Multiple Integrals,

Field Theory
and
Series

B.M. Budak, S.V. Fomin

MIR PUBLISHERS MOSCOW

B. M. EyAaK, C. B. <Pomhji

IxpaTHLie HHTerpajTw
II PH A M

II3A&TCJibCmo <Hayi<a»
MocKua
 B. M. Budak, S. V. Fomin

Multiple Integrals,
Field Theory and Series
An Advanced Course
in Higher Mathematics

Translated from, the Ilussian

by

V. M. VOLOSOV, D. Sc.

MIR PI/BL1S11KI!S . MOSCOW

UDC 517.3/517.52 (075.8) = 20

First published 1973

Revised from the 1907 Russian edition

Ha aMAuuenoM x-iuxe

(g) English translation, Mir Publishers, 1973

0223-197
OU (0i>—73
 Preface

The present book is based on lectures given by the authors at the

Physical Department of the Lomonosov State University of Moscow.
In the presentation of the material much emphasis has been placed
on the physical meaning of mathematical notions and their interrela­
tion as well as on the applications and computational aspects.
Chapters 1-0 and Supplement 2 (On Universal Digital Computers)
have been written by S.V. Fomin and Chapters 7-11 and Supple­
ment 1 (Asymptotic Expansions) by B.M. Budak. The authors have
discussed together the general plan of the book and many details
concerning the presentation of the material.
In the preparation of the book the authors have received valuable
advice from their colleagues V.A. Ilyin, E.G. Poznyak,
A-G- Sveshnikov and others. The authors owe very much to
A.N. Tikhonov for his helpful comments and aid- Some important
observations have been made by N.V. Yelimov and L.D. Kudryavtsev
who have read the manuscript of the book. To all of them the
authors express their warmest gratitude.

B . M . Bu dak
S.V. Fomin
 Contents

CHAPTER 1. DOUBLE IN TEG R A LS..................................................... 19

§ I. Auxiliary Notions. Area of a Plane F i g u r e ................................. 21
1. Interior and Boundary Points. D o m a in ................................ 21
2. Distance Between Two S e t s ..................................................... 22
3. Area of a Plane F ig u r e ............................................................. 23
4. Basic Properties of A r e a ............................................................. 27
5. The Concept of M easure............................................................. 28
§ 2. Definition and Basic Properties of Double I n t e g r a l ................ 29
1. Definition oT Double In te g r a l................................................. 29
2. Conditions for Existence of Double integral. Upper and Lower
Darboux Sums ............................................................................... 31
3. Some Important Classes of lnlegrableF u n ction s................... 38
4. Properties of Double In te g r a l................................................. 39

§ 3. Additive Set Functions. Derivative of a Set Function irith Hespect

to A r e a ................................................................................................. 41
1. Point Functions and SetF u n ction s........................................... 41
2. Double Integral as an Additive Function of Its Domain of
In teg ra tio n ..................................................................................... 42
3. Derivative of a Set Function with ltcspcct toArea . . . . 42
4. Derivative of a Double Integral with Respect to the Area
of Its Domain ofIn tegration ..................................................... 43
5. Reconstruction of an Additive Set Function from its Deri­
vative ............................................................................................. 44
6. Definite Integral of a Function of One Argument as a Function
of Its Interval of In tegration ..................................................... 4G
7. Extension of Additive Set F u n ctio n s..................................... 46

$ 4. Some Physical and Geometrical Applications of the Double 1 ntcgral 47

1. Evaluating V o lu m es............................................. . . . . 47
2. Computing A r e a s ............................. . . . . 48
3. Mass of a P l a t e ............................................................................. 48
4. Coordinates of the Centre of Gravity of aP l a t e .................... 49
5. Moments of Inertia of a P l a t e ................ ................................ 50
8 CONTENTS

G. Linninous Flux Incident on a P l a t e ........ . . 51

7. Flux of a Fluid Through the Cross Section of a Channel. . . 51

§ 5. Reducing Double Integral to a Twofold Iterated Integral . . . 52

1. Heuristic Considerations................................................. 52
2. The Case of a Rectangular Domain of Integration. . . . 54
3. The Case of a Curvilinear Domain*.............................. 56

§ 6. Change of Variables in Double I n t e g r a l ..................................... 61

1. Mapping of Plane F igu res........................................ 61
2. Curvilinear Coordinates ................................ ... . . 63
3. Polar Coordinates.............................................................. 64
4. Statement of the Problem of Changing Variables in the
Double I n te g r a l.................................................................. 66
5. Computing Area in Curvilinear Coordinates.............. 66
6. Change of Variables in Double In teg ra l...................... 74
7. Comparison with One-Dimensional Case. Integral Over an
Oriented D o m a i n ............................................................. 77

CHAPTER 2. TRIPLE INTEGRALS AND MULTIPLE INTEGRALS

OF HIGHER O R D E R ..................................................... 79
§ 1. Definition and Basic Properties of 'Triple I n t e g r a l ........... 79
1. Preliminary Observations. Volume of a Space Figure . . . 79
2. Definition of Triple In teg ra l.......................................... 81
3. Conditions for Existence of Triple Integral. Integrability
of Continuous F u nction s.................................................. 82
4. Properties of Triple In teg ra l...................................... 83
5. Triple Integral as an Additive Set F u nction.............. 84

§ 2. Some Applications of Triple Integral in Physics and Geometry . . 85

1. Computing V o lu m es.......................................................... 85
2. Finding tho Mass of a Solid from Its D e n sity .......... 85
3. Moment of I n e r tia .............................................................. 85
4. Determining the Coordinates of the Centre of Gravity . . . 86
5. Gravitational Attraction of a Material Point by a Solid . . . 86
§ 3. Evaluating Triple I n t e g r a l ..................................................... 87
1. Reducing Triple Integral Over a Rectangular Parallelepiped
to an Itoratml In to g rel.................................................. 88
2. Reducing Triple Integral Ovor a Curvilinear Domain to an
Iterated Integral ........................................................................ 90

§ -I. Change of Variables in Triple I n t e g r a l ......................................... 93

1. Mapping of Space F igu res............................................... 94
2. Curvilinear Coordinates in S p a c e .................................. 94
 CONTENTS 9

3. Cylindrical and Spherical Coordinates.................................... 95

4. Element of Volume in Curvilinear Coordinates................ 97
5. Change of Variables in Triple Integral. Geometric Meaning
of the J a co b ia n .................................................................... . 98
§ 5. Multiple Integrals of Higher O r d e r ............................................. 102
4. General R e m a r k s ........................................................................ 102
2. E xam p les......................................................................................... 403
CHAPTER 3. ELEMENTS OF DIFFERENTIAL GEOMETRY . . . 407
§ 7. Vector Function of a Scalar A r g u m e n t ......................................... 107
1. Definition of a Vector Function. Limit.Continuity . . . . 107
2. Differentiation of a Vector F u n ctio n ..................................... 108
3. llodograph. Singular P o in t s ..................................................... 110
4. Taylor’s F o r m u la ......................................................................... 141
5. Integral of a Vector Function with Respect to Scalar Argument 111
G. Vector Function of Several Scalar A rgum ents.................... 112
§ 2. Space C u r v e s ............................................................................. 113
1. Vector Equation of a C u rv e............................................ . 113
2. Moving T rihedron..................................................................... 115
3. Frenet-Serrct F orm ulas..................................................... 116
4. Evaluating Curvature and T orsion ............................. 117
5. Coordinate System Connected with Moving Trihedron 119
6. The Shape of a Curve in the Vicinity of Its Point . . . . 121
7. Curvature of a Plane C u rv e..................................................... 123
8. intrinsic Equations of a C u rv e................................. 124
9. Some Applications to M echanics................................. 126
§ 3. Parametric Equations of a S u r f a c e ................................................. 128
1. The Concept of a S u rfa ce..................................................... 128
2. Paramotrization of a S u rfa ce................................................. 130
3. Parametric Equations of a S u rfa ce..................................... 132
4. Curves on a S u rfa ce............................................................. 133
5. Tangent Plane ......................................................................... 133
6. Normal to a Surface ............................................................. 135
7. Coordinate Systems in Tangent P la n e s ................................. 135

£ 4. Determining Lengths , Angles and Areas on a Curvilinear Surface.

First Fundamental Quadratic Form of a S u r f a c e .................... 137
1. Affine Coordinate System in the P la n e ................................. 137
2. Arc Length of a Curve on a Surface. First Fundamental
Quadratic F o r m ............................................................................. 139
3. Anglo Between Two C urves......................................................... 141
4. Definition of Area of a Surface. TheSchwarz Example 142
5. Computing Area of a Smooth Surface . ................ 144
10 CONTENTS

$ 5. { urvaturc of Curves on a Surface . Second Fundamental Quadratic

Form of a Surface ............................................................................ 149
1. Normal Sections of a Surface and Their Curvature . . . . 149
2. Second Fundamental Quadratic Form of a Surface . . . 151
3. Dupin Indicatrix ........................................................................ 153
-4. Principal Directions and Principal Curvatures of a Surface.
Equation of E u le r ........................................................................ 154
5. Determining Principal Curvatures . . . . . . 150
0. Total Curvature and Mean Curvature . . . . 157
7. Classification of Points on a Surface . . . . 157
8. The First and the Second Fundamental Quadratic Forms
as Invariants of a S u rface........................................................ 159

§ 6 . Intrinsic Properties of a S u r f a c e .................................... 160

1. Applicable Surfaces. Necessary and Sufficient Condition for
A p p lic a b ility ................................................................................ 100
2. Intrinsic Properties of a S u rface........................ . . 101
3. Surfaces of Constant Curvature................................................ 163
CHAPTER 4. LINE INTEG RALS............................................................ 105
/. Line Integrals of the First T y p e ............................................... 1C5
1. Definition of Line Integral of the First T y p e .................... 105
2. Properties of Line In teg ra ls.................................................... 169
3. Some Applications of Line Integrals of the First Type . . . 170
4. Line Integrals of the First Type in S p a c e ............................ 172
§ 2. Line Integrals of the Second T y p e .................................... 173
1. Statement of the Problem. Work of a Field of Force . . . 173
2. Definition of Line Integral of the Second Type . . . 174
3. Connection Between Line Integrals of the First and the
Second T y p e s ........................................................ . . . 175
4. Evaluating Line Integral of the Second T y p e .................... 177
5. Dependence of Line Integral of the Second Type on tho
Orientation of the Path of Integration................................ 180
G. Lino Integrals Along Self-Intersecting and Closed Paths 180
7. Line Integral of the Second Typo Over a Space Curve . . . 181
§ 3. Green's Formula . . . . 183
1. Derivation of Green’s F orm u la................................................ 184
2. Application of Green’s Formula to Computing Areas . . . 189
§ 4. Conditions for a Line Integral of the Second Type lielng Path-
Independent . Integrating Total Differentials ................................ 190
1. Statement of the P rob lem ................................ . . . 190
2. The Case of a Simply Connected D o m a in ............................ 190
3. Reconstructing a Function from Its Total Differential . . . 193
4. Line Integrals in a Multiply Connected Domain . . . . 195
 CONTENTS 11

CHAPTER 5. SURFACE IN T E G R A L S..................................... 199

§ I. Surface Integral of the First T y p e ............................................. 199
1. Definition of Surface Integral of a Scalar Function . . . 199
2. Reducing Surface Integral to Double Integral . . . . 200
3. Some Applications of Surface Integrals to Mechanics . . . 204
4. Surface Integral of a Vector Function. General Concept
of Surface Integral of the First T y p e ..................................... 205
§ 2. Surface Integral of the Second Type . . . 207
1. One-Sided and Two-Sided S u rfaces............................... 207
2. Definition of Surface Integral of tho Second Type . . . 211
3. Reducing Surface Integral of the Second Type to Double
Integral ......................................................................................... 215
§ 3. Ostrogradsky Theorem . . ............................................. 218
1. Derivation of Ostrogradsky T h eorem ..................................... 218
2. Application of Ostrogradsky Theorem to Evaluating Surface
Integrals. Expressing Volumo of a Space Figure in the Form
of a Surface Integral . ..................................... . . . . 222
§ 4. Stokes' T h e o r e m ........................................................ . . . . 224
1. Derivation of Stokes' F orm ula................................................. 224
2. Application of Stokes' Theorem to Investigating Line Integ­
rals in S p a c e ................................................................................. 227

CHAPTER 6. FIELD T H E O R Y ................................................................. 230

§ 1. Scalar F i e l d ....................................................................... . 230
1. Definition and Examples of Scalar Field . . . 230
2. Level Surfaces and Level L in e s ......................... . 231
3. Various Types of Symmetry of F i e l d ......................... 232
4. Directional D eriv a tiv e............................................. . . . . 233
5. Gradient ofScalar F i e l d ............................................................... 234

§ 2. Vector F i e l d .......................... ................................................................ 236

1. Definition and Examples of Vector F i e l d ............................. 23G
2. Vector Lines and Vector Surfaces........................................ 237
3. Types of Symmetry of Vector F i e l d ..................................... 238
4. Field of Gradients. PotentialF i e l d .................... . . . 238

§ 3 . Flux of Vector Field. D ive rgen ce ........................... . . . . 240

1. Flux of Vector Field Across a S u rfa ce................................. 240
2. Divergence ......................................................................... 242
3. Physical Meaning of Divergence for Various Types of Field.
Examples . . . . 244
4. Solenoidal Field . . 246
12 CONTENTS

5. Equation of C o n t in u it y ............................................................ 248

6. Plane Fluid Flow. Ostrogradsky Theorem for Plane Field 249

§ 4. Circulation. R o t a t i o n ....................................................................... 250

1. Circulation of Vector F ie ld .................................................... 250
2. Rotation of Vector Field. Stokes’ Formula in Vector Notation 251
3. Symbolic Formula for R o ta tio n ................................................ 252
4. Physical Meaning of R o ta tio n ................................................ 253
5. More on Potential and Solonoidal F ie ld s ........................ 256

§ 5. Hamiltonian O p e r a t o r .......................................................................... 257

1. Symbolic Vector V . . . . . . . 257
2. Operations with Vector V . . . . . . 25$

§ 6. Repeated Operations Involving V. Laplacian Operator . . . 261

1. Repeated Differentiation............................................................ 2G1
2. Heat Conductivity E q u a tio n ................................................ 264
3. Stationary Distribution of Temperature. Harmonic Functions 205

§ 7. Expressing Field Operations in Curvilinear Orthogonal Coordinates 267

1. Statement oT the P rob lem .......................................................... 207
2. Curvilinear OrthogonalCoordinates inSpace . . . 267
3. Cylindrical andSpherical Coordinates . . . . . . 270
4. G r a d ie n t.......................................................................... . . . 271
5. D iv e r g e n c e ......................................................................... 271
G. R o t a t io n ..................................................................... 272
7. Laplace’s Operator .................................................................... 274
8. Basic Field Operations in Cylindrical and Spherical Coordi­
nates .................................................................................. •. 274

§ 8. Variable Fields in Continuous Media ..... ............................... 275

1. Partiul and Total Time D erivatives..................................... 275
2. Eulerian Equations of Motion of Ideal L iq u id .................... 277
3. Derivative with Respect to Time of an Integral Over a Fluid
V o lu m e............................................................................................ 279
4. Application to Deriving Equation of C on tin uity................ 282

CHAPTER 7. TEN SO R S.................................................................................... 283

§ 7. O r t h o g o n a l Affme. Tensor .................................................. 284
1. Transformation of Orthonormal B a s e s .................................... 284
2. Definition of Orthogonal Affine T en so r................................. 286

§ 2. Connection Between Tensors of Second Rank and Linear Operators 288

1. Linear Operator as a Tensor of Second R a n k .................... 288
2. Tencor of Second Rank as a Linear Operator........................ 290
 C O NT EN T S \:\
§ 3. Connection Between Tensors and Invariant Multilinear Forms 291
1. Tensors of Rank One and Invariant Linear Forms . . 2 9 1
2. Tensors of Rank Two and Invariant Bilinear Forms . . 292
3* Tensors of Arbitrary Rank p and Invariant MultilinearForms 294
§ 4. Strain T e n s o r ........................................................ 295
§ 5. Stress T e n s o r ............................
. 296
1, Definition of Stress T e n so r ......................................................... 296
2. Stress Tensor as a Linear Operator . . 298
§ 6. Algebraic Operations on T e n s o r s ..................................................... 300
1. Addition, Subtraction and Multiplication of Tensors . . . 300
2. Multiplying Tensor by Vector . 301
3- C ontraction........................... 302
4. interchanging Indices ..................................................... 302
5. Resolution of Tensor of Second Rank into Symmetric and
Antisymmetric P a r ts ......................................................... 302
§ 7. Tensor of Relative D isplacem ents ................................................. 303
§ 8. Tensor F i e l d ........................................................................ . . 305
1. Tensor Field. Divergence of T en so r................ 305
2. Ostrogradsky Theorem for Tensor Field . . 307
3. Equations of Motion of a Continuous M ed iu m ........ 308
§ 9. Principal Axes of Symmetric Tensor of Second R a n k ................ 309
§ 10. General T e n so rs ........................................................................................ 3 L1
t. Reciprocal Bases ................................................................ 311
2. Covariaut and CouLravariant Components of Vector. . . 312
3. Summation Convention.......................................................... 312
4. Transformation of Base V ec to rs.......................................... 313
5. Transformation of Covariant and Contravariant Components
of V e c to r ........................................................................................ 313
6. General Definition of T e n so r ................................................. 314
7. Operations on T en sors..................................... 316
8. Some Further Generalizations................................. 316

Appendix to Chapter 7. On Multiplication of M a t r i c e s ................. 317

CHAPTER 8. FUNCTIONAL SEQUENCES AND SEMES . . . 319
§ 1. Uniform Convergence. Tests for Uniform Convergence . . . 319
1. Convergence and Uniform Convergence.................................... 319
2. Jests for Uniform Convergence................................................. 325
2. Properties of Uniformly Convergent Functional Sequences and
Series . . . . . . . . 331
14 CONTEXTS

1. Continuity and Uniform Convergence.................................... 331

2. Passage to Limit Under the Sign of Integration and Tcrnnvise
Integration of a S e r ie s................................................................ 334
3. Passage to Limit Under the Sign of Differentiation and
Termwise Differentiation of a S e r ie s .................................... 337
4. Term-by-Term Passage to Limit in Functional Sequences
and Series . . . .................................... 339
§ 3. Power S e r i e s ................................................................... ... 341
1. Interval of Convergence of Power Series. Radius of Conver­
gence ................................................................................................ 341
2. On Uniform Convergence of a Power Series and Continuity
of Its S u m ........................................................................ 348
3. Differentiation and Integration of Power Series . . 351
4. Arithmetical Operations on Power S e r ie s ................ 352
§ 4. Expanding Functions in Power S e r i e s ............................ 354
1. Key Theorems on Expanding Functions in Power Series.
Expanding Elementary F u nction s................ 354
2. Some Applications of Power Series . . . . . . . 359
§ 5. Power Series in Complex Argument . ................ 362
§ 6. Convergence in the Mean . . . ................................ 366
1. Mean Square Deviation and Convergence in the Mean . 3GC
2. Cauchy-Bimyakovsky Inequality ............................................ 367
3. Integration of Sequences and Scries Convergent in the Mean 369
4. Connection Between Convergence in the Mean and Term-by-
Tcrm Differentiation of Sequences and S c r ie s .................... 371
5. Connection Between Convergence in the Mean and Other
Types of C onvergence.......................................... .. 372

Appendix J to Chapter 8. Criterion for Compactness of a Family of

F u n c t i o n s ............................................................................................ 374
Appendix 2 to Chapter 8. Weak Convergence and Delta Function 378
CHAPTER 9. IMPROPER INTEGRALS ............................ 383
§ 1. Integrals with Infinite Limits of I n t e g r a t i o n ................................ 383
1. Definitions. E xam p les........................................ ... 383

2. Reducing Improper Integral of the Form \ / (x) dx to Numc-

a
rical Sequence and Numerical S e r ie s .................................... 3SG
3. Cauchy Criterion for Improper In tegrals................................ 389
4. Absolute Convergence. Tests for Absolute Convergence 390
5. Conditional Convergence................................ . . . 397
 CONTENTS lo-

fi. Extending Methods of Evaluating Integrals to the Case of

Improper Integrals .................................................................... 309
^ 2. Integrals of Unbounded Functions with Finite and Infinite Limits
of Integration ..................................................................................... 400
§ 3. Cauchy's Principal Value of a Divergent Improper Integral 408
§ 4. Improper Multiple I n t e g r a l s ........................... 411
1. Integral of an Unbounded Function Over a Finite Domain 412
2. Integrals of Nonnegative Functions . . . . . . . 414
3. Absolute C o n v erg en ce.................................... 417
4. Tests for Absolute Convergence................................................. 419
5. Equivalence of Convergence and Absolute Convergence in
the Case of Improper Multiple In teg ra l................................. 421
6. Improper Integrals with Infinite Domain of Integration 424
7. Methods of Computing Improper Multiple Integrals . . . 423
CHAPTER 10. INTEGRALS DEPENDENT ON PARAMETER . . 427
§ 1. Proper and Simplest Improper Integrals Dependent on Parameter 427
1. Proper Integrals Dependent on Param eter......................... 427
2. Simplest Improper Integrals Dependent on Parameter . . . 432
.«? 2. Improper Integrals Dependent on P a r a m e t e r ................ 435
1. Uniform C o n v erg e n c e................................................................. 436
2. Reducing Improper Integral Dependent on Parameter to
a Functional S eq u en ce............................................................ 438
3. Propeities of Uniformly Convergent Improper Integrals De­
pendent on Param eter................................................................. 441
4. Tests for Uniform Convergence of Improppr Integrals Depen­
dent on Param eter......................................................................... 448
5. Examples of Evaluating Improper Integrals Dependent, on
Parameter hy Means of Differentiation and Integration with
Respect to P a r a m e te r ................................................................ 453
§ 3. Euler's I n t e g r a l s ...................................................................
4G0
1. Properties of Gamma F u n ction ....................................................... 460
2. Properties of Rota F u n ction ..................................................... 464
§ 4 . Multiple Integrals Dependent on P a r a m e t e r ................................. 468
CHAPTER 11. FOURIER SERIES AND FOURIER INTEGRAL 476
.«y /. Properties of Periodic Functions . Statement of the Key Problem 476
1. Periods of a Periodic F u n ctio n ....................... 476
2. Periodic Extension of a Nonperiodic Function . . . 477
3. Integral of a Periodic F u n ction ............................ 478
4. Arithmetical Operations on Periodic Functions . . 478
Ill CO NTENTS

5. Superposition ol Harmonies with Multiple Frequencies 479

6. Statement ol the Key P rob lem ................................................ 480
7. Orthogonality of Trigonometric System. Fourier Coefficients
and Fourier S c r ie s ........................................................................ 3S1
8. Expanding Even and Odd Functions in Fourier Series . . . 484
9. Expanding Functions in Fourier Scries on the Interval
I—n, J i l .................................................... . . . . 485
,\v 2. Fundamental Theorem on Convergence of Fourier Series . . . 486
1. Class of Piecewise Smooth F u n ction s........................................ 48G
2. Formulation of Fundamental Theorem on Convergence of
Fourier S c r ie s ................................................................................ 488
3. Key Lemma .................................................... 4S8
4. Proof of Convergence T heorem ................................................ 490
5. E x a m p les........................................................................................ 495
C. Fourier Sine and Cosine Series for Functions Defined on
Interval (0, /] . . 498
£ 7. Fourier Series with Respect to General Orthogonal Systems. Bes­
sel's I n e q u a l i t y ........................................ . . . . . . 501
1. Orthogonal Systems oT F u nction s............................................ 501
2. Fourier Coefficients and Fourier Series of a Function / (j )
with Respect to an Orthogonal S y s te m ................................ 503
3. Least Square Deviation. IVessel's In eq u ality........................ 504

/. Speed of Convergence of Fourier Series. Acceleration of Conver­

gence of Fourier S c r i e s .................... . . . SOS
1. Conditions for Uniform Convergence of Fourier Series . . . 508
2. Connor!ion Retween the Degree of Smoothness of a Function
and the Speed of Convergence of Its Fourier Scries . . . 512
3. Acceleration of Convergence of Fourier S e r ie s .................... 516

# 5. Uniform Approximation of Continuous Function by Trigono­

metric and Algebraic Polynomials. IVeierstrass' Approximation
Theorems ........................................................................................ 5t8
£ C. Complete and ClosedJjOrthogonal S y s t e m s .................................... 523
1. Complete Orthogonal S y s te m .................................................... 524
2. Parscval Relation as a Necessary and Sufficient Condition for
an Orthogonal System Being C om plete.................................... 524
3. Properties of Complete S y ste m s................................................ 525
4. Completeness of Trigonometi icS y s te m ...................................... 527
5. Completeness of Some Other Classical Orthogonal Systems 530

xv 7. Fourier Series in Orthogonal Systems of Complex Functions . . . 531

<■' S. Fourier Series for Functions of Several Independent Variables 535

 CONTENTS 17

£ fL Fourier I n t e g r a l ................................................................................. 538

1. Formal Derivation of Fourier integralF orm u la.................... 538
2. Proof of Fourier Integral T heorem ......................................... 540
3. Fourier Integral as an Expansion intoa Sumof Harmonics 344
4. Fourier Integral in Complex F o r m ............................................. 343
3. Fourier Transformation.......................... 54G
(1. Fourier Integral for Functions of Several Independent Varia­
bles ............................................................................................. 350
Appendix 1 io Chapter 11. On Legendre's P o l y n o m i a l s ..................... 5aG

.t ppi'itdix 2 to Chapter 11. Orthogonality with Weight Function and

Orthogoi.alization Proteus ................................................................. o5S

Appendix 7 to Chapter 11. Functional Space and Geometric Analogy 5H;»

Appendix ! to Chapter 11. Some Applications v{ Fourier Transforms 5tiS

Appendix o to Chapter 11. Expanding Delta Function in Fourier Series

and Fourier I n t e g r a l ......................................................................... 574

Appendix C /'* Chapter 11. Uniform A pproximation of Functions with

P o l y n o m i a l s ......................................................................................... 57»;

Appemiic 7 to Chapter 11. On Stable Summation of Fourier Series

with P a t min'd Coefficients ................................................................. 581
SUPPLEMENT 1. A S Y M P T O T I C EXPANSIONS ................................. SSCi
1. Examples of Asymptotic E x p a n sio n s .................... 580
1. mplolic Expulsions in the Neit>liimurliu«jd oi the Origin .>80
2. Asymptotic Expansions in tlie Neighbourhood of the Point
at I n f i n i t y ............................................................ . T>87
2. General Definitions and Theorem s ........................ 590
1. Order of Smallness. Asymptotic Kipiivalence . . . . 590
2. Asymptotic Expansions of F u nction s.................................... 502
3. Laplace Method for Deriving Asymptotic Expansions of Some
I ntegrals ............................................................................................. 598
SUPPLEMENT 2. ON UNIVERSAL DIGITAL COMPUTERS . C03
§ 1. Computers ............................................................ (503
t. Introduction .......................................................................... <503
2. Basic. Types ofC om puter............................................................... (503
3. Priii' ipal Components of a Computer ami Their Functions <504
Number Systems Usedin Computers........................................... (506
5. H-pre^nO in*; Numbers Within a C om puter............................ (508
J-os2t
18 contents

§ H. Basic Operations Executed by a Computer . I instructions . . . . 00$

1. Types of O p e r a tio n ........................................................ 008
2. Arithmetical O p e r a tio n s.................................................... 009
3. Additional ComputationalO p e r a tio n s...........................................010
4. Logical O perations........................................................................ 010
5. Input and Output O perations.................................................... Oil
0. Transfer of C on trol........................................................................ 012
7. Realization of OperationsWithina C oinpuler..................... 013
§ 3. Elements of P r o g r a m m in g ................................................... - Oil
1. General N o tio n s............................................................................ 014
2. Formula Programming................................................................ 015
3. Cyclic Processes............................................................................ 017
4. Flow-Chart. S u b r o u tin e s............................................................ 021
5. Instruction Codes. Operations onInstructions ................... 021
0. Automatic Programming............................................................ 023
$ 4. Organization of Computer W o r k ...........................................
. 024
1. Conditions for Effective Use of a Com puter........................ 024
2. Basic Stages of Solving a Problem on a Computer 024
3. Checking Computer Operation. Error Detection . . 025
B i b l i o g r a p h y ....................................................................................... 027
Name I n d e x ................................................................................................ 028
Subject I n d e x ................................................................................................ 029
1
Double
Integrals

The definite integral

b
J 1 (x) dx
a
is connected with the problems of determining the distance passed
over for a given speed, computing the area of a curvilinear trapezoid
etc. There are many similar problems involving functions dependent,
not on one but on many arguments. A typical problem of this kind
is to find the volume of a curvilinear cylinder (which is a three-
dimensional analogue of a curvilinear trapezoid).
By a curvilinear cylinder (cylindroid), with base F lying in the
x, [/-plane, we understand a solid T bounded by the base, a surface
z = / (a:, y) and the lateral cylindrical surface (Fig. 1.1). It seems

z z=f(T,y)

Fig. 1.1

natural to evaluate the volume of such a solid in the following way.

Divide the base F by a net of curves into elements, cells, Ff. This
results in breaking up the entire cylinder T into elementary cylin­
ders Ti whose bases are the cells /'V It is clear that the volume of the
cylinder T should be understood to be equal to the sum of volumes
of the elementary cylinders 7V
To find the volume of an elementary cylinder T,, we choose a point
(h, Vi) in Ff and replace the elementary cylinder Ti h a v in g curvi-
2*
20 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

linear upper base by a right, cylinder with constant altitude / (Ef, rp)
and the same lower base Ft. In other words, we consider the volume
of the elementary cylinder 7\ to be approximately equal to
/ (£,, r|i) ASt
where AS{ is the area of the element /p. Xow we take, as an appro­
ximate value of the volume of the whole cylinder 7\ the sum

tnh,r\i)*S, (1 .1)
i—1
extended over all the cells the base I' is divided into. It is intuitively
dear that sum (1.1) represents tlic volume of the cylinder T with
degree of accuracy increasing as the sizes of the cells /*’, are dimi­
nished. To obtain the precise value of the volume we must pass
to the limit in expression (1.1) by making the sizes of the elements b\
tend to zero.
'This passage to the limit leads to the notion of an integral of
a function / (x , i/) of two independent variables, i.e. to the double
integral.
There is an obvious analogy between the above (heuristic) conside­
rations concerning the double integral and the construction of the
definite integral of a function of one argument on an interval. The
only distinction between them is that in the former we consider func­
tions dependent not on one but on two arguments and that instead
of the lengths of suhintervaJs ,\j*f we lake the areas of the cell* Ff
into which the figure /'\ the base of the cylinder, is divided.
Besides Ilie problem of computing the volume of a curvilinear
cylinder there are many other problems connected with the concept
of the double integral. Some of them will be discussed in § 4 of this
clia pier.
Some physical and geometrical problems lead to the concept of
an integral of a function of three and more independent variables.
The next chapter is devoted to these integrals.
The problem of evaluating the volume of a curvilinear cylinder
indicates that the notion of a double integral is closely related to
the concept of area of a curvilinear plane figure because expres­
sion (11) involves the areas ASi of curvilinear plane elements b\
into which the base of tho cylinder has been broken up. Therefore,
although we suppose the reader to be familiar with the concept of
area*, we begin this chapter with a brief discussion of the basic
properties of area.

* E.g. see |Sl, Chapter It, § 2.

 CH. i. DO UBLE INTEGRALS 21

§ 1. AUXILIARY NOTIONS. AREA OF A l ’LAXE FIGURE

1. Interior ami Boundary Points. Domain. We arc going to remind
the reader of some notions which we shall need in what follows.
Let a be a point of the x. {/-plane. An open circle of radius e with
centre at the point a* is referred to as an c-neighbourhood, or
simply a neighbourhood, of the point. A point a of a given set A
is said to he its interior point if a “sufficiently small” e-neighbour­
hood of (lie point a entirely consists of points belonging to the set A.
A set whose all points are interior is called an open set. An open

set G is said to he (arewise) connected if each pair of its points can

he joined by a continuous curve entirely lying within G. An open
connected set is briefly referred to as a domain.
For instance, the collection of points whose coordinates satisfy
the inequality xl •*- y2 <Z 1 is a domain (see Fig. 1.2a). The set
consisting of the two circles xl T !/2 < 1 and (.r — 2)2 4 !/~ <Z 1 -

is not a domain because though it is open it is not connected (see

Fig. 126).
A point a is called a boundary point of the set A if its every neigh­
bourhood contains both points belonging and not belonging to A.
A boundary point itself may or may not belong to A. In particular,
an open set contains none of its boundary points. The collection
of all the boundary points of a set is called its boundary. A set con­
taining all its boundary points is called closed. Every set can be
turned into a closed set (called its closure) by adding all the boun­
dary points to it.** In particular, when adding to a domain G all
its boundary points we arrive at a set referred to as a closed domain.
A point a is called a limit point of a set A if in A there exists
an infinite sequence of pairwise distinct points ftt. ...................... . .
convergent to a. A limit point of a sot A may or may not be contained
in *1. A set contains all its boundary points if and only if it is closed.
(Prove it.)

* That is the totality of all points of the plane whose distances from a
arc less than s.
** An arbitrary sot may he, of course, neither open nor closed. The collec­
tion of all interior points of a set is referred to as its interior. — Tr.
22 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

We say that a set is bounded if it can be placed within a circie

of sufficiently largo radius. Let A be a bounded set. Denote by
P (®i» ^2) the distance between its two arbitrary points. Let now
a, and a2 independently run over the whole set A . The set of numbers
P (<2|t &z) is then obviously bounded above (because p (a,, a2) cannot
exceed the diameter of the circle in which A is contained). The least
upper bound of the set of numbers o (a,t a2) is called the diameter
d (A) of the set -1 (Fig. 1.3).

If a set A is a part of a set B or coincides with it (i.e. A is a subset

of B) we shall designate this fact, as usual, by the symbolic rela­
tion .4 n. B . If a point a belongs to a set A we write a £ A.
The union of two sets A and B , i.e. the collection of all points
belonging at least to one of the sets, will be denoted as A 4- B,
and the intersection (or product, or meet) of two sets A and B
which is the collection of all points simultaneously belonging to A
and B will he designated by A B .
2. Distance Between Two Sets. Let us introduce another notion which
will be applied to the proof of the theorem on the existence of a double
integral.
Let A and B be two arbitrary sets in the plane. We shall call the
number
p (A, B) - inf p (a, b) (1.2)
the distance between the sets A and B . In (1.2) the greatest lower
bound is taken with respect to all the pairs a £ A, b £ B . We clearly
have p (.4, B) — 0 if A and B have at least one point in common.
The converse does not hold in the gcuieral case; for instance, the
distance between the hyperbola y = -^-and the x-axis equals zero
although these two lines have no common points at all. At the same
time the following theorem (which will be needed in § 2) holds:
T h e o r e m 1.1 {On S e p a r a b i l i t y o f C lo s e d Sets). If P
and Q are two bounded closed sets with no points in common then
P (/>, Q) > 0 .
Proof. Assume the contrary, i.e. lot p (P, Q) = 0. Then, by the
definition of the distance between two sets, for each n — 1 ,2 , . . .
there are points pn £ P and gn £ (J such that

7r>) <~J- (1.3)

 C H . 1. D O UBL E INTEGRALS 23

But, {pn} being o bounded infinite sequence, we can choose, according

to the well known Bolzano-Weierstrass theorem (e.g. see [81, Chap-
ter 14, § 2), a subsequence
Pnj » P n „ » * • • > Pn^ » ♦ • •

of {pn} convergent to a point p 0. Then the corresponding points

of the sequence {?„} form a subsequence convergent, by (1.3), to the

same point p 0.
The point p 0 is sure to belong to the set P. In fact, there are two
possibilities here. Either the subsequence contains an infini­
tude of distinct points, and then p0 is a lim it point of P and p 0 6 P
because P is closed, or the subsequence {/>nft } is stabilized in the

Fig. 1.4

sense that all its points from some number onwards coincide, and
then they apparently coincide with n0 ar<l again p 0 £ P. By the
same argument, p 0 £ <7. But then P and Q have a common point,
which contradicts the hypothesis.
Exercise. Show that the theorem remains true when at least one
of the two closed sets P and Q is bounded.
3. Area of a Plane Figure. The concept of area of a polygonal
figure is well known from elementary geometry. (By a polygonal
figure we mean a set constituted by a finite number of bounded
polygons, see Fig. 1.4.) The area of a polygonal figure is a nonnega­
tive*' number possessing the following properties:
1 (monotonicity). If P and Q are two polygonal figures and P en­
tirely lies inside Q wo have
area of P ^ area of Q
2 (additivity). If P\ and P 2 are two polygonal figures without
common interior points and P t 4- P 2 is the union of the figures

* It can be equal to zero only if the polygonal figure degenerates into a

finite number of points or line segments.
24 M U L T I P L E IN T E G R A L S , FIELD THEORY AND S E R I E S

we have
area of (P^ -f P 2) = area of P j -f- area of P z*
3 (invariance). If two polygonal figures P t and P 2 are congruent
area of Pi* = area of Pn
Let us now extend the concept of area, preserving the throe pro­
perties. from polygonal figures to a wider class of plane figures.
This problem is solved as follows.
Let F he a plane figure**. Wo shall consider all the possible poly­
gonal figures P entirely lying inside F and the polygonal figures Q
entirely containing F. The former will he referred to as embedded
figures and the latter as enveloping ones. The areas of embedded
figures are bounded above (for instance, by the area of any enveloping
figure) and the areas of enveloping figures arc bounded below (e.g. by
the number zero). Therefore the set of areas of all polygonal figures
embedded in the figure F possesses the least upper bound***
S+ — S+ (F) — sup (area of P)
Per
and the set of areas of all enveloping figures possesses the greatest
lower bound
S* S* (F) — inf (area of Q)
Q^r
The quantity S„ is known as the interior (Jordan****) content,
of the figure F and S* as its exterior (Jordan) content. The .area
of any embedded figure not exceeding the area of any enveloping
figure, we have
S + < .V*
If S+ = S* — S their common value S is simply called the area
(the Jordan content) of the figure F. In this case the figure F is said
to be squarable.

* We can easily verify that the requirements 1 and 2 arc not independent
because mnnotonirily of area is implied by ils nonnogalivity and additivity.
Indeed, if a polygon.il figure P 1ms inside a polygonal figure Q we can represent
(J as the union of P and a polygonal figure which can be n a t u r a l l y called
the difference between the sets Q and P and designated as Q — P. Then, bv
additivity, we have area of Q ~ area of P ’ area of (Q — /J).“ fhit area of
(Q — P) ^ 0 and therefore area of Q ^ area of P .
** I.e. a bounded set of points in the plane.
*** If it is impossible to place any polygonal figure within the figure
F we put, by definition. S* - 0.
***« .Iordan, Camille (1838-1922), a French mathematician.
 CH 1 DO UBLE INTEGRALS 25

Thus, we have extended the concept of area from polygons to

a wider class of figures.* The retention of the basic properties
of area (i.e. additivity, monolonicity and invariance) will be proved
in Sec. 4.
We now establish the following necessary and sufficient condi­
tion for a figure being squarablc which will be of use for our further
aims.
T h e o r e m /.? . A figure F is squarable if and only if for every
e > 0 there exist two polygonal figures P d F and Q i d / such that
area of Q — area of P e (1.4)
Proof. In fact, if such figures exist it follows from
area of P ^ S+ ^ S* ^ area ol‘ Q
that
0 < S* — S+ < e
and therefore, since e > 0 is chosen arbitrarily, wc have S * [ = S+.
Conversely, if then, by definition, for any given f !> 0,
there is an embedded polygonal figure P and an enveloping figure
(J such that
S ¥— ^-<aren of P ^ S 09 £*■<area of O < S* ~~4r
which implies
area of Q — area of P <C e
The collrclion of the points belonging to () and, simultaneously,
not belonging to P is a polygonal figure of area (area of (J — area of P)

containing thc‘ boundary of the figure F. Consequently, the condition

of Theorem 1.2 implies that F is squarablo if and only if its boundary
can he embedded in a polygonal figure of an arbitrarily small area
(Tig. 1.5).

* Every polygonal figure is obviously squarablo. and the new definition

of area (introduced with the help of S t ami 5*) yields the original value of
its area.
26 MULTIPLE INTEGRALS, FIELD THEORY A N D jSERIES

The theorem enables us to establish the squarability of some

figures distinct from polygonal ones, for instance, the squarabililv
of the circle. For a circle we can take, as P and Q, a regular inscribed
and a regular circumscribed polygon with a sufficiently large number
of sides.
By the way, the derivation of the formula for the area of a circle
usually performed in elementary courses of geometry is based on the
same arguments which are given here in the general form.
Let us introduce the following terminology. We shall say that
a set and, in particular, a curve, is of area zero if it can be embedded
in a polygonal figure of an arbitrarily small area. This enables us
to rephrase Theorem 1.2 as follows:
T h e o r e m J .2 '. For a figure F to be squarable, it is necessary and
sufficient that its boundary be of area zero,
Based on the theorem, we now describe a sufficiently wide class
of squarable figures to which we shall restrict ourselves in our further
considerations.
Taiinn i a. Each r e c t i f i a b l e c u r v e * has zero area•
Proof. Let L be a rectifiable curve of length /. Divide the curve
into parts by n 1 points so that the length of each part is

less l linn —(of course, this is always possible) and construct a square
of side — n with centre at the kth point of division for each k —
— 1 ,2 ,. . ., n + 1 (see Fig. l.(>). The union of the squares is a poly­
gonal figure enveloping the curve L and the area of the polygonal
figure does not exceed the sum of the areas of the constituent squares,
i.e. is not greater than -r !)• Since I is fixed and n can be

* A rectifiable curve is the one that possesses a finite length. As is well

known (e g see |81, Chapter 11, § t), if a curve can be represented by parametric
equations of the form
x = cp (0, y — (0i ct ^ t ^ fi
where </) and if (f) are continuous functions with continuous (or piecewise
eontiiuiou.o derivatives it is recti liable.
 CH. 1. D O U B L E IN T E G R A L S 27

taken ns large as desired the curve L can be actually embedded

in a figure of an arbitrarily small area. The lemma has been proved.
From the lemma and Theorem 1.2' we conclude:
Every plane figure (i.e. a bounded plane set) whose boundary is
composed of one or several rectifiable curves is squarable.
It is this class of figures that, as a rule, we shall consider in what
follows.
Note. We can also point out another class of squarable plane
figures. Any curve which can be represented by an equation of the
form
y — f (x), a x < b
where / (r) is a continuous function or by an equation x — g (;/),
c ^ y ^ d where g (y) is also continuous is of zero area. (The proof
of this fact can be found, for instance, in 181, Chapter 11.) It follows,
by Theorem 1.2', that every figure with a boundary representable
as a union of finite number of continuous curves defined by equations
of the form y = f (x) or x = g (y) is squarable.
4. Basic Properties of Area. Now we are going to show that the
definition of the area of a plane figure thus introduced possesses the
properties of monotonicity, additivity and invariance.
Monotonicity is directly implied by the definition of area, and

Fig. 1.7

the proof of the property is left to the reader. Let ns establish

addilivilv, i.e. prove the following assertion:
(1) Let Fi and F2 be two squarable figures with no interior points
in common and F be their union (see Fig. 1.7); then F is also squarable
and
area of / ’ = area of F\ -f- area of F2 (1.5)
The squarability of the figure F follows from Theorem 1.2' and the
fact that the boundary of F is composed of sets of area zero which
are some parls of the union of the boundaries of the squarable figu­
res Fi and F%* Therefore, to complete the proof, we must only
deduce equality (1.5). To this end consider polygonal figures P t
and P2 embedded in P t and F2 and polygonal figures (?, and Q2
enveloping Ft and Fz, respectively. Since the figures P t and P2

* It appears obvious that every part of a set having zero area is a sot of
area zero.
28 MULTIPLE INTEGRALS, FI EL D T H E O R Y AND S E R I E S

do not intersect. the area of llie polygonal figure composed of P f

and P 2 equals area of P j — area of P z. The figures and Qz (which
may intersect) constitute their union Q whose area does not exceed
area of (7, t area of (J*. Thus, we have
area of P — area of P x 4- area oi P z ^. area of F ^
^ area of Q ^ area of Qx -r area of Qz
an d
area of P x -J- area of P z ^ area of Fx -f area of F2 ^
^ area of Qx -- area of Qz
Since the differences (area of Qx — area of P x) and (area of Qz —
— area of /*») can he made arbitrarily small it follows that equality
(1.5) holds. Additivity is thus proved.
Finally, the property of invariance of area immediately follows
from the invariance of the area of polygonal figures and from the

way the area of squaralde figures lias been introduced by means of

the areas of polygonal figures.
There is another property of the squarnble figures:
(2) The intersection of two squarahle. figures is <i squarahle figure.
Indeed, if Fj and Fz are two squarahle figures and F is their inter­
section (Fig. 1.8). each boundary point of F is a boundary point
at least of one of the figures Fxand Therefore our assert ion follow®
from Theorem 1.2' and the fact that the area of a union of sets of
zero area is equal to zero.
5. The Concept of Measure. As lias been mentioned above the
concept of area has been introduced here according to Jordan’s idea.
But this way of defining the measure of a set possesses certain disad­
vantages. Indeed, as has been shown, the union of two squarnble
figures is squarnble. This immediately implies that the union of any
finite mimher of squarahle figures is again a squarahle figure. But if
we take an infinite sequence of squarahle figures
/•V F: ni
their union may not he squarahle. Here is an example. Take the
square
it
in the x, //-plane and consider the collection of all its interior points
with rational coordinates. It can he easily shown that these points
 CH. 1. D O U B L E IXT EG H A LS 29

form a countable set, i.e. they can be arranged into a sequence

P\ = (^-li £/1)■» Pz = (**-2i Pz)* ■ • Pn = U n i [/n)» • • •
Now take a number e >* 0 and construct a closed circle of radius
j:
r, <C — with centre at the point p x lying within the square. Further.
take first of the points p2. />3, . . . which falls outside the circle and
construct- a new closed circle of radius r2 <Z ^ lying inside the
square ami not intersecting the former circle. Next wo find the
first of the remaining points lying outside the circles thus constructed
and take it as the centre of a circle of radius r3 < — contained
in the square and not intersecting the circles constructed before.
Let us infinitely continue the process in this fashion. We thus obtain
an intiuiIe sequence of nonoverlapping circles placed in the square,
their union being everywhere dense, in it.* Wc can easily show that
the union of the circles is a figure F which is nonsquarable in the
sense of Jordan (let the reader prove* it). On the other baud, it
appears natural to attribute to this figure an area equal to the sum
of the areas of the circles it is formed of. This sum is obviously
equal to

Such difficulties can bo avoided by introducing a more flexible

and perfect, concept of a Lebesgnc measure** hut we cannot discuss
it al liiigSli here.

§ 2. DEFINITION AND BASIC PROPERTIES

OF DO fliLE INTEGRAL
1. Definition of Double Integral. Let us now pass to the main
object of this chapter, i.e. the notion of a double integral. Let G he
a sqtiaraIde figure and / (:r, if) be a bounded function defined in G.
Divide G into a finite number of nonintersecting squarable parts Gx
forming a partition {(/*} of the figure G. Consider a sum of the form

<S - S / (Si, 11/) AS, (1.6)

i I

* This moans that the union of the circles is a set whose closure coincides
with the entire square.
** I.ohosguo, Henri Leon (1875 1941), a prominent French mathematician,
one of the founders of modern theory of functions.
30 M U L T I P L E IN T E G R A L S , FIELD THEORY AND SERIES

where A£\- is the area of Gj and (I*. q,) is an arbitrary point belonging*
to <?,* Sums of form (1.0) will be referred to as integral sums (asso­
ciated with the function / (z. //) and the figure G). We introduce the
following definition of the limit of integral sums (1.6).
D e f i n i t i o n J. Let D be the maximal of the diameters d (Gj) of the
figures O’, (the quantity Z), the maximal diameter of the partition {Gt},
is called the f i n e n e s s o f the p a r t i t i o n ) . A number J is said to
be the l i m i t of i n te f/r u l s u m s (1-0) as D — 0 if for every e > 0
there is 6 0 such that
Ia J |< c (1.7)
when
/)< 5 (1.8)
In other w«»rd«. inequality (1.7) mn«l hold for all integral sum*
a corresponding to the partitions G = Gt -f- Gz -j- • • - — Gn which
satisfy condition (1.8) irrespective of the way the figure G is broken
up into parts Gj and of the particular choice of a point (|f, q*) in
each element of the partition.
D e f i n it ion ?. ff the limit
n
lim >] / (|{, in) ASj
D->0 i~ I
of integral sums (l.(») exists it is called the d o u b l e i n t i ’(jru7 of
file f n net ion f (z, //) o v e r the f U j u r e G and denoted by the
symbol
( / (*» y)ds or | j / (z, y) dx dy
'G 'G
In this case the function / (z, //), the integrand, is said to he inle-
grable on the figure G and G is called the domain of integration.
The expression / (z, y) ds or / (z, y) dx dy is referred to as an element
of integration.
The notion of a double integral is sometimes introduced in a diffe­
rent manner. A figure G taken from a chosen class of figures is broken
into rectangular cells by means of straight lines parallel to tlie
coordinate axes (see Fig. 1.9). In each cell a point (If, i\ j ) is then
chosen and the sum a = y f (|j, q,-) A/>j is formed. The sum is
taken, say, over all the colls entirely lying within G disregarding
those adjoining the boundary of G (the total area of the latter is
small). Then the passage to the limit is performed as the maximal
diameter of the cells lends to zero. The imperfection of such a delini-
 CII. i. D O U BL E I N T E G H A L S 31

lion is that it is connected with a certain coordinate system in the

plane whereas it is intuitively clear that the integral / (.t, y) ds,
i.e. the volume of the corresponding cylindrical solid, must be inde­
pendent of the choice of the coordinate system. When the notion of
a double integral is introduced by means of such rectangular cells
the above fact should be additionally proved but our definition
implies it automatically. The definition given here has some other

Fig. 1.9 Fig. 1.10

advantages. Let, for instance, a function / (x , y) assume on G only

two values: </, and a 2 (Fig. J.10). If the parts Gx and G2 on which
/ (x, y) is equal to a x and rt2, respectively, are squaral.de our defini­
tion makes it possible to evaluate the integral / (2 , \j) ds without
passing to Ihe limit. Intuitively, it is apparent that

j J1 (*■ u ) * = (» * * » ° f G >) •«< + (arca of ^ 2 ) *a2

G

(prove it). But the definition based on forming rectangular cells

would need a sophisticated passage to the limit even in this simple
case.
At the same time it should bo noted that both definitions result
in the same notion of a double integral.
2. Conditions for existence of Double Integral. Upper and Lower
Darboux Sums. Let us iind out what requirements should be imposed
mi a function f (x, y) defined over a squarable figure G in order to
guarantee the existence of the double integral
U / (x, //) ds
v•
G
32 M U L T I P L E IN T E G R A L S , FIELD THEORY AND S E R I E S

In introducing the definition of the double integral we have

supposed the corresponding function / (x, y) to be bounded.* At the
same time we cau easily construct examples indicating that
an arbitrary bounded function is by far not always integrablo.**
To establish the integrnbility conditions it is convenient, as in
the case of one independent variable, to use the so-called Darboux***
upper and lower sums.
T.et f (r, y) be a bounded function defined on a squarabl© figure Gy
and {G {} be a partition of the figure. Denote by and //?/ the least
upper and the greatest lower bounds of the values of / {xy y) on the
element G;. The sums

v \ f . \ Si and o>= V. miASi

»=i

are, respectively, referred to as the upper .and the lower Darboux

sums of the function / (xy y) (corresponding to the given partition
{C,} of the figure G). \Vc obviously have Q ^ <o for any partition
{Gi).
Let us enumerate the basic properties of the upper and lower sums.
(1) For every partition {G^} of the figure G, the corresponding
upper and lower sums are, respectively, the least upper hound
and the greatest, lower hound for the integral sums
n
y / f£(,
i-=l

* As is knmvn, a fuin't.it>u nf «me variable winch is (Kiomaim) inlegrahlo

on an interval is necessarily bounded (e.g. sec 18], Chapter 10). Ihit the argu­
ment applied to proving this fact cannot he completely extended to the case
of two arguments. Actually, when talcing different partitions of a squnrahle
figure (i into squarjlde elements Gj we cannot, in general, avoid the cases in
which some of the elements arc of area zero. Rut this means that the corres­
ponding integral sums ^ ip) A S * must not necessarily be unbounded for
each partition even if the function / (i, //) is not bounded (because the function
may turn out to be unhounded only on those elements of partition whose area
equals zero). This cannot be the case for a function of one variable when we
break up the interval of integration into nonoverlapping half segments. It is
possible to avoid the appearance of elements of area zero for functions of two
(or several variables) by restricting both the class of figures and the class of
partitions in question. Another way out (which we follow incur presentation
of the theory) is to completely exclude unbounded functions.
* ’* An example of a bounded but nonintcgrable function of two variables
is the one defined oil the square 0 ^ x < 1, 0 < j/ < 1 in the following way:
1 (x. i/) — 1 if x and y are rat ional nil tubers and / (r, y) — 0 if otherwise.
'Plie proof of the fact that, the function thus constructed is nonintcgrable is left
to the render.
*** Darboux, Jean C.aston. a French mathematician (1842-1917).
 C ll. 1. D O UBL E IN T EG R AL S 33

associated with the partition {6,-} (for all the possible ways of
choosing the points (I/, >p)). In particular, we always have

w = 2 m tb S t < 2 /(!i, 2 M t& Si = Q

i=i i-1 i“ 1
Indeed, the inequality

v / d ,, ni) & S i < 2

t=i
obviously holds for any choice of points (£j, q,-) on Gt {i — 1 , 2 , . . .
. . ., n). On the other hand, by the definition of the least upper
bound, for every f > 0 it is possible to take a point n,) in each
element G, of the partition {G*} so that M — / (t*, c-
ij,) <C C
(where S is the area of the domain 6’). liut thenwe have
n n n

S i - S X S , . t|i)AS, = 2 ( A /i- Z d ,, ti,))AA’, < - | - 2 A 5‘ = *

i —1 t=l t=l

An analogous argument applies in the case of a lower sum.

A partition {G}} will he referred to as a refinement of a partition
(G/} if each element Gk of the latter is either an element of the former
or a union of several (dements of the former partition. In these terms
we can formulate the following assertion:
(2) If Q and io are the upper and the lower sums corresponding
to a partition (GJ, and Q' and ♦./ are the upper and the lower sums
for a refinement {G)) of {Gj}, then
to to' ^ Q' Q
that is the upper Darboux sum does not increase and the lower one
does not decrease as the partition is refined.
Actually, let (Gj} be a partition of the figure G and {Gj} be its
refinement. Then each element G< of the partition {G^} is the union
of some elements Gla, a = 1, 2, . . ., k t of the latter partition.
Furthermore we have
fei
A S ,= y A S ; a (1.9)
a=l
M i > A/*a, a ~ 1, 2, .. ., hi (1.10)
each element G) being a constituent of only one element G/. It
follows that

s > - V .l/iAG e> V 2 Mi*bSin = <y

i-1 l-ln = l
We similarly prove the inequality ro e /.
3-0824
 34 MULTIPLE IN T EG R AL S, FIELD THEORY AND S E R I E S

(3) Let {<?!} and {G) } be two arbitrary partitions of the figure
G, and Q', to' and £2", co" be, respectively, the upper and the lower
sums associated with the partitions. Then we have
Q' ^ co" and Q" ^ (o'
i.e. every lower sum (corresponding to a given function / (x, y))
does not exceed any upper sum (corresponding to the same function).
To prove the property we first of all note that for any two partitions
of the same figure G there exists their “common refinement”, i.e. a par­
tition such that it serves as a refinement, of each of the two parti­
tions. For instance, to construct such a common refinement we can
take, as its elements, the intersections of elements Gl of one parti­
tion with elements G) of the other (of course, we only take those
elements G\ and GJ which have common points).
Now consider the upper and the lower sums corresponding to the
partitions {CD> {GJ} and to tlieir common refinement {&>,}. Denote
them , respectively, as Q', co'; Q", co" and Q, co. T h e n ,;b y proper­
ty (2),
Q' > Q, Q" > O
and
co/ <1 co, co" ^ co
Besides, we obviously have the inequality

Hence, wc have
il' ^ Q ^ »o ^ co *
and, similarly,
Q" > Q > co > o'
The assertion has thus been proved.
The collection of all upper sums corresponding to a given func­
tion / (x, y) is bounded below (e.g. an upper sum cannot be less than
any lower sum) and the collection of all lower sums is hounded above
(e.g. a lower sum cannot exceed any upper sum). Therefore the
totality of the upper sums possesses the greatest lower bound which
we designate* as J and the totality of the lower sums has its least
upper bound. J . The numbers J and J are, respectively, called the
upper and the lower (Darboux) integrals (corresponding to the
domain G and the function / (x. jy)).
The upper and the lower integrals satisfy the inequality
J tC J
 CH. 1. DOUBL E INTEGRALS 35

In fact, assume the contrary, i.e. / > J . Then there exists a number
e > 0 such that
J —J > 0 (1.11)
Furthermore, by the definition of the least upper and greatest lower
bounds, there is an upper sum Qj and a lower sum such that
&i — J < y and £—
that is
Q( — (02 t ( ^ — J ) ^ ®
Consequently, by (1.11), we have
Q, - (o2 < 0
which contradicts property (3).
Properties (l)-(3) of the upper and the lower sums enable us to
establish the following necessary and sufficient condition for the
intogrability of a function j (j:, y) which is completely analogous
to the corresponding necessary and sufficient condition for the
existence of the definite integral of a Function of one argument
(e.g. sec [81, Chapter 10, Theorem 10.1):
'Jltrarem /..V. A bounded function f (x, y) defined on a squarable
figure G is integrablc over G if and only if for every e > 0 there exists
a partition of the figure G such that, the Darboux sums associated with
the partition satisfy the condition Q — <o <C e.
The proof of the theorem is based on the following (Durboux)
lemma:
D a r b o n x L e m m a . The upper (lower) integral J (J) is the
limit of the upper (lower) Darboux sum as D 0 (where D is the maxi­
mal of the diameters d (Gf) of the elements Gi oj the partition {Gf} of
the figure G).
For convenience, we introduce the notion of the boundary of a par­
tition. If we are given a partition {G,} of a figure G into squarable
parts Gj Hie union L of the boundaries L t of all the elements Gi will
be referred to as the boundary of the partition {G,}, i.e.
L ~ 1j\ -f- 1*2 " h • • • “f~ / 'i i

The boundaries Li being of area zero for every partition of the fig­
ure G inlo squarable parts G,-, Ihe boundary L of the partition {G,}
lias a zero area as well.
The boundary L is Ihe union of a finite number of closed sets L t
and therefore it is also closed. (This is a general property of a union
of a finite number of closed sets. I.el the render prove it.)
3-
30 MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

We now proceed Lo prove Darboux’s Jemma.

The proof of Darboux's lemma. By the definition of the upper inte­
gral J , for every e >• U there is a partition {Gf} of the figure G such
that the corresponding upper sum £2* satisfies the condition
0 < U * —/ < - f -
Embed the boundary L* of llie partition in a polygonal figure Q
of area less than — where M = sup | f (x, //)|, so that L*
, (* . y) ZG
is strictly contained in it. The houndary L* and the boundary
of the polygonal figure (J are two bounded closed sets having no points
in common (sec Fig. 1.11). Consequently, by Theorem 1.1, the
distance between them is a positive quantity a . Now consider an
arbitrary partition {Gh} of the figure G for which D < a . There is

Fig. 1.11

an obvious property of the elements Gtl of the partition: if Gh and L*

have at least one common point, then Gu lies entirely in the interior
of the figure Q. Such elements Gh will be called boundary elements,
and all the other will be called interior elements. Let us show that,
to every partition {Git} with D <C «, Ihere corresponds an upper
sum Q which differs from J by less Ihuu e. To this end, divide the
sum Q into two groups of terms:

o - >: M * \ S k = >]' Mit&Sk + S ' M itiS i

where the sum m ation in V ' is extended over all interior elements
and V " is taken over all the boundary elements of the partition
{Gft}. Let us separately estimate each sum. Every interior element
of the partition is strictly contained in an element, of the partition
{G*}. The corresponding least upper bound Mh apparently does not
exceed the least upper bound of the values of the function / {.c, y)
assumed on this element of the partition {G*}. It follows that
vrh\S'b <Q *
Furthermore, we have the evident inequalities
| d/h j-< M - sup | / (x, y) | (for all ft)
(x. y\£G
 CII. 1. D O UBL E IN TE GR ALS 37

and
2 AS; < area of
Consequently,
S ' M'kASI
and hence,
Q = 2 ' . V i . + 2" ■< £J* + f c J + £
which is what we set out to prove. The lower sums are considered
in a similar way.
Finally, we pass to the proof of Theorem 1.3.
Necessity. Let / (x , y) he intcgrablc and an arbitrary e > 0 be
given. Denote the integral of / (r, y) by the symbol J . From the
definition of the limit of integral sums, for any given e there exists
6 > 0 such that for each partition {£*} with D <C 6 the inequality
n
ili) ASi | < £ (1.12)
1=1
holds irrespective of the choice of the points (^ . ijt). Wc also know
that the upper and the lower sums corresponding to the partition
{Gj} are the least upper and the greatest lower hounds of the integral
«ums associated with the partition. Therefore, wo can take a fixed
partition and choose the points (Ef, t]*) and (£T i]i) within the
elements Gt of the partition so that the following inequalities are
fulfilled:
n n
Q - S / ( U . ti;)A .S ,< £ -: iia A S i-< 0 < - (l.i:s>
1=1 1=1
Lach of the two integral sums satisfying condition (1.12), we deduce,
from (1.13), the desired result:
Q — t»> < e
Sufficiency. If for every e > 0 there exists a partition such that
Q — (♦) <C r
we obviously have
J = J
Denote the common value of the quantities J and J by J. Let us
show that J is the limit of integral sums. i.e. the double integral
of the function 1 (x . y) over the* domain G |ty P n rb o n v ’c lemma
•-J8 M UL T I P L E INTEGRALS, MELD THEORY AND S E R I E S

J is the common limit of the upper and lower sums for D -+■ 0.
But since the value of any integral sum associated with a partition
is contained between the corresponding Darboux sums Q and co the
number J is the limit of the integral sums as D — 0. The theorem
has been proved.
3. Some Important Classes of Intcgrablc Functions. Applying
Theorem 1.3, wo shall now establish the integrability of some impor­
tant classes of functions, and, first of all, continuous functions.
In wlint follows we shall regard each function in question as being
defined on a bounded closed squarable domain.
Theorem Every continuous function / (x, y) defined in
a bounded closed* domain G is integrable on G.
Proof. Since / (x, y) is continuous on a bounded closed set it is
bounded and uniformly continuous on it.** The uniform continuity
of the function / (x, y) implies that for every e *-> 0 there is 6 0
such lluil if the figure G is divided into parts G, whose diameters
are less than <5 the oscillation of the function f (x, y) on each of the
parts, i.e. the difference M t — is less than k. But then
n n n
Q - U ) = y ATtASi — V] niiASi < e 5 AS f = eS
i—1 »—1 i—
«1

and hence the function / (x, y) is integrable.

The condition of continuity of the integrand is too restrictive.
Therefore the theorem below guaranteeing the existence of the
double integral for a class of discontinuous functions is important
fur applications
Theorem If a function f (x, y) is bounded over a bounded
closed domain G and is continuous throughout G possibly except a set
of area. zero the function is integrable on G.
Proof. Take an arbitrary e 0. By the hypothesis, / (x, y) is
bounded, that is there exists a number K such that | / (x, y) J< K.
Let us embed the set on which the function / (x, y) is discon-
tinuous in a polygonal figure Q of area less than — (see Fig. 1.12)
so that it should he strictly contained within the figure. Denote ns
G the part of the domain G not entering into the interior of Q. The
boundary points of the polygonal figure Q which belong to G lie in G,

* And, of course, squarable. In what follows we shall suppose, without

any further stipulation, that the condition of squarabilily is always fulfilled.
** p er coo fRJ rh eto r 14 ThoorfMn* 14 ft onH 14 8
 CH. 1. D O U BL E INT EGRALS 39

anti therefore G is closed. The function / (x, y) is continuous on the

closed set G and hence is uniformly continuous on it. Choose 6 > 0
so Ihat the oscillation of the function / (x, y) on any part of the
figure G with diameter less than 6 should be less than — (where
S is the area of G). Now, consider a partition {£?<} of the domain
G whose first element G^ coincides with Q and all the other elements

are of diameters less than 6. Let us estimate the difference <} — ai

for this partition. We have
n
Q — to = d/,AS, — w,AS, -j 2 W — ™f) AS] <
1=2

< ( > / , - / « , ) — h 2 li s ASi

1-2

Rut M x— m, 2A’ and ^ i\5; < .9, and thus

i=2
£2— a) < 2A. -}- S —b

The number e > 0 being chosen arbitrarily, the function / (x, y)

is integrable hv virtue of Theorem 1.3.
4. Properties of Double Integral. The basic properties of the
double integral are completely analogous to the corresponding pro­
perties of (he definite integral of a function of one independent
variable and therefore we shall only enumerate them without giving
the proofs.
1- If functions /, (x, ?/) and / 2 (x, y) are integrable over a domain
G their sum (difference) is also integrable on G and

f ( l/i (*' y) t / 2 y)]d s= \\ (x, y) ds ± f f / 2 (x, y) d,s

v_mi • • m) mi
G G G
'»0 MULTIPLE INTEGRALS. FIELD THEORY AND SERIES

2. It' k is a constant number and a function / (x, y ) is integrable

on G the function /c/(x, y) is also integrable on G and
j j kf (x*y) ds = k j j / (x, ij) ds
G G
These two properties express linearity of the integral.
3. If a domain G is a union of two domains G{ and 6’2 and a function
/ (x. y) is integrable on and (72 tlien the function is as well inte­
g r a te on G. If, besides, Gt and Gz have no interior points in common
we have
\ \ f {*, y) d s = \ \ f (x, y) ds~r j j I (x, y) ds
G Gt Gz
This property is referred to as additivity of the integral.
4. 11 /, (x, y) and / 2 (x, y) are integrate on G and (x, y) ^
< / 2 (x, y) Iben
j ] /i (x, y) d s < j j / 2 (x, y) ds
G G
This property is called monvtonicily of the integral; it implies pro­
perties 5 and ().
5 (estimation of the modulus of the integral). If / (x, y) is integrable
on G the function | / (x, y)\ is also integrable on G and
| jj/(x , j j |/( x , y)\ds
G G
(> (mean value theorem).If a function / (x, y) is integrable on fraud
satisfies the inequalities
m f (x, y) ^ .1/
we have
mS < j j / (x, y) ds < MS (1.14)
G
where S is the area of the figure G.
The assertion immediately follows from property 4 and an obvious
relation
j \ cds = cS, c —const
G

If / (x, y) is a continuous function the mean value theorem can

be stated as follows:
O'. Tn the domain (7, there is a point (£, q) such that
j j /(x \ y)ds = f(t, q)S (1.15)
G
 CH. 1. D O U B L E INT EGR ALS 41

Indeed, take respectively, as M and m, the least upper bound and

the greatest lower hound of the values of the function / (x , y) on G.
Then, according to (1.14), we have
f { x , y )d s^ .M

But, as is well known, a continuous function defined in a closed

domain assumes, at some points of the domain, the values equal
to its least upper bound AI and greatest lower bound m (e.g. see [81,
Chapter 14, § 3). Suppose, for simplicity, that the function / (x , y)
takes on the values M and m at the points (art, i/i) and (x2, y 2) lying
in the interior of the domain G (the argument becomes a little more
sophisticated if one of the points or both fall in the boundary of
the domain 6’). Every two points of a domain can be joined by
a broken line contained in the domain. Let us connect, by a broken
line contained in the domain G, the points (x,, i/<) and (x*, y?) at
which the function is, respectively, equal to AI and m. The function
/ (x, y) is continuous along this polygonal line and, consequently,
together with the values M and m, it assumes all the intermediate
values. In particular, we can find a point (denote it by {£, q)) at which

/ (£> '»l) = 4 * I j y} ds
and thus formula (l.lfi) has been proved.
§ 3. ADDITIVE SET FUNCTIONS.
DERIVATIVE OF A SET FUNCTION WITH RESPECT TO AREA
1. Point Functions and Set Functions. The notion of a function
is one of the most important in analysis. Wc have already dealt with
functions dependent on one, two or several arguments. Applying
geometrical terminology we can say that such functions are variable
quantities dependent on a point of the line (for one argument),
on a point in the plane (for two arguments), on a point of a three-
dimensional space (for three arguments) or on a point belonging to
a space of higher dimension. But in mathematical analysis and
its physical applications we often encounter functions of different type
for which the values of their arguments are not separate points but
certain sets, for instance, some plane or space geometric figures.
Functions of this type arc known as set functions.
As an example of a set function, we can take the area £ (G) of
a domain* G defined, in a manner described in § 1. for all squarable
* The term “domain” is understood here in a wider sense, not as an open
connected set but as a synonym for the term “sot”. A class of sets on which set
functions are considered can ho chosen in an arbitrary fashion. In this hook,
as a rule, we den! with set functions whose argument is a squarable plane figure
(domain).
42 m u l t ipl e in t e g r a l s, fie l d theory and se r ie s

domains in the plane. Take another example. Let a mass be distri­

buted in the x, y-plane. Then, to each domain G lying in the plane,
there corresponds a certain number, i.e. a mass p. (G) concentrated
on G. Here again we have a variable quantity dependent on a domain,
that is a set function.
Now we introduce an important definition.
D e f i n i t i o n . A set function F (G) is said to be atUHtire if the
following conditions hold:
(1) if F (G) is defined for domains Gx and G.. it is as well defined
for their union G, -f- G2;
(2) if Gi and G2 have no interior points in common we have
F (G, + G2) - F (GO + F (G2)*
The above functions, area and mass, possess these properties.
Wc can give many other examples of additive set functions: surface
charge, amount of light energy impinging on an illuminated surface,
fluid pressure acting upon the bottom of a vessel etc.
We can also indicate examples of nonadditive set functions. For
instance, if, with every squarable domain, we associate the square
of its area we obtain a set function which is not additive.
Additive functions whose argument is not a plane but a space
figure will be treated in the next chapter devoted to the triple integral.
2. Double Integral as an Additive Function of Its Domain of
Integration. Let us consider the double integral
J] / y) ds

in which the integrand / (x, y) is regarded as being fixed whereas the

domain of integration G is variable. Then the integral becomes
a function (G) of the domain G. By virtue of propc»rty 2 of the
double integral (see the foregoing section), this function is additive.
As a class of sets for which the function is defined, we can take the
totality of all squarable figures contained in the domain G0 on which
/ (x, y) is defined.
3. Derivative of a Set Function with Respect to Area. Take again
a function of type p, (G), i.e. a mass distribution in the plane. If G is
a squarable domain and S (G) its area the ratio
H(P (1.5G)
S( G)

* In particular, it follows tliat if G is of area zero, then F (G) = 0. In the

case of a mass this means that we only consider mass distributions having a two-
dimensional (surface) density (hut not concentralpd at separate points or curves).
 GH. 1. D O U BL E IN T E G R A L S 43

is the mean density of mass distribution in the domain G. We now

infinitely diminish the sizes of the domain G by contracting it to
a fixed point p 0. If. in this process, ratio (1.16) tends to a limit
p (pQ) the limit is called the density of mass distribution at the point p 0.
Thus, a mass distribution in the plane can be directly defined by
indicating an additive set function p (G) or characterized by the
corresponding density which is a point function.
We now pass from our concrete example (mass distribution) to
an arbitrary set function. Unlike mass distribution, an arbitrary set
function can assume both positive and negative values.
Let F (G) be an additive set function defined for all the squarable
domains*. We say that a number A is the limit of the ratio
F<G)
S(G)

(where S (G) is the area of the domain G), as the domain G is contract­
ed to a point p a, if for every e > 0 there is 6 > 0 such that
F{G)
S(G) — A
for each domain G entirely lying in the circle of radius 6 with centre
at the point p 0.
This limit will be denoted by the symbol
F(G) dF
lim S( G) or ds
G-*-7»o

and referred to as the derivative of the additive set function F (G)

with respect to area. The derivative is not a set function but an
ordinary point function, i.e. a variable quantity dependent on
a point.
Turning back to the above example, we can say that the density
p (p0) of a mass distribution in the plane is the derivative of mass
with respect to area.
4. Derivative of a Double Integral with Respect to the Area of
Its Domain of Integration. The mean value theorem for the double
integral (sec § 2, See. 4, property 6) implies the following result.
Take the integral
F (G) = j j / (x, y) ds (1.17)

where / (z, y) is a fixed function which is supposed to be continuous

throughout a chosen part of the plane. Let us show that the additive

• O r fo r nil «;fjnnrr*hlo dnrnninc c o n t a i n ^ in a fWnd domain.

44 M U L T IPL E INTEGRALS, FIELD THEORY AND S E R I E S

set function defined by relation (1.17) possesses the derivative with

respect to area which coincides with the integrand / (x, y).
Actually, let p Q be a fixed point and G be a domain lying within
a circle with centre at p 0. Denote by m and M the greatest lower
bound and the least upper bound of the values of the function / (x, y)
in the domain G. By virtue of the mean value theorem, we have
f \ ^ y)lds*£M
‘cv
When the domain G is contracted to the point p0, i.e. when the
radius of the circle tends to zero, the numbers m and M tend, because
of the continuity of / (x. y) at the point pn, to the same value, namely,
to the value taken by the function / (x, y) at the point. Consequently,
the ratio whose values lie between m and M tends to the same limit.
Hence, we really have
'dF
ds = /(* . V)
5. Reconstruction of an Additive Set Function from Its Derivative.
We have discussed the problem of finding the derivative of a set
function. Here wo shall consider the reverse problem: let a point
function / (x, y) be given and let it be necessary to determine a set
function F (G) whose derivative coincides with / (x, y). If the func­
tion / (x, y) is continuous wo can immediately indicate such a set
function, namely, the double integral
f j / (x, v) ds (1.18)
G
regarded as a function of G. It appears natural to pose the quc.sliun
on whether there exist some other set functions with the same deriva­
tive. Let us show that if f (x. //) is continuous there is only one addi­
tive set function whose derivative is / (x, y) (and which thus is
expressible in the form of double integral (1.18)).
If Fi (G) and F2 (G) are two additive set functions with the same
derivative with respect to area we have

It is therefore sufficient to prove the following assertion:

If = 0, then F ~ 0 . The proof is implied by (the lemma
stated below*.
dF
Fern tun. If the derivative of an additive set function F (D)
exists in a bounded closed domain D and is nonnegative, then
F( I ) ) > 0.
 CK. 1. D O U BL E INTEGRALS 45

Proof. Assume the contrary, i.e. let F(D)<C 0. Then there is

I <C 0 such that
F{D) < / < 0
S( D)
that is
F (D) < IS (D) fl.19
Further, take a sequence of positive numbers e2, . . . convergent
to zero and break up the domain D into a finite number of parts D t
so that their diameters are less than e{. Then at least for one of those
parts (denote it as D a>) we must have
F (/)(1>) < IS (Z><1>)
because if the opposite inequality
F (O') > IS (/>,)
were fulfilled for all we should sum up these inequalities over
all Di and thus arrive at a contradiction to inequality (1.19).
Now, divide Z)tl > into parts with diameters less than Among
them there is at least one (denote it by for which the inequality
F (Ot2>) < IS (0 <2>)
holds. Continuing in this manner we obtain the sequence {/.) }
which is a nested collection of closed and bounded domains (the
symbol Z)(n> designates the closure of Din\ and we apparently have
F (D*n)) = F (l){n))). The diameters of D{n) tending to zero, there
exists a single point belonging to all * (let us denote this
point by p0). By the hypothesis, the derivative ~ exists every­
where in D, and. in particular, at the point p0, and therefore
its value at the point can he expressed as
F (J9(n))
lim (t.20)
n—*oo S (Din))
But, according to the construction of the sequence the
F (Z?^nh
ratio —-— does not exceed the fixed negative number / for all n,
S (P<n)) fe
and thus limit (1.20) cannot be nonnegative. The lemma has
been proved.
Replacing
tn
F (G) by - - F (G) and applying the lemma we see
that if exists and is non positive, we have F (D) 0. Finally, if

* This is the two-dimensional analogue of the nested interval theorem (e g.

see [81. Chapter 3, § 3).
40 M ULTIPLE INTEGRALS, K1ELD T H E O R Y AND S E R I E S

that is if we simultaneously have

dF n , dF A
— >0 and -5 - < 0

then F (D ) — 0 for every bounded closed domain.

6. Definite Integral of a Function of One Argument as a Function
of Its Interval of Integration. Let us now compare what has been
said with the analogous facts of the theory of the definite integral
of a function of one independent variable. The definite integral
b
J / (I ) d l
a
can be regarded (for a fixed function /) as a function of the interval
[a. 61, i.e. as a set function on the line. Furthermore, the well known
properties of the definite integral imply the additivity of this set
function. But a line segment is completely specified by two points,
namely, by its end points. If one of the end points is fixed a function
of the line segment becomes an ordinary point function. We encounter
tills particular situation when wo consider the integral
X

j/(D «*S (*.21)

(for a fixed a) as a function of its upper limit of integration. If we

substitute another point a for the lower limit a function (1.21)
gains a constant (independent of x) increment, namely, the one
equal to
a'
j / © <ii
a
Thus, an integral of a function of one argument is a uniquely spe­
cified set function on the line. When we regard such a function as
a function of intervals it can be reduced to a function of one inde­
pendent variable determined to within an arbitrary (additive)
constant. The theorem on the derivative of a double integral with
respect to area and the theorem on the reconstruction of a set func­
tion from its derivative are, respectively, the two-dimensional
analogues of tho theorem on the derivative of a definite integral of
a continuous function with respect to its upper limit and of the one
asserting that an antiderivative is determined to within an arbitrary
constant summand.
7. Extension of Additive Set Functions. If a function is not given
for all possible values of its argument for which it may be defined
we can usually extend the function if some of its properties are
 CH. 1. D O U B L E I N T E G R A L S 47

known. For example, if a function / (x) is known to be linear, i.e. is

of the form
/ (x) = ax + b
then, to livid its values everywhere, it is sufficient to have at one’s
disposal the values of the function at any two distinct points.
If / (x) is a periodic function of a period T 7> 0, i.e. if it possesses
the property
/ (x 4- T) » / (x)
for all x, then, to find its values everywhere, it is sufficient to know
the values of the function on the interval [0, T1. For instance, if the
values of sin x are known for all x from 0 to 2n we can find the sine
of any angle. The set functions can he treated analogously. If a set
function F (G) is known to he additive and if its values assumed
on a certain class of sets are given we can sometimes uniquely extend
the function, preserving its additivity, to a wider class of sets. For
example, let F (G) he an additive set function defined on all triangles.
Then it can be extended, as an additive function, to all polygons
(and then to a wider class of sets).
We have dealt with a problem of this type in § I where we have
studied the notion of area. The area is an additive set function of
a domain which has been originally regarded as being defined on
polygons (or polygonal figures) and then extended, with preservation
of its additivity, to a wider class of figures which we have called
squarable.
The general problem of constructing on additive extension of a set
function and determining the widest class of sets for which the
function can be defined plays an important role in many divisions
of mathematics. Bill here we shall not. discuss these questions in
detail because it would involve the introduction and systematic
application of ideas and concepts of the general theory of measure.

§ 4. SOME PHYSICAL AND GEOMETHICAL APPLICATIONS

OF THE DOUBLE INTEGRAL
1. Evaluating Volumes. At the beginning of this chapter we have
already discussed a geometrical problem leading to the notion of
a double integral, that is the problem of finding the volume of a cur­
vilinear cylinder. We have seen that, fora cylindroid bounded below
by a closed domain G in x, //-plane and above by a surface z — f {x, y)
where / (x, //) is a nonnegative continuous function, the integral sum

3 /<!.’. T|,)-\Si (1-22)

t 1
gives an approximate value of the volume. (The sum is taken over
all elements G* of a pari i Iion of the figure G in to snuamhlo
48 M U L T I P L E IN T E G R A L S . FIELD THEORY AND S E R I E S

is the area of the element G,-, and (Hf, i],) £ Gf.) As has been said
in the introduction to this chapter, the exact value of the volume
equals the limit to which integral sums (1.22) tend as the fineness of
the partition tends to zero. But the limit of sums (1.22) is nothing
but the double integral of the function / (x , y) over G. Its existence
(under certain assumptions concerning / {x, y) and G) has already
been proved (Theorem 1.3). Hence, the volume V of a curvilinear
cylinder bounded below by a closed domain G and above by a surface
z — f (x, y) (where / 7> 0 is continuous) is represented by the double
integral
/ (x, y) da
G
Strictly speaking, the volume of a curvilinear cylinder must be
defined as the value of the double integral. The concept of the vol­
ume, clear though it may he from the geometrical point of view,
is not given beforehand and therefore our considerations only indicate
that such a definition looks natural and is coherent with geometric
intuition.
We shall consider here some other problems to which the notion
of double integral is applied.
2. Computing Areas. Assuming that the integrand / (x, y) of
a double integral is identically equal to unity we arrive at the
expression
f f ds (1.23)
G

which is obviously equal to the area of the figure G because each

integral sum corresponding to integral (1.23) equals that area. The
formula
S^^dx (1.24)
G
for computing the area is sometimes more convenient than the well
known formula
b
5 r-r j / (*) dx
a
expressing the area of a curvilinear trapezoid because formula (1.24)
is applicable not only to a curvilinear trapezoid hut also to any
squarablo figure occupying an arbitrary position with respect to the
coordinate axes.
3. Mass of a Plate. Consider a plate lying in the x , [/-plane, i.e.
a domain G in which a mass with surface density, (> (x. y) is distribul-
 CH. I. D O U BL E INTEGRALS 49

ed. Let us lind the mass of the plate from the given density p (a:, y)
under file hypothesis that p (x. y) is a continuous function in x and y.
lireak up G into parts Gi in an arbitrary way and take a point (c<, ip)
in each of the parts. The mass of each element Gi can be approxi­
mately regarded as equal to p (t*, rp) ASj (where AS* is the area
of G^ and the total mass of the plate as equal to the sum

2 (J(E|, »|,)AS,
i=l
(1.25)

taken over all the elements of the partition. To obtain the exact
value of the mass of the plate it is necessary to pass to the limit
in the sum as the maximal diameter of the partition {G,} of the
domain G is infinitely diminished. Then expression (1.25) turns
into the double integral
f f p ( x, y ) ds (1.26)

which gives the mass of the plate.

It appears clear that determining the mass of a plate from its
density is a particular case of the general problem of reconstructing
a set function from its derivative which has been discussed above
(see § 3).
4. Coordinates of the Centre of Gravity of a Plate. Let us determine
the coordinates of the centre of gravity of a plate occupying a domain
G in the x, {/-plane. Suppose that p (x, y) is the density of the plate
at the point (x, y). Divide the domain G into parts G,, choose a point
(li> *1») in each of the parts and consider the mass of each part to
be approximately equal to p ( | {, Tp) AS { where AA’j is the area of
the subdomain Gi. Each mass can be thought of as being concentrated
at one point, namely, at the point ( | f, ip). Then we can write the
well known expressions for the coordinates x c and y c of the centre
of gravity of the system of material points:
n
2 hP *li) ASi y "HiP (Si» *1*) A*Sj
i___________
n Sfc— ^ ------------------ (1.27)
2 P(SI. t i l A S i S p ni) as £
1=1 i= l
Expressions (1.27) are approximately equal to the coordinates of
the centre of gravity of the plate. To receive the exact values of
the coordinates we must pass to the limit in formulas (1.27) as the
partition is infinitely refined, i.e. Then the sums entering
into formulas (1.27) turn into the corresponding integrals and thus
we obtain the formulas for the coordinates of the centre of gravity
4-0824
30 MUI.TFFT.E INTE GRALS. FIELD T HE ORY AND SERIES

of tlie plate:
{ j xp (x, »/) ds J J yp (T>y)ds
__________ . (1.28)
•*'c
p (x» y) ds * ^ { p (x, y) ds
‘g 6
If the plate is homogeneous, i.e. p = const, the formulas for the
coordinates of the centre of gravity are simplified:
^ \ x ds
v v 5 5 lfds
G o (1.29)
He =
SJ* ’
5. Moments of Inertia of a Plate. As is well known, the moment
of inertia of a material point about an axis is equal to the product,
of the mass of the point by the square of its distance from the axis
and the moment of inertia of a system of material points (about
the same axis) equals the sum of the moments of inertia of the mass
points it is formed of. Let a domain G in the x, i/-plarie be occupied
by a plate of density p Or, y). Break up the domain G into parts G-t
with areas A*Sh, choose a point (£/, rp) iu every part and replace the
plate by the system of masses p (c*. ip) A c o n c e n tr a te d at the
points (c*, ip). Then the moment of inertia of this system of mate­
rial points about the y-axis is equal to

y , E?p (Ej, 11,) A.S'I

t-1
This expression is taken as an approximate value of the moment
of inertia of the plate, and the smaller the diameter of the partition,
the greater the accuracy of the approximation. Passing then to the
limit as the maximal diameter of the partition of the domain G
tends to zero we receive the following formula for the moment
of inertia of the plate about the y-axis:
j j x2p(x, y)ds
h (1.30)
G
Similarly, Ihe moment of inertia of the plate about the x-axis is
equal to
j j JTP(*» y)ds (1.31)
G
Now let us find the moment of inertia of the plate about the
origin of the coordinate system. Taking into account that the moment
of inertia of a material point of mass m (placed at the point (x, //))
 CIf. 1. DC11?HI. E JNTllCllALS 51

about the origin is equal to

m (.x2 i/2)
and applying the same arguments we find that.
/ o = j \ (x’ -r y2)p(*> 1/)^
g"
i.e.
^0 = "i" ^!f
6. Luminous Flux Incident on a Plate. Let a plate occupying
n domain C of the x. r/-plane he illuminated by a point source of
light placed at a point with the coordinates (0, 0. zv). .Suppose that
the light intensity of the source is the same in all directions and
denote it by /. Let us compute the luminous flux incident on the
plate.
The luminous flux (IF impinging on an elementary area ds is
equal to Idio where c/w is the solid angle at the point (U. 0. z0)
subtended by the surface ds. Furthermore, rfo) is equal to the product
of the ratio of the area ds to Ihe square of its distance from the source
by the cosine of the angle between the normal to the area and the
direction from the area to the source. The value of the derivative
jr
— at a point (x, y) of the plate is known as the intensity oj iliumt-
ds
nation at thp point (denote it by A (x, y)). It follows that
. \ df I dut Izn
A (x <y ) - s r - - 3 r - (i2+j,*4..j )3/2
The total luminous flux falling on the plate is equal to the double
integral of A (z, y) over the domain 6\ i.e. equal to7
_____ ds ______
Iz0
<*®H-|ZaH-*o)3/2
7. Flux of a Fluid Through the Cross Section of a Channel. Con­
sider a fluid flow in a channel, and take a cross section of the channel
perpendicular to the direction of the flow. Introducing a Cartesian
coordinate system x, y in the plane of the section we can regard
the speed V of the fluid, at each point of the section, as a function of
x and z/, i.e. V — V (x, y). Let us compute the amount of the fluid
passing across the section in unit time. Take an infinitesimal cle­
ment ds of the section. The quantity of fluid passing through the
element in unit time is obviously equal to I lie mass of the elemen­
tary fluid cylinder with base ds and altitude V (x. ?/). that is equal to
pF (x, y) ds (1.32)
where p is the density of the fluid. To find the amount of fluid passing
through the whole section we must sum up infinitesimal elements
4*
52 MULTIPLE IN T E G R A L S , FIELD THEORY AND S E R I E S

(1.32), i.e. take the double integral

f f pi7 (x, y) ds
a
over the section.
Note. In above considerations and, particularly, in the last pro­
blem we have used such terms as “an infinitesimal element of area”,
“an element of mass” and the like. This terminology is widely applied,
especially in physical literatu re. For instance, we say that, for
a plate with density p (x, y), the quantity
p (z, y) ds
is its “element of mass” (concentrated on “an element of area ds”)
and the total mass of the plate, that is the integral
j \ P (x, y) ds
c
is regarded as “the sum of the mass elements”.
The meaning of such statements is that we always imply the
corresponding processes of passing to the limit (from finite sums
to integrals) which have been encountered in the above problems.
In what follows we shall sometimes use this “physical” language
(keeping in mind its real sense based on the corresponding passage
to the limit).

§ 5. REDUCING DOUBLE INTEGRAL TO A TWOFOLD

ITERATED INTEGRAL
We have already discussed the definition and basic properties of
the double integral, the conditions for its existence and some physi­
cal and geometrical problems involving this notion but we have
not yet studied the practical ways of evaluating double integrals.
The most important role in the solution of this problem is played
by the theorem asserting that the evaluation of a double integral
can be reduced, under some general suppositions, to successive
separate integrations with respect to each variable. It is the proof
of the theorem that we are going to study in § 5.
1. Heuristic Considerations. The basic idea of the theorems proved
below ties in the following considerations. Lei. us regard the double
integral
f \ /(* , y) dxdy

as the volume of a curvilinear cylinder T hounded below by a domain

6?, above by a surface z — f (xy y) and on tlie sides by a cylindrical
 CIT. 1. D O t i n L E IN T E G R A L S 53

surface passing through the boundary of the domain G (see Fig. 1.13).
The solid T can be thought of as being composed of infinitely thin
layers parallel to the y, z-plane. The volume of each layer is equal
to the product
J (x) dx
where J (x) is the area of the corresponding section of the solid T
and dx is the width of the layer. Then the total volume of the solid T
is equal to
b
^ J( x) dx (1.33)
a
But the area J (x) (as the area of a curvilinear trapezoid) is given
by the integral

\ f ( x*y) dy (1-34)
Vt<*)
where x is regarded as a fixed quantity and the quantities y x (x)
and y2 (x) are the coordinates of the end points of the line segment

which is the projection of the section on the x, y-plane. Combining

(1.33) and (1.34) we see that the volume of the solid T can be expres­
sed in the form
b i/a(x)
\ dx [ f (x, y) dy
« tfl(x)
Hence, we obtain the relation
6 i2(ac)
/ (*> y) ds = j dx lj / (x, y) dy (1.35)
<* mV)
Ib is formula tells us that when a double integral is thought of as
a sum of the elements / (x, y) dx dy wc* can first perform the summa­
tion within the layers parallel to one of the coordinate planes and
then sum up the results corresponding i-.ich layer. As an algebraic
54 MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

analogue of relation (1.35), we can write the well known formula

2 2 ( 2 <*;*»)
i, k ik
It is clear that if we took the sections of a curvilinear cylinder
parallel to the xT z-plane instead of the y, z-plane this would result
in the equality
d *t(y)
f (x, y) ds = j dy j / (x, y) dx
c xj(y)
(see Fig. 1.14). Now let us pass from our heuristic considerations
to strict arguments.

2. The Case of a Rectangular Domain of Integration. We begin with

a double integral taken over a rectangle with sides parallel to the
coordinate axes.
T h e o r e m /.5 . If f (x. y) is a function defined in the rectangle
P = {a x b, c ^ y ^ d} (1.36)
for which the double integral
j j f{x,y)dxdy (1.37)
p
exists and if the onefold (single) integral
d
J (x) = $ / y) dy (1.38)

exists for each fixed value of x in the interval a ^ x ^ . b then the

i t e r a t e d { r e p e a te d ) i n t e g r a l
b d b
j dx j / (x, y ) d y = j J {x)dx (1.39)

also exists, and we have

b d

\" j / (x, y ) d x d y = ^ dx j / (x, y) dy (1.40)

 CH, 1. D O U B L E INT EG R AL S 55

Proof. Divide the rectangle P into rectangular subdomains P tJ

by breaking up its sides with the help of points a = x0 <C x4 <5
< x2 <C . . • < xft = b and, respectively, c = y0 ■< y t <C y* <C . . .
. . . < Vi = d. Thus, is the rectangle of the form P;j =
= {x^, ^ x ^ x {, y j - i ^ y ^ l / j } (see Fig. 1.15). Let mfi be

Fig, 1.15

the greatest lower bound and A/j7- the least upper bound, on the
rectangle of the values of the function / (x, y). Choose a point
£i in each subinterval [x*_i, X/). Since y) for
yj-\ ^ y ^ Vjy we have
VJ
j f (ti, y) dy*£Mi j Ayj (&yf = yj — ys-i) (1.41)
vJ~t
and the integral in (1.41) exists because, according to the hypothesis,
integral (1.38) taken over the whole interval [c, d\ exists for every x.
Summing up inequalities (1.41) with respect to / from 1 to / we derive
i d i
2 (!,-)= f /(& , Mi Ayj * * = 1 ,2 , . . . , k
i=i c i=i
Multiplying each of the last inequalities by Ax* — x L — x«_, and
summing them, with respect to i from 1 to k we deduce
h i k h i
2 Axi ^\ mijhyj*£ ^* J (£i) Axj <. Axi
i—1 j=l i=l •*=1 7=1
h
The expression 2 «J(£i)Axj entering into this relation is an
i=l
integral sum associated with the function J (x) whereas
h i h i
2 Ax i^ n t t j & y j and 2 Ax* 2 Mu&y} are the lower and
«= 1 j=l i—i j=*i
 50 MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

the upper Darboux sums corresponding to double integral (1.37).

Consequently,
A
c o < 2 J (SO A x * < Q
If we now make all Axt and Ayj tend to zero then, since we have
supposed that double integral (1.37) exists,* both the lower and
the upper sums will tend to the double integral. Hence, the integral
h
sums 2 J (tj) Ax, tend to the same limit. Thus, we have
t=i
b b d
j j / (*, y) dx dy = ^ J (x) dx = j dx j f (x, y) dy
P a u c

Interchanging the variables x and y ^and supposing that the

b
integral J t ( y ) = j f ( x , y ) d x existsJ we derive a similar relation
of the form
d b
j dy j / (x, y) dx = j j / (x, y) dx dy
c a
d
Finally, if both integrals /(x )= ^ f(x,y)dy and J x(y) —
c
b
= J f ( x , y ) d x exist, together with integral (1.37), wc obtain

6 of d b
j* j / (*> y) d x d y = j dx j / (x, y) dy = ^ dy j j (x, y) dx
a c c a

3. The Case of a Curvilinear Domain. We now pass to tlie question

on reducing a double integral to an iterated one for the case of a
curvilinear domain. Let a domain G be bounded by two continuous
curves y = y\{x) and y = y 2(x) (where yo(x) ^ i/t(x)) a,,d by
vertical line segments x = a and x — b (Fig. 1.1(5). Then the follow-

* By the hypothesis, double integral (1.37) exists and therefore, for any
way of partitioning the rectangle P into subdomains such that their maxima!
diameter tends to zero, the upper and the lower Parbonx sums tend to the com­
mon limit, i.e. to the corresponding double integral. This enables us to realize
the partition in any appropriate manner, and we have chosen the one performed
by means of vertical and horizontal straight lines.
 CII. 1. D O U BL E IN T E G R A L S 57

ing theorem takes place:

T h e o r e m 1JS. I f the double integral
j j / (x, y) dx dy

exists for a function f (x, y) defined in the domain G, and the

integral
M(*>
J ( x ) = [ / (x, y) dy
t/iV)
exists for each fixed value of x from the interval \a, 6] the iterated
integral
b Va(x)
j dx j f (x, y) dy
a Vi(*)
also exists and we have the equality
b t/Xx)
\ § f (x, y ) d x d y = j dx [ / (x, y) dy (1.42>

Proof. Put c — min i/,(x), d = max y2(x) and embed the domain
G in the rectangle P determined by the inequalities a x ^ by

0
d

c
Fig. 1.16 0 a b JT

c ^ y ^ dr (see Fig. 1.16). Consider the auxiliary function /*(x, y)

defined on the rectangle by the relations
j f (x , y) in G
r (*, y) \ 0 in P — G
where P — G is the difference between the sets P and 6, i.e. the
collection of all points of P not belonging to G.
The function /* (x, y) satisfies the conditions of the foregoing
theorem. In fact, since it coincides with / (x, //) on the domain G
it is intcgrable in G and it is identically equal to zero in P — G
and thus is integrable there too. Consequently, by the properly of
58 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

additivity (see § 5, Sec. 4, property 3), the function is integrable

over the entire rectangle P. Furthermore, we have
j j /*(*> y) dx dy — j \ f (.x, y) dx dy
JG JG
and
\ j f* (*> !/) dx dy = 0
P-G
whence
j \ /*(*. y) dx dy — j j /(x , y) d.z dy (1.43)

Besides, for each value of x lying between a and b% the integral

a vi(*) valx)
\ f* (*> y ) d y = j /* (x , y) J /* (*, #) +
lo(*)
d

+ j f* (*> y) dy (1.44)
1/2(X)
is sure to exist because each of the three integrals entering into the
right-hand side exists. Actually, the line segments connecting,
respectively, the points (x, c), (x, y\{x)) and (x, y 2(^))* (*> d)
in the x, y-plane lie outside the domain G and /* (x, y) equals zero
vb(*>
on them, and the integral I /* (x, y) dy coincides with the
IM(*>
integral
ya(*)
f / (x , y) dy
Vl(x)
which exists by the hypothesis. The first and the third integrals
entering into the right-hand side of (1-44) being equal to zero, we
finally obtain
d 1/tOO

j /* (x, y) dy = j /(x , y)dy (1.45)

C Vl(x)
We see that the function /* (x, y) defined in the rectangle P satisfies
the conditions of Theorem 1.5 and, consequently, the double integral
of it over P can be reduced to the iterated integral:
b d

j j /* (*, y) dx dy — j dx ^ /* (x, y) dy
P o c
 GH. 1. D O U BL E I N T E G R A L S 59

From the last relation and equalities (1.43) and (1.45) we deduce
6 y*(xl
j j / (*» 1/) dx dy = j dx j / (x , y) dy
G a yj(x)
which is what we set out to prove.
In Theorem 1.6 we have considered a domain G such that every
vertical straight line x — const cuts its boundary at no more than
two points (or, #j(x)) and (x, i/2W) and supposed that the integral

J ( x ) = \ 1( x, y ) dy (a*Cx<b)
yi(*>
exists. If we suppose that every straight line y — const has at
most two common points (x^y), y) and (x2(y), y) with the boun­
dary of a domain G (see Fig. 1.17) and require that the integral
**(!/)
\ / (^ 1 y) dx should exist for each fixed y we can prove the
*i(y)
existence of the iterated integral
d xj(!/)
\ d y \ f (x, y) dx
C x,(y)

and its coincidence with the double integral / (x, y) dx dy.

As was seen at the beginning of § 5, the geometric meaning
oF the formulas reducing a double integral to an iterated one is

Fig. 1.17 Fig. 1.18

that the volume of a solid is equal to the integral of the area of its
cross section (which is a function of the variable determining the
position of the cutting plane).
Note 1. If the domain G is such that there are straight lines (vertical
or horizontal), passing through interior points of the domain, which
have more than two common points with its boundary, then, to
represent the double integral taken over the domain in the form of
60 m u l t ip l e in t e g r a l s, fie l d theory and s e r ie s

an iterated integral, one should divide the domain G into parts

satisfying the conditions of Theorem 1.6 and separately reduce each
of the corresponding double integrals to an iterated one (see Fig. 1.18).
For example, let the domain of integration G he the unit circle
xr : y2 ^ 1 from which the ellipse x2 -r 2y 2 ^ 1 is cut out

Fig. t.19

(Fig. 1.19). Then the double integral over G can be, for instance
represented as
I 1 T—a2
j ] /(* • y ) d x d y = f dx f j{x,y)dy +
o - y J^

+ ] dx ( / (x, ;/) dy
_ yr \ - X 2

that is in the form of a sum of two iterated integrals.

Note 2. If a double integral can be reduced both to an iterated
b i/2(x)
integral of the form J dx ^ / (x, y) dy and to an iterated integral
« Vl(x)
d x t( y )
of the form ^ dy f / (£, y) dx then, when computing the double
c x t(v)
integral, we can use any of these representations. But it may well
happen that one of thorn is more convenient than the other, and
therefore, in concrete problems, an appropriate choice of the order
of integration (i.e- tlie order in which integrations with respect to x
and y are performed) may be of essential significance.
Exercise. Write the double integral
/ (x, y) dx dy
a
 CH. t. D O UBL E INT EG R AL S Cl

where G is the domain hounded by the curves y = ]/r2ax — x2 and

y = V 2ax and by the straight line x = 2a (Fig. 1.20) in the form

Fig. 1.20

of an iterated integral (for both possible orders of integration).

2a Vr 2ax

Answer j / (*, y ) dy
rgftx-x*
2a 2a a a — V e i l —j/^

( 2) j dy j / (x, y) dxA- j dy j f(x, y) dx +

a y 2 0 yt
2a la
a 2a

\ dy f / (x, y) dx
• »
0 a-f- V a J - y J

In the second case we have to break the integral into three summands
whereas the first case involves only one term.

$ fi. CHANGF! OF VARIABLES 1N DOUBLE TXT EC UAL

We often apply a change of variables when we integrate a function
of one independent variable, and this method is also very important
for evaluating double integrals. Before studying the problem of
changing variables in a double integral we shall discuss some ques­
tions related to mappings of domains.
1. Mapping of Plane Figures. Consider two planes with respective
Cartesian coordinates x, y and H. rj in them. Suppose that in the
x, y-plane we have a bounded closed domain G with boundary L
and in the q-plane a hounded closed domain I'* (see Fig. 1.21a
and b). Let
x = x (I, q), .y = y (I, q) (1.40)
be two functions defined in the domain F. Suppose that when the
point (£, q) runs over the domain Y the corresponding point (x, y)

* As before, we suppose that the domains G and T are squarable

02 MULTIPLE IXTEORAI^, FIELD THEORY AND S E R I E S

runs over the domain G. Thus, functions (1.46) define a mapping

of the domain T onto the domain G.
Let the mapping satisfy the following conditions:
(1) The mapping is one-to-one, which means that, to distinct
points of the domain T, there correspond distinct points of the
domain G. In other words, the solutions
% = I (*» y), i] = m (*» y) (1.47)
of equations (1.46) (obtained by resolving the equations in c and q)
are uniquely defined throughout the whole domain G.

Fig. 4.21 (a j (&J

(2) Functions (1.40) and (1.47) are continuous and possess conti­
nuous partial derivatives of the first order.
(3) The junctional determinant (Jacobian)
Ox d x
D (x, y) _ dl dT\
(1.48)
l) (5 * T)) ~ 0y dy
ol <?q
is different from zero everywhere in the domain I', and, consequently,
since the derivatives entering into the Jacobian are supposed to
be continuous, it retains its sign in F.
The Jacobian of inverse mapping (1.47) is connected
with Jacobian (1.48) by the relation
d y) 0 q) _ a
1> (£, t|) ’ D (*, y)
which is directly implied by the definition of the product of deter­
minants and the rules for differentiating a composite function.
Therefore the Jacobian ^ does not vanish in the domain G.
y)
If vve are given a smooth or piecewise smooth curve
I — I (0, q = 11 (0. a < t < (1
in the domain F, mapping (1.46) transforms it into the curve
-r = x a (t). t, it)) = x (o, y = y (I (/), ii (0) = y (0
 CH. 1. D O UBL E IN T E G R A L S 63

which is again smooth or piecewise smooth. Indeed, the deriva­

tives —
dt and — at existing and being continuous, the derivatives
dx _ dx , dx_ ti«] , dy_ dfi dy (in
dt ~~ d \ dt 5q dt dt 0\ dt dq dt
also exist and are continuous. Furthermore, if at least one of the
derivatives ^
dt and dt $ is different from zero the derivatives ^at and
do not vanish simultaneously since .. ff* ~f= 0.
dt ^ n* ^
We can also assert that the boundary A of the domain F is mapped
on the boundary L of the domain G. This follows from the theorem
on implicit functions (c.g. see 181, Chapter 15, § 2). Indeed, if,
to a point (:r0, y Q) belonging to L, an interior point (£0, “Ho) of the
domain F corresponded, the relations
x *= x (I, T|), y = y (H, ii)
would define the quantities H and q as functions of x and y in a
neighbourhood of the point (x0, y Q). But every neighbourhood of
a boundary point contains points not belonging to G, and hence the
point (£0, q0), an interior point of F, would possess a neighbourhood
lying in T and not mapped into G which contradicts the hypothesis.
2. Curvilinear Coordinates. Consider a straight line | = £0 in
the domain F (Fig. 1.21). It is mapped on u smooth curve lying
in the domain G and determined by the parametric equations
x = x (£„, q), y = y (£o» *l) (1-49)
(where q is the parameter). Similarly, to each straight line q •■= qrt
in F there corresponds a curve in the domain G with parametric
equations
x = X (1, q0), y — y H, q0) (1.50)
where £ is the parameter. Curves (1.49) and (1.50), lying in the
domain G, into which mapping (1.46) transforms straight lines
parallel to the coordinate axes £, q and belonging to F, are referred
to as the coordinate curves (^-curves and q-curves) in the domain 6’.
The mapping
x = x (£, q), y = y (£, q)
being one-to-one, it follows that there is a single curve of form
(1.49) passing through each point (jr, y) of the domain G which
corresponds to a given constant value of £ and a single curve of form
(1.50) corresponding to a constant value of q. Consequently, the
quantities £ and q can be regarded as the coordinates (different,
of course, from the Cartesian ones) of [mints belonging to the domain
G. The coordinate lines (1.49) and (1.50) corresponding to thp*o
<)4 MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

coordinates being curvilinear (but in the general case, not straight

lines as in the case of Cartesian coordinates), the quantities £ and r|
are called the curvilinear coordinates of the points of the domain G.
Thus, from the geometric point of view, the variables c and r|
are interpreted in a twofold sense: on the one hand, they are the
Cartesian coordinates of the points of the domain T and, oil the
other hand, they are the curvilinear coordinates of the points belong­
ing to the domain G. Accordingly, e v e r y relation of the form
<li (£, r|) = 0 can be regarded as an equation (in Cartesian coordi­
nates) determining a curve X lying in the domain T and also as an
equation (in curvilinear coordinates) of a curve £, the image of the
curve X under mapping (1.46), placed in G.
3. Polar Coordinates. The connection between polar coordina­
tes r, <p and Cartesian coordinates x , y is given by the relations
x = r cos q, y — r sin q (r ^ 0, 0 ^ q < 2n) (1.51)
if the pole coincides with the origin.
The coordinate curves of the polar system shown in Fig. 1.22 are
the concentric circles with centre at the origin (r = const) and the
rays starting from the centre (q — const). Mapping (1.51) transforms
the half-strip r ^ 0, 0 ^ q < 2ji onto the entire x, [/-plane. The
mapping is one-to-one everywhere except the point x = 0, y = 0
which is the imago of the half-segment r — 0, 0 q < 2jt of the
r, ip-plane. If we delete the point x = 0, y — 0 we can consider
the inverse of mapping (1.51) which transforms the punctured
x, [/-plane into the half-strip r !> 0, 0 ^ q <1 2;i. The inverse
mapping is continuous everywhere except the positive half of x-axis
because, all h o u g h the value q = 0 corresponds to all the points
of the semiaxis, the variable q tends to 2ji but not to zero as the
point M (x, y) approaches tlie seminxis from below. Thus, formulas
(1.51) define the mapping of the half-strip rl^O, 0 ^ <| <C 2n onto
the x, y-plane which is the one-to-one and continuous, together with
its inverse mapping, everywhere except the points at which r ~ 0
or q — 0.
We can visualize the transformation from the half-strip on the r,
q-plane to the x, [/-plane as “spreading a fan”. Imagine that we are
watching a film showing the process of spreading out the half-strip
0 ^ /• < oo, 0 ^ q < 'in as a fan which, when opened, covers the
x, [/-plane. Fig- 1.23a represents the first still of the film. Fig. 1.236
the second, Fig. 1.23c shows one of the final stills and Fig. 1.23d
the last one.
Consider an example. Lei a rectangular domain in the q-plane
be given by the relations 0 < « < r < 0 oc q (I < 2 ji.
The “fan spreading” procedure transforms it into an annular sector
in the x, //-plane (Fig. 1.24).
 y

'S
A

(3 ) (ft

£l
G6 MULTIPLE INTEGRALS, FIELD THEORY AN D S E R I E S

Let us find the Jacobian corresponding to the transformation

from Cartesian coordinates to polar ones, i.e. the Jacobian of func­
tions (1.51). We obtain
fD { x , y ) _ C0S<P — rsin<p
(£• il) sin <p rcos<p —
The Jacobian is different from 2 ero everywhere except the point
xA= 0 , y — 0 .
4. Statement of the Problem of Changing Variables in the Double
Integral. We now formulate the general problem of changing vari­
ables in the double integral. Let G be a closed domain bounded by
a piecewise smooth curve L and / (x, y) a bounded function,
defined in Gt which is continuous or has discontinuities forming
a set of area 2 ero. Further, let the functions
x = x T|), y = y (£, T])
determine a mapping of a domain V on the domain G and let the
mapping satisfy the conditions (l)-(3) enumerated in Sec. 1 . Our
aim is to represent the integral
| j / (x, y) dx dy

taken over the domain G as an integral over the domain T by trans­

forming its element of integration to the new variables | and tj.
5. Computing Area in Curvilinear Coordinates. In deriving the
formula for changing variables in a double integral the chief step
is to express the area of a domain in curvilinear coordinates. The
following theorem solves the problem:
T h e o r e m 1*7* Let x — x (£, tj), y — y (£, if]) be a continuous
and continuously differentiable one-to-one mapping of a domain V
of the £, r\-plane on a domain G in the x, y-plane. Suppose that the
Jacobian ^ is everywhere different from zero. Then
area of G = = (1.52)
c r
We shall preface the proof of the theorem with an intuitive argu­
ment (to which, by the way, the reader may restrict himself).
Consider two pairs of coordinate curves lying infinitely close to
each other in the domain G. Let the first pair correspond to the
values
£o « n <1 lo 4*
of the coordinate £ and the second pair to the values
tio and x)0 -{- dr\
 CH. 1. D O U B L E IN T E G K A L S 67

of the coordinate tj. These coordinate curves cut out of the domain G
an infinitesimal element of area A 0A iA 3A 2 (see Fig. 1.25). This
element can be apparently regarded as a parallelogram to within

*3

Fig. 1.25

infinitesimals of, order higher than the first. We see that the sides
of the parallelogram are the vectors

and

The area ds of the parallelogram A^A^AZA Z is equal to the modulus

of the determinant whose elements are the projections of the vectors
A qA i and A 0A Z on the coordinate axes, that is
dx «. dy

d s = modulus of y) dl dii (1.53)

dr\' — *>(h n)
01 | Ut\ dr\1
Hence, the total area S of the domain G is obtained by summing
up all the elements, i.e. is in fact representable in the form of a
double integral taken over the range T of the variables | and r|:
& (*> y) d^dr\
(&, *1)
We can* now corroborate our intuitive argument with a proof.
We shall leave out some details which we believe will not be difficult
for the reader to understand. Besides, to simplify the consideration
we shall suppose that the mapping is defined and satisfies the require­
ments of the theorem not only in the domain V but also in a wider
domain in which V is strictly contained together with its boundary.
Proof of Theorem 1.7. Wc first take a simple hut fundamental
case when the domain in the £, q-plane is a rectangle 11 with sides
parallel to the coordinate axes, and the mappirier of the rectanerle
5*
OS MULTIPLE IN T E G R A L S , FIELD T H E O R Y AND S E R I E S

on the x, //-plane is a linear one expressed by the formulas

x = x0 n c — Jii], y — y Q H- 4- fejT} (1.54)
a b
where x0, t/o» ai ai «nd bx are constants and 0.
a { 6,
As is well known from analytic geometry, the image of such a rec­
tangle II under a mapping of this type is a parallelogram P whose
area is connected with the area of the rectangle II by the relation
/ a b
area of P = ( modulus of (area of fl) (1.55)
at bi
(prove it). It follows that every squarablc figure (h lying in the
r, q-plane is mapped hv linear mapping (1.51) onto a squarable
figure F in the x, //-plane whose area is expressed as
f
a b
area of /■’— | modulus of (area of (|i) (1.50)
o>i b\
l»y the way, it is only relation (1.55) that we need for our further
aims.
Now let us consider an arbitrary (possibly nonlinear) mapping
.r = x (£, t]). y — // (£. i|). of a domain 1\ satisfying the conditions

of the theorem. Take a point (£0. r]0) belonging to the domain F

where the mapping is defined and consider the rectangle
lo ^ € < lo -i- ^ i» r |0 ^ i| t 1o h z

which we again denote by II (Fig. 1.2 0 ).

Applying Lagrange’s theorem on finite increments we rewrite the
equations defining the mapping of the rectangle into the x, //-plane
in the form
dx , dz , ,
x ^ x 0 -r*^|-asH--^-an r ^ lt !J - Uo + -Jr 1 dr\ ^ 2 (f 07)
where x0 —x (9 0 . q0), //o y (£01 %), the values of the derivatives are
taken at the point (£0, qa) and
a, — (x* (€*, il*) —xi (cUf q0)) dc. r (xj, (£*, if) - x^ (t0, r|0)) dx\
a2 no))d|-;-(//n (E*N ho))^l
t * ^ r** s; do<h* :'h; **
(Here we have ■1 )
 CII. 1, DOU DLE IN T EG R AL S G9

By the hypothesis, the first derivatives of x and y with respect

to £ and T) are continuous and thus they are uniformly continuous
in the hounded closed domain V. Consequently, for every e > 0
there exists a sufficiently small h > 0 such that if h{ -\-h2 < l h
the inequalities
I 4 (!* *1) — xi (£o> ho) I < e, | (£, q) — ( | 0, q0) | < e
hold for all ihe points (c, q) belonging: to the rectangle II, and
similar inequalities are true for y\ and y\x. e. being independent of
the particular choice of the point (£a, q0). These estimates show that
I aj | C eh, | a 2 | < eh (1.58)
Now compare nonlinear mapping (1.57) with the linear mapping
+ + y - y„ - If- d l \ dll (1.59)
which is obtained by dropping otj and r/.2 in formulas (1.57). As we
already know, a linear transformation of this kind maps the rec­
tangle II on a parallelogram which \vc again denote by P and, accor­
ding to ( 1 .5 5 ), we have
dx dx
area of P — modulus of 0y . (area of 1 1) (1.60)
^5 (M|
Nonlinear mapping (1.57) transforms II into a curvilinear figure oF.
Let us investigate what is the difference between its area and the
area of the parallelogram /A
By virtue of (1.58), for each point (c, q) 6 U we have
I x — x | = | a, | < ehy I y — y | =- I a 2 | < eh
i.e.
1/ (x — x)z -\- (y — y)2< V 2 e h
In other words, the distances between the images of the point
(£, q) 6 n under linear mapping (1.5ld) and nonlinear mapping
(1.57) are less than ^ 2 eh. Therefore, if we embed the boundary
of the parallelogram V in a strip of width J/IF eh the boundary
of the curvilinear figure cT will he strictly contained in the strip
(Fig. 1.27). It is clear that the difference hot ween the area of and
the area of P does not exceed the area of the strip. Performing simple
computations we find that llic area of the strip is not greater than
its width multiplied by the perimeter of the parallelogram P. The
perimeter can be easily estimated. Let a positive number Af be so
chosen that each of the derivatives ^ a n d ~ does not
dc */ri lit 'Oi
70 MULTIPLE IN T E G R A L S , FIELD THEORY AND S E R I E S

exceed A/ in its absolute value throughout the domain T (the deri­

vatives are continuous and therefore bounded in the bounded closed
domain T). Then (1.59) immediately implies that the sides of the
parallelogram P cannot exceed M h. Thus, the perimeter of P is

Fig. 1.27

not greater than AAIk, and the area of the strip in which the boundary
of P has been embedded does not exceed 4 2 eMk2, i.e. is not
greater than
eAf •(area of II)
Consequently,
area of & = area of P + y
or, by virtue of (1.60), we have
dx dx

area of eP= ^modulus of dy dy

(area of 11 ) + y (1.61)
W
where
| V I< V2 t M .(area of n) (1.62)

Let now iD be a polygonal figure lying within T and composed

of rectangles with sides parallel to the coordinate axes, and fp be
its image under the mapping x = x (£, -q), y = y (£, q). Break up
O into rectangles ITi so that the half-perimeter of each of them
should be less than h . The union of the images of the rectangles
is the figure jF, and the area of each i can be represented in the
form
area of &i = D (*1 y) *(area of IT#)-by# (1.63)
D (£, q)
*1= 0 #
where (£,-, q#) is a point belonging to the rectangle Ilf and
IYf I<c V"2eiM •(area of 17#)
 CH. i. DOUBLE INTEGRALS 71
«
Sum up equalities (1*63) over all the rectangles II This jdelds
n
2 (area of <0 ^) = area of JF =
i —1
n £«
Q(x, y)
£>(£* n) 6~ l|
• (area of n,)+ v< 2 (1.64)
i=l i=l»
The first summand on the right-hand side of equality (1.64) is
obviously an integral sum associated with the integral

and the second one does not exceed the expression

n
1^2 Ms 2 (area of Ilf) = 2 A/e*(area of O)
t=r.
where e can be. made arbitrarily small (because the diameter of the
partition of the figure <D can be chosen as small as desired). The
integrand being continuous, integral (1.65) is sure to exist. Conse­
quently, we can pass to the limit in relation (1.64) as the partition
of the figure CD is infinitely refined. We thus obtain
,rea o f j ^ = j j |

To complete the proof we should pass from the polygonal figure d>
embedded in the domain T to the domain T itself. This can be easily
performed. The domain T being squarable, we can find two figures
<t>! and <t>2 composed of rectangles,* the first of which is embedded
in r whereas the second envelops T, such that the difference between
their areas is less than a given positive number 6 . The mapping
x — x (E, r))f y = y ( |, q) transforms them into two squarable
figures t and the former being embedded in G and the latter
enveloping G. We can easily show that
| area of & x— area of <1 (2A/2 -f- y r2d /e)6
| prove it by applying relation (1.64) and the inequality
max d (*. y) < 2 Af2j . But then
d <£. *1)
| area of G — area of J<C (2M2 4 - Y 2 Mz) 6 (1 .6 6 )

* The enveloping figure Oj must lie in the domain in which V is strictly

contained and in which, by the hypothesis, the mapping is defined and satisfies
the conditions of the theorem.
 7 2 MUI-TIPLF. JNTECJKALS. KIEI.D T11EOHY AND SEK1ES

The figure i is the image of the polygonal figure Uh and therefore,

by what has been proved, we have
area of .Fi j \ to (*» {/) (1J>7)
f>ib n> i
*lM
Furthermore, by the mean value theorem, wo can write
to (*, „) d\dv\
I
HI SHSK'"-.!!
5
<l»|
to u)

= [j | |^ ^ ‘1 < 2A/;5 (1 .OK)

From (1.60) and (1.68), taking into account (1.67), we deduce

area of G — ( j | P | </g dr] I < { U P + V 2 AH) 6

"Flic number <S being arbitrarily small here, the proof of the theorem
lfills follows.
Mote I. The fundamental idea upon which both the above proof
and the foregoing intuitive? argument are based lies in the fact that
a nonlinear mapping x «* x (£, ri), jy — r/ (£, r]), when considered
in the small, can be approximated with a linear one, and the smaller
the domain, the greater the accuracy. 15y the way, properly speaking,
the substitution of a linear relation for a nonlinear one, considered
in the small, is, in general, the basic idea of mathematical analysis.
Example. Consider again polar coordinates. The cur\es r — r0,
r = rQ dr, cp •— <p0 and <| - <p0 c/cp cut out of the x, y-plane
an infinitesimal rectangle with sides dr and rp di\ (see Fig. 1.28).
Therefore the clement of area in polar coordinates is equal to r0 dt$ dr.
(The same result is, of course, implied by general formula (1.52)
since H}*1
t o (r, <p)
^ r.)' ConsequentIv,
*
the area in polar coordinates
is expressed by the formula
.S' = j j r dr d<( (1.6!<)
r
where P is the range of the variables r and cp. In particular, if the
domain G is bounded by two rays cp — cp] and cp = cp2 and by a
curve r - r (cp), i.e. has the form shown in Fig. 1.20 (represent, tin?
corresponding domain F in the r. (j-plane), then, reducing double
integral (1.611) to an iterated integral, we obtain
 C ll. 1. DOUULE IN T EG R AL S 73-

Performing integration with respect to r we find:

S = y j r* (<p) d<p
*1
This is the well known formula for area in polar coordinates ( e . g .
S e e 181. Chapter II, § 2).
Note 2, Formula (1.53) indicates the geometric meaning of the
absolute value of the Jacobian ^ x) . Actually. denote the
(s*'!)

Fig. 1.28 Fig. 1.29

Jacobian, for brevity, as J (£, i|) and consider the mapping of the
domain T on the domain G determined by the formulas
x = x (|, ii), y — y (|, l])
The mapping transforms the infinitesimal rectangle belonging to I1
(see Fig. 1.30), which is hounded hy the straight linos
I =- E„, I = H0 + d \ and q — *)0, q - i ]0 -r ch]
and has tlic area t/i), into a parallelogram of urea
I J ( 6 , »i) I d% rfn
Thus, the quantity | ./ (£, rj) | is the coefficient of area expansion
(iat the point (£, q)) for the mappingof the domain T on the domain G.
Note 3. In Theorem 1.7 we have supposed that the mapping
x = x {£, i]), y = y (£, ij)
of the domain F onto domain G is one-to-one. Hut expression (1.52)
for area in curvilinear coordinates remains true even when the
condition is violated at some separate points or on separate curves.
As a typical example of this kind, let us take the mapping of the
rectangle 0 ^ r ^ a, 0 ^ tp ^ 2 rc on the circle x 2 |- y 1 ^ a2,
determined by the formulas
x -- r cos (f , y — / sin cp (1.70)
which corresponds to the introduction of polar coordinates. The
mapping.satisfies the conditions of Theorem 1.7 everywhere except
74 MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

the points belonging to the line segment y — 0. 0 ^ x ^ a. Con­

sider tile rectangle p ^ r < ^ a , 0 ^ ip ^ 2 ji — e in the r, (p-plnne

(where 0 < p < a and 0 <C£<C 2 n) and its image under mapping
(1.70) in the x, i/-plane (see Fig. 1.81). Formula (1.52) holds here
because conditions (l)-(3) are fulhlled for these domains. Now,

Fig. 1.31

passing kto the limit, as p — 0 and t> 0 , we see that formula

(1.52) remains true for the entire circle r <^a.
Similar arguments are applicable in the general case of an arbi­
trary mapping which is one-to-one everywhere except separate
points or curves.
6 . Change of Variables in Double Integral. Expression (1.52)
obtained for area in curvilinear coordinates enables us to derive
the general formula for changing variables in a double integral.
Consider the integral
j j / (*t y) dx dy (1-71)
G
whore the domain G is bounded by a piecewise smooth contour L
and the function / (x, y) is continuous throughout the domain
(including its boundary) or is bounded and continuous in it every­
where possibly except a set of area zero.
[jet functions x — x (£, q) ami jy = y ( |, ij) define a one-to-one
correspondence between the points of the domain G and a domain T,
and let the mapping satisfy all the requirements under which the
validity of formula (1.52) expressing the area of the domain G in
 CH. 1. D O U B L E I N T E G R A L S 75

curvilinear coordinates has been established. Divide the domain T

into parts T* by means of a system of piecewise smooth curves. The
corresponding piecewise smooth curves in the x, y~plane (the images)
break the domain G into parts Gi of area AS-,. Choose an arbitrary
point (x<, ih) in each part G, ami form the integral sum
^ 1(xi, yi) A*5? (1.72)
i= l
associated with integral (1-71).
Applying formula (1.52) to each subdomain Gi we obtain

Denoting the Jacobian as / ( | , r|) and taking advantage

U (fc* *1)
of the mean value theorem we can write
AS* = | J (If, ^*) | Ad*
where Aa* is the area of the subdomain T*. Now we substitute the
above expression for the quantity ASi in integral sum (1.72)
and arrive at the sum

i=i
Tlie point (|*. r|f) appears when we apply the mean value theorem,
and lienee its position, within the subdomain T*, is preassigned by
the properties of the function and the subdomain, and we cannot
take it at pleasure. Hut the point (x,% f/ , ) , unlike (|*, t] * ) , is chosen
in the corresponding subdomain G, quite arbitrarily. Therefore we
can put
x t = x ( | f , T )f) , = ( | f , T |f )

i.e. take, as (x*, i/f), the point of the subdomain G, corresponding

to the point (|f, ijf) of the subdomain T,. Then the above integral
sum turns into the expression

> ]/(* (If, Tif), y (If, nf)) IJ (If, rif) | Ac,

t= l

which is nothing but an integral sum for the integral

M /( * ( S .n ) . 0(1. n ))!-'(5. T|) Ids dll (1.73)

r
T1io integrand being continuous in the domain F or bounded and
continuous everywhere in T except the points belonging to a set
71i MULTIPLE IN T E G R A L S , FIELD T H E O R Y AND S E R I E S

of area zero, the integral is sure to exist. If we now make the maximal
diameter of the partition of the domain T into the parts tend to
zero, the diameters of the subdomains Gt will also tend to zero.
In this limiting process, the integral sum under consideration must
tend both to double integral (1.71) and to integral (1.73). Hence,
the integrals are equal:
j j / (*> V) d x d y — ^j / (x (|, q), y (|, q)) | J (1. q) | dr\ (1.74)

It is this formula that describes the general case of changing variables

in the double integral.
Thus, if G is a bounded closed domain with a piecewise smooth
boundary and a function / (x, y ) defined in the domain is continuous
throughout the domain or bounded and continuous in it everywhere
except a set of area zero, and if the formulas
X — x (I, q), y = y (H, q)
determine a correspondence between the points of the domain G
and a domain F lying in the q plane which satisfies conditions
(i)-(3) of Sec. 1, we have formula (1.74) for changing variables in
the double integral.
Helation (1.74) also remains true when the condition that the
mapping of the domain G on the domain F is one-to-one, continuous
and continuously differentiable is violated at separate points or
on a finite number of curves with zero area.
As in the case of an ordinary definite integral of a function of one
argument, the method of changing variables is one of the most
powerful tools For reducing a double integral tu a form appropriate
for computing it. But it should be noted that in the case of two
independent variables there appears a new feature. The matter is
that a change of variable is introduced in a definite integral in
order to simplify the element of integration whereas when changing
variables in a double integral we try to simplify not only the inte­
grand but also the shape of the domain of integration. What has
been said is so important that it is sometimes advisable to complicate
the integrand when this yields a simplification of the domain of
integration.
Example. Evaluate ^ j dx dy where G is the domain bounded
*a
by the ellipse ^ ~ = 1. Here the integrand is identically equal
to unity but nevertheless it is expedient to perform the change
of variables
x — rtp cos y — b\> s i n ij; (1.75)
 CI1. 1. DOUHLE IN T E G R A L S 7

The Jacobian of this transformation is equal to abp. 'The domaii

of integration is changed under the mapping into the rectangh
0^ <r <C 2 ji, 1

Passing to tlie new variables and writing the double integral as ai

iterated one wo obtain
2 jt t
f I dx dy =» at \ d(( f p c/p «* nab
• v - •
C 0 0
Exercises
Compute the area of the domain bounded by Llie curves xy =
— 1 . xy = 2. y • x2 and y = 2 x2.
flint. Take, as new variables, the expressions
l-ry , (1.70
2. Draw the families of the coordinate curves corresponding t.i
transformations (1.75) and (1.76).
7. Comparison with One-Dimensional Case. Integral Over an Ori
ented Domain. Formula (1.74) is analogous to the formula for chang
ing variable in the definite integral:
*> &
J / (x) dx — ^ / (x (0) x (t ) df ( 1.77
<t a

The only difference between them is that in the case of one indc
pendent variable we do not take the absolute value of the derivativi
x' {() (which plays the role of the Jacobian here) but the derivative
itself. This is accounted for by tin* fact that the definite integro
6
^ / (x) dx is taken over an oriented interval fa, fr| and changes it
cl
sign when the limits of integration are reversed whereas the domaii
of integration of a double integral is not oriented. If. for the dofinib
integral, we introduce the condition that the limits of integralioi
must he so set that the lower limit should he not greater than th<
upper, formula (1.77) (in the case of a monotone function x — x (/)
takes the form
b ft

*\ / (x) d x = \If
a a
1
(r (0) | x' (t) Idt (1.78

(Check it un.)
On the other hand, for a double integral, we ran also introduc
the notion of an oriented domain and attach the sign plus or mi mi
to its area according to the oriental ion.
78 M UL TI PL E INTEGRALS. FIELD THEORY AND S E R I E S

We introduce the orientation of a domain by choosing a certain

direction of describing the contour of its boundary as positive.
Namely, the orientation of a domain is said to be positive if. when
describing its boundary, the domain is always kept on the left of
a person walking round the contour (see Fig. 1.32) and negative
if otherwise. If the area of a (nonoriented) domain G is equal to S

we assume that the area of the oriented domain is equal to S if the

orientation is positive and to —S if negative. It can bo shown that
a mapping x = x (£, q), y — y (£, q) of a domain T on a domain G
preserves the orientation if its Jacobian is positive and changes the
orientation to the opposite if ^ < 0. Therefore the formula
D <6 *’I) . .
expressing the area of the oriented domain G in curvilinear coordinates
is of the form

r
(i.e. the sign of modulus has heen omitted), and formula (1.74)
changes similarly.
2 Triple Integrals
and Multiple Integrals
oi‘ Higher Order

In the foregoing chapter we introduced the notion of a double

integral. Here we are going to define the integral of a function of
three independent variables, the so-called triple integral. Like
double integrals, triple integrals are widely applied to various
physical and geometrical problems. Some of the problems will be
considered in § 3.
Triple integrals are in many respects almost completely analogous
to double integrals and therefore we shall omit those proofs which
do not essentially differ from the corresponding proofs of the theory
of the double integral.
In § 5 of the present chapter we shall discuss the concept of mul­
tiple integrals of higher order, that is integrals of functions depen­
dent on an arbitrary number of arguments.

§ 1. DEFINITION AND BASIC PROPERTIES

OF TRIPLE INTEGRAL
1. Preliminary Observations. Volume of a Space Figure. The
definitions of an interior point of a domain, a boundary, a closed
domain, a diameter etc. given in § 1 of Chapter 1 for the plane are
transferred without any changes to the case of the three-dimensional
space.
When introducing the double integral we use the notion of area.
Similarly, the definition of the triple integral is based on the notion
of the volume of a space figure, a solid.
The reader is supposed to be familiar with the definition of the
volume of a polyhedron known from elementary geometry. The
extension of this notion to a wider class of figures can be performed
in the same manner as it was done in § 1 of Chapter I where the
notion of area was extended from polygonal figures to curvilinear
squarable figures. Here we shall briefly present the corresponding
arguments.
The volume V (P) of a polyhedral solid (polyhedral space figure),
i.e. a space figure composed of a finite number of polyhedrons, is
a nonnegative quantity possessing the following properties:
80 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

(monotonicity). If P nnd Q arc two polyhedral solids and P is

1
contained in Q then
V ( P ) ^ V (Q)
2 (additivity). If P and Q are two polyhedral solids without inte­
rior points in common we have
V { P f- ()) = V(P) -f- V( 0 )
3 (invariance). If polyhedral solids P and (J are congruent their
volumes are equal.
These three properties should be preserved when the concept of
volume is extended from polyhedral solids to a wider class of space
figures.
Take an arbitrary space figure* (I> and consider all the possible
polyhedral figures embedded in it. The least upper bound of their
volumes is referred to as the interior (Jordan) content of the figure
(if there is no nondegeiierate polyhedral figure that can he embedded
in the solid d> we attribute, by definition, a zero interior content
to <b). Similarly, the greatest lower bound of the volumes of all
polyhedral solids enveloping the figure <-L> is called its exterior (Jor­
dan) content. If the exterior and the interior contents of a figure <T>
coincide their common value is said to be the volume of <t>. The
following theorem is proved after a manner of Theorem 1.2:
Fh core at For a space figure to have volume it is necessary
nnd sufficient that for every r >> 0 there exist two polyhedral figures
I* CZ <I> and Q :z> such that
V (Q) — V (P) < c
We say that a set is of volume zero if it can be embedded in a pul>-
hedrcal solid of arbitrarily small volume. Using this notion we can
rephrase theorem 2 . 1 as To Hows:
For a space figure <b to have volume if is necessary and sufficient
that its boundary have zero volume.
This criterion enables us to establish the existence of volume
for some sufficiently wide classes of figures. For instance, the solids
composed of a finite number of curvilinear cylinders ha vi ng squarable
lower bases and bounded above by surfaces defined by equations
of the form z — f (x, y), where / (x, ?/) is a continuous function,
form such a class. The volume of each cylinder is given by the double
integral
\ ( / (*, !/) dx dy
G
taken over the base of the cylinder.
* Tlinl nri of pof « i
 CH. 2. T R I P L E IN T E G R A L S 81

Another important class of such figures consists of space figures

bounded by a finite number of smooth surfaces.* The proof of the
fact that a solid bounded by smooth surfaces has volume is almost
completely analogous to that of the fact that a smooth curve is
of area zero, but since this proof involves some more complicated
details we shall not present it here.
Repeating the argument given in § 1 , Sec. 4 we can easily establish
the validity of the following assertions:
(1) Let dJj and <X>2 be two space figures having volume. Then their
union O also has volume. and if the figures <Th and (1>2 have no interior
points in common the volume of <1 * is equal to the sum of the volumes
of (l>i and <P2.
(2) The intersection (the meet) of two figures having volume also
has volume.
Note. We should pay attention to the fact that we deal with two
different approaches to the concept of volume.
On the one hand, we have defined the volume of a curvilinear
cylinder with a squarable base G bounded above by a surface z —
= / (x, y) as the double integral
j j / (ar, y) dx dy
G
On the other hand, the concept of volume of a space figure has been
introduced by approximating the figure with embedding and
enveloping polyhedral figures. Hut it can be shown that these
approaches are equivalent for a sufficiently wide class of figures,
in particular, for the figures bounded by piecewise smooth surfaces.
2. Definition of Triple Integral. Let a bounded function / (.r, y, z)
he defined on a space figure V which has volume.** Break up V into
parts Vt, choose an arbitrary point (£*, ijt-, in each F; and form
the integral sum
T - £ / ( 6 j. >li. (2 . 1 )
i=l
where At?f is the volume of the element (subdomain) F, and the
sum is extended over all the elements of the partition. Let us intro­
duce the following definitions.
D e f i n i t i o n /. Let D be the maximal of the diameters d ( V,) of the
elements F* into which t.hc figure is divided (the f i n e n e s s of the par-

* A surface is said to he smooth if the tangent plane exists at each of its

points and if the position of the tangent plane varies continuously as thr point
moves on the surface.
** In what follows we shall always suppose, without any further stipula­
tion. that ihe space future* in question have volume.
6 —0824
82 M UL TIPLE INTEGRALS. FIELD THEORY AND S E R I E S

titiori). A number J is said to be the l i m i t o f i n t e g r a l s u m s (2.1),

as D —►0, if for every e > 0 there is 6 > 0 such that
IT - J |< e
if D < 6 .
In other words, the inequality | T — J | < e must hold for
every integral sum T associated with any partition {F*} for which
D < 6 irrespective of the choice of the points (£,, r\iy £,) in each F/.
D e f i n i t i o n 2. If the limit of integral sums (2.1) for D —*- 0 exists
it is called the t r i p l e i n t e g r a l o f the f u n c t i o n f (x , y, z)
o v e r V and denoted as

[ f j f(x> !/, z) dv or C^ f {x, y, z) dxdydz

* V J V

In this case the function / (x, y , z) is said to be intcgrable on F.

3. Conditions for Existence of Triple Integral. Integrability of
Continuous Functions. As in the case of a function of one or two
independent variables, an arbitrary bounded function / (i, y. z)
is by far not always integrable. To establish sufficient conditions
for the existence of the triple integral we shall use upper and lower
Darboux sums which are applied to single and double integrals.
Let / (x, j/, r) be a bounded function defined on a figure F and
{F, } be a partition of the figure. Denote by jl/j and w* the respective
upper and lower bounds of the values of the function / Or. //, z)
on Vi. The expressions
n n
2 M\\v-t and ra{Ai><
i= I 1= 1

(where Au* is the volume of Vi) are called, respectively, the upper
and the lower Darboux sums for the function / (x, z) associated
with the partition {F,} of the figure V. All the properties of the
upper and the lower Darboux sums given in § 2 of Chapter 1 for two
arguments arc completely transferred to the case of three arguments.
The following necessary and sufficient condition for the existence
of the triple integral is proved by applying arguments similar to
those of the proof of Theorem 1.3:
T h e o r e in 2.2. A bounded function f (x, y t z) defined on a space
figure V is integrable on V if and only if for every « > 0 there is a
partition of the figure V such that the difference between the upper and
the lower Darboux sums for the function f (x, i/, z) which correspond
to the partition is less than e.
This criterion implies the following theorems similar to Theorems
1 . 4 and 1 .4 ' proved for double integrals.
 CH. 2. T R I P L E I N T E G R A L S 83

T h e o r e m 2,:t. Each continuous function f (x, y, z) defined in

a bounded closed domain* V is integrable over the domain.
T h e o r e m 2,4. If a function f (x, y , z) is bounded throughout a
bounded closed domain, possibly except the points belonging to a set
of volume zero, the function f (x, y , z) is integrable on the domain.
4. Properties of Triple Integral. The basic properties of the triple
integral are completely analogous to those of the double integral.
We now enumerate them.
1-2 (linearity). If /, (x, y , z) and / 2 (x, y, z) are integrable over
a domain V and A:, and k z are constants the function k xf x -f- k ^ z
is also integrable on V and we have
\ \ j [> ./, (z, y , z) + A2/2 ( x , y , z) ]dy =

= kx j j j /, (x, y , z) d v + kz \ [ \ f2 {x, y , z) dv
" V v V*

3 {additivity). If V is the union of two space figures Vx and Vz

with no interior points in common and a function / (x, y . z) is
integrable over Vx and Vz, then / (x, y, z) is integrable over V and
j ( j / (*. z) dv = j j [ f (x, y, z) dv-'r j j j / {x, y , z) dy
vJ Kt* v*
4 (morcotometfy). If f\{x, y , z ) ^ f z {x, y , z) and both functions
arc integrable on P the inequality
JJJ/i (x, y , z) d y > jjj .V, z)dy
takes place.
5 (iestimation of the modulus of the integral). If f (x, y, z) is inte­
grable on V the function | / (x, y , z) | is also integrable and
ijjj
v
f{x, j j j (/(*» */» z ) \ d v
y,
Jv J
G {the mean value theorem). If a function / (x, y , z) is integrable
on V and satisfies the inequality m ^ / (x, y, z) M then
m yC j j j / {x, y , z) d y < A/y

where y is the volume of V.

* Hero and henceforward the terra “domain*’ is used as a synonym for a

space figure which has volume.
0*
84 M U L T IPL E INTEGRALS. FIELD THEORY AND S E R I E S

For the case of a continuous function the mean value theorem

can be rephrased as follows:
6 '. If a function / (x, y , z) is continuous and V is a bounded
closed connected domain, there is a point ( 5 , rj, Q in the domain V
such that
[ \ \ / (x, y, s) rfy = / ( il, s) v
V

5. Triple Integral as an Additive Set Function. By analogy with

a set function defined for plane figures, we can introduce the notion
of a set function whose argument is a space figure (domain). An
example of such a function (defined for each space figure that has
volume) is the volume of a domain. If the whole space or its part
is occupied by a substance we can associate with each domain the
mass contained in it and thus obtain a set function defined for domains
in space. Volume and mass possess the property of additivity which
is formulated in the same manner as in the case of plane figures:
a set function F (F) is said to be additive if, for any two domains
V\ and V2 having no interior points in common, on which F (F)
is defined, the value F (Fj + F2) is also defined and
F (Vt 4- F2) = F (F.) 4- F (F2)
If / (x, v» 2 ) is an integrable function the triple integral
!{*> !/* *)dv
*v
regarded as a function of its domain of integration is an additive
set fline lion (see property 3 in Sec. 4).
The notion of the derivative of an additive set function defined
in space with respect to volume is introduced by analogy with the
two-dimensional case. Namely, a number A is said to be the limit
of the quotient
F( F )
v
(where v is the volume of F). as F is contracted to a point M 0t
if for any e > 0 there exists 6 ^ > 0 such that

v
for every domain V entirely contained within the sphere of radius 6
with centre a I /V0. The limit is called the derivative of the junction
F (F) with respect to volume at the point A/ 0 and denoted by the
symbol
lini ‘F (F) or dF dv
V >A/h
 CH. 2. TRTPI.F. INTF.CiRAI.S 85

If F (V) is the mass contained in the domain V its derivative with

respect to volume (if it exists) is the volume density p (x, //, z)
of mass distribution in space.
From the mean value theorem for the triple integral it follows
that if the integrand is continuous the derivative of the integral
with respect to volume exists and coincides with the integrand:
/(* ’ s) do = f (x, ?/, s)
v
In this case the integral is the only additive set function in space
whose derivative with respect to volume is equal to the continuous
function / (x, y, z).

§ 2. SOME APPLICATIONS OF TIUPLE 1NT EG It AL

IN PHYSICS AND GEOMETRY
We now consider some typical problems involving computation
of triple integrals.
1 . Computing Volumes. If a space figure V has volume the triple
integral
f f \ c/x dy dz (2 .2 )
**V•' J

is equal to the volume. Indeed, each integral sum corresponding

to the integral equals the volume. Triple integrals are sometimes
more convenient than double integrals, for computing volumes,
because they enable us to put down the expression of the volume
not only for a curvilinear cylinder but. for an arbitrary solid as
well.
2. Finding the Mass of a Solid from Its Density. If we are given
a solid with a volume density p (x, y . z) of mass distribution which
is a continuous function the triple integral

Ivjj P ^ x% lJ ' d lJ d z

taken over the entire volume occupied by the solid gives the mass
of the solid. The derivation of this formula is completely analogous
to that of the formula for determining the mass of a plate from its
surface density.
3. Moment of Inertia. Performing the usual passage to the limit
from a system of mass points to a continuously distributed mass
we can easily derive the following expressions for the moments of
inertia of a solid with volume density P (x. y, z) about the coordi-
so M U LTIPI^ INTEGRALS, FIELD THEORY AND S E R I E

nate axes:
lx — \ \ \ (iji2 -f- z-) p {x, y , z) dx dy dz
V

ly = j \ j (x- r 22) P (-7 , y , 2 ) dx dy

r
=‘nV J + 2)p(*r*?/»s)
\ y (Jy i l z

The moment of inertia about the origin of coordinates is given

by the formula
h = \\ \ (* 2 \-y- + z*)p(x, y, z ) d x d y d z

A. Determining the Coordinates of the Centre of Gravity. The coor-

dinatesuf the centre of gravity of a solid with mass density p (x, y, z)
are expressed by the formulas
\ \ S *P (x, y, z) dx dy dz $ J $ yp (x, y, z) dx dy dz
' V '__________________ v _______________
Vc =
5 S J p (T' 2) r/v dz \ [ ? p (x, y, z) dx dy dz
*v ' * V '

£ J zp (x, y, z) dx dy dz
v '_______________
i J S P tx» 2) dx dz
v
which are received by means of the same arguments as in the case
of two dimensions. In particular, if the solid is homogeneous, i.c.
p (x. y. z) = const, the formulas for the coordinates of the centre
of gravity are simplified:
S$Sx dv SH y iiv SU zdy
2c
SSSrfi?
5. Gravitational Attraction of a Material Point by a Solid. Suppose
we are given a solid occupying a domain V and having a density
p (x, y. z) and a material point (lying outside V) of mass m with
the coordinates (x0. y$. z0). Let. us determine the gravitational force
with which the material point is attracted by the solid. Consider an
element of volume dv of the solid. The mass of the element is equal
to p (x. y, z) dv, and it attracts the material point with a force
whose numerical value is equal to
wp(x, y, z) dr
Y r*
 CH. 2. T R I P L E IN T E G R A L S 87

where y is the gravitational constant (whose value depends on the

choice of the system of units) and
r = V { x — *o) 2 H- (U— Vo) 2 -7- (s — 2 o) 2
the direction of the force coinciding with the direction of the vector r
joining the points (x0, i/0. z0) and (x, y , z). Consider the projection
of the force on the x-axis. It is equal to
y — t q ) w ip (x , y, z)dv ^2

^because the cosine of the angle between the axis and the vector r
is equal to —^J°j . To evaluate the projection Fx on the x-axis
of the force with which the entire solid attracts the material point
we in us L sum up elements (2.3), i.e. compute the corresponding
triple integral.
Thus, we have
yrnp (*, £/, 2 ) (x —x0)
dv

The other two projections are found similarly:

f
F„^ \ \ Y'"p {z- y' 5) i!l- t o ' A ,
«' J J
■yinp(x, y. z) (z — z0)
F .= dv

Note. It should be taken into account that, from the mathematical

point of view, the formulas obtained here and in § 4 of the preceding
chapter for similar problems an.- in fact Liu- dvdinitiuns of the corres­
ponding notions'(centre of gravitj’, moment of inertia etc.) for the
case of a continuously distributed mass. The justification of these
definitions docs not lie in logical arguments hut is based on the fact
that the results of the corresponding experiments are coincident
with those obtained by calculations performed according to the
formulas.
§ 3. EVALUATING TRIPLE INTEGRAL
As in the case of the double integral, the main technique used for
evaluating a triple integral is based on reducing the integral to
an iterated (repeated) one, i.e. on replacing the integration over
a space figure by successive separate integrations with respect to
each variable.*

* Here we menu the e.vail analytic computation of an integral. In practical

approximate calculations the reduction of a multiple integral to an iterated
one m rarelv applied.
88 MULTIPLE IN TEG R ALS, FIELD THEORY AN D S E R I E S

We shall first cousider a special case of the problem of reducing

a triple integral to an iterated one when the domain of integration

------------ Y
i Q
1
Lr— }------------ 1i
O. f1 1* ■ v1---■—
- >
P
Fiir. 2.1

is a rectangular parallelepiped with faces parallel to the coordinate

planes of a Cartesian coordinate system.
1. Reducing Triple Integral Over a Rectangular Parallelepiped
to an Iterated Integral. Consider a triple integral

jjj /(x , y, z ) d x d y d z
Q
whose domain of integration Q is a rectangular parallelepiped deter­
mined by inequalities
a x b, c y dy k ^ z ^ I
(see Tig. 2 . 1 ), the projection of the parallelepiped on the x. //-plane
being the rectangle P specified by the relations
a ^ x ^ by c ^ y ^ d
The following theorem takes place:
T h e o r e m '£.5. //, for a function f (x, y, z), the triple integral

( ( j f{Xy y, z)dv
‘Q
exists and if the integral
i
I (.3- * y) ~ ^ / (^» y\ *•) dz
h
exists for every fixed point (r, y) of /*, the iterated integral
i
I \ dxdy I / (Xy ijy z) dz
n n
 CH. 2. T R I P L E I N T E G R A L S 81

also exists and there is a relation of the form

i
/(•*» «/. z ) d u ~ f j dxdy [ / (x, y, z)dz (2.4)
JQ P h
The proof of the theorem is similar to that of the theorem on Ilit
reduction of a double integral to an iterated one (see Theorem 1 . 5 ).
It is based on establishing the fact that every integral sum corres­
ponding to the integral \ \ / (x, y) dx dy for an arbitrary parti-
p
tion lies between the upper and the lower Darboux sums associated
with the triple integral [ / (x, ;/, z) dv.
QJ
If we assume that the integral
a
J (x) = j / (x , y) dy

also exists for each fixed x from the interval a ^ x h we can

substitute, in formula (2.4), the successive integrations performed
first with respect to y and then to x for the integration over the
rectangle P. This enables us to rewrite equality (2.4) in the form
b il I
j ( j /(x , r/, z) dv = j dx j dy | / (x, y , z)dz (2.5;
Q a c h
It is this formula that reduces the evaluation of a triple integral ovei
a parallelepiped Q to successive separate integrations with respccl
to each variable. In the expression on the right-hand side of fonnu
la (2.5) the first integration is performed with respect to z, the seconc
with respect to y and, linally, the third with respect, to x. If wc sup­
pose that the integrals
b d
T i { y , z )“ j /( * , y , z)dx and / i ( z ) - j / i (y, z)clu
a c
exist we can derive the analogous formula
l d b

^ j j / (•*> i/, z) dv = j dz j dy j / (x, y t z) dx

Q h c u

Similarly, on condition that, the corresponding single and dmihh

integrals exist, we can establish analogous formulas reducing tin
triple integral to an iterated one taken with respect to x, y and :
in various orders. In particular, if the fund ion f (t . u. z) is ronti
AO MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

imous the triple integral and all the possible double and single
integrals are sure to exist and therefore when evaluating a triple
integral of a continuous function we can separately integrate with
respect to the variables x, ;/ and z in any orders and combinations.
In the general case of an arbitrary intcgrable function the orders are
not always interchangeable.
2. Reducing Triple Integral Over a Curvilinear Domain to an
Iterated Integral. We now consider a curvilinear domain F hounded
a Ido v e and below by the surfaces
z = Or, y) and s = z2 (x, //)
(z2 (x, i/) ^ 2 , (x, i/)) and on the sides bva cylindrical surface. Denote
by G Ilie projection of the domain I'" on the x, i/plane (Fig. 2.2).
A feature of such a domain is that each straight line parallel to the
\z=zz (x.{/)

z-axis and passing through a point of the domain has at most two
common points with its boundary. For brevity, we shall call such
a domain regular in the z-direction. Let a function / (x, y, z) defined
in Ilie domain V he inlegrable. Suppose that for every fixed point
(r. //) of G the single integral
22 (*, v)
j / (x, y, z)dz
Ti (ac. .v)
exists. Let us embed the domain V in a rectangular parallelepiped Q
determined by inequalities
a x b, c y d, k ^ z ^ /
and d e f i n e in Q an auxiliary function /* (x, //, z) I»y putting
/(x , i/, z) in V
 CH. 2. T R I P L E IN T E G R A L S 91

The function f*(xf //, z) is obviously integrable on Q and

(2 . 6 )

Applying formula (2 A) to /* (x, y, z) wo obtain

i

where P is the projection of Q on the x, {/-plane.

The function /* (.r, y, z) vanishing outside V, we thus have
^2 (*. y)
( 2 . 8)

Kxpression (2 .S) is a function of x and y which is identically equal

to zero outside the domain G. Therefore the double integral of the
expression taken over P (the projection of the parallelepiped Q
on the x, {/-plane) coincides with the integral of the expression taken
over G. Therefore, taking into account formulas (2 .6 ) and (2 .8 )
we can rewrite equality (2.7) in the form
V)
f (xy y , z)dz (2.9)
‘it*. lO
Hence, we have arrived at the following result:
T h eorem 2.6* If the triple integral

v
exists for a function f (x, y, z) defined in a domain F regular in
the z-direction and if the integral
*2 (*. {/)

exists for each fixed point (t, //) belonging to the projection G of
the domain V on the x, y-planey the iterated integral
!/)

O ;/)
also exists and euualitu rj.’H ha!da
92 MULTI P L E INTEGRALS, FIELD THEORY AND S E R I E S

The expression
*2 <*. v)
I{x, y ) = j / (x, y, z) dz
2 i(*. y)

is a function of two variables. If, for the function / (x, y. z) ami

for the domain G over which it is integrated, the conditions of Theo­
rem 1 . 6 are fulfilled, the double integral
^j / (x, y) dx dy
'gj
can be represented as an iterated integral taken, for instance, first
with respect- to y and then with respect to x. This results in the
rolat ion
>' 72(.T, y)
z)dv=\dx \ dy j / (x, -?/, z) dz (2 . 1 0 )
V o ykx) Zi ( x, u)

It is this final formula that reduces the triple integral to an iterated

one. The variables x, y and z can be interchanged if the corresponding
conditions hold. For example, we can reduce the triple integral
to an iterated one in which the integration is performed in a different
order, for instance, first with respect to x, then with respect to y
and, finally, with respect to z. In all the possible cases the limits
of integration with respect to a variable are dependent on those
variables with respect to which the integration has not yet boon
performed.
When deriving formulas (2.9) \vc have taken advantage of the
fa«'l that each straigh t lino parallel to the z-axi« and pacing through
a point of the domain G cuts its houndary at no more than two
points. If the domain is of a more complicated form then, to reduce
a triple integral taken over it to an iterated integral, we must break
up the domain into parts such that to each of them formula (2.9)
is applicable. We have already encountered such a situation in the
case of a double integral.
r*sow. .summing up, we can briefly formulate the rule of reducing
a triple integral to an iterated one (for definiteness, we suppose that
the iterated integral is first taken with respect to z and then with
respect to the other variables).
1 . Bleak the domain over which the triple integral is taken into
subdomains such that every vertical (paiailel to the z-axis) straight
line passing through a point of a subdomain inis at most t wo common
points with its boundary. In what follows we mention only one
such subdomain.
2. Fix arbitrary x and y and consider the corresponding straight
line parallel to the z-axis. Let z, (x, y) and z* (x. y) be the z-coordina-
 CH. 2. T R I P L E IN T EG R AL S 9;

tes of the points of intersection of the straight line with the boundary
of the domain (subdomain) of integration. The expressions Zj (x , y)
and z 2 Or. y) should be taken as the limits of integration with respect
to z.
3. Take the function of two variables x and y obtained after the
integration with respect to z has been performed. The domain ol
definition of the function is the projection of the space figure \
(or ol‘ the corresponding subdomain of V) on the x, y-plano. Finally,
replace the double integral of this function of two variables by the
corresponding iterated integral following the rules described in
§ 5 of Chapter 1.
The formula for reducing a triple integral to an iterated one is
essentially based on the process of grouping the summands which
lias already been dealt with. Instead of summing up the element?
of integration / (x, y , z ) dx dy dz in an arbitrary fashion l i e . evalua­
ting the integral S5J
J V
/ (x, y, z) dx dy dz j we first collect all th<
summands corresponding to one elementary cylinder with base in th<
vicinity of a point (x, y) (that is, we take the integral J / (x, y. z)dx)
then we add together the results corresponding to a section of tlu
domain V by a plane x ==■ const (which means that we compute
the integral \ dy ^ / (x, y, z) dz) and, finally, add up all the
y 'l(X )
quantities thus obtained that correspond to all the sections whicl
results in the formula
b V?(*) *2(x, 5/1

/(*> y> z ) d v = j dx \ dy j / (x, y, z) dz

V n yi\x) zi {x, y)

Exercise. Write the triple integral of a function /(x , y,z) takei

over the sphere (hall)
x 2 }- y2 + z2 ^ a2
in the form of an iterated integral.
A nswar.
a Va-—x* y a2 —x'- —
f (z, y, z) d v = ^ dz J dy \ f{x, y, z)dz
*9 ri/2+--'2 '5®2 ~a - J (i2 -x 5 - Ya*-xi-v*
§ 4. CHANCE OF VARIABLES IN TRIPLE INTEGRAL
We have alreadv% dealt with the method of changing ■» “ variable*
for the double integral (see § f> of (diopter 1). For the change ol
‘J 4 MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

variable in a single integral the reader can be referred to 181. Chap­

ter 6 , § 2. Here we shall consider the problem of changing variables
in Hie triple integral. The subject matter we are going to discuss is
in many respects similar to that of § 6 , Chapter 1.
1. Mapping of Space Figures. Consider two specimens of a three-
dimensional space. Introduce a Cartesian coordinate system x. y, z
in one of them and | , q, £ in the other. Furthermore, let V and Q
be domains belonging to the spaces, their boundaries being, respecti­
vely, some piecewise smooth surfaces S and 2 (Fig. 2.3). Suppose

that there is a correspondence between the points of the domains

which is one-to-one and continuous in both directions. The corres­
pondence can be expressed by means of three functions
x = X (£, rj, 0» y = y (£i n. t)t z = z (£, »1, £) (2.11)
or by the inverse functions
I ~ I (x* y, z), -- i| (a;, y, z), £ = &(■*, y, z) (2 . i 2 >
We shall consider functions (2 . 1 1 ) and (2 . 1 2 ) not only to be conti­
nuous but also possessing the continuous first-order partial deriva­
tives. Then the Jacobians
y> *> q n (1 D (Z- ^ £)
o 4 ^ (x* y . *)
exist and are continuous. Let each Jacobian be different from zero.
These conditions imply the relation
(j > .v« z) & (£» ii. £) = a fo i
»), £) * O (x, y , z) '
As in the two-dimensional case, we can show that under the mapping
specified by the correspondence which is determined by formu­
las (2 . 1 1 ) and (2 . 1 2 ) the interior points of one domain go into the
interior points of the other and the boundary points into the boun­
dary ones.
2. Curvilinear Coordinates in Space. Mapping (2.11) transforms
the domain £2 into V. Consequently, the specification of a point
 CH. 2. T R I P L E I N T E G R A L S 95

(|, q, £) belonging to Q uniquely determines the corresponding

point (a:, i/, z) of V. In other words, the quantities j\ and £ can
be regarded as coordinates (different from Cartesian ones, in the
general case) of the points of the domain V. They are called curvili­
near coordinates.
Consider, in the domain Q, a plane determined by the relation
| = E0, i.e. a plane parallel to the coordinate plane r|. £. Under
mapping (2.11), the plane goes into a surface lying in the domain V.
The Cartesian coordinates of the points lying on the surface are
expressed by the formulas
X = x (So, T], £), y = y (£0. *1. C). z = *(lo* 0* (2.14)
Making t 0 assume all the possible values \vc obtain a one-parameter
family of surfaces, the parameter being c. The planes tj = const and
£ = const are similarly mapped onto two families of surfaces lying
in the domain V. These three families form the set of the so-called

coordinate surfaces. On condition that mapping (2.11) is one-to-one

only a single surface belonging to one of the families passe* through
every given point of the domain V.
3. Cylindrical and Spherical Coordinates. We shall consider two
curvilinear coordinate systems in space which are used most fre­
quently, namely, cylindrical and spherical coordinates.
(a) Cylindrical coordinates. Let us specify the position of an arbit­
rary point M in space by means of its Cartesian coordinate z and
polar coordinates r, <p of its projection A/, on the x, z/-plane (see
Fig. 2.4). The quantities r, cp, z are referred to as the cylindrical
coordinates of the point A/. The figure directly implies that they are
connected with the Cartesian coordinates of the point ftt by the
relationship
x ~ r cos q>, y = rsin (p, z ^ z (2 . 1 ."»)

* (2.14) are the so-called parametric equations ot the suriace.

For a more detailed discussion of parametric equations of a surface see Chap­
ter 3.
•m MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

We have the following three families of coordinate surfaces correspon­

ding to the cylindrical coordinates:
(a) the cylinders r = const ( 0 ^ r <C oo).
((1 ) the vertical half-planes = const ( 0 ^ <p <C 2n),
(7 ) the horizontal planes z — const ( —0 0 < z <C 0 0 ).
The Jacobian of the transformation from the Cartesian coordinates
to the cylindrical ones is equal to
cos <p sin (p 0
i )(x, y. s)
— r sin tp r (os (p 0 = r ( 2 . 10)
U{ r, <(, z)
n 0 1

Formulas (2.15) expressing the relationship between Cartesian

and cylindrical coordinates determine a mapping of the domain
0 ^ r <T 00, 0 ^ <p <C 2ji, — 0 0 < z < 0 0 (2.17)
lying in the r, tp, z-space onto the entire x y jyt z-space. Under the
mapping, to each point (0 , 0 , z0) there corresponds an entire half­
segment of the form
r — 0, 0 ^ (p < C 2 j i , z — Zq

belonging to domain (2.17). Therefore the mapping is not one-to-one

at the points lying on the 2 -axis. But it is obviously one to-one
at all the oilier points of the x , y , z-space.
(b) Spherical coordinates. Let the position of the moving point M
in space be determined by the following three quantities:
(a) the distance p from the origin O to .17,
(|1) the angle 6 between the line segment OM and the direction
of the positive half-axis z,
(y) tin* angle (p formed by the projection OMx of the line seg­
ment OM 011 the 2 :, i/-plane and the positive direction of the x-axis

(see Fig. 2.5). The quantities p, 0 and cp are called the spherical
coordinates of the point M. The figure shows that the relationship
between the Cartesian coordinates of 1 lie point M and its spherical
 Clf. 2. TIIIPI.E INT EGRALS !)7

coordinates is expressed by the formulas

x = p sin 0 cos q , y = p sin 0 sin (p, z = p cos 0 (2.1 *S)
The three families of the coordinate surfaces corresponding to spheri­
cal coordinates are:
(a) the spheres p = const (0 ^ p <C o°),
((■&) the conical nappes (semi cones) 0 =- const (0 0 ^ n).
(y) the vertical half-planes (p = const (0 ^ q <C 2n).
'file Jacobian of the transformation of Cartesian coordinates into
spherical coordinates is equal to
sinOcosfp sin Bsin <p cos 0
//» ") peosOcoscp pcosOsinq — p sin 0 —o- sin 0
/ / ( p, <|, 0 )
—p sin 0 sin (p psinOcoscp 0
( 2 . 10)
Formulas (2.18) determine a mapping of the domain
—,I
(,4h seiiii-inlinitc rod”) of the p, 0, <p-space onto the entire x. y. z-space.
I.ike the mapping corresponding to cylindrical coordinates, this
mapping is one-to-one everywhere except the points lying on the
z-axis. Indeed, each point (0, 0, z0) is the image, under tile mapping,
of the half-segment p = z0, 0 « 0, 0 ^ ip < 2n for z0 > 0. and of
the half-segment p --=zti, 0 = n, 0 -sC ^ < 2.i for 2 O< 0 , and the
entire rectangle p — 0, 0 ^ 0 ^ jt . 0 ^ <p <C 2;t is mapped into
the poinl (0, 0, u).
4. Element of Volume in Curvilinear Coordinates. Let us finil the
expression of an element, of volume in curvilinear coordinates.
Consider again a space figure V in which some curvilinear coordina­
tes c. i). £ are introduced, and let the formulas
x x ( |, ip 0 , y — y (£, ip £), z = z (s, ip £) (2.20)
express the relationship between H, q, £ arid the Cartesian coordina­
tes x. //, z. The functions x (£, ip £), y (£, q, £) and z (|, ip £) are
supposed to he continuous and possess continuous derivatives.
W e also impose the condition that the Jacobian should he different
from zero.
Consider three pairs of coordinate surfaces drawn infinitely ciose
to each oilier. Let the first pair be given by fixed vaiues ol the first
coordinate which are equal, respectively, to c and | — <l\. Simi­
larly. let the second pair he specified by values q mid q --- c/ij of
the second coordinate, and the third one by values £ ami t -- dZ
of the third coordinate. These three pairs oi surlaccs cut out of spare
an infinitesimal curvilinear parallelepiped. We shall evaluate its
volume <h' mwjlortinnr Ipo form* of thr «econd ..ltd hi^lui order *>1
7-082;
98 MULTIPLE INTEGRALS, FIELD T H E O R Y AND S E R I E S

smallness relative to the volume. This parallelepiped coincides,

to within infinitesimals of order higher than the first, with the
rectilinear parallelepiped whose edges are the vectors P l \ . PP*
and P P 3 (Fig. 2.6). It can easily be shown that the vectors have
the following coordinates:

P/>1 = I 0y Jt
I r*
4
- g r d& -& * ) •
p p t = 1f t - * dy J*
an *1-
p p ,= \ J?Ldt
dt, a
where wo have again restricted ourselves to the principal terms
(i.e. of the first order of smallness). As is well known, the volume

Fig. 2.6

of a parallelepiped constructed on three vectors is equal to the

absolute value of the determinant with the coord in ates of the vectors
as its*'elements. Consequently, we have
f)rr dz Oz
- t dz dt
~w
dx dz dx dtt dz
dv — 0r\ dr\ dt di\ dZ.
-g -* ! ~dl] dr) dr\ dr\ dr)
dx dz 0ydx dz
"W OK ~ w
where the sign -j- or — is so chosen that the whole expression
is positive. Thus, we see that
d v = | / ( | , 7], g) \d%dy\dl (2.21)
where J (|, q, y. z) is the Jacobian of transforiun-
w ‘ D (I* 0
lion (2.20).
5. Change of Variables in Triple Integral. Geometric Meaning of
the Jacobian* We have shown that the volume of an infinitesimal
element is expressed in curvilinear coordinates by formula (2.21).
Ji follows directly that the velum** of ?> ftnito domain V enn he
 CH. 2. T R I P L E IN T E G R A L S 9$

written as the triple integral

T), £)|dg<hi<iC (2.22)

taken over the range 12, of the variables £, r\ and £t which is mapped
onto the domain V under mapping (2.20).*
On the basis of the expression for the volume we can derive the
formula of changing variables by means of the arguments given
below which are similar to the ones presented in § 6, Sec. 6 of
Chapter 1.
Let / (t. i/, z) be a continuous function defined in a bounded closed
domain V. Under these conditions, the integral
y , z)dxdydz (2.23)
v
exists and is equal to the lim it of integral sums of the form
n
2 /(* ,, m> Zi)&vt (2.24)
<—l
Suppose formula (2.20) establishes a correspondence between
the points of the domain V and a domain £2 which is the range of the
variables E, t] and £. Besides, let the correspondence satisfy the
conditions enumerated in Sec. 1. This correspondence attributes
to each partition {K,} of the domain V into parts Vt a certain
partition (£2.) of the domain £2 and vice versa. According to
(2.22), the volume A o f the subdomain V* is representable in
the form

QI
Applying the mean value theorem to the integral we receive
(£*» vft £i) I
where Aw* is the volume of the subdomain £2/ and (|f, itf, £?) is
a point belonging to Q4.
Each point (xf-. y,-, zt) entering into sum (2.24) can be chosen
quite arbitrarily within the corresponding subdomain Vt. In parti-

* Here we have left out the calculations which are similar to those written
in full in § G of Chapter 1 for the case of two independent variables.
If the reader carefullv studies Theorem 1.7 it will not be difficult for him
to prove formula (2.22). Here, as in the case of dimension 2, the basic idea
lies in approximating a nonlinear mapping of a small domain by an appropriate
linear mnppint?.
7*
100 MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

cular. we can take, as (x,. i/iy zr), the point whose curvilinear coordi­
nates are £*, i|* and £f. Therefore integral sum (2.24) can be- rewrit­
ten in the form
V /( x ( c ? , ii?, tf). */(£?. *nf, ;?), (It, ilf, it)) IJ (ff. ’I? it) | Aw,
(2.25)
that is as an integral sum corresponding to the integral
U j / ( * (I. n, t). y (I, ,|. S). S (I, Ti, 0 ) | j (I, T), Q I rf|rfnd i (2.2«>)
‘Q
The last integral is sure to exist because its integrand is continuous,
t'.onsider a sequence of partitions { V i } of the domain V which are
infinitely refined. Correspondence (2.20) determines a mapping
under which this sequence goes into a certain sequence of partitions
{Q,} of the domain Q, and since the maximal of the diameters of
Ilie subdomains F, tends to zero, the same is with the maximal of the
diameters of the subdomains Qf. The sequence of partitions of the
domain O generates the corresponding sequence of integral sums
each of which can lie put down both in form (2.24) and (2.25). The
limit of the sequence of integral sums of form (2.24) is equal to
integral (2.23) while the limit of sums (2.25) is equal to integral (2.2(>).
'Thus, integrals (2.23) and (2.2f>) are the limits of the same integral
sums and hence they coincide, that is
?/, z)dv =
t *
V

= ( ( f / (.<• a . .1, O. II (l, t\, t). : (E. r\. 1)) IJ (E, t), E) ] d»t (2.27)
»*J
Q
Consequently, if there is a one-to-one continuous and continuously
d iffererit iahle mapping of a hounded Hosed domain V onto a domain
12, with a nonzero Jacobian, and if / (x, y y z) is a continuous function
defined in the domain F, formula (2.27) for changing variables in
the triple integral is true.
We can easily show that the formula not only holds for a conti­
nuous function f but also for any hounded function continuous every­
where in V possibly except a set of points of volume zero.
Let us come hack to formulas (2.20) determining a correspondence
between the range V of the variables x, y, z and the range Q of the
variables c. q. t. The correspondence transforms an infinitesimal
rectilinear parallelepiped
lo ^ £ < Co I d l y ilo ^ n < ho 4- dr), t () < t < t n — r/£
of volume r/t»> = dc di) dt, lying iri Q into a curvilinear parallelepi­
ped specified hv Ilie same inequalities and belonging to V. The
 C il . 2. T R I P L E IN T EG R AL S 101

volume of the latter parallelepiped is equal to

dv = | J (£, q, t) | di c?q dt> (2.28)
Tluis, the absolute value [ / (c, q. £) | of the Jacobian in the
ratio of infinitesimal volumes corresponding to each other under
mapping (2.20) (see Fig. 2.7).
In some simpler cases the Jacobian associated with a change of
variables can be found on l-lie basis of expression (2.28) for an ele­
ment of volume, i.e. by means of purely geometric considerations

without computing the corresponding derivatives and the determi­

nant. We shall illustrate this technique by the examples of cylindrical
and spherical coordinates.
Cylindrical coordinates. Consider an element of volume contained
between three pairs of coordinate surfaces drawn infinitely close
to each other, namely the cylinders of radii r and r -p dr. two hoi i-
y.onla I planes correspondirig to certain values z and z -|- dz of the

coordinate z and two half-planes passing through the z-axis and

Forming angles and + dip with the x-axis. 'File volume (dement
hounded hv tin* surfaces is, to within infinitesimals of higher order
102 MULTIPLE INTEGRALS, F IE L D THEORY AND S E R I E S

of smallness, a rectangular parallelepiped with edges dr, dz and r dcp

(see Fig. 2.8). Its ^volume is equal to
r dr d<p dz
which implies that the Jacobian of transformation of Cartesian
coordinates into cylindrical ones is equal to r.
Spherical coordinates. Take a domain bounded by two spheres of
radii r and r -f- dr. two semicones determined by certain angles 0
and 6 dG (reckoned from the z-axis) and two half-planes forming

angles cp and cp -f dtp with the x, z-plane. The domain can be thought
of as a rectangular parallelepiped (to within infinitesimals of higher
order) with edges r dO, dr and r sin 0 d<p (see Fig. 2.9). Consequently,
its volume is equal to
r- sin 0 dr d0 dip
which shows that the corresponding Jacobian is equal to
rz sin 0

§ 5. MULTIPLE INTEGRALS OF HIGHER ORDER

1. General Remarks. The definitions and facts which have been
discussed in Chapter 1 for two variables, and in the present chapter
for three variables, can be transferred to the case of an arbitrary
number of variables. For this purpose w'c first of all define the volume
of an /z-dimensional parallelepiped.
As is weii known from analytic geometry, the area of a parallelo­
gram or the volume of a parallelepiped constructed on given vectors
 CM. 2. T R I P L E IN T E G R A L S 103

is expressed by the formulas representing it as the absolute value

of the determinant whose elements are the projections of the vectors
on the coordinate axes. Starting from these results, we define the
volume of an rc-dimonsional parallelepiped as the absolute value
of the determinant with rows (or columns) formed of the coordinates
of the vectors which are the edges of the parallelogram. Further,
based cm the volume of a parallelepiped, we can easily introduce
the concept of volume for polyhedral n-d intension;* I figures and then
define the volume for a wirier class of domains lying in on /?-dimcn-
sional space. After that the notion of an integral of a function
/ (X|. . . ., xn) dependent on arguments xx, x 2, . . . . is introduc­
ed as the limit of the corresponding integral sums. An /z-fold multiple
integral taken over an n-dimensional domain G is designated by
the symbol
^^ ^ / (^.li 3-2i * *•» ^n) d3Ti dx^ . . . (IXji
c
If the corresponding restrictions imposed on the domain G and
on the integrand are fulfilled, the tt-fold multiple integral can be
reduced to an iterated integral in which n successive integrations
with respect to each argument are performed:
t t - I f («3Tj, x%r «• •, «£n) dx\ dx 2 . . . dxn —
G
b x i2,(xj) , * n - i)
— ( dxt J dx2 . . - I /<* 2, ^2, . • ., :Cn) djCpi
a x*l><xj) Xnhxu . . ., * n - i >
'Hie formula for changing variables in the n-fold multiple intccrral
is analogous to the corresponding formulas for double and triple
integrals. Namely, if xt = xt (yit y z, . . y n) (i = 1, 2, . . ., n)
the integral is transformed according to the formula
f . . . | / ( 2*1» »• • » dx 1 . . . dx ^ —
-
c

= | • • • | / (^-i (yi» *• •» jn)» ♦• *! (l/i, yn)) X

^ (x jt • «», J n)
X dy± . . . dy n
where T is tlie range of the variables y t, . . yn.
2. Examples. All the basic facts of the theory of the double and
triple integrals remain true for the /i-dimensional case. Here we do
not dweii in more detaii 011 the general theory of /i-folu multiple
integrals and proceed to discuss some characteristic examples.
10't .ML'I.TIIM.E l . \ T E G H A l , S , FIELD THEUltY AND SK K IE S

(1) Gravitational attraction of two material t)odies. Although the

real physical space we live in has only three dimensions there are
various concrete problems involving multiple integrals of order
higher than the third. /Vs a simple example, let us derive the formula
for the force of mutual gravitational attraction of two finite material
solids. Let them occupy, respectively, domains G and G' and have
volume densities |» (x, y, z) and p' (xL y \ z') (these material bodies
lie in the same x, y , 2-space hut it is convenient to denote the coor­
dinates of their points by different symbols). According to Newton’s*
law of universal gravitation, the projection on the x-axis of the
force of attraction between two infinitesimal elements
ilv dx dy dz and dv' — dx dy' dz
of volume of the bodies is a quantity dFx equal to
y P------------
(*r* //, 2) p (x 1 y 1 z ) * /v 1 j . * / , /
*-5----— - ( x — x ) d x d y d z d x dy dz
* /
(2.2U)
where y is the constant of gravitation and
r y (x — x')2 + {y — y')1 j- (2 —z')-
To iind the total value of the projection F x of the force of interact ion
between the bodies we must sum up expressions (2.2D) over all the
volume elements of both bodies. In other wrords, the projection Fx
of the force of gravitational attraction between the bodies occupying
the domains G and G' is equal to
i* 1* i* i%
Y { x _ xt) dxd(fdzdx’ dy' d-J (2.20)
GxC
'I’fie other two projections are expressed similarly. Here the point
(x, jy, z) runs throughout the domain G and the point (.r', y \ z')
independently runs over the entire domain G’. Hence, integral (2.30)
is taken over a domain in the six-dimensional space of the vari­
ables x, //, z, x', y \ z'. Such a domain is usually denoted by the
symbol G X G’ and called the (Cartesian) product of the domains
O and <?'.
(2) Consider the integral
In = j^ . j dXi dxi . .. dxH (2.31)
G
taken over the domain G determined by the inequalities
x, ^ 0, x 2 0, . . ., x„ .> 0
x, xv ^1

* Mewtwii, I'.d.tk (1(142-1727;, tiic great English m athem atician ami phy­
sicist.
 Cil. 2. TH1P LE INT EU ItA LS u

deducing integral (2.31) to an iterated integral wo obtain

In ^ j . . . ^ dx i dxz . . dxji •"

c
1 l-xi . —x n~\
J dxi ^ dx2 . . .
o fl i dxn

Performing integration with respect to xn and substituting tli

corresponding limits of integration we derive
1 1—
Xi t—.Tfl—. . . —ln_2
I /i — | d x | ^ d x 2 . . . ^ (1 x i ■” . . . x n _j) dxi\ _j
0 0 0
Next, integrating with respect to xn_| and substituting the limit
of integration we arrive at the formula
1 1—jci 1 —’*1 " . *• _3
in — j dxt f dxz . . . j (1—* i - —xft_2) i
ft 0 0
Continuing these successive integrations we finally obtain

0 —r,)B-1 i
/n dx 1 ~ ~nT
J
0 (« —Dl

(3) The volume of an n-dimensional sphere. An ^-dimensional

sphere (hall) of radius a with centre at the origin of coordinates is.
by definition, the totality of ail the points of the /{-dimensional
space whose coordinates satisfy the condition
x\ + xl -r — ~r a2
The volume Vn of such a sphere is equal to the integral

J |%. . . ^ dx j dx2 . . . dxn

The integral can he transformed in the following way. Pulling

x^ (iUi (i ~ 1, 2, . . ., n) we can write
10b MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

where Un is the volume of unit //-dimensional sphere (of radius 1).

Furthermore, since
Un = . j dxi dxn . . . dxn —
XJ*‘’X5"^' ** 1
1
= ^ dxn I *** I * ’ ’ <^Xn~i ~
-1 +**-!< I-*n
1 «-!
■— | ( I — Xn) d X fj | ^ d x i ». • d x n —i

~ 1 *j+. | Cl
I »-t
^Un~\ J\ ( 1 — arfi) 2 d xn *
-1
wo can put x n = cos 0 and thus receive
n
2
f/n = 2f/„_i f sin" 0 dQ (2.32)
•*
0
Now taking into account that L\ = 2 (because one-dimensional unit
sphere with centre at the origin is nothing hut the line segment
( — 1 .1 ), and the corresponding one-dimensional volume is its
longtli) we can find, in succession. 7 U?t and so on.** / = .

* H ne, when com puling the integral

dxj we put Xi - ( 1 — x£) 2 iji (i = 1, 2 , . . .

Xi| P••. . 271 | 1. ^
. . . . n — 1 ) and thus obtain the expression
n- I n - t

(l-j-J) 1 j ... J </ !) , . . . d s /„ ., ^ ( l - i « ) z ] ... j d x , . . .

V, •+&« - * * l + . . . + * n - l :Sl
... dx^. |. ””"" r«
An explicit expression fur Un can be obtained with tlio help of so-called
Filler's integrals (see Chapter 10, § 3 and, in particular. Example 3).
 Elements
of Differential
Geometry

In this chapter we shall apply the differential and integral calculus

to studying geometric objects, namely curves and surfaces. The divi­
sion of mathematics in which various types of geometric configura­
tion are investigated by means of mathematical analysis is called
differential geometry.
In the framework of our course we can only present the funda­
mentals of differential geometry which is an extensive branch of
mathematics closely related to mechanics, the theory of differential
equations and other branches of knowledge.

§ VECTOR FUNCTION OF A SCALAR ARGUMENT

1. Definition of a Vector Function. Limit. Continuity. It is con­
venient to define curves and surfaces by means of functions taking
vectorial values (briefly, referred to as vector functions). Therefore
we begin this chapter with a brief review of basic applications
of mathematical analysis to vector functions. We shall not go into
particulars because there are only a few facts here distinct from
those of the theory of scalar functions.
D e f i n i t i o n , Let. to each value of a variable / belonging to
an interval la, 61, there correspond a vector
r (t) — x (£) i + y (t) j + z (/) k (3.1)
Such vector is said to be a v e c t o r (v e c t o r - v a l u e d ) f u n c t i o n
of the scalar argument t.
A vector function r (2) can be given the following visual inter­
pretation. If the vectors r (/) corresponding to all the possible values
of the argument t are laid off from the origin of coordinates their
tips trace a curve (the graph of the vector function) which is called
the hodograph of the function r (f) (Fig. 3.1). If the argument t is
considered to be time the hodograph of the function r (*) is inter­
preted as the trajectory of motion of a point.
A constant vector
R = ai + b\ -f- ck
is said to be the limit of r(*) for t 10 if
lim I r (/) — R | = 0 (3.2)
t-*to
108 MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

whore | r (t) — 11 | is Llie length (absolute value, or modulus) of the

vector r (/) — 11. Condition (3.2) is equivalent to tlie three scalar
relations
lim x ( /) —a, lim y (/) —&, lims(<) c (3.2')

A vector function r (/) is called continuous at a point /0 if

Iiin r (f ) r (/0)
A vector function r (/) is continuous at a point /0 if and only if its
projections, i.e. the three scalar functions x (0, y {t) and z (/), are
continuous at tQ (prove it). The sum. the difference, the scalar and
vector products of continuous vector functions are again continuous
(show it).
2. Differentiation of a Vector Function. A vector function r(/) is
said to be differentiable at a point t if the limit
r(f --l-AO—r (/>
lim = lim to
Af-*0 A!->0
exists. The limit is called the derivative of the vector function r(l)
and is denoted by the symbols ^ , r'(/) or r (/). It can be easily
proved that the existence of r' (/) is equivalent to the existence of the
throe derivatives x'(/), y'(/) and z'(t), these quantities being con­
nected by the relation
r' (/) = x' (/) i y' (/) j z (t) k
The vector is directed along the secant .lf.l/j of the hodograph
t t * t dv • • «
of the function r(/) (see Fig. 3.2), and the vector ^ is in the direc­
tion of the limiting straight line to whose position the secant tends
(Ir
as the point M x approaches M. Hence, ^ goes along the tangent
line to the hotlograph at the point :V/.
From the point of view of kinematics, r ' (/) is nothing but the
velocity of a point whose law of motion is r — r (<).
The following rules for differentiation of a vector function take
place:
(1) if r ( / ) — const we have r ' ( t ) e s 0;
(2) (/rr(/))' — k r ’(t) where k — const;
(3) ( w ( / ) r ( t ) ) ' = u ' ( f ) r ( / ) -f z/(r)r ' (/) where u (/) is a scalar
function;
('•) ( r , ( / ) ± r 2 (I))' = r ; ( 0 ± r'2 (t);
(•->) ( r , ! / ) . r 2( / ) ) ' — ( r j ( / ) , r 2( / ) ) ! ( r , ( / ) . r,(0 );
((>) I r,{/). r2(/)r lr;(/). rzD)l + I r,(0- »*;(/)| (it is necessary to
preserve the older of the factors here);
 CH. 3. E L E M E N T S OF D I F F E R E N T I A L G E O M E T R Y H

(7) if r = r (/) and t = t (x), then

dr dr dt
di dt dx
which expresses Ihe rule for differentiation of a composite veeto
function.
The proof of the rules is left to the reader.
‘Hie following special cases of differentiation of a vector funclioi
should he noted:
(a) The derivative of a vector function of a constant direction. Lc
a vector r(/) have an invariable direction in space (i.e. only it

Fi g. 3.1

length may depend on /). Then the vectors r(f) and r'(/) are colli
near.
Actually, in this case the vector r (/) can be put down in the lorn
r (f) —u (t) e
where u (/) is a scalar function and e is a constant vector. Thei
\vr have rr(t) u' (/) e, that is r‘ (/) — “ r (/).
(h) The derivative of a vector function having a constant length.
If ! r (0 ! const the vectors r(f) and i'(/) arc* orthogonal (perpen­
dicular) to each other.

Fig. 3.2

Indeed, hero wc have (r(/), r'f/)) --const. After this relation

lias been differentiated we obtain
2 (!•(/), r '(/))^ 0 , that is (r (/), r ' ( / ) ) - 0
which is what wc set out to prove.
'Hie geometric meaning of the last relation is quite clear. If | r (/) | =
— /?, the luidogra ph o il he fund ion r(/) cut in-ly lies on the sphere of
radio* ti v?*h centre .it llu oiiyin. The tangent losuth a curse lies
11U m u l t ipl e in te g r a l s , fie l d theory and se r ie s

in tlie tangent plane to the sphere and is therefore perpendicular

to the radius vector r(z) joining the origin with the point of tan-
gency.
The differential of a vector function r (t) is the vector
dr = dx • i -|- dy •j + dz«k
In other words, we have
dr = x’ (0 dt *i + y' (/) dt • j z (t) dt • k = r' (/) dt
which means that the differential of a vector function is equal to the
product of its derivative by the differential (i.e. increment) of the
independent variable. As in the case of a scalar function, the diffe­
rential dr of a vector function differs from its increment Ar by
a quantity of an order of smallness higher than the first relative
to A/.
3. Hodograph. Singular Points. We have defined the hodograph
of a vector function r (t) as a curve which is traced by the terminal
of the vector r (Z), when t varies, if its tail is always kept at a fixed
point.
As has been shown, if r(Z) is a differentiable vector function the
vector r'(/) is directed along the tangent to the hodograph at all
points where r ' ( / ) 0. The points at which the derivative r'(Z)

does not exist or vanishes are referred to as singular points. Here

we give several examples of singular points (see Fig. 3.3). In the
motion of a point according to a law r = r (/) the trajectory may
be a “smooth*’ curve but the velocity v(£) = r' (Z) may tend to zero
as t — /0. Then the material point is in an instantaneous state of
rest at the point r(/fl) at the moment t — t0. A singularity of this
kind characterizes the motion itself but not the geometric curve
along which the point moves (Fig. 3.3a). In some other cases this
can he followed by a change in the direction of motion (i.e. the
ii*cjr,<,tr*rv may he brokoi> :>i thr* pojr»t Fi? Tn this
 CM. 3. E L E M E N T S OF D I F F E R E N T I A L GEOMETRY 111

case we have a singularity both of the motion and of the correspond­

ing geometric curve. It may turn out that the trajectory is broken
at a point r(/0) but the velocity r'(*) does not tend to zero in approa­
ching the point (see again Fig. 3.36). In this case the moving point
r(/) is subjected to an impact (when it passes through the point r(/0))
which instantaneously changes its velocity in a jump-like fashion.
Further, the trajectory of motion can have a cusp (Fig. 3.3c) and
then the velocity of the material point may either lend to zero
in the vicinity of the cusp or change jump-like. Finally, the function
r(0 may turn into zero for / = /0 without assuming nonzero values
for the subsequent values of / >* t0. Then the point has a state of
rest at the end of the motion (Fig. 3.3d). These and many other
singularities of motion may he of interest for studying some concrete
cases hut at the same time they can rarely be treated by means of the
general methods. Therefore in what follows we shall exclude such
singularities from our discussion and consider the motions for which
r'(f) exists everywhere and does not vanish.
4. Taylor’s Formula. For a vector function we have Taylor’s
formula
r(/ A/) = r(/)-j-r (/) A/-f--jp r (/) Ai2-F • . . -j- -^j-(r<rt> (0 -r«) A/'*
(3.3)
where a is a vector which lends to zero as At -*■ 0. In fact, applying
Taylor’s formula to each projection £(/), t/(/) and 2(0* of the
vector r (0 we obtain the relation
x (t At ) —-x{t) -F x' (0 At x" (0 A<*-j- . . . — (x<n) (0 -F a,) AJ"
lor x(t) and two similar relations for y(t) and z(t). Multiplying
these relations, respectively, by i, j and k and adding up the results
we arrive at formula (3.3).
Thus we see that most of the basic notions and rules of differential
calculus are transferred without essential changes from scalar func­
tions to the vector-valued ones. But it should be noted that such
conclusions cannot be drawn automatically because there are excep­
tions to the rule. For instance, the well known theorem on finite
increments (Lagrange’s theorem; c.g. see (81, Chapter 8, § 0) is n o t
true for vector functions. (Let the reader construct an illustrative
example.)
5. Integral of a Vector Function with Respect to Scalar Argument.
As in the case of a scalar function, we can form integral sums for

* Here we suppose, of course, that each projection x(t ). y(0 and z(0 of
the vector function r (t ) satisfies the conditions for the applicability of Taylor’s
fo rm u la t o tf.
112 ML'I.Tri*LK^I\TE(iUALS, FIELD THEORY AND S 13H1 ES

a vector function r (() defined on an interval a t b and con­

sider their lim it as the maximal length of the segments forming
the partition of the interval [a, 6] lends to zero. The limit is re­
ferred to as the integral of r (/) over the interval la, b\ ami is desig­
nated as
b

j r(l)dt
a
liy analogy with scalar functions, we can easily prove that tin*
limit (i c- the integral) is sure to exist if r (/) is continuous on [a, £>|.
The existence of the limit of a vector integral sum
n
y> r(T|)(/i — /»_,)
i=l
(whore a /0< / , < . . . < /M 6. lj_t ^ ^ ij) is obviously
equivalent to tin* existence of the limits of the three scalar integral
sums corresponding to the projections x(t), y(t) and z(t) of the
vector function r (/). and we have
!• 1/ b b
^ r (/) dt — i • Jx (0 dt -f- j • j y (I) dt -| k •f z (i) dt
u a u a

The ordinary properties of the integrals of scalar fund ions are

easily extended to the case of the integrals of \reelor function*15. For
example, we have
b b

^ u* (t) r (/) dt —u (b) r (b) — u (a) r (a) — J a (/) r' (/) dt

il II
which is tin* formula of integration by parts where u(/) i* a scalar
function. The formulas connecting integration with basic operations
of xector algebra are also easily established. For instance,
b h
j (e, r (t)\ dt=* [c, j r(/) f//J
fi 41

where c is a constant vector.

(». Vector Functions of Several Scalar Arguments. We can also
consider vector functions dependent, not on one but on many
scalar arguments (in particular, we shall encounter vector fmiclions
of (wo scalar arguments in the present chapter in the study of sur­
faces). The concept of a partial derivative and oilier concepts of
analysis are easily generalized to these functions.
 CH. 3. ELEMENTS OF D I F F E R E N T I A L GEOMETRY 113

§ 2 . SPACE CURVES
1. Vector Equation of a Curve. Vector functions of a scalar argu­
ment provide a convenient method of determining space curves.*
Indeed, suppose we are given a continuous vector function r (/)
(« ^ b). Then, after constructing its hodogrnph. we obtain
a space curve V- Conversely, if a space curve y is defined in a certain
way we can try to determine it by means of a vector function.
We say that a curve y is represented p a r a m e t r i c a l l y if there is
a one-to-one correspondence which attributes a certain value of
a parameter t belonging to an interval I<z, b] to each point of the
curve, the correspondence being cont i nuous at each point of the
interval.** The latter condition means that ihe distance between
the points r{/0) amt r(0 of the curve lends to zero if / —►t 0. If
a curve y is represented parametrically the radius vector of each
of its points is uniquely determined by the corresponding value of
the parameter t. that is
r = r(/) (r ~ x i -f y\ + zk) (3.4)
Relation (3.4) is referred to as a parametric (vector) equation of the
curve y . Vector equation (3.4) can apparently he replaced by the
three scalar parametric equations
x = x(t), y = y{t), z=z(t)
Applying the terminology of § 1 we can say that a parametric
equation of a curve provides its representation as the hodograph
of a vector function r (t).
In what follows we shall only consider the curves and their para­
metric representations for which the corresponding vector functions
r(/) are triply continuously differentiable.
k.vutuple. Let us put
r(/) = \a cos / + j a sin / 4 - k hi (3.5)
This para metric, equation determines a curve called a screw line
(circular helix; see Fig. 3.4).
When considering a curve we can introduce its parametric repre­
sentation in various ways. For instance, if a curve y is given by an
equation r — r (/). a sC t b, we can put
! = t (t), a ^ t ^ [5

w h e r e / ( r ) i s a m o n o t o n e f u n c t i o n s u c h t h a t t ’(x) > 0 , t ( a ) = a
a m t t ( p ) -= b. a n d r e g a r d t a s a n e w p a r a m e t e r p r o v i d i n g t h e e q u a ­
tio n r r (/ ( t ) ) for t h e c u r v e y.

* We do not •specify tin* notion of a curve here. A discussion concerning

this question can ho fnnmi. for instance, in |8|, Chapter I t , § 1.
** The condition that the correspondence is one to-one means that we are
considering the curves without points of self-intersection.
8-082’,
1 i-i MULTIPLE INTEGRALS. FIELD T H E O R Y AND S E R I E S

lu many cases it is convenient to take as a parameter the arc

length of the curve reckoned from a iixed point. The transition from
an arbitrary parameter entering into a parametric representation of
a curve to the arc length of the curve can be performed as follows.

Let y be a curve and t a parameter entering into its parametric

equation. Choose a point 3 / 0 on y which corresponds to a certain
value t = t0 of the parameter and consider it to be an initial point.
Next take an arbitrary point M on y . The length I of the arc M 0M
is expressed by the well known formula
t t
I = J x'2 y'2- rz '2 dt*, i.e. l = ^\ t ’ (t)\dt
to h
where t is the value of the parameter corresponding to the point M.
The formula determines / as a single-valued and continuous function
of t: I = / (£). If the function r(/) is such that r'(£) does not vanish
anywhere we have l'(t) =7^ 0 at all the points and consequently t can
he represented as a single-valued continuous function of I: t = t(l).
(On the existence of an inverse function of type t = t (I) see, for
instance, [8 |, Chapter 11, § 1.) Now putting r = r (t (I)) we thus
represent r as a function of arc length /, i.e. obtain a parametric
equation of the curve in which the arc length serves as a parameter.
Example. Consider again circular helix (3.5). We have
dl = a2 sin214 - a2 cos21 4 - b2 dt = V a 24 - b2 dt
* The formula means in fact that the curve x = x (/). y = y (I)i z = 2 (t)
is regarded as being a “broken lino" with an infinite number of infinitesimal
segments (rfx, d y , dz). The length of a single segment is given by the Pythago­
rean theorem and is equal to
V(dx)* -h (dV)2 -I- (^T2 = >V (I»* + W (05*~+ <2 '(0 ) 2 dt
The “sum” of the lengths of the "segments”, that is the integral
1
^ 4 / x '2 4 * y *2 z ' 2 dt
io
i.> i-q.ial L» Mic length of the r»irvo.
 CH. 3. E L EM EN TS OF D I F F E R E N T I A L GEOMETRY 115

for it and hence I = y a1+ bl t . Passing to the parameter I we can

rewrite the equation of the circular helix in the form
r (2)
W —i a cos —. 1 ■■•-{- j a sin y .a2_l_^2
y alu-b2^* * — h k b y _L= r
a2-{-^
Note. If a parameter t entering into an equation
r = r (*)
is thought of as time the curve determined by the equation can be
regarded as the trajectory of a point moving from an initial position
with the velocity r'(/). But a point can be in various motions along
the same curve because when specifying a curve we only prescribe
the direction of the velocity at each moment but not its numerical
value. In particular, we can consider the case when the velocity rr
of motion is all the time identically equal to unity in its modulus.
It is just the case when the arc length / is taken as the parameter t.
Indeed, we have dr = i dx -f- j dy -f- k dz and consequently
dr _ Y ( d x )* -b dl
dl dl ~ dl
(3.G)
Thus, from the point of view of kinematics, various possible ways
of parametric representation of a curve can be interpreted as the
corresponding laws of motion of particles tracing the same trajectory
with different velocities. Then an equation of the form
r = r (0
where I is arc length describes the motion of a particle with unit
(in its modulus) velocity.
2. Moving Trihedron. Consider a curve given by an equation
r = r (I) (3.7)
At each point M of the curve (corresponding to a value f), the unit
vector

determines the direction of the tangent to the curve. The vector

r= x
is orthogonal to x (because it is the derivative of the vector x having
*•
constant length; see Sec. 2 in § 1). After the vector rhas been divided

* Here and henceforward \vc denote the derivatives of r with respect to

• ••
arc length by the symbols r, r etc. and use the notation r ' , r" etc. for the deri­
vatives with respect to an arbitrarv paramotor.
8*
116 MULTIPLE INTEGRALS, FIELD TH E O R Y AND S E R I E S

••
by | r | we arrive at the unit vector

orthogonal to t .* Furthermore, lot us take the vector

P = [t, v] (3.D)
where the square brackets designate the vector product. The vectors
t , v and (1 form a triad of mutually orthogonal unit vectors which
is referred to as the moving (natural) trihedron of curve (3.7) at.

Fig. 3.5

the point XI (Fig. 3.3). The trihedron is thought of ns being rigidly

connected with the curve at each of its points and therefore the
shape of the curve can be completely characterized by describing
the motion of the trihedron in space as ils vertex moves along the
curve.
It s h o u l d b e n o t e d t h a t b e s i d e s r e l a t i o n (3-3) t h e v e c t o r s t , v , p
satisfy th e tw o sim ilar relatio n s

(v, P ) = t , IP, x ] = v

The vectors r, v and p determine, respectively, the directions of

the straight lines called the tangent, the normal (principal normal)
and the binomial to the curve at the corresponding point. The
vectors are referred to as unit vectors in the direction of the tangent,
principal normal and binormal.
3. Frenct-Serret Formulas. The motion of the moving trihedron
of a curve is specified by the velocities characterizing the rate of
change of the vectors x, v and P, that is by the derivatives of the
vectors with respect to I. Let ils find the derivatives.
We have already dealt with the derivative of the vector x which
••
is the vector r. Introducing the notation
A* — | r |

* The vector v is not. defined for the points where r = R. Such points (calk’d
points of rorfifioolinn' wiU La oyr'link'd from nnr consideration.
 CH. 3. EL EM EN TS OP D IFFE R E N T IA L G EO M E TR Y 117

we rewrite the derivative in the form

= kv t

where k is a nonnegative number.

We now consider the vector p. Its derivative p is perpendicular
to it as a derivative of unit vector. Furthermore, it is also perpendicu-
• • • •
lar to t . In fact, wo have p -- [ t . v| and hpiir.e 0 = lx. vl -4- lx, vl =
• •
-= [kv. v| — lx. vl - lx, vl, the last vector being perpendicular
to t. 'Plie vector p is perpendicular to (i and x and thus collinear
to v. Consequently, wc can put
P — XV
where x is a numerical coefficient.*
Finally, compute v. We have
V ~ IP, T) (p, t! ; [p, t] = [ - k v , x] j |P , f c v | = — fcx -f-x P

Thus, we have obtained the following formulas for the derivatives

• • •
t , v and p:
t— kx (3.10)
v - —kx j - x p ( 3 * 11)

P -x v (3.12)
They are known as the Frcncl-Sorrel** formulas. The formulas involve
two scalar quantities, namely k and x. The quantity k is called
the curvature (the first curvature) of t he curve and x is called its torsion
(or the second curvature). The geometric meaning of the curvature
ami the torsion will he discussed later. The reciprocals of k and x
are referred to, respectively, as the radii of curvature and torsion.
4. Kvaluating Curvature ami Torsion. We have, by definition,
k — | *r| (3.13)
Therefore, to compute the curvature of a curve r = r (/) it is sufli-
••
cient to find tin- vector r (/) and determine its length.
To lind the torsion x we take the equalities
• ••
r —t and r —&v

* 'flu* ciieftieienL x can 1m * positive, negative or zero. We use the notation

x instead of y. because Ibis will be convenient, for our further aims.
** Frenel, Jean Frederic (181G-1900) anil Serrct, Joseph Alfred (1819 1885),
Krone h ina111em ali t ia ns.
118 MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

and differentiate the latter once again with respect to I. Applying

formula (3.11) for v we thus obtain
r = kv — hrx -+* A*x|l
The last three relations imply that
(r, r, r ) = / r x (3.14)
m mm • mm
whence x = ^ r*
k-
r *, i.e.

Formulas (3.13) and (3.15) enable us lo compute the curvature k

and the torsion x if the arc length has been chosen as the parameter.
In case a curve is given by an equation
r = r (/)
where r (£) is a triply continuously differentiable function of an
arbitrary parameter I we can regard I as being a function of arc
length I which yields
dr dr dt d 2r d-r / r f U 2 dr d 2t
~dT ~dt d f ' dl’1 “ I T l 1 \ ~ d l ) ' d T d l1 ’
(3.16)
d3r d3r / dt \ 3 « d2r dt dH ^ dr d3t
~dP“ = dt* \"dT / - dt 2 ~dl d l2 r ~dX dl3

The first relation can be rewritten as

t — r
/ dl_
dl

It follows that
dt 1
(3.17)
d l ~ I r ' (o |

because |r | = l ^here we assume that t and I vary in the same

direction, i.e. > Oj . Further, taking the vector product of the
first two equalities (3.1G) we derive
p dr d-r i f dr d'h / dt \ 3
l ." d T ’ ~d l 2~ J I ~ d i ~ ’ ~dt* ] \ dl j

or. since ( % , -g£- | =■ W,

 CH. 3. ELEMEN TS OF D l F F E R E N T f A L G EO M ET R Y 119

Since |0 | = 1, it follows from (3.17) and (3.18) that

llrMO, r" (011 (3.19)
IT' (0 13
Substituting expression (3.16) into relation (3.14) wc receive

(r’ (<). r'(<), r -(0 )(-^ -)° = * * x (3.20)

The last two equalities imply the final formula for the torsion:
( r ' (Q. r* (Qt r * (Q)
iir/(o. ^ w ir 1
Exercise. Find the curvature and the torsion of the screw line
r = i a cos t j a sin / | k 6/
Note. Let us come back to formulas (3.10). They indicate that the
■
vectors r' and r* are linearly expressible in terms of the vectors r
•»
and r. In other words, the vectors r# and t ” lie in the same plane as
■ ••
the vectors r and r. The plane is called the osculating plane. Hence,
the osculating plane of a curve (at a given point) can be defined
as a plane containing the vectors r' (f) and r" (/) (irrespective of the
specific choice of the parameter). If I is interpreted as Lime and the
equation
r = r (0
as the law of motion of a point we can say that the osculating plane
is the one that contains the velocity vector and the acceleration
vector.
5. Coordinate System Connected with Moving Trihedron. For a
curve r {I), the three vectors r, v and 0 specify a coordinate system
(for which they are the base vectors) at each point M of the curve,
the system varying, in the general case, as the point moves along the
curve. The axes of such a coordinate system are:
( 1 ) the tangent (its direction i s determined b y the vector t ),
(2) the principal normal (its direction coincides with that of the
vector v),
(3) the binormal (which goes along the vector |1).
The coordinate planus of the system are:
(1) t h e p l a n e d r a w n t h r o u g h t h e p o i n t M p e r p e n d i c u l a r l y t o t
(i.e. t h e p l a n e c o n t a i n i n g t h e p r i n c i p a l n o r m a l a n d t h e b i n o m i a l ) ;
i t i s c a l l e d t h e n o r m a l p l a n e t o t h e c u r v e r — r ( f ) a t t h e p o i n t M.
(2) the plane passing through the point M perpendicularly to v
which is referred to as tlie rectifying p l a n e .
120 M U L T I P L E IN T E G R A L S . FIF.LD T H E O R Y AND S E R I E S

(3) tlie plane passing through the point M and perpendicular to {3

• ••
(that is the plane in which r and r lie). This is the osculating plane
we have already dealt with.
The disposition ol' these straight lines (axes) and planes is depicted
in Fig. 3.(i.
Problem. Write the equations of the tangent, principal normal
and binomial for a curve r r (/). and also the equations of the
normal, rectifying and osculating planes, at?a point r0 = r (/0).

Solution. The vector equation of a straight line passing through

a point with radius vector r0 in the direct ion of a vector a is written as
P — i'o I ?.a, — oo < X <C o o

where p is the radius vector of the moving point and /. is a para­

meter; the e q u a t i o n of a plane drawn through a point with radius
vector r0 perpendicularly to a vector n is of the form
( P — **0 , n ) - 0

This immediately implies the following equations:

p— j Ar0 (tangent line);
«*
p - *Vt Ar0 (pri nci pa I norm at);
• • •

P — ro 1 M ro, **oI (bin onnal);

(p — *0. 1*0 ) 0 (normal plane);
(P — r0) - - 0 (rectifying plane);
• « •

(p — r0, I'o. **ol) (osculating plane)

where r„ - r (/„), r0 r (/0) and r0 r (/0).
 CH. 3. E L EM EN TS OF D I F F E R E N T I A L GEO METRY 121

Exercises
1. Write the equations of the tangent, principal normal and
binormal to the curve
r = r (/)
Ilinl. Note that the vector [ r', r"l is in the direction of the hi normal
and the vector I r \ [ r \ r ' l l goes along the principal normal.
2. Write the equations or the normal, rectifying and osculating
planes for a curve r — r (/).
3. Put down the equations of the tangent, principal normal and
binomial and also of the* normal, osculating and rectifying planes
for the circular helix
x = a cos /, y — a sin /, z = bl
at the point t — 0.
(). The Shape of a Curve in the Yrieinity of Its Point. To investi­
gate the shape of a curve in the vicinity of its point we shall lake
advantage of the coordinate system determined by the moving
trihedron of the curve.
Suppose the derivatives
ro - r (/0), r0= r{/0) and r0 - r ( /0)

are different from zero at a point r0 - r (!0) ami expand the func­
tion r(/) in a neighbourhood of the point /„ by means of Taylor’s
formula:
_l (yq
r(') r0\ / o r0AP j 0 { M ">t A/ —I —t0
Now take the coordinate system specified by the moving trihedron,
i.e. choose the point r0 as tin* origin of coordinates and the tangent,
principal normal and binomial lines as the x-a.\is. y-nxis and z-axis,
respectively. By applying the Frenot Serret formulas for computing
•« « »•
the derivatives r and r we can substitute the following three scalar
equa li Iies
j =A/ — AP + 0(A/*) (3.23a)

,j = — k \ l - + — A P -rO (A/J) (3.23b)

3 I O (A/1) (3.23c)

* Tim symbol O [At*) designates a quantity of the order of A t 4.

122 MULTIPLE IN TEG R ALS, FIELD T H E O RY AND S E R I E S

for vector relation (3.22). Let us investigate the projections of the

curve on the osculating and the rectifying planes.
We take equalities (3.23a) and (3.23b) and lim it ourselves to the
principal terms. The equalities then take the form
x = AI, y = — A'A/2
Eliminating At from these relations we obtain the equation of
a parabola (Fig. 3.7):
y = Y- k*1

which is, to within the terms of the order of A/3, the projection of the
curve r = r (/) on the osculating plane. The curvature k being, by
definition, positive, the parabola opens upwards or downwards

Fig. 3.7

according to the unit vector v, and the greater fr, the greater the
rate at which the branches of the parabola are turning out of the
tangent in the direction of the vector v.
Consider now the projection of the curve on the rectifying plane.
Taking formulas (3.23a) and (3.23c) and again limiting ourselves
to the principal terms we obtain

x=*A I, z = — kxAP

Eliminating AI from these relations wo arrive at the equation

of a cubic parabola:
z = -JrA:xx3 (3.24)
The sign of the coefficient in x2 coincides here with that of the torsion
(because the curvature is always positive). The corresponding para­
bolas are shown in Fig. 3.8 for x ;> 0 and x < 0. The signs of the
coordinates x and y being determined, for small values of A/, by the
signs of the corresponding principal terms, it follows from formu­
la (3.24) that:
 CH. 3. E L E M E N T S OP D I F F E R E N T I A L GEOMETRY 123

1. In the vicinity of a point at which the torsion is different

from zero the curve lies on both sides of the osculating plane.
2. The greater the absolute value of the torsion, the greater the
rate at which the curve is turning out of the osculating plane.
If x ;> 0 the curve is turning out of the osculating plane, as I increa­
ses, in the direction of the vector 0, and in the opposite direction
if otherwise.
Problems
1. Show that a curve whose curvature is identically equal to
zero is a straight line.

2. Prove that a curve whose torsion is identically equal to zero

is a plane curve, that is lies entirely in a fixed plane.
Solutions *• •
1. If k as 0 we have r = 0, i.e. r — e = const which implies
r = r0 + Ze0, the last relation being an equation of a straight line.
2. If x = 0, the third Frenet-Serret formula indicates that 0 = 0,
i.e. p = p0 = const. The vectors r and p0 being orthogonal, we have

(p, r) = 0

Hence, since P = Po = const, we can write

4 ( f t » r) = °

Consequently we arrive at the relation (P0, r) = const which is an

equation of a plane.
7. Curvature of a Plane Curve. Consider a curve lying in a fixed
plane. Introducing Cartesian coordinates x , y in the plane we can
write the equation of the curve in the form
x = x (/), y = y (t), z == 0 (3.25)
124 MULTIPLE INTEGRALS. FIELD T H E O R Y AND SE R IE S

Computing the curvature of Ihe curve by means of formula (3.19)

we find
|x y - * V |
(3.26)
(x'2+ y'zf<2
But (e.g. see [81, Chapter 16, § 3) the curvature of a plane curve
is usually defined with the absolute value sign removed, i.e. with
the sign 4- or — attached to it. Then the expression for the curva­
ture is written in the form
(3.27)

The matter is that in the case of the plane, unlike the case of the three-
dimensional space, we can speak not only about the absolute value
of the rate of turning of the tangent but also about its direction
(i.e. the clockwise or the counter-clockwise direction). It is the

direction of turning of the tangent that is indicated by the sign

of quantity (3.27). A c u r v e is said to be concave up if expression
(3.27) is positive (see Fig. 3.Via) and concave down if otherwise
(Fig. 3.96).
8. Intrinsic Equations of a Curve. Formulas (3.13) and (3.15) make
it possible to find the curvature and the torsion of a curve, given
by an equation r = r (/). as functions of I:
k -= k (/), x = x (/) (3.28)
These relations connecting the curvature and the torsion of a curve
with its arc length are known as intrinsic (natural) equations of tho
curve. We can now pose the quest ion as to what extent intrinsic
equations (3.28) determine Ihe curve itself. It turns out that every
curve is uniquely determined, to within its position in space, by
its intrinsic equations.
Indeed, let us he given two curves y and y,. Suppose it is possible
to represent ihe curves parametrically by introducing the corres­
ponding parameters I and /, (their arc lengths) in such a way that
 cil. 3. E L E M E N T S OK D I F F E R E N T I A L GEOMETRY I!

their curvatures k, k { and torsions x, Xj coincide at the point

for which the parameters take equal values. In this case, for / = /
the relations
k(l) = A*1(/1)f x(/) = x ,^ )
hold, and we say that the curves y and y, have the same intrinsi
equations. Let us show that when Ihis is so, one of the curves ca
be made to coincide with the other if wo move it in space, as a rigi
body, in an appropriate way. Actually, apply a point A of th
curve y corresponding to a value /° of the parameter / to the poir
A, of the curve y, associated with the same value of (lie para nick
/, — 1°. Further, turn the curve y so that the unit vectors x, v,
of its moving trihedron at the point A coincide with the correspom
ing unit vectors x,. vt, p, of the moving trihedron at the point A
of the curve y,. Obviously, this can always he achieved. Then w
have
T" = xj, v ° -= vj, po p: (3.2(.
where superscript “zero” indicates tliat the vectors are taken at th
corresponding points determined by the common value /° =
of the parameters. Fslablishing the correspondence between th
points .1/ and JU, of the curves y and yj For which / = /, we ca
consider both curves to be represented parametrically with (lie hel
of the same parameter I and thus regard x,, v, and p, as funetioi
of I. Now take the scalar function
o(l) *-(x, T ,)-f(\\ v j -l-fp, p,)
and find its derivative with respect to /. Taking advantage of tl
Frenel Serrel formulas we obtain
-jf = k (v » Ti) + k (vi *T) + ( — ^‘T*1 y-P» v t) -r
-f(v , — Ax, \ xpi) — x(v , P,) —x(v,, p)=- 0
nnd thus a does not in fact depend on I. Equalities (3.211) imply tin
the value of o corresponding to I = l0 is equal to three. Cons
quently,
a (/) = 3
Each of the three summands entering into o (/) is a scalar produ<
of two unit vectors and hence cannot be greater than unity. Tl
total sum being equal to three, each summand is exactly equal
unity. But a scalar product of two unit vectors is equal to unit
if and only if the vectors coincide. Therefore we have
T — T,, V == V , , p = p,

for all /, that is the moving trihedrons «»f the curves y and y, coincii
not, only at (he initial poinl /„ hut «l«n fnr nil ilu* r*f (!
126 MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

parameter I. It follows that the curves themselves coincide because

*
a curve can always be reconstructed from the vector t = r (/), namely
i
r (0 = r (l0) H- J T (X) dk
h
The converse is also true because if two curves differ only in their
position in space, they obviously have the same intrinsic equations.
Now it seems natural to ask whether there exists a curve for which
two arbitrarily chosen continuous functions
k (I) (k (I) > 0) and x (/)
are, respectively, its curvature and torsion. It turns out that the-
answer to the question is affirmative but we shall not present the
proof because this would involve some notions of the theory of diffe­
rential equations that we are not going to study here.
9. Some Applications to Mechanics. Consider a material point
moving along a trajectory. If r (£) is the radius vector of the point
at moment of time t the equation of the trajectory is written in the
form
r = r (0
The derivative
dr /.v
-a —
is the velocity of motion of the point along the trajectory. Intro­
ducing the arc length as the parameter we can write
m_ dr _ dr dl dl
^ ~ ~ d f ~ ~ d l ~ d t — x dt
Since t is a unit vector, we have
dl
v - dt

and consequently the derivative expresses the absolute value

of the velocity.
The second derivative
d2r
W— ~at*
of the radius vector with respect to t is the acceleration of the
point. It can be represented in the form
d2r ( dl \ 2 . d*l
 CH. 3. EL E M E N T S OP D I F F E R E N T I A L G E O M ET RY \n

Applying the Frenet-Serret formulas we obtain

dH . . / di \2
W = T
£/r- + « * ( # ) ■
Thus, the acceleration is resolved into the sum of two components
t ^ and \ k . The former is in the direction of the tangent
and is known as the tangential acceleration and the latter goes
along the principal normal and is called the normal (or centripetal)
(£-1
acceleration. The tangential acceleration wx = x ^ c a n also be
written as \vx = t cU $ where v = dt — is the absolute value of the
velocity which means that the tangential acceleration is a measure
of the rate of change of the absolute value of the velocity v. The formula
wv = vk for the normal acceleration is well known from
elementary courses of physics. Namely, when a point moves in a circle
of radius R with a velocity v constant in its absolute value, v = | v j.
\
V
\

the acceleration is always directed to the centre of the circle and

1 1
its absolute value is equal to v2 where is nothing but the curva
ture k of the circle. Hence, the resolution
, d*l . . / dl \2
W = W , + W V = T * 5- + V * ( - 3 J - )

can be interpreted as follows: an arbitrary curvilinear motion is

resolved, at a given moment of time, into an accelerated motion
along the tangent (which results in the appearance of the term wt
in the acceleration) and a uniform motion in a circle of radius R =
with the speed v — — (which yields the term wv in the accelera­
tion). Hence, the point simultaneously takes part in the two motions
(see Fig. 3.10).
Problem. A mass point moves under the action of a central force,
i.e. one whose line of action always passes through a fixed centre.
Prove that the trajectory is a plane curve.
128 MULTIPLE IN T E G R A L S , K1E1.I> TTIF.ORY AND S E R I E S

Solution. Take the centre as tlie origin of coordinates. Let

r = r (/>
be the equation of the trajectory of motion. The force acting upon
the moving point is directed toward the centre. Consequently, accord­
ing to Newton’s second law. the acceleration, that is the vector
r" (l). has the same direction. Therefore the vectors r and r " are
col linear and hence the relation
(r, r \ r") - 0
holds at each point of the trajectory. Differentiating the triple scalar
product witli respect to t we obtain
-jf (r, r \ r") = (r\ r', r")d-(r, r'\ r")-f-(r, r \ 0 ^ 0
The terms (r', r', r") and (r, r", r") are equal to zero and conse­
quently
(r, r', r'") =0
The vectors r and r" being coll inear, we thus li ve
( r \ r ” , r~) — 0

for all /. It follows that * = 0. the last relation indicating that the
trajectory is a plane curve (see § 2. Sec. t>).

§ 3. PARAMETRIC EQUATIONS OF A SURFACE

1. The Concept of a Surface*. The present and subsequent sections
of tills chapter are devoted to tlie application of mathematical analy­
sis to studying surfaces.
The concept of surface, although clear enough from an intuitive
point of view, can he defined with various degrees of genera lily.
In matheinaticaI analysis we often consider surfaces represented
by an equation of the form
z f (x, y)
where / (x, if) is a continuous function defined in a domain G. A wider
class of surfaces is described by equations of the form
F (.r, //, z) = 0
For such ail equation to determine a surface, as we intuitively
understand it. it is necessary ilia I the function F (x. y. z) satisfy
some additional requirements.
The definition of a surface as a set of points whoso coordinates
satisfy an equation of the form z f (x. //) or F (x. //. z) — 0 is
sometimes inconvenient because it is closely related to the specific.
 <.H. 3. ELEMENTS OF D I F F E R E N T I A L GE O ME TR Y 129

choice of the coordinate system. Therefore we shall give a definition

of the concept of surface without involving a coordinate system.
First of all we introduce an important notion of a simply connected
domain.
Let G be a domain in the plane. We say that the domain G is
simply connected if the following condition is satisfied: every closed
contour L lying inside the domain hounds a (finite) part of the plane
entirely contained in G.
In other words, a simply connected domain is one without ‘‘holes’*,
livery closed contour lying inside such a domain can be continuously
contracted to a point without falling outside the domain.
A domain which is not simply connected is referred to as a mul­
tiply connected domain.
Fxamples of a simply connected domain arc tlie circle, the triangle,
the square and so on. An annulus, that is a part of the plane hounded
by two concentric circles, is an example of a multiply connected

Fig. 3.11
domain. Indeed, the part of the plane bounded by the contour L
shown in Fig. 3.11 is by no means a part of the annulus lying bet­
ween tie* circles ( \ and
tty a simple surface we shall understand a set of points in a three
dimensional space which is representable as an image of a hounded
closed simply connected domain under a one-to-one bicontinuous
(i.c. continuous in both directions) mapping. Further, the term
surface will he applied to every union of a finite number of simple
surfaces. This also includes the case of self-intersecting surfaces.
For instance, we can consider such geometric configurations as the
one shown in Fig. 3.12.
If f (x. y) is a continuous function defined in a bounded dosed
domain G the equation
r - / (a:. ;/)
determines a simple surface, fn fact, the mapping
(;t\ y) -<■- (x, //, / (x. //))
specifies a one-to-one correspondence, continuous in both directions,
between the points (x. y) of the domain G and the points (r. y , z)
whose coordinates satisfy the orrnMim> r — / (x, y) (rhccK ii up).
s- osj ;
130 MULTIPLE IN T E G R A L S . FI E L D THEORY AND S E R I E S

Practically, in what follows we shall restrict our consideration

to surfaces representable as a union of a finite number of simple
surfaces determined by equations of the form z = i (x. y). Hesides
the condition of continuity, we shall usually impose some require­
ments specifying the smoothness of the corresponding functions /

Fig. 3 .1 2

(the existence and the continuity of its lirsl or first and second partial
derivatives). Such conditions will be explicitly stipulated when
necessary.
2. Paramctrization of a Surface. Although in mathematical ana­
lysis we very often deal with surfaces defined by equations of the
form z = f (x, y) or F (x. y. z) = 0 it is sometimes more convenient
to determine a surface by means of parametric equations. To write
down a parametric, representation of a surface we first introduce the
notion of coordinates on a surface.
Suppose then? is a one-parameter family of curves* lying on a sur­
face Z. YVe shall sav* that the family* is regular
• if. for every• ngiven
point of the surface, there is one and only one curve belonging to

the family which passes through the point. If there are two regular
families on a surface such that each curve of one family has a single
common point (point of intersection) with each curve of the other
family and the curves are not tangent to each oilier at the points
of intersection wo say that there is a system of parametric (coord i-

* This moans that each curve of the family is characterized by a certain

vr.lttc r.f o sing) 0
 CH. 3. E L E M E N T S OK D1FFEKKNT1AL GF.OMETHY 131

natc) curves on the surface. Let the curves of one family be determin­
ed by the values of a parameter u (we call them u-curves) and the
curves of the other family l*y the values of a parameter v (v-curves;
see Fig. 3.13). By the hypothesis, fur every given point of the sur­
face, there is a single curve of one family and a single curve of the
other family passing through tire point, and therefore the position
of each point on the surface is uniquely determined by certain values
u0 ami t' 0 <>f t he parameters u and v corresponding to the curves.
The parameters u and v whose values specify the curves are called
(curvilinear) coordinates on the surface.
Note. In § 0 of Chapter 1 we introduced curvilinear coordinates
in a plane region (domain)- Here we have repealed the construction
hut applied it to a curvilinear surface in space. The introduction
of coordinates on a surface is obviously equivalent lo the specifica­
tion of a one-to-one continuous mapping of the surface onto a part
of the plane where the Cartesian coordinates // and v have been
introduced. The parametric curves forming the system of coordinate
curves on the surface are the images of the straight lines parallel
to the coordinate axes in the //, r-plane.
Exam pies
1. A torus (anchor ring) is a surface generated by the rotation,
in space, of a circle about an axis in its plane hut not cutting the
circle. The position oT a point on the circle can he determined by an
angle ff ( 0 ^ q <C 2j0 reckoned from an initial point. The position

Fi g . 3 . 1 4

of the circle itself can he specified by the angle of turn if reckoned

from its initial position. Thus, the position of a current point on the
Inrus is determined by the two angles ff and \|: independently varying
within (lie limits from 0 to 2 :c. The curves ff — 0 and if 0 of the
corresponding families of parametric curves are depicted in Fig. 3.14.
2. Let a surface he represented hv an equation z = f {x. //).
In other words, let Ihere lie a one-to-one correspondence between its
points and the points of its projection on an ,r. //-plane. The curves
whose projections are Ihe straight linos x — const and y --- const
form the corresponding families of coordinate curves on the
surface z ■■ f x , (/) (see Fig. 3.15).
0*
132 MUI-Tl PI.I' 1XIEGRAI.S. I I Ki n Til KORY AXP SKUIKS

ll is clear that (here arc various families of coordinate curves

that can be constructed on tin* same surface.

3. Parametric Equations of a Surface. If some coordinates // and

v are introduced in a certain way on a surface we can write a so-
called parametric representation of the surface corresponding to the
parameters it and v. Each point of the surface can lie determined by
certain values of the parameters u and v but at the same time it
can be specified by its Cartesian coordinates. Consequently, the
Cartesian coordinates of the points of a surface, on which some
curvilinear coordinates a and v have been introduced, are functions
of the coordinates in tlie u, p-plane:
x = x (n, i>), if y (u, r), z z (n, v) (3.30)
The three scalar equations can be replaced by a single vector
equation:
r = r (//, v) (3.3(0
where r ii i/j -•* zk. Equations of form (3.30) or (3.30') will
bo referred to as parametric equations of the surface.
A ote /. When we write a parametric equation of a curve the coordi­
nates :c, //, z are functions of a single parameter. An equation r -
— r (n. v) of a surface naturally involves two parameters since
a surface is a two-dimensional geometric configuration.
Note 2. An equal ion z —/ (a*, //) can ho considered a special case
of a parametric equation because, if we choose x and y as parameters,
we can write
r — xi |- r/j -f / (x, ;/) k
Exercise. Write the parametric equations of a torus (see example 1
in Sec. 2) in coordinates and ij*.
As a rule, in what follows we shall consider surfaces represented
by parametric equations. The function r (/:, v) will he supposed
 CM. 3. EL1C ME NTS OP D I F F E R E N T I A ! . GEOMETRY 132

to be continuous together with its partial derivatives of the first

order with respect to u and r. In § 5 and in the subsequent sections
of the present chapter we shall additionally impose the condition
of existence and continuity of the partial derivatives of the second
order.
4. Curves on a Surface. Consider a curve on a surface determined
by equation (3.30'). If a parameter t is introduced on the curve
then, to each value of /. there corresponds a point of the surface,
i.e. certain values of u and r. Thus, the coordinates u and v become
Functions of the parameter t when considered along the curve:
it - it <0, v = v (/)
These equations are the equations of the curve on the surface. Substi­
tuting them into the. equation of llie surface we arrive at a para­
metric equation of the curve on the surface:
r - r (u (/), v (/)) (3.31)
Conversely, substituting arbitrary functions of a single variable /
for the independent variables u and v into equation (3.30') of the
surface we obtain the equation of a curve lying on the surface.
lad us consider the tangent to curve (3.31). Its direction is deter­
mined by the vector
dr Or du dr dr
dt Ou dt ’ dv dt

which is a linear combination of the vectors 0u — and dn— called base

vectors (for the curvilinear coordinates in question). At each point,
the veetors are tangent to the coordinate curves passing through
the point.
1
We denote them, for brevity, •
as r„u = —
On
and rr — — dv
.
5. Tangent Plane. Consider all the possible curves lying on a sur­
face and passing through a given point M and the tangent vector'
to the curves at the point (see Fig. 3.1G). Each vector can he expres­
sed linearly in terms of the vectors ru and rD, and lienee it lies in
the plane determined by the vectors. This plane is said to he the
tangent plane to the .surface at the point M. Let us form the equation
of the tangent plane. The vectors r„ and rr lying in the tangent
plane, the vector IS |rfJ. rf.| is orthogonal to the plane, and hence
the sought-for equation of the plane is
ip - r, N) - 0 (3.32)
where r is the radius vector of the point of tnngeur.y and p is the
radius vector of the moving point in tin* tangent, plane.*
* I loro and henceforward we exclude from our consideration tin; points
nt M'l»?r*l» Ir r I H
134 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

Now suppose that a surface is represented by an equation z =

= / (^» y) which can be written in vector form as
r = ix jy r k/ (x, y)
Let us write the equation of the tangent plane to such a surface.
We have
r* = i + k/x, r ^ j + k f'v
and. consequently.
N = [r*, - i / i — j/i + k (3.33)
Substituting the vector i (x — x0) -f j (y — yQ) -f- k {z — z0) for
p — r inlo equation (3.32) of the tangent plane, and expression (3.33)

for the vector N, we obtain the equation of the tangent plane to the
surface z — / (x, y) at the point (x0. y0, z0):
z — z0 = /* (x — x0) -T f'v (y — yo) (3.34)
where the values of the partial derivative® /I apH fy are taken at the
point (x0, y0) (the projection of the point of tangency (x0, </0» So)
on the x, y- plane).
If a surface is determined by an equation F (x, y, z) - 0, defi­
ning z as an implicit function of x and iy, we can write
OF dF
dz Ox. dz dy
dx OF * dy OF
dz dz

Substituting these expressions for fx and fy into equation (3.33)

we derive the equation of Hie tangent plane to the sur­
face F (x, y, z) = 0 at an arbitrary point (x0, j/0, (where
F (x0. (jv, z0) = 0):
(x — x0) Fx -f- (y — f/o) F'v + (s ~ Zo) F'z^O
The values /'*, and F': eiiUu'ing into the formula are taken at the
of it llig o ic j i v-4- u- ;/u» - u/*
 CH. 3. ELEMENTS OP DIFFERENTIAL G EO ME TR Y 135

6 . Normal to a Surface. Consider a surface r = r (u, v). Let us

lincl the direction cosines of the vector
N = [rtt, rr]
perpendicular to the tangent plane of the surface r = r (u, v). It
is called the vector of the normal (normal line) to the surface, which
is a straight line passing through the point r = r (u, u) of the surface
perpendicularly to its tangent plane. We have
/ dx dy dz \ , t dx dy dz \
r“ =(-SST* -55T' -Kr) and 1 7 ’ -57)
and consequently the projections of the vector [rtt, r„l are
dy dz dz dx dx dy
du du du du du 'du
A =
dz
• B =
dz dx
, c= dx i
(3.35)
dv Tv ~ dj do du dv

Therefore its direction cosines are respectively equal to the expres­

sions
B
cos (N, a;) — — — —, cos (N, y) = —. —•

cos (N, z) = — , C =■

In particular, if a surface is given by an equation

^ / {x, ij) 2

or, in vector form, by the corresponding equation

r = xi -f y\ + / (*, y) k
we have

= - r *XI 1

Thus, in this case the formulas for the direction cosines are
— i x _____
cos (N, x) = cos (N, y ) = —j= ~ ( « - -
V i + /? + /•; ' V i ' /«+/D
cos(N, z) = (3.30)
2 1 / '2
V i + / x • fy
7. Coordinate Systems in Tangent Planes. Consider a surface 2
having the tangent plane at each point A/. Ft is sometimes convenient
to Ihink of a surface as being covered by the “scales” formed of
tangent planes. Thus, the surface is interpreted as a curvilinear
mnnifnhl which i* iho carrier of its tangent planes, the latter being
13G MU LT IP LE INTEGRALS, FIELD THEORY AND S E R I E S

the carried linear (plane) manifolds. Such an approach will ho of

use for studying Ihe subject matter of this and the subsequent sections
of the present chapter.
('boose a pair of uoncollinenr vectors e, and e*» in each tangent
plane and consider them the base vectors of a coordinate system
in the plane. These base vectors can be taken at pleasure at each
point. Hut if l.lie surface is defined by parametric equations with
parameters u and i; we can conslmct a specific pair of base vectors
generated in a natural manner at each point by the parametric
representation of the surface, namely the vectors — — and
01* ^
!»„ — — . If we fix a value v = v0 of the parameter v and make

Fig. 3.17
the parameter u vary the radius vector r (u. i-*0) describes the coordi­
nate curve r — i'0 — const on the surface (Kig. 3.17). The tangent.
vector to the curve, i.e. ^ (m, t>0), lies in the tangent plane to Ilie
s u r f ac e (see Sec. 4). S i m i l a r l y - t he v e c t o r — al«o lies in t he t a n g e n t
7 *’ an
plane to the surface. As before, we suppose that only a single curve
of each of the families u — const and v — const passes through
every point. Therefore we have a uniquely defined pair of base
vectors rM, rc in each tangent plane. If the vectors are different
from zero they are noncollinear since, according to tin* h y p o t h e s i s ,
the curves u — const and v = const are at no point tangent to each
other. Hence, they may turn out to In* coJIinear only when one of
them or both vanish. In what follows we shall suppose that the
parametric representation is such that rfi =y= 0 and rt. U on the
piece of the surface we are dealing with.
Thus, every parametric representation of a surface with para­
meters // and i’ generates a uniquely determined pair of base vectors
<>, , r||. o, |-r. i.e. an affine coordinate system, in each tangent
plane to tlie* surface. _ _
If some other para meters u and v are chosen instead of n and r
we obtain another set of coordinate systems determined by She base
vec tors <*i — and 0 2 ^ in the tangent planes. J. he transition
from one parametric representation to another generates an ntfmc
! !*nncfopTV\ i o f j} o
 CH. 3. ELEMENTS OF D I F F E R E N T I A L GEOMETRY I.T

Actually, let
’ u — u (u. v), v — v (u. t )
Ik * the
expressions of the parameters it. v in terms of u. t*. According
to the rule of differentiation of a composite vector function we
obtain
(hi
------
Ov
r„ —
On * Ou
<fu dv (3.37)
r? i\.
or Ov

Consequently, the new base vectors Cj and e2 are expressed in

terms of e, and e2 by the formulas
Ott , Ov
«i — e, -J— —
tfU * (Jit
t r* o~' \
- Or
e2 - c'l - ‘‘2
r/t in
-

The formulas expressing the base \octors c, and e 2 as linear comhi-

nalions of e t and e 2 are similar to (3.37').§

§ 4. DETERMINING LENGTHS, ANGLES AND AREAS

ON A CURVILINEAR SURFACE.
FIRST FUNDAMENTAL QUADRATIC FORM OF A SURFACE
There are many physical, technical and geometric problems involv­
ing the computation of arc lengths for curves lying on a surface,
angles Lotween such curves and areas of various parts of the surface.
Here we shall discuss these questions. The key idea of all the con­
siderations given in § \ is essentially based on replacing an inlmitcs-
imal element of a smooth surface by the corresponding element of
its tangent plane. It is therefore expedient to begin with some formu­
las and notions related to determining lengths, angles and areas
in the plane.
I. Affine Coordinate System in the Plane. Consider a plane and
a pair of nonrollinear base vectors 0 | and e 2 lying in it. Kvery vector
in the plane is expressible in the form
r — i i *=2°3
Let us fin cl the square ol‘ the length of the vector r. We have
r2 - (r, r) = |; (e,, e,) -j- 3|,c 2 (e*, e2) 4- £; (e2> e2)
Introducing Ihe notation
(i4 t . (*.) e.*» —
, T — doi^ o.A (e»* oA
138 MULTIPLE IN T E G R A L S , F I E L D THEORY AND S E R I E S

we rewrite the last relation in the form

r3 (3.38)
The quantities g lit g i2 and g 22 (referred to as metric coefficients)
are specified by the choice of the base vectors Cf and e 2. It can be
easily shown that the lengths of the vectors lying in the plane,
the angles between the vectors and the areas of the parallelograms
constructed on the vectors are expressible in terms of these quantities
(ami. of course, in terms of the coordinates of the corresponding
vectors). Indeed, the expression for the length r of the vector r is
obtained from formula (3.38). Further, if
r = liCj + lnCo and p — i|jei + T)2e 2
we have
( rt p) = rttili'Hi ” £ 12 *1*12 ~r ^ial 2Tli t* £ 2 2 *2*12
Now applying the formula
/ \ ( r . P )

cos(r- p ) = "M T H
we can express the angle between the vectors r and p in terms of
their coordinates and the coefficients g ih.
Finally, let us find the area S of the parallelogram constructed
on the vectors r and p. .'Vs is well known,
S — | [ r, pi |
and therefore
S = | Itjc*, -1- £ 2e2, ilte, -}- ii 2 e 2l | = I iiq 2 — * 2*)i I I 1<-V <-‘2 ! I
Hut
I I«i. e2JH Ici 11 e 2 Isin (e,, e2) = | et | i e 2 1 V 1 — cos3 (c1? e2) =
“ V — (t»j, c2)a = v gwgzi—
Consequently,
S — V gugl'l— git I^1*12 £2*11
Thus, we can really lind the lengths, the angles and the areas on the
plane when the quantities g u , g 12 and g 2 2 arc known.*

* VVc sometimes use the notation

E — (ci, c,)» h' — (elf e.2), C = (e2, e2)
Hutting ^21 = gi 2 we can also write the quantities («, i — 1 , 2i in 11 to
lorm ot a matrix
( £11 £i 2 \
\ ^«ii !-**>*»/
 C1I- 3. EL EM EN TS OF D I F F E R E N T I A L G E O M ET RY 139

2. Arc Length of a Curve on a Surface. First Fundamental Quad­

ratic Form. Let us be given a surface
r — r (m, v)
Compute the arc length of a curve lying on the surface. Taking the
arc length of the curve as the parameter we can write its equation
in the form
r = r (k(0 , *(/))
The vector ^ lining of unit length, we have
dl 2 = dra
But
dr = ru du -f ru dv
and consequently
dl 2 — du* -f- 2 (ru, r„) du du -f- r* du2
Making use of the notation
Sn = rut Si 2 = (r,o re), g 22, = r%
we obtain
dl~ = g u du2 + 2 ", 2 du du 4- g 22 du2 (3.39)
This expression is a quadratic form (in the variables du and du)
which is apparently positiue definite*. It is called the first funda­
mental quadratic form of the surface r — r (u, u). The coefficients
"n> S\z and of the form are obviously the ones corresponding to
the base vectors ru and r„ in the tangent plane to the surface at the
point in question. The coefficients may vary as the point moves
on the surface. Besides, they are of course dependent on the specific
choice of the parametric representation of the surface.
The lirst fundamental quadratic form of a surface provides the
expression for the length of an infinitesimal arc. The length of
a finite curve lying on the surface is obtained from it by integration.
More precisely, if a curve on a surface is given by equations
u = u (/), u — v (/), /j ^ t < t2
its length is equal to
h _______________________________ _
f i / I du \2 tin du / dv \2
V g" \i u ) r 2"'°-nr — -r "2' (nr) ,u
*i
(the quantities gi2 and g22 become functions of the parameter t
when the current point moves along the curve).
n
* A quadratic form 2 is said to be positive definite if
i, k
n
2 :> (> for »il in .. In except || =* 5- = . . . — 6ri = °*
HO MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

Exam pies
1. Lot. in the piano. Ihero be given a coordinate system determin­
ed by two mutually orlhogonal unit vectors e, and e 2. If r0 is the
radius vector of the origin of the coordinates the radius vector of an
arbitrary point is equal to
r — r0 + ep/ -f c2u
We have thus obtained a parametric represent a Iion of the piano,
the parameters being the Cartesian coordinates u and v.
In this case we have
r t< — rc = e 2. Ifn = '!» £12 = *L U22 =* ^
and consequently the lirst fundamental quad ratio form of the plane
represented parametrically by means of its ('artesian coordinates
is written as
dl 2 = du2 - dv2
In this example the tangent plane coincides witli the surface
(which is also a plane) at all the points, and the pair of base vectors
generated in each tangent plane by the parametric representn Iion
coincides (to within a parallel translation) with the base vectors e,
and Co chosen in the plane.
2. Introduce polar coordinates p and cp in the plane. Then the radius
vector of an arbitrary point can be written in the form
r — r0 -}- p (e, cos rp -}- e 2 sin <|)
where e, and e 2 are again mutually orthogonal unit vectors. This
is tlie equation of tlie plane represented parametrically by moans
of polar coordinates. Here we have
rp = i»| cos cp + e 2 sin cp, r,, = p ( —*.*! sin <p -}- e 2 cos cp)
and consequently
£11= (ip, rP) - 1 , £ 1 2 = ( rP» >>) **
* 2 2 - (»V <V> = P8. dl* = dp2 t (»■ d ir
3. Consider a sphere of radius a ami take the longitude cp and
the latitude 0 as the parameters on it* (see Fig. 3.18). The equation
of the sphere in the coord i nut os cp and 0 is of the form
r = r0 |- «{(i cos <p -j- j sin cp) cos 0 + k sin 0 }
(check il up). It follows that
r0 _ r/ (i 0 0 s <| j j sin q) sin ll -|- ak cos 0

*'.p « (—i sin «| ! j cos <i) cos t)

* Let the laliindr 0 he reckoned from the equator, i.e. — -ry ^ h ^ -^7 *
 CH. 3. ELEMENTS t>K LUKFKHENTE A L C.EOMETUY 141

<iiicl lienee here we have

<//* = a * ( c / o 2 -i- c o s 2 o t f < r )

If a surface is given by an expression of z as an explicit function

3 = / (*. //)
that is
r = jr i -f- y \ -r / ( x , //) k

we can write the relations

r.v —i r k/xi *y —) -} k iy
and, consequently,
dt 2 = (!-L /i2) dx~ -r- 2l'xfy dxdy j- (1 j f j ) dy°~
Exercise. Write Ilie first fundamental quadratic form for a torus
in the coordinates cp and if (see the exercise in § 3, Sec. 3).
N

Fig. 3.18

3. Angle Between Two Curves. The angle between two intersecting

curve15 is, by definition, the one formed hy their tangent lines at the
point of intersection. Suppose two curve's lying on a surface have
a point in common. Let du and dv be the differentiaIs of the coordi­
nates corresponding to a displacement from the point of intersection
along one curve, and 6u and 6v be the differentials of the coordinates
corresponding |.o a displacement along the other curve. The displace­
ment vectors can he written as
dr = ru du -j- r,, dv, 6 r ■= ru 6u — rr 6 c
The angle cp between them is determined by the formula
idr, 6r)
cos cp= 1rfr 11 6r |

In particular, the angle o> between the coordinate curves, that is

between the vectors ru and rt>, i.*> determined by (lie formula
Cos CO -
142 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

If £ , 2 = 0 the coordinate curves on Ihe surface intersect at a right

angle. In (his case we have so-called curvilinear orthogonal coordi­
nates. The lirsl fundamental quadratic form is expressed, in ortho­
gonal coordinates, by the formula
dl2 = Sn dit* -f g2z dvz
4, Definition of Area of a Surface. The Schwarz Example. We
now proceed to study the notion of area for a curvilinear surface.
Before discussing the ways of computing the area we must give the

definition of the notion. It is introduced as follows. Let 2 be a smooth

surface bounded by a piecewise smooth contour L. Break up the
surface into “elements” 2 *, i — 1 , . . N, by means of a finite
number of piecewise smooth curves lying on the surface, and choose
a point M t in each part. Next draw the tangent plane to the surface
2 through each point Af, and project the elements 2 / on the corres­
ponding tangent planes. We thus obtain squarahle figures <S, in the
tangent planes (Fig. 3.19).
D e f i n i t i o n , The a r e a o f th e s u r f a c e 2 is the limit (provided
it exists) of the sum of the areas of the projections extended over all
the elements 2 f of the partition { 2 *} when the maximal of the
diameters of the elements tends to zero. A surface for which the limit
exists is said to be s q u a r a h l e ( r e c t i f i a b l e ) .
One may think that it would he more natural to define the area
of an arbitrary surface 2 as the limit to which the areas of the surfaces
of the polyhedrons inscribed in 2 tend, on condition that the maxi­
mal of the diameters of their faces tends to zero (by analogy with
the arc length of a curve which is the limit of the lengths of the
broken lines inscribed in the curve). But as early as the 19th century
it was found that such a definition was inconsistent. Consider the
following example of Schwarz*.
Let us inscribe a polyhedron in a cylinder of radius R and altitude
II in the following way. Divide the cylinder into m equal parts bv
//
means of horizontal planes, the altitude of each part being
(see Fig. 3.20). Break up eacli of the m }- 1 circles appearing in the
seel ions (including the upper and the lower bases of the original
cylinder) into n equal parts so that the points of division of each

* Schwarz, Hermann Amandus (1843-1921), a German mathematician.

 CH. 3. E L E M E N T S OF D I F F E R E N T I A L GEOMETRY 143

circle are placed above the midpoints of the arcs of the adjacent
circle. Now take two neighbouring points lying on a circle and the
midpoint of the corresponding arc of the nearest circle lying above
or below the circle and construct a triangle with vertices at the
three points. The union of these triangles is a polyhedral surface
shown in Fig. 3.21.
If now n and m are infinitely increased the sizes of all the triangles
(which are the faces of the polyhedral surface inscribed in the cylin­
der) tend to zero. But the total area of all the triangles by far not

Fig. 3.20

always tends to the quantity which is the lateral surface area

of the cylinder. Depending on the way n and in are varied the total
area can increase unlimitedly, tend to a finite limit different from
2a HII or have no limit at all.
In fact, simple calculations show that the area of one triangle
(for given m and n) is equal to

The total number of these triangles is obviously equal to 2nm

and therefore the sum of their areas is
= 211n sin -5- ( t — cos - ^ - ) 2 (3.40)
If now n and m tend to infinity so that m increases faster than n2
expression (3.40) increases unlimitedly. If n and m vary in such
a way that the ratio — tends to a finite limit o wc have
lim m ^1 — co.<-^-\= lim m2 sin 2 ="—7 - q
and, consequently.
14'i MULTIPLE INTEGRALS, FIELD TH E O RY AND S E R I E S

Selecting q in an appropriate manner we can make Ilie limit be

equal to any number greater than 2jtJiff (or equal to it), i.e. lo any
number greater than the “true*1 area of the lateral surface of the
cylinder. The true value of tlie area is obtained only in the case
q = U. i.e. if m increases slower than n*.
Thus, the attempt lo determine the area of a curvilinear surface
by means of inscribed polyhedrons has failed even in the case of
an ordinary circular cylinder. Hence, this method of defining the arc
length of a curve is inapplicable to the area of a surface. This admits
of a simple explanation. When the lint*ness of a partition of a curve
(which is supposed to he smooth) is small enough the direction of
the chord joining two neighbouring points of division is close to
the direction of the tangent drawn at any point of the corresponding
arc. Uut this is not the case lor a surface. Indeed, a polyhedral
plane area of arbitrarily small linear sizes can have all vertices
lying on a smooth surface and at the same time the angle between
the normal to the polyhedron and that of tlie surface can he large.
It is apparent that such a plane element cannot serve as a good
approximation to the corresponding curvilinear surface element.
This is just, the case in Schwarz’s example: if q ^ ^ is large the
triangles forming the inscribed polyhedral surface are almost per­
pendicular to l lie lateral surface of the cylinder. The polybed roll
composed of them forms a crinkled surface. This is why tin* area
of such a polyhedron can be many times that of the lateral surface
of tho cylinder.
5. Computing Area of a Smooth Surface. In the foregoing section
we have introduced the definition of the area of a curvilinear surface.
We are now going to prove the existence of area for a smooth surface
and deduce a formula for practical computation of the area.
T h e o r e m 3*1. Let a parametric equation
r — r (t/t v)
determine a smooth surface X bounded by a piece u-ise smooth contour.
Then the surface is squarable and its area is equal to
o= V gug 2 2 ~ g^du dv (S.'il)

irhere £,,, giZ and g *12 are the f n m U im e n tn l c o e f f i c i e n t s

(tj " u n f i t ies) o f th e f i r s t o r d e r of the surface, i.e. the coef­
ficients of its first fundamental quadratic form, and I) is the range
of the variables u and v.
Proof. Break up the surface X into parts X; (/ 1 . 2 .............u).
Choose a point .l/f in each part and draw the tangent plane at it.
 CH, 3. E L E M E N T S OF D I F F E R E N T I A L GEOMETRY 145

Next, introduce, in each tangent plane, a local coordinate system

with origin at the point M the normal to the surface at the point
being taken as a z-axis and the tangent plane as an x. t/-plane. The
coordinates x , y and z of an arbitrary point of the surface —, can
he written as functions of u and v:
x = x (u, i>), y — y (m, v), z = z (i/. v)*
The projection of the surface 2 * on the tangent plane at tlie point .!/<
is ^determined by the equations
x = x («, f) , y = y ( u , i;). z = 0
Taking advantage of tlie expression for I he area of a plane figure
in curvilinear coordinates (see Chapter 1 , § 0 ) we can write down
the area of the projection in the fomi
XU XC d u d v
Vu i/t>
D,
where D t is the range of the variables it, r as the point (a-, //, z) runs
over the element the sign -f- or —being so chosen that the
whole expression is positive.
The quantity
XIL *0
yu Uv
can be expressed iu a form irrelevant to (lie particular choice of the
coordinate system, namely
**
i/u !/n U1
If the surface elements (and, consequently, the domains Dj)
are sufficiently small we have
^ l| | l**«» **p1 Idu dv "{I [**ui *i’I | |u*u^, v—Vj "I ^i} di

* More correctly, we should have written x = xt («, r), tj — i/£ (//, i;),
z — zi («, v) because these equations are asocialed with the ith coordinate
system corresponding to the tangent plane and the normal at the point Mj.
** Let tj be the radius vector of the point Mj in the original coordinate
system in which the surface £ has the parametric equation r - r (i/. r). Deno­
ting the radius vector of a point A] (belonging to an element £,) in the local
coordinate system as p we can write r - r£ ! p. 't he vector r£ being considered
fixed (mid thus independent of u and c>, we have ru pr/. r(. = pr and hence
indeed
xu
^ l lf» ,p j I - I lr„. rcJ |
!/u J9v
where x. y (and z) are the coordinates of p relative to the local coordinate sys­
tem.- - Tr.
10 082-I
14C MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

where d t is the area of the domain D ,, t/j and v} are the coordinates
of the point Mi and max e, 0 when the partition of the surface 2
is infinitely refined. Therefore, the sum of the areas of the projections
of all the subdivisions 2 * of the surface 2 on the corresponding-
tangent planes is equal to the expression
71
2 (I [*•«, I |u.=ui. r=v-} d\ y. t;d s (3.42)
i=l 1 1 *=i
It is the limit of this expression, as the fineness of the partition
of the surface tends to zero, that has been called the area of the
surface. The limit exists and equals the integral
I [fat rel | du dv
D
because the first term in (3.42) is an integral sum for the integral
and the limit of the second term is zero. To complete the proof we
must show that
| [ru, r j | = V gugzz — g\« (3.43)
Let to be the angle between the vectors ru and r0. Then
I [**«, r 0] | = | r « | | r 0 | sin w = | ru 1 1 r„| V i — cos2 w ==

= V r l rl — rl rl cos2 (.) — vg\\gzz — £?*

and thus the theorem has been proved.
.’\ ote J. We have already dealt with tlie- vector lru, r„l (see § 3..
Sec. 3). As has been shown, its projections arc
Oy dz Dz dx I
du du du du du du
, *B** = and C
Qy dz i dz Dx Ox Oy
do Ov Ov Ov |I Ov tic
(where x, y and z are now the coordinates of r in the original
coordinate system) and hence the length of the vector is equal to
V A t + B' + C*
Consequently, formula (3.41) for the area of a surface can be
rewritten as follows:
a = f j VA* + H* - \ - C*du do (3 /.!')
D
Note 2. The geometric meaning of formula (3.41) lies in the fact
that the element of integration ’]/'gii —gf* du dr coincides, to
within infinitesimals of higher order, with the area of an “irifinh.esi-
in.il parr.Hologram" cut of ihe surface 2 hv two pairs of coordi-
 CH. 3. E L EM EN TS OK D I F F E R E N T I A L GE O M ET RY 147

nate curves u = u0, u — u0 -}- du and v = v0, v = v0 du drawn

iniinitely close to each other (see Fig. 3.22). In fact, the vertices
P 0, P t and P 2 of the parallelogram have the curvilinear coordinates

Fig. 3.22

(u0> yo)i (wo ~r du, v0) and (uQ, i>0 du), respectively. Therefore,
we have, to within infinitesimals of order higher than the first,
the relations
P 0P i = rUdu and P^P* — rudv
The area da of the parallelogram constructed on the vectors P 0Pi
and P 0 P 2 is equal to the absolute value of their vector product:
da = | Iru, r j | du dv
Finally, by virtue of formula (3.43), the last expression can be put
down as
da = V g u g Z2 — g';z du dv
Let us consider some important special cases of formula (3.41).
if a surface 2 is given by an equation
z = f (x. y)
expressing z as an explicit function of x and y we can write, as has
been shown (sec Sec. 2, example 4), the formulas
£11 - i + /x • £12 — f x f u £22 — f + fy

whence
V gl\g«2 g\2 = ^ f 'v
Thus, the area of a surface z = /(x , y) is expressed by the formula
CX= ^ y l + /* -I- ly dx dy (3.4 .'.)

where D is (for this particular case) Ihe projection of the entire

surface 2 on the x , [/-plane.
Note 1. Since we have
l / i + /* + /; ,- =
to*
US MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

(see § 3, Sec. 5), formula (3.44) can also be written in the form
dr. dy
cos (N, z)

This formula admits of a simple geometric interpretation: the area

of a surface element is equal to the area of its projection on the
x, p-planc divided by the cosine of the angle between the normal
to the element and that to the x, {/-plane.
Note 2. If a surface 2 is composed of a finite number of pieces
each of which is representable by an equation of the form z = / (i, //)
its area can be found by applying formula (3.44) separately to each
of the pieces.
Example. Find the area of the part of the paraboloid z = x2 v*
cut out by the cylinder x2 + ;/2 =
Solution. The soughl-for area is equal to
a— jl Y I -j /ixl -\-Ay'1dxdy

Passing to polar coordinates we obtain

2n a 2:t
a - f rf.p j V ^ T T r d r }2 f |(4r--|-l),/s | | ; <U, =
0 0 0

- l) -i)

Suppose now that a surface is dett'rmincd by an equal ion

F (x, i/, z) = 0 expressing an implicit functional relationship bet­
ween z and the variables x and y. If the surface is such that il is
possible to solve the equation in z, which is equivalent to the require­
ment that every vertical (i.e. parallel to the z-axis) straight line
lias at most one point in common with the surface, wc can apply
the rules for differentiation of an implicit function and thus write
dF dF
dz Ox dz dy
bx OF * dy OF
<)z dz
Substituting these expressions for f x’ and f'u into formula (3. V'.)
we derive

CI -- (3.45)
l r
 CU. 3 EL E M E N T S OF D I F F E R F. N'T1A I- G E O M E T R Y 149

Here again. as in formula (3.44). the integrand is nothing but the

reciprocal of the cosine of the angle between the normal to the surface
and the 3 -axis.
Exercise. Determine the area of the part of the conical surface
x 1 ~| y2 — z2 = 0 lying inside the cylinder x2 -f y2 — a2-
A nsw er. o = 2 2 rea 2.

§ 5. CURVATURE OF CURVES ON A SURFACE.

SECOND FUNDAMENTAL QUADRATIC FORM OF A SURFACE
In the foregoing sections we deduced the formulas for computing
lengths of curves on a surface, angles between the curves and areas
of surfaces. Hut these quantities do not completely characterize
the shape of the surface. For instance, a cylinder and a plane are
obviously different surfaces although a cylinder can be rolled out
on a plane so that all the angles, lengths and areas are preserved.
To investigate the shape of a surface we shall apply the following
method: we draw all the possible pianos passing through the normal
to the surface at a given point and consider Ihe shape of the sections,
i.e. the plane curves (called normal sections) thus obtained.
1. Normal Sections of a Surface and Their Curvature. Let us take
a surface 2 determined by an equation
r = r («, v)
Here and henceforward the vector function r (//, u) will be suppos­
ed to be doubly continuously differentiable. Choose a point M 0

on the surface and deline a certain direction on the normal to the

surface 2 at the point /V/0, i.e. draw a unit vector n along the normal
line. Let y be one of the normal sections passing through the point
M 0. Then the curve y lies in a plane passing through unil normal
vector n to 2 at the point M 0 (Fig. 3.23). Thus, y is a plane curve,
150 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

its shape in the vicinity of the point AI0 being completely charac­
terized by its curvature k at the point and by its direction of con­
cavity (relative to the chosen direction of the normal, at the point
A/o, specified by the vector n). To compute the curvature of the
curve y write the equation of the curve in the form
r = r (t/ (/), i;(f)> (3.46)
where I is its arc length and apply the first Frenet-Serrct formula
dx _ dar j
dl dl3 *■
It follows that
v) 0-47)
The unit vector v is apparently in the direction of the normal to
the surface 2 at Ilie point /l/0 and, consequently, it coincides with

n in case the direction of concavity of the section y coincides with

the direction chosen along the normal to 2 or differs from n in its
sign if these directions are opposite. In order to take into account
both the value of the curvature and the direction of concavity of
the section y we introduce the quantity
(3.48)

which will be referred to as the normal curvature of the surface 2

at the point Mo in the direction of the section y. It is clear that
k = | k j. If the plane in which the section y lies is rotated about
the vector n the normal curvature k = k (y) may vary. Its variation
indicates not only the shape of the normal section but also the
direction of its concavity. For instance, if a surface is of the form
of a saddle in the vicinity of the point M©, as shown in Fig. 3.24,
we have a positive normal curvature for thu section y t since the
vector Vi of the principal normal to y, coincides with n and a negative
normal curvature k z for the section y 2 since v 2 = —n.
 CH. 3. ELEMENTS OF D I F F E R E N T I A L GEO M ET RY 151

In what follows we shall always consider normal curvature (3.48)

hut not the curvature defined by formula (3.47). This normal cur­
vature will be denoted by the letter k instead of k.
The quantity

is known as the radius of normal curvature of the surface 2 (at the

corresponding point and in the given direction). The noiinegativc
quantity \ 11 | is obviously the radius of curvature of the corres­
ponding normal section. Since k may vanish the quantity R may
assume infinite values.
Let us now derive a formula for computing the normal curvature A\
For this purpose we take advantage of equation (3.46) of the curve y
d2r
and calculate . For brevity, we introduce the notation
_ d-v d-r
Fuji —■ ruv~inrou rt>»— do*
From equation (3.46) of the curve y we deduce
d -r d I _ du , dv \
dl* ~ ~dl \ r “ T°~df ) ~
/ du \2 du do / rv \ 2 d'2u d-v
= ruu\7 r) + 2 r “ w dT~dT + r ** U t J + r “ “d 7 r + r ° r f 7 r (3.40)
The vectors ru and rp lie in the tangent plane and hence they are
orthogonal to n, that is
(rUl n) = (rv, n) = 0
rf-r
Therefore, substituting expression (3.49) for into formula (3.48)
we obtain
1 __ / d2r
H ~ \ dl2
. \(d u \2 du dv .
—(<■«**» n) ( c[l ) -F2(rut), n) , (3.50)
2. Second Fundamental Quadratic Form of a Surface. Let us
transform formula (3.50) for the normal curvature to another form
which is more convenient. Introduce the notation
bjt — (fitio •*), ^12 — (i*upi n), £>22 — (Fdjm n) (3.«>t)
and rewrite equality (3.50) as follows:
I _ 1 _b\] du2 |- 26j^du du -f*^22 dv*
' = IT = dP (3.52)
L52 MULTIPLE IN T EG R AL S. FIELD THEORY AND S E R I E S

Here we have the expression d/2, i.e. the first fundamental quadra­
tic form of the surface, in the denominator. The numerator is also
a quadratic form (in the variables du and dv). It is called the second
fundamental quadratic form of the surface and plays a very impor­
tant role (together with the lirst fundamental quadratic form) in
the theory of surfaces. In what- follows, the second fundamental
quadratic form of a surface will he denoted by the symbol tp2.
Thus, we have
(p2 ~ 5,, duz -f- 2 6 , 2 du dv -f- b2Z dv2
where 6 tl, 6 , 2 and b2Z are determined by relations (3.51).
Example. Take a surface defined by an equation
z = / (*. y)
or, in vector form,
r ^ x i + i/j r / (-e. y) k
Here we have
r.v.v — /.vyk, r,vy == /xyk and ——/pyk
Consequently,
6 ,, - /!« cos (n, 2 ), 6 ,3 = fly cos (n, 3 ) and 622 = fyy cos (n, z)
that is
<p-»= {fix dx- -f 2fxy dx dy --{- f vv dy2) cos (n, z) (3.53)
Thus, in this case the second fundamental quadratic form is, to
within t h e factor cos (n, r). the sum of the second-order t e r m s in
the expansion of the function z — / (.r, //) by Taylor’s form a la.
;Vo/t\ As lias been shown, tlie first fundamental quadratic f o r m
of a surface determines its “metric”, i.e. such quantities as lengths,
angles and areas which are found by means of the form. The compu­
lation of those quantities is in fact based on the replacement, i n I h e
lirst approximation, of ail infinitesimal surface clement by the
corresponding element, of its tangent plane. The second fundamental
quadratic form of a surface characterizes the measure of the rate
at which the surface turns out of the tangent plane drawn t h r o u g h
its point, in the vicinity of the point.
To prove this, let us lind the distance from a point A/, of a surface
2 , lying close to a given point Af(n through which tin* tangent plane
to 2 is drawn, to the plane (see Fig. 3.25). Consider a normai section
passing through the points M 0 and M . The sought-for distance, is
apparently equal to (he distance AFP from M to the tangent line
to the curve y. This distance is equal, to within infinitesimals of
higher order (see § 2, Sec. 6), to
4- k dl'2 —-i- (6,, da2 }- 2612 du dv -J- 622 dv2)
 CH. 3. E L E M E N T S OF DIFFERENTIAL GEOMETRY 153

the sign of the lust expression indicating I lie direction in which

the surface is turning out of the tangent plane.
It is possible to give a definition of tire second fundamental quad­
ratic form of a surface (equivalent to the definition given above)

proceeding from the problem of calculating the distance from a point

of the surface to the tangent plane drawn through another point
taken close to the former.
Exercises
1. Prove that the second fundamental quadratic form of a plane
is identically equal to zero for all the possible parametric repre­
sentations of the plane.
2. Eind the second fundamental quadratic form of a torus in
coordinates <p and \\- (see Example 1 in § 3, Sec. 1).
3. Unpin Indicatrix. The radius of normal curvature li — — cor­
responding to a normal section y at a point A/n depends on the direc­
tion in which the section y is drawn. To represent (ho dependence

/ _ ------ ^ z

Fig. 3.26

in a visual manner we can apply the following technique. Lay off,

from the point 3 /0, in all the directions on the tangent plane, a radius
vector p whose length is equal to \ It | where /? is the radius of
normal curvature of the surface in the corresponding direction.
The vector can obviously he written in the form
i> = r r « iT
where r is the unit tangent vector to the normal section in question.
'Flic locus of t.ho tips of the vectors is a curve lying in the tangent
plane to the surface 2 at the point M0 (Tig. 3.2(5). 11 is called Du-
15'. MULTIPLE INTEGRALS. FIELD TIlEOltV AND SKKIES

pin’s* indicatrix of the surface 2 at the point. Let us deduce the

equation of the Du pin indicatrix.
Take the sectors ru and as base vectors of the coordinate system
in Ilie tangent plane. We have
rfr du , dv
T ^ n r = r“ I T -i" r° I T
and therefore
p= v w , % - t u - v m rt
Lhal is each point of tlic Dupin indicatrix has the coordinates

i = i ' T « i - 3 r and
relative to the basis we have chosen.
Next we take advantage of the relation

.Multiplying it hy | R | we see that

k ( v w \ - £ ) 5+ 2&„ ( v m — ) ( K m £ ) -i-
11
which implies that £ and q satisfy the equation
bn H2 \ L- *= ± 1 (d.r,.'.)
This is an equation of a central curve of tin1 second order with
centre at the origin of coordinates.**
Thus, Dupin’s indicatrix is a central second-degree curve with
centre at the corresponding point of the surface.***
4. Principal Directions and Principal Curvatures of a Surface.
Equation of Euler. The Dupin indicatrix being a central curve of the
second order, wo can pass to its principal axes, i.o. replace the
base vectors ru and r„, if necessary, hy a pair of unit base vectors
lying in the tangent plane which are mutually orthogonal and such
that the equation of the Dupin indicatrix in the new coordinates
does not contain the terms with the product of the coordinates.
The new base vectors must bo in the directions of the principal
axes of the Dupin indicatrix. We shall call the latter the principal
directions of Hi** surface (at the point in question).
* Dupin, Francois Piorrc Charles (1784-1875). n French mathematician.
** This is implied hy 1 he fact that there are no first-order terms in the
equation.
*** More precisely, there are two such curves here, namely blx~ - ^ 2! ~~
•}- />-*2 »l2 ~ I and />ui" ~r t />221!2 —• 1 whose equations only differ
in the signs of their constant terms. For more detail concerning the shape ol
the Dupin indicatrix see Sec. 7.
 CII. 3. E L E M E N T S OF D I F F E R E N T I A L G E O M ET RY 155

Kor such a choice of the coordinate system in the tangent plane,

Ihe equation of the Dupin indicatrix takes the form
px2 -[- qy2 = ± 1 (3-55)
Let cp bo the angle between the principal direction taken as the
direction of the z-nxis and an arbitrary normal section. Then \vc
obviously have
x — Y | /f | cos cp, y = | li | sin (p
where /? is the radius of curvature of the corresponding normal
section. Substituting these expressions of x and y into equation (3.55)
and bearing in mind that the right-hand side of the equation is
equal to the ratio of | | to H we obtain
pcos2cp [-7 sin 2 (p = = k (3.56)
Denote by A*i — — and k2 — the normal curvatures correspond­
ing to the principal directions of the Dupin indicatrix at the point
under consideration. These quantities are referred to as the principal
curvatures of the surface at the point. The principal directions are
determined by the values cp Oand cp — -^-in the coord inale system
we have chosen in the tangent plane. Consequently, we have
A'i = />, k2 = q
Therefore equality (3.60) takes the form
k-—kt cos2 cp•;* k2sin2 <p (3.57)
or
1 t (3..V7')
~ = — cos- 9 + 7 7 “ sin- cp
rormula (3.57) or (3.57') is known as the equation of Euler*. It
expresses the normal curvature in an arbitrary direction in terms
of the principal curvatures. From the equation of Euler it imincdia-
U*ly follows that the principal curvatures are the extremal values
of the normal curvature. Indeed, if A, = kz, the quantity k is inde­
pendent of cp, and all the directions can he regarded as being extremal
in this case.** Hut if k{ =?£= k2 we can put, for definiteness, A*, >* k 2t
and then A*j — A: 2 0 and the equation of Euler can be rewritten as
k (A*, — A*2) cos2 <| -f- k2 (cos2 cp -+- sin 2 cp) =
= (/c, — /c2) cos- cp r k 2
which shows that A*, ^ A* k 2 for every ip.
* Euler, Leonard (1707-1783), a great Russian mathematician (a Swiss
by birth).
** A point on a surface at which A:, = A*a == 0 is said to he an umbilical
(circular) point of the surface. It can be shown that the only surface whose all
points are umbilical is the sphere.
150 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

Tin* extremal properties of the principal curvatures provide

n convenient practical method for computing them.
5. Determining Principal Curvatures. Equation of Euler (3.57)
makes it. possible to visualize the dependence of the normal curvature
k (ip) on the direction specified by the angle q. This functional
relationship is represented graphically in Fig. 3.27. It shows that

for every given Ac. At > A0 > A2, there exist four values of the
angle q- for which k («j) = Ac. Since the angles which differ by 21
define the same direction, there are two normal sections corresponding
to each A0 for which the normal curvature is equal to k0. Hut when
k0 = A*j or A*o *= A*2 the two normal sections merge into one.
In other words, the principal curvatures are the values of the
normal curvature such that to each of them there corresponds one
and only one normal section of the surface. Formula (3.52) defining
the normal curvature as a function of the direction can he rewritten
as follows:
(Ajj “ * kg|j) dn~ -[* 2 (i/ | 2 kg 12) tin do {^22 — 22 ) dv~It

Now. dividing bv do2 and putting — = f (where the value t

determines the direction of the section) we obtain
(feu Agn) t2 — 2 (b{2 — Argi2) t T (6 3 2 — ^£22 ) —■ft (3.58)
According to the above discussion, this quadratic equation (in /)
lias n single root and not two distinct roots if and only if the para­
meter k entering into (3.58) takes on the values of the principal
curvatures. Then the corresponding values of the single root / deter­
mine the principal directions. Furthermore, for this being so it is
necessary and sufficient that the discriminant of equation (3.58)
turn into zero.
Thus, lo find the principal curvatures we must solve, in A*, the
following equation:
(b\-> — kgl2)* — (An — kgX\) {b22 — ^ 22 ) — ^ (3.5D)
or
An kg], A|2 — kg 12 q
(3.5<r)
b\2 — A:gt2 b22 — kg22
 CJI. 3. E L E M E N T S OF D I F F E R E N T I A L GEOMETRY 157

6. Total Curvature and Mean Curvature. In many cases it is more

convenient to consider tlio product
K = k\k2 (;;.ik>)
and the lialf-sum
H = j (fci -r *s) (3.01)
instead of the principal curvatures k t a n d k z. T h e q u a n t i t y K is
called the total (or (iaussiati) curvature of the surface, and II is
referred to as its mean curvature.
Quadratic equation (3.59f) immediately implies the formulas
11^22— 612 yy 22— 3^12 4~ #23^11 02)
gitffn — Kh ’ 2 (tfiu ,'2 2 — £ * 2 ) V * “
Example. Compute the total and the mean curvatures for the
hyperbolic paraboloid 2 = x1 — t/2.
Solution. We have g u — 1 4x2, g\2 = —4xi/, g22 1 + 4//2,
b\ 1 = 2. br> = 0 and b22 = —2. lienee,
11 = 14!- ix2 }-4</2
4
K = —
1-1-4j 2 l-4y- *
In particular, we have K — —4 and // = 0 at Ihe o r i g i n .
7. Classification of Points on a Surface. We have attributed a
certain plane curve, namely the Unpin i m l i c a t r i t o each point 3 /0
of a surface £ defined by equations involving doubly differentiable
functions. As has been shown, the equation of Dupiu's indicalii.v
can be transformed to the form
y o ■ J n I ■"J *0 .
tjX” -f“ /v'2y~ = i 1 (i.tiri)
where k t and kz arc the principal curvatures of the surface at the
point M (]. The type of curve (3.63) depends on Ihe sign of the pro­
duct kj%2. Let us consider the possible cases.
(1) k lk 2 !>• 0. We can pul, without loss of generality, A*, 0
and kz -J> 0, because, if otherwise, we can reverse the direction of
the normal vector 11 and thus change the signs of k t and k 2 lo Ihe
opposite. Equation (3.63) determines an ellipse for A, >• 0 and
A*2 >• 0 if we have -j 1 on its right-hand side and does not define
any curve at all when we have - 1.
The points For which ktk2 > 0 (i.e. when the Dupin indicatri.v
is an ellipse) are called elliptic points of the surface.
(2) A*i/r2 <C 0. In this case equation (3.G3) determines a hyperbola
or, more precisely, two hyperbolas with common asymptotes. One
of them corresponds to the term --1 on the right hand side and the
other to —1. A point at which A,A*2 <C 9 (when the Dupin indicalrix
is a pair of hyperbolas) is called a hyperbolic point.
158 MULTIPLE INTEGRALS, FIELD TH EO RY AND S E R I E S

(3) kxkz = 0 . If one of tlie principal curvatures is different from

zero equation (3.63) determines a pair of straight linos intersecting
at the point. The points at which kxk2 — 0 (but one of the prin­
cipal curvatures is nonzero) are said to be parabolic.
If k | = k2 — 0 the notion of Dupin’s indicatrix becomes sen­
seless. A point for which kx = k2 = 0 is referred to as a planar point
of the surface.
Tims, the type of the point is specified by tlie total curvature K —
= k xk2 at the point. Since

—*1*
and the quantity g ng22— g]2 is always positive, the type of the point
is determined by the sign of the discriminant of the second funda­
mental quadratic form.
We can easily visualize the shape of a surface in the vicinity of
its point belonging to each type. Let M0 be an elliptic point. Then

k x and kz are of the same sign and hence, by virtue of the equation
of Euler, all the normal curvatures have the same sign at the point.
This moans, geometrically, that all the normal sections at the point
have the same direction of concavity. In the vicinity of an elliptic
point the surface resembles a piece of an ellipsoid and looks as is
shown in Fig. 3-28a.
Consider now a hyperbolic point. The principal curvatures arc
of opposite signs at the point. Therefore in this case there are normal
sections of different directions of concavity. The surface is of the shape
of a saddle in the vicinity of such a point (see Fig. 3.286).
The structure of a surface in the vicinity of its parabolic point
can be of a more complicated nature. In this case there is a direction
in which the normal curvature is equal to zero, and the normal cur­
vatures arc nonzero and of the same sign in all the other directions.
A typical example of a parabolic point is any point of an ordinary
circular cylinder (see Fig. 3.28c) but there are many other possible
configurations which we shall not discuss herc-
Consider an example. Let a surface be determined by an equation
z = / (*, It)
and let. the* well known necessary conditions for extremum be
fulfilled at a point (a:,,, yft). i.e. ^ = Oand ^ = 0. Then the normal
 CII. 3. E L EM EN TS OF D I F F E R E N T I A L G E O M E T R Y 159

to the surface at the point coincides, in its direction, with the z-axis,
and, as it can be easily shown by means of simple calculations, the
coefficients of the second fundamental quadratic form at the point are
&H = fxx* hi- ~ ^21 = /* y » ^22 “ fy y
Consequently, wo have
bnb22- b [ 2= fxxfyy - fxy (3-64)
We see that the type of the point is determined by the sign of expres­
sion (3.64). But, as is well known, the sign of the expression specifies
the existence or nonexistence of an extremum at the point. Thus,
we have established the following relationship between the type
of the point and the existence or nonexistence of an extremum at it:
elliptic point. The condition fm xxfyy — fxy > 0 holds and there
is an extremum;
hyperbolic point. The condition fxxfyy — fxy<L U is then fulfilled
and there is no extremum;
parabolic point. The condition fxxfyy — fxy = 0 which takes place
here indicates the case when the question of an extremum at the
point (x0, ij0) remains open and requires further investigation.
Exercise. Determine the type of the points lying on the following
surfaces: (1) an ellipsoid, (2) a hyperboloid of two sheets, (3) a hyper­
boloid of one sheet, (4) an elliptic paraboloid and (5) a hyperbolic
cylinder.
8. The First and the Second Fundamental Quadratic Forms as
Invariants of a Surface. We have introduced the first fundamental
quadratic form of a surface and shown that it determines lengths,
angles and areas on the surface. Furthermore, we have proved that
the second fundamental quadratic form specifies the shape of the
surface in the vicinity of each point. .Now it is natural to ask as to
what extent a surface is determined by its two fundamental quadra­
tic forms. The answer to the question is given by the following
theorem.
T h e o r e m 3.2. If it is possible to introduce a curvilinear coordinate
system w, o on a surface 5 and a system u*, v* on a surface 2* so that
at the points where u = u* and v — v* the corresponding fundamental
quadratic forms also coincide (in the sense that the equalities
£ll = £ll»
£l2 = £i2> />22 = f>22*
&12= &22 =
hold at these points) the two surfaces are congruent, i.c. they
can on lit differ in their nasi Linn in snare.
Jl.l* M U L T I P L E IN T E G R A L S , F I E L D T H E O RY AND SERIES

Thus, tlie lirsl and Ihe second fundamental quadratic forms of

a surface play the same role for surfaces as the intrinsic equations
for curves, and lienee they form a complete system of invariants
which uniquely spcciiies the surface to within its position in space.
Wo shall not present the proof of the theorem because it can he
found in many courses in differential geometry.*

§ 6. (INTRINSIC pro perties of a s u r f a c e

1. Applicable Surfaces. Necessary and Sufficient Condition for

\pplicability. In the foregoing sections we regarded a surface as
a rigid body which can move in space but cannot change its shape.
L>ul it is sometimes convenient lo consider a surface as an inexten-
siblo hut absolutely flexible film. This leads to studying the pro­
perties of a surface which do not vary as tin* surface is subjected to
a bending, i.c. to a deformation which is not connected with stret­
ching or shrinking.
If a surface can be made coincident with another surface by
means of a bending, the surfaces a re said to be applicable (isometric).
In oilier words, two surfaces are called applicable if it is possible
lo establish a one-to-one correspondence between their points so
that the curves, lying on them, which are Iransfornm! into each
other by the correspondence, are of the same length.
It seems natural to raise the question as to what are the necessary
and sufficient conditions for two surfaces to be applicable. The
answer is given by the following theorem.
T h r o v e m 3.3. For two surfaces E and to be applicable it is neces­
sary and. sufficient that it he possible to inf rod'ire n porn mrtrizatinn
of the surfaces by means of the same parameters it and v so that their
fundamental quantities of the first order {the coefficients of their first
fundamental quadratic forms) should coincide at the points M £ 2
and AJ* £ having the same values of the coordinates u and v.
Proof. If the condition of the theorem holds \vc can establish
a one-to-one correspondence between the points of Ilie surfaces
having the same, coordinates u and and then their fundamental
coefficients oT the first order will coincide at the corresponding
points:
tfn ^ c>12 = £ 2 2 = £«

This makes it possible lo introduce parametric representations of

the curves, lying on Ilie surfaces, which are mapped onto each other,
by means of a common parameter ( such that the values of the para­
meter arc the same at the points of the curves which correspond
♦ T' - r |
» » ‘ A.
'1 • T r, <«A A
*1
I t 91
j.
A QQ
*
 Cl!. 3. ELEMENTS OF DIFFERENTIAL GEOMETRY 101

to each other. This implies

<i
i.e. the arcs are of equal lengths.
Conversely, if two surfaces I! and J * are applicable they can be
represented parametrically with the help of the same parameters
by introducing an arbitrary coordinate system ». v on the surface £
and attributing to each point M* £ £* the values of the coordinates
u and v of the point .1/ £ 2 to which M* corresponds. Take now an
arbitrary curve on the surface 5 and the corresponding curve on
the surface 2* and introduce a parameter t on them in such a way
that the points which coincide when the surfaces arc applied to
each other have the same values of the parameter. The arc lengths
of the curves being equal, we can write relation (3.05) for them.
Since the relation must hold for all the possible values /, and tz
of the parameter it follows that
gn du2 -f 2gtz du dv + gzz dv~ — g* du2 -f 2g* du dv -f g* dv2
The last equality is an identity in du and du since it is fulfilled
for any two curves corresponding to each other and passing through
any point in any direction. An identical equality of two quadratic
forms implies the coincidence of their coefficients and hence
g i i = £m. £12 = 2** = gU
which is what we set out to prove.
2. Intrinsic Properties of a Surface. The description of the pro­
perties of a surface which are invariable under a bending (i.e. are
preserved under an arbitrary isometric, length preserving, mapping)
constitutes the intrinsic geometry of the surface. Such properties
are referred to as intrinsic (absolute) properties of the surface. We
have proved that two surfaces are applicable if and only if it is
possible to introduce a first fundamental quadratic, form common
for them, nonce, a property belongs to the intrinsic geometry of
a surface if and only if it is expressible in terms of its first funda­
mental quadratic form.* Thus, the intrinsic geometry of a surface
is determined by its first fundamental quadratic form. Consequently,
the lengths of the curves lying on a surface are relevant to its inlriri-*1

* Here we mean, of course, the properties which arc related to the surface
itself hut not to the particular wav of in»rnd'i«*^rg {rioter7 ~r. :f.
1 1—0824
162 MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

sic geometry. Further, since the angle between two curves oil a sur­
face and the area of a surface are expressible in terms of the funda­
mental coefficients of the first order (see Secs. 2 and 4 in § 4), these
quantities also belong to the intrinsic geometry of the surface.
A remarkable fact is that the intrinsic geometry of a surface
includes its total (Gaussian) curvature A’. Indeed, there is a formula
obtained by Gauss which expresses the total curvature in curvilinear
orthogonal coordinates in the form
K _ _______ 1 f d / 1 dg2:>\ _j_ _d_ / 1 <?gn \ |
2 t du ' 3tt ' ' dv ' VeiiKji do > J
which involves only the fundamental quantities of the first order.
At the same time, neither the mean curvature nor the principal
curvatures are preserved under an arbitrary isometric transformation.
The term “intrinsic geometry” means that the properties under
consideration, preserved under an arbitrary isometric mapping,
pertain merely to the surface, not to its position in the surrounding
space.
We can illustrate this by means of the following “mental” expe­
riment. Imagine that there are some intelligent creatures inhabiting
a two-dimensional surface and that they cannot leave it and go
out into the surrounding space. They can construct the geometry
of their “world”, introduce the notion of a “straight line” passing
through two points by defining it as the shortest curve entirely lying
on the surface that joins the points (for instance, in the case of a
sphere such “straight, lines” are the arcs of the great circles) and
so on. They can introduce the notions of a “triangle”, “polygon” etc.
and study the properties of these figures (without going out into
tiie space surrounding the surface). These hypothetical creature*
cannot distinguish between this surface and any other surface
applicable to it.*
The geometry thus obtained is nothing but the intrinsic geometry
of the surface. For instance, the intrinsic geometry of the plane is
ordinary planimetry studied in elementary geometrical courses.
Hut all tiie theorems of planimetry remain true if the plane is replaced
by any surface applicable to it, say by a parabolic cylinder. Hut the
intrinsic geometry of the sphere essentially differs from that of
the plane. For example, the sum of the angles of a spherical triangle
is always greater than rt.
* The considerations concerning the possibility of distinguishing between
rectilinear and curvilinear geometric configurations on the basis of studying
their intrinsic properlies make sense not only for Iwo-dimcnsionai gcm etric
objects, i.o. surfaces, but also for the objects of higher dimensions, in particular
for the throe-dimensional space. These questions arc very important, for investi­
gating the general geometric properties of the universe but we cannot dwell in
ox* itioso emblems hero.
 CH. 3. ELEMENTS OF D I F F E R E N T I A L GEOMETRY 163

3. Surfaces of Constant Curvature. Consider a surface whose total

curvature K is the same at all its points. Such surfaces are called
the surfaces of constant curvature. The total curvature being inva­
riant under a bending, it follows that two surfaces or constant cur­
vature arc applicable to each other if and only if their total curva­
tures .arc equal. It can he shown that, conversely, any two surfaces
of the same constant total curvature are always applicable. Hence,

Fig. 3.2<J

such a surface is completely characterized, from the point of view

of its intrinsic geometry, by a single number, that is by its total
curvature A\
i 1• j I l i-»Vn o— of surfaces O f: LViictoiu
n r• m t n v
T h * r r p A j n n l r i n !» ^ r*I *
■»I C U I V a i u r C CSSOii
tially depend on the sign of the curvature, and therefore we must
separately consider the surfaces of a positive, negative and zero
curvature.
The plane is a surface of zero total curvature. Its intrinsic geo­
metry, as has already been mentioned, is ordinary planimetry. Any
oilier surface of zero total curvature has the same intrinsic geometry.
A sphere of radius /? can be regarded as a “canonical model” of
a surface with positive constant curvature K = ^ . The intrinsic.
geometry of this surface differs from planimetry which is familiar
to us. For example, if the arcs of the shortest lengths joining two
points (these are the arcs oT the great circles in the case of the sphere)
are understood as being “siraiglil lines” wo can assert that any two
“straight lines” intersect when infinitely continued, the sum of the
angles of any triangle exceeds n etc.
The so-called pseudosphere depicted in Fig. 3.29 is an example
of a surface of a negative constant curvature A' <T 0. This is a sur-
ii*
 MULTIPLE IN TE G R AL S, FIELD THEORY AND S E R I E S

face generated by revolution of a tractrix, i.e. a plane curve described

by the parametric equations

where x and y are Cartesian coordinates, about its asymptote y — 0.

The surface, as seen in Fig. 3.2b, is not smooth and has a cuspidal
edge. It is possible to prove that there cannot exist an infinitely
continuable smooth surface of constant negative curvature in the
three-dimensional space. The intrinsic geometry of the pseudosphere
differs from ordinary planimetry and from ihe intrinsic geometry
of the sphere. It coincides with the so-called Lohachevskian geomet­
ry* in which the sum of the angles of every triangle is less than n.
for each point there exists an infinitude of straight lines passing
through the point and not intersecting a given straight line etc.
Wc cannot discuss these problems at length here although they
are very important and closely related to modern ideas of physics
and, in particular, to the theory of relativity. For these questions
we refer the reader to special monographs**.

* N. I. Lobachevsky (1792-1856), a famous Russian mathematician, the

founder of non-Euclidean geometry.
** E g. see 118].
4 Line
Integrals

Such problems as determining the mass of a material line from

its density, computing the work of a held of force along a path and
many others require the introduction of the so-called line integrals
that is the integrals of functions defined over curves. The present
chapter is devoted to this notion which is important for ma hemati
cal analysis and its applications to physics.
Various physical problems involving integration of function;
defined along curves lead to two types of line integrals usual!)
referred to as line integrals of the first and of the second type. A:
will be shown, the integrals of these two types can be reduced U
each other.

§ 1. LINE INTEGRALS OF THE FIRST TYPE

1. Definition of Line Integral of Ihe First Type. Let A If be :

smooth or piecewise smooth* plane curve and / (M) be a funclioi
defined on the curve.
Consider a partition of the curve into parts by means o
points of division
. t = . l 0, t ............... 1* (4.1
and choose an arbitrary point M t on each arc .-I i„\A t-. Now forr
the sum
(4.2
i=l
where \ l t is the length of the arc (Fig. 4.1). We shall refe
to such sums as integral sums. Let us introduce the following deiini
1ion.

* A curve represented by equations x = q (?). // if (?) is said to be smnot

if the functions <p (?) and ij’ (?) are continuers and possess the continuous der
vutives rp'(/) and if*'(f) which do not. vanish simultaneously, i.e. if the cur\
has a t angerit at each point .ind the position of the tancenl continuously dcpcnc
on the point of tangency. A continuous curve composed of a finite number i
■**v, .. t!i v ,:rvrs *.s i.dhd pine wise ‘•mouth.
166 MU LT IP LE I N T E G R A L S . FI EL D THEORY AND SERIES

D e f i n i t o n . If integral sums (4.2) tend to a finite limit* J , as max Alt

approaches zero, the limit is called the l i n e i n t e g r a l o f the f i r s t

t y p e of the f u n c t i o n f (M ) o v e r th e c u r v e A B . We shall denote

the integral by the symbol
f / (M) dl (4.3)
AB

The points of the curve -42? being determined by their coordinates

(x, y ), we shall also designate the function f (M) defined over AB
by / (x, y) and write the integral ^ / (M) dl in the form
AB

\ i (*> y) di
AB

But the reader should bear in mind that the variables x and y are
not independent because the point (x, y) belongs to the curve AB.
We can easily show that the notion of a line integral of the first
type doe.® not in fact essentially differ from thot of a definite infegral
of a function of one independent variable and can be reduced to it.
Indeed, let us take the arc length I reckoned from the initial point
A as a parameter for the curve AB and write down the equations
of the curve in the form
x = x (Z), y = y (/), O ^ l^ L (4.4)
where L is the length of the entire curve AB. Then an arbitrary
function / (x, y) defined on AB reduces to a function / (x (I), y (/))
of a single variable I. Let I* he the value of the parameter I cor­
responding to the point .V/f, i = 1, 2. . . n. Then integral sum

* As in the case o f the definite integral (e.g. see [ 8 1 , Chapter 1 0 , § 1 ) , a num­

ber / is said to be the limit of the integral sums if, for evrvy e v O . the ine-
n
quality | J — V, / Cl/,-) A | < r holds when max Af,- becomes sufficiently
 CH. 4. L I N E IN T E G R A L S 167

(4.2) can be rewritten in the form

i (4.5)
1=1

which is nothing but an integral sum corresponding to the definite

integral
L

Integral sums (4.2) and (4.5) being equal, the integrals they are
related to are also equal. Thus, we have
L
f l ( M ) d l = j /(* ( /) , y(l))dl (4.6)
AB 0

and both integrals exist or do not exist simultaneously. Consequent­

ly, if the function f (M) is continuous* (or piecewise continuous and
bounded) along a piecewise smooth curve AB line integral (4.3)
is sure to exist because, under these conditions, the definite integral
on the right-liand side of equality (4.6) exists.
Note. Although, as has bet»n shown, the line integral of the first
type can be directly reduced to the definite integral there is a dis­
tinction between the two notions. The matter is that the quantities
Alj (the lengths of the arcs A l_iA i) are necessarily positive irrespec­
tive of which of the end-points A or B, of the curve AB, has been
chosen as the initial point. Hence, the orientation of the curve AB
(i.e. the choice of a certain direction on it from its initial point
to the terminal one) by no means affects the value of integral (4.3)
and consequently we have
j / (M) dl = f 1(M)dl (4.7)
AB BA
b
I3ut, as is known, the definite integral £ / {x) dx changes its sign
a
when the limits of integration are interchanged.
When reducing a line integral of the first type to the corresponding
definite integral we can as well use any arbitrary parameter / instead
of the arc length. Suppose a curve AB is given by parametric equa­
tions
_________ x = cp (t), u — \\> (0 (/0 < / < /•) (4.8)
* We say that a function / (M) defined on a rectifiable curve is continuous
on the curve if it is continuous as a function of the parameter I.
168 MULTIPLE INTEGRALS. FIELD T H E O R Y AND S E R I E S

xvher© Ih© fund ions cp (/) «aml ip (/) arc continuous. Iheir derivatives
<(/ (/) and if' (/) are piecewise continuous and bounded and xp'2 (/) -i-
- ’I '2 (0 *-> 0. Then we can introduce, as a new parameter, the arc
length I of the curve AB reckoned from a fixed point. Let us choose
the direction in which 1 is laid off so that the arc length /
increases when the parameter t increases. Then I becomes a monotone
increasing function of t. and we have
dl — ] / <p'2 (/) ip'2 (/) dt (4.9)
Taking advantage of equality (4.0) and formula of changing variable
in the definite integral we obtain
I* n ____________
j / (-’ /)d l = j / (•* Vh y W) d i = j / (t (0» ^ (0) V <p'* (0 4- Y A(0 dt
Alt 0 t(l

where /0 <c l x. Thus, we arrive at the following theorem.

T h e o r e m 4.1. Let AB be a smooth curve represented by equations
x = ip (/), if = ip (/) (t e f/0»
and f (x, if) be a junction defined a Ions* the curve. Then we have the
equality
ti ____________
\ /(* . y ) d i ^ j /(<p(0> »l*(0) V 'V ^ O l t ' 2(0<fr (4.10)
.w? t0
provided the integrals entering into it exist; the line integral on the
left-hand side exists i f and only if the definite integral on the right hand
side exists.
In particular, if the curve AB is determined by an equation of
the form
y — y (*) (a < x < 6)
expressing y as an explicit function of x formula (4.10) for reducing
the line integral to the definite integral takes the form
b _____
\ f (.1/) dl = \ f (x, y (x)) 1^1 -|~ y1* dx (4.11)
AD a

Kxercise. ( ’.onsider the line integral of a function / (x, y) over an

a i c .l / / represented by its polar equation
'■ r(«|)
in tie form of a definite integral with respect to «p.
 CH. 4. U N E IN T EG K A LS IB

A nsiver.
*2
j /(r , y ) d l = ^ / (r cos <p, rsin <f)]/ r2-Fr'<2dcp
AB <pf

Note. As is well known, the definite integral

b
j !{x)dx
Cl
of a noimegalive function cun Ik* interpreted as the area of the cur
vilinear trapezoid (see Fig. 4.2a). Similarly, the line integral
( / (■'/) cn
AH
can bo thought of, on condition that / (.1/) ^ 0. as the area of
piece of a cylindrical surface composed of the perpendiculars to ill

.r. //-plane erected at the points of the curve AB ant! having a vari
able length / (M) (Fig. 4.26).
2. Properties of Line Integrals. The basic properties of the 1in
integral of the first typo are almost completely analogous to thos
of tlie definite integral and are directly implied by formula (4.(1
which reduces the line integral to the definite one. Let us enumerat
t hern.
1 (linearity). If k = const and / (M) is integrable over AB wc liav.
f kj (M) dl —k [ / (.1/) di
An AB
ami the integral on the left-hand side exists.
2 (linearity). If / (d/) and g (d/) are integrable on AB the exp res
sion / (d/) lL g (M) is also integrable and the relation
\ (/ ( VO x g(\f)) dl--= j / (M) dl ± \ K (M) dl
AB AH AB
takes place.
170 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

3 (monotonicity). If f (M) is a nonnegative integrablc function

we always have

AB

4 (additivity). If an arc AB is composed of two arcs AC and CB

the equality
) ' / (■!/)dl = j / (M) d l - \ f (M) dl
AB AC CB
holds when the integrals entering into it exist, and the integral on
the left-hand side exists if and only if both integrals on the right-
hand side exist.
3 (estimation of the modulus of the integral). If / (J/) is integrabie
over AB the function | f (M) | is also integrable and there is an
inequality of the form
I j /(jW)<rt|< j \ H M ) \ d l
AB AB
6 (mean value theorem). If / (M ) is continuous on a curve AB
there is a point M* belonging to the curve such that
f / (M) dl —f (M*) L
AB
where L is the length of the arc AB.
7 (independence of the line integral of the first type of the orientation
of the path of integration). As has been mentioned, we always have
f / (.If) dl = j 1 (M) dl
AB BA
and hence the choice of the direction of the arc AB does not affect
the value of the integral of an arbitrary scalar function / (M) taken
along the arc.
3. Some Applications of Line Integrals of the First Type. We
shall point out some typical problems whose solution naturally
involves line integrals of the first type.
(1) Determining the mass of a material line from its density. A mate­
rial line will be understood as’a piecewise smooth curve along which
a mass is distributed. The linear density p (M) of a material line
ut a point M is the limit to which the ratio of the mass Ajx, carried
by an arc M M \ to the length of the arc M M * tends on condition
that the arc length tends to zero. In other words, if I is the length
of an arc A M and p(M) is the mass of the arc we have p(il/) =
= * It follows that the mass of the arc AB is expressed
 CH. 4. L I N E IN T E G R A L S 17!

t
hy the integral J p dl, that is by the line integral
o
\ p{M ) dl
AB
of the density p taken along the curve AB.
(2) Finding the coordinates of the centre of gravity of a material
line. Let a mass be distributed, with density p (x, i/),* along a curve
AB. After the curve has been broken up into parts of lengths Afj,
f= /?, and an arbitrary point (xf, y t) has been taken on
each part we can approximately represent the material line as
the system of mass points p(x*, t/f)AI/ placed at the points (x,-. ?/,-).
The centre of gravity of such a system of material points has the
coordinates
n n
2 xip to* yt) 2 wp(*i» yi) Az<
i= 1 ial___________
n » */c = n
2 Pto*0i)A/i 2 p to. Ali
»=1 i=*l
These expressions can be regarded as approximate values of the
coordinates xc and y e of the centre of gravity of the material line AB.
To obtain the exact values of the coordinates we must pass to the
limit as max Alt — 0- The passage to the limit results in
I x p { x t y)dl J yp to y)dl
AU A B ____________
xc Uc = (4.12)
J p (x , if) dl ’ I p(*,y)di
AB AB

In particular, for a homogeneous material line with p = const

we obtain
\ x dl \ ydi
AB AB
xc He — (4.13)
S* ’ I dl
AB AB

(3) Computing moments of inertia of a material line. The moment

of inertia of a system of mass points m t (i = 1, . . n) about
a straight line is equal lo
n
2 A mi
t=i
where r{ is the distance from the ith mass to the straight line. In
particular, for a system of material points lying in the x, (/-plane
* Here aod henceforth it appears natural to determine the points of a curve
by their ('artesian coordinates x and y (sec Sec. 1).
172 M U L T I P L E IN T E G R A L S . FIELD THEORY AND S E R I E S

the moments of inertia about the x-axis and the y-axis are respec­
tively equal to
n n

/ * = S y2inii an(1 fy= S

i=*l i= l
where (x*. yy,-) are the coordinates of the mass point ///,. To get the
moments of inertia, about the coordinate axes, of a material line AB
along which a mass is distributed with density p (x. y) we must
perform the passage to the limit similar to the one in the preceding
problem. Then, for the moments of inertia of the curve AB about
the coordinate axes, we deduce the expressions
Ax ^ j J/2P (*, y) dl, Iy = f x-p (x, y) dl (4.14)
AiJ AB
#(4) Gravitational attraction of a mass point by a material line.
Let AB he a material line with mass density p (x, yy). We now consi­
der the expressions of the projections on the coordinate axes of the
gravitational force F with which the material line attracts a mass
point m0 having the coordinates (x0, <ry0). Applying the argument
analogous to the one presented above we lind that the projections Hx
and by are given by the formulas
Fx - ym0 J pi'- dl, F„ = ym a f L1^* J’ ^Ul —Ve) dl
AD AH

where y is the constant of gravitation and r -- V (x — x0)2 (yy—y/0)-.

If integration of a vector is understood in the sense that each
projection is integrated (see $ I in Chapter 3) we can replace
the two scalar formulas by a single vector relation and thus the
force F with which the material point w 0 is attracted by the mate­
rial line AB is equal to
F - \tn0 ,J r dl j ^^ * (4.lo)
AB
where r is the vector whose projections are (x — x0) and (y — i/o)-
4. Line Integrals of the First Type in Space. The definition of
a line integral of the first type over a plane curve is directly trans­
ferred to the case of a function / (A/) defined along a space curve.
If a space curve is represented by parametric e q u a t i o n s
* (0» // = ip (')> z = '/ (*) A)
the line integral of the first type of the function / (.1/) taken along
the curve is reduced to the definite integral hv means of the formula
*1
f f(M)dl f/(< f(/), 'K 0, X(0) V'V* (0 ~ ' I (0 ~ x#* (0 dt
A Fi t r\
 CH. 4. LI K E IN T E G R A L S 173

Tiie conditions for the existence and basic properties of line

integrals in space are completely analogous to those for the case
of a plane. The line integrals of the first type in space are naturally
encountered in such problems as determining the mass of a space
curve (material line) from a given density, finding the coordinates
of the centre of gravity and the moments of inertia of the curve and
the like. By analogy with the arguments presented for the case of
a plane curve, the rcadrr can ea sily write tlie corresponding formulas.

§ 2. LINE INTEGRALS OF THE SECOND TYPE

1. Statement of the Problem. Work of a Field of Force. Now we
introduce integrals of another kind, the so-called line integrals
of the second type.
To begin with we shall take a concrete physical problem. Consider
a piano field of force, that is a domain in a plane at whose each point
M a force F (M) is determined. The projections of the force on the
x-axis and //-axis will he denoted, respectively, as P (M) and Q (M).
Let us find the work of the force held when a point, moves along
a curve A B .
If the force F is constant (both in its absolute value and direction)
and the path AB is rectilinear the corresponding work is equal to
the product of the value of the force by the path length and by the
cosine of the angle between the force and the direction of displace­
ment, i.e. to the scalar product.
U \ AH)
Let us now find the expression of the work in the general case
when the force F may ho variable and I be trajectory of motion cur
vilinear. Let AB be a smooth curve lying in the domain where the
held of force is defined. Divide the curve AB into parts by means
of points
.4 —j1/q, J \ ' f . . ., —B
and consider the broken lino with vertices at the points M-t (see
Fig. 4.3). We can approximately consider the force F to retain
a constant value along each segment of the broken line,
say equal to F (d/,), and compute the work corresponding to the
motion along the broken line. If (xf, //,-) are the coordinates of the
point Mi and
Axi = x i — x,_,f Atji = iji — f/i.,
tho work corresponding to the displacement, along the segment
M , . j.1/ 1 is equal to
(F {Mi), M t. {Mi) = P {Mi) Ar£ + Q (/I/,) A//,-
174 MULTIPLE IN TEG HALS, F I E L D THEORY AND S E R I E S

and the total work corresponding to the motion along the whole
broken line is equal to
2 (P(A /,)A *,
* + e(Jtf«)Ayi) (4.16)
4=1
The last sum can be approximately taken as the expression of the
work performed by the field of force F (M) along the curve AB. To
derive the exact value of the work we must pass to the limit in sum

Fig. 4.3

(4.1(1) by making the maximum of the lengths of the arcs

tend to zero. We now consider such a passage to the limit in the
general form.
2. Definition of Line Integral of the Second Type. Let AB be
a smooth curve and F (M) — (P {M)y Q (.If)) be a vector function
defined on the curve A B . Break up the curve into parts by means
of points
.1 ~ it/o, .V)• • • •» il/n = B
whose coordinates arc, respectively (x0, y0), (xM . . ., (xft, y„)
Tiiko lllP «iirn

T~, v \ P ( M , ) k x t + Q ( M t)b yi\ (4.17)

i-= l
wlu-iv \ x t — — x f_, and = yk —
If all such sums tend to a certain finite lim it when the maximum
of the lengths of the arcs Mi~\M* tends to zero the lim it is called
the line integral of the second type of the vector function F = {Py Q)
and is denoted* by the symbol
f P(M )dx+Q (M )dy (4.18)
AH
* We shall sometimes write P (xt iy) and Q ( .t, jj) instead of P (M) and
Q (Af) where x and ;/ should be understood as Cartesian coordinates of the moving
p n i nt M, We shall also simply denote the functions P (ilf) and Q {M) by P

and Q and write line integral (4.18) in the form ^ P dr -4- Q dy or ^ P dx 4-

AB A
-}- Q dtj unless this leads to misunderstanding.
 CH. LINE INTEGRALS 175

The integral is obviously the sum of the two integrals

j P(M)dx and £ Q( M) dy
AB AB

corresponding to the vectors (P, 0) and (0, Q), the components

into which the vector (P, Q) is resolved.
Note. The line integral of the second type must noL be confused
with an integral of a vector with respect to a scalar parameter (which
has been encountered in § 1, Sec. 5 of Chapter 3 and at the end
of Sec. 5, § 1 of the present chapter) when the projections of the
vector are integrated separately (e.g. we have had such a situation
in the computation of the force of attraction of a material point by
a material line).
3. Connection Between Line Integrals of the First and the Second
Types. The line integral of the second type can be easily reduced
to the line integral of the first type considered in § 1. The relation­
ship between the integrals is described by the following theorem:
T h e o v e m 4.2. Let AB be a smooth curve determined by equations
x = x (l)y y = y (0(4.19)
and F = (P. Q) be a vector junction defined on the curve and bounded*
on it. Then we have the equality
f P d x - { - Q d y = f (P co sa + ^Jsina) dl (4*20)
ad AB

where a = a (M) is the angle between the tangent to the curve AB

ca the point At and itic positive direction of the x-axisTprovided that
the integrals entering into (4.20) exist. Furthermore, the integral
on the left-hand side exists if and only if the line integral of the first,
type on the right-hand side of relation (4.20) exists.
Proof. Let us prove the equality
| P dx = | P cos a dl
AB AB
The equality
J Q dy = Q sin a dl
AB Ali
is proved similarly.

* A vector function (P, Q) is said to be bounded if the scalar functions

V and Q arc bounded.
17G M U L T IP L E IN T E G R A L S , F IE L D THEORY AND S E R IE S

The integral
\ Pdx
*
AB
is, by definition, the limit of the sums of the form
n
T = 2'
«—I
Let us compare the sum with the integral sum
n
r * = S / , (A/»)cos«(-WOA/,
t=l
associated (for the same partition of the curve AB) with the
integral
^ P cos a dl
AB
If x = x{l) the relation
-^ -= c o sc t(.l/) (4.21)
holds for each point M of the curve AB and, consequently, we
have
'i
S x t = J cos a dl
h-1
Applying the mean value theorem we thus obtain
Axt = cos a {M*) SIi
where M* is a point belonging to the arc Therefore,
n
| 7’— /•* | = | V p ( j / ,) ( c o s a ( A / i ) - c o s a (Mf)| M ,|<
n
y | P {M{) | • | cos cl (Mj) — cos a (Mf ) | Sli
i= i

The function cos a (/I/) is continuous on the smooth curve AB

and hence it is uniformly continuous because the curve is a hounded
closed set of points. Thus, given any e. /> 0, the* inequality
| cos cl {Mi) — cos a (.1/?) | <C e
is fulfilted for each partition of the curve AB whose lineness is small
enough. This implies
ft

17’— T ' l C C e 2 AZj = CeL

i
 CH. 4. LINE IN T E G R A L S 177

where C = sup | P | and L is the length of the curve AH. It follows

that if the integral sums T* have a limit the sums T tend to the
same limit. The theorem has thus been proved.
Note. The expression P cos a -f- Q sin a is the scalar product
(F, x) of the vector F = (P , Q) by the unit vector x = (cos a, sin a)
tangent to the curve AH and hence is equal to the projection of the
vector F —= (V, (/•) on the tangent line to AD. Denoting the pro­
jection by the symbol b\ ami taking advantage of equality (4.20)
we can write line integral (4.18) in the form

\Fxdl (4.22)
AB

This abridged notation will often be used in what follows and parti­
cularly in (Chapter l>. The integral is sometimes also written in the
form
I (I', d\) (4.23)
AB

where cil is understood as an infinitesimal vector whose projections

on the coordinate axes are
dr = dl cos a. and dtj = dl sin a
4. Evaluating Line Integral of the Second Type. Theorems 4.1
and 4.2 immediately imply the following result.
T h r o v e n t i.'d. Let
AH be a smooth curve represented by equations
* = *p(/), y = ip (/) (4.24)
and let F — ( / \ (V) be a vector function defined on the curve. Then we
have Ifie relation
11
( P d x + Qtly - \ l^(«p(0. ’I‘ (*))<p'( 0 + *1>( 0 ) 'I5' (01^*
An t0
(4.25)
provided that the integrals entering into it exist, and the integral on
the left-hand side is sure to exist if the definite integral on the right hand
sid" exists. Here, the value /0 °f parameter t corresponds to the point
A and /, to the point H.
Theorems 1-4.0 obviously remain true if the curve AD is not
smooth hut piecewise smooth.
Let us consider some import,ml special cases of formula (4.25).
If the curve AH is determined by an equation
y = jy (x) (4.9.1W
12 —0*24
178 MULTIPLE IN T E G R A L S , FIELD THEORY AND SERIES

expressing y as an explicit Function of the variable x which runs

over an interval [a, 6| formula (4.25) for reducing the line integral
of the second type to the definite integral turns into
h
J P d x + Qdy = | [P(x, y (x)) + Q (x, y ( x ) ) y f (x)]dx (4.27)
AD a

where the value x = a corresponds to the initial point A of the

curve AB and x = b to its terminal point B . In particular, if the
curve AB is a segment of a horizontal straight line y — y 0 we have
y ' = 0 along it and thus the integral
^ P dx - \- Q dy
AB

taken over such a segment simply reduces to the integral

b
^ P (x, y0) dx
a

Similarly, for a curve determined by an equation

x ~ x (y) (4.28)
where y ranges in an interval Ic. d\ we obtain
d
P d x + Q d y = \ [ l > ( x ( y ) , y) x' (y) + Q (.x (y), y)] dy (4.29)
ab c

If AB is a segment of a vertical straight line x = x0 we have x' = 0

and integral (4.20) takes the form

^ Q(xo, y) dy (4.30)
An

Examples
1. Evaluate the integral
J x2dx + xydy (4.31)
AB

taken along
(a) the line segment drawn from the point (1. 0) to the point
(0 , i ) .
(b) the quarter circle x — cos t, y — sin / j°>ninS
the same points (see Fig. 4.4).
 CH. *. LI N E INT E GR AL S 179

Solution•
o o
(a) ^ x2d x + xy dy — J (x2— x (1 — x ) ) d x = J (2x2—x)dx = ——
AB 1 •
£X
2
(b) £ x1dx 4* xy dy = ^ ( — cos2 / sin t -f- cos21 sin t) dt = 0
A/3 0
2. Evaluate the integral
£ 3x*y d x ( x * r i) dy (4.32)
ab
taken along
(a) the line segment drawn from the point (0, 0) to the point (1, 1),

(b) the arc of the parabola y — x- connecting the same points,

(c) the broken line passing through the points (0, 0), (1, 0)
and (1, 1) (Fig. 4.5).
Solution.
(a) j 3x2y dx 4- (.t34 (4x* 4 1) dx = 2
AB

(b) J 3x*y dx -I- (x3 4- (Ox44- 2x) d x = 2

Ali

(c) j 3x*ydx + (x3 | 1) dy =

AB
U%0) Cl, 1) f
= j 3x2y d x + f (x3-!- l ) d y = f 2 dy = 2
t o , 0) (i,o ) o
Note. The reader has undoubtedly noticed that the integral in
the second example assumes the same value when taken along the
three different paths connecting the two given points. This fact
is connected with an important property of the line integrals of
the second type which will be elucidated in § 4.
12*
»80 MULTIPLE IN T E G II AI.S, FIELD T H E O R Y AND S E R I E S

5. Dependence of Line Integral of tlie Second Type on the Orien­

tation of the Path of Integration. The definition of the line integral
f P d x + Qdy (4.33)
AB

automatically suggests that a constant factor can he taken outside

the integral sign, the integral of a sum of two vector functions is
equal to the sum of the integrals of the addends etc.
Another important property of integral (4.33) should be empha­
sized here: the line integral of the second type, in contrast to the
line integral of the first type, defined in § 1. depends on tlie orien­
tation of the curve AB it is taken over, namely, it changes its sign
when the orientation of the curve is reversed:
j / 1dx -\-Q dij — — j P dx | Q dy (4.34)
'a au
For, if wo change the direction in which the curve AB is traversed
we thus replace the quantities and Ay t entering into sum (4.17)
by —Ax-i and —Ay t. This changes the sign of integral sums (4.17)
and, consequently, the sign of their limit as well.
This property of the line integral of the second type is coherent
with the physical interpretation of the integral as the work of a
field of force along a path. Indeed, if the direction of tracing the
trajectory is reversed the work performed by the force held change's
its sign to the opposite.
6. Line Integrals Along Self-Intersecting and Closed Paths. For
the applications of the theory of line integrals, it is advisable to
include in our consideration the paths of integration which may
have points of self-intersection because such cases are encountered
in various problems Mathematically, (his means that when we
consider a curve determined by equations
x —x{t), y = y (t) (a < t < b)
we should not exclude the cases in which there may exist two (or
more) distinct values t x and /•> of the parameter I such that
* (*i) = z (t-i) and y (/,) = y (t2)
When studying such cases for the line integral of the second type
we must take into account that the specification of a path of inte­
gration includes the indication of a set of points constituting the
curve in question as well as a certain direction in which the path
is described. If a curve has points of self-intersection the way it is
traversed is not completely characterized by setting initial and
 CII. 4. I . I NK INTEGHALS 181

terminal points. For instance, the curves shown in Fig. 4.Ga and b
should be regarded as being two different paths although they coin­
cide as sets of points. What has been stipulated here pertains not
only to the plane curves but also to the space ones.
We also deal with line integrals taken over various closed con­
tours. A closed contour (in the plane) is understood as a curve
x = x (/), y = y (I), {a ^ t ^ b)
such that
x (,a) — x (b) and y {a) = y (b)
Here we can also have the cases when such a contour intersects itself,
that is there may exist various values of the parameter f, other
than t — a and / •— 6, for which the corresponding values of x
and y t respectively, coincide.

Fig. 4.6

If a closed contour, in the plane, has no points of self-intersection

there are only two possible directions of describing it: the contour
can be traced counterclockwise (the positive orientation) or clock­
wise (the negative oru m latiu u ).
If an integral of the second type

£ P dx-\- Q dy
c
is taken along such a contour its values corresponding to the two
different orientations of the contour C are equal in their absolute
values and have the opposite signs. In what follows, we shall, as
a rule, consider closed contours with the positive orientation and
replace line integrals of the second type taken along negatively
oriented contours hv the corresponding integrals taken in the posi­
tive direction with the minus sign attached to them.
A line integral over a closed contour is often denoted by the
symbol
P dx -f Q dy
b
 182 MU LTI PLE INTEG RAL S, FIEL D THEORY AND S E R I E S

7. Line Integral of the Second Type Over a Space Curve. We

have considered line integrals, of vector functions, taken along plane
curves. Most of the facts of the theory are automatically generalized
to the case of a space curve.
Let AR be a smooth space curve and F = (Pf R) be a con­
tinuous vector function defined on the curve. We divide the curve
AR into parts by means of points of division
A = .1/q, .1/|, . . AJn = B
having coordinates (x,-, y it zf). i = 1 , 2 ...........«, and form the
sum
71
2 {/' O M Ax, I- 0 ( M , ) b y , + n (.1/,) b z , }
i—l
where
Axi = x 4 — x ,.,, Ayl = r/i — i j i and Az, = z, — zt .i
flic limit of such sums will be referred to as the line integral of
the second type of the vector function F = (/*, Q, R) over the spare
curve AR and denoted hy

J P(M)clx + Q{M )d y + H ( \ f ) d : (/|.3.r>)

Alt
or*
j P (x, i/, z) dr !- Q (j, //, z) dy R (x, i/, z) dz
AH
Applying arguments completely similar to those ust:d in investi­
gating the case of a plane curve wo arrive at the following formula
reducing integral (4.35) to the line integral of the first type:
j P dx ! Qdy \ R d z — | P cos a -|- Q cos (5 • R cosy I dl
AU AD

where a, fi and y are the angles formed by tlie tangent line to the
curve AR (drawn in the direction of tracing the path of integration)
with the positive directions of the coordinate axes x, y and z.
If a smooth curve AB is given by equations
x — cp (/), ij — t|) (/), z = y. (0

* Kor brevity, wo shall sometimes writo the integral in the form ^ P dx -i

AD
fi
Q dy j /,’ dz or ^ P dx j- Q dy j- II dz unless this leads to misunder­

stand inir.
 Oil. 4. LI N E IN T EG R AL S 133

unci the value t = t0 of the parameter t corresponds to the point A

and t -- to the point B we have the relation
n
j P dx - f Q dy - f R dz = j (P (q> (£), \|: ( I), %(*)) W' ( 0 +
ab to
- K > ( c p ( 0 , t K 0 . x ( 0 ) ^ ( 0 + /? (q > ( 0 . ^ ( 0 . x ( 0 ) x ' ( 0 1 * (4 -3 6 )
which reduces the line integral of the second type to the definite
i ntegral.
The expression P coset f (J cos fi I- B cos y being the projec­
tion of the vector F - (/>, Q, Ji) on the tangent lineto the curve
AB, we can write line integral (4.35), as in the case of a plane curve,
in the form
\ t \ dl
AB
where Fx designates the projection of F on the tangent. All the
properties of the line integrals in the plane are automatically trans­
ferred to the spatial case. In particular, line integral (4.35) changes
its sign when the orientation of the path of integration is replaced
by the opposite one, i.e.
£ P dx + Q dy -j- R dz = — f P dx Q dy R dz
BA AB
Accordingly, in the definite integral entering into the right-hand
side of formula (4.30), Ihe lower limit of integration t0 should he
understood as the value of the parameter / corresponding to the
in itia l point A of the curve AR and the upper lim it /; as llm one
corresponding to the terminal point B irrespective of which of the
numbers t0 and is greater.

§ 3. GREEN’S FORMULA
Here we shall establish so-called Greens* formula which expresses
a relationship between a line integral
<£• P dx + Q dy
c
taken along the boundary of a domain and a double integral over
the domain. The formula is widely used in mathematical analysis
as well as in its applications. Some of the applications will be con­
sidered later.
* Green, George (1793-18'it), an English mathematician known for liis
results in the field of mathematical phvsics.
184 M U L T I P L E IN T E G R A L S . FIELD THEORY AND S E R I E S

1. Derivation of Green’s Formula. Let us first take a domain 6'.

of a particularly simple form, bounded by piecewise smooth curves
1/ — i/i (*), y = y 2 (*), (U2 (*) ^ V\ (*)) (4-37)
and by vertical line segments
x — a, x = b (6 > a) (4.38)
(see Fig. 4.7). We shall consider the boundary ABCDA of the domain
as being positively oriented in the sense that when describing the
contour thc^doinain is kept always on the left (in the particular

case of the domain G this means that the contour is traced counter­
clockwise). Let a function P (xt y) be delined and continuous
throughout the domain G including its boundary and possess the
OP
continuous partial derivative —• in this region.
Let us consider the double integral f f OP dxdy which we shall
c
try to transform into a line integral. To this end we reduce it to
b r a <x)
the iterated integral J dx £ ~ - d y and perform the integration
n Vx(T>
with respect to y. This yields
1) 2/2 (X )

(j a
f V i (x)
a ,j =
b
= J U* yz (*)) — v (•*% (*))] dx =
a
b b
= ( p C*, y 2 (•>•)) dx — \ P (X, y, (*)) dx (4.311)
a n
Each of the last definite integrals can he regarded as a line inte­
gral taken along the corresponding arc (see formula (4.27)),
namely as
b
j P(x, y2.(*))dz = | P (*» y) dx = — j P (z, y) dx
 CH. 4. LINE IN T E G R A L S 18!

ami
b
— j P (x, .Vi (x)) d x = — J P (x, y) dx
u AW
Now adding’ the two line integrals
—- I P (xy y)dx and — ^ P(i\y)dx
nc DA

to the right-hand side of equality (4.39) and taking into account

that both integrals are equal to zero (since dx = 0 on the vertical
line segments) we arrive at the relation
ff — | S' dx — I 'd x — f /> dx— ( S' dx
*G AR DC C l) D A

that is
f f ^ -dxdtj= — | Pdx (4.40)
G A DC DA

Kquality (•1.40) has been proved for a domain of a special form

bounded by Ihe curves of types (4.37) and (4.38). But the formula
can also be extended to an arbitrary domain which can be divided
into a finite number of parts of this particular form.
Kor, if a domain G with a boundary L is broken up into parts Giy
i — 1, 2.......... n, such that, for each part, the relation
if —
dxdy^-^P dx
o i-/
holds (where L t is the boundary of G,) we can sum up these relations
with respect to i from 1 In n and thus obtain the double integral
taken over the whole domain G on the left hand side and the sum
of the line integrals along the contours on the right-hand side.
But each contour L t consists of some arcs of the boundary of the
domain G and of some arcs of the auxiliary curves which divide
the domain G into parts. Kvory arc of each auxiliary curve being
a constituent of exactly two contours of the type Lf<t we thus see
that the line integral over such an arc is taken twice, the directions
of integration being opposite (see Fig. 4.8). Therefore, when we add
together integrals of ihe form
^ Pdx
h
the integrals taken over all the auxiliary arcs mutually cancel out
and ImWT 'Mily the integral along the boundary of the domain G
 MU L TI PL E INTEGRALS, FIELD THEORY AND S E R I E S

remains on the right-hand side. Thus, we obtain the equality

\ f ^ - d x d y = - f Pdx (4.41)
G
where L is the positively oriented* boundary of the domain G.

Let us now interchange the roles of the variables x and y and

Lake a domain G bounded by the horizontal line segments
y = c, y -= d (4/12)
and by the curves
x = x, (y), x = x2 (y) (4.43)
shown in Tig. 4.0. Let the function Q (*, y) and its derivative
^ be defined and continuous in the domain G (including its boun­
dary). Taking the double integral

o
in tlie form
rf *2(S/)

and performing the same calculations as in deducing formula

(4.40) wo receive the equality
^~~dxdy = f Qdy
ABCDA
analogous to (4.40) (with the only distinction that there is no minus
sign on tlu; right-hand side). Further, following the arguments

* As has boon mentioned, this means that the domain G is always kept
on the left of a person walking round the contour /, in the chosen (positive)
H ired inn tr*>*’ir?? Mm l»onri«lnr\F C.
 CI1. 4. LI NE IN T E G R A L S 187

similar to the ones given above we find that the relation

j I l i r dxdy = j Q dlJ (4.44)

G L
is true not only for the domains bounded by the curves of types
('4.42) and (4.43) but also for the unions of a finite number of such
domains.
A domain G which can be divided into parts with boundaries
of form (4.37), (4.38) and of form (4.42), (4.43) will bo referred to,
for brevity, as a simple domain. As has been shown, for a simple
domain, relations (4.41) and (4.44) are fulfilled. Subtracting (4.41)
from (4.44) we derive the formula
j !>dx + Q d y ~ j j ( - ^ d x d y (4.45)
G
where the line integral is taken along the boundary L of the domain G
in the positive direction. This very formula, which we set out to
prove, is called Green’s formula. Thus, we can state the following
result.
Theorem- 4.4. Let G be a simple domain and functions lJ (x, y)
and (J (x, y) and their derivatives be defined and continuous on the
closure of the domain G (i.e. on the union of G and its boundary). Then
Green's formula (4.45) holds.
ATofe /. If the boundary L of a domain C is compused of a finite
number of separate contours the symbol ^ P dx + Q dy should be
1.
understood as the sum of the integrals taken over all the contours

Fig. 4.10

the entire contour is formed of, and each contour is described in the
direction such lliat the domain G ahvavs remains on the left of the
contour (see Fig. 4.1U).
Note 2. hi deducing Green’s formula wc have supposed that the
functions P and Q and their partial derivatives ^ and ^ are con­
tinuous not only in the interior of the domain G hut also on its
fiftund iry Hot M t o m f o o l that it i.- .Tiilficu at iniin^c tin- couui-
188 MULTIPLE IN T E G R A L S . FIELD T H E O R Y AND S E R I E S

tion that the derivatives — Oy

and —ax
arc continuous and bounded
in the interior of the domain G. Indeed, take again a domain G
bounded by curves y — y\{x), y = ijz(x) and vertical line segment.4?
x - a, x = b (Fig. 4.7). Choose a sufficiently small number 6 > 0
and consider the domain Gf, bounded above and below by the curves
y = ij2 (x) — 6 and y = y,(x) -f- fi, respectively, and on the sides
by the vertical lino segments x = a 4- 6 and x = b — 6. For every
sufficiently small 6 > 0 such a domain G0 exists and is strictly
contained, together witli its boundary, within <7, and hence all the
conditions for which the validity of equality (4.41) has been estab­
lished are fulfilled for Gq. Therefore,

H dxdy ^ ” j P d x (4.4(>)
G6 /'0
where is the boundary of the domain The difference between
thb area of the domain G6 and that of the domain G not exceeding
the quantity 16 where I is the length of the boundary L of the
domain 6’, the integral on the left-hand side of relation (4.46) differs
from

G
not more than by the quantity 16M where M is the supreinum of
OP . . . ,,
inside 6. Furthermore, the function I* (x, y) is contin ous
On
and therefore uniformly continuous and bounded in the closure
of the domain G. It follows immediately that
j P dx j P dx as 6 - y 0
Ls l
Hence, we can pass to the limit in equality (4.46), for 6 —>0, and
thus prove that the relation

nG
- - j™, 1,
is true for the domain shown in Fig. 4.7 and, consequently, it
applies to any simple domain. The validity of the equality

is established in a similar way.

It should be noted that the requirement that the derivatives
fit/
and 4^-
rfx
are bounded in the domain G can be replaced bv the
 CH. 4. L I N E IN T E G R A L S 1}

condition that the integral f f ---- dxdy (understood as a

’G
improper double integral in case the derivatives are unbounded
exists.
Note 3 . Green’s formula has been proved here for the domaii
which we have agreed to call simple. All the polygonal figures belon
to this class of domains. Applying the technique of approximatin
a curvilinear domain by polygonal figures we can easily prove thr
Green’s formula is also valid for any domain bounded by a iinit
number of piecewise smooth curves.
2. Application of Green’s Formula to Computing Areas. Green
formula implies some useful formulas for computing the area <
a domain.
Cat G be a simple domain with a boundary L and let S be til
area of the domain. Consider the line integral
x dy

Applying Green’s formula to the integral we obtain

^ x dy = ^ J dx dy —S
/. **a
Next, we similarly derive the formula
S = — ^ y dx
L
Combining these formulas wc deduce another formula for compuliu.
areas in which the integrations with respect to x and y are involve!
symmetrically’"*:
S = y j x dy —y dx (\ A 7
L
Example. Compute the area of the domain bounded by the aslroit
x s= a cos3 t, y a sin3 t

* The notion of an improper integral will bo discussed in Chapter 9.

** It is .apparent that we ran receive infinitely many formulas of the fern
S — j P dx -f Qd y
L
To this end it is sufficient to take, for P and Q, any two functions satisfyinj
tho condition - - — — — I.
*U dy
19u MULTIPLE INTEGRALS, F IE LD THKORY AND S E M E S

Solution. Applying formula (4.47) wo obtain

2rc
.V - 4" J x dy ~ ydjr = Y ai ^ sin21 cos21 [cos2/ j- sin3 /] dt =
i. o
2ji
——■a2 J sin221dt = jui-
o

§ 4. CONDITIONS FOR A LINE INTEGRAL

OF THE SECOND TYPE BEING PATH-INDEPENDENT.
INTEGRATING TOTAL DIFFERENTIALS
1. Statement of the Problem. When studying the examples uf
line integrals in § 2 we have* noticed that in certain cases a line
integral
j P d x \-Qdy
AR
is independent oT the shape of the curve AB and is determined only
by the position of its end-points, that is .assumes the same values
for all the paths of integration joining the fixed points A and B.
Hero we are going to establish the general conditions guaranteeing
such independence of a line integral of the particular choice of the
path. This question is closely related to the problem of finding
a function of two independent variables from its total differential
which we are al*** gf,ing to consider hero.
2. The Case of a Simply Connected Domain. We remind the reader
(s<*«* Chapter 2) that a plane region (domain) G fc said to he aimply
connected if each closed contour L lying in the interior of the domain
hounds a (finite) part of the plane entirely belonging to 6T.
Theorem Let functions P (x , y) and Q {x%
y) and their partial
derivatives ~a y and —
d.r
he defined and continuous in a bounded closed
simply connected domain G. Then the following four conditions are
equivalent to each other (in the sense that the validity of each con-
dition implies the fulfilment of the other three):
1. The integral
<£> Pdx-\ Qdy
taken over an arbitrary closed contour lying within G is equal to zero.
2. The integral
^ P dx 4- Q dy
AH
 CH. 4. LINE INTEGRALS 191

is independent of Hie particular choice of the path of integration con­

necting the points A and IS (which are considered to be fixed but can
be chosen arbitrarily in the domain G).
3. The expression P dx — Q dy is the total (exact) differential of
a single-valued function defined on (he domain G.
/i. The relation
OP _ dQ
Otj Dx (4.48)
holds everywhere in the domain G.
Proof. We shall prove the theorem by following the logical scheme
4 -*■ 1
i.e. show Mint tile first- condition implies the second condition, tin*
second implies I lie third one, the third implies the fourth and the
fourth, in its turn, implies the first condition. Then the equivalence
of all the four conditions will be established.

Fig. 4.11

(a) i ►2. Take two arbitrary paths lying in the domain G ami
connecting the points A and IS. say the paths AC IS and ADIS shown
in Tig. 4.11. Xow consider the closed contour ACISDA composed
of them. Ity the hypothesis, the integral taken along any dosed
contour is equal In zero and thus
j P dx ’ Q dy 0
ACDDA
But wc have
| P dx + Q dy = ^ P dx — Q dy -{- ^ P dx Q dy -
A C■J*—
WA A C II IID A

— I" P dx-\-Q dy — ^ P dx -|- Q dy

A C ll AIUI
and consequently
^ P dx -+- Q dy ~ ^ P dx Q dy
actj a )j b

Hence the assertion “1 — 2V has been proved.*

* If the curves A CJi and ADR have some points in common other than
A and B (see Fig. 4.12) the same result can be established by applying a little
more complicated argument.
192 M U L T IP L E IN T E G R A L S , F IE L D THEORY AND S E R IE S

(b) 2 3. Let the integral

^ P dx -j- Q dy be independent.
An
of the path of integration. Then, if we fix the point A the integral
can be regarded as a single-valued function of the coordinates x
and y of the point B:
j P dx -b Q dy = U (x, y)
AD

Let us show that the function U (x . y) is differentiable and that

dll = P dx t- (J dy
ou
To prove this it is sufficient to show that the derivatives
OCJ
and exist and are, respectively, equal to P (x, y) and Q( x y y).*
Let us compute the limit
OU ___ j - V (c r y) — U (x, y)
0x ~ Ar-*0
The quantity V (x -f Ax, y) -- U (x, //) is equal to the integral
of P dx + O dy taken along an arbitrary path connecting the points

not depend on the shape of the curve joining the two points, lienee,
we can take the path coinciding with the horizontal line segment
B B i (Fig. L13). Therefore, applying the mean value theorem we
receive
f.\x . >)
AU
Aar ==-£• f I’ d x ' Q d , , ^ \ P (x, //) dx — P (x 4- 0Ax, y)
BDi X , p

where 0 <C 0 < 1. Consequently, we have

= lini P (r 0 \x , it) P (x, y)

tU A.V--0

* As is well known, a function possessing continuous partial derivatives

is differentiable.
 CH. 4. L I N E I N T E G R A L S 193

because P (.r, y) is continuous. The rotation ——- Q (.c, i/), is

proved similarly.
(c) 3 —►4. If the expression P dx -j- Q dij is (lie total differential
of a function (' (./■. ;/) we have
OU OU = (J
dx = /> oy
Then the well known theorem asserting that the mixed deriva-
lives —:- . - and —
OX Off
—— are equal when they are continuous implies
<nt ox
OQ j& L Q'-U or
t/j ox ()y Oy ox Oy
(d) 4-*■ 1. Let the equality
1
— — Oy bo fulfilled and let L he an
‘ ox
arbitrary closed contour lying in the domain G. 'flic* domain being
(by the hypothesis) simply connected, the part of the plane bounded
by the contour L belongs to the domain G in which Ilie rune-lions
P and (/ and their derivatives are defined and continuous. Therefor©,
by Green's formula, the line integral
^ P dx \-Qdy
L
can be transformed into the corresponding double integral

\
L
< 'tU ' IJ
) I
where D is t i l e domain bounded by the contour L. By virtue of
relation (4.48), the integral on the right-hand side is equal to zero.
Consequently, we have
^ P dx Q dy 0
1.
for any closed contour Ij lying within G. The proof of the theorem
has thus been completed.
3, Reconstructing a Function from Its Total Differential. In proving
Theorem 4.5 we have incidentally solved the following problem
(which will be again encountered in § 2, Sec. 4 of Chapter 6 ): given
an expression
P dx -f O dy
it is necessary to find a function whoso total differential coincides
with the expression. In this section we shall limit ourselves l.o the
r?P
case when iho functions P and Q and their partial derivatives —
1 3 - (182
194 M U L T I P L E I INTEGRALS, FIELD TH E O RY AND S E R I E S

and ^ a r c continuous over a simply connected domain G. As has

been proved, in these conditions, the expression P dx — Q dy is
the total differential of a function of two arguments if and only if
tlie equality
dl> OQ
t)y Ox
holds (see Theorem 4.5).
Furthermore, in the same proof we have shown that if the above
equality is fulfilled the relation
dC = P dx + Q dy (4.49)
is satisfied by the function

U (x ,y)=
J Pdx+Qdy
(*0 . t/0>
Finally, the formula of finite increments (e-g. see (81, Chapter 8 ,

Fig. 4.1-4 In
§ 9) suggests that two functions having the same total differen tial
may only differ by a constant term. Consequently, the formula
(X.v)
U (ar, y) — j P d x + Q d y —C (4.50)
(*0 , '/<!>
where (.x0, y 0) is a fixed point and C is an arbitrary constant des­
cribes a one-parameter family of functions which contains all the
functions satisfying condition (4.40). The integral entering into
equality (4.50) being path-independent, wc can take at pleasure the
curve connecting the points (x0t ?/o) and (x , y). For example, it is
convenient to choose, as the path of integration, the broken line
composed of the horizontal and the vertical line segments shown
in Fig. 4.14.* When the path of integration is chosen in this way
equality (4.50) takes the form
(* , yo> ex.^ j

V ( r, J/) - - \ P d x f- j Q dy ' C
__________________ Vo) (jc,j/o)
* PritvHwl il»o l.olone to If11>domain (I.
 CH. A. LI N E 1N TEUKAL S 195

The initial point (x0, y0) can bo arbitrarily taken within the domain
in which the functions P and Q arc defined. A change of the position
of the point (x0, y 0) is obviously equivalent to a variation of the
value of the arbitrary constant C.
Prnclicallv,
V 7 to determine a function from its total differential
it is convenient to apply the following technique. If we have

%r = P ’ T5T-<? (4 ‘51)
then, integrating the first equality with respect to a: and considering
tlie variable y entering into it to he a parameter, we obtain
V (*. y ) — f P d x + fi (4.52)
4r
where /, is independent of x (but, generally speaking, may depend
on y , i.e. /, = /, (y)). Further, integrating the second equality (4.51)
with respect to y while x, in its turn, is regarded as a parameter,
wc receive
£/(x, y ) = j Q d y - \ f 2 (4.53)
where / 2 = / 2 (x) If we now match the functions /, (t/) and / 2 (x)
in such a way that the right-hand sides of relations (4.52) and (4.53)
coincide this will result in a function whose total differential
is equal to tlie expression P dx 4- Q dy.
Example. Let
dU = (2xi/ 4- i) dx -}- (.r2 -f- 3r/2) dy
Integrating the coefficient in dx with respect to x we derive
j (2xy 1- 1 ) dx = xzy-\ x-f /i iy) (4.54)
The integration of the coefficient in dy with respect to y results in
j (x 2 I- 3y z) dy - - xhj -f y2 !- h (*) (4.55)
The right-hand sides of equalities (4.54) and (4.55) coincide if we
put
h (y) — y3 -i- C and / 2 (x) — x -J- C
Thus, we see that
U =- x2y 4- x 4- y3 -f C
4. Line Integrals in a Multiply Connected Domain. In the proof
of Theorem 4.5 we have taken advantage of the fact that the domain
C has been supposed to be simply connected when, on the basis of
the condition
OP dQ
190 MULTIPLE INTEGRALS, FIEL D THEORY AND SERIES

we have established the validity of the equality

P dx r Q dij —0 (4.57)
L
for any closed contour L lying in G.
Now we shall consider a simple example indicating that, gene­
rally speaking, equality (4.57) is nut implied by condition (4.56)
in the case of a multiply connected domain. Led
(4.58)

The integrand dues not make sense at the point. (0 . 0 ) and therefore
we shall delete a neighbourhood of the origin of coordinates. In
the plane witli the neighbourhood deleted Obis plane is now an
infinite multiply connected domain) the coefficients in dx and dij
are continuous and possess continuous partial derivatives. Besides,
we have
— I \ - 0 ( x
Oy \ / dx V * 2 -i if-
But integral (4.58) taken along a closed contour turns out to be
different from zero in the general case. Fur instance, if C is the
circle determined by Ilie equations
x = cos /, // =: sin /
we obtain
2-t
(4.59)

Now let us investigate the general properties of an integral

^ P d.c r - O dij
in case the functions P and Q satisfy the condition*
riP __ f)(J_
dy Ox
hut the domain G they are defined in is multiply connected. For
definiteness, let us lake the domain G depicted in Fig. 4.15, i.o.
the one having 1 liree “holes”, “lacunas1’, Gt . G2 and G3. Lot us first
consider a closed contour L which does not contain any lacuna inside
it. Then we can apply Green’s formula to the integral taken over
such a contour and thus we see that the integral is equal to zero.

* A s b e f o r e , w e s u p p o s e t h a t t h e f u n c t i o n s / J. 0 , — and ^ are c o n t i n u o u s
11 oy Ox
 CH. 4. LINE INTEGRALS 197

Now let L\ be a contour enveloping one of Ihe lacunas. say the

lacuna Gt (and not containing the other lacunas), Croon’s formula
is no longer applicable to this case and, generally speaking, the
integral taken over such a contour is not equal to zero (see the
above example).
Let us show that the value of this integral is independent of the
particular choice of such a contour containing tlu* lacuna. Let

and L\ be two such contours. Connecting them by an auxiliary curve

(ab) we obtain the contour
(ab) + /., (ba) - L\ (4.60)
where the minus sign in front of L\ means that the contour is tra­
versed in the negative direction. Contour (4.60) envelopes no lacunas
and hence the integral over it is equal to zero. But the integrals
taken along (ab) and (ba) are equal in their absolute values and
have the opposite signs. Thus we obtain
j P dx 4 Q dy [ j P dx |- Q dy ~ 0
U -C
that is
j P dx -i Q dy — I' P dx \ Q dy

Consequently, each lacuna G, (i = 1, 2t 3) in the domain G

can be characterized by a certain number on, namely by the value
of the line integral ^ P dx Q dy taken round an arbitrary closed
contour containing the lacuna inside it and not enveloping any
other lacuna. Now we can pul down the general expression for the
integral <^i P dx -f- Q dy taken over an arbitrary closed contour L
lying in 0. Suppose the contour L passes around the first lacuna A*j
times, around the second lacuna /»*« times and around the third one
/»*3 limes. Here, for kt (i — 1 , 2, 3) we each time lake into account
t he direction in which the moving point of the curve L passes around
198 MULT1FLE IN’TEGHAIS, FIELD THEORY AND SERIES

the ith lacuna, and thus every number is understood as an algeb­

raic sum equal to the difference between the number of times the
contour passes around the lacuna in the counterclockwise direction
and the number of times it passes around it in the clockwise direc­
tion. Then we obviously have
(^) P dx —
—^ dy — &,(!), —
j— -I—/ijWj
L

If we make the cuts I. I T and T i l , shown in Fig. A . lb, in the

domain G we obtain a simply connected domain in which we can
construct the single-valued function
<*. y)
U (*> y) = J P d x + Qdy (A.Gl)
(*o. vo)
But, according to what has been said above, the values of the func­
tion differ, on the opposite edges of the slit / , by the quantity coj.

on the edges of the slit IT by and on the edges of ITT by to3. If

\vc do not make the cuts expression ( 1 .0 1 ) again represents a func
tion whose total differential is equal lo P dx + Q dy hut in this
case the function is multiple-valued. Its values, at a fixed point,
corresponding to the contours passing around the lacunas several
times differ from each other by terms of the form
A*](i)j -j- h 2 U)2 A3 W3
where the numbers k^ k 2 and k 3 can be arbitrary integers (positive,
negative or zero).*
It appears clear that all that has been said here is automatically
extended to the general case of an arbitrary number of lacunas.

* I t m ay turn out, of course, th a t .all the numbers are equal to zero iti
(*, v)
a p articu lar case. Then the function U (r, y) = £ P d x -f- Q d y is single
<*0 . m) .,
valued even when the cuts are not made, and all the assertions of Theorem 4.o
r A,v'.rl i ? ♦ »■•» n fny* o m n l t fnl u P A t M i n r l o#l rlrw nnin
^ Surface
Integrals

In some physical problems we encounter functions defined on

various surfaces. Examples of such functions are the density of
a charge distribution over the surface of a conductor, the intensity
of illumination of a surface, the velocity of the particles of a fluid
passing through a surface and the like. The present chapter is devoted
to studying integrals of functions defined on surfaces, the so-called
surface integrals, and some of their applications.
The theory of surface integrals is in many respects analogous to
1 lie theory of line integrals presented in the foregoing chapter. In
particular, wc shall distinguish between the surface integrals of the
iirst and the second types.
When introducing the definition of a surface integral we shall use
some notions concerning surfaces which were discussed in §§ 3
and 4 of Chapter 3 and, particularly, the notion of area of a cur­
vilinear surface.

§ 1 . SURFACE INTEGRAL
UK THE FIRST TYPE
1. Definition of Surface Integral of a Scalar Function. Let 2
he a piecewise smooth surface hounded by a piecewise smooth con­
tour />.* Consider a hounded function / (A/) defined at the points
of the surface. Dreak up the surface 2 into parts 2 j, 2 2, • . 2 n
(Fig. 5.1) by means of piecewise smooth curves and denote the areas
nl the parts as o t (i •= 1, 2, . . ., n). Next wc choose a point M t
in each part 2 ,- and form the sum

T = 1' / ( Mi ) Oi (5.1)
i=l
which will be referred to as an integral sum corresponding to the
function / (A/) (for the partition { 2 *} (i — 1 , 2 , . . ., n) of the
surface 2 and for the* given choice of the points A/,).

* In particular, the surface ^ may he closed and have no boundary.

2 00 M ULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

Wc introduce the following

D e f i n i t i o n . Jf the integral sums T tend to a finite limit as the
maximal of the diameters of the parts of the surface 2 tends to
zero the limit is called the s u r f a c e i n t e g r a l o f th e f i r s t t y p e
o f the f u n c t i o n f (AT) o v e r th e s u r f a c e S and is denoted by
the symbol
\ f / (.10 do (5.2)
1 V
2
The moving point M on the surface 2! con he determined by
its Cartesian coordinates x, y and z. Therefore a function f (AI) defined
on 2 will also be designated as / (x, y y z) and the corresponding
surface integral as J J / (x, y t z) do. But when using the last
E

Fig. 5.1

notation one should bear in mind that the variables x, y and z are
not independent here but connected by the condition that the point
(x, y, z) belongs to the surface 2 .
2. Reducing Surface Integral to Double Integral. We have for­
mulated the definition of the surface integral of the first typo and
now we are going to discuss the conditions for its existence and the
methods of its practical computation.
Both questions are easily answered if we reduce the surface inte­
gral to the double integral.
To begin with, we take the simplest case when the surface in
question is represented by an equation in Cartesian coordinates.
'Theo rem t. Let 2 be a smooth surface determined by an equation
z = z (x, p), (x, y) £ Z>, where D is a bounded closed domain, and
lei f (x, i/, z) be a bounded function defined on the surface 2 . Then
we have the relation
j j / (*»*/> z) jf / y> 2 ( z , y ) ) V r l z * Z y l dx dy (5.3)
 CII. 5. SURFA CE IN T E G R A L S 201

provided that the integrals entering into it exist. The surface integral
on the left-hand side of equality (5-3) exists if the double integral on
the right-hand side does.
Proof. Divide the surface 2 into n parts 2f (i = 1. 2, . . . » /*)
by means of piecewise smooth curves. Projecting this partition on

the x, y-plane we obtain a partition of the domain D into squarable

parts Df (see Fig. 5.2) in which the diameter of each element D t
(e = 1 , 2 , . . ., n) does not exceed the diameter of the correspon­
ding element 2 j of the surface 2 .
Consider an integral sum
n
T— 2 / (*/i yi* zi) °i (5-4)
1=1
associated with the surface integral j ^ / ( j , j/, z)do. The area a-,
of the element 2 4 can be written in the form
= •r\ v\ 1^1 a z?-\- z'y dxdy

where z — z (x, y). Now taking advantage of the mean value theo­
rem for the double integral of a continuous function* we rewrite
the formula for a* as
ot = V l - {-z* (a:?, yt) -|- v ( zf, y*)Si
where (x'f, {/*) is a point belonging to the domain Di and Si is the
area of the domain. Consequently, integral sum (5.4) can he put
down in the form
rt

T = 2 /(* t, yi, s (x {, yi)) V i T Zx yt) 4- z* <xf, y t ) S t (5.4')

j=l
* The surface z ■=■ z (.r, y) is supposed to be smooth and hence the ex pres
sion ~\/1 4 - z!* (x, y) -J- (j, y) is a continuous function.
202 MULTIPLE IN T E G R A L S , F I E L D T H E O R Y AND S E R I E S

Compare this sum with the integral sum

n
T — y, 1(xi, y-t, z ( x i%iff)) K l+ Z x yi) + z'f{*i, y l) S l (5.5)
i=l
corresponding to the double integral on the right-hand side of equality
(5.3) (associated with the partition, of the domain Z), generated
by projecting the partition { 2 *} of the surface 2 ).
The only distinction between the sums (5 .4 ') and (5 .5 ) is that in
each summand entering into (5.5) the values of the function / and
of the expression V \ + z'x + z'y are taken at the same point (xty y t)
arbitrarily chosen within the clement D t whereas in (5.4') the values
of V i -i- Zx H- z'J are taken at the point (-r*, y*) whose position
is preassigned by the mean value theorem. Therefore, in the general
ease, the point (xf, ^f) does not coincide with the point (xit y t)
although it belongs to the same element D t.
The function Y 1 + zx -h z'y is continuous in the bounded closed
domain D and hence it is uniformly continuous there. Consequently,
given any e > 0 , there is 6 j > 0 such that
| Y 1 z .;4 ( x *, y t) + z * {Xi , iji) — \ f i yt) | < e
-f z? (xf, y\) -f- z '* ( x f ,
(5.6)
if the maximal of the diameters of the subdomains Z?j is less than
6 ,. By the hypothesis, the function / (x , y , z) is bounded, i.o

I / (*» y, z) | ^ K = const
aud therefore relation (5 .G) implies the inequality
tt
i T —T j = j S / (x/t yiy z (Xi , yt)) [ Y 1 -j- z x- ( x f , yt) ■+■Zy (xt, yt) —
i=l

— \fi + y d ^ ^ y i x i , y,)l Si |<A :e ^ S t = K z S (5.7)

where S is the area of the domain D.
Now we can easily complete the proof of the theorem. If the inte­
gral on the right-hand side of (5.3) exists, for every e > 0 there is
6 2 > 0 such that for any sum T corresponding to a partition (D*}
of the domain D whose elements are of diameters less than 6 2 we
have the inequality
j \ \ f ( x , y t z (x, y)) Y 1 -i- z'x (*» y) -h z'v2(*. i/) dx dy— T j < e (5.8)
'd
Let us take the number 5 = inin (6 j, d2) and consider the partitions
{ 2 *} of the surface 2 for which the diameters of all the elements 2 ,
arc less than 6 . Denote by {/Z*} the partitions of the domain D
 CH. 5. SURFACE I N T E G R A L S 203

corresponding to {2 ,}. Then the diameter of each Di is less than 6

and consequently inequalities (5.7) and (5.8) are fulfilled. The ine­
qualities imply that
/(x , y, z(x, y)) |A + 2 ;l (i,y ) + z'*(x, y) dxdy — T < e (1 + /C9)

for every partition of the surface 2 whose fineness is small enough.

It follows that the limit of the integral sums T exists and equals
the integral entering into the right-hand side of relation (5.3), and
hence the theorem has been proved.
CoroMa rf/. If the surface 2 is smooth and the function f (x, y , z)
is continuous the integral
j [ / (x, y, z) da
is sure to exist-
For, in this case we have a double integral of a continuous func­
tion on the right-hand side of equality (5-3) which exists and thus
the surface integral on the left-hand side also exists.
Note 7. We have
1
cos(n, z)
(see § 3, Sec. 6 in Chapter 3) and therefore equality (5.3) can be
rewritten as
j j /(*. y . 2 ) j / (x. y . z(*. </)) (5-9)
£ 1J

If a surface 2 is represented by an equation

x = x (y, z)
we can interchange the roles of the variables x t y and z and write
the relation
\ j /(*» If. *) d a = j j /(x (y , z), y, z) c ^ y{^ x) (5.9t)
2 D t

where D\ is the projection of the surface 2 on the y, z-plane. Simi­

larly, in the case of a surface 2 defined by an equation
y = y (z, x)
we have the equality
(5.9j)
n / ( ** y*z) d o = n / ( x ’ !/(j’ i ) ’ z ) ^ £5)
£ D2
where D is the projection of 2 on th'* "• x plane.
20'i MULTIPLE I N TE GR AL S . FIELD THEORY AND S E R I E S

Note 2. Suppose a surface 2 is composed of several pails each

of which can be represented by an equation of the form
* = x (y, z), ij = // (z, x) or 2 = z (x, i/)
Then we can take advantage of the fact that the surface integral
over 2 is equal to the sum of the integrals token over the parts the
original surface is formed of and apply formulas (5.9). (5.9,) or
(5.92) to each of the integrals sep arately and thus reduce the in tegral
over 2 ! to the sum of the double integrals.
In case a surface is represented by parametric equations we can
apply arguments essentially the same as above and thus prove the
following result.
T h e o r e m 5. Let 2 be a smooth surface represented by a (vector)
parametric equation
r = r (m, p)
and f (x, y , z) a bounded function defined on the surface. Then we
have the relation
\ \ /(x , y , z) do =
}vJ
9

-- ^ j f ( x ( u , V), //(«, 1;). -(«» 1;)) Vg\\X2!t — g\*<hl dv (5.10)

*D
provided that the integrals in (5.10) exist. The surface integral on the
right-hand side exists if the double integral on the left-hand side does-
Here D is the range of the parameters u, v and g,,, g 22 are
the fundamental coefficients of the first order of the surface 2 (see
$ 4. Sec. 1 in Chapter 3)- The expression "j/g^g^ — <?*f, du dv is an
element of surface area in the curvilinear coordinates u, v.
Hence, «formula
• (5.10) means that in order to write a surface
integral | j / (x, //, z) do, taken over a surface 2 which is .deter-
*v
mined by an equation r = r (u, v), in the form of a double integral
we must replace the Cartesian coordinates x, y and z of the points
of the surface by their expressions in terms of the curvilinear coor­
dinates u ami v and substitute, the above expression of on element
of surface area for do.
Formula (5.3) and formulas (5.9), (5.9,) and (5.92) are obviously
special cases of general formula (5.10). It can be easily shown that
these formulas are also valid when the surface is not smooth but
piecewise smooth.
3. Some Applications of Surface Integrals to Mechanics. Surface
integrals of the first type are frequently encountered in physical
problems. For instance, this is the case when we deal with a mass
 CI I . 5. S U B P A G E INTEGRALS 121

distribution over a surface and find the coordinates of its centre •

gravity, moments of inertia etc. The corresponding formulas ai
derived by essentially the same methods as those applied to studyin
mass distributions over a plane figure or along a curve (st
Secs. 3-h, § 4 in Chapter 1 and Sec. 3, § 1 in Chapter 4) and therefoi
we shall only present the linal results and leave the computatioi
to the reader.
Let a mass of areal density p (x , ?/, z) he distributed over a su
face Z which is smooth or piecewise smooth. We shall suppo;
that the function p (x, u, z) is continuous on Z and refer to sue
a surface, for brevity, as a material surface. Then we have the follov
ing results.
(J) The mass p of the material surface Z is equal to
p — i*
\ '*
\ P(r, i/, z) do
J
(2) The coordinates of the centre of gravity of the material surfai
are expressed by the formulas
\ J x p ( r , y , z) d o J J f/p ( r . r/f =) d o \ J 20 (x. y. =) d o
V V V
vp (jf. y , 2) d a ’ ^ J J p ( r . fir, z ) d o ' \ jk P ( r, 0. 2) d o
V *V *V

In particular, for a homogeneous surface (with p = const) we ha 1

$yS 1 in J ) do ft<»\ - da
V
zc

(3) The moment of inertia of the surface Z about the z-axis

equal to
f f ( z 2-^ij2)o(x, y, z)do

and the moments of inertia about the other two axes are expresse
si m ilarly.
4. Surface Integral of a Vector Function. General Concept ■
Surface Integral of the First Type. We have considered surfai
integrals of scalar functions. This notion can he easily generalize
to the vector functions. Let
F (M) = Pi h Qj + n k
he a vector function defined on a surface Z. Let us introduce tl
integral of such a function over the surface Z bv putting
do — i P{M)do ; j j f <)(.U )da-; k It(M)do

rr» I
2 0 t) MULTIPLE IN T E G R A L S , FIELD THEORY AND S E R I E S

Expression (3 - 1 1 ) will be called the surface integral of the first

type of the vector function F over the surface 2 . The value of such
an integral is a vector. The question of existence of a surface integral
of the first type of a vector function F, the problem of reducing it
to a double integral and the properties of the integral are investi­
gated on the basis of the corresponding facts concerning the integrals
of the scalar functions / \ Q and II which are the components (coor­
dinates) of the vector F.
To illustrate the application of this notion let us find the force
of gravitational attraction with which a material surface attracts
a material point.
Let p (x, y , z) be the density of mass distribution over a surface
2 and m0 be a mass concentrated at a point (x0, yo» zo) ,l°l belonging

to the surface. An element of area do carries the elementary mass

p (x, //, z) da and, by Newton’s law of gravitation, the elementary
force with which it attracts the mass point m0 is equal to
I* ^
dF = ym0p (x, y, z) do(5.12)

Here y is the constant of gravitation whose numerical value depends

on the choice of the system of units and r is the vector drawn from
the point (x0, y a, z0) to the point (x, y t z) (Fig. 5.3). The resultant
force F of attraction of the material point n? 0 by the entire surfaco 2
is equal to the sum of elementary forces (5 . 1 2 ), that is to the surface
integral

V'"o \ \ f>(*\ i/» 2)

A.’
 CH. 5. SUR FAC E I N T E G R A L S 207

Since we have r = (x—x0) i | (y — y0) j 4- (z — zQ) k, the expression

of the force can be written in the form

The last integral is sure to exist if the surface 2 is smooth or piece-

wise smooth and the surface density p (x, y f z) is a continuous
function on 2 .
An essential feature of the surface integrals of the first type is
that each element of integration of the form
f (M) da
depends solely on the magnitude of the element of area da .and on
the value of the function / (M) (which may be scalar or vector) at
the corresponding point M but is independent of the orientation
of the surface element da in the surrounding space. This is the case
in the physical problems wo have considered here because the mass
of an element of a material surface at a point M or the force with
which tho element attracts a material point does not vary if we
arbitrarily turn the element about the point M.
But there are problems of another kind in which the orientation
of the element da plays an important role. Such is the problem (to
bo considered below) of computing the amount of liquid passing
through a surface in unit time and some others. These problems
lead to another concept of a surface integral, namely to the so-called
surface integral of the second type which we shall deal with in $ 2 .
As will be shown, there are simple formulas expressing the rela­
tionship between the surface integrals of the first and the second
types.

§ 2. SURFACE INTEGRAL OF THE SECOND TYPE

1. One-Sided and Two-Sided Surfaces. To give the definition of
the surface integral of the second type we must first discuss the
question of choosing a side of a surface which is analogous to the
problem of introducing an orientation of a curve.
Let 2 be a smooth surface. Take an interior point M 0 on 2 . draw
the normal line to the surface at the point and choose one of the two
possible directions on the normal. This can he achieved by fixing
a certain uuil normal vector u to 2 at. the point jl/0. Next we con­
sider an arbitrary closed contour C lying on the surface 2 which
passes through the point ;l/ 0 and has no points in common with
the boundary of the surface. Imagine that we are carrying the unit
206 M ULTIPLE INTEGRALS, FI EL D T H E O R Y AN D S E R I E S

vector n along C (starting from the point M 0) in such a way that

the vector always remains perpendicular to 2 , i.e. perpen­
dicular to the tangent plane drawn to the surface through the origin
of n (which lies on 2 all the time), and so that the direction
of n continuously varies in this motion. The vector n always
remaining normal to the surface 2 , there are only two possibilities
here: (1) after the contour C has been traversed and the moving
point (the origin of n) has returned to the point M 0 the new position
of the vector n coincides with the original one; (2 ) after the contour
C has been traversed the vector n changes its direction to the oppo­
site. Accordingly, we introduce the following
D e f i n i t i o n . .4 smooth surface £ is said to be t iro-s if/eft if after
an arbitrary contour. lying on the surface 2 and having no common
points with Us boundary, has been described the normal vector to the
surface does not change its direction.
If there is a closed contour such that after it has been traversed the
normal vector changes its direction to the opposite the surface is called
onc-xi<teU.
If 2 is a two-sided surface we can choose a unit normal vector
n(:17) at each point .1/ of the surface so that n(4/) continuously
depends on the point AT. To construct such a vector function
n(;l/) wo can choose an initial point Afo on 2 and one of the two
possible directions of the normal vectors n(;l/0) at the point. Then
wc take an arbitrary point M on 2 , connect it with M 0 by a curve
L lying on 2 and carry over the vector n along L from \ f 0 to M
so that it always remains normal to the surface and its direc­
tion continuously varies in this motion on. The vector n(.W)
at the point. M thus obtained is independent of the choice of the
curve L joining tlie points M 0 and M . For, if two different curves
L t and yielded different results wc should have formed a closed
contour consisting of the curves L x and L 2 and thus obtained a closed
path C lying on 2 such that after C has been traversed the direction
of the normal vector is reversed. But this means that the surface
is not two-sided which contradicts the hypothesis.
It clearly follows that, on a two-sided surface 2 , there exist
exactly two functions of the type n{M) continuous throughout 2 .
Indeed, each function is completely specified by choosing one of the
two possible directions of the normal at an arbitrary initial point
on the surface. A function t\(M) of this kind will be referred to as
a continuous field of normals on 2 . In the case of a one-sided surface
it is obviously impossible to construct any continuous field of nor­
mals.
When we choose one of the two possible continuous fields of nor­
mals on a two-sided surface 2 we thus choose a certain side of the
surface (which is seen from the lip of the normal vector).
 CII. 5. SURFA CE IN TE GIi ALS 209

Examples
1 . A plane is a simple example of a two-sided surface. Any part
of a plane, e.g. a circle, is also a two-sided surface.
2 . Every smooth surface determined by an equation z — / (x, y)
is two-sided. In fact, if, at each point of the surface, we take the

normal vector whose direction forms an acute angle with the posi­
tive half-axis z we obtain one (upper) side of the surface, and the
oilier (lower) side if the orientation of tlie normal vector is reversed
(Fig. 5.4).
3. Any closed surface without self-intersections is two-sided,
e.g. a sphere, an ellipsoid etc. For instance, if we take, at each point
D C

Fig. 5.f» Fig.

of such a surface, the normal vector directed toward the interior

of the solid bounded by the surface wc thus indicate the inner side
of the surface.
4. The so-called Mobius* strip depicted in Fig. 5.5 is the simplest
example of a one-sided surface. It can he obtained by taking a rec­
tangular strip of paper ABCD (Fig. 5.0a) and pasting its two ends
together after giving it half a twist, i.c. so that the point .4
coincides with the point C ami the point // with the point I)
(Fig. 5.05). It is easily seen that after we traverse the centre line
of the Mobius strip the direction of tTie normal to the surface is
reversed, which means that the surface is in fact one-sided.

* Moltin*. A*i<tvc+ FeHtn-md 'ITf-O 1SCS), .. Hc-mian niaiiieiimiiciun.

14 — 0B2;i
2 1U MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

Note /. The two-sirled surfaces are also called orientable and

the process of choosing a certain side of a two-sided surface is referred
to as the orientation of the surface. The one-sided surfaces are said
to be nonorientable.
The reader should distinguish between the terms “an orientnble
surface” (which means that an orientation can bo introduced on it,
i.e. a certain side of the surface can lie chosen) arid ‘‘an oriented
surface” for which a certain side has already been chosen.
Note 2. In contrast to the so-called local properties, such as smooth­
ness of a surface, which are determined by the conditions that may

I'ig. G.7

or may not hold at separate points of a surface, the orientabilitv

(or nonoricnlability) of a surface is a global property characterizing
the surface as a whole. Indeed, a sufficiently small neighbourhood
of an interior point of the Mobius strip (or any other surface) proves
orienlable. In such a neighbourhood it is always possible to con­
struct a continuous held of normals although there exists no such
a held on the entire Mobius strip.
The concept of a side of a surface is closely related to the notion
of a coherently (eonoordanlly) oriented boundary of the surface
which we shall need later.*
Let 2 be an oriented surface bounded by a single or several con­
tours. We introduce, for each contour L entering into the boundary,
its orientation (coherent with the orientation of the surface 2 )
according to the following rule: a direction in which the contour L
is described is considered to be positive (i,c. coherent with the
orientation of 2 ) if the surface 2 is always kepi on the left of a
* This relationship is dependent on whether the coordinate system taken
in the three-dimensional space is right-handed or left handed. In what follows
w o «d*«H a l w a v v r h n o s e a right-handed coordinate system.
 CH. 5. SURFACE I N T E G R A L S 211

person who is placed on tlie surface so that the normal vector goes
from his feet to his head and who is walking round the contour in
this direction (see Fig. 5.7). The opposite direction is referred to as
the negative one.
If L is an arbitrary closed contour hounding a part of an oriented
surface 2 the positive direction of traversing the contour, coherent
with the orientation of the surface 2 , is again chosen in such a way

that this part of the surface (it is shaded in Fig- 5 .8 ) remains

on the left.* If an oriented plane is taken as 2 the definition of
a contour coherently oriented with a surface reduces to the well
known rule according to which a contour in the plane is regarded as
being positively oriented if it is described in the counterclockwise
direction and negatively oriented if otherwise.
Note 'V. The rule specifying the coherence between the orientations
of a surface 2 and of a contour L entering into the boundary of the
surface can also he Formulated as follows: let u he the unit normal
vector to the chosen side of the surface 2 at a point :\f belonging
to L. and let v he a vector which is perpendicular both to L and
to n and directed toward the surface 2 . Then the positive direction
of traversing the contour L coincides with that of the vector lv, n]**
(see Fig. 5.9).
2. Definition of Surface Integral of the Second Type. Let us first
consider a concrete problem involving the notion of a surface inte­
gral of the second type, namely the problem of computing the flux
of a liquid through a surface.
Let the space (or its part) be filled with a moving liquid (fluid),
the velocity of a particle of the liquid passing through an arbitrary

* If we take a left-handed coordinate system the rule changes to the oppn

site, that is the positive direction of tracing a contour I, lying on a surface X
is such that the pari of tlu* surface 2 hounded by /, remains on the right.
mm 'I'llis rule holds irrespective of whether a right-handed or a left-handed
coordinate system has been chosen in the space. The directions of the vectors
n and v arc independent of the choice of the coordinate system whereas the
vector product (v, u| changes its direction t«i the opposite when a right-handed
system is replaced by a left-handed one or v>r«> verc»».
212 MULTIPLE INTEGRALS, FIELD TH E O R Y AND S E R I E S

point (x, y t z ) being specified by a vector V (x, y, z) with projec­

tions (components) P = P (zt y, z), (J = Q (x, y> z) and R -=
= R (x, y, z) on the coordinate axes x, y and z. Let us find the
amount II of liquid passing in unit time through an oriented sur­
face £ .
Consider an infinitesimal element of area do of the surface E.
The quantity of liquid passing through do in unit time is obviously
equal to r/n = Vn do where Vn is the projection of the velocity V
on the direction of the normal n to da (Fig. 5.10). Expressing
T’n as t lie scalar product of the vector V by the normal vector n to
do we obtain
dll = [P cos (n, x) -f Q cos (n, y) -f- R cos (n, z)] do (5.13)
Formula (5.13) gives an elementary flux of liquid. To obtain

Fig. o.lo

the resultant flux, i.e. the amount of liquid flowing through the
whole surface 2 in unit lime, we must sum up expressions (5.13)
over all the elements da, that is take the integral
11= \ V [/> cos (n, x) -j- Q cos (n, y) ; R cos (n, z)l do
*vJ
According to the definition given in § 1, this is nothing hut the
surface integral of the first type of the function
P cos (n, x) + Q cos (n, y) R cos (nt z)
taken over the surface 2 . But an important thing is that here the
integrand depends not only on the vector function (P , <V, R) defined
over the surface £ hut also on the direction of the normal nt each
point of the surface.
Now we proceed to formulate the general definition. Let £ he
a smooth two-sided surface. Fix a certain side of the surface (that
is choose one of the two possible fields of normals n(.l/)) and con­
sider a vector function A — (/J, <V, /?) defined on X. Let us denote
by A n the projection of the vector A on the direction of the normal n
to the surface at an arbitrary point (x, y, z). The projection can
 CH. 5. SU R FA C E I N T E G R A L S 21

be written in the form

.4n = P cos (n, x) -f Q cos (n, y) + R cos (n, z)
where cos (n, 2:), cos (11, y) and cos (11, 2 ) are the cosines of tin
angles between the direction of the normal and the directions o
the coordinate axes, i.e. the components of the unit normal vector 11
The integral
j j [P cos (n, x) -j- Q cos ( 11, y) R cos (n, c)| da (.5.11

will be called the surface integral of the second type of the veclo
function A — (/J, Q, R) over the surface 2 (or, strictly speaking
over the chosen side of the surface —) and will be denoted as
j J P dy dz — Q dz dx -- R dx dy
v
Thus, by definition, we have the relation
^ I* P dy dz -j- Q dz dx -f- R dx dy
J
vJ
— \ j \P cos (n, x) r Q cos (n. y) -j- R cos (n, 2 )] da (5.1.5

If we pass lo the integration over the other side of the surfac-

the components of the unit normal vector change their ^igns to th
opposite and hence we have the same for integral (5.II). ll shoulr
he noted that for a one-sided surface the notion of a surface integra
of the second type is not introduced.
To achieve the generality guaranteeing the possibility of applying
the notion of a surface integral of the second type to a wide variety
of problems it is advisable to include the integrals over the surface
having self-intersections (an analogous situation was encounterei
in the theory of line integrals).
Note 1. If da is an infinitesimal element of area of a surface 2
the expressions
cos (n, x) dot cos (11, y) do and cos (n, z) da
are, respectively, the projections of the element da on the y. z-
z, x- and 2:, y-planes (sec Fig. 5.11). This is why wo denote then
as dy dz, dz dx and dx dy.
Note 2. We have defined the surface integral of the second typ
on the basis of the notion of the surface integral of the first type
But the surface integral of the second type can also be defined directl;
by means of the corresponding integral sums, which is performei
as follows.
214 MULTIPLE INTEGRALS. FIELD T H E O R Y AN D SEUIES

For brevity, let us consider only one of the projections of a con­

tinuous vector function (P, (V, If), say /?. Take an oriented smooth
surface 2 and form a partition of the surface into parts 2,-. Choosing
an arbitrary point (x*, s f) in each part we compose the integral
sum

2 It (H. yi. Si) St (5.16)

i=l
where is the area of the projection of on the x . //-plane taken
with the .sign — if, at the points belonging to 2 ,-. the normal to
the surface forms an acute angle with the positive direction of the

z-axis and the sign — if otherwise, i.e. if the angle is obtuse at each
point of the element 2 f.* We can easily verify that, for a continuous
function li (x. //, z) and a smooth surface 2 , 1 he limit of integral
sums (fi.1 0 ) (as the partitions of the surface ore infinitely refined)
exists and is equal to the integral
j j /? (x, y , z) dxdy
v
(compare this with the definition of the line integral of the second
typo given in § 2, Sec. 2 of ('diopter 4).

* A partition ,»f the surface 2 may include “irregular' elements, i.e. such
that the angle (n. ;) is acute at some of its points and obtuse at the other (see
Kig. 5.12). Wo can either exclude the partitions containing such elements or
attribute an arbitrary sign to the areas of their projections because this docs
not affect the result since the sum of the areas of the projections of these ele-
r u -“. ' r :• n c og iig
o i!.•«\ 9!m !!.
 CII. 5. SURFACE IN T E G R A L S 215

We can similarly define, by means of the corresponding integral

sums, the integrals
P (*, l/» z) dy dz and j j Q (x , j/, z) c/z dx

and, consequently, write the general integral

I* dy dz -p Q dzdx -' - Ti dx dy
z
ns the sum of the integrals of these three forms.
A»/(’ 3. The distinction between the surface integrals of the first
and the second types is that in the hitter each area element do is in
fact regarded not as a scalar quantity but as a vector do directed
along the normal n to the surface Z and having the components
do cos (n, x), do cos (n, y) and do cos (n, z)
Accordingly, the surface integral of the second type of a vector
function A = (/'. (J. li) is often written in the form
f j (A, do) (.5.17)

which is equivalent to
j*j (A, n) do (5.18)
£
Ae/c 4. B esides integrals (5.18) we encounter, in some problems,
integrals of the form
I \ (A, n) do (5.19)
v
The value of such an integral is not a scalar hut a vector. The com-
potation of integral (5.19) obviously reduces to the separate inte­
grations of the components (projections) of the vector [A. n 1. Ifere
llie integrand also depends, as in integral (5.18), on the normal ii
to the surface Z. and therefore integral (5.19) should be naturally
regarded as a surface integral of the second type (but as a “vector-
valued” one, in contrast to “scalar" integral (5.18)).
3. deducing Surface Integral of the Second Type to l>oublc
Integral. The definition of a surface integral of the second type
and Theorem 5.1 immediately imply the following result.
Let Z be a smooth (or piecewise smooth) surface determined by
an equation
< * . .7)
2Mi MULTIPLE INTEGRALS, FIELD TH EO K Y AND SERIES

and Jet li (x, y t z) be a bounded function defined on 2 . Then, lor

Ilie surface integral of the second type J j R (x, y, z) dx dy taken
over the upper side of the surface 2 , we have the relation
[ j R (*> U> ~) dx dy -= j j R (x , y, z (x, y)) dxdy (5.20)
‘s D
(where D is the projection of the surface 2 on the x, //-plane) pro­
vided that the integrals entering into (5.2U) exist. The surface inte­
gral on the left-hand side exists if the double integral on the right-
hand side of (5 . 2 0 ) exists.
Actually, the surface integral can be written in the form
j j H (x, i/, z) cos (n, z) do

Applying formula (5.9) to this expression we obtain the relation

we set out to prove.
Thus, in order to reduce a surface integral J J R (x, //, z) dx dy.
taken over the upper side of a surface 2 determined by an equation
z = z (x, //), to a double integral we must substitute the corres­
ponding function z — z (x, y) for z into the integrand and replace
ihe integration over the surface 2 by the integration over the p r o ­
ject ion D of 2 on the x, //-plane.
If the integral is taken over the lower side of the surface 2 we
obviously have

j j R (r, //, z) dx dy = — j ^ R (x, y } z (x, y)) dxdy

x n
We similarly derive the formulas
j j P ( x , y f z)d//dz = ± j j P (x(y, z), y „ z) dy dz (5.21)
E Di
and
I y, z ) d z d z — 1i ^ Q {x, y {z* x)i z)dzdx (5.22)
‘i *D2
where in the former 2 is understood as a surface represented by an
equation x = x (//. z) and in the latter as a surface determined by
an equation // = // (z, r). Accordingly, the plus sign is taken if
the normal to the surface x = x (//, z) (y — y (z, x)) forms an
acute angle with the positive direction of the x-axis (//-axis) and
the minus sign if the angle is old.use. The symbols />, and O- desig-
 CH. 5. SURFACE IN T E G R A L S 217

uatc, respectively, the projections of the .surface 2 on the if. z*

and z. x-axes.
Formula (5.20) can also he applied to reducing a surface integral
to a double one when an oriented surface 2 is composed of several
pieces each of which is determined by an equation of the form z =
= z (x. y). In this case the integral should he written as the sum
of the integrals corresponding to the pieces and then formula (5.20)
should be separately applied to each integral.
Exercise. Rewrite the integral
J — j \ R (x, y, z) dx dy
v
taken over the outer side of the sphere
x2 + IE ~ z2 = a2
in the form of a sum of double integrals.
A naiver.
/= ( l' II (a:, y, \ f a* — x2— y2) d x d y —
x2 +y2 ,;0a
— ff R (x, y, — Y d 1—xl — yL) dx dy
X2-}-y2. al
Here the iirsl summand is equal to the integral taken over the upper
side of the upper hemisphere and the second, with the minus sign
prefixed to it, is equal to the integral taken over the lower side
of the lower hemisphere because the two hemispheres, oriented in
this way, constitute the outer side of the entile sphere.
We have shown the way of reducing a surface integral of the
second type, taken over a surface determined by an equation in
(’artesian coordinates, to a double integral. For a surface represented
parametrically, the application of Theorem (5.1') immediately
implies the following result.
If a smooth (or piecewise smooth) surface 2 is represented by
a parametric equation
r = r (w, v)
and (/>, (?, R) is a bounded vector function defined on 2 we have
the re la ti oil
\ \ P dydz 4 - Q dz d;r 4 - R dx dy =
**
V

= \ \ [P cos(n, x)4-<?cos(», y)4-/*cos(n, z)J Vgttgzz — g^dudv

u
(5.221
218 Ml I.T1LM.K 1 NT lit* HALS, 1'IEI.D T l l K O R Y AND S E R IE S

where D is the range of Iho parameters it. v and gn, # 12 * fizz are
liu* fundamental quantities of the lirsi order of the .surface pro­
vided that the integrals entering into formula (5.23) exist. Thu
surface integral on the left-hand side of the formula exists if the
double integral on the right-hand side does.
Expression (5.23) can be transformed to another form. As is
known (see § 3. Sec. 5 in Chapter 3),
_____ A H_____
cos (si, x) --
Y A s + T f t + C* ’ cos(n ' y ) ~ v T > m --~c* ’
c
cos (n, 2) — (5.24)
V d ‘2 d -0 “ j-C*
where
Qy
du
az
du
0z
du
dx
d\i
i
<
'li.
du
dy
du
A =
Off dz
, B— dz dx
, c = Oj dy
(fV ov ~0v i)v Ur Ov
and
V gtig22 — gU = V A* +
W Q U O + - - O 12 * » — I

Therefore formula (5.23) can be put down as

Jj I'dydz Q d z d x -i ft dx dy - I' C(/M ' Qli \-HC\dudv (5.25)
L)
where
ft _ f> (x (m. i>). // (//, y), z («, t>)).
Q = (J (x (u, v)t y (u, i;), z (m, a)) and
/{ - H (x (r<, v ) . y (u, v ) , z ( u , i?))
l’h|uaIiI ies (5 .2 0 )-(5 . 2 2 ) are obviously special cases of general
formula (5.23).
§ 3. OSTROCIIADSKY TIIEOHHM
1. Derivation of Oslrogradsky Theorem. In the foregoing chapter
we established a formula connecting a double integral over a plant1
domain with a line integral taken along its boundary (dreen’s
rormuia). Mere we arc going to deduce a similar formula expressing
a relationship between a triple integral over a space figure and
a surface integral taken over the outer side of the surface hounding
Ihe ligurc. The result we arc speaking of is called the Oslrograthky
theorem {formula) *
* Ustrogradsky, Mikhail Y'asilycvich (1801 1802), a prominent Hessian
mathematician. This formula (also called the divergence theorem) was published
h i his article On Heal Theory in 1828. It is sometimes called the Gauss theorem
although Gauss obtained it considerably later (in 18*41).
 CH 5. SURFACE I N T E G R A L S 210

For convenience, we introduce the following terminology. A

spatial domain V bounded by two piecewise smooth surfaces
and Z 2 determined by equations
z — zx (x, y) and z = z2 (x, y) (z2 (*♦ y) ^ 2 * (x, y)) (5.20)
and by a lateral cylindrical surface S 3 with elements (generators)
parallel to the z-axis will be referred to as a domain regular in the
z-direction. The surfaces z — zx (x. y) and z — z2 (x, y) will be,
z * z z (x ,y )
z

respectively, calk'd the lower and the upper (curvilinear) bases of

the domain* (see Fig. 5.13). Similarly, a domain bounded by two
piecewise smooth surfaces
x = x, (//. z) and x — x 2 (y, z) (x 2 (y , z) ^ x x (//, z))
and l>y a cylindrical surface wilii generators parallel to tlie x-axis
will bo called a domain regular in the x-direction. A domain regular
in the y-direction is defined analogously.
Finally, a domain V will be called simple if it is possible to divide
it into a finite number of domains regular in the z-direction and
also into a finite number of domains regular in the other two direc­
tions.
Let now V be a domain, regular in the z-direction. with bases
and represented by equations (5.2b) ami a lateral cylindrical
surface —3 . The union of the three surfaces 2 ,. d,ld ^ -3 forms the
whole boundary of the domain V. We denote the boundary by 2
and consider its outer side. Take a function It (x, y, z) defined

* This definition also includes the cases where there can be no lateral sur­
face For instance, a three-dimensional sphere (hall) is considered to bo a
domain, regular in the z-direction, whoso bases arc its lower hemisphere - j
and upper hemisphere and whose lateral surface £ 3 is degenerated into
220 MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

in ihe domain V (including its boundary) and having Ilie conti-

nuous partial derivative — in the closure of V. We obviously have*
the equality
Z2(-v. V)
C OH
J a r dZT=Ii z 2(x*y))—R y •2i (x>y))
zi(.v. y)
Let us integrate this equality over tlie projection D of the domain
Y on the x, //-plane and replace the threefold iterated integral thus
obtained by Ihe corresponding triple integral:

\ \ \ i k d x d 'j d z = : \ \ R
V* *D
l J t (»
Zi x d x d y~~

— M /? (x, y , zi (x, y)) dxdi/ (5.27)

D
The first integral on the right-hand side of (5.27) can be written
(see formula (5.20)) in the form of the surface integral of the function
R (x, y, z) taken over the upper side of the surface
z z o (x, y)
Similarly, the second integral (x, //, 3j (x, y)) dx dy can
*D
he regarded as the surface inLegral of the function li (x, y y z) taken
over the upper side of the surface z ~ z t (x. y) or as the integral
over the lower side of the same surface z --- z { (x, y) with the minus
sign attached to it. Hence we obtain
\\[ — = j ] Hdxdy-Y^ \ l l d x d y (5 2 8 )
J VJ zi Si
where (lie first integral 011 the right-hand side is taken over the
upper side of the surface £ 2 and the second over the lower side of
the surface 2 ,. Adding the integral
j ^ dx dy
It
(which is obviously equal to zero) taken over the outer side of the
lateral surface 2.*$ to the right-hand side of formula (5.28) we obtain,
on the right-hand side, a surface integral over the outer side
of the entire surface 2 bounding the domain V'. 'Jims, we arrive
at the following relation:
dx dy dz = Jj R dx dy = ^ J R cos (11, z) do (5.29)
1/
 CII. 5. S U R FA C E IN T E G R A L S 221

Kelatiou (5.29) is also valid for any domain V which can bo broken
up into a finite number of domains regular in the 2 -direction. Indeed,
let us divide such a domain V into parts Vi regular in the z-direction
and write the relation of form (5.29) for each part. Next we sum
up all the relations. Then wc obtain, on the left-hand side, the triple
integral of dOHz taken over the whole domain V and, on the right-
hand side, the sum of the surface integrals over all the parts of the
surface ^ bounding the domain V and over the surfaces breaking up
the domain V into the parts Vr*. each of the latter integrals being
taken twice, that is over one side of the corresponding surface and
over its other side. Therefore, after the summation has been per­
formed all the integrals, taken over the surfaces by which the domain
I' is divided into the parts Vi* mutually cancel out and consequently
we derive the formula
^ ^ dx dy dz — j j R dx dy (5.30)
‘ v' 2

Let now V be a domain regular in the x-direclion, i.e. one hounded

by piecewise smooth surfaces (bases)
x = xt (y, z) and x = x2 (//, z)
ami hv a lateral cylindrical surface with elements parallel to tin*
j-axi>. Consider a function P (x , y , z) which is defined and con-
OP
tinuous together with its partial derivative in the domain V
(including its boundary). Applying arguments similar to those
presented above we obtain the equality

\ \ \ ^ r d x d y d z ~ \ \ P d y ds (5.3i)
*y *2
which also remains valid when the domain V consists of a liuile
number of domains regular in the x-direction.
We similarly obtain the relation
W ^^rdxdydz^^Q dzdx (5.32)
*V 2
for an arbitrary domain V which can he divided into a finite number
of parts regular in the ^-direction where Q (x, y , z) is a function
defined and continuous together with its partial derivative ^ in
oit
the closure of V.
Finally, let us take a simple domain V and consider three, func­
tions P, Q and R which are continuous in this domain (including
its boundary) and possess the continuous partial derivatives -J-.
 M U LT IP LE INTEGRALS, FIELD THEORY AND S E R IE S

~ and ^ in llie closure of V. Then all three relations (5.30). (5.31)

and (5.32) are fulfilled. Adding them together we obtain the equality
Ijj dxdiJ d z = \ \ P d y d z + Qdzdx { R dxdy
'v Z
(5.33)
or

= jj' ^ [P cos (n, x) Q cos (n, y) r R cos (n, z)\da (5.33')

Formula (5.33) (or (5.33')) expresses the above mentioned Ostro­

gradsky theorem.
Note. In deriving the Ostrogradsky formula we have supposed
OP
that the functions P , Q, R and their partial derivatives — ,
^ ^ an* continuous (and, consequently, bounded) in a closed
simple domain. But, applying arguments similar to those prr
sen ted in connection with Green’s formula (see Note 1 in § 3 of
Chapter /i) we can prove the validity of the Ostrogradsky formula
under more general conditions. Namely, the Ostrogradsky theorem
remains true if
1. V is a hounded domain whose boundary consists of a finite
number of piecewise smooth surfaces.
2. The functions P (x, ?/, 2 ), O (j:. y^ z) and R {x. y. c) are
continuous and. consequently, hounded in the closure of the domain
r.
3. The derivatives — , — , 0z exist and are continuous in
dx dy
the interior of the domain V (hut do not necessarily satisfy this
condition 011 Us boundary) and the integral

+ '- £ ) d * d y d z
V

(which is understood as an improper triple integral* in cast* condi­

tion 3 is violated on the boundary) exists.
2. Application of Ostrogradsky Theorem to Evaluating Surface
Integrals. Expressing Volume of a Space Figure in the Form of a

* Improper multiple integrals will be treated in Chapter 1).

 CII- V SUR FACE I N T E G R A L S 223

Surface Integral. Formula (5.10) established in § 1 makes it pos­

sible to reduce a surface integral of the second type to the corres­
ponding double integral. But there are cases when this way of com­
puting a surface integral proves to he practically inconvenient. In
particular, it is sometimes advisable to reduce a surface integral
over a closed surface to a triple integral by applying the Ostrogradsky
theorem.
Examples
1 . Evaluate the integral

J ■—j J x’Jdy dz -[- z/ydz dx —r 5 dx dy

z
over the sphere x1 \ yz r z2 — a1.
Solution. Taking advantage of the Ostrogradsky theorem \vcobtain
./= 3 (x2 } - y2-\-z2) dx dy dz

IVow introducing the spherical coordinates we receive

2n n n
* 4
J 3 ^ dip ^ dO r4 sin 0 r f r - Ta no 0
o «
2. Evaluate the integral
J - \ \ z dy dz !- x dz dx : y dx dy
•IvJ
taken over a closed surface 2 .
Solution. By the. Ostrogradsky theorem, the integral reduces
to the triple integral (over the domain bounded by the surface X)
whose integrand is identically equal to zero. Hence, we have ./ — 0
for any closed surface X.
In the preceding chapter we showed that Green’s formula made
it possible to express the area of a plane figure as a line integral
along its boundary (see formula (4.47)). Similarly, the Ostrogradsky
theorem yields an expression of the volume of a space figure V in
the form of a surface integral over the closed surface £ bounding
the figure. In fact, let 11s choose three functions Q and li so that.
OP ..
dx 1 dy 1 dz

Then we ohtam the relation

^j" P dy dz T- () dz dx -1- II dx dy = ^ ^ dx dy dz - V
*s v'
•>-K MULTIPLE INTEGMALS, 1’I E L D THEORY AND S E R I E S

where V is the volume of the domain hounded by 2 . The integral

is taken here over the outer side of the surface 2 . In particular,
putting
P —^ x t (^ = - 1/ and R —— z
wi: arrive at a convenient computational formula
I —-j x ihj dz -}- y dz dx -|- z dx dy (5.34)
v

§ 4. STORKS* THEOREM
1. Derivation of Stokes’ Formula. Stokes* formula expresses
a relationship between surface integrals and line integrals. It gene­
ralizes Green's theorem, the latter being a special case of the former
when the surface in question is a part of the x, //-plane. Like Green’s
formula and Oslrogradsky’s formula, Stokes’ formula is widely
applied in mathematical analysis and its applications.
Suppose we are given a smooth oriented surface 2 hounded hy
a coherently oriented contour A (see $ 2, Sec. 1). Get a vector func­
tion (/>. (J. R) be defined in a three-dimensional domain in which
the surface 2 is strictly contained and let the functions /•*. <J. It
and their lirst-order partial derivatives he continuous in the domain.
We shall transform the line integral
| I1dx + Q dy + R dz (o.5o)

taken along the contour A into a surface integral over the surface 2 .
We lirst consider the case when the surface 2 is represented in
(’artesian coordinates by an equation
z = z (x, y)
Denote by D the projection of the surface 2 on the x, p-plane, and
let L he the boundary of D, that is the projection of the contour A
(see Fig. 5.14). The transformation of line integral (5.35) into a sur­
face integral will he performed according to the scheme

a
-H I-
l n v

which means that we shall lirst transform the line integral over the
space curve A into a line integral along the plane contour /,. then

* Stokes, George Gabriel (ISltt-tllOli), an English mathematician and phy­

sicist.
 CH. 5. SURFACE I N T E G R A L S :>

reduce it (l>v means of Green s formula) to a double integral over

the domain D and. finally, transform the latter to a surface integral
over Z .
Next we proceed to calculate. To begin with, let us take an inte­
gral of the form
J x= j P dx
A
Observe that we have
J\ = \ P (x, ?/, z) dx - \ P (x , //, 3 (.r, y)) dx
A I
because the contour A lies on the surface Z determined by Ilie equ­
ation z -- z ( t . ij) . Now, applying Green s formula wo obtain
iP_ OP Oz
J\ ~ \ P Ih -(■*> y)) dx - — ( j ( -O'J Oz Oy j dxdy (5.3(5)
t 'D
where P is a composite function of x and y ami therefore its deri­
vative with respect to // has been computed in accordance with
the well known rule for differentiating a composite function.

Figf. .".14
Using expressions (3.3b) for the direction cosines of the normal
to a surface (represented by an equation z - z (.r, y)) we iiiul that
Oz cos(n, //)
~0y~ cos (n, z)
Thus,
OP cos(n, y)
oz cos(n, 2 )
Now. talcing advantage of formula (5.20) we can transform the
double integral into the corresponding surface integral. This yields
I f f / tfl* OP cos (n, //) \ . J
I * — 3r-35<nAF) <” *<“• da =

= — I j ( ~ c ° * ( n , z) — — (/)) da
2J0 MU I.Til'Ll-; I NTKtill ALS, FIELD THEORY AND SERIES

Henco7 we have

\ ? dx \ ] ( l 7 C0s(a * U) — ^ 7 c o s(n’ -)) dG (5*37)

A ‘z
Wo have supposed here that the surface Z is represented by an
equation of the form z = z (x, //). But the same result can be obtained
for a surface Z determined by an equation ij — y (z, x). To this
end we must consider the projection of Z on the z, arplane (instead
of x. //-plane) and apply arguments similar to the above. Further­
more, if Z is a part of a plane perpendicular to the jr-axis e(|uality
(5-37) remains valid (although in this case Z cannot bo projected
either on the x, //-plane or on the z. rr-plane so that the correspon­
dence between the points (z. y. z) £ Z and those of the coordinate
planes is one-to-one) since its left-hand and right-hand sides
are obviously equal to zero (check it upl). Finally, arguments ana­
logous to those applied to deducing Green’s formula and Ostrograd-
sky’s formula show that if the surface Z is composed of a finite
number of parts such that for each of them equality (5.37) holds
it also holds for the entire surface Z. Thus, relation (5.37) has been
established for any surface consisting of a finite number of surfaces
of the above types. Next we similarly derive the following two
equalities analogous to (5.37):
\ dy \\ (7 7 cos (n, z) — cos (n, a-)) do (5.38)
A *Z
a lid
\ H dz-^ ![ J (-— cos (n. *) — ~ cos (n, i/) ) do (5.31))
A
Adding together equalities (5.37), (p.38) and (5.39) we obtain

\ r <h <-><>!' f (-J7 ) eo» <"• =) J-

A *2

) ros x) ( 4 r - i l r ) cos<"■ 'J] j dG <5 /,0 >

which is Stokes’ formula that we set out to prove. It can he rewrit­
ten in the following form;

(o /il)

T o r e m e m b e r S l i d e s ' f o r m u l a o n e s h o u l d n o t i c e t h a t t h e first
s u m m a n d u n d e r t h e i n t e g r a l s i gn on t h e r i g h t - h a n d s i d e c o i n c i d e s
 CH. ft. SUltPACE I N T E G R A L S 227

with the expression under the integral sign in Grepn’s formula and
the second and the third summands can he obtained from the first
by means of circular permutation of the coordinates x, y , z and
the functions P. Q. R.
If the surface £ is a figure lying in the x, (/-plane the integrals
involving dz dx and dy dz vanish and thus Stokes’ formula turns
into Green s formula.
J\'ote 1. In deriving Stokes’ formula we have used a Cartesian
coordinate system. But neither the line integral nor the surface
integral entering into the formula depends on the way the surface 2

Fig. 5 .tfi

and its boundary A are represented. Therefore Stokes’ theorem

remains true for any oilier way of representing the surface, including
its parametric representation
r = r (m, v )

Note 2. Stokes’ theorem also applies when the boundary A of the

.surface Z is funned of several separate contours. In this case the
integral j P dx -- n dy -- R dz should he understood as the sum
A
of Ihe integrals taken over the contours coherently oriented with
the surface 2. For example, if 2 is the lateral surface of a cylinder
with an opening (see Fig- 5.15) and if wc consider the oulor side
of the surface, Stokes’ theorem expresses the relationship between
1 he integral over 2 ! and the line integral taken along the three con­
tours. forming the boundary of 2 !, in the directions indicated by
the arrows in Fig. 5.1.).
2. Application of Stokes’ Theorem to Investigating lane Integrals
in Space. Stokes’ theorem has various applications soiiim #»f ul»o*V»
15+
228 MU LT IP LE I X TE GU AL S , FIELD THEORY AN D SERIES

will !>«• considered iu the next chapter. Hero we are only going- lo
take advantage of tlie theorem in order to establish the conditions
for a line integral of the second type in space being independent
oT the path of integration. These conditions generalize the results
(obtained by means of (Ireen’s formula in § 4 of Chapter i) con­
cerning the question of path-independence of an integral over a
plane curve.
Let us introduce the following
Def i n i t i o n . .1 three-dimensional domain V is said to be s i m / R i /
con n eeded if. for any closed contour belonging to V. there exists
a surface, with the contour as its boundary, entirely lying in V.
Cvainplcs of simply connected domains are a sphere (ball), the
whole space, the domain lying between two concentric spheres etc.
As an example of a domain which is not simply connected (such
domains are referred to as multiply connected) wc can take a ball
with a cylindrical tunnel passing through it (see Fig. 5. lb).
Next we proceed lo establish the following result analogous lo
Theorem 4.3.
T h e o r e m o.'d. Iff* (:r, y , s). Q (x, y , z) and R (x. y , z) are con •
tinuous functions. defined iu a bounded closed simply connected domain
I 7, which possess the continuous first-order partial derivatives in. the
domain the following four assertions are equivalent to each oilier.
1. The integral ^ /' dx j (> dy \- R dz taken over any closed
contour lying inside V is equal to zero.
2. The integral j P dx -j- (J dy T R dz is independent of the
A 13
path of integration connecting two arbitrary fixed points A and B
3. The expression P dx Q dy \ R dz is the total differential of
n single valued function defined in V.
4. The conditions
QQ _ dV_ 0H_ _ dQ_ dP_ __ on_ r /9 .
dx dy ’ dy dz ’ dz dx >*"**"

are fulfilled at each point (x, y t z) of the domain V.

Proof. The theorem is proved according to the scheme 1 2
►3 — 4 1 which we followed when proving Theorem 4.5. We
leave tin* proof to the reader with the only hint. that, to deduce con­
dition 1 from condition 'i one must lake an arbitrary closed contour
A lying within V and consider a surface X entirely lying in V whose
boundary is A. suc h a surface? existing because of the condition that
V is a simply connected domain. Then the application of Stokes’
theorem to the line integral taken over A shows Hint condition (f»./i2)
 CH. 5. SlJltFACE INTKOHAI.S ■>*->!

implies the relation

j P dx V Qdy j R dz —0
A

If the expression P dx -j* (J dy -}- R dz is the total differential

of a function O' (x, i/t z) we can easily derive the formula
uc, v. *)
U (x, y, z) = j P dx - Q dy -- It dz ~C (o. 4:5
<*o. Un. *o)
analogous to formula ( 4 ..'*()) established in § 4 of (Chapter 4 for tin
(*. V. 2)
case of two independent variables. The symbol j designates
<*0. Vo* *o>
here the integral along an arbitrary path entirely lying in the domaii
I'and connecting an arbitrary fixed point (x0, y 0. z0) with a variahh
point (x. //, z)r C being an arbitrary constant.
If the functions f \ O and It satisfy conditions (o.42) but tin
domain they are defined in is not simply connected the properties
of the integral
\ Pdx ' Q d y Itdz
Aft
resemble those of the line integral | /* dx (J dy in a plane mul
.\n
liply connected domain. In particular, in the case of a multiply
connected domain, expression (.3.43) is also a function whose total
dilfcrcnliai coincides willi P dx (J dy It dz hut in the genera
case the function may be multiple-valued.
6

The concept of field forms llie basis for various notions of modern
physics- In this chapter we shall present the elements of mathe­
matical theory applied to investigating physical fields.
In physical problems we usually deal with quantities of two
basic typos, namely scalars and vectors.* Accordingly, we shall
consider two types of held, i.e. scalar fields and vector fields.

§ i. SCALAR FIELD
1. Definition and Examples of Scalar Field. Let £2 he a domain
in space. IT there is a correspondence which attributes a number
U {AI) to each point .1/ of the domain we say that there is a scalar
held defined in the domain 12.
Examples of scalar fields are a temperature held inside a body
subjected to heating (in this case at each point. M of the body the
corresponding temperature U (M) is specified), field of illumination
produced by a light source etc.
The density held of a mass d istrib u tio n we have already dealt
with is an important example of a scalar field. Let us come back
to this notion. Suppose? that a spatial domain 12 carries a conti­
nuously distributed mass. Associating with every subdomain V'
belonging to 12 the mass contained in V. we arrive at. an additive
set function p (L). If. at each p oin t AI £ 12, the set function possesses
the- derivative dv
—? with respect to volume the function 'n (A/) = tl\ 4-^
is called the density of mass and the values of the derivative form
a scalar held referred to as the density field of mass distribution.
Similarly, a continuous dislrihut ion of an electric charge yields
a scalar field of charge density. There are many other examples of
this kind.
* Bv the way. only in studying some simpler problems of physics can we
limit ourselves to scalar and vector quantities. In many branches of modern
theoretical physics, such as electrodynamics, the theory of relativity, the theory
of elementary particles etc., an essential role is played by the quantities of
a more complicated nature. In our course we shall deal with one type of such
quantities, namely with the so-called tensors which will be studied in the next
c hauler.
 CH. fi. FIELD THEORY 2:51

Besides the fields defined in spa Iini domains, wo often encounter

plane scalar fields. An example of such a field is the illumination of
a part of a plane produced by a light source.
2. Level Surfaces anti Level Lines. Let U(M) be a scalar field.
If we introduce a Cartesian coordinate system x. tj. z in tlie domain
of deli nit ion of the field this field can be represented by a scalar
function U {x. y. z) of the coordinates of tin* moving point .1/.*
In what follows, a Function I' (x, y. z) of thislype will be considered
to In* continuous and to have continuous partial derivatives of
the lirst order with respect to the variables x. y and 3 .
The specification of a scalar field by means of a fixed coordinate
system and the c o r r e s p o n d i n g function U (.r. //. z) is sometimes
insufficient for visualizing the structure of the field. To get a more
complete description it is convenient to use tin* so-called level
surfaces. A level surface of a scalar field U (M) is a locus of the points
at which the field U{M) assumes a given fixed value C. The equation
of a level surface is of the form
U (x, y, z) -= C** (0.1)
It is clear that the level surfaces (corresponding to all the possible
values of C) fill the entire domain in which the field is defined ami
11ml two surfaces
U (.r, y, z) = Cx and IJ (.r. y, z) — C2
have no points in common for ^ C«. The specification of all the
level surfaces and the corresponding values of C marked on them
is equivalent to the specification of the field U (.1/). The
d isp osition of l In* level surface® in space enables us I" visualize
the structure ol‘ the field.
This method of representing a field is especially convenient when
the field in question is defined in a plane region. In this case the
field is represented by a function U (x, y) of two variables. Gene­
rally speaking, an equality of tin* form (I (x. y) - C determine?

* The character of the function V (.r. y. is of course dependent ii"t

only on the field in question hut also on tile choice of the coordinate system.
If the coordinate system is regarded as being fixed the notion of a spatial scaler
field coincides with that of a function of three variables. Hut nevertheless, to
stress that, our discussion concerns, as a rule, the quant ities whiGi have a certain
physical significance independent of Hie choice of the coordinate system we
shall always use the term “field*'.
** I:ruler the a hove assumptions concerning the functions of typo f (;V)
equation (C*. I) in fact determines a smooth surface provided that, for the given
C. there exist points satisfying the equation and llio derivatives - - . - and
or </ tj

~ do not simultaneously vanish at. these points (e.g. see | 8 |. Chanter 1 ♦. 4t h.

2 3 2 MITI.TII'J.IS INTEGHAL5. 1'IKLO TIIKOMV AND SEKIES

a curve. Such curves arc called level lines of the plane scalar field
U (.1/). Level lines are widely applied in cartography for represen­
ting the relief of a terrain. For this purpose, to indicate altitude on
a topographic map. the contour lines (the horizontals) connecting
the points of the same elevation are drawn (see Fig. 0.1). This method
is also used for representing, on special maps, the distribution of

temperature, pressure, amount of precipitation and the like. The

corresponding level lines are then referred to as istf/erms in the case
of temperaluri*. isobars in the case of pressure etc.
3. Various Types of Symmetry of Field. In many physical pro­
blems we deal with fields possessing various types of symmetry.
The symmetry properties usually simplify the investigation of such
fields. I.et us indicate1 some* important special cases.
(a) r iro dimensional field. If there is a Cartesian coordinate1system
in which a scalar held U (A/) can he represented by a function
dependent not on three but on two coordinates (e.g. by a function
U (x. //)) the held is said to he two-dimensional (plane-parallel).
Iii other vvord>. a >caiar hold I {.\J) is called plane-parallel if there
is a direction in space such that the field goes into itself when being
translated along this direction. Tin* level surfaces of sue!1 .» held
form a family of cylindrical surfaces (Fig. (>.2). In an appropriately
chosen coordinate system the family is represented by an equation
of the* form I ’ (x. */) ~ C.
(b) Axially symmetric, field. If, Tor a given held U (A/)t there exists
a cylindrical coordinate system in which the held is represented hy
a function depending solely on the variables r — |/"j:2 y 1 ami z
(hut not on the angle q) the held is said to he axially symmetric.
This moans that a held U (A/) is axially symmetric if and only if
if goes into itself when being rotated (through an arbitrary angle)
about a lived straight line which is the axis of symmetry of the
held. The level surfaces of such a held are obviously surfaces of
revolution (Fig. l/.T). In case these surfaces of revolution are circular
cylinders (see Fig. t>.4), that is if fin* held U (A/) is represented, in
a specifically chosen coordinate system, by a function dependent
on only one coordinate r (which is the distance from the point. .1/
to the axis of symmetry of the held), I' (A/) is called a cylindrical
held.
 CII. ti. FIELD THEORY 23c

(c) Spherical field. If tho values of U (M) depend only on the

distance of the point M from a fixed point M 0 the field is said to he
spherical. The level surfaces of this field constitute a family of con­
centric spheres (Fig. (5.o).

A. Directional Derivative. Tho application of mathematical ana­

lysis to investigating a scalar held U {M) makes it possible to des­
cribe its local properties, i.e. the variation of V (M) in passing
from a given point M to the points A/' lying close to .1/.
To this end we shall use the notion of tho directional derivative
of a field. T.et £ (M) bo n scalar field. Consider two points M and M'
placed close to each other and form tho ratio
... l)x
n
23'i MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

where h is the length of the line segment M M '. Let the point M'
approach M so that the direction of the line segment M M f all the
time coincides with the direction of a fixed unit vector X. If. in this
process, ratio (6.2) tends to a finite limit we call this limit the

Fi-. r,
derivative of the scalar field U (•!/) at the point M along the direction
of the vector 7. (the directional derivative) and designate it as
tiU(M)
ok
OU
The derivative -jy characterizes the rate of change of the quantity
U(M) in the direction a.
To compute ^ we choose a coordinate system and represent
U(M) ns a function IJ (x. y. z).
Suppose the direclion a. (i.e. the vector 7J) forms angles
a, p ami y with the coordinate axes. Then we have
.1/1/' — h (i cos 2 : 1 j cos p | k cos y)
and
C (Af) — V (r -\ h cos a. t/ h cos {3, z -f- h cos y) (0.3)
Hence the derivative ^ co in cid es with the derivative of composite
function (6.3) with respect to h for h 0. Differentiating we thus
obtain
0U (M) <1U{M‘) dU . 01> 0 .00 lC
ok — — cos a -f -r - cos p |
- ---- • cos y (6.4)
oh ox Jy <iz 1
5. Gradient of Scalar Field. Lxpression (6.4) can lie regarded
as a scalar product of two vectors, namely of the unit vector
-- (cos ci, cos p. cos y)
determining° the direction in which the derivative o:. is taken and
the vector having t he components (projections on their-, y- and s-a\es)
ttf" ijf> , Of>
----. —— and
 OH. 6. FIELD THEORY 235

The latter vector is known as the gradient of the scalar held U and
is denoted bv the symbol
grad U
Tims, we can write
OU OU OU \
Ox * Oy 1 Oz )
(6.5)
and consequently
dU (6.6)
Ok
(gard U , >.)
Fig. 6.6 visually interprets the expression of the directional deri­
vative in the form of the projection of grad V on the vector
Formula (6.6) can also ho written as
= | gard U | cos (p
■where cp is the angle between grad V and the unit vector )». It follows
that at every point where grad U 0 there is a single direction

Fig. 0 .t>

OU
for which -r-
Ok
assumes its maximum value, i.e. a single
9
direction in
winch the function U increases with the maximum rate This direction
coincides with the direction of the vector grad U . Indeed, for this
direction cp = 0 and consequently
= | grad V |
whereas for all the other directions we have
OU
—= | grad V | cos cp<C | grad U \
Ok

Thus we see that the direction of the vector grad U is the one
in which the quantity V increases with the maximum rate and that
tin* magnitude (length) of the vector grad V equals the rate of in­
crease of 1lie quantity U> in this direction.
L*nt , obviously, neither the* direction of maximum rate of increase
of a function nor the value of iLs directional derivative in this direc­
tion depends on the choice of the coordinate system. Hence we come
to the conclusion that the gradient of a scalar field is specified by
i's'df ;*e d n-{ by flic rhcdc; ..file cimu di v-iem (<j huongli
2HG MULTIPLE INTEGRALS. FIELD TUEORV ANL> SE R I E S

this assertion is not directly implied by equality (!>•♦>) which we

consider to be a definition of the gradient).
As is known, the derivatives Ox
— ’, Otf
and Oz
at a point
1
.1/ are
the components (projections) of a vector normal to the surface
U (x. if. 2 ) ~ const passing through the point.* Therefore, the
gradient of a field U is directed along the normal to the level surface
at each point of the domain whole the field is defined.
livery curve for which the tangent- line at each point .1/ goes
along the vector grad U at the point Af will be referred lo as a
gradient lino of the field U.** Wo can now say that the gradient
lines of a field (■ are the curves along which the rate of change of
the held is maximal.
It can he proved that if I 7 (x. if, 2 ) is a function possessing
continuous partial derivatives up lo the second order inclusive for
every point .1/ of the domain of definition of the held V there exists
a single gradient lino passing through the point M . A gradient line
is orthogonal, at its every point, to the level surface passing through
tin* point.
§ 2. YlXfOti MKI.IJ
I. Definition and Examples of Vector Field. We say that there
is a vector held defined in a domain Q if a certain vector A (.1/)
is associated with each point .1/ of the domain.
An important, example of a vector field which will be many times
discussed in what follows is tin* field of velocities of a stationary
flow of a liquid. Lei a domain Q fie occupied by a liquid (fluid)
flowing, at each point., with a velocity v independent of time (but.
val ving, in the general case, from point to point). If we associate,
with each point .1/ of the domain 12, the vector v = v (:l/) we thus
arrive at a vector hold which is a field of velocities.
The field of gravitation of a mass distribution is another impor­
tant example of a vector field. Suppose a mass is distributee! in
space. Then a material point with unit mass placed at a point M
is subjected to a gravitational force. These forces (defined at every
point) form a vector field which is called Ihe held of gravitation
corresponding to the given niii<s distribution.

* Actually, if a straight line /. lies in the tangent pi ana to the surface

( r (r, 1/, s)
const at the point .1/ tin* derivative of 1/ in the direction of >.
is obviously equal to zero:
(grad f ’. ;.) = 0
Of.
Thus, tlu* vector grad V is perpendicular to any vector lying in the tangent
plane and hence it. goes along the uoriiud to the surface.
c i l i.' X'lLU lilt; ( l i l t l i l t Dsift (i Lt 1 i / f tU it ill $ -i.
 OH. 6. FI EL D THEORY

If there are electric charges distributed in space they act upon

unit charge placed at a point M with a certain force F (M). These
forces constitute a vector field which is referred to as an electrostatic
lie Id.
A Held of gravitation and an electrostatic held are examples
of a held of force.
Let A (.1/) be a vector held in space. Taking a Cartesian coordinate
system in space we can represent A (.1/) as a collection of three
scalar functions which are the components of the vector. As a rule,
in what follows these components (projections of A (.1/) on the
coordinate axes) will he denoted by I* ( x . if. z) . Q ( x. y . z) and
U (x, y. z). In this chapter we shall consider the vector lie Ids whose
components are continuous and possess continuous first-order
partial derivatives.*
2. Vector Lines and Vector Surfaces. Let a vector field A (A/)
he defined in a domain Q. A curve L lying in Q is called a vector
line if the direction of the tangent to the curve at. each point coin­
cides with the direction of 1ho vector A at the point. In particular,
if A is the held of velocit ies of a stationary flow of a liquid its vector
lines are the trajectories of the particles of the liquid.
in some questions related to investigating vector fields the pro­
blem of finding a vector line, of a field A, passing 1hrough a given
point .l/0 plays an important role. 'I’lie problem can be formulated
analytically: it is necessary to determine a vector function r (/)
satisfying flu* conditions
r' (✓ ) - k \ (fL7)
r (/0) ^ r0
where i\, is Ilie radius vector of the initial point /u is the initial
instant, of time and X is an arbitrary scalar parameter. It can be
.shown that if the components /L and Ji of the vector A are con­
tinuously differentiable functions of the coordinates which do not
simultaneously vanish, conditions (0.7) in fact determine a single
vector lino passing through the point ;V/0.**
A hounded surface 2 lying in a part, of space where a vector held
A is defined is said to he a vector surface if at each point of the sur­
face 2 the normal to 2 is orthogonal to the vector A at the point.
Thus we can say that every vector surface 2 is made up of vector
lines: each vector line either entirely lies on 2 or has no points in
common with it.

- 11 is c l e a r t h a t if t h i s c o n d i t i o n is fu lf il le d for o n e c o o r d i n a t e s y s t e m it
a u t o m a t i c a l l y h o l d s f»»r a n y o t h e r <«n>rdinate s y s t e m .
** Til is is a c o n s e q u e n c e o f t h e e x i s t e n c e and u n i q u e n e s s t h e o r e m for a s o l u ­
t i o n of a s y s t e m o f d i f f e r e n t i a l e q u a t i o n s wi t h g i v e n i n i t i a l c o n d i t i o n s ( e . g ,
<ee 15 1, C h a p t e r 1. § U).
 2:38 M l l.Tl e l .K INTEGRALS. FIELD THEORY A ND SERIES

A part of space lying in the domain of definition of a vector field

A and bounded by a tubular vector surface is referred to as a tube
of the vector field. Such a tube is entirely composed of vector lines,
and every vector line of the held A either entirely lies within the
lube or is placed outside it.
If a held A is thought of as a field of velocities of a stationary
fluid flow, a tube of the field is the part of space which is traced by
a lived volume of fluid in the process of motion.
3. Types of Symmetry of Vector Field. As in the case of a scalar
field, the investigation of a vector field is simplified when it possesses
symmetry properties. Let us enumerate some of the most important
special cases.
(a) Plane-parallel field. If, for a given vector field A, it is possible
to choose a C-arlesian coordinate system in which the components
of tlie field A have the form P (x , y), Q (x, y) and R (x, y), that
is are independent of z, the field A is said to be plane-parallel.
If, in addition, we have R (.r, y) — 0 the held A is called plane.
An example of such a held is the velocity held of a stationary flow
of a liquid whose particles have the velocities parallel to a fixed
plane and independent of the distance from a particle to the plane
(a plane tlow). The vector lines of such a field are plane curves whose
shape and disposition are the same in each plane parallel to the
given plane.
(b) Axially symmetric field. A vector field A is called axially
symmetric if there exists a cylindrical coordinate system r, <p, z
such that at each point Af the vector A (A/) depends only on r and z
ami <loes not depend on <| . In oilier words, such a field goes into
itself when it is rotated about the z-oxis. In case the vector A (A/)
depends solely on r the field is «aid in ho cylindrical.
(c) One-dimensional field. A one-dimensional field is characte­
rized by the possibility of choosing a Cartesian coordinate system
in which the components of the field have the form P (x), 0. 0.
The family of the vector lines of a one-dimensional field obviously
coincides with the totality of all straight lines parallel to the x-axis.
4. Field of Gradients. Potential Field. Consider a scalar field
U (A/). Constructing the vector grad (I at every point AT we arrive
a la vector field which is the field of gradients of the scalar quantity
U. Next we introduce the following
D e f i n i t i o n * A vector field A (.1/) is said to be p o t e n t i a l if it can
be represented as a field of gradients of a scalar field U (A/), i c.
A — grad V
In this case the scalar field U is called the potential (potential func­
tion) of the vector field A.
 CH. 0. FIELD THEORY 231)

Let us take the following example. Put U = / (r) where* r =

«* x- y yl -r 2 s. We thus obtain the spherical field U. To find
the gradient of U we differentiate with respect to x and thus obtain
OU ^ f \ dr , vx
= l {r^ = < W r
and, similarly.
dU f* / \ z
s r - i «7
Consequently, we have
grad U = /' (r) ± , r — xi -r i/J zk ( 0. 8)

If a vector field A has a potential function this function is

uniquely specified, to within an arbitrary constant addend, by the
field. Indeed, if scalar fields U and V have the same gradient we
cun write
grad (£/ — V) = 0
Put then the directional derivative of U — V is equal to zero
in nil directions at each point which immediately implies that
U — V — const
The vector lines of a potential field A are obviously the gradient
lines of its potential U, i.e. the curves along which the rale of change
of U is maximal.
It is now natural to ask what are the conditions for a vector field
being potential. The answer to the question was in fact found in
Chapter 5. indeed, as was shown (sec Theorem 5.2), an expression
P dx + Q dy + 7? dz
(where P, Q and /? are continuous functions possessing continuous
partial derivatives of Lhe first order) is the total differential of
a single-valued function U (x, y, z) if and only if P, (J and /? satisfy
the conditions*
op _ oQ_ on = on OR ^ OP r
dy Ox ’ Os 0\j * Ox Oz ' '
But the relation
P dx + Q dy R dz — dU
is equivalent to

* Hero wo of course suppose that the domain in which the vector field A
is considered is simply connected.
2-10 ML'LTH’Ll*: I N T R O l l A l . 5 , FI K L D T il Hi »K V AND SERIES

ami hence conditions (O.U) imply that the lie Id ( / \ <>, J{) is polen-
tial.
Thus, lor a vector lielcl A = (J\ Q, H) with continuous and
continuously differentiable components fJ (x, //, z), (J (x, y, z) ami
II (x, y, z) to be potential, it is necessary and sufficient that equali­
ties (b.‘J) be fullilled.
If we are given a potential vector held A the potential function
can practically he found from its total differential as was illu­
strated in the problem considered in $ 4 of Chapter 5 (formula (o.43))
for the case of three independent variables and in § 4 of Chapter 4
(formula (4.5U)) for two variables.
The notion of a potential held will he discussed again in Sec. 5
of § 4.
Example. Let a mass w he concentrated at the origin O. If we
now place unit mass at a point M (x, y, z) it will he acted upon
by the gravitational force

I*' = — Y 7 1 r (<* — •** //J i- sk )

These forces are determined at all points in space and thus form
a vector field which is the field of gravitation of Iho mass point m.
This lielcl can be represented as the gradient of the function
ym
r
known as the Newtonian potential of the mass point to. To verify
that this function is in fact the potential of the held we take advan­
tage of formula (li.8) and thus obtain
, vm m
grad
* ——
r = — v1 r* r
—

§ 3. FLUX OF VECTOR FIELD. DIVERGENCE

1. Flux of Vector Field Across a Surface. In the foregoing chapter
(§ 2) we showed that the amount oT Fluid passing through a given
(oriented) surface 2 in unit time is equal to the integral

A n clo

where A n is the projection of the velocity vector A = {I*. Q. II) on

the (outer) normal to the surface.
We called this quantity the flux of (he lufuid (through the surface
2).
 CH. 6. FI E L D THEORY 2 'i J

Now let A be an arbitrary vector held and 2 be an oriented surface.

The surface integral
j j Ando

will be called the flux of Che vector field A across the surface 2 .
Thus, if A is the velocity of a fluid flow the flux of the vector A
across a surface is equal to the quantity of the liquid passing through
the surface in unit time. For a vector field of some other nature the
flux of the field may have another physical meaning.
Example. Let U = U (x, y, z) be a temperature held inside a
physical body and let A = grad U. Denoting by k the coefficient
of thermal conductivity we can apply the Fourier* law of heat pro­
pagation and express the quantity of heat dQ passing in unit time
through an clement do of an oriented surface 2 in the form
dQ — — k -I—do (0.10)
0U
where — is the derivative of the field of temperature in the direction
of the (outer) normal to do. (The minus sign on the right-hand side
of equality (6.10) is due to the well known fact that heat travels,
within a heat conductor, from the regions of higher temperature
to those of lower temperature, i.e. in the direction of decrease of U.)
Since we have
dU
On
(grad C)H
equality (6.10) can be rewritten as
dQ = —k (grad U)h do
whence it follows that the amount of heat passing in unit time across
the whole surface 2 is equal to
Q = — ^j k (grad U)n do (6.11)
V

introducing the vector

q — —k grad U
known as the heat flux vector we obtain
Q = j j<7« do
V

* Fourier, lean Baptiste Joseph (1768-1830), a prominent French mathe­

matician ami physicist.
I6—082 k
242 M lI/ril'L E IN T E G R A L S , FIELD THEORY AND SERIES

Consequently, the quantity of heat passing through £ in unit

time is equal to the flux of the vector q across the surface £ .
2. Divergence. Let A be a vector field which will be thought
of as a velocity held of an incompressible fluid. The liquid being
incompressible, the flux
^4fi rfc

of the vector A across a closed surface £ * is obviously equal to the

amount of liquid which is introduced or removed, within the
domain 12 (bounded by the surface 2) in unit lime. This quantity
characterizes the total capacity of the sources, in case 11 >►0, or the
sinks (also termed negative sources), if 11 < 0, lying in the domain Q.
Consider the ratio
n
vju)
of the fiiiid flux through the surface £ to the volume V (12) of the
domain 12 bounded by the surface. It expresses the mean source
(sink) density, that is the amount of the liquid introduced (remo­
ved) within unit volume of the domain 12.
Finally, let us consider the limit
l S
v da
lim
Q-*AI F («)
of the ratio where Ihc sym b ol lim indicates the passage to the
limit as the domain 12 is contracted to a lixed point Af. This lim it
is Iho measure of source (sink) density of the fluid at the point M.
It is a scalar quantity which is an important characteristic of the
field.
Now we can pass to general definitions.
Let A be nil arbitrary vector field. With every spatial domain £2
bounded by a smooth or piecewise smooth surface 2 (lying in the
part of space where the field is defined), we now associate the quan­
tity
f \ An do
• *’
which is the flux of the vector A across the outer side of the surface 2 .
We thus arrive at a set function ib (12) = J j An do. It can lie
z
easily verified that the set function is additive.

* Here we consider the outer side of the surface.

 CII. ft. F IE L D TTTF.OTXY 243

D e f i n i t i o n . The derivative of the function <I> (Q) with respect to

volume, that is the limit
$ $ An do
V

lim (G.12)
Q-+M ~V(Q)
is termed the d i r e r (fence o f the v e c t o r f i e l d A and denoted by
the symbol
cliv A
Thus, the source density of the iiehl of velocities of a fluid flow
which has been considered above is equal to the divergence of the
field.
T h e o r e m 6*1. If A = ( / \ Q, R) is a vector field, defined in a do­
main Q, such that the functions P, Q, R are continuous and possess
continuous first-order partial derivatives in £2, the divergence div A
exists at all the points of the domain and is expressed, in every Cartesian
coordinate system, by the formula
dP OQ OR
div A dx ^ Off bz (6.13)
Proof. Let us apply the Oslrogradsky formula

v n
By virtue of the theorem on differentiating a triple integral with
respect to volume (see § 1, Sec. 5 of Chapter 2), the derivative with
respect to volume of the right-hand side of the formula exists and
is equal to — ox
4- —
oy
-b —
oz
. Consequently,
1
the derivative with
respect to volume of the left-hand side also exists and is equal to
the same expression. The latter derivative being, by definition, the
divergence div A of the vector iield A, the theorem has thus been
proved.
Note. The relation
OP OQ OR
div A — ox----b by Oz
is often taken as a definition of the divergence. But such definition
is less convenient than the one given here because it is based on the
choice of the cocudinate system and therefore the fact that the sum
—
Ox
! -Oy- ; —
dz
is independent
1
of the choice of the coordinate
syslem should be additionally proved whereas Ibo independence of
expression (b.12) of the choice is quite apparent.
Thus, with every vector field A whose components are continuous
and have continuous partial derivatives of the first order, we
1G*
244 MULTIPLE INTEGRALS. FIELD TH EORY AND S E R I E S

associate the scalar function div A, the divergence of the vector

field A. Using this notion we can rewrite the Ostrogradsky formula
as follows:
An c?cr = j d i v A f/y (0.14)

Hence the flux of a vector A through the outer side of every closed
surface 2 is equal to the integral of the divergence uf the field A
taken over the domain bounded by the surface 2 .
3. Physical Meaning of Divergence for Various Types of Field.
Examples.
(a) As was shown, Tor the velocity held A of an incompressible
liquid moving in a spatial domain the expression
\ f f div Ady
•• • J
<1
is the measure of the total capacity of sources (sinks) placed in the
domain Q, and div A is equal to the source intensity per unit volume.
In particular, if A is a field of velocities of an incompressible fluid
whose flow has neither sources nor sinks, we have
div A — 0
(b) Let us now consider the field of gravitation of a mass distri­
bution and elucidate the physical significance of the divergence of
such a field. To begin with, we take the field produced by a mass m0
concentrated at a point (x0, //<>» 2o)- Then the force acting upon unil
mass placed at a point (x, y, z) is equal to
F= V '^ - S r 2-) (6.15)

where r = V (x — x0)2 -r (y — y o)2 I- (z — z0)2 and y is the con­

stant of gravitation whose numerical value depends on the choice
of the system of units. In what follows we assume that the system
of units is so chosen that y = 1.
Let us compute the divergence of force field (O.lfi). At each point
distinct from (x0l y0, z0) wo h a v e
X—Xq r3 — 3 (x — .r0)2 r ro>2
r3 ra
and, similarly.
d r2—3(y —yi))2
(™o ? ) ■ -0 fb
and
d (m0 r~ — 3 (s —co)2
dz 7 ^ ) ='»«
 CH. 6. FIELD THEORY 245

Adding together the results we obtain

3rg — 3 (y — t q)8 — 3 (y — y 0) 2 — 3 ( z — zp)2
r5 -o
But these calculations do not apply in the case of the point
(x0, zo) at which the finite value of the divergence cannot bo
defined. Therefore the value of the integral

(which is an improper integral in this case, see Chapter 9) cannot

be obtained by a direct integration if the domain £2 contains the
point (x0, yo, Zo)- Therefore the expression on the right-hand side
of Ostrogradsky formula (6.14) is undetermined in this case. But
we can easily find the quantity on the left-hand side of the formula,
that is the flux of the vector F across the surface 2 bounding the
domain Q (and attribute the numerical value thus obtained to the
integral the point (x0, y0l z0) is contained
in Q).
We now proceed to calculate the flux. Let us first take, as the
surface 2 , a sphere of radius a with centre at the point (x0, y0, z0).
The direction of vector (6.15) coincides with the direction of the
normal to the sphere at each point of the sphere. Therefore, in this
case, the projection of vector (6.15) on the normal is equal to the
length of the vector, i.e. to the constant quantity — .Consequently
we have

Substituting any other closed surface 2 ,, containing the point

(xrt, Vo, z0) in the interior of the domain bounded by it, for tin
sphere 2 wo arrive at the same result. In fact, we can choose the
sphere 2 with a sufficiently small radius a > 0 so that it is
entirely contained within 2 ,. Then we have

since the left-hand side of this equality is equal to the flux of tin
vector F across the boundary of the spatial domain in which
div F = 0
Hence, wo obtain
246 MULTIPLE INTEG RAL S. FI ELD THEORY AMD S E R I E S

Now we shall consider the field of gravitation produced by several

mass points. This held is equal to the sum of the fields corresponding
to eacli separate mass. The flux of a sum of iiclds through a surface
being obviously equal to the sutn of the fluxes of the summands, it
follows that the flux of I lie held of gravitation, produced by a system
of mass points, across a closed surface is equal to the sum of the
masses contained inside the surface multiplied by 4n.
Applying the well known technique of passing to the limit from
a system of material points to a mass continuously distributed in
space with a volume density p (x, y y z) we can show* lhal for a
continuous mass distribution the flux of the field of gravitation
through a closed surface 2 is also equal to the total mass, contained
within this surface, multiplied by 4ji. But the total mass can be
represented as the integral of the density p (x, y, z) taken over the
volume Q bounded by the surface 2 . Therefore, denoting, as before,
the vectorial value of the gravitational field at an arbitrary point
(■£> y% z) by the symbol F (x, y , z) we can write
^ j F n (x, y, z) def = 4ji j j j p(x, y %z) dv

whence
SV $
4np (x, i/t 2 ) lim
y, *>
The integral on the right-hand side is the divergence of the vector
field F. Thus, we finally conclude that the divergence of the field
of gravitation produced by a continuous mass distribution is equal
l.o the volume density p (x, i/, z) oT the distribution multiplied
by 4n.
(c) The argument which has been applied to the field of gravita­
tion can also be used for investigating the electrostatic hold. This
results in the Bauss theorem which is widely applied to
various problems concerning electrostatic fields, e.g. to the problem
of determining the electric field intensity in the capacitors of various
types. The theorem states that the divergence of an electrostatic
field is equal to the charge density multiplied by 4.t .
4. Solcnoidal Field. A vector field whose divergence is identically
equal to zero is said to be solenoidal**. As was seen, the velocity

* The rigorous justification of such a passage to the limit is dealt with

in the so-called potential theory ami is based on the* Ihenry o[ integrals involving
parameters. The fundamentals of the theory of integrals dependent on a para­
meter are presented in Chapter 1 0 .
*'* F r o m t h e C r o o k w o rd a o i / . q v t u b e .
 OH. G. FI EL D THEORY 247

field of an incompressible fluid is an example of a solenoidal field

in case there are no sources and sinks, i.e. when there are no points
at which the fluid is in Ireduced or removed.
For the solenoidal fields, we have the so-called law of conservation
of intensity of a vector lube which we are going to deduce here. Let
A be a solenoidal field. Consider a tube of the vector held and take
its part contained between two sections 2^ and S 2 (see Fig. (5.7).

These sections together with the lateral surface form a closed

surface 2. The field being solenoidal, i.e. div A ^ 0, the Ostro-
gradsky theorem implies that
jjvl„da = 0
*2
But we have
^ j' An do — j j An do [- \ An d o -}- j j An do (6.1G)

where each integral is taken over the outer side of the corresponding
surface. The third summand on the right-hand side is equal to zero
since, by the definition of a tube of a vector, the vector held A is
directed perpendicularly to the normal to the surface 2* at each
point of this .surface and therefore, on I s . we have
An = 0
If we now take the inner side of the section i.e. reverse the direc­
tion of ils normal, and retain the outer side of the surface 2 2 equa­
lity (G. M>) turns into
= (6.17)
V, 12
Hence, the flux of a solenoidal vector field A across every section
of a vector tube has one and the same value. If the vector field A
is interpreted as the velocity field of a flow of an incompressible
liquid having neither sources nor sinks, relation (G17) shows that
the amount oF liquid flowing through a cross section of a vector
lube is the same for all the sections.
 248 M ULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

5. Eq uation ol Continuity. As an application of the above notions,

let us derive one of the basic equations of motion of fluid, the so-
c a l l e d equation of continuity. Let A be the held of velocities of a
moving fluid. We shall suppose that the domain in which the flow
js considered contains neither sources nor sinks, that is the fluid
does not concentrate toward or expand from any point of the domain.
13ut in contrast to our previous considerations, we shall not impose
the condition that the fluid is incompressible and therefore the
density p may depend not only on the coordinates x, y and z but
also on time t.
Let us investigate the relationship between the velocity of the
fluid and the variations of its density. For this purpose we pick out
a closed volume Q and find the increment AQ of the quantity cf
fluid (contained in £2) corresponding to time period At by applying
two different ways of computing A(J. Let p (x, y, z, t) be the den­
sity of the fluid at moment t at an arbitrary point (x, z) £ Q.
Then we obviously have

Q
On the other hand, the variation of the quantity of fluid contained
within the volume & is equal to the flux of the fluid through the
surface 2 (bounding the volume) multiplied by At. Therefore it
is equal to - At (pA)wdo where n is the outer normal, the
V
minus sign being taken because the quantity of fluid contained
within a volume decreases when the velocity is directed outward.
Transforming this surface integral by means of Lhe Ostrogradsky
theorem we obtain
AQ = — At j j ^ div (pA) dv

iSow equating the two expressions for AQ and cancelling out At

we reeei ve

Q Q
The last equality holding for any domain £2, the integrands are
equal, that is
— div (pA)
\Ve have thus derived an equation connecting the velocity and the
density of a moving liquid for any region which contains neither
sources nor sinks. Equation (G.18) is known as the equation of
continuity*
 CIT. 6 FIELD T I IE O fi Y 249

Introducing the vector J = pA (called the fluid-flux density vector)

we can rewrite the equation of continuity in the form
4 £ + d iv J = 0 (li. IS')
ot ' v '

6. Plane Fluid Flow. Ostrogradsky Theorem for Plane Field. Let

us consider a plane vector held, lhat is one whose components have
the form
P = P (x, y), Q = Q (i, y), ft = 0 (0.10)
in an appropriately chosen Cartesian coordinate system, (see § 2,
Sec. 3). The held can be thought of as a velocity field of a fluid
whose every particle moves in a plane parallel to the x, //-plane
with a velocity independent of its distance to the latter (this is
a so-called plane fluid flow). The divergence of such a field is equal to
dP , dQ
Ox * dy
Let S2 be a right cylinder of unit altitude with base G (lying in
the x, y-piane) and a lateral surface 2 (see Fig. 6.8). Now we put
down the Ostrogradsky formula for the domain Q. For this purpose
z ,

Fig. 0.8

we lake into account that the numerical value of the triple integral
dP dO
of — ~ over Q is equal to the double integral of this expression
over the plane region G and that the flux of vector (6.19) across the
surface 2 is equal to the line integral
\ [/^cosOi, x)-h(^cos(n, y)]dl
mi
I.
where n is the normal (in the x, y-plane) to the contour L hounding
the plane figure G. Besides, the fluxes through the upper and the
lower bases of the cylinder £2 are equal to zero, the latter assertion
being a consequence of the fact that vector (6.19) is perpendicular
to the z-axis. It follows that the Ostrogradsky theorem for the plane
held A and the cylindrical domain Q is expressed by the relation
J [ P cos (n, x) + Q cos (n, y)\dl = ( ~ ) dxdV ((;-20*
L 'C
250 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

Finally, %ve disregard the third coordinate z and consider (6.19)

to be a vector held defined in the x , y-plane. Let us call the line
integral
f IP cos (n, x) 7 Qcos (n, i/)| dl
- - (G-21)
L
the flux of vector field (6.19) through the contour L. Then formula
(t>.20) expressing the Ostrogradsky theorem for an arbitrary plane
held A can he interpreted as follows: the double integral of the diver­
g e n c e of the plane held A over a domain G is equal to the flux of the
vector A through the boundary L of the domain.
It can be easily verified that formula (6.20) is equivalent to
Green's formula (4.45). Actually, if we denote by a the angle bet­
ween the tangent to the curve L and the positive direction of the
x-axis we can write
cos (11, x) — —sin a and cos (11, y) = cos a
Therefore integral (6.21) turns into
^ (Q cos a — P sin a) dl
L
or
^ Q d x — P dy
L
Now transforming the last line integral into the corresponding
double integral by means of Green’s theorem we arrive at formula
( 6 . 20 ).
The above argument can he reversed, i.e. if equality (6.20) has
been established we can derive Green's formula from it.
Thus, Stokes’ theorem and Ostrogradsky’s theorem turn into
Green’s formula in the case of a plane field.

§ /i. CMICULATION.
ROTATION
1. Circulation of Vector Field. Let A — (P, (j, /?) be a vector
held and L be a smooth or piecewise smooth curve.
The line integral
\ P dx ~r Q dy -\- H dz
v
1.

which can also be written as

\ - I t dl
*
/.
 CH. C. FIELD THEORY 25!

where A x is the tangential component of the field A along the con­

tour L (i.e. the projection of A on the tangent line to L) is referred
to as the circulation of the vector held A over the closed curve L.
As was shown in § 2 of Chapter 4, if A = (/*, Q, H) is a field of
force, its circulation over a contour L is equal to the work of the
force along the path L. For the fields of other nature the circulation
may have another physical meaning.
2. Rotation of Vector Field. Stokes’ Formula in Vector Notation.
If L is a closed path the line integral
P dx -|- Q dy -j- H dz
L
can be transformed into the corresponding surface integral by apply­
ing Stokes’ formula (5.41):
§ P d x Q dy + It dz = —^ ) d x d V+

+ ( £- r)dtfdl+( r - |r ) d‘dz
4 4 4 (°-22>
The integral on the right-hand side of equality (6.22) is taken over
an arbitrary surface 2 “pulled over” the contour L (i.e. a surface
whose boundary is L). This integral can be interpreted as the flux,
across the surface 2 , of the vector

+ + (O'23)
which is called the rotation (curl) of the vector field A and is denoted
hy the symbol rot A (or curl A).
Thus, hy definition,

Using the notion of rotation we can rewrite Stokes’ formula in

tile following form:
(6.25)
§L
Thus, the circulation of a vector field A over a closed contour L
is equal to the flux of ihe rotation of the vector field across an arbi­
trary surface (lying in the domain of definition of the field) whose
boundary is L.
’Hie above definition of the rotation of a vector field A involves
not only the field itself hut also a certain coordinate system x , y , z.
liul the vector rut A does noL in fad depend on the choice of the
2.')2 MU LTI PLE INTEGRALS, FIELD THEORY AND SERIES

coortlinale system and is uniquely determined by the field A. To prove

this let us write Stokes’ formula (G.25) for a plane surface Z and
the contour L bounding this surface and then apply the mean value
theorem to the surface integral on the right-hand side of equality
(G.25). This results in
\ A x dl
( r o lA O T j .a

where M* is a point belonging to the surface 2 and a is its area.3

Let us now contract the surface E to a fixed point M so that
the normal n to the surface all the time retains its direction
invariable. In the limit we obtain
f -»TI"
(rot A (.l/))n lim a (6.26)
r—m
Since the circulation of the vector A over the contour is independent
of the choice of the coordinate system equality (G.2G) implies that
the projection of rot A on the normal u does not depend on the
choice. But the direction of the normal n can he chosen quite arbi­
trarily and thus the projection of the vector rot A on any fixed axis
is independent of the choice of the coordinate system. Consequently,
the same is true for the vector rot A itself.3**
3. Symbolic Formula for Rotation. The expression of the rotation
of a vector field A = (/■>, Q, /?) can be conveniently written as the
symbolic determinant
1 J• k
0 d d
Ox Oy dz
(G.27)
P n

where i, j and k are the unit vectors in the directions of the coordi*
d d ^
nate axes and the symbols
J
—
dx
, dy
and -r-
6z
are understood in the sense
that the multiplication of such a symbol by a function means the
differentiation with respect to the corresponding variable, e.g.
—0 ,,
(J means dQ .
dx v dx

* Here, as usual, we suppose that the first order partial derivatives of the
components P> Qy li of the vector field A with respect to r, //, z are continuous.
** It is supposed that \vc take only right handed coordinate systems. If
wo pass to n left-handed system (in which the positive direction of describing
the boundary of a surface is such that the surface is always kept on the right)
the direction of tho vector rot A changes to the opposite.
 CH. 6. FIELD TH E O R Y 2f»;

Indeed, (formally) expanding determinant (6.27) in minors ol

the first row we see that
i j k
O d d
Ox Oy dz
P Q R
4. Physical Meaning of Rotation. The physical meaning of rota
tion can be elucidated in the following way. Let us again interpre*
a vector field A as the velocity field of a fluid flow. Imagine that wt
put, in the flow, an “infinitesimal turbine” whose plane vanes (parallc
to the axis of the turbine) are placed round its circular periphery I
(sec Fig. 6.9). The fluid flow will make the turbine rotate about tin

Fig. 6.0

axis with an angular velocity which is, generally speaking, depen­

dent on the direction of the axis.
It appears natural to assume that the magnitude v of the lineal
velocity of each point, of the circle L is equal to the moan value o'
the projections of the vector A on the corresponding tangents to L
that is
= (6.28;
*1 .

IJy Stokes1 formula (0.25), line integral (6.28) can be transformec

into the surface integral
^ j j lr o t A ) .* (6.29;

taken over the surface 2 of the turbine. The turbine being cousiderec
infinitesimal, we can write the integral ^ j (rot A)n do as the pro
duel of1 the area of the turbine by the value of (rot A)n at its centre
i.e. in the form
jxR~ (rot A)*
2M MULTIPLE INTEGRALS, FIELD T H EO RY AND SERIES

Relation (0.28) then takes the form

v— (rot A),*
As is known, the projection of a vector assumes its maximal value
(equal to the modulus of the vector) when the axis the vector is
projected on is parallel to the vector and has the same direction.
Therefore, if we place the axis of the turbine so that its speed v
is maximal (i.e. so that the direction of the axis coincides with
the direction of rot A) we obtain
Vmax = ~2~| rot A |
or
|r o t A |= ^ p
But is equal to the angular speetl <o of the turbine anti hence we
have arrived at the following result: if the turbine is placed in the
flow so that its speed of rotation becomes maximal its angular speed
\
is equal to — | rot A | and the direction of its axis coincides with
the direction of the vector rot A.
Consequently, the vector rot A characterizes “the rotational
component” of the velocity held and is equal to the doubled angular
velocity of rotation of an infinitesimal particle of the fluid.
/examples
1. Consider the vector field A with the components
P - - —yw, (J = jcm, R =. 0
The field can be regarded as the velocity field corresponding to the
rotation of the entire space (filled with a fluid) about, the z-axis with
angular velocity to = (0, 0, o>). ll can be easily verified that, the
rotation of the vector field is equal to 2<ok, that is rot A (A =
= (P, I/)) is directed along the axis of revolution and its mag­
nitude is twice the angular speed « (see Fig. (5.10).
The physical significance of this result is that every particle (of
a fluid which is in a rolary motion about the z-axis) passing through
a point (x , y, z) at moment t simultaneously takes part in two
motions, namely in the instantaneous transient motion with vel oci t y
v — (—yo>, axn. 0) and in the instantaneous rolary motion. It is
obvious that the instantaneous angular velocity of rotation of
every particle coincides with the angular velocity of the macro­
scopic motion of the fluid as n whole. Hence, the field of the* rotation
rotA is also constant, and equal to 2io and the fluid can be thought
of as being entirely composed of infinitesimal curls (vortices).
 CU. is. FIELD THE O HY 2j5

2. Consider a liquid flowing in a constant direction with a con­

stant velocity, i.e. suppose that the functions /*, and R are iden­
tically constant. In this case we have rot A — 0.
3. Let P = y, Q = 0 and R = 0. Then
rot A = —k
Here the rotation is different from zero at each point although
all the vector lines are straight lines parallel to the y , 2-plane (see
Fig. 6.11). One may think that this result contradicts the assertion
that rot A characterizes the “rotational component” of the field A.
But in fact this is not so: in this example the “rotational component”

Fig. 0.10

is due not to the twist of the vector lines (which are rectilinear here)
but to the variation of the velocity which is dependent on the distance
from the x, z-plane. If we place an infinitesimal turbine in this

flow whose velocity is equal to (t/, 0, 0) at each point (x, //, 2 )

it will not apparently be in a stale of rest unless its axis of rotation
is perpendicular to the 2 -axis.
4. Let us take the vector field A with the components

Q -y iib - Jt = ° <G-30>
This field can ho regarded as the velocity field of a fluid whose par­
ticles move, in the x, jy-plane, along the hyperbolas xy = C (Fig. 6.12)
in such a way that the magnitude of the velocity is equal to unity
at each point. Compel imr tlie divergence and the rotation of ifip
25C MULTIPLE INTEGRALS, FIELD TH EO R Y AN D SERIES

field we find
div A = — ( —/ -T —) r -jr- { — ~~~'/ ) y*-**
OT v Vx - -}- J/- / <?i/ V j i J- yi / {xi + yi f V
and
2xy
,o l A — f — ( ..ZiL- j - l ( - - . 4 = | I
I \ya;4_- V ■y'xi-t-y2/ J
(x--- </2)3/“
The divergence is positive here for | y | > | x | and negative for
J y | <C | x j. From the physical point of view, this means that the
motion of an incompressible fluid described by field (6.30) is only
possible if there are sources in the regions where | y | > | x | and
sinks where \ y | < 1x |. The rotation of held (6.30) is directed
along the --axis at each point, this being so for every plane field
parallel to the z t y-plane. In the case of field (6.30) the vector rot A
goes along the positive direction of the z-axis in the second and
fourth quadrants and in the negative direction in the first and second
quadrants. Both divergence and rotation of field (6.30) tend to zero
as xl -p y~ —v oo, that is when the distance from the origin of coor­
dinates increases.
5. More on Potential and Solcnoidal Fields. The notion of rotation
discussed in Sec. 2 is directly related to the definitions of a potential
and of a solcnoidal field introduced above.
We have defined a potential vector field as one representable in
the form of the gradient of a scalar field. As was shown, a vector
field A = (/J, Q, R) is potential if and only if its components
satisfy* the conditions
dP _ oQ_ oQ_ __ on_ dll OP
(it) dx " f?2 f*tf 'It tlz

But these three conditions mean that all the three components of
the rotation of the field A are identically equal to zero.
Thus, we can state that for a vector field A to be potential it is
necessary and sufficient that the condition**
rot A = 0
be fulfilled. Hence, we always have r o t grad U — 0 for any func­
tion U.
The notion of a solenoidal field introduced in § 2 is also connected
with the notion of rotation. Indeed, the direct differentiation shows

* As before, \vc suppose that the functions P, Q and R are continuously

differentiable and that the domain in which the field is considered is simply
connected.
** In this case we suppose that the components of tho vector field A defined
in a simply connected domain possess the continuous partial derivatives of
<ho first rind s e c o n d o r d e r s with respect to x . u and z.
 CH- <5- FIELD THEORY 257

that, for cverv vector field A. we have the relation

Qli _o_ i 0Q_ op \
div (rot A) = -jtj ( 0\J oz \ Ox 0<j )
0

and hence any vector field representable as the rotation of another

vector field is always solenoidal.
It can be proved that, conversely, every solenoidal field can be
represented in the form of the rotation of a vector field. In other
words, for evers vector field A satisfying the condition div A -= U
y*

Pig. 6.12

there exists a field B such that V rot B. Such a vector hold B

is not uniquely determined by the field A but only to within a sum­
mand of the form grad ( being an arbitrary function.
If A = rot B the field B is termed a vector potential of the field A.
Although the potential and the solenoidal fields do not exhaust
all possible vector lield> it can lie shown that an arbitrary vector
field can be represented as a combination of the fields of these two
types. More precisely, it can be proved that every vector field A
can be represented in the form
A = B -f- C
where the field B is potential and C is solenoidal.

§ 5. HAMILTONIAN OPERATOR
1 . Symbolic Vector V« In § 1 wc introduced the notion of the
gradient of a scalar field. The process of obtaining the vector field
grad f' from a given scalar field V can be regarded as an operation
which is in many respects similar to differentiation with the only
difference that the latter transforms a scalar into a new scalar
whereas the former operation applied to a scalar yields a vector.
The operation of passing from U to grad V is usually designated
bvV the svmbol
t. V introduced bv■ Hamilton*. This symbol
* is read

Hamilton. William Rowan M80S-IXfim, an fr»=t> ? r > ♦»>•••»•

17-0824
25$ M U LT I PL E INTEGRALS, FIRM) THEORY AND SERIES

“nabla” ** (or “tier) and is called the Hamiltonian operator.

Thus, we have, by definition.
V (■' = grad U
It is convenient tointerpret
^ 0
theoperator
^
V as a symbolic vector
with thecomponents
1
— , ~Oy and —
Ox Oz
* *:

The application of this operator to a scalar function I' can be per­

formed as formal multiplication of the “vector” V by the scalar
U:
■. r 7 , r , / ^ ^ ^ \ ** Olj-OP - a
V6 grad
”
C- — i -' —
\ Ox Oy
j t- -oz - k f C == 1 —— 1- j
Ox • 0
— b k oz
dy
The vector V can also he conveniently used for writing formulas
involving many other operations of vector analysis. For instance,
if A = (/*, Q, 7?) then
div A = —
ox
P +* -!LQ : / /
Oy v • o z
(V, A)
that is the divergence of a vector lield A is the (formal) scalar pro­
duct of the symbolic vector V by Ilie vector A. Similarly, we have
rot

-IV , A |
i.e. the rotation of a vector held A is the (formal) vector product,
ut the vector V by the vector A.
2. Operations with Vector V. The expedience of the introduction
of the symbolic vector V lies in the fact that it enables us to deduce
and write various formulas of vector analysis in an abbreviated and
visual form. There are some simple examples below.
In Sec. "> we showed, by means of the direct calculations, that
rot grad (J —U
and
div ml A — 0

* This term, also introduced hv Hamilton, originates from the name

liable (Greek vccBa«) of an ancient musical instrument of triangular form.
** When writing the rotation as a symbolic determinant, vve have already
seen Hint it is convenient to think of the operation of differentiation as the (for­
mal) multiplication of the symbol 7- ^nr -- , 7- etc. j by tlu> function wlm.se
derivative is computed.
 C.H. 6. FIELD THEORY 2 5 9

Tlie.se identities can he rewritten willi ihe help of tlie vector V as

two relations of the form
lv , V l'\ = 0
and
(V, V, A) = 0
The left-hand side of the former is the “vector product” of two
’’vectors” V and y£* which only difier from each oilier in the
scalar factor 6 ', and we have the “triple scalar product” involving
two equal vectors on the left-hand side of Ihe latter relation. There­
fore these expressions turn out to he equal to zero in accordance
with the general rules of vector algebra.
We can directly verify that most of the basic operations performed
on the ordinary vectors can be extended to the symbolic vector V
which makes it possible to derive formulas of vector analysis
by applying the rules of vector algebra to expressions involving V-
l'3nI it should be noted that there is no complete analogy between
the symbolic vector V and “true*’ vectors. Namely, only the formulas
(containing the vector V) which do not involve the application of
the operator V to products of variable quantities (scalar or vector)
are completely analogous to the corresponding formulas of vector
algebra. Hut if an expression contains a product of two or more
variable quantities to which the operator V .should bo applied we
cannot follow the ordinary rules of vector algebra. To establish
the corresponding rules for application of the symbolic operator
V to such expressions we shall consider some typical examples.
1. Lot f t' (.<•. ij. c) ho ,\ scalar field and A ; A (x. z)
a vector field. We shall find div(£/A), i.e. ( V , UA). The application
of the vector operator V reduces to performing the operations of
differentiation involved in it. Hut. as is well known, when diffe­
rentiating a product, wo ran differentiate the first factor considering
the others to be constant, (hen differentiate the second factor as
if the other factors were constant and so on and tlien take the sum
of the expressions thus obtained.
Let us mark by the sign “j” the factor to which the operator V
should be applied. Thou, as it can lie easily verified, the expression
for div(£7A) can lie written as
(V, UA) = (V, UA) \ (V, £ A)
The factors, in each summand, that are not subjected to the operator
V can be taken outside the sign V. Consequently \u* obtain
(v , u \ ) = (v , / A) — (v, i t \ ) - ( y r , A) r <y, \)
which can be written down in the ordinary notation as
div (( A) — (A. <?rad U) - // die \
200 MULTIW.F. INTEGRALS. FIELD THEORY AND SERIES

2. Consider the expression

grad (UV)
which can be written symbolically in the form
VUV
Fo llowing the above rule we can write
V U V - v U V 4- VUV = W U + U v V
or, in Ilie ordinary notation,
grad(tfF) = V grad U + U gradl'
These examples enable us to formula Ie the rules according to
which the operator V should be applied: in the expressions containing
a single variable quantity the symbol V can be operated on as an
ordinary vector and in the expressions involving products of several
variable quantities the operator should he applied in accord with
the rule for differentiating a product. Finally, Ihe application of
V to a sum of any summands is performed termwise, i.e. V is sepa­
rately applied to each summand and the results are then added up.
In conclusion, we present a list of formulas connecting the opera­
tions of computing the gradient, the rotation and the divergence
with basic operations of vector algebra:
1. div (t/A) « (A, grad U) -f U div A,
2. grail (UV) « V grad U -|- U grad F,
3. rot ( U \ ) = U rot A r [grad U t Al,
A. div [A. B] — (B, rot A) — ( \ . rot B),
5. rot [A, Bl = (B, V) A — (A, V) B A div B B div A,
0. grad (A, B) - (B, V) A + (A, V) B + IB, rot A] +
-j- [A, roL B],
In particular, putting A -- B in the last formula we derive
grad = (A, V) A -r (A, i oL AJ
The first two of these formulas have already been deduced. The
other can be obtained by applying the operator V in accordance
with the above rules and ordinary formulas of vector algebra. In
particular, to liud the expression for rot (A, Bl wo symbolically
write it in the form
IV, [A, Bll
and apply the well known formula* for transforming a triple vector
product:
la, [b, cl I = b (a, c) — c (a, b)

* For a triple vector product of tlic form |[a, bl, c| the corresponding for-
iniihi is wrilton ns
IIn. !i|. cl = I) (a. cl — n Hi. cl
 CU V KIEL1» THEORY 261

An expression of the form (A. V) B encountered in formulas 5 ami (i

is understood as the vector quantity
oBx , t oBx oH, oB r/ C'By OB,
.•1. , Ax . .1 - , U
'>X !>~nr 01/ oz
dli. oB. , t itIt. '
A:c ~0z ,
tit "//
(where A x. . 1 ,,. A . and B x. B y. B t are the components of the
vectors A = (A T. .1,,, .1.) ami = (ftx. R y. B ,)) which can be
regarded as the result of applying the “scalar” operation
a
(a . v , A oz
to each component of the vector IT*

§ 6. REPEATED OPERATIONS INVOLVING V-

LA PL ACIAN OPERATOR
1. Repeated Differentiation. Tn §§ 3 5 we introduced the notions
of gradient, divergence and rotation. In applications of vector ana lysis
we deal not only with these basic operations blit also with their
combi nations. The most often used operations of this kind are
those containing second derivatives of the fields.
We can compose nine different combinations of the symbols grad,
rot and div involving second derivatives but some of them are
senseless. For instance, such is the operation
rot div
which cannot be applied either to a scalar field or to a vector held
because the notion oi the divergence has been introduced for the
vector fields A and the expression div A is always a scalar quantity
whereas the rotation was only defined for the vector fields and thus
it is senseless to speak about the rotation of a divergence. All pos­
sible combinations are given in the tabic below, the senseless com­
binations being marked by shading (he corresponding positions of
the table ('•an* the next page).
We see that, among these operations, there are only two that can
be applied to a scalar field V, namely
rot grad U
and
div grad U

* In the formula for (A. V IV) we have mooted the components of the vectors
A and IV liy liie same letters A and B with subscripts .r, 7 and 2 (designating
the coordinate axesi whirlt mates t.lie formula look symmetric, with respect
f r» flip Ir-ttnrs I n w l n f fpllmvc no clicill licit ticn iMc nnl <\t inn
2C
.2 MU LT IP LE IN T E G R A L S . FIEL D THEORY AND SERIES

The former is the rotation of the potential vector field grad V

and, as was shown, rot grad fl is always identically equal to zero.
The latter expression div grad U is not equal to zero in the
general case. It is called the Laplacian operator* and denoted by
AU. Taking advantage of the expressions for the gradient and rota­
tion in Cartesian coordinates (see formulas (6.5) and (0.23)) we
derive
rtif . <nj \ , ifiu < r-v
AU — div grad U Oy ^ ftz / 0.1 - OtJ- Oz'l
The divergence and the gradient being independent of the particular
choice of the coordinate system, the quantity A/-’ is llius completely
specified by the field £/ and does not depend on the coordinate system.
Later we shall dwell in more detail on Laplace’s operator.
'rite Laplacian operator A can be considered to be the (formal)
scalar square of the symbolic vector V. In fact, we have
. / ft \2 / o \2 / ,? v- o1 , o- o-

and hence
qv; q-u OV!
(V, V) u Ox- 1 Oy- Oz- AU, i,e. (V, V) -= A

It is .sometimes necessary to apply the operator A not to a scalar

quantity but to a vector. If
A - A x\ + A,J r A .k

* Xaiiu-d after P . 6 . Laplace ( 174‘J-l627), a prominent french matlicMuatician

and astronomer. Mon* precisely, the term “Laplacian operator** is used not
for the expression At' (which i« the result *d applying the operator to a scalar
field U ) hnt for the symbol A -= div grad.
 CU. «; FIELD THEG15Y 203

then, by definition, AA is understood as the vector

A.4 vi ■}■ A/I pj A.l zk
As will be shown later, this quantity does not in fact depend on
the choice of the coordinate system and is completely determined
by the vector held A-*
Let us now proceed to study the operations involving repeated
differentiation which make sense for the vector fields. There are
three such operations in the above table, namely
grad div A
rot rot A
div rol A
The expression div rol A was encountered in § \ when we deduced
the conditions for a held to be solenoidal. As was shown, we always
have
div rot A ~ 0
On the contrary, the expressions grad div A and rot rot A may
be nonzero. They are widely applied in various problems of
mechanics and electrodynamics.
Let us derive a formula connecting these quantities. For this
purpose we first consider the expression
rot rot A
which can lie wrillen, in symbolic form, as
lY, [V, All
Applying the formula for a triple vector product given in § f», Sec. 2.
we obtain
lY, (V, All = V (Y. A) - (V, V) A
that is
rol rot A — grad div A — A A (0.31)
In particular, it follows that the quantity AA defined above as
the result of application of the Laplacian operator to each component
of the vector function A is in fact independent of the choice of the
coordinate system because this is the case for Ihe quantities rot rot A
and grad div A.
Kxpression (15.31) involving only one variable quantity, we have
applied here the ordinary rules of vector algebra in operating on
tlie symbol V. AVe suggest that the reader directly derive equality
(0.31) without resorting lo the symbol V and compare the calculations

* II should bo nolorl that Ih<* original definition of A as A - div grad only

applies when we consider an expression of Ihe form S C when* C is a scalar.
 MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

with the above because this will show the advantage of introducing
the Hamiltonian operator.
2, Heat Conductivity Equation. To illustrate the possibility of
application of the concepts of vector analysis let us derive the equation
for a temperature field inside a physical body subjected to heating.
Denote by C (x, y, z> t) the temperature at an arbitrary point
(x, i/y z) of the body at moment 1. Let us (mentally) pick out.
within the body, a voiume Q bounded by a surface Z and compute,
by means of two different ways, the variation of the amount of
heat, contained in the volume, during an infinitesimal time period df.
In every infinitesimal element of volume dv taken inside the body
the increment of the temperature corresponding to the time interval
dt is equal to — 01 dl, the mass of the element being equal to <> dr
where p is the mass density. Consequently, the variation of the
quantity of heat in the volume element dr is expressed as

where c is the specific heat, the quantities c and p being considered

constant. Accordingly, the variation of the quantity of heat, con­
tained in the whole volume IT during time dt is equal to

Q
On the other hand, the quantity d(J can he found by computing
the amount of heat passing through the surface Z, hounding Ilie
volume o . during the time period from t to t \ dt. By Sec. 1 of
§ 2 , the amount of heat passing during the time interval dt through
an idcineiiLii y aiva d<J (in the direction of the outer normal to T)
is equa I to
- dt k (grad U)n do
and consequently the resultant increment of quantity of heat con­
tained within the volume Q bounded by the surface 2 is given by
the surface integral

Transforming this integral into the corresponding volume integral,

by applying the Ostrogradsky theorem, we receive

v Q O
 cn. c. riEi.n theory 2U

Now equating the two expressions obtained for do and cancelling

out dt we deduce the relation
»U
o< cpdv ( l l ' w ' r f r
a
which is valid for any spatial region Q taken within the body. It
follows that the integrands on Ilie left-hand and right-hand side*
are equal, that is
4£-««*A£*
Ot \
—Cp )
I '

We have thus derived the equation, called the heat conduclivit\

equation, for the function U describing distribution of tempera­
ture in a heat conductor.
3. Stationary Distribution of Temperature, Harmonic Functions
Wc have shown that a distribution of tempera lure inside a physica
body must satisfy equation (0.32). fn a particular case it may tun
out that the body is in a state of thermal equilibrium. i.e. its tempo
rature does not vary either on the boundary or in the interior of tin
OU
body. Then we must have — = 0, and equation (0.32) takes the fonr
At/= 0
or. in Cartesian coordinates,
, re c
= 0
0 c* ' 0y*
A state of thermal equilibrium can be thought of as follows. Sup post
that a constant temperature (independent of time) is kept at end
point of the boundary of the hotly (hut the temperature may vary
in the general case, from point to point, on the boundary). Then
after a sufficiently large time period has elapsed (strictly speaking
an infinite time period) there will he a stat ionary, time-independent,
temperature distribution. Such a distribution retains a constant
value (which may be different for different points) at each point
of the body. Apparently, such nil equilibrium temperature distri
bution is uniquely determined by tIn* temperature conditions on (li<
boundary of ihe body.
The equation
,\U = 0
is known as Laplace's equal ion. There are various stationary proces­
ses distinct from heat conduction which arc1 also described hv this
equation. Tor instance, those a r e a n equilibrium distri but ior
of electric charge over the surface of a conductor, a stationary fluid
flow in a closed region and the like. A scalar function V (r. //. z]
satisfying t h e ruination A /: — (t is called a harmonic lundim
2 t,fi M n . T I I ’I.r. IN T E G MAT..4'. FIF.T.D TTTEOEY A XL* S E T U E S

(Held). Thus, a stationary distribution of temperature within a

physical body is described by a harmonic function.
The function
-y (r = ]/ x~ y~ -- z-, ft = const)
is one of the most important examples of a harmonic function, it
can be interpreted as the potential function of 1 lie held of gravitation
(the electrostatic Jiehl) produced by a material point (point charge)
placed at the origin of coordinates. Let us verify that this is in
fact a harmonic function everywhere except the origin at which
it is not defined. Indeed, we have
<> /, A-J 0- k f r:* ^ A.r- — r-
e.t r r ;t ’ Oj - r * / <* * r*

and. similarly.
e- A- Ay- —/ - <t- k Az- —r -
----- r — • s-r" A* —■— .
0>J- r /•» — , Oz-
— — —" ft
r
.-------
whence
^ ( “7") “ ** (r ^

A funclion of Ihe form .— --- : is also harmonic for each lived

I r — rti |
r0 at every point whose radius vector r is distinct from r0. Conse­
quent ly, any linear combination of the form
n
'’Cl f'i
ZJ [r r,|
i -I
is a harmonic function (for r =^= ris i 1, 2 , . . n) which can be
interpreted as Ilie potential function of the held of gra vital ion (the
electrostatic field) produced by a system of material points (point
charges) located at the points with radius vectors r( (/ — 1 . 2 . . . .
. . ., n). The lieid tif gravitation produced by a mass continuously
distributed in space with volume density p. (x, //, z) can he obtained
by performing the passage to the limit in the above expression from
the system of mass points to a continuous mass distribut ion. The rigo­
rous mathematical justification of such a passage to the limit, which
looks quite natural from the point of view of physics, would involve
integrals dependent on parameters (see Chapter 10). We shall not
present such a proof here because these questions fall outside the
framework of our course and an* considered in the textbooks on
matheniatieaI physics.
Il xerchf ea
I. Write down Ihe expressions for tin* potential function of Ihe
held of gravitation produced by a mass continuously distributed
in spare with density u (x. //. 2 ).
 err. n. rn:u> tiieouv 267

2 . Tind ihe potential of the electrostatic field produced by an

infinite thread carrying a uniformly distributed charge. Is this
potential a harmonic function?

$ 7. EXPRESSING El ELD OPERATIONS

IN (It'll VI 1,1 NKAIt ORTHOGONAL COORDINATES
1. Statement of the Problem. Such quantities as the gradient,
divergence, rotation etc. of a field me widely applied in various
problems of theoretical and mathematical physics. In many cases
it appears necessary to express these quantities not only in ('.artesian
coordinates, as has been done in 3-6. but also in curvilinear
coordinate systems. For instance, if we deal with a spherically sym­
metric field, i.e. if. at every point in space, there is a scalar or vector
quantity dependent only on the distance from Ihe point to the
origin of coordinates, it is clear that all the formulas related to
such a field arc essentially simplified if we write them in spherical
coordinates instead of Cartesian ones. Of course, in various pro­
blems some other coordinate systems may prove to he more con­
venient.
Ifere wc shall derive the formulas for the gradient, divergence,
rotation and Laplace’s operator in general curvilinear orthogonal
coordinates and also consider Ihe special cases of spherical and cylin­
drical coordinates.
2. Curvilinear Orthogonal Coordinates in Space. Suppose we are
given a curvilinear coordinate system* r/2. q$ in space and let
the formulas connecting 7 ,. 7 * and q 3 with the Cartesian coordinates
x, if and z be of the form

x = x (7 ,. rj.,, 7 ,). if = !/ <7 i- 7-j- 7 a)* z ~ z (7 ,. 7 .>) (6.33)

We shall confine ourselves to the simplest case of orthogonal

coordinate systems which are particularly important for practical
applications. A curvilinear coordinate system is said to be ortho­
gonal if at each point the three coordinate curves passing through
the point form right angles with each other, i.e. the tangents to

* T ! u ‘ n o t i o n o f c u r v i l i n e a r c o n r d i m i l e s w a s d i s c u s s e d in C h a p t e r 2. W e
s h a l l s u p p o s e ilia ! t h e f u n c t i o n s e x p r e s s i n g c u r v i l i n e a r c o o r d i n a t e s in t e n n s
o f ( ' a r t e s i a n o n e s s a t i s f y I lie c o n d i t i o n s e n u 111 era ted in § A o f C h a p t e r 2. W e
s h a l l c h a n g e t h e n o t a t i o n i n t r o d u c e d in C h a p t e r 2 a n d d e s i g n a t e I h e c u r v i l i n e a r
c o o r d i n a t e s b y t h e s i n g l e l e t t e r r/ wi t h i h e s u b s c r i p t s I- 2 and 3 so Ilia! 111 0
f o r m u l a s tjike s y m m e t r i c f or m . S i m i l a r l y , for t l i e s a m e r e a s o n , t h e <*oni-
po iie nl> of a v e c t o r h e l d A (vvilli resp ect In I he c o o r d i n a t e s y s t e m in q u es t ion)
wi l l h e l:iln'lli»(l h v t . 1 ~ and 1
 >lt*l T I P L l i I NTF.CHAI. S. FIK1.D TI1 K 0K Y AND SFTIIF.S

the curves iiri1 mutually orthogonal at the point. In particular, the

spherical and the cylindrical coordinate systems in space which are
widely used in various problems also possess the orthogonality
property.
First of all. let us lind the expressions for the elements of length,
area and volume in orthogonal coordinates. For this purpose we
consider an infinitesimal curvilinear parallelepiped cut out of space
by throe pairs of coordinate surfaces corresponding, respectively, to
the values q ,. 7 , -1- dq,. <y2. r/ 2 ~ d q z am^ (h~ <7.** 'F d q d llu* coor­
dinates 7 ,. q z and q:i (see Fig. 0.13).
Consider the edge M M , of the parallelepiped. The point M lias
tlie curvilinear coordinates (7 ,. q z. q ?) and the point .1/j has the
coordinates (7 , d q lA q 2. 7 :)- Denoting the Cartesian coordinates
of the point .1/ as x. y. z and those of the point M , as x - - d x . y 4 d y y
z dz wo can write the expression
Y d x ' 1 rr d y 2 -{ dz*
for iho length d l, of the vector The coordinates x , y and z
are functions of a single parameter 7 , along the edge M. l f , since
7 2 and 7 ., are constant on . i f . If , . Consequently, in this case we have

dx — d q,. dy — dq,. dz — dqt

Off, '* Off, Oq, 1
and

" • -I 'U r - { % y + w * »

Similarly, for the lengths d h and r//n of the edges M M Z and

M we derive the expressions
r‘

a 1m1

" '•-I '' K ) =J ( ;):+ 5 (^ )

Introducing the notation

it - 1 (- z \ 2 (r>.34)
" r \ Oq* / 1 \ dq* ) ' I Oq* /

"»= \/ r i - Z J : (H)2!
we can rewrite the formulas for dl,, dl2 and d l z as follows:
,11 — fi./-is>.
k•
Al —? II- An„ A l* ~ // o da •» (ti .3,r>)
 CH. 0 FIKLD THKORY 200

The quantities II x, If* and / / 3 are referred to as scale factors

(Lame's* coeflicienU) associated with the curvilinear coordinates
qu f/ 2 * 7 3 ^ coordinate curve along which only one of the parameters
7 it <]z or <73 varies can be thought of as a curve on which the scale of
the corresponding parameter qt (i =* 1 . 2 or 3) is marked. The
multiplication by the factors ff ,t If* and / / 3 transforms infinitesimal
increments of the parameters qi, q2 and q3 into the corresponding
increments of the parameters l 2 and l 3 which are the arc lengths
of the coordinate curves.
By the hypothesis, the coordinate system being orthogonal, the
area dat of the face M 3 of the parallelepiped is equal to the
product of dl* by dl3. i.e.
do, — If* II3 dtj* dr/ 3
and similarly the areas do* and do3 of the other two faces are expres­
sed as
doz — U 3 IIx dq3 dqx and do3 — //, ff2 dqt dqz (0.36)
Finally, the volume of the infinitesimal parallelepiped is equal to
dv — dlx dl2 dl3 — f f x If* H:i £/r/, dqz dq3 (0.37)
Let us introduce, at each point A/, the orthonormal basis (i.e.
one whose base vectors are mutually orthogonal and have unit

lengths) determined by the three unit vectors e,. e 2. e 3 tangent tc

the coordinate curves passing through the point M. It should lx
noted that, unlike a Cartesian coordinate system specified by thret
constant (both in direction and magnitude) unit vectors i, j and k
the coordinate system determined by the basis e t, e 2, e 3 may vary
from point to point in the general case, that is the vectors e,. e.
and c 3 arc functions of the coordinates qt. 7 * and q 3. But this doe*
not prevent us from resolving an arbitrary vector given at any
point M as a linear combination of the corresponding unit vector.4
*q, e 2 and e3. Hence, every vector field can be (locally) resolved

* bam**. Cljibriel 11705 1870b a French <5rfenl»«i

L!70 -MULTIPLE INTEGRALS. FIEL D THEORY AND SEMES

into components along the directions of the vectors c»|. e 2 and e-,
at each point I/.
3. Cylindrical and Spherical Coordinates. Let us compute the
scale factors for the cylindrical and spherical coordinate systems
which arc the most important special cases of curvilinear orthogonal
coordinates.
Cylindrical coordinates r, <j, z are connected with Cartesian
coordinates x, y, z hv the formulas
x — r cos <p, y = r sin q
Applying formulas (6 .34) we obtain

"■~J '

( ^ ) ’ + ( ^ ) + ( ^ ? ) =
(6.38)
— Y f i s i n 31 «P ; r - COS2 q: = /•
rm
"» M S M - s r + r e r - i " - 1

Taking into account the geometric significance of the scale factors

//,, //2, Us we can directly derive formulas (6.38) without differen­
tiation. Indeed, consider an infinitesimal parallelepiped hounded
by the* three pairs of the coordinate surfaces corresponding to the
values r and r dr, q and cf j- dtp, z and z -f- dz of the cylindrical

n
efz
A

Fig. fi.14
0
coordinates r, rp, z (Fig. 6.14). The lengths d/lT dl2 and d/ 3 of the
edges AH, AC and AD of the parallelepiped are respectively equal
to dr, r dip and dz, which immediately implies formulas (6.3S).
Similarly, for the spherical coordinates determined by the relations
x = p cos cp sin 0 , y — p sin (p sin 0 , z = p cos 6
we find, by differentiating the formulas, the equalities
//, = 1, IIt = pt / / 3 = p sin 0 (6.311)
This result can again he obtained directly from Fig. 0.1 f> since the
lengths d/j. dl2 and d/ 3 of the edges AIJ, AC and AD of the parallele­
piped hounded by the coordinate surfaces specified by the value?
p and o • dp, q and cp dq, t) and 0 ' d 0 of the spherical coor-
 CH. 6. FI ELD TH EO RY 271

dinates p, <p. are respectively equal to

0

dp, p dO and p sin 0 dif

whence formulas (6.39) immediately follow.
4. Gradient. Let us find the expression of the gradient in curvi­
linear orthogonal coordinates. As is known, the projection of the
gradient of a scalar function U — U (c/t, q-z, on an arbitrary

axis coincides with the derivative of U along the direction of the

axis. Consequently, to compute the components of the vector grad (■
with respect to the basis iq, e 2, 1*3 wo must find the derivatives
of U along the directions specified by these vectors. Let AU be the
difference between the values of the function V at the points M t
and M. Then
M \r 1 ov
(grad (I, Cj) — 1iin <iix — Urn t l cA/1 //| »'</(
ilt j—
.0 <X?j —1»

Similarly, the other two components of the gradient are equal to

— —
/ / j 0 r/2
and — —
//.»(Jq:t

Jlcnce, we finally obtain

fW 1 uu 1 riu
grad
b
U — .. —
/ / 1 Oqi
Cj -j- “77—j,— 1*2 H-
1 1 / / 2 d q . c ' II3 dq3
5. Divergence. We now proceed to calculate the divergence of
a vector field A in coordinates 7 1 , q2, 7 3 . The expression of div A
at a point M was defined in § 2 by the formula
div A = lim 1 f 1 An do
q-+m v tul J J

Consequently, we can compute div A at an arbitrary point M

by evaluating the ratio of the flux of the vector A through the surface
of an infinitesimal parallelepiped shown in Fig. (>.13 to the volume
dv of the parallelepiped. Let A 1, A 2 and A3 be the coin po non Is
(coordinates, or projections) of the vector A in the basis e 1? e 3,
i.o. let
A = /IjCj ~r AoV2 I
To begin with, we find the Dux of the vector V through the two
faces perpendicular to the edge .1 / 1 /,. The outer normal lo the face
272 MULTIPLE INTEGRALS, FIELD TH E O R Y AND SERIES

M M iAF3 coincides with the vector —<*| hecause tlie vector ej

goes in the direction of tlie increase ol’ the coordinate q , whereas
the outer normal to this face is in the opposite direction. Hence,
the flux of the vector A through this face is given, to within infini­
tesimals of order higher than ihc first relative to the quantity dq2dq^,
by the expression
(A, —o j d(Tt = —A iFF2H3 dq2 dq3 (G.41)
where the values of . 1 1? II2 and fl:i are taken at the point (</,, qz. qs).
Tlie face M XA \ N N Z, opposite lo the face M M \ M differs from
the latter in the value of the coordinate qx which is now equal to
</i - d(ji instead of qt. Consequently, the quantity A x/ / 2/ / 3 takes,
on the face -l/^V^VA'.,, the value which differs from that on the face
M M ZN : M :i by the increment

^ ( . 4 , / / 2/ / 3)rf7.
Furthermore, tlie outer normal to the face ^/j/V3 A'*A‘7 is in the direc­
tion of the vector iq. Therefore the flux of the vector A through the
face .I/VV3ACV2 is equal to
| AUzlI:i~--— (/l,//2/ / 3) dqxJ dq2dqs (0.42)
Adding together expressions (0.41) and (G.42) we find that the flux
of the vector field A through the two opposite faces A1M2J\;XM2
ami U,AvVA'c is equal to
aq ( Ai Uvl I Jd q i d q z d q z
the last expression being correct lo within iiifinilesimaIs of higher
order relative to dv = l i xH2Hz dqx dq2 dq3.
Similarly, taking the other two pairs of opposite faces we find
the following expressions of the flux:
i) (. 4 «*y/ *|// j ) 7 j j . ^ (y ij/y 1//2) i j 1
----^ — - ikh (I(h dq* and ^ — dclx dq, dq:t
Adding iq> the three quantities thus found, dividing the sum by
dv and passing to the limit as Q —►AI we finally obtain the formula
1: v 1 f O(Alu . j i :t) . o (A2h 2if x) . o u » i i A ir2)
(6 / 1 0 )
~ dtjj
i i xi i J ! 2 I ’ fjqx Otfj
0. Rotation. As was shown in § 4, the projection (rot. A)n of
the rotation of a vector field A on the axis specified by an arbitrary
fixed vector n at a point M is expressed by the formula
J Ax cU
(rot A)n - lim L a
 CH. C. FIELD THEORY 27

where 2 is a surface perpendicular to the vector n at the point A/, c

being the area and L the boundary of 2 . Thus, wo can find the pro­
jection of rot A on the axis of Ihe vector e x by computing the ratio
of the circulation of A over the contour M M2N (shown in
Fig. 6.13) to the area a,. Let us present the circulation as the sum
of the four terms corresponding to the line segments MAf2. i\/ 2 iVt,
;ViA/3, A/3Af and compute each summand separately. We first
take The projection of the vector A on the line segment MM*
is equal to A 2 and consequently Ilie line integral of the projection
along the path M M 2 is given, to within infinitesimals of order oi
smallness higher than the first relative to the length of M M 2, by the
expression
dj = d j // *) d q 2 (6.44)
where the values of Ihe quantities A2 and / / 2 arc taken at the point
{(/;, fo)- The integral over ArxM3 differs from the above expres­
sion in the third coordinate, which assumes on Arxd/ 3 the value
(h d(h (instead of the value q$ taken on M M 2), and in the direc­
tion of integration since the direction of the line segment N XM.
is opposite to that of the vector e2. Therefore the integral of the
projection of A on A o v e r the path N i.\ls is equal to
— £ A2II2 4- UhIIi) J dq2 (6 •45)
Similarly, for the integrals taken along A/2 ;Vj and M 3M we obtain
the corresponding expressions
A'jIJa -t- ^ {A:iU 3) dq2 | dq3 (6.46)
and
A 3II3 dq3 (6.47)
Adding together expressions (0.44), (6 45), (6.46), and (6.17) \vc
find that the circulation of the vector A over the contour M M2N xj\I:lM
is equal to
0(A2It2) dq2 dq3-{■ d(dr,//3)
iiq.. dqi dq3
the last expression being correct to within infinitesimals of higher
order relative to dq2 dq3. Dividing this expression by II2H3 dq2 dq-,,
i.e. by the area of the face MM2N XM 3. and passing to the limit as
2 -v M we see that the component (rot A), of the vector rot A
in Ihe direction of the base vector e* is equal to
1 f (i (A3l f 3) _ d { A z f l 2)1
(0.48,)
^2^3 I ^ 2 /
The other two components are computed analogously:
(rot A) 2 =
1 ( W it h ) 0 ( A 3U 3) 1 ( 6.4 8 2)
1 I <*<h I
18 —0824
274 MLLTIIM.E INTEGRALS. FIELD THEORY AXI> S E R I E S

and
1 / d(A2u 2) _ d ( .y / t) ^
(rot A) 3 = //,/v 2 («.48a)
I Oq, <?72J
Finally wo can write the resultant expression of the rotation in cur­
vilinear orthogonal coordinates qu <y2, <?3*
^ l f < ? W / 3) 0 { A 2I1 2) ' \ _______l <7 (.4 3 / / :,) 1
yy2/ / 3 \ oq2 o<f9 J 1 ' * ~ - / / 3/ y 1 \ t/f/3 t/</i J e2-r
, _i _ f d(>l2//a) <> „
"r y /,y y , \ dqi oq2 J
7. Laplace’s Operator. Based on the expressions of grad i/
anil divA , we can write the following formula for Laplace’s
operator in curvilinear orthogonal coordinates qlt q2, q$:
t / y , y / 3 ou
SU -=- div grad U —
HiH2^7a dq\ ( y/, t)qx ) +
r. {I. j'/i
£_/ U3 Ui OU \ , 0 ( ___ 0(; ) 1 (ti AO)
dq
*72 V M'l <jq2 ) n Oqj H-a °<h / J
To memorize this formula note that the scale factors // ,, / / 2 and
/ / 3 entering into the denominators of the expressions written in
dU
front of the derivatives — (i — 1 , 2, 3) are due to the gradient,
°qi
the factors II2lf^ / / 3/ / t and / / , / / 2 in the numerators are geimratcd
by the expressions of the areas of the faces (which have been taken
\
when computing the fluxes) and the factor -rr-n— // j// 2 *n-
' 3 has appeared
when we have divided the resultant flux across the faces by the
volume of the infinitesimal parallelepiped.
8 . Basic Field Operations in Cylindrical and Spherical Coordinates.
In Sec. ?> wo found the scale factors for the cylindrical and spherical
coordinates. To write down the formulas expressing the gradient,
divergence, rotation and Laplacian operator iu these coordinates
wc must only substitute the corresponding scale factors into the
general formulas obtained in Secs. 4-7. This yields the following
results:
(a) In cylindrical coordinates we have
fl —
grad, t'== ou e ^. Tt — ou e< dU
—Oz er.
a;,, \ 1 0 ( rAr ) 1 OA fp OA.
 CH. 0. FIELD THEORY 2 /•)

where er, Oy and e 2 designate, respectively, the vectors cu e 2 and e 3

corresponding to the cylindrical coordinates /*, <p, z and A rt Av
and A t are the projections of A on cy, e<p and e 2.
(b) In spherical coordinates we obtain
. TT dU , l dU , t dU
grad
® U = -s— eD
dp p
A------
p 00 c 0 H----------:—rr ~r~
psmO v
1 1 0 (sin 0 -*lo) 1 OA<p
div A = p2 dp ' psinU m p sin 0 dtp 1
1 f d i A v si n 0) dA0 t dAp [ dfrAq)
rot A — \ e A- I ) e0 +
p sin 0 \ 00 dtp I <* r Vp sin 0i dtp p dp
d(P-4o) (1 cUp ,
1
T Ip
dp ~~ "p 0 0 .] 1<p
d I . dU I d-U
A£/ = 1 d / 2 < ^ \ ' p-sinO
1
ttt { sin 0
dtt \ ‘sin2 0 dip*
92 #9 ( ^ - * 1
where ep, e<>, ev denote the base vectors of the spherical coordinate
system and Apy A0, A <p are the corresponding projections of A.
In some problems related to the Laplacian operator, besides the
whole expression of AU in spherical coordinates, we encounter its
part of the form
1 6 i . t t dW
p2sinU 00 \ Sln 00 ) 1 p^sin‘ 0 dtp­
involving only the derivatives with respect to the “angular” variables
0 and tp.

Note. In § 7 we systematically used such terms as “an infinitesimal

parallelepiped”, “an clement of volume” and the like. It is clear that
here, as in many oilier similar cases, Lhe real sense of this termi­
nology is that we first take the corresponding geometrical and phy­
sical objects of finite sizes and then perform the passage to tin? limit
making the sizes tend to zero. We believe that the reader can easily
perform all the passages to the limit we are speaking about and
which have riot been written in full in the foregoing sections.

§ 8 . VARIABLE FIELDS IN CONTIMOUS MEDIA

So far, in studying various fields we. investigated lhe dependence
of the corresponding quantities (scalar or vector) on the spatial coor­
dinates. Here we are going to consider some questions related to
the dependence of fields on time.
1. Partial and Total Time Derivatives. Let us consider a fluid
motion whose velocity depends not only on the coordinates at each
point but also on time. Suppose there is a variable quantity tp related
to a fluid flow, for instance, temperature, pressure etc. The variation
18*
27G MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

of the quantity can be studied by means of two different approaches:

we can investigate its variation at a given point, i.e. for some fixed
values of x. y and z. or consider the values of the quantity connected
with a certain fluid particle whose coordinates are time-variable.
For example, if we deal with the temperature of a flow of fluid we
can measure it with a thermometer fixed at a certain point or with
one going downstream with the fluid.
The variation in time of a quantity cp (M , t) at a fixed point If
is characterized by the partial derivative (also termed explicit partial
deriva five)
0<f> _ | . <p(A/, 14 - A/) — <p(A/, 0
(G.50)
<" a'"o A/
expressing the local rate of change of <p, the coordinates x. y ami z
of the point M being regarded as constant parameters when compu­
ting (6.50).
The time-variation of the quantity <p (Af, /) connected with
a certain particle of fluid is specified by the total partial derivative
{particle derivative) of cp (Af, t) with respect to t which is defined
as follows.
Let a particle be at a point M at moment t and at a point M'
at moment t -r AL The total partial derivative of <p with respect to f
is the limit
]j,n t 0 (('>.51)
d* Ai-»0 At
To establish the connection between the partial and toLal time
derivatives wc note that when computing the total derivative we
must consider the coordinates x, y and z of the point M to be func­
tions of t whose derivatives with respect to t are the components
of the velocity of the fluid flow at the point:
(lx dy
u r -- I^.Vl fit
—vt
Therefore we must differentiate cp = tp (xt y, z, t) as a composite
function of t, which results in
. *<f dq>
dt ~~ i)l Ox dz
that is
ticp dip
dt dt
~b (v, grad (p) (<».52)
The notions of a partial and a total derivative with respect to lime
can be analogously introduced for an arbitrary vector quantity
A (Af, t) connected with a moving fluid. These derivatives are deter­
mined by the formulas
aA A (Af, I '-AO —A (Af, 0 rr
 Cil. G. FIELD THEORY 277

a n cl
dA ,. A (M , t -i- A/) — A (M, t)
—77- -- lim —i— -------- t~------i----- -- (0.54)
<it At->0 ^

which are similar to formulas (6.50) and (6.51). The relationship

between the derivatives is given by the formula
d\ OA , OA . dA , OA
~sf ~ ~ur H z'Vx m r Vy " r n r v* (6.55)
which is obtained by differentiating A (x, y, z, t) as a composite
function of t. Equalities (6.52) and (6.55) can be conveniently re­
written in the form
& ~ & + (v.V)«P (G.5G)
and
dA 0A . ,
V) A (0.57)
dt
where the expression (v, V) is understood as the operator
d_
Vx Oy

i.e. as the (formal) scalar product of the velocity vector v by the

symbolic vector V.
The expressions (v, V) <p and (v, V) A entering into formulas
(6.56) and (6.57) arc convective terms because they are connected with
convection of tlie particles. They only appear when wc deal with
a moving medium.
As an example, let us consider the acceleration of the particles
of a moving iiquid. It can be obtained as the total (particle) deri­
vative of the velocity with respect to time. Taking advantage of
formula (6.56) we derive

£ « - § f |-<v - V )v ((i-58>
or, in the coordinate notation.
dvx dvx 0vx dvx
■
+ Vx + Vy 0ox T V?
dt ~ dt dx Oy dz ’
dVy _ dVy OVy OVy
-L Vz Oz
dt ~ at 1 Vx dx H" Vy Oy
T
*

do? dr? Do? doz 0vx

Vx dx + * v -\ V?
dt Ot Oy OZ

. Eulcrian Equations of Motion of Ideal Liquid, hot us apply

2
the notions of partial and total time derivatives to o b tain in g the
278 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

equations of motion of an ideal liquid. We shall consider the so-called

Fulcrian specification of fluid motion playing an important role
in hydrodynamics.
Take a volume Q bounded by a surface 2 lying inside a moving
liquid. An element of area do of the surface is acted upon by a force
produced by the pressure. This force is directed along the normal n
to do* and is equal lo —pn do. (Here n is the unit vector in the
direction of the outer normal to 2 and p is the pressure, a scalar
quantity.) The resultant force F acting upon the whole surface 2
is expressed in the form
F= — f f pn do ((i.59)
v
• »

where, as before, the integral of the vector

pn = p cos (n, j:) i -r p cos (n, y) j -I p cos (n, z) k
is understood as a vector whose components are
J ^ p cos (n, x) do, j ^ />cos(n, y) do and J j pcos(n, z)da (G.GO)
2

Surface integral (6 .5 9 ) can be reduced to the corresponding volume

integral over Q by applying the Ostrogradsky theorem to each com­
ponent ((>.(>0 ) of the integral (which, as has been mentioned, is a
vector). This yields
dp .
—-
Oi/ £!*•> —

- kj j j 7 T - H Io Brad P d '"

and consequently an element of volume dta of the liquid is subjected

to the force
—grad p d<o
On the other hand, if p (A/, t) is the mass density of the liquid
at an arbitrary point Af at moment t and w is the acceleration of
a particle passing through the point Af, the quantity wp (A/, /) di<>
is the product of the mass contained in the volume d(o by the accele­
ration and hence, hv Newton’s second law, we obtain the equality
vv p dia = —grad p d (*>

* We suppose that the liquid is ideal which means that its v isc os ity factor
is considered to he equal to ze ro . Therefore the force acting upon an infinitesimal
area place* 1 inside the liquid is only produced by pressure and is dire* -ted per
peiui iciu.ii i v. tu tin- .ii ca.
 err. C. FIELD TIlEOliV 279

i.c.
wp = —grad p (0.61)
Tliis very equation yields an Eulerian description of motion of
an ideal liquid.* Here \v is understood as the acceleration of a particle
dv
of liquid, that is as the total derivative w = of the velocity v
with respect to time t. Using the expressions for the components of
the acceleration found in Sec. 1 we can rewrite equation (6 .0 1 ) in
coordinate form:
Ol'-X Ol'x . _L 0 lx \ dix Op
+ Vx
ot i Ox Vz Vy dy Ox
dz .) p - ’

du„ Ov,j 0v,j OVu \ dp

i
Ot + Vx Ox r Vy “ h Vz dz )p= - Oy 1

dot 1I 0vz dvz —L.

0vz 1\ ^ dp
1" Ox + i Vz
dz t) p =-—
0t dx oy
3. Derivative with Respect to Time of an Integral Over a Fluid
Volume. Let us consider a volume H of a moving continuous medium.
We shall refer to £2 as a fluid volume if it constantly consists of the
same particles of the fluid. A fluid volume thus may move and change
its shape in the process of motion. Consider the integral

J= ( 6 . 02 )

of a scalar function q> (d/, f) taken over such a volume. We shall

compute the derivative of this integral with respect to time.
In performing the computation we must take into account that
the variation of integral (0.62) in time is specified by two factors,
namely by the variation of the integrand connected with the incre­
ment of time t and by the change of the shape and position of the
spatial domain Q the integral is taken over.**
If the volume Q were invariable the function cp would gain the
increment — dl in lime dt and integral (6.62) would acquire the

* The minus sign on the right-hand side of equation (0 . 0 1 ) has an obvious

physical significance, namely it indicates that the acceleration of every particle
of liquid is in the direction of decrease of the pressure, i.e. opposite to the
gradient of p.
** Expression (6.62) is the so-called i n t e g r a l d e p e n d e n t on a p a r a m e t e r , the
parameter being t. (In (6.152) both the integrand and the domain of integration
depend on /.) The fundamentals of the theory of integrals involving a para­
meter will he presented in Chapter 10. Here, without resorting to the general
theory, we shall only consider the question of finding the derivative of integral
(6 . 0 2 ) with respect to limn, which is important for various nhvstVal nnnlimiinn*.
2 S0 MULTIPLE IN T EG R A L S, F I E L D THEORY AND S E R I E S

increment

12

Now consider the increment of integral (6.G2) produced by the

variation of the domain Q. Let 2 designate the surface bounding
the volume Q at moment t. The variation of the volume Q during
lime interval from / to t -{- dt is obviously due to the fact that some
particles of the fluid flow into or out of the volume £2 through the
surface 2 . The volume of the fluid flowing out through an element
do of the surface 2 during time dt is equal to vn dt do where vn
is the projection of the velocity of the fluid on the outer normal
to do. This variation of the volume gives, to integral (6.62). the
increment
qvn dt do
The resultant increment of integral (6.62) due to the variation of
the volume Q in time dl is equal to
<pt>n do
£
Thus, the total increment of integral (6.62) during time period dt
is equal to
dJ —dt ^ j - <2<o-J- dl I' j q t’n do
*h
and. consequently, we have
dJ_
(6.62,)
•it

Transforming the second summand on the right-hand side by the

Oslrogradsky theorem w’e obtain

w if i
12 Q
4 f+ div^v)jdi,“
Finally, taking advantage of the equality
div (cf.iv) — cp div v + (v, grad cp)
(see formula ((>.21))) and applying expression ((5.52) For Ihc total
derivative we receive the resulting formula
_d div v ) c/(0 (6.65)
dl
h* li
In particular, if div v = 0 , i.e. if we consider a motion of an
incompressible liquid without sources and sinks, formula (6.65)
 CH. 6. FI EL D THEORY 28

takes a simpler form

~di

Note. The problem of differentiating an integral taken over s

fluid volume is analogous to the following one-dimensional problen
(which we shall again deal with in Chapter 1 0 ): given an integra
tKO
J (t) = f i|>(x, /) dx
a(t)

il is required to find the derivative of J(t) with respect to t. ltegnr

ding J (t) as a composite function of t (dependent on t because botl
the integrand and the limits of integration a(t) and b(t) involve t
we easily find that
b(<)
= ) In d x ^ <r(fc(o. /)&'(<) -<p(« ( 0 , 0 o' (0
o(f)
m
Here again J'(t) is a sum of two terms
a(0
(J ^ dx and (rp (b, /) // —

<j (a, O'2*)), the former being due to the variation of the integrant
and the latter to the change of the interval of integration.
We have studied the integral of a scalar function over a fluic
volume. Similar techniques can be applied to studying an integra
of a vector function \ (,1 A. t). The same arguments yield the followin'
expression for the derivative of such an integral with respect to t

Q Q
+-a vj^ ^

The above discussion concerns integration over a volume of fluid

But in hydrodynamics and some other divisions of physics we a 1st
deal with surfaces and curves formed of particles of fluid whicl
change their position in space and their shape in accordance witl.
the motion of the particles. Surface and line integrals o
functions over such fluid surfaces and curves are again expression.'
whose dependence on time is specified by two factors, i.e. by tlu
variation, in lime, of the domains of integration and by that of tin
integrands. Using arguments similar to those applied to investi
gating integrals (l>.0 2 ) and (b.6 6 ) we can easily establish the corrcs
ponding formulas for differentiating these surface and line integrals
wiLh respect l.o time.
282 MUI.T1IM.K INTEGRALS, FI EL D THEORY AND -SERIES

4. Application to Deriving Equation of Continuity. Formula (G.63)

immediately implies the equation of continuity obtained in § 3,
Sec. 5. Indeed, let p (M, t) be the density of a moving liquid (which
may be compressible in the general case). The mass of the liquid
contained in a volume Q is equal to
j j j p r f ‘0
Q
If Q is a fluid volume which is thought of as consisting of the same
particles of the fluid in the process of motion, the mass it contains
remains constant. Consequently, by formula (6.64), we have

4r 1f1d<0= I \ j [ I f + div <pv>] dM= 0

i> Q
The volume Q being taken quite arbitrarily, we thus obtain the
relation
- |- -rdiv (pv) —0
which is the equation of continuity.
 Tensors

The investigation of phenomena in natural and engineering sciences

involves various mathematical quantities. The distinction between
these quantities lies in their analytic expressions and the laws of
transformation of the expressions when the coordinate system is
changed.
From the point of view of mathematics, the simplest physical
quantities are scalars (such as the mass of a body, the volume of
a body, the length of a vector etc.) which arc invariant with respect
to the transformations of coordinates. Every scalar quantity is
characterized, in any coordinate system, by a single numerical
value independent of the choice of the coordinate system.
The vector quantities, e.g. velocity, acceleration, force etc.,
are of a more complicated mathematical nature. A vector quantity
is specified, in the three-dimensional space, by a triple of numbers
with respect to each coordinate basis, namely by the three projections
of the vector on the coordinate axes,which are also referred to as the
coordinates (or components) of the vector. When a coordinate basis
is replaced by another one the components of a vector are transformed
according to a special law.
Still more complicated quantities, from the point of view of the
laws of their transformation, are the so-called tensors which, in
certain particular cases, are analogous to linear operators applied
to vectors (the notion of a linear operator is discussed in Sec. 1
of § 2 in the present chapter).
As an example of such a quantity, let us consider the so-called
conductivity tensor characterizing electric properties of an aniso­
tropic conductor. In an isotropic medium the current density vector
j and the electric intensity vector E are collinear. that is connected
by a relation
j=oE (7.1)
where or !> 0 is a scalar factor known as the specific conduct wily
of the medium. But in the general case of an anisotropic body the
vectors j and E are no longer collinear and therefore the factor o
should be regarded as a linear operator transforming the vector E
into the vector j. 'Phis operator is referred to as the conductivity
tensor.
2«4 MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

If we choose an arbitrary fixed basis et, e 2, e 3 in space and resolve

the vectors j and E with respect to the basis we can write
j = /iCi 4* / 2e 2 4- 73 C3 (7.2)
K = £ > , 4' ^2 ^2 4~ ^3*'3
and then replace relation (7.1) by the equivalent system of three
scalar equalities
3
jk — 5.' CkiEi. *-1,2,3 (7.1')
i-t
Thus, the conductivity tensor a is specified, in every basis <q, e 2, e3,
by the nine numbers oftf, k , i — 1 , 2, 3, termed the components
of the tensor a relative to this basis.
The definition of the notion of tensor should obviously include the
rule for transforming its components when the basis is changed.
In §§ 1-9 we shall only restrict ourselves to the orthonormal
bases (i.e. the ones consisting of mutually orthogonal unit vectors)
and their transformations connected with the notion of an orthogonal
affine tensor. In § 1 0 we shall briefly discuss the general tensors.

§ t. ORTHOGONAL AFFINE TENSOR

1. Transformation of Orthonormal Bases. Let us consider two
arbitrary orthonormal bases e,, e2, e 3 and ej, eL e' in a three-dimen­
sional Euclidean space. The bases being orthonormal, wo have the
following relations for Iheir scalar products:
f 0 for i k
(cq, c/ 0 = 5i/t, (el, ei) -S in , Si/, \ j fQr i==£ (7-3)

(fyft is called the Kroneeker* delta). We shall conditionally refer

to c*|, e 2, ea as the “old” basis and to ej, o', ejj as the “new” one.
Kesolving the base vectors of the new basis with respect to the
old basis we obtain the equalities
ej = -1- a 12 e 2 + a 13e3
e; = « 2 lC‘| 4" <*22C2 4- CC23<‘3 (7.4)
**i ~ 4~ ^32^*2 4* 0^33^3
or, in the contracted notation,

e '= V a,)Vh i — 1, 2, 3 (7.4')

i=i

* Kmn<*rk«*r. I.eopold (1823 1891), a German mathematician.

 Cll. 7. TE N SO R S 28.")

The matrix
« il «12 a 13
a 2i «22 a 23
a 3l a 32 a 33
is called the transformation matrix from the old basis e,, e2, e3
to the new basis e|, ej, e3.
Let us investigate the properties of this matrix. Multiplying
sealarly the vector ej = — ctf3 C3 (t = 1, 2, 3) by the
vector o) — a/je, -j- ccj2c 2 t a j 3e3 0 = 1 , 2 , 3) we receive
( 0 for j =7^=i
+ + j for . (7.(i)

Hence, the sum of the squares of the elements of every row of

the matrix || a tj || is equal to unity and the sum of the products of
the corresponding elements of two different rows is equal to zero.*
Next, computing the scalar products of expression (7.4') by vk
(/v = 1, 2, 3) we obtain**
(e;, c*jJ = <xitn i, k ^ 1 , 2, 3 (7.7)
We now proceed to find the expressions for the elements of the
inverse of matrix (7.5). For Ibis purpose we expand the base vectors
of the old basis cq, ts, e3 with respect tothe new basis and thus derive
C*1 = P llO j -T f>12c 8 T Pt3<*3

C2 ^ p 2 i*S I- P^ < ‘2 I P2 a«‘a (7.S)

C 3 — f’ 3ie i "I" P 32^*2 ’ P3:»**3
or, in the abbreviated notation,
a
Ch = 2 frjpj (7.8')
;=t
The matrix
P ll Pl2 Pl3
Pu 11 = P21 P22 P23 (7.9)
P3i P 32 P.13
is apparently the inverse of matrix (7.5). Performing scalar multi­
plication of expression (7.8') by c! (i = 1, 2, 3) we obtain
fa, ft*) = pftf, * , * = » ! , 2, 3 (7.10)

* A matrix |] a tj jj fur which relations (7.U) hold is called orthogonal

Thus, a transformation matrix from one ortlionormal basis to another is nn
orthogonal matrix.
** We obvionslv have n = (o'i. o- ' — v ’ cO f! 1 — 1 . *' 7 '
280 MULTIPLE INT Kf iH A LS , FIELD TI1EOFV AND SEU1ES

Comparing formulas (7.7) and (7.10) we find llie following rela­

tionship between the elements of matrices (7.5) and (7.9):
a , „ = p fci (i, fc = 1 , 2 , 3 ) (7.11)
Thus, matrix (7.9), the inverse of matrix (7.5). is the transpose
of matrix (7.5). It follows that the columns of an orthogonal matrix
possess the same properties as its rows (sec formula (7.6)) since the
inverse of matrix (7.5) is the transformation matrix from the ortho­
normal basis ej, c2, ej to the orlhonormnl basis e t. e2, o 3 and hence
it is an orthogonal matrix.
2. Definition of Orthogonal Affine Tensor. In constructing the
formal theory of tensors it appears expedient to include the invariant,
scalar quantities and vectors into the class of tensors. Thus, a scalar
quantity L invariant with respect to the transformations from one
orthonormal basis to another is called an orthogonal affine tensor
of rank zero.
Such quantities as temperature, mass, the length of a vector etc.
are examples of orthogonal affine tensors of rank zero. On the other
hand, although the projection of a vector on the ith (i — 1 , 2 , 3 )
coordinate axis (i.o. the axis determined by the ith base vector e*)
is specified by a single number in each basis e,, e2, e 3, it changes
in a certain manner when the basis is replaced and thus it is not
a tensor of rank zero.
To i nc lu d e the vectors into the class of tensors we define them as
tensors of rank {order) one.
I>efi n it ion /. Let a quantity L be determined. in every orthonormal
basis, by a triple of numbers, say by numbers L {, L2, L 3 in a basis
e ,, o3. by numbers L't. L\. L\ in a basis e^, o', ej etc. Suppose that
under the transformation from an arbitrary orthonvrmaL basis o,,
e2, «l;< to any other orthonormal basis ej. e2. e3 these numbers undergo
the transformation determined by the formulas
3
anU (i = 1, 2. 3) (7.12)
where || a Ul || is the transformation matrix from the basis c t. e2, e 3
to the basis e], e2, ej. Then the quantity L is said to bean o r t h o g o n a l
aff in e t e n s o r o f r a n k {o rd e r) o n e and is denoted by the symbol
(Li), i.e. L == (L^.
'fho nuinhers Lt, i - 1, 2, 3, are referred to as the components
(coordinates) of the tensor L, relative to the basis vt, e 2, 0 3 . the
numbers L\, i ~ 1. 2, 3, as the components in the basis ej, ej, e'
etc.
Let us prove llial every vector is an orthogonal affine tensor of
rank (order) one- Indeed, every vector x is determined by a triple
f*t niin»h**r< in each orthonormal basis e,, e2, e 3. these numbers being
 Oil. 7. TEN SORS :>s7
its coordinates relative to the basis, i.e. its projections on the cor­
responding coordinate axes. Besides, when the orthononnal basis
is changed the coordinates of the vector x are transformed according
to formulas of type (7.12). For, if we resolve x with respect to two
bases e f, e2, 0 3 and ej, e*, e '3 we obtain
x = x,e, — + *3<?3 — *je; + x'e' -f (7.13)
Multiplying scalarly equality (7.13) by vt\ (t = 1, 2 , 3) we lind,
with the help of formulas (7.3) and (7.7), that
3
x\ - a i7.x2 -\-(Xi2x 3 := V az/tXft,
4 i — 1, 2 . 3 (7.14)
ft— 1
Formulas (7.14) being of type (7.12), we thus see that every vector x
is in fact an orthogonal affine tensor of rank one.
Note 1 . It is obvious that, conversely, every orthogonal affine
tensor of rank one can be interpreted as a vector.
Note 2. Since the inverse of the matrix || a ^ || is its transpose,
relations (7.14) imply that

Xi =
a-jiXj. i = l,2, 3 (7-14')
j=\
Wo now proceed to define a tensor of second rank.
D e fin it i o n ‘2 . Let a quantity L be determined by nine numbers in
every orthanormal basis, these numbers being L iJy i. / — 1 , 2, 3, in a
basis c ,, e 2, c*, L\t , i, j = 1, 2 , 3. in another basis o|, e', e' etc.
Suppose that when an arbitrary orthonormal basis e,, o?. i>^ is trans­
formed to any other orthonormat. basis ej, cj, e' these numbers are
changed according to the formulas
3 3
L/ij — &ini,XinL/mn* j —1* 2, 3 (/.1 5)
77i— I n = l

where || a <y || is the transformation matrix from, the basis e,, o2, 0 3
to the basis ej, e', 0 3 . Then the quantity L is called an o rthof/onfti
ttffin e t e n s o r o f secon:f r a n k (orfter) and is denoted by the sym­
bol (//jj), i.e. L 7= (f/jj).
The numbers/yjy, i, f = I, 2, 3, arc referred to as the components
of the tensor L in the basis e,, c 2, e 3, the numbers L\j, i, y— 1, 2,3,
as its components in the basis e', e.^, e* etc.
In §§ 2-1) we shall dwell in more detail on the properties of the
orthogonal nfline tensors of second rank and on examples of such
tensors.
Now we give the definition of an orthogonal affine tensor of an
arbitrary rank (order) p 1.
288 MULTIPLE IN T EG R A L S, FIELD TH EO RY AND S E R I E S

D e f i n i t i o n ,7. Let us be given a quantity L which is specified by a

collection of 3P numbers Li 1>2 . . . j p , t, = 1, 2, 3, s = i , 2, . p
in every orthonormal basis Cj, e 2, e 3. If these numbers are transformed
according to the formulas
3^
Liii-2 . .. ip = . . .. i p 0 •^ 0
31)2 ... ;^=i
when an arbitrary basis Cj, c2, e3 is transformed to any other orthonor­
mal basis cj, ej, ej by means of a transformation matrix || |1, the
quantity L is called an o r t h o g o n a l a f f i n e t e n s o r o f r a n k p
and is denoted by the symbol {La i 2 . . . 4p), that is L — (Z«1*2
i . . . » jy)•
The numbers L i ^ t ^ - ’ i p are called I he components of the tensor L
relative lo the basis et, e 2, e 3, the numbers L \Vi 2 . . . *p are its com-
ponents in the basis cj, e', ej etc.
Note I. The definition of an orthogonal a lime tensor of order p
(p ^ J) can also be given in the following equivalent form.
We say that we are given an orthogonal affine tensor of rank
p ^ 1 (denoted as Ln \ . . . » ) if to every orthonormal basis clt e 2, e 3
there correspond 3P numbers L\^t...... { . i s = 1, 2, 3, s = 1, 2, . . .
. . p 7 which change in accord with the formulas
3
V
^ | I J . . . itt ^* 1 5 1 ^ 2 ^ 2 • • • 1*2 * * * ^ P ( ^ *i 0
502 . . . 5 ; . = i
when an arbitrary orlhonormal basis iq, c2, e3 is transformed to
another orlhonormal basis cj, cj, cj by means of a transformation
matrix || H.
We shall somelimes apply the latter form of definition to the
case p =* 1 and also to the case p = 2.
Note 2. Definitions 1, 2 and 3 have been given for a three-dimen­
sional space. Hut they can be rephrased in a completely analogous
manner for an A-dimcnsional space whose orlhonormal bases
I t l 2> c'v etc. contain Ar mutually orthogonal
unit vectors, the transformation from a basis ei% c 2» N to
another basis ej, e•»> being performed by the formulas
N
Cj — rx.tjCjy 1 = 1, 2, . . . , ./V
j~i
where t he transformation matrix || a i3* || is of order N.

$ 2. CONNECTION BETWEEN TENSORS OF SECOND HANK

AND LINEAR OPERATORS
1. Linear Operator as a Tensor of Second Hank. To begin with,
wL* remind thr* reader fhnl n linear operator ora linear vector function
 V.ll. 7. TKNSORS 2S‘J

is ft function
y = L (x)
which associates, with every vector x, a vector y in such a way that
the relation
L (C,x, T* C.»x2) — CjL (x,) -f* C 2 Li ( x 2) (7.18)
are fulfilled for any x, and x 2 and any constants Ct and C2.
Tile components (coordinates) of a linear operator L in a basis
Ci, e 2. t*3 nre the coefficients Li} appearing in the resolutions of the
images L (e,). L (e2) and L (e3) of the base vectors cj, c 2 and v :
relative to this basis:
L (e,) = t 7y2iC2 r ^ 3 i® 3

Ij (e2) ~ 4~ Tj 22^'2 4" 3 (7.10)

L (t'3) = X/13C1 7/2 3C2 H~ L 33C3
These formulas can be put down, in the abbreviated ^notation, as
3
r.(o ,)= V LujCi,, i -1 , 2, 3 (7.20)
*=i
Let us scalarly multiply both sides of equality (7.20) by the
vector v t (i — 1, 2 , 3). Then, based on relations (7.3), we obtain
L\i — (ei? L («/)), i, 7 — 1 , 2, 3 (7.21)
Similarly, for the components of the operaLor L in another basis
«ij, e', e ‘ we have
L\j = (c|, L (cj)), i, j = 1, 2, 3 (7.22)
Substituting the expressions
9
ei V (7.23)
771=1 n= 1
into formulas (7*22) we receive the relations
3 3
/4; = (c'i, M Ci) ) ^ ( ( S ( y fXjnL (e„)) =
m—1 n=1
3 3 3 3
= dim^Jn (**rrn lJ (®n)) == j
m —1 n —t rn— 1 n = l

i> i - 1, 2, 3 (7.24)
Formulas (7.24) coincidc’with formulas (7.15) ami consequently we
have proved that every linear operator L is an orthogonal affine
tensor of second rank.
19—0R24
290 m u l t ipl e in teg r a ls , fie l d theory and se r ie s

2. Tensor of Second Hank as a Linear Operator. An orthogonal

affine tensor (L^) of rank two can be interpreted as a linear operator
applied to the vectors of a Euclidean space. To show this we tako
an orthogonal affine tensor of second rank (L^) in a three-dimensional
space and define an operator y = L (x) by assuming that its appli­
cation to the base vectors e t, e2, e 3of an arbitrary orthonormal basis
is described by formulas (7.19) and that the expression L (x) (where
x = a:,e, x 2c 2 •+* 2 :3 0 3 is an arbitrary vector) is specified by the
formula
3

L(x) =* y} XjL(ei) (7.25)

i= l
Thus, we have associated, with every orthogonal tensor (Li;)
of second rank, a certain vector function (operator) y = L(x).
Now we arc going to prove that the operator thus defined is in fact
3

linear, i.e. satisfies condition (7.18). Let x and y =

t=i
3

— V] then we have
i= l
3

Ctx -hC2y V
/ i (CiZi + CayOei
i* t
Consequently, by virtue of definition (7.25), wc obtain (substituting
Ctx ~tC2y for x into (7.25)) the relation
3

L (6 'jX + C 2y) =* (CjXi -r C2I/O E (Ci) —

i— 1
3 3

-a Cj XjL (e*) -f- 6 2 t//L (ci) = C\L (x) -{-C2L (y)

i=l »=i
which coincides with (7.18). Hence, the linearity of the operator
has been proved.
It can he easily shown that the linear operator L defined above
by means of the tensor (Ltj) does not depend on the choice of the
basis Ci, e 2, 0 3 . In other words, if instead of the components L {j
of the tensor in Hie basis <?t, e2» e 3 we take its components L\j in
another basis o', o', and define a linear operator L' by means of
the relations

V (Ci) = TJ U,c;. (» = •!, 2, 3). V (x ). S x \U (ci) (7.25')

i=l
3
where x = xle\ wc shall have the identity
i= 1
U (x) = L (x) (7.2(>)
fui e«u Ii vector x.
 CH 7. TENSORS 291

Indeed, taking advantage of the relations

3 3 3 3
L \y~ 2 2
ft=l i = t
ctik'XjtLki, X i= 2
7 =1
etjix'j* ck = 2
«= 1
«i*ci
and formulas (7.19), (7.25) and (7.25') we obtain
3 3 3
L (X
)= 2i ^|L (ei) = i=
2i A
2=i L h i e h =
i=
= 2 xi 2 ( 2 2 a i k V y L t t i ) e't = 2 x) 2 L 'lrt —
3= 1 1 =1 h=l i= l 3 =1 /=1
3
= 2 (*}) = !/(* ) (7.27)
j= i
which is what we set out to prove.
Wo have thus shown that the operators 1/ and L coincide and
consequently relations (7.19) and (7.25) establish a one-to-one cor­
respondence between orthogonal affine tensors (L(j) of rank two
and the linear operators L associated with them. The linear operator
L can be identified with the corresponding tensor (L0 ), and hence
wc can consider an orthogonal affine tensor of second rank as being
a linear operator. This interpretation of an orthogonal affine tensor
of second rank is widely used in physics. Namely, in this manner
we interpret the conductivity tensor mentioned at the beginning of
the present chapter. The inertia tensor introduced in mechanics
and the stress tensor considered in the theory of elasticity (the latter
tensor will be treated in § 5*) are also understood in this way. But
there is another interpretation of a tensor of second rank which proves
to be useful in various applications. This new approach to the notiou
of tensor is discussed in § 3.

§ 3. CONNECTION BETWEEN TENSORS

AND INVARIANT MULTILINEAR FORMS
I. Tensors of Bank One and Invariant Linear Tonus. Let us bo
given, in every coordinate system, a triple of numbers a,,
a2, a 3. Suppose that when one coordinate system is transformed
* The reader will see in § 5 that the interpretation of the stress tensor is
connected with the notion of the adjoint operator . If, instead of relations (7.20)
and (7.25) specifying the linear operator L corresponding to a tensor (Lu ),
3 '
we take the formulas L* (e,) = 2 ^jkch i / = 1, 2 , 3, and L* (x) =
ft=»t
.1 3
sb ^ jT
jL* (Cj) ( x = V ^iCj) we arrive at the linear operator I>* which is the
i=i i=i
adjoint of L.
19*
292 .MLM.1IPLE INTEGRALS, FIELD T1IKOHY AN1» S E R I E S

to anotlier these numbers change in such a way that the linear

form aAx x — a.>x2 - • a^x3 (where xl% x2 and x 3 are the coordinates
of an arbitrary vector x) remains invariant. Then it turns out that
the quantities (i — 1 , 2 . 3) form a tensor of rank one. Indeed,
let an arbitrary vector x be resolved as x = x,ej x 2c2 -\~ x&a
relative to a basis et, e 2. c*3 in which the coefficients of a given linear
form are a a « . a3. and let the same vector x be expressed in tlie
form x = x|t*' 4- ;r'<*2 -{- jrjOj in an<ither basis ej. eL c, for which
the coefficients of the linear form are a\. a2 and a'. The linear form
being regarded as invariant, wc thus have the equality
a \x \ ~ a '.z2 a.jz'3 — a \X\ ~r -i 03*3(7 .28 )
for every vector x. Let us substitute the expression of x h in terms
3

of x'i (i.e. x h — V] ttihXi) into the right-hand side of equality (7.28).

i*=i
This results in
3 3 3 3 .1
^ aix'i = S (ih ]£ a,-*, jrt: V ( 2 OLihah) x'i
t= l ft—i i*=l i= l k= 1

The quantities x\%x’2 and x'3 being quite arbitrary here, we can write
the relation
3
= £ otii.aft a\ (7.29)
/tr-1
which is what we set out to prove.
2. Tensors of Hank Two and Invariant Bilinear Forms. We can
similarly prove that, the coefficients of an invariant bilinear Form

H aijxnjj (7.30)
U ;=1
(where and i = 1 , 2 . 3 , are, respectively, the coordinates of
variable vectors x and y) constitute a tensor of second rank. In
fact, let the bilinear form be expressed as (7.30) in a basis e2, e 3
and as
3
tiijXi i)i (7.31)
L 3 -1
in another basis e], o', o'. We suppose the form to be invariant, and
consequently
3 3

a\iX\y)~ y amnxmijn (7.32)

it I in . 7i— l

for any two vectors x and y. Substituting the expressions

Xni ^ U n. — CCj/tl/j (ttl, ft - 1, 2, 3) (7.33)

* - I
 CH TKNSORS 313

of the old coordinates of the vectors x and y (i.c. their coordinates

relative to the basis ct, o,, e*) in terms of the new coordinates (in
the basis o', e^, oj) into the right-hand side of equality (7.32) we
deduce, the relation
3 3 3 3

i.
5
1
S
vi, n —1
a ( V a .m X l)
t= l ;=* i
3 3
= & - i i n & j n @ n i n ) *^"i l f j (7.34;
i, j — 1 m , rv=l

The arbitrariness of x\ andjf/t (i —■1, 2, 3) suggests that

3
^ (/.OJ
m, n—1
which is what we set out to prove.
Note /. Equality (7.35) can be proved by considering only vector
of unit length. Indeed, putting
1 F«r i i0 . .
f 1 for / = ; 0
n r • _/ *
() for i r^r- i0 1 - 1 * 2 ,3 , \fO for } ~ j .o )■= 1, 2, c
yj
(7.3G
we deiive from equality (7.34) the relation
3
/ V
^t*j0“ y— uinUjoiiCLmn (*o* /o ~ 1* 2, 3)
m. m~ !
Eui llierinore, by equality (7.30), the vectors
*> ?
\i . . i xi ' '
x = 2_ i» ci y - 2j y&j
i-i i=1
are of uni I length because the basis oj, e,, ej is ortlionorinal (w
have agreed to restrict ourselves to such bases). Consequently,
a bilinear form is invariant on unit sphere, that is on conditio
that its values are considered only for the vectors of unit lengtl
its coefficients a:j (t, ; = 1 , 2 , 3) constitute an orthogonal aflii
tensor of second rank.
A bilinear form is said to he symmetric if its coefficient main
is symmetric, that is if a fJ aJit i, j — 1 , 2 , 3. (By virtue •
relation (7.35), we can assert that if <i-ti ajt at least in one orlh*
normal basis the matrix [| aif |j remains symmetric for any olh-
orlhonormal basis.) Putting y — x in a symmetric bilinear for
we obtain tin* so-called quadratic form
.14
(i fjj iif j ( i ••1
i. f—1
294 M U L T IP L E IN T E G R A L S , F IE L D THEO RY AND S E R IE S

A symmetric bilinear form is uniquely specified by the quadratic

form generated by it when we put x — y. Actually, if we substitute
the coordinates of the vector x + y for those of an arbitrary vector x
(where y is also an arbitrary vector) into formula (7.37) we obtain
3 3 3
2
;= i
a i i (x i -i-V i ) {x i 4-V i ) — 2t. i=t
a i j X i X j -J
- 2t,j=i
a *tytys +

3 3
-h 2 *ijXi!/j-r 2 *tjyi*j =
i, ;= 1 i, i=*l
3 3 3
= 2 &iiXiyj+ 2 <*ijyiyj + 2 2 a tw yj (7.38)
I. i=l i, j=l i, j=l
since atj = a}i. Consequently, we receive the equality
3

2 =
i. i*»l
3 3 3

= t { 2 au ( xi + yi)(xJ - r y } ) ~ 2 aHx ixi — 2 (7-39>

i, i —l 1, 7 = 1 :.i=l
which is what we set out to prove.
It follows that the coefficients of an invariant quadratic form
constitute an orthogonal affine tensor of second rank. Indeed, they
coincide with the ^coefficients of the corresponding invariant sym­
metric bilinear form for which we have already proved that the
collection of its coefficients is an orthogonal affine tensor of second
rank.
Note 2. On the basis of Note 1 we conclude that the coefficients
3
aU (a u = ajt\ L ) = 2, 3) of a quadratic form 2 aux ixj
i, i=t
which is defined and invariant on unit sphere constitute an orthogo­
nal affine tensor of second rank.
3. Tensors of Arbitrary Rank p and Invariant Multilinear Forms.
Let vectors %2* • • •» £/> he resolved with respect to a basis
®1» ^3*
I 7 = ijl^t + l> 2e 2 + £/3e3» 7 = 2, • P
Suppose that, for every basis eit e 2, e 3, there is a system of coeffi­
cients (where i s = 1, 2, 3, s = 1, 2, . . p ). Then
the function
 CH. 7. T E N S O R S 295

is said to be a multilinear form. As in the case of an invariant bili­

near form, it can be easily proved that the collection of the coeffi­
cients of an invariant multilinear form of an arbitrary order p ^ 1
is an orthogonal affine tensor of rank p.

§ 4. STRAIN TENSOR
Consider a deformable physical body whose arbitrary point is
specified by its position vector (radius vector) r = x ^ 4 ~ x 2e 2 4 -
4* x 3 e 3 in a coordinate system Oxtx^c3. If the radius vector of a
point M is equal to r we have OM = r. In this case we shall write
M(r).
Suppose the body is subjected to a deformation in which a point
M (r) is displaced by a vector u, that is passes to the new position
M ’( t -{- u) (see^Fig. 7.4). The deformation is specified by the field

of displaci'iiiciils u ~ UjVf ~r Let us consider a point

Afi (r — dr) lying close to the point M (r). After the body has
been deformed this point occupies the position r -|- dr -j- u -f-
4* <2 u). The deformation of the body in the vicinity of the point
M(r)j can be characterized hv the variations of the lengths of all
the line segments M M t (/ = 1, 2, 3, . . .) starting from this point,
their end points M i% M 2, . . . lying in a sufficiently small neigh­
bourhood of the point M {r).
Let us investigate the variation of the length of the line segment
M M Xdue to the deformation of the body. The length of the segment
AfMi in its original position is equal to [drj. The segment will
occupy, after the deformation, the position of the line segment
M'M\ whose length is equal to \ dr -f- duj. As a measure of change
of the length of the segment AfAfj, we^shall take the quantity
4 - { J T l r ' — AiM] } = i . {(dr + du)*— dr*} = — {2 du dr + du*} =
= 7*1*1 4- v*2*a 4 7*3*3 4~ 27*1x9 dxx dx2 4 ~
■
j- 2 yX |X 3 dx\ dx 3 -j- 2 yX sX 3 dx 2 dx 3
29ii MU LTI PLE INTEGRALS, FIELD TH EO RY AMD S E R I E S

where YxtXj ( l 7 = 1, 2, 3) are some coefficients and dx, (i =

= 1 ,2 , 3) are the components of dr. This expression is a quadratic
form in the variables dxt, dx2 and dx3. The construction of this
form implies that it is invariant. Consequently, its coefficients con­
stitute a tensor of rank two which is characterized by the matrix
Y * tX, Y xjx2 Y * |X 3

V *2*l Y x 2x 2 Y x 2X3

Y x ;ix i Y*3X2 Y xc X3

The tensor thus formed is called the strain tensor.

Suppose that the deformation is so small that the squares and
the products of the derivalives of u2 and z/ 3 with respect to x,,
xL and 1 3 are negligibly small compared with the first-order terms.
Then the matrix of the strain tensor can be written in the form

(7.41)

§ 5. STRESS TENSOR
1. Definition of Stress Tensor. Suppose we have an elastic body
which has been deformed. Let us mentally draw an elementary plane
surface of area cr through a point \ f of the body and erect at. M
a unit normal vector n to one of the two sides of the surface (see
Fig. 7.2). If we divide the resultant clastic force Fna (applied to the
chosen side of the surface element) by the area a we obtain the
s o - c a l l e d average (mean) stress (n-l_- — —^ o n t h e e l e m e n ta r y * a r e a
cr with normal 11 drawn through the point A/. Passing to the limit
as a is contracted toward the point M we arrive at the {total) stress
pA at the point M on an elementary area with normal n:
,,„ = |jm JI2 2 . (7.42)

Changing the direction of the normal n. that is turning the area

a about the point A7 it is drawn through, we obtain different values
of the vector prt at the same point M. Thus, a state of stress of an
elastic body at a given point A/ cannot be completely characterized
by a single vector. Hut it turns out that to obtain an exhaustive
description of such a state it is sufficient to determine the stresses
on three mutually perpendicular plane sections passing through
the point M because this makes it possible to find the stress at the
point M on an area (passing through A/) of arbitrary orientation.
 CI1. 7. TENSORS 2 9 7

Lot us establish the result stated above. Denote by pXl, px? and
Px3 the stresses at the point M on three elementary areas whose
normals go in the positive directions of the coordinate axes Oxu
Ox2 and Oxz. In other words, px is the stress on the area with unit
normal «»f, i — 1, 2, 3, where c, is the base vector of the
axis Oxi (Fig. 7.3). Consider a tetrahedron with one vertex at the
point AI and the edges M A y MB and MC parallel to the axes Oxit
Ox2 and Oxa. The outer normal n 2 to the face MAC of the tetrahed­
ron and the vector arc in the opposite directions. Hence, the

stress on tlie area MAC is equal to —pX2. Similarly, the stress on

the face BMC whose outer normal is iij — —oj is equal to —p.VJ,
and the stress on the area MAB corresponding to the outer normal
n 3 = - ( * , is equal to p*,. Let us designate by p„ the stress on the
area ABC with outer normal n and form the equation expressing
Newton’s second law for the tetrahedron MABC:

Py 4 r = CT,,a 0 cos (n * x i) P*« ~

— c cos (n, ;r2) px, —o cos (n, :r3) Pjc3 -}- y rr//pl' (7.43)
Here $at is the acceleration, a is the area of the face ABCy h is the
altitude of tlic tetrahedron (if the face ABC is taken as its base),
p is the volume mass density, f the volume force per unit mass (in
particular, it may bo the gravity force), the quantity y oh is the
volume of the tetrahedron MABC and the quantities cr cos (11, x,),
a cos (n, x2) and o cos (n, x 3) are, respectively, the areas of the
faces M BC, MAC and MAB. If we suppose that the acceleration
^ and the volume force f remain bounded] when o is made to lend
at f
to zero, then dividing equality (7.43) by o we obtain, in the limit,
the relation
0 = p., — p r cos fn. .r,> — p h* r_) p ^ cn* (n, .r J
298 MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

Consequently,
Pn = P.T, cos (n, x,) -r P i, cos (n, p*, cos (nf x3) (7.44)
Formula (7.44) expresses the stress pn on an area with an arbitrary
normal n in terms of the stresses on the areas whose normals go
along the coordinate axes.
Let us resolve the vectors p.t,, px, and pXf along the base vectors
Cj, e 2. e 3:
Px, = Pli^i + P 12e 2 + P 13°3
Px, = P21 C1 + P ZZC 2 -h P 23O3 (7.45)
P*. = Psiel P32e 2 + P:t9 C3
If, for a given point M y the matrix
Pll P\2 Pl3
II Pit II = P2 1 P22 P*23
P3i p32 Pas
is known, we can determine the stress on any area a passing through
the point M because the position of the area is specified by the
direction of its normal n and formulas (7.44) and (7.45) make it
possible to lind pn if the vector n is given. Thus, a state of stress
of an elastic body at a given point is completely characterized by
matrix (7.46).
Consider now the projection of the vector pn on the normal n.
The physical significance of the quantity thus obtained suggests
that it is independent of the choice of the coordinate system. To
find tlie projection we scalarly multiply both sides of equality
(7.44) by n .and apply formulas (7.45). This yields the expression
71

(pn, n) = y, pij cos (n, Xi)cos(nt Xj) (7.47)

t, ;'=1
and hence the sought-for quantity is given by a quadratic form
defined on unit sphere.
I fence, quadratic form (7.47) is invariant on unit sphere and
therefore its coefficients p Li (i, / = 1, 2 , 3) constitute an orthogonal
affine tensor of second rank n = (p tJ) (see Note 2 in § 3). This is
the so-called stress tensor
2. Stress Tensor as a Linear Operator. It is convenient to interpret
the stress tensor as a linear operator transforming the unit normal
n to an area into the vector pn (the total stress on the area).
Let us take the resolution of the^vector pft, with respect to the
basis e ,, o2t o3j which is of the form
Pr. — P ih^'l T P tia^Z
 CH. 7. TENSORS 2 9 9

and substitute it into the left-hand side of equality (7.44) and simul­
taneously substitute the expressions of the vectors p*,, pXt and pXl
in terras of the same basis (see relations (7.45)) into the right-hand
side of the equality. The resolution of p,, with respect to the basis
ei, e2, ° 3 being unique, we thus obtain the following system of three
scalar equations:
pm -PnCOs(n, xt) j-p21cos(n, x2) r Pzt cos (n, x3) '
P n 2 *— Pl 2 C-OS ( l l , X | ) - r P 2 ZCOS ( u , X2) Pz 2 C° S ( « , X3) > (7.48)
Pn3 — P 1 3 COS (n, Xi) p 2 3 cos (n, x2) - r p 33 cos (n, x 3) ,
These equations express the above mentioned linear operator in
the basis e2, e3. The unit normal vector n to an area a is expres­
sed as
n = ct cos (n, xt) -j- e 2 cos (11, x 2) + e 3cos (n, x 3)
Presenting the vector n as a row matrix of the form
n = || cos (11, x j , cos (n, x 2), cos (n, x 3) II
we can write (see Appendix to Chapter 7)
Pn = n || p u II (7.49)
where pn is interpreted as a row matrix || p nt, pn2, pn3 |[ and
II Pa II is the matrix corresponding to the stress tensor 0 = (Pa)
at the given point. When speaking about the multiplication of
a matrix corresponding to a tensor by a vector we simply say that
the tensor is multiplied by the vector.* Thus, to obtain the total
stress a I a point M on an area o with unit normal 11 we must take
the stress tensor 11 = (pki) at the point and multiply it 011 tho left
by the vector n;
p„ = n {pti) = nil (7.49')
Of course, we can also represent the unit normal 11 as a column
matrix
COS ( l l , X |)
11 — cos (n, x2)
cos (n, x3)
Pm
Then p„ (understood as a column matrix Pn 2 ) can be obtained
Pa 3
hy multiplying on the right the transpose Of die matrix (| p tj ||

* More precisely, we speak here about the multiplication of a tensor (inter­

preted as a linear operator) by a vector in the sense that tho corresponding
iinear operator i« annlifwl trv the ver*er> ?er> r>l50 the .'lul ->f !j.i=p aw liun.
300 MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

by the above vector 11. This corresponds to the application of the

adjoint operator (see footnote on page 291) associated with the ten­
sor II to the vector (column matrix) n.

§ G. ALGEBRAIC O P E RA T I O N S ON TENSORS
1. Addition, Subtraction and Multiplication of Tensors. The ope-
rations of addition and subtraction can be performed on tensors of
the same rank. For instance, the sum of two tensors of second rank
a-n and btJ is the tensor whose components are
Cij — -j- bij\ i, 7 =~ 1 , 2 , 3
and their difference is the tensor da r- a-ti — (t, j — 1. 2. 3).
It can be easily shown that the quantities c17- = a-ti bi} and
da = a,; — bjj are transformed according to the rule of transfor­
mation of tensors when the coordinate system is changed. The addi­
tion and subtraction of two tensors of an arbitrary rank are defined
similarly.
The product of tensors (also spoken of as the outer product) can
be defined for tensors of any rank. For example, the product of
a tensor of rank two atj by a tensor of rank three hmnp is a tensor of
rank five whose components ciJmnp are defined by the relation
C ijmnji = Q-ijbmnp" L / ’ P = 1* 3, o
We can easily prove that the above quantities C / ; , n n p are transformed
in accord with the rule of transformation of tensors when we
pa ss from one coordinate system to another. The (outer) product,
of two tensors of arbitrary ranks is defined similarly.
The product of a tensor by a number can he considered a special
case of the product of two tensors and is defined as follows: llie
product of a tensor aijh by a number C is the tensor with the com­
ponents bjj,. = CaijU. The fact that the quantities bt}h constitute
a tensor can be easily verified.
2. Multiplying Tensor by Vector. When studying the stress tensor
(see relation (7.‘19) in § 5), we dealt with a special case of multi­
plying a tensor of second rank (interpreted as a linear operator)
by a vector. TTcre we shall discuss this operation in the general form.
A tensor (£-,) can be multiplied by a vector x on the left or on
the right, i.e. we distinguish between the products x (L,j) and
{Ltj) x. T3ut in both cases the result of the operation is a vector
which is defined as follows. Let tlui matrix corresponding to the
tensor (/,j;) in the coordinate system specified by a basis e1( e2, v*
be equal to
//jo Lx*
I!/-;;!! = L*21 I j*2 T<23 (7.30)
Cl A;,-.
 C1I 7. TENSORS *,01

and let the resolution of x relative to the basis be of the form x =

a
= \ x fe {. Then the vector
y* = x {Lu) (7.51)
resulting from the multiplication on the left of the tensor
by the vector x is regarded as a row matrix || y*t y*. y * || whose
components i/f. y* and y* arc determined in this basis by the rela­
tion (see Appendix to Chapter 7)
1
^11 7-12 7-13

II U\» lit * If 3 II ~= 11 *^2 , ^ 3 II Z/21 X>22 7-23 (7.52)

7-31 7-32 ^33
The vector
y = (f 'u ) x (7.53)
resulting from the multiplication on the right of {Ltj) by x is given
bv the relation
«/l Ln 7^12 7-13 X,

y2 — I JO1 7-22 7-23 • x2 (7.54)

ys 7-31 7-32 7-33 *3

specifying y — ^ y ^ i in the same basis cj, e 2» 0 3 . Relations (7.52)

i= 1
and (7.54) can be written in the abridged notation
y* = x 1 1l u (7.52')
and
y = II L a II x (7.54')
The vectors x and y* entering into tlie former relation are interpreted
as row matrices and the vectors x and y in the latter relation as
column matrices.*
3. Contraction. Contraction of an orthogonal affine tensor is the
operation of putting one index equal to another and then summing
with respect to that index. For instance, if we take a tensor of rank
four put i = / and sum with respect to i we obtain the con­
tracted tensor of second rank whose components are given by
the equalities
3
&mn— Ci i nxn
t^l
The fact that the quantities amn constitute a~tensor of second rank
can be easily proved. If we take a tensor of an'even rank and perform
* Relations (7.51) and (7.53) (or, which is the same, relations (7.52') and
(7.5-4')) determine two linear operators which are the adjoinls of each other
(see footnote on page 291).
302 .MULTIPLE I NT E GR AL S , FI ELD T HE O RY AND SERIES

contraction operations on it as many times as possible we shall

arrive at a scalar, i.e. an invariant.
For example, if the contraction is performed on the product a fbj
of two tensors a t and bj of rank one the resultant tensor will be the
invariant scalar which is nothing but the scalar product of the
vector a (with components au a 2 and a 3) by the vector b (whose
components are 6,, bz and b3):
3
(a, b) = 2 aibi
i=*l

4. Interchanging Indices. Let us consider the operation of inter­

changing indices for an important special case, namely for an ortho­
gonal affine tensor of second rank Taking an arbitrary ortho­
normal basis ej, e 2, c 3 in which the components of the tensor are
L i} (i, / = 1, 2, 3) we put
j — L Jt
and thus arrive at the quantities Lfj (t, ; = 1. 2, 3) specified in
every orthonormal basis. It can be easily proved that L?, form an
orthogonal affine tensor of second rank. The tensor (L*j) constituted
by the quantities L*j (i, 7 = 1, 2, 3) is called the conjugate tensor
of (/>,;). The above operation is similarly performed on a tensor
of an arbitrary rank in which any two indices can be interchanged.
The resultant tensor obviously has the same rank
5. Resolution of Tensor of Second Rank into Symmetric and
Antisymmetric Parts. An orthogonal affine tensor of second rank
(Lij) is said to be symmetric if its matrix
j£ll L\2 L\z
1Lzt L22 L 23
\ L3\ L,t L 33
is symmetric in each orthonormal basis, that is if in every such
basis the relations L tj ~ LiX (t, / = 1, 2, 3) hold.
A tensor of rank two (Lij) is called antisymmetric (skew-symmetric)
if the elements of the matrix || L {j || corresponding to it satisfy
the conditions
Lij ~ Lji
in every orthonormal basis. The latter relations suggest that} for
an antisymmetric tensor (L^) we always have L it = —Lu , i.e.
2 L j,• — 0 and Ln = 0.
Thus, a symmetric tensor of second rank is completely specified
by its six components (since Li2 = ^ 2 i» “ L3l and L23 — L32
for such a tensor) whereas an antisymmetric tensor is characterized
l'v its throe nondingonal elements.
 CU. 7 TEN SOR S 303

The vector product of two vectors a and b is a simple example of

an antisymmetric tensor. Indeed, let the vectors a and b have the
resolutions
a = a,ct + a ze2 -f- a3 e3 and b= 6^ + b2o 2 + b3c3
in a basis et, c2, e3. Then the vector product [a, bl can be written as
ei e 2 e 3
Q\ az O’3
bi h b’A
= (aa53 —a3&2) e, -f- (a3&t — axb3) e2 4 - (ai&2 -'<r2&i) e3 (7.55)
Taking advantage of the fact that the vectors a and b are tensors
of rank one we can easily show that the nine quantities — a^bj —
— a}bt (i, j = 1, 2, 3) form a tensor of second rank. This tensor
is obviously antisymmetric because we have Lji = ajbf — a,bj —
= —(aibj — o,jbi) = —L ir Consequently, the tensor is completely
specified by its three components (a2b3 — a3b2). (a36 j — a,b3) and
{(i\bz — a zbx) entering into equality (7.5.5).
It can be easily proved that if the matrix corresponding to a tensor
of second rank is symmet ric (antisymmetric) in one orthonor­
mal basis it is also symmetric (antisymmetric) in any other ortho­
normal basis.
Finally, every tensor of second rank (Liy) can be represented in
the form of a sum of a symmetric tensor and an antisymmetric
tensor, namely as
Lij = — {Li) \ Lfji) h ~2 {Lij — (/.56)

where an(j [u ~ _Lf *_ ^ 2 , 3 ) are the components

of the symmetric and antisymmetric parts of (L^) which are
uniquely determined by the tensor (Li;).
In § 7 we shall consider an important example of resolution of
an orthogonal affine tensor of second rank into the sum of its sym­
metric and antisymmetric parts, namely we shall resolve the tensor
of relative displacements into the corresponding symmetric tensor
of pure deformation and antisymmetric tensor of rigid body rotation.

§ 7. TENSOR OK RELATIVE DISPLACEMENTS

Let us consider a state of strain of a physical body (see § 4).
Suppose that U = U (r) — c tut (xt, x 2, £ 3) 4- e 2u 2 (£1, x 2. x 3) 4-
q3u3 ( x j , x 2 , £ 3) is the displacement vector of a point specified
by the radius vector r 2 , 0 , 4 x 2e 2 A
r x 3 e3. Under the assumption
 U n.T lP L R INTERNALS, l-’IEI.n THEORY AND SE R IE S

ns u x% u .2 and
3 a re d iffe re n tia b le we can w rite
u
Ou , L <?"l d x ? -j- du. d x 3
dui =
C/r, d x , tfx2 dx3 t>

OU* , 0u2 du*

du 2= dxi dxz = d.r3 (7 .5 7 )
0*\ Ox*
Ou 3 On 3
dux = r/.r, 1 ^f/3 dx** d X n%
y
o*l 0.13

Passing from the Cartesian coordinates rr,, x 2. x 3 to new coordinate.'?

.3
a*', x', x', that is performing a substitution x* = a^x* where
*-*t
|| aji, || is an orthogonal matrix, we can easily verify that the quan­
tities , t, / = 1, 2, 3, constitute an orthogonal affine tensor of
rank two. This tensor is spoken of as the tensor of relative displa­
cements (corresponding to the state of strain in question). Intro­
ducing the notation for this tensor we can rewrite formulas
(7.57) in the form of the equality
<jU = ( i ^ ) d r (7.38)

Let us now resolve the tensor into its symmetric and

antisymmetric parts. Using the matrix notation we can w'rite down
this resolution as
’ du. «7u, du, |
0*i 0*2 0*3
0u? du* du*
0*1 0*2 0*3
On 3 du2 du3
dx, 0*2 0* 3
du, 1 / d u , , du2 \ 1 f 1 . du3 \
-f- -----
d x, 2 \ dx2 1 dx. ) 2 V dx3 dx, )
± |r ^2 du, \ du2 1 ( 0“2 , du3 \
■ 1 -I-
I
2 II 0 * , 0*2 / dx2 2 I 0*3 ‘ 0*2 /
± | ( 0 u 3 1 du, \ 1 / du3 du2 du3
2 'I <?x, 1 0 X 3 ) “2" V d x 2 1 d x3 J 0.1J
0 - — <i)3 <i)2
<1)3 0 — 0), (7.59)
-— 0)2 <0, 0

where <o,, a>2 and a>3 are the coordinates of the vector <0 — <0 , 0 !
<j>2c 2 ~\' ^ 3 ^ 3 which is equal to — rot U.
 CH 7 TENSORS 305

Till* first matrix on the right-hand side of equality (7.59) deter­

mines a symmetric tensor D describing a pure deformation (without
rotation) and the second matrix corresponds to an antisymmetric
tensor Q characterizing a rotation of the body as a whole (i.e. a
rigid body rotation without deformation). Relation (7.58) can now
be pul down in the form
dV = D dr + Q dr (7.60)
Pci forming direct calculations we can easily show that
Q dr = f-i- rot U, dr I
and consequently
dl. = D dr + £4- rot L, dr j (7.61)
Concluding our discussion of a state of strain of a deformable
physical body we indicate the following two special cases concerning
relative displacements r/L of Ihe points lying in the vicinity of
a point r: if the deformation is described by the displacement vec­
tors U( r) we note that
(1) in case rotU ~= 0 it follows from formula (7.01) that the
relative displacements c/U are due to a pure deformation;
(2) if D -= 0 (i.e. all the elements of the matrix corresponding
to the tensor D are equal to zero) the relative displacements r/U are
due lo a pure ndalion.

5 t>. TL'NSOIt KIKLH

I . Tensor Field. Divergence of Tensor. If to each point M belong­
ing to a domain (7 of space there corresponds a tensor (L{^) we say
that Ihere is a tensor field defined in the domain G. Here the
components />*; of the tensor (£,-,) are functions of the coordinates
of the variable point M (x)f x2, £ 3)-
Characteristic examples of a tensor field are the held of a strain
tensor and the field of a stress tensor describing a state of strain
and a slate of stress of an elastic body subjected to a deformation.
Indeed, in the general case a state of strain and a state of stress
of such a body vary from point to point and therefore the components
of the corresponding tensors depend on the coordinates of the vari­
able point {Xi. x2. x-i)
Let us suppose that the components Ltj of a tensor (A,ti) have
continuous partial derivatives of the first order with respect to
x l7 x z and x 3.

* For definiteness we shall deni with tensor fields constituted by tensors

of second rnnk.
20-082 v
30G MULTI P I , E tNTEU.KALS. F I E L D TIIEOPY AND S E H IE S

Take the matrix

L 12 L\3 14"
II LtJ l 22 L 23 (7.62)
1£ 3 1 L32 L33
whose elements are the components of the tensor (Lu) and form the
vectors
L| = I/hCi 4 •^'12^2 4" L|3C3
Ijq = ^21^1 ~T~ ^22^2 ~T* ^23^3 (7.tl3)
L3 = Lt 3jC j 4 " ^ 32l*2 4 “ L j $ :iC 3

The vector'—
-o x j
— —' 4 is called the divergence of the tensor
ox* 0-r3
(L^) and is designated by the symbol div (L^):
div (74j) = dx | / <^11 , <?L2i , \ ,
(?X*j 0 x3 \ dxt d x * 1 d x 3 ) '*

/OL 12 dL-y* , C^^32 \ / OL2.2 ^^"33 \ _

V dx\ *0x2 dxz / Cz \ dx | t>.r2 ' dx2 ) 3 ~
= (div {Lij^tCi - (div (Li j))2e2 -r (div (Lt j))2<i3 (7,(vi)
The above definition of the divergence of a tensor (£,,) is formal.
To justify the definition we must verify whether the divergence
thus defined is a vector or, which is the same, whether the quantities
(0LU 0L>\ , (0LiZ OL02 , "^32\ .,,.,1 fM' 13 ^L23 ,
17^ 7^ 77.- - — ) ’ 1 7 7 7 + 6^- + -a7T) “"<l \~UZ?~r 'o^:~r ~d^l
constitute a tensor of rank one. Thus, we must prove that the quan­
tities
$ 1 d/,2s 0L3t
(div(L u) ) ,= fixl ' dx-y + - dx3 ’ S = 1, 2, 3
are transformed like the components of a tensor of rank one when
the basis is changed. Let us rewrite expression (7.65) in the form
3

(div (Li,))s ^ 2 — , s-1,2, 3

h—1
and pass to a new coordinate system Ox\x'2z\. In the new system
we have
3

^mp &L’m p
(d iv (£ ,„ ));, 2
m — I
= 22
m—t «= 1
dxn 0*m *
p — 1, 2, 3 (T.Iili)

The passage from the old Cartesian coordinates to the new ones is
performed by means of an orthogonal matrix |I a nm ||. But, as is
innwn the inverse of an orthogonal matrix coincides with its traits
 CH- 7. TEN SOR S 30?

pose, and therefore we have

3
xn V a mnJ'm (7.67)
According to the definition of a tensor of rank two we can write
3 n
2 2 UmhOLplLhh /> = 1, 2, 3 (7.68)
h—1/—I
Substituting expressions (7.67) and (7.68) into formula (7.66) and
taking into account that
4 c _ M for " - *
/i —^txh — i rk r , ,
7/t— 1
1V 0 for n k
we obtain the relation
3 3 3 3

(div(/,ij));. J J S 2 ^
m — 1 n = l » i= t 1 1

- 2 S « r d 2 ( 2 “« ^ ) ^ r ' J -
ft- 1 f = l n-= l in=.\

Oxk ) =
Is- I N I n —1 1— 1 fr= t
3

= 2 «j>/(d»'■(£*/)/, p = t, 2, 3
i=i
which is what we set out to prove.
2. Oslrogrodsky Theorem for Tensor Field. Let the components
i — 1, 2, 3. / = 1, 2, 3, of a tensor (Li#) have continuous
first-order partial derivatives in a hounded closed domain Q whose
boundary cry is a piecewise smooth surface. We shall additionally
suppose that the domain Q satisfies the conditions under which
the Ostrogradsky theorem for vector functions holds.
Denote the unit outer normal vector to the surface oQ by n and
form, by analogy with relation (7.49') specifying the product
n {Pij), the vector u(L tj). Then we have the formula
f ^ n ( / , lJ) ( 1 o = \ ^j div (/,,j)du> (7.69)

« ‘

In other words, the flux of a tensor (Lfj) through a closed surface

Oq is equal to the triple integral of the divergence of the tensor
(Lij) over the volume Q hounded by the surface.
The flux of a tensor (Lij) through a surface On is equal, by defini­
tion, lo the surface integral on Ihe left-hand side of formula (7.69).
liu*
30S MI*I.TII’LK I N T E G R A L S . ITEM ) THEORY AND S E R I E S

Formula (7.60) expresses the Ostrograd-sky theorem for tensors.

Tlie proof of formula (7.69) reduces to applying the Ostrogrndsky
theorem for vectors established in Chapter 5 to each component
L\u cos (». a:j) -f L zk cos (n, x 2) - L Xk cos (n. x 3) {k = 1 , 2 , 3) of
the vector n
[ ^11 (L, j) <2o=c, j ^[Lji cos ( 11, z,)-;-L 2 1 cos(n, x2 )-i-L3i cos(n, x3)]do- 1-
Oq (T o

-t-C 2 j^ |L |2 cos (11, X j ) L2ocos (11, c o s ( n ta:a ]< 2c r - r

-b e3 ^ lL j 3cos (11, X|) -j~ L2 3 cos (n. *2 ) -b L3;}cos (n, ar3)lda =

Oo
= et j [ j (div (Lij))i d(0 b e 2 j ^ j (div (Lij))2di.o -f
1> s>J
-rC3 (div (L, ; ) ) 3 du* - (div (L,,)) d(o
<2 v a
Tims formula (7.69) has been proved.
3. Equations of Motion of a Continuous Medium. Let us apply
Ostrogradsky formula (7.69) to deriving an equation of motion of
a continuous medium. We mentally isolate an elementary domain

7.4

(Q) occupied by a moving continuous medium (Fig. 7.4). Denoting

the surface bounding the domain (Q) as oa we can write down the
equation expressing Newton's second law for the mass (distributed
with density p) carried by the domain (Q). We regard this mass of
the moving medium as being concentrated at its centre of gravity,
i.e. as a material point. Then the equation can he written in the
form

On
where Q is the volume of the domain (Q), f is the volume force
per unit mass and p* is the stress on an infinitesimal area do with
unit normal vector 11.
 Clf. 7. TKNSOHS 309

The surface integral on the right-hand side of equality (7.70)

is equal to
\ ^ prada *= ^j ii T1rfer = ^ ^ div il <Jm
Oq CJn Q

= Cj f f ( (div ri)| do) -h e2 f ( f (div l~l)2 do)-\-

»' .*•* *'j
n n
+ ea f f t (divII)3 *o (7.71)

where IT is the stress tensor. Applying the mean value theorem to

each integral on the right-hand side of equality (7.71) we obtain
p„d(r- - e,Q (div 11)1'-f e2Q (div Tl)* -} e3Q (div fl)J (7.72)

where (div IT){* (i - 1, 2. 3) is tlie value of (div TT)^ assumed

at a point M* £ (£2). Now substituting expression (7.72) into (7.70)
and passing to the limit as (£2) -■»- M (i.e. as the domain (£>) is con­
tracted toward an arbitrary fixed point .1/ £ (£2) we derive the vec tor
equation of motion of a continuous medium:
P -f Pf-: divl! (7.73)

Vector equation (7.73) can be rewritten in scalar form by pro­

jecting its left-hand and right-hand sides on the coordinate axes.
This yields the following system of three Scalar equations:
dv| dpi i $Pz\ | <*Pm
P ~jr ,°/i +* dx\ c?ar2 ' i)x3
H v-i r | d P t'l d p 22 t (?Pj2
‘ i}/ ‘ 2 < t?.i'j 1 Oun * t/X j (7.74)
.. d l a ^f • d P to i ° P r .t . <)pM
' dt t/xj • Jx2

§ 9. PRINCIPAL AXES OF SYMMETRIC TENSOR

OF SECOND HANK
Let us take an orthogonal affine tensor {Lii) which we shall inter­
pret as a linear operator (see Sec. 2 in § 2):
y - L (x) (7.75)
The eigenvectors and eigenvalues of the linear operator L (x) are
ref cured to as the eigenvectors and eigenvalues of the tensor (Z,^).
We remind the reader that an eigenvector o j a linear operator L (\)
is defined as a nonzero vector x satisfying the relation
L (x) = A\ (7.76)
310 MULTIPLE INTKC'.KALS, FIELD THEORY AND S E R I E S

where X is a scalar factor. The number X is called an eigenvalue of

the operator L (corresponding to the eigenvector x).
Passing from the vector x to its coordinates x lt x 2 and x 3 in a
basis Ci, c 2, t*3 , wc can replace vector relation (7.76) by the equiva­
lent system of scalar equalities

(7.77)

Lot us regard (7.77) as a system of equations in the unknowns

x,. x 2 and x 3. For this system to possess a nontrivial solution, that
is one for which x t, x 2 and x 3 are not simultaneously equal to zero,
it is necessary and sufficient that the determinant of system (7.77)
turn into zero, Hence, the eigenvalues X are determined by the
equation

(7.78)

As is well known, if {Li}) is a symmetric tensor, i.e. if its matrix

is symmetric in every orthonormal basis, all the roots Xi9 X2 and
?.:<of equation (7.7S) are real. In this case it is possible to construct
a system of throe unit eigenvectors cq, e 2 and e 3 associated with
the eigenvalues and X« such that they form an orthonormal
basis in which the matrix of the operator L takes the diagonal form
0 0 1
0 l2 0 (7.79)
0 0 L.s
The vectors elf e 2 and c 3 thus found specify the so-called prin­
cipal axes of the tensor (Ltj). As an example, we can mention the
principal axes of the conductivity tensor of a monocrystnl (that
is a homogeneous anisotropic body) which are the crystallographic
axes. The discussion of the properties of the principal axes of the
inertia tensor, strain tensor and stress tensor can he found in courses
of theoretical mechanics and mechanics of continue.

§ 10. GENERAL TENSORS

The notion of an orthogonal affine tensor discussed in the fore­
going sections is connected with the transformations of orthogonal
Cartesian coordinate systems and with the corresponding transfor-
 Oil. 7. TEKSUK5 311

Here we shall give the general definition of a tensor which invol­

ves oil the possible Cartesian coordinate systems (including the
oblique ones) specified by the arbitrary bases.
1. Reciprocal Bases. Let
e,, c2, c3 (7.80)
be three arbitrary noncomplanar vectors forming a basis in space.
For brevity, we shall denote such a basis by a single symbol ex.
Consider the triple scalar product of the vectors e„ ea and e3
which is equal to the volume V of the parallelepiped constructed
on these vectors (i.e. the one whose coterminal edges coincide
with the vectors c,. e2 and e3):
V = (et, e2, e3) (7.81)
The vectors eft(/r —1, 2, 3) determined by the relations
i __ 2. C3 I _ (e3» Ci1 _ 3 __ lc li cal
1 — y » c — y * c — y (7.82)

constitute a basis which is said to be reciprocal to e*.

We can easily show that, conversely, the basis e» is reciprocal
to the basis e,!. Indeed, the volume of the parallelepiped whose
edges are the vectors e* is equal to

l" ~ (e » . e2, e3) = ( - ^ N - , |cy e^ ) =

(l° 2 , e3] fe3, e,!, [e,, e2j J ) =

— ^(^2 » ^.1 ] (C1P1C3) ^ ^ ~ "j/iT= “y (7.83)

Therefore we have
W = 1 (7.84)
til us
[ c - , C'*| __j r J [c3, C |], [C j, e 2[ 1 __ V-Cj __
yt v J y-i | — y2 ci (7.85)
and, similarly.
, es, e1 , , cl. e2 ,
I yi J I y* J—2 (7.8B)
From relations (7.So) and (7.86) it follows that
U for i ^ k
(i-i, c*) = 6* = < (7.87)
1 for i — k
312 MU LT IP LE INTEGRALS, FIELD THEORY AND S E R I E S

It should be noted that every orthononiial basis ej, c3 coincides

with its reciprocal basis.
2 . Covariant and Contravariant Components of Vector. Let us
take a basis et- and its reciprocal basis el and write down the resolu­
tions of. an arbitrary vector x with respect to the bases:
3^ J
x— T x t ©* — x ' c t (7.HIS)
*=.1 i=»

he coefficients entering into the resolution of the vector x relative

1
to the given basis are called the contravariant components (contra-
variant coordinates) of the vector x in this basis. Thus, the numbers
x and x t are, respectively, the contravariant components of the
vector
rp I
x in #the bases elt
1
e "Zf e w3 and e l,7 e2, e3.
l ne covariant coinponenis of a vector x in a given basis are the
scalar products of the vector by the base vectors of the reciprocal
basis.
Multiplying scalarly equalities (7.88) by e* (vk) and taking advan­
tage of relations (7.87) we find that the covarianl components of
the vector x in the bases e1, e2, e 3 (oj. e 2, C3) are respectively equal lo
3 3

(x, C/t) = Xi (e \ e/{) = Xiblu= Th (7.8'd)

i«=I i =l
and
3 3

<x, o'1) _ V t-*) = ^ ~xk {7 .*><>)

t~l t=i
C.onsequently, the covarianl components of u \cctor in a given
basis are its contravariant components in the reciprocal basis.
d. Summation Convention. In tin* theory of tensors we usually
follow a summation convention (due lo A. Linslein*) which applies
as follows: if a subscript and a superscript entering into an expres­
sion are. labelled by the same symbol, this symbol (index) is under­
stood as denoting a summation with respect to that index over
its range.
In wliat follows the indices under consideration take on the values
J, 2 and 3 and hence a summation, if necessary, is carried out with
respect to an index ranging from one to three. For instance, applying
this rule to the sums entering into formula (7.88) we can write
down the resolutions of the vector x in the form
x = XjC1, x = x'e,- (7.91)

* hinstcin. Albert (1879-1955), the groat 20lh century physicist, the creator
of t l i P t lipnrv o f Vi * 1 t i v 51 /linm in D p r n i :t 11 \ \
 CH. 1. TENSORS 313

Similary, a bilinear form an..r'xh is written as

i, <i~l
aikx1** (7.1)2)
etc.
4. Transformation of Bast* Vectors. Lei us consider the transfor­
mation from an old basis <»,• to a new basis e»*. Using the summation
convention we can write
<V = a;.Ci, 2, 3 (1M\)
where the coefficients ccj, form the transformation matrix |Jaj. |j
new basis i.e.
*s- ]' oV*1'
ii = a2* CLTr <v3
*2’ (7.94)
a l a i oi'i.
«»
If we consider Lhe inverse transformation from the now basis
e, to the old basis c, which is written as
Cj=ct;'tY (7.!C>)
tlie matrix ||aj'j| is obviously the inverse of the matrix ||a{. ||.
Actually, substituting the expression Cj' = ccj.e, into the equality
eh — (7.1111)
we obtain
e ft —al/aj.Ci (7.f)7)
The resolution of each vector e>« {k = 1, 2, 3) with respect to
the base vectors <»,, e 2 and being unique, we derive from formulas
(7.97) the relations
0 for i=r=k
t (7.98)
< a i' 1 for i —k
which suggest llml the matrices |Jaj'|J and ||aj, || are mutually
inverse.
'>. Transformation of Covariant and Contra variant Components of
Vector. Let us first consider the transformation of the contra variant
components of an arbitrary vector
X TV; = (7.'.)!))

when a basis e, is transformed to a basis «v- Substituting the expiv.—

sion *'i — a) <v into formula (7.99) we see that
x -= .rVl'Cj' —
314 MULTIPLE IN'TEUUAl.?, FIELD TUF.OIIY A N D SEKTES

14y the uniqueness of the resolution of the vector x in the basis

c,-. Co', e 3', we can write
xv — aYx*
I (7 .100 )
'Thus, the “new” contravariant coordinates xr are expressed in
term s of the “old” contravariant coordinates xx by means of the
matrix || a\ || specifying the inverse transformation from the now
bii-is Ci' to the old basis This accounts for the term “contra-
variant components” which indicates that the expressions of vi
in terms of ep and of xl in /
terms of xx involve, respectively, Ilie
j
elements of the matrix [|ct* || and of its inverse.
Bv analogy with (7.100), we obtain the relation
x 1—aj,xr (7.101)
Let us now proceed to investigate the transformation of the cova-
riaril components x* of a vector x. We have
—(x, Of), T v - (x, Ci') (7.102)
and consequently
xp= (x, Ci-)= (x, a \ , V i ) —a’.Xi (7.103)
Similarly, by analogy with (7.103) we obtain
x\ — aj'xp (7.104)
'I'llus, the transformation of the covariant components of a vector
is performed by means of the same matrix as the transformation
of the base vectors. The tunus ‘Vovariant components” indicated
this coincidence of the matrices.
6. General Definition of Tensor. As before, we shall denote the
transformation matrix from an old basis e* to a new basis up by
|| <xl' || and the matrix of the inverse: r transformation from the new
basis up to the old basis e, by || a t |j.
.D e f i n i t i o n 1 . .4 quantity A which tsspecified in every basis e* {i —
= 1,7 2,1 3)' *bu * means of3 3piq numbers A ljl2’ 71,1''• -tli
Jfi where Ike indices
s = 1, 2, 3, . . p, and j t , t = 1, 2, . ! ., q, independently
assume the values 1, 2 and 3 is called a t e n s o r o f r a n k (or*(er)
{) q {p-fohl ea rn riu nt and q-fol(t. con t ra m r ittn t) if these
numbers undergo the transformation determined by the formulas
Aj\ * ’*'- Q = r j Xy‘f-t ** JP2 • . . ?<J (7.105)
*1 X3 a Za > H ■■ ■ 1*2 • • • »J#

* More precisely, formulas (7.100) show that the coefficients cuter lug ii*to
the expressions of x x in terms of x» constitute a matrix which is the transpose
of the inverse matrix j| a*'|l.
 CH. 7. TEVSORS 315

when we pass from an arbitrary basis (cj, e2, 0 3 ) to any other basis
tV, e2', <?3 ' where || a\' || is the transformation matrix from the
basis
_
Cj, c2, e 3 to ike» ■basis
•
e*-, e2', <±3 ' and |{ a} || is its inverse matrix.
The numbers -1^’. are spoken of as the components of the
tensor A relative to the basis e<. The superscripts j x, . . . . jq are
called the contravariant indices of the tensor and the subscripts
ij, . . ., ip are its covariant indices.
There is an alternative form of definition of a tensor (equivalent
to the above):
Let, in every basis tq. e 3, there be given a system of 3p+y numbers
A ,•j'jv l q w^iere the indices is, .9 — 1 , 2 , and ft , t =
= 1 , 2 , . . ., g, independently assume the values I, 2 and 3. If the
passage to any other basis e is c2'» Cv results in the transformation
of these numbers according to formulas (7.105):
k
12 - - >P 1 *2 ip J‘ J~

where || aj- || is the transformation matrix from the basis Cj, e 2, e 3

to the basis C|*. o2', 0 3 - and |( a* || is the inverse of || cc{' || we say
that we are given a tensor of rank p -1- q, p-fold covariant and q-fold
contravariant (or c o m rift n t o f r a n k p a n il co nt r a v a r i a n t of
r a n k q).
Examples
1. (a) The SVSt e j n of coefficient? atili of an invariant bilinear form
/ (x, y) = aixi2x^y^ = a^.x^y^, (7.10(5)
(x = xitCi, = ar'Jeij, y - y ^ c i2 ----- y V * ,
h = 1, 2, 3; i2= 1, 2, 3; t\ = i, 2 , 3; i ^ \ , 2, 3)
is a covariant tensor of second rank, i.e. having only covariant
indices.
Indeed, substituting the expressions
x't r- and y'* = ct$yt2 (7.107)
into formula (7-106) we obtain on identity involving the coordinates
of two arbitrary vectors x and y, which implies that
a ilia
*: 1- (7.108)
v 9
lb) In particular, if / (x, y) is equal lo the scalar product (x, y)
of two vectors x and \. tlie collect ion of the coefficients of
the bilinear form
3 iu .MILTIIM.K INTF.ORALS. PIE1.I* T H E O R Y AND S E R I E S

is termed a (fundamental) metric tensor or a eovariant inetrie tensor.

The elements of the inverse of the matrix ,».,{! (denoted by the symbols
guh) form a so-called contravariant metric tensor.
By the symmetry property of scalar product ((x, v) = (y, x)),
the tensors g(j and are symmetric, that is we have, in every
basis, the relations
8ij = g}n SiJ = ZH
2. Tlie elements L\ of the matrix of a linear operator G deter­
mined by the relations
L ( c , ) « L { e if i-1,2,3 (7.109)
constitute a tensor (of second order) eovariant of rank one and con­
travariant of rank one. For, in the new basis tv , we have
L (e^) = Li'Cj* (7.110)
On the oilier hand, we have
L (ej*) = L ( a j . C f ) —aj.L (c,) = a li,LjarcJ» (7.111)
Comparing (7.11U) with (7.111), by the uniqueness of resolution
of every vector L (e*) with respect to the basis tv , we obtain
l.{‘ = al-aj L> (7.112)
3. The covariant components of a vector x constitute a covariant
tensor of rank one, ami the contravariunl coordinates x%of x form
a contra variant tensor of rank one.
7. Operations on Tensors. In the general case Ihe operations on
tensors are defined in the same manner as for the orthogonal
affine tensors. The operations of addition and subtraction are natu­
rally defined only for the tensors of the same rank having the same
number /» of covariant indices and the same number q of contra-
variaiiL indices. The contraction is applied only to mixed tensors
(i.e. having both contravariunt and covariant indices) by pulling
one contravariant index equal to a eovariant index and summing
with respect, to that index.
There are also some other operations such as raising or lowering
indices by means of multiplying the fundamental (covariant) metric
tensor or the contravariant metric tensor by a given tensor T and
then contracting the product by pulling a eovariant. (contravariant)
index of gtJ (gi!) equal to a contravariant (covariant) index of T
and summing with respect to that index etc.
ft. Some Further Generalizations. Further generalizations are con­
nected with the introduction of curvilinear coordinates. This gives
rise to some new notions such as eovariant and contravariant diffe­
rentiation of a tensor and others. For ' li»* general theory of tensors
Wi. n«fi>r tl»n i * n n d c r I n f41. M ()| and Mil.
 CH. 7. TENSORS 317

A P P E N D IX TO CHAPTER 7

ON MULTIPLICATION OF MATRICES
Wo remind the reader that the product P*Q of two rectangular
matrices P and (J is only defined for the case when the number of
columns in the first factor equals the number of rows in the second
factor. If
P m P iz • ♦ • Pin Q11 (] 12 ♦ ' - <?i*
p -= P21 P22 ♦ • • Pin and Q = <?21 Qzz u 0 * <?2 s

Pm 1 P m2 ■• • Pmn <7nJ Qn2 - • • y ns

are such matrices their product is the matrix
^11 r a
. . . Tin
/*21 r 22 . . . /*2n

r,„ i r aj 2 • • * mn.
T

whose element rf; (i = 1, 2, . ., /»., j ~ 1, 2, . . #i) belonging

to the ith row and the ;th column is determined by tho formula
n
/*ij = S P*v(7vj
V— 1

Hence, if we interpret the elements of the ith row of P as llie

coordinates of an n-dimensional vector and the elements of the j i h
column of Q as the coordinates of another ^-dimensional vector
we can say that rtj is equal to the scalar product of the ith row of
the first factor by the ylh column of (he second factor.
Let us take two vectors x and y and represent them as the column
matrices
1
xx !/i
X — x<> and y — yz
*3 V*
where xu z 2, x 3 and y t, z/2, p 3 are the coordinates of x and y in
a given basis. If || |J is a matrix of the form
Ln ^ 1 2 7^13
L = 7^21 7^21 ^ 2 2

L Zl 7^32 L 3 3
then, by definition, tho equality y = Lx is equivalent to llie relation
Ux L n L \2 X,
—
y z 7^21 ^22 7^23 ( 1)
y 3 7>31 ^32 Z/33 | *3
318 M U L T l l ’I.E INTEGRALS, FIELD TI1E0KY AND SERIES

After the multiplication or the matrices on the right-hand side

of (1) has been performed we obtain

1 V L ikxk
h=l

ht 3
If— 33 LzhXk ( 2)
1/3 3
V>
**3/4 Xf, j
/i^l |
which is equivalent to the three scalar equalities
U\ — Li\Xi-r LiaT^ j
1/2 = L">\Xi - r " h £>2 3 X 3 r (3)
y 3 — £ 31^1 J"L&XZ ~T~Z/33X3 )
As has been said, relation (1) (or equivalent relations (2) and
(3)) is the definition of the multiplication (on the right) of a matrix
by a vector.
We can similarly lake two vectors x and y* and represent them
as the row matrices
x = II xu x 2, x3 II and y* = || y y ? t y*
'1 hen an equality of the form
Lu / / | 2 Z/J3
II.V11 'A> </$ || H I* !, **il f'Zi Is<\2 /-23 ('O
■^31 Z/32 ^33
can be rewritten as •V
3 3 3
yT* itzi y$ || = || 2 htyTii 2 ljizx u 2 Li3xt
i -1 i=l i —1

after the multiplication of the matrices on the right-hand side has

been performed. The latter relation is equivalent to the three scalar
equalities
y* = L>uX1 | £/21-3?2 + Z/31X3 1
y\ — Z/J0T1 -h Z/22-^2 1' / (5)
yt —L\%X\ ; L>Z3t 2 : I'M** J
In the contracted notation equality (1) is written in the form
y — hx (<>)
and equality (4) in tho form
v* = xL (/)
where. L is multiplied by x on the left.
 8 Functional
Sequences
and Series

In this chapter we shall study sequences and series (referred to

as functional sequences and series) whose members are functions.
in practical applications we often try to expand a given function
in a functional series whose terms are functions which are in a certain
sense simpler than the given function. Such an expansion facilitates
the investigation of the function, the computation of its values
and the integration. Functional series are also used in the theory
of differential equations and other divisions of mathematics and
its applications.
In investigating the properties of functional series and sequences
we introduce various types of convergence. Among them, uniform
convergence and convergence in the mean are of particular importance.

§ 1 . UNIFORM CONVERGENCE.
TESTS FUR UNIFORM CONVERGENCE
1. Convergence and Uniform Convergence. Let us consider a
set}uence of functions
/i (x)» /2 (*F)i •> /»i (x), • •• (£>-1)
defined on a closed interval a £ x > b.* If an arbitrary fixed value
x0 £ [fl. b] is substituted for the current variable x functional sequ­
ence (8.1) turns into a numerical sequence of the form
/l (*0), /*- (*^o)» * • fn (-^o) i • • • (8.2)
Functional sequence (8.1) is said to be convergent at a point xn
if number sequence (8.2) is convergent. Functional sequence. (8,1)
is said to be divergent at a point .r0 if sequence (8.2) is divergent.

* Instead of a closed interval a ^ x h we can take any other set X of

values of x, for instance, a <* t. <t b. n r h ^ i, z x —-.jo,
a x <z -f~oo, — x <l -•}-no etc. In what follows we shall stipulate the
ea«oc when such a replacement «>f a closed interval f«i. />| hv an nrhilmry set
is inadmissible.
32U M l'l .T ll’I.E I NTBllitALS. lUELU TIll-luHV AND SEKLES

Accordingly, in the former case x 0 called a point of convergence

of sequence (8 . 1 ) and in the latter case a point of divergence.*
If a functional sequence converges at each point x £ (a, 61 \vc
say that it converges on the interval [a, 61. If a sequence {/u (a)},
w ® 1, 2 , . . converges on (a , 61 there exists a certain limit
lim /„ (x) (which in the general case can vary from point to point)
n —> jo
a I each point x of the interval (c, 61. Therefore this limit is a func­
tion / (x) defined on la, 61. The function / (x) is called the limit
of functional sequence (S.l) and we write
i n (*) / {x) as n - ►—oo
or
liin fn (x) = / (x) on (a, 61 (8*3)
71-+CO

We can now formulate the following

l ) e f hi it i o n / . A functional sequence {fn (x)) is said to be c o n v e r ­
g e n t to a function f (x) on the interval [a, 6 | if for each fixed value
x £ [a, 61 the number sequence fn (x)t n = 1 . 2 , . . . . converges to
the number f (x), that is if for every e ;> U and every x 6 [a, 61 there
is a number N — N (e, x)** {dependent on e am/, generally speaking,
tm x) such that
I in (*) — f (x) I < e for every n > N (e, x) (8 ./i)
Among the convergent functional sequences the so-called uni­
formly convergent sequences are essentially important.
nit-ion A functional sequence {fn (x)} is cal led u n i fit rut I f/
c o n r e a j e n t on an interval [a, 61 to a function f(x) if, given any e > U ,
there exists a number A — N (e)*** {dependent on t, but independent
of x) such that the difference between fn (x) and f (x) satisfies the con­
dition
I f n (x) — / (x) | <C e for every n > N (e) (8 .0 )
for all x £ la, 61 simultaneously.
Tliis definition can be restated in an equivalent form:

* The set of all the points of convergence of functional sequence (8.1) is

referred to as the domain (or region) of convergence of the sequence The domain
of convergence of a functional sequence can be an arbitrary set of any complex
structure. It may coincide with the whole x-axis {as in the case of the sequence
\
fH {x) = — . —00 < x C -f-oo, n = 1, 2, . . ., convergent on the entire x-axis
to the function / (z )= 0 ) or he an empty set containing no points (e.g. for the
sequence fn (x ) = (—1)™, — 00 < x < f-0 0 , n -- 1 , 2 , . . ., which diverges
at every point x € (—0 0 , -(-oo)).
** The number N (e, x) may not be an integer.
*** The number N (c) is not necessarily an integer.
 CH. 8. FUNCTIONAL S E Q U E N C E S AND S E R IE S 321
D efin itio n A junctional sequence {fn (x)} is said to converge
uniformly to a function f (x) on an interval Ia, 61 if
sup | fn (x) - / (i) | -► 0 for to o (8.5')
a^x^b
that is if the least upper bound of | fn (x) — / (x) | (spoken of as the
m a x i m u m , d e v i a t i o n o f the f u n c t i o n / « (x) f r o m the
f u n c t i o n f (x) o n th e i n t e r v a l la, 61) tends to zero as n o o .
Indeed, if condition (8.5') is fulfilled then for every e 0 there
is N (e) such that for any n ;> ;V (e) the inequality
sup | f n (x) — / (x) | < E
a<x^b
holds for all x £ la, 61. But, by the definition of the least upper
bound, we have
I L (*) — / (*) K sup | /„ (x) — / (x) |
for all x 6 la, 6U Therefore relations (8.5) are also fulfilled.
Conversely, if relations (8.5) lake place we have
sup | fn (x) - / (x) | < 6
a*7:x^b
for every n JV (e), which implies (8.5') since e > 0 has been chosen
quite arbitrarily.
I 'nifoi jn convergence of a sequence {/„ (x)} to a funclion / (x)
on [a, 61 will be designated by the symbol relation
fn (x) =£ / (x) on [a, 6J (8.0)
The notion of uniform convergence admits of a simple geometric
interpretation. Isolation (8.5') means that llie least upper hound
of the deviation of the graph of the funclion // = f n (x) from that

of the function y = / (x) on the interval (a. 61 tends to zero ns

oo. In other words, if wo envelope the graph of the function
y f (x) by an “s-strip” (shown in Big. 8 1 ) determined hv the
relo lions
! (x) — e < . */ <. / (x; ; f, a <^. x ^ 0 (5.7)
21—0824
322 MCI.TIPLE IN T E G R A L S . F1EI-D T H E O R Y ANI) SERIES

then, beginning with a sufficiently large n% the graphs of all the

functions y — j n (x) entirely lie within the e-strip enveloping the
graph of the limit function / (x).
Exam pies
1. The sequence fn (x) — -i- sin nx converges to / (x) = 0 as
n -f- oo on the entire x-axis —oo < j < —oo. Here the conver­
gence is uniform because | fn (x) — / (r) | = -J- | sin nx | * -i- < e
for all x, —oo < x < : oo, siimiltaueouslv provided n > .V (e) —
1
e
2. The sequence /„ (x) — xn converges, as n —»- ; oo. o n llu?
interval 0 ^ x ^ 1 to the function / (x) determined by the relations
0 for 0^! x < 1
/ (*) = 1 for x = 1
But here the convergence is nommiform. For, if we take 0 < b < 1
and 0 < x < J the inequality | /„ (x) — / (x) | = x 11 <C e holds
only when n > N (e, x) = , and N (e, x) = Too
if x 1 — 0 for every fixed z £ (0, 1). Consequently, for every e
taken from the interval 0 < e < 1 there is no finite N (e) inde­
pendent of x such that the inequality | /„ (x) — / (x) | -- xn <C e

holds for every n > Ar (e) and for all x belonging to the half­
open interval 0 ^ x < 1. If we replace the segment 0 .< xl *; 1
hv a smaller segment 0 x 1 — ft with an arbitrarily small i),
0 <C <C 1. the sequence /„ (x) - xn converges uniformly to its
limit / (x) =r 0 on this smaller interval. Indeed, we have
In
A ( ‘'- *> = -}— « A' (*) - In (1 - 6 )
for 0 a 1 6
 CH. 8. FIrNOTION’AL S E Q U E N C E S AND SEHIF.S 3 23

and therefore |/„ ( x ) — /( x ) | = / < f for all * £ [ 0 ,1 — 61 when

« > A (e) ” An (i —fij •
From the geometrical point of view lliis example can be inter­
preted as follows. In Fig. 8.2 we see the graphs of several functions
f n (x) belonging to the sequence and the graph of the limiting func­
tion / (x). the latter being shown in the heavy line. The graph of
/ (x) consists of the half-segment 0 ^ x <C 1 (with the end point
x = 1 excluded) of the x-axis and an isolated point with the coor­
dinates (1, 1). Let us envelope the graph of the limit function by
an “e-strip”. 0 <C a < 1. The graph of every function /„ (x) = xn
starts from the origin of coordinates and its right end point lies
at the point (1. 1). Therefore the function /„ (x) — x'1 being con­
tinuous, its graph must leave the “e-slrip” at a point x, U < x C 1.
Hence, the sequence /„ (x) x'*, n = 1, 2, . . ., converges nonuni-
lornily on the interval 0 ^ x < 1.
3. The functional sequence j n (x) = — arc tan wx, —oo «< x <C
<C oot n = 1, 2, 3, . . ., converges to the function
( — 1 for — oo <C x <C U
/(x )= s g n x = \ 0 for x = 0
(. - 1 for 0 < x < !- oo
hut the sequence does not converge uniformly which can he easily
established by the geometric mol hod applied in the foregoing
example.
')njr
4. The sequence of functions /„ (x) — — 1 — Tt—T" , n — 1, 2, . .
converges 1«> the function / (.r) — 0 «n Hie positive x axis 0 ^ x
< oo. To find out whether the sequence is uniformly convergent
to its limit on the half-line 0 ^ x <! { oo we shall check up the
validity of relation (8.5'). Thus, we must verily if sup | fn (x) —
0 -f 00
— / (x) | —> 0 for n -f oo. r]’o this erul we evaluate the maximum
of q.H(x) = ] fn (x) — f (x) | = 1 _T” on the positive half of
x-axis. We have
. . , (1 0 1—«2>r2
(s) --------------------- *= 2u t r r ^ w
and hence <p«(x)=0 for 1 —n2x > - —0, i.e. for x « —--t
Consc-
([lienlly,

max <r*(x) — (—) = I -f~ 0 for n12 CO

21 *
324 m u l t ipl e in te g r a l s , f ie l d theory and se r ie s

Therefore the sequence does not converge uniformly. In Ibis example

iionuniform convergence is due to the fact that the maximum value
of fn (x) which is equal to unity coincides with the maximum deviation
of the graph of f n (x) from the graph of / (x) on the interval 0 ^
x < r ° ° , the point of maximum xn = moving to the left
as n — oo (see Fig. 8.3).

The notions introduced for the functional sequences can he easily

transferred to the functional series of the form

V, Uk{x) = M, (x) « 2 (x) 4 - - • • I Uh (*) '! • • • (8 .8 )

ft—1
where Ihe functions uh (x) are defined on a certain set, for instance,
on a closed interval la, 61.
D efin itio n A functional series (8.8) is said to be c o n n e r y e n t
if the sequence of its partial sums
Sa(x) - 2 uh (x), n — 1, 2, . . . (8.9)
h^l
converges.
The limit
S (x) =lim S n (x) (8.10)
n-^+oo
of the partial sums is called the s u m o f s e r i e s (8.8). / / series (8.8)
converges and its sum is equal to S (x) we write

S (x) — 2 (*) (811)

h=\
D e f in it ion .*1 convergent functional series (8.11) is said to con­
verge u n i f o r m ! y to its sum S (x) on an interval [a, b} if the sequ­
ence of the corresponding partial sums S n (x) is uniform lif convergent to
the sum S (x) 'on [a, b\. i.c. if for every r. > 0 there exists N — N (e)
such that the difference between S„ (x) and S (x) satisfies the inequality
oo

|S ( r ) —.<?„(*) H | (X 12)
k^n+ L
 CH. 8. F U N C T IO N A L SEQUENCES AND S E R IE S 325

/or all x £ [a, b\ simultaneously when n > A' (e) or, in other words', if
o
©

sup |S ( x ) — (x) | = sup | W | “ ^0 as n 4 oo (8.12')

-

ai^ x^ b a<jc:\ 6 fc^ n + 1

Examples of uniformly (nonuniformly) convergent functional
series can be easily constructed on the basis of uniformly (nonuni-
formly) convergent sequences. For, if we are given a functional
sequence
/ 1 (^)i fi fa*)» • • •> fn faO* * ♦ * (8.13)
\vc can take the series
f\ (x) "1* I/ 2 faO — ft fa-)) + I (x) — f 2 (x)l + - . •
/ 3

« • • f l/n (x) — fti-l OOJ *+* • • * (8-14)

for which (8.13) is the sequence of its partial sums. Therefore, if
sequence (8.13) converges uniformly (nonuniformly) then, according
to Definition 2t, series (8.14) is uniformly (nonuniformly) convergent.
It should be noted that, conversely, uniform (nonuniform) con­
vergence of series (8.14) implies, by virtue of Dclinition 2l} uniform
(nonuniform) convergence of sequence (8.13).
The following two assertions are an immediate consequence of
the definition of uniform convergence:
(1) The sum of a finite number oj uniformly convergent sequences
(series) is a uniformly convergent sequence (series).
(2) If all the terms of a uniformly convergent sequence (series) are
multiplied by a bounded function (x) (in particular, by a constant)
this does not affect the character of its convergence which remains
uniform.
The assertions are easily proved. For instance, the second one is
proved as follows. Let /„ (x) zZ / (x) on la. 61 and let C. 0 <T C C
< 4- 0 0 he a constant such that j tp (x) | < C for all x £ 1«, 61.
Suppose we are given an arbitrary k D. By the uniform conver­
gence of fn (x) to / (x), we can find N (e) such that j /„ (x) — / (x) ( <
<Zjr for all z £ la, 61 simultaneously w'heu n > Ar (e). Bui for
such values of n Ar (e) we then have
1<r (*) fn (x) — cp (x) / (x) | = | q> (x) H fn (x) — / (x) | <

for all x £ [a, 61 and hence <p (x) fn (x) ZZ V (2 ) / faO* as «

on the interval [a, 61. Regarding /„ (x) as being the nth partial
sum of a uniformly convergent functional series and / (x) as the
sum of this series w*e thus conclude that the assertion is valid for
the uniformly convergent scries as well.
2. Tests for Uniform Convergence. If the limit / (x) of functional
sequence (8.13) is known its uniform convergence can be tested on
326 M U LTIPLE INTEG RA LS, FIELD THEORY AND SER IES

the basis of Definitions 2 and 2t or by means of the corresponding

geometric interpretation as it was done when investigating exam­
ples 1-4.
But it sometimes turns expedient to reduce the question of uni­
form convergence of functional sequence (8.13) to testing uniform
convergence of corresponding functional series (8.14) for which the
sequence in question is the sequence of partial sums. Such a reduc­
tion may be useful because there arc various tests for uniform con­
vergence of series convenient for practical application.
One of the simplest and most commonly used tests of this kind
is the so-called iVetcrslrass* M-tesl based on comparing a given
functional series with a number series having nonnegative terms.
A number series
-1-30

^ M h = .1/, . i. ;1/* : . .. (8.15)

h—\
with nonnegative terms is said to be a d o m i n a n t s e r i e s for a fun­
ctional series

V Uh (x) = It 1 (x) -r u2{x)-\- . .. Mfe (x) -5• . . . (8.10)

h= I
on an interval a x ^ b if the inequalities
I (x) | k= 1,2,... (8.17)
hold for all x £ [a, 61 simultaneously. Series (8.10) is then spoken of
as a d o m i n a t e d s e r i e s ,
d ’eievstru n s* M - t e s t . If, for functional series (8.10) defined on an
interval la, 61, there exists a convergent dominant series of type (8.10)
the functional series is uniformly convergent on [a, 61.
Proof. Let an arbitrary e > 0 be given. Dominant scries (8.15)
being convergent, we have the inequality

V Mu < e
h=n4-l
for all sufficiently large n. By relations (8.17), for all sucli n the
inequalities
I y, uh (x) I < y |t/h (x )|< S (8.18)
fe=n-i-l *=-«+!

* Weierstrass, Karl Theodor Wilhelm (1815-1897), a prominent German

mathematician.
 CH. 8. FU NCTIONAL SEQUENCES AND SER IE S 327

hold for all x £ [a , b] simultaneously. Inequalities (8.18) indicate

that series (8.1li) is uniformly convergent and thus Weierstrass’
test has been proved.
Examples
~ 30

1. The series ^pAsin is uniformly convergent on the entire

7It=1
x-axi.s — oo <; x <; co because it can be dominated by the con-
vergent positive series yj -^r since we have
M=l
sin nx
/IT for — oo<Cx<Z oo
+°°
2. Consider the series V. , , J , .. ■ on the positive half of the
n—1
x-axis 0 < ! x O ; oo. Applying the well known techniques of diffc-
X 1
rential calculus we iind max 1 . Consequently,
1 - fr x "

1—n*^ for 0 < r < ; oo . The series 2 being convcr-

71 = 1
gent, we thus conclude, by Weierstrass’ test, that the series

■V
*—; —
1———
. n» uniformly• converges on the positive.r-axis
1 0 C x < - |- co.
»i—l
3. There is no convergent dominant number series for the
series 2 ( —1)« , 0 .<x <; -j- oo , since max ( —1)» == — and
n the
n U* x<+oo X
71- = l

|O
O
senes 2x’j 1 (the so-called harmonic series) is divergent. But, by
n=l
Leibniz’* test (e.g. see (8J, Chapter 13, § 5, inequality (13.80)),
the inequality
+ oo
1
2 JT-I k
h~n
X-|—/I

holds for every x £ 10, + ° ° ] and consequently the definition of

uniform convergence of a functional series (see relation (8.12))

* Leibniz. Gottfried Wilhelm (ir»4G-l71ti), the great German philosopher

anrl inatliemal ii iati.
:\2H MULTIPLE IN T E G R A L S , FIELD THEORY AND SERIES

^ / __ j ) n
implies that the series 2 j , converges uniformly on the entire
n~ !
positive half of the x-axis 0 ^ x < -r°° - This example shows that
Weicrstrass’ test provides only a sufficient condition for a functional
series to he uniformly convergent but not a necessary one.
Now let us proceed to formulate the basic (Cauchy) criterion for
uniform convergence which plays an important theoretical role
because, unlike Weicrstrass’ test, it gives us a necessary and suffi­
cient condition for uniform convergence and enables us to establish
more subtle sufficient conditions (compared with Weicrstrass’ test).
( f t . nrft y's* T e st ( f o r C n i f o r n t Convert,fence o f a'-Srf/tf c u ­
re). For a functional sequence (fn (x)} to be uniformly convergent to a
function f (x) on an interval [a, b) it is necessary and sufficient that
for every p. 0 there exist N — N (e) such that for each n -> N (e)
and all p 0 the inequality
1fn+r> {*) — A. (s) | < e (8.Id)
should hold for all x £ (a, 61 simultaneously.
Proof. Necessity. Let f„ (x) / (x) on la, 61. Then, given an
arbitrary e X ) , there is N (e) such that for all n .■> N (e) and all
p ;> 0 the inequalities
1I n { x ) — f { x ) | < 4 - and |/n ^ ,,(x) — / ( x ) | < 4 - -

hold for all x £ [a, 6|. Therefore we have | f n+j. (•*') — /„ (x) j-i^
< \fnu> (x) — f (r) | 1/ (x) — f n (*) I< 4* + 4 - = e for al1 11 > (f-)»
all p > I) and all x 6 l a, 6].
Sufficiency. If inequality (8.19) is fulfilled for all x £ (a, 61 it
follows Iha I for every fixed x £ (a, 61 the numerical sequence /„ (x),
n = 1, 2, . . is convergent because it is a Cauchy (fundamental)
sequence. Hence, the functional sequence /„ (x), n = 1, 2, . .
converges on the entire interval la, 61. Let the limit function he
denoted by / (x). Passing to the limit in inequality (8.19) as p
— -|-oo wo find that
I fn (X) — / (x) | < R
for all n > N (p) and for all x £ [a, 61 simultaneously. Put this
implies that /„ (j) / fi) on the interval (a, 61 and Cauchy’s
test has thus been proved.

* Cauchy, Augustin Louis (1789 1857), a famous French mathematician.

 CII. S. FUNCTIONAL SEQUENCES AND SERIES 321)

Applying Cauchy’s test for uniform convergence of a sequence to

the sequence of partial sums
S j (z) = // | (l), *S2 (z) — U\ (z) "I- IZj C^)t • • •» (j) =
= m, (x) 4- . . . un (x), . . .
4- -d o

of a functional series ^ ‘ ub (x) we arrive at

h=l
T e st ( f o r
( ' a u c / i f / ' x U n i f o r m Con.vevffen.ee of cl Series)..
A functional series

^ uk (x) = m, (x) —uz (x) -f . . -f- uh (x) -f- .. . (S.20)

ft—1
is uniformly convergent on an interval \a, b\ if and only if for every
e > 0 there exists N = N (e) such that for each n N (r) and all
p >> 0 the inequality

[ *^n+;j(x) — S a (x) J= | Uh (x) | = j U,n-j (x) -f- . . . -j- Un+p (x) | <C. E
ft=n-H
(8 .21 )
is simultaneously fulfilled for all x £ \a, 61.
On the basis of Cauchy’s test we can establish the following
AheUs* 'Test, ( f o r U n i f o r m Conver<jvnee o f a S e r i e s ) .
If the partial sums of a series
4 - DO

^ Un (x) - n, (x) tu (x) r . . . (8.22)

ft=I
are uniformly hounded on an interval fa, 61. i.c. if there is a constant
C, 0 <Z C <i -f oo, suck that
n
I$n (x) | ^ - | ^ Uh (x) | < C for n 1 , 2 , . . . (8.2.*»)
/l=s1
for all x 6 fa, 61, and if
ot, (x), a 2 (x), . . . a k (x), . . . (8.24)
is a monotone nonincreasing functional sequence uniformly convergent
to zero on the interval fa, 61, the series
-f.oo
^ rj.b (.r) Uh (x) fS.2‘>)
h—I
uniformly converges on fa, 61.

* AltuI, Niels Henrik (lSUlMiCiU), a famous Norwegian mathematician

 0 MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

Mofore proving A bels tesl let us consider an example ol its appli­

cation.
4. Wo take Ibe series ^ which can be regarded as the
*- 1
result of multiplying the terms of the series
-foo
V sin k\r —sin ,r - sin 2 x .. . 4~ n - • ••
is—1
by the members of the sequence
|’ i i ^ ) • • •» J_ » ... (8-27)
We have the inequality
I .s* (x) i X
(.z =^= 2.ttw, m — 1, 2, . . . )
Sill

for the partial sums of series (8.20) (e.g. see fSJ, Chapter lo, § 5)
and consequently the inequalities
\s» ( * ) | « - = const <C i-oo, « —1, 2,
sin
simultaneously hold for all x satisfying the conditions
2/n.T -f a (2/u i 1) n — a, (J < a < n,
m - U, d=l, ± 2 , ■ • - (8.2S)
Since (8.27) is n monotone decreasing number sequence converging
to zero it can he thought of as n uniformly convergent functional
sequence satisfying the conditions enumerated in Abel’s theorem.
Thus, on every interval determined by conditions (8.28) tbe serie s
«- DO

2 — satisfies the requirements of Abel’s test and lienee the

h~1
series is uniformly convergent on the interval.
Proof of Abel's test. To accomplish the proof we shall show' that
under the above assumptions series (8.25) satisfies the condition
of Cauchy’s test for uniform convergence. We have
^71+dhn 1 4" GCn, 2 ^'ri-i 2 ~
~1~ • • • “h t +pHn+p

= Gt/j + 1 I ^ n + I "f" & n + 2 I * ^ n i 2 + 1'

• *• & n + f> l*^w+p ' + == ®n+l*^n 4~

+ K +l — a n+2) Sn +1 4 • • -
• • • "1* ( « n +; ) - l ^n -jp) ‘S 'n + p -i ' h a ri+f>*-hi♦ /» ( 8 .2 0 )
 CH. 8. FUNCTIONAL S E Q U E N C E S A N D SEPTF.S 331

Making use of the inequalities otj (2 -) ^ a 2 (x) ^ ^ a n (x) ^

^ a nH (x) ^ . . . and the relation | S„ (x) | ^ C (fulfilled for
all n = 1. 2, . . . and all x 6 la. 61) we deduce from equality
(8.29) the relation
I ®n-dhi-fl 4- * • • “I” *■;/*«+j> I

< c {an+1 4 - (an+1 —«n+2) 4 - K: 2 —ccn+3) + . . .

• • - 4* (an+p-i — a n+;j) 4 a n+;>} - 2Can+1 ^ 2Cf„ m
where e„.t — sup a n+1 (x) —*- 0 as « —►oo> wliich holds for
all p > 0 and for all x £ |a. 6 1 simultaneously. Hence, under tin*
assumptions given in Abel’s tost, series (8.28) satisfies the condition
of Cauchy's test.

§ 2. PUOI’ERTIES UF UN 1FOHM t,Y CONVERGENT

FUNCTIONAL SEQUENCES AND SERIES
1. Continuity and Uniform Convergence.
T h e o r e m «V./. (A) If a sequence of continuous functions /, (x),
/ 2 (x), . . . , / „ (x), . . . uniformly converges on an interval |a. b]
to a function f (x) the limit function is also continuous on |a, 61.
(B) If all the terms of a series
-I.go
S (r) S uu U) (S-’iO)
h - \

are continuous functions on an interval la, 61 and ike series is uniformly

convergent on la, 6] its sum S (x) is also continuous on the interval.
Proof. (A) Take an arbitrary point x £ [a, 61 and let (x -[■ h) £
f In. 6|. We shall establish the continuity of f (x) at. the point, x.
For this purpose we estimate the difference / (x 4- k) — f (x). Sup­
pose we arc given an arbitrary e ^ 0. Lei us show that for all the
values of h sufficiently small in their moduli the modulus of the
difference is smaller than «. We have
| / (x + h) - / (x) I ^ I / (x 1 h) - fn (x + 6) 14-
4-1 / » ( * + h) - f n (x) | I /„ lx) - f (x) ( (8.31)
Taking a sufficiently large n wc obtain, by the uniform convergence
of /„ (x) to / (x) on (a, 6|, the inequalities
\ f ( x + h )—f n (x-f/i) [ < “!- for all (x 1 h) £ |a, b\ (8.32)
and
If {*) — fn (*) j < -j- for all x 6 fa. b\ (8.83)
Wc now lix the nhovc-choscn value of n and consider the term
| /„ (x -{■ h) — fn (x) | entering into the right-hand side of inequality
332 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

(8.31) , Since fn (x) is a continuous function, there is 6 = 6 (e) X >

such that for all h satisfying the inequality | k ( <C 6 (e) we have

\ f n (x + h) — f n (,x) \ C - j (8.34)

But then, by virtue of (8.32), (8.33) and (8.34), we derive from

(8.31) the relation | / (x -j- h) — / (x) | < g valid for all 6 satisfying
the condition | h \ <_ 6 (e), which means that / (x) is continuous
at the point x arbitrarily chosen on the interval [a , 6], Hence, / (x)
is continuous at each point x £ [a, 61, that is the function / (x)
is continuous on [a, 6].
It should be noted that if x is an end point of the interval [a, bI
then x can be given only a nonnegative increment if x = a and
only a nonpositive one if x = 6, and thus the above argument indi­
cates that / (x) is continuous on the left at x = b and on the right
at x — a.
71 -fo o

(B) Every partial sum Sn (x) 2] uk (x) of the series ^ uh (x)

h= i * = i
being a continuous function as a sum of a finite number of continuous
functions for any n = 1, 2, 3, . . we see that S (x) is continuous
because, by the hypothesis, the series is uniformly convergent and
hence Sn (r)z£ S (x) on la, 61 which implies, by (A), the continuity
of S (x). The theorem has thus been proved.
The condition of uniform convergence is only sufficient but not
necessary for the limit of a sequence of continuous fund ions to be
continuous. To illustrate this we can take example 4 in Sec. 1 of
§ 1 in which we (considered the sequence /„ (x) = j *
n — 1 , 2 , . . ., whose members are continuous functions. As was
shown, the sequence converges nonuniformly to the continuous
function / (x) - 0 on the positive pail of the x-axis 0 < i : < - r ° ° .
But there is a special class of sequences and series for which uni­
form convergence, as was proved by Dini*, is equivalent to con­
tinuity of the limit of the sequence or of the sum of the series.

T h e o r e m s .1' (J)ini*s Theorem , o n U n i / o v m C onvevijen.ee),

(A) If a sequence of continuous functions / n(x), » = 1 , 2, . . . , defined
on (a, 61** is nondecreasing, that i s f x (x) (x) ^ . . . < I/n (x) ..
on [a, 6l. and if f„ (x) converges to a continuous function f (x) the
convergence is uniform on [a, 61.

* Dini. 1‘lisse (1845-19181. an Italian mathematician.

** The condition that fa, b\ is a bounded and closed interval is essentially
used in the proof of this theorem which also remains true if [a, b] is replaced
by an arbitrary bounded closed set X,
 C1I. 8 . FUNC TIONA L SE Q U E N C E S AND S E R I E S 33^

(13) If the sum of a series S (x) = V uk (x) with nonnegativi

i»=i
continuous terms defined in an interval [a, 61 is continuous on [a, 6.
the series is uniformly convergent on the interval.
Proof. (A) Let us show that lor any e !> 0 there is n such that
0 < Rn (x) = / (.x) — fn (x) < c (8.35)
for all x 6 la, 61 simultaneously. Then, the sequence R n (x), n =
= 1, 2, . . being obviously monotone, that is
(*) ^ Rz (x) > • • • > {*) > • . . (8.36)
relation (8.35) must also hold for all sufficiently large ny which
implies the uniform convergence.
We shall prove the above assertion by contradiction. Suppose that
for a certain 0 there is no such n. Then for each n — 1, 2, . . .
there is xn £ [a, 61 such that
Rn (*„) > (8.37)
Applying the Bolzano*-Weierstrass theorem to the sequence of the
points xt, x 2, - - xn> . . . belonging to the interval [a, 61 we
can assert that there is a subsequence i , , xnjt, . . x„ft, . . .
convergent to a point x0 € [<*» 61. The function Rn (x) — / (x) —
— /n (x) (the difference of two continuous functions) is continuous
and therefore we can write, for every fixed m, the lelatiun
lim Rm (^njb) ' Rm teo)
nfc-»+aa
But for every m and any sufficiently large k we have n h ^>m and
consequently, by virtue of (8.36) and (8.37), we obtain
R m (^njj) ^ ^ nh ^
Passing to the limit in the last inequality as nh -foo we see
that R m (x0) ^ £o f°r any m ' But this contradicts the relation
lim Rm (x0) = 0 implied by the convergence of f m (x) to / (x)
^oo
at the point x0.
n
(B) The partial sums Sn (x) = >] Uh (x), n = i t 2.......... with
)i-i
nonnegative continuous terms uh (x) form a nondecreasing sequence
of continuous functions convergent, by the hypothesis, to a con­
tinuous function S (x). Therefore, by (A), the sequence converges
uniformly and thus the series is also uniformly convergent.
*Bolzano, Bernard (1781-184KV an ftnlHn n v » t , i fi'-ia:*.
334 MULTIPLE INTEGHALS, FIELD THEORY AND SERIES

2. Passage to Limit Under the Sign of Integration and Termwise

Integration of a Series. If

lim dc (8.38,)
7W+GO J / . (?)
*o
or

j { 2 »..(?)} S j* ® * (8.382)
X q k 1 l i = 1 Xfl

we say, accordingly, that it is permissible to pass to the limit under

X
the integral sign in the integral j fn (|) d% or that Ihe series
— x>
X0
T] uh (x) admits termwise {term-by-term) integration from x 0 lo x.
Relation (8.3S2) can be regarded as a generalization of the theorem
on an integral of a sum to the case of an infinite number of sum­
mands.
- >5
Replacing the functional sequence {/n (x)} by a series ^ uh (x)
h= I
for which it serves as the sequence of partial sums or, conversely,
-i-no
replacing the series ^ uh (x) by the sequence of its partial sums
A=l
we can easily transform relation (8.38,) to the form (8.32-j) and
(8.382) to (8.38,).
Hence, when investigating conditions for relation (8.38,) to be
valid we incidentally obtain the answer to the question of validity
of relation (8.38 a) and vice versa. It should be noted that, for rela­
tions (8.38,) and (8.382) to be true, the existence of the integrals
and convergence of the corresponding sequences and series are not
sufficient. This can be confirmed by the examples considered at
the end of the present section. For these relations to hold, an addi­
tional condition should be imposed. It turns out that such a sufficient
condition is (1) uniform convergence (this will be proved below)
or (2) convergence in the mean (which will be shown in § 0 of the
present chapter).
T h e o r e m 8.2, (A) If a sequence of continuous functions {fn (x) }
uniformly converges on an interval [a, b) to a function f (x), i.e.
fn (*) =£ / (x) on la, b] (8.30)

the sequence of integrals {

I fn (z) dz) uniformly converges {as a sequence
 CH. 8. FUNCTIONAL SEQUENCES AND S E R I E S 335-

of functions dependent on x) on the interval la, 6] to the integral

X

j / (z) dz, i.e.

*0 X X

[ /„ (z) dz ^ J / (s) (8.40)

*0 *0
for any x0 £ [a, 6].
(B) If a scries
— 30

S ( I ) = 5 « » w (8 .4 1 )
1
whose terms are continuous on an interval [a, M is uniformly convergent
on the interval we have the relation
x X

j S (z) dz = 2 ] uh (z) dz (8.42)

k = t x0

which means that series (8.41) can be integrated term-by-term within

the limits from x 0 to x for any x0 and x belonging to the interval (a, b\
and the series on the right-hand side of (8.42) uniformly converges
{with respect to x) on the interval |a, 61 for any x0 6 la> 61.
Proof. (A) Let there be given an arbitrary e > 0. Take N (e)
such that the inequality
| f n (*) — / ( * ) | < J — (^ -4 3 )

holds for all x t fa, 61 simultaneously when n I> A' (t). tho
choice of N (e) being possible by relation (8.34). According to
Theorem 8.1, the function f (x) is continuous as a limit of a uni­
formly convergent sequence of continuous functions. Therefore the
X

integral ^ / (z) dz exists for any x a and x belonging to la, 61. Let
*0 X X

us estimate the difference ^ /„ (z) d z — ^ / (z) dz. Dy (8.43), we

•V(| .v.»
have, for every n > N (e), the inequality
X

( f n (z) dz— j / (z) dz | j j |/„ (z) —j (s)l dz

xn .Vn

< j I f n ( 3 ) — f (-) I d z K 1 .1 — x 0 1 C t (8.44)

:3.% Ml.'LTI 1’I.R I N T E G K A U S . FIELD THEORY AND SERIES

which implies (8.40). It follows from relation (8.40) that

X x
lim
\ f n ( z ) d z = [ j (r) dz for x, x06[a, 6] (8.45)
n-»-foo J
*0 *0
Thus, if a sequence of continuous functions {/„ (x) } is uniformly
convergent on [a, 61 it is allowable to take the limit behind tlie
X

integral sign in the integral j* j n (z) dz for any x0 and x belonging

to |a, 61.
n
(13) The partial sum S n (x) = 2 uk (x) is continuous for every
h= i
n = 1. 2, . . . as a sum of a finite number of continuous functions.
13y the hypothesis, we have
Sn (x) =£ S (x) on la. 61
Then, by (A), we conclude that
X

f Sn (z) dz ZX f S (z) dz on [a, 6|

*0 *0
Observ e that
A ti JC 71 X

Sn (c) dz — f 2 Uh (z) dz — 2 \ u* <2) dz (•s.ir.)

*0 h= 1 h*=1xo
and t! lerefore relation (8.45) can he rewritten as
X

{ 2 j Uk w ■ =: :) dz on fa, 6] (8.47)
k= 1Xq *0
where the expression in the curly brackets is tho nth partial sum
of series (8.42). Consequently, equality (8.42) is true and Ilie series
on its right-hand side is uniformly convergent on la, 6), which is
what \vc set out to prove.
Note. The theorem also remains true when the functions j n (x),
n — 1, 2, . . may have discontinuities but arc integrablc. In such
a case the function / (x) is also integrable and relation (8.40) is
fulfilled provided that {/n (x)) converges uniformly.
The condition of uniform convergence is only sufficient but not
necessary for a series to admit term-by-term integra tion and for Ilie
passage to the limit under the sign of integration to be permissible.
For instance, the sequence j n (x) — xn converges nonuniforrnly on
the interval 0 < j < 1 to its limit
 CH. 8. FUNCTIONAL SEQUENCES AND SERIES 337

Cut at the same time

X X X

\ /„ (z)dz = j dz = J — o - j / (=) dz for » + oo

xn *o
and any x0 and x belonging to the closed interval |0. 1|. On the
other hand, there are eases when a nonunifornily convergent sequence
ot' integrnble functions Jn (x) converges to its limit in such a way
X .V

that Iiin
\ /„ (z) dz \ f (z) dz. For example, we have, for
/l-» x *
*r> -\\\
n — t ° ° . the relation
f„ (x) — 4nxV",,:vl —*■/ (x) = 0 , —oo < x < -j-oo
hut
i i
|| 4nxV ~ ” -v4 dx — 1 — e~n 7 ^ ^ 0 'd.r —0 as n- *~ i oo
o o
Passage to Limit Under the Sign of Differentiation and Terin-
wise DifVcrcnlialion of a Series. If
lim / ; ,( * ) - { lim /„ (x)}' (8.48)
M
—♦ oo n-*-.oo
or
-j-oo —
J-oo

I y, Ilk ( * ) ! '- V >‘u(x) (H.49)

h—1 U—1
we say, accordingly, that it is permissible tv pass to the limit under
the differentiation sign or that the scries V! m* (x) can he differentiated
a i
trnn-htf-tenn. \S

Relations (8.48) and (8.40) are equivalent in the same sense as

relations (8.88,) and (8.3o2)-
Fonnula (8.4b) can In* regarded as a generalization of the rule
for differentiating a sum to the case of an inlinito number of sum­
mands.
It turns out that ihe condition of existence of Ihe derivatives
and convergence of the corresponding sequences and series is not
sufficient for relations (8/i8) and (8./i9) to he valid and that some
additional requirements should he imposed to guarantee the vali­
dity. Sufficient conditions of this kind are given by
77t r a m t t S.lt. (A) !f ft sequence of continuously differentiable
iunctions* {/„ (.r)} ctmeerges to / (x) an la, ft I. i.e.
f„ (x) -+• f (x). .r £ Ia . ft I (8.50)

* A film tion / (r) is said to he canliuiioiisl \ iliffcrentiable if it i-<liftVriMifi.il>!:>

anil il*- derivative is a mill muons funriion
• J li - U S j
338 M U L T I P L E I N T E G R A L S , F I E L D T H E O R Y AND S E R I E S

and the sequence (x)} of their derivatives is uniformly convergent

to a function <p (x) on [a, 61, ie.
f\ (*). /a (*), . • f'n (x), • * * cp (x) on [a , 6] (8.51)
the function f (x) is also differentiable on la, 61 and
r (x) = <p (x) = lim /; (x) (8.52)
n -^ + o o

Hence, under these conditions, i£ is permissible to pass to the limit

under the differentiation sign.
(B) If a series
+°°
5 (*) = E (*) (8-53)
ft=l
with continuously differentiable terms converges on |a, 6] amf series
+ co

a ( x ) = 2 Uh(x) (8.54)
whose terms are the derivatives of uh (x), k = l t 2, . . . . converges
uniformly on Ia, 61 the sum S (x) of the former series is differentiable
on the interval (a, 6) and the equality
+°°
S' (x) = a (x) = 2 uh (x) (8.55)
h= 1
is fulfilled at each point of the interval. Thus, under the above
assumptions, series (8.53) can be differentiated termwise.
Proof. (A) By the hypothesis, the derivatives f'n (x) being conti­
nuous functions and the convergence being uniform on la, 61
(/» fc) T (t)), we conclude, on the basis of Theorem S.2, that
X X
lim I /n(z) d z = I q>(z)ds (8.56)
^0 ^0
i .e.
X

lim l/n {x) — fn (x0)I = f<p(z)dz (8.57)

Xo
Passing to tho limit in the left-hand side of equality (8.57) we
obtain
X

/ ( * ) —/ ( * o )= ( <p(z) &
and consequently

/ ( * ) = / ( * < > ) + j <p{z)dz (8.58)

*0
 CH. 8. FUNCTIONAL. S E Q U E N C E S AND SEKIES 339

Hence we see that the function / (x) is differentiable because it is

X t

equal to the sum of the constant / (x0) and the integral ^ <p (z) dz
which are differentiable functions. Now differentiating both sides
of equality (8.58) with respect to x we receive
/' (x) = tp (x) = lim /„ (x)
n-+ j-o©
n
(B) Putting Sn (x) = >], uk (x) we can write, by the hypothesis,
*=i
Sn (x) —►S (x) on la, 6]
and
S‘n (x) a (x) on [a, b\

and the function S n

’ (x) = 2.. a* (x) is continuous on [a, 6) for
every n = 1 ,2 ..............Consequently, by (A), S (x) is a differentiable
function on the interval [a, b) and the relation
S' (x) = o ( x ) = f u„ (X)

holds everywhere on [a, 61, which is what we set out to prove.

If a sequence of derivatives converges uoniiniforinly equality ($.52)
is not necessarily fulfilled. For instance, we have
/* (*) - ~ ln («* -1 y V . i 2— 1) - -
—>- / (x) = 0 for n — oo, — oo < i < l ^
but at the same time
lim /;(0 )= , lim ( I ) - 1 ¥=f (0) 0
n-> foo ^ \r -j- 1 |.x=.0 *

4. Term-by-Term Passage to Limit in Functional Sequences and

Series. Generally speaking, the well known theorem on a limit
of a sum is not true if the number of summands is infinite. Thus
for instance, every term entering into the series on the right-hand
side of the equality
+oo
knx
k sin ~l ’ I <C X Z, l ^izQ
►.=1
(proved in § 2, Sec. 5 of Chapter 11) and the sum of this series tend
to a finite limit as x —>■I — 0. But if we formally apply the theorem
on a limit of a sum to the above series for x —>- I — 0 we arrive
at an absurd equality 1 = 0.
22*
340 MULT II’LE INTEGRALS, FIELD THEORY AND S E R I E S

But the theorem on a limit of a sum can be extended to the case

of an infinite number of summands if some additional restrictions
are imposed. Namely, we have
T h e o r e m S.4. (A) Let a functional series
-J-SO
S (x) = uu (x) — u| (x) -r- u2 (r) uh (x) . . .(8.59)

converge uniformly in a neighbourhood of a point x0 and let

lim (x) = cft for k — 1, 2, . . . (8.GO)
A—*X(i
Then the number series ck is convergent and
h—\
-r- TX3 -J- 30

lim V Ch (8.G1)
.v“►
.\'o h -- 1 /t—1
which means that in a uniformly convergent series it is permissible
to perform a term-by-term passage to the limit.
(B) If a functional sequence /j (x), f 2 (x). (r), • - . is
uniformly convergent in a neighbourhood of a point xy and for every n
there exists a finite limit
lim fn (x) = A a

tinnumericalsequence .1,, A <>, . . ., .1n, . . . is also convergent and

lim lim fn (x) = lim lim /„ (x)
Y-* r,| 71 * 1 -w n 30 x —►A'n

Proof. Let an arbitrary t* > 0 be given. Series (8.59) being uni­

formly convergent in a neighbourhood of xy- there is *V (*■*) such
that for all n o .V (s) and all p 0 the inequality
I “fl-f1 (x) !- • • • r un.u,, (x) | < e («S.(>2)
holds for all x belonging to the neighbourhood. Passing to the limit
in inequality («S.G2). as x — x 0, wTe obtain the inequality
i cn f , -f - - - -!- cn+v | < t (8.G3)
which is valid [or all n >> .V (e) and all p > 0 . Consequently the
series cf. is convergent. Now. making the index p in (8.02) and
ii—t
(8.1)3) lend to infinity we derive the inequalities
co no
! y O fl^F . I V M ls (x ) |.< F (H.G'O
/, it-' I h n-' \
 CH. 8 . FUNCTIONAL S E Q U E N C E S AND S E R I E S 341

which are fulfilled for all n 2> N (e) and for all x belonging to the
chosen neighbourhood of x0. Let us fix an arbitrary n I> A (e) and
take 6 = 6 (e) such that the condition
n 71

| V Wft (x) — V Cft | <1« for 0 C | x — x01< 6 (?) ( 8 . 60 )

*= 1 k=z1

holds. Then, for 0 <C | x — x0 \ < 6 (e), we have, by virtue

of (8.(54) and (8.65), the relation
— CO 71 H

I >] u h (x) — y ck I< | v an (*>— y Ck |+

f<= l h~\ h— 1 ft— 1
-[oo -! oc
“ I -1 (X) | -• | y Ch | < K E~r R—0 6 (8 .66)
I h —u }-1

and hence part. (A) of the theorem lias been proved.

(13) This part of the ihoorem follows from (A) if we take the
series
fi {x) + I/-, (x) — fi (x)| + • - . -f [/n (x) — /«_i (^)l + . . .
whose partial sums form the sequence /, (x), / 2 (x), . . ., /„ (x). . .
nil the conditions of part (A) being fulfilled for this series.

§ 3. POVVElt SERIES
A functional series of the form
-f-oo
y cuxh -- c0 -r ctx -f- c2x2 + cnxu t- . . . (8.67)
ft=0
<jr of the form
-j oo
y Ch (x— x0)h^ C0 Cl ( X — X0) -i- ( \ (x— X0) 2 y ...
h--()

. . . [ - cti (x— x0)n \ . . . (8.68)

where ihe coefficients c0, c,, - - c„, . . . are constant mini hers
is called a power series. A simple change of variable of the form
x' = x -- x0 reduces series (8.08) to (8.67). Therefore in what follows
wo shall restrict ourselves to the series of form (8.07). The method
of representing a function in the form of a power series or, in oilier
words, rxpandini* a function info a poire r aeries, is widely applied
both in theoretical studies and in approximate ca Iculal ions. These
applications will bo discussed in more detail in § Here we are
going to investigate the basic properties of power series.
I. Interval of Convergence of Power Series. Radius of C o n \eig en ce.
W e sh all lirsl in v estigate Ihe structure of the domain of e o n \e i genre
342 MULTIPLE INTEGRALS, FIELD TH EO RY AND S E R I E S

of a power series. In contrast to general series whose domain of

convergence can be arbitrary sets of any complex structure* the
-rw
domain of convergence of a power series 2 cftx* *s always an interval
*=o
of the x-axis which can be a closed interval, a half-closed (half-open­
ed) interval or an open interval. This interval may also degenerate
into a single point (x = 0) or coincide with the whole x-axis. Every
power series 2 ck*h converges at the point x = 0 since at this
fc=0
point it turns into a numerical series of the form
Co H” C|*0 -|- c2-0 -f- . . . -J- cn *0 - r . . . = c0
There are power series which converge only at the point x = 0 .
"T°?
For instance, the series 2 n!xT'* is of this kind. Actually, for any
n=0
x =£ 0 we have
(a -f-1)! | x |n+i
lim «! 1x |»
lim (n -j-1 ) | x | = -J- oo
T1 n -»-l oo
and consequently, by the well known D'Alembert’s** test, the series
2 n\xn is divergent for x =7^ 0. There are also power series con-
n—0
+00
vergent on the whole x-axis, for example, the series 2 w^lose
n—0
convergence can be easily established for any x, with the help of
D ’Alembert’s test. Now let 11s take an example of a power serins
with a domain of convergence which neither coincides with the
entire x-axis nor degenerates into the point x = 0. Such is the
series 1 4 - x + x 2 4 . . 4 - xn + . . - (a geometric series) whose
- .

terms form a geometric progression with common ratio x. As is known,

this series converges for | x | < 1 1 and diverges for | x | ^ 1. Thus,
its domain of convergence is the finite (open) interval —i < i < 1
with centre at the point x == 0 . The following theorem indicates
that the above result is a special case of a general property of power
series.
Theorem If the domain of convergence of a power series
-f-OO
2 chxh neither degenerates into the point x — 0 nor coincides with

* We remind the reader that, by definition, 01 = 1.

** D’Alembert, Jean le Rond (1717-1783), a French philosopher and mathe-
 CH. 8 - FUNC TIONA L SE Q U E N CE S AND S E R I E S 343

the entire x-axis there is a finite open interval (—jR, /?), 0 <C R <Z +<*>
(termed the i n t e r v a l o f c o n v e r g e n c e o f the p o w e r s e r ie s )
such that the series is absolutely convergent at each interior point of the
interval and divergent at every point lying outside the closed interval
i - R , /?].*
The proof of the theorem is based on the following

-Lemma* If a power series 2 ChXk converges for x —a ^ O it

fe=0
converges absolutely for every x satisfying the condition |a r |< ;|a |.
+-
Proof of the lemma. The series 2 being convergent, it
A=0
follows that Ck<x -> 0 as k —*-+oo. Therefore there exists /! =
= const <C+ oo such that | | <1/I for all A:= 0, 1, 2, Now
IxI
let |x|<C |ot|. If we put q — -•—T-we obviously have 0 < ( ? < 1 arid
_ iii
hence | Cka,'* | = | c*ctfe |- —
GC I
for all A:=0, 1, 2, . . . . But the
-r30
series 2 Aqh converges as a geometric series with the modulus of
its common ratio less than unity and consequently, according to
the comparison test (e.g. see (8), Chapter 13, § 2), the series
-foo -Poo
2 |cft«£k| is also convergent. This means that the series 2 chXh
A=0
is absolutely convergent for the given x and the lemma has thus
been proved.
+®
The lemma implies that if a power scries 2 c***ch converges
A=0
for a value x=oc^= 0 it converges absolutely on the interval
-\-Oo
— | a | <; a:<C| ct |. In particular, if the series 2 is convergent
0
on the whole z-axis it is absolutely convergent there.
Proof of Theorem 8.5. Let us put R = sup |x'| where x' runs
througii the set of all the points of convergence of the series. We
obviously have R <C -\~oo because, if contrary, there would exist
points of convergence x ' with arbitrarily large moduli | x' \ and
hence the series would be absolutely convergent throughout the

* As will be shown later, if (—/?, /?) is the interval of convergence of

a power series 2 chxh the domain of convergence of the series may be the open
a^o
interval { —It , /{) or the closed interval f —It, /?) or one of the two half-open
intervals ( — It. It] ami I— It. H \
344 MU LTI PLE INTEGRALS, FIELD TH E O R Y AND S E R I E S

x-axis (by the above lemma), which contradicts the conditions of the
theorem. The dciinilion of the number ft suggests that the series
diverges for | x | :> ft. Let us prove that it converges absolutely for
| x | <C It. Take an arbitrary x with | x | -C It. Hy the definition
of the least upper bound oT a set of numbers, there is a point of
convergence x' such that | x | < | x' | < ft. Hut then the lemma
indicates that the series is absolutely convergent for this value of x.
Tlie theorem has been proved.
1he behaviour of a power series at an end point of its interval
of convergence is specified by the individual peculiarities of the
series. For instance, the series
1 -- x 4- x2 — . . . -| x" -p . . . (a)
T- 7*3 vfl
rn
x n (b)

(<-*)
and
fL .1 '
1 ~i~x •ri i ^1
32 *1 * • • I *» i W
have the common interval of convergence —1 < x <C 1. Series (a)
is a geometric series whose, interval of convergence — 1 < x < 1
has already been discussed and series (b), (c) and (d) are easily
investigated by applying D’Alembcrl’s test. Series (b) ((c)) converges,
by Leibniz* test, at the end point x — \ (x — —1) of the interval
and diverges at the other end point because it turns into the harmonic
series for x -= 1 (x - —1). Series (d) is convergent (by Cauchy s
integral test) at either end point of the interval of convergence.
1 •«
>
Thus, if the domain of convergence of a power series V Ch£k
k 0
is noL either the single point x — 0 or tin* whole x-axis there exists
a uniquely specified number/L 0 <C ft <L 0 0 , such that the domain
of convergence of the power series is one of the four intervals
(—ft. ft). (— ft. ft\. | - ft. ft) and (—ft. ft\. The number ft is referred
to as the radius of convergence of the power series.
It a power series is convergent only for x — 0 we put, by defini­
tion. ft =• 0 and if it converges throughout the x-axis wo write
ft --- -j-0 0 . Thjs convention makes it possible to apply IIn* notion
ol radius of convergence to all power series A powor scries and.
consequently. iN radius of convergence are completely specified
bv the sequence of its coefficients c0. ct. . . ., 1lieconclnd-
ing theorems of this section provide some methods for finding the
radius of convergence of a powor series from its coefficients.
 CH. 8. FUNCTIONAL SEQ U EN CE S AND SEIUES 34i>

T h r o v e hi S. 6‘|. If Mere exists a limit

lim 1f"*1,1 = / . *>()

Ic>' I
(finite or infinite) the radius of convergence of the series
2 is equal to /? —-j- ^the expression -J- is understood as being
k—n
equal to zero for I — oo ami as being equal to -f- oo i/ / U).
Proof. Applying D'Alembert’s test* to 1lie series wo linrl Unit

lim I cn -I | | * |><fI = | jt[ • lim

Ir n t |
I | | -r |“
7i * ; *30 71—
►-r» I cn |
= 1

If / — 0 we have | x [•/ — 0 and lienee the series converges absolu­

tely for any x in this case, i.e. II — -f-oo. If / = -f°° and x =^= 0
we have | x | • / — —oo and thus the series diverges for any x =f= 0,
i.e. R — o. Finally, if 0 <C / i 00 the series diverges in the case
/| x | I> 1 and converges absolutely in the case /) x | <C 1 and
consequently* R = -j- . The theorem has been proved.
The following theorem is proved after a manner of Theorem S.fq
by applying Cauchy’s root test.
'Throrem SJ>z* If there is a limit
lim y / \ c n\ — I, />-<)

“i ^
(finite or infinite) the radius of convergence of the series V r,hxh
h=.0
is equal to It = — ^where the expression -y is again understood as
being equal to zero for / = -) o o and as being equal to - oo for

Theorems S.f’q and <S.h2 only* apply when there is a limit (finite
or infinite) lim — or lim y ' \ c n |. The following theo-
))-► ; •> I' WI II- > - M
rein yields a more general result applicable to any power series.

* Whan l> Alembert's lest is applied to a series which is not positive «ui«v
must take the absolute values of its terms. The same refers to Cauchy”' r«»ul
I (*sl i is <*i | f<»r nrovinf* T lio riro m S O
346 MULTIPLE INTEGRALS, FIELD T H E O R Y A ND SERIES

T h e o r e m S ,03 ( th e C a u c h y - I I a d a m a r d * T h e o r e m ). The
+oo
radius of convergence of an arbitrary power series ^ chxk is equal to
h=o
= where / = lim ^r \cn\ (8.09)

where the expression -y- is put equal to zero in the case I = - r 00

and to -j- oo in the case 1= 0.
Xole. The symbol lim 7\/r\cn\ denotes the limit superior of
T»-*+oo _____ _____
the sequence of nounegative numbers |cf |, ]/ |c2|, J/~|c3 |, . . .
..., \ cn |, . . . . If the sequence is unbounded we put, by defini­
tion, lim j j /|c n| = + 0 0 . In case it is bounded the lim it superior
lim \ / |c re| is equal to the maximum of the abscissas of the limit
n->r=»
points of the sequence.
Proof of Theorem S.6j. Only the following three cases are pos­
sible here: (1) 0 < / < + oo, (2) 1= 0 and (3) l = + o o . We shall
separately consider each case.
(1) Let 0 <c I <z -1- oo. We shall prove that R = which is
+oo

equivalent to the following two assertions: (a) the series \ CkXh

0
I
converges for every xx satisfying the condition | X | | < — and (b)
the series is divergent for any x2 such that |x 2 | > - y .
(a) Let | xt , i.e. / 1xj | < 1. Then, given a sufficiently small
e>l>, we have (I e) | xi | = q < 1. The quantity 1 = lim J/' |c ri|
n~+-\-oo
being equal to the maximum abscissa of the limit points of the
sequence the inequality y ^ \c n\<Zl-\r £ holds beginning
with a sufficiently large n. Consequently, for all such n we have
v//r\ cn \ \X i\< .( l + e)\xi\ = q < l , i-e. |c n | |* i r < t f u
Therefore, by the comparison test, taking into account that the
-f-oo
geometric series 2 ?n *s convergent, we conclude that the series
n-=C
+oo +oo

Jc« I ! r t JM converges which means that the series V! cnx?

n —0 o
•converges absolutely.
* I . . l i w* S a l o mo n n865-!0C3h a noted Froncli mathematician.
 CH. B. FUNCTIONAL SE Q U E N C E S AND S E R I E S 347

(b) Let | x2| > - J -, he. J |x 2|; > l . Then For a sufficiently small
e, 0 < e < / . we can write (I— € )|x 2|>>1. The quantity I —
= lim | cn | being the (maximum) abscissa of a limiting point
of the sequence | | ), there is an infinite sequence of indices
/?i <C < 1 __ <C - such that n{ / Cnk > I—e, that is
n^ k n k |[x 2| > ( / — e )|x 2| > l and |cnfc 11x2 |" * > 1 - Thus, the
necessary condition for convergence of a series (requiring that
the general term of a convergent series should tend to zero) is
- 4-00

violated for the series 2 a°d hence it diverges.

(2) Let J= 0. We shall show that in this case /?= -{- oo which
is equivalent to the assertion that the series 2 CkXk is convergent
for all x, — oo < x<C-b c©. Take a value x0^ 0 . Since / = 0 =
— rim ;‘/T ^ n we have, beginning with a sufficiently large n, the
71—►-•-30

inequality ^ | cn | <1 which shows that for all such n we

can write J / fcTj I*o I < 4* and Ic,» | | x0|n < -~r ♦ Therefore, by vir-
+«
tue of the comparison test, the series 2 |cn ||x o |n converges and
-J-oo
thus the series 2 cnxo *s absolutely convergent.
»i—o ___
(3) Let I =z oo, i.e. let the number sequence be
unbounded. We shall prove that in this case /? —0 which means
-J.eo
that the series 2 diverges for any x0^=0. Suppose that the
n=0
-r »
series V cnx* is convergent for a value x0=?*=0. Then, by the above
n-=-0
mentioned necessary condition for convergence of a series, wc have
CftXj—►() as n —*■-f- oo and, consequently, there is a number A,
1 <C A < Z -j- oo, such that |c nx"|<C.4 for all n = 0, 1, 2...........
Therefore the inequality ^ { c n 11 x01^ y ^ A ^ A holds for all
n = 0, 1, 2, . . . and hence y \ cn | <C . A ■ which contradicts the
.___ IxoI
hypothesis that the sequence |c„J} is unbounded. The proof of
Uie theorem has thus been completed.
Note. The Cauchy-IIadamard theorem provides an approach
(other t h a n t h e onp applied in 3) lo proving t h e p n c c j p i i i t y
:v,s MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

of term-by-term differentiation and integration of a power series

because, on the basis of formula (8 -01 )), it can be easily verified that
the series obtained from a given power series by integration or
differentiation has the same radius of convergence as the original
series.
2. On Uniform Convergence of a Power Series and Continuity
of Us Sum. We have established that every power series is abso­
lutely convergent at the interior points of its interval of convergence.
We now proceed to investigate the properties of the power series
concerning their uniform convergence.
These properties are describecl by the following
-}•
T h e o v e tit &.7. Every power series ^ chxh converges uniformly
ft
on each closed suhinterval strictly contained within its interval of con­
vergence.
Proof. Let —It <C cl ^ x ^ (J» <T R where (—Hy M) is tlie interval
of convergence. We shall prove that the series is uniformly conver­
gent on the closed interval la, (J|. 'rake a value xQ max (| a |. |p|),
x0 € (—IE //)• Then we have the relation | x | <C ] jr0 | lor all x £
6 la. pi. Tliis implies the inequality 1c„xn | ^ | cfIa:{} |. I>ut the
r>o
number series ^ ^hX" is convergent and consequently, by Wrier-
tt=fl
-j-OO -J~c©
slrass’ test, the series ^ crixn and ^ | c(tx,x | converge iniifornily
n.-iO ?i^=0
on the interval la, fj]. Tlie theorem lias been proved.
Sole. A power series may not be uniformly convergent on the
entire interval or convergence. For example, the series
___!_______ I r .. J
,
t —X
■ 1 } X
y. >
r 00 • m •
1
I X I *• *

converges nonuniformlv on its interval of convergence — 1 <C x < 1

because the modulus of tin* difference between its sum and its
mb partial sum — 7 — 1 — x — x2 — . . . — xn can be made,
for any fixed w, to be arbitrarily large as x —►1 — 0 , and conse­
quently, the absolute value of the difference cannot remain smaller
than a finite number k 0 for all x belonging to the interval — 1 <C

On the other hand, the following theorem lakes place:

t-co
77# co re m V. « " l i a power series ^ e,.3:u converges at the end
h 0
point :r = It of ils in tonal of convergence (—IE H) it is uniformly
.
i- • • «1 l
r ^ , l ** 1 'c
w
f
i/a
•]
«/tc
-t.
!■ 1 ■ ' v .
.7 <, \.,.n 1 l a f>\
•*.
 CII. 8 . FUN C TIO N A L SE Q U E N C E S AND S E R I E S 349

Proof. We shall show that under the assumption of the theorem

the condition of Cauchy’s test, for uniform convergence is fulfilled
on the closed interval 10. R 1. This will imply the uniform conver­
gence of the series on [0, R I.
Let us introduce the notation
6’, a.i hi +p / r +i\ P * i } 2,
We obviously have
n n+1 S n 1
(«)
_
— c 2 c I■ « » + » « " « = s n. — s n> p-1
Let there lie given e > 0. Tlio number series V cJtRh being cou­
rt
vergent (by the hypothesis), Cauchy’s test for a convergent number
series suggests that there is N (f) such that we have
| Sn, h | <C « for all k 0, 1, 2, 3,
when n > .V (e). Taking into account that
jn -^ l for 0 and making use of (ct) and (P) we
obtain
\CnnX"" f W f4T . . . I Cn+pXn'P\ =

for all ti ^ A' (r), all p — 1. 2, . . . and simultaneously for all

.r belonging to the interval O^rr^T/?. which is what wo set out
to prove.
itoO MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

Note 1. We have a similar situation in case a series of the form

-r CO
2 chxh converges at the left end point of its interval of convergence
fc=U
(— R , R) or at both end points: in the former case the series is uni­
formly convergent on [ ~ R . 0 ) and in the latter it uniformly con­
verges on [ —R. /?!.
*r
Note 2. If a power series diverges at the end point x == R
h=O
of its interval of convergence (—/?, It) it cannot be uniformly con­
vergent on the interval 0 ^ x < R. For. assuming the contrary,
we would conclude, on the basis of the theorem on passing to limit
■+°°
in a uniformly convergent series, that the series V converges
le=0
at the end point x — R as well, which contradicts the original con­
dition.
The theorem below is a consequence of the theorems on uniform
convergence of a power series and of the fact that the terms of a power
series are continuous functions.
T h e o r e m S.S. The sum of a power series
+°°
S (x) = Chxn (8.70)
/l—0
is continuous at every interior point of its interval of convergence.*
Proof. If a point x lies in the interior of the interval of convergence
( - Jt, R) of series (8.70) the point can be embedded in a closed

Hig. tS.-'l QC O X p ft
■ _______« • 1 -1 1

interval la. p|. —R < a <. p <C R, strictly contained within the
interval of convergence (Fig. 8 / 1). Series (8.70) converging uniformly
on (he interval (a, PI, its sum is a continuous function on this
interval because the terms of the scries are continuous on fa, p|.
Hence, the sum S (x) is continuous at the point x £ [a, pi. The
theorem has been proved.
Note. If series (8.70) converges at an end point of its interval of
convergence ( - R, R) its sum is continuous at the end point. This
follows from the uniform convergence of series (8.70) on the corres­
ponding closed interval of the form I—if, 01 or 10, if I (see Theo­
rem 8.7,).

* Here and henceforward wc assume that the interval of convergence of

series (8.70) does not degenerate into the point x — 0.
 Cll. 8. FUNCTIONAL S E Q U E N C E S AND S E IU E S 331

3. Differentiation and Integration of Power Series.

T h e o r e m 8,9, Power series (8.70) admits term-by-term differen­
tiation at the interior points of its interval of convergence, that is its
sum S (;r) is differentiable, the relation
-j-oo

S' (x) = y kckxk~ 1 (8.71)

h^l
holds for each x £ (—11, R) and differentiated series (8.71) has the
same interval of convergence.
Proof. Let R and IV designate, respectively, the radii of conver­
gence of series (8.70) and (8.71). Let us first prove that R' = H.
* -f-ac
If x £ (—/?', /?') the series ^ k I ch || x |h-1 is convergent and
-t-
f-oo

therefore tlie .series V, k \ ch || x |h is convergent as well. Con-

1
+°°
sequently. by the comparison test, the series \ ch jjx |* also
ft:-0
converges and hence x £ (—R, R). Therefore R' ^ R. Furthermore,
if x £ (—/?» R) we can chooso x0 (j (—/?, R) such that the inequality
-30
1x0 1 Z> | x | (x0 0) is fulliHod. 'flic
series ^ chxJ| convor-
h=(»
ging, we have chxj —►0 as fc — -j oo. Hence, there is a constant
A 0 such that | | <C A for all k = 0. 1 . 2 , . . . . This
enables us to estimate the terms of series (8.71) ns follows:
X I t- 1 A ft- 1
| kc,lxh~ 1| = k | cux} | • < k (8 72)
~*0 IJo! Jo

For Ix 1< | x01, i.e. j < 1, the series k iI—x lft->

i conver-
J'o
1
ges, which can he easily established on the basis of D’Alembert’s
tost. Indeed, for /c-v - oo we have
A ft+t x X

-(<«*)
X
(* i 1) k <1
1Jo I ~*0 * Uol "jcT Jo Jo

Now applying the comparison test we conclude, by relation (8.72),

that series (8.71) also converges for the given x , i.e. x £ (—R \ I V )■
Consequently, R ^ I V . This result (together with the above inequa­
lity I V ^ R) implies that R' = R.
Now we can make use of the fact that series (8.70) and (8.71) with
continuous terms are uniformly convergent on every closed subin lar­
val strictly contained in the interior of their common interval of
convergence (—R, /?), which indicates that the conditions of Then-
3r>2 M U L TI PL E IN T EG R AL S. FIELD THBORV AND S E R I E S

rcm 8.3 on termwise differentiation of a functional series art* fulfilled

whence it follows that Theorem 8.9 is true.
-j-oo
C o r o l l a r y . The sum of a power senes S (x) = V ChX,h possesses
h=0
ihe derivatives of ail orders and

S {,,> (:r)^ k (k — 1) . . . (k — n - ljcvr"-". / i= 1, 2, (8.73)

h -rr
Resides, the radius of convergence of series (8.73) coincides with
that of series S (x) — N cfixh.
k—0
The proof of the corollary is based on Theorem 8.9 which is con­
secutively applied to differentiated series (8.71), then to the series
thus obtained and so on.
Tit core m S. to. It is permissible to integrate a power series

S (x) =-= chxh (8.70)

k --0
termwise between any limits of integration lying within its interval
of convergence. In particular, we have the relation
X -1-00
\ 5 (z) dz -Ch - ;-T . -* 6( - l i , K) (8.7'.)
U h ±0
Resides. the radii, of convergence of the integrated series and series (8.70)
coincide.
Proof. The terms of series (8.70) an* continuous functions and
hence the series can be integrated lenn-hy-terin on every interval
of uniform convergence, (liven an arbitrary point x £ ( —11. R).
we can embed it in a closed interval Ia. [11 strictly contained in the
interval (—R, R). Furlhermore, such an interval la, (11 can always
In; constructed so that it should contain the origin of coordinates,
l i e n e e , series (8.70) can in fact be integrated termwise. In particular,
•integrating series (8.70) from 0 to x on the corresponding interval
of type |a. |>| (—R, //) wo derive equal'll y (8.74) and hence the
proof has been completed.
A’vie. If series (8.70) converges at an end [mint of its interval
of convergence (—R. R) the limit of integration x in equality (8.74)
can he tnado to coincide with that end point because in this ease
series (S.7<0 is uniformly convergent on the corresponding closed
interval ( R. 01 or lU, //[.
4. Arithmetical Operations on Power Scries. Lei us first study
the operations of addition, subtract ion and multiplication. Consider
 CII. 8 . FUN CTI ON AL SE Q UE N CE S AND S E R I E S 353

two scries
-fM
/ ( 0 —a0 -. apr a2xz ~ .. . —a„ j ” &nxn (a)
*=o
and
"i'^
g (^) --- ; b\X t b2x2 - . . . — bnx" . . . - ^ bnx n (fi)
n=0
whose radii of convergence are, respectively, equal lo /i?a 0 and
> 0. Then
—>3
7 (c) zb g (•*) -• ^ (an ± bn) Xu for \ x \ < min (7?a, Jib) (v)
»=o
and

/ (.r) g (:r) = V (flo^'»-r«i^n-i-T- • • • -f«n&o)*n for | x | < min (/fat Kb) (b)
»i=0

Helation (y) is an apparent consequence of the corresponding

theorem on addition and subtraction of number series (e.g. sec 181,
Chapter 13. § 4). Helation (6) follows from the theorem on multi-
plicalion of absolutely convergent number series (e.g. see 181.
Chapter 13. § 4) since both series (a) and (p) are absolutely conver­
gent for | x | <C min (/?„, /?,,).
Finally, let us consider (lie operation of division. If f t
Ht, U and b0 =£- U we can write, for sufficiently small values
of | x |, a power series expansion of the quotient of series (*z) and ([>):
fiji j d|T *■ 4~ o nx n -f- . . .
= <‘o ' f'lT ; C2X-
f>0 b\.v - box'*—... b n * H r •
The coefficients c0, c,, . . ., c„, . . . of the expansion can be successi­
vely found by means of the recurrence formulas which are obtainoc
if wo perform the multiplication of the power series on the right-ham
side of the identity
fi0 j- a xx I- «aX2 + . . . s
— (&o b\X -f- b;>x“ - ( - . . . ) (c<> T ~j- Cnpcz -T • • •)
and compare the coefficients in like powers of x on the left-ham
-J-.v
and right-hand sides of the resulting relation. The series cnx'
n=r
can also bo obtained by dividing the series a0 j- «,x !- a2x2 . .
by the series b0 i bxx J b*>xr -j- . . . according lo the same rule
as those applied in dividing polynomials arranged in ascend in;
[mwers of the. variable. We shall not present here the proofs of tbes
assert ions.
:\5A MULTIPLE INTEC.ItALS, FIELD THEORY ANI» S E R I E S

§ A. EXI’A.NDlMi FUNCTIONS IN I'OWKII SFItIKS

Wc say that a function / (x) can be expanded in a power scries
\
y. cltxh on an interval (—r, r) if the series is convergent on Iho
k <»
interval and its sum is equal to / (x). i.c.

/(x)*= y Ct:Xh («S.7.“»)

h= I)
on the interval (—r, /•).* We suppose, of course, that the interval
(—r, r) does not degenerate into a point. We have already discussed
the importance of expansions of functions in power series and func­
tional series of other types (see the beginning of tlie presold chapter).
At the end of § 4 we shall present some characteristic examples of
applications of power series expansions for real values of Ilie* Nan-
able x. In $ 5 we shall consider some basic properties of power series
in a complex variable.
1. Key Theorems on Expanding Functions in Power Series. Expand­
ing Elementary Functions. First of all we prove that a function / (x)
cannot have two different expansions of form (8.75):
P h co v em *$'.//. A power series
-j oc
/(x ) = y* Oitf* (8.70)
/(=«
convergent on an interval (—It. It) (which dues not degenerate into
a point) is 'J'ft t / f o r ’s** .series for its sum f (x). that is the one
irfnis'e roefOernts are detenu ined htf Ta.ifl fir's f o r m it In
/«'*» (0) /• —n 1I * —
Cu —
k\ 01 (8.77)

Hence the coefficients of power series (8.76) are uniquely specified by

Us sum.
Proof. To prove the theorem wc can make use of the corollary
of the theorem on termwise differentiation of a power series. It follows
that the sum / (x) of series (8.75) is infinitely differentiable and the
cqunlit y
-foe
( *) - V k ( A - t ) . . . ( A- - n 1) o , n = I, 2, ... { 8 .7 8 )
h n

* A r u r i ' Uon / (x) w h i c h t a n h e expanded in a p o w e r series on an inlrrval

( — r. r) is said to h e an analjlu* l i m d i o n of t h e variable ,t on that interval.
** Taylor, brook (IGKa-l 731), an English umlhemalirian.
 CM. S. FUNCTION Ah SEQUENCES AND SERIES o•>
OO-

takes place. Putting z = 0 in (8.78) we obtain

/("> (0) - nlcn
ami, consequently.
/<"> (0)

which is what wc set out to prove. Thus, if a function / (z) can be

expanded into a power series convergent to the function Ibis series
is Taylor’s series for the function.
Now it appears natural to pose the question whether the converse
assertion is true. The problem can he stated as follows. Suppose
a function / (z) is infinitely differentiable on an interval (—R, R)
where It 0. Wc con formally construct the Taylor series for this
function:
V (») r (0) X~ — . / lw> (0) (8.7*1)
/ (0 ) 2| a!
1!
Now, does series (8.70) converge on the interval (—R, R) and will
its sum he equal to the function / (z) in case it does? It turns out
that in the general case the answer to the question is negative which
can be confirmed by the example of the function
i
e yi for x =r- 0
/ (x) ^
{ 0 for x = 0
(8.80)

In fact, it can he easily verified that the function is infinitely diffe­

rentiablethroughout the z axis and that at the origin we have
/ (0) = /' (0) = . . . = /<"> (0) = /<*+!> (0) — . . . = 0 (8.81)
C.onsequently. all the coefficients of the Taylor series of the function
are equal to zero. Thus, the Taylor series converges on the entire
z-axis and its sum is identically equal to zero whereas tlie f u n c t i o n
takes on a zero value only at the origin.
T h e o r e m S.l'i. A junction / (z) can be expanded in a power
series y chxk on an interval ( —11, R) ij and only if the junction / (z)
h—i*
possesses the derivatives of all orders on the interval and /Vn. Ihe
rent a note r o / t e r n t e r tits Rn in T a y l o r ' s fovtrt a h i,
I U) - / (0) /' (0) * - . . . -i — p x" H„ (8.82)
tends to zerofor every fixed x £ (—R, R) as n oo.
- oo

Proof. If a fund ion / (z) can he expanded in a power series y ct.xh

h -<>
on an interval (— R, R) it follows from the corollary of Thoornrn 8 P
23*
33li M U L T IP L E IN T E G R A L S , F IE L D T IIE O R V AND S E R IE S

Ihat / (x) has the derivatives of all orders and hence, by Theorem 8.11,
the equality / (x) = chxk can he rewritten in the form
k^O
; /tM>(0) „ (8.83)
I (x) •= / (») + T r X i-

Equality (8.83) suggests that the difference between the sum / (x)
and the nih partial sum of series (8.83) tends to zero, as n —►H-oo.
for all x £ ( /?, R). The difference being, by definition, the remaind*
or Rn of Taylor’s formula, the necessity has thus been proved.
Conversely, if / (x) possesses the derivatives of all orders on the
interval (—R, R) and if R n (the remainder in Taylor’s formula (8.82))
tends to zero as n -f-oo for every x £ (—/?, /?), wc have
/< * ) - [ / « > ) i - ™ x - . . . + ^ J(" ] | ^ 0
for n ►-r oo and each x £ (—R, R). Consequently, series (8.83)
converges on the interval (—R , R) and its sum is equal to / (x),
which is what we set out to prove.
Tlie following theorem yields some convenient sufficient conditions
for a function to have a power series expansion.
T h e o r e m S.IR. If a function f (x) has the derivatives of all orders
on an internal ( R, II) and if there exists a positive constant M such
that
I /<”> (x) | < M for n = 0, 1, 2, . . . and all x 6 (—/?, R) * (8.84)
that is if the family of the derivatives of all orders is uniformly bounded
on the interval ( fl. R). the function f (x) can be expanded into
a power series on ( —R , R ) m
Proof. The derivatives of all orders existing for the function / (x)
on the interval ( R , R), we can formally construct its Taylor series.
Let us prove that the series converges to / (x). According to Theo­
rem 8.12. for this purpose it is sufficient to show that the remainder
in Taylor’s formula (8.82) lends to zero, as n Too, for all x £
£ (—/?, /?). Applying Lagrange's form, of the remainder for Taylor's
theorem* we derive, on the basis of (8.84), the following inequality
for Rn:
j Kn | - for / / —0, 1 * € ( — //, R)
O H - 1 ) ! < 7 ^ ; i)i
( 0 < O C I) (8.85)

* Lagrange. .1oseph Louis (1736-1813), a famous french mathematician,

for Lagrange's form of Hits remaiiulor for Taylor’s theorem see. for instance.
 CIT 8. FUNCTIONAL SEQUENCES AND SER IES 357

Making use of D’Alembert’s test we can easily vorify that the series

(/i - 11)j! is convergent. Therefore, by the necessary condition

(ft ~
ti 0

for convergence of a series, we have (n 1)! 0 for n oo.

Hence, by virtue of inequality (8.85), Rn tends to zero, as n --oo,
for all x £ ( —R, R), which is what we set uuL to prove.
Lot us now proceed to consider some important examples of expand­
ing functions in power series, i.e. in Taylor’s series.
1. For the functions j (x) — sin x and / (x) — cos x we have
I /'"* (x) | ^ 1 for all n = 0. 1, 2, . . . and all x, —oo c x < --oo.
Therefore either function can be expanded in its Taylor series con­
vergent throughout the number axis.
Computing Taylor's coefficients ^ we obtain
x2n+l
in.r = x - - r + — a! - . . . - M - l ) " (2» •{-!)!
and
x2n
cos x I _____I ^___ jl. I _____ 1 _±__
x 2
2! t 4! 1 ' (2n)\
2. The derivatives of all orders of the function / (x) = ex satisfy
the inequality | / ‘,1‘ (x) | — ex ^ eR on an arbitrary interval
(—R t R). Consequently, the exponential function / (x) = ex is
expanded in a power series on every interval (—R, R) of the x-axis,
that is on the entire x-axis.
Computing Taylor’s coefficients we find
x rn
ex — 1 x • 21 ^ «j
3. To investigate the power scries expansion of the function
/ (x) In (1 -j- x) it is advisable to apply the following technique.
Differentiating with respect to x and expanding the derivative thus
obtained in a geometric series we find
/' (x) —~t1~-*—
J* — 1 —x 4- x 2- - x3 ‘r- . . . for — 1 < x < 1
Now, integrating this equality term-by-term we receive, on the
basis of Theorem 8.3, the formula
111 (I -|- x) = x -----f- —--------------j- . . . for — l < x < ! (8.80)
expansion (8.SO) also remains valid for x — 1. Indeed, by Leibniz’
test, series (S.S6) converges for x = 1 and therefore its sum
x'2
S ( x ) -- x ---- ^ —T
:l'i8 MULTIPLE INTEGRALS, FIELD TH KdHV AND S E t U E S

is continuous (see the note after Theorem 8.8) on the closed interval
10. 11. Consequently,
lim S (x)- 5(1) 1 ...
A - 4 - I- U ° 1

The function / (x) = In (1 4- x) is also continuous on that interval

and therefore In 2 = In (1 -+■ x) |x=1_0. Hut, according to (8.86),
we have the equality In (1 -r x) — S (x) for 0 ^ x <C 1. Hence,
hi 2 - lim In (1 — x) — lim 5 (x) = iS (1) = 1----t 'T““T-- r~~ • • •
\ *1-0 *-1-0 “ J ‘‘
4. The function f (x) = arc tan x can be .expanded in a power
series by applying the same procedure. Thus, differentiating and
expanding the derivative in a geometric series we get
/ ' (;r) = r ^ *1*^ = I x~ ; X 4 X* - p . * . for 1< X< 1

Performing termwise integration we find

a rc tan x = x — —
.t + -4-----
o -4-
/ -j- . . . for —
The validity of this expansion for the point x = 1 can be proved
after a manner of the above proof of the validity of the expansion
l,' 2 - I - - H - g ~ T + - - -
7). Let ns formally write Taylor’s formula for the function / (x) =
— (1 -- x)a where a is an arbitrary real number distinct from zero:
1 i OCX
*1* »)r2

I X * -:!). i ,
“ i « “ !)•••(“ —a ; I) h
m-------- J 2j -----------------it!----------------- 1
1 ■’
h i
For a positive integral a = n all the terms of series (8.87) from
the (n -f* 1) tli onwards turn into zero and thus we arrive at the
well known Newton binomial formula. Applying D ’Alembert’s test
to the case a n. n = 0, 1, 2, . . we can easily show that the
radius of convergence of series (8.S7) is equal to unity. Consequently,
outside the closed interval 1—1, 11, the function (1 x)a (for
a unequal to a positive integer) cannot he expanded into a power
series of the form TJ chxk, i.e. a series in positive integral powers
of x. Let us prove that inside the interval the expansion
ol (a —1) a ( a — l )...('x— A-r 1) h
(1 ;v)a - 1 ; a x i- VI = i i-2 ^ A! a
fe= t
 C.IJ. *. FI NCTIU.VAL SE Q U E N C E S ANI> SKIMES 359

is valid. According to Theorem 8.12, for this purpose it will suffice

to show that the remainder in Taylor's formula for the function
(1 r .r)"' tends to zero as n — ~-oo for all x belonging to the interval
( - 1 . 1).
VYc shall make use of Cauchy's form of the remainder for Taylor's
theorem’.
n„ • • (l /<»«■•. (O j), o < h < i
For the function (1 : x)rj this results in
J!„ „ 1 a ( a - I ) . . . ( a — >i)(1 =
} ( a - I) ( a — 2) . . (ct — n )
i — —1■■ ---------------------------- )" J- |ax(1 (8.8<l)

We consider the values x > — 1 and thus 0 <C ; , <Z I • Con.se-

(juenllv, we liuve 0 < *< 1 for oil w = 'l, 2, 3, . . . . Fur­
thermore, the quantity | ax \ (1 -}- Ox)a lies between the limits
|ct.rj(l— |.rj)a and | u.x | (1 + 1x |)a (independent of n) for all
.r£( -I. I) because we have the inequality 0 < C 0 < 1 . Finally,
the factor ——— ^ ‘ —") xn js j|lC (,orm Taylor's
n\
.series for tlm function (1 -j-j;)*” 1 whose convergence for — 1 <T
is easily established by D’Alembert’s test. Therefore,
for - I <; >• I. we conclude, by the necessary condition for con
vergencu of a series, that
a(a 1 ) . . . ( a — /»)
X V/
/<!
as n -*x*. Thus, the second and the thiril factors in the square
brackets on the right-hand side of relation (S.89) are hounded for
— 1 < .t < 1 whereas the first factor tends to zero as zi—►-f-oo.
Hence, /?„ —►U for every x £ (—1, 1) as n —»- -f-oo. This completes
the proof of the validity of expansion (8.88) on the interval (—1. 1).
2. Sonic Applications of Power Series.
(a) \ pf>{y i n y F oteev Sevieft to C o t n p r t t i n y I p j t r o x i -
■mate I’a / t t e s o f F a a c t i o n s . The scries
I T" < jit
eX I nr ( 8 .'JO)
J-3 .r" x -h ' 1
s m j: .i kT':' ^ t y\ ■■ • • • + < — ') 1)!
and
,.2h
cosx I— x»! — V -I- . . . — i - 1V4-(Hq\
- - '■ (Q “2)
300 MULTIPLE INTEGRALS, FIELD TIIKOItY AND Sl’.lUES

can be used for computing Ilie values ol <r\ sin x and cos x for any
values of x with an arbitrary accuracy since relations (8.90)-(S. 1)2)
bold throughout the x-axis for these functions.
Taking partial sums of scries (8.90)-(8.92) as approximate values
of the corresponding functions we can easily estimate the error
because, according to Leibniz’ test*, in the case of series (8.91) and
(8.92) the error docs not. exceed (in ils absolute value) the modulus
of the first discarded term.
Although Ilie series
x- x* h
111 (1 -- x) - x T } k\ (8.93)
T

for the logarithmic function is an alternating one it converges very

slowly. For the values x > I the series is divergent. To accelerate
the convergence of the series and apply it to computing logarithms
of numbers greater Iban unity we subtract the expansion
In ( I — x) = — x — — —— — . . . (8.93')
from expansion (8.93). This results in

1 "(f
= 9 ”2j (* +ir r-r+ •••) (8.94)

Putting x - — in (8.94) we obtain the formula

Hn-f- I
In fi +n 1 = In (n
' ~r 1)' — In n —
2 / I I v
= 'In -p I I 1 ~*~3 (2 h + lp 5 1)*“ '
(2/i-t- * ' )
(8.95)
Series (8.95) converges sufficiently fast, which makes it possible
to apply it to computing logarithms of natural numbers starting
from the value In 1 = 0 .
The series for the arctangent
arc lanx = x ------------------ ^—b ■ * • (8.96)
can be used for computing the number n with an arbitrary
accuracy. Namely, putting x = 1 in (8.96) we obtain

Since (8.97) is an alternating series, the error arising when its sum
is replaced by a partial sum can be easily estimated.

* E-S- 0 13 * r*
 CH. 8 . r i ' N C T I O N A I . SE Q U E N C E S AND SETUPS 3T,|

The series

(1 - X f , : I | V —(tt~ 'L j-j-t8 - * - 1) X’'

T(=l
can he applied to extracting roots. For instance,

r™- z y TTi ~. 2( i - r ±f =2\ 1- 14 - 4 ( 1) ...j

(8.98)
Since (8.98) is an alternating series, the error of an approximation
can be easily estimated. In particular, the sum of the terms written
in full on the right-hand side of (8.98) yields an approximation of
the root correct to the nearest ten thousandth.
(b) 1p p l ! /itlif f o t e e r S e r i e s K j ' p a n s i o n s to C o t n p n l i n f j
T n t r y r n f s J n e x p r e s s i b l e in T e r m s o f K l e m e n t n r y f u n c ­
t i o n s . As an example of such an application let us use series
expansion (8.91) of the sine for computing the integral
.t

S ix = I s d\ (the sine integral):

Jo s

(8.99)

It should ho noted that the division of series (8.91) by x is permissible

here for x 0 and therefore we have
Sill X
= 1 — — + - ---- •* (*)
:p 'I
ai
Putting -J— r- 1 at the point x - 0 we can retain equality (*)
for x — 0. The expression on the riglit-liand side of (8.99) is an
alternating series and hence the error of an approximation obtained
by replacing its sum by a partial sum can be easily estimated.
(c) A p p l y i u y 1*0 tee r S e r i e s to I nt e.yrnt i n y JJiffere.ut ia.l
f q n a t io n s . Power series and series in fractional powers of an
argument are widely applied for constructing solutions of differential
equations and introducing new classes ot* functions. For greater
detail concerning these questions the reader is referred to courses
on differential equations, e.g. 151. Here we shall restrict ourselves
to an elementary example. Suppose il is necessary to expand
A
into a power series the function V (a:) — e-A“ \ e^ d%. It can be
302 M tl.T lri.E IM EO IIA I.S, llK I.P TIIEOUN A\']> S E R I E S

easily verified that F(x) satisfies the differential equation

F' (x) + 2xF (x) = 1 (8.400)
with the initial condition F (0) = 0. Lei us try to lind the solution
of equation (8.100) satisfying the condition F (0) — 0 in the form
of a power series
-r JO

F (x) - V cnx" c0-f- c ,x 4- c«x2 4~ • • • -r Cn£u + • • • (8.101)

Substituting the series into equation (8.100) and equaling the

coefficients in the same powers of x on the left-hand and right-hand
sides of Ilie resulting equality we obtain
Cl =- 1, (n ~ 2) cnx 2 + 2c,t = 0, n - 1 , 2 , . . . (8.102)
The initial condition F (0) = 0 implies
c0 = 0 (8.108)
13v (8.102) and (8.103), it follows from (8.101) that
.V
( —I)f<l%us 1
F (x ) -- e ( 8 . 101)
1..5 . . . (2 n f)
n 0
Here we have first performed formal term-by-term differentiation
of series (8.1U1) hut now, us its coefficients are known, we sec? that
series (8.10-1) converges for alia:. —oo < i < -|-oo, and consequently
the termwise differentiation is in fact permissible for all the values
of x (sec Theorem 8.11).

§ 3. PUWKIt SKItlKS IN COMI’bKX AltGt'MbM

A sequence of complex numbers
s, — x, i\ju z2 = x2 -1- 0/2, . . ., zn = xn -h iyn, . . . (A)
is said to be convergent to a complex number z0 = x 0 4- iy0 as
n H-oo if
I3n — Zn \ = V ( xn — T0)2 - U/n — ;/o)' *-►o for n -j- oo
Therefore, for sequence (A) to converge to a number z0 — x(, -f- iya
it is necessary and sufficient that
x0 and //„ —►//„ for n 4*°°
A series with complex terms
 OH. S. F U N C TI O N A L SE Q U E N C E S AND S E R I E S 3r»a

where ah = a k — ifih. k = 1. 2, . . is called convergent if the

sequence of its partial sums converges.
The notions of absolute and conditional convergence as well as
Cauchy’s necessary and sufficient condition for convergence,
D’Alembert’s test and Cauchy's root lest are easily extended to the
series with complex terms.*
The domain of convergence of a power scries
c0 — ctz + c,z2 -I- cnztl t ••• (B)
where c0, Cj.......... c„, . . . are complex numbers and z — x -f* iy
is a complex variable is a circle (the circle of convergence) with
centre at the point 2 = 0 . The radius of this circle is the radius
of convergence of series (B). The circle may degenerate into the
point z — 0 or expand over the whole plane of the variable z =
— x -r iy. In the interior of the circle of convergence series (B)
is absolulely convergent. These assertions follow from the lemma
below.
L e m m a . If power series (B) converges for z — a 0 it is absolu­
tely convergent for every z satisfying the inequality \ z | <C | cc |,
i.c. in the circle of radius |a | with centre at the point z = 0 .* *
In every circle concentric with the circle of convergence and
strictly contained in it the power series converges uniformly and
ils sum is not only continuous but also infinitely differentiable.
Proceeding from the power series expansions of the elementary
functions of a real variable

eK I -p x + -Tjj--p • * • T * + ♦ • * (8.10.r>)

s i n . . . +(-1)*— •. • (8.100)
and

c o s x - 1 —-g- + ---------------------------- • ■■ (8.107)

we can define the elementary functions e' , cos z and sin z of a complex
variable z which coincide, respectively, with cx, cos x and sin x
for z = x. We remind the reader that series (8.105)-(8.107) are
convergent for all real values of x. Hence, by the above lemma, they
converge for all complex values of z if z is substituted for x into

* When applying 1)’Alembert's lest and Cauchy's root lest to a sprits whoso
levins are not. positive real numbers one should take the moduli of the terms.
** This lennna is a generalization of the one proved for power series in a
real argument x (see the proof of Theorem 8.13). The basic idea of the prooT
remains the same for the complex argument z — x -I- in.
3«/« m u l t i p l e i\*t e <;u \ i , s . K inil.n t h e o r y a n d sf .r ik s

tlie series. Therefore, pulling

; i z2 , , z» (8.108)
e 1rZ + 2! 1 *• * „i + ♦
-•2li-I
40
sin 2 — s ----Hf 4 - ■ • • + ( — !)* (5.109)
(2tt I)!
and
C053 2 — 1
(2k)\ ■h • • • (8.110)
. . . - i - ( — 1)

W(? obtain Uie corresponding functions of the complex variable

z ~ x ~r iy defined for all z. 13y the above lemma, the series thus
obtained are absolutely convergent. Therefore it is permissible to
perform the arithmetic operations of addition, subtraction and
multiplication on them. This enables us to establish the following
identities for the functions tr*, cos z and sin z of the complex variable z:
cz*•Cz* = C (8 .111)
and
cos2 z + sin2 z — 1 (8 . 112)
Substituting iz for z into (8.108), grouping separately the terms
containing and not containing i (after i2 has been replaced by —1
in all the lerins) and making use of formulas (8.109) and (8.110)
we arrive at Euler’s formula
eiz = cos z 4- i sin z (8.113)
valid for every complex z. Indeed, we have
a . . , *222 3-.3 I •j4 •J'-t I
l*>3' I . • . i-2
* *r 2 I•t223
1 | , : ' TT — r • • • — 1 -r *3 i—r j( "3T
O'2)2
-j- ‘II
= cosz 1- i sin z
Helations (8.109) and (8.110) implying cos (—z) = cos z and
sin (—z) = —sin z, we obtain (by substituting —z for z) the relation
e~i£ = cos z — i sin z (8.lid )
From relations (8.113) and (8.lid ) we derive
eiz A-e~lz —e~iz (8.1 IT,)
cos z —---- and sin z Ti
The last formulas are also spoken of as Ruler's formulas.

* Kquality (8.111) results from multiplying the power series for e*1 and
c2* according to Lin* rules IhnI. have been proved for a real argument hut remain
applicable in i h n <•?«<*< of ;» rnmnlev arrnirnent as well.
 CH. 8 . FUNCTIONAL S E Q U E N C E S AND S E IU E S Hlia

It follows from formulas (8.115) that the functions cos z and sin z
can assume arbitrarily large values in the complex plane. For
instance, putting z — —in where n is a natural number we obtain
£rl
cos( — in) —---- ►-J- 0 0 for n —►— oo
But nevertheless formula (8.112) remains true in this case.
Power series in a complex argument make it possible to introduce
many other functions de lined in domains lying in the complex plane
such as In (1 — z). arc tan z and the like.
The theory of functions of a complex variable is one of the most
important divisions of modern mathematics and is widely applied
in mathematical physics.
Functions of a complex variable unable ns to elucidate many
paradoxes of the theory of real functions. For instance, the left-hand
side of the equality
yq—t = 1 —z z + x x— (*)
is a bounded continuous function defined over the whole real axis
whereas the series on the right hand side diverges for \ x \ .
But if we consider the equality (*) for the complex values of x
we sec that the left-hand side of the equality turns into infinity
for x = i and hence the point x = i must lie on the circumference
of Ibe circle of convergence with centre at the origin. Indeed, if the
point x = i were placed inside the circle of convergence the function
-j— r would be continuous at the point, which is impossible since
it approaches infinity at this point.
In § i wo considered the function

/ \ \ e xi
T (>t ) = i
f°r
lO for jr--0
which lias derivatives of all the orders with respect to x at the origin
hut cannot, he expanded into a series in nonnegative integral powers
of x. This fact can be explained if we investigate the function
__i_
q (z) —c 22 {z =/- 0)
under the assumption that z can take on all the possible complex
values. Substituting z /// we obtain c = c*2 —- -j-oo for
_ r
// - >- 0 whereas c Y’ - 0 for x — 0- Consequently, the definition
of this function cannot be extended to the origin so that, the function
should become continuous. If a power series convergent to «(' (x)
 MULTI 1*1.1J I N T E G R A L S . FIELD THEORY AND SERIES

on an interval —f t < Z x < Z R existed the substitution of z for x

would result in a power scries converging to <p (z) in the circle J z \ <Z
< li and the function cp (z) would be continuous and even diffe­
rentiable with respect to z at the point z = 0, which contradicts the
fact that it has a discontinuity at that point. An exhaustive investi­
gation of this example can only be performed by applying the theory
of functions of a complex variable.

§ G. CO N V ERG ENCE IN T H E MEAN

In some divisions of mathematics and its applications a measure
of closeness of a function / (x) to a function g (x) is interpreted in
an “integral” sense. In such an interpretation the functions may
bo regarded as being close to each other despite the fact that the

absolute value of their difference / (x) — g (x) takes on large values

at separate points. Usually the so-called mean square deviation
is taken as such a measure, which leads to the notion of convergence
in Ike mean.
t. Mean Square Deviation ami Convergence in the Mean.
D e f i n i t i o n / . Given two functions f (x) and g (x), the nonnega­
tive quantity
b
g) - j I / (*) - g <*) |2 rfx*
P5 ( / , (8 .1 1 (i)
a
is called the m e a n s q u a r e d e v ia tio n of the function f (x) from
the function g (x) on the interval la, b].

* In § G wo suppose th a t all the functions mu lor consideration are integrablo

(in the ordinary sense) although most of the assertion:* and notions introduced
here also remain valid for the square intograble (siumnnble) functions (i.e. the
ones for which the integrals of the squares of their moduli exist as proper or
improper integrals).
 CIT. R. FUXCTf<»NAL S n n r F X C F S A ND SEMIES 307

We apparently have
P2 (/, S) = P2 (S, /)
The graphs of Uvo functions / (x) and g (x) which arc close to each
other in the sense of their mean square deviation may considerably
diverge at separate points (see Fig. 8.5).
f> efinit»ou A functional sequence
fi (^)i ft W* . . . . fn (x)> . . . (S .ll/)
is said to be eon re rye a t in the m e a n / i n the m e a n s q u a r e )
to a function f (a:) on an interval fa. 6) if
h
c"- { / . . / ) = i I /» (*) - / ( 0 e rfjr — 0 for >1 + oo (S. 118)

This kind of convergence will he designated by the symbolic relation

lim /„ (x) = / (x) on la, 61 (8.1 111)
n-*-‘ -jd
d efin itio n A functional series
-f-V
Hfc (x) (8.120)
h t
said to eon eeeqe in the m e a n ( s q u a r e) to S (x) on an interval
, 61 if the sequence of its partial sums
n
S.. I t ) -
v uh (,r). » - l . ^ * (8.121)
• ••

<<-•1
convergent in the mean to S (x) on Uitt interval la, 61, that ts
hI
n
12
>a(S (x), Sn ( * ) ) - f \ S { x ) - - 2 ] uh (x) ~ dx —►O as n-*- 1 oo (8.122)
•<1 * /<*i
fn such a case we shall write
-hoo
S(x) = uu (x) on [a , b\ (8.123)
h- -1
2- Cauchy-Bunyakovsky* Inequality. If two functions f (x) and
g (x) fulfill the above requirements on [a, 61 they satisfy the C a n -

* Bunyaknvskv, Victor Yakovlevich (1804-1839), a Russian mathematician.

This inequality is also spoken of as the Cauchy-Scluoarz inequality although
Runyakovsky called attention to it earlipr than Schwarz (in 1859).
3<i8 MULTIPLE IXTKUKALS, FIELD THEOHY AND SEUIES

r l i tf - ft u it t / a h o vsf» t f - S r h tr<t i'Z I n < ' q t t a f i t * /

] \ / (•*) H {*) dx
Z
j /- (.r) dx ♦ | /
/ b

\ g1 ( X ) dx (8.124)
i
Proof. Let us pul
b I, b
.1 = ^ f- (x) du\ B — ^ / (.r) g {x) dx and C = g~ (x) dx (8.12b)
« a a
arid consider the two cases which are possible here: (1) .4 = 6 * — 0
and (2) at least one of the numbers A and C is different from zero.
(1) If A = C = U. i.e.
b h
f /* (x) dx --- f g2 (x) dx 0
% v
a a
il follows, on the basis of Ilie obvious inequality
I/ (x) g ( x ) | < 4 - \ P {X) r g - (X) |
llia l

b h b

j I/ (*) g (*) Idx 1 [ ]' r- dx .• j g* U) dj j 0

a a

1>tiI we have
U ci
J ^ / (x) g (.? ) dx < l[ | / (x) g (x) | dx

and consequently />‘ = j / (x) g (x) dx — 0. Therefore inequality

a
(8.124) is fulfilled in this case because its right-hand and left-hand
sides are equal to zero.
(2) For definiteness, let A 0. Then we apply the following
technique. Taking a scalar parameter X we can write the relation
l>,/ (x) + g (* )!- > 0
which holds for all real values of the parameter. Thus, wo have
V-p (x) b 2If (x) g (x) g* {x) > 0
Integrating the last inequality with respect to x from a to b and
making use of notation (8.12b) we conclude that the inequality
,1a2 -b 2B k + C > 0 (8.12(i)
 CH. 8. KLW'CTIONAL SEQUE XCE S AND SERfES 369

(in which .1 >> U) holds for all real values of X. (Consequently the
quadratic trinomial /IX2 — 2Bh C cannot have two distinct
real roots Xj <C X2 because if otherwise it could be represented in
the form A (X — X,) (X — X2) and hence would assume negative
values for X satisfying the condition X! <X<CX2, which contradicts
inequality (8.126). But, as is well known, for a quadratic trinomial
to have no distinct real roots it is necessary and sufficient that the
discriminant of the trinomial be nonpositive:
B2 - A C < ^ 0 (8.127)
Transposing the product AC to the right-hand side of the inequality
and extracting the square root of either side we obtain | B |
^ \r | A (• \ ' | C |. Taking into account notation (8.125) we thus
arrive at the Cauchy-Bunyakovsky inequality.
3. Integration of Sequences mid Series Convergent in the Mean.
T h e o r e m S .1 4 t. If a functional sequence /, (x), / 2 (x). . . .
(a:), . . . converges in the mean to a function f (x) on an interval
la, bl we have, for any x0 and x belonging to la, b\. (he relation
X X

liin f f n (s) d z = [ / (z) dz* (8.128)

*0 *0 X

Moreover, for any x0(r[ai6]> the sequence { [ fn ( * ) ,

*0
n = 1, 2, . . . {regarded as a sequence of functions dependent on x)
X

uniformly converges to \ f (z) dz, i.e.

X
*0 X

j f„(z)dzZZ- j f(z)dz on |a, b] (8.129)

Proof. We have, by the hypothesis, the relation

b
ps </. /-.) = f l/0 O -/n ( z )] J<2z-*O, for « -*■ -f oo (8.130)
a

To prove (8.129) (and, consequently, (8.128) as well) we shall

X
estimate the integral ^ [/ (z) — fn. (2)1 dz. For t his purpose we rep-
*0

* We remind the reader that all the functions (including fn (x) and / (x),
considered in § 0 are supposed In lie inlegrable on [a, h| in the ordinary sense.
See also footnote on page
2 /. —0824
:J 7 0 MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

resent the integrand as the product

! / ( * ) - / » (Z)| = 1 [ / (Z)-/„(2)1
and apply the Cauchy-Bunyakovsky inequality:
X r X / ~x

Ij [/ (z) — In (z)l j/ j 12 d z * y j |/(z ) — fn (z)\2dz <

xo vo xq

^ V b —a - f ( z ) —fn (z)
a

= (/, /n) (8.131)

The right-hand side of inequality (8.131) being independent of x and
tending to zero as n + ° ° (by relation (8.130)), we can write
X

j (/(z) = /n (z)] d z ^ O on [at b]

*0
which means that relation (8.129) is fulfilled. Thus, the proof of the
theorem has been completed.
If we take the modulus of the difference | / (x) — /„ (x) j instead
oT / (x) — fn (x) and put x0 = a and x = b inequality (8.131) is
replaced by the relation
b
( | / W - / » (2) | d z < V b ^ a . Vp* (/, /„) (8.132)
a

The integral on the left-hand side of (8.132) expresses the area lying
between the graphs of f n (x) and / (x) and bounded on the left and
on the right by the respective vertical straight lines x = a and
x — b. Hence, in case fn (x) converges in the mean to / (x) on the
interval la, 61 the area tends to zero since the mean square deviation
P2 (/» fn) approaches zero. But at the same lime the maximum
deviation of fn (x) from / (x) on [a, 61, that is the quantity
sup | f n (x) — / (x) |, may infinitely increase since the values
n <x ^ b
of fn(x) and f (x) may considerably differ for any n at some separate
__t_ L n\i
points. For instance, the sequence fn (x) = n 8 2nx e 2
converges in the mean to / (x) s 0 on the closed interval 0 ^ x ^ 1
but
m a x \ f n (x) — /(x ) | (— ) = ( / - n s -> -- oo for n~+-r oo
 CH. 8. FUNCTIONAL S E Q U E N C E S A N D S E R IE S 371

Taking a series convergent in the mean we can apply Theo­

rem 8.14, to the sequence of its partial sums and thus obtain the
corresponding theorem concerning series:
-|~oo
T h e o r e m X.JJ?. If a functional series 2 (*) ivith integrable
terms converges in the mean to an integrable function S (x) on an
interval [a, 61 the relation
X X X

j S (z)dz— j k, (2 ) dz -{-• •• -r J un {z)dz-{- . . . (8.133)

*o xo *0

holds for any x0 and x belonging to |a, 6J and series (8.133) uni­
formly converges to its sum on [a, b).
Proof. By the hypothesis, Sn (x) converges in the mean to S (x)
n
on [a, 6} (where S n (x) = 2 «k(x))and therefore, by Theorem 8.14,,
*=i
we have
X X

^ S n (2 ) dz ZZ- j & (z) dz on [a, 6] (8.134)

*0 *0
But
n x

j S n (z)dz = 2 j uh(z)dz
X0 h—1 sc0
and consequently
n *
{ S j Uk (z) I t j S (2) dz
k—1xo xo
Hence, the sequence of partial sums of series (8.133) is uniformly
X

convergent to the function J S{z)dz which is what we set out to

xo
prove.
4. Connection Between Convergence in the Mean and Term-by-Tcrm
Differentiation of Sequences and Series.
Theove-m If a sequence of continuously differentiable
functions {fn (x)} is convergent in the mean to a function f (x) on an
interval fa, bI and the sequence of the derivatives (x)} is convergent
in the mean to a continuous function q> (a:) on (a, 61 the function / (x)
is differentiable on fa, 6] and
/' (r) = <P (*) = lim /; (x) on [a, 6) (8.135)
n—
>-}-oo
24*
372 MULTIPLE INTEGRALS. FI EL D THEORY AND SERIES

Proof. We have, for x, x0£[a, 6], tJie inequality

X X

f n ( * ) — f n ( X 0) — f M ( - ) * I= f [fn ( - ) “ (-)] dz
x0 *r>
/ x
* j

< | / j I’ r f r - y j |/;,( j )-< P (* )!* * <

*0 x0
< V / 6 —a •W 2(/«, <p)-► 0 (8.136)
as rc—*- oo. Consequently, passing to the limit and taking into
account that f n (x) f (x) and /„ (x0) —>*/ (x0) as n -{- oo we obtain
X X

/(* ) —/ (*o) = ( <p(z)ds, i-C. / (x) = / (x0) + j (p (z) ds (8.137)

It follows from equality (8.137) that / (x) is differentiable and that

equality (8.135) takes place. The theorem has been proved.
We similarly prove
Theorem «V.i.52. If a functional series
4-00

V 1/ 1, (x)
S ( x ) = (8.138)
*= 1
with continuously differentiable terms is convergent on Ia, 6] and the
series
o { x ) ~ ^ Uh (x) (8.139)
k=\
converges in the mean to a continuous function a (x) the sum S (x)
of series (8.138) is differentiable on la, 61 and

S' {X) = a (X) =L- S Hi, (x) (8.140)

fc.-I
The proof of the theorem is left to the reader.
5. Connection Between Convergence in the Mean and Other Types
of Convergence. If a sequence of functions /j (x), fz (a:). . .
f n ( ^ ) , . • • converges (in the ordinary sense) at each point of
an interval [a, 61 this does not implv, in the general case, its con-
---- —- —Ji 71
vergence in the mean. For example, we have fn (x) = V 2ttx e —►
—►/ (x) ^ 0 on the interval [0, 11 as n —*- -T°° whereas
i i
J [fn (o') — / (r)l2 dx = ^ 2nxe~uxi dx -=(1 — «TM ) 1 for n — -4- co
o n
 CH. 8 . F U N C T IO N A L SE Q U E N C E S AND S E U IE S

But if a sequence {fn (a:)} uniformly converges to a function / (.

on (a. 6) it necessarily converges in the mean to the same functioi
In fact, if for any e r>»■
0 —there is AT(e) such that the inequalit
I f n (x) — f ix) I < j / is fulfilled for all n >• TV (e) an
all x 6 [a, 6] we can square both sides of the inequality and integral
the result, which yields the relation
h
P* (/, I n ) = j | I n ( * ) - / ( X ) |* ( / » < — (b -a ) = 8
a
valid for all n Z> N (e). lienee p2 (/, /n)-^ 0 , as n —>■ -|-oo. whic
implies that / n (x) is convergent in the mean to / (x) on [a. 61.
Uniform convergence does not follow from convergence in th
mean; moreover, a sequence convergent in the mean on an interve

F'K- fe) ( fj
may even diverge at each point of the interval. As an example, lei
us construct a sequence of functions /, (x), j 2 (x). . . ., /„ (x), . .
convergent in the mean to f (x) ~ 0 on the interval 10, 11 which diver
ges at every point of the interval. To this end. we lirst divide tlu
interval 10, II into 2 equal parts and define the functions h (x \
and / 2 (x) as is shown in Fig. S.6a and b. The graphs of the functions
are <?iven in h*»avv li^oe the «=rnn1! nrr? indicate that tliv, cOiico
374 MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

ponding points do not belong to the adjoining pieces of the

graphs.
Next we divide the interval 10, 1] into 22 equal parts and define
the functions / 3 (or). / 4 (x). fh (x) and /* (a-) in a manner shown in
Fig. 8.6c, d. e and /. Continuing infinitely in this way we obtain
a functional sequence /, Or), / 2 (x). . . fn (x), . . . whose every
member (x) takes on only two values, namely 0 and 1, on the
interval [0, 1J. Therefore /?, (x) _ /„ (x) for all n — 1, 2, 3, . . .
on the interval 10. 1].
Now let us prove that /,, (x) converges in the mean to / (x) = 0
on 10, 11. We have
l i i
p2(/. fn) = f 1fn (x) — / (x) I4 dx = f fn (x) dx = f fn (x) dx 0
0 0 0

1-1
for n —►-j-oo. Indeed, the integral j fn (x) dx is equal to the area
o
shaded in Fig. 8.0, and when n -j-oo this area obviously tends to
zero. Convergence in the mean has thus been proved.
On the other hand, the functional sequence /, (x), / 2 (x). . .
fn (x), . . . thus constructed diverges at each point of the
interval [0, 11 because for every fixed point x belonging lo the
interval and for an arbitrarily large N > 0 there are functions fn>(x)
and f„- (x) with n \ n ” .V such that the former assumes the value
/w* (x) = 0 and the latter the value f n- (x) — 1 at llu* point x.

APPENDIX I TO CHAPTFU 8
CKITKIUUN lO ll COMPACTNESS OF A FAMILY
OF FUNCTIONS

The notion of compactness of a family of functions appears in

mathematical physics in connection with theorems on existence of
solutions of differential and integral equations and on convergence
cf various approximating processes for evaluating solutions of
such equations. But the definition of this notion does not involve
any concepts connected with Ilie theory of differentia! or integral
equations and is closely related to the questions discussed in the
present chapter.
D e f i n i t i o n /. .1 fnmifn rif functions {/ (x) } defined on a set
X of points x is said to be e o tn jta c t (in the sense of uniform con­
vergence) if for ever}/ in finite, sei/uence of functions /, (x). / 2 (x), . . .
• • •• in (x), • . . i/cthe f a n : ifIf ther* '•r»>/•.• n whvnonenre
 CH. 8 . FUNCTIONAL SEQUE NCE S AND S E U IE S 375

/„, (*)» /n, (*), • • •» fni( (z), • * • which is uniformly convergent on

the set X. *
JJcfin it ion A family of functions {/ (x)} defined on a set X
is said to be u n i f o r m h f b o n m i n i on this set if there is a constant C,
0 < C < + ° ° such that the inequality [ f (x) | ^ C is fulfilled for
all x £ X and all functions f (x) belonging to the family.
D e f i n i t i o n X. We say that a family of functions {f (x)} defined
on a set Ar is eqn i f o u t in non ft on the set if, given an arbitrary e > 0 ,
there exists 6 — 6 (e) 0 such that for every function f (x) £ {/ (x)}
and any z', x" £ X satisfying the condition \ x' — x" | < 6 (c)
we have the inequality
I / (*') — / {x”) | < e (1)
where 6(e) is independent of the choice of x' and x " and is determined
solely by the value of e.
T h e o r e m t ( A r z e l a ' s C r i t e r i o n f o r C o m p fic t ncsis** p
If a family of functions {/ (x)} defined on a closed interval a ^ x ^ b
is uniformly bounded and equicontinuous it is compact (in the sense
of uniform convergence).
Proof. Take an arbitrary countable set M of points x,, x 2, . . .
. . .. x,j, . . . (belonging to the interval a ^ x ^ b) which is every­
where dense in l«, b].’*** For instance, we can take, as x,, x 2, . . .
. . ., x„, . . ., the set of all rational points of the interval or the
set of all the points appearing in the process of successive division
of the interval into 2, 4, 8, . . 2n, . . . equal parts. Let us choose
a sequence of functions {fn (x)} belonging to the family. The family
being uniformly bounded, we have a relation | /„ (x) | <1 C —
= const < -j-co for all n — 1, 2, 3, . . . and all x £ M. In particu­
lar, the numerical sequence fn (x,), n = 1, 2, . . is bounded and
hence, by the I3olzano-Weierstrass theorem, there is a convergent,
subsequence of this numerical sequence whose members can be
numbered as (x,). / , 2 (x,), . - ., / tn (^i)t . . . . Thus, we have

* Uniform convergence on an arbitrary sot is defined in a manner completely

analogous to that of defining the notion of uniform convergence on an interval
In functional analysis the concept of compactness is also introduced, in n similar
fashion, for convergence in the mean and for other kinds of convergence but
we shall not dwell on these questions here.
\ e/> *« 1« P
** -\l«so k*i“\vn a? the Ascoli Arzela thcorer.......... ► »OW« V :..l:
' iu
*» A
ti O/.o 4onrk\
it< \ kCi * K/’
and Arzela, Ces.irc (1847-I912). Italian niathematkians.
*** A suh.cct B of a point set .1 is said to he (everywhere) dense in A if the
closure of B coincides with A (compare with footnote on pace 21M.
370 MULTIPLE IN T E G R A L S , FI ELD THEORY AND S ER IE S

taken, from the sequence {/n (x)}, the functional subsequence

/ll (^)> / 1 2 (•*•)» • • ♦» f i n (*^)» • - •
convergent at the point x, £ il/.
We can now similarly construct a subsequence
/21 /2 2 ( ^ ) i • • •» f 2 n ( ^ ) » • • • ( - 2)

of sequence (2j) convergent atthepoint x 26 il/- Subsequence (22)

is also convergent at the point Xj £ Ai since its members belong
to the sequence (2]), the latter being convergent at the point x,.
Thus, subsequence (22) converges at the points Xj and x 2 belonging
to the set Al.
Continuing unlimitedly this process of selecting subsequences
we arrive at an infinite table of the form
f i t C^) > f 12 (**•) 1 • • *r f \ n (*£)» • •*

/21 C ^)i /2 2 (*£)» • • f 2n ( ^ ) t ♦ •*

........................................................... (3)
f ni f c ) , f n 2 ( * ) » * • f nn ( * ) , • • •

whose «th row (n = 1, 2, 3, . . .) is a subsequence (of the original

sequence {/„ (x)}) converging al the points x 1? x 2, . . xn. Therefore
the “diagonal” subsequence,
/ 1| ( x ) , / 22 (x ), ■ • m f nn ( ^ ) 1 * • • ( ‘0

converges at each point xn £ A/.1

Let us prove that this sequence is uniformly convergent 011 [a, 61.
To this end, it is sufficient to show that sequence (4) satislies the
condition of Cauchy’s test for uni form convergence of a sequence.
Take an arbitrary c 0. Let us choose 6 (e) > 0 such that the
condition
| / (x') — / (x") | < e for | x — x" | < 6 (e) and x', x “ £ la. 61
holds, which is always possible since 1^1
the family is^ equiconti-
nuous. Next take a finite subset M lx,. x 2, . . ., xpl of the dense
set Al whose points break up the interval la, 6| into subintervals
of lengths less than 6 (e). Then, for each x £ (a. 61, there exists
a value x* £ M such that | x — x { ( < 8 (e). Furthermore, for the
given h 0, let \is lake N (e) independent of x*. i = 1, 2, . . ., p.
such that
I fmru ( *i ) — fan ( * « ) I < € ^ A ll ” > & (* 0 (3)
and ail 1 = i, 2, . . p. Then we have ihe inequality i / mm (*»') —
— / nr< (j-) j < for all m, n N ( b ) and all x £ l<*i 6|. Indeed,
eiipposo m. r? 'V (b'\ ami r f \o. h]. I hen there is a value x* 6 M
 CII. 8. FUNCTIONAL S E Q U E N C E S AND S E R I E S 377

such that | x — x* | < 6 (e). The family being equiconlinuous, it

follows that | fmm (^) fmm (•£/) I £ and | fnn (x) i nn (*£f) I
< e. Consequently, by virtue of (5), the inequality
I f m m (*e) fnn (•*•) I I f m m (x) f m m ( ^ - i ) I “f* I f m m O ^ i) fnn “H

1 fnn ( * i ) ” /n n W 1 < 8 + £ + 6 = 3fi

is fulfilled. Hence, sequence (4) satisfies the condition of Cauchy’s

test for uniform convergence and thus the theorem has been proved.
Arzela’s theorem has many applications in various divisions of
mathematics. In particular, it is used for proving existence theorems
of the theory of differential equations.
Let us prove another theorem widely applied in the theory of
integral equations.
T h e o r e m 2. If a functional sequence {fn (x)} is equiconlinuous
on [a, b\ and satisfies the condition
6
P2 ( / n > fm) = f l/m (*) — fn (x)l2dx->- 0 f o r « , /» -► + < » (l>)
a
it is uniformly convergent on the interval [a, 6] to a continuous func­
tion f (x).
Proof. We shall show that such a sequence satisfies the condition
of Cauchy’s test for uniform convergence, that is
Tn. rn (*) = Ifn (*) — /,„ (*)1 I t 0 on fa, b} as u, m -boo
;\ssume the contrary. Then there is e 0 I>()such that for an arbitrarily
large k there exist nu ^ h and x h £ la, b] for which
! <Pa. nh ( * f t ) I > 4f0 (7>

The sequence being equicontinuous, there is 6 = 6 (e0) such that

I f n (*) — fn (x,{) I < e0 for | x — x k | < 6 (e0) (8)
Consequently, for nh > k Z> N (f0) and | x — xk | <C ft (e0) wc have
I 9*. nh {*) — <Pa. nh i*h) I < I /a M — fk (*ft) I -1
~ I fnk (x) — fnh (xh) I < 2c0 (9)
Then, by virtue of (7) and (9), we can write
I 9ft. nh (*) I «|ft. nh (*ft) I “ I 9ft. nh (*> ~ 9'ft. nh (*h) I >

> — 2 s0 — 2t-0 {{•})

If we take 6 (e0) C b —a then at least half of the interval
Xh — 6 <T x < r 8 b e l o n g s to flip i n t e r v a l la M H e n c e , b y MOl
378 MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

the inequality
6
P2 (/nA, fk) = f <P*. nh (*) dx > 4eJ- 4" = 2ej6 = const > 0 (11)
a
is fulfilled for nh > k > X (s0), which contradicts condition (6)
since the number k can be chosen arbitrarily large. Therefore
<frn. rt (*) “ l/m (^) — /« (^)l ^ o on la, b| as n, m -*■ + °°* &>n
sequcmly, by virtue of Cauchy’s lest for uniform convergence of
a sequence, the sequence {/„ (x)} is uniformly convergent on la, 6]
to a function / (x) which is continuous as a limit of a uniformly
converging sequence of continuous functions. The theorem has been
proved.

APPENDIX 2 TO CHAPTER 8

WEAK CONVERGENCE AND DELTA FUNCTION

Besides uniform convergence and convergence in tbc mean, in

ma(hematics and mathematical physics the so-called weak con­
vergence also plays an important role.
D eft n i t t o n i . A sequence of functions
<Pl ( * ) , T 'i ( * ) , • • ■* <Pr, (X)y - • - lO

defined and integrable on an interval (a, b) is said to be a tceafcly

f u n d a m e n t a l s e q u e n c e on the interval (a, b) if for evert/
bounded and continuous function f (x) there exists a finite limit
0
Jim \ / (x) (p„ (x) dx (2)

D efin itio n A function ip (x) is called a w e a k l i m i t of func­

tional sequence (1) (or the sequence is said to be w e a k l y c o n v e r y e n t
to cp (x)) on an interval (a. b) if the relation

lim (3)
n—*-f-
holds for every bounded continuous function f (x) defined on. (a. b).*
If sequence (1) is uniformly convergent or convergent in the mean
to an integrable function *p (x) equality (3) is fulfilled for every

* More precisely, rp (.r) is a weak limit of sequence (1) relative to the class
<>f functions / (x) continuous on («/. ?j). The notion of weak convergence can
also ho defined for oilier rhi««sc«; of functions.
 CH. 8 . FUNCTIONAL SE Q U E N C E S AND S E R I E S 379

bounded continuous function / (a:) because this is implied by the

theorems on passing to limit under the sign of integration. Thus,
if a sequence {<pn (x)} converges uniformly or converges in the
mean to a limit function <p (x) it is always weakly convergent to
<f (x).
I f wc ha ve
lim \ f (x) q)n (x) dx —0

for any continuous hounded function / (x) defined on (a, b) the

sequence {q„ (x)} is weakly convergent to zero on (a, b) since equali­
ty (4) can be rewritten in Ilie form
b t>

lim i / (x) <pa (x) d x = f / ( x ) -0*dx (4f)

n—
►r jo a • v*
a
b

Applying Cauchy's test to the number sequence ( J / 0*0 cp„ (x) dxj
a
we arrive at the following test for a functional sequence to he weakly
fundamental:
Cu itc h y 's T e s t ( f o r ti S e q u e n c e to fie We ah’I y T n n <t <i-
m c n t a l ) . Sequence (1) is ireahly fundamental on (a, b) if and only
if for every t 0 and every continuous function } (x) there exists
a number N (e, /) such that
K
(X) lq>n (x) — Cp„, (x)l dx < e P)
for all n, m > N (f, /).
We remind the reader that a fundamental (Cauchy) sequence of
rational nntnhcrs may not converge to a rational number and this
leads to the irrational numbers which enlarge the class of rational
numbers so that every fundamental number sequence has a limit.
Similarly, there are weakly fundamental sequences of inte­
g r a te functions which do not weakly converge to an integrate
function and therefore it is advisable to enlarge the class of functions
under consideration in an appropriate manner. This loads to the
so-called generalized functions.
For instance, let us consider a sequence of functions {b„ (x0, x)}
determined by the relations
<*
{ n for Xo—
t <1 x < x0 +. 1

The seqiicnrp wojiklv fiiinlnnoMilal 1on <*vj*rv inlprval

1 to. b > . (G )
V *
0 for ~oc <C r <C Xu— — and x0f — < a : < - o o
380 MULTIPLE INTEGRALS. FIELD T H E O R Y AND S E R I E S

indeed, let a function / (x) be continuous on (a, b) and x0£(a, b).

Then, beginning with a sufficiently large n, the interval
1 \ \
( x0— Xq ‘2 ^’) is contained within (a, b) and hence, applying
Li

the mean value theorem to the integral j /(x )6 „ (x 0, x) dxf we oh-

a
tain
*0+2^
j f ( x)Sn (xQ'x)dx = n j /(x )d x = /(t),
a JI
X°” 2n
(li')
1 _ ,1

Passing to the lim it in the last equality,fas n —>■-r 00, we derive,

by the continuity of / (x), the relation
b
lim f / (x) 6„ (x0, x) dx = / (x0) (7)
►-{-00 J
It—

In case x0 lies outside the closed interval la, bjfweffiud

b
lim 1 / (x) 6„ (x0, x) dx = 0 (*)
n—►4*00 *

But of course there is no ordinary integrable function which

is a weak limit of the sequence {6P (x0. x)}. Therefore, to define
the weak lim it of this sequence we introduce a generalized function
6 (x0, x) called 6-function (the Dirac* delta functional or distri­
bution).
Making use of the definition of a weak limit (see equality (3))
and taking into account relations (7) and (8) we can write, for every
continuous function f i x) on (a, 6), the relation

J I 0 for x0£ [<7 , b]

Formal relation (9) is sometimes taken as a definition of the tle11a
function.
Note. Instead of a bounded continuous function / (x) defined
on (a. b) which enters into relations (7) and (8) we can obviously
take an arbitrary bounded piecewise coulimmua function / (2 .)

* D i r a c .1 P a n ! A d r i e n Mr.ur?co \ o *
 CH. 8. FUNCTIONAL SE Q UE N CE S AND S E R I E S 381

if ils value at each point of discontinuity x0 belonging to (a, b)

is defined as / (x0) = tS**— T^ . It is therefore natural
to extend relation (9) to such functions.*
We know that an irrational number can be determined by means
of various equivalent fundamental number sequences of rational
numbers. Similarly, a generalized function can be regarded as
a limit of different equivalent weakly fundamental sequences.
To define equivalence we introduce tile following
D e f in itio n «?. Two functional sequences {(p,, (x)} and {\|>n (x)}
weakly fundamental on (a . b) are called e q u iv a le n t if the relation

lim f / (x) (if n (x) —ifn Or)] dx = 0

Ti-*-p>a *

is fulfilled for every bounded continuous function f (x) defined on

In practical applications ol the delta function we often take
various weakly fundamental sequences (convergent to 6 (x0, x) in
the sense of relation (3)) equivalent to the sequence {6„ (x0. x ) } used
in this section (see Appendix 5 to Chapter 11). A feature of rela­
tion (9) which formally defines the delta fund ion is that when wc
multiply an arbitrary continuous function / (x) by 6 (x0, x) and
integrate the product over an interval in which / (x) is defined and
which contains x0 we “isolate" the value / (xtl) of tlie function / (x)
assumed at the point x0.
If to each function belonging to a certain class of functions defined
on (a, b) there corresponds a number we say that we have a f u n d io n a l
defined on that class of functions.

* If To is a point of discontinuity of a piecewise continuous function

/ (x) and / (xo) = ^ r°— — then to prove equality (7) for x0 £ (a, l>)

it is necessary to break up the integral ^ f ( x) dx in relation (S') into tfie

x« - ^
XO
two integrals ^ f ( j ) d. r and ^ /(r)rfr and separately apply tin mean
I To
value theorem to either integral.
382 M ULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

For example, let {/ (x)} he the class of all functions integrabie

b

on (a, 6). Then the integral ^ / (x) dx is a functional defined on

a
the class.
Integral (9) also determines, for every fixed x 0 6 (a, 6), a func­
tional defined on the class of all functions continuous on (a, b).
Indeed, relation (9) associates, with every continuous function
/ (x) defined on (a, b), a uniquely specified number, namely the
value / (x0) of this function at the point x0 £ (a, b).
6-function is sometimes identified with the functional determined
by relation (9). This is one of the possible interpretations of general­
ized functions which can serve as a basis for the theory of generalized
functions.
9 Improper
Integrals

The definition of the integral as the limit of the integral sums

b n
(9.1)

does not embrace those cases when the integrand is an unbounded

function or the interval of integration is infinite. In mathematics
and mathematical physics, however, we often use integrals of
unbounded functions or integrals with unbounded domains of
integration. Such integrals are called improper. To define them
it is not sufficient to apply a passage to the limit of type (9.1) but
it is necessary to use an additional passage to the limit involving
the domain of integration. The original domain of integration where
definition (9.1) does not apply is replaced by a subdomain where
it holds. Then this subdomain is made to extend so that it tends
to coincide with the original domain. The limit of the integral
taken over this subdomain is called the improper integral over
the original domain. This is the general idea the definition of the
improper integral is based on. The strict definition will he given
below.

§ I. INTEGRALS WITH INFINITE LIMITS

OF INTEGRATION
1. Definitions. Examples. Let a function f (x) be defined on an
B
interval a ^ x <Z + oo and let the integral f / (or) dx (determined
a
by relation (9.1)) exist for every B ;> a.
D efin itio n (. The improper integral ^ f (x) dx is understood
as Ike limit
(9.2)
 3S4 MULTIPLE IN TEG RALS. FIELD THEORY AND S E R I E S

-+-ao
If this limit exists and is finite the impoprer integral ^ / (x) dx is said
a
to be c o n v e r g e n t . If otherwise it is said to be d i v e r g e n t .
\'ote. Let Gj i> a. Then the equality
Ft <M fj

]* / (x) dx = \ / (x) dx -j- \ f (x) dx

a a ci
—-*3 -j-OO
implies that the integrals \ f (x) dx and I / (x) dx are simulta-
Ja J
ai
neously convergent or divergent. Thus, when testing the integral
-«~30

| / (x) dx for convergence, we can replace it by the integral

a
+*»
^ / (x) dx for any a, ;> a provided that the function / (x) satisfies
<n
the requirements of Definition 1.
Improper integral (9.2) can be interpreted geometrically in the
following way. Let / (x) be a continuous nonnegative function
defined for x ^ a. Consider the domain Q hounded below by

the port a <c^ x < ;-oo of tlie x-axis, above by the graph of the
function and on the left by the line segment x — a. (a).
The definition of squarability and the notion of the area of a plane
figure introduced in Sec. 3, § 1 of Chapter 1 arc inapplicable to the
domain Q because it is unbounded. Let us lake an arbitrary line
segment x = B ;.> a, 0 ^ y ^ / (/?). which cuts off a squarable
curvilinear trapezoid ABB*A* (see Tig. 9.1) whose area is equal
B
to the integral J / (x) dx. It is natural fo extend the notion of
a
squarability to the domain Q if the area of the trapezoid ABB9A*
tends to a finite limit as B -► -foo. In this case we say that Q is
squarable and call the limit of the area of the trapezoid ABB*A*
the area of the demain Q. This area is expressed by improper inte­
gral f9.2).
 CH. 9. IMIMiOFKH INTEOHA LS :5S;>

The improper integral of the form

a a
^ / (x) dx lim ^ / (x) dx (0.3)
A~+ — oo
—oo A
is defined by analogy with integral (0.2).
If both limits of integration are infinite we put, by definition,
-|-a> a -J-oo
^ f ( x ) d x — ^ f{x)dx-\~ j f ( x) dx (D.4)

where a is an arbitrary finite number, the integral I / (x) dx being

—50
regarded as c**on re r y e n t if and only if both integrals entering into
the right-hand side of (O/i) converge.
It can be easily shown that Ihe integral I / (x) dx is conver-
r
— OO
gent or divergent irrespective of the particular choice of the
point a, and if it is convergent its value does not depend on a.*
-{-oo
Thus, an integral of the form l / (x) dx is reduced to the inte-
—oo
+ 00
grals of 1lie form ^ / (x) dx and j f{x)dx. lint an integral of
a
•i-cv
* The integral | / (x) dx can be defined by the relation
v
— OO

■{-o
o J3
f f{x)dx - lim f f [x) dx <9.4')
J A-* - oo J
-°° n *i oo *
where A ami U tend to their limits independently. In fact, by virtue of
(9 2), (9.3) and (9.4) we have
-t-OO cl

j / (.r) dx — ^ f { x ) d x + ^ / (x) dx liin ^ / (*) dr i-

B t:
)- 1i in [ / (x) dx — lim f f {x) dx
li *>-foo J A -
>•—o
c> J
a B-++ -o
the last limit existing if and onlv if for .1 .x< and /> —*-4-oo the cor re:
«i1 * n^
j HHHl t ng l i m i t s lim \ / r r i ^ . r :tn<i lim \ / ( ri //r ovist wIiimi A zuul fi I«mw1
a » —oo J /j— %
t
to their limits independently.
*. \J O—t
 m u l t ipl e in te g h a l s, fie l d t iie o h y and sk u ies

tlie type j f ( x) dx is reduced to the corresponding integral of the

—w>
•1-30

form I / (x) dx by a simple change of variable if we substitute

Ja
— x for x y and therefore in what follows we shall limit our-
-f<V
selves to studying integrals of the form \ f(x)dx.
a
Let us consider some examples.
-f-sc ii

1. The integral ^ sin xd x — 1iin \siuxc/.<; = lim (1 — cos H)

0 J
u «— -X > »' JJ-. ! - -V

is divergent since cos D lias no limit and oscillates between — 1

and --1 as Ii —*---oo.
:*x

2. The integral j converges because the finite limit

—YV>
I!
lim \ x -- lim [arc tan Ii —arc tan .1] — — ( — ? ) Ji
«-.+00 • ^ ‘C Jf-k'-TC — ^ -“ I
A •- » A A . - JO
exists.
• The integral

f dx where C —const > 0 and a >* 0

J jra
is important for our further aims. It converges lor rx.Z> 1 and
diverges for a - < l. Indeed, we have
for a = 1

for a =?£ 1
and therefore
t a
C -----
a —1r for a > 1
! co for a < 1
This integral is widely used when applying the comparison test
to testing various improper integrals for convergence or divergence
(see Theorem 0.3 in Sec. 4).
-}->T
2. Reducing Improper Integral of the I’orm | /(.r ) <fx to Xu me-
 Cll. ». IMPROPER INTEGRALS 387

rical Sequence and Numerical Series. Testing an improper integral

J / (x) dx (or convergence can Ik* reduced to testing convergence
a
of number sequences or number series.
~\-oc
According to Definition 1, the improper integral \ / (x) dx
*.
'i
n
is the limit of th e function F {II) —V/ (x) dx for R -o o .
a
Applying to I'{13) the definition of 11le limit of a function ill terms
of sequences (e.g. see [8|, Chapter A. § 2) we arrive at the following
test:
For (fie integral ^ / (x) dx to be convergent if is necessary and
•
>
<1
sufficient that for an arbitrary choice of a sequence of points
B h J> <y, n = 1. 2 , . . li oo for n -+■ oo (0.5)
the numerical sequence
»n
j / (x) dx, n = 1, 2, Mt . . . (O.li)
a
•! OD
converge to the same finite limit. In case the integral ^ / (x) dx is
tl
convergent the limit of sequence (0.0) is equal to the value of the integral.
Note that (0 0) is a sequence of partial sums of the scries
ih nt Un
\ f (x) dx -j- / (x) dx -I- . . f /(x)tfx (0.7)
a li\ Bn - 1
and therefore ihe above test can be rephrased as follows:
+«
For the integral J / (x) dx to be convergent, it is necessary and
a
sufficient that for any choice of sequence of points (0.5) num ber
series (0.7) converge and its sum be independent of the particular choice
of the sequence. If the integral \ / (x) dx converges the sum of series (0.7)
tl
is equal to the integral.
Ante 1. If / (;r) is a function with alternating sign on the interval
a x <C t ° ° . the convergence of series (0.7) for a specific choice
of point sequence (9.5) does not imply, in the general case. Ihi* cmi-
-'.j-
388 MULTIPLE in t e g r a l s, fie l d theory and s e r ie s

+">
vergiMicc of tlie integral j / (x) dx. For instance, the integral
a
J sin x dx is divergent (see Example 1 in Sec. 1) although the
+oo 2ji(n+l)
series 2 J sin x dx converges because all its terms are equal
n=0 2:m
to zero.
/f a function f (x) retains its sign for all x ^ a, for instance, / (x) ^ U
oo
for all x ^ a, then for the integral j / (x) dx to be convergent it i>
a
necessary and sufficient that series (9.7) converge for at least one choice
of a monotone increasing sequence of type (9.5).*
The necessity of this test follows from the above. Let us establish
its sufficiency. Let / (x) ^ U for all x ^ a and let series (9.7) converge
For a monotone increasing sequence of type (9.5). The sequence (9.0)
of partial sums of the series is monotone increasing (or nondecreasing)
and tends to a finite limit / . We shall prove that for any other choice
of a sequence
hiin «, m = 1 , 2 , . . . ; B'm + oo for m -f oo {9.5')
the corresponding series
R:
^ f (x) dx -f- ^ / (x) d x . . . -j- ^ / (x) dx -f- . . . (9. / )

is converging and its sum is equal to J,

To prove the assertion we shall use partial sums ul series (9.7)
and (9.7'). Given e > 0, there is Bn<t such that the inequality
Br
J —£ < \ f(x)dzCJ

holds. Let us choose m0 such that for all m ^ m0 the inequality

B'm ^ B nrt is fulfilled. Furthermore, for any B'm there exists B nm>
B't,,, and therefore we have the inequality
®t»o s 'm
J — e < ^ /(x )d x < J /(x )d x < f{x)dz^-J j
11

* Compare with the well known Cauchy integral lest for convergence of
on infinite coring fo.c*. <*00 (8|. Chapter 13. 5 2. Sec. 4)*
 CH. 9. IM P R O P E R IN T E G R A L S 389

for all m > m0 since / (x) is nonnegative. Consequently,

B

lim \ f ( x) dx = J y which is what we set out to prove.

fl-*—OOJa
Example. Take the function
2" for n^.x4^.n n = 1,2,
/w = <
0 for n -f < z < « + 1, n — 1 ,2 , . .
Then the integral
+ 0O -|-oo 11+J .L- 30

j f ( x ) d x = 2 j /(* ) < f a - 2 4 r = 1
1 n= 1 n n= 1
is converging.
Note 2 . The above example indicates that even if a function / (x)
-}-ao

is nonnegative the fact that the integral ^ / (2 :) dx is convergent

a
does not imply that / (x) — 0 for x —►-j-0 0 .
3. Cauchy Criterion for Improper Integrals. The convergence of
an improper integral
+00 B

f f( x) dx = lim \ f ( x ) d x (0.2)
Ja » aJ
is equivalent to the existence of a (mite limit of the function F (£) =
B
= J f (x) dx for B —+■ + 00. According to the Cauchy general cri­
terion (e.g. see 181, Chapter 8, § 1, Sec. 2), F {B) tends to a finite
limit as B —*■ -f 0 0 if and only if for every e > 0 there exists B (e)
such that I F { B”) — F {B’) | < e for all B' > B (e) and B" >
B

> B (c). Taking the expression J / (x) dx as the function F (5) we

a
obtain the following
C a u c h y ’s T e s t { f or C o n v e rg e n c e o f a n I m p r o p e r I n te -
-J-00 +OO

g r a l of T y p e J f ( x) ( 1x) . An integral j / (x) dx converges if

a o
and only if for every e z> 0 there exists B (e.) such that the inequality

f (r) r f i l c e (9.8)
it'
:wi> MULTIPLE INTEGRALS. FIELD THEORY AND SERIES

holds for all B ', B " > B (c), which is equivalent to the requirement
that the integral
b-
f / (x) rfx (it .S')
h
tends to zero for //' -»■ - o c , #" -poo.
('auchy test (‘J.8) may sometimes In* directly applied to testing
soiin1 integrals for convergence.
.-c

Example, Let us consider the integral \ —Mir d r . Its integrand

0
5 ||| 7* ^
/ (x) = — j — can be regarded as a continuous function if \ve put
/ (x) — 1 for x = = 0. Integrating by parts we get
II
hi

It"
f sin x , cos IS' cos IS" {* cos x ,
f\) —
s,nz
J
dx = IF------) — dx
/V ti­
Therefore
tf
r dr , - if1
( sin . _i rL ^1 ros .i ,
— - dx
tr3
.i'­
ll •
tr
I 1 . I1 ff r/.r
f I ... 2 2 n» — V
IV :'■ B" 1
11I JJ ■J- j is ' 13" * ol * ’ ^
fi*
-^no
lienee, t lie integral ^ -a *“ x dx is convergenl.
o
Imt it slioiiId be noted that in many important applications the
Cauchy test, is more effectively used not for investigating concrete
integrals but for establishing general tests providing some sufficient
conditions for convergence wliicli can be applied to practical pro­
blems. Before proceeding to study such tests we shall introduce
the notion of an absolutely convergenl improver integral which is
analogous to the notion of an absolutely convergent numerical
series.
h. Absolute Convergence. Tests for Absolute Convergence.
D el in it ion Vf. Let a function f (x) be integrable in the ordinary
sense over every interval a ^ x //. a. < />’ < - o o .* Improper

* As is known, inlograbili ty ot f(x) in tlx* ordinary sense over a finite

; I, ; \ J •in ji! ;<.- ;* ••£*• i tijl i i v 1’> Hi,* nrf| inn rv »»f I // rl I uver I tieinterval.
 CII. i>. IMPROPER INTEGRALS 31J1

integral (0.2) is said to be aIt.sol utchy con c e n jc n t if the integral

-i-oo

\ \ f { x) \ dx (!».'•')
*
•I
converges,
T ltc o rc m ft,/. If integral (0.2) converges absolutely it is con­
vergent.
Proof. Indeed, integral (0.10) being convergent, it follows that
n
for every e > 0 there is II (e) such that ^ | / (x) | dx < t- for
D‘
all B ', (r). Hut we always have
ir U"
j \ / (x) dx *> | \ \f (;r) | dx (0.10)
jy /r
which implies
B"
| [ f (x) dx < | \ 1/ (:c) Idx | e for all B \ IP > B (e)
Ir f'r

Hence the conditions of (Cauchy’s test are fulfilled for integral (11.2).
Consequently, integral (11.2) is convergent and thus the theorem
has been proved.
Note /. The fact that integral (11.2) converges does not imply
r* ^71T1 *
its absolute convergence. For example, the integral \ : - —dx
o
; eo
converges (see Sec. 3) while the integral ^ ^ I dx is divergent.
*0
To prove the latter it is sufficient (see Sec. 2) to show that the
-f-oojr(n+l)
number series V, ^ JS 1 \ dx is divergent which can easily he
11=^0 mi
done bv applying the comparison test for numerical series. In
fact., we have
ji <n -M ) L)
f | sin t | , . 1 | r . , ____ 2_____
J\ J-------
r Lc i > -(«—r~n—
—l)n I .1 \ sin xdx t" 1 1) ~t for n .>• I
.IN .Trt
392 MULT IPLE INTEGRALS. FI ELD T H E O R Y AND SERIES

1-00
-> „ 1
ami the series -U
^ .1n ~=— V. —
n xJ n is divergent because it differs from
n —I n= 1
2 * **
the harmonic series onlv in the constant factor n
Note 2. Let f (x) be a function defined for a ^ x < L - f-oo and
integrable over every finite interval Then, for any
4*00

a, we can assert that if the integral ^ / (x) dx is absolutely

-f-SO
convergent the integral j f {x) dx is also absolutely convergent
a
4-oo +°°
localise for any of the integrals |/(x)|fL r and ^ | / (x) | dx to
onvergent it is necessary and sufficient that
B"

j | / (x) | d x —►O for B \ B* OO

B'
o test an improper integral for absolute convergence we usually
apply comparison tests.
T h e o r e m V.'Z ( G e n e r a l ('o n tp a r is o n T e st). Suppose that
the inequality
I / (*) I < g to 0>-ii)
holds for all sufficiently large x. Then if the integral
-4«>
j eU) dj c (fl.12)
a
is convergent the integral
4-00
^ / (x) dx (if 13)
a
is absolutely convergent.

-l-oo

* If an integral of the form ^ f ( x) dx is convergent and the integral

a
^ ; f (x) I dx is divergent the former is called conditionally convergent. Such
*1
integ ra ls are I run led in Se e. 3.
** In Theorems 9.2, 9.3, 9.3', 9.3" and 9.3"' «ve suppose that the function
t (x ) is Integra hie in the ordinary sense over every finite interval // <1 x By
/r - H ro.
 C ll . 9. IMPHOPEK INTEGRALS 30 i

Proof. The conditions of Cauchy’s test holding for convergent

integral (9.1). inequality (9.11) implies that for any there
exists B (e) such that

j | / (x) | dx for all B \ £ " > (e)

B'
i.e. the conditions of Cauchy’s lest also hold for the integral
4 °°
\ | / (x) | dx. Consequently, this integral is convergent whence
a
it follows that integral (9.13) is absolutely convergent. The theorem
has been proved.
In Example 3 of Sec. 1 we showed that
„l-a
r c dx C ^—
a —1
tr for a > l
Ju
4 -0 0 lor a-^Cl
for a > 0 and C > 0. Therefore, on the basis of the general compari­
son test we obtain
T h eo rem i f j i (S p e c ia l C o m p a r is o n Test ). Let an improper
4 -0 0

integral V / (.r) dx be given.

V
a
Q
1. If | / (or) | < 1 —~ for all sufficiently large values of x where
x fl

C ^ Q and a > l , the integral converges absolutely.

2. If for all sufficiently large values of x the function f (x)
c c
satisfies the inequality /(a) 5 or f (x) - < -----— where 0 and
X®' ra
a^< 1 , the integral is divergent.
Proof. (1 ) Putting g (x) ---------
_.0t
in the general comparison test and

taking into account that the integral f C dx Cu1-*

a —1
converges
J ra
_.^o
for a > > l and a > 0 * we see that the integral ^ f ( x) dx is ubso-
it
lutely convergent.
* We suppose that because, if otherwise, a can be replaced hv
|-oc JU

«i >• 0, amt the integrals j f(x)d> ami £ f ( r \ ' f r are siinultaneously ah«o-
«1 'o
O ile lv c o n v e rg e n t n r H iv 'r'v o ri
o94 MULTIPLE IN T E G R A L S . FIELD THEORY AND S E R I E S

(2) Let /(x)5>-^- where C > 0 ami 1 for all x ^ a t > a .

B
Integrating between the limits at and B we see that \ f {x) d x ^
V

B “ *

^ — — »- 4- oo for Z?—►-}-oo since a ^ l. Hence, the integral

-T->J —3C
\ f ( x) dx diverges, which implies that the integral \ f { x ) d x is
ul a
also divergent.
If —for all .t ><*,;> a > 0 , C > 0 and 1 we can
JC*
£
put /*(x) = — f{x) and then f*(i: ) > — for all x > a , > a > 0 , and
X*
+ ?°
consequently the integral f f*(x)dx diverges. Therefore the integral
a

ll u
f (.a:) d x = 1im j f { x ) d .r = — 1 i in j /* (x) dx
1 it -»-}■* n
a n

also diverges.
Note 1. Part 2 of Theorem 11.3 can be equivalently formulated
C
as follows: if a function f (x) retains its sign and |/( x ) |> - —5 -
for all sufficiently large x ( x > a ) inhere C >■ 0 and a < ! l, the integ-
+°°
ral J f (x) dx diverges.
a

Note 2. The fact that the condition | / (x) | * £ > 0 , cc < 1

holds for a function /(x ) for all sufficiently large x ^ a may not
-Lao

imply that the integral J f { x) dx is divergent. This integral may

a
turn out to be convergent if /(x ) is a function with alternating
+«
sign. For example, we can easily show that the integral \ f (x) dx
i
where the integrand is determined by the relation /(x ) —
= ( — l ) n+1 — . « < x < r i } 1 , n --1, 2, 3. . . . , 0 < a < J , conver-
 CH. 9. IM PROPER IN T E G R A L S 395

ges al though | / (x) | - — and t lie integral

f | / (x) | dx diverges for
.f J
i
0 < a-< 1.
£
The test based on comparison with the function — can be reph­
rased as follows:
T h eo rem 9.H' ( Mo d i f i e d S p e c ia l C o m p a r is o n T est). Let
*+*»
the integrand in the integral j / (x) dx be representable, for all
a
sufficiently large x, in the form f (x) — g ^ . Then (1) if g (x) is
bounded in its modulus and a > l , the integral is absolutely con­
vergent, (2) if g(x) retains its sign, | g (x) | >• const > 0 and a < 1
die integral diverges.
The following form of the comparison test based on comparison
C
with the function —- sometimes proves to be convenient:
xa

T h eo rem ihli” (th e L i m i t i n y F orm o f S p e c ia l C o m p a ­

rison. T est), Let the limit lim ( / (x) | xa = C exist. Then (1) if
JC-+4-O0
0 ^ C <Z. -}-oo and a > 1 the integral ^ / (x) dx is absolutely
a
convergent, and (2) if 0 <C C ^ + oo, a ^ 1 and f (x) retains its
sign for all sufficiently large x, the integral is divergent.
Proof. (1) If 0 ^ C <Z -f- oo, we have, for all sufficiently large x,
the inequality
2r
|/( x ) |x » < 2 C , i.e. | / (*) | < -= --, for C > 0
TU
and the inequality
|/ ( x ) |x a < l , i.e. | / ( * ) | < ~ - * for C = 0
X
+oo
Therefore, by Theorem 9.3. the integral J / (x) dx converges
a
absolutely.
(2) If 0 <; C <1 -j-oo and gj ^ 1, we have
|/ ( x ) |x a > - ^ -, i.e. | / ( ^ ) | > - —^-, for C < - f oo
 MULTIPLE INT E GR AL S, FIELD THEORY AND S E R I E S

and
| / (ar) Iz a > 1, i | / (x) | > — , for C = -f- oo
I
for all sufficiently large x. and hence, on the basis of Note 1 after
-J-oo
Theorem 9.3', we conclude that the integral ^ / (x) dx diverges.
a
The theorem has been proved.
Note 3. Theorem 9.3" (the modified comparison test in the limit
form) embraces a narrower class of functions than Theorem 9.3'
(the modified comparison test) because Theorem 9.3", unlike Theo­
rem 9.3' is only applicable if a finite or infinite limit of | f (x) \ xP-
exists for x -}-oo.
Theorem 9.3" implies
T h e o re m iK’T (S p e c ia l C o m p a r is o n T est in T e r m s of'
O r d e r s o f I ti f i n i t e s i tn a l s j , Let \j (x) | be an infinitesimal of
the order of ~ for a: -> -foo. Then (1) if a j> 1 the integral
+QO
^ / (x) dx converges absolutely and (2) if <x ^ 1 and f (x) retains its
a
-f-ao
sign for all sufficiently large x, the integral j / (x) dx diverges.
a
We remind the reader that / (x) is said to be an infinitesimal of the
order of (ct ;> U) lor x —y ;-oo if
xa
lim —4p-L-■= liin \ f ( x ) \ x a = C where O c C ^ r 00
• ) ■ » I _______ \
X X —» 4 -° °
\ xa i
Note 4. It is obvious that Theorem 9.3” is applicable to a still
narrower cla.ss of functions than Theorem 9.3" since its conditions
include the existence of a finite limit of | / (x) lx®, for x j-oo,
different from zero and infinity.
Exam pies
-f-OO
1. By Theorem 9.3, the Eulcr-Poisson* integral j e~*2 dx is con-
o
vergeut since the exponential function e~xi decreases faster than
any negative power of x as x —j- J-oo and, consequently, we have

r-iisa u i., Deni** (17.Q1 t Q/in y F ro n rh r n n tt io m .it ir.ia n

 CH. ». IM l’KOPEH INTEGRALS 397

for all sufficiently large x where C = const ;> 0 (here wc have put
a = 2 but any other number exceeding 1 can also be taken as a).
Theorem 9.3' also indicates that this integral is convergent because
we have
Iim xze~*~ = 0 (a = 2)
X - ¥ — OC

-r*
2. The integral I e~xx7'~l dx converges for all real values of />.
l
Indeed, to prove this we can apply, for instance, Theorem 9.3*
since the relation
lim x2e~x:x,l~1 = 0
X-+-r»
is fulfilled for all such p.
+*»
f Xm
«’». Let us consider the integral j yW dx for We have
1
•rm _ T™ 1 _ K (x) 1
[ — " ~ , r n 1-L rn U (+) 1 ! C~"

where -4- ^ g (-r) ^ 1 for x ^ 1. Therefore, by Theorem 9.3', the

integral is convergent for n — m „> 1 and divergent for n — m ^ 1.
5. Conditional Convergence.
I) t / i n it ioa .7. . 1n intcgral
+00

f j ( x) t l x (9.1-1)
a

is said to be condiI ionally convergent if it co/merges while the integral

+oo
j I / M l dx ( 9 .1 5 )
a
diverges.
The Abel theorem given below makes it possible to lest some
conditionally convergent integrals for convergence.
77leoren* tt. l (. \UeVx Test ). Let <p (a*) he continuous and g (x)
be continuously differentiable on the infinite interval a x < -J o o .
n
If the a n t i d e r i v a t i v e < 9 ( / > ) = ^ ij> ( . / ) <Ir is honndetl on. tii.e in te r v a l
a
a />' <[ -j-oo and g (x) is a monotone decreasing function which
 MULTH'LE IN T E G R A L S , FIELD THEORY AND SERIES

tends to zero for x — -foo, the integral

1 cp (x) g (x) dx (‘J.lC>i

a
is convergent.
Proof. We shall show that, under the conditions of the theorem,
the requirements ol' Cauchy's lest are IuHi 1led for integral (ti.10).
Integrating hy parts we obtain
ir i-r
C (x) g (x) dx = (I) (H ') g (// ”) _ <I> (B ') g ( H ' ) - \ <H (x) g' (x) <fx
ir ?;*
The function g (x) being monotone ami decreasing as x —►-j-oo,
we have gf (x)<^10t arid it is therefore allowable to apply the genera­
lized mean value theorem to the last integral. This results in
B”
( (x) gr (x) dx -= T> (?) j g (x) dx = t]> (|) [g (IV) — g (ZT) |
tf B’
where c lies between ft* and H". Consequently.
iv
( <r (*) g (*) d x ^ 4> (B■■) g ( « " ) - <I>(B') g (B')- 4> (5) | g ( B " ) - g (B')\
h’
The antiderivative <J> (/?) being bounded, it follows that
B"
C cp (x) g (x) dx~+ 0 for B \ B " ^ ~ o o
a*
since g {B) tends to zero as li —*- T oo. The theorem has thus been
proved.
Exam pies
+ oe>
1. The integral ^ S1”— dx is convergent for a > 0 because
k X
if we put ff (x) = sin x and g{x)-~ — , we have
JCa
X X

| Cl) (x) | = j I cp (x) dx — j sin x dx = | cos ji — cos x | < 2

jt n

\
for jx ^ x ^ -|oo, and g (x) *=-— is a monotone decreasing func-
xa
lion tending to zero for x -}-oo and a > 0 .
 Cll. i). IMPHOl’EH INTEGRALS 39U

iXote. We can prove lhat the above integral is convergent without

resorting to Abel’s test if we apply Cauchy’s lest and integration
by parts as it was done at Ihe end of Sec. 4.
2. Taking the integral

lest, it is convergent.
— -V.
%
3. Let us consider tlie Fresnel* integral I sin (a-) dx which is
o
used in optics. Putting z 1 = I we obtain

The Abel test indicates that the bitter integral converges.

(i. Extending Methods of Evaluating Integrals to the Case of Impro­
per Integrals. When evaluating improper integrals we can change
variables, iutegrale by parts and represent the integral of a sum
of functions as the sum of the integrals of these functions, that is
apply all Ibe methods of evaluating proper integrals. The correspon­
ding formulas are valid provided all the integrals entering into them
are convergent.
Let us illustrate the significance of the latter requirement by
taking a concrete example. The integral
vcrgonl (which, for instance, can be established on tHe basis of the
special comparison test). Let us write down IHe decomposition of the
integrand into partial fractions (e.g. see 18], Chapter 7, § 7):

x- —x — 2 3 (x 2) ^ 3 (x —1) (*)

It is obvious that the integrals V and | 'jTZ'f are divergent.

3
therefore the equality

* Fresnel, Augustin Joan (1788-1827), a French optician ami goometnr

400 MULTIPLE INTEGRALS, FIELD THEORY AND S E R IE S

is incorrect. To use decomposition (*) for evaluating the integral

we can integrate (*) from 0 to A and then transform the right-
hand side of the resulting equality to the form
A A A
I f dx I f dx 1 | 5 A — 1 j

a" J x -2 "^"3 J or — f = T l n T A ~2 J
3 3 3
Now passing to Lthe limit in the last equality as yl—^ + o o we
obtain

§ 2. INTEGRALS OF UNROUNDED FUNCTIONS

WITH FINITE AND INFINITE LIMITS OF INTEGRATION
We shall First consider the integrals with finite limits of integra­
tion. Let a function / (x) be defined everywhere on an interval (a, hi

except possibly at a finite number of points. A point x 0 £ h!

will be referred to as a singular point of / (x) if / (x) is unbounded
in every neighbourhood of the point x 0. For example, the function
for 0 < x ^ 1
/(* ) =
I
0 for x = 0
X

has the singular point x = 0 (sec Fig. 0.2).

The point x — 0 is also a singular point for the function
f —s in - - for 0<Cx<TI
/w H *
l O for x —0
(Fig. 0.3). 11, should be noted Ihat. in the last example the func­
tion / (x) does not have an infinite limit when x ->- U because it turns
 Cl 1. 9. IM PR OP E R I NT E GR AL S ■m

into zero iuljnilcly many limes in every arbitrarily small neighbour­

hood of the point i = 0.
D e f i n i t i o n . Let a function f (x) be defined on an interval [a. b\
everywhere except possibly at a finite number of points. If x — b is
b -u

a singular point of the function f (a:) and the integral ^ / (xl dx

a
exists for every p, 0 < ft < 6 — a, the improper integral
t
\ / (x) dx is understood as the limit
b-ti
f / (x) dx — lim ( f (x) dx (‘b 17)
u n
If this finite limit exists integral (9.17) is said to be convergent,
if otherwise it is called divergent.
We similarly define the improper integral in the case when the
singular point of / (x) coincides with the left end point x — a of the
interval of integration fa, b]:
U v
[ f (x) dx — lim f / (x) dx (9.18)
io *!-?i .
(l-J
If both end points x —a and x —b are singular points of / (x)
the integral is defined as
b-U
f /( x ) d x = lim ( f ( x) dx (9.19)
J \-++o
X”*+0 J .

If an interior point x = c, a < c < i , o fjth c interval [a, 6| is

a singular point of / (x) \vc put
6 <?-u i,b
f f ( x ) d x = lim [ j f ( x ) d x ~ ~ j f (x) d z J (9.20)
a |X—
>—
J—
0 11 C-fA
Let us now discuss the conditions guaranteeing the convergence
of an improper integral of ari unbounded function. Applying Cauchy’s
criterion to the function
b —n

*» = ( I W dx (0.2t)

for ii —►-f-0 we obtain

2 C—0824
402 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

C a u c h y 's T e s t ( f o r I tu p r o p e r I n t e g r a l (9.17),). For inte­

gral (9.17) to be convergent it is necessary and sufficient that for every
e ;> 0 there exist 6 = 6 (e) > 0 such that
b -n m

J '<*>dx
b -H '
<e for all 0 < p r, ji" < 6 ( e)

Cauchy’s test for integrals (9.18)-(9.20) can be formulated in

a similar manner. It can easily be shown that integrals (9.19) and
c b
(9.20) converge if and only if both integrals j / (x) dx and j / (x) dx
a c
are convergent, and that in the case of integral (9.19) the point c can
be chosen arbitrarily. If integrals (9.19) and (9.20) are convergent
we have the equality
c 6
\
f (x) dx « J f (x) dx + J / (x) dx (9.22)

for both integrals.

The same idea can be used for defining an improper integral with
an integrand having a finite number of singular points on its interval
of integration (a, 61: the interval is broken up into a finite number
of subintervals in such a way that the function / (x) has a singular
point at only one of the end points of each subinterval.
Thus, the general case is reduced to integrals (9.17) and (9.18).
Rut. integral (9.18) reduces to the corresponding integral (9.17)
if we substitute —x for x. Therefore we shall limit ourselves to
studying integrals of form (9.17).
Absolute convergence and conditional convergence are defined
for integrals of unbounded functions like in the case of integrals
with infinite limits of integrations. Cauchy’s test makes it possible
to prove that an absolutely convergent integral is convergent and
also to establish the following test.
U e n e r a l C o m p a r is o n T e s t . If b is the only singular point of
f (x) on {a, b\ and | / (x) | ^ g (x) for all x £ Ul 6) lying sufficiently
b
close to b and if the integral J g (x) dx converges, the integral
a
b
J / (x) dx is absolutely convergent.
a
Lot us now formulate a test based on comparison with the func-
C r y
tion (6
—-----
—i)a — which is analogous to Theorem 9.3.
 CH. 9. IMPROPER INTEGRALS 403

S p e c ia l C om p a r is o u T ent. Let a function / (x) defined on

[a, b] have a singular point at the end point x = b and let the
b-n
integral ^ f ( x) dx exist for every p, 0 < l p < lh — a. Then
a
(1) if we have
| / (x) | ^ ----— — where 0 < C c — or», a < 1(9.23)
b
for all x6[<z, b) lying sufficiently close to b, the integral J f(x)dx
a
converges absolutely;
(2) if we have
f (x) > C where C > 0, a :> 1 (9.24)

for all x £ [a , b) which are sufficiently close to b or

/ (x) < — , where C > 0, a>l (9.24')

for all x £ |a, 6) lying sufficiently close to b, this'integral diverges.

Proof. (1) In this case we have
b-ur t-|l'
I
b—u'
!{x)
dx < | j |/( x ) |d x
b-ii*
cl r — d-^ ~

_ ^ l 90*"“ —(fO1-01
—° | 1—a
b
for a < l and p'. Consequently, the integrals j / (x) dx
b
and j | / (x) | dx are convergent.
a
(2) Let us suppose that / (x) is nonnegative.* Then we have
Q
fW > — for a < f l , < x < h , a > l

* If / (x) is nonpositive we introduce the function /*(x) = —/(x) which

6
is nonnegative, and if the integral J /*(x)<fx diverges the integral
a
b b
J / (x) dx= — J /* (x) dx also diverges.

20*
404 MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

and
6--p
p ii 6—(i
C f ( x ) d x > T — —---- d x —*-- L <x> for li—* -0 + 0 and a^ -1
J J (6—x)a ’ r
«1
aI a*
because

<*~*«>l "a 1 for a > 1

c , I l l —a 1 —a J
f —£
J
-—— dx
(6 —x) 1 C In Ln_£f for a = l
b
Hence, the integral ^f ( x ) d x diverges and thus the integral
at
J / (x) dx also diverges
a
Note. Part (2) of the above theorem can be formulated equiva-
Q
lently in the following form: if \f for all x lying
(*—b)
sufficiently close to b where C > 0, o c ^ i and f (x) retains ils sign
b
for these values of x, the integral J / (x) dx is divergent.
a
The above special comparison test can be rephrased as follows:
M o d if ie d S p e c ia l C o m p a r is o n T e st. Let a function f (x)
be integrable in the ordinary sense on every finite interval a ^ x <1
^ b — X where 0 < X < b — a . Suppose that this function can be
represented in the form f (x) = in a neighbourhood of b
{i.e. for b — 5 < z < b, 6 _> 0, 6 < 6 — a). Then
(1) if g (x) is bounded in its modulus and a <Z 1, the integral
b
J f (x)'dx converges absolutely;
a
(2) if g (x) retains its sign in a neighbourhood of b, \ g (x) | ^
t
> const ;> 0 and a ^ 1, the integral J / (x) dx diverges.
a
The modified special comparison test can be formulated in like
manner in the case when the only singular point of / (x) on the
interval la, M is located at the end point x = a.
It is also possible to formulate and prove the special comparison
test in the limiting form, which we leave to the reader.
 CH. 9. IM PR OPE R IN T E G R A L S 405

We remind the reader that \f(x)\ is said to be of the order of

1 for x —>b— 0 if

lim I/ (T) I _ n ra (b — a:)a —^ where 0 <C £ < oo

*-+6—0 ( 1 I x~*b—0
I {b-x)a >
Let us now formulate
S p e c ia l C o m p a r is o n T e s t in T e r m s o f O r d e r s of I n ­
f i n i t i e s . Let | f (x) | be an infinitely large quantity of the order of
1
(a > 0) for x--*- b — 0.* Then
b
(1) if a C 1 the integral I / (x) dx is absolutely convergent;
a
(2) if a ^ 1 and f (x) retains its sign in a neighbourhood of x = b
(ii.e. for b — — a) this integral is divergent.
This test is formulated similarly when / (a:) has a singular point
not at the end point x = b but at the end point x = a of the inter­
val [a, b\.
Examples
C dx
1. The integral J ^ converges since we have

1 1 1 1
/(* ) = ■
V i — (1 - x ) l/Z ( 1 4 z + r 0- ) 1/ 2 ( I — r l 1/ 2
.?(*)
where g W = j is a bounded function. Here wo have
<z = 0, 6 = 1 and a = -£-•

2. Consider’ the integral j 31^ )- dx. We have

sin * sin j: i , .
/<*) = =-3=rA
xPxP- f(x) xP~l x

where g (x) = —n—

* is a function bounded in its modulus, and
sin x ^ g (x) ^ 1. Here we have a — 0, 6 ^ 1 and cc = p — 1.
Therefore the integral converges for — 1 < 1 and diverges
i
sin x ,
si
—-=—
xp ax is convergent
C*
S
for p «< 2 and divergent for p ^ 2.
* The function / (j:) is supposed to he inlegrable in lliu ordinary sense over
every interval a ^ j < 6 — a, 0 <r X <r b — n.
40 G MULTIPLE IN T EG R AL S, FIELD THEORY AND S E R I E S

Abel’s test, for improper integrals with finite limits of integration

can be formulated and proved by analogy with Abel’s test for
improper integrals with infinite limits of integration (see § 1, Sec. 5)
and we leave this to the reader.
Finally, we can change variables, integrate by parts and break
up integrals into sums of integrals in the case of improper integrals
with finite limits of integration under the same conditions as in the
case of improper integrals with infinite limits of integration (see § 1,
Sec. 6).
Now we shall briefly discuss the integrals with infinite lim its
of integration of unbounded functions having a finite number of
singular points. If an improper integral is taken over an interval
a ^ x <Z. r ° ° or — oo <; x ^ a or over the whole x-axis —oo <
<1 x <i f-oo we break up the interval of integration by means of one
or two points of division into one finite interval containing all the
singular points of the integrand / (x) and one or two semi-infinite
intervals without singular points of f (x). This reduces the improper
integral of the general type to the above special cases. The original
integral is, by definition, understood as being equal to the sura
of the integrals taken over the subintervals the original interval
of integration is broken into.
The original integral is said to be convergent if and only if all
the integrals over the subintervals are convergent. If at least one
of these integrals is divergent, the original integral is regarded as
divergent.
It can be shown that the definition of convergence of the original
integral and its numerical value (provided this integral is convergent)
are independent of the choice of the points of division.
Exam pies
+°°
3. Take the integral \ e-*xv~l dx. If p — 1«<0, the integrand
i
has a singular point at x = 0. Therefore let us divide the interval
of integration into two intervals, e.g. [0, 11 and (1, -f-oo), by
means of the point x = l. This results in
-foo 1 -1-30
e~*xp~l dx = e~xxp~l dx -J- ^ e~xxv~l dx
o *1
i i
The integral \ e~xxp~l d x ~ \ -^ -^ dx converges for 1— p <i 1, that
o *0
is for p^> 0, and diverges for p*£.0. As has been shown (see § 1,
4-00
(•
Sec. 4), the integral \ c~xxv~l dx converges for all values of pt
 CI1. 9. IMPROPER IN T E G R A L S 407

+ 30
— oo < p <C -r oo , and, consequently, the integral \ e~xxp~l dx*
o
converges for all p > 0 and diverges for all p ^ .0 .
i
4. Consider the integral ^ xplnff — dx. Performing the substitu-
h
lion ln*^-—/ = x --=e~l and d x = — dt j we oblaiu
I 0 “ -oo

I xp In'1 -i- dx = — f dt = j ^ e-(P+t)< *

o + oo i)
The integrand in the last integral has a singular point t —0 and
therefore we break up the integral into two integrals:
•i-oo 1 - f oo

j *<Je-(p+i>r<fr = j dt -f j

(* (* (p-M)t
The integral \ e~^p ~^Hq dt = \ — —— dt is convergent only for
o «
— q C l, i.e. for r/ > — 1, irrespective of the values of p. The
integral j ^-(p+o^i rff (with q > — 1) converges only if p -1 1 >• 0,
l
i
that is only i Consequently, the integral C xp h\<1— dx
J X
0
converges for — 1 and q > — 1 and diverges for all the other
values of p and q.
+ oo
5. Taking the integral | xi>(\n x)<i (In hi x)r antl I,urforming the
e
change of variable In In x —t we find that the integral converges
for p > 1 (and arbitrary q) only if r < ; l. If p = 1 it converges
only if r < l and qZ> 0. Finally, if p <Z 1 the integral diverges
for any r and q.

* This integral is known as Euler’s integral of the second kind (the

-i-30
gamma function) and is designated by F (/>), i.e. F (p) = ^ e~xxfi~x dx (see
G
$ 3 of Charter 10L
4U8 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

§ 3. CAUCHY’S PRINCIPAL VALUE OF A DIVERGENT

IMPROPER INTEGRAL
Let a function / (x) be integrable in the ordinary sense on every
. b

finite interval of the x-axis. If the lim it

lim \ j ( x) dx does not
A— ►
—oo J
exist when A and B tend independently to their limits, i.e. the
+ 00 A
integral \ / (x) dx diverges, but the limit lim I / (x) dx exists,
J
—00 dA
—
the latter is called the (Cauchy) principal value of the divergent
— OO

integral j / (x) dx. In this case we write

— OO
-ro c A

v.p. f f ( x ) d x — lim f } { x) dx (9.25)

J
—00
A~~p-j-oo —A
J

(the notation v.p. originates from French valeur principale prin­

cipal value).
Now let /(x ) be a function defined on an interval (a, 6] and
having only one singular point c, a < l c < i b . Suppose that the
integral
b c-7. b

f/(x)c/x= lim { f f ( x ) d x -f f /(x )d x )

J ?-+0+0 t J J J
« H -* 0 + 0 u c _ rM-

is divergent, that is the lim it of the expression in curly brackets

does not exist when >. > 0 and p.;>0 independently tend to zero.
Then if the limit of this expression exists for X= p —►O-M) it »c
b

called the principal value of the divergent integral \ /(x) dx

a

and denoted as
b C -A b

v.p. f / (x) dx = lim { j f ( x ) d x + j f { x ) d x } (9.2U)

J a-*0+0 <1 c-t-A
Examples
1. If /(x) is an odd junction (i.e. / ( — x) ss —/(x); see Sec. 6 in
§ 1 of Chapter 11) defined on ( — 0 0 , ~j- 0 0 ) the principal value
OO A
v.p. j ( x) dx - lim f / (x) dx = 0
- .4
• » • . ►
d I O 11 j ' i A l v
 CH. 9. IMPROPER INTEGRALS 409

If / (x) is an even function on ( — oo, + oo) (i.e. / ( - * ) =

2.
= /(x); see Sec. 6 in § 1 of Chapter 11) we have
a a o
I / (x) dx 2 ^ / (x) dx = 2 J / (x) dx
-A b - a

4 “oo

for any .4. Therefore, if the integral J / (x) dx of an even func-

—oo
0
tion diverges, i.e. if at least one of the integrals f f { x) dx and
4" +OQ
^ / (x) dx does not exist, the principal value v.p. J /(x )d x d o e s
0 — oo
not exist either.
3. Let us apply tlie notion of the principal value of a divergent
improper integral to evaluating the integral
.1 OO
• ^

2J — \ "fVgtw c^c» "'here n and in are integers and 0 < m < « (9.27)
—ao
This integral plays an important role in the theory of Euler's
integrals (see § 3 of Chapter 10). We have
x2m I ^
—■ —7— < —r- for ± o o , where C —const > 0
. (2fe+l)n
Besides, all the roots xu = e 2n =ah- t i bk. k —0 , 1 , .... 2 n— 1 ,
of the equation l^ -x 2,i = 0 are not real. Therefore the integrand
has no singular points on the x-axis and the integral thus con-
verges. rTaking the decomposition of the l
rational fraction ^ Z * 7ng--
into partial fractions and integrating from —I to I ( / > 0 ) we find*
2n-l I 2 n —I

-I fc—0 -I 0 —I
2n - 1
= y Ak { f £f --------
^ I J( (i —ak)2-’r bh J, (x —»h)--| bh )

* An integral of a complex function u (x)- tr(x) of a real variable x

where u (x) and i>(x) are real functions is defined as ^ [a (x)-J-t'y (x)j dx =

= ^ u(x)dx-l-t ^ v(x)dx.
410 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

2n—1
((-<■>)*-)- bi
= A=0
2 Ak {ln
+ 1 [ arc tan T T + arc Un - ^ T -]}
x2m *
where Ah =- 2nizu-i • = — ^ x f i7"41 s*nce — — 1- Passing to the

limit for / —v -t-oo we obtain

+ co 2 n —l
r x-m vi
\ T T ^ dx= 2 ± n U *
-oo /.=0
where the plus sign corresponds to ^ > 0 and the minus sign to
bh c 0. The integrals
-4- OO -{-O© 4 -0 0

(i —flit) dx , . r bh dx \
ic —0 f 1, •••
( z - « h)2 -r^ ^ J (x-ah^-rfrhJ
— OO

dx
lim
are obviously divergent and the numbers -*i niAh
/— I
f-f-OO —i * —
are their principal values.
Now note that h* > 0 for A: = 0, 1,n — 1 and bh<C 0 for
k ~ n , n ' 1, 2n — 1. Hence, we can write
n - 1 2n —1
x- m
I-]-*2" dx = ni j 2 >•<.
k=0
s (A)

where
n —I n —1 n—1 (27q+l)(2fe+1)
1 2n
S —s- s 4 -- = 2/l
>e-=G h=0 ft=0
2 m-{-1
*— =-*—x .i-----—
2n
(2iw-f1){2ti*|'I)
^----- -n
2n
1 e
~2n~ 2w»+1
t2 — — — j i
1—e 2n
. 2m -M.t
.1—zn r---
n 2m-M (B)
1—e *" 2n a
 CH . 9. IMPROPER INTEGRALS 411

because e»(2 m + i):i_— l. Next, putting k = k -j- n we obtain

2n—1 2n-i 2 n -1 ( 2 w + IK 2 f c - H )
2n
S Ah = — s r 2 « m+, = - i ! r S
h=n fc=n
n —1 . ( 2 x n + t) ( 2 * '+ t>
2n

A'=*0
«— I . < 2 m + l) ( 2 f c ' + l)
2n
jt
= -sr 2 e' (<’)
fc‘= 0

and the last sum differs from (B) only in its sign.
From (A), taking advantage of (B) and (C), we deduce
2 in *- l
-l-oo
r2m 2rcc <? 31 1
2i = T i^ r -dx = — n 42 2m+171~ n 31T1 2m-1-1
1 — e 2n In
x2m
Thus, tlie integrand / (.r) — ^_^an- being an even function, we
have (see Example 2)
+<» xint
J= Jl (9.27'1)
0
1-f x2n Z f
n . 2m 4-1
sin — 2n
-------Jt

A. The improper integral

b c->. b
b— c h_
= lim c —it
+ lim In
>.-►0+0 c-n \-*-n+o
H-*0+0 n-^o+o
where a < i c < i b is divergent. But if we put X = and pass
to the lim it as 2i-*-0 4-0 we see that this integral possesses the
principal value
b
i* dx . b — c
v .p . I -------- =2 In ------- ( f l < C < fc)
j x —c c —a
a

§ IMPROPER MULTIPLE INTEGRALS

We shall lirsi consider the case’ when the integrand is unbounded
and the domain of integration is finite (bounded) and then pass
to the case of an i n f i n i t p Oinhmmrjod) dnrn.ii;. of L, U - i u L i o n . I or
412 MULTIPLE IN T E G R A L S . FI EL D THEORY AND S E R I E S

simplicity's sake, we shall take double integrals although triple

integrals and -Y-foId multiple integrals are considered similarly.
1. Integral of an Unbounded Function Over a Finite Domain.
Let a function / (./l/) = / (x, y) he defined over a finite domain 12
of the x, //-plane. We shall suppose that / (x, y) is unbounded in
every neighbourhood of a point Jl/0 (x0, </o) £ Q and that for any

domain 12 — gj* containing the point Af0 in its interior the func­
tion / (x, y) is bounded and integrahle in the ordinary sense over
the domain 12 — <j>6 (shaded in Fig. 9.4a and b). This means that
the integral \ j / (AJ) c/w is the limit of the corresponding
mJ
“ ' w6
integral sums (according to Definition 1 in § 2 of ('.hapier 1).*
The subscript ft denotes the positive diameter of the domain o>/>.
If ft —- 0 the domain o* is coni parted toward the point
D r / i ni t ton 1 . The i m p r o p e r -in te q ru t o f the fu n et ion
/ ( • * ') = / (x , y) o r c r th e d o m a i n Q is equal to the limit

ini { ( / ( J / ) r f w= f f f ( M ) d u > ( 9 .2 8 )
>-►1) ,J J J
!- <u6
Ij this limit exists, is finite and does not depend on the way the domain
g>6 is contracted toward the point A10, improper integral (9 .2 8 ) is said
to be e o n v e r y e n t. Ij otherwise, it is called d *r ev p en t.
We say that the integral / (A/) dco tends to a finite limit J ,
tl-O),
as ft -v 0, which is independent of the way the domain o>6 is con-

* The dom ains Q ami and all the other dom ains which ore considered
in $ 4 are supposed to he squarable. The symbol 12 denotes tlie closure of 12,
i.e. a closed set which is the union of Q and its boundary- The point jl/ 0 may
belong to the boundary of Q or to its interior but it m ust be an interior poinr
of o>6. The symbol 12 - (05 designates the set of all points belonging to Q and
v. >t Hohmpinsr to cn*. If 12 aDd 0)5 are squarable* the dom ain 12 — (Og is also
squarable.
 CII. 9. IMPROPER INTEGRALS 41'.

Irncted to the point M 0 if for every sequence of domains

(u^|9 • • *9 j ••• (0129^
each of which contains the point M 0 in its interior anti whose diame­
ters satisfy the condition
6n —>- 0 for n -j-oo* (9.30)
the corresponding number sequence
f j /(.l/) c K t(M)d<•>.......... J / (M) doi, . . . (SJ.31)
Q-to6| n l«6n
converges to one and the same limit J irrespective of the choice
of sequence (9.29).
Note 1. For an integral taken over a line segment fa, b\ (i.e. in
the case N — 1) we took tlie intervals of the form [a, b — k\ (which
arc connected point sets) as the domains Q — ^ ^ 2
the domains Q — o>dn and nrc not necessarily supposed lo be
connected.
D e f in itio n 2. Let tke point M n lie in the interior or £2. Suppose
that integral (9.28) is divergent but sequence (9.31) tends to one and
the same limit when (9.29) is an arbitrary sequence of concentric circles
with centre at .1/0 which are contracted toward .V/0. Then this limit
is called the j >v i ne i j j at v a l u e o f d iv e r g e n t i n t e g r a l (9.28).**
Principal values of divergent improper double (and triple) integrals
are applicable lo some problems of mathematical physics.
Sole 2. If 3/q is an interior point of £2, then, when testing the
integral \ I / (3f) dm for convergence, we can replace Q by any
Q*
subdomain Qr a Q containing the point M 0 in its interior and
consider the integral \ \ f (M) dm instead of the original integral
*• *
O'
(compare this with the note after Definition 1 in § J, Sec. 1). If the
singular point M 0 belongs to the boundary of <2 we can take, as £2#,
any subdomain which is the intersection of the domain 12 with
an arbitrary domain £2* for which M0 is an interior point.

* Here we tin not suppose that (9.29) is n monotone sequence of sets, that
is the one satisfying the condition <o(5i id to6> id . . . id td . . . . The
only requirement is that condition (9.30) must be fulfilled.
** Accordingly, the definition of the principal value of a divergent, improper
A"-fold multiple integral involves the sequences of .V-dimensional balls, instead
of the sequences of circles which are contracted to the corresponding point.
4 14 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

Note 3. The case when f (A/) has an arbitrary finite number of

singular points belonging to the domain & or to its boundary reduces
to the case treated in Definition 1 if we appropriately break up the
domain Q into parts as it was done for onefold improper integrals.
2. Integrals of Nonnegative Functions. In this section we shall
consider integrals of nonnegative functions which are simpler to
investigate. The results obtained here will be further used for integrals
of functions with alternating sign.
T h e o re m ;>.«>. Let the integrand f (M) = / (x, y) in inte­
gral (9.28) be a nonnegative function and let us take an arbitrary
monotone sequence of concentric circles with centre at Mo contracting
toward Mo a sequence of type (9.29). Denoting a circle of radius 6
with centre at Mo by K& we can write this sequence in the form
K zd z d . . . i d K 6'
n z d . . . where 6^. —*-0 for n~+- + oo (9.29')
Under these conditions, for integral (9.28) to be convergent it is
necessary and sufficient that the corresponding number sequence
dw, J J / (A/) do 1)' /(AO do, . . . (9.31*)
Q -K Q -K Q-K,.
*>n
be bounded.
Proof. Necessity directly follows from the definition of convergence
of integral (9.28) because if integral (9.28) is convergent sequen­
ce (9.31') must also be convergent, and hence it is bounded.
Sufficiency. Let sequence (9.31') be bounded. Sequence (9.29')
being monotone and contracting to Af<j» the sequence of domains
of integration of integrals (9.31') is monotone increasing, i.e. we hove
£2— cr Q — cz • . . cr Q — cr. ..
Therefore number sequence (9.31') is nondecreasing because the
integrand / (M) = / (x, y) is a nonnegative function. But this
number sequence is supposed to be bounded and, consequently,
it converges to a finite limit J:
lim ff f(M)d<a = J (9.32)
n-^+oo
Q -K

and / (A/) do ^ / . To complete the proof of the theorem

Q -K ,/
On
wc must show that for any other choice of a sequence of domains (9.29)
contracting toward Af0 the corresponding number sequence of form
(9.31) converges to the same limit J. For every domain <0 6 n entering
into sequence f9.291 there exist circles and K&* belonging
 CH. 9. IMPROPER INTEGRALS 415
to sequence (9.29') whose radii 6p and 6'q tend to zero as 6„ -*■ 0
such that
K.*Op Z)0)6un z) (9.33)
Relation (9.33) implies
Q — K Op
. ‘ c z Q —cosn c r Q— K Oq (9.34)
The integrand / (M) being nonnegative, it follows from (9.34) that
H
a -x
/< "> * •< ft /(")«*»<n-Jlka. / (A/) d<a (9.35)

But
lim ft /(.l/)d c o = lim
f f /(M )doi = /
5i-a
V fi-K
Q ^ .,
vp v9
and, consequently, relation (9.35) implies that
lim f f /(.l/)do> = /
n-ui6n
which is what we set out to prove.
The following more general theorem is a direct consequence of
Theorem 9.5.
T h e o re m 9M. Let the integrand f (M) = /(* , y) in integral (9.28)
he a nonnegative function and let (9.29) be an arbitrary sequence of
domains contracting to M 0 (see footnote on page 413). Then inte­
gral (9.28) converges if and only if the corresponding numerical sequence
of form (9.31) is bounded.
Proof. Necessity is proved as in the foregoing theorem. To prove
sufficiency let us take a monotone sequence of circles (9.29") con­
tracting toward Mo and show that number sequence (9.31') corres­
ponding to this sequence of circles is bounded provided that se­
quence (9.31) is bounded. Then Theorem 9.5 will imply that integral
(9.28) is convergent.
The fact that number sequence (9.31') is bounded is proved a
follows. Suppose that
f (M) = const < 4- oo (9.36)

for all m = 1, 2, 3, . . . . Since 8 rn 0 for m — -f-oo we can assert

that for any n there is m such that relation
(9.37)
41G MULTIPLE INTEGRALS. FIELD THEORY AND SERIES

holds. Hence it follows that

Q — K(f‘n a Q — o>5n (9.38)
Therefore we obtain the inequality
I'j /(.!/) do> (9.39)
Q-K.» m
6n
because / (.1/) = / (.r, i/) is nonnegative. From (9.36) and (9.39)
we conclude that the inequality
j j / ( > / ) dm ^ C = const < -4- o o (9.40)
q- ka»
Or
is fulfilled for all » which is what we set out to prove.
Example. Let us prove that the integral
C
wliere C = const > 0 and r = J^(X — x0)3-r (*/— I/o)2
rc
(9.41)
taken over a finite domain Q containing A/0= (x0, //0) as its interior
point converges for a < 2 and diverges for a ^ 2.
According to Note 2 in Sec. I. integral (9.41) over the domain Q
can be replaced by the integral ^ dx dy taken over an arbitrary
sr r
subdomain Q' d Q for which A/0 is an interior point. Let us take
a circle K R of a sufficiently small radius /? with centre at the point
?\1q <10 cl Subdomain Q'. Thus, we must investigate the integral
~ d x d y , C > 0, r ~ Y { x — x0)3 + (y — f/0)2, a = const (9.42)
KR
To do this let us choose a monotone sequence of circles
f<6] . . . 9 M0 where 6n~*-0 for n — “-tt— O O (9.43)
which contracts to anil consider the integral

Kn~Kf>.
II 11
4 d x d y
(9.44)

Transforming this integral to polar coordinates wo obtain

f i
Kn~ Ktn
4 d x d,j =Kn~IIhQn - 4 r d r d<f = *I° d< i’ I - ^ r r d r = -

. 2- a ,r = /f
2nC —----- for a. ^£=2
= ZjiC’ ^ rJ dr = [ 2 — a J r -A (9.451
2nC (In for cc = 2
 UH. 9. IMPROPER IN T E G R A L S 417

Now, passing to the limit in (9.45) as 6n — 0 we sec that inte­

gral (9.40) is hounded for cc <C 2 and approaches infinity for a ^ 2.
(Consequently, integral (9.42) is convergent for a < 2 and divergent
for a ^ 2, and hence the same conclusion applies to integral (9.41).
Similarly, in the case of A' independent variables Xi, x 2, - . x Y
the A'-fold multiple integral

r
converges for a <C -V and diverges for a ^ Ar if AI0 = (x°, . . xn)
is an interior point of the A’-dimensiona 1 domain Q. Tims, the value
a = A’ (equal to the dimension of the space) is a critical one in the
sense that it separates the values of a (a <; A') for which integral
(9.40) converges from the values of a (a ^ N) for which it diverges,
the value a = A' corresponding to a divergent integral.
3. Absolute Convergence. Let a point Af0 belonging to a domain
Q be the only singular point of a function / (M) defined in CS. The
point M 0 may he interior or belong to the boundary of the domain £2.
We suppose that for every domain co for which M 0 is an inferior
point the function / (.\f) is integrable in the ordinary sense over the
domain £2 — to.
D e f i n i t i o n .7. The integral f f / (Al) dto is said to be nbso

converges.

Theorem 9.7. If integral

it is convergent.
Before proceeding to prove Theorem 9.7 we shall indicate some
general properties of convergent improper integrals. We know that
the limit of a sum is equal to the sum of the corresponding limits
and that a constant factor can be taken outside the limit sign. There­
fore we can assert that

12
are also convergent and the
O
equality

holds;
27—0824
418 ML'l.TIIM.E IXTKr.n.M-S. FIELD T1IF.OHV -AND SEItIKS

(2) if an integral j ^ / (.U) dto converges^ the integral H Cf{M)da>

‘if *n
where C — const also converges and
[ ( c i (.1/) do = c ^ ( / (.1/) dm
‘si ‘n
Wc can now prove Tlieorem 9.7. Let us represent the integrand
/ (.1/) as a difference of two noimegative functions f t (M) = | / (AI) |
and / a (-17) = \ f ( M ) \ - j (M):
1 O/) = 1/ (M) 1 - 1 1 / U/) I - / (M)\ = /, 0 /) - ft (M) 0.-47)
lly tile hypothesis, Ihe iiitcgi.il I t IS
D *£i
convergent. We have
ft O W = I / (■»/) I - 1 (M) < 2 | / (,!/) |
mid, according to the conditions of the theorem, the integral
j j 2 |/( A /) |d o - 2 f $ |/( .V ) |d o
Q Q
converges. Therefore, by Theorem 9.0, for every contracting sequen­
ce (9.29) the corresponding sequence of integrals if
2 | / 01/) |do>
Ql"6"
is bounded, burlhormurc, wo have an obvious inequality
j j /.(.V ) dto ^^ j / («^^) Jdisi
Q-tAA/t Q—a>5«

and hence the sequence of integrals f 2 (A/) dw is also bounded.

n
Thus, by Theorem 9.6, the integral 55 /* w dm is convergent.
a
Uut then relation (9.47) implies that the integral ^ / (-‘17) du> is also
a
convergent and the equality
j [ / (M) dm = [ j /, (A/) d o - j j / 2 (A/) do (9.48)
n *u n
is fulfilled. The theorem has thus been proved.
Note. In the case of an improper Ar-fold multiple integral, for
A 2, Llie nmsci.-c o f the a b o v e theorem i«= al<o Irue. that is con-
 C1I. 9. IMPROPER INTEGRALS 419

vergence and absolute convergence are equivalent in the case N ^ 2

(see Sec. 5).
4. Tests for Absolute Convergence.
T h e o r e m 9.8 ( G e n e r a l C o m p a r i s o n Test). Let the inequality
0 < | / (M) I ^ g (M) (9.49)
hold everywhere in the domain (,). Besides, suppose that M 0 is the only
singular point of the functions f (M) and g{M) in the domain Q which
may belong to its boundary or be an interior point. Then
(!) if the integral ^^ g (A/) eta converges the integral / (A/) eta
'd a
converges absolutely;
(2) if the integral j^ / (M) dm diverges the integral g(AJ) eta
n n
also diverges.
Proof. Let us take an arbitrary sequence of domains (9.29) con­
tracting toward AI0. From inequality (9.49) it follows that
|/ ( M ) |d 0 ) < g(M)da> (9.50)
52—“6. q- o6

(1) If the integral g(Af)d(o is convergent the sequence

52
5J 18 (AT) | do>j- is bounded and then, by inequality (9.50),
Q- “«n
the sequence j j j | / (.1/) | dw} is also bounded. Consequently, by

Theorem 9.6, the integral ^J | / (M) | eta is convergent.

Q.
(2) If the integral J j /(;!/) dm diverges the integral J J \j{M)\du>
Q Q
also diverges. Indeed, if the latter were convergent, the integral
\ \ f(M)dto would also be convergent (by Theorem 9.7). The inte-
* »'
£2
gral M |/(.l/)|e ta diverging, Theorem 9.6 implies that for any
* 4'
n
choice of a sequence of domains (9.29) contracting to M0 the
sequence M \ / m |eta is unbounded, liut then, by inequality

(9.50), the sequence g(M)dio is also unbounded, and conse-

27*
420 MULTIPLE I N TE GR AL S , FIELD THEORY AND S E R IE S

qucntly the integral jjg(A /)d<o diverges, which is what we set

Q
out to prove.
T h e o r e m thO ( S p e c i a l C o m p a r i s o n Test), If a junction f (.1/)
defined in Q and having only one singular point A/0 (x0, J7o) belonging
to the interior of Q or to its boundary satisfies the inequality
l/(d O l = l/(*» 0 ) | < “7T.
r C —const ;> 0,
(9.51)
r = V ( x — x0)2 -}- (y — y0)z
where a < 2, the integral j f / (At) dm is absolutely convergent.
o'
Proof. Integral (9.41) being convergent for a < 2, we conclude.
by Theorem 9.S and inequality (9.51), that the integral J ( I / (A/)|do)
d
is also convergent, which is what we set out 1o prove.
iVote. In the case of an improper iV-fold multiple integral
/ (X|, . . ., J.v) dx\ . . . dx'y

taken over an Ar-diinensional domain 12 where the function / (A/) —

— f (xt> . . ., x >v) has only one singular point M q — (x“, . . . » i \ )
in the interior of the domain Q or on its boundary we should lake,
in the special comparison lest (Theorem 9.9), r —
— (x\ — -f* • • • -h(xlV—x'jv)" and a < N.
Example. Let us compute the force with which a material point
M n (x0. y „. z„) of unit mass is attracted by a material body,
occupying a domain 12 in the x, //, 2-space. with volume mass density
P (Af) ^ P (z, //. z)-
Let us write down the expressions of the projections on the coordi­
nate axes of the force of attraction (see Sec. 5 in § 2 of Chapter 2):
^ v : - [ \ \ p (M) J?-p r fL dxdy dz

Fy = %f *\ v\ P (A/) — rJ— dx dy dz (9.52)

O

- j j j Vm dx dy dz
O*
where
/• ] r{ r 'of* ! (if - ; /A 2 i - b . — M = f.r. a . z)
 err. 9 IM P R O P E R INTEGRALS 421

In Chapter 2 we limited ourselves to the case when the point

M0 (x0, i/ot z0) lies outside the body Q. If A/0 belongs to the
boundary of Q or is its interior point the integrals (9.52) are improper
in the general case. Suppose that the density p (A/) = p (x, y, z)
is hounded on Q, that is 0 ^ p (A/) ^ p0 = const for all the points
.1/ 6 Then
—*0 I 1 x — x0
von X

r» I Po r3 r
since r

Mere we have a = 2 < l \ ■=■ 3 and therefore, l>y the special compari­
son test, the lirst integral (9.52) is absolutely convergent. The other
two integrals (9.52) are also absolutely convergent., which is proved
similarly.
In the case N ^ 2 the. improper iV-fold multiple integrals possess
a remarkable property which does not extend to the case N = 1.
Namely, if N ^ 2, ordinary convergence of an improper integral
implies its absolute convergence, that is the converse of Theorem 9.7
is Irue in this case.
5. Equivalence of Convergence and Absolute Convergence in the
Case of Improper Multiple Integral. An improper integral of a func­
tion f (M) converges for Ar ^ 2 if and only if the integral of |/(A /) |
converges. This follows from Theorem 9.7 and the following theorem.
N

T h c o r e m it. 1 it. I f N 2 and. the in t e g r a l . . . J / (.U) dx^..dxN

N

converges, the integral f f . . . \ \ f (A'f) \ dxx . . . dx} also converges.

*J a J
Proof. To simplify the notation we shall take the case N — 2
although the argument below is valid for any N ^ 2. Let a singular
point M 0 of u function / (A/) = / (x, y) be an interior point of a plane
region Q the function is defined in.* Suppose the integral
^ \ f (A/) do) is convergent while the integral ( J I / (A/) | c/o>
n
is divergent. Then we can take an arbitrary sequence {A’n} of con­
centric circles
Q zd A ', zd I \ z zd . . . zd K n zd . . . } A/ 0 (9 .5 3 )

* Wiiliont loss of generality (see Note 3 in § 4, Sec. 1). we can suppose

ihat Ma is the only singular point of / {.!/) in 12- If M<\ lies on the boundary
of we simply lake, instead of A'n in (0.53). the intersections of Ihe circles
A n wilti the domain 12, i.e. their parts belonging to 12-
A‘2 2 MULTIPI.K INTEGRALS. FIELD THEORY AND SERIES

with centre at A/0 contracting toward M 0 and write the relation

lim \ |/( : l/) |d w « -:-oo (9.54)
n-> *30n-X'n
because |/(.17) | is normegative. Sequence (9.53) can therefore be
chosen in such a way that the inequalities
|/( .l/) |r f w > 2 f j |/(.l/)|tfw -j 2«, h = 1, 2, . . . (9.55)
A n — iw i+ t £2 —K u

hold. Let us introduce the functions

/ . ( Vf) = 4 - 1 1 / ( V/) | + /(-U )| ami /_(.W) = - l [ |/ ( , U ) |- / ( . l / ) |( y . 5 t y
We obviously have / + (A/) ^ 0, /_ (A/) ^ 0 and
/ (M) = / + (A/) - /_ (A/), | / (A/) | = /+ (A/) + /_ (A7) (9.57)
From (9.57) it follows that
jj | / (9/) | riu>= f /. (.1/) d'M— jj /_ (.!/■) do) (9.5S)
/Cn —K f i f t /v « —K a i l A n —iC n+ i

We shall suppose that sequence (9.53) is chosen in such a way that

jj f r (M)d«>> f-(M)d<o (9.59)
T\n —K p j-t K n-ivii+l

(if otherwise we can pass from sequence (9.53) to an appropriate

subsequence and replace, if necessary, / (A/) bv —/ (A/)). 3’hen rela­
tions (9.55) and (9.58) imply that
fj /. (.)/)*■ •> | ( |/U/)|rf<"-; », n = 1 . 2, . . . (9.00)
f\n —An+1 &—/Oi

If we break up tlie annulus A’n — A’n+i iulo sufficiently small

squarahle cells Aon the lower Darhoux sum correspon­
ding to the integral I I f t (j\I)dio satisfies, by (9.(>0), the ine-
K/i —JCrtti
quality
y, > jf ! / ( . ! / ) | d<» + n, « = ■!, 2, . . . (9.(11)
O- Kn

* Hero «?[* designates the greatest lower hound of f+ (M) on the cell Ato, d
ci An Anjj.
 CH. 9. IM P R O P E R INT E GR AL S

Wo have m{+ ^ 0 for all these cells since /+ ^ 0 everywhere in Q.

Let us delete all the terms in the sum V wi!+ Ato, for which m{+ — 0
{this of course does not affect the validity of inequality (9.61)).
Denoting by Gn the domain which is the union of the remaining
cells, we can write f (M) = /+ (AI) for M 6 Gn, and
fj | f(M)
‘ n O- Kn
n = 1, 2, . . . (9.62)
Furthermore, we have
j j /(I /) i « = 1, 2, . . . (9.63)
G—i\ n G—A'n
Adding up (9.62) and (9.63) wc derive
(9.64)

where IIn — (12 — A'n) — Gn. If we denote by con the difference

Q — //„. the diameter of o>n tends to zero for n -f-oo. Consequent­
ly, (9.64) implies that the integral (AI) d(o diverges, which
"a
contradicts the hypothesis. Thus, supposing that the integral
I I | / (M) | </o> is divergent we arrive at a contradiction and hence
a
it is convergent. The theorem has thus been proved.
iXa/e. If, in the case JV 2^ 2, wo consider the domains — <»$,,
entering into the dclinil i o n o f an improper iY-fold multiple integral
to be connected Theorem 9.10 remains true. In fact, the domain
IIn — (Q — A’n) — Gn (« = 1 , 2 , . . . ) appearing in the above
proof of Theorem 9.1U can be made connected vvithouL violating the
validity of inequality (9.64). For this purpose it is sufficient to
join together the connected components constituting IIn by squarnhle
strips of sufficiently small total area. The possibility of construc­
ting such strips becomes obvious if we break up the annulus A‘n —
— A'n m into squarable parts entering into the integral sum
rn[*2u0i l>y means of rays starting from the centre M 0 of
the annulus anu concentric circles with centre at. A/0.
In contrast to this, if in the case N = 1, i.e. for an improper
b
integral \ j (z) dx taken over an interval \a. 61, wc take, instead
41
of the sequences of intervals of the form | a , 6 — /*! (a > 0 , K
Al'i MULTI PI.E INTEGRALS, FIELD THEORY AND SERIES

<C. b — «) entering into the corresponding definition (see rela­

tion (9.17) in § 2), exhaustive sequences of arbitrary disconnected
domains, this will lead to a narrower class of functions for which
the improper integrals exist. Indeed, in this case only the functions
absolutely intograble in the improper sense will constitute the
class of functions with the convergent improper integrals and thus
the functions the integrals of which are conditionally convergent
will be excluded (the classes of functions absolutely integrabie in the
improper sense will obviously be the same in both definitions).
6. Improper Integrals with Infinite Domain of Integration. The
integrals over unbounded domains whose integrands are functions
bounded in any finite domain are investigated in quite analogous
fashion. /Vs an example, we shall formulate the definition of an
improper integral over an unbounded domain and a sufficient condi­
tion for convergence.
Dcf in it ion 4. Let Q he an infinite {unbounded) domain. A sequen­
ce of finite (bounded) subdomains
S22, . . ., £>n, . . . (9.or>)
is said to be e x h a u s t i v e if for any It ;> 0 there is m = m {ft)
such that all the points of the domain il lying within the circle of radius ft
with centre at. the origin belong to all for u > m (It).
D e f i n i t i o n .5. Lei a function f (M) defined in an infinite domain Q
be integrabie in the ordinary sense on every finite subdomain. If for
any choice of exhaustive sequence (9.65) the corresponding number
sequence
j j / (A/) <K f j / 0 0 dio, . . ., t/vc), ...
121 12* Q/i
converges to one and the same finite limit J the integral f J / (M) d«>
is said to be c o n v e r g e n t : if otherwise the integral is called d i v e r ­
gen t.
S u f f i c i e n t i'ontlition. f o r C on.re r g c n c e . If f (A/) - / {x, y)
satisfies the requirements of the foregoing definition and the inequality
I/ 0 0 | '£ —— where C —const > (J,
ra

a - consi > 2 and r —Y \x — x0)a i (y — i/o)3

is fuifilled where l / 0 — (.r0, y 0) is a fixed point belonging to Q the
integral \ j / (A/) day is convergent.
 CH. 9. IMPROPER INTEGRALS 421

In conclusion we note that the general theorems analogous tc

Theorems 9.5, 9.6, 9-7, 9.S and 9.10 also apply to improper integral?
with unbounded domains of integration.
7. Methods of Computing Improper Multiple Integrals. A convergent
improper double integral can be reduced to the corresponding twofold
iterated integral like a proper double integral under the following
conditions:
(1) if the integrand is a nonnegative (nonpositive) function it is
required that the iterated integral of this function be convergent;
(2) if the integrand is a function with alternating sign it is supposed
that the iterated integral of its modulus converges.*
The method of changing variables is applied to a convergent
improper A'-fold multiple integral according to the same rules as
in the case of a proper A’-fold multiple integral.
Here we do not present the proofs of these general assertions and
confine ourselves to an example in which we encounter reduction
of an improper double integral to an iterated one and change of
variables in an improper integral.
Let it be necessary to evaluate the Ktiler-Poisson integral J —
= ^ e ~ x2 d x (see also Example 1 in Sec. 4, § 1 where the conver-
b
genre of this integral was established). The value of a definite integral
remains Ilie same when the notation of the variable of integration
-L-zo -i-oo
is changed and therefore r ~ x~ d r = i e~u~ dij. lienee we
*
o o
can write
-J-flO -I-50 -|-00 +c*>
J 2 — ^ e~x2 d r ^ e~y~dij ^ c ~ y2d \ j ^ d x —
b b o n
-joo -[-o©

= f dx ^
i) b
and the iterated integral is convergent. The double integral
-j-OO+OC
j' f e -^'-'ru-idx dy
o b

* The situation is similar in t.be rose of on i mi proper V-folrf multiple integral

for .v >
426 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

is also convergent which is implied by the sufficient condition for

convergence given in Sec. 6. Consequently,

J3 dx dy

Passing to polar coordinates we obtain

n
J8 j «-rW r— £
0
Hence, we have

/= j e - * ’- d x = X p -
0
This technique of evaluating the integral was developed by Poisson.
 Integrals Dependent
10 on Parameter

In this chapter we consider the properties of integrals dependent

on a parameter which are effectively used in analytical methods
of mathematics and mathematical physics. Such important integrals
as Euler's integrals of the first and the second kind (the beta and
gamma functions'^ see § 3), integrals of the type of a potential function
etc. belong to the class of integrals dependent on a parameter,

§ 1. PROPER AND SIMPLEST IMPROPER INTEGRALS

DEPENDENT ON PARAMETER
1. Proper Integrals Dependent on Parameter. Let u — f (x, y)
be a function defined in a rectangle II: a ^ x ^ b> c ^ y ^ d.
We suppose that this function is integrable with respect to x on the
interval a ^ x ^ b for every value of y belonging to the interval
c ^ y ^ d. Then the integral
b
J iy) ~ j / (x > y ) (io.i)
a
(dependent on the parameter y) is a function of the parameter y
on the interval c ^ y ^ d. Here we shall study the properties of
integrals (10.1).
T h e o r e m U h l ( O n the C o n t i n u i t y of a n I n t e g r a l D e ­
p e n d e n t on a P a r a m e t e r wit h l t e s p e c t to the P a r a m e t e r )•
/ / the function f (x, y) is continuous in the closed rectangle Tl: a ^
b
C x ^ by c ^ y ^ d, the integral J (y) — ^ / (x, y) dx is a con-
a
tinuous function of the parameter y on the interval c^Ly ^ d.
Proof. Since the function / (x, y) is continuous in the closed rect­
angle II it is uniformly continuous. Hence, for every e ;> 0 (here is
6 — 5 (e) 0 such that Lhe in eq u a lities
\x' — x" | < 6 (e) and I y f — I/" | < 6 (e) (1U.2)
42S MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

imply the inequality

\f(x', y ' ) - f ( / , / ) | C j ^ (10.3)
(here 6 (e) depends solely on e and is independent of the positions
occupied by the points (x', if) and (x", y") within the rectangle n
provided that inequalities (10.2) are fulfilled). In particular, putting
x — x* — x we see that for any if and i/" belonging to the interval
c ^ y ^ d and satisfying the inequality
|;/' - y " | < 5 ( f . ) (10.2')
and for all x, a ^ x ^ by the inequality
| f ( x , y ,) - f ( x , u" ) \ < ^ (10.3')
holds. Therefore, for any y' and y" belonging to the interval c ^ y ^
<1 d and satisfying inequality (10.2') the inequality
b
IJ (y') —J (*/") I= [ j If (*, y') —/ (3 %y")\d.%

y ' ) — f ( x, if) | (b — a) — ?

is fulfilled, which means that J {y) is uniformly continuous on the

interval c y ^ d. The theorem has been proved.
f-oroflart/. Under the conditions of Theorem 10.1, the function
L
F (u, y) = j / (x. y) dx is continuous in the closed rectangle
U

n*:a ^ u. ^ 6 , a ^ v ^ 6 , c ^ y ^ d
Proof. The function / (x, y) being continuous in the closed rectan­
gle II, there is a constant C, 0 < 6’ < -b00*such that | / (x, y) | <C C
everywhere in II. Therefore, for any points (u \ vf, y') and (u-", v ”, y “)
of 11* we have the following inequality:
if v

| F (it\ v \ y ) — F (u\ v \ if | = I f / (x, y ) dx — f / (x, y") dx

Iu'•/ *
uw
V* u"

< | j 1/ (*, y ) — i (3*, y ‘) 1dx | 1j / (x, if) dx + j

r'
■t j \ / v ) dx < I f i / (x, if) — / (x, if) i </x |-f
\ r#
J Ium
41 *

- I T ; |a " - ! / | 4 c |ir - ,/| (10./,)

 CH. lf>. IN T E G R A L S D E P E N D E N T ON P A R A M E T E R 421

Now lei the point («', i/, if) be fixed and the point (u. v ”, y
tend to (a', vr, {/'): (u ”, v ”, y ") (»'. v \ y ‘). Then, by Theorem 10.1
the first term on the right-hand side of (10.4) tends to zero. Th<
second and the third terms on the right-hand side of (10.4) alsc
tending to zero .is (u”, v", y ”) —►(u'. v , if), the theorem has thus
been proved.
T h e o r e m 10.2 (O n D i f f e r e n t i a t i o n o f a n I ntef/vul
D e p e n d e n t on a 7>a r u n i e t e r wit h R e s p e c t to the P a r a -
ntetevj. If the function f (x, y) and its partial derivative f'y (x, y) are
continuous in the rectangle II: a ^ x ^ b, c ^ y ^ d, the integ-
t.
ral J {y) = ^ f (x, y) dx is a differentiable function of the parameter
a
y on the interval c y ^ d, and the relation
b b
f L L ^ ^ ^ f ( x , y ) d x = /y(x, y)d:c (10.5)
a ci

is valid for all y belonging to this interval.

Note. Formula (10.5) is known as Leibniz’ rule (formula) for
differentiating on integral with respect to the parameter it depends
on: the derivative of the integral with respect to the parameter is equal
to the integral of the derivative of the integrand with respect to this
parameter.
Proof. We must show that
b
,. J (y ! - A i / ) — J ( y ) ( ,, , . ,
lim
\ It. %
fi — 5— ^ ------- - = 1
i M * . y)dx

For this purpose we shall prove that the difference between the
variable quantity ^ (\ y Qj and t|ie integral
b
( f y Or, jf ) dx tends to zero when S y —* 0. By virtue of Lagrange’s
a
formula of finite increments, we have
J (// — A //) — J (y) /(•<■. y - r A / / ) —/ U , y ) J f /- /
\>t -----------a?---------------------/»(*- y 0 A//) dx

where 0 < l ) < 1. (Consequently, we can write

./ {u —a»/)—j r .., j r* w / ~ .

--------- ^*1 ------ :-------%1 b y) d r = »>1 Ih (*. .'/ OAy) — /;• (X, y)l dx
Ct a
4W MULTIPLE INTEGRALS, FIELD T H E O R Y AND S E R IE S

Lot us estimate the above difference for sufficiently small values

of | A#/ |. Let there be given c 0. Since the derivative f'v (x, y)
is continuous on the closed rectangle H it. is uniformly continuous
on it and therefore there is 6 (e) > 0 such that, for | Ay | < 5 (e),
the inequality
!/«(*» y i-Ay) —/;(x, y)|<^—
liolds for all x £ (a, 61 and any y am! y Ay belonging to the inter­
val [c, d |. We have 0 <C 0 < 1 and hence for all x, y and y -|- Ay
mentioned above the inequality
I/v (r, y i-0Ay) — U (*» It) I< ] ~ Z
is fulfilled. Tims, by (10.0), we have

J (.v —J (i/) — ( /*; (*> y) dx j = j j l/y (x, y i- OAy) — f v (x, y)l dx

A.v

j I /y (J, y — o Ay) — fy (x, y) dx

0— u
(b— a) —c
for all | Ay J< 6 (c). The theorem has been proved.
The*} r e m JOJi ( O n I> iff ev e n t i a t in*j a n I n t e g r a l l>e-
p en f fe n t o ft a f*a ra m e t e r iff* o.se TJnt i ts of' t ntetf ration.
A l s o I)epen*t on the T a r a m e t e r teitl* lie.spect to the.
T a r a m e t e r ) , Let f (x, y) and ftJ(x, y) be continuous in the rectangle II:
a x 5 ^ 6, c ^ y ^ d, and let x = x, (y) and x — x-> (y) be diffe­
rentiable functions defined on the interval c ^ y ^ d and satisfying
the condition a <C xt (y) < 6 (t — 1, 2). Then the derivative of the
integral
xHtl)
J (y) ~ j J(x, y) dx (10.7)

with respect to the parameter y exists and is equal to

*2(;o
(.'/) j iy U , y) dx r / (^s (//), y) — / (*i (y), y) (10.8)
*^i(.</)
Troof W e have
./ fy) - /’’ (x, (y), x- (//), //) ( 1 0 .0 )
 CH. 10. I N T E G R A L S D E P E N D E N T ON P A H A J IE T E H 431

where Ilie function F («, v, y) = J / (x, y) dx considered for a ^

U
^ u ^ b. a ^ p ^ b and c ^ y ^ d possesses the continuous par­
tial deriva liv e s

F'u - — / (w, ij ) , F’v = / (i>, i/), -= j Fj (x, */) fix (10.10)

U
By Theorem 10.3. the partial derivative F'y exists, and the Corollary
of Theorem (10.1) implies that it is continuous. The functions x =■
— x, (//) and x = x-> (//) being differentiable, we can apply the rule
fur differentiating a composite function to integral (1U.S) which
results in equality (lU.tS). The theorem lias been proved.
T h e o r e m 10 .4 ( O n I n t e p r a t i o n o f a n I n t e g r a l D e p e n ­
den t on a i*nru m e t e r teith lie-sped to the P a r a m e t e r ) .
It j (x, y) is a continuous junction in the rectangle II: a ^ x sC 6,
c^ y d, we have
d if h b il
j J (u) dij = f (xy y ) d x = j dx j /(x , y) dy (10.11)
c c u a c
h
which means that to integrate the integral J (y) — j / (x, y) dx
a
with respect to the parameter y ut can integrate the integrand j (x, y)
with respect to this parameter.
/Von/. Kquality (10.11) is a consequence of the theorem on reduc­
ing a double integral to an iterated one (see Chapter 1, § 5).
We shall present here another proof of the theorem which can
easily he extended In the case of an arbitrary dimension A' (sec § 4).
Moreover, instead of equality (10.11) we shall establish a more
general relation
a t t d
j dy j / (x, y) dx = j dx \ j (x, y) dy for a t (10.12)
c a a c

Introducing the notation

d t t d
T (/) -= j dy ^ j (x, y) dxy i|? (0 = j dx j / (x, y) dy (10.13)
c a a c

we see that it, is sufficient to prove that q>'(/) = i[-'(/) for a <Z t b
and that q-- (a) = if (a) because this obviously implies the identity
q> (0 ~ (/)* t C Ia. b\.
432 MULTIPLE INTEGRALS, FI EL D T H E O R Y AND S E R I E S

The equality(p («) = t|? (a) is apparent since <p (a) = 0 and
t
^ (a) = 0. Putting F (J, y) = j / (x, y) dx we can write cp (/) —
n
d
= J F (F y) dy where the function F (/. y) is continuous in the
C
rectangle II*: a ^ b. c ^ y ^ d, by virtue of the Corollary
of Theorem 10.1. Furthermore, by the hypothesis of the theorem,
the derivative F\ (/, y) — / (t, y) is continuous and, consequently,
by Theorem 10.2, we have
ri ci

<p' (0 = j Ft (C V) dy = j / (C y) dy (10.14)
c c
d t
Putting £ (x) — f / (x, y) dy we obtain ij> (/) = f c (x) dx. The
* w
c a
function £ (x) being continuous in x on the interval a ^ x ^ b
(by Theorem 10.1), the theorem on differentiating a definite integral
with respect to its upper limit of integration implies the relation
t d

'f' (0 = \ dxs= l (/) = j /(*, y)(iy (10.15)

u c

From (10.14) and (10.15) it follows that cp'(/.) ==il>'(/) fora ^ ^ 5.

Hence, hy the equality «p (a) — i|? (a), wc have cp (l) a sif (/) for
a * £ / ^ b . In particular, <p (6) — ij> (b), that is equality (10-11)
holds, which is what we set out to prove.
2. Simplest Improjier Integrals Dependent on Parameter. Theorems
10.1, 10.2 and 10.4 can easily he extended to improper integrals
of tlie following special type:
b
J (i/) = j / (*, //) g (*) dx (10.10)
a
where llie function / (x, //) is continuous and the function g (x)
raav be discontinuous, in the general case, but such that the integral
b
j I g (*) | dx is convergent, including the case when one or both
a
limits of integration are intiriite.
Now we proceed to formulate the exact conditions of these general­
ized theorems*.

* Theorems 10.T, 10.2' ami 10.4' are used in mathematical physics and
iu the theory T r-'ori^r’^ intoernl.
 C1I. 10 IN T E G R A L S D E P E N D E N T ON P A R A M E T E R 433

T h e o r e m 1 0 . T ((! e i i e r a l i n e d T h e o r e m oil the ( 'o u t f ­

it n o u s D e p e n d e n c e o f a n I n t e g r a l o n the T a r a m e t e r ) .
It the function f(x. tj) is continuous and bounded for a*Cx<C. + °o,
-f »
c^Cy^Cd, and the integral j \ g (x) \dx converges, the integral
a
+3°
J ( y ) = \ /(*> ij)g{x)dx (10.17)
n
is a continuous function of y on the interval c ^ // ^ d.
T h e o r e m lO.'T ( G e n e r a l i z e d - T h e o r e m on D i f f e r e n t i a t i n g
a n I n t e g r a l trifh Itesgect to the T a r a m e t e r ) . If the function
f(x, y) and its partial derivative fu (zt y) are continuous and bounded
•i *>
for a^.x<Z ; oo, c-f.y-s'-d, and the integral ^ | g (x) | dx is con-
a
vergenl, integral (1U.17) is a differentiable function of the parameter y
for c gC. y ^ d and the equality
4-00
J ’ { y ) = ft'\ f u (•*> y ) z (*)dx (io .i8 )
a
holds for all y (; Ic, <21.
T h e o r e m 1 0 . 1' ( G e n e r a l i z e d T h e o r e m on I n t e g r a t i n g
a n I n t e g r a l trifh lie s fieri to the T a r a m e t e r ) * Under the
conditions of Theorem 10.1 integral (0.17) is an integrable function
of the parameter y on the interval c ^ . y g ^ d and
d d
j J (u) dy = j dy j / (x, y) g (x) dx =
c c ti
-OO (i
= j (g (*) j /(*> y)dy) dx (10.19)
a c

As an instance, let us prove Theorem 10.17. Let | / (x, y) | < C —

-(’OO
= const for < i < i < ~ o o , c ■<.(/< <2 and ^ |g(;c) | <2x< K < -J- oo.
a
-f-oo
C
Let €>>0 he taken arbitrarily. rl he integral t \ g {x) \ dx being
*'
a
convergent, there exists a sufficiently large / >* a such that
■\00
2C f ||f(x) |< i x < 4 . Taking such a fixed value of / and choosing
Ji *
is -Do^i
 434 MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

arbitrary values y' and i/" belonging to the interval c ^ y ^ d

we can represent the difference J {y') — J ( y “) in the form
t
J — J (!/") = j [/(*, y ' ) ~ f ( x , y"))g(x) dx---
a

4- \ [f(x, y') — /(x , y’’)\g(x) dx (10.20)

«
/
Since the function / (x, y) is continuous in the rectangle a ^ x ^ /,
c ^ y ^ dy it is uniformly continuous. Therefore there exists 6 =
= d (e) :> 0 such that for any y' and y" belonging to the interval
c - ^ y - ^ d and satisfying the inequality I if — U" l’< & (e) we
have the inequality

I y ' ) —f{z* y*)l < 2 ^-

for all x, a * C x ^ l . Then equality (10.20) implies the relation
i
IJ ( y ' ) - J ( y l I | / ( x . / ) - / ( * , </") I \ g (*) | dx +
a
-foo
i- j {I/( * . *')! + !/(* . le (*)!<**«

/ -f'OO

a /

< 2^ A ' J - ^ = £ for | j/' — | < 6 (e)

which means that the integral J (y) is a continuous function of y
on the interval c ^ y ^ d.
Let the reader prove Theorems 10.2' and 10.4' for integrals (10.17)
and also rephrase and prove Theorems 10-1', 10.2' and 10.4' for
integrals (10.1(5). We only note that in the case of integral (10.1G)
the boundedness of / (xt y) and fy (x, y) is implied by the conti­
nuity of / (x, y) and j'y (x, y) in the domain a^CxsCb, c * C y ^ d ,
and that it is unnecessary to break up the interval of integration
a ^ x ^ b into parts as it was done in the proof of Theorem 10.1'
for integrals of type (10.17).
Differentiation and integration of integrals with respect to para­
meters are widely applied lo evaluating integrals dependent on
parameters and also for computing integrals not involving para­
meters after a parameter is appropriately introduced.
 CH. 10. IN T E G R A L S D E P E N D E N T ON P A R A M E T E R 43;,

Example. Let us evaluate the integral

J (y) = J e~ax dx where a ~ const > 0 , — A-g^y^A

° ( 10. 21 )
Fulling / (x, y) = - nj-xy-
C and g (x) = e~ax we see that /(x , y) and
}'v (zi u) are continuous and bounded for 0 ^ x < C + o o , —
-|-oo —poo
and the integral J j g ( x ) |d x = j e-°-x dx —— is conver-
o 5
gent. Therefore wo can apply generalized Theorem 10.2' for the
integrals of form (10.17). Performing differentiation with respect
to the parameter under the sign of integration we obtain
+oo
J* (y) = J e~ax cos xy dx
o

Integrating by parts with respect to x in the latter integral twice

we find
55^55- <1(>-22>
By (10.21), we have / (0) = 0 and, therefore, integrating (10.22)
from 0 to y we derive
v
a
w - J (X2 y *'22 dy = arc tan ~a
There is another way of completing the evaluation of this integral.
Namely, after relation (10.22) has been obtained, we can integrate
it with respect to y which results in
J (y) = arc tan —
06 -j- C, C — const
But, by (10.21), wc have J (0) — 0, and, consequently, putting
y = 0 in the above relation involving the constant C we see that
(7 = 0. Both approaches involve a known value of the integral
in question for a particular value of the parameter y.

§ 2. IMPROPER INTEGRALS DEPENDENT

ON PARAMETER
Let a function u — f (x, y) be defined for 0 ^ x < C - p o o ,c « ^ y ^ d
and let, for every value of y £ fc, d], the integral

J (y) = f / y) dx (10.2.°,)

28*
436 MULTIPLE IN T E G R A L S , PfELD THEORY AND S E R I E S

be convergent. Then J (y ) is a function of y defined on the interval

(c, d\. According to the definition of the improper integral, we have
-J-oo I
J (y) = f /(x , y ) d x = lim f / (x, y) dx (10.24)
J £~fr.J«90 J
a a

In the case of integrals of unbounded functions we have a similar

situation. For instance, let a function a — / (x, y) defined for a ^
< b, c ^ y < d be uul»ounded for x b — 0 and let the
integral
b b-h
J* (y) = f / (*, y) dx = lim \ f (x, y) dx (10.25)
J
a
>.-►« f-o J
a

be convergent for every value of y belonging to the interval [c, d]

Then ./* (y) is a function of y defined on the interval (c, d\.
1. Uniform Convergence. The notion of uniform convergence plays
an important role in the theory of improper integrals dependent
on a parameter. Uniformly convergent improper integrals can he
operated on like proper ones (see Sec. 3 of § 2). We shall begin
with the definition of uniform convergence for integrals with infinite
limits of integration.
D e f i n i t i o n 1. We say that integral (10.23) is u n i f o r m l y
e o n r e r y e n f wit h r e s p e c t to the p a r a m e t e r y o n the i n t e r -
rat c y ^ d if, given au arbitrary e ;> 0, there is f, — t* (s)
suck that the inequality

J (y) -- \ f (*, y) dx I= I \ f (x, ij) dx ( 10. 20 )

•'
a I I «
is fulfilled for all I z> L (e) and all y £ Ic, d\ simultaneously.
Uniform convergence of an integral of an unbounded function
is defined in a similar fashion:
D e f i n i t i o n 2. Integral (10.25) is said to c o n e c r y c u n i f o r ­
m l y i c i t h r e s p e r f to the p a r a m e t e r y o n the i n t e r e a l Ic, r/]
if for every e > 0 there exists 6 = 6 (c) > 0 such that the inequality
b~\
I I* (y) — \ / (x, It) dt (10.27)

holds for all K<C b — a satisfying the condition 0 < X < 5 (b) and
for all y £ |c, d\ simultaneously.
 CH. 10. I N T E G R A L S D E P E N D E N T ON PA R A M E T E R 43’i

Examples
-f-oo
1. The integral J (y) — \ ye~xv dx is convergent for every y
o
belonging lo the interval 0 C y 1 but not uniformly convergent.
Indeed, we have
-J-eo 4~°°
f ye~xydx = f e~tdt = e~ty
*' »'
l ly
Therefore, for an arbitrarily large fixed value / >* 0 the latlei
integral exceeds for all values of y located sufficiently closi
\
to zero and, consequently, for e = ~ there is no L (e) such that
for Z:> L (e) and for all y belonging to the interval 0 ^ y ^ 1
the inequality
+ 00
j ye~xydx < E - 2
i
is fulfilled.
But if the interval 0 < y CJl is replaced by an interval of llu
4-ao
form 0 <C */<: 1, 6 < 1, the integral J (y) = ^ ye~xy'da
o
is uniformly convergent on the latter interval. In fact, we ha\t
•f-oo -f-oo
C ye~x,J dx — f e~l dt = e“hj^.e~16 for U <C ^ 1 anc
i *u
therefore the inequality
-J-oo
f ye~xydx < e

0
holds for / > —^— . 0 < e < 1, and all y belonging to the inter­
val 0 <i ft < y ^ I-
i
/i
2. The integral J (y) — \ yxv~ 1 dx is convergent for every y be
A*
o
longing to the interval but not uniformly convergent
To show this, we take into account that here the integrand is un
x
bounded for x >>0 ! 0. Let ns estimate the integral \ y x y~] dx —
438 MULTIPLE INTEGRALS, PIELD THEORY AND S E R I E S

= Xy — kv. For any arbitrarily small and fixed X > 0 this integ­
ral tends to unity as y —*-0-|-0. Hence, for e = -s-, there is no

5 = 5 (e) such that the inequality \ \ y x v~i dx < e = y simultane­

ously holds for all y belonging to the interval
But if we replace the interval (X ^ j/^ l by an interval 0<C8o-^
i
60<Cl, the integral / (y) = J yx*'-1 dx converges uni-
o
x
formly on the .latter. Indeed, we have £ yx v~ 1 dx = XvO . 6® for

OCkCi and C l . Therefore, if 0 < e < C l and A<Ce°°,

we obtain the inequality \ yx^~i dx < £ for all y belonging to the
o
interval
2.* Reducing Improper Integral Dependent on Parameter to a
Functional Sequence. An improper integral dependent on a para­
meter can be reduced to a functional sequence which makes it possible
to prove the fundamental theorems concerning such integrals on the
basis of the corresponding theorems on functional sequences.
If an integral
+oo
• '( » ) = j /(* , y)dx (10.28)

converges for every y £ [c, dl, then, for an arbitrary number sequence
lu l 2y . . /*, . . liiii Jft = where lh ^ a for k =
= 1, 2, . . . » the functional sequence
*n
j
Fh (y) = / (z, y) dx, k = 1, 2 , . . . , c<y<d
a

is obviously convergent to J (y) on the interval [c, d\.

The theorem below holds under the condition that integral (10.28)
converges for every y belonging to the interval c ^ y ^ d.
4-»
T h e o r e m 10.5. For the integral J (y) = ^ / (x, y) dy to be
a
uniformly convergent with respect to the parameter y on the interval
 CH. 10. I N T E G R A L S D E P E N D E N T ON P A R A M E T E R 439

[c, dl, it is necessary and sufficient that tke functional sequence

ik
Fk ( y ) = [ f (x , g)dx, k= 1,2,... (10.29)
o
converge uniformly to J(y) on the interval c ^ y ^ d for any choice
of the sequence llt I21 • . ., lh* • * -♦ lim In = + 0 0 .
k^-^oo
Proof. Necessity. Suppose integral (10.28) is uniformly convergent
on the interval c -^ y ^ d . Then, given an arbitrary e > 0, there
is L (e) such that, for all I z > L (e), the inequality
1

(y) —j / ( * . y)'dx
a
is satisfied for all y £ lc, dl simultaneously.
Let lh — + 0 0 for k —v -poo (where lk ^ a, k = 1, 2, . . Then,
there exists N (e) such that lh > L (e) for all k ^ N (e). Conse­
quently, for all such fc, we have, by the manner L (e) has been chosen,
the inequality

UO/) — J (y) —j /(^i y)dx <e

which holds for all y £ (c, dl. This means that sequence (10.29)
is uniformly convergent to integral (10.28) on the interval c ^ y ^ d.
Sufficiency. Let us show that if every sequence of form (10.29)
where lim lh = + 0 0 converges uniformly to J (y) on the
h—*100
interval c * £ y ^ . d y integral (10.28) is uniformly convergent with
respect to the parameter y on this interval. In fact, if we suppose
that integral (10.28) (which is, by the hypothesis, convergent for
every y £ (c, d\) converges nonuniformly with respect to y on the
interval c<Cy<:d, there must exist e0 such that for an arbitrarily
large L there arc / > L and y £ [c, dl such that the inequality
1

J (y) —j / (*, y ) d x >?0

a
is satisfied. Then, making L assume the values L — 1, 2, 3,
. . A*, . . . we obLain the corresponding number sequence ltn k -
— 1, 2, . . ., ;> /c, and a sequer.ee y h £ [c, dl, k = 1, 2, • • • I
for which

(yh) |> * c.
A
 440 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

lk
This means that the functional sequence Fk (y) = ^ / (x, y) dx,
a
k = 1, 2, . . thus constructed, converges nonuniformly on the
interval c*£y ^ d, which contradicts the hypothesis. The theorem
has been proved.
Note 1 . If / (x, y) is a function, retaining its sign, for instance,
4-f°
a nonnegalive one, then for the integral d (y) = j / (x, y) dx
a
to he uniformly convergent with respect to y on the interval c ^ y * £
<ld, it is sufficient that functional sequence (10.29) converge to
J(y) uniformly for at least one particular choice of the number seque­
nce /j, Z2, • • •> ....
1

Indeed, if / (x, y) is nonnegative we have j / ( x , t/)dx ^

a
lh
/ (x, y) dx for all I ^ Z/t. Consequently,
a

4 n
— (*> y) d* < J (y) — j / (*i y) dx <C r for a 11 / > h
a

and for all y £ k, d I simultaneously provided that lfi is sufficiently

large.
Note 2. If the function / (x, y) is continuous for a .r -r°°,
c ^ . y ^ d and retains its sign (for instance, is nonnegative) and
-]-ao
the integral 1 / (x, y) dx is a continuous function of the parame-
a
ter y on the interval k , d], this integral is uniformly convergent
on k , d\.
In fact, taking an increasing number sequence /,, Z2> • • •-*
. . ., lim l h — -l-oo (lh a, k — 1, 2, . . .) we arrive at the
-00
functional sequence
i*
Fh { y ) ~ j /(-r, y) dx, /t=l,2, ... (A)

The function / (x, y) being nonnegative, sequence (A) is monotone

nondccrcasing, and, by Theorem 10.1, the functions Ffi {y)f k —
= 1 2 ............ are roriMnnou«. Upside**. Ihis sequence converges to the
 CII. 10. IN T E G R A L S D E P E N D E N T ON P A K A M ET E R 44

continuous function
+oo

J (y) = j / (*, y) dx (b ;
G

on the interval But then Dini’s theorem (see Chapter 8,

§ 2, Sec. 1) implies that sequence (A) converges uniformly lo its
limit (B) on the interval lcf d\ and, consequently, by Note I, the
-1-00
integral J {y) = J / (x, y) dx is uniformly convergent on this
G
interval.
Note 3. The improper integral
h b-J.

J* (x/) = f / (*, y ) dx = lim f / (*> y) dx

J
ci
[X->0+0 a^
can be similarly reduced lo a functional sequence f t (y) =

/ (x, */) <7x where Xh —>-0-1-0 for Zr - } - c © but we shall

not go into particulars here.
3. Properties of Uniformly Convergent Improper Integrals Depen­
dent on Parameter.
T h e o r e m 10 M. If / (x, y) is a continuous function defined in
a domain a x <. -j-oo, c ^ y ^ d, and the integral J (y) =
+ 00
= ^ / (x. i f ) dx converges uniformly with respect to the parameter y
a
on the interval y ^ <7, the function J(y) is continuous on this
interval.
Proof. Take an arbitrary numerical sequence Z1? Z2, • • •» Z/., • • •
. . ., lim /ii—►-f-oo, (lu ^ a), and consider the sequence of
h—►-f-oo
functions
{h
F h ( y ) = \ f (x, y)dx, h — 1 , 2 , . . . , c < / / < d
a
By Theorem 10.1 on the continuous dependence of a proper integral
on the parameter, the functions f\{y)y k *=■ 1, 2, . . .. are con­
tinuous on the interval y ^ d . But Theorem 10.5 implies
that this sequence is uniformly convergent to the integral J(y) =
+ oo
/ (x, y) dx on the interval c ^ y ^ d> and conseqin>iii lv tho
442 M ULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

function J (y) is continuous as the limit of a uniformly convergent

functional sequence. The theorem has been proved.
T h e o r e m 10.7 ( O n D i f f e r e n t i a t i o n o f a n I m p r o p e r I n t e ­
g r a l w i t h I t e s p e c t to the P a r a m e t e r ) . Let /(x , y) andf'v (x, y)
he continuous for c ^ y ^ d, a ^ x «< -h°°» and let the integral
+06
J (y) = j / (*, y) dx (10.28')

be convergent on the interval c ^ y ^ d. Suppose that the integral

«f-oo

j ft,(x, y)dx (10.30)

a

converges uniformly on this interval. Then J(y) is a differentiable

function of y for y £ (c, d\ and
-4-00 -J-oo

^ 7 5=='^' j /(* , y ) d x = j fy (X| y)dx (10.31)

a a

Proof. Take an arbitrary number sequence Z|, Z2, . . lk, .

. . ., iim — -fco, lh ^ a, and consider the sequence of
functions
*K
Fh(y) = ^ f (x, y ) d x t k = i t 2t * . . , c < y < : d
+»
-which is convergent to the integral J (y) = J / (x, y) dx on the
a
interval lc, <21. Theorem 10.3 on differentiating a proper integral
with respect to the parameter implies
*h ln
F'*(y'> = l & \ /.(*. y)d**= y)dx, fc = l , 2, . . . .
a a
c y^ d
where all the functions F£ (y), fc = 1, 2, . . are continuous on
lc, dl. Integral (10.30) converging uniformly, the sequence of func­
tions Fh (y) is uniformly convergent to integral (10.30) on the inter­
val lc, d J, Hence, we have
+°°
Fu (y) J (y) on lc, d|, F'h (y) j /y(x, y)dx on [c, dl
 CH. 10. N T E G R A L S D E P E N D E N T ON P A R A M E T E R 443

where the symbol indicates uniform convergence, and (y)

are continuous on lc, d\. Consequently, by the theorem on diffe­
rentiating a functional sequence (see Sec. 3 in § 2 of Chapter 8),
J (y) is a differentiable function on lc, d\ and the relation
+°°
J '( ,y ) = \ f 'A * ' y ) d * (10.31)
a

holds for every y belonging to the interval c ^ y d which is

what we set out to prove.
Th e o r e m 10.8 ( O n I n t e g r a t i o n o f an. I m p r o p e r I n t e g r a l
w i t h R e s p e c t to the P a r a m e t e r ) . If / (x, y) is a continuous
function in the domain a ^ x <C + ° ° . c ^ y ^ d, and the integral
+oo
/ (y) = j / (*, y) dx (io.2S")
a

converges uniformly on the interval c ^ y d, then

it d -}-oo -j-oo d
\ J {y ) dy = ^ dy j / (x, y)dx — \ dx j f (x, y)dy (10.32)
c c a a c

Proof. For any number sequence lif 12 , . . . . . (lJt ^ a,

lim If, = -J-oo), the corresponding sequence of functions
k-+G0
**
*'h {y) = j / (*. y)dx, k= 1,2,...
a

ii uniformly convergent to J(y) on lc, d\ which is implied by

Theorem 10.5 on reducing a uniformly convergent integral to the
corresponding functional sequence. By Theorem 10.1 on the con­
tinuity of a proper integral as a function of its parameter, all Fh (r/),
k = 1, 2, . . ., are continuous functions on the interval c ^ y ^ d.
Hence, by the theorem on Integrating a functional sequence (see
d d
Chapter 8, § 2, Sec. 2), we have Urn f Fk (y) dy = f / (y) dy.
h' » I OOJ J
C C
But the theorem on integration of a proper integral with respect
to the parameter implies that
d <1 lh lh d
j Fh (y) d y = [ dy j / (x, y) dx = \ dx \ f (x, y) dy
u u a c
444 MULTIPLE IN T E G R A L S , FIELD THEORY AND S E R I E S

Consequently, for any choice of the number sequence 2*

. . . » />,, . . . . lim 4 = -j- oo, we have
h->-j-cv

k
•foo ti
m C
'this means that the integral \ dx \ f (ar, y)dy converges and the
V 4-'
a c
equality
+ 00 tl it -}-oo

( d x ^ f ( x , y ) d y = ^ d y \ / ( x, y) dx
a c c a

is full! Iled. The theorem has been proved.

t ' a n d l a r g . If / (x , y) is a continuous function retaining its sign
and defined for d ^ i < 4 ® , c ^ y d (e.g. f (x, y) is non ne­
gative) and if the integral
-foo
J { y ) — j /(•*> y)d*

is a continuous function of y for c y d, relation (10.32) is valid.

Proof. In fuel. Note 2 alter Theorem 10.5 implies that the integral
+«“
J (r/) I j (x, y) dx converges uniformly on llie interval c ^ ? / ^
C
^ d ami. consequently, by Theorem 10.8, equality (10.32) is valid.
For a function / (x, y) retaining its sign \se can prove the following
' t h e o r e m tit.it ( O n R e v e r s i n g the O r d e r o f i n t e g r a t i o n
in tin I m p r o p e r I t e r a t e d I n t e g r a / ) . Let j (x, y) he a con­
tinuous function of constant sign defined for c ^ j / < ---oo, a
x <Z -1 oo, and let the integrals
• do - f oo

J { y ) = [ f (x, y) dx and J * (x) = \ f (x, //) dy

<1 C

regarded as functions of the corresponding parameters be. respectively,

continuous for c ^ // <C j oo and a ^ x <C -|-co. Then, if ui least
one of the iterated integrals
-f-OO - |.o o -j oo -{ oo

\ dy j j (x, y) dx and ^ dx \ f (x. y) dy

c a a c

ccvycrg™. ofhnr in form I nls» conferees and (heir values coincide.

 CH. 10. I N T E G R A L S DEPENDENT ON PA R A M E T E R 445

Proof. We shall present the proof for the case of a nonnegative

function / (x, y) defined for c y < -f°o , a ^ x <C -foo. hut
us suppose that the iterated integral
+» -H
J — ^ dy \ f (x, y) dx (10.33)

is convergent. Then we must prove that

I 4* W pM
lim f dx f / (x, y) dy = J — f dy \ f ( x , y ) d x (10.34)
/—
>-4-00 Ja Jc J
c
J
a

Let there be given £ > 0 . We shall show that, for all suffi­
ciently large I, thedifference between the variable quantity
I -}-0O -L 4'C°
\ dx ^ / (x, y) dy and the constant quantity \ dy 1 / (x, y)dx is
a « a c
less than £ in its absolute value.
We have, by the Corollary of Theorem 10.8, the relation
/ -1-00 4-30 I
\ dx \ f (x, y) dy — \ dy ^ / (x, //) dx
a
The function /(x , y) being nonnegativo, we can write
4-00 4-00 / 4-00
0 < j dy j / (x, y) dx — ^ dx j / (x, y) dy =
C il n
4 -co 4-00

= j dy j /(x , t/)dx =
C I

C|4-00 4 -m 4-00

^ / ( x, y) c/x 1- \ dy \ / (x, #)c/x<

c f Cl I
Ct +00 +» + ”
< j dy \ /(x , y ) d x ± ^ dy ^ / (x, »/) tfx (10.33)
c l ci o
where c <C < -foo. We begin with estimating the second sum­
mand on the right-hand side of inequality (10.33). Since iterated
integral (10.33) is convergent, there is c, c such that
-J-X> - ‘-Vi

\ dy \ ! (x. y) dx < -!j- (10.30)

• •
Ci <1
44G MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

Let us fix Ci c in such a way that inequality (10.30) holds and

then proceed to estimate the first summand on the right-hand side
of inequality (10.35). By the hypothesis, the integral ^ / (x , y) dx
a
is a continuous function for c ^ y <Z -Too and, consequently, it
converges uniformly because the function / (x, y) is nonnegative
(see Note 2 after Theorem 10.5). Therefore there exists L (e) such
-few

that the inequality J / (x, y) dx <i 9 ^ is fulfilled for all

i
I ;> L (e) and for all y £ lc, c j simultaneously. But then we have

J / (*• y ) d x < ~ r <10-37>

C I

for all Z > L ( e). N ow, taking into account (10.35), (10.36) and
(10.37) we conclude that
+-oo +- co / -+no

0 < j dy j /(x , y)dx — j dx j /(x , y ) d y < e

c a a c

for all I L (e) which is what wc set out to prove.

If / (x, y) is a function that may have alternating sign the theo­
rem on reversing the order of integration in an improper iterated
integral can be formulated as follows.
T heo re m, A O » U '» Let the function f (x, y) be continuous for a^.x< C
< -too, c ^ y <; -foo and let the integrals
+ 30 -+00
J /(x , y)dy and ^ f(x,y)dx (10.38)
c a

be% respectively, uniformly convergent on every finite interval a ^

^ x ^ A and on every finite interval c ^ y ^ C. Then, if at least
one of the iterated integrals
-+00 -+co

J dx j |/(x , p) | dy and ^ dy j | /(x , y) | dx (10.39)

a c c a

converges, the iterated integrals

-f-ro +-oo +-oo -+oo

f dx f /(x , y) dy and f dy f / (x, y) dx (10.40)

J k *'
rt c c a

are a/so convergent and their values are equal.

 CH. 10. I N T E G R A L S D E P E N D E N T ON PA R AM E TE R 447

Proof. For definiteness, suppose that the second integral (10.39)

converges. Then, applying the comparison test to the functions
/ (x, y) and | / (x , y) j and to the functions j / (i, ij) dx and
u
j I / (x » y) I dx we conclude that the second integral (10.40)
a
also converges.
Thus, we must only prove that
I -J-OO -1-00 -J-OO
lim j dx j /(x , y ) d y = j dy j /(x , y)dx (10.41)
I->4-0Oa t C O
-foo
The integral j f ( x , y ) d y being uniformly convergent, we have
C
i -j oo +oo I

j dx ^ /(x , y) dy = j dy j / (x, y)dx (10.42)

a c c a
for every finite I a. Letus estimatethe difference between the
I -}-oo
variable quantity ^ dx ^ / (x, y) dy and the constant quantity
a c
-f-cv: -j-oo

j' dy J / (x, y) dx entering into relation (10.41). Taking ad van-

c a
Inge of equality (IU.42), we see that
—j-DO /4*°°
| j dy j / (x, y) d x — j dx j / (x, ij) dy
c a a c
-f-oo + co I
-f-oo
= | j dy j /(x , y ) d x — j dy j / (x, y) dx
c a c a
+■» -!-«■ Cj -f-oo

= | \ dy | / (x, y) dx = | j dy / (x, y)\dx -p j

C I C I

-1-00 -J-OO Cl +ao

*F j dy J / (x, y) dx | •< ^ dy j / (x, y) dx j 4-

CJ
4"°° Cl -J oo —j—oo

+ 1 rf.V j I/(* , ;/)|rfjr<j \ d y | f(x, y) dx ' \ ihj j \ / (,r, //) | * •

Ci n
(10/i3)
448 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

for any Since, by Uie hypothesis, the iterated integral

-1oo -rOO
\ dy \ | / ( x, y)\dx converges, for every e > 0 there is cx> c such
c a
that
-7-00 -}-oo
j <*</ j* | / ( ^ i y ) | r f a ; < — (10.44)
ci a
jSow, li.xing a value or ct ;> c for which inequality (10.44) holds
!O O
and taking into account that the integral f (xt y) dx is uniformly"
<1
convergent, we choose, as in the proof of Theorem 10.0, a quantity
L (e) such that inequality J j / (x, y) dx 2(^-6-) is full!Hod
1
lor all I L (e) and for all y £ k , cj- Then wo have
Cl
t (f ! — C) K (10.45)
] dy j / (x, y) dx ^ ^
2 (V
cC ii ~
- cc
)/
T
-
c J
for all I L (e) and, consequently, hyr virtue of (10.43), (10.44)
and (10.45), wc obtain the inequality
-f-OO -f-OO / t»

M f /(* , i/)d x— Vda: f / (-r, y)dy

!« «' « •"
a

for all such /, which is wlial we set out to prove.

It should be noted that similar theorems are also valid for impro­
per integrals of unbounded functions involving a parameter.
4. Tests for Uniform Convergence of Improper Integrals Dependent
on Parameter.
O
('ci n c h i/’s T e s t . For an integral \ f (x, y) dx to converge uni-
m■
a
formly on an interval [c, d\ it is necessary and sufficient that for every
e 0 there exist L — L (e) such that the inequality
r
I / (x, y) dx < e (10.46)
r
holds for all T, I " L (e) and for all y £ |c. d) simultaneously.
Proof. Necessity. If the integral is uniformly convergent, then,
for (?very c 0 there is Ij — L (e) such that for all VL: , ' L (e),
 CUT. 10. INTEGRALS nKPKN’DEN'T O.V PARAMETER 440

/" > L (c) and ij £ If. d\ the inequalities

— 1C

| j /(•*. li)dx and f{x, y)dx < -

r
are fulfilled. Therefore, for all /" >• L (e) and all y 6 [c, d\
we obtain the inequality

/ (.z, y) dx --1 ^ / (x. y) dx — [ / (x, //) r/.r

- 00
/(*» y) dx \ /(./, y)dx
mJ
■) • •> = F
r
Sufficiency. If inequality (10.4b) holds for all I' Z> L (e), /* >
I«
L (e) arid all y £ l c , dl. the integral ^ / (x , //) dx c o n v e r g e s
a
for every y £ Ic, tfl (see Chapter 0, $ 1, Sec. 3). Therefore, passing
to the limit as Too we obtain, lor all /' > L (e), the ine­
quality
-|*JO
j { j \ y) dx
V
which holds for all y 6 Ic, d I simultaneously. 'The theorem has
been proved.
tl'eie r.strt t,#&* T e s t ( S u f f i c i e n t ('a n d it ion f o r I n i f o r in
('on v e r (fence o f a n I nil t r o v e r I n t e g r a l ) . Ij | / (x, y) \ 5^ g (x)
;^
for a x <1 l-oo and the integral j g (x) dx is convergent, the
a
•(•*>0 ‘f^
integrals ^ / (x, //) dx and j | / (x, //) | dx are uniformly Conner-
*1 O
gent on the interval c ^ // ^ d.
•f •*>
/Von/. Let an arbitrary r ^>0 he given. The integral \ g (x) dx
J
rt
converging, there is L « L (e) such that the condition
r
J g (x) dx e.
 450 MULTIPLE IN T E G R A L S . IIELD THEORY AND SElilES

is satisfied for all V, /" L (s) (I” /'). This implies that, for all
Z" j> L (e) {V 'Jr> Z'), the inequnlilies
t" i~ i"
| j / (x, y ) dx | -< j / {*> V ) d x < j g (x) dx < e (I " > V)
i' r r
arc fulfilled for all y £ Ic, d 1 simultaneously. Consequently, by
-f-3 0 TO

Cauchy’s test, the integrals ^ / (x, y) dx and ^ | / (x, p) | dx

a a
arc uniformly convergent on the interval c ^ y ^ d which is what
we sol out. to prove.
The corresponding tests for uniform convergence of improper
integrals with unbounded integrands and finite limits of integration
are formulated and proved in a similar way. As an instance, let
us formulate
t 'l in c h f/*s 1 'e.st ( S e.cessar y a n i l Sufficient- C o n d itio n ,
f o r C n i f o r m Con re rife nee of a n I m p r o p e r I nteif r a l o f
an U a b o u n d e d C n n e t ion). An Improper integral of the form
b b
J * (y) = \ / (J’»y ) “ hill 1 / ( x , ;/)dx, c < y < d ( 1 0 .4 7 )
ft rt
is uniformly convergent, with respect to the parameter y on the interval
c ^ y J^d if and only if for crcry c ~>() there exists 6 - 6 (e) Z> O
suck that for all ltd and a" belonging to the interval 0 < A <C
<C min {b — a , 6 (t.)) the inergiality
b-\~
j /(.r, y)d x < e (10.48)
b
is fulfilled for all y £ 1c, d\ simultaneously.
Exam pies
i
1. It is clear that the integral J (p) --= j xv~y dx is convergent
o
for p >• 0 and divergent for p 0. Let p n > 0. Then the inequality
xn“ ‘ ^.xr>n~i holds on the interval 0 < j < 1 for all p ^ /?rt. There­
fore, we can put / (x, p) — x ”-1 and g (x) x^0-1 and apply Weier-
strass’ test and thus conclude that, since the integral
t 1
Xu
\ g (x) dx ^ xPt>
“ 1dx =
o i) ."o
 CH. 10. INTERNALS DEPENDENT ON PARAMETER 451

\ I
converges, the integral J {p) =- \ / (x. />) dx = \ xy ~l dx converges
o o
uniformly with respect to the parameter p on the interval 0 <
<C p 0 -tC p <Z. Too for every arbitrarily small ami fixed p 0 0.
Let us test this integral for uniform convergence on the interval
0 <c p < -foo, For this purpose we shall study the behaviour of
x
the integral ^ x{,~1 dx for p —►0 t d- We have x'~l dx = —---- ►
n f>
oo for p —+■0 f 0 ami for any arbitrarily small fixed /. 0.
Consequently, for any e T* t) and any arbitrarily small /. i> 0 ,
the inequality

f x,,_1 dx < 8
o
caimol be valid for all p belonging to the interval 0 < p < - oo.
i
This means that the integral J (/>) = \ x1’"1 dx converges nonnni-
o
formiy on the interval 0 <1 p <C f oo.
:^
2. The integral J (a) — ^ e~axi dx is uiiiformly convergent
o
for <> -< olv ^ a <1 i oo when* ct0 Wis an arbitrarily small fixed
positive number. This is implied by Weierstrass’ test, if wo put
/ (x, a) = e~ux~ and g hr) — c~°nV' since we have | j (x, ct) j —
_ e-ux* ^ c7 (x) — e~ct»x- for () ot0<^a < oo, ()< x < -j oo,
J oo -f"co

ami the integral |* g (x) dx - ^ c-/,!o-vt! dx is converging.

6 o
Let us prove that if we substitute the interval 0 < a < -f-oo for
the interval 0 < a 0 < a < T ° ° and thus consider the values of a
which can be arbitrarily close to zero, i.e. take the maximal range
-f'XO
/•
of a for which the integral J (a) = ^ e~axZ dx is convergent, the
o
uniform convergence of the integral is violated.
To show that the integral under consideration is not uniformly
convergent on the interval « <C ; we take into account
-1’00
that | c~vi dt is a positive constant as an integral of a nonnegalive
o
continuous function which is not identically* equal u> zero. Thio
29*
4 >2 MULTIPLE IXTliOU \1.S. 1'IELO THEOIIY AMI) SKIUES

enables us to estimate the integral j <?'ax2 dx fur an arbitrarily

i __ __
large lived I and 0<1 cc< - 7 - 0 0 . Pulling I — z Y rjL (dt — lAx dx)
we obtain

\ r « ! rf.r - - \ e~i2 i- 0 0 for a 0 -1- 0

* V os «'
1 )a
si m e
liin ^ e,~1' dt - i|' e~li tll -const ;>()
ft »0;-0 1I a_
Hence, for any / Z> 0 and any arbitrarily small lived k ;> 0 Ilie
inequality
+o»
e ~ uxl d x -U f - ‘s dt < t-
v « I -L
Ia
cainiul be valid simultaneously for all « belonging to the interval
0 < cc< ; 0 0 , which means that the integral does not converge
uniformly on the whole inlcrval 0 <C ot <C -poo.
3. Let us prove thal the integral J (cc) = I' e ~ar s*n ^r r/j: is
0
uniformly convergent with respect t.o the parameter nc. for
<C -p 0 0 and any lived p-y==l). To do this, let us estimate the in teg-
— x>

ral g-o* (jx^ 1 ^ 0 putting rfw - e ''•* sin (H dx

ami integrating hy parts we obtain

+*> - c
-a x siu(Vr
► __ t* a ‘r sin (p x | <p) i- 0 0 0 0
sin (Px ; <p) d r
x \ / a'2 p |V- x^l - I - z* l / a - ‘ r p -

wbere ip is an auxiliary angle determined the relations by

a . p -M sill
c (Vr
cos a —— , ■ and sin <i>= —. - . We have <
V** -p- V a- ^p~ V * 4-i b-
^ 1 , . . „ . .. , „
i p | ^°r all and all a > 0 and, consequently, the
inequali ly
I 30
<xx
, sin px - 1 1 r ^
dx
I " ‘ H P I : "I Pi J

= 7 tV < k ftn* ; > I p 11:

 CIf. JO. 1NTKGNA1.S PF .r F N P F . NT < N PAI! AMETl'.U

is fulfilled for all a, 0 ^ a <C -foo, wlticii moans that the integi
in question is uniformly convergent.
II' wo have an improper integral dependent on a paramoler
y 6 [c. d\. whose one or bolli limits of integration an* inliuilo a
whose integrand has one or more singular points, we break up t
interval of integration (provided this is possible) in such a w.
that the integral taken over each part has either one infinite lin
or integration or ono singular point. thou llu» original integral
said to he uniform ly convergent with respect to the parameter y
the interval c ^ y - ^ d if and only if each of tin* constituent integr.
taken over the parts the original interval of integration is divid
into is uniformly convergent for c y ■<(1.
5. Examples of Evaluating Improper Integrals Dependent on I'm
meter by Means of Differentiation and Integration with Rcspc
to Parameter. The integrals below not. only demonstrate some tec
niques, but are also used in various divisions of mat hematics ai
physics.
1. Knowing that

fI l--x
. ' Sm,w
2n (Is. ■= —In -----J1m
---------------------------------
1
0 sm ~ .t
for m <i n where m and n are positive integers (see Example 3 i
§ 3 of Chapter 0), we shall prove, on the basis of the theorem c
continuous dependence of an integral on a parameter, that
i-°°

.1 l t * am pa
0
for 0 < p c.' I. For this purpose we substitute x = t*n into (
ami thus obtain
2n n
dl - Ii«i —1 (I
I H * sin —^ 2-----
n a
^-i
The function / (/, p) =
1-f * is continuous For 0 <C / OC

D < p < I. Breaking up the interval ol integral ion 0 ^ t < 4-c

into two parts (e.g- 1 and 1 ^ t <C oo) and applyin
Weiorstrass’ test to the integrals over [0. 11 and 11, { oo) with th
pi~ i tT2-l
function g (/) equal, respectively, to t and < p x< 0
1
r->3
1 t
r lP~l
1*2 ^ see that the integral j — ( <tf is uniformly con
454 MULTIPLE IN T E G R A L S , FIEL D THEORY AND SERIES

vergeut with respect to p on every interval of the form 0 < p t

-} -D O

r t!‘~l
^ P ^ Pi <C 1- Hence, the ini eg rail ^ ^ _ dt is a continuous
o
function of the parameter p for 0 <C p <Z 1. Every number p belon­
ging to the interval (0, 1) can bo regarded as the limit of a subse­
quence of the number sequence » 0 <C m < n, m, n =
= 1, 2, . . and therefore, performing an appropriate passage
to the lim it in relation (H), we arrive at relation (A), which we set
out to prove. Equality (A) is used in the theory of Euler’s integrals
(sec § \ of the present chapter).
•r*>
2. Let us evaluate Ilie integral | sil^ 'r dx. ft cannot be diffc-
6
rentialed Mirectly with respect to the parameter P under the
integral sign but we Unow (see Sec. 2 of § 1) that the more general
-j-co
integral ^ e~ax dx differing from the above integral in
o
the factor e~axj a ;> 0 (which guaranlees uniform convergence of
the latter integral) can be evaluated by means of differentiation
with respect to the parameter. This results in
Joo
C sin fir , . p
\J e-a x -----5
x
— dx — arc tan -1
a
—
n
As was proved (seo Example 3 in Sec. 4), this integral is uniformly
convergent with respect to m. 0 for any fixed (T f’ou-
soquenlly, it is a continuous function of the parameter a. for 0 ^
< — {—oo. ri’hereforc,

f Inn T d i *-
J
o * a *0 |-0 J
~ for P > 0
= Jim arc tan — 0 for 0 = 0 (10.49)
ct -O : o a— =
----5- for [1 < 0
In particular, we obtain
-TO
C sill jr ,
=-f (10™ )
6
The last integral is used in the theory of Fourier’s series and inte­
grals.
 CII. 10. integrals dependent on parameter 455

3. Let us evaluate the Culer-Poisson integral

+»
(10.51)

(see also the end of Chapter !)). Its convergence was established in
Chapter D, § 1, Sec. 4. Putting x — ut, dx «— u dt, we obtain

^ e~uZts it dt
*
Multiplying both sides of this equality by e~ui we find
i»
J e - U’ - e - f + i-)^udt (10.52)
o
Integrating equality (10.52) with respect to u we obtain
-{-30
J'- = J j e- “2du = j da j c - < « W « dt (10.53)
0 0 0
Here the integrand / (/, it) — u. is a nonnegative and
conlimious function for 0<^f<C -foo, <C -| oo. The inner
integral in (1U.53) is a continuous function of u for 0 u <Z -Too
which is implied by (10.52). If we formally reverse the order of
integration we arrive at the iterated integral
-f-TO oo

j dt j e-vw « 'u d u (1 0 .5 4 )
o b
whoso inner integral

r e-<i+4i)««„du 1 e -(H «2)U2 U— -)-0O 1 1

2" i ; (10.55)
tt=0
u
is a continuous function of I for 0 <^/ <C + oo. Hence, by Theorem
10.1) on reversing the order of integration in an improper iterated
integral with a nonnegativc integrand, integral (10.54) is also con­
vergent and equal to the integral (10.53). Consequently, by (10.55),
we obtain
-j-OO +oo -}-30
i r dt
f dt \ e - v [L-)"*u dit = 2“ J i - T
• **
0 0 o
Thus,
45U MUl.TlI'i.E INTEGRALS, IIKLD THEORY AND SERIES

This integral has various applications lo the theory of Iunit conduc­

tivity. probability theory, statistical physics elc.
4. Evaluate the integral
-J-OO
J (fi) ^ j c~'xxi cos fix tlx where a —const ;> 0 (10.57)
o
which is used in tin* theory of heat comlm l i\ it y ami statistical
physics. JIs convergence follows, for instance, from (lie fact that
the integral ^ e_<xx~ dx is convergent. Performing formal ilil’fe-
o
rentialion with respect to J1! we obtain tin* equality
-feo
— =■ ^ e~uxi ( — x) sin fix dx (10.58)
o
Equality (10.58) can easily he justilied. Indeed, we see that (1) the
functions e~av'2 cos fix and c"a*2 x sin |ix are continuous for —oo ■<
< fi <C -|-oo. 0-^ x <; -j-oo. (2) integral (IU.57) is convergent for
—oo <C (i < -j-oo and integral (10.58) converges uniformly with
respect to fi for —oo •<! fi <C T°° hy Wo iers trass’ test (in which we
can put g (x) — c-etxS). Therefore, hy the theorem on differentiation
of an improper integral with respect to the parameter, equality
(10.58) is in fact valid. Now, integrating hy parts in (10.58) (with
respect to x) we obtain
dJ — p —a x - _____
___ sin ffir *“ -1 00 ft ”^*x> JL
^ c~<xx~cos fix dx - 2a
rffl 2a x = o
o
Separating variables in the differential equal ion thus obtained we
find
dJ _ p dfi
(10.50)
J ~~ 2a
Integrating (10.50) we obtain
_Ji
./ (fi) —Ce <«, C const (10.00)
Lot us determine the constant C. According lo (10.5b), we have

7 (0 )-
(10.61)
Consoquen Ily. hy (10.00) and (10.01). we can write
/ (0 )—c 4 / / " Ta
 CII. iO. I N T E G R A L S DEI'KM>l-:.NT ON P A R AM E TE R 4a

Substituting this result into (10.OU) we finally derive

/ — fl-

J (p) e~ ax~cos p^: dx ——- y —e ',a (10.02

5. Let us evaluate the Fresnel integrals ^ sin(.r-)d.r am

«
0
rv
\ cos(x*)dx which are used in optics. Flitting x~ —l we obtain
*7

j sin (**>dr - t - r 1 ti
sp* r tU' 1
Ik
ixt)du - 4 - $
r.
- 1j t
As an instance, let us evaluate the first integral. Noting that
- oo
I
—— - “ - ( e~tu~dit (see 10.01) we derive
V'' v» i

0 V T
*
1
~ ~ y t
v (jI ”• 0j v
S dt J
(l" 03)
11’ it wore easy to justify the possibility of reversing the order of
integration in integral (10.03) we could complete our calculations
in a simple way but it turns out that this involves some complicated
techniques. Therefore, as in Kxamplc 1, we shall introduce the
factor e~ht, k = const 0, and consider the integral
oo +*> -loo

f f dt { e-tfcru^sin tdu =
J Vi I n J J
-T»
= l 4 ) du 1 .1 i i - ( M - y 00-84)
1 0 0 r I)
In tiles latter case it is easy to show that the order of integration can
bo reversed on the basis of Theorem 10.9. Since the integral
-1-00
-ht sin--dt
/ converges uniformly for u ^ /«;<<-; oo, and its
Vi
integrand is continuous for 0 ^ k <C t o o , 0 ^ / < ^ oo, this
integral is a continuous function of k on the interval 0 <- k < r^o.
Therefore, passing to the limit for k ^ 0 -- 0 we obtain
458 MULTIPLE INTEGRALS, FIELD T H E O R Y AND S E R I E S

Taking the decomposition of the fraction __ ^ into partial frac-

lions and performing integration we finally obtain

j sill ( I 1) dx^ — y -2- (10.65)

The relation
+ 30 y ----
f cos (x2) dx = y ( 10 . 66 )

is proved in a similar fashion.

0. Let us consider the Frullani* integral
+°°
f / (M —/(ax) £ x wjlerc 0 < i b< C .— oo (10.67)
J X
0
We sliall limit ourselves to the following two cases.
(1) If /'(x) is continuous and integrable on the interval 0 < lx *<
< t ° ° and f{x) tends to a finite lim it / (-Too) as x — ►
- T°°» i.e.

j f(* )d * = /( + o o )-/(0 )
ill
then the integral

Jj /' (ui) dx (10.08)

o
is uniformly convergent with respect to the parameter u on the
interval 0 < a < u < 6 . Indeed, since / (x) tends to a finite limit
/ (4-oo) as x —►-Too, the necessary and sufficient condition of
Cauchy’s criterion for existence of a limit of a function is fulfilled
for / (x), that is for every e > 0 there is N (e) such that
I / (x ) — / (x") J < e for all x't x" >> N (e). But then we have
A" A "u

| j f ( u x ) d x | - | T j f ' W dt u
-I' A’u

\
for all .4', — Ar(e) and for all u belonging to the interval
■a-tCu^. b. Therefore integral (10.67) can be evaluated by inte-

* rr :lhi:ii. C.iubann M795-1834V an Italian matlicmalician.

 CII. 10. I N T E G R A L S D E P E N D E N T ON P A R A M E T E R 459

grating (1U.08) with respect to the parameter u from a to b:

-]-w -f-O O b b -j-a >

( dx =*j dx j /' (ux) du = j du j /' (mi) dx =

0 0 tt a 0
h

= J / ( ^°°>~/(0> du = |/(---oc)-/(0)]ln 4 - <10.09)

Cl
(2) If there is no finite limit of f(x) for x —>- + oo hut the integral
-fo e

j m a r , A > 0 . is convergent, an,I the derivative /' (0) exists.

A
we have

j / (M ~ / <**) d x = __ j (Q) in ± . (10.70)

o
In fact, we can write

| f(,)- /(0) d t - j I M - M d z it-a x )

and

j js n = m _ * -1 u - b x )
o o
Consequently,
s (is
j =j m a (0) =
0 at as

as

whence we obtain (10.70) by passing to the limit for s —►-r«>.

Frullani’s equalities (10.69) and (10.70) can bo applied to evaluat­
ing various concrete integrals. For instance, applying (10.69), for
0 < a < b, to the function / (x) — e~x we find
*|*oo 4*»
\ e~bx — e - a x a , f arc tan bx —arc tan ax ,
\ ------ ox —ln-^- and \ ------------------- a i=
o b
400 MULTIPLE IN T E G R A L S , FIET.D T H E O R Y AND S E R I E S

Similarly, the application of (10.70) to tlio function / (x) = .sin x

results in
-i-.v
siu bx — sin ax , f cos bx — «*<>•?</.;• . , a
x
dx —-0 and \ --------- ---------- dx - in
(I o

§ 3. El'LEK’S INTEGRA OS
Killer’s: integral of the first kind
i
13 ( p , 11) - ^ x**- 1 (1 — a 1)'*” 1 d x
5
called the beta function of p and q and Euler’s integral of the second’
kind

f (p) — f x»-'c-*d£
0
called tlie gamma functioii of p play an important role in various
divisions of mathematics and mathematical physics. As will be
shown, the beta function is expressed in terms of the gamma func­
tion (see relation (10.81)) and therefore we begin with the properties
of the gamma function.
1. Properties of Gamma Function.
-{-OO
(1) The integral T {p) — ^ xp~1e~x dr converges for 0 <C p <T.
6
< i oo and diverges for p <1 0 (see the end of § 2, Chapter 9). This
integral is improper for p <C I not only because its interval of inte­
gration is infinite but also because the integrand approaches infinity
for /) < 1 as x —►0 -| 0.
-Joo
Let us prove that the integral F (p) — \ xi'~ye~<dx is uniformly
o
convergent with respect to the parameter p on every finite interval
0 <Z po p ■/\)<C oo. As in the case of testing this integral for
ordinary convergence, we break up the interval of integration
{(), ; oo) into two intervals, namely ( ) < r < 1 and 1 ' x <L :-oo,
and test, for uniform convergence, the corresponding integrals
! -},ocs 1
| xr le r dx and ^ dx. Thu integral x•J ' c/jf* converges
<1 \ 0
uniformly for U < />o^ x <C ~r oo by Weierstrass’ test siiin1t'"'/1"1
 C.Il. 10. INT E GR AL S DEPENDENT ON PA RAM ET ER 4E1

I
for 0 < , r < 1 ami />!>/>o and the integral \ 1dx con-

verges lor estimating the integral ^ x1' le~x dx for —0

o
aiul X-= const > 0 we see that.
/.e
^ d:*' ^ rr~te~1dx —e~l ^ z 1' 1dx = ~ -I- oo
y*
5 6 o

and, consequently, the integral ^ x7' V v(/a* (loos not converge uni-
o
formly on the interval ( ) < / ) < | oo.
| OO

Weierstrass’ test indicates that the integral \ ar ~1e~x dx con-

1
verges uniformly for —o o < p < f J0< -f-o o where / >0 is an arbit­
rary fixed number because we have
x v~le~x .rr° - 'e~x for 1 C x <C ■-}- oo, - oo < /) < /^

and the integral I z p°~1e~x dx is convergent. Hut. the integral

**
I
“I OO

|j a*'"1c~x dx docs not converge uniforinly on the interval — oo <Cp<i

*i
i «>
<C ?-oo. Jo show this let us estimate tlie integral j xv~le~xdx for
i
an arbitrary fixed /~> 1 as p —*- j oo. For any integer ;V > 0 we
have p — 1 ;> A’, from some value of p on, when p ->■ i-co, and
therefore, for such />, we can write
-t-*> +*>
j* xp~le~x dx > J xNe~xd x = —f ' V p I A ’ J xs - ]e~' dx -
'i i t
[lK A - \ ’l s - ' j - A r (/ V — 1 ) / A' - U .. . A’ 11 c~* f oo
for /V j- oo
Consequent ly,
-f oo
Iim \ xv xc~x d x — -f- oo
J— . j OO J

for any fixed I <J.

402 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

I
Thus, the integral^ e~xxp~l dx converges uniformly on every
o«
interval 0 < p9< p < -poo where p 0 is an arbitrary lived positive
Moo
number and the integral [ xp~le~xdx converges uniformly on
l
every interval —oo < < -j-oo where P 0 is an arbitrary
finite number. Hence, both integrals simultaneously converge
uniformly on every interval of the form 0 < Po^. p < <C - f 00
and, consequently, the integral P (/>) — I xn~lo Xdx also conver-
0
ges uniformly with respect to p on every such interval.
(2) Since the integrand / (i, p) xp~ie~x is continuous for
0 -i oo, 0 ; p <C -j-oo, and the integral ^ xv~xc~x d.rr
understood as the limit
/ +»
lim \ x i * - ' e ~ x d x - = ^ x f>" 1e~ ‘ <(x
—*—
Una v
>’-*d+o ®
is uniformly convergent with respect to p on every Unite interval
*
0 < Po- ^ -P ^ P q <C H-oo, the integral
Y (p) —
■ I xp~'c,~xdx is-
o
a continuous function on every such inlerxal, i.e. a continuous
function for all p satisfying the condition U < p <
-[-OO
(M) Differentiating Y (p) ==■ J xp~le~x dx formally with respect
o
to p under the integral sign we obtain
+oo
T '(/> )- \ xp~' (In x)e~x dx (10.71)
o
Hut equality (10.71) can he justified because we can easily prove
that Ilie integral on the right-hand side of (10.71) is uniformly
convergent on every finite interval 0 <C Pn -<Cp^ fJ0 < —oo, and
the partial derivative jp (a:, p) = xv ~l (In x) e~x (where / (x. p) s=
== xv ~l(T~':) is continuous for 0 < i < oo. C) < /) <C oo. The
fact- that, the integral on the right-hand side of (10.71) converges
uniformly is proved by applying V\ cierslrnss’ test to the integrals
i -oo

^ a**1 (lux) e”* dx and x'” 1(In jt) e~x dx

o 1
 Oil. 10. IN T E G R A L S D E P E N D E N T ON PARAMETER

for which wc can put, respectively, g {x) — x^o-1 | In x | and g {x) =

= | lti x | c~x.
We likewise prove that the derivatives F'fc> (p ) of all orders
k = 1, 2, 3, . . . exist and are expressed hy the formulas
-Loo
:= ( X T- ' (Inx)*e-x dx, k ~ 1 , 2 , . . . (10.72)
0
(4) Performing integration by parts we liml
-f-oo Too
pV(p) = p f dx * xve-* \f°% f xpc~x dx
6 5
that is
r (p -r i) = />i' ( p ) (10.73)
Applying recurrence formula (10.73) repeatedly we can reduce
evaluation of F (a -j- «•) where 0 <C a <-CI and n is an arbitrary
natural number to evaluation of F (a):
r (a //) -= (a -f n — 1) (a — 2) . . . (a -r 1) gT (a) (10.74)
If we pul a = 1 and take into account that
-J-oo
r (1) — ^ c~x dx - 1 (10.75)
b
formula (10.74) results in
r (n -i- i) — n (n 1) . . . 2 1 ^ nl tiu./b)
(5) Let us evaluate
-L/v* j

r ^ —- ) = ^ j: - e~xdx
Putting .r — /2 we obtain

r (4 -)= 2 ( e - ‘'-<n a L i- y .n (10.77)

((.*) Let us consider the graph of the fund ion !'(/?) (see Fig. 10.1).
For p — 0 “j—0 and p - r ° ° we have F(/>) — -i-oo. The values
of F(/>) for natural p are given hy formula (10.70). We have* V (1) —
= F (2) = 1, and therefore, hy Hollo’s theorem, the derivative
1 '(/>) turns into zero at a point belonging to the interval 1 <C p < 2.
Let p Q he such a point. Since T" (/>) = ^ xp~l (In x)2 e~x 0
o
for all />, 0 < p << + ° ° , the derivative I’' (p) is n nKmotone increas­
ing function for 0 < p <C -\ oo. Consequently, the derivative I’Vp't
'il'..'* MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

lias no roots on Ilie interval 0 < / ? < -f-oo other than p0. Besides.
L'(p) < 0 for p <C p0 ami F'(p) !> 0 for p > p0 because I '(p)
is a monotone increasing function. Hence, the function V (/>) 1ms
r'r’iI

only one extremal value on the interval U < p < oo, namely the
mitiimuin attained at the point p - p0.
Lot us consider the corresponding numerical data:
p0 l.'dilG, min T (p) > F (po) ■0.«S8f>G
Since the gainma function increases tor p ^ 2 we have F (p) ;>
^ P [n -- I) —«! for p w -I- 1 where /i 1. lienee. I’ (p)—*-
—►-f-oo sis p -► -i-oo. Furthermore, we can write
n /> + n
r(p) /»
for p > 0 and hence F (p) — — —— >■-j- oo for p — 0 ; 0 because
F (p ' 1)--*• 1'(1) — 1 for /;-> 0-{-0.
2. Properties of Beta Function.
i
(1) Tin? integral B (p, //)-■- f a:7*-1 (1 — x)q~l dx converges for
o
p ;> U and q > 0.
(2) The change of variable x — 1 — t shows that
It <p, q) = B fo, p) ( 1 0 .7 8 )
( -onse<|uent lv. the hela function B (p. r/) is a symmetric, function
of p and r/.
 Cll. 10. INTEGRALS D E P E N D E N T ON PARAMETER 41j 5

(3) For q >> 1 wc have

1
13 (p, ? ) = ( ( l - x ) « - * d ( — ) =
*0
xP (1 — + 2 n l f i>’(1 —x f - ^ d x ^
0 P J '

I 1
Jlu L xP- 1 (1 — x y ' - d x — f x”-' (1 —x ) i- ' rfx =
6 ft

= qZ X 15 ( P ’ — q)

Thus, we have

H</>, 7 ) - 7 ^ —r ,-» (P . 9 - ! ) for 9 > 1 (10.79)

The beta function being symmetric, relation (tU.79) implies, for

p 1, the equality

B(/>, (l) " P ^ ~~1’ ^ for /? > 1 (10.79')

(4)
'
Performing the change of variable x = -^--— 1 -j- r
in the integral
1
n (/), q) --- ^ a*,,"1(1—x)'l~l dx wc arrive at another analytical rop-
b
refutation of the beta function:
zP*i
V,(p, q) -= ^ (1 dz (10.<SO)

(y) Let us deduce a relation connecting the beta and the gamma
functions. Namely, we shall prove that
„ x _ n f» r < 7)
],{p - q)- T i F T W (10.81)
for p > t> and q > 0. The change of variable x — tz \t. > n, dx
+oo
c
— / tiz) iii Ihe integral 1 (/') - J xv“le~x dx results in
o
r (/>)
ip
0
\ lit* MII.TIPLE INTEGRALS, FIELD THEORY AND SER IES

Substituting 1 -p / for t ami p r q for p we obtain from (10.82)

the relation
+“
r ( p - g) _ f -I'tQ- 1e-( | j- t):(lz
(1 (10.83)
0
Multiplying both sides of the latter equality by lp~l and integrating
with respect to t from 0 to r oo we derive
j-<X> -f- 30

r </>-'/) jn ~(T+V)7;~ tU = f dt f
0 0
liy (10.80), the last relation can be rewritten in the form
-(-cio 4-oc
V(p I </)B(p, <1 ) = f dt [ ' ^ dz (10.84)
v
0 0
Now let us prove that it is allowable to reverse the order of inte­
gration in integral (1U.84) for p !> 1 and q^> 1. Indeed, we can
easily show that the conditions of Theorem 10.9 on reversing the
order of integration in an improper iterated integral are fulfilled
here:
(a) the function
/ (z, I) —zp • ^’ '/p-V <* '• ^ 0
is continuous for O ^ z <C -f-oo, 0 ^ /< C -j-co;
(b) if p > 1 and q 1 integral (10.84) is convergent;
(c.) the integral
-} -o O

\ dt == l'(p -i- v)
0
is a continuous function of t for 0 ^ /< C | oo and the integral
-|-ou
\ tv izP-M7-ie-(i.fo* dt = T (p ) zq-'e-'
o
is a continuous function of z for 0 ^ z < - {
-4-ou -\ oo
Hence, the iterated integral C dz ^ zr,‘Q 1 ff,~ l"-( l“- t)r dt
Jo *n
is convergent and equal to integral (It).84) (Theorem 10.(J).
 CH. 10. IN T E G R A L S D E P E N D E N T ON PARAMETER 407

Consequently,
-J*oo -4i*oc
<i) ii (p. <i) = \ dz m f
)
dt-=
0 0
—oo -j-oo -f no
= f z!)+q-'e dz f f
J0 *(•J
0
Too
= r{ p ) j z ^ c ' </z = r { p ) r (f/)

We have thus proved that equality (10.81) is valid for p > 1

and q > 1. To extend relation (10.81) to all /> > 0 and q > 0 we
simply write this relation for p > 1 and q ;> 1 and then apply
recurrence formulas (10.79) ami (10.79') to its left-hand side and
recurrence formula (10-73) to its right-hand side.
(6) Let. us derive the formula
Jl (10.85)
H(/>, 1 - 7 ) = Sill for 0 < /; < 1

Substituting q = 1 —* p into formula (10.(80) we obtain

f oo
13 (p, !-< /)= j ~ - d z . 0 < p < l (10.80)

Hut in Sec. 5 {if § 2 (Example I) we showed that integral (10.86)

is equal to S1
—n-—
pH for 0 < p < 1, and this implies relation (10.85).
.Making use of formulas (10.81) and (10.85) we obtain the formula

r (/') r (1 - p) = -— p for 0 < p < 1 (10.87)

which plays an important role in the theory of the gamma function.

There are many integrals that we can evaluate by reducing them
to Euler’s integrals.
Examples

30*
4W* M ULTIPLE IN T EG R A L S, FIELD THEORY AND S E R I E S

ji
r - r
2. Let us evaluate the integral / ~ \ sin2 xcos2 x dx. rutting

sin2x = 2 we obtain sin x — z2, cosx = (1 — z)2 and dz = 2 sinxcosxdx.

Consequently, taking into account the foregoing example and
formula (10.79') we find
3 it y 2
y= 1U
3. Let us consider the integral
ji
z
— J sin,J_I x cos^”1xdx, pZ> 0, 7 > 0
o
Putting sin2x = z we get

C 1 c 2-i 2-t
I sin1' 1 x cos'' ‘ x d x ^ -g - J z 2 (1— z)- dz =

t
2

In particular, for 7 = 1 we obtain the formula

y ji
sin*5"1 x dx =
I 2
)
§ 4. MULTIPLE INTEGRALS DEPENDENT
ON PARAMETER
For definiteness, we shall consider triple integrals dependent on
a parameter although the results obtained below hold for multiple
integrals of an arbitrary order except certain cases which will be
stipulated in what follows.
Let a fund ion / (x, y, z, ot, p, v) be defined for (x, y, z) £ Q
and (a, p, y) £ Q* where Q and Q* are, respectively, domains
of the x, y, z-spaco and the a, p, y-space. Suppose thaL the integral
•J(a. P> y) = f^j/(^ V» z, a, p, y)d x d y d z (10.SS)
 CH. tO. I N T E G R A L S D E P E N D E N T ON P A R A M E T E R 46

exists as a proper or improper integral for all (a, p, y) £ Q*. Thei

it is i\ function of the parameters oc, p, y in the domain Q*.
As in the case of a onefold integral, we can easily prove the follow
ing assertions:
(1) If f (x, y 7 z. a, p, y) is continuous as a function defined if
the domain Q x Q* where Q. and Q* are bounded closed domains
the integral / ( a , p, y) is a continuous function of the parameter.
a, p, y in the domain Q.^.*
(2) //, in addition, the derivative f'a (x , y 7 z, a, p, y) exists anc
is continuous in Q X Q*, the integral J (a, p, y) can be differen­
tiated according to the rule
= /(* , P* Y)d x d y d z =
JQ
= y » 2. a, p, y ) d x d y d z (10.89)

The derivatives of J with respect to p and y are expressed simi­

larly provided that the derivatives /p and f'y exist and are conti­
nuous.
(3) If the conditions of Assertion (1) are fulfilled. it is permissible
to integrate the integral / (a, P, y) with respect to the parameters
a, p and y under the integral sign.
Assertions (1)-(3) are proved by analogy with the case of a single
integral. They can also he easily extended to integrals of the form
J (a, p, V ) = j j j / ( x> z»a * P. V)£(*> */» z)d x d y d z (10.90)
QJ
where / (x, y 7 z, a, p, y) satisfies the above requirements and
j*J J | g (x, y 7 z) | dx dy d z < K = const <14- oo
a
(here the integral g (*. y i z) | dx dy dz may be proper
or improper).
Example. The potential function of the field of gravitation gene­
rated by a material body £1 with volume mass density p (M) =
« p (x, y 7 z) is expressed, at every point (J (x0, i/0, z0) iying outside

_ *JFor given domains £2 and Q* ((j, y, z) (a, B. y) £Q*), the domain

2 X Q* is the set of all points (x, y, z, ce, fi, y) (belonging to the correspond­
r/>

ing 6-dimensional Euclidean space) obtained when the poiuts (x, y, z) ami
{a, p, y) independently run through their domain's O and O*
470 MULTIRLE INTEGRALS, FIELD THEO RY AND S E ItIK S

tlie body, by the integral

U (Q) = U (x(J, iy0, z0) = jj j \ — ^ dxdydz (10.91)
Si ‘
where ri>v = [ (x — xtt)'2 T (i/ — y 0)~ -f- (z — zl()‘“ is the distance
between the points P (x, i/, z) and Q (x0, y 0, z0). If the point (V
is at a positive distance from the body Q the function / (x, //, z , x0,
//o» ^o) = ----- ii* continuous and possesses the continuous partial
rl*(j
derivatives —f-. - ami The density (> (z, z) is supposed
to be an absolutely inlegrahte function in Q. Differentiating (10.91)
with respect to x0, y0 and z0 according to Leibniz’ rule (see relation
(10.89)) we find the projections on the coordinate axes of the force
of attraction with which the body Q acts upon unit mass located at
the point Q (*„, y 0, z0):

r * «?) =^ =ini^ (-r_-r,) d ld y d z

i',j«?) (10.92)

F , «?)
u
If the point Q (x0. t/0, z0) lies inside the body Q we have (I
when I* coincides with Q. Hence, in this case Q is a singular point
of the integrand in the integrals (10.91) and (10.92) and thus these
integrals are improper even if p {P) ■— p (x, iy, z) is a bounded
inlegrable function in Q. A feature of these improper integrals
dependent on the parameters x 0, //0r z0 (on the point <9 (x0. y v. zG))
is that the coordinates of the singular point of the integrand depend
on these parameters, namely, are equal to them. Here we shall limit
ourselves to studying improper integrals dependent on parameters
which are of the form
/<<?)_ <?) / (/») dx dy dz (10.93)
Ci
where
P (*, i/; z) 6 »nd <9 (x0, y 0i z0) £ il
The function F ( / \ (J) is supposed to lie continuous for P <9
and unbounded for P -*■ O while / (P) is a hounded inlegrahle
function on Q (integrals (10.91) and (10.92) are special cases ot
integral (10.93)).
 CII. 10. INT E GR AL S DEPENDENT ON PA R AM E TE R 471

D e f i n i t i o n . We say that integral (10.93) is u n i f o r m ? if ( O l i ­

v e r yen t a t a p o i n t Q (.r0, Uq , z 0 ) £ 0 if for every e > 0 there is
6 = 6 (e) !> 0 such that, the inequality
Q ) / (P ) d x d y dz (10.94)

is fulfilled for any domain Q&(£; °f diameter less than 6(e), belonging
to Q and containing the point (J, and for any point Q* whose distance
from Q is less than 6 (e).
S u f f i c i e n t ( ' o n d i t i o n f o r V n i f o r m C o n r e r y e n v c . If there
exist a neighbourhood of the point Q (x0, y 0. z0) £ O and constants
C ;> 0 and X < 3 such that for all P and ( / belonging to this neigh­
bourhood the inequality
| F{P, ^ f) |< l—j-— (a const <[ 3, 0 = const < - • °°) (10.95)
ri*Q'
holds, integral (10.93) is uniformly convergent at the point (J(x0, f/0, z0).
Proof. By the hypothesis, we have | / (P) | << K = const < -foo
everywhere in Q. Consequently, if a hull S6(e, (Q) of radius 6(e)
with centre at Q lies within the above neighbourhood of the point
Q. we can write the inequality
| <?') / (P) dz dy dz
u»ie.
c d r r t t j f t 7.
(10.90)
K dx dy d
r1'Q’ t>
* t ’Q'

for every domain Q$(e> of diameter less than 6(e) containing the
point Q and for any point Q' £ «S$tC) "’here S 26ie.) ((/) is a hall
of radius 26(e) with centre at Q* (see Fig. 10.2). Passing to spherical
coordinates with origin at the point (j* we obtain
2n ft 2f t ( e )
f r C <£r dy dz
) •’j 0 0 0
•iftie)
An
= 4n r2-^rfr = [26 (e)| 3 - ) . (10.90')
o
It follows from (ID.90) and (10.90') that
An
| (?) i (P) dxdy dz :t - /. [26(f )]‘- , CA (10.97)
ft<C>
Since 3 — ?<. ;> 0. the right-hand side of (10.97) is less Ilian e if
6 (e) is sufficiently small, which is what we «el »ml i n prove*.
472 M ULTIPLE INTEGRALS, FIELD TH E O RY AND S E R I E S

Note. In the case of an improper A'-fold multiple integral of type

(10.93) (iV > 1) the exponent X must satisfy the inequality a, < A*.
If mass density p (P) in integrals (10.91) and (10.92) is a bounded
inlegrable function in Q the above test implies that these integrals
are uniformly convergent at every point Q (x0, i/o» zo) £

Kig. 10.2
The implications of uniform convergence are Ihe same as in the
case of onefold integrals. As an instance, let us consider the question
of continuity and differentiability of an integral as a function of its
parameters.
T h e o r e m JO.tth If integral (10.93) is uniformly convergent
at a point Q £ 9 and the functions F (/*, (J) and f (P) satisfy the above
conditions, integral (10.93) is a. continuous function at the point
Proof. Let us show that for every e z> 0 there is fi = 6 (e) > 0
such that the inequality | tqq* | <C 6 («) implies the inequality

I J (0) — J (CO I < c. For this purpose we take a ball S^e) (O)
of a sufficiently small radius 6 (c) with centre at Q lying inside ii
(see Fig. 10.3) and represent each of the integrals J (CO and J (O')
as a sum of two terms, namely an integral over the domain
5 6(e) (0) and another integral J z over Lhe domain Q — *$6 (0 (CO
Then wc have
I j (0) - J (</) I < I •/.> « ? W 2 (O') 1 + I / i (COI -i \Jt (Q')\
(10.98)
 CI1. 10. INTEGRALS DEPENDENT ON PARAMETER 473

If 6 (e) ;> 0 is sufficiently small the second and the third terms
on the right-hand side of (10.98) are less than because the integral
/ , is uniformly convergent at the pointy. Let us choose an arbitrary
positive S'(e) <C ^ 6 (e). Then, if the distance between Q and Q*
satisfies the inequality
IW | < 6' (e) (10.90)
the integrals in the first summand on the right-hand side of (10.98)
are proper. Consequently, if 6'(e) (0 < 6'(e) < 6(e)) is suffi­
ciently small the theorem on the continuous dependence of a proper
integral on parameters implies that the first summand on the right-
hand side of (10.98) is also less than provided that condi­
tion (10.99) is fulfilled. These results indicate that (10.OS) and
(10.90) imply the inequality
I/(CO - / «?') t < £ (10.100)
which is what we set out to prove.
Here we do not discuss the general problem of differentiability
of improper integrals of form (10.03) with respect to parameters
and refer the reader to special courses (e.g. see 115], Lecture 7, § 2).
We shall onlv illustrate the corresponding techniques in connection
with a special case when F (P, Q) — —— and / (P) = p (P) where
p (P) is a bounded (| p (P) J < C — const < -Too) and differen­
tiable function in Q, that is when the integral in question is of
form (10.01). This case is important for the theory of potential
functions (see 1171).
If the point (J (x0» 2o) lies outside Q (and P (x, i/, z) runs
through Q) integral (10.91) is proper and, as lias been shown, equali­
ties (10.92) arc valid. Let us prove that relations (10.92) remain
valid when the point Q (x0, y0, z0) belongs to Q. We shall limit
ourselves to the first equality (10.92). To prove this equality we musL
show that the difference
^ (^o-f Ax0> y0t s0) —U (r0, i/o» z0)
Ax0
—H i ~ J°^ dx dy rfz, Axo^O (10.101;

lends to zero for every fixed point Q {x0, y Q, z0) belonging to Q as

Az0 —>- 0. Let there he given an arbitrary e >> O. Take a ball
(CO °f a sufficiently small radius <S (?) lying entirely inside O
and denote, respectively, by (arft, t/«, z0) and b \ (x*. f/«. z*) llm
474 MULTIPLE IN T E G R A L S , FI EL D THEORY AND SERIES

integrals of type (10.91) over the domains = S 6(e) ( 0 and Q2 =

= Q — Qt = Q — S 6(e) ( 0 . We obviously have V — Ul -f- U2
and therefore difference (10.101) can be rewritten in the form

iU w
Let us estimate the first term on the right-hand side of (10.102).
We have
AU x t 1
Ajo Ax0 dxdy dz —
rPQ'

Ajy Jf Jf Jf p(P) - r PQr

Q r- Q' dxdydz (10.103)
PQ*

where the point Qr belongs to £2, = S ^ e) ( 0 and has the coordinates

(x0 -f- Ax0, y o, z0). The sides of the triangle QPQ' are of lengths
rPQ* rpQ- and |Ax0| and therefore
I r pQ' — r PQ I ^ I Ax0 | (10.104)
Taking into account inequality (10.104) and the apparent relation
i \
rPQrK'’
PQr K ‘r *^ '' r/'Q
PQ ^ r% Q'' f
' PQ
we derive the following estimate for (10.103):
A6r,
Ax( ^ *i \ \ \ \ -2
'Q t'Q'
dx dy dz
since |p ( 0 | .<1 C. The integral converges um-
* ht rPQ
foriniy on Q, = Sa{e) ( 0 (see the sufficient condition for uniform
convergence) and, consequently, the right-hand side of inequali­
ty (10.105) is less than y if 6 = 6 (e) > 0 is sufficiently small.
The second term on the right-hand side of (10.102) is a uniformly
convergent improper integral, which is implied by the inequality
— 1 and Ihe sufficient condition for uniform convergence.
rpQ
6 *
Hence, this term is less than -g- in its modulus for all sufficiently
small 6 = 6 (c) 0. Finally, let us estimate the third term on the
right-hand side of (10.102). Since the integral U2 (2 *0 . .Voi 2«) —
~ dx dy dz is of form (10.90) (because the point
 CH. 10. I N T E G R A L S D E P E N D E N T ON P A R A M E T E R 475

Q (x0t y 0. z0) lies outside Q2) it can be differentialed with respect

to the parameter xp under the sign of integral ion and. consequently,
fur all sufficiently small |Ax0| < 6 = 6 (e) we have
P (p ) (x ~ *o) dxdy dz (10.100)
rl>Q
Thus, for all sufficiently small 6 = 6 (e) and |Ax0|, difference
(10.101) is less lhan z . which is what we set out to prove.
Note. The results obtained here for volume integrals can easily
be extended lo line integrals and surface integrals. Let the reader
modify appropriately the formulations and proofs of the above
theorems.
 | /j Fourier Series
and Fourier Integral

In natural and engineering sciences we often deal with periodic

processes such as oscillatory or rotary motion of various parts
of machines and apparatuses, periodic motion of heavenly bodies
and elementary particles, acoustic and electromagnetic vibrations
etc.
These processes are described mathematically by means of periodic
functions. A function f (/) of one independent variable t is said to be
p e r i o d ic if there is a number T ^=- 0 (called its p e r i o d ) such that
f (t ~r T) ~ f (/) for all values of t, —oc < ; / < - ] -oc (11.1)
The well known trigonometric functions sin t and cos t with
period T ~ 2ji are the simplest periodic functions.
In physics the simplest periodic processes are described by means
of Iho function
£ (t) -- -d sin (c*>^ F <|), —oo <; / < -b°° (11.2)
called a harmonic (or a harmonic vibration). We have
c(I ^ ==£(/) for — o o < < < 4 - o o (11.3)
*)-j
and therefore the quantity I ~ l-s period of this harmonic.
The constants A and to are called, respectively, the amplitude and
Ilie frequency of the harmonic, the expression tot -F t(; is called its
phase and the constant, <[ is the initial phase.
Tim fundamental question this chapter is devoted to concerns
the problem of representing an arbitrary periodic function in the
form of a sum of harmonics.

§ I. PROPERTIES OF PERIODIC FUNCTIONS.

STATEMENT OK THE KEY PROHEEM
1. Periods of a Periodic Function. Fel f (£) he a periodic function
with period T #= 0, that is
/ (/ i- 7’) - / (/) for all /, - o o < / < + oo (11.4)
 CH. 11. FOURIER SERIES AND FOURIER INTEGRA!. 41 i

Then every integer multiple k'l\ k — ± 1 , ± 2 , . . of the period T

also serves as a period of this function.
Indeed, if T is a period we have, for any integer k 1, the relation
/ {t 4- AT) = / l M - (A* - 1) T -} T\ =
= / U + (A — I) 74 — / (t) (M.5)
for all t, —oo < / < 4-oot which means that kT is a period of / (/).
Furthermore, we have
f (t — T) = f [(t — T) 4- 74 = f (l) for all /,
—OO < t < 4-00 (11.0)
and consequently the number —T is a period of / (/)• But then it
follows from (11.5) that the number k (—T) = —kT is also a period
of / (/) for any integer A: > 1. Thus, the above assertion has been
proved.
We can now easily verify that if two numbers Tx and T2 are
periods of a function / (1) the numbers rl \ dz T 2 arc also periods
of this function.
A constant quantity can obviously be regarded as a periodic
function with an arbitrary period, i.e. every number is its period.
/ / / (0 is a continuous periodic function which is not identically
equal to a constant it possesses the f e a s t p o s i t ir e p e r i o d called
its p r i m i t ire p e r i o d * When speaking abuut the period of a function
we usually mean its primitive period.
2. Periodic Extension of a Nonperiodic Function, raking an arbi­
trary nonperiodic function / (x)** defined on an interval a^xssQ
a 4 - T we can construct a periodic function F{x) with period T
which coincides with / (a:) on this interval. Geometrically, to obtain
the graph of F (x) we must translate the graph of the function / (x)
(constructed 011 the interval a ^ x ^ a + 7*) along the x-a.xis to
the left and right by shifting this graph, in succession, by
distances T, 27’, 37', . . nT, . . . as is shown in Fig. 11.1.
This process (and the resulting function F(x) itself) is called the

* If we suppose that a continuous periodic function / (t) different from

a constant does not have the least positive period, it can easily he shown that
there must exist a sequence of its positive periods Tt, . . Tn , . . .con­
vergent to zero. Thu set of alt the integer multiples of these periods constitute
ail evciywhere dense point set lying on the t- axis, — 00 < t c j-0 0 , and,
consequently, the values of / (i ) assumed at the points oT this everywhere dense
set arc equal to its value at the origin of coordinalos. Therefore / (j) jg identi­
cally equal to the constant / (0) on this set and hence, by the continuity of the
function. »1 must he identically equal U» this constant on the whole /-axis,
—<x> c t c i-o°» which contradicts the hypothesis. Hence, the smallest posi-
live period must exist.
** 111 v hat follows we shall denote *nd«p'»*’darj» vriri..!.!c ..
478 MULTIIM-E IN T E G R A L S , FIELD T H E O R Y AND S E R I E S

periodic extension, with period 7’, ot* the function / (x) from the
interval a 4 ^ x ^ a -{- T to ttie x-axis. In the general case this
y

--1______ i -
Fig. ti.t u^ZT ~a^f a+T a+ZT a+JT jc

process does not uniquely specify ihe values of F{x) at the points
x a ± kTy k = 1, 2, 3, . . . .
3. Integral of a Periodic Function. If f(x) is a periodic integrahle
«+r r
function with period T we have for any a.
— 00 00 . In fact, we can write
<i + r r «+r t «t
| / (x) dx .= j / (x) dx -J- ^ / (x) dx — j / (x) dx ~~ f / (x)dx = \ I (>z)dx
\« n r <1 0 5

because the periodicity of 1lie function implies that

a+T a

| j (x) dx = ^ / (x — /') dx — ^ / (x') dx' where x = x — T

Thus, the integral of a periodic function with period T taken over

an arbitrary interval of length T has one and the same value.
4. Arithmetical Operations 011 Periodic Functions. It is apparent
that a sum, a difference, a product and a ratio of periodic functions
with the same period T are again periodic functions with period T.
Let / (x) and g (x) he two periodic functions. If their periods Tf
if p
and T„ are commensurable, that is — where p ami and q are
integers, the number 7’* = p T g ~ qTf is a period of both functions
/ (x) and g(x). Consequently, the surn, the difference, the product
and the ratio of these functions are also periodic functions with
period T*.
Hut if the periods Tf ami TR of the functions / (x) and g (x) are
incommensurable their sum is no longer a periodic function but
a so-called almost periodic function.
Let us give the definition of an almost periodic function. *1 func­
tion f (x) continuous on the entire real axis ---0 0 < x <C 1 0 0 is said
to be f i t p e r i o d i c if for every e > 0 there exists a number
L =« L (t?) >- 0 such that for every interval of length L, a x ^
I / . .- ' o ^ ^ .I 0 0 thrtrn if rj f fottyf n)7c flf/m tu T T — T (p ^
 CH. II. FO URI ER SERIES AM) FOURIER INTEGRAL 4?y

such that
I / (x r X (0) - / (x) I < 6
for all x , —oo <; x <C -j-oo.
Periodic funclions are obviously a special case of almost periodic
functions. It can be proved that a sum, a difference, a product and
a ratio of almost periodic functions (under the condition that in the
latter the divisor is different from zero) are almost periodic functions.
Hence, the set of all almost periodic functions (ill contrast to the
set of all periodic functions) is closed with respect to the fundamental
operations of arithmetic.
5. Superposition of Harmonics with Multiple Frequencies. Let
us consider a sequence of harmonics
x -p (pfj J , k -= 1, 2, . . . , — o o < r < ~ o o , T> 0
(11.7)
T
Obviously, the number Th = - y is the period of the Ath harmonic.*
Consequently, the number T — kTh is a common period for all the
harmonics entering into sequence (11.7) ^but it is the least period
only for the first harmonic A x sin x -r (f i) ) • The frequency
of the Ath harmonic is equal to Xh = —j — . A* — 1 , 2 , . . . Thus,
the frequencies of the harmonics belonging to sequence (11.7) are
integer mnli iples of one and the same m im ber-y- . The least positive
T
(primitive) period of the A*tli harmonic is equal to T ti — — o
, k ■=
= 1, 2, . . ., and, consequently, we have Xk = - * — and Xh
*U
Such harmonics will be referred to as the ones with multiftit*, frequen­
cies.
A .s<um (orT in physical terminology, a superposition) of a finite
number of harmonics of the form
TV
4 / 2icfr . \ (11.8)
fy (>■£) — Afi | >j .4hsn. ( T X - H-h)
h 1

* In fsirl w«* have ?iu . T\ . 1

X~ ~ ) r * 1" i ( T
. / 2a*
- 5‘"
480 MULTI I»LE I N T E G R A L S , FIELD THEORY AND S E R I E S

is a periodic function with period T since the number T is a period

of all the harmonics* and the least period of the first harmonic.
Similarly, a superposition of an infinite number of such harmonics,
or, more precisely, the sum of a convergent series of the form

/(x ) = 4 0 - j - 2 -4* sin (11.9)

k= 1
is also a periodic function with period T.
Equalities (11.8) and (11.9) can be transformed in the following
way. Taking into account that
. . / 2ak \ A 2rtA *>nk
.ljc sin ( —j —x - q.-hj = Ah sm <(&cos 4k cos (pn sin —f —x
we put
"o 40, ah = Ah sin ip*, bh — Ah cos q>h, 21 = T
anil rewrite (11.8) and (11.9) in the form
N
vi / Aar . . . knx \
/<v (•*) v ■2 j ^a/t cos ^------ 1 6* s in —j - J (11.10)
/i=i
and
f kzix , , A-ax ^
/(x ) = "o
•> S
I ah cos —— --- bh sin T y ( 11. 11 )

All the lunctioiis on the right- and left-hand sides of (11.10) and
(11.11) are periodic, with period 21.
It should be noted that the functions / v (r) and / (x) are of a more
. . . nature in comparison
complicated . • . the
with , i
harmonics. fcji.r—
cos —j
A.tx _
and sin—'j— , k — 1, 2, . . tliey are formed of (e.g. see Kig. 11.7b).
The series on the right-hand side of (11.11) is called a t v i y o n o -
o*ctrie. s e r i e s . A relation of form (11.11) (provided it is valid)
is called the e x p a n s i o n o f the f i u u t i o n f (x) i n t o a t r i
yon o m et r i c s e r i e s .
5. Statement of the Key Problem. The main aim of the present
chapter is to elucidate the following questions:
(1) What are the periodic functions with period 21 which can
he expanded in trigonometric series (11.11), i.e. can he represented
as a sum of this kind?

* Wc remind the reader that the constant -In he regarded as a periodic

function with an arbitrary period, in particular, with period T .
 OH. li. FOURIER SERIFS AND FO URI ER INTEGRAL /|SI

(2) How cun wo determine the coefficients a0, ah and bt.y k =

= 1, 2. . . of e.\(illusion (11.11) if it is valid?
(3) What is 1lie conneclion between tlie character of convergence
of series (11.11) and the properties oT the function j (x)?
7, Orthogonality of Trigonometric System, Fourier Coefficients and
Fourier Series. Relation (11.11) is the expansion of the function
/ (x) into a scries with respect to the system of functions
I nr . nx knx . knx
—, c o s — , sill— , cos — , s i n —j ~ . . . (11.12)

which will he referred to as the trigonometric system.

Trigonometric system (11.11) possesses the properly that the
integral of the product of any two different functions of this system
taken over the interval (—/, l\ is equal to zero. This properly is
referred to as the orthogonality of the system (the general definition
of orthogonal functions will be given in Sec. 1 of § 3). Besides,
the integral of the square of every function belonging to (11.11)
is unequal to zero.
Indeed, we have

r 1 Aax 1 I Aax
) t cos I
dx —T An sin x = —/
=0
-/
/
f 1 sili knx dx — 1 I x~l (1113.)
.U I *+0> kn *COS

= - 4 2 - ^Hrjt i ( - i ) h- ( - i ) ki = o
and
I An
knx . nnx
J C0S — S i n dj =
-i

= 4" ^ [ cos —— x -- cos —— x Jdj: = 0 for k ^ n ( i i . i;m

-i
Similarly,
Ii \
/far . nnx
si n —-—si 11 —— d x = 0 for k ^ '=n

I
(M .I3y)
sin cos "7!— dx —0 for anv k and n
- 1
n-tu»2i
482 MU1.TI1M.1-: 1NTKGHAIJ?. FIELD T HE OR Y AND SERIES

Finally,
i I-r COS 2
knx
{ cos2 dx — f ------- ----- -— d x = I
-i
-/ -/
knx
i
^ .Avrx , fLI — cos 2 I.
^ Sill-
si i —j — d x ■- \ -------------------- dx = Z (11.1-4)
-i -!
i

-1
Now lot us discuss the problem of determining llm coefficients o„,
a fl* b;., A
-' 1, 2, . . ., of expansion (11.11).
As was proved, a functional series convergent in tlie mean or
uniformly convergent can be integrated term-by-term (see §§ 2 and t>
of simpler 8). Lot us suppose that series (11.11) converges uniformly
or in the mean to the function f(x) on the interval I—Z, /|. It HMiiains
convergent (in the same sense) if we multiply both sides of rela­
tion (11.11) by any continuous function. This properly and the
orthogonality of system (11.12) enable us to perform the operations
given below and to determine the sought.-for coefficients a0, <7/.. (>n,
k = 1, 2, . . . .
Integrating equality (11.11) termwise we find
( l -[-30 I I
\j(x)dx - — • ^d x ^ I a'< \ c o s - — d x - r b i t J sin dxj aj
-t —i ft - 0 -i
whence

a t> -Jr j j (x) d x (M .1 U

«.njf
To determine the coefficient. a n in cos—— we multiply equality
f? n .r
( l i . l l ) by cos " I and integrate it wiLh respect to x from Z
to /. This results (by (11. ld,)-( 11. l o 4)) in
l l • 00 I
i* >inx <j,i f uni , v'« f k s ix n ru ,
I f (x) cos —j —d x i — \ cos ~— d x \ an \ cos —— cos - — d r
-t ~ -i /t=i -i
i
r \ - ^1/ ft*1.1 » I ^ *i tuKJ* j|
* iiJXX
+ 2 j bn \ s i n — j— c o s — — u j ^ a n \ c o s - — j —d x = a nt
n = i 2 / - i
 CH. 11. FOURIER SERIES AND FOURIER INTEGRA!. 483

anil thus
i
an - | f / (x) cos -2221 dx (11.14,)
/J JT
Similarly, to find the coefficient bn in sin —^— we multiply equa­
lity (1 1.1 1) by sin —j— and integrate it with respect to x from —/ to /,
which yields (by (11.13,)-(11.134)) the relation
i
\ i (X) s i n ~ d x r-. b„l
-I
and th u s
/
bu =r y J /(x )sin ( ll.M 2)

Dej in it ion /. The numbers a0, a„ and bn, n = 1, 2, . . . . deter­

mined by formulas (11.110), (11.I'll) and (11.11 called the
f ' o u r i e r eoeffieient.s o f the f u net ion / (x) frith ve sp er t
to t r i y o n o m e t vie system- (11.12).
D e f i n i t i o n i. The trigonometric series

—-r- 2 ( a,t c w s ~r bh sin (I Mb)

M=I
whose coefficients are specified by the function f (x) according to formu­
las (11. Ho), (11.111) and (11.H*) is called the T o u v i e r s e r i e s
ttf the f n net ion /(x ).
It should be noted that the Fourier coefficients a0, bh, k -
— 1, 2, . . . . determined bv Formulas (11.14 0) * (11.14() and (1l.14o)
can bo defined williout imposing the requirement that scries (11.11)
is convergent. Indeed, for integrals (11.140)-(11.142) to exist it is
sufficient that the function / (x) he integrable on the interval I— I, /!.
Therefore, to every function / (x) integrable on the interval (— /, /)
there corresponds its Fourier series. This correspondence can he
written in the form
-}-oo

/ (x) ~ y -2 ( ait C0S ~T~ hk sin ~7~ ) (11-1 fi)

h= I
where the series on the right-hand side is the trigonometric series
whose coefficients are determined by formulas (11.140), (11.1',)
and (ll.ld o ). Rut in the general case when the function f (x) satisfies
nnly the condition that it is integrable on the interval I - /, /1 the
31 *
4S4 MULTIPLE IN T E G R A L S , FIELD THEORY AND S E R I E S

sign of correspondence ~ in relation (11.16) cannot be replaced by

the sign of equality. In § 2 we shall introduce some sufficient condi­
tions for expansion (11.11) to be valid.
8. Expanding] Even and Odd Functions in Fourier Series. A func­
tion / (x) defined on an interval I —Z, l\ is called even if
/ (—x) = / (x) for all x £ I—Z, Z1 (11.17)
If a function / (x) defined on an interval I—/, /I satisfies the con-
dit ion
/ (—x) = —/ (x) for all x £ ( - - / , Z1 (11.18)
we say that it is an odd function.
These definitions imply that the graph of an oven function is
symmetric with respect to the axis of ordinates and the graph of an
odd function is symmetric with respect to the origin.
If / (x) is an arbitrary function defined on an interval 1—Z, Z] we
can form the functions
/<-*> and / 2(x) = / ( * ) — / ( — x ) (11.19)
where the former is even and the latter is odd. We obviously have
/ (x) = /, (x) 4- / 2 (x) for all x 6 I—Z, *1 (11.20)
and, consequently, every function / (x) defined on an interval of the
form I I, /| can be represented as the sum of the corresponding
even and odd functions.
If f(.r) i s inlegrable for x £ f —/, 11 wo obtain
i o i i
f / (r) tlx -- \ / (X) dx ' - [ i (.r)dr = 0 |/(x ) \-/( — x)\dx (11.21)
-I ~l 0 0
because substituting —x for x we get
o o i
^ /(x ) c?x = — j / ( - x ) d x -~ J / ( — x)dx
- t i o
ll follows from relation (11.21) that
i
2 ( / (x) dx if the function / (.r) is even
j / (x) dx - ( 11. 22 )
6
^0 if the function /(x) is odd
, . 1 rr.r 2ax Ar.xr
Tlie functions —, cos——, cos—-—, . . . , cos—j—, . . . are
. c ,. . :ix . 2ns . /,-n.r
even and the functions s m —p , s in —— , . . . , s in —— , ...
are odd.
 CII. 11. FOURIER SERIES AND FOURIER INTEGRAL 485

Let / (x) be integrable on Ihe interval I—/, 11. Then if this func­
tion is even its Fourier series is of the form
4*00
kr\x
! + 2k=i
ttkCOS (11.23)

and if the function / (x) is odd its Fourier series lias tlie form
oo
knx
bh sin (11.24)
fc=l
knx
an
Indeed, if the function / (z) is even the expression /(x) cos —j
~T
k’l
is also an even function whereas the function / (x) sin — is odd.
and therefore iu this case
i t \
Uo j { = f $/(£)<*!
-(
I
bn j / ( y s i n M - J s - 0 , * = 1, 2, . . . (11.25)

/
«k = T j / ( |) ( .° s * |id g , k —i, 2.
-I o
knx
Accordingly, if the function / (x) is odd Ilie function /(x) cos —f
is also odd and tin* function / (x) sin —^— is even, and hence
i b
a .--!" ak = - j HI) cos dS O
-I i - I-
I (11.211)
6-, = - j / ( 6 ) » i n - - ^ d | = y j / ( |) s i n d \, k = 1 , 2 , . . .

9. Expanding Functions in Fourier Scries on the Interval | —5t, jt|.

If it is required to expand a function / (x) defined on the interval
| —jt. ji 1 into Fourier’s series we put / = ji in formulas (11.13) and
(11.17) and tlius obtain the following expressions for the Fourier
coefficients and Fourier scries:
JT
« o j / (£) °n = — ] / (I) cos n i d i
—n
-1 -p o o

>,t ~ J / (s) sin / (*) ~ i' 2 (a k cos ^'x 1“ s l’ 11 k & )

H- I
• 'i S f i MU LTI PLE 1N T BG HA LS , FIELD T H E O R Y AND SERIES

The general case when a function / (x) is originally defined on

an interval I—/, /1 is reduced to the above by means of the change
^j
of the independent variable x ^ facl* if the function f (x)
is defined on the interval [—1. l\ the new function q.- (s') = /
is tie lined for —j i^ x ' -<;x. But in view of the applications of
Fourier series to the problems of mathematical physics we shall
deal with the general case of an arbitrary interval [—L 11 to unify
the notation and facilitate the reader to learn the corresponding
(echini] ues.

§ 2. FUNDAMENTAL THEOREM ON CONVERGENCE

OF FOURIER SERIES

In $ 2 our main aim is to prove that Fourier series (11.10) of

a periodic piecewise smooth function / (x) with period 21 converges
to / (x) at each point of continuity of / (x).
1. Class of Piecewise Smooth Functions. A function f (x) is said
to be i n e r a r i . s e c o n t i n u o u s on a n i n t e r r a f [a, 6| if it is
continuous everywhere on this interval except possibly at a finite number
of points of ordinary discontinuity (i.e. discontinuity of the first kind
or jump discontinuity).
Such a function has finite left-hand and right-hand limits
/ (x — 0) - lim /(x i z), /(x -r -0 )-■-=l i m / ( x - r -) (11.27)
; —-0 z—►O
t^-0 ?>0

at each interior point x of the interval [a, 61 and Imilo limits

/ (« r U) and j (6 — 0) at the end points a and b of the interval
61. At each point of continuity of / (x) we have / (x — 0) =
= / (x -f- 0) -= / (x), and thus / (x — 0) and / (x -f- 0) do not
coincide* only at a finite number of points at which / (x) has the
jumps / (x 0) — / (x — 0) ^ 0.
.1 pieceu-isc continuous function f(x) defined on an interval la, 61,
a < 6. is said to be j> ie r e w is e s m o o t h if the derivative /'(x)
exists and is continuous everywhere on (a, 61 except possibly at a finite
number of points at which finite right-hand and left-hand Limits
i'{x i 0) lim /'(x -h c ), f (x — 0) ~ lim f* (x — z) (11.28)
.0 r-t-O
;>0 s>0
e;r.,st. ft is als<$ supposed that finite limits f'(a j ()) and j'(b - 0)
exist at the end points of the interval la, 61.
 Oil 11. FOUHIEK SERIES AND FOURIER INTEGRA!. 4S7

A piecewise smooth function / (or) possesses finite lei'I-ham! ami

right-hand derivatives
f)(x) = lim
— 2

<11-2U)
/ « * ) - iim n*+!)-n*+o)
0

at each point x of the interval [a, 6|. In fact, applying the formula
of finite increments* and using relation (11.28) we obtain
lim /( J ~ -)-~ /(;c- n) = lim / ' {x ± 0z) -= / ' ( x ± 0) (11.3tJ)
;-*0 z-*0
2 >0 2 >0

and lienee ihc derivatives f\. (x) and f\i (x) exist and besides 1lie
rein lions
Vl ( * ) = r {X - 0) and f ‘H (x) - f (x + 0)
hold.
The graph of a piecewise smooth function / (.r) possesses a uni­
quely specified tangent line at each point except possibly at a finite

number of points but at these there must exist uniquely specified

left-hand and right-hand tangent lines (see Fig. 11.2).
if / (.r) is a piecewise smooth function on [a, 61 we can break
up llu* interval (a, 6| into a finite number of subintervals
U 0>"J* (^1>^ 21» • • •J f^t»^i+ll» • • •» ^.v+tl
where
« < ffj < . . . < « / < +J < . . . <C a x <C «.v+ i = I*

* Move we apply tlie formula of finite increments written in the form

f (r z) — f (.r -f- 0 ) — z f (x - - 0 c) where 0 < 0 -r: 1 which a n lie derived
in tin’ following manner: wc lake f t . 0 < : 6 < ' z. and write down the ordinary
(<»rmnla of finite increments
/ ( I T * ) - / (r ft) - (z — ft) f {T -4- I)
xvi.f. .. ., n x - s ' .d tluj. lu till- iihiiI 0 — 0.
4<SU MULTIPLE INTEGRALS, FI EL D T H E O R Y AND S E R I E S

such that- the functions /(x) ami f{x) are continuous at the interior
points of each interval Iflj, i = 0, . . A\ and lend lo finite
limits
fiai ~r 0), / f(«| + 0) and f(<*n t — 0), /'(«,• + 1 — 0)
when x approaches « t* from the right and a i+i from Ihe left. It follows
that the functions f(z) and /'(x) arc bounded on each interval
frtj, and, consequently, they are bounded on the whole'interval
[a. b \ *
2. Formulation of Fundamental Theorem on Convergence of Fourier
Series.
T h e o r e m / / . / . // f{x) is a piecewise smooth junction on an
interval —l < ' x ^ /, its Fourier series converges at each point x of
the interval and the sum

(a /tcos bfi sin (11.HI)

ft—i
of this series satisfies the following relations:
(1) S (x) = f{x) if —I <C. x <i I and f(x) is continuous at the
point x,
(2) S(x) / (r 4- 0) -- / (,r —0) ^ — l < x < i l and x is a point of
discontinuity of f (x),
(3 ) S( - / ) = - - £ ( / ) - : i{ - / ( / “ 0) .

Note. If /(x) is continuous at a point x. ~ l < Z x < Z /. we have

f(x — 0) — f(x | If) “ / (x) and, consequently,
/(J .i 0) t - / ( . r - 0 ) _ | 2 / (x) ^ 1 ^

Therefore equalities (1) and (2) can be replaced by the relation

6’ (*) = U * + °H -/( * - » ) ( ,, 32 )

which holds for every interior point x of the interval I—/, l\.
3. Key Lemma. To prove the theorem we need the following lemma

* In fact, if we redefine / (x) and /'(x) at the end points of an interval

la*, "f+th * -= 0. 1................A\ by putting / (a,) — / {tij | 0), /(rfr+i ) -
~ / («a +t -- 0). /'(«,-) - /'(«,- 4- (l) and f'(an i ) i ' b i m — 0). the functions
I f.r) and C {.rl become continuous on the closed interval frtj. and. conse­
quently, bounded on it. But, in this construction, the values of f (x) and f'(r)
may only change at Ihe end points of the interval [n; . /i.^] and lienee the ori­
ginal functions / (x) and /'(x) are also hounded on
 CH tl. FOURIER SERIES AND FOURIER INTEGRAL /.Sf

K e y X cm nuf. If f (x) is a piecewise smooth fmiction on m,

interval a ^ x ^ b wc have
b
lini f / (jt) siRaz<fe = 0 (I I . 3.T|
U^oo •a
Proof. Let us divide (a, b] into ports
[iZqj flj|, 1 f l 2 K • ■ *» il»• • •> f^.v ’ «\- : 1^
where a ^ a0 < a, < . . . < a(. < aM., < . . . < fl v < a.v-ti ~ fc
in such <1 wav that the functions f(x) and f'(x) are continuous in
the interior of each suhinterval Ia,, (7,^,1, i =r 0, 1, . . . . A", and
tend to finite limits
/ (a* -I- 0), /'(«, -|- 0) and / - 0). /' <«,*, - 0)
when x tends to at+\ from the left and to at from the right, respecti­
vely. Since
b X u i+ I
f / (a:) sin ax cl.r = 2 f f ix) n dx (11 •3/i)
* »
a i -0 «^

it is sufficient to prove that

°i+l
lim f / (z) sin axdx —0 (11.35)

for 0 We can regard / (x) and f'(x) as being continuous

on the closed interval jl (see footnote on page 488), and there­
fore it is allowable to integrate by parts, which gives
°i+l
/ (x) cos ax x=a i+I -0
^ / (x) sin ax d x = a j* f' (x) cos a x d x
.v = a .+ 0
•»/
(11.30)
The functions f (x) and f'(z) being bounded on l«, b], there exist
constants \ [ and M f such that \j (a:)K[.'l/ and |
everywhere on la, b]. Hence, relation (11.30) implies the inequality
i+i
^ 2,1/ U' lai^- ui )
/ (jt) sin ax dx rL (11.37>

Passing to the limit in equality (11.37) for a —* 0 0 we obtain rela­

tion (11.35). The key lemma has thus been proved.
4'JI I MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

Avti' 7. IL can be shown that the above lemma applies to a consi­

derably wider class of functions. For instances if / (a:) is an absolutely
b
inlcgrable function on la, fr], that is the integral J | / (x) | dx <1 -r-oo
a
(which may be improper) exists, the lemma remains valid.
JVote 2. We can similarly take cos ax under the sign of integration
instead of sin ax.
4. Proof of Convergence Theorem. Let / (x) be a piecewise con­
tinuous and piecewise smooth function on an interval —Z ^ x ^ Z .
We shall extend the function / (x) as a periodic function with
period 21 from the interval I—Z, Z1 to the entire x-axis and prove that
| 5 a(J) _ l i 3r- 0H / ,(*+*» j _^0 for h —►-i- oo (11.38)

for every x where

S„(*) = ~ l 2 («*cos^2£. + f t * * iii~ .) (11.30)

/l“ 1
is the nth partial sum of the Fourier series of the function / (x)
corresponding to the interval —/^ x < lZ .* Substituting the expres­
sions of the Fourier coefficients
i
« „ - T* •/(/(S M ?
i
ah - - j / (I) cos M / .= I, 2, (11.40)
-t
I
= j j / ( I ) sin / f c - 1 , 2,
-I1*

* After the periodic extension lias boon performed, /(*) becomes a perio-
1 J IX TC£
die function with period 2/, and the functions — , cos —— , sin —j - , . . .
Jcj\X
. . , cos —-— , sin . . . are also periodic with period 27. Therefore
ihe integrals appearing when \vc compute the Fourier coefficients «<)-—
i t i
1 f , . %. I I* . . . nxx , . I f . . . . nitx
- — \f{x)dx. On —— J /(x )c o s —j— dx, ] /(.r) sm —
-/ -I -l
1, 2, . . . , can be taken not only over Ibo original interval ( — 7, 7| but
alsu over any other interval of length 21, which does not affect the values of
those coefficients.
 CH. 11. FOURIER SERIES AND FOURIER INTEGRAL 491

into (11.39) we obtain

t n I
5 "<x) = i r j / ® de - T 2 j m f c o s - ^ c o s - ^ 1
—I k=i - t
n
4 „ M s i n i 2 £ . ] d| = | C/ ( | ) [ | + 2 c o s =
—/ #t=1
/-X n
= 1 j / ( l -L2) [ i . + 2 c o s - ^ J ^ (11.41)
-l-x h—1

where z = | — z. Now let us derive an expression for the sum

/ \ 1 * yi knz
kn
>n (’) = — + 2j cosCO® — (11.42)
fc=*i
Miitiplying both sides of (11.42) by 2 sin 4^- we obtain
n
n / \ • JI4* • 71Z 't*! <-| • Tfj* ^
2on (c) sin — = s i n ^ + 2 j 2 s i n - ^ c o s - ^ =

= sin ^ - 4 - ^ [ si" (* ■‘"t ) "7~—s'n (* J") 'T‘] = s i n ( ” + 4 ' ) “T

ft---I
whence
. / 1 \ nz
/v 1 1 vi kjiz *,n \ n ~r 2 ) I
On (z) -b 2 j COS—p = 0 . nz
( 1 1 .4 3 )
ft=t 2 s,n ^ r
Substituting (11.43) into (11.41) we arrive at the following formula
for the /ah partial sum of the Fourier series:
l-x ( - M nz
1 r Sinl n+ T ) T (11.44)
Sn(x) = T \ f { * + z) ---- ------- H
nz - ± ~ d z
2 si n
— l —x ~U
The function / (a:) (after it has been periodically extended from
the inlerval ( —7, /I to the whole x-axis) is periodic with period 2/,
sin ( n - y ) -7 1
and the expression ------------------------- is, by (11.43), aperiodic
2 an
fuuction of z with period 2/. Therefore 1)10 product of / (.< j)
/.M2 MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

sin ^ n -f- ~ j —j~

b y ------------------- is also a periodic function of z with period 21.
20 sin
. —n*
and the integral of this product taken over any interval of the
s-axis of length 21 has one and the same value. Consequently, the
limits of integration —/ — x and I — x in integral (11.44) (which
is taken over an interval of length 20 can be replaced by the limits
— I and /. This results in

, r sin ( -j-
S„ (x) — T \ / (r - s) ------ ------- dz (11.45)
-t 2 sin

Integrating (11.43) wil.li respect to z from —I to / we deduce

1 \ nt
i r sin("-T)
7! (11.415)

since J cos dz —0 for k — 1, 2, 3, . . . . Hut the expression

si„ |__L\ zh.

as — is an even function of 2 (see § 1, Sec. 8), and
_ sin
~2l
consequently (11.46) implies that

, 5 sin ( n - r - M - y - ' sin (n + y ) V ,

T *•
.
2 sin —
~J a"<* 1,1 % 2 sin
ns
l ! ^“4-
(11.47)

Multiplying the hrst equality (11.47) by / ( x — 0) and the second

by / (x 0) and adding together the results we obtain

/ ( . - 0 ) l - / ( x - i 0) = i j / ( l _ 0) si" ^ dz _
“ I J ,. . A*
—I 2 Sin ------
21
I ' sir: (» + 4-)
~ | / i j.- 0 ) dz (11.48)
-

.1 2
2 sin HT
 ON 11 . FOURIER SERIES AM) FOUR! EH INTEGRAL 493

Now subtracting (11.48) from (11.45) wo got

S „ ( . r ) - “ * - 0-> 4 - ; ( , + 0 ) «

0
= - M 1 / ( * - - ) —/ ( * —o)i
-I
I
- y j |/U - ; ) - / ( t -0)1 (ll.'tU)
0
Let us prove lliat both integrals on the right-haml .side of equali­
ty (II . i t ) ) tend to zero when n —*■ -• o o . For instance, lot ns lake
tlie second integral. We can represent the integral in question in the
form
nz
1 { T - z ) - j { x 1-0)
Jn dz *-

f(x-±-z)-f(x- 1-0)
nz '■-Jn (11.50)
2 sin
IT
where 0 < ft <C L Suppose we are given an arbitrary e > 0 . Lot
us show that for sufficiently small ft 0 the first integral in (11.50)
i.> less than in its modulus forall n ~ 1, 2, . . . . In fact,
/ (•*-t- z) _J (.r , 0)— ^ j, ^ 0) as z — 0 -}~ 0*, and therefore, for
a sufficiently small 6 ;v> 0 and all z belonging to the interval
0 < z < ft, we have the inequality
/ ( x - r z) — f ( x -1- 0)
< | / ' ( z - ' r 0)| i-1
nz
IT
Furthermore, ns 1 for z~>0 and, consequently, for a suffi­
sin
~2T
ciently small ft ;> 0 and all z from the interval f / < * < 6 we ha\e
the inequality
nz
•Jr
l < —:—n z < 2
sin
2T

* See relation (1 1 .3 0 1.
 MULTIPLE 1XT MORALS, F I E L D THEORY AMJ SERIES

Finally, for all real z and all « —1, 2, . . . we can write

1 \ 7TZ
sin (m h T ) ^ i < 1

Tims, the relation

6 ns
(Xa-s)_ f (x + 0) HT
sin ( « -r -■) <h <
sin
UT
26
< ^ l | / ' ( x - 0 ) | ?-IJ
is fulfilled for all n = 1, 2, . . . provided that 6 >> 0 is sufficiently
small. Taking a sufficiently small ft .> 0 such that 11 }' —0)|-j-
11 < -rr we see that
IJn
’ I < — for all n 1,2,... (11.51)
We now fix a value of Z> 0 thus determined and proceed to esti­
mate the second integral in (11.50). We have
I f /(x-uz)-/(x -L (■)
s in (» L )f< /z
'■ -t ) 2 sin
nz

where the function ^ji is piecewise continuous and

2 sin ~2L
piecewise smooth on the interval iS ^ z <5 / (lor 6 U) because
hern the numerator is piecewise smooth and the denominator is
a continuously differentiable function which does not vanish on that
interval. Then, by the key lemma, we have J'n —*- U for n —►+ o o
and, consequently, for all sufficiently large values of n we obtain
tin! inequality
\Jn\<4r (1 1.52)
Taking into account (11.50) we conclude, on the basis of (11.51)
and (11.52), that
| / „ K | / ; | J' l j ; i < 7 - - T —e (11.53)
Tor all sufficient Iy large n which means that lim = U.
fl
We similarly prove that the first integral on the right-hand side
of (ll.V.I) also tends to zero for n — +oc , and thus
l i m 5 . (x) - / | r ~ f l ) ~:' , ('r — 0 ). (11.54)
>1 — T£>
 CTI. 11. FO URIER SERIES AND FOURIER INTEGHAI VI')

Wo remind l he render that the function / (x) is regarded as being

periodically extended, with period 21, from the interval I—/, l\
to llii* whole x-axis. Therefore
/ (/ _ ( ) ) = / (_ / ; 0) and / ( - / — 0) « / (J - 0) (1 1 ..M)

Now substituting x = —/ and .r — I into (1 l.o'i) and taking advan­

tage of relation (ll/io ) we obtain
litn .S'„ ( — /) — lim £ „ (/) , ' ‘ • - * > > - /« -» ) ( 11 .r,li)

and thus the proof of the theorem has been completed.

5. Examples. 1. Eel us take the function f (x) = x regarded as
being defined on the interval I—/. I\ ami expand it into a Kmirier
series. This function is odd, and hence
f
0 ,-ft. «„ = 0, f r . - ~ T f /(•*■) s i l l » - l , 2, . . .
*0
Integrating by pails we obtain
t
nns f
bn =
0
x sin — dx —
fj
/
t I
(\ w
nn COS
uzvr \
— ■
— )
I
I ) |x--n n:I J
\
U
cos
I
dx \
I r
l’l
nn

Consequently, according to the convergence theorem, we have

c/ , 21 vT (— l),e-1 • knx _
(*) 2 j -— ;i:----- sin I ~
u l
/ (x —0) h /(*-'- 0)
•)
for — /< x< I
(11.07)
/( —<—0)-:-/(/ ft) = 0 for
2 2
r= ± /
The graphs of j ( x ) - ^ x and 5(x) are shown in Fig. 11.3.
The function .ST(x) is periodic with period 21, and .S(x) — x
only for —I <C x <C I At the points x = (2k -■ 1)Z. k 0. ~ 1 ,
d:2. • • the function .S'(x) has discontinuities, its left-hand and
right-hand limits being, respectively, equal to / [(2/c + 1) I — 0| =
= —1 and / l(2/.‘ ; 1) / -1- 0| — 1. k — 0, zb-A. ± 2 , . . . . at I he>e
puinls. At. every point r -- (2k [ I) /, k — 0, j l. dr2 ..........
Mu* function S(:r) assumes the value which is the half-sum of the
left-haml and right hand limits, i.e. equal to zero. In Fig. I 1. \
4% MULTIPLE INTEGRALS. F I E L 1) T H E O R Y AND S E R I E S

we see the graph of the partial sum .V3 (x)

'll (-1 ) A+l sin A.*tx
- S k T
constructed on the interval —I <. x<<l.

2. Let us take 1lie function /(x) = x2 on the interval I—/, /].

Tills function being even, we have
i i
aQ= — f x“ dxy an —-J- j* x2 cos dxy bn ~ 0, n — 1, 2, . . .
o o
After the coefficients aQand an, n 1, 2, . . ., liave been computed,
we find, by the convergence theorem, tlint

£ ( * ) = -y •^TT“ cosJT ^ for (11-38)

1
 nil. It. FOURIER SERIES AND FOURIER INTEGRAL

The graphs of the function / (x) = x2 and of the suin S (x) of

Fourier’s series (11.58) are depicted in Fig. 11.5. The function S (x)
is continuous everywhere including the points x = (2A* -\- 1) jx,
k — 0, -jr 1. since for the function f (x) — x2 we have
/ ( —/ + 0) = / ( / - 0 ) = r-.
If we apply this expansion to the interval I —n , 5i|, that is for
I = a, equality (11.58) takes the form

x--i = —-
-T" - , A/ (*“ 1)^ /
> , ---- T7,---- c o s h , 71
.
X < 71 ( 1 J .5 1 ) )

h—1
Substituting x — U into (11.511) we obtain the useful relation

72 = ' - T - T - 17j— ••• 1 ' ( 1 1 . 0 0 )

3. Let / (x) he defined on the interval I—/, 21 by the relations

j c, for —/ < x < 0
1 l'r) ( c2 for 0 < r < J

Then
/
«o - J - ^ / (x) dx —-j- j Ci dx -i y ^ C2 dx ~^C\ 4 cz
-i -i b
/
1 /,fv , I f n :ix ,
a„ -- y j / (x) cos —— ax = y ^ ct cos —-— ax f
-1
i
l f
r y | C2C0S —j —dx —0, « —1, 2, . . .

and

bn - y f / (x) sin ^^-c?x = y j cl sin dx h

*/ -l

1I . flJtx . l o
1 y ^ Cj S i l l -j— CIX, a — 1, X, .

Consequently, //„ - 0 for even u ami l>„ .m - for odd a. Thero-

32_0t»2 i
 498 MULTIPLE INTEGRALS, FI EL D T H E O R Y AND SERIES

fore, by the convergence theorem, we obtain

e / r\ _f 2 (cs—ci) f_._ j ix , 1 . 3ju: . 1 . 5ax .
1 ( , ) ------2 ~ + — S----j s i n - p + -j-sm —j— r -j s m - j — r
c\ for —Z < x < 0

for 0 < x < Z

± e z for
~+~ I, x = 0

The graph of S(x) is shown in Pig. 1 1 .0 .

y
<5 C,*C3
V
J-, 1
Fig. 11.0 S i -SI ~4l S I -21 -I O I 21 31 41 SI SI

In Fig. 11.7 we see the graphs of the partial sums S t{x)} S 2 (x)
and S 2 (x) for the case ct = —1 and c 2 — - h i . In this case tlie series
turns into
4 1 ■ nx t . 3nx , 1 .
3 sin , + 5 sin ,
,
+ . . . j-\
S (*) = ~n l s l" T +
for — Z<C x < 0
f for 0< x< Z
Io for x = 0, x = dr I
6. Fourier Sine and Cosine Series for Functions Defined on Interval
| 0 , f |. Let a piecewise continuous and piecewise smooth function
/ (x) be dehned on an interval 0 <!x<;Z. It can be extended in
various ways to the interval — ^ 0 , in particular, (1 ) as an
even function or (2) as an odd function. In the former case we obtain
an even function on the interval I—I , Z], for which
/ i
<*o - 7 - j / (I) d& ak - - j - U (&) cos ^ r ~ 6 *= 0 , /c = l , 2 , . . .
0 0
(11.61)
and whose Fourier series, oii the interval [—Z, Zj, is of the form

k.l.i:
/(* ) 1 + 2I dh cos I (11.62)
 CH. U. FOURIER SERIES AND FOURIER INTEGRAL 499

In the latter case we have an odd function on the interval 1— f, J]

with Fourier’s coefficients

Oq 0, &h 0* 6* = 7 - ( / ( i ) 3 i n - ^ - d i (11.63)
0
and Fourier’s series
+°°
knx
/(* ) 2 ^ s in T (11.64)
fe=i
Each of the series (11.62) and (11.64) converges to / (x) on the
interval 0 < x <C / at the points of continuity of the function / (x).

Tlius, we see that every piecewise smooth function / (x) defined

on an interval 0 ^ x ^ I can he expanded both into a series of
form (11.62) involving only cosines whose coefficients are detrrminori
5U0 MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

by formulas (11.01) and into a series of type (11.64) containing

only sines with coefficients found by formulas (11.63). Series (11.62)
and (11.04) arc termed, respectively, o (Fourier) cosine series and
sine scries (they ore also referred to os Fourier’s half-range or incom­
plete series).
For example, let f(x) = x on the interval 0 •< x ^ L If f(x)
is extended “in odd fashion” we arrive at the function / (x) = x
defined on Ilie interval —I ^ x I whose Fourier series has already

been studied (see Example 1 and Fig. 11.3). Extending f(x) “in oven
fashion” we obtain the function / (x) = ( x | on the interval —I ^
Expanding / (x) = |x | into a cosine series on the inter­
val O ^ x ^ l we obtain, by formulas (11.01) and (11.62), the
relation
-f ■»
x — -f- V an cos for 0 x /
;t -\
where

flo = y J x dx -- /, an =■ y ^ x cos dx =

f 0° for even n
41 odd n
I n2«* for
Consequently,
3nz i 5jix .
! i —
■ a2 cos i +
for O^Cx^.1 (11.65)
The validity of equality (11.05) for x — 0 and x = I can easily
be established if we regard (11.65) as Fourier’s series of the even
function / (x) — | x | defined on the interval 1 -Z, /!, Such an inter­
pretation of series (11.05) enables us to construct the graph of its
sum for any values of x (see Fig. 11.8).
Generally, when an arbitrary piecewise continuous ami piecewise
smooth function f(x) defined on an interval |0. l\ is extended in
even fashion to the interval I—/, 01 we always have
/ (—0) = / ( [ 0) and / ( - / i ( ) ) = / ( / — 0) (1 I00)
 CH 11. FOURIER SERIES AND FO UR IE R INTEGRA!, 501
Consequently, the sura of the corresponding Fourier series is conti­
nuous at the points x — 0 and x = zkl, and its values at these
points are, respectively, equal to / (-F0) =■ / (—0) and / (—I + 0) —
= / (I — 0). If, in addition, the original function / {x) is continuous
at the end points of the interval (0. f), that is / ( h0) = / (0) and
/ (/ — 0) = / (/), the sum of its cosine series is equal to / (a:) at these
end points.
On the contrary, if we extend a function / (x) (originally defined
lor U in odd fashion to the interval —/ ^ x ^ 0, the sum
of the corresponding sine series may have discontinuities at the

points x — 0 and x = ± t despite the continuity and smoothness

of / (x) on the interval 0 - ^ x ^ . l . In Fig. 11.9 we see the graph
of l lie sum of I lie Fourier series corresponding to the function / (x) =
— x -f 1 originally defined for (J ^ x C /a n d then extended in odd
fashion. In the general case, when a function / (x) is extended as
an odd function from an interval 10, /) to the interval I—f, Oj we
always have / ( —0) = —/ (H-0) and / (—I I- 0) = —/ (I — 0). lint
for the sum of the corresponding Fourier series to be continuous at
the points x = 0 and x = ±Z it is necessary that the relations
/ ( 0) = / (-| 0) and f (—1 -1 0) = / (/ — 0) hold, and lienee this
sum is continuous only if
/ (-1- 0 ) = o and / (/ - 0) = 0 (11.07)

§ 3. FOURIER SERIES WITH RESPECT TO GENERAL

ORTHOGONAL SYSTEMS.
BESSEL’S INEQUALITY
Fourier’s trigonometric series expansion of a function / (x) is
a special case of expansion of / (x) in a series with respect to an
orthogonal system of functions.
I. Orthogonal Systems of Functions. Tiro functions q (x) and if(x)
inlegrahle* on a n internal If/, h\ a r e said to he o r t h o i / o n a f on

* In what follows, unless otherwise staled, integrahility is understood

it* tlm 9pr^p n( nrn*>nr iptocrrql ♦iri ) flip fi*T>r1?np^ htp quppmCp H f n ho rp.q|
502 MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

l<7, 6] if
V

^ <p (x) \p (x) dx = 0 (11.08)

A system of functions
<Pl (* )< <[7 ( * ) * • * ■» <fn ( * ) , • ■ • (IIB )
integrable on l a , b\ is called o r t h o y o n c i l on [a, 61 if
' r 0 for i k
\ <f« (*) <P* (*) dx = | ^ 0 (11.70)
for i= k
Let us consider some examples of orthogonal systems.
(1) The trigonometric system
1_ nx 71X knr Ajix
2 ’ cos T ’ sin COS
T * sin “7“ ’ (11.71)
is orthogonal on I—/, Zj.
(2) Rach of tin* systems of functions
j x 1 TT.r 2 j tx /fjtx
(a) — , cos — , cos -j— , . . ., cos —y ~
and (11.72)
2nx . knx
(b) sin , sin —— Sill —

is orthogonal on tlie interval |(J, /!.

(3) The system of Legendre's* polynomials
« = !. 2, . . . . e ,( * ) = l (11.73)

is ortlmgonaI on the interval I— 1, 11 (see Appendix 1 to Chapter 11).

f f a function (p ( t ) is integrable on fa, 61 the nonnegative quantity
b
Vi
II <PII= ( \ <P2(*) <**) (11.74)
u
is called the n or m, o f kp (x) on {a, 6|. According to relat ions (11.7U),
the norms of all the functions of an orthogonal system are positive.
Using the notation (11.74) we can rewrite relations (11.70) as
for i^ k
f Ti (?) Tft {*) dx = J ° (11.75)
* l n *r h n - for i= k

* Legendre, Adrien Marie (1752-1833), a noted French mathematician.

 oil. II. FOURIER SERIES AND FOURIER INTEGRAL 503

Wc now consider the norms of the fund ions of some orthogonal

systems.
(1) By definition (11.74) and relations (11.132). the norms of the
functions forming trigonometric system (11.71) are
k.ix i/7 1 An.r
1 T COS — j— = Vl | sin —j— = y 7, * = i, 2 ,...
(11.70)
(2) It can he easily verified that the norms of the functions of
systems (11.72) considered on the interval (0. /I are equal to
1 _ V'i Aaj
2 •} cos—j— k t. o
and (11.77)

(b)
A: = 1. 2,

(3) The norms of Legendre’s polynomials on the interval

| —1, 1) are
ii i --------
II Pn (*) II = ( f = y-zrrT I11-78)
-1
(the calculations connected with computing || J}n (x) || are given
in Appendix 1 to Chapter 11).
2. Fourier Coefficients and Fourier Series of a Function /' (.*•)
with Hespect to an Orthogonal System. Lot a function / (x) be integra­
t e over la. 61 mid let tho equality
- f oo
/ ( * ) = Y ah([,h{x) (11.79)
*&=»t
hold where ah are constant numbers and cph(x) arc the functions
of an orthogonal system of type (11.09) on the interval (a, 61. The
coefficients ah can he easily expressed in terms of f(x) if we multiply
equality (11.79) hy cpn(x), n 1, 2, . . and integrate the rela­
tion thus obtained term-by-term over (a, 61.* Indeed, multiplying
equality (11.79) by <pn (x) and integrating with respect to x from a
to 6 we obtain the relation
b -poo i

j / (<*) 0 n (*) dx — 2 ah J (^) (^c) dx, n = 1,2,... (11.80)

11 h=^J a

* For termwise integration to he permissible it is suflicienl that series

(It.70) converge uniformly or in the mean to its sum on the interval \n. M.
504 M ULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

All Ilie integrals on the right-hand side of equality (11.80) are equal
to zero for k n (see orthogonality relations (11.70)). Consequently,
we have
6 b
I / ('^) (^) ^ ■—(Iti f tPn (*f) dx = Q.n || (f n. 11“t ^ “ 11 2, . . .
*
n
J
a

whence
b
1 f
dji = || ^ jj]T J / (•!’) (•£) c l x t tl = 1* 2 , . * . (1 I . 8 1 )
a
The numbers an determined by formulas (11.81) are called Fourier's
coefficients of the function / (x) with respect to orthogonal system
(11.69) and series (11.79) whose coefficients are specified by for­
mulas (11.81) is referred to as the Fourier scries of the function
/ (x) with respect to orthogonal system (11.69).
For the numbers ah determined by formulas (11.81) to exist it
is sufficient that the function / (x) be inlograble on la, 6) (because
the integrability of the functions (p/, (x) is one of the conditions
of the definition of an orthogonal system). Thus, with every func­
tion f (x) integrable on (a, 61, we can associate its Fourier series
xvilh respect to system (11.69) orthogonal on (a, 61:

/ ( a ) ~ >]ar<‘|fc(x) (11.82)
h=\
where the coefficients of series (11.82) arc determined by formu­
las (11.81).
The conditions under which a given function j{x) can be expanded
into series (11.82) (i.e. the sign ~ can be replaced by the sign of
equality) depend on the properties of the orthogonal system (cfh (x)}.
In the case when we expand a given function with respect to Ilie
trigonometric system tlie conditions of the convergence theorem
proved in § 2, Sec. A arc sufficient for such an expansion to be valid.
Similar convergence theorems for other orthogonal systems involve
special investigation which we shall not present here.
3. Least Square Deviation. Bessel’s Inequality. Consider an arbi­
trary fixed number of functions belonging to system (11.09) ortho­
gonal on la, 61:
Ti (*). T2 (*), • * T h (x ) (11.83)
A linear combination
n
S WVh (x) (11 M )
h=\
 CH . Ii. FOUBIER SERIES AND FOURIER INTEGRAL 5

of functions (11.83) with arbitrary numerical coefficients a*., k

= 1, 2, . . . . n, is called a polynomial of order n with respect
orthogonal s>slein (1I.C9).
Let f(x) be an integrable function on (a, 61 and let i( be neccssai
to solve the following problem.
It is required to determine the values of the coefficients k =
= 1 , 2 , . . . , n, for which the corresponding polynomial of for;
(11.84) has the least mean square deviation (see Chapter 8, § 6. Sec.
from the function f (?) on the interval (a, 6]. Tims, we must fin
the values of the coefficients ctt, a 2? • • *» ot„ for which the quantit
n b rt

P2 ( /, 2 ah<l *) = ( [ / <*) — 2 U) ]* dx = / — 2 ry'u(t h

.1 h=l
(11.85
achieves its absolute mini mum.
Opening the square brackets we obtain
n i) n b

P2 ( /» 2 ) = \ /' (J) dx — 2 2 a '< ( / (*) Th (*) dx -f

k= 1 it to la
b n it

-f \ ( 2 w Y d x —\ / 2(* )dx 2 2 rz,,ah ii rf'h ii2 J' 2 a,‘ iif| h ii5

a ^= 1 a U~\ k*=1
(11.80)
o
because, by ( 1 i .8 I ), we have J / (x) q.* ( x ) dx = <Z/4 || ||8 a n d
a
b *
j <Pf ( * ) Cfh (x) dx = 0 for i ^ k and j <p£ (x) dx ~ || qpi, ||*

Completing the square we derive from (11.80) the relation

n h n n

P2 ( /, 2 “ h<Pfc) = J /*(*) dx— 2 al II Tt II2-f 2 (a h—akf II q* IIs

A= i n I /(I
(11.87)
Among the terms on the right-hand side of (11.87), only I lie last,
sum depends on the coefficients a J(. This sum being nomiegativo,
its greatest lower bon ml (equal to zero) is attained for a -- ah.
7t

In this case the mean square deviation p2 (/, 2 r*irPr) hHmV vo*?
h=I
50(1 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

its absolute minimum which is equal to

n b n

rninp2 (/, 2 ^ [ /( * ) — 2 **<!*(*)] d x ~

h— \ a fc = l
b n

— ( /» (x) d x — 2 HTa II2(11.88)

a h=i
The polynomial

2a*<ph(x) (11.89)
A=l
whoso coefficients a* are the Fourier coefficients of the function
f (x) with respect to the given orthogonal system (cpA (x)} is called
the Fourier polynomial of the function / (x) with respect to the
orthogonal system (i|j4 (x)}.
Thus, among all the polynomials of order n (where n is arbitrary
but fixed) corresponding to the given system {<pfc(x)} orthogonal on
la, hi, the Fourier polynomial of the function /(x) with respect to
the system (cp* (x)} has the least square deviation from the func­
tion /(x) on the interval la, hi.
Equality (11.88) expressing the least square deviation of the
Fourier polynomial of a function /(x) from this function is known as
Bessel's* identity.
Noting that the left-hand side of equality (11.88) is nonnegalive
we obtain the inequality
n b

2 i r < j /•(*)«** (H.90)

A=*i a
which holds for all n ^ 1 because its right-hand side is independent
of n. The sum on the left-hand side of inequality (11.90) is non­
decreasing as n -{-oo and therefore, since it is bounded from
b
above by the integral J f 2 (x) dx, it tends to a finite limit for
a
ti — f-°°- Fussing to the limit in inequality (11.90) as n —*~ + o °
we arrive at the so-called Bessel’s inequality:
+ oo b

2 ii /*(*)** ( u - 91)
a

* Hcssel, Prtedrieli Wilhelm (1784-1846), a German astronomer and mathe-

T&'X riyi
 CH. ||. FOURIER SERIES AND FOURIER INTEGRAL 507

By relations (11.76), in the case of the trigonometric system inequa­

lity (11.91) turns into
2 -r** 1
if- + 2 (“i + « ) < ! j /* (* )dx <11-92)
k—l ~l
It should be noted that we have established Bessel's inequali­
ty (11.91) for any function /(x) which is integrable in the ordinary
sense on the interval (a, b] but it can be generalized to a wider class
of functions.
We say that a function / (x) is square-integrable or square-sum-
mable (quadratically integrable or summable) if the integrals
b b
j f { x ) d x and J f2(x)dx (11.93)
a a
exist as proper or improper integrals.
It turns out that Bessel’s inequality (11.91) and, consequently,
inequality (11.92) remain valid for any function / (x) square-inte­
grable on [a, b].
Moreover, Bessel’s inequality (11.91) is also valid in the case when
the functions <pk(x) of the orthogonal system are square-integrable
t
on (a, 6|. In fact, the convergence of the integrals j /* (x) dx and
a
b

1<p* Or) dx and the evident inequalities

/- (x) -j- <p? (x) ipf (x)-\-ip\(x)*
| / (x) <f7, (x) | < ------ ^------- and <W(*) <pk (*) | < 2
imply (by the comparison test) that the integrals
b b
^ / (x) <jph(x) dx and J <p(x) tp* (x) dx
a a
are absolutely convergent. But when deducing Bessel’s inequality
we deal only with such integrals.**
Using the notion of a square-integrable function and introducing
the concept of orthogonality with weight function (see Appendix 2
to Chapter 11) we can derive the generalized Bessel's inequality which
holds for a still wider class of functions.

* In fact, wp Imvo 0 < (1 / (r) t — | <ph (x) IVs — /* (x) — 2 | / (x) q>h (.t ) +
-f (x) whence I / (x) <p* (x) I < - y \f* (x) 4 - (x)J.
** On some further generalizations. see Appendix ° to tt
508 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

Bessel’s inequality for the trigonometric system (see inequali-

a1 +°°
ty (11.92)) implies that the series 4f-T- V («ir b‘h) is convergent
h^l
and. consequently, we have
/
lim (ih—-J- lim I / (x) cos dx = 0
J< -f- 30 k~>~\ OO *^ ^

and {II 94)

i
lim bn —— 1iin [ f (x) sin dx —0
' h~+-rOO j 1

Relations (11.94) are a special case of the more general relations

t t
lim \ f (x) cos a x d x —0 and lim \ / (x) sin ax dx = 0 (11.95)
Ct^+ao j OL—►+c’c> j
which, according to Note 1 after the key lemma proved in § 2,
are valid for any absolutely integrable function / (or).

§ 4. SPEED OF CONVERGENCE OF FOURIER SERIES.

ACCELERATION OF CONVERGENCE OF FOURIER SERIES
We shall first consider the conditions guaranteeing uniform con­
vergence of a Fourier series and establish the relationship between
the degree of smoothness of a function / (x) and the rate at which
the coefficients ah and bh of its Fourier scries decrease as k —► r °°.
This will enable us to estimate the speed of convergence of the
scries.
1. Conditions for Uniform Convergence of Fourier Scries. We
shall prove that Fourier’s trigonometric series of a continuous and
piecewise smooth function / (x) satisfying anjadditional necessary
condition is uniformly convergent. We remind the reader that a func­
tion / (x) is said to be continuous and piecewise smooth on an interval
I—/, /I if it is continuous and its derivative / f(x) is piecewise con­
tinuous on this interval.
T h e o r e m / / . ? . If a continuous and piecewise smooth function
f (x) defined on an interval I — Zl assumes equal values at the end
points of the interval, i.e. f (—I) — f (I), its Fourier scries
+»
ao V {t qk cos knx , knx \
S (x) 9 — Sl"— ) m .9 ii)
 ch . li. F o u m r.n seu ies and fo u k iek in t e g h a i. 50!)

in w h ich
i I
o» = y ( a k ^ y j /(S )cos^ tZ 5.
-I -I

Ok - | f / (I) sir, - ^ d I, h - 1, 2, . .. (11.97)

-I
is uniformly convergent on the interval (—1> l\ and S{x) — / (x)
at each point of this interval.
Proof. To establish the uniform convergence of series (11.96)
it is sufficient 1o prove that the dominant number series
4-00

H- S (I «* I r |6* I) (11-08)
k= I
or, which is the same, the series
-*'OO
S (I a,. I ! | Oh |) (11-99)
h—1
is convergent. Indeed, this will imply, by YVoierstrass* test (see § I,
Sec. 2 in Chapter «S), that series (11.96) is uniformly convergent
on the whole x-axis since we have
knx knx
a «cos ■<|<2 a| and bh s i n bhf h —■1, 2, ..
for all x, —oo < i < oo. Let us denote the Fourier coefficients
of the derivative /'(x) by a'tt and b'k, i.e.
i i
a'k - y j /' (z) cos — d x , b ’k —j [ /' (x ) sin dx
-i -t
Integrating by parts in the formulas expressing the Fourier coef­
ficients ot the function / (x) we obtain
i
ah = y ^ / (x) cos dx =
r
1 I . , , . k n x *=■* I If ,, , \ kru . Ib’h
T*3T/(*> Sln — x , -» - ~r ~
l - I / ( r ) si n
/ b J ' w l
d x = -----------
a k f-
-i
{since sin —j X- turns into zero for x —± l \ nd

I I , knx .t=r
hu T i n ^ O) cos ~ r X - . - t ‘
510 MUI/l'IPLE INTEGRALS, FI ELD T H E O R Y A ND SERIES

because, in view of the condition / ( / ) = / ( — /), we have

knx x= i
cos /(* ) x = - l

= ( - l ) h[ / ( / ) - / ( - 0 1 = 0
Therefore
^ K l , \*k\ (11.100)
au\ . I ]■< n y—£ 1--- j - j i k 11 2, * •

But, hv the conditions of the theorem, the function i'{x) is piecewise

continuous on the interval [ —/, l\ and hence it is intcgrable on this
interval anil Bessel’s inequality
2 ‘h°° 1
4 « ’ + « ’> < t J /•* (* ) &
A— 1 - i

is fulfilled for f'{x). Consequently, the number series

.a + 00
—|—b 2 ia'h t b'k ) (11.101)
ft-i
converges. Furthermore, the inequalities
1 \2 /2 „ K I , l

ami

imply the relations

k i^ i
h < 4 - ( aiS + 'p -) and -^ < 4 (^ 1 + ^ )
Thus, we have
I caI . 1**1 1 t ,..*x i i (11.102)
fc k ^ 2 ^'h ' k2
• oo

The series jV. ~ is convergent

LJ kz *
(this can easily he established on
h—1
the basis of Cauchy’s integral test), and therefore, since series
(11.101) is also convergent, we conclude that the series
-t-*> j
V | 2 -hi,*) ! converges. Consequently, by the compnri-
u t
 CII. 11. F O U R I E R SERIES AND FOURIER INTEGRAL 511

son test and inequalities (11.100) and (11.102), the series

-j-oo

k=L
(which is dominant for the Fourier series of the function / (x)) also
converges. It follows that Fourier’s trigonometric series of the
function / (x) is uniformly convergent to its sum &(x) on the entire
x-a.vis. The validity of the equality S{x) = / (a:) on the interval
(—/, Z| is implied by the convergence theorem proved in Sec. 4
of § 2 because the conditions of this theorem are fulfil led here.
Thus, the theorem has been proved.
Note. The condition that the values of the function / (x) at the
end points of the interval I—/, /I arc equal is necessary for Fourier’s
series (11.96) of the function / (x) to be convergent to / (x) at the
end points. In fact, if the sum of the series satisfies the equalities
S( - / ) « / ( -/), (11.103)
the relation
5 ( - 0 = S (I) (11.101)
(which is a consequence nf the periodicity of the sum of series (11.96)
whose all terms are periodic with period 21) implies the equality
/ (—0 = / (/) (11.106)
Therefore, for Fourier series (11.96) of the function / (x) to be uni­
formly convergent l.o/ (.r) on the closed interval ( l\ it is necessary
that equality (11.105) hold.
If the values of a continuous ami piecewise smooth fund ion / (x),
which is defined on an interval I—/, /I and assumes equal values
at its end points, are changed at a finite number of points this results
in a discontinuous function whose values at the end points of the
interval may uo longer coincide. Hut the Fourier coefficients of the
modified function remain the same since the integrals expressing
these coefficients do not change. Consequently, by the estimates
obtained in the proof of Theorem 11.2, the series ^ (|aft| -1- |6*|)
k--.I
is convergent and thus the Fourier series of the modified piecewise
continuous ami piecewise smooth function is also uniformly con­
vergent on the interval [ /, /) and its sum is equal to the original
function but not to the modified function. Hut it can bo proved that
for the Fourier series of a piecewise continuous and piecewise smooth
function / (x) defined oil an interval 1- /. /] to be uniformly con­
vergent on that inh;r\al, it is necessary that all the discontinuities
of the function be removable and that, the equality of the limiting
values at the end points of the interval hold. i.e. / (—/ - 0)
512 MUL TU' I-i : IN T E C HALS, KIEL!) TTIF.OISY AND S E R I E S

= I (I - U). This can easily be proved on Ilie basis of the theorems

on conlinuily of the sum of a uniformly convergent series and
Ilie possibility of termwise passage to the limit in such a series
(see § 2 of Chapter 8).
Theorem 11.2 can be stated in a different manner. First of all
note that if a function / (x) is continuous on an interval I- I, l\
and takes oil equal values at its end points, the function obtained
when / (x) is periodically extended with period 21 is continuous
throughout the x-axis.
Next, let us agree that a function / (x) will be called piecewise
smooth on the whole x-axis if it is piecewise smooth on every finite
interval of the x-axis. Then Theorem 11.2 can he restated as follows:
If a periodic function / (x) with period 21 is continuous and piecewise
smooth on the entire x-axis its Fourier series (1 I .06) converges uniformly
to this function on the x-axis.
2. Connection Between the Degree of Smoothness of a Function
and the Speed of Convergence of Its Fourier Series. The investigation
of the connection between the properties of a function / (x) and the
speed of convergence of its Fourier series is important for applica­
tions of Fourier’s scries to some problems of mathematical physics
in which the sum of such a series is replaced by its nth partial sum.
It is also important in view of the possibility of term-by-term di­
fferentiation of such series etc.
T h eo r em , l l . l l . If a function f (x) and its derivatives /'(x), . . .
. . .. /(m)(x) (w ^ 0 ) are continuous on an interval I—l y l\ and
assume equal values at its end points, that is
/ ( i) - / (/). r ( o - r (/), • . / (m) < -/) = / (m) w (ii-io 6 )
and the derivative /(m«1} (x) is piecewise continuous on the interval
I —/, /I, the Fourier coefficients
i i
ah = — j / (|) cos — -dl and bh= -J- j / (E) sin d%
-t -t
of the function f (x) satisfy the relations

(ip k ) f°r ( u . m

and the series

^ hv (I a* I ‘ ! h |) , v - 0 , 1 , 2 .........m (11.108)
 Cl!. 11. FOURIER SERIES AND FOURIER INTEGRAL 51:

are convergent. *
Proof. Integrating by parts, as was done in the proof ol
Theorem 11.1, we obtain
t

« > .= ! f / (?) COS di =

In these calculations wc take into account that (1) condition (11.10)

of lhe theorem is fulfilled, (2) tho cosine is an even function and (3)
the sine of the argument , k — 1, 2, . . ., turns into zero at
the end points of the interval I—/, /) and therefore we have the
relations
A.r£
/'” (I) cos = 0 and / ‘^ (£) sin h =0
i ' ' / ||= -f
for 0 ^ s ^ m
The curly brackets under the sign of integration in (11.109) mean
that, depending on the number of times the integration by parts
lias been performed, the factor by which the function /*m+1) (|) is
multiplied is either cos-— or sin—? .

* delations (11.107) mean that

lim ——— = 0 and lim bh = 0
1
/.-mil /-m-t-1
 514 MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

In just tlic same way we obtain

, f . LiZ
, /mt L i r —
cos——
A'-T- tfc (11.110)
-I
From (11.109) and (11.110) we derive
J I , I t I lm+i //|ajlmtl>{
1a h I ; I| |
I\
1^1 1^1 ^niTl ( £m+l 1 J (11.111)
where a|tm+1> and 6jlm+n are the Fourier coefficients of / <n,fl,(x).
Since a™*1* and &hm+,) tend to zero as k-+- — oor it follows
from (11.111) that
ao = o ( s s r i ) . 6 * = ° ( iF f l) for I/-*---co (11.112)
Relation (11.111) implies that
,mM , | ,J—.Cm+n<I I
fcm (| au | 4 -16ft I X f[ T lt+ l ( )
< {I r- -r 1 I 2 r ^ -} ( 11. 11 :>.)
because
b t m+i > |
i «r*” i i <
4 - ( lW " !* + * •)
Therefore, by. Bessel’s inequality
•f O
O
| «#"+»> |2 ^im+n
2
A=1
2<l
f - H 6 r +1,i* ) « T I i/'",+" w i 2 d*
-l
-fo a

and by the convergence of the series 2 ~£T we conclude that the

h^\
-f oo
series 2 k™ ( [ah| 4- |6h|) is convergent. It follows that all
A= 1
the scries (11.10S) are convergent and thus the theorem has been
proved.
Note 1, To apply the above theorem to the expansion of a func­
tion / (x) which is defined only on an interval [0, /I into a series
with respect to the system of functions sin —^ , k = 1, 2, . .
we must verify the validity of the conditions of Theorem 11.3 for
the function F(x) defined on the entire interval I—J, I] which is
obtained when f{x) is extended as an odd function. In particular,
for the function F(x) to be continuous at the point .r = 0 it is neces-
 CH. 11. FO U RIER SERIES AND FOURIER INTEGRAL 515
sary that the equality /(0) = 0 hold because, if otherwise, IHe
extended function would have a discontinuity at x — 0. Analogously,
at the point x = I we must have / (/) = 0 because the extended
function should satisfy the equality F(—/) = F(Z). The derivative
of an odd function being even, the condition F'(—I) — F'(l) auto­
matically holds for the derivative of the function F(x).
Generally, for the derivatives of the function F(x) to be conti­
nuous on the interval 1—I, Z1 and to have equal values at its end
points we should impose the conditions
/ (v) (0) = / (v> (I) = 0 for v = 0, 2, 4, . . .
Then the corresponding conditions of Theorem 11.3 automatically
hold for the derivatives of F (x) of odd orders.
In particular, for the series

2 V l M . v - 0 , 1, 2
k=\
I
(where bh — -j- J / (|) sin d \ are the Fourier coefficients of
o
a function / (x) defined on the interval (0, /)) to be convergent
it is sufficient that the function / (x) satisfy the following conditions:
(1) the function / (x) and its derivatives /' (x) and /" (x) are
continuous and the derivative fm (x) is piecewise continuous on the
interval 10, Jl;
(2) / (0) = / (I) and r (0) - T (0 = 0.
Xote 2. If a function / (x) satisfying the conditions of Theorem 11.3
on an interval I—/, 11 is periodically extended with period 21 from
the interval I—Z, l\ to the whole x-axis, the extended function and
its derivatives up to the order m inclusive are continuous and periodic
with period 21 for all x, —oo < x <C -{-oo. Therefore Theorem 11.3
can be rephrased as follows:
If a periodic function f (x) with period 21 and its derivatives up to
the mth order inclusive (m ^ 0) are continuous for —oo < x < -f-oo
and the (m + 1)th derivative (x) is piecewise continuous1 the
Fourier coefficients ak and bk of this function with respect to trigono­
metric system (11.12) satisfy the following conditions:
(1) ah—o and b|<= 0 (jtsn ) for k — -\- oo
-(-oo

(2) the series 2 &v (I <**. I+ I I), v —0, 1,

are convergent.
Hence, this theorem establishes a relationship between the degree
of smoothness of a periodic function and the speed of convergence
of its Fourier series.
33*
516 MULTIPLE INTEGRALS. FIELD TIIEOnV AND SERIES

Note 3. If llie conditions of Theorem 11.3 ore fulfilled for m Z> 0

the Fourier series of the fund ion f{x) can be differentiated term-by-
term at least m times, that is we have the relations

f{S) (*) = S (°* cos bh sin n r )<3) (1 1.114)

for 1 ^ s ^ ni, —/ < x < i

since the dominant series
+ CO
— A:' (I rth ! -1-1bk J), 1 A *< ni
i
are convergent.
Note 4. Tlie proof of Theorem 11.3 enables us to estimate the
speed of convergence of a Fourier series, i.e. to obtain an estimation
for the error arising when the sum of the Fourier series is replaced
by its partial sum. In fact, under the conditions of Theorem 11.3,
we can apply relation (11.111), the Oauchy-Bunyakovsky inequality
for sums, Bessel’s inequality for Fourier’s coefficients of the func­
tion (x) and the obvious inequality
4-00 + 00
dx
y,
'*0
x'-m - I
which leads to the relation
-r°° +^ 0
J 2j
1
-T&fcSin — j j <
Jl=&o+l
(I H I^ I)

/m+1 //--- ;---

V S i V 2 2 <l«8", T - H « r ,T )
/t-/L 04 -l

< S^rrr C \ i « i ) ( f \ I^ (x) I*dx ) ’* “

/t0 ~t

™4~t-
---- ---- ~
— rr ( i I (*)I2 d x )' !
= ° ( —
a ”*+i(2m-r l)v * ' j , I »>+-i- j
’
^0 ' t'O /
3. Acceleration of Convergence of Fourier Stories. A Fourier series
with respect to the trigonometric system appearing in a concrete
problem may turn out to converge so slowly Lhat it is impossible
to use it for practical purposes when its sum is unknown.
 CII. 11. FOURIER SERIES AND FO U RIER IN TEG R A L 517

In this connection it is natural to pose the question whether

it is possible to choose for a given slowly converging Fourier series
-pOO
/ (*) —4^- + 2 { ah C0S^F + 6*s i n , — / < x < Z (A)
n—t
a new slowly converging trigonometric series whose sum cp (x) is
known such that the series entering into the relation
-1-30

/ (x) = <f (x) 2 (a»C0S — <B)

A=1
converges fast enough, that is such that its coefficients o.h and p*
tend to zero sufficiently fast as k — H-°°-
When passing from representation (A) of the function / (x) to
representation (B), we say that the convergence of series (A) is
a ccelera ted (improved).
If some singularities of the function / (x) (such as the limiting
values of / (x) and of the derivatives /<*> (x). s - - 1, 2, . . w,
at their points of discontinuity and for x ± 1 etc.) are known the
convergence can be accelerated by means of a simple function <j> (x)
(which is known) having the same singularities as f (x).
For instance, suppose we know that / (x) is continuously differen­
tiable on the interval I—I, /I and 1iin / (x) = A-l. The values
X~+ ‘ (
of / (x) assumed al the end points of the interval 1—/, I] being une­
qual, series (A) converges nonuniformly on this interval. Let iis
put (x) — x. The latter function is also continuously differen­
tiable on the interval 1—/, /] and possesses the same limiting values
at its end points as / (x). Therefore the function / (x) — x is con­
tinuously differentiable on the interval I—/, /1 and its end point
limiting values are equal. Consequently, the series entering into
representation (B) of the function / (x) for <p (x) = x is uniformly
convergent on the interval I—/, /).
Now let us discuss another method of improving convergence
of series (A) which is applicable when we have no additional infor­
mation concerning its sum. This approach was suggested by
A. NT. Krylov*. The basic idea of the method is that we select those
coefficients an and bn of series (A) which contain the lowest powers
of the quantity ami try to compute the sum of the auxiliary
series with these coefficients by using the tables of Fourier series
expansions of various functions whose Fourier series are slowly
convergent.

* Krylov, Aleksei Nikolayevich (l8C.3-104.r»), a prominent Kussian mathe-

nialjrian and engineer.
518 M U LTIPLE INTEG RA LS. FIELD THEORY AND SER IES

For example, let it be necessary to improve the convergence of the

seri es
-OO
/ (*) ^ ( — 1)n sin nx, — n < x < J i (M.115)
n—2
We have
«3 1 .1
r»4—1 n ' «’»—n
and therefore
—oo
-n < x < n
n=2 n=2
But we have already found (see Example 1 in § 2, Sec. 5) that
-U^o

X= 2 2 (— —a < x < n
n=*l
(to obtain the last expression we must put / = jt in the above-men­
tioned example). Hence, we can write
^
-^»-oo
, . vTI sin n.r x . __ _
2 j ( — 1)" — = -y j-s in i, — J l< X < JX
n —2

Consequently,

/(* ) = — -f- -r?>» *-i- 2 ( ~ 1)" —n < x < n (11.116)

2

Series (11.110) converges much faster than the original series (11.115).

§ o. UNIFORM APPROXIMATION OF CONTINUOUS FUNCTION

BY TRIGONOMETRIC AND ALGEBRAIC POLYNOMIALS.
WEIERSTRASS* APPROXIMATION THEOREMS
Let e >> 0 be an arbitrary fixed number. If the inequality
i <p (x) — i|) (x) 1 < e
is fulfilled for all x £ la, simultaneously we say that the function
<p (x) is uniformly approximated on the interval la, b\ by the function
ij> (x) to an accuracy of e.
T h e o r e m 1 1 .4 (rH e i e r s t r a p s 9 T r i g o n o m e t r i c A p p r o x i ­
m a t i o n T h e o r e m ). If a function f (x) is continuous on on
interval —I ^ x I and assumes equal values at its end points, then
 CH. 11. FOURTER SERIES AND FOURIER I N T E G R A L 511)

for every s U there is a tv iffo u o n te tr ic p o l t /n o tn ia l

T„(x) = ^ - + % ( « * c o s ^ + 6 * 8 i» ^ - ) (U .H 7 )
ft= l
which uniformly approximates the function f (x) on the interval
•—/ ^ x ^ I to an accuracy of e.
The proof of this theorem is based on the following
Tjentnia,. For every continuous function f(x) defined on an interval
a ^ x ^ h and for every e > 0 there exists a continuous piecewise
smooth function ge(x) defined on this interval such that
| } (x) — ge (x) | < -|- tor all X e la, 61 (11.118)
and
& (« )= /(« ). & (6 )= /(6 ) (11.119)
Proof of the lemma. Since the function / (x) is continuous on the
closed inlorval la, b] it is uniformly continuous on it, that is for
every &> 0 there is d = 6 (e) >• 0 such that the inequality
|/( x ') - /( x " ) |< |- (11.120)
is fulfilled for any x' and x" belonging to the interval [a, 61 and
satisfying the condition | x' — x* | < 6 (e). Therefore, if we break
up the interval (a, 61 into subintervals lx*, x i+1l, i = 0, 1, . . n,
of lengths less than 6 by means of points of division x0 = a *<
. < I,- < Xi+jt C . . . < xn+i = 6, inequality (11.120)
bolds for any two points x' and x 0 belonging to one and the same
subinterval |Xj, x{+1).
Let us construct a continuous piecewise smooth function y —
— Se (x) defined on the interval fa, 61 by pulling gt (xf) = / (xj)
for i = 0, 1, . . n-\-i and ge(x) = gE(x|) - f ( x —x*)
*i +l — x i
for Xj ^ x ^ x i+1. This means that the function y — ge (x)
is linear on each subinterval (xlt x l+Jl, i = 0, 1, . . . » n,
and its graph is a polygonal line inscribed in the graph of the func­
tion y — } (x). According to the construction of gz (x) w c have
S t («) = / («). S t (*>) = / (6)
Now we can show that

I f i x ' ) - S * ( * ' ) I .< t

for any x' £ [a, 6l. Indeed, let, for instance, x* £ [xf, x i+1l. The
function g* (x) being linear on the subintcrval (x*, z (+1), the value
?v fy ) Hoc bcMvooii the value? (*,) - / (xt) and (xi+1) =
520 MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

= / (xi+l). As is known, a continuous function assumes all inter­

mediate values lying between its end-point values and therefore
the function / (x) takes on all intermediate values between / (x*)
anil / (x* + j) on the interval la:*, x*+1). Hence, there is x " £ lx*, x i+1l
such that / (x") — g£ (x'). Consequently,
I i <4
since x', x ” £ lx*, x/+1), which is what we set out to prove.
Proof of Theorem 11A . By the hypothesis, the function / (x) is
continuous on the interval I—/, /] and has equal values at its end
points: / (—/) = / (I). Let us take an arbitrary c >►0. According
to the above lemma, there is a continuous piecewise smooth func­
tion ge (x) defined on I—/, l\ such that
1/(*) —£«(*) | < - r for allx6l —L l\ (11.121)
and
«e ( - 1) = 1 ( - D , g. U> = / (0
which implies
gt ( - 1) = gi (I) ( 11. 122)
since
( - 1) = / (0 f

By Theorem 11.2, Fourier’s series of the function g, (x) is uni­

formly convergent to g. (x) on the interval I—I, 11. Hence, for
a sufficiently large n, the nth partial sum Tn (x) = -vr +
n
T (a* cos —y-~ -| bh sin —y—) of this series satisfies the rela-
ft=t
tion
| ge (x) — Tn (x) | < y for all x £ ( — L /] (11.123)
From (11.121) and (11.123) we conclude that
| / (X) — 7’rt (X) | < | / (X) — g e (X) | - f | g e (X) — ( X) | < - | - f Y :£
(11.124)
for all x £ I— /, I]. The theorem has been proved.
Note. Taking a number sequence Cj, e2i • • •» . . . convergent
to zero we can construct the corresponding sequence of trigonometric
polynomials Tnx (#)* Tna (x), . . . uniformly convergent to the
function / (x) on the interval [—ly I]. But in the general case these
trigonometric polynomials are not partial sums of one and the
t A
iiiiiit I r i ^ u i i B i i i u U ’i c . S f r ic . i. In ftK I , t h it .- *n>] ^ i i u l i M . i i
 CH. II. FOURIER SERIES AND FOURIER INTEGRAL 021

ponding to a given e > 0) which enters into inequality (11.124)

is Fourier’s polynomial of the auxiliary continuous piecewise smooth
function ge(;r) which changes when e varies and lienee the coef­
ficients of the polynomial Tn{x) also vary. But it should be noted
that the above fact is not a consequence of the specific method used
for constructing the polynomials Tn(x) in the proof of Theorem 11.4
because it can be shown that in the general case a continuous func­
tion / (x) may not be the limit of a uniformly convergent sequence
of partial sums of one and the same trigonometric series. Indeed,
if f (x) were the limit of a uniformly convergent sequence of partial
sums of a trigonometric series

(afcC O s^ + PfcSin— )
*=i
on the interval I—/, l\ this series would necessarily be the Fourier
series of / (a;). But there are examples of continuous functions / {x)
defined on an interval I—I, l\ whose Fourier series diverge at a finite
(and even at on infinite) number of points belonging to the interval
(—/, /l. Such examples are rather difficult to construct and we shall
not consider them here.*
T h e o r e m 1 /-«5. ( W c ier&t r u s s ’ l * o ( y n o m i a l Tpp-rojcimu,-
tio n T h e o r e m ) . If f (x) is continuous on an interval a ^ x ^ b,
then for every e >» 0 there exists an algebraic polynomial
f}m (r) = A 0 ~~ A ,x -f -42x2 -|- . . . -}- A mxm
such that
! / ( * ) - Pm(x) 1< e (11.125)
on the interval (a, b].
Proof. Take a sufficiently large I > 0 such that the interval (a, b\
is strictly contained within the interval I—/, /|. Lcl us construct
a continuous function F(x) defined on [—/, l\ by putting F(x) =
= f (x) for a < * ^ b, F (—1) = F (I) = 0,
F (ar) = ——j (x 4* /) for — 1-^.x^a and F(x) — -j—^ (l — x)
for b ^ x ^ / (i.e. F(x) is a linear function on the intervals I—ly a\
and fby /I). By the Weierstrass’ trigonometric approximation theo­
rem. for every c ;> 0 there is a trigonometric polynomial
n

T,4(r) = "°- -f 2 (<**« cos ~~ si n ~ r ~) (11.12fi)

fe=i

'L.g. acu i»j.

022 MULTIPLE INTEGRALS. FIELD THEORY AND SERIES

such that for all —l ^ x ^ l we have

|f ( x ) - 7 - » ( x ) |< i. (11.127)
Let us expand into Taylor’s series the trigonometric functions
entering into (11.127):
cos ka* _ , s i kW x‘— (11.128)
I 2)1* X 1
and
A3a3 -.3 ' A3.t3 (11.12D)
S il l — r—
3U3 "olW
Power series (11.128) and (11.129) are absolutely convergent for
all x, —oo < z < -j-oo, which can be, for instance, proved by
applying D’Alembert’s tesL. Substituting (11.128) and (11.120)
into (11.126) we arrive at a power series
T'n (x) ^ .40 A. iX -f- AgX® -j- • • . -f- «rlmxm (11.1*30)
which is convergent for all x , —oo < x < -foo. Consequently,
series (11.130) converges uniformly on every finite interval of the
x-axis and, in particular, on the interval [—I, l\. Therefore, taking
a sufficiently large m we obtain the mth partial sum
P m (x) = A 0 -h A tx + . . . + A mxm
of series (11.130) such that
| Tn(x) — P m(x) |C - |- for all x £ | — /, I] (11.131)
Prom inequalities (11.127) and (11.131) it follows that
I F (x) — P m (z) I < IF ( x ) - 7 \ (x) I+ 1r„ (X) - P m ( x ) |< - - + y - e
(11.132)
for all x 6 I—/, l\ and, in particular, for all x £ [a, 61. But we have
F (x) = / (x) for all x £ (a, 6] and thus inequality (11.132) turns
into the relation
I /( * ) - Pm (*) l < e (11.133)
which is what we set out to prove.
Note. If we take a number sequence elf e2, . . b>(1 . . . conver­
gent to zero we can construct the corresponding sequence of algebraic
polynomials P mi (x), P mi (x), . . . uniformly convergent to / (x)
on [a, 61. But in the general case these polynomials may nol bo
partial sums of one and the same power scries which can be confirmed
bv an argument similar to the presented above in connection with
trigonometric pr»!vnomin!c f«oe the note oftor Theorem 11
 CH. 11. P0UK1EK SERIES AND lO lIU E R INTEGRAL

VVeierstrass' theorems do not provide effective means for practical

construction of polynomials uniformly approximating a given
continuous function to any preassigned degree of accuracy e ;> 0.
The problem of constructing polynomials yielding the best appro­
ximation was posed and solved by Chebyshev*. Cliebyshev’s polyno­
mials make it possible to construct effectively polynomials appro­
ximating a given continuous function.
We now denote by H n the totality of all algebraic polynomials P m
of degree m ^ n. Let P m (x) 6 Hn and let / (*) be a continuous func­
tion oil \a, 61. The number
E </, P m (*)) = max | / (*) - P m (x) \

is termed the deviation of P m (x) from / (a) on [a, 61. The greatest
lower bound En (/) of the values of the quantity E (/, P m) when
Pm (x) runs through the whole set H n is called the least deviation.
Chebyshev proved the existence and uniqueness theorem for a poly­
nomial of the best uniform approximation, that is a polynomial
P in (*) € Hn such that
E (/, P m) » En {/)
and also developed some methods of constructing such polynomials.
He found the so-called polynomials of least deviation from zero,
also known as Chebyshev’s polynomials (sec* [21, vol. 2, Chapter 4).
In practical problems of constructing a polynomial uniformly
approximating n continuous function / (.r) on an interval (<?, 6]
to an accuracy of e it is important to obtain a polynomial whose
degree is as low as possible. It is obvious that such a polynomial
is Ihe polynomial of the best approximation (for the function / (x)
on the interval la, 61) belonging to the totality Hn with n such
that
En ( / ) < e < E n_x </)

§ 6. COMPLETE AND CLOSED OHTIIOGONAL SYSTEMS

Let us denote by Q [a, 6] the class of functions containing all the
piecewise continuous functions defined oil [a, 6l. We shall introduce
the notions of complete and closed orthogonal systems for the func­
tions belonging to the class Q Ia, 61. Basic theorems concerning
these notions will also be proved for this class of functions (Theo­
rems 11.(5-11.10). But it should be noted that this theory can be
generalized to a considerably wider class of functions, namely to
the squarc-intcgrable functions on la, 61 and even to the functions

* Chebyshev, Pafnutv Lvovich (1821-1894L a famous Hussian matliernali-

ci.ui.
324 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

which are square-inlegrable with a weight function p (x) on [<z, 61

(see Appendi-v 2 to Chapter 11).
1. Complete Orthogonal System. An orthogonal system of functions
^fl (**01 (*^)i • • ♦♦ *Pn (*^)» • • • (11.134)
on 1cl* 61 is said to be complete if for every function / (x) belonging
to (J [a, 61 its Fourier series with respect to orthogonal system (11.134)
-f-oo
2 0,(f*(x) (11.133)
k=l
where
b b
Ch = if— ,i J / (*) <P* (*) dx, II <Pfc II = ( j <pfc(*) dx)'* (11.13(1)
a a
is convergent in the mean to / (x) on [a, 6]t that is
n l / 7 i

p2 (/, 2
Jt= l
= J [ /(* ) — 2 W * (*) J dx^O forn- CO

(11.137)
In this case we say that system (11.134) is a basis of the functional
space <J [a, 6l since, in the case of completeness, for each “element”
/ (x) € Q l«. 61 we can write the generalized equality
-:-ao b
f (x) — 2 c,<cp,t (x) where ch = ^ / (x) <pft (x) dx (11.138)
1 a
which should he understood in the sense of convergence in the mean
on the interval Ia, 61, i.e. in the sense that relation (11.137) is ful­
filled (see § G, Sec. 1 of Chapter 8).
2. Parseval Relation as a Necessary and Sufficient Condition for
an Orthogonal System Being Complete. Let us make use of Bessel’s
identity
n b n

p2 (/, 2 Ci^ k) = \ \ f ( x) — 2 ]"d x =

fc»=l a fc=l
b ti
= 2 ci II'PhII* (11.13*))
a k=i
(see § 3, Sec. 3). Passing to the limit in (11.139) for n -► 4-oo we
obtain
71 It-j* 0 0
I o
n n p- ( /. 2 c“‘fx) = j f * ( x ) d x - 2 I11*11* (11.140)
71 v to
• I .• t
 CII. 11. FOURIER SERIES AND FOURIER INTEGRAL 525

whence it follows that the relation

1i m p2 (/, S c*<l>* ) = 0 <11.141)

7l“>-pOO /<=1
is equivalent to the equality
b -i-o°
f P ( x ) d x = 2 t C£II*P*II* (11.142)
a k —\

Equality (11.142) is known as Parseval’s relation. Thus, orthogonal

system (11.134) is complete if and only if for any function / (x) £
6 Q l^» 61 Parseval’s relation (11.142) is fulfilled.
3. Properties of Complete Systems. An orthogonal system
<Pi (x)t (*), • * •» (*)• • * • (11.143)
on an interval la, 61 is said to be cloxetl if every function f (x) 6
6 (J Itf, which is orthogonal to all the junctions of system (11.143)
is the zero element of the functional space Q (a, hi, i.e. f (x) is equal
to zero at all its points of continuity and hence is different from zero
only at a finite number of points of the interval la, 61.
Theorem 11M. If system (11.143) is orthogonal and complete
on (a, 61 it is closed.
Proof. Let a piecewise continuous function / (x) be orthogonal
on la, 6] to all the functions of system (11.143), i.e.
b
( / (-?) (^) dx = 0 for k - 1 , 2 , . . .
ml
a

Then the Fourier coefficients of the function / (x) with respect

to system (11.143} are equal to zero:
b
Ch = || if 1 \[- i ^ ‘f ,l (*) d* " 0, k =* 1, 2, . . . (11.144)
a

By the completeness of system (11.143), Parsevnl's relation

b +oo
J /* (,r) dx = 2 c*II 'I1'* IIs (11.145)
a h-^i
is fulfilled for any function / (x) £ Q la, 61. But then, by virtue
of (11.144), equality (11.145) implies that

| / 3 (x) dx = 0 (il.1 10)

521* .MULTIPLE INTKUU\LS. PIEL1> T H E O R Y A XL SERIES

Suppose dial j (x0) ^ 0 where x0 £ b\ is a point at which / (x)

is continuous. Let us embed the point x 0 in an interval la', bf I
(Iving within {a, 6]) on which / (x) is continuous. The function / 2 (x)
6'
being continuous and nonnegative oil la', b‘ 1, we have j f 1 (x) dx >
a'
b
> 0 since f 2 (x0) >►0. Bui this implies that I /- (x) dx > U whicli
J
a
contradicts equality (11.146). Consequently, / (x) == 0 at all the
points of continuity on [a, id, and thus the theorem lias been proved.
T h e o r e m i f . 7. Ij two junctions f (x) and g (x) belonging to
Q la, b\ have the same Fourier series with respect to complete orthogonal
system (11.143) on la, b], these functions coincide as elements of the
space <J [a, 6], that is they may differ only at a finite number of points
of the interval [a, b|.
Proof. The function (x) — (/ (x) — g (x)) £ (J [a. 61 is orthogonal
to all the functions of system (11.143) on la, id- In fact, we have
It b o
j 4’ (x) (f/ft (x) dx = j / (x) <p*J(x) d x — j g (x) <f/, (x) dx =
a n a

= c|||tpAl|2 = (ci-—c?)||^Jt||2, k - = 1 ,2 ,3 , . . . (11.147)

where ck are the Fourier coefficients of the function / (x) and c*
are the Fourier coefficients of the function g (x). By the hypothesis,
the Fourier series of these functions coincide, i.e. cj = cft for k =
— 1, 2, . . . » ami hence it follows from (11.147) that

J <J-(x) rph (x) dx = 0 for k = 1, 2, 3, • (11.148)

a

But. then, by the foregoing theorem, the difference \\> (x) = / (x) —
— g (x) is identically equal to zero on [a, id at all the points of
continuity of \\; (x) arid thus may he different from zero only at
a finite number of points of the interval la, id which is what we
set out to prove*
T h e o r e m t i . S . If system (11.143) defined on the interval [a, 6i
is orthogonal and complete on la, 6], then, for any two junctions j (x)
and g (x) belonging to O la. id ire have the f/e t i e r a l i p e d P a r s e r u t
rela tio n
b -i »
\ f (x ) g (x) dx = 2 cl cf II <Vk II" (11.149)
<< U~1
 Cll. ] 1. FOUIUEP SERIES AND FOURIER IN T E G R A L 527

where c* (cjj) are the Fourier coefficients of f (x) (g (x)) with respect
to orthogonal system (11.143).
Proof. Equality (11.149) is obtained if we write Parseval’s rela­
tion for the functions / (x) ~ g (x) and / (x) — g (x) and then sub­
tract the latter from the former and take half of the result.
T h e o r e m I t . 9. Let f (x) £ Q \a, 6]. If an orthogonal system
of functions {cpf (x)} defined on la, hi is complete, the Fourier series
of the function f (x) with respect to the system (q?/ (x)} can be integrated
term-by-term, that is
X X

( / © d |= 2 ft. J <f* © d | (11.150)

x<> h= 1 x0

for any x0 and x belonging to the interval (a, hi.

Proof. This theorem follows from the fact that the Fourier series
+»
2 chiph (x) converges in the mean to / (x) on [a, fr|. Indeed, as
was proved, it is allowable to integrate termwise the series which
converge in the mean (sec Theorem 8.142 in Chapter 8, § 6, Sec. 3).
4. Completeness of Trigonometric System.
T h e o r e m 11.10. The trigonometric system
1 nx . ax k:xx k:i.r
— ♦ cos — , sin — , . . . , cos —j—, sm —j— . . . (11.1.11)

is complete.
Proof. It is required to prove that the relation
i
P~(/> T!tx) - \ [/(*) — Tn(x)]z dx-*- 0 for « -> oo (11.1.>2)
-I
holds for every piecewise continuous function / (x) defined on [—/, ZJ
where

Tn (*) ■= -5- S
iv= r\
( a,t cos *^T bh sin )

is the Fourier polynomial of the function / (x) with respect to

system (11.151).
Let |/ (x)| <C M on [ - /. Z1 and let e > 0 he an arbitrary nuinher.
Without loss of generality, we can suppose that / (x) has a single
point of discontinuity x 0 lying in the interior of 1 Z, Zl. We shall
construct a continuous function g (x) on the interval I -Z, ZJ such.
528 MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

that it assumes equal values g (—/) = g (I) at the end points of this
interval and satisfies the inequality
t
P’- ( / , g ) = j \ f ( x ) - g ( x ) f d x < - i (11.153)
-I
For this purpose we take a sufficiently small 6 >►0 and put g (x) =
= / (x) for —I ^ x ^ x0 — 6 and for x0 -{- 6 ^ x ^ I — 6, g (I) —
= / (—/), and consider g (x) being linear on the intervals x 0 — 6 ^
^ x ^ x0 ~~~ 0) and I — 6 ^ x ^ I (see Fig. 11.10 where the graph

of / (x) is shown in the continuous line and the graph of g (x) in the
doited line). According to the way g (x) has been constructed, we
have g ( —/) = g (I) =■ / ( —Z), and the difference / (x) — g (x) may
be different from zero only for x0 — 6 < x < x0 -f 8 and / — 6 <C
<C x < /. Therefore we can write
l xo-j-6
•y-U. &) - j \ H 4 - s ( x) \ - dx - j \!(X) - B ( x ) \ * d x
—I *0—6
t *0+6
+ I \ ! { x ) - g ( x ) \ * d x < , j { |/( x ) | rlfiW Ip d * -!-
1-6 *0-6
I
-I f {| / (X) | + | g (x) |}2 4jW826 i 4 M 26 = 12jI#*6 c \
1-6
provided that 5 > 0 is sufficiently small. Since the function g (x)
is continuous on the interval I—Z, Z) and takes equal values at its
end points {g (—Z) = g (Z)), Theorem 11.4 (W’eierstrass’ trigono­
metric approximation theorem) implies that there is a trigonometric
polynomial
r„„ (x) = ^ -)- 2 ( “ ft COS— + sin - j - )
A=1
such that
I? for all I 6 t - f . f l (11.154)
Consequently,
V
P* Of. g -
( l« (X) - r», (x)|2 dx < ( rfx = £ (II. 155.
-/ -/
Now taking advantage of the inequality
(a b f < 2a* - r 262
and pulling a = j (x) — g (x) a n d b = g (x) Tna (x) we oblaic
{/ (x) - Tno (x)}2 < 2 {[/ (x) - g (x)l2 -4- Ig (x) - Tno (x)|*)
ll follows that
i /
!>"-(/. ?'»„)’ f 1/ (*) — Tna (x)|2<fx < 2 \ U(u:)-U'(.c)\2<h
-t -I

■ 2 | | g ( i )--y „ „ (x )l2d x < 2 i - ;

-I
If we substitute the Courier trigonometric polynomial T‘n (x) cor
responding to the function / (x) for the trigonometric polynomial
Tno (x) in the last inequality we shall have
I>* (f. T l j ^ r. (11.1511)
because Tn^ gives the minimum to the moan square ilevialiou.
Making u^“ *»f Bessel's identity we can rewrite inequality fi l.lbli)
in the form
l Tlq

Ps (/, ’/'!.») = i" (x) dx - l {4 - i- S (« i i «> } < e (11.157)

-I 1
Consequently, we have
i n
(>*(/, 7’,',) = f / 2(x)rfx —/ { ^ - I- 2 (<tf &?<)} < * (11.15X)
—I k—1
for all n ^ n0. Since 8 I> 0 has heen chosen quite arbitrarily, il
follows that fr (/, Th) -► 0 for rc -■* } oo which is wiiat we set out
to prove.
The completeness of the trigonometric system which lias been
established here implies that this system is closed. It also moans
that a piecewise continuous function / (x) is uniquely specified by
its Connor series with respool to the trigonometric system every­
where on the interval I I. /) except possibly at a linile number ol
points (at which / (x) is discontiuiious). The completeness of the
14 OS-J'i
530 MULTIPLE I N T E C F A L S , FI E L D THEORY AND S E R I E S

trigonometric system was for the first time proved by A. M. Lya­

punov*.
5. Completeness of Some Other Classical Orthogonal Systems. In
mathematical physics we deal with various orthogonal systems,
other than the trigonometric system, whose completeness is establish­
ed similarly. As an example, let us prove that the system of Legen­
dre’s polynomials is complete. Consider a piecewise continuous
function / (x) defined on the interval [ — 1, 1|. Suppose we are given
an arbitrary r ;> 0. By analogy with the proof of the completeness
of the trigonometric system, let us construct a function gt (x) which
is continuous on 1— 1, 11 and satisfies the relation
i'
i’2 (/\ S e ) = j d / U ) — g t (2-))2 d x < — ( 1 1 . loll)
-I*
(but here the requirement that ge (x) must assume equal values at
the end [joints of the interval I—1, 1| is dropped). Theorem 11.5
(WeiersLrass* polynomial approximation theorem) indicates that
then* exists an algebraic polynomial
Qm (x) = -f A iX -f- AvX2 -f- . . . -f- A mxm
satisfying the inequality

for all —1 < x < 1. This implies the inequality

i
f>2 (£*, Qm) - j [g, (x) - Qm (x)l2 dx (11.160)
-i
The functions 1, x, x2, . . ., xm are linear combinations of
Legendre’s polynomials** and therefore
Q m (*) = ^ 0 + ffit’i (*) + B zP7 (x) + . . . + HmP m (x)

where P^ (x), . . P m (x) are Legendre’s polynomials.

Taking advantage of the inequality (a + b)z ^ 2a2 -f- 2b2 we
conclude, on the basis of relations (11.159) and (11.100), that
n2(/. Cm)•< 2f*»(/, ff,) + 2p*{/r„ <?„,)«2 4 + 2 4 (11.161)
Let us substitute the Fourier coefficients of the function / (x) with
respect to the system of Legendre’s polynomials for the coefficients

* Lyapunov, Aleksandr Mikhailovich (1857-1918), a prominent Russian

mat hem at irian.
** See Appendix 1 tu Chapter 11.
 CH. 11. FOURIER SERIES AND FOURIER INTEGRAL 531

B .............. B m in t h e e x p r e s s i o n of p 2 (/, (Jm), i . e . r e p l a c e B hJ k =

= 0. 1, b y the q u a n titie s
i
cu — | | ^ t ]p ' \ f ( x) Ph{ x) dx, k —0, 1 ( 1 1 . 1 0 2 )
- 1
S i n c e F o u r i e r ’s p o l y n o m i a l s m i n i m i z e t h e m e a n s q u a r e d e v i a t i o n ,
the a b o v e s u b s titu tio n c a n n o t increase the m ean sq u a re d e v ia tio n .
Therefore, in tro d u cin g the n o ta tio n
Vin (*) - *0 T cir l (x) 4- . . . - r cmP M (x) (1 1 .1 0 3 )
we a r r i v e at the in equality
PM /. «?,'■)<«: (11.1134)
Fnrtlierm ore, by B e s s e l ’s i d e n t i t y
t >i
P2 ( / , \ p { x ) d x ~ Y 1 c l | | v* |P (H.10r>)
-1
w e c o n c l u d e t h a t r e l a t i o n ( 1 1 .1 0 1 ) i m p l i e s t h e v a l i d i t y of t h e i n e ­
quality
(>*(/, ( > ' ) < e (1 1 .1 0 0 )
for a l l n ^ m. S i n c e e > • 0 h a s b e e n t a k e n a r b i t r a r i l y , it f o l l o w s t h a t
p2 (/> Qh) 0 for n +oo (1 1 .1 0 7 )
a n d th u s wc h a v e e s ta b li s h e d t h e c o m p l e t e n e s s of t h e s y s t e m of
L e g e n d r e ’s po I ynorn i a Is. *§

§ 7. KOI: It I Kit S E R I E S IN ORTHOGONAL SYSTEMS

OF COMPLEX FUNCTIONS
Hero, b esid es re a l f u n c tio n s , we s h a l l also c o n s id e r c o m p le x
f u n c t i o n s of a r e a l i n d e p e n d e n t v a r i a b l e x w h i c h a r e of t h e f o r m
<p (x) = q>* (x) 4" i q * * (*) (11.16S)
w h e r e cp* (x) a n d cp** (x) are real fu n ctio n s. T h e fu n c tio n w hich
is t h e c o m p l e x c o n j u g a t e o f <p (x) ( i.e . t h e o n e w h i c h d i f f e r s f r o m
<p (x) o n l y in t h e s ig n o f i t s i m a g i n a r y p a r t ) w ill be d e n o t e d b y
<p (x). T h u s ,
q> (x) — <p* (x) — i q * * (x) ( 1 1 . 1 6 8 ')
It s h o u l d be n o t e d th at
<|> <*)q (*) - lq* (*)l2 ~r lq** (*)l2 = | <| (x) j2> 0 (11.11311)
Af u n c t i o n q
(x) — q * (x) + iq)** (x) is s a i d to b e continuous
{piecewise continuous) on |« , M if i t s r e a l a n d i m a g i n a r y p a r t s , I h a t
34*
532 MULTIPLE IN 'T E G H .\ I jS, FI EL D T H E O K Y AND SEHIES

is the functions q* (x) and q** (x), are continuous (piecewise con­
tinuous) on fa. b\.
The derivative und the integral of a function q> (x) = q;* (x) -j-
-f- iq>** (x) are defined, respectively, by the equalities
<i(f dq>* , . £?<(•**
dx dx 1 dx
(11.170)
and
b b b

[ qp(x) dx — j q>* (x) dx-J- t j <p** (x) dx (11.171)

a a
and the function q> (x) = q,* (x) dp** (x) is said lo be differen­
tiable {integrable) if q>* (x) and (p** (x) are differentiable (integrable).
If two functions <p (x) = q>* (x) -f dp** (x) and ip (x) = (x) -f-
r ill’** (x) are integrable on l«, 6), it is obvious that the function
q (x) q>(x) is also integrable on Ia, b\. In particular, if q'(x) is
integrable on la, 6] the function q> (x) q> (x) is also integrable on
la, 61 and
b b i.
j fp (x) <p(x) d x - ^ l<p (x) |2dx — f {lq>* (x)]2 -l Jcp** (x)]-} dx > 0
a n ri
Two functions q (x) and i|* (x) integrable on an interval (a, b\
are said to be orthogonal on this interval if

\ <! (•*) »r(x)dx --0 (I 1. 172)

«*
A system of complex functions
T i (^) * q 2 (*^) * ♦•*i *1H (^) i ••• (I 1.172)
inlegrahlc on (a, b\ is called orthogonal on (a, 61 if
t \~ / \ j for /st=/>’1 (11.174)
‘IV (*) 'I k (x) fix - (n ^ ||f > 0 for . _ ,J

■b%
where IJfp,»; || -- \ ] <|/c (x)\- dx is the norm of q<,. r.enerally, the norm
of an integrable complex function rp (x) is defined as the nminegn-
tive (jurmlilv
0 b
!! q 11= ( \ <| (x) q* (x) d.r j “ - ( \ | fp (x) I2 tlx) (11.17:»)
tJ *}
 CII. 11. FOURIER SERIES AND FOURIER INTEGRAL 53:t

One of the most important examples of orthogonal systems of

complex functions is the system
nnr

H = 0, 1, + 2, . (11.170)
which is orthogonal on the interval [ —Z, ZJ. The orthogonality
, k jtX , U TIX

of e 1 and e 1 for k ^ t i is established directly by integra­

ting the product
I. — :— 1. TI3T3C
hZlX . fenx . ictx . Cfc—
tonx (fc—n)Lnj_ ^.
-----t —]— *• *, = COS
rn« '------
e 1 (e 1 ) = e 1e ' = e =

( A — « ) JTX
—i sin
fl.TA
over the interval [ — Z, Z]. For the norm of the function e
we obtain the expression
. 71JUC . . nnx . nnx . .
/ / " = (J e ' e 1 c/x) = ( ( dx) = V i‘ l (11.177)
-I -I
The Fourier coefficients of a function / (x) (inlegrnhle on l«. bI)
with respect to orthogonal system (11.173) are determined by the
formulas
b
ch 1|(^ ^ j / (■*) (*) dxi t 1, 2, . . . ( 11 .1
a

The Fourier series of an integralde function / (x) with respect

to orthogonal system (11.173) is, by definition, the series
-{-no
f{z)~ ciMh (*) (11 170)

whose coefficients ck are determined by formulas (11.17S).

In particular, the Fourier coefficients of / (x) with respect to
system (11.176) arc equal to
i r • *t7lX
ChT=^ r \ / M * 1 A*= 0, zLl. =h 2, . . . (11.17«S')
-1
iind the corresponding Fourier series of f (x) with respect lo this
system is the two way series
-few TtTlX

/ (*) ~ cne I (11.17!tr)

T> —
534 MULTIPLE IN T E G R A L S . FIELD THEORY AND S E R I E S

Lei us prove that, if the function / (x) is real on the interval I—f. Jl,
relations (11.178") and (11.179") are equivalent to the relations
/
afc —— ^ / (.r) cos c/ r> /,- —0, 1. 2, . . .
(H. ISO)
b{. f /(x ) sin dx. k -- 1, 2. . . .
-t
ami

/ (*) ^ -? r -r 2 \ a,i cos~ T ~ ^"^i‘ s,n —t 1- ) (11.181)

h—l
I Ini I is relations (11.178") and (11.179") are the complex form of
Fourier's coefficients and series of the function / (x) with respect
to the trigonometric system.
A p p ly in g K i l l e r 's form ula
<
>»<
p _ cos ,j_ i s in (|

to (11.178") we obtain

Co = 4 - ( = ^ (11.182)
-I
. hnx

Ch - 4 r ( f (x) a 1 dx ~

t f »- fr .i.r . . A-a
--2T ] / ( * ) j cos — ----- i s m - j dx ~

I ^
)c — \17 —? •• • (11.183)
an

c- h= 4 r f f ( x) e' 1 dx = ~ w \ f ^ f cos “T ~
-i -i
. . knx | , ah ' ihii k. —
L—1 I ^ *>y • • • (11.184)
i sin —;—
/ Idx -
/%

Scries (11.179') can he rewritten as

V
ru tx + 3D . t a -j-so —Ij/mx
_I = ca -t- 2 ' -r 2 C-ft* (11.185)
n *= — - jo h-=\ ft
Substituting expressions (11.182). (11.183) and (11.1S1) of the
coefficients cl{ and into (11.185) and taking advantage ol
 CH. II . FOUKIF.lt SEMES AND FOURIER INTEGRAL 5!15

Ruler’s formulas
•*.*-»*
cos f| = , sin <p=

we arrive at the equality

‘ 30 .. njtx
TtJtX 'v . kax “ ^ —i kax
V „ Z~t— _ <*o ■ X* ah—*Ok — V <ih—ibh
Z j Cfl€ ~ ~ ~ * r Z j -----2 1 Zl 2
7)—’ - Aw *—l
/m.v - 1 Jrnx
4 + 2 h -
fi= 1
hax . hax

, j • < i «»,
t i . „ ---------- 7i---------- J “ T +
+
. / A'rfi/ * • A*ji<r % (I I .-ISO)
+ 2 j ^ttfcCOS — J------ Oj, SHI — j — )

which is what we set out to prove.

If the function / (a?) is not only inlegrable but also piecewise smooth
on I —I, 11, the theorem on convergence of a Fourier series ami r«‘la-
lion (11.ISO) enable us to write down the expansion
k]+op . tint!
f ( x)^- 2 c»c' ij <11.1K7)
n --oo
wli(»ro the Fourier coefficients ore determined by Ilie formulas (11.178).
At t!ic points ol discontinuity of the function / (i) the left-hand
side of equality (11.187) should be replaced by —— 1' ——
if -l< x < l and by /(<-"> + /(--<•+0) ;I ;.= ± /
Complex form (11.178'), (11.179') of expansion of functions into
trigonometric series is widely used in mathemalica I pli ysics and
its applications. This form is especially convenient for performing
calculations and transformations, in particular, when the expres­
sions we deal with involve the products of Fourier’s series and also
for Fourier’s series of functions of several independent variables.

§ 8 . FOURIER SERIES FOR FUNCTIONS

OF SEVERAL INDEPENDENT VARIABLES
Lei a function / (x, y) he defined in a rectangle —/, x / 1,
— fa ^ ;/ ^ /?,. We shall suppose that for every y £ | — \z. L) the
function / (x, y) satisfies Ilie conditions under which it can he oxpan-
o.Hi MUI.TIIH.K J M ’BOHALS. F1K1.I) THEOHY AND SERIES

dec! into Fourier's series as a function of x on the interval (—Zlt />].

Then, using the complex form of Fourier's series we can write
+oo . nnx
/(* . if)= S cn (y)e <> (11.188)
n = — oo

where
11 __i r»n|
°n U/) = j /(I , y) e '* dc, H = 0 , + 1 , ± 2 , . . . (11.189)
-i i
Lei each of the functions £„(!/), its turn, have the Fourier
series expansion
+oo . mny
Cn{y)= h * n = 0, =fc 1, - t 2, . . . (11.190)

on the interval —12 ^ y $$ lz. The coefficients cnm are determined

by the formulas
12 . mnn
i r _ 1—i—
c nm j C tl (l)) 6* 2 dij, n, m = 0, t 1, ± 2 , . . . (11.191)
—io
.Now, substituting (11.ISO) into (11.191) and (11.190) into (11.188)
we obtain
-1 -0 0 . / n n .v in jiy

/ (*i y) -=n=i]—» > ]

C n m e K '■ '* (11.192)
= — oo

where
/1 it / ru t* , m nr) \

(11.193)
I J
-/i - /2
;» ^ l) e ' " '• \

Thus. we have obtained, in complex form, the Fourier series expan­

sion for the function / (x , y) of two independent variables
Making use of Euler's formula ei,p = cos cp -j- i sin cp we can
rewrite expansion (11.192) in the form
- p JO
i, \
/ (X, //) \
n , W = — no
(/ cos mu , . nnx \ /
^ 1sin ~~r~ I \ cos ~ r r +
mnr/ .

-f-oo
r l S ill
mny \ - r mix wmv .
) = Zi I (ltun COS COS —j— - f
/j- r m
m ,. n — —cw

n n x nnx . m7ty ,
gmn sin COS i T ^»in COS 11
U . Sill --- r ^ - -4-

H T tr mrt// ] (11.104)
1tn n sin —
sin
 CH. 11. FOURIER SERIES AND FOURIER INTEGRAL >)3«

where

-J- for m-=n = 0

— .* ~ for m = 0, n > 0 anti m > 0, n —0 (11.195)
I1 for m > 0 , « > 0
and

timn = ( j / (*, */) COS cos dx dy

"Ji —
<i /?
l ttJUT mny
bmn — lxU \ \ i (*. //) sim h COS lz
—fi —
(11.19(1)
i nnx mni/
Cmn — ixi2 j f f{Xy y) cos sin
h l2
-it —it
/j iz
r. nai sin tnny
dn\n — iti. j [ f ( x . //) sin U

If / (xy y) is on even function with respect to ouch argument, that

is if’
f x. a) = f (x, —.y) = 1 (Xy //)
we can easily see that 6 mu = c'„m — d mn — 0 and therefore the
Fourier series of such a function takes the form
+30
// \ vi nJl'r tuny , it , ilOV
/ j J/) — / | ^>njt&inn COS • * OS ; (1 1 • 1 ‘ ^ )
^
m # 7 i= 0
M *2

If / (ar, #) is odd with respect to each argument x and y. only

its coefficients d mn can be different from zero and, consequently,
its Fourier series is of the form

/ (-**> lf)= dfim Sin sin — ( 1 1 . 1 !H»)

T7i# 0

H / (Xy y) is even in y ami odd in x its expansion contains only

the functions sin co.s and if it is an odd function of u and
an even function of x the expansion involves the function';
nnx mnif
cos —;— sin —7 —.
I J /**
..;*S MUI.TIPLE INTEGRALS. FIELD THEOKY .AND SERIES

More we do not investigate the conditions guaranteeing the possi­

bility of expanding a function of two variables / (x, y) into a I'ourier
uuilliple scries a m i limit ourselves to the formulation (without proof)
ol the following° assertion: if f (x , Ju).
' <t >
anti ° Ox
otj
f—Oyarc con-
tinuous junctions periodic in x with period 2/j and in y with period 2tz,
the Fourier double series of the f unction / (x. //) con verges to f (x. //)
at every point.

%9. FOUtlKH INTKC.ItAI,

1. Formal Derivation of Fourier Integral Formula. In this section
we present, a formal derivation of Fourier's integral formula which
is obtained if we take the Fourier series of a function / (x) on an
inlerva! I—1%t\ and tiion extend this interval to infinity, i.e. pass
to the limit as I j-°°- Then the Fourier series turns into the
Fourier integral.
When we consider a function / (x) defined on an arbitrary finite
interval (—I. 11, if is expanded into a sum o f ‘'harmonic oscillations’'
whose frequencies form a descrete number sequence. I»ut- when wo
take the limit fur I —*- ~.-oo we pass to a function defined over the
whole x-axis (or on the semi-infinite interval [0, +ooj of the x-axis)
which is presented by an inlegral, i.e. a sum el* “harmonic vibra­
tions" whose frequencies ^ constitute the continuous interval U
. a, *< : oo. Let us consider this formal passage to the limit from
Ilie Fourier scries to the I'ourier integral.
lari. / (x) hi* defined on the entire x axis. We shall suppose that
the function / (x) is piecewise smooth on every finite interval [—I. 11,
flu'ii. hv the convergence theorem for trigonometric series, we have,
for ciery I ;> 0. the expansion
I-X >
kn
V ( a,, cos T . . t o \
(11.200)
/ (■■'■) *> h>‘ s,n j~ )
u- i
where

( 11. 201)

’’'quality (11.20(1) is valid if x is an interior point, ol llie interval

i —L l\ at which / (x) is con I inuous. If x is an interior point of the
interval at which the function / (x) has a discontinuity, the left-hand
side of equality (11.200) must lie replaced by liic expression
 CU. II. FOURIER SERIES AND FOURIER INTKGRAT. 5iW

/ (r- 0 ) / (x— 0 )^ Substituting formulas (11.201) into (1 1 .2 0 0 )

wo obtain
i f
/w = ^ r t / (I) cos a - x) <i\ (n .202)
-1 h=zI -I
II the function / (x) is absolutely integrable over the whole .r-axis,
that is
-1-00
j |/ ( i ) |r f j = (> < f-oo (M . 2 (i:i)

then, by virtue of condition (11.203), the lirst summand on the

right-hand side of ( 1 1 .2 0 2 ) tends to zero for / -foo. (Consequently,
we have
-*-ao I
/ (.r) = lim -j- Y \ 1(1) cos — a - x ) d i (11.204)
/->4-oo 1 J
/l-l -/ 1
kn
flJlr J"J
Putting 2111 —/v;, and -f- — _\a* we can rewrite relation (11.20-1)
in the form
• <x> f

/ (r) = lim — 2 ( / (?) ros hfi (? — J*) <&> ( I 1.2) In)

/ . 4-:<© ’^ *.
AA.fc-*0 h 1

\\ r can now conclude, intuitively (without any rigorous argument).

Ih-t I
(
( 1 ) Ihe integral \ / (H) cos ?./, (c —-r) dt can be rejdared, for large

values of L by the integral \ f (t) cos kt, ( 9 —a) o?H, and

—oo
-<(-00 4*°°
( 2 ) the expression ^ \ /(£) cos A/t (9 — x) dc is an integral
i<=i
4-00
sum corresponding- to the integral f dk f /(£)eos/. (£— j;) 1/ 9 , and
0 —30
llms. relation (11.20-3) implies the formula
-f-OO f-OO
/ (j:) — | dh ^ / (c ) c o s k ( 9 — x) d£ ( 1 1 . 2 0 l»)
(I —go
If u: is a point of discontinuity of f(x). the left-hand side of (ll.20fi)
should he replaced by the expression '-1 ~ ()1 —/(« Itidali-
540 MULTIPLE IN T E G R A L S , FIELD THEORY AND S E R I E S

on (11.200) is called Fourier’s integral formula (theorem) and the

integral on the right-hand side of (11.200) is called the Fourier
integral.
2. Proof of Fourier Integral Theorem. The above derivation of
formula (11.200) is formal because it involves some passages to
the limit which have not been justified rigorously. But it turns
out that it is easier to present a direct proof of formula (11.200)
than to justify these passages to the limit.
T h e o r e m 11. I I , If f (x) is a piecewise smooth function on every
finite interval of the x-axis absolutely integrable on the whole x-axis,
-i-OQ

i.c. I | f (x) 1 dx is convergent, then

f -f-oo

Ii m - — f dX f / (E) cos X (| — x) dt —Zif-Z 0 ) + / ( r - 0 ) (11.207)

/-•■J-AC J J
+°°
Proof. Note that the integral ^ /(£)cosX (E— x ) d \ dependent
—oo
on the paranieter /. is uniformly convergent with respect to /. for
0 v' X <c - oo since | / (E) cos X ($ — x) \ | / (£) | and the integral
T^ [ / (c) | <1% is supposed to bo convergent. Consequently, it is per-
— OO
inissiblc to reverse the order of integration (see See. 3, § 2, Chap-
I -[-oo

ter 10) and rewrite the integral | dX j /i(|) cos A,(£ — x)riE in
the form
r i-oo f-OQ /

X - ]' <n. \ / (I) cos x (I - x) di = - L j di ( / (5) cos.x (5 - 1 ) dk =

0 —oo — ■DO 0

= 4 - ( / ( ; ) si" / g — *) = — \ l(x L j ;) liiiL d £ (11.208)

J
—C*J
b X 71 J
— oo
b
where — r, <VX ~'/E. Hence we must only prove that
1 f / ( x -- 0 )
1
Inn — \ / j(x j, £)
... Sill I t
1
—y.— dt, -= —— (11.200)
;-»• i-=c -i c,
and
+ >»
I• 1 r r/ , sin It. f i(xJ
* - 3- 0 ) (11.210)
11 n * ^
 Cl!. It. FOURIER SERIES AND FOURIER IN T K fi K A I.

For this purpose we shall taJke advantage of the relation

5
1
n T*>«*_-*. ( 11.21

which is implied by formula (10.50) derived in § 2, Sec. 5

Chapter 10.
As an instance, wc shall establish the validity of relation (11.21
(formula (11.209) is proved similarly). By equality (11.211). •
can write
/<■r-M>) I I* S i n 11
= - \ /< * -» ) (K (I 1.21
oo
v , sin 11
Therefore the ilifferoncc between the variable | / (j* - 1) si>! 1*
‘o
and the constant entering iuLo relation (11.210) is equ
to the expression

Jo. ~ ( / (j + O - ^ - r f S - =

j 1/(1 I O - / U ^ ) ) |^ < /; (11.21:

6
Thu:-. we muM show that the integral on the light hand side •
(11.210) tends to zero as I ->■ -j oo. For this purpose we di\iih- tl
interval of integration 0 ^ < , oo into three parts, iiamei
0 £ :$C 6t b ^ ^ A and A ^ £ <C r ° ° - Then the integr
we are interested in is presented as the sum of ihe correspondii
integrals:
J 0, +ou " J .6 ”1' 6, A —• ^A, +3® ( 1 1.—
1
We shall first, lake an arbitrary and prove that for all su
ficiently small d 0 and all sufficiently large A > <S we have tl
inequalities
| a! < 4 and IJ I< — ( 11.2 1;
which are fulfilled Tor all I I simultaneously. Then we shall It
some values of b and \ for which inequalities (11.215) hold an
then choose a sufficient l v large I's^ 1 for which Lho relat ion \ L\ \<
«f
is satisfied. The key lemma (see § o) indicates that sue
a choice of / 1 is possible. Then, by (M.2M), wc shall conclmi
that | Jo. .j.*, I <C s for all sufficient Ivr larire / '> 1
542 MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

Tims, let us begin with estimating the integral

y0.
> f /<* + P rw ' ( « + °) S,n licit
** J
0
For all sufficiently small v c have
Hx + D - H ' ;~0> | < I/;, (a) I 1 for all £€(0, 8)

Consequently,
A .» K -{ l/k W I v 'I X j for oi l 8 < 3 ^ ■ (11.211',,

and for all values of /.

Next, let us estimate the integral

*^A, +oo :
We ean write
r/C . \!{s »-n)| | f sin C
Ij *. + - 1 - \ i/o i-0 1 4 -i ji
dl
C
•foo
1
• n\ j\ •'
I H'* ^ *') |1d w
tT• *
/A
^ + |O j -’ “>l I f * " I L r f I where $♦ « ( 1 1 . 2 1 7 )
nA I
/A
»

By condition (11.203), \vc have Q = ^ j / (x) j d x <C oo and, there-

*
— QL«

fore, for all sufficiently large A > U the inequality

f sin
is fulfilled for all I. Furthermore, the integral ^~ —d^m being
convergent, we have
I f ir ^-U)| b\
n r* ¥
IS
for a l l s u f f i c i e n t l y l a r g e A >» 0 a n d al l / ^ 1. lienee, hv virtue
of ( 1 1 . 2 1 7 ) , th e i n e q u a l i t y
( U .2 ik>
holds for all sufficient Iv large A > 0 and all / ^ 1.
 CH. 11. FOURIER SERIES AND FOURIER I N T E R UAL 5 4 3

Finally, let us estimate the integral

y s i n / : (/; (11 . 210)

b

The expression “ ---- ^ — —------------------- is a piecewise smoothfunction

of the argument ^ on the interval 6 ^ \. Consequently, by the
key lemma (see § n). tlio inequality
1' (11.22H)
I'^6. ^I ^ 3
is valid for all sufficiently large / ^ 1. On the basis of (11.21b).
(11.218) and (11.220) we conclude that, for all sufficiently large
/ ^ 1, wo ha\c the relation
IA + - K * (11.221)
which is what we set out to prove.
Aro/<?. Tin* above theorem on Fourier’s integral can he proved under
more general conditions imposed on the function / (a:). Namely,
if the function f (:r) is absolutely integrable over the. x-axis and satisfies
the conditions that (1) it is piecewise smooth on every finite interval
of the x-axis and (2) the expression - ^ „ 1 ^J-r^ is bounded
W

for any fixed x and alt sufficiently small £ ;> 0, the above theorem
remains valid.
Indeed, the p ro o f ol the theorem reduces to estimating the three
integrals J&. A. and ./A, ^ for J o,-h*. and the three integrals
J - a._ n and which aro considered similarly, liy tin*
absolute intcgrability of / (a:), the integral /„.$ *s small for all
sufficient ly large A. The integral /<,•« is smail for all sufficiently
fix (x+ 0);
small h ^ 0 provided that the expression
is hounded for every fixed x and all sufficiently small t, 0- To esti-
mate the integral
, 1 i* /(*+&)--/(*4-0)
J 6, a ~ —) —---------- =------------ s »n L, at
A
vve note that the function ip (£) — / + 0 / (*4 b) pj(,cevs ,*st.
continuous on the interval 0 < 6 I; ^ A for anyfixed x. Let
fa, 6] be an interval on which tp (£) is continuous. Take an arbitrary
s 0. lad us construct a piecewise smooth function gt (x) sm h
that the inequality
I <P (£) — ge Q I < - r r j j r - ^ r f o r N < r * <" t <' A
5M MULTIPLE INTEGRALS, FIELD THEORY A ND SERIES

holds (which can be done as in the proof of Weiorslrass1 trigono­

metric approximation theorem). Then we have
b b
jj <p ( y s m / u s | < j h » ( ? ) - g E (;)|rfs-^
a a

o
+ 1j g, & sin l l d\ —e

for all suflicienlly large / ;> 0 since the key lemma is valid for the
piecewise smooth function ge(£). breaking up the integral J A
into the sum ol integrals taken over the intervals of continuity
of «p(£) we see that /$. for / — — whi ch completes the
proof of the theorem.
3. Fourier Integral as an Expansion into a Sum of Harmonics*
Fourier’s integral formula (11.200) can be rewritten as
+eo
j (x) - - ^ [A ( a ) co s Xx B (7.) sin 7.j.) dX (I 1.222)

where
«v> -r ^
.1 ( 0 - - 4 - \ « (0 = 4 [ fG isiu X idi (11.223)
• * ♦'

Kelniinti (11.222) is analogous to an expansion of a function into

a trigonometric series, and expressions (11-223) are similar tu llie
formulas for llie Fourier coefficients. Let us transform expression
(11.222). We have
A (7.) cos f*x r B (A.) sin Xx = N (7.) sin {Xx -j- qx) (11.224
where
A 0.)
A' (/.) - \ A1 (7.) i rr- (X), cos <{>K= A‘ (>.» sm rf7. = NBO)
(X)
(11.225)
Thus, relation (1 1.222) can ho interpreted as an expansion of a func­
tion / lx) defined for all x, -oo < x <1 -r°°» into a sum of harmonic
oscillations whose frequencies A continuously cover the semi-infinite
interval 0 :<C X <C T"»- The functions A (?-.) and B (a) (see (11.225))
give us the law of distribution of the amplitudes and initial pha­
ses fp;.. when X varies over the positive half-axis X, 0 7w<C i-oo.
If a function fi x) is defined mi a liuMe interval ( —/. 1I it can
bo cvnnndcd. under the conditions slated above, into harmonic
 CH. 11 . FOUKIEB SEHIES AND FOURIER INTEGRAL 545

use illations:

/ (*) = i 2 ( a'1cos ^T~ '• bh sin ~T~ ) =

h=.i
+cc
= - f - - 2 A'* sin (>•** <Pa) (11.226)
^jX
whore the frequencies Aft= — , Ar= l, 2t form an arithmetical
p rogression.
4. Fourier Integral in Complex Form. The Fourier integral formula
can be rewritten in the complex form
•^•oo -{—
o©
/(* )-4 j j /(« )* * * -» < £ (11.227)
— 30 —OC
4 -0 0

equivalent to (11.2U13): the integral j / (t) cos a (x — £)dE

—oo
4“
is an even function of 7. and the integral j /(£)sin>. (x — h) d\
—oc
is an odd function of A, and therefore, we have
; OO -{-30 4-00 -f co
.*i !j* dk j / (E) cos A ( | — a) c/| -L. j dX j /(E) cos ) . ( z - l ) d l
— *30 —OO

and
4*OC 4-00
— j dX j / (s) sin X (i — |) d |= 0
—O
O —oo

Consequently, by Euler’s formula

eiX{x-5) — cos — q q_ i sin X (a: — |)
we can write
-f-O0 4-00 4“00 4-00
4 j rfX j /(E) dE = _L | d>. { / (I) cos X ( i — E) --
— OO — oo — oo — oo

4"oo 4~oo
I
I — j d>. f / (I) sin X( i — I) d t —
—oo —
oo

4“30 4~oo
= — j <1X j / (J) cos X(x — I) dt
0 —oo

35—0824
54G MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

whence it follows that formulas (11.206) and (11.227) are equivalent.

But it should be noted that in the general case the integral
; 90 +OO
j d>. j / (I) sin X (x - 1) d | = 0 entering into the relation is
—oo —oo
understood in the sense of Cauchy’s principal value (see § 3 of
Chapter 9), that is
-j-a o -f-ao

j j /( |) s in X(x — Q d l =
—oo -oo
I
— lim j dk
^ /(£)sinX(x —£)d| = 0
—I —oo
5. Fourier Transformation. Equality (11.227) can be rewritten
in the form
+oo
1 1
/ ( * ) « q/2H f dl (
V2it j /( » * - * «
(11.228)
—oo

Introducing the notation

f
— ao
(11.229)

we obtain, by (11.228), the formula

+oo
/ (x) = — I— f 7 (X) e'>* d k (11.230)
— oo

The function / (X) is called the Fourier transform (or spectral characte­
ristic) of the function / (x) defined on the real x-axis, —oo<C x<4-oo.
The transformation from fix) to / (X) performed according to for­
mula (11.229) is called the Fourier transformation. Formula (11.230)
(Fourier’s inversion formula) expressing the original function f (x)
in terms of its Fourier transform / (X) describes the (Fourier) inverse
transformation, and / (x) is termed the Fourier inverse transform
of / (X).
Now we can rephrase Theorem 11.11 on Fourier’s integral as
follows:
T h eorem , l i . t ' i . If f (x) is an absolutely integrable function
{on the whole x-axis) piecewise smoot h on every finite interval of the
x-axis, then (1) the Fourier transform determined by formula (11.229)
exists, and (2) we have inversion formula (11.230) which should be
 CH. 11. FOURIER SERIES AND FOURIER INTEGRAL 547

understood as the limiting relation

/(x ) = lim eikx dk

V2n f-H-OO- it T W

Note. According to the note after Theorem 11.11, we can assert

that Theorem 11.12 remains valid for every function / (x) which
is absolutely integrable over the entire x-axis and piecewise con­
tinuous on every finite interval of the x-axis provided the expression
/<* + D -/< * + 0) is bounded for every fixed x and all sufficien­
tly small | £ | ^ 0. Fourier transforms of functions defined for
—co < x < -t-oo are widely applied to various problems of mathe­
matics and mathematical physics (see Appendix 4 to Chapter It).
For functions defined over the semi-infinite interval 0 ^ x <C+ °°
we also use the so-called Fourier sine transform and Fourier cosine
transform. Let us dwell in more detail on these notions. Applying
the formula
cos k (! — x) = cos X£ cos kx sin XI sin kx
to the integrand in formula (11.200) we deduce
-l-oo -J-oc
/ (x) —-i- f dk j / (|) cos X! cos kx d| -f-
0 — oc
+00 -4* co
J dk ^ / (|) sin X!sin kx (11.231)
0 —oo
where, by the absolute integrabilily of / (x) over the whole x-axis,
both integrals are convergent. If / (!) is an even function the product
f (I) sin X! is an odd function while the product / (!) cos XI is an
even function. Therefore the second term on the right-hand side
of (11.231) turns into zero and we thus obtain
"4“oo -{-oo
/(x) = j cos kx dk J / (!) cos X! d! (11.232)

Similarly, if / (x) is an odd function wc find that

-Lq
o | oo

/ (x) - 4 f sinXxdX ^ /(!)sinX |rf£ (11.233)

If x is a point of discontinuity of the function / (x) the left-hand

sides of equalities (11.231), (11.232) and (11.233) should he replaced
by the expression ——:— J~y ------ .
35*
548 MULTIPLE INTEGRALS, FIELD TJIEOHY AND SEMES

Now suppose that a function /(x ) is defined only on the interval

0 ^ x < 4 - 0 0 . Then it can bo extended to the negative half of
x-axis in such a way that we obtain an even or an odd function
defined for all x, —oo < x <z -.-oo. If / (x) is extended in even
fashion we obtain the representation
___ -r 30 -f-OO

/ (x) = \ / — [ cos > . x d k [ y — ij /(?)cos/.£d?) (11.234)

6 6
i
and if it is extended in odd fashion \vc get another representation:
__ -f-JO 4-00
/(* ) = } / - 1 ^ Pin },xd \ ( j / C /(E)sin XEd l \ (11.235)
o o
If the function / (x) defined on the semi-axis 0 ^ x <C + ° ° is
continuous at the point x = 0 then, after it has been extended as
an even function to the entire x-axis, it remains continuous at the
point x = 0, and therefore relation (11.234) will also hold for x = 0
in this case. On the contrary, for the function /(x) extended in odd
fashion relation (11.235) does not hold for x — 0 in the general
case even if the original function is continuous at the point x = 0.
It is clear that this relation may only hold for x — 0 if / (0) — 0
since we have the equality ^( j ^(~l°j — Q for every odd
function.
Equality (11.234) ran he rewritten in another form. Puttincr

/ — +a>
U (? ) = | 7 4 f / (?) m s « rfg (11.23(1)
6
we derive from (11.234) the formula
H*>
f(x) = y | fc (>■) cos Xxdl (11.237)
0
which is equivalent to (11.234). The function fc (>.) is called the
Fourier cosine transform of the function / (x) defined on the semi­
axis 0 ^ i < l-oo. Accordingly, the transformation from f (x) to
fc {}*) performed by formula (11.236) is termed the Fourier cosine
transformation. The Fourier inverse cosine transformation described
by formula (11.237) yields the expression of / (x) (which is Fourier’s
inverse cosine transform of fc (X)) in terms of the function fc(X).
We see that transformations (11.236) and (11.237) are inverse
to each other.
 CH. 11. FOURIER SERIES AND FOURIER INTEGRAL

Similarly, instead of relation (11.235) we can write the formulas

(11.238)
and
(11.239)
o
which are equivalent to (11.235). The function fs (h) is known a*
the Fourier sine transform of the function / (x) defined on the semi­
infinite interval 0 ^ x < j - o o , and transformation (11.238) frorr
/ (x) to fs (/.) is called the Fourier sine transformation. Formula
(11.239) describes Fourier's inverse sine transformation which expres­
ses the original function / (a;) (Fourier’s inverse sine transform o!
/* (X)) in terms of f s (a).
Examples
1. Consider the function

Its Fourier cosine transform is the function

Applying formula (11.237) wo obtain

1 for 0 < x a

— for x —a
0 for x Z> a
2. For the function / (x) = e~ax, a > 0, x ^ 0, we find, per
forming integration by parts in formulas (11.236) and (11.238)
the expressions

%
n
and
550 M U L T IP L E IN T E G R A L S , F IE L D TH EO RY AND S E R IE S

Accordingly, applying formulas (11.237) and (11.239) to the above

equalities we obtain
+ 80
2a f <‘0s A. x
Jl J a 1 — a* dX - <raac, x> 0
o
and
+oo
}. sin >.x
dX = e- a x » x > 0
0
We see that Fourier’s cosine and sine transformations enable us
to find the values of some integrals dependent on a parameter.
But the main application of these transformations lies in using
them for solving various problems of mathematical physics (see
Appendix 3 to Chapter 11).
6. Fourier Integral for Functions of Several Independent Variables.
We shall begin with the case of two independent variables. Let
a function / (xj, x z) be defined for — oo < i | < - f- o o t — oo < x2 <
-< -f-oo. We shall suppose that f fo , x 2) is absolutely integrable
with respect to each variable xt and x 2 from —oo to + oo for every
fixed value of the other variable. If, in addition, the function / (xf, x 2)
is continuous and piecewise smooth with respect to X \ (x2) for every
fixed value of x 2 (xt) we can apply Fourier’s integral formula to each
variable for every fixed value of the other. Fixing an arbitrary
value of x 2 and applying Fourier’s formula (11.206) (for the variable
X|) we obtain
—|-OO Iuu
/ (X |, = | j dX, j / (£„ x2) cos X, ( X i — El) d | I (11.240)
0 —
oo

Similarly, we can fix x, — Ei and apply Fourier’s formula (11.206)

to the variable x 2 which results in
- |- D O - f«

f {11, * 2 ) = 4" f dKz j / Si» &) cos d^ 1 *241)

Substituting (11.241) into (11.240) we derive the formula

—
J—
00 -|-00 +«>+°°
f {xi, x 2) = -j — j dXi j cos Xt ( X i — li) dlt j dX2 j / (61 * £2) X
0
-I-00 +00 +00 + 00

X cos (jrg — £2) ^ 2 ~ ^^ dE\ ^ dhq J / (£i* ^2 ) ^

• 0 —00 0 —00

X cos X, (xj — It) cos X2 (x2—E2) 0 * -242)

 CH. II. FOURIER S E R IE S AND FOURIER INTEGRAL 551

If f (x«, x2) is an even function of each argument xi and x2 formu­

la (11.242) turns into
-^•oo

/ (xlt x2) = j cos X{Xi dXJ j cos A^ dll X

0 o
-f- 00 ( OO

X f cos ?.2x2dXj | /(?„ (11.243)

0 6
Similarly, if / (xt, x 2) is an odd function of each argument X| and x2
we obtain
-j-OO 4"00
/ ( x i , x 2) — ^ sinA^dAt ^ sm A i^ c ^ x
0 0
-p VJ -)-W
sinA^dA^ j /i(£i, £2) sin A2£2d£2 (11.244)

Passing to the complex form of Fourier’s integral we can rewrite

formula (11.242)in the form
/ (*n *2 ) =
-j-OO -j-0 0 -|~ao 4 *-00

= p - r j dX. j d \ t j dk, j / ( |„ yeUW-.-E.H-X.tx.- toldg, (11.245)

— 00 — 00 — 00 —00

where the integrals taken with respect to Art and X2 should be under­
stood, in the general case, in the sense of Cauchy’s principal value
(see § 3 of Chapter 9 and § 9, Sec. 4 of the present chapter). If it
is allowable to reverse the order of integration with respect to £1
and A2, formula (11.245) turns to be equivalent to the following two
formulas:
-}- cpo +00

T a 1 , ^> = -^r i d l ' j I*) e-^it'+W d dg, (11.246)

— 00 — 00

and
+O O + 0O

/<*., * * )= -s r j j 7(*i, **) s‘(A.x.+^.l dX, (11.247)

— OO — CO

Formula (11.246) expresses the Fourier transformation from / (xlf x2)

to / (A,, A2) and formula (11.247) describes Fourier’s inverse transfor­
mation. Accordingly, / (Aj, A2) is the (two-dimensional) Fourier
transform of / (xj, x2), and / (xi, x2) is the Fourier inverse trans­
form of f fAi, A*»).
 f> 5 2 MULTIPLE IN T E G R A L S , FIELD THEORY AND S E R I E S

In thejmse of three or more independent variables the Fourier

transformations are constructed in a similar manner. Here we shall
give the corresponding formulas for functions of three independent
variables. The Fourier integral formula is written as
-J-OO -}-<» -J-/X) -J-OO -J-OQ -f-OO

t ( x it ^*2 » ^ 3 ) ~ “^ 3 " ^ ^ ^ ^ dA.3 j } (^J, £3) X

0 —0 0 0 — 00 0 — 00

X cos /.! (x, — ct) cos /v2 (x2— c2) cos (x3— (11.248)
or
-(“ DO -}~DO •
; OO -f- 30 -|-oO-j-OO

/( * i. x2, g a )= j d>-i j j ^ 2 j d\« j dhj j / (|i, | 2, Is) X

0 — DO () — OO — OO — OO

^ £t[M(xi-|i) fXa(x2-£s)+k3(X3—Jiall ^ £ 3 (11.249)

in complex form.
If the conditions which guarantee the possibility of reversing
the order of integration are fulfilled, formula (11.249) is equivalent
to the following two formulas:
_ +30 -|-ao +00

/(>.,. h . X-i) = - ( y = j r j di, j d |2 j /(£,. Sj, y x

x d |3 (11.250)
and
1>
5. | x

i (xu x2, x:f) = ( y .^ \3 j dh, j dl^ j 7 (^ , ^2 , a3) X

~ -lO —OO —no
X eil*-i*i+*-2*H-A.3*3l c?A,3 (11.251)
In the general case the integrals on the right-hand sides of (11.247)
and (11.251) are understood in the sense of (Cauchy's principal value.
We now briefly discuss the justification of formulas (11.240)
and (11.247). Formulas (11.250) and (11.251) are proved in a similar
way.
T h e o r e m 11.IS. Let a function f (x t, x z) be continuous throughout
the x ,, xz~plane. Suppose that the following conditions are fulfilled:
(1) 77te integrals
~j-00 -J-OO
j I / Ui, xz) \d x x and j | / (xl5 xJIcte, (11.252)

are uniformly convergent with respect to x z and x, on every finite interval

x 2 ^ x2 ^ x 2 and x, x, X|.
 U I. 11. FOURIER SERIES AND FOURIER INTEGRAL DO

(2) The iterated integral

*4-00 -{-X>
j dx2 j X2) \d x t (11.253
—oo ~ oo

is convergent.
(3) For alt sufficiently small | £ ] ^ 0 the inequality
I Xt>~-ii£1~ u- *•> | < C, = const

holds for every fixed Xi and all x 2.

(4) For all sufficiently small J £ | 0 the relation
/ ( J i» J it "p b ) ~ ~ / ( ^ l ~f~0) i ^
s I
is fulfilled for every fixed x z and all xt where C2 (^i) is a function sucf
4-00
that the integral j C2 fo) dxt converges.
—OQ
Then the two-dimensional Fourier transform
—
j-oo -|-oo

7 (> ■ „ > * ) = 4 rj J /®«. (11.254;

of the function f (xu x2) exists, arad //ie F o u r i e r
i n r c r s i on f o r m u Ia

/(x „ X2) = - ^ j‘ dx, j 7(1„ a.) e'C ^ '+ ^ l dXj ,(11.255)

—O
O —oo

which is understood in the sense of the limiting relation

f.l I** _
/(Xu *2) = -£T lim I d ^ f 7(X„ k 2) eU^*.+x2x2] dk2 (i j .250)
-*i -h
where we first pass to the limit for l2 - *■ -j^oo and then for lt —oo.
/V 0 0 /. The one-dimensional Fourier transform (with respect to the
argument xt)
-f-OO
7(>.„ X.) = —1=. j / ( |„ r2) di, (11.257)
—oo

exists since the first integral (11.252) is convergent. The uniform

convergence of this integral implies that the integral on the right-
hand side of (11.257) is uniformly convergent and therefore the
function /(/.,. x?) is ront.i minus in r;. By tli^ hyp^thcri?, into-
554 M U LTIPLE IN TEG RA LS, FIELD THEORY AND SER IES

gral (11.253) converges and therefore the integral

+ oo -f-O© ao
1
J |7(?m, xz)[dx2 = j d i t \ j /( ! „ *2) d l
1 ~[/Zn
(11.258)

also converges.
Taking into account conditions (1) and (3), the continuity of
/ (xi, as 0 function of x x and the note after Theorem 11.10 (see
Sec. 4 of § 6), we see that the inversion formula

f ( x lt x2) = - J = = j 7(X„ x2) eiXlXl dXx=

— 00
fj
= —)=■ lim f J(k„ Xz)e*<*<dX, (11.259)
~y 2 IX Ij_>_^_oc J
is valid here. By condition (4) and equality (11.257), we can write
I / (^i» xz H"C) — / (^ 1 » x2 4" 0) |
-J-oo +°°
« j !/(?.. * i + 0 - / ( l i . i , + 0 ) | r f | , < | 5 | j C2 (i,)d5,
— oo —oc

that is
/(*■>» Ti 4- £) — / (kj« -r 2 ~ t ~ 0 )
^ ^2 (il) ^£l (11.260)

Since integral (11.258) is convergent the two-dimensional Fourier

transform
_ + °°
7(?.„ >„) = — = - ) 7(?-!. y « - « * (i51=
—no
-j-OO - |- o o

= W j J /(I .. fc) «-UX.S.+M»J d£, (11.261)

— OO — oo

of the function / (xit x2) exists, and the inversion formula

~|~oo

7(*l, Xz) = -y 1— f 7(?U, X2)^®**d^2 =

“ OO
_
——7=- lim f j Vki, A,2) (11.262)
y2it is-j-oo J
—«2
is valid because the function /(X ,t a:2) is continuous in x 2 and condi­
tion (11.260) is fulfilled. Substituting (11.262) into (11.259) we
 CH. II. FOURIER SERIES AND FOURIER INTEGRAL ooa

obtain
'1
/ (&u X2) = lim f e®**' dk{ { li'u f / (Xj, X2) «*•*»
“ -1 li-+ J
1 • -It
-'-X >
1 -i.
(11.263)
or, which is the same,
<1 <2
/(*„ x 2) = J - lim f dXt f / (>*, e*fx*xl+,-*T*l dX2 (11.264)
r.. 1^ + - _J , t j.
where the passage to the limit is first performed with respect to
12 and then to lx. The theorem has thus been proved.
In the case of three independent variables the formulas for the
three-dimensional Fourier transform and the corresponding inverse
transform are written in the form
-f-M-pgo-|-g
/ (>.». k.) = j j J /(& .& . w x
* —OO—00 —00
x e-i txiii+*»5.-Hjfa) d.%, ci&<%3 (H.265)
and
-|-00 -pQQ -J-oo _

f { xu x 2, X3) = ( y l ^ ) a ' 1 1 1 /(^l* **)X

Y ~ — 00 — 00 — 00

X ei [Xi*i+X***+X3*3] dX2 dk2 (11.266)

Substituting (11.265) into (11.266) we arrive at the relation
j CO J- OO ^ 30 —
^ OO —
f-OO J OO

f{x it X2y Xa) = - ^ 3- J j j j j j f i t t» ^2’ ^3 ) X

— OO — OO — OO — OO — OO — OO

x *« IM(X1-ID+X2 (*a-g2)+X3 C**-E»)1 di, d|a dkt dkz dk2 (11.267)

In the general case oi N (N > 3) independent variables the corres­
ponding formulas for Fourier’s transforms can be easily written
down by analogy with the above formulas.
To justify equalities (11.265) and (11.266) we should introduce
the condition that the corresponding integrals (analogous to those
entering into the formulation of the foregoing theorem) are conver­
gent and, in addition, impose the requirement that for all sufficiently
small | £| ^ 0 the following inequalities hold:
*2» x9) — f ( z 1 + 0 * X 2, x3)
(1) 1 = const
t
fur every fixed x^ and all x 2 and x 3,
/ (Tj, *a)—/(*!* *2-rO» Ja) < C 2 (x1)
(2) y
55G MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

for every fixed x 2 and all x t and x 3.

/ (*[. s-z, — / ( j |, gg, *r3~r 0)
(3) v C3 (*1» *2)
s
for every fixed x 3 and all xt and x 2 where C2 (xi) and 6 ’ 3 (xlt x2)
+O C
are some functions such that the integrals I C2 (xj) dxt and
+°° +«°
^ c/x2 ^ C 3 (xj, x2) dxt are convergent. Then, under these res-
__0 0 — 00
trictions, the three-dimensional Fourier transform of the function
/ ( X | , x 2, x3) determined
by formula (11.265) exists and equality
(11.266) holds, tlie latter being understood in the sense of the limi­
ting relation
it it
/(X i, x2, X3) = — —1iin I d ll \ lira \ d l 2 'A
2

*3 _
x | Imi f /(/.„ ?.2, X3) rf>.3j } (11.268)

where the passage to the limit is first performed with respect to

/ 3? then to l 2 and finally with respect to
The general case of A independent variables (A > 3) is treated
sim ilarly.

APPENDIX 1 TO CHAPTER 11

ON LEGENDRE’S POLYNOMIALS

Here we shall prove Ihat Legendre’s polynomials

P o (* )^ i, «= 1 .2 — (i)
are orthogonal on the interval I—1 , 11, i.e.
1
j Pn (x) Pm (x) dx = 0 for n (2)
-1

Since m and n are involved equivalently in (2) it is sufficient 1 o prove

that relation (2) holds for tn «< n. For this purpose we must only show
that
1

^ P n (x)xmdx s = 0 for tn<^n (3)

 CH. U. FOURIER SERIES AND FOURIER INTEGRAL 55?

where m is a nonnegative integer. Hutting

w = i * s- i ] n
we can write

\ P„ (x) x '" ^ = 3 ^ r \ z m <Ix (4 )

-1 -1
Integrating by parts in (4) m -f- i times and taking into account that
U„ ( ± 1) -= Un ( t 1 )= 0
we arrive at equality (3) and thus the orthogonality of Legendre’s
polynomials on the interval I—1, 1] has been established.
Now let us compute the norm of the nth polynomial Pn (x). For this
purpose we integrate by parts in the integral

II/ ’.<*>I P - 2^ r 1 <5>

Integrating by parts n times and taking into account that un (x)

is a polynomial of degree 2n and un (± 1 ) = nn(iiil) = . . .
. . . = nj1” 1) ( i l ) — 11 we obtain
l i
( t ) d n ^ u n (x) ^ _

-1 -1
-( n2 f (x)dn+2«n (tI j___
11 J dx'"- UX — . . . —
-1

= ( - l) U j «n (*) — dX
-l
I
= (2n)\ ( (1—.r)"(l \ - x y d z (IS)
-1
But we have
i
j ( l - x ) n (! ; x)nd i = -^-j- ( (1 - x )"-1 (1 -| x)’" l rf.r= . . . =
-1 -1

. i-l)
(« tVw(n A-Z) ... (2u) Jf (I
v - x)*n
' dx
-I
(«02 1
(2«)! (2n 1) (7)
558 MULTIPLE IN T E G R A L S , FIELD THEORY AND S E R I E S

and therefore, substituting (6) and (7) into (5), wc get the for­
mula
i
II M ir = j V>« <*)ll dx = 5 - ^ (8)
-1
Consequently, the norm of the /?th polynomial Pn (x) is ex­
pressed as
II P n (*) || = | (9)
It should be noted that the nth polynomial P n (x) is of degree n for
n = 0, 1, . . . . Legendre's polynomials P 0 (x), P x (x), . . ., P n (x)
are orthogonal on the interval I 1, II and hence they are linearly
independent. Consequently, the system of Legendre’s polynomials
is a basis of the space of all algebraic polynomials of degree not
greater than n. It follows that every polynomial of degree not greater
than n can bo represented in the form of a linear combination of
Legendre’s polynomials P 0 (x), P t (x), . . ., P n (x). In particular,
we have
*n = « 0 n^ 0 (x) + Ctm^I (*) + - • • + CtnaPn (x)
(see Appendix 2 to Chapter 11).

APPENDIX 2 TO CHAPTER 11

ORTHOGONALITY WITH WEIGHT FUNCTION

AND ORTHOGONALIZATION PROCESS

The concept of orthogonality with weight function is a generaliza-

tion of the concept of orthogonality of functions in the sense of
relation (ll.(iS).
Let p (x) be a nonnegative function which is not identically equal
to zero. We shall suppose that this function is continuous in an
open interval («, b) and that the integral
b
\ p (x) dx (1)
v
a
exists (as proper or improper integral) and is positive.* The func­
tion p (x) will be referred to as a weight function.

* This integral may turn out to be improper under these conditions if the
function p (x) is unbounded for x u -•!- 0 or x -> t> 0. We encounter sin­
gularities of p (.r) of this type when studying some important classes of special
functions fe.g. snob a weight function is used at the end of this appendix when
we consider Chebyshev’s polynomials).
 CII. 11. FOURIER SERIES AND FOURIER INTEGRAL

Lot / (x) be a function defined on (a, 61 such that the integrals

b b
jp (x )/(x )r fx and J p (x) [f (x)p dx (2)
Cl fl
exist (as proper or improper integrals). Then the function / (x) is
said to be square-integrable with weight function p (x) on the interval
fa, 6). In particular, if p (x) == 1 we come back to the ordinary
definition of a square-integrable function gi ven in Sec. 3 of Chapter 11
(see relation (11.93)).
Let
<fi ( * ) » <P2 ( * ) .................<fn ( * ) » • • • (3 )

be a system of functions defined on (a, 61 which are square-integrable

with weight function p (x) on [a, 6), that is the integrals
6 i>
\ p (x) (pr, (x) dx and I p (x) [cpn (x)l2dx, « = l t 2. . . . (4)
a a

(understood as proper or improper integrals) exist. If the interval

[a, 6j is finite the existence of integrals (2) and (4) and the obvious
inequalities
|/W<p»WKjl/’W+fiWl
and (5)
I<Pn (x) (x) | < (cp£ (x) -4 cp*, (x)l
imply that the integrals
b i.
j p (x) / (x) (pn (x) dx and \ p (x) tpn (x) <pTO(x) dx (6)
a a

also exist. Let us agree that if the interval |a, 6] is infinite we additio­
nally impose the requirement that integrals (G) exist.
In what follows we shall suppose that every function wc deal
with is continuous everywhere on |a, 6] except possibly at a finite
number of points which, in particular, may be singular points of the
functions.
We sap that two functions (pa (x) and cpm (x) are o r t h o g o n a l on
the i n t e r v a l Ia. 6| w ith -weight f u n c t i o n p (x) if
b
[ p (x) (x) <fm(x) dx = 0 for m n (7)
n
System (2) of functions square-integrable with weight function p (x)
on the interval (a, 6| is said to be o r t h o g o n a l on. la, 6| irit/i
5(50 MULTIPLE INTEGRALS. l'TELD T H B O H Y AND S E R I E S

w i y h t f u n c t i o n p (x) if
o
j p (X) <pn (x) <pm (x) dx = 0 for n ^ rn (8)
a
and
J p (x) ffn (x) dx 0 for 1 , 2 , . . . (9)
a

hi the case p (x) s 1 Ilie definition of orthogonality with weight

lunclion turns into the definition of (ordinary) orthogonality given
in § 3, Sec. 1 of Chapter 11.
Let / (x) be a function square-inlegrable on la, 61 with weight
function p (x) and let system (3) he orthogonal on (a, 61 with weight
function p (x).
A scries of the form
Cicpi (x) -f c2q>2 {x) -f . . . + cnq„ {x) -f . . - (A)
with coefficients cn, n = 1, 2, . . determined by the formulas
u
j P ( 0 / <*) q-n (x) dx
= — I-------------------- (B)
5P U ) <tn (*)<*-t
a

is referred to as the F o u r i e r s c r i e s o f th e f u n c t i o n f (x)

with r e s p e c t to s y s t e m (3), and we write
f (^) /'s’' £lfPl (*^) “h • • • i“ (■/iTn (^0 “I' • * *
We say that series (A) i s c o n v e r g e n t i n th e m e a n on [a, 61
tcith w e i g h t / u n c t i o n p (x) to th e f u n c t i o n f (x) if
b rn

lim ( p(x) I /( x ) — 2 ch(ph(x) \2 dx = 0 (C)

7>l-» 1.00 J ( '
1 a h= 1

If scries (A) converges uniformly or in the mean (with weight

function p (x)) to the function / (x) on the interval la, 61 its coef­
ficients are uniquely specified by formulas (B). 1ti fact, under the
given conditions, the Cauchy-Bunyakovsky inequality implies that
b to h
[ p ( l ) i p n (*) | / (*)
a
— 2
h— 1
Ch‘, h j< ( \ P M rPn ( x) d x ) 1U X
a
b t>i

X ( j p (x) [ f ( x ) — y \ ct,q>A ( x ) j^r/x) l/“ -^0

a h 1
 CII. 11. FOURIER S E R IE S AND FO UR IE R I NT E GR AL 50 i

for m —*■ r oo and any fixed n. Oil the other hand, the functions
i — 1, 2, being orthogonal on the interval [o, 6] with
weight function p (x), we have the relation
b 7/1 b

j P (*) <rM(■*') [ / (*) — (*) j dx = j p (x) / (x) <pa (x) dx—

a Jt = 1 c

/>
— cn j /; (x) <1 (x) dx =■-: const
a

where tn ^ i i and n is iixed. Consequently,

b b
j p (x) f (x) q>„ (x) rfx— cn j p (x) q * (x) dx = 0
a a

which implies (IS).

The definitions ol complete and closed systems (see § b of Cliap-
lor It) and also the basic theorems related to these notions (see
Theorems 11.4-11.7 in §§ 5, (i of Chapter 11) are easily generalized
to the case of systems orthogonal with weight function.
In mathematical physics expansions of functions into series with
respect to systems of functions orthogonal with a weight function
are widely applied to various problems. Among the most important
systems of functions orthogonal with a weight function we can
mention various systems of special polynomials (which will be
d is c u s s e d at t h e en d of lhi< appendix) and also the systems of eigen­
functions used in studying the problems of vibration of a circular or
a ring-shaped membrane, the systems of eigenfunctions for a sphere
and for a spherical layer etc. (see 117)).
Systems ol functions orthogonal with weight function (and. in
particular, with weight function p (x) — I) are especially convenient
for expanding functions into series because the corresponding coef­
ficients of such expansions are easily found.
An orthogonal (with weight function) system of functions can he
constructed by applying the so-called orthogonalization process
to a given system of linearly' independent functions.
We say that functions* «pt (x), q;2 (z), • • <|„ (x) are N n e a r l y
d e p e n d e n t on (a, b\ if there are constants C\. C 2 , . . Cn, not all
zero, such that the Linear combination of the functions q 5 (x), q>2 (x), . . .
. . ., q n (x) with the coefficients Cu C2, • • ., Cn is identically equal
to zero on (a, b\ except possibly at the points of discontinuity of the

* Wc remind the reader that all the functions under consideration are
supposed to he continuous everywhere on (a, 5| except possibly at a finite
number of points.
3U~0ts24
 502 MtJl.TI I'l.E IM K G U A I.S. FIELD THEORY AND S E R I E S

junctions <| 4 (x), <f2 (x:), . . qn (x):

^i<fi (*) t C2«fo (x) -r - . . — C„(pn (x) == U (10)
It identity (10) (understood in the above sense) implies (hat all the
coefficients 6’,. C \...........Cn of this linear combination are equal to zero
the functions q , (x), cj 2 (x), . . q,t (x) are said to be f i n e a r l i /
h t f h ' p r n . f l a n t on f a , 6 |.
If a system of functions if, (x). fp2 (x:), • * <f„ (x) is orthogonal
on the interval | a , 6) with weight function p (x) the functions of this
system are I i n e a r l t / in*tejten*leut on la, 6 ]. Indeed, suppose that
^VfY(x) -f C\n 2 (x) -r • • • 4 O j r« U) 0 on [a, 61 (1 1 )
everywhere except possibly at a finite mirnber of points. Multiplying
equality ( 1 1 ) by p (x) q m (x), m -- 1 , 2 , . . integrating with
respect to x from a to b and taking into account that the functions
(t 1 (x), . . <|'„ (x) arc orthogonal on (a, b\ with weight function
p (x), we obtain

C,„ f p (x) fp?,, (x) dx =■■0, r n = 1 , 2 , . . . (12)

•I
a
The integral in (12) being different from zero, we see that C,n — 0 ,
m — 1 , 2 , . . . . Thus we conclude that C\ — Cz «= . * . = Cn ~ 0 .
If we are given a system of linearly independent functions if, (x),
»p2 (x). - • »ln (x) we can easily construct an orthogonal system
(with an arbitrary weight function p (x)) q 1 (x), q 2 (x)i • * •» (tn fa)
such that
Tifa) -ip, fa)
<p2 (x) - »p2 (x) -! A2 ,if, (x)
M'3 (^ ) —: *l:» (•i ) r ^'3i*t 1 fa ) i" ^'32*1 2 (^ ) ( l -t)(x)

(x) — \pfl (x) Amif 1 (x) /v*2'|2 (x) -r . . . -b (x)

Before proceeding to prove this assertion we note that relations (13)
(provided they are valid) imply that every function q>ft (x), k
•= I, 2, . . . , / « is a linear combination of the functions i| , (x), . . .
• • •» 'Pk fa) in which the coefficient in ^ (x) is equal to unity and.
consequently, q>fc fa), k -- 1, 2, . . «, may vanish only at the
points of discontinuity of the functions ip, (x), . . ip,. (x). In fact,
if otherwise, the functions if, (x), . . ip,, (x) would he linearly
dependent on la. hi. which contradicts the hypothesis. Thus, every
f u n c t i o n q h ( x ) . k - 1 . 2 ................... « . is d i f f e r e n t from zero at each
of its p o in ts of continuity, and hence
>1
^ p (x) qo. (.r) dx > 0. k -= 1, 2. . . ., n (14)
 CH. 11 . FOURIER SERIES AND FOURIETt INTEGRAL 503

It should also be noted that it follows from relations (13) that

every function ij,( (x). A; = 1. 2, . . . . n. is a linear combination
of tlio functions <pt (x), cp2 (^)- • • •» (| ft (x) 'v^th the coefficient
in <pfc (a:) equal to unity.
To justify formulas (13) (i.e. to prove the possibility of construc­
ting the desired system {(ph (a:)} orthogonal on [a, 61 with weight
function p (a:)) we must show that the numbers entering into (13)
ran in fact be found. This can easily he shown by induction. Moreo­
ver, it turns out that formulas (13) uniqutdy specify Ihe quantities X/;.
Indeed, multiplying both sides of the second equality (13) by
p (x) cf, (a*) and integrating with respect to x from a to b wc derive
b *>
\ p (x) <| , (x) <p2 (x) d x = \ p (x) cp, (x) <p2 (x) dx —
• 4
n a
h
-j- ?*2i ^ P ( x ) fFi fa) d x - 0
fi

Consequently, by the hypothesis that the functions <p# (x), i =

= 1 , 2 . . . . » n, are orthogonal on (a, b\ with weight function p (x),
we obtain
b
y p (x) <fi (r) (\z (j ) dx

J /*(■') 7 i (r) (lx

a

and thus IIn* rooflicient /.21 has been determined (and is uniquely
specified by (13)). Now suppose that the coefficients Xij, i =
= 1, 2, . k 1, / -- 1, 2, . . . . i — 1, have already been
computed and that the functions <| t (a). . . ., (x) thus obtained
are pairwise orthogonal on (a, 61 with weight function p (x). Then the
conditions of orthogonality
>> »
\ P fa) <1u fa) Mj fa) dx - I p (x) i|> (x) <p, (x) dx —
fi a

k- 1 ^
! 2 /'hi j P fa) fl i fa) ‘I j fa;) dx - ~
i--\ a
i-• h
r
— \ p (•') 'I ' ( ') <1 I (./■) (I r / \ p (x) rpj (a ) dx — 0,
% ' 4
•: n
/= 1, 2 , . k 1
5G4 MULTIPLE INTEGRALS, FIELD THEORY AXI) S E R I E S

yield
h
\ P (*) (*) <1j (*) djr
/.Itj — ------------- j;------------------------------------------ , / — 1 , 2 , . . . , fi — 1

J P (x) <Pj {*) d-c

a
This determines the function <| j, (j:) = ). (x) — (x) -t- . .
. . . -f ai»tl the system of functions r.f, (.r), . . .
. . ., <P/» (<£) is orthogonal on la. b\ with weight function p (x). Thus,
the assertion has been proved.
The above process of constructing an orthogonal system ij-i (x), . . .
. . <pn (x) from a given system tyi (x), . . il>n (x) of linearly inde­
pendent functions performed according to formulas (13) is referred
to as the orlhogonaiization process. In just the same way this process
cun he applied to infinite systems of functions.
Let us consider the system ol functions
1, x, x2, . . xn, . . . (15)
consisting of integral nonnogative powers of the argument x. As is
known, these functions are linearly independent. Applying to (15)
the orlhogonaiization process for the interval I 1. 1) with the
weight function p (x) 1 we obtain a system of polynomials which
are orthogonal on [— 1, 1J. We do not write down Lhe formulas for
these polynomials because they differ only in constant factors from
Legendre’s polynomials determined by the formulas
/•’ll (-f) =2^7T(-^r, •)"!. " = 1 , 2 , . . . ; /'«(*) I (16)
Relations (1G) are known as Rodrigues’* formulas.
Taking the weight function p (x) — —jzL-=^ and applying lhe
y1 j-
orlhogonaiization process to the same system of powers (15) on the
interval [—1, 11 we arrive at the system of Chebyshev s polynomials
of Uie first kind, ff we take the weight function p (x) = | 1 — x*
we obtain (.hebyshev’s polynomials of the second kind.
Considering system (15) on the semi-infinite interval 10, -f-oo)
and performing the orlhogonaiization process with the weight
function p (x) — e~x vve obtain the .system of the so-called Chebyshev-
Laguerrc** polynomials. If we take the weight function /> (x) =
= x V - \ s > — 1, the orlhogonaiization process (on the same inter­
val 10, -|-oo)) results in the generalized Chobyshev-Laguerrc poly­
nomials.

* Uodriguos, Olinde (1704 1851), a French mathematician.

** I.afnmrro. Edmond Nicolas (1834 188G), a French mathematician.
 CH. II. FOURIER SERIES AND FOURIER INTEGRAL 5U5

F i n a l l y , t a k i n g , f o r t h e s a m e s y s t e m (15), t h e i n t e r v a l — oc <
< 5 < -roo a n d t h e w e i g h t f u n c t i o n p (x) = e ~ x2 w e o b t a i n t h e
s y s t e m of t h e C h e b y s h e v - l l e r m i t e * p o l y n o m i a l s .
There are convenient general formulas (sim ilar to R odrigues’
f o r m u l a s (15)) for a l l s p e c i a l p o l y n o m i a l s .
T h e s y s t e m s of o r t h o g o n a l p o l y n o m i a l s e n u m e r a t e d a b o v e ar e
w i d e l y a p p l i e d t o v a r i o u s p r o b l e m s of m a t h e m a t i c a l p h y s i c s
( e . g. see 1171).

APPENDIX 3 TO CHAPTER 11

FI XCTIONAL SPACE AND GEOMETRIC ANALOGY

T h e s e t of f u n c t i o n s Q [a, frl d e f i n e d in § G c a n b e r e g a r d e d as
a f u n c tio n a l space w h o se e l e m e n t s a r e f u n c tio n s . T w o function?
if (x) a n d i|: (x) b e l o n g i n g t o <J | « . b] a r e c o n s i d e r e d a s r e p r e s e n t i n g
t h e s a m e e l e m e n t ('‘v e c t o r ") o f t h i s s p a c e i f t h e y d i f f e r a t no mo r i
t h a n a finite n u m b e r of p o i n t s of t h e i n t e r v a l (a, b]. I n w h a t f o l l o w s
we s h a l l d e n o t e t h e e l e m e n t s of t h e s p a c e (J [a, b\ c o r r e s p o n d i n g
t o t h e f u n c t i o n s if (x), »| (x), q (x), . . . b y <p, q, . . . o m i t t i n f
t h e a r g u m e n t x.
T h e s u m if — \[- of t w o e l e m e n t s ip a n d a n d t h e p r o d u c t A<p o
a n e l e m e n t ij b y a n u m b e r X a r e d e f i n e d as t h e e l e m e n t s represent lei
b y t h e s u m ip (x) \|- (x) a n d b y t h e p r o d u c t Xcp (x). T h e n t h e spac«
(J ( a, b] w i t h t h e o p e r a t i o n s of a d d i t i o n a n d m u l t i p l i c a t i o n b y a mi n i
h e r is a n a l o g o u s t o t h e E u c l i d e a n s p a c e of a l l t h r e e - d i m e n s i o n a
v e c t o r s w i t h o r d i n a r y o p e r a t i o n s of a d d i t i o n of v e c t o r s a n d m u l t i
p l i c a t i o n of v e c t o r s b y s c a l a r s . T h e z e r o e l e m e n t 0 of t h e s p e c
(J ( a, 6| is r e p r e s e n t e d b y a n y f u n c t i o n w h i c h is i d e n t i c a l l y e q u a
t o z e r o on t h e i n t e r v a l (a, b 1 e x c e p t p o s s i b l y a t a f i n i t e n u m b e r o
points.
W e n o w d e f i n e t h e s c a l a r p r o d u c t of t wo e l e m e n t s <|> a n d \\>b e l o n g i n j
to (J I a, 6| b y p u t t i n g

(T> 'l‘) = j <P(*H*(*M* (1

a

O n e can easily v e r if y t h a t the scala r p r o d u c t t h u s defined satisfie

t h e o r d i n a r y c o n d i t i o n s f or t h e s c a l a r p r o d u c t of g e o m e t r i c vec t or ?
namely:
(1) (q>, i|') = (if, <|),
(2) (Xip,iJ:) — X (<p, \j) w h e r e X is an a r b i t r a r y r eal n u m b e r .
(3) (f| , ij i T if 2 ) = W* Mt) (<P, ^‘2 ),
(\) (<f . i|) ^ G, a n d ( i j , i | ) = 0 if a n d o n l y if q = 0. W h e n pr<

* Herniite, Charles {IS22-1001). a French mathematician.

 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

ving Theorem H.(» in § (i, Sec. 3. we established the validity of

condition (“i), and conditions (1)-(3) are evident.
Thus, we see that the space fix. (A with the scalar product dclined
by equality (1) is closely analogous to the Euclidean space of all
three-dimensional vectors whose scalar product is deli nod in the
ordinary way.
The “vectors” <|> and \p of \a, M are said to he* orthogonal if their
scalar product is equal to zero, that is
b
(<t. T) = %\ *r ( 0 »l (*) ==0 (2)
a
The norm or “length” of a vector q £ (J fix, b| can now he dehned
by the equality
ll <i ll - l^ (?7 T ) (3)
If || tp || =?fe 0 wc can pul t|' (x) - and thus obtain, according
to Delia it ion 3, the relation
M 1
Ml2= ( - ^
1 \ il II Til II <1 II2 (<f. <f) = l (4 )

which moans that \|‘ is a “unit” vector. The cosine of the

angle between / (x) and g (x) is defined by the relation

c«s(/, » )- ||/{|'|^ || (o)

The justification of this definition lies in the fact that, according
to the Cauchy-BunyaUovsky inequality (see $ I), Sec. 2 of Chapter 8),
we aI wavs have
I (/, g) | < II / II I! g II (“)
and hence relation (5) actually determines a unique angle in the
interval 10, jt|.
The projection of / on g (where g =£= 0) is defined as the scalar
quantity

II/IR o-M /T « )= t - f (7)

Now let us define the notion of convergence for the space (J fix. b\.
We say that q>„ converges to . i.e. «pn —» q- for n —*■ —oo. if
v
IK« -<i It ~ ( f IU« (^*) — <p(x)|arfz)l/2--^0 («S)
a
for w —oo. Hence. the refill ion lim q lt = q means that the
14 >o

functional sequence {qu (:c) } converges in the mean to q (.r) on (ex. 6|.
 CH. ti. FOURIER SERIES AND FOURIER INTEGRAL 507

Similarly, tin* relation

f — / 1 / 2 ~7~ • • * i fn ••
is understood in the sense that
u b n
( 10)
/ - £ h |j = ( j [ / ( ^ ) - 2 -0
/<=I f{=*1

for n —►l oc, that is the series 2! h ( x) *•** convergent in the mean
h - 1‘
tv f {*) on In. /;).
Wo now proceed to establish the analogy between the resolution
of a vector x of tin* three-dimensional Euclidean space with respect
to an orthogonal basis rf. o2» c 3 and the expansion of a function
/ (*) £ Q Itf, b\ i n t o a Fourier series with respect to a complete
orthogonal system
*Pl (■T)i 2 (*^*)» • ♦ n (^)i • • •i k (^-) ^ Q 1^? ^1?
A- 1 , 2 , . . . (II)
For every vector x belonging to tbe three-dimensional Euclidean
space there exists a unique resolution of llie form
x = i te, -|- -{■ x 3e 3 (12)
with respect to any fixed orthogonal basis o,. c 2. o3. The coeffi­
cients of this resolution can bo easily found if we use the notion
of scalar product. Indeed, multiplying scnlarly equality (12) by e4.
/ — 1. 2. 3. wo obtain
(x, c\) - Ti ((*,, «\) = x 4 || e, ||2. / -* 1, 2. 3 (13)
since the basis e,, «■,, e 3 is supposed to be orthogonal.* From (13)
we derive
X*“ F * T F ’ i = i . 2, 3 (U )
The quantities
tt||c»|| = ^ ' , A - I, 2, 3 (IS)
(whore the symbol || || designates the length of a vector) are the pro­
ject ions of the vector x on the axes whoso directions are specified
by the vectors v,(. />• — 11 2. 3.
Taking the scalar squares of both sides of equality (12) we arrive
at the relation
II x II2 J* II c, if -f 4 || c . If 4 i i i p (16)

* In lh<> m*i n*r«i 1 rii^e I lx* nrl iiMijiiiiiil v(>chii< p,. p» run! p_, air* of nrhilrary
Iwurths.
oOS MULTIPLE IN T E G R A L S . FIELD THEORY AND S E R I E S

which expresses the Pythagoras* theorem for the three-dimensional

case: the square of the length of a vector is equal to tlie sum ol the
squares of its projections on any three mutually orthogonal axes.
For the space Q (a, id we have a completely analogous situation;
every “vector” /can be uniquely expanded with respect to the “vectors”
of a complete orthogonal system {<£*}, i.e. represented in the form
/ = W i — c2(f2 . ..— . . . (17)

(see § 6, Sec. 1 of Chapter 11)where the coefficients e* of

expansion (17) are determined by the formulas

* = 1 , 2 > 3 ’ ••• (18)

Thus, every “vector” / belonging to Q la, 6) is uniquely specified

by the infinite sequence of its “coordinates” {c*} in every “orthogonal
basis” {cpj,}.
The system {cpft} being complete. we have ParscuaVs relation
II / II3 - rj II fPi IP r c\ || <,:2 ||2 r ■• • r c\ || q n ||2 - . . . (1»)
(see § 6, Sec. 2) which can be interpreted as Pythagoras* theorem
for the functional space Q (a, b],
if we take not all the base vectors c /. = l t 2, 3, but only some
of them, for instance, e, and c 2, equality (Hi) is replaced by the
inequality
II x IP > * J || e , |F - ^ || e 2 IF (2 0 )

Similarly, if an orthogonal system {<p„ } of Ihe space <J la, id is not

complete, Parseval’s relation (ID) is replaced by Bessel s inequality
II / IF > II <p2 IF -r 1n <fn ( 21)

These geometric ideas and analogues are used in the theory of the
so-called Hilbert** spaces which is widely applied to various problems
of quantum mechanics and mathematical physics.

APPENDIX 4 TO CHAPTER 11

SOME APPLICATIONS OF FOURIER TRANSFORMS

We now discuss some applications of Fourier’s transformation.

Many physical devices can be interpreted as the so-called input-
output systems. An input signal described by functions /, (/), / 2 (/), . . .
is applied to the input of such a system which results in the nppearan-

* Pythagoras, a Creak philosopher and mathematician of the 6th renturv

B.C.
** Hilbert, David (1862 l!143). a famous Dorman mathematician.
 CH. 11. FOURIER SERIES AND FOURIER INTEGRA!. •jl'*!

ce of an output signal dEscribed by another system of f 11net ion?

Xj (0. x 2 (f)» • ♦ • • Thus, such a device can be regarded as a conver­
tor which transforms input functions into output functions. Km
instance, various amplifiers can be regarded as such systems trails
forming the voltage / (2) of an alternating electric current applied
to the input terminals into an alternating voltage appearing at
the output terminals.
An input-output system is called linear if the following two
conditions are fulfilled:
(1) if /(/) is transformed into x (0 then c /(/) (where c is an arbitra­
ry constant) is transformed into c x(t).
(2) if f i ( l ) and f 2 (/) are transformed, respectively, into x,(/)
anti x2(/) then / t (/) — /•_.(/) is transformed into x, (/) — x 2 (/)•
If conditions (i) and (2) are satisfied we say that the principle
of superposition holds for the system in question.
We shall also suppose that every steady-stale process of harmonic
vibration with frequency i» is transformed into another steady-stale
vibrational process with the same frequency. This means that we
impose one more condition;
(3) every function of the form goes into a function of the
form A (ui)eiiot.
In the general case tlie factor of proportionality .!(<•>) is depen­
dent on the frequency co, which means that an input-output system
may transform harmonic oscillations with different frequencies in
a different way. The function A — /t(o>) is called the spectral
characteristic of the system. I 11 the general case this function may
assume complex values:
-1 (<o) = H (<o) e*^') where H (co) -• |.-1(ro)l and
<p (co) ~ arg
Consequently, harmonic oscillations described by a function eu'}i
are transformed into harmonic oscillations of the form -d(<o) =
= n (o j)
The modulus 7? (co) — \A (<o) j of the spectral characteristic
is known as the fvetju en ct/ chn r a f t e r i s tic of the system. The
amplitude of the output harmonic signal is H(o>) times that of the
input harmonic signal with Ilie given frequency co. The argument
<j (<») a r g ,l(o ) of the spectral characteristic /I(to) is called the
c h a r a e t e r i s t i r of the input-output system. It determines
the pha.se displacement of a harmonic signal of a given frequency
oj when i t is transformed by tin; system.
If Ilie spectral characteristic of a linear inptil output s\.-1em
is known we can solve the following two problems:
I- lliven an input function / (/). it is required to determine the
corresponding output function ,r 1/V
570 -MULTIPLE I N T E G R A L S , F I E L D T H E O R Y AND S E R I E S

2. It is necessary to find (lie input function / (/) from a given

output function x(/).
The second problem is reverse with respect to the first. We shall
begin with the first problem. Let an input signal described by
a function / (/) be applied to the input of the system. The Fourier
transform of the function / (I) is determined by the relation
—»»
) /< T )e-» * "* d T (1)
— CO

ami the function /(/), the inverse transform, is represented by

Fourier’s inversion formula:

/(/) = eiMi ck*> (2)

The integral on the right-hand side of equality (2) can be regarded

as a sum of infinitely many infinitesimal harmonic oscillations of the
form
- ]_ i (to) e“',<c/w (3 )
yin
13ut every function of the form eiiut is transformed into the function
/l(co)t'i<d< and. consequently, according to property 1 of the system
the harmonic signal / (co) tfto) ei6>* goes into the harmonic
signal
-4= / M ) A (01) e™1= - 4 = A (to) / (to) e d o ) (4)
V la f \ 2a
Property (2) of the system implies that the sum of harmonic vibra­
tions of form (3) is transformed into the sum of functions (4) and
hence the function / (t) determined by relation (2) is transformed
into the function
x (/) = j A (co) / (co) eiMl do> (5)

Formula (5) gives the solution of Problem 1.

Relation (5) indicates that the Fourier transform x (co) of the
function x (f) is expressed as
x (co) — A (co) / (o>) R>)
Thus, to find the F o u r i e r transform x (o>) of tlie output function x (t)
we must multiply the Fourier transform / (co) of the input func­
tion f(t\ by the <pertral clunacterist ic of the input-output system.
 CH. 11 . FOUHIER SERIES AND F O U R I E R INTEGRAL 571

To solve the reverse Problem 2 we find the function

(7)
from relation (G) and tlien apply the Fourier inverse transformation
to equality (7), which results in

—ac

Relation (8) expresses the solution of Problem 2. It enables us to

reconstruct the input function / (*) corresponding to the output
function x (/). To this end, we first apply the Fourier transformation
and determine the Fourier transform x (<o) of the function x (0
and then use formula (8).
Such problems are encountered in radiophysics, radio engineering,
in the theory of control systems and the like.
Fourier’s transformation is also widely applied to solving various
boundary value problems of mathematical physics. The matter is,
that the Fourier transform of a sought-for function may satisfy
an equation which is much simpler than the original equation for
the unknown function. Therefore the boundary value problems
of mathematical physics are solved by means of the Fourier trans­
formation according to the following scheme: we first apply the
Fourier transformation to Lite equation which is satisfied by the
soughl-for function and thus obtain an equation for the Fourier
transform of the unknown function, then we find the Fourier trans­
form of the sought-for function from the equation thus obtained
and finally determine the solution of the original problem by moans
of the Fourier inverse transformation (see 117)).
Let ns consider an example. Suppose that it is required to find
the distribution of temperature in an infinite rod at an arbitrary
moment of time t > 0 from a given temperature distribution at
the moment t — 0. Let the x-axis be directed along the rod. Then
the temperature u ala point x of the rod at moment t is described by
a function u — u (x, t), — oo < i <C-j- oo, 0 < t < + ° ° , of two
variables. As is known (e.g. see (17)), the temperature u(xt I) satisfies
1he heat conductivity equation
0 (9)
The initial temperature distribution being known, we can write

of tempera lure in 11n* unbounded rod at an nrbilrarv time moment t

572 MULTIPLE INTEGRALS. FIELD THEORY AND S E R I E S

we must find the solution of equation (9) which assumes the initial
values (10) at the moment / = 0.
Lot us solve this problem by applying the Fourier transformation
to the function u {z, t) regarded as a function of the argument z
for every fixed value of t. Designating the Fourier transform of the
function u (x, /) by u (X, /) we can write

u( Xf / ) « — j u (z, t) e~i)xdx (11)

Y —oo
Let us multiply both sides of equation (9) by ■ *_ e~i>x and integra­
te the resulting relation with respect to x from —oo to -f oo under
the assumption that the function u and its derivatives tend to zero
sufficiently fast as x — zfcoo. Then, integrating by parts, we obtain
0 l
UT V'In
— OO

-r°°
du (X, /) 2 f d*u
dt
1
e~i>'xdx —
y2k J
— OO

1 |X=+oo
a2 1
-L
y i n Ox | —oo

1
>

I=a-j»30
1
>
*-

H
c*
♦

V' iji —cu

-\-ro
— a2X2 ~ 4= f ite~ikx dx = — ().. f)
V2ji —JOO '
since the expression e~iXac is bounded* and (by the above hypothesis)
the function u and its derivative dx — tend to zero when j - + ± o o .
Consequently, the Fourier transform of the sought-for function
satisfies the equation
+ aWu = 0 ( 12)
dt
which is considerably simpler than equation (9). From equality (11)
we find, by putting t = 0, tlic initial condition for a (X, t):
-r*> 4 -00
(X, 0) = —ip- f a { x , 0) Y - f ^ f ( x ) e - i 7xd x = : t ( l ) (13)
y in J y in J

* In fact, we h a v e \e w-x | = | c o s / j - S - i s i n X x | = l / c o s 2 X x 4 - s i U “ Xx— J.

 CH. 11. FOURIER SERIES AND FOURIER INTEGRAL 57«

Let us solve equation (12) with initial condition (13). From (12
we line!
du — a"X2 dt
U
and, consequently, we have
In n = —a2X2/ j In C
and
it (a. /) — Ce
where C is independent of / and may only depend on X. The quantity C
can be determined by means of initial condition (13):
u (X, 0) =r C = / (X)
Substituting the value of C thus found into the foregoing equality
we obtain the following expression for the Fourier transform of tin
unknown function:
u (X, /) = / (X) (14
To determine u (x, t), we apply the Fourier inverse transfer
mation to equality (14) in which wo substitute the expressioi
1 30
/(X). This results in
w 3 to
+» ^
1
u (x, t) — J u (X, t) cikx dk
Y2n —oo

= — \ f ( l ) d i j «— ></>.=
—O
O —oo
-{-OO *f«'
= - - j f ( l ) d l j r nIX,<c o s X ( j - 6) dl =
—oo 0
Loo
iaH c%
2a y ^ t J ' '
because we have (see Chapter 10. § 2, Sec. 5)
+j° r— _
f e- a2?-2cospXdX = i . |/
o
Thus, the solution of equation (0) with initial conditions (10) h
expressed by the formula
~T x

u (x, t) — e *«“ dc,

574 MULTIPLE INTEGRALS. FIELD TH E O RY AN D S E R I E S

Fourier’s sine and cosine transforms are similarly applied to solving

various boundary value problems for the .semi-infinite interval
0 ^ x <C —oo (e.g. see [171).
The application of Fourier’s multiple integrals and the corres­
ponding Fourier transforms makes it possible to solve some boundary
value problems for unbounded regions in the plane and in space,
for instance, such as the whole plane, half-plane, a quadrant, the
entire three-dimensional space, half-space etc.

APPENDIX 5 TO CHAPTER 11
E X P A N D IN G D E L T A F U N C T IO N IN F O U R IE R S E R IE S
A N D F O U R IE R IN T E G R A L

Let us take the della f u n c t i o n <5 (x0, x) and compute (formally)

its Fourier coefficients by using the ordinary formulas. This yields,
for xn £ (—I, /), the expressions
i
t f An* 1 ^ Avlt () , ,
Oh = — \ o (x0, c.) cos—— at ——cos—^— . /»*=!, _\ . . .
-/
/
= y \& (X(r I) 7 -

i
j 6(x0, sin —— = -j-si u . k — 1, 2. . . .
-i
Consequently, (formal) Fourier’s series of the delta function 6 (x0, x)
is of the form
r ^ v 1 i t * A rcx ij • A t t * \ ___
+ ( cos ~ cos — + sm —— sin — ) =
k=\
An
=4r+4 S c o s 1 0) (I)
?|rr,|
or, in the complex form.
I ^ i/t —
* ( * - . v 0)
6 (x(>, x) — 2 i
2 (2 )
/it**-)0
Lot us consider the sequence of partial sums of this series:
n
tf 1 . 1 yi / kjT.tii A\ix * Amt,) . knx \
6„ (x0. x) = -j;- r - 2 ( « * — c o s - ; - sin — sin — ) , (3)
«t—I
n — 1. 2, . . .
 CH. 11. FOURIER SERIES AND FOURIER INTEGRAL 575

The complex form of tins sequence is

if*—
6n (x0, x) 21 e 1 2 n = 1, 9 (4)
k=—U
If / (x) is an arbitrary piecewise smooth function on an interval
(—l y I) we obviously have
/
lim \ f (z) K (z0, z )d x = f (x0), * o € < - / , 0 (5)
n-» \ jo -I V

provided that the function / (x) has been redefined at every point
of discontinuity x* according to the relation / (x*) = /<** 0)+/
2
(x*+0)
We can therefore say (see the note after relations (7), (8) and (9)
in Appendix 2 to Chapter S and footnote on page 381) that the
sequence {6n (x0, x)} is weakly convergent to the delta function 6 (x0. x)
(relative to the class of piecewise smooth functions on (—I. I)) or,
in other words, series (1) is weakly convergent to 6 (x0. x). This fact
can be expressed symbolically by the relation
kjZTQ knx knx \
®(*^o» z) COS
T -j-sin I sin T }
k^l
Multiplying both sides of equality (6) by an arbitrary piecewise
smooth inaction / (x) and performing term-by-term integration
with respect to x from —/ to / we arrive at tlie equality

;/(* .) = - ¥ - + 2 ( « * c o s ^ + Afcs i n ^ p ) (7)

n=i
where the coefficients aln bh arc determined by the ordinary formu­
las for Fourier’s coefficients of the trigonometric series of the piece-
wise smooth function / (x) on the interval (—Z, I). The validity
of equality (7) was proved in § 2, Sec. 4 of Chapter 11.
Equality (6) understood in the sense of weak convergence is
called the expansion of the delta function 6 (x0t x) in the Fourier
series with respect to the trigonometric system.
Similarly, the delta function c'i (x0, x) can be written in the form
of Fourier’s integral. Applying (formally) the Fourier integral
formula (see § 9) to fi (x0, x) we obtain

6(j-0, ar) = -L f d). j 5(*„, J ) c o s /.( | — i ) d s

0 -X )
570 MULTIPLE INTEGRALS. FIELD THEORY AND SE R I E S

which results in
-J- DO
| f
6 (x0, x) —— \ cos a. (x0 - X) d/. (8)
o
This relation is called the representation of the delta function ft (x0, a-)
as a Fourier integral. It should also he understood in the sense
of weak convergence. Multiplying both sides of (8) by an arbitrary
function / (x) (absolutely inlrgrablc over the* x-axis from «—oo to
4-oo and piecewise smooth mi every finite interval) and reversing
the order of integration with respect to x and /. on the right-hand
side we arrive at the relation
-J-OO 4-JO
/(*(.) ^4" j d* ( / ( l ) C°S>-(^0— 1 ) ^ (9)
0 — DO

which was proved, for a function / (a:) of this type, in § 1), Sec. 2
of Chapter 11.
Thus. when expanding the delta function ft (x0. x) in Fourier
scries (b) or representing it as Fourier integral (8) we can operate
on ft (x0. x) as if it were an ordinary piecewise smooth function
(absolutely inlegrable from —oo to : oo in the case of Fourier’s
integral). We can also say that expansions (b) and (8) (understood
in the sense of weak convergence) can be dealt with, to some extent,
as ordinary equalities.
For greater detail on the delta function we refer the reader to 171,
191 and [17|.

APPENDIX ti 10 QiAPTKIl It

UNIFORM APPROXIMATION OF FUNCTIONS

WITH p o l y n o m ia l s

Here wc present another proof of YVeierstrass’ polynomial appro­

ximation theorem (sec § 5 of Chapter 11) on approximating a con­
tinuous function with algebraic polynomials which can easily bo
extended to functions of several independent variables. In the case
oT a function / (x) possessing the continuous derivatives / (s)(x), s —
— 1, . . . » /V, this proof makes it possible to construct an algebraic
polynomial approximating uniformly the function / in such a way
that the derivatives of this polynomial up lo the order ;V inclusive
uniformly approximate the corresponding derivatives of the func­
tion /.
T h e o r e m / ( W c i e r s t r a s s ’ f*ofj/aom in/ A p p r o x i m a t i o n
T h e o r e m f o r f'tr a c t i o n s o f O n e I n d e p e m l e n t I' ( t r i a b l e ).
 CH. 11. FOURJEK SERIES AND I'O l. K IE R INTEGRAL

Any junction j (x) continuous on a bounded closed interval a<'.x^.b

can be uniformly approximated with an algebraic polynomial to an
arbitrary degree oj accuracy.
Proof. We shall suppose, without loss of generality, that the
interval Ia, 61 is such that 0 < u < 6 < 1 (if otherwise, we can
perform the corresponding change of the independent variable 2 *).
Let us lake arbitrary numbers a and p such that U < a < f l < ( / <
<C |> <C 1. Now, considering these o. and p to be fixed, let us continuo­
usly extend the f u n c t i o n / (x) to I be whole interval 0 ^ ^ I
in such a way that / (z) is identically equal to zero (/ (jr) ~ U) foi
0 ^ x ^ go and P ^ x ^ 1.
We shall prove that the expression

— i------------------- (i:
J (1 —ua)n <iu
-l
(which is an algebraic polynomial in x of degree 2n) uniformly
approximates, for a sufficiently large «, the function / (x) on tin
interval [a, 61 to an arbitrary accuracy. To this end we note that
i i
I
J „ = f (1 — »a)" d i » J (1 -v )" du =
n- I u= 0

and
J*n = j (1 —V-)" d v < ( I — 6*)“

for any 6, 0 < 6 < : 1 . Consequently, we have

0 < 4 = - < - ( l —6 T ( » i 1) — o for n - y ! oo 12]
J il
provided that 6 = const, 0 < 6 < 1• Performing the change ol
variable u — x — v we can rewrite expression (1) in the form
P-*
a—x
Pn { X) = (3:
j <i I-2)- j f
-1
Lot us now estimate the difference
$~x 1
J / (C f-.T) (I —!•-)» dv— J f(j)(\ -r-)ndv
c„(.r) i(x) = 2=Z n Tj ^ — (/.
37 — 0824
o7S MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

on the interval a x ^ b. Given e > 0 , we can choose 6 , 0 < 6 < 1,

such that the inequality

| / (x -j- v) — / (x) | <C - 7 for a x ^ b and ]v | 6 (3)

is fulfilled and the relation 0 < i -f r < 1 holds for a ^ x ^ b

and | v j ^ 6 . The numerator of fraction (4) can be represented as
-6 fl-x

j t ( v ~ x ) ( 1- V -)" dii -r \ i {v -r x) ( I — r ;)" dv

a-x

6 —6
+ j 1/ ( x - r f ) — / (i')l (1 — j /(x ) ( 1 — t “)n i / e - r
-6 - 1

r j / W ( l - ^ ' dv («>

Kurt her more, by (5) we have

ft

J \ (/(x-:-i/) /(-^OlO v~)'du 2 1-/^ J nl (7)

-6
i\.»u pulling .1/ m a x j / (x) | \vc o b t a i n the e s tim a tio n s
tl '7x<b

I | /(v I x) (I — u'-)ndv ; .i/y?.

a-x
and («)
0-*
t*
\ j ( u - t x ) ( \ - v 2r dv

since we have — 1 <Z rx — x < 0 and 0 < [t — x <i 1 for a x ^ 0

and the inequalities
ft |
I f /(a-) (I -w T ^ I al"1 | ( / {-r) (I - I’T'di'l MJf, O')
^I
T l i c i v f . i i <■ t l i c im m c r : il. ir of ( ru c tio n (() *!.**•>. not cxocc.l l l i c q u n n t i l y

2 (III)
 CH. 11. FOURIER SERIES AND FOURIER INTEGRAL r>79

in its mod ulus. Ilencc, for ('i) we can write the inequality

—— • 9 v _£«. 00
2 n
for all x £ l«, 6). But, by (2), the second summand on the right-hand
side of inequality (11) is less than — for all sufficiently large n.
Consequently, for all sufficiently large n we have
I (x) — f ix) I <1 € for all £ £ la. 61 (12)
and hence the theorem has been proved.
For functions of several variables we can similarly prove
T h r o v e n i 2 (II e i c r s t r a .sx* P o l 1/ n u m i a f .1 p p r o x i m a t i o u
T h e o r e m f o r Fn tte tio n s o f Sere, r u t / t u lip e m l e u t I ( t r i a b ­
le#). Suppose a junction j (z,, x*........ x,„) is continuous in an m-dimen-
sionalparallelepiped 11: i —1,2,. . ., m, where i)<C.al<C.bi<Zj\ ■
Let us continuously extend / to the entire tn-dimensional unit cube F^:
0 1, i — 1, 2, in such a way that / is identically
equal to zero outside a wider m-dimensional parallelepiped II*: a* ^
^ Xi ^ (J,- where 0 <C a* <C a* < 6j <i Pj 1, i — 1, 2, . . . » m.
Then the algebraic polynomial of degree urn in the variables xu x 2, - - -
. . xm expressed by the formula
P n (zlT x2, .. ., xtti) —
Pi Pm
\ . . . V / (f<i , . . . , u m ) (1 — ( f / i - r,)-j ' . . . { I - ( « m — J m ) 2 ] 11 < h i\ - - • d u
at am _______________________________________ ______________________________
\
(13)
[ J (I - I * * ) |
-1
uniformly approximates the function / (xt, . . ., z m) to an arbitrarily
chosen accuracy in the domain J1 for sufficiently large n.
A-ole J. If the function / (z) (/ (zlt . . z ffl)) possesses con­
tinuous derivatives up to an order ;V inclusive the derivatives of
/\, (z) (Pn (z(, . . . » z„,)) up to Ihe order iV inclusive uniformly
approximate the corresponding derivatives of / (z) (/ (zt, . . xm))
on the interval (a, h\ (in the parallelepiped II) to an arbitrary degree
of accuracy for all sufficiently large n.
Wo shall illustrate the proof of this assertion by taking the simple
case of a function / (z) which is coni innous on an interval a ^ x <1 fr,
0 < a <C b < 1, and has the continuous derivative j' (z) on [a, />}.
hot us extend / (x) to the whole inlerval 0 I in such a way
llml / (z) and f(x) are continuous on this inlerval and identically
equal to zero outside an interval n. r p whore 0 <C <C n <C
<C h <; p I. DilTorenl iating polynomial (U with respect 1o .r and
37*
580 M U LT IP LE INTEGRALS. FIELD THEORY AND SERIES

integrating by parts we obtain

ft ft
/ (w) [1—(H—j)2]n t/u j / [ l _ ( u_..T*2jr,dW

<■*>- — ------- 57;--------------- - ----------- ir„-------------

ft ft
- ] I («) (/'(•<)!!- ( u - x ) 21" r f li
a
n u n
ft - V

a - a:
^ ( I — f 2)n d r

-1
Now the difference
ft-* 1
^ / ' ( x + i>)(1— v2) n d v — ^ f (x) (1 — u - ) n d r
Pn ( x ) - r (x) = ^ -------------------- 2 7 ;— —-------------------
can be estimated as was done for difference (4) in Theorem 1,
and this completes the proof of our assertion.
Note 2. The theorem below is a direct consequence of Theorem 2.
T h e o r e m *i ( Meier.stru ss* T r i y t m o n t t t r i e .-Iy>y>ro<cim a-
t l o n T h e o r e m ) , If f (x) is a continuous (unction on an interval
Z<^x which assumes equal values at its end points. i.e. / ( —/) =
= / (/), it can be uniformly approximated on this interval, to an arbi­
trary accuracy, by a trigonometric polynomial of the. form

-y- + 2 ( ah cos -^7 ^-4- bk sin ) (14)

fc=l
Proof. Let us put ^ = 0. Then the function F (0) = / J—
= / (x) is continuous in the interval —n ^ 0 ^ ji, and F (—n) =
= F (ji). Let us introduce, on the plane with Cartesian coordinates
»], the polar coordinates 0, p: £ = p cos 0, q -= p sin 0. Now
consider the function q> (£, q) = pF (0). This function is continuous
throughout the £, q-plane and coincides with F (0) for p - 1, that
is rp (£, q) = F (0) on the circumference of tln> circle £2 -h q? 1.
By Theorem 2, the function (p (£, q) can lie uniformly approximated
in the square —1 \ 1.—1 ^ q ^ 1 by an algebraic polynomial
P„ (£, q) to an arbitrary accuracy. Hence, putting p = 1 we arrive
 CH. 11. FOURIER SERIES AND FOURIER INTEGRAL 38:

at the trigonometric polynomial P n (cos 0, sin G) which approxima-

tes the function / (a) with the same accuracy on the interval —;i ^
^ 6 ^ ji. Next we return to the argument x = ■- and obtain tin
trigonometric polynomial P n ^ co s^ -, sin-^r-j which uniformK
approximates the function / (x) on the interval —I ^ x ^ / wilt
the same accuracy. Replacing in the latter polynomial the product?
JTwT
and the higher powers of cos -j- and sin -y by the corresponding
linear combinations of sines and cosines of multiple arcs wo reduce
this polynomial to form (14). The theorem lias been proved.

APPENDIX 7 TO CHAPTER It

ON STABLE SUMMATION OF FOURIER SERIES

WITH PERTURBED COEFFICIENTS
Suppose that we know the exact values ch of the Fourier coefficient?
of a square-integrahie function / (x) on an interval la, b] with respect
to an orthonormal* system {<pfc (x)}:
b
Ch = j / (x) (f'ft <■*) dx, k — 1, 2, . . . (1)
Cl
If the equality
+ 00
/(* )“ opM- t) (2)
h= i
is fulfilled for a value of x belonging to [a, b\ we can replace in (2)
the sum of the series by its nth partial sum and thus obtain the
approximate equality
n
/ {*) ~ ^ o,«ffc (*) (3)
*= i v
which goes into exact equality (2) when n — -foo. Thus, increasing
the number of terms n we can obtain equality (3) which approxi­
mates exact equality (2) at the point x to an arbitrary accuracy.
But in practical problems we usually deal with approximate
values of the Fourier coefficients, i.e. with the numbers
ch = Ch f (4)

* An orthogonal system {tf/jkO} is called orlhnnonual if the norms ol

all tlie functions <\ft ( j -) . k = t. 2, . . .. nre equal In unity, i.e. || (r ]|
i, >i<
= j T* (*) R *‘=1. . -—7T.
5*2 MULTIPLE INTEGKAI.S, FIELD THEORY AND SERIES

Quantities ('i) are termed perturbed coefficients and the discrepan­

cies Ac* * arc the errors or perturbations.
If we substitute the approximate values ct, for the exact values clt
of the Fourier coefficients into approximate equality (3) this will
result in the approximate equality
n
K x) ~ c/,<fk(a-) (5)
whose accuracy may even be worsened when the number of terms n
is too great. If the perturbations Ac* are not subjected to any res­
trictions this assertion appears quite clear. Usually we impose the
following restriction on the quantities Ac*:

V (Ac*)4< 6 - (C)
h=l
where 62 is a sufficiently small number. If condition (0) is fulfilled
for a given finite value of 6 it follows that, there exists a square
-j- jo
intcgrable function / (x) on |«, M for which the series V c*q* (x)
*•-i
is its Fourier series* and. besides, by Parseval’s relation
b +°o -\-oo
f i / o ) - / (x)r-dx = 2 fh -c„)-
2 0)
a /*=1
the mean square deviation of / (x) from / (x) on la, &) is less than <V-.
Cut the validity of condition (0) for mi arbitrarily small ii dm-*
not guarantee the convergence of series
■4”00 ^
cuVu (*) (8)
A«1

+ *> -i 00
* Indeed, inequality V (Ac*)’- < - r °°» Bessel’s inequality cl <
k-i
b
< ^ f ~(x)dx and the evident inequality eg -< 2 (eg -r(Ac*)‘2 J, h= 1, 2 , ....

imply tlie convergence of the series ^ eg. But in the theory of functions
ft*“ll
of a real argument (e.g. see the Hiosz-Fisher theorem in [ 1 1 ]) it is proved
-i-JO ^
that a series oT form is convergent if and only if there exists a fun-
1
^ *^vX> —
ction Y(x) squnre-integrable on [«, b\ for which the series N' c;t<p/t (x) is its
Fourier series.
 C1I. 11. FOURIER SERIES AND FOURIER INTEGRAL 5«3

willi tlie perturbed coefficients. Therefore, even if condition (G)

is fulfilled, the accuracy of approximate equality (5) may even
be worsened if the number of terms it is loo large.
For example, let us consider the complete orthonormal system
\
*Pi (*^) = /—* q:it+i (■i) : 1/ “ Cos/*x, A’ = 1, 2, . . . (9)
y rt r -1
on the interval 0 ^ x ^ n. Suppose that a function / (x) is arbitra­
rily smooth on the interval l<) ^ x ^ nl ami that its exact Fourier
series
-f-»= r—
2 (-r) —y=- -F 2 f<Hi ]' -jeos/.a: (10)
h=l *^ h—I
converges to it arbitrarily fast at the point x — 0. Now let us put

Ac, —>0 , Ac*,, for A'- 1, 2 , .... <V> 0 (1 1 )

Then we have

2 (Ac*)2 =- —jf— < (12 )

^=1
which means that condition (G) is fulfilled for the perturbations At;..
Hut the series
-[-OO CO
2 (*) -- 2 (c * ■&Ck) <(;», { x )
><—i /i —i
is divergent at the point x —0. Indeed, by the hypothesis, the
series 2 (0) converges to / (0) while the series
h=t
___
2 Ac„ fp„ (0 ) - 6 / 4 2 , r 6K ^ i
h=\ k^\ 1
diverges hecause it (lifters from the divergent harmonic series
2 i~ only in the constant factor ft 1/ — ( 6 ^ 0 ) .
/<=i k r 7i
Hut nevertheless, if the exact Fourier series of a function / (x)
converges to / (x) at a point x £ l«, b) the validity of condition (6)
for an arbitrarily small ft > U enables us to make the quantity
A '(fl)

I / (-r)— 2 (.r) 1
h—1
become arbitrarily small if the number of terms A(ft) is chosen
in an appropriate manner. Thus, this makes it possible In obtain
 S4 MULTIPLE INTEGRALS. 1TELD THEORY AX11 SERIES

.it the point x, the approximate equality

N<6) _
/ ' ( x ) « Y ck<fk (x) (5‘)
A=1
which is as close as desired to the exact one. Moreover, it turns out
that it is unnecessary to impose any additional restrictions on Ilie
decree of snioothuess of the function / (x) and on the speed of con­
vergence of its exact Fourier series at the point x. Namely, the
following theorem is valid:
TUfioretn. Let {<p*(x)} be an ortho normal system on an interval
(a, bI which satisfies the condition
| (x) | A = const < -f-oo for k = 1, 2, . . . ;
a ^ x ^ b (13)
Suppose that for every b2 entering into the right-hand side of equality
(G) the number N (6) is chosen in such a way that the conditions
A; (6) -► “j-oo for 6 — 0 (141)
ana
62N (6) -*■ 0 for 6 -► 0 (142)
are fulfilled. Then wc have
JV(6) _
lim | / (x) — V awn (x) | = 0 (15)
6-.0 k—1

fur every a: t k , b\ which satisfies equality (2).

Proof. By relations (2), ( 4 ) , ((>), ( 1 4 , ) and (142) and the Cauchy-
I)nn> akovsky inequality for sums we have

I/ ( x) — 2] Ch<V\ (x) i — I ^ (x) — V Cfttpfc (x) I <

1 fc = l A= I
N lfi) -foo
C | — y Ac,«cph (x) I+ 1 Y <W< (*) I<
ft-1 ft=N(6)+1
/w(6) pm i<*>
Cr V Y
k=\ (Ac*)2r k--s\
v ^ (P*( T) +1/i=.V(6)+l (X) I<
_________ + 01’
< ^ T 6 2A'(6) + I V (x) | (10)
‘ fc.A'UO+1
The term A I'^ iV (6) on [the right-hand side of inequality (Hi)
l»*nds to zero ns 5 - + 0 which is implied by condition (142). and
•~nO
the term | chi\ u (x) | tends to zero for 6 - * 0 hecanse wc have
fc=A’(6}-i I
 CH. 11. FO UI U EK SERIES AND FOURIER INTEGRAL

+00

condition (14,) and Ilie scries 2 (a:) converges at the point a.

The theorem has been proved.
We can now draw an important conclusion from the above proof.
Let equality (2) be fulfilled at a point x £ [a, b\. Suppose that
condition (6) imposed on the quantities Ack is satisfied for a given
6 Z> 0 (where Acfi. k = 1, 2, . . . . are the perturbations of the
Fourier coefficients). It appears clear that to minimize the quantity
I / (x) — 2 (*) I {Ck = Ck + Ac,{, A* - 1, 2, . . .) at the
h=\
point x £ |<z, b\ wo must choose the number of terms ;Y (6) in the
.Y«5)
partial sum 21 ch(lh (^) so that it should not be too small (because
h= l
+oo
2 (*) on the right-hand side of inequality (16) must
h -Y(6)+l
be small) and too large (because the term A |'/ 6*Ar (6) on the
right hand side of (16) must also be sufficiently small).
fivery method of reconstructing a function / (a:), to any given
accuracy, from its Fourier series with perturbed coefficients satis­
fying condition (6) for an arbitrarily small 6 > 0 is referred to as
a stable method for the summation of Fourier series with perturbed
coefficients.
Thus, the above theorem shows that if the quantity 62 on Lhe
right-hand side of relation (6) can be made arbitrarily small it is
possible to choose /V (6) in approximate equality (.V) in an appro­
priate manner so as to perforin stable summation of Fourier’s series
with perturbed coefficients.
Stable methods for the summation of Fourier’s series with per­
turbed coefficients were developed by Tikhonov* in llul. Tikho­
nov’s methods make it possible to reconstruct, from a given Fourier’s
series with perturbed coefficients, the corresponding function / (x)
and also its derivatives but we shall not discuss here these more
complicated questions.

* Tikhonov, Andrei Nikolayevich (born in 1900), a prominent Soviet mathe­

matician.
 SUPPLEMENT 1

A symJj t o ti e Expansions

In many problems of mat hemal ics and mallienuitical physics

the investigation and computation of a function / (x) in the neigh­
bourhood of a finite point x0 or in the neighbourhood of the point
at infinity* is connected with considerable difficulties. These ilif'
lieu 11ics may often be overcome by moans of an asymptotic expansion
which substitutes a simpler function for the given function / (x).
This simpler function is chosen in such a manner that it can he inve­
stigated and computed in an easier way than the original function
/ (x) which it. approximates to an arbitrary accuracy when x tends
to x0 or approaches infinity.
We shall begin with some examples of asymptotic expansions
(§ 1) without giving general definitions, then we shall dwell in
more detail on some general definitions and theorems (§ 2) and,
finally, wo shall illustrate Laplace’s method of asymptotic repre­
sentation of some integrals hy deriving asymptotic formulas for the
gamma function (§ 3).

§ I. EXAM PLES OF ASYMPTOTIC EXPANSIONS

1. Asymptotic Expansions in the Neighbourhood of the Origin.

Taylor’s formula (e.g. see 18|. Chapter 8) yields the well known
asymptotic expansions for the following elementary functions:
r2 rU
cx — 2! 0 (:r ')
u- I
2 n-l
COS X — 2! » 4! 41 ’ # - 1 '( - i ) — 1 )! +
-|-0(xu) (where n is odd) (1)
n
.r** rO •> rH-*l
Sill a1 - X- :t! * K5! T - *- •
* *• +‘ (\ - ! *) / • ( n — 1)!
-|-o(xT') (where n is even)

* 'Flic expression “ in the neighbourhood of the point at infinity'’ means

** for x *■ [-oo” or “for x — —oo“ or “ for | .r | —
*■ ao“.
 SUPPLEMENT I. ASYMPTOTIC E X P A N S I O N S 587

In (14^) = ^ 4 + T - " • - H - 1)"-1—

( l - f ! ) “ - ! - ! — 2 + a(g2r 1) xz + . . . (I)

- « ( « - » ) - ( < » - -+ 1)y. 0 (x..}

n! v 7
where the symbol o (xn) designates a quantity of a higher order
of smallness than xn for Expansions (I) are used for eva­
luating limits of various elementary Junctions for x — U when they
cannot be found directly from the given analytical expressions of
the functions.
2. Asymptotic Expansions in the Neighbourhood of the Point at
Infinity. Wo now consider some asymptotic expansions in the neigh­
bourhood of the point at infinity which are used in various appli­
cations.
In mathematical physics we are often interested in the values of
the funct ion

(t)
for large values of the argument x > 0.* Let us derive the asymptotic
expansion of the function Mf (x) in the neighbourhood of the point
at infinity, i.e. for x —► - -|-°o. Integrating by parts we find
-J-o© +oo 4-oo
J
-£2
e *
J f n lit C. -X
r W
►^
e

Again integrating by parts in the integral j etc. we oh-

X

tain, after the integration by parts has been repeatedly perfor-

* This function can bo represented in Ilie form T —7 ^ C

y zi J
.v x
— 1 — <T»(.r) whore <|) (x)------------c“ “2 r/|. Thu integral is called the
0 0
error funetion and is denoted hy Krf (x). It plays an important role in the
probability theory, theory of heal eomluctiv ity, slut istical physics etc. There
are various tables of its values for difforeni value* oi tho i
r>ss MULTIPLE INTEGHALS, FI ELD THEORY AND SEfUES

med (n — 1) Iirues, the following asymptotic expansion for IF (x)

e ~ xZ\ . A . 1 - 3 ... (2 * l)
-y r J .i i - S ( - i) — ■f / M * ) ] ( 2)
v h=i
where
1-3 . . . (2 n -i -l) +°° -52
ex”x
Rn (•■**) — ( — 1)n+1 9ti \
X
p i*
The remainder term R n (z) satisfies the following apparent ine­
quality:
1.3 ... (2« - I)
/?„ (j ) 1< 1 ‘V j «-s*2|d i = 2n+lx2M+2 (4)

Let us designate by for i->-oo, every quantity which

satisfies the relation
O const*-— for
(?)| x —* o o

Then, by estimation (4), expansion (2) can be rewritten as

- x2 - - (
,|rw t v i r i!- s ( - 1)'

We have O (vx-} <•r- f —*0 for r — oo ami hence, discarding the

term O 2-^—j , we can write the relation
—X^ n
h 1.3... {2k—1)
xIr (*)
X* l/? t 1-1 A=
2 (-!)
1
(2x*)** J for x oo (6)

Relations (5) and (6) represent the asymptotic expansion of the

+«►
function xY { x ) - - —^= f e~Z* dt for {- o o , t h a t is i n t h e n e i g h -
l/;c J
X
b o u r h o o d of th e p o i n t a t in f in ity .
K x t e n d i n g t h e s u m m a t i o n w i t h r e s p e c t t o t l i e i n d e x k Vo a l l t h e
v a l u e s k = 1 . 2 . . . . we o b t a i n t h e a s y m p t o t i c e x n a n s i o n

—X-
-f^OO
1 - 3 . . .(2A;~1)
T (*> x f/.n 1 + 2
h=l
( --* ) (2r~)h J (7)
 SUPPLEMENT 1. ASYMPTOTIC E X P A N S I O N S 58

where the series on the right-hand side of (7) is an asymptotic seriei

This series is divergent for every value of x. Taking an arbitrar
fixed and sufficiently large value of x \vc can easily see that whe
the number k increases the modulus of the kth term of the seri<
first decreases and then, after its minimum value lias been attainec
increases and tends to infinity. But, according to inequality (4
the difference between lF (x) and the nth partial sum of this scri<
satisfies the inequality
n
T (x) -
. - .v-

[ s< l)
1-3..
2hT~k
1)

^ €~xl 1.3 . . . (2n L i )

2,i+,i 2n'r2
In other words, estimate (8) shows that the error arising win
the function T (x) is replaced by the nth partial sum of series (
does not exceed, in its absolute value, the modulus of 1lie first di
carded term and thus it tends to zero fast, as x -*■ 1 oo.
The repeated application of integration by parts is one of ll
general methods of deriving asymptotic expansions. Using th
technique we can obtain asymptotic expansions for the exponenti
integral
Ei (x) = £ — oo <Cx <c 0 for x — o o

cosine integral
+°°
Ci (x) = ^ C°r ■ 0 < i < ! oo for x —>- + co
X

and sine integral

X

Si (x) — ^ ‘21B-I <2^ — co <; x <C r o o for \ z\-+- -}- oo

6
There are various asymptotic expansions which can be obtain*
by means of some other elementary methods (e.g. see 131 or [61
Let us consider a simple example. In some problems connect*
with studying the propagation of electromagnetic waves near tl
earth surface and in some other problems we encounter the funot I*
X

F(x) = c-*2 (
0
This function can easily be expanded into a convergent power seri
(see § d, Sec. 2 of Chapter 8) but this scries cannot be convenient
:>9o MU LTI PLE IN T E G R A L S , FIEL D THEORY AND S E R IE S

used for large values ol' x. Lot us derive an asymptotic representation

of b' (x) for x — H-oo. Multiplying equality (U) by 2x amt applying
LTlospital’s rule twice for x —*■ -?-oo we obtain
lim Ixh' (x) — 1 (10)
X—
►;-C»C
Consequently. we can write the following asymptotic representation
for I' (x):
11 i o('l)] for x —►*- 00 (II)
where the symbol o (1), for x —►—oo. designates a quantity which
tends to zero as x —*■ -j oo. Instead of ( ii) we can also write
/' (x) » ^ for x -*» -i- oo (12)
There arc many other simple asymptotic expansions that are obtained
by means of elementary met bods but here we shall limit ourselves
to the above examples.

§ 2. CE NKKAL DICK! M U O N S AND T1 HCOiWC.US

We shall consider functions / (x), g (x), . . . defined on a point
set M belonging to the real x-axis. For instance, as a set M we can
lake a Unite interval, a semi-i ni'mile interval, the svhole x-axis etc.
I. Order of Smallness. Asymptotic Equivalence. We begin with
discussing the relations of the form / (x) = o (g (x)) and / (x) =
O (g (*)).
O r f i n it ion 1. If
lim f U) = 0 (13)
x-+xo .*(•0
x£M
we say (hat. f (x) i s o f ft it / h e r o r d e r o f s o m f In css t h a n g (x) on
th e s e t M f o r x —►x0 and write
f (x) = o (g (x)) for x —>■Xq on M {I \)
Mote. Itelntion (13) means that for every e ]>» 0 there is 6 =
— (e) > 0 such that
| / (x) | < e | g (x) | for all x £ M which satisfy the inequality
| x — x0 | <C 6 (1-0
I)'>f i n it i o n ?. If there is a constant C , t) <C C < -r°°, such that,
for off ,r r M belonging to a sufficiently small neighbourhood of .r9
the itfeiiuulity
I]_[x±
<C (1*0
I tfO)
 S U P P L E M E N T 1. ASYMPTOTIC E X P A N S I O N S 59 I

holds ire say that j (x) is o f the o r d e r o f g (x) on th e set M f o r

x —*-x0 and write
f (x) — O («■ (x)) for x - > x0 on M (lt>)
D e f i n i t i o n *2j. / / inequality (lo) is fulfilled for all x £ M um
e say
that f (x) is o f the o r d e r o f g (x) on the s e t M and mite
f (x) = O {g (x)) for x C -V (10')
If / (x) -- u{g (x)) lor x x0 on .1/. Definitions I and 2 indicate
that we c<iiialso write / (x) — O (g (x)) for x — x n but of course
the relation / (x)— O (g (x)) does not imjily, in the general case,
that / (x) - o (£(x)).
A’o/r. If M is an unbounded sel we can .similarly define the rela­
tions J (x) =- o (g (x)) and / (x) — O {g (x)) for x oo (or
x —*- —o o or |x| -*- o o ) .
In concrete problems the structure of the corresponding set .1/
may be obvious a ml then we may not indicate the set d/ when writing
the above relations.
/Examples
(1) ex — 1 — O (x) for x 0,
(2) sin x = () (x) for x -► 0,
(3) cos x — 0 ( 1) for x - - U,
(1) sin'2 x — o (x) for x - 0 ,
(5) x~ o (x) for x 0,
(t)) x-O (x) for x —►U.
(7) c~xZ —o (x l) for x - *“ oo,
X

(S) /•'(x) e xl j* e*2f/‘“ —O ^ j for x —*- ■ oo,

o
Now let us formulate the definition of asymptotic equivalence.
D e f i n i t i o n .7. We say that two functions j (x) and g (x) are
(tist/rn p to ficn ifi/) e i/tr im te n t (e q n n /) f o r x x t} on the s e t
M and write
f (x) ~ g (x) for x - * - x 0 on M (17)
if
I for x-*-x0 on d! (IS)
It is evident that instead of relation (17) \vc can write the relation
/ (x) — g (x) 11 i o ( l) l for x x0 on M (Hiy
592 MULTIPLE INTEGRALS, FIELD THEORY A ND SERIES

which is called an asymptotic representation of / (a:) in the neigh­

bourhood of the point x0 on the set M .
'Hie notion of functions asymptotically equivalent on an unboun­
ded set M in the neighbourhood of the point at infinity (i.e. for
x —*- -?-oo or x —►—oo or |x | — »►oo) is defined similarly.
Examples
(9) sin x ~ x for x — 0,
X

(10) F (x) -= e~x' j e*2 d£ ~ for /-> "j-oo.

0
2. Asymptotic Expansions of Eunctions. Opening the brackets in
relation (19) we get the equality
/ (x) ~ g {x) — o (g (x)) for x Xq on M (19')
which is a simple asymptotic expansion of / (x) in the neighbourhood
of x0 on the set M .
We now proceed to formulate the general definition of an asympto­
tic expansion which also embraces the special cases considered in
§ 1 of Supplement 1. We shall begin with the definitions of an asymp­
totic sequence and asymptotic series.
JJefru it i o n /. A finite or infinite sequence of functions {<(n (x)}
defined on a set M is called an a s i / m p t o t i c s e q u e n c e on M f o r
x —►x0 (x -v oo) if the relations <p„+i (x) = o (<pn (x)) for x —>- x0
(x -► oo) hold for all n = 1 , 2 , . . . .
Here arc some examples of sequences which are obviously asynip
totic:
(1) 1, x, x% . . . . xa, . . . for x 0,
(2) 1, x?,7 x*-, . . x>n . . . (0 < ?w, <; y.L• < ; . . . <c
Xn «< . . .) for x — U,
(3) 1, (x — Xj|). (x — x0)2, . . . » (x — x0)n, . . . for x -► x0,
(4) 1, x-?'i, . . ., x"?'«, . . . (0 < >., < < . . .
. . . <C <C . . .) for x —►—oo,
(5) ex, exx~^y exx~>J2, . . . e*x~>n, . . . (0 <C X, <C a2 <C . . -
for X — OO,
(l>) 1, x -1, x -2, . . x “n, . . . for x —*- -}-oo.
Sequences (1). (3) and (<i) are referred lo as power sequences and
sequences (2), ('i) and (5) in which X,; are real numbers that may not
be integers are called generalized power sequences.
 S U P P L E M E N T 1. ASYMPTOTIC E X P A N S I O N S sn

D e f i n i t i o n .>. / / {(pn (x)} is an infinite asymptotic sequence (Oi

-r<*>
the set A!) for x —►x0 (x —b -f-oo) the series 2 cH(r„ (x) with arbitral',
«-=i
constant coefficients cz, - * •, cn, . . . is said to be an a s y m p t o t i c
s e r i e s (on th e set M) for x x0 (x — t o o ) .
D e f i n it i o n . <>. Let {cp„ (x)} be a finite or infinite asymptoti
sequence on the set M for r x 0 (x —*- c o ) . If a function f (x) defma
on M satisfies the relation
N
f (*) — 2 (x) -}- o (<p.Y (x)) for X - V X 0 (x-^ oo) (20
h —1
where aj, a2. . . . » a x are some constants this relation is termed tin
a s y m p t o t i c , e x p a n s i o n o f f (x) on M f o r x — x0 (x —►cx>) nj
to th e : \ f h t e r m i n c f u s i v e irith r e s p e c t to the sequence
{Tn (*)}•
Kxpansion (20) can also be written in Ilie form
iv
/ ( x) j^ 2 a*<P* (*) for x-»-x0 (x->-oo) (21)
If asymptotic expansion (20) holds we can obviously write <lo\vr
the asymptotic expansions which are obtained from (20) by sub­
stituting k = 1, 2, . . . » A 1 for Ar.
- oo
D e f i n i t i o n 7* Let 2 (*) l)f> an asymptotic series on the
*= i
set M for x — x0 (x oo) and let asymptotic expansion (20) he ealio
for a function f (x) defined on M for every N — 1, 2, M, . . Then
this series is called an a s y m p t o t i c (ex p an sio n of' the f u n c t i o n
f (x) f o r x - ►x0 (x -+- oo) on M and. tee write
-1-00

/ (x) ^ 2 (x) f°r x~+x0 (x — oo) (22)

><=i
The asymptotic expansions considered in § 1 apparently satisfy
the conditions of Definitions 0 and 7.
It should be noted that an asymptotic expansion in the neigh­
bourhood of the point, x = 0 (x — x0 =f= 0) can be reduced to an
asymptotic expansion in the neighbourhood of the point at infinity
and vice versa by a change of variable of the form z = (z =
(' \ but this may not be convenient in some practical
—------l
x — x0 ;
problems.
Let us discuss the difference between an expansion of a function
/ (x) in a functional series convergent to this function and its asymp-
3 8 -0 S 2 4
 594 MULTIPLE INTEGRALS. FIELD THEORY AND SERIES

lolic expansion, in the former case the difference between / (x)

and the nth partial sum of the convergent functional series tends
to zero for every fixed x as N —►
- oo. In the latter case the difference
N
/ W — S a/«rPfc (x) tends to zero, as x-*- x0 (x-*- oo), for every
k—1
fixed A\ its order of smallness beingliigher than that of the last term
in the partial sum.
The examples of asymptotic series considered in § 1 show that
an asymptotic series may be convergent or divergent. Moreover, if
an asymptotic series for a given function / (x) is convergent this
may not imply that its sum coincides with / (x). For instance, we
have the following simple asymptotic expansion for the function
/ (x) = e~x:
c~x 0 *1 ~p 0 -x"1 -j- . . . -j~ 0 'X~n -f- . . .
(for x 4-oo) (23)
and asymptotic series (23) is convergent but its sum is unequal to
e~x for all x.
For practical purposes it is important to estimate the error arising
when / (x) is replaced by the A'th partial sum 2 ^ (x) of its
h=i
asymptotic series (22), that is to estimate the remainder term
(cp v (x)) in expansion (20) for x —>• x0 (x oo).
Analytical estimation of a remainder term is often connected
with considerable difficulties and therefore, in problem solving
practice, we usually try to use various computational techniques
based on simpler methods, which prove to be sufficient in many
cases. For example, suppose we know that the absolute value of the
remainder | o (<p,v (x)) | tends to zero and is a monotone decreasing
function for x x0. If wc manage to compute the value of / (x)
at a point x lying sufficiently close to x0 and if it turns out (hat the
N
difference between this value and the value of \ (x) at the
k~ t
same point is less than e >> 0 (in its modulus) we can conclude that
for all values of the independent variable lying closer to x0 than x
N
the modulus of the difference / (x) — ]>] (x) remains less
*= i
than e. A similar situation occurs when we have an inequality of
the form 1 o (q\v (x))| ^ ' l \ v (x) where i|)y (x) is a monotone decreas­
ing function which tends to zero for x —► - x 0.
If there exists an asymptotic expansion of a function / (x) with
respect to a given asymptotic sequence* {<pn (x)} this expansion is
uniquely specified by / (x). Namely, wc shall prove the following
theorem.
 SUPPLEMENT 1 ASYMPTOTIC E X P A N S I O N S 595

T h e o r e m 1. Let every member of an asymptotic sequence

{tfa C*1)} be different from zero for all x belonging to a sufficiently
small neighbourhood of x0 {or for x — oo) and let an asymptotic expan-
sion of form (20) hold for a function f {x). Then the coefficients a*t
k = 1, 2, . . . . A*, of this expansion are uniquely determined by the
form u las
nr 1
/(•*■)— y, tffc<rh w
an = lim ------------------- forn=*1,2, . . . , N (24)
.t-+ X (i T n \x )

Proof. Hcplaci ng N by rc<iV in relation (20) we put il in the

form
n—1
/(*)= S + (?«(*)). 1 n N
h—1
and lind
n—1
/(*>— y OAfM(x)
fe=1________ ■ 0 (?#»(*)) I < n < ;V
—
«fn (*) Cz> ’
which implies the validity of formulas (24). The theorem has been
proved.
It should be noted that the converse of the above theorem is not
true. Namely, a function / (x) is not uniquely determined by its
asymptotic expansions, that is there may exist different functions
having the same asymptotic expansion. For example, the functions
/ (x) =.= e~x and g (x) = 0 have the same asymptotic expansion (2'1)
with respect to the sequence 1, x"1, x"2, . . ., x '\ . . . for x —*•
-f-OO.
D e f i n i t i o n S. Two functions f (x) and g (x) arc said to be
a s i / m p t o t i c a l l // e q u i v a l e n t ( e q u a l ) with respect to a given asym­
ptotic sequence (<pn (x)} for x — Xq ( x o o ) if the relation
f M — S {x) = o (((V (x)) for x —►Xo (x ► - oo) (25)
is fulfilled for all n.
It appears obvious that two functions / (x) and g (x) possessing
the same asymptotic expansion with respect to an asymptotic
sequence are equivalent (relative to this sequence).
We can easily show that for two functions / (x) and g (x) to have
the same coefficients of their asymptotic expansions with respect
to one and the same asymptotic sequence {cf„ (x)} for x —►x0
(x —y- oo) il is necessary and sufficient that these functions he asymp­
totically equal with respect to the sequence {q»n (x)} for x -> x0
(x — oo).
38*
596 MULTIPLE 1N TE G R A 1 S . FI EL D THEORY AND SERIES

We now discuss some operations on asymptotic expansions.

If we have asymptotic expansions
-f-oo -lo o

/(x )» 2 (x) and g (x) » 2 bh<ph (x ) for x —►*o (x —►oo)

f t= ! h = l

(20 )
we can obviously write down the asymptotic expansion
-•-OO
af (x) H- fig (x) « V. (aah 4- p6h) ip>t (x) for x - v .r0 (x co) (27)

which is valid for any constants a and P.

In the general case it is not allowable to multiply the asymptotic
expansions of two functions / (x) and g (x) with respect to one and
the same asymptotic sequence {cpn(x)| because it may turn out that
the products q>m(x) <pH(x) cannot he arranged into an asymptotic
sequence.
Let us dwell on the question of term-by-term integration of asymp­
totic expansions.
T h e o r e m ?. Lei a sequence {cp„ (x)} of positive functions of
a real variable x defined on an interval a < x <C b be an asymptotic
sequence for x -► b — 0. Suppose we have an asymptotic expansion
• oo

/ (x) ^ >] f7/.<p/t(x) for x~*-b — 0* (28)

h= \
If the integrals
b b
amt j<r* (!)<*!, fr -1 ,2 , . . . (20)
X X

are convergent the asymptotic expansion

b +» b
( / (£) dc ^ 2 Ch j V* (0 for x - * b — 0 (30)
x h — \ X

is also valid.
Proof. The relation <pn+i (x) = o (<pn (x)) means that for every
e > 0 we have the inequality | (x) | < e | q>n (x) | provided
that n is sufhcicntly large.** The functions rp;< (x), k - 1 , 2 , . .
being positive, we can drop the sign of modulus and write
<P„-f i ( * ) < (*) for x - + b — 0 (*)

* Here b i^ a finite number nr -f-oo.

** See the note after Definition 1.
 SUPPLEMENT 1. ASYMPTOTIC E X P A N S I O N S 5 9

for any e > 0. Integrating (*) from x to 6 (which is permissibl

since the integrals are convergent) we see that
b b
0 < j (p,, +1 (!) ds < e ^ <fn (f) c/b <31
X* X
b
for any e > 0 and x-*-b — U whence it follows that
X
is an asymptotic sequence for x — *~b —0. To prove relation (30
we must show that
b jv b
j / (!) d\ « ^ Ch j <('* (I) fife for x - > 6 — 0 (30'
X k — \ X

for every A' = 1, 2, 3, . . . . By tlic positivity of the function:

<Pn UT n = 1, 2, . . ., relation (20) (whose validity is implied In
(28)) can be rewritten in ttic form
N
!/(* )— (x) | < e<pv (i) for x —^b — 0
h-=\
for any e 0 where Af = 1, 2, 3, . . . . Consequently, we ha vi

and any s > U for every A' = !, 2, 3, . . . . But this exactly

means that asymptotic expansion (3U) holds for every A^ — 1. 2,
3, . . which is wluil we set out to prove.
The assertion below is a direct consequence of Theorem 2:
/ / we have a power asymptotic expansion
- f oo
f (x) ^ afti"'1 for x + oo (32)
h~0
+oo
and the integral j [/(£) —a0 — dt converges, the asymptotic

expansion
+w -f-00
j l/(£ )—#0— V aic (33)
_A- + 1
X h-2
is also valid.
598 MULTIPLE INTEGRALS, FIELD THEORY- AND SERIES

Iii the general case an asymptotic expansion cannot be differen­

tiated termwise. I3nt there are some special cases of asymptotic
expansions when term-by-term differentiation is permissible. For
instance, let f (x) have a power asymptotic expansion of form (32).
Suppose that its derivative f'(x) also possesses a power asymptotic
expansion in which there are no terms containing x 1 and x ' 1:
a? 2<v?
r w .r* x:i i'1' 1 for x -f- oo (31)
Integrating (31) we obtain, by the foregoing assertion, the relation
a? . a* “n for x — (35)
/ ( * ) « / ( -j- oo + X>*
o o

Hut since the function / (x) uniquely specifies its asymptotic series
with respect to the sequence 1, x -1, x “-, x~;\ . . x~‘\ . . .
expansion (35) must coincide with expansion (32) and thus we have
the equalities / (-i-oo) -= «0, — a ,, . . = a/£................ Sub­
stituting the above values into (31) we obtain
/'(*) x nri for x — oo (31')

where the latter asymptotic expansion ran he directly obtained

from asymptotic relation (32) by termwise differentiation.
We shall limit ourselves to this short review of some general
properties of asymptotic expansions. In § 3 we shall describe an
important method of construe ling asymptotic expansions of some
integrals.

§3. LAP LACK METHOD TOR DERIVING ASYMPTOTIC

EXPANSIONS OF SOMP INTEGRALS
Let it be required to derive an asymptotic represenintion of an
integral
t,
J (/) ^ ^ / (x, t) dx for t —►d- oo (30)
a
under the assumption that, for large values of /, the integrand has
a sharp extremum in a neighbourhood of a point x = x0 anil that
its modulus is very small outside this neighbourhood. Then it may
turn out that the integral taken over this neighbourhood of the point
x0 will he almost equal lo the integral taken from a to b for large
values of f. If, in addition, it is possible to replace the function
/ (x, /) in this neighbourhood, to a sufficient degree of accuracy,
by a simple function which can easily he integrated. ;*m! if the
difference between the original integral (30) and the inlegral of this
auxiliary function tends lo zero as t ! oo, wo can thus derive
 SUPPLEMENT 1. ASYMPTOTIC E X P A N S I O N S .*>M

an asymptotic expansion of integral (3G) for large values of t. This

is the general idea of the method suggested by Laplace for deriving
asymptotic expansions of integrals of the above type.
As an illustrative instance, we shall apply this method to deriving
an asymptotic expansion of the gamma function
-Lco
C du (37)
o
for t — H-oo. The asymptotic representation which we shall thus
obtain for T (t -r I) for large values of t yields, in the case when
t —n i) is an integer, the so-called Stirling* formula which
gives an asymptotic representation of /?! for large \ a lues of n.
Performing the substitution u = t (1 — x) we reduce integral
(37) to the form
^\-rx> -f-o©
r ( < + l ) = e-,i'*1 f |e~*(l -j-a-)]‘ d x = e 'll M ( ethw tlx (3.S)
-1 -1
where h (x) = —x -j- In (1 x). The function (1 - j - x) =
. ,,-x| iii(i ;*> = t(ains its maximum at the point x = 0 at
which h (x) has its only maximum. Indeed, the derivative h'(x) —
- 1 1—■ — is positive for -1 < x < 0 ami negative for
*■“ ■!>
0 •< x <C t ° o . At the point of maximum of h (x) we have h (U) — 0
and hence the function h (x) is negative for all the values of x belon­
ging to ihc intervals - 1 < r < <t and 0 <c! X . o^. Therefore,
(lie function c,,,l<v) t e n d s to zero ns t —>■ -|-oo for — I < x <C 0 and
0 <C X < ;-oo. Conse(|ueiilly, it is advisable to try to apply Lap­
lace's method.
I'aking a sufficiently small 6 > 0 (6 < 1) we represent the integral
m qiif.st-ion as the sum
1ac -6 0 -!-•»
1 gii>(x) __ f etft(x) tfx _i_ f C eift^ d x (31>)
J •» *
-1 -I -6 i>
To estimate each integral on the right hand side of equality (3!))
we investigate h{x) on the intervals — I < i ^ —ft, —ft ^ x ^ ft
and 6 ^ x <C -Loo. Expanding In (1 x) into Taylor’s series in
the interval —f t ^ x ^ f t we obtain
x- r'f r4 z5
h (x) = — ♦>
M •/ ■ o
oyri »
2 L '3 i - v - J

* Stirling. Jamo« flCift^-1770^ nr> T**i»: \: lu.illn u.a! LCiali.

GOO MULT IPLE INTEGRALS, FIELD THEORY AND SERIES

The series in the square brackets satisfies Ilie conditions of Leibniz’

test because | x | < c A < C l . Therefore if we discard all the terms
beginning with — — this results in an error whose modulus does
2 x2 2z2 0 Ir ]
not exceed ~r-
4 . Tor sufficiently" small 6 > 0 we have -H
7- <
O
for all x belonging to the interval —6 ^ x ^ 6 . Consequently, we
can write
—I-|-~ 6 j for — 6 < i < 6 , -y6 < 1

m
The function k (x) increases on the interval —1 ^ x ^ —6 and
assumes its greatest value h (—A) < 0 at the point x —6 . On
the interval A ^ x <C * 0 0 the function k (x) decreases and its
greatest value h (A) < 0 is attained at the point x — 6 . Thus,
h (x) ^ h (—6 ) < 0 for —1 < 1 ^ 6 ('ll)
and
h (x) < h (A) < 0 for A ^ x < -|- 0 0 (42)
Now let us estimate the integrals on the right-hand side of (39).
By inequality (41), we have
—6 -ft
f ««*<*> f ^',<-fi)dx = ^ ( - 6 )(l->6)---<9(e^-6))-^0 (43)
• •’
-1 -1

for t \ 0 0 . Furthermore, for A ^ x <C -Too and / > 1 we have,

by (42). the inequalities th (x) th (A) and th (x) h (x). Adding
them up we get the relation
th (x)<^ — [th (6) I h (x)| for / > 0 and A^x<C {-00 (44)
Therefore
4“00 -00 *
j e * » d x < j =
6
I •!-*» 1 hix)
■—c y i<6) r J \ e- dx = 0 ( e > iH6y) ^ 0 (45)

for t —►+ 00 . Finally, by inequality (40), we can write

J ,e - 4 i ,+S*i( f a < j dx (4f»)

-6 -6 -A
 S U P P L E M E N T 1. ASYMPTOTIC E X P A N S I O N S GC

On the basis of estimates (44) and (45) we obtain the relation

(47
1
for t —►-j- oo where a (6) is n positive constant dependent on
and independent of t. i\ow let us evaluate the integral
-f-oo
\ dx

i
Performing the change of variable | = r f 4 - ( l ± -g-6j I" weobtair

j dx = (2xft~ * ( l ± 2 (48;
— OO

From (47) and (48) it follows, for / —^-J-oo, that

A x2 4 1 1 1_
f c- , T l ,±5*)dJ;_ ( 2 n ) 5 r S ( l ± - g - 6 ) ’ !!-|-<?(e-««»') (V.I)
-6
It is obvious that
_i i
( l t - 5 ) 2 = 1 _ f)(-6) and = = 1 : r 2 (6 )

where £| (fi) and e2 (6) are positive and tend to zero for 6 -► 0.
If 0 is lixcd we h a v e , for sufficiently large I ;> 0 , the inequality
1 _2
| O (e~°!<W) < min (6| (6), k2 (<*>)) ( - 71)2 { 1
since the quantity —a (fi) is negative. Therefore from (46) and (46)
we deduce the relation
i _i 6 I
(2n f t 2 |l — 2e, (6)1 < C e‘*<*>dx< (2 n )'V 2 |1 + 2 e , (6)1 (50)
-4
for sufficiently large / ;> 0. Taking into account relations (43)
and (15) and tlie fact that the inequalities
i _i
| O (e'M-61) | < {2 jif t 2 min (e, (6), e, (6 ))
and
l _i
| () | <; (2ji)“ / 2 rnin (6), (7))|
002 MUI-TIPI.E I N TE GR AL S , 1'IKI.D T H E O R Y AND SERIES

hold for sufficiently largo t ;> 0 wo arrive at the inequality

1 -I 1 _i
(2n f t 5 (1 — 3 e ,(6 ))< f e"‘Wf7x< (2n)z t 3 (1 ->-3e.(6)) (51)
41
-I
which holds for all sufficiently largo t ;> 0. Consequently, since
c t (6) and t*2 (6) arc arbitrarily small if 6 is sufficiently small we
can write, taking into account, the definition of the symbol o (1),
Iho relation
-t-» i _
C eiti(x) ~ (2ji)~ t 2 11 ! a (1)1 for / —»- ~r oo (52)
-1
Substituting (52) into (38) wo finally obtain the souglil-for asymp­
totic representation of V (t p 1) for /-*- -foo;
,. i i
r(/-M ) -■<?-</ ‘- (2n)J[1 - 0 (1 )1, (53)
Lulling t — n whore n is a positive integer wo derive from relation
(53) Stirling’s formula
,,4 -1 r
«! = e“"« 2(2n)" 11-}-o (1)1 for n — (54)
which is widely used in mathematics and its applications.
The Laplace method makes it possible to obtain asymptotic
expansions of a more general type. It can also be applied to multiple
integrals (see 131). Laplace’s method is also generalized to thucarw
oT integrals of functions of a complex variable. This generalization
is known as the saddle-point method which, like Laplace’s method,
is used in various divisions oT mathematics and mathematical phy­
sics. On ilie Saddle-point method we refer the reader to 13! and 112!
 SUPPLEMENT 2

On Universal Digital Computers

This supplement provides ail introduction to modern digital

computers, their operation and use. It cannot he regarded as a
systematic account of the computer theory and programming me­
thods, and for greater detail we refer the reader to special books.

§ 1. COMPUTERS
1. Introduction. Many problems of modern science and engineering
require extensive calculations to obtain results of practical impor­
tance. The amount of work may he so large that the calculations
either cannot he carried out manually or lake so much time that
the result becomes useless. For example, it makes no sense to fore­
cast the next day’s weather by applying a method that lakes a mouth
of computational work.
The number of problems requiring large-scale calculations as well
as prompt answers has recently increased in the light of such tech­
nical needs as automation of manufacturing processes.
Some technical devices which facilitate compulations were inven­
ted long ago hut the last two decades are connected with a radical
turn in ill is held owing Lo the appearance of high speed coni puling
machines based on elccLronics.
The new technique lias achieved a marvellous success in a short
time. Modern computers can do hundreds of thousands of arithme­
tical operations per second. This makes it possible to succeed in
solving such problems that could not even be slated before.
Computers not only increased the range of mathematical appli­
cations hut also influenced the development of mathematics itself.
In matiiematicaI logic and numerical analysis there have arisen
new problems and mnv trends. The computer theory and programming
(see § 3) are very important fields of modern mat hemal ics.
Nowadays computers are used almost everywhere. Therefore
specialists in various branches of science, physicists in particular,
should he lam iliac with the mode of operation of those machines,
with their properties and capabilities.
2. Basie Typos of Computer. Computing machines are. divided
into two basic classes: machines of discrete nnoraiien referred w*
(JU4 M U LT IP LE INTEGRALS, FIELD THEO RY AND S E R IE S

as digital and machines of continuous operation called analogue.

A digital computer operates with numbers represented in a positio­
nal number system. An analogue computer represents variables by
means of some physical processes and quantities (such as electric
currents, voltages, mechanical displacements and so on) iliat may
vary continuously. In analogue computers only the final results
take Ihe digital form. Analogue computers are widely applied
(mostly when high accuracy is not required) hut in modern compu­
tational mathematics they are of less importance than digital
machines.
In what follows we shall deal with digital computers to which
the general term “computer” will he applied.
There are two types of digital computers: special purpose and
general purpose. Special purpose computers are designed to solve
a specific class of problems. A computer is said to he general purpose
or universal when it can be used to solve a wide variety of problems,
arithmetical as well as logical (for instance, the translation of
languages). It is the universal computers that we are going to con­
sider here.
Versatility is a great advantage of universal computers. However,
it should ho noted that the computer itself can only perform a res­
tricted number of elementary operations (see § 2), and therefore
to solve a problem on a computer wc must reduce it to a sequence
of elementary steps, that is a routine corresponding to the problem
must he prepared for the machine. The process of preparing a routine
(programming) may be rather complicated. Some general notions
related to programming and examples of elementary routines art;
given in § 3 of this supplement.
3. Principal Components of a Computer and Their Functions. As
has been mentioned, every computer can perform some elementary
operations, arithmetical and logical. Furthermore, facilities must
be provided for performing some additional operations, namely
for entering initial data and instructions into the machine, storing
this information and intermediate results and withdrawing final
results from the computer. Accordingly, every universal digital
computer includes, irrespective of its construction, the following
functional units:
1. Input unit.
2. Storage unit (memory).
3. Arithmetic unit.
A. Control unit.
5 . Output unit.
A block diagram of a typical computing machine system is shown
below.
 S U P P L E M E N T 2. ON U N I V E R S A L D I G I T A L COMPUTERS oo:

Direction of information flow is shown by continuous arrows am

that of control signals by doited ones.
Let us briefly consider the function of each unit.
1. Input is a device used for transferring initial dola and instruc­
tions into a computer. Information must be prepared in a forn
intelligible to the machine. In general, it processes some physica
medium such as magnetic tape, punch tape and punch cards.
On a magnetic tape data are stored as small magnetized spot?
arranged in column form across the width of the tape.
On a punch tape and punch cards storage is in the form of hole:
punched at definite places. A punch card is shown below. Suel
cards are widely used.

1 1 II III II 1 1 1
> 1 1 *4 . V-j * »* :* x .** O «• <4 1 1 1 II♦: ^ V 1M — i II„> -i ■)
50 '.7 -» it.
i ;Tvf3e63cojoJ#P2C«jcfcti<mjo«c#aaoi»Jnc!iJ.of. fljir: o«11:'
: . ; . • i . > >. ii i t 11 i I i i : m i '• i : i : n n 11 11 n n t i 111 m : i > u 11 ii 1111 111111 i : t 11: n i ; i

j 31Jn j Ji j j : J: j s: |>.»j j ms t ,tt j i| - >n mu ^| | * j t «»»:: i! ?»rrjr ’ *' - ’ ;

llll
6MS£6>C>f:!iTt(i>vit{<<6&l6MiSf£CGt(i$tS£6CCt(b35;(!j6(!<6C566»C&til9Lt.Av', otiSltiSSi
7??777»7rmjlZ7M7?l7T777JD77??7T7J7J7J77J3J7/7M7T7;777777JJ?/Mr/?M/;j37777
uiuseitim tM M iiitsuiiim uM m eam uiiK ttm iiiH uuuini'im ett
>i « ( t n t; :i »i u rj r? .t •) i i m s * ® c « « «i * a x si a a v w « >* n » » s s »
S 3 9 r 3 M 3 3 < ; « » f , t n ? 9 5 S 5 3 9 ; 5 9 9 j ; } S 3 3 J 5 M 3 J 3 3 9 9 3 3 S 9 S J 9 9 J 9 7 3 ) J 5 3 3 9 J 3 M n ,n ,n 9 $ 9 9 9 0 9 »

2. Storage* (memory) is the section in which instructions, initia

data and intermediate results arc stored and from which they car
be taken. A storage is a set of locations, cells, each of which is iden
titled hv a number. Kach location is of a fixed digit capacity ant
can be used for storing numbers as well as instructions. As is seer
from the block diagram, the storage unit is directly connected wit!
the other units. From the input device it accepts the initial dalr
and the set of instructions (the program) corresponding to the pro
blein which is to he solved. The storage sends numbers to the nrith
GOG MULTIPLE IN T E G R A L S . FIEL D THEORY AND SERIES

iiielic unit whore arithmetical operations are carried out. The results
conic back to the storage.
Complicated problems require storage, devices of great capacity.
At the same time it is important that the stored data could be rapidly
read out. To satisfy both conditions storage devices are usually
made of two blocks: internal and external memory. An external
(auxiliary) storage bolds much larger amount of information than
the main, internal, storage hut its access time is greater. The external
storage is not connected directly to the arithmetic unit; if necessary
it feeds information to the internal storage and only Ihen this infor­
mation can be processed by the arithmetic unit,
3. Arithmetic unit is a part of a computer in which arithmetic
operations are performed. The set of these operations will he spe­
edlied below.
4. Control is the section which interprets the instructions involved
in a routine (i.e. converts them into pulses) and theu sends appro­
priate signals to other machine blocks. The control unit determines,
in accordance with a given routine, the operation of all other pails
of the computer.
Since no human operator can achieve the speed at which a com­
puter works a control unit must be automatic; this is one of lhe
main principles of computer organization.
5. Output unit is meant for accepting, in a suitable form, the
solution of a problem ami some intermediate results that are of
interest. A few output devices are card punches, paper tape punches.
p riiiler> and m a g n etic tap e unit.**.
4. .Number Systems Used in Computers. When we operate on
number'-- (no mat lor whet her we utilize a computer or not) we have
to represent them in a certain system of notation, i.e. in a number
system. Tho system practised on the greatest scale nowadays is the
decimal system in which any numerical quantity is represented
as a sequence of coefficients in the successive powers of ten. For
example, the decimal expression 2548 denotes the number
2 *10:s -{- 5-1U2 4 -101 8-lb°
The decimal number system uses the digits 0, 1, 2, 3, 4, 5, 0, 7,
8, and 1); operations on n u m b e r s comply with the well known rules.
Any other positive integer except unity may also he used as
a base of a number system.* Logically, the simplest is the binary
system in which every number is represented as a combination
of powers of two. For example, 13 — 1 *23 j- 1 -22 ••!- I)-2l -f* 1 *2°

* Indeed, there were nations that used non-dccimal number systems. From
ilie mathematical point of view the decimal system has no special advantages,
and its general uso stems from the fact that rnnn has ten fingers.
 S U P P L E M E N T 2. ON U N I V E R S A L DIGITAL COMPUTERS (ill?

mid lienee llie decimal number “thirteen” is represented os the binaiw

number 1101. This notation uses only two digits 0 and 1, tho number
two being the unity of the next (to tho left) position.
The reader might well ask at this point which system of notation
is the best suited for electronic computers. Note that if wo use a num­
ber system with base p the corresponding digits can have one of the
values 0, 1 . 2 .......... p — 1 (for instance, the decimal system uses
1U digits, tho binary system uses 2 digits etc.). In order to be able
to lix p different digits a machine must involve some devices that
have p stable stales, each representing finite digit. The speed
with which a modern computer works (as a rule, hundreds of thou­
sands of operations a second) makes it impossible to use any mecha­
nical device for fixing numbers. On the other bund such speed can
he easily achieved by electronic circuits which are practically
inertia less. Most electronic elements (valves, transistors etc.) are
bi stable. For instance, a valve assumes either of two stable states:
“on” (the current flows through it) and “off” (the current does not
flow through it). Owing to these features of electronic devices, the
binary number system proves to be the most applicable for modern
calculating equipment.
The binary system has also the advantage of simpler arithmetic.
For example, the “multiplication table”, in the binary notation,
consists of the following lour items:
0-0 - 0, 0-1 - U,
1 .0 ^ 0, 1 -1 _ t
The binary system has some disadvantage because it requires
converting initial data written to the base ten to tho equivalent
numbers VM'illeU to the base two and converting computed results
hack to the decimal form. This operation is however not complicated
and may easily lie automatized.
Besides the binary system, computers also use the octal system
(based on a radix of eight) and the binary coded decimal notation
(binary-decinial system). In the latter a number is fust written
in the decimal notation and then each decimal digit is represented
by the corresponding binary number. For instance, the number
538fi
is represented in the binary-decimal notation as
0101, (HIM, 1000, 0110
It is clear that each decimal digit (i.o. the digits 0, 1. 2, . . . . 0)
can he represented by a four-digit binary vintuber.
There are computers with elements having three stable slates (for
example, the current Hows through an (dement, the current flow's
008 MULTIPLE INTE GR AL S, FIELD THEORY AND SERIES

m the opposite direction and the current does not flow). Arithmetic
in these machines is based on the ternary number system.
5. Representing Numbers Within a Computer. Any computer ope­
rates with quantities having a definite (for each machine) number of
digits. If a number is shorter than tlie location in a given machine,
zeros are put to the left of significant digits. If a number is longer
it must be rounded off by deleting less significant digits. The number
of binary places in each location limits the precision of representing
the results and thus restricts the accuracy of compulations.
Since we deal with negative quantities as well as with positive
ones, the computer must contain some means of distinguishing
between them. A certain binary position is usually assigned for
this purpose and the codes U ami 1 are interpreted, respectively, as
positive and negative signs of the quantities.
Moreover, since we have to deal with mixed numbers the machine
must he able to separate integral and fractional parts by a “binary
point”. The position of the radix point may either vary in the course
of calculation (floating-point machines) or be constant (lixed-point
machines). In the latter case the integral part of any numerical
quantity is expressed by a predetermined number of digits. All
quantities occurring in a problem to he solved on a fixed-point
computer must, he converted into the desired range of magnitude
by means of ’•scale-factors'1 specific for each problem. This makes
fixed-point, machines less convenient than floating-point ones;
however, their logic and hardware are simpler.
§ 2. BASIC OPERATI ONS EXECUTE 1) RY A COMPUTER.
INSTRUCTIONS
1. Types of Operation. We have already mentioned that any com­
puter is designed to perform a certain restricted number of basic
operations. Although this number can be *-=»i 11 reduced, it. would
be inconvenient for the programmer and user. On the other hand,
the larger the number of different elementary operations, the more
complicated the construction of the machine. A compromise must
therefore be reached between the claims of the designer and of the
programmer; the point at which the balance is struck varies from
one machine to another, every computer still being able to carry
out the following operations:
(1) arithmeticai operations;
(2) additional computational operations;
(d) logical operations;
(4) transfer operations (including conditional transfer of control);
(5) input and output operations.
One point must be emphasized at the outset. Wc assume all ope­
rands to be stored in the memory locations, each location being
identified with an address, i.c. a certain serial number. When
 S U P P L E M E N T 2. ON U N I V E R S A L D I G I T A L C O M P U T E R S 009

an operation on^two quantities is performed the following must be

indicated:
(1) the addresses ol' the operands;
(2) the command to be executed (addition, multiplication etc.);
(3) the address for storing the result.
Consequently, each instruction that causes a computer to perform
ari operation contains three addresses of the locations involved in
the operation and specifics the operation which is to be performed.
In other words, such nil instruction is of the form*

opera l ion 1st address 2ml address 3rd address

It is important to note here that it is not the operands but their

addresses that are indicated in the instruction. This enables us to
prepare a program (a sequence of instructions) without knowing
the specific values of the quantities to be dealt with.
Now let ns describe the basic operations.
2. Arithmetical Operations. Arithmetical operations arc of the
following four types:
(1) Addition. This operation reads: ‘Take the number stored in
location a, add it to the number stored in location P and store the
result in local ion y.” Symbolically:

A d d i I ion I*

(2) Multiplication. “Multiply the number in location a by the

number in location p and store the result in location y . n Symboli­
cally: ______________
Mult iplical ion a f) y

(3) Subtraction: “Subtract the number in location P from the

number in location a and store tlie result in location y .” Symbo­
lically:
i------------------
Snitl rai Lioa oc f* Y

* Kor Ihr: sake of simplicity, we consider computers referred to as three-

address machines. There exist, machines that involve one-, two-, or four-address
instructions but we shall not discuss them.
;e.j—u.sj-'i
010 MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

(4) Division: “Divide the number in local ion a by the number in

location p and store the result in location y ” Symbolically:

Division a 0 M
1

3. Additional Computational Operations. The set of these operations

may vary from one machine to another. Here are some examples:
(I) Maximum: “From two numbers stored in locations « and (5
take tlie greater one and place it in location y”, that is

M a x im u m a r> Y

(2) Minimum: is debited similarly.

(3) Magnitude: “Take the absolute value of the number in loca­
tion <x. and store it in location y”, i.c.

A b s o lu te v a lu e ot y

The latter operation (and some other) involves only two (and
not three) addresses.
The greater the number of various additional operations, the
easier the process of programming. Many modern computers have
built-in instructions for square root operation, sine evaluation ami
so on, although these operations are in fact reduced to certain com­
bi rial ions of « limited variety of elementary computing steps,
(see § 3).
4. Logical Operations. These operations on numbers are jrerformed
on a digil-by-dlgit basis without carry. A few examples will illu­
strate this type of operation.
(1) Logical addition. It is an operation in which the numbers
placed in locations a and P arc added bit-lo-bil*, i.e. each digit
in a is added to the corresponding digit in p according to the follow­
ing rules:
0 + 0 = 0, 0 - M - 1 , 1 + 0 = 1, 1 + 1 = 0
The result is placed in location y. We denote this operation sym­
bolically as

B i•-1o-l>iL addition a p M
1

* The term bit means binary unit (of information) or binary digit.
 S U P P L E M E N T 2. ON U N IV E R S A L D I G I T A L CO MP UT ER S Cll

(2) Comparison. This operation is concerned with the determina­

tion of similarity or dissimilarity of the corresponding digits in
the numbers stored in locations a and |1. If the digits in a certain
place in ct and are alike the result in this place at y is equal to
unity, if otherwise, the result is equal to zero.* For example, the
result of comparing the numbers
10 1 1 0 1 1 0 1
1 1 0 10 0 10 0
is the number
1 0 0 1 1 0 1 10
Let us symbolically denote this operation by

Comparison a p i

(3) Logical Negation. The operation is performed as follows: if

the number stored in location a contains zero in a certain position
the corresponding binary place in location y contains unity and
vice versa. Symbolically:

Logical negation a 1

The set of available logical operations depends upon the type

of a computer.
5. Input and Output Operations. These are the following opera­
tions: input, writing (transferring a number from the internal storage
lr> the external storage), reading (transferring a number in the reverse
direction), printing and stopping.
(1) Input instruction is denoted as

Input n—1 a -f 1

It means: “Transfer n numbers (or instructions) from the input

device (e-g. punch cards or punch tape) into n memory locations
a + 1, a -j- 2, . . a -j n."
(2) Print. This is an instruction of the form

Print, a. n—1

* Some computers use the opposite rule, i.e. 0 dcnolrs the fact of coinci­
dence and 1 of discrepancy.
30“
012 MULTIPLE INTEGRALS, FI EL D THEORY AND SERIES

which means “Print (in Ilie decimal system) the numbers stored in n
consecutive locations beginning with location a .”
(3) Stop. This is an instruction of the form

Stop

which causes the computer to stop working.

Kxlernnl storage devices are used in large-scale computations
requiring long programs and largo amounts of initial data. We shall
not dwell on them here.
(5. Transfer of Control. We have pointed out that it is not the
operands hut their addresses that are indicated in an instinct ion.
This enables us to plan Ilie whole procedure for solving a problem
before starting compulations. However, in a computational process
there may occur a situation in which the further course of compu­
tation depends upon the result we have obtained at a certain stage.
For example, if a quadratic equation is being solved the course of
computing depends on the sign of its discriminant. If some alter­
native courses of action must be lakcn depending on a relationship
obtained at a certain step the so-called operations of conditional
transfer of control (conditional jump) arc used.
The examples below illustrate this kind of operation.
(1) Transfer of control d e p e n d i n g o n relative value of two numbers
taken with their algebraic signs. This is an instruction of the form

When readied in the course of a program it causes the computer

to compare the numbers stored in locations a and 0. If the former
is greater than the latter the computer executes the next instruction
in the original sequence, if otherwise it goes to the instruction
stored in location /:.
(2) Transfer of control depending on relative magnitude of two
numbers. We denote lids distinction by
1
1 '\ jump a 0 k

and understand it in the sense of the previousjoperation with the

only difference that it is the absolute values of the numbers whic h
must be compared.
 S U P P L E M E N T 2. ON U N IV E R S A L D I G I T A L CO MP UT ER S 0

(3) Transfer of control depending on the sign of a number (ph

jump).
r.
Plus jump I 'x “1

This operation means: “If the number in location cc is positi'

execute the instruction stored in location /»j, otherwise execute ll
instruction stored in location A*a.”
The latter instruction may be used for the operation of uncoml
tional transfer of control to location k. For this purpose it is sufficic
to form the following instruction:

Plus jum p | rt k A-
i
This instruction transfers control to the instruction stored in loc
lion k irrespective of the number stored in location cc.
A computer performs any of the available operations after tl
corresponding instruction has been received- The instructions a
represented by binary numbers and stored in computer’s liiem oi
as well as initial data. The control unit interprets these instru
(ions, that is decodes them and applies the proper signals to tl
other parts of the machine.
7. Realization of Operations Within a Computer. From the stan
point of engineering every process of computing consists in co
veiling the corresponding pulses by electronic circuits. We a
not going to discuss this question at length, that is to go into part
culars concerning the elements neeiieii lor performing specific op
rations. We shall only limit ourselves to the operation of adriilk
of two positive quantities l>v way of illustration.
Operation of addition in the binary system, as in any other poi
tional number system, involves the bil-lo-bit addition and, win
necessary, a carry to the next higher digit place. The bit-lo-b
addition complies with the following rules:
0 | 0 -- 0; 1 -{- 0 = 0 -|- 1 = 1 and 1 -p I = 0
plus next place unity.
Let a and b be the digits we have to add together when perform!!
the operation of addition on a certain digit, place and let r be tl
carry from the foregoing place. To perform Iho operation of add
lion on a digit place means the following: given «, b, and c whit
may take one of the two values zero ami unity, it is required
find s (the digit to be written into this place) and j> (the carry
the next place). The variety of all the situations that may ari
here is c o n f i n e d to t h e following table:
014 M U LTIPLE INTEG RA LS, FIELD THEORY AND SER IE S

a 0 i 0 0 1 1 0 1

b 0 0 1 0 1 0 t 1

c u 0 0 1 0 1 1 1

s u 1 1 1 0 0 0 1

p u 0 0 0 1 I i 1

It follows that to realize the addition on one digit place the

machine must have a device with three values (a, b, and c) at its
input and two values (s and p) at its output. The operation of this
device must be in accordance with the above table. Therefore, when
voltages at the inputs are equal to zero the voltages at the outputs
do not exist either; when voltage is fed to one of the inputs the
output voltage is produced at s while that at p is equal to zero,
and so on.
Such a device is referred to as an adder. It can be easily realized
as a circuit composed of electronic valves or transistors but we
shall not discuss this question in detail here.

§ 3. ELEMENTS OF PROGRAMMING
1. Genera! Notions. A problem to be solved on a computer must
be expressed as a sequence of elementary operations which the
machine is able to perforin. Each operation is specified by the cor­
responding instruction, and the complete sequence of instructions
necessary to solve the problem is referred to as the program or routine.
Programming, that is preparing a routine, is one of the main stages
of computing. It is clear that prior to this stage an appropriate
mathematical method for solving the problem must be chosen and
specific formulas for computing should bo obtained.
The form of a routine depends on the numerical method chosen
for solving the problem (for instance, we may compute an integral
by means of the trapezoid rule, rectangular rule or some other for­
mula of approximate integration) and on the type of the machine,
i.e. on the set of operations it can perform. When the numerical
method and the type of the machine arc fixed the routine is not
yet uniquely determined because there is a variety of ways for
reducing the calculations to elementary operations. The selection
of the most suitable routine depends, to a considerable extent, on
 S U P P L E M E N T 2. ON U N IV E R S A L D IG IT A L C O M P U T E R S 6 15

the skill of the programmer. Here we shall not go into particulars

and coniine ourselves to some elementary examples.
2. Formula Programming. Problems involving formulas are the
simplest for solving on a computer. In this case programming is
reduced to representing Ihe formula as a sequence of elementary
operations in a reasonable way and to placing the corresponding
instructions and initial data into the storage. Let us consider a
simple example.
Example. Given x r it is required to evaluate the quantity y
expressed by the formula

5 x -f-1

The calculations can be represented as the following sequence of

elementary steps:
(1) A = 2 x , (2) A2 —Ai~{- 3, (3) 5x,
(4) Bz = B\-\~ (5) y =
To perform these operations on a computer we arrange initial
data in five storage locations (e.g. locations with numbers from
n -}-'l to n 5). This can be written down as
S to red S to red
L o c a tio n n u m b er L o c a tio n num ber
ti

n + 1 X n -f A 5
n -f 2 2 n -i- 5 \
« -H 3
3

Lot us place the instructions corresponding to the above operations

into some other five locations. We thus obtain the following sequence
of instructions:

1st a d ­ 2nd a d ­ 3 rd a d ­ ltfSUj t o f

L o c a tio n Operation dress dress dress o p e ra tio n

tn 1 Multiplication n -J- i rt + 2 n+ 2 2x
tv t-2 Addition n+ 2 M -3 n 4- 2 2x+3
tn -j~ 3 Multiplical ion « -} -l n+ 4 n 4- 1 5x
tn J- A Addition rt -j- 1 n -j- 5 n4 I 5 i 4- t
2x + 3
nr -j~5 Division It 4- 2 n -j-1
5x 1
C1G MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

When an intermediate result is not needed for further computations

it may be erased so that the storage location can be used for now
data. For instance, the execution of the first instruction involves
such a procedure: tile product of the numbers stored in locations
n r 1 and n — 2 is again placed into location n 2. This makes
the use of the storage more effective without overloading it with
unnecessary data.
The program wo have composed must be further supplemented
with an input instruction which transfers the initial data and
instruction codes into the storage and with the instruction fordeci-
ina l-lo-binary conversion because the machine uses Ilie binary
system while initial data are usually written and fed to tin* machine
in the decimal notation. Instruction “division” written in location
m + 5 must he followed by three more instructions, namely, for
converting the result of computation to the decimal system, printing
the answer and slopping the machine.
The hnal stage of preparing a program involves replacing symbolic
addresses by the corresponding concrete numbers.
These numbers, the absolute addresses, are four-digit mini hers
which are written in the octal number system beginning with 0000.
Several initial storage locations are commonly used as standard
operating cells (for input operations, number system conversion etc.).
Suppose we have a machine in which the first eleven locations are
used for this purpose (locations with numbers from 0001 to 0013
in the octal notation)*. Then we begin the storage allocation with
location 0014 and thus write down our program in tin* following
form:
Operation or nimikr isi I'nil | ad
L o ca t io n ad(lrc?« arldrcs.i j adilivss
i
i
00 It Input (in locations 0015 through i
0032) 0015 0015
0015 Conversion from decimal system to
binary (locations 0026-0032) 0026 0004 0020
0016 Multiplication 0020 0027 UU27
0017 Addition 0027 0030 0027
0020 Multiplication 0020 0031 0026
0021 Addition 0026 0032 0020
0022 Division 0027 0026 0027
0023 Conversion from binary system to
decimal 0027 0027
i. *0024 Print 0027
1

* Let the location identified by 0000 contain tlie number zero.

 S U P P L E M E N T 2. ON U N IV E R S A L D I G I T A L C O M P UT ER S G1
Continued

L o c a tio n O p e r a tio n or num ber

1st 2nd :!id
ad d ress ad d ress ad d ress

0 O 2 . r> S to p

01120 X

0 0 2 7 2

0 0 3 0 3

0 0 3 1 T>

0 0 3 2 1

0 0 3 3

3. Cyclic Processes. It is clear that in such a simple case as w<

have just considered there is no practical sense in using a computer
This elementary example has been taken with the only purpos<
to demonstrate how some familiar expressions are translated to tin
language intelligible to the' machine. The application of a digita
computer is effective, only when the number of operations that tin
machine carries out is much greater than the number of instruction:
we must feed into its memory. An instruction can he reused man\
times when a repetitious scries of the same operations is presen
in a problem. In such cases routines are said to contain loops. The
action of performing each operation in one traversal of a loop i;
called a loop cycle. \Y'e now consider two elementary examples o
cyclic programs.
(1) Compulation of the Square Root. Suppose it is necessary l<
approximate the square root of a. a positive number, to a giver
accuracy. For solving this problem we may take advantage of tin
following fact (o.g. sue ISI, Chapter 3). The sequence
3*1 —*T * u:2— ( X14" -Jj") * •••

•••» ^n-t-1 “ ~2 (^H "H"x^”) * (—

)

converges to a for every positive number a. Calculating the succes­

sive approximations x1} x 2, . . . and so on we may continue the
process until a sufficiently accurate value is obtained. For testing
the accuracy of an approximation we can compare xn with x„_,;
when the difference is less than the preassigned value we stop the
iterative process.
Now we see that to compute V a wc must place into three storage
locations the three numbers: a ~ xe (assumed to be an initial a p p r o ­
ximation), s (assessing the degree of accuracy) and . The further
compulation of 1/ a is perform ed a c c o r d in g to th e f o ll o w i n g r o u tin e :
OlS MULTIPLE INTEGRALS, FIELD THEORY AND S E R I E S

Loca­ 1st 2n d 3rd D e sc rip tio n of o p e r a t i o n s a n d

tio n <I>j>^rat L'^ri o r n u m b e r ad d ress address address th e ir re s u lts

00 \ 'i Input 0016 0015 Inputting numbers and ins­

truction codes
0015 Dec iinn 1-to-b inary 0030 1)003 0030 Converting input data to bi­
conversion nary system
0010 Bit-to-bit ad- 0031 0027 Sending x n from location
dition 0031 (where x n remains
to be stored) to location
0027
a
0017 Division 0030 0027 0031
x n.

0020 Addition 0031 0027 0031

X n J r xX n

0021 Multiplication 0031 0033 0031 X f l r l ^ Y ( X' * + I 7 )

0022 Subtraction 0031 0027 0027 41 x n
0023 1 1 jump 0032 0027 0016 Checking whether the given
accuracy is achieved by
comparing |x„_t — a:n |
with e and completing the
cycle if the accuracy is
attained

002-,» Bmai y-L>-doci- 0031 0031 Converting the result to deci­

inal conversion mal system
0025 Print 0031 Printing the result
0020 Stop
0027 Operating coll
0030
0031 *0
0032 K

0033 1
2
0 :M
I

(2) Tabulating Functions. Another typical example of a cyclic

process is evaluating elementary functions for different values of
their arguments, i.e. compiling tables.
For example, let us consider the function sin x. From Taylor’s
formula it follows that
r3 f1
sin x —x ----+ . . . + ( — 1)” - J-------------L
<2» Lit! nn f i (3)
 S U P P L E M E N T 2. ON U N IV E R S A L D IG IT A L C O M P U T ER S CIO

where the remainder /?n+ i satisfies the inequality

x-nf3
I 1 ^ (2*4-3)!
Denoting the Arlli term of the sum on the right-hand side of (3) by
uk and setting sh = u, -f- u2 4- . . . -f- uk we get
uk+i = — . where ak — 2A: (2k 1) (4)
and
"S), ! i (^v 11 • • *)i ^ (*j)
Finally, it is readily seen that
a-h+1 = ah Sf, -f- 6, ui = 6 (k = 1, 2, . . .) (0)
We thus arrive at. the following computational procedure: we choose
a rough approximation Sj = x for sin x and compute the successive
approximations sk (k = 1, 2, . . .) by using formulas (5).
At each step, after the Arlli approximation sk has been computed,
we first find a*.fi (by (6)), then W/,+i (by (4)) and finally $k + If
appears to be less than the preassigned number e the machine
regards sh as the sought-for value of sin x and proceeds to compute
sin x for another x. This computational scheme may he realized
by means of the following program:

Loca­ 1st 2nd 3 rd D c sc rijitio n o f o |> en itio n s

tio n O p e ra tio n or n u m b er ad d re ss ad d ress address and their results

non Input 0043 0015 Inputting numbers and

instruction codes
0015 Decimal-lo-binary con­ 0044 0010 00-44 Converting input data
version to binary system
oo lo Hit to bit addition 0050 0002 Transferring x to the
standard location
for Ufi
0017 Bit-lo bit addition 0002 0003 Transferring Uj to the
standard location
for s
0020 Multiplication 0002 0U02 0004 .T“
0021 Subtraction 0004 0004
0022 Multiplication 0045 0000 0005 8 'A - 1)
0(123 Addition 0005 0040 0005 8 (A- 1) -; 0
0024 Addition 0057 0005 0057 °h
0025 Addition 0000 0044 ooon k
0020 Division 0002 0057 0005
<>h
020 MULTIPLE INTEGRALS. FIELD THEO RY AND SE R IE S

Con tinued

J.oca-
O i<cration o r n u m b e r
1st 2nd 3 rd D e sc rip tio n of o p e ra tio n s
t ion a d d re ss address ad d ress a n d th e ir re s u lts

0027 Mimplication 0005 0004 0002 — “Jim

ah
003o Addition 0003 00G2 0ut>3 4>+l = Sk~r "h-i
(Htil K l j«mp 0047 0002 0022 Completing computa­
tion Of S i l l Tf
0032 Hit-to-bit addition U0C3 0050 Transfer: sin xy -► xy
0033 Bil-lo-bil addition 0000 Transfer: 0-*-fc
0034 Bit-to-bit addition 0057 Transfer: O-*-^)
U(»3o Address modification UO10 0050 0010 Modifying the 1st ad­
dress of the instru­
ction in location
on ll>
0U3G Add ress mod i ficn t ion 0032 0055 0032 Mwlifying the 3rd ad­
dress of the instru­
ction in location
0032: i i -f 1
0037 Addition 0001 0044 00C1
0040 Transfer of control 0001 0054 0010 Completing tuluilalion
0041 Binary-to-decimal con­ 0050 (►003 0050
version
0042 Print 0050 0003
0043 Stop
0044 1

0045 8
0040 0
0047 e.
0050 X,
0051 in
0052 *3
0053 *4
0054 4
0055 1 in the 3rd address
0050 1 in the 1st address
0057 0 (0 * )
0000 0<Ar)
0001 Operating cell f o r i
0002 Standard location for
0003 Standard location for $
0004 Standard location for —x2 1
OOf.r* 1
1 I 1 1
 SUPPLEMENT 2. ON UNIVERSAL DIGITAL COMPUTERS 621

Similar routines can be written for evaluating the other elemen­

tary functions (cos x, ex> In x etc.).
4. Flow-chart. Subroutines. When preparing a program for solving
a complex problem it is helpful to represent it as a sequence of
blocks corresponding to individual problems, that is to draw a
flow’-chart. This makes programming easier; moreover, one and
the same block may bo involved in different routines as a subroutine.
Let us consider an elementary example- Suppose we have to compute
an approximate value of an integral
b
J = \ f { x) dx (7)
a
by using the rectangular formula (e.g. see 181, Chapter 12). It is
natural to break down the computation into two steps (two blocks):
(1) Computing the values of the function / (x) for the points Xi
involved in the rectangular formula.
(2) Computing the sum

(s )
i=*l

which approximates integral (7). The routine for computing / (xj

depends upon the form of / (x), while that for evaluating sum (8)
is irrelevant of the choice of / (x).
The absolute error of the result (i.e. the absolute value of 1lie
difference J — S) depends on two factors: the accuracy of the rec­
tangular formula* and the accuracy with which the values of /
at the points x { are determined.
5. Instruction Codes. Operations on Instructions. In writing the
above programs we referred to machine operations by symbolic
names such as “addition”, “multiplication” etc. These symbols,
however, must be replaced by the corresponding binary numbers
to be entered into the machine, i.e. by instruction codes. These
binary numbers are fed to the memory locations in the same way
as data, that is in the form of a sequence of the symbols 0 and 1.
Each location has several predetermined binary places for storing
them.
Let us consider an imaginary three-address computer. Suppose
its locations are of 42-digit capacity, a 0-digit number being used
for the operation code and three 12-digit numbers representing the
addresses. Then the instruction “add the numbers in locations 7

* For estimation of errors of various formulas for approximate integration

see (S|. Chapter 2, § 2.
022 MULTIPLE INTEGR AL S, HELD THEORY AND SERIES

and 12 together and store the result in location 13” is represented

in the code of our machine as
1
0 a D0 0 310 0 0 1 1 0 0 0 a a ala i i 0 0 0 0 0 0 0 0 0 0 1 1 Q1 0 0 a 0 0 I
i
V---------- „----------- a ----------- „------------^ ---v-----'
1st address 2nd address 3rd aduress code
where 000001 is assumed to designate the operation of addition.
The fact that instructions and numerical data, when put into the
machine, are of similar form presents no difficulties. Moreover,
this enables us to treat instructions as ordinary numbers, e.g. to
add* one instruction to another, which turns out to simplify pro­
gramming. To illustrate this let us consider a simple example. Sup­
pose we have to determine the sum of a thousand numbers. We
may of course place them in the storage, e g. in the locations with
numbers from n -f- 1 to n 1000, and then compile the following
program:
1st instruction:
Addition n -1 n-L 2 n-\- 2

2nd instruction:

Addi tion n -2 «+ 3 n f- 3

OOO/h instruction:

Addition n + ! W‘J n-\ 1000 n \ IU00

But we can solve this problem more economically: we again

place the numbers in locations n -{- 1 through n 1000 and then
write (e.g. in location n 1001) the following “instruction”:

Now let

* Tt should Ihj noted that operalions «m instructions involve special types

of addition, namely, operation code addition, address-lo-address addition and
bit-to-l>it addition.
 SUPPLEMENT 2. ON UNIVERSAL DIGITAL COMPUTERS 123

be stored in location m I- 1. This instruction results in adding

together the first two numbers. In the next location m -j- 2 we
place the following instruction:

Address-to-address addition rn -p 1 n -f 1001 m -T- 1

In location m + 3 we put the instruction of transferring control

to location m 1 which will at this stage contain an instruction
of the form:

A d d itio n H- 1 /i - r d n—6

The latter instruction causes the machine to add the third number
to the sum of the first two. It is clear that this loop of three instruc­
tions will provide adding together all the numbers stored in locations
n -j- 1 through n } 1000. It remains to provide the corresponding
instructions for printing the answer and slopping the machine when
the work is completed.
Thus, the use of operation of addition of instructions has made
it possible to perform address modification and thus replace a long
chain of similar instructions by a small number of operations.
t>. Automatic Programming. Modern computer techniques include
libraries of subroutines and other programming facilities but never­
theless the procedure of programming takes much time and effort.
The stage of programming may require much more time than that
of machine operation. This primarily applies to modern large high
speed computers, and particular attention is therefore attached to
various methods of automatic programming.
Here we cannot go into particulars concerning these methods.
Their basic idea is to use the computer for translating a routine
written in a symbolic language into a machine language. In other
words, a mathematician describes the procedure for solving a pro­
blem in a special language using words and symbols taken from a
fixed set of terms (vocabulary). Each character in this text when
transferred into a machine is represented by a definite combination
of zeros and ones. A special routine called the translator is then
used to translate this program into a machine language program
that can l>e run on a computer. It is necessary that the procedure
be described in accordance with some formal rules of syntax by
means of a clear cut set of available words. There exist several con­
ventional languages designed for automatic programming. The
most widely used languages are ALGOL and EOHTHAN*. Each

* Abbreviations for “Algorithmic Language" and “Formula 'I ranslatiiur".

024 MULTIPLE INTEGRALS, FIELD THEORY AND SERIES

language can be used independently of the computer at hand while

the choice of the translator is governed by the choice,of a language
and the type of a machine (but not by the problem to be solved).
The use of algorithmic languages and translators reduces tedious
and labour-consuming work of a programmer.

§ 4. ORGANIZATION OF COMPUTER WORK

1. Conditions for Effective Use of a Computer. As has been said,
the solution of a problem on a computer requires an adequate routine,
i.e. a sequence of coded instructions causing the computer to perforin
(lie elementary operations which the problem is reduced to. If
the number of individual instructions in a routine wore equal to
the number of operations necessary to solve the problem, the use
of computers would not be effective since the process of programming
would take almost as much time as performing all the calculations
manually. However, every complicated problem which is to be
solved on a computer contains specific groups of operations that
must be repeated(ovor and over again. This was shown in the instance
of the routino^for computing the square root; to a greater extent
this is associated with more complex problems. The number of
instructions in a reasonable routine is therefore much less than the
number of operations it causes the computer to perform. The use
of a computer is particularly effective in the problems where different
sets of initial data must, be processed in one and the same manner.
On the other hand, there exist problems in which the application
of a computer useless b e c a u s e t hey require a long and led ions
process of programming while the amount of calculations is compa­
ratively small.
The question whet her the computer methods are suited for a par­
ticular problem is of primary importance for effective use of com­
puters.
2. Basic Stages of Solving a Problem on a Computer. Once tho
decision to compute has been made, the problem moves through the
following stages:
(1) Problem Formulation. In the first place any problem to be
solved on a computer must be slated precisely in the mathematical
form. In other words, the problem that has arisen in physics, engi­
neering or elsewhere must be described as that of solving eq u ation s,
evaluating integrals and so on. It should be stressed that this stage
is mainly based on collaboration of physicists or engineers setting
the problem and specialists in computational mathematics and is
therefore connected with considerable difficulties. To overcome them
successfully it is necessary that the mathematicians be familiar
with the nature of the problem they deal with. On llu* other hand,
tho engineers and physicists must have a clear idea about compu-
 SUPPLEMENT L\ OX UXIVERSAL DIGITAL COMPUTERS 625

lational methods and the advantages they pul at his dispo­

sal.
(2) Problem Analysis {Selection of an Algorithm). Computer cannot
deal with such terms as the solution of an equation, integral, func­
tion etc. in which we commonly stale a problem Hence, before
translating the problem into a language acceptable to the machine,
a suitable numerical method must be selected for such operations
as evaluation of derivatives and integrals, solution of equations,
etc. I'or example, d eriv a tiv es are replaced by the corresponding
divided finite differences, integrals are evaluated by means of some
approximate formulas (such as the trapezoid rule or Simpson’s rule)
and so on. Thus, tin* problem is reduced to a finite sequence of arith­
metical operations. It is clear that one and the same problem can
be solved by using different numerical methods. Computer efficiency
essentially depends upon the proper choice between alternative
methods of computation.
(.’!) Programming. This stage comes when the problem analysis
has l.»ee11 accomplished. i.e. when a suitable algorithm consisting
of a sequence of elementary operations has been chosen for each
step. The routine, however, may be written in a variety of ways.
The choice of the most adequate program, of the best way of using
the storage and other machine’s facilities requires professional skill
of the programmer, his experience and familiarity will) the type
of the machine.
Ci) Machine flan. When a program lias been completed the process
of coinpulal.ion.s reduces to a standard machine operation. The person
who iiiaiiipulales the controls of a computer (operator) may be
unfamiliar with the particulars of the problem he deals with.
M. (’.hoi king Computer Operation. Error Detection. Mass-scale cal­
culation on a computer involves millions of elementary operations.
It is therefore a complicated problem to prevent computer errors.
Wrong results may be obtained for various reasons. To begin with,
the program itself may have errors. A single mistake in a routine
may result in (he failure of the whole compulation process. 1Ienee.
nnv mistakes in a computer program must he located ami corrected
before starting compulations. 'This is the so-called checkout process.
Sometimes it is helpful to perform a step of calculation manually
and then compare the result with the machine result at the same
step. There exist some other systematical methods of checking pro­
grams which however we shall not discuss here.
Another thing we most make sure* of is the proper hardware ope­
ration. This is usually cheeked by putting through a number of test
routines. Some of these an* specially designed to lest all the main
units and functions of tin* computer, lint it should be taken into
account that a failure of equipment may occur in the process of cal­
culation. Such faults an* delected and nli mine l«*d in fin* lo!bv;;*’T
t O - U a J-i
020 M U L T I I’ll'" INTEGRATE, FIELD THEORY AND S E R I E S

manner: the niachine is made lo store an intermediate result and

to repeat computation once again. If the two results coincide the
machine proceeds lo the next step.
Finally, it may be a rounding error that involves wrong results.
The finite capacity of storage locations puls a limit to the precision
of computations since all the quantities must be rounded off by
deleting less significant digits. These deviations from the theoretical­
ly correct values may occur in the course of a long computation and
yield an incorrect result without anv fault in the. routine or hard-
“ ft

ware.
There exist different methods Tor increasing the accuracy with
which the computed results are obtained. For instance, we may
write numbers having twice as many digits as are normally handled
in a given computer by using two locations instead of one for each
number (the so-called double-precision method).
 BIBLIOGRAPHY

(Starred items are in Russian)

1. * Bari. N. K., Triyouotueiric Series. Fizmolgiz, Moscow, 1961.

2. * Berezin. 1. S.. Zhidkov. N. P., Computational Methods* vol. 2, Chapter 4,
Fizmntpiz. 1962.
3. De Brujn. N. O., Asymptotic Methods in Analysis, Amsterdam, North*
Itollnnd Publishing. 19(11.
4. Eddington. A. S.. The Mathematical Theory of Relativity, Cambridge,
Cambridge University Press. 1924.
o. Elsgolts. L. K.. Differential Equations and Cnlculus of Variations , Mir
Publishers. Moscow. 1970.
6. F.rdely. A., Asymptotic Expansions, New York, Dover, 1956.
7 * Gelfand. I. M., Shilov, 0 . E., Generalized Functions, vols. 1, 2, 3, Fizmat-
giz. 105?<.
The English translation: Generalized Functions, Academic Press, Inc.,
New York. 1901.
8. * Ilyin. V. A.. Poznvnk. P. O.. Fundamentals of Mathematical Analysis,
“Nauka‘*. Moscow. 1907.
9 . * Ivanenko, D. D-. Sokolov, A. A., Classical Field Theory. GTTI, Moscow,
1901.
|n. * Koclun. N. E.. Vector Calculus and Elements of Tensor Calculus, GONTI,
Moscow. 1938.
11. * Knlnioj'tinix, A- N-. F o m i n . S. V., F i c m e n t s of the Th eory o f F u n c t i o n s
and Functional Analysis, *‘Nauka‘\ Moscow. 1908.
The English translation: Introductory Real Analysis, Prentice-Hall, Inc.,
Englewood Clifts, N-3 , 1070.
12. * Lavrentyev. M. A., Shahnt, R. V.. Methods of the Theory of Functions
of a Complex Variable , “Nauka”, Moscow, 1965.
13. * Nnrden, A- P.. Differential Geometry, Uchpedgiz, Moscow, 1948, p. 188.
14. * Rashevsky. P- K., lliemnnuian Geometry and Tensor A nalysis, “Nauka**,
Moscow, 1964.
15.* Sobolev, S. L., Equations of Mathematical Physics, GTTI, Moscow, 1954.
16.* Tikhonov, A. N*. Stable Methods for the Summation of Fourier Series ,
Doklady Akademii Naiik SSSR, vol, 156. N 2, 1964. 268-271.
17. * Tikhonov, A. N., Samarsky, A. A.. Equations of Mathematical Physics,
“Nauka”, Moscow. 1906.
The English translation: Equations of Mathematical Physics, New York,
Pergamen Press, 1963.
18. * Yefimov. N. V.. Higher Geometry, Fizmatgiz, Moscow, 1961.

40*
 NAMK

Abel. X. 329 Kocbin. N. R. 027

Arzel.i, .{7.i Ki)] mogfni'i iv. \ \ li‘^7
Ascoli. ('■. 375 KI (meeker. F.. 281
Krylov, A. \ . 517
Bari, K. 027 Kudryavtsev, L. I). 5
Berezin. I- S. 027
Bessel. F. \V. 7.00 Lagrange, J. L. 350
Bolzano, B. 23. 375 Lai'uirre, K. A. 501
Btnlak. B. M. a Faimc. G. 2(59
Bunyaknvsky, V. Va. 307 Laplace, P. S. 202, 205, 59!)
Lavrentyev. M. A. 027
Cauchy. A. I.. 328. 320. 315, 310, Lcbesguc. M. L. 29
3.70. 307, :’.70, 389, 1l»2, Htt. 1',8, Legendre, A. M. 5o2
/,5n Leibniz, G. W. 327, 357
Chebyshov. P. I,. 323, 5(55 Lobar lie vsky, X. I. 101
Lvapunov, A M. 530
P ’Alembert, J. L. B. 3'»2 Mobius, A. F. 209
parboilx, J. G. 32, 3'., 35, 82
pe Brujn, X. G. 027 Newton. I. 101, 128, 2<>0. 210, 278,
pini, U- 332 3i is
J)ira<\ P. A. M. 380 Norden, A. F. (527
pupin, F. P. O. 151 l)<trograd.sky, M. V. 218, 308
Par.seval, M. A. 525
£ d d in g to n . A. 8. 027 Poisson. S. I). 3!)ii, 12(5
B instei n . V 312 ro/nvak, I., G. .i, 027
j'lsgol Is, L. K. 027 Pythagoras, 508
Erddlv. A. 027
Bnler. L. HHi. 135, 278, 301 Basin*v<kv. P. K. 027
Bi(Miiann, (5. F. B. 32
Fomin. S. V. 5, 027 Itodrigues, (L 501
Fourier, J. B. 211 Samarsky, A. A. (527
f r e n e l . .1. F. 11 7 Schwarz, L. A. 1-12, 307
Fresnel. A. .T. .399, 357 Senvt, J. A. 117
Bi'iilliini, G. 35S Shaba I, It. V. (527
Shilov, G. K. 027
Gauss, C. F. 137, 102, 210 Sobolev, S. L. 027
Gelsfand, I. M. (527 Sokolov, A. A. (527
Green, G. 183 Stirling, J. 599
Stokes, G. G. 221
Ifadamaid, J. S. 3-10 Svoslmikov, A. G. 5
11ami 10>n, \\\ B. 257
llemiile, ('. 505 Taylor. It. 351
Gilbert. D. 508 Tikhonov, A. N. 585, 027
Weierstrass. K. T. \Y. 23. .120. 375,
Ilyin, V. A. 5, 027 119
Ivanenko, 13. D. 027
Yefimov, A. V. 5, 027
Jordan, C. 21 Zhidkov, \ . P. 027
 SIBJKCT INDEX

Abel’s test in polar coordinates 72

for (conditional) convergence of an invariance of 24. 28
improper integral .'107 monotnnieily of 23. 27
for uniform convergence of a seri­ of an oriented domain 77f
es 329 of a plane figure 21. 48
Absolute (intrinsic) properties of a of a polygonal ligure 23
surface 101 of a surface 14211
Acceleration definition of 142
of a particle of a fluid 277 formulas for I44ff
of a point 120 properties of 27f
normal (centripetal) 127 Arithmetic unit of a computer 1*0(5
resolution of 127 Ascoli-Arzela theorem 375
tangential 127 Asymptotic
Adder (in a computer) till equivalence 501
Addition of tensors 3(Xi expansion 593
Additive set function 421T. t>if representation of a function 592
derivative of 43, 811 sequence 592
e.\ tension of 46f series 59,3
reconstruction from its derivative Axially symmetric field 232. *238
of 4 iff. 85
Additivity Base vectors of a curvilinear coordi­
of area 23 nate system
of the integral 10, S3, 170 in space 2(59
of volume 80 on a surface 13.3
Address of a location in a computer Basis
009 of a functional space 521, 50S
modification of 023 orthoumiiiul 2S3
Adjoint operator 291 transformation of 285. 31 it
Algebraic operations on tensors 300 reciprocal 311
AIAiOL 023 liessol s
Algorithm 025 identity 508
Almost periodic function 478 inequality 501*
Amplitude of a harmonic 470 Beta function—$«*<* Euler’s integral
Analogue computer 004 of the 1st kind
Analytic function 354 Bilinear invariant form 292
Anchor ring (torus) 131 symmetric 293
Angle between elements of a func­ Binormal 110, 119
tional space 5l>0 Bit GKJ
Angle between two curves on a stir- Bolzano-Woiorstrass theorem 28
face 141 f Boundary
Antisymmetric (screw-syui met t ie) of a pat tit ion 35
tensor of rank two 3U2 of a set 21
Applicable (isometric) surfaces Hit) point 21
Arc length of a curve on a surface 139 Bourded sot 22
Area
additivity of 22,. 27 (.artesian product 101
element of 07, 20i C.nochv-I l a d a m a r d llieorom 348
»•« V U i \ JI j I l t i t t UOU t.,aiu.t»y-DuiiyukoYsky inequality 3071
030 SUHJRCT INDEX

Cauchy’s general purpose (universal) 604

form of the remainder for Taylor's high speed electronic 603ff
series 359 input of 605
principal value of an improper memory (storage) of 605f
integral 408 location (cell) of 605
test output of 606
for a sequence to be weald y fun­ representing numbers in 60S
damental 379 special purpose 604
for improper integrals 389, 402, tiiree-aduress 609
408 universal digital 604
for uniform convergence of a Condition(s)
sequence 328 for existence of a double integral
for uniform convergence of a 31 ff
series 329 necessary and sufficient
Centro of gravity for a line integral of the second
of a material line 171 type being path-inde­
of a material surface 205 pendent l§0rr, 228f
of a plate 49 for integrability of a function 35
of a solid 86 for squarahility 25
Change of variables Conditional transfer of control (con­
in double integral 74ff ditional jump) 612
formula for 76 Conductivity
in triple integral 98ff specific 283
formula for 100 tensor 283, 291
Chcbyshev-Hermite polynomials 565 thermal, coefficient of 241
Chebyshev’s polynomials 523, 564 Conjugate tensor 302
Circular helix (screw-line) 113 Connection between line integrals of
Circulation of a vector field over a the 1st and 2nd types 175
contour 251 Constant of gravitation 87
Closed Content (Jordan)
set 21 exterior 24, 80
system of functions 525 interior 24, 80
Cneffi/’iontfs) Continuously differentiable [unction
fundamental of the 1st order, of 113, 337
a surface 144 of a plane figure 24
Lame’s 269f! of a space figure 80
metric 138 Contour lines (horizontals) 232
of area expansion 73 Contracted tensor 301
of thermal conductivity 241 Contraction of a tensor 301
Coherently (conuordantly) oriented Contravariant
contour on a surface 21D componont of a vector 312
Compactness of a family of functions transformation of 313f
374 indices of a tensor 315
criterion for (Ascoli-Arzela theorem) metric tensor 316
375 Convection 277
Completeness 524 Convergence
of Legendre’s polynomials 530 in a functional space 566
of trigonometric system 527ff in the mean 207, 334, 366f
Complex function of a real argument connection with other types of
531 ff convergence of 372ff
Components (of a vector, of a tensor) of a sequence 367
283, 284, 286, 287, 283 of a scries 367
Computer with weight function 560
analogue 604 of a functional sequence 320
arithmetic unit of GOG of a functional series 324
control unit of 606 of an improper integral 384, 385,
digital 601 401
 SUBJECT INDEX <>31

Convergence, in a functional space point of rectification of 116

absolute 390, 391 principal normal of 116, 119
conditional 392. 397 rectifiable 26
uni form 436. 471 rectifying piano of 119
ol a power series shape of, in the vicinitv of a point
circle of 363 121ff
interval nf 343 smooth 105
radius of 341. 363 tangent of 116, 119
of a sequence ol complex numbers Curvilinear cylinder (cylindroid) 19
362 Cusp 111
of a series with complex terms 363 Cuspidal edge 164
uniform 213, 320. 436. 471
weak 378 Darboux’s
Coordinate integrals, upper and lower 34
curve 63, I30f lemma 35
surface 95 proof of 36
Coordinates upper and lower sums 32, 82
connected with the moving tri­ Del—see Hamiltonian operator
hedron 119f Delta function (Dirac’s) 380
curvilinear 64, 95, 131 expansion as Fourier’s integral of
in space 95 576
in the plane 64 expansion in Fourier scries of 575
orthogonal 142. 267 Density
cylindrical 95. 270 field 230
on a surface 131 of mass distribution 43, 85
spherical 96, 270f Derivative
Cosine integral 589 left-hand 487
Covariant of a double integral with respect to
components of a vector 312 the area of domain of in­
transformation of 314 tegration 43f
indices of a tensor 315 of a set function 43
tensor 315 of a triple integral with respect
Curvntnre(s) to volume 84f
mean, of a surfacr 157, 162 of a vector function 108
normal, of a surface, in a given right-hand 487
direction 150 Deviation of functions
radius of 151 least 523
of a plane curve 1231 maximum 321, 370
of a space- curve mean square 3G6
first 117 Diameter of a set 22
second (torsion) 117 Difference between two sets 24
principal, of a surface at a point 155 Differential geometry 107
total (Gaussian) of a surface 157, Differential of a vector function 110
162 Digital computer 604
Curve(s) fixed-point 608
binormal of 116. 119 floating-point 608
concave down (up) 124 Dini’s theorom 332
intrinsic (natural) equation of 124 Direction cosines of the normal to a
moving (natural) trihedron of 116 surface 135
normal piano to 119 Directional derivative 233, 234
on a surface 133 Distance between two sets 22
onc-narameter family of 130 Distribution (generalized function) 380
osculating plane of 119 Divergence
parametric (coordinate) 130f expression in general curvilinear
parametric representation of 113 orthogonal coordinates of
parametric (vector) equation of 113 271 f
piecewise smooth 165 of a tensor 306
G32 SLIUECT INDEX

()i verge nc ik Error function 587

of a vector Held 213 Euler-Poisson integral 396, 125, 455
physical meaning of 21-Iff Euler’s
theorem (Ost rogradsky’s formula) formulas 361, 531
218. 222 integral of Iho 1st kind (beta func­
Domain 21 tion) 460
arewise connected 21 properties of 464ff
closed 21 integral of the 2nd kind (gamma
multiply connected 129, 228 function) 407, 160
of integration 30 properties of 46017
oriented 77f specification of motion of a fluid
regular in a certain direction 90. 219 278f
dimply connected 129, 190. 22S Even function 181
Double integral 20, 30 Everywhere dense set 29. 375
additivity of 40 Exhaustive sequence of sets 121
applications of 47ft Expansion of a function into a power
change of variables in 74IT series 311
conditions for existence of 31 ff Expansion of a function into a tri­
definition of 30 gonometric series ISO
evaluation of 52ff Explicit partial derivative 276
linearity of 40 Exponential integral 589
monotonic.ity of 10 Extension of a set function 40f
over an oriented domain 77f External memory of a computer Ooo
properties of 39ff
reduction to an iterated integral Eield 230
of 52ff of a tensor 305
Double-precision method 62G of force 173
Du pin's indiculrix 153f of normals on a surface 208
operations
Eigenvalue of a linear operator 310 in cylindrical coordinates 271
Eigenvector of a linear operator 310 in general curvilinear orthogonal
Element coordinates 27Iff
of a functional «paee 593 in spheiical c o o n i inates 273
of area in curvilinear coordinates scalar 230ff
67, 2G9 axially symmetric 232
of i nl egrat i«*n 30 charge-density 250
of length in curvilinear orthogonal cylindrical 232
coordinates 268 directional derivative of 233. 231
of volume in curvilinear coordina­ gradient lines of 236
tes 97 f, 269 gradient o) 235
Elliptic point of a surface 157 illumination 230
Equal ion(s) level lines of 232
of continuity 248, 282 level surfaces of 231
of Euler 155 mass-density 230
of heal conduction 265, 571 spherical 233
of motion of a continuous medium symmetry of 232f
308f two-dimensional (piane-parallel)
of the tangent plane to a surface 232
133 f theory 230IT
parametric vector 230
of a curve on a surface 133 axially symmetric 238
of a space curve 113 circulation over a contour of 251
of a surface 132 cylindrical 238
Eniiicontiimnus family of fimc.linn« divergence of 213
375 electrostatic 237
Equivalent, weakly fundamental fun­ (lux across a surface of 211
ctional sequences 381 Newtonian potential of 210
 sun. f i xt im > i:\ n.83

Field. v e c t o r acceleration of convergence of

of displacements 205 51 Off
of force 237 conditions for uniform conver­
of grail ienls 233 gence of 5081"
of gravilatiou 236 cosine 500
of a ma.^ distribution 236 for functions of several inde­
of a mass point 240 pendent variables 535ff
of velocities 236 half-range (incomplete) 5ou
ono-d i uiensiouul 238 in complex functions 531 ff
plane 23.S si ne 500
plane-parallel 233 speed of convergence of 51 iff
potential 238 stable methods for summation of
potential function of 238. 240 585
rotation (curl) of 251 theorem on convergence of 488
sink of 242 with respect to an orthogonal
solenutdal 246 system of functions 504
source of 242 with respect to a system of func­
symmetry of 238 tions orthogonal with
tube of 238 weight function 56i»
vector lines of 236. 237 sine transform 546
vector potential oT 2.57 transform 546
vector surfaces of 237 applications of 568ff
Fineness of a partition 30. SI 111tee-d i inension aI 555
First curvature—sre Curvature two-dimensional 551
First fuiiilamenl.il quadratic form of transformation 546
a surface 130 for functions of several indepen­
expression in orthogonal coordina­ dent variables 55Iff
tes of 142 Frequency of a harmonic 476
Flo\v-chart 621 Frenet -Serrcl fonnulas 117
Fluid How 212. 240, 218 Fresnel’s integral(s) 399, 457
plane 24flf Fridlani inlegral 458
Fluicl-lln.x density (vector) 246 Function(s)
Flux ,m.d\lk 25 4
Inniinous, incident on a plane re­ continuously differentiable 113. 337
gion 51 Dirac (delta function) 380
of a fluid, through the cross section even 484
of a channel 51. 212. expansion into Fourier’s series of
24(1 483, 484
of a plane liehl through a contour family of
250 compact 374
of a tensor 308 equicontiniKius 375
of a vector liehl 241 uniformly hounded 375
of heat 24If generalized 379ff
FORTRAN 623 harmonic 265
Fourier’s i ntegrabic 30, 32. 381. 82
coefficients 483. 504. 533 left-hand derivative of 487
cosine transform 548 linearly dependent (independent) 562
integral 540 odd 484
for functions of several inde­ orthogonal 481. 501, 532
pendent variables 55011" orthogonal with weight function
in complex form 545f 559. 560
integral formula (theorem) 540 oscillation of 38
inverse transform 546 periodic 476f
i fiversion formula 546, 553 piecewise continuous 136
law of heat propagation 241 piecewise smooth 486
polynomial 506 right hand derivative of 487
series 483. 504. 533, 5(40 set 41
634 SUUJECT INDEX

FuuctUm(s) Heal flux (vector) 211

square-intcgrable (suimnable) 360, High speed electronic computer G03ff
507 Hilbert space 568
witli weight function 559 llodograpli 107
vector 107 Hyperbolic point of a surface 157
Functional 381
determinant (Jacobian) 62
sequence 319 Ideal liquid 278
convergence in the mean of 367 Inpul-output system 568
convergence of 319 frequency characteristic of 569
series 324 linear 569
convergence in the mean of 367 phase characteristic of 309
space 5G5ft principle of superposition for 569
basis of 524, 568 spectral characteristic, of 569
Fundamental coeftieienls (quantities) Input uniL of a computer 605
of the 1st order of a surface Instruction (in a program for a
144 computer) 601
Fundamental (covariant) metric ten­ Instruction code 021
sor 316 Integrahle function 30, 82
Integral(s)
additivity of 40
Gamma function —see Euler’s inte­ Darhoux’s, upper and lower 34
gral of the 2nd kind dependent on a parameter 279
Gaussian (total) curvature of a sur­ improper 435ff
face 157, 162 Cauchy's test for uniform con­
Gauss’ theorem 246 vergence of 448, 450
Generalized function 3711fl evaluation of 453ff
General purpose (universal) computer reduction to a functional sequen­
604 ce of 438Cf
Geometry uniformly convergent 43G, 471
differential 107 Wciorslrass’ test for uniform
intrinsic, of a surface 161 convergence of 449
Lobachevskian 164 properties of 44Iff
Gradient 235 tests for 4 48If
expression in general curvilinear multiple 4G8ff
orthogonal coordinates of proper 4261?
271 double 20, 30
Gravitational attraction estimation of the modulus of 40,
between a mass point and a mate­ 83 170
rial line 172 Kulur-Poisson 396, 425, 455
between a mass point and a mate­ 1-2ul or’s 106
rial surface 206 improper 883, 401
between a mass point and a solid absolutely convergent 390. 391,
8Gf, 420f, 470 417
of two material bodies 104 Cauchy’s criterion for conver­
Green’s formula 183R, 187 gence of 3S9, 402
application to computing areas 1891 comparison tests for convergence
of 392, 393, 395, 396,
Hamiltonian operator (nabla or del) 402fi. 4199
258 conditionally convergent 392, 397
operations with 258ff Abel’s test for convergence of
repeated application of 261 ff 397
rules for application of 2581T convergent 384f, 412, 424
Harmonic divergent 384. 412. 424
function 2G5 Caiichy’s principal value of
series 327 408, 413
vibration 47fi methods of evaluation of 399f
Heat conduct!vit.v mention 265. 571 multiple 103ff
 SUBJECT INDEX G35

Inlegral(s), improper Lagrange's theorem on finite incre­

of uonnegative function 414ff ments 111. 194
of unbounded function 400ff Laguerre’s polynomials 564
over unbounded domain 424f Lame’s coefticients (scale factors) 269IT
with infinite limits of integra­ Laplace’s
tion 383ff equation 265
iterated (repeated) 54. S9 method 598ff
linearity of 40 Laplacian operator 262
monotonicity of 40 expression in general curvilinear
multiple, of higher order 103ff coordinates of 274
of a vector function 112 Lohesgue’s measure 29
over a fluid volume, time-derivati­ Legendre’s polynomials 502, 556
ve of 279ff orthogonality of 502. 556
triple 82 Leibniz rule for differentiating an in­
sum 30, 81, 166. 190 tegral dependent on a para­
limit of 30, 82, 166 meter with respect to the
surface 199fT parameter 429
Integrand 30 Level line 232
Intensity of illumination 51 Level surface 231
Interchanging indices in a tensor 302 Limit
Interior of a set 21 of a vector function 107
Internal memory of a computer 606 of integral sums 30, 82
Intersection of sets (or product or point 21
meet) 22 superior 346
Interval of convergence of a power Linear operator 288
series 343 Linearity of the integral 40, 83, 169
Intrinsic Line integral 165
(absolute) properties of a surface 161 of the first, type 165f, 172f
geometry of a surface 161 applications of 170ff
(natural) equations of a curve 124 independence of the orientation of
Invariance the path of integration of
of area 24 170
of volume 80 properties of 169f
Invariant forrn(s) reduction to the definite integral
bilinear 292 of 1G7L 172f
symmetric 293 of the second type 165, 173f, 130-183
connection with tensors of 291 ff dependence on the orientation
linear 292 of the path of 180
multilinear 295 evaluation of 177, 183
quadratic 293 in a multiply connected domain
Invariants of a surface 159f 195ff
Inverse (Fourier) transform 546 reduction to the line integral of
Inverse mapping 62 the first type of 175fl, 182
Isobar 232 Lobachcvskinn geometry 164
Isometric Logical
(length preserving) mapping 161 addition 610
surfaces 160 negation 611
Isotherm 232 Lowering indices in a tensor 316
Luminous flux 51
Jacobian 62, 94. 101, 102
geometric meaning of 73. 101
Jordan content 24. 80 Mapping 62. 94
inverse 62
Kronccker delta 284 isometric (length preserving) 161
Krylov's method 517f linear 68
of a plane figure 62
Lagrange’s form of the remainder for of a space figure y4
Tavlor’s theorem 356 ati«»
636 SUHJECT INDEX

Material surface 205 Newtonian potential of a mass point

centre of gravity of 205 240
mass of 205 Newton’s
moment of inertia of 205 binomial formula 358
Matrix law of universal gravitation 104, 106
orthogonal 2455 second law of dynamics 128. 278,
inverse of 286 297, 308
transformation, from one basis to Normal
another 285 acceleration 127
transpose of 2S6 curvature of a surface in a given
Mass direction 150
of a material line 1701 curvature, radius of 151
of a material surface 205 plane to a curve 119
of a plate -48f section of a surface 149
of a solid 85 to a surface 135
Mass distribution 43, 85 Norm of a function 502. 532. 566
density at a point of 43. 85 Number systems 606. 607, 60S
moan density of 43
Maximum deviation of functions 321, Odd function 484
370 Ouc-parameter family of curves 130
Mean One-to-one mapping 62, 94
curvature of a surface 157. 162 Open
s q u a r e deviation of functions 3 6 6
value theorem i0. 831, 170 circle 21
Measure 20 set 21
Metric Operating cell (in memory of a com­
coefficients 138 puter) 610
tensor 316 Operations performed l»v a computer
Mixed tensor 316 668 ff
Moment of inertia Operator
of a material line 17 1f Laplace's 262
of a material surface 265 linear 288
ol a plate 50 adjoint 291
eigenvalues and eigenvectors of
of a solid S5f 310
Monoton icily Or de r o f s m a l l n e s s 5901
ol area 23 Orientation
of the integral 40. 83. 170 of a boundary of a surface 21 Of
of volume SO of a dosed contour on a surface 211
Moving (natural) trihedron 116 of a domain in Hie plane 771
iMdhius strip 209 of a surface 210
Multilinear invariant form 295 Oriented domain 77f
Multiple integrals of higher order 103fT Ort hogonal
Multiplication afline tensor 284. 286
of a matrix by a vector (on the coordinates 1 42, 267
right or on the loft) 318 matrix 285
of a tensor by a number 300 inverse of 286
of a tensor l>v a vector 300f system of functions 592. 559f
of matrices 817f Orthogonal it y
of tensors 3t>0 in a functional space 566
Multiply connected domain 129 cif complex functions 532
of trigonometric system 481
Nabln—see Hamiltonian operator with weight function 559. 569
/i-dimeusioual sphere lo.'*f Orthogonali/.alioii process 564
Neighbourhood 21 Orthonormnl basis 284
Nested transformation of 2.S4ff
collection of domains 15 Oscillation of a function 38
I uit;l \ <11 i inwlVIH I.» Oscillating plane 119
 XIUJKCT 1NDKX 037

Ostrogradsky theorem 218. 222 theory 24t>

applications of 22211 Power series 311
for tensors 308 applications of 35911
Parabolic point, of a surface I58 arithmetical operations on 352ff
Parametric curves oil a surface I3i.)f differentiation of 348, 351
Parametric (vector) equations expanding a function into 341,
«.»t a curve 113 35111. 357 ff
of a surface 1)5. 132 in a complex variable 3t»2fT
Parsevnl’s r e l a t i o n 525, 520 circle of honvergence of 3(>3
Particle derivative 27(5 integration of 318. 332
Partition of a domain 29, 81 interval of convergence of 313
fineness of 30. 81 radius of convergence of 311, 3t53
refinement, of 33, 34 remainder of 335ff
Periodic extension of a function 178 uniform convergence of 318ff
Periodic function 470 Principal
integral of 478 axes of a symmetric tensor 31 (>
period of 470 curvatures of a surface at a point 155
primitive 477 direction of a surface at a point 15-1
Phase of n harmonic 4715 normal ilt>. 119
initial 17(> value of a divergent im p r o p e r in­
Piecewise smooth curve 1(>3 tegral 408
Planar point, of a surface 158 Principle of superposition 5(59
Plane I’rod net
curve, concave up (down) 121 of matrices 317
figure 21 of tensor by a scalar 30l)
area of 24, IS (miter) of tensors 300
content (Jordan) of 21 Program (505, 1514
of area zero 26 cyclic (loop) (>I7
sqtinrahle 21 Programming 01 Iff
flow 238 automatic (»23f
normal to a curve 119 Projection of an element of a func-
osculating 119 11<11lil 1 •‘pace .,Ui
rectifying 119 Pseudosphere 1(53
tangent to a surface 133f Pythagoras’ theorem lor a functional
Point space 5(58
boundary 21
elliptic, of a surface 157 Quadralically integrable (sumuiahle)
hyperbolic, of a surface 157 funct ion —.*<•<• Squarr-i ulcgra-
limit 21 Ide (sumtnahlc) function
of rectification llli Quadratic form 293
parabolic, of a surface 158 first, fundamental of a surface 139
planar, of a surface 158 positive definite 139
umbilical (circular), of a surface 155 second, fundamental of a surface 152
Polar coordinates (jiff
Polygonal figure 23 Radius
Polyhedral solid 79 of convergence 311
Polynomials of curvature of a curve 117
Chebysliev’s 523, 505 of normal curvature of a surface 151
Herinite’s 5(55 of torsion (of second curvature) of
Kaguerre’s 501 a curve 117
of least deviation from zero 523 Raising indices in a tensor 31(5
oT order a with respect to an orthogo­ Reciprocal bases 311
nal system of functions 505 Reconstructing a lurirtion from its
Potent ial total differential I93ff
field 238 Rectifiable curve 2(5
function of a vector field 238. 2lo, surface 1 12
•me ite<lif\mg plane tin
 638 SUBJECT I N D E X

Refinement of a partition 33 termwise passage to limit in 369ff

Relative displacements, tensor of 303f uniformly convergent 324
Remainder of a power series 355ff comparison (Weierstrass’) test for
in Cauchy’s form 359 326
in Lagrange's form 356 properties of 331 ff
Resolution of a tensor ol second rank Set(s)
into symmetric and antisym­ arewise connected 21
metric parts 303 bounded 22
Rodrigues’ formulas 564 closed 21
Rotation (curl) of a vector held 251 closure of 21
expression in general curvilinear diameter of 22
orthogonal coordinates of difference between 24
272ff distance between 22
physical meaning of 253f (everywhere) dense 29, 375
symbolic formula for 252 function 4 Iff, 34ff
interior of 21
Saddle-point method Gl)2 intersection (or product or meet) of
Scalar product (in a functional space) 22
565 open 21
Scalar field—see f ield separability of 22
Scale factor 209(1, 608 subset of 22
Schwarz example 142ff union of 22
Screw line (circular helix) 113 Shape of a curve in the vicinity of its
Second curvature—see torsion point 121 ff
Sequence, functional 319 Simply connected domain 129,190, 228
convergent at a point 319 Sine integral iii.
convergent in the mean 367 Singular point 110, 400
integration and differentiation of Sink (negative source) 242
369 ff Skew-symmetric (antisymmetric) ten­
convergent on an interval 320 sor 302
convergent to a function 320 Smooth curve 165
divergent 319 Source of a field 242
domain of convergence of 320 Space figure 8ti
limit function of 320 oT volume zero 80
passage to limit in 369ff Special purpose computer 604
point of convergence of 520 Specilic conductivity 283
point of divergence of uniformly Specific heat 264
V t/* * * W* « ^ v V Spectral characteristic 569
properties of 33Iff Spherical coordinates 96
weak limit of 378 Squarahle figure 24, 384
weakly convergent 378 Squarc-iiitcgrable (summable) func­
weakly fundamental 378 tion 366
equivalent 381 Stationary distribution of temperature
Series, functional 324 265
convergent 324 Stirling’s formula 599, 602
convergent in the mean 367 Stokes^ theorem (formula) 224, 226
integration and differentiation of applications of 227ff
371 ff Storage (memory) of a computer 605f
dominant series of 32C Strain tensor 296
dominated 32G Stress 296
power 341 tensor 291, 298ff
sum of 324 Subroutine 621
Taylor’s 354 Subtraction of tensors 300
term-by-term passage to limit in Summation convention 312
339ff Superposition of harmonics 479
termwise differentiation of 337T, 372 Surrace(s)
termwise inteirration of 334ff. 371 applicable (isometr><0 160
 SUBJECT INDEX 639

Surfaces) orthogonal affine 284, 286

area of 142ff connection with invariant forms
classification of the points of 157ff of 292ff
coordinates on 131 of rank one 286
definition of 129 of rank p 288
lirst fundamental quadratic form of of rank two 287
139 connection with linear opera­
integral 199 tors of 28Sfl. 298ff
applications of 204 of rank zero 286
of the first type 200. 200 raisine indices in 316
of the second type 207, 213 resolution into symmetric and an­
reduction to double integral of tisymmetric parts of 303
200ff, 215ff skew-symmetric (antisymmetric) 302
iiitiin.-dc (absolute) properties of strain 296
mean curvature of 137, 162 stress 291. 298ff
normal to 135 symmetric 302
of constant curvature 163 principal axes of 310
one-sided (nonorientable) 208, 210 Thermal equilibrium 265
parametric equations of 132 Torsion (second curvature)
parametric representation of 132 of a space curve 117
second fundamental quadratic form radius of 117
of 152 Torus (anchor ring) 131
smooth SI Total differential, integration of 193ff
squnrahle (rectifiable) 112 Total (Gaussian) curvature of a sur­
total (Gaussian) curvature of 157, face 157, 102
if.2 Total partial derivative 276
two-sided (orient able) 208, 210 Tract rix 101
symmetric tensor 302 Transfer of control 012f
Transformation matrix from one or­
Tangent thogonal basis to another 285
plane to a surface 1331 Transformation of base vectors 285, 313
to a curve lit'*. 119 Translator f*23
Tangential acceleration 127 Transpose of a matrix 2S0
Taylor’s Trigonometric
formula 351 polynomial 519
series 480
series 354 system 481
Tensor(s) 230, 284 completeness of 527ff
algebraic operations on 300 orthogonality of 481
components of 283, 284, 2SG, 287, Triple integral 82
288 applications of 85ff
conductivity 283, 291 change of variables in 98ff
conjugate 302 conditions for existence of 82f
contracted 301 evaluation of 87fi
contraction of 301 properties of 83f
contravariant 316 reduction to an iterated integral
covariant 315 of 88fr
held 305
divergence of 306
flux of 308 Umbilical (circular) point of a sur­
general, definition of 314. 315 face 155
interchanging indices in 302 Unconditional transfer of control G13
lowering indices in 316 Uniform approximation of functions
metric 316 witii algebraic and trigono­
mixed 316 metric polynomials 51bff
multiplication by a vector of 300f Uniform convergence, tests for 320ff
of pure deformation 303. 305 Uniformly bounded familv of f u n c ­
>'4 u u J ) y lu i i i l l u l i isUif, tions 375
 SUIUKC.T IM )i:\

I nioii of sc Is 22 in curvilinear coordinates 071

Universal digital computer GO4 of a curvilinear cylinder 1Of. '*7f
of an n-dirnensioiia! parallelepiped
Veclor(s) 102 f
base, t‘<*r curvilinear coordinates on of an /i-dimensional sphere lord
a surface 133 of a polyhedral solid 79f
be Ul—see Field of a space figure SO, S3, 223f
function 107, 112. 28$
continuous 10$ Weak
derivative of 108 convergence 378
different! .able 1U8 limit 378
differential of 110 Weakly convergent sequence 378
integral of 112 Weakly fundamental sequence 37S
limit of 107 Weierst lass’
singular point of 11(1 approximation theorems 5|8lf, 37tiff
line —srr Field .t/ test 32G
potential of a snlcnoidal field 237 test (for integrals) 410
surface—see Field Weight function 558
Viscosity factor 278 Work of a field of force 173f
Volume
element of 071

TO THE HEADER

Mir Publishers would be grateful for your com­

ment;* on the content, translation and design of this
book. We would also be pleased to receive any oilier
?ngffe«tion« you may wish to make.
Our address is:
USSR, 120820, Moscow 1-110, GSP
Porvy Hiy.li«ky Pereulok, 2
MIR PUBLISHERS

Printed in the Union of Soviet Socialist Republics