is clear that
ô ’o(t, h) = al (/, h) (18)
(with theexception of the points ( = 0 and t — h). Passing to the

limit as h —* 0 inéquation (18), we see that cr0 (t, h )—*aa(t) and
we write •
h) —►a'0 (t) as h —<-0 -
The right member of (18) o1(t, h)—>-ô(7) as h —*0. Thus, équation
(18) passes into the conditional équation (17).
Find solutions to the following équations for the indicated initial conditions:
W
2y Wy
1. - jto + 3 tt4-2jc = 0, x = \, x' = 2 for t = 0. Ans. x = 4e~t —3e~2t.
a tL dt
H*r S x
2. — ——= 0 , x = 2, x f = 0, x" = 1 for t = 0. Ans. x = \ —
dt3 d t2
3. 2 a ^ + ( a * + t»2)x = 0, x = xc, x '= x '0 for / = 0 .
Ans. x = ~y \xob cos bt + (*o —*ofl) sin bt ] .

Hx 1 1 2
4. T+2x = e5t>x==l>x' = 2 fort = 0- Ans• x= ï2 ebt+ T et+T e2t-
d"X *
5. - ^ - \ - m 2x = a cos nt, x = x Qt x' = x0 for i = 0.
Ans. x = —ê——if cos nt — cos m t)4 -x0 cos m / + — sin m t.

m2—n2 1m
6. X = 0 , x' = 0 for 7 = 0. Ans. * = 2e<— 1 -7 » -7 8- 2 / - 2 .
d t£ dt 6
Sx 1
7. — 4 -jc = — t 2el , x = x' = x ”= 0 for t = 0.
dt3 2
^ — K « --« h 4 ) ' - à ' - ' - K - T 1 - ^

o - 7 t t +i * = i1, *0 = *o' = *o' = iw
8- 0 for tt = n0. Aü n s.x = ,1 - r 1- r '<— 2r- e— / -----
2 cos —
dt* 6 6 2
dxx d2x * „
9 . - ^ — 2*^r + j f = s i n / , *0 = *o = *o = *o = 0 f o r / = 0 .
>l/w. jc = i - [ ^ ( ^ - 2 ) + e -^ (/ + 2) + 2sin t}.

10. Find solutions to the System'of differential équations
d2jc . f d2y . A
•5 ^ + y - 1- •5 72 + -* -°
that satisfy the initial conditions x0 = y0 — x'0 — yo = 0 for / = 0.
■Ans. x(t) = —ÿ c o s +
Jf(0 = —y c o s / - - - e < - i - e - ' + l .

CHAPTER 8
E L E M E N T S O F T H E T H E O R Y OF P R OB A B I L I TY
A N D M A TH EM A TICA L STAT1STICS
Our daily expérience in ordinary life, in practical situations, and

also in scientific investigations constantly supplies us with examples
of the breakdown of the familiar regularities of strict determinism
that we are accustomed to. For instance, suppose we wish to know
how many téléphoné calls a first-aid station will receive within
twenty four hours.
Long-term observations indicate that there is noway of predicting
the number of such calls. This number is subject to appréciable
(and, what is more, random) fluctuations. Likewise, the time the
doctor will hâve to spend with the patient in each instance is quite
random.
Suppose we want to test N machine parts, ail items having been
manufactured under identical conditions and out of the same ma-
terials. The time from start of testing to their breakdown (failure)
turns out to be a random quantity subject to an extremely broad
spread of values.
In shooting at a target we hâve what is called shell dispersion.
The departure of the point of impact of a shell from the centre
of the target cannot be predicted because it is a random quan
tity.
It is not sufficient merely to indicate the fact of randomness in
order to make use of a particular phenomenon of nature or to control
a technological process. We hâve to learn to evaluate random events
numerically and predict the course they will take. Such, at the
présent time, are the insistent demands of theoretical and practical
problems. Two divisions of mathematics are engaged in the solution
of such problems and in constructing the requisite general mathe-
matical theory: they are the theory of probability and mathema-
tical statistics.
In this chapter we deal with the basic essentials of probability
theory and mathematical statistics.
8 .! The Subject of Probability Theory 445
8.1 RANDOM EVENT.

RELATIVE FREQUENCY OF A RANDOM EVENT.
THE PROBABILITY O f AN EVENT.
THE SUBJECT OF PROBABILITY THEORY
The basic concept of probability theory is that of a random

(chance) event. A random event is an event which may occur or
fail to occur under the realization of a certain set of conditions.
Example 1. In coin tossing, the occurrence of heads is a random event.
Example 2. In firing at a target from a particular gun, hitting the target or
a given area on it is a random event.
Example 3. In manufacturing a cylinder with a specified diameter of 20 cm,
errors less than 0 . 2 mm represent a random event for a given set of production
conditions.
Définition 1. The relative frequency p* of a random event A is
the ratio of the number m* of occurrences of the .given event to the
total number n* of identical trials, in each of which the given
event could occur or fail to occur. We will Write
p *(i4)= /> •= -£ - (n
Example 4. Suppose, under identical conditions, we fire 6 sequences of shots
at a given target:
in the first sequence there wcre 5 shots and 2 hits,
in the second sequence there were 10 shots and 6 hits,
in the third sequence there were 12 shots and 7 hits,
in the fourth sequence there were 50 shots and 27 hits,
in the fifth sequence there were 100 shots and 49 hits,
in the sixth sequence there were 2 0 0 shots and 102 hits.
Event A is a hit. The relative frequency of hits in the sequences will be
2
first, 4 = 0 - 4 0 ,
second.
re“ 0 6 0 ’
third, ^ - 0 .5 8 .
27
fourth,
® r 054-
fifth,
T O " 0-49’
sixth,
» = 0-5'-
From observations of a variety of phenomena, we can conclude
that if the number of trials in each sequence is small, then the
relative frequencies of the occurrence of event A in the different
sequences can differ substantially from one another. However, if
446 Ch. 8 The Theory of Probability
the number of trials in the sequences is great, then, as a rule, the

relative frequencies of the occurrence of event A in different sequ
ences will differ but slightly, and the différence is the smaller, the
greater the number of trials in the sequences. We say that the rela
tive frequency in a large number of trials ceases more and more to be
accidentai (of a random nature). However, it must be noted that
there are events the relative frequency of which is not stable and
may expérience considérable fluctuations even in sequences with a
large number of trials.
Experiments show that in most cases there is a constant p such
that the relative frequencies of occurrence of an event A, given a
large number of trials, differ but slightly from p, except in rare cases.
This experimental fact is symbolized as follows:
The number p is called the probability of occurrence of a ran

dom event A. This statement is symbolized as
P(A) = p (3)
The probability p is an objective characteristic of the possibi-
lity of occurrence of event A under given trials. It is determined by
the nature of A.
Given a large number of trials, the relative frequency difFers
very slightly from the probability, except in rare cases, which may
be ignored.
Relation (2) can be formulated briefly as follows.
I f the number n* of trials is increased without bound, the re
lative frequency of event A converges to the probability p of occur
rence of the event.
Note. In the foregoing, we postulated relation (2) on the basis
of experiments. But other natural conditions that follow from ex-
periment hâve been postulated. From them relation (2) is derived,
which then becomes a theorem. This is the familiar theorem of
probability theory that bears the name of Jakob Bernoulli (1654—
1705).
Since probability is an objective characteristic of the possibility
of occurrence of a certain event, to predict the course of numerous
processes that one has to consider in military affairs, in the or-
ganization of production, in économie situations, etc., it is neces-
sary to be able to détermine the probability of occurrence of cer
tain compound events. Determining the probability of occurrence
of an event on the basis of the probabilities of the elementary
events governing the given compound event, and the study of
probabilistic regularities of various random events constitute the
subject of the theory of probability.
8.2 Définition of Probability and the Calculation of Probabilities 447
8.2 THE CLASSICAL DEFINITION OF PROBABILITY

AND THE CALCULATION OF PROBABILITIES
In many cases it is possible to calcfllate the, probability of a

random event by proceeding from an analysis of the trial. An
example will help to illustrate this idea.
Example 1. A homogeneous cube with faces labelled 1 to 6 is called a die.
We will consider the random event of the occurrence of a number / ( l < ; / < : 6 )
on the upper face for each throw of the die. By virtue of th.e symmetry of the
die, the events (the appearance of any number from 1 to 6) are equally probable.
Hence they are called equally probable events. Given a large number of throws,
n, it can be expected that the number / (and any other number from 1 to 6)
will turn up in roughly n/6 cases. Experiments corroborate this fact.
The relative frequency will be close to the number p * = l/6 . It is thereforc
considered that the probability of the number / ( l ^ / ^ 6 ) tuming up is equal
to 1/6.
Below, we will analyze random events whose probability can be
calculated directly.
Définition 1. Random events in a given trial are called d i s j o i n t
( m u tu a lly exclusive) if no two can occur at the same time.
Définition 2 . We will say that random events form a c om plété
g r o u p if in each trial any one of them can occur but no disjoint
event can occur.
We consider a complété group of equally probable disjoint ran
dom events. We give the name cases to such events.
An event (case) of such a group is termed favou rable to the oc
currence of event A if the occurrence of the case implies the oc
currence of A .
Example 2. We hâve 8 balls in an urn. Each bail is numbered from 1 to 8.
Balls labelled 1, 2, 3 are red, the others are black. The occurrence of abail
label led 1 (or 2 or 3) isan event favourable to the occurrence of a redbail.
For this case, we can give a définition of probability that differs
from that given in Sec. 8.1.
Définition 3. The p r o b a b ility p of event A is the ratio of the
number m of favourable cases to the number n of ail possible cases
forming a complété group of equally probable disjoint events, or,
symbol ically,
= 0 )
Définition 4. If relative to some event there are n favourable
cases forming a complété group of equally probable disjoint events,
then such an event is called a c e r ta in event. A certain event has
probability p = 1.
If not a single one of n cases forming a complété group of equa
lly probable disjoint events is favourable to an event, then it is
termed an im possible event and has probability p = 0.
Note 1. T he con v erse assertio n s a lso hold true here. In other

ca ses, h ow ever, for in sta n ce in the case of a c o n tin u o n s random v a ri
a b le (Sec. 8 .1 2 ), the con verse a ssertio n s m ay n ot hold true; in
oth er w ord s, from th e fact th a t the p r o b a b ility of som e e v e n t is
eq u a l to 1 or 0, it d oes n o t y e t fo llo w th a t th is ev e n t is ce rta in
or im p o ssib le .
F rom th e d éfin itio n of p r o b a b ility it fo llo w s th a t th e re la tio n
1
h o ld s tru e.
Example 3. One card is drawn from a deck of 36 cards. What is the proba
bility of drawing a spade?
Solution. This is a scheme of cases. Event A is the occurrence of a spade.
We hâve a total of n = 36 cases. The number of cases favouring A is m = 9.
Conséquently, p = ^ = l - .
Example 4. Two coins are tossed at the same time. What is the probability
of getting 2 heads?
Solution. Let us compile a scheme of ail possible cases.
First coin Second coin
lst case head head

2nd case head tail
3rd case ta il head
4th case tail tail
There are 4 cases in ail, of which one is favourable. Hence, the probability
of obtaining two heads is
P= 4
g
Example 5. The probability of hitting a target from one gun is , from
7
another gun, . Find the probability of destroying the target in a simultaneous
firing from both guns. The target will be destroyed if at least one of the guns
makes a hit.
Solution. This problem can be modelled as follows. Two urns contain 10
balls each, numbered from 1 to 10, with 8 red and 2 black in the first urn,
and 7 red and 3 black in the second. One bail is drawn from each urn. What
is the probability that at least one of 2 drawn balls will be red?
Since each bail of the first urn can be drawn with any bail of the second
urn, there will be a total of 100 cases: n= 100.
Let us calculate the favourable cases.
Wtien each of the 8 red balls of the first urn is drawn together with any
bail of the second urn, we will hâve at least one red bail, drawn. There will
be 10x8 = 80 such cases. Drawing each of the two black balls of the first urn
together with any one of the 7 red balls of the second urn gives us one red
bail. There will be 2 x 7 = 1 4 such cases. This makes a total of m = 80+ 14 = 94
favourable cases.
8.3 The Addition of Probabilities. Complémentary Random Events 449
The probability that there will be at least one red bail from among those
drawn is
m 94
n 100 •
Such also is the probability of destroying the target.
Note 2 . In this example, we reduced the problem of probabi

lity in gunfire to a problem of the probability of a given bail
being drawn from an urn. Many problems in probability theory
can be reduced to the “urn model”. This permits us to regard urn-
model problems (drawing balls from urns) as generalized problems.
Example 6. Ten items out of a set of 100 are defective. What is the proba
bility that 3 out of any 4 chosen items will not be defective?
Solution. Four items out of 100 can be chosen in the following number of
ways: n = C}0o* The number of cases where 3 out of 4 items are nondefective
is equal to m = C20*C}0-
The desired probability is
m _ Cf!o • C jp 1424
0.3
" Cioo *753
8.3 THE ADDITION OF PROBABILITIES.

COMPLEMENTARY RANDOM EVENTS
Définition 1. The logical sum (union) of two events A1 and A2

is an event C consisting in the occurrence of at least one of the
events.
Below we consider the probability of the union of two disjoint
events Ax and A2. The union of these events is denoted by
+ A2
or
A t or A *
The following theorem, which is called the theorem on the ad
dition of probabilities, holds true.
Theorem 1 . Suppose, in a given trial (phenomenon, experiment),
a random event A1 can occurwith probability P (AJ and an event Â2
with probability P (/l2). The events A x and A2 are exclusive. Then
the probability of the union of the events, that is, the probability
that eitker event A x or event A2 will take place, is computed from
the formula
P (A, or i4#) = P(.41) + P(i4#) (1)
♦ Note here that the "or” is inclusive and signifies the occurrence of at least
one of the events, in accord with Définition 1.
2 9 -2 0 8 2
Proof. Suppose
p (4 ) = - . p(A )=-
Since the events A 1 and A2 are exclusive, it follows that for a
total number n of cases, the number of cases favouring the occur
rence of Ax and A2 together is equal to 0 , and the number of ca
ses favouring the occurrence of Ax or Â2 is equal to m1+ m2.
Consequently,
P {A, or A2) = ^ ^ ^ + ^ = P( Al) + P(At)
and the proof is complété.
The proof of this theorem is the same for any number of terms:
P (A or A2 or .. .or An) = P (/!,) + P (4,) + . . . + P {An) (V)
This équation may be written thus:
p (i//)= J iP M /) (O
Note. We proved the addition theorem for a scheme of cases

where the probability is determined by direct computation. In the
sequel we will assume that the addition
theorem holds true also for the case where
direct computation of probabilités is im
possible. This assertion is based on the
following reasoning. For large numbers of
trials, the probabilities of events are (with
rare exceptions) close to the relative fre-
quencies, and for the latter the proof is the
same as that given above. This remark
will hold true also in the proof of subsé
quent theorems that we will prove by
means of the urn scheme.
Example I. Shots are fired at a certain domain D consisting of three non-
5
overlapping zones (Fig. 156). The probability of hitting Zone f is P ^ ^ y ^ ,
Zone II, P ( i42) = - j ^ , and Zone IJI, P(y43) = ~ . What is the probability of
hitting D? Event A is a hit scored in domain D. By formula (T) we hâve
p M , ) + p M * )+ p ('43 ) = T g g + î ^ + i ^ = ^
Définition 2. Two events are called complementary events if they

are exclusive and form a complété group.
8.3 The Addition of Probabilities. Complementary Random Events 451
If one event is denoted by A, the complément (complementary

event) is denoted by A.
Let the probability of the occurrence of event A be p, the pro-
bability of the nonoccurrence of event that is, the probability
of the occurrence of event A, be P(A) = q.
On trial, either A or À will occur, therefore Theorem 1 gives
P(X) + P ( J ) = 1
That is, the union of the probabilities of complementary events is
equal to unit y:
P+q= i (2)
Example 2. A shot is fired at a target. Event A represents a hit. The proba
bility p of a hit is P (A) = p. Détermine the probability of a miss. A miss is
given by event A, the complément of A. The probability of a miss is thus
<7=1—P-
Example 3. A measurement is made. Event A will dénoté an error less
tlian X. Let P (A)=--p. The complementary event is an error exceeding X or
equal to X, and it is denoted by event A. The probability of this event is
P(Â) = < 7 = l - p .
Corollary 1 . I f random events A lt A........ . An forma complété
group of exclusive events, then the following équation holds true:
P ( i 4 1) + P ( i 4 1) + . . . + P ( i 4 li) = l ( 3)
Proof. Since the events A ly A2t . . . , An form a complété group

of events, the occurrence of one of them is a certain event. Conse-
quently,
P (A, or A2 or . . . or A n) = 1
Transforming the left member by formula (T), we get (3).
Définition 3. Random events A and B are called compatible if
in a given trial both events can occur, which is to say we hâve
a logical product (intersection) of events A and B .
The event which consists in the intersection of A and B is de
noted by (A and B) or (AB). The probability of the intersection
of events A and B will be denoted by P (A and B).
Theorem 2 . The probability of the union of compatible events is
computed from the formula
P(A or B) = P (t4)-(-P (B) —P (A and B) (4)
The truth of formula (4) can be illustrated geometrically. We
first give the following définition.
Définition 4. Given a certain domain D with area 5. Consider
a subdomain d of D. Let 5 be the area of d. Then the probabi
lity of a point falling in d (the falling of a point in D is taken
29 *
to be a certain event) is, by définition, 5/5, or /j = 5/5. This is

called géométrie probability.
Then, assuming as certain the falling of a point in a square
with unit side, we hâve (Fig. 157):
P (A or fl) = area abeda,
P (A) =area abfda,
P (S) = area bedeb,
P (A andB) — area debfd
We clearly hâve the équation
area abeda — area abfda
+ area bedeb—area debfd
Putting into this équation the left mem-
Fig. 157 bers of (5), we get (4). In similar fashion
we can compute the probability of the
union of any number of compatible random events.
Note that Theorem 2 can be proved by proceeding from the fo-
regoing définitions and rules of the operations.
8.4 MULTIPLICATION OF PROBABILITES

OF INDEPENDENT EVENTS
Définition I. An event A is said to be independent of B if the

probability of occurrence of A does not dépend on whether event B
took place or not.
Theorem 1. If random events A and B are independent, then the
probability of the intersection of events A and B is equal to the
product of the probabilities of occurrence of A and B:
P (A and B )-= P (A) •P (fl) (1)
Proof. Let us carry out the proof by the urn scheme. Each of
two urns has, respectively, nl and n2 balls. There are ml red balls
in the first urn, and the remaining balls are black. There are m2
red balls in the second urn, ail others being black. One bail is
drawn from each urn. What is the probability that both balls will
be red?
Let event A be a drawing of a red bail from the first urn. Event B
will be a drawing of a red bail from the second urn. These events
are independent. Clearly,
P(A) = ^/ii , P ( B ) -n2
^ (2)
Altogether, there will be nin2 possible cases of simultaneous

drawing of one bail from each urn. The number of cases favouring
8.4 Multiplication of Probabilities of Independent Events 453
red balls being drawn from both urns will be mlm2. The probabi-
lity of the intersection of events A and B will be
P (A and 7 = nxn 2 n{ m2
V 7
If in this formula we replace and —- by their expressions
in (2 ), we get ( 1). The theorem is illustrated in Fig. 158.
ni
Fig. 158
If we hâve n independent events At, A2, . . . , A„, then in simi-

lar fashion we can prove the validity of the équation
P (A, and At and . . . and ^l,I) = P(>4,)-P(yiJ) - . . . -P(4„) (3)
Example 1. Two tanks are firing at one and the same target. Tank One has
9 5
a probability of — of hitting the target, Tank Two, a probability of — . One
shot is fired from each tank at the same time. Détermine the probability tliât
two hits will be scored.
9 5
Solution. Here, P(y4)=^ — , P (£) = — ; P (A and B) is the probability of two
hits and is found from formula (1):
P M and ô ) = ^ . | = 4
Example 2 . No-failure operation of a device is determined by trouble-free

operation of each of three component units. The probabilities of no-failure ope
ration of the units during a certain cycle are to pt ^= 0.6, p2 = 0.7, p.z = 0.9.
Find the probability that the device will not break down during the indicated
operation cycle.
Solution. By Theorem (3) on the multiplication of probabilities, we hâve
p-= px. p2. p3 = 0.6 x 0.7 x 0.9 = 0.378
Note. Theorem 2, Sec. 8.3 [formula (4)J, on the probability of
the union of compatible events, with regard for formula ( 1) of this
section, is then written as
P M or B) = P(A) + P(B) — P (A)-P (B) (4)
Example 3. Solve the problem given in Example 5, Sec. 8.2, using for
mula (4).
Solution. Event A is a hit scored by the first gun. Event B is a hit scored
by the second gun. It is obvious that
P ( A) - W' P(B) = ÏÔ’

n / A rn 8 , 7 8 7 94
P (A or fl) = ï ô + Tô _ î ô ' ÏÔ = ÎÔÔ
We naturally obtained the same resuit as before.
Example 4. The probability of destroying a target in one shut is equal to p.
Détermine the number n of shots needed to destroy the target with probability
greater than or equal to Q.
Solution. By the theorems on the union and intersection of probabilities we
can Write
Q ^ l-(l-p )"
Solving this inequality for n, we get
■^ lo g io d - Q )
" lo g io (1 - P )
A problem with this kind of analytic solution can readily be stated in terms
of the urn model.
8.5 DEPENDENT EVENTS.
CONDITIONAL PROBABILITY. TOTAL PROBABILITY
Définition 1. Event A is said to be dépendent on event B if the

probability of occurrence of A dépends on whether B took place
or not.
The probability that event A occurred, provided that B took
place, will be denoted by P(A/B) and will be called the conditio-
nal probability of event A provided that B has occurred.
Example 1. An urn contains 3 white balls and 2 black balls. One bail is
drawn (first drawing) and then a second is drawn (second drawing). Event B is
the occurrence of a white bail in the first drawing, event A, the occurrence of
a white bail in the second drawing.
It is clear that the probability of A, provided B has occurred, is
p (^ ) = t = 4
The probability of event A, provided B has not taken place (a black bail
appears in the first drawing), will be
P(d/5)=-|-
We see that
P (A /£ )# P (A /B )
Theorem l. The probability of the intersection of two event s is
equal to the product (logical intersection) of the probability of one by
the conditional probability of the other computed on the condition
that the first event has taken place, that is,
P (A and B) = P(B) P(A/B) (n
8.5 Dépendent Events. Conditional Probability. Total Probability 455
------------------------V------------------------------------------------------------
Proof. We carry out the proof for events that reduce to the urn
model (that is, for the case vvhere the classical définition of pro
bability is applicable). •
An urn contains n balls, of which n 1 are white and n2 are
black. Suppose the nx white balls include / balls marked with
a star (Fig. 159).
n
L
Fig. 159
One bail is drawn from the urn. What is the probability of

drawing a marked white bail?
Let B be the event of a white bail being drawn, A the event
of a marked white bail being drawn. It is then clear that
P(fl)-? (2 )
The probability of a marked white bail being drawn, provided
that a white bail has already been drawn, is
P =- (3)
The probability of a marked white bail being drawn is P
(A and B). Obviously,
P (A and B) — -^ (4)
But
Substituting into (5) the left-hand members of (2 ), (3) and (4),

we get
P(i4 and B) = P (B)-P(A/B)
Equation (1) is proved.
If the events under considération do not fit the classical scheme,
then formula (1) serves to define conditional probability. Namely,
the conditional probability of A, provided event B has occurred, is
determined by means of the formula
p {A/B) = PMpa(g B) [for P (B) ^ 0]
Note 1. Let us apply this formula to the expression P (B and A):
P (B and A) = P (A)-P (B/A) (6 )
In équations (1) and (6 ), the left-hand members are equal, since

this is one and the same probability; hence, the right-hand members
are also equal. We can therefore write the équation
P (A and B) = P (B) •P ( AiB) = P ( A) •P ( B / A ) (7)
Example 2. For the case of Example 1, given at the beginning of this sec
tion we hâve
P<B)==|-, P(^/fl) = ±
P(A and Æ) = - ^ . y = = —
The probability P (A and B) can also be readily calculated directly.

Example 3. The probability of manufacturing a nondefective (acceptable)
item by a given machine is equal to 0.9. The probability of the occurrence of
qualitv articles of grade one among the nondefective items is 0.8. Détermine
the probability of turning out grade-one articles by this machine.
Solution. Évent B stands for manufacturing a nondefective article on the
given machine. Etent A stands for the occurrence of a grade-one item.
Here, P (B) = 0.9, P (/t/£ ) = 0.8.
Substituting into formula (1), we get the desired probability:
P ( A and £) = 0.9-0.8 = 0.72
T h eo rem 2. If euent A can be realized only when one of the
events Blt B 2t ...» B n , which form a complété group of exclusive
events, occurs, the probability of event A is computed from the formula
P ( A) = P ( Bj ) • P ( A l B , ) + P ( B 2) P (A/B2) + + P ( f l J P ( A / B n) (8 )
Formula (8 ) is called the formula of total probability.
Event A can occur if one of the compatible events
P ro o f.
( B x and A ), (B 2 and A ), . . . , (B n and A )
is realized. Consequently, by the theorem of addition of probabi
lités, we get
P ( A) = P (Bj and A ) + P ( B 2 and A ) + . . . + P ( B n and A)
Replacing the terms of the right side in accordance with for
mula ( 1), we get équation (8 ).
Example 4. Three shots are fired at a target in succession. The probability
of a hit in the first shot is px — 0.3, in the second, p2 = 0.6, in the third,
p3= 0.8. In the case of one hit, the probability of destroying the target is
^ = 0.4, in the case of two hits, a2 = 0.7, in the case of three hits, À,3= 1 .0 .
Détermine the probability of destroying the target in three shots (event A ) .
Solution. Let us consider the complété group of exclusive events:
Bx means one hit,
B%means two hits,
B3 mfcans three hits,
fî4 means no hits.
8.6 Probability of Causes. Bayes*s Formula 457
We détermine the probability of each event. One hit willoccur if: either
the first shot scores a hit and the second and third miss; or the first shot is
a miss, the second is a hit, and the third is a miss; or the first shot is a miss,
the second is a miss, and the third is a hit» Therefore, by the theorem of
multiplication arrd addition of probabilités, we get the following expression for
the probability of one hit:
P( Bl) = p1( \ - p 2) ( \ - p 3) + ( \ - p 1) p 2( l - p 3) + ( \ - p l) ( \ - p 2) p 9 = 0 M2
Reasoning in like fashion, we hâve
P (B2) = P1 P2 — Pn) ~r Pi 0 ~Pi ) P3 + 0 —Pi) P2 P3 — 0.468
P(^3) = ^lP2P3^ 0144
P ( * 4) = (1 - Pi) 0 P2 ) (1 - p a) = 0.056
Let us write down the conditional probabilities of destroying the target upon
the realization of each of these events:
P (A/ BJ = 0.4, P (AlB.2) - 0.7, P ( A/£3) - 1.0, P ( A/ BJ = 0
Substituting the resulting expressions into formula (8 ), we get the probabi
lity of target destruction:
P(i4) = P ( f l 1)-P(i4/fl1) + P(fl,)-P(i4/flt ) + P(fl3)*P(»/Ba)
P (B4) ■P (A/B4) = 0.332 •0.4 + 0.468 •0.7
+ 0.144 • 1.0 + 0.056 •0 = 0.6044
Note 2. If event A does not dépend on event B, then
P(A/B) = P(A)
and formula ( 1) takes the form
P (A and B) - P (fl) •P (4)
That is, we obtain formula (1), Sec. 8.4.
8.6 PROBABILITY OF CAUSES.

BAYES’S FORMULA
Statement of the problem. As in Theorem 2, Sec. 8.5, we will

consider a complété group of exclusive events Blt B2, . . . , Bny the
probabilities of occurrence of which are P(Bj), P(B2), . . . , P (B J.
Event A caru, occur only together with some one of the events
Bly B2, ...» B„, which we will call causes.
The probability of the occurrence of event A is, in accord with
formula (8 ) of Sec. 8.5,
P (A) - P (Bj) P (A/Bt) + P (B2) P (A/B2) + • • • + P (Bn) P (A/Bn) (1)
Suppose that event A has taken place. This fact will alter the
probability of the causes, PtB ^, . . . , P (B,J. It is required to
détermine the conditional probabilities of the realization of these
causes on the assumption that event A has occurred, that is, to
détermine
P (BJA)% P (B2M )......... P(BtJA)
S o lu tio n of th e p ro b lem . By formula (7), Sec. 8.5, we will find

the probability P(/4 and B,):
P (A and B J = P (BJ-P (A/BJ = P(A)-P(BjA)
whence
Substituting for P (4) its expression (1), we get

P (B J A) = .(2 )
2 P (B i).P(AiBi)
t=l
Thç probabilities P(B2/A), P {BJ A), . . . , P( Bj A) are determi-
ned in similar fashion:
P (B„/A) = ~?..{-Bk) P{AlBk) ■ (3 )
2 Pffl+PM/fl,-)
Formula (2) is called Bayes's formula or the theorem of causes.

(Bayes's rule for the probability of causes.)
Note. From formula (3) it follows that in the expression of the
probability P(Bk/ A ) — the probability of the realization of cause Bk
provided that event A has occurred — the denominator is indepen-
dent of the number k.
Example 1. Suppose that prior to an experiment there were four equally
probable causes:
Blt B2t B3, Ba\ P( B1) = P( B2) = P( B3) = P( Ba) = 0.25
The conditional probabilities of the occurrence of event A are, respectively,
equal to
P (A/Bi) = 0.7, P( A/ B2) = 0A
P( A/ B3) = 0At P (A/ BJ = 0.02
Suppose the outcome of a trial yields event A. Then, by formula (3), we get
0.25 0.7 0.175
P (S i/4 ) 0.25-0.7 + 0.25-0.1 +0.25-0.1 +0.25-0.02 0.23
0.76
P(flsM)=^ w 1 = 0 -11
p (53M) = — 2^ i = 0-1l
p (BJA) = — = 002
Here P (/?!) = 0.25, P (Bi/A) = 0.76 became greater because event A occurred.
In this case, the probability P (A/Bi) = 0.7 is greater than the other conditional
probabilities.
Example 2. Each of two tanks fired independently at a target. The probabi
lity of the first tank destroying the target is ^ = 0.8, that of the second,
8.6 Probability of Causes. Bayes's Formula 459
p 2 = 0,4. The target is destroyed by a single hit. Détermine the probability

that it was destroyed by the first tank.
Solution. Event A is the destruction of the target by a single hit. Prior to
firing, we hâve the following possible causes: •
Blf bot h tanks missed,
£ 2, both tanks scored a hit,
£ 3, the first tank scored a hit, the second missed,
£ 4 , the first tank missed, the second scored a hit.
We find the probabilités of these causes by the theorem of multiplication
of probabilités:
P (£ x) = ( 1—p2) ( 1—p2) = 0.2 0.6 = 0.12
P (£2) = PiP2 = 0 .8 -0.4 = 0.32
P (£,) = Pi (1 — p2) = 0.8-0.6 = 0.48
p (B4) = ( 1—Pî) p2 = 0.2• 0.4 = 0.08
We détermine the conditional probabilities of occurrence of the event:
P (A/B,) = 0, P (A/B2) = 0, P (AlB3) = 1, P (A/Ba) — 1
By formula (2) we find the conditional probability of the causes:
0 . 12-0
P (B1/A) =
0.12 0 4-0.32-0 + 0.48-1 +0.08-1 0.56 —0
0.32-0
P (BdA ) = -0
0.56
0.48-1 6
P (£ 3/ A) = 0.56 7
0.08-1 1
P (£ 4/^) = 0.56 7
Example 3. At a factory, 30% of the instruments are assembled by specia-
lists of high qualification, 70% by those of medium qualification. The reliabi-
lity of an instrument assembled by the former is 0.90, that assembled by the
latter, 0.80. An instrument picked off the shelf turns out to be retable. Déter
mine the probability that it was assembfed by the specialist of higher qualifi
cation.
Solution. Event A stands for retable operation of the instrument. Prior to
checking the instrument, we hâve the following possible causes:
£ j, the instrument is assembled by the highly qualified specialist,
£ 2, the instrument is assembled by the specialist of medium qualification.
We write down the probabilities of these causes:
P (£ 1) = 0.3, P (£a) = 0.7
The conditional probabilités of the event are
P (i4/£1) = 0.9, P (A/BJ = 0.8
We now détermine the probabilities of the causes Bx and £ 2, provided that
event A has occurred.
By formula (2) we hâve
P ^ ) = 0T O T ï O T 8 = m = 0-325
p^ > = ô Æ o . 8= ë =a675
8.7 A DISCRETE RANDOM VARIABLE.

THE DISTRIBUTION LAW OF A DISCRETE RANDOM VARIABLE
D é fin itio n . A variable quantity x which, in a trial, assumes one

value out of a finite or infinité sequence x2, xkt . . . is
called a discrète random quantity (or variable), if to each value xk
there corresponds a definite probability pk that the variable x will
assume the value xk.
It follows from the définition that to every value xk there cor
responds a probability pk.
The functional relationship between pk and xk is called the distri
bution law of probabilities of a discrète random variable x: *
Possible values of the ran

dom variable *1 *2 ... Xk
Probabilities of these values î Pt ... Pk
The distribution law can also be represented graphically in the

form of a polygon of probability distribution (also called a Jrequency
polygon): in a rectangular coordinate
system, points are constructed with
coordinates (xk, pk) and are joined
by a polygonal line (Fig. 160).
The distribution law can also be
specified analytically:
pk= f (*k)
The fact that a random variable x
assumes one of the values of a sequence
* 1, • • • , **, • • • IS a certain event, and so the condition
£ p ,== i o)
<=i
must hold true for the case of a finite sequence of N values, or
= 1 (!')
<=1
for the case of an infinité sequence. The value x,- of the random
variable having the greatest probability is called the mode. The
random variable x depicted in Fig. 160 has mode x2.
* Or. briefly, the distribution law of a random variable.
8.7 A Discrète Random Variable 461
Example 1. A variable x dénotés the number that turns up in a single throw

of a die. The variable x can assume one of the following values: 1, 2, 3, 4, 5, 6.
The probability of any one value turning up is 1/6. We thus get the following
table for the distribution of this random variable:
X 1 2 3 4 5 6
p 1/6 1/6 1/6 1/6 1/6 1/6
Example 2. The probability of the occurrence of event A in each of an

infinité sequence of trials is equal to p. The random variable x is the number
of the trial in which A occurred for the first time. Find the law of distribution of x.
Solution. The random variable x can assume any intégral positive value
1, 2, 3..........The probability p x that event A will occur in the first trial is
Pl = V(A) = p
The probability p2 that the event will not occur in the first trial but will occur
in the second is
p2 — ? ( À and A) = ( l —p) p
The probability p2 that A will not occur in the first or the second trial but
will occur in the third is
p3 = P ( A and A and A) = ( \ — p ) ( \ —p) p = ( \ —p)i p
and so on,
P* = (l —P)k ~l P ( 2)
Here is the table of distribution of probabilities:
X 1 2 3 k
Ph p o - p ) p ( i - p ) ’ p ... (i —p ) * - * p
We also hâve
*=i
Z
*=i
p » = £ ( i - p ) * - | p = i _ '(î _ p j = i
Problem on firing to a first hit. The foregoing problem flnds

applications, in particular in the theory of gunfire.
Suppose we hâve a case of gunfire to a first hit. The probability
of a hit scored on each shot is p.
The random variable x is the number of the shot in which the
first hit was scored. The distribution table of the probabilities of
this random variable will be the same as In Example 2.
Example 3. The probability of a hit on each shot is p = 0.8. We hâve three
shells. lietermine the probability that one, two and three shells will be used it
the gunfire is continued to a first hit or to a miss by ail three shells; compile
a table of the distribution of the random variable x (the number of shells used).
Solution. Let x be a random variable indicating the number of shells fired;

P (je = jc,) is the probability that x l shells will be used. Then P ( x = l ) = p = 0.8
is the probability of a hit scored on the first shot.
P ( a: = 2) = ( 1— p) p = (1—0.8).0.8 = 0.16
is the probability that the first shot was a miss, the second a hit.
P (* = 3) = (1—p)2= (l —0 .8 )(1 —0.8) = 0.2-0.2 = 0.04
because there are only three shells and gunfire ceases irrespective of whether
the third shot is a hit or a miss. This last probability could also hâve been
computed as a différence:
1—P (* = 1) —P (* = 2) = 1—0.8 —0.16 = 0.04
The distribution table will look like this:
X 1 2 3
V(x = x„) 0.8 0.16 0.04
Note. This problem can be stated in terms of the urn model, which means
it can also be applied to other problems. This remark applies to certain
other problems as well.
8.8 RELATIVE FREQUENCY AND THE PROBABILITY

OF RELATIVE FREQUENCY
» IN REPEATED TRIALS
Suppose we hâve a sequence of n trials, in each of which event A

canoccur with probability p. Let x be a random variable denoting the
relative frequency of occurrence of event A in the sequence con-
sisting of n trials. It is required to détermine the law of distribu
tion of the random variable x in the «-trial sequence.
It is quite obvious that in n trials x will assume one of the
following values:
0^ J_ 2_ JL
n ' n ' n ’ *' ' ’ n
Theorem 1. The probability P ( x= — \ that the random variable x

\ n J
will assume the value — , that is, that in n trials event A will
occur m times and the event A ( nonoccurrence of A) will occur n —m
times is equal to C^pmqn~m, where C% is the number of combinations
of n éléments taken m at a time; p is the probability of the occur
rence of event A, p — P(A); q is the probability of the nonoccurrence
of event A, that is, q = l — p = P(A).
8.8 Relative Frequency and the Probability of Relative Frequency 463
Proof. Event A will occur m times in n trials, for instance, if

the alternation of events A and Â is as îven below:
AA . . . AÂ J . . . Â
m n -m
which is to say, A occurs in the first m trials, and does not occur
in the next n — m trials (event A occurs). Since
P(A) = p, P(Â)--=rl— p = q
then by the multiplication theorem the probability of such an
alternation of events A and À will be
p m• q n ~m
But event A can appear m times in n trials if we hâve a

different sequence of events A and A. For example, given the alter
nation AA . . . A AA . . . Â A. But it is necessary that A occur m
m- 1 h- m '
times, and Â, n —m times. The probability of the occurrence of
this kind of alternation of events A and À is
p m - l q t - m p _ pm gti-m
How many distinct alternations of events A and Â can there be

in n trials in which A occurs m times? It is clearly equal to the
number of combinations of n éléments taken m at a time:
n m ^ n ( n — \)(n — 2) . . . [n —(m —1)]
1-2-3 . . . m
Thus, by the addition theorem, we hâve
p mg n ~ m —
(- p m q n ~ m —
|— -J- p m q t i —m
or
' s r"
p[x = ^ = CZpmqn~m ( 1)
The proof of the theorem is complété.

Having proved the theorem, we hâve thus determined the law
of distribution of the random variable x\ we give the law in the
form of a table:
0 1 2 m n
X
n n n n n
1 qn C ' pqn -' C \p iq', ~'i C ™ p mq n ~ m 1 •p n

464 Ch. 8 The Theory of Probabilité
This distribution law is known as the binomial distribution be-

cause the probabilities P ^ x = - ^ are equal to the corresponding
ternis in the binomial expansion of the expression (q + p)n:
(V+ p)m= 2 CnPmqn~m (2)

m=o
As was to be expected, the union (logical sum) of the probabi
lities of ail possible values of the random variable is equal to
1 since (p + q)n= 1" = 1 .
Note, in the study of many problems it is often necessary to
détermine the probability that event A will occur “at least once”, that
is, the relative frequency of the event, x ^ - - - . It is quite obvious
that this probability P canbe determined from the équation
From the distribution table it also follows that the probability

that the event will occur no less than k times is found
from the formula
p ( ^ i ) :=É o t -* (4)
m-k
or
k- 1
p ( x ^ 4 ) = 1~ Z
' ' m=0
Example 1. Give a graphical représentation of the binomial distribution of
the random variable x for n = 8, p = l/2 , <7 = 1/ 2 .
Solution. We déterminé ail the values of the probabilities that appear in the
table:
'•«-O»
P ( x = i ) = Cj
]_
256= 32
P (* = j ) = d f ! Y — ± —JL
\ 2J 1•2 ’ 28 64
8 8-7-6 1 7
P
'1-2-3 29 32
8 -7 -6 5 35
P
=1-2-3-4 * 2«“ 128
P
( * = 8') = C®2»=:32
8.8 Relative Frequency and the Probability of Relative Frequency 465
v 8 y *2»—64
P(* = ir ) = C* 2à= 32
Pf^ «8 )U c8 28'i _- 2561
Let us construct the distribution polygon, or frequency polygon (Fig. 161).
Example 2. What is the probability that event A will occur twice (a) in two
trials, (b) in three trials, (c) in 10 trials, if the probability of the occurrence of
the event in each trial is equal to 0.4?
Solution, (a) Here, n = 2, p = 0.4, (7 = 0 .6 :
p ( jc=t ) = c *p V = i ^ (0-4 ) S = 0 -16

(b) here, n = 3, p = 0.4, (7 = 0 .6 :
p(■*=-§-)=c*pV=ri (0-4>î 0 -6= 0288

(c) here, n = 1 0 , p = 0.4, (7 = 0 .6 :
p ( , = f 0) = cî.pV = - ^ ( 0 .4)M0.6)»=0.,2«
Example 3. Five independent shots are fired at a target. The probability of
a hit in each shot is 0.2. Three hits suffice to destroy the target. Détermine the
probability of target destruction.
Solution. Here, n = 5, p = 0 .2 , <7 = 0.8. It is clear that the probability of
target destruction should be computed from the formula
p<«= p ( * = 4 ) + p ( x“ t ) + p (*= t )
or from the formula
P«<*,= l - [ p ( * = y ) + p ( * = 3 - ) + p ( * = |- ) ]
30—2082
By the first formula we hâve
Pdes = d P V + C lp V + Cjp* = • (0.2)» •(0.8)*
+ (0.2)«0.8 + 1 (0.2)*=0.05792 « 0.06

Example 4. Four independent trials are carried out. The probability of the
occurrence of event A in each trial is 0.5. Détermine the probability that A
will occur at least twice.
Solution. Here, n = 4. p — 0.5, (7= 0.5:
p ( ^ 4 ) = p ( ^ = 4 ) + ,,( ^ 4 ) + p ( * = 4 )
or
We compute the probability

P < - ) = P (* = - ) + P = - ) = ?* + 4qV = (0.5)* + 4 (0.5)« = 0.3125
Hence, by the second formula, we obtain
P(* -?-) = l — [(0.5)« + 4 (0.5)*] = 0.6875 « 0.69
Example 5. The probability of defective items in a given batch is p = 0.1.

What is the probability that in a batch of three items there will be m = 0f
1, 2, 3 defective items?
Solution.
P = -J ) = = 1•0.9» = 0.729
P ^ = C\pq* = y -0.1 •0.9* = 0.243
P 1 J = Cîp’-q = — . 0.1*• 0.9 = 0.027
p (jc = |-^ = iC § p 3= 1 0 .1 » = 0 .0 0 1
8.9 THE MATHEMATICAL EXPECTATION

OF A DISCRETE RANDOM VARIABLE
Suppose we hâve a discrète random variable x with an appro-
priate distribution law:
X *1 x2 Xk */!
P (*=**) P i P2 ... Pk ... Pn
Définition 1. The mathematical expectation (or, simply, expecta

tion) of a random variable x (we symbqlize expectation by M [x]
8.9 The Mathematical Expectation of a Discrète Random Variable 467
or mx) is the sum of the products of ail possible values of the

random variable by the probabilities of these values:
M [x] = x,p 1 + x.lpi + . . . + xnpn
or, briefly,
M [x] = 2 **/>* (1)

k—\
n
Here, as has already been pointed out, 2 P* = 1.
k= 1
If the values of the random variable form an infinité sequence,
then
oo
= 2 xkPk (O
We will consider only those random variables for which the
sériés converges.
We will now see what connection there is between the mathema
tical expectation of a random variable and the arithmetic mean of
that variable for a large number of trials; namely, we will demon-
strate that in a large number of trials, the arithmetic mean of the
observed values is close to the expectation; or, in the ternis of Sec. 8.1,
we can say that the arithmetic 1mean of the observed values of a
random variable tends to the expectation when the number of trials
increases without bound.
Suppose that N independent experiments are performed and that
the value xx occurred n1 times,
the value x2 occurred n2 times,
the value xv occurred nv times.

The random variable x assumes the values xlt x2, . . . , xv .
We compute the arithmetic mean of the variables x (we dénoté
it by M [x] or mx):
— *1n1+ x îM - . . . + . * v /lv , nv
( 2)
However, since in the case of a large number of trials N the rela

tive frequency tends to the probabil ity of the occurrence of
the value xk, it folloiVs that
\
2
k=l k=\
xkPk
Under rather natural assumptions we hâve

M [ x ] — M[x] (3)
n —«
Note I. If we ccnsidered an um scheme with N balls where there
are n1 balls labelled xlt ri2 balls labelled x2 and so on, the expected
value on drawing one bail will be expressed by formula (2 ), that
is, it will be equal to mx.
Example I. Détermine the expectation of the random variable x of the num-
ber of hits in three shots if the probability of a hit in each shot is p = 0.4.
Solution. The random variable x can take on the values
*i = 0 , **= 1, *3 = 2 , *4 = 3
We set up a distribution table of the given random variable.
The probabilités of these values are found by the theorem on repeated trials
(n = 3, p = 0.4, <7= 0.6):
P (x = 0) = CS (0.6)3 = 0.216
P (x = l) = Cj (0.4) (0.6)J = 0.432
P (x = 2) = C1 (0.4)*. (0.6) - 0.288
P (x = 3) = C® (0.4)* = 0.064
The distribution table of the random variable is
X 0 1 2 3
10
x'
0.216 0.432 0.228 0.064

II
è
We calculate the expectation from formula (1):

m* = 0-0.216 + 1-0.432 + 2-0.288 + 3-0.064= 1.2 hits
Example 2. One shot is fired at a target. The probability of a hit is p.
Détermine the expectation of the random variable x of the number of hits.
Form the distribution table of the random variable
x 0 1
Pk l —p P
Consequently, mx = 0 (1 — p) + l- p = p.
Note 2 . In the sequel it will be established that the expectation
M (*] of the number of occurrences of event A ity n independent
trials is equal to the product of the number of tr/als by the pro
bability p of occurrence of A in each trial:
M [*] = np (4)
8.9 The Mathematical Expectation of a Discrète Random Variable 469
If in (4), n is the number of shots and p is the probability of

a hit, then the solution of the problem of Example 1 will be:
M [x] = tip — 3 0A = L.2 hits
If in formula (4) we know M [x] and pf then we can find n — the
number of trials that yields the given expectation of the number
of the occurrences of the event:
M [x]
n = ——
p
Example 3. The probability of scoring a hit in one shot is p = 0.2. Déter
mine the shell consumption that will ensure expectation 5 of the number of hits:
n = j^ 2 = 25 shells
(Note once again that similar problems occur in a variety of investigations.
The word “hit” is replaced by “occurrence of an event” and “shot” is replaced
by “trial”.)
Example 4. Détermine the expectation of the random variable x with the
following distribution table:
x 1 1 2 3 ... k ...
Pk II p ( i —p ) p (i- p? p ... ( i —p )* -1 p ...
See Example 2, Sec. 8.7

Solution. From formula (1) we hâve (denoting \ ^ - p = q)
mx = 1-p + 2qp + 3q*p+ . . . + k q k~ 1p + . . .
= p ( \ + 2 q + Z q * + . . . + k q * - '+ ...)
= P(<7+<72 + <78+ - • • + < 7 * + • • • ) ' = / > ( f Z ^ )
1—<7+<7^ p _p __ 1
p ( i — <7)a ( * — q)2 P'1 P
Note that
mx —►1 as p —►1
rnx —» oo as p —►0
These relations can be explained by the essence of the problem.
Indeed, if the probability of occurrence of event A in each trial
is close to 1 (p « 1), then it may be expected that A will occur
in the first trial (mx « 1). But if the probability p is small (p « 0),
then one can expect that in order for event A to occur it will be
necessary to perform a large number of trials (mx » oo).
The mathematical expeatation of a random variable x is called
the centre of probability distribution of the random variable x.
Note 3. The term “centre of probability distribution” is intro-

duced by analogy with the term “centre of gravity”. If on the
je-axis we place masses p lt p i t . . . , p n at points with abscissas
xlt xt, . . . . xn, then from analytic geometry it is clear that the
abscissa of the centre of gravity of these masses is determined from
the formula
n
*kPk
n
2 Ph
*=1
If 2 p * = l , then
*=i n
* c =*=
2 i XkPk (5)
Formula (5) coïncides (in appearance) with formula (1) for mathe-
matical expectation.
We hâve thus established that the centre of gravity of a sériés
of masses and the expectation are computed from analogous for
mulas, whence the term “centre of probability distribution”.
\
P
p\
Pi
r f l n _ 1^1 Pi
0 x, X2 x3 Xg Xg Xg Xj X x? x \ xgo x4°x | xg Xf X*
Fig. 162 Fig. 163
Suppose we hâve a random variable x with air associated distri

bution (Fig. 162); let its expectation be mx. \Now consider the
différence between the random variable x and its expectation: x —mx.
We will call this quantity the centred random variable, or the
déviation, and we will dénoté it by x°. /
It is quite clear that the distribution law/of this random vari
able will be /
X i = X i — mx x l = x2— mx / ...
1
II
&
a
£
Pk Pl p% / ... Pk
(see Fig. 163). 7

S. 10 Variance. Root-Mean-Square (Standard) Déviation. Moments 471
Let us find the expectation of a centred random variable:

n n n
M [x— mx] = 2 (xk— mx)p„ = 2 2 mxp„

*=1 *=1 A£=1
n
= mx— mx 2 Pk = mx —mx - 1 = 0
A=1
Thus, the expectation of a centred random variable is equal to zéro.
Note 4. It is sometimes advisable to regard a nonrandom constant
(a certain quantity) c as a random variable which assumes the
value c with probability 1 and assumes ail other values with pro-
bability 0 .
Then it is meaningful to speak of the expectation of a constant:
M [c]=c-l=c (6)
The expectation of a constant is equal to that constant.
8.10 VARIANCE.
ROOT-MEAN-SQUARE (STANDARD) DEVIATION. MOMENTS
In addition to the expectation of a random variable x, which

defines the position of the centre of a probability distribution, a
distribution is further characterized quantitatively by the variance
of the random variable x.
The variance is denoted by D [x] or aj.
The word variance means dispersion. Variance is a numerical
characteristic of the dispersion, or spread of values, of a random
variable about its mathematical expectation.
Définition. The variance of a random variable x is the expecta
tion of the square of the différence between x and its expectation
(that is, the expectation of the square of the appropriate centred
random variable):
D [x] =M ((x—mxf] ( 1)
or
D[x] = 2 (Xk— mxy p k (2 )

k—1
Variance has the dimensions of the square of the random vari

able. It is sometimes more convenient in describing dispersion tù
make use of a quantity whose dimensions coïncide with those of
the random variable. This quantity is termed the root-mean-square
déviation (standard déviation). \
or, in expanded form,
° M = 1f S (xk— mxY pk (3)

f *=l
The standard déviation is also denoted by ox.
Note I. In computing variance, it is often convenient to trans-
form formula ( 1) as follows:
n n n n
D[.ï] =k=l
2 (x n —m x ) i Pk =fe=l
2 xlPk — 2 *=1
2 Xkmxpk +k=l
2 tti'xPk
n n n
= 2 x%pk — 2mx 2 xkpk + m% 2 Pk

k=l k=\ /Z=l
—M [x 2] —2mx -mx -\-ml• 1 = M [x2] — ml
Thus
D [.v] = M [x2] —ml (4)
That is, the variance is equal to the différence between the expec
tation of the square of the random variable and the square of the
expectation of the random variable.
Example 1. One shot is fired at a target. The probability of a hit is p.
Détermine the expectation, the variance and the standard déviation.
Solution. We construct a table of the values of the number of hits:
X 1 0
I
Pk 1 P <7
q = 1 —p. Hence, /
W [ * ] = l.p + 0.? = p /
D(*) =(1 — PÏ1P + (0 — p)2q = q2p / p * q = pq
o (x) = V'pq /
Let us consider some more exam ple/in order to get a good idea
of what the concepts of variance and/standard déviation mean as
measures of dispersion of a random/variable.
Example 2. A random variable x-is s/ecifi'ed by the following distribution
law (see table and Fig. 164). /
3 4
* h 7
Pk 0.3 0.4 0.3

8.10 Variance. Root-Mean-Square (Standard) Déviation. Moments 473
Détermine the (1) expectation, (2) variance, (3) standard déviation.

Solution.
1. M [x] = 2 0.3 + 3 0.4 + 4 0.3 = 3,
2. D [x] = (2 —3)2 •0.3 + (3 —3)2•0.4 + (4 —3)2.0«3 = 0.6,
3. c r[* ]= / D l ï j = ]/lT 6 = 0.77.
P
P
0 .3 0M 0 .3
0 .3 0A 0 .3
1 ■
0 2\ jr i '4 X 0 1 i? 5 x
Fig. 164 Fig. 165
Example 3. A random variable x has the distribution law (see table and
Fig. 165):
x 1 3 5
Pk 0.3 0.4 0.3

Solution.
1. M[x] = 1 0.3 + 3 -0 .4 + 5 0.3 = 3,
2. D [jc] = (l —3)2-0.3+ (3 —3)2 0.4+ (5 —3)2-0.3 = 2.4,
3. o [jc1= 1^2.4=1.55.
The dispersion of the random variable in the first example is less than that
in the second example (see Figs. 164, 165). The variances of these quantities
are equal to 0.6 and 2.4, respectively.
Example 4. A random variable x has the following distribution law (see
table and Fig. 166):
P1 ti
x 3
l
P l
0 3 x
Fig. 166

Solution.
1. M [x] = 3 • 1 = 3,
2. D [x] = (3—3)2-1 = 0 , i
3. o [ x 1= 0. \ /
There is no dispersion of the random variable in thiycase.
Note 2 . If we regard a constant as a random variable which

assumes a value c with probability 1 , then it is easy to demon-
strate that D [c] = 0.
Proof. It has been shown [see (6 ), Sec. 8.9] that M [c]= c. From
formula ( 1) we get
D [c] —M [(c—c)2] —M [0] = 0
which complétés the proof.
Note 3. By analogy with the terminology of mechanics, we can
call the expectations of the quantities (x—mx), (x—mx)2 the first
central moment and the second central moment of the variable x.
A third central moment is also considered:
n
2 (Xk — mx)3Pk
k=l
If a random variable is distributed symmetrically about the centre
of the probability distribution (Fig. 162), it is then obvious that
its third central moment will beequal to zéro. If the third moment
is different from zéro, the random variable cannot be distributed
symmetrically.
8.11 FUNCTIONS OF RANDOM VARIABLES

Let a random variable x be given by the following distribution
law:
X Xi x2 Xk xn
Pk Pl Pi ^ Pk Pu
We consider a^tunction of the random variable x:
The values of the function yk ~ f ( x h) will be the values of the

random variable y.
If ail the values #*--/(**) are distinct, then the distribution
law of the random variable y is given by the table
y = f(x) y\ = f (*i) Vt = f (*t) yk = f (**) ... y n = f (x „ )
Pk Pl Pi Pk Pn
8.ÎJ Functions of Random Variables 475
If some of the values yk = f (xk) are equal, the appropriate co-

lumns may be combined into one by combining the corresponding
probabilities. .
The expectation of the function y ~ f ( x ) of the random variable x
will be found from a formula similar to formula (1) of Sec. 8.9:
M [/(*)] = £ /(* * ) P* (1)

k=\
In similar fashion we détermine the variance of the function:
D [/ (x)] = M [(/ (x )— M [ / (x)])*] = £ (f (xk) - m , pk

k=\
Example. A random variable <p is given by the following distribution law:
n jt Jl Jl
<p 0
“ T “ T T T
Pk 0.1 0.1 0.2 0.3 0.3
We consider a function of this random variable:

y = A sin cp
We form the distribution table for the random variable y:
y —A A V~2 0 AV 2 A
2 2
Pk 0.1 0.1 0.2 0.3 0.3
Let us find the expectation of the function:
M [A sin<p] = —^ 0.1— - 0 . 1 + 0 - 0 . 2 + ^ - p 0.3 + /10.3
= A ^0.2 + ^ Ç ~ 0.2^ = A (0.2 + 0.14) = 0.34.4

Problems of this kind occur in the considération of oscillatory processes.
8.12 CONTINUOUS RANDOM VARIABLE.

PROBABILITY DENSITY FUNCTION OF A CONTINUOUS RANDOM
VARIABLE. THE PROBABILITY OF THE RANDOM VARIABLE
FALLING IN A SPECIFIED INTERVAL.
An example will help to cast some light on the problem at hand.
Example. The amount of wear of a cylinder is measured after a certain
period of operation. This quantity is determined by the value of the increase
in diameter of the cylinder. We dénoté it by x. From the essence of the problem,
it follows that x can assume any \value in a certain intetval (a, b) of possible
values. \ /
This quantity is termed a œntinuous randotn variable.

We consider the continuous random variable x specified on a
certain interval {a, b) which can also be an infinité interval,
(—oo, +oo). We divide this interval into subintervals of length
Axi„l = xi —x,_i by the arbitrary points x0, x1( x2, . . . , x„.
Suppose we know the probability that the random variable x
will fall in the interval (x,-^, x,). We dénoté this probability by
P (x,_, < x < Xj) and represent it as the area of a rectangle with
base Ax,- (Fig. 167).
X0X,
i.
xi-lx i x n-ix n x
Fig. 167
For each interval (x,_,, x/) we détermine the probability of x

falling in this interval and, hence, we can construct an appropriate
rectangle. The resuit is a step-like broken line.
Définition I. If there exists a function y = f(x) such that
Iim p f* < * < *+ A*) _ ^ | (i)
Ax -* 0 Ax
then this function f(x) isêrine&~tfrê~probability densityJfundioa
(or, simply, denstîÿ function) of the random variable x, or the
distribution. (It is also called the fre-
quency function, distribution density, or
the probability density.) We will use x
to dénoté the continuous random varia
ble, x or xk to dénoté the values of this
Fig. 168 random variable. If it is clear from the
context, we will occasionally drop the
bar on x. The curve y = f(x) is called the probability curve or the dis
tribution curve (Fig. 168). Using the définition of limit, from
équation ( 1) follows, to within infinitesimals of higher order than
Ax, the approximate équation
P (x < x < x + Ax) » f (x) Ax ( 2)
We will now prove the following theorem.
8.12 Continuous Random Variable 477
Theorem 1 . Let f (x) be the detisiiy function of the random vari

able x. Then the probability that a value of the random variable x
will fall in some interval (a, P) is equal to the definite intégral
of the function f (x) from a to P; that îs, we hâve the following
équation:
_ fl
P (a < x < p) = ^ / (x) dx (3)
a
Proof. Partition the interval (a, P) into n subintervals by the

points a = x„ x2, . . . , xn+l = P (Fig. 169). Apply formula (2) to
each subinterval:
P (xt < X < X2) » / (Xt) âXj
P (x2 < x < x3) » / (x,) Ax,
P ( x „ < x < x n+ l) « / (-«n) àX„

Adding the left and right membersof the équations, we will clearly
get P (a < x < P) on the left. We thus hâve the approximate équation
n
P (a < X < P) w 2 / (*,) Ax,-

i = 1
Passing to the limit in the right member as maxA*,-—►(), we get,

on the basis of the properties of intégral sums, the exact équation
n
P (a < x < P) = lim 2 f (*/) Axt

m a x Ax • -► 0 i = 1
(We assume that f(x) is such that the limit on the right exists.)
The limit on the right is the definite intégral of the function f (x)
from a to p. Thus
P (a < x < P) = j / (x)
a
which complétés the proof. /
Thus, knowing the probability density function of a random

variable, we can détermine the probability that a value of the
random variable will lie in a given interval. Geometrically, this
probability is equal to the area of the resulting curvilinear tra-
pezoid (Fig. 170).
y N ote. In the case of a continuous
random variable, the probability of
an event that x = x0 will be equal to
A zéro.
0 oc P JC Indeed, putting .c = * 0 in équation
Fig. 170
(2 ), we get
P(x0 < x < x 0+ A x ) æ f (*0) Ax

whence
Iim P (jt0 < x < x0+ Ax) = 0
A* -►0
or
P ( jc= .*:0) = 0
(Also see Note 1 on page 448.) For this reason, in (3) and the
preceding équations we can Write not only P ( a < x < P ) but also
P (a ^ .* ;< P ), since
P (ocîSÇa; < P) = P W = a) + P (a < x < P) + P (x = P) = P (a < x < P)
If ail possible values of the random variable x lie in the inter-
val (a, b), then
[ f ( x ) dx = 1 (4)
a
since we are certain that a value of the random variable will lie
in the interval (a, b).
If the interval of possible values is (— oo, +<»). then
^ f (x)dx= 1 (5)
Note that if it follows from the essence of the problem that the func
tion f (x) is defined on the finite interval (a, b), then we can regard
the function to be defined on the whole infinité interval (— oo,
-f- oo), but
ï(x) = 0
outside the interval (a, b). In this case, both (4) and (5) hold
true. The density function of the random variable fully détermines
the random variable.
8.13 The Distribution Function. Law of Uniform Distribution 479
8.13 THE DISTRIBUTION FUNCTION.

LAW OF UNIFORM DISTRIBUTION
Définition 1 . Let f (x) be the density fimction of some random
variable x ( — oo < x < + oo); then the function
F(x)= $ f(x)dx ( 1)
—oo
is called the distribution function.
For a discrète random variable, the distribution function is
equal to the sum of the probabilities of those values xk which are
less than x:
F(x)= 2 Pk
xk < x
from équation (3), Sec. 8.12, it follows that the distribution

function F (%) is the probability that the random variable x will
assume a value less than x (Fig. 172):
F (x) = P (— oo < x < x) (2)
It follows from Fig. 171 that for a given value of x, the value
of the distribution function is numerically equal to the area bounded
by the distribution curve lying to the left of the ordinate drawn
through the point x.
The graph of the function F (x) is termed the probability distri

bution curve (Fig. 172).
Passing to the limit as x —►+-oo in (1), with regard for (5),
Sec. 8 . 12 , we get
* oc
lim F (x) = lim f f ( x ) dx = C / (*) dx 1
» *-++« •)- X
J — 30
Now let us prove the following theorem. _

Theorem 1. The probability of a random variable x lying in a
given interval (a, P) is equal to the incrément in the distribution
function over that interval:
P ( a < * < p ) = F (P )-F (c()
Proof. Let us express the probability of x to assume a value

lying in the given interval (a, P). We rewrite formula (3),
Sec. 8 . 1 2 , as
p p a
P ( a < I < p ) = \ ) f(x)dx = j f (x)dx— J f(x)dx
(seeîg. 173). Or, using ( 1), we can Write

P ( a < ï < P ) = F(p)-F(a)
which\completes the proof (see Fig. 174).
9(<X<z<fl’ F(p)-Fÿ.)
F (fi)
Note tha't the density function f (x) and the corresponding dis
tribution function F (x) are connectée! by the relation
F'(x) = f(x) (3)
This follows from (1) and the theorem on differentiating a definite
intégral with respect to the upper limit.
We now consider a random variable with a uniform distribution
of probabilities. The distribution law, or the density function,
f (x) of such a random variable is de-
fined as follows:
f(x) = 0 for x < a
f(x) = c for a < x < b
f(x) — 0 for b < x
On the interval (a, b) the density func-
F»g. 175 tion f(x) has a constant value c (Fig.
175); outside this interval it is zéro.
Such a distribution is also called the law of uniform distribution.
00
From the condition J f ( x ) dx = 1 we find the value of c:
—ao
00 b
J f (x)dx — ^ cdx = c(b—a) = 1
8.13 The Distribution Function. Law of Uniform Distribution 481
hence
From the last équation it follows that the interval (a, b) of uni
form distribution is necessarily finite. Let us détermine the pro-
bability that the random variable x will assume a value lying in
the interval (a, P):
P (a < * < P) = j / (x) dx = J — • dx P—«

b —a
a a
Thus, the desired probability is
P ( a < * < |) ) - L ï
(this relation is similar to the définition of geometrical probability
for the two-dimensional case given on page 452).
We now détermine the distribution function
X
F( x) = J f(x)dx
— CD
I! x < a , then f(x) = 0 , and, hence,

F(x) = 0
If a < x < b, then f (x) = and, hence,
X
If b < x, then
/(* ) = 0 , \f(x)dx = 0
b
hep—
(see Fig. 176). Fig. 176

The following are examples of random
variables that obey the law of uniform distribution.
Example 1. The measurements of a certain quantity are rounded to the
nearest scale division. The rounding errors are a random variable with uniform
distribution of probabilités. If 21 is the number of certain units in one scale
division, then the density function of this random variable is

/ u0 îor
for x < — /
/(*> - U 21 « < x< l

0 for l < X
i
Here, a = — l , b = l , c = — .
Example 2. A symmetric wheel in rotation cornes to a stop due to friction.
When the wheel cornes to rest, the angle 0 formed by a certain definite moving
radius of the wheel with a fixed radius is a random variable with a density
function
0 for 0 < 0
W H i for 0 < 0 < 2ji

for 2 ji < 0
8.14 NUMERICAL CHARACTERISTICS

OF A CONTINUOUS RANDOM VARIABLE
In the same manner that we considered a discrète (discontinuons)
random variable, let us now examine the numerical characteristics
of a continuous random variable x with density function f(x).
Définition 1 . The mathematical expectation of a continuous ran
dom variable x with density function f (x) is the expression
00
M [je] — ^ xf (x)dx ( 1)
—ao
If a random variable x can assume values only on a finite inter-
val [a, b], then the expectation M [je] is expressed by the formula
_ b
M [je] = J xf (je) dx 11 ')
a
Formula (T) may be regarded as a generalization of formula (1),
Sec. 8.9.
Indeed, partition [a, b] into subintervals (je*^, je*), in each of
which we take a point £*. Consider the auxiliary discrète random
variable £ which can take on the values
^1» ^2* • *• * • • •»
Let the probabilities of thé corresponding values of the discrète
random variable be plt p2, . . . , pk, . . . , p„:
p t = f ( & A x 2, . . . .
_____________P* = /(ift) Ax*. . . . . p„ —f (i„) Axn *
* At the same time, /(g*) A** is the probability that the continuous random
variable jc wi 11 assume a value in the interval **).
8.14 Numerical Characteristics of a Continuous Random Variable 433
The expectation of the given discrète variable l will be
M [l] = £ l kpk-
k= 1
or
m[s] = y (ix) a*+ y (g.) a*, +... + y (g*) a**+...
+ U ( U a xn= t u ( i * ) Ax*
*=i
Passing to the limit as max Ax*—<-0, we get
l'm S Ikf (tk) Ax* = Çxf (x) dx

m a x Ajcfc -► 0
r k= l aJ
The expression on the right is the expectation of the continuous
random variable x which can take on any value of x in the inter-
val [a, b]. The reasoning is similar for an infinité interval, that
is, for the expression (1). Formulas (1) and (T) are similar to for
mula ( 1) of Sec. 8.9 for a discrète random variable. We use the
symbol mx for the expectation.
The expectation is called the centre of probability distribution
of the random variable x (Fig. 177). If the distribution curve is
symmetric about the #-axis, that is, if f(x) is an even function,
then clearly
00
M [x] = ^ x/ (x) dx = 0
—oo
In this case the centre of the distribution will coïncide with the
coordinate origin (Fig. 178). Let us consider a centred random
variable x —mx. We will find its expectation:
oo oo oo
M [x— mx\ = J (x— mx) f ( x ) d x = J xf (x)dx— mx J f(x)dx
—oo —oo —oo
== mx—mx• 1 = 0
The expectation of a centred random variable is zéro. ,
31
484 Ch. 8 The Theory of Prohability
Définition 2 . The variance of a randorn variable x is the expec

tation of the square of the corresponding centred randorn variable:
où
D [x] = J (x— mx)-f(x)dx (2 )
—x
Formula (2) is similar to formula (2) of Sec. 8.10.

Définition 3. The standard déviation of a randorn variable x is
the square root of the variance:
- '-----=r f 00
a [x] = Ÿ D [x] - \ J (x— mxy f (x) dx (3)
— 00
This formula is similar to formula (3) of Sec. 8.10. When we

consider some illustrative examples, vve wi 11 see that as in the
case of a discrète randorn variable,
the variance and the standard dé
viation are measures of the disper
sion of values of the randorn va
riable.
Définition 4. The value of the
randorn variable x for which the
density function is a maximum
is termed the mode (symbolized
by M0). For the randorn variable x, the distribution curve of
which is shown in Figs. 177 and 178, the mode coïncides with
the expectation.
Définition 5. A number (symbolized by AI*) is called the médian
if it satisfies the équation
$ f (x) dx — 5 f ( x ) d x ^ - j (4)
- X Àf
e
(Fig. 179). This équation may be rewritten as
P ( x < M , ) = P ( A 4 e <7) - - , i -
This means that there is an equal probability that the randorn

variable x will assume a value less than M e and greater than M e.
Note that the randorn variable x may not take on the value M e
a t a il.
8.Î5 Normal Distribution. The Expectation of a Normal Distribution 485
8.15 NORMAL DISTRIBUTION.

THE EXPECTATION OF A NORMAL DISTRIBUTION
Studies of various phenomena show that many random variables,
such, for example, as measurement errors, the latéral déviation
and range déviation of the point of impact from a certain centre
in gunfire, and the amount of wear in machine parts, hâve a den-
( i «« 11 /« iirA ii U ir 4- (/M*mi<1 a
no. — j
distributed (the term Gaussian dis- Fig. 180
tribution is also used). The so-called
normal curve (normal distribution curve) in shown Fig. 180. A table
of the values of the function ( 1) for a = 0 , a = l is given in the
Appendix, Table 2. A similar curve was investigated in detail
in Sec. 5.9 of Vol. I.
First of ail, we will show that the density function (1) satisfies
the basic relation (5) of Sec. 8.12:
$ f{x)dx*= 1
— QD
Indeed, introducing the notation

^ r = t, dx = a V 2 d t
o V2
we can Write
since (see Sec. 2.5, Example 2)

oo
J e -1 ' d t —V n
We will détermine the expectation of a random variable with

normal distribution [see (1)]. By formula (I) of Sec. 8.14 we hâve
oo
( 2)
Making the change of variable

486 Ch. H The Tlieory of Prububility
we get
+ V 2 a/, dx = o \ f 2 d t
Hence
CO
= j (a + V 2 o t)e -‘*dt
—ce
CO CO
The first intégral on the right is equal to V n. Let us compute

the second intégral:
co
J t-e~tld t = — j e - 1* — 00
= 0
Th us
mx = a (3 )
The value of the parameter a in formula (1) is equal to the

expectation of the random variable under considération. The point
x = a is the centre of the distribution or the centre of dispersion.
When x —a the function f (x) has a maximum and so the value
x —a is the mode of the random variable. Since the curve of ( 1)
is symmetric about the straight line x = a, then
a oo
J f ( x ) d x ~ ^ f (x) dx
—co a
That is, the value x = a is the médian of the normal distribution.
If in formula ( 1) we put a = 0 , then we
get
/(*) = ■ V 2.1 -exp — 2a* (4 )
The corresponding curve is symmetric

about the ÿ-axis. The function f (x) is the
density function of normal distribution
of a random variable with centre of pro-
bability distribution coincident with the
origin (Fig. 181). The numerical characteristics of random va
riables with- distribution laws (1) and (4) defining the type of
dispersion of values of the random variable about the centre of
dispersion are determined by the shape of the curve, which is
independent of a, and therefore coincide. The quantity a déter
mines the amount of shift of the curve ( 1) to the right (for a > 0 )
8.16 Variance and Standard Déviation 487
or to the left (for a < 0). To save space, in most of our discussions
we will henceforth use the probability density function defined by
formula (4).
8.16 VARIANCE AND STANDARD DEVIATION
OF A NORMALLY DISTRIBUTED RANDOM VARIABLE
Suppose the probability density function (the frequency function)
of the random variable x is given by
, w “ ï 7 f r exp( - - & ) <■>

The variance of a continuous random variable is found from
formula (2), Sec. 8.14.
In our case
mx = a = 0
We hâve
Making the substitution -yz= = t, we get
D [* ]= — f t2 e~t,d t = f t-2t-e~1' dt
~t%
y n J yn J
—QO —QO
Integratiiig by parts, we obtain
Since
lim t-e~t* = 0 , ? e-*'dt = V n
t -►00 J
we finally hâve
D [x] = a* (2)
The standard déviation, in accordance with formula (3), Sec. 8.14, is

a [x] = V D [x] = a (8 )
Thus, the variance is equal to the parameter oa in the densïty-
function formula (1). We hâve already pointed out that the variance
488 Ch. 8 The Theory of Probabiliiy
characterizes the dispersion of values of the random variable about

the centre of dispersion. Let us now see how the value of the
parameter a- affects the shape of the distribution curve (frequency
curve). Fig. 182 gives the dis
tribution curves for the va-
lues a = - j , ct = 1, ct = 2 . We
see here that the smaller
ct, the greater the maximum
of the function f(x), also the
greater the probability of va
lues close to the centre of dis
persion (jc = 0 ) and the small
er the probability of values
distant from the origin. We
can thus say that the smaller the variance ct2, the smaller the
dispersion of values of the random variable.
8.17 THE PROBABILITY OF A VALUE OF THE RANDOM

VARIABLE FALLING IN A GIVEN INTERVAL.
THE LAPLACE FUNCTION.
NORMAL DISTRIBUTION FUNCTION
In accordance with formula (3), Sec. 8 . 12 , we détermine the
probability that a value of the random variable x having the
density function
l {x)~ ^ 7 ü exp
will fall in the interval (a, P)
8
P (a < x < P) = ^ f(x )d x
a
or
P ( a < ï < |)) = - ^ = . j ' e x p [ — !ïj2 îi] i , ( ,)
a
(Fig. 183). Making the change of variable
r~2
we get P-a
G V~2
l
P (a < x < P) =
va
J e~<'dt ( 1')
GV 2
8.17 The Laplace Function. Normal Distribution Function 489
The intégral on the right is not expressible in terms of elementary

functions. The values of this intégral can be expressed in terms
of the values of the probability intégral ,
a>(x) = ^ r ^ e ~ ‘t dt (2 )
0
Here are some of the properties of the function 0 (* ) which we

will use later on.
1 . <D(x) is defined for ail values of x.
2 . G (0 ) = 0 .
2 Ÿ~n .
3. d>( + oo) = - ^
y r ~ ~ '
o
4. O (x) is monotonie increasing on the interval (0, oo).
5. <D(x) is an odd function since
0 (_ * )= _ O (* ) (3)
6. The graph of the function O (jc) is shown in Fig. 184.
Detailed tables of the values of this function hâve been compiled
(a short table is given in the Appendix, Table 1).
Rewrite équation (T) using the theorem on the partition of the

interval of intégration:
P-a
0 a V2
1
P (a < x < P) = J e~‘, d t+ J e-t'd t
T Ï a-a 0
lO V~2
1 a V2 aV2
— j e -‘'d t+ j e-i'd t
Ÿ'n
o Q
This équation can be rewritten as follows:

a -a
P (a < * < P) = y e~‘‘ dt
With the aid of the function ®(x) [see (2)], we give a final
expression for the probability of the normally distributed random
variable x falling in the interval (a, p):
P (a < x < p) = (4)
K k r )-* (r ? ï)
When a = 0 we hâve
P (a < x < P) = y (5)
Equating the right members of ( 1) for the case a = 0 and of (5),
we get
P
J a V 2n (5')
a
We frequently hâve to compute the probability that a value of the

random variable will fall in the interval (a—l, a + l) symmetric
about the point x = a (Fig. 185). Formula (4) then takes the form
f ( a - , < ; < „ + , ) = £ [ ® ( - p T ) —'® ( - ï t t ) ;
Noting that <P ^ -----y=^j = —(P ^ j [see formula (3)], we

finally get
P (a — 1< x < a+1) = (P — ) ( 6)
8.17 The Laplace Function. Normal Distribution Function 491
The right side does not dépend on the position of the centre of
dispersion, and so for a = 0 as well we get
P H < ï< 0 =® (^ -) (7)
Example 1. A random variable x obeys the normal distribution law with centre
of dispersion a = 0.5 and variance o2 = -^-. Détermine the probability that a
8
value of the random variable x will fall in the interval (0.4, 0.6) (Fig. 186).
Solution. Here ----r=r = 2. By formula (4) we hâve

o y 2
P (0.4 < x < 0.6) = ÿ { ® [ 2 (0.6—0.5)]-« P [2 (0.4—0.5)]}
= 1 {0 (0 .2)- ® ( - 0 .2)}
But <D( — 0.2)= — CD(0.2) [see formula (3)] and so we can write
P (0.4 < x < 0.6) = -i- [O (0.2) + O (0.2)] = ^ (0-2)

Using the table of values of the function <D(x) in the Appendix, we find
P (0.4 < x < 0.6) = 0.223
Example 2. The length of an item manufactured on an automatic machine
tool is a normally distributed random variable with parameters M [x] = 10,
a2 = ^ . Find the probability of defective production if the tolérance is
200
10 ± 0.05.
1
Solution. Here, a = 1 0 , 10, a = ---- ^ =- . The probability of defective
a V~2 ’ 10 V 2
production, pê/ is, in accordance with (4),
Pde/= i — P (9-95 < x < 10.05)
= 1 (10 (10.05— 10)]— ® [10 (9.95— 10)]}
== 1- {O (0.5)—O (-0 .5 )} = 1 - O (0.5) = 1—0.52 = 0.48

Example 3. Find the probability of hitting a strip of width 2 /= 3.5 mètres

if the errors of gunfire obey the normal distribution law with parameters a = 0,
<J= 1.9.
Solution. Here, a = — 1.75, p = 1.75, — \^=r = 0.372. From formula (7) we get
oV 2
P ( — 1.75 < x < 1.75) = O (1.75*0.372) = O (0.651) = 0.643
Note. In place of the function <D(jc), (2), frequent use is made
of the Laplace function
x
2 dt 8
( )
The Laplace function is connected with the function <D(x) by

a simple relation. In (8 ) make the change of variablç yL^- = z to
get
X
Thus
= (9)
and, obviously,
Ô ( x / 2 ) =i-<t>(x) ( 10 )
Using the function (x) and relation (9), weWrite (5) as follows:
P(a<x<P) = o ( | ) - ô ( £ ) (11)
and when a = 1
P ( a < x < p)=<b(p) — Ô (a)
A table of the values of the Laplace function lî>(x) is given in
the Appendix, Table 3.
Now let us define the normal distribution function. From formula
( 1), Sec. 8.13, we hâve
X
F (x) =- J / (x) dx
—»
1
a V ïïï I exP [ - T ^ l dx = P (—oo < x < x)
Using formula (4) for the case a — —oo, P = x, we get
8.18 Probable Error 493
but ®( — oo) = — 1 [see formula (3)]. y

Hence, i
Si
F(x) = ir '® (fFï) + 1] <12> Z
0 X
The graph of the function F (x) for
a —0 is shown in Fig. 187. Fig. 187
8.18 PROBABLE ERROR
The probable error (or ttieati error) is a measure of dispersion

that has found numerous applications in the probability, gunfire
and errors théories.
D é fin itio n 1. The probable error is a number E such that the

probability that a random variable obeying the normal distribu
tion law
, w = ï ? i ï exp ( “ ■»■)
will fall in the interval ( — E, E) is ÿ (Fig. 188); that is,
P(-£ < * < £ ) = ! ' ( 1)

For any random variable x subject to the normal distribution
law with centre of dispersion at x = a, the probable error E
(Fig. 189) satisfies the relation
P (a —E < x < a + E) = Y ( 2)
Let us express the standard déviation o in terms of the pro

bable error E.
We express the left member of ( 1) in terms of the function O(x);
E
r < -£ < ;< £ )= J (3)

-£
494 Ch. 8 The Theory of Probabitiiy
From formula (7), Sec. 8.17, we get

p< -£ < ;< £ ) (4)
The left members of ( 1) and (4) are equal, and so also are the
right members:
Using the table of values of the function O(x), we find the

value of the argument x ^0.4769 for which 0 (x ) = ÿ . Hence
— 5 ^ = 0.4769
aY 2
The number 0.4769 is ordinarily denoted by p:
E
0.4769 ( 6)
a V"2 ~ P '
whence
E =ga V 2 ï
(7)
G== P V"2 J
8.19 EXPRESSING THE NORMAL DISTRIBUTION

IN TERMS OF THE PROBABLE ERROR.
THE REDUCED LAPLACE FUNCTION
Expressing the parameter a in terms of the parameter E by
formula (7), Sec. 8.18, and substituting into (4), Sec. 8.15, we
get an expression of the distribution in terms of the probable error:
fM = r h e x p ( - p' p ) (l)
The probabilityof the random variable (an error, for example)
falling in the interval (a, P) will, in accordance with (5), Sec.
8.17, be
P ( a < * (p4) (3)
B oc
The numbers and -g- in the right member of (2) are determined
by the nature of the problem; p = 0.4769.
8.19 The Reduced Laplace Function 495
In order to avoid multiplying by p in each computation, tables

hâve been compiled for the function Ofpx). This function is de-
noted by <D(x):
* Ô (x) = 0 (px) (4)
and is called the reduced Laplace function. See Table 1 in the
Appendix for values of this function.
On the basis of (2 ), Sec. 8.17, the function ô (je) is defined by
the intégral
p*
Making the change of variable t — pz, we get

X
Ô>(x) = -^=r [ e - p ^ d z (5 )
’ 31 o
Let us express the right member of (2) in ternis of the reduced
Laplace function-
P ( a < * < P) = ! [ è ( - | ) - Ô ( ! - ) ] (6 )
In particular, the probability of a value of a random variable

falling in a symmetric interval about the centre of dispersion
( - / , /) is, by (3),
H1
©>
A
A
II
(7)
î
P(0 < * < / ) = -• < î> (-) (8 )
Note that if the expectation a=£0, the probability of the ran

dom variable x falling in the interval (a, P), will be expressed
as follows in ternis of the probable error E [see formula (4),
Sec. 8.17]:
P (a< ^ < p ) = ! [ t f ( p ^ ) - < D ( p ^ ) ] (9)
This équation is expressed in terms of the reduced Laplace func

tion as follows:
p (a < i < W - 4 - [ é ( t £ ) - ( i . ( 2 = 2 ) ( 10)
8.20 THE THREE-SIGMA RULE.

ERROR DISTRIBUTION
In practical computations, the unit of measurement of the dévia
tion of a normally distributed random variable from the centre of
dispersion (the mathematical expectation) is taken to be
the root-mean-square (standard) déviation a. Then, by formula (7),
Sec. 8.17, we get some useful équations:
P ( — cr < x < o ) = <& = 0-683
P ( —2ct < je < 2o) = <D = 0.954
P ( —3 c t< x < 3ct) = (D (-— ) = 0.997
These results are depicted geometrically in Fig. 190.
We can be almost certain that the random variable (error) will
not départ from the absolute value of the expectation by more
than 3o. This proposition is called the three-sigma rule.
In gunfire theory and in the processing of various statistical

data it is often useful to know the probability of the random
variable x falling in the intervals (0, £), (£, 2£), (2£, 3£), (3£, 4£),
(4 £, 5£) for the density function defined by (1), Sec. 8.19. In
many cases, a knowledge of these probabilities reduces computa
tions and helps in the analysis of the phenomena.
In computing these probabilities, we will use formula (8 ), Sec. 8.19,
and the table of the function (D(x):
P (0 < 3c < £) = ÿ ô ( l ) = 0.2500
P (£ < x < 2£) =ÿ[<D (2) — <D(l)]-0.1613
P(2£ < x < 3 £ ) = 4 [<î> (3)—Ô (2)] = 0.0672
P (3£ < x < 4£) = ÿ [cî> (4)— CD(3)] -0.0180
P (4£ < x < cx>) = 1 [Ô (oo) - Ô (4)] = - ( 1-0.9930)
= 0.0035
8.2Î Meari Arithmetic Error 497
The results of the computations are geometrically represented in

Fig. 191, which is called the error distribution. These calculations
imply that it is practically a certainty that a value of the
random variable will fall in the interval (-—4£, 4£). The proba-
bility that this value of the random variable will fall outside that
interval is less than 0 .0 1 .
Example 1. One shot is fired at a strip 100 métrés wide aimed at the middle
line of the strip perpendicular to the plane of the path of the projectile. Dis
persion obeys the normal law with lon
gitudinal probable error £ = 20 métrés.
Détermine the probability of hitting the
strip (Fig. 192). In gunfire theory, the
longitudinal probable error is denoted by
B i o n , the latéral probable error by B l a f.
Solution. Take advantage of formula
(7), Sec. 8.19. In our case 7 = 50 métrés,
£ = Blon = 20 métrés. Hence
P (-5 0 < * < 5 0 ) = Ô ( g )
= 0 (2 .5 ) = 0.9082 «0.91 Fig. 192
Note. An approximate solution
may be obtained without resorting to tables of the function d> (jc),
simply by using the error distribution (Fig. 191). In our case,
Z= 2.5£, hence
P ( — 50 < x < 50) = 2 (0.25+0.16+0.01) -0 .9 0
Example 2. Experiment shows that the error of an instrument used for
measuring distances obeys the normal distribution law with probable error £ = 10
métrés. Find the probability that the distance determined by this instrument
will départ from the true distance by no more than 15 métrés.
Solution. Here, / = 15 m, £ = 10m. By formula (7), Sec. 8.19, we get
P ( - 1 5 < x < 15) = Ô ( ^ + 0 ( 1 . 5 ) = 0.6883 «0.69
8.21 MEAN ARITHMETIC ERROR
To characterize errors we introduce the concept of the mean

arithmetic error, which is equal to the expectation of the absolute
value of the errors. We dénoté the mean arithmetic error by d.
Let us détermine the mean arithmetic error if the errors x obey
the normal law (4), Sec. 8.15. By a formula similar to (2) of
Sec. 8.15, we get (a --- 0)
d= J j 1 - E » ) 'd x - 7 7 5 1 *'exp (■~ £ ) ' ix
- 7 7 5 [ “ 0’exp( - w - ) ] . ' “ -pH

32—2082
Thus, the mean arithmetic error is expressed in terms of the root-

mean-square (standard) déviation a thus:
d O)
8.22 MODULUS OF PRECISION.

RELATIONSHIPS BETWEEN THE CHARACTERISTICS
OF THE DISTRIBUTION OF ERRORS
In many processes, particularly in gunfire theory, the probabi

lity density function of the normal law is given as
f(x )= -£ = e -» * ( 1)
Vn
Comparing (4), Sec. 8.15, and (1) of this section we see that the
newly introduced parameter h is expressed in terms of the para-
meter a thus:
The quantity h is inversely proportional to a, that is, it is inve--

sely proportional to the mean-square error or the standard dévi
ation. The smaller the variance a 2, that is, the smaller the disper
sion, the greater the value of h. For this reason, h is termed the
modulus of précision.
From (2) of this section and ( 1) of Sec. 8.21 we get
° hV T (3)
d—-U
h VU
(4)
The probable error E is expressed in terms of the modulus of

précision h via formula (7), Sec. 8.18, and (3):
E -$ (5)
It is sometimes necessary to express one characteristic of error

distribution in terms of another. The following équations are found
to be useful:
£ = p 1/2=0.6745, = p V n = 0.8453, - j = j / y 1.2533
a_____ 1 _ d 1 [ ( 6)
= 1.4826, 1.1829
E p y j E p j/7 ï
8.23 Two-Dimensional Random Variables 499
8.23 TWO-DIMENSIONAL RANDOM VARIABLES
Two-dimensional random variables are involved for example when

considering the process of target destruction on a plane (xOy).
The value of a two-dimensional random variable is determined
by two numbers: x and y . We will dénoté the two-dimensional
random variable itself by (x , y). Let x and y assume discrète values
Xi and y{. Let each pair of values (xh y() of a certain set be asso-
ciated with a definite probability pif. We can form a table of
probability distribution of the discrète two-dimensional random
variable:
m n
2 2
/=1 /=1
p /y = l (U
Now let us détermine a continuous two-dimensional random vari

able. The probability that a value of the two-dimensional random
variable (a:, y) satisfies the inequalities x < x < x + Ax, y < y <
< y-{- Ay will be denoted thus: P (x < x < x + Ax, y < y < y + Ay).
Définition 1. The function f(x> y) is called the probability density
function of the two-dimensional random variable (x, y) if, to within
infinitesimals of higher order than Ap = |/Ajc2 + A y2, the following
équation holds true:.
P ( x < x < x + Ax, y < ÿ < y + Ay) « / ( x , y) Ax Ay (2)
Formula (2) is similar to formula (2) of Sec. 8.12.
We consider a rectangular coordinate system (xOy). If the values
of the random variable (x, y) are denoted by points of the plane
having the appropriate coordinates x and y , then the expression
50Ü Ch. 8 The Theory of Probabilïty
P (x < x < x + Ax, y < y < y + Ay) dénotés the probability that the
two-dimensional random variable (x, y) will assume a value denoted
by a point lying in the hatched rectangle As (Fig. 193). We will
say that the “value of the random va
riable lies in the domain As”.*
The probability P (x < x: < x + Ax,
y < y < y + Ay) will also be denoted
by P [(at, i/)cAs]. Using this nota
tion, we can rewrite (2 ) as
P [(a, i/)cA s] œ f ( x , y) As (3)
We will now prove the following the-
orem, which is analogous to Theorem 1, Sec. 8.12.
Theorem I. The probability P [(x ,ÿ )c D ] that a two-dimensional
random variable (x, y) with density function f (x, y) will be in domain
D is expressed by the double intégral of the function f (x, y) over D,
that is,
P [(x, ÿ)cD] = SJ /(*• y)dxdy (4)
D
Proof. Partition D into subdomains As, as is done in the theory

of double intégrais. For each subdomain we write (3) and add the
left-hand and right-hand members of the resulting équations. Since
2 A s = D and 2 P [(*. f/)c As] = P [(x, y)<=D]
we get an approximate équation to within infinitesimals of higher
order than As:
P [(x, y) cD] y) As
Passing to the limit in the right-hand member of this équation
as As—>-0, we get on the right a double intégral and, on the
basis of the properties of an intégral sum, the exact équation
P [(À, ÿ)<=D] = $$ f(x, y)dxdy

*D
The proof is complété.
Note 1 . If the domain D is a rectangle bounded by the straight
Unes x = a, jc = P, y = y, y = 6 (Fig. 194), then
M
P [ a < x < p , y < y < 6 ] = J $ /(* , y)dxdy (5)
a v
* The shape of the subdomain in (3) may be arbitrary.

S.23 Two-Dimensional II undont Variables 501
Note 2. Like (1), the following équation

oc oc
S S f(x,y)dxdy~ 1 (6 )
—oc —ac
holds true since it is a certainty that the two-dimensional variable
will assume some value. We put f(x, y) = 0 wherever the function
f{x,y) is not defined by the meaning of the problem.
y>
9
y w m m .
0 oc p X
Fig. 194 Fig. 195
If the domain D is the sum of rectangles of the kind shown

in Fig. 195, then the probability of the random variable falling
in such a domain is defined as the sum of the probabilités for
the separate rectangles, that is, as the sum of definite intégrais
over each rectangle:
P [(x, y)<z.D] — P [(x, ÿ)cD ,] + P [(x, ÿ ) c D 2] + P [(x, ÿ)<zD3]
Example. The density function of a two-dimensional random variable is
given by the formula
f(x' ^)= ji* (1 + xa) (1 +y2)

Find the probability that a value of the random variable will fall in the rectangle
bounded by the straight lines x = 0, * = 1 , y = - ^ f= , y = V T .
Solution. By formula (5) we obtain
1y3
l dxdy
^ 1
A
A
Ÿ s]
’ y 'T l - i I I (!+**) (!+*•)
V3
1 VT
_ î C dx _ r dy _ 1 arctanx VT
arctan y 1
Jl- J l+ A i+ y * Jl*
o i VT
VT
= _ L / 'ü _ o U - - - W -
Jis V4 JV3 6J 24
D é fin itio n 2 . The function

y x
F(x,y)=\ J f(u, v)dudv (7)
—00—00
is termed the distribution function of probabilités of the two-
dimensional random variable (x, y).
It is clear that the distribution function expresses the probabi
lity that x < x, y < y , that is,
y> F {x, y) = P (x < x, y < y)
Geometrically, the distribution function
* expresses the probability that a two-di-
mensional random variable lies in the
infinité rectangle hatched in Fig. 196.
By the theorem on the différentiation
Fig. 196
of a definite intégral with respect to a pa-
rameter, a relation is established between
the density function and the distribution function:
w = S j (x' v)dv
( 8)
The probability density function of a two-dimensional random

variable is the mixed second dérivative of the distribution function.
8.24 NORMAL DISTRIBUTION IN THE PLANE
Définition 1. The distribution of a two-dimensional random

variable is termed normal if the density function of this variable
is expressed by the formula
f (*» y) — 2n(Jxay exP ( — ~ 2^j)
The graph of this function is a surface (see Fig. 197).
The centre of dispersion of a random variable with distribution
law ( 1) is the point (0 , 0 ).* ax and ay are called the principal
standard déviations.
* If the centre of dispersion lies at the point (a, b), then the distribution
is given by the formula
/ (x, y) (* —a)2 ( ÿ - t )81
(1')
2 noxoy eXp 2o| 2ol J
8.24 Normal Distribution in the Plane 503
Rewrite formula (1) as follows:
(2)
Thus, f (x, y) may be regarded as a product of Jwo density func-
tions of normally distributed random variables x and y. As in the
case of a one-dimensional random variable, we define the principal

probable errors, Ex and Ey [see formula (7), Sec. 8.18] as follows:
Ex — P°x V'^'t Ev = P°yV2 (3)
Sutistituting ox and oy, expressed in terms of Ex and £ , into
formula ( 1), we get
/ (x , y) (4)
r /xp[ - p! (-§+€>;
Let us consider the level Unes of the surface (4):
- S + -fï = * 2 = const ant (5 )
Ey
[here, f (x, y) = const], The level Unes are ellipses with semiaxes
equal to kEx and kEy. The centres of the ellipses coïncide with
the centre of dispersion. These ellipses are called ellipses of disper
sion. Their axes are termed axes of dispersion. The unit ellipse of
dispersion is an ellipse whose semiaxes are equal to the probable
errors Ex and Ey. The équation of the unit ellipse is obtained by
putting k = \ in équation (5):
( 6)
The total ellipse of dispersion is an ellipse whose semiaxes are

equal to 4Ex and 4Ey. The équation of this ellipse is
=1 (7)
In the next section we will establish the fact that the proba-
bility of a two-dimensional random variable falling in the total
ellipse of dispersion is equal to 0.97, which makes a hit a practical
certainty.
8.25 THE PROBABILITY OF A TWO-DIMENSIONAL
RANDOM VARIABLE FALLING IN A RECTANGLE WITH SIDES PARALLEL
TO THE PRINCIPAL AXES OF DISPERSION UNDER THE NORMAL
DISTRIBUTION LAW
Let
By formula (5), Sec. 8.23 (see Fig. 194), the probability of a random
variable falling in the rectangle bounded by the straight Unes
x = a, x = y —y, is expressed thus:
P (a < x < p, y < ÿ < 6) = j j ^ ^ e x p [ “ P * ( | k j | ) ] dxdy (1)

a y
Representing the integrand as a product of two functions, we can
Write
P (a < x < P, y < y < ô)
a
yfe:exp(-p!liM "Hh;exp(-p‘| ) *
V
( 2)
and, on the basis of formula (6 ), Sec. 8.19, we finally get

P (a < x < P, y < y < 6 )
If we put a = — / „ P = /x, y = — ^2. à = l t here, that is, if we

regard a rectangle with centre at the origin, then by (7), Sec. 8.19,
formula (3) will take the form
P ( - / l < î < / 1, - Z , < ÿ < ^ = d » ( ^ ) ô ( A ) (4)
Note. The problem on the probability of a random variable
falling in a rectangle with sides parallel to the coordinate axes
could also be solved in the following manner. Falling in the
rectangle is a compound event consisting in the coincidence of
two independent events: falling in the strip — lt < x < Zt and
falling in the strip — /, < ÿ < / s. (To save space, we consider a
rectangle with centre at the origin.) Let the density function of
8.25 Probability of a Two-Dimensional Random Variable 505
the random variable x be
The density function of the random variable y is
We compute the probability of the random variable falling in

the strip — lY< x < l x and in the strip —/2 < { / < / 2. By formula
(7), Sec. 8.19, we get
P ( - / 1 (^ )

The probability of a compound event — falling in the rectangle —
is equal to the product of the probabilités:
P (a < * < P, y < y < 6) = P (— U < x < lx) P (— /2 < y < /2)
=H k ) H t y
We hâve obtained formula (4).
Example. Shots are fired at a rectangle with sides 200 m and 100 m bounded
by the Unes
x ——100, ^=100, y = —50, y = 50
The principal mean errors are equal to Ex = Blon = 50 métrés and E y — Biaf
= 10 métrés, respectively. Find the probability of hitting the rectangle in a
single shot (Fig. 198).
Solution. In our case

l x= 100, /2 = 50, Ex = 50, £ y= 10
Putting these values into (4) and using the table of values of the function O (x)
(see Table 1 of the Appendix), we find
p = Ô » ^ ^ ^ . ( î i ^ j = ( î ) ( 2 ) . < i ( 5 ) = 0.8227 0.9993 = 0.8221
8.26 THE PROBABILITY OF A TWO-DIMENSIONAL RAN DOM

VARIABLE FALLING IN THE ELLIPSE OF DISPERSION
In the theory of errors one cornes up against the following problem.
It is required to compute the probability that a random variable,
say, an error in the plane, will fall in the ellipse of dispersion
if the density function is given by formula (4), Sec. 8.24. By (4)

of Sec. 8.23, we get
dxdy (2)
[(î' »,cDW J ^ " p[ - ,>,(‘â +i ï ) ’
'e l
where the domain Del is bounded by the ellipse (1). We make

the change of variables
x = Exu, y = Eyv
Then the ellipse Del will be carried into the circle
u2+ v2= k 2 (3)
Since the Jacobian of the transformation is equal to I = Ex Eyy
équation (2) becomes
P[(x, y)czDel] p2e“pt<ut+vt^dudv (4)
" Dk
In the last intégral we pass to polar coordinates
a ^ rc o sq p , u ^ rs in c p
Then the right member of (4) takes the form
2j i k
P[(x, ~y)<=Del] = i ^ p * e - e ,r‘ r d r d y
oo
Carrying out the computations in the right member, we get the
expression of the probability of falling in the ellipse of dispersion:
= l - r P ,‘1
P [ ( ^ ÿ ) c D ((] (5)
Let us consider some particular cases. The probability of falling
in the unit ellipse of dispersion occurs if we put k = 1 in formula (5)
P[(x, ÿ ) c D J A=1 = l- e - P * -0 .2 0 3 (6)
The probability of falling in the total ellipse of dispersion (7),
Sec. 8.24, occurs if in (5) we put k=- A:
P[(x, y )c D el]k=t = 1—e - 16P’ = 0.974 (7)
8.27 Problems of Mathematical Statistics 507
Let us consider the particular case where in (4), Sec. 8.24,

Ex ~ Eu — E. The ellipse of dispersion (5), Sec. 8.24, turns into
circle .
x* + 0 * -A * £ * (8)
with radius R --k E . The probability of a two-dimensional random
variable falling in a circle of radius R is, in accord with (5),
P[(*. ÿ)cO s ] = 1—exp( — p*-|î) (9)
D é fin itio n 1. The mean radial error is a number ER such that
the probability of a two-dimensional variable falling in a circle
of radius R = Er is equal to
The définition implies that the quantity R = E r is determined
from the relation
1 —exp ( — p2— ) = y
From a table of values of the exponential function we find
Er = 1.75 E
8.27 PROBLEMS OF MATHEMATICAL STATISTICS.

STATISTICAL DATA
Statistical data are obtained as a resuit of observation and record-
ing of mass random (chance) phenomena. An instance of such data
are the errors of measurement.
If an observed quantity is a random variable, then itisstudied
by the methods of probability theory. To grasp the nature of this
random variable, one has to knovv its distribution law. Determin-
ing the laws of distribution of such quantifies and estimating the
values of the parameters of distribution on the basis of observed
values constitutes a problem in mathematical statistics.
Another problem of mathematical statistics is the establishment
of methods of Processing and analyzing statistical data to obtain
certain conclusions necessary in organizing an optimal process
involving the quantities under considération.
The following are some illustrative examples of observations of
phenomena that yield statistical data.
Example 1. The distance to an object is measured a large number of times
with the aid of a mcasuring instrument yielding a séries of values of the observed
quantity. They are called observed values (we wi 11 use this term for any value
obtained in studying any phenomenon).
The values thus obtained require systématisation and processing
before any conclusions can be drawn from them.
508 C h . 8 T h e T h e o r y o f P r o b a b i l i t y
As has already been pointed out, the différence 6 between an

observed value x and the true value of the observed quantity
a ( x — a = 6) is called the e r r o r o f m e a s u r e m e n t . The errors of
measurement require mathematical Processing in order to obtain
spécifie conclusions.
Examp'.e 2. In mass production one has to consider the amount of déviation
of a certain dimension of a finished article (say, length) from a given dimension
(fabrication error).
Example 3. The différence between the coordinate of the point of impact in
gunfire and the coordinate of the target is the error of gunfire (dispersion). These
errors require mathematical investigation.
Example 4. The results of measurements of the amount of departure of the
dimensions of an item after a period of operation from its dimensions prior to
operation (design dimensions) require mathematical analysis. Such déviations
can also be regarded as “errors”.
It is clear from the foregoing that these quantities are random
variables and each observed value should be regarded as a particular
value of the random variable.
For instance, the range error (dispersion) in gunfire is determined
by the error obtained when weighing the charge, the error in weight
in the manufacturing process, the aiming error, the range determin-
ing error, variations in weather conditions, and so forth. AU these
are random variables, and dispersion, as the resuit of their joint
action, is also a random variable.
6.28 STATISTICAL SERIES. HISTOGRAM

The statistical findings obtained in observations (measurements)
are tabulated in two rows. The first row contains the number i
of the measurement, the second, the value x t of the quantity x
being measured:
i 1 2 3 ... i ... n
X i * 2 * 3 ... X i ... X n
This table is called a s i m p l e s t a t i s t i c a l s é r i é s . If there are a

large number of measurements, the statistical data is not readily
surveyable and an analysis is complicated. Therefore, the data
obtained from a simple statistical sériés are given in a f r e q u e n c y
t a b l e . This is done in the following manner.
The whole interval of values of the quantity x is partitioned
into subintervals of equal length (a0, a , ) , ( a l t a 2) .......................( a * - i , a * .)
and then we count the number of values of x falling in the m k
interval (ak^ v ak). The values at the end of the interval are referred
8.28 Stalisiical Sériés. Histogram 509
eitlier to the left or right intervals (sometimes half is placed in

the left interval and half in the right). The nuinber
^ = * ( 1)
is the relative frequency corresponding to the interval (a*.,, ak).

Clearly,
2 pü = i
*=i
(2)
Having processed the data in this manner, \ve then construct
a table consisting of three rows. The first row indicates the intervals
in order of increasing akJ the second row, the corresponding numbers
mk, and the third row, the frequencies pk ^-,^— \
This is what is called the frequency table. It may be converted

into a diagram, as follows. On the .v-axis we mark of! points aw,
alf . . . , ak1 ...» a?.. On each interval [ak_ly ak] as a base, we
construct a rectangle whose area is equal top£. The resuiting figure
is called a histogram (Fig. 199).
P '
~ V » ,
a 0 a, 0 aZ a3 a« a5 a6 a7 a8 X
Fig. 199
On the basis of the frequency table and the histogram we can

give an approximate construction of a statistical distribution function.
The Processing is continued as follows. The midvalues of the
intervals (<!*_,, ak) are denoted by xk and this value is taken to
be the value of the measurement, which is repeated mk times. Then
in place of the frequency table we construct the following table;
510 Ch. 8 Thé Theory of Probability
This procedure is based on the fact that ail values in the interval
(a*_lf ak) are close-lying values and are therefore considered to be
equal to the abscissa of the midvalue of the interval x\.
Example. One hundred range measurements were made yielding the following
results (frequency table):
O O
290-320
140-170
260-290
170-200
200-230
230-260
Intervals 1
1
O O
00
mk 2 5 16 24 28 18 6 1
Pk 0.02 0 05 0.16 0.24 0.28 0.18 0.06 0.01
From this frequency table we compile a histogram (Fig. 200).

8.29 Determining a Suitable Value of a Measured Quantity 511
We then construct the following table:
*k 95 125 155 185 215 245 275 305

•
mk 2 5 16 24 28 18 6 1
Pk 0.02 0.05 0.16 0.24 0.28 0.18 0.06 0.01
8.29 DETERMINING A SUITABLE VALUE

OF A MEASURED QUANTITY
Suppose that in measuring a certain quantity we obtained the
measurements xlt x2i . . . , xn. These values may be regarded as
particular values of the random variable x. For the suitable value
of the quantity being measured we take the arithmetic mean of
the values:
n
2 xi
i= 1 / 1\
The quantity ml is called the statistical mean.

If the number of measurements n is great, then use is made of
the tabulated data considered in Sec. 8.28, and ml is computed
as follows:
* _ + x 2m2 + . . . + xkmk + . . . + XytnK
or, using the notation (1), Sec. 8.28,

x
== 2 ~
x kPi (2)
k—1
This value is termed the weighted mean.
Note. From now on the results of computations via formulas (1)
and (2) will be denoted bv the same letter. This wi 11 also be true
of formulas (3) and (4).
It can be proved that, under certain restrictions, the statistical
mean tends in probability to the expectation of the random variable
x as n —►o o . This assertion follows from the Chebyshev theorem.
Let us define the statistical variance: *
2 (Xi — mx)*
= n-----
~ (3)
* Actually, it is better to compute the statistical variance by means of for
mula (5) of Sec. 8.30.
512 Ch. 8 The Theory of Probabilily
This quantity characterizes the dispersion of values of an observed

variable.
If we make use of the material of the tables in Sec. 8.28, then
the statistical variance is determined from the formula
mlYPk (4)
k= 1
This formula is similar to formula (2) of Sec. 8.10.
Example. Détermine the statistical mean and the statistical variance on the
basis of the statistical data of the Example in Sec. 8.28.
Solution. By formula (2) we get
n
*>
m £= — = 2 1 *aPÂ= 95-0.02+ 125-0.05+155-0.16
k= 1
+ 185-0.24+ 215-0.28+ 245*0.18 + 275-0.06+ 305-0.01 __ 201.20
By formula (4) we obtain
n
2 (**—m i)2 x x
D * [x] = — ---- - --------- = X = X * k P k —m * x
k= 1 k= 1
= 952*0.02+ 1253*0.05+ 1552*0.16+ 1852*0.24 + 2152*0.28
+ 2452*0.18 + 2752•0.06 + 3052•0.01 — (201.20)2= 1753.56
8.30 DETERMINING THE PARAMETERS

OF A DISTRIBUTION LAW. LYAPUNOV’S THEOREM.
LAPLACE’S THEOREM
Let x be a random variable, say, the resuit of a measurement,

let a be the quantity being measured, and 6, the error of measu-
rement. Then these quantifies are connected by the relation
6 = x — a, jc= a + 6 (1)
Numerous experiments and observations hâve shown that the errors
of measurements (after excluding crude errors and after excluding
any systematic error, that is, an error that is constant for ail measu
rements—say the instrument error—or that varies by a known law
from measurement to measurement) obey the normal distribution
law with centre of distribution at the origin. This is confirmed by
theoretical arguments as well.
If a random variable is the sum of a large number of random
variables, then for certain restrictions the sum obeys the normal
distribution law. This assertion is stated in the so-called central
limit theorem of A. M. Lyapunov (1857-1918). The theorem is
given here in simplified form.
8.30 Determinin% the Parameters of a Distribution Law 513
Theorem 1. If independent random variables xv x2, x„ obey

one and the sanie distribution law with expectation a (without loss
of generality, we can assume a = 0) and variance o2, then, as n
2 aï
increases without bound, the sum y„ = l~lp= is asymptotically nor-
oy n
mally distributed (y„ is normalized so that M [yn\ = 0, D [yn\ = 1).
The practical significance of the Lyapunov theorem consists in
the following. We consider a random variable, say, the déviation
of a certain quantity from the given quantity. This déviation is
due to many factors, each of which makes a certain contribution
to the déviation, say in the case of gunfire, the departure of the
point of impact from the aiming point is due to the aiming error,
the range error, the shell-fabrication error, etc. We do not even
know ail of the contributing factors, neither do we know the
distribution laws of the component random variables. But from the
Lyapunov theorem it follows that the random variable of the total
déviation obeys the normal distribution law.
Lyapunov’s theorem implies that if xlt xt, . . . . xn are the
results of measurements of a certain quantity (each of the x,- is
a random variable), then the random variable, which is the arith-
metic mean,
~ X, - ! - T ; + ■ . ■+ X n
n
is asymptotically normally distributed for sufficiently large n if

the random variables x-, obey one and the same distribution law.
The theorem holds true also for a sum of random variables
having unlike distribution laws, subject to certain additional con
ditions, which as a rule are fuifilled for random variables ordina-
rily encountered in practical situations. Expérience shows that for
the number of terms of the order of 10, the sum may be regarded
as normally distributed.
Let us dénoté by a and o2 the approximate values of the expec
tation and variance. We can then write down approximately the
laws of distribution of the random variables ô and x:
(3)
On the basis of experimental findings, the parameter a is deter-

mined from formula (1), Sec. 8.29:
a= (4 )
This follows from the so-called theorem of Chebyshev (1821-1894).

Without dwelling on the proof, we note that the parameter a is
more naturally determined from the following formula instead of
from (3) of Sec. 8.29:
2 (xi— <j)a
<•=■ /K\
Note that the right member in (5) and the right member of (3),
Sec. 8.29, differ by the factor , which is close to 1 in prac-
tical problems.
Example 1. Write down the expression for the distribution law of the random
variable on the basis of the measurements given in the Example of Sec. 8.28,
and the results of computations carried out in the Example of Sec. 8.29.
Solution. On the basis of the computations carried out in the Example of
Sec. 8.29, we obtain
a = m* = 201
âî = — ^— D* = ~ . 1754=1771
n—1 99
ô = yT77l«:41
r/ x 1 [ (x —201 )21
Note. If a statistical distribution function has been obtained

for some random variable xt then the question of whether the
given random variable obeys the normal law or not is sometimes
settled in the following manner.
Suppose we hâve the following values of a random variable:
X» *2, • • - ,
Find the arithmetic mean of a by formula (4). Find the values
of the centred random variable
V u y 2» • • • . y n
The absolute values of are then arranged in a sériés in in-
creasing order. If n is odd, then for the probable (mean) error Em
in the constructed sériés of absolute values, we take the absolute
8.30 Determining the Parameters of a Distribution Law 515
value | y rn | that occupies the ) th + 1 site; if n iseven, thenfor E m

we take the arithmetic mean of the absolute values occupying
sites with numbers y and y + 1 .
We then form the mean arithmetic error using the formula
j <=i
2 I y>I
(6)
d==— h—
From (5) we détermine the standard déviation
a - (7)
1 n— 1
E E
And then we find the ratios -?■ and — .
d a
£
For a random variable that obeys the normal law, the ratios y
£•
and — are equal, respectively, to 0.8453 and 0.6745 [see formula
CT
(6), Sec. 8.22]. If the ratios ^ and — differ from 0.8453 and
CT
0.6745 by a quantity of the order of 10%, then it is agreed, con-
ditionally, that the random variable y obeys the normal law.
A corollary to the central limit theorem is an important theorem
of Laplace conceming the probability that an event will occur not
less than a times and no more than p times. We give this theorem
without proof.
Theorem 2 (Laplace). I f n in d e p e n d e n t t r i a l s a r e m o d e in each
o f w h ich th e p r o b a b i l i t y o f o cc u rre n c e of even t A is p , th e n th e
f o llo w in g r e la tio n h o ld s tr u e :
^ < - < ^ [ H 7 f m ) - H ? T 7 k ) } <8>

w h erern is th e n u m b e r o f o c c u rre n c e s o f A , q = 1 — /?, a n d P (a < m < P )
is th e p r o b a b il ity th a t th e n u m b e r o f o cc u rren ce s o f A lie s b etw e en
a a n d f5.
The function O ( x ) is defined in Sec. 8.17.
We will demonstrate the use of the Laplace theorem in the
solution of problems.
Example 2. The probability of defective items in the production of certain
articles is p — 0.01. Détermine the probability that in 1000 articles therewillbe
no more than 20 defective items.
33
Solution. Here,
n = 1000, p = 0 .01 , (7= 0.99, a = 0 , P = 20
We then find
a — np _ 0 — 10 = _ 2 25
Ÿ"2 yltp q V"2 I / 9 T9
V>~nP _ 2 0 -1 0 _ „ 25
Y ' 2 ^n p q Y ' 2 f 9.9
P ( 0 < m < 2 0 ) = — j^<D (2.25) — <D (- 2 .2 5 )] = <D(2.25)
Using the table of the function <D(x), we find

P ( 0 < m < 2 0 ) = 0.9985
Note that the theorems of Bernoulli, Lyapunov, Chebvshev, and Laplace dis-
cussed above constitute what is known as the law of large numbers in probability
theory.
!. Two dice arc thrown at one time. Find the probability that a sum of 5
will turn up. Ans. 1/9.
2. There are 10 tickets in a lottery: 5 wins and 5 losses. Take two tickets.
What is the probability of a win? Ans. 7/9.
3. A die is thrown 5 times. What is the probability that a number of four
will turn up at least once? ,4ns. 0.99987.
4. The probability of hitting an aircraft with a rifle is 0.004. How many
men must shoot so that the probability of a hit becomes > 70%? Ans. n > 300.
5. Two guns tire at the same target. The probability of a hit from the first
gun is 0.7, from the second one, 0.6. Détermine the probability of at least one
hit. Ans. 0.88.
6. One hundred cards are numbered from 1 to 100. Find the probability
that a randomly chosen card has the digit 5. Ans. 0,19.
7. There are 4 machines. The probability that a machineis in operation at
an arbitrary time t is equal to 0.9. Find the probability that at time t at
least one machine is working Ans. 0.9999.
8. The probability of hitting a target is p = 0.9. Find the probability that
in three shots there wiH'be tbree hits. Ans. «0.73.
9. Box One contains 30% first-grade articles, Box Two contains 40% first-
grade articles. One article is drawn from each box. Find the probability that
bot h drawn articles are first-grade. Ans. 0.12.
10. A mechanism consists of three parts. The probability of defective items
in the manufacture of the first part is px — 0.008, the probability of defectives
in the second part is p2~ 0.012, the probability of defectives in the third part
is p3 = 0.01. Find the probability that the whole mechanism is defective.
Ans. 0.03.
11. The probability of a hit in a single shot is p — 0.6. Détermine the pro
bability that three shots will yield at least one hit. Ans. 0.936.
12. Out of a total of 350 machine^, there are 160 of grade One, 110 of grade
Two, and 80 of grade Three. The probability of defectives in the grade-one
category is 0.01, in the grade-two category, 0.02, in the grade-three category,
0.04. Take one machine. Détermine the probability that it is acceptable.
Ans. 0.98.
13. Suppose it is known that due to errors committcd in préparation, for gunfire,
the centre of dispersion of the shells (CDS) in the first shot can lie at (in range)
one of fïve points. The probabiIities that the CDS will lie at these points are equal
to Pi —~0.1, /)2- 0 .2 , p z--0 A , pA— 0.2, ph — 0.\. It is also known that if the
CDS lies in the first point, the probability (in range) of hitting the target will
be ^ —0.15. For the othcr points \ve will hâve for this case: p2 — 0.25, p3 =
= 0.60, pA— 0.25, pb = 0.15.
In the initial adjustment of sight, a shot was fired with a miss in fange.
Déterminé the probability that the shot is made at adjustment of sight corre-
sponding to each of the five indicated points of the CDS, that is, déterminé the
probabilités of causes (hypothèses) concerning distinct errors in the position of
the CDS after a trial (shot). Ans. 0.85, 0.75, 0.40, 0.75, 0.85.
14. A die is thrown 5 times. What is the probability that a six will turn
up twice and a non-six three times? Ans. 625 3888.
15. Six shots are fired. Find the probability that not ail are plus rounds if
the probability of a plus round is p = 1/2, the probability of a minus round is
q — 1/2 (gunfire at a “narrow” target). Ans. 31/32.
16. Referring to the conditions of the preceding problem, détermine the pro
bability that there will be three plus rounds and three minus rounds. Ans. 5/16,
17. Find the mathematical expectation of the number that tums up in a
single toss of a die. ,4ns. 7/2.
18. Find the variance of a random variable x given by the following distri
bution table*
jc 2 3 5
p 0.1 0.6 0.3
Ans. 1.05.
19. The probability of occurrence of event A in a single trial is equal to 0.4.
Five independent trials are carried out. Find the variance of the number of
occurrences of A. Ans. 1.2.
20. In a case of gunfire at a target, the probability of a hit is 0.8. The fire
is kept up till a first hit. There are 4 shells. Find the expectation of the number
of shells used. Ans. 1.242.
21. In firing at a “narrow” target, the probability of a plus round is p — 1/4,
the probability of a minus round is q = 3,4. Find the probability of a combi
nation of two plus rounds and 4 minus rounds in six shots. Ans. 0.297.
22. The probability that an article will be defective is p = 0.01. What is the
probability that a batch of 10 articles will yield 0, 1, 2, 3 defectives? Ans. 0.9045,
Q,0904, 0.0041, 0.0011.
23. Find the probabilités of obtaining at least one hit in the case of 10 shots
if the probability of hitting the target in a single shot is p = 0.15. Ans. 1 —
— (0.85)10 « 0.803.
24. A random variable x is given by the distribution function
Find the density f unction / ( jc ) , M [ jc ], D[ jc ] .
0 fo r jc < 0
1 fo r 0 < jc C 1,
0 fo r 1 < jc
25. A random variable x obeys the normal law with expectation 30 and
variance 100. Find the probability that the value of the random variable lies
in the interval (10, 50). Ans. 0.954.
26. A random variable obeys the normal distribution law with variance
a 2 = 0.16. Find the probability that a value of the random variable will differ
in absolute value from the mathematical expectation by less than0.3. Ans. 0.5468.
27. A random variable x is normally distributed with centre of dispersion
a = 0.3 and modulus of précision. h = 2. Find the probability of a hit in the
interval (0.5, 2.0). Ans. 0.262.
28. Gunfire is conducted at a strip of width 4 métrés. There is a syste-
matic error in aiming of 1 métré (undershooting). The probable error is 5 métrés.
Find the probability of hitting the strip in the case of normal dispersion.
Ans. 0.211.
29. Fire is conducted at a rectangle bounded by the straight lines x t = 10 m,
xa = 20m, y x= 15m, y2 = 35 m, in the direction of the straight line bisecting
the short side. The probable errors of normal distribution on the plane are
Ex = 5 m, £ J/= 1 0 m . Find the probability of a hit in a single shot. Ans. 0.25.
30. An error in the fabrication of an article with a given length 20 cm is
a normally distributed random variable; a = 0.2 cm. Détermine the probability
that the length of a manufactured article will differ from the given value by
less than 0.3 cm. Ans. 0.866.
31. Under the hypothèses of Example 30, détermine the fabrication error that
has probability 0.95 of not being exceeded. Ans. 0.392.
32. A random variable x is normally distributed with parameters M[*] = 5
and a = 2. Détermine the probability that the random variable will lie in the
interval (1, 10). Make a drawing. Ans. 0.971.
33. The length of an article made on an automatic machine is a normally
distributed random variable with parameters M [x |= 15, o = 0.2. Find the pro
bability of defective output if the tolérance is 15 ± 0.3. What accuracy of length
of the article can be guaranteed with probability 0.97? Make a drawing.
34. In measuring a certain quantity we get the following statistical sériés;
X 1 2 3 4
Relative 20 15 10 5
frequency
Détermine the statistical mean and the statistical variance. Ans. 2 and 1.
35. The results of measurements are tabulated as follows.
x 0.18 0.20 0.22 0.24 0.26 0.28
Relative 4 18 33 35 9 1
frequency
Find the statistical mean a and the statistical variance o2. Ans. 0.226, 0.004.
36. The probability of defective production is p = 0.02. Find the probability
that in a batch of 400 parts there will be from 7 to 10 defectives. Ans. 0.414.
37. The probability of a hit is P = y • What is the probability that in 250
shots the number of hits will lie between 100 and 150? Ans. 0.998.
38. The probability of defective production is p = 0.02. Find the probability
that among 1000 items there will be not more than 25 defectives. i4ns. 0.87.
CH A PT E.R 9
M A TRICES
9.1 LINEAR TRANSFORMATIONS. MATRIX NOTATION
C onsider tw o p la n es P and Q. G iv en in the P p lan e is a rectan-

gu lar S ystem of coord in ates x fix 2, in the Q p la n e, a S ystem y lOyi .
The P and Q p la n es can be brought to c o in c id e n c e and so can
the coord in ate S ystem s. L et us con sid er the fo llo w in g S ystem of
éq u a tio n s:
01 = * 1 1*1+ 01.*. \
y 2 ~ *21*1 + *22*2 )
By (1), to each point M (xt, x2) of the .Vjjr2-plane there corresponds
a point M (yt, y2) of the y ty2-plane.
We say that équations (1) constitute linear transformations of
the coordinates. These équations map the x^-plane onto the
plane (not necessarily onto the whole plane). Since the équations
(1) are linear, the mapping is termed a linear mapping.
Ui
0 xj 0 y(
Fig. 201
If in the x,.v2-plane we consider some domain D, then using (1)

we can define some set of points D of the ÿ^-plane (Fig. 201).
Note that one can also consider nonlinear mappings
y , = *22'
A rectangular array of these coefficients can be written hus:
*11*12 *11 a
or
*21 *22 ,*21 a
520 Ch. 9 Matrices
It is called a matrix of the mapping (1). The symbol || || or ( )

is the matrix symbol.
Matrices are also designated by a single letter, for instance
A or || A ||:
Û12
a..2
f l, ,
A déterminant made up of the éléments of a matrix, in the order

they are given in the matrix, is called the déterminant of the matrix
[we dénoté the déterminant by A (/!)]:
Example 1. The mapping

y l = x l cos a — x2 sin a
y2 = x t sin a + x2 cos a
is a rotation through the angle a . In this mapping, every point M with polar
coordinatcs (p, 0) is carricd into point M with polar coordinates (p, 0-}-a) if
the coordinate syfctems {x{Ox2) and
{y\0y2) are brought to coincidence (Fig.
202).
The matrix of this mapping is
cos a —sin a
A—
sin a cos a
y\ = k*i
y 2= x 2
represents a stretching along the x r axis

with stretching factor k (Fig. 203).

y l = kxl
yt = **2
is a k-fold stretching bot h in the direction of the x r axis and in the direction
of the Xg-axis (Fig. 204).
9.1 Linear Transformations. Matrix Notation 521
Example 4. The transformation

y i = — *i
y2 — x 2
is called a mirror reflection in the x2-axis (Fig. 205).
x 2 1yz *2\y 2
o ■ ï 0
Fig. 205 Fig. 206
The matrix of tins transformation is

ÿl = *1+^*2
02= *2
is termed a simple shear transformation along the xt -axis (Fig. 206). The matrix
of tliis transformation is
We can consider a linear transformation vvith an arbitrary number

of variables.
Thus, the transformation
( 1)
is a mapping of three-dimensional space (x1( x2, x3) into the three-

dimensional space (t/u y t, y3). The matrix of this transformation-is
° ll ° 1 2 °13
A — ü 3i 0,33 ^23 (5)

^31 ^32
31 m CL.3 3
32 w
lt is also possible to consider linear transformations with a non-

square matrix, which is a matrix where the number of rows differs
522 Ch. 9 Matrices
from the number of columns. The transformation

Hi = üj 2x 2 )
9t= auxl + attxt l (6)
9 ,= a ,iX l + anxt J
is a mapping of the x,x2-plane into a certain set of points in the
space (ÿ„ yt, y3).
Here, the matrix of the transformation is
a,,11 “a12
û 2i #22 (7)
# 5 1 # 5 2
It is also possible to consider matrices with arbitrary numbers

of rows and columns. Matrices find application outside of linear
transformations. For this reason, a matrix is an independent mathe-
matical concept similar to that of a déterminant. We now give
some définitions involving the matrix concept.
9.2 GENERAL DEFINITIONS INVOLVING MATRICES
D é fin itio n 1. A rectangular array of mn numbers consisting of m
rows and n columns,
an #12 * .. aln
A = ( i)
a»i ■** #/»/!
is termed a matrix. A matrix is also denoted briefly as
A = || au || (t = 1, 2, . . m\ i = 1, 2........ n) (2)
where a,;- are the entries (éléments) of the matrix.
If the number of rows in a matrix is equal to the number of
columns, m = n, then we hâve a square matrix:
#11 #12 *• • fli»
#21 #22 •
A = (3)
a„i a„2 • •' * #/f«
2. A déterminant consisting of the éléments of a square
D éfinition
matrix (in the order given in the matrix) is called the déterminant
of the matrix. We dénoté it by A (A):
#11 #12 * • * # 1 /1
A (A ) = #21 #22 * • * #2 n (4)

# » 1 # /î2 * '' * # ///*
9.2 General Définitions Involving Matrices 523
Note that a nonsquare matrix does not hâve a déterminant.

Définition 3. A matrix A * is called the transpose of A if the
columns of A are the rows of A * . .
Example. Let
a ll a l2
A= a 2 l a 22
a 9l a 32
The transposed matrix A* is then

a h a21 cts i
A* =
a l2 a 22 a 32
D é finition 4. A square matrix A is said to be symmetric about

the principal diagonal if a/;-= a;7. It»is clear from this that a
symmetric matrix coïncides with its transpose.
D é fin itio n 5. A square matrix whose éléments outside the prin
cipal diagonal are ail zéro is called a diagonal matrix. If the
éléments of a diagonal matrix on the principal diagonal are ail
unity, then the matrix is termed a unit matrix (identity matrix)
and is denoted by E :
1 0 .. . 0
0 1 .. . 0
O •
0 .. . 1
D é fin itio n 6. We can also consider matrices consisting of a single
column or a single row:
X-i
x2
X y = \ \ y i ÿ * • • • 0.11 (6)
Xm
The former is called a column matrix, the latter, a row matrix.

D éfinition 7. Two matrices A and B are considered equal if they
hâve the same number of rows and columns and if the correspond-
ing éléments are equal:
A = B (7)
or
IK /IM IM I (» = 1, 2, . . m; / = 1, 2.........n) (8)
if
a it = b a (9)
It is sometimes convenient to identify a column matrix and a
vector in a space with the corresponding number of dimensions,
524 Ch. 9 Matricés
where the éléments of the matrix are the projections of the vector
on the appropriate coordinate axes. Thus, we can write
*1
—Xl/ -j- x^j -J- x jt ( 10)
xt
A row matrix is also occasionally conveniently identified with
a vector.
9.3 INVERSE TRANSFORMATION
From équations (1) of Sec. 9.1,
Ui ~ #11*1 4 “ #12*2
y* ~ #21*1 4“ #22*2 i
\
(O
it follows that the mapping of x^-plane into y xy2-plane is one-to-
one, since each point of the x^-plane is associated with a unique
point of the y xy2-plane.
If the déterminant of the transformation matrix is nonzero,
#11 #12
A (A) = :0 or «11«22 «21«12 ^ ( 2)
«21 «22
then, as we know, the system of équations (1) has a unique solu

tion for xl and x2:
ÿ \ û 12 "il ÿl
yt °22 a2* y 2
or, expanded,
■y2
°2I
(3)
To each point M ( y l, y i) of the y ty2-plane there corresponds a

definite point M (xt, x2) of the x,x2-plane. In this case, the mapp
ing (1) is termed a one-to-one (nonsingular) mapping. The trans
formation of coordinates (y,, y2) into the coordinates (je,, x2) (3) is
called an inverse transformation. Here, the inverse mapping is
linear. Note that a linear nonsingular mapping is called an affine
mapping. The matrix of an inverse transformation is a matrix which
we dénoté by A ' 1:
#22 -- #12
A A
—#21 #n (4)
A A
9.3 Inverse Transformation 525
If the déterminant of the matrix A is equal to zéro,

^81^1» = 0 (5)
then the transformation (1) is said to be singular. In that case, it
is not a one-to-one transformation.
We will prove this. Let us consider two possible cases:
(1) If a,, = al2 = a21= a22= 0, then for arbitrary x, and x2 it will
be true that y l = 0, ÿ2= 0. In this case, any point (x„ x2) of the
XjXj-plane will pass into the origin of the y ^ - plane.
(2) Let at least one of the coefficients of the transformation be
different from zéro, say an =£0.
Multiplying the first équation of (1) by an , the second by au
and then subtracting, we get [with regard for équation (5)]
Û21 yi = anXl + ai**i
011 y2 = a81x 1 + a22x. ( 6)
atiÿi —an ÿt —0
Thus, given arbitrary xlt x2, we get (6) for the values yx and yt,
that is, the corresponding point of the x,x2,-plane faits ôn the straight
line (6) of the plane. It is clear that this mapping is not a
one-to-one mapping since each point of the straight line (6) of the
t/,ÿ2-plane is associated with a set of points of the XjX2-plane lying
on the straight line y l = allx1+ a12xI.
In neither case is the mapping one-to-one.
0i = 2*i+*«
y2 = x l — x2
is one-to-one since the déterminant A (A) of the transformation matrix A is
nonzero:
12 11
A <*) = h _ , I = —3
The inverse transformation is
*i = “3'0i+-3 y%
x* = T y i ~ T y*
The matrix of the inverse transformation îsr by formula (4),
1 1
3 3
A~l =
1 2
3 3
Example 2. The linear transformation
= * i + 2 x2
0i = 2*i + 4*a
526 C7i. 9 Matrices
in singular, since the déterminant of the transformation matrix

12
A (A) =
2 4
This transformation carries ail points of the x ix2-plane into the straight line
y* — 2 ^ = 0 of the y xy2-plane.
9.4 OPERATIONS ON MATRICES.

ADDITION OF MATRICES
Définition 1. The sum of two matrices ||af-y|| and ||6,y|| having

the same number of rows and the same number of columns is a
matrix ||ci; || whose element c;j is the sum -f biS of the correspond-
ing éléments of the matrices ||a(/|| and ||6,; ||; that is,
Il aij II + Il bij II = Il cij II (1)
if
a ,7 + b.V= c <7 ( i = 1, 2, . . m; / = 1, 2, . . n) (2)
Example 1.
an a 12 bu b\2 lû ll + ftll Q\2 “f* b\2
+
f l2 l û 22 b2i b22 | û 21 + ^21 û 22 + ^22
The différence of two matrices is defined in similar fashion.

The expedience of such a définition of the sum of two matrices
follows, in particular, from the conception of a vector as a column
matrix.
Multiplication of a matrix by a scalar (number). To multiply
a matrix by a scalar X, multiply each element of the matrix by
that scalar:
k IIan\\ = 11k*//1| (3)
If X is intégral, then formula (3) is obtained as a conséquence
of the rule for adding matrices.
Example 2.
aU a\2 II_ Xan ka12
a21 a22II Xa2\ ha2 2
Product of two matrices. Suppose we hâve a linear transforma
tion of the x^-plane into the yxy2-plane:
Ui = ci11xl -f- ü12x2 \
(4)
y 2 ~ ^21^1 4“ ^22^2 1
with the transformation matrix
9.4 Operations on Matrices. Addition of Matrices 527
Furthermore, suppose a linear transformation has been performed

of the y^-plane into the ztza-plane:
^ 1 = ûih “F ^
^2 =: ^ 2lUl “1“ ^22^2 J
with transformation matrix
K l *>12 I
B (7)
^21 ^22I
It is required to détermine the matrix of the transformation of
the x^-plane into the z^-plane. Substituting the expressions (4)
into (6) we get
zx= bn (allxl + al2x2) + b12 (a21x, + a22x2)
Z j — b 2 i (^11^*1 “ F 2 ^ 2 ) “F ^22 (^ 2 1 ^ 1 + ^22^2)
or
Zj = (bilall ~Fb12a21) x, -F (^nîî ~F^12^22) * 2 \
( 8)
z2= (b2lalt -F ^22^21) *1 + (K 1& 12 "F^ 2 2 ^ 2 2 ) % 2 J
The matrix of the resulting transformation is
^11®U "F^lAl ^lia!2 "F ^12û22 II
c= (9)
^21^11 ~F^22^21 ^21^12 + t>22a 22 ||
or, briefly,
C11 C12
C = ( 10)
C nn
The matrix (9) is called the product of the matrices (7) and (5)
and we write
b21b12 «u a,2 ^11^11 “1"bl2a2l ^liai2 4“^12^22
( 11)
bl2 b22 <*21^22 ^2iail 4“^22^21 ^21^12 4“ ^22a22
or, briefly,
B A = C ( 12)
Let us now state the rule for multiplying two matrices B and .4
if the first contains m rows and k columns, and the second, k rows
and n columns.
Schematically we hâve the équation
bu ^12 • • • bik all CLi2 . . . aij . . . avi Cl l ............. Cin
Û21 ^22 • • * ^2j • • • ^2n = . . •
bii bî2 • • * : • • • cu ■■; (13)
bmi bm2 • • • bmk akl ak2 • • • akj • • • akn Cml............. Cmn

528 Ch. 9 Matrices
Elément c,y of matrix C, which is the product of matrix B by

matrix A, is equal to the sum of the products of the éléments of
row i of matrix B by the corresponding éléments of column j of
matrix A\ that is,
k
c,7 = bnflu (i= 1, 2, . . . , m; / = 1, 2........ n)
Example 3. Suppose
10 0 0
B= , A=
, 0 0 10
then
10 0 01 0 0
( 1) ba =
0 0 10 0 0
0 0 1 01 0 0
(2) AB =
10 0 o| 10
In this example,
BA 7= AB
We therefore conclude that in matrix multiplication the commu
tative law does not hold true.
Example 4. Given the matrices
10 0 0 10
A= 0 2 1 . B= 2 0 1
3 0 0 l 0 1
Find AB and BA.
1 0 + 0-2 + 0-1 1 • 1 + 0-0 + 0-0 1-0 + 0 - 1 + 0-1 0 1 0
AB = 0-0 + 2 -2 + 1-1 0 1 + 2 -0 + 1-0 0-0 + 2-1 + 1-1 — 5 0 3
3-0 + 0-2 + 0 1 3 1 + 0 - 0 + 0-0 3 0 + 0 -1+ 0-1 0 3 0
0-1 + 1-0 + 0-3 0 -0 + 1-2 + 0-0 0 - 0 + 1 - 1 + 0 0 0 2 1
BA = 2 -1 + 0 -0 + 1 -3 2-0 + 0 - 2 + 1 0 2 0 + 0-1 + 1-0 = 5 0 0
1 1 + 0 - 0 + 1 3 1-0 + 0 2 + 1 - 0 1-0 + 0-1 + 1 0 4 0 0
Example 5. Find the product of the matrices
b\\ bl2 b13
a l l a 12 a l 3
b2i b22 b23
a 2 l a 22 a 23
^31 ^32 ^33
__ flll^ll + fl12^2l~f~al3^3J a11^12+ û12^22+ fl13^32 a\1^13+ °12^23+ û13^33II
fl21^11+ ^22^21~t~°23^31 û21^12+ û22^22+ û23^32 fl21^134" °22^23+ Û23^33 II
By direct vérification we see that the following relations for
matrices are valid (k is a scalar, A, B, C are matrices):
( k A ) B = A-(kB) (14)
(A ^ B) C = A C + B C (15)
C ( A + B ) = CA + CB (16)
A(BC) = (AB)C (17)
9.5 Transforming a Vector into Another Vector 529
From the rules for multiplying a square matrix A by a scalar k

and from the rule for taking out a common factor of the éléments
of the columns of the déterminant of a matrix of order n follows
A (kA) = kn A {A) (18)
Since multiplication of two square matrices A and B yields a
square matrix whose éléments are formed by the rule for multi-
plying déterminants, the following équation is clearly valid:
A (AB) — A (A) •A (B) (19)
Multiplication by the unit matrix. A square matrix whose élé
ments on the principal diagonal are ail equal to unity and else-
where are zéro (as mentioned above) is a unit matrix.
For instance,
is a unit matrix of order two.

By the rule for multiplication of matrices we get
û]2 1 0 an an
A E = «n
avi 0 1 a2l a.it
That is,
AE = A (21)
and also
EA = A (22)
It is easy to see that the product of a square matrix of any
order by the corresponding unit matrix is equal to the original
matrix, whîch means (21) and (22) hold true. In matrix multipli
cation, the unit matrix plays the rôle of unity, whence the namé
“unit matrix” (or identity matrix).
The unit matrix (2) is associated with the transformation
yi = *i
y >= * 2
This is the identity transformation. Conversely, to an identity
transformation there corresponds a unit (identity) matrix. In simi-
lar fashion we define the identity transformation of any number of
variables.
9.5 TRANSFORMING A VECTOR INTO ANOTHER VECTOR

BY MEANS OF A MATRIX
Suppose we hâve a vector

X = x J + x„J + x.Jt
34—2082
530 Ch. 9 Matrices
which we will write in the form of a column matrix:
(1)
We transform the projections of this vector with the aid of the

matrix
a i l û 1 2 û 13
^22 ^23 (2)

®31 ^32 ^33
to get
^1=01X^1 + ^ 2 + 0 1■o13x3
3^3 ^
y 2 = o 21Xj 4- + -at3x3
1 [• (3)
y3 OjjJCj-j“Ü33X3 jj j +"O
^33X
a3 J
This yields a new vector
y= yJ +y J + y 3^
which can be written as a column vector thus:
yi ^11*1+ ^12-^2+ ^13*3
Y =
y% ---- <*13*.
0 * 1 * 1 + « * * * * + (4)
y3 f l * l * l + fl32**+û.3*3
Using the matrix-multiplication rule, we can write down the

transformation operation as follows:
+ 1#12 û 13 *1 <*nXi + <*itXt + al9x9
a.n a*2 ^ 2 3 x 2• = &21X 1 + & 22X 2 ^ 2 3 X 3 (5)
a,01
31 ^ 3 2 ^ 3 3 X 3 a 3 l X l + ^32*^2 + Û 33 * 3
That is,
Y=AX (6)
Multiplication of a square matrix by a column matrix yields a
column matrix with the same number of rows.
Clearly, the transformation of a three-dimensional vector X into
the vector Y is another way of saying that three-dimensional space
is transformed into three-dimensional space.
Note that the System of équations (3) is obtained from the mat
rix équation (4) by equating the éléments of the left-hand matrix
and the right-hand matrix.
Equation (4) yields a transformation of the vector X into the
vector Y by the matrix A.
Ail the reasoning pertaining to a vector in three-dimensional
space can be earried over to a transformation of vectors in a space
of any number of dimensions.
9.6 Inverse Matrix 531
9.6 INVERSE MATRIX
Suppose we hâve a vector X. Let us transform it by means of

a square matrix A to get a vector Y:
Y —A X (1)
Let the déterminant of A be nonzero, A (i4 )# 0 . Then the in
verse transformation of the vector Y into the vector X exists.
This transformation is found by solving the System of équations (3),
Sec. 9.5, for xlt x2, x3. The matrix of an inverse transformation is
an inverse matrix of A. The inverse of A is denoted by A~x. We
can thus Write
X = A~XY (2)
Here, X is a column matrix, K is a column matrix, A X is a
column matrix, A~x is a square matrix. Substituting into the right
member of (2) the right member of (1) in place of Y, we get
X=A~'AX (3)
Thus, on the vector X we performed a succession of transformations
with the matrices A and A~l. That is, we carried out a transfor
mation by means of a matrix that is equal to a product of matri
ces (A~XA). The resuit is an identity transformation. Hence, the
matrix A~XA is an identity (unit) matrix:
A~lA = E (4)
Equation (3) takes the form
X = EX (5)
Theorem 1.If the matrix A~x is theinverse of the matrix A,
then A is the inverse of A~x, and the following équation holds true:
A XA = A A ~ X= E (6)
Proof. Transform both members of (3) by means of A:
AX=A(A~'A)X
Using the associative property for matrix multiplication, we can
write the last équation as
A X = ( A A ~ X) A X
From this we get
AA~ X= E (7)
which complétés the proof.
From équations (4) and (7) it follows that the matrices A and
A~x are inverses of each other. It also follows therefrom that
( A - X) - X= A (8 )
34 *
532 C /i. 9 Matrices
Indeed, équation (7) implies that
Comparing this équation with (4), we get équation (8).
9.7 MATRIX INVERSION
Suppose we hâve a nonsingular matrix

ûll
M l ûis ûl1 3
u 12 W
A sa, ( 1)
ai û 32 Q-à3
^11 a i2 a i3
A = A (i4) = û■*9i210*®
“ 22 “a ,2» ¥ = 0 (2)
® 81 û »8 ^ 3 »
We will prove that the inverse matrix A~l will be
4s»
A A A
4*2 4 s*
A~ l A A A (3)
4|S 4*3 4*3
A A A
where A(f is the cofactor of the element a,-.- of the déterminant

A = A(4).
We find the matrix C as the product of the matrices 4 4 -1:
4 11 4 ?l 4 31
ûu Ü12 û jj A A A 1 0 0
4]2 4 22 4 32
0 = 4 4 -* = Û21 ^22 ^22 0 1 0
A A A
f l 31 û 8 2 û 83 4 js 4?s 4 s3 0 0 1
A A A
Indeed, by the matrix-multiplication rule, the diagonal éléments
of matrix C are each equal to the sum of the products of the
éléments of a row of the déterminant A by the corresponding cofactors
divided by the déterminant A, that is, equal to unity. For example,
the element cu is defined as follows:
An •, A| A ,, au A u -\-attA 12~t~^13*^13
^11 = fll l I- û 18 A = 1
Each off-diagonal element is the sum of the products of the éléments

of a row by the cofactorS of another row divided by the determi-
9.7 Matrix Inversion 533
nant A; thus, for example, element c23 is

^31 Û j l î l _l û -d»l a2\Az\ ~hflg2^32 ~^~fl23^33 0 q
C93 — U22 £ T u23 £ Â- ' A_
This complétés the proof.
Note. The matrix
An A2l A9ï
A= A 12 -^22 ^32 (4)
A 1 3 A 2 3 A :j3
is the adjoint of i4. The inverse matrix i4-1 is expressed in tenus

of the adjoint A as follows:
l
A '1
A (A)
A (5)
The truth of this équation follows from (3).

Example. Given a matrix
12 0
A= 0 3 1
0 12
Find the inverse A - 1 and the adjoint matrix A.

Solution. We find the déterminant of A:
A (A) = 5
Then we find the cofactors
i4n = 5, î2~0, >4|3 = 0

= —4, /422 == 2, i423 = ““l
^31 = 2. ^32 = —i» ^33 = 3
Hence, by (3),
5 4 2
5 5 5
2 1
0
5 5
1 3
0
5 5
Using formula (4) we find the adjoint matrix
5 —4 2
A= 0 2 —1
0 —1 3
534 Ch. 9 Matrices
9.8 MATRIX NOTATION FOR SYSTEMS

OF LINEAR EQUATIONS AND SOLUTIONS OF SYSTEMS
OF LINEAR EQUATIONS
The discussion will be conducted for the case of three-dimensional

space. Suppose we hâve the following System of linear équations:
^ll'X'l 4" #12-^2 4“ ^13-^3 =
^21*1 + Û22^2 + ^23*3 — do ^ ( 1)
^31^1 4“ #32^2 -f- # 33X3 = d 3
We consider three matrices:
Il ûa.,
#n a.
12 w13
a 2i #22 û 2<1 (2)
a 3\ a 32 a 33
X,
Xe, (3)
(4)
Then, using the matrix-multiplication rule, we can write system (1)

in matrix form as follows:
«Il #12 #13 di
fl2l û22û23 • = d2 (5)
a3l G32 û33 *3 d3
Indeed, in (5), on the left we hâve a product of two matrices
which is equal to a column matrix whose éléments are defined by (5).
On the right we again hâve a column matrix. The two matrices
are equal if their éléments are equal. Equating the corresponding
éléments, we get the system of équations (1). The matrix équation (5)
is compactly written as
AX = D (6)
Example. Write the following system of équations in matrix form:
*i -f 2*0 =5
3*2 + *3 = 9
*2 - f 2*3 = 8
Solution. Write matrix A of the system, the solution matrix X and the
matrix D of constant terms:
12 0 xl 5
A = 0 3 1 , X = *2 , D = 9
0 12 X3 8
9.9 Solving Systems of Linear Equations by the M atrix Method 535
In matrix form, this System of linear équations can be written tlius:
120 SI
0 3 1 *2 = 9
0 12 *3 8
9.9 SOLVING SYSTEMS OF LINEAR EQUATIONS

BY THE MATRIX METHOD
Let the déterminant of a matrix A be A (A)^=0. Premultiply

the left and right members of (6), Sec. 9.8, by A ' 1, the inverse
of A, to get
A l A X = A~lD (1)
But
A 1A = E , EX = X
and so from (1) follows
X = A~lD (2)
This équation can, with regard for (5) of Sec. 9.7, be written as
X - I W AD (3)
or, expanded,
An A2l A31 dt
1
*AM 2 À 22 A 3 2 d2
y
X 2
A (A)
u (4)
x3 ^ 4 1 3 A 23 A 33 d3
Multiplying together the matrices on the right, we get
^-\A \\ H" ^2^21 "t“ ^3^31

_ __ 1__
X2 ^1^12 d2A22“h d9A32 (5)
“ A (A)
*3 ^1^13 “h ^2^23 ^3^33
Equating the éléments of the matrices on the left and on the right,
we get
« _ ^1^11 + ^2^21 + ^3^31
A
diA i2-\-d2A 22-\-dÂ22
*2— A (6)
_ _d{Ai3-{-d2A 23-\-d3As3
*« — a
536 Ch. 9 Matrices
The solution (6) can be written in terms of déterminants

di al2 flij ûn di 013
1
0‘>2023 02J 023
d3 032 033 031 d$ 033
»
a\\ a\2fl13 » 011 0J2 ÛJ3
a.n a22 023 a2l a22a23
Û31 032 033 a31a32a33
(7)
an
021 a22d2
v8- asi a32dn
011 012 0J3
02i 022 023
031 032 033 >
Example 1. Solve the System of équations
*i + 2*2 =5
3*2+ *3 = 9
*2 + 2*8= 8
by the matrix method.
Solution. Let us find the déterminant of the matrix of the System:
12 0
A(4) = 0 3 1 = 5
0 12
Let us détermine the inverse matrix by formula (3), Sec. 8.7:
2_
5
J_
5
j3
5
The matrix D will be
5
Z>= 9
8
We \vrite down the solution in matrix form, via formula (2), as lollows:
4 2 4 2
*1 I 5
5 5 l- 8 - T - 9 + T - 8
2 1
*2 = 0 9 = ° .5 + 4 . 9 - 1 . 8
T 5
î 3
*3 0 8 °-5 ~ f 9 + f 8
5 5
9.10 Orthogonal Mappings. Orthogonal Matrices 537
Equating tJie rows of tlie matrices on the left and on the right, \ve get
*, = 0 . 5 + | ~ 9 — - - 8 = 2
*a = 0 5 - - i - 9 + | - - 8 = . 3
Example 2. Solve the System of équations
xx-\-2x.î + xz —0
*2 +*3= 1
*i-|-3*2 + *3 - 2
by the matrix method.
Solution. Find the déterminant of the matrix of the System
1 2 1
A (A) = 2 1 1 = 1 9*Û
1 3 1
We find the inverse matrix

—2 1 1
A~l = —1 0 1
5 — 1 — 3
Then Write the solution of the System in matrix form:

*1 — 2 1 1 0 3
*2 — 1 0 1 1 — 2
*3 1 5 — 1 — 3 2 — 7
Equating the rows of the matrices on the right and on the left, we get
*i = 3, *a = 2, *a = — 7
0.10 ORTHOGONAL MAPPINGS.

ORTHOGONAL MATRICES
S u p p ose in th r ee -d im en sio n a l sp a c e w e h â v e tw o recta n g u la r
S y stem s of co o r d in a tes (x19 x2, x3) and ( x[, x2, x'3) h a v in g a com m on
o rig in O. L et a p o in t M h â v e co o rd in a tes (x1% x2% x3) and (xu x2t
x3) in th e first and secon d co o r d in a te S ystem s (th e o r ig in s n eed n o t
c o ïn c id e ).
Dénoté by elt e2, e3 the unit vectors on the coordinate axes in
the first coordinate system, by e[y e2, e3 the unit vectors in the
second coordinate system. The vectors elf e2y e3 are base vectors
in the system (jtlf x2, x3), the vectors e[9 ë2% e3 are base vectors
in the system (x[y x2f x3). ___
Then, in the first system the vector OM will be written thus:
OM = x1el + x2e2+ x3e3 ( 1)
538 Ch. 9 Matrices
in th e secon d S ystem thus:
OM = x\e\ + x2e2+ x'3e3 (2 )

We consider a transformation of the coordinates xlt x., x3 of
an arbitrary point M into the coordinates x[, x2, x3 of that point.
We can say that we are considering a transformation of the space
(xlt x2, xs) into the space (x[, x3, x3).
This transformation has the property that a line-segment of
length / is carried into a line-segment of the same length /.
A triangle is carried into an equal triangle and hence two vectors
emanating from a single point with an angle t|> between them go
into two vectors of the same length and with the same angle be
tween them.
A transformation having this property is called an orthogonal
transformation.
We can say that under an orthogonal transformation, the whole
space is displaced like a rigid solid, or we hâve a displacement
and a mirror reflection. We now détermine the matrix of this trans
formation.
We express the unit vectors e[, e 3, e3 in terms of the unit vec
tors e lt e t , e 3:
+ + \
C2 -■ “f“ -f- &32@3 I (3)
B3 I “h &33&2 “ h & 33C 3 J
Here,
OCjj = COS ( f i j , £ j.)* ^ ia = COS £ 2)» ^13 COS ( £ | , &3) |
a 2l = cos(e2, eï), a 22 = cos(e2, e 3), a 23 = cos(e2, e 3) l (4)
cx31 = cos(e3, e ; , ) , a 32 = cos(e3, e 2), a 33 = cos(e3, e 3) j
We write the nine direction cosines in the form ofa matrix:
«11 «21 «31
«12 «22 «32 (5)

“ 13 «23 «33
Using the relations (4), we can also write

e t ~ &U&X “1“ ® 12^ 2 “1“ «13^3
&3~ “h CC22B3“h 0^23é?3 (6)
C3= &3l@l -f- &32@3“h «33^3
It is clear that the matrix
“ il «12 «13
S* = «21 «22 «23 (7)

«31 «32 «33
9.10 Orthogonal Mappings. Orthogonal Matrices 539
is the transpose of S. Since e[t e ’2J e'3 are mutually perpendicular

unit vectors, their triple scalar product is equal to ± 1 . Thus,
21 ^31
(&1&2&3) «12 «22 «32 = ±1 (8)
3 ^ ‘23 0&33
Similarly
a 11
tl ^12
a,, a 13
(e&es) = A (S*) = 21 « 2 2 =± 1 (9)
«31 «32 «33
We compute the product of the matrices

«11 «21 «31 « I l “ 12 « 1 3 1 0 0
SS* = «12 «22 «32

. «21 «22 «23
— 0 1 0 (10)
«13 «23 «33 «31 «32 «33 0 0 1
Indeed, if we dénoté by ci} the éléments of the matrix of the

product, then
Cii = a îi + a n + a !i= 1 )
c32= a,2n + al3+ alt = 1 \ (H)
C33 = a w + a l3 + a l 3 = 1 J
c12= a u a 12 + a 21a 22 + a 31a S2 = (e[e3) = 0
Similarly
ci. = e'le', = 0 for i ^ j ( i = l , 2, 3; /'= 1 ,2 ,3 ) (12)
Thus
SS* = E .(13)
The transpose S* coïncides with the inverse S -1:
5* = S -1 (14)
A matrix satisfying the conditions (13) or (14) — that is, the in
verse of its transpose— is called an orthogonal matrix. Let us now
find the formulas for transforming the coordinates (xlt x„, xs) into the
coordinates (xJ, x2, x3), and vice versa. By virtue of formulas (3)
and (6 ), the right members of ( 1) and (2 ) can be expressed either
in terms of the basis (e,, e2, e3) or in terms of the basis {élt e3,
e3). Hence, we can write the équation
X2£ 2 “f- X3C3 = Xi&i -f- X2e 2 -f- X363 (18)
Multiplying in succession ail terms of (15) by the vector e[, by
the vector e2, by the vector e'3 and noting that
e’fi'i = 0 for i j
e'ie) — 1 for i = / (16)
«(«/ = <*//
540 Ch. 9 Matrices
we get
*î = « n * i+ '“ 21*2 + « 8.*s ]
X* = a 1tX1+ aitx. -f a 33x3 l (17)
^ = « 1 3 * 1 + “ 2 3 * 2 + «33*3 J
Multiplying the terms of (15) in succession by ev et, e„ we get

*i = “ n*l + “ i2*; + “ i3*3 \
*2 = “ 21 * I + «22*2 + “ t3*i > (1 8 )
x3= a3lx[ + ct32x'2+ a 33x'3 J
Thus, S is the matrix of orthogonal transformations (17) and S*
is the matrix of the inverse transformation (18).
It is thus proved that in a Cartesian coordinate System, an
orthogonal transformation is associated with an orthogonal matrix.
It can be proved that if the matrices of direct and inverse trans
formations (17) and (18) satisfy the relation (13) or (14), that is,
are orthogonal, then the transformation is orthogonal as well.
If we introduce the column matrices
X'x *.
X ' = x't , x= *2
x3
*3
then the Systems (17) and (18) can be written thus:
X' = SX (20)
X ^ S - 'X ’ (2 1 )
If we introduce the transposed matrices of (19),
X'* = || x[ x ’3 x ’3H, X t = \\x1 x3 x31| (22)
then we can write
X ,, = X*S-i. X* = X'*S (23)
9.11 THE EIGENVECTOR OF A L1NEAR TRANSFORMATION
Définition 1. Suppose we hâve a vector X:
Xi
X= (1)
w here
X* + Xj + Xj 0
If after transforming the vector X by means of matrix .4
[see (2), Sec. 9.5] we get a vector Y
Y = AX (2)
9.11 The Eigenvector of a Linear Transformation 541
parallel to X,
Y=XX (3)
where X is a scalar, then the vector X is termed the eigenvector
of matrix A or the eigenvector of the given linear transformation;
the scalar X is termed the eigenvalue.
Let us find the eigenvector
X= X2
*3
of the given linear transformation or of the given matrix A. For

X to be an eigenvector of matrix A, it is necessary that (2) and
(ô) hold true. Equating the right members of these équations,
we get
A X = XX (4)
or
A X —XEX
that is,
(A — XE)X = 0 (5)
From this équation it follows that the vector X is defined to
within a constant.
Equation (4) can clearly be expanded to
0\ " f ûiax2 4" •Xx3
an xt + aMxt + at3x 3 = Xx2 . (6)
« 3 1 ^ 1 XX3 ^
* 1 ” ^ 3 2 * ^ 2 ” 1” « 3 3 * ^ 3
and (5) to
(au —X) xy + a13x3+ a, 3x, = 0
Aji-Ci + (as2— X) x2 + al3x3= 0 (7)
«31*^1 4 “ « 3 2 ^ 2 4 “ (« 3 3 ^ 3 ^
We obtain a System of homogeneous linear équations for deter-

mining the coordinates xx, xt, x3 of the vector X. For System (7)
to hâve nonzero solutions, it is necessary and sufficient for the
déterminant of the system to be equal to zéro:
— X a i2 «13
a u ü 22 ^ a 23 (8)
«31 &32 û 33 *
or
A ( 4 —XE) = 0 (9)
This is a cubic équation in X. It is called the characteristic equa-
542 Ch. 9 Matrices
tion of the matrix A. From this équation we can find the eigen-
values k.
Let us consider a case where ail roots of the characteristic équa
tion are real and distinct. We dénoté them by klt k2, k3.
To each eigenvalue k there corresponds an eigenvector whose
coordinates are determined from the System (7) for an appropriate
value of k. Let us dénoté the eigenvectors by xlt t 2, t 3. It can
be shown that these vectors are linearly independent, that is to
say, none can be expressed in terms of the others. Hence, any
vector can be expressed in terms of the vectors t lf t 2, t 3, that is,
they can be taken for base vectors.
We note without proof that ail roots of the* characteristic équa
tion of a symmetric matrix are real.
Example 1. Find the eigenvectors and the corresponding eigenvalues of the
matrix
1 1
8 3
Solution. Form the characteristic équation and find the eigenvalues:
1—k 1
= 0 , .e., X2—4X — 5 = 0, Aa = — -1, k2 = 5
8 3—k
We find the eigenvector corresponding to the eigenvalue ^ = —1 from the
appropriate System of équations (7):
(1—X2) *i+ *2 —0 or 2xi -f- —0
8î (3 — Àj) X 2 —
—0 8*i 4x2 ~ 0
Solving this System, we get x l = m, x2 = —2m, where m is an arbitrary num-
ber. The eigenvector is
x1= mi — 2mj
For the eigenvalue X2 = 5, we write the System of équations
—4xi + *2 = 0
Sxi —2*2 = 0
The eigenvector is
x2 = mi + 4m j
Example 2. Find the eigenvalues and the eigenvectors of the matrix
7 —2 0
—2 6 —2
0*—2 5
Solution. We write the characteristic équation
7—k —2 0
—2 6 -A, —2 = 0, i.e., -A ,3+18A,2-9 9 A ,+ 162 = 0
0 —2 5 —Â
The roots of this équation are: k x = 3t À2 = 6, Pi3 = 9.
9.12 The M atrix of a Linear Transformation 543
For X,j = 3 the eigenvector is determined from the system of équations

4x1— 2xi =0
—2*i + 3*j—2*8 =0
- 2*8 + 2*, = 0
Setting *j —m, we get * , = 2 m, *s = 2 m. The eigenvector
Tj = mi + 2 /n /+ 2mA
Similarly we find
t 2 = m/+ y m]—mk
x3= — m i+ m j— ^ m k
9.12 THE MATRIX OF A LINEAR TRANSFORMATION

UNDER WHICH THE BASE VECTORS ARE EIGENVECTORS
We now define the matrix of a linear transformation when the

basis consists of eigenvectors r lt t,. The following relations
must hold under this transformation:
»î = V i '
T, ^ 2^2 * ( 1)
T3 = ^ 3^3 ^
where t ’, tJ, tJ are the images of the vectors xlt t 2, T,.
Let the transformation matrix be
ail &12&13
A ' = Û21 #22 #23
/ » t (2)
#31 a32a33
Let us détermine the éléments of this matrix. Relative to the
basis t x, t2, t„ we can write
1
* 1 = l- T i + O-Tg + O -T, = 0
0
Since the vector t x goes into the vector t j = X1t l after a trans
formation by the matrix A':
Tî —^jTx + 0 •T2 + 0 •T,
we can write
tj = = A ' t1
Hence
#1 1 #12 #13 1
*1
0 — #21 #22 #23 0

(3)
0 # 3 1 #32 #33 0
544 Ch. 9 Matrices
or, a s a S y s te m o f é q u a t io n s
Â .j = f l j i • 1 -j” O u * 0 - f - f lj 3 • 0 I
0 == a» • 1 + •0 -f- ^ 2.1 •0 > (4)

0 = fl8i • 1 - ( - û 8 j • 0 -J - f l j j • 0 J
From this System we find
O u = A.1 û 2l = 0 , flji = 0
On the basis of the relations

T, = Ta = XjTj
we find in similar fashion
Su —0 , U22 — üjs —0
ûjS—0 , #23 *0 » O33 A.j
Thus, the matrix of the transformation is of the form
*2 0 0
A’ 0 ^ 0 (5)
0 0 X,
The linear transformation is
Ui=
ifa — hfXt (6)
y» ~ ^ 3-^3
If = A* = As = A\ then the linear transformation assumes the form
y \ = >*x[
y\ —X*x\
y3= x*x3
This is what is called a similarity transformation with a coeffi
cient A*. Under this transformation, every vector of the space is
an eigenvector with eigenvalue A*.
9.13 TRANSFORMING THE MATRIX OF A LINEAR TRANSFORMATION

WHEN CHANGING THE BASIS
Let X be an arbitrary vector

Xj
X = *3 ^ 2î "1“ X2€3”1“ X3C2 O)
X»
given in the basis (elt e2, e3). The vector X is transformed by
9.13 Transforming the M atrix of a Linear Transformation 545
the matrix A into the vector Y:

0i
Y — yt = 01 * ,+ y te » + y ^ . (2)
y3
Y —A X (3)
In the space under considération, we introduce a new basis
(e[, e't, e3) connected with the old basis by the change-of-basis
formulas
*1 ^11*1 "b ^21*2 “b ^31*3 |
*2 = ^ 1 2 * 1 ~ b ^ 2 2 * 2 ”b ^ 3 2 * 3 j ^
*3 = ^ 1 3 * 1 “ 1“ ^ 2 3 * 2 “b ^ 3 3 * 3 /
Let the vector X be written as follows in the new basis:
X' —X[e'i + *2*2 + *3*3 = *2 (5)
We can Write the équation

* , * , “b * 2 * 2 + *3*3 * 1 * 1 b ' * 2 *2 “ b * 3 * 3 (b )
where the expressions (4) are substituted into the right-hand side.
Equating the coefficients of the vectors elt e3, e3 on the right and
left, we get the équations
x 1= bllx[ + bl3xî + biix3
x<i —b3lx3+ b33x3-b b33x3 (7)
x3 ^ b 3lx[+ b3tx3-\-b33x3
or, briefly,
X=BX' ( 8)
where
bn K 2 bi3
b3i b» bu (9)
bn ^32 b33
This is a nonsingular matrix that has an inverse, B -1, since
System (7) has a definite solution for x\, x’t, x3. If we write the
vector Y in the new basis
Y '= y \e \+ y 3ë3+ y 3ë3
then quite obviously the équation
Y = BY' (10)
35—2082
546 Ch. 9 Matrices
will hold. Substituting (8 ) and (10) into (3), we get

BY' = ABX' (11)
Multiplying both mémbers of the équation by B~l, we obtain
Y ’ = B -l ABX' (12)
Hence, the transformation matrix A' in the new basis will be
A ’ = B~lA B (13)
Example. Suppose, using the matrix A,
1 10
10 1
0 1 1
we transform a vector in the basis (eL, e2t e3). Détermine the transformation
matrix A' in the basis (e[, e2, e3) if
*j = *i + 2i?2 +é ?3
' *2 = 2*J + *2 +
*3 = *1 + *2 + *3
Solution. Here, matrix B is [see formulas (4) and (9)]

12 1
B = 2 11
13 1
We find the inverse matrix [A(2?) = l]
II - 2 1 1
-1 0 1
and then we find

I 5 - -1 - -3
—1 - 1 2
B -M = -1 0 11
4 2 —4
Finally, by formula (13), we get
-1 3 01
A f — B - 1 AB = 0 10
4 — 2 2\
Let us nôw prove the following theorem.

Theorem 1 . The characteristic polynomial [the left member of
(8 ), Sec. 9.11] remains unchanged for any choice of basis under
a given linear transformation.
Proof. We write two matrix équations:
A ' = B~lA B
E —B~lEB
9.14 Quadratic Forms and Their Transformation 547
where A and A ' are matrices corresponding to different bases for

one and the same linear transformation, B is the change-of-basis
matrix from new coordinates to old coordinates, and E is the unit
(identity) matrix. .
On the basis of the last two équations we get
A ' — % E = B 1(A — kE) B
Passing from matrices to déterminants and using the rule for

multiplication of matrices and déterminants, we get
A (A ' — k E ) = A ( B -* ( A — k E ) B ) = A ( B ~ 1) - A ( A — k E ) A ( B )
But
A ( B - 1) A ( B ) = A ( B - * B ) = A ( E ) — l
Hence
A { A '— hE) = A ( A — kE)
On the left and on the right we hâve the characteristic polynomials of

the transformation matrices, which proves the theorem.
9.14 QUADRATIC FORMS AND THEIR TRANSFORMATION
Définition 1. A quadratic form in several variables is a homo-

geneous polynomial of degree two in these variables.
A quadratic form in three variables xlt x2, x„ is of the form
F —ûj -f “T "1” 2a^3x^x3 2a^3x2x3 (1)
where a(/ are specified numbers and the coefficients 2 are takenso
as to simplify subséquent formulas.
Equation (1) can be written thus:
F = x i (fljjXj flj -f- a13x3)
■f-^2 (^21-^1 "Ha3ix3 ^ 23^3) (2 )
+ Xs («31*1 + «32*2 + «33*3)
where «/; (t = 1, 2, 3; / = 1, 2, 3) are given numbers, and
« U = «2J> «13 = « 3 l> «23 = «32 (3)
The matrix
« I l «12 «13
A = «21 «22 «23 (4 )

«31 «32 «33
is called the matrix of the quadratic form ( 1). The given matrix
is symmetric.
We will regard (xlt x2, x3) as the coordinates of a point of space
or the coordinates of a vector in the orthogonal basis (elt e 2, e„),
where e t , e 2, e 3 are unit vectors.
35
548 Ch. 9 Matricés
Let us consider a linear transformation in the basis (elt e2, e3):

x[ = au x1+ a l2xï + alix3
X? ” UjjXj -j~ CL22 X 2 “t“ @2 3 X 3 (5)
x3 ^ ailx1+ a3iX2 + a33x3 ,
The matrix of this transformation coincides with the matrix of the

quadratic form.
Let us now define two vectors
*1
X = x.. (6)
4
X' A'j (7)
x'3
We write transformation (5) as
X' = AX ( 8)
Then the quadratic form (2 ) ean be represented as a scalar pro-

duct of these vectors
F= X A X (9)
Let e[, e'2, e3 be orthogonal eigenvectors of the transformation (8 )
corresponding to the eigenvalues Xlt A2, A3. It may be proved that
if the matrix is symmetric, then there exists an orthogonal basis
composed of the eigenvectors of matrix A. Let us carry out the
transformation (8 ) ’in the basis (e[, e\, e3). Then the transformation
matrix in this basis will be diagonal (see Sec. 9.12)-
0 0
0 %2 0 ( 10)
0 0 A,,
It may be shown that by applying this transformation to the

quadratic form ( 1), we can reduce the latter to the form
F —AxXj -)- %2X\ •+■AjXg (II)
The directions of the eigenvectors e[, e\, e3 are called the prin
cipal directions of the quadratic form.
9.15 The Rank of a M atrix 549
9.15 THE RANK OF A MATRIX.

THE EXISTENCE OF SOLUTIONS
OF A SYSTEM OF LINEAR.EQUATIONS
Définition 1. The minor of a given matrix A is the déterminant
composed of éléments of the matrix left after striking out certain
rows and columns.
Example 1. Suppose we hâve a matrix
fll l f l 12 f l 13 a u
a 2\ Û 2 Î f l23 ^24
Û 31 f l32 û 33 û 34
Third-order minors of this matrix are obtained by striking out one column and
replacing the matrix symbol || || by the sign | | of the déterminant. There are
four. Second-order minors are obtained by striking out iwo columns and one
row. There are 18. First-order minors number 12.
Définition 2 . The rank of a matrix A is the highest order of

a nonzero minor of A.
Example 2. It is easy to verifv that the rank of the matrix
1 —1 0
2 C 1
1 1 1
is equal to 2.
Example 3. The rank of the matrix
12 3
3 6 9
is 1.
If a matrix A is square and of order n, the rank k satisfies the
relation k ^ .n . As pointed out above, if k = n, the matrix is
nonsingular, if k < n, the matrix is singular.
For instance, the matrix
0 10
0 0 1
l 10
is nonsingular since A ( 4 ) = l 0; the matrix in Example 2 is singular, since
there n = 3 and k — 2.
The concept of the rank of a matrix is widely used in the
theory of Systems of linear équations. The following theorem is
valid.
Theorem 1. Given a systern of linear équations:
«11*1 «12*2 «13*3 ~ |
«21*1 + «22*2 + «23*3 = &2 f O)

«31*1 « 3 2 * 2 ~f~ « 3 3 * 3 ~ ^3 /
550 Ch. 9 Matrices
We in trod u ce th e m a tr ix o! the System :
° 1 1 a i t a is
A= Û J I ü 22 Û 2 3 ( 2)
ûjt ûjj Ûjs
and the augmented matrix
®11 ®1* ^13 ^1
5= ûji ü22 ûs» bt (3)
atl a32 a33 ^3
The system (1) has solutions if the rank of the matrix A is
equal to the rank of the matrix B. The system does not hâve any
solutions if the rank of A is less than the rank of B. If the rank
of A and the rank of B is 3, then the system has a unique solution.
If the rank of the matrices A and B is equal to 2, then the
system has an infinitude of solutions; here, two unknowns are expres-
sed in terms of the third, which has an arbitrary value.
I f the rank of the matrices A and B is equal to 1, then the
system has an infinitude of solutions and two unknowns hâve
arbitrary values, the third being expressed in terms of these two.
The validity o! this theorem is readily established on the basis
of an analysis (familiar in algebra) of the solutions of a system
of équations. This theorem holds true for a system of any number
of équations.
9.16 DIFFERENTIATION AND INTEGRATION OF MATRICES

Suppose we hâve a matrix ||a/y(/)|| where the entries (éléments)
üif(t) of the matrix are functions of a certain argument t:
ail ( 0 a\2 ( 0 < * ln
Il fl/ / (011 =
a 21(0 a22 (0 a2n (t)
(l)
ami ( 0 î»2 ( 0 • • • Q/nn ( 0

We can Write this m ore c o m p a c tly as
Il fl (011 = 11fl// (011 ( < = 1 . 2 ........... « ; / = ! . 2 ..........n) (2 )
Let the éléments of the matrix hâve dérivatives
«tou(0 àamn(t)
dt * — dt
Définition 1 . The dérivative of a matrix ||a (/)|| is a matrix, we
dénoté it by -gj- || a (t) ||, whose entries are dérivatives of the
9.16 Différentiation and Intégration of Matrices 551
éléments of the matrix ||a(f)||; that is,

dan dalt datn
dt dt ■V dt
don da22 da.in
dt dt * • ’ dt
3 7 1 1 ° ( 011 = (3)
dam\ damt damn

dt dt ’ • ’ dt
Note that this définition of the dérivative of a matrix cornes
quite naturally if to the operations of subtraction of matrices and
multiplication by a scalar introduced in Sec. 9.4 we adjoin the
operation of passage to a limit:
j_
lim a ij( t + U ) \ \ - \ \ a u (t)\\\
àt o
aiJ(t + to)-aij(t) aij(i + to) — dij(t)
-= lim lim
At - 0 to to At - 0
We can write équation (3) more compactly in the following sym-
bolic form:
i\\a ( t) \\= é a if ( 0 (4)
or || a (0 1| = (5)
In place of the différentiation symbol ^ the symbol D is often

used, and then (5) can be written thus:
D ||a || = ||£>û|| (6 )
D é fin itio n 2. The intégral of the matrix ||a (f)|| is a matrix,
which we dénoté as
J || a (z) || dz
10
whose éléments are equal to the intégrais of the éléments of the
given matrix:
t t
an (z)dz ...
Sain (z)dz
> t.
t t
a2l(z)dz ... Sa»n(*)dz
S
U
Ila (z) \\dz = > t. (7)
f t
ûml(Z)dz ... J&mn(2 )dz
> U
552 Ch. 9 Matrices
More compacty.
(8)
or
J || a (z) || dz (9)
1o
The symbol $ ( )dz is sometimes denoted by a single Iet-

u
ter, say S, and then we can write équation (9), like we did (6 ),
thus:
S ||û || = ||5û|| ( 10)
9.17 MATRIX NOTATION FOR SYSTEMS OF DIFFERENTIAL EQUATIONS

AND SOLUTIONS OF SYSTEMS OF DIFFERENTIAL EQUATIONS WITH
CONSTANT COEFFICIENTS
We consider a System of n linear differential équations with n

unknown functions x1(/), xt (/), . . . , x„ (/):
dx '
-jjj- = ÛhJCj + £12X2 + • • • + ainxn
^ - = ailx1+ aîtxî + . . . + a tnxn
d\ ...............................................^ ( 1)
= ®m xl "t" ®n2 X 2 “h • • • 4 " a nnx n
The coefficients atJ- are constants. We introduce the notation

(0
ii*ii- x: m ( 2)
xn V )
This is the solution matrix or the vector solution of System (1).

We then define the matrix of dérivatives of the solutions:
dXy
dt
dx2
dx Ht
dt (3 )
dx„
dt
9.17 Matrix Notation for Systems of Differential Equations 553
Let us Write down the matrix of coefficients of the System of di

fïerential équations:
a n «12 * • • a i n 1
«21 a 22 . . * ^2n
Il a II = 11^ ,•11 =
• a m Û/X2 * * ■ a nn
Using the rule for matrix multiplication (see Sec. 9.4), we can
write the System of difïerential équations (1) in matrix form:
dxi
dt a 12 . . • a in
«U
dx 2
c i22 . . • ^2 n
©2
~dt •
(5)
dxn «ni ««2 •• ■ a nn

*,«
dt
or, more compactly on the basis of the rule for differentiating
matrices
£ ||* ( 0 ! I = IM M M I (6 )
This équation can also be written as
W = ax (7)
where x is also called the vector solution; a is short for the
matrix ||a (/ ||.
Suppose we hâve
«i
tt2
||a || = ce = ( 8)
«„
where a,- are certain scalars.
The set of solutions of a System of difïerential équations will
be sought in the form [see formula (2), Sec. 1.30]
| x 1| = e*' | a (9)
or
X = ekta ( 10)
Substituting ( 10 ) into (7) [or (9) into (6 )], via the rule for multi
plication of a matrix by a scalar and the rule for differentiating
matrices, we get
kekta — aekla (H)
554 Ch. 9 Matrices
whence we obtain
ka = art
or
a a — ka = 0 ( 12 )
Recall that in the last équation, a is the matrix (4), k is a scalar,
and a is the column matrix (8 ). The matrix in the left member
of ( 12 ) can be written as
(a — kE) a = 0 (13)
where E is an identity (unit) matrix of order n. In expanded form,
équation (13) can be rewritten thus:
au — tl cl12 •• <*ln “i
flji ait— k . • • n a2
(14)
a . . a nn— k a n
Equation (12) shows that the vector a can be transformed by the

matrix a into a parai le 1 vector ka. Hence, the vector a is an
eigenvector of the matrix a corresponding to the eigenvalue k
(see Sec. 9.11).
In scalar form, équation (12) is written as a System of algebraic
équations [see System (3), Sec. 1.30]. The scalar k must be deter-
mined from équation (5) of Sec. 1.30, which in matrix form can
be written
A (a — kE) = 0 (15)
That is, the déterminant
an —k au • .. aln
d2\ Û22 ^ *•• atn
=--0 (16)
a„i ani • • • # n n ^
must be equal to zéro.
Let ail the roots of équation (16),
k^t k%
2t >*• , kn
be distinct. For each value kt of the System (13) we define a ma
trix of the values of a:
a n(h
9.17 M atrix Notation for Systems oj Differential Equations 555
(one of these values is arbitrary). Consequently, in matrix form,

the solution of System ( 1) can be written.thus:
a',” a*,2’ ... a}"*
C,e*>'
a"’ ai® ... a'n> (17)
a«‘>a<2> ... a<">C„eknl
where C,- are arbitrary constants, or, briefly,
11*Il = 1a I I Ce** Il (18)
In scalar form, the solutions are given by formulas (6 ), Sec. 1.30.
Example 1. Write down in matrix form the System and the solution of the
System of linear differential équations
dxj_= 2x1+ 2xa
dt
Solution. We write the matrix of the System

2 2
13
In matrix form, the System of équations is written as [see équation (5)]
dx i
dt 2 21
dx2 13
dt
We now form the characteristic équation (15) and find its roots:
2—k 2
= 0, i.e.t Æ2—5Æ+ 4 = 0
1 3 —k
hence
^ = 1, k2 = 4
We set up System (14) to détermine the values a?*, for the root ^*= 1:
(2 -l)a { x,+ 2ail) = 0
a îl>+ (3 — 1) a*11= 0
Setting we get 0 2 *= — i-.
In similar fashion we find a[2> and ai*’ which correspond to the root *. = 4.
We obtain
aj2) = 1, «i2, = l
Now we can write the solution of the System in matrix form [formula (17)J
1 1
I
* C*e«|
U'
556 Ch. 9 Matrices
or tu the usual form:

x ^ C ^ + C ^t
Xis= —ÿ c ie '+ c ,e 4<
Example 2. Write in matrix form the svstem and the solution of the System
of differential équations
dxi =*i
dt
dt +
^ - = x x-{-Xi + 3xa
Solution. We Write down the matrix of the System:
10 0
A=12 0
1 13
Thus, in matrix form, the System of équations isw rittenas [see équation (5)J:
dx,
1 0 1
dt
dx2
= 1 2 0 *2
dt
dx3
1 1 3 *3
dt
Let us form the characteristic équation (16) and find its roots:
1- Jf e 0 0
1 2— k 0 = 1 , that is, (l — k) (2 —£) (3 —fc) = 0

1 1 3 — A:
conséquent ly,
= 1, ^2 = 2, ^3 = 3
We détermine a î1), o^1*, a ^ , which correspond to the root k x— \ t fromthe
Systemof équations (14):
aî1) -+- o4l) = 0
a<i)+ a ^)+ 2a ^ = 0
to ftnd
a i1) = 1, a il) = - l f a ? ) = 0
We détermine a i2), a j 2*, as2), which correspond to the root k8 = 2t from the
System
— a{2) =0
aj2> =0
a(«+ a i*)+ a J*)= 0
to find
aj*’ = 0, o ? » = l . a?» = - i
9.18 M atrix Notation for a Linear Equation of Order n 657
We détermine ai*0, a 23\ o49\ whieh correspond to the root ks = $, from the System
- 2 a ia) =0 #
ai3,- a ? ) = 0
cxi3, + a <
28)= 0
to find
a<3) = 0 , a23) = 0 , a<3) = l
Now let us \vrite down in matrix form the solution of the System [see formula (17)J:
*i 1 0 0 Cte*
x2 = — 1 1 0 Cte**
*3 0 — 1 1
or in ordinary form
j = — C,e‘ + C ,e st
*, = - C te*t+Cse»t
9.18 MATRIX NOTATION FOR A LINEAR EQUATION

OF ORDER n
Suppose we hâve an nth order linear difîerential équation with

constant coefficients:
dnx dn~ 1x d”~ 2x
d tn dta~l ' ^n—1 d tn~ 2 .. . + 0 ^ ( 1)
We will see later on that this way of numbering the coefficients

is convenient. Set x = xx and
dxL
dt = **
dx2
~dt
( 2)
fon- . =X„
dt
^ ûjXj -f-. . . +a„x„
Let us Write down the matrix of coefficients of the System:

0 1 0 0 ... 0
0 0 1 0 ... 0
(3)
0 0 0 0 ... 1
al al a* . . . a„
558 Ch. 9 Matrices
Tlien, like formula (5) of Sec. 9.17, the System (2) can be written
as follows:
dxi
0 1 0 . , ,. 0
~dt
dxi
0 0 1 . , ,. 0
dt
dx„-1
0 0 0.. . 1
dt
dxn
dt a » a a • • • an
or, briefly.
a-Il-<11=Il “•INI •«il (5)
The subséquent solution is carried through as in Sec. 9.17, since
the matrix équation (5) is a particular case of équation (6 ), Sec. 9.17.
Example. Write the équation
d 2x dx
di*= p W +qx
in matrix form.
Solution. Put x = x x . Then
d x x
d t ~ ~ * 2
t + Q x i
The System o( équations looks like this in matrix forai:

d x x
0 1
d t
d x 2
d t
<7 P X 2
9.1» SOLVING A SYSTEM OP LINEAR DIFFERENTIAL EQUATIONS WITH

VARIABLE COEFFICIENTS BY THE METHOD OF SUCCESSIVE
APPROXIMATIONS USING MATRIX NOTATION
Let it be required to find the solution of the System of linear
differential équations with variable coefficients
^jf- = a 11( t ) x l + a u ( t ) x l + . . . + a in (0 x tt
~ d f = ^*i ( 0 xi H- (0 x î 4" • • • + a t„ ( t)x n

0 )
TT “ (*) x i + ( 0 *» + • • • + a nn(t)xn
9.19 Solving a System of Linear Differential Equations 559
that satisfy the initial conditions
Xl —•tfjot —^ 2o> • • ' > Xrm f®** ^ (2 )
If, b e sid es the m a tr ix of co e ffic ie n ts o f th e S y stem and the m a

tr ix of s o lu tio n , w e in tro d u ce th e m a tr ix of in itia l c o n d itio n s
(3)
then the System of équations ( 1) with initial conditions (2 ) can be

written thus:
e II*IMI«(OIHMI (4)
for the initial conditions
IM IH K II for t = tu (5)
Here, ||a ( 0 || fs again the matrix of coefficients of the System.
We will solve the problem by the method of successive appro
ximations.
To get a better grasp of the material that follows let us apply
the method of successive approximations first to a single linear
équation of the first order (see Sec. 4.26).
It is required to find the solution of the single équation
^ = a (t)x (6 )

*= for t = t0(7 )
We will assume that a (t) is a continuous function. As was point-

ed out in Sec. 4.26, the solution of the differential équation (6 )
for initial conditions (7) reduces to the solution of the intégral
équation
t
x = x„+ â (z)x(z)d z (*)
u
560 Ch. 9 Matrices
We will solve this équation by the method of successive appro

ximations:
t
X1 = x0 + S o (z) x„ dz
t,
t
(z)*i (z)dz
10 (9)
t
xm= xo + l a (z)xm -i(z)dz

U
/
To save space we introduce the operator S (the intégration ope-
rator):
t
S()=$()dz (10)
U
Using this operator S, we can write the équations (9) as follows:

—jc#-|- S (âXy)
xt = xo+ S (ûjcJ = x0+ 5 (a (x0+ 5 (ax0)))
x3= xt + S(u (x0+ S (a (x0+ S (ax0)))))
xm— x0-\-S(a (x0 + S (a (x0+ S (a (xu S (a . . .)))))))

Expanding, we get
xm= x0 Sax0+ Sa Saxa-f Sa Sa Sax0+ . . . -f Sa Sa S a .. .Sax0
m tim
es
Taking x0 outside the brackets (x0 constant), we obtain
xm= [ 1-f- Su Su Su -f- . . . -f- Su S u . . . Sa] x0(11)
m tim
es
It has been proved (see Sec. 4.26) that if u(t) is a continuous
function, then the sequence {xm) converges. The limit of this
sequence is a convergent sériés:
*= [1 - f Su + S u S u + . . . ] x0 ( 12)
9.19 Solving a System of Linear Differential Equations 561
Note. If a(<) = const, then formula (12) assumes a simple form.

Indeed, by (10) we can write
Sa = aS 1 = a (t —1„)
Sa Sa = a*S (t —10) - a2 - ~ <(l-
Sa S a .. .Sa = am- —m!

'----- v------
m times
In this case, ( 12) takes the form
or
*= (13)
The method of solving the single équation (6 ) that we hâve just

reviewed is carried over in its entirety to the solution of system ( 1)
for the initial conditions (2 ).
In matrix form, system (1) with initial conditions (2) can be
written as
i- I M M M O I I - I M I (14)
11*11 = 11*11 for * = U (15)
If we use the rule of matrix multiplication and matrix intégra
tion, the solution of system (14), given condition (15), can be
reduced to the solution of the matrix intégral équation
t
l l * ( 0 l l = l |* , I H - S l l « ( 2 ) I M I * ( 2 ) l l r f * (1 6 )
We find the successive approximations
Il * . (0 II = Il *„ Il + S II a (Z) II-1| * . - 1 (Z) Il dz (17)
By successive substitution of the successive approximations under

the intégral, the solution of the system cornes out like this in
562 Ch. 9 Matrices
matrix form:
ll* (0 II = Il *0 II + S II a (Z|) Il ( || *,11

/o
+ | Il <*(**) II(il *o II+ J IIa (z,) || (.. . ) d 2 ^ j d 2 ^ d z t
ll*W II = IUo II + S II a (zt) l|-||*.ll<k»

t,
* 2x
+ S ||a (z ,)|| S \\a(zi)\\-\\x0\\dzt dzl + . . . (18)
u u
Using the intégration operator S, we can Write (18) as
||* ( / ) ||= [ ||£ || + 5 ||a || + S | | a | | S | | a | | + . . . ] | | x 0|| (19)
The operator in square brackets can be denoted by a single letter.
We dénoté it by tfiVSwn- Then équation (19) is compactly written as
\\x(t) || = 4 ^ 0 nl Kl l (2 0 )
It is interesting to note that if the coefficients of system (1) are
constants, then, using the rule for taking a common factor of ail
entries of the matrix outside the matrix symbol,* we can write
S |M I = ^ I M |
S |M |S I M I = ^ i ^ l M l a
S ||a ||S ||a ||S ||a || = — j ^ | | a ||3 and so on
In the case of constant coefficients, formula (19) assumes the form
Ilx (0 II = [ Il £ Il+ —p 2 IIII + Ila II2+ • • •
+ iLr - \ \ a Wm+ - - - ] \ \ xo\\ (2 i)

This équation can be symbolized in compact form as
Il * ( 0 II = * < '-'•> 11011 II * 0 II (2 2 )
* Here we do not discuss the question of passage to a limit for operations

performed on matrices.
1. Find the matrix of the inverse transformation for the linear transformation
= 3 *x4 - 2 x 2, y 2= 7 xf + 5 xa
5-2
-7 3 •
2. Find the matrix of the inverse transformation
yi = *i —**. V2= xl
0 1
Ans.
-1 1
l 2 2 0
3. Find the product of the matrices
3 4 3 —1
8 —2
Ans.
18 —4
4. Given the matrices
1 2 3 5 0 7
2 0 1 and B= 12 3
3 —1 1 —1 0 2
Find the matrices A B and BA.
4 4 19 26 3 221
Ans. 9 0 16 and 14—1
8
13 —2 20 5 —4 - > l
12 3
5. Given: A — 5 7 10 and E, a third-order unit matrix. Détermine the
4 3 1
matrix A + 2E.
3 2 3
Ans. 5 9 10
4 3 3
12
6. Given the matrix A = . Find the matrix Aa.
3 4
I 7 10
Ans.
115 22
7. Suppose A =
12 20
1 3 4 1| ' F'11** * + M-
Ans.
30 42
1 1 1
8. Find the inverse of A = 5 4 3 .
10 5 1
11—4 1
Ans —25 9 —2 .
1 5 -5 1
36 *
564 Ch. 9 Matrices
9. Write down the solutions of the System of équations

*i + 2x2+ *3= 10
2x, + x2+ x3 = 20
Xi + 3x2+ *3 = 30
in matrix form and find x lf x?, x3.
*1 —2 1 1 10
Ans. x2 _ — 1 0 1 .1 20 x x= 30, x2= 20, *3 = — 60.
*3 5 —1 - 3 i 3J
10. Write down in matrix form the solution of the System of équations
*1+ *3+ *3 = 3
5xx + 4x2 + 3*3 = 11
10*i + 5*2+ * 3 = 11.5
and find x2, x 3.
Xi 11—4 1 3
Ans. x2 _ —25 9 —2 11
*3 15—5 1 11.5
11. Find the eigenvalues and the eigenvectors of the matrix
1 2 —4
2 —2 —2
—4 —2 1
Ans. kj = 6, k2= k3 = — 3; xt = m i+ y m j— mk, t* is any vector satisfying

the condition ( t ^ ^ O , and m is an arbitrary scalar.
12. Find the eigenvectors of the matrix
cos a sin a
sin a cos a
Ans. They do not exist.
4 0 0
13. Find the eigenvectors of the matrix 0 4 0 .
0 0 4
Ans. Ail the vectors are eigenvectors.
14. Solve the following System of linear differential équations by the matrix
method:
W + * - 0
Ans. iï1= C1e * H C te - « , x2= - 2

APPENDIX
TABLE 1 • x
The values of the function O (x) = - J._f e~ i2dt
K » J0
and the reduced Lapi ace function <î>(je) = <D(px)
X OU) A &(x) A X OU) A OU) A
0 .0 0 0 .0 0 0 0 0 .0 0 0 0 okq 2 .2 5 0 .9 9 8 5 Q 0 .8 7 0 9 QQ
OCVt O OO
0 .0 5 0 .0 5 6 4 0 .0 2 6 9 2 .3 0 0 .9 9 8 8 0 .8 7 9 2
0 .1 0 0 .1 1 2 5 561 0 .0 5 3 8 269 2 .3 5 0 .9 9 9 1 3 0 .8 8 7 1 79
0 .1 5 0 .1 6 8 0 555 0 .0 8 0 6 268 2 .4 0 0 .9 9 9 3 2 0 .8 9 4 5 74
0 .2 0 0 .2 2 2 7 547 0 .1 0 7 3 267 2 .4 5 0 9995 2 0 .9 0 1 6 71
0 .2 5 0 .2 7 6 3 536 0 .1 3 3 9 2ô6 2 50 0 .9 9 9 6 1 0 /9 0 8 2 66
0 .3 0 0 .3 2 8 6 523 0 .1 6 0 4 265 2 .5 5 0 .9 9 9 7 1 0 .9 1 4 6 64
0 .3 5 0 .3 7 9 4 508 0 .1 8 6 6 262 2 .6 0 0 .9 9 9 8 1 0 .9 2 0 5 59
0 .4 0 0 .4 2 3 4 490 0 .2 1 2 7 261 2 .6 5 0 .9 9 9 8 0 0 .9 2 6 1 56
0 .4 5 0 .4 7 5 5 471 0 .2 3 8 5 258 2 .7 0 0 .9 9 9 9 1 0 .9 3 1 4 53
0 .5 0 0 .5 2 0 5 450 0 .2 6 4 1 256 2 .7 5 0 .9 9 9 9 0 0 .9 3 6 4 50
0 .5 5 0 .5 6 3 3 428 0 .2 8 9 3 252 2 .8 0 0 .9 9 9 9 0 0 .9 4 1 0 46
0 .6 0 0 .6 0 3 9 406 0 .3 1 4 3 250 2 .8 5 1 0 .9 4 5 4 44
0 .6 5 0 .6 4 2 0 381 0 .3 3 8 9 246 2 .9 J 0 .9 4 9 5 41
0 .7 0 0 .6 7 7 8 358 0 .3 6 3 2 243 2 .9 5 0 .9 5 3 4 39
0 .7 5 0 .7 1 1 2 334 0 .3 8 7 0 238 3 .0 0 1 .0 0 0 0 0 .9 5 7 0 36
0 .8 0 0 .7 4 2 1 309 0 .4 1 0 5 235 3 .0 5 0 .9 6 0 3 33
0 .8 5 0 .7 7 0 7 286 0 .4 3 3 6 231 3 .1 0 0 .9 6 3 5 32
0 .9 0 0 .7 9 6 9 262 0 .4 5 6 2 226 3 .1 5 0 .9 6 6 4 29
0 .9 5 0 .8 2 0 9 240 0 .4 7 8 3 221 3 .2 0 0 .9 6 9 1 27
1 .0 0 0 .8 4 2 7 218 0 .5 0 0 0 217 3 .2 5 0 9716 25
1 .0 5 0 .8 6 2 4 197 0 .5 2 1 2 212 3 .3 0 0 .9 7 4 0 24
1 .1 0 0 .8 8 0 2 178 0 .5 4 1 9 207 3 .3 5 0 .9 7 6 1 21
1 .1 5 0 .8 9 6 1 159 0 .5 6 2 0 201 3 .4 0 0 .9 7 8 2 21
1 .2 0 0 .9 1 0 3 142 0 .5 8 1 7 197 3 .5 0 0 .9 8 1 8 36
1 .2 5 0 .9 2 2 9 126 0 .6 0 0 8 191 3 .6 0 0 .9 8 4 8 30
1 .3 0 0 .9 3 4 0 111 0 .6 1 9 4 186 3 .7 0 0 .9 8 7 4 26
1 .3 5 0 .9 4 3 8 98 0 .6 3 7 5 181 3 .8 0 0 .9 8 9 6 22
1 .4 0 0 .9 5 2 3 85 0 .6 5 5 0 175 3 .9 0 0 .9 9 1 5 19
1 .4 5 0 .9 5 9 7 74 0 .6 7 1 9 169 4 .0 0 0 .9 9 3 0 15
1 .5 0 0 .9 6 6 1 64 0 .6 8 8 3 164 4 .1 0 0 .9 9 4 3 13
1 .5 5 0 .9 7 1 6 55 0 .7 0 4 2 159 4 .2 0 0 .9 9 5 4 11
1 .6 0 0 .9 7 3 6 47 0 .7 1 9 5 153 4 .3 0 0 .9 9 6 3 9
1 .6 5 0 .9 8 0 4 41 0 .7 3 4 2 147 4 .4 0 0 .9 9 7 0 7
1 .7 0 0 .9 8 3 8 34 0 .7 4 8 5 143 4 .5 0 0 .9 9 7 6 6
1 .7 5 0 .9 8 6 7 29 0 .7 6 2 1 136 4 .6 0 0 .9 9 8 1 5
1 .8 0 0 .9 8 9 1 24 0 .7 7 5 3 132 4 .7 0 0 .9 9 8 5 4
1 .8 5 0 .9 9 1 1 20 0 .7 8 7 9 126 4 .8 0 0 .9 9 8 8 3
1 .9 0 0 .9 9 2 8 17 0 .8 0 0 0 121 4 .9 0 0 .9 9 9 1 3
1 .9 5 0 .9 9 4 2 14 0 .8 1 1 6 116 5 .0 0 0 .9 9 9 3 2
2 .0 0 0 .9 9 5 3 11 0 .8 2 2 7 111 5 .1 0 0 .9 9 9 4 1
2 .0 5 0*9963 10 0 .8 3 3 2 105 5 .2 0 0 .9 9 9 6 2
2 .1 0 0 .9 9 7 0 7 0 .8 4 3 4 102 5 .3 0 0 .9 9 9 7 1
2 .1 5 0 .9 9 7 6 6 0 .8 5 3 0 96 5 .4 0 0 .9 9 9 7 0
2 .2 0 0 .9 9 8 1 5 0 .8 6 2 2 92
4 87
TABLE 2 The values of the function / ( * ) = — ___ ■■■* 2
____________________ y 2n_____
X / w X f(x) X | fM
0 .0 0 0 .3 9 8 9 1 .3 5 0 .1 6 0 4 • 2 .7 0 0 .0 1 0 4
0 .0 5 0 .3 9 8 4 1 .4 0 0 .1 4 9 7 2 .7 5 0 .0 0 9 1
0 .1 0 0 .3 9 7 0 1 .4 5 0 .1 3 9 4 2 .8 0 0 .0 0 7 9
0 .1 5 0 .3 9 4 5 1 .5 0 0 .1 2 9 5 2 .8 5 0 .0 0 6 9
0 .2 0 0 .3 9 1 0 1 .5 5 0 .1 2 0 0 2 .9 0 0 .0 0 6 0
0 .2 5 0 .3 8 6 7 1 .6 0 0 .1 1 0 9 2 .9 5 0 .0 0 5 1
0 .3 0 0 .3 8 1 4 1 .6 5 0 .1 0 2 3 3 .0 0 0 .0 0 4 4
0 .3 5 0 .3 7 5 2 1 .7 0 0 .0 9 4 0 3 .0 5 0 .0 0 3 8
0 .4 0 0 .3 6 8 3 1 .7 5 0 .0 8 6 3 3 .1 0 0 .0 0 3 3
0 .4 5 0 .3 6 0 5 1 .8 0 0 .0 7 9 0 3 .1 5 0 .0 0 2 8
0 .5 0 0 .3 5 2 1 1 .8 5 0 .0 7 2 1 3 .2 0 0 .0 0 2 4
0 .5 5 0 .3 4 2 9 1 .9 0 0 .0 6 5 6 3 .2 5 0 .0 0 2 0
0 .6 0 0 .3 3 3 2 1 .9 5 0 .0 5 9 6 3 .3 0 0 .0 0 1 7
0 .6 5 0 .3 2 3 0 2 .0 0 0 .0 5 4 0 3 .3 5 0 .0 0 1 5
0 .7 0 0 .3 1 2 3 2 .0 5 0 .0 4 8 8 3 .4 0 0 .0 0 1 2
0 .7 5 0 .3 0 1 1 2 .1 0 0 .0 4 4 0 3 .4 5 0 .0 0 1 0
0 .8 0 0 .2 8 9 7 2 .1 5 0 .0 3 9 6 3 .5 0 0 .0 0 0 9
0 .8 5 0 .2 7 8 0 2 .2 0 0 .0 3 5 5 3 .5 5 0 .0 0 0 7
0 .9 0 0 .2 6 6 1 2 .2 5 0 .0 3 1 7 3 .6 0 0 .0 0 0 6
0 .9 5 0 .2 5 4 1 2 .3 0 0 .0 2 8 3 3 .6 5 0 .0 0 0 5
1 .0 0 0 .2 4 2 0 2 .3 5 0 .0 2 5 2 3 .7 0 0 .0 0 0 4
1 .0 5 0 .2 2 9 9 2 .4 0 0 .0 2 2 4 3 .7 5 0 .0 0 0 4
1 .1 0 0 .2 1 7 9 2 .4 5 0 .0 1 9 8 3 .8 0 0 .0 0 0 3
1 .1 5 0 .2 0 5 9 2 .5 0 0 .0 1 7 5 3 .8 5 0 .0 0 0 2
1 .2 0 0 .1 9 4 2 2 .5 5 0 .0 1 5 4 3 .9 0 0 .0 0 0 2
1 .2 5 0 .1 8 2 6 2 .6 0 0 .0 1 3 6 3 .9 5 0 .0 0 0 2
1 .3 0 0 .1 7 1 4 2 .6 5 0 .0 1 1 9 4 .0 0 0 .0 0 0 1
x
_L2
TABLE 3 The values of the function <D(jt) = — --= l e 2 dz
Ÿ2* J
o
X c D (r) X 4> (X) X | <!><*)
0 .0 0 0.0000 0 .9 5 0 .3 2 8 9 1 .9 0 0 .4 7 1 3
0 .0 1 0 .0 0 4 0 1 .0 0 0 .3 4 1 3 2 .0 0 0 .4 7 7 2
0 .0 5 0 .0 1 9 9 1 .0 5 0 .3 5 3 1 2 .1 0 0 .4 8 2 1
0 .1 0 0 .0 3 9 8 1 .1 0 0 .3 6 4 3 2 .2 0 0 .4 8 6 1
0 .1 5 0 .0 5 9 6 1 .1 5 0 .3 7 4 9 2 .3 0 0 .4 8 9 3
0 .2 0 0 .0 7 9 3 1 .2 0 0 .3 8 4 9 2 .4 0 0 .4 9 1 8
0 .2 5 0 .0 9 8 7 1 .2 5 0 .3 9 4 4 2 .5 0 0 .4 9 3 8
0 .3 0 0 .1 1 7 9 1 .3 0 0 .4 0 3 2 2 .6 0 0 .4 9 5 3
0 .3 5 0 .1 3 6 8 1 .3 5 0 .4 1 1 5 2 .7 0 0 .4 9 6 5
0 .4 0 0 .1 5 5 4 1 .4 0 0 .4 1 9 2 2 .8 0 0 .4 9 7 4
0 .4 5 0 .1 7 3 6 1 .4 5 0 .4 2 6 5 2 .9 0 0 .4 9 8 1
0 .5 0 0 .1 9 1 5 1 .5 0 0 .4 3 3 2 3 .0 0 0 .4 9 8 6 5
0 .5 5 0 .2 0 8 8 1 .5 5 0 .4 3 9 4 3 .2 0 0 .4 9 9 3 1
0 .6 0 0 .2 2 5 7 1 .6 0 0 .4 4 5 2 3 .4 0 0 .4 9 9 6 6
0 .6 5 0 .2 4 2 2 1 .6 5 0 .4 5 0 5 3 .6 0 0 .4 9 9 8 4 1
0 .7 0 0 .2 5 8 0 1 .7 0 0 .4 5 5 4 3 .8 0 0 .4 9 9 9 2 7
0 .7 5 0 .2 7 3 4 1 .7 5 0 .4 5 9 9 4 .0 0 0 .4 9 9 9 6 8
0 .8 0 0 .2 8 8 1 1 .8 0 0 .4 6 4 1 4 .5 0 0 .4 9 9 9 9 7
0 .8 5 0 .3 0 2 3 1 .8 5 0 .4 6 7 8 5 .0 0 0 .5 0 0 0 0 0
0 .9 0 0 .3 1 5 9
INDEX
AbeFs theorem 283, 284 Bunyakovsky, V. Ya. 195, 196

absolutely convergent sériés 272, 310 Bunyakovsky’s inequality 196
Adam’s formula 138
addition of probabilities 449 calculus, operational 411
theorem on 449 cases (events) 447 '
adjoint 533 favourable 447
adjoint matrix 533 catenary 13
affine mapping 524 Cauchy’s test 264, 265
altemating sériés 269 causes
amplitude 100, 101 probability of 457
complex 358 theorem of 458
analysis, harmonie 355 central limit theorem 512
analytic function of a complex variab central moment (first, second, third)
le 310 474
analyzer, Fourier 356 centre 128
angular velocity 100 of dispersion 486
approximation(s) of distribution 486
first 312 of gravity 470
second 313 coordinates of (of a solid) 298
successive, method of 312 of probability distribution 469, 470,
zeroth 312 483
arithmetic mean 467 centred random variable 470
expectation of random variable com- certain events 447
pared with 466 characteristic équation 541, 542
augmented matrix 550 characteristic polynomial 546
autonomous System 119 Chebyshev, P. L. 514
auxiliary équation 76, 112, 116, 423 theorem of 514
axes of dispersion 503 circle of convergence 310
circulation of a vector 219
base vector 537 Clairaut’s équation 46-48
basis 539 closed contour 218
Bayes’s formula 457, 458 closed domain 158
Bayes’s rule for the probability of coefficient(s)
causes 458 Fourier (see Fourier coefficients) 330
bending moment 58 of thermal conductivity 382, 384
Bernoulli, Jakob 446 of a trigonométrie sériés 327
Bernoulli équation 32-34 cofactor 532
Bernoulli theorem 446 Collatz, Lothar 389
Bessel équation 303-305 collection, discrète 358
Bessel function of the first kind 306 column matrix 523
Bessel function of the second kind 307 common logarithms 299
Bessel functipns, System of 366 compatible random events 451
Bessel inequality 346 complementary events 450
Bezikovich, Ya. S, 137 complementary random events 449
binomial distribution 464 complété group of random events
binomial sériés 295-297 447
boundary conditions 376, 377, 384, 386 complété intégral (of a differential
boundary-value conditions 376 équation) 16, 56
boundary-value problem complété orthogonal System of functions
first 383, 395 366
second 395 complété sequence of functions 364
568 Index
complété System of functions 365 delay theorem 439, 440

complex amplitude 101, 358 delta function 440
complex form of Fourier sériés 357 density
complex Fourier coefficients 357 distribution 476
complex solution 101 law of uniform 480
complex variable probability 476
analytic function of 310 spectral 364
exponential function of 310 surface 160
cosine of 311 density function 470, 476
power sériés in 309, 310 of normal distribution 486
sine of 311 probability 476
condition(s) dépendent events 454
boundary 376, 377, 384, 386 dérivative of a matrix 550
boundary-value 376 déterminant
initial 16, 55, 376, 377, 384, 386 functional 184
Lipschitz 314 of a matrix 520, 522
conditional probability 454 Wronskian 70
conditionally convergent sériés 272 déviation 471
continuous random variable 475, 476 maximum 373
numerical characteristics of 482 root-mean-square 343, 471
probability density functions of 475 standard (see standard déviation)
continuous spectrum 364 471, 484
contour, closed 218 diagonal matrix 523
convergence diameter of a subdomain 198
circle of 310 die 447
domain of 310 difference(s)
necessary condition for 256 first 135
radius of 285, 310 of matrices 526
of a sériés 309 second 135
convergent sériés 309 of third order 136
absolutely 272 differential equation(s) 11, 14
conditionally 272 exact 34
convolution 431 existence of solution of 313
convolution formula 431 first-order 15-20
convolution theorem 429 higher-order 55-56
coordinate(s) linear 68, 69
of centre of* gravity of a solid 298 ordinary 14
curvilinear 182 différentiation of matrices 550
correspondence, one-to-one 182 Dirac delta function 441
cosine of a complex variable 311 Dirac function 441
curl (of a vector function) 239 direction of circulation 218
curve(s) direction field 18
distribution 476 Dirichlet intégral 350
intégral 17, 56, 121 Dirichlet-Neumann problem 398, 401
normal 485 Dirichlet problem 395
normal distribution 485 discrète collection 358
probability 476 discrète random variable 460
probability distribution 479 distribution law of 460
résonance 104 discrète spectrum 358
smooth 67 disjoint events 447
solution 121 dispersion
curvilinear coordinates 182 axes of 503
centre of 486
d’Alembert’s test 260 ellipses of 503
data, statistical 507 error scale of distribution 496, 497
décomposition theorem 426 sheil 444
degenerate saddle point 131 total ellipse of 503
del operator 244 unit ellipse of 503
in d e x 569
distribution 476 differential 11, 14

binomial (see binomial distribution) of elliptic type 374
centre of 486 exaçt differential 34
Gaussian 485 first-order linear 24
normal (see normal distribution) of forced oscillations 98
probability (see probability distri Fourier (for heat conduction) 374
bution) of free oscillations 98
uniform (see uniform distribution) heat-conduction 374, 375, 382, 383,
distribution curve 476 386
distribution density 476 of heat transfer in a plane 387
distribution function 479, 480, 481 of heat transfer in a rod 387
normal 488 higher-order differential 55-56
statistical 509 homogeneous 24
distribution law of discrète random homogeneous linear 69
variable 460 of hyperbolic type 374
Ditkin, V. A. 411 intégral 312
divergence (of a vector) 243 Lagrange’s 48-50
divergent sériés 309 Laplace’s 247, 374, 394
domain in cylindrical coordinates 400
closed 158 linear 29
of convergence (of a sériés) 274, 310 linear differential 68, 69
of intégration 159 Lyapunov-Parseval 346
irregular three-dimensional 199 nonhomogeneous linear 68
regular 161 nonhomogeneous second-order linear
regular three-dimensional 199 82
regular in the ^-direction 161 ordinary differential 14
regular in the y-direction 161 of parabolic type 374
dominance (see majorization) partial differential 14
dominated sériés 275-277 with separated variables 20
dot product 368 System of (see System of équations)
double intégral 159 telegraph 378
double root 88 transform 423
with variables separable 20
eigenfunetions 380 of the vibrating string 375
eigenvalues 380, 541 wave 376
eigenvector equipotential Unes 51
of a linear transformation 540 error
of a matrix 541 mean 493
element(s) mean arithmetic 497
of a matrix 522 of measurement 508
in a space 367, 369 probable 493
ellipse error distribution 496, 497
of dispersion 503 error estimation in approximate solu
total 503 tions 313
unit 503 escape-velocity problem 65
of inertia 193-195 Euclidean space, n-dimensional 367
entry of a matrix 522 Euler’s formula 294
envelope (of a family of curves) 39, 42 Euler’s formula for a complex z 311
equal matrices 523 Euler’s method (of approximate solu
equally probable events 447 tion of first-order differential équ
equation(s) ations) 135
auxiliary 76, 112, 116, 423 Euler’s polygon 134
Bernoulli’s 32-34 event(s)
Bessel’s 303-305 certain 447
characteristic 541, 542 complementary 450
Clairaut’s 46-48 complementary random 449
of continuity 395, 397 dépendent 454
of a compressible liquid 397 disjoint 447
570 In d ex
event(s) Fourier intégral 358-371

equally probable 447 in complex form 362-371
impossible 447 Fourier inverse transform 364
independent (see independent events) Fourier method 378
intersection of 451 Fourier sériés 327-371
mutually exclusive 447 in complex form 356
random (see random events) expansion of functions in a (compared
exact differential équation 34 with décomposition of vectors)
existence of solution of a differential 367
équation 312 Fourier sine transform 361
existence theorem of a line intégral 219 Fourier transform 364
expectation 446 free oscillations 98, 435
of a centred random variable 471 équation of 98
of a constant 471 frequency 100
• mathematical 466 relative (see relative frequency)
of normal distribution 485 frequency function 476
of a random variable compared with frequency polygon 460
arithmetic mean 467 frequency table 508, 509
exponential function of a complex function(s)
variable 310 analytic (of a complex variable)
310
factor Bessel, of the first kind 306
integrating 37-38 Bessel, of the second kind 307
stretching 520 Bessel, System of 366
Faltung 431 complété System of 365
family continued in even fashion 342
of curves 17 continued in odd fashion 342
one-parameter 39 delta 440
of orthogonal trajectories 51 density (see density function)
favourable cases 447 Dirac 441
Fikhtengolts, G. M. 273 Dirac delta 441
flow Unes 51 distribution (see distribution func
flux (of a vector field through a sur tion)
face) 233 exponential (of a complex variable)
force, restoring 97 310
forced oscillations 98, 102-105 frequency 476
équation of 98 harmonie 247, 394
form, quadratic (see quadratic form) Heaviside unit 413, 442
formula(s) homogeneous 24
Adams* 138 Laplace 488, 492
Bayes’s 457, 458 Laplace, reduced 494, 495
convolution 431 linearly dépendent 80
Euler’s 294 linearly independent 80
for a complex 2 311 normal (of normal distribution) 492
Green’s 225-227 original 412
Liouville’s 71 orthogonal with weight x 367
Ostrogradsky 241-244 pieeewise continuous 347
Ostrogradsky-Gauss 243 pieeewise monotonie 330
Stokes* 236-241 probability density 476
of total probability 456 of a random variable 474
for transformation of coordinates in reduced Laplace 494, 495
a double intégral 185 sequence of (orthogonal) 370
Fourier analyzer 356 complété 364
Fourier coefficients 330, 366 sériés of 274
complex 357 space of 369
Fourier cosine transform 361 spectral 364
Fourier équation for heat conduction statistical distribution 509
374 unit impulse 440
In d ex 571
function space Fourier 358-360

linear 367 Fourier, in complex form 363
linear (with quadratic metric) 370 improper iterated 180
functional déterminant 184 iterated 161
functional sériés 274 line 216, 219-220
of a matrix 551
Gauss (Ostrogradsky-Gauss formula) particular 16
243 Poisson’s 179, 393, 405
Gaussian distribution 485 probability 489
general solution (of a differential équ threefold iterated 198-204
ation) 16, 17, 56 triple (see triple intégral)
general ized problems 449 intégral curves 17, 56, 121
géométrie probability 452 intégral équation 312
géométrie progression 253 intégral sum 158
gravity, centre of 470 intégral test for convergence 266-268
Green, George 227 integrate a differential équation 11
Green’s formula 225-227 integrating factor 37-38
intégration
Hamiltonian operator 244 domain of 159
harmonie analysis 355 of matrices 550
harmonie function 247, 394 région of 159
harmonie oscillations 100 sense of 217
harmonie sériés 257, 258 interior point (of a région or domain)
harmonies 358 161
heat-conduction équation 374, 375, 382, intersection of events 451
383, 386 inverse Fourier transform 364
heat transfer inverse matrix 531
équation of (in a rod) 387 inverse transformation 524
in space 384 inversion, matrix 532
Heaviside unit function 413, 442 irregular domain 199
histogram 508, 509 irrotational vector field 246
homogeneous équation 24 isocline 18
homogeneous function 24 isogonal trajectories 50, 53-54
homogeneous linear équation 69 iterated intégral 161
hypothèses (see causes and probability évaluation of 165
of causes) 457 improper 180
threefold 199-203
identity matrix 523, 529 itération, method of 312
identity transformation 529 Jacobi, G. G. 184
impossible events 447 Jacobian 184, 207
improper iterated intégral 180
inclusive “or” 449 Kuznetsov, P. I. 411
independent events 452
multiplication of probabilities 452 L-transform 412
inequality Lagrange’s équation 48-50
Bessel’s 346 Laplace équation 247, 374, 394
Bunyakovsky’s 196 in cylindrical coordinates 400
Schwarz’ 196 Laplace function 488, 492
inertia, moment of (see moment of reduced 494, 495
inertia) Laplace theorem 512, 515, 516
initial condition(s) 16, 55, 376, 377, Laplacian operator 247, 250, 354
384, 386 large numbers, law of 516
initial phase 100 law
inner product 368 of large numbers 516
intégral (s) 14 of uniform distribution 479, 481
complété 16, 56 Leibniz’ theorem 269, 270
Dirichlet’s 350 length of a vector 368
double 159 line (see Unes)
572 In d ex
line intégral 216, 219-220 intégration of 550

linear differential équation 68, 69 inverse 531
linear équation 29 minor of 549
linear function space 367 multiplication of by a scalar 526
with quadratic metric 370 multiplication of by unit 529
linear map 517 operations on 526
linear transformation 517 orthogonal 537, 539
eigenvector of 540 product of 526, 527
linearity property (of a transform) 415 of quadratic form 547
linearly dépendent functions 80 rank of 549
linearly dépendent solutions 69 row 523
linearly independent functions 80 solution 552
linearly independent solutions 69 square 522
Unes sum of 526
equipotential 51 transforming a vector into a vector
flow 51 by 529
Liouville’s formula 71 transpose of 523
Lipschitz condition 314 transposed 523
localization, principle of 352 unit 523, 529
logarithm, natural 299 matrix inversion 532
logical product 451 matrix notation 517
logical sum 449 for a linear équation of order n 557
Lurle, A. I. 411 for a System of linear differential
Lyapunov, A. M. 118, 132, 512 équations with variable coefficients
Lyapunov-Parseval équation 346 558
Lyapunov stable (about solutions, con for Systems of differential équations
ditions) 118 552
Lyapunov theorem 512, 513 for Systems of linear équations 534
Lyapunov’s theory of stability 117 matrix symbol 520
maximum déviation 343
mean
Maclaurin’s sériés 290-291 arithmetic (see arithmetic mean)
majorization 283 statistical 511
map, linear 517 weighted 511
mapping(s) mean arithmetic error 497
affine 524 mean error 493
one-to-one 182, 524 mean-value theorem 166, 201
orthogonal 537 measurement, error of 508
mathematical expectation 466 médian 484
mathematical statistics 444 of normal distribution 486
problems of 507 method
matrices (see matrix; différence (for solving Systems of
matrix (matrices) 517, 522 équations) 142
addition of 526 Euler’s 133-135
adjoint 533 Fourier 378
augmented 550 of itération 312
column 523 of successive approximation 312
dérivative of 550 of variation of arbitrary constants
déterminant of 520, 522 (parameters) 84
diagonal 523 metric (of a space) 369
différence of 526 Mikusinski, J an G. 411
différentiation of 550 minor of a matrix 549
eigenvector of 541 mirror reflection 521
element of 522 mode 460, 484, 486
eritry of 522 model, urn 449
equal 523 modulus
identity 523, 529 of précision 498
intégral of 551 of a vector 368
In d ex 573
moment(s) orthogonal matrix 537, 539

bending 58 orthogonal sequence of functions 370
central (see central moment) orthogonal svstem of functions, comp
of inertia lété 366
of area of a plane figure 191ff orthogonal trajectories 50-53
of a solid 207 orthogonal transformation 538
static 197 orthogonal vectors 368
motion oscillations
stationary 51 forced 98, 102-105
steady-state 51 free 98, 435
multiplication of probabilités of inde- harmonie 100
pendent events 452 Ostrogradsky, M. V. 185, 227, 243
mutually exclusive events 447 Ostrogradsky formula 241-244
Ostrogradsky-Gauss formula 243
natural logarithms 299 Panov, D. Yu. 389
n-dimensional Euclidean space 367 parabola, safety 43
Neumann problem 395 Parseval (Lyapunov-Parseval - équation)
nodal point 131 346
nonhomogeneous linear équation 68 partial differential équations 14
nonhomogeneous second-order linear équ particular intégral 16
ations 82 particular solution 16, 56
norm 369 Peano’s theorem 317
normal curve 485 period of oscillation 100
normal distribution 485 Petrovsky, I. G. 117, 134, 319
density function of 486 phase, initial 100
expectation of 485 phase plane 121
médian of 486 pieeewise continuous function 347
in the plane 502 pieeewise monotonie function 330
normal distribution curve 485 plane, phase 121
normal distribution function 488, 492 plus-and-minus sériés 271
normal System of équations 106 point (s)
nth partial sum of a sériés 253 degenerate saddle 131
number(s) interior 161
large, law of 516 nodal 131
law of large 516 ordinary 46
wave 358, 365 saddle 124
numerical sériés 253 singular 46
of a space 367, 369, 370
stable focal 125, 126
observed values 507 stable nodal 122, 123
one-parameter family of curves 39 unstable focal 126, 127
one-to-one correspondence 182 unstable nodal 123
one-to-one mapping 182, 524 Poisson’s intégral 179, 393, 405
operational calculus 411 polygon
operator Euler’s 134
y-operator 244 frequency 460
del 244 of probability distribution 460
Hamiltonian 244 polynomial(s)
Laplacian 247, 250, 394 characteristic 546
“or” (inclusive) 449 orthogonal Legendre 367
order of a differential équation 14 potential
ordinary differential équation 14 of a gravitational field 241
ordinary point 46 of a vector 230
original function 412 potential vector field 245
original-transform tables 413 power sériés 283
orthogonal Legendre polynomials 367 in a complex variable 309, 310
orthogonal mapping 537 précision, modulus of 498
574 In d ex
principal directions of quadratic forms quadratic forms 547

548 matrix of 547
principal standard déviations 502 principal directions of 548
principal value of an intégral 363
principle of localization 352
probable error 493 radium, rate of decay of 22
probabilities (see probability) radius of convergence 285, 310
probability(-ties) radius vector 205
addition of 449 random event(s) 445
Bayes’s rule for (of causes) 458 complementary 449
calculation of 447 complété group of 447
of causes 457 compatible 451
Bayes’s rule for 458 mutually exclusive 447
classical définition of 447 probability of 445, 446
conditional 454 relative frequency of 445
géométrie 452 random variable(s)
multiplication of (of independent centred 470
events) 452 continuous 475, 476
of random event 445, 446 discrète 460
of relative frequency (in repeated functions of 474
trials) 462 two-dimensional 499
theorem on addition of 449 probability density function of 499
total 454 rank of a matrix 549
formula of 456 rate of decay of radium 22
probability curve 476 ratio of a géométrie progression 253
probability density 476 reduced Laplace function 494, 495
probability density function 476 reflection, mirror 521
of continuous random variable 475, région of intégration 159
476 regular domain 161, 199
of two-dimensional random variables relative frequency 462
499 probability of (in repeated trials) 462
probability distribution of random event 445
centre of 469, 470, 483 résonance 105, 439
curve of 479 résonance curves 104
polygon of 460 restoring force 97
probability intégral 489 résultant 431
probability theory 444 rigi'dity, spring 97
subject of 445, 447 root, double 88
problem(s) simple (single) 87
Dirichlet 395 root-mean-square déviation 343, 471
Dirichlet-Neumann 398, 401 row matrix 523
escape-velocity 65 rule
on firing to a first hit 461 Bayes’s (for the probability of causes)
first boundary-value 383, 395 458
généra lized 449 three-sigma 496
Neumann 395
second boundary-value 395
of a simple pendulum 62-66 saddle point 124
urn-model 449 safety parabola 43
product Samarsky, A. A. 382
dpt 368 scalar product 368
inner 368 Schwarz inequality 196
logical 451 sense of description 218
of matrices 526, 527 sense of intégration 217
scalar 368 separated variables 21
progression, géométrie 253 sequence of functions
property, linearity (of a transform) 415 complété 364
Prudnikov, A. P. 411 orthogonal 370
In d ex 575
sériés spectral density 364

absolutely convergent 272, 310 spectral function 364
altemating 269 spectrum 358
binomial 295-297 continuas 364
with complex terms 308 discrète 358
conditionally convergent 272 spring rigidity 97
convergence of 309 square matrix 522
convergent 309 stable, Lyapunov (about solutions, con
divergent 309 ditions) 118
dominated 275-277 stable focal point 125, 126
Fourier 327-371 stable nodal point 122, 123
définition of 330 stable solution 118, 122, 125
functional 274 standard deviation(s) 471, 484
of functions 274 of normally distributed random va
harmonie 257, 258 riable 487
Maclaurin’s 290-291 principal 502
numerical 253 static moments 197
plus-and-minus 271 stationary motion 51
power 283 statistical data 507
power (in a complex variable) 309, 310 statistical distribution function 509
statistical 508, 509 statistical sériés 508
sum of 309 statistical mean 511
Taylor’s 290, 291 statistical variance 511
trigonométrie 327 statistics
shell dispersion 444 mathematical 444
shift theorem 416 problem of 507
similarity transformation 544 steady-state motion 51
simple pendulum, problem of 62-66 Stokes, G. H. 239
simple root 87 Stokes’ formula 236-241
simple shear transformation 521 Stokes* theorem 239
sine of a complex variable 311 stretching factor 520
single root 87 subdomain 158
singular point 46 sum
singular solution of differential équ intégral 158
ation 45 logical 449
singular transformation 525 of matrices 526
Smimov, V. I. 412 nth partial 253
smooth curve 67 of a sériés 253, 309
solenoidal vector field 246 surface density 190
solution(s) symbolic vector 244
complex 101 System
of a differential équation 14 autonomous 119
general 16, 17, 56 of Bessel functions 366
linearly dépendent 69 of équations, différence method for
linearly independent 69 solving 142
particular 16, 56 of functions, complété 365
singular 45 of homogeneous linear differential
stable 118, 122, 125 équations with constant coef
unstable 123, 124 ficients lllff
vector 552, 553
solution curves 121
solution matrix 552 table(s)
solve (a differential équation) 18 frequency 508, 509
space original-transform 413
of functions 369 of trarsforms 421-422
linear function 367 Taylor’s sériés 290, 291
with quadratic metric 370 telegraph équations 378
n-dimensional Euclidean 367 terms of a sériés 253
576 In d ex
test simple shear 521

Cauchy’s 264, 265 singular 525
cTAlembert's 260 transpose of a matrix 523
intégral (for convergence) 266-268 transposed matrix 523
theorem trigonométrie sériés 327
Abel’s 283, 284 triple intégral 197, 198, 201, 204, 205,
on addition of probabilities 449 206
Bemoullf 446 change of variables in 204
of causes 458 in cylindrical coordinates 204
central limit 512 in spherical coordinates 205
of Chebyshev 514 two-dimensional random variables 499
convoi ution 429 probability density function of 499
décomposition 426
delay 439, 440 uniform distribution, law of 479, 481
existence (of a line integra!) 219 union 449
of existence and uniqueness of solu uniqueness theorem of solution of a
tion of a differentiai équation 15 differentiai équation 318
Laplace 512, 515, 516 unit ellipse of dispersion 503
Leibniz 269, 270 unit impulse function 440
Lyapunov’s 512, 513 unit matrix 523, 529
mean-value 166, 201 unit vector 537
Peano’s 317 unstable focal point 126, 127
shift 416 unstable nodal point 127
Stokes’ 239 unstable solution 123, 124
uniqueness (of solution of differen urn model 449
tiai équation) 413 urn-model problems 449
work-kinet ic energy 240
theory valuc(s)
probability 444 determining a suitable (of a measured
subject of 445, 447 quantity) 511
of stability, Lyapunov’s 117 observed 507
thermal conductivity, coefficient of principal (of an intégral) 363
382, 384 variables, separated 21
threefold iterated intégral 199-203 variables separable 21
three-sigma rule 496 variance 471, 472, 484
Tikhonov, A. N. 382 or normally distributed random va
total ellipse of dispersion 503 riable 487
total probability 454 statistical 511
formula of 456 vector(s) 370
trajectories base 537
isogonal 50, 53-54 length of 368
orthogonal 50-53 modulus of 368
transform(s) 413, 414 orthogonal 368
of dérivatives 419, 420 symbolic 244
différentiation of 417-419 unit 537
Fourier 364 vector field
Fourier cosine 361 irrotational 246
Fourier sine 331 potential 245
inverse Fourier 364 solenoidal 246
Laplace 412 vector solution 552, 553
L-transform 412 velocity, angular 100
transform équation 423 vibrating string, équation of 375
transformation(s)
identity 529 wave équation 374, 376
inverse 524 wave numbers 358, 365
linear 517 weighted mean 511
orthogonal 538 Wronskian 70-73, 83, 84, 94
similarity 544 Wronskian déterminant 70
ABOUT THE PUBLISHERS
Mir Publishers of Moscow pu-

blish Soviet scientific and
technical literature in sixteen
languages-English, German,
French, Italian, Spanish, Portu-
guese, Czech, Slovak, Finnish,
H ungarian, Mongolian, Arabie,
Persian, H indi, Vietnamese and
T am il. T itles include textbooks for
higher technical schools and
vocational schools, literature
on the natural sciences and
m edicine, including textbooks
for medical schools, popular
science and science fiction.
The contributors to Mir Pub
lishers' list are leading Soviet
scientists and engineers in ail
fields of science and technology
and include more than 40 Mem-
bers and Corresponding Mem-
bers of the USSR Academy of
Sciences.
Skilled translators provide ,
a high standard of translation
from the original Russian.
Many of the titles already
issued by Mir Publishers hâve
been adopted as textbooks and
m anuals at educîonal establish
ments in France, Switzerland,
Cuba, Syria, India, and many
other countries.
Mir P ublishers’ books in
foreign languages are exported
by V/O “Mezhdunarodnaya
Kniga” and can be purchased
or ordered through booksellers
in your country dealing with
V/O “Mezhdunarodnaya Kniga”.
Other books
for your library
3
A BOOK OF PROBLEMS IN
ORDINARY DIFFERENTIAL
EQUATIONS
b y M . K r a s n o v a n d o th e r s
VECTOR ANALYSIS
b y M . K r a s n o v a n d o th e r s
A COURSE OF MATHEMATICAL
ANALYSIS, VOLS. M I
b y S . N ik o ls k y
M IR P U B U S H E R S

Piskunov - Differential and Integral Calculus - Volume 2 - Mir

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Piskunov - Differential and Integral Calculus - Volume 2 - Mir

Uploaded by

Copyright:

Available Formats

N.

MIR PUBLISHERS • M OSCOW

Textbook by the late Prof. Nikolai

H3ZIATEJlbCTBO «HAYK A» MOCKBA

Translatée! from the Russian

© Engli^h translation, Mir Publishers, 1974

CHAPTER 1 D1FFERENTIAL EQUATIONS

1.1 Statement of the problem. The équation of motion of a body with

CHAPTER 2 MULTIPLE INTEGRALS

2.1 Double in té g r a is ...................................................................... 158

CHAPTER 3 LINE INTEGRALS AND SURFACE INTEGRALS

4.20 Expansion of the function ln (l-fx ) in a power sériés. Computing

CHAPTER 5 FOU R 1ER SERIES

CHAPTER 6 EQUATIONS OF MATHEMATICAL PHYSICS

6.10 The solution of Dirichlet’s problem for a circle 401

CHAPTER 7 OPERATIONAL CALCULAIS AND CERTAIN

8.15 Normal distribution. The expectation of a normal distribution . . . 485

1.1 STATEMENT OF THE PROBLEM. THE EQUATION

where is the accélération of a moving body (the dérivative of the velocity

Thus, the constant C fs found, and the sought-for dependence of v on / is

Example 1. Suppose we hâve the équation

The functions y = sin*, y = 2 cos x, t/ = 3 sin x — cos x and, in general,

1.3 F1RST-ORDER DIFFERENTIAL EQUATIONS

The géométrie meaning of the theorem consists in the tact that

the general solution is a family of functions y = — ; this can be checked by

From the géométrie viewpoint, the general solution (complété

équation but also the associated intégral curve is a solution of

and let t/ = <p(x, C) be the general solution of this équation. This

The isoclines of the given differential équation are

This is a family of straight lines. They are constrycted in Fig. 3.

Let there be given a family of functions that dépends on a

Hence, to establish a relationship between *, y and ^ that is,

Putting the value C = — into

1.4 EQUATIONS WITH SEPARATED AND SEPARABLE VARIABLES.

where the right side is a product of a function dépendent only

which is a relationship connecting the solution of y, the indepen-

is called an équation with separated variables. Front what has

This is the équation of a family of concentric

2. An équation of the form

is called an équation with separable

Example 3. Given the équation

* Having in view subséquent transformations, we denoted the arbitrary

Some other problems of physics and chemistry can be reduced to équations

Its general solution is of the form

1.5 HOMOGENEOUS FIRST-ORDER EQUATIONS

is called homogeneous in x and y if the function f(x, y) is a ho­

Thus, a homogeneous function of degree zéro is dépendent only

Making the substitution

Putting this expression of the dérivative into équation (T), we

This is an équation with variables separable

Putting the ratio — in place of U after Intégration, we get the

Whence, integrating, we find

Substituting u = - j , we get the general solution of the original équation:

It is impossible here to get y as an explicit function of x in terms of elemen-

1.6 EQUATIONS REDUCIBLE TO HOMOGENEOUS EQUATIONS

If c1= c = 0, then équation (1) is obviously homogeneous. Now let

Putting into (2) the expressions for x, y , and , we obtain

Choose h and k so that the following équations are fulfilled:

Solving this équation and passing once again to x and y by

the équation is reduced to one with variables separable.

Putting into (5) expressions (6) and (7), we get

which is an équation with variables separable.

and we get an équation with variables separable

cannot be solved by the substitution x = x x+ h9 y = y x + k t since in this case

is called homogeneous in x and y if the function f(x, y) is a ho

Note. In applications, one frequently encounters linear équa

It is obvious that any function p(x, y) that satisfies this équa

On the other hand, from the same équa

As has already been indicated above (see Sec. 1.2), a differen