Introduction to Methods of Applied Mathematics

or
Advanced Mathematical Methods for Scientists and Engineers

Sean Mauch

April 26, 2001

Contents

1 Anti-Copyright 22

2 Preface 23
2.1 Advice to Teachers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3 Warnings and Disclaimers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.4 Suggested Use . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.5 About the Title . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

I Algebra 1
3 Sets and Functions 2
3.1 Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
3.2 Single Valued Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3.3 Inverses and Multi-Valued Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.4 Transforming Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

4 Vectors 11
4.1 Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1
 4.1.1 Scalars and Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.1.2 The Kronecker Delta and Einstein Summation Convention . . . . . . . . . . . . . . . . . . 14
4.1.3 The Dot and Cross Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.2 Sets of Vectors in n Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.4 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.5 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

II Calculus 36
5 Differential Calculus 37
5.1 Limits of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.2 Continuous Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5.3 The Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.4 Implicit Differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.5 Maxima and Minima . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.6 Mean Value Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.6.1 Application: Using Taylor’s Theorem to Approximate Functions. . . . . . . . . . . . . . . . 57
5.6.2 Application: Finite Difference Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.7 L’Hospital’s Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.9 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.10 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

6 Integral Calculus 100

6.1 The Indefinite Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.2 The Definite Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.2.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

2
 6.3 The Fundamental Theorem of Integral Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.4 Techniques of Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.4.1 Partial Fractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.5 Improper Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
6.7 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
6.8 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

7 Vector Calculus 134

7.1 Vector Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
7.2 Gradient, Divergence and Curl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
7.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
7.4 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
7.5 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

III Functions of a Complex Variable 150

8 Complex Numbers 151
8.1 Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
8.2 The Complex Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
8.3 Polar Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
8.4 Arithmetic and Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
8.5 Integer Exponents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
8.6 Rational Exponents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
8.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
8.8 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
8.9 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179

3
9 Functions of a Complex Variable 202
9.1 Curves and Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
9.2 Cartesian and Modulus-Argument Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
9.3 Graphing Functions of a Complex Variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
9.4 Trigonometric Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
9.5 Inverse Trigonometric Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
9.6 Branch Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
9.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
9.8 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
9.9 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258

10 Analytic Functions 303

10.1 Complex Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
10.2 Cauchy-Riemann Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
10.3 Harmonic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
10.4 Singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
10.4.1 Categorization of Singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
10.4.2 Isolated and Non-Isolated Singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
10.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
10.6 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
10.7 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334

11 Analytic Continuation 356

11.1 Analytic Continuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
11.2 Analytic Continuation of Sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
11.3 Analytic Functions Defined in Terms of Real Variables . . . . . . . . . . . . . . . . . . . . . . . . 360
11.3.1 Polar Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
11.3.2 Analytic Functions Defined in Terms of Their Real or Imaginary Parts . . . . . . . . . . . 369
11.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
11.5 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375

4
 11.6 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376

12 Contour Integration and Cauchy’s Theorem 381

12.1 Line Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
12.2 Under Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386
12.3 Cauchy’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 388
12.4 Indefinite Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
12.5 Contour Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
12.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
12.7 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
12.8 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398

13 Cauchy’s Integral Formula 403

13.1 Cauchy’s Integral Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
13.2 The Argument Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
13.3 Rouche’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
13.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
13.5 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
13.6 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 418

14 Series and Convergence 422

14.1 Series of Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
14.1.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
14.1.2 Special Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
14.1.3 Convergence Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
14.2 Uniform Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432
14.2.1 Tests for Uniform Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
14.2.2 Uniform Convergence and Continuous Functions. . . . . . . . . . . . . . . . . . . . . . . . 435
14.3 Uniformly Convergent Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436
14.4 Integration and Differentiation of Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443

5
 14.5 Taylor Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446
14.5.1 Newton’s Binomial Formula. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
14.6 Laurent Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452
14.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455
14.8 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 462
14.9 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465

15 The Residue Theorem 490

15.1 The Residue Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 490
15.2 Cauchy Principal Value for Real Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 498
15.2.1 The Cauchy Principal Value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 498
15.3 Cauchy Principal Value for Contour Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503
15.4 Integrals on the Real Axis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 507
15.5 Fourier Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 512
15.6 Fourier Cosine and Sine Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515
15.7 Contour Integration and Branch Cuts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 517
15.8 Exploiting Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 521
15.8.1 Wedge Contours . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 521
15.8.2 Box Contours . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 524
15.9 Definite Integrals Involving Sine and Cosine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525
15.10Infinite Sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 528
15.11Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 532
15.12Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 546
15.13Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553

IV Ordinary Differential Equations 634

16 First Order Differential Equations 635
16.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 635

6
 16.2 One Parameter Families of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 637
16.3 Exact Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 639
16.3.1 Separable Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 643
16.3.2 Homogeneous Coefficient Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 645
16.4 The First Order, Linear Differential Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 649
16.4.1 Homogeneous Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 649
16.4.2 Inhomogeneous Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 650
16.4.3 Variation of Parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 651
16.5 Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 653
16.5.1 Piecewise Continuous Coefficients and Inhomogeneities . . . . . . . . . . . . . . . . . . . . 654
16.6 Well-Posed Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 659
16.7 Equations in the Complex Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 661
16.7.1 Ordinary Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 661
16.7.2 Regular Singular Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 664
16.7.3 Irregular Singular Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 669
16.7.4 The Point at Infinity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 671
16.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 674
16.9 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 680
16.10Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 683

17 First Order Systems of Differential Equations 705

17.1 Matrices and Jordan Canonical Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 705
17.2 Systems of Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 713
17.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 719
17.4 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 725
17.5 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 727

18 Theory of Linear Ordinary Differential Equations 757

18.1 Nature of Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 758
18.2 Transformation to a First Order System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 761

7
 18.3 The Wronskian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 762
18.3.1 Derivative of a Determinant. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 762
18.3.2 The Wronskian of a Set of Functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 763
18.3.3 The Wronskian of the Solutions to a Differential Equation . . . . . . . . . . . . . . . . . . 765
18.4 Well-Posed Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 768
18.5 The Fundamental Set of Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 770
18.6 Adjoint Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 773
18.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 776
18.8 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 778
18.9 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 780

19 Techniques for Linear Differential Equations 786

19.1 Constant Coefficient Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 786
19.1.1 Second Order Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 787
19.1.2 Higher Order Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 791
19.1.3 Real-Valued Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 793
19.2 Euler Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 795
19.2.1 Real-Valued Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 798
19.3 Exact Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 801
19.4 Equations Without Explicit Dependence on y . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 802
19.5 Reduction of Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 803
19.6 *Reduction of Order and the Adjoint Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 804
19.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 807
19.8 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 814
19.9 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 817

20 Techniques for Nonlinear Differential Equations 842

20.1 Bernoulli Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 842
20.2 Riccati Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 844
20.3 Exchanging the Dependent and Independent Variables . . . . . . . . . . . . . . . . . . . . . . . . 848

8
 20.4 Autonomous Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 850
20.5 *Equidimensional-in-x Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 854
20.6 *Equidimensional-in-y Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 856
20.7 *Scale-Invariant Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 859
20.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 860
20.9 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 864
20.10Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 866

21 Transformations and Canonical Forms 878

21.1 The Constant Coefficient Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 878
21.2 Normal Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 881
21.2.1 Second Order Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 881
21.2.2 Higher Order Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 883
21.3 Transformations of the Independent Variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 885
21.3.1 Transformation to the form u” + a(x) u = 0 . . . . . . . . . . . . . . . . . . . . . . . . . . 885
21.3.2 Transformation to a Constant Coefficient Equation . . . . . . . . . . . . . . . . . . . . . . 886
21.4 Integral Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 888
21.4.1 Initial Value Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 889
21.4.2 Boundary Value Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 891
21.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 894
21.6 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 896
21.7 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 897

22 The Dirac Delta Function 904

22.1 Derivative of the Heaviside Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 904
22.2 The Delta Function as a Limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 906
22.3 Higher Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 908
22.4 Non-Rectangular Coordinate Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 908
22.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 910
22.6 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 911

9
 22.7 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 912

23 Inhomogeneous Differential Equations 915

23.1 Particular Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 915
23.2 Method of Undetermined Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 917
23.3 Variation of Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 921
23.3.1 Second Order Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 921
23.3.2 Higher Order Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 925
23.4 Piecewise Continuous Coefficients and Inhomogeneities . . . . . . . . . . . . . . . . . . . . . . . . 927
23.5 Inhomogeneous Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 931
23.5.1 Eliminating Inhomogeneous Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . 931
23.5.2 Separating Inhomogeneous Equations and Inhomogeneous Boundary Conditions . . . . . . 933
23.5.3 Existence of Solutions of Problems with Inhomogeneous Boundary Conditions . . . . . . . 934
23.6 Green Functions for First Order Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 936
23.7 Green Functions for Second Order Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 939
23.7.1 Green Functions for Sturm-Liouville Problems . . . . . . . . . . . . . . . . . . . . . . . . . 949
23.7.2 Initial Value Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 953
23.7.3 Problems with Unmixed Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . 956
23.7.4 Problems with Mixed Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 958
23.8 Green Functions for Higher Order Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 962
23.9 Fredholm Alternative Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 968
23.10Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 975
23.11Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 982
23.12Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 986

24 Difference Equations 1027

24.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1027
24.2 Exact Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1029
24.3 Homogeneous First Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1030
24.4 Inhomogeneous First Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1032

10
 24.5 Homogeneous Constant Coefficient Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1035
24.6 Reduction of Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1038
24.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1040
24.8 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1041
24.9 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1042

25 Series Solutions of Differential Equations 1046

25.1 Ordinary Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1046
25.1.1 Taylor Series Expansion for a Second Order Differential Equation . . . . . . . . . . . . . . 1051
25.2 Regular Singular Points of Second Order Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 1060
25.2.1 Indicial Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1063
25.2.2 The Case: Double Root . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1065
25.2.3 The Case: Roots Differ by an Integer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1069
25.3 Irregular Singular Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1079
25.4 The Point at Infinity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1079
25.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1082
25.6 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1087
25.7 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1089

26 Asymptotic Expansions 1113

26.1 Asymptotic Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1113
26.2 Leading Order Behavior of Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 1117
26.3 Integration by Parts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1126
26.4 Asymptotic Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1133
26.5 Asymptotic Expansions of Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1135
26.5.1 The Parabolic Cylinder Equation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1135

27 Hilbert Spaces 1141

27.1 Linear Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1141
27.2 Inner Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1143

11
 27.3 Norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1144
27.4 Linear Independence. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1147
27.5 Orthogonality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1147
27.6 Gramm-Schmidt Orthogonalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1147
27.7 Orthonormal Function Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1151
27.8 Sets Of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1152
27.9 Least Squares Fit to a Function and Completeness . . . . . . . . . . . . . . . . . . . . . . . . . . . 1158
27.10Closure Relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1161
27.11Linear Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1166
27.12Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1167
27.13Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1168
27.14Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1169

28 Self Adjoint Linear Operators 1171

28.1 Adjoint Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1171
28.2 Self-Adjoint Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1172
28.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1175
28.4 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1176
28.5 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1177

29 Self-Adjoint Boundary Value Problems 1178

29.1 Summary of Adjoint Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1178
29.2 Formally Self-Adjoint Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1179
29.3 Self-Adjoint Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1182
29.4 Self-Adjoint Eigenvalue Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1183
29.5 Inhomogeneous Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1188
29.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1191
29.7 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1192
29.8 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1193

12
30 Fourier Series 1195
30.1 An Eigenvalue Problem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1195
30.2 Fourier Series. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1198
30.3 Least Squares Fit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1204
30.4 Fourier Series for Functions Defined on Arbitrary Ranges . . . . . . . . . . . . . . . . . . . . . . . 1207
30.5 Fourier Cosine Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1210
30.6 Fourier Sine Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1211
30.7 Complex Fourier Series and Parseval’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1212
30.8 Behavior of Fourier Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1215
30.9 Gibb’s Phenomenon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1224
30.10Integrating and Differentiating Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1224
30.11Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1229
30.12Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1238
30.13Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1241

31 Regular Sturm-Liouville Problems 1291

31.1 Derivation of the Sturm-Liouville Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1291
31.2 Properties of Regular Sturm-Liouville Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1294
31.3 Solving Differential Equations With Eigenfunction Expansions . . . . . . . . . . . . . . . . . . . . 1305
31.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1311
31.5 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1315
31.6 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1317

32 Integrals and Convergence 1342

32.1 Uniform Convergence of Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1342
32.2 The Riemann-Lebesgue Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1343
32.3 Cauchy Principal Value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1345
32.3.1 Integrals on an Infinite Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1345
32.3.2 Singular Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1345

13
33 The Laplace Transform 1347
33.1 The Laplace Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1347
33.2 The Inverse Laplace Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1349
33.2.1 F(s) with Poles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1352
33.2.2 fˆ(s) with Branch Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1357
33.2.3 Asymptotic Behavior of F(s) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1360
33.3 Properties of the Laplace Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1362
33.4 Constant Coefficient Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1366
33.5 Systems of Constant Coefficient Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . 1368
33.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1370
33.7 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1378
33.8 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1382

34 The Fourier Transform 1415

34.1 Derivation from a Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1415
34.2 The Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1417
34.2.1 A Word of Caution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1420
34.3 Evaluating Fourier Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1421
34.3.1 Integrals that Converge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1421
34.3.2 Cauchy Principal Value and Integrals that are Not Absolutely Convergent. . . . . . . . . . 1424
34.3.3 Analytic Continuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1426
34.4 Properties of the Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1428
34.4.1 Closure Relation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1429
34.4.2 Fourier Transform of a Derivative. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1430
34.4.3 Fourier Convolution Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1431
34.4.4 Parseval’s Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1435
34.4.5 Shift Property. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1436
34.4.6 Fourier Transform of x f(x). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1437
34.5 Solving Differential Equations with the Fourier Transform . . . . . . . . . . . . . . . . . . . . . . 1437
34.6 The Fourier Cosine and Sine Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1440

14
 34.6.1 The Fourier Cosine Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1440
34.6.2 The Fourier Sine Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1441
34.7 Properties of the Fourier Cosine and Sine Transform . . . . . . . . . . . . . . . . . . . . . . . . . . 1442
34.7.1 Transforms of Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1442
34.7.2 Convolution Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1444
34.7.3 Cosine and Sine Transform in Terms of the Fourier Transform . . . . . . . . . . . . . . . . 1446
34.8 Solving Differential Equations with the Fourier Cosine and Sine Transforms . . . . . . . . . . . . . 1447
34.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1449
34.10Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1455
34.11Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1458

35 The Gamma Function 1484

35.1 Euler’s Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1484
35.2 Hankel’s Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1486
35.3 Gauss’ Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1488
35.4 Weierstrass’ Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1490
35.5 Stirling’s Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1492
35.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1497
35.7 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1498
35.8 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1499

36 Bessel Functions 1501

36.1 Bessel’s Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1501
36.2 Frobeneius Series Solution about z = 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1502
36.2.1 Behavior at Infinity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1505
36.3 Bessel Functions of the First Kind . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1507
36.3.1 The Bessel Function Satisfies Bessel’s Equation . . . . . . . . . . . . . . . . . . . . . . . . 1508
36.3.2 Series Expansion of the Bessel Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1509
36.3.3 Bessel Functions of Non-Integral Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1512
36.3.4 Recursion Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1515

15
 36.3.5 Bessel Functions of Half-Integral Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1518
36.4 Neumann Expansions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1519
36.5 Bessel Functions of the Second Kind . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1523
36.6 Hankel Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1525
36.7 The Modified Bessel Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1525
36.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1529
36.9 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1534
36.10Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1536

V Partial Differential Equations 1559

37 Transforming Equations 1560
37.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1561
37.2 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1562
37.3 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1563

38 Classification of Partial Differential Equations 1564

38.1 Classification of Second Order Quasi-Linear Equations . . . . . . . . . . . . . . . . . . . . . . . . 1564
38.1.1 Hyperbolic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1565
38.1.2 Parabolic equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1570
38.1.3 Elliptic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1570
38.2 Equilibrium Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1572
38.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1574
38.4 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1575
38.5 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1576

39 Separation of Variables 1580

39.1 Eigensolutions of Homogeneous Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1580
39.2 Homogeneous Equations with Homogeneous Boundary Conditions . . . . . . . . . . . . . . . . . . 1580

16
 39.3 Time-Independent Sources and Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . 1582
39.4 Inhomogeneous Equations with Homogeneous Boundary Conditions . . . . . . . . . . . . . . . . . 1585
39.5 Inhomogeneous Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1587
39.6 The Wave Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1589
39.7 General Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1593
39.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1594
39.9 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1608
39.10Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1613

40 Finite Transforms 1690

40.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1694
40.2 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1695
40.3 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1696

41 Waves 1701
41.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1702
41.2 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1708
41.3 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1710

42 The Diffusion Equation 1727

42.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1728
42.2 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1730
42.3 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1731

43 Similarity Methods 1734

43.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1739
43.2 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1740
43.3 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1741

17
44 Method of Characteristics 1743
44.1 The Method of Characteristics and the Wave Equation . . . . . . . . . . . . . . . . . . . . . . . . 1743
44.2 The Method of Characteristics for an Infinite Domain . . . . . . . . . . . . . . . . . . . . . . . . . 1745
44.3 The Method of Characteristics for a Semi-Infinite Domain . . . . . . . . . . . . . . . . . . . . . . 1746
44.4 Envelopes of Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1747
44.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1750
44.6 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1752
44.7 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1753

45 Transform Methods 1759

45.1 Fourier Transform for Partial Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 1759
45.2 The Fourier Sine Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1761
45.3 Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1762
45.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1763
45.5 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1768
45.6 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1771

46 Green Functions 1794

46.1 Inhomogeneous Equations and Homogeneous Boundary Conditions . . . . . . . . . . . . . . . . . 1794
46.2 Homogeneous Equations and Inhomogeneous Boundary Conditions . . . . . . . . . . . . . . . . . 1795
46.3 Eigenfunction Expansions for Elliptic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1797
46.4 The Method of Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1802
46.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1803
46.6 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1808
46.7 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1810

47 Conformal Mapping 1851

47.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1852
47.2 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1855
47.3 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1856

18
48 Non-Cartesian Coordinates 1864
48.1 Spherical Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1864
48.2 Laplace’s Equation in a Disk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1865
48.3 Laplace’s Equation in an Annulus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1868

VI Calculus of Variations 1872

49 Calculus of Variations 1873
49.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1874
49.2 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1891
49.3 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1897

VII Nonlinear Differential Equations 1990

50 Nonlinear Ordinary Differential Equations 1991
50.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1992
50.2 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1997
50.3 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1999

51 Nonlinear Partial Differential Equations 2021

51.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2022
51.2 Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2025
51.3 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2026

VIII Appendices 2044

A Greek Letters 2045

19
B Notation 2047

C Formulas from Complex Variables 2049

D Table of Derivatives 2052

E Table of Integrals 2056

F Definite Integrals 2060

G Table of Sums 2063

H Table of Taylor Series 2066

I Table of Laplace Transforms 2069

J Table of Fourier Transforms 2074

K Table of Fourier Transforms in n Dimensions 2077

L Table of Fourier Cosine Transforms 2078

M Table of Fourier Sine Transforms 2080

N Table of Wronskians 2082

O Sturm-Liouville Eigenvalue Problems 2084

P Green Functions for Ordinary Differential Equations 2086

Q Trigonometric Identities 2089

Q.1 Circular Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2089

20
 Q.2 Hyperbolic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2091

R Bessel Functions 2094

R.1 Definite Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2094

S Formulas from Linear Algebra 2095

T Vector Analysis 2097

U Partial Fractions 2099

V Finite Math 2103

W Probability 2104
W.1 Independent Events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2104
W.2 Playing the Odds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2105

X Economics 2106

Y Glossary 2107

21
Chapter 1

Anti-Copyright

Anti-Copyright @ 1995-2000 by Mauch Publishing Company, un-Incorporated.

No rights reserved. Any part of this publication by be reproduced, stored in a retrieval system, transmitted or
desecrated without permission.

22
Chapter 2

Preface

During the summer before my final undergraduate year at Caltech I set out to write a math text unlike any
other, namely, one written by me. In that respect I have succeeded beautifully. Unfortunately, the text is neither
complete nor polished. I have a “Warnings and Disclaimers” section below that is a little amusing, and an
appendix on probability that I feel concisesly captures the essence of the subject. However, all the material in
between is in some stage of development. I am currently working to improve and expand this text.
This text is freely available from my web set. Currently I’m at http://www.its.caltech.edu/˜sean. I post
new versions a couple of times a year.

2.1 Advice to Teachers

If you have something worth saying, write it down.

2.2 Acknowledgments
I would like to thank Professor Saffman for advising me on this project and the Caltech SURF program for
providing the funding for me to write the first edition of this book.

23
2.3 Warnings and Disclaimers
• This book is a work in progress. It contains quite a few mistakes and typos. I would greatly appreciate
your constructive criticism. You can reach me at ‘[email protected]’.

• Reading this book impairs your ability to drive a car or operate machinery.

• This book has been found to cause drowsiness in laboratory animals.

• This book contains twenty-three times the US RDA of fiber.

• Caution: FLAMMABLE - Do not read while smoking or near a fire.

• If infection, rash, or irritation develops, discontinue use and consult a physician.

• Warning: For external use only. Use only as directed. Intentional misuse by deliberately concentrating
contents can be harmful or fatal. KEEP OUT OF REACH OF CHILDREN.

• In the unlikely event of a water landing do not use this book as a flotation device.

• The material in this text is fiction; any resemblance to real theorems, living or dead, is purely coincidental.

• This is by far the most amusing section of this book.

• Finding the typos and mistakes in this book is left as an exercise for the reader. (Eye ewes a spelling
chequer from thyme too thyme, sew their should knot bee two many misspellings. Though I ain’t so sure
the grammar’s too good.)

• The theorems and methods in this text are subject to change without notice.

• This is a chain book. If you do not make seven copies and distribute them to your friends within ten days
of obtaining this text you will suffer great misfortune and other nastiness.

• The surgeon general has determined that excessive studying is detrimental to your social life.

24
 • This text has been buffered for your protection and ribbed for your pleasure.

• Stop reading this rubbish and get back to work!

2.4 Suggested Use

This text is well suited to the student, professional or lay-person. It makes a superb gift. This text has a boquet
that is light and fruity, with some earthy undertones. It is ideal with dinner or as an apertif. Bon apetit!

2.5 About the Title

The title is only making light of naming conventions in the sciences and is not an insult to engineers. If you want to
find a good math text to learn a subject, look for books with “Introduction” and “Elementary” in the title. If it is
an “Intermediate” text it will be incomprehensible. If it is “Advanced” then not only will it be incomprehensible,
it will have low production qualities, i.e. a crappy typewriter font, no graphics and no examples. There is an
exception to this rule when the title also contains the word “Scientists” or “Engineers”. Then an advanced book
may be quite suitable for actually learning the material.

25
Part I

Algebra

1
Chapter 3

Sets and Functions

3.1 Sets
Definition. A set is a collection of objects. We call the objects, elements. A set is denoted by listing the
elements between braces. For example: {e, i, π, 1}. We use ellipses to indicate patterns. The set of positive
integers is {1, 2, 3, . . . }. We also denote a sets with the notation {x|conditions on x} for sets that are more easily
described than enumerated. This is read as “the set of elements x such that x satisfies . . . ”. x ∈ S is the notation
for “x is an element of the set S.” To express the opposite we have x 6∈ S for “x is not an element of the set S.”

Examples. We have notations for denoting some of the commonly encountered sets.

• ∅ = {} is the empty set, the set containing no elements.

• Z = {. . . , −1, 0, 1 . . . } is the set of integers. (Z is for “Zahlen”, the German word for “number”.)

• Q = {p/q|p, q ∈ Z, q 6= 0} is the set of rational numbers. (Q is for quotient.)

• R = {x|x = a1 a2 · · · an .b1 b2 · · · } is the set of real numbers, i.e. the set of numbers with decimal expansions.

2
 • C = {a + ib|a, b ∈ R, i2 = −1} is the set of complex numbers. i is the square root of −1. (If you haven’t
seen complex numbers before, don’t dismay. We’ll cover them later.)

• Z+ , Q+ and R+ are the sets of positive integers, rationals and reals, respectively. For example, Z+ =
{1, 2, 3, . . . }.

• Z0+ , Q0+ and R0+ are the sets of non-negative integers, rationals and reals, respectively. For example,
Z0+ = {0, 1, 2, . . . }.

• (a . . . b) denotes an open interval on the real axis. (a . . . b) ≡ {x|x ∈ R, a < x < b}

• We use brackets to denote the closed interval. [a . . . b] ≡ {x|x ∈ R, a ≤ x ≤ b}

The cardinality or order of a set S is denoted |S|. For finite sets, the cardinality is the number of elements
in the set. The Cartesian product of two sets is the set of ordered pairs:

X × Y ≡ {(x, y)|x ∈ X, y ∈ Y }.

The Cartesian product of n sets is the set of ordered n-tuples:

X1 × X2 × · · · × Xn ≡ {(x1 , x2 , . . . , xn )|x1 ∈ X1 , x2 ∈ X2 , . . . , xn ∈ Xn }.

Equality. Two sets S and T are equal if each element of S is an element of T and vice versa. This is denoted,
S = T . Inequality is S 6= T , of course. S is a subset of T , S ⊆ T , if every element of S is an element of T . S is
a proper subset of T , S ⊂ T , if S ⊆ T and S 6= T . For example: The empty set is a subset of every set, ∅ ⊆ S.
The rational numbers are a proper subset of the real numbers, Q ⊂ R.

Operations. The union of two sets, S ∪ T , is the set whose elements are in either of the two sets. The union
of n sets,

∪nj=1 Sj ≡ S1 ∪ S2 ∪ · · · ∪ Sn

3
is the set whose elements are in any of the sets Sj . The intersection of two sets, S ∩ T , is the set whose elements
are in both of the two sets. In other words, the intersection of two sets in the set of elements that the two sets
have in common. The intersection of n sets,

∩nj=1 Sj ≡ S1 ∩ S2 ∩ · · · ∩ Sn

is the set whose elements are in all of the sets Sj . If two sets have no elements in common, S ∩ T = ∅, then the
sets are disjoint. If T ⊆ S, then the difference between S and T , S \ T , is the set of elements in S which are not
in T .

S \ T ≡ {x|x ∈ S, x 6∈ T }

The difference of sets is also denoted S − T .

Properties. The following properties are easily verified from the above definitions.
• S ∪ ∅ = S, S ∩ ∅ = ∅, S \ ∅ = S, S \ S = ∅.
• Commutative. S ∪ T = T ∪ S, S ∩ T = T ∩ S.
• Associative. (S ∪ T ) ∪ U = S ∪ (T ∪ U ) = S ∪ T ∪ U , (S ∩ T ) ∩ U = S ∩ (T ∩ U ) = S ∩ T ∩ U .
• Distributive. S ∪ (T ∩ U ) = (S ∪ T ) ∩ (S ∪ U ), S ∩ (T ∪ U ) = (S ∩ T ) ∪ (S ∩ U ).

3.2 Single Valued Functions

Single-Valued Functions. A single-valued function or single-valued mapping is a mapping of the elements
f
x ∈ X into elements y ∈ Y . This is expressed notationally as f : X → Y or X → Y . If such a function is
well-defined, then for each x ∈ X there exists a unique element of y such that f (x) = y. The set X is the domain
of the function, Y is the codomain, (not to be confused with the range, which we introduce shortly). To denote
the value of a function on a particular element we can use any of the notations: f (x) = y, f : x 7→ y or simply
x 7→ y. f is the identity map on X if f (x) = x for all x ∈ X.

4
 Let f : X → Y . The range or image of f is

f (X) = {y|y = f (x) for some x ∈ X}.

The range is a subset of the codomain. For each Z ⊆ Y , the inverse image of Z is defined:

f −1 (Z) ≡ {x ∈ X|f (x) = z for some z ∈ Z}.

Examples.

• Finite polynomials and the exponential function are examples of single valued functions which map real
numbers to real numbers.

• The greatest integer function, b·c, is a mapping from R to Z. bxc in the greatest integer less than or equal
to x. Likewise, the least integer function, dxe, is the least integer greater than or equal to x.

The -jectives. A function is injective if for each x1 6= x2 , f (x1 ) 6= f (x2 ). In other words, for each x in the
domain there is a unique y = f (x) in the range. f is surjective if for each y in the codomain, there is an x such
that y = f (x). If a function is both injective and surjective, then it is bijective. A bijective function is also called
a one-to-one mapping.

Examples.

• The exponential function y = ex is a bijective function, (one-to-one mapping), that maps R to R+ . (R is

the set of real numbers; R+ is the set of positive real numbers.)

• f (x) = x2 is a bijection from R+ to R+ . f is not injective from R to R+ . For each positive y in the range,
there are two values of x such that y = x2 .

• f (x) = sin x is not injective from R to [−1..1]. For each y ∈ [−1, 1] there exists an infinite number of values
of x such that y = sin x.

5
 Injective Surjective Bijective

Figure 3.1: Depictions of Injective, Surjective and Bijective Functions

3.3 Inverses and Multi-Valued Functions

If y = f (x), then we can write x = f −1 (y) where f −1 is the inverse of f . If y = f (x) is a one-to-one function,
then f −1 (y) is also a one-to-one function. In this case, x = f −1 (f (x)) = f (f −1 (x)) for values of x where both
f (x) and f −1 (x) are defined. For example log x, which maps R+ to R is the inverse of ex . x = elog x = log( ex )
for all x ∈ R+ . (Note the x ∈ R+ ensures that log x is defined.)

If y = f (x) is a many-to-one function, then x = f −1 (y) is a one-to-many function. f −1 (y) is a multi-valued

function. We have x = f (f −1 (x)) for values of x where f −1 (x) is defined, however x 6= f −1 (f (x)). There are
diagrams showing one-to-one, many-to-one and one-to-many functions in Figure 3.2.

Example 3.3.1 y = x2 , a many-to-one function has the inverse x = y 1/2 . For each positive y, there are two
values of x such that x = y 1/2 . y = x2 and y = x1/2 are graphed in Figure 3.3.

there are two branches of y = x1/2

We say that √ √: the positive
√ and the negative branch. We denote the positive
√
1/2
branch as y = x; the negative branch is y = − x. We √ 2 call x the principal branch of x . Note that
√ x
1/2 2 2 1/2 2 1/2
is a one-to-one function. Finally, x = (x ) since (± x) = x, but x 6= (x ) since (x ) = ±x. y = x is

6
 one-to-one many-to-one one-to-many

domain range domain range domain range

Figure 3.2: Diagrams of One-To-One, Many-To-One and One-To-Many Functions

Figure 3.3: y = x2 and y = x1/2

graphed in Figure 3.4.

√
Figure 3.4: y = x

7
 Now consider the many-to-one function y = sin x. The inverse is x = arcsin y. For each y ∈ [−1, 1] there are
an infinite number of values x such that x = arcsin y. In Figure 3.5 is a graph of y = sin x and a graph of a few
branches of y = arcsin x.

Figure 3.5: y = sin x and y = arcsin x

Example 3.3.2 arcsin x has an infinite number of branches. We will denote the principal branch by Arcsin x
which maps [−1, 1] to − π2 , π2 . Note that x = sin(arcsin x), but x 6= arcsin(sin x). y = Arcsin x in Figure 3.6.

Figure 3.6: y = Arcsin x

Example 3.3.3 Consider 11/3 . Since x3 is a one-to-one function, x1/3 is a single-valued function. (See Figure 3.7.)
11/3 = 1.

8
 Figure 3.7: y = x3 and y = x1/3

Example 3.3.4 Consider arccos(1/2). cos x and a few branches of arccos x are graphed in Figure 3.8. cos x = 1/2

Figure 3.8: y = cos x and y = arccos x

has the two solutions x = ±π/3 in the range x ∈ [−π, π]. Since cos(x + π) = − cos x,

arccos(1/2) = {±π/3 + nπ}.

3.4 Transforming Equations

We must take care in applying functions to equations. It is always safe to apply a one-to-one function to an
equation, (provided it is defined for that domain). For example, we can apply y = x3 or y = ex to the equation
x = 1. The equations x3 = 1 and ex = e have the unique solution x = 1.

9
 If we apply a many-to-one function to an equation, we may introduce spurious solutions. Applying y = x2 and
2
y = sin x to the equation x = π2 results in x2 = π4 and sin x = 1. The former equation has the two solutions
x = ± π2 ; the latter has the infinite number of solutions x = π2 + 2nπ, n ∈ Z.

We do not generally apply a one-to-many function to both sides of an equation as this rarely is useful. Consider
the equation

sin2 x = 1.

Applying the function f (x) = x1/2 to the equation would not get us anywhere

(sin2 x)1/2 = 11/2 .

Since (sin2 x)1/2 6= sin x, we cannot simplify the left side of the equation. Instead we could use the definition of
f (x) = x1/2 as the inverse of the x2 function to obtain

sin x = 11/2 = ±1.

Then we could use the definition of arcsin as the inverse of sin to get

x = arcsin(±1).

x = arcsin(1) has the solutions x = π/2 + 2nπ and x = arcsin(−1) has the solutions x = −π/2 + 2nπ. Thus
π
x= + nπ, n ∈ Z.
2
Note that we cannot just apply arcsin to both sides of the equation as arcsin(sin x) 6= x.

10
Chapter 4

Vectors

4.1 Vectors
4.1.1 Scalars and Vectors
A vector is a quantity having both a magnitude and a direction. Examples of vector quantities are velocity, force
and position. One can represent a vector in n-dimensional space with an arrow whose initial point is at the origin,
(Figure 4.1). The magnitude is the length of the vector. Typographically, variables representing vectors are often
written in capital letters, bold face or with a vector over-line, A, a, ~a. The magnitude of a vector is denoted |a|.
A scalar has only a magnitude. Examples of scalar quantities are mass, time and speed.

Vector Algebra. Two vectors are equal if they have the same magnitude and direction. The negative of a
vector, denoted −a, is a vector of the same magnitude as a but in the opposite direction. We add two vectors a
and b by placing the tail of b at the head of a and defining a + b to be the vector with tail at the origin and
head at the head of b. (See Figure 4.2.)
The difference, a − b, is defined as the sum of a and the negative of b, a + (−b). The result of multiplying
a by a scalar α is a vector of magnitude |α| |a| with the same/opposite direction if α is positive/negative. (See
Figure 4.2.)

11
 z

Figure 4.1: Graphical Representation of a Vector in Three Dimensions

b 2a
a
a
a+b -a

Figure 4.2: Vector Arithmetic

Here are the properties of adding vectors and multiplying them by a scalar. They are evident from geometric
considerations.

a+b=b+a αa = aα commutative laws

(a + b) + c = a + (b + c) α(βa) = (αβ)a associative laws
α(a + b) = αa + αb (α + β)a = αa + βa distributive laws

12
Zero and Unit Vectors. The additive identity element for vectors is the zero vector or null vector. This is a
vector of magnitude zero which is denoted as 0. A unit vector is a vector of magnitude one. If a is nonzero then
a/|a| is a unit vector in the direction of a. Unit vectors are often denoted with a caret over-line, n̂.

Rectangular Unit Vectors. In n dimensional Cartesian space, Rn , the unit vectors in the directions of the
coordinates axes are e1 , . . . en . These are called the rectangular unit vectors. To cut down on subscripts, the unit
vectors in three dimensional space are often denoted with i, j and k. (Figure 4.3).

j
y

Figure 4.3: Rectangular Unit Vectors

Components of a Vector. Consider a vector a with tail at the origin and head having the Cartesian coordinates
(a1 , . . . , an ). We can represent this vector as the sum of n rectangular component vectors, a = a1 e1 + · · · + an en .
(See Figure 4.4.) Another p notation for the vector a is ha1 , . . . , an i. By the Pythagorean theorem, the magnitude
of the vector a is |a| = a21 + · · · + a2n .

13
 z

a
a3 k
y
a1 i

a2 j

Figure 4.4: Components of a Vector

4.1.2 The Kronecker Delta and Einstein Summation Convention

The Kronecker Delta tensor is defined
(
1 if i = j,
δij =
0 if i 6= j.

This notation will be useful in our work with vectors.

Consider writing a vector in terms of its rectangular

Pn components. Instead of using ellipses: a = a1 e1 +· · ·+an en ,
we could write the expression as a sum: a = i=1 ai ei . We can shorten this notation by leaving out the sum:
a = ai ei , where it is understood that whenever an index is repeated in a term we sum over that index from 1 to
n. This is the Einstein summation convention. A repeated index is called a summation index or a dummy index.
Other indices can take any value from 1 to n and are called free indices.

14
Example 4.1.1 Consider the matrix equation: A · x = b. We can write out the matrix and vectors explicitly.
    
a11 · · · a1n x1 b1
 .. . .. ..   ..  =  ... 
.   .  
 . 
an1 · · · ann xn bn

This takes much less space when we use the summation convention.

aij xj = bi

Here j is a summation index and i is a free index.

4.1.3 The Dot and Cross Product

Dot Product. The dot product or scalar product of two vectors is defined,

a · b ≡ |a||b| cos θ,

where θ is the angle from a to b. From this definition one can derive the following properties:
• a · b = b · a, commutative.
• α(a · b) = (αa) · b = a · (αb), associativity of scalar multiplication.
• a · (b + c) = a · b + a · c, distributive.
• ei ej = δij . In three dimension, this is

i · i = j · j = k · k = 1, i · j = j · k = k · i = 0.

• a · b = ai bi ≡ a1 b1 + · · · + an bn , dot product in terms of rectangular components.

• If a · b = 0 then either a and b are orthogonal, (perpendicular), or one of a and b are zero.

15
The Angle Between Two Vectors. We can use the dot product to find the angle between two vectors, a and
b. From the definition of the dot product,

a · b = |a||b| cos θ.

If the vectors are nonzero, then

 
a·b
θ = arccos .
|a||b|

Example 4.1.2 What is the angle between i and i + j?

 
i · (i + j)
θ = arccos
|i||i + j|
 
1
= arccos √
2
π
= .
4

Parametric Equation of a Line. Consider a line that passes through the point a and is parallel to the vector
t, (tangent). A parametric equation of the line is

x = a + ut, u ∈ R.

Implicit Equation of a Line. Consider a line that passes through the point a and is normal, (orthogonal,
perpendicular), to the vector n. All the lines that are normal to n have the property that x · n is a constant,
where x is any point on the line. (See Figure 4.5.) x · n = 0 is the line that is normal to n and passes through
the origin. The line that is normal to n and passes through the point a is

x · n = a · n.

16
 x n=1 x n= a n

n a
x n=0

x n=-1

Figure 4.5: Equation for a Line

The normal to a line determines an orientation of the line. The normal points in the direction that is above
the line. A point b is (above/on/below) the line if (b − a) · n is (positive/zero/negative). The signed distance of
a point b from the line x · n = a · n is
n
(b − a) · .
|n|

Implicit Equation of a Hyperplane. A hyperplane in Rn is an n − 1 dimensional “sheet” which passes

through a given point and is normal to a given direction. In R3 we call this a plane. Consider a hyperplane that
passes through the point a and is normal to the vector n. All the hyperplanes that are normal to n have the
property that x · n is a constant, where x is any point in the hyperplane. x · n = 0 is the hyperplane that is
normal to n and passes through the origin. The hyperplane that is normal to n and passes through the point a is

x · n = a · n.

The normal determines an orientation of the hyperplane. The normal points in the direction that is above the
hyperplane. A point b is (above/on/below) the hyperplane if (b − a) · n is (positive/zero/negative). The signed

17
distance of a point b from the hyperplane x · n = a · n is
n
(b − a) · .
|n|

Right and Left-Handed Coordinate Systems. Consider a rectangular coordinate system in two dimensions.
Angles are measured from the positive x axis in the direction of the positive y axis. There are two ways of labeling
the axes. (See Figure 4.6.) In one the angle increases in the counterclockwise direction and in the other the angle
increases in the clockwise direction. The former is the familiar Cartesian coordinate system.

y x

θ
θ x y

Figure 4.6: There are Two Ways of Labeling the Axes in Two Dimensions.

There are also two ways of labeling the axes in a three-dimensional rectangular coordinate system. These are
called right-handed and left-handed coordinate systems. See Figure 4.7. Any other labelling of the axes could be
rotated into one of these configurations. The right-handed system is the one that is used by default. If you put
your right thumb in the direction of the z axis in a right-handed coordinate system, then your fingers curl in the
direction from the x axis to the y axis.

Cross Product. The cross product or vector product is defined,

a × b = |a||b| sin θ n,

where θ is the angle from a to b and n is a unit vector that is orthogonal to a and b and in the direction such
that a, b and n form a right-handed system.

18
 z z

k k
j y i x
i j
x y

Figure 4.7: Right and Left Handed Coordinate Systems

You can visualize the direction of a × b by applying the right hand rule. Curl the fingers of your right hand
in the direction from a to b. Your thumb points in the direction of a × b. Warning: Unless you are a lefty, get
in the habit of putting down your pencil before applying the right hand rule.

The dot and cross products behave a little differently. First note that unlike the dot product, the cross product
is not commutative. The magnitudes of a × b and b × a are the same, but their directions are opposite. (See
Figure 4.8.)
Let
a × b = |a||b| sin θ n and b × a = |b||a| sin φ m.
The angle from a to b is the same as the angle from b to a. Since {a, b, n} and {b, a, m} are right-handed systems,
m points in the opposite direction as n. Since a × b = −b × a we say that the cross product is anti-commutative.

Next we note that since

|a × b| = |a||b| sin θ,
the magnitude of a × b is the area of the parallelogram defined by the two vectors. (See Figure 4.9.) The area of
the triangle defined by two vectors is then 12 |a × b|.

19
 a b

a
b a

Figure 4.8: The Cross Product is Anti-Commutative.

b b
b sin θ
a a

Figure 4.9: The Parallelogram and the Triangle Defined by Two Vectors

From the definition of the cross product, one can derive the following properties:

• a × b = −b × a, anti-commutative.

• α(a × b) = (αa) × b = a × (αb), associativity of scalar multiplication.

• a × (b + c) = a × b + a × c, distributive.

• (a × b) × c 6= a × (b × c). The cross product is not associative.

• i × i = j × j = k × k = 0.

20
 • i × j = k, j × k = i, k × i = j.

•
 
 i j k
 
a × b = (a2 b3 − a3 b2 )i + (a3 b1 − a1 b3 )j + (a1 b2 − a2 b1 )k = a1 a2 a3  ,
 b1 b2 b3 

cross product in terms of rectangular components.

• If a · b = 0 then either a and b are parallel or one of a or b is zero.

Scalar Triple Product. Consider the volume of the parallelopiped defined by three vectors. (See Figure 4.10.)
The area of the base is ||b||c| sin θ|, where θ is the angle between b and c. The height is |a| cos φ, where φ is the
angle between b × c and a. Thus the volume of the parallelopiped is |a||b||c| sin θ cos φ.

b c
a
φ
c
θ

Figure 4.10: The Parallelopiped Defined by Three Vectors

Note that

|a · (b × c)| = |a · (|b||c| sin θ n)|

= ||a||b||c| sin θ cos φ| .

21
Thus |a · (b × c)| is the volume of the parallelopiped. a · (b × c) is the volume or the negative of the volume
depending on whether {a, b, c} is a right or left-handed system.
Note that parentheses are unnecessary in a·b×c. There is only one way to interpret the expression. If you did
the dot product first then you would be left with the cross product of a scalar and a vector which is meaningless.
a · b × c is called the scalar triple product.

Plane Defined by Three Points. Three points which are not collinear define a plane. Consider a plane that
passes through the three points a, b and c. One way of expressing that the point x lies in the plane is that the
vectors x − a, b − a and c − a are coplanar. (See Figure 4.11.) If the vectors are coplanar, then the parallelopiped
defined by these three vectors will have zero volume. We can express this in an equation using the scalar triple
product,

(x − a) · (b − a) × (c − a) = 0.

c
a
b

Figure 4.11: Three Points Define a Plane.

22
4.2 Sets of Vectors in n Dimensions
Orthogonality. Consider two n-dimensional vectors

x = (x1 , x2 , . . . , xn ), y = (y1 , y2 , . . . , yn ).

The inner product of these vectors can be defined

n
X
hx|yi ≡ x · y = xi y i .
i=1

The vectors are orthogonal if x·y = 0. The norm of a vector is the length of the vector generalized to n dimensions.
√
kxk = x·x

Consider a set of vectors

{x1 , x2 , . . . , xm }.

If each pair of vectors in the set is orthogonal, then the set is orthogonal.

xi · xj = 0 if i 6= j

If in addition each vector in the set has norm 1, then the set is orthonormal.
(
1 if i = j
xi · xj = δij =
0 6 j
if i =

Here δij is known as the Kronecker delta function.

23
Completeness. A set of n, n-dimensional vectors

{x1 , x2 , . . . , xn }

is complete if any n-dimensional vector can be written as a linear combination of the vectors in the set. That is,
any vector y can be written
n
X
y= ci xi .
i=1

Taking the inner product of each side of this equation with xm ,

n
!
X
y · xm = ci xi · xm
i=1
n
X
= ci xi · xm
i=1
= cm xm · xm
y · xm
cm =
kxm k2
Thus y has the expansion
n
X y · xi
y= xi .
i=1
kxi k2

If in addition the set is orthonormal, then

n
X
y= (y · xi )xi .
i=1

24
4.3 Exercises
The Dot and Cross Product
Exercise 4.1
Prove the distributive law for the dot product,

a · (b + c) = a · b + a · c.

Exercise 4.2
Prove that

a · b = ai bi ≡ a1 b1 + · · · + an bn .

Exercise 4.3
What is the angle between the vectors i + j and i + 3j?

Exercise 4.4
Prove the distributive law for the cross product,

a × (b + c) = a × b + a × b.

Exercise 4.5
Show that
 
 i j k
 
a × b = a1 a2 a3 
 b1 b2 b3 

25
Exercise 4.6
What is the area of the quadrilateral with vertices at (1, 1), (4, 2), (3, 7) and (2, 3)?

Exercise 4.7
What is the volume of the tetrahedron with vertices at (1, 1, 0), (3, 2, 1), (2, 4, 1) and (1, 2, 5)?

Exercise 4.8
What is the equation of the plane that passes through the points (1, 2, 3), (2, 3, 1) and (3, 1, 2)? What is the
distance from the point (2, 3, 5) to the plane?

26
4.4 Hints
The Dot and Cross Product
Hint 4.1
First prove the distributive law when the first vector is of unit length,
n · (b + c) = n · b + n · c.
Then all the quantities in the equation are projections onto the unit vector n and you can use geometry.

Hint 4.2
First prove that the dot product of a rectangular unit vector with itself is one and the dot product of two distinct
rectangular unit vectors is zero. Then write a and b in rectangular components and use the distributive law.

Hint 4.3
Use a · b = |a||b| cos θ.

Hint 4.4
First consider the case that both b and c are orthogonal to a. Prove the distributive law in this case from
geometric considerations.
Next consider two arbitrary vectors a and b. We can write b = b⊥ + bk where b⊥ is orthogonal to a and bk
is parallel to a. Show that
a × b = a × b⊥ .
Finally prove the distributive law for arbitrary b and c.

Hint 4.5
Write the vectors in their rectangular components and use,
i × j = k, j × k = i, k × i = j,

27
and,

i × i = j × j = k × k = 0.

Hint 4.6
The quadrilateral is composed of two triangles. The area of a triangle defined by the two vectors a and b is
1
2
|a · b|.

Hint 4.7
Justify that the area of a tetrahedron determined by three vectors is one sixth the area of the parallelogram
determined by those three vectors. The area of a parallelogram determined by three vectors is the magnitude of
the scalar triple product of the vectors: a · b × c.

Hint 4.8
The equation of a line that is orthogonal to a and passes through the point b is a · x = a · b. The distance of a
point c from the plane is
 
(c − b) · a 
 
 |a| 

28
4.5 Solutions
The Dot and Cross Product
Solution 4.1
First we prove the distributive law when the first vector is of unit length, i.e.,

n · (b + c) = n · b + n · c. (4.1)

From Figure 4.12 we see that the projection of the vector b + c onto n is equal to the sum of the projections b · n
and c · n.
c
n
b b+c

nc
nb n (b+c)

Figure 4.12: The Distributive Law for the Dot Product

Now we extend the result to the case when the first vector has arbitrary length. We define a = |a|n and
multiply Equation 4.1 by the scalar, |a|.

|a|n · (b + c) = |a|n · b + |a|n · c

a · (b + c) = a · b + a · c.

29
Solution 4.2
First note that
ei · ei = |ei ||ei | cos(0) = 1.
Then note that that dot product of any two distinct rectangular unit vectors is zero because they are orthogonal.
Now we write a and b in terms of their rectangular components and use the distributive law.
a · b = ai ei · bj ej
= ai bj ei · ej
= ai bj δij
= ai bi

Solution 4.3
Since a · b = |a||b| cos θ, we have
 
a·b
θ = arccos
|a||b|
when a and b are nonzero.
    √ !
(i + j) · (i + 3j) 4 2 5
θ = arccos = arccos √ √ = arccos ≈ 0.463648
|i + j||i + 3j| 2 10 5

Solution 4.4
First consider the case that both b and c are orthogonal to a. b + c is the diagonal of the parallelogram defined
by b and c, (see Figure 4.13). Since a is orthogonal to each of these vectors, taking the cross product of a with
these vectors has the effect of rotating the vectors through π/2 radians about a and multiplying their length by
|a|. Note that a × (b + c) is the diagonal of the parallelogram defined by a × b and a × c. Thus we see that the
distributive law holds when a is orthogonal to both b and c,
a × (b + c) = a × b + a × c.

30
 a

b a c

b+c c
a b

a (b+c)

Figure 4.13: The Distributive Law for the Cross Product

Now consider two arbitrary vectors a and b. We can write b = b⊥ + bk where b⊥ is orthogonal to a and bk
is parallel to a, (see Figure 4.14).

b
b
θ

Figure 4.14: The Vector b Written as a Sum of Components Orthogonal and Parallel to a

31
 By the definition of the cross product,

a × b = |a||b| sin θ n.

Note that

|b⊥ | = |b| sin θ,

and that a × b⊥ is a vector in the same direction as a × b. Thus we see that

a × b = |a||b| sin θ n
= |a|(sin θ|b|)n
= |a||b⊥ |n = |a||b⊥ | sin(π/2)n

a × b = a × b⊥ .

Now we are prepared to prove the distributive law for arbitrary b and c.

a × (b + c) = a × (b⊥ + bk + c⊥ + ck )
= a × ((b + c)⊥ + (b + c)k )
= a × ((b + c)⊥ )
= a × b⊥ + a × c⊥
=a×b+a×c

a × (b + c) = a × b + a × c

Solution 4.5
We know that

i × j = k, j × k = i, k × i = j,

32
and that
i × i = j × j = k × k = 0.
Now we write a and b in terms of their rectangular components and use the distributive law to expand the cross
product.
a × b = (a1 i + a2 j + a3 k) × (b1 i + b2 j + b3 k)
= a1 i × (b1 i + b2 j + b3 k) + a2 j × (b1 i + b2 j + b3 k) + a3 k × (b1 i + b2 j + b3 k)
= a1 b2 k + a1 b3 (−j) + a2 b1 (−k) + a2 b3 i + a3 b1 j + a3 b2 (−i)
= (a2 b3 − a3 b2 )i − (a1 b3 − a3 b1 )j + (a1 b2 − a2 b1 )k
Next we evaluate the determinant.
 
 i j k      
a1 a2 a3  = i a2 a3  − j a1 a3  + k a1 a2 
       
   b2 b3   b1 b3   b1 b2 
 b1 b2 b3 
= (a2 b3 − a3 b2 )i − (a1 b3 − a3 b1 )j + (a1 b2 − a2 b1 )k
Thus we see that,
 
 i j k
 
a × b = a1 a2 a3 
 b1 b2 b3 

Solution 4.6
The area area of the quadrilateral is the area of two triangles. The first triangle is defined by the vector from
(1, 1) to (4, 2) and the vector from (1, 1) to (2, 3). The second triangle is defined by the vector from (3, 7) to (4, 2)
and the vector from (3, 7) to (2, 3). (See Figure 4.15.) The area of a triangle defined by the two vectors a and b
is 12 |a · b|. The area of the quadrilateral is then,
1 1 1 1
|(3i + j) · (i + 2j)| + |(i − 5j) · (−i − 4j)| = (5) + (19) = 12.
2 2 2 2

33
 y (3,7)

(2,3)
(4,2)
(1,1) x

Figure 4.15: Quadrilateral

Solution 4.7
The tetrahedron is determined by the three vectors with tail at (1, 1, 0) and heads at (3, 2, 1), (2, 4, 1) and
(1, 2, 5). These are h2, 1, 1i, h1, 3, 1i and h0, 1, 5i. The area of the tetrahedron is one sixth the area of the
parallelogram determined by these vectors. (This is because the area of a pyramid is 31 (base)(height). The base
of the tetrahedron is half the area of the parallelogram and the heights are the same. 12 13 = 16 ) Thus the area of
a tetrahedron determined by three vectors is 61 |a · b × c|. The area of the tetrahedron is
1 1 7
|h2, 1, 1i · h1, 3, 1i × h1, 2, 5i| = |h2, 1, 1i · h13, −4, −1i| =
6 6 2

Solution 4.8
The two vectors with tails at (1, 2, 3) and heads at (2, 3, 1) and (3, 1, 2) are parallel to the plane. Taking the cross
product of these two vectors gives us a vector that is orthogonal to the plane.

h1, 1, −2i × h2, −1, −1i = h−3, −3, −3i

We see that the plane is orthogonal to the vector h1, 1, 1i and passes through the point (1, 2, 3). The equation of

34
the plane is

h1, 1, 1i · hx, y, zi = h1, 1, 1i · h1, 2, 3i,

x + y + z = 6.

Consider the vector with tail at (1, 2, 3) and head at (2, 3, 5). The magnitude of the dot product of this vector
with the unit normal vector gives the distance from the plane.
  √

h1, 1, 2i · h1, 1, 1i  4
= √ = 4 3
 |h1, 1, 1i|  3 3

35
Part II

Calculus

36
Chapter 5

Differential Calculus

5.1 Limits of Functions

Definition of a Limit. If the value of the function y(x) gets arbitrarily close to η as x approaches the point ξ,
then wee say that the limit of the function as x approaches ξ is equal to η. This is written:

lim y(x) = η
x→ξ

To make the notion of “arbitrarily close” precise: for any  > 0 there exists a δ > 0 such that |y(x) − η| <  for
all 0 < |x − ξ| < δ. That is, there is an interval surrounding the point x = ξ for which the function is within
 of η. See Figure 5.1. Note that the interval surrounding x = ξ is a deleted neighborhood, that is it does not
contain the point x = ξ. Thus the value function at x = ξ need not be equal to η for the limit to exist. Indeed
the function need not even be defined at x = ξ.
To prove that a function has a limit at a point ξ we first bound |y(x) − η| in terms of δ for values of x satisfying
0 < |x − ξ| < δ. Denote this upper bound by u(δ). Then for an arbitrary  > 0, we determine a δ > 0 such that
the the upper bound u(δ) and hence |y(x) − η| is less than .

37
 y

η+ε

η−ε

x
ξ−δ ξ+δ

Figure 5.1: The δ neighborhood of x = ξ such that |y(x) − η| < .

Example 5.1.1 Show that

lim x2 = 1.
x→1

Consider any  > 0. We need to show that there exists a δ > 0 such that |x2 − 1| <  for all |x − 1| < δ. First we
obtain a bound on |x2 − 1|.
|x2 − 1| = |(x − 1)(x + 1)|
= |x − 1||x + 1|
< δ|x + 1|
= δ|(x − 1) + 2|
< δ(δ + 2)
Now we choose a positive δ such that,
δ(δ + 2) = .
We see that
√
δ= 1 +  − 1,

38
is positive and satisfies the criterion that |x2 − 1| <  for all 0 < |x − 1| < δ. Thus the limit exists.

Note that the value of the function y(ξ) need not be equal to limx→ξ y(x). This is illustrated in Example 5.1.2.

Example 5.1.2 Consider the function

(
1 for x ∈ Z,
y(x) =
0 for x 6∈ Z.
For what values of ξ does limx→ξ y(x) exist?
First consider ξ 6∈ Z. Then there exists an open neighborhood a < ξ < b around ξ such that y(x) is identically
zero for x ∈ (a, b). Then trivially, limx→ξ y(x) = 0.
Now consider ξ ∈ Z. Consider any  > 0. Then if |x − ξ| < 1 then |y(x) − 0| = 0 < . Thus we see that
limx→ξ y(x) = 0.
Thus, regardless of the value of ξ, limx→ξ y(x) = 0.

Left and Right Limits. With the notation limx→ξ+ y(x) we denote the right limit of y(x). This is the limit
as x approaches ξ from above. Mathematically: limx→ξ+ exists if for any  > 0 there exists a δ > 0 such that
|y(x) − η| <  for all 0 < ξ − x < δ. The left limit limx→ξ− y(x) is defined analogously.
sin x
Example 5.1.3 Consider the function, |x|
, defined for x 6= 0. (See Figure 5.2.) The left and right limits exist
as x approaches zero.
sin x sin x
lim+ = 1, lim− = −1
x→0 |x| x→0 |x|
However the limit,
sin x
lim ,
x→0 |x|

does not exist.

39
 Figure 5.2: Plot of sin(x)/|x|.

Properties of Limits. Let limx→ξ u(x) and limx→ξ v(x) exist.

• limx→ξ (au(x) + bv(x)) = a limx→ξ u(x) + b limx→ξ v(x).
• limx→ξ (u(x)v(x)) = (limx→ξ u(x)) (limx→ξ v(x)).
 
u(x) lim u(x)
• limx→ξ v(x) = limx→ξx→ξ v(x)
if limx→ξ v(x) 6= 0.

Example 5.1.4 Prove that if limx→ξ u(x) = µ and limx→ξ v(x) = ν exist then
  
lim (u(x)v(x)) = lim u(x) lim v(x) .
x→ξ x→ξ x→ξ

Assume that µ and ν are nonzero. (The cases where one or both are zero are similar and simpler.)
|u(x)v(x) − µν| = |uv − (u + µ − u)ν|
= |u(v − ν) + (u − µ)ν|
= |u||v − ν| + |u − µ||ν|
A sufficient condition for |u(x)v(x) − µν| <  is
 
|u − µ| < and |v − ν| <  .
2|ν| 2 |µ| + 
2|ν|

40
Since the two right sides of the inequalities are positive, there exists δ1 > 0 and δ2 > 0 such that the first inequality
is satisfied for all |x − ξ| < δ1 and the second inequality is satisfied for all |x − ξ| < δ2 . By choosing δ to be the
smaller of δ1 and δ2 we see that

|u(x)v(x) − µν| <  for all |x − ξ| < δ.

Thus

  
lim (u(x)v(x)) = lim u(x) lim v(x) = µν.
x→ξ x→ξ x→ξ

41
 Result 5.1.1 Definition of a Limit. The statement:

lim y(x) = η
x→ξ

means that y(x) gets arbitrarily close to η as x approaches ξ. For any  > 0 there exists
a δ > 0 such that |y(x) − η| <  for all x in the neighborhood 0 < |x − ξ| < δ. The left
and right limits,

lim y(x) = η and lim y(x) = η

x→ξ − x→ξ +

denote the limiting value as x approaches ξ respectively from below and above. The
neighborhoods are respectively −δ < x − ξ < 0 and 0 < x − ξ < δ.
Properties of Limits. Let limx→ξ u(x) and limx→ξ v(x) exist.
• limx→ξ (au(x) + bv(x)) = a limx→ξ u(x) + b limx→ξ v(x).
• limx→ξ (u(x)v(x)) = (limx→ξ u(x)) (limx→ξ v(x)).
 
u(x) lim u(x)
• limx→ξ v(x) = limx→ξx→ξ v(x)
if limx→ξ v(x) 6= 0.

5.2 Continuous Functions

Definition of Continuity. A function y(x) is said to be continuous at x = ξ if the value of the function is
equal to its limit, that is, limx→ξ y(x) = y(ξ). Note that this one condition is actually the three conditions: y(ξ)
is defined, limx→ξ y(x) exists and limx→ξ y(x) = y(ξ). A function is continuous if it is continuous at each point
in its domain. A function is continuous on the closed interval [a, b] if the function is continuous for each point

42
x ∈ (a, b) and limx→a+ y(x) = y(a) and limx→b− y(x) = y(b).

Discontinuous Functions. If a function is not continuous at a point it is called discontinuous at that point.
If limx→ξ y(x) exists but is not equal to y(ξ), then the function has a removable discontinuity. It is thus named
because we could define a continuous function
(
y(x) for x 6= ξ,
z(x) =
limx→ξ y(x) for x = ξ,

to remove the discontinuity. If both the left and right limit of a function at a point exist, but are not equal, then
the function has a jump discontinuity at that point. If either the left or right limit of a function does not exist,
then the function is said to have an infinite discontinuity at that point.
sin x
Example 5.2.1 x
has a removable discontinuity at x = 0. The Heaviside function,

0
 for x < 0,
H(x) = 1/2 for x = 0,

1 for x > 0,


1
has a jump discontinuity at x = 0. x
has an infinite discontinuity at x = 0. See Figure 5.3.

Figure 5.3: A Removable discontinuity, a Jump Discontinuity and an Infinite Discontinuity

43
Properties of Continuous Functions.

Arithmetic. If u(x) and v(x) are continuous at x = ξ then u(x) ± v(x) and u(x)v(x) are continuous at x = ξ.
u(x)
v(x)
is continuous at x = ξ if v(ξ) 6= 0.

Function Composition. If u(x) is continuous at x = ξ and v(x) is continuous at x = µ = u(ξ) then u(v(x)) is
continuous at x = ξ. The composition of continuous functions is a continuous function.

Boundedness. A function which is continuous on a closed interval is bounded in that closed interval.

Nonzero in a Neighborhood. If y(ξ) 6= 0 then there exists a neighborhood (ξ − , ξ + ),  > 0 of the point ξ such
that y(x) 6= 0 for x ∈ (ξ − , ξ + ).

Intermediate Value Theorem. Let u(x) be continuous on [a, b]. If u(a) ≤ µ ≤ u(b) then there exists ξ ∈ [a, b]
such that u(ξ) = µ. This is known as the intermediate value theorem. A corollary of this is that if u(a) and
u(b) are of opposite sign then u(x) has at least one zero on the interval (a, b).

Maxima and Minima. If u(x) is continuous on [a, b] then u(x) has a maximum and a minimum on [a, b]. That
is, there is at least one point ξ ∈ [a, b] such that u(ξ) ≥ u(x) for all x ∈ [a, b] and there is at least one point
η ∈ [a, b] such that u(η) ≤ u(x) for all x ∈ [a, b].

Piecewise Continuous Functions. A function is piecewise continuous on an interval if the function is bounded
on the interval and the interval can be divided into a finite number of intervals on each of which the function is
continuous. For example, the greatest integer function, bxc, is piecewise continuous. (bxc is defined to the the
greatest integer less than or equal to x.) See Figure 5.4 for graphs of two piecewise continuous functions.

Uniform Continuity. Consider a function f (x) that is continuous on an interval. This means that for any
point ξ in the interval and any positive  there exists a δ > 0 such that |f (x) − f (ξ)| <  for all 0 < |x − ξ| < δ. In
general, this value of δ depends on both ξ and . If δ can be chosen so it is a function of  alone and independent
of ξ then the function is said to be uniformly continuous on the interval. A sufficient condition for uniform
continuity is that the function is continuous on a closed interval.

44
 Figure 5.4: Piecewise Continuous Functions

5.3 The Derivative

Consider a function y(x) on the interval (x . . . x + ∆x) for some ∆x > 0. We define the increment ∆y =
∆y
y(x+∆x)−y(x). The average rate of change, (average velocity), of the function on the interval is ∆x . The average
rate of change is the slope of the secant line that passes through the points (x, y(x)) and (x + ∆x, y(x + ∆x)).
See Figure 5.5.

∆y
∆x
x

Figure 5.5: The increments ∆x and ∆y.

If the slope of the secant line has a limit as ∆x approaches zero then we call this slope the derivative or
dy
instantaneous rate of change of the function at the point x. We denote the derivative by dx , which is a nice

45
 ∆y
notation as the derivative is the limit of ∆x
as ∆x → 0.

dy y(x + ∆x) − y(x)

≡ lim .
dx ∆x→0 ∆x
dy
∆x may approach zero from below or above. It is common to denote the derivative dx d
by dx y, y 0 (x), y 0 or Dy.
A function is said to be differentiable at a point if the derivative exists there. Note that differentiability implies
continuity, but not vice versa.

Example 5.3.1 Consider the derivative of y(x) = x2 at the point x = 1.

y(1 + ∆x) − y(1)

y 0 (1) ≡ lim
∆x→0 ∆x
(1 + ∆x)2 − 1
= lim
∆x→0 ∆x
= lim (2 + ∆x)
∆x→0
=2

Figure 5.6 shows the secant lines approaching the tangent line as ∆x approaches zero from above and below.

Example 5.3.2 We can compute the derivative of y(x) = x2 at an arbitrary point x.

d  2 (x + ∆x)2 − x2
x = lim
dx ∆x→0 ∆x
= lim (2x + ∆x)
∆x→0
= 2x

46
 4 4
3.5 3.5
3 3
2.5 2.5
2 2
1.5 1.5
1 1
0.5 0.5
0.5 1 1.5 2 0.5 1 1.5 2

Figure 5.6: Secant lines and the tangent to x2 at x = 1.

Properties. Let u(x) and v(x) be differentiable. Let a and b be constants. Some fundamental properties of
derivatives are:
d du dv
(au + bv) = a +b Linearity
dx dx dx
d du dv
(uv) = v+u Product Rule
dx dx dx
d u
  v − u dv
du
= dx 2 dx Quotient Rule
dx v v
d a du
(u ) = aua−1 Power Rule
dx dx
d du dv
(u(v(x))) = = u0 (v(x))v 0 (x) Chain Rule
dx dv dx
These can be proved by using the definition of differentiation.

47
Example 5.3.3 Prove the quotient rule for derivatives.

u(x+∆x) u(x)
d u v(x+∆x)
− v(x)
= lim
dx v ∆x→0 ∆x
u(x + ∆x)v(x) − u(x)v(x + ∆x)
= lim
∆x→0 ∆xv(x)v(x + ∆x)
u(x + ∆x)v(x) − u(x)v(x) − u(x)v(x + ∆x) + u(x)v(x)
= lim
∆x→0 ∆xv(x)v(x)
(u(x + ∆x) − u(x))v(x) − u(x)(v(x + ∆x) − v(x))
= lim
∆x→0 ∆xv 2 (x)
u(x+∆x)−u(x) v(x+∆x)−v(x)
lim∆x→0 ∆x
v(x) − u(x) lim∆x→0 ∆x
=
v 2 (x)
v du
dx
dv
− u dx
=
v2

48
Trigonometric Functions. Some derivatives of trigonometric functions are:

d d 1
sin x = cos x arcsin x =
dx dx (1 − x2 )1/2
d d 1
cos x = − sin x arccos x = −
dx dx (1 − x2 )1/2
d 1 d 1
tan x = arctan x =
dx cos2 x dx 1 + x2
d x d 1
e = ex log x =
dx dx x
d d 1
sinh x = cosh x arcsinh x = 2
dx dx (x + 1)1/2
d d 1
cosh x = sinh x arccosh x = 2
dx dx (x − 1)1/2
d 1 d 1
tanh x = arctanh x =
dx cosh2 x dx 1 − x2

Example 5.3.4 We can evaluate the derivative of xx by using the identity ab = eb log a .

d x d x log x
x = e
dx dx
d
= ex log x (x log x)
dx
1
= xx (1 · log x + x )
x
x
= x (1 + log x)

49
Inverse Functions. If we have a function y(x), we can consider x as a function of y, x(y). For example, if
√
y(x) = 8x3 then x(y) = 2 3 y; if y(x) = x+2
x+1
then x(y) = 2−y
y−1
. The derivative of an inverse function is

d 1
x(y) = dy .
dy dx

Example 5.3.5 The inverse function of y(x) = ex is x(y) = log y. We can obtain the derivative of the logarithm
from the derivative of the exponential. The derivative of the exponential is
dy
= ex .
dx
Thus the derivative of the logarithm is
d d 1 1 1
log y = x(y) = dy = x = .
dy dy dx
e y

5.4 Implicit Differentiation

An explicitly defined function has the form y = f (x). A implicitly defined function has the form f (x, y) = 0. A
few examples of implicit functions are x2 + y 2 − 1 = 0 and x + y + sin(xy) = 0. Often it is not possible to write
an implicit equation in explicit form. This is true of the latter example above. One can calculate the derivative
of y(x) in terms of x and y even when y(x) is defined by an implicit equation.

Example 5.4.1 Consider the implicit equation

x2 − xy − y 2 = 1.
This implicit equation can be solved for the dependent variable.
1 √ 
y(x) = −x ± 5x2 − 4 .
2

50
We can differentiate this expression to obtain
 
0 1 5x
y = −1 ± √ .
2 5x2 − 4
One can obtain the same result without first solving for y. If we differentiate the implicit equation, we obtain
dy dy
2x − y − x − 2y = 0.
dx dx
dy
We can solve this equation for dx
.

dy 2x − y
=
dx x + 2y
We can differentiate this expression to obtain the second derivative of y.

d2 y (x + 2y)(2 − y 0 ) − (2x − y)(1 + 2y 0 )

=
dx2 (x + 2y)2
5(y − xy 0 )
=
(x + 2y)2

Substitute in the expression for y 0 .

10(x2 − xy − y 2 )
=−
(x + 2y)2

Use the original implicit equation.

10
=−
(x + 2y)2

51
5.5 Maxima and Minima
A differentiable function is increasing where f 0 (x) > 0, decreasing where f 0 (x) < 0 and stationary where f 0 (x) = 0.
A function f (x) has a relative maxima at a point x = ξ if there exists a neighborhood around ξ such that
f (x) ≤ f (ξ) for x ∈ (x − δ, x + δ), δ > 0. The relative minima is defined analogously. Note that this definition
does not require that the function be differentiable, or even continuous. We refer to relative maxima and minima
collectively are relative extrema.

Relative Extrema and Stationary Points. If f (x) is differentiable and f (ξ) is a relative extrema then x = ξ
is a stationary point, f 0 (ξ) = 0. We can prove this using left and right limits. Assume that f (ξ) is a relative
maxima. Then there is a neighborhood (x − δ, x + δ), δ > 0 for which f (x) ≤ f (ξ). Since f (x) is differentiable
the derivative at x = ξ,
f (ξ + ∆x) − f (ξ)
f 0 (ξ) = lim ,
∆x→0 ∆x
exists. This in turn means that the left and right limits exist and are equal. Since f (x) ≤ f (ξ) for ξ − δ < x < ξ
the left limit is non-positive,
f (ξ + ∆x) − f (ξ)
f 0 (ξ) = lim − ≤ 0.
∆x→0 ∆x
Since f (x) ≤ f (ξ) for ξ < x < ξ + δ the right limit is nonnegative,

f (ξ + ∆x) − f (ξ)
f 0 (ξ) = lim + ≥ 0.
∆x→0 ∆x
Thus we have 0 ≤ f 0 (ξ) ≤ 0 which implies that f 0 (ξ) = 0.

It is not true that all stationary points are relative extrema. That is, f 0 (ξ) = 0 does not imply that x = ξ is
an extrema. Consider the function f (x) = x3 . x = 0 is a stationary point since f 0 (x) = x2 , f 0 (0) = 0. However,
x = 0 is neither a relative maxima nor a relative minima.

52
 It is also not true that all relative extrema are stationary points. Consider the function f (x) = |x|. The point
x = 0 is a relative minima, but the derivative at that point is undefined.

First Derivative Test. Let f (x) be differentiable and f 0 (ξ) = 0.

• If f 0 (x) changes sign from positive to negative as we pass through x = ξ then the point is a relative maxima.

• If f 0 (x) changes sign from negative to positive as we pass through x = ξ then the point is a relative minima.

• If f 0 (x) is not identically zero in a neighborhood of x = ξ and it does not change sign as we pass through
the point then x = ξ is not a relative extrema.

Example 5.5.1 Consider y = x2 and the point x = 0. The function is differentiable. The derivative, y 0 = 2x,
vanishes at x = 0. Since y 0 (x) is negative for x < 0 and positive for x > 0, the point x = 0 is a relative minima.
See Figure 5.7.

Example 5.5.2 Consider y = cos x and the point x = 0. The function is differentiable. The derivative, y 0 =
− sin x is positive for −π < x < 0 and negative for 0 < x < π. Since the sign of y 0 goes from positive to negative,
x = 0 is a relative maxima. See Figure 5.7.

Example 5.5.3 Consider y = x3 and the point x = 0. The function is differentiable. The derivative, y 0 = 3x2 is
positive for x < 0 and positive for 0 < x. Since y 0 is not identically zero and the sign of y 0 does not change, x = 0
is not a relative extrema. See Figure 5.7.

Concavity. If the portion of a curve in some neighborhood of a point lies above the tangent line through that
point, the curve is said to be concave upward. If it lies below the tangent it is concave downward. If a function
is twice differentiable then f 00 (x) > 0 where it is concave upward and f 00 (x) < 0 where it is concave downward.
Note that f 00 (x) > 0 is a sufficient, but not a necessary condition for a curve to be concave upward at a point. A

53
 Figure 5.7: Graphs of x2 , cos x and x3 .

curve may be concave upward at a point where the second derivative vanishes. A point where the curve changes
concavity is called a point of inflection. At such a point the second derivative vanishes, f 00 (x) = 0. For twice
continuously differentiable functions, f 00 (x) = 0 is a necessary but not a sufficient condition for an inflection point.
The second derivative may vanish at places which are not inflection points. See Figure 5.8.

Figure 5.8: Concave Upward, Concave Downward and an Inflection Point.

Second Derivative Test. Let f (x) be twice differentiable and let x = ξ be a stationary point, f 0 (ξ) = 0.

• If f 00 (ξ) < 0 then the point is a relative maxima.

• If f 00 (ξ) > 0 then the point is a relative minima.

• If f 00 (ξ) = 0 then the test fails.

54
Example 5.5.4 Consider the function f (x) = cos x and the point x = 0. The derivatives of the function are
f 0 (x) = − sin x, f 00 (x) = − cos x. The point x = 0 is a stationary point, f 0 (0) = − sin(0) = 0. Since the second
derivative is negative there, f 00 (0) = − cos(0) = −1, the point is a relative maxima.

Example 5.5.5 Consider the function f (x) = x4 and the point x = 0. The derivatives of the function are
f 0 (x) = 4x3 , f 00 (x) = 12x2 . The point x = 0 is a stationary point. Since the second derivative also vanishes at
that point the second derivative test fails. One must use the first derivative test to determine that x = 0 is a
relative minima.

5.6 Mean Value Theorems

Rolle’s Theorem. If f (x) is continuous in [a, b], differentiable in (a, b) and f (a) = f (b) = 0 then there exists
a point ξ ∈ (a, b) such that f 0 (ξ) = 0. See Figure 5.9.

Figure 5.9: Rolle’s Theorem.

To prove this we consider two cases. First we have the trivial case that f (x) ≡ 0. If f (x) is not identically
zero then continuity implies that it must have a nonzero relative maxima or minima in (a, b). Let x = ξ be one
of these relative extrema. Since f (x) is differentiable, x = ξ must be a stationary point, f 0 (ξ) = 0.

55
Theorem of the Mean. If f (x) is continuous in [a, b] and differentiable in (a, b) then there exists a point x = ξ
such that
f (b) − f (a)
f 0 (ξ) = .
b−a
That is, there is a point where the instantaneous velocity is equal to the average velocity on the interval.

Figure 5.10: Theorem of the Mean.

We prove this theorem by applying Rolle’s theorem. Consider the new function
f (b) − f (a)
g(x) = f (x) − f (a) − (x − a)
b−a
Note that g(a) = g(b) = 0, so it satisfies the conditions of Rolle’s theorem. There is a point x = ξ such that
g 0 (ξ) = 0. We differentiate the expression for g(x) and substitute in x = ξ to obtain the result.
f (b) − f (a)
g 0 (x) = f 0 (x) −
b−a

f (b) − f (a)
g 0 (ξ) = f 0 (ξ) − =0
b−a

f (b) − f (a)
f 0 (ξ) =
b−a

56
Generalized Theorem of the Mean. If f (x) and g(x) are continuous in [a, b] and differentiable in (a, b), then
there exists a point x = ξ such that

f 0 (ξ) f (b) − f (a)

0
= .
g (ξ) g(b) − g(a)

We have assumed that g(a) 6= g(b) so that the denominator does not vanish and that f 0 (x) and g 0 (x) are not
simultaneously zero which would produce an indeterminate form. Note that this theorem reduces to the regular
theorem of the mean when g(x) = x. The proof of the theorem is similar to that for the theorem of the mean.

Taylor’s Theorem of the Mean. If f (x) is n + 1 times continuously differentiable in (a, b) then there exists
a point x = ξ ∈ (a, b) such that

(b − a)2 00 (b − a)n (n) (b − a)n+1 (n+1)

f (b) = f (a) + (b − a)f 0 (a) + f (a) + · · · + f (a) + f (ξ). (5.1)
2! n! (n + 1)!

For the case n = 0, the formula is

f (b) = f (a) + (b − a)f 0 (ξ),

which is just a rearrangement of the terms in the theorem of the mean,

f (b) − f (a)
f 0 (ξ) = .
b−a

5.6.1 Application: Using Taylor’s Theorem to Approximate Functions.

One can use Taylor’s theorem to approximate functions with polynomials. Consider an infinitely differentiable
function f (x) and a point x = a. Substituting x for b into Equation 5.1 we obtain,

(x − a)2 00 (x − a)n (n) (x − a)n+1 (n+1)

f (x) = f (a) + (x − a)f 0 (a) + f (a) + · · · + f (a) + f (ξ).
2! n! (n + 1)!

57
If the last term in the sum is small then we can approximate our function with an nth order polynomial.
0 (x − a)2 00 (x − a)n (n)
f (x) ≈ f (a) + (x − a)f (a) + f (a) + · · · + f (a)
2! n!
The last term in Equation 5.6.1 is called the remainder or the error term,
(x − a)n+1 (n+1)
Rn = f (ξ).
(n + 1)!
Since the function is infinitely differentiable, f (n+1) (ξ) exists and is bounded. Therefore we note that the error
must vanish as x → 0 because of the (x − a)n+1 factor. We therefore suspect that our approximation would be a
good one if x is close to a. Also note that n! eventually grows faster than (x − a)n ,
(x − a)n
lim = 0.
n→∞ n!
So if the derivative term, f (n+1) (ξ), does not grow to quickly, the error for a certain value of x will get smaller
with increasing n and the polynomial will become a better approximation of the function. (It is also possible that
the derivative factor grows very quickly and the approximation gets worse with increasing n.)

Example 5.6.1 Consider the function f (x) = ex . We want a polynomial approximation of this function near
the point x = 0. Since the derivative of ex is ex , the value of all the derivatives at x = 0 is f (n) (0) = e0 = 1.
Taylor’s theorem thus states that
x2 x 3 xn xn+1 ξ
ex = 1 + x + + + ··· + + e,
2! 3! n! (n + 1)!
for some ξ ∈ (0, x). The first few polynomial approximations of the exponent about the point x = 0 are
f1 (x) = 1
f2 (x) = 1 + x
x2
f3 (x) = 1 + x +
2
x 2 x3
f4 (x) = 1 + x + +
2 6

58
The four approximations are graphed in Figure 5.11.
2.5 2.5 2.5 2.5
2 2 2 2
1.5 1.5 1.5 1.5
1 1 1 1
0.5 0.5 0.5 0.5
-1 -0.5 0.5 1 -1 -0.5 0.5 1 -1 -0.5 0.5 1 -1 -0.5 0.5 1

Figure 5.11: Four Finite Taylor Series Approximations of ex

Note that for the range of x we are looking at, the approximations become more accurate as the number of
terms increases.

Example 5.6.2 Consider the function f (x) = cos x. We want a polynomial approximation of this function near
the point x = 0. The first few derivatives of f are
f (x) = cos x
f 0 (x) = − sin x
f 00 (x) = − cos x
f 000 (x) = sin x
f (4) (x) = cos x
It’s easy to pick out the pattern here,
(
(−1)n/2 cos x for even n,
f (n) (x) = (n+1)/2
(−1) sin x for odd n.
Since cos(0) = 1 and sin(0) = 0 the n-term approximation of the cosine is,
x2 x 4 x6 x2(n−1) x2n
cos x = 1 − + − + · · · + (−1)2(n−1) + cos ξ.
2! 4! 6! (2(n − 1))! (2n)!

59
Here are graphs of the one, two, three and four term approximations.

1 1 1 1
0.5 0.5 0.5 0.5
-3 -2 -1 1 2 3 -3 -2 -1 1 2 3 -3 -2 -1 1 2 3 -3 -2 -1 1 2 3
-0.5 -0.5 -0.5 -0.5
-1 -1 -1 -1

Figure 5.12: Taylor Series Approximations of cos x

Note that for the range of x we are looking at, the approximations become more accurate as the number of
terms increases. Consider the ten term approximation of the cosine about x = 0,

x2 x 4 x18 x20
cos x = 1 − + − ··· − + cos ξ.
2! 4! 18! 20!

Note that for any value of ξ, | cos ξ| ≤ 1. Therefore the absolute value of the error term satisfies,
 20
 |x|20

x
|R| = 
 cos ξ  ≤ .
20! 20!

x20 /20! is plotted in Figure 5.13.

Note that the error is very small for x < 6, fairly small but non-negligible for x ≈ 7 and large for x > 8. The
ten term approximation of the cosine, plotted below, behaves just we would predict.
The error is very small until it becomes non-negligible at x ≈ 7 and large at x ≈ 8.

Example 5.6.3 Consider the function f (x) = log x. We want a polynomial approximation of this function near

60
 1

0.8

0.6

0.4

0.2

2 4 6 8 10

Figure 5.13: Plot of x20 /20!.

1

0.5

-10 -5 5 10

-0.5

-1

-1.5

-2

Figure 5.14: Ten Term Taylor Series Approximation of cos x

the point x = 1. The first few derivatives of f are

f (x) = log x
1
f 0 (x) =
x
1
f 00 (x) = − 2
x
2
f 000 (x) = 3
x
3
f (4) (x) = − 4
61 x
The derivatives evaluated at x = 1 are
f (0) = 0, f (n) (0) = (−1)n−1 (n − 1)!, for n ≥ 1.
By Taylor’s theorem of the mean we have,
(x − 1)2 (x − 1)3 (x − 1)4 (x − 1)n (x − 1)n+1 1
log x = (x − 1) − + − + · · · + (−1)n−1 + (−1)n .
2 3 4 n n + 1 ξ n+1
Below are plots of the 2, 4, 10 and 50 term approximations.

2 2 2 2
1 1 1 1
-1 0.5 1 1.5 2 2.5 3 -1 0.5 1 1.5 2 2.5 3 -1 0.5 1 1.5 2 2.5 3 -1 0.5 1 1.5 2 2.5 3
-2 -2 -2 -2
-3 -3 -3 -3
-4 -4 -4 -4
-5 -5 -5 -5
-6 -6 -6 -6

Figure 5.15: The 2, 4, 10 and 50 Term Approximations of log x

Note that the approximation gets better on the interval (0, 2) and worse outside this interval as the number
of terms increases. The Taylor series converges to log x only on this interval.

5.6.2 Application: Finite Difference Schemes

Example 5.6.4 Suppose you sample a function at the discrete points n∆x, n ∈ Z. In Figure 5.16 we sample the
function f (x) = sin x on the interval [−4, 4] with ∆x = 1/4 and plot the data points.
We wish to approximate the derivative of the function on the grid points using only the value of the function
on those discrete points. From the definition of the derivative, one is lead to the formula
f (x + ∆x) − f (x)
f 0 (x) ≈ . (5.2)
∆x

62
 1

0.5

-4 -2 2 4

-0.5

-1

Figure 5.16: Sampling of sin x

Taylor’s theorem states that

∆x2 00
f (x + ∆x) = f (x) + ∆xf 0 (x) + f (ξ).
2

Substituting this expression into our formula for approximating the derivative we obtain

2
f (x + ∆x) − f (x) f (x) + ∆xf 0 (x) + ∆x2 f 00 (ξ) − f (x) ∆x 00
= = f 0 (x) + f (ξ).
∆x ∆x 2

Thus we see that the error in our approximation of the first derivative is ∆x 2
f 00 (ξ). Since the error has a linear
factor of ∆x, we call this a first order accurate method. Equation 5.2 is called the forward difference scheme for
calculating the first derivative. Figure 5.17 shows a plot of the value of this scheme for the function f (x) = sin x
and ∆x = 1/4. The first derivative of the function f 0 (x) = cos x is shown for comparison.
Another scheme for approximating the first derivative is the centered difference scheme,

f (x + ∆x) − f (x − ∆x)
f 0 (x) ≈ .
2∆x

63
 1

0.5

-4 -2 2 4

-0.5

-1

Figure 5.17: The Forward Difference Scheme Approximation of the Derivative

Expanding the numerator using Taylor’s theorem,

f (x + ∆x) − f (x − ∆x)
2∆x
∆x2 00 ∆x3 000 ∆x2 00 ∆x3 000
f (x) + ∆xf 0 (x) + 2
f (x) + 6
f (ξ) − f (x) + ∆xf 0 (x) − 2
f (x) + 6
f (η)
=
2∆x
∆x2 000
= f 0 (x) + (f (ξ) + f 000 (η)).
12
The error in the approximation is quadratic in ∆x. Therefore this is a second order accurate scheme. Below is a
plot of the derivative of the function and the value of this scheme for the function f (x) = sin x and ∆x = 1/4.
Notice how the centered difference scheme gives a better approximation of the derivative than the forward
difference scheme.

5.7 L’Hospital’s Rule

Some singularities are easy to diagnose. Consider the function cosx x at the point x = 0. The function evaluates
to 10 and is thus discontinuous at that point. Since the numerator and denominator are continuous functions and

64
 1

0.5

-4 -2 2 4

-0.5

-1

Figure 5.18: Centered Difference Scheme Approximation of the Derivative

the denominator vanishes while the numerator does not, the left and right limits as x → 0 do not exist. Thus
the function has an infinite discontinuity at the point x = 0. More generally, a function which is composed of
continuous functions and evaluates to a0 at a point where a 6= 0 must have an infinite discontinuity there.

Other singularities require more analysis to diagnose. Consider the functions sinx x , sin
|x|
x sin x
and 1−cos x
at the point
0
x = 0. All three functions evaluate to 0 at that point, but have different kinds of singularities. The first has
a removable discontinuity, the second has a finite discontinuity and the third has an infinite discontinuity. See
Figure 5.19.

An expression that evaluates to 00 , ∞∞

, 0 · ∞, ∞ − ∞, 1∞ , 00 or ∞0 is called an indeterminate. A function
f (x) which is indeterminate at the point x = ξ is singular at that point. The singularity may be a removable
discontinuity, a finite discontinuity or an infinite discontinuity depending on the behavior of the function around
that point. If limx→ξ f (x) exists, then the function has a removable discontinuity. If the limit does not exist, but
the left and right limits do exist, then the function has a finite discontinuity. If either the left or right limit does
not exist then the function has an infinite discontinuity.

65
 sin x sin x sin x
Figure 5.19: The functions x
, |x| and 1−cos x
.

L’Hospital’s Rule. Let f (x) and g(x) be differentiable and f (ξ) = g(ξ) = 0. Further, let g(x) be nonzero in a
deleted neighborhood of x = ξ, (g(x) 6= 0 for x ∈ 0 < |x − ξ| < δ). Then

f (x) f 0 (x)
lim = lim 0 .
x→ξ g(x) x→ξ g (x)

To prove this, we note that f (ξ) = g(ξ) = 0 and apply the generalized theorem of the mean. Note that

f (x) f (x) − f (ξ) f 0 (η)

= = 0
g(x) g(x) − g(ξ) g (η)

for some η between ξ and x. Thus

f (x) f 0 (η) f 0 (x)

lim = lim 0 = lim 0
x→ξ g(x) η→ξ g (η) x→ξ g (x)

provided that the limits exist.

L’Hospital’s Rule is also applicable when both functions tend to infinity instead of zero or when the limit
point, ξ, is at infinity. It is also valid for one-sided limits.
L’Hospital’s rule is directly applicable to the indeterminate forms 00 and ∞
∞
.

66
 sin x sin x sin x
Example 5.7.1 Consider the three functions x
, |x| and 1−cos x
at the point x = 0.

sin x cos x
lim = lim =1
x→0 x x→0 1
sin x
Thus x
has a removable discontinuity at x = 0.

sin x sin x
lim+ = lim+ =1
x→0 |x| x→0 x

sin x sin x
lim− = lim− = −1
x→0 |x| x→0 −x
sin x
Thus |x|
has a finite discontinuity at x = 0.

sin x cos x 1
lim = lim = =∞
x→0 1 − cos x x→0 sin x 0
sin x
Thus 1−cos x
has an infinite discontinuity at x = 0.

Example 5.7.2 Let a and d be nonzero.

ax2 + bx + c 2ax + b
lim = lim
x→∞ dx2 + ex + f x→∞ 2dx + e
2a
= lim
x→∞ 2d
a
=
d

67
Example 5.7.3 Consider
cos x − 1
lim .
x→0 x sin x

This limit is an indeterminate of the form 00 . Applying L’Hospital’s rule we see that limit is equal to

− sin x
lim .
x→0 x cos x + sin x

This limit is again an indeterminate of the form 00 . We apply L’Hospital’s rule again.

− cos x 1
lim =−
x→0 −x sin x + 2 cos x 2

Thus the value of the original limit is − 21 . We could also obtain this result by expanding the functions in Taylor
series.
 
x2 x4
cos x − 1 1 − 2
+ 24
− · · · −1
lim = lim 3 5
x→0 x x − x + x − · · ·


x→0 x sin x
6 120
2 x4
− x2 + 24
− ···
= lim x 4 x 6
2
x→0 x − + 120 − · · ·
6
2
− 12 + x24 − · · ·
= lim 2 4
x→0 1 − x + x − · · ·
6 120
1
=−
2
We can apply L’Hospital’s Rule to the indeterminate forms 0 · ∞ and ∞ − ∞ by rewriting the expression in
a different form, (perhaps putting the expression over a common denominator). If at first you don’t succeed, try,
try again. You may have to apply L’Hospital’s rule several times to evaluate a limit.

68
Example 5.7.4
 
1 x cos x − sin x
lim cot x − = lim
x→0 x x→0 x sin x
cos x − x sin x − cos x
= lim
x→0 sin x + x cos x
−x sin x
= lim
x→0 sin x + x cos x
−x cos x − sin x
= lim
x→0 cos x + cos x − x sin x

=0
You can apply L’Hospital’s rule to the indeterminate forms 1∞ , 00 or ∞0 by taking the logarithm of the
expression.

Example 5.7.5 Consider the limit,

lim xx ,
x→0

which gives us the indeterminate form 00 . The logarithm of the expression is

log(xx ) = x log x.
As x → 0 we now have the indeterminate form 0 · ∞. By rewriting the expression, we can apply L’Hospital’s rule.
log x 1/x
lim = lim
x→0 1/x x→0 −1/x2

= lim (−x)
x→0
=0
Thus the original limit is
lim xx = e0 = 1.
x→0

69
5.8 Exercises
Limits and Continuity
Exercise 5.1
Does
 
1
lim sin
x→0 x

exist?

Exercise 5.2
Is the function sin(1/x) continuous in the open interval (0, 1)? Is there a value of a such that the function defined
by
(
sin(1/x) for x 6= 0,
f (x) =
a for x = 0

is continuous on the closed interval [0, 1]?

Exercise 5.3
Is the function sin(1/x) uniformly continuous in the open interval (0, 1)?

Exercise 5.4 √ 1
Are the functions x and x
uniformly continuous on the interval (0, 1)?

Exercise 5.5
Prove that a function which is continuous on a closed interval is uniformly continuous on that interval.

70
Definition of Differentiation
Exercise 5.6 (mathematica/calculus/differential/definition.nb)
Use the definition of differentiation to prove the following identities where f (x) and g(x) are differentiable functions
and n is a positive integer.
d
a. dx
(xn ) = nxn−1 , (I suggest that you use Newton’s binomial formula.)
dg
b. d
dx
(f (x)g(x)) = f dx + g df
dx
d
c. dx
(sin x) = cos x. (You’ll need to use some trig identities.)
d. d
dx
(f (g(x))) = f 0 (g(x))g 0 (x)

Rules of Differentiation
Exercise 5.7 (mathematica/calculus/differential/rules.nb)
Find the first derivatives of the following:
a. x sin(cos x)
b. f (cos(g(x)))
1
c. f (log x)
x
d. xx
e. |x| sin |x|

Exercise 5.8 (mathematica/calculus/differential/rules.nb)

Using
d d 1
sin x = cos x and tan x =
dx dx cos2 x
find the derivatives of arcsin x and arctan x.

71
Implicit Differentiation
Exercise 5.9 (mathematica/calculus/differential/implicit.nb)
Find y 0 (x), given that x2 + y 2 = 1. What is y 0 (1/2)?

Exercise 5.10 (mathematica/calculus/differential/implicit.nb)

Find y 0 (x) and y 00 (x), given that x2 − xy + y 2 = 3.

Maxima and Minima

Exercise 5.11 (mathematica/calculus/differential/maxima.nb)
Identify any maxima and minima of the following functions.
a. f (x) = x(12 − 2x)2 .
b. f (x) = (x − 2)2/3 .

Exercise 5.12 (mathematica/calculus/differential/maxima.nb)

A cylindrical container with a circular base and an open top is to hold 64 cm3 . Find its dimensions so that the
surface area of the cup is a minimum.

Mean Value Theorems

Exercise 5.13
Prove the generalized theorem of the mean. If f (x) and g(x) are continuous in [a, b] and differentiable in (a, b),
then there exists a point x = ξ such that
f 0 (ξ) f (b) − f (a)
0
= .
g (ξ) g(b) − g(a)
Assume that g(a) 6= g(b) so that the denominator does not vanish and that f 0 (x) and g 0 (x) are not simultaneously
zero which would produce an indeterminate form.

72
Exercise 5.14 (mathematica/calculus/differential/taylor.nb)
1
Find a polynomial approximation of sin x on the interval [−1, 1] that has a maximum error of 1000 . Don’t use any
more terms that you need to. Prove the error bound. Use your polynomial to approximate sin 1.

Exercise 5.15 (mathematica/calculus/differential/taylor.nb)

You use the formula f (x+∆x)−2f (x)+f (x−∆x)
∆x2
to approximate f 00 (x). What is the error in this approximation?

Exercise 5.16
The formulas f (x+∆x)−f
∆x
(x)
and f (x+∆x)−f
2∆x
(x−∆x)
are first and second order accurate schemes for approximating
0
the first derivative f (x). Find a couple other schemes that have successively higher orders of accuracy. Would
these higher order schemes actually give a better approximation of f 0 (x)? Remember that ∆x is small, but not
infinitesimal.

L’Hospital’s Rule
Exercise 5.17 (mathematica/calculus/differential/lhospitals.nb)
Evaluate the following limits.
x−sin x
a. limx→0 x3
1


b. limx→0 csc x − x
1 x


c. limx→+∞ 1 + x

d. limx→0 csc2 x − x12 . (First evaluate using L’Hospital’s rule then using a Taylor series expansion. You will

find that the latter method is more convenient.)

Exercise 5.18 (mathematica/calculus/differential/lhospitals.nb)

Evaluate the following limits,
 a bx
lim xa/x , lim 1 + ,
x→∞ x→∞ x

73
where a and b are constants.

74
5.9 Hints
Limits and Continuity
Hint 5.1
Apply the , δ definition of a limit.

Hint 5.2
The composition of continuous functions is continuous. Apply the definition of continuity and look at the point
x = 0.

Hint 5.3
1 1
Note that for x1 = (n−1/2)π
and x2 = (n+1/2)π
where n ∈ Z we have | sin(1/x1 ) − sin(1/x2 )| = 2.

Hint 5.4 √ √
Note that the function x + δ − x is a decreasing function of x and an increasing function of δ for positive x
and δ. Bound this function for fixed δ.
Consider any positive δ and . For what values of x is

1 1
− > .
x x+δ

Hint 5.5
Let the function f (x) be continuous on a closed interval. Consider the function

e(x, δ) = sup |f (ξ) − f (x)|.

|ξ−x|<δ

Bound e(x, δ) with a function of δ alone.

75
Definition of Differentiation
Hint 5.6
a. Newton’s binomial formula is
n  
n
X n n−k k n(n − 1) n−2 2
(a + b) = a b = an + an−1 b + a b + · · · + nabn−1 + bn .
k=0
k 2

Recall that the binomial coefficient is

 
n n!
= .
k (n − k)!k!

b. Note that
 
d f (x + ∆x)g(x + ∆x) − f (x)g(x)
(f (x)g(x)) = lim
dx ∆x→0 ∆x
and
   
0 0 f (x + ∆x) − f (x) g(x + ∆x) − g(x)
g(x)f (x) + f (x)g (x) = g(x) lim + f (x) lim .
∆x→0 ∆x ∆x→0 ∆x
Fill in the blank.

c. First prove that

sin θ
lim = 1.
θ→0 θ
and
 
cos θ − 1
lim = 0.
θ→0 θ

76
 d. Let u = g(x). Consider a nonzero increment ∆x, which induces the increments ∆u and ∆f . By definition,

∆f = f (u + ∆u) − f (u), ∆u = g(x + ∆x) − g(x),

and ∆f, ∆u → 0 as ∆x → 0. If ∆u 6= 0 then we have

∆f df
= − → 0 as ∆u → 0.
∆u du
If ∆u = 0 for some values of ∆x then ∆f also vanishes and we define  = 0 for theses values. In either case,

df
∆y = ∆u + ∆u.
du
Continue from here.

Rules of Differentiation
Hint 5.7
a. Use the product rule and the chain rule.

b. Use the chain rule.

c. Use the quotient rule and the chain rule.

d. Use the identity ab = eb log a .

e. For x > 0, the expression is x sin x; for x < 0, the expression is (−x) sin(−x) = x sin x. Do both cases.

Hint 5.8
Use that x0 (y) = 1/y 0 (x) and the identities cos x = (1 − sin2 x)1/2 and cos(arctan x) = 1
(1+x2 )1/2
.

Implicit Differentiation

77
Hint 5.9
Differentiating the equation

x2 + [y(x)]2 = 1

yields

2x + 2y(x)y 0 (x) = 0.

Solve this equation for y 0 (x) and write y(x) in terms of x.

Hint 5.10
Differentiate the equation and solve for y 0 (x) in terms of x and y(x). Differentiate the expression for y 0 (x) to
obtain y 00 (x). You’ll use that

x2 − xy(x) + [y(x)]2 = 3

Maxima and Minima

Hint 5.11
a. Use the second derivative test.
b. The function is not differentiable at the point x = 2 so you can’t use a derivative test at that point.

Hint 5.12
Let r be the radius and h the height of the cylinder. The volume of the cup is πr2 h = 64. The radius and height
64 2 2 128
are related by h = πr 2 . The surface area of the cup is f (r) = πr + 2πrh = πr + r . Use the second derivative

test to find the minimum of f (r).

Mean Value Theorems

Hint 5.13
The proof is analogous to the proof of the theorem of the mean.

78
Hint 5.14
The first few terms in the Taylor series of sin(x) about x = 0 are

x3 x5 x7 x9
sin(x) = x − + − + + ··· .
6 120 5040 362880
When determining the error, use the fact that | cos x0 | ≤ 1 and |xn | ≤ 1 for x ∈ [−1, 1].

Hint 5.15
The terms in the approximation have the Taylor series,

∆x2 00 ∆x3 000 ∆x4 0000

f (x + ∆x) = f (x) + ∆xf 0 (x) + f (x) + f (x) + f (x1 ),
2 6 24
∆x2 00 ∆x3 000 ∆x4 0000
f (x − ∆x) = f (x) − ∆xf 0 (x) + f (x) − f (x) + f (x2 ),
2 6 24
where x ≤ x1 ≤ x + ∆x and x − ∆x ≤ x2 ≤ x.

Hint 5.16

L’Hospital’s Rule
Hint 5.17
a. Apply L’Hospital’s rule three times.

b. You can write the expression as

x − sin x
.
x sin x

c. Find the limit of the logarithm of the expression.

79
 d. It takes four successive applications of L’Hospital’s rule to evaluate the limit.
For the Taylor series expansion method,

1 x2 − sin2 x x2 − (x − x3 /6 + O(x5 ))2

csc2 x − = =
x2 x2 sin2 x x2 (x + O(x3 ))2

Hint 5.18
To evaluate the limits use the identity ab = eb log a and then apply L’Hospital’s rule.

80
5.10 Solutions
Limits and Continuity
Solution 5.1
Note that in any open neighborhood of zero, (−δ, δ), the function sin(1/x) takes on all values in the interval
[−1, 1]. Thus if we choose a positive  such that  < 1 then there is no value of η for which | sin(1/x) − η| <  for
all x ∈ (−, ). Thus the limit does not exist.

Solution 5.2
Since x1 is continuous in the interval (0, 1) and the function sin(x) is continuous everywhere, the composition
sin(1/x) is continuous in the interval (0, 1).
Since limx→0 sin(1/x) does not exist, there is no way of defining sin(1/x) at x = 0 to produce a function that
is continuous in [0, 1].

Solution 5.3
1 1
Note that for x1 = (n−1/2)π and x2 = (n+1/2)π where n ∈ Z we have | sin(1/x1 ) − sin(1/x2 )| = 2. Thus for any
0 <  < 2 there is no value of δ > 0 such that | sin(1/x1 ) − sin(1/x2 )| <  for all x1 , x2 ∈ (0, 1) and |x1 − x2 | < δ.
Thus sin(1/x) is not uniformly continuous in the open interval (0, 1).

Solution 5.4 √ √ √
First consider the function x. Note that the function x + δ − x is a decreasing function of x and an increasing
√ √ √
function of δ for positive x and δ. Thus for any fixed δ, the√maximum
√ value of x + δ − x is bounded by
√ δ.
2
Therefore on the interval (0, 1), a sufficient condition for | x − ξ| <  is |x − ξ| <  . The function x is
uniformly continuous on the interval (0, 1).
Consider any positive δ and . Note that

1 1
− >
x x+δ

81
for
r !
1 4δ
x< δ2 + −δ .
2 

Thus there is no value of δ such that

 
 1 1
 − <
x ξ 

1
for all |x − ξ| < δ. The function x
is not uniformly continuous on the interval (0, 1).

Solution 5.5
Let the function f (x) be continuous on a closed interval. Consider the function

e(x, δ) = sup |f (ξ) − f (x)|.

|ξ−x|<δ

Since f (x) is continuous, e(x, δ) is a continuous function of x on the same closed interval. Since continuous
functions on closed intervals are bounded, there is a continuous, increasing function (δ) satisfying,

e(x, δ) ≤ (δ),

for all x in the closed interval. Since (δ) is continuous and increasing, it has an inverse δ(). Now note that
|f (x) − f (ξ)| <  for all x and ξ in the closed interval satisfying |x − ξ| < δ(). Thus the function is uniformly
continuous in the closed interval.

Definition of Differentiation

82
Solution 5.6
a.

(x + ∆x)n − xn
 
d n
(x ) = lim
dx ∆x→0 ∆x
 n(n−1) n−2
 
xn + nxn−1 ∆x + 2
x ∆x2 + · · · + ∆x n
− xn
= lim  
∆x→0 ∆x
 
n−1 n(n − 1) n−2 n−1
= lim nx + x ∆x + · · · + ∆x
∆x→0 2
= nxn−1

d n
(x ) = nxn−1
dx

b.
 
d f (x + ∆x)g(x + ∆x) − f (x)g(x)
(f (x)g(x)) = lim
dx ∆x→0 ∆x
 
[f (x + ∆x)g(x + ∆x) − f (x)g(x + ∆x)] + [f (x)g(x + ∆x) − f (x)g(x)]
= lim
∆x→0 ∆x
   
f (x + ∆x) − f (x) g(x + ∆x) − g(x)
= lim [g(x + ∆x)] lim + f (x) lim
∆x→0 ∆x→0 ∆x ∆x→0 ∆x
0 0
= g(x)f (x) + f (x)g (x)

d
(f (x)g(x)) = f (x)g 0 (x) + f 0 (x)g(x)
dx

83
c. Consider a right triangle with hypotenuse of length 1 in the first quadrant of the plane. Label the vertices
A, B, C, in clockwise order, starting with the vertex at the origin. The angle of A is θ. The length of a
circular arc of radius cos θ that connects C to the hypotenuse is θ cos θ. The length of the side BC is sin θ.
The length of a circular arc of radius 1 that connects B to the x axis is θ. (See Figure 5.20.)
B

θ cos θ sin θ

θ
A
C

Figure 5.20:

Considering the length of these three curves gives us the inequality:

θ cos θ ≤ sin θ ≤ θ.
Dividing by θ,
sin θ
cos θ ≤ ≤ 1.
θ
Taking the limit as θ → 0 gives us
sin θ
lim = 1.
θ→0 θ

84
 One more little tidbit we’ll need to know is
   
cos θ − 1 cos θ − 1 cos θ + 1
lim = lim
θ→0 θ θ→0 θ cos θ + 1
2
 
cos θ − 1
= lim
θ→0 θ(cos θ + 1)

− sin2 θ
 
= lim
θ→0 θ(cos θ + 1)
   
− sin θ sin θ
= lim lim
θ→0 θ θ→0 (cos θ + 1)
 
0
= (−1)
2
= 0.
Now we’re ready to find the derivative of sin x.
 
d sin(x + ∆x) − sin x
(sin x) = lim
dx ∆x→0 ∆x
 
cos x sin ∆x + sin x cos ∆x − sin x
= lim
∆x→0 ∆x
   
sin ∆x cos ∆x − 1
= cos x lim + sin x lim
∆x→0 ∆x ∆x→0 ∆x
= cos x

d
(sin x) = cos x
dx

d. Let u = g(x). Consider a nonzero increment ∆x, which induces the increments ∆u and ∆f . By definition,
∆f = f (u + ∆u) − f (u), ∆u = g(x + ∆x) − g(x),

85
 and ∆f, ∆u → 0 as ∆x → 0. If ∆u 6= 0 then we have

∆f df
= − → 0 as ∆u → 0.
∆u du

If ∆u = 0 for some values of ∆x then ∆f also vanishes and we define  = 0 for theses values. In either case,

df
∆y = ∆u + ∆u.
du

We divide this equation by ∆x and take the limit as ∆x → 0.

df ∆f
= lim
dx ∆x→0  ∆x 
df ∆u ∆u
= lim +
∆x→0 du ∆x ∆x
     
df ∆f ∆u
= lim + lim  lim
du ∆x→0 ∆x ∆x→0 ∆x→0 ∆x
 
df du du
= + (0)
du dx dx
df du
=
du dx

Thus we see that

d
(f (g(x))) = f 0 (g(x))g 0 (x).
dx

Rules of Differentiation

86
Solution 5.7
a.
d d d
[x sin(cos x)] = [x] sin(cos x) + x [sin(cos x)]
dx dx dx
d
= sin(cos x) + x cos(cos x) [cos x]
dx
= sin(cos x) − x cos(cos x) sin x

d
[x sin(cos x)] = sin(cos x) − x cos(cos x) sin x
dx

b.
d d
[f (cos(g(x)))] = f 0 (cos(g(x))) [cos(g(x))]
dx dx
d
= −f 0 (cos(g(x))) sin(g(x)) [g(x)]
dx
= −f (cos(g(x))) sin(g(x))g 0 (x)
0

d
[f (cos(g(x)))] = −f 0 (cos(g(x))) sin(g(x))g 0 (x)
dx
c.
  d
d 1 dx
[f (log x)]
=−
dx f (log x) [f (log x)]2
f 0 (log x) dx
d
[log x]
=−
[f (log x)]2
f 0 (log x)
=−
x[f (log x)]2

87
 f 0 (log x)
 
d 1
=−
dx f (log x) x[f (log x)]2

d. First we write the expression in terms exponentials and logarithms,

x
xx = xexp(x log x) = exp(exp(x log x) log x).
Then we differentiate using the chain rule and the product rule.
d d
exp(exp(x log x) log x) = exp(exp(x log x) log x) (exp(x log x) log x)
dx  dx 
x x d 1
=x exp(x log x) (x log x) log x + exp(x log x)
dx x
 
x 1
= xx xx (log x + x ) log x + x−1 exp(x log x)
x
x
= xx xx (log x + 1) log x + x−1 xx


x
= xx +x x−1 + log x + log2 x

d xx x
x = xx +x x−1 + log x + log2 x


dx

e. For x > 0, the expression is x sin x; for x < 0, the expression is (−x) sin(−x) = x sin x. Thus we see that
|x| sin |x| = x sin x.
The first derivative of this is
sin x + x cos x.

d
(|x| sin |x|) = sin x + x cos x
dx

88
Solution 5.8
Let y(x) = sin x. Then y 0 (x) = cos x.

d 1
arcsin y = 0
dy y (x)
1
=
cos x
1
=
(1 − sin2 x)1/2
1
=
(1 − y 2 )1/2

d 1
arcsin x =
dx (1 − x2 )1/2

Let y(x) = tan x. Then y 0 (x) = 1/ cos2 x.

d 1
arctan y = 0
dy y (x)
= cos2 x
= cos2 (arctan y)
 
1
=
(1 + y 2 )1/2
1
=
1 + y2

d 1
arctan x =
dx 1 + x2

89
Implicit Differentiation
Solution 5.9
Differentiating the equation

x2 + [y(x)]2 = 1

yields

2x + 2y(x)y 0 (x) = 0.

We can solve this equation for y 0 (x).

x
y 0 (x) = −
y(x)
To find y 0 (1/2) we need to find y(x) in terms of x.
√
y(x) = ± 1 − x2

Thus y 0 (x) is
x
y 0 (x) = ± √ .
1 − x2
y 0 (1/2) can have the two values:
 
0 1 1
y = ±√ .
2 3

Solution 5.10
Differentiating the equation

x2 − xy(x) + [y(x)]2 = 3

90
yields

2x − y(x) − xy 0 (x) + 2y(x)y 0 (x) = 0.

Solving this equation for y 0 (x)

y(x) − 2x
y 0 (x) = .
2y(x) − x

Now we differentiate y 0 (x) to get y 00 (x).

(y 0 (x) − 2)(2y(x) − x) − (y(x) − 2x)(2y 0 (x) − 1)
y 00 (x) = ,
(2y(x) − x)2

xy 0 (x) − y(x)
y 00 (x) = 3 ,
(2y(x) − x)2

00
x y(x)−2x
2y(x)−x
− y(x)
y (x) = 3 ,
(2y(x) − x)2

x(y(x) − 2x) − y(x)(2y(x) − x)

y 00 (x) = 3 ,
(2y(x) − x)3

x2 − xy(x) + [y(x)]2
y 00 (x) = −6 ,
(2y(x) − x)3

−18
y 00 (x) = ,
(2y(x) − x)3

91
Maxima and Minima
Solution 5.11
a.

f 0 (x) = (12 − 2x)2 + 2x(12 − 2x)(−2)

= 4(x − 6)2 + 8x(x − 6)
= 12(x − 2)(x − 6)

There are critical points at x = 2 and x = 6.

f 00 (x) = 12(x − 2) + 12(x − 6) = 24(x − 4)

Since f 00 (2) = −48 < 0, x = 2 is a local maximum. Since f 00 (6) = 48 > 0, x = 6 is a local minimum.

b.
2
f 0 (x) = (x − 2)−1/3
3
The first derivative exists and is nonzero for x 6= 2. At x = 2, the derivative does not exist and thus x = 2
is a critical point. For x < 2, f 0 (x) < 0 and for x > 2, f 0 (x) > 0. x = 2 is a local minimum.

Solution 5.12
Let r be the radius and h the height of the cylinder. The volume of the cup is πr2 h = 64. The radius and height
64 2 2 128
are related by h = πr 2 . The surface area of the cup is f (r) = πr + 2πrh = πr + r . The first derivative of the

surface area is f 0 (r) = 2πr − 128

r2
. Finding the zeros of f 0 (r),
128
2πr − = 0,
r2

2πr3 − 128 = 0,

92
 4
r= √ .
3
π
The second derivative of the surface area is f 00 (r) = 2π + 256
r3
. Since f 00 ( √4
3 π ) = 6π, r =
4
√
3π is a local minimum of
f (r). Since this is the only critical point for r > 0, it must be a global minimum.
4 4
The cup has a radius of √ 3 π cm and a height of √ 3 π.

Mean Value Theorems

Solution 5.13
We define the function
f (b) − f (a)
h(x) = f (x) − f (a) − (g(x) − g(a)).
g(b) − g(a)
Note that h(x) is differentiable and that h(a) = h(b) = 0. Thus h(x) satisfies the conditions of Rolle’s theorem
and there exists a point ξ ∈ (a, b) such that
f (b) − f (a) 0
h0 (ξ) = f 0 (ξ) − g (ξ) = 0,
g(b) − g(a)

f 0 (ξ) f (b) − f (a)

0
= .
g (ξ) g(b) − g(a)

Solution 5.14
The first few terms in the Taylor series of sin(x) about x = 0 are
x3 x5 x7 x9
sin(x) = x − + − + + ··· .
6 120 5040 362880
The seventh derivative of sin x is − cos x. Thus we have that
x3 x5 cos x0 7
sin(x) = x − + − x,
6 120 5040

93
where 0 ≤ x0 ≤ x. Since we are considering x ∈ [−1, 1] and −1 ≤ cos(x0 ) ≤ 1, the approximation

x3 x5
sin x ≈ x − +
6 120
1
has a maximum error of 5040
≈ 0.000198. Using this polynomial to approximate sin(1),

13 15
1− + ≈ 0.841667.
6 120

To see that this has the required accuracy,

sin(1) ≈ 0.841471.

Solution 5.15
Expanding the terms in the approximation in Taylor series,

∆x2 00 ∆x3 000 ∆x4 0000

f (x + ∆x) = f (x) + ∆xf 0 (x) + f (x) + f (x) + f (x1 ),
2 6 24
∆x2 00 ∆x3 000 ∆x4 0000
f (x − ∆x) = f (x) − ∆xf 0 (x) + f (x) − f (x) + f (x2 ),
2 6 24
where x ≤ x1 ≤ x + ∆x and x − ∆x ≤ x2 ≤ x. Substituting the expansions into the formula,

f (x + ∆x) − 2f (x) + f (x − ∆x) 00 ∆x2 0000

= f (x) + [f (x1 ) + f 0000 (x2 )].
∆x2 24
Thus the error in the approximation is

∆x2 0000
[f (x1 ) + f 0000 (x2 )].
24

94
Solution 5.16

L’Hospital’s Rule

Solution 5.17
a.

   
x − sin x 1 − cos x
lim = lim
x→0 x3 x→0 3x2
 
sin x
= lim
x→0 6x
h cos x i
= lim
x→0 6
1
=
6

 
x − sin x 1
lim 3
=
x→0 x 6

95
b.

   
1 1 1
lim csc x − = lim −
x→0 x x→0 sin x x
 
x − sin x
= lim
x→0 x sin x
 
1 − cos x
= lim
x→0 x cos x + sin x
 
sin x
= lim
x→0 −x sin x + cos x + cos x
0
=
2
=0

 
1
lim csc x − =0
x→0 x

96
c.

  x    x 
1 1
log lim 1+ = lim log 1+
x→+∞ x x→+∞ x
  
1
= lim x log 1 +
x→+∞ x
" #
log 1 + x1


= lim
x→+∞ 1/x
"
−1
#
1 + x1 − x12
= lim
x→+∞ −1/x2
" −1 #
1
= lim 1+
x→+∞ x
=1

Thus we have

 x 
1
lim 1+ = e.
x→+∞ x

97
d. It takes four successive applications of L’Hospital’s rule to evaluate the limit.

x2 − sin2 x
 
2 1
lim csc x − 2 = lim 2 2
x→0 x x→0 x sin x

2x − 2 cos x sin x
= lim 2
x→0 2x cos x sin x + 2x sin2 x

2 − 2 cos2 x + 2 sin2 x
= lim 2
x→0 2x cos2 x + 8x cos x sin x + 2 sin2 x − 2x2 sin2 x
8 cos x sin x
= lim
x→0 12x cos2 x + 12 cos x sin x − 8x2 cos x sin x − 12x sin2 x

8 cos2 x − 8 sin2 x
= lim
x→0 24 cos2 x − 8x2 cos2 x − 64x cos x sin x − 24 sin2 x + 8x2 sin2 x
1
=
3

It is easier to use a Taylor series expansion.

x2 − sin2 x
 
2 1
lim csc x − 2 = lim 2 2
x→0 x x→0 x sin x

x2 − (x − x3 /6 + O(x5 ))2
= lim
x→0 x2 (x + O(x3 ))2
x2 − (x2 − x4 /3 + O(x6 ))
= lim
x→0 x4 + O(x6 )
 
1 2
= lim + O(x )
x→0 3
1
=
3

98
Solution 5.18
To evaluate the first limit, we use the identity ab = eb log a and then apply L’Hospital’s rule.
a log x
lim xa/x = lim e x
x→∞ x→∞
 
a log x
= exp lim
x→∞ x
 
a/x
= exp lim
x→∞ 1
0
= e

lim xa/x = 1
x→∞

We use the same method to evaluate the second limit.

 a bx   a 
lim 1 + = lim exp bx log 1 +
x→∞ x x→∞ x 
  a
= exp lim bx log 1 +
x→∞ x 
log(1 + a/x)
= exp lim b
x→∞ 1/x
 2

−a/x
1+a/x
= exp  lim b 
x→∞ −1/x2
 
a
= exp lim b
x→∞ 1 + a/x

 a bx
lim 1 + = eab
x→∞ x

99
Chapter 6

Integral Calculus

6.1 The Indefinite Integral

The opposite of a derivative is the anti-derivative or the indefinite integral. The indefinite integral of a function
f (x) is denoted,
Z
f (x) dx.

It is defined by the property that

Z
d
f (x) dx = f (x).
dx

While a function f (x) has a unique derivative if it is differentiable, it has an infinite number of indefinite integrals,
each of which differ by an additive constant.

Zero Slope Implies a Constant Function. If the value of a function’s derivative is identically zero, df dx
= 0,
then the function is a constant, f (x) = c. To prove this, we assume that there exists a non-constant differentiable

100
function whose derivative is zero and obtain a contradiction. Let f (x) be such a function. Since f (x) is non-
constant, there exist points a and b such that f (a) 6= f (b). By the Mean Value Theorem of differential calculus,
there exists a point ξ ∈ (a, b) such that

f (b) − f (a)
f 0 (ξ) = 6= 0,
b−a
which contradicts that the derivative is everywhere zero.

Indefinite Integrals Differ by an Additive Constant. Suppose that F (x) and G(x) are indefinite integrals
of f (x). Then we have

d
(F (x) − G(x)) = F 0 (x) − G0 (x) = f (x) − f (x) = 0.
dx
Thus Rwe see that F (x) − G(x) = c and the two indefinite integrals must differ by a constant. For example, we
have sin x dx = − cos x + c. While every function that can be expressed in terms of elementary functions, (the
exponent, logarithm, trigonometric functions, etc.), has a derivative Rthat can be written explicitly in terms of
elementary functions, the same is not true of integrals. For example, sin(sin x) dx cannot be written explicitly
in terms of elementary functions.

Properties. Since the derivative is linear, so is the indefinite integral. That is,
Z Z Z
(af (x) + bg(x)) dx = a f (x) dx + b g(x) dx.

d
For each derivative identity there is a corresponding integral identity. Consider the power law identity, dx
(f (x))a =
a(f (x))a−1 f 0 (x). The corresponding integral identity is

(f (x))a+1
Z
(f (x))a f 0 (x) dx = + c, a 6= −1,
a+1

101
 d f 0 (x)
where we require that a 6= −1 to avoid division by zero. From the derivative of a logarithm, dx
ln(f (x)) = f (x)
,
we obtain,
Z 0
f (x)
dx = ln |f (x)| + c.
f (x)
d 1 1
Note the absolute value signs. This is because dx
ln |x| = x
for x 6= 0. In Figure 6.1 is a plot of ln |x| and x
to
reinforce this.

Figure 6.1: Plot of ln |x| and 1/x.

Example 6.1.1 Consider

Z
x
I= dx.
(x2 + 1)2
We evaluate the integral by choosing u = x2 + 1, du = 2x dx.
Z
1 2x
I= dx
2 (x + 1)2
2
Z
1 du
=
2 u2
1 −1
=
2 u
1
=− 2
.
2(x + 1)

102
Example 6.1.2 Consider
Z Z
sin x
I= tan x dx = dx.
cos x
By choosing f (x) = cos x, f 0 (x) = − sin x, we see that the integral is
− sin x
Z
I=− dx = − ln | cos x| + c.
cos x

Change of Variable. The differential of a function g(x) is dg = g 0 (x) dx. Thus one might suspect that for
ξ = g(x),
Z Z
f (ξ) dξ = f (g(x))g 0 (x) dx, (6.1)

since dξ = dg = g 0 (x) dx. This turns out to be true. To prove it we will appeal to the the chain rule for
differentiation. Let ξ be a function of x. The chain rule is
d
f (ξ) = f 0 (ξ)ξ 0 (x),
dx

d df dξ
f (ξ) = .
dx dξ dx
We can also write this as
df dx df
= ,
dξ dξ dx
or in operator notation,
d dx d
= .
dξ dξ dx

103
Now we’re ready to start. The derivative of the left side of Equation 6.1 is
Z
d
f (ξ) dξ = f (ξ).
dξ
Next we differentiate the right side,
Z Z
d 0 dx d
f (g(x))g (x) dx = f (g(x))g 0 (x) dx
dξ dξ dx
dx
= f (g(x))g 0 (x)
dξ
dx dg
= f (g(x))
dg dx
= f (g(x))
= f (ξ)
to see that it is in fact an identity for ξ = g(x).

Example 6.1.3 Consider

Z
x sin(x2 ) dx.

We choose ξ = x2 , dξ = 2xdx to evaluate the integral.

Z Z
2 1
x sin(x ) dx = sin(x2 )2x dx
2
Z
1
= sin ξ dξ
2
1
= (− cos ξ) + c
2
1
= − cos(x2 ) + c
2

104
Integration by Parts. The product rule for differentiation gives us an identity called integration by parts. We
start with the product rule and then integrate both sides of the equation.
d
(u(x)v(x)) = u0 (x)v(x) + u(x)v 0 (x)
Z dx
(u0 (x)v(x) + u(x)v 0 (x)) dx = u(x)v(x) + c
Z Z
u (x)v(x) dx + u(x)v 0 (x)) dx = u(x)v(x)
0

Z Z
u(x)v (x)) dx = u(x)v(x) − v(x)u0 (x) dx
0

The theorem is most often written in the form

Z Z
u dv = uv − v du.

So what is the usefulness of this? Well, it may happen for some integrals and a good choice of u and v that the
integral on the right is easier to evaluate than the integral on the left.
R
Example 6.1.4 Consider x ex dx. If we choose u = x, dv = ex dx then integration by parts yields
Z Z
x x
x e dx = x e − ex dx = (x − 1) ex .

Now notice what happens when we choose u = ex , dv = x dx.

Z Z
x 1 2 x 1 2 x
x e dx = x e − x e dx
2 2
The integral gets harder instead of easier.

When applying integration by parts, one must choose u and dv wisely. As general rules of thumb:

105
 • Pick u so that u0 is simpler than u.
• Pick dv so that v is not more complicated, (hopefully simpler), than dv.
Also note that you may have to apply integration by parts several times to evaluate some integrals.

6.2 The Definite Integral

6.2.1 Definition
The area bounded by the x axis, the vertical lines x = a and x = b and the function f (x) is denoted with a
definite integral,
Z b
f (x) dx.
a
The area is signed, that is, if f (x) is negative, then the area is negative. We measure the area with a divide-and-
conquer strategy. First partition the interval (a, b) with a = x0 < x1 < · · · < xn−1 < xn = b. Note that the area
under the curve on the subinterval is approximately the area of a rectangle of base ∆xi = xi+1 − xi and height
f (ξi ), where ξi ∈ [xi , xi+1 ]. If we add up the areas of the rectangles, we get an approximation of the area under
the curve. See Figure 6.2

Z b n−1
X
f (x) dx ≈ f (ξi )∆xi
a i=0

As the ∆xi ’s get smaller, we expect the approximation of the area to get better. Let ∆x = max0≤i≤n−1 ∆xi . We
define the definite integral as the sum of the areas of the rectangles in the limit that ∆x → 0.
Z b n−1
X
f (x) dx = lim f (ξi )∆xi
a ∆x→0
i=0

The integral is defined when the limit exists. This is known as the Riemann integral of f (x). f (x) is called the
integrand.

106
 f(ξ1 )

a x1 x2 x3 x n-2 x n-1 b
∆ xi

Figure 6.2: Divide-and-Conquer Strategy for Approximating a Definite Integral.

6.2.2 Properties
Linearity and the Basics. Because summation is a linear operator, that is

n−1
X n−1
X n−1
X
(cfi + dgi ) = c fi + d gi ,
i=0 i=0 i=0

definite integrals are linear,

Z b Z b Z b
(cf (x) + dg(x)) dx = c f (x) dx + d g(x) dx.
a a a

One can also divide the range of integration.

Z b Z c Z b
f (x) dx = f (x) dx + f (x) dx
a a c

107
We assume that each of the above integrals exist. If a ≤ b, and we integrate from b to a, then each of the ∆xi
will be negative. From this observation, it is clear that
Z b Z a
f (x) dx = − f (x) dx.
a b

If we integrate any function from a point a to that same point a, then all the ∆xi are zero and
Z a
f (x) dx = 0.
a

Bounding the Integral. Recall that if fi ≤ gi , then

n−1
X n−1
X
fi ≤ gi .
i=0 i=0

Let m = minx∈[a,b] f (x) and M = maxx∈[a,b] f (x). Then

n−1
X n−1
X n−1
X
(b − a)m = m∆xi ≤ f (ξi )∆xi ≤ M ∆xi = (b − a)M
i=0 i=0 i=0

implies that
Z b
(b − a)m ≤ f (x) dx ≤ (b − a)M.
a

Since
 n−1  n−1
X  X
fi  ≤ |fi |,
 

 
i=0 i=0

we have
Z b  Z b
 

 f (x) dx ≤ |f (x)| dx.
a a

108
Mean Value Theorem of Integral Calculus. Let f (x) be continuous. We know from above that

Z b
(b − a)m ≤ f (x) dx ≤ (b − a)M.
a

Therefore there exists a constant c ∈ [m, M ] satisfying

Z b
f (x) dx = (b − a)c.
a

Since f (x) is continuous, there is a point ξ ∈ [a, b] such that f (ξ) = c. Thus we see that

Z b
f (x) dx = (b − a)f (ξ),
a

for some ξ ∈ [a, b].

6.3 The Fundamental Theorem of Integral Calculus

Definite Integrals with Variable Limits of Integration. Consider a to be a constant and x variable, then
the function F (x) defined by

Z x
F (x) = f (t) dt (6.2)
a

109
is an anti-derivative of f (x), that is F 0 (x) = f (x). To show this we apply the definition of differentiation and the
integral mean value theorem.
F (x + ∆x) − F (x)
F 0 (x) = lim
∆x→0 ∆x
R x+∆x Rx
f (t) dt − f (t) dt
= lim a a
∆x→0 ∆x
R x+∆x
f (t) dt
= lim x
∆x→0 ∆x
f (ξ)∆x
= lim , ξ ∈ [x, x + ∆x]
∆x→0 ∆x
= f (x)

The Fundamental Theorem of Integral Calculus. Let F (x) be any anti-derivative of f (x). Noting that
all anti-derivatives of f (x) differ by a constant and replacing x by b in Equation 6.2, we see that there exists a
constant c such that
Z b
f (x) dx = F (b) + c.
a

Now to find the constant. By plugging in b = a,

Z a
f (x) dx = F (a) + c = 0,
a

we see that c = −F (a). This gives us a result known as the Fundamental Theorem of Integral Calculus.
Z b
f (x) dx = F (b) − F (a).
a

We introduce the notation

[F (x)]ba ≡ F (b) − F (a).

110
Example 6.3.1
Z π
sin x dx = [− cos x]π0 = − cos(π) + cos(0) = 2
0

6.4 Techniques of Integration

6.4.1 Partial Fractions
A proper rational function
p(x) p(x)
=
q(x) (x − a)n r(x)
Can be written in the form
 
p(x) a0 a1 an−1
= + + ··· + + (· · · )
(x − α)n r(x) (x − α) n (x − α)n−1 x−α
where the ak ’s are constants and the last ellipses represents the partial fractions expansion of the roots of r(x).
The coefficients are
1 dk p(x) 
 
ak = .
k! dxk r(x) x=α

Example 6.4.1 Consider the partial fraction expansion of

1 + x + x2
.
(x − 1)3
The expansion has the form
a0 a1 a2
3
+ 2
+ .
(x − 1) (x − 1) x−1

111
The coefficients are
1
a0 = (1 + x + x2 )|x=1 = 3,
0!
1 d
a1 = (1 + x + x2 )|x=1 = (1 + 2x)|x=1 = 3,
1! dx
1 d2 1
a2 = 2
(1 + x + x2 )|x=1 = (2)|x=1 = 1.
2! dx 2
Thus we have

1 + x + x2 3 3 1
3
= 3
+ 2
+ .
(x − 1) (x − 1) (x − 1) x−1

Example 6.4.2 Suppose we want to evaluate

1 + x + x2
Z
dx.
(x − 1)3

If we expand the integrand in a partial fraction expansion, then the integral becomes easy.

1 + x + x2
Z Z  
3 3 1
dx. = + + dx
(x − 1)3 (x − 1)3 (x − 1)2 x − 1
3 3
=− − + ln(x − 1)
2(x − 1)2 (x − 1)

Example 6.4.3 Consider the partial fraction expansion of

1 + x + x2
.
x2 (x − 1)2

112
The expansion has the form
a0 a1 b0 b1
2
+ + 2
+ .
x x (x − 1) x−1
The coefficients are
1 1 + x + x2 
 
a0 = = 1,
0! (x − 1)2 
x=0
1 d 1 + x + x2  2(1 + x + x2 ) 
   
1 + 2x
a1 = = − = 3,
1! dx (x − 1)2 
x=0 (x − 1)2 (x − 1)3 
x=0
1 1 + x + x2 
 
b0 = = 3,
0! x2 
x=1
1 d 1 + x + x2  1 + 2x 2(1 + x + x2 ) 
   
b1 = = − = −3,
1! dx x2 
x=1 x2 x3 
x=1

Thus we have
1 + x + x2 1 3 3 3
2 2
= 2+ + 2
− .
x (x − 1) x x (x − 1) x−1
If the rational function has real coefficients and the denominator has complex roots, then you can reduce the
work in finding the partial fraction expansion with the following trick: Let α and α be complex conjugate pairs
of roots of the denominator.
 
p(x) a0 a1 an−1
= + + ··· +
(x − α)n (x − α)n r(x) (x − α)n (x − α)n−1 x−α
 
a0 a1 an−1
+ + + ··· + + (· · · )
(x − α)n (x − α)n−1 x−α
Thus we don’t have to calculate the coefficients for the root at α. We just take the complex conjugate of the
coefficients for α.

113
Example 6.4.4 Consider the partial fraction expansion of
1+x
.
x2 + 1
The expansion has the form
a0 a0
+
x−i x+i
The coefficients are
 
1 1 + x  1
a0 = = (1 − i),
0! x + i x=i 2

1 1
a0 = (1 − i) = (1 + i)
2 2
Thus we have
1+x 1−i 1+i
= + .
x2 + 1 2(x − i) 2(x + i)

6.5 Improper Integrals

Rb
If the range of integration is infinite or f (x) is discontinuous at some points then a
f (x) dx is called an improper
integral.

Discontinuous Functions. If f (x) is continuous on the interval a ≤ x ≤ b except at the point x = c where
a < c < b then
Z b Z c−δ Z b
f (x) dx = lim+ f (x) dx + lim+ f (x) dx
a δ→0 a →0 c+

provided that both limits exist.

114
Example 6.5.1 Consider the integral of ln x on the interval [0, 1]. Since the logarithm has a singularity at x = 0,
this is an improper integral. We write the integral in terms of a limit and evaluate the limit with L’Hospital’s
rule.
Z 1 Z 1
ln x dx = lim ln x dx
0 δ→0 δ
= lim[x ln x − x]1δ
δ→0
= 1 ln(1) − 1 − lim(δ ln δ − δ)
δ→0
= −1 − lim(δ ln δ)
δ→0
 
ln δ
= −1 − lim
δ→0 1/δ
 
1/δ
= −1 − lim
δ→0 −1/δ 2
= −1

Example 6.5.2 Consider the integral of xa on the range [0, 1]. If a < 0 then there is a singularity at x = 0. First
assume that a 6= −1.
Z 1  a+1 1
a x
x dx = lim+
0 δ→0 a+1 δ
1 δ a+1
= − lim+
a + 1 δ→0 a + 1
This limit exists only for a > −1. Now consider the case that a = −1.
Z 1
x−1 dx = lim+ [ln x]1δ
0 δ→0

= ln(0) − lim+ ln δ
δ→0

115
This limit does not exist. We obtain the result,
Z 1
1
xa dx = , for a > −1.
0 a+1

Infinite Limits of Integration. If the range of integration is infinite, say [a, ∞) then we define the integral as
Z ∞ Z α
f (x) dx = lim f (x) dx,
a α→∞ a

provided that the limit exists. If the range of integration is (−∞, ∞) then
Z ∞ Z a Z β
f (x) dx = lim f (x) dx + lim f (x) dx.
−∞ α→−∞ α β→+∞ a

Example 6.5.3
∞ ∞  
d −1
Z Z
ln x
dx = ln x dx
1 x2 1 dx x
 ∞ Z ∞
−1 −1 1
= ln x − dx
x 1 1 x x
   ∞
ln x 1
= lim − −
x→+∞ x x 1
 
1/x 1
= lim − − lim + 1
x→+∞ 1 x→∞ x

=1

116
Example 6.5.4 Consider the integral of xa on [1, ∞). First assume that a 6= −1.
∞ β
xa+1
Z 
a
x dx = lim
1 β→+∞ a + 1
1
β a+1 1
= lim −
β→+∞ a + 1 a+1
The limit exists for β < −1. Now consider the case a = −1.
Z ∞
x−1 dx = lim [ln x]β1
1 β→+∞
1
= lim ln β −
β→+∞ a+1
This limit does not exist. Thus we have
Z ∞
1
xa dx = − , for a < −1.
1 a+1

117
6.6 Exercises
Fundamental Integration Formulas
ExerciseR6.1 (mathematica/calculus/integral/fundamental.nb)
Evaluate (2x + 3)10 dx.

ExerciseR6.2 (mathematica/calculus/integral/fundamental.nb)
2
Evaluate (lnxx) dx.

ExerciseR6.3√(mathematica/calculus/integral/fundamental.nb)
Evaluate x x2 + 3 dx.

ExerciseR6.4 (mathematica/calculus/integral/fundamental.nb)
Evaluate cos x
sin x
dx.

ExerciseR6.5 (mathematica/calculus/integral/fundamental.nb)
2
Evaluate x3x−5 dx.

Integration by Parts
ExerciseR6.6 (mathematica/calculus/integral/parts.nb)
Evaluate x sin x dx.

ExerciseR6.7 (mathematica/calculus/integral/parts.nb)
Evaluate x3 e2x dx.

118
Partial Fractions
ExerciseR6.8 (mathematica/calculus/integral/partial.nb)
Evaluate x21−4 dx.

ExerciseR6.9 (mathematica/calculus/integral/partial.nb)
x+1
Evaluate x3 +x 2 −6x dx.

Definite Integrals
Exercise 6.10 (mathematica/calculus/integral/definite.nb)
Use the result
Z b N
X −1
f (x) dx = lim f (xn )∆x
a N →∞
n=0
b−a
where ∆x = N
and xn = a + n∆x, to show that
Z 1
1
x dx = .
0 2

Exercise 6.11 (mathematica/calculus/integral/definite.nb) Rπ

Evaluate the following integral using integration by parts and the Pythagorean identity. 0 sin2 x dx

Exercise 6.12 (mathematica/calculus/integral/definite.nb)

Prove that
Z f (x)
d
h(ξ) dξ = h(f (x))f 0 (x) − h(g(x))g 0 (x).
dx g(x)
(Don’t use the limit definition of differentiation, use the Fundamental Theorem of Integral Calculus.)

119
Improper Integrals
ExerciseR6.13 (mathematica/calculus/integral/improper.nb)
4 1
Evaluate 0 (x−1)2 dx.

ExerciseR6.14 (mathematica/calculus/integral/improper.nb)
1
Evaluate 0 √1x dx.

ExerciseR6.15 (mathematica/calculus/integral/improper.nb)
∞
Evaluate 0 x21+4 dx.

Taylor Series
Exercise 6.16 (mathematica/calculus/integral/taylor.nb)
a. Show that
Z x
f (x) = f (0) + f 0 (x − ξ) dξ.
0

b. From the above identity show that

Z x
0
f (x) = f (0) + xf (0) + ξf 00 (x − ξ) dξ.
0

c. Using induction, show that

Z x
1 1 1 n (n+1)
f (x) = f (0) + xf (0) + x2 f 00 (0) + · · · + xn f (n) (0) +
0
ξ f (x − ξ) dξ.
2 n! 0 n!

120
6.7 Hints
Fundamental Integration Formulas
Hint 6.1
Make the change of variables u = 2x + 3.

Hint 6.2
Make the change of variables u = ln x.

Hint 6.3
Make the change of variables u = x2 + 3.

Hint 6.4
Make the change of variables u = sin x.

Hint 6.5
Make the change of variables u = x3 − 5.

Integration by Parts
Hint 6.6
Let u = x, and dv = sin x dx.

Hint 6.7
Perform integration by parts three successive times. For the first one let u = x3 and dv = e2x dx.

Partial Fractions

121
Hint 6.8
Expanding the integrand in partial fractions,

1 1 a b
= = +
x2 −4 (x − 2)(x + 2) (x − 2) (x + 2)

1 = a(x + 2) + b(x − 2)

Set x = 2 and x = −2 to solve for a and b.

Hint 6.9
Expanding the integral in partial fractions,

x+1 x+1 a b c
= = + +
x3 + x2 − 6x x(x − 2)(x + 3) x x−2 x+3

x + 1 = a(x − 2)(x + 3) + bx(x + 3) + cx(x − 2)

Set x = 0, x = 2 and x = −3 to solve for a, b and c.

Definite Integrals
Hint 6.10

Z 1 N
X −1
x dx = lim xn ∆x
0 N →∞
n=0
N
X −1
= lim (n∆x)∆x
N →∞
n=0

122
Hint 6.11 Rπ
Let u = sin x and dv = sin x dx. Integration by parts will give you an equation for 0 sin2 x dx.

Hint 6.12
Let H 0 (x) = h(x) and evaluate the integral in terms of H(x).

Improper Integrals
Hint 6.13

Z 4 Z 1−δ Z 4
1 1 1
dx = lim+ dx + lim+ dx
0 (x − 1)2 δ→0 0 (x − 1)2 →0 1+ (x − 1)2

Hint 6.14

Z 1 Z 1
1 1
√ dx = lim √ dx
0 x →0+  x

Hint 6.15

Z
1 1 x
dx = arctan
x2 + a2 a a

Taylor Series
Hint 6.16
a. Evaluate the integral.

123
b. Use integration by parts to evaluate the integral.
1 n
c. Use integration by parts with u = f (n+1) (x − ξ) and dv = n!
ξ .

124
6.8 Solutions
Fundamental Integration Formulas
Solution 6.1

Z
(2x + 3)10 dx

Let u = 2x + 3, g(u) = x = u−3

2
, g 0 (u) = 12 .

Z Z
10 1
(2x + 3) dx = u10 du
2
11
u 1
=
11 2
(2x + 3)11
=
22

Solution 6.2

(ln x)2
Z Z
d(ln x)
dx = (ln x)2 dx
x dx
(ln x)3
=
3

125
Solution 6.3

Z √ Z √
1 d(x2 )
x x2 + 3 dx = x2 + 3 dx
2 dx
1 (x2 + 3)3/2
=
2 3/2
(x + 3)3/2
2
=
3

Solution 6.4

Z Z
cos x 1 d(sin x)
dx = dx
sin x sin x dx
= ln | sin x|

Solution 6.5

x2 1 1 d(x3 )
Z Z
dx = dx
x3 − 5 x3 − 5 3 dx
1
= ln |x3 − 5|
3

Integration by Parts

126
Solution 6.6
Let u = x, and dv = sin x dx. Then du = dx and v = − cos x.
Z Z
x sin x dx = −x cos x + cos x dx

= −x cos x + sin x + C

Solution 6.7
Let u = x3 and dv = e2x dx. Then du = 3x2 dx and v = 12 e2x .
Z Z
3 2x 1 3 2x 3
x e dx = x e − x2 e2x dx
2 2

Let u = x2 and dv = e2x dx. Then du = 2x dx and v = 21 e2x .

Z  Z 
3 2x 1 3 2x 3 1 2 2x 2x
x e dx = x e − x e − x e dx
2 2 2

Z Z
1 3 3
3 2x
x e dx = x3 e2x − x2 e2x + x e2x dx
2 4 2

Let u = x and dv = e2x dx. Then du = dx and v = 21 e2x .

Z  Z 
3 2x 1 3 2x 3 2 2x 3 1 2x 1 2x
x e dx = x e − x e + xe − e dx
2 4 2 2 2

Z
1 3 3 3
x3 e2x dx = x3 e2x − x2 e2x + x e2x − e2x + C
2 4 4 8

Partial Fractions

127
Solution 6.8
Expanding the integrand in partial fractions,
1 1 A B
= = +
x2 −4 (x − 2)(x + 2) (x − 2) (x + 2)

1 = A(x + 2) + B(x − 2)
Setting x = 2 yields A = 14 . Setting x = −2 yields B = − 14 . Now we can do the integral.
Z Z  
1 1 1
dx = − dx
x2 − 4 4(x − 2) 4(x + 2)
1 1
= ln |x − 2| − ln |x + 2| + C
4  4
1 x − 2
 
=  +C
4 x + 2

Solution 6.9
Expanding the integral in partial fractions,
x+1 x+1 A B C
= = + +
x3 2
+ x − 6x x(x − 2)(x + 3) x x−2 x+3

x + 1 = A(x − 2)(x + 3) + Bx(x + 3) + Cx(x − 2)

Setting x = 0 yields A = − 61 . Setting x = 2 yields B = 103
. Setting x = −3 yields C = − 15 2
.
Z Z  
x+1 1 3 2
3 2
dx = − + − dx
x + x − 6x 6x 10(x − 2) 15(x + 3)
1 3 2
= − ln |x| + ln |x − 2| − ln |x + 3| + C
6 10 15
|x − 2|3/10
= ln 1/6 +C
|x| |x + 3|2/15

128
Definite Integrals

Solution 6.10

Z 1 N
X −1
x dx = lim xn ∆x
0 N →∞
n=0
N
X −1
= lim (n∆x)∆x
N →∞
n=0
N
X −1
2
= lim ∆x n
N →∞
n=0
N (N − 1)
= lim ∆x2
N →∞ 2
N (N − 1)
= lim
N →∞ 2N 2
1
=
2

129
Solution 6.11
Let u = sin x and dv = sin x dx. Then du = cos x dx and v = − cos x.
Z π π
Z π
2
cos2 x dx

sin x dx = − sin x cos x 0 +
0 0
Z π
= cos2 x dx
Z0 π
= (1 − sin2 x) dx
0
Z π
=π− sin2 x dx
0

Z π
2 sin2 x dx = π
0

Z π
π
sin2 x dx =
0 2

Solution 6.12
Let H 0 (x) = h(x).
Z f (x)
d d
h(ξ) dξ = (H(f (x)) − H(g(x)))
dx g(x) dx
= H 0 (f (x))f 0 (x) − H 0 (g(x))g 0 (x)
= h(f (x))f 0 (x) − h(g(x))g 0 (x)

Improper Integrals

130
Solution 6.13

Z 4 Z 1−δ Z 4
1 1 1
dx = lim+ dx + lim+ dx
0 (x − 1)2 δ→0 0 (x − 1) 2 →0 1+ (x − 1)
2
 1−δ  4
1 1
= lim+ − + lim+ −
δ→0 x−1 0 →0 x − 1 1+
   
1 1 1
= lim+ − 1 + lim+ − +
δ→0 δ →0 3 
=∞+∞

The integral diverges.

Solution 6.14

Z 1 Z 1
1 1
√ dx = lim √ dx
0 x →0+  x
 √ 1
= lim+ 2 x 
→0
√
= lim+ 2(1 − )
→0
=2

131
Solution 6.15

Z ∞ Z α
1 1
2
dx = lim dx
0 x +4 α→∞ 0 +4 x2

1  x α
= lim arctan
α→∞ 2 2 0
1 π
 
= −0
2 2
π
=
4

Taylor Series
Solution 6.16
a.
Z x
f (0) + f 0 (x − ξ) dξ = f (0) + [−f (x − ξ)]x0
0
= f (0) − f (0) + f (x)
= f (x)

b.
Z x Z x
0 00 0 0 x
f (0) + xf (0) + ξf (x − ξ) dξ = f (0) + xf (0) + [−ξf (x − ξ)]0 − −f 0 (x − ξ) dξ
0 0
= f (0) + xf 0 (0) − xf 0 (0) − [f (x − ξ)]x0
= f (0) − f (0) + f (x)
= f (x)

132
c. Above we showed that the hypothesis holds for n = 0 and n = 1. Assume that it holds for some n = m ≥ 0.
Z x
0 1 2 00 1 n (n) 1 n (n+1)
f (x) = f (0) + xf (0) + x f (0) + · · · + x f (0) + ξ f (x − ξ) dξ
2 n! 0 n!
 x
0 1 2 00 1 n (n) 1 n+1 (n+1)
= f (0) + xf (0) + x f (0) + · · · + x f (0) + ξ f (x − ξ)
2 n! (n + 1)! 0
Z x
1
− − ξ n+1 f (n+2) (x − ξ) dξ
0 (n + 1)!
1 1 1
= f (0) + xf 0 (0) + x2 f 00 (0) + · · · + xn f (n) (0) + xn+1 f (n+1) (0)
2 n! (n + 1)!
Z x
1
+ ξ n+1 f (n+2) (x − ξ) dξ
0 (n + 1)!

This shows that the hypothesis holds for n = m + 1. By induction, the hypothesis hold for all n ≥ 0.

133
Chapter 7

Vector Calculus

7.1 Vector Functions

Vector-valued Functions. A vector-valued function, r(t), is a mapping r : R 7→ Rn that assigns a vector to
each value of t.

r(t) = r1 (t)e1 + · · · + rn (t)en .

An example of a vector-valued function is the position of an object in space as a function of time. The function
is continous at a point t = τ if

lim r(t) = r(τ ).

t→τ

This occurs if and only if the component functions are continuous. The function is differentiable if

dr r(t + ∆t) − r(t)

≡ lim
dt ∆t→0 ∆t
exists. This occurs if and only if the component functions are differentiable.

134
 If r(t) represents the position of a particle at time t, then the velocity and acceleration of the particle are
dr d2 r
and ,
dt dt2
respectively. The speed of the particle is |r0 (t)|.

Differentiation Formulas. Let f (t) and g(t) be vector functions and a(t) be a scalar function. By writing out
components you can verify the differentiation formulas:
d
(f · g) = f 0 · g + f · g0
dt
d
(f × g) = f 0 × g + f × g0
dt
d
(af ) = a0 f + af 0
dt

7.2 Gradient, Divergence and Curl

Scalar and Vector Fields. A scalar field is a function of position u(x) that assigns a scalar to each point in
space. A function that gives the temperature of a material is an example of a scalar field. In two dimensions, you
can graph a scalar field as a surface plot, (Figure 7.1), with the vertical axis for the value of the function.
A vector field is a function of position u(x) that assigns a vector to each point in space. Examples of vectors
fields are functions that give the acceleration due to gravity or the velocity of a fluid. You can graph a vector
field in two or three dimension by drawing vectors at regularly spaced points. (See Figure 7.1 for a vector field in
two dimensions.)

Partial Derivatives of Scalar Fields. Consider a scalar field u(x). The partial derivative of u with respect
to xk is the derivative of u in which xk is considered to be a variable and the remaining arguments are considered
to be parameters. The partial derivative is denoted ∂x∂ k u(x), ∂x
∂u
k
or uxk and is defined
∂u u(x1 , . . . , xk + ∆x, . . . , xn ) − u(x1 , . . . , xk , . . . , xn )
≡ lim .
∂xk ∆x→0 ∆x

135
 1
0.5
6
0
-0.5
4
-1
0

2 2
4

6 0

Figure 7.1: A Scalar Field and a Vector Field

Partial derivatives have the same differentiation formulas as ordinary derivatives.

Consider a scalar field in R3 , u(x, y, z). Higher derivatives of u are denoted:

∂2u ∂ ∂u
uxx ≡ 2
≡ ,
∂x ∂x ∂x
∂2u ∂ ∂u
uxy ≡ ≡ ,
∂x∂y ∂x ∂y
∂4u ∂ 2 ∂ ∂u
uxxyz ≡ ≡ .
∂x2 ∂y∂z ∂x2 ∂y ∂z

136
If uxy and uyx are continuous, then

∂2u ∂2u
= .
∂x∂y ∂y∂x

This is referred to as the equality of mixed partial derivatives.

Partial Derivatives of Vector Fields. Consider a vector field u(x). The partial derivative of u with respect
to xk is denoted ∂x∂ k u(x), ∂x
∂u
k
or uxk and is defined

∂u u(x1 , . . . , xk + ∆x, . . . , xn ) − u(x1 , . . . , xk , . . . , xn )

≡ lim .
∂xk ∆x→0 ∆x

Partial derivatives of vector fields have the same differentiation formulas as ordinary derivatives.

Gradient. We introduce the vector differential operator,

∂ ∂
∇≡ e1 + · · · + en ,
∂x1 ∂xn

which is known as del or nabla. In R3 it is

∂ ∂ ∂
∇≡ i+ j + k.
∂x ∂y ∂z

Let u(x) be a differential scalar field. The gradient of u is,

∂u ∂u
∇u ≡ e1 + · · · + en ,
∂x1 ∂xn

137
Directional Derivative. Suppose you are standing on some terrain. The slope of the ground in a particular
direction is the directional derivative of the elevation in that direction. Consider a differentiable scalar field, u(x).
The derivative of the function in the direction of the unit vector a is the rate of change of the function in that
direction. Thus the directional derivative, Da u, is defined:
u(x + a) − u(x)
Da u(x) = lim
→0 
u(x1 + a1 , . . . , xn + an ) − u(x1 , . . . , xn )
= lim
→0 
(u(x) + a1 ux1 (x) + · · · + an uxn (x) + O(2 )) − u(x)
= lim
→0 
= a1 ux1 (x) + · · · + an uxn (x)

Da u(x) = ∇u(x) · a.

Tangent to a Surface. The gradient, ∇f , is orthogonal to the surface f (x) = 0. Consider a point ξ on the
surface. Let the differential dr = dx1 e1 + · · · dxn en lie in the tangent plane at ξ. Then
∂f ∂f
df = dx1 + · · · + dxn = 0
∂x1 ∂xn
since f (x) = 0 on the surface. Then
 
∂f ∂f
∇f · dr = e1 + · · · + en · (dx1 e1 + · · · + dxn en )
∂x1 ∂xn
∂f ∂f
= dx1 + · · · + dxn
∂x1 ∂xn
=0

Thus ∇f is orthogonal to the tangent plane and hence to the surface.

138
Example 7.2.1 Consider the paraboloid, x2 + y 2 − z = 0. We want to find the tangent plane to the surface at
the point (1, 1, 2). The gradient is

∇f = 2xi + 2yj − k.

At the point (1, 1, 2) this is

∇f (1, 1, 2) = 2i + 2j − k.

We know a point on the tangent plane, (1, 1, 2), and the normal, ∇f (1, 1, 2). The equation of the plane is

∇f (1, 1, 2) · (x, y, z) = ∇f (1, 1, 2) · (1, 1, 2)

2x + 2y − z = 2

The gradient of the function f (x) = 0, ∇f (x), is in the direction of the maximum directional derivative. The
magnitude of the gradient, |∇f (x)|, is the value of the directional derivative in that direction. To derive this,
note that

Da f = ∇f · a = |∇f | cos θ,

where θ is the angle between ∇f and a. Da f is maximum when θ = 0, i.e. when a is the same direction as ∇f .
In this direction, Da f = |∇f |. To use the elevation example, ∇f points in the uphill direction and |∇f | is the
uphill slope.

Example 7.2.2 Suppose that the two surfaces f (x) = 0 and g(x) = 0 intersect at the point x = ξ. What is the
angle between their tangent planes at that point? First we note that the angle between the tangent planes is by
definition the angle between their normals. These normals are in the direction of ∇f (ξ) and ∇g(ξ). (We assume
these are nonzero.) The angle, θ, between the tangent planes to the surfaces is
 
∇f (ξ) · ∇g(ξ)
θ = arccos .
|∇f (ξ)| |∇g(ξ)|

139
Example 7.2.3 Let u be the distance from the origin:
√ √
u(x) = x · x = xi xi .

In three dimensions, this is

p
u(x, y, z) = x2 + y 2 + z 2 .

The gradient of u, ∇(x), is a unit vector in the direction of x. The gradient is:
 
x1 xn xi e i
∇u(x) = √ ,... √ =√ .
x·x x·x xj x j

In three dimensions, we have

* +
x y z
∇u(x, y, z) = p ,p ,p .
x2 + y 2 + z 2 x2 + y 2 + z 2 x2 + y 2 + z 2

This is a unit vector because the sum of the squared components sums to unity.
xi ei xk e k xi xi
∇u · ∇u = √ ·√ =1
xj xj xl x l xj xj

Figure 7.2 shows a plot of the vector field of ∇u in two dimensions.

Example 7.2.4 Consider an ellipse. An implicit equation of an ellipse is

x2 y 2
+ 2 = 1.
a2 b
We can also express an ellipse as u(x, y) + v(x, y) = c where u and v are the distance from the two foci. That is,
an ellipse is the set of points such that the sum of the distances from the two foci is a constant. Let n = ∇(u + v).

140
 Figure 7.2: The gradient of the distance from the origin.

This is a vector which is orthogonal to the ellipse when evaluated on the surface. Let t be a unit tangent to the
surface. Since n and t are orthogonal,
n·t=0
(∇u + ∇v) · t = 0
∇u · t = ∇v · (−t).
Since these are unit vectors, the angle between ∇u and t is equal to the angle between ∇v and −t. In other
words: If we draw rays from the foci to a point on the ellipse, the rays make equal angles with the ellipse. If the
ellipse were a reflective surface, a wave starting at one focus would be reflected from the ellipse and travel to the
other focus. See Figure 8.3. This result also holds for ellipsoids, u(x, y, z) + v(x, y, z) = c.
We see that an ellipsoidal dish could be used to collect spherical waves, (waves emanating from a point). If
the dish is shaped so that the source of the waves is located at one foci and a collector is placed at the second,
then any wave starting at the source and reflecting off the dish will travel to the collector. See Figure 7.4.

141
 n
v

-t u
θ
θ θ
θ t
u v

Figure 7.3: An ellipse and rays from the foci.

Figure 7.4: An elliptical dish.

7.3 Exercises

Vector Functions

142
Exercise 7.1
Consider the parametric curve
   
t t
r = cos i + sin j.
2 2
dr d2 r
Calculate dt
and dt2
. Plot the position and some velocity and acceleration vectors.

Exercise 7.2
Let r(t) be the position of an object moving with constant speed. Show that the acceleration of the object is
orthogonal to the velocity of the object.

Vector Fields
Exercise 7.3
Consider the paraboloid x2 + y 2 − z = 0. What is the angle between the two tangent planes that touch the surface
at (1, 1, 2) and (1, −1, 2)? What are the equations of the tangent planes at these points?

Exercise 7.4
Consider the paraboloid x2 + y 2 − z = 0. What is the point on the paraboloid that is closest to (1, 0, 0)?

143
7.4 Hints
Vector Functions
Hint 7.1
Plot the velocity and acceleration vectors at regular intervals along the path of motion.

Hint 7.2
If r(t) has constant speed, then |r0 (t)| = c. The condition that the acceleration is orthogonal to the velocity can
be stated mathematically in terms of the dot product, r00 (t) · r0 (t) = 0. Write the condition of constant speed in
terms of a dot product and go from there.

Vector Fields
Hint 7.3
The angle between two planes is the angle between the vectors orthogonal to the planes. The angle between the
two vectors is
 
h2, 2, −1i · h2, −2, −1i
θ = arccos
|h2, 2, −1i||h2, −2, −1i|

The equation of a line orthogonal to a and passing through the point b is a · x = a · b.

Hint 7.4
Since the paraboloid is a differentiable surface, the normal to the surface at the closest point will be parallel to
the vector from the closest point to (1, 0, 0). We can express this using the gradient and the cross product. If
(x, y, z) is the closest point on the paraboloid, then a vector orthogonal to the surface there is ∇f = h2x, 2y, −1i.
The vector from the surface to the point (1, 0, 0) is h1 − x, −y, −zi. These two vectors are parallel if their cross
product is zero.

144
7.5 Solutions
Vector Functions
Solution 7.1
The velocity is
   
0 1 t 1 t
r = − sin i + cos j.
2 2 2 2

The acceleration is
   
0 1 t 1 t
r = − cos i − sin j.
4 2 4 2

See Figure 7.5 for plots of position, velocity and acceleration.

Figure 7.5: A Graph of Position and Velocity and of Position and Acceleration

145
Solution 7.2
If r(t) has constant speed, then |r0 (t)| = c. The condition that the acceleration is orthogonal to the velocity can
be stated mathematically in terms of the dot product, r00 (t) · r0 (t) = 0. Note that we can write the condition of
constant speed in terms of a dot product,
p
r0 (t) · r0 (t) = c,

r0 (t) · r0 (t) = c2 .

Differentiating this equation yields,

r00 (t) · r0 (t) + r0 (t) · r00 (t) = 0

r00 (t) · r0 (t) = 0.

This shows that the acceleration is orthogonal to the velocity.

Vector Fields
Solution 7.3
The gradient, which is orthogonal to the surface when evaluated there is ∇f = 2xi + 2yj − k. 2i + 2j − k and
2i − 2j − k are orthogonal to the paraboloid, (and hence the tangent planes), at the points (1, 1, 2) and (1, −1, 2),
respectively. The angle between the tangent planes is the angle between the vectors orthogonal to the planes.
The angle between the two vectors is
 
h2, 2, −1i · h2, −2, −1i
θ = arccos
|h2, 2, −1i||h2, −2, −1i|

 
1
θ = arccos ≈ 1.45946.
9

146
Recall that the equation of a line orthogonal to a and passing through the point b is a · x = a · b. The equations
of the tangent planes are

h2, ±2, −1i · hx, y, zi = h2, ±2, −1i · h1, ±1, 2i,

2x ± 2y − z = 2.

The paraboloid and the tangent planes are shown in Figure 7.6.
-1
0
1

1
0
-1

Figure 7.6: Paraboloid and Two Tangent Planes

Solution 7.4
Since the paraboloid is a differentiable surface, the normal to the surface at the closest point will be parallel to
the vector from the closest point to (1, 0, 0). We can express this using the gradient and the cross product. If
(x, y, z) is the closest point on the paraboloid, then a vector orthogonal to the surface there is ∇f = h2x, 2y, −1i.
The vector from the surface to the point (1, 0, 0) is h1 − x, −y, −zi. These two vectors are parallel if their cross
product is zero,

h2x, 2y, −1i × h1 − x, −y, −zi = h−y − 2yz, −1 + x + 2xz, −2yi = 0.

147
This gives us the three equations,

−y − 2yz = 0,
−1 + x + 2xz = 0,
−2y = 0.

The third equation requires that y = 0. The first equation then becomes trivial and we are left with the second
equation,

−1 + x + 2xz = 0.

Substituting z = x2 + y 2 into this equation yields,

2x3 + x − 1 = 0.

The only real valued solution of this polynomial is

√
2/3
6−2/3 9 + 87 − 6−1/3
x= √
1/3 ≈ 0.589755.
9 + 87

Thus the closest point to (1, 0, 0) on the paraboloid is


−2/3
√
2/3 −1/3 −2/3
√
2/3 −1/3
!2 
6 9 + 87 −6 6 9 + 87 −6
 √
1/3 , 0, √
1/3  ≈ (0.589755, 0, 0.34781).
9 + 87 9 + 87

The closest point is shown graphically in Figure 7.7.

148
 1-1
1
0.5 -0.5
0 0
-0.5 0.5
-1 1
2

1.5

0.5

Figure 7.7: Paraboloid, Tangent Plane and Line Connecting (1, 0, 0) to Closest Point

149
 Part III

Functions of a Complex Variable

150
Chapter 8

Complex Numbers

For every complex problem, there is a solution that is simple, neat, and wrong.

- H. L. Mencken

8.1 Complex Numbers

When you started algebra, you learned that the quadratic equation: x2 + 2ax + b = 0 has either two, one or no
solutions. For example:

• x2 − 3x + 2 = 0 has the two solutions x = 1 and x = 2.

• x2 − 2x + 1 = 0 has the one solution x = 1.

• x2 + 1 = 0 has no solutions.

151
This is a little unsatisfactory. We can formally solve the general quadratic equation.

x2 + 2ax + b = 0
(x + a)2 = a2 − b
√
x = −a ± a2 − b
√
However, the solutions are defined only when
√ a2 ≥ b. The square root function, x, is a bijection from R0+ to
R0+ . We cannot solve x2 = −1 because −1 is not defined.
√ To overcome this apparent shortcoming of the real
2
number system, we create a new symbolic constant, i ≡ −1.
Now we can express the solutions of x = −1 as
2
√ 2 2
√
2
x = i and x = −i. These satisfy the equation since i = −1 = −1 and (−i) = − −1 = −1. Note that
√ √ √ √
we can express the square root of any negative real number in terms of i: −r = −1 r = i r. We call any
number of the form ib, b ∈ R, a pure imaginary number. 1 We call numbers of the form a + ib, where a, b ∈ R,
complex numbers 2

√
The quadratic with real coefficients, x2 + 2ax + b = 0, has solutions x = −a ± a2 − b. The solutions are
real-valued only if a2 − b ≥ 0. If not,
√ then we can define solutions as complex numbers. If the discriminant is
negative, then we write x = −a ± i b − a2 . Thus every quadratic polynomial has exactly two solutions, counting
multiplicities. The fundamental theorem of algebra states that an nth degree polynomial with complex coefficients
has n, not necessarily distinct, complex roots. We will prove this result later using the theory of functions of a
complex variable.

Consider the complex number z = x + iy, (x, y ∈ R). The real part of z is <(z) = x; the imaginary part of z is
=(z) = y. Two complex numbers, z1 = x1 + iy1 and z2 = x2 + iy2 , are equal if and only if x1 = x2 and y1 = y2 .
The complex conjugate 3 of z = x + iy is z = x − iy. The notation z ∗ = x − iy is also used.
1
“Imaginary” is an unfortunate term. Real numbers are artificial; constructs of the mind. Real numbers are no more real than
imaginary numbers.
2
Here complex means “composed of two or more parts”, not “hard to separate, analyze, or solve”. Those who disagree have a
complex number complex.
3
Conjugate: having features in common but opposite or inverse in some particular.

152
 The set of complex numbers, C, form a field. That essentially means that we can do arithmetic with complex
numbers. We treat i as a symbolic constant with the property that i2 = −1. The field of complex numbers satisfy
the following properties: (Let z, z1 , z2 , z3 ∈ C.)

1. Closure under addition and multiplication.

z1 + z2 = (x1 + iy1 ) + (x2 + iy2 )

= (x1 + x2 ) + i(y1 + y2 ) ∈ C
z1 z2 = (x1 + iy1 )(x2 + iy2 )
= (x1 x2 − y1 y2 ) + i(x1 y2 + x2 y1 ) ∈ C

2. Commutativity of addition and multiplication. z1 + z2 = z2 + z1 . z1 z2 = z2 z1 .

3. Associativity of addition and multiplication. (z1 + z2 ) + z3 = z1 + (z2 + z3 ). (z1 z2 )z3 = z1 (z2 z3 ).

4. Distributive law. z1 (z2 + z3 ) = z1 z2 + z1 z3 .

5. Identity with respect to addition and multiplication. z + 0 = z. z(1) = z.

6. Inverse with respect to addition. z + (−z) = (x + iy) + (−x − iy) = 0.

7. Inverse with respect to multiplication for nonzero numbers. zz −1 = 1, where

1 1 x − iy x y
z −1 = = = 2 2
= 2 2
−i 2
z x + iy x +y x +y x + y2

Complex Conjugate. Using the field properties of complex numbers, we can derive the following properties
of the complex conjugate, z = x − iy.

1. (z) = z,

2. z + ζ = z + ζ,

153
 3. zζ = zζ,
 
z z
4. = .
ζ ζ

8.2 The Complex Plane

We can denote a complex number z = x + iy as an ordered pair of real numbers (x, y). Thus we can represent
a complex number as a point in R2 where the x component is the real part and the y component is the imaginary
part of z. This is called the complex plane or the Argand diagram. (See Figure 8.1.)

Im(z)

(x,y)
r
θ Re(z)

Figure 8.1: The Complex Plane

There are two ways of describing a point in the complex plane: an ordered pair of coordinates (x, y) that give
the horizontal and vertical offset from the origin or the distance r from the origin and the angle θ from the positive
horizontal axis. The angle θ is not unique. It is only determined up to an additive integer multiple of 2π.

Modulus. The magnitudepor modulus of a complex number is the distance of the point from the origin. It is
defined as |z| = |x + iy| = x2 + y 2 . Note that zz = (x + iy)(x − iy) = x2 + y 2 = |z|2 . The modulus has the
following properties.

154
 1. |z1 z2 | = |z1 | |z2 |
 
 z1  |z1 |
2.   = for z2 6= 0.
z2 |z2 |
3. |z1 + z2 | ≤ |z1 | + |z2 |
4. |z1 + z2 | ≥ ||z1 | − |z2 ||
We could prove the first two properties by expanding in x + iy form, but it would be fairly messy. The proofs
will become simple after polar form has been introduced. The second two properties follow from the triangle
inequalities in geometry. This will become apparent after the relationship between complex numbers and vectors
is introduced. One can show that
|z1 z2 · · · zn | = |z1 | |z2 | · · · |zn |
and
|z1 + z2 + · · · + zn | ≤ |z1 | + |z2 | + · · · + |zn |
with proof by induction.

Argument. The argument of a complex number is the angle that the vector with tail at the origin and head
at z = x + iy makes with the positive x-axis. The argument is denoted arg(z). Note that the argument is defined
for all nonzero numbers and is only determined up to an additive integer multiple of 2π. That is, the argument
of a complex number is the set of values: {θ + 2πn | n ∈ Z}. The principal argument of a complex number is that
angle in the set arg(z) which lies in the range (−π, π]. The principal argument is denoted Arg (z). We prove the
following identities in Exercise 8.7.
arg(zζ) = arg(z) + arg(ζ)
Arg (zζ) 6= Arg (z) + Arg (ζ)
arg(z 2 ) = arg(z) + arg(z) 6= 2 arg(z)

155
Example 8.2.1 Consider the equation |z − 1 − i| = 2. The set of points satisfying this equation is a circle of
radius 2 and center at 1 + i in the complex plane. You can see this by noting that |z − 1 − i| is the distance from
the point (1, 1). (See Figure 8.2.)

-1 1 2 3

-1

Figure 8.2: Solution of |z − 1 − i| = 2

Another way to derive this is to substitute z = x + iy into the equation.

|x + iy − 1 − i| = 2
p
(x − 1)2 + (y − 1)2 = 2
(x − 1)2 + (y − 1)2 = 4

This is the analytic geometry equation for a circle of radius 2 centered about (1, 1).

Example 8.2.2 Consider the curve described by

|z| + |z − 2| = 4.

156
Note that |z| is the distance from the origin in the complex plane and |z − 2| is the distance from z = 2. The
equation is

(distance from (0, 0)) + (distance from (2, 0)) = 4.

√
From geometry, we know that this is an ellipse with foci at (0, 0) and (2, 0), major axis 2, and minor axis 3.
(See Figure 8.3.)

-1 1 2 3

-1

-2

Figure 8.3: Solution of |z| + |z − 2| = 4

157
 We can use the substitution z = x + iy to get the equation an algebraic form.
|z| + |z − 2| = 4
|x + iy| + |x + iy − 2| = 4
p p
x2 + y 2 + (x − 2)2 + y 2 = 4
p
x2 + y 2 = 16 − 8 (x − 2)2 + y 2 + x2 − 4x + 4 + y 2
p
x − 5 = −2 (x − 2)2 + y 2
x2 − 10x + 25 = 4x2 − 16x + 16 + 4y 2
1 1
(x − 1)2 + y 2 = 1
4 3
Thus we have the standard form for an equation describing an ellipse.

8.3 Polar Form

Polar Form. A complex number written as z = x + iy is said to be in Cartesian form, or a + ib form.
We can convert this representation to polar form, z = r(cos θ + i sin θ), using trigonometry. Here r = |z| is
the modulus and θ = arctan(x, y) is the argument of z. The argument is the angle between the x axis and
the vector with its head at (x, y). (See Figure 8.4.) Note that θ is not unique. If z = r(cos θ + i sin θ) then
z = r(cos(θ + 2nπ) + i sin(θ + 2nπ)) for any n ∈ Z.

The Arctangent. Note
that arctan(x, y) is not the same thing as the old arctangent that you learned about
in trigonometry, arctan xy . For example,
π −3π
arctan(1, 1) = + 2nπ and arctan(−1, −1) = + 2nπ,
4 4
whereas
   
−1 1
arctan = arctan = arctan(1).
−1 1

158
 Im( z ) (x,y)

r r sinθ
θ Re(z )
r cos θ

Figure 8.4: Polar Form

Euler’s Formula. Euler’s formula, eiθ = cos θ + i sin θ, allows us to write the polar form more compactly.
Expressing the polar form in terms of the exponential function of imaginary argument makes arithmetic with
complex numbers much more convenient. (See Exercise 8.14 for a proof of Euler’s formula.)

z = r(cos θ + i sin θ) = r eiθ

Arithmetic With Complex Numbers. Note that it is convenient to add complex numbers in Cartesian form.

(x1 + iy1 ) + (x2 + iy2 ) = (x1 + x2 ) + i(y1 + y2 )

However, it is difficult to multiply or divide them in Cartesian form.

(x1 + iy1 )(x2 + iy2 ) = (x1 x2 − y1 y2 ) + i(x1 y2 + x2 y1 )

x1 + iy1 (x1 + iy1 )(x2 − iy2 ) x1 x2 + y 1 y 2 x2 y1 − x1 y2

= = 2 2
+i
x2 + iy2 (x2 + iy2 )(x2 − iy2 ) x2 + y2 x22 + y22

159
On the other hand, it is difficult to add complex numbers in polar form.

r1 eiθ1 + r2 eiθ2 = r1 (cos θ1 + i sin θ1 ) + r2 (cos θ2 + i sin θ2 )

= r1 cos θ1 + r2 cos θ2 + i(r1 sin θ1 + r2 sin θ2 )
p
= (r1 cos θ1 + r2 cos θ2 )2 + (r1 sin θ1 + r2 sin θ2 )2
× ei arctan(r1 cos θ1 +r2 cos θ2 ,r1 sin θ1 +r2 sin θ2 )
q
= r12 + r22 + 2 cos(θ1 − θ2 ) ei arctan(r1 cos θ1 +r2 cos θ2 ,r1 sin θ1 +r2 sin θ2 )

However, it is convenient to multiply and divide them in polar form.

r1 eiθ1 r2 eiθ2 = r1 r2 ei(θ1 +θ2 )

r1 eiθ1 r1 i(θ1 −θ2 )

= e
r2 eiθ2 r2
Keeping this in mind will make working with complex numbers a shade or two less grungy.
Result 8.3.1 To change between Cartesian and polar form, use the identities

r eiθ = r cos θ + ir sin θ,

p
x + iy = x2 + y 2 ei arctan(x,y) .

Cartesian form is convenient for addition. Polar form is convenient for multiplication
and division.

Example 8.3.1 The polar form of 5 + 7i is

√
5 + 7i = 74 ei arctan(5,7) .

160
2 eiπ/6 in Cartesian form is
π  π 
iπ/6
2e = 2 cos + 2i sin
√ 6 6
= 3 + i.

Example 8.3.2 We will show that

1 1 3
cos4 θ = cos 4θ + cos 2θ + .
8 2 8
Recall that
eiθ + e−iθ eiθ − e−iθ
cos θ = and sin θ = .
2 2i

4
eiθ + e−iθ

4
cos θ =
2
1 4iθ
= ( e + 4 e2iθ + 6 + 4 e−2iθ + e−4iθ )
16
1 e4iθ + e−4iθ 1 e2iθ + e−2iθ
  
3
= + +
8 2 2 2 8
1 1 3
= cos 4θ + cos 2θ +
8 2 8

By the definition of exponentiation, we have einθ = ( eiθ )n We apply Euler’s formula to obtain a result which is
useful in deriving trigonometric identities.

cos(nθ) + i sin(nθ) = (cos(θ) + i sin(θ))n

161
 a
Result 8.3.2 DeMoivre’s Theorem.

cos(nθ) + i sin(nθ) = (cos(θ) + i sin(θ))n

a
It’s amazing what passes for a theorem these days. I would think that this would be a corollary at most.

Example 8.3.3 We will express cos 5θ in terms of cos θ and sin 5θ in terms of sin θ.
We start with DeMoivre’s theorem.
ei5θ = ( eiθ )5

cos 5θ + i sin 5θ = (cos θ + i sin θ)5

       
5 5 5 4 5 3 2 5
= cos θ + i cos θ sin θ − cos θ sin θ − i cos2 θ sin3 θ
0 1 2 3
   
5 4 5
+ cos θ sin θ + i sin5 θ
4 5
= cos θ − 10 cos θ sin θ + 5 cos θ sin4 θ + i 5 cos4 θ sin θ − 10 cos2 θ sin3 θ + sin5 θ
5 3 2

Equating the real and imaginary parts we obtain

cos 5θ = cos5 θ − 10 cos3 θ sin2 θ + 5 cos θ sin4 θ
sin 5θ = 5 cos4 θ sin θ − 10 cos2 θ sin3 θ + sin5 θ
Now we use the Pythagorean identity, cos2 θ + sin2 θ = 1.
cos 5θ = cos5 θ − 10 cos3 θ(1 − cos2 θ) + 5 cos θ(1 − cos2 θ)2
cos 5θ = 16 cos5 θ − 20 cos3 θ + 5 cos θ
sin 5θ = 5(1 − sin2 θ)2 sin θ − 10(1 − sin2 θ) sin3 θ + sin5 θ
sin 5θ = 16 sin5 θ − 20 sin3 θ + 5 sin θ

162
8.4 Arithmetic and Vectors
Addition. We can represent the complex number z = x + iy = r eiθ as a vector in Cartesian space with tail at
the origin and head at (x, y), or equivalently, the vector of length r and angle θ. With the vector representation,
we can add complex numbers by connecting the tail of one vector to the head of the other. The vector z + ζ is
the diagonal of the parallelogram defined by z and ζ. (See Figure 8.5.)

Negation. The negative of z = x + iy is −z = −x − iy. In polar form we have z = r eiθ and −z = r ei(θ+π) ,
(more generally, z = r ei(θ+(2n+1)π) , n ∈ Z. In terms of vectors, −z has the same magnitude but opposite direction
as z. (See Figure 8.5.)

Multiplication. The product of z = r eiθ and ζ = ρ eiφ is zζ = rρ ei(θ+φ) . The length of the vector zζ is the
product of the lengths of z and ζ. The angle of zζ is the sum of the angles of z and ζ. (See Figure 8.5.)
Note that arg(zζ) = arg(z) + arg(ζ). Each of these arguments has an infinite number of values. If we write
out the multi-valuedness explicitly, we have

{θ + φ + 2πn : n ∈ Z} = {θ + 2πn : n ∈ Z} + {φ + 2πn : n ∈ Z}

The same is not true of the principal argument. In general, Arg (zζ) 6= Arg (z) + Arg (ζ). Consider the case
z = ζ = ei3π/4 . Then Arg (z) = Arg (ζ) = 3π/4, however, Arg (zζ) = −π/2.

Multiplicative Inverse. Assume that z is nonzero. The multiplicative inverse of z = r eiθ is z1 = 1r e−iθ . The
length of z1 is the multiplicative inverse of the length of z. The angle of z1 is the negative of the angle of z. (See
Figure 8.6.)

Division. Assume that ζ is nonzero. The quotient of z = r eiθ and ζ = ρ eiφ is zζ = ρr ei(θ−φ) . The length of the
vector zζ is the quotient of the lengths of z and ζ. The angle of zζ is the difference of the angles of z and ζ. (See
Figure 8.6.)

163
 Im(z ) Im( z ) z ζ=(xξ− y η)+ i(xη+y ξ)
Im( z )
z+ ζ=(x+ξ)+i(y+η) =rρe
z=x+iy i(θ+φ ) ζ=ξ+iη=ρei φ
=reiθ z=x+iy =reiθ
ζ=ξ+iη z=x+iy

Re(z ) Re(z ) Re(z )

-z=-x-iy
=rei(θ+π)

Figure 8.5: Addition, Negation and Multiplication

Complex Conjugate. The complex conjugate of z = x + iy = r eiθ is z = x − iy = r e−iθ . z is the mirror image
of z, reflected across the x axis. In other words, z has the same magnitude as z and the angle of z is the negative
of the angle of z. (See Figure 8.6.)

8.5 Integer Exponents

Consider the product (a + b)n , n ∈ Z. If we know arctan(a, b) then it will be most convenient to expand the
product working in polar form. If not, we can write n in base 2 to efficiently do the multiplications.
√
Example 8.5.1 Suppose that we want to write ( 3 + i)20 in Cartesian form. 4 We can do the multiplication √ n
directly. Note that 20 is 10100 in base 2. That is, 20 = 24 +22 . We first calculate the powers of the form ( 3+i)2

4
No, I have no idea why we would want to do that. Just humor me. If you pretend that you’re interested, I’ll do the same. Believe
me, expressing your real feelings here isn’t going to do anyone any good.

164
 Im( z ) Im(z ) ζ=ρ eiφ Im(z )
z=reiθ z=x+iy=reiθ
z=reiθ

Re(z ) Re(z ) Re(z )

z r
- =- ei (θ−φ)
-z1 = −e
1 -i θ ζ ρ
r
-z=x-iy=re -i θ

Figure 8.6: Multiplicative Inverse, Division and Complex Conjugate

by successive squaring.

√ √
( 3 + i)2 = 2 + i2 3
√ √
( 3 + i)4 = −8 + i8 3
√ √
( 3 + i)8 = −128 − i128 3
√ √
( 3 + i)16 = −32768 + i32768 3

√ √
Next we multiply ( 3 + i)4 and ( 3 + i)16 to obtain the answer.

√ √ √ √
( 3 + i)20 = (−32768 + i32768 3)(−8 + i8 3) = −524288 − i524288 3

√
Since we know that arctan( 3, 1) = π/6, it is easiest to do this problem by first changing to modulus-argument

165
form.
!20
√ 20 √ 2
r 
√
i arctan( 3,1)
3+i = 3 + 12 e

20
= 2 eiπ/6
= 220 ei4π/3
√ !
1 3
= 1048576 − − i
2 2
√
= −524288 − i524288 3

Example 8.5.2 Consider (5+7i)11 . We will do the exponentiation in polar form and write the result in Cartesian
form.
√ 11
(5 + 7i)11 = 74 ei arctan(5,7)
√
= 745 74(cos(11 arctan(5, 7)) + i sin(11 arctan(5, 7)))
√ √
= 2219006624 74 cos(11 arctan(5, 7)) + i2219006624 74 sin(11 arctan(5, 7))
The result is correct, but not very satisfying. This expression could be simplified. You could evaluate the
trigonometric functions with some fairly messy trigonometric identities. This would take much more work than
directly multiplying (5 + 7i)11 .

8.6 Rational Exponents

In this section we consider complex numbers with rational exponents, z p/q , where p/q is a rational number.
First we consider unity raised to the 1/n power. We define 11/n as the set of numbers {z} such that z n = 1.
11/n = {z | z n = 1}

166
We can find these values by writing z in modulus-argument form.
zn = 1
rn einθ = 1
rn = 1 nθ = 0 mod 2π
r=1 θ = 2πk for k ∈ Z
There are only n distinct solutions as a result of the 2π periodicity of eiθ . Thus
11/n = { ei2πk/n | k = 0, . . . , n − 1}.
These values are equally spaced points on the unit circle in the complex plane.

Example 8.6.1 11/6 has the 6 values,

ei0 , eiπ/3 , ei2π/3 , eiπ , ei4π/3 , ei5π/3 .


In Cartesian form this is

( √ √ √ √ )
1 + i 3 −1 + i 3 −1 − i 3 1 − i 3
1, , , −1, , .
2 2 2 2
The sixth roots of unity are plotted in Figure 8.7.

The nth roots of the complex number c = α eiβ are the set of numbers z = r eiθ such that
z n = c = α eiβ
rn einθ = α eiβ
√
n
r= α nθ = β mod 2π
√
r= nα θ = (β + 2πk)/n for k = 0, . . . , n − 1.
Thus
√ p
c1/n = { n α ei(β+2πk)/n | k = 0, . . . , n − 1} = { n |c| ei( Arg (c)+2πk)/n | k = 0, . . . , n − 1}

167
 1

-1 1

-1

Figure 8.7: The Sixth Roots of Unity.

Principal Roots. The principal nth root is denoted

√ √
n
z ≡ n z ei Arg (z)/n .

Thus the principal root has the property

√

−π/n < Arg n
z ≤ π/n.
√
This is consistent with the notation you learned back in algebra where n x denoted the positive nth root of a
positive
√ real number. We adopt the convention that z 1/n denotes the nth roots of z, which is a set of n numbers
and z is the principal nth root of z, which is a single number. With the principal root we can write,
n

√
z 1/n = n r ei( Arg (z)+2πk)/n | k = 0, . . . , n − 1}
√
= n z ei2πk/n | k = 0, . . . , n − 1}

√
z 1/n = n
z11/n .

That is, the nth roots of z are the principal nth root of z times the nth roots of unity.

168
Rational Exponents. We interpret z p/q to mean z (p/q) . That is, we first simplify the exponent, i.e. reduce
the fraction, before carrying out the exponentiation. Therefore z 2/4 = z 1/2 and z 10/5 = z 2 . If p/q is a reduced
fraction, (p and q are relatively prime, in other words, they have no common factors), then
z p/q ≡ (z p )1/q .
Thus z p/q is a set of q values. Note that for an un-reduced fraction r/s,

r
(z r )1/s 6= z 1/s .
The former expression is a set of s values while the latter is a set of no more that s values. For instance,
(12 )1/2 = 11/2 = ±1 and (11/2 )2 = (±1)2 = 1.

Example 8.6.2 Consider 21/5 , (1 + i)1/3 and (2 + i)5/6 .

√
21/5 = 2 ei2πk/5 , for k = 0, 1, 2, 3, 4
5

√ 1/3
(1 + i)1/3 = 2 eiπ/4
√
= 2 eiπ/12 ei2πk/3 ,
6
for k = 0, 1, 2
√ 5/6
5/6 i Arctan (2,1)
(2 + i) = 5e
√ 1/6
= 55 ei5 Arctan (2,1)
√ i 5 Arctan (2,1) iπk/3
12
= 55 e 6 e , for k = 0, 1, 2, 3, 4, 5

Example 8.6.3 The roots of the polynomial z 5 + 4 are

1/5
(−4)1/5 = 4 eiπ
√
= 4 eiπ(1+2k)/5 ,
5
for k = 0, 1, 2, 3, 4.

169
8.7 Exercises
Complex Numbers
Exercise 8.1
Verify that:

1 + 2i 2 − i 2
1. + =−
3 − 4i 5i 5

2. (1 − i)4 = −4

Exercise 8.2
Write the following complex numbers in the form a + ib.
 √ −10
1. 1+i 3

2. (11 + 4i)2

Exercise 8.3
Write the following complex numbers in the form a + ib
 2
2+i
1.
i6 − (1 − i2)

2. (1 − i)7

Exercise 8.4
If z = x + iy, write the following in the form u(x, y) + iv(x, y).

170
  
z
1.
z
z + 2i
2.
2 − iz

Exercise 8.5
Quaternions are sometimes used as a generalization of complex numbers. A quaternion u may be defined as

u = u0 + iu1 + ju2 + ku3

where u0 , u1 , u2 and u3 are real numbers and i, j and k are objects which satisfy

i2 = j 2 = k 2 = −1, ij = k, ji = −k

and the usual associative and distributive laws. Show that for any quaternions u, w there exists a quaternion v
such that

uv = w

except for the case u0 = u1 = u2 = u3 .

Exercise 8.6
Let α 6= 0, β 6= 0 be two complex numbers. Show that α = tβ for some real number t (i.e. the vectors defined by
α and β are parallel) if and only if =(αβ) = 0.

The Complex Plane

Exercise 8.7
Prove the following identities.
1. arg(zζ) = arg(z) + arg(ζ)

171
 2. Arg (zζ) 6= Arg (z) + Arg (ζ)

3. arg(z 2 ) = arg(z) + arg(z) 6= 2 arg(z)

Exercise 8.8
Show, both by geometric and algebraic arguments, that for complex numbers z1 and z2 the inequalities

||z1 | − |z2 || ≤ |z1 + z2 | ≤ |z1 | + |z2 |

hold.

Exercise 8.9
Find all the values of

1. (−1)−3/4

2. 81/6

and show them graphically.

Exercise 8.10
Find all values of

1. (−1)−1/4

2. 161/8

and show them graphically.

172
Exercise 8.11
Sketch the regions or curves described by

1. 1 < |z − 2i| < 2

2. |<(z)| + 5|=(z)| = 1

Exercise 8.12
Sketch the regions or curves described by

1. |z − 1 + i| ≤ 1

2. |z − i| = |z + i|

3. <(z) − =(z) = 5

4. |z − i| + |z + i| = 1

Exercise 8.13
Solve the equation

| eiθ − 1| = 2

for θ (0 ≤ θ ≤ π) and verify the solution geometrically.

Polar Form
Exercise 8.14
Prove Euler’s formula, eiθ = cos θ + i sin θ. Consider the Taylor series of ez to be the definition of the exponential
function.

173
Exercise 8.15
Use de Moivre’s formula to derive the trigonometric identity

cos(3θ) = cos3 (θ) − 3 cos(θ) sin2 (θ).

Exercise 8.16
Establish the formula

1 − z n+1
1 + z + z2 + · · · + zn = , (z 6= 1),
1−z
for the sum of a finite geometric series; then derive the formulas
1 sin((n + 1/2))
1. 1 + cos θ + cos 2θ + · · · + cos nθ = +
2 2 sin(θ/2)

1 θ cos((n + 1/2))
2. sin θ + sin 2θ + · · · + sin nθ = cot −
2 2 2 sin(θ/2)
where 0 < θ < 2π.

Arithmetic and Vectors

Exercise 8.17  
|z1 |
Prove |z1 z2 | = |z1 ||z2 | and  zz12  = using polar form.
 
|z2 |

Exercise 8.18
Prove that

|z + ζ|2 + |z − ζ|2 = 2 |z|2 + |ζ|2 .

Interpret this geometrically.

174
Integer Exponents
Exercise 8.19
Write (1 + i)10 in Cartesian form with the following two methods:

1. Just do the multiplication. If it takes you more than four multiplications, you suck.

2. Do the multiplication in polar form.

Rational Exponents
Exercise 8.20
Show that each of the numbers z = −a + (a2 − b)1/2 satisfies the equation z 2 + 2az + b = 0.

175
8.8 Hints
Complex Numbers
Hint 8.1

Hint 8.2

Hint 8.3

Hint 8.4

Hint 8.5

Hint 8.6

The Complex Plane

Hint 8.7
Write the multivaluedness explicitly.

176
Hint 8.8
Consider a triangle with vertices at 0, z1 and z1 + z2 .

Hint 8.9

Hint 8.10

Hint 8.11

Hint 8.12

Hint 8.13

Polar Form
Hint 8.14
Find the Taylor series of eiθ , cos θ and sin θ. Note that i2n = (−1)n .

Hint 8.15

177
Hint 8.16

Arithmetic and Vectors

Hint 8.17
| eiθ | = 1.

Hint 8.18
Consider the parallelogram defined by z and ζ.

Integer Exponents
Hint 8.19
For the first part,

10

2
2 2
(1 + i) = (1 + i) (1 + i)2 .

Rational Exponents
Hint 8.20
Substitite the numbers into the equation.

178
8.9 Solutions

Complex Numbers

Solution 8.1
1.

1 + 2i 2 − i 1 + 2i 3 + 4i 2 − i −i
+ = +
3 − 4i 5i 3 − 4i 3 + 4i 5i −i
−5 + 10i −1 − 2i
= +
25 5
2
=−
5

2.

(1 − i)4 = (−2i)2 = −4

179
Solution 8.2
1. First we do the multiplication in Cartesian form.

√ −10 √ 2  √ 8 −1
  
1+i 3 = 1+i 3 1+i 3

√  √ 4 −1
 
= −2 + i2 3 −2 + i2 3

√  √ 2 −1
 
= −2 + i2 3 −8 − i8 3
 √  √ −1
= −2 + i2 3 −128 + i128 3
 √ −1
= −512 − i512 3
1 −1
= √
512 1 + i 3
√
1 −1 1 − i 3
= √ √
512 1 + i 3 1 − i 3
√
1 3
=− +i
2048 2048

180
 Now we do the multiplication in modulus-argument, (polar), form.
 √ −10
−10
1+i 3 = 2 eiπ/3
= 2−10 e−i10π/3
    
1 10π 10π
= cos − + i sin −
1024 3 3
    
1 4π 4π
= cos − i sin
1024 3 3
√ !
1 1 3
= − +i
1024 2 2
√
1 3
=− +i
2048 2048
2.

(11 + 4i)2 = 105 + i88

Solution 8.3
1.
 2  2
2+i 2+i
=
i6 − (1 − i2) −1 + i8
3 + i4
=
−63 − i16
3 + i4 −63 + i16
=
−63 − i16 −63 + i16
253 204
=− −i
4225 4225

181
 2.

(1 − i)7 = ((1 − i)2 )2 (1 − i)2 (1 − i)

= (−i2)2 (−i2)(1 − i)
= (−4)(−2 − i2)
= 8 + i8

Solution 8.4
1.

   
z x + iy
=
z x + iy
 
x − iy
=
x + iy
x + iy
=
x − iy
x + iy x + iy
=
x − iy x + iy
x2 − y 2 2xy
= 2 + i
x + y2 x2 + y 2

182
 2.
z + 2i x + iy + 2i
=
2 − iz 2 − i(x − iy)
x + i(y + 2)
=
2 − y − ix
x + i(y + 2) 2 − y + ix
=
2 − y − ix 2 − y + ix
x(2 − y) − (y + 2)x x2 + (y + 2)(2 − y)
= + i
(2 − y)2 + x2 (2 − y)2 + x2
−2xy 4 + x2 − y 2
= + i
(2 − y)2 + x2 (2 − y)2 + x2

Solution 8.5
Method 1. We expand the equation uv = w in its components.

uv = w
(u0 + iu1 + ju2 + ku3 )(v0 + iv1 + jv2 + kv3 ) = w0 + iw1 + jw2 + kw3

(u0 v0 − u1 v1 − u2 v2 − u3 v3 ) + i(u1 v0 + u0 v1 − u3 v2 + u2 v3 ) + j(u2 v0 + u3 v1 + u0 v2 − u1 v3 )

+ k(u3 v0 − u2 v1 + u1 v2 + u0 v3 ) = w0 + iw1 + jw2 + kw3

We can write this as a matrix equation.

    
u0 −u1 −u2 −u3 v0 w0
u1 u0 −u3 u2  v1  w1 
   =  
u2 u3 u0 −u1  v2  w2 
u3 −u2 u1 u0 v3 w3

183
This linear system of equations has a unique solution for v if and only if the determinant of the matrix is nonzero.
2
The determinant of the matrix is (u20 + u21 + u22 + u23 ) . This is zero if and only if u0 = u1 = u2 = u3 = 0. Thus
there exists a unique v such that uv = w if u is nonzero. This v is

v = (u0 w0 + u1 w1 + u2 w2 + u3 w3 ) + i(−u1 w0 + u0 w1 + u3 w2 − u2 w3 ) + j(−u2 w0 − u3 w1 + u0 w2 + u1 w3 )

+ k(−u3 w0 + u2 w1 − u1 w2 + u0 w3 ) /(u20 + u21 + u22 + u23 )

Method 2. Note that uu is a real number.

uu = (u0 − iu1 − ju2 − ku3 )(u0 + iu1 + ju2 + ku3 )

= (u20 + u21 + u22 + u23 ) + i(u0 u1 − u1 u0 − u2 u3 + u3 u2 )
+ j(u0 u2 + u1 u3 − u2 u0 − u3 u1 ) + k(u0 u3 − u1 u2 + u2 u1 − u3 u0 )
= (u20 + u21 + u22 + u23 )

uu = 0 only if u = 0. We solve for v by multiplying by the conjugate of u and divide by uu.

uv = w
uuv = uw
uw
v=
uu
(u0 − iu1 − ju2 − ku3 )(w0 + iw1 + jw2 + kw3 )
v=
u20 + u21 + u22 + u23

v = (u0 w0 + u1 w1 + u2 w2 + u3 w3 ) + i(−u1 w0 + u0 w1 + u3 w2 − u2 w3 ) + j(−u2 w0 − u3 w1 + u0 w2 + u1 w3 )

+ k(−u3 w0 + u2 w1 − u1 w2 + u0 w3 ) /(u20 + u21 + u22 + u23 )

Solution 8.6
If α = tβ, then αβ = t|β|2 , which is a real number. Hence =(αβ) = 0.

184
 Now assume that =(αβ) = 0. This implies that αβ = r for some r ∈ R. We multiply by β and simplify.

α|β|2 = rβ
r
α= β
|β|2
r
By taking t = |β|2
We see that α = tβ for some real number t.

The Complex Plane

Solution 8.7
Let z = r eiθ and ζ = ρ eiϑ .
1.

arg(zζ) = arg(z) + arg(ζ)

arg(rρ ei(θ+ϑ) ) = {θ + 2πm} + {ϑ + 2πn}
{θ + ϑ + 2πk} = {θ + ϑ + 2πm}

2.

Arg (zζ) 6= Arg (z) + Arg (ζ)

Consider z = ζ = −1. Arg (z) = Arg (ζ) = π, however Arg (zζ) = Arg (1) = 0. The identity becomes
0 6= 2π.

3.

arg(z 2 ) = arg(z) + arg(z) 6= 2 arg(z)

arg(r2 ei2θ ) = {θ + 2πk} + {θ + 2πm} =
6 2{θ + 2πn}
{2θ + 2πk} = {2θ + 2πm} 6= {2θ + 4πn}

185
 z1+z2
|z2|
z1

|z | |z +z2|
1 1

Figure 8.8: Triangle Inequality

Solution 8.8
Consider a triangle in the complex plane with vertices at 0, z1 and z1 + z2 . (See Figure 8.8.)

The lengths of the sides of the triangle are |z1 |, |z2 | and |z1 + z2 | The second inequality shows that one side of
the triangle must be less than or equal to the sum of the other two sides.

|z1 + z2 | ≤ |z1 | + |z2 |

The first inequality shows that the length of one side of the triangle must be greater than or equal to the difference
in the length of the other two sides.

|z1 + z2 | ≥ ||z1 | − |z2 ||

186
 Now we prove the inequalities algebraically. We will reduce the inequality to an identity. Let z1 = r1 eiθ1 ,
z2 = r2 eiθ2 .

||z1 | − |z2 || ≤ |z1 + z2 | ≤ |z1 | + |z2 |

|r1 − r2 | ≤ |r1 eiθ1 + r2 eiθ2 | ≤ r1 + r2
(r1 − r2 )2 ≤ (r1 eiθ1 + r2 eiθ2 )(r1 e−iθ1 + r2 e−iθ2 ) ≤ (r1 + r2 )2
r12 + r22 − 2r1 r2 ≤ r12 + r22 + r1 r2 ei(θ1 −θ2 ) + r1 r2 ei(−θ1 +θ2 ) ≤ r12 + r22 + 2r1 r2
−2r1 r2 ≤ 2r1 r2 cos(θ1 − θ2 ) ≤ 2r1 r2
−1 ≤ cos(θ1 − θ2 ) ≤ 1

Solution 8.9
1.

1/4
(−1)−3/4 = (−1)−3
= (−1)1/4
= ( eiπ )1/4
= eiπ/4 11/4
= eiπ/4 eikπ/2 , k = 0, 1, 2, 3
= eiπ/4 , ei3π/4 , ei5π/4 , ei7π/4

 
1 + i −1 + i −1 − i 1 − i
= √ , √ , √ , √
2 2 2 2

See Figure 8.9.

187
 1

-1 1

-1

Figure 8.9: (−1)−3/4

2.

√
81/6 = 811/6
6

√ ikπ/3
= 2e , k = 0, 1, 2, 3, 4, 5
n√ √ √ √ √ √ o
= 2, 2 eiπ/3 , 2 ei2π/3 , 2 eiπ , 2 ei4π/3 , 2 ei5π/3
( √ √ √ √ )
√ 1 + i 3 −1 + i 3 √ −1 − i 3 1 − i 3
= 2, √ , √ ,− 2 √ , √
2 2 2 2

See Figure 8.10.

188
 2

-2 -1 1 2

-1

-2

Figure 8.10: 81/6

Solution 8.10
1.

(−1)−1/4 = ((−1)−1 )1/4

= (−1)1/4
= ( eiπ )1/4
= eiπ/4 11/4
= eiπ/4 eikπ/2 , k = 0, 1, 2, 3
= eiπ/4 , ei3π/4 , ei5π/4 , ei7π/4

 
1 + i −1 + i −1 − i 1 − i
= √ , √ , √ , √
2 2 2 2

See Figure 8.11.

189
 1

-1 1

-1

Figure 8.11: (−1)−1/4

2.

√
161/8 = 16 11/8
8

√
= 2 eikπ/4 , k = 0, 1, 2, 3, 4, 5, 6, 7
n√ √ √ √ √ √ √ √ o
= 2, 2 eiπ/4 , 2 eiπ/2 , 2 ei3π/4 , 2 eiπ , 2 ei5π/4 , 2 ei3π/2 , 2 ei7π/4
n√ √ √ √ o
= 2, 1 + i, 2i, −1 + i, − 2, −1 − i, − 2i, 1 − i

See Figure 8.12.

Solution 8.11
1. |z − 2i| is the distance from the point 2i in the complex plane. Thus 1 < |z − 2i| < 2 is an annulus. See
Figure 8.13.

190
 1

-1 1

-1

Figure 8.12: 16−1/8

5
4
3
2
1
-3 -2 -1 1 2 3
-1

Figure 8.13: 1 < |z − 2i| < 2

191
 2.

|<(z)| + 5|=(z)| = 1
|x| + 5|y| = 1

In the first quadrant this is the line y = (1 − x)/5. We reflect this line segment across the coordinate axes
to obtain line segments in the other quadrants. Explicitly, we have the set of points: {z = x + iy : −1 <
x < 1 ∧ y = ±(1 − |x|)/5}. See Figure 8.14.

0.4

0.2

-1 1
-0.2

-0.4

Figure 8.14: |<(z)| + 5|=(z)| = 1

Solution 8.12
1. |z − 1 + i| is the distance from the point (1 − i). Thus |z − 1 + i| ≤ 1 is the disk of unit radius centered at
(1 − i). See Figure 8.15.

2. The set of points equidistant from i and −i is the real axis. See Figure 8.16.

192
 1

-1 1 2 3

-1

-2

-3

Figure 8.15: |z − 1 + i| < 1

-1 1

-1

Figure 8.16: |z − i| = |z + i|

193
3.

<(z) − =(z) = 5
x−y =5
y =x−5

See Figure 8.17.

-10 -5 5 10

-5

-10

-15

Figure 8.17: <(z) − =(z) = 5

4. Since |z − i| + |z + i| ≥ 2, there are no solutions of |z − i| + |z + i| = 1.

194
Solution 8.13

| eiθ − 1| = 2
( eiθ − 1)( e−iθ − 1) = 4
1 − eiθ − e−iθ + 1 = 4
−2 cos(θ) = 2
θ=π

{ eiθ | 0 ≤ θ ≤ π} is a unit semi-circle in the upper half of the complex plane from 1 to -1. The only point on
this semi-circle that is a distance 2 from the point 1 is the point -1, which corresponds to θ = π.

Polar Form
Solution 8.14
The Taylor series expansion of ex is
∞
x
X xn
e = .
n=0
n!

Taking this as the definition of the exponential function for complex argument,
∞
iθ
X (iθ)n
e =
n=0
n!
∞
X θn
= in
n=0
n!
∞ ∞
X θ2n X θ2n+1
= (−1)n +i (−1)n .
n=0
(2n)! n=0
(2n + 1)!

195
The sine and cosine have the Taylor series

∞ ∞
X θ2n X θ2n+1
cos θ = (−1)n , sin θ = (−1)n ,
n=0
(2n)! n=0
(2n + 1)!

Thus eiθ and cos θ + i sin θ have the same Taylor series expansions about θ = 0. Since the radius of convergence
of the series is infinite we conclude that,

eiθ = cos θ + i sin θ.

Solution 8.15

cos(3θ) + i sin(3θ) = (cos(θ) + i sin(θ))3

cos(3θ) + i sin(3θ) = cos3 (θ) + i3 cos2 (θ) sin(θ) − 3 cos(θ) sin2 (θ) − i sin3 (θ)

We equate the real parts of the equation.

cos(3θ) = cos3 (θ) − 3 cos(θ) sin2 (θ)

Solution 8.16
Define the partial sum,

n
X
Sn (z) = zk .
k=0

196
Now consider (1 − z)Sn (z).

n
X
(1 − z)Sn (z) = (1 − z) zk
k=0
n
X n+1
X
(1 − z)Sn (z) = zk − zk
k=0 k=1
n+1
(1 − z)Sn (z) = 1 − z

We divide by 1 − z. Note that 1 − z is nonzero.

1 − z n+1
Sn (z) =
1−z
1 − z n+1
1 + z + z2 + · · · + zn = , (z 6= 1)
1−z

Now consider z = eiθ where 0 < θ < 2π so that z is not unity.

n
n+1
X
iθ k

1 − eiθ
e =
k=0
1 − eiθ
n
X 1 − ei(n+1)θ
eikθ =
k=0
1 − eiθ

197
In order to get sin(θ/2) in the denominator, we multiply top and bottom by e−iθ/2 .
n
X e−iθ/2 − ei(n+1/2)θ
(cos kθ + i sin kθ) =
k=0
e−iθ/2 − eiθ/2
n n
X X cos(θ/2) − i sin(θ/2) − cos((n + 1/2)θ) − i sin((n + 1/2)θ)
cos kθ + i sin kθ =
k=0 k=0
−2i sin(θ/2)
n n  
X X 1 sin((n + 1/2)θ) 1 cos((n + 1/2)θ)
cos kθ + i sin kθ = + +i cot(θ/2) −
k=0 k=1
2 sin(θ/2) 2 sin(θ/2)
1. We take the real and imaginary part of this to obtain the identities.
n
X 1 sin((n + 1/2)θ)
cos kθ = +
k=0
2 2 sin(θ/2)

2.
n
X 1 cos((n + 1/2)θ)
sin kθ = cot(θ/2) −
k=1
2 2 sin(θ/2)

Arithmetic and Vectors

Solution 8.17

|z1 z2 | = |r1 eiθ1 r2 eiθ2 |

= |r1 r2 ei(θ1 +θ2 ) |
= |r1 r2 |
= |r1 ||r2 |
= |z1 ||z2 |

198
  z1   r1 eiθ1 
   
 = 
 z2   r2 eiθ2 
 
 r1 i(θ −θ ) 
=  e 1 2 
r
 2
 r1 
=  
r2
|r1 |
=
|r2 |
|z1 |
=
|z2 |

Solution 8.18

|z + ζ|2 + |z − ζ|2 = (z + ζ) z + ζ + (z − ζ) z − ζ

= zz + zζ + ζz + ζζ + zz − zζ − ζz + ζζ
= 2 |z|2 + |ζ|2

Consider the parallelogram defined by the vectors z and ζ. The lengths of the sides are z and ζ and the lengths of
the diagonals are z + ζ and z − ζ. We know from geometry that the sum of the squared lengths of the diagonals
of a parallelogram is equal to the sum of the squared lengths of the four sides. (See Figure 8.18.)

Integer Exponents

199
 z- ζ
z
z+ζ

Figure 8.18: The parallelogram defined by z and ζ.

Solution 8.19
1.

10

2
2 2
(1 + i) = (1 + i) (1 + i)2

2
= (i2)2 (i2)
= (−4)2 (i2)
= 16(i2)
= i32

2.
√ 10
(1 + i)10 = 2 eiπ/4
√ 10
= 2 ei10π/4
= 32 eiπ/2
= i32

200
Rational Exponents
Solution 8.20
We substitite the numbers into the equation to obtain an identity.

z 2 + 2az + b = 0
(−a + (a2 − b)1/2 )2 + 2a(−a + (a2 − b)1/2 ) + b = 0
a2 − 2a(a2 − b)1/2 + a2 − b − 2a2 + 2a(a2 − b)1/2 + b = 0
0=0

201
Chapter 9

Functions of a Complex Variable

If brute force isn’t working, youŕe not using enough of it.

-Tim Mauch

In this chapter we introduce the algebra of functions of a complex variable. We will cover the trigonometric and
inverse trigonometric functions. The properties of trigonometric function carry over directly from real-variable
theory. However, because of multi-valuedness, the inverse trigonometric functions are significantly trickier than
their real-variable counterparts.

9.1 Curves and Regions

In this section we introduce curves and regions in the complex plane. This material is necessary for the study
of branch points in this chapter and later for contour integration.

202
Curves. Consider two continuous functions, x(t) and y(t), defined on the interval t ∈ [t0 . . . t1 ]. The set of
points in the complex plane

{z(t) = x(t) + iy(t) | t ∈ [t0 . . . t1 ]}

defines a continuous curve or simply a curve. If the endpoints coincide, z(t0 ) = z(t1 ), it is a closed curve. (We
assume that t0 6= t1 .) If the curve does not intersect itself, then it is said to be a simple curve.
If x(t) and y(t) have continuous derivatives and the derivatives do not both vanish at any point 1 , then it is
a smooth curve. This essentially means that the curve does not have any corners or other nastiness.
A continuous curve which is composed of a finite number of smooth curves is called a piecewise smooth curve.
We will use the word contour as a synonym for a piecewise smooth curve.
See Figure 9.1 for a smooth curve, a piecewise smooth curve, a simple closed curve and a non-simple closed
curve.

(a) (b) (c) (d)

Figure 9.1: (a) Smooth Curve, (b) Piecewise Smooth Curve, (c) Simple Closed Curve, (d) Non-Simple Closed
Curve

Regions. A region R is connected if any two points in R can be connected by a curve which lies entirely in
R. A region is simply-connected if every closed curve in R can be continuously shrunk to a point without leaving
R. A region which is not simply-connected is said to be multiply-connected region. Another way of defining
1
Why is it necessary that the derivatives do not both vanish?

203
simply-connected is that a path connecting two points in R can be continuously deformed into any other path
that connects those points. Figure 9.2 shows a simply-connected region with two paths which can be continuously
deformed into one another and a multiply-connected region with paths which cannot be deformed into one another.

Figure 9.2: Simply-connected and multiply-connected regions.

Jordan Curve Theorem. A continuous, simple, closed curve is known as a Jordan curve. The Jordan Curve
Theorem, which seems intuitively obvious but is difficult to prove, states that a Jordan curve divides the plane
into a simply-connected, bounded region and an unbounded region. These two regions are called the interior and
exterior regions, respectively. The two regions share the curve as a boundary. Points in the interior are said to
be inside the curve; points in the exterior are said to be outside the curve.

Traversal of a Contour. Consider a Jordan curve. If you traverse the curve in the positive direction, then the
inside is to your left. If you traverse the curve in the opposite direction, then the outside will be to your left and
you will go around the curve in the negative direction. For circles, the positive direction is the counter-clockwise
direction. The positive direction is consistent with the way angles are measured in a right-handed coordinate
system, i.e. for a circle centered on the origin, the positive direction is the direction of increasing angle. For an
oriented contour C, we denote the contour with opposite orientation as −C.

204
Boundary of a Region. Consider a simply-connected region. The boundary of the region is traversed in the
positive direction if the region is to the left as you walk along the contour. For multiply-connected regions, the
boundary may be a set of contours. In this case the boundary is traversed in the positive direction if each of the
contours is traversed in the positive direction. When we refer to the boundary of a region we will assume it is
given the positive orientation. In Figure 9.3 the boundaries of three regions are traversed in the positive direction.

Figure 9.3: Traversing the boundary in the positive direction.

Two Interpretations of a Curve. Consider a simple closed curve as depicted in Figure 9.4a. By giving it
an orientation, we can make a contour that either encloses the bounded domain Figure 9.4b or the unbounded
domain Figure 9.4c. Thus a curve has two interpretations. It can be thought of as enclosing either the points
which are “inside” or the points which are “outside”. 2

2
A farmer wanted to know the most efficient way to build a pen to enclose his sheep, so he consulted an engineer, a physicist
and a mathematician. The engineer suggested that he build a circular pen to get the maximum area for any given perimeter. The
physicist suggested that he build a fence at infinity and then shrink it to fit the sheep. The mathematician constructed a little fence
around himself and then defined himself to be outside.

205
 (a) (b) (c)

Figure 9.4: Two interpretations of a curve.

9.2 Cartesian and Modulus-Argument Form

We can write a function of a complex variable z as a function of x and y or as a function of r and θ with the
substitutions z = x + iy and z = r eiθ , respectively. Then we can separate the real and imaginary components or
write the function in modulus-argument form,

f (z) = u(x, y) + iv(x, y), or f (z) = u(r, θ) + iv(r, θ),

f (z) = ρ(x, y) eiφ(x,y) , or f (z) = ρ(r, θ) eiφ(r,θ) .

1
Example 9.2.1 Consider the functions f (z) = z, f (z) = z 3 and f (z) = 1−z
. We write the functions in terms of
x and y and separate them into their real and imaginary components.

f (z) = z
= x + iy

206
 f (z) = z 3
= (x + iy)3
= x3 + ix2 y − xy 2 − iy 3
= (x3 − xy 2 ) + i(x2 y − y 3 )

1
f (z) =
1−z
1
=
1 − x − iy
1 1 − x + iy
=
1 − x − iy 1 − x + iy
1−x y
= + i
(1 − x)2 + y 2 (1 − x)2 + y 2

1
Example 9.2.2 Consider the functions f (z) = z, f (z) = z 3 and f (z) = 1−z
. We write the functions in terms of
r and θ and write them in modulus-argument form.

f (z) = z
= r eiθ

f (z) = z 3

3
= r eiθ
= r3 ei3θ

207
 1
f (z) =
1−z
1
=
1 − r eiθ
1 1
=
1 − r eiθ 1 − r e−iθ
1 − r e−iθ
=
1 − r eiθ − r e−iθ + r2
1 − r cos θ + ir sin θ
=
1 − 2r cos θ + r2

Note that the denominator is real and non-negative.

1
= |1 − r cos θ + ir sin θ| ei arctan(1−r cos θ,r sin θ)
1 − 2r cos θ + r2 q
1
= (1 − r cos θ)2 + r2 sin2 θ ei arctan(1−r cos θ,r sin θ)
1 − 2r cos θ + r2
1 p
= 2
1 − 2r cos θ + r2 cos2 θ + r2 sin2 θ ei arctan(1−r cos θ,r sin θ)
1 − 2r cos θ + r
1
=√ ei arctan(1−r cos θ,r sin θ)
1 − 2r cos θ + r2

9.3 Graphing Functions of a Complex Variable

We cannot directly graph a function of a complex variable as they are mappings from R2 to R2 . To do so would
require four dimensions. However, we can can use a surface plot to graph the real part, the imaginary part, the
modulus or the argument of a function of a complex variable. Each of these are scalar fields, mappings from R2
to R.

208
Example 9.3.1 Consider the identity function, f (z) = z. In Cartesian coordinates and Cartesian form, the
function is f (z) = x + iy. The real and imaginary components are u(x, y) = x and v(x, y) = y. (See Figure 9.5.)
In modulus argument form the function is

2 2
1 2 1 2
0 0
-1 1 -1 1
-2 -2
-2 0 y -2 0 y
-1 -1
-1 -1
x0 1 x0 1
2-2 2-2

Figure 9.5: The real and imaginary parts of f (z) = z = x + iy

p
f (z) = z = r eiθ = x2 + y 2 ei arctan(x,y) .
The modulus of f (z) is a single-valued function which is the distance from the origin. The argument of f (z) is a
multi-valued function. Recall that arctan(x, y) has an infinite number of values each of which differ by an integer
multiple of 2π. A few branches of arg(f (z)) are plotted in Figure 9.6. The modulus and principal argument of
f (z) = z are plotted in Figure 9.7.

Example 9.3.2 Consider the function f (z) = z 2 . In Cartesian coordinates and separated into its real and
imaginary components the function is
f (z) = z 2 = (x + iy)2 = (x2 − y 2 ) + i2xy.
Figure 9.8 shows surface plots of the real and imaginary parts of z 2 . The magnitude of z 2 is
p
|z 2 | = z 2 z 2 = zz = (x + iy)(x − iy) = x2 + y 2 .

209
 y 12
0
-1
-2
5
0
-5
-2
-1
0
x 1
2

Figure 9.6: A Few Branches of arg(z)

2 2 2
0 2
1 1 -2 1
0
-2 0 y -2 0 y
-1 -1
-1 -1
x0 1 x0 1
2-2 2 -2

Figure 9.7: Plots of |z| and Arg (z)

Note that
z 2 = (r eiθ )2 = r2 ei2θ .

210
 4
2 5
2 2
0 0
-2 1 -5 1
-4
-2 0 y -2 0 y
-1 -1
0 -1 0 -1
x 1 x 1
2 -2 2 -2

Figure 9.8: Plots of <(z 2 ) and =(z 2 )

In Figure 9.9 are plots of |z 2 | and a branch of arg(z 2 ).

8 5
6 2 2
4 0
2 1 1
0 -5
-2 0 y -2 0 y
-1 -1
0 -1 0 -1
x 1 x 1
2 -2 2 -2

Figure 9.9: Plots of |z 2 | and a branch of arg(z 2 )

211
9.4 Trigonometric Functions
The Exponential Function. Consider the exponential function ez . We can use Euler’s formula to write
ez = ex+iy in terms of its real and imaginary parts.
ez = ex+iy = ex eiy = ex cos y + i ex sin y
From this we see that the exponential function is i2π periodic: ez+i2π = ez , and iπ odd periodic: ez+iπ = − ez .
Figure 9.10 has surface plots of the real and imaginary parts of ez which show this periodicity.

20 20
10 10
0 5 0 5
-10 -10
-20 -20
0 y 0 y
-2 -2
0 0
x -5 x -5
2 2

Figure 9.10: Plots of <( ez ) and =( ez )

The modulus of ez is a function of x alone.

 
| ez | =  ex+iy  = ex
The argument of ez is a function of y alone.
arg ( ez ) = arg ex+iy = {y + 2πn | n ∈ Z}

In Figure 9.11 are plots of | ez | and a branch of arg( ez ).

212
 20 5
15 5 0 5
10
5 -5
0
0 y 0 y
-2 -2
x0 -5 x0 -5
2 2

Figure 9.11: Plots of | ez | and a branch of arg( ez )

Example 9.4.1 Show that the transformation w = ez maps the infinite strip, −∞ < x < ∞, 0 < y < π, onto
the upper half-plane.
Method 1. Consider the line z = x + ic, −∞ < x < ∞. Under the transformation, this is mapped to

w = ex+ic = eic ex , −∞ < x < ∞.

This is a ray from the origin to infinity in the direction of eic . Thus we see that z = x is mapped to the positive,
real w axis, z = x + iπ is mapped to the negative, real axis, and z = x + ic, 0 < c < π is mapped to a ray with
angle c in the upper half-plane. Thus the strip is mapped to the upper half-plane. See Figure 9.12.
Method 2. Consider the line z = c + iy, 0 < y < π. Under the transformation, this is mapped to

w = ec+iy + ec eiy , 0 < y < π.

This is a semi-circle in the upper half-plane of radius ec . As c → −∞, the radius goes to zero. As c → ∞, the
radius goes to infinity. Thus the strip is mapped to the upper half-plane. See Figure 9.13.

213
 3 3
2 2
1 1

-3 -2 -1 1 2 3 -3 -2 -1 1 2 3

Figure 9.12: ez maps horizontal lines to rays.

3 3
2 2
1 1

-1 1 -3 -2 -1 1 2 3

Figure 9.13: ez maps vertical lines to circular arcs.

The Sine and Cosine. We can write the sine and cosine in terms of the exponential function.

eiz + e−iz cos(z) + i sin(z) + cos(−z) + i sin(−z)

=
2 2
cos(z) + i sin(z) + cos(z) − i sin(z)
=
2
= cos z

214
 eiz − e−iz cos(z) + i sin(z) − cos(−z) − i sin(−z)
=
i2 2
cos(z) + i sin(z) − cos(z) + i sin(z)
=
2
= sin z
We separate the sine and cosine into their real and imaginary parts.
cos z = cos x cosh y − i sin x sinh y sin z = sin x cosh y + i cos x sinh y
For fixed y, the sine and cosine are oscillatory in x. The amplitude of the oscillations grows with increasing |y|.
See Figure 9.14 and Figure 9.15 for plots of the real and imaginary parts of the cosine and sine, respectively.
Figure 9.16 shows the modulus of the cosine and the sine.

5 5
2.5 2.5
0 2 0 2
-2.5 1 -2.5 1
-5 -5
0 y 0 y
-2 -2
0 -1 0 -1
x x
2 -2 2 -2

Figure 9.14: Plots of <(cos(z)) and =(cos(z))

The Hyperbolic Sine and Cosine. The hyperbolic sine and cosine have the familiar definitions in terms of
the exponential function. Thus not surprisingly, we can write the sine in terms of the hyperbolic sine and write
the cosine in terms of the hyperbolic cosine. Below is a collection of trigonometric identities.

215
 5 5
2.5 2.5
0 2 0 2
-2.5 1 -2.5 1
-5 -5
0 y 0 y
-2 -2
0 -1 0 -1
x x
2 -2 2 -2

Figure 9.15: Plots of <(sin(z)) and =(sin(z))

4 4
2 2
2 2
1 1
0
0 y 0 y
-2 -2
0 -1 0 -1
x x
2 -2 2 -2

Figure 9.16: Plots of | cos(z)| and | sin(z)|

Result 9.4.1

ez = ex (cos y + i sin y)
eiz + e−iz eiz − e−iz
cos z = 216sin z =
2 2i
cos z = cos x cosh y − i sin x sinh y sin z = sin x cosh y + i cos x sinh y
ez + e−z ez − e−z
cosh z = sinh z =
2 2
cosh z = cosh x cos y + i sinh x sin y sinh z = sinh x cos y + i cosh x sin y
sin iz = i sinh z sinh iz = i sin z
cos iz = cosh z cosh iz = cos z
log z = Log |z| + i arg(z) = Log |z| + i Arg (z) + 2iπn
9.5 Inverse Trigonometric Functions
The Logarithm. The logarithm, log(z), is defined as the inverse of the exponential function ez . The exponential
function is many-to-one and thus has a multi-valued inverse. From what we know of many-to-one functions, we
conclude that
elog z = z, but log( ez ) 6= z.
This is because elog z is single-valued but log( ez ) is not. Because ez is i2π periodic, the logarithm of a number is
a set of numbers which differ by integer multiples of i2π. For instance, ei2πn = 1 so that log(1) = {i2πn : n ∈ Z}.
The logarithmic function has an infinite number of branches. The value of the function on the branches differs
by integer multiples of i2π. It has singularities at zero and infinity. | log(z)| → ∞ as either z → 0 or z → ∞.
We will derive the formula for the complex variable logarithm. For now, let Log (x) denote the real variable
logarithm that is defined for positive real numbers. Consider w = log z. This means that ew = z. We write
w = u + iv in Cartesian form and z = r eiθ in polar form.
eu+iv = r eiθ
We equate the modulus and argument of this expression.
eu = r v = θ + 2πn
u = Log r v = θ + 2πn
With log z = u + iv, we have a formula form the logarithm.

log z = Log |z| + i arg(z)

If we write out the multi-valuedness of the argument function we note that this has the form that we expected.
log z = Log |z| + i( Arg (z) + 2πn), n∈Z
We check that our formula is correct by showing that elog z = z
elog z = e Log |z|+i arg(z) = e Log r+iθ+i2πn = r eiθ = z

217
Note again that log( ez ) 6= z.

log( ez ) = Log | ez | + i arg( ez ) = Log ( ex ) + i arg( ex+iy ) = x + i(y + 2πn) = z + i2nπ 6= z

The real part of the logarithm is the single-valued Log r; the imaginary part is the multi-valued arg(z). We
define the principal branch of the logarithm Log z to be the branch that satisfies −π < =( Log z) ≤ π. For
positive, real numbers the principal branch, Log x is real-valued. We can write Log z in terms of the principal
argument, Arg z.

Log z = Log |z| + i Arg (z)

See Figure 9.17 for plots of the real and imaginary part of Log z.

1
2
0 2 2
0
-1 1 1
-2
-2
-2 0 y -2 0 y
-1 -1
0 -1 0 -1
x 1 x 1
2 -2 2 -2

Figure 9.17: Plots of <( Log z) and =( Log z).

The Form: ab . Consider ab where a and b are complex and a is nonzero. We define this expression in terms of
the exponential and the logarithm as

ab = eb log a .

218
Note that the multi-valuedness of the logarithm may make ab multi-valued. First consider the case that the
exponent is an integer.
am = em log a = em( Log a+i2nπ) = em Log a ei2mnπ = em Log a
Thus we see that am has a single value where m is an integer.
Now consider the case that the exponent is a rational number. Let p/q be a rational number in reduced form.
p p p
ap/q = e q log a = e q ( Log a+i2nπ) = e q Log a ei2npπ/q .
This expression has q distinct values as
ei2npπ/q = ei2mpπ/q if and only if n = m mod q.
Finally consider the case that the exponent b is an irrational number.
ab = eb log a = eb( Log a+i2nπ) = eb Log a ei2bnπ
Note that ei2bnπ and ei2bmπ are equal if and only if i2bnπ and i2bmπ differ by an integer multiple of i2π, which
means that bn and bm differ by an integer. This occurs only when n = m. Thus ei2bnπ has a distinct value for
each different integer n. We conclude that ab has an infinite number of values.
You may have noticed something a little fishy. If b is not an integer and a is any non-zero complex number,
then ab is multi-valued. Then why have we been treating eb as single-valued, when it is merely the case a = e?
The answer is that in the realm of functions of a complex variable, ez is an abuse of notation. We write ez when
we mean exp(z), the single-valued exponential function. Thus when we write ez we do not mean “the number e
raised to the z power”, we mean “the exponential function of z”. We denote the former scenario as (e)z , which
is multi-valued.

Logarithmic Identities. Back in high school trigonometry when you thought that the logarithm was only
defined for positive real numbers you learned the identity log xa = a log x. This identity doesn’t hold when the
logarithm is defined for nonzero complex numbers. Consider the logarithm of z a .
log z a = Log z a + i2πn

219
 a log z = a( Log z + i2πn) = a Log z + i2aπn

Note that

log z a 6= a log z

Furthermore, since

Log z a = Log |z a | + i Arg (z a ), a Log z = a Log |z| + ia Arg (z)

and Arg (z a ) is not necessarily the same as a Arg (z) we see that

Log z a 6= a Log z.

Consider the logarithm of a product.

log(ab) = Log |ab| + i arg(ab)

= Log |a| + Log |b| + i arg(a) + i arg(b)
= log a + log b

There is not an analogous identity for the principal branch of the logarithm since Arg (ab) is not in general the
same as Arg (a) + Arg (b).
Pn
Using log(ab) = log(a) + log(b) we can deduce that log(an ) = k=1 log a = n log a, where n is a positive
integer. This result is simple, straightforward and wrong. I have led you down the merry path to damnation. 3
In fact, log(a2 ) 6= 2 log a. Just write the multi-valuedness explicitly,

log(a2 ) = Log (a2 ) + i2nπ, 2 log a = 2( Log a + i2nπ) = 2 Log a + i4nπ.

3
Don’t feel bad if you fell for it. The logarithm is a tricky bastard.

220
 You can verify that

 
1
log = − log a.
a

We can use this and the product identity to expand the logarithm of a quotient.

a
log = log a − log b
b

For general values of a, log z a 6= a log z. However, for some values of a, equality holds. We already know
that a = 1 and a = −1 work. To determine if equality holds for other values of a, we explicitly write the
multi-valuedness.

log z a = log ea log z = a log z + i2πk,

k∈Z
a log z = a Log |z| + ia Arg z + ia2πm, m∈Z

We see that log z a = a log z if and only if

{am | m ∈ Z} = {am + k | k, m ∈ Z}.

The sets are equal if and only if a = 1/n, n ∈ Z± . Thus we have the identity:

1
log z 1/n = log z, n ∈ Z±
n

221
 Result 9.5.1 Logarithmic Identities.

ab = eb log a
elog z = e Log z = z
log(ab) = log a + log b
log(1/a) = − log a
log(a/b) = log a − log b
  1
log z 1/n
= log z, n ∈ Z±
n
Logarithmic Inequalities.

Log (uv) 6= Log (u) + Log (v)

log z a 6= a log z
Log z a 6= a Log z
log ez 6= z

Example 9.5.1 Consider 1π . We apply the definition ab = eb log a .

1π = eπ log(1)
= eπ( Log (1)+i2nπ)
2
= ei2nπ

Thus we see that 1π has an infinite number of values, all of which lie on the unit circle |z| = 1 in the complex
plane. However, the set 1π is not equal to the set |z| = 1. There are points in the latter which are not in the

222
former. This is analogous to the fact that the rational numbers are dense in the real numbers, but are a subset
of the real numbers.

Example 9.5.2 We find the zeros of sin z.

eiz − e−iz
sin z = =0
2i
eiz = e−iz
ei2z = 1
2z mod 2π = 0
z = nπ, n∈Z

Equivalently, we could use the identity

sin z = sin x cosh y + i cos x sinh y = 0.

This becomes the two equations (for the real and imaginary parts)

sin x cosh y = 0 and cos x sinh y = 0.

Since cosh is real-valued and positive for real argument, the first equation dictates that x = nπ, n ∈ Z. Since
cos(nπ) = (−1)n for n ∈ Z, the second equation implies that sinh y = 0. For real argument, sinh y is only zero at
y = 0. Thus the zeros are

z = nπ, n∈Z

Example 9.5.3 Since we can express sin z in terms of the exponential function, one would expect that we could

223
express the sin−1 z in terms of the logarithm.
w = sin−1 z
z = sin w
eiw − e−iw
z=
2i
e2iw − 2iz eiw − 1 = 0
√
eiw = iz ± 1 − z 2
 √ 
w = −i log iz ± 1 − z 2

Thus we see how the multi-valued sin−1 is related to the logarithm.

−1
 √ 
sin z = −i log iz ± 1 − z 2

Example 9.5.4 Consider the equation sin3 z = 1.

sin3 z = 1
sin z = 11/3
eiz − e−iz
= 11/3
2i
eiz − 2i(1)1/3 − e−iz = 0
e2iz − 2i(1)1/3 eiz − 1 = 0
p
iz 2i(1)1/3 ± −4(1)2/3 + 4
e =
q2
eiz = i(1)1/3 ± 1 − (1)2/3
 p 
z = −i log i(1)1/3 ± 1 − 12/3

224
Note that there are three sources of multi-valuedness in the expression for z. The two values of the square root are
shown explicitly. There are three cube roots of unity. Finally, the logarithm has an infinite number of branches.
To show this multi-valuedness explicitly, we could write
 p 
i2mπ/3 i4mπ/3
z = −i Log i e ± 1− e + 2πn, m = 0, 1, 2, n = . . . , −1, 0, 1, . . .

Example 9.5.5 Consider the harmless looking equation, iz = 1.

Before we start with the algebra, note that the right side of the equation is a single number. iz is single-valued
only when z is an integer. Thus we know that if there are solutions for z, they are integers. We now proceed to
solve the equation.
iz = 1
( eiπ/2 )z = 1
Use the fact that z is an integer.
eiπz/2 = 1
iπz/2 = 2inπ, for some n ∈ Z
z = 4n, n∈Z
Here is a different approach. We write down the multi-valued form of iz . We solve the equation by requiring
that all the values of iz are 1.
iz = 1
ez log i = 1
z log i = 2πin, for some n ∈ Z
 π 
z i + 2πim = 2πin, ∀m ∈ Z, for some n ∈ Z
2
π
i z + 2πimz = 2πin, ∀m ∈ Z, for some n ∈ Z
2

225
The only solutions that satisfy the above equation are

z = 4k, k ∈ Z.

Now let’s consider a slightly different problem: 1 ∈ iz . For what values of z does iz have 1 as one of its values.

1 ∈ iz
1 ∈ ez log i
1 ∈ { ez(iπ/2+i2πn) | n ∈ Z}
z(iπ/2 + i2πn) = i2πm, m, n ∈ Z
4m
z= , m, n ∈ Z
1 + 4n

There are an infinite set of rational numbers for which iz has 1 as one of its values. For example,

i4/5 = 11/5 = 1, ei2π/5 , ei4π/5 , ei6π/5 , ei8π/5



9.6 Branch Points

Example 9.6.1 Consider the function z 1/2 . For each value of z, there are two values of z 1/2 . We write z 1/2 in
modulus-argument and Cartesian form.
p
z 1/2 = |z| ei arg(z)/2
p p
z 1/2 = |z| cos(arg(z)/2) + i |z| sin(arg(z)/2)

Figures 9.18 and 9.19 show the real and imaginary parts of z 1/2 from three different viewpoints. The second
and third views are looking down the x axis and y axis, respectively. Consider <(z 1/2 ). This is a double layered
sheet which intersects itself on the negative real axis. (=(z 1/2 ) has a similar structure, but intersects itself on the

226
positive real axis.) Let’s start at a point on the positive real axis on the lower sheet. If we walk around the origin
once and return to the positive real axis, we will be on the upper sheet. If we do this again, we will return to the
lower sheet.
Suppose we are at a point in the complex plane. We pick one of the two values of z 1/2 . If the function varies
continuously as we walk around the origin and back to our starting point, the value of z 1/2 will have changed.
We will be on the other branch. Because walking around the point z = 0 takes us to a different branch of the
function, we refer to z = 0 as a branch point.

0121
-1
-2 -1
-20121
x y
1 0 0
0 2
-1 1 -1 -1
-2 0 y -2 -1 0 1 2 2 1 0 -1 -2
-1 y x
0 -1
x 1
2 -2

Figure 9.18: Plots of <(z 1/2 ) from three viewpoints.

Now consider the modulus-argument form of z 1/2 :

p
z 1/2 = |z| ei arg(z)/2 .
Figure 9.20 shows the modulus and the principal argument of z 1/2 . We see that each time we walk around the
origin, the argument of z 1/2 changes by π. This means that the value of the function changes by the factor
eiπ = −1, i.e. the function changes sign. If we walk around the origin twice, the argument changes by 2π, so that
the value of the function does not change, ei2π = 1.
z 1/2 is a continuous function except at z = 0. Suppose we start at z = 1 = ei0 and the function value
( e ) = 1. If we follow the first path in Figure 9.21, the argument of z varies from up to about π4 , down to
i0 1/2

about − π4 and back to 0. The value of the function is still ( ei0 )1/2 .

227
 0121
-1
-2 -1
-20121
x y
1 0 0
0 2
-1 1 -1 -1
-2 0 y -2 -1 0 1 2 2 1 0 -1 -2
-1 y x
0 -1
x 1
2 -2

Figure 9.19: Plots of =(z 1/2 ) from three viewpoints.

1 2
2 0 2
0.5 1 -2 1
0
-2 0 y -2 0 y
-1 -1 -1
x0 -1
1 x0 1
2 -2 2 -2

Figure 9.20: Plots of |z 1/2 | and Arg (z 1/2 ).

Now suppose we follow a circular path around the origin in the positive, counter-clockwise, direction. (See the
second path in Figure 9.21.) The argument of z increases by 2π. The value of the function at half turns on the

228
 Im(z) Im(z)

Re(z) Re(z)

Figure 9.21: A path that does not encircle the origin and a path around the origin

path is

( ei0 )1/2 = 1,
( eiπ )1/2 = eiπ/2 = i,
( ei2π )1/2 = eiπ = −1

As we return to the point z = 1, the argument of the function has changed by π and the value of the function
has changed from 1 to −1. If we were to walk along the circular path again, the argument of z would increase
by another 2π. The argument of the function would increase by another π and the value of the function would
return to 1.

( e4πi )1/2 = e2πi = 1

In general, any time we walk around the origin, the value of z 1/2 changes by the factor −1. We call z = 0 a
branch point. If we want a single-valued square root, we need something to prevent us from walking around the
origin. We achieve this by introducing a branch cut. Suppose we have the complex plane drawn on an infinite
sheet of paper. With a scissors we cut the paper from the origin to −∞ along the real axis. Then if we start

229
at z = ei0 , and draw a continuous line without leaving the paper, the argument of z will always be in the range
−π < arg z < π. This means that − π2 < arg(z 1/2 ) < π2 . No matter what path we follow in this cut plane, z = 1
has argument zero and (1)1/2 = 1. By never crossing the negative real axis, we have constructed a single valued
branch of the square root function. We call the cut along the negative real axis a branch cut.

Example 9.6.2 Consider the logarithmic function log z. For each value of z, there are an infinite number of
values of log z. We write log z in Cartesian form.
log z = Log |z| + i arg z
Figure 9.22 shows the real and imaginary parts of the logarithm. The real part is single-valued. The imaginary
part is multi-valued and has an infinite number of branches. The values of the logarithm form an infinite-layered
sheet. If we start on one of the sheets and walk around the origin once in the positive direction, then the value
of the logarithm increases by i2π and we move to the next branch. z = 0 is a branch point of the logarithm.

1
0 2 5 2
-1 0
1 -5 1
-2
-2 0 y -2 0 y
-1 -1
-1 -1
x0 1 x0 1
2-2 2 -2

Figure 9.22: Plots of <(log z) and a portion of =(log z).

The logarithm is a continuous function except at z = 0. Suppose we start at z = 1 = ei0 and the function
value log( ei0 ) = Log (1) + i0 = 0. If we follow the first path in Figure 9.21, the argument of z and thus the

230
imaginary part of the logarithm varies from up to about π4 , down to about − π4 and back to 0. The value of the
logarithm is still 0.
Now suppose we follow a circular path around the origin in the positive direction. (See the second path in
Figure 9.21.) The argument of z increases by 2π. The value of the logarithm at half turns on the path is
log( ei0 ) = 0,
log( eiπ ) = iπ,
log( ei2π ) = i2π
As we return to the point z = 1, the value of the logarithm has changed by i2π. If we were to walk along the
circular path again, the argument of z would increase by another 2π and the value of the logarithm would increase
by another i2π.

Result 9.6.1 A point z0 is a branch point of a function f (z) if the function changes
value when you walk around the point on any path that encloses no singularities other
than the one at z = z0 .

Branch Points at Infinity : Mapping Infinity to the Origin. Up to this point we have considered only
branch points in the finite plane. Now we consider the possibility of a branch point at infinity. As a first method
of approaching this problem we map the point at infinity to the origin with the transformation ζ = 1/z and
examine the point ζ = 0.

Example 9.6.3 Again consider the function z 1/2 . Mapping the point at infinity to the origin, we have f (ζ) =
(1/ζ)1/2 = ζ −1/2 . For each value of ζ, there are two values of ζ −1/2 . We write ζ −1/2 in modulus-argument form.
1
ζ −1/2 = p e−i arg(ζ)/2
|ζ|
Like z 1/2 , ζ −1/2 has a double-layered sheet of values. Figure 9.23 shows the modulus and the principal argument
of ζ −1/2 . We see that each time we walk around the origin, the argument of ζ −1/2 changes by −π. This means

231
that the value of the function changes by the factor e−iπ = −1, i.e. the function changes sign. If we walk around
the origin twice, the argument changes by −2π, so that the value of the function does not change, e−i2π = 1.

3 2
2.5 2 2
2 0
1.5 1 -2 1
1
-2 0y -2 0 y
-1 -1 -1
x0 1 x0
-1
2 -2 1
2 -2

Figure 9.23: Plots of |ζ −1/2 | and Arg (ζ −1/2 ).

Since ζ −1/2 has a branch point at zero, we conclude that z 1/2 has a branch point at infinity.

Example 9.6.4 Again consider the logarithmic function log z. Mapping the point at infinity to the origin, we
have f (ζ) = log(1/ζ) = − log(ζ). From Example 9.6.2 we known that − log(ζ) has a branch point at ζ = 0. Thus
log z has a branch point at infinity.

Branch Points at Infinity : Paths Around Infinity. We can also check for a branch point at infinity by
following a path that encloses the point at infinity and no other singularities. Just draw a simple closed curve
that separates the complex plane into a bounded component that contains all the singularities of the function in
the finite plane. Then, depending on orientation, the curve is a contour enclosing all the finite singularities, or
the point at infinity and no other singularities.

232
Example 9.6.5 Once again consider the function z 1/2 . We know that the function changes value on a curve that
goes once around the origin. Such a curve can be considered to be either a path around the origin or a path
around infinity. In either case the path encloses one singularity. There are branch points at the origin and at
infinity. Now consider a curve that does not go around the origin. Such a curve can be considered to be either a
path around neither of the branch points or both of them. Thus we see that z 1/2 does not change value when we
follow a path that encloses neither or both of its branch points.

Example 9.6.6 Consider f (z) = (z 2 − 1)1/2 . We factor the function.

f (z) = (z − 1)1/2 (z + 1)1/2

There are branch points at z = ±1. Now consider the point at infinity.

f (ζ −1 ) = (ζ −2 − 1)1/2 = ±ζ −1 (1 − ζ 2 )1/2

Since f (ζ −1 ) does not have a branch point at ζ = 0, f (z) does not have a branch point at infinity. We could
reach the same conclusion by considering a path around infinity. Consider a path that circles the branch points
at z = ±1 once in the positive direction. Such a path circles the point at infinity once in the negative direction.
In traversing this path, the value of f (z) is multiplied by the factor ( ei2π )1/2 ( ei2π )1/2 = ei2π = 1. Thus the value
of the function does not change. There is no branch point at infinity.

Diagnosing Branch Points. We have the definition of a branch point, but we do not have a convenient
criterion for determining if a particular function has a branch point. We have seen that log z and z α for non-
integer α have branch points at zero and infinity. The inverse trigonometric functions like the arcsine also have
branch points, but they can be written in terms of the logarithm and the square root. In fact all the elementary
functions with branch points can be written in terms of the functions log z and z α . Furthermore, note that the
multi-valuedness of z α comes from the logarithm, z α = eα log z . This gives us a way of quickly determining if and
where a function may have branch points.

233
 Result 9.6.2 Let f (z) be a single-valued function. Then log(f (z)) and (f (z))α may
have branch points only where f (z) is zero or singular.

Example 9.6.7 Consider the functions,

1. (z 2 )1/2
2. (z 1/2 )2
3. (z 1/2 )3
Are they multi-valued? Do they have branch points?
1.
√
(z 2 )1/2 = ± z 2 = ±z
Because of the (·)1/2 , the function is multi-valued. The only possible branch points are at zero and infinity.
If (( ei0 )2 )1/2 = 1, then (( e2πi )2 )1/2 = ( e4πi )1/2 = e2πi = 1. Thus we see that the function does not change
value when we walk around the origin. We can also consider this to be a path around infinity. This function
is multi-valued, but has no branch points.
2.
√
(z 1/2 )2 = (± z)2 = z
This function is single-valued.
3.
√ √
(z 1/2 )3 = (± z)3 = ±( z)3
This function is multi-valued. We consider the possible branch point at z = 0. If (( e0 )1/2 )3 = 1, then
(( e2πi )1/2 )3 = ( eπi )3 = e3πi = −1. Since the function changes value when we walk around the origin, it has
a branch point at z = 0. Since this is also a path around infinity, there is a branch point there.

234
 1 1


Example 9.6.8 Consider the function f (z) = log z−1 . Since z−1 is only zero at infinity and its only singularity
is at z = 1, the only possibilities for branch points are at z = 1 and z = ∞. Since
 
1
log = − log(z − 1)
z−1

and log w has branch points at zero and infinity, we see that f (z) has branch points at z = 1 and z = ∞.

Example 9.6.9 Consider the functions,

1. elog z

2. log ez .

Are they multi-valued? Do they have branch points?

1.

elog z = exp( Log z + 2πin) = e Log z e2πin = z

This function is single-valued.

2.

log ez = Log ez + 2πin = z + 2πim

This function is multi-valued. It may have branch points only where ez is zero or infinite. This only
occurs at z = ∞. Thus there are no branch points in the finite plane. The function does not change when
traversing a simple closed path. Since this path can be considered to enclose infinity, there is no branch
point at infinity.

235
 Consider (f (z))α where f (z) is single-valued and f (z) has either a zero or a singularity at z = z0 . (f (z))α
may have a branch point at z = z0 . If f (z) is not a power of z, then it may be difficult to tell if (f (z))α changes
value when we walk around z0 . Factor f (z) into f (z) = g(z)h(z) where h(z) is nonzero and finite at z0 . Then
g(z) captures the important behavior of f (z) at the z0 . g(z) tells us how fast f (z) vanishes or blows up. Since
(f (z))α = (g(z))α (h(z))α and (h(z))α does not have a branch point at z0 , (f (z))α has a branch point at z0 if and
only if (g(z))α has a branch point there.
Similarly, we can decompose

log(f (z)) = log(g(z)h(z)) = log(g(z)) + log(h(z))

to see that log(f (z)) has a branch point at z0 if and only if log(g(z)) has a branch point there.

Result 9.6.3 Consider a single-valued function f (z) that has either a zero or a singu-
larity at z = z0 . Let f (z) = g(z)h(z) where h(z) is nonzero and finite. (f (z))α has a
branch point at z = z0 if and only if (g(z))α has a branch point there. log(f (z)) has a
branch point at z = z0 if and only if log(g(z)) has a branch point there.

Example 9.6.10 Consider the functions,

1. sin z 1/2

2. (sin z)1/2

3. z 1/2 sin z 1/2

4. (sin z 2 )1/2
Find the branch points and the number of branches.

1.
√ √
sin z 1/2 = sin(± z) = ± sin z

236
 sin z 1/2 is multi-valued. It has two branches. There may be branch points at zero and infinity. Consider
the unit circle which is a path around the origin or infinity. If sin(( ei0 )1/2 ) = sin(1), then sin(( ei2π )1/2 ) =
sin( eiπ ) = sin(−1) = − sin(1). There are branch points at the origin and infinity.

2.
√
(sin z)1/2 = ± sin z

The function is multi-valued with two branches. The sine vanishes at z = nπ and is singular at infinity.
There could be branch points at these locations. Consider the point z = nπ. We can write
sin z
sin z = (z − nπ)
z − nπ
sin z
Note that z−nπ
is nonzero and has a removable singularity at z = nπ.

sin z cos z
lim = lim = (−1)n
z→nπ z − nπ z→nπ 1

Since (z − nπ)1/2 has a branch point at z = nπ, (sin z)1/2 has branch points at z = nπ.
Since the branch points at z = nπ go all the way out to infinity. It is not possible to make a path that
encloses infinity and no other singularities. The point at infinity is a non-isolated singularity. A point can
be a branch point only if it is an isolated singularity.

3.
√ √
z 1/2 sin z 1/2 = ± z sin(± z)
√ √
= ± z(± sin z)
√ √
= z sin z

The function is single-valued. Thus there could be no branch points.

237
4.
√
(sin z 2 )1/2 = ± sin z 2

This function is multi-valued. Since sin z 2 = 0 at z = (nπ)1/2 , there may be branch points there. First
consider the point z = 0. We can write

sin z 2
sin z 2 = z 2
z2
where sin(z 2 )/z 2 is nonzero and has a removable singularity at z = 0.

sin z 2 2z cos z 2
lim = lim = 1.
z→0 z 2 z→0 2z

Since (z 2 )1/2 does not have a branch point at z = 0, (sin z 2 )1/2 does not have a branch point there either.
√
Now consider the point z = nπ.

√ sin z 2
sin z 2 = (z − nπ) √
z − nπ
√ √
sin(z 2 )/(z − nπ) in nonzero and has a removable singularity at z = nπ.

sin z 2 2z cos z 2 √
lim
√ √ = lim
√ = 2 nπ(−1)n
z→ nπ z − nπ z→ nπ 1
√ √
Since (z − nπ)1/2 has a branch point at z = nπ, (sin z 2 )1/2 also has a branch point there.
Thus
√ we √
see that (sin√ z 2 )1/2 has 1/2
√ branch points at z = (nπ) for n ∈ Z \ {0}. This is the set of numbers:
{± π, ± 2π, . . . , ±i π, ±i 2π, . . . }. The point at infinity is a non-isolated singularity.

238
Example 9.6.11 Find the branch points of
f (z) = (z 3 − z)1/3 .
√
3
Introduce branch cuts. If f (2) = 6 then what is f (−2)?
We expand f (z).
f (z) = z 1/3 (z − 1)1/3 (z + 1)1/3 .
There are branch points at z = −1, 0, 1. We consider the point at infinity.
   1/3  1/3  1/3
1 1 1 1
f = −1 +1
ζ ζ ζ ζ
1
= (1 − ζ)1/3 (1 + ζ)1/3
ζ
Since f (1/ζ) does not have a branch point at ζ = 0, f (z) does not have a branch point at infinity. Consider the
three possible branch cuts in Figure 9.24.

Figure 9.24: Three Possible Branch Cuts for f (z) = (z 3 − z)1/3

The first and the third branch cuts will make the function single valued, the second will not. It is clear that
the first set makes the function single valued since it is not possible to walk around any of the branch points.

239
 The second set of branch cuts would allow you to walk around the branch points at z = ±1. If you walked
around these two once in the positive direction, the value of the function would change by the factor e4πi/3 .
The third set of branch cuts would allow you to walk around all three branch points together. You can verify
that if you walk around the three branch points, the value of the function will not change√( e6πi/3 = e2πi = 1).
Suppose we introduce the third set of branch cuts and are on the branch with f (2) = 3 6.
√
f (2) = (2 ei0 )1/3 (1 ei0 )1/3 (3 ei0 )1/3 = 6.
3

The value of f (−2) is

f (−2) = (2 eiπ )1/3 (3 eiπ )1/3 (1 eiπ )1/3
√ √ √
= 2 eiπ/3 3 eiπ/3 1 eiπ/3
3 3 3

√
= 6 eiπ
3

√
3
= − 6.

Example 9.6.12 Find the branch points and number of branches for
2
f (z) = z z .

2
z z = exp(z 2 log z)
There may be branch points at the origin and infinity due to the logarithm. Consider walking around a circle of
radius r centered at the origin in the positive direction. Since the logarithm changes by i2π, the value of f (z)
2
changes by the factor ei2πr . There are branch points at the origin and infinity. The function has an infinite
number of branches.

Example 9.6.13 Construct a branch of

f (z) = (z 2 + 1)1/3

240
such that
1 √
f (0) = (−1 + 3i).
2
First we factor f (z).

f (z) = (z − i)1/3 (z + i)1/3

There are branch points at z = ±i. Figure 9.25 shows one way to introduce branch cuts.

φ
ρ

r
θ

Figure 9.25: Branch Cuts for f (z) = (z 2 + 1)1/3

Since it is not possible to walk around any branch point, these cuts make the function single valued. We
introduce the coordinates:

z − i = ρ eiφ , z + i = r eiθ .

f (z) = (ρ eiφ )1/3 (r eiθ )1/3

√
= 3 ρr ei(φ+θ)/3

241
 The condition
1 √
f (0) = (−1 + 3i) = ei(2π/3+2πn)
2
can be stated
√
1 ei(φ+θ)/3 = ei(2π/3+2πn)
3

φ + θ = 2π + 6πn
The angles must be defined to satisfy this relation. One choice is
π 5π π 3π
<φ< , − <θ< .
2 2 2 2

Principal Branches. We construct the principal branch of the logarithm by putting a branch cut on the
negative real axis choose z = r eiθ , θ ∈ (−π, π). Thus the principal branch of the logarithm is
Log z = Log r + iθ, −π < θ < π.
Note that the if x is a negative real number, (and thus lies on the branch cut), then Log x is undefined.
The principal branch of z α is
z α = eα Log z .
Note that there is a branch cut on the negative real axis.
−απ < arg( eα Log z ) < απ
√ √
The principal branch of the z 1/2
is denoted z. The principal branch of z 1/n is denoted n z.
√
Example 9.6.14 Construct 1 − z 2 , the principal branch of (1 − z 2 )1/2 .
First note that since (1 − z 2 )1/2 = (1 − z)1/2 (1 + z)1/2 there are branch points at z = 1 and z = −1. The
principal branch of the square root has a branch cut on the negative real axis. 1 − z 2 is a negative real number
for z ∈ (−∞, −1) ∪ (1, ∞). Thus we put branch cuts on (−∞, −1] and [1, ∞).

242
9.7 Exercises
Cartesian and Modulus-Argument Form
Exercise 9.1
For a given real number φ, 0 ≤ φ < 2π, find the image of the sector 0 ≤ arg(z) < φ under the transformation
w = z 4 . How large should φ be so that the w plane is covered exactly once?
Hint, Solution

Trigonometric Functions
Exercise 9.2
In Cartesian coordinates, z = x + iy, write sin(z) in Cartesian and modulus-argument form.
Hint, Solution

Exercise 9.3
Show that ez in nonzero for all finite z.
Hint, Solution

Exercise 9.4
Show that
 
 z2  2
 e  ≤ e|z| .

When does equality hold?

Hint, Solution

Exercise 9.5
Solve coth(z) = 1.
Hint, Solution

243
Exercise 9.6
Solve 2 ∈ 2z . That is, for what values of z is 2 one of the values of 2z ? Derive this result then verify your answer
by evaluating 2z for the solutions that your find.
Hint, Solution

Exercise 9.7
Solve 1 ∈ 1z . That is, for what values of z is 1 one of the values of 1z ? Derive this result then verify your answer
by evaluating 1z for the solutions that your find.
Hint, Solution

Logarithmic Identities
Exercise 9.8
Find the fallacy in the following arguments:
1


1. log(−1) = log −1 = log(1) − log(−1) = − log(−1), therefore, log(−1) = 0.

2. 1 = 11/2 = ((−1)(−1))1/2 = (−1)1/2 (−1)1/2 = ii = −1, therefore, 1 = −1.

Hint, Solution

Exercise 9.9
Write the following expressions in modulus-argument or Cartesian form. Denote any multi-valuedness explicitly.
√
22/5 , 31+i , ( 3 − i)1/4 , 1i/4 .

Hint, Solution

Exercise 9.10
Solve cos z = 69.
Hint, Solution

244
Exercise 9.11
Solve cot z = i47.
Hint, Solution

Exercise 9.12
Determine all values of
1. log(−i)

2. (−i)−i

3. 3π

4. log(log(i))
and plot them in the complex plane.
Hint, Solution

Exercise 9.13
Determine all values of ii and log((1 + i)iπ ) and plot them in the complex plane.
Hint, Solution

Exercise 9.14
Find all z for which
1. ez = i

2. cos z = sin z

3. tan2 z = −1
Hint, Solution

245
Exercise 9.15
Show that
 
−1 i i+z
tan (z) = log
2 i−z

and
 
−1 1 1+z
tanh (z) = log .
2 1−z

Hint, Solution

Branch Points and Branch Cuts

Exercise 9.16
Determine the branch points of the function

f (z) = (z 3 − 1)1/2 .

Construct cuts and define a branch so that z = 0 and z = −1 do not lie on a cut, and such that f (0) = −i. What
is f (−1) for this branch?
Hint, Solution

Exercise 9.17
Determine the branch points of the function

w(z) = ((z − 1)(z − 6)(z + 2))1/2

Construct cuts and define a branch so that z = 4 does not lie on a cut, and such that w = 6i when z = 4.
Hint, Solution

246
Exercise 9.18
Give the number of branches and locations of the branch points for the functions
1. cos z 1/2
2. (z + i)−z
Hint, Solution

Exercise 9.19
Find the branch points of the following functions in the extended complex plane, (the complex plane including
the point at infinity).
1. (z 2 + 1)1/2
2. (z 3 − z)1/2
3. log(z 2 − 1)
 
z+1
4. log
z−1
Introduce branch cuts to make the functions single valued.
Hint, Solution

Exercise 9.20
Find all branch points and introduce cuts to make the following functions single-valued: For the first function,
choose cuts so that there is no cut within the disk |z| < 2.

1/2
1. f (z) = z 3 + 8
 1/2 !
z+1
2. f (z) = log 5 +
z−1

247
 3. f (z) = (z + i3)1/2

Hint, Solution

Exercise 9.21
Let f (z) have branch points at z = 0 and z = ±i, but nowhere else in the extended complex plane. How does
the value and argument of f (z) change while traversing the contour in Figure 9.26? Does the branch cut in
Figure 9.26 make the function single-valued?

Figure 9.26: Contour Around the Branch Points and Branch Cut.

Hint, Solution

Exercise 9.22
Let f (z) be analytic except for no more than a countably infinite number of singularities. Suppose that f (z) has
only one branch point in the finite complex plane. Does f (z) have a branch point at infinity? Now suppose that
f (z) has two or more branch points in the finite complex plane. Does f (z) have a branch point at infinity?
Hint, Solution

248
Exercise 9.23
Find all branch points of (z 4 + 1)1/4 in the extended complex plane. Which of the branch cuts in Figure 9.27
make the function single-valued.

Figure 9.27: Four Candidate Sets of Branch Cuts for (z 4 + 1)1/4

Hint, Solution

Exercise 9.24
Find the branch points of
 1/3
z
f (z) = 2
z +1
in the extended complex plane. Introduce branch cuts that make the function single-valued
√ and such that the
3
function is defined on the positive real axis. Define a branch such that f (1) = 1/ 2. Write down an explicit
formula for the value of the branch. What is f (1 + i)? What is the value of f (z) on either side of the branch
cuts?
Hint, Solution

Exercise 9.25
Find all branch points of

f (z) = ((z − 1)(z − 2)(z − 3))1/2

249
in the extended complex plane. Which of the branch cuts in Figure 9.28 will √ make the function single-valued.
Using the first set of branch cuts in this figure define a branch on which f (0) = i 6. Write out an explicit formula
for the value of the function on this branch.

Figure 9.28: Four Candidate Sets of Branch Cuts for ((z − 1)(z − 2)(z − 3))1/2

Hint, Solution

Exercise 9.26
Determine the branch points of the function

1/3
w = (z 2 − 2)(z + 2) .
Construct and define a branch so that the resulting cut is one line of finite extent and w(2) = 2. What is w(−3)
for this branch? What are the limiting values of w on either side of the branch cut?
Hint, Solution

Exercise 9.27
Construct the principal branch of arccos(z). ( Arccos (z) has the property that if x ∈ [−1, 1] then Arccos (x) ∈
[0, π]. In particular, Arccos (0) = π2 ).
Hint, Solution

250
Exercise 9.28
Find the branch points of (z 1/2 − 1)1/2 in the finite complex plane. Introduce branch cuts to make the function
single-valued.
Hint, Solution

Exercise 9.29
For the linkage illustrated in Figure 9.29, use complex variables to outline a scheme for expressing the angular
position, velocity and acceleration of arm c in terms of those of arm a. (You needn’t work out the equations.)

b
a
c φ
θ
l

Figure 9.29: A linkage

Hint, Solution

Exercise 9.30
Find the image of the strip |<(z)| < 1 and of the strip 1 < =(z) < 2 under the transformations:
1. w = 2z 2

z+1
2. w = z−1

Hint, Solution

251
Exercise 9.31
Locate and classify all the singularities of the following functions:

(z + 1)1/2
1.
z+2
 
1
2. cos
1+z
1
3.
(1 − ez )2
In each case discuss the possibility of a singularity at the point ∞.
Hint, Solution

Exercise 9.32
Describe how the mapping w = sinh(z) transforms the infinite strip −∞ < x < ∞, 0 < y < π into the w-plane.
Find cuts in the w-plane which make the mapping continuous both ways. What are the images of the lines (a)
y = π/4; (b) x = 1?
Hint, Solution

252
9.8 Hints
Cartesian and Modulus-Argument Form
Hint 9.1

Trigonometric Functions
Hint 9.2
Recall that sin(z) = 1
2i
( eiz − e−iz ). Use Result 8.3.1 to convert between Cartesian and modulus-argument form.

Hint 9.3
Write ez in polar form.

Hint 9.4
The exponential is an increasing function for real variables.

Hint 9.5
Write the hyperbolic cotangent in terms of exponentials.

Hint 9.6
Write out the multi-valuedness of 2z . There is a doubly-infinite set of solutions to this problem.

Hint 9.7
Write out the multi-valuedness of 1z .

Logarithmic Identities

253
Hint 9.8
Write out the multi-valuedness of the expressions.

Hint 9.9
Do the exponentiations in polar form.

Hint 9.10
Write the cosine in terms of exponentials. Multiply by eiz to get a quadratic equation for eiz .

Hint 9.11
Write the cotangent in terms of exponentials. Get a quadratic equation for eiz .

Hint 9.12

Hint 9.13
ii has an infinite number of real, positive values. ii = ei log i . log((1 + i)iπ ) has a doubly infinite set of values.
log((1 + i)iπ ) = log(exp(iπ log(1 + i))).

Hint 9.14

Hint 9.15

Branch Points and Branch Cuts

254
Hint 9.16

Hint 9.17

Hint 9.18

Hint 9.19
1. (z 2 + 1)1/2 = (z − i)1/2 (z + i)1/2

2. (z 3 − z)1/2 = z 1/2 (z − 1)1/2 (z + 1)1/2

3. log(z 2 − 1) = log(z − 1) + log(z + 1)

4. log z+1


z−1
= log(z + 1) − log(z − 1)

Hint 9.20

Hint 9.21
Reverse the orientation of the contour so that it encircles infinity and does not contain any branch points.

Hint 9.22
Consider a contour that encircles all the branch points in the finite complex plane. Reverse the orientation of
the contour so that it contains the point at infinity and does not contain any branch points in the finite complex

255
plane.

Hint 9.23
Factor the polynomial. The argument of z 1/4 changes by π/2 on a contour that goes around the origin once in
the positive direction.

Hint 9.24

Hint 9.25
To define the branch, define angles from each of the branch points in the finite complex plane.

Hint 9.26

Hint 9.27

Hint 9.28

Hint 9.29

256
Hint 9.30

Hint 9.31

Hint 9.32

257
9.9 Solutions
Cartesian and Modulus-Argument Form
Solution 9.1
We write the mapping w = z 4 in polar coordinates.

w = z 4 = (r eiθ )4 = r4 ei4θ

Thus we see that

w : {r eiθ | r ≥ 0, 0 ≤ θ < φ} → {r4 ei4θ | r ≥ 0, 0 ≤ θ < φ} = {r eiθ | r ≥ 0, 0 ≤ θ < 4φ}.

We can state this in terms of the argument.

w : {z | 0 ≤ arg(z) < φ} → {z | 0 ≤ arg(z) < 4φ}

If φ = π/2, the sector will be mapped exactly to the whole complex plane.

Trigonometric Functions
Solution 9.2

1 iz
e − e−iz


sin z =
2i
1 −y+ix
− ey−ix


= e
2i
1 −y
e (cos x + i sin x) − ey (cos x − i sin x)


=
2i
1 −y
e (sin x − i cos x) + ey (sin x + i cos x)


=
2
= sin x cosh y + i cos x sinh y

258
 q
sin z = sin2 x cosh2 y + cos2 x sinh2 y exp(i arctan(sin x cosh y, cos x sinh y))
q
= cosh2 y − cos2 x exp(i arctan(sin x cosh y, cos x sinh y))
r
1
= (cosh(2y) − cos(2x)) exp(i arctan(sin x cosh y, cos x sinh y))
2

Solution 9.3
In order that ez be zero, the modulus, ex must be zero. Since ex has no finite solutions, ez = 0 has no finite
solutions.

Solution 9.4

   
 z2   (x+iy)2 
e  = e 
 
 x2 −y2 +i2xy 
= e 
2 −y 2
= ex

2 2 2 +y 2
e|z| = e|x+iy| = ex

The exponential function is an increasing function for real variables. Since x2 − y 2 ≤ x2 + y 2 ,

 
 z2  2
 e  ≤ e|z| .

Equality holds when y = 0.

259
Solution 9.5

coth(z) = 1

( ez + e−z )/2
=1
( ez − e−z )/2

ez + e−z = ez − e−z

e−z = 0
There are no solutions.

Solution 9.6
We write out the multi-valuedness of 2z .
2 ∈ 2z
e Log 2 ∈ ez log(2)
e Log 2 ∈ { ez( Log (2)+i2πn) | n ∈ Z}
Log 2 ∈ z{ Log 2 + i2πn + i2πm | m, n ∈ Z}
 
Log (2) + i2πm
z= | m, n ∈ Z
Log (2) + i2πn

We verify this solution. Consider m and n to be fixed integers. We express the multi-valuedness in terms of k.
2( Log (2)+i2πm)/( Log (2)+i2πn) = e( Log (2)+i2πm)/( Log (2)+i2πn) log(2)
= e( Log (2)+i2πm)/( Log (2)+i2πn)( Log (2)+i2πk)

For k = n, this has the value, e Log (2)+i2πm = e Log (2) = 2.

260
Solution 9.7
We write out the multi-valuedness of 1z .

1 ∈ 1z
1 ∈ ez log(1)
1 ∈ { ezi2πn | n ∈ Z}

The element corresponding to n = 0 is e0 = 1. Thus 1 ∈ 1z has the solutions,

z ∈ C.

That is, z may be any complex number. We verify this solution.

1z = ez log(1) = ezi2πn

For n = 0, this has the value 1.

Logarithmic Identities
Solution 9.8
1. The algebraic manipulations are fine. We write out the multi-valuedness of the logarithms.
 
1
log(−1) = log = log(1) − log(−1) = − log(−1)
−1
{iπ + i2πn : n ∈ Z} = {iπ + i2πn : n ∈ Z} = {i2πn : n ∈ Z} − {iπ + i2πn : n ∈ Z} = {−iπ − i2πn : n ∈ Z}

Thus log(−1) = − log(−1). However this does not imply that log(−1) = 0. This is because the logarithm
is a set-valued function log(−1) = − log(−1) is really saying:

{iπ + i2πn : n ∈ Z} = {−iπ − i2πn : n ∈ Z}

261
 2. We consider
1 = 11/2 = ((−1)(−1))1/2 = (−1)1/2 (−1)1/2 = ii = −1.
There are three multi-valued expressions above.
11/2 = ±1
((−1)(−1))1/2 = ±1
(−1)1/2 (−1)1/2 = (±i)(±i) = ±1
Thus we see that the first and fourth equalities are incorrect.
1 6= 11/2 , (−1)1/2 (−1)1/2 6= ii

Solution 9.9

22/5 = 41/5
√
= 4 11/5
5

√
= 4 ei2nπ/5 ,
5
n = 0, 1, 2, 3, 4

31+i = e(1+i) log 3

= e(1+i)( Log 3+i2πn)
= e Log 3−2πn ei( Log 3+2πn) , n∈Z

√
( 3 − i)1/4 = (2 e−iπ/6 )1/4
√
= 2 e−iπ/24 11/4
4

√
= 2 ei(πn/2−π/24) ,
4
n = 0, 1, 2, 3

262
 1i/4 = e(i/4) log 1
= e(i/4)(i2πn)
= e−πn/2 , n∈Z

Solution 9.10

cos z = 69
e + e−iz
iz
= 69
2
ei2z − 138 eiz + 1 = 0
1 √ 
eiz = 138 ± 1382 − 4
2  √ 
z = −i log 69 ± 2 1190
  √  
z = −i Log 69 ± 2 1190 + i2πn
 √ 
z = 2πn − i Log 69 ± 2 1190 , n ∈ Z

263
Solution 9.11

cot z = i47
( e + e−iz )/2
iz
= i47
( eiz − e−iz )/(2i)
eiz + e−iz = 47( eiz − e−iz )
46 ei2z − 48 = 0
24
i2z = log
23
i 24
z = − log
 2 23
i 24
z=− Log + i2πn , n∈Z
2 23
i 24
z = πn − Log , n∈Z
2 23

Solution 9.12
1.

log(−i) = Log (| − i|) + i arg(−i)

 π 
= Log (1) + i − + 2πn , n∈Z
2

π
log(−i) = −i + i2πn, n∈Z
2

These are equally spaced points in the imaginary axis. See Figure 9.30.

264
 10

-1 1
-10

Figure 9.30: log(−i)

2.

(−i)−i = e−i log(−i)

= e−i(−iπ/2+i2πn) , n∈Z

(−i)−i = e−π/2+2πn , n∈Z

These are points on the positive real axis with an accumulation point at the origin. See Figure 9.31.

3.

3π = eπ log(3)
= eπ( Log (3)+i arg(3))

3π = eπ( Log (3)+i2πn) , n∈Z

These points all lie on the circle of radius |eπ | centered about the origin in the complex plane. See Figure 9.32.

265
 1

-1

Figure 9.31: (−i)−i

10

-10 -5 5 10

-5

-10

Figure 9.32: 3π

4.
 π 
log(log(i)) = log i + 2πm , m ∈ Z
π 2   π 
= Log  + 2πm + i Arg i + 2πm + i2πn, m, n ∈ Z
 
π2 2
 π
= Log  + 2πm + i sign (1 + 4m) + i2πn, m, n ∈ Z
 
2 2

266
 These points all lie in the right half-plane. See Figure 9.33.

20
10

1 2 3 4 5
-10
-20

Figure 9.33: log(log(i))

Solution 9.13

ii = ei log(i)
= ei( Log |i|+i Arg (i)+i2πn) , n∈Z
i(iπ/2+i2πn)
= e , n∈Z
= e−π(1/2+2n) , n∈Z

These are points on the positive real axis. There is an accumulation point at z = 0. See Figure 9.34.

267
 1

25 50 75 100

-1

Figure 9.34: ii

log (1 + i)iπ = log eiπ log(1+i)

= iπ log(1 + i) + i2πn, n ∈ Z
= iπ( Log |1 + i| + i Arg (1 + i) + i2πm) + i2πn, m, n ∈ Z
 
1 π
= iπ Log 2 + i + i2πm + i2πn, m, n ∈ Z
2 4
   
2 1 1
= −π + 2m + iπ Log 2 + 2n , m, n ∈ Z
4 2

See Figure 9.35 for a plot.

268
 10
5

-40 -20 20
-5
-10

Figure 9.35: log((1 + i)iπ )

Solution 9.14
1.

ez = i
z = log i
z = Log (|i|) + i arg(i)
π 
z = Log (1) + i + 2πn , n ∈ Z
2
π
z = i + i2πn, n ∈ Z
2

269
2. We can solve the equation by writing the cosine and sine in terms of the exponential.
cos z = sin z
eiz + e−iz eiz − e−iz
=
2 i2
(1 + i) e = (−1 + i) e−iz
iz

−1 + i
ei2z =
1+i
ei2z = i
i2z = log(i)
π
i2z = i + i2πn, n ∈ Z
2
π
z = + πn, n ∈ Z
4
3.
tan2 z = −1
sin2 z = − cos2 z
cos z = ±i sin z
e + e−iz
iz
eiz − e−iz
= ±i
2 2i
−iz −iz
e = −e or e = − eiz
iz

e−iz = 0 or eiz = 0
ey−ix = 0 or e−y+ix = 0
ey = 0 or e−y = 0
z=∅
There are no solutions for finite z.

270
Solution 9.15
First we consider tan−1 (z).

w = tan−1 (z)
z = tan(w)
sin(w)
z=
cos(w)
( e − e−iw )/(2i)
iw
z=
( eiw + e−iw )/2
z e + z e−iw = −i eiw + i e−iw
iw

(i + z) ei2w = (i − z)
 1/2
iw i−z
e =
i+z
 1/2
i−z
w = −i log
i+z
 
−1 i i+z
tan (z) = log
2 i−z

271
 Now we consider tanh−1 (z).
w = tanh−1 (z)
z = tanh(w)
sinh(w)
z=
cosh(w)
( e − e−w )/2
w
z= w
( e + e−w )/2
z ew + z e−w = ew − e−w
(z − 1) e2w = −z − 1
 1/2
w −z − 1
e =
z−1
 1/2
z+1
w = log
1−z
 
−1 1 z+1
tanh (z) = log
2 1−z

Branch Points and Branch Cuts

Solution 9.16
The cube roots of 1 are
( √ √ )
−1 + i 3 −1 − i 3
1, ei2π/3 , ei4π/3 =

1, , .
2 2

Thus we can write

√ !1/2 √ !1/2

1/2 1−i 3 1+i 3
z3 − 1 = (z − 1)1/2 z+ z+ .
2 2

272
There are three branch points on the circle of radius 1.
( √ √ )
−1 + i 3 −1 − i 3
z= 1, ,
2 2

We examine the point at infinity.

f (1/ζ) = (1/ζ 3 − 1)1/2 = ζ −3/2 (1 − ζ 3 )1/2

Since f (1/ζ) has a branch point at ζ = 0, f (z) has a branch point at infinity.
There are several ways of introducing branch cuts to separate the branches of the function. The easiest
approach is to put a branch cut from each of the three branch points in the finite complex plane out to the
branch point at infinity. See Figure 9.36a. Clearly this makes the function single valued as it is impossible to
walk around any of the branch points. Another approach is to have a branch cut from one of the branch points
in the finite plane to the branch point at infinity and a branch cut connecting the remaining two branch points.
See Figure 9.36bcd. Note that in walking around any one of the finite branch points, (in the positive direction),
the argument of the function changes by π. This means that the value of the function changes by eiπ , which is
to say the value of the function changes sign. In walking around any two of the finite branch points, (again in
the positive direction), the argument of the function changes by 2π. This means that the value of the function
changes by ei2π , which is to say that the value of the function does not change. This demonstrates that the latter
branch cut approach makes the function single-valued.
Now we construct a branch. We will use the branch cuts in Figure 9.36a. We introduce variables to measure
radii and angles from the three finite branch points.

z − 1 = r1 eiθ1 , 0 < θ1 < 2π

√
1−i 3 2π π
z+ = r2 eiθ2 , − < θ2 <
2√ 3 3
1+i 3 π 2π
z+ = r3 eiθ3 , − < θ3 <
2 3 3

273
 a b c d

Figure 9.36: (z 3 − 1)1/2

We compute f (0) to see if it has the desired value.

√
f (z) = r1 r2 r3 ei(θ1 +θ2 +θ3 )/2
f (0) = ei(π−π/3+π/3)/2 = i

Since it does not have the desired value, we change the range of θ1 .

z − 1 = r1 eiθ1 , 2π < θ1 < 4π

f (0) now has the desired value.

f (0) = ei(3π−π/3+π/3)/2 = −i

We compute f (−1).
√ √
f (−1) = 2 ei(3π−2π/3+2π/3)/2 = −i 2

Solution 9.17
First we factor the function.

w(z) = ((z + 2)(z − 1)(z − 6))1/2 = (z + 2)1/2 (z − 1)1/2 (z − 6)1/2

274
There are branch points at z = −2, 1, 6. Now we examine the point at infinity.
     1/2    1/2
1 1 1 1 −3/2 2 1 6
w = +2 −1 −6 =ζ 1+ 1− 1−
ζ ζ ζ ζ ζ ζ ζ

Since ζ −3/2 has a branch point at ζ = 0 and the rest of the terms are analytic there, w(z) has a branch point at
infinity.
Consider the set of branch cuts in Figure 9.37. These cuts let us walk around the branch points at z = −2
and z = 1 together or if we change our perspective, we would be walking around the branch points at z = 6 and
z = ∞ together. Consider a contour in this cut plane that encircles the branch points at z = −2 and z = 1. Since
the argument of (z − z0 )1/2 changes by π when we walk around z0 , the argument of w(z) changes by 2π when we
traverse the contour. Thus the value of the function does not change and it is a valid set of branch cuts.

Figure 9.37: Branch Cuts for ((z + 2)(z − 1)(z − 6))1/2

Now to define the branch. We make a choice of angles.

z + 2 = r1 eiθ1 , θ1 = θ2 for z ∈ (1..6),

iθ2
z − 1 = r2 e , θ2 = θ1 for z ∈ (1..6),
z − 6 = r3 eiθ3 , 0 < θ3 < 2π

275
The function is

1/2 √
w(z) = r1 eiθ1 r2 eiθ2 r3 eiθ3 = r1 r2 r3 ei(θ1 +θ2 +θ3 )/2 .
We evaluate the function at z = 4.
p
w(4) = (6)(3)(2) ei(2πn+2πn+π)/2 = i6
We see that our choice of angles gives us the desired branch.

Solution 9.18
1.
√ √
cos z 1/2 = cos(± z) = cos( z)
This is a single-valued function. There are no branch points.
2.
(z + i)−z = e−z log(z+i)
= e−z( Log |z+i|+i Arg (z+i)+i2πn) , n∈Z
There is a branch point at z = −i. There are an infinite number of branches.

Solution 9.19
1.
f (z) = (z 2 + 1)1/2 = (z + i)1/2 (z − i)1/2
We see that there are branch points at z = ±i. To examine the point at infinity, we substitute z = 1/ζ and
examine the point ζ = 0.
 2 !1/2
1 1
+1 = 2 1/2 (1 + ζ 2 )1/2
ζ (ζ )

276
 Since there is no branch point at ζ = 0, f (z) has no branch point at infinity.
A branch cut connecting z = ±i would make the function single-valued. We could also accomplish this with
two branch cuts starting z = ±i and going to infinity.
2.
f (z) = (z 3 − z)1/2 = z 1/2 (z − 1)1/2 (z + 1)1/2
There are branch points at z = −1, 0, 1. Now we consider the point at infinity.
   3 !1/2
1 1 1
f = − = ζ −3/2 (1 − ζ 2 )1/2
ζ ζ ζ
There is a branch point at infinity.
One can make the function single-valued with three branch cuts that start at z = −1, 0, 1 and each go to
infinity. We can also make the function single-valued with a branch cut that connects two of the points
z = −1, 0, 1 and another branch cut that starts at the remaining point and goes to infinity.
3.
f (z) = log(z 2 − 1) = log(z − 1) + log(z + 1)
There are branch points at z = ±1.
   
1 1
f = log 2
− 1 = log(ζ −2 ) + log(1 − ζ 2 )
ζ ζ
log(ζ −2 ) has a branch point at ζ = 0.
log(ζ −2 ) = Log |ζ −2 | + i arg(ζ −2 ) = Log |ζ −2 | − i2 arg(ζ)
Every time we walk around the point ζ = 0 in the positive direction, the value of the function changes by
−i4π. f (z) has a branch point at infinity.
We can make the function single-valued by introducing two branch cuts that start at z = ±1 and each go
to infinity.

277
 4.
 
z+1
f (z) = log = log(z + 1) − log(z − 1)
z−1
There are branch points at z = ±1.
     
1 1/ζ + 1 1+ζ
f = log = log
ζ 1/ζ − 1 1−ζ
There is no branch point at ζ = 0. f (z) has no branch point at infinity.
We can make the function single-valued by introducing two branch cuts that start at z = ±1 and each go
to infinity. We can also make the function single-valued with a branch cut that connects the points z = ±1.
This is because log(z + 1) and − log(z − 1) change by i2π and −i2π, respectively, when you walk around
their branch points once in the positive direction.

Solution 9.20
1. The cube roots of −8 are
n √ √ o
−2, −2 ei2π/3 , −2 ei4π/3 = −2, 1 + i 3, 1 − i 3 .


Thus we can write

1/2 √ √
z3 + 8 = (z + 2)1/2 (z − 1 − i 3)1/2 (z − 1 + i 3)1/2 .

There are three branch points on the circle of radius 2.

n √ √ o
z = −2, 1 + i 3, 1 − i 3 .

We examine the point at infinity.

f (1/ζ) = (1/ζ 3 + 8)1/2 = ζ −3/2 (1 + 8ζ 3 )1/2

278
 Since f (1/ζ) has a branch point at ζ = 0, f (z) has a branch point at infinity.
There are several ways of introducing branch cuts outside of the disk |z| < 2 to separate the branches of
the function. The easiest approach is to put a branch cut from each of the three branch points in the finite
complex plane out to the branch point at infinity. See Figure 9.38a. Clearly this makes the function single
valued as it is impossible to walk around any of the branch points. Another approach is to have a branch cut
from one of the branch points in the finite plane to the branch point at infinity and a branch cut connecting
the remaining two branch points. See Figure 9.38bcd. Note that in walking around any one of the finite
branch points, (in the positive direction), the argument of the function changes by π. This means that the
value of the function changes by eiπ , which is to say the value of the function changes sign. In walking
around any two of the finite branch points, (again in the positive direction), the argument of the function
changes by 2π. This means that the value of the function changes by ei2π , which is to say that the value of
the function does not change. This demonstrates that the latter branch cut approach makes the function
single-valued.

a b c d

Figure 9.38: (z 3 + 8)1/2

2.
 1/2 !
z+1
f (z) = log 5 +
z−1

279
First we deal with the function

 1/2
z+1
g(z) =
z−1

Note that it has branch points at z = ±1. Consider the point at infinity.

 1/2  1/2
1/ζ + 1 1+ζ
g(1/ζ) = =
1/ζ − 1 1−ζ

Since g(1/ζ) has no branch point at ζ = 0, g(z) has no branch point at infinity. This means that if we
walk around both of the branch points at z = ±1, the function does not change value. We can verify this
with another method: When we walk around the point z = −1 once in the positive direction, the argument
of z + 1 changes by 2π, the argument of (z + 1)1/2 changes by π and thus the value of (z + 1)1/2 changes
by eiπ = −1. When we walk around the point z = 1 once in the positive direction, the argument of z − 1
changes by 2π, the argument of (z − 1)−1/2 changes by −π and thus the value of (z − 1)−1/2 changes by
e−iπ = −1. f (z) has branch points at z = ±1. When we walk around both points z = ±1 once in the

1/2
positive direction, the value of z+1
z−1
does not change. Thus we can make the function single-valued with
a branch cut which enables us to walk around either none or both of these branch points. We put a branch
cut from −1 to 1 on the real axis.

f (z) has branch points where

 1/2
z+1
5+
z−1

is either zero or infinite. The only place in the extended complex plane where the expression becomes infinite

280
is at z = 1. Now we look for the zeros.
 1/2
z+1
5+ = 0.
z−1
 1/2
z+1
= −5.
z−1
z+1
= 25.
z−1
z + 1 = 25z − 25
13
z=
12

Note that
 1/2
13/12 + 1
= 251/2 = ±5.
13/12 − 1

On one branch, (which we call the positive branch), of the function g(z) the quantity
 1/2
z+1
5+
z−1

is always nonzero. On the other (negative) branch of the function, this quantity has a zero at z = 13/12.
The logarithm introduces branch points at z = 1 on both the positive and negative branch of g(z). It
introduces a branch point at z = 13/12 on the negative branch of g(z). To determine if additional branch
cuts are needed to separate the branches, we consider
 1/2
z+1
w =5+
z−1

281
and see where the branch cut between ±1 gets mapped to in the w plane. We rewrite the mapping.
 1/2
2
w =5+ 1+
z−1

The mapping is the following sequence of simple transformations:

(a) z 7→ z − 1
1
(b) z 7→
z
(c) z 7→ 2z
(d) z 7→ z + 1
(e) z 7→ z 1/2
(f) z 7→ z + 5

We show these transformations graphically below.

-1 1 -2 0 -1/2 -1

1
z 7→ z − 1 z 7→ z 7→ 2z z 7→ z + 1
z

z 7→ z 1/2 z 7→ z + 5
For the positive branch of g(z), the branch cut is mapped to the line x = 5 and the z plane is mapped to
the half-plane x > 5. log(w) has branch points at w = 0 and w = ∞. It is possible to walk around only one

282
 of these points in the half-plane x > 5. Thus no additional branch cuts are needed in the positive sheet of
g(z).
For the negative branch of g(z), the branch cut is mapped to the line x = 5 and the z plane is mapped to
the half-plane x < 5. It is possible to walk around either w = 0 or w = ∞ alone in this half-plane. Thus
we need an additional branch cut. On the negative sheet of g(z), we put a branch cut beteen z = 1 and
z = 13/12. This puts a branch cut between w = ∞ and w = 0 and thus separates the branches of the
logarithm.
Figure 9.39 shows the branch cuts in the positive and negative sheets of g(z).

Im(z) Im(z)
g(13/12)=5 g(13/12)=-5
Re(z) Re(z)

 
z+1 1/2


Figure 9.39: The branch cuts for f (z) = log 5 + z−1
.

3. The function f (z) = (z + i3)1/2 has a branch point at z = −i3. The function is made single-valued by
connecting this point and the point at infinity with a branch cut.

Solution 9.21
Note that the curve with opposite orientation goes around infinity in the positive direction and does not enclose
any branch points. Thus the value of the function does not change when traversing the curve, (with either
orientation, of course). This means that the argument of the function must change my an integer multiple of 2π.
Since the branch cut only allows us to encircle all three or none of the branch points, it makes the function single
valued.

283
Solution 9.22
We suppose that f (z) has only one branch point in the finite complex plane. Consider any contour that encircles
this branch point in the positive direction. f (z) changes value if we traverse the contour. If we reverse the
orientation of the contour, then it encircles infinity in the positive direction, but contains no branch points in the
finite complex plane. Since the function changes value when we traverse the contour, we conclude that the point
at infinity must be a branch point. If f (z) has only a single branch point in the finite complex plane then it must
have a branch point at infinity.
If f (z) has two or more branch points in the finite complex plane then it may or may not have a branch point
at infinity. This is because the value of the function may or may not change on a contour that encircles all the
branch points in the finite complex plane.

Solution 9.23
First we factor the function,
 1/4  1/4  1/4  1/4
4

1/4 1+i −1 + i −1 − i 1−i
f (z) = z + 1 = z− √ z− √ z− √ z− √ .
2 2 2 2
±1±i
There are branch points at z = √ .
2
We make the substitution z = 1/ζ to examine the point at infinity.
   1/4
1 1
f = +1
ζ ζ4
1
1/4
= 1/4
1 + ζ4
(ζ 4 )

(ζ 1/4 )4 has a removable singularity at the point ζ = 0, but no branch point there. Thus (z 4 + 1)1/4 has no branch
point at infinity.
Note that the argument of (z 4 − z0 )1/4 changes by π/2 on a contour that goes around the point z0 once in the
positive direction. The argument of (z 4 + 1)1/4 changes by nπ/2 on a contour that goes around n of its branch
points. Thus any set of branch cuts that permit you to walk around only one, two or three of the branch points
will not make the function single valued. A set of branch cuts that permit us to walk around only zero or all four

284
of the branch points will make the function single-valued. Thus we see that the first two sets of branch cuts in
Figure 9.27 will make the function single-valued, while the remaining two will not.
Consider the contour in Figure ??. There are two ways to see that the function does not change value while
traversing the contour. The first is to note that each of the branch points makes the argument of the function
increase by π/2. Thus the argument of (z 4 + 1)1/4 changes by 4(π/2) = 2π on the contour. This means that the
value of the function changes by the factor ei2π = 1. If we change the orientation of the contour, then it is a
contour that encircles infinity once in the positive direction. There are no branch points inside the this contour
with opposite orientation. (Recall that the inside of a contour lies to your left as you walk around it.) Since there
are no branch points inside this contour, the function cannot change value as we traverse it.

Solution 9.24

 1/3
z
f (z) = 2
= z 1/3 (z − i)−1/3 (z + i)−1/3
z +1
There are branch points at z = 0, ±i.
1/3
ζ 1/3
  
1 1/ζ
f = =
ζ (1/ζ)2 + 1 (1 + ζ 2 )1/3
There is a branch point at ζ = 0. f (z) has a branch point at infinity.
We introduce branch cuts from z = 0 to infinity on the negative real axis, from z = i to infinity on the positive
imaginary axis and from z = −i to infinity on the negative imaginary axis. As we cannot walk around any of the
branch points, this makes the function single-valued.
We define a branch by defining angles from the branch points. Let

z = r eiθ − π < θ < π,

iφ
(z − i) = s e − 3π/2 < φ < π/2,
(z + i) = t eiψ − π/2 < ψ < 3π/2.

285
With
f (z) = z 1/3 (z − i)−1/3 (z + i)−1/3
√ 1 −iφ/3 1 −iψ/3
= 3 r eiθ/3 √ e √3
e
3
s t
r
r i(θ−φ−ψ)/3
= 3 e
st
we have an explicit formula for computing the value of the function for this branch. Now we compute f (1) to see
if we chose the correct ranges for the angles. (If not, we’ll just change one of them.)
s
1 1
f (1) = 3 √ √ ei(0−π/4−(−π/4))/3 = √ 3
2 2 2
We made the right choice for the angles. Now to compute f (1 + i).
s√ r
3 2 i(π/4−0− Arctan (2))/3 6 2 i(π/4− Arctan (2))/3
f (1 + i) = √ e = e
1 5 5
Consider the value of the function above and below the branch cut on the negative real axis. Above the branch
cut the function is
r
x
f (−x + i0) = 3 √ √ ei(π−φ−ψ)/3
2
x +1 x +12

Note that φ = −ψ so that

r r √
x x 1+i 3
f (−x + i0) = 3
2
ei(π)/3 = 3
.
x +1 x2 + 1 2
Below the branch cut θ = −π and
r r √
x x 1−i 3
f (−x − i0) = 3
2
ei(−π)/3 = 3
.
x +1 x2 + 1 2

286
For the branch cut along the positive imaginary axis,

r
y
f (iy + 0) = 3 ei(π/2−π/2−π/2)/3
(y − 1)(y + 1)
r
y
= 3 e−iπ/6
(y − 1)(y + 1)
√
3−i
r
y
= 3 ,
(y − 1)(y + 1) 2

r
y
f (iy − 0) = 3 ei(π/2−(−3π/2)−π/2)/3
(y − 1)(y + 1)
r
y
= 3 eiπ/2
(y − 1)(y + 1)
r
y
=i3 .
(y − 1)(y + 1)

For the branch cut along the negative imaginary axis,

r
y
f (−iy + 0) = 3 ei(−π/2−(−π/2)−(−π/2))/3
(y + 1)(y − 1)
r
y
= 3 eiπ/6
(y + 1)(y − 1)
r √
y 3+i
= 3 ,
(y + 1)(y − 1) 2

287
 r
y
f (−iy − 0) = 3 ei(−π/2−(−π/2)−(3π/2))/3
(y + 1)(y − 1)
r
y
= 3 e−iπ/2
(y + 1)(y − 1)
r
y
= −i 3 .
(y + 1)(y − 1)

Solution 9.25
First we factor the function.

f (z) = ((z − 1)(z − 2)(z − 3))1/2 = (z − 1)1/2 (z − 2)1/2 (z − 3)1/2

There are branch points at z = 1, 2, 3. Now we examine the point at infinity.

     1/2    1/2
1 1 1 1 −3/2 1 2 3
f = −1 −2 −3 =ζ 1− 1− 1−
ζ ζ ζ ζ ζ ζ ζ

Since ζ −3/2 has a branch point at ζ = 0 and the rest of the terms are analytic there, f (z) has a branch point at
infinity.
The first two sets of branch cuts in Figure 9.28 do not permit us to walk around any of the branch points,
including the point at infinity, and thus make the function single-valued. The third set of branch cuts lets us
walk around the branch points at z = 1 and z = 2 together or if we change our perspective, we would be walking
around the branch points at z = 3 and z = ∞ together. Consider a contour in this cut plane that encircles the
branch points at z = 1 and z = 2. Since the argument of (z − z0 )1/2 changes by π when we walk around z0 , the
argument of f (z) changes by 2π when we traverse the contour. Thus the value of the function does not change
and it is a valid set of branch cuts. Clearly the fourth set of branch cuts does not make the function single-valued
as there are contours that encircle the branch point at infinity and no other branch points. The other way to see
this is to note that the argument of f (z) changes by 3π as we traverse a contour that goes around the branch
points at z = 1, 2, 3 once in the positive direction.

288
 Now to define the branch. We make the preliminary choice of angles,

z − 1 = r1 eiθ1 , 0 < θ1 < 2π,

iθ2
z − 2 = r2 e , 0 < θ2 < 2π,
z − 3 = r3 eiθ3 , 0 < θ3 < 2π.

The function is

1/2 √
f (z) = r1 eiθ1 r2 eiθ2 r3 eiθ3 = r1 r2 r3 ei(θ1 +θ2 +θ3 )/2 .

The value of the function at the origin is

√ i(3π)/2 √
f (0) = 6e = −i 6,

which is not what we wanted. We will change range of one of the angles to get the desired result.

z − 1 = r1 eiθ1 , 0 < θ1 < 2π,

iθ2
z − 2 = r2 e , 0 < θ2 < 2π,
iθ3
z − 3 = r3 e , 2π < θ3 < 4π.

√ i(5π)/2 √
f (0) = 6e = i 6,

Solution 9.26

1/3 √ √
w = (z 2 − 2)(z + 2) (z + 2)1/3 (z − 2)1/3 (z + 2)1/3
√
There are branch points at z = ± 2 and z = −2. If we walk around any one of the branch points once in the
positive direction, the argument of w changes by 2π/3 and thus the value of the function changes by ei2π/3 . If we

289
walk around all three branch points then the argument of w changes by 3 × 2π/3
√ = 2π. The value of the function
i2π
is unchanged as e = 1. Thus the branch cut on the real axis from −2 to 2 makes the function single-valued.
Now we define a branch. Let
√ √
z− 2 = a eiα , z+ 2 = b eiβ , z + 2 = c eiγ .

We constrain the angles as follows: On the positive real axis, α = β = γ. See Figure 9.40.

Im(z)

c b a

β α Re(z)

1/3
Figure 9.40: A branch of ((z 2 − 2)(z + 2)) .

Now we determine w(2).

√ √
w(2) = (2 − 2)1/3 (2 + 2)1/3 (2 + 2)1/3
√ i0 3 √ √
q q
3
2 + 2 ei0 4 ei0
3
= 2 − 2e
√3
√3
= 2 4
= 2.

Note that we didn’t have to choose the angle from each of the branch points as zero. Choosing any integer multiple
of 2π would give us the same result.

290
 √ √
w(−3) = (−3 − 2)1/3 (−3 + 2)1/3 (−3 + 2)1/3
√ iπ/3 3 √ √
q q
3
3 − 2 eiπ/3 1 eiπ/3
3
= 3 + 2e
√
= 7 eiπ
3

√
3
=− 7

The value of the function is

√
3
w= abc ei(α+β+γ)/3 .
√ √
Consider the interval (− 2 . . . 2). As we approach the branch cut from above, the function has the value,
√
3 iπ/3
q√
3
√
w= abc e = ( 2 − x)(x + 2)(x + 2) eiπ/3 .

As we approach the branch cut from below, the function has the value,
√ q√ √
w = abc e−iπ/3 = ( 2 − x)(x + 2)(x + 2) e−iπ/3 .
3 3

√
Consider the interval (−2 . . . − 2). As we approach the branch cut from above, the function has the value,
√
3 i2π/3
q√
3
√
w= abc e = ( 2 − x)(−x − 2)(x + 2) ei2π/3 .

As we approach the branch cut from below, the function has the value,
√ −i2π/3
q√ √
= ( 2 − x)(−x − 2)(x + 2) e−i2π/3 .
3 3
w = abc e

291
 3
2.5
2
1.5
1
0.5
-1 -0.5 0.5 1

Figure 9.41: The Principal Branch of the arc cosine, Arccos(x).

Solution 9.27
Arccos (x) is shown in Figure 9.41 for real variables in the range [−1, 1].
First we write arccos(z) in terms of log(z). If cos(w) = z, then w = arccos(z).
cos(w) = z
e + e−iw
iw
=z
2
( eiw )2 − 2z eiw + 1 = 0
eiw = z + (z 2 − 1)1/2
w = −i log(z + (z 2 − 1)1/2 )
Thus we have
arccos(z) = −i log(z + (z 2 − 1)1/2 ).

Since Arccos (0) = π2 , we must find the branch such that

−i log(0 + (02 − 1)1/2 ) = 0

−i log((−1)1/2 ) = 0.

292
Since
 π  π
−i log(i) = −i i + i2πn = + 2πn
2 2
and
 π  π
−i log(−i) = −i −i + i2πn = − + 2πn
2 2
we must choose the branch of the square root such that (−1)1/2 = i and the branch of the logarithm such that
log(i) = i π2 .
First we construct the branch of the square root.

(z 2 − 1)1/2 = (z + 1)1/2 (z − 1)1/2

We see that there are branch points at z = −1 and z = 1. In particular we want the Arccos to be defined for
z = x, x ∈ [−1, 1]. Hence we introduce branch cuts on the lines −∞ < x ≤ −1 and 1 ≤ x < ∞. Define the local
coordinates

z + 1 = r eiθ , z − 1 = ρ eiφ .

With the given branch cuts, the angles have the possible ranges

{θ} = {. . . , (−π..π), (π..3π), . . . }, {φ} = {. . . , (0..2π), (2π..4π), . . . }.

Now we choose ranges for θ and φ and see if we get the desired branch. If not, we choose a different range for one
of the angles. First we choose the ranges

θ ∈ (−π..π), φ ∈ (0..2π).

If we substitute in z = 0 we get

(02 − 1)1/2 = (1 ei0 )1/2 (1 eiπ )1/2 = ei0 eiπ/2 = i

293
 θ=π φ=0
θ=−π φ=2π

Figure 9.42: Branch Cuts and Angles for (z 2 − 1)1/2

Thus we see that this choice of angles gives us the desired branch.
Now we go back to the expression
arccos(z) = −i log(z + (z 2 − 1)1/2 ).
We have already seen that there are branch points at z = −1 and z = 1 because of (z 2 − 1)1/2 . Now we must
determine if the logarithm introduces additional branch points. The only possibilities for branch points are where
the argument of the logarithm is zero.
z + (z 2 − 1)1/2 = 0
z2 = z2 − 1
0 = −1
We see that the argument of the logarithm is nonzero and thus there are no additional branch points. Introduce
the variable, w = z + (z 2 − 1)1/2 . What is the image of the branch cuts in the w plane? We parameterize the
branch cut connecting z = 1 and z = +∞ with z = r + 1, r ∈ [0, ∞).
w = r + 1 + ((r + 1)2 − 1)1/2
p
= r + 1 ± r(r + 2)
p
= r(1 ± r 1 + 2/r) + 1

294
 p p
r(1 + 1 + 2/r) + 1 is the interval [1, ∞); r(1 − 1 + 2/r) + 1 is the interval (0, 1]. Thus we see that this branch
cut is mapped to the interval (0, ∞) in the w plane. Similarly, we could show that the branch cut (−∞, −1] in
the z plane is mapped to (−∞, 0) in the w plane. In the w plane there is a branch cut along the real w axis from
−∞ to ∞. Thus cut makes the logarithm single-valued. For the branch of the square root that we chose, all the
points in the z plane get mapped to the upper half of the w plane.
With the branch cuts we have introduced so far and the chosen branch of the square root we have

arccos(0) = −i log(0 + (02 − 1)1/2 )

= −i log i
 π 
= −i i + i2πn
2
π
= + 2πn
2

Choosing the n = 0 branch of the logarithm will give us Arccos (z). We see that we can write

Arccos (z) = −i Log (z + (z 2 − 1)1/2 ).

Solution 9.28
We consider the function f (z) = (z 1/2 − 1)1/2 . First note that z 1/2 has a branch point at z = 0. We place a branch
cut on the negative real axis to make it single valued. f (z) will have a branch point where z 1/2 − 1 = 0. This
occurs at z = 1 on the branch of z 1/2 on which 11/2 = 1. (11/2 has the value 1 on one branch of z 1/2 and −1 on
the other branch.) For this branch we introduce a branch cut connecting z = 1 with the point at infinity. (See
Figure 9.43.)

Solution 9.29
The distance between the end of rod a and the end of rod c is b. In the complex plane, these points are a eiθ and

295
 1/2 1/2
1 =1 1 =-1

Figure 9.43: Branch Cuts for (z 1/2 − 1)1/2

l + c eiφ , respectively. We write this out mathematically.

 
l + c eiφ − a eiθ  = b
l + c eiφ − a eiθ l + c e−iφ − a e−iθ = b2

l2 + cl e−iφ − al e−iθ + cl eiφ + c2 − ac ei(φ−θ) − al eiθ − ac ei(θ−φ) + a2 = b2

1 2
b − a2 − c2 − l2


cl cos φ − ac cos(φ − θ) − al cos θ =
2

This equation relates the two angular positions. One could differentiate the equation to relate the velocities and
accelerations.

Solution 9.30
1. Let w = u + iv. First we do the strip: |<(z)| < 1. Consider the vertical line: z = c + iy, y ∈ R. This line is
mapped to

w = 2(c + iy)2
w = 2c2 − 2y 2 + i4cy
u = 2c2 − 2y 2 , v = 4cy

This is a parabola that opens to the left. For the case c = 0 it is the negative u axis. We can parametrize

296
the curve in terms of v.
1 2
u = 2c2 − v , v∈R
8c2
The boundaries of the region are both mapped to the parabolas:
1
u = 2 − v2, v ∈ R.
8
The image of the mapping is
 
1 2
w = u + iv : v ∈ R and u < 2 − v .
8

Note that the mapping is two-to-one.

Now we do the strip 1 < =(z) < 2. Consider the horizontal line: z = x + ic, x ∈ R. This line is mapped to
w = 2(x + ic)2
w = 2x2 − 2c2 + i4cx
u = 2x2 − 2c2 , v = 4cx
This is a parabola that opens upward. We can parametrize the curve in terms of v.
1 2
u= v − 2c2 , v∈R
8c2
The boundary =(z) = 1 is mapped to
1
u = v 2 − 2, v ∈ R.
8
The boundary =(z) = 2 is mapped to
1 2
u= v − 8, v∈R
32

297
 The image of the mapping is
 
1 2 1 2
w = u + iv : v ∈ R and v −8<u< v −2 .
32 8

2. We write the transformation as

z+1 2
=1+ .
z−1 z−1
Thus we see that the transformation is the sequence:
(a) translation by -1
(b) inversion
(c) magnification by 2
(d) translation by 1
Consider the strip |<(z)| < 1. The translation by −1 maps this to −2 < <(z) < 0. Now we do the
inversion. The left edge, <(z) = 0, is mapped to itself. The right edge, <(z) = −2, is mapped to the circle
|z + 1/4| = 1/4. Thus the current image is the left half plane minus a circle:
 
 1  1
<(z) < 0 and z +  > .

4 4
The magnification by 2 yields
 
 1  1
<(z) < 0 and z + > .
 2 2
The final step is a translation by 1.
 
 1  1
<(z) < 1 and z − > .
 2 2

298
 Now consider the strip 1 < =(z) < 2. The translation by −1 does not change the domain. Now we do the
inversion. The bottom edge, =(z) = 1, is mapped to the circle |z + i/2| = 1/2. The top edge, =(z) = 2, is
mapped to the circle |z + i/4| = 1/4. Thus the current image is the region between two circles:
   
 i  1
z +  <
 i  1
and z +  > .
 2  2 4 4
The magnification by 2 yields
 
 i  1
|z + i| < 1 and z + > .
 2 2
The final step is a translation by 1.
 
 i  1
|z − 1 + i| < 1 and z − 1 + > .
 2 2

Solution 9.31
1. There is a simple pole at z = −2. The function has a branch point at z = −1. Since this is the only
branch point in the finite complex plane there is also a branch point at infinity. We can verify this with the
substitution z = 1/ζ.

(1/ζ + 1)1/2
 
1
f =
ζ 1/ζ + 2
ζ (1 + ζ)1/2
1/2
=
1 + 2ζ
Since f (1/ζ) has a branch point at ζ = 0, f (z) has a branch point at infinity.

2. cos z is an entire function with an essential singularity at infinity. Thus f (z) has singularities only where
1/(1 + z) has singularities. 1/(1 + z) has a first order pole at z = −1. It is analytic everywhere else,

299
 including the point at infinity. Thus we conclude that f (z) has an essential singularity at z = −1 and is
analytic elsewhere. To explicitly show that z = −1 is an essential singularity, we can find the Laurent series
expansion of f (z) about z = −1.
∞
(−1)n
  X
1
cos = (z + 1)−2n
1+z n=0
(2n)!

3. 1 − ez has simple zeros at z = i2nπ, n ∈ Z. Thus f (z) has second order poles at those points.
The point at infinity is a non-isolated singularity. To justify this: Note that
1
f (z) =
(1 − ez )2
1
has second order poles at z = i2nπ, n ∈ Z. This means that f (1/ζ) has second order poles at ζ = i2nπ ,
n ∈ Z. These second order poles get arbitrarily close to ζ = 0. There is no deleted neighborhood around
ζ = 0 in which f (1/ζ) is analytic. Thus the point ζ = 0, (z = ∞), is a non-isolated singularity. There is no
Laurent series expansion about the point ζ = 0, (z = ∞).
The point at infinity is neither a branch point nor a removable singularity. It is not a pole either. If it were,
there would be an n such that limz→∞ z −n f (z) = const 6= 0. Since z −n f (z) has second order poles in every
deleted neighborhood of infinity, the above limit does not exist. Thus we conclude that the point at infinity
is an essential singularity.

Solution 9.32
We write sinh z in Cartesian form.

w = sinh z = sinh x cos y + i cosh x sin y = u + iv

Consider the line segment x = c, y ∈ (0 . . . π). Its image is

{sinh c cos y + i cosh c sin y | y ∈ (0 . . . π)}.

300
This is the parametric equation for the upper half of an ellipse. Also note that u and v satisfy the equation for
an ellipse.
u2 v2
+ =1
sinh2 c cosh2 c
The ellipse starts at the point (sinh(c), 0), passes through the point (0, cosh(c)) and ends at (−sinh(c), 0). As c
varies from zero to ∞ or from zero to −∞, the semi-ellipses cover the upper half w plane. Thus the mapping is
2-to-1.
Consider the infinite line y = c, x ∈ (−∞ . . . ∞).Its image is

{sinh x cos c + i cosh x sin c | x ∈ (−∞ . . . ∞)}.

This is the parametric equation for the upper half of a hyperbola. Also note that u and v satisfy the equation for
a hyperbola.
u2 v2
− + =1
cos2 c sin2 c
As c varies from 0 to π/2 or from π/2 to π, the semi-hyperbola cover the upper half w plane. Thus the mapping
is 2-to-1.
We look for branch points of sinh−1 w.

w = sinh z
ez − e−z
w=
2
e2z − 2w ez − 1 = 0
ez = w + (w2 + 1)1/2
z = log w + (w − i)1/2 (w + i)1/2

There are branch points at w = ±i. Since w + (w2 + 1)1/2 is nonzero and finite in the finite complex plane, the
logarithm does not introduce any branch points in the finite plane. Thus the only branch point in the upper

301
half w plane is at w = i. Any branch cut that connects w = i with the boundary of =(w) > 0 will separate the
branches under the inverse mapping.
Consider the line y = π/4. The image under the mapping is the upper half of the hyperbola

2u2 + 2v 2 = 1.

Consider the segment x = 1.The image under the mapping is the upper half of the ellipse

u2 v2
+ = 1.
sinh2 1 cosh2 1

302
Chapter 10

Analytic Functions

Students need encouragement. So if a student gets an answer right, tell them it was a lucky guess. That way,
they develop a good, lucky feeling. 1
-Jack Handey

10.1 Complex Derivatives

Functions of a Real Variable. The derivative of a function of a real variable is
d f (x + ∆x) − f (x)
f (x) = lim .
dx ∆x→0 ∆x
If the limit exists then the function is differentiable at the point x. Note that ∆x can approach zero from above
or below. The limit cannot depend on the direction in which ∆x vanishes.
Consider f (x) = |x|. The function is not differentiable at x = 0 since
|0 + ∆x| − |0|
lim + =1
∆x→0 ∆x
1
Quote slightly modified.

303
and
|0 + ∆x| − |0|
lim − = −1.
∆x→0 ∆x

Analyticity. The complex derivative, (or simply derivative if the context is clear), is defined,
d f (z + ∆z) − f (z)
f (z) = lim .
dz ∆z→0 ∆z
The complex derivative exists if this limit exists. This means that the value of the limit is independent of the
manner in which ∆z → 0. If the complex derivative exists at a point, then we say that the function is complex
differentiable there.
A function of a complex variable is analytic at a point z0 if the complex derivative exists in a neighborhood
about that point. The function is analytic in an open set if it has a complex derivative at each point in that set.
Note that complex differentiable has a different meaning than analytic. Analyticity refers to the behavior of a
function on an open set. A function can be complex differentiable at isolated points, but the function would not
be analytic at those points. Analytic functions are also called regular or holomorphic. If a function is analytic
everywhere in the finite complex plane, it is called entire.

Example 10.1.1 Consider z n , n ∈ Z+ , Is the function differentiable? Is it analytic? What is the value of the
derivative?
We determine differentiability by trying to differentiate the function. We use the limit definition of differenti-
ation. We will use Newton’s binomial formula to expand (z + ∆z)n .
d n (z + ∆z)n − z n
z = lim
dz ∆z→0
 ∆z 
n(n−1) n−2
z n + nz n−1 ∆z + 2
z ∆z 2 + · · · + ∆z n − z n
= lim
∆z→0
 ∆z 
n−1 n(n − 1) n−2 n−1
= lim nz + z ∆z + · · · + ∆z
∆z→0 2
= nz n−1

304
The derivative exists everywhere. The function is analytic in the whole complex plane so it is entire. The value
d
of the derivative is dz = nz n−1 .

Example 10.1.2 We will show that f (z) = z is not differentiable. The derivative is,
d f (z + ∆z) − f (z)
f (z) = lim .
dz ∆z→0 ∆z

d z + ∆z − z
z = lim
dz ∆z→0 ∆z
∆z
= lim
∆z→0 ∆z

If we take ∆z = ∆x, the limit is

∆x
lim = 1.
∆x→0 ∆x

If we take ∆z = i∆y, the limit is

−i∆y
lim = −1.
∆y→0 i∆y

Since the limit depends on the way that ∆z → 0, the function is nowhere differentiable. Thus the function is not
analytic.

Complex Derivatives in Terms of Plane Coordinates. Let z = ζ(ξ, η) be a system of coordinates in the
complex plane. (For example, we could have Cartesian coordinates z = ζ(x, y) = x + iy or polar coordinates
z = ζ(r, θ) = r eiθ ). Let f (z) = ψ(ξ, η) be a complex-valued function. (For example we might have a function

305
in the form ψ(x, y) = u(x, y) + iv(x, y) or ψ(r, θ) = R(r, θ) eiΘ(r,θ) .) If f (z) = ψ(ξ, η) is analytic, its complex
derivative is equal to the derivative in any direction. In particular, it is equal to the derivatives in the coordinate
directions.
 −1
df f (z + ∆z) − f (z) ψ(ξ + ∆ξ, η) − ψ(ξ, η) ∂ζ ∂ψ
= lim = lim ∂ζ
= ,
dz ∆ξ→0,∆η=0 ∆z ∆ξ→0
∂ξ
∆ξ ∂ξ ∂ξ
 −1
df f (z + ∆z) − f (z) ψ(ξ, η + ∆η) − ψ(ξ, η) ∂ζ ∂ψ
= lim = lim ∂ζ
= .
dz ∆ξ=0,∆η→0 ∆z ∆η→0
∂η
∆η ∂η ∂η

Example 10.1.3 Consider the Cartesian coordinates z = x + iy. We write the complex derivative as derivatives
in the coordinate directions for f (z) = ψ(x, y).
 −1
df ∂(x + iy) ∂ψ ∂ψ
= = ,
dz ∂x ∂x ∂x
 −1
df ∂(x + iy) ∂ψ ∂ψ
= = −i .
dz ∂y ∂y ∂y
We write this in operator notation.
d ∂ ∂
= = −i .
dz ∂x ∂y

d n
Example 10.1.4 In Example 10.1.1 we showed that z n , n ∈ Z+ , is an entire function and that dz z = nz n−1 .
Now we corroborate this by calculating the complex derivative in the Cartesian coordinate directions.
d n ∂
z = (x + iy)n
dz ∂x
= n(x + iy)n−1
= nz n−1

306
 d n ∂
z = −i (x + iy)n
dz ∂y
= −i(i)n(x + iy)n−1
= nz n−1

Complex Derivatives are Not the Same as Partial Derivatives Recall from calculus that

∂f ∂g ∂s ∂g ∂t
f (x, y) = g(s, t) ⇒ = +
∂x ∂s ∂x ∂t ∂x
Do not make the mistake of using a similar formula for functions of a complex variable. If f (z) = ψ(x, y) then

df ∂ψ ∂x ∂ψ ∂y
6= + .
dz ∂x ∂z ∂y ∂z
d
This is because the dz operator means “The derivative in any direction in the complex plane.” Since f (z) is
0
analytic, f (z) is the same no matter in which direction we take the derivative.

Rules of Differentiation. For an analytic function defined in terms of z we can calculate the complex derivative
using all the usual rules of differentiation that we know from calculus like the product rule,

d
f (z)g(z) = f 0 (z)g(z) + f (z)g 0 (z),
dz
or the chain rule,

d
f (g(z)) = f 0 (g(z))g 0 (z).
dz
This is because the complex derivative derives its properties from properties of limits, just like its real variable
counterpart.

307
Result 10.1.1 The complex derivative is,
d f (z + ∆z) − f (z)
f (z) = lim .
dz ∆z→0 ∆z
The complex derivative is defined if the limit exists and is independent of the manner
in which ∆z → 0. A function is analytic at a point if the complex derivative exists in a
neighborhood of that point.
Let z = ζ(ξ, η) be coordinates in the complex plane. The complex derivative in the
coordinate directions is
 −1  −1
d ∂ζ ∂ ∂ζ ∂
= = .
dz ∂ξ ∂ξ ∂η ∂η
In Cartesian coordinates, this is
d ∂ ∂
= = −i .
dz ∂x ∂y
In polar coordinates, this is
d ∂ i ∂
= e−iθ = − e−iθ
dz ∂r r ∂θ
Since the complex derivative is defined with the same limit formula as real derivatives,
all the rules from the calculus of functions of a real variable may be used to differentiate
functions of a complex variable.

308
 d n
Example 10.1.5 We have shown that z n , n ∈ Z+ , is an entire function. Now we corroborate that dz
z = nz n−1
by calculating the complex derivative in the polar coordinate directions.
d n ∂
z = e−iθ rn einθ
dz ∂r
= e−iθ nrn−1 einθ
= nrn−1 ei(n−1)θ
= nz n−1

d n i ∂
z = − e−iθ rn einθ
dz r ∂θ
i −iθ n
= − e r in einθ
r
= nrn−1 ei(n−1)θ
= nz n−1

Analytic Functions can be Written in Terms of z. Consider an analytic function expressed in terms of x
and y, ψ(x, y). We can write ψ as a function of z = x + iy and z = x − iy.
 
z+z z−z
f (z, z) = ψ ,
2 i2

We treat z and z as independent variables. We find the partial derivatives with respect to these variables.
 
∂ ∂x ∂ ∂y ∂ 1 ∂ ∂
= + = −i
∂z ∂z ∂x ∂z ∂y 2 ∂x ∂y
 
∂ ∂x ∂ ∂y ∂ 1 ∂ ∂
= + = +i
∂z ∂z ∂x ∂z ∂y 2 ∂x ∂y

309
Since ψ is analytic, the complex derivatives in the x and y directions are equal.
∂ψ ∂ψ
= −i
∂x ∂y
The partial derivative of f (z, z) with respect to z is zero.
 
∂f 1 ∂ψ ∂ψ
= +i =0
∂z 2 ∂x ∂y
Thus f (z, z) has no functional dependence on z, it can be written as a function of z alone.
If we were considering an analytic function expressed in polar coordinates φ(r, θ), then we could write it in
Cartesian coordinates with the substitutions:
p
r = x2 + y 2 , θ = arctan(x, y).
Thus we could write φ(r, θ) as a function of z alone.

Result 10.1.2 Any analytic function ψ(x, y) or φ(r, θ) can be written as a function of z
alone.

10.2 Cauchy-Riemann Equations

If we know that a function is analytic, then we have a convenient way of determining its complex derivative.
We just express the complex derivative in terms of the derivative in a coordinate direction. However, we don’t
have a nice way of determining if a function is analytic. The definition of complex derivative in terms of a limit
is cumbersome to work with. In this section we remedy this problem.
Consider a function f (z) = ψ(x, y). If f (z) is analytic, the complex derivative is equal to the derivatives in
the coordinate directions. We equate the derivatives in the x and y directions to obtain the Cauchy-Riemann
equations in Cartesian coordinates.
ψx = −iψy (10.1)

310
This equation is a necessary condition for the analyticity of f (z).
Let ψ(x, y) = u(x, y) + iv(x, y) where u and v are real-valued functions. We equate the real and imaginary
parts of Equation 10.1 to obtain another form for the Cauchy-Riemann equations in Cartesian coordinates.
u x = vy , uy = −vx .
Note that this is a necessary and not a sufficient condition for analyticity of f (z). That is, u and v may satisfy
the Cauchy-Riemann equations but f (z) may not be analytic. The Cauchy-Riemann equations give us an easy
test for determining if a function is not analytic.

Example 10.2.1 In Example 10.1.2 we showed that z is not analytic using the definition of complex differenti-
ation. Now we obtain the same result using the Cauchy-Riemann equations.
z = x − iy
ux = 1, vy = −1
We see that the first Cauchy-Riemann equation is not satisfied; the function is not analytic at any point.

A sufficient condition for f (z) = ψ(x, y) to be analytic at a point z0 = (x0 , y0 ) is that the partial derivatives
of ψ(x, y) exist and are continuous in some neighborhood of z0 and satisfy the Cauchy-Riemann equations there.
If the partial derivatives of ψ exist and are continuous then
ψ(x + ∆x, y + ∆y) = ψ(x, y) + ∆xψx (x, y) + ∆yψy (x, y) + o(∆x) + o(∆y).
Here the notation o(∆x) means “terms smaller than ∆x”. We calculate the derivative of f (z).
f (z + ∆z) − f (z)
f 0 (z) = lim
∆z→0 ∆z
ψ(x + ∆x, y + ∆y) − ψ(x, y)
= lim
∆x,∆y→0 ∆x + i∆y
ψ(x, y) + ∆xψx (x, y) + ∆yψy (x, y) + o(∆x) + o(∆y) − ψ(x, y)
= lim
∆x,∆y→0 ∆x + i∆y
∆xψx (x, y) + ∆yψy (x, y) + o(∆x) + o(∆y)
= lim .
∆x,∆y→0 ∆x + i∆y

311
Here we use the Cauchy-Riemann equations.

(∆x + i∆y)ψx (x, y) o(∆x) + o(∆y)

= lim + lim .
∆x,∆y→0 ∆x + i∆y ∆x,∆y→0 ∆x + i∆y
= ψx (x, y)

Thus we see that the derivative is well defined.

Cauchy-Riemann Equations in General Coordinates Let z = ζ(ξ, η) be a system of coordinates in the

complex plane. Let ψ(ξ, η) be a function which we write in terms of these coordinates, A necessary condition for
analyticity of ψ(ξ, η) is that the complex derivatives in the coordinate directions exist and are equal. Equating
the derivatives in the ξ and η directions gives us the Cauchy-Riemann equations.

 −1  −1
∂ζ ∂ψ ∂ζ ∂ψ
=
∂ξ ∂ξ ∂η ∂η

We could separate this into two equations by equating the real and imaginary parts or the modulus and argument.

312
 Result 10.2.1 A necessary condition for analyticity of ψ(ξ, η), where z = ζ(ξ, η), at
z = z0 is that the Cauchy-Riemann equations are satisfied in a neighborhood of z = z0 .
 −1  −1
∂ζ ∂ψ ∂ζ ∂ψ
= .
∂ξ ∂ξ ∂η ∂η
(We could equate the real and imaginary parts or the modulus and argument of this to
obtain two equations.) A sufficient condition for analyticity of f (z) is that the Cauchy-
Riemann equations hold and the first partial derivatives of ψ exist and are continuous in
a neighborhood of z = z0 .
Below are the Cauchy-Riemann equations for various forms of f (z).

f (z) = ψ(x, y), ψx = −iψy

f (z) = u(x, y) + iv(x, y), ux = vy , uy = −vx
i
f (z) = ψ(r, θ), ψr = − ψ θ
r
1
f (z) = u(r, θ) + iv(r, θ), ur = vθ , uθ = −rvr
r
R 1
f (z) = R(r, θ) eiΘ(r,θ) , Rr = Θθ , Rθ = −RΘr
r r

Example 10.2.2 Consider the Cauchy-Riemann equations for f (z) = u(r, θ) + iv(r, θ). From Exercise 10.2 we
know that the complex derivative in the polar coordinate directions is

d ∂ i ∂
= e−iθ = − e−iθ .
dz ∂r r ∂θ

313
From Result 10.2.1 we have the equation,
∂ i ∂
e−iθ [u + iv] = − e−iθ [u + iv].
∂r r ∂θ
We multiply by eiθ and equate the real and imaginary components to obtain the Cauchy-Riemann equations.
1
u r = vθ , uθ = −rvr
r

Example 10.2.3 Consider the exponential function.

ez = ψ(x, y) = ex (cos y + i sin(y))

We use the Cauchy-Riemann equations to show that the function is entire.

ψx = −iψy
e (cos y + i sin(y)) = −i ex (− sin y + i cos(y))
x

ex (cos y + i sin(y)) = ex (cos y + i sin(y))

Since the function satisfies the Cauchy-Riemann equations and the first partial derivatives are continuous every-
where in the finite complex plane, the exponential function is entire.
Now we find the value of the complex derivative.
d z ∂ψ
e = = ex (cos y + i sin(y)) = ez
dz ∂x
The differentiability of the exponential function implies the differentiability of the trigonometric functions, as they
can be written in terms of the exponential.

In Exercise 10.11 you can show that the logarithm log z is differentiable for z 6= 0. This implies the differen-
tiability of z α and the inverse trigonometric functions as they can be written in terms of the logarithm.

314
Example 10.2.4 We compute the derivative of z z .

d z d z log z
(z ) = e
dz dz
= (1 + log z) ez log z
= (1 + log z)z z
= z z + z z log z

10.3 Harmonic Functions

A function u is harmonic if its second partial derivatives exist, are continuous and satisfy Laplace’s equation
∆u = 0. 2 (In Cartesian coordinates the Laplacian is ∆u ≡ uxx + uyy .) If f (z) = u + iv is an analytic function
then u and v are harmonic functions. To see why this is so, we start with the Cauchy-Riemann equations.

ux = vy , uy = −vx

We differentiate the first equation with respect to x and the second with respect to y. (We assume that u and v
are twice continuously differentiable. We will see later that they are infinitely differentiable.)

uxx = vxy , uyy = −vyx

Thus we see that u is harmonic.

∆u ≡ uxx + uyy = vxy − vyx = 0

One can use the same method to show that ∆v = 0.

2
The capital Greek letter ∆ is used to denote the Laplacian, like ∆u(x, y), and differentials, like ∆x.

315
 If u is harmonic on some simply-connected domain, then there exists a harmonic function v such that f (z) =
u + iv is analytic in the domain. v is called the harmonic conjugate of u. The harmonic conjugate is unique up
to an additive constant. To demonstrate this, let w be another harmonic conjugate of u. Both the pair u and v
and the pair u and w satisfy the Cauchy-Riemann equations.
ux = vy , uy = −vx , ux = wy , uy = −wx
We take the difference of these equations.
vx − wx = 0, vy − wy = 0
On a simply connected domain, the difference between v and w is thus a constant.
To prove the existence of the harmonic conjugate, we first write v as an integral.
Z (x,y)
v(x, y) = v(x0 , y0 ) + vx dx + vy dy
(x0 ,y0 )

On a simply connected domain, the integral is path independent and defines a unique v in terms of vx and vy .
We use the Cauchy-Riemann equations to write v in terms of ux and uy .
Z (x,y)
v(x, y) = v(x0 , y0 ) + −uy dx + ux dy
(x0 ,y0 )

Changing the starting point (x0 , y0 ) changes v by an additive constant. The harmonic conjugate of u to within
an additive constant is
Z
v(x, y) = −uy dx + ux dy.

This proves the existence 3 of the harmonic conjugate. This is not the formula one would use to construct the
harmonic conjugate of a u. One accomplishes this by solving the Cauchy-Riemann equations.
3
A mathematician returns to his office to find that a cigarette tossed in the trash has started a small fire. Being calm and a
quick thinker he notes that there is a fire extinguisher by the window. He then closes the door and walks away because “the solution
exists.”

316
 Result 10.3.1 If f (z) = u + iv is an analytic function then u and v are harmonic
functions. That is, the Laplacians of u and v vanish ∆u = ∆v = 0. The Laplacian in
Cartesian and polar coordinates is
∂2 ∂2 1 ∂2
 
1 ∂ ∂
∆ = 2 + 2, ∆= r + 2 2.
∂x ∂y r ∂r ∂r r ∂θ
Given a harmonic function u in a simply connected domain, there exists a harmonic
function v, (unique up to an additive constant), such that f (z) = u + iv is analytic in
the domain. One can construct v by solving the Cauchy-Riemann equations.

Example 10.3.1 Is x2 the real part of an analytic function?

The Laplacian of x2 is

∆[x2 ] = 2 + 0

x2 is not harmonic and thus is not the real part of an analytic function.

Example 10.3.2 Show that u = e−x (x sin y − y cos y) is harmonic.

∂u
= e−x sin y − ex (x sin y − y cos y)
∂x
= e−x sin y − x e−x sin y + y e−x cos y

∂2u
= − e−x sin y − e−x sin y + x e−x sin y − y e−x cos y
∂x2
= −2 e−x sin y + x e−x sin y − y e−x cos y

317
 ∂u
= e−x (x cos y − cos y + y sin y)
∂y

∂2u
2
= e−x (−x sin y + sin y + y cos y + sin y)
∂y
= −x e−x sin y + 2 e−x sin y + y e−x cos y
∂2u ∂2u
Thus we see that ∂x2
+ ∂y 2
= 0 and u is harmonic.

Example 10.3.3 Consider u = cos x cosh y. This function is harmonic.

uxx + uyy = − cos x cosh y + cos x cosh y = 0

Thus it is the real part of an analytic function, f (z). We find the harmonic conjugate, v, with the Cauchy-Riemann
equations. We integrate the first Cauchy-Riemann equation.

vy = ux = − sin x cosh y
v = − sin x sinh y + a(x)

Here a(x) is a constant of integration. We substitute this into the second Cauchy-Riemann equation to determine
a(x).

vx = −uy
− cos x sinh y + a0 (x) = − cos x sinh y
a0 (x) = 0
a(x) = c

Here c is a real constant. Thus the harmonic conjugate is

v = − sin x sinh y + c.

318
The analytic function is

f (z) = cos x cosh y − i sin x sinh y + ic

We recognize this as

f (z) = cos z + ic.

Example 10.3.4 Here we consider an example that demonstrates the need for a simply connected domain.
Consider u = Log r in the multiply connected domain, r > 0. u is harmonic.

1 ∂2
 
1 ∂ ∂
∆ Log r = r Log r + 2 2 Log r = 0
r ∂r ∂r r ∂θ

We solve the Cauchy-Riemann equations to try to find the harmonic conjugate.

1
ur = vθ , uθ = −rvr
r
vr = 0, vθ = 1
v =θ+c

We are able to solve for v, but it is multi-valued. Any single-valued branch of θ that we choose will not be
continuous on the domain. Thus there is no harmonic conjugate of u = Log r for the domain r > 0.
If we had instead considered the simply-connected domain r > 0, | arg(z)| < π then the harmonic conjugate
would be v = Arg (z) + c. The corresponding analytic function is f (z) = Log z + ic.

Example 10.3.5 Consider u = x3 − 3xy 2 + x. This function is harmonic.

uxx + uyy = 6x − 6x = 0

319
Thus it is the real part of an analytic function, f (z). We find the harmonic conjugate, v, with the Cauchy-Riemann
equations. We integrate the first Cauchy-Riemann equation.
vy = ux = 3x2 − 3y 2 + 1
v = 3x2 y − y 3 + y + a(x)
Here a(x) is a constant of integration. We substitute this into the second Cauchy-Riemann equation to determine
a(x).
vx = −uy
6xy + a0 (x) = 6xy
a0 (x) = 0
a(x) = c
Here c is a real constant. The harmonic conjugate is
v = 3x2 y − y 3 + y + c.
The analytic function is
f (z) = x3 − 3xy 2 + x + i(3x2 y − y 3 + y) + ic
f (z) = x3 + i3x2 y − 3xy 2 − iy 2 + x + iy + ic
f (z) = z 3 + z + ic

10.4 Singularities
Any point at which a function is not analytic is called a singularity. In this section we will classify the different
flavors of singularities.
Result 10.4.1 Singularities. If a function is not analytic at a point, then that point
is a singular point or a singularity of the function.

320
10.4.1 Categorization of Singularities
Branch Points. If f (z) has a branch point at z0 , then we cannot define a branch of f (z) that is continuous
in a neighborhood of z0 . Continuity is necessary for analyticity. Thus all branch points are singularities. Since
function are discontinuous across branch cuts, all points on a branch cut are singularities.

Example 10.4.1 Consider f (z) = z 3/2√. The origin and infinity are branch points and are thus singularities of
f (z). We choose the branch g(z) = z 3 . All the points on the negative real axis, including the origin, are
singularities of g(z).

Removable Singularities.

Example 10.4.2 Consider

sin z
f (z) = .
z

This function is undefined at z = 0 because f (0) is the indeterminate form 0/0. f (z) is analytic everywhere in
the finite complex plane except z = 0. Note that the limit as z → 0 of f (z) exists.

sin z cos z
lim = lim =1
z→0 z z→0 1

If we were to fill in the hole in the definition of f (z), we could make it differentiable at z = 0. Consider the
function
(
sin z
z
z=6 0,
g(z) =
1 z = 0.

321
We calculate the derivative at z = 0 to verify that g(z) is analytic there.
f (0) − f (z)
f 0 (0) = lim
z→0 z
1 − sin(z)/z
= lim
z→0 z
z − sin(z)
= lim
z→0 z2
1 − cos(z)
= lim
z→0 2z
sin(z)
= lim
z→0 2
=0
We call the point at z = 0 a removable singularity of sin(z)/z because we can remove the singularity by defining
the value of the function to be its limiting value there.

Consider a function f (z) that is analytic in a deleted neighborhood of z = z0 . If f (z) is not analytic at z0 ,
but limz→z0 f (z) exists, then the function has a removable singularity at z0 . The function
(
f (z) z 6= z0
g(z) =
limz→z0 f (z) z = z0
is analytic in a neighborhood of z = z0 . We show this by calculating g 0 (z0 ).
g(z0 ) − g(z)
g 0 (z0 ) = lim
z→z0 z0 − z
−g 0 (z)
= lim
z→z0 −1
0
= lim f (z)
z→z0

This limit exists because f (z) is analytic in a deleted neighborhood of z = z0 .

322
Poles. If a function f (z) behaves like c/(z − z0 )n near z = z0 then the function has an nth order pole at that
point. More mathematically we say
lim (z − z0 )n f (z) = c 6= 0.
z→z0

We require the constant c to be nonzero so we know that it is not a pole of lower order. We can denote a removable
singularity as a pole of order zero.
Another way to say that a function has an nth order pole is that f (z) is not analytic at z = z0 , but (z −z0 )n f (z)
is either analytic or has a removable singularity at that point.

Example 10.4.3 1/ sin(z 2 ) has a second order pole at z = 0 and first order poles at z = (nπ)1/2 , n ∈ Z± .
z2 2z
lim 2
= lim
z→0 sin(z ) z→0 2z cos(z 2 )
2
= lim
z→0 2 cos(z 2 ) − 4z 2 sin(z 2 )

=1

z − (nπ)1/2 1
lim 2
= lim
z→(nπ)1/2 sin(z ) z→(nπ)1/2 2z cos(z 2 )
1
=
2(nπ) (−1)n
1/2

Example 10.4.4 e1/z is singular at z = 0. The function is not analytic as limz→0 e1/z does not exist. We check
if the function has a pole of order n at z = 0.
eζ
lim z n e1/z = lim
z→0 ζ→∞ ζ n

eζ
= lim
ζ→∞ n!

323
Since the limit does not exist for any value of n, the singularity is not a pole. We could say that e1/z is more
singular than any power of 1/z.

Essential Singularities. If a function f (z) is singular at z = z0 , but the singularity is not a branch point, or
a pole, the the point is an essential singularity of the function.

The point at infinity. We can consider the point at infinity z → ∞ by making the change of variables z = 1/ζ
and considering ζ → 0. If f (1/ζ) is analytic at ζ = 0 then f (z) is analytic at infinity. We have encountered
branch points at infinity before (Section 9.6). Assume that f (z) is not analytic at infinity. If limz→∞ f (z) exists
then f (z) has a removable singularity at infinity. If limz→∞ f (z)/z n = c 6= 0 then f (z) has an nth order pole at
infinity.

Result 10.4.2 Categorization of Singularities. Consider a function f (z) that has a

singularity at the point z = z0 . Singularities come in four flavors:
Branch Points. Branch points of multi-valued functions are singularities.
Removable Singularities. If limz→z0 f (z) exists, then z0 is a removable singularity. It
is thus named because the singularity could be removed and thus the function made
analytic at z0 by redefining the value of f (z0 ).
Poles. If limz→z0 (z − z0 )n f (z) = const 6= 0 then f (z) has an nth order pole at z0 .
Essential Singularities. Instead of defining what an essential singularity is, we say
what it is not. If z0 neither a branch point, a removable singularity nor a pole, it is
an essential singularity.

324
 A pole may be called a non-essential singularity. This is because multiplying the function by an integral power
of z − z0 will make the function analytic. Then an essential singularity is a point z0 such that there does not exist
an n such that (z − z0 )n f (z) is analytic there.

10.4.2 Isolated and Non-Isolated Singularities

Result 10.4.3 Isolated and Non-Isolated Singularities. Suppose f (z) has a singu-
larity at z0 . If there exists a deleted neighborhood of z0 containing no singularities then
the point is an isolated singularity. Otherwise it is a non-isolated singularity.

If you don’t like the abstract notion of a deleted neighborhood, you can work with a deleted circular neighbor-
hood. However, this will require the introduction of more math symbols and a Greek letter. z = z0 is an isolated
singularity if there exists a δ > 0 such that there are no singularities in 0 < |z − z0 | < δ.

Example 10.4.5 We classify the singularities of f (z) = z/ sin z.

z has a simple zero at z = 0. sin z has simple zeros at z = nπ. Thus f (z) has a removable singularity at z = 0
and has first order poles at z = nπ for n ∈ Z± . We can corroborate this by taking limits.
z 1
lim f (z) = lim = lim =1
z→0 z→0 sin z z→0 cos z

(z − nπ)z
lim (z − nπ)f (z) = lim
z→nπ z→nπ sin z
2z − nπ
= lim
z→nπ cos z
nπ
=
(−1)n
6= 0

325
 Now to examine the behavior at infinity. There is no neighborhood of infinity that does not contain first order
poles of f (z). (Another way of saying this is that there does not exist an R such that there are no singularities
in R < |z| < ∞.) Thus z = ∞ is a non-isolated singularity.
We could also determine this by setting ζ = 1/z and examining the point ζ = 0. f (1/ζ) has first order poles
at ζ = 1/(nπ) for n ∈ Z \ {0}. These first order poles come arbitrarily close to the point ζ = 0 There is no deleted
neighborhood of ζ = 0 which does not contain singularities. Thus ζ = 0, and hence z = ∞ is a non-isolated
singularity.
The point at infinity is an essential singularity. It is certainly not a branch point or a removable singularity.
It is not a pole, because there is no n such that limz→∞ z −n f (z) = const 6= 0. z −n f (z) has first order poles in any
neighborhood of infinity, so this limit does not exist.

326
10.5 Exercises
Complex Derivatives
Exercise 10.1
Show that if f (z) is analytic and ψ(x, y) = f (z) is twice continuously differentiable then f 0 (z) is analytic.

Exercise 10.2
Find the complex derivative in the coordinate directions for f (z) = ψ(r, θ).
Hint, Solution

Exercise 10.3
Show that the following functions are nowhere analytic by checking where the derivative with respect to z exists.
1. sin x cosh y − i cos x sinh y
2. x2 − y 2 + x + i(2xy − y)
Hint, Solution

Exercise 10.4
f (z) is analytic for all z, (|z| < ∞). f (z1 + z2 ) = f (z1 )f (z2 ) for all z1 and z2 . (This is known as a functional
equation). Prove that f (z) = exp(f 0 (0)z).
Hint, Solution

Cauchy-Riemann Equations
Exercise 10.5
Find the Cauchy-Riemann equations for

f (z) = R(r, θ) eiΘ(r,θ) .

327
Hint, Solution

Exercise 10.6
Let
(
x4/3 y 5/3 +ix5/3 y 4/3
x2 +y 2
for z 6= 0,
f (z) =
0 for z = 0.
Show that the Cauchy-Riemann equations hold at z = 0, but that f is not differentiable at this point.
Hint, Solution

Exercise 10.7
Consider the complex function
x3 (1+i)−y 3 (1−i)
(
x2 +y 2
for z 6= 0,
f (z) = u + iv =
0 for z = 0.
Show that the partial derivatives of u and v with respect to x and y exist at z = 0 and that ux = vy and uy = −vx
there: the Cauchy-Riemann equations are satisfied at z = 0. On the other hand, show that
f (z)
lim
z→0 z

does not exist, that is, f is not complex-differentiable at z = 0.

Hint, Solution

Exercise 10.8
Show that the function
( −4
e−z for z 6= 0,
f (z) =
0 for z = 0.

328
satisfies the Cauchy-Riemann equations everywhere, including at z = 0, but f (z) is not analytic at the origin.
Hint, Solution

Exercise 10.9
1. Show that ez is not analytic.

2. f (z) is an analytic function of z. Show that f (z) = f (z) is also an analytic function of z.
Hint, Solution

Exercise 10.10
1. Determine all points z = x + iy where the following functions are differentiable with respect to z:

x−1 y
(i) x3 + y 3 (ii) 2 2
−i
(x − 1) + y (x − 1)2 + y 2

2. Determine all points z where the functions in part (a) are analytic.

3. Determine which of the following functions v(x, y) are the imaginary part of an analytic function u(x, y) +
iv(x, y). For those that are, compute the real part u(x, y) and re-express the answer as an explicit function
of z = x + iy:

(i) x2 − y 2 (ii) 3x2 y

Hint, Solution

Exercise 10.11
Show that the logarithm log z is differentiable for z 6= 0. Find the derivative of the logarithm.
Hint, Solution

329
Exercise 10.12
Show that the Cauchy-Riemann equations for the analytic function f (z) = u(r, θ) + iv(r, θ) are

ur = vθ /r, uθ = −rvr .

Hint, Solution

Exercise 10.13
w = u + iv is an analytic function of z. φ(x, y) is an arbitrary smooth function of x and y. When expressed in
terms of u and v, φ(x, y) = Φ(u, v). Show that (w0 6= 0)
 −1  
∂Φ ∂Φ dw ∂φ ∂φ
−i = −i .
∂u ∂v dz ∂x ∂y
Deduce
 −2  2
∂ 2 Φ ∂ 2 Φ  dw  ∂ φ ∂2φ

+ =  + .
∂u2 ∂v 2 dz ∂x2 ∂y 2
Hint, Solution

Exercise 10.14 p
Show that the functions defined by f (z) = log |z| + i arg(z) and f (z) = |z| ei arg(z)/2 are analytic in the sector
|z| > 0, | arg(z)| < π. What are the corresponding derivatives df /dz?
Hint, Solution

Exercise 10.15
Show that the following functions are harmonic. For each one of them find its harmonic conjugate and form the
corresponding holomorphic function.
1. u(x, y) = x Log (r) − y arctan(x, y) (r 6= 0)

330
 2. u(x, y) = arg(z) (| arg(z)| < π, r 6= 0)

3. u(x, y) = rn cos(nθ)

4. u(x, y) = y/r2 (r 6= 0)

Hint, Solution

331
10.6 Hints
Complex Derivatives
Hint 10.1
Start with the Cauchy-Riemann equation and then differentiate with respect to x.

Hint 10.2
Read Example 10.1.3 and use Result 10.1.1.

Hint 10.3
Use Result 10.1.1.

Hint 10.4
Take the logarithm of the equation to get a linear equation.

Cauchy-Riemann Equations
Hint 10.5
Use the result of Exercise 10.2.

Hint 10.6
To evaluate ux (0, 0), etc. use the definition of differentiation. Try to find f 0 (z) with the definition of complex
differentiation. Consider ∆z = ∆r eiθ .

Hint 10.7
To evaluate ux (0, 0), etc. use the definition of differentiation. Try to find f 0 (z) with the definition of complex
differentiation. Consider ∆z = ∆r eiθ .

332
Hint 10.8

Hint 10.9
Use the Cauchy-Riemann equations.

Hint 10.10

Hint 10.11

Hint 10.12

Hint 10.13

Hint 10.14

Hint 10.15

333
10.7 Solutions
Complex Derivatives
Solution 10.1
We start with the Cauchy-Riemann equation and then differentiate with respect to x.

ψx = −iψy
ψxx = −iψyx

We interchange the order of differentiation.

(ψx )x = −i(ψx )y
(f 0 )x = −i(f 0 )y

Since f 0 (z) satisfies the Cauchy-Riemann equation and its partial derivatives exist and are continuous, it is
analytic.

Solution 10.2
The complex derivative in the coordinate directions is
−1
∂r eiθ

df ∂ψ ∂ψ
= = e−iθ ,
dz ∂r ∂r ∂r
−1
∂r eiθ

df ∂ψ i ∂ψ
= = − e−iθ .
dz ∂θ ∂θ r ∂θ
We write this in operator notation.
d ∂ i ∂
= e−iθ = − e−iθ
dz ∂r r ∂θ

334
Solution 10.3
1. Consider f (x, y) = sin x cosh y − i cos x sinh y. The derivatives in the x and y directions are
∂f
= cos x cosh y + i sin x sinh y
∂x
∂f
−i = − cos x cosh y − i sin x sinh y
∂y
These derivatives exist and are everywhere continuous. We equate the expressions to get a set of two
equations.
cos x cosh y = − cos x cosh y, sin x sinh y = − sin x sinh y
cos x cosh y = 0, sin x sinh y = 0
 π 
x = + nπ and (x = mπ or y = 0)
2
The function may be differentiable only at the points
π
x= + nπ, y = 0.
2
Thus the function is nowhere analytic.
2. Consider f (x, y) = x2 − y 2 + x + i(2xy − y). The derivatives in the x and y directions are
∂f
= 2x + 1 + i2y
∂x
∂f
−i = i2y + 2x − 1
∂y
These derivatives exist and are everywhere continuous. We equate the expressions to get a set of two
equations.
2x + 1 = 2x − 1, 2y = 2y.
Since this set of equations has no solutions, there are no points at which the function is differentiable. The
function is nowhere analytic.

335
Solution 10.4

f (z1 + z2 ) = f (z1 )f (z2 )

log(f (z1 + z2 )) = log(f (z1 )) + log(f (z2 ))

We define g(z) = log(f (z)).

g(z1 + z2 ) = g(z1 ) + g(z2 )

This is a linear equation which has exactly the solutions:

g(z) = cz.
Thus f (z) has the solutions:
f (z) = ecz ,
where c is any complex constant. We can write this constant in terms of f 0 (0). We differentiate the original
equation with respect to z1 and then substitute z1 = 0.
f 0 (z1 + z2 ) = f 0 (z1 )f (z2 )
f 0 (z2 ) = f 0 (0)f (z2 )
f 0 (z) = f 0 (0)f (z)

We substitute in the form of the solution.

c ecz = f 0 (0) ecz

c = f 0 (0)
Thus we see that
0
f (z) = ef (0)z .

336
Cauchy-Riemann Equations
Solution 10.5
We find the Cauchy-Riemann equations for

f (z) = R(r, θ) eiΘ(r,θ) .

From Exercise 10.2 we know that the complex derivative in the polar coordinate directions is
d ∂ i ∂
= e−iθ = − e−iθ .
dz ∂r r ∂θ
We equate the derivatives in the two directions.
∂  iΘ  i ∂  iΘ 
e−iθ Re = − e−iθ Re
∂r r ∂θ
i
(Rr + iRΘr ) eiΘ = − (Rθ + iRΘθ ) eiΘ
r
We divide by eiΘ and equate the real and imaginary components to obtain the Cauchy-Riemann equations.
R 1
Rr = Θθ , Rθ = −RΘr
r r

Solution 10.6

( (
x4/3 y 5/3 x5/3 y 4/3
x2 +y 2
if z 6= 0, x2 +y 2
if z 6= 0,
u= , v=
0 if z = 0. 0 if z = 0.

The Cauchy-Riemann equations are

u x = vy , uy = −vx .

337
The partial derivatives of u and v at the point x = y = 0 are,

u(∆x, 0) − u(0, 0)
ux (0, 0) = lim
∆x→0 ∆x
0−0
= lim
∆x→0 ∆x
= 0,

v(∆x, 0) − v(0, 0)
vx (0, 0) = lim
∆x→0 ∆x
0−0
= lim
∆x→0 ∆x
= 0,

u(0, ∆y) − u(0, 0)

uy (0, 0) = lim
∆y→0 ∆y
0−0
= lim
∆y→0 ∆y

= 0,

v(0, ∆y) − v(0, 0)

vy (0, 0) = lim
∆y→0 ∆y
0−0
= lim
∆y→0 ∆y

= 0.

Since ux (0, 0) = uy (0, 0) = vx (0, 0) = vy (0, 0) = 0 the Cauchy-Riemann equations are satisfied.

338
 f (z) is not analytic at the point z = 0. We show this by calculating the derivative.
f (∆z) − f (0) f (∆z)
f 0 (0) = lim = lim
∆z→0 ∆z ∆z→0 ∆z
iθ
Let ∆z = ∆r e , that is, we approach the origin at an angle of θ. Then x = ∆r cos θ and y = ∆r sin θ.
f (∆r eiθ )
f 0 (0) = lim
∆r→0 ∆r eiθ
∆r4/3 cos4/3 θ∆r 5/3 sin5/3 θ+i∆r5/3 cos5/3 θ∆r 4/3 sin4/3 θ
∆r 2
= lim
∆r→0 ∆r eiθ
4/3 5/3 5/3 4/3
cos θ sin
θ + i cos θ sin θ
= lim
∆r→0 eiθ
The value of the limit depends on θ and is not a constant. Thus this limit does not exist. The function is not
differentiable at z = 0.

Solution 10.7

( (
x3 −y 3 x3 +y 3
x2 +y 2
for z 6= 0, x2 +y 2
for z 6= 0,
u= , v=
0 for z = 0. 0 for z = 0.
The Cauchy-Riemann equations are
u x = vy , uy = −vx .
The partial derivatives of u and v at the point x = y = 0 are,
u(∆x, 0) − u(0, 0)
ux (0, 0) = lim
∆x→0 ∆x
∆x − 0
= lim
∆x→0 ∆x
= 1,

339
 v(∆x, 0) − v(0, 0)
vx (0, 0) = lim
∆x→0 ∆x
∆x − 0
= lim
∆x→0 ∆x
= 1,

u(0, ∆y) − u(0, 0)

uy (0, 0) = lim
∆y→0 ∆y
−∆y − 0
= lim
∆y→0 ∆y
= −1,

v(0, ∆y) − v(0, 0)

vy (0, 0) = lim
∆y→0 ∆y
∆y − 0
= lim
∆y→0 ∆y
= 1.
We see that the Cauchy-Riemann equations are satisfied at x = y = 0
f (z) is not analytic at the point z = 0. We show this by calculating the derivative.
f (∆z) − f (0) f (∆z)
f 0 (0) = lim = lim
∆z→0 ∆z ∆z→0 ∆z
iθ
Let ∆z = ∆r e , that is, we approach the origin at an angle of θ. Then x = ∆r cos θ and y = ∆r sin θ.
f (∆r eiθ )
f 0 (0) = lim
∆r→0 ∆r eiθ
∆r 3 cos3 θ(1+i)−∆r3 sin3 θ(1−i)
∆r 2
= lim
∆r→0 ∆r eiθ
3 3
cos θ(1 + i) − sin θ(1 − i)
= lim
∆r→0 eiθ

340
The value of the limit depends on θ and is not a constant. Thus this limit does not exist. The function is not
differentiable at z = 0. Recall that satisfying the Cauchy-Riemann equations is a necessary, but not a sufficient
condition for differentiability.

Solution 10.8
First we verify that the Cauchy-Riemann equations are satisfied for z 6= 0. Note that the form
fx = −ify
will be far more convenient than the form
ux = vy , uy = −vx
for this problem.
−4
fx = 4(x + iy)−5 e−(x+iy)
−4 −4
−ify = −i4(x + iy)−5 i e−(x+iy) = 4(x + iy)−5 e−(x+iy)
The Cauchy-Riemann equations are satisfied for z 6= 0.
Now we consider the point z = 0.
f (∆x, 0) − f (0, 0)
fx (0, 0) = lim
∆x→0 ∆x
−∆x−4
e
= lim
∆x→0 ∆x
=0

f (0, ∆y) − f (0, 0)

−ify (0, 0) = −i lim
∆y→0 ∆y
−4
e−∆y
= −i lim
∆y→0 ∆y
=0

341
The Cauchy-Riemann equations are satisfied for z = 0.
f (z) is not analytic at the point z = 0. We show this by calculating the derivative.
f (∆z) − f (0) f (∆z)
f 0 (0) = lim = lim
∆z→0 ∆z ∆z→0 ∆z
iθ
Let ∆z = ∆r e , that is, we approach the origin at an angle of θ.
f (∆r eiθ )
f 0 (0) = lim
∆r→0 ∆r eiθ
−4 −i4θ
e−r e
= lim
∆r→0 ∆r eiθ
For most values of θ the limit does not exist. Consider θ = π/4.
−4
er
f 0 (0) = lim =∞
∆r→0 ∆r eiπ/4

Because the limit does not exist, the function is not differentiable at z = 0. Recall that satisfying the Cauchy-
Riemann equations is a necessary, but not a sufficient condition for differentiability.

Solution 10.9
1. A necessary condition for analyticity in an open set is that the Cauchy-Riemann equations are satisfied in
that set. We write ez in Cartesian form.
ez = ex−iy = ex cos y − i ex sin y.
Now we determine where u = ex cos y and v = − ex sin y satisfy the Cauchy-Riemann equations.
ux = vy , uy = −vx
x x
e cos y = − e cos y, − ex sin y = ex sin y
cos y = 0, sin y = 0
π
y = + πm, y = πn
2
Thus we see that the Cauchy-Riemann equations are not satisfied anywhere. ez is nowhere analytic.

342
 2. Since f (z) = u + iv is analytic, u and v satisfy the Cauchy-Riemann equations and their first partial
derivatives are continuous.
f (z) = f (z) = u(x, −y) + iv(x, −y) = u(x, −y) − iv(x, −y)
We define f (z) ≡ µ(x, y) + iν(x, y) = u(x, −y) − iv(x, y). Now we see if µ and ν satisfy the Cauchy-Riemann
equations.
µx = νy , µy = −νx
(u(x, −y))x = (−v(x, −y))y , (u(x, −y))y = −(−v(x, −y))x
ux (x, −y) = vy (x, −y), −uy (x, −y) = vx (x, −y)
ux = vy , uy = −vx
Thus we see that the Cauchy-Riemann equations for µ and ν are satisfied if and only if the Cauchy-Riemann
equations for u and v are satisfied. The continuity of the first partial derivatives of u and v implies the same
of µ and ν. Thus f (z) is analytic.

Solution 10.10
1. The necessary condition for a function f (z) = u + iv to be differentiable at a point is that the Cauchy-
Riemann equations hold and the first partial derivatives of u and v are continuous at that point.
(a)
f (z) = x3 + y 3 + i0
The Cauchy-Riemann equations are
ux = vy and uy = −vx
3x2 = 0 and 3y 2 = 0
x = 0 and y = 0
The first partial derivatives are continuous. Thus we see that the function is differentiable only at the
point z = 0.

343
 (b)
x−1 y
f (z) = 2 2
−i
(x − 1) + y (x − 1)2 + y 2
The Cauchy-Riemann equations are
ux = vy and uy = −vx
−(x − 1)2 + y 2 −(x − 1)2 + y 2 2(x − 1)y 2(x − 1)y
2 2 2
= 2 2 2
and 2 2 2
=
((x − 1) + y ) ((x − 1) + y ) ((x − 1) + y ) ((x − 1)2 + y 2 )2
The Cauchy-Riemann equations are each identities. The first partial derivatives are continuous every-
where except the point x = 1, y = 0. Thus the function is differentiable everywhere except z = 1.
2. (a) The function is not differentiable in any open set. Thus the function is nowhere analytic.
(b) The function is differentiable everywhere except z = 1. Thus the function is analytic everywhere except
z = 1.
3. (a) First we determine if the function is harmonic.
v = x2 − y 2
vxx + vyy = 0
2−2=0
The function is harmonic in the complex plane and this is the imaginary part of some analytic function.
By inspection, we see that this function is
iz 2 + c = −2xy + c + i(x2 − y 2 ),
where c is a real constant. We can also find the function by solving the Cauchy-Riemann equations.
ux = vy and uy = −vx
ux = −2y and uy = −2x

344
 We integrate the first equation.

u = −2xy + g(y)

Here g(y) is a function of integration. We substitute this into the second Cauchy-Riemann equation
to determine g(y).

uy = −2x
−2x + g 0 (y) = −2x
g 0 (y) = 0
g(y) = c
u = −2xy + c
f (z) = −2xy + c + i(x2 − y 2 )
f (z) = iz 2 + c
(b) First we determine if the function is harmonic.
v = 3x2 y
vxx + vyy = 6y
The function is not harmonic. It is not the imaginary part of some analytic function.

Solution 10.11
We show that the logarithm log z = ψ(r, θ) = Log r + iθ satisfies the Cauchy-Riemann equations.
i
ψr = − ψθ
r
1 i
=− i
r r
1 1
=
r r

345
Since the logarithm satisfies the Cauchy-Riemann equations and the first partial derivatives are continuous for
z 6= 0, the logarithm is analytic for z 6= 0.
Now we compute the derivative.
d ∂
log z = e−iθ ( Log r + iθ)
dz ∂r
1
= e−iθ
r
1
=
z

Solution 10.12
The complex derivative in the coordinate directions is

d ∂ i ∂
= e−iθ = − e−iθ .
dz ∂r r ∂θ
We substitute f = u + iv into this identity to obtain the Cauchy-Riemann equation in polar coordinates.
∂f i ∂f
e−iθ = − e−iθ
∂r r ∂θ
∂f i ∂f
=−
∂r r ∂θ
i
ur + ivr = − (uθ + ivθ )
r
We equate the real and imaginary parts.

1 1
u r = vθ , vr = − uθ
r r
1
ur = vθ , uθ = −rvr
r

346
Solution 10.13
Since w is analytic, u and v satisfy the Cauchy-Riemann equations,

ux = vy and uy = −vx .

Using the chain rule we can write the derivatives with respect to x and y in terms of u and v.
∂ ∂ ∂
= ux + vx
∂x ∂u ∂v
∂ ∂ ∂
= uy + vy
∂y ∂u ∂v
Now we examine φx − iφy .

φx − iφy = ux Φu + vx Φv − i(uy Φu + vy Φv )
φx − iφy = (ux − iuy )Φu + (vx − ivy )Φv
φx − iφy = (ux − iuy )Φu − i(vy + ivx )Φv

We use the Cauchy-Riemann equations to write uy and vy in terms of ux and vx .

φx − iφy = (ux + ivx )Φu − i(ux + ivx )Φv

Recall that w0 = ux + ivx = vy − iuy .

dw
φx − iφy = (Φu − iΦv )
dz

Thus we see that,

 −1  
∂Φ ∂Φ dw ∂φ ∂φ
−i = −i .
∂u ∂v dz ∂x ∂y

347
 We write this in operator notation.
 −1  
∂ ∂ dw ∂ ∂
−i = −i
∂u ∂v dz ∂x ∂y

The complex conjugate of this relation is

 −1  
∂ ∂ dw ∂ ∂
+i = +i
∂u ∂v dz ∂x ∂y

Now we apply both these operators to Φ = φ.

    −1   −1  
∂ ∂ ∂ ∂ dw ∂ ∂ dw ∂ ∂
+i −i Φ= +i −i φ
∂u ∂v ∂u ∂v dz ∂x ∂y dz ∂x ∂y

∂2 ∂2 ∂2 ∂2
 
+i −i + Φ
∂u2 ∂u∂v ∂v∂u ∂v 2
 −1 "    −1 !    −1   #
dw ∂ ∂ dw ∂ ∂ dw ∂ ∂ ∂ ∂
= +i −i + +i −i φ
dz ∂x ∂y dz ∂x ∂y dz ∂x ∂y ∂x ∂y

(w0 )−1 is an analytic function. Recall that for analytic functions f , f 0 = fx = −ify . So that fx + ify = 0.
−1 "
−1  2 #
∂2Φ ∂2Φ ∂2

dw dw ∂
+ = + φ
∂u2 ∂v 2 dz dz ∂x2 ∂y 2
 −2  2
∂ 2 Φ ∂ 2 Φ  dw  ∂ φ ∂2φ

+ =  +
∂u2 ∂v 2 dz ∂x2 ∂y 2

348
Solution 10.14
1. We consider

f (z) = log |z| + i arg(z) = log r + iθ.

The Cauchy-Riemann equations in polar coordinates are

1
ur = vθ , uθ = −rvr .
r
We calculate the derivatives.
1 1 1
ur = , vθ =
r r r
uθ = 0, −rvr = 0

Since the Cauchy-Riemann equations are satisfied and the partial derivatives are continuous, f (z) is analytic
in |z| > 0, | arg(z)| < π. The complex derivative in terms of polar coordinates is
d ∂ i ∂
= e−iθ = − e−iθ .
dz ∂r r ∂θ
We use this to differentiate f (z).
df ∂ 1 1
= e−iθ [log r + iθ] = e−iθ =
dz ∂r r z

2. Next we consider
p √ iθ/2
f (z) = |z| ei arg(z)/2 = re .

The Cauchy-Riemann equations for polar coordinates and the polar form f (z) = R(r, θ) eiΘ(r,θ) are
R 1
Rr = Θθ , Rθ = −RΘr .
r r

349
 √
We calculate the derivatives for R = r, Θ = θ/2.
1 R 1
Rr = √ , Θθ = √
2 r r 2 r
1
Rθ = 0, −RΘr = 0
r
Since the Cauchy-Riemann equations are satisfied and the partial derivatives are continuous, f (z) is analytic
in |z| > 0, | arg(z)| < π. The complex derivative in terms of polar coordinates is
d ∂ i ∂
= e−iθ = − e−iθ .
dz ∂r r ∂θ
We use this to differentiate f (z).
df ∂ √ 1 1
= e−iθ [ r eiθ/2 ] = iθ/2 √ = √
dz ∂r 2e r 2 z

Solution 10.15
1. We consider the function

u = x Log r − y arctan(x, y) = r cos θ Log r − rθ sin θ

We compute the Laplacian.

1 ∂2u
 
1 ∂ ∂u
∆u = r + 2 2
r ∂r ∂r r ∂θ
1 ∂ 1
= (cos θ(r + r Log r) − θ sin θ) + 2 (r(θ sin θ − 2 cos θ) − r cos θ Log r)
r ∂r r
1 1
= (2 cos θ + cos θ Log r − θ sin θ) + (θ sin θ − 2 cos θ − cos θ Log r)
r r
=0

350
The function u is harmonic. We find the harmonic conjugate v by solving the Cauchy-Riemann equations.

1
vr = − uθ , vθ = rur
r
vr = sin θ(1 + Log r) + θ cos θ, vθ = r(cos θ(1 + Log r) − θ sin θ)

We integrate the first equation with respect to r to determine v to within the constant of integration g(θ).

v = r(sin θ Log r + θ cos θ) + g(θ)

We differentiate this expression with respect to θ.

vθ = r(cos θ(1 + Log r) − θ sin θ) + g 0 (θ)

We compare this to the second Cauchy-Riemann equation to see that g 0 (θ) = 0. Thus g(θ) = c. We have
determined the harmonic conjugate.

v = r(sin θ Log r + θ cos θ) + c

The corresponding analytic function is

f (z) = r cos θ Log r − rθ sin θ + i(r sin θ Log r + rθ cos θ + c).

On the positive real axis, (θ = 0), the function has the value

f (z = r) = r Log r + ic.

We use analytic continuation to determine the function in the complex plane.

f (z) = z log z + ic

351
2. We consider the function

u = Arg (z) = θ.

We compute the Laplacian.

1 ∂2u
 
1 ∂ ∂u
∆u = r + 2 2 =0
r ∂r ∂r r ∂θ

The function u is harmonic. We find the harmonic conjugate v by solving the Cauchy-Riemann equations.

1
vr = − uθ , vθ = rur
r
1
vr = − , vθ = 0
r
We integrate the first equation with respect to r to determine v to within the constant of integration g(θ).

v = − Log r + g(θ)

We differentiate this expression with respect to θ.

vθ = g 0 (θ)

We compare this to the second Cauchy-Riemann equation to see that g 0 (θ) = 0. Thus g(θ) = c. We have
determined the harmonic conjugate.

v = − Log r + c

The corresponding analytic function is

f (z) = θ − i Log r + ic

352
 On the positive real axis, (θ = 0), the function has the value
f (z = r) = −i Log r + ic
We use analytic continuation to determine the function in the complex plane.

f (z) = −i log z + ic

3. We consider the function

u = rn cos(nθ)
We compute the Laplacian.
1 ∂2u
 
1 ∂ ∂u
∆u = r + 2 2
r ∂r ∂r r ∂θ
1 ∂
= (nrn cos(nθ)) − n2 rn−2 cos(nθ)
r ∂r
= n2 rn−2 cos(nθ) − n2 rn−2 cos(nθ)
=0
The function u is harmonic. We find the harmonic conjugate v by solving the Cauchy-Riemann equations.
1
vr = − uθ , vθ = rur
r
vr = nrn−1 sin(nθ), vθ = nrn cos(nθ)
We integrate the first equation with respect to r to determine v to within the constant of integration g(θ).
v = rn sin(nθ) + g(θ)
We differentiate this expression with respect to θ.
vθ = nrn cos(nθ) + g 0 (θ)

353
 We compare this to the second Cauchy-Riemann equation to see that g 0 (θ) = 0. Thus g(θ) = c. We have
determined the harmonic conjugate.

v = rn sin(nθ) + c

The corresponding analytic function is

f (z) = rn cos(nθ) + irn sin(nθ) + ic

On the positive real axis, (θ = 0), the function has the value

f (z = r) = rn + ic

We use analytic continuation to determine the function in the complex plane.

f (z) = z n

4. We consider the function

y sin θ
u= 2
=
r r
We compute the Laplacian.

1 ∂2u
 
1 ∂ ∂u
∆u = r + 2 2
r ∂r ∂r r ∂θ
 
1 ∂ sin θ sin θ
= − − 3
r ∂r r r
sin θ sin θ
= 3 − 3
r r
=0

354
The function u is harmonic. We find the harmonic conjugate v by solving the Cauchy-Riemann equations.
1
vr = − uθ , vθ = rur
r
cos θ sin θ
vr = − 2 , v θ = −
r r
We integrate the first equation with respect to r to determine v to within the constant of integration g(θ).
cos θ
v= + g(θ)
r
We differentiate this expression with respect to θ.
sin θ
vθ = − + g 0 (θ)
r
We compare this to the second Cauchy-Riemann equation to see that g 0 (θ) = 0. Thus g(θ) = c. We have
determined the harmonic conjugate.
cos θ
v= +c
r
The corresponding analytic function is
sin θ cos θ
f (z) = +i + ic
r r
On the positive real axis, (θ = 0), the function has the value
i
f (z = r) =+ ic.
r
We use analytic continuation to determine the function in the complex plane.
i
f (z) = + ic
z

355
Chapter 11

Analytic Continuation

I’m about two beers away from fine.

11.1 Analytic Continuation

Suppose there is a function, f1 (z) that is analytic in the domain D1 and another analytic function, f2 (z) that
is analytic in the domain D2 . (See Figure 11.1.)
If the two domains overlap and f1 (z) = f2 (z) in the overlap region D1 ∩ D2 , then f2 (z) is called an analytic
continuation of f1 (z). This is an appropriate name since f2 (z) continues the definition of f1 (z) outside of its
original domain of definition D1 . We can define a function f (z) that is analytic in the union of the domains
D1 ∪ D2 . On the domain D1 we have f (z) = f1 (z) and f (z) = f2 (z) on D2 . f1 (z) and f2 (z) are called function
elements. There is an analytic continuation even if the two domains only share an arc and not a two dimensional
region.

With more overlapping domains D3 , D4 , . . . we could perhaps extend f1 (z) to more of the complex plane.
Sometimes it is impossible to extend a function beyond the boundary of a domain. This is known as a natural
boundary. If a function f1 (z) is analytically continued to a domain Dn along two different paths, (See Figure 11.2.),

356
 Im(z)

D1
D2
Re(z)

Figure 11.1: Overlapping Domains

then the two analytic continuations are identical as long as the paths do not enclose a branch point of the function.
This is the uniqueness theorem of analytic continuation.

Dn
D1

Figure 11.2: Two Paths of Analytic Continuation

Consider an analytic function f (z) defined in the domain D. Suppose that f (z) = 0 on the arc AB, (see
Figure 11.3.) Then f (z) = 0 in all of D.
Consider a point ζ on AB. The Taylor series expansion of f (z) about the point z = ζ converges in a circle C
at least up to the boundary of D. The derivative of f (z) at the point z = ζ is
f (ζ + ∆z) − f (ζ)
f 0 (ζ) = lim
∆z→0 ∆z

357
 D
C
B
A ζ

Figure 11.3: Domain Containing Arc Along Which f (z) Vanishes

If ∆z is in the direction of the arc, then f 0 (ζ) vanishes as well as all higher derivatives, f 0 (ζ) = f 00 (ζ) = f 000 (ζ) =
· · · = 0. Thus we see that f (z) = 0 inside C. By taking Taylor series expansions about points on AB or inside of
C we see that f (z) = 0 in D.

Result 11.1.1 Let f1 (z) and f2 (z) be analytic functions defined in D. If f1 (z) = f2 (z)
for the points in a region or on an arc in D, then f1 (z) = f2 (z) for all points in D.
To prove Result 11.1.1, we define the analytic function g(z) = f1 (z) − f2 (z). Since g(z) vanishes in the region
or on the arc, then g(z) = 0 and hence f1 (z) = f2 (z) for all points in D.

Result 11.1.2 Consider analytic functions f1 (z) and f2 (z) defined on the domains D1
and D2 , respectively. Suppose that D1 ∩ D2 is a region or an arc and that f1 (z) = f2 (z)
for all z ∈ D1 ∩ D2 . (See Figure 11.4.) Then the function
(
f1 (z) for z ∈ D1 ,
f (z) =
f2 (z) for z ∈ D2 ,

is analytic in D1 ∪ D2 .
Result 11.1.2 follows directly from Result 11.1.1.

358
 D1 D2 D1 D2

Figure 11.4: Domains that Intersect in a Region or an Arc

11.2 Analytic Continuation of Sums

Example 11.2.1 Consider the function
∞
X
f1 (z) = zn.
n=0

The sum converges uniformly for D1 = |z| ≤ r < 1. Since the derivative also converges in this domain, the
function is analytic there.

Re(z) Re(z)

D2
D1
Im(z) Im(z)

P∞
Figure 11.5: Domain of Convergence for n=0 zn.

359
 Now consider the function
1
f2 (z) = .
1−z
This function is analytic everywhere except the point z = 1. On the domain D1 ,
∞
1 X
f2 (z) = = z n = f1 (z)
1−z n=0

Analytic continuation tells us that there is a function that is analytic on the union of the two domains. Here,
the domain is the entire z plane except the point z = 1 and the function is
1
f (z) = .
1−z
1
P∞
1−z
is said to be an analytic continuation of n=0 zn.

11.3 Analytic Functions Defined in Terms of Real Variables

Result 11.3.1 An analytic function, u(x, y) + iv(x, y) can be written in terms of a

function of a complex variable, f (z) = u(x, y) + iv(x, y).
Result 11.3.1 is proved in Exercise 11.1.

Example 11.3.1

f (z) = cosh y sin x(x ex cos y − y ex sin y) − cos x sinh y(y ex cos y + x ex sin y)
+ i cosh y sin x(y ex cos y + x ex sin y) + cos x sinh y(x ex cos y − y ex sin y)
 

360
is an analytic function. Express f (z) in terms of z.
On the real line, y = 0, f (z) is

f (z = x) = x ex sin x

(Recall that cos(0) = cosh(0) = 1 and sin(0) = sinh(0) = 0.)

The analytic continuation of f (z) into the complex plane is

f (z) = z ez sin z.

Alternatively, for x = 0 we have

f (z = iy) = y sinh y(cos y − i sin y).

The analytic continuation from the imaginary axis to the complex plane is

f (z) = −iz sinh(−iz)(cos(−iz) − i sin(−iz))

= iz sinh(iz)(cos(iz) + i sin(iz))
= z sin z ez .

Example 11.3.2 Consider u = e−x (x sin y − y cos y). Find v such that f (z) = u + iv is analytic.
From the Cauchy-Riemann equations,

∂v ∂u
= = e−x sin y − x e−x sin y + y e−x cos y
∂y ∂x

∂v ∂u
=− = e−x cos y − x e−x cos y − y e−x sin y
∂x ∂y

361
Integrate the first equation with respect to y.
v = − e−x cos y + x e−x cos y + e−x (y sin y + cos y) + F (x)
= y e−x sin y + x e−x cos y + F (x)
F (x) is an arbitrary function of x. Substitute this expression for v into the equation for ∂v/∂x.
−y e−x sin y − x e−x cos y + e−x cos y + F 0 (x) = −y e−x sin y − x e−x cos y + e−x cos y
Thus F 0 (x) = 0 and F (x) = c.
v = e−x (y sin y + x cos y) + c

Example 11.3.3 Find f (z) in the previous example. (Up to the additive constant.)

Method 1
f (z) = u + iv
= e−x (x sin y − y cos y) + i e−x (y sin y + x cos y)
e − e−iy e + e−iy e − e−iy e + e−iy
  iy   iy    iy   iy 
−x −x
= e x −y + ie y +x
2i 2 2i 2
= i(x + iy) e−(x+iy)
= iz e−z

Method 2 f (z) = f (x + iy) = u(x, y) + iv(x, y) is an analytic function.

On the real axis, y = 0, f (z) is
f (z = x) = u(x, 0) + iv(x, 0)
= e−x (x sin 0 − 0 cos 0) + i e−x (0 sin 0 + x cos 0)
= ix e−x

362
Suppose there is an analytic continuation of f (z) into the complex plane. If such a continuation, f (z), exists,
then it must be equal to f (z = x) on the real axis An obvious choice for the analytic continuation is

f (z) = u(z, 0) + iv(z, 0)

since this is clearly equal to u(x, 0) + iv(x, 0) when z is real. Thus we obtain

f (z) = iz e−z

Example 11.3.4 Consider f (z) = u(x, y) + iv(x, y). Show that f 0 (z) = ux (z, 0) − iuy (z, 0).

f 0 (z) = ux + ivx
= ux − iuy

f 0 (z) is an analytic function. On the real axis, z = x, f 0 (z) is

f 0 (z = x) = ux (x, 0) − iuy (x, 0)

Now f 0 (z = x) is defined on the real line. An analytic continuation of f 0 (z = x) into the complex plane is

f 0 (z) = ux (z, 0) − iuy (z, 0).

Example 11.3.5 Again consider the problem of finding f (z) given that u(x, y) = e−x (x sin y − y cos y). Now we
can use the result of the previous example to do this problem.

∂u
ux (x, y) = = e−x sin y − x e−x sin y + y e−x cos y
∂x
∂u
uy (x, y) = = x e−x cos y + y e−x sin y − e−x cos y
∂y

363
 f 0 (z) = ux (z, 0) − iuy (z, 0)
= 0 − i(z e−z − e−z )
= i(−z e−z + e−z )

Integration yields the result

f (z) = iz e−z + c

Example 11.3.6 Find f (z) given that

u(x, y) = cos x cosh2 y sin x + cos x sin x sinh2 y

v(x, y) = cos2 x cosh y sinh y − cosh y sin2 x sinh y

f (z) = u(x, y) + v(x, y) is an analytic function. On the real line, f (z) is

f (z = x) = u(x, 0) + iv(x, 0)
= cos x cosh2 0 sin x + cos x sin x sinh2 0 + i(cos2 x cosh 0 sinh 0 − cosh 0 sin2 x sinh 0)
= cos x sin x

Now we know the definition of f (z) on the real line. We would like to find an analytic continuation of f (z) into
the complex plane. An obvious choice for f (z) is

f (z) = cos z sin z

Using trig identities we can write this as

sin 2z
f (z) = .
2

364
Example 11.3.7 Find f (z) given only that

u(x, y) = cos x cosh2 y sin x + cos x sin x sinh2 y.

Recall that

f 0 (z) = ux + ivx
= ux − iuy

Differentiating u(x, y),

ux = cos2 x cosh2 y − cosh2 y sin2 x + cos2 x sinh2 y − sin2 x sinh2 y

uy = 4 cos x cosh y sin x sinh y

f 0 (z) is an analytic function. On the real axis, f 0 (z) is

f 0 (z = x) = cos2 x − sin2 x

Using trig identities we can write this as

f 0 (z = x) = cos(2x)

Now we find an analytic continuation of f 0 (z = x) into the complex plane.

f 0 (z) = cos(2z)

Integration yields the result

sin(2z)
f (z) = +c
2

365
11.3.1 Polar Coordinates
Example 11.3.8 Is


u(r, θ) = r log r cos θ − θ sin θ
the real part of an analytic function?
The Laplacian in polar coordinates is
1 ∂2φ
 
1 ∂ ∂φ
∆φ = r + 2 2.
r ∂r ∂r r ∂θ
Calculating the partial derivatives of u,
∂u
= cos θ + log r cos θ − θ sin θ
∂r
∂u
r = r cos θ + r log r cos θ − rθ sin θ
 ∂r

∂ ∂u
r = 2 cos θ + log r cos θ − θ sin θ
∂r ∂r
 
1 ∂ ∂u 1

r = 2 cos θ + log r cos θ − θ sin θ
r ∂r ∂r r
∂u

= −r θ cos θ + sin θ + log r sin θ
∂θ
∂2u

2
= r − 2 cos θ − log r cos θ + θ sin θ
∂θ
1 ∂2u 1

= − 2 cos θ − log r cos θ + θ sin θ
r2 ∂θ2 r
From the above we see that
1 ∂2u
 
1 ∂ ∂u
∆u = r + 2 2 = 0.
r ∂r ∂r r ∂θ
Therefore u is harmonic and is the real part of some analytic function.

366
Example 11.3.9 Find an analytic function f (z) whose real part is


u(r, θ) = r log r cos θ − θ sin θ .
Let f (z) = u(r, θ) + iv(r, θ). The Cauchy-Riemann equations are
vθ
ur = , uθ = −rvr .
r
Using the partial derivatives in the above example, we obtain two partial differential equations for v(r, θ).
uθ
vr = − = θ cos θ + sin θ + log r sin θ
r

vθ = rur = r cos θ + log r cos θ − θ sin θ
Integrating the equation for vθ yields


v = r θ cos θ + log r sin θ + F (r)
where F (r) is a constant of integration.
Substituting our expression for v into the equation for vr yields
θ cos θ + log r sin θ + sin θ + F 0 (r) = θ cos θ + sin θ + log r sin θ
F 0 (r) = 0
F (r) = const
Thus we see that
f (z) = u + iv



= r log r cos θ − θ sin θ + ir θ cos θ + log r sin θ + const
f (z) is an analytic function. On the line θ = 0, f (z) is



f (z = r) = r log r + ir 0 + const
= r log r + const

367
The analytic continuation into the complex plane is

f (z) = z log z + const

Example 11.3.10 Find the formula in polar coordinates that is analogous to

f 0 (z) = ux (z, 0) − iuy (z, 0).

We know that
df ∂f
= e−iθ .
dz ∂r
If f (z) = u(r, θ) + iv(r, θ) then
df
= e−iθ (ur + ivr )
dz
From the Cauchy-Riemann equations, we have vr = −uθ /r.
df −iθ
 uθ 
= e ur − i
dz r
f 0 (z) is an analytic function. On the line θ = 0, f (z) is

uθ (r, 0)
f 0 (z = r) = ur (r, 0) − i
r
The analytic continuation of f 0 (z) into the complex plane is

i
f 0 (z) = ur (z, 0) − uθ (z, 0).
r

368
Example 11.3.11 Find an analytic function f (z) whose real part is


u(r, θ) = r log r cos θ − θ sin θ .

ur (r, θ) = (log r cos θ − θ sin θ) + cos θ

uθ (r, θ) = r(− log r sin θ − sin θ − θ cos θ

i
f 0 (z) = ur (z, 0) − uθ (z, 0)
r
= log z + 1

Integrating f 0 (z) yields

f (z) = z log z + ic.

11.3.2 Analytic Functions Defined in Terms of Their Real or Imaginary Parts

Consider an analytic function: f (z) = u(x, y) + iv(x, y). We differentiate this expression.

f 0 (z) = ux (x, y) + ivx (x, y)

We apply the Cauchy-Riemann equation vx = −uy .

f 0 (z) = ux (x, y) − iuy (x, y). (11.1)

Now consider the function of a complex variable, g(ζ):

g(ζ) = ux (x, ζ) − iuy (x, ζ) = ux (x, ξ + iη) − iuy (x, ξ + iη).

369
This function is analytic where f (ζ) is analytic. To show this we first verify that the derivatives in the ξ and η
directions are equal.
∂
g(ζ) = uxy (x, ξ + iη) − iuyy (x, ξ + iη)
∂ξ
∂
−i g(ζ) = −i(iuxy (x, ξ + iη) + uyy (x, ξ + iη)) = uxy (x, ξ + iη) − iuyy (x, ξ + iη)
∂η

Since these partial derivatives are equal and continuous, g(ζ) is analytic. We evaluate the function g(ζ) at ζ = −ix.
(Substitute y = −ix into Equation 11.1.)

f 0 (2x) = ux (x, −ix) − iuy (x, −ix)

We make a change of variables to solve for f 0 (x).

x x x x
0
f (x) = ux , −i − iuy , −i .
2 2 2 2
If the expression is nonsingular, then this defines the analytic function, f 0 (z), on the real axis. The analytic
continuation to the complex plane is
z z z z
f 0 (z) = ux , −i − iuy , −i .
2 2 2 2
d
Note that dz
2u(z/2, −iz/2) = ux (z/2, −iz/2) − iuy (z/2, −iz/2). We integrate the equation to obtain:
z z
f (z) = 2u , −i + c.
2 2
We know that the real part of an analytic function determines that function to within an additive constant.
Assuming that the above expression is non-singular, we have found a formula for writing an analytic function
in terms of its real part. With the same method, we can find how to write an analytic function in terms of its
imaginary part, v.

370
 We can also derive formulas if u and v are expressed in polar coordinates:
f (z) = u(r, θ) + iv(r, θ).

Result 11.3.2 If f (z) = u(x, y) + iv(x, y) is analytic and the expressions are non-
singular, then
z z
f (z) = 2u , −i + const (11.2)
2z 2z 
f (z) = i2v , −i + const. (11.3)
2 2
If f (z) = u(r, θ) + iv(r, θ) is analytic and the expressions are non-singular, then
 
1/2 i
f (z) = 2u z , − log z + const (11.4)
2
 
i
f (z) = i2v z 1/2 , − log z + const. (11.5)
2

Example 11.3.12 Consider the problem of finding f (z) given that u(x, y) = e−x (x sin y − y cos y).
z z
f (z) = 2u , −i
2  2 
−z/2 z z z  z 
= 2e sin −i + i cos −i +c
2  z 2 2
z  2
= iz e−z/2 i sin i + cos −i +c
2 2
= iz e−z/2 ( e−z/2 ) + c
= iz e−z + c

371
Example 11.3.13 Consider
1
Log x2 + y 2 + i Arctan (x, y).


Log z =
2
We try to construct the analytic function from it’s real part using Equation 11.2.
z z
f (z) = 2u , −i +c
2 2
1  z 2  z 2 
= 2 Log + −i +c
2 2 2
= Log (0) + c

We obtain a singular expression, so the method fails.

Example 11.3.14 Again consider the logarithm, this time written in terms of polar coordinates,

Log z = Log r + iθ.

We try to construct the analytic function from it’s real part using Equation 11.4.
 
1/2 i
f (z) = 2u z , −i log z + c
2
1/2


= 2 Log z +c
= Log z + c

With this method we recover the analytic function.

372
11.4 Exercises
Exercise 11.1
Consider two functions, f (x, y) and g(x, y). They are said to be functionally dependent if there is a an h(g) such
that

f (x, y) = h(g(x, y)).

f and g will be functionally dependent if and only if their Jacobian vanishes.

If f and g are functionally dependent, then the derivatives of f are

fx = h0 (g)gx
fy = h0 (g)gy .

Thus we have
 
∂(f, g) fx fy 
= = fx gy − fy gx = h0 (g)gx gy − h0 (g)gy gx = 0.
∂(x, y) g x g y


If the Jacobian of f and g vanishes, then

fx gy − fy gx = 0.

This is a first order partial differential equation for f that has the general solution

f (x, y) = h(g(x, y)).

Prove that an analytic function u(x, y) + iv(x, y) can be written in terms of a function of a complex variable,
f (z) = u(x, y) + iv(x, y).

Exercise 11.2
Which of the following functions are the real part of an analytic function? For those that are, find the harmonic
conjugate, v(x, y), and find the analytic function f (z) = u(x, y) + iv(x, y) as a function of z.

373
 1. x3 − 3xy 2 − 2xy + y

2. ex sinh y

3. ex (sin x cos y cosh y − cos x sin y sinh y)

Exercise 11.3
For an analytic function, f (z) = u(r, θ) + iv(r, θ) prove that under suitable restrictions:
 
1/2 i
f (z) = 2u z , − log z + const.
2

374
11.5 Hints
Hint 11.1
Show that u(x, y) + iv(x, y) is functionally dependent on x + iy so that you can write f (z) = f (x + iy) =
u(x, y) + iv(x, y).

Hint 11.2

Hint 11.3
Check out the derivation of Equation 11.2.

375
11.6 Solutions
Solution 11.1
u(x, y) + iv(x, y) is functionally dependent on z = x + iy if and only if

∂(u + iv, x + iy)

= 0.
∂(x, y)

 
∂(u + iv, x + iy) ux + ivx uy + ivy 
=
∂(x, y) 1 i 
= −vx − uy + i(ux − vy )

Since u and v satisfy the Cauchy-Riemann equations, this vanishes

=0

Thus we see that u(x, y) + iv(x, y) is functionally dependent on x + iy so we can write

f (z) = f (x + iy) = u(x, y) + iv(x, y).

Solution 11.2
1. Consider u(x, y) = x3 − 3xy 2 − 2xy + y. The Laplacian of this function is

∆u ≡ uxx + uyy
= 6x − 6x
=0

Since the function is harmonic, it is the real part of an analytic function. Clearly the analytic function is of
the form,

az 3 + bz 2 + cz + id,

376
with a, b and c complex-valued constants and d a real constant. Substituting z = x + iy and expanding
products yields,

a(x3 + i3x2 y − 3xy 2 − iy 3 ) + b(x2 + i2xy − y 2 ) + c(x + iy) + id.

By inspection, we see that the analytic function is

f (z) = z 3 + iz 2 − iz + id.

The harmonic conjugate of u is the imaginary part of f (z),

v(x, y) = 3x2 y − y 3 + x2 − y 2 − x + d.

We can also do this problem with analytic continuation. The derivatives of u are

ux = 3x2 − 3y 2 − 2y,
uy = −6xy − 2x + 1.

The derivative of f (z) is

f 0 (z) = ux − iuy = 3x2 − 2y 2 − 2y + i(6xy − 2x + 1).

On the real axis we have

f 0 (z = x) = 3x2 − i2x + i.

Using analytic continuation, we see that

f 0 (z) = 3z 2 − i2z + i.

Integration yields

f (z) = z 3 − iz 2 + iz + const

377
2. Consider u(x, y) = ex sinh y. The Laplacian of this function is

∆u = ex sinh y + ex sinh y
= 2 ex sinh y.

Since the function is not harmonic, it is not the real part of an analytic function.

3. Consider u(x, y) = ex (sin x cos y cosh y − cos x sin y sinh y). The Laplacian of the function is
∂
∆u = ( ex (sin x cos y cosh y − cos x sin y sinh y + cos x cos y cosh y + sin x sin y sinh y))
∂x
∂
+ ( ex (− sin x sin y cosh y − cos x cos y sinh y + sin x cos y sinh y − cos x sin y cosh y))
∂y
= 2 ex (cos x cos y cosh y + sin x sin y sinh y) − 2 ex (cos x cos y cosh y + sin x sin y sinh y)
= 0.

Thus u is the real part of an analytic function. The derivative of the analytic function is

f 0 (z) = ux + ivx = ux − iuy

From the derivatives of u we computed before, we have

f (z) = ( ex (sin x cos y cosh y − cos x sin y sinh y + cos x cos y cosh y + sin x sin y sinh y))
− i ( ex (− sin x sin y cosh y − cos x cos y sinh y + sin x cos y sinh y − cos x sin y cosh y))

Along the real axis, f 0 (z) has the value,

f 0 (z = x) = ex (sin x + cos x).

By analytic continuation, f 0 (z) is

f 0 (z) = ez (sin z + cos z)

378
 We obtain f (z) by integrating.
f (z) = ez sin z + const.
u is the real part of the analytic function

f (z) = ez sin z + ic,

where c is a real constant. We find the harmonic conjugate of u by taking the imaginary part of f .
f (z) = ex (cosy + i sin y)(sin x cosh y + i cos x sinh y) + ic

v(x, y) = ex sin x sin y cosh y + cos x cos y sinh y + c

Solution 11.3
We consider the analytic function: f (z) = u(r, θ) + iv(r, θ). Recall that the complex derivative in terms of polar
coordinates is
∂ i ∂
= e−iθ = − e−iθ .
z ∂r r ∂θ
The Cauchy-Riemann equations are
1 1
ur = vθ , vr = − uθ .
r r
We differentiate f (z) and use the partial derivative in r for the right side.
f 0 (z) = e−iθ (ur + ivr )
We use the Cauchy-Riemann equations to right f 0 (z) in terms of the derivatives of u.
 
0 −iθ 1
f (z) = e ur − i uθ (11.6)
r

379
 Now consider the function of a complex variable, g(ζ):
   
−iζ 1 η−iξ 1
g(ζ) = e ur (r, ζ) − i uθ (r, ζ) = e ur (r, ξ + iη) − i uθ (r, ξ + iη)
r r
This function is analytic where f (ζ) is analytic. It is a simple calculus exercise to show that the complex derivative
∂ ∂
in the ξ direction, ∂ξ , and the complex derivative in the η direction, −i ∂η , are equal. Since these partial derivatives
are equal and continuous, g(ζ) is analytic. We evaluate the function g(ζ) at ζ = −i log r. (Substitute θ = −i log r
into Equation 11.6.)
 
0 i(−i log r)

−i(−i log r) 1
f re = e ur (r, −i log r) − i uθ (r, −i log r)
r
1
rf 0 r2 = ur (r, −i log r) − i uθ (r, −i log r)


r
If the expression is nonsingular, then it defines the analytic function, f 0 (z), on a curve. The analytic continuation
to the complex plane is
1
zf 0 z 2 = ur (z, −i log z) − i uθ (z, −i log z).


z
We integrate to obtain an expression for f (z 2 ).
1
f z 2 = u(z, −i log z) + const


2
We make a change of variables and solve for f (z).
 
1/2 i
f (z) = 2u z , − log z + const.
2
Assuming that the above expression is non-singular, we have found a formula for writing the analytic function in
terms of its real part, u(r, θ). With the same method, we can find how to write an analytic function in terms of
its imaginary part, v(r, θ).

380
Chapter 12

Contour Integration and Cauchy’s Theorem

Between two evils, I always pick the one I never tried before.

- Mae West

12.1 Line Integrals

In this section we will recall the definition of a line integral of real-valued functions in the Cartesian plane. We
will use this to define the contour integral of complex-valued functions in the complex plane.

Definition. Consider a curve C in the Cartesian plane joining the points (a0 , b0 ) and (a1 , b1 ). Partition the
curve into n + 1 segments with the points (x0 , y0 ), . . . , (xn , yn ) where the first and last points are at the endpoints
of the curve. Define ∆xk = xk+1 − xk and ∆yk = yk+1 − yk . Let (ξk , ηk ) be points on the curve between (xk , yk )
and (xk+1 , yk+1 ). (See Figure 12.1.)
Consider the sum
n−1
X
(P (ξk , ηk )∆xk + Q(ξk , ηk )∆yk ) ,
k=0

381
 y

(ξ1 ,η1) (x2 ,y2 ) (xn ,yn )

(x1 ,y1 )
(ξ2 ,η2) (ξ n-1 ,ηn-1 )
(ξ0 ,η0)
(x0 ,y0 ) (xn-1 ,yn-1 )
x

Figure 12.1: A curve in the Cartesian plane.

where P and Q are continuous functions on the curve. In the limit as each of the ∆xk and ∆yk approach zero
the value of the sum, (if the limit exists), is denoted by
Z
P (x, y) dx + Q(x, y) dy.
C

This is a line integral along the curve C. The value of the line integral depends on the functions P (x, y) and
Q(x, y), the endpoints of the curve and the curve C. One can also write a line integral in vector notation,
Z
f (x) · dx,
C

where x = (x, y) and f (x) = (P (x, y), Q(x, y)).

Evaluation. Let the curve C be parametrized by x = x(t), y = y(t) for t0 ≤ t ≤ t1 . The differentials on the
curve are dx = x0 (t) dt and dy = y 0 (t) dt. Thus the line integral is
Z t1
P (x(t), y(t))x0 (t) + Q(x(t), y(t))y 0 (t) dt,


t0

382
which is a definite integral.

Example 12.1.1 Consider the line integral

Z
x2 dx + (x + y) dy,
C
where C is the semi-circle from (1, 0) to (−1, 0) in the upper half plane. We parameterize the curve with x = cos t,
y = sin t for 0 ≤ t ≤ π.
Z Z π
2
cos2 t(− sin t) + (cos t + sin t) cos t dt


x dx + (x + y) dy =
C 0
π 2
= −
2 3

Complex Line Integrals. Consider a curve C in the complex plane joining the points c0 and c1 . Partition the
curve into n + 1 segments with the points z0 , . . . , zn where the first and last points are at the endpoints of the
curve. Define ∆zk = zk+1 − zk . Let ζk be points on the curve between zk and zk+1 . Consider the sum
n−1
X
f (ζk )∆zk ,
k=0
where f is a continuous, complex-valued function on the curve. In the limit as each of the ∆zk approach zero the
value of the sum, (if the limit exists), is denoted by
Z
f (z) dz.
C
This is a complex line integral along the curve C.
We can write a complex line integral in terms of real line integrals. Let f (z) = u(x, y) + iv(x, y).
Z Z
f (z) dz = (u(x, y) + iv(x, y))( dx + i dy)
Z ZC C
Z
f (z) dz = (u(x, y) dx − v(x, y) dy) + i (v(x, y) dx + u(x, y) dy). (12.1)
C C C

383
Evaluation. Let the curve C be parametrized by z = z(t) for t0 ≤ t ≤ t1 . Then the complex line integral is
Z t1
f (z(t))z 0 (t) dt,
t0

which is a definite integral of a complex-valued function.

Example 12.1.2 Let C be the positively oriented unit circle about the origin in the complex plane. Evaluate:
R
1. C z dz

2. C z1 dz
R

3. C z1 |dz|
R

1. We parameterize the curve and then do the integral.

z = eiθ , dz = i eiθ dθ

Z Z 2π
z dz = eiθ i eiθ dθ
C 0
 2π
1 i2θ
= e
2 0
 
1 i4π 1 i0
= e − e
2 2
=0

2.
Z Z 2π Z 2π
1 1 iθ
dz = i e dθ = i dθ = i2π
C z 0 eiθ 0

384
 3.
   
|dz| = i eiθ dθ = i eiθ  |dθ| = |dθ|

Since dθ is positive in this case, |dθ| = dθ.

Z Z 2π
1  −iθ 2π
1/z |dz| = dθ = ie 0
=0
C 0 eiθ

Maximum Modulus Integral Bound. The absolute value of a real integral obeys the inequality
Z b  Z b
 
 f (x) dx ≤ |f (x)| |dx| ≤ (b − a) max |f (x)|.

a a a≤x≤b

Now we prove the analogous result for the modulus of a complex line integral.
Z   n−1
X


  
 f (z) dz  =  lim f (ζk )∆zk 

  ∆z→0
C 
k=0
n−1
X
≤ lim |f (ζk )| |∆zk |
∆z→0
k=0
Z
= |f (z)| |dz|
C
Z  
≤ max |f (z)| |dz|
C z∈C
 Z
= max |f (z)| |dz|
z∈C C
 
= max |f (z)| × (length of C)
z∈C

385
 Result 12.1.1 Maximum Modulus Integral Bound.
Z  Z  
 
 f (z) dz  ≤ |f (z)| |dz| ≤ max |f (z)| (length of C)

C

C z∈C

12.2 Under Construction

Cauchy’s Theorem. Let f (z) be analytic in a compact, closed, connected domain D. Consider the integral of
f (z) on the boundary of the domain.
Z Z Z
f (z) dz = ψ(x, y) (dx + idy) = ψ dx + iψ dy
∂D ∂D ∂D

Recall Green’s Theorem.

Z Z
P dx + Q dy = (Qx − Py ) dxy
∂D D

We apply Green’s Theorem to the integral of f (z) on ∂D.

Z Z Z
f (z) dz = ψ dx + iψ dy = (iψx − ψy ) dxy
∂D ∂D D

Since f (z) is analytic, ψx = −iψy . The integrand iψx − ψy is zero. Thus we have
Z
f (z) dz = 0.
∂D

This is known as Cauchy’s Theorem.

386
Fundamental Theorem of Calculus. First note that <(·) and =(·) commute with derivatives and integrals.
Let P (x, y) and Q(x, y) be defined on a simply connected domain. A necessary and sufficient condition for the
existence of a primitive φ is that Py = Qx . The primitive satisfies
dφ = P dx + Q dy.
Definite integral can be evaluated in terms of the primitive.
Z (c,d)
P dx + Q dy = φ(c, d) − φ(a, b)
(a,b)

Now consider integral along the contour C of the complex-valued function ψ(x, y).
Z Z
ψ dz = ψ dx + iψ dy
C C

If ψ(x, y) is analytic then there exists a function Ψ such that

dΨ = ψ dx + iψ dy.
Then ψ satisfies the Cauchy-Riemann equations. How do we find the primitive Ψ that satisfies Ψx = ψ and
Ψy = iψ? Note that choosing Ψ(x, y) = F (z) where F (z) is an anti-derivative of f (z), F 0 (z) = f (z), does the
trick.
F 0 (z) = Ψx = −iΨy = f = ψ
The differential of Ψ is
dΨ = Ψx dx + Ψy dy = ψ dx + ψ dy.
We can evaluate a definite integral of f in terms of F .
Z b
f (z) dz = F (b) − F (a).
a

This is the Fundamental Theorem of Calculus for functions of a complex variable.

387
12.3 Cauchy’s Theorem

Result 12.3.1 Cauchy’s Theorem. If f (z) is analytic in a compact, closed, connected

domain D then the integral of f (z) on the boundary of the domain vanishes.
I XI
f (z) dz = f (z) dz = 0
∂D k Ck

Here the set of contours {Ck } make up the positively oriented boundary ∂D of the domain
D.
This result follows from Green’s Theorem. Since Green’s theorem holds for both simply and multiply connected
domains, so does Cauchy’s theorem.

Proof of Cauchy’s Theorem. We will assume that f 0 (z) is continuous. This assumption is not necessary,
but it allows us to use Green’s Theorem, which makes for a simpler proof. We consider the integral of f (z) =
u(x, y) + iv(x, y) along the boundary of the domain. From Equation 12.1 we have,
Z Z Z
f (z) dz = (u dx − v dy) + i (v dx + u dy)
∂D ∂D ∂D

We use Green’s theorem to write this as an area integral.

Z Z Z
f (z) dz = (−vx − uy ) dx dy + i (ux − vy ) dx dy
∂D D D

Since u and v satisfy the Cauchy-Riemann Equations, ux = vy and uy = −vx , the two integrands on the right
side are identically zero. Thus the two area integrals vanish and Cauchy’s theorem is proved.
As a special case of Cauchy’s theorem we can consider a simply-connected region. For this the boundary is a
Jordan curve. We can state the theorem in terms of this curve instead of referring to the boundary.

388
 Result 12.3.2 Cauchy’s Theorem for Jordan Curves. If f (z) is analytic inside and
on a simple, closed contour C, then
I
f (z) dz = 0
C

Example 12.3.1 In Example 12.1.2 we calculated that

Z
z dz = 0
C

where C is the unit circle about the origin. Now we can evaluate the integral without parameterizing the curve.
We simply note that the integrand is analytic inside and on the circle, which is simple and closed. By Cauchy’s
Theorem, the integral vanishes.
We cannot apply Cauchy’s theorem to evaluate

Z
1
dz = i2π
C z

as the integrand is not analytic at z = 0.

Morera’s Theorem. The converse of Cauchy’s theorem, is Morera’s Theorem. If the integrals of a continuous
function f (z) vanish along all possible simple, closed contours in a domain, then f (z) is analytic on that domain.
To prove Morera’s Theorem we will assume that first partial derivatives of f (z) = u(x, y)+iv(x, y) are continuous,
although the result can be derived without this restriction. Let the simple, closed contour C be the boundary of

389
D which is contained in the domain Ω.
I I
f (z) dz = (u + iv)( dx + i dy)
C
IC I
= u dx − v dy + i v dx + u dy
C C
Z Z
= (−vx − uy ) dx dy + i (ux − vy ) dx dy
D D
=0

Since the two integrands are continuous and vanish for all C in Ω, we conclude that the integrands are identically
zero. This implies that the Cauchy-Riemann equations,

u x = vy , uy = −vx ,

are satisfied. f (z) is analytic in Ω.

Result 12.3.3 Morera’s Theorem. If f (z) is continuous in a simply connected domain

Ω and
I
f (z) dz = 0
C

for all possible simple, closed contours C in the domain, the f (z) is analytic in Ω.

12.4 Indefinite Integrals

Consider a function f (z) which is analytic in a domain D. An anti-derivative or indefinite integral (or simply
integral ) is a function F (z) which satisfies F 0 (z) = f (z). This integral exists and is unique up to an additive

390
constant. Note that if the domain is not connected, then the additive constants in each connected component are
independent. The indefinite integrals are denoted:
Z
f (z) dz = F (z) + c.

We will prove existence in the next section by writing an indefinite integral as a contour integral. We consider
uniqueness here. Let F (z) and G(z) be integrals of f (z). Then F 0 (z) − G0 (z) = f (z) − f (z) = 0. One can use
this to show that F (z) − G(z) is a constant on each connected component of the domain. This demonstrates
uniqueness.
Integrals of analytic functions have all the nice properties of integrals of functions of a real variables. All the
formulas from integral tables, including things like integration by parts, carry over directly.

12.5 Contour Integrals

Result 12.5.1 Path Independence. Let f (z) be analytic on a simply connected do-
main. For points a and b in the domain, the contour integral,
Z b
f (z) dz
a

is independent of the path connecting the points.

(Here we assume that the paths lie entirely in the domain.) This result is a direct consequence of Cauchy’s
Theorem. Let C1 and C2 be two different paths connecting the points. Let −C2 denote the second curve with the
opposite orientation. Let C be the contour which is the union of C1 and −C2 . By Cauchy’s theorem, the integral
along this contour vanishes.
Z Z
f (z) dz + f (z) dz = 0
C1 −C2

391
This implies that
Z Z
f (z) dz = f (z) dz.
C1 C2

Thus contour integrals on simply connected domains are independent of path. This result does not hold for
multiply connected domains.

Result 12.5.2 Constructing an Indefinite Integral. If f (z) is analytic in a simply

connected domain D and a is a point in the domain, then
Z z
F (z) = f (ζ) dζ
a

is analytic in D and is an indefinite integral of f (z), (F 0 (z) = f (z)).

To prove this, we use the limit definition of differentiation.
F (z + ∆z) − F (z)
F 0 (z) = lim
∆z→0 ∆z
Z z+∆z Z z 
1
= lim f (ζ) dζ − f (ζ) dζ
∆z→0 ∆z a a
Z z+∆z
1
= lim f (ζ) dζ
∆z→0 ∆z z

The integral is independent of path. We choose a straight line connecting z and z + ∆z. We add and subtract
R z+∆z
∆zf (z) = z f (z) dζ from the expression for F 0 (z).
 Z z+∆z 
0 1
F (z) = lim ∆zf (z) + (f (ζ) − f (z)) dζ
∆z→0 ∆z z
Z z+∆z
1
= f (z) + lim (f (ζ) − f (z)) dζ
∆z→0 ∆z z

392
Since f (z) is analytic, it is certainly continuous. This means that
lim f (ζ) = 0.
ζ→z

The limit term vanishes as a result of this continuity.

 Z z+∆z 
 1  1
lim  (f (ζ) − f (z)) dζ  ≤ lim |∆z| max |f (ζ) − f (z)|
∆z→0 ∆z z ∆z→0 |∆z| ζ∈[z...z+∆z]

= lim max |f (ζ) − f (z)|

∆z→0 ζ∈[z...z+∆z]

=0
Thus F 0 (z) = f (z).
This results demonstrates the existence of the indefinite integral. We will use this to prove the Fundamental
Theorem of Calculus for functions of a complex variable.

Result 12.5.3 Fundamental Theorem of Calculus. If f (z) is analytic in a simply

connected domain D then
Z b
f (z) dz = F (b) − F (a)
a

where F (z) is any indefinite integral of f (z).

From Result 12.5.2 we know that
Z b
f (z) dz = F (b) + c.
a
(Here we are considering b to be a variable.) The case b = a determines the constant.
Z a
f (z) dz = F (a) + c = 0
a
c = −F (a)

393
This proves the Fundamental Theorem of Calculus for functions of a complex variable.

Example 12.5.1 Consider the integral

Z
1
dz
C z−a
where C is any closed contour that goes around the point z = a once in the positive direction. We use the
Fundamental Theorem of Calculus to evaluate the integral. We start at a point on the contour z − a = r eiθ .
When we traverse the contour once in the positive direction we end at the point z − a = r ei(θ+2π) .
Z
1 ei(θ+2π)
dz = [log(z − a)]z−a=r
z−a=r eiθ
C z −a
= Log r + i(θ + 2π) − ( Log r + iθ)
= i2π

394
12.6 Exercises
Exercise 12.1
C is the arc corresponding to the unit semi-circle, |z| = 1, =(z) ≥ 0, directed from z = −1 to z = 1. Evaluate
Z
1. z 2 dz
C
Z
 2
2. z  dz
C
Z
3. z 2 |dz|
C
Z
 2
4. z  |dz|
C

Exercise 12.2
Evaluate
Z ∞
2 +bx)
e−(ax dx,
−∞

where a, b ∈ C and <(a) > 0. Use the fact that

Z ∞ √
2
e−x dx = π.
−∞

Exercise 12.3
Evaluate
Z ∞ Z ∞
−ax2 2
2 e cos(ωx) dx, and 2 x e−ax sin(ωx) dx,
0 0

395
where <(a) > 0 and ω ∈ R.

396
12.7 Hints
Hint 12.1

Hint 12.2
Let C be the parallelogram in the complex plane with corners at ±R and ±R + b/(2a). Consider the integral of
2
e−az on this contour. Take the limit as R → ∞.

Hint 12.3
Extend the range of integration to (−∞ . . . ∞). Use eiωx = cos(ωx) + i sin(ωx) and the result of Exercise 12.2.

397
12.8 Solutions
Solution 12.1
We parameterize the path with z = eiθ , with θ ranging from π to 0.

dz = i eiθ dθ

|dz| = |i eiθ dθ| = |dθ| = −dθ

1.

Z Z 0
2
z dz = ei2θ i eiθ dθ
C
Zπ 0
= i ei3θ dθ
π
 0
1 i3θ
= e
3 π
1 i0
e − ei3π


=
3
1
= (1 − (−1))
3
2
=
3

398
2.
Z Z 0
2
|z | dz = | ei2θ |i eiθ dθ
C
Zπ 0
= i eiθ dθ
π
 iθ 0
= e π
= 1 − (−1)
=2
3.
Z Z 0
2
z |dz| = ei2θ |i eiθ dθ|
C
Zπ 0
= − ei2θ dθ
π
 0
i i2θ
= e
2 π
i
= (1 − 1)
2
=0
4.
Z Z 0
2
|z | |dz| = | ei2θ ||i eiθ dθ|
C
Zπ 0
= − dθ
π
= [−θ]0π
=π

399
Solution 12.2

Z ∞
2 +bx)
I= e−(ax dx
−∞

First we complete the square in the argument of the exponential.

Z ∞
b2 /(4a) 2
I= e e−a(x+b/(2a)) dx
−∞
2
Consider the parallelogram in the complex plane with corners at ±R and ±R + b/(2a). The integral of e−az on
this contour vanishes as it is an entire function. We can write this as
Z R+b/(2a) Z −R Z R Z R+b/(2a) !
2 2
e−az dz = + + e−az dz.
−R+b/(2a) −R+b/(2a) −R R

The first and third integrals on the right side vanish as R → ∞ because the integrand vanishes and the lengths
of the paths of integration are finite. Taking the limit as R → ∞ we have,
Z ∞+b/(2a) Z ∞ Z ∞
−az 2 −a(x+b/(2a))2 2
e dz ≡ e dx = e−ax dx.
−∞+b/(2a) −∞ −∞

Now we have
Z ∞
b2 /(4a) 2
I= e e−ax dx.
−∞
√
We make the change of variables ξ = ax.
Z ∞
b2 /(4a) 1 2
I= e √ e−ξ dx
a −∞

∞
Z r
−(ax2 +bx) π b2 /(4a)
e dx = e
−∞ a

400
Solution 12.3
Consider
Z ∞
2
I=2 e−ax cos(ωx) dx.
0

Since the integrand is an even function,

Z ∞
2
I= e−ax cos(ωx) dx.
−∞
2
Since e−ax sin(ωx) is an odd function,
Z ∞
2
I= e−ax eiωx dx.
−∞

We evaluate this integral with the result of Exercise 12.2.

∞
Z r
−ax2 π −ω2 /(4a)
2 e cos(ωx) dx = e
0 a

Consider
Z ∞
2
I=2 x e−ax sin(ωx) dx.
0

Since the integrand is an even function,

Z ∞
2
I= x e−ax sin(ωx) dx.
−∞
2
Since x e−ax cos(ωx) is an odd function,
Z ∞
2
I = −i x e−ax eiωx dx.
−∞

401
We add a dash of integration by parts to get rid of the x factor.
 ∞ Z ∞ 
1 −ax2 iωx 1 −ax2 iωx
I = −i − e e +i − e iω e dx.
2a −∞ −∞ 2a
ω ∞ −ax2 iωx
Z
I= e e dx.
2a −∞

∞
Z r
−ax2 ω π −ω2 /(4a)
2 xe sin(ωx) dx. = e
0 2a a

402
Chapter 13

Cauchy’s Integral Formula

If I were founding a university I would begin with a smoking room; next a dormitory; and then a decent reading
room and a library. After that, if I still had more money that I couldn’t use, I would hire a professor and get
some text books.

- Stephen Leacock

403
13.1 Cauchy’s Integral Formula

Result 13.1.1 Cauchy’s Integral Formula. If f (ζ) is analytic in a compact, closed,

connected domain D and z is a point in the interior of D then
I I
1 f (ζ) 1 X f (ζ)
f (z) = dζ = dζ. (13.1)
i2π ∂D ζ − z i2π Ck ζ − z k

Here the set of contours {Ck } make up the positively oriented boundary ∂D of the domain
D. More generally, we have
I XI
n! f (ζ) n! f (ζ)
f (n) (z) = dζ = dζ. (13.2)
i2π ∂D (ζ − z)n+1 i2π Ck (ζ − z)n+1
k

Cauchy’s Formula shows that the value of f (z) and all its derivatives in a domain are determined by the value
of f (z) on the boundary of the domain. Consider the first formula of the result, Equation 13.1. We deform the
contour to a circle of radius δ about the point ζ = z.
I I
f (ζ) f (ζ)
dζ = dζ
C ζ −z Cδ ζ − z
f (ζ) − f (z)
I I
f (z)
= dζ + dζ
Cδ ζ − z Cδ ζ −z

We use the result of Example 12.5.1 to evaluate the first integral.

f (ζ) − f (z)
I I
f (ζ)
dζ = i2πf (z) + dζ
C ζ −z Cδ ζ −z

404
The remaining integral along Cδ vanishes as δ → 0 because f (ζ) is continuous. We demonstrate this with the
maximum modulus integral bound. The length of the path of integration is 2πδ.
I   
 f (ζ) − f (z)  1
lim  dζ  ≤ lim (2πδ) max |f (ζ) − f (z)|

δ→0  C
δ
ζ −z δ→0 δ |ζ−z|=δ
 
≤ lim 2π max |f (ζ) − f (z)|
δ→0 |ζ−z|=δ

=0

This gives us the desired result.

I
1 f (ζ)
f (z) = dζ
i2π C ζ −z

We derive the second formula, Equation 13.2, from the first by differentiating with respect to z. Note that the
integral converges uniformly for z in any closed subset of the interior of C. Thus we can differentiate with respect
to z and interchange the order of differentiation and integration.

1 dn
I
(n) f (ζ)
f (z) = n
dζ
i2π dz C ζ − z
dn f (ζ)
I
1
= dζ
i2π C dz n ζ − z
I
n! f (ζ)
= dζ
i2π C (ζ − z)n+1

Example 13.1.1 Consider the following integrals where C is the positive contour on the unit circle. For the
third integral, the point z = −1 is removed from the contour.
I
1. sin(cos(z 5 )) dz
C

405
 I
1
2. dz
C (z − 3)(3z − 1)
√
Z
3. z dz
C

1. Since sin(cos(z 5 )) is an analytic function inside the unit circle,

I
sin(cos(z 5 )) dz = 0
C

1
2. has singularities at z = 3 and z = 1/3. Since z = 3 is outside the contour, only the singularity at
(z−3)(3z−1)
z = 1/3 will contribute to the value of the integral. We will evaluate this integral using the Cauchy integral
formula.
I  
1 1 πi
dz = i2π =−
C (z − 3)(3z − 1) (1/3 − 3)3 4
√
3. Since the curve is not closed, we cannot apply the Cauchy integral formula. Note that z is single-valued
and analytic in the complex plane with a branch cut on the negative real axis. Thus we use the Fundamental
Theorem of Calculus.
 √  eiπ
√
Z
2 3
z dz = z
C 3 e−iπ
2 i3π/2
− e−i3π/2


= e
3
2
= (−i − i)
3
4
= −i
3

406
Cauchy’s Inequality. Suppose the f (ζ) is analytic in the closed disk |ζ − z| ≤ r. By Cauchy’s integral formula,
I
(n) n! f (ζ)
f (z) = dζ,
i2π C (ζ − z)n+1

where C is the circle of radius r centered about the point z. We use this to obtain an upper bound on the modulus
of f (n) (z).
I 
f (z) = n!  f (ζ)
 (n)   
n+1
dζ 
2π C (ζ − z)

 
n!  f (ζ) 
≤ 2πr max  
2π |ζ−z|=r (ζ − z)n+1 
n!
= max |f (ζ)|
rn |ζ−z|=r

Result 13.1.2 Cauchy’s Inequality. If f (ζ) is analytic in |ζ − z| ≤ r then

 (n)  n!M
 
f (z) ≤ n
r
where |f (ζ)| ≤ M for all |ζ − z| = r.

Liouville’s Theorem. Consider a function f (z) that is analytic and bounded, (f (z) ≤ M ), in the complex
plane. From Cauchy’s inequality,

M
|f 0 (z)| ≤
r
for any positive r. By taking r → ∞, we see that f 0 (z) is identically zero for all z. Thus f (z) is a constant.

407
 Result 13.1.3 Liouville’s Theorem. If f (z) is analytic and bounded in the complex
plane then f (z) is a constant.

The Fundamental Theorem of Algebra. We will prove that every polynomial of degree n ≥ 1 has exactly
n roots, counting multiplicities. First we demonstrate that each such polynomial has at least one root. Suppose
that an nth degree polynomial p(z) has no roots. Let the lower bound on the modulus of p(z) be 0 < m ≤ |p(z)|.
The function f (z) = 1/p(z) is analytic, (f 0 (z) = p0 (z)/p2 (z)), and bounded, (|f (z)| ≤ 1/m), in the extended
complex plane. Using Liouville’s theorem we conclude that f (z) and hence p(z) are constants, which yields a
contradiction. Therefore every such polynomial p(z) must have at least one root.

Now we show that we can factor the root out of the polynomial. Let

n
X
p(z) = pk z k .
k=0

We note that

n−1
X
n n
(z − c ) = (z − c) cn−1−k z k .
k=0

408
Suppose that the nth degree polynomial p(z) has a root at z = c.
p(z) = p(z) − p(c)
Xn n
X
k
= pk z − pk ck
k=0 k=0
Xn
= pk (z k − ck )
k=0
Xn k−1
X
= pk (z − c) ck−1−j z j
k=0 j=0

= (z − c)q(z)
Here q(z) is a polynomial of degree n − 1. By induction, we see that p(z) has exactly n roots.

Result 13.1.4 Fundamental Theorem of Algebra. Every polynomial of degree n ≥

1 has exactly n roots, counting multiplicities.

Gauss’ Mean Value Theorem. Let f (ζ) be analytic in |ζ − z| ≤ r. By Cauchy’s integral formula,
I
1 f (ζ)
f (z) = dζ,
2πi C ζ − z
where C is the circle |ζ − z| = r. We parameterize the contour with ζ = z + r eiθ .
Z 2π
1 f (z + r eiθ )
f (z) = ir eiθ dθ
2πi 0 r eiθ
Writing this in the form,
Z 2π
1
f (z) = f (z + r eiθ )r dθ,
2πr 0

409
we see that f (z) is the average value of f (ζ) on the circle of radius r about the point z.

Result 13.1.5 Gauss’ Average Value Theorem. If f (ζ) is analytic in |ζ − z| ≤ r

then
Z 2π
1
f (z) = f (z + r eiθ ) dθ.
2π 0
That is, f (z) is equal to its average value on a circle of radius r about the point z.

Extremum Modulus Theorem. Let f (z) be analytic in closed, connected domain, D. The extreme values
of the modulus of the function must occur on the boundary. If |f (z)| has an interior extrema, then the function
is a constant. We will show this with proof by contradiction. Assume that |f (z)| has an interior maxima at the
point z = c. This means that there exists an neighborhood of the point z = c for which |f (z)| ≤ |f (c)|. Choose
an  so that the set |z − c| ≤  lies inside this neighborhood. First we use Gauss’ mean value theorem.
Z 2π
1
f c +  eiθ dθ


f (c) =
2π 0
We get an upper bound on |f (c)| with the maximum modulus integral bound.
Z 2π
1 
f c +  eiθ  dθ


|f (c)| ≤
2π 0
Since z = c is a maxima of |f (z)| we can get a lower bound on |f (c)|.
Z 2π
1 
f c +  eiθ  dθ


|f (c)| ≥
2π 0
If |f (z)| < |f (c)| for any point on |z − c| = , then the continuity of f (z) implies that |f (z)| < |f (c)| in a
neighborhood of that point which would make the value of the integral of |f (z)| strictly less than |f (c)|. Thus we

410
conclude that |f (z)| = |f (c)| for all |z − c| = . Since we can repeat the above procedure for any circle of radius
smaller than , |f (z)| = |f (c)| for all |z − c| ≤ , i.e. all the points in the disk of radius  about z = c are also
maxima. By recursively repeating this procedure points in this disk, we see that |f (z)| = |f (c)| for all z ∈ D.
This implies that f (z) is a constant in the domain. By reversing the inequalities in the above method we see that
the minimum modulus of f (z) must also occur on the boundary.

Result 13.1.6 Extremum Modulus Theorem. Let f (z) be analytic in a closed,

connected domain, D. The extreme values of the modulus of the function must occur on
the boundary. If |f (z)| has an interior extrema, then the function is a constant.

13.2 The Argument Theorem

Result 13.2.1 The Argument Theorem. Let f (z) be analytic inside and on C except
for isolated poles inside the contour. Let f (z) be nonzero on C.
Z 0
1 f (z)
dz = N − P
i2π C f (z)
Here N is the number of zeros and P the number of poles, counting multiplicities, of
f (z) inside C.

First we will simplify the problem and consider a function f (z) that has one zero or one pole. Let f (z) be
analytic and nonzero inside and on A except for a zero of order n at z = a. Then we can write f (z) = (z − a)n g(z)

411
 f 0 (z)
where g(z) is analytic and nonzero inside and on A. The integral of f (z)
along A is

f 0 (z)
Z Z
1 1 d
dz = (log(f (z))) dz
i2π A f (z) i2π A dz
Z
1 d
= (log((z − a)n ) + log(g(z))) dz
i2π A dz
Z
1 d
= (log((z − a)n )) dz
i2π A dz
Z
1 n
= dz
i2π A z − a
=n

Now let f (z) be analytic and nonzero inside and on B except for a pole of order p at z = b. Then we can write
g(z) f 0 (z)
f (z) = (z−b) p where g(z) is analytic and nonzero inside and on B. The integral of f (z) along B is

f 0 (z)
Z Z
1 1 d
dz = (log(f (z))) dz
i2π B f (z) i2π B dz
Z
1 d
log((z − b)−p ) + log(g(z)) dz


=
i2π B dz
Z
1 d
log((z − b)−p )+ dz


=
i2π B dz
−p
Z
1
= dz
i2π B z − b
= −p

Now consider a function f (z) that is analytic inside an on the contour C except for isolated poles at the points
b1 , . . . , bp . Let f (z) be nonzero except at the isolated points a1 , . . . , an . Let the contours Ak , k = 1, . . . , n,
be simple, positive contours which contain the zero at ak but no other poles or zeros of f (z). Likewise, let the

412
contours Bk , k = 1, . . . , p be simple, positive contours which contain the pole at bk but no other poles of zeros of
f (z). (See Figure 13.1.) By deforming the contour we obtain
n p Z
f 0 (z) f 0 (z) f 0 (z)
Z X Z X
dz = dz + dz.
C f (z) j=1 Aj f (z) k=1 B j
f (z)

From this we obtain Result 13.2.1.

A1 C
B3
B1
A2
B2

Figure 13.1: Deforming the contour C.

13.3 Rouche’s Theorem

Result 13.3.1 Rouche’s Theorem. Let f (z) and (g) be analytic inside and on a
simple, closed contour C. If |f (z)| > |g(z)| on C then f (z) and f (z) + g(z) have the
same number of zeros inside C and no zeros on C.
First note that since |f (z)| > |g(z)| on C, f (z) is nonzero on C. The inequality implies that |f (z) + g(z)| > 0
on C so f (z) + g(z) has no zeros on C. We well count the number of zeros of f (z) and g(z) using the Argument

413
Theorem, (Result 13.2.1). The number of zeros N of f (z) inside the contour is
I 0
1 f (z)
N= dz.
i2π C f (z)

Now consider the number of zeros M of f (z) + g(z). We introduce the function h(z) = g(z)/f (z).
I 0
1 f (z) + g 0 (z)
M= dz
i2π C f (z) + g(z)
I 0
1 f (z) + f 0 (z)h(z) + f (z)h0 (z)
= dz
i2π C f (z) + f (z)h(z)
I 0
h0 (z)
I
1 f (z) 1
= dz + dz
i2π C f (z) i2π C 1 + h(z)
1
=N+ [log(1 + h(z))]C
i2π
=N

(Note that since |h(z)| < 1 on C, <(1 + h(z)) > 0 on C and the value of log(1 + h(z)) does not not change in
traversing the contour.) This demonstrates that f (z) and f (z) + g(z) have the same number of zeros inside C
and proves the result.

414
13.4 Exercises
Exercise 13.1
What is

(arg(sin z))C
where C is the unit circle?

Exercise 13.2
Let C be the circle of radius 2 centered about the origin and oriented in the positive direction. Evaluate the
following integrals:
1. C zsin z
H
2 +5 dz

2. C z2z+1 dz
H

H 2
3. C z z+1 dz

Exercise 13.3
Let f (z) be analytic and bounded (i.e. |f (z)| < M ) for |z| > R, but not necessarily analytic for |z| ≤ R. Let the
points α and β lie inside the circle |z| = R. Evaluate
I
f (z)
dz
C (z − α)(z − β)

where C is any closed contour outside |z| = R, containing the circle |z| = R. [Hint: consider the circle at infinity]
Now suppose that in addition f (z) is analytic everywhere. Deduce that f (α) = f (β).

Exercise 13.4
Using Rouche’s theorem show that all the roots of the equation p(z) = z 6 − 5z 2 + 10 = 0 lie in the annulus
1 < |z| < 2.

415
Exercise 13.5
Evaluate as a function of t

ezt
I
1
ω= dz,
i2π C z 2 (z 2 + a2 )

where C is any positively oriented contour surrounding the circle |z| = a.

416
13.5 Hints
Hint 13.1
Use the argument theorem.

Hint 13.2

Hint 13.3
To evaluate the integral, consider the circle at infinity.

Hint 13.4

Hint 13.5

417
13.6 Solutions
Solution 13.1
Let f (z) be analytic inside and on the contour C. Let f (z) be nonzero on the contour. The argument theorem
states that
Z 0
1 f (z)
dz = N − P,
i2π C f (z)
where N is the number of zeros and P is the number of poles, (counting multiplicities), of f (z) inside C. The
theorem is aptly named, as
Z 0
1 f (z) 1
dz = [log(f (z))]C
i2π C f (z) i2π
1
= [log |f (z)| + i arg(f (z))]C
i2π
1
= [arg(f (z))]C .
2π
Thus we could write the argument theorem as
Z 0
1 f (z) 1
dz = [arg(f (z))]C = N − P.
i2π C f (z) 2π
Since sin z has a single zero and no poles inside the unit circle, we have
1 
arg(sin(z))C = 1 − 0
2π

arg(sin(z))C = 2π

Solution 13.2

418
 √
1. Since the integrand zsin z
2 +5 is analytic inside and on the contour, (the only singularities are at z = ±i 5 and
at infinity), the integral is zero by Cauchy’s Theorem.

2. First we expand the integrand in partial fractions.

z a b
= +
z2 +1 z−i z+i

 
z  1 z  1
a= = , b= =
z + i z=i 2
 z − i z=−i 2


Now we can do the integral with Cauchy’s formula.

Z Z Z
z 1/2 1/2
dz = dz + dz
C z2 + 1 C z−i C z+i
1 1
= i2π + i2π
2 2
= i2π

3.

z2 + 1
Z Z  
1
dz = z+ dz
C z C z
Z Z
1
= z dz + dz
C C z
= 0 + i2π
= i2π

419
Solution 13.3
Let C be the circle of radius r, (r > R), centered at the origin. We get an upper bound on the integral with the
Maximum Modulus Integral Bound, (Result 12.1.1).
I   
 f (z)   f (z)  M

 (z − α)(z − β)  dz  ≤ 2πr max   ≤ 2πr
C |z|=r (z − α)(z − β)
  (r − |α|)(r − |β|)
By taking the limit as r → ∞ we see that the modulus of the integral is bounded above by zero. Thus the integral
vanishes.
Now we assume that f (z) is analytic and evaluate the integral with Cauchy’s Integral Formula. (We assume
that α 6= β.)
I
f (z)
dz = 0
C (z − α)(z − β)
I I
f (z) f (z)
dz + dz = 0
C (z − α)(α − β) C (β − α)(z − β)
f (α) f (β)
i2π + i2π =0
α−β β−α
f (α) = f (β)

Solution 13.4
Consider the circle |z| = 2. On this circle:
|z 6 | = 64
| − 5z 2 + 10| ≤ | − 5z 2 | + |10| = 30
Since |z 6 | < | − 5z 2 + 10| on |z| = 2, p(z) has the same number of roots as z 6 in |z| < 2. p(z) has 6 roots in |z| < 2.
Consider the circle |z| = 1. On this circle:
|10| = 10
|z 6 − 5z 2 | ≤ |z 6 | + | − 5z 2 | = 6

420
Since |z 6 − 5z 2 | < |10| on |z| = 1, p(z) has the same number of roots as 10 in |z| < 1. p(z) has no roots in |z| < 1.
On the unit circle,

|p(z)| ≥ |10| − |z 6 | − |5z 2 | = 4.

Thus p(z) has no roots on the unit circle.

We conclude that p(z) has exactly 6 roots in 1 < |z| < 2.

Solution 13.5
We evaluate the integral with Cauchy’s Integral Formula.

ezt
I
1
ω= dz
2πi C z 2 (z 2 + a2 )
I  zt
i ezt i ezt

1 e
ω= + − dz
2πi C a2 z 2 2a3 (z − ia) 2a3 (z + ia)
d ezt i eiat i e−iat
 
ω= + −
dz a2 z=0 2a3 2a3
t sin(at)
ω= 2−
a a3
at − sin(at)
ω=
a3

421
Chapter 14

Series and Convergence

You are not thinking. You are merely being logical.

- Neils Bohr

14.1 Series of Constants

14.1.1 Definitions
Convergence of Sequences. The infinite sequence {an }∞
n=0 ≡ a0 , a1 , a2 , . . . is said to converge if

lim an = a
n→∞

for some constant a. If the limit does not exist, then the sequence diverges. Recall the definition of the limit in
the above formula: For any  > 0 there exists an N ∈ Z such that |a − an | <  for all n > N .

Example 14.1.1 The sequence {sin(n)} is divergent. The sequence is bounded above and below, but bounded-
ness does not imply convergence.

422
Cauchy Convergence Criterion. Note that there is something a little fishy about the above definition. We
should be able to say if a sequence converges without first finding the constant to which it converges. We fix this
problem with the Cauchy convergence criterion. A sequence {an } converges if and only if for any  > 0 there
exists an N such that |an − am | <  for all n, m > N . The Cauchy convergence criterion is equivalent to the
definition we had before. For some problems it is handier to use. Now we don’t need to know the limit of a
sequence to show that it converges.
P∞ PN −1
Convergence of Series. The series n=1 an converges if the sequence of partial sums, SN = n=0 an , con-
verges. That is,
N
X −1
lim SN = lim an = constant.
N →∞ N →∞
n=0

If the limit does not exist, then the series diverges. A necessary condition for the convergence of a series is that

lim an = 0.
n→∞

Otherwise the sequence of partial sums would not converge.

Example 14.1.2 The series ∞ n

P
n=0 (−1) = 1 − 1 + 1 − 1 + · · · is divergent because the sequence of partial sums,
{SN } = 1, 0, 1, 0, 1, 0, . . . is divergent.

P∞ P∞
Tail of a Series. An infinite
P∞ series, n=0 an , converges or diverges with its tail. That is, for fixed N , n=0 an
converges if and only if n=N an converges. This is because the sum of the first N terms of a series is just a
number. Adding or subtracting a number to a series does not change its convergence.

Absolute Convergence. The series ∞

P P∞
n=0 an converges absolutely if n=0 |an | converges. Absolute convergence
implies convergence. If a series is convergent, but not absolutely convergent, then it is said to be conditionally
convergent.

423
 The terms of an absolutely convergent series can be rearranged in any order and the series will still converge
to the same sum. This is not true of conditionally convergent series. Rearranging the terms of a conditionally
convergent series may change the sum. In fact, the terms of a conditionally convergent series may be rearranged
to obtain any desired sum.

Example 14.1.3 The alternating harmonic series,

1 1 1
1− + − + ··· ,
2 3 4

converges, (Exercise 14.2). Since

1 1 1
1+ + + + ···
2 3 4

diverges, (Exercise 14.3), the alternating harmonic series is not absolutely convergent. Thus the terms can be
rearranged to obtain any sum, (Exercise 14.4).

P∞
Finite Series and Residuals. Consider the series f (z) = n=0 an (z). We will denote the sum of the first N
terms in the series as
N
X −1
SN (z) = an (z).
n=0

We will denote the residual after N terms as

∞
X
RN (z) ≡ f (z) − SN (z) = an (z).
n=N

424
14.1.2 Special Series
1
Geometric Series. One of the most important series in mathematics is the geometric series,
∞
X
zn = 1 + z + z2 + z3 + · · · .
n=0

The series
Pclearly diverges for |z| ≥ 1 since the terms do not vanish as n → ∞. Consider the partial sum,
N −1 n
SN (z) ≡ n=0 z , for |z| < 1.

N
X −1
(1 − z)SN (z) = (1 − z) zn
n=0
N
X −1 XN
= zn − zn
n=0 n=1
= (1 + z + · · · + z N −1 ) − (z + z 2 + · · · + z N )
= 1 − zN

N −1
X 1 − zN 1
zn = → as N → ∞.
n=0
1−z 1−z

1
The limit of the partial sums is 1−z
.

∞
X 1
zn = for |z| < 1
n=0
1−z

1
The series is so named because the terms grow or decay geometrically. Each term in the series is a constant times the previous
term.

425
Harmonic Series. Another important series is the harmonic series,
∞
X 1 1 1
α
= 1 + α + α + ··· .
n=1
n 2 3

The series is absolutely convergent for <(α) > 1 and absolutely divergent for <(α) ≤ 1, (see the Exercise 14.6).
The Riemann zeta function ζ(α) is defined as the sum of the harmonic series.
∞
X 1
ζ(α) =
n=1
nα
The alternating harmonic series is
∞
X (−1)n+1 1 1 1
=1− + − + ··· .
n=1
nα 2 α 3 α 4α

Again, the series is absolutely convergent for <(α) > 1 and absolutely divergent for <(α) ≤ 1.

14.1.3 Convergence Tests

P∞
The
P∞ Comparison Test. The series of positive
P∞ terms n=0 an converges if there exists a convergent
P∞ series
n=0 bn such that an ≤ bn for all n. Similarly, n=0 an diverges if there exists a divergent series n=0 bn such
that an ≥ bn for all n.

Example 14.1.4 Consider the series

∞
X 1
.
n=1
2n2
We can rewrite this as
∞
X 1
.
n=1
2n
n a perfect square

426
Then by comparing this series to the geometric series,
∞
X 1
n
= 1,
n=1
2

we see that it is convergent.

Integral Test. If the coefficients an of a series ∞

P
n=0 an are monotonically decreasing and can be extended to
a monotonically decreasing function of the continuous variable x,

a(x) = an for x ∈ Z0+ ,

then the series converges or diverges with the integral

Z ∞
a(x) dx.
0

P∞ 1
Example 14.1.5 Consider the series n=1 n2 . Define the functions sl (x) and sr (x), (left and right),

1 1
sl (x) = , sr (x) = .
(dxe)2 (bxc)2

Recall that bxc is the greatest integer function, the greatest integer which is less than or equal to x. dxe is the
least integer function, the least integer greater than or equal to x. We can express the series as integrals of these
functions.
∞ Z ∞ Z ∞
X 1
= sl (x) dx = sr (x) dx
n=1
n2 0 1

427
In Figure 14.1 these functions are plotted against y = 1/x2 . From the graph, it is clear that we can obtain a lower
and upper bound for the series.
Z ∞ ∞ Z ∞
1 X 1 1
dx ≤ ≤1+ dx
1 x2 n=1
n 2
1 x2
∞
X 1
1≤ ≤2
n=1
n2

1 1

1 2 3 4 1 2 3 4
P∞
Figure 14.1: Upper and Lower bounds to n=1 1/n2 .

In general, we have
Z ∞ ∞
X Z ∞
a(x) dx ≤ an ≤ am + a(x) dx.
m n=m m

Thus we see that the sum converges or diverges with the integral.
P∞
The Ratio Test. The series n=0 an converges absolutely if
 
 an+1 
lim   < 1.
n→∞  an 

428
If the limit is greater than unity, then the series diverges. If the limit is unity, the test fails.
If the limit is greater than unity, then the terms are eventually increasing with n. Since the terms do not
vanish, the sum is divergent. If the limit is less than unity, then there exists some N such that
 
 an+1 
 an  ≤ r < 1 for all n ≥ N.
 

P∞
From this we can show that n=0 an is absolutely convergent by comparing it to the geometric series.
∞
X ∞
X
|an | ≤ |aN | rn
n=N n=0
1
= |aN |
1−r

Example 14.1.6 Consider the series,

∞
X en
.
n=1
n!

We apply the ratio test to test for absolute convergence.

n+1
 
 an+1 
lim   = lim e n!
n→∞ an  n→∞ en (n + 1)!
e
= lim
n→∞ n + 1

=0

The series is absolutely convergent.

429
Example 14.1.7 Consider the series,

∞
X 1
,
n=1
n2

which we know to be absolutely convergent. We apply the ratio test.

2
 
 an+1 
lim   = lim 1/(n + 1)
n→∞ an  n→∞ 1/n2
n2
= lim 2
n→∞ n + 2n + 1
1
= lim
n→∞ 1 + 2/n + 1/n2

=1

The test fails to predict the absolute convergence of the series.

P∞
The Root Test. The series n=0 an converges absolutely if

lim |an |1/n < 1.

n→∞

If the limit is greater than unity, then the series diverges. If the limit is unity, the test fails.
If the limit is greater than unity, then the terms in the series do not vanish as n → ∞. This implies that the
sum does not converge. If the limit is less than unity, then there exists some N such that

|an |1/n ≤ r < 1 for all n ≥ N.

430
We bound the tail of the series of |an |.
∞
X ∞
X
|an | = (|an |1/n )n
n=N n=N
X∞
≤ rn
n=N
N
r
=
1−r
P∞
n=0 an is absolutely convergent.

Example 14.1.8 Consider the series

∞
X
na bn ,
n=0

where a and b are real constants. We use the root test to check for absolute convergence.

lim |na bn |1/n < 1

n→∞

|b| lim na/n < 1

n→∞ 
1 log n
|b| exp lim <1
n→∞ n
|b| e0 < 1
|b| < 1

Thus we see that the series converges absolutely for |b| < 1. Note that the value of a does not affect the absolute
convergence.

431
Example 14.1.9 Consider the absolutely convergent series,
∞
X 1
.
n=1
n2

We aply the root test.

 1/n
1/n
1
lim |an | = lim  2 
n→∞ n→∞ n

= lim n−2/n
n→∞
2
= lim e− n log n
n→∞
0
= e
=1

It fails to predict the convergence of the series.

14.2 Uniform Convergence

Continuous Functions. A function f (z) is continuous in a closed domain if, given any  > 0, there exists a
δ > 0 such that |f (z) − f (ζ)| <  for all |z − ζ| < δ in the domain.
An equivalent definition is that f (z) is continuous in a closed domain if

lim f (ζ) = f (z)

ζ→z

for all z in the domain.

432
Convergence. Consider a series in which the terms are functions of z, ∞
P
n=0 an (z). The series is convergent
P∞ in a
domain if the series converges for each point z in the domain. We can then define the function f (z) = n=0 an (z).
We can state the convergence criterion as: For any given  > 0 there exists a function N (z) such that
 
 N (z)−1 
 X 
|f (z) − SN (z) (z)| = f (z) − an (z) < 
 n=0 

for all z in the domain. Note that the rate of convergence, i.e. the number of terms, N (z) required for for the
absolute error to be less than , is a function of z.
P∞
Uniform Convergence. Consider a series n=0 an (z) that is convergent in some domain. If the rate of
convergence is independent of z then the series is said to be uniformly convergent. Stating this a little more
mathematically, the series is uniformly convergent in the domain if for any given  > 0 there exists an N ,
independent of z, such that
 N

 X 
|f (z) − SN (z)| = f (z) − an (z) < 
 
 
n=1

for all z in the domain.

14.2.1 Tests for Uniform Convergence

Weierstrass
P∞ M-test. The Weierstrass M-test is useful in determining if a series is uniformly convergent. The
series Pn=0 an (z) is uniformly and absolutely convergent in a domain if there exists a convergent series of positive
terms ∞ n=0 Mn such that |an (z)| ≤ Mn for all z in the domain. This condition first implies that the series is
absolutely convergent for all z in the domain. The condition |an (z)| ≤ Mn also ensures that the rate of convergence
is independent of z, which is the criterion for uniform convergence.
Note that absolute convergence and uniform convergence are independent. A series of functions may be
absolutely convergent without being uniformly convergent or vice versa. The Weierstrass M-test is a sufficient

433
but not a necessary condition for uniform convergence. The Weierstrass M-test can succeed only if the series is
uniformly and absolutely convergent.

Example 14.2.1 The series

∞
X sin x
f (x) =
n=1
n(n + 1)

sin x 1
P∞ 1
is uniformly and absolutely convergent for all real x because | n(n+1) |< n2
and n=1 n2 converges.

Dirichlet Test. Consider a sequence of monotone decreasing, positive constants cn with limit zero. If all the
partial sums of an (z) are bounded in some closed domain, that is
 
X N 
an (z) < constant
 

 n=1 

for all N , then ∞

P
n=1 cn an (z) is uniformly convergent in that closed domain. Note that the Dirichlet test does not
imply that the series is absolutely convergent.

Example 14.2.2 Consider the series,

∞
X sin(nx)
.
n=1
n

We cannot use the Weierstrass M-test to determine if the series is uniformly convergent on an interval. While it
is easy to bound the terms with | sin(nx)/n| ≤ 1/n, the sum
∞
X 1
n=1
n

434
 PN −1
does not converge. Thus we will try the Dirichlet test. Consider the sum n=1 sin(nx). This sum can be
evaluated in closed form. (See Exercise 14.7.)
−1
N
(
X 0 for x = 2πk
sin(nx) = cos(x/2)−cos((N −1/2)x)
n=1 2 sin(x/2)
for x 6= 2πk
The partial sums have infinite discontinuities at x = 2πk, k ∈ Z. The partial sums arePbounded on any closed
interval that does not contain an integer multiple of 2π. By the Dirichlet test, the sum ∞n=1
sin(nx)
n
is uniformly
convergent on any such closed interval. The series may not be uniformly convergent in neighborhoods of x = 2kπ.

14.2.2 Uniform Convergence and Continuous Functions.

Consider a series f (z) = ∞
P
n=1 an (z) that is uniformly convergent in some domain and whose terms an (z) are
continuous functions. Since the series is uniformly convergent, for any given  > 0 there exists an N such that
|RN | <  for all z in the domain.
Since the finite sum SN is continuous, for that  there exists a δ > 0 such that |SN (z) − SN (ζ)| <  for all ζ
in the domain satisfying |z − ζ| < δ.
Combining these two results,
|f (z) − f (ζ)| = |SN (z) + RN (z) − SN (ζ) − RN (ζ)|
≤ |SN (z) − SN (ζ)| + |RN (z)| + |RN (ζ)|
< 3 for |z − ζ| < δ.
Thus f (z) is continuous.

Result 14.2.1 A uniformly convergent series of continuous terms represents a continuous

function.

Example 14.2.3 Again consider ∞ sin(nx)

P
n=1 n
. In Example 14.2.2 we showed that the convergence is uniform in
any closed interval that does not contain an integer multiple of 2π. In Figure 14.2 is a plot of the first 10 and

435
then 50 terms in the series and finally the function to which the series converges. We see that the function has
jump discontinuities at x = 2kπ and is continuous on any closed interval not containing one of those points.

P∞ sin(nx)
Figure 14.2: Ten, Fifty and all the Terms of n=1 n
.

14.3 Uniformly Convergent Power Series

Power Series. Power series are series of the form
∞
X
an (z − z0 )n .
n=0

Domain of Convergence of a Power Series Consider the series ∞ n

P
n=0 an z . Let the series converge at some
point z0 . Then |an z0n | is bounded by some constant A for all n, so
 n  n
n  z 
n
  z
|an z | = |an z0 |   < A  
z0 z0

This comparison test shows that the series converges absolutely for all z satisfying |z| < |z0 |.

436
 Suppose that the series diverges at some point z1 . Then the series could not converge for any |z| > |z1 | since
this would imply convergence at z1 . Thus there exists some circle in the z plane such that the power series
converges absolutely inside the circle and diverges outside the circle.

Result 14.3.1 The domain of convergence of a power series is a circle in the complex
plane.

Radius of Convergence of Power Series. Consider a power series

∞
X
f (z) = an z n
n=0

Applying the ratio test, we see that the series converges if

|an+1 z n+1 |
lim <l
n→∞ |an z n |
|an+1 |
lim |z| < 1
n→∞ |an |

|an |
|z| < lim
n→∞ |an+1 |

437
 Result 14.3.2 The radius of convergence of the power series
∞
X
f (z) = an z n
n=0

is
|an |
R = lim
n→∞ |an+1 |

when the limit exists.

Result 14.3.3 Cauchy-Hadamard formula. The radius of convergence of the power

series:
∞
X
an z n
n=0

is
1
R= p .
lim sup n
|an |

Absolute Convergence of Power Series. Consider a power series

∞
X
f (z) = an z n
n=0

438
that converges for z = z0 . Let M
P∞be the value of the greatest term, an z0n . Consider any point z such that |z| < |z0 |.
We can bound the residual of n=0 |an z n |,
∞
X
RN (z) = |an z n |
n=N
∞
 an z n 
X 
n
=  an z n  |an z0 |
 
n=N 0
∞
Xz
 n
≤M  
 z0 
n=N

Since |z/z0 | < 1, this is a convergent geometric series.

 N
z 1
= M  
z0 1 − |z/z0 |
→ 0 as N → ∞
Thus the power series is absolutely convergent for |z| < |z0 |.
P∞ n
Result 14.3.4 If the power series n=0 an z converges for z = z0 , then the series con-
verges absolutely for |z| < |z0 |.

Example 14.3.1 Find the radii of convergence of

∞
X ∞
X ∞
X
n n
1) nz , 2) n!z , 3) n!z n!
n=1 n=1 n=1

1. Applying the formula for the radius of convergence,

 
 an 
R = lim   = lim n = 1
n→∞ an+1  n→∞ n + 1

439
2. Applying the ratio test to the second series,
 
 n! 
R = lim 
 
n→∞ (n + 1)! 
1
= lim
n→∞ n + 1

=0

Thus we see that the second series has a vanishing radius of convergence.

3. The third series converges when

 (n + 1)!z (n+1)! 
 
lim  <1
n→∞  n!z n! 
lim (n + 1)|z|(n+1)!−n! < 1
n→∞

lim (n + 1)|z|(n)n! < 1

n→∞
lim (log(n + 1) + (n)n! log |z|) < 0
n→∞
− log(n + 1)
log |z| < lim
n→∞ (n)n!
log |z| < 0
|z| < 1

Thus the radius of convergence for the third series is 1.

Alternatively we could determine the radius of convergence of the third series with the comparison test. We
know that
∞
X ∞
 n!  X
n!z  ≤ |nz n |
n=1 n=1

440
 P∞
n=1 nz n has a radius of convergence of 1. Thus the third sum must have a radius of convergence of at
least 1. Note that if |z| > 1 then the terms in the third series do not vanish as n → ∞. Thus the series
must diverge for all |z| > 1. We see that the radius of convergence is 1.

P∞ n
Uniform Convergence of Power Series. Consider a power series n=0 an z that converges in the disk
n n
|z|
P∞< r0 . nThe sum converges absolutely for z in the closed disk, |z| ≤ r < r0 . Since |an z | ≤ |an r | and
n=0 |an r | converges, the power series is uniformly convergent in |z| ≤ r < r0 .
P∞ n
Result 14.3.5 If the power series n=0 an z converges for |z| < r0 then the series con-
verges uniformly for |z| ≤ r < r0 .

Example 14.3.2 Convergence and Uniform Convergence. Consider the series

∞
X zn
log(1 − z) = − .
n=1
n

This series converges for |z| ≤ 1, z 6= 1. Is the series uniformly convergent in this domain? The residual after N
terms RN is
∞
X zn
RN (z) = .
n=N +1
n

We can get a lower bound on the absolute value of the residual for real, positive z.
∞
X xn
|RN (x)| =
n=N +1
n
Z ∞ α
x
≤ dα
N +1 α
= − Ei ((N + 1) log x)

441
The exponential integral function, Ei (z), is defined
∞
e−t
Z
Ei (z) = − dt.
−z t

The exponential integral function is plotted in Figure 14.3. Since Ei (z) diverges as z → 0, by choosing x
sufficiently close to 1 the residual can be made arbitrarily large. Thus this series is not uniformly convergent in
the domain |z| ≤ 1, z 6= 1. The series is uniformly convergent for |z| ≤ r < 1.

-4 -3 -2 -1
-0.25

-0.5

-0.75

-1

-1.25

-1.5

-1.75

-2

Figure 14.3: The Exponential Integral Function.

Analyticity.
H Recall that a sufficient condition for the analyticity of a function f (z) in a domain is that
C
f (z) dz = 0 for all simple, closed contours in the domain.

442
 Consider a power series fH(z) = ∞ n
P
n=0 an z that is uniformly convergent in |z| ≤ r. If C is any simple, closed
contour in the domain then C f (z) dz exists. Expanding f (z) into a finite series and a residual,
I I
f (z) dz = [SN (z) + RN (z)] dz.
C C
Since the series is uniformly convergent, for any given  > 0 there exists an N such that |RN | <  for all z in
|z| ≤ r. If L is the length of the contour C then
I 
 
 RN (z) dz  ≤ L → 0 as N → ∞.
 
C

−1
I I N
!
X
f (z) dz = lim an z n + RN (z) dz
C N →∞ C n=0
∞
I X
= an z n
C n=0
∞
X I
= an z n dz
n=0 C

= 0.
Thus f (z) is analytic for |z| < r.

Result 14.3.6 A power series is analytic in its domain of uniform convergence.

14.4 Integration and Differentiation of Power Series

Consider a power series f (z) = ∞ n
P
n=0 an z that is convergent in the disk |z| < r0 . Let C be any contour of finite
length L lying entirely within the closed domain |z| ≤ r < r0 . The integral of f (z) along C is
Z Z
f (z) dz = [SN (z) + RN (z)] dz.
C C

443
Since the series is uniformly convergent in the closed disk, for any given  > 0, there exists an N such that

|RN (z)| <  for all |z| ≤ r.

Bounding the absolute value of the integral of RN (z),

Z  Z
 
 RN (z) dz  ≤ |RN (z)| dz
 
C C
< L
→ 0 as N → ∞

Thus
Z N
Z X
f (z) dz = lim an z n dz
C N →∞ C n=0
N
X Z
= lim an z n dz
N →∞ C
n=0
∞
X Z
= an z n dz
n=0 C

ResultP14.4.1 If C is a contour lying in the domain of uniform convergence of the power

∞ n
series n=0 an z then
Z X ∞ X∞ Z
n
an z dz = an z n dz.
C n=0 n=0 C

In the domain of uniform convergence of a series we can interchange the order of summation and a limit

444
process. That is,
∞
X ∞
X
lim an (z) = lim an (z).
z→z0 z→z0
n=0 n=0

We can do this because the rate of convergence does not depend on z. Since differentiation is a limit process,
d f (z + h) − f (z)
f (z) = lim ,
dz h→0 h
we would expect that we could differentiate a uniformly convergent series.
Since we showed that a uniformly convergent power series is equal to an analytic function, we can differentiate
a power series in it’s domain of uniform convergence.
Result 14.4.2 In the domain of uniform convergence of a power series
∞ ∞
d X X
an z n = (n + 1)an+1 z n .
dz n=0 n=0

Example 14.4.1 Differentiating a Series. Consider the series from Example 14.3.2
∞
X zn
log(1 − z) = − .
n=1
n

Differentiating this series yields

∞
1 X
− =− z n−1
1−z n=1
∞
1 X
= zn.
1−z n=0

445
We recognize this as the geometric series, which is convergent for |z| < 1 and uniformly convergent for |z| ≤ r < 1.
Note that the domain of convergence is different than the series for log(1 − z). The geometric series does not
converge for |z| = 1, z 6= 1. However, the domain of uniform convergence has remained the same.

14.5 Taylor Series

Result 14.5.1 Taylor’s Theorem. Let f (z) be a function that is single-valued and
analytic in |z − z0 | < R. For all z in this open disk, f (z) has the convergent Taylor series
∞
X f (n) (z0 )
f (z) = (z − z0 )n . (14.1)
n=0
n!

We can also write this as

∞
f (n) (z0 )
I
X 1 f (z)
f (z) = an (z − z0 )n , an = = dz, (14.2)
n=0
n! i2π C (z − z0 )n+1

where C is a simple, positive, closed contour in 0 < |z − z0 | < R that goes once around
the point z0 .

Proof of Taylor’s Theorem. Let’s see why Result 14.5.1 is true. Consider a function f (z) that is analytic in
|z| < R. (Considering z0 6= 0 is only trivially more general as we can introduce the change of variables ζ = z − z0 .)
According to Cauchy’s Integral Formula, (Result ??),
I
1 f (ζ)
f (z) = dζ, (14.3)
i2π C ζ − z

446
where C is a positive, simple, closed contour in 0 < |ζ − z| < R that goes once around z. We take this contour
to be the circle about the origin of radius r where |z| < r < R. (See Figure 14.4.)

Im(z)

r R
Re(z)
C z

Figure 14.4: Graph of Domain of Convergence and Contour of Integration.

1
We expand ζ−z
in a geometric series,
1 1/ζ
=
ζ −z 1 − z/ζ
∞  n
1X z
= , for |z| < |ζ|
ζ n=0 ζ
∞
X zn
= n+1
, for |z| < |ζ|
n=0
ζ

We substitute this series into Equation 14.3.

∞
!
f (ζ)z n
I
1 X
f (z) = dζ
i2π C n=0
ζ n+1

447
The series converges uniformly so we can interchange integration and summation.

∞ I
X 1 f (ζ)
= n+1
dζ z n
n=0
i2π C ζ

Now we have derived Equation 14.2. To obtain Equation 14.1, we apply Cauchy’s Integral Formula.

∞
X f (n) (0)
= zn
n=0
n!

There is a table of some commonly encountered Taylor series in Appendix H.

Example 14.5.1 Consider the Taylor series Pexpansion of 1/(1 − z) about z = 0. Previously, we showed that this
∞ n
function is the sum of the geometric series n=0 z and we used the ratio test to show that the series converged
absolutely for |z| < 1. Now we find the series using Taylor’s theorem. Since the nearest singularity of the function
is at z = 1, the radius of convergence of the series is 1. The coefficients in the series are

1 dn 1
 
an =
n! dz n 1 − z z=0
 
1 n!
=
n! (1 − z)n z=0
=1

Thus we have
∞
1 X
= zn, for |z| < 1.
1−z n=0

448
14.5.1 Newton’s Binomial Formula.

Result 14.5.2 For all |z| < 1, a complex:

     
a a a 3
(1 + z)a = 1 + z+ z2 + z + ···
1 2 3
where
 
a a(a − 1)(a − 2) · · · (a − r + 1)
= .
r r!
If a is complex, then the expansion is of the principle branch of (1 + z)a . We define
     
r 0 0
= 1, = 0, for r 6= 0, = 1.
0 r 0

Example 14.5.2 Evaluate limn→∞ (1 + 1/n)n .

First we expand (1 + 1/n)n using Newton’s binomial formula.

 n        
1 n n 2 n 3
lim 1 + = lim 1 + 1/n + 1/n + 1/n + · · ·
n→∞ n n→∞ 1 2 3
 
n(n − 1) n(n − 1)(n − 2)
= lim 1 + 1 + + + ···
n→∞ 2! n2 3! n3
 
1 1
= 1 + 1 + + + ···
2! 3!

449
We recognize this as the Taylor series expansion of e1 .

=e

We can also evaluate the limit using L’Hospital’s rule.

  x 
1
Log lim 1 + = lim Log ((1 + 1/x)x )
x→∞ x x→∞

= lim x Log (1 + 1/x)

x→∞
Log (1 + 1/x)
= lim
x→∞ 1/x
−1/x2
1+1/x
= lim
x→∞ −1/x2

=1

 x
1
lim 1 + = e1
x→∞ x

Example 14.5.3 Find the Taylor series expansion of 1/(1 + z) about z = 0.

For |z| < 1,
     
1 −1 −1 2 −1 3
=1+ z+ z + z + ···
1+z 1 2 3
= 1 + (−1)1 z + (−1)2 z 2 + (−1)3 z 3 + · · ·
= 1 − z + z2 − z3 + · · ·

450
Example 14.5.4 Find the first few terms in the Taylor series expansion of

1
√
2
z + 5z + 6

about the origin.

We factor the denominator and then apply Newton’s binomial formula.

1 1 1
√ =√ √
z 2 + 5z + 6 z+3 z+2
1 1
=√ p √ p
3 1 + z/3 2 1 + z/2
          
1 −1/2 z −1/2  z 2 −1/2 z −1/2  z 2
= √ 1+ + + ··· 1 + + + ···
6 1 3 2 3 1 2 2 2
z z2 z 3z 2
  
1
= √ 1− + + ··· 1 − + + ···
6 6 24 4 32
 
1 5 17 2
= √ 1 − z + z + ···
6 12 96

451
14.6 Laurent Series

Result 14.6.1 Let f (z) be single-valued and analytic in the annulus R1 < |z − z0 | < R2 .
For points in the annulus, the function has the convergent Laurent series
∞
X
f (z) = an z n ,
n=−∞

where
I
1 f (z)
an = dz
2πi C (z − z0 )n+1
and C is a positively oriented, closed contour around z0 lying in the annulus.
To derive this result, consider a function f (ζ) that is analytic in the annulus R1 < |ζ| < R2 . Consider any
point z in the annulus. Let C1 be a circle of radius r1 with R1 < r1 < |z|. Let C2 be a circle of radius r2 with
|z| < r2 < R2 . Let Cz be a circle around z, lying entirely between C1 and C2 . (See Figure 14.5 for an illustration.)
Consider the integral of fζ−z
(ζ)
around the C2 contour. Since the the only singularities of f (ζ)
ζ−z
occur at ζ = z and
at points outside the annulus,
I I I
f (ζ) f (ζ) f (ζ)
dζ = dζ + dζ.
C2 ζ −z Cz ζ −z C1 ζ −z

By Cauchy’s Integral Formula, the integral around Cz is

I
f (ζ)
dζ = i2π f (z).
Cz ζ −z

452
This gives us an expression for f (z).
I I
1 f (ζ) 1 f (ζ)
f (z) = dζ − dζ (14.4)
2πi C2 ζ −z 2πi C1 ζ −z
On the C2 contour, |z| < |ζ|. Thus
1 1/ζ
=
ζ −z 1 − z/ζ
∞  n
1X z
= , for |z| < |ζ|
ζ n=0 ζ
∞
X zn
= , for |z| < |ζ|
n=0
ζ n+1

On the C1 contour, |ζ| < |z|. Thus

1 1/z
− =
ζ −z 1 − ζ/z
∞  n
1X ζ
= , for |ζ| < |z|
z n=0 z
∞
X ζn
= n+1
, for |ζ| < |z|
n=0
z
−1
X zn
= , for |ζ| < |z|
n=−∞
ζ n+1

We substitute these geometric series into Equation 14.4.

∞
! −1
!
f (ζ)z n f (ζ)z n
I I
1 X 1 X
f (z) = dζ + dζ
i2π C2 n=0 ζ n+1 i2π C1 n=−∞
ζ n+1

453
Since the sums converge uniformly, we can interchange the order of integration and summation.
∞ I −1 I
1 X f (ζ)z n 1 X f (ζ)z n
f (z) = dζ + dζ
i2π n=0 C2 ζ n+1 i2π n=−∞ C1 ζ n+1

Since the only singularities of the integrands lie outside of the annulus, the C1 and C2 contours can be deformed
to any positive, closed contour C that lies in the annulus and encloses the origin. (See Figure 14.5.) Finally, we
combine the two integrals to obtain the desired result.
∞ I 
X 1 f (ζ)
f (z) = n+1
dζ z n
n=−∞
i2π C ζ

For the case of arbitrary z0 , simply make the transformation z → z − z0 .

Example 14.6.1 Find the Laurent series expansions of 1/(1 + z).

For |z| < 1,
     
1 −1 −1 2 −1 3
=1+ z+ z + z + ···
1+z 1 2 3
= 1 + (−1)1 z + (−1)2 z 2 + (−1)3 z 3 + · · ·
= 1 − z + z2 − z3 + · · ·

For |z| > 1,

1 1/z
=
1+z 1 + 1/z
     
1 −1 −1 −1 −2
= 1+ z + z + ···
z 1 2
= z −1 − z −2 + z −3 − · · ·

454
 Im(z) Im(z)

r2 R2 R2

r1 R1 R1
Re(z) Re(z)
C1
z C
C2 Cz

Figure 14.5: Contours for a Laurent Expansion in an Annulus.

14.7 Exercises
Exercise 14.1 (mathematica/fcv/series/constants.nb)
Does the series
∞
X 1
n=2
n log n

converge?
Hint, Solution

455
Exercise 14.2 (mathematica/fcv/series/constants.nb)
Show that the alternating harmonic series,

∞
X (−1)n+1 1 1 1
=1− + − + ··· ,
n=1
n 2 3 4

is convergent.
Hint, Solution

Exercise 14.3 (mathematica/fcv/series/constants.nb)

Show that the series

∞
X 1
n=1
n

is divergent with the Cauchy convergence criterion.

Hint, Solution

Exercise 14.4
The alternating harmonic series has the sum:

∞
X (−1)n
= log(2).
n=1
n

Show that the terms in this series can be rearranged to sum to π.

Hint, Solution

456
Exercise 14.5 (mathematica/fcv/series/constants.nb)
Is the series,
∞
X n!
n
,
n=1
n
convergent?
Hint, Solution

Exercise 14.6
Show that the harmonic series,
∞
X 1 1 1
α
= 1 + α + α + ··· ,
n=1
n 2 3
converges for α > 1 and diverges for α ≤ 1.
Hint, Solution

ExerciseP 14.7
−1
Evaluate N n=1 sin(nx).
Hint, Solution

Exercise 14.8
Using the geometric series, show that
∞
1 X
= (n + 1)z n , for |z| < 1,
(1 − z)2 n=0

and
∞
X zn
log(1 − z) = − , for |z| < 1.
n=1
n

457
Hint, Solution

Exercise 14.9
1
Find the Taylor series of 1+z 2 about the z = 0. Determine the radius of convergence of the Taylor series from the

singularities of the function. Determine the radius of convergence with the ratio test.
Hint, Solution

Exercise 14.10
Use two methods to find the Taylor series expansion of log(1 + z) about z = 0 and determine the circle of
convergence. First directly apply Taylor’s theorem, then differentiate a geometric series.
Hint, Solution

Exercise 14.11
Find the Laurent series about z = 0 of 1/(z − i) for |z| < 1 and |z| > 1.
Hint, Solution

Exercise 14.12
Evaluate

n
X n
X
k
kz and k2z k
k=1 k=1

for z 6= 1.
Hint, Solution

Exercise 14.13
Find the circle of convergence of the following series.

458
 z2 z3 z4
1. z + (α − β) + (α − β)(α − 2β) + (α − β)(α − 2β)(α − 3β) + · · ·
2! 3! 4!
∞
X n
2. n
(z − i)n
n=1
2
∞
X
3. nn z n
n=1

∞
X n! n
4. n
z
n=1
n
∞
X
5. [3 + (−1)n ]n z n
n=1

∞
X
6. (n + αn ) z n (|α| > 1)
n=1

Hint, Solution

Exercise 14.14
Let f (z) = (1 + z)α be the branch for which f (0) = 1. Find its Taylor series expansion about z = 0. What is the
radius of convergence of the series? (α is an arbitrary complex number.)
Hint, Solution

Exercise 14.15
Obtain the Laurent expansion of
1
f (z) =
(z + 1)(z + 2)

459
centered on z = 0 for the three regions:
1. |z| < 1
2. 1 < |z| < 2
3. 2 < |z|
Hint, Solution

Exercise 14.16
By comparing the Laurent expansion of (z + 1/z)m , m ∈ Z+ , with the binomial expansion of this quantity, show
that
(
π m
Z 2π

m 2m−1 (m−n)/2
−m ≤ n ≤ m and m − n even
(cos θ) cos(nθ) dθ =
0 0 otherwise
Hint, Solution

Exercise 14.17
The function f (z) is analytic in the entire z-plane, including ∞, except at the point z = i/2, where it has a simple
pole, and at z = 2, where it has a pole of order 2. In addition
I I I
f (z) dz = 2πi, f (z) dz = 0, (z − 1)f (z) dz = 0.
|z|=1 |z|=3 |z|=3

Find f (z) and its complete Laurent expansion about z = 0.

Hint, Solution

Exercise 14.18
3 z k
Let f (z) = ∞
P

k=1 k 3
. Compute each of the following, giving justification in each case. The contours are
circles of radius one about the origin.

460
 Z
1. eiz f (z) dz
|z|=1
Z
f (z)
2. dz
|z|=1 z4

f (z) ez
Z
3. dz
|z|=1 z2
Hint, Solution

461
14.8 Hints
Hint 14.1
Use the integral test.

Hint 14.2
Group the terms.

1 1
1− =
2 2
1 1 1
− =
3 4 12
1 1 1
− =
5 6 30
···

Hint 14.3
Show that

1
|S2n − Sn | > .
2

Hint 14.4
The alternating harmonic series is conditionally convergent. Let {an } and {bP
n } be the positive and negative terms
in the sum, respectively, ordered in decreasing magnitude. Note that both ∞ n=1 an and sumoinbn are divergent.
Devise a method for alternately taking terms from {an } and {bn }.

Hint 14.5
Use the ratio test.

462
Hint 14.6
Use the integral test.

Hint 14.7
Note that sin(nx) = =( einx ). This substitute will yield a finite geometric series.

Hint 14.8
Differentiate the geometric series. Integrate the geometric series.

Hint 14.9
The Taylor series is a geometric series.

Hint 14.10

Hint 14.11

Hint 14.12
Let Sn be the sum. Consider Sn − zSn . Use the finite geometric sum.

Hint 14.13

463
Hint 14.14

Hint 14.15

Hint 14.16

Hint 14.17

Hint 14.18

464
14.9 Solutions
Solution
P∞ 14.1
Since n=2 is a series of positive, monotone decreasing terms, the sum converges or diverges with the integral,
Z ∞ Z ∞
1 1
dx = dξ
2 x log x log 2 ξ

Since the integral diverges, the series also diverges.

Solution 14.2

∞ ∞ 
(−1)n+1

X X 1 1
= −
n=1
n n=1
2n − 1 2n
∞
X 1
=
n=1
(2n − 1)(2n)
∞
X 1
<
n=1
(2n − 1)2
∞
1 X 1
<
2 n=1
n2
2
π
=
12

Thus the series is convergent.

465
Solution 14.3
Since
2n−1 
X 1
|S2n − Sn | = 
 
j


j=n
2n−1
X 1
≥
j=n
2n − 1
n
=
2n − 1
1
>
2
the series does not satisfy the Cauchy convergence criterion.

Solution 14.4
The alternating harmonic series is conditionally convergent. That is, the sum is convergent but not absolutely
convergent. Let {an } and {bP
n } be the positive
P∞ and negative terms in the sum, respectively, ordered in decreasing
magnitude. Note that both ∞ a
n=1 n and n=1 bn are divergent. Otherwise the alternating harmonic series would
be absolutely convergent.
To sum the terms in the series to π we repeat the following two steps indefinitely:

1. Take terms from {an } until the sum is greater than π.

2. Take terms from {bn } until the sum is less than π.

P∞ P∞
Each of these steps can always be accomplished because the sums, n=1 an and n=1 bn are both divergent.
Hence the tails of the series are divergent. No matter how many terms we take, the remaining terms in each series
are divergent. In each step a finite, nonzero number of terms from the respective series is taken. Thus all the
terms will be used. Since the terms in each series vanish as n → ∞, the running sum converges to π.

466
Solution 14.5
Applying the ratio test,
n
 
 an+1 
lim   = lim (n + 1)!n
n→∞ an  n→∞ n!(n + 1)(n+1)
nn
= lim
n→∞ (n + 1)n
 n
n
= lim
n→∞ (n + 1)
1
=
e
< 1,

we see that the series is absolutely convergent.

Solution 14.6
The harmonic series,
∞
X 1 1 1
α
= 1 + α + α + ··· ,
n=1
n 2 3

converges or diverges absolutely with the integral,

x]∞
(
Z ∞
1
Z ∞
1 [log 1 i for <(α) = 1,
dx = dx = h 1−<(α)
x
∞
1 |xα | 1 x<(α) 1−<(α)
for <(α) 6= 1.
1

The integral converges only for <(α) > 1. Thus the harmonic series converges absolutely for <(α) > 1 and
diverges absolutely for <(α) ≤ 1.

467
Solution 14.7

N
X −1 N
X −1
sin(nx) = sin(nx)
n=1 n=0
N
X −1
= =( einx )
n=0
−1
N
!
X
== ( eix )n
n=0
(
=(N
 ) iN x  for x = 2πk
=
= 1− e
1− eix
for x 6= 2πk
(
0 for x = 2πk
= e−ix/2 − ei(N −1/2)x

= e−ix/2 − eix/2
for x 6= 2πk
(
0 for x = 2πk
= −ix/2 ei(N −1/2)x

= e −2i−sin(x/2) for x 6= 2πk
(
0 for x = 2πk
= −ix/2 − ei(N −1/2)x

< e 2 sin(x/2) for x 6= 2πk

−1
N
(
X 0 for x = 2πk
sin(nx) = cos(x/2)−cos((N −1/2)x)
n=1 2 sin(x/2)
6 2πk
for x =

468
Solution 14.8
The geometric series is

∞
1 X
= zn.
1−z n=0

This series is uniformly convergent in the domain, |z| ≤ r < 1. Differentiating this equation yields,
∞
1 X
= nz n−1
(1 − z)2 n=1
∞
X
= (n + 1)z n for |z| < 1.
n=0

Integrating the geometric series yields

∞
X z n+1
− log(1 − z) =
n=0
n+1

∞
X zn
log(1 − z) = − , for |z| < 1.
n=1
n

Solution 14.9

∞ ∞
1 X
2 n
X
= (−z ) = (−1)n z 2n
1 + z2 n=0 n=0

469
 1 1
The function 1+z 2 = (1−iz)(1+iz) has singularities at z = ±i. Thus the radius of convergence is 1. Now we use the

ratio test to corroborate that the radius of convergence is 1.

 
 an+1 (z) 
lim  <1
n→∞  an (z) 

 (−1)n+1 z 2(n+1) 
 
lim  <1
n→∞  (−1)n z 2n 
 
lim z 2  < 1
n→∞
|z| < 1

Solution 14.10

Method 1.
 2
z2
  
d z d
log(1 + z) = [log(1 + z)]z=0 + log(1 + z) + log(1 + z) + ···
dz z=0 1! dz 2 z=0 2!
z2 z3
     
1 z −1 2
=0+ + + + ···
1 + z z=0 1! (1 + z)2 z=0 2! (1 + z)3 z=0 3!
z2 z3 z4
=z− + − + ···
2 3 4
∞
X zn
= (−1)n+1
n=1
n

Since the nearest singularity of log(1 + z) is at z = −1, the radius of convergence is 1.

Method 2. We know the geometric series
∞
1 X
= (−1)n z n
1+z n=0

470
converges for |z| < 1. Integrating this equation yields

∞ n+1 ∞
X
n z
X zn
log(1 + z) = (−1) = (−1)n+1
n=0
n + 1 n=1 n

for |z| < 1. We calculate the radius of convergence with the ratio test.

   
 an   −(n + 1) 
R = lim   = lim  =1
n→∞ an+1  n→∞  n 

Thus the series converges absolutely for |z| < 1.

Solution 14.11
For |z| < 1:

1 i
=
z−i 1 + iz
X∞
=i (−iz)n
n=0

(Note that |z| < 1 ⇔ | − iz| < 1.)

For |z| > 1:

1 1 1
=
z−i z (1 − i/z)

471
(Note that |z| > 1 ⇔ | − i/z| < 1.)
∞  n
1X i
=
z n=0 z
0
1 X −n n
= i z
z n=−∞
0
X
= (−i)n z n−1
n=−∞
−1
X
= (−i)n+1 z n
n=−∞

Solution 14.12
Let
n
X
Sn = kz k .
k=1

n
X n
X
k
Sn − zSn = kz − kz k+1
k=1 k=1
Xn n+1
X
= kz k − (k − 1)z k
k=1 k=2
Xn
= z k − nz n+1
k=1
z − z n+1
= − nz n+1
1−z

472
 n
X z(1 − (n + 1)z n + nz n+1 )
kz k =
k=1
(1 − z)2

Let

n
X
Sn = k2z k .
k=1

n
X
Sn − zSn = (k 2 − (k − 1)2 )z k − n2 z n+1
k=1
Xn n
X
k
=2 kz − z k − n2 z n+1
k=1 k=1
z(1 − (n + 1)z n + nz n+1 ) z − z n+1
=2 − − n2 z n+1
(1 − z)2 1−z

n
X z(1 + z − z n (1 + z + n(n(z − 1) − 2)(z − 1)))
k2z k =
k=1
(1 − z)3

473
Solution 14.13
1. We assume that β 6= 0. We determine the radius of convergence with the ratio test.
 
 an 
R = lim  
n→∞ an+1 
 
 (α − β) · · · (α − (n − 1)β)/n! 
= lim  
n→∞ (α − β) · · · (α − nβ)/(n + 1)! 
 
 n+1 
= lim  
n→∞ α − nβ 
1
=
|β|
The series converges absolutely for |z| < 1/|β|.
2. By the ratio test formula, the radius of absolute convergence is
n
 
 n/2 
R = lim  
n→∞ (n + 1)/2n+1 
 
 n 
= 2 lim  
n→∞ n + 1 

=2

By the root test formula, the radius of absolute convergence is

1
R= p
limn→∞ n |n/2n |
2
= √
limn→∞ n n
=2

The series converges absolutely for |z − i| < 2.

474
3. We determine the radius of convergence with the Cauchy-Hadamard formula.

1
R= p
lim sup n |an |
1
= p
lim sup n |nn |
1
=
lim sup n
=0

The series converges only for z = 0.

475
4. By the ratio test formula, the radius of absolute convergence is
n
 
 n!/n 
R = lim  
n→∞ (n + 1)!/(n + 1)n+1 
 (n + 1)n 
 
= lim  
n→∞ nn 
 n
n+1
= lim
n→∞ n
  n 
n+1
= exp lim log
n→∞ n
  
n+1
= exp lim n log
n→∞ n
 
log(n + 1) − log(n)
= exp lim
n→∞ 1/n
 
1/(n + 1) − 1/n
= exp lim
n→∞ −1/n2
 
n
= exp lim
n→∞ n + 1
1
= e
The series converges absolutely in the circle, |z| < e.
5. By the Cauchy-Hadamard formula, the radius of absolute convergence is
1
R=
lim sup | (3 + (−1)n )n |
pn

1
=
lim sup (3 + (−1)n )
1
=
4

476
 Thus the series converges absolutely for |z| < 1/4.

6. By the Cauchy-Hadamard formula, the radius of absolute convergence is

1
R= p
lim sup n |n + αn |
1
= p
lim sup |α| |1 + n/αn |
n

1
=
|α|

Thus the sum converges absolutely for |z| < 1/|α|.

Solution 14.14
The Taylor series expansion of f (z) about z = 0 is

∞
X f (n) (0)
f (z) = zn.
n=0
n!

The derivatives of f (z) are

n−1
!
Y
(n)
f (z) = (α − k) (1 + z)α−n .
k=0

Thus f (n) (0) is

n−1
Y
(n)
f (0) = (α − k).
k=0

477
If α = m is a non-negative integer, then only the first m + 1 terms are nonzero. The Taylor series is a polynomial
and the series has an infinite radius of convergence.
m Qn−1
k=0 (α − k) n
X
m
(1 + z) = z
n=0
n!

If α is not a non-negative integer, then all of the terms in the series are non-zero.
∞ Qn−1
k=0 (α − k) n
X
α
(1 + z) = z
n=0
n!

The radius of convergence of the series is the distance to the nearest singularity of (1 + z)α . This occurs at z = 1.
Thus the series converges for |z| < 1. We can corroborate this with the ratio test. The radius of convergence is
 Qn−1

−
 

 Q k=0 (α k) /n!  n+1
 = | − 1| = 1.
R = lim  n  = lim 

k=0 (α − k)) /(n + 1)! 
n→∞  ( n→∞ α − n 

If we define the binomial coefficient,

  Qn−1
α (α − k)
≡ k=0 ,
n n!
then we can write the series as
∞  
α
X α n
(1 + z) = z .
n=0
n

Solution 14.15
We expand the function in partial fractions.
1 1 1
f (z) = = −
(z + 1)(z + 2) z+1 z+2

478
The Taylor series about z = 0 for 1/(z + 1) is
1 1
=
1+z 1 − (−z)
X∞
= (−z)n , for |z| < 1
n=0
∞
X
= (−1)n z n , for |z| < 1
n=0

The series about z = ∞ for 1/(z + 1) is

1 1/z
=
1+z 1 + 1/z
∞
1X
= (−1/z)n , for |1/z| < 1
z n=0
∞
X
= (−1)n z −n−1 , for |z| > 1
n=0
−1
X
= (−1)n+1 z n , for |z| > 1
n=−∞

The Taylor series about z = 0 for 1/(z + 2) is

1 1/2
=
2+z 1 + z/2
∞
1X
= (−z/2)n , for |z/2| < 1
2 n=0
∞
X (−1)n
= zn, for |z| < 2
n=0
2n+1

479
The series about z = ∞ for 1/(z + 2) is

1 1/z
=
2+z 1 + 2/z
∞
1X
= (−2/z)n , for |2/z| < 1
z n=0
∞
X
= (−1)n 2n z −n−1 , for |z| > 2
n=0
−1
X (−1)n+1 n
= z , for |z| > 2
n=−∞
2n+1

To find the expansions in the three regions, we just choose the appropriate series.

1.

1 1
f (z) = −
1+z 2+z
∞ ∞
X
n n
X (−1)n n
= (−1) z − n+1
z , for |z| < 1
n=0 n=0
2
∞  
X 1
= (−1) 1 − n+1 z n , for |z| < 1
n

n=0
2

∞
X 2n+1 − 1
f (z) = (−1)n n+1 z n , for |z| < 1
n=0
2

480
 2.
1 1
f (z) = −
1+z 2+z
−1 ∞
X X (−1)n
f (z) = (−1)n+1 z n − zn, for 1 < |z| < 2
n=−∞ n=0
2n+1

3.
1 1
f (z) = −
1+z 2+z
−1 −1
X X (−1)n+1 n
= (−1)n+1 z n − n+1
z , for 2 < |z|
n=−∞ n=−∞
2

−1
X 2n+1 − 1 n
f (z) = (−1)n+1 z , for 2 < |z|
n=−∞
2n+1

Solution 14.16
Laurent Series. We assume that m is a non-negative integer and that n is an integer. The Laurent series about
the point z = 0 of
 m
1
f (z) = z +
z
is
∞
X
f (z) = an z n
n=−∞

481
where
I
1 f (z)
an = dz
i2π C z n+1
and C is a contour going around the origin once in the positive direction. We manipulate the coefficient integral
into the desired form.
(z + 1/z)m
I
1
an = dz
i2π C z n+1
Z 2π iθ
1 ( e + e−iθ )m iθ
= i e dθ
i2π 0 ei(n+1)θ
Z 2π
1
= 2m cosm θ e−inθ dθ
2π 0
2m−1 2π
Z
= cosm θ(cos(nθ) − i sin(nθ)) dθ
π 0

Note that cosm θ is even and sin(nθ) is odd about θ = π.

2m−1 2π
Z
= cosm θ cos(nθ) dθ
π 0

Binomial Series. Now we find the binomial series expansion of f (z).

 m X m    n
1 m m−n 1
z+ = z
z n=0
n z
m  
X m m−2n
= z
n=0
n
m  
X m
= zn
n=−m
(m − n)/2
m−n even

482
 P∞
The coefficients in the series f (z) = n=−∞ an z n are
(
m


(m−n)/2
−m ≤ n ≤ m and m − n even
an =
0 otherwise
By equating the coefficients found by the two methods, we evaluate the desired integral.
(
π m
Z 2π

m m−1 −m ≤ n ≤ m and m − n even
(cos θ) cos(nθ) dθ = 2 (m−n)/2
0 0 otherwise

Solution 14.17
First we write f (z) in the form
g(z)
f (z) = .
(z − i/2)(z − 2)2
g(z) is an entire function which grows no faster that z 3 at infinity. By expanding g(z) in a Taylor series about
the origin, we see that it is a polynomial of degree no greater than 3.
αz 3 + βz 2 + γz + δ
f (z) = .
(z − i/2)(z − 2)2
Since f (z) is a rational function we expand it in partial fractions to obtain a form that is convenient to integrate.
a b c
f (z) = + + +d
z − i/2 z − 2 (z − 2)2
We use the value of the integrals of f (z) to determine the constants, a, b, c and d.
I  
a b c
+ + + d dz = i2π
|z|=1 z − i/2 z − 2 (z − 2)2
i2πa = i2π
a=1

483
 I  
1 b c
+ + +d dz = 0
|z|=3 z − i/2 z − 2 (z − 2)2
i2π(1 + b) = 0
b = −1

Note that by applying the second constraint, we can change the third constraint to
I
zf (z) dz = 0.
|z|=3

I  
1 1 c
z − + + d dz = 0
|z|=3 z − i/2 z − 2 (z − 2)2
 
(z − i/2) + i/2 (z − 2) + 2 c(z − 2) + 2c
I
− + dz = 0
|z|=3 z − i/2 z−2 (z − 2)2
 
i
i2π −2+c =0
2
i
c=2−
2
Thus we see that the function is

1 1 2 − i/2
f (z) = − + + d.
z − i/2 z − 2 (z − 2)2

where d is an arbitrary constant. We can also write the function in the form:

dz 3 + 15 − i8
f (z) = .
4(z − i/2)(z − 2)2

484
 Complete Laurent Series. We find the complete Laurent series about z = 0 for each of the terms in the
partial fraction expansion of f (z).

1 i2
=
z − i/2 1 + i2z
∞
X
= i2 (−i2z)n , for | − i2z| < 1
n=0
∞
X
=− (−i2)n+1 z n , for |z| < 1/2
n=0

1 1/z
=
z − i/2 1 − i/(2z)
∞  n
1X i
= , for |i/(2z)| < 1
z n=0 2z
∞  n
X i
= z −n−1 , for |z| < 2
n=0
2
−1  −n−1
X i
= z n , for |z| < 2
n=−∞
2
−1
X
= (−i2)n+1 z n , for |z| < 2
n=−∞

485
 1 1/2
− =
z−2 1 − z/2
∞
1 X  z n
= , for |z/2| < 1
2 n=0 2
∞
X zn
= , for |z| < 2
n=0
2n+1

1 1/z
− =−
z−2 1 − 2/z
∞  n
1X 2
=− , for |2/z| < 1
z n=0 z
∞
X
=− 2n z −n−1 , for |z| > 2
n=0
−1
X
=− 2−n−1 z n , for |z| > 2
n=−∞

486
 2 − i/2 1
2
= (2 − i/2) (1 − z/2)−2
(z − 2) 4
∞  
4 − i X −2  z n
= − , for |z/2| < 1
8 n=0 n 2
∞
4−iX
= (−1)n (n + 1)(−1)n 2−n z n , for |z| < 2
8 n=0
∞
4−iXn+1 n
= z , for |z| < 2
8 n=0 2n

 −2
2 − i/2 2 − i/2 2
2
= 1−
(z − 2) z2 z
∞    n
2 − i/2 X −2 2
= 2
− , for |2/z| < 1
z n=0
n z
∞
X
= (2 − i/2) (−1)n (n + 1)(−1)n 2n z −n−2 , for |z| > 2
n=0
−2
X
= (2 − i/2) (−n − 1)2−n−2 z n , for |z| > 2
n=−∞
−2
X n+1 n
= −(2 − i/2) z , for |z| > 2
n=−∞
2n+2

We take the appropriate combination of these series to find the Laurent series expansions in the regions:

487
|z| < 1/2, 1/2 < |z| < 2 and 2 < |z|. For |z| < 1/2, we have
∞ ∞ ∞
X X zn 4−iXn+1 n
f (z) = − (−i2)n+1 z n + n+1
+ n
z +d
n=0 n=0
2 8 n=0
2
∞  
X
n+1 1 4−in+1
f (z) = −(−i2) + n+1 + n
zn + d
n=0
2 8 2
∞   
X
n+1 1 4−i
f (z) = −(−i2) + n+1 1 + (n + 1) z n + d, for |z| < 1/2
n=0
2 4

For 1/2 < |z| < 2, we have

−1 ∞ ∞
X X zn 4−iXn+1 n
f (z) = (−i2)n+1 z n + n+1
+ n
z +d
n=−∞ n=0
2 8 n=0
2
−1 ∞   
X
n+1 n
X 1 4−i
f (z) = (−i2) z + n+1
1+ (n + 1) z n + d, for 1/2 < |z| < 2
n=−∞ n=0
2 4

For 2 < |z|, we have

−1 −1 −2
X
n+1 n
X
−n−1 n
X n+1 n
f (z) = (−i2) z − 2 z − (2 − i/2) z +d
n=−∞ n=−∞ n=−∞
2n+2
−2  
X 1
f (z) = (−i2)n+1 − (1 + (1 − i/4)(n + 1)) z n + d, for 2 < |z|
n=−∞
2n+1

488
Solution 14.18
The radius of convergence of the series for f (z) is
3 k  k3 
   
 k /3 
R = lim   = 3 lim   = 3.
n→∞ (k + 1)3 /3k+1  n→∞  (k + 1)3 

Thus f (z) is a function which is analytic inside the circle of radius 3.

1. The integrand is analytic. Thus by Cauchy’s theorem the value of the integral is zero.
I
eiz f (z) dz = 0
|z|=1

2. We use Cauchy’s integral formula to evaluate the integral.

i2π 3! 33
I
f (z) i2π (3)
4
dz = f (0) = = i2π
|z|=1 z 3! 3! 33
I
f (z)
4
dz = i2π.
|z|=1 z

3. We use Cauchy’s integral formula to evaluate the integral.

f (z) ez 1! 13
I
i2π d z 

dz = (f (z) e ) z=0
= i2π
|z|=1 z2 1! dz 31
f (z) ez
I
i2π
2
dz =
|z|=1 z 3

489
Chapter 15

The Residue Theorem

Man will occasionally stumble over the truth, but most of the time he will pick himself up and continue on.

- Winston Churchill

15.1 The Residue Theorem

We will find that many integrals on closed contours may be evaluated in terms of the residues of a function. We
first define residues and then prove the Residue Theorem.

490
 Result 15.1.1 Residues. Let f (z) be single-valued an analytic in a deleted neighbor-
hood of z0 . Then f (z) has the Laurent series expansion
∞
X
f (z) = an (z − z0 )n ,
n=−∞

1
The residue of f (z) at z = z0 is the coefficient of the z−z0 term:

Res (f (z), z0 ) = a−1 .

The residue at a branch point or non-isolated singularity is undefined as the Laurent

series does not exist. If f (z) has a pole of order n at z = z0 then we can use the Residue
Formula:
dn−1 
 
1
(z − z0 )n f (z) .

Res (f (z), z0 ) = lim n−1
z→z0 (n − 1)! dz

See Exercise 15.1 for a proof of the Residue Formula.

Example 15.1.1 In Example 10.4.5 we showed that f (z) = z/ sin z has first order poles at z = nπ, n ∈ Z \ {0}.

491
Now we find the residues at these isolated singularities.
 z   z 
Res , z = nπ = lim (z − nπ)
sin z z→nπ sin z
z − nπ
= nπ lim
z→nπ sin z
1
= nπ lim
z→nπ cos z
1
= nπ
(−1)n
= (−1)n nπ

Residue Theorem. We can evaluate many integrals in terms of the residues of a function. Suppose f (z) has
only one singularity, (at z = z0 ), inside the simple, closed, positively oriented contour C. f (z) has a convergent
Laurent
R series in some deleted disk about z0 . We deform C to lie in the disk. See Figure 15.1. We now evaluate
C
f (z) dz by deforming the contour and using the Laurent series expansion of the function.

C B

Figure 15.1: Deform the contour to lie in the deleted disk.

492
 Z Z
f (z) dz = f (z) dz
C B
Z ∞
X
= an (z − z0 )n dz
B n=−∞
∞ r ei(θ+2π)
(z − z0 )n+1

X i(θ+2π)
= an + a−1 [log(z − z0 )]rr eeiθ
n=−∞
n+1 r eiθ
n6=−1

= a−1 i2π

Z
f (z) dz = i2π Res (f (z), z0 )
C

Now assume that f (z) has n singularities at {z1 , . . . , zn }. We deform C to n contours C1 , . . . , Cn which enclose
the singularities and lie inRdeleted disks about the singularities in which f (z) has convergent Laurent series. See
Figure 15.2. We evaluate C f (z) dz by deforming the contour.
Z n Z
X n
X
f (z) dz = f (z) dz = i2π Res (f (z), zk )
C k=1 Ck k=1

Now instead let f (z) be analytic outside and on C except for isolated singularities at {ζn } in the domain
R outside
C and perhaps an isolated singularity at infinity. Let a be any point in the interior of C. To evaluate C f (z) dz
we make the change of variables ζ = 1/(z − a). This maps the contour C to C 0 . (Note that C 0 is negatively
oriented.) All the points outside C are mapped to points inside C 0 and vice versa. We can then evaluate the
integral in terms of the singularities inside C 0 .

493
 C2

C1 C3
C

Figure 15.2: Deform the contour n contours which enclose the n singularities.

 
−1
I I
1
f (z) dz = f +a dζ
C C0 ζ ζ2
I  
1 1
= 2
f + a dz
−C 0 z z
       
X 1 1 1 1 1
= i2π Res 2
f +a , + i2π Res 2
f + a ,0 .
n
z z ζ n − a z z

494
 C

a C’

Figure 15.3: The change of variables ζ = 1/(z − a).

Result 15.1.2 Residue Theorem. If f (z) is analytic in a compact, closed, connected

domain D except for isolated singularities at {zn } in the interior of D then
Z XI X
f (z) dz = f (z) dz = i2π Res (f (z), zn ).
∂D k Ck n

Here the set of contours {Ck } make up the positively oriented boundary ∂D of the domain
D. If the boundary of the domain is a single contour C then the formula simplifies.
I X
f (z) dz = i2π Res (f (z), zn )
C n

If instead f (z) is analytic outside and on C except for isolated singularities at {ζn } in
the domain outside C and perhaps an isolated singularity at infinity then
I        
X 1 1 1 1 1
f (z) dz = i2π Res 2
f +a , + i2π Res 2
f + a ,0 .
C n
z z ζ n − a z z

Here a is a any point in the interior of C. 495

Example 15.1.2 Consider
Z
1 sin z
dz
2πi C z(z − 1)

where C is the positively oriented circle of radius 2 centered at the origin. Since the integrand is single-valued
with only isolated singularities, the Residue Theorem applies. The value of the integral is the sum of the residues
from singularities inside the contour.
The only places that the integrand could have singularities are z = 0 and z = 1. Since
sin z cos z
lim = lim = 1,
z→0 z z→0 1
there is a removable singularity at the point z = 0. There is no residue at this point.
Now we consider the point z = 1. Since sin(z)/z is analytic and nonzero at z = 1, that point is a first order
pole of the integrand. The residue there is
 
sin z sin z
Res , z = 1 = lim(z − 1) = sin(1).
z(z − 1) z→1 z(z − 1)

There is only one singular point with a residue inside the path of integration. The residue at this point is
sin(1). Thus the value of the integral is
Z
1 sin z
dz = sin(1)
2πi C z(z − 1)

Example 15.1.3 Evaluate the integral

Z
cot z coth z
dz
C z3

where C is the unit circle about the origin in the positive direction.

496
 The integrand is
cot z coth z cos z cosh z
3
= 3
z z sin z sinh z
sin z has zeros at nπ. sinh z has zeros at inπ. Thus the only pole inside the contour of integration is at z = 0.
Since sin z and sinh z both have simple zeros at z = 0,
sin z = z + O(z 3 ), sinh z = z + O(z 3 )
the integrand has a pole of order 5 at the origin. The residue at z = 0 is
1 d4 1 d4
 
5 cot z coth z
z 2 cot z coth z


lim 4
z 3
= lim 4
z→0 4! dz z z→0 4! dz

1
= lim 24 cot(z) coth(z) csc(z)2 − 32 z coth(z) csc(z)4
4! z→0
− 16 z cos(2 z) coth(z) csc(z)4 + 22 z 2 cot(z) coth(z) csc(z)4
+ 2 z 2 cos(3 z) coth(z) csc(z)5 + 24 cot(z) coth(z) csch (z)2
+ 24 csc(z)2 csch (z)2 − 48 z cot(z) csc(z)2 csch (z)2
− 48 z coth(z) csc(z)2 csch (z)2 + 24 z 2 cot(z) coth(z) csc(z)2 csch (z)2
+ 16 z 2 csc(z)4 csch (z)2 + 8 z 2 cos(2 z) csc(z)4 csch (z)2
− 32 z cot(z) csch (z)4 − 16 z cosh(2 z) cot(z) csch (z)4
+ 22 z 2 cot(z) coth(z) csch (z)4 + 16 z 2 csc(z)2 csch (z)4

2 2 4 2 5
+ 8 z cosh(2 z) csc(z) csch (z) + 2 z cosh(3 z) cot(z) csch (z)
 
1 56
= −
4! 15
7
=−
45

497
 Since taking the fourth derivative of z 2 cot z coth z really sucks, we would like a more elegant way of finding
the residue. We expand the functions in the integrand in Taylor series about the origin.
  
z2 z4 z2 z4
cos z cosh z 1 − 2
+ 24
− · · · 1 + 2
+ 24
+ · · ·
3
= z 3 z 5
z 3 z5


z sin z sinh z 3
z z − 6 + 120 − · · · z + 6 + 120 + ···
z4
1− 6
+ ···
= −1 1



z3 z2 + z6 36
+ 90
+ ···
4
1 1 − z6 + · · ·
= z4
z 5 1 − 90 + ···
z4 z4
  
1
= 1− + ··· 1+ + ···
z5 6 90
 
1 7
= 5
1 − z4 + · · ·
z 45
1 7 1
= − + ···
z 5 45 z
7
Thus we see that the residue is − 45 . Now we can evaluate the integral.
Z
cot z coth z 14
3
dz = −i π
C z 45

15.2 Cauchy Principal Value for Real Integrals

15.2.1 The Cauchy Principal Value
First we recap improper integrals. If f (x) has a singularity at x0 ∈ (a . . . b) then
Z b Z x0 − Z b
f (x) dx ≡ lim+ f (x) dx + lim+ f (x) dx.
a →0 a δ→0 x0 +δ

498
For integrals on (−∞ . . . ∞),
Z ∞ Z b
f (x) dx ≡ lim f (x) dx.
−∞ a→−∞, b→∞ a

R1 1
Example 15.2.1 −1 x
dx is divergent. We show this with the definition of improper integrals.
Z 1 Z − Z 1
1 1 1
dx = lim+ dx + lim+ dx
−1 x →0 −1 x δ→0 δ x
= lim+ [ln |x|]−
−1 + lim+ [ln |x|]1δ
→0 δ→0
= lim+ ln  − lim+ ln δ
→0 δ→0

The integral diverges because  and δ approach zero independently.

Since 1/x is an odd function, it appears that the area under the curve is zero. Consider what would happen if
 and δ were not independent. If they approached zero symmetrically, δ = , then the value of the integral would
be zero.
Z − Z 1 
1
lim+ + dx = lim+ (ln  − ln ) = 0
→0 −1  x →0

We could make the integral have any value we pleased by choosing δ = c. 1
Z − Z 1 
1
lim+ + dx = lim+ (ln  − ln(c)) = − ln c
→0 −1 c x →0

We have seen it is reasonable that

Z 1
1
dx
−1 x
1
This may remind you of conditionally convergent series. You can rearrange the terms to make the series sum to any number.

499
has some meaning, and if we could evaluate the integral, the most reasonable value would be zero. The Cauchy
principal value provides us with a way of evaluating such integrals. If f (x) is continuous on (a, b) except at the
point x0 ∈ (a, b) then the Cauchy principal value of the integral is defined
Z b Z x0 − Z b 
− f (x) dx = lim+ f (x) dx + f (x) dx .
a →0 a x0 +

The Cauchy principal value is obtained by approaching the singularity symmetrically. The principal value of
the integral may exist when the integral diverges. If the integral exists, it is equal to the principal value of the
integral. R1
The Cauchy principal value of −1 x1 dx is defined
Z 1 Z − Z 1 
1 1 1
− dx ≡ lim+ dx + dx
−1 x →0 −1 x  x
−
[log |x|]1


= lim+ [log |x|]−1
→0
= lim+ (log | − | − log ||)
→0
= 0.
R
(Another notation for the principal value of an integral is PV f (x) dx.) Since the limits of integration approach
zero symmetrically, the two halves of the integral cancel. If the limits of integration approached zero independently,
(the definition of the integral), then the two halves would both diverge.
R∞
Example 15.2.2 −∞ x2x+1 dx is divergent. We show this with the definition of improper integrals.
Z ∞ Z b
x x
2
dx = lim 2
dx
−∞ x + 1 a→−∞, b→∞ a x + 1
 b
1 2
= lim ln(x + 1)
a→−∞, b→∞ 2
 2 a
1 b +1
= lim ln
2 a→−∞, b→∞ a2 + 1

500
The integral diverges because a and b approach infinity independently. Now consider what would happen if a and
b were not independent. If they approached zero symmetrically, a = −b, then the value of the integral would be
zero.

b2 + 1
 
1
lim ln =0
2 b→∞ b2 + 1

We could make the integral have any value we pleased by choosing a = −cb.
R∞
We can assign a meaning to divergent integrals of the form −∞
f (x) dx with the Cauchy principal value. The
Cauchy principal value of the integral is defined

Z ∞ Z a
− f (x) dx = lim f (x) dx.
−∞ a→∞ −a

The Cauchy principal value is obtained by approaching infinity symmetrically.

R∞
The Cauchy principal value of −∞ x2x+1 dx is defined

Z ∞ Z a
x x
− 2
dx = lim dx
−∞ x +1 a→∞ −a +1 x2

a
 
1 2
= lim ln x + 1
a→∞ 2
−a
= 0.

501
 Result 15.2.1 Cauchy Principal Value. If f (x) is continuous on (a, b) except at the
point x0 ∈ (a, b) then the integral of f (x) is defined
Z b Z x0 − Z b
f (x) dx = lim+ f (x) dx + lim+ f (x) dx.
a →0 a δ→0 x0 +δ

The Cauchy principal value of the integral is defined

Z b Z x0 − Z b 
− f (x) dx = lim+ f (x) dx + f (x) dx .
a →0 a x0 +

If f (x) is continuous on (−∞, ∞) then the integral of f (x) is defined

Z ∞ Z b
f (x) dx = lim f (x) dx.
−∞ a→−∞, b→∞ a

The Cauchy principal value of the integral is defined

Z ∞ Z a
− f (x) dx = lim f (x) dx.
−∞ a→∞ −a

The principal value of the integral may exist when the integral diverges. If the integral
exists, it is equal to the principal value of the integral.
R∞
Example 15.2.3 Clearly −∞
x dx diverges, however the Cauchy principal value exists.
Z ∞  2
x
− x dx = lim a=0
−∞ a→∞ 2 −a

502
In general, if f (x) is an odd function with no singularities on the finite real axis then
Z ∞
− f (x) dx = 0.
−∞

15.3 Cauchy Principal Value for Contour Integrals

Example 15.3.1 Consider the integral
Z
1
dz,
Cr z−1

where Cr is the positively oriented circle of radius r and center at the origin. From the residue theorem, we know
that the integral is

Z (
1 0 for r < 1,
dz =
Cr z−1 2πi for r > 1.

When r = 1, the integral diverges, as there is a first order pole on the path of integration. However, the principal
value of the integral exists.

Z Z 2π−
1 1
− dz = lim+ ieiθ dθ
Cr z−1 →0  −1 eiθ
 iθ
2π−
= lim+ log(e − 1) 
→0

503
We choose the branch of the logarithm with a branch cut on the positive real axis and arg log z ∈ (0, 2π).

= lim+ log ei(2π−) − 1 − log ei − 1

→0
= lim+ log 1 − i + O(2 ) − 1 − log 1 + i + O(2 ) − 1






→0
= lim+ log −i + O(2 ) − log i + O(2 )




→0
= lim+ Log  + O(2 ) + i arg −i + O(2 ) − Log  + O(2 ) − i arg i + O(2 )






→0
3π π
=i −i
2 2
= iπ
Thus we obtain

Z
1 0
 for r < 1,
− dz = πi for r = 1,
Cr z−1 
2πi for r > 1.


In the above example we evaluated the contour integral by parameterizing the contour. This approach is
only feasible when the integrand is simple. We would like to use the residue theorem to more easily evaluate the
principal value of the integral. But before we do that, we will need a preliminary result.

Result 15.3.1 Let f (z) have a first order pole at z = z0 and let (z − z0 )f (z) be analytic
in some neighborhood of z0 . Let the contour C be a circular arc from z0 +eiα to z0 +eiβ .
(We assume that β > α and β − α < 2π.)
Z
lim+ f (z) dz = i(β − α) Res (f (z), z0 )
→0 C

The contour is shown in Figure 15.4. (See Exercise 15.6 for a proof of this result.)

504
 Cε
β−α
z0 ε

Figure 15.4: The C Contour

Example 15.3.2 Consider

Z
1
− dz
C z−1

where C is the unit circle. Let Cp be the circular arc of radius 1 that starts and ends a distance of  from z = 1.
Let C be the positive, circular arc of radius  with center at z = 1 that joins the endpoints of Cp . Let Ci , be the
union of Cp and C . (Cp stands for Principal value Contour; Ci stands for Indented Contour.) Ci is an indented
contour that avoids the first order pole at z = 1. Figure 15.5 shows the three contours.

Cp

Cε

Figure 15.5: The Indented Contour.

505
 Note that the principal value of the integral is

Z Z
1 1
− dz = lim+ dz.
C z−1 →0 Cp z−1

We can calculate the integral along Ci with the residue theorem.

Z
1
dz = 2πi
Ci z−1

We can calculate the integral along C using Result 15.3.1. Note that as  → 0+ , the contour becomes a semi-circle,
a circular arc of π radians.
Z  
1 1
lim dz = iπ Res ,1 = iπ
→0+ C z−1 z−1

Now we can write the principal value of the integral along C in terms of the two known integrals.

Z Z Z
1 1 1
− dz = dz − dz
C z−1 Ci z − 1 C z − 1
= i2π − iπ
= iπ

In the previous example, we formed an indented contour that included the first order pole. You can show that
if we had indented the contour to exclude the pole, we would obtain the same result. (See Exercise 15.8.)
We can extend the residue theorem to principal values of integrals. (See Exercise 15.7.)

506
 Result 15.3.2 Residue Theorem for Principal Values. Let f (z) be analytic inside
and on a simple, closed, positive contour C, except for isolated singularities at z1 , . . . , zm
inside the contour and first order poles at ζ1 , . . . , ζn on the contour. Further, let the
contour be C 1 at the locations of these first order poles. (i.e., the contour does not have
a corner at any of the first order poles.) Then the principal value of the integral of f (z)
along C is
Z Xm Xn
− f (z) dz = i2π Res (f (z), zj ) + iπ Res (f (z), ζj ).
C j=1 j=1

15.4 Integrals on the Real Axis

Example 15.4.1 We wish to evaluate the integral
Z ∞
1
dx.
−∞ x2 +1

We can evaluate this integral directly using calculus.

Z ∞
1
dx = [arctan x]∞
−∞
−∞ x2 +1
=π

Now we will evaluate the integral using contour integration. Let CR be the semicircular arc from R to −R in the
upper half plane. Let C be the union of CR and the interval [−R, R].
We can evaluate the integral along C with the residue theorem. The integrand has first order poles at z = ±i.

507
For R > 1, we have
Z  
1 1
2
dz = i2π Res ,i
C z +1 z2 + 1
1
= i2π
2i
= π.

Now we examine the integral along CR . We use the maximum modulus integral bound to show that the value of
the integral vanishes as R → ∞.
Z   
 1   1 
 dz ≤ πR max  2
  
 z2 + 1 
CR z∈CR z + 1 

1
= πR
R2
−1
→ 0 as R → ∞.

Now we are prepared to evaluate the original real integral.

Z
1
2
dz = π
C z +1
Z R Z
1 1
2
dx + 2
dz = π
−R x + 1 CR z + 1

We take the limit as R → ∞.

Z ∞
1
dx = π
−∞ x2 +1

We would get the same result by closing the path of integration in the lower half plane. Note that in this case
the closed contour would be in the negative direction.

508
 If you are really observant, you may have noticed that we did something a little funny in evaluating
Z ∞
1
2
dx.
−∞ x + 1

The definition of this improper integral is

Z ∞ Z 0 Z b
1 1 1
2
dx = lim 2
dx+ = lim 2
dx.
−∞ x + 1 a→+∞ −a x + 1 b→+∞ 0 x + 1

In the above example we instead computed

Z R
1
lim dx.
R→+∞ −R x2 +1
x
Note that for some integrands, the former and latter are not the same. Consider the integral of x2 +1
.
Z ∞ Z 0 Z b
x x x
2
dx = lim 2
dx + lim 2
dx
−∞ x + 1 a→+∞ −a x + 1 b→+∞ 0 x + 1
   
1 2 1 2
= lim log |a + 1| + lim − log |b + 1|
a→+∞ 2 b→+∞ 2
Note that the limits do not exist and hence the integral diverges. We get a different result if the limits of
integration approach infinity symmetrically.
Z R  
x 1 2 2
lim dx = lim (log |R + 1| − log |R + 1|)
R→+∞ −R x2 + 1 R→+∞ 2
=0
(Note that the integrand is an odd function, so the integral from −R to R is zero.) We call this the principal
value of the integral and denote it by writing “PV” in front of the integral sign or putting a dash through the
integral.
Z ∞ Z ∞ Z R
PV f (x) dx ≡ − f (x) dx ≡ lim f (x) dx
−∞ −∞ R→+∞ −R

509
 The principal value of an integral may exist when the integral diverges. If the integral does converge, then it
is equal to its principal value.

We can use the method of Example 15.4.1 to evaluate the principal value of integrals of functions that vanish
fast enough at infinity.

510
Result 15.4.1 Let f (z) be analytic except for isolated singularities, with only first order
poles on the real axis. Let CR be the semi-circle from R to −R in the upper half plane.
If
 
lim R max |f (z)| = 0
R→∞ z∈CR

then
Z ∞ m
X n
X
− f (x) dx = i2π Res (f (z), zk ) + iπ Res (f (z), xk )
−∞ k=1 k=1

where z1 , . . . zm are the singularities of f (z) in the upper half plane and x1 , . . . , xn are
the first order poles on the real axis.
Now let CR be the semi-circle from R to −R in the lower half plane. If
 
lim R max |f (z)| = 0
R→∞ z∈CR

then
Z ∞ m
X n
X
− f (x) dx = −i2π Res (f (z), zk ) − iπ Res (f (z), xk )
−∞ k=1 k=1

where z1 , . . . zm are the singularities of f (z) in the lower half plane and x1 , . . . , xn are
the first order poles on the real axis.

511
 This result is proved in Exercise 15.9. Of course we can use this result to evaluate the integrals of the form

Z ∞
f (z) dz,
0

where f (x) is an even function.

15.5 Fourier Integrals

In order to do Fourier transforms, which are useful in solving differential equations, it is necessary to be able to
calculate Fourier integrals. Fourier integrals have the form

Z ∞
eiωx f (x) dx.
−∞

We evaluate these integrals by closing the path of integration in the lower or upper half plane and using techniques
of contour integration.
Consider the integral

Z π/2
e−R sin θ dθ.
0

Since 2θ/π ≤ sin θ for 0 ≤ θ ≤ π/2,

e−R sin θ ≤ e−R2θ/π for 0 ≤ θ ≤ π/2

512
 Z π/2 Z π/2
−R sin θ
e dθ ≤ e−R2θ/π dθ
0 0
h π iπ/2
−R2θ/π
= − e
2R 0
π −R
= − ( e − 1)
2R
π
≤
2R
→ 0 as R → ∞

We can use this to prove the following Result 15.5.1. (See Exercise 15.13.)

Result 15.5.1 Jordan’s Lemma.

Z π
π
e−R sin θ dθ < .
0 R
Suppose that f (z) vanishes as |z| → ∞. If ω is a (positive/negative) real number and
CR is a semi-circle of radius R in the (upper/lower) half plane then the integral
Z
f (z) eiωz dz
CR

vanishes as R → ∞.

R∞
We can use Jordan’s Lemma and the Residue Theorem to evaluate many Fourier integrals. Consider −∞ f (x) eiωx dx
where ω is a positive real number. Let f (z) be analytic except for isolated singularities, with only first order poles
on the real axis. Let C be the contour from −R to R on the real axis and then back to −R along a semi-circle in
the upper half plane. If R is large enough so that C encloses all the singularities of f (z) in the upper half plane

513
then
Z m
X n
X
iωz iωz
f (z) e dz = i2π Res (f (z) e , zk ) + iπ Res (f (z) eiωz , xk )
C k=1 k=1

where z1 , . . . zm are the singularities of f (z) in the upper half plane and x1 , . . . , xn are the first order poles on the
real axis. If f (z) vanishes as |z| → ∞ then the integral on CR vanishes as R → ∞ by Jordan’s Lemma.
Z ∞ Xm n
X
f (x) eiωx dx = i2π Res (f (z) eiωz , zk ) + iπ Res (f (z) eiωz , xk )
−∞ k=1 k=1

For negative ω we close the path of integration in the lower half plane. Note that the contour is then in the
negative direction.

Result 15.5.2 Fourier Integrals. Let f (z) be analytic except for isolated singularities,
with only first order poles on the real axis. Suppose that f (z) vanishes as |z| → ∞. If ω
is a positive real number then
Z ∞ Xm n
X
iωx iωz
f (x) e dx = i2π Res (f (z) e , zk ) + iπ Res (f (z) eiωz , xk )
−∞ k=1 k=1

where z1 , . . . zm are the singularities of f (z) in the upper half plane and x1 , . . . , xn are
the first order poles on the real axis. If ω is a negative real number then
Z ∞ X m n
X
iωx iωz
f (x) e dx = −i2π Res (f (z) e , zk ) − iπ Res (f (z) eiωz , xk )
−∞ k=1 k=1

where z1 , . . . zm are the singularities of f (z) in the lower half plane and x1 , . . . , xn are
the first order poles on the real axis.

514
15.6 Fourier Cosine and Sine Integrals
Fourier cosine and sine integrals have the form,
Z ∞ Z ∞
f (x) cos(ωx) dx and f (x) sin(ωx) dx.
0 0

If f (x) is even/odd then we can evaluate the cosine/sine integral with the method we developed for Fourier
integrals.

Let f (z) be analytic except for isolated singularities, with only first order poles on the real axis. Suppose that
f (x) is an even function and that f (z) vanishes as |z| → ∞. We consider real ω > 0.
Z ∞
1 ∞
Z
− f (x) cos(ωx) dx = − f (x) cos(ωx) dx
0 2 −∞

Since f (x) sin(ωx) is an odd function,

1 ∞
Z
− f (x) sin(ωx) dx = 0.
2 −∞
Thus
Z ∞
1 ∞
Z
− f (x) cos(ωx) dx = − f (x) eiωx dx
0 2 −∞
Now we apply Result 15.5.2.
Z ∞ m n
X iπ X
− f (x) cos(ωx) dx = iπ Res (f (z) eiωz , zk ) + Res (f (z) eiωz , xk )
0 k=1
2 k=1

where z1 , . . . zm are the singularities of f (z) in the upper half plane and x1 , . . . , xn are the first order poles on the
real axis.

515
 If f (x) is an odd function, we note that f (x) cos(ωx) is an odd function to obtain the analogous result for
Fourier sine integrals.

Result 15.6.1 Fourier Cosine and Sine Integrals. Let f (z) be analytic except for
isolated singularities, with only first order poles on the real axis. Suppose that f (x) is
an even function and that f (z) vanishes as |z| → ∞. We consider real ω > 0.
Z ∞ m n
X
iωz iπ X
− f (x) cos(ωx) dx = iπ Res (f (z) e , zk ) + Res (f (z) eiωz , xk )
0 2
k=1 k=1

where z1 , . . . zm are the singularities of f (z) in the upper half plane and x1 , . . . , xn are
the first order poles on the real axis. If f (x) is an odd function then,
Z ∞ µ n
X
iωz πX
− f (x) sin(ωx) dx = π Res (f (z) e , ζk ) + Res (f (z) eiωz , xk )
0 2
k=1 k=1

where ζ1 , . . . ζµ are the singularities of f (z) in the lower half plane and x1 , . . . , xn are the
first order poles on the real axis.

Now suppose that f (x) is neither even nor odd. We can evaluate integrals of the form:

Z ∞ Z ∞
f (x) cos(ωx) dx and f (x) sin(ωx) dx
−∞ −∞

516
by writing them in terms of Fourier integrals
Z ∞
1 ∞ 1 ∞
Z Z
f (x) cos(ωx) dx = iωx
f (x) e dx + f (x) e−iωx dx
2 2
Z −∞
∞ Z−∞ Z−∞∞
i ∞ i
f (x) sin(ωx) dx = − f (x) eiωx dx + f (x) e−iωx dx
−∞ 2 −∞ 2 −∞

15.7 Contour Integration and Branch Cuts

Example 15.7.1 Consider
∞
x−a
Z
dx, 0 < a < 1,
0 x+1
where x−a denotes exp(−a ln(x)). We choose the branch of the function
z −a
f (z) = |z| > 0, 0 < arg z < 2π
z+1
with a branch cut on the positive real axis.
Let C and CR denote the circular arcs of radius  and R where  < 1 < R. C is negatively oriented; CR
is positively oriented. Consider the closed contour C that is traced by a point moving from C to CR above the
branch cut, next around CR , then below the cut to C , and finally around C . (See Figure 15.10.)
We write f (z) in polar coordinates.
exp(−a log z) exp(−a(log r + iθ))
f (z) = =
z+1 r eiθ + 1
We evaluate the function above, (z = r ei0 ), and below, (z = r ei2π ), the branch cut.
exp[−a(log r + i0)] r−a
f (r ei0 ) = =
r+1 r+1
−a −i2aπ
exp[−a(log r + i2π)] r e
f (r ei2π ) = = .
r+1 r+1

517
 CR

Cε

Figure 15.6:

We use the residue theorem to evaluate the integral along C.

I
f (z) dz = i2π Res (f (z), −1)
C
Z R −a Z Z R −a −i2aπ Z
r r e
dr + f (z) dz − dr + f (z) dz = i2π Res (f (z), −1)
 r+1 CR  r+1 C

The residue is
Res (f (z), −1) = exp(−a log(−1)) = exp(−a(log 1 + iπ)) = e−iaπ .
We bound the integrals along C and CR with the maximum modulus integral bound.
−a
1−a
Z 
 ≤ 2π 
 
 f (z) dz = 2π
  1− 1−
Z C −a
R1−a

  R
 f (z) dz  ≤ 2πR
 = 2π

CR R−1 R−1
Since 0 < a < 1, the values of the integrals tend to zero as  → 0 and R → ∞. Thus we have
Z ∞ −a
r e−iaπ
dr = i2π
0 r+1 1 − e−i2aπ

518
 ∞
x−a
Z
π
dx =
0 x+1 sin aπ

519
Result 15.7.1 Integrals from Zero to Infinity. Let f (z) be a single-valued analytic
function with only isolated singularities and no singularities on the positive, real axis,
[0, ∞). Let a 6∈ Z. If the integrals exist then,
Z ∞ Xn
f (x) dx = − Res (f (z) log z, zk ) ,
0 k=1

Z ∞ n
a i2π X
x f (x) dx = i2πa
Res (z a f (z), zk ) ,
0 1− e
k=1

Z ∞ n n
1X X
Res f (z) log2 z, zk + iπ


f (x) log x dx = − Res (f (z) log z, zk ) ,
0 2
k=1 k=1

Z ∞ n
a i2π X
x f (x) log x dx = i2πa
Res (z a f (z) log z, zk )
0 1− e
k=1
n
π2a X
+ 2 Res (z a f (z), zk ) ,
sin (πa) k=1

n
!
Z ∞ m
∂ i2π X
xa f (x) logm x dx = Res (z a f (z), zk ) ,
0 ∂am 1 − ei2πa
k=1

where z1 , . . . , zn are the singularities of f (z) and there is a branch cut on the positive
real axis with 0 < arg(z) < 2π. 520
15.8 Exploiting Symmetry

We have already used

R ∞symmetry of the integrand to evaluate certain integrals. For f (x) an even function we
were able to evaluate 0 f (x) dx by extending the range of integration from −∞ to ∞. For

Z ∞
xα f (x) dx
0

we put a branch cut on the positive real axis and noted that the value of the integrand below the branch cut
is a constant multiple of the value of the function above the branch cut. This enabled us to evaluate the real
integral with contour integration. In this section we will use other kinds of symmetry to evaluate integrals. We
will discover that periodicity of the integrand will produce this symmetry.

15.8.1 Wedge Contours

We note that z n = rn einθ is periodic in θ with period 2π/n. The real and imaginary parts of z n are odd periodic
in θ with period π/n. This observation suggests that certain integrals on the positive real axis may be evaluated
by closing the path of integration with a wedge contour.

Example 15.8.1 Consider

Z ∞
1
dx
0 1 + xn

521
where n ∈ N, n ≥ 2. We can evaluate this integral using Result 15.7.1.
Z ∞ n−1  
1 X log z iπ(1+2k)/n
dx = − Res ,e
0 1 + xn k=0
1 + zn
n−1
(z − eiπ(1+2k)/n ) log z
X  
=− lim
k=0
z→ eiπ(1+2k)/n 1 + zn
n−1
log z + (z − eiπ(1+2k)/n )/z
X  
=− lim
k=0
z→ eiπ(1+2k)/n nz n−1
n−1  
X iπ(1 + 2k)/n
=−
k=0
n eiπ(1+2k)(n−1)/n
n−1
iπ X
=− (1 + 2k) ei2πk/n
n2 eiπ(n−1)/n k=0
n−1
iπ/n X
i2π e
= k ei2πk/n
n2 k=1
iπ/n
i2π e n
= 2 i2π/n
n e −1
π
=
n sin(π/n)
This is a bit grungy. To find a spiffier way to evaluate the integral we note that if we write the integrand as a
function of r and θ, it is periodic in θ with period 2π/n.
1 1
n
=
1+z 1 + rn einθ
The integrand along the rays θ = 2π/n, 4π/n, 6π/n, . . . has the same value as the integrand on the real axis.
Consider the contour C that is the boundary of the wedge 0 < r < R, 0 < θ < 2π/n. There is one singularity

522
inside the contour. We evaluate the residue there.
z − eiπ/n
 
1 iπ/n
Res , e = lim
1 + zn z→ eiπ/n 1 + z
n

1
= lim n−1
z→ eiπ/n nz
eiπ/n
=−
n
We evaluate the integral along C with the residue theorem.

−i2π eiπ/n
Z
1
n
dz =
C 1+z n
Let CR be the circular arc. The integral along CR vanishes as R → ∞.
Z   
 1  2πR  1 

n
dz  ≤ max  

CR 1 + z n z∈CR  1 + z n 
2πR 1
≤
n Rn − 1
→ 0 as R → ∞

We parametrize the contour to evaluate the desired integral.

Z ∞ Z 0
1 1 i2π/n −i2π eiπ/n
dx + e dx =
0 1 + xn ∞ 1+x
n n
Z ∞ iπ/n
1 −i2π e
n
dx =
0 1+x n(1 − ei2π/n )
Z ∞
1 π
n
dx =
0 1+x n sin(π/n)

523
15.8.2 Box Contours
Recall that ez = ex+iy is periodic in y with period 2π. This implies that the hyperbolic trigonometric functions
cosh z, sinh z and tanh z are periodic in y with period 2π and odd periodic in y with period π. We can exploit
this property to evaluate certain integrals on the real axis by closing the path of integration with a box contour.

Example 15.8.2 Consider the integral

Z ∞    ∞
1 iπ x
dx = i log tanh +
−∞ cosh x 4 2 −∞
= i log(1) − i log(−1)
= π.

We will evaluate this integral using contour integration. Note that

ex+iπ + e−x−iπ
cosh(x + iπ) = = − cosh(x).
2

Consider the box contour C that is the boundary of the region −R < x < R, 0 < y < π. The only singularity
of the integrand inside the contour is a first order pole at z = iπ/2. We evaluate the integral along C with the
residue theorem.
I  
1 1 iπ
dz = i2π Res ,
C cosh z cosh z 2
z − iπ/2
= i2π lim
z→iπ/2 cosh z
1
= i2π lim
z→iπ/2 sinh z

= 2π

524
The integrals along the sides of the box vanish as R → ∞.
Z ±R+iπ   
 1   1 
 dz  ≤ π
 max  

±R cosh z z∈[±R...±R+iπ]  cosh z 
 
 2 
≤ π max  ±R+iy

∓R−iy

y∈[0...π] e +e 
2
=
eR− e−R
π
≤
sinh R
→ 0 as R → ∞

The value of the integrand on the top of the box is the negative of its value on the bottom. We take the limit as
R → ∞.
Z ∞ Z −∞
1 1
dx + dx = 2π
−∞ cosh x ∞ − cosh x
Z ∞
1
dx = π
−∞ cosh x

15.9 Definite Integrals Involving Sine and Cosine

Example 15.9.1 For real-valued a, evaluate the integral:
Z 2π
dθ
f (a) = .
0 1 + a sin θ
What is the value of the integral for complex-valued a.
Real-Valued a. For −1 < a < 1, the integrand is bounded, hence the integral exists. For |a| = 1, the
integrand has a second order pole on the path of integration. For |a| > 1 the integrand has two first order poles

525
on the path of integration. The integral is divergent for these two cases. Thus we see that the integral exists for
−1 < a < 1.
For a = 0, the value of the integral is 2π. Now consider a 6= 0. We make the change of variables z = eiθ . The
real integral from θ = 0 to θ = 2π becomes a contour integral along the unit circle, |z| = 1. We write the sine,
cosine and the differential in terms of z.
z − z −1 z + z −1 dz
sin θ = , cos θ = , dz = i eiθ dθ, dθ =
2i 2 iz
We write f (a) as an integral along C, the positively oriented unit circle |z| = 1.
I I
1/(iz) 2/a
f (a) = −1
dz = 2
dz
C 1 + a(z − z )/(2i) C z + (2i/a)z − 1

We factor the denominator of the integrand.

I
2/a
f (a) = dz
C (z − z1 )(z − z2 )
 √   √ 
−1 + 1 − a2 −1 − 1 − a2
z1 = i , z2 = i
a a
Because |a| < 1, the second root is outside the unit circle.
√
1 + 1 − a2
|z2 | = > 1.
|a|
Since |z1 z2 | = 1, |z1 | < 1. Thus the pole at z1 is inside the contour and the pole at z2 is outside. We evaluate the
contour integral with the residue theorem.
I
2/a
f (a) = 2
dz
C z + (2i/a)z − 1
2/a
= i2π
z1 − z2
1
= i2π √
i 1 − a2

526
 2π
f (a) = √
1 − a2
Complex-Valued a. We note that the integral converges except for real-valued a satisfying |a| ≥ 1. On
any closed subset of C \ {a ∈ R | |a| ≥ 1} the integral is uniformly convergent. Thus except for the values
{a ∈ R | |a| ≥ 1}, we can differentiate the integral with respect to a. f (a) is analytic in the complex plane except
for the set of points on the real axis: a ∈ (−∞ . . . − 1] and a ∈ [1 . . . ∞). The value of the analytic function f (a)
on the real axis for the interval (−1 . . . 1) is
2π
f (a) = √ .
1 − a2
By analytic continuation we see that the value of f (a) in the complex plane is the branch of the function
2π
f (a) =
(1 − a2 )1/2
where f (a) is positive, real-valued for a ∈ (−1 . . . 1) and there are branch cuts on the real axis on the intervals:
(−∞ . . . − 1] and [1 . . . ∞).

Result 15.9.1 For evaluating integrals of the form

Z a+2π
F (sin θ, cos θ) dθ
a

it may be useful to make the change of variables z = eiθ . This gives us a contour integral
along the unit circle about the origin. We can write the sine, cosine and differential in
terms of z.
z − z −1 z + z −1 dz
sin θ = , cos θ = , dθ =
2i 2 iz

527
15.10 Infinite Sums
The function g(z) = π cot(πz) has simple poles at z = n ∈ Z. The residues at these points are all unity.

π(z − n) cos(πz)
Res (π cot(πz), n) = lim
z→n sin(πz)
π cos(πz) − π(z − n) sin(πz)
= lim
z→n π cos(πz)
=1

Let Cn be the square contour with corners at z = (n + 1/2)(±1 ± i). Recall that

cos z = cos x cosh y − i sin x sinh y and sin z = sin x cosh y + i cos x sinh y.

First we bound the modulus of cot(z).

 
 cos x cosh y − i sin x sinh y 
| cot(z)| =  
sin x cosh y + i cos x sinh y 
s
cos2 x cosh2 y + sin2 x sinh2 y
=
sin2 x cosh2 y + cos2 x sinh2 y
s
cosh2 y
≤
sinh2 y
= | coth(y)|

The hyperbolic cotangent, coth(y), has a simple pole at y = 0 and tends to ±1 as y → ±∞.
Along the top and bottom of Cn , (z = x ± i(n + 1/2)), we bound the modulus of g(z) = π cot(πz).
 
|π cot(πz)| ≤ π  coth(π(n + 1/2))

528
Along the left and right sides of Cn , (z = ±(n + 1/2) + iy), the modulus of the function is bounded by a constant.
 
 cos(π(n + 1/2)) cosh(πy) ∓ i sin(π(n + 1/2)) sinh(πy) 
|g(±(n + 1/2) + iy)| = π 
sin(π(n + 1/2)) cosh(πy) + i cos(π(n + 1/2)) sinh(πy) 
= |∓iπ tanh(πy)|
≤π

Thus the modulus of π cot(πz) can be bounded by a constant M on Cn .

Let f (z) be analytic except for isolated singularities. Consider the integral,
I
π cot(πz)f (z) dz.
Cn

We use the maximum modulus integral bound.

I 
 
 π cot(πz)f (z) dz  ≤ (8n + 4)M max |f (z)|

Cn
 z∈Cn

Note that if

lim |zf (z)| = 0,

|z|→∞

then
I
lim π cot(πz)f (z) dz = 0.
n→∞ Cn

This implies that the sum of all residues of π cot(πz)f (z) is zero. Suppose further that f (z) is analytic at
z = n ∈ Z. The residues of π cot(πz)f (z) at z = n are f (n). This means
∞
X
f (n) = −( sum of the residues of π cot(πz)f (z) at the poles of f (z) ).
n=−∞

529
 Result 15.10.1 If

lim |zf (z)| = 0,

|z|→∞

then the sum of all the residues of π cot(πz)f (z) is zero. If in addition f (z) is analytic
at z = n ∈ Z then
∞
X
f (n) = −( sum of the residues of π cot(πz)f (z) at the poles of f (z) ).
n=−∞

Example 15.10.1 Consider the sum

∞
X 1
, a 6∈ Z.
n=−∞
(n + a)2

By Result 15.10.1 with f (z) = 1/(z + a)2 we have

∞  
X 1 1
2
= − Res π cot(πz) , −a
n=−∞
(n + a) (z + a)2
d
= −π lim cot(πz)
z→−a dz
−π sin2 (πz) − π cos2 (πz)
= −π .
sin2 (πz)

∞
X 1 π2
=
n=−∞
(n + a)2 sin2 (πa)

530
Example 15.10.2 Derive π/4 = 1 − 1/3 + 1/5 − 1/7 + 1/9 − · · · .
Consider the integral
Z
1 dw
In =
2πi Cn w(w − z) sin w

where Cn is the square with corners at w = (n + 1/2)(±1 ± i)π, n ∈ Z+ . With the substitution w = x + iy,

| sin w|2 = sin2 x + sinh2 y,

we see that |1/ sin w| ≤ 1 on Cn . Thus In → 0 as n → ∞. We use the residue theorem and take the limit n → ∞.
∞ 
(−1)n (−1)n

X 1 1
0= + + − 2
n=1
nπ(nπ − z) nπ(nπ + z) z sin z z

∞
1 1 X (−1)n
= − 2z
sin z z n=1
n2 π 2 − z 2
∞ 
1 X (−1)n (−1)n

= − −
z n=1 nπ − z nπ + z

We substitute z = π/2 into the above expression to obtain

π/4 = 1 − 1/3 + 1/5 − 1/7 + 1/9 − · · ·

531
15.11 Exercises
The Residue Theorem
Exercise 15.1
Let f (z) have a pole of order n at z = z0 . Prove the Residue Formula:

dn−1
 
1 n
Res (f (z), z0 ) = lim [(z − z0 ) f (z)] .
z→z0 (n − 1)! dz n−1
Hint, Solution

Exercise 15.2
Consider the function

z4
f (z) = .
z2 + 1
Classify the singularities of f (z) in the extended complex plane. Calculate the residue at each pole and at infinity.
Find the Laurent series expansions and their domains of convergence about the points z = 0, z = i and z = ∞.
Hint, Solution

Exercise 15.3
Let P (z) be a polynomial none of whose roots lie on the closed contour Γ. Show that
Z 0
1 P (z)
dz = number of roots of P (z) which lie inside Γ.
i2π P (z)
where the roots are counted according to their multiplicity.
Hint: From the fundamental theorem of algebra, it is always possible to factor P (z) in the form P (z) =
(z − z1 )(z − z2 ) · · · (z − zn ). Using this form of P (z) the integrand P 0 (z)/P (z) reduces to a very simple expression.
Hint, Solution

532
Exercise 15.4
Find the value of
ez
I
dz
C (z − π) tan z
where C is the positively-oriented circle
1. |z| = 2
2. |z| = 4
Hint, Solution

Cauchy Principal Value for Real Integrals

Solution 15.1
Show that the integral
Z 1
1
dx.
−1 x
is divergent. Evaluate the integral
Z 1
1
dx, α ∈ R, α 6= 0.
−1 x − iα
Evaluate
Z 1
1
lim+ dx
α→0 −1 x − iα
and
Z 1
1
lim− dx.
α→0 −1 x − iα

533
The integral exists for α arbitrarily close to zero, but diverges when α = 0. Plot the real and imaginary part of
the integrand. If one were to assign meaning to the integral for α = 0, what would the value of the integral be?

Exercise 15.5
Do the principal values of the following integrals exist?
R1 1
1. −1 x2
dx,
R1 1
2. −1 x3
dx,
R1f (x)
3. −1 x3
dx.

Assume that f (x) is real analytic on the interval (−1, 1).

Hint, Solution

Cauchy Principal Value for Contour Integrals

Exercise 15.6
Let f (z) have a first order pole at z = z0 and let (z − z0 )f (z) be analytic in some neighborhood of z0 . Let the
contour C be a circular arc from z0 + eiα to z0 + eiβ . (Assume that β > α and β − α < 2π.) Show that
Z
lim f (z) dz = i(β − α) Res (f (z), z0 )
→0+ C

Hint, Solution

Exercise 15.7
Let f (z) be analytic inside and on a simple, closed, positive contour C, except for isolated singularities at
z1 , . . . , zm inside the contour and first order poles at ζ1 , . . . , ζn on the contour. Further, let the contour be C 1 at

534
the locations of these first order poles. (i.e., the contour does not have a corner at any of the first order poles.)
Show that the principal value of the integral of f (z) along C is
Z Xm n
X
− f (z) dz = i2π Res (f (z), zj ) + iπ Res (f (z), ζj ).
C j=1 j=1

Hint, Solution

Exercise 15.8
Let C be the unit circle. Evaluate
Z
1
− dz
C z−1
by indenting the contour to exclude the first order pole at z = 1.
Hint, Solution

Integrals from −∞ to ∞
Exercise 15.9
Prove Result 15.4.1.
Hint, Solution

Exercise 15.10
Evaluate
Z ∞
2x
− .
−∞ x2 +x+1
Hint, Solution

535
Exercise 15.11
Use contour integration to evaluate the integrals
Z ∞
dx
1. 4
,
−∞ 1 + x
Z ∞
x2 dx
2. 2 2
,
−∞ (1 + x )
Z ∞
cos(x)
3. 2
dx.
−∞ 1 + x

Hint, Solution

Exercise 15.12
Evaluate by contour integration
∞
x6
Z
dx.
0 (x4 + 1)2
Hint, Solution

Fourier Integrals
Exercise 15.13
Suppose that f (z) vanishes as |z| → ∞. If ω is a (positive / negative) real number and CR is a semi-circle of
radius R in the (upper / lower) half plane then show that the integral
Z
f (z) eiωz dz
CR

vanishes as R → ∞.
Hint, Solution

536
Exercise 15.14
Evaluate by contour integration
Z ∞
cos 2x
dx.
−∞ x − iπ
Hint, Solution

Fourier Cosine and Sine Integrals

Exercise 15.15
Evaluate
Z ∞
sin x
dx.
−∞ x
Hint, Solution

Exercise 15.16
Evaluate
∞
1 − cos x
Z
dx.
−∞ x2
Hint, Solution

Exercise 15.17
Evaluate
Z ∞
sin(πx)
dx.
0 x(1 − x2 )
Hint, Solution

537
Contour Integration and Branch Cuts
Exercise 15.18
Show that
Z ∞
(ln x)2 π3
1. dx = ,
0 1 + x2 8
Z ∞
ln x
2. dx = 0.
0 1 + x2
Hint, Solution

Exercise 15.19
By methods of contour integration find
Z ∞
dx
0 x2 + 5x + 6
R
[ Recall the trick of considering Γ
f (z) log z dz with a suitably chosen contour Γ and branch for log z. ]
Hint, Solution

Exercise 15.20
Show that
∞
xa
Z
πa
2
dx = for − 1 < <(a) < 1.
0 (x + 1) sin(πa)
From this derive that
∞ ∞
log2 x π2
Z Z
log x
dx = 0, dx = .
0 (x + 1)2 0 (x + 1)2 3
Hint, Solution

538
Exercise 15.21
Consider the integral
∞
xa
Z
I(a) = dx.
0 1 + x2

1. For what values of a does the integral exist?

2. Evaluate the integral. Show that

π
I(a) =
2 cos(πa/2)

3. Deduce from your answer in part (b) the results

∞ ∞
log2 x π3
Z Z
log x
dx = 0, dx = .
0 1 + x2 0 1 + x2 8

You may assume that it is valid to differentiate under the integral sign.

Hint, Solution

Exercise 15.22
Let f (z) be a single-valued analytic function with only isolated singularities and no singularities on the positive
real axis, [0, ∞). Give sufficient conditions on f (x) for absolute convergence of the integral
Z ∞
xa f (x) dx.
0

Assume that a is not an integer. Evaluate the integral by considering the integral of z a f (z) on a suitable contour.
(Consider the branch of z a on which 1a = 1.)
Hint, Solution

539
Exercise 15.23
Using the solution to Exercise 15.22, evaluate
Z ∞
xa f (x) log x dx,
0

and
Z ∞
xa f (x) logm x dx,
0

where m is a positive integer.

Hint, Solution

Exercise 15.24
Using the solution to Exercise 15.22, evaluate
Z ∞
f (x) dx,
0

i.e. examine a = 0. The solution will suggest Ra way to evaluate the integral with contour integration. Do the
∞
contour integration to corroborate the value of 0 f (x) dx.
Hint, Solution

Exercise 15.25
Let f (z) be an analytic function with only isolated singularities and no singularities on the positive real axis,
[0, ∞). Give sufficient conditions on f (x) for absolute convergence of the integral
Z ∞
f (x) log x dx
0

Evaluate the integral with contour integration.

Hint, Solution

540
Exercise 15.26
For what values of a does the following integral exist?
Z ∞
xa
dx.
0 1 + x4

Evaluate the integral. (Consider the branch of xa on which 1a = 1.)

Hint, Solution

Exercise 15.27
By considering the integral of f (z) = z 1/2 log z/(z + 1)2 on a suitable contour, show that
∞ ∞
x1/2 log x x1/2
Z Z
π
dx = π, dx = .
0 (x + 1)2 0 (x + 1) 2 2

Hint, Solution

Exploiting Symmetry
Exercise 15.28
Evaluate by contour integration, the principal value integral
Z ∞
eax
I(a) = − x −x
dx
−∞ e − e

for a real and |a| < 1. [Hint: Consider the contour that is the boundary of the box, −R < x < R, 0 < y < π, but
indented around z = 0 and z = iπ.
Hint, Solution

Exercise 15.29
Evaluate the following integrals.

541
 Z ∞
dx
1. ,
0 (1 + x2 )2
Z ∞
dx
2. .
0 1 + x3
Hint, Solution

Exercise 15.30
Find the value of the integral I
Z ∞
dx
I=
0 1 + x6

by considering the contour integral

Z
dz
Γ 1 + z6

with an appropriately chosen contour Γ.

Hint, Solution

Exercise 15.31
2
Let C be the boundary of the sector 0 < r < R, 0 < θ < π/4. By integrating e−z on C and letting R → ∞ show
that
Z ∞ Z ∞ Z ∞
1 2
2
cos(x ) dx = 2
sin(x ) dx = √ e−x dx.
0 0 2 0

Hint, Solution

542
Exercise 15.32
Evaluate
Z ∞
x
dx
−∞ sinh x

using contour integration.

Hint, Solution

Exercise 15.33
Show that
∞
eax
Z
π
x
dx = for 0 < a < 1.
−∞ e +1 sin(πa)

Use this to derive that

Z ∞
cosh(bx) π
dx = for − 1 < b < 1.
−∞ cosh x cos(πb/2)

Hint, Solution

Exercise 15.34
Using techniques of contour integration find for real a and b:
Z π
dθ
F (a, b) =
0 (a + b cos θ)2

What are the restrictions on a and b if any? Can the result be applied for complex a, b? How?
Hint, Solution

543
Exercise 15.35
Show that
Z ∞
cos x π
dx =
−∞ ex + e−x eπ/2 + e−π/2

[ Hint: Begin by considering the integral of eiz /( ez + e−z ) around a rectangle with vertices: ±R, ±R + iπ.]
Hint, Solution

Definite Integrals Involving Sine and Cosine

Exercise 15.36
Use contour integration to evaluate the integrals
Z 2π
dθ
1. ,
0 2 + sin(θ)
Z π
cos(nθ)
2. dθ for |a| < 1, n ∈ Z0+ .
−π 1 − 2a cos(θ) + a2

Hint, Solution

Exercise 15.37
By integration around the unit circle, suitably indented, show that
Z π
cos(nθ) sin(nα)
− dθ = π .
0 cos θ − cos α sin α

Hint, Solution

544
Exercise 15.38
Evaluate
1
x2
Z
√ dx.
0 (1 + x2 ) 1 − x2

Hint, Solution

Infinite Sums
Exercise 15.39
Evaluate
∞
X 1
.
n=1
n4

Hint, Solution

Exercise 15.40
Sum the following series using contour integration:
∞
X 1
n=−∞
n2 − α 2

Hint, Solution

545
15.12 Hints
The Residue Theorem
Hint 15.1
Substitute the Laurent series into the formula and simplify.

Hint 15.2
Use that the sum of all residues of the function in the extended complex plane is zero in calculating the residue
at infinity. To obtain the Laurent series expansion about z = i, write the function as a proper rational function,
(numerator has a lower degree than the denominator) and expand in partial fractions.

Hint 15.3

Hint 15.4

Cauchy Principal Value for Real Integrals

Hint 15.5

Hint 15.6
For the third part, does the integrand have a term that behaves like 1/x2 ?

Cauchy Principal Value for Contour Integrals

546
Hint 15.7
Expand f (z) in a Laurent series. Only the first term will make a contribution to the integral in the limit as
 → 0+ .

Hint 15.8
Use the result of Exercise 15.6.

Hint 15.9
Look at Example 15.3.2.

Integrals from −∞ to ∞
Hint 15.10
Close the path of integration in the upper or lower half plane with a semi-circle. Use the maximum modulus
integral bound, (Result 12.1.1), to show that the integral along the semi-circle vanishes.

Hint 15.11
Make the change of variables x = 1/ξ.

Hint 15.12
Use Result 15.4.1.

Hint 15.13

Fourier Integrals

547
Hint 15.14
Use
Z π
π
e−R sin θ dθ < .
0 R

Hint 15.15

Fourier Cosine and Sine Integrals

Hint 15.16
eix
Consider the integral of ix
.

Hint 15.17
Show that
∞ Z ∞
1 − cos x 1 − eix
Z
dx = − dx.
−∞ x2 −∞ x2

Hint 15.18
Show that
∞
i ∞ eix
Z Z
sin(πx)
dx = − − dx.
0 x(1 − x2 ) 2 −∞ x(1 − x2 )

Contour Integration and Branch Cuts

Hint 15.19
Integrate a branch of (log z)2 /(1 + z 2 ) along the boundary of the domain  < r < R, 0 < θ < π.

548
Hint 15.20

Hint 15.21
Note that
Z 1
xa dx
0

converges for <(a) > −1; and

Z ∞
xa dx
1

converges for <(a) < 1.

Consider f (z) = z a /(z + 1)2 with a branch cut along the positive real axis and the contour in Figure 15.10 in
the limit as ρ → 0 and R → ∞.
To derive the last two integrals, differentiate with respect to a.

Hint 15.22

Hint 15.23
Consider the integral of z a f (z) on the contour in Figure 15.10.

Hint 15.24
Differentiate with respect to a.

549
Hint 15.25
Take the limit as a → 0. Use L’Hospital’s rule. To corroborate the result, consider the integral of f (z) log z on
an appropriate contour.

Hint 15.26
Consider the integral of f (z) log2 z on the contour in Figure 15.10.

Hint 15.27
Consider the integral of
za
f (z) =
1 + z4
on the boundary of the region  < r < R, 0 < θ < π/2. Take the limits as  → 0 and R → ∞.

Hint 15.28
Consider the branch of f (z) = z 1/2 log z/(z + 1)2 with a branch cut on the positive real axis and 0 < arg z < 2π.
Integrate this function on the contour in Figure 15.10.

Exploiting Symmetry
Hint 15.29

Hint 15.30
For the second part, consider the integral along the boundary of the region, 0 < r < R, 0 < θ < 2π/3.

Hint 15.31

550
Hint 15.32
To show that the integral on the quarter-circle vanishes as R → ∞ establish the inequality,

4 π
cos 2θ ≥ 1 − θ, 0≤θ≤ .
π 4

Hint 15.33
Consider the box contour C this is the boundary of the rectangle, −R ≤ x ≤ R, 0 ≤ y ≤ π. The value of the
integral is π 2 /2.

Hint 15.34
Consider the rectangular contour with corners at ±R and ±R + i2π. Let R → ∞.

Hint 15.35

Hint 15.36

Definite Integrals Involving Sine and Cosine

Hint 15.37

Hint 15.38

551
Hint 15.39
Make the changes of variables x = sin ξ and then z = eiξ .

Infinite Sums
Hint 15.40
Use Result 15.10.1.

Hint 15.41

552
15.13 Solutions

The Residue Theorem

Solution 15.2
Since f (z) has an isolated pole of order n at z = z0 , it has a Laurent series that is convergent in a deleted
neighborhood about that point. We substitute this Laurent series into the Residue Formula to verify it.

dn−1
 
1 n
Res (f (z), z0 ) = lim [(z − z0 ) f (z)]
z→z0 (n − 1)! dz n−1
" ∞
#!
1 dn−1 n
X
k
= lim (z − z0 ) ak (z − z0 )
z→z0 (n − 1)! dz n−1 k=−n
"∞ #!
1 dn−1 X
= lim ak−n (z − z0 )k
z→z0 (n − 1)! dz n−1 k=0
∞
!
1 X k!
= lim ak−n (z − z0 )k−n+1
z→z0 (n − 1)! k=n−1 (k − n + 1)!
∞
!
1 X (k + n − 1)!
= lim ak−1 (z − z0 )k
z→z0 (n − 1)! k=0 k!
1 (n − 1)!
= a−1
(n − 1)! 0!
= a−1

This proves the Residue Formula.

553
Solution 15.3
Classify Singularities.

z4 z4
f (z) = = .
z2 + 1 (z − i)(z + i)

There are first order poles at z = ±i. Since the function behaves like z 2 at infinity, there is a second order pole
there. To see this more slowly, we can make the substitution z = 1/ζ and examine the point ζ = 0.

ζ −4
 
1 1 1
f = −2 = 2 4
= 2
ζ ζ +1 ζ +ζ ζ (1 + ζ 2 )

f (1/ζ) has a second order pole at ζ = 0, which implies that f (z) has a second order pole at infinity.
Residues. The residues at z = ±i are,
 4
z4

z i
Res 2
, i = lim =− ,
z +1 z→i z + i 2

z4 z4
 
i
Res 2
, −i = lim = .
z +1 z→−i z−i 2

The residue at infinity is

   
−1 1
Res (f (z), ∞) = Res f ,ζ = 0
ζ2 ζ
−1 ζ −4
 
= Res ,ζ = 0
ζ 2 ζ −2 + 1
ζ −4
 
= Res − ,ζ = 0
1 + ζ2

554
Here we could use the residue formula, but it’s easier to find the Laurent expansion.
∞
!
X
= Res −ζ −4 (−1)n ζ 2n , ζ = 0
n=0
=0
We could also calculate the residue at infinity by recalling that the sum of all residues of this function in the
extended complex plane is zero.
−i i
+ + Res (f (z), ∞) = 0
2 2

Res (f (z), ∞) = 0
Laurent Series about z = 0. Since the nearest singularities are at z = ±i, the Taylor series will converge
in the disk |z| < 1.
z4 1
2
= z4
z +1 1 − (−z)2
X∞
4
=z (−z 2 )n
n=0
∞
X
= z4 (−1)n z 2n
n=0
∞
X
= (−1)n z 2n
n=2

This geometric series converges for | − z 2 | < 1, or |z| < 1. The series expansion of the function is
∞
z4 X
= (−1)n z 2n for |z| < 1
z 2 + 1 n=2

555
 Laurent Series about z = i. We expand f (z) in partial fractions. First we write the function as a proper
rational function, (i.e. the numerator has lower degree than the denominator). By polynomial division, we see
that
1
f (z) = z 2 − 1 + .
z2 +1
Now we expand the last term in partial fractions.
−i/2 i/2
f (z) = z 2 − 1 + +
z−i z+i
Since the nearest singularity is at z = −i, the Laurent series will converge in the annulus 0 < |z − i| < 2.

z 2 − 1 = ((z − i) + i)2 − 1
= (z − i)2 + i2(z − i) − 2

i/2 i/2
=
z+i i2 + (z − i)
1/4
=
1 − i(z − i)/2
∞  n
1 X i(z − i)
=
4 n=0 2
∞
1 X in
= (z − i)n
4 n=0 2n

This geometric series converges for |i(z − i)/2| < 1, or |z − i| < 2. The series expansion of f (z) is
∞
−i/2 1 X in
f (z) = − 2 + i2(z − i) + (z − i)2 + (z − i)n .
z−i 4 n=0 2n

556
 ∞
z4 −i/2 2 1 X in
= − 2 + i2(z − i) + (z − i) + (z − i)n for |z − i| < 2
z2 + 1 z−i 4 n=0 2n

Laurent Series about z = ∞. Since the nearest singularities are at z = ±i, the Laurent series will converge
in the annulus 1 < |z| < ∞.

z4 z2
=
z2 + 1 1 + 1/z 2
∞  n
2
X 1
=z − 2
n=0
z
0
X
= (−1)n z 2(n+1)
n=−∞
1
X
= (−1)n+1 z 2n
n=−∞

This geometric series converges for | − 1/z 2 | < 1, or |z| > 1. The series expansion of f (z) is

1
z4 X
= (−1)n+1 z 2n for 1 < |z| < ∞
z 2 + 1 n=−∞

Solution 15.4
Method 1: Residue Theorem. We factor P (z). Let m be the number of roots, counting multiplicities, that

557
lie inside the contour Γ. We find a simple expression for P 0 (z)/P (z).
n
Y
P (z) = c (z − zk )
k=1
n
XY n
P 0 (z) = c (z − zj )
k=1 j=1
j6=k

Pn Qn
0 c k=1 j=1 (z − zj )
P (z) j6=k
=
c nk=1 (z − zk )
Q
P (z)
Qn
Xn j=1 (z − zj )
j6=k
= Qn
k=1 j=1 (z − zj )
n
X 1
=
k=1
z − zk
Now we do the integration using the residue theorem.
n
P 0 (z)
Z Z X
1 1 1
dz = dz
i2π Γ P (z) i2π Γ k=1 z − zk
n Z
X 1 1
= dz
k=1
i2π Γ z − zk
Z
X 1 1
= dz
zk inside Γ
i2π Γ z − zk
X
= 1
zk inside Γ

=m

558
 Qn
Method 2: Fundamental Theorem of Calculus. We factor the polynomial, P (z) = c k=1 (z − zk ). Let
m be the number of roots, counting multiplicities, that lie inside the contour Γ.

P 0 (z)
Z
1 1
dz = [log P (z)]C
i2π Γ P (z) i2π
" n
#
1 Y
= log (z − zk )
i2π k=1
" n #C
1 X
= log(z − zk )
i2π k=1
C

The value of the logarithm changes by i2π for the terms in which zk is inside the contour. Its value does not
change for the terms in which zk is outside the contour.
" #
1 X
= log(z − zk )
i2π zk inside Γ C
1 X
= i2π
i2π zk inside Γ

=m

Solution 15.5
1.

ez ez cos z
I I
dz = dz
C (z − π) tan z C (z − π) sin z

The integrand has first order poles at z = nπ, n ∈ Z, n 6= 1 and a double pole at z = π. The only pole

559
 inside the contour occurs at z = 0. We evaluate the integral with the residue theorem.
ez cos z ez cos z
I  
dz = i2π Res ,z = 0
C (z − π) sin z (z − π) sin z
ez cos z
= i2π lim z
z=0 (z − π) sin z
z
= −i2 lim
z=0 sin z
1
= −i2 lim
z=0 cos z
= −i2

ez
I
dz = −i2
C (z − π) tan z

2. The integrand has a first order poles at z = 0, −π and a second order pole at z = π inside the contour. The
value of the integral is i2π times the sum of the residues at these points. From the previous part we know
that residue at z = 0.
ez cos z
 
1
Res ,z = 0 = −
(z − π) sin z π
We find the residue at z = −π with the residue formula.
ez cos z ez cos z
 
Res , z = −π = lim (z + π)
(z − π) sin z z→−π (z − π) sin z
−π
e (−1) z+π
= lim
−2π z→−π sin z
e−π 1
= lim
2π z→−π cos z
e−π
=−
2π

560
We find the residue at z = π by finding the first few terms in the Laurent series of the integrand.
ez cos z ( eπ + eπ (z − π) + O ((z − π)2 )) (1 + O ((z − π)2 ))
=
(z − π) sin z (z − π) (−(z − π) + O ((z − π)3 ))
− eπ − eπ (z − π) + O ((z − π)2 )
=
−(z − π)2 + O ((z − π)4 )
eπ eπ
(z−π)2
+ z−π + O(1)
=
1 + O ((z − π)2 )
eπ eπ
 
+ O(1) 1 + O (z − π)2



= 2
+
(z − π) z−π
π
e eπ
= + + O(1)
(z − π)2 z − π
With this we see that
ez cos z
 
Res ,z = π = eπ .
(z − π) sin z

The integral is
ez cos z ez cos z ez cos z
I     
dz = i2π Res , z = −π + Res ,z = 0
C (z − π) sin z (z − π) sin z (z − π) sin z
ez cos z
 
+ Res ,z = π
(z − π) sin z
e−π
 
1 π
= i2π − − +e
π 2π

ez
I
dz = i 2π eπ − 2 − e−π


C (z − π) tan z

561
Cauchy Principal Value for Real Integrals

Solution 15.6
Consider the integral

Z 1
1
dx.
−1 x

By the definition of improper integrals we have

Z 1 Z − Z 1
1 1 1
dx = lim+ dx + lim+ dx
−1 x →0 −1 x δ→0 δ x
= lim+ [log |x|]−
−1 + lim+ [log |x|]1δ
→0 δ→0
= lim+ log  − lim+ log δ
→0 δ→0

This limit diverges. Thus the integral diverges.

Now consider the integral

Z 1
1
dx
−1 x − iα

562
where α ∈ R, α 6= 0. Since the integrand is bounded, the integral exists.
Z 1 Z 1
1 x + iα
dx = 2 2
dx
−1 x − iα −1 x + α
Z 1
iα
= 2 2
dx
−1 x + α
Z 1
α
= 2i 2 2
dx
0 x +α
Z 1/α
1
= 2i 2
dξ
0 ξ +1
= 2i [arctan ξ]1/α
 0
1
= 2i arctan
α
Note that the integral exists for all nonzero real α and that
Z 1
1
lim+ dx = iπ
α→0 −1 x − iα

and
Z 1
1
lim− dx = −iπ.
α→0 −1 x − iα

The integral exists for α arbitrarily close to zero, but diverges when α = 0. The real part of the integrand
is an odd function with two humps that get thinner and taller with decreasing α. The imaginary part of the
integrand is an even function with a hump that gets thinner and taller with decreasing α. (See Figure 15.7.)
   
1 x 1 α
< = 2 2
, = = 2
x − iα x +α x − iα x + α2

563
 Figure 15.7: The real and imaginary part of the integrand for several values of α.

Note that
Z 1
1
< dx → +∞ as α → 0+
0 x − iα
and
Z 0
1
< dx → −∞ as α → 0− .
−1 x − iα
However,
Z 1
1
lim < dx = 0
α→0 −1 x − iα
because the two integrals above cancel each other.
Now note that when α = 0, the integrand is real. Of course the integral doesn’t converge for this case, but if
we could assign some value to
Z 1
1
dx
−1 x

564
it would be a real number. Since
Z 1  
1
lim < dx = 0,
α→0 −1 x − iα

This number should be zero.

Solution 15.7
1.
Z 1 Z − Z 1 
1 1 1
− 2 dx = lim+ 2
dx + 2
dx
−1 x →0 −1 x  x
 −  1 !
1 1
= lim+ − + −
→0 x −1 x 
 
1 1
= lim+ −1−1+
→0  

The principal value of the integral does not exist.

2.
Z 1 Z − Z 1 
1 1 1
− 3 dx = lim+ 3
dx + 3
dx
−1 x →0 −1 x  x
 −  1 !
1 1
= lim+ − 2 + − 2
→0 2x −1 2x 
 
1 1 1 1
= lim+ − + − +
→0 2(−)2 2(−1)2 2(1)2 22
=0

565
 3. Since f (x) is real analytic,
∞
X
f (x) = f n xn for x ∈ (−1, 1).
n=1

We can rewrite the integrand as

f (x) f0 f1 f2 f (x) − f0 − f1 x − f2 x2
= + + + .
x3 x3 x2 x x3
Note that the final term is real analytic on (−1, 1). Thus the principal value of the integral exists if and
only if f2 = 0.

Cauchy Principal Value for Contour Integrals

Solution 15.8
We can write f (z) as

f0 (z − z0 )f (z) − f0
f (z) = + .
z − z0 z − z0
Note that the second term is analytic in a neighborhood of z0 . Thus it is bounded on the contour. Let M be the
maximum modulus of (z−z0z−z
)f (z)−f0
0
on C . By using the maximum modulus integral bound, we have
Z 
 (z − z0 )f (z) − f0 
 dz  ≤ (β − α)M

C z − z0
→ 0 as  → 0+ .

Thus we see that

Z Z
f0
lim+ f (z) dz lim+ dz.
→0 C →0 C z − z0

566
We parameterize the path of integration with

z = z0 + eiθ , θ ∈ (α, β).

Now we evaluate the integral.

Z Z β
f0 f0
lim+ dz = lim+ ieiθ dθ
→0 C z − z0 →0 α eiθ
Z β
= lim+ if0 dθ
→0 α
= i(β − α)f0
≡ i(β − α) Res (f (z), z0 )

This proves the result.

Solution 15.9
Let Ci be the contour that is indented with circular arcs or radius  at each of the first order poles on C so as to
enclose these poles. Let A1 , . . . , An be these circular arcs of radius  centered at the points ζ1 , . . . , ζn . Let Cp be
the contour, (not necessarily connected), obtained by subtracting each of the Aj ’s from Ci .
Since the curve is C 1 , (or continuously differentiable), at each of the first order poles on C, the Aj ’s becomes
semi-circles as  → 0+ . Thus

Z
f (z) dz = iπ Res (f (z), ζj ) for j = 1, . . . , n.
Aj

567
 The principal value of the integral along C is
Z Z
− f (z) dz = lim+ f (z) dz
C →0 Cp
Z n Z
!
X
= lim+ f (z) dz − f (z) dz
→0 Ci Aj
j=1
m n
! n
X X X
= i2π Res (f (z), zj ) + Res (f (z), ζj ) − iπ Res (f (z), ζj )
j=1 j=1 j=1

Z m
X n
X
− f (z) dz = i2π Res (f (z), zj ) + iπ Res (f (z), ζj ).
C j=1 j=1

Solution 15.10
Consider
Z
1
− dz
C z−1

where C is the unit circle. Let Cp be the circular arc of radius 1 that starts and ends a distance of  from z = 1.
Let C be the negative, circular arc of radius  with center at z = 1 that joins the endpoints of Cp . Let Ci , be the
union of Cp and C . (Cp stands for Principal value Contour; Ci stands for Indented Contour.) Ci is an indented
contour that avoids the first order pole at z = 1. Figure 15.8 shows the three contours.

Figure 15.8: The Indented Contour.

568
 Note that the principal value of the integral is
Z Z
1 1
− dz = lim+ dz.
C z−1 →0 Cp z − 1

We can calculate the integral along Ci with Cauchy’s theorem. The integrand is analytic inside the contour.
Z
1
dz = 0
Ci z − 1

We can calculate the integral along C using Result 15.3.1. Note that as  → 0+ , the contour becomes a semi-circle,
a circular arc of π radians in the negative direction.
Z  
1 1
lim dz = −iπ Res , 1 = −iπ
→0+ C z − 1 z−1

Now we can write the principal value of the integral along C in terms of the two known integrals.
Z Z Z
1 1 1
− dz = dz − dz
C z−1 Ci z − 1 C z − 1
= 0 − (−iπ)
= iπ

Integrals from −∞ to ∞
Solution 15.11
Let CR be the semicircular arc from R to −R in the upper half plane. Let C be the union of CR and the interval
[−R, R]. We can evaluate the principal value of the integral along C with Result 15.3.2.
Z m
X n
X
− f (x) dx = i2π Res (f (z), zk ) + iπ Res (f (z), xk )
C k=1 k=1

569
We examine the integral along CR as R → ∞.
Z 
 
 f (z) dz  ≤ πR max |f (z)|

CR
 z∈CR

→ 0 as R → ∞.

Now we are prepared to evaluate the real integral.

Z ∞ Z R
− f (x) dx = lim − f (x) dx
−∞ R→∞ −R
Z
= lim − f (z) dz
R→∞ C
Xm n
X
= i2π Res (f (z), zk ) + iπ Res (f (z), xk )
k=1 k=1

If we close the path of integration in the lower half plane, the contour will be in the negative direction.
Z ∞ Xm n
X
− f (x) dx = −i2π Res (f (z), zk ) − iπ Res (f (z), xk )
−∞ k=1 k=1

Solution 15.12
We consider
Z ∞
2x
− dx.
−∞ x2 +x+1

With the change of variables x = 1/ξ, this becomes

Z −∞
2ξ −1
 
−1
− −2 −1
dξ,
∞ ξ + ξ + 1 ξ2

570
 Z ∞
2ξ −1
− dξ
−∞ ξ2 + ξ + 1
√
There are first order poles at ξ = 0 and ξ = −1/2 ± i 3/2. We close the path of integration in the upper half
plane with a semi-circle. Since the integrand decays like ξ −3 the integrand along the semi-circle vanishes as the
radius tends to infinity. The value of the integral is thus
√ !
2z −1 2z −1
 
1 3
iπ Res 2
, z = 0 + i2π Res 2
,z = − + i
z +z+1 z +z+1 2 2

2z −1
   
2
iπ lim + i2π lim√ √
z→0 z2 + z + 1 z→(−1+i 3)/2 z + (1 + i 3)/2

Z ∞
2x 2π
− dx = − √
−∞ x2 + x + 1 3

Solution 15.13
1. Consider
Z ∞
1
dx.
−∞ x4 + 1

The integrand z41+1 is analytic on the real axis and has isolated singularities at the points z = { eiπ/4 , ei3π/4 , ei5π/4 ,
Let CR be the semi-circle of radius R in the upper half plane. Since
    
 1  1
lim R max  4  = lim R = 0,
R→∞ z∈CR z + 1  R→∞ R4 − 1

571
we can apply Result 15.4.1.
Z ∞     
1 1 iπ/4 1 i3π/4
4
dx = i2π Res ,e + Res ,e
−∞ x + 1 z4 + 1 z4 + 1
The appropriate residues are,
z − eiπ/4
 
1
Res , eiπ/4 = lim
z4 + 1 4
z→ eiπ/4 z + 1
1
= lim 3
z→ eiπ/4 4z
1
= e−i3π/4
4
−1 − i
= √ ,
4 2
 
1 1
Res 4
, ei3π/4 =
z +1 4( ei3π/4 )3
1 −iπ/4
= e
4
1−i
= √ ,
4 2
We evaluate the integral with the residue theorem.
Z ∞  
1 −1 − i 1 − i
4
dx = i2π √ + √
−∞ x + 1 4 2 4 2

Z ∞
1 π
dx = √
−∞ x4 +1 2

572
2. Now consider
∞
x2
Z
dx.
−∞ (x2 + 1)2
The integrand is analytic on the real axis and has second order poles at z = ±i. Since the integrand decays
sufficiently fast at infinity,
2 2
    
 z  R
lim R max  2  = lim R =0
R→∞ z∈CR (z + 1)2  R→∞ (R2 − 1)2
we can apply Result 15.4.1.
∞
x2 z2
Z  
dx = i2π Res ,z = i
−∞ (x2 + 1)2 (z 2 + 1)2

z2 z2
   
d 2
Res ,z = i = lim (z − i) 2
(z 2 + 1)2 z→i dz (z + 1)2
z2
 
d
= lim
z→i dz (z + i)2
(z + i)2 2z − z 2 2(z + i)
 
= lim
z→i (z + i)4
i
=−
4

∞
x2
Z
π
2 2
dx =
−∞ (x + 1) 2

3. Since
sin(x)
1 + x2

573
 is an odd function,
∞ ∞
eix
Z Z
cos(x)
dx = dx
−∞ 1 + x2 −∞ 1 + x2

Since eiz /(1 + z 2 ) is analytic except for simple poles at z = ±i and the integrand decays sufficiently fast in
the upper half plane,
  iz   
 e  1
lim R max   = lim R =0
R→∞ z∈CR 1 + z 2  R→∞ R2 − 1
we can apply Result 15.4.1.
∞
eix eiz
Z  
dx = i2π Res ,z = i
−∞ 1 + x2 (z − i)(z + i)
e−1
= i2π
i2

Z ∞
cos(x) π
dx =
−∞ 1 + x2 e

Solution 15.14
Consider the function

z6
f (z) = .
(z 4 + 1)2
The value of the function on the imaginary axis:
−y 6
(y 4 + 1)2

574
is a constant multiple of the value of the function on the real axis:

x6
.
(x4 + 1)2

Thus to evaluate the real integral we consider the path of integration, C, which starts at the origin, follows the
real axis to R, follows a circular path to iR and then follows
√ the imaginary axis back √down to the origin. f (z)
has second order poles at the fourth roots of −1: (±1 ± i)/ 2. Of these only (1 + i)/ 2 lies inside the path of
integration. We evaluate the contour integral with the Residue Theorem. For R > 1:

z6 z6
Z  
iπ/4
dz = i2π Res ,z = e
C (z 4 + 1)2 (z 4 + 1)2
z6
 
d iπ/4 2
= i2π lim (z − e ) 4
z→ eiπ/4 dz (z + 1)2
z6
 
d
= i2π lim
z→ eiπ/4 dz (z − ei3π/4 )2 (z − ei5π/4 )2 (z − ei7π/4 )2
 z6
= i2π lim i3π/4 )2 (z − ei5π/4 )2 (z − ei7π/4 )2
z→ eiπ/4 (z − e
 
6 2 2 2
− − −
z z − ei3π/4 z − ei5π/4 z − ei7π/4
√ √ !
−i 6 2 2 2 2 2
= i2π −√ − − √
(2)(4i)(−2) 1 + i 2 2 + i2 i 2
3 √
= i2π (1 − i) 2
32
3π
= √ (1 + i)
8 2

The integral along the circular part of the contour, CR , vanishes as R → ∞. We demonstrate this with the

575
maximum modulus integral bound.

z6 6
Z   
  πR z
 dz  ≤ max

CR (z 4 + 1)2  4 z∈CR (z 4 + 1)2
πR R6
=
4 (R4 − 1)2
→ 0 as R → ∞

Taking the limit R → ∞, we have:

∞ Z 0
x6 (iy)6
Z
3π
4 2
dx + 4 2
i dy = √ (1 + i)
(x + 1) ∞ ((iy) + 1) 8 2
Z0 ∞ 6 Z ∞ 6
x y 3π
4 2
dx + i 4 2
dy = √ (1 + i)
0 (x + 1) 0 (y + 1) 8 2
Z ∞ 6
x 3π
(1 + i) 4 2
dx = √ (1 + i)
0 (x + 1) 8 2
Z ∞ 6
x 3π
4 2
dx = √
0 (x + 1) 8 2

Fourier Integrals
Solution 15.15
We know that

Z π
π
e−R sin θ dθ < .
0 R

576
First take the case that ω is positive and the semi-circle is in the upper half plane.
Z  Z 
 iωz
  iωz

 f (z) e dz ≤ e dz  max |f (z)|
    z∈CR
CR CR
Z π 
 iωR eiθ iθ 
≤ e R e  dθ max |f (z)|
0 z∈CR
Z π
 −ωR sin θ 
=R e  dθ max |f (z)|
0 z∈CR
π
<R max |f (z)|
ωR z∈CR
π
= max |f (z)|
ω z∈CR
→ 0 as R → ∞

The procedure is almost the same for negative ω.

Solution 15.16
First we write the integral in terms of Fourier integrals.
∞ ∞ ∞
ei2x e−i2x
Z Z Z
cos 2x
dx = dx + dx
−∞ x − iπ −∞ 2(x − iπ) −∞ 2(x − iπ)
1
Note that 2(z−iπ) vanishes as |z| → ∞. We close the former Fourier integral in the upper half plane and the latter
in the lower half plane. There is a first order pole at z = iπ in the upper half plane.
Z ∞
ei2x ei2z
 
dx = i2π Res , z = iπ
−∞ 2(x − iπ) 2(z − iπ)
e−2π
= i2π
2

577
There are no singularities in the lower half plane.
Z ∞
e−i2x
dx = 0
−∞ 2(x − iπ)
Thus the value of the original real integral is
Z ∞
cos 2x
dx = iπ e−2π
−∞ x − iπ

Fourier Cosine and Sine Integrals

Solution 15.17
We are considering the integral
Z ∞
sin x
dx.
−∞ x
The integrand is an entire function. So it doesn’t appear that the residue theorem would directly apply. Also the
integrand is unbounded as x → +i∞ and x → −i∞, so closing the integral in the upper or lower half plane is
not directly applicable. In order to proceed, we must write the integrand in a different form. Note that
Z ∞
cos x
− dx = 0
−∞ x
since the integrand is odd and has only a first order pole at x = 0. Thus
Z ∞ Z ∞ ix
sin x e
dx = − dx.
−∞ x −∞ ix

Let CR be the semicircular arc in the upper half plane from R to −R. Let C be the closed contour that is the
union of CR and the real interval [−R, R]. If we close the path of integration with a semicircular arc in the upper
half plane, we have
Z ∞ Z iz
eiz
Z 
sin x e
dx = lim − dz − dz ,
−∞ x R→∞ C iz CR iz

578
provided that all the integrals exist.
The integral along CR vanishes as R → ∞ by Jordan’s lemma. By the residue theorem for principal values
we have
Z iz  iz 
e e
− dz = iπ Res , 0 = π.
iz iz

Combining these results,

Z ∞
sin x
dx = π.
−∞ x

Solution 15.18
Note that (1 − cos x)/x2 has a removable singularity at x = 0. The integral decays like x12 at infinity, so the
integral exists. Since (sin x)/x2 is a odd function with a simple pole at x = 0, the principal value of its integral
vanishes.
Z ∞
sin x
− 2
dx = 0
−∞ x
Z ∞ Z ∞ Z ∞
1 − cos x 1 − cos x − i sin x 1 − eix
dx = − dx = − dx
−∞ x2 −∞ x2 −∞ x2

Let CR be the semi-circle of radius R in the upper half plane. Since

 1 − eiz 
  
lim R max    = lim R 2 = 0
R→∞ z∈CR z 2  R→∞ R2

the integral along CR vanishes as R → ∞.

1 − eiz
Z
dz → 0 as R → ∞
CR z2

579
We can apply Result 15.4.1.
Z ∞
1 − eix 1 − eiz 1 − eiz −i eiz
 
− dx = iπ Res , z = 0 = iπ lim = iπ lim
−∞ x2 z2 z→0 z z→0 1

∞
1 − cos x
Z
dx = π
−∞ x2

Solution 15.19
Consider
Z ∞
sin(πx)
dx.
0 x(1 − x2 )
Note that the integrand has removable singularities at the points x = 0, ±1 and is an even function.
Z ∞
1 ∞ sin(πx)
Z
sin(πx)
dx = dx.
0 x(1 − x2 ) 2 −∞ x(1 − x2 )
cos(πx)
Note that is an odd function with first order poles at x = 0, ±1.
x(1 − x2 )
Z ∞
cos(πx)
− 2
dx = 0
−∞ x(1 − x )
Z ∞
i ∞ eiπx
Z
sin(πx)
dx = − − dx.
0 x(1 − x2 ) 2 −∞ x(1 − x2 )
Let CR be the semi-circle of radius R in the upper half plane. Since
 eiπz 
  
1
lim R max  2
 = lim R
2
=0
R→∞ z∈CR z(1 − z )  R→∞ R(R − 1)

580
the integral along CR vanishes as R → ∞.

eiπz
Z
dz → 0 as R → ∞
CR z(1 − z 2 )

We can apply Result 15.4.1.

i ∞ eiπx eiz eiz

    
−i
Z
− − dx = iπ Res , z = 0 + Res ,z = 1
2 −∞ x(1 − x2 ) 2 z(1 − z 2 ) z(1 − z 2 )
eiz
 
+ Res , z = −1
z(1 − z 2 )
eiπz eiπz eiπz
 
π
= lim − lim + lim
2 z→0 1 − z 2 z→0 z(1 + z) z→0 z(1 − z)
 
π −1 −1
= 1− +
2 2 −2

Z ∞
sin(πx)
dx = π
0 x(1 − x2 )

Contour Integration and Branch Cuts

Solution 15.20
Let C be the boundary of the region  < r < R, 0 < θ < π. Choose the branch of the logarithm with a branch cut
on the negative imaginary axis and the angle range −π/2 < θ < 3π/2. We consider the integral of (log z)2 /(1+z 2 )

581
on this contour.

(log z)2 (log z)2

I  
dz = i2π Res ,z = i
C 1 + z2 1 + z2
(log z)2
= i2π lim
z→i z + i
(iπ/2)2
= i2π
2i
π3
=−
4

Let CR be the semi-circle from R to −R in the upper half plane. We show that the integral along CR vanishes as
R → ∞ with the maximum modulus integral bound.

(log z)2   (log z)2 

Z   

 dz  ≤ πR max  

CR 1 + z2 z∈CR 1 + z 2 

(ln R)2 + 2π ln R + π 2
≤ πR
R2 − 1
→ 0 as R → ∞

Let C be the semi-circle from − to  in the upper half plane. We show that the integral along C vanishes as
 → 0 with the maximum modulus integral bound.

(log z)2   (log z)2 

Z   

 dz  ≤ π max  

C 1 + z2 z∈C 1 + z 2 

(ln )2 − 2π ln  + π 2
≤ π
1 − 2
→ 0 as  → 0

582
 Now we take the limit as  → 0 and R → ∞ for the integral along C.

(log z)2 π3
I
2
dz = −
C 1+z 4
Z ∞ 2 Z 0
(ln r) (ln r + iπ)2 π3
dr + dr = −
0 1 + r2 ∞ 1 + r2 4
∞ 2 ∞ Z ∞
π3
Z Z
(ln x) ln x 2 1
2 dx + i2π dx = π dx − (15.1)
0 1 + x2 0 1 + x2 0 1 + x2 4

We evaluate the integral of 1/(1 + x2 ) by extending the path of integration to (−∞ . . . ∞) and closing the
path of integration in the upper half plane. Since
    
 1  1
lim R max    ≤ lim R 2 = 0,
R→∞ z∈CR 1 + z 2  R→∞ R −1

the integral of 1/(1 + z 2 ) along CR vanishes as R → ∞. We evaluate the integral with the Residue Theorem.
∞
π2 ∞ 1
Z Z
2 1
π dx = dx
0 1 + x2 2 −∞ 1 + x2
π2
 
1
= i2π Res ,z = i
2 1 + z2
1
= iπ 3 lim
z→i z + i
π3
=
2
Now we return to Equation 15.1.
∞ ∞
(ln x)2 π3
Z Z
ln x
2 dx + i2π dx =
0 1 + x2 0 1 + x2 4

583
We equate the real and imaginary parts to solve for the desired integrals.

∞
(ln x)2 π3
Z
dx =
0 1 + x2 8
Z ∞
ln x
dx = 0
0 1 + x2

Solution 15.21
We consider the branch of the function

log z
f (z) =
z2 + 5z + 6

with a branch cut on the real axis and 0 < arg(z) < 2π.
Let C and CR denote the circles of radius  and R where  < 1 < R. C is negatively oriented; CR is positively
oriented. Consider the closed contour, C, that is traced by a point moving from  to R above the branch cut,
next around CR back to R, then below the cut to , and finally around C back to . (See Figure 15.10.)

CR

Cε

Figure 15.9: The path of integration.

584
 We can evaluate the integral of f (z) along C with the residue theorem. For R > 3, there are first order poles
inside the path of integration at z = −2 and z = −3.
Z     
log z log z log z
2
dz = i2π Res , z = −2 + Res , z = −3
C z + 5z + 6 z 2 + 5z + 6 z 2 + 5z + 6
 
log z log z
= i2π lim + lim
z→−2 z + 3 z→−3 z + 2
 
log(−2) log(−3)
= i2π +
1 −1
= i2π (log(2) + iπ − log(3) − iπ)
 
2
= i2π log
3
In the limit as  → 0, the integral along C vanishes. We demonstrate this with the maximum modulus
theorem.
Z   
 log z   log z 

2
dz  ≤ 2π max  2
  

C z + 5z + 6 z∈C z + 5z + 6 
2π − log 
≤ 2π
6 − 5 − 2
→ 0 as  → 0

In the limit as R → ∞, the integral along CR vanishes. We again demonstrate this with the maximum modulus
theorem.
Z   
 log z   log z 

2 + 5z + 6
dz  ≤ 2πR max  
 z 2 + 5z + 6 

CR z  z∈CR

log R + 2π
≤ 2πR 2
R − 5R − 6
→ 0 as R → ∞

585
 Taking the limit as  → 0 and R → ∞, the integral along C is:
Z Z ∞ Z 0
log z log x log x + i2π
2
dz = 2
dx + 2
dx
C z + 5z + 6 0 x + 5x + 6 ∞ x + 5x + 6
Z ∞
log x
= −i2π 2
dx
0 x + 5x + 6
Now we can evaluate the real integral.
Z ∞  
log x 2
−i2π 2
dx = i2π log
0 x + 5x + 6 3
Z ∞  
log x 3
2
dx = log
0 x + 5x + 6 2

Solution 15.22
We consider the integral
∞
xa
Z
I(a) = dx.
0 (x + 1)2
To examine convergence, we split the domain of integration.
Z ∞ Z 1 Z ∞
xa xa xa
dx = dx + dx
0 (x + 1)2 0 (x + 1)
2
1 (x + 1)2
First we work with the integral on (0 . . . 1).
Z 1 a
 Z 1
 xa 

 x 

2
dx ≤  (x + 1)2  |dx|
 
0 (x + 1)

0
Z 1
x<(a)
= 2
dx
0 (x + 1)
Z 1
≤ x<(a) dx
0

586
This integral converges for <(a) > −1.
Next we work with the integral on (1 . . . ∞).
Z ∞  Z ∞
xa  xa 

 

2
dx ≤

 (x + 1)2  |dx|
 

1 (x + 1) 1
Z ∞
x<(a)
= dx
1 (x + 1)2
Z ∞
≤ x<(a)−2 dx
1

This integral converges for <(a) < 1.

Thus we see that the integral defining I(a) converges in the strip, −1 < <(a) < 1. The integral converges
uniformly in any closed subset of this domain. Uniform convergence means that we can differentiate the integral
with respect to a and interchange the order of integration and differentiation.
Z ∞ a
0 x log x
I (a) = dx
0 (x + 1)2

Thus we see that I(a) is analytic for −1 < <(a) < 1.

For −1 < <(a) < 1 and a 6= 0, z a is multi-valued. Consider the branch of the function f (z) = z a /(z + 1)2
with a branch cut on the positive real axis and 0 < arg(z) < 2π. We integrate along the contour in Figure 15.10.
The integral on C vanishes as  → 0. We show this with the maximum modulus integral bound. First we
write z a in modulus-argument form, z =  eiθ , where a = α + iβ.

z a = ea log z
= e(α+iβ)(ln +iθ)
= eα ln −βθ+i(β ln +αθ)
= α e−βθ ei(β log +αθ)

587
Now we bound the integral.

za  za 
Z   
 
 dz  ≤ 2π max  

C (z + 1)2  z∈C (z + 1)2 

α e2π|β|
≤ 2π
(1 − )2
→ 0 as  → 0

The integral on CR vanishes as R → ∞.

za  za 
Z   
 
 dz  ≤ 2πR max  

CR (z + 1)2  z∈CR (z + 1)2 

Rα e2π|β|
≤ 2πR
(R − 1)2
→ 0 as R → ∞

Above the branch cut, (z = r ei0 ), the integrand is

ra
f (r ei0 ) = .
(r + 1)2

Below the branch cut, (z = r ei2π ), we have,

ei2πa ra
f (r ei2π ) = .
(r + 1)2

588
Now we use the residue theorem.
Z ∞ Z 0 i2πa a
ra za
 
e r
dr + dr = i2π Res , −1
0 (r + 1)2 ∞ (r + 1)
2 (z + 1)2

∞ ra
Z
i2πa d a
1− e 2
dr = i2π lim (z )
0 (r + 1) z→−1 dz
Z ∞
ra a eiπ(a−1)
dr = i2π
(r + 1)2 1 − ei2πa
Z0 ∞ a
r −i2πa
2
dr = −iπa
0 (r + 1) e − eiπa
Z ∞
xa πa
2
dx = for − 1 < <(a) < 1, a 6= 0
0 (x + 1) sin(πa)
The right side has a removable singularity at a = 0. We use analytic continuation to extend the answer to a = 0.
Z ∞ (
a πa
x sin(πa)
for − 1 < <(a) < 1, a 6= 0
I(a) = dx =
0 (x + 1)2 1 for a = 0

We can derive the last two integrals by differentiating this formula with respect to a and taking the limit
a → 0.
Z ∞ a Z ∞ a 2
0 x log x 00 x log x
I (a) = 2
dx, I (a) = dx
0 (x + 1) 0 (x + 1)2
Z ∞ Z ∞
0 log x 00 log2 x
I (0) = dx, I (0) = dx
0 (x + 1)2 0 (x + 1)2
We can find I 0 (0) and I 00 (0) either by differentiating the expression for I(a) or by finding the first few terms in
the Taylor series expansion of I(a) about a = 0. The latter approach is a little easier.
∞
X I (n) (0)
I(a) = an
n=0
n!

589
 πa
I(a) =
sin(πa)
πa
=
πa − (πa)3 /6 + O(a5 )
1
=
1 − (πa) /6 + O(a4 )
2

π 2 a2
=1+ + O(a4 )
6

Z ∞
0 log x
I (0) = dx = 0
0 (x + 1)2
∞
log2 x π2
Z
I 00 (0) = dx =
0 (x + 1)2 3

Solution 15.23
1. We consider the integral

∞
xa
Z
I(a) = dx.
0 1 + x2

To examine convergence, we split the domain of integration.

∞ 1 ∞
xa xa xa
Z Z Z
dx = dx + dx
0 1 + x2 0 1 + x2 1 1 + x2

590
 First we work with the integral on (0 . . . 1).
Z 1 a
 Z 1 a 
 x 
≤
 x 
dx  1 + x2  |dx|
  
2
0 1+x
 
0
Z 1 <(a)
x
= 2
dx
0 1+x
Z 1
≤ x<(a) dx
0

This integral converges for <(a) > −1.

Next we work with the integral on (1 . . . ∞).
Z ∞  Z ∞ a 
 xa 
≤
 x 
dx  1 + x2  |dx|
  
 1 + x 2 
1 1
Z ∞ <(a)
x
= dx
1 1 + x2
Z ∞
≤ x<(a)−2 dx
1

This integral converges for <(a) < 1.

Thus we see that the integral defining I(a) converges in the strip, −1 < <(a) < 1. The integral converges
uniformly in any closed subset of this domain. Uniform convergence means that we can differentiate the
integral with respect to a and interchange the order of integration and differentiation.
Z ∞ a
0 x log x
I (a) = dx
0 1 + x2
Thus we see that I(a) is analytic for −1 < <(a) < 1.
2. For −1 < <(a) < 1 and a 6= 0, z a is multi-valued. Consider the branch of the function f (z) = z a /(1 + z 2 )
with a branch cut on the positive real axis and 0 < arg(z) < 2π. We integrate along the contour in
Figure 15.10.

591
 CR

Cε

Figure 15.10:

The integral on Cρ vanishes are ρ → 0. We show this with the maximum modulus integral bound. First we
write z a in modulus-argument form, where z = ρ eiθ and a = α + iβ.

z a = ea log z
= e(α+iβ)(log ρ+iθ)
= eα log ρ−βθ+i(β log ρ+αθ)
= ρa e−βθ ei(β log ρ+αθ)

Now we bound the integral.

Z 
a
 a 
 z   z 
dz ≤ 2πρ max
   
 Cρ 1 + z 2  z∈Cρ  1 + z 2 
 

ρα e2π|β|
≤ 2πρ
1 − ρ2
→ 0 as ρ → 0

592
The integral on CR vanishes as R → ∞.

za
Z   a 
 ≤ 2πR max  z 
   
 dz

CR 1 + z2  z∈CR  1 + z 2 

Rα e2π|β|
≤ 2πR
R2 − 1
→ 0 as R → ∞

Above the branch cut, (z = r ei0 ), the integrand is

ra
f (r ei0 ) = .
1 + r2

Below the branch cut, (z = r ei2π ), we have,

ei2πa ra
f (r ei2π ) = .
1 + r2

593
 Now we use the residue theorem.
Z ∞ Z 0 i2πa a
ra
  a   a 
e r z z
dr + dr = i2π Res , i + Res , −i
0 1 + r2 ∞ 1+r
2 1 + z2 1 + z2

∞ xa za za
Z  
i2πa
1− e dx = i2π lim + lim
0 1 + x2 z→i z + i z→−i z − i
Z ∞
xa eiaπ/2 eia3π/2
 
i2πa


1− e dx = i2π +
0 1 + x2 2i −2i
Z ∞ a iaπ/2 ia3π/2
x e −e
2
dx = π
0 1+x 1 − ei2aπ
Z ∞
xa eiaπ/2 (1 − eiaπ )
dx = π
0 1 + x2 (1 + eiaπ )(1 − eiaπ )
Z ∞
xa π
2
dx = −iaπ/2
0 1+x e + eiaπ/2
Z ∞
xa π
2
dx = for − 1 < <(a) < 1, a 6= 0
0 1+x 2 cos(πa/2)
We use analytic continuation to extend the answer to a = 0.
Z ∞
xa π
I(a) = 2
dx = for − 1 < <(a) < 1
0 1+x 2 cos(πa/2)

3. We can derive the last two integrals by differentiating this formula with respect to a and taking the limit
a → 0.
Z ∞ a Z ∞ a 2
0 x log x 00 x log x
I (a) = 2
dx, I (a) = dx
0 1+x 0 1 + x2
Z ∞ Z ∞
0 log x 00 log2 x
I (0) = dx, I (0) = dx
0 1 + x2 0 1 + x2

594
 We can find I 0 (0) and I 00 (0) either by differentiating the expression for I(a) or by finding the first few terms
in the Taylor series expansion of I(a) about a = 0. The latter approach is a little easier.
∞
X I (n) (0)
I(a) = an
n=0
n!

π
I(a) =
2 cos(πa/2)
π 1
=
2 1 − (πa/2) /2 + O(a4 )
2
π
1 + (πa/2)2 /2 + O(a4 )


=
2
π π 3 /8 2
= + a + O(a4 )
2 2
Z ∞
0 log x
I (0) = dx = 0
0 1 + x2
∞
log2 x π3
Z
00
I (0) = dx =
0 1 + x2 8

Solution 15.24
Convergence. If xa f (x)  xα as x → 0 for some α > −1 then the integral
Z 1
xa f (x) dx
0

will converge absolutely. If xa f (x)  xβ as x → ∞ for some β < −1 then the integral
Z ∞
xa f (x)
1

595
will converge absolutely. These are sufficient conditions for the absolute convergence of
Z ∞
xa f (x) dx.
0

Contour Integration. We put a branch cut on the positive real axis and choose 0 < arg(z) < 2π. We
consider the integral of z a f (z) on the contour in Figure 15.10. Let the singularities of f (z) occur at z1 , . . . , zn .
By the residue theorem,
Z n
X
a
z f (z) dz = i2π Res (z a f (z), zk ) .
C k=1

On the circle of radius , the integrand is o(−1 ). Since the length of C is 2π, the integral on C vanishes
as  → 0. On the circle of radius R, the integrand is o(R−1 ). Since the length of CR is 2πR, the integral on CR
vanishes as R → ∞.
The value of the integrand below the branch cut, z = x ei2π , is

f (x ei2π ) = xa ei2πa f (x)

In the limit as  → 0 and R → ∞ we have

Z ∞ Z 0 n
X
a a i2πa
x f (x) dx + x e f (x) dx = i2π Res (z a f (z), zk ) .
0 −∞ k=1

Z ∞ n
a i2π X
x f (x) dx = Res (z a f (z), zk ) .
0 1 − ei2πa k=1

596
Solution 15.25
In the interval of uniform convergence of th integral, we can differentiate the formula
Z ∞ n
a i2π X
x f (x) dx = i2πa
Res (z a f (z), zk ) ,
0 1− e k=1

with respect to a to obtain,

Z ∞ n n
i2π X 4π 2 a ei2πa X
xa f (x) log x dx = a
Res (z f (z) log z, zk ) , − Res (z a f (z), zk ) .
0 1 − ei2πa k=1 (1 − ei2πa )2 k=1

∞ n n
π2a X
Z
a i2π X a
x f (x) log x dx = Res (z f (z) log z, zk ) , + 2 Res (z a f (z), zk ) ,
0 1 − ei2πa k=1 sin (πa) k=1

Differentiating the solution of Exercise 15.22 m times with respect to a yields

n
!
∞
∂m
Z
i2π X
xa f (x) logm x dx = m Res (z a f (z), zk ) ,
0 ∂a 1 − ei2πa k=1

Solution 15.26
Taking the limit as a → 0 ∈ Z in the solution of Exercise 15.22 yields
Z ∞  Pn a

k=1 Res (z f (z), zk )
f (x) dx = i2π lim
0 a→0 1 − ei2πa
The numerator vanishes because the sum of all residues of z n f (z) is zero. Thus we can use L’Hospital’s rule.
Z ∞  Pn a

k=1 Res (z f (z) log z, zk )
f (x) dx = i2π lim
0 a→0 −i2π ei2πa

597
 Z ∞ n
X
f (x) dx = − Res (f (z) log z, zk )
0 k=1

This suggests that we could have derived the result directly by considering the integral of f (z) log z on the contour
in Figure 15.10. We put a branch cut on the positive real axis and choose the branch arg z = 0. Recall that we
have assumed that f (z) has only isolated singularities and no singularities on the positive real axis, [0, ∞). By
the residue theorem,
Z n
X
f (z) log z dz = i2π Res (f (z) log z, z = zk ) .
C k=1

By assuming that f (z)  z α as z → 0 where α > −1 the integral on C will vanish as  → 0. By assuming that
f (z)  z β as z → ∞ where β < −1 the integral on CR will vanish as R → ∞. The value of the integrand below
the branch cut, z = x ei2π is f (x)(log x + i2π). Taking the limit as  → 0 and R → ∞, we have
Z ∞ Z 0 n
X
f (x) log x dx + f (x)(log x + i2π) dx = i2π Res (f (z) log z, zk ) .
0 ∞ k=1

Thus we corroborate the result.

Z ∞ n
X
f (x) dx = − Res (f (z) log z, zk )
0 k=1

Solution 15.27
Consider the integral of f (z) log2 z on the contour in Figure 15.10. We put a branch cut on the positive real axis
and choose the branch 0 < arg z < 2π. Let z1 , . . . zn be the singularities of f (z). By the residue theorem,
Z n
X
2
Res f (z) log2 z, zk .


f (z) log z dz = i2π
C k=1

598
If f (z)  z α as z → 0 for some α > −1 then the integral on C will vanish as  → 0. f (z)  z β as z → ∞ for some
β < −1 then the integral on CR will vanish as R → ∞. Below the branch cut the integrand is f (x)(log x + i2π)2 .
Thus we have
Z ∞ Z 0 n
X
2 2 2
Res f (z) log2 z, zk .


f (x) log x dx + f (x)(log x + i4π log x − 4π ) dx = i2π
0 ∞ k=1

Z ∞ Z ∞ n
X
2
Res f (z) log2 z, zk .


−i4π f (x) log x dx + 4π f (x) dx = i2π
0 0 k=1

Z ∞ n n
1X 2

X
f (x) log x dx = − Res f (z) log z, zk + iπ Res (f (z) log z, zk )
0 2 k=1 k=1

Solution 15.28
Convergence. We consider
∞
xa
Z
dx.
0 1 + x4
Since the integrand behaves like xa near x = 0 we must have <(a) > −1. Since the integrand behaves like xa−4
at infinity we must have <(a − 4) < −1. The integral converges for −1 < <(a) < 3.
Contour Integration. The function
za
f (z) =
1 + z4
√
has first order poles at z = (±1 ± i)/ 2 and a branch point at z = 0. We could evaluate the real integral by
putting a branch cut on the positive real axis with 0 < arg(z) < 2π and integrating f (z) on the contour in
Figure 15.11.

599
 CR

Cε

za
Figure 15.11: Possible Path of Integration for f (z) = 1+z 4

Integrating on this contour would work because the value of the integrand below the branch cut is a constant
times the value of the integrand above the branch cut. After demonstrating that the integrals along C and CR
vanish in the limits as  → 0 and R → ∞ we would see that the value of the integral is a constant times the sum
of the residues at the four poles. However, this is not the only, (and not the best), contour that can be used to
evaluate the real integral. Consider the value of the integral on the line arg(z) = θ.
ra eiaθ
f (r eiθ ) =
1 + r4 ei4θ
If θ is a integer multiple of π/2 then the integrand is a constant multiple of
ra
f (x) = .
1 + r4
Thus any of the contours in Figure 15.12 can be used to evaluate the real integral. The only difference is how
many residues we have to calculate. Thus we choose the first contour in Figure 15.12. We put a branch cut on
the negative real axis and choose the branch −π < arg(z) < π to satisfy f (1) = 1.
We evaluate the integral along C with the Residue Theorem.
za
Z  a 
z 1+i
4
dz = i2π Res ,z = √
C 1+z 1 + z4 2

600
 CR CR CR

Cε Cε Cε

za
Figure 15.12: Possible Paths of Integration for f (z) = 1+z 4

Let a = α + iβ and z = r eiθ . Note that

|z a | = |(r eiθ )α+iβ | = rα e−βθ .
The integral on C vanishes as  → 0. We demonstrate this with the maximum modulus integral bound.
a
Z   a 
 z  π
≤
 z 

4
dz max  

C 1 + z
 2 z∈C  1 + z 4 
π α eπ|β|/2
≤
2 1 − 4
→ 0 as  → 0
The integral on CR vanishes as R → ∞.
za
Z   a 
  πR  z 
 dz ≤ max  

CR 1 + z4  2 z∈CR  1 + z 4 
πR Rα eπ|β|/2
≤
2 R4 − 1
→ 0 as R → ∞

601
The value of the integrand on the positive imaginary axis, z = x eiπ/2 , is

(x eiπ/2 )a xa eiπa/2
= .
1 + (x eiπ/2 )4 1 + x4

We take the limit as  → 0 and R → ∞.

∞ Z 0 a iπa/2
xa
Z  a 
x e iπ/2 z iπ/4
dx + e dx = i2π Res ,e
0 1 + x4 ∞ 1+x
4 1 + z4

∞ xa
 a
z (z − eiπ/2 )
Z 
iπ(a+1)/2
1− e dx = i2π lim
0 1 + x4 z→ eiπ/4 1 + z4
Z ∞
xa az a (z − eiπ/2 ) + z a
 
i2π
dx = lim
0 1 + x4 1 − eiπ(a+1)/2 z→ eiπ/4 4z 3
Z ∞
xa i2π eiπa/4
dx =
1 + x4 1 − eiπ(a+1)/2 4 ei3π/4
Z ∞0
xa −iπ
4
dx = −iπ(a+1)/4
0 1+x 2( e − eiπ(a+1)/4 )
Z ∞
xa
 
π π(a + 1)
dx = csc
0 1 + x4 4 4

Solution 15.29
Consider the branch of f (z) = z 1/2 log z/(z + 1)2 with a branch cut on the positive real axis and 0 < arg z < 2π.
We integrate this function on the contour in Figure 15.10.

602
 We use the maximum modulus integral bound to show that the integral on Cρ vanishes as ρ → 0.
Z 
1/2
 1/2 
 z log z   z log z 
dz  ≤ 2πρ max 
  
 Cρ (z + 1)2 (z + 1)2 

 Cρ

ρ1/2 (2π − log ρ)

= 2πρ
(1 − ρ)2
→ 0 as ρ → 0

The integral on CR vanishes as R → ∞.

z 1/2 log z 
Z   1/2 
  z log z 

2
dz  ≤ 2πR max 
 
2 

CR (z + 1) CR (z + 1)
R1/2 (log R + 2π)
= 2πR
(R − 1)2
→ 0 as R → ∞

Above the branch cut, (z = x ei0 ), the integrand is,

x1/2 log x
f (x ei0 ) = .
(x + 1)2

Below the branch cut, (z = x ei2π ), we have,

−x1/2 (log x + iπ)

f (x ei2π ) = .
(x + 1)2

Taking the limit as ρ → 0 and R → ∞, the residue theorem gives us

Z ∞ 1/2 Z 0
−x1/2 (log x + i2π)
 1/2 
x log x z log z
dx + dx = i2π Res , −1 .
0 (x + 1)2 ∞ (x + 1)2 (z + 1)2

603
 ∞ ∞
x1/2 log x x1/2
Z Z
d 1/2
2 dx + i2π dx = i2π lim (z log z)
0 (x + 1)2 0 (x + 1)2 z→−1 dz

∞ ∞
x1/2 log x x1/2
Z Z  
1 −1/2 1
2 dx + i2π dx = i2π lim z log z + z 1/2
0 (x + 1)2 0 (x + 1)2 z→−1 2 z

∞ ∞
x1/2 log x x1/2
Z Z  
1
2 dx + i2π dx = i2π (−i)(iπ) − i
0 (x + 1)2 0 (x + 1)2 2

∞ ∞
x1/2 log x x1/2
Z Z
2 dx + i2π dx = 2π + iπ 2
0 (x + 1)2 0 (x + 1) 2

Equating real and imaginary parts,

∞ ∞
x1/2 log x x1/2
Z Z
π
dx = π, dx = .
0 (x + 1)2 0 (x + 1) 2 2

Exploiting Symmetry
Solution 15.30
Convergence. The integrand,

eaz eaz
= ,
ez − e−z 2 sinh(z)

has first order poles at z = inπ, n ∈ Z. To study convergence, we split the domain of integration.
Z ∞ Z −1 Z 1 Z ∞
= + +
−∞ −∞ −1 1

604
The principal value integral
Z 1
eax
− dx
−1 ex − e−x
exists for any a because the integrand has only a first order pole on the path of integration.
Now consider the integral on (1 . . . ∞).
Z ∞ ax
 Z ∞ (a−1)x
 e  e
 dx = dx

1 ex − e−x 
1 1 − e−2x
Z ∞
1
≤ e(a−1)x dx
1 − e−2 1
This integral converges for a − 1 < 0; a < 1.
Finally consider the integral on (−∞ . . . − 1).
Z −1 ax
 Z −1 (a+1)x
 e  e

x −x
dx =
2x
dx
−∞ e − e −∞ 1 − e
 
Z −1
1
≤ e(a+1)x dx
1 − e−2 −∞
This integral converges for a + 1 > 0; a > −1.
Thus we see that the integral for I(a) converges for real a, |a| < 1.
Choice of Contour. Consider the contour C that is the boundary of the region: −R < x < R, 0 < y < π.
The integrand has no singularities inside the contour. There are first order poles on the contour at z = 0 and
z = iπ. The value of the integral along the contour is iπ times the sum of these two residues.
The integrals along the vertical sides of the contour vanish as R → ∞.
Z R+iπ az  eaz 
  
 e 
 dz  ≤ π max  z 

R ez − e−z  z∈(R...R+iπ) e − e−z 

eaR
≤π
eR − e−R
→ 0 as R → ∞

605
 −R+iπ
eaz  eaz 
Z   
 
 dz  ≤ π max  

−R ez − e−z  z∈(−R...−R+iπ)  ez − e−z 

e−aR
≤ π −R
e − eR
→ 0 as R → ∞

Evaluating the Integral. We take the limit as R → ∞ and apply the residue theorem.

∞ −∞+iπ
eax eaz
Z Z
dx + dz
−∞ ex − e−x ∞+iπ ez − e−z
eaz eaz
   
= iπ Res , z = 0 + iπ Res , z = iπ
ez − e−z ez − e−z

∞ Z −∞
eax ea(x+iπ z eaz (z − iπ) eaz
Z
−x
dx + dz = iπ lim + iπ lim
x
−∞ e − e ∞ ex+iπ − e−x−iπ z→0 2 sinh(z) z→iπ 2 sinh(z)
Z ∞
eax eaz + az eaz eaz + a(z − iπ) eaz
(1 + eiaπ ) x −x
dx = iπ lim + iπ lim
−∞ e − e z→0 2 cosh(z) z→iπ 2 cosh(z)
Z ∞ ax iaπ
e 1 e
(1 + eiaπ ) x −x
dx = iπ + iπ
−∞ e − e 2 −2
Z ∞ ax iaπ
e iπ(1 − e )
x −x
dx =
−∞ e − e 2(1 + eiaπ )
Z ∞
eax π i( e−iaπ/2 − eiaπ/2 )
x −x
dx =
−∞ e − e 2 eiaπ/2 + eiaπ/2
Z ∞
eax π  aπ 
x −x
dx = tan
−∞ e − e 2 2

606
Solution 15.31
1.
Z ∞ Z ∞
dx 1 dx
2 =
0 (1 + x2 ) 2 −∞ (1 + x2 )2
We apply Result 15.4.1 to the integral on the real axis. First we verify that the integrand vanishes fast
enough in the upper half plane.
    
 1  1
lim R max    = lim R =0
R→∞ z∈CR (1 + z 2 )2  R→∞ (R2 − 1)2
Then we evaluate the integral with the residue theorem.
Z ∞  
dx 1
= i2π Res ,z = i
2 2 (1 + z 2 )2
−∞ (1 + x )
 
1
= i2π Res ,z = i
(z − i)2 (z + i)2
d 1
= i2π lim
z→i dz (z + i)2
−2
= i2π lim
z→i (z + i)3
π
=
2
Z ∞
dx π
2 =
0 (1 + x2 ) 4

2. We wish to evaluate
Z ∞
dx
.
0 x3+1

607
Let the contour C be the boundary of the region 0 < r < R, 0 < θ < 2π/3. We factor the denominator of
the integrand to see that the contour encloses the simple pole at eiπ/3 for R > 1.

z 3 + 1 = (z − eiπ/3 )(z + 1)(z − e−iπ/3 ),

We calculate the residue at that point.

   
1 iπ/3 iπ/3 1
Res ,z = e = lim (z − e ) 3
z3 + 1 z→ eiπ/3 z +1
 
1
= lim
z→ eiπ/3 (z + 1)(z − e−iπ/3 )
1
= iπ/3
(e + 1)( eiπ/3 − e−iπ/3 )

We use the residue theorem to evaluate the integral.

I
dz i2π
= iπ/3
3
C z +1 (e + 1)( eiπ/3 − e−iπ/3 )

Let CR be the circular arc portion of the contour.

Z Z R Z Z R 2iπ/3
dz dx dz e dx
= + −
C z3 + 1 3
0 x +1
3
CR z + 1 0 x3 + 1
Z R Z
−iπ/3 dx dz
= (1 + e ) 3
+ 3
0 x +1 CR z + 1

We show that the integral along CR vanishes as R → ∞ with the maximum modulus integral bound.
Z 
 dz  2πR 1
≤
 → 0 as R → ∞
 z3 + 1 
CR 3 R3 − 1

608
 We take R → ∞ and solve for the desired integral.

Z ∞
−iπ/3 dx i2π
(1 + e ) =
0 +1 x3(1 + e )( eiπ/3 − e−iπ/3 )
iπ/3
Z ∞
dx i2π
= −iπ/3
0
3
x +1 (1 + e )(1 + eiπ/3 )( eiπ/3 − e−iπ/3 )
Z ∞
dx i2π
3
= 1 √ 1 √ √
0 x +1 2
(3 + i 3) 2 (3 − i 3)(i 3)
Z ∞
dx 2π
3
= √
0 x +1 3 3

Solution 15.32
Method 1: Semi-Circle Contour. We wish to evaluate the integral

Z ∞
dx
I= .
0 1 + x6

We note that the integrand is an even function and express I as an integral over the whole real axis.

Z ∞
1 dx
I=
2 −∞ 1 + x6

Now we will evaluate the integral using contour integration. We close the path of integration in the upper half
plane. Let ΓR be the semicircular arc from R to −R in the upper half plane. Let Γ be the union of ΓR and the
interval [−R, R]. (See Figure 15.13.)
We can evaluate the integral along Γ with the residue theorem. The integrand has first order poles at

609
 Figure 15.13: The semi-circle contour.

z = eiπ(1+2k)/6 , k = 0, 1, 2, 3, 4, 5. Three of these poles are in the upper half plane. For R > 1, we have

Z 2  
1 X 1 iπ(1+2k)/6
dz = i2π Res ,e
Γ z6 + 1 k=0
z 6+1

2
X z − eiπ(1+2k)/6
= i2π lim
k=0
z→ eiπ(1+2k)/6 z6 + 1

610
Since the numerator and denominator vanish, we apply L’Hospital’s rule.
2
X 1
= i2π lim
k=0
z→ eiπ(1+2k)/6 6z 5
2
iπ X
= e−iπ5(1+2k)/6
3 k=0
iπ −iπ5/6
+ e−iπ15/6 + e−iπ25/6


= e
3
iπ −iπ5/6
+ e−iπ/2 + e−iπ/6


= e
3
√ √ !
iπ − 3 − i 3−i
= −i+
3 2 2
2π
=
3
Now we examine the integral along ΓR . We use the maximum modulus integral bound to show that the value of
the integral vanishes as R → ∞.
Z   
 1   1 
 dz ≤ πR max  6
  
 z6 + 1 
ΓR z∈ΓR z + 1 

1
= πR
R6
−1
→ 0 as R → ∞.
Now we are prepared to evaluate the original real integral.
Z
1 2π
6
dz =
Γ z +1 3
Z R Z
1 1 2π
6
dx + 6
dz =
−R x + 1 ΓR z + 1 3

611
We take the limit as R → ∞.
Z ∞
1 2π
dx =
x6 +1 3
Z−∞
∞
1 π
6
dx =
0 x +1 3

We would get the same result by closing the path of integration in the lower half plane. Note that in this case
the closed contour would be in the negative direction.
Method 2: Wedge Contour. Consider the contour Γ, which starts at the origin, goes to the point R along
the real axis, then to the point R eiπ/3 along a circle of radius R and then back to the origin along the ray θ = π/3.
(See Figure 15.14.)

Figure 15.14: The wedge contour.

We can evaluate the integral along Γ with the residue theorem. The integrand has one first order pole inside

612
the contour at z = eiπ/6 . For R > 1, we have
Z  
1 1 iπ/6
6
dz = i2π Res ,e
Γ z +1 z6 + 1
z − eiπ/6
= i2π lim 6
z→ eiπ/6 z + 1

Since the numerator and denominator vanish, we apply L’Hospital’s rule.

1
= i2π lim
z→ eiπ/6 6z 5
iπ −iπ5/6
= e
3
π
= e−iπ/3
3
Now we examine the integral along the circular arc, ΓR . We use the maximum modulus integral bound to show
that the value of the integral vanishes as R → ∞.
Z   
 1  πR  1 
 dz  ≤ max  

ΓR z6 + 1  3 z∈ΓR  z 6 + 1 
πR 1
=
3 R6 − 1
→ 0 as R → ∞.
Now we are prepared to evaluate the original real integral.
Z
1 π
6
dz = e−iπ/3
Γ z +1 3
Z R Z Z 0
1 1 1 π
6
dx + 6
dz + 6
dz = e−iπ/3
0 x +1 ΓR z + 1 R eiπ/3 z + 1 3
Z R Z Z 0
1 1 1 π
6
dx + 6
dz + 6
eiπ/3 dx = e−iπ/3
0 x +1 ΓR z + 1 R x +1 3

613
We take the limit as R → ∞.

Z ∞
1 π
iπ/3
dx = e−iπ/3


1− e
0 +1 x6 3
Z ∞ −iπ/3
1 π e
6
dx =
0 x +1 3 1 − eiπ/3
∞
√
π (1 − i 3)/2
Z
1
dx = √
0 x6 + 1 3 1 − (1 + i 3)/2
Z ∞
1 π
6
dx =
0 x +1 3

Solution 15.33
First note that

4 π
cos(2θ) ≥ 1 − θ, 0≤θ≤ .
π 4

These two functions are plotted in Figure 15.15. To prove this inequality analytically, note that the two functions
are equal at the endpoints of the interval and that cos(2θ) is concave downward on the interval,

d2 π
2
cos(2θ) = −4 cos(2θ) ≤ 0 for 0 ≤ θ ≤ ,
dθ 4

while 1 − 4θ/π is linear.

Let CR be the quarter circle of radius R from θ = 0 to θ = π/4. The integral along this contour vanishes as

614
 Figure 15.15: cos(2θ) and 1 − π4 θ

R → ∞.
Z  Z π/4  
−z 2  −(R eiθ )2  
  

 e dz  ≤
 e  Ri eiθ  dθ
CR 0
Z π/4
2
≤ R e−R cos(2θ) dθ
0
Z π/4
2
≤ R e−R (1−4θ/π) dθ
0
h π −R2 (1−4θ/π) iπ/4
= R 2
e
4R 0
π  2

= 1 − e−R
4R
→ 0 as R → ∞

Let C be the boundary of the domain 0 < r < R, 0 < θ < π/4. Since the integrand is analytic inside C the
integral along C is zero. Taking the limit as R → ∞, the integral from r = 0 to ∞ along θ = 0 is equal to the
integral from r = 0 to ∞ along θ = π/4.
Z ∞ Z ∞ 2
1
+i

−x2 − 1+i
√ x
e dx = e 2 √ dx
0 0 2

615
 Z ∞ Z ∞
−x2 1+i 2
e dx = √ e−ix dx
0 2 0

Z ∞ Z ∞
−x2 1+i
cos(x2 ) − i sin(x2 ) dx


e dx = √
0 2 0

Z ∞ Z ∞ Z ∞  Z ∞ Z ∞ 
−x2 1 2 2 i 2 2
e dx = √ cos(x ) dx + sin(x ) dx + √ cos(x ) dx − sin(x ) dx
0 2 0 0 2 0 0

We equate the imaginary part of this equation to see that the integrals of cos(x2 ) and sin(x2 ) are equal.
Z ∞ Z ∞
2
cos(x ) dx = sin(x2 ) dx
0 0

The real part of the equation then gives us the desired identity.
Z ∞ Z ∞ Z ∞
1 2
2
cos(x ) dx = 2
sin(x ) dx = √ e−x dx
0 0 2 0

Solution 15.34
Consider the box contour C that is the boundary of the rectangle −R ≤ x ≤ R, 0 ≤ y ≤ π. There is a removable
singularity at z = 0 and a first order pole at z = iπ. By the residue theorem,
Z
z  z 
− dz = iπ Res , iπ
C sinh z sinh z
z(z − iπ)
= iπ lim
z→iπ sinh z
2z − iπ
= iπ lim
z→iπ cosh z
2
=π

616
The integrals along the side of the box vanish as R → ∞.
Z ±R+iπ 
 z   z 
dz  ≤ π max
   
sinh z z∈[±R,±R+iπ] sinh z
  
±R
R+π
≤π
sinh R
→ 0 as R → ∞

The value of the integrand on the top of the box is

x + iπ x + iπ
=− .
sinh(x + iπ) sinh x
Taking the limit as R → ∞,
Z ∞ Z ∞
x x + iπ
dx + − − dx = π 2 .
−∞ sinh x −∞ sinh x
Note that
Z ∞
1
− dx = 0
−∞ sinh x
as there is a first order pole at x = 0 and the integrand is odd.
Z ∞
x π2
dx =
−∞ sinh x 2

Solution 15.35
First we evaluate
∞
eax
Z
dx.
−∞ ex + 1

617
Consider the rectangular contour in the positive direction with corners at ±R and ±R + i2π. With the maximum
modulus integral bound we see that the integrals on the vertical sides of the contour vanish as R → ∞.

Z R+i2π az
 aR
 e  ≤ 2π e

 dz → 0 as R → ∞

R ez + 1  eR − 1
Z −R az
 −aR
 e  ≤ 2π e

 dz → 0 as R → ∞

−R+i2πez + 1  1 − e−R

In the limit as R tends to infinity, the integral on the rectangular contour is the sum of the integrals along the
top and bottom sides.

Z ∞ Z −∞ a(x+i2π)
eaz eax
Z
e
z
dz = x
dx + dx
C e +1 −∞ e + 1 ∞ ex+i2π + 1
Z ∞
eaz eax
Z
−i2aπ
z
dz = (1 − e ) x
dx
C e +1 −∞ e + 1

The only singularity of the integrand inside the contour is a first order pole at z = iπ. We use the residue theorem
to evaluate the integral.

eaz eaz
Z  
dz = i2π Res , iπ
C ez + 1 ez + 1
(z − iπ) eaz
= i2π lim
z→iπ ez + 1
a(z − iπ) eaz + eaz
= i2π lim
z→iπ ez
= −i2π eiaπ

618
We equate the two results for the value of the contour integral.
Z ∞
−i2aπ eax
(1 − e ) x
dx = −i2π eiaπ
−∞ e + 1
Z ∞ ax
e i2π
dx = iaπ
x
−∞ e + 1 e − e−iaπ
Z ∞
eax π
x
dx =
−∞ e + 1 sin(πa)
Now we derive the value of,
Z ∞
cosh(bx)
dx.
−∞ cosh x
First make the change of variables x → 2x in the previous result.
Z ∞
e2ax π
2x + 1
2 dx =
−∞ e sin(πa)
Z ∞ (2a−1)x
e π
x −x
dx =
−∞ e + e sin(πa)
Now we set b = 2a − 1.
∞
ebx
Z
π π
dx = = for − 1 < b < 1
−∞ cosh x sin(π(b + 1)/2) cos(πb/2)
Since the cosine is an even function, we also have,
Z ∞ −bx
e π
dx = for − 1 < b < 1
−∞ cosh x cos(πb/2)
Adding these two equations and dividing by 2 yields the desired result.
Z ∞
cosh(bx) π
dx = for − 1 < b < 1
−∞ cosh x cos(πb/2)

619
Solution 15.36
Real-Valued Parameters. For b = 0, the integral has the value: π/a2 . If b is nonzero, then we can write the
integral as

1 π
Z
dθ
F (a, b) = 2 .
b 0 (a/b + cos θ)2

We define the new parameter c = a/b and the function,

Z π
2 dθ
G(c) = b F (a, b) = .
0 (c + cos θ)2

If −1 ≤ c ≤ 1 then the integrand has a double pole on the path of integration. The integral diverges. Otherwise
the integral exists. To evaluate the integral, we extend the range of integration to (0..2π) and make the change
of variables, z = eiθ to integrate along the unit circle in the complex plane.
Z 2π
1 dθ
G(c) =
2 0 (c + cos θ)2

For this change of variables, we have,

z + z −1 dz
cos θ = , dθ = .
2 iz

Z
1 dz/(iz)
G(c) = −1 2
2 C (c + (z + z )/2)
Z
z
= −i2 2 2
dz
C (2cz + z + 1)
Z
z
= −i2 √ √ dz
C (z + c + c − 1) (z + c − c2 − 1)2
2 2

620
 √ √
If c > 1, then −c − c2 − 1 is outside the unit circle and −c + c2 − 1 is inside the unit circle. The integrand
has a second order pole inside the path of integration. We evaluate the integral with the residue theorem.

√
 
z
G(c) = −i2i2π Res √ √ , z = −c + c2 − 1
(z + c + c2 − 1)2 (z + c − c2 − 1)2
d z
= 4π lim
√ √
z→−c+ c2 −1 dz (z + c + c2 − 1)2
 
1 2z
= 4π lim
√ √ − √
z→−c+ c2 −1 (z + c + c2 − 1)2 (z + c + c2 − 1)3
√
c + c2 − 1 − z
= 4π lim
√ √
z→−c+ c2 −1 (z + c + c2 − 1)3
2c
= 4π √
(2 c2 − 1)3
πc
=p
(c − 1)3
2

621
 √ √
If c < 1, then −c − c2 − 1 is inside the unit circle and −c + c2 − 1 is outside the unit circle.

√
 
z
G(c) = −i2i2π Res √ √ , z = −c − c2 − 1
(z + c + c2 − 1)2 (z + c − c2 − 1)2
d z
= 4π lim
√ √
z→−c− c2 −1 dz (z + c − c2 − 1)2
 
1 2z
= 4π lim
√ √ − √
z→−c− c2 −1 (z + c − c2 − 1)2 (z + c − c2 − 1)3
√
c − c2 − 1 − z
= 4π lim
√ √
z→−c− c2 −1 (z + c − c2 − 1)3
2c
= 4π √
(−2 c2 − 1)3
πc
= −p
(c − 1)3
2

Thus we see that


= √ πc for c > 1,
(c2 −1)3



G(c) = − √ πc for c < 1,
 (c2 −1)3

for − 1 ≤ c ≤ 1.

is divergent

In terms of F (a, b), this is


= √ aπ for a/b > 1,
(a2 −b2 )3



F (a, b) = − √ aπ for a/b < 1,
 (a2 −b2 )3

for − 1 ≤ a/b ≤ 1.

is divergent

622
 Complex-Valued Parameters. Consider
Z π
dθ
G(c) = ,
0 (c + cos θ)2

for complex c. Except for real-valued c between −1 and 1, the integral converges uniformly. We can interchange
differentiation and integration. The derivative of G(c) is

d π
Z
0 dθ
G (c) =
dc (c + cos θ)2
Z π 0
−2
= 3
dθ
0 (c + cos θ)

Thus we see that G(c) is analytic in the complex plane with a cut on the real axis from −1 to 1. The value of
the function on the positive real axis for c > 1 is
πc
G(c) = p .
(c2 − 1)3

We use analytic continuation to determine G(c) for complex c. By inspection we see that G(c) is the branch of
πc
,
(c2 − 1)3/2

with a branch cut on the real axis from −1 to 1 and which is real-valued and positive for real c > 1. Using
F (a, b) = G(c)/b2 we can determine F for complex-valued a and b.

Solution 15.37
First note that
∞ ∞
eix
Z Z
cos x
dx = dx
−∞ e + e−x
x
−∞ ex + e−x

623
since sin x/( ex + e−x ) is an odd function. For the function
eiz
f (z) =
ez + e−z
we have
eix−π −π eix
f (x + iπ) = x+iπ −x−iπ
= − e x −x
= − e−π f (x).
e +e e +e
Thus we consider the integral
eiz
Z
dz
C ez + e−z
where C is the box contour with corners at ±R and ±R + iπ. We can evaluate this integral with the residue
theorem. We can write the integrand as
eiz
.
2 cosh z
We see that the integrand has first order poles at z = iπ(n + 1/2). The only pole inside the path of integration
is at z = iπ/2.
eiz eiz
Z  
iπ
−z
dz = i2π Res ,z =
z
C e + e ez + e−z 2
iz
(z − iπ/2) e
= i2π lim
z→iπ/2 ez + e−z
eiz + i(z − iπ/2) eiz
= i2π lim
z→iπ/2 ez − e−z
e−π/2
= i2π iπ/2
e − e−iπ/2
= π e−π/2

624
The integrals along the vertical sides of the box vanish as R → ∞.
Z ±R+iπ iz
  iz

 e 
≤π
 e 
 dz max  

±R ez + e−z  z∈[±R...±R+iπ] ez + e−z 

 
 1 
≤ π max  R+iy 
y∈[0...π] e + e−R−iy 
 
 1 
≤ π max  R 
y∈[0...π] e + e−R−i2y 
1
=π
2 sinh R
→ 0 as R → ∞
Taking the limit as R → ∞, we have
Z ∞ Z −∞+iπ
eix eiz
dx + dz = π e−π/2
−∞ ex + e−x ∞+iπ e z + e−z
Z ∞
eix
(1 + e−π ) x + e−x
dx = π e−π/2
−∞ e
Z ∞
eix π
−x
dx =
x
−∞ e + e eπ/2 + e−π/2
Finally we have,
Z ∞
cos x π
−x
dx = π/2 .
−∞ ex+e e + e−π/2

Definite Integrals Involving Sine and Cosine

Solution 15.38
1. Let C be the positively oriented unit circle about the origin. We parametrize this contour.
z = eiθ , dz = i eiθ dθ, θ ∈ (0 . . . 2π)

625
 We write sin θ and the differential dθ in terms of z. Then we evaluate the integral with the Residue theorem.
Z 2π I
1 1 dz
dθ =
0 2 + sin θ 2 + (z − 1/z)/(2i) iz
IC
2
= 2
dz
C z + i4z − 1
I
2
= √

√

dz
C z +i 2+ 3 z+i 2− 3
  √    √   √ 
= i2π Res z+i 2+ 3 z + i 2 − 3 , z = i −2 + 3
2
= i2π √
i2 3
2π
=√
3

2. First consider the case a = 0.

(
π
for n ∈ Z+
Z
0
cos(nθ) dθ =
−π 2π for n = 0

Now we consider |a| < 1, a 6= 0. Since

sin(nθ)
1 − 2a cos θ + a2
is an even function,
π π
einθ
Z Z
cos(nθ)
dθ = dθ
−π 1 − 2a cos θ + a2 −π 1 − 2a cos θ + a2
Let C be the positively oriented unit circle about the origin. We parametrize this contour.
z = eiθ , dz = i eiθ dθ, θ ∈ (−π . . . π)

626
 We write the integrand and the differential dθ in terms of z. Then we evaluate the integral with the Residue
theorem.
Z π
einθ zn
I
dz
2
dθ = 2
−π 1 − 2a cos θ + a C 1 − a(z + 1/z) + a iz
zn
I
= −i dz
−az 2 + (1 + a2 )z − a
IC
i zn
= dz
a C z 2 − (a + 1/a)z + 1
zn
I
i
= dz
a C (z − a)(z − 1/a)
zn
 
i
= i2π Res ,z = a
a (z − a)(z − 1/a)
2π an
=−
a a − 1/a
2πan
=
1 − a2

We write the value of the integral for |a| < 1 and n ∈ Z0+ .
(
Z π
cos(nθ) 2π for a = 0, n = 0
dθ = 2πan
−π 1 − 2a cos θ + a2 1−a2
otherwise

Solution 15.39
Convergence. We consider the integral
Z π
cos(nθ) sin(nα)
I(α) = − dθ = π .
0 cos θ − cos α sin α

627
We assume that α is real-valued. If α is an integer, then the integrand has a second order pole on the path of
integration, the principal value of the integral does not exist. If α is real, but not an integer, then the integrand
has a first order pole on the path of integration. The integral diverges, but its principal value exists.

Contour Integration. We will evaluate the integral for real, non-integer α.

Z π
cos(nθ)
I(α) = − dθ
0 cos θ − cos α
1 2π cos(nθ)
Z
= − dθ
2 0 cos θ − cos α
Z 2π
1 einθ
= <− dθ
2 0 cos θ − cos α

We make the change of variables: z = eiθ .

zn
Z
1 dz
I(α) = < −
2 (z + 1/z)/2 − cos α iz
Z C
−iz n
=<− iα −iα )
dz
C (z − e )(z − e

628
Now we use the residue theorem.

zn
   
iα
= < iπ(−i) Res ,z = e
(z − eiα )(z − e−iα )
zn
  
−iα
+ Res ,z = e
(z − eiα )(z − e−iα )
zn zn
 
= π< limiα + lim
z→ e z − e−iα z→ e−iα z − eiα

einα e−inα
 
= π< + −iα
eiα − e−iα e − eiα
e − e−inα
 inα 
= π<
eiα − e−iα
 
sin(nα)
= π<
sin(α)

Z π
cos(nθ) sin(nα)
I(α) = − dθ = π .
0 cos θ − cos α sin α

Solution 15.40
Consider the integral
1
x2
Z
√ dx.
0 (1 + x2 ) 1 − x2
We make the change of variables x = sin ξ to obtain,
Z π/2
sin2 ξ
p cos ξ dξ
0 (1 + sin2 ξ) 1 − sin2 ξ

629
 π/2
sin2 ξ
Z
dξ
0 1 + sin2 ξ

π/2
1 − cos(2ξ)
Z
dξ
0 3 − cos(2ξ)

2π
1 − cos ξ
Z
1
dξ
4 0 3 − cos ξ

Now we make the change of variables z = eiξ to obtain a contour integral on the unit circle.
 
1 − (z + 1/z)/2 −i
Z
1
dz
4 C 3 − (z + 1/z)/2 z

−i (z − 1)2
Z
√ √ dz
4 C z(z − 3 + 2 2)(z − 3 − 2 2)
There are two first order poles inside the contour. The value of the integral is
(z − 1)2 (z − 1)2 √
    
−i
i2π Res √ √ , 0 + Res √ √ ,z = 3 − 2 2
4 z(z − 3 + 2 2)(z − 3 − 2 2) z(z − 3 + 2 2)(z − 3 − 2 2)

(z − 1)2 (z − 1)2
    
π
lim √ √ + lim√ √ .
2 z→0 (z − 3 + 2 2)(z − 3 − 2 2) z→3−2 2 z(z − 3 − 2 2)

1
√
x2 (2 − 2)π
Z
√ dx =
0 (1 + x2 ) 1 − x2 4

630
Infinite Sums
Solution 15.41
From Result 15.10.1 we see that the sum of the residues of π cot(πz)/z 4 is zero. This function has simples poles
at nonzero integers z = n with residue 1/n4 . There is a fifth order pole at z = 0. Finding the residue with the
formula
1 d4
lim 4 (πz cot(πz))
4! z→0 dz
would be a real pain. After doing the differentiation, we would have to apply L’Hospital’s rule multiple times. A
better way of finding the residue is with the Laurent series expansion of the function. Note that
1 1
=
sin(πz) πz − (πz)3 /6 + (πz)5 /120 − · · ·
1 1
=
πz 1 − (πz) /6 + (πz)4 /120 − · · ·
2
 2 2 !
π4 4
  2
1 π 2 π 2 π4 4
= 1+ z − z + ··· + z − z + ··· + ··· .
πz 6 120 6 120
Now we find the z −1 term in the Laurent series expansion of π cot(πz)/z 4 .
2 !
π2 2 π4 4
 2
π4 4
  2
π4 4
 
π cos(πz) π 1 π 2 π 2
= 4 1 − z + z − ··· 1+ z − z + ··· + z − z + ··· + ···
z 4 sin(πz) z 2 24 πz 6 120 6 120
π4 π4 π4 π4
   
1 4
= 5 ··· + − + − + z + ···
z 120 36 12 24
π4 1
= ··· − + ···
45 z
Thus the residue at z = 0 is −π 4 /45. Summing the residues,
−1 ∞
X 1 π4 X 1
− + = 0.
n=−∞
n4 45 n=1 n4

631
 ∞
X 1 π4
=
n=1
n4 90

Solution 15.42
For this problem we will use the following result: If
lim |zf (z)| = 0,
|z|→∞

then the sum of all the residues of π cot(πz)f (z) is zero. If in addition, f (z) is analytic at z = n ∈ Z then
∞
X
f (n) = −( sum of the residues of π cot(πz)f (z) at the poles of f (z) ).
n=−∞

We assume that α is not an integer, otherwise the sum is not defined. Consider f (z) = 1/(z 2 − α2 ). Since
 
 1 
lim z 2  = 0,
|z|→∞ z − α2 
and f (z) is analytic at z = n, n ∈ Z, we have
∞
X 1
= −( sum of the residues of π cot(πz)f (z) at the poles of f (z) ).
n=−∞
n2 − α2

f (z) has first order poles at z = ±α.

∞    
X 1 π cot(πz) π cot(πz)
= − Res , z = α − Res , z = −α
n=−∞
n − α2
2 z 2 − α2 z 2 − α2
π cot(πz) π cot(πz)
= − lim − lim
z→α z+α z→−α z−α
π cot(πα) π cot(−πα)
=− −
2α −2α

632
∞
X 1 π cot(πα)
=−
n=−∞
n2 −α 2 α

633
 Part IV

Ordinary Differential Equations

634
Chapter 16

First Order Differential Equations

Don’t show me your technique. Show me your heart.

-Tetsuyasu Uekuma

16.1 Notation
A differential equation is an equation involving a function, it’s derivatives, and independent variables. If there is
only one independent variable, then it is an ordinary differential equation. Identities such as

d dy dx
f 2 (x) = 2f (x)f 0 (x),


and =1
dx dx dy

are not differential equations.

The order of a differential equation is the order of the highest derivative. The following equations are first,
second and third order, respectively.

• y 0 = xy 2

635
 • y 00 + 3xy 0 + 2y = x2

• y 000 = y 00 y
The degree of a differential equation is the highest power of the highest derivative in the equation. The
following equations are first, second and third degree, respectively.
• y 0 − 3y = sin x

• (y 00 )2 + 2xy = ex

• (y 0 )3 + y 5 = 0
An equation is said to be linear if it is linear in the dependent variable.
• y 00 cos x + x2 y = 0 is a linear differential equation.

• y 0 + xy 2 = 0 is a nonlinear differential equation.

A differential equation is homogeneous if it has no terms that are functions of the independent variable alone.
Thus an inhomogeneous equation is one in which there are terms that are functions of the independent variables
alone.
• y 00 + xy + y = 0 is a homogeneous equation.

• y 0 + y + x2 = 0 is an inhomogeneous equation.
A first order differential equation may be written in terms of differentials. Recall that for the function y(x)
the differential dy is defined dy = y 0 (x) dx. Thus the differential equations

y 0 = x2 y and y 0 + xy 2 = sin(x)

can be denoted:

dy = x2 y dx and dy + xy 2 dx = sin(x) dx.

636
 A solution of a differential equation is a function which when substituted into the equation yields an identity.
For example, y = x log x is a solution of
y
y0 − =1
x
and y = c ex is a solution of

y 00 − y = 0

for any value of the parameter c.

16.2 One Parameter Families of Functions

Consider the equation

F (x, y(x); c) = 0, (16.1)

which implicitly defines a one-parameter family of functions y(x). (We assume that F has a non-trivial dependence
on y, that is Fy 6= 0.) Differentiating this equation with respect to x yields

Fx + Fy y 0 = 0.

This gives us two equations involving the independent variable x, the dependent variable y(x) and its derivative
and the parameter c. If we algebraically eliminate c between the two equations, the eliminant will be a first order
differential equation for y(x). Thus we see that every equation of the form (16.1) defines a one-parameter family
of functions y(x) which satisfy a first order differential equation. This y(x) is the primitive of the differential
equation. Later we will discuss why y(x) is the general solution of the differential equation.

Example 16.2.1 Consider the family of circles of radius c centered about the origin,

x2 + y 2 = c2 .

637
Differentiating this yields,

2x + 2yy 0 = 0.

It is trivial to eliminate the parameter and obtain a differential equation for the family of circles.

x + yy 0 = 0.

We can see the geometric meaning in this equation by writing it in the form
x
y0 = − .
y

The slope of the tangent to a circle at a point is the negative of the cotangent of the angle.

Example 16.2.2 Consider the one-parameter family of functions,

y(x) = f (x) + cg(x),

where f (x) and g(x) are known functions. The derivative is

y 0 = f 0 + cg 0 .

Eliminating the parameter yields

gy 0 − g 0 y = gf 0 − g 0 f
g0 g0f
y0 − y = f 0 − .
g g

Thus we see that y(x) = f (x) + cg(x) satisfies a first order linear differential equation.

638
 We know that every one-parameter family of functions satisfies a first order differential equation. The converse
is true as well.
Result 16.2.1 Every first order differential equation has a one-parameter family of so-
lutions, y(x), defined by an equation of the form:

F (x, y(x); c) = 0.

This y(x) is called the general solution. If the equation is linear then the general solution
expresses the totality of solutions of the differential equation. If the equation is nonlinear,
there may be other special singular solutions, which do not depend on a parameter.

This is strictly an existence result. It does not say that the general solution of a first order differential equation
can be determined by some method, it just says that it exists. There is no method for solving the general first
order differential equation. However, there are some special forms that are soluble. We will devote the rest of
this chapter to studying these forms.

16.3 Exact Equations

Any first order ordinary differential equation of the first degree can be written as the total differential equation,

P (x, y) dx + Q(x, y) dy = 0.

If this equation can be integrated directly, that is if there is a primitive, u(x, y), such that

du = P dx + Q dy,

then this equation is called exact. The (implicit) solution of the differential equation is

u(x, y) = c,

639
where c is an arbitrary constant. Since the differential of a function, u(x, y), is
∂u ∂u
du ≡ dx + dy,
∂x ∂y
P and Q are the partial derivatives of u:
∂u ∂u
P (x, y) = , Q(x, y) = .
∂x ∂y

In an alternate notation, the differential equation

dy
P (x, y) + Q(x, y) = 0, (16.2)
dx
is exact if there is a primitive u(x, y) such that
du ∂u ∂u dy dy
≡ + = P (x, y) + Q(x, y) .
dx ∂x ∂y dx dx
The solution of the differential equation is u(x, y) = c.

Example 16.3.1
dy
x+y =0
dx
is an exact differential equation since
 
d 1 2 2 dy
(x + y ) = x + y
dx 2 dx
The solution of the differential equation is
1 2
(x + y 2 ) = c.
2

640
Example 16.3.2 , Let f (x) and g(x) be known functions.
g(x)y 0 + g 0 (x)y = f (x)
is an exact differential equation since
d
(g(x)y(x)) = gy 0 + g 0 y.
dx
The solution of the differential equation is
Z
g(x)y(x) = f (x) dx + c
Z
1 c
y(x) = f (x) dx + .
g(x) g(x)

A necessary condition for exactness. The solution of the exact equation P + Qy 0 = 0 is u = c where u is
the primitive of the equation, dudx
= P + Qy 0 . At present the only method we have for determining the primitive
is guessing. This is fine for simple equations, but for more difficult cases we would like a method more concrete
than divine inspiration. As a first step toward this goal we determine a criterion for determining if an equation
is exact.
Consider the exact equation,
P + Qy 0 = 0,
with primitive u, where we assume that the functions P and Q are continuously differentiable. Since the mixed
partial derivatives of u are equal,
∂2u ∂2u
= ,
∂x∂y ∂y∂x
a necessary condition for exactness is
∂P ∂Q
= .
∂y ∂x

641
A sufficient condition for exactness. This necessary condition for exactness is also a sufficient condition.
We demonstrate this by deriving the general solution of (16.2). Assume that P + Qy 0 = 0 is not necessarily exact,
but satisfies the condition Py = Qx . If the equation has a primitive,
du ∂u ∂u dy dy
≡ + = P (x, y) + Q(x, y) ,
dx ∂x ∂y dx dx
then it satisfies
∂u ∂u
= P, = Q. (16.3)
∂x ∂y
Integrating the first equation of (16.3), we see that the primitive has the form
Z x
u(x, y) = P (ξ, y) dξ + f (y),
x0

for some f (y). Now we substitute this form into the second equation of (16.3).
∂u
= Q(x, y)
∂y
Z x
Py (ξ, y) dξ + f 0 (y) = Q(x, y)
x0

Now we use the condition Py = Qx .

Z x
Qx (ξ, y) dξ + f 0 (y) = Q(x, y)
x0
Q(x, y) − Q(x0 , y) + f 0 (y) = Q(x, y)
f 0 (y) = Q(x0 , y)
Z y
f (y) = Q(x0 , η) dη
y0

642
Thus we see that
Z x Z y
u= P (ξ, y) dξ + Q(x0 , η) dη
x0 y0

is a primitive of the derivative; the equation is exact. The solution of the differential equation is
Z x Z y
P (ξ, y) dξ + Q(x0 , η) dη = c.
x0 y0

Even though there are three arbitrary constants: x0 , y0 and c, the solution is a one-parameter family. This is
because changing x0 or y0 only changes the left side by an additive constant.
Result 16.3.1 Any first order differential equation of the first degree can be written in
the form
dy
P (x, y) + Q(x, y) = 0.
dx
This equation is exact if and only if

Py = Q x .

In this case the solution of the differential equation is given by

Z x Z y
P (ξ, y) dξ + Q(x0 , η) dη = c.
x0 y0

16.3.1 Separable Equations

Any differential equation that can written in the form
P (x) + Q(y)y 0 = 0

643
is a separable equation, (because the dependent and independent variables are separated). We can obtain an
implicit solution by integrating with respect to x.
Z Z
dy
P (x) dx + Q(y) dx = c
dx
Z Z
P (x) dx + Q(y) dy = c

Result 16.3.2 The general solution to the separable equation P (x) + Q(y)y 0 = 0 is
Z Z
P (x) dx + Q(y) dy = c

Example 16.3.3 Consider the equation y 0 = xy 2 .

dy
= xy 2
dx
y −2 dy = x dx
Z Z
−2
y dy = x dx + c
1
−y −1 = x2 + c
2
−1
y= 1 2
2
x +c

Example 16.3.4 The equation

y0 = y − y2,

644
is separable.

y0
=1
y − y2

We expand in partial fractions and integrate.

 
1 1
− y0 = 1
y y−1
log(y) − log(y − 1) = x + c

Then we solve for y(x).

 
y
log =x+c
y−1
y
= ex+c
y−1
ex+c
y = x+c
e −1

Finally we substitute a = e−c to write the solution in a nice form.

1
y=
1 − a e−x

16.3.2 Homogeneous Coefficient Equations

Euler’s Theorem on Homogeneous Functions. The function F (x, y) is homogeneous of degree n if

F (λx, λy) = λn F (x, y).

645
From this definition we see that
 y
n
F (x, y) = x F 1, .
x
(Just formally substitute 1/x for λ.) For example,

x2 y + 2y 3
xy 2 , , x cos(y/x)
x+y

are homogeneous functions of orders 3, 2 and 1, respectively.

Euler’s theorem for a homogeneous function of order n is:

xFx + yFy = nF.

To prove this, we define ξ = λx, η = λy. From the definition of homogeneous functions, we have

F (ξ, η) = λn F (x, y).

We differentiate this equation with respect to λ.

∂F (ξ, η) ∂ξ ∂F (ξ, η) ∂η
+ = nλn−1 F (x, y)
∂ξ ∂λ ∂η ∂λ
xFξ + yFη = nλn−1 F (x, y)

Setting λ = 1, (and hence ξ = x, η = y), proves Euler’s theorem.

Result 16.3.3 Euler’s Theorem. If F (x, y) is a homogeneous function of degree n,

then

xFx + yFy = nF.

646
Homogeneous Coefficient Differential Equations. If the coefficient functions P (x, y) and Q(x, y) are ho-
mogeneous of degree n then the differential equation,
dy
Q(x, y) + P (x, y) = 0,
dx
is called a homogeneous coefficient equation. They are often referred to as simply homogeneous equations. We can
write the equation in the form,
 y  y  dy
xn Q 1, + xn P 1, = 0,
x x dx
 y  y  dy
Q 1, + P 1, = 0.
x x dx
y(x)
This suggests the change of dependent variable u(x) = x
.
 
du
Q(1, u) + P (1, u) u + x =0
dx
This equation is separable.
du
Q(1, u) + uP (1, u) + xP (1, u) =0
dx
1 P (1, u) du
+ =0
x Q(1, u) + uP (1, u) dx
Z
1
log x + du = c
u + Q(1, u)/P (1, u)

By substituting log c for c, we can write this in the form,

Z
1 c
du = log .
u + Q(1, u)/P (1, u) x

647
Example 16.3.5 Consider the homogeneous coefficient equation
dy
x2 − y 2 + xy = 0.
dx
The solution for u(x) = y(x)/x is determined by
Z
1 c
2 du = log
u + 1−u
u
x
Z c
u du = log
x
1 2 c
u = log
2 p x
u = ± 2 log(c/x)
Thus the solution of the differential equation is
p
y = ±x 2 log(c/x)

Result 16.3.4 Homogeneous Coefficient Differential Equations. If P (x, y) and

Q(x, y) are homogeneous functions of degree n, then the equation
dy
P (x, y) + Q(x, y) =0
dx
y(x)
is made separable by the change of independent variable u(x) = x . The solution is
determined by
Z
1 c
du = log .
u + Q(1, u)/P (1, u) x

648
16.4 The First Order, Linear Differential Equation
16.4.1 Homogeneous Equations
The first order, linear, homogeneous equation has the form
dy
+ p(x)y = 0.
dx
Note that this equation is separable.
y0
= −p(x)
y
Z
log(y) = − p(x) dx + a
R
y = e− p(x) dx+a
R
y = c e− p(x) dx

Example 16.4.1 Consider the equation

dy 1
+ y = 0.
dx x
R
y(x) = c e− 1/x dx

y(x) = c e− log x

c
y(x) =
x

649
16.4.2 Inhomogeneous Equations
The first order, linear, inhomogeneous differential equation has the form
dy
+ p(x)y = f (x). (16.4)
dx
This equation is not separable. Note that it is similar to the exact equation we solved in Example 16.3.2,
g(x)y 0 (x) + g 0 (x)y(x) = f (x).
To solve Equation 16.4, we multiply by an integrating factor. Multiplying a differential equation by its integrating
factor changes it to an exact equation. Multiplying Equation 16.4 by the function, I(x), yields,
dy
I(x) + p(x)I(x)y = f (x)I(x).
dx
In order that I(x) be an integrating factor, it must satisfy
d
I(x) = p(x)I(x).
dx
This is a first order, linear, homogeneous equation with the solution
R
p(x) dx
I(x) = c e .
This is an integrating factor for any constant c. For simplicity we will choose c = 1.
R
To solve Equation 16.4 we multiply by the integrating factor and integrate. Let P (x) = p(x) dx.
dy
eP (x) + p(x) eP (x) y = eP (x) f (x)
dx
d
eP (x) y = eP (x) f (x)


dx Z
y = e−P (x) eP (x) f (x) dx + c e−P (x)

y ≡ yp + c yh

650
Note that the general solution is the sum of a particular solution, yp , that satisfies y 0 + p(x)y = f (x), and an
arbitrary constant times a homogeneous solution, yh , that satisfies y 0 + p(x)y = 0.

Example 16.4.2 Consider the differential equation

1
y 0 + y = x2 .
x
The integrating factor is
Z 
1
I(x) = exp dx = elog x = x.
x
Multiplying by the integrating factor and integrating,
d
(xy) = x3
dx
1
xy = x4 + c
4
1 c
y = x3 + .
4 x
We see that the particular and homogeneous solutions are
1 1
y p = x3 and yh = .
4 x
Note that the general solution to the differential equation is a one-parameter family of functions. The general
solution is plotted in Figure 16.1 for various values of c.

16.4.3 Variation of Parameters.

We could also have found the particular solution with the method of variation of parameters. Although we
can solve first order equations without this method, it will become important in the study of higher order

651
 10

7.5

2.5

-1 -0.5 0.5 1
-2.5

-5

-7.5

-10

Figure 16.1: Solutions to y 0 + y/x = x2 .

inhomogeneous equations. We begin by assuming that the particular solution has the form yp = u(x)yh (x) where
u(x) is an unknown function. We substitute this into the differential equation.

d
yp + p(x)yp = f (x)
dx
d
(uyh ) + p(x)uyh = f (x)
dx
u0 yh + u(yh0 + p(x)yh ) = f (x)

652
Since yh is a homogeneous solution, yh0 + p(x)yh = 0.

f (x)
u0 =
yh
Z
f (x)
u= dx
yh (x)

Recall that the homogeneous solution is yh = e−P (x) .

Z
u= eP (x) f (x) dx

Thus the particular solution is

Z
−P (x)
yp = e eP (x) f (x) dx.

16.5 Initial Conditions

In physical problems involving first order differential equations, the solution satisfies both the differential
equation and a constraint which we call the initial condition. Consider a first order linear differential equation
subject to the initial condition y(x0 ) = y0 . The general solution is
Z
−P (x)
y = yp + cyh = e eP (x) f (x) dx + c e−P (x) .

For the moment, we will assume that this problem is well-posed. A problemR is well-posed if there is a unique
solution to the differential equation that satisfies
R x P (ξ)the constraint(s). Recall that eP (x) f (x) dx denotes any integral
P (x)
of e f (x). For convenience, we choose x0 e f (ξ) dξ. The initial condition requires that
Z x0
−P (x0 )
y(x0 ) = y0 = e eP (ξ) f (ξ) dξ + c e−P (x0 ) = c e−P (x0 ) .
x0

653
Thus c = y0 eP (x0 ) . The solution subject to the initial condition is
Z x
−P (x)
y= e eP (ξ) f (ξ) dξ + y0 eP (x0 )−P (x) .
x0

Example 16.5.1 Consider the problem

y 0 + (cos x)y = x, y(0) = 2.
From Result 16.5.1, the solution subject to the initial condition is
Z x
− sin x
y= e ξ esin ξ dξ + 2 e− sin x .
0

16.5.1 Piecewise Continuous Coefficients and Inhomogeneities

If the coefficient function p(x) and the inhomogeneous term f (x) in the first order linear differential equation
dy
+ p(x)y = f (x)
dx
are continuous, then the solution is continuous and has a continuous first derivative. To see this, we note that
the solution
Z
−P (x)
y= e eP (x) f (x) dx + c e−P (x)

is continuous since the integral of a piecewise continuous function is continuous. The first derivative of the solution
can be found directly from the differential equation.
y 0 = −p(x)y + f (x)
Since p(x), y, and f (x) are continuous, y 0 is continuous.
If p(x) or f (x) is only piecewise continuous, then the solution will be continuous since the integral of a piecewise
continuous function is continuous. The first derivative of the solution will be piecewise continuous.

654
Example 16.5.2 Consider the problem

y 0 − y = H(x − 1), y(0) = 1,

where H(x) is the Heaviside function.

(
1 for x > 0,
H(x) =
0 for x < 0.

To solve this problem, we divide it into two equations on separate domains.

y10 − y1 = 0, y1 (0) = 1, for x < 1

y20 − y2 = 1, y2 (1) = y1 (1), for x > 1

With the condition y2 (1) = y1 (1) on the second equation, we demand that the solution be continuous. The
solution to the first equation is y = ex . The solution for the second equation is
Z x
y= e x
e−ξ dξ + e1 ex−1 = −1 + ex−1 + ex .
1

Thus the solution over the whole domain is

(
ex for x < 1,
y=
(1 + e−1 ) ex − 1 for x > 1.

The solution is graphed in Figure 16.2.

Example 16.5.3 Consider the problem,

y 0 + sign (x)y = 0, y(1) = 1.

655
 8

-1 -0.5 0.5 1 1.5 2

Figure 16.2: Solution to y 0 − y = H(x − 1).

Recall that

−1
 for x < 0
sign x = 0 for x = 0

1 for x > 0.


Since sign x is piecewise defined, we solve the two problems,

0
y+ + y+ = 0, y+ (1) = 1, for x > 0
0
y− − y− = 0, y− (0) = y+ (0), for x < 0,

656
and define the solution, y, to be

(
y+ (x), for x ≥ 0,
y(x) =
y− (x), for x ≤ 0.

The initial condition for y− demands that the solution be continuous.

Solving the two problems for positive and negative x, we obtain

(
e1−x , for x > 0,
y(x) =
e1+x , for x < 0.

This can be simplified to

y(x) = e1−|x| .

This solution is graphed in Figure 16.3.

657
 2.5

1.5

0.5

-3 -2 -1 1 2 3

Figure 16.3: Solution to y 0 + sign (x)y = 0.

Result 16.5.1 Existence, Uniqueness Theorem. Let p(x) and f (x) be piecewise
continuous on the interval [a, b] and let x0 ∈ [a, b]. Consider the problem,
dy
+ p(x)y = f (x), y(x0 ) = y0 .
dx
The general solution of the differential equation is
Z
y = e−P (x) eP (x) f (x) dx + c e−P (x) .

The unique, continuous solution of the differential equation subject to the initial condition
is 658
Z x
y = e−P (x) eP (ξ) f (ξ) dξ + y0 eP (x0 )−P (x) ,
x0
R
where P (x) = p(x) dx.
16.6 Well-Posed Problems
Example 16.6.1 Consider the problem,

1
y 0 − y = 0, y(0) = 1.
x

The general solution is y = cx. Applying the initial condition demands that 1 = c · 0, which cannot be satisfied.
The general solution for various values of c is plotted in Figure 16.4.

0.6

0.4

0.2

-1 -0.5 0.5 1

-0.2

-0.4

-0.6

Figure 16.4: Solutions to y 0 − y/x = 0.

659
Example 16.6.2 Consider the problem
1 1
y0 − y = − , y(0) = 1.
x x
The general solution is

y = 1 + cx.

The initial condition is satisfied for any value of c so there are an infinite number of solutions.

Example 16.6.3 Consider the problem

1
y 0 + y = 0, y(0) = 1.
x
The general solution is y = xc . Depending on whether c is nonzero, the solution is either singular or zero at the
origin and cannot satisfy the initial condition.

The above problems in which there were either no solutions or an infinite number of solutions are said to be
ill-posed. If there is a unique solution that satisfies the initial condition, the problem is said to be well-posed. We
should have suspected that we would run into trouble in the above examples as the initial condition was given at
a singularity of the coefficient function, p(x) = 1/x.
Consider the problem,

y 0 + p(x)y = f (x), y(x0 ) = y0 .

We assume that f (x) bounded in a neighborhood of x = x0 . The differential equation has the general solution,
Z
−P (x)
y= e eP (x) f (x) dx + c e−P (x) .

660
If the homogeneous solution, e−P (x) , is nonzero and finite at x = x0 , then there is a unique value of c for which
the initial condition is satisfied. If the homogeneous solution vanishes at x = x0 then either the initial condition
cannot be satisfied or the initial condition is satisfied for all values of c. The homogeneous solution can vanish or
be infinite only if P (x) → ±∞ as x → x0 . This can occur only if the coefficient function, p(x), is unbounded at
that point.
Result 16.6.1 If the initial condition is given where the homogeneous solution to a first
order, linear differential equation is zero or infinite then the problem may be ill-posed.
This may occur only if the coefficient function, p(x), is unbounded at that point.

16.7 Equations in the Complex Plane

16.7.1 Ordinary Points
Consider the first order homogeneous equation
dw
+ p(z)w = 0,
dz
where p(z), a function of a complex variable, is analytic in some domain D. The integrating factor,
Z 
I(z) = exp p(z) dz ,

is an analytic function in that domain. As with the case of real variables, multiplying by the integrating factor
and integrating yields the solution,
 Z 
w(z) = c exp − p(z) dz .

We see that the solution is analytic in D.

661
Example 16.7.1 It does not make sense to pose the equation
dw
+ |z|w = 0.
dz
For the solution to exist, w and hence w0 (z) must be analytic. Since p(z) = |z| is not analytic anywhere in the
complex plane, the equation has no solution.

Any point at which p(z) is analytic is called an ordinary point of the differential equation. Since the solution
is analytic we can expand it in a Taylor series about an ordinary point. The radius of convergence of the series
will be at least the distance to the nearest singularity of p(z) in the complex plane.

Example 16.7.2 Consider the equation

dw 1
− w = 0.
dz 1−z
c
The general solution is w = 1−z
. Expanding this solution about the origin,
∞
c X
w= =c zn.
1−z n=0

The radius of convergence of the series is,

 
 an 
R = lim 
  = 1,
n→∞ an+1 

1
which is the distance from the origin to the nearest singularity of p(z) = 1−z
.

We do not need to solve the differential equation to find the Taylor series expansion of the homogeneous
solution. We could substitute a general Taylor series expansion into the differential equation and solve for the
coefficients. Since we can always solve first order equations, this method is of limited usefulness. However, when
we consider higher order equations in which we cannot solve the equations exactly, this will become an important
method.

662
Example 16.7.3 Again consider the equation

dw 1
− w = 0.
dz 1−z

P∞
Since we know that the solution has a Taylor series expansion about z = 0, we substitute w = n=0 an z n into
the differential equation.

∞ ∞
d X n
X
(1 − z) an z − an z n = 0
dz n=0 n=0
∞
X ∞
X ∞
X
n−1 n
nan z − nan z − an z n = 0
n=1 n=1 n=0
∞
X ∞
X ∞
X
(n + 1)an+1 z n − nan z n − an z n = 0
n=0 n=0 n=0
∞
X
((n + 1)an+1 − (n + 1)an ) z n = 0.
n=0

Now we equate powers of z to zero. For z n , the equation is (n + 1)an+1 − (n + 1)an = 0, or an+1 = an . Thus we
have that an = a0 for all n ≥ 1. The solution is then

∞
X
w = a0 zn,
n=0

which is the result we obtained by expanding the solution in Example 16.7.2.

663
 Result 16.7.1 Consider the equation
dw
+ p(z)w = 0.
dz
If p(z) is analytic at z = z0 then z0 is called an ordinary point of the differential equation.
The Taylor series expansion of the solution can be found by substituting w = ∞
P
n=0 an (z−
n
z0 ) into the equation and equating powers of (z − z0 ). The radius of convergence of the
series is at least the distance to the nearest singularity of p(z) in the complex plane.

16.7.2 Regular Singular Points

If the coefficient function p(z) has a simple pole at z = z0 then z0 is a regular singular point of the first order
differential equation.

Example 16.7.4 Consider the equation

dw α
+ w = 0, α 6= 0.
dz z
This equation has a regular singular point at z = 0. The solution is w = cz −α . Depending on the value of α, the
solution can have three different kinds of behavior.

α is a negative integer. The solution is analytic in the finite complex plane.

α is a positive integer The solution has a pole at the origin. w is analytic in the annulus, 0 < |z|.

α is not an integer. w has a branch point at z = 0. The solution is analytic in the cut annulus 0 < |z| < ∞,
θ0 < arg z < θ0 + 2π.

664
 Consider the differential equation
dw
+ p(z)w = 0,
dz
where p(z) has a simple pole at the origin and is analytic in the annulus, 0 < |z| < r, for some positive r. Recall
that the solution is
 Z 
w = c exp − p(z) dz
 Z 
b0 b0
= c exp − + p(z) − dz
z z
 
zp(z) − b0
Z
= c exp −b0 log z − dz
z
 Z 
−b0 zp(z) − b0
= cz exp − dz
z
The exponential factor has a removable singularity at z = 0 and is analytic in |z| < r. We consider the
following cases for the z −b0 factor:
b0 is a negative integer. Since z −b0 is analytic at the origin, the solution to the differential equation is analytic
in the circle |z| < r.
b0 is a positive integer. The solution has a pole of order −b0 at the origin and is analytic in the annulus
0 < |z| < r.
b0 is not an integer. The solution has a branch point at the origin and thus is not single-valued. The solution
is analytic in the cut annulus 0 < |z| < r, θ0 < arg z < θ0 + 2π.
Since the exponential factor has a convergent Taylor series in |z| < r, the solution can be expanded in a series
of the form
X∞
−b0
w=z an z n , where a0 6= 0 and b0 = lim z p(z).
z→0
n=0

665
In the case of a regular singular point at z = z0 , the series is
∞
X
−b0
w = (z − z0 ) an (z − z0 )n , where a0 6= 0 and b0 = lim (z − z0 ) p(z).
z→z0
n=0

Series of this form are known as Frobenius series. Since we can write the solution as
 Z   
−b0 b0
w = c(z − z0 ) exp − p(z) − dz ,
z − z0

we see that the Frobenius expansion of the solution will have a radius of convergence at least the distance to the
nearest singularity of p(z).

Result 16.7.2 Consider the equation,

dw
+ p(z)w = 0,
dz
where p(z) has a simple pole at z = z0 , p(z) is analytic in some annulus, 0 < |z − z0 | < r,
and limz→z0 (z − z0 )p(z) = β. The solution to the differential equation has a Frobenius
series expansion of the form
∞
X
−β
w = (z − z0 ) an (z − z0 )n , a0 6= 0.
n=0

The radius of convergence of the expansion will be at least the distance to the nearest
singularity of p(z).

Example 16.7.5 We will find the first two nonzero terms in the series solution about z = 0 of the differential

666
equation,
dw 1
+ w = 0.
dz sin z
First we note that the coefficient function has a simple pole at z = 0 and
z 1
lim = lim = 1.
z→0 sin z z→0 cos z

Thus we look for a series solution of the form

X∞
−1
w=z an z n , a0 6= 0.
n=0

The nearest singularities of 1/ sin z in the complex plane are at z = ±π. Thus the radius of convergence of the
series will be at least π.
Substituting the first three terms of the expansion into the differential equation,
d 1
(a0 z −1 + a1 + a2 z) + (a0 z −1 + a1 + a2 z) = O(z).
dz sin z
Recall that the Taylor expansion of sin z is sin z = z − 61 z 3 + O(z 5 ).

z3
 
z− + O(z ) (−a0 z −2 + a2 ) + (a0 z −1 + a1 + a2 z) = O(z 2 )
5
6
 a0 
−a0 z −1 + a2 + z + a0 z −1 + a1 + a2 z = O(z 2 )
6
 a0 
a1 + 2a2 + z = O(z 2 )
6
a0 is arbitrary. Equating powers of z,
z0 : a1 = 0.
a0
z1 : 2a2 + = 0.
6

667
Thus the solution has the expansion,
 z
w = a0 z −1 − + O(z 2 ).
12
In Figure 16.5 the exact solution is plotted in a solid line and the two term approximation is plotted in a dashed
line. The two term approximation is very good near the point x = 0.

1 2 3 4 5 6

-2

Figure 16.5: Plot of the Exact Solution and the Two Term Approximation.

Example 16.7.6 Find the first two nonzero terms in the series expansion about z = 0 of the solution to
cos z
w0 − i w = 0.
z

668
 cos z
Since z
has a simple pole at z = 0 and limz→0 −i cos z = −i we see that the Frobenius series will have the form
∞
X
i
w=z an z n , a0 6= 0.
n=0

P∞ (−1)n z 2n
Recall that cos z has the Taylor expansion n=0 (2n)!
. Substituting the Frobenius expansion into the differential
equation yields
∞ ∞
! ∞
! ∞
!
X X X (−1)n z 2n X
z iz i−1 an z n + z i nan z n−1 −i zi an z n =0
n=0 n=0 n=0
(2n)! n=0
∞ ∞
! ∞
!
X X (−1)n z 2n X
(n + i)an z n − i an z n = 0.
n=0 n=0
(2n)! n=0

Equating powers of z,

z 0 : ia0 − ia0 = 0 ⇒ a0 is arbitrary

1
z : (1 + i)a1 − ia1 = 0 ⇒ a1 = 0
i i
z 2 : (2 + i)a2 − ia2 + a0 = 0 ⇒ a2 = − a0 .
2 4
Thus the solution is
 
i 2 i 3
w = a0 z 1 − z + O(z ) .
4

16.7.3 Irregular Singular Points

If a point is not an ordinary point or a regular singular point then it is called an irregular singular point. The
following equations have irregular singular points at the origin.

669
 √
• w0 + zw = 0

• w0 − z −2 w = 0

• w0 + exp(1/z)w = 0

Example 16.7.7 Consider the differential equation

dw
+ αz β w = 0, α 6= 0, β 6= −1, 0, 1, 2, . . .
dz

This equation has an irregular singular point at the origin. Solving this equation,
 Z  
d β
exp αz dz w = 0
dz
∞ n
(−1)n
  
α β+1 X α
w = c exp − z =c z (β+1)n .
β+1 n=0
n! β+1

If β is not an integer, then the solution has a branch point at the origin. If β is an integer, β < −1, then
the solution
P∞ has nan essential singularity at the origin. The solution cannot be expanded in a Frobenius series,
λ
w=z n=0 an z .

Although we will not show it, this result holds for any irregular singular point of the differential equation. We
cannot approximate the solution near an irregular singular point using a Frobenius expansion.
Now would be a good time to summarize what we have discovered about solutions of first order differential
equations in the complex plane.

670
 Result 16.7.3 Consider the first order differential equation
dw
+ p(z)w = 0.
dz
Ordinary Points If p(z) is analytic at z = z0 then z0 is an ordinary point of
the differential
P∞ equation. The solution can be expanded in the Taylor series
n
w = n=0 an (z − z0 ) . The radius of convergence of the series is at least the
distance to the nearest singularity of p(z) in the complex plane.
Regular Singular Points If p(z) has a simple pole at z = z0 and is analytic in some
annulus 0 < |z−z0 | < r then z0 is a regular singular point of the differential equation.
The solution at z0 will either be analytic, have a pole, or have a branch
P∞ point. Then
−β
solution can be expanded in the Frobenius series w = (z − z0 ) n=0 an (z − z0 )
where a0 6= 0 and β = limz→z0 (z−z0 )p(z). The radius of convergence of the Frobenius
series will be at least the distance to the nearest singularity of p(z).
Irregular Singular Points If the point z = z0 is not an ordinary point or a regular
singular point, then it is an irregular singular point of the differential equation. The
solution cannot be expanded in a Frobenius series about that point.

16.7.4 The Point at Infinity

Now we consider the behavior of first order linear differential equations at the point at infinity. Recall from
complex variables that the complex plane together with the point at infinity is called the extended complex plane.
To study the behavior of a function f (z) at infinity, we make the transformation z = ζ1 and study the behavior of
f (1/ζ) at ζ = 0.

671
Example 16.7.8 Let’s examine the behavior of sin z at infinity. We make the substitution z = 1/ζ and find the
Laurent expansion about ζ = 0.

∞
X (−1)n
sin(1/ζ) =
n=0
(2n + 1)! ζ (2n+1)

Since sin(1/ζ) has an essential singularity at ζ = 0, sin z has an essential singularity at infinity.

We use the same approach if we want to examine the behavior at infinity of a differential equation. Starting
with the first order differential equation,

dw
+ p(z)w = 0,
dz

we make the substitution

1 d d
z= , = −ζ 2 , w(z) = u(ζ)
ζ dz dζ

to obtain

du
−ζ 2 + p(1/ζ)u = 0
dζ
du p(1/ζ)
− u = 0.
dζ ζ2

672
 Result 16.7.4 The behavior at infinity of
dw
+ p(z)w = 0
dz
is the same as the behavior at ζ = 0 of
du p(1/ζ)
− u = 0.
dζ ζ2

Example 16.7.9 Classify the singular points of the equation

dw 1
+ 2 w = 0.
dz z +9
Rewriting this equation as
dw 1
+ w = 0,
dz (z − 3i)(z + 3i)

we see that z = 3i and z = −3i are regular singular points. The transformation z = 1/ζ yields the differential
equation
du 1 1
− 2 u=0
dζ ζ (1/ζ)2 + 9
du 1
− 2 u=0
dζ 9ζ + 1
Since the equation for u has a ordinary point at ζ = 0, z = ∞ is a ordinary point of the equation for w.

673
16.8 Exercises
Exact Equations
Exercise 16.1 (mathematica/ode/first order/exact.nb)
Find the general solution y = y(x) of the equations
dy x2 + xy + y 2
1. = ,
dx x2
2. (4y − 3x) dx + (y − 2x) dy = 0.
Hint, Solution

Exercise 16.2 (mathematica/ode/first order/exact.nb)

Determine whether or not the following equations can be made exact. If so find the corresponding general solution.
1. (3x2 − 2xy + 2) dx + (6y 2 − x2 + 3) dy = 0
dy ax + by
2. =−
dx bx + cy
Hint, Solution

Exercise 16.3 (mathematica/ode/first order/exact.nb)

Find the solutions of the following differential equations which satisfy the given initial condition. In each case
determine the interval in which the solution is defined.
dy
1. = (1 − 2x)y 2 , y(0) = −1/6.
dx
2. x dx + y e−x dy = 0, y(0) = 1.
Hint, Solution

674
Exercise 16.4
Show that

1
µ(x, y) =
xM (x, y) + yN (x, y)
is an integrating factor for the homogeneous equation,
dy
M (x, y) + N (x, y) .
dx
Hint, Solution

Exercise 16.5
Are the following equations exact? If so, solve them.
1. (4y − x)y 0 − (9x2 + y − 1) = 0

2. (2x − 2y)y 0 + (2x + 4y) = 0.

Hint, Solution

Exercise 16.6
Solve the following differential equations by inspection. That is, group terms into exact derivatives and then
integrate. f (x) and g(x) are known functions.
1. g(x)y 0 (x) + g 0 (x)y(x) = f (x)
y 0 (x)
2. y(x)
= f (x)

3. y α (x)y 0 (x) = f (x)

y0
4. cos x
+ y tan x
cos x
= cos x

675
Hint, Solution

Exercise 16.7 (mathematica/ode/first order/exact.nb)

Suppose we have a differential equation of the form dy/dt = f (y/t). Differential equations of this form are called
homogeneous equations. Since the right side only depends on the single variable y/t, it suggests itself to make
the substitution y/t = v or y = tv.

1. Show that this substitution replaces the equation dy/dt = f (y/t) by the equivalent equation tdv/dt + v =
f (v), which is separable.

2. Find the general solution of the equation dy/dt = 2(y/t) + (y/t)2 .

Hint, Solution

Exercise 16.8 (mathematica/ode/first order/exact.nb)

Find all functions f (t) such that the differential equation

dy
y 2 sin t + yf (t) =0 (16.5)
dt
is exact. Solve the differential equation for these f (t).
Hint, Solution

The First Order, Linear Differential Equation

Exercise 16.9 (mathematica/ode/first order/linear.nb)
Solve the differential equation
y
y0 + = 0.
sin x
Hint, Solution

676
Exercise 16.10 (mathematica/ode/first order/linear.nb)
Solve the differential equation

1
y 0 − y = xα .
x
Hint, Solution

Initial Conditions
Exercise 16.11 (mathematica/ode/first order/exact.nb)
Find the solutions of the following differential equations which satisfy the given initial conditions:

dy
1. + xy = x2n+1 , y(1) = 1, n∈Z
dx
dy
2. − 2xy = 1, y(0) = 1
dx
Hint, Solution

Exercise 16.12 (mathematica/ode/first order/exact.nb)

Show that if α > 0 and λ > 0, then for any real β, every solution of

dy
+ αy(x) = β e−λx
dx
satisfies limx→+∞ y(x) = 0. (The case α = λ requires special treatment.) Find the solution for β = λ = 1 which
satisfies y(0) = 1. Sketch this solution for 0 ≤ x < ∞ for several values of α. In particular, show what happens
when α → 0 and α → ∞.
Hint, Solution

Well-Posed Problems

677
Exercise 16.13
Find the solutions of

dy
t + Ay = 1 + t2
dt
which are bounded at t = 0. Consider all (real) values of A.
Hint, Solution

Equations in the Complex Plane

Exercise 16.14
Find the Taylor series expansion about the origin of the solution to

dw 1
+ w=0
dz 1−z
with the substitution w = ∞ n
P
n=0 an z . What is the radius of convergence of the series? What is the distance to
1
the nearest singularity of 1−z ?
Hint, Solution

Exercise 16.15
Classify the singular points of the following first order differential equations, (include the point at infinity).

1. w0 + sin z
z
w =0

2. w0 + 1
z−3
w =0

3. w0 + z 1/2 w = 0

Hint, Solution

678
Exercise 16.16
Consider the equation

w0 + z −2 w = 0.

The point z = 0 is an irregular singular point of the differential equation. Thus we know that we cannot expand
the solution about z = 0 in a Frobenius series. Try substituting the series solution
∞
X
λ
w=z an z n , a0 6= 0
n=0

into the differential equation anyway. What happens?

Hint, Solution

679
16.9 Hints
Exact Equations
Hint 16.1
1.

2.

Hint 16.2
1. The equation is exact. Determine the primitive u by solving the equations ux = P , uy = Q.

2. The equation can be made exact.

Hint 16.3
1. This equation is separable. Integrate to get the general solution. Apply the initial condition to determine
the constant of integration.

2. Ditto. You will have to numerically solve an equation to determine where the solution is defined.

Hint 16.4

Hint 16.5

Hint 16.6

680
 1. d
dx
[uv] = u0 v + uv 0
d 1
2. dx
log u = u

3. d c
dx
u = uc−1 u0

Hint 16.7

Hint 16.8

The First Order, Linear Differential Equation

Hint 16.9
Look in the appendix for the integral of csc x.

Hint 16.10
Make sure you consider the case α = 0.

Initial Conditions
Hint 16.11

Hint 16.12

Well-Posed Problems

681
Hint 16.13

Equations in the Complex Plane

Hint 16.14
1
The radius of convergence of the series and the distance to the nearest singularity of 1−z
are not the same.

Hint 16.15

Hint 16.16
Try to find the value of λ by substituting the series into the differential equation and equating powers of z.

682
16.10 Solutions
Exact Equations
Solution 16.1
1.

dy x2 + xy + y 2
=
dx x2

Since the right side is a homogeneous function of order zero, this is a homogeneous differential equation.
We make the change of variables u = y/x and then solve the differential equation for u.

xu0 + u = 1 + u + u2
du dx
2
=
1+u x
arctan(u) = ln |x| + c
u = tan(ln(|cx|))
y = x tan(ln(|cx|))

2.

(4y − 3x) dx + (y − 2x) dy = 0

Since the coefficients are homogeneous functions of order one, this is a homogeneous differential equation.

683
 We make the change of variables u = y/x and then solve the differential equation for u.
 y  y 
4 − 3 dx + − 2 dy = 0
x x
(4u − 3) dx + (u − 2)(u dx + x du) = 0
(u2 + 2u − 3) dx + x(u − 2) du = 0
dx u−2
+ du = 0
x (u + 3)(u − 1)
 
dx 5/4 1/4
+ − du = 0
x u+3 u−1
5 1
ln(x) + ln(u + 3) − ln(u − 1) = c
4 4
4 5
x (u + 3)
=c
u−1
x4 (y/x + 3)5
=c
y/x − 1
(y + 3x)5
=c
y−x

Solution 16.2
1.
(3x2 − 2xy + 2) dx + (6y 2 − x2 + 3) dy = 0
We check if this form of the equation, P dx + Q dy = 0, is exact.
Py = −2x, Qx = −2x
Since Py = Qx , the equation is exact. Now we find the primitive u(x, y) which satisfies
du = (3x2 − 2xy + 2) dx + (6y 2 − x2 + 3) dy.

684
 The primitive satisfies the partial differential equations

ux = P, uy = Q. (16.6)

We integrate the first equation of 16.6 to determine u up to a function of integration.

ux = 3x2 − 2xy + 2
u = x3 − x2 y + 2x + f (y)

We substitute this into the second equation of 16.6 to determine the function of integration up to an additive
constant.

−x2 + f 0 (y) = 6y 2 − x2 + 3
f 0 (y) = 6y 2 + 3
f (y) = 2y 3 + 3y

The solution of the differential equation is determined by the implicit equation u = c.

x3 − x2 y + 2x + 2y 3 + 3y = c

2.
dy ax + by
=−
dx bx + cy
(ax + by) dx + (bx + cy) dy = 0

We check if this form of the equation, P dx + Q dy = 0, is exact.

Py = b, Qx = b

Since Py = Qx , the equation is exact. Now we find the primitive u(x, y) which satisfies

du = (ax + by) dx + (bx + cy) dy

685
 The primitive satisfies the partial differential equations

ux = P, uy = Q. (16.7)

We integrate the first equation of 16.7 to determine u up to a function of integration.

ux = ax + by
1
u = ax2 + bxy + f (y)
2

We substitute this into the second equation of 16.7 to determine the function of integration up to an additive
constant.

bx + f 0 (y) = bx + cy
f 0 (y) = cy
1
f (y) = cy 2
2

The solution of the differential equation is determined by the implicit equation u = d.

ax2 + 2bxy + cy 2 = d

Solution 16.3
Note that since these equations are nonlinear, we cannot predict where the solutions will be defined from the
equation alone.

686
1. This equation is separable. We integrate to get the general solution.

dy
= (1 − 2x)y 2
dx
dy
= (1 − 2x) dx
y2
1
− = x − x2 + c
y
1
y= 2
x −x−c

Now we apply the initial condition.

1 1
y(0) = =−
−c 6
1
y= 2
x −x−6
1
y=
(x + 2)(x − 3)

The solution is defined on the interval (−2 . . . 3).

2. This equation is separable. We integrate to get the general solution.

x dx + y e−x dy = 0
x ex dx + y dy = 0
1
(x − 1) ex + y 2 = c
p 2
y = 2(c + (1 − x) ex )

687
 We apply the initial condition to determine the constant of integration.
p
y(0) = 2(c + 1) = 1
1
c=−
2
p
y = 2(1 − x) ex − 1
The function 2(1 − x) ex − 1 is plotted in Figure 16.6. We see that the argument of the square root in
the solution is non-negative only on an interval about the origin. Because 2(1 − x) ex − 1 == 0 is a mixed
algebraic / transcendental equation, we cannot solve it analytically. The solution of the differential equation
is defined on the interval (−1.67835 . . . 0.768039).

1
-5 -4 -3 -2 -1 1
-1
-2
-3

Figure 16.6: The function 2(1 − x) ex − 1.

Solution 16.4
We consider the homogeneous equation,
dy
M (x, y) + N (x, y) .
dx

688
That is, both M and N are homogeneous of degree n. Multiplying by
1
µ(x, y) =
xM (x, y) + yN (x, y)
will make the equation exact. To prove this we use the result that
dy
P (x, y) + Q(x, y) =0
dx
is exact if and only if Py = Qx .
 
∂ M
Py =
∂y xM + yN
My (xM + yN ) − M (xMy + N + yNy )
=
(xM + yN )2

 
∂ N
Qx =
∂x xM + yN
Nx (xM + yN ) − N (M + xMx + yNx )
=
(xM + yN )2

My (xM + yN ) − M (xMy + N + yNy ) = Nx (xM + yN ) − N (M + xMx + yNx )

yMy N − yM Ny = xM Nx − xMx N
xMx N + yMy N = xM Nx + yM Ny

With Euler’s theorem, this reduces to the identity,

nM N = nM N.

Thus the equation is exact. µ(x, y) is an integrating factor for the homogeneous equation.

689
Solution 16.5
1. We consider the differential equation,

(4y − x)y 0 − (9x2 + y − 1) = 0.

∂
1 − y − 9x2 = −1


Py =
∂y
∂
Qx = (4y − x) = −1
∂x
This equation is exact. It is simplest to solve the equation by rearranging terms to form exact derivatives.

4yy 0 − xy 0 − y + 1 − 9x2 = 0
d  2
2y − xy + 1 − 9x2 = 0

dx
2y 2 − xy + x − 3x3 + c = 0
1 p 
y= x ± x2 − 8(c + x − 3x3 )
4

2. We consider the differential equation,

(2x − 2y)y 0 + (2x + 4y) = 0.

∂
Py = (2x + 4y) = 4
∂y
∂
Qx = (2x − 2y) = 2
∂x
Since Py 6= Qx , this is not an exact equation.

690
Solution 16.6

1.
g(x)y 0 (x) + g 0 (x)y(x) = f (x)
d
[g(x)y(x)] = f (x)
dx
Z
1 c
y(x) = f (x) dx +
g(x) g(x)

2.
y 0 (x)
= f (x)
y(x)
d
log(y(x)) = f (x)
dx Z
log(y(x)) = f (x) dx + c
R
f (x) dx+c
y(x) = e
R
f (x) dx
y(x) = a e

3.
y α (x)y 0 (x) = f (x)
y α+1 (x)
Z
= f (x) dx + c
α+1
 Z 1/(α+1)
y(x) = (α + 1) f (x) dx + a

691
 4.

y0 tan x
+y = cos x
cos x cos x
d  y 
= cos x
dx cos x
y
= sin x + c
cos x
y(x) = sin x cos x + c cos x

Solution 16.7
1. We substitute y = tv into the differential equation and simplify.

y 
y0 = f
t
0
tv + v = f (v)
tv 0 = f (v) − v
v0 1
= (16.8)
f (v) − v t

The final equation is separable.

2. We start with the homogeneous differential equation:

dy  y   y 2
=2 + .
dt t t

692
 We substitute y = tv to obtain Equation 16.8, and solve the separable equation.
v0 1
2
=
v +v t
v0 1
=
v(v + 1) t
0 0
v v 1
− =
v v+1 t
log v − log(v + 1) = log t + c
 
v
log = log(ct)
v+1
v
= ct
v+1
ct
v=
1 − ct
t
v=
c−t
t2
y=
c−t

Solution 16.8
Recall that the differential equation
P (x, y) + Q(x, y)y 0 = 0
is exact if and only if Py = Qx . For Equation 16.5, this criterion is
2y sin t = yf 0 (t)
f 0 (t) = 2 sin t
f (t) = 2(a − cos t).

693
In this case, the differential equation is
y 2 sin t + 2yy 0 (a − cos t) = 0.
We can integrate this exact equation by inspection.
d 2

y (a − cos t) = 0
dt
y 2 (a − cos t) = c
c
y = ±√
a − cos t

The First Order, Linear Differential Equation

Solution 16.9
Consider the differential equation
y
y0 + = 0.
sin x
The solution is
R
−1/ sin x dx
y = ce
= c e− log(tan(x/2))

x
y = c cot .
2

Solution 16.10

1
y 0 − y = xα
x

694
The integrating factor is
Z 
1 1
exp − dx = exp (− log x) = .
x x

1 0 1
y − 2 y = xα−1
x  x
d 1
y = xα−1
dx x
Z
1
y = xα−1 dx + c
x
Z
y = x xα−1 dx + cx

(
xα+1
α
+ cx for α 6= 0,
y=
x log x + cx for α = 0.

Initial Conditions
Solution 16.11
1.

y 0 + xy = x2n+1 , y(1) = 1, n∈Z

The integrating factor is

2 /2
R
x dx
I(x) = e = ex .

695
 We multiply by the integrating factor and integrate. Since the initial condition is given at x = 1, we will
take the lower bound of integration to be that point.
d  x2 /2  2
e y = x2n+1 ex /2
dx Z
x
−x2 /2 2 /2 2 /2
y= e ξ 2n+1 eξ dξ + c e−x
1

We choose the constant of integration to satisfy the initial condition.

Z x
−x2 /2 2 /2 2 )/2
y= e ξ 2n+1 eξ dξ + e(1−x
1

If n ≥ 0 then we can use integration by parts to write the integral as a sum of terms. If n < 0 we can write
the integral in terms of the exponential integral function. However, the integral form above is as nice as any
other and we leave the answer in that form.

2.
dy
− 2xy(x) = 1, y(0) = 1.
dx
The integrating factor is
2
R
−2x dx
I(x) = e = e−x .

d  −x2  2
e y = e−x
dx Z
x
x2 2 2
y= e e−ξ dξ + c ex
0

696
 We choose the constant of integration to satisfy the initial condition.
 Z x 
x2 −ξ 2
y= e 1+ e dξ
0

We can write the answer in terms of the Error function,

Z x
2 2
erf (x) ≡ √ e−ξ dξ.
π 0

 √ 
x2 π
y= e 1+ erf (x)
2

Solution 16.12
The integrating factor is,
R
α dx
I(x) = e = eαx .

d
( eαx y) = β e(α−λ)x
dx Z
y = β e−αx e(α−λ)x dx + c e−αx

For α 6= λ, the solution is

e(α−λ)x
y = β e−αx + c e−αx
α−λ
β
y= e−λx + c e−αx
α−λ
Clearly the solution vanishes as x → ∞.

697
 For α = λ, the solution is

y = β e−αx x + c e−αx
y = (c + βx) e−αx

We use L’Hospital’s rule to show that the solution vanishes as x → ∞.

c + βx β
lim = lim =0
x→∞ eαx x→∞ α eαx

For β = λ = 1, the solution is

(
1
α−1
e−x + c e−αx for α 6= 1,
y=
(c + x) e−x for α = 1.

The solution which satisfies the initial condition is

(
1
α−1
( e−x + (α − 2) e−αx ) for α 6= 1,
y=
(1 + x) e−x for α = 1.

In Figure 16.7 the solution is plotted for α = 1/16, 1/8, . . . , 16.

Consider the solution in the limit as α → 0.

1
e−x + (α − 2) e−αx


lim y(x) = lim
α→0 α→0 α − 1

= 2 − e−x

698
 1

4 8 12 16

Figure 16.7: The Solution for a Range of α

In the limit as α → ∞ we have,

1
e−x + (α − 2) e−αx


lim y(x) = lim
α→∞ α→∞ α − 1
α − 2 −αx
= lim e
α→∞ α − 1
(
1 for x = 0,
=
0 for x > 0.

This behavior is shown in Figure 16.8. The first graph plots the solutions for α = 1/128, 1/64, . . . , 1. The second
graph plots the solutions for α = 1, 2, . . . , 128.

Well-Posed Problems

699
 1

1 2 3 4 1 2 3 4

Figure 16.8: The Solution as α → 0 and α → ∞

Solution 16.13
First we write the differential equation in the standard form.

dy A 1
+ y = +t
dt t t

The integrating factor is

R
A/t dt
I(t) = e = eA log t = tA

700
We multiply the differential equation by the integrating factor and integrate.

dy A 1
+ y = +t
dt t t
d A

t y = tA−1 + tA+1
Adt
t tA+2
 A + A+2 + c,
 A 6= 0, −2
A 1 2
t y = log t + 2 t + c, A=0

 1 −2
− t + log t + c, A = −2
 2 2
1 t −A
 A + A+2 + ct ,
 A 6= −2
1 2
y = log t + 2 t + c, A=0

 1 2 2
− 2 + t log t + ct , A = −2

For positive A, the solution is bounded at the origin only for c = 0. For A = 0, there are no bounded solutions.
For negative A, the solution is bounded there for any value of c and thus we have a one-parameter family of
solutions.
In summary, the solutions which are bounded at the origin are:


1 t2
 A + A+2 ,
 A>0
1 t2 −A
y = A + A+2 + ct , A < 0, A 6= −2

 1 2 2
− 2 + t log t + ct , A = −2

Equations in the Complex Plane

701
Solution 16.14 P
We substitute w = ∞ n
n=0 an z into the equation
dw
dz
+ 1
1−z
w = 0.
∞ ∞
d X n 1 X
an z + an z n = 0
dz n=0 1 − z n=0
∞
X ∞
X
n−1
(1 − z) nan z + an z n = 0
n=1 n=0
∞
X ∞
X ∞
X
n n
(n + 1)an+1 z − nan z + an z n = 0
n=0 n=0 n=0
∞
X
((n + 1)an+1 − (n − 1)an ) z n = 0
n=0

Equating powers of z to zero, we obtain the relation,

n−1
an+1 = an .
n+1
a0 is arbitrary. We can compute the rest of the coefficients from the recurrence relation.
−1
a1 = a0 = −a0
1
0
a2 = a1 = 0
2
We see that the coefficients are zero for n ≥ 2. Thus the Taylor series expansion, (and the exact solution), is

w = a0 (1 − z).

1
The radius of convergence of the series in infinite. The nearest singularity of 1−z is at z = 1. Thus we see the
radius of convergence can be greater than the distance to the nearest singularity of the coefficient function, p(z).

702
Solution 16.15

1. Consider the equation w0 + sinz z w = 0. The point z = 0 is the only point we need to examine in the finite
plane. Since sinz z has a removable singularity at z = 0, there are no singular points in the finite plane. The
substitution z = ζ1 yields the equation

sin(1/ζ)
u0 − u = 0.
ζ

Since sin(1/ζ)
ζ
has an essential singularity at ζ = 0, the point at infinity is an irregular singular point of the
original differential equation.

2. Consider the equation w0 + z−31 1

w = 0. Since z−3 has a simple pole at z = 3, the differential equation has a
regular singular point there. Making the substitution z = 1/ζ, w(z) = u(ζ)
1
u0 − u=0
ζ 2 (1/ζ − 3)
1
u0 − u = 0.
ζ(1 − 3ζ)
Since this equation has a simple pole at ζ = 0, the original equation has a regular singular point at infinity.

3. Consider the equation w0 + z 1/2 w = 0. There is an irregular singular point at z = 0. With the substitution
z = 1/ζ, w(z) = u(ζ),

ζ −1/2
u0 − u=0
ζ2
u0 − ζ −5/2 u = 0.

We see that the point at infinity is also an irregular singular point of the original differential equation.

703
Solution 16.16
We start with the equation

w0 + z −2 w = 0.
P∞
Substituting w = z λ n=0 an z n , a0 6= 0 yields
∞
! ∞
d λ
X
n −2 λ
X
z an z +z z an z n = 0
dz n=0 n=0
∞
X ∞
X ∞
X
λ−1 n λ n−1 λ
λz an z + z nan z +z an z n−2 = 0
n=0 n=1 n=0

The lowest power of z in the expansion is z λ−2 . The coefficient of this term is a0 . Equating powers of z demands
that a0 = 0 which contradicts our initial assumption that it was nonzero. Thus we cannot find a λ such that the
solution can be expanded in the form,
∞
X
λ
w=z an z n , a0 6= 0.
n=0

704
Chapter 17

First Order Systems of Differential Equations

We all agree that your theory is crazy, but is it crazy enough?

- Niels Bohr

17.1 Matrices and Jordan Canonical Form

Functions of Square Matrices. Consider a function f (x) with a Taylor series.
∞
X f (n) (0)
f (x) = xn
n=0
n!
We can define the function to take square matrices as arguments. The function of the square matrix A is defined
in terms of the Taylor series.
∞
X f (n) (0) n
f (A) = A
n=0
n!
(Note that this definition is usually not the most convenient method for computing a function of a matrix. Use
the Jordan canonical form for that.)

705
Eigenvalues and Eigenvectors. Consider a square matrix A. A nonzero vector x is an eigenvector of the
matrix with eigenvalue λ if

Ax = λx.

Note that we can write this equation as

(A − λI)x = 0.

This equation has solutions for nonzero x if and only if A − λI is singular, (det(A − λI) = 0). We define the
characteristic polynomial of the matrix χ(λ) as this determinant.

χ(λ) = det(A − λI)

The roots of the characteristic polynomial are the eigenvalues of the matrix. The eigenvectors of distinct eigen-
values are linearly independent. Thus if a matrix has distinct eigenvalues, the eigenvectors form a basis.
If λ is a root of χ(λ) of multiplicity m then there are up to m linearly independent eigenvectors corresponding
to that eigenvalue. That is, it has from 1 to m eigenvectors.

Diagonalizing Matrices. Consider an n × n matrix A that has a complete set of n linearly independent
eigenvectors. A may or may not have distinct eigenvalues. Consider the matrix S with eigenvectors as columns.


S = x1 x2 · · · xn

A is diagonalized by the similarity transformation:

Λ = S−1 AS.

Λ is a diagonal matrix with the eigenvalues of A as the diagonal elements. Furthermore, the k th diagonal element
is λk , the eigenvalue corresponding to the the eigenvector, xk .

706
Generalized Eigenvectors. A vector xk is a generalized eigenvector of rank k if

(A − λI)k xk = 0 but (A − λI)k−1 xk 6= 0.

Eigenvectors are generalized eigenvectors of rank 1. An n × n matrix has n linearly independent generalized
eigenvectors. A chain of generalized eigenvectors generated by the rank m generalized eigenvector xm is the set:
{x1 , x2 , . . . , xm }, where

xk = (A − λI)xk+1 , for k = m − 1, . . . , 1.

Computing Generalized Eigenvectors. Let λ be an eigenvalue of multiplicity m. Let n be the smallest

integer such that

rank ( nullspace ((A − λI)n )) = m.

Let Nk denote the number of eigenvalues of rank k. These have the value:

Nk = rank nullspace (A − λI)k − rank nullspace (A − λI)k−1

.

One can compute the generalized eigenvectors of a matrix by looping through the following three steps until
all the the Nk are zero:

1. Select the largest k for which Nk is positive. Find a generalized eigenvector xk of rank k which is linearly
independent of all the generalized eigenvectors found thus far.

2. From xk generate the chain of eigenvectors {x1 , x2 , . . . , xk }. Add this chain to the known generalized
eigenvectors.

3. Decrement each positive Nk by one.

707
Example 17.1.1 Consider the matrix
 
1 1 1
A =  2 1 −1 .
−3 2 4

The characteristic polynomial of the matrix is

 
1 − λ 1 1 

χ(λ) =  2 1 − λ −1 
 −3 2 4 − λ
= (1 − λ)2 (4 − λ) + 3 + 4 + 3(1 − λ) − 2(4 − λ) + 2(1 − λ)
= −(λ − 2)3 .

Thus we see that λ = 2 is an eigenvalue of multiplicity 3. A − 2I is

 
−1 1 1
A − 2I =  2 −1 −1
−3 2 2

The rank of the nullspace space of A − 2I is less than 3.

 
0 0 0
(A − 2I)2 = −1 1 1
1 −1 −1

The rank of nullspace ((A − 2I)2 ) is less than 3 as well, so we have to take one more step.
 
0 0 0
(A − 2I)3 = 0 0 0
0 0 0

708
The rank of nullspace ((A−2I)3 ) is 3. Thus there are generalized eigenvectors of ranks 1, 2 and 3. The generalized
eigenvector of rank 3 satisfies:

(A − 2I)3 x3 = 0
 
0 0 0
0 0 0 x3 = 0
0 0 0

We choose the solution

 
1
x3 = 0 .

0

Now to compute the chain generated by x3 .

 
−1
x2 = (A − 2I)x3 =  2 
−3
 
0
x1 = (A − 2I)x2 = −1
1

Thus a set of generalized eigenvectors corresponding to the eigenvalue λ = 2 are

     
0 −1 1
x1 = −1 , x2 =  2 , x3 = 0 .

1 −3 0

709
Jordan Block. A Jordan block is a square matrix which has the constant, λ, on the diagonal and ones on the
first super-diagonal:
 
λ 1 0 ··· 0 0
0 λ 1 · · · 0 0
...
 
0 0 λ 0 0
 
. . .
 .. .. . . . . . . . . ... 

 

0 0 0 . . . λ 1


0 0 0 ··· 0 λ

Jordan Canonical Form. A matrix J is in Jordan canonical form if all the elements are zero except for Jordan
blocks Jk along the diagonal.
 
J1 0 · · · 0 0
.
 0 J2 . . 0 0
 
. . ... .. .. 
J=  .
. . . . .
 . .

0 0 . Jn−1 0 
0 0 ··· 0 Jn

The Jordan canonical form of a matrix is obtained with the similarity transformation:

J = S−1 AS,

where S is the matrix of the generalized eigenvectors of A and the generalized eigenvectors are grouped in chains.

Example 17.1.2 Again consider the matrix

 
1 1 1
A =  2 1 −1 .
−3 2 4

710
Since λ = 2 is an eigenvalue of multiplicity 3, the Jordan canonical form of the matrix is

 
2 1 0
J = 0 2 1 .
0 0 2

In Example 17.1.1 we found the generalized eigenvectors of A. We define the matrix with generalized eigenvectors
as columns:

 
0 −1 1
S = −1 2 0 .
1 −3 0

We can verify that J = S−1 AS.

J = S−1 AS
   
0 −3 −2 1 1 1 0 −1 1
= 0 −1 −1  2 1 −1 −1 2 0
1 −1 −1 −3 2 4 1 −3 0
 
2 1 0
= 0 2 1

0 0 2

711
Functions of Matrices in Jordan Canonical Form. The function of an n × n Jordan block is the upper-
triangular matrix:
 0 00

f (n−2) (λ) f (n−1) (λ)
f (λ) f 1!(λ) f 2!(λ) ··· (n−2)! (n−1)! 
0 f (n−3) (λ) f (n−2) (λ) 

 0
 f (λ) f 1!(λ) ··· (n−3)! (n−2)! 

 0 ... f (n−4) (λ) f (n−3) (λ) 

0 f (λ)
f (Jk ) = 
 .
(n−4)! (n−3)! 
 .. .
.. ... ... ... .. 
 . 


 0 ... f 0 (λ) 
0 0 f (λ) 1!

0 0 0 ··· 0 f (λ)

The function of a matrix in Jordan canonical form is

 
f (J1 ) 0 ··· 0 0
...
 0 f (J2 ) 0 0
 

 . ... ... ..  ..
f (J) =  .
 . . 
 .

 0 ... 
0 f (Jn−1 ) 0 
0 0 ··· 0 f (Jn )
The Jordan canonical form of a matrix satisfies:

f (J) = S−1 f (A)S,

where S is the matrix of the generalized eigenvectors of A. This gives us a convenient method for computing
functions of matrices.

Example 17.1.3 Consider the matrix exponential function eA for our old friend:
 
1 1 1
A=  2 1 −1 .
−3 2 4

712
In Example 17.1.2 we showed that the Jordan canonical form of the matrix is
 
2 1 0
J = 0 2 1 .
0 0 2

Since all the derivatives of eλ are just eλ , it is especially easy to compute eJ .

 2 2 2 
e e e /2
eJ =  0 e2 e2 
0 0 e2

We find eA with a similarity transformation of eJ . We use the matrix of generalized eigenvectors found in
Example 17.1.2.

eA = S eJ S−1
  2 2 2  
0 −1 1 e e e /2 0 −3 −2
eA = −1 2 0  0 e2 e2  0 −1 −1
1 −3 0 0 0 e2 1 −1 −1
 
0 2 2
A e2
e =  3 1 −1
2
−5 3 5

17.2 Systems of Differential Equations

The homogeneous differential equation

x0 (t) = Ax(t)

713
has the solution

x(t) = eAt c

where c is a vector of constants. The solution subject to the initial condition, x(t0 ) = x0 is

x(t) = eA(t−t0 ) x0 .

The homogeneous differential equation

1
x0 (t) = Ax(t)
t
has the solution

x(t) = tA c ≡ eA Log t c,

where c is a vector of constants. The solution subject to the initial condition, x(t0 ) = x0 is
 A
t
x(t) = x0 ≡ eA Log (t/t0 ) x0 .
t0

The inhomogeneous problem

x0 (t) = Ax(t) + f (t), x(t0 ) = x0

has the solution

Z t
x(t) = e A(t−t0 )
x0 + e At
e−Aτ f (τ ) dτ.
t0

714
Example 17.2.1 Consider the system
 
1 1 1
dx 
= 2 1 −1 x.
dt
−3 2 4

The general solution of the system of differential equations is

x(t) = eAt c.

In Example 17.1.3 we found eA . At is just a constant times A. The eigenvalues of At are {λk t} where {λk } are
the eigenvalues of A. The generalized eigenvectors of At are the same as those of A.
Consider eJt . The derivatives of f (λ) = eλt are f 0 (λ) = t eλt and f 00 (λ) = t2 eλt . Thus we have
 2t 
e t e2t t2 e2t /2
eJt =  0 e2t t e2t 
0 0 e2t
 
1 t t2 /2
eJt = 0 1 t  e2t
0 0 1

We find eAt with a similarity transformation.

eAt = S eJt S−1

    
0 −1 1 1 t t2 /2 0 −3 −2
eAt = −1 2 0 0 1 t  e2t 0 −1 −1
1 −3 0 0 0 1 1 −1 −1
 
1−t t t
eAt =  2t − t2 /2 1 − t + t2 /2 −t + t2 /2  e2t
−3t + t2 /2 2t − t2 /2 1 + 2t − t2 /2

715
The solution of the system of differential equations is
      
1−t t t
x(t) = c1  2t − t2 /2  + c2 1 − t + t2 /2 + c3  −t + t2 /2  e2t
−3t + t2 /2 2t − t2 /2 1 + 2t − t2 /2

Example 17.2.2 Consider the Euler equation system

 
dx 1 1 1 0
= Ax ≡ x.
dt t t 1 1

The solution is x(t) = tA c. Note that A is almost in Jordan canonical form. It has a one on the sub-diagonal
instead of the super-diagonal. It is clear that a function of A is defined
 
f (1) 0
f (A) = .
f 0 (1) f (1)

The function f (λ) = tλ has the derivative f 0 (λ) = tλ log t. Thus the solution of the system is
      
t 0 c1 t 0
x(t) = = c1 + c2
t log t t c2 t log t t

Example 17.2.3 Consider an inhomogeneous system of differential equations.

   −3 
dx 4 −2 t
= Ax + f (t) ≡ x+ , t > 0.
dt 8 −4 −t−2

The general solution is

Z
At
x(t) = e c + e At
e−At f (t) dt.

716
First we find homogeneous solutions. The characteristic equation for the matrix is
 
4 − λ −2 
χ(λ) =   = λ2 = 0
8 −4 − λ
λ = 0 is an eigenvalue of multiplicity 2. Thus the Jordan canonical form of the matrix is
 
0 1
J= .
0 0
Since rank ( nullspace (A − 0I)) = 1 there is only one eigenvector. A generalized eigenvector of rank 2 satisfies

(A − 0I)2 x2 = 0
 
0 0
x =0
0 0 2
We choose
 
1
x2 =
0
Now we generate the chain from x2 .
 
4
x1 = (A − 0I)x2 =
8
We define the matrix of generalized eigenvectors S.
 
4 1
S=
8 0

The derivative of f (λ) = eλt is f 0 (λ) = t eλt . Thus

 
1 t
Jt
e =
0 1

717
The homogeneous solution of the differential equation system is xh = eAt c where

eAt = S eJt S−1

    
4 1 1 t 0 1/8
eAt = .
8 0 0 1 1 −1/2
 
At 1 + 4t −2t
e =
8t 1 − 4t

The general solution of the inhomogeneous system of equations is

Z
At
x(t) = e c + e At
e−At f (t) dt
   Z    −3 
1 + 4t −2t 1 + 4t −2t 1 − 4t 2t t
x(t) = c+ dt
8t 1 − 4t 8t 1 − 4t −8t 1 + 4t −t−2
2 − 2 Log t + 6t − 2t12
     
1 + 4t −2t
x(t) = c1 + c2 +
8t 1 − 4t 4 − 4 Log t + 13t

We can tidy up the answer a little bit. First we take linear combinations of the homogeneous solutions to obtain
a simpler form.

2 − 2 Log t + 6t − 2t12
     
1 2t
x(t) = c1 + c2 +
2 4t − 1 4 − 4 Log t + 13t

Then we subtract 2 times the first homogeneous solution from the particular solution.

−2 Log t + 6t − 2t12
     
1 2t
x(t) = c1 + c2 +
2 4t − 1 −4 Log t + 13t

718
17.3 Exercises
Exercise 17.1 (mathematica/ode/systems/systems.nb)
Find the solution of the following initial value problem. Describe the behavior of the solution as t → ∞.
   
0 −2 1 1
x = Ax ≡ x, x(0) = x0 ≡
−5 4 3
Hint, Solution

Exercise 17.2 (mathematica/ode/systems/systems.nb)

Find the solution of the following initial value problem. Describe the behavior of the solution as t → ∞.
   
1 1 2 2
0
x = Ax ≡  0 2 2 x, x(0) = x0 ≡ 0
 
−1 1 3 1
Hint, Solution

Exercise 17.3 (mathematica/ode/systems/systems.nb)

Find the solution of the following initial value problem. Describe the behavior of the solution as t → ∞.
   
0 1 −5 1
x = Ax ≡ x, x(0) = x0 ≡
1 −3 1
Hint, Solution

Exercise 17.4 (mathematica/ode/systems/systems.nb)

Find the solution of the following initial value problem. Describe the behavior of the solution as t → ∞.
   
−3 0 2 1
0
x = Ax ≡  1 −1 0 x, x(0) = x0 ≡ 0
 
−2 −1 0 0

719
Hint, Solution

Exercise 17.5 (mathematica/ode/systems/systems.nb)

Find the solution of the following initial value problem. Describe the behavior of the solution as t → ∞.
   
0 1 −4 3
x = Ax ≡ x, x(0) = x0 ≡
4 −7 2
Hint, Solution

Exercise 17.6 (mathematica/ode/systems/systems.nb)

Find the solution of the following initial value problem. Describe the behavior of the solution as t → ∞.
   
−1 0 0 −1
x0 = Ax ≡ −4 1 0 x, x(0) = x0 ≡  2 
3 6 2 −30
Hint, Solution

Exercise 17.7
1. Consider the system
 
1 1 1
x0 = Ax =  2 1 −1 x. (17.1)
−3 2 4
(a) Show that λ = 2 is an eigenvalue of multiplicity 3 of the coefficient matrix A, and that there is only
one corresponding eigenvector, namely
 
0
ξ (1) =  1  .
−1

720
 (b) Using the information in part (i), write down one solution x(1) (t) of the system (17.1). There is no
other solution of a purely exponential form x = ξ eλt .
(c) To find a second solution use the form x = ξt e2t + η e2t , and find appropriate vectors ξ and η. This
gives a solution of the system (17.1) which is independent of the one obtained in part (ii).
(d) To find a third linearly independent solution use the form x = ξ(t2 /2) e2t + ηt e2t + ζ e2t . Show that ξ,
η and ζ satisfy the equations
(A − 2I)ξ = 0, (A − 2I)η = ξ, (A − 2I)ζ = η.
The first two equations can be taken to coincide with those obtained in part (iii). Solve the third
equation, and write down a third independent solution of the system (17.1).
2. Consider the system
 
5 −3 −2
x0 = Ax =  8 −5 −4 x. (17.2)
−4 3 3
(a) Show that λ = 1 is an eigenvalue of multiplicity 3 of the coefficient matrix A, and that there are only
two linearly independent eigenvectors, which we may take as
   
1 0
(1) (2)
ξ = 0 , ξ =
   2
2 −3
Find two independent solutions of equation (17.2).
(b) To find a third solution use the form x = ξt et + ηet ; then show that ξ and η must satisfy
(A − I)ξ = 0, (A − I)η = ξ.
Show that the most general solution of the first of these equations is ξ = c1 ξ1 + c2 ξ 2 , where c1 and
c2 are arbitrary constants. Show that, in order to solve the second of these equations it is necessary
to take c1 = c2 . Obtain such a vector η, and use it to obtain a third independent solution of the
system (17.2).

721
Hint, Solution

Exercise 17.8 (mathematica/ode/systems/systems.nb)

Consider the system of ODE’s

dx
= Ax, x(0) = x0
dt
where A is the constant 3 × 3 matrix
 
1 1 1
A= 2 1 −1
−8 −5 −3

1. Find the eigenvalues and associated eigenvectors of A. [HINT: notice that λ = −1 is a root of the charac-
teristic polynomial of A.]

2. Use the results from part (a) to construct eAt and therefore the solution to the initial value problem above.

3. Use the results of part (a) to find the general solution to

dx 1
= Ax.
dt t

Hint, Solution

Exercise 17.9 (mathematica/ode/systems/systems.nb)

1. Find the general solution to

dx
= Ax
dt

722
 where
 
2 0 1
A = 0 2 0
0 1 3

2. Solve
dx
= Ax + g(t), x(0) = 0
dt
using A from part (a).
Hint, Solution

Exercise 17.10
Let A be an n × n matrix of constants. The system

dx 1
= Ax, (17.3)
dt t
is analogous to the Euler equation.

1. Verify that when A is a 2 × 2 constant matrix, elimination of (17.3) yields a second order Euler differential
equation.

2. Now assume that A is an n × n matrix of constants. Show that this system, in analogy with the Euler
equation has solutions of the form x = atλ where a is a constant vector provided a and λ satisfy certain
conditions.

3. Based on your experience with the treatment of multiple roots in the solution of constant coefficient systems,
what form will the general solution of (17.3) take if λ is a multiple eigenvalue in the eigenvalue problem
derived in part (b)?

723
 4. Verify your prediction by deriving the general solution for the system
 
dx 1 1 0
= x.
dt t 1 1

Hint, Solution

Exercise 17.11
Use the matrix form of the method of variation of parameters to find the general solution of
   −3 
dx 4 −2 t
= x+ , t > 0.
dt 8 −4 −t−2

Hint, Solution

724
17.4 Hints
Hint 17.1

Hint 17.2

Hint 17.3

Hint 17.4

Hint 17.5

Hint 17.6

Hint 17.7

Hint 17.8

725
Hint 17.9

Hint 17.10

Hint 17.11

726
17.5 Solutions
Solution 17.1
We consider an initial value problem.
   
0 −2 1 1
x = Ax ≡ x, x(0) = x0 ≡
−5 4 3
Method 1. Find Homogeneous Solutions. The matrix has the distinct eigenvalues λ1 = −1, λ2 = 3. The
corresponding eigenvectors are
   
1 1
x1 = , x2 = .
1 5
The general solution of the system of differential equations is
   
1 −t 1
x = c1 e + c2 e3t .
1 5
We apply the initial condition to determine the constants.
    
1 1 c1 1
=
1 5 c2 3
1 1
c1 = , c2 =
2 2
The solution subject to the initial condition is
   
1 1 −t 1 1
x= e + e3t
2 1 2 5

For large t, the solution looks like

 
1 1
x≈ e3t .
2 5

727
 10

7.5

2.5

-10 -7.5 -5 -2.5 2.5 5 7.5 10

-2.5

-5

-7.5

-10

Figure 17.1: Homogeneous solutions in the phase plane.

Both coordinates tend to infinity.

Figure 17.1 show some homogeneous solutions in the phase plane.

Method 2. Use the Exponential Matrix. The Jordan canonical form of the matrix is

 
−1 0
J= .
0 3

728
The solution of the initial value problem is x = eAt x0 .
x = eAt x0
= S eJt S−1 x0
   −t    
1 1 e 0 1 5 −1 1
= 3t
1 5 0 e 4 −1 1 3
 −t 3t

1 e +e
=
2 e−t + 5 e3t

   
1 1 −t 1 1
x= e + e3t
2 1 2 5

Solution 17.2
We consider an initial value problem.
   
1 1 2 2
x0 = Ax ≡  0 2 2 x, x(0) = x0 ≡ 0

−1 1 3 1
Method 1. Find Homogeneous Solutions. The matrix has the distinct eigenvalues λ1 = 1, λ2 = 2,
λ3 = 3. The corresponding eigenvectors are
     
0 1 2
x1 = −2 , x2 = 1 , x3 = 2 .
    
1 0 1
The general solution of the system of differential equations is
     
0 1 2
x = c1 −2 e + c2 1 e + c3 2 e3t .
  t   2t 
1 0 1

729
We apply the initial condition to determine the constants.
    
0 1 2 c1 2
−2 1 2   c2 = 0
 
1 0 1 c3 1
c1 = 1, c2 = 2, c3 = 0

The solution subject to the initial condition is


  
0 1
x = −2 e + 2 1 e2t .
  t 
1 0

As t → ∞, all coordinates tend to infinity.

Method 2. Use the Exponential Matrix. The Jordan canonical form of the matrix is
 
1 0 0
J = 0 2 0 .
0 0 3

The solution of the initial value problem is x = eAt x0 .

x = eAt x0
= S eJt S−1 x0
  t    
0 1 2 e 0 0 1 −1 0 2
1
=  −2 1 2   0 e2t 0   4 −2 −4 0
2
1 0 1 0 0 e3t −1 1 2 1
 
2 e2t
= −2 et + 2 e2t 
et

730
    
0 2
x = −2 e + 2 e2t .
  t 
1 0

Solution 17.3
We consider an initial value problem.
   
0 1 −5 1
x = Ax ≡ x, x(0) = x0 ≡
1 −3 1
Method 1. Find Homogeneous Solutions. The matrix has the distinct eigenvalues λ1 = −1 − i, λ2 =
−1 + i. The corresponding eigenvectors are
   
2−i 2+i
x1 = , x2 = .
1 1
The general solution of the system of differential equations is
   
2−i (−1−i)t 2+i
x = c1 e + c2 e(−1+i)t .
1 1
We can take the real and imaginary parts of either of these solution to obtain real-valued solutions.
     
2+i 2 cos(t) − sin(t) −t cos(t) + 2 sin(t)
e(−1+i)t
= e +i e−t
1 cos(t) sin(t)
   
2 cos(t) − sin(t) −t cos(t) + 2 sin(t)
x = c1 e + c2 e−t
cos(t) sin(t)
We apply the initial condition to determine the constants.
    
2 1 c1 1
=
1 0 c2 1
c1 = 1, c2 = −1

731
The solution subject to the initial condition is
 
cos(t) − 3 sin(t)
x= e−t .
cos(t) − sin(t)

Plotted in the phase plane, the solution spirals in to the origin at t increases. Both coordinates tend to zero as
t → ∞.
Method 2. Use the Exponential Matrix. The Jordan canonical form of the matrix is
 
−1 − i 0
J= .
0 −1 + i

The solution of the initial value problem is x = eAt x0 .

x = eAt x0
= S eJt S−1 x0
   (−1−i)t    
2−i 2+i e 0 1 i 1 − i2 1
= (−1+i)t
1 1 0 e 2 −i 1 + i2 1
−t
 
(cos(t) − 3 sin(t)) e
=
(cos(t) − sin(t)) e−t

   
1 −t 3
x= e cos(t) − e−t sin(t)
1 1

Solution 17.4
We consider an initial value problem.
   
−3 0 2 1
x0 = Ax ≡  1 −1 0 x, x(0) = x0 ≡ 0

−2 −1 0 0

732
 Method
√ 1. Find√Homogeneous Solutions. The matrix has the distinct eigenvalues λ1 = −2, λ2 =
−1 − i 2, λ3 = −1 + i 2. The corresponding eigenvectors are
   √   √ 
2 2 + i √2 2 − i √2
x1 = −2 , x2 = −1 + i 2 , x3 = −1 − i 2 .
1 3 3

The general solution of the system of differential equations is

   √   √ 
2 2 + i √2 √ 2 − i √2 √
x = c1 −2 e−2t + c2 −1 + i 2 e(−1−i 2)t + c3 −1 − i 2 e(−1+i 2)t .
1 3 3

We can take the real and imaginary parts of the second or third solution to obtain two real-valued solutions.
 √   √ √ √  √ √ √ 
2 + i √2 √ 2 cos( √2t) + √2 sin( √2t) √ 2 cos( √2t) − 2 sin(√ 2t)
(−1−i 2)t −t  e−t
−1 + i 2 e = − cos( 2t) +√ 2 sin( 2t) e + i  2 cos( 2t) + √ sin( 2t)
3 3 cos( 2t) −3 sin( 2t)
   √ √ √  √ √ √ 
2 2 cos( √2t) + √2 sin( √2t) √2 cos( √2t) − 2 sin(
√ 2t)
−2t −t  e−t
x = c1 −2 e + c2 − cos( 2t) +√ 2 sin( 2t) e + c3
     2 cos( 2t) +√ sin( 2t)
1 3 cos( 2t) −3 sin( 2t)

We apply the initial condition to determine the constants.

 √    
2 2 √2 c1 1
−2 −1 2   c2
 = 0
1 3 0 c3 0
1 1 5
c1 = , c 2 = − , c 3 = √
3 9 9 2

733
The solution subject to the initial condition is
  √ √ √ 
2 2 cos( √2t) − 4√ 2 sin(√ 2t)
1 1
x = −2 e−2t +  4 cos( √2t) + √ 2 sin( √2t)  e−t .
3 6
1 −2 cos( 2t) − 5 2 sin( 2t)

As t → ∞, all coordinates tend to infinity. Plotted in the phase plane, the solution would spiral in to the origin.
Method 2. Use the Exponential Matrix. The Jordan canonical form of the matrix is
 
−2 0 √ 0
J =  0 −1 − i 2 0 √ .
0 0 −1 + i 2

The solution of the initial value problem is x = eAt x0 .

x = eAt x0
= S eJt S−1 x0
 √ √   −2t 
6 2 + i √2 2 − i √2 e 0√ 0
1 (−1−i 2)t
= −6 −1 + i 2 −1 − i 2   0 e 0√ 
3
3 3 3 0 0 e(−1+i 2)t
  
2√ −2√ −2√ 1
1
−1 − i5√2/2 1 − i2√2 4 + i√2 0
6
−1 + i5 2/2 1 + i2 2 4 − i 2 0

  √ √ √ 
2 2 cos( √2t) − 4√ 2 sin(√ 2t)
1 1
x = −2 e−2t +  4 cos( √2t) + √ 2 sin( √2t)  e−t .
3 6
1 −2 cos( 2t) − 5 2 sin( 2t)

734
Solution 17.5
We consider an initial value problem.
   
0 1 −4 3
x = Ax ≡ x, x(0) = x0 ≡
4 −7 2
Method 1. Find Homogeneous Solutions. The matrix has the double eigenvalue λ1 = λ2 = −3. There
is only one corresponding eigenvector. We compute a chain of generalized eigenvectors.

(A + 3I)2 x2 = 0
0x2 = 0
 
1
x2 =
0
(A + 3I)x2 = x1
 
4
x1 =
4
The general solution of the system of differential equations is
     
1 −3t 4 1
x = c1 e + c2 t+ e−3t .
1 4 0
We apply the initial condition to determine the constants.
    
1 1 c1 3
=
1 0 c2 2
c1 = 2, c2 = 1

The solution subject to the initial condition is

 
3 + 4t
x= e−3t .
2 + 4t

735
Both coordinates tend to zero as t → ∞.
Method 2. Use the Exponential Matrix. The Jordan canonical form of the matrix is
 
−3 1
J= .
0 −3

The solution of the initial value problem is x = eAt x0 .

x = eAt x0
= S eJt S−1 x0
  −3t
t e−3t
   
1 1/4 e 0 1 3
= −3t
1 0 0 e 4 −4 2

 
3 + 4t
x= e−3t .
2 + 4t

Solution 17.6
We consider an initial value problem.
   
−1 0 0 −1
x0 = Ax ≡ −4 1 0 x, x(0) = x0 ≡  2 
3 6 2 −30

Method 1. Find Homogeneous Solutions. The matrix has the distinct eigenvalues λ1 = −1, λ2 = 1,
λ3 = 2. The corresponding eigenvectors are
     
−1 0 0
x1 = −2 , x2 = −1 , x3 = 0 .
    
5 6 1

736
The general solution of the system of differential equations is
     
−1 0 0
−t
x = c1 −2 e + c2 −1 e + c3 0 e2t .
    t 
5 6 1
We apply the initial condition to determine the constants.
    
−1 0 0 c1 −1
−2 −1 0 c2  =  2 
5 6 1 c3 −30
c1 = 1, c2 = −4, c3 = −11
The solution subject to the initial condition is
     
−1 0 0
x = −2 e−t − 4 −1 et − 11 0 e2t .
5 6 1

As t → ∞, the first coordinate vanishes, the second coordinate tends to ∞ and the third coordinate tends to −∞
Method 2. Use the Exponential Matrix. The Jordan canonical form of the matrix is
 
−1 0 0
J =  0 1 0 .
0 0 2
The solution of the initial value problem is x = eAt x0 .
x = eAt x0
= S eJt S−1 x0
   −t    
−1 0 0 e 0 0 −1 0 0 −1
1
= −2 −1 0  0 et 0   2 −1 0  2 
2
5 6 1 0 0 e2t −7 6 1 −30

737
      
−1 0 0
−t
x = −2 e − 4 −1 e − 11 0 e2t .
    t 
5 6 1

Solution 17.7
1. (a) We compute the eigenvalues of the matrix.
 
1 − λ 1 1 

χ(λ) =  2 1 − λ −1  = −λ3 + 6λ2 − 12λ + 8 = −(λ − 2)3
 −3 2 4 − λ
λ = 2 is an eigenvalue of multiplicity 3. The rank of the null space of A − 2I is 1. (The first two rows
are linearly independent, but the third is a linear combination of the first two.)
 
−1 1 1
A − 2I =  2 −1 −1
−3 2 2
Thus there is only one eigenvector.
  
−1 1 1 ξ1
 2 −1 −1 ξ2  = 0
−3 2 2 ξ3
 
0
(1)
ξ =  1
−1

(b) One solution of the system of differential equations is

 
0
x(1) =  1  e2t .
−1

738
(c) We substitute the form x = ξt e2t + η e2t into the differential equation.
x0 = Ax
ξ e2t + 2ξt e2t + 2η e2t = Aξt e2t + Aη e2t
(A − 2I)ξ = 0, (A − 2I)η = ξ
We already have a solution of the first equation, we need the generalized eigenvector η. Note that η
is only determined up to a constant times ξ. Thus we look for the solution whose second component
vanishes to simplify the algebra.
(A − 2I)η = ξ
    
−1 1 1 η1 0
 2 −1 −1  0  =  1 
−3 2 2 η3 −1
−η1 + η3 = 0, 2η1 − η3 = 1, −3η1 + 2η3 = −1
 
1
η = 0

1
A second linearly independent solution is
   
0 1
x(2) =  1 t e + 0 e2t .
 2t 
−1 1

(d) To find a third solution we substutite the form x = ξ(t2 /2) e2t + ηt e2t + ζ e2t into the differential
equation.
x0 = Ax
2ξ(t2 /2) e2t + (ξ + 2η)t e2t + (η + 2ζ) e2t = Aξ(t2 /2) e2t + Aηt e2t + Aζ e2t
(A − 2I)ξ = 0, (A − 2I)η = ξ, (A − 2I)ζ = η

739
 We have already solved the first two equations, we need the generalized eigenvector ζ. Note that ζ
is only determined up to a constant times ξ. Thus we look for the solution whose second component
vanishes to simplify the algebra.

(A − 2I)ζ = η
    
−1 1 1 ζ1 1
 2 −1 −1  0  = 0
−3 2 2 ζ3 1
−ζ1 + ζ3 = 1, 2ζ1 − ζ3 = 0, −3ζ1 + 2ζ3 = 1
 
1
ζ = 0

2
A third linearly independent solution is
     
0 1 1
(3)
x =   2 2t   2t
1 (t /2) e + 0 t e + 0 e2t

−1 1 2

2. (a) We compute the eigenvalues of the matrix.

 
5 − λ −3 −2 

χ(λ) =  8 −5 − λ −4  = −λ3 + 3λ2 − 3λ + 1 = −(λ − 1)3
 −4 3 3 − λ

λ = 1 is an eigenvalue of multiplicity 3. The rank of the null space of A − I is 2. (The second and
third rows are multiples of the first.)
 
4 −3 −2
A−I=  8 −6 −4
−4 3 2

740
 Thus there are two eigenvectors.
  
4 −3 −2 ξ1
 8 −6 −4 ξ2  = 0
−4 3 2 ξ3
   
1 0
ξ (1) = 0 , ξ (2) =  2 
2 −3

Two linearly independent solutions of the differential equation are

   
1 0
x(1) = 0 et ,
 x(2) =  2  et .
2 −3

(b) We substitute the form x = ξt et + η et into the differential equation.

x0 = Ax
ξ et + ξt et + η et = Aξt et + Aη et
(A − I)ξ = 0, (A − I)η = ξ

The general solution of the first equation is a linear combination of the two solutions we found in the
previous part.

ξ = c1 ξ 1 + c2 ξ 2

Now we find the generalized eigenvector, η. Note that η is only determined up to a linear combination

741
 of ξ 1 and ξ 2 . Thus we can take the first two components of η to be zero.
      
4 −3 −2 0 1 0
 8 −6 −4  0  = c1 0 + c2  2 
−4 3 2 η3 2 −3
−2η3 = c1 , −4η3 = 2c2 , 2η3 = 2c1 − 3c2
c1
c1 = c2 , η3 = −
2
We see that we must take c1 = c2 in order to obtain a solution. We choose c1 = c2 = 2 A third linearly
independent solution of the differential equation is
   
2 0
x(3) =   t
4 t e + 0  et .

−2 −1

Solution 17.8
1. The characteristic polynomial of the matrix is
 
1 − λ 1 1 
 
χ(λ) =  2 1−λ −1 
 −8 −5 −3 − λ
= (1 − λ)2 (−3 − λ) + 8 − 10 − 5(1 − λ) − 2(−3 − λ) − 8(1 − λ)
= −λ3 − λ2 + 4λ + 4
= −(λ + 2)(λ + 1)(λ − 2)

Thus we see that the eigenvalues are λ = −2, −1, 2. The eigenvectors ξ satisfy

(A − λI)ξ = 0.

742
For λ = −2, we have
(A + 2I)ξ = 0.
    
3 1 1 ξ1 0
2 3 −1   ξ 2 = 0
 
−8 −5 −1 ξ3 0
If we take ξ3 = 1 then the first two rows give us the system,
    
3 1 ξ1 −1
=
2 3 ξ2 1
which has the solution ξ1 = −4/7, ξ2 = 5/7. For the first eigenvector we choose:
 
−4
ξ=  5
7

For λ = −1, we have

(A + I)ξ = 0.
    
2 1 1 ξ1 0
2 2 −1  ξ 2 = 0
 
−8 −5 −2 ξ3 0
If we take ξ3 = 1 then the first two rows give us the system,
    
2 1 ξ1 −1
=
2 2 ξ2 1
which has the solution ξ1 = −3/2, ξ2 = 2. For the second eigenvector we choose:
 
−3
ξ=  4
2

743
 For λ = 2, we have

(A + I)ξ = 0.
    
−1 1 1 ξ1 0
 2 −1 −1 ξ2  = 0
−8 −5 −5 ξ3 0
If we take ξ3 = 1 then the first two rows give us the system,
    
−1 1 ξ1 −1
=
2 −1 ξ2 1
which has the solution ξ1 = 0, ξ2 = −1. For the third eigenvector we choose:
 
0
ξ = −1
1

In summary, the eigenvalues and eigenvectors are

     
 −4 −3 0 
λ = {−2, −1, 2}, ξ=  5 , 4 , −1
   
7 2 1
 

2. The matrix is diagonalized with the similarity transformation

J = S−1 AS,

where S is the matrix with eigenvectors as columns:

 
−4 −3 0
S= 5 4 −1
7 2 1

744
 The matrix exponential, eAt is given by
eA = S eJ S−1 .
   −2t   
−4 −3 0 e 0 0 6 3 3
1 
eA =  5 4 −1  0 e−t 0 −12 −4 −4 .
2t 12
7 2 1 0 0 e −18 −13 −1

−2 e−2t + 3 e−t − e−2t + e−t − e−2t + e−t

 
−2t −t t 15 e−2t −16 e−t +13 et 15 e−2t −16 e−t + et 
eAt =  5 e −8 e +3 e
2 12 12
7 e−2t −4 e−t −3 et 21 e−2t −8 e−t −13 et 21 e−2t −8 e−t − et
2 12 12

The solution of the initial value problem is eAt x0 .

3. The general solution of the Euler equation is
     
−4 −3 0
−2 −1
c1 5 t + c2 4 t + c3 −1 t2 .
    
7 2 1

We could also write the solution as

x = tA c ≡ eA log t c,

Solution 17.9
1. The characteristic polynomial of the matrix is
 
2 − λ 0 1 

χ(λ) =  0 2−λ 0 
 0 1 3 − λ
= (2 − λ)2 (3 − λ)

745
Thus we see that the eigenvalues are λ = 2, 2, 3. Consider
 
0 0 1
A − 2I = 0 0 0 .
0 1 3

Since rank ( nullspace (A − 2I)) = 1 there is one eigenvector and one generalized eigenvector of rank two for
λ = 2. The generalized eigenvector of rank two satisfies

(A − 2I)2 ξ 2 = 0
 
0 1 1
0 0 0 ξ 2 = 0
0 1 1

We choose the solution

 
0
ξ 2 = −1 .
1

The eigenvector for λ = 2 is

 
1
ξ 1 = (A − 2I)ξ 2 = 0 .

0

The eigenvector for λ = 3 satisfies

(A − 3I)2 ξ = 0
 
−1 0 1
 0 −1 0 ξ = 0
0 1 0

746
We choose the solution
 
1
ξ = 0 .

1
The eigenvalues and generalized eigenvectors are
     
 1 0 1 
λ = {2, 2, 3}, ξ = 0 , −1 , 0 .
0 1 1
 

The matrix of eigenvectors and its inverse is

   
1 0 1 1 −1 −1
S = 0 −1 0 , S−1 = 0 −1 0  .
0 1 1 0 1 1
The Jordan canonical form of the matrix, which satisfies J = S−1 AS is
 
2 1 0
J= 0 2
 0
0 0 3
Recall that the function of a Jordan block is:
  f 0 (λ) f 00 (λ) f 000 (λ)
 
λ 1 0 0 f (λ) 1! 2! 3!
 0 λ 1 0   f 0 (λ) f 00 (λ) 
0 f (λ)
f 0 0 λ 1  = 
1! 2! ,
   
f 0 (λ) 
 0 0 f (λ) 1!

0 0 0 λ 0 0 0 f (λ)
and that the function of a matrix in Jordan canonical form is
   
J1 0 0 0 f (J1 ) 0 0 0
 0 J2 0 0  =  0 f (J2 ) 0 0 
 
f 
 0 0
.
J3 0   0 0 f (J3 ) 0 
0 0 0 J4 0 0 0 f (J4 )

747
 We want to compute eJt so we consider the function f (λ) = eλt , which has the derivative f 0 (λ) = t eλt .
Thus we see that
 2t 
e t e2t 0
eJt =  0 e2t 0
0 0 e3t
The exponential matrix is
eAt = S eJt S−1 ,
 
e2t −(1 + t) e2t + e3t − e2t + e3t
eAt = 0 e2t 0 .
0 − e2t + e3t e3t

The general solution of the homogeneous differential equation is

x = eAt C.

2. The solution of the inhomogeneous differential equation subject to the initial condition is
Z t
At
x= e 0+ e At
e−Aτ g(τ ) dτ
0
Z t
x = eAt e−Aτ g(τ ) dτ
0

Solution 17.10
1.
dx 1
= Ax
 0  dt  t   
x a b x1
t 10 =
x2 c d x2

748
 The first component of this equation is

tx01 = ax1 + bx2 .

We differentiate and multiply by t to obtain a second order coupled equation for x1 . We use (17.3) to
eliminate the dependence on x2 .

t2 x001 + tx01 = atx01 + btx02

t2 x001 + (1 − a)tx01 = b(cx1 + dx2 )
t2 x001 + (1 − a)tx01 − bcx1 = d(tx01 − ax1 )
t2 x001 + (1 − a − d)tx01 + (ad − bc)x1 = 0

Thus we see that x1 satisfies a second order, Euler equation. By symmetry we see that x2 satisfies,

t2 x002 + (1 − b − c)tx02 + (bc − ad)x2 = 0.

2. We substitute x = atλ into (17.3).

1
λatλ−1 = Aatλ
t
Aa = λa

Thus we see that x = atλ is a solution if λ is an eigenvalue of A with eigenvector a.

3. Suppose that λ = α is an eigenvalue of multiplicity 2. If λ = α has two linearly independent eigenvectors,
a and b then atα and btα are linearly independent solutions. If λ = α has only one linearly independent
eigenvector, a, then atα is a solution. We look for a second solution of the form

x = ξtα log t + ηtα .

Substituting this into the differential equation yields

αξtα−1 log t + ξtα−1 + αηtα−1 = Aξtα−1 log t + Aηtα−1

749
 We equate coefficients of tα−1 log t and tα−1 to determine ξ and η.

(A − αI)ξ = 0, (A − αI)η = ξ

These equations have solutions because λ = α has generalized eigenvectors of first and second order.
Note that the change of independent variable τ = log t, y(τ ) = x(t), will transform (17.3) into a constant
coefficient system.
dy
= Ay
dτ
Thus all the methods for solving constant coefficient systems carry over directly to solving (17.3). In the
case of eigenvalues with multiplicity greater than one, we will have solutions of the form,

ξtα , ξtα log t + ηtα , ξtα (log t)2 + ηtα log t + ζtα , ... ,

analogous to the form of the solutions for a constant coefficient system,

ξ eατ , ξτ eατ + η eατ , ξτ 2 eατ + ητ eατ + ζ eατ , ... .

4. Method 1. Now we consider

 
dx 1 1 0
= x.
dt t 1 1

The characteristic polynomial of the matrix is

 
1 − λ 0 
χ(λ) =  = (1 − λ)2 .
1 1 − λ

λ = 1 is an eigenvalue of multiplicity 2. The equation for the associated eigenvectors is

    
0 0 ξ1 0
= .
1 0 ξ2 0

750
There is only one linearly independent eigenvector, which we choose to be
 
0
a= .
1

One solution of the differential equation is

 
0
x1 = t.
1

We look for a second solution of the form

x2 = at log t + ηt.

η satisfies the equation

   
0 0 0
(A − I)η = η= .
1 0 1

The solution is determined only up to an additive multiple of a. We choose

 
1
η= .
0

Thus a second linearly independent solution is

   
0 1
x2 = t log t + t.
1 0

The general solution of the differential equation is

      
0 0 1
x = c1 t + c2 t log t + t .
1 1 0

751
Method 2. Note that the matrix is lower triangular.
 0   
x1 1 1 0 x1
0 = (17.4)
x2 t 1 1 x2

We have an uncoupled equation for x1 .

1
x01 = x1
t
x1 = c 1 t

By substituting the solution for x1 into (17.4), we obtain an uncoupled equation for x2 .
1
x02 = (c1 t + x2 )
t
1
x02 − x2 = c1
 t0
1 c1
x2 =
t t
1
x2 = c1 log t + c2
t
x2 = c1 t log t + c2 t

Thus the solution of the system is

 
c1 t
x= ,
c1 t log t + c2 t
   
t 0
x = c1 + c2 ,
t log t t

which is equivalent to the solution we obtained previously.

752
Solution 17.11
Homogeneous Solution, Method 1. We designate the inhomogeneous system of differential equations

x0 = Ax + g(t).

First we find homogeneous solutions. The characteristic equation for the matrix is
 
4 − λ −2 
χ(λ) =   = λ2 = 0
8 −4 − λ

λ = 0 is an eigenvalue of multiplicity 2. The eigenvectors satisfy

    
4 −2 ξ1 0
= .
8 −4 ξ2 0

Thus we see that there is only one linearly independent eigenvector. We choose
 
1
ξ= .
2

One homogeneous solution is then

   
1 0t 1
x1 = e = .
2 2

We look for a second homogeneous solution of the form

x2 = ξt + η.

We substitute this into the homogeneous equation.

x02 = Ax2
ξ = A(ξt + η)

753
We see that ξ and η satisfy

Aξ = 0, Aη = ξ.

We choose ξ to be the eigenvector that we found previously. The equation for η is then
    
4 −2 η1 1
= .
8 −4 η2 2

η is determined up to an additive multiple of ξ. We choose

 
0
η= .
−1/2

Thus a second homogeneous solution is

   
1 0
x2 = t+ .
2 −1/2

The general homogeneous solution of the system is

   
1 t
xh = c1 + c2
2 2t − 1/2

We can write this in matrix notation using the fundamental matrix Ψ(t).
  
1 t c1
xh = Ψ(t)c =
2 2t − 1/2 c2

Homogeneous Solution, Method 2. The similarity transform C−1 AC with

 
1 0
C=
2 −1/2

754
will convert the matrix
 
4 −2
A=
8 −4

to Jordan canonical form. We make the change of variables,

 
1 0
y= x.
2 −1/2

The homogeneous system becomes

   
dy 1 0 4 −2 1 0
= y
dt 4 −2 8 −4 2 −1/2
 0   
y1 0 1 y1
0 =
y2 0 0 y2

The equation for y2 is

y20 = 0.
y2 = c2

The equation for y1 becomes

y10 = c2 .
y1 = c1 + c2 t

The solution for y is then

   
1 t
y = c1 + c2 .
0 1

755
We multiply this by C to obtain the homogeneous solution for x.
   
1 t
xh = c1 + c2
2 2t − 1/2

Inhomogeneous Solution. By the method of variation of parameters, a particular solution is

Z
xp = Ψ(t) Ψ−1 (t)g(t) dt.
 Z    −3 
1 t 1 − 4t 2t t
xp = dt
2 2t − 1/2 4 −2 −t−2
−2t−1 − 4t−2 + t−3
 Z  
1 t
xp = dt
2 2t − 1/2 2t−2 + 4t−3
−2 log t + 4t−1 − 21 t−2
  
1 t
xp =
2 2t − 1/2 −2t−1 − 2t−2
−2 − 2 log t + 2t−1 − 21 t−2
 
xp =
−4 − 4 log t + 5t−1

By adding 2 times our first homogeneous solution, we obtain

−2 log t + 2t−1 − 21 t−2

 
xp =
−4 log t + 5t−1

The general solution of the system of differential equations is

−2 log t + 2t−1 − 12 t−2

     
1 t
x = c1 + c2 +
2 2t − 1/2 −4 log t + 5t−1

756
Chapter 18

Theory of Linear Ordinary Differential

Equations

A little partyin’ is good for the soul.

-Matt Metz

757
18.1 Nature of Solutions

Result 18.1.1 Consider the nth order ordinary differential equation of the form
dn y dn−1 y dy
L[y] = n + pn−1 (x) n−1 + · · · + p1 (x) + p0 (x)y = f (x). (18.1)
dx dx dx
If the coefficient functions pn−1 (x), . . . , p0 (x) and the inhomogeneity f (x) are continuous
on some interval a < x < b then the differential equation subject to the conditions,

y(x0 ) = v0 , y 0 (x0 ) = v1 , ... y (n−1) (x0 ) = vn−1 , a < x0 < b,

has a unique solution on the interval.

Linearity of the Operator. The differential operator L is linear. To verify this,

dn dn−1 d
L[cy] = n
(cy) + p n−1 (x) n−1
(cy) + · · · + p1 (x) (cy) + p0 (x)(cy)
dx dx dx
dn dn−1 d
= c n y + cpn−1 (x) n−1 y + · · · + cp1 (x) y + cp0 (x)y
dx dx dx
= cL[y]
dn dn−1 d
L[y1 + y2 ] = n (y1 + y2 ) + pn−1 (x) n−1 (y1 + y2 ) + · · · + p1 (x) (y1 + y2 ) + p0 (x)(y1 + y2 )
dx dx dx
dn dn−1 d
= n (y1 ) + pn−1 (x) n−1 (y1 ) + · · · + p1 (x) (y1 ) + p0 (x)(y1 )
dx dx dx
dn dn−1 d
+ n (y2 ) + pn−1 (x) n−1 (y2 ) + · · · + p1 (x) (y2 ) + p0 (x)(y2 )
dx dx dx
= L[y1 ] + L[y2 ].

758
Homogeneous Solutions. The general homogeneous equation has the form

dn y dn−1 y dy
L[y] = n
+ pn−1 (x) n−1
+ · · · + p1 (x) + p0 (x)y = 0.
dx dx dx
From the linearity of L, we see that if y1 and y2 are solutions to the homogeneous equation then c1 y1 + c2 y2 is
also a solution, (L[c1 y1 + c2 y2 ] = 0).
On any interval where the coefficient functions are continuous, the nth order linear homogeneous equation has
n linearly independent solutions, y1 , y2 , . . . , yn . (We will study linear independence in Section 18.3.) The general
solution to the homogeneous problem is then

yh = c1 y1 + c2 y2 + · · · + cn yn .

Particular Solutions. Any function, yp , that satisfies the inhomogeneous equation, L[yp ] = f (x), is called a
particular solution or particular integral of the equation. Note that for linear differential equations the particular
solution is not unique. If yp is a particular solution then yp + yh is also a particular solution where yh is any
homogeneous solution.
The general solution to the problem L[y] = f (x) is the sum of a particular solution and a linear combination
of the homogeneous solutions

y = yp + c1 y1 + · · · + cn yn .

Example 18.1.1 Consider the differential equation

y 00 − y 0 = 1.

You can verify that two homogeneous solutions are ex and 1. A particular solution is −x. Thus the general
solution is

y = −x + c1 ex + c2 .

759
Real-Valued Solutions. If the coefficient function and the inhomogeneity in Equation 18.1 are real-valued,
then the general solution can be written in terms of real-valued functions. Let y be any, homogeneous solution,
(perhaps complex-valued). By taking the complex conjugate of the equation L[y] = 0 we show that ȳ is a
homogeneous solution as well.

L[y] = 0
L[y] = 0
y (n) + pn−1 y (n−1) + · · · + p0 y = 0
ȳ (n) + pn−1 ȳ (n−1) + · · · + p0 ȳ = 0
L [ȳ] = 0

For the same reason, if yp is a particular solution, then yp is a particular solution as well.
Since the real and imaginary parts of a function y are linear combinations of y and ȳ,

y + ȳ y − ȳ
<(y) = , =(y) = ,
2 i2

if y is a homogeneous solution then both <y and =(y) are homogeneous solutions. Likewise, if yp is a particular
solution then <(yp ) is a particular solution.

 
yp + yp f f
L [<(yp )] = L = + =f
2 2 2

Thus we see that the homogeneous solution, the particular solution and the general solution of a linear differential
equation with real-valued coefficients and inhomogeneity can be written in terms of real-valued functions.

760
 Result 18.1.2 The differential equation
dn y dn−1 y dy
L[y] = n + pn−1 (x) n−1 + · · · + p1 (x) + p0 (x)y = f (x)
dx dx dx
with continuous coefficients and inhomogeneity has a general solution of the form

y = yp + c1 y1 + · · · + cn yn

where yp is a particular solution, L[yp ] = f , and the yk are linearly independent homoge-
neous solutions, L[yk ] = 0. If the coefficient functions and inhomogeneity are real-valued,
then the general solution can be written in terms of real-valued functions.

18.2 Transformation to a First Order System

Any linear differential equation can be put in the form of a system of first order differential equations. Consider

y (n) + pn−1 y (n−1) + · · · + p0 y = f (x).

We introduce the functions,

y1 = y, y2 = y 0 , ,... , yn = y (n−1) .

The differential equation is equivalent to the system

y10 = y2
y20 = y3
.. ..
.=.
0
yn = f (x) − pn−1 yn − · · · − p0 y1 .

761
 The first order system is more useful when numerically solving the differential equation.

Example 18.2.1 Consider the differential equation

y 00 + x2 y 0 + cos x y = sin x.

The corresponding system of first order equations is

y10 = y2
y20 = sin x − x2 y2 − cos x y1 .

18.3 The Wronskian

18.3.1 Derivative of a Determinant.
Before investigating the Wronskian, we will need a preliminary result from matrix theory. Consider an n × n
matrix A whose elements aij (x) are functions of x. We will denote the determinant by ∆[A(x)]. We then have
the following theorem.

Result 18.3.1 Let aij (x), the elements of the matrix A, be differentiable functions of x.
Then
n
d X
∆[A(x)] = ∆k [A(x)]
dx
k=1

where ∆k [A(x)] is the determinant of the matrix A with the k th row replaced by the
derivative of the k th row.

762
Example 18.3.1 Consider the the matrix
 
x x2
A(x) =
x2 x4
The determinant is x5 − x4 thus the derivative of the determinant is 5x4 − 4x3 . To check the theorem,
 
d d  x x2 
∆[A(x)] =
dx dx x2 x4 
   
 1 2x  x x2 
=  2 4  +  
x x 2x 4x3 
= x4 − 2x3 + 4x4 − 2x3
= 5x4 − 4x3 .

18.3.2 The Wronskian of a Set of Functions.

A set of functions {y1 , y2 , . . . , yn } is linearly dependent on an interval if there are constants c1 , . . . , cn not all zero
such that
c1 y1 + c2 y2 + · · · + cn yn = 0 (18.2)
identically on the interval. The set is linearly independent if all of the constants must be zero to satisfy c1 y1 +
· · · cn yn = 0 on the interval.
Consider a set of functions {y1 , y2 , . . . , yn } that are linearly dependent on a given interval and n − 1 times
differentiable. There are a set of constants, not all zero, that satisfy equation 18.2
Differentiating equation 18.2 n − 1 times gives the equations,
c1 y10 + c2 y20 + · · · + cn yn0 = 0
c1 y100 + c2 y200 + · · · + cn yn00 = 0
···
(n−1) (n−1)
c1 y1 + c2 y2 + · · · + cn yn(n−1) = 0.

763
We could write the problem to find the constants as
  
y1 y2 ... yn c1
 y10 y 0
2 . . . y 0  
n   c2 
 00 00
 y1 y2 ... yn00 
  c3  = 0
 
 .. ..

...  . 
 . . . . .   .. 
(n−1) (n−1) (n−1)
y1 y2 . . . yn cn
From linear algebra, we know that this equation has a solution for a nonzero constant vector only if the determinant
of the matrix is zero. Here we define the Wronskian ,W (x), of a set of functions.
 
 y1 y 2 . . . y n

 0 0 0 

 y1 y2 ... yn 
W (x) =  .. ..

... 
 . . ... 
 (n−1) (n−1) (n−1) 

 y1 y2 . . . yn
Thus if a set of functions is linearly dependent on an interval, then the Wronskian is identically zero on that
interval. Alternatively, if the Wronskian is identically zero, then the above matrix equation has a solution for a
nonzero constant vector. This implies that the the set of functions is linearly dependent.
Result 18.3.2 The Wronskian of a set of functions vanishes identically over an interval
if and only if the set of functions is linearly dependent on that interval. The Wronskian of
a set of linearly independent functions does not vanish except possibly at isolated points.

Example 18.3.2 Consider the set, {x, x2 }. The Wronskian is

 
x x2 
W (x) = 
 
1 2x
= 2x2 − x2
= x2 .

764
Thus the functions are independent.

Example 18.3.3 Consider the set {sin x, cos x, eix }. The Wronskian is
 
ix 
 sin x cos x e
 
W (x) =  cos x − sin x i eix  .
− sin x − cos x − eix 
Since the last row is a constant multiple of the first row, the determinant is zero. The functions are dependent.
We could also see this with the identity eix = cos x + i sin x.

18.3.3 The Wronskian of the Solutions to a Differential Equation

Consider the nth order linear homogeneous differential equation
y (n) + pn−1 (x)y (n−1) + · · · + p0 (x)y = 0.
Let {y1 , y2 , . . . , yn } be any set of n linearly independent solutions. Let Y (x) be the matrix such that W (x) =
∆[Y (x)]. Now let’s differentiate W (x).
d
W 0 (x) = ∆[Y (x)]
dx
Xn
= ∆k [Y (x)]
k=1

We note that the all but the last term in this sum is zero. To see this, let’s take a look at the first term.
 0 0

0 
 y1
 0 y 2 · · · y n 
0 0 
 y1 y 2 · · · y n 
∆1 [Y (x)] =  .. .. .. 

...
 . . . 
 (n−1) (n−1) (n−1)

 y1 y2 · · · yn 

765
The first two rows in the matrix are identical. Since the rows are dependent, the determinant is zero.
The last term in the sum is
 
 y1 y2 ··· yn 
 .. .. .. .. 

 . . . . 
∆n [Y (x)] =  (n−2) (n−2) (n−2)  .
 (n) y2 (n) · · · yn (n) 
 y1 
 y
1 y2 · · · yn 
(n) (n−1)
In the last row of this matrix we make the substitution yi = −pn−1 (x)yi − · · · − p0 (x)yi . Recalling that
we can add a multiple of a row to another without changing the determinant, we add p0 (x) times the first row,
and p1 (x) times the second row, etc., to the last row. Thus we have the determinant,
 
 y1 y2 ··· yn 
.. .. .. ..
 
0

. . . .

W (x) = 
 
(n−2) (n−2) (n−2) 
 y1 y2 ··· yn 
−pn−1 (x)y (n−1) −pn−1 (x)y (n−1) · · · −pn−1 (x)yn(n−1) 
 
1 2
 
 y1 y2 ··· yn 
 .. .. ... .. 

 . . . 
= −pn−1 (x)  (n−2) (n−2) (n−2) 
 (n−1) y2(n−1) · · · yn(n−1) 
y1 
y
1 y2 · · · yn 
= −pn−1 (x)W (x)

Thus the Wronskian satisfies the first order differential equation,

W 0 (x) = −pn−1 (x)W (x).

Solving this equation we get a result known as Abel’s formula.

 Z 
W (x) = c exp − pn−1 (x) dx

766
Thus regardless of the particular set of solutions that we choose, we can compute their Wronskian up to a constant
factor.
Result 18.3.3 The Wronskian of any linearly independent set of solutions to the equa-
tion

y (n) + pn−1 (x)y (n−1) + · · · + p0 (x)y = 0

is, (up to a multiplicative constant), given by

 Z 
W (x) = exp − pn−1 (x) dx .

Example 18.3.4 Consider the differential equation

y 00 − 3y 0 + 2y = 0.

The Wronskian of the two independent solutions is

 Z 
W (x) = c exp − −3 dx

= c e3x .

For the choice of solutions { ex , e2x }, the Wronskian is

 x 
e e2x 
W (x) =  x
 = 2 e3x − e3x = e3x .
e 2 e2x 

767
18.4 Well-Posed Problems
Consider the initial value problem for an nth order linear differential equation.
dn y dn−1 y dy
n
+ p n−1 (x) n−1
+ · · · + p1 (x) + p0 (x)y = f (x)
dx dx dx
y(x0 ) = v1 , y 0 (x0 ) = v2 , . . . , y (n−1) (x0 ) = vn
Since the general solution to the differential equation is a linear combination of the n homogeneous solutions plus
the particular solution
y = yp + c1 y1 + c2 y2 + · · · + cn yn ,
the problem to find the constants ci can be written
      
y1 (x0 ) y2 (x0 ) ... yn (x0 ) c1 yp (x0 ) v1
 y10 (x0 ) 0 0 0
y (x
2 0 ) . . . y (x
n 0 ) y (x
  c2   p 0   v2 
    )   
.. .. + . = . .

.. . ..
.   ..     .. 
      
 . . ...
(n−1) (n−1) (n−1) (n−1)
y1 (x0 ) y2 (x0 ) . . . yn (x0 ) cn yp (x0 ) vn
From linear algebra we know that this system of equations has a unique solution only if the determinant of the
matrix is nonzero. Note that the determinant of the matrix is just the Wronskian evaluated at x0 . Thus if the
Wronskian vanishes at x0 , the initial value problem for the differential equation either has no solutions or infinitely
many solutions. Such problems are said to be ill-posed. From Abel’s formula for the Wronskian
 Z 
W (x) = exp − pn−1 (x) dx ,

we see that the only way the Wronskian can vanish is if the value of the integral goes to ∞.

Example 18.4.1 Consider the initial value problem

2 2
y 00 − y 0 + 2 y = 0, y(0) = y 0 (0) = 1.
x x

768
The Wronskian
 Z 
2
W (x) = exp − − dx = exp (2 log x) = x2
x

vanishes at x = 0. Thus this problem is not well-posed.

The general solution of the differential equation is

y = c 1 x + c 2 x2 .

We see that the general solution cannot satisfy the initial conditions. If instead we had the initial conditions
y(0) = 0, y 0 (0) = 1, then there would be an infinite number of solutions.

Example 18.4.2 Consider the initial value problem

2
y 00 − y = 0, y(0) = y 0 (0) = 1.
x2

The Wronskian
 Z 
W (x) = exp − 0 dx = 1

does not vanish anywhere. However, this problem is not well-posed.

The general solution,

y = c1 x−1 + c2 x2 ,

cannot satisfy the initial conditions. Thus we see that a non-vanishing Wronskian does not imply that the problem
is well-posed.

769
 Result 18.4.1 Consider the initial value problem
dn y dn−1 y dy
+ p n−1 (x) + · · · + p 1 (x) + p0 (x)y = 0
dxn dxn−1 dx
y(x0 ) = v1 , y 0 (x0 ) = v2 , . . . , y (n−1) (x0 ) = vn .

If the Wronskian
 Z 
W (x) = exp − pn−1 (x) dx

vanishes at x = x0 then the problem is ill-posed. The problem may be ill-posed even if
the Wronskian does not vanish.

18.5 The Fundamental Set of Solutions

Consider a set of linearly independent solutions {u1 , u2 , . . . , un } to an nth order linear homogeneous differential
equation. This is called the fundamental set of solutions at x0 if they satisfy the relations

u1 (x0 ) = 1 u2 (x0 ) = 0 ... un (x0 ) = 0

u01 (x0 ) = 0 u02 (x0 ) = 1 ... u0n (x0 ) = 0
.. .. ... ..
. . .
(n−1) (n−1) (n−1)
u1 (x0 ) = 0 u2 (x0 ) = 0 . . . un (x0 ) = 1

Knowing the fundamental set of solutions is handy because it makes the task of solving an initial value problem
trivial. Say we are given the initial conditions,

y(x0 ) = v1 , y 0 (x0 ) = v2 , ... , y (n−1) (x0 ) = vn .

770
If the ui ’s are a fundamental set then the solution that satisfies these constraints is just

y = v1 u1 (x) + v2 u2 (x) + · · · + vn un (x).

Of course in general, a set of solutions is not the fundamental set. If the Wronskian of the solutions is
nonzero and finite we can generate a fundamental set of solutions that are linear combinations of our original set.
Consider the case of a second order equation Let {y1 , y2 } be two linearly independent solutions. We will generate
the fundamental set of solutions, {u1 , u2 }.

    
u1 c11 c12 y1
=
u2 c21 c22 y2

For {u1 , u2 } to satisfy the relations that define a fundamental set, it must satisfy the matrix equation

u1 (x0 ) u01 (x0 ) y1 (x0 ) y10 (x0 )

      
c11 c12 1 0
= =
u2 (x0 ) u02 (x0 ) c21 c22 y2 (x0 ) y20 (x0 ) 0 1

−1
y1 (x0 ) y10 (x0 )
  
c11 c12
=
c21 c22 y2 (x0 ) y20 (x0 )

If the Wronskian is non-zero and finite, we can solve for the constants, cij , and thus find the fundamental set of
solutions. To generalize this result to an equation of order n, simply replace all the 2 × 2 matrices and vectors
of length 2 with n × n matrices and vectors of length n. I presented the case of n = 2 simply to save having to
write out all the ellipses involved in the general case. (It also makes for easier reading.)

Example 18.5.1 Two linearly independent solutions to the differential equation y 00 + y = 0 are y1 = eix and
y2 = e−ix .

y1 (0) y10 (0)

   
1 i
=
y2 (0) y20 (0) 1 −i

771
To find the fundamental set of solutions, {u1 , u2 }, at x = 0 we solve the equation
   −1
c11 c12 1 i
=
c21 c22 1 −i

   
c11 c12 1 i i
=
c21 c22 2i 1 −1

The fundamental set is

eix + e−ix eix − e−ix
u1 = , u2 = .
2 2i
Using trigonometric identities we can rewrite these as

u1 = cos x, u2 = sin x.

Result 18.5.1 The fundamental set of solutions at x = x0 , {u1 , u2 , . . . , un }, to an nth

order linear differential equation, satisfy the relations
u1 (x0 ) = 1 u2 (x0 ) = 0 ... un (x0 ) = 0
u01 (x0 ) = 0 u02 (x0 ) = 1 ... u0n (x0 ) = 0
.. .. ... ..
. . .
(n−1) (n−1) (n−1)
u1 (x0 ) = 0 u2 (x0 ) = 0 . . . un (x0 ) = 1.
If the Wronskian of the solutions is nonzero and finite at the point x0 then you can
generate the fundamental set of solutions from any linearly independent set of solutions.

772
18.6 Adjoint Equations
For the nth order linear differential operator
dn y dn−1 y
L[y] = pn + pn−1 + · · · + p0 y
dxn dxn−1
(where the pj are complex-valued functions) we define the adjoint of L
dn n−1 d
n−1
L∗ [y] = (−1)n (p n y) + (−1) (pn−1 y) + · · · + p0 y.
dxn dxn−1
Here f denotes the complex conjugate of f .

Example 18.6.1
1
L[y] = xy 00 + y 0 + y
x
has the adjoint
d2
 
∗ d 1
L [y] = 2 [xy] − y +y
dx dx x
1 1
= xy 00 + 2y 0 − y 0 + 2 y + y
 x x  
00 1 0 1
= xy + 2 − y + 1 + 2 y.
x x
Taking the adjoint of L∗ yields
d2
    
∗∗ d 1 1
L [y] = 2 [xy] − 2− y + 1+ 2 y
dx dx x x
     
00 0 1 0 1 1
= xy + 2y − 2 − y − y+ 1+ 2 y
x x2 x
1
= xy 00 + y 0 + y.
x

773
Thus by taking the adjoint of L∗ , we obtain the original operator.

In general, L∗∗ = L.

Consider L[y] = pn y (n) + · · · + p0 y. If each of the pk is k times continuously differentiable and u and v are n
times continuously differentiable on some interval, then on that interval

d
vL[u] − uL∗ [v] = B[u, v]
dx
where B[u, v], the bilinear concomitant, is the bilinear form
n
X X
B[u, v] = (−1)j u(k) (pm v)(j) .
m=1 j+k=m−1
j≥0,k≥0

This equation is known as Lagrange’s identity. If L is a second order operator then

d
up1 v + u0 p2 v − u(p2 v)0

vL[u] − uL∗ [v] =
dx
= u00 p2 v + u0 p1 v + u − p2 v 00 + (−2p02 + p1 )v 0 + (−p002 + p01 )v .
 

Example 18.6.2 Verify Lagrange’s identity for the second order operator, L[y] = p2 y 00 + p1 y 0 + p0 y.
 2 
00 0 d d
vL[u] − uL∗ [v] = v(p2 u + p1 u + p0 u) − u (p2 v) − (p1 v) + p0 v
dx2 dx
= v(p2 u00 + p1 u0 + p0 u) − u(p2 v 00 + (2p2 0 − p1 )v 0 + (p2 00 − p1 0 + p0 )v)
= u00 p2 v + u0 p1 v + u − p2 v 00 + (−2p02 + p1 )v 0 + (−p002 + p01 )v .
 

We will not verify Lagrange’s identity for the general case.

774
Integrating Lagrange’s identity on its interval of validity gives us Green’s formula.
Z b   
vL[u] − uL∗ [v] dx = B[u, v]x=b − B[u, v]x=a
a

Result 18.6.1 The adjoint of the operator

dn y dn−1 y
L[y] = pn n + pn−1 n−1 + · · · + p0 y
dx dx
is defined
∗ dn n n−1 d
n−1
L [y] = (−1) (pn y) + (−1) (pn−1 y) + · · · + p0 y.
dxn dxn−1
If each of the pk is k times continuously differentiable and u and v are n times continuously
differentiable, then Lagrange’s identity states
n
d d X X
vL[y] − uL∗ [v] = B[u, v] = (−1)j u(k) (pm v)(j) .
dx dx m=1
j+k=m−1
j≥0,k≥0

Integrating Lagrange’s identity on it’s domain of validity yields Green’s formula,

Z b   
∗
vL[u] − uL [v] dx = B[u, v]x=b − B[u, v]x=a .
a

775
18.7 Exercises
Exercise 18.1
Determine a necessary condition for a second order linear differential equation to be exact.
Determine an equation for the integrating factor for a second order linear differential equation.
Hint, Solution

Exercise 18.2
Show that

y 00 + xy 0 + y = 0

is exact. Find the solution.

Hint, Solution

Nature of Solutions
Exercise 18.3
On what intervals do the following problems have unique solutions?

1. xy 00 + 3y = x

2. x(x − 1)y 00 + 3xy 0 + 4y = 2

3. ex y 00 + x2 y 0 + y = tan x

Hint, Solution

Exercise 18.4
Suppose you are able to find three linearly independent particular solutions u1 (x), u2 (x) and u3 (x) of the second
order linear differential equation L[y] = f (x). What is the general solution?
Hint, Solution

776
Transformation to a First Order System
The Wronskian
Well-Posed Problems
The Fundamental Set of Solutions
Exercise 18.5
Two solutions of y 00 − y = 0 are ex and e−x . Show that the solutions are independent. Find the fundamental set
of solutions at x = 0.
Hint, Solution

Adjoint Equations
Exercise 18.6
Find the adjoint of the Bessel equation of order ν,

x2 y 00 + xy 0 + (x2 − ν 2 )y = 0,

and the Legendre equation of order α,

(1 − x2 )y 00 − 2xy 0 + α(α + 1)y = 0.

Hint, Solution

Exercise 18.7
Find the adjoint of

x2 y 00 − xy 0 + 3y = 0.

Hint, Solution

777
18.8 Hints
Hint 18.1

Hint 18.2

Nature of Solutions
Hint 18.3

Hint 18.4
The difference of any two of the ui ’s is a homogeneous solution.

Transformation to a First Order System

The Wronskian
Well-Posed Problems
The Fundamental Set of Solutions
Hint 18.5

Adjoint Equations
Hint 18.6

778
Hint 18.7

779
18.9 Solutions
Solution 18.1
The second order, linear, homogeneous differential equation is

P (x)y 00 + Q(x)y 0 + R(x)y = 0. (18.3)

The second order, linear, homogeneous, exact differential equation is

 
d dy d
P (x) + [f (x)y] = 0. (18.4)
dx dx dx

P (x)y 00 + (P 0 (x) + f (x)) y 0 + f 0 (x)y = 0

Equating the coefficients of Equations 18.3 and 18.4 yields the set of equations,

P 0 (x) + f (x) = Q(x), f 0 (x) = R(x).

We differentiate the first equation and substitute in the expression for f 0 (x) from the second equation to determine
a necessary condition for exactness.

P 00 (x) − Q0 (x) + R(x) = 0

We multiply Equation 18.3 by the integrating factor µ(x) to obtain,

µ(x)P (x)y 00 + µ(x)Q(x)y 0 + µ(x)R(x)y = 0. (18.5)

The corresponding exact equation is of the form,

 
d dy d
µ(x)P (x) + [f (x)y] = 0. (18.6)
dx dx dx

780
 µ(x)P (x)y 00 + (µ0 (x)P (x) + µ(x)P 0 (x) + f (x)) y 0 + f 0 (x)y = 0

Equating the coefficients of Equations 18.5 and 18.6 yields the set of equations,

µ0 P + µP 0 + f = µQ, f 0 = µR.

We differentiate the first equation and substitute in the expression for f 0 from the second equation to find a
differential equation for µ(x).

µ00 P + µ0 P 0 + µ0 P 0 + µP 00 + µR = µ0 Q + µQ0
P µ00 + (2P 0 − Q)µ0 + (P 00 − Q0 + R)µ = 0

Solution 18.2
We consider the differential equation,

y 00 + xy 0 + y = 0.

Since

(1)00 − (x)0 + 1 = 0

we see that this is an exact equation. We rearrange terms to form exact derivatives and then integrate.

(y 0 )0 + (xy)0 = 0
y 0 + xy = c
d h x2 /2 i 2
e y = c ex /2
dx
Z
−x2 /2 2 2
y = ce ex /2 dx + d e−x /2

781
Nature of Solutions
Solution 18.3
Consider the initial value problem,

y 00 + p(x)y 0 + q(x)y = f (x),

y(x0 ) = y0 , y 0 (x0 ) = y1 .

If p(x), q(x) and f (x) are continuous on an interval (a . . . b) with x0 ∈ (a . . . b), then the problem has a unique
solution on that interval.

1.

xy 00 + 3y = x
3
y 00 + y = 1
x
Unique solutions exist on the intervals (−∞ . . . 0) and (0 . . . ∞).
2.

x(x − 1)y 00 + 3xy 0 + 4y = 2

3 4 2
y 00 + y0 + y=
x−1 x(x − 1) x(x − 1)
Unique solutions exist on the intervals (−∞ . . . 0), (0 . . . 1) and (1 . . . ∞).
3.

ex y 00 + x2 y 0 + y = tan x
y 00 + x2 e−x y 0 + e−x y = e−x tan x
 
(2n−1)π (2n+1)π
Unique solutions exist on the intervals 2
... 2 for n ∈ Z.

782
Solution 18.4
We know that the general solution is

y = yp + c1 y1 + c2 y2 ,

where yp is a particular solution and y1 and y2 are linearly independent homogeneous solutions. Since yp can be
any particular solution, we choose yp = u1 . Now we need to find two homogeneous solutions. Since L[ui ] = f (x),
L[u1 − u2 ] = L[u2 − u3 ] = 0. Finally, we note that since the ui ’s are linearly independent, y1 = u1 − u2 and
y2 = u2 − u3 are linearly independent. Thus the general solution is

y = u1 + c1 (u1 − u2 ) + c2 (u2 − u3 ).

Transformation to a First Order System

The Wronskian
Well-Posed Problems
The Fundamental Set of Solutions
Solution 18.5
The Wronskian of the solutions is
 x 
e e−x 
W (x) =  x
 = −2.
e − e−x 

Since the Wronskian is nonzero, the solutions are independent.

The fundamental set of solutions, {u1 , u2 }, is a linear combination of ex and e−x .
    x 
u1 c11 c12 e
=
u2 c21 c22 e−x

783
The coefficients are
   0 −1
c11 c12 e e0
=
c21 c22 e−0 − e−0
 −1
1 1
=
1 −1
 
1 −1 −1
=
−2 −1 1
 
1 1 1
=
2 1 −1

1 1
u1 = ( ex + e−x ), u2 = ( ex − e−x ).
2 2
The fundamental set of solutions at x = 0 is

{cosh x, sinh x}

Adjoint Equations
Solution 18.6
1. The Bessel equation of order ν is

x2 y 00 + xy 0 + (x2 − ν 2 )y = 0.

The adjoint equation is

x2 µ00 + (4x − x)µ0 + (2 − 1 + x2 − ν 2 )µ = 0

x2 µ00 + 3xµ0 + (1 + x2 − ν 2 )µ = 0.

784
 2. The Legendre equation of order α is

(1 − x2 )y 00 − 2xy 0 + α(α + 1)y = 0

The adjoint equation is

(1 − x2 )µ00 + (−4x + 2x)µ0 + (−2 + 2 + α(α + 1))µ = 0

(1 − x2 )µ00 − 2xµ0 + α(α + 1)µ = 0

Solution 18.7
The adjoint of

x2 y 00 − xy 0 + 3y = 0

is
d2 2 d
2
(x y) + (xy) + 3y = 0
dx dx
(x2 y 00 + 4xy 0 + 2y) + (xy 0 + y) + 3y = 0
x2 y 00 + 5xy 0 + 6y = 0.

785
Chapter 19

Techniques for Linear Differential Equations

My new goal in life is to take the meaningless drivel out of human interaction.
-Dave Ozenne

The nth order linear homogeneous differential equation has the form
y (n) + an−1 (x)y (n−1) + · · · + a1 (x)y 0 + a0 (x)y = 0.
In general it is not possible to solve second order and higher linear differential equations. In this chapter we will
examine equations that have special forms which allow us to either reduce the order of the equation or solve it.

19.1 Constant Coefficient Equations

The nth order constant coefficient differential equation has the form
y (n) + an−1 y (n−1) + · · · + a1 y 0 + a0 y = 0.
We will find that solving a constant coefficient differential equation is no more difficult than finding the roots of
a polynomial.

786
19.1.1 Second Order Equations
Factoring the Differential Equation. Consider the second order constant coefficient differential equation

y 00 + 2ay 0 + by = 0. (19.1)

Just as we can factor the polynomial,

λ2 + 2aλ + b = (λ − α)(λ − β), (19.2)

where
√ √
α = −a + a2 − b and β = −a − a2 − b,

we can factor the differential equation.

 2    
d d d d
+ 2a +b y = −α −β y
dx2 dx dx dx

Once we have factored the differential equation, we can solve it by solving a series of of two first order differential
d


equations. We set u = dx − β y to obtain a first order equation,
 
d
− α u = 0,
dx

which has the solution

u = c1 eαx .

To find the solution of Equation 19.1, we solve

 
d
− β y = u = c1 eαx .
dx

787
We multiply by the integrating factor and integrate.
d
e−βx y = c1 e(α−β)x


dx Z
βx
y = c1 e e(α−β)x dx + c2 eβx

We first consider the case when α and β are distinct.

1
y = c1 eβx e(α−β)x + c2 eβx
α−β
We choose new constants to write the solution in a better form.
y = c1 eαx + c2 eβx
Now we consider the case α = β.
Z
αx
y = c1 e 1 dx + c2 eαx

y = c1 x eαx + c2 eαx .
The solution of Equation 19.1 is
(
c1 eαx + c2 eβx , α 6= β,
y= αx αx
c1 e + c2 x e , α = β.

Example 19.1.1 Consider the differential equation: y 00 + y = 0. We factor the equation.

  
d d
−i +i y =0
dx dx
The general solution of the differential equation is
y = c1 eix + c2 e−ix .

788
Example 19.1.2 Consider the differential equation: y 00 = 0. We factor the equation.
  
d d
−0 −0 y =0
dx dx
The general solution of the differential equation is

y = c1 e0x + c2 x e0x
y = c1 + c2 x.

Substituting the Form of the Solution into the Differential Equation. Note that if we substitute y = eλx
into the differential equation 19.1, we will obtain the quadratic polynomial equation 19.2 for λ.

y 00 + 2ay 0 + by = 0
λ2 eλx + 2aλ eλx + b eλx = 0
λ2 + 2aλ + b = 0.

This gives us a superficially different method for solving constant coefficient equations. We substitute y = eλx
into the differential equation. Let α and β be the roots of the quadratic in λ. If the roots are distinct, then the
linearly independent solutions are y1 = eαx and y2 = eβx . If the quadratic has a double root at λ = α, then the
linearly independent solutions are y1 = eαx and y2 = x eαx .

Example 19.1.3 Consider the equation

y 00 − 3y 0 + 2y = 0.

The substitution y = eλx yields

λ2 − 3λ + 2 = (λ − 1)(λ − 2) = 0.

Thus the solutions are ex and e2x .

789
Example 19.1.4 Consider the equation

y 00 − 2y 0 + 4y = 0.

The substitution y = eλx yields

λ2 − 2λ + 4 = (λ − 2)2 = 0.

Thus the solutions are e2x and x e2x .

Shift Invariance. Note that if u(x) is a solution of a constant coefficient equation, then u(x + c) is also a
solution. This is useful in applying initial or boundary conditions.

Example 19.1.5 Consider the problem

y 00 − 3y 0 + 2y = 0, y(0) = a, y 0 (0) = b.

We know that the general solution is

y = c1 ex + c2 e2x .

Applying the initial conditions, we obtain the equations,

c1 + c2 = a, c1 + 2c2 = b.

The solution is

y = (2a − b) ex + (b − a) e2x .

Now suppose we wish to solve the same differential equation with the boundary conditions y(1) = a and y 0 (1) = b.
All we have to do is shift the solution to the right.

y = (2a − b) ex−1 + (b − a) e2(x−1) .

790
 Result 19.1.1 . Consider the second order constant coefficient equation

y 00 + 2ay 0 + by = 0.

The general solution of this differential equation is

  √ √ 
−ax a2 −b x − a2 −b x
e

 c1 e + c2 e if a2 > b,
√ √
y = e−ax c1 cos( b − a2 x) + c2 sin( b − a2 x)


 if a2 < b,
 e−ax (c + c x)

if a2 = b.
1 2

The fundamental set of solutions at x = 0 is

n  √
−ax cosh( a2 − b x) + √ a
√  √ o
 e sinh( a2 − b x) , e−ax √ 1 sinh( a 2 − b x) if a2 > b,
 a2 −b a2 −b
n
−ax
 √ a
√ 
−ax 1
√ o
 e cos( b − a2 x) + √
b−a2
sin( b − a2 x) , e √
b−a2
sin( b − a2 x) if a2 < b,

 −ax
{(1 + ax) e , x e }−ax if a2 = b.

To obtain the fundamental set of solutions at the point x = ξ, substitute (x − ξ) for x

in the above solutions.

19.1.2 Higher Order Equations

The constant coefficient equation of order n has the form

L[y] = y (n) + an−1 y (n−1) + · · · + a1 y 0 + a0 y = 0. (19.3)

791
The substitution y = eλx will transform this differential equation into an algebraic equation.

L[ eλx ] = λn eλx + an−1 λn−1 eλx + · · · + a1 λ eλx + a0 eλx = 0

λn + an−1 λn−1 + · · · + a1 λ + a0 eλx = 0

λn + an−1 λn−1 + · · · + a1 λ + a0 = 0

Assume that the roots of this equation, λ1 , . . . , λn , are distinct. Then the n linearly independent solutions of
Equation 19.3 are

eλ1 x , . . . , eλn x .

If the roots of the algebraic equation are not distinct then we will not obtain all the solutions of the differential
equation. Suppose that λ1 = α is a double root. We substitute y = eλx into the differential equation.

L[ eλx ] = [(λ − α)2 (λ − λ3 ) · · · (λ − λn )] eλx = 0

Setting λ = α will make the left side of the equation zero. Thus y = eαx is a solution. Now we differentiate both
sides of the equation with respect to λ and interchange the order of differentiation.
 
d λx d λx
= L x eλx
 
L[ e ] = L e
dλ dλ
 λx 
Let p(λ) = (λ − λ3 ) · · · (λ − λn ). We calculate L x e by applying L and then differentiating with respect to λ.
d
L x eλx = L[ eλx ]
 
dλ
d
= [(λ − α)2 (λ − λ3 ) · · · (λ − λn )] eλx
dλ
d
= [(λ − α)2 p(λ)] eλx
dλ
= 2(λ − α)p(λ) + (λ − α)2 p0 (λ) + (λ − α)2 p(λ)x eλx


= (λ − α) [2p(λ) + (λ − α)p0 (λ) + (λ − α)p(λ)x] eλx

792
Since setting λ = α will make this expression zero, L[x eαx ] = 0, x eαx is a solution of Equation 19.3. You can
verify that eαx and x eαx are linearly independent. Now we have generated all of the solutions for the differential
equation.
If λ = α is a root of multiplicity m then by repeatedly differentiating with respect to λ you can show that the
corresponding solutions are
eαx , x eαx , x2 eαx , . . . , xm−1 eαx .

Example 19.1.6 Consider the equation

y 000 − 3y 0 + 2y = 0.
The substitution y = eλx yields
λ3 − 3λ + 2 = (λ − 1)2 (λ + 2) = 0.
Thus the general solution is

y = c1 ex + c2 x ex + c3 e−2x .

19.1.3 Real-Valued Solutions

If the coefficients of the differential equation are real, then the solution can be written in terms of real-valued
functions (Result 18.1.2). For a real root λ = α of the polynomial in λ, the corresponding solution, y = eαx , is
real-valued.
Now recall that the complex roots of a polynomial with real coefficients occur in complex conjugate pairs.
Assume that α ± iβ are roots of
λn + an−1 λn−1 + · · · + a1 λ + a0 = 0.
The corresponding solutions of the differential equation are e(α+iβ)x and e(α−iβ)x . Note that the linear combinations
e(α+iβ)x + e(α−iβ)x e(α+iβ)x − e(α−iβ)x
= eαx cos(βx), = eαx sin(βx),
2 i2

793
are real-valued solutions of the differential equation. We could also obtain real-valued solution by taking the real
and imaginary parts of either e(α+iβ)x or e(α−iβ)x .

< e(α+iβ)x = eαx cos(βx), = e(α+iβ)x = eαx sin(βx)

Result 19.1.2 Consider the nth order constant coefficient equation

dn y dn−1 y dy
+ an−1 + · · · + a1 + a0 y = 0.
dxn dxn−1 dx
Let the factorization of the algebraic equation obtained with the substitution y = eλx be

(λ − λ1 )m1 (λ − λ2 )m2 · · · (λ − λp )mp = 0.

A set of linearly independent solutions is given by

{ eλ1 x , x eλ1 x , . . . , xm1 −1 eλ1 x , . . . , eλp x , x eλp x , . . . , xmp −1 eλp x }.

If the coefficients of the differential equation are real, then we can find a real-valued set
of solutions.

Example 19.1.7 Consider the equation

d4 y d2 y
+ 2 + y = 0.
dx4 dx2

The substitution y = eλx yields

λ4 + 2λ2 + 1 = (λ − i)2 (λ + i)2 = 0.

794
Thus the linearly independent solutions are
eix , x eix , e−ix and x e−ix .
Noting that
eix = cos(x) + i sin(x),
we can write the general solution in terms of sines and cosines.
y = c1 cos x + c2 sin x + c3 x cos x + c4 x sin x

Example 19.1.8 Consider the equation

y 00 − 2y 0 + 2y = 0.
The substitution y = eλx yields
λ2 − 2λ + 2 = (λ − 1 − i)(λ − 1 + i) = 0.
The linearly independent solutions are
e(1+i)x , and e(1−i)x .
We can write the general solution in terms of real functions.
y = c1 ex cos x + c2 ex sin x

19.2 Euler Equations

Consider the equation
d2 y dy
L[y] = x2 + ax + by = 0, x > 0.
dx2 dx

795
Let’s say, for example, that y has units of distance and x has units of time. Note that each term in the differential
equation has the same dimension.

(distance) (distance)
(time)2 2
= (time) = (distance)
(time) (time)

Thus this is a second order Euler, or equidimensional equation. We know that the first order Euler equation,
xy 0 + ay = 0, has the solution y = cxa . Thus for the second order equation we will try a solution of the form
y = xλ . The substitution y = xλ will transform the differential equation into an algebraic equation.

d2 λ d λ
L[xλ ] = x2 [x ] + ax [x ] + bxλ = 0
dx2 dx
λ(λ − 1)xλ + aλxλ + bxλ = 0
λ(λ − 1) + aλ + b = 0

Factoring yields

(λ − λ1 )(λ − λ2 ) = 0.

If the two roots, λ1 and λ2 , are distinct then the general solution is

y = c1 xλ1 + c2 xλ2 .

If the roots are not distinct, λ1 = λ2 = λ, then we only have the one solution, y = xλ . To generate the other
solution we use the same approach as for the constant coefficient equation. We substitute y = xλ into the
differential equation and differentiate with respect to λ.

d d
L[xλ ] = L[ xλ ]
dλ dλ
= L[ln x xλ ]

796
Note that
d λ d λ ln x
x = e = ln x eλ ln x = ln x xλ .
dλ dλ
Now we apply L and then differentiate with respect to λ.
d d
L[xλ ] = (λ − α)2 xλ
dλ dλ
= 2(λ − α)xλ + (λ − α)2 ln x xλ

Equating these two results,

L[ln x xλ ] = 2(λ − α)xλ + (λ − α)2 ln x xλ .

Setting λ = α will make the right hand side zero. Thus y = ln x xα is a solution.
If you are in the mood for a little algebra you can show by repeatedly differentiating with respect to λ that if
λ = α is a root of multiplicity m in an nth order Euler equation then the associated solutions are

xα , ln x xα , (ln x)2 xα , . . . , (ln x)m−1 xα .

Example 19.2.1 Consider the Euler equation

y
xy 00 − y 0 + = 0.
x
The substitution y = xλ yields the algebraic equation

λ(λ − 1) − λ + 1 = (λ − 1)2 = 0.

Thus the general solution is

y = c1 x + c2 x ln x.

797
19.2.1 Real-Valued Solutions

If the coefficients of the Euler equation are real, then the solution can be written in terms of functions that are
real-valued when x is real and positive, (Result 18.1.2). If α ± iβ are the roots of

λ(λ − 1) + aλ + b = 0

then the corresponding solutions of the Euler equation are

xα+iβ and xα−iβ .

We can rewrite these as

xα eiβ ln x and xα e−iβ ln x .

Note that the linear combinations

xα eiβ ln x + xα e−iβ ln x xα eiβ ln x − xα e−iβ ln x

= xα cos(β ln x), and = xα sin(β ln x),
2 2i

are real-valued solutions when x is real and positive. Equivalently, we could take the real and imaginary parts of
either xα+iβ or xα−iβ .

< xα eiβ ln x = xα cos(β ln x), = xα eiβ ln x = xα sin(β ln x)

798
 Result 19.2.1 Consider the second order Euler equation

x2 y 00 + (2a + 1)xy 0 + by = 0.

The general solution of this differential equation is

  √ √ 
−a a2 −b − a2 −b
x

 c1 x + c2 x if a2 > b,
√ √
y = x−a c1 cos b − a2 ln x + c2 sin b − a2 ln x




 if a2 < b,
x−a (c + c ln x)

if a2 = b.
1 2

The fundamental set of solutions at x = ξ is

n  −a  √  √ 
x 2 − b ln x + √ a 2 − b ln x

 ξ
cosh a sinh a ,
√ ξ a2 −b ξ


  −a o
√ ξ
 x
sinh a2 − b ln xξ if a2 > b,



 ξ a2 −b

n 
x
−a   √  √ 
cos b − a2 ln x + √ a sin b − a2 ln xξ ,
y= ξ ξ b−a 2
  −a √ o
√ ξ
 x
sin b − a2 ln xξ if a2 < b,




 ξ b−a2
  −a    −a 
 x
1 + a ln ξ , xξ
x
ξ ln xξ if a2 = b.


 ξ

Example 19.2.2 Consider the Euler equation

x2 y 00 − 3xy 0 + 13y = 0.

The substitution y = xλ yields

λ(λ − 1) − 3λ + 13 = (λ − 2 − i3)(λ − 2 + i3) = 0.

799
The linearly independent solutions are

x2+i3 , x2−i3 .


We can put this in a more understandable form.

x2+i3 = x2 ei3 ln x = x2 cos(3 ln x) + x2 sin(3 ln x)

We can write the general solution in terms of real-valued functions.

y = c1 x2 cos(3 ln x) + c2 x2 sin(3 ln x)

Result 19.2.2 Consider the nth order Euler equation

n n−1
nd y n−1 d y dy
x + an−1 x + · · · + a1 x + a0 y = 0.
dxn dxn−1 dx
Let the factorization of the algebraic equation obtained with the substitution y = xλ be

(λ − λ1 )m1 (λ − λ2 )m2 · · · (λ − λp )mp = 0.

A set of linearly independent solutions is given by

{xλ1 , ln x xλ1 , . . . , (ln x)m1 −1 xλ1 , . . . , xλp , ln x xλp , . . . , (ln x)mp −1 xλp }.

If the coefficients of the differential equation are real, then we can find a set of solutions
that are real valued when x is real and positive.

800
19.3 Exact Equations
Exact equations have the form

d
F (x, y, y 0 , y 00 , . . . ) = f (x).
dx
If you can write an equation in the form of an exact equation, you can integrate to reduce the order by one, (or
solve the equation for first order). We will consider a few examples to illustrate the method.

Example 19.3.1 Consider the equation

y 00 + x2 y 0 + 2xy = 0.

We can rewrite this as

d 0
y + x2 y = 0.

dx
Integrating yields a first order inhomogeneous equation.

y 0 + x 2 y = c1
R
We multiply by the integrating factor I(x) = exp( x2 dx) to make this an exact equation.

d  x3 /3  3
e y = c1 ex /3
dx Z
3 /3 3 /3
ex y = c1 ex dx + c2
Z
3 /3 3 /3 3 /3
y = c1 e−x ex dx + c2 e−x

801
 Result 19.3.1 If you can write a differential equation in the form
d
F (x, y, y 0 , y 00 , . . . ) = f (x),
dx
then you can integrate to reduce the order of the equation.
Z
F (x, y, y 0 , y 00 , . . . ) = f (x) dx + c

19.4 Equations Without Explicit Dependence on y

Example 19.4.1 Consider the equation
√
y 00 + xy 0 = 0.
This is a second order equation for y, but note that it is a first order equation for y 0 . We can solve directly for y 0 .
   
d 2 3/2 0
exp x y =0
dx 3
 
0 2 3/2
y = c1 exp − x
3
Now we just integrate to get the solution for y.
Z  
2 3/2
y = c1 exp − x dx + c2
3

Result 19.4.1 If an nth order equation does not explicitly depend on y then you can
consider it as an equation of order n − 1 for y 0 .

802
19.5 Reduction of Order
Consider the second order linear equation
L[y] ≡ y 00 + p(x)y 0 + q(x)y = f (x).
Suppose that we know one homogeneous solution y1 . We make the substitution y = uy1 and use that L[y1 ] = 0.
L[uy1 ] = 0u00 y1 + 2u0 y10 + uy100 + p(u0 y1 + uy10 ) + quy1 = 0
u00 y1 + u0 (2y10 + py1 ) + u(y100 + py10 + qy1 ) = 0
u00 y1 + u0 (2y10 + py1 ) = 0
Thus we have reduced the problem to a first order equation for u0 . An analogous result holds for higher order
equations.

Result 19.5.1 Consider the nth order linear differential equation

y (n) + pn−1 (x)y (n−1) + · · · + p1 (x)y 0 + p0 (x)y = f (x).

Let y1 be a solution of the homogeneous equation. The substitution y = uy1 will trans-
form the problem into an (n − 1)th order equation for u0 . For the second order problem

y 00 + p(x)y 0 + q(x)y = f (x)

this reduced equation is

u00 y1 + u0 (2y10 + py1 ) = f (x).

Example 19.5.1 Consider the equation

y 00 + xy 0 − y = 0.

803
By inspection we see that y1 = x is a solution. We would like to find another linearly independent solution. The
substitution y = xu yields
xu00 + (2 + x2 )u0 = 0
 
00 2
u + + x u0 = 0
x

The integrating factor is I(x) = exp(2 ln x + x2 /2) = x2 exp(x2 /2).

d  2 x2 /2 0 
x e u =0
dx
2
u0 = c1 x−2 e−x /2
Z
2
u = c1 x−2 e−x /2 dx + c2
Z
2
y = c1 x x−2 e−x /2 dx + c2 x

Thus we see that a second solution is

Z
2 /2
y2 = x x−2 e−x dx.

19.6 *Reduction of Order and the Adjoint Equation

Let L be the linear differential operator
dn y dn−1 y
L[y] = pn + p n−1 + · · · + p0 y,
dxn dxn−1
where each pj is a j times continuously differentiable complex valued function. Recall that the adjoint of L is
dn n−1 d
n−1
L∗ [y] = (−1)n (p n y) + (−1) (pn−1 y) + · · · + p0 y.
dxn dxn−1

804
If u and v are n times continuously differentiable, then Lagrange’s identity states
d
vL[u] − uL∗ [v] = B[u, v],
dx
where
n
X X
B[u, v] = (−1)j u(k) (pm v)(j) .
m=1 j+k=m−1
j≥0,k≥0

For second order equations,

B[u, v] = up1 v + u0 p2 v − u(p2 v)0 .
(See Section 18.6.)
If we can find a solution to the homogeneous adjoint equation, L∗ [y] = 0, then we can reduce the order of the
equation L[y] = f (x). Let ψ satisfy L∗ [ψ] = 0. Substituting u = y, v = ψ into Lagrange’s identity yields
d
ψL[y] − yL∗ [ψ] = B[y, ψ]
dx
d
ψL[y] = B[y, ψ].
dx
The equation L[y] = f (x) is equivalent to the equation
d
B[y, ψ] = ψf
dx Z
B[y, ψ] = ψ(x)f (x) dx,

which is a linear equation in y of order n − 1.

Example 19.6.1 Consider the equation

L[y] = y 00 − x2 y 0 − 2xy = 0.

805
Method 1. Note that this is an exact equation.
d 0
(y − x2 y) = 0
dx
y 0 − x2 y = c 1
d  −x3 /3  3
e y = c1 e−x /3
dx
Z
x3 /3 3 3
y = c1 e e−x /3 dx + c2 ex /3

Method 2. The adjoint equation is

L∗ [y] = y 00 + x2 y 0 = 0.

By inspection we see that ψ = (constant) is a solution of the adjoint equation. To simplify the algebra we will
choose ψ = 1. Thus the equation L[y] = 0 is equivalent to

B[y, 1] = c1
d d
y(−x2 ) + [y](1) − y [1] = c1
dx dx
0 2
y − x y = c1 .

By using the adjoint equation to reduce the order we obtain the same solution as with Method 1.

806
19.7 Exercises
Constant Coefficient Equations
Exercise 19.1 (mathematica/ode/techniques linear/constant.nb)
Find the solution of each one of the following initial value problems. Sketch the graph of the solution and describe
its behavior as t increases.
1. 6y 00 − 5y 0 + y = 0, y(0) = 4, y 0 (0) = 0

2. y 00 − 2y 0 + 5y = 0, y(π/2) = 0, y 0 (π/2) = 2

3. y 00 + 4y 0 + 4y = 0, y(−1) = 2, y 0 (−1) = 1
Hint, Solution

Exercise 19.2 (mathematica/ode/techniques linear/constant.nb)

Substitute y = eλx to find two linearly independent solutions to

y 00 − 4y 0 + 13y = 0.

that are real-valued when x is real-valued.

Hint, Solution

Exercise 19.3 (mathematica/ode/techniques linear/constant.nb)

Find the general solution to

y 000 − y 00 + y 0 − y = 0.

Write the solution in terms of functions that are real-valued when x is real-valued.
Hint, Solution

807
Exercise 19.4
Substitute y = eλx to find the fundamental set of solutions at x = 0 for the equations:

1. y 00 + y = 0,

2. y 00 − y = 0,

3. y 00 = 0.

What are the fundamental sets of solutions at x = 1 for these equations.

Hint, Solution

Exercise 19.5
Find the general solution of

y 00 + 2ay 0 + by = 0

for a, b ∈ R. There are three distinct forms of the solution depending on the sign of a2 − b.
Hint, Solution

Exercise 19.6
Find the fundamental set of solutions of

y 00 + 2ay 0 + by = 0

at the point x = 0, for a, b ∈ R. Use the general solutions obtained in Exercise 19.5.
Hint, Solution

Exercise 19.7
Consider a ball of mass m hanging by an ideal spring of spring constant k. The ball is suspended in a fluid
which damps the motion. This resistance has a coefficient of friction, µ. Find the differential equation for the

808
displacement of the mass from its equilibrium position by balancing forces. Denote this displacement by y(t). If
the damping force is weak, the mass will have a decaying, oscillatory motion. If the damping force is strong, the
mass will not oscillate. The displacement will decay to zero. The value of the damping which separates these two
behaviors is called critical damping.
Find the solution which satisfies the initial conditions y(0) = 0, y 0 (0) = 1. Use the solutions obtained in
Exercise 19.6 or refer to Result 19.1.1.
Consider the case m = k = 1. Find the coefficient of friction for which the displacement of the mass decays
most rapidly. Plot the displacement for strong, weak and critical damping.
Hint, Solution

Exercise 19.8
Show that y = c cos(x − φ) is the general solution of y 00 + y = 0 where c and φ are constants of integration. (It
is not sufficient to show that y = c cos(x − φ) satisfies the differential equation. y = 0 satisfies the differential
equation, but is is certainly not the general solution.) Find constants c and φ such that y = sin(x).
Is y = c cosh(x − φ) the general solution of y 00 − y = 0? Are there constants c and φ such that y = sinh(x)?
Hint, Solution

Exercise 19.9 (mathematica/ode/techniques linear/constant.nb)

Let y(t) be the solution of the initial-value problem

y 00 + 5y 0 + 6y = 0; y(0) = 1, y 0 (0) = V.

For what values of V does y(t) remain nonnegative for all t > 0?
Hint, Solution

Exercise 19.10 (mathematica/ode/techniques linear/constant.nb)

Find two linearly independent solutions of

y 00 + sign (x)y = 0, −∞ < x < ∞.

809
where sign (x) = ±1 according as x is positive or negative. (The solution should be continuous and have a
continuous first derivative.)
Hint, Solution

Euler Equations
Exercise 19.11
Find the general solution of

x2 y 00 + xy 0 + y = 0, x > 0.

Hint, Solution

Exercise 19.12
Substitute y = xλ to find the general solution of

x2 y 00 − 2xy + 2y = 0.

Hint, Solution

Exercise 19.13 (mathematica/ode/techniques linear/constant.nb)

Substitute y = xλ to find the general solution of

1
xy 000 + y 00 + y 0 = 0.
x
Write the solution in terms of functions that are real-valued when x is real-valued and positive.
Hint, Solution

810
Exercise 19.14
Find the general solution of

x2 y 00 + (2a + 1)xy 0 + by = 0.

Hint, Solution

Exercise 19.15
Show that

eαx − e−αx
y1 = eax , y2 = lim
α→a α
are linearly indepedent solutions of

y 00 − a2 y = 0

for all values of a. It is common to abuse notation and write the second solution as
eax − e−ax
y2 =
a
where the limit is taken if a = 0. Likewise show that
xa − x−a
y 1 = xa , y2 =
a
are linearly indepedent solutions of

x2 y 00 + xy 0 − a2 y = 0

for all values of a.

Hint, Solution

811
Exercise 19.16 (mathematica/ode/techniques linear/constant.nb)
Find two linearly independent solutions (i.e., the general solution) of

(a) x2 y 00 − 2xy 0 + 2y = 0, (b) x2 y 00 − 2y = 0, (c) x2 y 00 − xy 0 + y = 0.

Hint, Solution

Exact Equations
Exercise 19.17
Solve the differential equation

y 00 + y 0 sin x + y cos x = 0.

Hint, Solution

Equations Without Explicit Dependence on y

Reduction of Order
Exercise 19.18
Consider

(1 − x2 )y 00 − 2xy 0 + 2y = 0, −1 < x < 1.

Verify that y = x is a solution. Find the general solution.

Hint, Solution

Exercise 19.19
Consider the differential equation

x+1 0 1
y 00 − y + y = 0.
x x

812
 x+1 1
Since the coefficients sum to zero, (1 − x
+ x
= 0), y = ex is a solution. Find another linearly independent
solution.
Hint, Solution

Exercise 19.20
One solution of

(1 − 2x)y 00 + 4xy 0 − 4y = 0

is y = x. Find the general solution.

Hint, Solution

Exercise 19.21
Find the general solution of

(x − 1)y00 − xy0 + y = 0,

given that one solution is y = ex . (you may assume x > 1)

Hint, Solution

*Reduction of Order and the Adjoint Equation

813
19.8 Hints
Constant Coefficient Equations
Hint 19.1

Hint 19.2

Hint 19.3
It is a constant coefficient equation.

Hint 19.4
Use the fact that if u(x) is a solution of a constant coefficient equation, then u(x + c) is also a solution.

Hint 19.5
Substitute y = eλx into the differential equation.

Hint 19.6
The fundamental set of solutions is a linear combination of the homogeneous solutions.

Hint 19.7
The force on the mass due to the spring is −ky(t). The frictional force is −µy 0 (t).
Note that the initial conditions describe the second fundamental solution at t = 0.
Note that for large t, t eαt is much small than eβt if α < β. (Prove this.)

814
Hint 19.8
By definition, the general solution of a second order differential equation is a two parameter family of functions
that satisfies the differential equation. The trigonometric identities in Appendix Q may be useful.

Hint 19.9

Hint 19.10

Euler Equations
Hint 19.11

Hint 19.12

Hint 19.13

Hint 19.14
Substitute y = xλ into the differential equation. Consider the three cases: a2 > b, a2 < b and a2 = b.

Hint 19.15

815
Hint 19.16

Exact Equations
Hint 19.17
It is an exact equation.

Equations Without Explicit Dependence on y

Reduction of Order
Hint 19.18

Hint 19.19
Use reduction of order to find the other solution.

Hint 19.20
Use reduction of order to find the other solution.

Hint 19.21

*Reduction of Order and the Adjoint Equation

816
19.9 Solutions
Constant Coefficient Equations
Solution 19.1
1. We consider the problem

6y 00 − 5y 0 + y = 0, y(0) = 4, y 0 (0) = 0.

We make the substitution y = eλx in the differential equation.

6λ2 − 5λ + 1 = 0
(2λ − 1)(3λ − 1) = 0
 
1 1
λ= ,
3 2

The general solution of the differential equation is

y = c1 et/3 + c2 et/2 .

We apply the initial conditions to determine the constants.

c1 c2
c1 + c2 = 4, + =0
3 2
c1 = 12, c2 = −8

The solution subject to the initial conditions is

y = 12 et/3 − 8 et/2 .

The solution is plotted in Figure 19.1. The solution tends to −∞ as t → ∞.

817
 1 2 3 4 5
-5
-10
-15
-20
-25
-30

Figure 19.1: The solution of 6y 00 − 5y 0 + y = 0, y(0) = 4, y 0 (0) = 0.

2. We consider the problem

y 00 − 2y 0 + 5y = 0, y(π/2) = 0, y 0 (π/2) = 2.
We make the substitution y = eλx in the differential equation.
λ2 − 2λ + 5 = 0
√
λ=1± 1−5
λ = {1 + i2, 1 − i2}
The general solution of the differential equation is
y = c1 et cos(2t) + c2 et sin(2t).
We apply the initial conditions to determine the constants.
y(π/2) = 0 ⇒ −c1 eπ/2 = 0 ⇒ c1 = 0
y 0 (π/2) = 2 ⇒ −2c2 eπ/2 = 2 ⇒ c2 = − e−π/2

818
 The solution subject to the initial conditions is

y = − et−π/2 sin(2t).

The solution is plotted in Figure 19.2. The solution oscillates with an amplitude that tends to ∞ as t → ∞.

50
40
30
20
10
3 4 5 6
-10

Figure 19.2: The solution of y 00 − 2y 0 + 5y = 0, y(π/2) = 0, y 0 (π/2) = 2.

3. We consider the problem

y 00 + 4y 0 + 4y = 0, y(−1) = 2, y 0 (−1) = 1.

We make the substitution y = eλx in the differential equation.

λ2 + 4λ + 4 = 0
(λ + 2)2 = 0
λ = −2

819
 The general solution of the differential equation is
y = c1 e−2t + c2 t e−2t .

We apply the initial conditions to determine the constants.

c1 e2 − c2 e2 = 2, −2c1 e2 + 3c2 e2 = 1
c1 = 7 e−2 , c2 = 5 e−2

The solution subject to the initial conditions is

y = (7 + 5t) e−2(t+1)

The solution is plotted in Figure 19.3. The solution vanishes as t → ∞.

7 + 5t 5
lim (7 + 5t) e−2(t+1) = lim = lim =0
t→∞ t→∞ e2(t+1) t→∞ 2 e2(t+1)

Solution 19.2

y 00 − 4y 0 + 13y = 0.

With the substitution y = eλx we obtain

λ2 eλx − 4λ eλx + 13 eλx = 0
λ2 − 4λ + 13 = 0
λ = 2 ± 3i.

Thus two linearly independent solutions are

e(2+3i)x , and e(2−3i)x .

820
 2

1.5

0.5

-1 1 2 3 4 5

Figure 19.3: The solution of y 00 + 4y 0 + 4y = 0, y(−1) = 2, y 0 (−1) = 1.

Noting that
e(2+3i)x = e2x [cos(3x) + i sin(3x)]
e(2−3i)x = e2x [cos(3x) − i sin(3x)],
we can write the two linearly independent solutions
y1 = e2x cos(3x), y2 = e2x sin(3x).

Solution 19.3
We note that
y 000 − y 00 + y 0 − y = 0
is a constant coefficient equation. The substitution, y = eλx , yields
λ3 − λ2 + λ − 1 = 0
(λ − 1)(λ − i)(λ + i) = 0.

821
The corresponding solutions are ex , eix , and e−ix . We can write the general solution as

y = c1 ex + c2 cos x + c3 sin x.

Solution 19.4
We start with the equation y 00 + y = 0. We substitute y = eλx into the differential equation to obtain

λ2 + 1 = 0, λ = ±i.

A linearly independent set of solutions is

{ eix , e−ix }.

The fundamental set of solutions has the form

y1 = c1 eix + c2 e−ix ,
y2 = c3 eix + c4 e−ix .

By applying the constraints

y1 (0) = 1, y10 (0) = 0,

y2 (0) = 0, y20 (0) = 1,

we obtain
eix + e−ix
y1 = = cos x,
2
eix + e−ix
y2 = = sin x.
2i
Now consider the equation y 00 − y = 0. By substituting y = eλx we find that a set of solutions is

{ ex , e−x }.

822
By taking linear combinations of these we see that another set of solutions is

{cosh x, sinh x}.

Note that this is the fundamental set of solutions.

Next consider y 00 = 0. We can find the solutions by substituting y = eλx or by integrating the equation twice.
The fundamental set of solutions as x = 0 is

{1, x}.

Note that if u(x) is a solution of a constant coefficient differential equation, then u(x + c) is also a solution.
Also note that if u(x) satisfies y(0) = a, y 0 (0) = b, then u(x − x0 ) satisfies y(x0 ) = a, y 0 (x0 ) = b. Thus the
fundamental sets of solutions at x = 1 are

1. {cos(x − 1), sin(x − 1)},

2. {cosh(x − 1), sinh(x − 1)},

3. {1, x − 1}.

Solution 19.5
We substitute y = eλx into the differential equation.

y 00 + 2ay 0 + by = 0
λ2 + 2aλ + b = 0
√
λ = −a ± a2 − b

If a2 > b then the two roots are distinct and real. The general solution is
√ √
y = c1 e(−a+ a2 −b)x
+ c2 e(−a− a2 −b)x
.

823
If a2 < b then the two roots are distinct and complex-valued. We can write them as
√
λ = −a ± i b − a2 .

The general solution is

√ √
y = c1 e(−a+i b−a2 )x
+ c2 e(−a−i b−a2 )x
.

By taking the sum and difference of the two linearly independent solutions above, we can write the general solution
as
√  √ 
y = c1 e−ax cos b − a2 x + c2 e−ax sin b − a2 x .

If a2 = b then the only root is λ = −a. The general solution in this case is then

y = c1 e−ax + c2 x e−ax .

In summary, the general solution is

  √ √ 
−ax a2 −b x − a2 −b x
e

 c1 e + c2 e if a2 > b,
√ √
y = e−ax c1 cos b − a2 x + c2 sin b − a2 x




 if a2 < b,
 e−ax (c + c x)

if a2 = b.
1 2

Solution 19.6
First we note that the general solution can be written,

−ax
√
√


e c1 cosh a2 − b x + c2 sinh a2 − b x if a2 > b,
√ √

y = e−ax c1 cos b − a2 x + c2 sin b − a2 x




if a2 < b,

 −ax
e (c1 + c2 x) if a2 = b.

824
We first consider the case a2 > b. The derivative is
 √  √   √  √ 
y 0 = e−ax −ac1 + a2 − b c2 cosh a2 − b x + −ac2 + a2 − b c1 sinh a2 − b x .

The conditions, y1 (0) = 1 and y10 (0) = 0, for the first solution become,
√
c1 = 1, −ac1 + a2 − b c2 = 0,
a
c1 = 1, c2 = √ .
a2 − b
The conditions, y2 (0) = 0 and y20 (0) = 1, for the second solution become,
√
c1 = 0, −ac1 + a2 − b c2 = 1,
1
c1 = 0, c2 = √ .
2
a −b
The fundamental set of solutions is
  √  a √  1 √ 
−ax −ax
e cosh a − bx + √
2 sinh a − bx , e √
2 sinh 2
a − bx .
a2 − b a2 − b
Now consider the case a2 < b. The derivative is
 √  √   √  √ 
y 0 = e−ax −ac1 + b − a2 c2 cos b − a2 x + −ac2 − b − a2 c1 sin b − a2 x .

Clearly, the fundamental set of solutions is

√ √ √
    
−ax
 a −ax 1
e cos b − a2 x + √ sin b − a2 x ,e √ sin b − a2 x .
b − a2 b − a2
Finally we consider the case a2 = b. The derivative is

y 0 = e−ax (−ac1 + c2 + −ac2 x).

825
The conditions, y1 (0) = 1 and y10 (0) = 0, for the first solution become,

c1 = 1, −ac1 + c2 = 0,
c1 = 1, c2 = a.

The conditions, y2 (0) = 0 and y20 (0) = 1, for the second solution become,

c1 = 0, −ac1 + c2 = 1,
c1 = 0, c2 = 1.

The fundamental set of solutions is

(1 + ax) e−ax , x e−ax .



In summary, the fundamental set of solutions at x = 0 is

n
−ax
 √
a
√
 −ax 1 √
o
n e

 2 √ 2
cosh a − b x + a2 −b sinh a − b x , e √
a2 −b
2
sinh a − b x if a2 > b,
−ax
 √ √ 
−ax √ 1
√
o
2x + √ a



 e cos b − a b−a2
sin b − a 2x , e b−a2
sin b − a 2x if a2 < b,

−ax −ax
{(1 + ax) e , x e } if a2 = b.


Solution 19.7
Let y(t) denote the displacement of the mass from equilibrium. The forces on the mass are −ky(t) due to the
spring and −µy 0 (t) due to friction. We equate the external forces to my 00 (t) to find the differential equation of
the motion.

my 00 = −ky − µy 0
µ 0 k
y 00 + y + y=0
m m

826
The solution which satisfies the initial conditions y(0) = 0, y 0 (0) = 1 is
 p 
−µt/(2m) √ 2m

 e sinh µ 2 − 4km t/(2m) if µ2 > km,

 µ2 −4km p 
y(t) = e−µt/(2m) √ 2m sin 4km − µ 2 t/(2m) if µ2 < km,

 4km−µ2
t e−µt/(2m)

if µ2 = km.

We respectively call these cases: strongly damped, weakly damped and critically damped. In the case that
m = k = 1 the solution is
 p 
−µt/2 √ 2 2 − 4 t/2

 e sinh µ if µ > 2,

 µ2 −4 p 
y(t) = e−µt/2 √ 2 sin 4 − µ2 t/2 if µ < 2,

 4−µ2
t e−t

if µ = 2.
Note that when t is large, t e−t is much smaller than e−µt/2 for µ < 2. To prove this we examine the ratio of these
functions as t → ∞.
t e−t t
lim −µt/2 = lim (1−µ/2)t
t→∞ e t→∞ e
1
= lim
t→∞ (1 − µ/2) e(1−µ)t

=0
Using this result, we see that the critically damped solution decays faster than the weakly damped solution.
We can write the strongly damped solution as
2  √ √ 2 
2
e−µt/2 p e µ −4 t/2 − e− µ −4 t/2 .
µ2 − 4
√ 
µ2 −4−µ t/2
For large t, the dominant factor is e . Note that for µ > 2,
p p
µ2 − 4 = (µ + 2)(µ − 2) > µ − 2.

827
Therefore we have the bounds
p
−2 < µ2 − 4 − µ < 0.
This shows that the critically damped solution decays faster than the strongly damped solution. µ = 2 gives the
fastest decaying solution. Figure 19.4 shows the solution for µ = 4, µ = 1 and µ = 2.

0.5
Strong Damping
0.4
Weak Damping
0.3 Critical Damping
0.2

0.1

2 4 6 8 10
-0.1

Figure 19.4: Strongly, weakly and critically damped solutions.

Solution 19.8
Clearly y = c cos(x − φ) satisfies the differential equation y 00 + y = 0. Since it is a two-parameter family of
functions, it must be the general solution.
Using a trigonometric identity we can rewrite the solution as
y = c cos φ cos x + c sin φ sin x.
Setting this equal to sin x gives us the two equations
c cos φ = 0,
c sin φ = 1,

828
which has the solutions c = 1, φ = (2n + 1/2)π, and c = −1, φ = (2n − 1/2)π, for n ∈ Z.
Clearly y = c cosh(x − φ) satisfies the differential equation y 00 − y = 0. Since it is a two-parameter family of
functions, it must be the general solution.
Using a trigonometric identity we can rewrite the solution as
y = c cosh φ cosh x + c sinh φ sinh x.
Setting this equal to sinh x gives us the two equations
c cosh φ = 0,
c sinh φ = 1,
which has the solutions c = −i, φ = i(2n + 1/2)π, and c = i, φ = i(2n − 1/2)π, for n ∈ Z.

Solution 19.9
We substitute y = eλt into the differential equation.
λ2 eλt + 5λ eλt + 6 eλt = 0
λ2 + 5λ + 6 = 0
(λ + 2)(λ + 3) = 0
The general solution of the differential equation is
y = c1 e−2t + c2 e−3t .
The initial conditions give us the constraints:
c1 + c2 = 1,
−2c1 − 3c2 = V.
The solution subject to the initial conditions is
y = (3 + V ) e−2t − (2 + V ) e−3t .
This solution will be non-negative for t > 0 if V ≥ −3.

829
Solution 19.10
For negative x, the differential equation is
y 00 − y = 0.
We substitute y = eλx into the differential equation to find the solutions.
λ2 − 1 = 0
λ = ±1
y = ex , e−x


We can take linear combinations to write the solutions in terms of the hyperbolic sine and cosine.
y = {cosh(x), sinh(x)}
For positive x, the differential equation is
y 00 + y = 0.
We substitute y = eλx into the differential equation to find the solutions.
λ2 + 1 = 0
λ = ±i
 ix −ix
y= e ,e
We can take linear combinations to write the solutions in terms of the sine and cosine.
y = {cos(x), sin(x)}
We will find the fundamental set of solutions at x = 0. That is, we will find a set of solutions, {y1 , y2 } that
satisfy the conditions:
y1 (0) = 1 y10 (0) = 0
y2 (0) = 0 y20 (0) = 1

830
Clearly these solutions are
( (
cosh(x) x < 0 sinh(x) x < 0
y1 = y2 =
cos(x) x≥0 sin(x) x≥0

Euler Equations
Solution 19.11
We consider an Euler equation,
x2 y 00 + xy 0 + y = 0, x > 0.
We make the change of independent variable ξ = ln x, u(ξ) = y(x) to obtain
u00 + u = 0.
We make the substitution u(ξ) = eλξ .
λ2 + 1 = 0
λ = ±i
A set of linearly independent solutions for u(ξ) is
{ eiξ , e−iξ }.
Since
eiξ + e−iξ eiξ − e−iξ
cos ξ = and sin ξ = ,
2 2i
another linearly independent set of solutions is
{cos ξ, sin ξ}.
The general solution for y(x) is
y(x) = c1 cos(ln x) + c2 sin(ln x).

831
Solution 19.12
Consider the differential equation
x2 y 00 − 2xy + 2y = 0.
With the substitution y = xλ this equation becomes
λ(λ − 1) − 2λ + 2 = 0
λ2 − 3λ + 2 = 0
λ = 1, 2.
The general solution is then

y = c 1 x + c 2 x2 .

Solution 19.13
We note that
1
xy 000 + y 00 + y 0 = 0
x
is an Euler equation. The substitution y = xλ yields
λ3 − 3λ2 + 2λ + λ2 − λ + λ = 0
λ3 − 2λ2 + 2λ = 0.
The three roots of this algebraic equation are
λ = 0, λ = 1 + i, λ=1−i

The corresponding solutions to the differential equation are

y = x0 y = x1+i y = x1−i
y=1 y = x ei ln x y = x e−i ln x .

832
We can write the general solution as
y = c1 + c2 x cos(ln x) + c3 sin(ln x).

Solution 19.14
We substitute y = xλ into the differential equation.
x2 y 00 + (2a + 1)xy 0 + by = 0
λ(λ − 1) + (2a + 1)λ + b = 0
λ2 + 2aλ + b = 0
√
λ = −a ± a2 − b
For a2 > b then the general solution is
√ √
a2 −b a2 −b
y = c1 x−a+ + c2 x−a− .
For a2 < b, then the general solution is
√ √
b−a2 b−a2
y = c1 x−a+i + c2 x−a−i .
By taking the sum and difference of these solutions, we can write the general solution as
√  √ 
y = c1 x−a cos b − a2 ln x + c2 x−a sin b − a2 ln x .
For a2 = b, the quadratic in lambda has a double root at λ = a. The general solution of the differential equation
is
y = c1 x−a + c2 x−a ln x.
In summary, the general solution is:
  √ √ 
−a a2 −b − a2 −b

 x c 1 x + c 2 x if a2 > b,
√ √

y= x −a




 c 1 cos b − a 2 ln x + c sin
2 b − a2 ln x if a2 < b,
x−a (c + c ln x)

if a2 = b.
1 2

833
Solution 19.15
For a 6= 0, two linearly independent solutions of

y 00 − a2 y = 0

are

y1 = eax , y2 = e−ax .

For a = 0, we have

y1 = e0x = 1, y2 = x e0x = x.

In this case the solution are defined by

 
ax d ax
y1 = [ e ]a=0 , y2 = e .
da a=0

By the definition of differentiation, f 0 (0) is

f (a) − f (−a)
f 0 (0) = lim .
a→0 2a
Thus the second solution in the case a = 0 is
eax − e−ax
y2 = lim
a→0 a
Consider the solutions
eαx − e−αx
y1 = eax , y2 = lim .
α→a α
Clearly y1 is a solution for all a. For a 6= 0, y2 is a linear combination of eax and e−ax and is thus a solution.
Since the coefficient of e−ax in this linear combination is non-zero, it is linearly independent to y1 . For a = 0, y2
is one half the derivative of eax evaluated at a = 0. Thus it is a solution.

834
 For a 6= 0, two linearly independent solutions of

x2 y 00 + xy 0 − a2 y = 0

are

y 1 = xa , y2 = x−a .

For a = 0, we have
 
a d a
y1 = [x ]a=0 = 1, y2 = x = ln x.
da a=0

Consider the solutions

a xa − x−a
y1 = x , y2 =
a
Clearly y1 is a solution for all a. For a 6= 0, y2 is a linear combination of xa and x−a and is thus a solution. For
a = 0, y2 is one half the derivative of xa evaluated at a = 0. Thus it is a solution.

Solution 19.16
1.

x2 y 00 − 2xy 0 + 2y = 0

We substitute y = xλ into the differential equation.

λ(λ − 1) − 2λ + 2 = 0
λ2 − 3λ + 2 = 0
(λ − 1)(λ − 2) = 0
y = c1 x + c2 x2

835
 2.

x2 y 00 − 2y = 0

We substitute y = xλ into the differential equation.

λ(λ − 1) − 2 = 0
λ2 − λ − 2 = 0
(λ + 1)(λ − 2) = 0
c1
y= + c 2 x2
x

3.

x2 y 00 − xy 0 + y = 0

We substitute y = xλ into the differential equation.

λ(λ − 1) − λ + 1 = 0
λ2 − 2λ + 1 = 0
(λ − 1)2 = 0

Since there is a double root, the solution is:

y = c1 x + c2 x ln x.

Exact Equations
Solution 19.17
We note that

y 00 + y 0 sin x + y cos x = 0

836
is an exact equation.

d 0
[y + y sin x] = 0
dx
y 0 + y sin x = c1
d  − cos x 
ye = c1 e− cos x
dx
Z
y = c1 ecos x
e− cos x dx + c2 ecos x

Equations Without Explicit Dependence on y

Reduction of Order
Solution 19.18

(1 − x2 )y 00 − 2xy 0 + 2y = 0, −1 < x < 1

We substitute y = x into the differential equation to check that it is a solution.

(1 − x2 )(0) − 2x(1) + 2x = 0

We look for a second solution of the form y = xu. We substitute this into the differential equation and use the

837
fact that x is a solution.

(1 − x2 )(xu00 + 2u0 ) − 2x(xu0 + u) + 2xu = 0

(1 − x2 )(xu00 + 2u0 ) − 2x(xu0 ) = 0
(1 − x2 )xu00 + (2 − 4x2 )u0 = 0
u00 2 − 4x2
=
u0 x(x2 − 1)
u00 2 1 1
0
=− + −
u x 1−x 1+x
0
ln(u ) = −2 ln(x) − ln(1 − x) − ln(1 + x) + const
 
0 c
ln(u ) = ln
x2 (1 − x)(1 + x)
c
u0 = 2
x (1 − x)(1 + x)
 
0 1 1 1
u =c + +
x2 2(1 − x) 2(1 + x)
 
1 1 1
u = c − − ln(1 − x) + ln(1 + x) + const
x 2 2
  
1 1 1+x
u = c − + ln + const
x 2 1−x

A second linearly independent solution is

 
x 1+x
y = −1 + ln .
2 1−x

838
Solution 19.19
We are given that y = ex is a solution of

x+1 0 1
y 00 − y + y = 0.
x x
To find another linearly independent solution, we will use reduction of order. Substituting

y = u ex
y 0 = (u0 + u) ex
y 00 = (u00 + 2u0 + u) ex

into the differential equation yields

x+1 0 1
u00 + 2u0 + u − (u + u) + u = 0.
x x
x − 1
u00 + u0 = 0
 Z  x  
d 0 1
u exp 1− dx =0
dx x
u0 ex−ln x = c1
u0 = c1 x e−x
Z
u = c1 x e−x dx + c2

u = c1 (x e−x + e−x ) + c2
y = c1 (x + 1) + c2 ex

Thus a second linearly independent solution is

y = x + 1.

839
Solution 19.20
We are given that y = x is a solution of
(1 − 2x)y 00 + 4xy 0 − 4y = 0.
To find another linearly independent solution, we will use reduction of order. Substituting
y = xu
y 0 = xu0 + u
y 00 = xu00 + 2u0
into the differential equation yields
(1 − 2x)(xu00 + 2u0 ) + 4x(xu0 + u) − 4xu = 0,
(1 − 2x)xu00 + (4x2 − 4x + 2)u0 = 0,
u00 4x2 − 4x + 2
= ,
u0 x(2x − 1)
u00 2 2
0
=2− + ,
u x 2x − 1
ln(u0 ) = 2x − 2 ln x + ln(2x − 1) + const,
 
0 2 1
u = c1 − e2x ,
x x2
1
u = c1 e2x + c2 ,
x
y = c1 e2x + c2 x.

Solution 19.21
One solution of
(x − 1)y00 − xy0 + y = 0,

840
is y1 = ex . We find a second solution with reduction of order. We make the substitution y2 = u ex in the
differential equation. We determine u up to an additive constant.

(x − 1)(u00 + 2u0 + u) ex − x(u0 + u) ex + u ex = 0

(x − 1)u00 + (x − 2)u0 = 0
u00 x−2 1
0
=− = −1 +
u x−1 x−1
0
ln |u | = −x + ln |x − 1| + c
u0 = c(x − 1) e−x
u = −cx e−x

The second solution of the differential equation is y2 = x.

*Reduction of Order and the Adjoint Equation

841
Chapter 20

Techniques for Nonlinear Differential

Equations

In mathematics you don’t understand things. You just get used to them.

- Johann von Neumann

20.1 Bernoulli Equations

Sometimes it is possible to solve a nonlinear equation by making a change of the dependent variable that converts
it into a linear equation. One of the most important such equations is the Bernoulli equation

dy
+ p(t)y = q(t)y α , α 6= 1.
dt

The change of dependent variable u = y 1−α will yield a first order linear equation for u which when solved will
give us an implicit solution for y. (See Exercise ??.)

842
 Result 20.1.1 The Bernoulli equation y 0 + p(t)y = q(t)y α , α 6= 1 can be transformed to
the first order linear equation
du
+ (1 − α)p(t)u = (1 − α)q(t)
dt
with the change of variables u = y 1−α .

Example 20.1.1 Consider the Bernoulli equation

2
y0 = y + y2.
x

First we divide by y 2 .

2 −1
y −2 y 0 = y +1
x

We make the change of variable u = y −1 .

2
−u0 = u+1
x
2
u0 + u = −1
x

843
 2
R
The integrating factor is I(x) = exp( x
dx) = x2 .
d 2
(x u) = −x2
dx
1
x2 u = − x3 + c
3
1 c
u=− x+ 2
3 x
 −1
1 c
y = − x+ 2
3 x
Thus the solution for y is

3x2
y= .
c − x2

20.2 Riccati Equations

Factoring Second Order Operators. Consider the second order linear equation
 2 
d d
L[y] = + p(x) + q(x) y = y 00 + p(x)y 0 + q(x)y = f (x).
dx2 dx
If we were able to factor the linear operator L into the form
  
d d
L= + a(x) + b(x) , (20.1)
dx dx
then we would be able to solve the differential equation. Factoring reduces the problem to a system of first order
equations. We start with the factored equation
  
d d
+ a(x) + b(x) y = f (x).
dx dx

844
 d 
We set u = dx
+ b(x) y and solve the problem
 
d
+ a(x) u = f (x).
dx

Then to obtain the solution we solve

 
d
+ b(x) y = u.
dx

Example 20.2.1 Consider the equation

   
00 1 0 1
y + x− y + − 1 y = 0.
x x2

Let’s say by some insight or just random luck we are able to see that this equation can be factored into
  
d d 1
+x − y = 0.
dx dx x

We first solve the equation

 
d
+ x u = 0.
dx
u0 + xu = 0
d  x2 /2 
e u =0
dx
2
u = c1 e−x /2

845
Then we solve for y with the equation
 
d 1 2
− y = u = c1 e−x /2 .
dx x
1 2
y 0 − y = c1 e−x /2
x
d 2
x−1 y = c1 x−1 e−x /2


dx
Z
2
y = c1 x x−1 e−x /2 dx + c2 x

If we were able to solve for a and b in Equation 20.1 in terms of p and q then we would be able to solve any
second order differential equation. Equating the two operators,

d2
  
d d d
+p +q = +a +b
dx2 dx dx dx
d2 d
= 2 + (a + b) + (b0 + ab).
dx dx

Thus we have the two equations

a + b = p, and b0 + ab = q.

Eliminating a,

b0 + (p − b)b = q
b0 = b2 − pb + q

Now we have a nonlinear equation for b that is no easier to solve than the original second order linear equation.

846
Riccati Equations. Equations of the form

y 0 = a(x)y 2 + b(x)y + c(x)

are called Riccati equations. From the above derivation we see that for every second order differential equation
there is a corresponding Riccati equation. Now we will show that the converse is true.
We make the substitution
u0 u00 (u0 )2 a0 u0
y=− , y0 = − + + 2 ,
au au au2 au
in the Riccati equation.

y 0 = ay 2 + by + c
u00 (u0 )2 a0 u0 (u0 )2 u0
− + + = a − b +c
au au2 a2 u a2 u2 au
u00 a0 u0 u0
− + 2 +b −c=0
au  a u  au
a0
u00 − + b u0 + acu = 0
a
Now we have a second order linear equation for u.
0
u
Result 20.2.1 The substitution y = − au transforms the Riccati equation

y 0 = a(x)y 2 + b(x)y + c(x)

into the second order linear equation

 0 
00 a
u − + b u0 + acu = 0.
a

847
Example 20.2.2 Consider the Riccati equation
1 1
y0 = y2 + y + 2 .
x x
0
With the substitution y = − uu we obtain

1 1
u00 − u0 + 2 u = 0.
x x
This is an Euler equation. The substitution u = xλ yields

λ(λ − 1) − λ + 1 = (λ − 1)2 = 0.

Thus the general solution for u is

u = c1 x + c2 x log x.
0
Since y = − uu ,

c1 + c2 (1 + log x)
y=−
c1 x + c2 x log x
1 + c(1 + log x)
y=−
x + cx log x

20.3 Exchanging the Dependent and Independent Variables

Some differential equations can be put in a more elementary form by exchanging the dependent and independent
variables. If the new equation can be solved, you will have an implicit solution for the initial equation. We will
consider a few examples to illustrate the method.

848
Example 20.3.1 Consider the equation

1
y0 = .
y3 − xy 2

Instead of considering y to be a function of x, consider x to be a function of y. That is, x = x(y), x0 = dx

dy
.

dy 1
= 3
dx y − xy 2
dx
= y 3 − xy 2
dy
x0 + y 2 x = y 3

Now we have a first order equation for x.

d  y3 /3  3
e x = y 3 ey /3
dy
Z
−y 3 /3 3 3
x= e y 3 ey /3 dy + c e−y /3

Example 20.3.2 Consider the equation

y
y0 = .
y2 + 2x

849
Interchanging the dependent and independent variables yields

1 y
0
= 2
x y + 2x
x
x0 = y + 2
y
x
x0 − 2 = y
y
d −2
(y x) = y −1
dy
y −2 x = log y + c
x = y 2 log y + cy 2

Result 20.3.1 Some differential equations can be put in a simpler form by exchanging
the dependent and independent variables. Thus a differential equation for y(x) can be
written as an equation for x(y). Solving the equation for x(y) will give an implicit solution
for y(x).

20.4 Autonomous Equations

Autonomous equations have no explicit dependence on x. The following are examples.

• y 00 + 3y 0 − 2y = 0

• y 00 = y + (y 0 )2

• y 000 + y 00 y = 0

850
 The change of variables u(y) = y 0 reduces an nth order autonomous equation in y to a non-autonomous
equation of order n − 1 in u(y). Writing the derivatives of y in terms of u,

y 0 = u(y)
d
y 00 = u(y)
dx
dy d
= u(y)
dx dy
= y 0 u0
= u0 u
y 000 = (u00 u + (u0 )2 )u.

Thus we see that the equation for u(y) will have an order of one less than the original equation.

Result 20.4.1 Consider an autonomous differential equation for y(x), (autonomous e-

quations have no explicit dependence on x.) The change of variables u(y) = y 0 reduces
an nth order autonomous equation in y to a non-autonomous equation of order n − 1 in
u(y).

Example 20.4.1 Consider the equation

y 00 = y + (y 0 )2 .

With the substitution u(y) = y 0 , the equation becomes

u0 u = y + u2
u0 = u + yu−1 .

851
We recognize this as a Bernoulli equation. The substitution v = u2 yields
1 0
v =v+y
2
v 0 − 2v = 2y
d
e−2y v = 2y e−2y


dy
Z
2y
v(y) = c1 e + e 2y
2y e−2y dy
 Z 
2y 2y −2y −2y
v(y) = c1 e + e −y e + e dy
 
2y 2y −2y 1 −2y
v(y) = c1 e + e −y e − e
2
1
v(y) = c1 e2y − y − .
2
Now we solve for u.
 1/2
2y 1
u(y) = c1 e − y − .
2
 1/2
dy 2y 1
= c1 e − y −
dx 2
This equation is separable.
dy
dx =
1 1/2


c1 e2y − y − 2
Z
1
x + c2 = dy
1 1/2


c1 e2y − y − 2

852
Thus we finally have arrived at an implicit solution for y(x).

Example 20.4.2 Consider the equation

y 00 + y 3 = 0.

With the change of variables, u(y) = y 0 , the equation becomes

u0 u + y 3 = 0.

This equation is separable.

u du = −y 3 dy
1 2 1
u = − y 4 + c1
2 4
 1/2
1 4
u = 2c1 − y
2
 1/2
0 1 4
y = 2c1 − y
2
dy
= dx
(2c1 − 12 y 4 )1/2

Integrating gives us the implicit solution

Z
1
dy = x + c2 .
(2c1 − 12 y 4 )1/2

853
20.5 *Equidimensional-in-x Equations
Differential equations that are invariant under the change of variables x = c ξ are said to be equidimensional-in-x.
For a familiar example from linear equations, we note that the Euler equation is equidimensional-in-x. Writing
the new derivatives under the change of variables,
d 1 d d2 1 d2
x = c ξ, = , = 2 2, ... .
dx c dξ dx2 c dξ

Example 20.5.1 Consider the Euler equation

2 3
y 00 + y 0 + 2 y = 0.
x x
Under the change of variables, x = c ξ, y(x) = u(ξ), this equation becomes
1 00 2 1 0 3
2
u + u + 2 2u = 0
c cξ c c ξ
2 3
u00 + u0 + 2 u = 0.
ξ ξ
Thus this equation is invariant under the change of variables x = c ξ.

Example 20.5.2 For a nonlinear example, consider the equation

y 00 y0
y 00 y 0 + + 2 = 0.
xy x
With the change of variables x = c ξ, y(x) = u(ξ) the equation becomes
u00 u0 u00 u0
+ + =0
c2 c c3 ξ u c3 ξ 2
u00 u0
u00 u0 + + 2 = 0.
ξu ξ
We see that this equation is also equidimensional-in-x.

854
 You may recall that the change of variables x = et reduces an Euler equation to a constant coefficient
equation. To generalize this result to nonlinear equations we will see that the same change of variables reduces
an equidimensional-in-x equation to an autonomous equation.
Writing the derivatives with respect to x in terms of t,
d dt d d
x = et , = = e−t
dx dx dt dt

d d
x =
dx dt

d2 d2
 
2 d d d d
x = x x −x = 2− .
dx2 dx dx dx dt dt

Example 20.5.3 Consider the equation in Example 20.5.2

y 00 y0
y 00 y 0 + + 2 = 0.
xy x
Applying the change of variables x = et , y(x) = u(t) yields an autonomous equation for u(t).
x2 y 00
x2 y 00 x y 0 + + x y0 = 0
y
u00 − u0
(u00 − u0 )u0 + + u0 = 0
u

Result 20.5.1 A differential equation that is invariant under the change of variables
x = c ξ is equidimensional-in-x. Such an equation can be reduced to autonomous equation
of the same order with the change of variables, x = et .

855
20.6 *Equidimensional-in-y Equations
A differential equation is said to be equidimensional-in-y if it is invariant under the change of variables y(x) =
c v(x). Note that all linear homogeneous equations are equidimensional-in-y.

Example 20.6.1 Consider the linear equation

y 00 + p(x)y 0 + q(x)y = 0.

With the change of variables y(x) = cv(x) the equation becomes

cv 00 + p(x)cv 0 + q(x)cv = 0
v 00 + p(x)v 0 + q(x)v = 0

Thus we see that the equation is invariant under the change of variables.

Example 20.6.2 For a nonlinear example, consider the equation

y 00 y + (y 0 )2 − y 2 = 0.

Under the change of variables y(x) = cv(x) the equation becomes.

cv 00 cv + (cv 0 )2 − (cv)2 = 0
v 00 v + (v 0 )2 − v 2 = 0.

Thus we see that this equation is also equidimensional-in-y.

856
 The change of variables y(x) = eu(x) reduces an nth order equidimensional-in-y equation to an equation of
order n − 1 for u0 . Writing the derivatives of eu(x) ,
d u
e = u0 eu
dx
d2 u
e = (u00 + (u0 )2 ) eu
dx2
d3 u
e = (u000 + 3u00 u00 + (u0 )3 ) eu .
dx3

Example 20.6.3 Consider the linear equation in Example 20.6.1

y 00 + p(x)y 0 + q(x)y = 0.
Under the change of variables y(x) = eu(x) the equation becomes
(u00 + (u0 )2 ) eu + p(x)u0 eu + q(x) eu = 0
u00 + (u0 )2 + p(x)u0 + q(x) = 0.
Thus we have a Riccati equation for u0 . This transformation might seem rather useless since linear equations are
usually easier to work with than nonlinear equations, but it is often useful in determining the asymptotic behavior
of the equation.

Example 20.6.4 From Example 20.6.2 we have the equation

y 00 y + (y 0 )2 − y 2 = 0.
The change of variables y(x) = eu(x) yields
(u00 + (u0 )2 ) eu eu + (u0 eu )2 − ( eu )2 = 0
u00 + 2(u0 )2 − 1 = 0
u00 = −2(u0 )2 + 1

857
 v0
Now we have a Riccati equation for u0 . We make the substitution u0 = 2v
.

v 00 (v 0 )2 (v 0 )2
− = −2 +1
2v 2v 2 4v 2
v 00 − 2v = 0
√ √
v = c1 e 2x + c2 e− 2x
√ √
√ c 1 e 2x
− c 2 e− 2x
u0 = 2 2 √ √
c1 e 2x + c2 e− 2x
√ √ √ √
c1 2 e 2x − c2 2 e− 2x
Z
u=2 √ √ dx + c3
c1 e 2x + c2 e− 2x
 √ √ 
u = 2 log c1 e 2x + c2 e− 2x + c3
 √ √ 2
y = c1 e 2x + c2 e− 2x ec3

The constants are redundant, the general solution is

 √ √ 2
y = c1 e 2x + c2 e− 2x

Result 20.6.1 A differential equation is equidimensional-in-y if it is invariant under the

change of variables y(x) = cv(x). An nth order equidimensional-in-y equation can be
reduced to an equation of order n − 1 in u0 with the change of variables y(x) = eu(x) .

858
20.7 *Scale-Invariant Equations

Result 20.7.1 An equation is scale invariant if it is invariant under the change of vari-
ables, x = cξ, y(x) = cα v(ξ), for some value of α. A scale-invariant equation can be trans-
formed to an equidimensional-in-x equation with the change of variables, y(x) = xα u(x).

Example 20.7.1 Consider the equation

y 00 + x2 y 2 = 0.

Under the change of variables x = cξ, y(x) = cα v(ξ) this equation becomes
cα 00
v (ξ) + c2 x2 c2α v 2 (ξ) = 0.
c2
Equating powers of c in the two terms yields α = −4.
Introducing the change of variables y(x) = x−4 u(x) yields

d2  −4
+ x2 (x−4 u(x))2 = 0

x u(x)
dx2
x−4 u00 − 8x−5 u0 + 20x−6 u + x−6 u2 = 0
x2 u00 − 8xu0 + 20u + u2 = 0.

We see that the equation for u is equidimensional-in-x.

859
20.8 Exercises
Exercise 20.1
1. Find the general solution and the singular solution of the Clairaut equation,

y = xp + p2 .

2. Show that the singular solution is the envelope of the general solution.
Hint, Solution

Bernoulli Equations
Exercise 20.2 (mathematica/ode/techniques nonlinear/bernoulli.nb)
Consider the Bernoulli equation

dy
+ p(t)y = q(t)y α .
dt
1. Solve the Bernoulli equation for α = 1.

2. Show that for α 6= 1 the substitution u = y 1−α reduces Bernoulli’s equation to a linear equation.

3. Find the general solution to the following equations.

dy
t2 + 2ty − y 3 = 0, t > 0
dt
(a)
dy
+ 2xy + y 2 = 0
dx
(b)

Hint, Solution

860
Exercise 20.3
Consider a population, y. Let the birth rate of the population be proportional to y with constant of proportionality
1. Let the death rate of the population be proportional to y 2 with constant of proportionality 1/1000. Assume
that the population is large enough so that you can consider y to be continuous. What is the population as a
function of time if the initial population is y0 ?
Hint, Solution

Exercise 20.4
Show that the transformation u = y 1−n reduces the equation to a linear first order equation. Solve the equations

dy
1. t2 + 2ty − y 3 = 0 t > 0
dt
dy
2. = (Γ cos t + T ) y − y 3 , Γ and T are real constants. (From a fluid flow stability problem.)
dt
Hint, Solution

Riccati Equations
Exercise 20.5
1. Consider the Ricatti equation,

dy
= a(x)y 2 + b(x)y + c(x).
dx
Substitute
1
y = yp (x) +
u(x)

into the Ricatti equation, where yp is some particular solution to obtain a first order linear differential
equation for u.

861
 2. Consider a Ricatti equation,

y 0 = 1 + x2 − 2xy + y 2 .

Verify that yp (x) = x is a particular solution. Make the substitution y = yp + 1/u to find the general
solution.
What would happen if you continued this method, taking the general solution for yp ? Would you be able
to find a more general solution?

3. The substitution
u0
y=−
au
gives us the second order, linear, homogeneous differential equation,
 0 
00 a
u − + b u0 + acu = 0.
a
The general solution for u has two constants of integration. However, the solution for y should only have
one constant of integration as it satisfies a first order equation. Write y in terms of the solution for u and
verify tha y has only one constant of integration.
Hint, Solution

Exchanging the Dependent and Independent Variables

Exercise 20.6
Solve the differential equation
√
0 y
y = .
xy + y
Hint, Solution

862
Autonomous Equations
*Equidimensional-in-x Equations
*Equidimensional-in-y Equations
*Scale-Invariant Equations

863
20.9 Hints
Hint 20.1

Bernoulli Equations
Hint 20.2

Hint 20.3
The differential equation governing the population is

dy y2
=y− , y(0) = y0 .
dt 1000
This is a Bernoulli equation.

Hint 20.4

Riccati Equations
Hint 20.5

Exchanging the Dependent and Independent Variables

Hint 20.6
Exchange the dependent and independent variables.

Autonomous Equations

864
*Equidimensional-in-x Equations
*Equidimensional-in-y Equations
*Scale-Invariant Equations

865
20.10 Solutions
Solution 20.1
We consider the Clairaut equation,

y = xp + p2 . (20.2)

1. We differentiate Equation 20.2 with respect to x to obtain a second order differential equation.

y 0 = y 0 + xy 00 + 2y 0 y 00
y 00 (2y 0 + x) = 0

Equating the first or second factor to zero will lead us to two distinct solutions.
x
y 00 = 0 or y 0 = −
2
If y 00 = 0 then y 0 ≡ p is a constant, (say y 0 = c). From Equation 20.2 we see that the general solution is,

y(x) = cx + c2 . (20.3)

Recall that the general solution of a first order differential equation has one constant of integration.
If y 0 = −x/2 then y = −x2 /4 + const. We determine the constant by substituting the expression into
Equation 20.2.
x2  x   x 2
− +c=x − + −
4 2 2
Thus we see that a singular solution of the Clairaut equation is

1
y(x) = − x2 . (20.4)
4
Recall that a singular solution of a first order nonlinear differential equation has no constant of integration.

866
2. Equating the general and singular solutions, y(x), and their derivatives, y 0 (x), gives us the system of
equations,
1 1
cx + c2 = − x2 , c = − x.
4 2
Since the first equation is satisfied for c = −x/2, we see that the solution y = cx + c2 is tangent to the
solution y = −x2 /4 at the point (−2c, −|c|). The solution y = cx + c2 is plotted for c = . . . , −1/4, 0, 1/4, . . .
in Figure 20.1.
2

-4 -2 2 4

-1

-2

-3

-4

Figure 20.1: The Envelope of y = cx + c2 .

The envelope of a one-parameter family F (x, y, c) = 0 is given by the system of equations,

F (x, y, c) = 0, Fc (x, y, c) = 0.
For the family of solutions y = cx + c2 these equations are
y = cx + c2 , 0 = x + 2c.
Substituting the solution of the second equation, c = −x/2, into the first equation gives the envelope,
   2
1 1 1
y = − x x + − x = − x2 .
2 2 4
Thus we see that the singular solution is the envelope of the general solution.

867
Bernoulli Equations
Solution 20.2
1.

dy
+ p(t)y = q(t)y
dt
dy
= (q − p) dt
y
Z
ln y = (q − p) dt + c
R
(q−p) dt
y = ce

2. We consider the Bernoulli equation,

dy
+ p(t)y = q(t)y α , α 6= 1.
dt
We divide by y α .

y −α y 0 + p(t)y 1−α = q(t)

This suggests the change of dependent variable u = y 1−α , u0 = (1 − α)y −α y 0 .

1 d 1−α
y + p(t)y 1−α = q(t)
1 − α dt
du
+ (1 − α)p(t)u = (1 − α)q(t)
dt
Thus we obtain a linear equation for u which when solved will give us an implicit solution for y.

868
3. (a)
dy
t2 + 2ty − y 3 = 0, t > 0
dt
y0 1
t2 3 + 2t 2 = 1
y y
We make the change of variables u = y −2 .
1
− t2 u0 + 2tu = 1
2
4 2
u0 − u = − 2
t t
The integrating factor is
R
µ= e (−4/t) dt
= e−4 ln t = t−4 .
We multiply by the integrating factor and integrate to obtain the solution.
d −4

t u = −2t−6
dt
2
u = t−1 + ct4
5
2
y −2 = t−1 + ct4
5
√
1 5t
y = ±q y = ±√
2 −1
t + ct4 2 + ct5
5

(b)
dy
+ 2xy + y 2 = 0
dx
y0 2x
+ = −1
y2 y

869
 We make the change of variables u = y −1 .
u0 − 2xu = 1
The integrating factor is
2
R
µ= e (−2x) dx
= e−x .
We multiply by the integrating factor and integrate to obtain the solution.
d  −x2  2
e u = e−x
dx Z
2 2 2
u = ex e−x dx + c ex
2
e−x
y = R −x2
e dx + c

Solution 20.3
The differential equation governing the population is
dy y2
=y− , y(0) = y0 .
dt 1000
We recognize this as a Bernoulli equation. The substitution u(t) = 1/y(t) yields
du 1 1
− =u− , u(0) = .
dt 1000 y0
1
u0 + u =
1000
t
e−t
Z
1 −t
u= e + eτ dτ
y0 1000 0
 
1 1 1
u= + − e−t
1000 y0 1000

870
Solving for y(t),
   −1
1 1 1 −t
y(t) = + − e .
1000 y0 1000

As a check, we see that as t → ∞, y(t) → 1000, which is an equilibrium solution of the differential equation.
dy y2
=0=y− ⇒ y = 1000.
dt 1000

Solution 20.4
1.
dy
t2 + 2ty − y 3 = 0
dt
dy
+ 2t−1 y = t−2 y 3
dt
We make the change of variables u(t) = y −2 (t).

u0 − 4t−1 u = −2t−2
This gives us a first order, linear equation. The integrating factor is
−4t−1 dt
R
I(t) = e = e−4 log t = t−4 .
We multiply by the integrating factor and integrate.
d −4

t u = −2t−6
dt
2
t−4 u = t−5 + c
5
2 −1
u = t + ct4
5

871
 Finally we write the solution in terms of y(t).
1
y(t) = ± q
2 −1
5
t + ct4
√
5t
y(t) = ± √
2 + ct5

2.
dy
− (Γ cos t + T ) y = −y 3
dt
We make the change of variables u(t) = y −2 (t).

u0 + 2 (Γ cos t + T ) u = 2
This gives us a first order, linear equation. The integrating factor is
R
2(Γ cos t+T ) dt
I(t) = e = e2(Γ sin t+T t)
We multiply by the integrating factor and integrate.
d
e2(Γ sin t+T t) u = 2 e2(Γ sin t+T t)


dt Z 
−2(Γ sin t+T t) 2(Γ sin t+T t)
u = 2e e dt + c

Finally we write the solution in terms of y(t).

eΓ sin t+T t
y = ±q R

2 e2(Γ sin t+T t) dt + c

872
Riccati Equations
Solution 20.5
We consider the Ricatti equation,

dy
= a(x)y 2 + b(x)y + c(x). (20.5)
dx
1. We substitute
1
y = yp (x) +
u(x)

into the Ricatti equation, where yp is some particular solution.

u0
   
yp 1 1
yp0 2
− 2 = +a(x) yp + 2 + 2 + b(x) yp + + c(x)
u u u u
u0
 
1 yp 1
− 2 = b(x) + a(x) 2 + 2
u u u u
u0 = − (b + 2ayp ) u − a

We obtain a first order linear differential equation for u whose solution will contain one constant of integra-
tion.

2. We consider a Ricatti equation,

y 0 = 1 + x2 − 2xy + y 2 . (20.6)

We verify that yp (x) = x is a solution.

1 = 1 + x2 − 2xx + x2

873
Substituting y = yp + 1/u into Equation 20.6 yields,

u0 = − (−2x + 2x) u − 1
u = −x + c
1
y =x+
c−x

1
What would happen if we continued this method? Since y = x + c−x
is a solution of the Ricatti equation
we can make the substitution,

1 1
y =x+ + , (20.7)
c − x u(x)

which will lead to a solution for y which has two constants of integration. Then we could repeat the process,
substituting the sum of that solution and 1/u(x) into the Ricatti equation to find a solution with three
constants of integration. We know that the general solution of a first order, ordinary differential equation
has only one constant of integration. Does this method for Ricatti equations violate this theorem? There’s
only one way to find out. We substitute Equation 20.7 into the Ricatti equation.
  
0 1
u = − −2x + 2 x + u−1
c−x
2
u0 = − u−1
c−x
2
u0 + u = −1
c−x

The integrating factor is

1
I(x) = e2/(c−x) = e−2 log(c−x) = .
(c − x)2

874
 Upon multiplying by the integrating factor, the equation becomes exact.
 
d 1 1
u = −
dx (c − x)2 (c − x)2
−1
u = (c − x)2 + b(c − x)2
c−x
u = x − c + b(c − x)2
Thus the Ricatti equation has the solution,
1 1
y =x+ + .
c − x x − c + b(c − x)2
It appears that we we have found a solution that has two constants of integration, but appearances can be
deceptive. We do a little algebraic simplification of the solution.
1 1
y =x+ +
c − x (b(c − x) − 1)(c − x)
(b(c − x) − 1) + 1
y =x+
(b(c − x) − 1)(c − x)
b
y =x+
b(c − x) − 1
1
y =x+
(c − 1/b) − x

This is actually a solution, (namely the solution we had before), with one constant of integration, (namely
c − 1/b). Thus we see that repeated applications of the procedure will not produce more general solutions.
3. The substitution
u0
y=−
au

875
 gives us the second order, linear, homogeneous differential equation,

a0
 
00
u − + b u0 + acu = 0.
a

The solution to this linear equation is a linear combination of two homogeneous solutions, u1 and u2 .

u = c1 u1 (x) + c2 u2 (x)

The solution of the Ricatti equation is then

c1 u01 (x) + c2 u02 (x)

y=− .
a(x)(c1 u1 (x) + c2 u2 (x))

Since we can divide the numerator and denominator by either c1 or c2 , this answer has only one constant of
integration, (namely c1 /c2 or c2 /c1 ).

Exchanging the Dependent and Independent Variables

Solution 20.6
Exchanging the dependent and independent variables in the differential equation,

√
0 y
y = ,
xy + y

yields

x0 (y) = y 1/2 x + y 1/2 .

876
This is a first order differential equation for x(y).

x0 − y 1/2 x = y 1/2
2y 3/2 2y 3/2
    
d 1/2
x exp − = y exp −
dy 3 3
2y 3/2 2y 3/2
   
x exp − = − exp − + c1
3 3
 3/2 
2y
x = −1 + c1 exp
3
 3/2 
x+1 2y
= exp
c1 3
 
x+1 2
log = y 3/2
c1 3
  2/3
3 x+1
y= log
2 c1
 2/3
3
y = c + log(x + 1)
2

Autonomous Equations
*Equidimensional-in-x Equations
*Equidimensional-in-y Equations
*Scale-Invariant Equations

877
Chapter 21

Transformations and Canonical Forms

Prize intensity more than extent. Excellence resides in quality not in quantity. The best is always few and
rare - abundance lowers value. Even among men, the giants are usually really dwarfs. Some reckon books by
the thickness, as if they were written to exercise the brawn more than the brain. Extent alone never rises above
mediocrity; it is the misfortune of universal geniuses that in attempting to be at home everywhere are so nowhere.
Intensity gives eminence and rises to the heroic in matters sublime.

-Balthasar Gracian

21.1 The Constant Coefficient Equation

The solution of any second order linear homogeneous differential equation can be written in terms of the solutions
to either

y 00 = 0, or y 00 − y = 0

Consider the general equation

y 00 + ay 0 + by = 0.

878
We can solve this differential equation by making the substitution y = eλx . This yields the algebraic equation

λ2 + aλ + b = 0.

1 √ 
λ= 2
−a ± a − 4b
2
There are two cases to consider. If a2 6= 4b then the solutions are
√ √
a2 −4b)x/2 a2 −4b)x/2
y1 = e(−a+ , y2 = e(−a−

If a2 = 4b then we have

y1 = e−ax/2 , y2 = x e−ax/2

Note that regardless of the values of a and b the solutions are of the form

y = e−ax/2 u(x)

We would like to write the solutions to the general differential equation in terms of the solutions to simpler
differential equations. We make the substitution

y = eλx u

The derivatives of y are

y 0 = eλx (u0 + λu)

y 00 = eλx (u00 + 2λu0 + λ2 u)

Substituting these into the differential equation yields

u00 + (2λ + a)u0 + (λ2 + aλ + b)u = 0

879
In order to get rid of the u0 term we choose
a
λ=− .
2
The equation is then
a2
 
00
u + b− u = 0.
4
There are now two cases to consider.

Case 1. If b = a2 /4 then the differential equation is

u00 = 0
which has solutions 1 and x. The general solution for y is then
y = e−ax/2 (c1 + c2 x).

Case 2. If b 6= a2 /4 then the differential equation is

 2 
00 a
u − − b u = 0.
4
We make the change variables
u(x) = v(ξ), x = µξ.
The derivatives in terms of ξ are
d dξ d 1 d
= =
dx dx dξ µ dξ
2
d 1 d 1 d 1 d2
= = .
dx2 µ dξ µ dξ µ2 dξ 2

880
The differential equation for v is

a2
 
1 00
v − −b v =0
µ2 4

a2
 
00 2
v −µ −b v =0
4
We choose
−1/2
a2

µ= −b
4
to obtain

v 00 − v = 0

which has solutions e±ξ . The solution for y is

y = eλx c1 ex/µ + c2 e−x/µ

 √ √ 
−ax/2 a2 /4−b x − a2 /4−b x
y= e c1 e + c2 e

21.2 Normal Form

21.2.1 Second Order Equations
Consider the second order equation

y 00 + p(x)y 0 + q(x)y = 0. (21.1)

881
Through a change of dependent variable, this equation can be transformed to

u00 + I(x)y = 0.

This is known as the normal form of (21.1). The function I(x) is known as the invariant of the equation.
Now to find the change of variables that will accomplish this transformation. We make the substitution
y(x) = a(x)u(x) in (21.1).

au00 + 2a0 u0 + a00 u + p(au0 + a0 u) + qau = 0

 0  00
pa0
 
00 a 0 a
u + 2 +p u + + +q u=0
a a a

To eliminate the u0 term, a(x) must satisfy

a0
2 +p=0
a

1
a0 + pa = 0
2

 Z 
1
a = c exp − p(x) dx .
2

For this choice of a, our differential equation for u becomes

p2 p0
 
00
u + q− − u = 0.
4 2

Two differential equations having the same normal form are called equivalent.

882
 Result 21.2.1 The change of variables
 Z 
1
y(x) = exp − p(x) dx u(x)
2
transforms the differential equation

y 00 + p(x)y 0 + q(x)y = 0

into its normal form

u00 + I(x)u = 0

where the invariant of the equation, I(x), is

p2 p0
I(x) = q − − .
4 2

21.2.2 Higher Order Differential Equations

Consider the third order differential equation

y 000 + p(x)y 00 + q(x)y 0 + r(x)y = 0.

883
We can eliminate the y 00 term. Making the change of dependent variable
 Z 
1
y = u exp − p(x) dx
3
   Z 
0 0 1 1
y = u − pu exp − p(x) dx
3 3
   Z 
00 00 2 0 1 2 0 1
y = u − pu + (p − 3p )u exp − p(x) dx
3 9 3
   Z 
00 000 00 1 2 0 0 1 0 00 3 1
y = u − pu + (p − 3p )u + (9p − 9p − p )u exp − p(x) dx
3 27 3

yields the differential equation

1 1
u000 + (3q − 3p0 − p2 )u0 + (27r − 9pq − 9p00 + 2p3 )u = 0.
3 27

Result 21.2.2 The change of variables

 Z 
1
y(x) = exp − pn−1 (x) dx u(x)
n
transforms the differential equation

y (n) + pn−1 (x)y (n−1) + pn−2 (x)y (n−2) + · · · + p0 (x)y = 0

into the form

u(n) + an−2 (x)u(n−2) + an−3 (x)u(n−3) + · · · + a0 (x)u = 0.

884
21.3 Transformations of the Independent Variable
21.3.1 Transformation to the form u” + a(x) u = 0
Consider the second order linear differential equation

y 00 + p(x)y 0 + q(x)y = 0.

We make the change of independent variable

ξ = f (x), u(ξ) = y(x).

The derivatives in terms of ξ are

d dξ d d
= = f0
dx dx dξ dξ
2 2
d 0 d 0 d 0 2 d d
2
= f f = (f ) 2
+ f 00
dx dξ dξ dξ dξ
The differential equation becomes

(f 0 )2 u00 + f 00 u0 + pf 0 u0 + qu = 0.

In order to eliminate the u0 term, f must satisfy

f 00 + pf 0 = 0

 Z 
0
f = exp − p(x) dx

Z  Z 
f= exp − p(x) dx dx.

885
The differential equation for u is then
q
u00 + u=0
(f 0 )2
 Z 
00
u (ξ) + q(x) exp 2 p(x) dx u(ξ) = 0.

Result 21.3.1 The change of variables

Z  Z 
ξ = exp − p(x) dx dx, u(ξ) = y(x)

transforms the differential equation

y 00 + p(x)y 0 + q(x)y = 0

into
 Z 
00
u (ξ) + q(x) exp 2 p(x) dx u(ξ) = 0.

21.3.2 Transformation to a Constant Coefficient Equation

Consider the second order linear differential equation
y 00 + p(x)y 0 + q(x)y = 0.
With the change of independent variable
ξ = f (x), u(ξ) = y(x),

886
the differential equation becomes

(f 0 )2 u00 + (f 00 + pf 0 )u0 + qu = 0.

For this to be a constant coefficient equation we must have

(f 0 )2 = c1 q, and f 00 + pf 0 = c2 q,

for some constants c1 and c2 . Solving the first condition,

√
f 0 = c q,

Z p
f =c q(x) dx.

The second constraint becomes

f 00 + pf 0
= const
q
1 −1/2 0
2
cq q + pcq 1/2
= const
q
q 0 + 2pq
= const.
q 3/2

887
 Result 21.3.2 Consider the differential equation

y 00 + p(x)y 0 + q(x)y = 0.

If the expression
q 0 + 2pq
q 3/2
is a constant then the change of variables
Z p
ξ=c q(x) dx, u(ξ) = y(x),

will yield a constant coefficient differential equation. (Here c is an arbitrary constant.)

21.4 Integral Equations

Volterra’s Equations. Volterra’s integral equation of the first kind has the form
Z x
N (x, ξ)f (ξ) dξ = f (x).
a

The Volterra equation of the second kind is

Z x
y(x) = f (x) + λ N (x, ξ)y(ξ) dξ.
a

N (x, ξ) is known as the kernel of the equation.

888
Fredholm’s Equations. Fredholm’s integral equations of the first and second kinds are
Z b
N (x, ξ)f (ξ) dξ = f (x),
a

Z b
y(x) = f (x) + λ N (x, ξ)y(ξ) dξ.
a

21.4.1 Initial Value Problems

Consider the initial value problem
y 00 + p(x)y 0 + q(x)y = f (x), y(a) = α, y 0 (a) = β.
Integrating this equation twice yields
Z xZ η Z x Z η
00 0
y (ξ) + p(ξ)y (ξ) + q(ξ)y(ξ) dξ dη = f (ξ) dξ dη
a a a a

Z x Z x
00 0
(x − ξ)[y (ξ) + p(ξ)y (ξ) + q(ξ)y(ξ)] dξ = (x − ξ)f (ξ) dξ.
a a

Now we use integration by parts.

Z x Z x
0
x 0
x
[(x − ξ)p0 (ξ) − p(ξ)]y(ξ) dξ
 
(x − ξ)y (ξ) a − −y (ξ) dξ + (x − ξ)p(ξ)y(ξ) a −
a a
Z x Z x
+ (x − ξ)q(ξ)y(ξ) dξ = (x − ξ)f (ξ) dξ.
a a

Z x
0
− (x − a)y (a) + y(x) − y(a) − (x − a)p(a)y(a) − [(x − ξ)p0 (ξ) − p(ξ)]y(ξ) dξ
a
Z x Z x
+ (x − ξ)q(ξ)y(ξ) dξ = (x − ξ)f (ξ) dξ.
a a

889
We obtain a Volterra integral equation of the second kind for y(x).
Z x Z x
(x − ξ)[p0 (ξ) − q(ξ)] − p(ξ) y(ξ) dξ.

y(x) = (x − ξ)f (ξ) dξ + (x − a)(αp(a) + β) + α +
a a

Note that the initial conditions for the differential equation are “built into” the Volterra equation. Setting
x = a in the Volterra equation yields y(a) = α. Differentiating the Volterra equation,
Z x Z x
0
y (x) = f (ξ) dξ + (αp(a) + β) − p(x)y(x) + [p0 (ξ) − q(ξ)] − p(ξ)y(ξ) dξ
a a
and setting x = a yields
y 0 (a) = αp(a) + β − p(a)α = β.
(Recall from calculus that
Z x Z x
d ∂
g(x, ξ) dξ = g(x, x) + [g(x, ξ)] dξ.)
dx ∂x

Result 21.4.1 The initial value problem

y 00 + p(x)y 0 + q(x)y = f (x), y(a) = α, y 0 (a) = β.

is equivalent to the Volterra equation of the second kind

Z x
y(x) = F (x) + N (x, ξ)y(ξ) dξ
a

where
Z x
F (x) = (x − ξ)f (ξ) dξ + (x − a)(αp(a) + β) + α
a
N (x, ξ) = (x − ξ)[p0 (ξ) − q(ξ)] − p(ξ).

890
21.4.2 Boundary Value Problems
Consider the boundary value problem
y 00 = f (x), y(a) = α, y(b) = β. (21.2)
To obtain a problem with homogeneous boundary conditions, we make the change of variable
β−α
y(x) = u(x) + α + (x − a)
b−a
to obtain the problem
u00 = f (x), u(a) = u(b) = 0.
Now we will use Green’s functions to write the solution as an integral. First we solve the problem
G00 = δ(x − ξ), G(a|ξ) = G(b|ξ) = 0.
The homogeneous solutions of the differential equation that satisfy the left and right boundary conditions are
c1 (x − a) and c2 (x − b).
Thus the Green’s function has the form
(
c1 (x − a), for x ≤ ξ
G(x|ξ) =
c2 (x − b), for x ≥ ξ
Imposing continuity of G(x|ξ) at x = ξ and a unit jump of G(x|ξ) at x = ξ, we obtain
(
(x−a)(ξ−b)
b−a
, for x ≤ ξ
G(x|ξ) = (x−b)(ξ−a)
b−a
, for x ≥ ξ
Thus the solution of the (21.2) is
b
β−α
Z
y(x) = α + (x − a) + G(x|ξ)f (ξ) dξ.
b−a a

891
 Now consider the boundary value problem

y 00 + p(x)y 0 + q(x)y = 0, y(a) = α, y(b) = β.

From the above result we can see that the solution satisfies

b
β−α
Z
y(x) = α + (x − a) + G(x|ξ)[f (ξ) − p(ξ)y 0 (ξ) − q(ξ)y(ξ)] dξ.
b−a a

Using integration by parts, we can write

Z b Z b 
0
 b ∂G(x|ξ) 0
− G(x|ξ)p(ξ)y (ξ) dξ = − G(x|ξ)p(ξ)y(ξ) a + p(ξ) + G(x|ξ)p (ξ) y(ξ) dξ
a a ∂ξ
Z b 
∂G(x|ξ) 0
= p(ξ) + G(x|ξ)p (ξ) y(ξ) dξ.
a ∂ξ

Substituting this into our expression for y(x),

b Z b 
β−α
Z
∂G(x|ξ) 0
y(x) = α + (x − a) + G(x|ξ)f (ξ) dξ + p(ξ) + G(x|ξ)[p (ξ) − q(ξ)] y(ξ) dξ,
b−a a a ∂ξ

we obtain a Fredholm integral equation of the second kind.

892
Result 21.4.2 The boundary value problem

y 00 + p(x)y 0 + q(x)y = f (x), y(a) = α, y(b) = β.

is equivalent to the Fredholm equation of the second kind

Z b
y(x) = F (x) + N (x, ξ)y(ξ) dξ
a

where
Z b
β−α
F (x) = α + (x − a) + G(x|ξ)f (ξ) dξ,
b−a a
Z b
N (x, ξ) = H(x|ξ)y(ξ) dξ,
(a
(x−a)(ξ−b)
b−a , for x ≤ ξ
G(x|ξ) = (x−b)(ξ−a)
, for x ≥ ξ,
( b−a
(x−a) (x−a)(ξ−b) 0
b−a p(ξ) + b−a [p (ξ) − q(ξ)] for x ≤ ξ
H(x|ξ) = (x−b) (x−b)(ξ−a) 0
b−a p(ξ) + b−a [p (ξ) − q(ξ)] for x ≥ ξ.

893
21.5 Exercises
The Constant Coefficient Equation
Normal Form
Exercise 21.1
Solve the differential equation
 
4 1
y + 2 + x y0 +
00
24 + 12x + 4x2 y = 0.


3 9

Hint, Solution

Transformations of the Independent Variable

Integral Equations
Exercise 21.2
Show that the solution of the differential equation

y 00 + 2(a + bx)y 0 + (c + dx + ex2 )y = 0

can be written in terms of one of the following canonical forms:

v 00 + (ξ 2 + A)v = 0
v 00 = ξv
v 00 + v = 0
v 00 = 0.

Hint, Solution

894
Exercise 21.3
Show that the solution of the differential equation
   
00 b 0 d e
y +2 a+ y + c+ + 2 y =0
x x x
can be written in terms of one of the following canonical forms:
 
00 A B
v + 1+ + 2 v =0
ξ ξ
 
1 A
v 00 + + v=0
ξ ξ2
A
v 00 + 2 v = 0
ξ
Hint, Solution

Exercise 21.4
Show that the second order Euler equation
2
y2d dy
x 2
+ a1 x + a0 y = 0
dx dx
can be transformed to a constant coefficient equation.
Hint, Solution

Exercise 21.5
Solve Bessel’s equation of order 1/2,
 
1 0
00 1
y + y + 1 − 2 y = 0.
x 4x
Hint, Solution

895
21.6 Hints
The Constant Coefficient Equation
Normal Form
Hint 21.1
Transform the equation to normal form.

Transformations of the Independent Variable

Integral Equations
Hint 21.2
Transform the equation to normal form and then apply the scale transformation x = λξ + µ.

Hint 21.3
Transform the equation to normal form and then apply the scale transformation x = λξ.

Hint 21.4
Make the change of variables x = et , y(x) = u(t). Write the derivatives with respect to x in terms of t.
x = et
dx = et dt
d d
= e−t
dx dt
d d
x =
dx dt

Hint 21.5
Transform the equation to normal form.

896
21.7 Solutions
The Constant Coefficient Equation
Normal Form
Solution 21.1

 
4 1
y + 2 + x y0 +
00
24 + 12x + 4x2 y = 0


3 9
To transform the equation to normal form we make the substitution
 Z   
1 4
y = exp − 2 + x dx u
2 3
2 /3
= e−x−x u
The invariant of the equation is
 2  
1
1 4 1 d 4
I(x) = 24 + 12x + 4x2 − 2+ x − 2+ x
9 4 3 2 dx 3
= 1.
The normal form of the differential equation is then
u00 + u = 0
which has the general solution
u = c1 cos x + c2 sin x
Thus the equation for y has the general solution
2 /3 2 /3
y = c1 e−x−x cos x + c2 e−x−x sin x.

897
Transformations of the Independent Variable
Integral Equations
Solution 21.2
The substitution that will transform the equation to normal form is
 Z 
1
y = exp − 2(a + bx) dx u
2
2 /2
= e−ax−bx u.

The invariant of the equation is

1 1 d
I(x) = c + dx + ex2 − (2(a + bx))2 − (2(a + bx))
4 2 dx
= c − b − a2 + (d − 2ab)x + (e − b2 )x2
≡ α + βx + γx2

The normal form of the differential equation is

u00 + (α + βx + γx2 )u = 0

We consider the following cases:

γ = 0.
β = 0.
α = 0. We immediately have the equation

u00 = 0.

α 6= 0. With the change of variables

v(ξ) = u(x), x = α−1/2 ξ,

898
 we obtain

v 00 + v = 0.

β 6= 0. We have the equation

y 00 + (α + βx)y = 0.

The scale transformation x = λξ + µ yields

v 00 + λ2 (α + β(λξ + µ))y = 0

v 00 = [βλ3 ξ + λ2 (βµ + α)]v.

Choosing
α
λ = (−β)−1/3 , µ=−
β
yields the differential equation

v 00 = ξv.

γ 6= 0. The scale transformation x = λξ + µ yields

v 00 + λ2 [α + β(λξ + µ) + γ(λξ + µ)2 ]v = 0

v 00 + λ2 [α + βµ + γµ2 + λ(β + 2γµ)ξ + λ2 γξ 2 ]v = 0.

Choosing
β
λ = γ −1/4 , µ=−
2γ

899
 yields the differential equation

v 00 + (ξ 2 + A)v = 0

where
1
A = γ −1/2 − βγ −3/2 .
4

Solution 21.3
The substitution that will transform the equation to normal form is
 Z   
1 b
y = exp − 2 a+ dx u
2 x
= x−b e−ax u.

The invariant of the equation is

  2   
d e 1 b 1 d b
I(x) = c + + 2 − 2 a+ − 2 a+
x x 4 x 2 dx x
2
d − 2ab e + b − b
= c − ax + +
x x2
β γ
≡ α + + 2.
x x
The invariant form of the differential equation is
 
00 β γ
u + α + + 2 u = 0.
x x

We consider the following cases:

900
α = 0.

β = 0. We immediately have the equation

γ
u00 + u = 0.
x2

β 6= 0. We have the equation

 
00 β γ
u + + 2 u = 0.
x x

The scale transformation u(x) = v(ξ), x = λξ yields

 
00 βλ γ
v + + 2 u = 0.
ξ ξ

Choosing λ = β −1 , we obtain
 
00 1 γ
v + + 2 u = 0.
ξ ξ

α 6= 0. The scale transformation x = λξ yields

 
00 2 βλ γ
v + αλ + + 2 v = 0.
ξ ξ

Choosing λ = α−1/2 , we obtain

α−1/2 β
 
00 γ
v + 1+ + 2 v = 0.
ξ ξ

901
Solution 21.4
We write the derivatives with respect to x in terms of t.

x = et
dx = et dt
d d
= e−t
dx dt
d d
x =
dx dt
d 2
Now we express x2 dx 2 in terms of t.

d2 d2
 
2 d d d d
x 2
=x x −x = 2−
dx dx dx dx dt dt

Thus under the change of variables, x = et , y(x) = u(t), the Euler equation becomes

u00 − u0 + a1 u0 + a0 u = 0
u00 + (a1 − 1)u0 + a0 u = 0.

Solution 21.5
The transformation
 Z 
1 1
y = exp − dx = x−1/2 u
2 x

will put the equation in normal form. The invariant is

   
1 1 1 1 −1
I(x) = 1 − 2 − 2
− = 1.
4x 4 x 2 x2

902
Thus we have the differential equation

u00 + u = 0,

with the solution

u = c1 cos x + c2 sin x.

The solution of Bessel’s equation of order 1/2 is

y = c1 x−1/2 cos x + c2 x−1/2 sin x.

903
Chapter 22

The Dirac Delta Function

I do not know what I appear to the world; but to myself I seem to have been only like a boy playing on a
seashore, and diverting myself now and then by finding a smoother pebble or a prettier shell than ordinary, whilst
the great ocean of truth lay all undiscovered before me.
- Sir Issac Newton

22.1 Derivative of the Heaviside Function

The Heaviside function H(x) is defined
(
0 for x < 0,
H(x) =
1 for x > 0.
The derivative of the Heaviside function is zero for x 6= 0. At x = 0 the derivative is undefined. We will represent
the derivative of the Heaviside function by the Dirac delta function, δ(x). The delta function is zero for x 6= 0 and
infinite at the point x = 0. Since the derivative of H(x) is undefined, δ(x) is not a function in the conventional
sense of the word. One can derive the properties of the delta function rigorously, but the treatment in this text
will be almost entirely heuristic.

904
 The Dirac delta function is defined by the properties
(
∞
for x 6= 0,
Z
0
δ(x) = and δ(x) dx = 1.
∞ for x = 0, −∞

The second property comes from the fact that δ(x) represents the derivative of H(x). The Dirac delta function
is conceptually pictured in Figure 22.1.

Figure 22.1: The Dirac Delta Function.

Let f (x) be a continuous function that vanishes at infinity. Consider the integral
Z ∞
f (x)δ(x) dx.
−∞

905
Using integration by parts,
Z ∞ ∞
Z ∞
f 0 (x)H(x) dx

f (x)δ(x) dx = f (x)H(x) −∞ −
−∞ −∞
Z ∞
=− f 0 (x) dx
0
= [−f (x)]∞
0
= f (0).
We assumed that f (x) vanishes at infinity in order to use integration by parts to evaluate the integral. However,
since the delta function is zero for x 6= 0, the integrand is nonzero only at x = 0. Thus the
R∞behavior of the function
at infinity should not affect the value of the integral. Thus it is reasonable that f (0) = −∞ f (x)δ(x) dx holds for
all continuous functions. Changing variables and noting that δ(x) is symmetric we have
Z ∞
f (x) = f (ξ)δ(x − ξ) dξ.
−∞

This formula is very important in solving inhomogeneous differential equations.

22.2 The Delta Function as a Limit

Consider a function b(x, ) defined by
(
0 for |x| > /2
b(x, ) = 1

for |x| < /2.
The graph of b(x, 1/10) is shown in Figure 22.2.
The Dirac delta function δ(x) can be thought of as b(x, ) in the limit as  → 0. Note that the delta function
so defined satisfies the properties,
( Z ∞
0 for x 6= 0
δ(x) = and δ(x) dx = 1
∞ for x = 0 −∞

906
 10

-1 1
Figure 22.2: Graph of b(x, 1/10).

Delayed Limiting Process. When the Dirac delta function appears inside an integral, we can think of the
delta function as a delayed limiting process. That is,
Z ∞ Z ∞
f (x)δ(x) dx ≡ lim f (x)b(x, ) dx.
−∞ →0 −∞

Let f (x) be a continuous function and let F 0 (x) = f (x). The integral of f (x)δ(x) is then
Z ∞
1 /2
Z
f (x)δ(x) dx = lim f (x) dx
−∞ →0  −/2

1 /2
= lim [F (x)]−/2
→0 
F (/2) − F (−/2)
= lim
→0 
0
= F (0)
= f (0).

907
22.3 Higher Dimensions
We can define a Dirac delta function in n-dimensional Cartesian space, δn (x), x ∈ Rn . It is defined by the
following two properties.

δn (x) = 0 for x 6= 0
Z
δn (x) dx = 1
Rn

It is easy to verify, that the n-dimensional Dirac delta function can be written as a product of 1-dimensional Dirac
delta functions.
n
Y
δn (x) = δ(xk )
k=1

22.4 Non-Rectangular Coordinate Systems

We can derive Dirac delta functions in non-rectangular coordinate systems by making a change of variables in
the relation,
Z
δn (x) dx = 1
Rn

Where the transformation is non-singular, one merely divides the Dirac delta function by the Jacobian of the
transformation to the coordinate system.

Example 22.4.1 Consider the Dirac delta function in cylindrical coordinates, (r, θ, z). The Jacobian is J = r.
Z ∞ Z 2π Z ∞
δ3 (x − x0 ) r dr dθ dz = 1
−∞ 0 0

908
For r0 6= 0, the Dirac Delta function is
1
δ3 (x − x0 ) = δ (r − r0 ) δ (θ − θ0 ) δ (z − z0 )
r
since it satisfies the two defining properties.
1
δ (r − r0 ) δ (θ − θ0 ) δ (z − z0 ) = 0 for (r, θ, z) 6= (r0 , θ0 , z0 )
r

Z ∞ Z 2π Z ∞
1
δ (r − r0 ) δ (θ − θ0 ) δ (z − z0 ) r dr dθ dz
−∞ 0 0 r
Z ∞ Z 2π Z ∞
= δ (r − r0 ) dr δ (θ − θ0 ) dθ δ (z − z0 ) dz = 1
0 0 −∞

For r0 = 0, we have
1
δ3 (x − x0 ) = δ (r) δ (z − z0 )
2πr
since this again satisfies the two defining properties.
1
δ (r) δ (z − z0 ) = 0 for (r, z) 6= (0, z0 )
2πr
Z ∞ Z 2π Z ∞ Z ∞ Z 2π Z ∞
1 1
δ (r) δ (z − z0 ) r dr dθ dz = δ (r) dr dθ δ (z − z0 ) dz = 1
−∞ 0 0 2πr 2π 0 0 −∞

909
22.5 Exercises
Exercise 22.1
Let f (x) be a function that is continuous except for a jump discontinuity at x = 0. Using a delayed limiting
process, show that
Z ∞
f (0− ) + f (0+ )
= f (x)δ(x) dx.
2 −∞

Hint, Solution

Exercise 22.2
Let y = y(x) be defined on some interval Assume y(x) is continuously differentiable and that y 0 (x) 6= 0. Show
that
 −1
dy
δ(x − x0 ) = δ(y − y0 )
dx

where y0 = y(x0 ).
Hint, Solution

Exercise 22.3
Determine the Dirac delta function in spherical coordinates, (r, θ, φ).

x = r cos θ sin φ, y = r sin θ sin φ, z = r cos φ

Hint, Solution

910
22.6 Hints
Hint 22.1

Hint 22.2
Make a change of variables in the integral
Z
δ(x − x0 ) dx.

Hint 22.3
Consider the special cases φ0 = 0, π and r0 = 0.

911
22.7 Solutions
Solution 22.1
Let F 0 (x) = f (x).
Z ∞ Z ∞
1
f (x)δ(x) dx = lim f (x)b(x, ) dx
−∞ →0  −∞
!
Z 0 Z /2
1
= lim f (x)b(x, ) dx + f (x)b(x, ) dx
→0  −/2 0
1
= lim ([F (0) − F (−/2)] + [F (/2) − F (0)])
→0 
 
1 F (0) − F (−/2) F (/2) − F (0)
= lim +
→0 2 /2 /2
0 − 0 +
F (0 ) + F (0 )
=
2
f (0− ) + f (0+ )
=
2

Solution 22.2
We prove the identity by making a change of variables in the integral of δ(x − x0 ).

Z b Z y(b)  −1
dy
δ(x − x0 ) dx = δ(y − y0 ) dy
a y(a) dx
 −1
dy
δ(x − x0 ) = δ(y − y0 )
dx

912
Solution 22.3
We consider the Dirac delta function in spherical coordinates, (r, θ, φ). The Jacobian is J = r2 sin(φ).
Z π Z 2π Z ∞
δ3 (x − x0 ) r2 sin(φ) dr dθ dφ = 1
0 0 0

For r0 6= 0, and φ0 6= 0, π, the Dirac Delta function is

1
δ3 (x − x0 ) = δ (r − r0 ) δ (θ − θ0 ) δ (φ − φ0 )
r2 sin(φ)
since it satisfies the two defining properties.
1
δ (r − r0 ) δ (θ − θ0 ) δ (φ − φ0 ) = 0 for (r, θ, φ) 6= (r0 , θ0 , φ0 )
r2 sin(φ)

Z π Z 2π Z ∞
1
δ (r − r0 ) δ (θ − θ0 ) δ (φ − φ0 ) r2 sin(φ) dr dθ dφ
0 0 0 r2 sin(φ)
Z ∞ Z 2π Z π
= δ (r − r0 ) dr δ (θ − θ0 ) dθ δ (φ − φ0 ) dφ = 1
0 0 0

For φ0 = 0 or φ0 = π, the Dirac delta function is

1
δ3 (x − x0 ) = δ (r − r0 ) δ (φ − φ0 ) .
2πr2 sin(φ)
We check that the value of the integral is unity.
Z π Z 2π Z ∞
1
δ (r − r0 ) δ (φ − φ0 ) r2 sin(φ) dr dθ dφ
0 0 0 2πr2 sin(φ)
Z ∞ Z 2π Z π
1
= δ (r − r0 ) dr dθ δ (φ − φ0 ) dφ = 1
2π 0 0 0

913
For r0 = 0 the Dirac delta function is
1
δ3 (x) = δ (r)
4πr2
We verify that the value of the integral is unity.
Z π Z 2π Z ∞ Z ∞ Z 2π Z π
1 2 1
δ (r − r0 ) r sin(φ) dr dθ dφ = δ (r) dr dθ sin(φ) dφ = 1
0 0 0 4πr2 4π 0 0 0

914
Chapter 23

Inhomogeneous Differential Equations

Feelin’ stupid? I know I am!

-Homer Simpson

23.1 Particular Solutions

Consider the nth order linear homogeneous equation
L[y] ≡ y (n) + pn−1 (x)y (n−1) + · · · + p1 (x)y 0 + p0 (x)y = 0.
Let {y1 , y2 , . . . , yn } be a set of linearly independent homogeneous solutions, L[yk ] = 0. We know that the general
solution of the homogeneous equation is a linear combination of the homogeneous solutions.
n
X
yh = ck yk (x)
k=1

Now consider the nth order linear inhomogeneous equation

L[y] ≡ y (n) + pn−1 (x)y (n−1) + · · · + p1 (x)y 0 + p0 (x)y = f (x).

915
Any function yp which satisfies this equation is called a particular solution of the differential equation. We want
to know the general solution of the inhomogeneous equation. Later in this chapter we will cover methods of
constructing this solution; now we consider the form of the solution.
Let yp be a particular solution. Note that yp +h is a particular solution if h satisfies the homogeneous equation.

L[yp + h] = L[yp ] + L[h] = f + 0 = f

Therefore yp + yh satisfies the homogeneous equation. We show that this is the general solution of the inhomoge-
neous equation. Let yp and ηp both be solutions of the inhomogeneous equation L[y] = f . The difference of yp
and ηp is a homogeneous solution.

L[yp − ηp ] = L[yp ] − L[ηp ] = f − f = 0

yp and ηp differ by a linear combination of the homogeneous solutions {yk }. Therefore the general solution of
L[y] = f is the sum of any particular solution yp and the general homogeneous solution yh .
n
X
yp + yh = yp (x) + ck yk (x)
k=1

Result 23.1.1 The general solution of the nth order linear inhomogeneous equation
L[y] = f (x) is

y = yp + c1 y1 + c2 y2 + · · · + cn yn ,

where yp is a particular solution, {y1 , . . . , yn } is a set of linearly independent homogeneous

solutions, and the ck ’s are arbitrary constants.

Example 23.1.1 The differential equation

y 00 + y = sin(2x)

916
has the two homogeneous solutions

y1 = cos x, y2 = sin x,

and a particular solution

1
yp = − sin(2x).
3
We can add any combination of the homogeneous solutions to yp and it will still be a particular solution. For
example,
1 1
ηp = − sin(2x) − sin x
3  3  
2 3x x
= − sin cos
3 2 2
is a particular solution.

23.2 Method of Undetermined Coefficients

The first method we present for computing particular solutions is the method of undetermined coefficients. For
some simple differential equations, (primarily constant coefficient equations), and some simple inhomogeneities
we are able to guess the form of a particular solution. This form will contain some unknown parameters. We
substitute this form into the differential equation to determine the parameters and thus determine a particular
solution.
Later in this chapter we will present general methods which work for any linear differential equation and any
inhogeneity. Thus one might wonder why I would present a method that works only for some simple problems.
(And why it is called a “method” if it amounts to no more than guessing.) The answer is that guessing an answer
is less grungy than computing it with the formulas we will develop later. Also, the process of this guessing is not
random, there is rhyme and reason to it.

917
 Consider an nth order constant coefficient, inhomogeneous equation.

L[y] ≡ y (n) + an−1 y (n−1) + · · · + a1 y 0 + a0 y = f (x)

If f (x) is one of a few simple forms, then we can guess the form of a particular solution. Below we enumerate
some cases.

f = p(x). If f is an mth order polynomial, f (x) = pm xm + · · · + p1 x + p0 , then guess

y p = c m xm + · · · c 1 x + c 0 .

f = p(x) eax . If f is a polynomial times an exponential then guess

yp = (cm xm + · · · c1 x + c0 ) eax .

f = p(x) eax cos (bx). If f is a cosine or sine times a polynomial and perhaps an exponential, f (x) = p(x) eax cos(bx)
or f (x) = p(x) eax sin(bx) then guess

yp = (cm xm + · · · c1 x + c0 ) eax cos(bx) + (dm xm + · · · d1 x + d0 ) eax sin(bx).

Likewise for hyperbolic sines and hyperbolic cosines.

Example 23.2.1 Consider

y 00 − 2y 0 + y = t2 .

The homogeneous solutions are y1 = et and y2 = t et . We guess a particular solution of the form

yp = at2 + bt + c.

918
We substitute the expression into the differential equation and equate coefficients of powers of t to determine the
parameters.
yp00 − 2yp0 + yp = t2
(2a) − 2(2at + b) + (at2 + bt + c) = t2
(a − 1)t2 + (b − 4a)t + (2a − 2b + c) = 0
a − 1 = 0, b − 4a = 0, 2a − 2b + c = 0
a = 1, b = 4, c = 6
A particular solution is
yp = t2 + 4t + 6.

If the inhomogeneity is a sum of terms, L[y] = f ≡ f1 +· · ·+fk , you can solve the problems L[y] = f1 , . . . , L[y] =
fk independently and then take the sum of the solutions as a particular solution of L[y] = f .

Example 23.2.2 Consider

L[y] ≡ y 00 − 2y 0 + y = t2 + e2t . (23.1)
The homogeneous solutions are y1 = et and y2 = t et . We already know a particular solution to L[y] = t2 . We
seek a particular solution to L[y] = e2t . We guess a particular solution of the form
yp = a e2t .
We substitute the expression into the differential equation to determine the parameter.
yp00 − 2yp0 + yp = e2t
4ae2t − 4a e2t + a e2t = e2t
a=1
A particular solution of L[y] = e2t is yp = e2t . Thus a particular solution of Equation 23.1 is
yp = t2 + 4t + 6 + e2t .

919
 The above guesses will not work if the inhomogeneity is a homogeneous solution. In this case, multiply the
guess by the lowest power of x such that the guess does not contain homogeneous solutions.

Example 23.2.3 Consider

L[y] ≡ y 00 − 2y 0 + y = et .
The homogeneous solutions are y1 = et and y2 = t et . Guessing a particular solution of the form yp = a et would
not work because L[ et ] = 0. We guess a particular solution of the form
yp = at2 et
We substitute the expression into the differential equation and equate coefficients of like terms to determine the
parameters.
yp00 − 2yp0 + yp = et
(at2 + 4at + 2a) et − 2(at2 + 2at) et + at2 et = et
2a et = et
1
a=
2
A particular solution is
t2 t
yp = e.
2

Example 23.2.4 Consider

1 1
y 00 + y 0 + 2 y = x, x > 0.
x x
The homogeneous solutions are y1 = cos(ln x) and y2 = sin(ln x). We guess a particular solution of the form
yp = ax3

920
We substitute the expression into the differential equation and equate coefficients of like terms to determine the
parameter.
1 1
yp00 + yp0 + 2 yp = x
x x
6ax + 3ax + ax = x
1
a=
10
A particular solution is

x3
yp = .
10

23.3 Variation of Parameters

In this section we present a method for computing a particular solution of an inhomogeneous equation given
that we know the homogeneous solutions. We will first consider second order equations and then generalize the
result for nth order equations.

23.3.1 Second Order Differential Equations

Consider the second order inhomogeneous equation,

L[y] ≡ y 00 + p(x)y 0 + q(x)y = f (x), on a < x < b.

We assume that the coefficient functions in the differential equation are continuous on [a . . . b]. Let y1 (x) and
y2 (x) be two linearly independent solutions to the homogeneous equation. Since the Wronskian,
 Z 
W (x) = exp − p(x) dx ,

921
is non-vanishing, we know that these solutions exist. We seek a particular solution of the form,

yp = u1 (x)y1 + u2 (x)y2 .

We compute the derivatives of yp .

yp0 = u01 y1 + u1 y10 + u02 y2 + u2 y20

yp00 = u001 y1 + 2u01 y10 + u1 y100 + u002 y2 + 2u02 y20 + u2 y200

We substitute the expression for yp and its derivatives into the inhomogeneous equation and use the fact that y1
and y2 are homogeneous solutions to simplify the equation.

u001 y1 + 2u01 y10 + u1 y100 + u002 y2 + 2u02 y20 + u2 y200 + p(u01 y1 + u1 y10 + u02 y2 + u2 y20 ) + q(u1 y1 + u2 y2 ) = f
u001 y1 + 2u01 y10 + u002 y2 + 2u02 y20 + p(u01 y1 + u02 y2 ) = f

This is an ugly equation for u1 and u2 , however, we have an ace up our sleeve. Since u1 and u2 are undetermined
functions of x, we are free to impose a constraint. We choose this constraint to simplify the algebra.

u01 y1 + u02 y2 = 0

This constraint simplifies the derivatives of yp ,

yp0 = u01 y1 + u1 y10 + u02 y2 + u2 y20

= u1 y10 + u2 y20
yp00 = u01 y10 + u1 y100 + u02 y20 + u2 y200 .

We substitute the new expressions for yp and its derivatives into the inhomogeneous differential equation to obtain
a much simpler equation than before.

u01 y10 + u1 y100 + u02 y20 + u2 y200 + p(u1 y10 + u2 y20 ) + q(u1 y1 + u2 y2 ) = f (x)
u01 y10 + u02 y20 + u1 L[y1 ] + u2 L[y2 ] = f (x)
u01 y10 + u02 y20 = f (x).

922
 With the constraint, we have a system of linear equations for u01 and u02 .
u01 y1 + u02 y2 = 0
u01 y10 + u02 y20 = f (x).
  0  
y1 y2 u1 0
0 0 0 =
y1 y2 u2 f
We solve this system using Kramer’s rule. (See Appendix S.)
f (x)y2 f (x)y1
u01 = − u02 =
W (x) W (x)
Here W (x) is the Wronskian.
 
 y1 y2 
W (x) =  0 
y1 y20 
We integrate to get u1 and u2 . This gives us a particular solution.
Z Z
f (x)y2 (x) f (x)y1 (x)
yp = −y1 dx + y2 dx.
W (x) W (x)

Result 23.3.1 Let y1 and y2 be linearly independent homogeneous solutions of

L[y] = y 00 + p(x)y 0 + q(x)y = f (x).

A particular solution is
Z Z
f (x)y2 (x) f (x)y1 (x)
yp = −y1 (x) dx + y2 (x) dx,
W (x) W (x)
where W (x) is the Wronskian of y1 and y2 .

923
Example 23.3.1 Consider the equation,

y 00 + y = cos(2x).

The homogeneous solutions are y1 = cos x and y2 = sin x. We compute the Wronskian.
 
 cos x sin x 
W (x) =   = cos2 x + sin2 x = 1
− sin x cos x

We use variation of parameters to find a particular solution.

Z Z
yp = − cos(x) cos(2x) sin(x) dx + sin(x) cos(2x) cos(x) dx
Z Z
1
1

= − cos(x) sin(3x) − sin(x) dx + sin(x) cos(3x) + cos(x) dx
2 2
   
1 1 1 1
= − cos(x) − cos(3x) + cos(x) + sin(x) sin(3x) + sin(x)
2 3 2 3
1
1
sin2 (x) − cos2 (x) +


= cos(3x) cos(x) + sin(3x) sin(x)
2 6
1 1
= − cos(2x) + cos(2x)
2 6
1
= − cos(2x)
3

The general solution of the inhomogeneous equation is

1
y = − cos(2x) + c1 cos(x) + c2 sin(x).
3

924
23.3.2 Higher Order Differential Equations
Consider the nth order inhomogeneous equation,
L[y] = y(n) + pn−1 (x)y (n−1) + · · · + p1 (x)y 0 + p0 (x)y = f (x), on a < x < b.
We assume that the coefficient functions in the differential equation are continuous on [a . . . b]. Let {y1 , . . . , yn }
be a set of linearly independent solutions to the homogeneous equation. Since the Wronskian,
 Z 
W (x) = exp − pn−1 (x) dx ,

is non-vanishing, we know that these solutions exist. We seek a particular solution of the form
yp = u1 y1 + u2 y2 + · · · + un yn .
Since {u1 , . . . , un } are undetermined functions of x, we are free to impose n − 1 constraints. We choose these
constraints to simplify the algebra.
u01 y1 +u02 y2 + · · ·+u0n yn =0
u01 y10 +u02 y20 + · · ·+u0n yn0 =0
.. . . .
. + .. + .. + .. =0
(n−2) (n−2)
u01 y1 +u02 y2 + · · ·+u0n yn(n−2) =0
We differentiate the expression for yp , utilizing our constraints.
yp =u1 y1 +u2 y2 + · · ·+un yn
yp0 =u1 y10 +u2 y20 + · · ·+un yn0
yp00 =u1 y100 +u2 y200 + · · ·+un yn00
.. . . . .
. = .. + .. + .. + ..
(n) (n) (n−1) (n−1)
yp(n) =u1 y1 +u2 y2 + · · ·+un yn(n) + u01 y1 + u02 y2 + · · · + u0n yn(n−1)

925
We substitute yp and its derivatives into the inhomogeneous differential equation and use the fact that the yk are
homogeneous solutions.

(n) (n−1) (n−1)

u1 y1 + · · · + un yn(n) + u01 y1 + · · · + u0n yn(n−1) + pn−1 (u1 y1 + · · · + un yn(n−1) ) + · · · + p0 (u1 y1 + · · · un yn ) = f
(n−1) (n−1)
u1 L[y1 ] + u2 L[y2 ] + · · · + un L[yn ] + u01 y1 + u02 y2 + · · · + u0n yn(n−1) = f
(n−1) (n−1)
u01 y1 + u02 y2 + · · · + u0n yn(n−1) = f.

With the constraints, we have a system of linear equations for {u1 , . . . , un }.

    
y1 y2 ··· yn u01 0
 y10 y20 ··· yn0   u0   .. 
  2  . 
.. .. ..   ..  =   .

..
.

 . . .   .  0
(n−1) (n−1) (n−1)
y1 y2 · · · yn u0n f

We solve this system using Kramer’s rule. (See Appendix S.)

W [y1 , . . . , yk−1 , yk+1 , . . . , yn ]

u0k = (−1)n+k+1 f, for k = 1, . . . , n,
W [y1 , y2 , . . . , yn ]

Here W is the Wronskian.

We integrating to obtain the uk ’s.

Z
n+k+1 W [y1 , . . . , yk−1 , yk+1 , . . . , yn ](x)
uk = (−1) f (x) dx, for k = 1, . . . , n
W [y1 , y2 , . . . , yn ](x)

926
 Result 23.3.2 Let {y1 , . . . , yn } be linearly independent homogeneous solutions of

L[y] = y(n) + pn−1 (x)y (n−1) + · · · + p1 (x)y 0 + p0 (x)y = f (x), on a < x < b.

A particular solution is

yp = u1 y1 + u2 y2 + · · · + un yn .

where
Z
n+k+1 W [y1 , . . . , yk−1 , yk+1 , . . . , yn ](x)
uk = (−1) f (x) dx, for k = 1, . . . , n,
W [y1 , y2 , . . . , yn ](x)
and W [y1 , y2 , . . . , yn ](x) is the Wronskian of {y1 (x), . . . , yn (x)}.

23.4 Piecewise Continuous Coefficients and Inhomogeneities

Example 23.4.1 Consider the problem

y 00 − y = e−α|x| , y(±∞) = 0, α > 0, α 6= 1.

927
The homogeneous solutions of the differential equation are ex and e−x . We use variation of parameters to find a
particular solution for x > 0.
Z x −ξ −αξ Z x ξ −αξ
x e e −x e e
yp = − e dξ + e dξ
−2 −2
1 x x −(α+1)ξ 1 −x x (1−α)ξ
Z Z
= e e dξ − e e dξ
2 2
1 1
=− e−αx + e−αx
2(α + 1) 2(α − 1)
−αx
e
= 2 , for x > 0
α −1
A particular solution for x < 0 is
eαx
yp = , for x < 0.
α2 − 1
Thus a particular solution is
e−α|x|
yp = 2 .
α −1
The general solution is
1
y= e−α|x| + c1 ex + c2 e−x .
α2 −1
Applying the boundary conditions, we see that c1 = c2 = 0. Apparently the solution is
e−α|x|
y= .
α2 − 1
This function is plotted in Figure 23.1. This function satisfies the differential equation for positive and negative
x. It also satisfies the boundary conditions. However, this is NOT a solution to the differential equation. Since

928
the differential equation has no singular points and the inhomogeneous term is continuous, the solution must be
twice continuously differentiable. Since the derivative of e−α|x| /(α2 − 1) has a jump discontinuity at x = 0, the
second derivative does not exist. Thus this function could not possibly be a solution to the differential equation.
In the next example we examine the right way to solve this problem.

-4 -2 2 4
0.3
-0.05
0.25
-0.1
0.2
-0.15
0.15
-0.2
0.1
-0.25
0.05
-0.3

-4 -2 2 4

Figure 23.1: The Incorrect and Correct Solution to the Differential Equation.

Example 23.4.2 Again consider

y 00 − y = e−α|x| , y(±∞) = 0, α > 0, α 6= 1.

Separating this into two problems for positive and negative x,

00
y− − y− = eαx , y− (−∞) = 0, on − ∞ < x ≤ 0,
00 −αx
y+ − y+ = e , y+ (∞) = 0, on 0 ≤ x < ∞.

929
In order for the solution over the whole domain to be twice differentiable, the solution and it’s first derivative
must be continuous. Thus we impose the additional boundary conditions
0 0
y− (0) = y+ (0), y− (0) = y+ (0).

The solutions that satisfy the two differential equations and the boundary conditions at infinity are

eαx e−αx
y− = 2
+ c− e x , y+ = 2
+ c+ e−x .
α −1 α −1
The two additional boundary conditions give us the equations

y− (0) = y+ (0) → c− = c+
0 0 α α
y− (0) = y+ (0) → + c − = − − c+ .
α2 − 1 α2 − 1
We solve these two equations to determine c− and c+ .
α
c− = c+ = −
α2 −1
Thus the solution over the whole domain is
(
eαx −α ex
α2 −1
for x < 0,
y= e−αx −α e−x
α2 −1
for x > 0

e−α|x| − α e−|x|
y= .
α2 − 1

This function is plotted in Figure 23.1. You can verify that this solution is twice continuously differentiable.

930
23.5 Inhomogeneous Boundary Conditions
23.5.1 Eliminating Inhomogeneous Boundary Conditions
Consider the nth order equation

L[y] = f (x), for a < x < b,

subject to the linear inhomogeneous boundary conditions

Bj [y] = γj , for j = 1, . . . , n,

where the boundary conditions are of the form

B[y] ≡ α0 y(a) + α1 y 0 (a) + · · · + yn−1 y (n−1) (a) + β0 y(b) + β1 y 0 (b) + · · · + βn−1 y (n−1)

Let g(x) be an n-times continuously differentiable function that satisfies the boundary conditions. Substituting
y = u + g into the differential equation and boundary conditions yields

L[u] = f (x) − L[g], Bj [u] = bj − Bj [g] = 0 for j = 1, . . . , n.

Note that the problem for u has homogeneous boundary conditions. Thus a problem with inhomogeneous bound-
ary conditions can be reduced to one with homogeneous boundary conditions. This technique is of limited
usefulness for ordinary differential equations but is important for solving some partial differential equation prob-
lems.

Example 23.5.1 Consider the problem

y 00 + y = cos 2x, y(0) = 1, y(π) = 2.

x
g(x) = π
+ 1 satisfies the boundary conditions. Substituting y = u + g yields
x
u00 + u = cos 2x − − 1, y(0) = y(π) = 0.
π

931
Example 23.5.2 Consider
y 00 + y = cos 2x, y 0 (0) = y(π) = 1.
g(x) = sin x − cos x satisfies the inhomogeneous boundary conditions. Substituting y = u + sin x − cos x yields
u00 + u = cos 2x, u0 (0) = u(π) = 0.
Note that since g(x) satisfies the homogeneous equation, the inhomogeneous term in the equation for u is the
same as that in the equation for y.

Example 23.5.3 Consider

2 4
y 00 + y = cos 2x, y(0) = , y(π) = − .
3 3
1 1
g(x) = cos x − 3
satisfies the boundary conditions. Substituting y = u + cos x − 3
yields
1
u00 + u = cos 2x + , u(0) = u(π) = 0.
3

Result 23.5.1 The nth order differential equation with boundary conditions

L[y] = f (x), Bj [y] = bj , for j = 1, . . . , n

has the solution y = u + g where u satisfies

L[u] = f (x) − L[g], Bj [u] = 0, for j = 1, . . . , n

and g is any n-times continuously differentiable function that satisfies the inhomogeneous
boundary conditions.

932
23.5.2 Separating Inhomogeneous Equations and Inhomogeneous Boundary Con-
ditions
Now consider a problem with inhomogeneous boundary conditions

L[y] = f (x), B1 [y] = γ1 , B2 [y] = γ2 .

In order to solve this problem, we solve the two problems

L[u] = f (x), B1 [u] = B2 [u] = 0, and

L[v] = 0, B1 [v] = γ1 , B2 [v] = γ2 .

The solution for the problem with an inhomogeneous equation and inhomogeneous boundary conditions will be
the sum of u and v. To verify this,

L[u + v] = L[u] + L[v] = f (x) + 0 = f (x),

Bi [u + v] = Bi [u] + Bi [v] = 0 + γi = γi .

This will be a useful technique when we develop Green functions.

Result 23.5.2 The solution to

L[y] = f (x), B1 [y] = γ1 , B2 [y] = γ2 ,

is y = u + v where

L[u] = f (x), B1 [u] = 0, B2 [u] = 0, and

L[v] = 0, B1 [v] = γ1 , B2 [v] = γ2 .

933
23.5.3 Existence of Solutions of Problems with Inhomogeneous Boundary Con-
ditions
Consider the nth order homogeneous differential equation

L[y] = y (n) + pn−1 y (n−1) + · · · + p1 y 0 + p0 y = f (x), for a < x < b,

subject to the n inhomogeneous boundary conditions

Bj [y] = γj , for j = 1, . . . , n

where each boundary condition is of the form

B[y] ≡ α0 y(a) + α1 y 0 (a) + · · · + αn−1 y (n−1) (a) + β0 y(b) + β1 y 0 (b) + · · · + βn−1 y (n−1) (b).

We assume that the coefficients in the differential equation are continuous on [a, b]. Since the Wronskian of the
solutions of the differential equation,
 Z 
W (x) = exp − pn−1 (x) dx ,

is non-vanishing on [a, b], there are n linearly independent solution on that range. Let {y1 , . . . , yn } be a set
of linearly independent solutions of the homogeneous equation. From Result 23.3.2 we know that a particular
solution yp exists. The general solution of the differential equation is

y = yp + c1 y1 + c2 y2 + · · · + cn yn .

The n boundary conditions impose the matrix equation,

     
B1 [y1 ] B1 [y2 ] · · · B1 [yn ] c1 γ1 − B1 [yp ]
 B2 [y1 ] B2 [y2 ] · · · B2 [yn ]   c2   γ2 − B2 [yp ] 
 ..  = 
     
 .. .. ... ..  .. 
 . . .  .  . 
Bn [y1 ] Bn [y2 ] · · · Bn [yn ] cn γn − Bn [yp ]

934
This equation has a unique solution if and only if the equation

     
B1 [y1 ] B1 [y2 ] ··· B1 [yn ] c1 0
 B2 [y1 ] B2 [y2 ] ··· B2 [yn ]   c2  0
 ..  =  .. 
     
 .. .. ... .. 
 . . .   .  .
Bn [y1 ] Bn [y2 ] · · · Bn [yn ] cn 0

has only the trivial solution. (This is the case if and only if the determinant of the matrix is nonzero.) Thus the
problem

L[y] = y (n) + pn−1 y (n−1) + · · · + p1 y 0 + p0 y = f (x), for a < x < b,

subject to the n inhomogeneous boundary conditions

Bj [y] = γj , for j = 1, . . . , n,

has a unique solution if and only if the problem

L[y] = y (n) + pn−1 y (n−1) + · · · + p1 y 0 + p0 y = 0, for a < x < b,

subject to the n homogeneous boundary conditions

Bj [y] = 0, for j = 1, . . . , n,

has only the trivial solution.

935
 Result 23.5.3 The problem

L[y] = y (n) + pn−1 y (n−1) + · · · + p1 y 0 + p0 y = f (x), for a < x < b,

subject to the n inhomogeneous boundary conditions

Bj [y] = γj , for j = 1, . . . , n,

has a unique solution if and only if the problem

L[y] = y (n) + pn−1 y (n−1) + · · · + p1 y 0 + p0 y = 0, for a < x < b,

subject to

Bj [y] = 0, for j = 1, . . . , n,

has only the trivial solution.

23.6 Green Functions for First Order Equations

Consider the first order inhomogeneous equation

L[y] ≡ y 0 + p(x)y = f (x), for x > a, (23.2)

subject to a homogeneous initial condition, B[y] ≡ y(a) = 0.

The Green function G(x|ξ) is defined as the solution to

L[G(x|ξ)] = δ(x − ξ) subject to G(a|ξ) = 0.

936
We can represent the solution to the inhomogeneous problem in Equation 23.2 as an integral involving the Green
function. To show that
Z ∞
y(x) = G(x|ξ)f (ξ) dξ
a

is the solution, we apply the linear operator L to the integral. (Assume that the integral is uniformly convergent.)
Z ∞  Z ∞
L G(x|ξ)f (ξ) dξ = L[G(x|ξ)]f (ξ) dξ
a a
Z ∞
= δ(x − ξ)f (ξ) dξ
a
= f (x)
The integral also satisfies the initial condition.
Z ∞  Z ∞
B G(x|ξ)f (ξ) dξ = B[G(x|ξ)]f (ξ) dξ
a a
Z ∞
= (0)f (ξ) dξ
a
=0
Now we consider the qualitiative behavior of the Green function. For x 6= ξ, the Green function is simply a
homogeneous solution of the differential equation, however at x = ξ we expect some singular behavior. G0 (x|ξ)
will have a Dirac delta function type singularity. This means that G(x|ξ) will have a jump discontinuity at x = ξ.
We integrate the differential equation on the vanishing interval (ξ − . . . ξ + ) to determine this jump.
G0 + p(x)G = δ(x − ξ)
Z ξ+
+ −
G(ξ |ξ) − G(ξ |ξ) + p(x)G(x|ξ) dx = 1
ξ−
G(ξ + |ξ) − G(ξ − |ξ) = 1 (23.3)

937
 The homogeneous solution of the differential equation is
R
yh = e− p(x) dx

Since the Green function satisfies the homogeneous equation for x 6= ξ, it will be a constant times this homogeneous
solution for x < ξ and x > ξ.
( R
c1 e− p(x) dx a<x<ξ
G(x|ξ) = R
c2 e− p(x) dx ξ<x

In order to satisfy the homogeneous initial condition G(a|ξ) = 0, the Green function must vanish on the interval
(a . . . ξ).
(
0 a<x<ξ
G(x|ξ) = R
c e− p(x) dx ξ<x

The jump condition, (Equation 23.3), gives us the constraint G(ξ + |ξ) = 1. This determines the constant in the
homogeneous solution for x > ξ.
(
0 R a<x<ξ
G(x|ξ) = x
e− ξ p(t) dt ξ<x

We can use the Heaviside function to write the Green function without using a case statement.
Rx
G(x|ξ) = e− ξ p(t) dt
H(x − ξ)

Clearly the Green function is of little value in solving the inhomogeneous differential equation in Equation 23.2,
as we can solve that problem directly. However, we will encounter first order Green function problems in solving
some partial differential equations.

938
 Result 23.6.1 The first order inhomogeneous differential equation with homogeneous
initial condition

L[y] ≡ y 0 + p(x)y = f (x), for a < x, y(a) = 0,

has the solution

Z ∞
y= G(x|ξ)f (ξ) dξ,
a

where G(x|ξ) satisfies the equation

L[G(x|ξ)] = δ(x − ξ), for a < x, G(a|ξ) = 0.

The Green function is

Rx
− p(t) dt
G(x|ξ) = e ξ H(x − ξ)

23.7 Green Functions for Second Order Equations

Consider the second order inhomogeneous equation
L[y] = y 00 + p(x)y 0 + q(x)y = f (x), for a < x < b, (23.4)
subject to the homogeneous boundary conditions
B1 [y] = B2 [y] = 0.
The Green function G(x|ξ) is defined as the solution to
L[G(x|ξ)] = δ(x − ξ) subject to B1 [G] = B2 [G] = 0.

939
The Green function is useful because you can represent the solution to the inhomogeneous problem in Equation 23.4
as an integral involving the Green function. To show that
Z b
y(x) = G(x|ξ)f (ξ) dξ
a

is the solution, we apply the linear operator L to the integral. (Assume that the integral is uniformly convergent.)
Z b  Z b
L G(x|ξ)f (ξ) dξ = L[G(x|ξ)]f (ξ) dξ
a a
Z b
= δ(x − ξ)f (ξ) dξ
a
= f (x)

The integral also satisfies the boundary conditions.

Z b  Z b
Bi G(x|ξ)f (ξ) dξ = Bi [G(x|ξ)]f (ξ) dξ
a a
Z b
= [0]f (ξ) dξ
a
=0

One of the advantages of using Green functions is that once you find the Green function for a linear operator
and certain homogeneous boundary conditions,

L[G] = δ(x − ξ), B1 [G] = B2 [G] = 0,

you can write the solution for any inhomogeneity, f (x).

L[f ] = f (x), B1 [y] = B2 [y] = 0

940
You do not need to do any extra work to obtain the solution for a different inhomogeneous term.
Qualitatively, what kind of behavior will the Green function for a second order differential equation have?
Will it have a delta function singularity; will it be continuous? To answer these questions we will first look at the
behavior of integrals and derivatives of δ(x).
The integral of δ(x) is the Heaviside function, H(x).
Z x (
0 for x < 0
H(x) = δ(t) dt =
−∞ 1 for x > 0

The integral of the Heaviside function is the ramp function, r(x).

Z x (
0 for x < 0
r(x) = H(t) dt =
−∞ x for x > 0

The derivative of the delta function is zero for x 6= 0. At x = 0 it goes from 0 up to +∞, down to −∞ and then
back up to 0.
In Figure 23.2 we see conceptually the behavior of the ramp function, the Heaviside function, the delta function,
and the derivative of the delta function.

d
Figure 23.2: r(x), H(x), δ(x) and dx
δ(x)

941
 We write the differential equation for the Green function.

G00 (x|ξ) + p(x)G0 (x|ξ) + q(x)G(x|ξ) = δ(x − ξ)

we see that only the G00 (x|ξ) term can have a delta function type singularity. If one of the other terms had a delta
function type singularity then G00 (x|ξ) would be more singular than a delta function and there would be nothing
in the right hand side of the equation to match this kind of singularity. Analogous to the progression from a
delta function to a Heaviside function to a ramp function, we see that G0 (x|ξ) will have a jump discontinuity and
G(x|ξ) will be continuous.
Let y1 and y2 be two linearly independent solutions to the homogeneous equation, L[y] = 0. Since the Green
function satisfies the homogeneous equation for x 6= ξ, it will be a linear combination of the homogeneous solutions.
(
c1 y1 + c2 y2 for x < ξ
G(x|ξ) =
d1 y1 + d2 y2 for x > ξ

We require that G(x|ξ) be continuous.

 
G(x|ξ)x→ξ− = G(x|ξ)x→ξ+

We can write this in terms of the homogeneous solutions.

c1 y1 (ξ) + c2 y2 (ξ) = d1 y1 (ξ) + d2 y2 (ξ)

We integrate L[G(x|ξ)] = δ(x − ξ) from ξ − to ξ+.

Z ξ+ Z ξ+
00 0
[G (x|ξ) + p(x)G (x|ξ) + q(x)G(x|ξ)] dx = δ(x − ξ) dx.
ξ− ξ−

942
Since G(x|ξ) is continuous and G0 (x|ξ) has only a jump discontinuity two of the terms vanish.

Z ξ+ Z ξ+
0
p(x)G (x|ξ) dx = 0 and q(x)G(x|ξ) dx = 0
ξ− ξ−
Z ξ+ Z ξ+
00
G (x|ξ) dx = δ(x − ξ) dx
ξ− ξ−
ξ + ξ +
G0 (x|ξ)
 
ξ−
= H(x − ξ) ξ−
G0 (ξ + |ξ) − G0 (ξ − |ξ) = 1

We write this jump condition in terms of the homogeneous solutions.

d1 y10 (ξ) + d2 y20 (ξ) − c1 y10 (ξ) − c2 y20 (ξ) = 1

Combined with the two boundary conditions, this gives us a total of four equations to determine our four constants,
c1 , c2 , d1 , and d2 .

943
 Result 23.7.1 The second order inhomogeneous differential equation with homogeneous
boundary conditions

L[y] = y 00 + p(x)y 0 + q(x)y = f (x), for a < x < b, B1 [y] = B2 [y] = 0,

has the solution

Z b
y= G(x|ξ)f (ξ) dξ,
a

where G(x|ξ) satisfies the equation

L[G(x|ξ)] = δ(x − ξ), for a < x < b, B1 [G(x|ξ)] = B2 [G(x|ξ)] = 0.

G(x|ξ) is continuous and G0 (x|ξ) has a jump discontinuity of height 1 at x = ξ.

Example 23.7.1 Solve the boundary value problem

y 00 = f (x), y(0) = y(1) = 0,
using a Green function.
A pair of solutions to the homogeneous equation are y1 = 1 and y2 = x. First note that only the trivial
solution to the homogeneous equation satisfies the homogeneous boundary conditions. Thus there is a unique
solution to this problem.
The Green function satisfies
G00 (x|ξ) = δ(x − ξ), G(0|ξ) = G(1|ξ) = 0.
The Green function has the form
(
c1 + c2 x for x < ξ
G(x|ξ) =
d1 + d2 x for x > ξ.

944
Applying the two boundary conditions, we see that c1 = 0 and d1 = −d2 . The Green function now has the form
(
cx for x < ξ
G(x|ξ) =
d(x − 1) for x > ξ.

Since the Green function must be continuous,

ξ
cξ = d(ξ − 1) ⇒ d=c .
ξ−1
From the jump condition,
d ξ  d 
c (x − 1) − cx =1

dx ξ − 1 x=ξ dx x=ξ
ξ
c −c=1
ξ−1
c = ξ − 1.
Thus the Green function is
(
(ξ − 1)x for x < ξ
G(x|ξ) =
ξ(x − 1) for x > ξ.

The Green function is plotted in Figure 23.3 for various values of ξ. The solution to y 00 = f (x) is

Z 1
y(x) = G(x|ξ)f (ξ) dξ
0

Z x Z 1
y(x) = (x − 1) ξf (ξ) dξ + x (ξ − 1)f (ξ) dξ.
0 x

945
 0.1 0.1 0.1 0.1
0.5 1 0.5 1 0.5 1 0.5 1
-0.1 -0.1 -0.1 -0.1
-0.2 -0.2 -0.2 -0.2
-0.3 -0.3 -0.3 -0.3

Figure 23.3: Plot of G(x|0.05),G(x|0.25),G(x|0.5) and G(x|0.75).

Example 23.7.2 Solve the boundary value problem

y 00 = f (x), y(0) = 1, y(1) = 2.

In Example 23.7.1 we saw that the solution to

u00 = f (x), u(0) = u(1) = 0

is
Z x Z 1
u(x) = (x − 1) ξf (ξ) dξ + x (ξ − 1)f (ξ) dξ.
0 x

Now we have to find the solution to

v 00 = 0, v(0) = 1, u(1) = 2.

The general solution is

v = c1 + c2 x.

Applying the boundary conditions yields

v = 1 + x.

946
 Thus the solution for y is
Z x Z 1
y = 1 + x + (x − 1) ξf (ξ) dξ + x (ξ − 1)f ( xi) dξ.
0 x

Example 23.7.3 Consider

y 00 = x, y(0) = y(1) = 0.

Method 1. Integrating the differential equation twice yields

1
y = x3 + c 1 x + c 2 .
6
Applying the boundary conditions, we find that the solution is

1
y = (x3 − x).
6

Method 2. Using the Green function to find the solution,

Z x Z 1
2
y = (x − 1) ξ dξ + x (ξ − 1)ξ dξ
0 x
 
1 3 1 1 1 3 1 2
= (x − 1) x + x − − x + x
3 3 2 3 2

1
y = (x3 − x).
6

947
Example 23.7.4 Find the solution to the differential equation

y 00 − y = sin x,

that is bounded for all x.

The Green function for this problem satisfies

G00 (x|ξ) − G(x|ξ) = δ(x − ξ).

The homogeneous solutions are y1 = ex , and y2 = e−x . The Green function has the form
(
c1 ex + c2 e−x for x < ξ
G(x|ξ) = x −x
d1 e + d2 e for x > ξ.

Since the solution must be bounded for all x, the Green function must also be bounded. Thus c2 = d1 = 0. The
Green function now has the form
(
c ex for x < ξ
G(x|ξ) = −x
de for x > ξ.

Requiring that G(x|ξ) be continuous gives us the condition

c eξ = d e−ξ ⇒ d = c e2ξ .

G(x|ξ) has a jump discontinuity of height 1 at x = ξ.

 
d 2ξ −x  d x 
ce e  − ce  =1
dx x=ξ dx x=ξ

−c e2ξ e−ξ − c eξ = 1
1
c = − e−ξ
2

948
The Green function is then
(
− 12 ex−ξ for x < ξ
G(x|ξ) =
− 12 e−x+ξ for x > ξ

1
G(x|ξ) = − e−|x−ξ| .
2

A plot of G(x|0) is given in Figure 23.4. The solution to y 00 − y = sin x is

Z ∞
1
y(x) = − e−|x−ξ| sin ξ dξ
−∞ 2
Z x Z ∞ 
1 x−ξ −x+ξ
=− sin ξ e dξ + sin ξ e dξ
2 −∞ x
1 sin x + cos x − sin x + cos x
= − (− + )
2 2 2

1
y= sin x.
2

23.7.1 Green Functions for Sturm-Liouville Problems

Consider the problem
0
L[y] = (p(x)y 0 ) + q(x)y = f (x), subject to
B1 [y] = α1 y(a) + α2 y 0 (a) = 0, B2 [y] = β1 y(b) + β2 y 0 (b) = 0.

949
 0.6

0.4

0.2

-4 -2 2 4
-0.2

-0.4

-0.6

Figure 23.4: Plot of G(x|0).

This is known as a Sturm-Liouville problem. Equations of this type often occur when solving partial differential
equations. The Green function associated with this problem satisfies
L[G(x|ξ)] = δ(x − ξ), B1 [G(x|ξ)] = B2 [G(x|ξ)] = 0.
Let y1 and y2 be two non-trivial homogeneous solutions that satisfy the left and right boundary conditions,
respectively.
L[y1 ] = 0, B1 [y1 ] = 0, L[y2 ] = 0, B2 [y2 ] = 0.
The Green function satisfies the homogeneous equation for x 6= ξ and satisfies the homogeneous boundary condi-
tions. Thus it must have the following form.
(
c1 (ξ)y1 (x) for a ≤ x ≤ ξ,
G(x|ξ) =
c2 (ξ)y2 (x) for ξ ≤ x ≤ b,

950
Here c1 and c2 are unknown functions of ξ.
The first constraint on c1 and c2 comes from the continuity condition.

G(ξ − |ξ) = G(ξ + |ξ)

c1 (ξ)y1 (ξ) = c2 (ξ)y2 (ξ)

We write the inhomogeneous equation in the standard form.

00 p0 0 q δ(x − ξ)
G (x|ξ) + G (x|ξ) + G(x|ξ) =
p p p
The second constraint on c1 and c2 comes from the jump condition.
1
G0 (ξ + |ξ) − G0 (ξ − |ξ) =
p(ξ)
1
c2 (ξ)y20 (ξ) − c1 (ξ)y10 (ξ) =
p(ξ)
Now we have a system of equations to determine c1 and c2 .

c1 (ξ)y1 (ξ) − c2 (ξ)y2 (ξ) = 0

1
c1 (ξ)y10 (ξ) − c2 (ξ)y20 (ξ) = −
p(ξ)
We solve this system with Kramer’s rule.
y2 (ξ) y1 (ξ)
c1 (ξ) = − , c2 (ξ) = −
p(ξ)(−W (ξ)) p(ξ)(−W (ξ))
Here W (x) is the Wronskian of y1 (x) and y2 (x). The Green function is
( y (x)y (ξ)
1 2
p(ξ)W (ξ)
for a ≤ x ≤ ξ,
G(x|ξ) = y2 (x)y1 (ξ)
p(ξ)W (ξ)
for ξ ≤ x ≤ b.

951
The solution of the Sturm-Liouville problem is

Z b
y= G(x|ξ)f (ξ) dξ.
a

Result 23.7.2 The problem

0
L[y] = (p(x)y 0 ) + q(x)y = f (x), subject to
B1 [y] = α1 y(a) + α2 y 0 (a) = 0, B2 [y] = β1 y(b) + β2 y 0 (b) = 0.

has the Green function

(y
1 (x)y2 (ξ)
p(ξ)W (ξ) for a ≤ x ≤ ξ,
G(x|ξ) = y2 (x)y1 (ξ)
p(ξ)W (ξ) for ξ ≤ x ≤ b,

where y1 and y2 are non-trivial homogeneous solutions that satisfy B1 [y1 ] = B2 [y2 ] = 0,
and W (x) is the Wronskian of y1 and y2 .

Example 23.7.5 Consider the equation

y 00 − y = f (x), y(0) = y(1) = 0.

A set of solutions to the homogeneous equation is { ex , e−x }. Equivalently, one could use the set {cosh x, sinh x}.
Note that sinh x satisfies the left boundary condition and sinh(x − 1) satisfies the right boundary condition. The

952
Wronskian of these two homogeneous solutions is
 
 sinh x sinh(x − 1) 
W (x) =  
cosh x cosh(x − 1)
= sinh x cosh(x − 1) − cosh x sinh(x − 1)
1 1
= [sinh(2x − 1) + sinh(1)] − [sinh(2x − 1) − sinh(1)]
2 2
= sinh(1).
The Green function for the problem is then
( sinh x sinh(ξ−1)
sinh(1)
for 0 ≤ x ≤ ξ
G(x|ξ) = sinh(x−1) sinh ξ
sinh(1)
for ξ ≤ x ≤ 1.

The solution to the problem is

x 1
sinh(x − 1)
Z Z
sinh(x)
y= sinh(ξ)f (ξ) dξ + sinh(ξ − 1)f (ξ) dξ.
sinh(1) 0 sinh(1) x

23.7.2 Initial Value Problems

Consider
L[y] = y 00 + p(x)y 0 + q(x)y = f (x), for a < x < b,
subject the the initial conditions
y(a) = γ1 , y 0 (a) = γ2 .
The solution is y = u + v where
u00 + p(x)u0 + q(x)u = f (x), u(a) = 0, u0 (a) = 0,

953
and
v 00 + p(x)v 0 + q(x)v = 0, v(a) = γ1 , v 0 (a) = γ2 .
Since the Wronskian
 Z 
W (x) = c exp − p(x) dx

is non-vanishing, the solutions of the differential equation for v are linearly independent. Thus there is a unique
solution for v that satisfies the initial conditions.

The Green function for u satisfies

G00 (x|ξ) + p(x)G0 (x|ξ) + q(x)G(x|ξ) = δ(x − ξ), G(a|ξ) = 0, G0 (a|ξ) = 0.
The continuity and jump conditions are
G(ξ − |ξ) = G(ξ + |ξ), G0 (ξ − |ξ) + 1 = G0 (ξ + |ξ).
Let u1 and u2 be two linearly independent solutions of the differential equation. For x < ξ, G(x|ξ) is a linear
combination of these solutions. Since the Wronskian is non-vanishing, only the trivial solution satisfies the
homogeneous initial conditions. The Green function must be
(
0 for x < ξ
G(x|ξ) =
uξ (x) for x > ξ,

where uξ (x) is the linear combination of u1 and u2 that satisfies

uξ (ξ) = 0, u0ξ (ξ) = 1.
Note that the non-vanishing Wronskian ensures a unique solution for uξ . We can write the Green function in the
form
G(x|ξ) = H(x − ξ)uξ (x).

954
This is known as the causal solution. The solution for u is
Z b
u= G(x|ξ)f (ξ) dξ
a
Z b
= H(x − ξ)uξ (x)f (ξ) dξ
Za x
= uξ (x)f (ξ) dξ
a

Now we have the solution for y,

Z x
y=v+ uξ (x)f (ξ) dξ.
a

Result 23.7.3 The solution of the problem

y 00 + p(x)y 0 + q(x)y = f (x), y(a) = γ1 , y 0 (a) = γ2 ,

is
Z x
y = yh + yξ (x)f (ξ) dξ
a

where yh is the combination of the homogeneous solutions of the equation that satisfy
the initial conditions and yξ (x) is the linear combination of homogeneous solutions that
satisfy yξ (ξ) = 0, yξ0 (ξ) = 1.

955
23.7.3 Problems with Unmixed Boundary Conditions
Consider

L[y] = y 00 + p(x)y 0 + q(x)y = f (x), for a < x < b,

subject the the unmixed boundary conditions

α1 y(a) + α2 y 0 (a) = γ1 , β1 y(b) + β2 y 0 (b) = γ2 .

The solution is y = u + v where

u00 + p(x)u0 + q(x)u = f (x), α1 u(a) + α2 u0 (a) = 0, β1 u(b) + β2 u0 (b) = 0,

and

v 00 + p(x)v 0 + q(x)v = 0, α1 v(a) + α2 v 0 (a) = γ1 , β1 v(b) + β2 v 0 (b) = γ2 .

The problem for v may have no solution, a unique solution or an infinite number of solutions. We consider only
the case that there is a unique solution for v. In this case the homogeneous equation subject to homogeneous
boundary conditions has only the trivial solution.

The Green function for u satisfies

G00 (x|ξ) + p(x)G0 (x|ξ) + q(x)G(x|ξ) = δ(x − ξ),

α1 G(a|ξ) + α2 G0 (a|ξ) = 0, β1 G(b|ξ) + β2 G0 (b|ξ) = 0.

The continuity and jump conditions are

G(ξ − |ξ) = G(ξ + |ξ), G0 (ξ − |ξ) + 1 = G0 (ξ + |ξ).

956
Let u1 and u2 be two solutions of the homogeneous equation that satisfy the left and right boundary conditions,
respectively. The non-vanishing of the Wronskian ensures that these solutions exist. Let W (x) denote the
Wronskian of u1 and u2 . Since the homogeneous equation with homogeneous boundary conditions has only the
trivial solution, W (x) is nonzero on [a, b]. The Green function has the form
(
c1 u1 for x < ξ,
G(x|ξ) =
c2 u2 for x > ξ.

The continuity and jump conditions for Green function gives us the equations

c1 u1 (ξ) − c2 u2 (ξ) = 0
c1 u01 (ξ) − c2 u02 (ξ) = −1.

Using Kramer’s rule, the solution is

u2 (ξ) u1 (ξ)
c1 = , c2 = .
W (ξ) W (ξ)
Thus the Green function is
(u
1 (x)u2 (ξ)
W (ξ)
for x < ξ,
G(x|ξ) = u1 (ξ)u2 (x)
W (ξ)
for x > ξ.

The solution for u is

Z b
u= G(x|ξ)f (ξ) dξ.
a

Thus if there is a unique solution for v, the solution for y is

Z b
y=v+ G(x|ξ)f (ξ) dξ.
a

957
 Result 23.7.4 Consider the problem

y 00 + p(x)y 0 + q(x)y = f (x),

α1 y(a) + α2 y 0 (a) = γ1 , β1 y(b) + β2 y 0 (b) = γ2 .

If the homogeneous differential equation subject to the inhomogeneous boundary condi-

tions has the unique solution yh , then the problem has the unique solution
Z b
y = yh + G(x|ξ)f (ξ) dξ
a

where
(u
1 (x)u2 (ξ)
W (ξ) for x < ξ,
G(x|ξ) = u1 (ξ)u2 (x)
W (ξ) for x > ξ,

u1 and u2 are solutions of the homogeneous differential equation that satisfy the left and
right boundary conditions, respectively, and W (x) is the Wronskian of u1 and u2 .

23.7.4 Problems with Mixed Boundary Conditions

Consider
L[y] = y 00 + p(x)y 0 + q(x)y = f (x), for a < x < b,
subject the the mixed boundary conditions
B1 [y] = α11 y(a) + α12 y 0 (a) + β11 y(b) + β12 y 0 (b) = γ1 ,

958
 B2 [y] = α21 y(a) + α22 y 0 (a) + β21 y(b) + β22 y 0 (b) = γ2 .

The solution is y = u + v where

u00 + p(x)u0 + q(x)u = f (x), B1 [u] = 0, B2 [u] = 0,

and

v 00 + p(x)v 0 + q(x)v = 0, B1 [v] = γ1 , B2 [v] = γ2 .

The problem for v may have no solution, a unique solution or an infinite number of solutions. Again we consider
only the case that there is a unique solution for v. In this case the homogeneous equation subject to homogeneous
boundary conditions has only the trivial solution.
Let y1 and y2 be two solutions of the homogeneous equation that satisfy the boundary conditions B1 [y1 ] = 0
and B2 [y2 ] = 0. Since the completely homogeneous problem has no solutions, we know that B1 [y2 ] and B2 [y1 ] are
nonzero. The solution for v has the form

v = c1 y1 + c2 y2 .

Applying the two boundary conditions yields

γ2 γ1
v= y1 + y2 .
B2 [y1 ] B1 [y2 ]

The Green function for u satisfies

G00 (x|ξ) + p(x)G0 (x|ξ) + q(x)G(x|ξ) = δ(x − ξ), B1 [G] = 0, B2 [G] = 0.

The continuity and jump conditions are

G(ξ − |ξ) = G(ξ + |ξ), G0 (ξ − |ξ) + 1 = G0 (ξ + |ξ).

959
We write the Green function as the sum of the causal solution and the two homogeneous solutions

G(x|ξ) = H(x − ξ)yξ (x) + c1 y1 (x) + c2 y2 (x)

With this form, the continuity and jump conditions are automatically satisfied. Applying the boundary conditions
yields

B1 [G] = B1 [H(x − ξ)yξ ] + c2 B1 [y2 ] = 0,

B2 [G] = B2 [H(x − ξ)yξ ] + c1 B2 [y1 ] = 0,

B1 [G] = β11 yξ (b) + β12 yξ0 (b) + c2 B1 [y2 ] = 0,

B2 [G] = β21 yξ (b) + β22 yξ0 (b) + c1 B2 [y1 ] = 0,

β21 yξ (b) + β22 yξ0 (b) β11 yξ (b) + β12 yξ0 (b)
G(x|ξ) = H(x − ξ)yξ (x) − y1 (x) − y2 (x).
B2 [y1 ] B1 [y2 ]

Note that the Green function is well defined since B2 [y1 ] and B1 [y2 ] are nonzero. The solution for u is

Z b
u= G(x|ξ)f (ξ) dξ.
a

Thus if there is a unique solution for v, the solution for y is

Z b
γ2 γ1
y= G(x|ξ)f (ξ) dξ + y1 + y2 .
a B2 [y1 ] B1 [y2 ]

960
Result 23.7.5 Consider the problem

y 00 + p(x)y 0 + q(x)y = f (x),

B1 [y] = α11 y(a) + α12 y 0 (a) + β11 y(b) + β12 y 0 (b) = γ1 ,

B2 [y] = α21 y(a) + α22 y 0 (a) + β21 y(b) + β22 y 0 (b) = γ2 .

If the homogeneous differential equation subject to the homogeneous boundary conditions

has no solution, then the problem has the unique solution
Z b
γ2 γ1
y= G(x|ξ)f (ξ) dξ + y1 + y2 ,
a B2 [y1 ] B1 [y2 ]
where
β21 yξ (b) + β22 yξ0 (b)
G(x|ξ) = H(x − ξ)yξ (x) − y1 (x)
B2 [y1 ]
β11 yξ (b) + β12 yξ0 (b)
− y2 (x),
B1 [y2 ]
y1 and y2 are solutions of the homogeneous differential equation that satisfy the first and
second boundary conditions, respectively, and yξ (x) is the solution of the homogeneous
equation that satisfies yξ (ξ) = 0, yξ0 (ξ) = 1.

961
23.8 Green Functions for Higher Order Problems
Consider the nth order differential equation

L[y] = y (n) + pn−1 (x)y (n−1) + · · · + p1 (x)y 0 + p0 y = f (x) on a < x < b,

subject to the n independent boundary conditions

Bj [y] = γj

where the boundary conditions are of the form

n−1
X n−1
X
B[y] ≡ αk y (k) (a) + βk y (k) (b).
k=0 k=0

We assume that the coefficient functions in the differential equation are continuous on [a, b]. The solution is
y = u + v where u and v satisfy

L[u] = f (x), with Bj [u] = 0,

and

L[v] = 0, with Bj [v] = γj

From Result 23.5.3, we know that if the completely homogeneous problem

L[w] = 0, with Bj [w] = 0,

has only the trivial solution, then the solution for y exists and is unique. We will construct this solution using
Green functions.

962
 First we consider the problem for v. Let {y1 , . . . , yn } be a set of linearly independent solutions. The solution
for v has the form

v = c1 y1 + · · · + cn yn

where the constants are determined by the matrix equation

     
B1 [y1 ] B1 [y2 ] · · · B1 [yn ] c1 γ1
 B2 [y1 ] B2 [y2 ] · · · B2 [yn ]   c2   γ2 
 ..  =  ..  .
     
 .. .. . . .. 
 . . . .  . .
Bn [y1 ] Bn [y2 ] · · · Bn [yn ] cn γn

To solve the problem for u we consider the Green function satisfying

L[G(x|ξ)] = δ(x − ξ), with Bj [G] = 0.

Let yξ (x) be the linear combination of the homogeneous solutions that satisfy the conditions

yξ (ξ) = 0
yξ0 (ξ) = 0
.. .
. = ..
(n−2)
yξ (ξ) = 0
(n−1)
yξ (ξ) = 1.

The causal solution is then

yc (x) = H(x − ξ)yξ (x).

The Green function has the form

G(x|ξ) = H(x − ξ)yξ (x) + d1 y1 (x) + · · · + dn yn (x)

963
The constants are determined by the matrix equation

     
B1 [y1 ] B1 [y2 ] ··· B1 [yn ] d1 −B1 [H(x − ξ)yξ (x)]
 B2 [y1 ] B2 [y2 ] ··· B2 [yn ]   d2   −B2 [H(x − ξ)yξ (x)] 
 ..  =  .
     
 .. .. .. ..  ..
 . . . .  .  . 
Bn [y1 ] Bn [y2 ] · · · Bn [yn ] dn −Bn [H(x − ξ)yξ (x)]

The solution for u then is

Z b
u= G(x|ξ)f (ξ) dξ.
a

964
 Result 23.8.1 Consider the nth order differential equation

L[y] = y (n) + pn−1 (x)y (n−1) + · · · + p1 (x)y 0 + p0 y = f (x) on a < x < b,

subject to the n independent boundary conditions

Bj [y] = γj

If the homogeneous differential equation subject to the homogeneous boundary conditions

has only the trivial solution, then the problem has the unique solution
Z b
y= G(x|ξ)f (ξ) dξ + c1 y1 + · · · cn yn
a

where

G(x|ξ) = H(x − ξ)yξ (x) + d1 y1 (x) + · · · + dn yn (x),

{y1 , . . . , yn } is a set of solutions of the homogeneous differential equation, and the con-
stants cj and dj can be determined by solving sets of linear equations.

Example 23.8.1 Consider the problem

y 000 − y 00 + y 0 − y = f (x),

y(0) = 1, y 0 (0) = 2, y(1) = 3.

The completely homogeneous associated problem is
w000 − w00 + w0 − w = 0, w(0) = w0 (0) = w(1) = 0.

965
The solution of the differential equation is

w = c1 cos x + c2 sin x + c2 ex .

The boundary conditions give us the equation

    
1 0 1 c1 0
 0 1 1 c2  = 0 .
cos 1 sin 1 e c3 0

The determinant of the matrix is e−cos 1−sin 1 6= 0. Thus the homogeneous problem has only the trivial solution
and the inhomogeneous problem has a unique solution.
We separate the inhomogeneous problem into the two problems

u000 − u00 + u0 − u = f (x), u(0) = u0 (0) = u(1) = 0,

v 000 − v 00 + v 0 − v = 0, v(0) = 1, v 0 (0) = 2, v(1) = 3,

First we solve the problem for v. The solution of the differential equation is

v = c1 cos x + c2 sin x + c2 ex .

The boundary conditions yields the equation

    
1 0 1 c1 1
 0 1 1   c2 = 2 .
 
cos 1 sin 1 e c3 3

The solution for v is

1
(e + sin 1 − 3) cos x + (2e − cos 1 − 3) sin x + (3 − cos 1 − 2 sin 1) ex .
 
v=
e − cos 1 − sin 1

966
 Now we find the Green function for the problem in u. The causal solution is
1
H(x − ξ)uξ (x) = H(x − ξ) (sin ξ − cos ξ) cos x − (sin ξ + cos ξ) sin ξ + e−ξ ex ,

2
1
H(x − ξ)uξ (x) = H(x − ξ) ex−ξ − cos(x − ξ) − sin(x − ξ) .
 
2
The Green function has the form
G(x|ξ) = H(x − ξ)uξ (x) + c1 cos x + c2 sin x + c3 ex .
The constants are determined by the three conditions
c1 cos x + c2 sin x + c3 ex x=0 = 0,
 
 
∂ x
(c1 cos x + c2 sin x + c3 e ) = 0,
∂x x=0
uξ (x) + c1 cos x + c2 sin x + c3 ex x=1 = 0.
 

The Green function is

1  cos(1 − ξ) + sin(1 − ξ) − e1−ξ 
G(x|ξ) = H(x − ξ) ex−ξ − cos(x − ξ) − sin(x − ξ) + cos x + sin x − ex
 
2 2(cos 1 + sin 1 − e)
The solution for v is
Z 1
v= G(x|ξ)f (ξ) dξ.
0
Thus the solution for y is

Z 1
1 
y= G(x|ξ)f (ξ) dξ + (e + sin 1 − 3) cos x
0 e − cos 1 − sin 1
+ (2e − cos 1 − 3) sin x + (3 − cos 1 − 2 sin 1) ex .


967
23.9 Fredholm Alternative Theorem
Orthogonality. Two real vectors, u and v are orthogonal if u · v = 0. Consider two functions, u(x) and v(x),
defined in [a, b]. The dot product in vector space is analogous to the integral
Z b
u(x)v(x) dx
a

in function space. Thus two real functions are orthogonal if

Z b
u(x)v(x) dx = 0.
a

Consider the nth order linear inhomogeneous differential equation

L[y] = f (x) on [a, b],
subject to the linear inhomogeneous boundary conditions
Bj [y] = 0, for j = 1, 2, . . . n.
The Fredholm alternative theorem tells us if the problem has a unique solution, an infinite number of solutions,
or no solution. Before presenting the theorem, we will consider a few motivating examples.

No Nontrivial Homogeneous Solutions. In the section on Green functions we showed that if the completely
homogeneous problem has only the trivial solution then the inhomogeneous problem has a unique solution.

Nontrivial Homogeneous Solutions Exist. If there are nonzero solutions to the homogeneous problem
L[y] = 0 that satisfy the homogeneous boundary conditions Bj [y] = 0 then the inhomogeneous problem L[y] =
f (x) subject to the same boundary conditions either has no solution or an infinite number of solutions.
Suppose there is a particular solution yp that satisfies the boundary conditions. If there is a solution yh to the
homogeneous equation that satisfies the boundary conditions then there will be an infinite number of solutions
since yp + cyh is also a particular solution.

968
 The question now remains: Given that there are homogeneous solutions that satisfy the boundary conditions,
how do we know if a particular solution that satisfies the boundary conditions exists? Before we address this
question we will consider a few examples.

Example 23.9.1 Consider the problem

y 00 + y = cos x, y(0) = y(π) = 0.
The two homogeneous solutions of the differential equation are
y1 = cos x, and y2 = sin x.
y2 = sin x satisfies the boundary conditions. Thus we know that there are either no solutions or an infinite number
of solutions. A particular solution is
cos2 x
Z Z
cos x sin x
yp = − cos x dx + sin x dx
1 1
Z Z  
1 1 1
= − cos x sin(2x) dx + sin x + cos(2x) dx
2 2 2
 
1 1 1
= cos x cos(2x) + sin x x + sin(2x)
4 2 4
1 1 
= x sin x + cos x cos(2x) + sin x sin(2x)
2 4
1 1
= x sin x + cos x
2 4
The general solution is
1
y = x sin x + c1 cos x + c2 sin x.
2
Applying the two boundary conditions yields
1
y = x sin x + c sin x.
2

969
Thus there are an infinite number of solutions.

Example 23.9.2 Consider the differential equation

y 00 + y = sin x, y(0) = y(π) = 0.

The general solution is

1
y = − x cos x + c1 cos x + c2 sin x.
2

Applying the boundary conditions,

y(0) = 0 ⇒ c1 = 0
1
y(π) = 0 ⇒ − π cos(π) + c2 sin(π) = 0
2
π
⇒ = 0.
2

Since this equation has no solution, there are no solutions to the inhomogeneous problem.

In both of the above examples there is a homogeneous solution y = sin x that satisfies the boundary con-
ditions. In Example 23.9.1, the inhomogeneous term is cos x and there are an infinite number of solutions. In
Example 23.9.2, the inhomogeneity is sin x and there are no solutions. In general, if the inhomogeneous term
is orthogonal to all the homogeneous solutions that satisfy the boundary conditions then there are an infinite
number of solutions. If not, there are no inhomogeneous solutions.

970
 Result 23.9.1 Fredholm Alternative Theorem. Consider the nth order inhomoge-
neous problem

L[y] = f (x) on [a, b] subject to Bj [y] = 0 for j = 1, 2, . . . , n,

and the associated homogeneous problem

L[y] = 0 on [a, b] subject to Bj [y] = 0 for j = 1, 2, . . . , n.

If the homogeneous problem has only the trivial solution then the inhomogeneous prob-
lem has a unique solution. If the homogeneous problem has m independent solutions,
{y1 , y2 , . . . , ym }, then there are two possibilities:
• If f (x) is orthogonal to each of the homogeneous solutions then there are an infinite
number of solutions of the form
m
X
y = yp + cj yj .
j=1

• If f (x) is not orthogonal to each of the homogeneous solutions then there are no
inhomogeneous solutions.

Example 23.9.3 Consider the problem

y 00 + y = cos 2x, y(0) = 1, y(π) = 2.
cos x and sin x are two linearly independent solutions to the homogeneous equation. sin x satisfies the homogeneous
boundary conditions. Thus there are either an infinite number of solutions, or no solution.

971
 x
To transform this problem to one with homogeneous boundary conditions, we note that g(x) = π
+ 1 and
make the change of variables y = u + g to obtain
x
u00 + u = cos 2x − − 1, y(0) = 0, y(π) = 0.
π
Since cos 2x − πx − 1 is not orthogonal to sin x, there is no solution to the inhomogeneous problem.
To check this, the general solution is
1
y = − cos 2x + c1 cos x + c2 sin x.
3
Applying the boundary conditions,
4
y(0) = 1 ⇒ c1 =
3
1 4
y(π) = 2 ⇒ − − = 2.
3 3
Thus we see that the right boundary condition cannot be satisfied.

Example 23.9.4 Consider

y 00 + y = cos 2x, y 0 (0) = y(π) = 1.

There are no solutions to the homogeneous equation that satisfy the homogeneous boundary conditions. To check
this, note that all solutions of the homogeneous equation have the form uh = c1 cos x + c2 sin x.

u0h (0) = 0 ⇒ c2 = 0
uh (π) = 0 ⇒ c1 = 0.

From the Fredholm Alternative Theorem we see that the inhomogeneous problem has a unique solution.

972
 To find the solution, start with
1
y = − cos 2x + c1 cos x + c2 sin x.
3

y 0 (0) = 1 ⇒ c2 = 1
1
y(π) = 1 ⇒ − − c1 = 1
3
Thus the solution is
1 4
y = − cos 2x − cos x + sin x.
3 3

Example 23.9.5 Consider

2 4
y 00 + y = cos 2x, y(0) = , y(π) = − .
3 3
cos x and sin x satisfy the homogeneous differential equation. sin x satisfies the homogeneous boundary conditions.
Since g(x) = cos x − 1/3 satisfies the boundary conditions, the substitution y = u + g yields
1
u00 + u = cos 2x + , y(0) = 0, y(π) = 0.
3
Now we check if sin x is orthogonal to cos 2x + 13 .
Z π   Z π
1 1 1 1
sin x cos 2x + dx = sin 3x − sin x + sin x dx
0 3 2 2 3
0 π
1 1
= − cos 3x + cos x
6 6 0
=0

973
Since sin x is orthogonal to the inhomogeneity, there are an infinite number of solutions to the problem for u,
(and hence the problem for y).
As a check, then general solution for y is
1
y = − cos 2x + c1 cos x + c2 sin x.
3
Applying the boundary conditions,
2
y(0) = ⇒ c1 = 1
3
4 4 4
y(π) = − ⇒ − =− .
3 3 3
Thus we see that c2 is arbitrary. There are an infinite number of solutions of the form

1
y = − cos 2x + cos x + c sin x.
3

974
23.10 Exercises
Undetermined Coefficients
Exercise 23.1 (mathematica/ode/inhomogeneous/undetermined.nb)
Find the general solution of the following equations.

1. y 00 + 2y 0 + 5y = 3 sin(2t)

2. 2y 00 + 3y 0 + y = t2 + 3 sin(t)

Hint, Solution

Exercise 23.2 (mathematica/ode/inhomogeneous/undetermined.nb)

Find the solution of each one of the following initial value problems.

1. y 00 − 2y 0 + y = t et + 4, y(0) = 1, y 0 (0) = 1

2. y 00 + 2y 0 + 5y = 4 e−t cos(2t), y(0) = 1, y 0 (0) = 0

Hint, Solution

Variation of Parameters
Exercise 23.3 (mathematica/ode/inhomogeneous/variation.nb)
Use the method of variation of parameters to find a particular solution of the given differential equation.

1. y 00 − 5y 0 + 6y = 2 et

2. y 00 + y = tan(t), 0 < t < π/2

3. y 00 − 5y 0 + 6y = g(t), for a given function g.

Hint, Solution

975
Exercise 23.4 (mathematica/ode/inhomogeneous/variation.nb)
Solve
y 00 (x) + y(x) = x, y(0) = 1, y 0 (0) = 0.
Hint, Solution

Exercise 23.5 (mathematica/ode/inhomogeneous/variation.nb)

Solve
x2 y 00 (x) − xy 0 (x) + y(x) = x.
Hint, Solution

Exercise 23.6 (mathematica/ode/inhomogeneous/variation.nb)

1. Find the general solution of y 00 + y = ex .
2. Solve y 00 + λ2 y = sin x, y(0) = y 0 (0) = 0. λ is an arbitrary real constant. Is there anything special about
λ = 1?
Hint, Solution

Exercise 23.7 (mathematica/ode/inhomogeneous/variation.nb)

Consider the problem of solving the initial value problem
y 00 + y = g(t), y(0) = 0, y 0 (0) = 0.
1. Show that the general solution of y 00 + y = g(t) is
 Z t   Z t 
y(t) = c1 − g(τ ) sin τ dτ cos t + c2 + g(τ ) cos τ dτ sin t,
a b

where c1 and c2 are arbitrary constants and a and b are any conveniently chosen points.

976
 2. Using the result of part (a) show that the solution satisfying the initial conditions y(0) = 0 and y 0 (0) = 0 is
given by
Z t
y(t) = g(τ ) sin(t − τ ) dτ.
0

Notice that this equation gives a formula for computing the solution of the original initial value problem for
any given inhomogeneous term g(t). The integral is referred to as the convolution of g(t) with sin t.
3. Use the result of part (b) to solve the initial value problem,

y 00 + y = sin(λt), y(0) = 0, y 0 (0) = 0,

where λ is a real constant. How does the solution for λ = 1 differ from that for λ 6= 1? The λ = 1 case
provides an example of resonant forcing. Plot the solution for resonant and non-resonant forcing.
Hint, Solution

Exercise 23.8
Find the variation of parameters solution for the third order differential equation

y 000 + p2 (x)y 00 + p1 (x)y 0 + p0 (x)y = f (x).

Hint, Solution

Green Functions
Exercise 23.9
Use a Green function to solve

y 00 = f (x), y(−∞) = y 0 (−∞) = 0.

Verify the the solution satisfies the differential equation.

Hint, Solution

977
Exercise 23.10
Solve the initial value problem

1 1
y 00 + y 0 − 2 y = x2 , y(0) = 0, y 0 (0) = 1.
x x
First use variation of parameters, and then solve the problem with a Green function.
Hint, Solution

Exercise 23.11
What are the continuity conditions at x = ξ for the Green function for the problem

y 000 + p2 (x)y 00 + p1 (x)y 0 + p0 (x)y = f (x).

Hint, Solution

Exercise 23.12
Use variation of parameters and Green functions to solve

x2 y 00 − 2xy 0 + 2y = e−x , y(1) = 0, y 0 (1) = 1.

Hint, Solution

Exercise 23.13
Find the Green function for

y 00 − y = f (x), y 0 (0) = y(1) = 0.

Hint, Solution

978
Exercise 23.14
Find the Green function for

y 00 − y = f (x), y(0) = y(∞) = 0.

Hint, Solution

Exercise 23.15
Find the Green function for each of the following:

a) xu00 + u0 = f (x), u(0+ ) bounded, u(1) = 0.

b) u00 − u = f (x), u(−a) = u(a) = 0.

c) u00 − u = f (x), u(x) bounded as |x| → ∞.

d) Show that the Green function for (b) approaches that for (c) as a → ∞.
Hint, Solution

Exercise 23.16
1. For what values of λ does the problem

y 00 + λy = f (x), y(0) = y(π) = 0, (23.5)

have a unique solution? Find the Green functions for these cases.

2. For what values of α does the problem

y 00 + 9y = 1 + αx, y(0) = y(π) = 0,

have a solution? Find the solution.

979
 3. For λ = n2 , n ∈ Z+ state in general the conditions on f in Equation 23.5 so that a solution will exist. What
is the appropriate modified Green function (in terms of eigenfunctions)?
Hint, Solution

Exercise 23.17
Show that the inhomogeneous boundary value problem:
Lu ≡ (pu0 )0 + qu = f (x), a < x < b, u(a) = α, u(b) = β
has the solution:
Z b
u(x) = g(x; ξ)f (ξ) dξ − αp(a)gξ (x; a) + βp(b)gξ (x; b).
a
Hint, Solution

Exercise 23.18
The Green function for
u00 − k 2 u = f (x), −∞ < x < ∞
subject to |u(±∞)| < ∞ is
1 −k|x−ξ|
G(x; ξ) = −
e .
2k
(We assume that k > 0.) Use the image method to find the Green function for the same equation on the
semi-infinite interval 0 < x < ∞ satisfying the boundary conditions,
i) u(0) = 0 |u(∞)| < ∞,
ii) u0 (0) = 0 |u(∞)| < ∞.

Express these results in simplified forms without absolute values.

Hint, Solution

980
Exercise 23.19
1. Determine the Green function for solving:

y 00 − a2 y = f (x), y(0) = y 0 (L) = 0.

2. Take the limit as L → ∞ to find the Green function on (0, ∞) for the boundary conditions: y(0) = 0,
y 0 (∞) = 0. We assume here that a > 0. Use the limiting Green function to solve:

y 00 − a2 y = e−x , y(0) = 0, y 0 (∞) = 0.

Check that your solution satisfies all the conditions of the problem.
Hint, Solution

981
23.11 Hints
Undetermined Coefficients
Hint 23.1

Hint 23.2

Variation of Parameters
Hint 23.3

Hint 23.4

Hint 23.5

Hint 23.6

Hint 23.7

982
Hint 23.8
Look for a particular solution of the form

yp = u1 y1 + u2 y2 + u3 y3 ,

where the yj ’s are homogeneous solutions. Impose the constraints

u01 y1 + u02 y2 + u03 y3 = 0

u01 y10 + u02 y20 + u03 y30 = 0.

To avoid some messy algebra when solving for u0j , use Kramer’s rule.

Green Functions
Hint 23.9

Hint 23.10

Hint 23.11

Hint 23.12

Hint 23.13
cosh(x) and sinh(x−1) are homogeneous solutions that satisfy the left and right boundary conditions, respectively.

983
Hint 23.14
sinh(x) and e−x are homogeneous solutions that satisfy the left and right boundary conditions, respectively.

Hint 23.15
The Green function for the differential equation

d
L[y] ≡ (p(x)y 0 ) + q(x)y = f (x),
dx
subject to unmixed, homogeneous boundary conditions is

y1 (x< )y2 (x> )

G(x|ξ) = ,
p(ξ)W (ξ)

(y
1 (x)y2 (ξ)
p(ξ)W (ξ)
for a ≤ x ≤ ξ,
G(x|ξ) = y1 (ξ)y2 (x)
p(ξ)W (ξ)
for ξ ≤ x ≤ b,

where y1 and y2 are homogeneous solutions that satisfy the left and right boundary conditions, respectively.
Recall that if y(x) is a solution of a homogeneous, constant coefficient differential equation then y(x + c) is
also a solution.

Hint 23.16
The problem has a Green function if and only if the inhomogeneous problem has a unique solution. The inhomo-
geneous problem has a unique solution if and only if the homogeneous problem has only the trivial solution.

Hint 23.17
Show that gξ (x; a) and gξ (x; b) are solutions of the homogeneous differential equation. Determine the value of
these solutions at the boundary.

984
Hint 23.18

Hint 23.19

985
23.12 Solutions
Undetermined Coefficients
Solution 23.1
1. We consider
y 00 + 2y 0 + 5y = 3 sin(2t).
We first find the homogeneous solution with the substitition y = eλt .
λ2 + 2λ + 5 = 0
λ = −1 ± 2i
The homogeneous solution is
yh = c1 e−t cos(2t) + c2 e−t sin(2t).
We guess a particular solution of the form
yp = a cos(2t) + b sin(2t).
We substitute this into the differential equation to determine the coefficients.
yp00 + 2yp0 + 5yp = 3 sin(2t)
−4a cos(2t) − 4b sin(2t) − 4a sin(2t) + 4b sin(2t) + 5a cos(2t) + 5b sin(2t) = −3 sin(2t)
(a + 4b) cos(2t) + (−3 − 4a + b) sin(2t) = 0
a + 4b = 0, −4a + b = 3
12 3
a=− , b=
17 17
A particular solution is
3
yp = (sin(2t) − 4 cos(2t)).
17

986
 The general solution of the differential equation is
3
y = c1 e−t cos(2t) + c2 e−t sin(2t) + (sin(2t) − 4 cos(2t)).
17

2. We consider
2y 00 + 3y 0 + y = t2 + 3 sin(t)
We first find the homogeneous solution with the substitition y = eλt .
2λ2 + 3λ + 1 = 0
λ = {−1, −1/2}
The homogeneous solution is
yh = c1 e−t + c2 e−t/2 .
We guess a particular solution of the form
yp = at2 + bt + c + d cos(t) + e sin(t).
We substitute this into the differential equation to determine the coefficients.
2yp00 + 3yp0 + yp = t2 + 3 sin(t)

2(2a − d cos(t) − e sin(t)) + 3(2at + b − d sin(t) + e cos(t))

+ at2 + bt + c + d cos(t) + e sin(t) = t2 + 3 sin(t)

(a − 1)t2 + (6a + b)t + (4a + 3b + c) + (−d + 3e) cos(t) − (3 + 3d + e) sin(t) = 0

a − 1 = 0, 6a + b = 0, 4a + 3b + c = 0, −d + 3e = 0, 3 + 3d + e = 0
9 3
a = 1, b = −6, c = 14, d = − , e = −
10 10

987
 A particular solution is

3
yp = t2 − 6t + 14 − (3 cos(t) + sin(t)).
10

The general solution of the differential equation is

3
y = c1 e−t + c2 e−t/2 + t2 − 6t + 14 − (3 cos(t) + sin(t)).
10

Solution 23.2
1. We consider the problem

y 00 − 2y 0 + y = t et + 4, y(0) = 1, y 0 (0) = 1.

First we solve the homogeneous equation with the substitution y = eλt .

λ2 − 2λ + 1 = 0
(λ − 1)2 = 0
λ=1

The homogeneous solution is

yh = c1 et + c2 t et .

We guess a particular solution of the form

yp = at3 et + bt2 et + 4.

988
 We substitute this into the inhomogeneous differential equation to determine the coefficients.
yp00 − 2yp0 + yp = t et + 4
(a(t3 + 6t2 + 6t) + b(t2 + 4t + 2)) et − 2(a(t2 + 3t) + b(t + 2)) et at3 et + bt2 et + 4 = t et + 4
(6a − 1)t + 2b = 0
6a − 1 = 0, 2b = 0
1
a= , b=0
6
A particular solution is
t3 t
yp = e + 4.
6
The general solution of the differential equation is
t3 t
y = c1 et + c2 t et + e + 4.
6
We use the initial conditions to determine the constants of integration.
y(0) = 1, y 0 (0) = 1
c1 + 4 = 1, c1 + c2 = 1
c1 = −3, c2 = 4
The solution of the initial value problem is

t3
 
y= + 4t − 3 et + 4.
6

2. We consider the problem

y 00 + 2y 0 + 5y = 4 e−t cos(2t), y(0) = 1, y 0 (0) = 0.

989
First we solve the homogeneous equation with the substitution y = eλt .

λ2 + 2λ + 5 = 0
√
λ = −1 ± 1 − 5
λ = −1 ± i2

The homogeneous solution is

yh = c1 e−t cos(2t) + c2 e−t sin(2t).

We guess a particular solution of the form

yp = t e−t (a cos(2t) + b sin(2t))

We substitute this into the inhomogeneous differential equation to determine the coefficients.

yp00 + 2yp0 + 5yp = 4 e−t cos(2t)

e−t ((−(2 + 3t)a + 4(1 − t)b) cos(2t) + (4(t − 1)a − (2 + 3t)b) sin(2t))
+ 2 e−t (((1 − t)a + 2tb) cos(2t) + (−2ta + (1 − t)b) sin(2t))
+ 5( e−t (ta cos(2t) + tb sin(2t))) = 4 e−t cos(2t)

4(b − 1) cos(2t) − 4a sin(2t) = 0

a = 0, b = 1

A particular solution is

yp = t e−t sin(2t).

990
 The general solution of the differential equation is

y = c1 e−t cos(2t) + c2 e−t sin(2t) + t e−t sin(2t).

We use the initial conditions to determine the constants of integration.

y(0) = 1, y 0 (0) = 0
c1 = 1, −c1 + 2c2 = 0
1
c1 = 1, c2 =
2
The solution of the initial value problem is

1 −t
y= e (2 cos(2t) + (2t + 1) sin(2t)) .
2

Variation of Parameters
Solution 23.3
1. We consider the equation

y 00 − 5y 0 + 6y = 2 et .

We find homogeneous solutions with the substitution y = eλt .

λ2 − 5λ + 6 = 0
λ = {2, 3}

The homogeneous solutions are

y1 = e2t , y2 = e3t .

991
 We compute the Wronskian of these solutions.
 2t 
3t 
 e e
W (t) =  2t  = e5t
2 e 3 e3t 
We find a particular solution with variation of parameters.
2 et e3t 2 et e2t
Z Z
2t 3t
yp = − e dt + e dt
e5t e5t
Z Z
= −2 e2t e−t dt + 2 e3t e−2t dt

= 2 et − et

yp = et

2. We consider the equation

π
y 00 + y = tan(t), 0<t< .
2
We find homogeneous solutions with the substitution y = eλt .

λ2 + 1 = 0
λ = ±i

The homogeneous solutions are

y1 = cos(t), y2 = sin(t).

We compute the Wronskian of these solutions.

 
 cos(t) sin(t) 
W (t) =   = cos2 (t) + sin2 (t) = 1
− sin(t) cos(t)

992
 We find a particular solution with variation of parameters.
Z Z
yp = − cos(t) tan(t) sin(t) dt + sin(t) tan(t) cos(t) dt

sin2 (t)
Z Z
= − cos(t) dt + sin(t) sin(t) dt
cos(t)
  
cos(t/2) − sin(t/2)
= cos(t) ln + sin(t) − sin(t) cos(t)
cos(t/2) + sin(t/2)

 
cos(t/2) − sin(t/2)
yp = cos(t) ln
cos(t/2) + sin(t/2)

3. We consider the equation

y 00 − 5y 0 + 6y = g(t).

The homogeneous solutions are

y1 = e2t , y2 = e3t .

The Wronskian of these solutions is W (t) = e5t . We find a particular solution with variation of parameters.

g(t) e3t g(t) e2t

Z Z
2t 3t
yp = − e dt + e dt
e5t e5t

Z Z
−2t
yp = − e 2t
g(t) e dt + e3t
g(t) e−3t dt

993
Solution 23.4
Solve

y 00 (x) + y(x) = x, y(0) = 1, y 0 (0) = 0.

The solutions of the homogeneous equation are

y1 (x) = cos x, y2 (x) = sin x.

The Wronskian of these solutions is

 
 cos x sin x 
W [cos x, sin x] =  
− sin x cos x
= cos2 x + sin2 x
= 1.

The variation of parameters solution for the particular solution is

Z Z
yp = − cos x x sin x dx + sin x x cos x dx
 Z   Z 
= − cos x −x cos x + cos x dx + sin x x sin x − sin x dx

= − cos x (−x cos x + sin x) + sin x (x sin x + cos x)

= x cos2 x − cos x sin x + x sin2 x + cos x sin x
=x

The general solution of the differential equation is thus

y = c1 cos x + c2 sin x + x.

Applying the two initial conditions gives us the equations

c1 = 1, c2 + 1 = 0.

994
The solution subject to the initial conditions is
y = cos x − sin x + x.

Solution 23.5
Solve
x2 y 00 (x) − xy 0 (x) + y(x) = x.
The homogeneous equation is
x2 y 00 (x) − xy 0 (x) + y(x) = 0.
Substituting y = xλ into the homogeneous differential equation yields
x2 λ(λ − 1)xλ−2 − xλxλ + xλ = 0

λ2 − 2λ + 1 = 0

(λ − 1)2 = 0

λ = 1.
The homogeneous solutions are
y1 = x, y2 = x log x.
The Wronskian of the homogeneous solutions is
 
x x log x 
W [x, x log x] = 
 
1 1 + log x
= x + x log x − x log x
= x.

995
Writing the inhomogeneous equation in the standard form:
1 1 1
y 00 (x) − y 0 (x) + 2 y(x) = .
x x x
Using variation of parameters to find the particular solution,
Z Z
log x 1
yp = −x dx + x log x dx
x x
1
= −x log2 x + x log x log x
2
1
= x log2 x.
2
Thus the general solution of the inhomogeneous differential equation is

1
y = c1 x + c2 x log x + x log2 x.
2

Solution 23.6
1. First we find the homogeneous solutions. We substitute y = eλx into the homogeneous differential equation.

y 00 + y = 0
λ2 + 1 = 0
λ = ±i
 ix −ix
y= e ,e

We can also write the solutions in terms of real-valued functions.

y = {cos x, sin x}

996
 The Wronskian of the homogeneous solutions is
 
 cos x sin x 
W [cos x, sin x] = 
  = cos2 x + sin2 x = 1.
− sin x cos x

We obtain a particular solution with the variation of parameters formula.

Z Z
yp = − cos x e sin x dx + sin x ex cos x dx
x

1 1
yp = − cos x ex (sin x − cos x) + sin x ex (sin x + cos x)
2 2
1 x
yp = e
2
The general solution is the particular solution plus a linear combination of the homogeneous solutions.

1 x
y= e + cos x + sin x
2

2.

y 00 + λ2 y = sin x, y(0) = y 0 (0) = 0

Assume that λ is positive. First we find the homogeneous solutions by substituting y = eαx into the
homogeneous differential equation.

y 00 + λ2 y = 0
α 2 + λ2 = 0
α = ±iλ
 iλx −iλx
y= e ,e
y = {cos(λx), sin(λx)}

997
The Wronskian of these homogeneous solution is
 
 cos(λx) sin(λx) 
W [cos(λx), sin(λx)] = 
 = λ cos2 (λx) + λ sin2 (λx) = λ.
−λ sin(λx) λ cos(λx)

We obtain a particular solution with the variation of parameters formula.

Z Z
sin(λx) sin x cos(λx) sin x
yp = − cos(λx) dx + sin(λx) dx
λ λ

We evaluate the integrals for λ 6= 1.

cos(x) sin(λx) − λ sin x cos(λx) cos(x) cos(λx) + λ sin x sin(λx)

yp = − cos(λx) 2
+ sin(λx)
λ(λ − 1) λ(λ2 − 1)
sin x
yp = 2
λ −1
The general solution for λ 6= 1 is
sin x
y= + c1 cos(λx) + c2 sin(λx).
λ2 − 1
The initial conditions give us the constraints:

c1 = 0,
1
+ λc2 = 0,
λ2 −1
For λ 6= 1, (non-resonant forcing), the solution subject to the initial conditions is

λ sin(x) − sin(λx)
y= .
λ(λ2 − 1)

998
 Now consider the case λ = 1. We obtain a particular solution with the variation of parameters formula.
Z Z
2
yp = − cos(x) sin (x) dx + sin(x) cos(x) sin x dx
 
1 1 2
yp = − cos(x) (x − cos(x) sin(x)) + sin(x) − cos (x)
2 2
1
yp = − x cos(x)
2
The general solution for λ = 1 is
1
y = − x cos(x) + c1 cos(x) + c2 sin(x).
2
The initial conditions give us the constraints:
c1 = 0
1
− + c2 = 0
2
For λ = 1, (resonant forcing), the solution subject to the initial conditions is
1
y = (sin(x) − x cos x).
2

Solution 23.7
1. A set of linearly independent, homogeneous solutions is {cos t, sin t}. The Wronskian of these solutions is
 
 cos t sin t 
W (t) =   = cos2 t + sin2 t = 1.
− sin t cos t
We use variation of parameters to find a particular solution.
Z Z
yp = − cos t g(t) sin t dt + sin t g(t) cos t dt

999
 The general solution can be written in the form,
 Z t   Z t 
y(t) = c1 − g(τ ) sin τ dτ cos t + c2 + g(τ ) cos τ dτ sin t.
a b

2. Since the initial conditions are given at t = 0 we choose the lower bounds of integration in the general
solution to be that point.
 Z t   Z t 
y = c1 − g(τ ) sin τ dτ cos t + c2 + g(τ ) cos τ dτ sin t
0 0

The initial condition y(0) = 0 gives the constraint, c1 = 0. The derivative of y(t) is then,
Z t  Z t 
0
y (t) = −g(t) sin t cos t + g(τ ) sin τ dτ sin t + g(t) cos t sin t + c2 + g(τ ) cos τ dτ cos t,
0 0
Z t  Z t 
0
y (t) = g(τ ) sin τ dτ sin t + c2 + g(τ ) cos τ dτ cos t.
0 0

The initial condition y 0 (0) = 0 gives the constraint c2 = 0. The solution subject to the initial conditions is
Z t
y= g(τ )(sin t cos τ − cos t sin τ ) dτ
0
Z t
y= g(τ ) sin(t − τ ) dτ
0

3. The solution of the initial value problem

y 00 + y = sin(λt), y(0) = 0, y 0 (0) = 0,
is
Z t
y= sin(λτ ) sin(t − τ ) dτ.
0

1000
For λ 6= 1, this is

1 t
Z


y= cos(t − τ − λτ ) − cos(t − τ + λτ ) dτ
2 0
 t
1 sin(t − τ − λτ ) sin(t − τ + λτ )
= − +
2 1+λ 1−λ 0
 
1 sin(t) − sin(−λt) − sin(t) + sin(λt)
= +
2 1+λ 1−λ

λ sin t sin(λt)
y=− 2
+ . (23.6)
1−λ 1 − λ2

The solution is the sum of two periodic functions of period 2π and 2π/λ. This solution is plotted in
Figure 23.5 on the interval t ∈ [0, 16π] for the values λ = 1/4, 7/8, 5/2.

Figure 23.5: Non-resonant Forcing

1001
For λ = 1, we have
1 t
Z


y= cos(t − 2τ ) − cos(tau) dτ
2 0
 t
1 1
= − sin(t − 2τ ) − τ cos t
2 2 0

1
y= (sin t − t cos t) . (23.7)
2
The solution has both a periodic and a transient term. This solution is plotted in Figure 23.5 on the interval
t ∈ [0, 16π].

Figure 23.6: Resonant Forcing

Note that we can derive (23.7) from (23.6) by taking the limit as λ → 0.
sin(λt) − λ sin t t cos(λt) − sin t
lim 2
= lim
λ→1 1−λ λ→1 −2λ
1
= (sin t − t cos t)
2

1002
Solution 23.8
Let y1 , y2 and y3 be linearly independent homogeneous solutions to the differential equation

L[y] = y 000 + p2 y 00 + p1 y 0 + p0 y = f (x).

We will look for a particular solution of the form

yp = u1 y1 + u2 y2 + u3 y3 .

Since the uj ’s are undetermined functions, we are free to impose two constraints. We choose the constraints to
simplify the algebra.

u01 y1 + u02 y2 + u03 y3 = 0

u01 y10 + u02 y20 + u03 y30 = 0

Differentiating the expression for yp ,

yp0 = u01 y1 + u1 y10 + u02 y2 + u2 y20 + u03 y3 + u3 y30

= u1 y10 + u2 y20 + u3 y30
yp00 = u01 y10 + u1 y100 + u02 y20 + u2 y200 + u03 y30 + u3 y300
= u1 y100 + u2 y200 + u3 y300
yp000 = u01 y100 + u1 y1000 + u02 y200 + u2 y2000 + u03 y300 + u3 y3000

Substituting the expressions for yp and its derivatives into the differential equation,

u01 y100 + u1 y1000 + u02 y200 + u2 y2000 + u03 y300 + u3 y3000 + p2 (u1 y100 + u2 y200 + u3 y300 ) + p1 (u1 y10 + u2 y20 + u3 y30 )
+ p0 (u1 y1 + u2 y2 + u3 y3 ) = f (x)

u01 y100 + u02 y200 + u03 y300 + u1 L[y1 ] + u2 L[y2 ] + u3 L[y3 ] = f (x)
u01 y100 + u02 y200 + u03 y300 = f (x).

1003
With the two constraints, we have the system of equations,

u01 y1 + u02 y2 + u03 y3 = 0

u01 y10 + u02 y20 + u03 y30 = 0
u01 y100 + u02 y200 + u03 y300 = f (x)

We solve for the u0j using Kramer’s rule.

(y2 y30 − y20 y3 )f (x) (y1 y30 − y10 y3 )f (x) (y1 y20 − y10 y2 )f (x)
u01 = , u02 = − , u03 =
W (x) W (x) W (x)
Here W (x) is the Wronskian of {y1 , y2 , y3 }. Integrating the expressions for u0j , the particular solution is

(y2 y30 − y20 y3 )f (x) (y3 y10 − y30 y1 )f (x) (y1 y20 − y10 y2 )f (x)
Z Z Z
yp = y1 dx + y2 dx + y3 dx.
W (x) W (x) W (x)

Green Functions
Solution 23.9
We consider the Green function problem

G00 = f (x), G(−∞|ξ) = G0 (−∞|ξ) = 0.

The homogeneous solution is y = c1 + c2 x. The homogeneous solution that satisfies the boundary conditions is
y = 0. Thus the Green function has the form
(
0 x < ξ,
G(x|ξ) =
c1 + c2 x x > ξ.

The continuity and jump conditions are then

G(ξ + |ξ) = 0, G0 (ξ + |ξ) = 1.

1004
Thus the Green function is
(
0 x < ξ,
G(x|ξ) = = (x − ξ)H(x − ξ).
x−ξ x>ξ

The solution of the problem

y 00 = f (x), y(−∞) = y 0 (−∞) = 0.
is
Z ∞
y= f (ξ)G(x|ξ) dξ
−∞
Z ∞
y= f (ξ)(x − ξ)H(x − ξ) dξ
−∞
Z x
y= f (ξ)(x − ξ) dξ
−∞

We differentiate this solution to verify that it satisfies the differential equation.

Z x Z x
0 ∂
y = [f (ξ)(x − ξ)]ξ=x + (f (ξ)(x − ξ)) dξ = f (ξ) dξ
−∞ ∂x −∞
y 00 = [f (ξ)]ξ=x = f (x)

Solution 23.10
Since we are dealing with an Euler equation, we substitute y = xλ to find the homogeneous solutions.
λ(λ − 1) + λ − 1 = 0
(λ − 1)(λ + 1) = 0
1
y1 = x, y2 =
x

1005
Variation of Parameters. The Wronskian of the homogeneous solutions is
 
x 1/x  1 1 2
W (x) =  2 = − − =− .

1 −1/x x x x

A particular solution is

x2 (1/x) x2 x
Z Z
1
yp = −x dx + dx
−2/x x −2/x
x2 x4
Z Z
1
= −x − dx + − dx
2 x 2
4 4
x x
= −
6 10
4
x
= .
15
The general solution is

x4 1
y= + c1 x + c2 .
15 x
Applying the initial conditions,

y(0) = 0 ⇒ c2 = 0
y 0 (0) = 0 ⇒ c1 = 1.

Thus we have the solution

x4
y= + x.
15

1006
Green Function. Since this problem has both an inhomogeneous term in the differential equation and inho-
mogeneous boundary conditions, we separate it into the two problems

1 1
u00 + u0 − 2 u = x2 , u(0) = u0 (0) = 0,
x x
1 1
v 00 + v 0 − 2 v = 0, v(0) = 0, v 0 (0) = 1.
x x

First we solve the inhomogeneous differential equation with the homogeneous boundary conditions. The Green
function for this problem satisfies

L[G(x|ξ)] = δ(x − ξ), G(0|ξ) = G0 (0|ξ) = 0.

Since the Green function must satisfy the homogeneous boundary conditions, it has the form
(
0 for x < ξ
G(x|ξ) =
cx + d/x for x > ξ.

From the continuity condition,

0 = cξ + d/ξ.

The jump condition yields

c − d/ξ 2 = 1.

Solving these two equations, we obtain

(
0 for x < ξ
G(x|ξ) = 1 ξ2
2
x − 2x
for x > ξ

1007
Thus the solution is
Z ∞
u(x) = G(x|ξ)ξ 2 dξ
Z0 x 
ξ2

1
= x− ξ 2 dξ
0 2 2x
1 4 1
= x − x4
6 10
x4
= .
15

Now to solve the homogeneous differential equation with inhomogeneous boundary conditions. The general
solution for v is

v = cx + d/x.

Applying the two boundary conditions gives

v = x.

Thus the solution for y is

x4
y =x+ .
15

Solution 23.11
The Green function satisfies

G000 (x|ξ) + p2 (x)G00 (x|ξ) + p1 (x)G0 (x|ξ) + p0 (x)G(x|ξ) = δ(x − ξ).

First note that only the G000 (x|ξ) term can have a delta function singularity. If a lower derivative had a delta
function type singularity, then G000 (x|ξ) would be more singular than a delta function and there would be no other

1008
term in the equation to balance that behavior. Thus we see that G000 (x|ξ) will have a delta function singularity;
G00 (x|ξ) will have a jump discontinuity; G0 (x|ξ) will be continuous at x = ξ. Integrating the differential equation
from ξ − to ξ + yields
Z ξ+ Z ξ+
000
G (x|ξ) dx = δ(x − ξ) dx
ξ− ξ−

G00 (ξ + |ξ) − G00 (ξ − |ξ) = 1.

Thus we have the three continuity conditions:

G00 (ξ + |ξ) = G00 (ξ − |ξ) + 1

G0 (ξ + |ξ) = G0 (ξ − |ξ)
G(ξ + |ξ) = G(ξ − |ξ)

Solution 23.12
Variation of Parameters. Consider the problem

x2 y 00 − 2xy 0 + 2y = e−x , y(1) = 0, y 0 (1) = 1.

Previously we showed that two homogeneous solutions are

y1 = x, y2 = x2 .

The Wronskian of these solutions is

 
x x2 
W (x) =   = 2x2 − x2 = x2 .
1 2x

1009
In the variation of parameters formula, we will choose 1 as the lower bound of integration. (This will simplify the
algebra in applying the initial conditions.)
Z x −ξ 2 Z x −ξ
e ξ 2 e ξ
yp = −x 4
dξ + x dξ
1 ξ 1 ξ4
Z x −ξ Z x −ξ
e 2 e
= −x dξ + x dξ
1 ξ2 1 ξ3
Z x −ξ 
e−x e−x 1 x e−ξ
  −x Z 
−1 e 2 e
= −x e − − dξ + x − 2 + dξ
x 1 ξ 2x 2x 2 1 ξ
 Z x −ξ
x + x2

−1 1 −x e
= −x e + (1 + x) e + dξ
2 2 1 ξ
If you wanted to, you could write the last integral in terms of exponential integral functions.
The general solution is
 Z x −ξ
x2

2 −1 1 −x e
y = c1 x + c2 x − x e + (1 + x) e + x + dξ
2 2 1 ξ
Applying the boundary conditions,
y(1) = 0 ⇒ c1 + c2 = 0
y 0 (1) = 1 ⇒ c1 + 2c2 = 1,
we find that c1 = −1, c2 = 1.
Thus the solution subject to the initial conditions is
 Z x −ξ
x2

−1 1 2 −x e
y = −(1 + e )x + x + (1 + x) e + x + dξ
2 2 1 ξ

Green Functions. The solution to the problem is y = u + v where

2 2 e−x
u00 − u0 + 2 u = 2 , u(1) = 0, u0 (1) = 0,
x x x

1010
and
2 2
v 00 − v 0 + 2 v = 0, v(1) = 0, v 0 (1) = 1.
x x
The problem for v has the solution
v = −x + x2 .
The Green function for u is
G(x|ξ) = H(x − ξ)uξ (x)
where
uξ (ξ) = 0, and u0ξ (ξ) = 1.
Thus the Green function is
x2
 
G(x|ξ) = H(x − ξ) −x + .
ξ
The solution for u is then
∞
e−ξ
Z
u= G(x|ξ) 2 dξ
ξ
Z1 x 
x2 e−ξ

= −x + dξ
1 ξ ξ2
 Z x −ξ
x2

−1 1 −x e
= −x e + (1 + x) e + x + dξ.
2 2 1 ξ
Thus we find the solution for y is
 Z x −ξ
x2

−1 1 2 −x e
y = −(1 + e )x + x + (1 + x) e + x + dξ
2 2 1 ξ

1011
Solution 23.13
The differential equation for the Green function is

G00 − G = δ(x − ξ), Gx (0|ξ) = G(1|ξ) = 0.

Note that cosh(x) and sinh(x − 1) are homogeneous solutions that satisfy the left and right boundary conditions,
respectively. The Wronskian of these two solutions is
 
cosh(x) sinh(x − 1) 
W (x) =  
sinh(x) cosh(x − 1)
= cosh(x) cosh(x − 1) − sinh(x) sinh(x − 1)
1
ex + e−x ex−1 + e−x+1 − ex − e−x ex−1 − e−x+1






=
4
1 1
e + e−1


=
2
= cosh(1).

The Green function for the problem is then

cosh(x< ) sinh(x> − 1)
G(x|ξ) = ,
cosh(1)

( cosh(x) sinh(ξ−1)
cosh(1)
for 0 ≤ x ≤ ξ,
G(x|ξ) = cosh(ξ) sinh(x−1)
cosh(1)
for ξ ≤ x ≤ 1.

Solution 23.14
The differential equation for the Green function is

G00 − G = δ(x − ξ), G(0|ξ) = G(∞|ξ) = 0.

1012
Note that sinh(x) and e−x are homogeneous solutions that satisfy the left and right boundary conditions, respec-
tively. The Wronskian of these two solutions is
 
 sinh(x) e−x 
W (x) = 
 
cosh(x) − e−x 
= − sinh(x) e−x − cosh(x) e−x
1 1 x
= − ex − e−x e−x − e + e−x e−x



2 2
= −1
The Green function for the problem is then
G(x|ξ) = − sinh(x< ) e−x>

(
− sinh(x) e−ξ for 0 ≤ x ≤ ξ,
G(x|ξ) =
− sinh(ξ) e−x for ξ ≤ x ≤ ∞.

Solution 23.15

a) The Green function problem is

xG00 (x|ξ) + G0 (x|ξ) = δ(x − ξ), G(0|ξ) bounded, G(1|ξ) = 0.
First we find the homogeneous solutions of the differential equation.
xy 00 + y 0 = 0
This is an exact equation.
d
[xy 0 ] = 0
dx

1013
 c1
y0 =
x
y = c1 log x + c2
The homogeneous solutions y1 = 1 and y2 = log x satisfy the left and right boundary conditions, respectively.
The Wronskian of these solutions is
 
1 log x 1
W (x) =  = .
0 1/x  x
The Green function is
1 · log x>
G(x|ξ) = ,
ξ(1/ξ)

G(x|ξ) = log x> .

b) The Green function problem is

G00 (x|ξ) − G(x|ξ) = δ(x − ξ), G(−a|ξ) = G(a|ξ) = 0.
{ ex , e−x } and {cosh x, sinh x} are both linearly independent sets of homogeneous solutions. sinh(x + a) and
sinh(x − a) are homogeneous solutions that satisfy the left and right boundary conditions, respectively. The
Wronskian of these two solutions is,
 
 sinh(x + a) sinh(x − a) 
W (x) =  
cosh(x + a) cosh(x − a)
= sinh(x + a) cosh(x − a) − sinh(x − a) cosh(x + a)
= sinh(2a)
The Green function is
sinh(x< + a) sinh(x> − a)
G(x|ξ) = .
sinh(2a)

1014
c) The Green function problem is

G00 (x|ξ) − G(x|ξ) = δ(x − ξ), G(x|ξ) bounded as |x| → ∞.

ex and e−x are homogeneous solutions that satisfy the left and right boundary conditions, respectively. The
Wronskian of these solutions is
 x 
e e−x 
W (x) =  x
 = −2.
e − e−x 
The Green function is
ex< e−x>
G(x|ξ) = ,
−2

1
G(x|ξ) = − ex< −x> .
2

d) The Green function from part (b) is,

sinh(x< + a) sinh(x> − a)
G(x|ξ) = .
sinh(2a)
We take the limit as a → ∞.
sinh(x< + a) sinh(x> − a) ( ex< +a − e−x< −a ) ( ex> −a − e−x> +a )
lim = lim
a→∞ sinh(2a) a→∞ 2 ( e2a − e−2a )
− ex< −x> + ex< +x> −2a + e−x< −x> −2a − e−x< +x> −4a
= lim
a→∞ 2 − 2 e−4a
x< −x>
e
=−
2
Thus we see that the solution from part (b) approaches the solution from part (c) as a → ∞.

1015
Solution 23.16
1. The problem,

y 00 + λy = f (x), y(0) = y(π) = 0,

has a Green function if and only if it has a unique solution. This inhomogeneous problem has a unique
solution if and only if the homogeneous problem has only the trivial solution.
First consider the case λ = 0. We find the general solution of the homogeneous differential equation.

y = c1 + c2 x

Only the trivial solution satisfies the boundary conditions. The problem has a unique solution for λ = 0.
Now consider non-zero λ. We find the general solution of the homogeneous differential equation.
√  √ 
y = c1 cos λx + c2 sin λx .

The solution that satisfies the left boundary condition is

√ 
y = c sin λx .

We apply the right boundary condition and find nontrivial solutions.

√ 
sin λπ = 0
λ = n2 , n ∈ Z+

Thus the problem has a unique solution for all complex λ except λ = n2 , n ∈ Z+ .
Consider the case λ = 0. We find solutions of the homogeneous equation that satisfy the left and right
boundary conditions, respectively.

y1 = x, y2 = x − π.

1016
 We compute the Wronskian of these functions.
 
x x − π 
W (x) = 
  = π.
1 1 

The Green function for this case is

x< (x> − π)
G(x|ξ) = .
π
We consider the case λ =6 n2 , λ 6= 0. We find the solutions of the homogeneous equation that satisfy the
left and right boundary conditions, respectively.
√  √ 
y1 = sin λx , y2 = sin λ(x − π) .

We compute the Wronskian of these functions.

 √  √  
 sin λx sin λ(x − π)  √ √ 
 
W (x) = √
 √  √ √  = λ sin λπ
 λ cos λx λ cos λ(x − π) 


The Green function for this case is

√  √ 
sin λx< sin λ(x> − π)
G(x|ξ) = √ √  .
λ sin λπ

2. Now we consider the problem

y 00 + 9y = 1 + αx, y(0) = y(π) = 0.

The homogeneous solutions of the problem are constant multiples of sin(3x). Thus for each value of α,
the problem either has no solution or an infinite number of solutions. There will be an infinite number of

1017
 solutions if the inhomogeneity 1 + αx is orthogonal to the homogeneous solution sin(3x) and no solution
otherwise.
Z π
πα + 2
(1 + αx) sin(3x) dx =
0 3

The problem has a solution only for α = −2/π. For this case the general solution of the inhomogeneous
differential equation is
 
1 2x
y= 1− + c1 cos(3x) + c2 sin(3x).
9 π

The one-parameter family of solutions that satisfies the boundary conditions is

 
1 2x
y= 1− − cos(3x) + c sin(3x).
9 π

3. For λ = n2 , n ∈ Z+ , y = sin(nx) is a solution of the homogeneous equation that satisfies the boundary
conditions. Equation 23.5 has a (non-unique) solution only if f is orthogonal to sin(nx).
Z π
f (x) sin(nx) dx = 0
0

The modified Green function satisfies

sin(nx) sin(nξ)
G00 + n2 G = δ(x − ξ) − .
π/2

We expand G in a series of the eigenfunctions.

∞
X
G(x|ξ) = gk sin(kx)
k=1

1018
 We substitute the expansion into the differential equation to determine the coefficients. This will not
determine gn . We choose gn = 0, which is one of the choices that will make the modified Green function
symmetric in x and ξ.
∞ ∞
X
2 2

2X
gk n − k sin(kx) = sin(kx) sin(kξ)
k=1
π k=1
k6=n
∞
2 X sin(kx) sin(kξ)
G(x|ξ) =
π k=1 n2 − k 2
k6=n

The solution of the inhomogeneous problem is

Z π
y(x) = f (ξ)G(x|ξ) dξ.
0

Solution 23.17
We separate the problem for u into the two problems:
Lv ≡ (pv 0 )0 + qv = f (x), a < x < b, v(a) = 0, v(b) = 0
Lw ≡ (pw0 )0 + qw = 0, a < x < b, w(a) = α, w(b) = β
and note that the solution for u is u = v + w.
The problem for v has the solution,
Z b
v= g(x; ξ)f (ξ) dξ,
a

with the Green function,

(v
1 (x)v2 (ξ)
v1 (x< )v2 (x> ) p(ξ)W (ξ)
for a ≤ x ≤ ξ,
g(x; ξ) = ≡ v1 (ξ)v2 (x)
p(ξ)W (ξ) p(ξ)W (ξ)
for ξ ≤ x ≤ b.

1019
Here v1 and v2 are homogeneous solutions that respectively satisfy the left and right homogeneous boundary
conditions.
Since g(x; ξ) is a solution of the homogeneous equation for x 6= ξ, gξ (x; ξ) is a solution of the homogeneous
equation for x 6= ξ. This is because for x 6= ξ,
 
∂ ∂ ∂
L g = L[g] = δ(x − ξ) = 0.
∂ξ ∂ξ ∂ξ

If ξ is outside of the domain, (a, b), then g(x; ξ) and gξ (x; ξ) are homogeneous solutions on that domain. In
particular gξ (x; a) and gξ (x; b) are homogeneous solutions,

L [gξ (x; a)] = L [gξ (x; b)] = 0.

Now we use the definition of the Green function and v1 (a) = v2 (b) = 0 to determine simple expressions for these
homogeneous solutions.

v10 (a)v2 (x) (p0 (a)W (a) + p(a)W 0 (a))v1 (a)v2 (x)
gξ (x; a) = −
p(a)W (a) (p(a)W (a))2
0
v (a)v2 (x)
= 1
p(a)W (a)
v10 (a)v2 (x)
=
p(a)(v1 (a)v20 (a) − v10 (a)v2 (a))
v10 (a)v2 (x)
=−
p(a)v10 (a)v2 (a)
v2 (x)
=−
p(a)v2 (a)

We note that this solution has the boundary values,

v2 (a) 1 v2 (b)
gξ (a; a) = − =− , gξ (b; a) = − = 0.
p(a)v2 (a) p(a) p(a)v2 (a)

1020
We examine the second solution.
v1 (x)v20 (b) (p0 (b)W (b) + p(b)W 0 (b))v1 (x)v2 (b)
gξ (x; b) = −
p(b)W (b) (p(b)W (b))2
0
v1 (x)v2 (b)
=
p(b)W (b)
v1 (x)v20 (b)
=
p(b)(v1 (b)v20 (b) − v10 (b)v2 (b))
v1 (x)v20 (b)
=
p(b)v1 (b)v20 (b)
v1 (x)
=
p(b)v1 (b)
This solution has the boundary values,
v1 (a) v1 (b) 1
gξ (a; b) = = 0, gξ (b; b) = = .
p(b)v1 (b) p(b)v1 (b) p(b)
Thus we see that the solution of

Lw = (pw0 )0 + qw = 0, a < x < b, w(a) = α, w(b) = β,

is

w = −αp(a)gξ (x; a) + βp(b)gξ (x; b).

Therefore the solution of the problem for u is

Z b
u= g(x; ξ)f (ξ) dξ − αp(a)gξ (x; a) + βp(b)gξ (x; b).
a

1021
 -4 -2 2 4

-0.1

-0.2

-0.3

-0.4

-0.5

Figure 23.7: G(x; 1) and G(x; −1)

Solution 23.18
Figure 23.7 shows a plot of G(x; 1) and G(x; −1) for k = 1.
First we consider the boundary condition u(0) = 0. Note that the solution of
G00 − k 2 G = δ(x − ξ) − δ(x + ξ), |G(±∞; ξ)| < ∞,
satisfies the condition G(0; ξ) = 0. Thus the Green function which satisfies G(0; ξ) = 0 is
1 −k|x−ξ| 1 −k|x+ξ|
G(x; ξ) = − e + e .
2k 2k
Since x, ξ > 0 we can write this as
1 1 −k(x+ξ)
G(x; ξ) = − e−k|x−ξ| + e
( 2k 2k
1 −k(ξ−x) 1 −k(x+ξ)
− 2k e + 2k e , for x < ξ
= 1 −k(x−ξ) 1 −k(x+ξ)
− 2k e + 2k e , for ξ < x
(
− k1 e−kξ sinh(kx), for x < ξ
=
− k1 e−kx sinh(kξ), for ξ < x

1022
 1
G(x; ξ) = − e−kx> sinh(kx< )
k
Now consider the boundary condition u0 (0) = 0. Note that the solution of
G00 − k 2 G = δ(x − ξ) + δ(x + ξ), |G(±∞; ξ)| < ∞,
satisfies the boundary condition G0 (x; ξ) = 0. Thus the Green function is
1 −k|x−ξ| 1 −k|x+ξ|
G(x; ξ) = − e − e .
2k 2k
Since x, ξ > 0 we can write this as
1 1 −k(x+ξ)
G(x; ξ) = − e−k|x−ξ| − e
( 2k 2k
1 −k(ξ−x) 1 −k(x+ξ)
− 2k e − 2k e , for x < ξ
= 1 −k(x−ξ) 1 −k(x+ξ)
− 2k e − 2k e , for ξ < x
(
− k1 e−kξ cosh(kx), for x < ξ
=
− k1 e−kx cosh(kξ), for ξ < x

1
G(x; ξ) = − e−kx> cosh(kx< )
k
The Green functions which satisfies G(0; ξ) = 0 and G0 (0; ξ) = 0 are shown in Figure 23.8.

Solution 23.19
1. The Green function satisfies
g 00 − a2 g = δ(x − ξ), g(0; ξ) = g 0 (L; ξ) = 0.

1023
 1 2 3 4 5 1 2 3 4 5
-0.1 -0.1
-0.2
-0.2 -0.3
-0.3 -0.4
-0.4 -0.5

Figure 23.8: G(x; 1) and G(x; −1)

We can write the set of homogeneous solutions as

eax , e−ax

or {cosh(ax), sinh(ax)} .

The solutions that respectively satisfy the left and right boundary conditions are

u1 = sinh(ax), u2 = cosh(a(x − L)).

The Wronskian of these solutions is

 
sinh(ax) cosh(a(x − L))
W (x) = = −a cosh(aL).
a cosh(ax) a sinh(a(x − L))

Thus the Green function is

(
− sinh(ax) cosh(a(ξ−L))
a cosh(aL)
for x ≤ ξ, sinh(ax< ) cosh(a(x> − L))
g(x; ξ) = =− .
− sinh(aξ) cosh(a(x−L))
a cosh(aL)
for ξ ≤ x. a cosh(aL)

1024
2. We take the limit as L → ∞.
sinh(ax< ) cosh(a(x> − L))
g(x; ξ) = lim −
L→∞ a cosh(aL)
sinh(ax< ) cosh(ax> ) cosh(aL) − sinh(ax> ) sinh(aL)
= lim −
L→∞ a cosh(aL)
sinh(ax< )
=− (cosh(ax> ) − sinh(ax> ))
a

1
g(x; ξ) = − sinh(ax< ) e−ax>
a
The solution of

y 00 − a2 y = e−x , y(0) = y 0 (∞) = 0

is
Z ∞
y= g(x; ξ) e−ξ dξ
0
1 ∞
Z
=− sinh(ax< ) e−ax> e−ξ dξ
a 0
Z x Z ∞ 
1 −ax −ξ −aξ −ξ
=− sinh(aξ) e e dξ + sinh(ax) e e dξ
a 0 x

We first consider the case that a 6= 1.

 −ax 
1 e −x

1 −(a+1)x
=− −a + e (a cosh(ax) + sinh(ax)) + e sinh(ax)
a a2 − 1 a+1
e−ax − e−x
=
a2 − 1

1025
For a = 1, we have
 
1 −2x

1 −2x
y=− e−x −1 + 2x + e + e sinh(x)
4 2
1
= − x e−x .
2
Thus the solution of the problem is
(
e−ax − e−x
a2 −1
for a 6= 1,
y=
− 12 x e−x for a = 1.

We note that this solution satisfies the differential equation and the boundary conditions.

1026
Chapter 24

Difference Equations

Televisions should have a dial to turn up the intelligence. There is a brightness knob, but it doesn’t work.

-?

24.1 Introduction
Example 24.1.1 Gambler’s ruin problem. Consider a gambler that initially has n dollars. He plays a game
in which he has a probability p of winning a dollar and q of losing a dollar. (Note that p+q = 1.) The gambler has
decided that if he attains N dollars he will stop playing the game. In this case we will say that he has succeeded.
Of course if he runs out of money before that happens, we will say that he is ruined. What is the probability of
the gambler’s ruin? Let us denote this probability by an . We know that if he has no money left, then his ruin is
certain, so a0 = 1. If he reaches N dollars he will quit the game, so that aN = 0. If he is somewhere in between
ruin and success then the probability of his ruin is equal to p times the probability of his ruin if he had n + 1
dollars plus q times the probability of his ruin if he had n − 1 dollars. Writing this in an equation,

an = pan+1 + qan−1 subject to a0 = 1, aN = 0.

1027
This is an example of a difference equation. You will learn how to solve this particular problem in the section on
constant coefficient equations.

Consider the sequence a1 , a2 , a3 , . . . Analogous to a derivative of a continuous function, we can define a discrete
derivative on the sequence

Dan = an+1 − an .

The second discrete derivative is then defined as

D2 an = D[an+1 − an ] = an+2 − 2an+1 + an .

The discrete integral of an is

n
X
ai .
i=n0

Corresponding to
Z β
df
dx = f (β) − f (α),
α dx
in the discrete realm we have
β−1 β−1
X X
D[an ] = (an+1 − an ) = aβ − aα .
n=α n=α

Linear difference equations have the form

Dr an + pr−1 (n)Dr−1 an + · · · + p1 (n)Dan + p0 (n)an = f (n).

From the definition of the discrete derivative an equivalent form is

an+r + qr−1 (n)anr −1 + · · · + q1 (n)an+1 + q0 (n)an = f (n).

1028
 Besides being important in their own right, we will need to solve difference equations in order to develop series
solutions of differential equations. Also, some methods of solving differential equations numerically are based on
approximating them with difference equations.
There are many similarities between differential and difference equations. Like differential equations, an rth
order homogeneous difference equation has r linearly independent solutions. The general solution to the rth
order inhomogeneous equation is the sum of the particular solution and an arbitrary linear combination of the
homogeneous solutions.
For an rth order difference equation, the initial condition is given by specifying the values of the first r an ’s.

Example 24.1.2 Consider the difference equation an−2 −an−1 −an = 0 subject to the initial condition a1 = a2 = 1.
Note that although we may not know a closed-form formula for the an we can calculate the an in order by
substituting into the difference equation. The first few an are 1, 1, 2, 3, 5, 8, 13, 21, . . . We recognize this as the
Fibonacci sequence.

24.2 Exact Equations

Consider the sequence a1 , a2 , . . . . Exact difference equations on this sequence have the form

D[F (an , an+1 , . . . , n)] = g(n).

We can reduce the order of, (or solve for first order), this equation by summing from 1 to n − 1.
n−1
X n−1
X
D[F (aj , aj+1 , . . . , j)] = g(j)
j=1 j=1
n−1
X
F (an , an+1 , . . . , n) − F (a1 , a2 , . . . , 1) = g(j)
j=1
n−1
X
F (an , an+1 , . . . , n) = g(j) + F (a1 , a2 , . . . , 1)
j=1

1029
 Result 24.2.1 We can reduce the order of the exact difference equation

D[F (an , an+1 , . . . , n)] = g(n), for n ≥ 1

by summing both sides of the equation to obtain

n−1
X
F (an , an+1 , . . . , n) = g(j) + F (a1 , a2 , . . . , 1).
j=1

Example 24.2.1 Consider the difference equation, D[nan ] = 1. Summing both sides of this equation

n−1
X n−1
X
D[jaj ] = 1
j=1 j=1

nan − a1 = n − 1
n + a1 − 1
an = .
n

24.3 Homogeneous First Order

Consider the homogeneous first order difference equation

an+1 = p(n)an , for n ≥ 1.

1030
We can directly solve for an .
an−1 an−2 a1
an = an ···
an−1 an−2 a1
an an−1 a2
= a1 ···
an−1 an−2 a1
= a1 p(n − 1)p(n − 2) · · · p(1)
n−1
Y
= a1 p(j)
j=1

Alternatively, we could solve this equation by making it exact. Analogous to an integrating factor for differ-
ential equations, we multiply the equation by the summing factor
" n #−1
Y
S(n) = p(j) .
j=1

an+1 − p(n)an = 0
an+1 an
Qn − Qn−1 =0
j=1 p(j) j=1 p(j)
" #
an
D Qn−1 =0
j=1 p(j)

Now we sum from 1 to n − 1.

an
Qn−1 − a1 = 0
j=1 p(j)
n−1
Y
an = a1 p(j)
j=1

1031
 Result 24.3.1 The solution of the homogeneous first order difference equation

an+1 = p(n)an , for n ≥ 1,

is
n−1
Y
an = a1 p(j).
j=1

Example 24.3.1 Consider the equation an+1 = nan with the initial condition a1 = 1.
n−1
Y
an = a1 j = (1)(n − 1)! = Γ(n)
j=1

Recall that Γ(z) is the generalization of the factorial function. For positive integral values of the argument,
Γ(n) = (n − 1)!.

24.4 Inhomogeneous First Order

Consider the equation

an+1 = p(n)an + q(n) for n ≥ 1.

hQ i−1
n
Multiplying by S(n) = j=1 p(j) yields

a a q(n)
Qn n+1 − Qn−1n = Qn .
j=1 p(j) j=1 p(j) j=1 p(j)

1032
The left hand side is a discrete derivative.
" #
an q(n)
D Qn−1 = Qn
j=1 p(j) j=1 p(j)

Summing both sides from 1 to n − 1,

n−1
" #
an X q(k)
Qn−1 − a1 = Qk
j=1 p(j) k=1 j=1 p(j)

" n−1 # " n−1 " # #

Y X q(k)
an = p(m) Qk + a1 .
m=1 k=1 j=1 p(j)

Result 24.4.1 The solution of the inhomogeneous first order difference equation

an+1 = p(n)an + q(n) for n ≥ 1

is
" n−1 # " n−1 " # #
Y X q(k)
an = p(m) Qk + a1 .
m=1 k=1 j=1 p(j)

Example 24.4.1 Consider the equation an+1 = nan + 1 for n ≥ 1. The summing factor is
" n #−1
Y 1
S(n) = j = .
j=1
n!

1033
Multiplying the difference equation by the summing factor,
an+1 an 1
− =
n! (n − 1)! n!
 
an 1
D =
(n − 1)! n!
n−1
an X 1
− a1 =
(n − 1)! k=1
k!
" n−1 #
X 1
an = (n − 1)! + a1 .
k=1
k!

Example 24.4.2 Consider the equation

an+1 = λan + µ, for n ≥ 0.
From the above result, (with the products and sums starting at zero instead of one), the solution is
" n−1 # " n−1 " # #
Y X µ
a0 = λ Qk + a0
m=0 k=0 j=0 λ
" n−1 #
Xh µ i
n
=λ + a0
k=0
λk+1
 −n−1
− λ−1

n λ
=λ µ + a0
λ−1 − 1
 −n 
n λ −1
=λ µ + a0
1−λ
1 − λn
=µ + a 0 λn .
1−λ

1034
24.5 Homogeneous Constant Coefficient Equations
Homogeneous constant coefficient equations have the form

an+N + pN −1 an+N −1 + · · · + p1 an+1 + p0 an = 0.

The substitution an = rn yields

rN + pN −1 rN −1 + · · · + p1 r + p0 = 0
(r − r1 )m1 · · · (r − rk )mk = 0.

If r1 is a distinct root then the associated linearly independent solution is r1n . If r1 is a root of multiplicity
m > 1 then the associated solutions are r1n , nr1n , n2 r1n , . . . , nm−1 r1n .

Result 24.5.1 Consider the homogeneous constant coefficient difference equation

an+N + pN −1 an+N −1 + · · · + p1 an+1 + p0 an = 0.

The substitution an = rn yields the equation

(r − r1 )m1 · · · (r − rk )mk = 0.

A set of linearly independent solutions is

{r1n , nr1n , . . . , nm1 −1 r1n , . . . , rkn , nrkn , . . . , nmk −1 rkn }.

Example 24.5.1 Consider the equation an+2 − 3an+1 + 2an = 0 with the initial conditions a1 = 1 and a2 = 3.
The substitution an = rn yields

r2 − 3r + 2 = (r − 1)(r − 2) = 0.

1035
Thus the general solution is
an = c1 1n + c2 2n .
The initial conditions give the two equations,
a1 = 1 = c1 + 2c2
a2 = 3 = c1 + 4c2
Since c1 = −1 and c2 = 1, the solution to the difference equation subject to the initial conditions is
an = 2n − 1.

Example 24.5.2 Consider the gambler’s ruin problem that was introduced in Example 24.1.1. The equation for
the probability of the gambler’s ruin at n dollars is
an = pan+1 + qan−1 subject to a0 = 1, aN = 0.
We assume that 0 < p < 1. With the substitution an = rn we obtain
r = pr2 + q.
The roots of this equation are
√
1± 1 − 4pq
r=
2p
p
1 ± 1 − 4p(1 − p)
=
2p
p
1 ± (1 − 2p)2
=
2p
1 ± |1 − 2p|
= .
2p
We will consider the two cases p 6= 1/2 and p = 1/2.

1036
p 6= 1/2. If p < 1/2, the roots are
1 ± (1 − 2p)
r=
2p
1−p q
r1 = = , r2 = 1.
p p
If p > 1/2 the roots are
1 ± (2p − 1)
r=
2p
−p + 1 q
r1 = 1, r2 = = .
p p
Thus the general solution for p 6= 1/2 is
 n
q
an = c1 + c2 .
p
The boundary condition a0 = 1 requires that c1 + c2 = 1. From the boundary condition aN = 0 we have
 N
q
(1 − c2 ) + c2 =0
p
−1
c2 =
−1 + (q/p)N
pN
c2 = N .
p − qN
Solving for c1 ,
pN
c1 = 1 −
pN − q N
−q N
c1 = N .
p − qN

1037
 Thus we have
 n
−q N pN q
an = N + .
p − qN pN − q N p

p = 1/2. In this case, the two roots of the polynomial are both 1. The general solution is
an = c1 + c2 n.
The left boundary condition demands that c1 = 1. From the right boundary condition we obtain
1 + c2 N = 0
1
c2 = − .
N
Thus the solution for this case is
n
an = 1 − .
N
N/2
As a check that this formula makes sense, we see that for n = N/2 the probability of ruin is 1 − N
= 12 .

24.6 Reduction of Order

Consider the difference equation
(n + 1)(n + 2)an+2 − 3(n + 1)an+1 + 2an = 0 for n ≥ 0 (24.1)
We see that one solution to this equation is an = 1/n!. Analogous to the reduction of order for differential
equations, the substitution an = bn /n! will reduce the order of the difference equation.
(n + 1)(n + 2)bn+2 3(n + 1)bn+1 2bn
− + =0
(n + 2)! (n + 1)! n!
bn+2 − 3bn+1 + 2bn = 0 (24.2)

1038
At first glance it appears that we have not reduced the order of the equation, but writing it in terms of discrete
derivatives

D2 bn − Dbn = 0

shows that we now have a first order difference equation for Dbn . The substitution bn = rn in equation 24.2 yields
the algebraic equation

r2 − 3r + 2 = (r − 1)(r − 2) = 0.

Thus the solutions are bn = 1 and bn = 2n . Only the bn = 2n solution will give us another linearly independent
solution for an . Thus the second solution for an is an = bn /n! = 2n /n!. The general solution to equation 24.1 is
then

1 2n
an = c 1 + c2 .
n! n!

Result 24.6.1 Let an = sn be a homogeneous solution of a linear difference equation.

The substitution an = sn bn will yield a difference equation for bn that is of order one less
than the equation for an .

1039
24.7 Exercises
Exercise 24.1
Find a formula for the nth term in the Fibonacci sequence 1, 1, 2, 3, 5, 8, 13, . . . .
Hint, Solution

Exercise 24.2
Solve the difference equation

2
an+2 = an , a1 = a2 = 1.
n
Hint, Solution

1040
24.8 Hints
Hint 24.1
The difference equation corresponding to the Fibonacci sequence is

an+2 − an+1 − an = 0, a1 = a2 = 1.

Hint 24.2
Consider this exercise as two first order difference equations; one for the even terms, one for the odd terms.

1041
24.9 Solutions
Solution 24.1
We can describe the Fibonacci sequence with the difference equation

an+2 − an+1 − an = 0, a1 = a2 = 1.

With the substitution an = rn we obtain the equation

r2 − r − 1 = 0.

This equation has the two distinct roots

√ √
1+ 5 1− 5
r1 = , r2 = .
2 2

Thus the general solution is

√ !n √ !n
1+ 5 1− 5
an = c 1 + c2 .
2 2

From the initial conditions we have

c1 r1 +c2 r2 = 1
c1 r12 +c2 r22 = 1.

Solving for c2 in the first equation,

1
c2 = (1 − c1 r1 ).
r2

1042
We substitute this into the second equation.

1
c1 r12 + (1 − c1 r1 )r22 = 1
r2
c1 (r12 − r1 r2 ) = 1 − r2

1 − r2
c1 =
r12− r1 r2
√
1− 5
1− 2
= √ √
1+ 5
2
5
√
1+ 5
2
= √ √
1+ 5
2
5
1
=√
5

Substitute this result into the equation for c2 .

 
1 1
c2 = 1 − √ r1
r2 5
√ !
2 1 1+ 5
= √ 1− √
1− 5 5 2
√ !
2 1− 5
=− √ √
1− 5 2 5
1
= −√
5

1043
Thus the nth term in the Fibonacci sequence has the formula
√ !n √ !n
1 1+ 5 1 1− 5
an = √ −√ .
5 2 5 2

It is interesting to note that although the Fibonacci sequence is defined in terms of integers, one cannot express
the formula form the nth element in terms of rational numbers.

Solution 24.2
We can consider
2
an+2 = an , a1 = a2 = 1
n
to be a first order difference equation. First consider the odd terms.
a1 = 1
2
a3 =
1
22
a5 =
31
2(n−1)/2
an =
(n − 2)(n − 4) · · · (1)
For the even terms,
a2 = 1
2
a4 =
2
22
a6 =
42
2(n−2)/2
an = .
(n − 2)(n − 4) · · · (2)

1044
Thus
(
2(n−1)/2
(n−2)(n−4)···(1)
for odd n
an = 2(n−2)/2
(n−2)(n−4)···(2)
for even n.

1045
Chapter 25

Series Solutions of Differential Equations

Skill beats honesty any day.

-?

25.1 Ordinary Points

Big O and Little o Notation. The notation O(z n ) means “terms no bigger than z n .” This gives us a convenient
shorthand for manipulating series. For example,
z3
sin z = z − + O(z 5 )
6

1
= 1 + O(z)
1−z
The notation o(z n ) means “terms smaller that z n .” For example,

cos z = 1 + o(1)

1046
 ez = 1 + z + o(z)

Example 25.1.1 Consider the equation

w00 (z) − 3w0 (z) + 2w(z) = 0.

The general solution to this constant coefficient equation is

w = c1 ez + c2 e2z .

The functions ez and e2z are analytic in the finite complex plane. Recall that a function is analytic at a point z0
if and only if the function has a Taylor series about z0 with a nonzero radius of convergence. If we substitute the
Taylor series expansions about z = 0 of ez and e2z into the general solution, we obtain
∞ ∞
X zn X 2n z n
w = c1 + c2 .
n=0
n! n=0
n!

Thus we have a series solution of the differential equation.

Alternatively, we could tryPsubstituting a Taylor series into the differential equation and solving for the
coefficients. Substituting w = ∞ n
n=0 an z into the differential equation yields
∞ ∞ ∞
d2 X n d X n
X
an z − 3 an z + 2 an z n = 0
dz 2 n=0 dz n=0 n=0
∞
X ∞
X ∞
X
n−2 n−1
n(n − 1)an z −3 nan z +2 an z n = 0
n=2 n=1 n=0
∞
X X∞ ∞
X
n n
(n + 2)(n + 1)an+2 z − 3 (n + 1)an+1 z + 2 an z n = 0
n=0 n=0 n=0
∞ h
X i
(n + 2)(n + 1)an+2 − 3(n + 1)an+1 + 2an z n = 0.
n=0

1047
Equating powers of z, we obtain the difference equation
(n + 2)(n + 1)an+2 − 3(n + 1)an+1 + 2an = 0, n ≥ 0.
We see that an = 1/n! is one solution since
(n + 2)(n + 1) n+1 1 1−3+2
−3 +2 = = 0.
(n + 2)! (n + 1)! n! n!
We use reduction of order for difference equations to find the other solution. Substituting an = bn /n! into the
difference equation yields
bn+2 bn+1 bn
(n + 2)(n + 1) − 3(n + 1) +2 =0
(n + 2)! (n + 1)! n!
bn+2 − 3bn+1 + 2bn = 0.
At first glance it appears that we have not reduced the order of the difference equation. However writing this
equation in terms of discrete derivatives,
D2 bn − Dbn = 0
we see that this is a first order difference equation for Dbn . Since this is a constant coefficient difference equation
we substitute bn = rn into the equation to obtain an algebraic equation for r.
r2 − 3r + 2 = (r − 1)(r − 2) = 0
Thus the two solutions are bn = 1n b0 and bn = 2n b0 . Only bn = 2n b0 will give us a second independent solution
for an . Thus the two solutions for an are
a0 2 n a0
an = and an = .
n! n!
Thus we can write the general solution to the differential equation as
∞ ∞
X zn X 2n z n
w = c1 + c2 .
n=0
n! n=0
n!

1048
We recognize these two sums as the Taylor expansions of ez and e2z . Thus we obtain the same result as we did
solving the differential equation directly.

Of course it would be pretty silly to go through all the grunge involved in developing a series expansion of the
solution in a problem like Example 25.1.1 since we can solve the problem exactly. However if we could not solve
a differential equation, then having a Taylor series expansion of the solution about a point z0 would be useful in
determining the behavior of the solutions near that point.
For this method of substituting a Taylor series into the differential equation to be useful we have to know at
what points the solutions are analytic. Let’s say we were considering a second order differential equation whose
solutions were
1
w1 = , and w2 = log z.
z
Trying to find a Taylor series expansion of the solutions about the point z = 0 would fail because the solutions
are not analytic at z = 0. This brings us to two important questions.
1. Can we tell if the solutions to a linear differential equation are analytic at a point without knowing the
solutions?
2. If there are Taylor series expansions of the solutions to a differential equation, what are the radii of conver-
gence of the series?
In order to answer these questions, we will introduce the concept of an ordinary point. Consider the nth order
linear homogeneous equation
dn w dn−1 w dw
n
+ pn−1 (z) n−1
+ · · · + p1 (z) + p0 (z)w = 0.
dz dz dz
If each of the coefficient functions pi (z) are analytic at z = z0 then z0 is an ordinary point of the differential
equation.
For reasons of typography we will restrict our attention to second order equations and the point z0 = 0 for a
while. The generalization to an nth order equation will be apparent. Considering the point z0 6= 0 is only trivially
more general as we could introduce the transformation z − z0 → z to move the point to the origin.

1049
 In the chapter on first order differential equations we showed that the solution is analytic at ordinary points.
One would guess that this remains true for higher order equations. Consider the second order equation
y 00 + p(z)y 0 + q(z)y = 0,
where p and q are analytic at the origin.
∞
X ∞
X
n
p(z) = pn z , and q(z) = qn z n
n=0 n=0

Assume that one of the solutions is not analytic at the origin and behaves like z α at z = 0 where α 6= 0, 1, 2, . . . .
That is, we can approximate the solution with w(z) = z α + o(z α ). Let’s substitute w = z α + o(z α ) into the
differential equation and look at the lowest power of z in each of the terms.
∞ ∞
α−2 α−2 α−1 α−1
X n α
Xα
qn z n = 0.
   
α(α − 1)z + o(z ) + αz + o(z ) pn z + z + o(z )
n=0 n=0
α
We see that the solution could not possibly behave like z , α 6= 0, 1, 2, · · · because there is no term on the left to
cancel out the z α−2 term. The terms on the left side could not add to zero.
You could also check that a solution could not possibly behave like log z at the origin. Though we will not
prove it, if z0 is an ordinary point of a homogeneous differential equation, then all the solutions are analytic at
the point z0 . Since the solution is analytic at z0 we can expand it in a Taylor series.

Now we are prepared to answer our second question. From complex variables, we know that the radius of
convergence of the Taylor series expansion of a function is the distance to the nearest singularity of that function.
Since the solutions to a differential equation are analytic at ordinary points of the equation, the series expansion
about an ordinary point will have a radius of convergence at least as large as the distance to the nearest singularity
of the coefficient functions.

Example 25.1.2 Consider the equation

1
w00 + w0 + z 2 w = 0.
cos z

1050
If we expand the solution to the differential equation in Taylor series about z = 0, the radius of convergence will
be at least π/2. This is because the coefficient functions are analytic at the origin, and the nearest singularities
of 1/ cos z are at z = ±π/2.

25.1.1 Taylor Series Expansion for a Second Order Differential Equation

Consider the differential equation
w00 + p(z)w0 + q(z)w = 0
where p(z) and q(z) are analytic in some neighborhood of the origin.
∞
X ∞
X
n
p(z) = pn z and q(z) = qn z n
n=0 n=0

We substitute a Taylor series and it’s derivatives

X∞
w= an z n
n=0
X∞ ∞
X
w0 = nzn z n−1 = (n + 1)an+1 z n
n=1 n=0
∞
X ∞
X
00 n−2
w = n(n − 1)an z = (n + 2)(n + 1)an+2 z n
n=2 n=0

into the differential equation to obtain

∞ ∞
! ∞
!
X X X
(n + 2)(n + 1)an+2 z n + pn z n (n + 1)an+1 z n
n=0 n=0 n=0
∞
! ∞
!
X X
+ qn z n an z n =0
n=0 n=0

1051
 ∞ ∞ n
! ∞ n
!
X X X X X
n
(n + 2)(n + 1)an+2 z + (m + 1)am+1 pn−m z n + am qn−m z n = 0
n=0 n=0 m=0 n=0 m=0
∞
" n
#
X X
z n = 0.


(n + 2)(n + 1)an+2 + (m + 1)am+1 pn−m + am qn−m
n=0 m=0

Equating coefficients of powers of z,

n
X

(n + 2)(n + 1)an+2 + (m + 1)am+1 pn−m + am qn−m = 0 for n ≥ 0.
m=0

We see that a0 and a1 are arbitrary and the rest of the coefficients are determined by the recurrence relation
n
1 X
an+2 =− ((m + 1)am+1 pn−m + am qn−m ) for n ≥ 0.
(n + 1)(n + 2) m=0

Example 25.1.3 Consider the problem

1 0
y 00 + y + ex y = 0, y(0) = y 0 (0) = 1.
cos x
Let’s expand the solution in a Taylor series about the origin.
∞
X
y(x) = a n xn
n=0

Since y(0) = a0 and y 0 (0) = a1 , we see that a0 = a1 = 1. The Taylor expansions of the coefficient functions are

1
= 1 + O(x), and ex = 1 + O(x).
cos x

1052
Now we can calculate a2 from the recurrence relation.
0
1 X
a2 = − ((m + 1)am+1 p0−m + am q0−m )
1 · 2 m=0
1
= − (1 · 1 · 1 + 1 · 1)
2
= −1
Thus the solution to the problem is
y(x) = 1 + x − x2 + O(x3 ).
In Figure 25.1 the numerical solution is plotted in a solid line and the sum of the first three terms of the Taylor
series is plotted in a dashed line.

The general recurrence relation for the an ’s is useful if you only want to calculate the first few terms in the
Taylor expansion. However, for many problems substituting the Taylor series for the coefficient functions into the
differential equation will enable you to find a simpler form of the solution. We consider the following example to
illustrate this point.

Example 25.1.4 Develop a series expansion of the solution to the initial value problem
1
w00 + 2 w = 0, w(0) = 1, w0 (0) = 0.
(z + 1)

Solution using the General Recurrence Relation. The coefficient function has the Taylor expansion
∞
1 X
= (−1)n z 2n .
1 + z2 n=0

From the initial condition we obtain a0 = 1 and a1 = 0. Thus we see that the solution is
X∞
w= an z n ,
n=0

1053
 1.2

1.1

0.2 0.4 0.6 0.8 1 1.2 1.4

0.9

0.8

0.7

Figure 25.1: Plot of the Numerical Solution and the First Three Terms in the Taylor Series.

where
n
1 X
an+2 = − am qn−m
(n + 1)(n + 2) m=0

and
(
0 for odd n
qn =
(−1)(n/2) for even n.

Although this formula is fine if you only want to calculate the first few an ’s, it is just a tad unwieldy to work
with. Let’s see if we can get a better expression for the solution.

1054
Substitute the Taylor Series into the Differential Equation. Substituting a Taylor series for w yields
∞ ∞
d2 X n 1 X
an z + 2 an z n = 0.
dz 2 n=0 (z + 1) n=0

Note that the algebra will be easier if we multiply by z 2 + 1. The polynomial z 2 + 1 has only two terms, but the
Taylor series for 1/(z 2 + 1) has an infinite number of terms.
∞ ∞
2 d2 X n
X
(z + 1) 2 an z + an z n = 0
dz n=0 n=0
∞
X ∞
X ∞
X
n n−2
n(n − 1)an z + n(n − 1)an z + an z n = 0
n=2 n=2 n=0
∞
X ∞
X ∞
X
n(n − 1)an z n + (n + 2)(n + 1)an+2 z n + an z n = 0
n=0 n=0 n=0
∞ h
X i
(n + 2)(n + 1)an+2 + n(n − 1)an + an z n = 0
n=0

Equating powers of z gives us the difference equation

n2 − n + 1
an+2 = − an , for n ≥ 0.
(n + 2)(n + 1)

From the initial conditions we see that a0 = 1 and a1 = 0. All of the odd terms in the series will be zero. For
the even terms, it is easier to reformulate the problem with the change of variables bn = a2n . In terms of bn the
difference equation is

(2n)2 − 2n + 1
bn+1 = − bn , b0 = 1.
(2n + 2)(2n + 1)

1055
This is a first order difference equation with the solution
n 
4j 2 − 2j + 1
Y 
bn = − .
j=0
(2j + 2)(2j + 1)

Thus we have that

(Qn/2  
4j 2 −2j+1
j=0 − (2j+2)(2j+1)
for even n,
an =
0 for odd n.

Note that the nearest singularities of 1/(z 2 + 1) in the complex plane are at z = ±i. Thus the radius of
convergence must be at least 1. Applying the ratio test, the series converges for values of |z| such that
 an+2 z n+2 
 
lim  <1
n→∞  an z n 
n2 − n + 1  2
 

lim −
 |z| < 1
n→∞  (n + 2)(n + 1) 

|z|2 < 1.

The radius of convergence is 1.

The first few terms in the Taylor expansion are

1 1 13 6
w = 1 − z2 + z4 − z + ··· .
2 8 240

In Figure 25.2 the plot of the first two nonzero terms is shown in a short dashed line, the plot of the first four
nonzero terms is shown in a long dashed line, and the numerical solution is shown in a solid line.

In general, if the coefficient functions are rational functions, that is they are fractions of polynomials, multi-
plying the equations by the quotient will reduce the algebra involved in finding the series solution.

1056
 0.2 0.4 0.6 0.8 1 1.2
0.9

0.8

0.7

0.6

0.5

0.4

0.3

Figure 25.2: Plot of the solution and approximations.

Example 25.1.5 If we were going to find the Taylor series expansion about z = 0 of the solution to
z 1
w00 + w0 + w = 0,
1+z 1 − z2
we would first want to multiply the equation by 1 − z 2 to obtain

(1 − z 2 )w00 + z(1 − z)w00 + w = 0.

Example 25.1.6 Find the series expansions about z = 0 of the fundamental set of solutions for

w00 + z 2 w = 0.

1057
Recall that the fundamental set of solutions {w1 , w2 } satisfy

w1 (0) = 1 w2 (0) = 0
w10 (0) = 0 w20 (0) = 1.

Thus if
∞
X ∞
X
w1 = an z n and w2 = bn z n ,
n=0 n=0

then the coefficients must satisfy

a0 = 1,
a1 = 0, and b0 = 0, b1 = 1.

Substituting the Taylor expansion w = ∞ n

P
n=0 cn z into the differential equation,
∞
X ∞
X
n−2
n(n − 1)cn z + cn z n+2 = 0
n=2 n=0
∞
X ∞
X
n
(n + 2)(n + 1)cn+2 z + cn−2 z n = 0
n=0 n=2
∞ h
X i
2c2 + 6c3 z + (n + 2)(n + 1)cn+2 + cn−2 z n = 0
n=2

Equating coefficients of powers of z,

z 0 : c2 = 0
z 1 : c3 = 0
z n : (n + 2)(n + 1)cn+2 + cn−2 = 0, for n ≥ 2
cn
cn+4 = −
(n + 4)(n + 3)

1058
 For our first solution we have the difference equation
an
a0 = 1, a1 = 0, a2 = 0, a3 = 0, an+4 = − .
(n + 4)(n + 3)

For our second solution,

bn
b0 = 0, b1 = 1, b2 = 0, b3 = 0, bn+4 = − .
(n + 4)(n + 3)

The first few terms in the fundamental set of solutions are

1 4 1 8 1 5 1 9
w1 = 1 − z + z − ··· , w2 = z − z + z − ··· .
12 672 20 1440

In Figure 25.3 the five term approximation is graphed in a coarse dashed line, the ten term approximation is
graphed in a fine dashed line, and the numerical solution of w1 is graphed in a solid line. The same is done for
w2 .

Result 25.1.1 Consider the nth order linear homogeneous equation

dn w dn−1 w dw
+ p n−1 (z) + · · · + p 1 (z) + p0 (z)w = 0.
dz n dz n−1 dz
If each of the coefficient functions pi (z) are analytic at z = z0 then z0 is an ordinary
point of the differential equation. The solution is analytic in some region containing z0
and can be expanded in a Taylor series. The radius of convergence of the series will be
at least the distance to the nearest singularity of the coefficient functions in the complex
plane.

1059
 1.5 1.5

1 1

0.5 0.5

1 2 3 4 5 6 1 2 3 4 5 6

-0.5 -0.5

-1 -1

Figure 25.3: The graph of approximations and numerical solution of w1 and w2 .

25.2 Regular Singular Points of Second Order Equations

Consider the differential equation
p(z) 0 q(z)
w00 + w + w = 0.
z − z0 (z − z0 )2
If z = z0 is not an ordinary point but both p(z) and q(z) are analytic at z = z0 then z0 is a regular singular
point of the differential equation. The following equations have a regular singular point at z = 0.
• w00 + z1 w0 + z 2 w = 0
• w00 + 1
sin z
w0 −w =0

1060
 • w00 − zw0 + 1
z sin z
w =0
Concerning regular singular points of second order linear equations there is good news and bad news.

The Good News. We will find that with the use of the Frobenius method we can always find series expansions
of two linearly independent solutions at a regular singular point. We will illustrate this theory with several
examples.

The Bad News. Instead of a tidy little theory like we have for ordinary points, the solutions can be of several
different forms. Also, for some of the problems the algebra can get pretty ugly.

Example 25.2.1 Consider the equation

3(1 + z)
w00 + w = 0.
16z 2
P∞
We wish to find series solutions about the point z = 0. First we try a Taylor series w = n=0 an z n . Substituting
this into the differential equation,
∞ ∞
2
X
n−2 3 X
z n(n − 1)an z + (1 + z) an z n = 0
n=2
16 n=0
∞ ∞ ∞
X 3 X 3 X
n(n − 1)an z n + an z n + an+1 z n = 0.
n=0
16 n=0 16 n=1

Equating powers of z,
z0 : a0 = 0
 
3 3
zn : n(n − 1) + an + an+1 = 0
16 16
 
16
an+1 = n(n − 1) + 1 an .
3

1061
This difference equation has the solution an = 0 for all n. Thus we have obtained only the trivial solution to the
differential equation. We must try an expansion of a more general form. We recall that for regular
P singular points
of first order equations we can always find a solution in the form of a Frobenius series w = z α ∞ a
n=0 n z n
, a0 6= 0.
We substitute this series into the differential equation.
∞ ∞
2
X  n+α−2 3 α
X
an z n = 0

z α(α − 1) + 2αn + n(n − 1) an z + (1 + z)z
n=0
16 n=0
∞ ∞ ∞
X 3 X 3 X
α(α − 1) + 2n + n(n − 1) an z n + an z n + an−1 z n = 0
 
n=0
16 n=0 16 n=1

Equating the z 0 term to zero yields the equation

 
3
α(α − 1) + a0 = 0.
16
Since we have assumed that a0 6= 0, the polynomial in α must be zero. The two roots of the polynomial are
p p
1 + 1 − 3/4 3 1 − 1 − 3/4 1
α1 = = , α2 = = .
2 4 2 4
Thus our two series solutions will be of the form
∞
X ∞
X
3/4 n 1/4
w1 = z an z , w2 = z bn z n .
n=0 n=0

Substituting the first series into the differential equation,

∞   ∞
X 3 3 n 3 X
− + 2n + n(n − 1) + an z + an−1 z n = 0.
n=0
16 16 16 n=1

Equating powers of z, we see that a0 is arbitrary and

3
an = − an−1 for n ≥ 1.
16n(n + 1)

1062
This difference equation has the solution
n  
Y 3
an = a0 −
j=1
16j(j + 1)
 n Yn
3 1
= a0 −
16 j=1 j(j + 1)
 n
3 1
= a0 − for n ≥ 1.
16 n!(n + 1)!
Substituting the second series into the differential equation,
∞   ∞
X 3 3 n 3 X
− + 2n + n(n − 1) + bn z + bn−1 z n = 0.
n=0
16 16 16 n=1

We see that the difference equation for bn is the same as the equation for an . Thus we can write the general
solution to the differential equation as
∞  n ! ∞  n !
X 3 1 X 3 1
w = c1 z 3/4 1 + − z n + c2 z 1/4 1 + − zn
n=1
16 n!(n + 1)! n=1
16 n!(n + 1)!

∞  n !
X 3 1
c1 z 3/4 + c2 z 1/4
zn .


1+ −
n=1
16 n!(n + 1)!

25.2.1 Indicial Equation

Now let’s consider the general equation for a regular singular point at z = 0
p(z) 0 q(z)
w00 + w + 2 w = 0.
z z

1063
Since p(z) and q(z) are analytic at z = 0 we can expand them in Taylor series.
∞
X ∞
X
p(z) = pn z n , q(z) = qn z n
n=0 n=0

α
P∞
Substituting a Frobenius series w = z n=0 an z n , a0 6= 0 and the Taylor series for p(z) and q(z) into the
differential equation yields
∞ h ∞
! ∞
! ∞
! ∞
!
X i X X X X
(α + n)(α + n − 1) an z n + pn z n (α + n)an z n + qn z n an z n =0
n=0 n=0 n=0 n=0 n=0

∞ h
X i
(α + n) − (α + n) + p0 (α + n) + q0 an z n
2

n=0
∞
! ∞
! ∞
! ∞
!
X X X X
+ pn z n (α + n)an z n + qn z n an z n =0
n=1 n=0 n=1 n=0

∞ h ∞ n−1
! ∞ n−1
!
X i X X X X
2 n
(α + n) + (p0 − 1)(αn ) + q0 an z + (α + j)aj pn−j zn + aj qn−j zn = 0
n=0 n=1 j=0 n=1 j=0

Equating powers of z,
h i
z0 : α2 + (p0 − 1)α + q0 a0 = 0
h i n−1 h
X i
n 2
z : (α + n) + (p0 − 1)(α + n) + q0 an = − (α + j)pn−j + qn−j aj .
j=0

Let
I(α) = α2 + (p0 − 1)α + q0 = 0.

1064
This is known as the indicial equation. The indicial equation gives us the form of the solutions. The equation
for a0 is I(α)a0 = 0. Since we assumed that a0 is nonzero, I(α) = 0. Let the two roots of I(α) be α1 and α2
where <(α1 ) ≥ <(α2 ).
Rewriting the difference equation for an (α),
n−1 h
X i
I(α + n)an (α) = − (α + j)pn−j + qn−j aj (α) for n ≥ 1. (25.1)
j=0

If the roots are distinct and do not differ by an integer then we can use Equation 25.1 to solve for an (α1 ) and
an (α2 ), which will give us the two solutions
∞
X ∞
X
α1 n α2
w1 = z an (α1 )z , and w2 = z an (α2 )z n .
n=0 n=0

If the roots are not distinct, α1 = α2 , we will only have one solution and will have to generate another. If
the roots differ by an integer, α1 − α2 = N , there is one solution corresponding to α1 , but when we try to solve
Equation 25.1 for an (α2 ), we will encounter the equation
N
X −1 h i
I(α2 + N )aN (α2 ) = I(α1 )aN (α2 ) = 0 · aN (α2 ) = − (α + n)pn−j + qn−j aj (α2 ).
j=0

If the right side of the equation is nonzero, then aN (α2 ) is undefined. On the other hand, if the right side is zero
then aN (α2 ) is arbitrary. The rest of this section is devoted to considering the cases α1 = α2 and α1 − α2 = N .

25.2.2 The Case: Double Root

Consider a second order equation L[w] = 0 with a regular singular point at z = 0. Suppose the indicial equation
has a double root.

I(α) = (α − α1 )2 = 0

1065
One solution has the form
∞
X
α1
w1 = z an z n .
n=0

In order to find the second solution, we will differentiate with respect to the parameter, α. Let an (α) satisfy
Equation 25.1 Substituting the Frobenius expansion into the differential equation,
" ∞
#
X
L zα an (α)z n = 0.
n=0

Setting α = α1 will make the left hand side of the equation zero. Differentiating this equation with respect to α,
" ∞
#
∂ X
L zα an (α)z n = 0.
∂α n=0

Interchanging the order of differentiation,

" ∞ ∞
#
α
X
n α
X dan (α) n
L log z z an (α)z + z z = 0.
n=0 n=0
dα

Since setting α = α1 will make the left hand side of this equation zero, the second linearly independent solution
is
∞ ∞

α1
X
n α1
X da n (α) 
w2 = log z z an (α1 )z + z zn

dα


n=0 n=0 α=α1

∞
X
w2 = w1 log z + z α1
a0n (α1 )z n .
n=0

1066
Example 25.2.2 Consider the differential equation
1+z
w00 + w = 0.
4z 2
There is a regular singular point at z = 0. The indicial equation is
 2
1 1
α(α − 1) + = α − = 0.
4 2
One solution will have the form
∞
X
w1 = z 1/2 an z n , a0 6= 0.
n=0

Substituting the Frobenius expansion

∞
X
zα an (α)z n
n=0

into the differential equation yields

1
z 2 w00 + (1 + z)w = 0
4
∞ ∞ ∞
X 1X 1X
α(α − 1) + 2αn + n(n − 1) an (α)z n+α + an (α)z n+α + an (α)z n+α+1 = 0.
 
n=0
4 n=0
4 n=0

Divide by z α and adjust the summation indices.

∞ ∞ ∞
X 1X n n 1X
[α(α − 1) + 2αn + n(n − 1)] an (α)z + an (α)z + an−1 (α)z n = 0
n=0
4 n=0 4 n=1
  ∞   
1 X 1 1
α(α − 1)a0 + a0 + α(α − 1) + 2n + n(n − 1) + an (α) + an−1 (α) z n = 0
4 n=1
4 4

1067
Equating the coefficient of z 0 to zero yields I(α)a0 = 0. Equating the coefficients of z n to zero yields the difference
equation
 
1 1
α(α − 1) + 2n + n(n − 1) + an (α) + an−1 (α) = 0
4 4
 
n(n + 1) α(α − 1) 1
an (α) = − + + an−1 (α).
4 4 16

The first few an ’s are

    
9 25 9
a0 , − α(α − 1) + a0 , α(α − 1) + α(α − 1) + a0 , . . .
16 16 16

Setting α = 1/2, the coefficients for the first solution are

5 105
a0 , − a0 , a0 , ...
16 16

The second solution has the form

∞
X
w2 = w1 log z + z 1/2 a0n (1/2)z n .
n=0

Differentiating the an (α),

   
da0 da1 (α) da2 (α) 9 25
= 0, = −(2α − 1)a0 , = (2α − 1) α(α − 1) + + α(α − 1) + a0 , ...
dα dα dα 16 16

Setting α = 1/2 in this equation yields

a00 = 0, a01 (1/2) = 0, a02 (1/2) = 0, ...

1068
Thus the second solution is
w2 = w1 log z.
The first few terms in the general solution are
 
5 105 2
(c1 + c2 log z) 1 − z + z − ··· .
16 16

25.2.3 The Case: Roots Differ by an Integer

Consider the case in which the roots of the indicial equation α1 and α2 differ by an integer. (α1 − α2 = N ) Recall
the equation that determines an (α)
h i n−1 h
X i
2
I(α + n)an = (α + n) + (p0 − 1)(α + n) + q0 an = − (α + j)pn−j + qn−j aj .
j=0

When α = α2 the equation for aN is

N
X −1 h i
I(α2 + N )aN (α2 ) = 0 · aN (α2 ) = − (α + j)pN −j + qN −j aj .
j=0

If the right hand side of this equation is zero, then aN is arbitrary. There will be two solutions of the Frobenius
form.
∞
X ∞
X
α1 n α2
w1 = z an (α1 )z and w2 = z an (α2 )z n .
n=0 n=0

If the right hand side of the equation is nonzero then aN (α2 ) will be undefined. We will have to generate the
second solution. Let
∞
X
α
w(z, α) = z an (α)z n ,
n=0

1069
where an (α) satisfies the recurrence formula. Substituting this series into the differential equation yields

L[w(z, α)] = 0.

We will multiply by (α − α2 ), differentiate this equation with respect to α and then set α = α2 . This will generate
a linearly independent solution.
 
∂ ∂
L[(α − α2 )w(z, α)] = L (α − α2 )w(z, α)
∂α ∂α
" ∞
#
∂ X
=L (α − α2 )z α an (α)z n
∂α n=0
" ∞ ∞
#
X X d
= L log z z α (α − α2 )an (α)z n + z α [(α − α2 )an (α)]z n
n=0 n=0
dα

Setting α = α2 with make this expression zero, thus

∞ ∞  
α
X
n α2
X d
log z z lim {(α − α2 )an (α)} z + z lim [(α − α2 )an (α)] z n
n=0
α→α2
n=0
α→α2 dα

is a solution. Now let’s look at the first term in this solution

∞
X
α
log z z lim {(α − α2 )an (α)} z n .
α→α2
n=0

The first N terms in the sum will be zero. That is because a0 , . . . , aN −1 are finite, so multiplying by (α − α2 ) and
taking the limit as α → α2 will make the coefficients vanish. The equation for aN (α) is
N
X −1 h i
I(α + N )aN (α) = − (α + j)pN −j + qN −j aj (α).
j=0

1070
Thus the coefficient of the N th term is
" N −1
#
(α − α2 ) X h i
lim (α − α2 )aN (α) = − lim (α + j)pN −j + qN −j aj (α)
α→α2 α→α2 I(α + N ) j=0
" N −1 h
#
(α − α2 ) X i
= − lim (α + j)pN −j + qN −j aj (α)
α→α2 (α + N − α1 )(α + N − α2 ) j=0

α−α2
Since α1 = α2 + N , limα→α2 α+N −α1
= 1.

N −1 h
1 X i
=− (α2 + j)pN −j + qN −j aj (α2 ).
(α1 − α2 ) j=0

Using this you can show that the first term in the solution can be written
d−1 log z w1 ,
where d−1 is a constant. Thus the second linearly independent solution is
∞
X
w2 = d−1 log z w1 + z α2 dn z n ,
n=0

where
N −1 h
1 1 X i
d−1 = − (α2 + j)pN −j + qN −j aj (α2 )
a0 (α1 − α2 ) j=0

and
 i
d h
dn = lim (α − α2 )an (α) for n ≥ 0.
α→α2 dα

1071
Example 25.2.3 Consider the differential equation
 
00 2 2
w + 1− w0 + 2 w = 0.
z z
The point z = 0 is a regular singular point. In order to find series expansions of the solutions, we first calculate
the indicial equation. We can write the coefficient functions in the form
p(z) 1 q(z) 1
= (−2 + z), and 2
= 2 (2).
z z z z
Thus the indicial equation is
α2 + (−2 − 1)α + 2 = 0
(α − 1)(α − 2) = 0.

The First Solution. The first solution will have the Frobenius form
∞
X
2
w1 = z an (α1 )z n .
n=0

Substituting a Frobenius series into the differential equation,

z 2 w00 + (z 2 − 2z)w0 + 2w = 0
∞
X X∞ ∞
X
(n + α)(n + α − 1)z n+α + (z 2 − 2z) (n + α)z n+α−1 + 2 an z n = 0
n=0 n=0 n=0
∞ h
X i
[α2 − 3α + 2]a0 + (n + α)(n + α − 1)an + (n + α − 1)an−1 − 2(n + α)an + 2an z n = 0.
n=1

Equating powers of z,
h i
(n + α)(n + α − 1) − 2(n + α) + 2 an = −(n + α − 1)an−1
an−1
an = − .
n+α−2

1072
 Setting α = α1 = 2, the recurrence relation becomes

an−1 (α1 )
an (α1 ) = −
n
(−1)n
= a0 .
n!
The first solution is
∞
X (−1)n
w1 = a0 z n = a0 e−z .
n=0
n!

The Second Solution. The equation for a1 (α2 ) is

0 · a1 (α2 ) = 2a0 .

Since the right hand side of this equation is not zero, the second solution will have the form
∞  
α2
X d
w2 = d−1 log z w1 + z lim [(α − α2 )an (α)] z n
n=0
α→α2 dα

First we will calculate d−1 as we defined it previously.

1 1
d−1 = − a0 = −1.
a0 2 − 1

The expression for an (α) is

(−1)n a0
an (α) = .
(α + n − 2)(α + n − 1) · · · (α − 1)

1073
The first few an (α) are

a0
a1 (α) = −
α−1
a0
a2 (α) =
α(α − 1)
a0
a3 (α) = − .
(α + 1)α(α − 1)

We would like to calculate

 i
d h
dn = lim (α − 1)an (α) .
α→1 dα

1074
The first few dn are

 
d h i
d0 = lim (α − 1)a0
α→1 dα
= a0
   
d a0
d1 = lim (α − 1) −
α→1 dα α−1
 h 
d i
= lim − a0
α→1 dα
=0
   
d a0
d2 = lim (α − 1)
α→1 dα α(α − 1)
 
d a0
h i
= lim
α→1 dα α
= −a0
   
d a0
d3 = lim (α − 1) −
α→1 dα (α + 1)α(α − 1)
  
d a0
= lim −
α→1 dα (α + 1)α
3
= a0 .
4

It will take a little work to find the general expression for dn . We will need the following relations.

n−1
0
X 1
Γ(n) = (n − 1)!, Γ (z) = Γ(z)ψ(z), ψ(n) = −γ + .
k=1
k

1075
See the chapter on the Gamma function for explanations of these equations.
(−1)n a0
  
d
dn = lim (α − 1)
α→1 dα (α + n − 2)(α + n − 1) · · · (α − 1)
(−1)n a0
  
d
= lim
α→1 dα (α + n − 2)(α + n − 1) · · · (α)
d (−1)n a0 Γ(α)
  
= lim
α→1 dα Γ(α + n − 1)
 
n Γ(α)ψ(α) Γ(α)ψ(α + n − 1)
= (−1) a0 lim −
α→1 Γ(α + n − 1) Γ(α + n − 1)
 
n Γ(α)[ψ(α) − ψ(α + n − 1)]
= (−1) a0 lim
α→1 Γ(α + n − 1)
ψ(1) − ψ(n)
= (−1)n a0
(n − 1)!
n−1
(−1)n+1 a0 X 1
=
(n − 1)! k=0 k

Thus the second solution is

∞ n−1
!
X (−1)n+1 a0 X 1
w2 = − log z w1 + z zn.
n=0
(n − 1)! k=0 k

The general solution is

∞ n−1
!
X (−1)n+1 X 1
w = c1 e−z − c2 log z e−z + c2 z zn.
n=0
(n − 1)! k=0 k

We see that even in problems that are chosen for their simplicity, the algebra involved in the Frobenius method
can be pretty involved.

1076
Example 25.2.4 Consider a series expansion about the origin of the equation
1−z 0 1
w00 + w − 2 w = 0.
z z
The indicial equation is

α2 − 1 = 0
α = ±1.

Substituting a Frobenius series into the differential equation,

∞
X ∞
X ∞
X
z2 (n + α)(n + α − 1)an z n−2 + (z − z 2 ) (n + α)an z n−1 − an z n = 0
n=0 n=0 n=0
∞
X ∞
X ∞
X ∞
X
n n n
(n + α)(n + α − 1)an z + (n + α)an z − (n + α − 1)an−1 z − an z n = 0
n=0 n=0 n=1 n=0
h i ∞ h
X i
α(α − 1) + α − 1 a0 + n + α)(n + α − 1)an + (n + α − 1)an − (n + α − 1)an−1 z n = 0.
n=1

Equating powers of z to zero,

an−1 (α)
an (α) = .
n+α+1
We know that the first solution has the form
∞
X
w1 = z an z n .
n=0

Setting α = 1 in the reccurence formula,

an−1 2a0
an = = .
n+2 (n + 2)!

1077
Thus the first solution is
∞
X 2a0
w1 = z zn
n=0
(n + 2)!
∞
1 X z n+2
= 2a0
z n=0 (n + 2)!
∞
!
2a0 X zn
= −1−z
z n=0
n!
2a0 z
= ( e − 1 − z).
z
Now to find the second solution. Setting α = −1 in the reccurence formula,
an−1 a0
an = = .
n n!
We see that in this case there is no trouble in defining a2 (α2 ). The second solution is
∞
a0 X z n a0 z
w2 = = e.
z n=0 n! z

Thus we see that the general solution is

c1 z c2
w= ( e − 1 − z) + ez
z z

 
d1 z 1
w= e + d2 1 + .
z z

1078
25.3 Irregular Singular Points
If a point z0 of a differential equation is not ordinary or regular singular, then it is an irregular singular point.
At least one of the solutions at an irregular singular point will not be of the Frobenius form. We will examine
how to obtain series expansions about an irregular singular point in the chapter on asymptotic expansions.

25.4 The Point at Infinity

If we want to determine the behavior of a function f (z) at infinity, we can make the transformation t = 1/z and
examine the point t = 0.

Example 25.4.1 Consider the behavior of f (z) = sin z at infinity. This is the same as considering the point
t = 0 of sin(1/t), which has the series expansion

∞
(−1)n
  X
1
sin = .
t n=0
(2n + 1)!t2n+1

Thus we see that the point t = 0 is an essential singularity of sin(1/t). Hence sin z has an essential singularity at
z = ∞.

Example 25.4.2 Consider the behavior at infinity of z e1/z . With the transformation t = 1/z the function is

∞
1 t 1 X tn
e = .
t t n=0 n!

Thus z e1/z has a pole of order 1 at infinity.

1079
 In order to classify the point at infinity of a differential equation in w(z), we apply the transformation t = 1/z,
u(t) = w(z). Writing the derivatives with respect to z in terms if t yields
1
z=
t
1
dz = − 2 dt
t
d d
= −t2
dz dt
d2
 
2 d 2 d
= −t −t
dz 2 dt dt
2
d d
= t4 2 + 2t3 .
dt dt
Applying the transformation to the differential equation
w00 + p(z)w0 + q(z)w = 0
yields
t4 u00 + 2t3 u0 + p(1/t)(−t2 )u0 + q(1/t)u = 0
 
00 2 p(1/t) q(1/t)
u + − 2
u0 + u = 0.
t t t4

Example 25.4.3 Classify the singular points of the differential equation

1
w00 + w0 + 2w = 0.
z
There is a regular singular point at z = 0. To examine the point at infinity we make the transformation
t = 1/z, u(t) = w(z). The equation in u is
 
00 2 1 2
u + − u0 + 4 u = 0
t t t

1080
 1 2
u00 + u0 + 4 u = 0.
t t
Thus we see that the differential equation for w(z) has an irregular singular point at infinity.

1081
25.5 Exercises
Exercise 25.1 (mathematica/ode/series/series.nb)
f (x) satisfies the Hermite equation

d2 f df
− 2x + 2λf = 0.
dx2 dx

Construct two linearly independent solutions of the equation as Taylor series about x = 0. For what values of x
do the series converge?
Show that for certain values of λ, called eigenvalues, one of the solutions is a polynomial, called an eigenfunc-
tion. Calculate the first four eigenfunctions H0 (x), H1 (x), H2 (x), H3 (x), ordered by degree.
Hint, Solution

Exercise 25.2
Consider the Legendre equation

(1 − x2 )y 00 − 2xy 0 + α(α + 1)y = 0.

1. Find two linearly independent solutions in the form of power series about x = 0.

2. Compute the radius of convergence of the series. Explain why it is possible to predict the radius of conver-
gence without actually deriving the series.

3. Show that if α = 2n, with n an integer and n ≥ 0, the series for one of the solutions reduces to an even
polynomial of degree 2n.

4. Show that if α = 2n + 1, with n an integer and n ≥ 0, the series for one of the solutions reduces to an odd
polynomial of degree 2n + 1.

1082
 5. Show that the first 4 polynomial solutions Pn (x) (known as Legendre polynomials) ordered by their degree
and normalized so that Pn (1) = 1 are

P0 = 1 P1 = x
1 1
P2 = (3x2 − 1) P4 = (5x3 − 3x)
2 2

6. Show that the Legendre equation can also be written as

((1 − x2 )y 0 )0 = −α(α + 1)y.

Note that two Legendre polynomials Pn (x) and Pm (x) must satisfy this relation for α = n and α = m
respectively. By multiplying the first relation by Pm (x) and the second by Pn (x) and integrating by parts
show that Legendre polynomials satisfy the orthogonality relation
Z 1
Pn (x)Pm (x) dx = 0 if n 6= m.
−1

If n = m, it can be shown that the value of the integral is 2/(2n + 1). Verify this for the first three
polynomials (but you needn’t prove it in general).
Hint, Solution

Exercise 25.3
Find the forms of two linearly independent series expansions about the point z = 0 for the differential equation

1 1−z
w00 + w0 + 2 w = 0,
sin z z
such that the series are real-valued on the positive real axis. Do not calculate the coefficients in the expansions.
Hint, Solution

1083
Exercise 25.4
Classify the singular points of the equation

w0
w00 + + 2w = 0.
z−1
Hint, Solution

Exercise 25.5
Find the series expansions about z = 0 for

5 0 z−1
w00 + w + w = 0.
4z 8z 2
Hint, Solution

Exercise 25.6
Find the series expansions about z = 0 of the fundamental solutions of

w00 + zw0 + w = 0.

Hint, Solution

Exercise 25.7
Find the series expansions about z = 0 of the two linearly independent solutions of

1 0 1
w00 + w + w = 0.
2z z
Hint, Solution

1084
Exercise 25.8
Classify the singularity at infinity of the differential equation
 
00 2 3 1
w + + 2 w0 + 2 w = 0.
z z z
Find the forms of the series solutions of the differential equation about infinity that are real-valued when z is
real-valued and positive. Do not calculate the coefficients in the expansions.
Hint, Solution

Exercise 25.9
Consider the second order differential equation

d2 y dy
x 2
+ (b − x) − ay = 0,
dx dx
where a, b are real constants.
1. Show that x = 0 is a regular singular point. Determine the location of any additional singular points and
classify them. Include the point at infinity.

2. Compute the indicial equation for the point x = 0.

3. By solving an appropriate recursion relation, show that one solution has the form
ax (a)2 x2 (a)n xn
y1 (x) = 1 + + + ··· + + ···
b (b)2 2! (b)n n!

where the notation (a)n is defined by

(a)n = a(a + 1)(a + 2) · · · (a + n − 1), (a)0 = 1.

Assume throughout this problem that b 6= n where n is a non-negative integer.

1085
 4. Show that when a = −m, where m is a non-negative integer, that there are polynomial solutions to this
equation. Compute the radius of convergence of the series above when a 6= −m. Verify that the result you
get is in accord with the Frobenius theory.

5. Show that if b = n + 1 where n = 0, 1, 2, . . . , then the second solution of this equation has logarithmic terms.
Indicate the form of the second solution in this case. You need not compute any coefficients.

Hint, Solution

Exercise 25.10
Consider the equation

xy 00 + 2xy 0 + 6 ex y = 0.

Find the first three non-zero terms in each of two linearly independent series solutions about x = 0.
Hint, Solution

1086
25.6 Hints
Hint 25.1

Hint 25.2

Hint 25.3

Hint 25.4

Hint 25.5

Hint 25.6

Hint 25.7

Hint 25.8

1087
Hint 25.9

Hint 25.10

1088
25.7 Solutions
Solution 25.1
f (x) is a Taylor series about x = 0.
∞
X
f (x) = an xn
n=0
∞
X
f 0 (x) = nan xn−1
n=1
∞
X
= nan xn−1
n=0
∞
X
f 00 (x) = n(n − 1)an xn−2
n=2
∞
X
= (n + 2)(n + 1)an+2 xn
n=0

We substitute the Taylor series into the differential equation.

f 00 (x) − 2xf 0 (x) + 2λf = 0

∞
X X∞ ∞
X
n n
(n + 2)(n + 1)an+2 x − 2 nan x + 2λ a n xn
n=0 n=0 n=0

Equating coefficients gives us a difference equation for an :

(n + 2)(n + 1)an+2 − 2nan + 2λan = 0

n−λ
an+2 = 2 an .
(n + 1)(n + 2)

1089
The first two coefficients, a0 and a1 are arbitrary. The remaining coefficients are determined by the recurrence
relation. We will find the fundamental set of solutions at x = 0. That is, for the first solution we choose a0 = 1
and a1 = 0; for the second solution we choose a0 = 0, a1 = 1. The difference equation for y1 is

n−λ
an+2 = 2 an , a0 = 1, a1 = 0,
(n + 1)(n + 2)

which has the solution

Qn
2n k=0 (2(n− k) − λ)
a2n = , a2n+1 = 0.
(2n)!

The difference equation for y2 is

n−λ
an+2 = 2 an , a0 = 0, a1 = 1,
(n + 1)(n + 2)

which has the solution

Qn−1
2n k=0 (2(n − k) − 1 − λ)
a2n = 0, a2n+1 = .
(2n + 1)!

A set of linearly independent solutions, (in fact the fundamental set of solutions at x = 0), is

∞ ∞
2n nk=0 (2(n − k) − λ) 2n 2n n−1 (2(n − k) − 1 − λ)
X Q X Q
y1 (x) = x , y2 (x) = k=0
x2n+1 .
n=0
(2n)! n=0
(2n + 1)!

Since the coefficient functions in the differential equation do not have any singularities in the finite complex plane,
the radius of convergence of the series is infinite.
If λ = n is a positive even integer, then the first solution, y1 , is a polynomial of order n. If λ = n is a positive

1090
odd integer, then the second solution, y2 , is a polynomial of order n. For λ = 0, 1, 2, 3, we have

H0 (x) = 1
H1 (x) = x
H2 (x) = 1 − 2x2
2
H3 (x) = x − x3
3

Solution 25.2
1. First we write the differential equation in the standard form.

(1 − x2 )y 00 − 2xy 0 + α(α + 1)y = 0 (25.2)

2x α(α + 1)
y 00 − 2
+ y = 0. (25.3)
1−x 1 − x2
Since the coefficients of y 0 and y are analytic in a neighborhood of x = 0, We can find two Taylor series
solutions about that point. We find the Taylor series for y and its derivatives.
∞
X
y= an xn
n=0
∞
X
y0 = nan xn−1
n=1
X∞
y 00 = (n − 1)nan xn−2
n=2
∞
X
= (n + 1)(n + 2)an+2 xn
n=0

1091
Here we used index shifting to explicitly write the two forms that we will need for y 00 . Note that we can
take the lower bound of summation to be n = 0 for all above sums. The terms added by this operation are
zero. We substitute the Taylor series into Equation 25.2.
∞
X ∞
X ∞
X ∞
X
n n n
(n + 1)(n + 2)an+2 x − (n − 1)nan x − 2 nan x + α(α + 1) an xn = 0
n=0 n=0 n=0 n=0
∞ 
X

(n + 1)(n + 2)an+2 − (n − 1)n + 2n − α(α + 1) an xn = 0
n=0

We equate coefficients of xn to obtain a recurrence relation.

(n + 1)(n + 2)an+2 = (n(n + 1) − α(α + 1))an

n(n + 1) − α(α + 1)
an+2 = an , n ≥ 0.
(n + 1)(n + 2)

We can solve this difference equation to determine the an ’s. (a0 and a1 are arbitrary.)

n−2
 a0 Y

k(k + 1) − α(α + 1) , even n,


 n! k=0



even k
an = n−2
 a1 Y

k(k + 1) − α(α + 1) , odd n



 n!


k=1
odd k

We will find the fundamental set of solutions at x = 0, that is the set {y1 , y2 } that satisfies

y1 (0) = 1 y10 (0) = 0

y2 (0) = 0 y20 (0) = 1.

1092
 For y1 we take a0 = 1 and a1 = 0; for y2 we take a0 = 0 and a1 = 1. The rest of the coefficients are
determined from the recurrence relation.
 
∞ n−2
X 1 Y
k(k + 1) − α(α + 1)  xn


y1 =
n!

n=0 k=0
even n even k
 
∞ n−2
X 1 Y
k(k + 1) − α(α + 1)  xn


y2 =
n! k=1

n=1
odd n odd k

2. We determine the radius of convergence of the series solutions with the ratio test.
 an+2 xn+2 
 
lim  <1
n→∞  a n xn 
 n(n+1)−α(α+1) 
n+2 

 (n+1)(n+2) a n x
lim  <1

n→∞  an x n 
 
 n(n + 1) − α(α + 1)   2 
lim   x  < 1
n→∞ (n + 1)(n + 2) 
 2
x  < 1

Thus we see that the radius of convergence of the series is 1. We knew that the radius of convergence would
be at least one, because the nearest singularities of the coefficients of (25.3) occur at x = ±1, a distance
of 1 from the origin. This implies that the solutions of the equation are analytic in the unit circle about
x = 0. The radius of convergence of the Taylor series expansion of an analytic function is the distance to
the nearest singularity.

3. If α = 2n then a2n+2 = 0 in our first solution. From the recurrence relation, we see that all subsequent

1093
 coefficients are also zero. The solution becomes an even polynomial.
 
2n
X  1 Y m−2
k(k + 1) − α(α + 1)  xm


y1 =
m! k=0

m=0
even m even k

4. If α = 2n + 1 then a2n+3 = 0 in our second solution. From the recurrence relation, we see that all subsequent
coefficients are also zero. The solution becomes an odd polynomial.
 
2n+1
X  1 Y m−2
k(k + 1) − α(α + 1)  xm


y2 =
m! k=1

m=1
odd m odd k

5. From our solutions above, the first four polynomials are

1
x
1 − 3x2
5
x − x3
3
To obtain the Legendre polynomials we normalize these to have value unity at x = 1
P0 = 1
P1 = x
1
3x2 − 1


P2 =
2
1
5x3 − 3x


P3 =
2
These four Legendre polynomials are plotted in Figure 25.4.

1094
 Figure 25.4: The First Four Legendre Polynomials

6. We note that the first two terms in the Legendre equation form an exact derivative. Thus the Legendre
equation can also be written as

0
(1 − x2 )y 0 = −α(α + 1)y.

Pn and Pm are solutions of the Legendre equation.

0
0
(1 − x2 )Pn0 = −n(n + 1)Pn , (1 − x2 )Pm0 = −m(m + 1)Pm (25.4)

We multiply the first relation of Equation 25.4 by Pm and integrate by parts.

0
(1 − x2 )Pn0 Pm = −n(n + 1)Pn Pm
Z 1 Z 1
2 0 0


(1 − x )Pn Pm dx = −n(n + 1) Pn Pm dx
−1 −1
Z 1 Z 1
 2 0

1 2 0 0
(1 − x )Pn Pm −1 − (1 − x )Pn Pm dx = −n(n + 1) Pn Pm dx
−1 −1
Z 1 Z 1
2 0 0
(1 − x )Pn Pm dx = n(n + 1) Pn Pm dx
−1 −1

1095
 We multiply the
R 1 secord2 relation of Equation 25.4 by Pn and integrate by parts. To obtain a different
0 0
expression for −1 (1 − x )Pm Pn dx.
Z 1 Z 1
(1 − x 2
)Pm0 Pn0 dx = m(m + 1) Pm Pn dx
−1 −1
R1
We equate the two expressions for −1
(1 − x2 )Pm0 Pn0 dx. to obtain an orthogonality relation.
Z 1
(n(n + 1) − m(m + 1)) Pn Pm dx = 0
−1
Z 1
Pn (x)Pm (x) dx = 0 if n 6= m.
−1

We verify that for the first four polynomials the value of the integral is 2/(2n + 1) for n = m.
Z 1 Z 1
P0 (x)P0 (x) dx = 1 dx = 2
−1 −1
1 11
x3
Z Z 
22
P1 (x)P1 (x) dx = x dx = =
−1 −1 3 −1 3
Z 1 Z 1   5 1
1 4 2

1 9x 3 2
P2 (x)P2 (x) dx = 9x − 6x + 1 dx = − 2x + x =
−1 −1 4 4 5 −1 5
Z 1 Z 1 1
1 25x7
  
1 2
25x6 − 30x4 + 9x2 dx = − 6x5 + 3x3


P3 (x)P3 (x) dx = =
−1 −1 4 4 7 −1 7

Solution 25.3
The indicial equation for this problem is

α2 + 1 = 0.

1096
Since the two roots α1 = i and α2 = −i are distinct and do not differ by an integer, there are two solutions in the
Frobenius form.
∞
X X∞
w1 = z i an z n , w1 = z −i bn z n
n=0 n=0

However, these series are not real-valued on the positive real axis. Recalling that
z i = ei log z = cos(log z) + i sin(log z), and z −i = e−i log z = cos(log z) − i sin(log z),
we can write a new set of solutions that are real-valued on the positive real axis as linear combinations of w1 and
w2 .
1 1
u1 = (w1 + w2 ), u2 = (w1 − w2 )
2 2i
∞
X ∞
X
n
u1 = cos(log z) cn z , u1 = sin(log z) dn z n
n=0 n=0

Solution 25.4
Consider the equation w00 + w0 /(z − 1) + 2w = 0.
We see that there is a regular singular point at z = 1. All other finite values of z are ordinary points of the
equation. To examine the point at infinity we introduce the transformation z = 1/t, w(z) = u(t). Writing the
derivatives with respect to z in terms of t yields
d d d2 4 d
2
3 d
= −t2 , = t + 2t .
dz dt dz 2 dt2 dt
Substituting into the differential equation gives us
t2 u0
t4 u00 + 2t3 u0 − + 2u = 0
1/t − 1
 
00 2 1 2
u + − u0 + 4 u = 0.
t t(1 − t) t

1097
Since t = 0 is an irregular singular point in the equation for u(t), z = ∞ is an irregular singular point in the
equation for w(z).

Solution 25.5
Find the series expansions about z = 0 for
5 0 z−1
w00 + w + w = 0.
4z 8z 2
We see that z = 0 is a regular singular point of the equation. The indicial equation is
1 1
α2 + α − = 0
 4 8 
1 1
α+ α− = 0.
2 4
Since the roots are distinct and do not differ by an integer, there will be two solutions in the Frobenius form.
∞
X X∞
1/4 n −1/2
w1 = z an (α1 )z , w2 = z an (α2 )z n
n=0 n=0
2
We multiply the differential equation by 8z to put it in a better form. Substituting a Frobenius series into
the differential equation,
∞
X ∞
X ∞
X
2 n+α−2 n+α−1
8z (n + α)(n + α − 1)an z + 10z (n + α)an z + (z − 1) an z n+α
n=0 n=0 n=0
∞
X ∞
X ∞
X ∞
X
8 (n + α)(n + α − 1)an z n + 10 (n + α)an z n + an−1 z n − an z n .
n=0 n=0 n=1 n=0

Equating coefficients of powers of z,

[8(n + α)(n + α − 1) + 10(n + α) − 1] an = −an−1
an−1
an = − 2
.
8(n + α) + 2(n + α) − 1

1098
The First Solution. Setting α = 1/4 in the recurrence formula,
an−1
an (α1 ) = −
8(n + 1/4)2 + 2(n + 1/4) − 1
an−1
an (α1 ) = − .
2n(4n + 3)
Thus the first solution is
∞  
1/4
X
n 1/4 1 1 2
w1 = z an (α1 )z = a0 z 1− z+ z + ··· .
n=0
14 616

The Second Solution. Setting α = −1/2 in the recurrence formula,

an−1
an = −
8(n − 1/2)2 + 2(n − 1/2) − 1
an−1
an = −
2n(4n − 3)
Thus the second linearly independent solution is
∞  
−1/2
X
n −1/2 1 1 2
w2 = z an (α2 )z = a0 z 1 − z + z + ··· .
n=0
2 40

Solution 25.6
We consider the series solutions of,

w00 + zw0 + w = 0.

We would like to find the expansions of the fundamental set of solutions about z = 0. Since z = 0 is a regular
point, (the coefficient functions are analytic there), we expand the solutions in Taylor series. Differentiating the

1099
series expansions for w(z),

∞
X
w= an z n
n=0
∞
X
w0 = nan z n−1
n=1
X∞
w00 = n(n − 1)an z n−2
n=2
∞
X
= (n + 2)(n + 1)an+2 z n
n=0

We may take the lower limit of summation to be zero without changing the sums. Substituting these expressions
into the differential equation,

∞
X ∞
X ∞
X
n n
(n + 2)(n + 1)an+2 z + nan z + an z n = 0
n=0 n=0 n=0
∞
X
(n + 2)(n + 1)an+2 + (n + 1)an z n = 0.


n=0

Equating the coefficient of the z n term gives us

(n + 2)(n + 1)an+2 + (n + 1)an = 0, n≥0

an
an+2 = − , n ≥ 0.
n+2

a0 and a1 are arbitrary. We determine the rest of the coefficients from the recurrence relation. We consider the

1100
cases for even and odd n separately.
a2n−2
a2n = −
2n
a2n−4
=
(2n)(2n − 2)
a0
= (−1)n
(2n)(2n − 2) · · · 4 · 2
a0
= (−1)n Qn , n≥0
m=1 2m

a2n−1
a2n+1 = −
2n + 1
a2n−3
=
(2n + 1)(2n − 1)
a1
= (−1)n
(2n + 1)(2n − 1) · · · 5 · 3
a1
= (−1)n Qn , n≥0
m=1 (2m + 1)

If {w1 , w2 } is the fundamental set of solutions, then the initial conditions demand that w1 = 1 + 0 · z + · · · and
w2 = 0 + z + · · · . We see that w1 will have only even powers of z and w2 will have only odd powers of z.

∞ ∞
X (−1)n 2n X (−1)n
w1 = Qn z , w2 = Qn z 2n+1
n=0 m=1 2m n=0 m=1 (2m + 1)

Since the coefficient functions in the differential equation are entire, (analytic in the finite complex plane), the
radius of convergence of these series solutions is infinite.

1101
Solution 25.7

1 0 1
w00 +w + w = 0.
2z z
We can find the indicial equation by substituting w = z + O(z α+1 ) into the differential equation.
α

1
α(α − 1)z α−2 + αz α−2 + z α−1 = O(z α−1 )
2
Equating the coefficient of the z α−2 term,
1
α(α − 1) + α = 0
2
1
α = 0, .
2
Since the roots are distinct and do not differ by an integer, the solutions are of the form
∞
X ∞
X
w1 = an z n , w2 = z 1/2 bn z n .
n=0 n=0

Differentiating the series for the first solution,

∞
X
w1 = an z n
n=0
∞
X
w10 = nan z n−1
n=1
X∞
= (n + 1)an+1 z n
n=0
∞
X
w100 = n(n + 1)an+1 z n−1 .
n=1

1102
Substituting this series into the differential equation,
∞ ∞ ∞
X
n−1 1 X n 1X
n(n + 1)an+1 z + (n + 1)an+1 z + an z n = 0
n=1
2z n=0 z n=0

∞  
X 1 1 1
n(n + 1)an+1 + (n + 1)an+1 + an z n−1 + a1 + a0 = 0.
n=1
2 2z z

Equating powers of z,
a1
z −1 : + a0 = 0 ⇒ a1 = −2a0
2 
1 an
z n−1 : n+ (n + 1)an+1 + an = 0 ⇒ an+1 = − .
2 (n + 1/2)(n + 1)

We can combine the above two equations for an .

an
an+1 = − , for n ≥ 0
(n + 1/2)(n + 1)

Solving this difference equation for an ,

n−1
Y −1
an = a0
j=0
(j + 1/2)(j + 1)

n−1
(−1)n Y 1
an = a0
n! j=0 j + 1/2

1103
 Now let’s find the second solution. Differentiating w2 ,
∞
X
w20 = (n + 1/2)bn z n−1/2
n=0
X∞
w200 = (n + 1/2)(n − 1/2)bn z n−3/2 .
n=0

Substituting these expansions into the differential equation,

∞ ∞ ∞
X
n−3/2 1X n−3/2
X
(n + 1/2)(n − 1/2)bn z + (n + 1/2)bn z + bn−1 z n−3/2 = 0.
n=0
2 n=0 n=1

Equating the coefficient of the z −3/2 term,

 
1 1 11
− b0 + b0 = 0,
2 2 22
we see that b0 is arbitrary. Equating the other coefficients of powers of z,
1
(n + 1/2)(n − 1/2)bn + (n + 1/2)bn + bn−1 = 0
2
bn−1
bn = −
n(n + 1/2)
Calculating the bn ’s,
b0
b1 = −
1 · 32
b0
b2 =
1 · 2 · 32 · 52
(−1)n 2n b0
bn =
n! · 3 · 5 · · · (2n + 1)

1104
Thus the second solution is
∞
X (−1)n 2n z n
w2 = b0 z 1/2 .
n=0
n! 3 · 5 · · · (2n + 1)

Solution 25.8

 
00 2 3 1
w + + 2 w0 + w = 0.
z z z2
In order to analyze the behavior at infinity we make the change of variables t = 1/z, u(t) = w(z) and examine
the point t = 0. Writing the derivatives with respect to z in terms if t yields
1
z=
t
1
dz = − 2 dt
t
d d
= −t2
dz dt

d2
 
2 d 2 d
= −t −t
dz 2 dt dt
2
d d
= t4 2 + 2t3 .
dt dt
The equation for u is then
t4 u00 + 2t3 u0 + (2t + 3t2 )(−t2 )u0 + t2 u = 0
1
u00 + −3u0 + 2 u = 0
t

1105
We see that t = 0 is a regular singular point. To find the indicial equation, we substitute u = tα + O(tα+1 ) into
the differential equation.

α(α − 1)tα−2 − 3αtα−1 + tα−2 = O(tα−1 )

Equating the coefficients of the tα−2 terms,

α(α − 1) + 1 = 0
√
1±i 3
α=
2
Since the roots of the indicial equation are distinct and do not differ by an integer, a set of solutions has the form
( ∞ ∞
)
√ X √ X
t(1+i 3)/2 an tn , t(1−i 3)/2 bn tn .
n=0 n=0

Noting that
√ ! √ !
√ i 3 √ i 3
(1+i 3)/2
t = t1/2 exp log t , and t (1−i 3)/2
= t1/2 exp − log t .
2 2

We can take the sum and difference of the above solutions to obtain the form
√ ! ∞ √ ! ∞
3 X 3 X
u1 = t1/2 cos log t an tn , u1 = t1/2 sin log t bn tn .
2 n=0
2 n=0

Putting the answer in terms of z, we have the form of the two Frobenius expansions about infinity.
√ ! ∞ √ ! ∞
3 X an 3 X bn
w1 = z −1/2 cos log z n
, w1 = z −1/2 sin log z n
.
2 n=0
z 2 n=0
z

1106
Solution 25.9
1. We write the equation in the standard form.

b−x 0 a
y 00 + y − y=0
x x

Since b−x
x
has no worse than a first order pole and xa has no worse than a second order pole at x = 0, that
is a regular singular point. Since the coefficient functions have no other singularities in the finite complex
plane, all the other points in the finite complex plane are regular points.

Now to examine the point at infinity. We make the change of variables u(ξ) = y(x), ξ = 1/x.

dξ d 1
y0 = u = − 2 u0 = −ξ 2 u0
dx dξ x
 
d d
y 00 = −ξ 2 −ξ 2 u = ξ 4 u00 + 2ξ 3 u0
dξ dξ

The differential equation becomes

xy 00 + (b − x)y 0 − ay
 
1 4 00 3 0 1
−ξ 2 u0 − au = 0



ξ u + 2ξ u + b −
ξ ξ
ξ u + (2 − b)ξ + ξ u0 − au = 0
3 00 2


 
00 2−b 1 a
u + + 2 − 3u = 0
ξ ξ ξ

Since this equation has an irregular singular point at ξ = 0, the equation for y(x) has an irregular singular
point at infinity.

1107
2. The coefficient functions are

∞
1X 1
p(x) ≡ pn xn = (b − x),
x n=1 x
∞
1 X 1
q(x) ≡ 2 qn xn = 2 (0 − ax).
x n=1 x

The indicial equation is

α2 + (p0 − 1)α + q0 = 0
α2 + (b − 1)α + 0 = 0
α(α + b − 1) = 0.

3. Since one of the roots of the indicial equation is zero, and the other root is not a negative integer, one of

1108
the solutions of the differential equation is a Taylor series.
∞
X
y1 = c k xk
k=0
∞
X
y10 = kck xk−1
k=1
X∞
= (k + 1)ck+1 xk
k=0
∞
X
= kck xk−1
k=0
X
y100 = k(k − 1)ck xk−2
k=2
∞
X
= (k + 1)kck+1 xk−1
k=1
∞
X
= (k + 1)kck+1 xk−1
k=0

We substitute the Taylor series into the differential equation.

xy 00 + (b − x)y 0 − ay = 0
∞
X ∞
X X∞ ∞
X
(k + 1)kck+1 xk + b (k + 1)ck+1 xk − kck xk − a c k xk = 0
k=0 k=0 k=0 k=0

We equate coefficients to determine a recurrence relation for the coefficients.

(k + 1)kck+1 + b(k + 1)ck+1 − kck − ack = 0
k+a
ck+1 = ck
(k + 1)(k + b)

1109
 For c0 = 1, the recurrence relation has the solution
(a)k xk
ck = .
(b)k k!
Thus one solution is
∞
X (a)k k
y1 (x) = x .
k=0
(b)k k!

4. If a = −m, where m is a non-negative integer, then (a)k = 0 for k > m. This makes y1 a polynomial:
m
X (a)k k
y1 (x) = x .
k=0
(b)k k!

5. If b = n + 1, where n is a non-negative integer, the indicial equation is

α(α + n) = 0.

For the case n = 0, the indicial equation has a double root at zero. Thus the solutions have the form:
m ∞
X (a)k k X
y1 (x) = x , y2 (x) = y1 (x) log x + d k xk
k=0
(b)k k! k=0

For the case n > 0 the roots of the indicial equation differ by an integer. The solutions have the form:
m ∞
X (a)k k X
y1 (x) = x , y2 (x) = d−1 y1 (x) log x + x−n dk xk
k=0
(b)k k!
k=0

The form of the solution for y2 can be substituted into the equation to determine the coefficients dk .

1110
Solution 25.10
We write the equation in the standard form.

xy 00 + 2xy 0 + 6 ex y = 0
ex
y 00 + 2y 0 + 6 y = 0
x
We see that x = 0 is a regular singular point. The indicial equation is

α2 − α = 0
α = 0, 1.

The first solution has the Frobenius form.

y1 = x + a2 x2 + a3 x3 + O(x4 )

We substitute y1 into the differential equation and equate coefficients of powers of x.

xy 00 + 2xy 0 + 6 ex y = 0

x(2a2 + 6a3 x + O(x2 )) + 2x(1 + 2a2 x + 3a3 x2 + O(x3 ))

+ 6(1 + x + x2 /2 + O(x3 ))(x + a2 x2 + a3 x3 + O(x4 )) = 0

(2a2 x + 6a3 x2 ) + (2x + 4a2 x2 ) + (6x + 6(1 + a2 )x2 ) = O(x3 ) = 0

17
a2 = −4, a3 =
3
17
y1 = x − 4x2 + x3 + O(x4 )
3

1111
Now we see if the second solution has the Frobenius form. There is no a1 x term because y2 is only determined
up to an additive constant times y1 .
y2 = 1 + O(x2 )
We substitute y2 into the differential equation and equate coefficients of powers of x.
xy 00 + 2xy 0 + 6 ex y = 0
O(x) + O(x) + 6(1 + O(x))(1 + O(x2 )) = 0
6 = O(x)
The substitution y2 = 1 + O(x) has yielded a contradiction. Since the second solution is not of the Frobenius
form, it has the following form:
y2 = y1 ln(x) + a0 + a2 x2 + O(x3 )
The first three terms in the solution are
y2 = a0 + x ln x − 4x2 ln x + O(x2 ).
We calculate the derivatives of y2 .
y20 = ln(x) + O(1)
1
y200 = + O(ln(x))
x
We substitute y2 into the differential equation and equate coefficients.
xy 00 + 2xy 0 + 6 ex y = 0
(1 + O(x ln x)) + 2 (O(x ln x)) + 6 (a0 + O(x ln x)) = 0
1 + 6a0 = 0
1
y2 = − + x ln x − 4x2 ln x + O(x2 )
6

1112
Chapter 26

Asymptotic Expansions

The more you sweat in practice, the less you bleed in battle.

-Navy Seal Saying

26.1 Asymptotic Relations

The  and ∼ symbols. First we will introduce two new symbols used in asymptotic relations.

f (x)  g(x) as x → x0 ,

is read, “f (x) is much smaller than g(x) as x tends to x0 ”. This means

f (x)
lim = 0.
x→x0 g(x)

The notation

f (x) ∼ g(x) as x → x0 ,

1113
is read “f (x) is asymptotic to g(x) as x tends to x0 ”; which means

f (x)
lim = 1.
x→x0 g(x)

A few simple examples are

• − ex  x as x → +∞

• sin x ∼ x as x → 0

• 1/x  1 as x → +∞

• e−1/x  x−n as x → 0+ for all n

An equivalent definition of f (x) ∼ g(x) as x → x0 is

f (x) − g(x)  g(x) as x → x0 .

Note that it does not make sense to say that a function f (x) is asymptotic to zero. Using the above definition
this would imply

f (x)  0 as x → x0 .

If you encounter an expression like f (x) + g(x) ∼ 0, take this to mean f (x) ∼ −g(x).

The Big O and Little o Notation. If |f (x)| ≤ m|g(x)| for some constant m in some neighborhood of the
point x = x0 , then we say that

f (x) = O(g(x)) as x → x0 .

We read this as “f is big O of g as x goes to x0 ”. If g(x) does not vanish, an equivalent definition is that f (x)/g(x)
is bounded as x → x0 .

1114
 If for any given positive δ there exists a neighborhood of x = x0 in which |f (x)| ≤ δ|g(x)| then
f (x) = o(g(x)) as x → x0 .
This is read, “f is little o of g as x goes to x0 .”
For a few examples of the use of this notation,
• e−x = o(x−n ) as x → ∞ for any n.
• sin x = O(x) as x → 0.
• cos x − 1 = o(1) as x → 0.
• log x = o(xα ) as x → +∞ for any positive α.

Operations on Asymptotic Relations. You can perform the ordinary arithmetic operations on asymptotic
relations. Addition, multiplication, and division are valid.
You can always integrate an asymptotic relation. Integration is a smoothing operation. However, it is necessary
to exercise some care.

Example 26.1.1 Consider

1
f 0 (x) ∼ as x → ∞.
x2
This does not imply that
−1
f (x) ∼ as x → ∞.
x
We have forgotten the constant of integration. Integrating the asymptotic relation for f 0 (x) yields
−1
f (x) ∼ + c as x → ∞.
x
If c is nonzero then
f (x) ∼ c as x → ∞.

1115
 It is not always valid to differentiate an asymptotic relation.
1 1
Example 26.1.2 Consider f (x) = x
+ x2
sin(x3 ).

1
f (x) ∼ as x → ∞.
x
Differentiating this relation yields
1
f 0 (x) ∼ − as x → ∞.
x2
However, this is not true since
1 2
f 0 (x) = − 2
− 3 sin(x3 ) + 2 cos(x3 )
x x
1
6 − 2 as x → ∞.
∼
x

The Controlling Factor. The controlling factor is the most rapidly varying factor in an asymptotic relation.
Consider a function f (x) that is asymptotic to x2 ex as x goes to infinity. The controlling factor is ex . For a few
examples of this,

• x log x has the controlling factor x as x → ∞.

• x−2 e1/x has the controlling factor e1/x as x → 0.

• x−1 sin x has the controlling factor sin x as x → ∞.

The Leading Behavior. Consider a function that is asymptotic to a sum of terms.

f (x) ∼ a0 (x) + a1 (x) + a2 (x) + · · · , as x → x0 .

1116
where

a0 (x)  a1 (x)  a2 (x)  · · · , as x → x0 .

The first term in the sum is the leading order behavior. For a few examples,

• For sin x ∼ x − x3 /6 + x5 /120 − · · · as x → 0, the leading order behavior is x.

• For f (x) ∼ ex (1 − 1/x + 1/x2 − · · · ) as x → ∞, the leading order behavior is ex .

26.2 Leading Order Behavior of Differential Equations

It is often useful to know the leading order behavior of the solutions to a differential equation. If we are considering
a regular point or a regular singular point, the approach is straight forward. We simply use a Taylor expansion or
the Frobenius method. However, if we are considering an irregular singular point, we will have to be a little more
creative. Instead of an all encompassing theory like the Frobenius method which always gives us the solution, we
will use a heuristic approach that usually gives us the solution.

Example 26.2.1 Consider the Airy equation

y 00 = xy.

We 1 would like to know how the solutions of this equation behave as x → +∞. First we need to classify the
point at infinity. The change of variables

1 d d d2 4 d
2
d
x= , y(x) = u(t), = −t2 , 2
= t 2
+ 2t3
t dx dt dx dt dt
1
Using ”We” may be a bit presumptuous on my part. Even if you don’t particularly want to know how the solutions behave, I
urge you to just play along. This is an interesting section, I promise.

1117
yields
1
t4 u00 + 2t3 u0 = u
t
2 1
u00 + u0 − 5 u = 0.
t t
Since the equation for u has an irregular singular point at zero, the equation for y has an irregular singular point
at infinity.

The Controlling Factor. Since the solutions at irregular singular points often have exponential behavior, we
make the substitution y = es(x) into the differential equation for y.

d2  s 
2
e = x es
 00dx
s + (s0 )2 es = x es


s00 + (s0 )2 = x

The Dominant Balance. Now we have a differential equation for s that appears harder to solve than our
equation for y. However, we did not introduce the substitution in order to obtain an equation that we could solve
exactly. We are looking for an equation that we can solve approximately in the limit as x → ∞. If one of the
terms in the equation for s is much smaller that the other two as x → ∞, then dropping that term and solving the
simpler equation may give us an approximate solution. If one of the terms in the equation for s is much smaller
than the others then we say that the remaining terms form a dominant balance in the limit as x → ∞.
Assume that the s00 term is much smaller that the others, s00  (s0 )2 , x as x → ∞. This gives us

(s0 )2 ∼ x
√
s0 ∼ ± x
2
s ∼ ± x3/2 as x → ∞.
3

1118
Now let’s check√our assumption that the s00 term is small. Assuming that we can differentiate the asymptotic
relation s0 ∼ ± x, we obtain s00 ∼ ± 12 x−1/2 as x → ∞.

s00  (s0 )2 , x ⇒ x−1/2  x as x → ∞

Thus we see that the behavior we found for s is consistent with our assumption. The controlling factors for
solutions to the Airy equation are exp(± 32 x3/2 ) as x → ∞.

The Leading Order Behavior of the Decaying Solution. Let’s find the leading order behavior as x goes
to infinity of the solution with the controlling factor exp(− 23 x3/2 ). We substitute

2
s(x) = − x3/2 + t(x), where t(x)  x3/2 as x → ∞
3

into the differential equation for s.

s00 + (s0 )2 = x
1
− x−1/2 + t00 + (−x1/2 + t0 )2 = x
2
1
t00 + (t0 )2 − 2x1/2 t0 − x−1/2 = 0
2

Assume that we can differentiate t  x3/2 to obtain

t0  x1/2 , t00  x−1/2 as x → ∞.

1119
Since t00  − 12 x−1/2 we drop the t00 term. Also, t0  x1/2 implies that (t0 )2  −2x1/2 t0 , so we drop the (t0 )2 term.
This gives us
1
−2x1/2 t0 − x−1/2 ∼ 0
2
1
t0 ∼ − x−1
4
1
t ∼ − log x + c
4
1
t ∼ − log x as x → ∞.
4
Checking our assumptions about t,

t0  x1/2 ⇒ x−1  x1/2

t00  x−1/2 ⇒ x−2  x−1/2

we see that the behavior of t is consistent with our assumptions.

So far we have
 
2 3/2 1
y(x) ∼ exp − x − log x + u(x) as x → ∞,
3 4

where u(x)  log x as x → ∞. To continue, we substitute t(x) = − 14 log x + u(x) into the differential equation
for t(x).
1
t00 + (t0 )2 − 2x1/2 t0 − x−1/2 = 0
2
 2  
1 −2 00 1 −1 0 1 −1 1
x +u + − x +u − 2x 1/2
− x + u − x−1/2 = 0
0
4 4 4 2
 
1 5
u00 + (u0 )2 + − x−1 − 2x1/2 u0 + x−2 = 0
2 16

1120
Assume that we can differentiate the asymptotic relation for u to obtain
u0  x−1 , u00  x−2 as x → ∞.
We know that − 21 x−1 u0  −2x1/2 u0 . Using our assumptions,
5 −2
u00  x−2 ⇒ u00  x
16
5
u0  x−1 ⇒ (u0 )2  x−2 .
16
Thus we obtain
5 −2
−2x1/2 u0 + x ∼0
16
5 −5/2
u0 ∼ x
32
5
u ∼ − x−3/2 + c
48
u ∼ c as x → ∞.
Since u = c + o(1), eu = ec + o(1). The behavior of y is
 
−1/4 2 3/2
y∼x exp − x ( ec + o(1)) as x → ∞.
3
Thus the full leading order behavior of the decaying solution is
 
−1/4 2 3/2
y ∼ (const)x exp − x as x → ∞.
3
You can show that the leading behavior of the exponentially growing solution is
 
−1/4 2 3/2
y ∼ (const)x exp x as x → ∞.
3

1121
Example 26.2.2 The Modified Bessel Equation. Consider the modified Bessel equation

x2 y 00 + xy 0 − (x2 + ν 2 )y = 0.

We would like to know how the solutions of this equation behave as x → +∞. First we need to classify the point
at infinity. The change of variables x = 1t , y(x) = u(t) yields
 
1 4 00 3 0 1 2 0 1 2
(t u + 2t u ) + (−t u ) − 2 + ν u = 0
t2 t t
2
 
1 1 ν
u00 + u0 − 4 + 2 u = 0
t t t

Since u(t) has an irregular singular point at t = 0, y(x) has an irregular singular point at infinity.

The Controlling Factor. Since the solutions at irregular singular points often have exponential behavior, we
make the substitution y = es(x) into the differential equation for y.

x2 (s00 + (s0 )2 ) es + xs0 es − (x2 + ν 2 ) es = 0

1 ν2
s00 + (s0 )2 + s0 − (1 + 2 ) = 0
x x
We make the assumption that s00  (s0 )2 as x → ∞ and we know that ν 2 /x2  1 as x → ∞. Thus we drop these
two terms from the equation to obtain an approximate equation for s.
1
(s0 )2 + s0 − 1 ∼ 0
x
This is a quadratic equation for s0 , so we can solve it exactly. However, let us try to simplify the equation even
further. Assume that as x goes to infinity one of the three terms is much smaller that the other two. If this is
the case, there will be a balance between the two dominant terms and we can neglect the third. Let’s check the
three possibilities.

1122
 1.
1 1
1 is small. ⇒ (s0 )2 + s0 ∼ 0 ⇒ s0 ∼ − , 0
x x
1
1 6 x2
,0 as x → ∞ so this balance is inconsistent.
2.
1 0
s is small. ⇒ (s0 )2 − 1 ∼ 0 ⇒ s0 ∼ ±1
x
1
This balance is consistent as x
 1 as x → ∞.
3.
1 0
(s0 )2 is small. ⇒ s −1∼0 ⇒ s0 ∼ x
x
This balance is not consistent as x2 6 1 as x → ∞.
The only dominant balance that makes sense leads to s0 ∼ ±1 as x → ∞. Integrating this relationship,

s ∼ ±x + c
∼ ±x as x → ∞.

Now let’s see if our assumption that we made to get the simplified equation for s is valid. Assuming that we can
differentiate s0 ∼ ±1, s00  (s0 )2 becomes
d   2
± 1 + o(1)  ± 1 + o(1)
dx
0 + o(1/x)  1

Thus we see that the behavior we obtained for s is consistent with our initial assumption.
We have found two controlling factors, ex and e−x . This is a good sign as we know that there must be two
linearly independent solutions to the equation.

1123
Leading Order Behavior. Now let’s find the full leading behavior of the solution with the controlling factor
e−x . In order to find a better approximation for s, we substitute s(x) = −x + t(x), where t(x)  x as x → ∞,
into the differential equation for s.
ν2
 
00 0 2 1 0
s + (s ) + s − 1 + 2 = 0
x x
ν2
 
00 0 2 1 0
t + (−1 + t ) + (−1 + t ) − 1 + 2 = 0
x x
2
   
00 0 2 1 0 1 ν
t + (t ) + −2 t − + =0
x x x2
1 ν2 1
We know that x
 2 and x2
 x
as x → ∞. Dropping these terms from the equation yields
1
t00 + (t0 )2 − 2t0 − ∼ 0.
x
Assuming that we can differentiate the asymptotic relation for t, we obtain t0  1 and t00  1
x
as x → ∞. We
can drop t00 . Since t0 vanishes as x goes to infinity, (t0 )2  t0 . Thus we are left with
1
−2t0 − ∼ 0, as x → ∞.
x
Integrating this relationship,
1
t ∼ − log x + c
2
1
∼ − log x as x → ∞.
2
Checking our assumptions about the behavior of t,
1
t0  1 ⇒ − 1
2x
1 1 1
t00  ⇒ 2

x 2x x

1124
we see that the solution is consistent with our assumptions.
The leading order behavior to the solution with controlling factor e−x is
 
1
y(x) ∼ exp −x − log x + u(x) = x−1/2 e−x+u(x) as x → ∞,
2
where u(x)  log x. We substitute t = − 21 log x + u(x) into the differential equation for t in order to find the
asymptotic behavior of u.
1 ν2
   
00 0 2 1 0
t + (t ) + −2 t − + =0
x x x2
2 
1 ν2
    
1 00 1 0 1 1 0
+u + − +u + −2 − +u − + =0
2x2 2x x 2x x x2
1 ν2
u00 + (u0 )2 − 2u0 + 2 − 2 = 0
4x x
Assuming that we can differentiate the asymptotic relation for u, u0  1
x
and u00  1
x2
as x → ∞. Thus we see
that we can neglect the u00 and (u0 )2 terms.
 
0 1 2 1
−2u + −ν ∼0
4 x2

 
0 1 1 2 1
u ∼ −ν
2 4 x2
 
1 1 1
u∼ ν2 − +c
2 4 x
u ∼ c as x → ∞

Since u = c + o(1), we can expand eu as ec + o(1). Thus we can write the leading order behavior as

y ∼ x−1/2 e−x ( ec + o(1)).

1125
Thus the full leading order behavior is

y ∼ (const)x−1/2 e−x as x → ∞.

You can verify that the solution with the controlling factor ex has the leading order behavior

y ∼ (const)x−1/2 ex as x → ∞.

Two linearly independent solutions to the modified Bessel equation are the modified Bessel functions, Iν (x)
and Kν (x). These functions have the asymptotic behavior

√
1 π
Iν (x) ∼ √ ex , Kν (x) ∼ √ e−x as x → ∞.
2πx 2x
√
π −x
In Figure 26.1 K0 (x) is plotted in a solid line and √2x e is plotted in a dashed line. We see that the leading
order behavior of the solution as x goes to infinity gives a good approximation to the behavior even for fairly
small values of x.

26.3 Integration by Parts

Example 26.3.1 The complementary error function
Z ∞
2 2
erfc (x) = √ e−t dt
π x

1126
 2

1.75

1.5

1.25

0.75

0.5

0.25

0 1 2 3 4 5

Figure 26.1: Plot of K0 (x) and it’s leading order behavior.

is used in statistics for its relation to the normal probability distribution. We would like to find an approximation
to erfc (x) for large x. Using integration by parts,

Z ∞ 
2 −1 2

erfc (x) = √ −2t e−t dt
π x 2t
 ∞ Z ∞
2 −1 −t2 2 1 −2 −t2
=√ e −√ t e dt
π 2t x π x 2
Z ∞
1 −1 −x2 1 2
=√ x e −√ t−2 e−t dt.
π π x

1127
We examine the residual integral in this expression.
Z ∞
−1 −3 ∞
Z
1 −2 −t2 2
√ t e dt ≤ √ x −2t e−t dt
π x 2 π x
1 −3 −x2
= √ x e .
2 π
Thus we see that
Z ∞
1 −1 −x2 1 2
√ x e √ t−2 e−t dt as x → ∞.
π π x
Therefore,
1 2
erfc (x) ∼ √ x−1 e−x as x → ∞,
π
2
and we expect that √1 x−1 e−x would be a good approximation to erfc (x) for large x. In Figure 26.2 log( erfc (x))
π  
2
is graphed in a solid line and log √1π x−1 e−x is graphed in a dashed line. We see that this first approximation
to the error function gives very good results even for moderate values of x. Table 26.1 gives the error in this first
approximation for various values of x.
If we continue integrating by parts, we might get a better approximation to the complementary error function.
Z ∞
1 −1 −x2 1 2
erfc (x) = √ x e −√ t−2 e−t dt
π π x
 ∞ Z ∞
1 −1 −x2 1 1 −3 −t2 1 3 −4 −t2
=√ x e −√ − t e +√ t e dt
π π 2 x π x 2
  Z ∞
1 −x2 −1 1 −3 1 3 −4 −t2
=√ e x − x +√ t e dt
π 2 π x 2
   ∞ Z ∞
1 −x2 −1 1 −3 1 3 −5 −t2 1 15 −6 −t2
=√ e x − x +√ − t e −√ t e dt
π 2 π 4 x π x 4
  Z ∞
1 −x2 −1 1 −3 3 −5 1 15 −6 −t2
=√ e x − x + x −√ t e dt
π 2 4 π x 4

1128
 2

0.5 1 1.5 2 2.5 3

-2

-4

-6

-8

-10

Figure 26.2: Logarithm of the Approximation to the Complementary Error Function.

The error in approximating erfc (x) with the first three terms is given in Table 26.1. We see that for x ≥ 2 the
three terms give a much better approximation to erfc (x) than just the first term.

At this point you might guess that you could continue this process indefinitely. By repeated application of
integration by parts, you can obtain the series expansion

∞
X (−1)n (2n)!
2 2
erfc (x) = √ e−x .
π n=0
n!(2x)2n+1

1129
 x erfc (x) One Term Relative Error Three Term Relative Error
1 0.157 0.3203 0.6497
2 0.00468 0.1044 0.0182
3 2.21 × 10−5 0.0507 0.0020
4 1.54 × 10−8 0.0296 3.9 · 10−4
5 1.54 × 10−12 0.0192 1.1 · 10−4
6 2.15 × 10−17 0.0135 3.7 · 10−5
7 4.18 × 10−23 0.0100 1.5 · 10−5
8 1.12 × 10−29 0.0077 6.9 · 10−6
9 4.14 × 10−37 0.0061 3.4 · 10−6
10 2.09 × 10−45 0.0049 1.8 · 10−6

Table 26.1:

This is a Taylor expansion about infinity. Let’s find the radius of convergence.

 (−1)n+1 (2(n + 1))! n!(2x)2n+1 

   
 an+1 (x) 
lim   < 1 ⇒ lim   <1
n→∞  an (x)  n→∞ (n + 1)!(2x)2(n+1)+1 (−1)n (2n)! 
 
 (2n + 2)(2n + 1) 
⇒ lim   <1
n→∞ (n + 1)(2x)2 
 
 2(2n + 1) 
⇒ lim  <1
n→∞ (2x)2 
 
1
⇒   = 0
x

Thus we see that our series diverges for all x. Our conventional mathematical sense would tell us that this series
is useless, however we will see that this series is very useful as an asymptotic expansion of erfc (x).

1130
 Say we are working with a convergent series expansion of some function f (x).
∞
X
f (x) = an (x)
n=0

For fixed x = x0 ,

N
X
f (x0 ) − an (x0 ) → 0 as N → ∞.
n=0

P∞
For an asymptotic series we have a quite different behavior. If g(x) is asymptotic to n=0 bn (x) as x → x0 then
for fixed N ,

N
X
g(x) − bn (x)  bN (x) as x → x0 .
0

For the complementary error function,

N
X (−1)n (2n)!
2 2
For fixed N , erfc (x) − √ e−x 2n+1
 x−2N −1 as x → ∞.
π n=0
n!(2x)

We say that the error function is asymptotic to the series as x goes to infinity.
∞
X (−1)n (2n)!
2 2
erfc (x) ∼ √ e−x as x → ∞
π n=0
n!(2x)2n+1

In Figure 26.3 the logarithm of the difference between the one term, ten term and twenty term approximations
and the complementary error function are graphed in coarse, medium, and fine dashed lines, respectively.

1131
 1 2 3 4 5 6

-20

-40

-60

Figure 26.3: log(error in approximation)

*Optimal Asymptotic Series. Of the three approximations, the one term is best for x . 2, the ten term
is best for 2 . x . 4, and the twenty term is best for 4 . x. This leads us to the concept of an optimal
asymptotic approximation. An optimal asymptotic approximation contains the number of terms in the series that
best approximates the true behavior.
In Figure 26.4 we see a plot of the number of terms in the approximation versus the logarithm of the error for
x = 3. Thus we see that the optimal asymptotic approximation is the first nine terms. After nine terms the error
gets larger. It was inevitable that the error would start to grow after some point as the series diverges for all x.
A good rule of thumb for finding the optimal series is to find the smallest term in the series and take all of the
terms up to but not including the smallest term as the optimal approximation. This makes sense, because the
nth term is an approximation of the error incurred by using the first n − 1 terms. In Figure 26.5 there is a plot

1132
 -12

-14

-16

-18

5 10 15 20 25

Figure 26.4: The logarithm of the error in using n terms.

of n versus the logarithm of the nth term in the asymptotic expansion of erfc (3). We see that the tenth term is
the smallest. Thus, in this case, our rule of thumb predicts the actual optimal series.

26.4 Asymptotic Series

P∞
A function f (x) has an asymptotic series expansion about x = x0 , n=0 an (x), if

N
X
f (x) − an (x)  aN (x) as x → x0 for all N.
n=0

1133
 -12

-14

-16

5 10 15 20 25

Figure 26.5: The logarithm of the nth term in the expansion for x = 3.

An asymptotic series may be convergent or divergent. Most of the asymptotic series you encounter will be
divergent. If the series is convergent, then we have that
N
X
f (x) − an (x) → 0 as N → ∞ for fixed x.
n=0

Let n (x) be some set of gauge functions. The example that we are most familiar with is n (x) = xn . If we
say that
∞
X ∞
X
an n (x) ∼ bn n (x),
n=0 n=0

1134
then this means that an = bn .

26.5 Asymptotic Expansions of Differential Equations

26.5.1 The Parabolic Cylinder Equation.
Controlling Factor. Let us examine the behavior of the bounded solution of the parabolic cylinder equation
as x → +∞.
 
00 1 1 2
y + ν+ − x y=0
2 4
This equation has an irregular singular point at infinity. With the substitution y = es , the equation becomes
1 1 2
s00 + (s0 )2 + ν + − x = 0.
2 4
We know that
1 1
ν+  x2 as x → +∞
2 4
so we drop this term from the equation. Let us make the assumption that
s00  (s0 )2 as x → +∞.
Thus we are left with the equation
1 2
(s0 )2 ∼ x
4
1
s0 ∼ ± x
2
1
s ∼ ± x2 + c
4
1
s ∼ ± x2 as x → +∞
4

1135
Now let’s check if our assumption is consistent. Substituting into s00  (s0 )2 yields 1/2  x2 /4 as x → +∞
which is true. Since the equation for y is second order, we would expect that there are two different behaviors
as x → +∞. This is confirmed by the fact that we found two behaviors for s. s ∼ −x2 /4 corresponds to the
2
solution that is bounded at +∞. Thus the controlling factor of the leading behavior is e−x /4 .

Leading Order Behavior. Now we attempt to get a better approximation to s. We make the substitution
s = − 41 x2 + t(x) into the equation for s where t  x2 as x → +∞.

1 1 1 1
− + t00 + x2 − xt0 + (t0 )2 + ν + − x2 = 0
2 4 2 4

t00 − xt0 + (t0 )2 + ν = 0

Since t  x2 , we assume that t0  x and t00  1 as x → +∞. Note that this in only an assumption since it is not
always valid to differentiate an asymptotic relation. Thus (t0 )2  xt0 and t00  xt0 as x → +∞; we drop these
terms from the equation.
ν
t0 ∼
x
t ∼ ν log x + c
t ∼ ν log x as x → +∞

Checking our assumptions for the derivatives of t,

1 1
t0  x ⇒ x t00  1 ⇒  1,
x x2
we see that they were consistent. Now we wish to refine our approximation for t with the substitution t(x) =
ν log x + u(x). So far we have that
 2   2 
x ν x
y ∼ exp − + ν log x + u(x) = x exp − + u(x) as x → +∞.
4 4

1136
We can try and determine u(x) by substituting the expression t(x) = ν log x + u(x) into the equation for t.

ν 00 0 ν 2 2ν 0
− + u − (ν + xu ) + + u + (u0 )2 + ν = 0
x2 x2 x
After suitable simplification, this equation becomes

ν2 − ν
u0 ∼ as x → +∞
x3
Integrating this asymptotic relation,

ν − ν2
u∼ + c as x → +∞.
2x2
2
Notice that ν−ν
2x2
 c as x → +∞; thus this procedure fails to give us the behavior of u(x). Further refinements
to our approximation for s go to a constant value as x → +∞. Thus we have that the leading behavior is
 2
ν x
y ∼ cx exp − as x → +∞
4

Asymptotic Expansion Since we have factored off the singular behavior of y, we might expect that what is
left over is well behaved enough to be expanded in a Taylor series about infinity. Let us assume that we can
expand the solution for y in the form
 2  2X ∞
x x
ν
y(x) ∼ x exp − ν
σ(x) = x exp − an x−n as x → +∞
4 4 n=0
 2
where a0 = 1. Differentiating y = xν exp − x4 σ(x),
 
0 1 ν+1 −x2 /4 2
y = νxν−1
− x e σ(x) + xν e−x /4 σ 0 (x)
2

1137
    
00 ν−2 1 ν 1 ν 1 ν+2 −x2 /4 ν−1 1 ν+1 −x2 /4 0
y = ν(ν − 1)x − νx − (ν + 1)x + x e σ(x) + 2 νx − x e σ (x)
2 2 4 2
2 /4
+ xν e−x σ 00 (x).

Substituting this into the differential equation for y,

     
−2 1 1 2 −1 1 0 00 1 1 2
ν(ν − 1)x − (ν + ) + x σ(x) + 2 νx − x σ (x) + σ (x) + ν + − x σ(x) = 0
2 4 2 2 4
σ 00 (x) + (2νx−1 − x)σ 0 (x) + ν(ν − 1)x−2 σ = 0
x2 σ 00 (x) + (2νx − x3 )σ 0 (x) + ν(ν − 1)σ(x) = 0.

Differentiating the expression for σ(x),

∞
X
σ(x) = an x−n
n=0
∞
X ∞
X
0 −n−1
σ (x) = −nan x = −(n + 2)an+2 x−n−3
n=1 n=−1
X∞
σ 00 (x) = n(n + 1)an x−n−2 .
n=1

Substituting this into the differential equation for σ(x),

∞
X ∞
X ∞
X ∞
X
n(n + 1)an x−n + 2ν −nan x−n − −(n + 2)an+2 x−n + ν(ν − 1) an x−n = 0.
n=1 n=1 n=−1 n=0

Equating the coefficient of x1 to zero yields

a1 x = 0 ⇒ a1 = 0.

1138
Equating the coefficient of x0 ,
1
2a2 + ν(ν − 1)a0 = 0 ⇒ a2 = − ν(ν − 1).
2
From the coefficient of x−n for n > 0,
n(n + 1)an − 2νnan + (n + 2)an+2 + ν(ν − 1)an = 0
(n + 2)an+2 = −[n(n + 1) − 2νn + ν(ν − 1)]an
(n + 2)an+2 = −[n2 + n − 2νn + ν(ν − 1)]an
(n + 2)an+2 = −(n − ν)(n − ν + 1)an .
Thus the recursion formula for the an ’s is
(n − ν)(n − ν + 1)
an+2 = − an , a0 = 1, a1 = 0.
n+2
The first few terms in σ(x) are
ν(ν − 1) −2 ν(ν − 1)(ν − 2)(ν − 3) −4
σ(x) ∼ 1 − x + x − · · · as x → +∞
21 1! 22 2!
If we check the radius of convergence of this series
 an+2 x−n−2 
   
 (n − ν)(n − ν + 1) −2 
lim   < 1 ⇒ lim −  x  < 1
n→∞  an x−n  n→∞ n+2
1
⇒ =0
x
we see that the radius of convergence is zero. Thus if ν 6= 0, 1, 2, . . . our asymptotic expansion for y
 
ν −x2 /4 ν(ν − 1) −2 ν(ν − 1)(ν − 2)(ν − 3) −4
y∼x e 1− x + x − ···
21 1! 22 2!
diverges for all x. However this solution is still very useful. If we only use a finite number of terms, we will get a
very good numerical approximation for large x.
In Figure 26.6 the one term, two term, and three term asymptotic approximations are shown in rough, medium,
and fine dashing, respectively. The numerical solution is plotted in a solid line.

1139
 6

1 2 3 4 5 6

-2

Figure 26.6: Asymptotic Approximations to the Parabolic Cylinder Function.

1140
Chapter 27

Hilbert Spaces

An expert is a man who has made all the mistakes which can be made, in a narrow field.
- Niels Bohr

WARNING: UNDER HEAVY CONSTRUCTION.

In this chapter we will introduce Hilbert spaces. We develop the two important examples: l2 , the space of
square summable infinite vectors and L2 , the space of square integrable functions.

27.1 Linear Spaces

A linear space is a set of elements {x, y, z, . . . } that is closed under addition and scalar multiplication. By closed
under addition we mean: if x and y are elements, then z = x + y is an element. The addition is commutative and
associative.
x+y =y+x
(x + y) + z = x + (y + z)

1141
Scalar multiplication is associative and distributive. Let a and b be scalars, a, b ∈ C.
(ab)x = a(bx)
(a + b)x = ax + bx
a(x + y) = ax + ay
All the linear spaces that we will work with have additional properties: The zero element 0 is the additive
identity.
x+0=x
Multiplication by the scalar 1 is the multiplicative identity.
1x = x
Each element x and the additive inverse, −x.
x + (−x) = 0
Consider a set of elements {x1 , x2 , . . . }. Let the ci be scalars. If
y = c1 x1 + c2 x2 + · · ·
then y is a linear combination of the xi . A set of elements {x1 , x2 , . . . } is linearly independent if the equation
c1 x1 + c2 x2 + · · · = 0
has only the trivial solution c1 = c2 = · · · = 0. Otherwise the set is linearly dependent.
Let {e1 , e2 , · · · } be a linearly independent set of elements. If every element x can be written as a linear
combination of the ei then the set {ei } is a basis for the space. The ei are called base elements.
X
x= ci ei
i

The set {ei } is also called a coordinate system. The scalars ci are the coordinates or components of x. If the set
{ei } is a basis, then we say that the set is complete.

1142
27.2 Inner Products
hx|yi is an inner product of two elements x and y if it satisfies the properties:

1. Conjugate-commutative.

hx|yi = hx|yi

2. Linearity in the second argument.

hx|ay + bzi = ahx|yi + bhx|yi

3. Positive definite.

hx|xi ≥ 0
hx|xi = 0 if and only if x = 0

From these properties one can derive the properties:

1. Conjugate linearity in the first argument.

hax + by|zi = ahx|zi + bhx|zi

2. Schwarz Inequality.

|hx|yi|2 ≤ hx|xihy|yi

One inner product of vectors is the Euclidean inner product.

n
X
hx|yi ≡ x · y = xi y i .
i=0

1143
 One inner product of functions defined on (a . . . b) is
Z b
hu|vi = u(x)v(x) dx.
a

If σ(x) is a positive-valued function, then we can define the inner product:

Z b
hu|σ|vi = u(x)σ(x)v(x) dx.
a

This is called the inner product with respect to the weighting function σ(x). It is also denoted hu|viσ .

27.3 Norms
A norm is a real-valued function on a space which satisfies the following properties.
1. Positive.

kxk ≥ 0

2. Definite.

kxk = 0 if and only if x = 0

3. Multiplication my a scalar, c ∈ C.

kcxk = |c|kxk

4. Triangle inequality.

kx + yk ≤ kxk + kyk

1144
Example 27.3.1 Consider a vector space, (finite or infinite dimension), with elements x = (x1 , x2 , x3 , . . . ). Here
are some common norms.

• Norm generated by the inner product.

p
kxk = hx|xi

• The lp norm.

∞
!1/p
X
kxkp = |xk |p
k=1

There are three common cases of the lp norm.

– Euclidian norm, or l2 norm.

v
u∞
uX
kxk2 = t |xk |2
k=1

– l1 norm.
∞
X
kxk1 = |xk |
k=1

– l∞ norm.

kxk∞ = max |xk |

k

Example 27.3.2 Consider a space of functions defined on the interval (a . . . b). Here are some common norms.

1145
 • Norm generated by the inner product.
p
kuk = hu|ui

• The Lp norm.
Z b 1/p
p
kukp = |u(x)| dx
a

There are three common cases of the Lp norm.

– Euclidian norm, or L2 norm.

s
Z b
kuk2 = |u(x)|2 dx
a

– L1 norm.
Z b
kuk1 = |u(x)| dx
a

– L∞ norm.

kuk∞ = lim sup |u(x)|

x∈(a...b)

Distance. Using the norm, we can define the distance between elements u and v.

d(u, v) ≡ ku − vk

Note that d(u, v) = 0 does not necessarily imply that u = v. CONTINUE.

1146
27.4 Linear Independence.
27.5 Orthogonality
Orthogonality.

hφj |φk i = 0 if j 6= k

Orthonormality.

hφj |φk i = δjk

Example 27.5.1 Infinite vectors. ej has all zeros except for a 1 in the j th position.

ej = (0, 0, . . . 0, 1, 0, . . . )

Example 27.5.2 L2 functions on (0 . . . 2π).

1
φj = √ eijx , j∈Z
2π

1 (1) 1 (1) 1
φ0 = √ , φj = √ cos(jx), φj = √ sin(jx), j ∈ Z+
2π π π

27.6 Gramm-Schmidt Orthogonalization

Let {ψ1 (x), . . . , ψn (x)} be a set of linearly independent functions. Using the Gramm-Schmidt orthogonalization
process we can construct a set of orthogonal functions {φ1 (x), . . . , φn (x)} that has the same span as the set of

1147
ψn ’s with the formulas

φ 1 = ψ1
hφ1 |ψ2 i
φ 2 = ψ2 − φ1
kφ1 k2
hφ1 |ψ3 i hφ2 |ψ3 i
φ 3 = ψ3 − 2
φ1 − φ2
kφ1 k kφ2 k2
···
n−1
X hφj |ψn i
φ n = ψn − φj .
j=1
kφj k2

You could verify that the φm are orthogonal with a proof by induction.

Example 27.6.1 Suppose we would like a polynomial approximation to cos(πx) in the domain [−1, 1]. One way
to do this is to find the Taylor expansion of the function about x = 0. Up to terms of order x4 , this is

(πx)2 (πx)4
cos(πx) = 1 − + + O(x6 ).
2 24

In the first graph of Figure 27.1 cos(πx) and this fourth degree polynomial are plotted. We see that the approx-
imation is very good near x = 0, but deteriorates as we move away from that point. This makes sense because
the Taylor expansion only makes use of information about the function’s behavior at the point x = 0.
As a second approach, we could find the least squares fit of a fourth degree polynomial to cos(πx). The set
of functions {1, x, x2 , x3 , x4 } is independent, but not orthogonal in the interval [−1, 1]. Using Gramm-Schmidt

1148
orthogonalization,
φ0 = 1
h1|xi
φ1 = x − =x
h1|1i
h1|x2 i hx|x2 i 1
φ2 = x2 − − x = x2 −
h1|1i hx|xi 3
3
φ3 = x3 − x
5
6 3
φ4 = x − x2 −
4
7 35
A widely used set of functions in mathematics is the set of Legendre polynomials {P0 (x), P1 (x), . . . }. They
differ from the φn ’s that we generated only by constant factors. The first few are
P0 (x) = 1
P1 (x) = x
3x2 − 1
P2 (x) =
2
3
5x − 3x
P3 (x) =
2
35x4 − 30x2 + 3
P4 (x) = .
8
Expanding cos(πx) in Legendre polynomials
4
X
cos(πx) ≈ cn Pn (x),
n=0

and calculating the generalized Fourier coefficients with the formula

hPn | cos(πx)i
cn = ,
hPn |Pn i

1149
yields

15 45(2π 2 − 21)
cos(πx) ≈ − P2 (x) + P4 (x)
π2 π4
105
= 4 [(315 − 30π 2 )x4 + (24π 2 − 270)x2 + (27 − 2π 2 )]
8π

The cosine and this polynomial are plotted in the second graph in Figure 27.1. The least squares fit method uses
information about the function on the entire interval. We see that the least squares fit does not give as good
an approximation close to the point x = 0 as the Taylor expansion. However, the least squares fit gives a good
approximation on the entire interval.
In order to expand a function in a Taylor series, the function must be analytic in some domain. One advantage
of using the method of least squares is that the function being approximated does not even have to be continuous.

1 1

0.5 0.5

-1 -0.5 0.5 1 -1 -0.5 0.5 1

-0.5 -0.5

-1 -1

Figure 27.1: Polynomial Approximations to cos(πx).

1150
27.7 Orthonormal Function Expansion
Let {φj } be an orthonormal set of functions on the interval (a, b). We expand a function f (x) in the φj .
X
f (x) = cj φj
j

We choose the coefficients to minimize the norm of the error.

2 *  +
X X  X
f − cj φj = f − cj φj f − cj φj


j j j
*  + *  + *X X +
 X X  
= kf k2 − f  cj φj − cj φj f + cj φj  cj φj
j j j j
X X X
2 2
= kf k + |cj | − cj hf |φj i − cj hφj |f i
j j j

2
X X X X
2
f − c j j = kf k +
φ |cj |2 − cj hφj |f i − cj hφj |f i (27.1)



j j j j

P
To complete the square, we add the constant j hφj |f ihφj |f i. We see the values of cj which minimize
X
kf k2 + |cj − hφj |f i|2 .
j

Clearly the unique minimum occurs for

cj = hφj |f i.

1151
We substitute this value for cj into the right side of Equation 27.1 and note that this quantity, the squared norm
of the error, is non-negative.
X X X
kf k2 + |cj |2 − |cj |2 − |cj |2 ≥ 0
j j j
X
kf k2 ≥ |cj |2
j

This is known as Bessel’s Inequality. If the set of {φj } is complete then the norm of the error is zero and we
obtain Bessel’s Equality.
X
kf k2 = |cj |2
j

27.8 Sets Of Functions

Orthogonality. Consider two complex valued functions of a real variable φ1 (x) and φ2 (x) defined on the interval
a ≤ x ≤ b. The inner product of the two functions is defined
Z b
hφ1 |φ2 i = φ1 (x)φ2 (x) dx.
a
p
The two functions are orthogonal if hφ1 |φ2 i = 0. The L2 norm of a function is defined kφk = hφ|φi.
Let {φ1 , φ2 , φ3 , . . . } be a set of complex valued functions. The set of functions is orthogonal if each pair of
functions is orthogonal. That is,
hφn |φm i = 0 if n 6= m.
If in addition the norm of each function is 1, then the set is orthonormal. That is,
(
1 if n = m
hφn |φm i = δnm =
0 if n 6= m.

1152
Example 27.8.1 The set of functions
(r r r )
2 2 2
sin(x), sin(2x), sin(3x), . . .
π π π
is orthonormal on the interval [0, π]. To verify this,
*r r +
2 π 2
Z
2  2

sin(nx) sin(nx) = sin (nx) dx
π  π π 0
=1
If n 6= m then
*r r +
2 π
Z
2  2
sin(nx) sin(mx) = sin(nx) sin(mx) dx

π  π π 0
1 π
Z
= (cos[(n − m)x] − cos[(n + m)x]) dx
π 0
= 0.

Example 27.8.2 The set of functions

1 1 1 1
{. . . , √ e−ix , √ , √ eix , √ e2ix , . . . },
2π 2π 2π 2π
is orthonormal on the interval [−π, π]. To verify this,
*  + Z π
1 inx  1 inx 1
√ e √ e = e−inx einx dx
2π  2π 2π −π
Z π
1
= dx
2π −π
= 1.

1153
If n 6= m then
*  + Z π
1 inx  1 imx 1
√ e √ e = e−inx eimx dx
2π  2π 2π −π
Z π
1
= ei(m−n)x dx
2π −π
= 0.

Orthogonal with Respect to a Weighting Function. Let σ(x) be a real-valued, positive function on the
interval [a, b]. We introduce the notation
Z b
hφn |σ|φm i ≡ φn σφm dx.
a

If the set of functions {φ1 , φ2 , φ3 , . . . } satisfy

hφn |σ|φm i = 0 if n 6= m

then the functions are orthogonal with respect to the weighting function σ(x).
If the functions satisfy

hφn |σ|φm i = δnm

then the set is orthonormal with respect to σ(x).

Example 27.8.3 We know that the set of functions

(r r r )
2 2 2
sin(x), sin(2x), sin(3x), . . .
π π π

1154
is orthonormal on the interval [0, π]. That is,
Z πr r
2 2
sin(nx) sin(mx) dx = δnm .
0 π π
√
If we make the change of variables x = t in this integral, we obtain
Z π2 r
√
r
1 2 2 √
√ sin(n t) sin(m t) dt = δnm .
0 2 t π π
Thus the set of functions
(r )
√
r
√
r
1 1 1 √
sin( t), sin(2 t), sin(3 t), . . .
π π π
1
is orthonormal with respect to σ(t) = √
2 t
on the interval [0, π 2 ].

Orthogonal Series. Suppose that a function f (x) defined on [a, b] can be written as a uniformly convergent
sum of functions that are orthogonal with respect to σ(x).
X∞
f (x) = cn φn (x)
n=1

We can solve for the cn by taking the inner product of φm (x) and each side of the equation with respect to σ(x).
*   ∞ +
 X
hφm |σ|f i = φm σ  c φ
 
  n=1 n n
∞
X
hφm |σ|f i = cn hφm |σ|φn i
n=1
hφm |σ|f i = cm hφm |σ|φm i
hφm |σ|f i
cm =
hφm |σ|φm i

1155
The cm are known as Generalized Fourier coefficients. If the functions in the expansion are orthonormal, the
formula simplifies to

cm = hφm |σ|f i.

Example 27.8.4 The function f (x) = x(π − x) has a uniformly convergent series expansion in the domain [0, π]
of the form
∞ r
X 2
x(π − x) = cn sin(nx).
n=1
π

The Fourier coefficients are

*r  +
2 
cn = sin(nx)x(π − x)

π 
r Z π
2
= x(π − x) sin(nx) dx
π 0
r
2 2
= (1 − (−1)n )
π n3
(q
2 4
π n3
for odd n
=
0 for even n

Thus the expansion is

∞
X 8
x(π − x) = sin(nx) for x ∈ [0, π].
n=1
πn3
oddn

In the first graph of Figure 27.2 the first term in the expansion is plotted in a dashed line and x(π − x) is
plotted in a solid line. The second graph shows the two term approximation.

1156
 2 2

1 1

1 2 3 1 2 3

Figure 27.2: Series Expansions of x(π − x).

√ √ √ √
Example 27.8.5 The set {. . . , 1/ 2π e−ix , 1/ 2π, 1/ 2π eix , 1/ 2π e2ix , . . . } is orthonormal on the interval
[−π, π]. f (x) = sign (x) has the expansion


∞
* +
X 1  1
sign (x) ∼ √ einξ  sign (ξ) √ einx

n=−∞
2π  2π
∞ Z π
1 X
= e−inξ sign (ξ) dξ einx
2π n=−∞ −π
∞ Z 0 Z π 
1 X −inξ −inξ
= −e dξ + e dξ einx
2π n=−∞ −π 0
∞
1 X 1 − (−1)n inx
= e .
π n=−∞ in

1157
In terms of real functions, this is
∞
1 X 1 − (−1)n
= (cos(nx) + i sin(nx))
π n=−∞ in
∞
2 X 1 − (−1)n
= sin(nx)
π n=1 in

∞
4X1
sign (x) ∼ sin(nx).
π n=1 n
oddn

27.9 Least Squares Fit to a Function and Completeness

Let {φ1 , φ2 , φ3 , . . . } be a set of real, square integrable functions that are orthonormal with respect to the weighting
function σ(x) on the interval [a, b]. That is,

hφn |σ|φm i = δnm .

Let f (x) be some square integrable function defined on the same interval. We would like to approximate the
function f (x) with a finite orthonormal series.
N
X
f (x) ≈ αn φn (x)
n=1

f (x) may or may not have a uniformly convergent expansion in the orthonormal functions.
We would like to choose the αn so that we get the best possible approximation to f (x). The most common
measure of how well a series approximates a function is the least squares measure. The error is defined as the

1158
integral of the weighting function times the square of the deviation.
Z b N
!2
X
E= σ(x) f (x) − αn φn (x) dx
a n=1

The “best” fit is found by choosing the αn that minimize E. Let cn be the Fourier coefficients of f (x).
cn = hφn |σ|f i
we expand the integral for E.
N
!2
Z b X
E(α) = σ(x) f (x) − αn φn (x) dx
a n=1
 N
X   N
X 
 
= f− αn φn  σ f−
 αn φn
n=1 n=1
N N
  N
X    X  X 
 
= hf |σ|f i − 2 αn φn σ f + αn φn σ  αn φn
n=1 n=1 n=1
N
X N
XX N
= hf |σ|f i − 2 αn hφn |σ|f i + αn αm hφn |σ|φm i
n=1 n=1 m=1
XN N
X
= hf |σ|f i − 2 αn cn + αn2
n=1 n=1
N
X N
X
2
= hf |σ|f i + (αn − cn ) − c2n
n=1 n=1
Each term involving αn in non-negative and is minimized for αn = cn . The Fourier coefficients give the least
squares approximation to a function. The least squares fit to f (x) is thus
N
X
f (x) ≈ hφn |σ|f i φn (x).
n=1

1159
 Result 27.9.1 If {φ1 , φ2 , φ3 , . . . } is a set of real, square integrable functions that are
orthogonal with respect to σ(x) then the least squares fit of the first N orthogonal
functions to the square integrable function f (x) is
N
X hφn |σ|f i
f (x) ≈ φn (x).
n=1
hφ n |σ|φ n i

If the set is orthonormal, this formula reduces to

N
X
f (x) ≈ hφn |σ|f i φn (x).
n=1

Since the error in the approximation E is a nonnegative number we can obtain on inequality on the sum of
the squared coefficients.
N
X
E = hf |σ|f i − c2n
n=1
N
X
c2n ≤ hf |σ|f i
n=1

P is2 known as Bessel’s Inequality. Since hf |σ|f i is just a nonnegative number, independent of N ,
This equation
the sum ∞ n=1 cn is convergent and cn → 0 as n → ∞

Convergence in the Mean. If the error E goes to zero as N tends to infinity

Z b N
!2
X
lim σ(x) f (x) − cn φn (x) dx = 0,
N →∞ a n=1

1160
then the sum converges in the mean to f (x) relative to the weighting function σ(x). This implies that

N
!
X
lim hf |σ|f i − c2n =0
N →∞
n=1
∞
X
c2n = hf |σ|f i.
n=1

This is known as Parseval’s identity.

Completeness. Consider a set of functions {φ1 , φ2 , φ3 , . . . } that is orthogonal with respect to the weighting
function σ(x). If every function f (x) that is square integrable with respect to σ(x) has an orthogonal series
expansion

∞
X
f (x) ∼ cn φn (x)
n=1

that converges in the mean to f (x), then the set is complete.

27.10 Closure Relation

Let {φ1 , φ2 , . . . } be an orthonormal, complete set on the domain [a, b]. For any square integrable function f (x)
we can write
∞
X
f (x) ∼ cn φn (x).
n=1

1161
Here the cn are the generalized Fourier coefficients and the sum converges in the mean to f (x). Substituting the
expression for the Fourier coefficients into the sum yields
∞
X
f (x) ∼ hφn |f iφn (x)
n=1
X∞ Z b 
= φn (ξ)f (ξ) dξ φn (x).
n=1 a

Since the sum is not necessarily uniformly convergent, we are not justified in exchanging the order of summation
and integration . . . but what the heck, let’s do it anyway.

∞
!
Z b X
= φn (ξ)f (ξ)φn (x) dξ
a n=1
∞
!
Z b X
= φn (ξ)φn (x) f (ξ) dξ
a n=1

The sum behaves like a Dirac delta function. Recall that δ(x − ξ) satisfies the equation
Z b
f (x) = δ(x − ξ)f (ξ) dξ for x ∈ (a, b).
a

Thus we could say that the sum is a representation of δ(x − ξ). Note that a series representation of the delta
function could not be convergent, hence the necessity of throwing caution to the wind when we interchanged the
summation and integration in deriving the series. The closure relation for an orthonormal, complete set states
∞
X
φn (x)φn (ξ) ∼ δ(x − ξ).
n=1

1162
 Alternatively, you can derive the closure relation by computing the generalized Fourier coefficients of the delta
function.
X∞
δ(x − ξ) ∼ cn φn (x)
n=1

cn = hφn |δ(x − ξ)i

Z b
= φn (x)δ(x − ξ) dx
a
= φn (ξ)
∞
X
δ(x − ξ) ∼ φn (x)φn (ξ)
n=1

Result 27.10.1 If {φ1 , φ2 , . . . } is an orthogonal, complete set on the domain [a, b], then
∞
X φn (x)φn (ξ)
∼ δ(x − ξ).
n=1
kφn k2

If the set is orthonormal, then

∞
X
φn (x)φn (ξ) ∼ δ(x − ξ).
n=1

Example 27.10.1 The integral of the Dirac delta function is the Heaviside function. On the interval x ∈ (−π, π)
Z x (
1 for 0 < x < π
δ(t) dt = H(x) =
−π 0 for − π < x < 0.

1163
 Consider the orthonormal, complete set {. . . , √12π e−ix , √12π , √12π eix , . . . } on the domain [−π, π]. The delta
function has the series

∞ ∞
X 1 int 1 −in0 1 X int
δ(t) ∼ √ e √ e = e .
n=−∞
2π 2π 2π n=−∞

We will find the series expansion of the Heaviside function first by expanding directly and then by integrating
the expansion for the delta function.

Finding the series expansion of H(x) directly. The generalized Fourier coefficients of H(x) are

Z π
1
c0 = √ H(x) dx
−π 2π
Z π
1
=√ dx
2π 0
r
π
=
2
Z π
1
cn = √ e−inx H(x) dx
−π 2π
Z π
1
=√ e−inx dx
2π 0
1 − (−1)n
= √ .
in 2π

1164
Thus the Heaviside function has the expansion
r ∞
π 1 X 1 − (−1)n 1 inx
H(x) ∼ √ + √ √ e
2 2π n=−∞ in 2π 2π
n6=0
∞
1 1 X 1 − (−1)n
= + sin(nx)
2 π n=1
n

∞
1 2X1
H(x) ∼ + sin(nx).
2 π n=1 n
oddn

Integrating the series for δ(t).

Z x Z x X ∞
1
δ(t) dt ∼ eint dt
−π 2π −π n=−∞
 
∞  x
1  X 1
= (x + π) + eint

2π in

n=−∞ −π
n6=0
 
∞
1  X 1 inx
e − (−1)n 


= (x + π) +
2π n=−∞
in
n6=0
∞
x 1 1 X 1 inx
e − e−inx − (−1)n + (−1)n


= + +
2π 2 2π n=1
in
∞
x 1 1X1
= + + sin(nx)
2π 2 π n=1 n

1165
 x
Expanding 2π
in the orthonormal set,
∞
x X 1
∼ cn √ einx .
2π n=−∞ 2π

Z π
1 x
c0 = √ dx = 0
2π 2π
Z−π
π
1 x i(−1)n
cn = √ e−inx dx = √
−π 2π 2π n 2π

∞ ∞
x X i(−1)n 1 inx 1X
∼ √ √ e =− (−1)n sin(nx)
2π n=−∞ n 2π 2π π n=1
n6=0
x
Substituting the series for 2π
into the expression for the integral of the delta function,
Z x ∞
1 1 X 1 − (−1)n
δ(t) dt ∼ + sin(nx)
−π 2 π n=1 n

Z x ∞
1 2X1
δ(t) dt ∼ + sin(nx).
−π 2 π n=1 n
oddn

Thus we see that the series expansions of the Heaviside function and the integral of the delta function are the
same.

27.11 Linear Operators

1166
27.12 Exercises
Exercise 27.1
1. Suppose {φk (x)}∞
k=0 is an orthogonal system on [a, b]. Show that any finite set of the φj (x) is a linearly
independent set on [a, b]. That is, if {φj1 (x), φj2 (x), . . . , φjn (x)} is the set and all the jν are distinct, then

a1 φj1 (x) + a2 φj2 (x) + · · · + an φjn (x) = 0 on a ≤ x ≤ b

is true iff: a1 = a2 = · · · = an = 0.
RL
2. Show that the complex functions φk (x) ≡ eikπx/L , k = 0, 1, 2, . . . are orthogonal in the sense that −L
φk (x)φ∗n (x) d
0, for n 6= k. Here φ∗n (x) is the complex conjugate of φn (x).
Hint, Solution

1167
27.13 Hints
Hint 27.1

1168
27.14 Solutions
Solution 27.1
1.

a1 φj1 (x) + a2 φj2 (x) + · · · + an φjn (x) = 0

X n
ak φjk (x) = 0
k=1

Rb
We take the inner product with φjν for any ν = 1, . . . , n. (hφ, ψi ≡ a
φ(x)ψ ∗ (x) dx.)
* n
+
X
ak φjk , φjν =0
k=1

We interchange the order of summation and integration.

n
X
ak hφjk , φjν i = 0
k=1

hφjk φjν i = 0 for j 6= ν.

aν hφjν φjν i = 0

hφjν φjν i 6= 0.

aν = 0

Thus we see that a1 = a2 = · · · = an = 0.

1169
2. For k 6= n, hφk , φn i = 0.
Z L
hφk , φn i ≡ φk (x)φ∗n (x) dx
−L
Z L
= eikπx/L e−inπx/L dx
−L
Z L
= ei(k−n)πx/L dx
−L
L
ei(k−n)πx/L

=
i(k − n)π/L −L
i(k−n)π −i(k−n)π
e −e
=
i(k − n)π/L
2L sin((k − n)π)
=
(k − n)π
=0

1170
Chapter 28

Self Adjoint Linear Operators

28.1 Adjoint Operators

The adjoint of an operator, L∗ , satisfies

hv|Lui − hL∗ v|ui = 0

for all elements u an v. This is known as Green’s Identity.

The adjoint of a matrix. For vectors, one can represent linear operators L with matrix multiplication.

Lx ≡ Ax

1171
Let B = A∗ be the adjoint of the matrix A. We determine the adjoint of A from Green’s Identity.

hx|Ayi − hBx|yi = 0
x · Ay = Bx · y
T
xT Ay = Bx y
T
xT Ay = xT B y
T T
yT A x = yT BxB = A

T
Thus we see that the adjoint of a matrix is the conjugate transpose of the matrix, A∗ = A . The conjugate
transpose is also called the Hermitian transpose and is denoted AH .

The adjoint of a differential operator. Consider a second order linear differential operator acting on C 2
functions defined on (a . . . b) which satisfy certain boundary conditions.

Lu ≡ p2 (x)u00 + p1 (x)u0 + p0 (x)u

28.2 Self-Adjoint Operators

T
Matrices. A matrix is self-adjoint if it is equal to its conjugate transpose A = AH ≡ A . Such matrices are
called Hermitian. For a Hermitian matrix H, Green’s identity is

hy|Hxi = hHy|xi
y · Hx = Hy · x

1172
The eigenvalues of a Hermitian matrix are real. Let x be an eigenvector with eigenvalue λ.

hx|Hxi = hHx|xi
hx|λxi − hλx|xi = 0
(λ − λ)hx|xi = 0
λ=λ

The eigenvectors corresponding to distinct eigenvalues are distinct. Let x and y be eigenvectors with distinct
eigenvalues λ and µ.

hy|Hxi = hHy|xi
hy|λxi − hµy|xi = 0
(λ − µ)hy|λxi = 0
(λ − µ)hy|xi = 0
hy|xi = 0

Furthermore, all Hermitian matrices are similar to a diagonal matrix and have a complete set of orthogonal
eigenvectors.

Trigonometric Series. Consider the problem

−y 00 = λy, y(0) = y(2π), y 0 (0) = y 0 (2π).

1173
 d2
We verify that the differential operator L = − dx 2 with periodic boundary conditions is self-adjoint.

hv|Lui = hv| − u00 i

2π
= [−vu0 ]0 − hv 0 | − u0 i
= hv 0 |u0 i
 2π
= v 0 u 0 − hv 00 |ui
= h−v 00 |ui
= hLv|ui

The eigenvalues and eigenfunctions of this problem are

λ0 = 0, φ0 = 1
λn = n 2 , φ(1)
n = cos(nx), φ(2)
n = sin(nx), n ∈ Z+

1174
28.3 Exercises

1175
28.4 Hints

1176
28.5 Solutions

1177
Chapter 29

Self-Adjoint Boundary Value Problems

Seize the day and throttle it.

-Calvin

29.1 Summary of Adjoint Operators

The adjoint of the operator
dn y dn−1 y
L[y] = pn + p n−1 + · · · + p0 y,
dxn dxn−1
is defined
dn n−1 d
n−1
L∗ [y] = (−1)n (p n y) + (−1) (pn−1 y) + · · · + p0 y
dxn dxn−1
If each of the pk is k times continuously differentiable and u and v are n times continuously differentiable on
some interval, then on that interval Lagrange’s identity states
d
vL[u] − uL∗ [v] = B[u, v]
dx

1178
where B[u, v] is the bilinear form
n
X X
B[u, v] = (−1)j u(k) (pm v)(j) .
m=1 j+k=m−1
j≥0,k≥0

If L is a second order operator then

vL[u] − uL∗ [v] = u00 p2 v + u0 p1 v + u − p2 v 00 + (−2p02 + p1 )v 0 + (−p002 + p01 )v .

 

Integrating Lagrange’s identity on its interval of validity gives us Green’s formula.

Z b 
vL[u] − uL∗ [v] dx = hv|L[u]i − hL∗ [v]|ui = B[u, v]x=b − B[u, v]x=a
 
a

29.2 Formally Self-Adjoint Operators

Example 29.2.1 The linear operator

L[y] = x2 y 00 + 2xy 0 + 3y

has the adjoint operator

d2 2 d
L∗ [y] = (x y) − (2xy) + 3y
dx2 dx
= x2 y 00 + 4xy 0 + 2y − 2xy 0 − 2y + 3y
= x2 y 00 + 2xy 0 + 3y.

In Example 29.2.1, the adjoint operator is the same as the operator. If L = L∗ , the operator is said to be
formally self-adjoint.

1179
 Most of the differential equations that we study in this book are second order, formally self-adjoint, with
real-valued coefficient functions. Thus we wish to find the general form of this operator. Consider the operator
L[y] = p2 y 00 + p1 y 0 + p0 y,
where the pj ’s are real-valued functions. The adjoint operator then is
d2 d
L∗ [y] = 2
(p2 y) − (p1 y) + p0 y
dx dx
= p2 y 00 + 2p02 y 0 + p002 y − p1 y 0 − p01 y + p0 y
= p2 y 00 + (2p02 − p1 )y 0 + (p002 − p01 + p0 )y.
Equating L and L∗ yields the two equations,
2p02 − p1 = p1 , p002 − p01 + p0 = p0
p02 = p1 , p002 = p01 .
Thus second order, formally self-adjoint operators with real-valued coefficient functions have the form
L[y] = p2 y 00 + p02 y 0 + p0 y,
which is equivalent to the form
d
L[y] = (py 0 ) + qy.
dx

Any linear differential equation of the form

L[y] = y 00 + p1 y 0 + p0 y = f (x),
where each pj is j times continuously differentiable and real-valued, can be written as a formally self adjoint
equation. We just multiply by the factor,
Z x
P (x)
e = exp( p1 (ξ) dξ)

1180
to obtain

exp [P (x)] (y 00 + p1 y 0 + p0 y) = exp [P (x)] f (x)

d
(exp [P (x)] y 0 ) + exp [P (x)] p0 y = exp [P (x)] f (x).
dx

Example 29.2.2 Consider the equation

1
y 00 + y 0 + y = 0.
x

Multiplying by the factor

Z x 
1
exp dξ = elog x = x
ξ

will make the equation formally self-adjoint.

xy 00 + y 0 + xy = 0
d
(xy 0 ) + xy = 0
dx

1181
 Result 29.2.1 If L = L∗ then the linear operator L is formally self-adjoint. Second
order formally self-adjoint operators have the form
d
L[y] = (py 0 ) + qy.
dx
Any differential equation of the form

L[y] = y 00 + p1 y 0 + p0 y = f (x),

where each pj is j times continuously differentiable and real-valued, can be Rwritten as a

x
formally self adjoint equation by multiplying the equation by the factor exp( p1 (ξ) dξ).

29.3 Self-Adjoint Problems

Consider the nth order formally self-adjoint equation L[y] = 0, on the domain a ≤ x ≤ b subject to the boundary
conditions, Bj [y] = 0 for j = 1, . . . , n. where the boundary conditions can be written
n
X
Bj [y] = αjk y (k−1) (a) + βjk y (k−1) (b) = 0.
k=1

If the boundary conditions are such that Green’s formula reduces to

hv|L[u]i − hL[v]|ui = 0

then the problem is self-adjoint

Example 29.3.1 Consider the formally self-adjoint equation −y 00 = 0, subject to the boundary conditions y(0) =

1182
y(π) = 0. Green’s formula is

hv| − u00 i − h−v 00 |ui = [u0 (−v) − u(−v)0 ]π0

= [uv 0 − u0 v]π0
= 0.

Thus this problem is self-adjoint.

29.4 Self-Adjoint Eigenvalue Problems

Associated with the self-adjoint problem

L[y] = 0, subject to Bj [y] = 0,

is the eigenvalue problem

L[y] = λy, subject to Bj [y] = 0.

This is called a self-adjoint eigenvalue problem. The values of λ for which there exist nontrivial solutions to
this problem are called eigenvalues. The functions that satisfy the equation when λ is an eigenvalue are called
eigenfunctions.

Example 29.4.1 Consider the self-adjoint eigenvalue problem

−y 00 = λy, subject to y(0) = y(π) = 0.

First consider the case λ = 0. The general solution is

y = c1 + c2 x.

1183
Only the trivial solution satisfies the boundary conditions. λ = 0 is not an eigenvalue. Now consider λ 6= 0. The
general solution is
√  √ 
y = c1 cos λx + c2 sin λx .

The solution that satisfies the left boundary condition is

√ 
y = c sin λx .

For non-trivial solutions, we must have

√ 
sin λπ = 0,

λ = n2 , n ∈ N.

Thus the eigenvalues λn and eigenfunctions φn are

λn = n 2 , φn = sin(nx), for n = 1, 2, 3, . . .

Self-adjoint eigenvalue problems have a number a interesting properties. We will devote the rest of this section
to developing some of these properties.

Real Eigenvalues. The eigenvalues of a self-adjoint problem are real. Let λ be an eigenvalue with the eigen-
function φ. Green’s formula states

hφ|L[φ]i − hL[φ]|φi = 0
hφ|λφi − hλφ|φi = 0
(λ − λ)hφ|φi = 0

Since φ 6≡ 0, hφ|φi > 0. Thus λ = λ and λ is real.

1184
Orthogonal Eigenfunctions. The eigenfunctions corresponding to distinct eigenvalues are orthogonal. Let λn
and λm be distinct eigenvalues with the eigenfunctions φn and φm . Using Green’s formula,
hφn |L[φm ]i − hL[φn ]|φm i = 0
hφn |λm φm i − hλn φn |φm i = 0
(λm − λn )hφn |φm i = 0.

Since the eigenvalues are real,

(λm − λn )hφn |φm i = 0.

Since the two eigenvalues are distinct, hφn |φm i = 0 and thus φn and φm are orthogonal.

*Enumerable Set of Eigenvalues. The eigenvalues of a self-adjoint eigenvalue problem form an enumerable
set with no finite cluster point. Consider the problem
L[y] = λy on a ≤ x ≤ b, subject to Bj [y] = 0.
Let {ψ1 , ψ2 , . . . , ψn } be a fundamental set of solutions at x = x0 for some a ≤ x0 ≤ b. That is,
(k−1)
ψj (x0 ) = δjk .
The key to showing that the eigenvalues are enumerable, is that the ψj are entire functions of λ. That is, they
are analytic functions of λ for all finite λ. We will not prove this.
The boundary conditions are
n
X
αjk y (k−1) (a) + βjk y (k−1) (b) = 0.
 
Bj [y] =
k=1

The eigenvalue problem has a solution for a given value of λ if y = nk=1 ck ψk satisfies the boundary conditions.
P
That is,
" n # n
X X
Bj c k ψk = ck Bj [ψk ] = 0 for j = 1, . . . , n.
k=1 k=1

1185
 Define an n × n matrix M such that Mjk = Bk [ψj ]. Then if ~c = (c1 , c2 , . . . , cn ), the boundary conditions can
be written in terms of the matrix equation M~c = 0. This equation has a solution if and only if the determinant of
the matrix is zero. Since the ψj are entire functions of λ, ∆[M ] is an entire function of λ. The eigenvalues are real,
so ∆[M ] has only real roots. Since ∆[M ] is an entire function, (that is not identically zero), with only real roots,
the roots of ∆[M ] can only cluster at infinity. Thus the eigenvalues of a self-adjoint problem are enumerable and
can only cluster at infinity.
An example of a function whose roots have a finite cluster point is sin(1/x). This function, (graphed in
Figure 29.1), is clearly not analytic at the cluster point x = 0.

-1 1

Figure 29.1: Graph of sin(1/x).

1186
Infinite Number of Eigenvalues. Though we will not show it, self-adjoint problems have an infinite number
of eigenvalues. Thus the eigenfunctions form an infinite orthogonal set.

Eigenvalues of Second Order Problems. Consider the second order, self-adjoint eigenvalue problem

L[y] = (py 0 )0 + qy = λy, on a ≤ x ≤ b, subject to Bj [y] = 0.

Let λn be an eigenvalue with the eigenfunction φn .

hφn |L[φn ]i = hφn |λn φn i

hφn |(pφ0n )0 + qφn i = λn hφn |φn i
Z b
φn (pφ0n )0 dx + hφn |q|φn i = λn hφn |φn i
a
Z b
0 b 0
φn pφ0n dx + hφn |q|φn i = λn hφn |φn i
 
φn pφn a −
a

[pφn φ0n ]ba − hφ0n |p|φ0n i + hφn |q|φn i

λn =
hφn |φn i

Thus we can express each eigenvalue in terms of its eigenfunction. You might think that this formula is just
a shade less than worthless. When solving an eigenvalue problem you have to find the eigenvalues before you
determine the eigenfunctions. Thus this formula could not be used to compute the eigenvalues. However, we can
often use the formula to obtain information about the eigenvalues before we solve a problem.

Example 29.4.2 Consider the self-adjoint eigenvalue problem

−y 00 = λy, y(0) = y(π) = 0.

1187
The eigenvalues are given by the formula
b
(−1)φφ0 a − hφ0n |(−1)|φ0n i + hφn |0|φn i

λn =
hφn |φn i
0 0
0 + hφn |φn i + 0
= .
hφn |φn i
We see that λn ≥ 0. If λn = 0 then hφ0n |φ0n i = 0,which implies that φn = const. The only constant that satisfies the
boundary conditions is φn = 0 which is not an eigenfunction since it is the trivial solution. Thus the eigenvalues
are positive.

29.5 Inhomogeneous Equations

Let the problem,
L[y] = 0, Bk [y] = 0,
be self-adjoint. If the inhomogeneous problem,
L[y] = f, Bk [y] = 0,
has a solution, then we we can write this solution in terms of the eigenfunction of the associated eigenvalue
problem,
L[y] = λy, Bk [y] = 0.
We denote the eigenvalues as λn and the eigenfunctions as φn for n ∈ Z+ . For the moment we assume that
λ = 0 is not an eigenvalue and that the eigenfunctions are real-valued. We expand the function f (x) in a series
of the eigenfunctions.
X hφn |f i
f (x) = fn φn (x), fn =
kφn k

1188
We expand the inhomogeneous solution in a series of eigenfunctions and substitute it into the differential equation.

L[y] = f
hX i X
L yn φn (x) = fn φn (x)
X X
λn yn φn (x) = fn φn (x)
fn
yn =
λn

The inhomogeneous solution is

X hφn |f i
y(x) = φn (x). (29.1)
λn kφn k

As a special case we consider the Green function problem,

L[G] = δ(x − ξ), Bk [G] = 0,

We expand the Dirac delta function in an eigenfunction series.

X hφn |δi X φn (ξ)φn (x)

δ(x − ξ) = φn (x) =
kφn k kφn k

The Green function is

X φn (ξ)φn (x)
G(x|ξ) = .
λn kφn k

1189
We corroborate Equation 29.1 by solving the inhomogeneous equation in terms of the Green function.
Z b
y= G(x|ξ)f (ξ) dξ
a
Z bX
φn (ξ)φn (x)
y= f (ξ) dξ
a λn kφn k
X b φn (ξ)f (ξ) dξ
R
a
y= φn (x)
λn kφn k
X hφn |f i
y= φn (x)
λn kφn k

Example 29.5.1 Consider the Green function problem

G00 + G = δ(x − ξ), G(0|ξ) = G(1|ξ) = 0.

First we examine the associated eigenvalue problem.

φ00 + φ = λφ, φ(0) = φ(1) = 0

φ00 + (1 − λ)φ = 0, φ(0) = φ(1) = 0
λn = 1 − (nπ)2 , φn = sin(nπx), n ∈ Z+

We write the Green function as a series of the eigenfunctions.

∞
X sin(nπξ) sin(nπx)
G(x|ξ) = 2
n=1
1 − (nπ)2

1190
29.6 Exercises
Exercise 29.1
Show that the operator adjoint to

Ly = y (n) + p1 (z)y (n−1) + p2 (z)y (n−2) + · · · + pn (z)y

is given by

M y = (−1)n u(n) + (−1)n−1 (p1 (z)u)(n−1) + (−1)n−2 (p2 (z)u)(n−2) + · · · + pn (z)u.

Hint, Solution

1191
29.7 Hints
Hint 29.1

1192
29.8 Solutions
Solution 29.1
Consider u(x), v(x) ∈ C n . (C n is the set of n times continuously differentiable functions). First we prove the
preliminary result

n−1
(n) n (n) d X
uv − (−1) u v= (−1)k u(k) v (n−k−1) (29.2)
dx k=0

by simplifying the right side.

n−1 n−1
d X X
(−1)k u(k) v (n−k−1) = (−1)k u(k) v (n−k) + u(k+1) v (n−k−1)


dx k=0 k=0
n−1
X n−1
X
k (k) (n−k)
= (−1) u v − (−1)k+1 u(k+1) v (n−k−1)
k=0 k=0
n−1
X n
X
k (k) (n−k)
= (−1) u v − (−1)k u(k) v (n−k)
k=0 k=1
0 (0) n−0 n (n) (n−n)
= (−1) u v − (−1) u v
(n) n (n)
= uv − (−1) u v

We define p0 (x) = 1 so that we can write the operators in a nice form.

n
X n
X
(n−m)
Ly = pm (z)y , Mu = (−1)m (pm (z)u)(n−m)
m=0 m=0

1193
Now we show that M is the adjoint to L.
n
X n
X
(n−m)
uLy − yM u = u pm (z)y −y (−1)m (pm (z)u)(n−m)
m=0 m=0
n
X
upm (z)y (n−m) − (pm (z)u)(n−m) y


=
m=0

We use Equation 29.2.

n n−m−1
X d X
= (−1)k (upm (z))(k) y (n−m−k−1)
m=0
dz k=0

n n−m−1
d X X
uLy − yM u = (−1)k (upm (z))(k) y (n−m−k−1)
dz m=0 k=0

1194
Chapter 30

Fourier Series

Every time I close my eyes

The noise inside me amplifies
I can’t escape
I relive every moment of the day
Every misstep I have made
Finds a way it can invade
My every thought
And this is why I find myself awake

-Failure
-Tom Shear (Assemblage 23)

30.1 An Eigenvalue Problem.

A self adjoint eigenvalue problem. Consider the eigenvalue problem

y 00 + λy = 0, y(−π) = y(π), y 0 (−π) = y 0 (π).

1195
We rewrite the equation so the eigenvalue is on the right side.
L[y] ≡ −y 00 = λy
We demonstrate that this eigenvalue problem is self adjoint.
hv|L[u]i − hL[v]|ui = hv| − u00 i − h−v 00 |ui
= [−v̄u0 ]π−π + hv 0 |u0 i − [−v̄ 0 u]−π π − hv 0 |u0 i
= −v(π)u0 (π) + v(−π)u0 (−π) + v 0 (π)u(π) − v 0 (−π)u(−π)
= −v(π)u0 (π) + v(π)u0 (π) + v 0 (π)u(π) − v 0 (π)u(π)
=0
Since Green’s Identity reduces to hv|L[u]i − hL[v]|ui = 0, the problem is self adjoint. This means that the
eigenvalues are real and that eigenfunctions corresponding to distinct eigenvalues are orthogonal. We compute
the Rayleigh quotient for an eigenvalue λ with eigenfunction φ.
−[φ̄φ0 ]π−π + hφ0 |φ0 i
λ=
hφ|φi
−φ(π)φ0 (π) + φ(−π)φ0 (−π) + hφ0 |φ0 i
=
hφ|φi
−φ(π)φ0 (π) + φ(π)φ0 (π) + hφ0 |φ0 i
=
hφ|φi
0 0
hφ |φ i
=
hφ|φi
We see that the eigenvalues are non-negative.

Computing the eigenvalues and eigenfunctions. Now we find the eigenvalues and eigenfunctions. First we
consider the case λ = 0. The general solution of the differential equation is
y = c1 + c2 x.

1196
The solution that satisfies the boundary conditions is y = const.
Now consider λ > 0. The general solution of the differential equation is
√  √ 
y = c1 cos λx + c2 sin λx .
We apply the first boundary condition.
y(−π) = y(π)
 √   √  √  √ 
c1 cos − λπ + c2 sin − λπ = c1 cos λπ + c2 sin λπ
√  √  √  √ 
c1 cos λπ − c2 sin λπ = c1 cos λπ + c2 sin λπ
√ 
c2 sin λπ = 0
Then we apply the second boundary condition.
y 0 (−π) = y 0 (π)
√  √  √  √  √ √  √ √ 
−c1 λ sin − λπ + c2 λ cos − λπ = −c1 λ sin λπ + c2 λ cos λπ
√  √  √  √ 
c1 sin λπ + c2 cos λπ = −c1 sin λπ + c2 cos λπ
√ 
c1 sin λπ = 0
√ 
To satisify the two boundary conditions either c1 = c2 = 0 or sin λπ = 0. The former yields the trivial
solution. The latter gives us the eigenvalues λn = n2 , n ∈ Z+ . The corresponding solution is
yn = c1 cos(nx) + c2 sin(nx).
There are two eigenfunctions for each of the positive eigenvalues.
We choose the eigenvalues and eigenfunctions.
1
λ0 = 0, φ0 =
2
2
λn = n , φ2n−1 = cos(nx), φ2n = sin(nx), for n = 1, 2, 3, . . .

1197
Orthogonality of Eigenfunctions. We know that the eigenfunctions of distinct eigenvalues are orthogonal.
In addition, the two eigenfunctions of each positive eigenvalue are orthogonal.
Z π  π
1 2
cos(nx) sin(nx) dx = sin (nx) =0
−π 2n −π

Thus the eigenfunctions { 21 , cos(x), sin(x), cos(2x), sin(2x)} are an orthogonal set.

30.2 Fourier Series.

A series of the eigenfunctions

1
φ0 = , φ(1)
n = cos(nx), φ(2)
n = sin(nx), for n ≥ 1
2
is
∞
1 X

a0 + an cos(nx) + bn sin(nx) .
2 n=1

This is known as a Fourier series. (We choose φ0 = 21 so all of the eigenfunctions have the same norm.) A fairly
general class of functions can be expanded in Fourier series. Let f (x) be a function defined on −π < x < π.
Assume that f (x) can be expanded in a Fourier series
∞
1 X

f (x) ∼ a0 + an cos(nx) + bn sin(nx) . (30.1)
2 n=1

Here the “∼” means “has the Fourier series”. We have not said if the series converges yet. For now let’s assume
that the series converges uniformly so we can replace the ∼ with an =.

1198
 We integrate Equation 30.1 from −π to π to determine a0 .
Z π Z π Z π ∞
1 X
f (x) dx = a0 dx + an cos(nx) + bn sin(nx) dx
−π 2 −π −π n=1
Z π ∞ 
X Z π Z π 
f (x) dx = πa0 + an cos(nx) dx + bn sin(nx) dx
−π n=1 −π −π
Z π
f (x) dx = πa0
−π
1 π
Z
a0 = f (x) dx
π −π

Multiplying by cos(mx) and integrating will enable us to solve for am .

Z π Z π
1
f (x) cos(mx) dx = a0 cos(mx) dx
−π 2 −π
X∞  Z π Z π 
+ an cos(nx) cos(mx) dx + bn sin(nx) cos(mx) dx
n=1 −π −π

All but one of the terms on the right side vanishes due to the orthogonality of the eigenfunctions.
Z π Z π
f (x) cos(mx) dx = am cos(mx) cos(mx) dx
−π −π
Z π Z π 
1
f (x) cos(mx) dx = am + cos(2mx) dx
−π −π 2
Z π
f (x) cos(mx) dx = πam
−π
1 π
Z
am = f (x) cos(mx) dx.
π −π

1199
Note that this formula is valid for m = 0, 1, 2, . . . .

Similarly, we can multiply by sin(mx) and integrate to solve for bm . The result is

Z π
1
bm = f (x) sin(mx) dx.
π −π

an and bn are called Fourier coefficients.

Although we will not show it, Fourier series converge for a fairly general class of functions. Let f (x− ) denote
the left limit of f (x) and f (x+ ) denote the right limit.

Example 30.2.1 For the function defined

(
0 for x < 0,
f (x) =
x+1 for x ≥ 0,

the left and right limits at x = 0 are

f (0− ) = 0, f (0+ ) = 1.

1200
 Rπ
Result 30.2.1 Let f (x) be a 2π-periodic function for which −π |f (x)| dx exists. Define
the Fourier coefficients
1 π 1 π
Z Z
an = f (x) cos(nx) dx, bn = f (x) sin(nx) dx.
π −π π −π
If x is an interior point of an interval on which f (x) has limited total fluctuation, then
the Fourier series of f (x)
∞
a0 X

+ an cos(nx) + bn sin(nx) ,
2 n=1

converges to 21 (f (x− ) + f (x+ )). If f is continuous at x, then the series converges to f (x).

Periodic Extension of a Function. Let g(x) be a function that is arbitrarily defined on −π ≤ x < π. The
Fourier series of g(x) will represent the periodic extension of g(x). The periodic extension, f (x), is defined by the
two conditions:

f (x) = g(x) for − π ≤ x < π,

f (x + 2π) = f (x).

The periodic extension of g(x) = x2 is shown in Figure 30.1.

Limited Fluctuation. A function that has limited total fluctuation can be written f (x) = ψ+ (x) − ψ− (x),
where ψ+ and ψ− are bounded, nondecreasing functions. An example of a function that does not have limited
total fluctuation is sin(1/x), whose fluctuation is unlimited at the point x = 0.

1201
 10

-5 5 10
-2

Figure 30.1: The Periodic Extension of g(x) = x2 .

Functions with Jump Discontinuities. Let f (x) be a discontinuous function that has a convergent Fourier
series. Note that the series does not necessarily converge to f (x). Instead it converges to fˆ(x) = 21 (f (x− )+f (x+ )).

Example 30.2.2 Consider the function defined by

(
−x for − π ≤ x < 0
f (x) =
π − 2x for 0 ≤ x < π.

1202
The Fourier series converges to the function defined by


0 for x = −π

−x for −π <x<0
fˆ(x) =


π/2 for x=0
π − 2x

for 0 < x < π.

The function fˆ(x) is plotted in Figure 30.2.

-3 -2 -1 1 2 3

-1

-2

-3

Figure 30.2: Graph of fˆ(x).

1203
30.3 Least Squares Fit
Approximating a function with a Fourier series. Suppose we want to approximate a 2π-periodic function
f (x) with a finite Fourier series.

N
a0 X
f (x) ≈ + (an cos(nx) + bn sin(nx))
2 n=1

Here the coefficients are computed with the familiar formulas. Is this the best approximation to the function?
That is, is it possible to choose coefficients αn and βn such that

N
α0 X
f (x) ≈ + (αn cos(nx) + βn sin(nx))
2 n=1

would give a better approximation?

Least squared error fit. The most common criterion for finding the best fit to a function is the least squares
fit. The best approximation to a function is defined as the one that minimizes the integral of the square of the
deviation. Thus if f (x) is to be approximated on the interval a ≤ x ≤ b by a series

N
X
f (x) ≈ cn φn (x), (30.2)
n=1

the best approximation is found by choosing values of cn that minimize the error E.

Z b  N
2
X 
E≡ f (x) − cn φn (x) dx
 
a 
n=1


1204
Generalized Fourier coefficients. We consider the case that the φn are orthogonal. For simplicity, we also
assume that the φn are real-valued. Then most of the terms will vanish when we interchange the order of
integration and summation.
Z b N N N
!
X X X
2
E= f − 2f cn φn + cn φn cm φm dx
a n=1 n=1 m=1
Z b N
X Z b N
XXN Z b
2
E= f dx − 2 cn f φn dx + cn cm φn φm dx
a n=1 a n=1 m=1 a
Z b N
X Z b N
X Z b
2
E= f dx − 2 cn f φn dx + c2n φ2n dx
a n=1 a n=1 a
Z b N
X Z b Z b 
2
E= f dx + c2n φ2n dx − 2cn f φn dx
a n=1 a a

We complete the square for each term.

 Rb !2 Rb !2 
b Z b N
f φ dx f φn dx
Z
n
X
E= f 2 dx +  φ2n dx cn − Ra b − Ra b 
a a φ 2 dx 2
φ dx
n=1 a n a n

Each term involving cn is non-negative, and is minimized for

Rb
f φn dx
cn = Ra b . (30.3)
φ 2 dx
a n

We call these the generalized Fourier coefficients.

For such a choice of the cn , the error is
Z b N
X Z b
2 2
E= f dx − cn φ2n dx.
a n=1 a

1205
Since the error is non-negative, we have
Z b N
X Z b
2
f dx ≥ c2n φ2n dx.
a n=1 a

This is known as Bessel’s Inequality. If the series in Equation 30.2 converges in the mean to f (x), lim N → ∞E =
0, then we have equality as N → ∞.
Z b X∞ Z b
2 2
f dx = cn φ2n dx.
a n=1 a

This is Parseval’s equality.

Fourier coefficients. Previously we showed that if the series,

∞
a0 X
f (x) = + (an cos(nx) + bn sin(nx),
2 n=1

converges uniformly then the coefficients in the series are the Fourier coefficients,
1 π 1 π
Z Z
an = f (x) cos(nx) dx, bn = f (x) sin(nx) dx.
π −π π −π
Now we show that by choosing the coefficients to minimize the squared error, we obtain the same result. We
apply Equation 30.3 to the Fourier eigenfunctions.
Rπ 1
f dx 1 π
Z
−π 2
a0 = R π 1 = f (x) dx
−π 4
dx π −π
Rπ
f cos(nx) dx 1 π
Z
−π
an = R π 2 (nx) dx
= f (x) cos(nx) dx
cos π −π
R −π
π
f sin(nx) dx 1 π
Z
−π
bn = π = f (x) sin(nx) dx
sin2 (nx) dx
R
−π
π −π

1206
30.4 Fourier Series for Functions Defined on Arbitrary Ranges
If f (x) is defined on c − d ≤ x < c + d and f (x + 2d) = f (x), then f (x) has a Fourier series of the form
∞    
a0 X nπ(x + c) nπ(x + c)
f (x) ∼ + an cos + bn sin .
2 n=1
d d
Since
Z c+d   Z c+d  
2 nπ(x + c) 2 nπ(x + c)
cos dx = sin dx = d,
c−d d c−d d
the Fourier coefficients are given by the formulas
1 c+d
Z  
nπ(x + c)
an = f (x) cos dx
d c−d d
1 c+d
Z  
nπ(x + c)
bn = f (x) sin dx.
d c−d d

Example 30.4.1 Consider the function defined by


x + 1
 for − 1 ≤ x < 0
f (x) = x for 0 ≤ x < 1

3 − 2x for 1 ≤ x < 2.


This function is graphed in Figure 30.3.

The Fourier series converges to fˆ(x) = (f (x− ) + f (x+ ))/2,


 − 12 for x = −1

x + 1 for − 1 < x < 0



fˆ(x) = 2 1
for x = 0

x for 0 < x < 1





3 − 2x for 1 ≤ x < 2.

1207
fˆ(x) is also graphed in Figure 30.3.

1 1

0.5 0.5

-1 -0.5 0.5 1 1.5 2 -1 -0.5 0.5 1 1.5 2

-0.5 -0.5

-1 -1

Figure 30.3: A Function Defined on the range −1 ≤ x < 2 and the Function to which the Fourier Series Converges.

1208
The Fourier coefficients are
Z 2  
1 2nπ(x + 1/2)
an = f (x) cos dx
3/2 −1 3
2 5/2
Z  
2nπx
= f (x − 1/2) cos dx
3 −1/2 3
2 1/2 2 3/2
Z   Z  
2nπx 2nπx
= (x + 1/2) cos dx + (x − 1/2) cos dx
3 −1/2 3 3 1/2 3
2 5/2
Z  
2nπx
+ (4 − 2x) cos dx
3 3/2 3
 
1 2nπ h n
 nπ i
=− sin 2(−1) nπ + 9 sin
(nπ)2 3 3

Z 2  
1 2nπ(x + 1/2)
bn = f (x) sin dx
3/2 −1 3
2 5/2
Z  
2nπx
= f (x − 1/2) sin dx
3 −1/2 3
2 1/2 2 3/2
Z   Z  
2nπx 2nπx
= (x + 1/2) sin dx + (x − 1/2) sin dx
3 −1/2 3 3 1/2 3
2 5/2
Z  
2nπx
+ (4 − 2x) sin dx
3 3/2 3
2 2 nπ
 h
n
 nπ   nπ i
=− sin 2(−1) nπ + 4nπ cos − 3 sin
(nπ)2 3 3 3

1209
30.5 Fourier Cosine Series
If f (x) is an even function, (f (−x) = f (x)), then there will not be any sine terms in the Fourier series for f (x).
The Fourier sine coefficient is
1 π
Z
bn = f (x) sin(nx) dx.
π −π
Since f (x) is an even function and sin(nx) is odd, f (x) sin(nx) is odd. bn is the integral of an odd function from
−π to π and is thus zero. We can rewrite the cosine coefficients,
1 π
Z
an = f (x) cos(nx) dx
π −π
2 π
Z
= f (x) cos(nx) dx.
π 0

Example 30.5.1 Consider the function defined on [0, π) by

(
x for 0 ≤ x < π/2
f (x) =
π−x for π/2 ≤ x < π.

The Fourier cosine coefficients for this function are

2 π/2 2 π
Z Z
an = x cos(nx) dx + (π − x) cos(nx) dx
π 0 π π/2
(
π
for n = 0,
= 48 nπ

2 nπ

πn2
cos 2 sin 4 for n ≥ 1.

In Figure 30.4 the even periodic extension of f (x) is plotted in a dashed line and the sum of the first five nonzero
terms in the Fourier cosine series are plotted in a solid line.

1210
 1.5

1.25

0.75

0.5

0.25

-3 -2 -1 1 2 3

Figure 30.4: Fourier Cosine Series.

30.6 Fourier Sine Series

If f (x) is an odd function, (f (−x) = −f (x)), then there will not be any cosine terms in the Fourier series. Since
f (x) cos(nx) is an odd function, the cosine coefficients will be zero. Since f (x) sin(nx) is an even function,we can
rewrite the sine coefficients

Z π
2
bn = f (x) sin(nx) dx.
π 0

1211
Example 30.6.1 Consider the function defined on [0, π) by
(
x for 0 ≤ x < π/2
f (x) =
π−x for π/2 ≤ x < π.
The Fourier sine coefficients for this function are
2 π/2 2 π
Z Z
bn = x sin(nx) dx + (π − x) sin(nx) dx
π 0 π π/2
16  nπ 
3 nπ
 
= cos sin
πn2 4 4
In Figure 30.5 the odd periodic extension of f (x) is plotted in a dashed line and the sum of the first five nonzero
terms in the Fourier sine series are plotted in a solid line.

30.7 Complex Fourier Series and Parseval’s Theorem

By writing sin(nx) and cos(nx) in terms of einx and e−inx we can obtain the complex form for a Fourier series.
∞ ∞  
a0 X
a0 X 1 inx −inx 1 inx −inx
+ an cos(nx) + bn sin(nx) = + an ( e + e ) + bn ( e − e )
2 n=1
2 n=1
2 2i
∞  
a0 X 1 inx 1 −inx
= + (an − ibn ) e + (an + ibn ) e
2 n=1
2 2
∞
X
= cn einx
n=−∞

where

1
 2 (an − ibn )
 for n ≥ 1
a0
cn = 2
for n = 0

1
(a
2 −n
+ ib−n ) for n ≤ −1.

1212
 1.5

0.5

-3 -2 -1 1 2 3

-0.5

-1

-1.5

Figure 30.5: Fourier Sine Series.

The functions {. . . , e−ix , 1, eix , e2ix , . . . }, satisfy the relation

Z π Z π
inx −imx
e e dx = ei(n−m)x dx
−π
(−π
2π for n = m
=
0 for n 6= m.

Starting with the complex form of the Fourier series of a function f (x),
∞
X
f (x) ∼ cn einx ,
−∞

1213
we multiply by e−imx and integrate from −π to π to obtain
Z π Z π ∞
X
−imx
f (x) e dx = cn einx e−imx dx
−π −π −∞
Z π
1
cm = f (x) e−imx dx
2π −π

If f (x) is real-valued then

Z π Z π
1 imx 1
c−m = f (x) e dx = f (x)( e−imx ) dx = cm
2π −π 2π −π

where z̄ denotes the complex conjugate of z.

Assume that f (x) has a uniformly convergent Fourier series.

∞
! ∞
!
Z π Z π X X
f 2 (x) dx = cm eimx cn einx dx
−π −π m=−∞ n=−∞
∞
X
= 2π cn c−n
n=−∞
−1   ∞  !
X 1 a0 a0 X 1
= 2π (a−n + ib−n )(a−n − ib−n ) + + (an − ibn )(an + ibn )
n=−∞
4 2 2 n=1
4
∞
!
a20 1 X 2
= 2π + (a + b2n )
4 2 n=1 n

This yields a result known as Parseval’s theorem which holds even when the Fourier series of f (x) is not uniformly
convergent.

1214
 Result 30.7.1 Parseval’s Theorem. If f (x) has the Fourier series
∞
a0 X
f (x) ∼ + (an cos(nx) + bn sin(nx)),
2 n=1

then
Z π ∞
2 π 2 X
f (x) dx = a0 + π (a2n + b2n ).
−π 2 n=1

30.8 Behavior of Fourier Coefficients

Before we jump hip-deep into the grunge involved in determining the behavior of the Fourier coefficients, let’s
take a step back and get some perspective on what we should be looking for.
One of the important questions is whether the Fourier series converges uniformly. From Result 14.2.1 we know
that a uniformly convergent series represents a continuous function. Thus we know that the Fourier series of a
discontinuous function cannot be uniformly convergent. From Section 14.2 we know that a series is uniformly
convergent if it can be bounded by a series of positive terms.
P∞ If the Fourier coefficients, an and bn , are O(1/nα )
where α > 1 then the series can be bounded by (const) n=1 1/nα and will thus be uniformly convergent.

Let f (x) be a function that meets the conditions for having a Fourier series and in addition is bounded. Let
(−π, p1 ), (p1 , p2 ), (p2 , p3 ), . . . , (pm , π) be a partition into a finite number of intervals of the domain, (−π, π) such
that on each interval f (x) and all it’s derivatives are continuous. Let f (p− ) denote the left limit of f (p) and f (p+ )
denote the right limit.

f (p− ) = lim+ f (p − ), f (p+ ) = lim+ f (p + )

→0 →0

1215
Example 30.8.1 The function shown in Figure 30.6 would be partitioned into the intervals

(−2, −1), (−1, 0), (0, 1), (1, 2).

0.5

-2 -1 1 2

-0.5

-1

Figure 30.6: A Function that can be Partitioned.

Suppose f (x) has the Fourier series

∞
a0 X
f (x) ∼ + an cos(nx) + bn sin(nx).
2 n=1

1216
We can use the integral formula to find the an ’s.

1 π
Z
an = f (x) cos(nx) dx
π −π
Z p1 Z p2 Z π 
1
= f (x) cos(nx) dx + f (x) cos(nx) dx + · · · + f (x) cos(nx) dx
π −π p1 pm

Using integration by parts,

h iπ 
1 ip1 h ip2 h
= f (x) sin(nx) + f (x) sin(nx) + · · · + f (x) sin(nx)
nπ −π p1 pm
Z p1 Z p2 Z π 
1 0 0 0
− f (x) sin(nx) dx + f (x) sin(nx) dx + f (x) sin(nx) dx
nπ −π p1 pm
1 n −  − o
f (p1 ) − f (p+ +
 
= 1 ) sin(np 1 ) + · · · + f (pm ) − f (p m ) sin(np m )
nπ Z π
11
− f 0 (x) sin(nx) dx
n π −π
1 1
= An − b0n
n n
where
m
1X
sin(npj ) f (p− +
 
An = j ) − f (p j )
π j=1

and the b0n are the sine coefficients of f 0 (x).

Since f (x) is bounded, An = O(1). Since f 0 (x) is bounded,

1 π 0
Z
0
bn = f (x) sin(nx) dx = O(1).
π −π

1217
Thus an = O(1/n) as n → ∞. (Actually, from the Riemann-Lebesgue Lemma, b0n = O(1/n).)
Now we repeat this analysis for the sine coefficients.
1 π
Z
bn = f (x) sin(nx) dx
π −π
Z p1 Z p2 Z π 
1
= f (x) sin(nx) dx + f (x) sin(nx) dx + · · · + f (x) sin(nx) dx
π −π p1 pm
−1 n p1  p  π o
= f (x) cos(nx) −π + f (x) cos(nx) p21 + · · · + f (x) cos(nx) pm
nπ Z p1 Z p2 Z π 
1 0 0 0
+ f (x) cos(nx) dx + f (x) cos(nx) dx + f (x) cos(nx) dx
nπ −π p1 pm
1 1
= − Bn + a0n
n n
where
m
(−1)n   1X
cos(npj ) f (p− +
 
Bn = f (−π) − f (π) − j ) − f (pj )
π π j=1

and the a0n are the cosine coefficients of f 0 (x).

Since f (x) and f 0 (x) are bounded, Bn , a0n = O(1) and thus bn = O(1/n) as n → ∞.
With integration by parts on the Fourier coefficients of f 0 (x) we could find that
1 0 1
a0n = An − b00n
n n
Pm
where A0n = 1
π j=1 sin(npj )[f 0 (p− 0 + 00 00
j ) − f (pj )] and the bn are the sine coefficients of f (x), and

1 1
b0n = − Bn0 + a00n
n n
(−1)n Pm
where Bn0 = π
[f 0 (−π)−f 0 (π)]− π1 j=1 cos(npj )[f 0 (p− 0 + 00 00
j )−f (pj )] and the an are the cosine coefficients of f (x).

1218
 Now we can rewrite an and bn as
1 1 1
an = An + 2 Bn0 − 2 a00n
n n n
1 1 0 1
bn = − Bn + 2 An − 2 b00n .
n n n
(j) (j)
Continuing this process we could define An and Bn so that
1 1 1 1
an = An + 2 Bn0 − 3 A00n − 4 Bn000 + · · ·
n n n n
1 1 0 1 00 1
bn = − Bn + 2 An + 3 Bn − 4 A000 − ··· .
n n n n n
For any bounded function, the Fourier coefficients satisfy an , bn = O(1/n) as n → ∞. If An and Bn are zero
then the Fourier coefficients will be O(1/n2 ). A sufficient condition for this is that the periodic extension of f (x)
is continuous. We see that if the periodic extension of f 0 (x) is continuous then A0n and Bn0 will be zero and the
Fourier coefficients will be O(1/n3 ).

Result 30.8.1 Let f (x) be a bounded function for which there is a partition of the
range (−π, π) into a finite number of intervals such that f (x) and all it’s derivatives are
continuous on each of the intervals. If f (x) is not continuous then the Fourier coefficients
are O(1/n). If f (x), f 0 (x), . . . , f (k−2) (x) are continuous then the Fourier coefficients are
O(1/nk ).

P∞If the periodic extension of f (x) is continuous,Pthen the Fourier coefficients will be O(1/n2 ). The series
∞ 2
n=1 |an cos(nx)bn sin(nx)| can be bounded by M n=1 1/n where M = max(|an | + |bn |). Thus the Fourier
n
series converges to f (x) uniformly.

Result 30.8.2 If the periodic extension of f (x) is continuous then the Fourier series of
f (x) will converge uniformly for all x.

1219
 If the periodic extension of f (x) is not continuous, we have the following result.

Result 30.8.3 If f (x) is continuous in the interval c < x < d, then the Fourier series is
uniformly convergent in the interval c + δ ≤ x ≤ d − δ for any δ > 0.

Example 30.8.2 Different Rates of Convergence.

A Discontinuous Function. Consider the function defined by

(
−1 for − 1 < x < 0
f1 (x) =
1, for 0 < x < 1.

This function has jump discontinuities, so we know that the Fourier coefficients are O(1/n).
Since this function is odd, there will only be sine terms in it’s Fourier expansion. Furthermore, since the
function is symmetric about x = 1/2, there will be only odd sine terms. Computing these terms,
Z 1
bn = 2 sin(nπx) dx
0
 1
−1
=2 cos(nπx)
nπ 0
(−1)n −1
 
=2 − −
nπ nπ
(
4
for odd n
= nπ
0 for even n.

The function and the sum of the first three terms in the expansion are plotted, in dashed and solid lines
respectively, in Figure 30.7. Although the three term sum follows the general shape of the function, it is clearly
not a good approximation.

1220
 1 0.4

0.5 0.2

-1 -0.5 0.5 1 -1 -0.5 0.5 1

-0.5 -0.2

-1 -0.4

Figure 30.7: Three Term Approximation for a Function with Jump Discontinuities and a Continuous Function.

A Continuous Function. Consider the function defined by


−x − 1
 for − 1 < x < −1/2
f2 (x) = x for − 1/2 < x < 1/2

−x + 1 for 1/2 < x < 1.


1221
 1 0.5
0.2

0.1 1 1

-1 -0.5 0.5 1 0.25 0.1

-0.1
1
-0.2
0
0.1

Figure 30.8: Three Term Approximation for a Function with Continuous First Derivative and Comparison of the
Rates of Convergence.

Since this function is continuous, the Fourier coefficients will be O(1/n2 ). Also we see that there will only be odd
sine terms in the expansion.
Z −1/2 Z 1/2 Z 1
bn = (−x − 1) sin(nπx) dx + x sin(nπx) dx + (−x + 1) sin(nπx) dx
−1 −1/2 1/2
Z 1/2 Z 1
=2 x sin(nπx) dx + 2 (1 − x) sin(nπx) dx
0 1/2
4
= sin(nπ/2)
(nπ)2
(
4
(nπ)2
(−1)(n−1)/2 for odd n
=
0 for even n.
1222
 The function and the sum of the first three terms in the expansion are plotted, in dashed and solid lines respec-
tively, in Figure 30.7. We see that the convergence is much better than for the function with jump discontinuities.

A Function with a Continuous First Derivative. Consider the function defined by

(
x(1 + x) for − 1 < x < 0
f3 (x) =
x(1 − x) for 0 < x < 1.

Since the periodic extension of this function is continuous and has a continuous first derivative, the Fourier
coefficients will be O(1/n3 ). We see that the Fourier expansion will contain only odd sine terms.

Z 0 Z 1
bn = x(1 + x) sin(nπx) dx + x(1 − x) sin(nπx) dx
−1 0
Z 1
=2 x(1 − x) sin(nπx) dx
0
4(1 − (−1)n )
=
(nπ)3
(
4
(nπ)3
for odd n
=
0 for even n.

The function and the sum of the first three terms in the expansion are plotted in Figure 30.8. We see that the
first three terms give a very good approximation to the function. The plots of the function, (in a dashed line),
and the three term approximation, (in a solid line), are almost indistinguishable.
In Figure 30.8 the convergence of the of the first three terms to f1 (x), f2 (x), and f3 (x) are compared. In the
last graph we see a closeup of f3 (x) and it’s Fourier expansion to show the error.

1223
30.9 Gibb’s Phenomenon
The Fourier expansion of
(
1 for 0 ≤ x < 1
f (x) =
−1 for − 1 ≤ x < 0

is
∞
4X1
f (x) ∼ sin(nπx).
π n=1 n

For any fixed x, the series converges to 21 (f (x− ) + f (x+ )). For any δ > 0, the convergence is uniform in the
intervals −1 + δ ≤ x ≤ −δ and δ ≤ x ≤ 1 − δ. How will the nonuniform convergence at integral values of x
affect the Fourier series? Finite Fourier series are plotted in Figure 30.9 for 5, 10, 50 and 100 terms. (The plot
for 100 terms is closeup of the behavior near x = 0.) Note that at each discontinuous point there is a series of
overshoots and undershoots that are pushed closer to the discontinuity by increasing the number of terms, but
do not seem to decrease in height. In fact, as the number of terms goes to infinity, the height of the overshoots
and undershoots does not vanish. This is known as Gibb’s phenomenon.

30.10 Integrating and Differentiating Fourier Series

Integrating Fourier Series. Since integration is a smoothing operation, any convergent Fourier series can be
integrated term by term to yield another convergent Fourier series.

Example 30.10.1 Consider the step function

(
π for 0 ≤ x < π
f (x) =
−π for − π ≤ x < 0.

1224
 1 1

1 1

1 1.2

1 0.1

0.8

Figure 30.9:

Since this is an odd function, there are no cosine terms in the Fourier series.

2 π
Z
bn = π sin(nx) dx
π 0
 π
1
= 2 − cos(nx)
n 0
2
= (1 − (−1)n )
n
(
4
for odd n
= n
0 for even n.

1225
 ∞
X 4
f (x) ∼ sin nx
n=1
n
oddn

Integrating this relation,

Z x Z x ∞
X 4
f (t) dt ∼ sin(nt) dt
−π −π n=1
n
oddn
∞
4 x
X Z
F (x) ∼ sin(nt) dt
n=1
n −π
oddn
∞  x
X 4 1
= − cos(nt)
n=1
n n −π
oddn
∞
X 4
= (− cos(nx) + (−1)n )
n=1
n2
oddn
∞ ∞
X −1 X cos(nx)
=4 2
−4
n=1
n n=1
n2
oddn oddn

Since this series converges uniformly,

∞ ∞
(
X −1 X cos(nx) −x − π for − π ≤ x < 0
4 2
−4 = F (x) =
n=1
n n=1
n2 x−π for 0 ≤ x < π.
oddn oddn

The value of the constant term is

∞
−1 2 π
Z
X 1
4 2
= F (x) dx = − .
n=1
n π 0 π
oddn

1226
Thus
∞
(
1 X cos(nx) −x − π for − π ≤ x < 0
− −4 =
π n=1
n2 x−π for 0 ≤ x < π.
oddn

Differentiating Fourier Series. Recall that in general, a series can only be differentiated if it is uniformly
convergent. The necessary and sufficient condition that a Fourier series be uniformly convergent is that the
periodic extension of the function is continuous.

Result 30.10.1 The Fourier series of a function f (x) can be differentiated only if the
periodic extension of f (x) is continuous.

Example 30.10.2 Consider the function defined by

(
π for 0 ≤ x < π
f (x) =
−π for − π ≤ x < 0.

f (x) has the Fourier series

∞
X 4
f (x) ∼ sin nx.
n=1
n
oddn

The function has a derivative except at the points x = nπ. Differentiating the Fourier series yields
∞
X
0
f (x) ∼ 4 cos(nx).
n=1
oddn

1227
For x 6= nπ, this implies
∞
X
0=4 cos(nx),
n=1
oddn

which is false. The series does not converge. This is as we expected since the Fourier series for f (x) is not
uniformly convergent.

1228
30.11 Exercises
Exercise 30.1
1. Consider a 2π periodic function f (x) expressed as a Fourier series with partial sums

N
a0 X
SN (x) = + an cos(nx) + bn sin(nt).
2 n=1

Assuming that the Fourier series converges in the mean, i.e.

Z π
lim (f (x) − SN (x))2 dx = 0,
N →∞ −π

show
∞ π
a20 X 2
Z
1
+ an + b2n = f (x)2 dx.
2 n=1
π −π

This is called Parseval’s equation.

2. Find the Fourier series for f (x) = x on −π ≤ x < π (and repeating periodically). Use this to show
∞
X 1 π2
2
= .
n=1
n 6

3. Similarly, by choosing appropriate functions f (x), use Parseval’s equation to determine

∞ ∞
X 1 X 1
and .
n=1
n4 n=1
n6

1229
Exercise 30.2
Consider the Fourier series of f (x) = x on −π ≤ x < π as found above. Investigate the convergence at the points
of discontinuity.
1. Let SN be the sum of the first N terms in the Fourier series. Show that
N + 12 x



dSN N cos
= 1 − (−1) .
cos x2


dx

2. Now use this to show that

1
x



sin N+ (ξ − π)
Z
2
x − SN = ξ−π

dξ.
0 sin 2

3. Finally investigate the maxima of this difference around x = π and provide an estimate (good to two decimal
places) of the overshoot in the limit N → ∞.

Exercise 30.3
Consider the boundary value problem on the interval 0 < x < 1

y 00 + 2y = 1 y(0) = y(1) = 0.

1. Choose an appropriate periodic extension and find a Fourier series solution.

2. Solve directly and find the Fourier series of the solution (using the same extension). Compare the result to
the previous step and verify the series agree.

Exercise 30.4
Consider the boundary value problem on 0 < x < π

y 00 + 2y = sin x y 0 (0) = y 0 (π) = 0.

1230
 1. Find a Fourier series solution.
2. Suppose the ODE is slightly modified: y 00 + 4y = sin x with the same boundary conditions. Attempt to find
a Fourier series solution and discuss in as much detail as possible what goes wrong.

Exercise 30.5
Find the Fourier cosine and sine series for f (x) = x2 on 0 ≤ x < π. Are the series differentiable?

Exercise 30.6
Find the Fourier series of cosn (x).

Exercise 30.7
For what values of x does the Fourier series
∞
π2 X (−1)n
+4 2
cos nx = x2
3 n=1
n
converge? What is the value of the above Fourier series for all x? From this relation show that
∞
X 1 π2
=
n=1
n2 6
∞
X (−1)n+1 π2
=
n=1
n2 12

Exercise 30.8
1. Compute the Fourier sine series for the function
2x
f (x) = cos x − 1 + , 0 ≤ x ≤ π.
π

1231
 2. How fast do the Fourier coefficients an where
∞
X
f (x) = an sin nx
n=1

decrease with increasing n? Explain this rate of decrease.

Exercise 30.9
Determine the cosine and sine series of
f (x) = x sin x, (0 < x < π).
Estimate before doing the calculation the rate of decrease of Fourier coefficients, an , bn , for large n.

Exercise 30.10
Determine the Fourier cosine series of the function
f (x) = cos νx, 0 ≤ x ≤ π,
where ν is an arbitrary real number. From this series deduce that for ν 6= n
∞  
π 1 X n 1 1
= + (−1) +
sin πν ν n=1 ν−n ν+n
∞  
1 X 1 1
π cot πν = + +
ν n=1 ν − n ν + n

Integrate the last formula with respect to ν from ν = 0 to ν = θ, (0 < θ < 1), to show that
∞ 
θ2

sin πθ Y
= 1− 2
πθ n=1
n

The symbol ∞
Q
1 un denotes the infinite product u1 u2 u3 · · · .

1232
Exercise 30.11
1. Show that
∞
x X (−1)n
log cos = − log 2 − cos nx, −π < x < π
2 n=1
n
Hint: use the identity
∞
X zn
log(1 − z) = − for |z| ≤ 1, z 6= 1.
n=1
n

2. From this series deduce that

Z π  x
log cos dx = −π log 2.
0 2
3. Show that
  ∞
1  sin((x + ξ)/2)  X sin nx sin nξ
log 
  = , x 6= ±ξ + 2kπ.
2 sin((x − ξ)/2) 
n=1
n

Exercise 30.12
Solve the problem
y 00 + αy = f (x), y(a) = y(b) = 0,
with an eigenfunction expansion. Assume that α 6= nπ/(b − a), n ∈ N.

Exercise 30.13
Solve the problem
y 00 + αy = f (x), y(a) = A, y(b) = B,
with an eigenfunction expansion. Assume that α 6= nπ/(b − a), n ∈ N.

1233
Exercise 30.14
Find the trigonometric series and the simple closed form expressions for A(r, x) and B(r, x) where z = r eix and
|r| < 1.
1
a) A + iB ≡ 2
= 1 + z2 + z4 + · · ·
1−z
1 1
b) A + iB ≡ log(1 + z) = z − z 2 + z 3 − · · ·
2 3
Find An and Bn , and the trigonometric sum for them where:

c) An + iBn = 1 + z + z 2 + · · · + z n .

Exercise 30.15
1. Is the trigonometric system

{1, sin x, cos x, sin 2x, cos 2x, . . . }

orthogonal on the interval [0, π]? Is the system orthogonal on any interval of length π? Why, in each case?
2. Show that each of the systems

{1, cos x, cos 2x, . . . }, and {sin x, sin 2x, . . . }

are orthogonal on [0, π]. Make them orthonormal too.

Exercise 30.16
Let SN (x) be the N th partial sum of the Fourier series for f (x) ≡ |x| on −π < x < π. Find N such that
|f (x) − SN (x)| < 10−1 on |x| < π.

Exercise 30.17
The set {sin(nx)}∞
n=1 is orthogonal and complete on [0, π].

1234
 1. Find the Fourier sine series for f (x) ≡ 1 on 0 ≤ x ≤ π.

2. Find a convergent series for g(x) = x on 0 ≤ x ≤ π by integrating the series for part (a).

3. Apply Parseval’s relation to the series in (a) to find:

∞
X 1
n=1
(2n − 1)2

Check this result by evaluating the series in (b) at x = π.

Exercise 30.18
1. Show that the Fourier cosine series expansion on [0, π] of:

π
1, 0 ≤ x < 2 ,

f (x) ≡ 12 , x = π2 ,

0, π2 < x ≤ π,


is
∞
1 2 X (−1)n
S(x) = + cos((2n + 1)x).
2 π n=0 2n + 1

2. Show that the N th partial sum of the series in (a) is

1 1 x−π/2 sin((2(N + 1)t)

Z
SN (x) = − dt.
2 π 0 sin t

( Hint: Consider the difference of 2N

P +1 iy n PN i2y n
n=1 ( e ) and n=1 ( e ) , where y = x − π/2.)
nπ
3. Show that dSN (x)/dx = 0 at x = xn = 2(N +1)
for n = 0, 1, . . . , N, N + 2, . . . , 2N + 2.

1235
 4. Show that at x = xN , the maximum of SN (x) nearest to π/2 in (0, π/2) is
Z πN
1 1 2(N +1) sin(2(N + 1)t)
SN (xN ) = + dt.
2 π 0 sin t
Clearly xN ↑ π/2 as N → ∞.
5. Show that also in this limit,
Z π
1 1 sin t
SN (xN ) → + dt ≈ 1.0895.
2 π 0 t
How does this compare with f (π/2 − 0)? This overshoot is the Gibbs phenomenon that occurs at each
discontinuity. It is a manifestation of the non-uniform convergence of the Fourier series for f (x) on [0, π].

Exercise 30.19
Prove the Isoperimetric Inequality: L2 ≥ 4πA where L is the length of the perimeter and A the area of any
piecewise smooth plane figure. Show that equality is attained only for the circle. (Hints: The closed curve is
represented parametrically as
x = x(s), y = y(s), 0≤s≤L
where s is the arclength. In terms of t = 2πs/L we have
 2  2  2
dx dy L
+ = .
dt dt 2π
Integrate this relation over [0, 2π]. The area is given by
Z 2π
dy
A= x dt.
0 dt
Express x(t) and y(t) as Fourier series and use the completeness and orthogonality relations to show that L2 −
4πA ≥ 0.)

1236
Exercise 30.20
1. Find the Fourier sine series expansion and the Fourier cosine series expansion of

g(x) = x(1 − x), on 0 ≤ x ≤ 1.

Which is better and why over the indicated interval?

2. Use these expansions to show that:

∞ ∞ ∞
X 1 π2 X (−1)k π2 X (−1)k π3
i) = , ii) =− , iii) = − .
k=1
k2 6 k=1
k2 12 k=1
(2k − 1)2 32

Note: Some useful integration by parts formulas are:

Z Z
1 x 1 x
x sin(nx) = 2 sin(nx) − cos(nx); x cos(nx) = 2 cos(nx) + sin(nx)
n n n n
2 2
n x −2
Z
2x
x2 sin(nx) = 2 sin(nx) − cos(nx)
n n3
n 2 x2 − 2
Z
2x
x2 cos(nx) = 2 cos(nx) + sin(nx)
n n3

1237
30.12 Hints
Hint 30.1

Hint 30.2

Hint 30.3

Hint 30.4

Hint 30.5

Hint 30.6
Expand
 n
1
cos (x) = ( eix + e−ix )
n
2

Using Newton’s binomial formula.

Hint 30.7

1238
Hint 30.8

Hint 30.9

Hint 30.10

Hint 30.11

Hint 30.12

Hint 30.13

Hint 30.14

Hint 30.15

1239
Hint 30.16

Hint 30.17

Hint 30.18

Hint 30.19

Hint 30.20

1240
30.13 Solutions
Solution 30.1
1. We start by assuming that the Fourier series converges in the mean.

∞
!2
Z π
a0 X
f (x) − − (an cos(nx) + bn sin(nx)) = 0
−π 2 n=1

We interchange the order of integration and summation.

Z π Z π ∞ 
X Z π Z π 
2
(f (x)) dx − a0 f (x) dx − 2 an f (x) cos(nx) dx + bn f (x) sin(nx)
−π −π n=1 −π −π
∞ Z π
πa20 X
+ + a0 (an cos(nx) + bn sin(nx)) dx
2 n=1 −π
∞ X
X ∞ Z π
+ (an cos(nx) + bn sin(nx))(am cos(mx) + bm sin(mx)) dx = 0
n=1 m=1 −π

Most of the terms vanish because the eigenfunctions are orthogonal.

Z π Z π ∞ 
X Z π Z π 
2
(f (x)) dx − a0 f (x) dx − 2 an f (x) cos(nx) dx + bn f (x) sin(nx)
−π −π n=1 −π −π
∞ π
πa2 X
Z
+ 0+ (a2n cos2 (nx) + b2n sin2 (nx)) dx = 0
2 n=1 −π

1241
 We use the definition of the Fourier coefficients to evaluate the integrals in the last sum.
Z π ∞ ∞
X
πa2 X
(f (x))2 dx − πa20 − 2π a2n + b2n + 0 + π a2n + b2n = 0


−π n=1
2 n=1
∞
a20 X 2
1 π
Z
+ an + b2n = f (x)2 dx
2 n=1
π −π

2. We determine the Fourier coefficients for f (x) = x. Since f (x) is odd, all of the an are zero.
1 π
Z
b0 = x sin(nx) dx
π −π
 π Z π
1 1 1
= − x cos(nx) + cos(nx) dx
π n −π −π n
2(−1)n+1
=
n
The Fourier series is
∞
X 2(−1)n+1
x= sin(nx) for x ∈ (−π . . . π).
n=1
n

We apply Parseval’s theorem for this series to find the value of ∞ 1

P
n=1 n2 .
∞
1 π 2
Z
X 4
2
= x dx
n=1
n π −π
∞
X 4 2π 2
=
n=1
n2 3
∞
X 1 π2
=
n=1
n2 6

1242
3. Consider f (x) = x2 . Since the function is even, there are no sine terms in the Fourier series. The coefficients
in the cosine series are

2 π 2
Z
a0 = x dx
π 0
2π 2
=
3Z
2 π 2
an = x cos(nx) dx
π 0
4(−1)n
= .
n2

Thus the Fourier series is

∞
2 π2 X (−1)n
x = +4 cos(nx) for x ∈ (−π . . . π).
3 n=1
n2

P∞ 1
We apply Parseval’s theorem for this series to find the value of n=1 n4 .

∞
2π 4 1 π 4
Z
X 1
+ 16 4
= x dx
9 n=1
n π −π
∞
2π 4 X 1 2π 4
+ 16 =
9 n=1
n4 5
∞
X 1 π4
=
n=1
n4 90

1243
Now we integrate the series for f (x) = x2 .

x ∞
π2 (−1)n x
Z   X Z
2
ξ − dξ = 4 cos(nξ) dξ
0 3 n=1
n2 0
∞
x3 π 2 X (−1)n
− x=4 sin(nx)
3 3 n=1
n3

P∞ 1
We apply Parseval’s theorem for this series to find the value of n=1 n6 .

∞ 2
1 π x3 π 2
Z 
X 1
16 6
= − x dx
n=1
n π −π 3 3
∞
X 1 16π 6
16 =
n=1
n6 945
∞
X 1 π6
=
n=1
n6 945

1244
Solution 30.2
1. We differentiate the partial sum of the Fourier series and evaluate the sum.

N
X 2(−1)n+1
SN = sin(nx)
n=1
n
XN
0
SN =2 (−1)n+1 cos(nx)
n=1
N
!
X
0
SN = 2< (−1)n+1 einx
n=1
1 − (−1)N +2 ei(N +1)x
 
0
SN = 2<
1 + eix
1 + e−ix − (−1)N ei(N +1)x − (−1)N eiN x
 
0
SN = <
1 + cos(x)
0 cos((N + 1)x) + cos(N x)
SN = 1 − (−1)N
1 + cos(x)
1 x




0 cos N + x cos
SN = 1 − (−1)N 2 2
cos2 x2

1



dSN cos N + x
= 1 − (−1)N x

2
dx cos 2

1245
 0
2. We integrate SN .
1



x
(−1)N cos N+ ξ
Z
2
SN (x) − SN (0) = x − ξ

dξ
0 cos 2
1
x



sin N+ (ξ − π)
Z
2
x − SN = ξ−π

dξ
0 sin 2

3. We find the extrema of the overshoot E = x − SN with the first derivative test.
sin N + 12 (x − π)



0
E = =0
sin x−π


2

We look for extrema in the range (−π . . . π).

 
1
N+ (x − π) = −nπ
2
 
n
x=π 1− , n ∈ [1 . . . 2N ]
N + 1/2
The closest of these extrema to x = π is
 
1
x=π 1−
N + 1/2
Let E0 be the overshoot at this point. We approximate E0 for large N .
sin N + 21 (ξ − π)
Z π(1−1/(N +1/2))


E0 = dξ
sin ξ−π


0 2

We shift the limits of integration.

1
π



sin N+ ξ
Z
2
E0 = ξ

dξ
π/(N +1/2) sin 2

1246
We add and subtract an integral over [0 . . . π/(N + 1/2)].

1 1
π


π/(N +1/2)



sin N+ ξ sin N+ ξ
Z Z
2 2
E0 = ξ

dξ − ξ

dξ
0 sin 2 0 sin 2

We can evaluate the first integral with contour integration on the unit circle C.

1
π


π
sin N+ ξ
Z Z
2 sin ((2N + 1) ξ)
ξ

dξ = dξ
0 sin 2 0 sin (ξ)
Z π
1 sin ((2N + 1) ξ)
= dξ
2 −π sin (ξ)


= z 2N +1
Z
1 dz
= −
2 C (z − 1/z)/(i2) iz
z 2N +1
Z 
== − 2 dz
C (z − 1)
z 2N +1 z 2N +1
    
= = iπ Res , 1 + iπ Res , −1
(z + 1)(z − 1) (z + 1)(z − 1)
 2N +1
(−1)2N +1

1
= π< +
2 −2
=π

1247
 We approximate the second integral.

1
π/(N +1/2)


Z π
sin N+ ξ
Z
2 2 sin(x)
dξ =
dx
ξ 2N + 1 0 sin 2Nx+1


0 sin 2
Z π
sin(x)
≈2 dx
0 x
Z π X ∞
1 (−1)n x2n+1
=2 dx
0 x n=0 (2n + 1)!
∞ Z π
X (−1)n x2n
=2 dx
n=0 0 (2n + 1)!
∞
X (−1)n π 2n+1
=2 dx
n=0
(2n + 1)(2n + 1)!
≈ 3.70387

In the limit as N → ∞, the overshoot is

|π − 3.70387| ≈ 0.56.

Solution 30.3
1. The eigenfunctions of the self-adjoint problem

−y 00 = λy, y(0) = y(1) = 0,

are

φn = sin(nπx), n ∈ Z+

1248
We find the series expansion of the inhomogeneity f (x) = 1.
∞
X
1= fn sin(nπx)
n=1
Z 1
fn = 2 sin(nπx) dx
0
 1
cos(nπx)
fn = 2 −
nπ 0
2
fn = (1 − (−1)n )
(nπ
4
nπ
for odd n
fn =
0 for even n
We expand the solution in a series of the eigenfunctions.
∞
X
y= an sin(nπx)
n=1

We substitute the series into the differential equation.

y 00 + 2y = 1
∞ ∞ ∞
X
2 2
X X 4
− an π n sin(nπx) + 2 an sin(nπx) = sin(nπx)
n=1 n=1 n=1
nπ
odd n
(
4
2 2 for odd n
an = nπ(2−π n )
0 for even n
∞
X 4
y= sin(nπx)
n=1
nπ(2 − π 2 n2 )
odd n

1249
2. Now we solve the boundary value problem directly.
y 00 + 2y = 1 y(0) = y(1) = 0
The general solution of the differential equation is
√  √  1
y = c1 cos 2x + c2 sin 2x + .
2
We apply the boundary conditions to find the solution.
1 √  √  1
c1 + = 0, c1 cos 2 + c2 sin 2 + =0
2 √
2
1 cos 2 − 1
c1 = − , c 2 = √

2 2 sin 2
√  cos √2
− 1
!
1 √ 
y= 1 − cos 2x + √
sin 2x
2 sin 2

We find the Fourier sine series of the solution.

∞
X
y= an sin(nπx)
n=1
Z 1
an = 2 y(x) sin(nπx) dx
0
√  cos √2
− 1
!
Z 1 √ 
an = 1 − cos 2x + √
sin 2x sin(nπx) dx
0 sin 2
2(1 − (−1)2
an =
nπ(2 − π 2 n2 )
(
4
nπ(2−π 2 n2 )
for odd n
an =
0 for even n

1250
 We obtain the same series as in the first part.

Solution 30.4
1. The eigenfunctions of the self-adjoint problem
−y 00 = λy, y 0 (0) = y 0 (π) = 0,
are
1
φ0 = , φn = cos(nx), n ∈ Z+
2
We find the series expansion of the inhomogeneity f (x) = sin(x).
∞
f0 X
f (x) = + fn cos(nx)
2 n=1
2 π
Z
f0 = sin(x) dx
π 0
4
f0 =
π
2 π
Z
fn = sin(x) cos(nx) dx
π 0
2(1 + (−1)n )
fn =
π(1 − n2 )
(
4
2 for even n
fn = π(1−n )
0 for odd n
We expand the solution in a series of the eigenfunctions.
∞
a0 X
y= + an cos(nx)
2 n=1

1251
 We substitute the series into the differential equation.

y 00 + 2y = sin(x)
∞ ∞ ∞
X X 2 X 4
− an n2 cos(nx) + a0 + 2 an cos(nx) = + 2)
cos(nx)
n=1 n=1
π n=2
π(1 − n
even n
∞
1 X 4
y= + 2 )(2 − n2 )
cos(nx)
π n=2
π(1 − n
even n

2. We expand the solution in a series of the eigenfunctions.

∞
a0 X
y= + an cos(nx)
2 n=1

We substitute the series into the differential equation.

y 00 + 4y = sin(x)
∞ ∞ ∞
X X 2 X 4
− an n2 cos(nx) + 2a0 + 4 an cos(nx) = + 2)
cos(nx)
n=1 n=1
π n=2
π(1 − n
even n

It is not possible to solve for the a2 coefficient. That equation is

4
(0)a2 = − .
3π
This problem is to be expected, as this boundary value problem does not have a solution. The solution of
the differential equation is
1
y = c1 cos(2x) + c2 sin(2x) + sin(x)
3

1252
 The boundary conditions give us an inconsistent set of constraints.

y 0 (0) = 0, y 0 (π) = 0
1 1
c2 + = 0, c2 − = 0
3 3

Thus the problem has no solution.

Solution 30.5
Cosine Series. The coefficients in the cosine series are

2 π 2
Z
a0 = x dx
π 0
2π 2
=
3Z
2 π 2
an = x cos(nx) dx
π 0
4(−1)n
= .
n2

Thus the Fourier cosine series is

∞
π 2 X 4(−1)n
f (x) = + 2
cos(nx).
3 n=1
n

In Figure 30.10 the even periodic extension of f (x) is plotted in a dashed line and the sum of the first five terms
in the Fourier series is plotted in a solid line. Since the even periodic extension is continuous, the cosine series is
differentiable.

1253
 10 10

8
5

-3 -2 -1 1 2 3
4

-5
2

-3 -2 -1 1 2 3 -10

Figure 30.10: The Fourier Cosine and Sine Series of f (x) = x2 .

Sine Series. The coefficients in the sine series are

Z π
2
bn = x2 sin(nx) dx
π 0
2(−1)n π 4(1 − (−1)n )
=− −
( n πn3
n
− 2(−1)
n
π
for even n
= n
2(−1) π 8
− n − πn3 for odd n.

1254
Thus the Fourier sine series is

∞ 
2(−1)n π 4(1 − (−1)n )
X 
f (x) ∼ − + sin(nx).
n=1
n πn3

In Figure 30.10 the odd periodic extension of f (x) and the sum of the first five terms in the sine series are plotted.
Since the odd periodic extension of f (x) is not continuous, the series is not differentiable.

Solution 30.6
We could find the expansion by integrating to find the Fourier coefficients, but it is easier to expand cosn (x)
directly.

 n
n 1 ix −ix
cos (x) = ( e + e )
2
        
1 n inx n i(n−2)x n −i(n−2)x n −inx
= n e + e + ··· + e + e
2 0 1 n−1 n

1255
If n is odd,

"   
1 n −inx n
cosn (x) = n inx
(e + e )+ ( ei(n−2)x + e−i(n−2)x ) + · · ·
2 0 1
  #
n
+ ( eix + e−ix )
(n − 1)/2
      
1 n n n
= n 2 cos(nx) + 2 cos((n − 2)x) + · · · + 2 cos(x)
2 0 1 (n − 1)/2
(n−1)/2  
1 X n
= n−1 cos((n − 2m)x)
2 m=0
m
n  
1 X n
= n−1 cos(kx).
2 k=1
(n − k)/2
odd k

1256
If n is even,
"   
1 n −inx n
cosn (x) = n inx
(e + e )+ ( ei(n−2)x + e−i(n−2)x ) + · · ·
2 0 1
   #
n n
+ ( ei2x + e−i2x ) +
n/2 − 1 n/2
       
1 n n n n
= n 2 cos(nx) + 2 cos((n − 2)x) + · · · + 2 cos(2x) +
2 0 1 n/2 − 1 n/2
  (n−2)/2  
1 n 1 X n
= n + n−1 cos((n − 2m)x)
2 n/2 2 m=0
m
  n  
1 n 1 X n
= n + n−1 cos(kx).
2 n/2 2 k=2
(n − k)/2
even k

We may denote,

n
n a0 X
cos (x) = ak cos(kx),
2 k=1

where

1 + (−1)n−k 1
 
n
ak = .
2 2n−1 (n − k)/2

1257
Solution 30.7
We expand f (x) in a cosine series. The coefficients in the cosine series are

2 π 2
Z
a0 = x dx
π 0
2π 2
=
3Z
2 π 2
an = x cos(nx) dx
π 0
4(−1)n
= .
n2
Thus the Fourier cosine series is
∞
π2 X (−1)n
f (x) = +4 2
cos(nx).
3 n=1
n

The Fourier series converges to the even periodic extension of

f (x) = x2 for 0 < x < π,

which is
  2
x+π
fˆ(x) = x − 2π .
2π

(b·c denotes the floor or greatest integer function.) This periodic extension is a continuous function. Since x2 is
an even function, we have
∞
π2 X (−1)n
+4 2
cos nx = x2 for − π ≤ x ≤ π.
3 n=1
n

1258
 We substitute x = π into the Fourier series.
∞
π2 X (−1)n
+4 2
cos(nπ) = π 2
3 n=1
n
∞
X 1 π2
=
n=1
n2 6

We substitute x = 0 into the Fourier series.

∞
π2 X (−1)n
+4 2
=0
3 n=1
n
∞
X (−1)n+1 π2
=
n=1
n2 12

Solution 30.8
1. The Fourier sine coefficients are

2 π
Z
an = f (x) sin(nx) dx
π 0
2 π
Z  
2x
= cos x − 1 + sin(nx) dx
π 0 π

2(1 + (−1)n )
an =
π(n3 − n)

2. From our work in the previous part, we see that the Fourier coefficients decay as 1/n3 . The Fourier sine
series converges to the odd periodic extension of f (x). We can determine the rate of decay of the Fourier

1259
 coefficients from the smoothness of fˆ(x). For −π < x < π, the odd periodic extension of f (x) is defined
(
f+ (x) = cos(x) − 1 + 2x 0 < x < π,
fˆ(x) = π
f− (x) = −f+ (−x) = − cos(x) + 1 + 2x
π
−π < x < 0.

Since
f+ (0) = f− (0) = 0 and f+ (π) = f− (−π) = 0
fˆ(x) is continuous, C 0 . Since
2 2
f+0 (0) = f−0 (0) = and f+0 (π) = f−0 (−π) =
π π
fˆ(x) is continuously differentiable, C 1 . However, since
f+00 (0) = −1, f−00 (0) = 1

fˆ(x) is not C 2 . Since fˆ(x) is C 1 we know that the Fourier coefficients decay as 1/n3 .

Solution 30.9
Cosine Series. The even periodic extension of f (x) is a C 0 , continuous, function (See Figure 30.11. Thus the
coefficients in the cosine series will decay as 1/n2 . The Fourier cosine coefficients are
2 π
Z
a0 = x sin x dx
π 0
=2
Z π
2
a1 = x sin x cos x dx
π 0
1
=−
2

1260
 2 π
Z
an = x sin x cos(nx) dx
π 0
2(−1)n+1
= , for n ≥ 2
n2 − 1
The Fourier cosine series is
∞
ˆ 1 X 2(−1)n
f (x) = 1 − cos x − 2 cos(nx).
2 n=2
n2 − 1

-5 5

Figure 30.11: The even periodic extension of x sin x.

Sine Series. The odd periodic extension of f (x) is a C 1 , continuously differentiable, function (See Fig-
ure 30.12. Thus the coefficients in the cosine series will decay as 1/n3 . The Fourier sine coefficients are
1 π
Z
a1 = x sin x sin x dx
π 0
π
=
2

1261
 Z π
2
an = x sin x sin(nx) dx
π 0
4(1 + (−1)n )n
=− , for n ≥ 2
π(n2 − 1)2

The Fourier sine series is

∞
ˆ π 4 X (1 + (−1)n )n
f (x) = sin x − cos(nx).
2 π n=2 (n2 − 1)2

-5 5

Figure 30.12: The odd periodic extension of x sin x.

Solution 30.10
If ν = n is an integer, then the Fourier cosine series is cos(|n|x).
We note that for ν 6= n, the even periodic extension of cos(νx) is C 0 so that the series converges to cos(νx)

1262
for −π ≤ x ≤ π and the coefficients decay as 1/n2 . If ν is not an integer, then then Fourier cosine coefficients are
2 π
Z
a0 = cos(νx) dx
π 0
2
= sin(πν)
πν
2 π
Z
an = cos(νx) cos(nx) dx
π 0
 
n 1 1
= (−1) + sin(πν)
ν−n ν+n
The Fourier cosine series is
∞  
1 X
n 1 1
cos(νx) = sin(πν) + (−1) + sin(πν) cos(nx).
πν n=1
ν − n ν + n

For ν 6= n we substitute x = 0 into the Fourier cosine series.

∞  
1 X
n 1 1
1= sin(πν) + (−1) + sin(πν)
πν n=1
ν − n ν + n
∞  
π 1 X n 1 1
= + (−1) +
sin πν ν n=1 ν−n ν+n

For ν 6= n we substitute x = π into the Fourier cosine series.

∞  
1 X
n 1 1
cos(νπ) = sin(πν) + (−1) + sin(πν)(−1)n
πν n=1
ν − n ν + n
∞  
1 X 1 1
π cot πν = + +
ν n=1 ν − n ν + n

1263
 We write the last formula as
∞  
1 X 1 1
π cot πν − = +
ν n=1
ν−n ν+n
We integrate from ν = 0 to ν = θ < 1.
  θ X ∞ 
sin(πν) θ θ

log = [log(n − ν)]0 + [log(n + ν)]0
ν 0 n=1
  ∞     
sin(πθ) X n−θ n+θ
log − log(π) = log + log
θ n=1
n n
∞
θ2
  X  
sin(πθ)
log = log 1 − 2
πθ n=1
n
∞  !
θ2
 
sin(πθ) Y
log = log 1− 2
πθ n=1
n
∞ 
θ2

sin(πθ) Y
= 1− 2
πθ n=1
n

Solution 30.11
1. We will consider the principal branch of the logarithm, −π < arg(log z) ≤ π. For −π < x < π, cos(x/2) is
positive so that log(cos(x/2)) is real-valued. At x = ±π, log(cos(x/2)) is singular. However, the function is
integrable so it has a Fourier series which converges except at x = (2k + 1)π, k ∈ Z.
+ e−ix/2
 ix/2 
 x e
log cos = log
2 2
= − log 2 + log e−ix/2 1 + eix



x
= − log 2 − i + log 1 + eix


2

1264
 Since | eix | ≤ 1 and eix 6= −1 for =(x) ≥ 0, x 6= (2k + 1)π, we can expand the last term in a Taylor series
in that domain.
∞
x X (−1)n ix
n
= − log 2 − i − e
2 n=1 n
∞ ∞
!
X (−1)n x X (−1)n
= − log 2 − cos nx − i + sin nx
n=1
n 2 n=1 n
For −π < x < π, log(cos(x/2)) is real-valued. We equate the real parts of the equation on this domain to
obtain the Fourier series of log(cos(x/2)).
∞
 x X (−1)n
log cos = − log 2 − cos(nx), −π < x < π.
2 n=1
n

The domain of convergence for this series is =(x) = 0, x 6= (2k + 1)π. The Fourier series converges to the
periodic extension of the function.
∞
 x  X (−1)n
log cos  = − log 2 − cos(nx), x 6= (2k + 1)π, k ∈ Z

2 n=1
n

2. Now we integrate the function from 0 to π.

∞
Z π Z π !
 x X (−1)n
log cos dx = − log 2 − cos(nx) dx
0 2 0 n=1
n
∞ π
(−1)n
X Z
= −π log 2 − cos(nx) dx
n=1
n 0
∞ π
(−1)n sin(nx)
X 
= −π log 2 − dx
n=1
n n 0

1265
 Z π  x
log cos dx = −π log 2
0 2

3.
 
1  sin((x + ξ)/2)  1
 = log |sin((x + ξ)/2)| − 1 log |sin((x − ξ)/2)|
log 
2 sin((x − ξ)/2)  2 2

Consider the function log | sin(y/2)|. Since sin(x) = cos(x − π/2), we can use the result of part (a) to obtain,
 
 y  y − π 
log sin  = log cos
  
2 2 
∞
X (−1)n
= − log 2 − cos(n(y − π))
n=1
n
∞
X 1
= − log 2 − cos(ny), for y 6= 2πk, k ∈ Z.
n=1
n

We return to the original function:

∞ ∞
  !
1  sin((x + ξ)/2)  1 X 1 X 1
log 
 = − log 2 − cos(n(x + ξ)) + log 2 + cos(n(x − ξ)) ,
2 sin((x − ξ)/2)  2 n=1
n n=1
n

for x ± ξ 6= 2πk, k ∈ Z.
  ∞
1  sin((x + ξ)/2)  X sin nx sin nξ
log 
  = , x 6= ±ξ + 2kπ.
2 sin((x − ξ)/2) 
n=1
n

1266
Solution 30.12
The eigenfunction problem associated with this problem is
φ00 + λ2 φ = 0, φ(a) = φ(b) = 0,
which has the solutions,
 
nπ nπ(x − a)
λn = , φn = sin , n ∈ N.
b−a b−a
We expand the solution and the inhomogeneity in the eigenfunctions.
∞  
X nπ(x − a)
y(x) = yn sin
n=1
b−a

∞   b  
nπ(x − a) nπ(x − a)
Z
X 2
f (x) = fn sin , fn = f (x) sin dx
n=1
b−a b−a a b−a

Since the solution y(x) satisfies the same homogeneous boundary conditions as the eigenfunctions, we can differ-
entiate the series. We substitute the series expansions into the differential equation.
y 00 + αy = f (x)
∞
X X ∞
−λ2n + α sin (λn x) =


yn fn sin (λn x)
n=1 n=1
fn
yn =
α − λ2n
Thus the solution of the problem has the series representation,
∞  
X nπ(x − a)
λ2n


y(x) = α− sin .
n=1
b−a

1267
Solution 30.13
The eigenfunction problem associated with this problem is
φ00 + λ2 φ = 0, φ(a) = φ(b) = 0,
which has the solutions,
 
nπ nπ(x − a)
λn = , φn = sin , n ∈ N.
b−a b−a
We expand the solution and the inhomogeneity in the eigenfunctions.
∞  
X nπ(x − a)
y(x) = yn sin
n=1
b−a
∞   b  
nπ(x − a) nπ(x − a)
Z
X 2
f (x) = fn sin , fn = f (x) sin dx
n=1
b−a b−a a b−a
Since the solution y(x) does not satisfy the same homogeneous boundary conditions as the eigenfunctions, we can
differentiate the series. We multiply the differential equation by an eigenfunction and integrate from a to b. We
use integration by parts to move derivatives from y to the eigenfunction.
y 00 + αy = f (x)
Z b Z b Z b
00
y (x) sin(λm x) dx + α y(x) sin(λm x) dx = f (x) sin(λm x) dx
a a a
Z b
0 b b−a b−a
[y sin(λm x)]a − y 0 λm cos(λm x) dx + α ym = fm
a 2 2
Z b
b b−a b−a
− [yλm cos(λm x)]a − yλ2m sin(λm x) dx + α ym = fm
a 2 2
b−a b−a
−Bλm (−1)m + Aλm (−1)m+1 − λ2m ym + α ym = fm
2 2
fm + (−1)m λm (A + B)
ym =
α − λ2m

1268
Thus the solution of the problem has the series representation,

∞
fm + (−1)m λm (A + B)
 
X nπ(x − a)
y(x) = sin .
n=1
α − λ2m b−a

Solution 30.14
1.

1
A + iB =
1 − z2
X∞
= z 2n
n=0
X∞
= r2n ei2nx
n=0
∞
X ∞
X
2n
= r cos(2nx) + i r2n sin(2nx)
n=0 n=1

∞
X ∞
X
2n
A= r cos(2nx), B= r2n sin(2nx)
n=0 n=1

1269
 1
A + iB =
1 − z2
1
=
1 − r2 e2ix
1
=
1 − r2 cos(2x) − ir2 sin(2x)
1 − r2 cos(2x) + ir2 sin(2x)
=
(1 − r2 cos(2x))2 + (r2 sin(2x))2

1 − r2 cos(2x) r2 sin(2x)
A= , B=
1 − 2r2 cos(2x) + r4 1 − 2r2 cos(2x) + r4

2. We consider the principal branch of the logarithm.

A + iB = log(1 + z)
∞
X (−1)n+1 n
= z
n=1
n
∞
X (−1)n+1
= rn einx
n=1
n
∞
X (−1)n+1 n

= r cos(nx) + i sin(nx)
n=1
n

∞ ∞
X (−1)n+1 n
X (−1)n+1
A= r cos(nx), B= rn sin(nx)
n=1
n n=1
n

1270
 A + iB = log(1 + z)
= log 1 + r eix

= log (1 + r cos x + ir sin x)

= log |1 + r cos x + ir sin x| + i arg (1 + r cos x + ir sin x)
p
= log (1 + r cos x)2 + (r sin x)2 + i arctan (1 + r cos x, r sin x)

1
log 1 + 2r cos x + r2 ,


A= B = arctan (1 + r cos x, r sin x)
2

3.
n
X
An + iBn = zk
k=1
1 − z n+1
=
1−z
1 − rn+1 ei(n+1)x
=
1 − r eix
1 − r e−ix − rn+1 ei(n+1)x + rn+2 einx
=
1 − 2r cos x + r2

1 − r cos x − rn+1 cos((n + 1)x) + rn+2 cos(nx)

An =
1 − 2r cos x + r2

r sin x − rn+1 sin((n + 1)x) + rn+2 sin(nx)

Bn =
1 − 2r cos x + r2

1271
 n
X
An + iBn = zk
k=1
Xn
= rk eikx
k=1

n
X n
X
k
An = r cos(kx), Bn = rk sin(kx)
k=1 k=1

Solution 30.15
1.
Z π
1 · sin x dx = [− cos x]π0 = 2
0

Thus the system is not orthogonal on the interval [0, π]. Consider the interval [a, a + π].
Z a+π
1 · sin x dx = [− cos x]a+π
a = 2 cos a
Za a+π
1 · cos x dx = [sin x]a+π
a = −2 sin a
a

Since there is no value of a for which both cos a and sin a vanish, the system is not orthogonal for any
interval of length π.

2. First note that

Z π
cos nx dx = 0 for n ∈ N.
0

1272
If n 6= m, n ≥ 1 and m ≥ 0 then
Z π
1 π
Z


cos nx cos mx dx = cos((n − m)x) + cos((n + m)x) dx = 0
0 2 0
Thus the set {1, cos x, cos 2x, . . . } is orthogonal on [0, π]. Since
Z π
dx = π
Z π 0
π
cos2 (nx) dx = ,
0 2
the set
(r r r )
1 2 2
, cos x, cos 2x, . . .
π π π

is orthonormal on [0, π].

If n 6= m, n ≥ 1 and m ≥ 1 then
Z π
1 π
Z


sin nx sin mx dx = cos((n − m)x) − cos((n + m)x) dx = 0
0 2 0
Thus the set {sin x, sin 2x, . . . } is orthogonal on [0, π]. Since
Z π
π
sin2 (nx) dx = ,
0 2
the set
(r r )
2 2
sin x, sin 2x, . . .
π π

is orthonormal on [0, π].

1273
Solution 30.16
Since the periodic extension of |x| in [−π, π] is an even function its Fourier series is a cosine series. Because of the
anti-symmetry about x = π/2 we see that except for the constant term, there will only be odd cosine terms. Since
the periodic extension is a continuous function, but has a discontinuous first derivative, the Fourier coefficients
will decay as 1/n2 .

∞
X
|x| = an cos(nx), for x ∈ [−π, π]
n=0

π  π
1 x2
Z
1 π
a0 = x dx = =
π 0 π 2 0 2

2 π
Z
an = x cos(nx) dx
π 0
π
2 π sin(nx)
 Z
2 sin(nx)
= x − dx
π n 0 π 0 n
 π
2 cos(nx)
=−
π n2 0
2
= − 2 (cos(nπ) − 1)
πn
2(1 − (−1)n )
=
πn2

∞
π 4 X 1
|x| = + cos(nx) for x ∈ [−π, π]
2 π n=1 n2
odd n

1274
Define RN (x) = f (x) − SN (x). We seek an upper bound on |RN (x)|.
 
 ∞

4 X 1 
|RN (x)| =  cos(nx)
 
π n 2 
 n=N +1 
odd n
∞
4 X 1
≤
π n=N +1 n2
odd n
∞ N
4 X 1 4 X 1
= −
π n=1
n2 π n=1 n2
odd n odd n

Since
∞
X 1 π2
=
n=1
n2 8
odd n

We can bound the error with,

N
π 4 X 1
|RN (x)| ≤ − .
2 π n=1 n2
odd n

N = 7 is the smallest number for which our error bound is less than 10−1 . N ≥ 7 is sufficient to make the error
less that 0.1.
 
π 4 1 1 1
|R7 (x)| ≤ − 1+ + + ≈ 0.079
2 π 9 25 49
N ≥ 7 is also necessary because.
∞
4 X 1
|RN (0)| = .
π n=N +1 n2
odd n

1275
Solution 30.17
1.

∞
X
1∼ an sin(nx), 0≤x≤π
n=1

Since the odd periodic extension of the function is discontinuous, the Fourier coefficients will decay as 1/n.
Because of the symmetry about x = π/2, there will be only odd sine terms.

2 π
Z
an = 1 · sin(nx) dx
π 0
2
= (− cos(nπ) + cos(0))
nπ
2
= (1 − (−1)n )
nπ

∞
4 X sin(nx)
1∼
π n=1 n
odd n

2. It’s always OK to integrate a Fourier series term by term. We integrate the series in part (a).

x ∞ Z
4 X x sin(nξ)
Z
1 dx ∼ dx
a π n=1 a n
odd n
∞
4 X cos(na) − cos(nx)
x−a∼
π n=1
n2
odd n

1276
 Since the series converges uniformly, we can replace the ∼ with =.

∞ ∞
4 X cos(na) 4 X cos(nx)
x−a= −
π n=1 n2 π n=1 n2
odd n odd n

Now we have a Fourier cosine series. The first sum on the right is the constant term. If we choose a = π/2
this sum vanishes since cos(nπ/2) = 0 for odd integer n.

∞
π 4 X cos(nx)
x= −
2 π n=1 n2
odd n

3. If f (x) has the Fourier series

∞
a0 X
f (x) ∼ + (an cos(nx) + bn sin(nx)),
2 n=1

then Parseval’s theorem states that

Z π ∞
2 π 2 X
f (x) dx = a0 + π (a2n + b2n ).
−π 2 n=1

1277
 We apply this to the Fourier sine series from part (a).
Z π ∞  2
2
X 4
f (x) dx = π
−π n=1
πn
odd n
Z 0 Z π ∞
2 2 16 X 1
(−1) dx + (1) dx =
−π 0 π n=1 (2n − 1)2
∞
X 1 π2
=
n=1
(2n − 1)2 8

We substitute x = π in the series from part (b) to corroborate the result.

∞
π 4 X cos((2n − 1)x)
x= −
2 π n=1 (2n − 1)2
∞
π 4 X cos((2n − 1)π)
π= −
2 π n=1 (2n − 1)2
∞
X 1 π2
=
n=1
(2n − 1)2 8

Solution 30.18
1.
∞
X
f (x) ∼ a0 + an cos(nx)
n=1

Since the periodic extension of the function is discontinuous, the Fourier coefficients will decay like 1/n.
Because of the anti-symmetry about x = π/2, there will be only odd cosine terms.
1 π
Z
1
a0 = f (x) dx =
π 0 2

1278
 2 π
Z
an = f (x) cos(nx) dx
π 0
2 π/2
Z
= cos(nx) dx
π 0
2
= sin(nπ/2)
πn
(
2
πn
(−1)(n−1)/2 , for odd n
=
0 for even n

The Fourier cosine series of f (x) is

∞
1 2 X (−1)n
f (x) ∼ + cos((2n + 1)x).
2 π n=0 2n + 1

2. The N th partial sum is

N
1 2 X (−1)n
SN (x) = + cos((2n + 1)x).
2 π n=0 2n + 1

We wish to evaluate the sum from part (a). First we make the change of variables y = x − π/2 to get rid

1279
of the (−1)n factor.

∞
X (−1)n
cos((2n + 1)x)
n=0
2n + 1
N
X (−1)n
= cos((2n + 1)(y + π/2))
n=0
2n + 1
N
X (−1)n
= (−1)n+1 sin((2n + 1)y)
n=0
2n + 1
N
X 1
=− sin((2n + 1)y)
n=0
2n + 1

1280
We write the summand as an integral and interchange the order of summation and integration to get rid of
the 1/(2n + 1) factor.
XN Z y
=− cos((2n + 1)t) dt
n=0 0
Z N
yX
=− cos((2n + 1)t) dt
0 n=0
2N +1 N
!
Z y X X
=− cos(nt) − cos(2nt) dt
0 n=1 n=1
2N +1 N
!
Z y X X
=− < eint − ei2nt dt
0 n=1 n=1
y it i(2N +2)t
ei2t − ei2(N +1)t
 
e −e
Z
=− < − dt
0 1 − eit 1 − ei2t
Z y  it
( e − ei2(N +1)t )(1 − ei2t ) − ( ei2t − ei2(N +1)t )(1 − eit )

=− < dt
0 (1 − eit )(1 − ei2t )
Z y  it
e − ei2t + ei(2N +4)t − ei(2N +3)t

=− < dt
0 (1 − eit )(1 − ei2t )
Z y  it
e − ei(2N +3)t

=− < dt
0 1 − ei2t
Z y  i(2N +2)t 
e −1
=− < dt
0 eit − e−it
Z y 
−i ei2(N +1)t + i

=− < dt
0 2 sin t
1 y sin(2(N + 1)t)
Z
=− dt
2 0 sin t
1 x−π/2 sin(2(N + 1)t)
Z
=− dt
2 0 sin t
1281
 Now we have a tidy representation of the partial sum.
Z x−π/2
1 1 sin(2(N + 1)t)
SN (x) = − dt
2 π 0 sin t

dSN (x)
3. We solve dx
= 0 to find the relative extrema of SN (x).
0
SN (x) = 0
1 sin(2(N + 1)(x − π/2))
− =0
π sin(x − π/2)
(−1)N +1 sin(2(N + 1)x)
=0
− cos(x)
sin(2(N + 1)x)
=0
cos(x)
nπ
x = xn = , n = 0, 1, . . . , N, N + 2, . . . , 2N + 2
2(N + 1)

Note that xN +1 = π/2 is not a solution as the denominator vanishes there. The function has a removable
singularity at x = π/2 with limiting value (−1)N .

4.
πN
Z −π/2
1 1 2(N +1) sin(2(N + 1)t)
SN (xN ) = − dt
2 π 0 sin t

We note that the integrand is even.

πN
Z
2(N +1)
−π/2 Z − 2(Nπ+1) Z π
2(N +1)
= =−
0 0 0

1282
 Z π
1 1 2(N +1) sin(2(N + 1)t)
SN (xN ) = + dt
2 π 0 sin t

5. We make the change of variables 2(N + 1)t → t.

Z π
1 1 sin(t)
SN (xN ) = + dt
2 π 0 2(N + 1) sin(t/(2(N + 1)))

Note that

sin(t) t cos(t)
lim = lim =t
→0  →0 1

Z π
1 1 sin(t)
SN (xN ) → + dt ≈ 1.0895 as N → ∞
2 π 0 t

This is not equal to the limiting value of f (x), f (π/2 − 0) = 1.

Solution 30.19
With the parametrization in t, x(t) and y(t) are continuous functions on the range [0, 2π]. Since the curve is
closed, we have x(0) = x(2π) and y(0) = y(2π). This means that the periodic extensions of x(t) and y(t) are
continuous functions. Thus we can differentiate their Fourier series. First we define formal Fourier series for x(t)

1283
and y(t).

∞
a0 X

x(t) = + an cos(nt) + bn sin(nt)
2 n=1
∞
c0 X

y(t) = + cn cos(nt) + dn sin(nt)
2 n=1
∞
X
0


x (t) = nbn cos(nt) − nan sin(nt)
n=1
∞
X
y 0 (t) =


ndn cos(nt) − ncn sin(nt)
n=1

In this problem we will be dealing with integrals on [0, 2π] of products of Fourier series. We derive a general
formula for later use.
∞
Z 2π Z 2π ! ∞
!
a0 X
c0 X

xy dt = + an cos(nt) + bn sin(nt) + cn cos(nt) + dn sin(nt) dt
0 0 2 n=1
2 n=1
∞
Z 2π !
a0 c0 X
an cn cos2 (nt) + bn dn sin2 (nt)


= + dt
0 4 n=1
∞
!
1 X
=π a0 c 0 + (an cn + bn dn )
2 n=1

In the arclength parametrization we have

 2  2
dx dy
+ = 1.
ds ds

1284
In terms of t = 2πs/L this is
 2  2  2
dx dy L
+ = .
dt dt 2π

We integrate this identity on [0, 2π].

2π 2 2 !
L2
Z  
dx dy
= + dt
2π 0 dt dt
∞ ∞
!
X
X
2 2
(ndn )2 + (−ncn )2


=π (nbn ) + (−nan ) +
n=1 n=1
∞
X
=π n2 (a2n + b2n + c2n + d2n )
n=1

∞
X
L2 = 2π 2 n2 (a2n + b2n + c2n + d2n )
n=1

We assume that the curve is parametrized so that the area is positive. (Reversing the orientation changes the
sign of the area as defined above.) The area is
Z 2π
dy
A= x dt
0 dt
∞
Z 2π ! ∞ !
a0 X
X

= + an cos(nt) + bn sin(nt) ndn cos(nt) − ncn sin(nt) dt
0 2 n=1 n=1
∞
X
=π n(an dn − bn cn )
n=1

1285
 1 2
Now we find an upper bound on the area. We will use the inequality |ab| ≤ 2
|a + b2 |, which follows from
expanding (a − b)2 ≥ 0.
∞
πX
n a2n + b2n + c2n + d2n


A≤
2 n=1
∞
πX 2 2
n an + b2n + c2n + d2n


≤
2 n=1

We can express this in terms of the perimeter.

L2
=
4π

L2 ≥ 4πA

Now we determine the curves for which L2 = 4πA. To do this we find conditions for which A is equal to the
upper bound we obtained for it above. First note that
∞
X ∞
X
a2n b2n c2n d2n n2 a2n + b2n + c2n + d2n



n + + + =
n=1 n=1

implies that all the coefficients except a0 , c0 , a1 , b1 , c1 and d1 are zero. The constraint,
∞ ∞
X πX
n a2n + b2n + c2n + d2n


π n(an dn − bn cn ) =
n=1
2 n=1

then becomes

a1 d1 − b1 c1 = a21 + b21 + c21 + d21 .

1286
This implies that d1 = a1 and c1 = −b1 . a0 and c0 are arbitrary. Thus curves for which L2 = 4πA have the
parametrization
a0 c0
x(t) = + a1 cos t + b1 sin t, y(t) = − b1 cos t + a1 sin t.
2 2
Note that
 a0 2  c0 2
x(t) − + y(t) − = a21 + b21 .
2 2
p
The curve is a circle of radius a21 + b21 and center (a0 /2, c0 /2).

Solution 30.20
1. The Fourier sine series has the form
∞
X
x(1 − x) = an sin(nπx).
n=1

The norm of the eigenfunctions is

Z 1
1
sin2 (nπx) dx = .
0 2
The coefficients in the expansion are
Z 1
an = 2 x(1 − x) sin(nπx) dx
0
2
= (2 − 2 cos(nπ) − nπ sin(nπ))
π 3 n3
4
= (1 − (−1)n ).
π 3 n3

1287
Thus the Fourier sine series is
∞ ∞
8 X sin(nπx) 8 X sin((2n − 1)πx)
x(1 − x) = 3 = 3 .
π n=1 n3 π n=1 (2n − 1)3
odd n

The Fourier cosine series has the form

∞
X
x(1 − x) = an cos(nπx).
n=0

The norm of the eigenfunctions is

Z 1 Z 1
2 1
1 dx = 1, cos2 (nπx) dx = .
0 0 2
The coefficients in the expansion are
Z 1
1
a0 = x(1 − x) dx = ,
0 6
Z 1
an = 2 x(1 − x) cos(nπx) dx
0
2 4 sin(nπ) − nπ cos(nπ)
=− +
π 2 n2 π 3 n3
2
= − 2 2 (1 + (−1)n )
π n
Thus the Fourier cosine series is
∞ ∞
1 4 X cos(nπx) 1 1 X cos(2nπx)
x(1 − x) = − 2 = − 2 .
6 π n=1 n2 6 π n=1 n2
even n

1288
 0.2 0.2
0.1 0.1

-1 -0.5 0.5 1 -1 -0.5 0.5 1

-0.1 -0.1
-0.2 -0.2

Figure 30.13: The odd and even periodic extension of x(1 − x), 0 ≤ x ≤ 1.

The Fourier sine series converges to the odd periodic extension of the function. Since this function is C 1 ,
continuously differentiable, we know that the Fourier coefficients must decay as 1/n3 . The Fourier cosine
series converges to the even periodic extension of the function. Since this function is only C 0 , continuous,
the Fourier coefficients must decay as 1/n2 . The odd and even periodic extensions are shown in Figure 30.13.
The sine series is better because of the faster convergence of the series.

2. (a) We substitute x = 0 into the cosine series.

∞
1 1 X 1
0= − 2
6 π n=1 n2
∞
X 1 π2
=
n=1
n2 6

1289
(b) We substitute x = 1/2 into the cosine series.
∞
1 1 1 X cos(nπ)
= − 2
4 6 π n=1 n2
∞
X (−1)n π2
=−
n=1
n2 12

(c) We substitute x = 1/2 into the sine series.

∞
1 8 X sin((2n − 1)π/2)
= 3
4 π n=1 (2n − 1)3
∞
X (−1)n π3
= −
n=1
(2n − 1)3 32

1290
Chapter 31

Regular Sturm-Liouville Problems

I learned there are troubles

Of more than one kind.
Some come from ahead
And some come from behind.

But I’ve bought a big bat.

I’m all ready, you see.
Now my troubles are going
To have troubles with me!
-I Had Trouble in Getting to Solla Sollew
-Theodor S. Geisel, (Dr. Suess)

31.1 Derivation of the Sturm-Liouville Form

Consider the eigenvalue problem on the finite interval [a, b]
p2 (x)y 00 + p1 (x)y 0 + p0 (x)y = µy,

1291
subject to the homogeneous unmixed boundary conditions

α1 y(a) + α2 y 0 (a) = 0, β1 y(b) + β2 y 0 (b) = 0.

Here the pj ’s are real and continuous and p2 > 0 on the interval [a, b]. The αj ’s and βj ’s are real. (Note that if
p2 were negative we could multiply the equation by (−1) and replace µ by −µ.)
We would like to write this problem in a form that can be used to obtain qualitative information about the
problem. First we will write the operator in self-adjoint form. Since p2 is positive on the interval,
p1 0 p0 µ
y 00 + y + y = y.
p2 p2 p2
Multiplying by the factor
Z x 
p1
exp dξ = eP (x)
p2

yields
 
P (x) 00 p1 0 p0 µ
e y + y + y = eP (x) y
p2 p2 p2
0 p0 µ
eP (x) y 0 + eP (x) y = eP (x) y.


p2 p2
We define the following functions
p0 1
p = eP (x) , q = eP (x) , σ = eP (x) , λ = −µ.
p2 p2
Since the pj ’s are continuous and p2 is positive, p, q, and σ are continuous. p and σ are positive functions. The
problem now has the form

(py 0 )0 + qy + λσy = 0,

1292
subject to the boundary conditions

α1 y(a) + α2 y 0 (a) = 0, β1 y(b) + β2 y 0 (b) = 0.

This is known as a Regular Sturm-Liouville problem. We will devote much of this chapter to studying the
properties of this problem. We will encounter many results that are analogous to the properties of self-adjoint
eigenvalue problems.

Example 31.1.1

 
d dy
log x + λxy = 0, y(1) = y(2) = 0
dx dx

is not a regular Sturm-Liouville problem since log x vanishes at x = 1.

Result 31.1.1 Any eigenvalue problem of the form

p2 y 00 + p1 y 0 + p0 y = µy, for a ≤ x ≤ b,
α1 y(a) + α2 y 0 (a) = 0, β1 y(b) + β2 y 0 (b) = 0,

where the pj ’s are real and continuous, p2 > 0 on [a, b], and the αj ’s and βj ’s are real can
be written in the form of a regular Sturm-Liouville problem

(py 0 )0 + qy + λσy = 0, on a ≤ x ≤ b,
α1 y(a) + α2 y 0 (a) = 0, β1 y(b) + β2 y 0 (b) = 0.

1293
31.2 Properties of Regular Sturm-Liouville Problems
Self-Adjoint. Writing the problem in the form
L[y] = (py 0 )0 + qy = −λσy,
we see that the operator is formally self-adjoint. Now to see if the problem is self-adjoint.
hv|L[u]i − hL[v]|ui = hv|(pu0 )0 + qui − h(pv 0 )0 + qv|ui
= [vpu0 ]ba − hv 0 |pu0 i + hv|qui − [pv 0 u]ba + hpv 0 |u0 i − hqv|ui
= [vpu0 ]ba − [pv 0 u]ba
= p(b) v(b)u0 (b) − v 0 (b)u(b) + p(a) v(a)u0 (a) − v 0 (a)u(a)



     
β1 β1
= p(b) v(b) − u(b) − − v(b)u(b)
β2 β2
     
α1 α1
+ p(a) v(a) − u(a) − − v(a)u(a)
α2 α2
=0
Note that αi and βi are real so
       
α1 α1 β1 β1
= , =
α2 α2 β2 β2
Thus L[y] subject to the boundary conditions is self-adjoint.

Real Eigenvalues. Let λ be an eigenvalue with the eigenfunction φ. Starting with Green’s formula,
hφ|L[φ]i − hL[φ]|φi = 0
hφ| − λσφi − h−λσφ|φi = 0
−λhφ|σ|φi + λhφ|σ|φi = 0
(λ − λ)hφ|σ|φi = 0.

1294
Since hφ|σ|φi > 0, λ − λ = 0. Thus the eigenvalues are real.

Infinite Number of Eigenvalues. There are an infinite of eigenvalues which have no finite cluster point. This
result is analogous to the result that we derived for self-adjoint eigenvalue problems. When we cover the Rayleigh
quotient, we will find that there is a least eigenvalue. Since the eigenvalues are distinct and have no finite cluster
point, λn → ∞ as n → ∞. Thus the eigenvalues form an ordered sequence,

λ1 < λ2 < λ3 < · · · .

Orthogonal Eigenfunctions. Let λ and µ be two distinct eigenvalues with the eigenfunctions φ and ψ. Green’s
formula states

hψ|L[φ]i − hL[ψ]|φi = 0.
hψ| − λσφi − h−µσψ|φi = 0
−λhψ|σ|φi + µhψ|σ|φi = 0
(µ − λ)hψ|σ|φi = 0

Since the eigenvalues are distinct, hψ|σ|φi = 0. Thus eigenfunctions corresponding to distinct eigenvalues are
orthogonal with respect to σ.

Unique Eigenfunctions. Let λ be an eigenvalue. Suppose φ and ψ are two independent eigenfunctions corre-
sponding to λ. The eigenfunctions satisfy the equations

L[φ] + λσφ = 0
L[ψ] + λσψ = 0.

1295
Taking the difference of ψ times the first equation and φ times the second equation gives us

ψL[φ] − φL[ψ] = 0
ψ(pφ0 )0 − φ(pψ 0 )0 = 0
(p(ψφ0 − ψ 0 φ))0 = 0
p(ψφ0 − ψ 0 φ) = const.

In order to satisfy the boundary conditions, the constant must be zero.

p(ψφ0 − ψ 0 φ) = 0

Since p > 0,

ψφ0 − ψ 0 φ = 0
φ0 ψ 0 φ
− 2 =0
ψ ψ
 
d φ
=0
dx ψ
φ
= const.
ψ
φ and ψ are not independent. Thus each eigenvalue has a unique, (to within a multiplicative constant),
eigenfunction.

Real Eigenfunctions. If λ is an eigenvalue with eigenfunction φ, then

(pφ0 )0 + qφ + λσφ = 0.

Taking the complex conjugate of this equation,

 0 0
pφ + qφ + λσφ = 0.

1296
Thus φ is also an eigenfunction corresponding to λ. Are φ and φ independent functions, or do they just differ by
a multiplicative constant? (For example, eix and e−ix are independent functions, but ix and −ix are dependent.)
From our argument on unique eigenfunctions, we see that
φ = (const)φ.
Since φ and φ only differ by a multiplicative constant, the eigenfunctions can be chosen so that they are real-valued
functions.

Rayleigh’s Quotient. Let λ be an eigenvalue with the eigenfunction φ.

hφ|L[φ]i = hφ| − λσφi
hφ|(pφ0 )0 + qφi = −λhφ|σ|φi
b
φpφ0 − hφ0 |p|φ0 i + hφ|q|φi = −λhφ|σ|φi

a
b
− pφφ0 a + hφ0 |p|φ0 i − hφ|q|φi

λ=
hφ|σ|φi

This is known as Rayleigh’s quotient. It is useful for obtaining qualitative information about the eigenvalues.

Minimum Property of Rayleigh’s Quotient. Note that since p, q, σ and φ are bounded functions, the
Rayleigh quotient is bounded below. Thus there is a least eigenvalue. If we restrict u to be a real continuous
function that satisfies the boundary conditions, then
−[puu0 ]ba + hu0 |p|u0 i − hu|q|ui
λ1 = min ,
u hu|σ|ui
where λ1 is the least eigenvalue. This form allows us to get upper and lower bounds on λ1 .
To derive this formula, we first write it in terms of the operator L.
−hu|L[u]i
λ1 = min
u hu|σ|ui

1297
Since u is continuous and satisfies the boundary conditions, we can expand u in a series of the eigenfunctions.

P∞  P∞ 
hu|L[u]i n=1 cn φ n
L [
m=1 cm φm ]
− = −
P∞   P∞ 
hu|σ|ui n=1 cn φn σ
 
m=1 cm φm

P∞  P∞ 
c φ −
n=1 n n m=1 cm λm σφm
= −
P∞   P∞ 
n=1 cn φn σ m=1 cm φm
 

Assuming that we can interchange summation and integration,

P∞ P∞
n=1 P m=1 cn cm λn hφm |σ|φn i
= P ∞ ∞
n=1 m=1 cn cm hφm |σ|φn i
P∞ 2
n=1 |cn | λn hφn |σ|φn i
= P ∞
|cn |2 hφn |σ|φn i
Pn=1
∞
|cn |2 hφn |σ|φn i
≤ λ1 Pn=1
∞ 2
n=1 |cn | hφn |σ|φn i
= λ1 .

We see that the minimum value of Rayleigh’s quotient is λ1 . The minimum is attained when cn = 0 for all n ≥ 2,
that is, when u = c1 φ1 .

Completeness. The set of the eigenfunctions of a regular Sturm-Liouville problem is complete. That is, any
piecewise continuous function defined on [a, b] can be expanded in a series of the eigenfunctions
∞
X
f (x) ∼ cn φn (x),
n=1

where the cn are the generalized Fourier coefficients

hφn |σ|f i
cn = .
hφn |σ|φn i

1298
Here the sum is convergent in the mean. For any fixed x, the sum converges to 12 (f (x− ) + f (x+ )). If f (x) is
continuous and satisfies the boundary conditions, then the convergence is uniform.

1299
Result 31.2.1 Properties of regular Sturm-Liouville problems.
• The eigenvalues λ are real.
• There are an infinite number of eigenvalues

λ1 < λ2 < λ3 < · · · .

There is a least eigenvalue λ1 but there is no greatest eigenvalue, (λn → ∞ as

n → ∞).
• For each eigenvalue, there is one unique, (to within a multiplicative constant), eigen-
function φn . The eigenfunctions can be chosen to be real-valued. (Assume the φn
following are real-valued.) The eigenfunction φn has exactly n − 1 zeros in the open
interval a < x < b.
• The eigenfunctions are orthogonal with respect to the weighting function σ(x).
Z b
6 m.
φn (x)φm (x)σ(x) dx = 0 if n =
a

• The eigenfunctions are complete. Any piecewise continuous function f (x) defined
on a ≤ x ≤ b can be expanded in a series of eigenfunctions
∞
X
f (x) ∼ cn φn (x),
n=1

where
Rb
f (x)φn (x)σ(x) dx
cn = a
R b 1300 .
2
a φn (x)σ(x) dx

The sum converges to 21 (f (x− ) + f (x+ )).

• The eigenvalues can be related to the eigenfunctions with a formula known as the
Rayleigh quotient.
b Z b    2 
−pφn dφ
dx a +
n
p dφ n
− qφ2n dx

dx
a
λn = Rb
2
a φn σ dx
Example 31.2.1 A simple example of a Sturm-Liouville problem is
 
d dy
+ λy = 0, y(0) = y(π) = 0.
dx dx

Bounding The Least Eigenvalue. The Rayleigh quotient for the first eigenvalue is
Rπ 0 2
(φ ) dx
λ1 = 0R π 12 .
0
φ1 dx
Rπ
Immediately we see that the eigenvalues are non-negative. If 0 (φ01 )2 dx = 0 then φ = (const). The only constant
that satisfies the boundary conditions is φ = 0. Since the trivial solution is not an eigenfunction, λ = 0 is not an
eigenvalue. Thus all the eigenvalues are positive.
Now we get an upper bound for the first eigenvalue.
Rπ 0 2
(u ) dx
λ1 = min R0 π 2
u
0
u dx

where u is continuous and satisfies the boundary conditions. We choose u = x(x − π) as a trial function.
Rπ 0 2
(u ) dx
λ1 ≤ R0 π 2
u dx
R π0
(2x − π)2 dx
= R π0 2
0
(x − πx)2 dx
π 3 /3
= 5
π /30
10
= 2
π
≈ 1.013

1301
Finding the Eigenvalues and Eigenfunctions. We consider the cases of negative, zero, and positive eigen-
values to check our results above.

λ < 0. The general solution is

√ √
−λx
y = ce + d e− −λx
.

The only solution that satisfies the boundary conditions is the trivial solution, y = 0. Thus there are no
negative eigenvalues.

λ = 0. The general solution is

y = c + dx.

Again only the trivial solution satisfies the boundary conditions, so λ = 0 is not an eigenvalue.

λ > 0. The general solution is

√ √
y = c cos( λx) + d sin( λx).

Applying the boundary conditions,

y(0) = 0 ⇒ c=0
√
y(π) = 0 ⇒ d sin( λπ) = 0

The nontrivial solutions are

√
λ = n = 1, 2, 3, . . . y = d sin(nπ).

Thus the eigenvalues and eigenfunctions are

λn = n 2 , φn = sin(nx), for n = 1, 2, 3, . . .

1302
 We can verify that this example satisfies all the properties listed in Result 31.2.1. Note that there are an
infinite number of eigenvalues. There is a least eigenvalue λ1 = 1 but there is no greatest eigenvalue. For each
eigenvalue, there is one eigenfunction. The nth eigenfunction sin(nx) has n − 1 zeroes in the interval 0 < x < π.
Since a series of the eigenfunctions is the familiar Fourier sine series, we know that the eigenfunctions are
orthogonal and complete. Checking Rayleigh’s quotient,
π Z π 
2 
dφn 
−pφn dx  + p dφ
dx
n
− qφ 2
n dx
0 0
λn = R π 2
0
φn σ dx
π Z π  2 
d sin(nx)  d sin(nx)
− sin(nx) dx  + dx
dx
0 0
= Rπ 2
0
sin (nx)dx
Rπ 2
n cos2 (nx) dx
= 0
π/2
2
=n .

Example 31.2.2 Consider the eigenvalue problem

x2 y 00 + xy 0 + y = µy, y(1) = y(2) = 0.
Since x2 > 0 on [1, 2], we can write this problem in terms of a regular Sturm-Liouville eigenvalue problem.
Dividing by x2 ,
1 1
y 00 + y 0 + 2 (1 − µ)y = 0.
x x
Rx 1
We multiply by the factor exp( ξ dξ) = exp(log x) = x and make the substitution, λ = 1 − µ to obtain the
Sturm-Liouville form
1
xy 00 + y 0 + λ y = 0
x
1
(xy 0 )0 + λ y = 0.
x

1303
We see that the eigenfunctions will be orthogonal with respect to the weighting function σ = 1/x.
From the Rayleigh quotient,
b
− pφφ0 a + hφ0 |x|φ0 i

λ=
hφ| x1 |φi
hφ0 |x|φ0 i
= .
hφ| x1 |φi

If φ0 = 0, then only the trivial solution, φ = 0, satisfies the boundary conditions. Thus the eigenvalues λ are
positive.
Returning to the original problem We see that the eigenvalues, µ, satisfy µ < 1. Since this is an Euler equation,
the substitution y = xα yields

α(α − 1) + α + 1 − µ = 0
α2 + 1 − µ = 0.

Since µ < 1,
p
α = ±i 1 − µ.

The general solution is

√ √
y = c1 xi 1−µ
+ c2 x−i 1−µ
.

We know that the eigenfunctions can be written as real functions. We can rewrite the solution as
√ √
y = c1 ei 1−µ log x
+ c2 e−i 1−µ log x
.

An equivalent form is
p p
y = c1 cos( 1 − µ log x) + c2 sin( 1 − µ log x).

1304
Applying the boundary conditions,

y(1) = 0 ⇒ c1 = 0
p
y(2) = 0 ⇒ sin( 1 − µ log 2) = 0
p
⇒ 1 − µ log 2 = nπ, for n = 1, 2, . . .

Thus the eigenvalues and eigenfunctions are

 2  
nπ log x
µn = 1 − , φn = sin nπ for n = 1, 2, . . .
log 2 log 2

31.3 Solving Differential Equations With Eigenfunction Expansions

Linear Algebra. Consider the eigenvalue problem,

Ax = λx.

If the matrix A has a complete, orthonormal set of eigenvectors {ξ k } with eigenvalues {λk } then we can represent
any vector as a linear combination of the eigenvectors.
n
X
y= ak ξ k , ak = ξ k · y
k=1
n
X
y= (ξ k · y) ξ k
k=1

This property allows us to solve the inhomogeneous equation

Ax − µx = b. (31.1)

1305
Before we try to solve this equation, we should consider the existence/uniqueness of the solution. If µ is not an
eigenvalue, then the range of L ≡ A − µ is Rn . The problem has a unique solution. If µ is an eigenvalue, then the
null space of L is the span of the eigenvectors of µ. That is, if µ = λi , then nullspace (L) = span (ξ i1 , ξ i2 , . . . , ξ im ).
({ξ i1 , ξ i2 , . . . , ξ im } are the eigenvalues of λi .) If b is orthogonal to nullspace (L) then Equation 31.1 has a solution,
but it is not unique. If y is a solution then we can add any linear combination of {ξ ij } to obtain another solution.
Thus the solutions have the form
m
X
x=y+ cj ξij .
j=1

If b is not orthogonal to nullspace (L) then Equation 31.1 has no solution.

Now we solve Equation 31.1. We assume that µ is not an eigenvalue. We expand the solution x and the
inhomogeneity in the orthonormal eigenvectors.
n
X n
X
x= ak ξ k , b= bk ξ k
k=1 k=1

We substitute the expansions into Equation 31.1.

n
X n
X n
X
A ak ξ k − µ ak ξ k = bk ξ k
k=1 k=1 k=1
n
X n
X n
X
ak λk ξ k − µ ak ξ k = bk ξ k
k=1 k=1 k=1
bk
ak =
λk − µ
The solution is
n
X bk
x= ξ .
k=1
λk − µ k

1306
Inhomogeneous Boundary Value Problems. Consider the self-adjoint eigenvalue problem,

Ly = λy, a < x < b,

B1 [y] = B2 [y] = 0.

If the problem has a complete, orthonormal set of eigenfunctions {φk } with eigenvalues {λk } then we can represent
any square-integrable function as a linear combination of the eigenfunctions.

X Z b
f= fk φk , fk = hφk |f i = φk (x)f (x) dx
k a
X
f= hφk |f iφk
k

This property allows us to solve the inhomogeneous differential equation

Ly − µy = f, a < x < b, (31.2)

B1 [y] = B2 [y] = 0.

Before we try to solve this equation, we should consider the existence/uniqueness of the solution. If µ is not
an eigenvalue, then the range of L − µ is the space of square-integrable functions. The problem has a unique
solution. If µ is an eigenvalue, then the null space of L is the span of the eigenfunctions of µ. That is, if µ = λi ,
then nullspace (L) = span (φi1 , φi2 , . . . , φim ). ({φi1 , φi2 , . . . , φim } are the eigenvalues of λi .) If f is orthogonal to
nullspace (L − µ) then Equation 31.2 has a solution, but it is not unique. If u is a solution then we can add any
linear combination of {φij } to obtain another solution. Thus the solutions have the form
m
X
y =u+ cj φij .
j=1

If f is not orthogonal to nullspace (L − µ) then Equation 31.2 has no solution.

1307
 Now we solve Equation 31.2. We assume that µ is not an eigenvalue. We expand the solution y and the
inhomogeneity in the orthonormal eigenfunctions.
X X
y= yk φk , f= fk φk
k k

It would be handy if we could substitute the expansions into Equation 31.2. However, the expansion of a function
is not necessarily differentiable. Thus we demonstrate that since y is C 2 (a . . . b) and satisfies the boundary
conditions B1 [y] = B2 [y] = 0, we are justified in substituting it into the differential equation. In particular, we
will show that
" #
X X X
L[y] = L yk φk = yk L [φk ] = y k λk φ k .
k k k
2
To do this we will use Green’s identity. If u and v are C (a . . . b) and satisfy the boundary conditions B1 [y] =
B2 [y] = 0 then
hu|L[v]i = hL[u]|vi.
First we assume that we can differentiate y term-by-term.
X
L[y] = y k λk φ k
k

Now we directly expand L[y] and show that we get the same result.
X
L[y] = ck φk
k

ck = hφk |L[y]i
= hL[φk ]|yi
= hλk φk |yi
= λk hφk |yi
= λk yk

1308
 X
L[y] = yk λφk
k

The series representation of y may not be differentiable, but we are justified in applying L term-by-term.
Now we substitute the expansions into Equation 31.2.
" #
X X X
L yk φk − µ yk φk = fk φk
k k k
X X X
λk y k φ k − µ yk φk = fk φk
k k k
fk
yk =
λk − µ
The solution is
X fk
y= φk
k
λk − µ

Consider a second order, inhomogeneous problem.

L[y] = f (x), B1 [y] = b1 , B2 [y] = b2

We will expand the solution in an orthogonal basis.

X
y= an φn
n

We would like to substitute the series into the differential equation, but in general we are not allowed to differentiate
such series. To get around this, we use integration by parts to move derivatives from the solution y, to the φn .

Example 31.3.1 Consider the problem,

y 00 + αy = f (x), y(0) = a, y(π) = b,

1309
where α 6= n2 , n ∈ Z+ . We expand the solution in a cosine series.
∞ r
y0 X 2
y(x) = √ + yn cos(nx)
π n=1 π

We also expand the inhomogeneous term.

∞ r
f0 X 2
f (x) = √ + fn cos(nx)
π n=1 π

We multiply the differential

√ equation by the orthonormal functions and integrate over the interval. We neglect
the special case φ0 = 1/ π for now.
Z πr Z πr Z πr
2 2 2
cos(nx)y 00 dx + α cos(nx)y dx = f (x) dx
0 π 0 π 0 π
"r #π Z r
π
2 2
cos(nx)y 0 (x) + n sin(nx)y 0 (x) dx + αyn = fn
π 0 π
r "r 0 #π Z r
π
2 2 2 2
((−1)n y 0 (π) − y 0 (0)) + n sin(nx)y(x) − n cos(nx)y(x) dx + αyn = fn
π π 0 π
0
r
2
((−1)n y 0 (π) − y 0 (0)) − n2 yn + αyn = fn
π

Unfortunately we don’t know the values of y 0 (0) and y 0 (π).

CONTINUE HERE

1310
31.4 Exercises
Exercise 31.1
Find the eigenvalues and eigenfunctions of
y 00 + 2αy 0 + λy = 0, y(a) = y(b) = 0,
where a < b.
Write the problem in Sturm Liouville form. Verify that the eigenvalues and eigenfunctions satisfy the properties
of regular Sturm-Liouville problems. Find the coefficients in the expansion of an arbitrary function f (x) in a series
of the eigenfunctions.
Hint, Solution

Exercise 31.2
Find the eigenvalues and eigenfunctions of the boundary value problem
λ
y 00 + y=0
(z + 1)2
on the interval 1 ≤ z ≤ 2 with boundary conditions y(1) = y(2) = 0. Discuss how the results confirm the concepts
presented in class relating to boundary value problems of this type.
Hint, Solution

Exercise 31.3
Find the eigenvalues and eigenfunctions of
2α + 1 0 λ
y 00 + y + 2 y = 0, y(a) = y(b) = 0,
x x
where 0 < a < b. Write the problem in Sturm Liouville form. Verify that the eigenvalues and eigenfunctions
satisfy the properties of regular Sturm-Liouville problems. Find the coefficients in the expansion of an arbitrary
function f (x) in a series of the eigenfunctions.
Hint, Solution

1311
Exercise 31.4
Find the eigenvalues and eigenfunctions of

y 00 − y 0 + λy = 0, y(0) = y(1) = 0.

Find the coefficients in the expansion of an arbitrary, f (x), in a series of the eigenfunctions.
Hint, Solution

Exercise 31.5
Find the eigenvalues and eigenfunctions for,

y 00 + λy = 0, y(0) = 0, y(1) + y 0 (1) = 0.

Show that the transcendental equation for λ has infinitely many roots λ1 < λ2 < λ3 < · · · . Find the limit of λn
as n → ∞. How is this approached?
Hint, Solution

Exercise 31.6
Consider

y 00 + y = f (x) y(0) = 0 y(1) + y 0 (1) = 0.

Find the eigenfunctions for this problem and the equation which the eigenvalues satisfy. Give the general solution
in terms of these eigenfunctions.
Hint, Solution

Exercise 31.7
Show that the eigenvalue problem,

y 00 + λy = 0, y(0) = 0, y 0 (0) − y(1) = 0,

1312
(note the mixed boundary condition), has only one real eigenvalue. Find it and the corresponding eigenfunction.
Show that this problem is not self-adjoint. Thus the proof, valid for unmixed, homogeneous boundary conditions,
that all eigenvalues are real fails in this case.
Hint, Solution

Exercise 31.8
Determine the Rayleigh quotient, R[φ] for,
1
y 00 + y 0 + λy = 0, |y(0)| < ∞, y(1) = 0.
x
Use the trial function φ = 1 − x in
√ R[φ] to deduce that the smallest zero of J0 (x), the Bessel function of the first
kind and order zero, is less than 6.
Hint, Solution

Exercise 31.9
Discuss the eigenvalues of the equation
y 00 + λq(z)y = 0, y(0) = y(π) = 0
where
(
a > 0, 0 ≤ z ≤ l
q(z) =
b > 0, l < z ≤ π.
This is an example that indicates that the results we obtained in class for eigenfunctions and eigenvalues with
q(z) continuous and bounded also hold if q(z) is simply integrable; that is
Z π
|q(z)| dz
0

is finite.
Hint, Solution

1313
Exercise 31.10
1. Find conditions on the smooth real functions p(x), q(x), r(x) and s(x) so that the eigenvalues, λ, of:

Lv ≡ (p(x)v 00 (x))00 − (q(x)v 0 (x))0 + r(x)v(x) = λs(x)v(x), a<x<b

v(a) = v 00 (a) = 0
v 00 (b) = 0, p(b)v 000 (b) − q(b)v 0 (b) = 0

are positive. Prove the assertion.

2. Show that for any smooth p(x), q(x), r(x) and s(x) the eigenfunctions belonging to distinct eigenvalues are
orthogonal relative to the weight s(x). That is:
Z b
vm (x)vk (x)s(x) dx = 0 if λk 6= λm .
a

3. Find the eigenvalues and eigenfunctions for:

d4 φ nφ(0) = φ00 (0) = 0,

= λφ,
dx4 φ(1) = φ00 (1) = 0.

Hint, Solution

1314
31.5 Hints
Hint 31.1

Hint 31.2

Hint 31.3

Hint 31.4
Write the problem in Sturm-Liouville form to show that the eigenfunctions are orthogonal with respect to the
weighting function σ = e−x .

Hint 31.5
Note that the solution is a regular Sturm-Liouville problem and thus the eigenvalues are real. Use the Rayleigh
quotient to show that there are only positive eigenvalues. Informally show that there are an infinite number of
eigenvalues with a graph.

Hint 31.6

Hint 31.7
Find the solution for λ = 0, λ < 0 and λ > 0. A problem is self-adjoint if it satisfies Green’s identity.

1315
Hint 31.8
Write the equation in self-adjoint form. The Bessel equation of the first kind and order zero satisfies the problem,

1
y 00 + y 0 + y = 0, |y(0)| < ∞, y(r) = 0,
x
where r is a positive root of J0 (x). Make the change of variables ξ = x/r, u(ξ) = y(x).

Hint 31.9

Hint 31.10

1316
31.6 Solutions
Solution 31.1
Recall that constant coefficient equations are shift invariant. If u(x) is a solution, then so is u(x − c).
We substitute y = eγx into the constant coefficient equation.
y 00 + 2αy 0 + λy = 0
γ 2 + 2αγ + λ = 0
√
γ = −α ± α2 − λ
First we consider the case λ = α2 . A set of solutions of the differential equation is
 −αx
e , x e−αx

The homogeneous solution that satisfies the left boundary condition y(a) = 0 is
y = c(x − a) e−αx .
Since only the trivial solution with c = 0 satisfies the right boundary condition, λ = α2 is not an eigenvalue.
Next we consider the case λ 6= α2 . We write
√
γ = −α ± i λ − α2 .
√
Note that <( λ − α2 ) ≥ 0. A set of solutions of the differential equation is
n √ o
(−α±i λ−α2 )x
e

By taking the sum and difference of these solutions we obtain a new set of linearly independent solutions.
n √  √ o
−αx 2 −αx 2
e cos λ−α x , e sin λ−α x

The solution which satisfies the left boundary condition is

√ 
−αx 2
y = ce sin λ − α (x − a) .

1317
For nontrivial solutions, the right boundary condition y(b) = 0 imposes the constraint
√ 
−αb 2
e sin λ − α (b − a) = 0
√
λ − α2 (b − a) = nπ, n ∈ Z

We have the eigenvalues

 2
2 nπ
λn = α + , n∈Z
b−a

with the eigenfunctions

 
−αx x−a
φn = e sin nπ .
b−a

To write the problem in Sturm-Liouville form, we multiply by the integrating factor

R
2α dx
e = e2αx .

0
e2αx y 0 + λ e2αx y = 0, y(a) = y(b) = 0

Now we verify that the Sturm-Liouville properties are satisfied.

• The eigenvalues
 2
2 nπ
λn = α + , n∈Z
b−a

are real.

1318
• There are an infinite number of eigenvalues
λ1 < λ2 < λ3 < · · · ,
 2  2  2
π 2π 3π
α2 + 2
<α + <α + 2
< ··· .
b−a b−a b−a
There is a least eigenvalue
 2
2 π
λ1 = α + ,
b−a
but there is no greatest eigenvalue, (λn → ∞ as n → ∞).
• For each eigenvalue, we found one unique, (to within a multiplicative constant), eigenfunction φn . We were
able to choose the eigenfunctions to be real-valued. The eigenfunction
 
−αx x−a
φn = e sin nπ .
b−a
has exactly n − 1 zeros in the open interval a < x < b.
• The eigenfunctions are orthogonal with respect to the weighting function σ(x) = e2ax .
Z b Z b    
−αx x−a −αx x−a
φn (x)φm (x)σ(x) dx = e sin nπ e sin mπ e2ax dx
a a b − a b − a
Z b    
x−a x−a
= sin nπ sin mπ dx
a b−a b−a
b−a π
Z
= sin(nx) sin(mx) dx
π 0
b−a π
Z
= (cos((n − m)x) − cos((n + m)x)) dx
2π 0
= 0 if n 6= m

1319
• The eigenfunctions are complete. Any piecewise continuous function f (x) defined on a ≤ x ≤ b can be
expanded in a series of eigenfunctions
∞
X
f (x) ∼ cn φn (x),
n=1

where
Rb
a
f (x)φn (x)σ(x) dx
cn = Rb .
φ 2 (x)σ(x) dx
a n

The sum converges to 12 (f (x− ) + f (x+ )). (We do not prove this property.)

• The eigenvalues can be related to the eigenfunctions with the Rayleigh quotient.
b R b  dφn
2 
−pφn dφ
 2
n
dx a
+ a
p dx
− qφ n dx
λn = Rb
φ2 σ dx
a n

R b 2αx −αx nπ


2 
x−a x−a


a
e e b−a
cos nπ b−a
− α sin nπ b−a
dx
= Rb 2 2αx
e−αx sin nπ x−a



a b−a
e dx
R b  nπ
2 2 x−a

nπ x−a

x−a

2 2 x−a


a b−a
cos nπ b−a − 2α b−a cos nπ b−a sin nπ b−a + α sin nπ b−a dx
= Rb 2
sin nπ x−a


a b−a
dx
 
nπ 2
Rπ
2 nπ 2 2
0 b−a
cos (x) − 2α b−a
cos(x) sin(x) + α sin (x) dx
= Rπ 2
0
sin (x) dx
 2
nπ
= α2 +
b−a

1320
 Now we expand a function f (x) in a series of the eigenfunctions.
∞  
X
−αx x−a
f (x) ∼ cn e sin nπ ,
n=1
b−a

where
Rb
f (x)φn (x)σ(x) dx
a
cn = Rb
φ2 (x)σ(x) dx
a n
Z b  
2n αx x−a
= f (x) e sin nπ dx
b−a a b−a

Solution 31.2
This is an Euler equation. We substitute y = (z + 1)α into the equation.

λ
y 00 + y=0
(z + 1)2
α(α − 1) + λ = 0
√
1 ± 1 − 4λ
α=
2
First consider the case λ = 1/4. A set of solutions is
n√ √ o
z + 1, z + 1 log(z + 1) .

Another set of solutions is

√ √
  
z+1
z + 1, z + 1 log .
2

1321
The solution which satisfies the boundary condition y(1) = 0 is

√
 
z+1
y = c z + 1 log .
2

Since only the trivial solution satisfies the y(2) = 0, λ = 1/4 is not an eigenvalue.
Now consider the case λ 6= 1/4. A set of solutions is
n √ √ o
(z + 1)(1+ 1−4λ)/2
, (z + 1)(1− 1−4λ)/2
.

We can write this in terms of the exponential and the logarithm.

 √  √
√ √
  
4λ − 1 4λ − 1
z + 1 exp i log(z + 1) , z + 1 exp −i log(z + 1) .
2 2

Note that
 √  √
√ √
    
4λ − 1 z+1 4λ − 1 z+1
z + 1 exp i log , z + 1 exp −i log .
2 2 2 2

is also a set of solutions. The new factor of 2 in the logarithm just multiplies the solutions by a constant. We
write the solution in terms of the cosine and sine.
√ √
√ √
    
4λ − 1 z+1 4λ − 1 z+1
z + 1 cos log , z + 1 sin log .
2 2 2 2

The solution of the differential equation which satisfies the boundary condition y(1) = 0 is
√
√
 
1 − 4λ z+1
y = c z + 1 sin log .
2 2

1322
Now we use the second boundary condition to find the eigenvalues.
y(2) = 0
√  
4λ − 1 3
sin log =0
2 2
√  
4λ − 1 3
log = nπ, n ∈ Z
2 2
 2 !
1 2nπ
λ= 1+ , n∈Z
4 log(3/2)
n = 0 gives us a trivial solution, so we discard it. Discarding duplicate solutions, The eigenvalues and eigenfunc-
tions are
2
√
  
1 nπ log((z + 1)/2)
λn = + , yn = z + 1 sin nπ , n ∈ Z+ .
4 log(3/2) log(3/2)
Now we verify that the eigenvalues and eigenfunctions satisfy the properties of regular Sturm-Liouville prob-
lems.
• The eigenvalues are real.
• There are an infinite number of eigenvalues
λ1 < λ2 < λ3 < · · ·
 2  2  2
1 π 1 2π 1 3π
+ < + < + < ···
4 log(3/2) 4 log(3/2) 4 log(3/2)
There is a least least eigenvalue
 2
1 π
λ1 = + ,
4 log(3/2)
but there is no greatest eigenvalue.

1323
 • The eigenfunctions are orthogonal with respect to the weighting function σ(z) = 1/(z + 1)2 . Let n 6= m.
Z 2
yn (z)ym (z)σ(z) dz
1
√ log((z + 1)/2) √
Z 2    
log((z + 1)/2) 1
= z + 1 sin nπ z + 1 sin mπ dz
1 log(3/2) log(3/2) (z + 1)2
Z π
log(3/2)
= sin(nx) sin(mx) dx
0 π
log(3/2) π
Z
= (cos((n − m)x) − cos((n + m)x)) dx
2π 0
=0

• The eigenfunctions are complete. A function f (x) defined on (1 . . . 2) has the series representation
∞ ∞
√
 
X X log((z + 1)/2)
f (x) ∼ cn yn (x) = cn z + 1 sin nπ ,
n=1 n=1
log(3/2)
where
2
hyn |1/(z + 1)2 |f i
Z  
2 log((z + 1)/2) 1
cn = 2
= sin nπ f (x) dz
hyn |1/(z + 1) |yn i log(3/2) 1 log(3/2) (z + 1)3/2

Solution 31.3
Recall that Euler equations are scale invariant. If u(x) is a solution, then so is u(cx) for any nonzero constant c.
We substitute y = xγ into the Euler equation.
2α + 1 0 λ
y 00 + y + 2y = 0
x x
γ(γ − 1) + (2α + 1)γ + λ = 0
γ 2 + 2αγ + λ = 0
√
γ = −α ± α2 − λ

1324
 First we consider the case λ = α2 . A set of solutions of the differential equation is
 −α −α
x , x log x
The homogeneous solution that satisfies the left boundary condition y(a) = 0 is
x
y = cx−α (log x − log a) = cx−α log .
a
Since only the trivial solution with c = 0 satisfies the right boundary condition, λ = α2 is not an eigenvalue.
Next we consider the case λ 6= α2 . We write
√
γ = −α ± i λ − α2 .
√
Note that <( λ − α2 ) ≥ 0. A set of solutions of the differential equation is
n √ o
2
x−α±i λ−α
n √ o
−α ±i λ−α2 log x
x e .

By taking the sum and difference of these solutions we obtain a new set of linearly independent solutions.
n √  √  o
−α 2 −α 2
x cos λ − α log x , x sin λ − α log x ,

The solution which satisfies the left boundary condition is

√  x 
−α 2
y = cx sin λ − α log .
a
For nontrivial solutions, the right boundary condition y(b) = 0 imposes the constraint
√
  
−α 2
b
b sin λ − α log
a
√
 
b
λ − α2 log = nπ, n ∈ Z
a

1325
We have the eigenvalues
 2
2 nπ
λn = α + , n∈Z
log(b/a)

with the eigenfunctions

 
−α log(x/a)
φn = x sin nπ .
log(b/a)

To write the problem in Sturm-Liouville form, we multiply by the integrating factor

R
(2α+1)/x dx
e = e(2α+1) log x = x2α+1 .

0
x2α+1 y 0 + λx2α−1 y = 0, y(a) = y(b) = 0

Now we verify that the Sturm-Liouville properties are satisfied.

• The eigenvalues
 2
2 nπ
λn = α + , n∈Z
log(b/a)

are real.

• There are an infinite number of eigenvalues

λ1 < λ2 < λ3 < · · · ,

 2  2  2
π 2π 3π
α2 + 2
<α + <α + 2
< ···
log(b/a) log(b/a) log(b/a)

1326
 There is a least eigenvalue
 2
2 π
λ1 = α + ,
log(b/a)
but there is no greatest eigenvalue, (λn → ∞ as n → ∞).
• For each eigenvalue, we found one unique, (to within a multiplicative constant), eigenfunction φn . We were
able to choose the eigenfunctions to be real-valued. The eigenfunction
 
−α log(x/a)
φn = x sin nπ .
log(b/a)
has exactly n − 1 zeros in the open interval a < x < b.
• The eigenfunctions are orthogonal with respect to the weighting function σ(x) = x2α−1 .
Z b Z b    
−α log(x/a) −α log(x/a)
φn (x)φm (x)σ(x) dx = x sin nπ x sin mπ x2α−1 dx
a a log(b/a) log(b/a)
Z b    
log(x/a) log(x/a) 1
= sin nπ sin mπ dx
a log(b/a) log(b/a) x
log(b/a) π
Z
= sin(nx) sin(mx) dx
π 0
log(b/a) π
Z
= (cos((n − m)x) − cos((n + m)x)) dx
2π 0
= 0 if n 6= m

• The eigenfunctions are complete. Any piecewise continuous function f (x) defined on a ≤ x ≤ b can be
expanded in a series of eigenfunctions
∞
X
f (x) ∼ cn φn (x),
n=1

1327
 where
Rb
a
f (x)φn (x)σ(x) dx
cn = Rb .
φ 2 (x)σ(x) dx
a n

The sum converges to 12 (f (x− ) + f (x+ )). (We do not prove this property.)

• The eigenvalues can be related to the eigenfunctions with the Rayleigh quotient.
b R b  dφn
2 
−pφn dφ
 2
n
dx a
+ a
p dx
− qφ n dx
λn = Rb
φ2 σ dx
a n
 2 
R b 2α+1  −α−1  nπ 
log(x/a)
 
log(x/a)
a
x x log(b/a)
cos nπ log(b/a) − α sin nπ log(b/a) dx
= Rb  2
x −α sin nπ log(x/a) x2α−1 dx
a log(b/a)
 
R b  nπ 2
a log(b/a)
cos (·) − 2α log(b/a) cos (·) sin (·) + α sin (·) x−1 dx
2 nπ 2 2

= R b 2  log(x/a) 
a
sin nπ log(b/a) x−1 dx
 
R π  nπ 2 2 nπ 2 2
0 log(b/a)
cos (x) − 2α log(b/a) cos(x) sin(x) + α sin (x) dx
= Rπ 2
0
sin (x) dx
 2
nπ
= α2 +
log(b/a)

Now we expand a function f (x) in a series of the eigenfunctions.

∞  
X
−α log(x/a)
f (x) ∼ cn x sin nπ ,
n=1
log(b/a)

1328
where
Rb
a
f (x)φn (x)σ(x) dx
cn = Rb
φ2 (x)σ(x) dx
a n
Z b  
2n α−1 log(x/a)
= f (x)x sin nπ dx
log(b/a) a log(b/a)

Solution 31.4

y 00 − y 0 + λy = 0, y(0) = y(1) = 0.

The factor that will put this equation in Sturm-Liouville form is

Z x 
F (x) = exp −1 dx = e−x .

The differential equation becomes

d
e−x y 0 + λ e−x y = 0.


dx
Thus we see that the eigenfunctions will be orthogonal with respect to the weighting function σ = e−x .
Substituting y = eαx into the differential equation yields

α2 − α + λ = 0
√
1 ± 1 − 4λ
α=
2
1 p
α = ± 1/4 − λ.
2

1329
 If λ < 1/4 then the solutions to the differential equation are exponential and only the trivial solution satisfies
the boundary conditions.
If λ = 1/4 then the solution is y = c1 ex/2 + c2 x ex/2 and again only the trivial solution satisfies the boundary
conditions.
Now consider the case that λ > 1/4.

1 p
α= ± i λ − 1/4
2

The solutions are

p p
ex/2 cos( λ − 1/4 x), ex/2 sin( λ − 1/4 x).

The left boundary condition gives us

p
y = c ex/2 sin( λ − 1/4 x).

The right boundary condition demands that

p
λ − 1/4 = nπ, n = 1, 2, . . .

Thus we see that the eigenvalues and eigenfunctions are

1
λn = + (nπ)2 , yn = ex/2 sin(nπx).
4

If f (x) is a piecewise continuous function then we can expand it in a series of the eigenfunctions.
∞
X
f (x) = an ex/2 sin(nπx)
n=1

1330
The coefficients are
R1
f (x) e−x ex/2 sin(nπx) dx
an = R0 1
e−x ( ex/2 sin(nπx))2 dx
R 10
f (x) e−x/2 sin(nπx) dx
= 0 R1 2
sin (nπx) dx
Z 1 0
=2 f (x) e−x/2 sin(nπx) dx.
0

Solution 31.5
Since this is a Sturm-Liouville problem, there are only real eigenvalues. By the Rayleigh quotient, the eigenvalues
are
1 R 

1 dφ 2
−φ dφ + 0 dx

dx  dx
0
λ= R1 ,
φ 2 dx
0

R 1  dφ
2 
φ2 (1) + 0 dx
dx
λ= R1 .
φ 2 dx
0

This demonstrates that there are only positive eigenvalues. The general solution of the differential equation for
positive, real λ is
√  √ 
y = c1 cos λx + c2 sin λx .

The solution that satisfies the left boundary condition is

√ 
y = c sin λx .

1331
For nontrivial solutions we must have
√  √ √ 
sin λ + λ cos λ =0

√ √ 
λ = − tan λ .
√ 
The positive solutions of this equation are eigenvalues with corresponding eigenfunctions sin λx . In Fig-
ure 31.1 we plot the functions x and − tan(x) and draw vertical lines at x = (n − 1/2)π, n ∈ N.

Figure 31.1: x and − tan(x).

From this we see that there are an infinite number of eigenvalues, λ1 < λ2 < λ3 < · · · . In the limit as n → ∞,
λn → (n − 1/2)π. The limit is approached from above.

Solution 31.6
Consider the eigenvalue problem
y 00 + y = µy y(0) = 0 y(1) + y 0 (1) = 0.

1332
From Exercise 31.5 we see that the eigenvalues satisfy
p p 
1 − µ = − tan 1−µ

and that there are an infinite number of eigenvalues. For large n, µn ≈ 1 − (n − 1/2)π. The eigenfunctions are
p 
φn = sin 1 − µn x .

To solve the inhomogeneous problem, we expand the solution and the inhomogeneity in a series of the eigen-
functions.
∞ R1
X f (x)φn (x) dx
f= fn φn , fn = 0 R 1
n=1 φ2 (x) dx
0 n
∞
X
y= yn φn
n=1

We substitite the expansions into the differential equation to determine the coefficients.

y 00 + y = f
∞
X X∞
µn yn φn = fn φn
n=1 n=1
∞
X fn p 
y= sin 1 − µn x
µ
n=1 n

Solution 31.7
First consider λ = 0. The general solution is

y = c1 + c2 x.

1333
y = cx satisfies the boundary conditions. Thus λ = 0 is an eigenvalue.
Now consider negative real λ. The general solution is
√  √ 
y = c1 cosh −λx + c2 sinh −λx .

The solution that satisfies the left boundary condition is

√ 
y = c sinh −λx .

For nontrivial solutions of the boundary value problem, there must be negative real solutions of
√ √ 
−λ − sinh −λ = 0.

Since x = sinh x has no nonzero real solutions, this equation has no solutions for negative real λ. There are no
negative real eigenvalues.
Finally consider positive real λ. The general solution is
√  √ 
y = c1 cos λx + c2 sin λx .

The solution that satisfies the left boundary condition is

√ 
y = c sin λx .

For nontrivial solutions of the boundary value problem, there must be positive real solutions of
√ √ 
λ − sin λ = 0.

Since x = sin x has no nonzero real solutions, this equation has no solutions for positive real λ. There are no
positive real eigenvalues.
There is only one real eigenvalue, λ = 0, with corresponding eigenfunction φ = x.

1334
 The difficulty with the boundary conditions, y(0) = 0, y 0 (0) − y(1) = 0 is that the problem is not self-adjoint.
We demonstrate this by showing that the problem does not satisfy Green’s identity. Let u and v be two functions
that satisfy the boundary conditions, but not necessarily the differential equation.

hu, L[v]i − hL[u], vi = hu, v 00 i − hu00 , vi

1 1
= [uv 0 ]0 − hu0 , v 0 i − hu0 , v 0 i − [u0 v]0 + hu0 , v 0 i − hu0 , v 0 i
= u(1)v 0 (1) − u0 (1)v(1)

Green’s identity is not satisfied,

hu, L[v]i − hL[u], vi 6= 0;

The problem is not self-adjoint.

Solution 31.8
First we write the equation in formally self-adjoint form,

L[y] ≡ (xy 0 )0 = −λxy, |y(0)| < ∞, y(1) = 0.

Let λ be an eigenvalue with corresponding eigenfunction φ. We derive the Rayleigh quotient for λ.

hφ, L[φ]i = hφ, −λxφi

hφ, (xφ0 )0 i = −λhφ, xφi
1
[φxφ0 ]0 − hφ0 , xφ0 i = −λhφ, xφi

We apply the boundary conditions and solve for λ.

hφ0 , xφ0 i
λ=
hφ, xφi

1335
The Bessel equation of the first kind and order zero satisfies the problem,

1
y 00 + y 0 + y = 0, |y(0)| < ∞, y(r) = 0,
x
where r is a positive root of J0 (x). We make the change of variables ξ = x/r, u(ξ) = y(x) to obtain the problem

1 00 1 1 0
2
u + u + u = 0, |u(0)| < ∞, u(1) = 0,
r rξ r

1
u00 + u0 + r2 u = 0, |u(0)| < ∞, u(1) = 0.
ξ

Now r2 is the eigenvalue of the problem for u(ξ). From the Rayleigh quotient, the minimum eigenvalue obeys the
inequality

hφ0 , xφ0 i
r2 ≤ ,
hφ, xφi

where φ is any test function that satisfies the boundary conditions. Taking φ = 1 − x we obtain,
R1
(−1)x(−1) dx
r2 ≤ R 1 0
= 6,
0
(1 − x)x(1 − x) dx

√
r≤ 6
√
Thus the smallest zero of J0 (x) is less than or equal to 6 ≈ 2.4494. (The smallest zero of J0 (x) is approximately
2.40483.)

1336
Solution 31.9
We assume that 0 < l < π.
Recall that the solution of a second order differential equation with piecewise continuous coefficient functions
is piecewise C 2 . This means that the solution is C 2 except for a finite number of points where it is C 1 .
First consider the case λ = 0. A set of linearly independent solutions of the differential equation is {1, z}. The
solution which satisfies y(0) = 0 is y1 = c1 z. The solution which satisfies y(π) = 0 is y2 = c2 (π − z). There is a
solution for the problem if there are there are values of c1 and c2 such that y1 and y2 have the same position and
slope at z = l.
y1 (l) = y2 (l), y10 (l) = y20 (l)
c1 l = c2 (π − l), c1 = −c2
Since there is only the trivial solution, c1 = c2 = 0, λ = 0 is not an eigenvalue.
Now consider λ 6= 0. For 0 ≤ z ≤ l a set of linearly independent solutions is
n √ √ o
cos( aλz), sin( aλz) .

The solution which satisfies y(0) = 0 is

√
y1 = c1 sin( aλz).
For l < z ≤ π a set of linearly independent solutions is
n √ √ o
cos( bλz), sin( bλz) .

The solution which satisfies y(π) = 0 is

√
y2 = c2 sin( bλ(π − z)).
λ 6= 0 is an eigenvalue if there are nontrivial solutions of
y1 (l) = y2 (l), y10 (l) = y20 (l)
√ √ √ √ √ √
c1 sin( aλl) = c2 sin( bλ(π − l)), c1 aλ cos( aλl) = −c2 bλ cos( bλ(π − l))

1337
 p
We divide the second equation by (λ) since λ 6= 0 and write this as a linear algebra problem.
 √ √    
sin( √aλl) √ − sin( √bλ(π − l)) c1 0
√ =
a cos( aλl) b sin( bλ(π − l)) c2 0

This system of equations has nontrivial solutions if and only if the determinant of the matrix is zero.
√ √ √ √ √ √
b sin( aλl) sin( bλ(π − l)) + a cos( aλl) sin( bλ(π − l)) = 0

We can use trigonometric identities to write this equation as

√ √ √ √ √  √ √ √ √ √ 
( b − a) sin λ(l a − (π − l) b) + ( b + a) sin λ(l a + (π − l) b) = 0

Clearly this equation has an infinite number of solutions for real, positive λ. However, it is not clear that this
equation does not have non-real solutions. In order to prove that, we will show that the problem is self-adjoint.
Before going on to that we note that the eigenfunctions have the form
( √

sin aλn z 0≤z≤l
φn (z) = √

sin bλn (π − z) l < z ≤ π.

Now we prove that the problem is self-adjoint. We consider the class of functions which are C 2 in (0 . . . π)
except at the interior point x = l where they are C 1 and which satisfy the boundary conditions y(0) = y(π) = 0.
Note that the differential operator is not defined at the point x = l. Thus Green’s identity,

hu|q|Lvi = hLu|q|vi

is not well-defined. To remedy this we must define a new inner product. We choose
Z l Z π
hu|vi ≡ uv dx + uv dx.
0 l

1338
This new inner product does not require differentiability at the point x = l.
The problem is self-adjoint if Green’s indentity is satisfied. Let u and v be elements of our class of functions.
In addition to the boundary conditions, we will use the fact that u and v satisfy y(l− ) = y(l+ ) and y 0 (l− ) = y 0 (l+ ).

Z l Z π
00
hv|Lui = vu dx + vu00 dx
0 l
Z l Z π
0 l 0 0 0 π
= [vu ]0 − v u dx + [vu ]l − v 0 u0 dx
0 l
Z l Z π
0 0 0 0
= v(l)u (l) − v u dx − v(l)u (l) − v 0 u0 dx
0 l
Z l Z π
=− v 0 u0 dx − v 0 u0 dx
0 l
Z l Z π
0 l 00 0 π
= − [v u]0 + v u dx − [v u]l + v 00 u dx
0 l
Z l Z π
0 00 0
= −v (l)u(l) + v u dx + v (l)u(l) + v 00 u dx
0 l
Z l Z π
= v 00 u dx + v 00 u dx
0 l
= hLv|Lui

The problem is self-adjoint. Hence the eigenvalues are real. There are an infinite number of positive, real
eigenvalues λn .

Solution 31.10
1. Let v be an eigenfunction with the eigenvalue λ. We start with the differential equation and then take the

1339
 inner product with v.

(pv 00 )00 − (qv 0 )0 + rv = λsv

hv, (pv 00 )00 − (qv 0 )0 + rvi = hv, λsvi

We use integration by parts and utilize the homogeneous boundary conditions.

b b
[v(pv 00 )0 ]a − hv 0 , (pv 00 )0 i − [vqv 0 ]a + hv 0 , qv 0 i + hv, rvi = λhv, svi
b
− [v 0 pv 00 ]a + hv 00 , pv 00 i + hv 0 , qv 0 i + hv, rvi = λhv, svi
hv 00 , pv 00 i + hv 0 , qv 0 i + hv, rvi
λ=
hv, svi

We see that if p, q, r, s ≥ 0 then the eigenvalues will be positive. (Of course we assume that p and s are not
identically zero.)

2. First we prove that this problem is self-adjoint. Let u and v be functions that satisfy the boundary conditions,
but do not necessarily satsify the differential equation.

hv, L[u]i − hL[v], ui = hv, (pu00 )00 − (qu0 )0 + rui − h(pv 00 )00 − (qv 0 )0 + rv, ui

Following our work in part (a) we use integration by parts to move the derivatives.

= (hv 00 , pu00 i + hv 0 , qu0 i + hv, rui) − (hpv 00 , u00 i + hqv 0 , u0 i + hrv, ui)
=0

This problem satisfies Green’s identity,

hv, L[u]i − hL[v], ui = 0,

and is thus self-adjoint.

1340
 Let vk and vm be eigenfunctions corresponding to the distinct eigenvalues λk and λm . We start with Green’s
identity.

hvk , L[vm ]i − hL[vk ], vm i = 0

hvk , λm svm i − hλk svk , vm i = 0
(λm − λk )hvk , svm i = 0
hvk , svm i = 0

The eigenfunctions are orthogonal with respect to the weighting function s.

3. From part (a) we know that there are only positive eigenvalues. The general solution of the differential
equation is

φ = c1 cos(λ1/4 x) + c2 cosh(λ1/4 x) + c3 sin(λ1/4 x) + c4 sinh(λ1/4 x).

Applying the condition φ(0) = 0 we obtain

φ = c1 (cos(λ1/4 x) − cosh(λ1/4 x)) + c2 sin(λ1/4 x) + c3 sinh(λ1/4 x).

The condition φ00 (0) = 0 reduces this to

φ = c1 sin(λ1/4 x) + c2 sinh(λ1/4 x).

We substitute the solution into the two right boundary conditions.

c1 sin(λ1/4 ) + c2 sinh(λ1/4 ) = 0
−c1 λ1/2 sin(λ1/4 ) + c2 λ1/2 sinh(λ1/4 ) = 0

We see that sin(λ1/4 ) = 0. The eigenvalues and eigenfunctions are

λn = (nπ)4 , φn = sin(nπx), n ∈ N.

1341
Chapter 32

Integrals and Convergence

Never try to teach a pig to sing. It wastes your time and annoys the pig.

-?

32.1 Uniform Convergence of Integrals

Consider the improper integral
Z ∞
f (x, t) dt.
c

The integral is convergent to S(x) if, given any  > 0, there exists T (x, ) such that
Z τ 
 

 f (x, t) dt − S(x) <  for all τ > T (x, ).

c

The sum is uniformly convergent if T is independent of x.

1342
 Similar to the Weierstrass M-test for infinite sums we have a uniform
R∞ convergence test for integrals.
R∞ If there
exists a continuous function M (t) such that |f (x, t)| ≤ M (t) and c M (t) dt is convergent, then c f (x, t) dt is
uniformly convergent.
R∞
If c f (x, t) dt is uniformly convergent, we have the following properties:

• If f (x, t) is continuous for x ∈ [a, b] and t ∈ [c, ∞) then for a < x0 < b,
Z ∞ Z ∞  
lim f (x, t) dt = lim f (x, t) dt.
x→x0 c c x→x0

• If a ≤ x1 < x2 ≤ b then we can interchange the order of integration.

Z x2 Z ∞  Z ∞ Z x2 
f (x, t) dt dx = f (x, t) dx dt
x1 c c x1

∂f
• If ∂x
is continuous, then
Z ∞ Z ∞
d ∂
f (x, t) dt = f (x, t) dt.
dx c c ∂x

32.2 The Riemann-Lebesgue Lemma

Rb
Result 32.2.1 If a |f (x)| dx exists, then
Z b
f (x) sin(λx) dx → 0 as λ → ∞.
a

1343
 Before we try to justify the Riemann-Lebesgue lemma, we will need a preliminary result. Let λ be a positive
constant.
Z b   b 
   1
sin(λx) dx =  − cos(λx) 
 

a  λ a

2
≤ .
λ
We will prove the Riemann-Lebesgue lemma for the case when f (x) has limited total fluctuation on the interval
(a, b). We can express f (x) as the difference of two functions

f (x) = ψ+ (x) − ψ− (x),

where ψ+ and ψ− are positive, increasing, bounded functions.

From the mean value theorem for positive, increasing functions, there exists an x0 , a ≤ x0 ≤ b, such that
Z b   Z b 
   

 ψ + (x) sin(λx) dx = ψ+ (b)
  sin(λx) dx

a x0
2
≤ |ψ+ (b)| .
λ
Similarly,
Z b 
  2
 ψ− (x) sin(λx) dx ≤ |ψ− (b)| .

a λ

Thus
Z b 
  2
 f (x) sin(λx) dx ≤ (|ψ+ (b)| + |ψ− (b)|)
  λ
a
→ 0 as λ → ∞.

1344
32.3 Cauchy Principal Value
32.3.1 Integrals on an Infinite Domain
R∞
The improper integral −∞
f (x) dx is defined
Z ∞ Z 0 Z b
f (x) dx = lim f (x) dx + lim f (x) dx,
−∞ a→−∞ a b→∞ 0

when these limits exist. The Cauchy principal value of the integral is defined
Z ∞ Z a
PV f (x) dx = lim f (x) dx.
−∞ a→∞ −a

The principal value may exist when the integral diverges.

R∞
Example 32.3.1 −∞ x dx diverges, but
Z ∞ Z a
PV x dx = lim x dx = lim (0) = 0.
−∞ a→∞ −a a→∞

If the improper integral converges, then the Cauchy principal value exists and is equal to the value of the
integral. The principal value of the integral of an odd function is zero. If the principal value of the integral of an
even function exists, then the integral converges.

32.3.2 Singular Functions

Let f (x) have a singularity at x = 0. Let a and b satisfy a < 0 < b. The integral of f (x) is defined
Z b Z 1 Z b
f (x) dx = lim− f (x) dx + lim+ f (x) dx,
a 1 →0 a 2 →0 2

1345
when the limits exist. The Cauchy principal value of the integral is defined
Z b Z − Z b 
PV f (x) dx = lim+ f (x) dx + f (x) dx ,
a →0 a 

when the limit exists.

Example 32.3.2 The integral

Z 2
1
dx
−1 x

diverges, but the principal value exists.

Z 2 Z − Z 2 
1 1 1
PV dx = lim+ dx + dx
−1 x →0 −1 x  x
 Z 1 Z 2 
1 1
= lim+ − dx + dx
→0  x  x
Z 2
1
= dx
1 x
= log 2

1346
Chapter 33

The Laplace Transform

33.1 The Laplace Transform

The Laplace transform of the function f (t) is defined
Z ∞
L[f (t)] = e−st f (t) dt,
0

for all values of s for which the integral exists. The Laplace transform of f (t) is a function of s which we will
denote fˆ(s). 1
A function f (t) is of exponential order α if there exist constants t0 and M such that

|f (t)| < M eαt , for all t > t0 .

Rt
If 0 0 f (t) dt exists and f (t) is of exponential order α then the Laplace transform F (s) exists for <(s) > α.
Here are a few examples of these concepts.

• sin t is of exponential order 0.

1
Denoting the Laplace transform of f (t) as F (s) is also common.

1347
 • t e2t is of exponential order α for any α > 2.
2
• et is not of exponential order α for any α.
• tn is of exponential order α for any α > 0.
• t−2 does not have a Laplace transform as the integral diverges.

Example 33.1.1 Consider the Laplace transform of f (t) = 1. Since f (t) = 1 is of exponential order α for any
α > 0, the Laplace transform integral converges for <(s) > 0.
Z ∞
ˆ
f (s) = e−st dt
0 ∞
1 −st
= − e
s 0
1
=
s

Example 33.1.2 The function f (t) = t et is of exponential order α for any α > 1. We compute the Laplace
transform of this function.
Z ∞
ˆ
f (s) = e−st t et dt
Z0 ∞
= t e(1−s)t dt
0 ∞ Z ∞
1 (1−s)t 1
= te − e(1−s)t dt
1−s 1 − s
 0
∞ 0
1
=− e(1−s)t
(1 − s)2 0
1
= for <(s) > 1.
(1 − s)2

1348
Example 33.1.3 Consider the Laplace transform of the Heaviside function,
(
0 for t < c
H(t − c) =
1 for t > c,
where c > 0.
Z ∞
L[H(t − c)] = e−st H(t − c) dt
Z0 ∞
= e−st dt
c −st ∞
e
=
−s c
−cs
e
= for <(s) > 0
s

Example 33.1.4 Next consider H(t − c)f (t − c).

Z ∞
L[H(t − c)f (t − c)] = e−st H(t − c)f (t − c) dt
Z0 ∞
= e−st f (t − c) dt
Zc ∞
= e−s(t+c) f (t) dt
0
= e−cs fˆ(s)

33.2 The Inverse Laplace Transform

The inverse Laplace transform in denoted
f (t) = L−1 [fˆ(s)].

1349
We compute the inverse Laplace transform with the Mellin inversion formula.
Z α+i∞
1
f (t) = est fˆ(s) ds
i2π α−i∞

Here α is a real constant that is to the right of the singularities of fˆ(s).

To see why the Mellin inversion formula is correct, we take the Laplace transform of it. Assume that f (t) is
of exponential order α. Then α will be to the right of the singularities of fˆ(s).
 Z α+i∞ 
−1 ˆ 1 zt ˆ
L[L [f (s)]] = L e f (z) dz
i2π α−i∞
Z ∞ Z α+i∞
−st 1
= e ezt fˆ(z) dz dt
0 i2π α−i∞

We interchange the order of integration.

Z α+i∞ Z ∞
1
= fˆ(z) e(z−s)t dt dz
i2π α−i∞ 0

Since <(z) = α, the integral in t exists for <(s) > α.

α+i∞
fˆ(z)
Z
1
= dz
i2π α−i∞ s−z
We would like to evaluate this integral by closing the path of integration with a semi-circle of radius R in the right
half plane and applying the residue theorem. However, in order for the integral along the semi-circle to vanish as
R → ∞, fˆ(z) must vanish as |z| → ∞. If fˆ(z) vanishes we can use the maximum modulus bound to show that
the integral along the semi-circle vanishes. This we assume that fˆ(z) vanishes at infinity.
Consider the integral,
I ˆ
1 f (z)
dz,
i2π C s − z

1350
 Im(z)
α+iR

s
Re(z)

α-iR

Figure 33.1: The Laplace Transform Pair Contour.

where C is the contour that starts at α − iR, goes straight up to α + iR, and then follows a semi-circle back down
to α − iR. This contour is shown in Figure 33.1.
If s is inside the contour then
I ˆ
1 f (z)
dz = fˆ(s).
i2π C s − z
Note that the contour is traversed in the negative direction. Since fˆ(z) decays as |z| → ∞, the semicircular
contribution to the integral will vanish as R → ∞. Thus
Z α+i∞ ˆ
1 f (z)
dz = fˆ(s).
i2π α−i∞ s − z
Therefore, we have shown than
L[L−1 [fˆ(s)]] = F (s).

1351
 f (t) and fˆ(s) are known as Laplace transform pairs.

33.2.1 F(s) with Poles

Example 33.2.1 Consider the inverse Laplace transform of 1/s2 . s = 1 is to the right of the singularity of 1/s2 .
  Z 1+i∞
−1 1 1 1
L 2
= est 2 ds
s i2π 1−i∞ s

Let BR be the contour starting at 1 − iR and following a straight line to 1 + iR; let CR be the contour starting at
1 + iR and following a semicircular path down to 1 − iR. Let C be the combination of BR and CR . This contour
is shown in Figure 33.2.

Im(s)
α+iR
CR
BR
Re(s)

α-iR

Figure 33.2: The Path of Integration for the Inverse Laplace Transform.

1352
 Consider the line integral on C for R > 1.
I  
1 1
st st 1
e 2 ds = Res e 2 , 0
i2π C s s
d st 

= e 
ds s=0
=t

If t ≥ 0, the integral along CR vanishes as R → ∞. We parameterize s.

π 3π
s = 1 + R eiθ , ≤θ≤
2 2
 st   t(1+R eiθ ) 
e  = e  = et etR cos θ ≤ et

Z  Z  
 1   1 
 est 2 ds ≤  est  ds
 s2 

CR s CR
1
≤ πR et
(R − 1)2
→ 0 as R → ∞

Thus the inverse Laplace transform of 1/s2 is

 
−1 1
L = t, for t ≥ 0.
s2

Let fˆ(s) be analytic except for isolated poles at s1 , s2 , . . . , sN and let α be to the right of these poles. Also,
let fˆ(s) → 0 as |s| → ∞. Define BR to be the straight line from α − iR to α + iR and CR to be the semicircular

1353
path from α + iR to α − iR. If R is large enough to enclose all the poles, then
I N
1 st
X
e F (s) ds = Res ( est F (s), sn )
i2π BR +CR n=1
Z N Z
1 st
X
st 1
e F (s) ds = Res ( e F (s), sn ) − est F (s) ds.
i2π BR n=1
i2π CR

Now let’s examine the integral along CR . Let the maximum of |fˆ(s)| on CR be MR . We can parameterize the
contour with s = α + R eiθ , π/2 < θ < 3π/2.
Z  Z 3π/2 

t(α+R eiθ ) ˆ
 st
 iθ iθ
e F (s) ds =  e f (α + R e )Ri e dθ
   

CR  π/2 
Z 3π/2
≤ eαt etR cos θ RMR dθ
π/2
Z π
= RMR e αt
e−tR sin θ dθ
0

If t ≥ 0 we can use Jordan’s Lemma to obtain,

π
< RMR eαt .
tR
αt π
= MR e
t

We use that MR → 0 as R → ∞.

→ 0 as R → ∞

1354
Thus we have an expression for the inverse Laplace transform of fˆ(s).
Z α+i∞ N
1 X
e fˆ(s) ds =
st
Res ( est fˆ(s), sn )
i2π α−i∞ n=1
N
X
L−1 [fˆ(s)] = Res ( est fˆ(s), sn )
n=1

Result 33.2.1 If fˆ(s) is analytic except for poles at s1 , s2 , . . . , sN and fˆ(s) → 0 as

|s| → ∞ then the inverse Laplace transform of fˆ(s) is
N
X
f (t) = L [fˆ(s)] =
−1
Res ( est fˆ(s), sn ), for t > 0.
n=1

1
Example 33.2.2 Consider the inverse Laplace transform of s3 −s2
.
First we factor the denominator.
1 1 1
= 2 .
s3 −s 2 s s−1
Taking the inverse Laplace transform,
     
−1 1 st 1 1 st 1 1
L = Res e 2 , 0 + Res e 2 ,1
s3 − s3 s s−1 s s−1
d est 

= + et
ds s − 1 s=0
−1 t
= 2
+ + et
(−1) −1

1355
Thus we have that
 
−1 1
L = et − t − 1, for t > 0.
s3 − s2

Example 33.2.3 Consider the inverse Laplace transform of

s2 + s − 1
.
s3 − 2s2 + s − 2

We factor the denominator.

s2 + s − 1
.
(s − 2)(s − i)(s + i)

Then we take the inverse Laplace transform.

s2 + s − 1 s2 + s − 1 s2 + s − 1
     
−1 st st
L = Res e , 2 + Res e ,i
s3 − 2s2 + s − 2 (s − 2)(s − i)(s + i) (s − 2)(s − i)(s + i)
s2 + s − 1
 
st
+ Res e , −i
(s − 2)(s − i)(s + i)
1 −1
= e2t + eit + e−it
i2 i2
Thus we have

s2 + s − 1
 
−1
L 3 2
= sin t + e2t , for t > 0.
s − 2s + s − 2

1356
33.2.2 fˆ(s) with Branch Points
√
Example 33.2.4 Consider the inverse Laplace transform of √1 . s denotes the principal branch of s1/2 . There
s
is a branch cut from s = 0 to s = −∞ and

1 e−iθ/2
√ = √ , for − π < θ < π.
s r

Let α be any positive number. The inverse Laplace transform of √1 is

s

Z α+i∞
1 1
f (t) = est √ ds.
i2π α−i∞ s

We will evaluate the integral by deforming it to wrap around the branch cut. Consider the integral on the contour
shown in Figure 33.3. CR+ and CR− are circular arcs of radius R. B is the vertical line at <(s) = α joining the two
arcs. C is a semi-circle in the right half plane joining i and −i. L+ and L− are lines joining the circular arcs
at =(s) = ±.
Since there are no residues inside the contour, we have
Z Z Z Z Z Z !
1 1
+ + + + + est √ ds = 0.
i2π B +
CR L+ C L− −
CR s

We will evaluate the inverse Laplace transform for t > 0.

First we will show that the integral along CR+ vanishes as R → ∞. As  → 0, we have
Z Z π/2 Z π
· · · ds = · · · dθ + · · · dθ.
+
CR π/2−δ π/2

The first integral vanishes by the maximum modulus bound. Note that the length of the path of integration is

1357
 CR+
B

π/2−δ
L+ Cε
L- −π/2+δ

CR-

√
Figure 33.3: Path of Integration for 1/ s

less than 2α.

Z  !
 π/2  
 st 1 
· · · dθ ≤ max  e √  (2α)
 
s

 π/2−δ  s∈C + R

1
= eαt √ (2α)
R
→ 0 as R → ∞

1358
The second integral vanishes by Jordan’s Lemma. A parameterization of CR+ is s = R eiθ .
Z π  Z π  
 iθ
Re t 1   iθ
 eR e t √ 1 
 e √ dθ ≤  dθ
R eiθ R eiθ
   
π/2 π/2
Z π
1
≤√ eR cos(θ)t dθ
R π/2
Z π/2
1
≤√ e−Rt sin(φ) dφ
R 0
1 π
<√
R 2Rt
→ 0 as R → ∞

We could show that the integral along CR− vanishes by the same method. Now we have
Z Z Z Z 
1 1
+ + + est √ ds = 0.
i2π B L+ C L− s

We can show that the integral along C vanishes as  → 0 with the maximum modulus bound.
Z    
 st 1   st 1 
 e √ ds ≤ max  e √  (π)
 sC s∈C s
1
< et √ π

→ 0 as  → 0

Now we can express the inverse Laplace transform in terms of the integrals along L+ and L− .
Z α+i∞ Z Z
1 1 st 1 st1 1 1
f (t) ≡ e √ ds = − e √ ds − est √ ds
i2π α−i∞ s i2π L+ s i2π L− s

1359
On L+ , s = r eiπ , ds = eiπ dr = −dr; on L− , s = r e−iπ , ds = e−iπ dr = −dr. We can combine the integrals along
the top and bottom of the branch cut.
Z 0 Z ∞
1 −rt −i 1 i
f (t) = − e √ (−1) dr − e−rt √ (−1) dr
i2π ∞ r i2π 0 r
Z ∞
1 i2
= e−rt √ dr
i2π 0 r

We make the change of variables x = rt.

Z ∞
1 1
= √ e−x √ dx
π t 0 x

We recognize this integral as Γ(1/2).

1
= √ Γ(1/2)
π t
1
=√
πt
Thus the inverse Laplace transform of √1 is
s

1
f (t) = √ , for t > 0.
πt

33.2.3 Asymptotic Behavior of F(s)

Consider the behavior of
Z ∞
fˆ(s) = e−st f (t) dt
0

1360
as s → +∞. Assume that f (t) is analytic in a neighborhood of t = 0. Only the behavior of the integrand near
t = 0 will make a significant contribution to the value of the integral. As you move away from t = 0, the e−st
term dominates. Thus we could approximate the value of fˆ(s) by replacing f (t) with the first few terms in its
Taylor series expansion about the origin.

∞
t2 00
Z  
fˆ(s) ∼ −st
e 0
f (0) + tf (0) + f (0) + · · · dt as s → +∞
0 2

Using

n!
L [tn ] =
sn+1

we obtain

f (0) f 0 (0) f 00 (0)

fˆ(s) ∼ + 2 + 3 + ··· as s → +∞.
s s s

Example 33.2.5 The Taylor series expansion of sin t about the origin is

t3
sin t = t − + O(t5 ).
6

Thus the Laplace transform of sin t has the behavior

1 1
L[sin t] ∼ − + O(s−6 ) as s → +∞.
s2 s4

1361
We corroborate this by expanding L[sin t].

1
L[sin t] =
s2
+1
s−2
=
1 + s−2
∞
X
−2
=s (−1)n s−2n
n=0
1 1
= 2
− 4 + O(s−6 )
s s

33.3 Properties of the Laplace Transform

In this section we will list several useful properties of the Laplace transform. If a result is not derived, it is shown
in the Problems section. Unless otherwise stated, assume that f (t) and g(t) are piecewise continuous and of
exponential order α.

• L[af (t) + bg(t)] = aL[f (t)] + bL[g(t)]

• L[ ect f (t)] = F (s − c) for s > c + α

dn ˆ
• L[tn f (t)] = (−1)n dsn [f (s)] for n = 1, 2, . . .
Rβ f (t)
• If 0 t
dt exists for positive β then
  Z ∞
f (t)
L = F (σ) dσ.
t s

hR i
t fˆ(s)
• L 0
f (τ ) dτ = s

1362
 f (t) = sfˆ(s) − f (0)
d 
• L dt
h i
d2
L dt2
f (t) = s2 fˆ(s) − sf (0) − f 0 (0)

To derive these formulas,

  Z ∞
d
L f (t) = e−st f 0 (t) dt
dt 0
 −st ∞
Z ∞
= e f (t) 0 − −s e−st f (t) dt
0
= −f (0) + sfˆ(s)

d2
 
L f (t) = sL[f 0 (t)] − f 0 (0)
dt2
= s2 fˆ(s) − sf (0) − f 0 (0)

• Let f (t) and g(t) be continuous. The convolution of f (t) and g(t) is defined

Z t Z t
h(t) = (f ∗ g) = f (τ )g(t − τ ) dτ = f (t − τ )g(τ ) dτ
0 0

The convolution theorem states

ĥ(s) = fˆ(s)ĝ(s).

1363
 To show this,
Z ∞ Z t
−st
ĥ(s) = e f (τ )g(t − τ ) dτ dt
Z0 ∞ Z ∞
0

= e−st f (τ )g(t − τ ) dt dτ
Z0 ∞ τ
Z ∞
−sτ
= e f (τ ) e−s(t−τ ) g(t − τ ) dt dτ
Z0 ∞ τ
Z ∞
−sτ
= e f (τ ) dτ e−sη g(η) dη
0 0
= fˆ(s)ĝ(s)

• If f (t) is periodic with period T then

RT
0
e−st f (t) dt
L[f (t)] = .
1 − e−sT

1
Example 33.3.1 Consider the inverse Laplace transform of s3 −s2
. First we factor the denominator.

1 1 1
= 2
s3 −s2 s s−1

We know the inverse Laplace transforms of each term.

   
−1 1 −1 1
L = t, L = et
s2 s−1

1364
We apply the convolution theorem.
  Z t
−1 1 1
L 2
= τ et−τ dτ
s s−1 0
Z t
−τ t
t t
− e−τ dτ
 
= e −τ e 0 − e
0
= −t − 1 + et

 
−1 1 1
L = et − t − 1.
s2 s − 1

Example 33.3.2 We can find the inverse Laplace transform of

s2 + s − 1
s3 − 2s2 + s − 2
with the aid of a table of Laplace transform pairs. We factor the denominator.
s2 + s − 1
(s − 2)(s − i)(s + i)
We expand the function in partial fractions and then invert each term.
s2 + s − 1 1 i/2 i/2
= − +
(s − 2)(s − i)(s + i) s−2 s−i s+i
2
s +s−1 1 1
= + 2
(s − 2)(s − i)(s + i) s−2 s +1
 
1 1
L−1 + 2 = e2t + sin t
s−2 s +1

1365
33.4 Constant Coefficient Differential Equations
Example 33.4.1 Consider the differential equation
y 0 + y = cos t, for t > 0, y(0) = 1.
We take the Laplace transform of this equation.
s
sŷ(s) − y(0) + ŷ(s) =
+1 s2
s 1
ŷ(s) = 2
+
(s + 1)(s + 1) s + 1
1/2 1 s+1
ŷ(s) = + 2
s+1 2s +1
Now we invert ŷ(s).
1 −t 1 1
y(t) = e + cos t + sin t, for t > 0
2 2 2
Notice that the initial condition was included when we took the Laplace transform.

One can see from this example that taking the Laplace transform of a constant coefficient differential equation
reduces the differential equation for y(t) to an algebraic equation for ŷ(s).

Example 33.4.2 Consider the differential equation

y 00 + y = cos(2t), for t > 0, y(0) = 1, y 0 (0) = 0.
We take the Laplace transform of this equation.
s
s2 ŷ(s) − sy(0) − y 0 (0) + ŷ(s) =
s2 +4
s s
ŷ(s) = + 2
(s2 2
+ 1)(s + 4) s + 1

1366
From the table of Laplace transform pairs we know
   
−1 s −1 1 1
L = cos t, L = sin(2t).
s2 + 1 2
s +4 2
We use the convolution theorem to find the inverse Laplace transform of ŷ(s).
Z t
1
y(t) = sin(2τ ) cos(t − τ ) dτ + cos t
0 2
1 t
Z
= sin(t + τ ) + sin(3τ − t) dτ + cos t
4 0
 t
1 1
= − cos(t + τ ) − cos(3τ − t) + cos t
4 3 0
 
1 1 1
= − cos(2t) + cos t − cos(2t) + cos(t) + cos t
4 3 3
1 4
= − cos(2t) + cos(t)
3 3
Alternatively, we can find the inverse Laplace transform of ŷ(s) by first finding its partial fraction expansion.
s/3 s/3 s
ŷ(s) = − +
s2 + 1 s2 + 4 s2 + 1
s/3 4s/3
=− 2 + 2
s +4 s +1

1 4
y(t) = − cos(2t) + cos(t)
3 3

Example 33.4.3 Consider the initial value problem

y 00 + 5y 0 + 2y = 0, y(0) = 1, y 0 (0) = 2.

1367
Without taking a Laplace transform, we know that since

y(t) = 1 + 2t + O(t2 )

the Laplace transform has the behavior

1 2
ŷ(s) ∼ + 2 + O(s−3 ), as s → +∞.
s s

33.5 Systems of Constant Coefficient Differential Equations

The Laplace transform can be used to transform a system of constant coefficient differential equations into a
system of algebraic equations. This should not be surprising, as a system of differential equations can be written
as a single differential equation, and vice versa.

Example 33.5.1 Consider the set of differential equations

y10 = y2
y20 = y3
y30 = −y3 − y2 − y1 + t3

with the initial conditions

y1 (0) = y2 (0) = y3 (0) = 0.

We take the Laplace transform of this system.

sŷ1 − y1 (0) = ŷ2

sŷ2 − y2 (0) = ŷ3
6
sŷ3 − y3 (0) = −ŷ3 − ŷ2 − ŷ1 +
s4

1368
The first two equations can be written as
ŷ3
ŷ1 =
s2
ŷ3
ŷ2 = .
s
We substitute this into the third equation.
ŷ3 ŷ3 6
sŷ3 = −ŷ3 − − 2+ 4
s s s
6
(s3 + s2 + s + 1)ŷ3 = 2
s
6
ŷ3 = 2 3 .
s (s + s2 + s + 1)
We solve for ŷ1 .
6
ŷ1 =
s4 (s3
+ s2 + s + 1)
1 1 1 1−s
ŷ1 = 4 − 3 + +
s s 2(s + 1) 2(s2 + 1)
We then take the inverse Laplace transform of ŷ1 .
t3 t2 1 −t 1 1
y1 = − + e + sin t − cos t.
6 2 2 2 2
We can find y2 and y3 by differentiating the expression for y1 .
t2 1 1 1
y2 = − t − e−t + cos t + sin t
2 2 2 2
1 −t 1 1
y3 = t − 1 + e − sin t + cos t
2 2 2

1369
33.6 Exercises
Exercise 33.1
Find the Laplace transform of the following functions:

1. f (t) = eat

2. f (t) = sin(at)

3. f (t) = cos(at)

4. f (t) = sinh(at)

5. f (t) = cosh(at)

sin(at)
6. f (t) =
t
Z t
sin(au)
7. f (t) = du
0 u
(
1, 0 ≤ t < π
8. f (t) =
0, π ≤ t < 2π
and f (t + 2π) = f (t) for t > 0. That is, f (t) is periodic for t > 0.

Hint, Solution

Exercise 33.2
Show that L[af (t) + bg(t)] = aL[f (t)] + bL[g(t)].
Hint, Solution

1370
Exercise 33.3
Show that if f (t) is of exponential order α,

L[ ect f (t)] = F (s − c) for s > c + α.

Hint, Solution

Exercise 33.4
Show that

dn ˆ
L[tn f (t)] = (−1)n [f (s)] for n = 1, 2, . . .
dsn
Hint, Solution

Exercise 33.5
Rβ f (t)
Show that if 0 t
dt exists for positive β then
  Z ∞
f (t)
L = F (σ) dσ.
t s

Hint, Solution

Exercise 33.6
Show that
t
fˆ(s)
Z 
L f (τ ) dτ = .
0 s
Hint, Solution

1371
Exercise 33.7
Show that if f (t) is periodic with period T then
RT
0
e−st f (t) dt
L[f (t)] = .
1 − e−sT
Hint, Solution

Exercise 33.8
The function f (t) t ≥ 0, is periodic with period 2T ; i.e. f (t + 2T ) ≡ f (t), and is also odd with period T ; i.e.
f (t + T ) = −f (t). Further,
Z T
f (t) e−st dt = ĝ(s).
0

Show that the Laplace transform of f (t) is fˆ(s) = ĝ(s)/(1 + e−sT ). Find f (t) such that fˆ(s) = s−1 tanh(sT /2).
Hint, Solution

Exercise 33.9
Find the Laplace transform of tν , ν > −1 by two methods.

1. Assume that s is complex-valued. Make the change of variables z = st and use integration in the complex
plane.

2. Show that the Laplace transform of tν is an analytic function for <(s) > 0. Assume that s is real-valued.
Make the change of variables x = st and evaluate the integral. Then use analytic continuation to extend
the result to complex-valued s.

Hint, Solution

1372
Exercise 33.10 (mathematica/ode/laplace/laplace.nb)
Show that the Laplace transform of f (t) = ln t is
Z ∞
Log s γ
fˆ(s) = − − , where γ = − e−t ln t dt.
s s 0

[ γ = 0.5772 . . . is known as Euler’s constant.]

Hint, Solution

Exercise 33.11
Find the Laplace transform of tν ln t. Write the answer in terms of the digamma function, ψ(ν) = Γ0 (ν)/Γ(ν).
What is the answer for ν = 0?
Hint, Solution

Exercise 33.12
Find the inverse Laplace transform of

1
fˆ(s) =
s3 − 2s2 +s−2

with the following methods.

1. Expand fˆ(s) using partial fractions and then use the table of Laplace transforms.

2. Factor the denominator into (s − 2)(s2 + 1) and then use the convolution theorem.

3. Use Result 33.2.1.

Hint, Solution

1373
Exercise 33.13
Solve the differential equation

y 00 + y 0 + y = sin t, y(0) = y 0 (0) = 0, 0<1

using the Laplace transform. This equation represents a weakly damped, driven, linear oscillator.
Hint, Solution

Exercise 33.14
Solve the problem,

y 00 − ty 0 + y = 0, y(0) = 0, y 0 (0) = 1,

with the Laplace transform.

Hint, Solution

Exercise 33.15
Prove the following relation between the inverse Laplace transform and the inverse Fourier transform,

1 ct −1 ˆ
L−1 [fˆ(s)] = e F [f (c + iω)],
2π

where c is to the right of the singularities of fˆ(s).

Hint, Solution

Exercise 33.16 (mathematica/ode/laplace/laplace.nb)

Show by evaluating the Laplace inversion integral that if
 π 1/2 1/2 √
fˆ(s) = e−2(as) , s1/2 = s for s > 0,
s

1374
 √
then f (t) = e−a/t
R /∞ t.−axHint: cut the s-plane along the negative real axis and deform the contour onto the cut.
2 2
p
Remember that 0 e cos(bx) dx = π/4a e−b /4a .
Hint, Solution

Exercise 33.17 (mathematica/ode/laplace/laplace.nb)

Use Laplace transforms to solve the initial value problem

d4 y
− y = t, y(0) = y 0 (0) = y 00 (0) = y 000 (0) = 0.
dt4
Hint, Solution

Exercise 33.18 (mathematica/ode/laplace/laplace.nb)

Solve, by Laplace transforms,
Z t
dy
= sin t + y(τ ) cos(t − τ ) dτ, y(0) = 0.
dt 0

Hint, Solution

Exercise 33.19 (mathematica/ode/laplace/laplace.nb)

Suppose u(t) satisfies the difference-differential equation

du
+ u(t) − u(t − 1) = 0, t ≥ 0,
dt
and the ‘initial condition’ u(t) = u0 (t), −1 ≤ t ≤ 0, where u0 (t) is given. Show that the Laplace transform û(s)
of u(t) satisfies
Z 0
u0 (0) e−s
û(s) = −s
+ −s
e−st u0 (t) dt.
1+s− e 1+s− e −1

1375
Find u(t), t ≥ 0, when u0 (t) = 1. Check the result.
Hint, Solution

Exercise 33.20
Let the function f (t) be defined by
(
1 0≤t<π
f (t) =
0 π ≤ t < 2π,

and for all positive values of t so that f (t + 2π) = f (t). That is, f (t) is periodic with period 2π. Find the solution
of the intial value problem

d2 y
− y = f (t); y(0) = 1, y 0 (0) = 0.
dt2

Examine the continuity of the solution at t = nπ, where n is a positive integer, and verify that the solution is
continuous and has a continuous derivative at these points.
Hint, Solution

Exercise 33.21
Use Laplace transforms to solve
Z t
dy
+ y(τ ) dτ = e−t , y(0) = 1.
dt 0

Hint, Solution

1376
Exercise 33.22
An electric circuit gives rise to the system

di1
L + Ri1 + q/C = E0
dt
di2
L + Ri2 − q/C = 0
dt
dq
= i1 − i2
dt
with initial conditions
E0
i1 (0) = i2 (0) = , q(0) = 0.
2R
Solve the system by Laplace transform methods and show that
E0 E0 −αt
i1 = + e sin(ωt)
2R 2ωL
where
R 2
α= and ω 2 = − α2 .
2L LC
Hint, Solution

1377
33.7 Hints
Hint 33.1
Use the differentiation and integration properties of the Laplace transform where appropriate.

Hint 33.2

Hint 33.3

Hint 33.4
∂g
If the integral is uniformly convergent and ∂s
is continuous then
Z b Z b
d ∂
g(s, t) dt = g(s, t) dt
ds a a ∂s

Hint 33.5

Z ∞
1 −sx
e−tx dt = e
s x

Hint 33.6
Use integration by parts.

1378
Hint 33.7

Z ∞ Z ∞ (n+1)T
X
−st
e f (t) dt = e−st f (t) dt
0 n=0 nT

The sum can be put in the form of a geometric series.

∞
X 1
αn = , for |α| < 1
n=0
1−α

Hint 33.8

Hint 33.9
Write the answer in terms of the Gamma function.

Hint 33.10

Hint 33.11

Hint 33.12

1379
Hint 33.13

Hint 33.14

Hint 33.15

Hint 33.16

Hint 33.17

Hint 33.18

Hint 33.19

Hint 33.20

1380
Hint 33.21

Hint 33.22

1381
33.8 Solutions
Solution 33.1
1.
Z ∞
at
e−st eat dt
 
L e =
Z0 ∞
= e−(s−a)t dt
0 −(s−a)t ∞
e
= − , for <(s) > <(a)
s−a 0

1
L eat =
 
s−a

2.
Z ∞
L[sin(at)] = e−st sin(at) dt
0
1 ∞ (−s+ia)t
Z
− e(−s−ia)t dt


= e
2i 0
∞
1 − e(−s+ia)t e(−s−ia)t

= +
2i s − ia s + ia 0
 
1 1 1
= −
2i s − ia s + ia

a
L[sin(at)] = .
s2 + a2

1382
3.
 
d sin(at)
L[cos(at)] = L
dt a
 
sin(at)
= sL − sin(0)
a

s
L[cos(at)] =
s2 + a2

4.
Z ∞
L[sinh(at)] = e−st sinh(at) dt
0
1 ∞ (−s+a)t
Z
− e(−s−a)t dt


= e
2 0
∞
1 − e(−s+a)t e(−s−a)t

= + , for <(s) > |<(a)|
2 s−a s+a 0
 
1 1 1
= −
2 s−a s+a

a
L[sinh(at)] = , for <(s) > |<(a)|
s2 − a2

5.
 
d sinh(at)
L[cosh(at)] = L
dt a
 
sinh(at)
= sL − sinh(0)
a

1383
 s
L[cosh(at)] =
s2 − a2

6. First note that

  Z ∞
sin(at)
L (s) = L[sin(at)](σ) dσ.
t s

Now we use the Laplace transform of sin(at) to compute the Laplace transform of sin(at)/t.
  Z ∞
sin(at) a
L = dσ
t σ + a2
2
Zs ∞
1 dσ
=
(σ/a)2 + 1 a
hs  σ i∞
= arctan
a  s
π s
= − arctan
2 a

 
sin(at) a
L = arctan
t s

7.
Z t   
sin(aτ ) 1 sin(at)
L dτ = L
0 τ s t

Z t 
sin(aτ ) 1 a
L dτ = arctan
0 τ s s

1384
 8.
R 2π
0
e−st f (t) dt
L[f (t)] =
R π1 − e−2πs
e−st dt
= 0
1 − e−2πs
1 − e−πs
=
s(1 − e−2πs )

1
L[f (t)] =
s(1 + e−πs )

Solution 33.2

Z ∞
e−st af (t) + bg(t) dt


L[af (t) + bg(t)] =
0
Z ∞ Z ∞
−st
=a e f (t) dt + b e−st g(t) dt
0 0
= aL[f (t)] + bL[g(t)]

Solution 33.3
If f (t) is of exponential order α, then ect f (t) is of exponential order c + α.
Z ∞
ct
L[ e f (t)] = e−st ect f (t) dt
Z0 ∞
= e−(s−c)t f (t) dt
0
= fˆ(s − c) for s > c + α

1385
Solution 33.4
First consider the Laplace transform of t0 f (t).

L[t0 f (t)] = fˆ(s)

Now consider the Laplace transform of tn f (t) for n ≥ 1.

Z ∞
n
L[t f (t)] = e−st tn f (t) dt
0
d ∞ −st n−1
Z
=− e t f (t) dt
ds 0
d
= − L[tn−1 f (t)]
ds

Thus we have a difference equation for the Laplace transform of tn f (t) with the solution

n dn
n
L[t f (t)] = (−1) n
L[t0 f (t)] for n ∈ Z0+ ,
ds

dn ˆ
L[tn f (t)] = (−1)n f (s) for n ∈ Z0+ .
dsn

Solution
R β f (t) 33.5
If 0 t dt exists for positive β and f (t) is of exponential order α then the Laplace transform of f (t)/t is defined

1386
for s > α.

 Z ∞

f (t) 1
L = e−st f (t) dt
t t
Z0 ∞ Z ∞
= e−σt dσ f (t) dt
0 s
Z ∞Z ∞
= e−σt f (t) dt dσ
Zs ∞ 0
= fˆ(σ) dσ
s

Solution 33.6

Z t  Z ∞ Z t
−st
L f (τ ) dτ = e f (τ ) dτ dx
0 0 0
 −st t ∞ Z ∞ t
e−st d
Z Z 
e
= − f (τ ) dτ − − f (τ ) dτ dt
s s dt
Z ∞ 0 0 0 0
1
= e−st f (t) dt
s 0
1
= fˆ(s)
s

1387
Solution 33.7
f (t) is periodic with period T .

Z ∞
L[f (t)] = e−st f (t) dt
0
Z T Z 2T
−st
= e f (t) dt + e−st f (t) dt + · · ·
0 T
∞ Z
X (n+1)T
= e−st f (t) dt
n=0 nT
X∞ Z T
= e−s(t+nT ) f (t + nT ) dt
n=0 0

X∞ Z T
−snT
= e e−st f (t) dt
n=0 0
Z T ∞
X
−st
= e f (t) dt e−snT
0 n=0
RT
0
e−st f (t) dt
=
1 − e−sT

1388
Solution 33.8

Z ∞
fˆ(s) = e−st f (t) dt
0
n Z
X (n+1)T
= e−st f (t) dt
0 nT
n
X T
Z
= e−s(t+nT ) f (t + nT ) dt
0 0
n
X Z T
−snT
= e e−st (−1)n f (t) dt
0 0
Z T n
−st
X
n
= e f (t) dt (−1)n e−sT
0 0

ĝ(s)
fˆ(s) = , for <(s) > 0
1 + e−sT

Consider fˆ(s) = s−1 tanh(sT /2).

esT /2 − e−sT /2
s−1 tanh(sT /2) = s−1
esT /2 + e−sT /2
1 − e−sT
= s−1
1 + e−sT

We have
T
1 − e−st
Z
ĝ(s) ≡ f (t) e−st dt = .
0 s

1389
By inspection we see that this is satisfied for f (t) = 1 for 0 < t < T . We conclude:
(
1 for t ∈ [2nT . . . (2n + 1)T ),
f (t) =
−1 for t ∈ [(2n + 1)T . . . (2n + 2)T ),

where n ∈ Z.

Solution 33.9
The Laplace transform of tν , ν > −1 is
Z ∞
fˆ(s) = e−st tν dt.
0

Assume s is complex-valued. The integral converges for <(s) > 0 and ν > −1.

Method 1. We make the change of variables z = st.

Z  z ν 1
fˆ(s) = e−z dz
C s s
Z
−(ν+1)
=s e−z z ν dz
C

C is the path from 0 to ∞ along arg(z) = arg(s). (Shown in Figure 33.4).

Since the integrand is analytic in the domain  < r < R, 0 < θ < arg(s), the integral along the boundary of
this domain vanishes.
Z i arg(s) Z i arg(s) Z !
Z R Re e 
+ + + e−z z ν dz = 0
 R R ei arg(s)  ei arg(s)

1390
 Im(z)

arg(s)
Re(z)

Figure 33.4: The Path of Integration.

We show that the integral along CR , the circular arc of radius R, vanishes as R → ∞ with the maximum modulus
integral bound.
Z 
−z ν  ≤ R| arg(s)| max  e−z z ν 
   
 e z dz

CR
 z∈CR

= R| arg(s)| e−R cos(arg(s)) Rν

→ 0 as R → ∞.

The integral along C , the circular arc of radius , vanishes as  → 0. We demonstrate this with the maximum
modulus integral bound.
Z 
−z ν  ≤ | arg(s)| max  e−z z ν 
   
 e z dz

C
 z∈C

= | arg(s)| e− cos(arg(s)) ν

→ 0 as  → 0.

1391
 Taking the limit as  → 0 and R → ∞, we see that the integral along C is equal to the integral along the real
axis.
Z Z ∞
−z ν
e z dz = e−z z ν dz
C 0
ν
We can evaluate the Laplace transform of t in terms of this integral.
Z ∞
−(ν+1)
ν
L [t ] = s e−t tν dt
0

Γ(ν + 1)
L [tν ] =
s−(ν+1)
In the case that ν is a non-negative integer ν = n > −1 we can write this in terms of the factorial.
n!
L [tn ] =
s−(n+1)

Method 2. First note that the integral

Z ∞
fˆ(s) = e−st tν dt
0

exists for <(s) > 0. It converges uniformly for <(s) ≥ c > 0. On this domain of uniform convergence we can
interchange differentiation and integration.
dfˆ d ∞ −st ν
Z
= e t dt
ds ds 0
Z ∞
∂
e−st tν dt


=
∂s
Z0 ∞
= −t e−st tν dt
0
Z ∞
=− e−st tν+1 dt
0

1392
Since fˆ0 (s) is defined for <(s) > 0, fˆ(s) is analytic for <(s) > 0.
Let σ be real and positive. We make the change of variables x = σt.

Z ∞  x ν 1
fˆ(σ) = e −x
dx
0 σ σ
Z ∞
= σ −(ν+1) e−x xν dx
0
Γ(ν + 1)
=
σ ν+1

Note that the function

Γ(ν + 1)
fˆ(s) =
sν+1

is the analytic continuation of fˆ(σ). Thus we can define the Laplace transform for all complex s in the right half
plane.

Γ(ν + 1)
fˆ(s) =
sν+1

Solution 33.10
Note that fˆ(s) is an analytic function for <(s) > 0. Consider real-valued s > 0. By definition, fˆ(s) is

Z ∞
fˆ(s) = e−st ln t dt.
0

1393
We make the change of variables x = st.
Z ∞  x  dx
fˆ(s) = e−x ln
0 s s
Z ∞
1
= e−x (ln x − ln s) dx
s 0
ln |s| ∞ −x 1 ∞ −x
Z Z
=− e dx + e ln x dx
s 0 s 0
ln s γ
=− − , for real s > 0
s s

The analytic continuation of fˆ(s) into the right half-plane is

Log s γ
fˆ(s) = − − .
s s

Solution 33.11
Define
Z ∞
fˆ(s) = L[t ln t] = ν
e−st tν ln t dt.
0

This integral defines fˆ(s) for <(s) > 0. Note that the integral converges uniformly for <(s) ≥ c > 0. On this
domain we can interchange differentiation and integration.
Z ∞ Z ∞
ˆ0 ∂ −st ν
t e−st tν Log t dt


f (s) = e t ln t dt = −
0 ∂s 0

Since fˆ0 (s) also exists for <(s) > 0, fˆ(s) is analytic in that domain.

1394
 Let σ be real and positive. We make the change of variables x = σt.
fˆ(σ) = L [tν ln t]
Z ∞
= e−σt tν ln t dt
Z0 ∞
−x x
 ν x 1
= e ln dx
0 σ σσ
Z ∞
1
= ν+1 e−x xν (ln x − ln σ) dx
σ
0Z ∞ Z ∞ 
1 −x ν −x ν
= ν+1 e x ln x dx − ln σ e x dx
σ
Z0 ∞ 0

1 ∂ −x ν


= ν+1 e x dx − ln σΓ(ν + 1)
σ 0 ∂ν
 Z ∞ 
1 d −x ν
= ν+1 e x dx − ln σΓ(ν + 1)
σ dν 0
 
1 d
= ν+1 Γ(ν + 1) − ln σΓ(ν + 1)
σ dν
 0 
1 Γ (ν + 1)
= ν+1 Γ(ν + 1) − ln σ
σ Γ(ν + 1)
1
= ν+1 Γ(ν + 1) (ψ(ν + 1) − ln σ)
σ
Note that the function
1
fˆ(s) = Γ(ν + 1) (ψ(ν + 1) − ln s)
sν+1
is an analytic continuation of fˆ(σ). Thus we can define the Laplace transform for all s in the right half plane.
1
L[tν ln t] = Γ(ν + 1) (ψ(ν + 1) − ln s) for <(s) > 0.
sν+1

1395
For the case ν = 0, we have
1
L[ln t] = Γ(1) (ψ(1) − ln s)
s1
−γ − ln s
L[ln t] = ,
s

where γ is Euler’s constant

Z ∞
γ= e−x ln x dx = 0.5772156629 . . .
0

Solution 33.12
Method 1. We factor the denominator.

1 1
fˆ(s) = 2
=
(s − 2)(s + 1) (s − 2)(s − i)(s + i)

We expand the function in partial fractions and simplify the result.

1 1/5 (1 − i2)/10 (1 + i2)/10

= − −
(s − 2)(s − i)(s + i) s−2 s−i s+i
1 1 1 s+2
fˆ(s) = − 2
5s−2 5s +1
We use a table of Laplace transforms to do the inversion.
1 s 1
L[ e2t ] = , L[cos t] = , L[sin t] =
s−2 s2 +1 s2 +1
1 2t

f (t) = e − cos t − 2 sin t
5

1396
 Method 2. We factor the denominator.
1 1
fˆ(s) = 2
s−2s +1
From a table of Laplace transforms we note
1 1
L[ e2t ] = , L[sin t] = .
s−2 s2 + 1
We apply the convolution theorem.
Z t
f (t) = sin τ e2(t−τ ) dτ
0
1 2t

f (t) = e − cos t − 2 sin t
5

Method 3. We factor the denominator.

1
fˆ(s) =
(s − 2)(s − i)(s + i)

fˆ(s) is analytic except for poles and vanishes at infinity.

est
X  
f (t) = Res , sn
s =2,i,−i
(s − 2)(s − i)(s + i)
n

e2t eit e−it

= + +
(2 − i)(2 + i) (i − 2)(i2) (−i − 2)(−i2)
e2t (−1 + i2) eit (−1 − i2) e−it
= + +
5 10 10
e2t eit + e−it eit − e−it
= +− +i
5 10 5

1397
 1 2t

f (t) = e − cos t − 2 sin t
5

Solution 33.13

y 00 + y 0 + y = sin t, y(0) = y 0 (0) = 0, 0<1

We take the Laplace transform of this equation.

(s2 ŷ(s) − sy(0) − y 0 (0)) + (sŷ(s) − y(0)) + ŷ(s) = L[sin(t)]

(s2 + s + 1)ŷ(s) = L[sin(t)]
1
ŷ(s) = 2 L[sin(t)]
s + s + 1
1
ŷ(s) = 2 L[sin(t)]
(s + 2 ) + 1 − 4
 2

We use a table of Laplace transforms to find the inverse Laplace transform of the first term.
" # r !
1 1 2
L−1  2 2
=q e−t/2 sin 1− t
(s + 2 ) + 1 − 4 1− 2 4
4

We define
r
2
α= 1−
4

1398
to get rid of some clutter. Now we apply the convolution theorem to invert 2 ŷs.
Z t
1 −τ /2
y(t) = e sin (ατ ) sin(t − τ ) dτ
0 α
 
−t/2 1 1 1
y(t) = e cos (αt) + sin (αt) − cos t
 2α 

The solution is plotted in Figure 33.5 for  = 0.05.

15

10

20 40 60 80 100
-5

-10

-15

Figure 33.5: The Weakly Damped, Driven Oscillator

2
Evaluate the convolution integral by inspection.

1399
Solution 33.14
We consider the solutions of
y 00 − ty 0 + y = 0, y(0) = 0, y 0 (0) = 1
which are of exponential order α for any α > 0. We take the Laplace transform of the differential equation.
d
s2 ŷ − 1 + (sŷ) + ŷ = 0
 ds 
2 1
ŷ 0 + s + ŷ =
s s
2
1 e−s /2
ŷ(s) = 2 + c 2
s s
We use that
y(0) y 0 (0)
ŷ(s) ∼ + 2 + ···
s s
to conclude that c = 0.
1
ŷ(s) =
s2
y(t) = t

Solution 33.15

Z c+i∞
1
L [fˆ(s)] =
−1
est fˆ(s) ds
i2π c−i∞

First we make the change of variable s = c + σ.

Z i∞
1 ct
L [fˆ(s)] =
−1
e eσt F (c + σ) dσ
i2π −i∞

1400
Then we make the change of variable σ = iω.
1 ct ∞ iωt
Z
−1ˆ
L [f (s)] = e e F (c + iω) dω
2π −∞
1 ct −1
L−1 [fˆ(s)] = e F [F (c + iω)]
2π

Solution 33.16 √
We assume that <(a) ≥ 0. We are considering the principal branch of the square root: s1/2 = s. There is a
branch cut on the negative real axis. fˆ(s) is singular at s = 0 and along the negative real axis. Let α be any

1/2 −2(as)1/2
positive number. The inverse Laplace transform of πs e is
Z α+i∞
1  π 1/2 1/2
f (t) = e st
e−2(as) ds.
i2π α−i∞ s

We will evaluate the integral by deforming it to wrap around the branch cut. Consider the integral on the
contour shown in Figure 33.6. CR+ and CR− are circular arcs of radius R. B is the vertical line at <(s) = α joining
the two arcs. C is a semi-circle in the right half plane joining i and −i. L+ and L− are lines joining the circular
arcs at =(s) = ±.
Since there are no residues inside the contour, we have
Z Z Z Z Z Z !
1  π 1/2 1/2
+ + + + + est e−2(as) ds = 0.
i2π B +
CR L+ C L− −
CR s

We will evaluate the inverse Laplace transform for t > 0.

First we will show that the integral along CR+ vanishes as R → ∞. We parametrize the path of integration
with s = R eiθ and write the integral along CR+ as the sum of two integrals.
Z Z π/2 Z π
· · · ds = · · · dθ + · · · dθ
+
CR π/2−δ π/2

1401
 CR+
B

π/2−δ
L+ Cε
L- −π/2+δ

CR-

Figure 33.6: Path of Integration

The first integral vanishes by the maximum modulus bound. Note that the length of the path of integration is
less than 2α.

Z  
 π/2     
 st π 1/2 −2(as)1/2 
· · · dθ ≤ max e e  (2α)
 
s

 π/2−δ  θ∈[π/2−δ...π/2]  
√
π
= eαt √ (2α)
R
→ 0 as R → ∞

1402
The second integral vanishes by Jordan’s Lemma.
Z π √ √
 Z π  √ 
iθ π iθ iθ π √ √ iθ/2 
e−2 aR e dθ ≤ e−2 a R e  dθ
  
 eR e t √  eR e t √

R eiθ 
R e iθ
π/2 π/2
√ Z π
π
≤√ eR cos(θ)t dθ
R π/2
√ Z π/2
π
≤√ e−Rt sin(φ) dφ
R 0
√
π π
<√
R 2Rt
→ 0 as R → ∞
We could show that the integral along CR− vanishes by the same method.
Now we have
Z Z Z Z 
1  π 1/2 1/2
+ + + est e−2(as) ds = 0.
i2π B L+ C L− s
We show that the integral along C vanishes as  → 0 with the maximum modulus bound.
Z  π 1/2      
 st −2(as)1/2   st π 1/2 −2(as)1/2 
 e e ds ≤ max  e e  (π)

C s s∈C s 
√
t π
≤ e √ π

→ 0 as  → 0.
Now we can express the inverse Laplace transform in terms of the integrals along L+ and L−
Z α+i∞
1  π 1/2 1/2
f (t) ≡ est e−2(as) ds
i2π α−i∞ s
Z Z
1 st π
 1/2
−2(as)1/2 1  π 1/2 1/2
=− e e ds − est e−2(as) ds.
i2π L+ s i2π L− s

1403
On L+ , s = r eiπ , ds = eiπ dr = −dr; on L− , s = r e−iπ , ds = e−iπ dr = −dr. We can combine the integrals along
the top and bottom of the branch cut.
Z 0 √ Z ∞ √
1 −rt π −i2√a√r 1 −rt π √ √
f (t) = − e √ e (− dr) − e √ ei2 a r (− dr)
i2π ∞ i r i2π 0 −i r
Z ∞ √ √ √ √
1 1  
= √ e−rt √ e−i2 a r + ei2 a r dr
2 π 0 r
Z ∞
1 1 √ √

= √ √ e−rt 2 cos 2 a r dr
2 π 0 r
√
We make the change of variables x = r.
Z ∞
1 1 −tx2 √

=√ e cos 2 ax 2x dx
π 0 x
Z ∞
2 2 √

=√ e−tx cos 2 ax dx
π 0
r
2 π −4a/(4t)
=√ e
π 4t
e−a/t
= √
t
Thus the inverse Laplace transform is

e−a/t
f (t) = √
t

Solution 33.17
We consider the problem
d4 y
− y = t, y(0) = y 0 (0) = y 00 (0) = y 000 (0) = 0.
dt4

1404
We take the Laplace transform of the differential equation.

1
s4 ŷ(s) − s3 y(0) − s2 y 0 (0) − sy 00 (0) − y 000 (0) − ŷ(s) =
s2
1
s4 ŷ(s) − ŷ(s) =
s2
1
ŷ(s) =
s2 (s4 − 1)

There are several ways in which we could carry out the inverse Laplace transform to find y(t). We could expand
the right side in partial fractions and then use a table of Laplace transforms. Since the function is analytic except
for isolated singularities and vanishes as s → ∞ we could use the result,

N
X  
L [fˆ(s)] =
−1
Res est fˆ(s), sn ,
n=1

where {sk }nk=1 are the singularities of fˆ(s). Since we can write the function as a product of simpler terms we
could also apply the convolution theorem.
We will first do the inverse Laplace transform by expanding the function in partial fractions to obtain simpler
rational functions.

1 1
= 2
s2 (s4 − 1) s (s − 1)(s + 1)(s − i)(s + i)
a b c d e f
= 2+ + + + +
s s s−1 s+1 s−i s+i

1405
  
1
a= 4 = −1
s − 1 s=0
 
d 1
b= =0
ds s4 − 1 s=0
 
1 1
c= 2 =
s (s + 1)(s − i)(s + i) s=1 4
 
1 1
d= 2 =−
s (s − 1)(s − i)(s + i) s=−1 4
 
1 1
e= 2 = −i
s (s − 1)(s + 1)(s + i) s=i 4
 
1 1
f= 2 =i
s (s − 1)(s + 1)(s − i) s=−i 4

Now we have simple functions that we can look up in a table.

1 1/4 1/4 1/2

ŷ(s) = − 2 + − + 2
 s s − 1 s + 1 s + 1
1 t 1 −t 1
y(t) = −t + e − e + sin t H(t)
4 4 2
 
1
y(t) = −t + (sinh t + sin t) H(t)
2

We can also do the inversion with the convolution theorem.

1 1 1 1
= 2 2
s2 (s4− 1) 2
s s +1s −1

1406
From a table of Laplace transforms we know,

 
−1 1
L = t,
s2
 
−1 1
L = sin t,
s2 + 1
 
−1 1
L = sinh t.
s2 − 1

Now we use the convolution theorem to find the solution for t > 0.

  Z t
−1 1
L = sinh(τ ) sin(t − τ ) dτ
s4 − 1 0
1
= (sinh t − sin t)
2

  Z t
−1 1 1
L = (sinh τ − sin τ ) (t − τ ) dτ
s2 (s4 − 1) 0 2
1
= −t + (sinh t + sin t)
2

1407
Solution 33.18

Z t
dy
= sin t + y(τ ) cos(t − τ ) dτ
dt 0
1 s
sŷ(s) − y(0) = + ŷ(s) 2
+1s2 s +1
(s3 + s)ŷ(s) − sŷ(s) = 1
1
ŷ(s) = 3
s
t2
y(t) =
2

Solution 33.19
The Laplace transform of u(t − 1) is

Z ∞
L[u(t − 1)] = e−st u(t − 1) dt
Z0 ∞
= e−s(t+1) u(t) dt
−1
Z 0 Z ∞
−s −st −s
= e e u(t) dt + e e−st u(t) dt
−1 0
Z 0
= e−s e−st u0 (t) dt + e−s û(s).
−1

1408
 We take the Laplace transform of the difference-differential equation.
Z 0
−s
sû(s) − u(0) + û(s) − e e−st u0 (t) dt + e−s û(s) = 0
−1
Z 0
−s −s
(1 + s − e )û(s) = u0 (0) + e e−st u0 (t) dt
−1
0
e−s
Z
u0 (0)
û(s) = + e−st u0 (t) dt
1 + s − e−s 1 + s − e−s −1

Consider the case u0 (t) = 1.

Z 0
1 e−s
û(s) = −s
+ −s
e−st dt
1+s− e 1+s− e
−s
 −1 
1 e 1 1 s
û(s) = + − + e
1 + s − e−s 1 + s − e−s s s
−s
1/s + 1 − e /s
û(s) =
1 + s − e−s
1
û(s) =
s
u(t) = 1

Clearly this solution satisfies the difference-differential equation.

Solution 33.20
We consider the problem,

d2 y
− y = f (t), y(0) = 1, y 0 (0) = 0,
dt2

1409
where f (t) is periodic with period 2π and is defined by,
(
1 0 ≤ t < π,
f (t) =
0 π ≤ t < 2π.

We take the Laplace transform of the differential equation.

s2 ŷ(s) − sy(0) − y 0 (0) − ŷ(s) = fˆ(s)

s2 ŷ(s) − s − ŷ(s) = fˆ(s)
s fˆ(s)
ŷ(s) = +
s2 − 1 s2 − 1
By inspection, (of a table of Laplace transforms), we see that
 
−1 s
L = cosh(t)H(t),
s2 − 1
 
−1 1
L = sinh(t)H(t).
s2 − 1

Now we use the convolution theorem.

" # Z
fˆ(s) t
L−1 = f (τ ) sinh(t − τ ) dτ
s2 − 1 0

The solution for positive t is

Z t
y(t) = cosh(t) + f (τ ) sinh(t − τ ) dτ.
0

1410
Clearly the solution is continuous because the integral of a bounded function is continuous. The first derivative
of the solution is
Z t
0
y (t) = sinh t + f (t) sinh(0) + f (τ ) cosh(t − τ ) dτ
0
Z t
0
y (t) = sinh t + f (τ ) cosh(t − τ ) dτ
0

We see that the first derivative is also continuous.

Solution 33.21
We consider the problem
Z t
dy
+ y(τ ) dτ = e−t , y(0) = 1.
dt 0

We take the Laplace transform of the equation and solve for ŷ.
ŷ 1
sŷ − y(0) + =
s s+1
s(s + 2)
ŷ =
(s + 1)(s2 + 1)
We expand the right side in partial fractions.
1 1 + 3s
ŷ = − +
2(s + 1) 2(s2 + 1)
We use a table of Laplace transforms to do the inversion.

1 1
y = − e−t + (sin(t) + 3 cos(t))
2 2

1411
Solution 33.22
We consider the problem

di1
L + Ri1 + q/C = E0
dt
di2
L + Ri2 − q/C = 0
dt
dq
= i1 − i2
dt
E0
i1 (0) = i2 (0) = , q(0) = 0.
2R

We take the Laplace transform of the system of differential equations.

 
E0 q̂ E0
L sî1 − + Rî1 + =
2R C s
 
E0 q̂
L sî2 − + Rî2 − = 0
2R C
sq̂ = î1 − î2

We solve for î1 , î2 and q̂.

 
E0 1 1/L
î1 = +
2 Rs s2 + Rs/L + 2/(CL)
 
E0 1 1/L
î2 = −
2 Rs s2 + Rs/L + 2/(CL)
 
CE0 1 s + R/L
q̂ = −
2 s s2 + Rs/L + 2/(CL)

1412
We factor the polynomials in the denominators.
 
E0 1 1/L
î1 = +
2 Rs (s + α − iω)(s + α + iω)
 
E0 1 1/L
î2 = −
2 Rs (s + α − iω)(s + α + iω)
 
CE0 1 s + 2α
q̂ = −
2 s (s + α − iω)(s + α + iω)
Here we have defined
R 2
α= and ω 2 = − α2 .
2L LC
We expand the functions in partial fractions.
  
E0 1 i 1 1
î1 = + −
2 Rs 2ωL s + α + iω s + α − iω
  
E0 1 i 1 1
î2 = − −
2 Rs 2ωL s + α + iω s + α − iω
  
CE0 1 i α + iω α − iω
q̂ = + −
2 s 2ω s + α − iω s + α + iω
Now we can do the inversion with a table of Laplace transforms.
 
E0 1 i (−α−iω)t (−α+iω)t


i1 = + e −e
2 R 2ωL
 
E0 1 i (−α−iω)t (−α+iω)t


i2 = − e −e
2 R 2ωL
 
CE0 i (−α+iω)t (−α−iω)t


q= 1+ (α + iω) e − (α − iω) e
2 2ω

1413
We simplify the expressions to obtain the solutions.
 
E0 1 1 −αt
i1 = + e sin(ωt)
2 R ωL
 
E0 1 1 −αt
i2 = − e sin(ωt)
2 R ωL
CE0   α 
q= 1 − e−αt cos(ωt) + sin(ωt)
2 ω

1414
Chapter 34

The Fourier Transform

34.1 Derivation from a Fourier Series

Consider the eigenvalue problem

y 00 + λy = 0, y(−L) = y(L), y 0 (−L) = y 0 (L).

The eigenvalues and eigenfunctions are

 nπ 2
λn = for n ∈ Z0+
L
π inπx/L
φn = e , for n ∈ Z
L
The eigenfunctions form an orthogonal set. A piecewise continuous function defined on [−L . . . L] can be expanded
in a series of the eigenfunctions.
∞
X π inπx/L
f (x) ∼ cn e
n=−∞
L

1415
The Fourier coefficients are
D  E
π inπx/L 
L
e f (x)
cn = D  E
π
einπx/L  Lπ einπx/L

L
Z L
1
= e−inπx/L f (x) dx.
2π −L

We substitute the expression for cn into the series for f (x).

∞  Z L 
X 1 −inπξ/L
f (x) ∼ e f (ξ) dξ einπx/L .
n=−∞
2L −L

We let ωn = nπ/L and ∆ω = π/L.

∞  Z L 
X 1 −iωn ξ
f (x) ∼ e f (ξ) dξ eiωn x ∆ω.
ω =−∞
2π −L
n

In the limit as L → ∞, (and thus ∆ω → 0), the sum becomes an integral.

Z ∞ Z ∞ 
1 −iωξ
f (x) ∼ e f (ξ) dξ eiωx dω.
−∞ 2π −∞

Thus the expansion of f (x) for finite L

∞
X π inπx/L
f (x) ∼ cn e
n=−∞
L
Z L
1
cn = e−inπx/L f (x) dx
2π −L

1416
in the limit as L → ∞ becomes
Z ∞
f (x) ∼ fˆ(ω) eiωx dω
−∞
Z ∞
1
fˆ(ω) = f (x) e−iωx dx.
2π −∞

Of course this derivation is only heuristic. In the next section we will explore these formulas more carefully.

34.2 The Fourier Transform

R∞
Let f (x) be piecewise continuous and let −∞
|f (x)| dx exist. We define the function I(x, L).
Z L Z ∞ 
1
I(x, L) = f (ξ) e iωξ
dξ e−iωx dω.
2π −L −∞

Since the integral in parentheses is uniformly convergent, we can interchange the order of integration.
Z ∞ Z L 
1 iω(ξ−x)
= f (ξ) e dω dξ
2π −∞ −L
Z ∞ L
1 eiω(ξ−x)
= f (ξ) dξ
2π −∞ i(ξ − x) −L
Z ∞
1 1
eiL(ξ−x) − e−iL(ξ−x) dξ


= f (ξ)
2π −∞ i(ξ − x)
Z ∞
1 sin(L(ξ − x))
= f (ξ) dξ
π −∞ ξ−x
1 ∞
Z
sin(Lξ)
= f (ξ + x) dξ.
π −∞ ξ

1417
In Example 34.3.3 we will show that
Z ∞
sin(Lξ) π
dξ = .
0 ξ 2

Continuous Functions. Suppose that f (x) is continuous.

1 ∞
Z
sin(Lξ)
f (x) = f (x) dξ
π −∞ ξ
1 ∞ f (x + ξ) − f (x)
Z
I(x, L) − f (x) = sin(Lξ) dξ.
π −∞ ξ
R ∞  f (x+ξ)−f (x) 
If f (x) has a left and right derivative at x then f (x+ξ)−f
ξ
(x)
is bounded and −∞  ξ  dξ < ∞. We use the
Riemann-Lebesgue lemma to show that the integral vanishes as L → ∞.

1 ∞ f (x + ξ) − f (x)
Z
sin(Lξ) dξ → 0 as L → ∞.
π −∞ ξ

Now we have an identity for f (x).

Z ∞ Z ∞ 
1
f (x) = f (ξ) e iωξ
dξ e−iωx dω.
2π −∞ −∞

Piecewise Continuous Functions. Now consider the case that f (x) is only piecewise continuous.

f (x+ ) 1 ∞
Z
sin(Lξ)
= f (x+ ) dξ
2 π 0 ξ
f (x− ) 1 0
Z
sin(Lξ)
= f (x− ) dξ
2 π −∞ ξ

1418
 0
f (x+ ) + f (x− ) f (x + ξ) − f (x− )
Z  
I(x, L) − = sin(Lξ) dξ
2 −∞ ξ
Z ∞
f (x + ξ) − f (x+ )

− sin(Lξ) dξ
0 ξ

If f (x) has a left and right derivative at x, then

f (x + ξ) − f (x− )
is bounded for ξ ≤ 0, and
ξ
f (x + ξ) − f (x+ )
is bounded for ξ ≥ 0.
ξ

Again using the Riemann-Lebesgue lemma we see that

∞ ∞
f (x+ ) + f (x− )
Z Z 
1
= f (ξ) e iωξ
dξ e−iωx dω.
2 2π −∞ −∞

1419
 R∞
Result 34.2.1 Let f (x) be piecewise continuous with −∞ |f (x)| dx < ∞. The Fourier
transform of f (x) is defined
Z ∞
1
fˆ(ω) = F[f (x)] = f (x) e−iωx dx.
2π −∞
We see that the integral is uniformly convergent. The inverse Fourier transform is defined
Z ∞
f (x+ ) + f (x− ) −1 ˆ
= F [f (ω)] = fˆ(ω) eiωx dω.
2 −∞

If f (x) is continuous then this reduces to

Z ∞
f (x) = F −1
[fˆ(ω)] = fˆ(ω) eiωx dω.
−∞

34.2.1 A Word of Caution

Other texts may define the Fourier transform differently. The important relation is
Z ∞  Z ∞ 
1 ∓iωξ
f (x) = f (ξ) e dξ e±iωx dω.
−∞ 2π −∞

Multiplying the right side of this equation by 1 = α1 α yields

Z ∞  Z ∞ 
1 α ∓iωξ
f (x) = f (ξ) e dξ e±iωx dω.
α −∞ 2π −∞

1420
 √
Setting α = 2π and choosing sign in the exponentials gives us the Fourier transform pair
Z ∞
ˆ 1
f (ω) = √ f (x) e−iωx dx
2π −∞
Z ∞
1
f (x) = √ fˆ(ω) eiωx dω.
2π −∞
Other equally valid pairs are
Z ∞
fˆ(ω) = f (x) e−iωx dx
−∞
Z ∞
1
f (x) = fˆ(ω) eiωx dω,
2π −∞

and
Z ∞
fˆ(ω) = f (x) eiωx dx
−∞
Z ∞
1
f (x) = fˆ(ω) e−iωx dω.
2π −∞

Be aware of the different definitions when reading other texts or consulting tables of Fourier transforms.

34.3 Evaluating Fourier Integrals

34.3.1 Integrals that Converge
If the Fourier integral
Z ∞
1
F[f (x)] = f (x) e−iωx dx,
2π −∞

1421
converges for real ω, then finding the transform of a function is just a matter of direct integration. We will
consider several examples of such garden variety functions in this subsection. Later on we will consider the more
interesting cases when the integral does not converge for real ω.

Example 34.3.1 Consider the Fourier transform of e−α|x| , where α > 0. Since the integral of e−α|x| is absolutely
convergent, we know that the Fourier transform integral converges for real ω. We write out the integral.

Z ∞
−α|x| 1
e−α|x| e−iωx dx
 
F e =
2π −∞
Z 0 Z ∞
1 1
= e αx−iωx
dx + e−αx−iωx dx
2π −∞ 2π 0
Z 0 Z ∞
1 (α−i<(ω)+=(ω))x 1
= e dx + e(−α−i<(ω)+=(ω))x dx
2π −∞ 2π 0

The integral converges for |=(ω)| < α. This domain is shown in Figure 34.1.

Im(z)

Re(z)

Figure 34.1: The Domain of Convergence

1422
 Now We do the integration.
Z 0 Z ∞
−α|x| 1 1
(α−iω)x
e−(α+iω)x dx
 
F e = e dx +
2π −∞ 2π 0
 (α−iω)x 0  −(α+iω)x ∞
1 e 1 e
= + −
2π α − iω −∞ 2π α + iω 0
 
1 1 1
= +
2π α − iω α + iω
1 α
= , for |=(ω)| < α
π π(ω + α2 )
2

We can extend the domain of the Fourier transform with analytic continuation.
α
F e−α|x| =
 
, for ω 6= ±iα
π(ω + α2 )
2

1
Example 34.3.2 Consider the Fourier transform of f (x) = x−iα , α > 0.
  Z ∞
1 1 1
F = e−iωx dx
x − iα 2π −∞ x − iα
The integral converges for =(ω) = 0. We will evaluate the integral for positive and negative real values of ω.

For ω > 0, we will close the path of integration in the lower half-plane. Let CR be the contour from x = R
to x = −R following a semicircular path in the lower half-plane. The integral along CR vanishes as R → ∞ by
Jordan’s Lemma.
Z
1
e−iωx dx → 0 as R → ∞.
CR x − iα

Since the integrand is analytic in the lower half-plane the integral vanishes.
 
1
F =0
x − iα

1423
 For ω < 0, we will close the path of integration in the upper half-plane. Let CR denote the semicircular contour
from x = R to x = −R in the upper half-plane. The integral along CR vanishes as R goes to infinity by Jordan’s
Lemma. We evaluate the Fourier transform integral with the Residue Theorem.
   −iωx 
1 1 e
F = 2πi Res , iα
x − iα 2π x − iα
= i eαω

We combine the results for positive and negative values of ω.

  (
1 0 for ω > 0,
F =
x − iα ie αω
for ω < 0

34.3.2 Cauchy Principal Value and Integrals that are Not Absolutely Convergent.
That the integral of f (x) is absolutely R ∞convergent is a sufficient but not a Rnecessary condition that the Fourier
−iωx ∞
transform of f (x) exists. The integral −∞ f (x) e dx may converge even if −∞ |f (x)| dx does not. Furthermore,
if the Fourier transform integral diverges, its principal value may exist. We will say that the Fourier transform of
f (x) exists if the principal value of the integral exists.
Z ∞
F[f (x)] = − f (x) e−iωx dx
−∞

Example 34.3.3 Consider the Fourier transform of f (x) = 1/x.

Z ∞
ˆ 1 1 −iωx
f (ω) = − e dx
2π −∞ x
If ω > 0, we can close the contour in the lower half-plane. The integral along the semi-circle vanishes due to
Jordan’s Lemma.
Z
1 −iωx
lim e dx = 0
R→∞ C x
R

1424
We can evaluate the Fourier transform with the Residue Theorem.
   
1 −1 1
fˆ(ω) = (2πi) Res e−iωx
,0
2π 2 x
i
fˆ(ω) = − , for ω > 0.
2
The factor of −1/2 in the above derivation arises because the path of integration is in the negative, (clockwise),
direction and the path of integration crosses through the first order pole at x = 0. The path of integration is
shown in Figure 34.2.

Im(z)

Re(z)

Figure 34.2: The Path of Integration

If ω < 0, we can close the contour in the upper half plane to obtain
i
fˆ(ω) = , for ω < 0.
2
1
For ω = 0 the integral vanishes because x
is an odd function.
Z ∞
ˆ 1 1
f (0) = =− dx = 0
2π −∞ x

1425
We collect the results in one formula.
i
fˆ(ω) = − sign (ω)
2

We write the integrand for ω > 0 as the sum of an odd and and even function.
Z ∞
1 1 −iωx i
− e dx = −
2π −∞ x 2
Z ∞ Z ∞
1 −i
− cos(ωx) dx + − sin(ωx) dx = −iπ
−∞ x −∞ x

The principal value of the integral of any odd function is zero.

Z ∞
1
− sin(ωx) dx = π
−∞ x

If the principal value of the integral of an even function exists, then the integral converges.
Z ∞
1
sin(ωx) dx = π
−∞ x
Z ∞
1 π
sin(ωx) dx =
0 x 2

Thus we have evaluated an integral that we used in deriving the Fourier transform.

34.3.3 Analytic Continuation

Consider the Fourier transform of f (x) = 1. The Fourier integral is not convergent, and its principal value
does not exist. Thus we will have to be a little creative in order to define the Fourier transform. Define the two

1426
functions

 
1
 for x > 0 0
 for x > 0
f+ (x) = 1/2 for x = 0 , f− (x) = 1/2 for x = 0 .
 
0 for x < 0 1 for x < 0
 

Note that 1 = f− (x) + f+ (x).

The Fourier transform of f+ (x) converges for =(ω) < 0.

Z ∞
1
F[f+ (x)] = e−iωx dx
2π 0
Z ∞
1
= e(−i<(ω)+=(ω))x dx.
2π 0
∞
1 e−iωx

=
2π −iω 0
i
=− for =(ω) < 0
2πω

Using analytic continuation, we can define the Fourier transform of f+ (x) for all ω except the point ω = 0.

i
F[f+ (x)] = −
2πω

1427
 We follow the same procedure for f− (x). The integral converges for =(ω) > 0.
Z 0
1
F[f− (x)] = e−iωx dx
2π −∞
Z 0
1
= e(−i<(ω)+=(ω))x dx
2π −∞
0
1 e−iωx

=
2π −iω −∞
i
= .
2πω
Using analytic continuation we can define the transform for all nonzero ω.
i
F[f− (x)] =
2πω

Now we are prepared to define the Fourier transform of f (x) = 1.

F[1] = F[f− (x)] + F[f+ (x)]
i i
=− +
2πω 2πω
= 0, for ω 6= 0
When ω = 0 the integral diverges. When we consider the closure relation for the Fourier transform we will see
that
F[1] = δ(ω).

34.4 Properties of the Fourier Transform

In this section we will explore various properties of the Fourier Transform. I would like to avoid stating assumptions
on various functions at the beginning of each subsection. Unless otherwise indicated, assume that the integrals

1428
converge.

34.4.1 Closure Relation.

Recall the closure relation for an orthonormal set of functions {φ1 , φ2 , . . . },

∞
X
φn (x)φn (ξ) ∼ δ(x − ξ).
n=1

There is a similar closure relation for Fourier integrals. We compute the Fourier transform of δ(x − ξ).

Z ∞
1
F[δ(x − ξ)] = δ(x − ξ) e−iωx dx
2π −∞
1 −iωξ
= e
2π

Next we take the inverse Fourier transform.

Z ∞
1 −iωξ iωx
δ(x − ξ) ∼ e e dω
−∞ 2π
Z ∞
1
δ(x − ξ) ∼ eiω(x−ξ) dω.
2π −∞

Note that the integral is divergent, but it would be impossible to represent δ(x − ξ) with a convergent integral.

1429
34.4.2 Fourier Transform of a Derivative.
Consider the Fourier transform of y 0 (x).
Z ∞
0 1
F[y (x)] = y 0 (x) e−iωx dx
2π −∞
 ∞ Z ∞
1 −iωx 1
= y(x) e − (−iω)y(x) e−iωx dx
2π −∞ 2π −∞
Z ∞
1
= iω y(x) e−iωx dx
2π −∞
= iωF[y(x)]

Next consider y 00 (x).

 
00 d 0
F[y (x)] = F (y (x))
dx
= iωF[y 0 (x)]
= (iω)2 F[y(x)]
= −ω 2 F[y(x)]

In general,

F y (n) (x) = (iω)n F[y(x)].

 

Example 34.4.1 The Dirac delta function can be expressed as the derivative of the Heaviside function.
(
0 for x < c,
H(x − c) =
1 for x > c

1430
Thus we can express the Fourier transform of H(x − c) in terms of the Fourier transform of the delta function.
F[δ(x − c)] = iωF[H(x − c)]
Z ∞
1
δ(x − c) e−iωx dx = iωF[H(x − c)]
2π −∞
1 −icω
e = iωF[H(x − c)]
2π

1 −icω
F[H(x − c)] = e
2πiω

34.4.3 Fourier Convolution Theorem.

Consider the Fourier transform of a product of two functions.
Z ∞
1
F[f (x)g(x)] = f (x)g(x) e−iωx dx
2π −∞
Z ∞ Z ∞ 
1 ˆ
= f (η) e dη g(x) e−iωx dx
iηx
2π −∞ −∞
Z ∞ Z ∞ 
1 ˆ i(η−ω)x
= f (η)g(x) e dx dη
2π −∞ −∞
Z ∞  Z ∞ 
1
= fˆ(η) g(x) e −i(ω−η)x
dx dη
−∞ 2π −∞
Z ∞
= fˆ(η)G(ω − η) dη
−∞

The convolution of two functions is defined

Z ∞
f ∗ g(x) = f (ξ)g(x − ξ) dξ.
−∞

1431
Thus
Z ∞
F[f (x)g(x)] = fˆ ∗ ĝ(ω) = fˆ(η)ĝ(ω − η) dη.
−∞

Now consider the inverse Fourier Transform of a product of two functions.

Z ∞
−1 ˆ
F [f (ω)ĝ(ω)] = fˆ(ω)ĝ(ω) eiωx dω
Z−∞∞  Z ∞ 
1 −iωξ
= f (ξ) e dξ ĝ(ω) eiωx dω
−∞ 2π
Z ∞ Z−∞ ∞ 
1 iω(x−ξ)
= f (ξ)ĝ(ω) e dω dξ
2π −∞ −∞
Z ∞ Z ∞ 
1 iω(x−ξ)
= f (ξ) ĝ(ω) e dω dξ
2π −∞ −∞
Z ∞
1
= f (ξ)g(x − ξ) dξ
2π −∞
Thus
Z ∞
1 1
F −1
[fˆ(ω)ĝ(ω)] = f ∗ g(x) = f (ξ)g(x − ξ) dξ,
2π 2π −∞

F[f ∗ g(x)] = 2π fˆ(ω)ĝ(ω).

These relations are known as the Fourier convolution theorem.

Example 34.4.2 Using the convolution theorem and the table of Fourier transform pairs in the appendix, we
can find the Fourier transform of
1
f (x) = .
x4 + 5x2 + 4

1432
We factor the fraction.

1
f (x) =
(x2 + 1)(x2 + 4)

From the table, we know that

 
2c
F 2 2
= e−c|ω| for c > 0.
x +c

We apply the convolution theorem.

 
1 2 4
F[f (x)] = F
8 x 2 + 1 x2 + 4
Z ∞ 
1 −|η| −2|ω−η|
= e e dη
8 −∞
Z 0 Z ∞ 
1 η −2|ω−η| −η −2|ω−η|
= e e dη + e e dη
8 −∞ 0

First consider the case ω > 0.

Z 0 Z ω Z ∞ 
1 −2ω+3η −2ω+η 2ω−3η
F[f (x)] = e dη + e dη + e dη
8 −∞ 0 ω
 
1 1 −2ω −ω −2ω 1 −ω
= e +e −e + e
8 3 3
1 1 −2ω
= e−ω − e
6 12

1433
Now consider the case ω < 0.
Z ω Z 0 Z ∞ 
1 −2ω+3η 2ω−η 2ω−3η
F[f (x)] = e dη + e dη + e dη
8 −∞ ω 0
 
1 1 ω 2ω ω 1 2ω
= e −e +e + e
8 3 3
1 1 2ω
= eω − e
6 12

We collect the result for positive and negative ω.

1 −|ω| 1 −2|ω|
F[f (x)] = e − e
6 12

A better way to find the Fourier transform of

1
f (x) =
x4 + 5x2 + 4

is to first expand the function in partial fractions.

1/3 1/3
f (x) = − 2
x2+1 x +4

   
1 2 1 4
F[f (x)] = F 2 − F 2
6 x +1 12 x +4
1 1 −2|ω|
= e−|ω| − e
6 12

1434
34.4.4 Parseval’s Theorem.
Recall
P∞ Parseval’s theorem for Fourier series. If f (x) is a complex valued function with the Fourier series
inx
c
n=−∞ n e then
∞
X Z π
2
2π |cn | = |f (x)|2 dx.
n=−∞ −π

Analogous to this result is Parseval’s theorem for Fourier transforms.

Let f (x) be a complex valued function that is both absolutely integrable and square integrable.
Z ∞ Z ∞
|f (x)| dx < ∞ and |f (x)|2 dx < ∞
−∞ −∞

The Fourier transform of f (−x) is fˆ(ω).

Z ∞
h i 1
F f (−x) = f (−x) e−iωx dx
2π −∞
Z −∞
1
=− f (x) eiωx dx
2π ∞
Z ∞
1
= f (x) e−iωx dx
2π −∞
= fˆ(ω)
We apply the convolution theorem.
Z ∞
F −1
[2π fˆ(ω)fˆ(ω)] =f (ξ)f (−(x − ξ)) dξ
−∞
Z ∞ Z ∞
ˆ ˆ iωx
2π f (ω)f (ω) e dω = f (ξ)f (ξ − x) dξ
−∞ −∞

1435
We set x = 0.
Z ∞ Z ∞
2π fˆ(ω)fˆ(ω) dω = f (ξ)f (ξ) dξ
−∞ −∞
Z ∞ Z ∞
2π |fˆ(ω)|2 dω = |f (x)|2 dx
−∞ −∞

This is known as Parseval’s theorem.

34.4.5 Shift Property.

The Fourier transform of f (x + c) is
Z ∞
1
F[f (x + c)] = f (x + c) e−iωx dx
2π −∞
Z ∞
1
= f (x) e−iω(x−c) dx
2π −∞

F[f (x + c)] = eiωc fˆ(ω)

The inverse Fourier transform of fˆ(ω + c) is

Z ∞
F −1
[fˆ(ω + c)] = fˆ(ω + c) eiωx dω
Z−∞
∞
= fˆ(ω) ei(ω−c)x dω
−∞

F −1 [fˆ(ω + c)] = e−icx f (x)

1436
34.4.6 Fourier Transform of x f(x).
The Fourier transform of xf (x) is
Z ∞
1
F[xf (x)] = xf (x) e−iωx dx
2π −∞
Z ∞
1 ∂
= if (x) ( e−iωx ) dx
2π −∞ ∂ω
 Z ∞ 
∂ 1 −iωx
=i f (x) e dx
∂ω 2π −∞

∂ fˆ
F[xf (x)] = i .
∂ω
Similarly, you can show that

n n∂ fˆ
n
F[x f (x)] = (i) .
∂ω n

34.5 Solving Differential Equations with the Fourier Transform

The Fourier transform is useful in solving some differential equations on the domain (−∞ . . . ∞) with homogeneous
boundary conditions at infinity. We take the Fourier transform of the differential equation L[y] = f and solve
for ŷ. We take the inverse transform to determine the solution y. Note that this process is only applicable if the
Fourier transform of y exists. Hence the requirement for homogeneous boundary conditions at infinity.
We will use the table of Fourier transforms in the appendix in solving the examples in this section.

Example 34.5.1 Consider the problem

y 00 − y = e−α|x| , y(±∞) = 0, α > 0, α 6= 1.

1437
We take the Fourier transform of this equation.
α/π
−ω 2 ŷ(ω) − ŷ(ω) =
ω2+ α2
We take the inverse Fourier transform to determine the solution.
−α/π
ŷ(ω) =
(ω 2 + α2 )(ω 2 + 1)
 
−α 1 1 1
= −
π α2 − 1 ω 2 + 1 ω 2 + α2
 
1 α/π 1/π
= 2 −α 2
α − 1 ω 2 + α2 ω +1

e−α|x| − α e−|x|
y(x) =
α2 − 1

Example 34.5.2 Consider the Green function problem

G00 − G = δ(x − ξ), y(±∞) = 0.

We take the Fourier transform of this equation.

−ω 2 Ĝ − Ĝ = F[δ(x − ξ)]
1
Ĝ = − 2 F[δ(x − ξ)]
ω +1

We use the Table of Fourier transforms.

Ĝ = −πF e−|x| F[δ(x − ξ)]

 

1438
We use the convolution theorem to do the inversion.

Z ∞
1
G = −π e−|x−η| δ(η − ξ) dη
2π −∞
1
G(x|ξ) = − e|x−ξ|
2

The inhomogeneous differential equation

y 00 − y = f (x), y(±∞) = 0,

has the solution

Z ∞
1
y=− f (ξ) e−|x−ξ| dξ.
2 −∞

When solving the differential equation L[y] = f with the Fourier transform, it is quite common to use the
convolution theorem. With this approach we have no need to compute the Fourier transform of the right side.
We merely denote it as F[f ] until we use f in the convolution integral.

1439
34.6 The Fourier Cosine and Sine Transform
34.6.1 The Fourier Cosine Transform
Suppose f (x) is an even function. In this case the Fourier transform of f (x) coincides with the Fourier cosine
transform of f (x).
Z ∞
1
F[f (x)] = f (x) e−iωx dx
2π −∞
Z ∞
1
= f (x)(cos(ωx) − i sin(ωx)) dx
2π −∞
Z ∞
1
= f (x) cos(ωx) dx
2π −∞
1 ∞
Z
= f (x) cos(ωx) dx
π 0
The Fourier cosine transform is defined:
Z ∞
1
Fc [f (x)] = fˆc (ω) = f (x) cos(ωx) dx.
π 0

Note that fˆc (ω) is an even function. The inverse Fourier cosine transform is
Z ∞
−1 ˆ
Fc [fc (ω)] = fˆc (ω) eiωx dω
Z−∞
∞
= fˆc (ω)(cos(ωx) + i sin(ωx)) dω
Z−∞
∞
= fˆc (ω) cos(ωx) dω
−∞
Z ∞
=2 fˆc (ω) cos(ωx) dω.
0

1440
 Thus we have the Fourier cosine transform pair
Z ∞ Z ∞
−1 ˆ 1
f (x) = Fc [fc (ω)] = 2 fˆc (ω) cos(ωx) dω, fˆc (ω) = Fc [f (x)] = f (x) cos(ωx) dx.
0 π 0

34.6.2 The Fourier Sine Transform

Suppose f (x) is an odd function. In this case the Fourier transform of f (x) coincides with the Fourier sine
transform of f (x).
Z ∞
1
F[f (x)] = f (x) e−iωx dx
2π −∞
Z ∞
1
= f (x)(cos(ωx) − i sin(ωx)) dx
2π −∞
i ∞
Z
=− f (x) sin(ωx) dx
π 0

Note that fˆ(ω) = F[f (x)] is an odd function of ω. The inverse Fourier transform of fˆ(ω) is
Z ∞
−1 ˆ
F [f (ω)] = fˆ(ω) eiωx dω
−∞
Z ∞
= 2i fˆ(ω) sin(ωx) dω.
0

Thus we have that

∞
i ∞
Z  Z 
f (x) = 2i − f (x) sin(ωx) dx sin(ωx) dω
0 π 0
Z ∞ Z ∞ 
1
=2 f (x) sin(ωx) dx sin(ωx) dω.
0 π 0

1441
This gives us the Fourier sine transform pair
Z ∞ Z ∞
1
f (x) = Fs−1 [fˆs (ω)] =2 fˆs (ω) sin(ωx) dω, fˆs (ω) = Fs [f (x)] = f (x) sin(ωx) dx.
0 π 0

Result 34.6.1 The Fourier cosine transform pair is defined:

Z ∞
f (x) = Fc−1 [fˆc (ω)] = 2 fˆc (ω) cos(ωx) dω
Z ∞0
1
fˆc (ω) = Fc [f (x)] = f (x) cos(ωx) dx
π 0
The Fourier sine transform pair is defined:
Z ∞
f (x) = Fs−1 [fˆs (ω)]
=2 fˆs (ω) sin(ωx) dω
Z ∞0
1
fˆs (ω) = Fs [f (x)] = f (x) sin(ωx) dx
π 0

34.7 Properties of the Fourier Cosine and Sine Transform

34.7.1 Transforms of Derivatives

Cosine Transform. Using integration by parts we can find the Fourier cosine transform of derivatives. Let
y be a function for which the Fourier cosine transform of y and its first and second derivatives exists. Further

1442
assume that y and y 0 vanish at infinity. We calculate the transforms of the first and second derivatives.

1 ∞ 0
Z
0
Fc [y ] = y cos(ωx) dx
π 0
∞ ω ∞
Z
1
= y cos(ωx) 0 + y sin(ωx) dx
π π 0
1
= ω ŷc (ω) − y(0)
π
1 ∞ 00
Z
00
Fc [y ] = y cos(ωx) dx
π 0
∞ ω ∞ 0
Z
1 0
= y cos(ωx) 0 + y sin(ωx) dx
π π 0
∞ ω 2 ∞
Z
1 0 ω
= − y (0) + y sin(ωx) 0 − y cos(ωx) dx
π π π 0
1
= −ω 2 fˆc (ω) − y 0 (0)
π

Sine Transform. You can show, (see Exercise 34.3), that the Fourier sine transform of the first and second
derivatives are

Fs [y 0 ] = −ω fˆc (ω)
ω
Fs [y 00 ] = −ω 2 ŷc (ω) + y(0).
π

1443
34.7.2 Convolution Theorems

Cosine Transform of a Product. Consider the Fourier cosine transform of a product of functions. Let f (x)
and g(x) be two functions defined for x ≥ 0. Let Fc [f (x)] = fˆc (ω), and Fc [g(x)] = ĝc (ω).

1 ∞
Z
Fc [f (x)g(x)] = f (x)g(x) cos(ωx) dx
π 0
1 ∞
Z  Z ∞ 
= 2 fˆc (η) cos(ηx) dη g(x) cos(ωx) dx
π 0 0
2 ∞ ∞ ˆ
Z Z
= fc (η)g(x) cos(ηx) cos(ωx) dx dη
π 0 0

We use the identity cos a cos b = 21 (cos(a − b) + cos(a + b)).

1 ∞ ∞ ˆ
Z Z


= fc (η)g(x) cos((ω − η)x) + cos((ω + η)x) dx dη
π
Z ∞0 0
 Z ∞ Z ∞ 
1 1
= fˆc (η) g(x) cos((ω − η)x) dx + g(x) cos((ω + η)x) dx dη
0 π 0 π 0
Z ∞
fˆc (η) ĝc (ω − η) + ĝc (ω + η) dη


=
0

ĝc (ω) is an even function. If we have only defined ĝc (ω) for positive argument, then ĝc (ω) = ĝc (|ω|).

Z ∞
fˆc (η) ĝc (|ω − η|) + ĝc (ω + η) dη


=
0

1444
Inverse Cosine Transform of a Product. Now consider the inverse Fourier cosine transform of a product of
functions. Let Fc [f (x)] = fˆc (ω), and Fc [g(x)] = ĝc (ω).
Z ∞
Fc−1 [fˆc (ω)ĝc (ω)] =2 fˆc (ω)ĝc (ω) cos(ωx) dω
Z0 ∞  Z ∞ 
1
=2 f (ξ) cos(ωξ) dξ ĝc (ω) cos(ωx) dω
0 π 0
2 ∞ ∞
Z Z
= f (ξ)ĝc (ω) cos(ωξ) cos(ωx) dω dξ
π 0 0
1 ∞ ∞
Z Z


= f (ξ)ĝc (ω) cos(ω(x − ξ)) + cos(ω(x + ξ)) dω dξ
π 0
Z ∞0  Z ∞ Z ∞ 
1
= f (ξ) 2 ĝc (ω) cos(ω(x − ξ)) dω + 2 ĝc (ω) cos(ω(x + ξ)) dω dξ
2π 0 0 0
Z ∞
1

= f (ξ) g(|x − ξ|) + g(x + ξ) dξ
2π 0

Sine Transform of a Product. You can show, (see Exercise 34.5), that the Fourier sine transform of a product
of functions is
Z ∞
fˆs (η) ĝc (|ω − η|) − ĝc (ω + η) dη.


Fs [f (x)g(x)] =
0

Inverse Sine Transform of a Product. You can also show, (see Exercise 34.6), that the inverse Fourier sine
transform of a product of functions is
Z ∞
1
Fs−1 [fˆs (ω)ĝc (ω)]


= f (ξ) g(|x − ξ|) − g(x + ξ) dξ.
2π 0

1445
 Result 34.7.1 The Fourier cosine and sine transform convolution theorems are
Z ∞
fˆc (η) ĝc (|ω − η|) + ĝc (ω + η) dη
 
Fc [f (x)g(x)] =
0 Z ∞
−1 ˆ 1

Fc [fc (ω)ĝc (ω)] = f (ξ) g(|x − ξ|) + g(x + ξ) dξ
2π
Z ∞ 0
fˆs (η) ĝc (|ω − η|) − ĝc (ω + η) dη


Fs [f (x)g(x)] =
0 Z ∞
1
Fs−1 [fˆs (ω)ĝc (ω)] =


f (ξ) g(|x − ξ|) − g(x + ξ) dξ
2π 0

34.7.3 Cosine and Sine Transform in Terms of the Fourier Transform

We can express the Fourier cosine and sine transform in terms of the Fourier transform. First consider the Fourier
cosine transform. Let f (x) be an even function.
1 ∞
Z
Fc [f (x)] = f (x) cos(ωx) dx
π 0
We extend the domain integration because the integrand is even.
Z ∞
1
= f (x) cos(ωx) dx
2π −∞
R∞
Note that −∞ f (x) sin(ωx) dx = 0 because the integrand is odd.
Z ∞
1
= f (x) e−iωx dx
2π −∞
= F[f (x)]

1446
 Fc [f (x)] = F[f (x)], for even f (x).
For general f (x), use the even extension, f (|x|) to write the result.
Fc [f (x)] = F[f (|x|)]
There is an analogous result for the inverse Fourier cosine transform.
h i h i
Fc−1 fˆ(ω) = F −1 fˆ(|ω|)

For the sine series, we have

h i h i
Fs [f (x)] = iF [ sign (x)f (|x|)] Fs−1 ˆ
f (ω) = −iF −1 ˆ
sign (ω)f (|ω|)

Result 34.7.2 The results:

h i h i
Fc [f (x)] = F [f (|x|)] Fc−1 ˆ
f (ω) = F −1 ˆ
f (|ω|)

h i h i
Fs [f (x)] = iF [ sign (x)f (|x|)] Fs−1 ˆ
f (ω) = −iF −1 ˆ
sign (ω)f (|ω|)

allow us to evaluate Fourier cosine and sine transforms in terms of the Fourier transform.
This enables us to use contour integration methods to do the integrals.

34.8 Solving Differential Equations with the Fourier Cosine and

Sine Transforms
Example 34.8.1 Consider the problem
y 00 − y = 0, y(0) = 1, y(∞) = 0.

1447
Since the initial condition is y(0) = 1 and the sine transform of y 00 is −ω 2 ŷc (ω) + ωπ y(0) we take the Fourier sine
transform of both sides of the differential equation.
ω
−ω 2 ŷc (ω) + y(0) − ŷc (ω) = 0
π
2 ω
−(ω + 1)ŷc (ω) = −
π
ω
ŷc (ω) =
π(ω 2 + 1)
We use the table of Fourier Sine transforms.
y = e−x

Example 34.8.2 Consider the problem

y 00 − y = e−2x , y 0 (0) = 0, y(∞) = 0.
Since the initial condition is y 0 (0) = 0, we take the Fourier cosine transform of the differential equation. From
the table of cosine transforms, Fc [ e−2x ] = 2/(π(ω 2 + 4)).
1 2
−ω 2 ŷc (ω) − y 0 (0) − ŷc (ω) = 2
π π(ω + 4)

2
ŷc (ω) = −
π(ω 2
+ 4)(ω 2 + 1)
 
−2 1/3 1/3
= −
π ω2 + 1 ω2 + 4
1 2/π 2 1/π
= −
3 ω + 4 3 ω2 + 1
2

1 −2x 2 −x
y= e − e
3 3

1448
34.9 Exercises
Exercise 34.1
Show that
sin(cω)
H(x + c) − H(x − c) = .
πω

Exercise 34.2
Using contour integration, find the Fourier transform of
1
f (x) = ,
x2 + c2
where <(c) 6= 0

Exercise 34.3
Find the Fourier sine transforms of y 0 (x) and y 00 (x).

Exercise 34.4
Prove the following identities.
1. F[f (x − a)] = e−iωa fˆ(ω)
1 ˆ ω 
2. F[f (ax)] = f
|a| a

Exercise 34.5
Show that
Z ∞
fˆs (η) ĝc (|ω − η|) − ĝc (ω + η) dη.


Fs [f (x)g(x)] =
0

1449
Exercise 34.6
Show that
Z ∞
1
Fs−1 [fˆs (ω)ĝc (ω)]


= f (ξ) g(|x − ξ|) − g(x + ξ) dξ.
2π 0

Exercise 34.7
Let fˆc (ω) = Fc [f (x)], fˆc (ω) = Fs [f (x)], and assume the cosine and sine transforms of xf (x) exist. Express
Fc [xf (x)] and Fs [xf (x)] in terms of fˆc (ω) and fˆc (ω).

Exercise 34.8
Solve the problem

y 00 − y = e−2x , y(0) = 1, y(∞) = 0,

using the Fourier sine transform.

Exercise 34.9
Show that
h i h i
Fs [f (x)] = iF[ sign (x)f (|x|)] Fs−1 fˆ(ω) = −iF −1 sign (ω)fˆ(|ω|)

Exercise 34.10
Let fˆc (ω) = Fc [f (x)] and fˆc (ω) = Fs [f (x)]. Show that
∂ ˆ
1. Fc [xf (x)] = f (ω)
∂ω c

∂ ˆ
2. Fs [xf (x)] = − ∂ω fc (ω)

1450
 3. Fc [f (cx)] = 1c fˆc ω


c
for c > 0

4. Fs [f (cx)] = 1c fˆc ω


c
for c > 0.

Exercise 34.11
Solve the integral equation,
Z ∞
2 2
u(ξ) e−a(x−ξ) dξ = e−bx ,
−∞

where a, b > 0, a 6= b, with the Fourier transform.

Exercise 34.12
Evaluate
Z ∞
1 1 −cx
e sin(ωx) dx,
π 0 x
where ω is a positive, real number and <(c) > 0.

Exercise 34.13
Use the Fourier transform to solve the equation

y 00 − a2 y = e−a|x|
on the domain −∞ < x < ∞ with boundary conditions y(±∞) = 0.

Exercise 34.14
1. Use the cosine transform to solve
y 00 − a2 y = 0 on x ≥ 0 with y 0 (0) = b, y(∞) = 0.

1451
 2. Use the cosine transform to show that the Green function for the above with b = 0 is
1 −a|x−ξ| 1 −a(x−ξ)
G(x, ξ) = − e − e .
2a 2a

Exercise 34.15
1. Use the sine transform to solve

y 00 − a2 y = 0 on x ≥ 0 with y(0) = b, y(∞) = 0.

2. Try using the Laplace transform on this problem. Why isn’t it as convenient as the Fourier transform?
3. Use the sine transform to show that the Green function for the above with b = 0 is
1
e−a(x−ξ) − e−a|x+ξ|


g(x; ξ) =
2a

Exercise 34.16
1. Find the Green function which solves the equation

y 00 + 2µy 0 + (β 2 + µ2 )y = δ(x − ξ), µ > 0, β > 0,

in the range −∞ < x < ∞ with boundary conditions y(−∞) = y(∞) = 0.

2. Use this Green’s function to show that the solution of

y 00 + 2µy 0 + (β 2 + µ2 )y = g(x), µ > 0, β > 0, y(−∞) = y(∞) = 0,

with g(±∞) = 0 in the limit as µ → 0 is

Z x
1
y= g(ξ) sin[β(x − ξ)]dξ.
β −∞

You may assume that the interchange of limits is permitted.

1452
Exercise 34.17
Using Fourier transforms, find the solution u(x) to the integral equation
Z ∞
u(ξ) 1
2 2
dξ = 2 0 < a < b.
−∞ [(x − ξ) + a ] x + b2

Exercise 34.18
The Fourer cosine transform is defined by
Z ∞
1
Fc (ω) = f (x) cos(ωx) dx.
π 0

1. From the Fourier theorem show that the inverse cosine transform is given by
Z ∞
f (x) = 2 Fc (ω) cos(ωx) dω.
0

2. Show that the cosine transform of f 00 (x) is

f 0 (0)
−ω 2 Fc (ω) − .
π

3. Use the cosine transform to solve

y 00 − a2 y = 0 on x > 0 with y 0 (0) = b, y(∞) = 0.

Exercise 34.19
The Fourier sine transform is defined by
Z ∞
1
Fs (ω) = f (x) sin(ωx) dx.
π 0

1453
 1. Show that the inverse sine transform is given by
Z ∞
f (x) = 2 Fs (ω) sin(ωx) dω.
0

2. Show that the sine transform of f 00 (x) is

ω
f (0) − ω 2 Fs (ω).
π

3. Use this property to solve the equation

y 00 − a2 y = 0 on x > 0 with y(0) = b, y(∞) = 0.

4. Try using the Laplace transform on this problem. Why isn’t it as convenient as the Fourier transform?

Exercise 34.20
Show that
1
F[f (x)] = (Fc [f (x) + f (−x)] − iFs [f (x) − f (−x)])
2
where F, Fc and Fs are respectively the Fourier transform, Fourier cosine transform and Fourier sine transform.

Exercise 34.21
Find u(x) as the solution to the integral equation:
Z ∞
u(ξ) 1
2 2
dξ = 2 , 0 < a < b.
−∞ (x − ξ) + a x + b2
Use Fourier transforms and the inverse transform. Justify the choice of any contours used in the complex plane.

1454
34.10 Hints
Hint 34.1

(
1 for |x| < c,
H(x + c) − H(x − c) =
0 for |x| > c

Hint 34.2
Consider the two cases <(ω) < 0 and <(ω) > 0, closing the path of integration with a semi-circle in the lower or
upper half plane.

Hint 34.3

Hint 34.4

Hint 34.5

Hint 34.6

Hint 34.7

1455
Hint 34.8

Hint 34.9

Hint 34.10

Hint 34.11
2
The left side is the convolution of u(x) and e−ax .

Hint 34.12

Hint 34.13

Hint 34.14

Hint 34.15

1456
Hint 34.16

Hint 34.17

Hint 34.18

Hint 34.19

Hint 34.20

Hint 34.21

1457
34.11 Solutions
Solution 34.1

Z ∞
1
F[H(x + c) − H(x − c)] = (H(x + c) − H(x − c)) e−iωx dx
2π −∞
Z c
1
= e−iωx dx
2π −c
c
1 e−iωx

=
2π −iω −c
 −iωc
eiωc

1 e
= −
2π −iω −iω

sin(cω)
F[H(x + c) − H(x − c)] =
πω

Solution 34.2

  Z ∞
1 1 1
F 2 2
= e−iωx dx
x +c 2π −∞ x + c2
2
Z ∞
1 e−iωx
= dx
2π −∞ (x − ic)(x + ic)

If <(ω) < 0 then we close the path of integration with a semi-circle in the upper half plane.

e−iωx
   
1 1 1 cω
F 2 2
= 2πi Res , x = ic = e
x +c 2π (x − ic)(x + ic) 2c

1458
If ω > 0 then we close the path of integration in the lower half plane.

e−iωx
   
1 1 1 −cω
F 2 = − 2πi Res , −ic = e
x + c2 2π (x − ic)(x + ic) 2c

Thus we have that

 
1 1 −c|ω|
F 2 2
= e , for <(c) 6= 0.
x +c 2c

Solution 34.3

1 ∞ 0
Z
0
Fs [y ] = y sin(ωx) dx
π 0
1h i∞ ω Z ∞
= y sin(ωx) − y cos(ωx) dx
π 0 π 0
= −ω ŷc (ω)
1 ∞ 00
Z
00
Fs [y ] = y sin(ωx) dx
π 0
1h 0 i∞ ω Z ∞
= y sin(ωx) − y 0 cos(ωx) dx
π 0 π 0
ω2 ∞
∞
Z
ω h i
= − y cos(ωx) − y sin(ωx) dx
π 0 π 0
ω
= −ω 2 ŷs (ω) + y(0).
π

1459
Solution 34.4
1.
Z ∞
1
F[f (x − a)] = f (x − a) e−iωx dx
2π −∞
Z ∞
1
= f (x) e−iω(x+a) dx
2π −∞
Z ∞
−iωa 1
= e f (x) e−iωx dx
2π −∞

F[f (x − a)] = e−iωa fˆ(ω)

2. If a > 0, then
Z ∞
1
F[f (ax)] = f (ax) e−iωx dx
2π −∞
Z ∞
1 1
= f (ξ) e−iωξ/a dξ
2π −∞ a
1  ω 
= fˆ .
a a
If a < 0, then
Z ∞
1
F[f (ax)] = f (ax) e−iωx dx
2π −∞
Z −∞
1 1
= e−iωξ/a dξ
2π ∞ a
1  ω 
= − fˆ .
a a

1460
 Thus

1 ˆ ω 
F[f (ax)] = f .
|a| a

Solution 34.5

1 ∞
Z
Fs [f (x)g(x)] = f (x)g(x) sin(ωx) dx
π 0
1 ∞
Z  Z ∞ 
= 2 ˆ
fs (η) sin(ηx) dη g(x) sin(ωx) dx
π 0 0
2 ∞ ∞ ˆ
Z Z
= fs (η)g(x) sin(ηx) sin(ωx) dx dη
π 0 0

Use the identity, sin a sin b = 12 [cos(a − b) − cos(a + b)].

Z ∞ Z ∞
1 ˆ
h i
= fs (η)g(x) cos((ω − η)x) − cos((ω + η)x) dx dη
π 0 0
∞  Z ∞
1 ∞
Z Z 
ˆ 1
= fs (η) g(x) cos((ω − η)x) dx − g(x) cos((ω + η)x) dx dη
0 π 0 π 0

Z ∞
fˆs (η) Gc (|ω − η|) − Gc (ω + η) dη
 
Fs [f (x)g(x)] =
0

1461
Solution 34.6

Z ∞
Fs−1 [fˆs (ω)Gc (ω)] =2 fˆs (ω)Gc (ω) sin(ωx) dω
0
Z ∞ Z ∞ 
1
=2 f (ξ) sin(ωξ) dξ Gc (ω) sin(ωx) dω
π
Z0 ∞ Z ∞ 0
2
= f (ξ)Gc (ω) sin(ωξ) sin(ωx) dω dξ
π 0
Z ∞ Z0 ∞
1 h i
= f (ξ)Gc (ω) cos(ω(x − ξ)) − cos(ω(x + ξ)) dω dξ
π 0
Z ∞0  Z ∞ Z ∞ 
1
= f (ξ) 2 Gc (ω) cos(ω(x − ξ)) dω − 2 Gc (ω) cos(ω(x + ξ)) dω) dξ
2π 0 0 0
Z ∞
1
= f (ξ)[g(x − ξ) − g(x + ξ)] dξ
2π 0

Z ∞
1
Fs−1 [fˆs (ω)Gc (ω)]
 
= f (ξ) g(|x − ξ|) − g(x + ξ) dξ
2π 0

1462
Solution 34.7

1 ∞
Z
Fc [xf (x)] = xf (x) cos(ωx) dx
π 0
1 ∞
Z
∂
= f (x) (sin(ωx)) dx
π 0 ∂ω
Z ∞
∂ 1
= f (x) sin(ωx) dx
∂ω π 0
∂ ˆ
= fs (ω)
∂ω

1 ∞
Z
Fs [xf (x)] = xf (x) sin(ωx) dx
π 0
1 ∞
Z
∂
= f (x) (− cos(ωx)) dx
π 0 ∂ω
Z ∞
∂ 1
=− f (x) cos(ωx) dx
∂ω π 0
∂
= − fˆc (ω)
∂ω

Solution 34.8

y 00 − y = e−2x , y(0) = 1, y(∞) = 0

We take the Fourier sine transform of the differential equation.

ω 2ω/π
−ω 2 ŷs (ω) + y(0) − ŷs (ω) = 2
π ω +4

1463
 ω/π ω/π
ŷs (ω) = − + 2
(ω 2 2
+ 4)(ω + 1) (ω + 1)
ω/(3π) ω/(3π) ω/π
= 2 − 2 + 2
ω +4 ω +1 ω +1
2 ω/π 1 ω/π
= +
3 ω + 1 3 ω2 + 4
2

2 −x 1 −2x
y= e + e
3 3

Solution 34.9
Consider the Fourier sine transform. Let f (x) be an odd function.

1 ∞
Z
Fs [f (x)] = f (x) sin(ωx) dx
π 0

Extend the integration because the integrand is even.

Z ∞
1
= f (x) sin(ωx) dx
2π −∞

R∞
Note that −∞
f (x) cos(ωx) dx = 0 as the integrand is odd.
Z ∞
1
= f (x)i e−iωx dx
2π −∞
= iF[f (x)]

Fs [f (x)] = iF[f (x)], for odd f (x).

1464
For general f (x), use the odd extension, sign (x)f (|x|) to write the result.

Fs [f (x)] = iF[ sign (x)f (|x|)]

Now consider the inverse Fourier sine transform. Let fˆ(ω) be an odd function.
h i Z ∞
Fs−1 fˆ(ω) = 2 fˆ(ω) sin(ωx) dω
0

Extend the integration because the integrand is even.

Z ∞
= fˆ(ω) sin(ωx) dω
−∞

R∞
Note that −∞
fˆ(ω) cos(ωx) dω = 0 as the integrand is odd.
Z ∞
= fˆ(ω)(−i) eiωx dω
−∞
h i
= −iF −1 fˆ(ω)

h i h i
Fs−1 ˆ
f (ω) = −iF −1 ˆ
f (ω) , for odd fˆ(ω).

For general fˆ(ω), use the odd extension, sign (ω)fˆ(|ω|) to write the result.
h i h i
Fs−1 fˆ(ω) = −iF −1 sign (ω)fˆ(|ω|)

1465
Solution 34.10

1 ∞
Z
Fc [xf (x)] = xf (x) cos(ωx) dx
π 0
1 ∞
Z
∂
= f (x) sin(ωx) dx
π 0 ∂ω
∂ 1 ∞
Z
= f (x) sin(ωx) dx
∂ω π 0
∂ ˆ
= fs (ω)
∂ω

1 ∞
Z
Fs [xf (x)] = xf (x) sin(ωx) dx
π 0
1 ∞
Z
∂
= f (x) (− cos(ωx)) dx
π 0 ∂ω
∂ 1 ∞
Z
=− f (x) cos(ωx) dx
∂ω π 0
∂
= − fˆc (ω)
∂ω

1 ∞
Z
Fc [f (cx)] = f (cx) cos(ωx) dx
π 0
1 ∞
Z  ω  dξ
= f (ξ) cos ξ
π 0 c c
1  ω 
= fˆc
c c

1466
 1 ∞
Z
Fs [f (cx)] = f (cx) sin(ωx) dx
π 0
1 ∞
Z  ω  dξ
= f (ξ) sin ξ
π 0 c c
1  ω 
= fˆs
c c

Solution 34.11

Z ∞
2 2
u(ξ) e−a(x−ξ) dξ = e−bx
−∞

We take the Fourier transform and solve for U (ω).

h 2
i h 2
i
2πU (ω)F e−ax = F e−bx
1 2 1 2
2πU (ω) √ e−ω /(4a) = √ e−ω /(4b)
4πa 4πb
r
1 a −ω2 (a−b)/(4ab)
U (ω) = e
2π b
Now we take the inverse Fourier transform.
r p
1 a 4πab/(a − b) −ω2 (a−b)/(4ab)
U (ω) = p e
2π b 4πab/(a − b)

a 2
u(x) = p e−abx /(a−b)
π(a − b)

1467
Solution 34.12

Z ∞
1 1 −cx
I= e sin(ωx) dx
π 0 x
Z ∞ Z ∞ 
1 −zx
= e dz sin(ωx) dx
π 0 c
Z ∞Z ∞
1
= e−zx sin(ωx) dx dz
π
Zc ∞ 0
1 ω
= dz
π c z + ω2
2

1 h  z i∞
= arctan
π ω c
1 π  c 
= − arctan
π 2 ω
1 ω 
= arctan
π c

Solution 34.13
We consider the differential equation

y 00 − a2 y = e−a|x|
on the domain −∞ < x < ∞ with boundary conditions y(±∞) = 0. We take the Fourier transform of the
differential equation.
a
−ω 2 ŷ − a2 ŷ =
π(ω + a2 )
2

We solve for ŷ(ω).

a
ŷ(ω) = −
π(ω 2 + a2 )2

1468
We take the inverse Fourier transform to find the solution of the differential equation.
Z ∞
a
y(x) = − eixω dω
−∞ π(ω 2 2
+a ) 2

Note that since ŷ(ω) is a real-valued, even function, y(x) is a real-valued, even function. Thus we only need to
evaluate the integral for positive x. If we replace x by |x| in this expression we will have the solution that is valid
for all x.
For x > 0, we evaluate the integral by closing the path of integration in the upper half plane and using the
Residue Theorem and Jordan’s Lemma.
a ∞
Z
1
y(x) = − eixω dω
π −∞ (ω − ia) (ω + ia)2
2
 
a 1 ixω
= −i2π Res e , ω = ia
π (ω − ia)2 (ω + ia)2
eixω
 
d
= −i2a lim
ω→ia dω (ω + ia)2
ix eixω 2 eixω
 
= −i2a lim −
ω→ia (ω + ia)2 (ω + ia)3
ix e−ax 2 e−ax
 
= −i2a −
−4a2 −i8a3
(1 + ax) e−ax
=− , for x ≥ 0
2a2
The solution of the differential equation is

1
y(x) = − 2
(1 + a|x|) e−a|x| .
2a

1469
Solution 34.14
1. We take the Fourier cosine transform of the differential equation.
b
−ω 2 ŷ(ω) − − a2 ŷ(ω) = 0
π
b
ŷ(ω) = −
π(ω + a2 )
2

Now we take the inverse Fourier cosine transform. We use the fact that ŷ(ω) is an even function.
 
−1 b
y(x) = Fc −
π(ω 2 + a2 )
 
−1 b
=F −
π(ω 2 + a2 )
 
b 1 iωx
= − i2π Res e , ω = ia
π ω 2 + a2
 iωx 
e
= −i2b lim , for x ≥ 0
ω→ia ω + ia

b
y(x) = − e−ax
a

2. The Green function problem is

g 00 − a2 g = δ(x − ξ) on x, ξ > 0, g 0 (0; ξ) = 0, g(∞; ξ) = 0.
We take the Fourier cosine transform and solve for ĝ(ω; ξ).
−ω 2 ĝ − a2 ĝ = Fc [δ(x − ξ)]
1
ĝ(ω; ξ) = − 2 Fc [δ(x − ξ)]
ω + a2

1470
 We express the right side as a product of Fourier cosine transforms.

π
ĝ(ω; ξ) = − Fc [ e−ax ]Fc [δ(x − ξ)]
a

Now we can apply the Fourier cosine convolution theorem,

Z ∞
1
Fc−1


[Fc [f (x)]Fc [g(x)]] = f (t) g(|x − t|) + g(x + t) dt,
2π 0

to obtain

Z ∞
π 1
δ(t − ξ) e−a|x−t| + e−a(x+t) dt


g(x; ξ) = −
a 2π 0
1
e−a|x−ξ| + e−a(x+ξ)


g(x; ξ) = −
2a

Solution 34.15
1. We take the Fourier sine transform of the differential equation.

bω
−ω 2 ŷ(ω) + − a2 ŷ(ω) = 0
π
bω
ŷ(ω) =
π(ω + a2 )
2

1471
 Now we take the inverse Fourier sine transform. We use the fact that ŷ(ω) is an odd function.
 
−1 bω
y(x) = Fs
π(ω 2 + a2 )
 
−1 bω
= −iF
π(ω 2 + a2 )
 
b ω iωx
= −i i2π Res e , ω = ia
π ω 2 + a2
 iωx 
ωe
= 2b lim
ω→ia ω + ia
−ax
= be for x ≥ 0

y(x) = b e−ax

2. Now we solve the differential equation with the Laplace transform.

y 00 − a2 y = 0
s2 ŷ(s) − sy(0) − y 0 (0) − a2 ŷ(s) = 0

We don’t know the value of y 0 (0), so we treat it as an unknown constant.

bs + y 0 (0)
ŷ(s) =
s 2 − a2
y 0 (0)
y(x) = b cosh(ax) + sinh(ax)
a
In order to satisfy the boundary condition at infinity we must choose y 0 (0) = −ab.

y(x) = b e−ax

1472
 We see that solving the differential equation with the Laplace transform is not as convenient, because
the boundary condition at infinity is not automatically satisfied. We had to find a value of y 0 (0) so that
y(∞) = 0.

3. The Green function problem is

g 00 − a2 g = δ(x − ξ) on x, ξ > 0, g(0; ξ) = 0, g(∞; ξ) = 0.

We take the Fourier sine transform and solve for ĝ(ω; ξ).

−ω 2 ĝ − a2 ĝ = Fs [δ(x − ξ)]
1
ĝ(ω; ξ) = − 2 Fs [δ(x − ξ)]
ω + a2

We write the right side as a product of Fourier cosine transforms and sine transforms.

π
ĝ(ω; ξ) = − Fc [ e−ax ]Fs [δ(x − ξ)]
a
Now we can apply the Fourier sine convolution theorem,
Z ∞
−1 1

Fs [Fs [f (x)]Fc [g(x)]] = f (t) g(|x − t|) − g(x + t) dt,
2π 0

to obtain
Z ∞
π 1
δ(t − ξ) e−a|x−t| − e−a(x+t) dt


g(x; ξ) = −
a 2π 0
1
e−a(x−ξ) − e−a|x+ξ|


g(x; ξ) =
2a

1473
Solution 34.16
1. We take the Fourier transform of the differential equation, solve for Ĝ and then invert.
G00 + 2µG0 + (β 2 + µ2 )G = δ(x − ξ)
e−iωξ
−ω 2 Ĝ + i2µω Ĝ + (β 2 + µ2 )Ĝ =
2π
e−iωξ
Ĝ = −
2π(ω 2 − i2µω − β 2 − µ2 )
Z ∞
e−iωξ eiωx
G= − dω
−∞ 2π(ω 2 − i2µω − β 2 − µ2 )
Z ∞
1 eiω(x−ξ)
G=− dω
2π −∞ (ω + β − iµ)(ω − β − iµ)
For x > ξ we close the path of integration in the upper half plane and use the Residue theorem. There are
two simple poles in the upper half plane. For x < ξ we close the path of integration in the lower half plane.
Since the integrand is analytic there, the integral is zero. G(x; ξ) = 0 for x < ξ. For x > ξ we have

eiω(x−ξ)
 
1
G(x; ξ) = − i2π Res , ω = −β + iµ
2π (ω + β − iµ)(ω − β − iµ)
!
eiω(x−ξ)

+ Res , ω = −β − iµ
(ω + β − iµ)(ω − β − iµ)

ei(−β+iµ)(x−ξ) ei(β+iµ)(x−ξ)
 
G(x; ξ) = −i
−2β 2β
1
G(x; ξ) = e−µ(x−ξ) sin(β(x − ξ)).
β
Thus the Green function is
1 −µ(x−ξ)
G(x; ξ) = e sin(β(x − ξ))H(x − ξ).
β

1474
 2. The solution of the inhomogeneous equation
y 00 + 2µy 0 + (β 2 + µ2 )y = g(x), y(−∞) = y(∞) = 0,
is
Z ∞
y(x) = g(ξ)G(x; ξ) dξ
−∞
Z ∞
1
y(x) = g(ξ) e−µ(x−ξ) sin(β(x − ξ)) dξ
−∞ β
Z x
1
y(x) = g(ξ) e−µ(x−ξ) sin(β(x − ξ)) dξ.
β −∞
Taking the limit µ → 0 we have
1 x
Z
y= g(ξ) sin(β(x − ξ))dξ.
β −∞

Solution 34.17
First we consider the Fourier transform of f (x) = 1/(x2 + c2 ) where <(c) > 0.
 
ˆ 1
f (ω) = F 2
x + c2
Z ∞
1 1
= e−iωx dx
2π −∞ x + c2
2
Z ∞
1 e−iωx
= dx
2π −∞ (x − ic)(x + ic)
If ω < 0 then we close the path of integration with a semi-circle in the upper half plane.
e−iωx
 
ˆ 1
f (ω) = 2πi Res , x = ic
2π (x − ic)(x + ic)
ecω
= , for ω < 0
2c

1475
Note that f (x) = 1/(x2 + c2 ) is an even function of x so that fˆ(ω) is an even function of ω. If fˆ(ω) = g(ω) for
ω < 0 then f (ω) = g(−|ω|) for all ω. Thus

 
1 1 −c|ω|
F 2 2
= e .
x +c 2c

Now we consider the integral equation

Z ∞
u(ξ) 1
2 2
dξ = 2 0 < a < b.
−∞ [(x − ξ) + a ] x + b2

We take the Fourier transform, utilizing the convolution theorem.

e−a|ω| e−b|ω|
2πû(ω) =
2a 2b
−(b−a)|ω|
ae
û(ω) =
2πb
a 1
u(x) = 2(b − a) 2
2πb x + (b − a)2
a(b − a)
u(x) =
πb(x2 + (b − a)2 )

1476
Solution 34.18
1. Note that Fc (ω) is an even function. The inverse Fourier cosine transform is
f (x) = Fc−1 [Fc (ω)]
Z ∞
= Fc (ω) eiωx dω
Z−∞
∞
= Fc (ω)(cos(ωx) + i sin(ωx)) dω
−∞
Z ∞
= Fc (ω) cos(ωx) dω
−∞
Z ∞
=2 Fc (ω) cos(ωx) dω.
0

2.
1 ∞ 00
Z
00
Fc [y ] = y cos(ωx) dx
π 0
∞ ω ∞ 0
Z
1 0
= y cos(ωx) 0 + y sin(ωx) dx
π π 0
∞ ω 2 ∞
Z
1 0 ω
= − y (0) + y sin(ωx) 0 − y cos(ωx) dx
π π π 0

y 0 (0)
Fc [y 00 ] = −ω 2 Fc (ω) −
π

3. We take the Fourier cosine transform of the differential equation.

b
−ω 2 ŷ(ω) − − a2 ŷ(ω) = 0
π
b
ŷ(ω) = −
π(ω + a2 )
2

1477
 Now we take the inverse Fourier cosine transform. We use the fact that ŷ(ω) is an even function.
 
−1 b
y(x) = Fc −
π(ω 2 + a2 )
 
−1 b
=F −
π(ω 2 + a2 )
 
b 1 iωx
= − i2π Res e , ω = ia
π ω 2 + a2
 iωx 
e
= −i2b lim , for x ≥ 0
ω→ia ω + ia

b
y(x) = − e−ax
a

Solution 34.19
1. Suppose f (x) is an odd function. The Fourier transform of f (x) is
Z ∞
1
F[f (x)] = f (x) e−iωx dx
2π −∞
Z ∞
1
= f (x)(cos(ωx) − i sin(ωx)) dx
2π −∞
i ∞
Z
=− f (x) sin(ωx) dx.
π 0
Note that F (ω) = F[f (x)] is an odd function of ω. The inverse Fourier transform of F (ω) is
Z ∞
−1
F [F (ω)] = F (ω) eiωx dω
−∞
Z ∞
= 2i F (ω) sin(ωx) dω.
0

1478
 Thus we have that
∞
i ∞
 Z Z 
f (x) = 2i − f (x) sin(ωx) dx sin(ωx) dω
0 π 0
Z ∞ Z ∞ 
1
=2 f (x) sin(ωx) dx sin(ωx) dω.
0 π 0
This gives us the Fourier sine transform pair
Z ∞ Z ∞
1
f (x) = 2 Fs (ω) sin(ωx) dω, Fs (ω) = f (x) sin(ωx) dx.
0 π 0

2.
1 ∞ 00
Z
00
Fs [y ] = y sin(ωx) dx
π 0
1h 0 i∞ ω Z ∞
= y sin(ωx) − y 0 cos(ωx) dx
π 0 π 0
ωh i∞ ω 2 Z ∞
= − y cos(ωx) − y sin(ωx) dx
π 0 π 0

ω
Fs [y 00 ] = −ω 2 Fs (ω) + y(0)
π

3. We take the Fourier sine transform of the differential equation.

bω
−ω 2 ŷ(ω) + − a2 ŷ(ω) = 0
π
bω
ŷ(ω) =
π(ω + a2 )
2

1479
 Now we take the inverse Fourier sine transform. We use the fact that ŷ(ω) is an odd function.
 
−1 bω
y(x) = Fs
π(ω 2 + a2 )
 
−1 bω
= −iF
π(ω 2 + a2 )
 
b ω iωx
= −i i2π Res e , ω = ia
π ω 2 + a2
 iωx 
ωe
= 2b lim
ω→ia ω + ia
−ax
= be for x ≥ 0

y(x) = b e−ax

4. Now we solve the differential equation with the Laplace transform.

y 00 − a2 y = 0
s2 ŷ(s) − sy(0) − y 0 (0) − a2 ŷ(s) = 0

We don’t know the value of y 0 (0), so we treat it as an unknown constant.

bs + y 0 (0)
ŷ(s) =
s 2 − a2
y 0 (0)
y(x) = b cosh(ax) + sinh(ax)
a
In order to satisfy the boundary condition at infinity we must choose y 0 (0) = −ab.

y(x) = b e−ax

1480
 We see that solving the differential equation with the Laplace transform is not as convenient, because
the boundary condition at infinity is not automatically satisfied. We had to find a value of y 0 (0) so that
y(∞) = 0.

Solution 34.20
The Fourier, Fourier cosine and Fourier sine transforms are defined:

Z ∞
1
F[f (x)] = f (x) e−iωx dx,
2π −∞
1 ∞
Z
F[f (x)]c = f (x) cos(ωx) dx,
π 0
1 ∞
Z
F[f (x)]s = f (x) sin(ωx) dx.
π 0

We start with the right side of the identity and apply the usual tricks of integral calculus to reduce the expression
to the left side.

1
(Fc [f (x) + f (−x)] − iFs [f (x) − f (−x)])
2

Z ∞ Z ∞ Z ∞ Z ∞ 
1
f (x) cos(ωx) dx + f (−x) cos(ωx) dx − i f (x) sin(ωx) dx + i f (−x) sin(ωx) dx
2π 0 0 0 0
Z ∞ Z −∞ Z ∞ Z −∞ 
1
f (x) cos(ωx) dx − f (x) cos(−ωx) dx − i f (x) sin(ωx) dx − i f (x) sin(−ωx) dx
2π 0 0 0 0
Z ∞ Z 0 Z ∞ Z 0 
1
f (x) cos(ωx) dx + f (x) cos(ωx) dx − i f (x) sin(ωx) dx − i f (x) sin(ωx) dx
2π 0 −∞ 0 −∞

1481
 Z ∞ Z ∞ 
1
f (x) cos(ωx) dx − i f (x) sin(ωx) dx
2π −∞ −∞
Z ∞
1
f (x) e−iωx dx
2π −∞
F[f (x)]

Solution 34.21
We take the Fourier transform of the integral equation, noting that the left side is the convolution of u(x) and
1
x2 +a2
.
   
1 1
2πû(ω)F 2 =F 2
x + a2 x + b2
1
We find the Fourier transform of f (x) = x2 +c ˆ
2 . Note that since f (x) is an even, real-valued function, f (ω) is

an even, real-valued function.

  Z ∞
1 1 1
F 2 2
= 2 2
e−iωx dx
x +c 2π −∞ x + c

For x > 0 we close the path of integration in the upper half plane and apply Jordan’s Lemma to evaluate the
integral in terms of the residues.

e−iωx
 
1
= i2π Res , x = ic
2π (x − ic)(x + ic)
e−iωic
=i
2ic
1 −cω
= e
2c

1482
Since fˆ(ω) is an even function, we have
 
1 1 −c|ω|
F 2 2
= e .
x +c 2c

Our equation for û(ω) becomes,

1 −a|ω| 1 −b|ω|
2πû(ω) e = e
2a 2b
a −(b−a)|omega|
û(ω) = e .
2πb
We take the inverse Fourier transform using the transform pair we derived above.

a 2(b − a)
u(x) =
2πb x2 + (b − a)2
a(b − a)
u(x) =
πb(x2 + (b − a)2 )

1483
Chapter 35

The Gamma Function

35.1 Euler’s Formula

For non-negative, integral n the factorial function is

n! = n(n − 1) · · · (1), with 0! = 1.

We would like to extend the factorial function so it is defined for all complex numbers.
Consider the function Γ(z) defined by Euler’s formula
Z ∞
Γ(z) = e−t tz−1 dt.
0

(Here we take the principal value of tz−1 .) The integral converges for <(z) > 0. If <(z) ≤ 0 then the integrand
will be at least as singular as 1/t at t = 0 and thus the integral will diverge.

1484
Difference Equation. Using integration by parts,
Z ∞
Γ(z + 1) = e−t tz dt
0
h i∞ Z ∞
−t z
= −e t − − e−t ztz−1 dt.
0 0

Since <(z) > 0 the first term vanishes.

Z ∞
=z e−t tz−1 dt
0
= zΓ(z)

Thus Γ(z) satisfies the difference equation

Γ(z + 1) = zΓ(z).

For general z it is not possible to express the integral in terms of elementary functions. However, we can
evaluate the integral for some z. The value z = 1 looks particularly simple to do.
Z ∞ h i∞
Γ(1) = e−t dt = − e−t = 1.
0 0

Using the difference equation we can find the value of Γ(n) for any positive, integral n.

Γ(1) = 1
Γ(2) = 1
Γ(3) = (2)(1) = 2
Γ(4) = (3)(2)(1) = 6
··· = ···
Γ(n + 1) = n!.

1485
 Thus the Gamma function, Γ(z), extends the factorial function to all complex z in the right half-plane. For
non-negative, integral n we have

Γ(n + 1) = n!.

Analyticity. The derivative of Γ(z) is

Z ∞
0
Γ (z) = e−t tz−1 log t dt.
0

Since this integral converges for <(z) > 0, Γ(z) is analytic in that domain.

35.2 Hankel’s Formula

We would like to find the analytic continuation of the Gamma function into the left half-plane. We accomplish
this with Hankel’s formula
Z
1
Γ(z) = et tz−1 dt.
2i sin(πz) C

Here C is the contour starting at −∞ below the real axis, enclosing the origin and returning to −∞ above the
real axis. A graph of this contour is shown in Figure 35.1. Again we use the principle value of tz−1 so there is a
branch cut on the negative real axis.
The integral in Hankel’s formula converges for all complex z. For non-positive, integral z the integral does
not vanish. Thus because of the sine term the Gamma function has simple poles at z = 0, −1, −2, . . . . For
positive, integral z, the integrand is entire and thus the integral vanishes. Using L’Hospital’s rule you can show
that the points, z = 1, 2, 3, . . . are removable singularities and the Gamma function is analytic at these points.
Since the only zeroes of sin(πz) occur for integral z, Γ(z) is analytic in the entire plane except for the points,
z = 0, −1, −2, . . . .

1486
 Figure 35.1: The Hankel Contour.

Difference Equation. Using integration by parts we can derive the difference equation from Hankel’s formula.
Z
1
Γ(z + 1) = et tz dt
2i sin(π(z + 1)) C
h i−∞+0i Z 
1 t z t z−1
= et − e zt dt
−2i sin(πz) −∞−0i C
Z
1
= z et tz−1 dt
2i sin(πz) C
= zΓ(z).

Evaluating Γ(1),
R
et tz−1 dt
C
Γ(1) = lim .
z→1 2i sin(πz)

1487
Both the numerator and denominator vanish. Using L’Hospital’s rule,
R t z−1
e t log t dt
= lim C
z→1 2πi cos(πz)
R t
e log t dt
= C
2πi
Let Cr be the circle of radius r starting at −π radians and going to π radians.
Z −r Z Z −∞ 
1 t t t
= e [log(−t) − πi] dt + e log t dt + e [log(−t) + πi] dt
2πi −∞ Cr −r
Z −∞ Z −∞ Z 
1 t t t
= e [− log(−t) + πi] dt + e [log(−t) + πi] dt + e log t dt
2πi −r −r Cr
Z −∞ Z 
1 t t
= e 2πi dt + e log t dt
2πi −r Cr

The integral on Cr vanishes as r → 0.

Z −∞
1
= 2πi et dt
2πi 0
= 1.
Thus we obtain the same value as with Euler’s formula. It can be shown that Hankel’s formula is the analytic
continuation of the Gamma function into the left half-plane.

35.3 Gauss’ Formula

Gauss defined the Gamma function as an infinite product. This form is useful in deriving some of its properties.
We can obtain the product form from Euler’s formula. First recall that
 n
−t t
e = lim 1 − .
n→∞ n

1488
Substituting this into Euler’s formula,
Z ∞
Γ(z) = e−t tz−1 dt
0
Z n n
t
= lim 1− tz−1 dt.
n→∞ 0 n

With the substitution τ = t/n,

Z 1
= lim (1 − τ )n nz−1 τ z−1 n dτ
n→∞ 0
Z 1
z
= lim n (1 − τ )n τ z−1 dτ.
n→∞ 0

Let n be an integer. Using integration by parts we can evaluate the integral.

1 1 Z 1
(1 − τ )n τ z τz
Z 
n z−1
(1 − τ ) τ dτ = − −n(1 − τ )n−1 dτ
0 z 0 0 z
Z 1
n
= (1 − τ )n−1 τ z dτ
z 0
n(n − 1) 1
Z
= (1 − τ )n−2 τ z+1 dτ
z(z + 1) 0
Z 1
n(n − 1) · · · (1)
= τ z+n−1 dτ
z(z + 1) · · · (z + n − 1) 0
 z+n 1
n(n − 1) · · · (1) τ
=
z(z + 1) · · · (z + n − 1) z + n 0
n!
=
z(z + 1) · · · (z + n)

1489
 Thus we have that
n!
Γ(z) = lim nz
n→∞ z(z + 1) · · · (z + n)
1 (1)(2) · · · (n)
= lim nz
z n→∞ (z + 1)(z + 2) · · · (z + n)
1 1
= lim nz
z n→∞ (1 + z)(1 + z/2) · · · (1 + z/n)
1 1 2z 3 z · · · n z
= lim
z n→∞ (1 + z)(1 + z/2) · · · (1 + z/n) 1z 2z · · · (n − 1)z

(n+1)z
Since limn→∞ nz
= 1 we can multiply by that factor.

1 1 2z 3z · · · (n + 1)z
= lim
z n→∞ (1 + z)(1 + z/2) · · · (1 + z/n) 1z 2z · · · nz
∞ 
(n + 1)z

1Y 1
=
z n=1 1 + z/n nz

Thus we have Gauss’ formula for the Gamma function

∞  z 
1Y 1  z −1
Γ(z) = 1+ 1+ .
z n=1 n n

We derived this formula from Euler’s formula which is valid only in the left half-plane. However, the product
formula is valid for all z except z = 0, −1, −2, . . . .

35.4 Weierstrass’ Formula

1490
The Euler-Mascheroni Constant. Before deriving Weierstrass’ product formula for the Gamma function we
will need to define the Euler-Mascheroni constant

  
1 1 1
γ = lim 1 + + + ··· + − log n = 0.5772 · · · .
n→∞ 2 3 n

In deriving the Euler product formula, we had the equation

 
z n!
Γ(z) = lim n .
n→∞ z(z + 1) · · · (z + n)
 
−1
 z −1  z −1  z −1 z
= lim z 1+ 1+ ··· 1 + n
n→∞ 1 2 n
1 h  z z  z  −z log n i
= lim z 1 + 1+ ··· 1 + e
Γ(z) n→∞ 1 2 n
    
z  −z  z  −z/2  z  −z/n 1 1
= lim z 1 + e 1+ e ··· 1 + e exp 1 + + · · · + − log n z
n→∞ 1 2 n 2 n

Weierstrass’ formula for the Gamma function is then

∞ h
1 Y z  −z/n i
= z eγz 1+ e .
Γ(z) n=1
n

Since the product is uniformly convergent, 1/Γ(z) is an entire function. Since 1/Γ(z) has no singularities, we
see that Γ(z) has no zeros.

1491
 Result 35.4.1 Euler’s formula for the Gamma function is valid for <(z) > 0.
Z ∞
Γ(z) = e−t tz−1 dt
0

Hankel’s formula defines the Γ(z) for the entire complex plane except for the points
z = 0, −1, −2, . . . .
Z
1
Γ(z) = et tz−1 dt
2i sin(πz) C
Gauss’ and Weierstrass’ product formulas are, respectively
∞  z 
1Y 1  z −1
Γ(z) = 1+ 1+ and
z n=1 n n

∞ h
1 γz
Y z  −z/n i
= ze 1+ e .
Γ(z) n=1
n

35.5 Stirling’s Approximation

In this section we will try to get an approximation to the Gamma function for large positive argument. Euler’s
formula is
Z ∞
Γ(x) = e−t tx−1 dt.
0

1492
We could first try to approximate the integral by only looking at the domain where the integrand is large. In
Figure 35.2 the integrand in the formula for Γ(10), e−t t9 , is plotted.

40000

30000

20000

10000

5 10 15 20 25 30

Figure 35.2: Plot of the integrand for Γ(10)

We see that the ”important” part of the integrand is the hump centered around x = 9. If we find where the
integrand of Γ(x) has its maximum

d
e−t tx−1 = 0


dx
− e−t tx−1 + (x − 1) e−t tx−2 = 0
(x − 1) − t = 0
t = x − 1,

we see that the maximum varies with x. This could complicate our analysis. To take care of this problem we

1493
introduce the change of variables t = xs.
Z ∞
Γ(x) = e−xs (xs)x−1 x ds
0
Z ∞
=x x
e−xs sx s−1 ds
Z0 ∞
= xx e−x(s−log s) s−1 ds
0

The integrands, ( e−x(s−log s) s−1 ), for Γ(5) and Γ(20) are plotted in Figure 35.3.

0.007 2·10
-9
0.006
-9
0.005 1.5·10
0.004 -9
1·10
0.003
0.002 -10
5·10
0.001
1 2 3 4 1 2 3 4

Figure 35.3: Plot of the integrand for Γ(5) and Γ(20).

We see that the important part of the integrand is the hump that seems to be centered about s = 1. Also
note that the the hump becomes narrower with increasing x. This makes sense as the e−x(s−log s) term is the
most rapidly varying term. Instead of integrating from zero to infinity, we could get a good approximation to the
integral by just integrating over some small neighborhood centered at s = 1. Since s − log s has a minimum at
s = 1, e−x(s−log s) has a maximum there. Because the important part of the integrand is the small area around

1494
s = 1, it makes sense to approximate s − log s with its Taylor series about that point.
1
s − log s = 1 + (s − 1)2 + O (s − 1)3
 
2
Since the hump becomes increasingly narrow with increasing x, we will approximate the 1/s term in the integrand
with its value at s = 1. Substituting these approximations into the integral, we obtain
Z 1+
2
Γ(x) ∼ x x
e−x(1+(s−1) /2) ds
1−
Z 1+
x −x 2
=x e e−x(s−1) /2 ds
1−

As x → ∞ both of the integrals

Z 1− Z ∞
−x(s−1)2 /2 2 /2
e ds and e−x(s−1) ds
−∞ 1+

are exponentially small. Thus instead of integrating from 1 −  to 1 +  we can integrate from −∞ to ∞.
Z ∞
x −x 2
Γ(x) ∼ x e e−x(s−1) /2 ds
Z−∞
∞
2
= xx e−x e−xs /2 ds
r−∞
2π
= xx e−x
x

√
Γ(x) ∼ 2πxx−1/2 e−x as x → ∞.

This is known as Stirling’s approximation to the Gamma function. In the table below, we see that the
approximation is pretty good even for relatively small argument.

1495
 √
n Γ(n) 2πxx−1/2 e−x relative error
5 24 23.6038 0.0165
15 8.71783 · 1010 8.66954 · 1010 0.0055
25 6.20448 · 1023 6.18384 · 1023 0.0033
35 2.95233 · 1038 2.94531 · 1038 0.0024
45 2.65827 · 1054 2.65335 · 1054 0.0019

In deriving Stirling’s approximation to the Gamma function we did a lot of hand waving. However, all of
the steps can be justified and better approximations can be obtained by using Laplace’s method for finding the
asymptotic behavior of integrals.

1496
35.6 Exercises
Exercise 35.1
Given that
Z ∞ √
2
e−x dx = π,
−∞

deduce the value of Γ(1/2). Now find the value of Γ(n + 1/2).

ExerciseR35.2
∞ 3
Evaluate 0 e−x dx in terms of the gamma function.

Exercise 35.3
Show that
Z ∞
Γ(i) + Γ(−i)
e−x sin(log x) dx = .
0 2

1497
35.7 Hints
Hint 35.1
Use the change of variables, ξ = x2 in the integral. To find the value of Γ(n + 1/2) use the difference relation.

Hint 35.2
Make the change of variable ξ = x3 .

1498
35.8 Solutions
Solution 35.1

Z ∞ √ 2
e−x dx =
π
−∞
Z ∞ √
−x2 π
e dx =
0 2

Make the change of variables ξ = x2 .

Z ∞ √
−ξ 1 −1/2 π
e ξ dξ =
0 2 2
√
Γ(1/2) = π

Recall the difference relation for the Gamma function Γ(z + 1) = zΓ(z).

Γ(n + 1/2) = (n − 1/2)Γ(n − 1/2)

2n − 1
= Γ(n − 1/2)
2
(2n − 3)(2n − 1)
= Γ(n − 3/2)
22
(1)(3)(5) · · · (2n − 1)
= Γ(1/2)
2n

(1)(3)(5) · · · (2n − 1) √
Γ(n + 1/2) = π
2n

1499
Solution 35.2
We make the change of variable ξ = x3 , x = ξ 1/3 , dx = 13 ξ −2/3 dξ.
Z ∞ Z ∞
−x3 1
e dx = e−ξ ξ −2/3 dξ
0 0 3
 
1 1
= Γ
3 3

Solution 35.3

Z ∞ Z ∞
−x 1 i log x
e−x − e−i log x dx


e sin(log x) dx = e
0 0 2i
1 ∞
Z
e−x xi − x−i dx


=
2i 0
1
= (Γ(1 + i) − Γ(1 − i))
2i
1
= (iΓ(i) − (−i)Γ(−i))
2i
Γ(i) + Γ(−i)
=
2

1500
Chapter 36

Bessel Functions

Ideas are angels. Implementations are a bitch.

36.1 Bessel’s Equation

A commonly encountered differential equation in applied mathematics is Bessel’s equation

ν2
 
001 0
y + y + 1 − 2 y = 0.
z z

For our purposes, we will consider ν ∈ R0+ . This equation arises when solving certain partial differential equations
with the method of separation of variables in cylindrical coordinates. For this reason, the solutions of this equation
are sometimes called cylindrical functions.
This equation cannot be solved directly. However, we can find series representations of the solutions. There is
a regular singular point at z = 0, so the Frobenius method is applicable there. The point at infinity is an irregular
singularity, so we will look for asymptotic series about that point. Additionally, we will use Laplace’s method to
find definite integral representations of the solutions.

1501
 Note that Bessel’s equation depends only on ν 2 and not ν alone. Thus if we find a solution, (which of course
depends on this parameter), yν (z) we know that y−ν (z) is also a solution. For this reason, we will consider
ν ∈ R0+ . Whether or not yν (z) and y−ν (z) are linearly independent, (distinct solutions), remains to be seen.

Example 36.1.1 Consider the differential equation

1 ν2
y 00 + y 0 + 2 y = 0
z z
One solution is yν (z) = z ν . Since the equation depends only on ν 2 , another solution is y−ν (z) = z −ν . For ν 6= 0,
these two solutions are linearly independent.
Now consider the differential equation
y 00 + ν 2 y = 0
One solution is yν (z) = cos(νz). Therefore, another solution is y−ν (z) = cos(−νz) = cos(νz). However, these two
solutions are not linearly independent.

36.2 Frobeneius Series Solution about z = 0

We note that z = 0 is a regular singular point, (the only singular point of Bessel’s equation in the finite complex
plane.) We will use the Frobenius method at that point to analyze the solutions. We assume that ν ≥ 0.
The indicial equation is
α(α − 1) + α − ν 2 = 0
α = ±ν.
If ±ν do not differ by an integer, (that is if ν is not a half-integer), then there will be two series solutions of the
Frobenius form.
∞
X X∞
ν k −ν
y1 (z) = z ak z , y2 (z) = z bk z k
k=0 k=0

1502
If ν is a half-integer, the second solution may or may not be in the Frobenius form. In any case, then will always
be at least one solution in the Frobenius form. We will determine that series solution. y(z) and it derivatives are
∞
X ∞
X ∞
X
y= ak z k+ν , y0 = (k + ν)ak z k+ν−1 , y 00 = (k + ν)(k + ν − 1)ak z k+ν−2 .
k=0 k=0 k=0

We substitute the Frobenius series into the differential equation.

z 2 y 00 + zy 0 + z 2 − ν 2 y = 0


∞
X X ∞ X∞ ∞
X
k+ν k+ν k+ν+2
(k + ν)(k + ν − 1)ak z + (k + ν)ak z + ak z − ν 2 ak z k+ν = 0
k=0 k=0 k=0 k=0
∞
X ∞
X
k 2 + 2kν ak z k + ak−2 z k = 0


k=0 k=2

We equate powers of z to obtain equations that determine the coefficients. The coefficient of z 0 is the equation
0 · a0 = 0. This corroborates that a0 is arbitrary, (but non-zero). The coefficient of z 1 is the equation

(1 + 2ν)a1 = 0
a1 = 0

The coefficient of z k for k ≥ 2 gives us

k 2 + 2kν ak + ak−2 = 0.


ak−2 ak−2
ak = − 2 =−
k + 2kν k(k + 2ν)

From the recurrence relation we see that all the odd coefficients are zero, a2k+1 = 0. The even coefficients are

a2k−2 (−1)k a0
a2k = − = 2k
4k(k + ν) 2 k!Γ(k + ν + 1)

1503
Thus we have the series solution
∞
X (−1)k
y(z) = a0 z 2k .
k=0
22k k!Γ(k + ν + 1)
−ν
a0 is arbitrary. We choose a0 = 2 . We call this solution the Bessel function of the first kind and order ν and
denote it with Jν (z).
∞
X (−1)k  z 2k+ν
Jν (z) =
k=0
k!Γ(k + ν + 1) 2

Recall that the Gamma function is non-zero and finite for all real arguments except non-positive integers.
Γ(x) has singularities at x = 0, −1, −2, . . . . Therefore, J−ν (z) is well-defined when ν is not a positive integer.
Since J−ν (z) ∼ z −ν at z = 0, J−ν (z) is clear linearly independent to Jν (z) for non-integer ν. In particular we note
that there are two solutions of the Frobenius form when ν is a half odd integer.
∞
X (−1)k  z 2k−ν
J−ν (z) = , for ν 6∈ Z+
k=0
k!Γ(k − ν + 1) 2

Of course for ν = 0, Jν (z) and J−ν (z) are identical. Consider the case that ν = n is a positive integer. Since
Γ(x) → +∞ as x → 0, −1, −2, . . . we see the the coefficients in the series for J−nu (z) vanish for k = 0, . . . , n − 1.
∞
X (−1)k  z 2k−n
J−n (z) =
k=n
k!Γ(k − n + 1) 2
∞
X (−1)k+n  z 2k+n
J−n (z) =
k=0
(k + n)!Γ(k + 1) 2
∞
n
X (−1)k  z 2k+n
J−n (z) = (−1)
k=0
k!(k + n)! 2
J−n (z) = (−1)n Jn (z)
Thus we see that J−n (z) and Jn (z) are not linearly independent for integer n.

1504
36.2.1 Behavior at Infinity
With the change of variables z = 1/t, w(z) = u(t) Bessel’s equation becomes
t4 u00 + 2t3 u0 + t(−t2 )u0 + (1 − ν 2 t2 )u = 0
ν2
 
00 1 0 1
u + u + 4 − 2 u = 0.
t t t
The point t = 0 and hence the point z = ∞ is an irregular singular point. We will find the leading order
asymptotic behavior of the solutions as z → +∞.

Controlling Factor. Starting with Bessel’s equation for real argument

ν2
 
00 1 0
y + y + 1 − 2 y = 0,
x x
we make the substitution y = es(x) to obtain
00 1 0
0 2 ν2
s + (s ) + s + 1 − 2 = 0.
x x
ν2
We know that x2
 1 as x → ∞; we will assume that s00  (s0 )2 as x → ∞. This gives us
1
(s0 )2 + s0 + 1 ∼ 0 as x → ∞.
x
To simplify the equation further, we will try the possible two-term balances.
1. (s0 )2 + x1 s0 ∼ 0 ⇒ s0 ∼ − x1 This balance is not consistent as it violates the assumption that 1 is
smaller than the other terms.
2. (s0 )2 + 1 ∼ 0 ⇒ s0 ∼ ±i This balance is consistent.
1 0
3. x
s +1∼0 ⇒ s0 ∼ −x This balance is inconsistent as (s0 )2 isn’t smaller than the other terms.
Thus the only dominant balance is s0 ∼ ±i. This balance is consistent with our initial assumption that
s00  (s0 )2 . Thus s ∼ ±ix and the controlling factor is e±ix .

1505
Leading Order Behavior. In order to find the leading order behavior, we substitute s = ±ix + t(x) where
t(x)  x as x → ∞ into the differential equation for s. We first consider the case s = ix + t(x). We assume that
t0  1 and t00  1/x.

00 0 2 1 0 ν2
t + (i + t ) + (i + t ) + 1 − 2 = 0
x x
2
i 1 ν
t00 + 2it0 + (t0 )2 + + t0 − 2 = 0
x x x

Using our assumptions about the behavior of t0 and t00 ,

i
2it0 + ∼0
x
1
t0 ∼ −
2x
1
t ∼ − log x as x → ∞.
2
This asymptotic behavior is consistent with our assumptions.
Substituting s = −ix + t(x) will also yield t ∼ − 21 log x. Thus the leading order behavior of the solutions is
1
y ∼ c e±ix− 2 log x+u(x) = cx−1/2 e±ix+u(x) as x → ∞,

where u(x)  log x as x → ∞.

By substituting t = − 12 log x + u(x) into the differential equation for t, you could show that u(x) → const as
x → ∞. Thus the full leading order behavior of the solutions is

y ∼ cx−1/2 e±ix+u(x) as x → ∞

where u(x) → 0 as x → ∞. Writing this in terms of sines and cosines yields

y1 ∼ x−1/2 cos(x + u1 (x)), y2 ∼ x−1/2 sin(x + u2 (x)), as x → ∞,

1506
where u1 , u2 → 0 as x → ∞.

Result 36.2.1 Bessel’s equation for real argument is

ν2
 
00 1 0
y + y + 1 − 2 y = 0.
x x
If ν is not an integer then the solutions behave as linear combinations of

y 1 = xν , and y2 = x−ν

at x = 0. If ν is an integer, then the solutions behave as linear combinations of

y1 = x ν , and y2 = x−ν + cxν log x

at x = 0. The solutions are asymptotic to a linear combination of

y1 = x−1/2 sin(x + u1 (x)), and y2 = x−1/2 cos(x + u2 (x))

as x → +∞, where u1 , u2 → 0 as x → ∞.

36.3 Bessel Functions of the First Kind

Consider the function exp( 12 z(t − 1/t)). We can expand this function in a Laurent series in powers of t,

∞
1
X
z(t−1/t)
e 2 = Jn (z)tn ,
n=−∞

1507
where the coefficient functions Jn (z) are
I
1 1
Jn (z) = τ −n−1 e 2 z(τ −1/τ ) dτ.
2πi
Here the path of integration is any positive closed path around the origin. exp( 21 z(t − 1/t)) is the generating
function for Bessel function of the first kind.

36.3.1 The Bessel Function Satisfies Bessel’s Equation

We would like to expand Jn (z) in powers of z. The first step in doing this is to make the substitution τ = 2t/z.
I  −n−1   
1 2t 1 2t z 2
Jn (z) = exp z − dt
2πi z 2 z 2t z
I
1  z n 2
= t−n−1 et−z /4t dt
2πi 2
Differentiating the expression for Jn (z),

1 nz n−1
 
−2z
I I
0 −n−1 t−z 2 /4t 1  z n −n−1 t−z 2 /4t
Jn (z) = t e dt + t e dt
2πi 2n 2πi 2 4t
I
1  z n  n z  −n−1 t−z2 /4t
= − t e dt
2πi 2 z 2t

I     
1  z n n n z n 1 z n z  −n−1 t−z2 /4t
Jn00 (z) = − + − 2− − − t e dt
2πi 2 z z 2t z 2t 2t z 2t
I  2
z 2 −n−1 t−z2 /4t

1  z n n nz n 1 nz
= − − − − + t e dt
2πi 2 z2 2zt z 2 2t 2zt 4t2
z 2 −n−1 t−z2 /4t
I  
1  z n n(n − 1) 2n + 1
= − + 2 t e dt.
2πi 2 z2 2t 4t

1508
 Substituting Jn (z) into Bessel’s equation,

n2
 
00 1 0
Jn + Jn + 1 − 2 Jn
z z
z2 n2
     
n(n − 1) 2n + 1
I
1 z n
  n 1 −n−1 t−z 2 /4t
= − + + − + 1 − t e dt
2πi 2 z2 2t 4t2 z 2 2t z2
z 2 −n−1 t−z2 /4t
I  
1  z n n+1
= 1− + 2 t e dt
2πi 2 t 4t
I
1  z n d  −n−1 t−z2 /4t 
= t e dt
2πi 2 dt
2 /4t
Since t−n−1 et−z is analytic in 0 < |t| < ∞ when n is an integer, the integral vanishes.

= 0.

Thus for integral n, Jn (z) satisfies Bessel’s equation.

Jn (z) is called the Bessel function of the first kind. The subscript is the order. Thus J1 (z) is a Bessel function
of order 1. J0 (x) and J1 (x) are plotted in the first graph in Figure 36.1. J5 (x) is plotted in the second graph in
Figure 36.1. Note that for non-negative, integral n, Jn (z) behaves as z n at z = 0.

36.3.2 Series Expansion of the Bessel Function

Expanding exp(−z 2 /4t) in the integral expression for Jn ,
I
1  z n 2
Jn (z) = t−n−1 et−z /4t dt
2πi 2
∞ 
!
2 m

−z
I
1  z n X 1
= t−n−1 et dt
2πi 2 m=0
4t m!

1509
 1

0.8 0.3

0.6 0.2

0.4 0.1

0.2
5 10 15 20
2 4 6 8 10 12 14 -0.1
-0.2
-0.2
-0.4

Figure 36.1: Plot of J0 (x), J1 (x) and J5 (x).

For the path of integration, we are free to choose any contour that encloses the origin. Consider the circular path
on |t| = 1. Since the integral is uniformly convergent, we can interchange the order of integration and summation.

∞
1  z n X (−1)m z 2m
I
Jn (z) = t−n−m−1 et dt
2πi 2 m=0 22m m!

1510
 If n is a non-negative integer,
dn+m z
I  
1 −n−m−1 t 1
t e dt = lim (e )
2πi z→0 (n + m)! dz n+m
1
= .
(n + m)!
Thus we have the series expansion
∞
X (−1)m  z n+2m
Jn (z) = for n ≥ 0.
m=0
m!(n + m)! 2

Now consider J−n (z), (n positive).

∞
1  z −n X (−1)m z 2m
I
J−n (z) = tn−m−1 et dt
2πi 2 m=1
22m m!
For m ≥ n, the integrand has a pole of order m − n + 1 at the origin.
(
1
for m ≥ n
I
1
tn−m−1 et dt = (m−n)!
2πi 0 for m < n
The expression for J−n is then
∞
X (−1)m  z −n+2m
J−n (z) =
m=n
m!(m − n)! 2
∞
X (−1)m+n  z n+2m
=
m=0
(m + n)!m! 2
= (−1)n Jn (z).
Thus we have that
J−n (z) = (−1)n Jn (z) for integral n.

1511
36.3.3 Bessel Functions of Non-Integral Order
The generalization of the factorial function is the Gamma function. For integral values of n, n! = Γ(n + 1). The
Gamma function is defined for all complex-valued arguments. Thus one would guess that if the Bessel function
of the first kind were defined for non-integral order, it would have the definition,
∞
X (−1)m  z ν+2m
Jν (z) = .
m=0
m!Γ(ν + m + 1) 2

The Integrand for Non-Integral ν. Recall the definition of the Bessel function
I
1  z ν 2
Jν (z) = t−ν−1 et−z /4t dt.
2πi 2
When ν is an integer, the integrand is single valued. Thus if you start at any point and follow any path around
the origin, the integrand will return to its original value. This property was the key to Jn satisfying Bessel’s
equation. If ν is not an integer, then this property does not hold for arbitrary paths around the origin.

A New Contour. First, since the integrand is multiple-valued, we need to define what branch of the function
we are talking about. We will take the principal value of the integrand and introduce a branch cut on the negative
real axis. Let C be a contour that starts at z = −∞ below the branch cut, circles the origin, and returns to the
point z = −∞ above the branch cut. This contour is shown in Figure 36.2.
Thus we define
I
1  z ν 2
Jν (z) = t−ν−1 et−z /4t dt.
2πi 2 C

Bessel’s Equation. Substituting Jν (z) into Bessel’s equation yields

ν2
  I
00 1 0 1  z ν d  −ν−1 t−z2 /4t 
Jν + Jν + 1 − 2 Jν = t e dt.
z z 2πi 2 C dt
2
Since t−ν−1 et−z /4t is analytic in 0 < |z| < ∞ and | arg(z)| < π, and it vanishes at z = −∞, the integral is zero.
Thus the Bessel function of the first kind satisfies Bessel’s equation for all complex orders.

1512
 Figure 36.2: The Contour of Integration.

Series Expansion. Because of the et factor in the integrand, the integral defining Jν converges uniformly.
2
Expanding e−z /4t in a Taylor series yields
∞
1  z ν X (−1)m z 2m
I
Jν (z) = t−ν−m−1 et dt
2πi 2 m=0 22m m! C

Since
I
1 1
= t−α−1 et dt,
Γ(α) 2πi C

we have the series expansion of the Bessel function

∞
X (−1)m  z ν+2m
Jν (z) = .
m=0
m!Γ(ν + m + 1) 2

1513
Linear Independence. The Wronskian of Bessel’s equation is
 Z z 
1 1
W (z) = exp − dζ = e− log z = .
ζ z

Thus to within a function of ν, the Wronskian is 1/z. For any given ν, there are two linearly independent
solutions. Note that Bessel’s equation is unchanged under the transformation ν → −ν. Thus both Jν and J−ν
satisfy Bessel’s equation. Now we must determine if they are linearly independent. We have already shown that
for integral values of ν they are not independent. (J−n = (−1)n Jn .) Assume that ν is not an integer. The
Wronskian of Jν and J−ν is
 
Jν J−ν 
W [Jν , J−ν ] =  0 0 

Jν J−ν
0
= Jν J−ν − J−ν Jν0

Substituting in the expansion for Jν ,

∞
!∞
!
X (−1)m  z ν+2m X (−1)n (−ν + 2n)  z −ν+2n−1
=
m=0
m!Γ(ν + m + 1) 2 n=0
n!Γ(−ν + n + 1)2 2
∞
! ∞
!
X (−1)m  z −ν+2m X (−1)n (ν + 2n)  z ν+2n−1
− .
m=0
m!Γ(−ν + m + 1) 2 n=0
n!Γ(ν + n + 1)2 2

Since the Wronskian is a function of ν times 1/z the coefficients of all of the powers of z except 1/z must vanish.

−ν ν
= −
zΓ(ν + 1)Γ(−ν + 1) zΓ(−ν + 1)Γ(ν + 1)
2
=−
zΓ(ν)Γ(1 − ν)

1514
Using an identity for the Gamma function simplifies this expression.

2
=− sin(πν)
πz
Since the Wronskian is nonzero for non-integral ν, Jν and J−ν are independent functions when ν is not an integer.
The general solution to the equation is then aJν + bJ−ν .

36.3.4 Recursion Formulas

In showing that Jν satisfies Bessel’s equation for arbitrary complex ν, we obtained
I
d  −ν t−z2 /4t 
t e dt = 0.
C dt

Expanding the integral,

z 2 −ν−2
I  
−ν −ν−1 2
t + t − νt et−z /4t dt = 0.
C 4
z 2 −ν−2
I  
1 z
 ν
−ν −ν−1 2
t + t − νt et−z /4t dt = 0.
2πi 2 C 4

2 /4t
1
t−ν−1 et−z
H
Since Jν (z) = 2πi
(z/2)ν C
dt,
"  #
−1   2
2 2 z
Jν−1 + Jν+1 − νJν = 0.
z z 4
2ν
Jν−1 + Jν+1 = Jν
z

1515
Differentiating the integral expression for Jν ,

1 νz ν−1
I I
−ν−1 t−z 2 /4t 1  z ν  z
t−z 2 /4t
Jν0 (z)= t e dt + t−ν−1
− e dt
2πi 2ν C 2πi 2 C 2t
I I
0 ν 1  z ν −ν−1 t−z 2 /4t 1  z ν+1 2
Jν (z) = t e dt − t−ν−2 et−z /4t dt
z 2πi 2 C 2πi 2 C
ν
Jν0 = Jν − Jν+1
z

From the two relations we have derived you can show that

1 ν
Jν0 = (Jν−1 + Jν+1 ) and Jν0 = Jν−1 − Jν .
2 z

1516
Result 36.3.1 The Bessel function of the first kind, Jν (z), is defined,
I
1  z ν 2
Jν (z) = t−ν−1 et−z /4t dt.
2πi 2 C

The Bessel function has the expansion,

∞
X (−1)m  z ν+2m
Jν (z) = .
m=0
m!Γ(ν + m + 1) 2

The Wronskian of Jν (z) and J−ν (z) is

2
W (z) = − sin(πν).
πz
Thus Jν (z) and J−ν (z) are independent when ν is not an integer. The Bessel functions
satisfy the recursion relations,
2ν ν
Jν−1 + Jν+1 = Jν Jν0 = Jν − Jν+1
z z
1 ν
Jν0 = (Jν−1 − Jν+1 ) Jν0 = Jν−1 − Jν .
2 z

1517
36.3.5 Bessel Functions of Half-Integral Order
Consider J1/2 (z). Start with the series expansion

∞
X (−1)m  z 1/2+2m
J1/2 (z) = .
m=0
m!Γ(1/2 + m + 1) 2

(1)(3)···(2n−1) √
Use the identity Γ(n + 1/2) = 2n
π.

∞
X (−1)m 2m+1  z 1/2+2m
= √
m=0
m!(1)(3) · · · (2m + 1) π 2
∞  1/2+m
X (−1)m 2m+1 1
= √ z 1/2+2m
m=0
(2)(4) · · · (2m) · (1)(3) · · · (2m + 1) π 2
 1/2 X ∞
2 (−1)m 2m+1
= z
πz m=0
(2m + 1)!

We recognize the sum as the Taylor series expansion of sin z.

 1/2
2
= sin z
πz

Using the recurrence relations,

ν ν
Jν+1 = Jν − Jν0 and Jν−1 = Jν + Jν0 ,
z z

we can find Jn+1/2 for any integral n.

1518
Example 36.3.1 To find J3/2 (z),
1/2 0
J3/2 (z) = J1/2 (z) − J1/2 (z)
z
 1/2    1/2  1/2
1/2 2 −1/2 1 2 −3/2 2
= z sin z − − z sin z − z −1/2 cos z
z π 2 π π
= 2−1/2 π −1/2 z −3/2 sin z + 2−1/2 π −1/2 z −3/2 sin z − 2−1/2 π −1/2 cos z
 1/2  1/2
2 −3/2 2
= z sin z − z −1/2 cos z
π π
 1/2
2
z −3/2 sin z − z −1/2 cos z .


=
π
You can show that
 1/2
2
J−1/2 (z) = cos z.
πz
Note that at a first glance it appears that J3/2 ∼ z −1/2 as z → 0. However, if you expand the sine and cosine
you will see that the z −1/2 and z 1/2 terms vanish and thus J3/2 (z) ∼ z 3/2 as z → 0 as we showed previously.
Recall that we showed the asymptotic behavior as x → +∞ of Bessel functions to be linear combinations of
x−1/2 sin(x + U1 (x)) and x−1/2 cos(x + U2 (x))
where U1 , U2 → 0 as x → +∞.

36.4 Neumann Expansions

Consider expanding an analytic function in a series of Bessel functions of the form
∞
X
f (z) = an Jn (z).
n=0

1519
If f (z) is analytic in the disk |z| ≤ r then we can write
I
1 f (ζ)
f (z) = dζ,
2πi ζ −z
1
where the path of integration is |ζ| = r and |z| < r. If we were able to expand the function ζ−z in a series of Bessel
functions, then we could interchange the order of summation and integration to get a Bessel series expansion of
f (z).

1
The Expansion of 1/(ζ − z). Assume that ζ−z
has the uniformly convergent expansion
∞
1 X
= c0 (ζ)J0 (z) + 2 cn (ζ)Jn (z),
ζ −z n=1

where each cn (ζ) is analytic. Note that

 
∂ ∂ 1 −1 1
+ = + = 0.
∂ζ ∂z ζ − z (ζ − z)2 (ζ − z)2
Thus we have
 " ∞
#
∂ ∂ X
+ c0 (ζ)J0 (z) + 2 cn (ζ)Jn (z) = 0
∂ζ ∂z n=1
" ∞
# " ∞
#
X X
c00 J0 + 2 c0n Jn + c0 J00 + 2 cn Jn0 = 0.
n=1 n=1

Using the identity 2Jn0 = Jn−1 − Jn+1 ,

" ∞
# " ∞
#
X X
c00 J0 + 2 c0n Jn + c0 (−J1 ) + cn (Jn−1 − Jn+1 ) = 0.
n=1 n=1

1520
Collecting coefficients of Jn ,
∞
X
(c00 + c1 )J0 + (2c0n + cn+1 − cn−1 )Jn = 0.
n=1

Equating the coefficients of Jn , we see that the cn are given by the relations,

c1 = −c00 , and cn+1 = cn−1 − 2c0n .

We can evaluate c0 (ζ). Setting z = 0,

∞
1 X
= c0 (ζ)J0 (0) + 2 cn (ζ)Jn (0)
ζ n=1
1
= c0 (ζ).
ζ

Using the recurrence relations we can calculate the cn ’s. The first few are:
−1 1
c1 = − 2
= 2
ζ ζ
1 −2 1 4
c2 = − 2 3 = + 3
ζ ζ ζ ζ
 
1 −1 12 3 24
c3 = 2 − 2 2
− 4 = 2 + 4.
ζ ζ ζ ζ ζ

We see that cn is a polynomial of degree n + 1 in 1/ζ. One can show that

  
 2n−1 n! ζ2 ζ4 ζn
ζ n+1
1 + 2(2n−2) + 2·4·(2n−2)(2n−4) + · · · + 2·4···n·(2n−2)···(2n−n) for even n
cn (ζ) =  
 2 n+1n! 1 + ζ 2 +
n−1 ζ4
+ · · · + ζ n−1
for odd n
ζ 2(2n−2) 2·4·(2n−2)(2n−4) 2·4···(n−1)·(2n−2)···(2n−(n−1))

1521
 1
Uniform Convergence of the Series. We assumed before that the series expansion of ζ−z
is uniformly
convergent. The behavior of cn and Jn are

2n−1 n! zn
cn (ζ) = + O(ζ −n ), Jn (z) = + O(z n+1 ).
ζ n+1 2n n!

This gives us
 n  n+1 !
1 z 1 z
cn (ζ)Jn (z) = +O .
2ζ ζ ζ ζ
 
If  zζ  = ρ < 1 we can bound the series with the geometric series
P n
ρ . Thus the series is uniformly convergent.
 

Neumann Expansion of an Analytic Function. Let f (z) be a function that is analytic in the disk |z| ≤ r.
Consider |z| < r and the path of integration along |ζ| = r. Cauchy’s integral formula tells us that
I
1 f (ζ)
f (z) = dζ.
2πi ζ −z

1
Substituting the expansion for ζ−z
,

∞
I !
1 X
= f (ζ) co (ζ)J0 (z) + 2 cn (ζ)Jn (z) dζ
2πi n=1
I ∞ I
1 f (ζ) X Jn (z)
= J0 (z) dζ + cn (ζ)f (ζ) dζ
2πi ζ n=1
πi
∞ I
X Jn (z)
= J0 (z)f (0) + cn (ζ)f (ζ) dζ.
n=1
πi

1522
 Result 36.4.1 let f (z) be analytic in the disk, |z| ≤ r. Consider |z| < r and the path
of integration along |ζ| = r. f (z) has the Bessel function series expansion
∞ I
X Jn (z)
f (z) = J0 (z)f (0) + cn (ζ)f (ζ) dζ,
n=1
πi

where the cn satisfy

∞
1 X
= c0 (ζ)J0 (z) + 2 cn (ζ)Jn (z).
ζ −z n=1

36.5 Bessel Functions of the Second Kind

When ν is an integer, Jν and J−ν are not linearly independent. In order to find an second linearly independent
solution, we define the Bessel function of the second kind, (also called Weber’s function),

(J
ν (z) cos(νπ)−J−ν (z)
sin(νπ)
when ν is not an integer
Yν =
limµ→ν Jµ (z) cos(µπ)−J
sin(µπ)
−µ (z)
when ν is an integer.

Jν and Yν are linearly independent for all ν.

In Figure 36.3 Y0 and Y1 are plotted in solid and dashed lines, respectively.

1523
 0.5

0.25

5 10 15 20
-0.25

-0.5

-0.75

-1

Figure 36.3: Bessel Functions of the Second Kind

Result 36.5.1 The Bessel function of the second kind, Yν (z), is defined,
( J (z) cos(νπ)−J (z)
ν −ν
sin(νπ) when ν is not an integer
Yν = J (z) cos(µπ)−J−µ (z)
limµ→ν µ sin(µπ) when ν is an integer.

The Wronskian of Jν (z) and Yν (z) is

2
W [Jν , Yν ] = .
πz
Thus Jν (z) and Yν (z) are independent for all ν. The Bessel functions of the second kind
satisfy the recursion relations,
1524
2ν ν
Yν−1 + Yν+1 = Yν Yν0 = Yν − Yν+1
z z
1 ν
Yν0 = (Yν−1 − Yν+1 ) Yν0 = Yν−1 − Yν .
2 z
36.6 Hankel Functions
Another set of solutions to Bessel’s equation is the Hankel functions,

Hν(1) (z) = Jν (z) + iYν (z),

Hν(2) (z) = Jν (z) − iYν (z)

Result 36.6.1 The Hankel functions are defined

Hν(1) (z) = Jν (z) + iYν (z),

Hν(2) (z) = Jν (z) − iYν (z)
(1) (2)
The Wronskian of Hν (z) and Hν (z) is
4i
W [Hν(1) , Hν(2) ] = − .
πz
The Hankel functions are independent for all ν. The Hankel functions satisfy the same
recurrence relations as the other Bessel functions.

36.7 The Modified Bessel Equation

The modified Bessel equation is

ν2
 
1 0
00
w + w − 1 + 2 w = 0.
z z

1525
This equation is identical to the Bessel equation except for a sign change in the last term. If we make the change
of variables ξ = iz, u(ξ) = w(z) we obtain the equation
ν2
 
00 1 0
−u − u − 1 − 2 u = 0
ξ ξ
2
 
00 1 0 ν
u + u + 1 − 2 u = 0.
ξ ξ
This is the Bessel equation. Thus Jν (iz) is a solution to the modified Bessel equation. This motivates us to define
the modified Bessel function of the first kind
Iν (z) = i−ν Jν (iz).
Since Jν and J−ν are linearly independent solutions when ν is not an integer, Iν and I−ν are linearly independent
solutions to the modified Bessel equation when ν is not an integer.
The Taylor series expansion of Iν (z) about z = 0 is
Iν (z) = i−ν Jν (iz)
∞  ν+2m
−ν
X (−1)m iz
=i
m=0
m!Γ(ν + m + 1) 2
∞
X (−1)m iν i2m  z ν+2m
= i−ν
m=0
m!Γ(ν + m + 1) 2
∞
X 1  z ν+2m
=
m=0
m!Γ(ν + m + 1) 2

Modified Bessel Functions of the Second Kind. In order to have a second linearly independent solution
when ν is an integer, we define the modified Bessel function of the second kind
( I −I
π −ν ν
when ν is not an integer,
Kν (z) = 2 sin(νπ) π I−µ −Iµ
limµ→ν 2 sin(µπ) when ν is an integer.

1526
Iν and Kν are linearly independent for all ν. In Figure 36.4 I0 and K0 are plotted in solid and dashed lines,
respectively.

10

1 2 3 4

Figure 36.4: Modified Bessel Functions

1527
Result 36.7.1 The modified Bessel functions of the first and second kind, Iν (z) and
Kν (z), are defined,

Iν (z) = i−ν Jν (iz).

π I−ν −Iν
(
2 sin(νπ) when ν is not an integer,
Kν (z) = π I−µ −Iµ
limµ→ν 2 sin(µπ) when ν is an integer.

The modified Bessel function of the first kind has the expansion,
∞
X 1  z ν+2m
Iν (z) =
m=0
m!Γ(ν + m + 1) 2

The Wronskian of Iν (z) and I−ν (z) is

2
W [Iν , I−ν ] = − sin(πν).
πz
Iν (z) and I−ν (z) are linearly independent when ν is not an integer. The Wronskian of
Iν (z) and Kν (z) is
1
W [Iν , Kν ] = − .
z
Iν (z) and Kν (z) are independent for all ν. The modified Bessel functions satisfy the
recursion relations,
2ν ν
Aν−1 − Aν+1 = A0ν = Aν+1 + Aν
Aν
z z
1 1528 ν
A0ν = (Aν−1 + Aν+1 ) A0ν = Aν−1 − Aν .
2 z
where A stands for either I or K.
36.8 Exercises
Exercise 36.1
Consider Bessel’s equation
z 2 y 00 (z) + zy 0 (z) + z 2 − ν 2 y = 0

where ν ≥ 0. Find the Frobenius series solution that is asymptotic to tν as t → 0. By multiplying this solution
by a constant, define the solution
∞
X (−1)k  z 2k+ν
Jν (z) = .
k=1
k!Γ(k + ν + 1) 2

This is called the Bessel function of the first kind and order ν. Clearly J−ν (z) is defined and is linearly independent
to Jν (z) if ν is not an integer. What happens when ν is an integer?

Exercise 36.2
Consider Bessel’s equation for integer n,
z 2 y 00 + zy 0 + (z 2 − n2 )y = 0.
Using the kernel
1 1
K(z, t) = e 2 z(t− t ) ,
find two solutions of Bessel’s equation. (For n = 0 you will find only one solution.) Are the two solutions linearly
independent? Define the Bessel function of the first kind and order n,
I
1 1
Jn (z) = t−n−1 e 2 z(t−1/t) dt,
i2π C
where C is a simple, closed contour about the origin. Verify that
∞
1
X
z(t−1/t)
e 2 = Jn (z)tn .
n=−∞

1529
This is the generating function for the Bessel functions.

Exercise 36.3
Use the generating function
∞
1
X
z(t−1/t)
e 2 = Jn (z)tn
n=−∞

to show that Jn satisfies Bessel’s equation

z 2 y 00 + zy 0 + (z 2 − n2 )y = 0.

Exercise 36.4
Using
2n n
Jn−1 + Jn+1 = Jn and Jn0 = Jn − Jn+1 ,
z z
show that
1 n
Jn0 = (Jn−1 − Jn+1 ) and Jn0 = Jn−1 − Jn .
2 z

Exercise 36.5
Find the general solution of
 
1 0
00 1
w + w + 1 − 2 w = z.
z 4z

Exercise 36.6
Show that Jν (z) and Yν (z) are linearly independent for all ν.

1530
Exercise 36.7
Compute W [Iν , I−ν ] and W [Iν , Kν ].

Exercise 36.8
Using the generating function,
   +∞
z 1 X
exp t− = Jn (z)tn ,
2 t n=−∞

verify the following identities:

1.
2n
Jn (z) = Jn−1 (z) + Jn+1 (z).
z
This relation is useful for recursively computing the values of the higher order Bessel functions.

2.
1
Jn0 (z) = (Jn−1 − Jn+1 ) .
2
This relation is useful for computing the derivatives of the Bessel functions once you have the values of the
Bessel functions of adjacent order.

3.
d −n
z Jn (z) = −z −n Jn+1 (z).


dz

1531
Exercise 36.9
Use the Wronskian of Jν (z) and J−ν (z),
2 sin νπ
W [Jν (z), J−ν (z)] = − ,
πz
to derive the identity
2
J−ν+1 (z)Jν (z) + J−ν (z)Jν−1 (z) = sin νπ.
πz

Exercise 36.10
Show that, using the generating function or otherwise,
J0 (z) + 2J2 (z) + 2J4 (z) + 2J6 (z) + · · · = 1
J0 (z) − 2J2 (z) + 2J4 (z) − 2J6 (z) + · · · = cos z
2J1 (z) − 2J3 (z) + 2J5 (z) − · · · = sin z
J0 (z) + 2J12 (z) + 2J22 (z) + 2J32 (z) + · · · = 1
2

Exercise 36.11
It is often possible to “solve” certain ordinary differential equations by converting them into the Bessel equation
by means of various transformations. For example, show that the solution of
y 00 + xp−2 y = 0,
can be written in terms of Bessel functions.
   
1/2 2 p/2 1/2 2 p/2
y(x) = c1 x J1/p x + c2 x Y1/p x
p p
Here c1 and c2 are arbitrary constants. Thus show that the Airy equation,
y 00 + xy = 0,
can be solved in terms of Bessel functions.

1532
Exercise 36.12
The spherical Bessel functions are defined by
r
π
jn (z) = Jn+1/2 (z),
2z
r
π
yn (z) = Yn+1/2 (z),
2z
r
π
kn (z) = Kn+1/2 (z),
2z
r
π
in (z) = In+1/2 (z).
2z
Show that
sin z cos z
j1 (z) = − ,
z2 z
sinh z
i0 (z) = ,
z
π
k0 (z) = exp(−z).
2z

Exercise 36.13
Show that as x → ∞,

e−x 4n2 − 1 (4n2 − 1)(4n2 − 9)

 
Kn (x) ∝ √ 1+ + + ··· .
x 8x 128x2

1533
36.9 Hints
Hint 36.2

Hint 36.3

Hint 36.4
Use the generating function
∞
1
X
z(t−1/t)
e 2 = Jn (z)tn
n=−∞

to show that Jn satisfies Bessel’s equation

z 2 y 00 + zy 0 + (z 2 − n2 )y = 0.

Hint 36.6
Use variation of parameters and the Wronskian that was derived in the text.

Hint 36.7
Compute the Wronskian of Jν (z) and Yν (z). Use the relation
2
W [Jν , J−ν ] = − sin(πν)
πz

Hint 36.8
Derive W [Iν , I−ν ] from the value of W [Jν , J−ν ]. Derive W [Iν , Kν ] from the value of W [Iν , I−ν ].

1534
Hint 36.9

Hint 36.10

Hint 36.11

Hint 36.12

Hint 36.13

Hint 36.14

1535
36.10 Solutions
Solution 36.1
Bessel’s equation is
L[y] ≡ z 2 y 00 + zy 0 + (z 2 − n2 )y = 0.
We consider a solution of the form
Z
1
y(z) = e 2 z(t−1/t) v(t) dt.
C

We substitute the form of the solution into Bessel’s equation.

Z h 1 i
z(t−1/t)
L e 2 v(t) dt = 0
C
Z  2  2 !
21 1 1 1 2 2 1
e 2 z(t−1/t) v(t) dt = 0


z t+ +z t− + z −n (36.1)
C 4 t 2 t

By considering
   
d 1 z(t−1/t) 1 1 1
t e2 = x t+ + 1 e 2 z(t−1/t)
dt 2 t
2 !
d2 2 1 z(t−1/t)
  
1 2 1 1 1
t e2 = x t+ + x 2t + + 2 e 2 z(t−1/t)
dt2 4 t t

we see that
d2 2
 
h 1
z(t−1/t)
i d 1
L e 2 = 2
t − 3 t + (1 − n ) e 2 z(t−1/t) .
2
dt dt
Thus Equation 36.1 becomes
Z  2 
d 2 1 z(t−1/t) d 1 z(t−1/t) 2 1
z(t−1/t)
t e2 − 3 t e2 + (1 − n ) e 2 v(t) dt = 0
C dt2 dt

1536
We apply integration by parts to move derivatives from the kernel to v(t).
h 1 i h 1 i h i Z
1 1
0
2 2 z(t−1/t) z(t−1/t) z(t−1/t)
e 2 z(t−1/t) t2 v 00 (t) + 3tv(t) + (1 − n2 )v(t) dt = 0


t e v(t) − t e 2 v (t) + −3t e 2 v(t) +
C C C C

In order that the integral vanish, v(t) must be a solution of the differential equation
t2 v 00 + 3tv + (1 − n2 )v = 0.
This is an Euler equation with the solutions {tn−1 , t−n−1 } for non-zero n and {t−1 , t−1 log t} for n = 0.
Consider the case of non-zero n. Since
1
e 2 z(t−1/t) (t2 − 3t)v(t) − tv 0 (t)

is single-valued and analytic for t 6= 0 for the functions v(t) = tn−1 and v(t) = t−n−1 , the boundary term will
vanish if C is any closed contour that that does not pass through the origin. Note that the integrand in our
solution,
1
e 2 z(t−1/t) v(t),
is analytic and single-valued except at the origin and infinity where it has essential singularities. Consider a
simple closed contour that does not enclose the origin. The integral along such a path would vanish and give us
y(z) = 0. This is not an interesting solution. Since
1
e 2 z(t−1/t) v(t),
has non-zero residues for v(t) = tn−1 and v(t) = t−n−1 , choosing any simple, positive, closed contour about the
origin will give us a non-trivial solution of Bessel’s equation. These solutions are
Z Z
n−1 21 z(t−1/t) 1
y1 (t) = t e dt, y2 (t) = t−n−1 e 2 z(t−1/t) dt.
C C

Now consider the case n = 0. The two solutions above concide and we have the solution
Z
1
y(t) = t−1 e 2 z(t−1/t) dt.
C

1537
Choosing v(t) = t−1 log t would make both the boundary terms and the integrand multi-valued. We do not pursue
the possibility of a solution of this form.
The solution y1 (t) and y2 (t) are not linearly independent. To demonstrate this we make the change of variables
t → −1/t in the integral representation of y1 (t).
Z
1
y1 (t) = tn−1 e 2 z(t−1/t) dt
ZC
1 −1
= (−1/t)n−1 e 2 z(−1/t+t) 2 dt
t
ZC
1
= (−1)n t−n−1 e 2 z(t−1/t) dt
C
= (−1)n y2 (t)

Thus we see that a solution of Bessel’s equation for integer n is

Z
1
y(t) = t−n−1 e 2 z(t−1/t) dt
C

where C is any simple, closed contour about the origin.

Therefore, the Bessel function of the first kind and order n,
I
1 1
Jn (z) = t−n−1 e 2 z(t−1/t) dt
i2π C

is a solution of Bessel’s equation for integer n. Note that Jn (z) is the coefficient of tn in the Laurent series of
1
e 2 z(t−1/t) . This establishes the generating function for the Bessel functions.
∞
1
X
z(t−1/t)
e 2 = Jn (z)tn
n=−∞

1538
Solution 36.2
The generating function is
∞
z
X
(t−1/t)
e 2 = Jn (z)tn .
n=−∞

In order to show that Jn satisfies Bessel’s equation we seek to show that

∞
X
z 2 Jn00 (z) + zJn (z) + (z 2 − n2 )Jn (z) tn = 0.


n=−∞

To get the appropriate terms in the sum we will differentiate the generating function with respect to z and t.
First we differentiate it with respect to z.
  ∞
1 1 z
X
t− e 2
(t−1/t)
= Jn0 (z)tn
2 t n=−∞
 2 ∞
1 1 z
X
t− e 2
(t−1/t)
= Jn00 (z)tn
4 t n=−∞

Now we differentiate with respect to t and multiply by t get the n2 Jn term.

  ∞
z 1 z
(t−1/t)
X
1+ 2 e 2 = nJn (z)tn−1
2 t n=−∞
  ∞
z 1 z
X
t+ e 2 (t−1/t) = nJn (z)tn
2 t n=−∞
  2
 2 ∞
z 1 z
(t−1/t) z 1 z
(t−1/t)
X
1 − 2 e2 + t+ e2 = n2 Jn (z)tn−1
2 t 4 t n=−∞
  2
 2 ∞
z 1 z
(t−1/t) z 1 z
(t−1/t)
X
t− e2 + t+ e2 = n2 Jn (z)tn
2 t 4 t n=−∞

1539
Now we can evaluate the desired sum.
∞
X
z 2 Jn00 (z) + zJn (z) + (z 2 − n2 )Jn (z) tn


n=−∞
2 2 !
z2 2
     
1 z 1 z 1 z 1 z
= t− + t− + z2 − t− − t+ e 2 (t−1/t)
4 t 2 t 2 t 4 t

∞
X
z 2 Jn00 (z) + zJn (z) + (z 2 − n2 )Jn (z) tn = 0


n=−∞

z 2 Jn00 (z) + zJn (z) + (z 2 − n2 )Jn (z) = 0

Thus Jn satisfies Bessel’s equation.

Solution 36.3

n
Jn0 = Jn − Jn+1
z
1
= (Jn−1 + Jn+1 ) − Jn+1
2
1
= (Jn−1 − Jn+1 )
2
n
Jn0 = Jn − Jn+1
z  
n 2n
= Jn − Jn − Jn−1
z z
n
= Jn−1 − Jn
z

1540
Solution 36.4
The linearly independent homogeneous solutions are J1/2 and J−1/2 . The Wronskian is

2 2
W [J1/2 , J−1/2 ] = − sin(π/2) = − .
πz πz

Using variation of parameters, a particular solution is

z Z z
ζJ−1/2 (ζ) ζJ1/2 (ζ)
Z
yp = −J1/2 (z) dζ + J−1/2 (z) dζ
−2/πζ −2/πζ
Z z Z z
π 2 π
= J1/2 (z) ζ J−1/2 (ζ) dζ − J−1/2 (z) ζ 2 J1/2 (ζ) dζ.
2 2

Thus the general solution is

Z z Z z
π 2 π
y = c1 J1/2 (z) + c2 J−1/2 (z) + J1/2 (z) ζ J−1/2 (ζ) dζ − J−1/2 (z) ζ 2 J1/2 (ζ) dζ.
2 2

We could substitute

 1/2  1/2
2 2
J1/2 (z) = sin z and J−1/2 = cos z
πz πz

into the solution, but we cannot evaluate the integrals in terms of elementary functions. (You can write the
solution in terms of Fresnel integrals.)

1541
Solution 36.5

 
Jν Jν cot(νπ) − J−ν csc(νπ)
W [Jν , Yν ] =  0 
Jν Jν0 cot(νπ) − J−ν 0
csc(νπ)
   
Jν Jν  Jν J−ν 
= cot(νπ)  0  − csc(νπ)  0 
Jν Jν0  0 
Jν J−ν
−2
= − csc(νπ) sin(πν)
πz
2
=
πz
Since the Wronskian does not vanish identically, the functions are independent for all values of ν.

Solution 36.6

Iν (z) = i−ν Jν (iz)

 
Iν I−ν 
W [Iν , I−ν ] =  0 
0 
Iν I−ν
 −ν 
 i Jν (iz) iν J−ν (iz) 
=  −ν 0
 0

i iJν (iz) iν iJ−ν (iz)
 
Jν (iz) J−ν (iz)
= i 0 0

Jν (iz) J−ν (iz)
−2
=i sin(πν)
πiz
2
= − sin(πν)
πz

1542
  
Iν π csc(πν)(I−ν − Iν )
W [Iν , Kν ] =  0 π2 0

Iν 2 csc(πν)(I−ν − Iν0 )
   
π Iν I−ν  Iν Iν 
= csc(πν)  0 0 − 0
  
2 Iν I−ν Iν Iν0 
π −2
= csc(πν) sin(πν)
2 πz
1
=−
z

Solution 36.7
1. We diferentiate the generating function with respect to t.
∞
z
X
(t−1/t)
e 2 = Jn (z)tn
n=−∞
  ∞
z 1 z
(t−1/t)
X
1+ 2 e 2 = Jn (z)ntn−1
2 t n=−∞
  X ∞ ∞
1 n 2 X
1+ 2 Jn (z)t = Jn (z)ntn−1
t n=−∞ z n=−∞
∞ ∞ ∞
X
n
X
n−2 2 X
Jn (z)t + Jn (z)t = Jn (z)ntn−1
n=−∞ n=−∞
z n=−∞
∞ ∞ ∞
X X 2 X
Jn−1 (z)tn−1 + Jn+1 (z)tn−1 = Jn (z)ntn−1
n=−∞ n=−∞
z n=−∞
2
Jn−1 (z) + Jn+1 (z) = Jn (z)n
z
2n
Jn (z) = Jn−1 (z) + Jn+1 (z)
z

1543
2. We diferentiate the generating function with respect to z.
∞
z
X
e 2 (t−1/t) = Jn (z)tn
n=−∞
  ∞
1 1 z
X
t− e2 (t−1/t)
= Jn0 (z)tn
2 t n=−∞
  X ∞ ∞
1 1 X
t− n
Jn (z)t = Jn0 (z)tn
2 t n=−∞ n=−∞
∞ ∞
! ∞
1 X X X
Jn (z)t n+1
− Jn (z)tn−1
= Jn0 (z)tn
2 n=−∞ n=−∞ n=−∞
∞ ∞
! ∞
1 X X X
Jn−1 (z)tn − Jn+1 (z)tn = Jn0 (z)tn
2 n=−∞ n=−∞ n=−∞
1
(Jn−1 (z) − Jn+1 (z)) = Jn0 (z)
2
1
Jn0 (z) = (Jn−1 − Jn+1 )
2
3.
d −n
z Jn (z) = −nz −n−1 Jn (z) + z −n Jn0 (z)


dz
1 2n 1
= − z −n Jn (z) + z −n (Jn−1 (z) − Jn+1 (z))
2 z 2
1 −n 1
= − z (Jn+1 (z) + Jn−1 (z)) + z −n (Jn−1 (z) − Jn+1 (z))
2 2

d −n
z Jn (z) = −z −n Jn+1 (z)


dz

1544
Solution 36.8
For this part we will use the identities
ν ν
Jν0 (z) = Jν (z) − Jν+1 (z), Jν0 (z) = Jν−1 (z) − Jν (z).
z z
 
 = − 2 sin(νπ)
Jν (z) J−ν (z)
 0 0
Jν (z) J−ν (z) πz
 
 J ν (z) J −ν (z)  2 sin(νπ)
Jν−1 (z) − ν Jν − ν J−ν (z) − J−ν+1 (z) = − πz
 
z z
   
 Jν (z) J −ν (z)  ν Jν (z) J−ν (z) 2 sin(νπ)
Jν−1 (z) −J−ν+1 (z) − z Jν (z) J−ν (z) = − πz
   

2 sin(νπ)
−Jν+1 (z)Jν (z) − Jν (z)Jν−1 (z) = −
πz
2
J−ν+1 (z)Jν (z) + J−ν (z)Jν−1 (z) = sin νπ
πz

Solution 36.9
The generating function for the Bessel functions is
∞
1
X
z(t−1/t)
e 2 = Jn (z)tn . (36.2)
n=−∞

1. We substitute t = 1 into (36.2).

∞
X
Jn (z) = 1
n=−∞
∞
X ∞
X
J0 (z) + Jn (z) + J−n (z) = 1
n=1 n=1

1545
 We use the identity J−n = (−1)n Jn .
∞
X
J0 (z) + (1 + (−1)n )Jn (z) = 1
n=1
∞
X
J0 (z) + 2 Jn (z) = 1
n=2
even n

∞
X
J0 (z) + 2 J2n (z) = 1
n=1

2. We substitute t = i into (36.2).

∞
X
Jn (z)in = eiz
n=−∞
∞
X ∞
X
J0 (z) + Jn (z)i + n
J−n (z)i−n = eiz
n=1 n=1
∞
X ∞
X
J0 (z) + Jn (z)in + (−1)n Jn (z)(−i)n = eiz
n=1 n=1

∞
X
J0 (z) + 2 Jn (z)in = eiz (36.3)
n=1

Substituting t = −i into (36.2) yields

∞
X
J0 (z) + 2 (−1)n Jn (z)in = e−iz (36.4)
n=1

1546
 Dividing the sum of (36.3) and (36.4) by 2 gives us the desired identity.

∞
X
J0 (z) + (1 + (−1)n )Jn (z)in = cos z
n=1
∞
X
J0 (z) + 2 Jn (z)in = cos z
n=2
even n
∞
X
J0 (z) + 2 (−1)n/2 Jn (z) = cos z
n=2
even n
∞
X
J0 (z) + 2 (−1)n J2n (z) = cos z
n=1

3. Dividing the difference of (36.3) and (36.4) by 2i gives us the other identity.

∞
X
−i (1 − (−1)n )Jn (z)in = sin z
n=1
∞
X
2 Jn (z)in−1 = sin z
n=1
odd n
∞
X
2 (−1)(n−1)/2 Jn (z) = sin z
n=1
odd n
∞
X
2 (−1)n J2n+1 (z) = sin z
n=0

1547
 4. Substituting −t for t in (36.2) yields
∞
1
X
e− 2 z(t−1/t) = Jn (z)(−t)n . (36.5)
n=−∞

We take the product of (36.2) and (36.5) to obtain the final identity.
∞
! ∞
!
1 1
X X
Jn (z)tn Jm (z)(−t)m = e 2 z(t−1/t) e− 2 z(t−1/t) = 1
n=−∞ m=−∞

Note that the coefficients of all powers of t except t0 in the product of sums must vanish.
∞
X
Jn (z)tn J−n (z)(−t)−n = 1
n=−∞
∞
X
Jn2 (z) = 1
n=−∞
∞
X
J02 (z) +2 Jn2 (z) = 1
n=1

Solution 36.10
First we make the change of variables y(x) = x1/2 v(x). We compute the derivatives of y(x).

1
y 0 = x1/2 v 0 + x−1/2 v,
2
1
y 00 = x1/2 v 00 + x−1/2 v 0 − x−3/2 v.
4

1548
We substitute these into the differential equation for y.

y 00 + xp−2 y = 0
1
x1/2 v 00 + x−1/2 v 0 − x−3/2 v + xp−3/2 v = 0
4 
2 00 0 p 1
x v + xv + x − v=0
4

Then we make the change of variables v(x) = u(ξ), ξ = p2 xp/2 . We write the derivatives in terms of ξ.

d dξ d d p d
x =x = xxp/2−1 = ξ
dx dx dξ dξ 2 dξ
2
d d d d p d p d p2 d2 p2 d
x2 2 + x =x x = ξ ξ = ξ2 2 + ξ
dx dx dx dx 2 dξ 2 dξ 4 dξ 4 dξ

We write the differential equation for u(ξ).

p2 2 00 p2 0
 2 
p 2 1
ξ u + ξu + ξ − u=0
4 4 4 4
 
00 1 0 1
u + u + 1− 2 2 u=0
ξ pξ

This is the Bessel equation of order 1/p. We can write the general solution for u in terms of Bessel functions of
the first kind if p 6= ±1. Otherwise, we use a Bessel function of the second kind.

u(ξ) = c1 J1/p (ξ) + c2 J−1/p (ξ) for p 6= 0, ±1

u(ξ) = c1 J1/p (ξ) + c2 Y1/p (ξ) for p 6= 0

1549
We write the solution in terms of y(x).

√ √
  
2 p/2 2 p/2
y(x) = c1 xJ1/p x + c2 xJ−1/p x for p 6= 0, ±1
p p
√ √
   
2 p/2 2 p/2
y(x) = c1 xJ1/p x + c2 xY1/p x for p 6= 0
p p

The Airy equation y 00 + xy = 0 is the case p = 3. The general solution of the Airy equation is

√ √
   
2 3/2 2 3/2
y(x) = c1 xJ1/3 x + c2 xJ−1/3 x .
3 3

Solution 36.11
Consider J1/2 (z). We start with the series expansion.
∞
X (−1)m  z 1/2+2m
J1/2 (z) = .
m=0
m!Γ(1/2 + m + 1) 2

(1)(3)···(2n−1) √
Use the identity Γ(n + 1/2) = 2n
π.

∞
X (−1)m 2m+1  z 1/2+2m
= √
m=0
m!(1)(3) · · · (2m + 1) π 2
∞  1/2+m
X (−1)m 2m+1 1
= √ z 1/2+2m
m=0
(2)(4) · · · (2m) · (1)(3) · · · (2m + 1) π 2
 1/2 X ∞
2 (−1)m 2m+1
= z
πz m=0
(2m + 1)!

1550
We recognize the sum as the Taylor series expansion of sin z.
 1/2
2
= sin z
πz
Using the recurrence relations,
ν ν
Jν+1 = Jν − Jν0 and Jν−1 = Jν + Jν0 ,
z z
we can find Jn+1/2 for any integral n.
We need J3/2 (z) to determine j1 (z). To find J3/2 (z),

1/2 0
J3/2 (z) = J1/2 (z) − J1/2 (z)
z
 1/2    1/2  1/2
1/2 2 −1/2 1 2 −3/2 2
= z sin z − − z sin z − z −1/2 cos z
z π 2 π π
= 2−1/2 π −1/2 z −3/2 sin z + 2−1/2 π −1/2 z −3/2 sin z − 2−1/2 π −1/2 cos z
 1/2  1/2
2 −3/2 2
= z sin z − z −1/2 cos z
π π
 1/2
2
z −3/2 sin z − z −1/2 cos z .


=
π

The spherical Bessel function j1 (z) is

sin z cos z
j1 (z) = − .
z2 z
The modified Bessel function of the first kind is

Iν (z) = i−ν Jν (iz).

1551
We can determine I1/2 (z) from J1/2 (z).
r
2
I1/2 (z) = i−1/2 sin(iz)
πiz
r
2
= −i i sinh(z)
πz
r
2
= sinh(z)
πz
The spherical Bessel function i0 (z) is
sinh z
i0 (z) = .
z
The modified Bessel function of the second kind is
π I−µ − Iµ
Kν (z) = lim
µ→ν 2 sin(µπ)
Thus K1/2 (z) can be determined in terms of I−1/2 (z) and I1/2 (z).
π

K1/2 (z) = I−1/2 − I1/2
2
We determine I−1/2 with the recursion relation
ν
Iν−1 (z) = Iν0 (z) + Iν (z).
z

0 1
I−1/2 (z) = I1/2 (z) + I1/2 (z)
r 2z r r
2 −1/2 1 2 −3/2 1 2 −1/2
= z cosh(z) − z sinh(z) + z sinh(z)
π 2 π 2z π
r
2
= cosh(z)
πz

1552
Now we can determine K1/2 (z).
r r !
π 2 2
K1/2 (z) = cosh(z) − sinh(z)
2 πz πz
r
π −z
= e
2z
The spherical Bessel function k0 (z) is
π −z
k0 (z) = e .
2z

Solution 36.12
The Point at Infinity. With the change of variables z = 1/t, w(z) = u(t) the modified Bessel equation becomes
n2
 
00 1 0
w + w − 1+ 2 w =0
z z
4 00 3 0 2 0
t u + 2t u + t(−t )u − (1 + n2 t2 )u = 0
n2
 
00 1 0 1
u + u − 4 − 2 u = 0.
t t t
The point t = 0 and hence the point z = ∞ is an irregular singular point. We will find the leading order
asymptotic behavior of the solutions as z → +∞.
Controlling Factor. Starting with the modified Bessel equation for real argument
n2
 
00 1 0
y + y − 1 + 2 y = 0,
x x
we make the substitution y = es(x) to obtain
1 n2
s00 + (s0 )2 + s0 − 1 − 2 = 0.
x x

1553
 n2
We know that x2
 1 as x → ∞; we will assume that s00  (s0 )2 as x → ∞. This gives us
1
(s0 )2 + s0 − 1 ∼ 0 as x → ∞.
x
To simplify the equation further, we will try the possible two-term balances.
1. (s0 )2 + x1 s0 ∼ 0 ⇒ s0 ∼ − x1 This balance is not consistent as it violates the assumption that 1 is
smaller than the other terms.

2. (s0 )2 − 1 ∼ 0 ⇒ s0 ∼ ±1 This balance is consistent.

1 0
3. x
s −1∼0 ⇒ s0 ∼ x This balance is inconsistent as (s0 )2 isn’t smaller than the other terms.
Thus the only dominant balance is s0 ∼ ±1. This balance is consistent with our initial assumption that
s  (s0 )2 . Thus s ∼ ±x and the controlling factor is e±x . We are interested in the decaying solution, so we will
00

work with the controlling factor e−x .

Leading Order Behavior. In order to find the leading order behavior, we substitute s = −x + t(x) where
t(x)  x as x → ∞ into the differential equation for s. We assume that t0  1 and t00  1/x.

1 n2
t00 + (−1 + t0 )2 + (−1 + t0 ) − 1 − 2 = 0
x x
2
1 1 n
t00 − 2t0 + (t0 )2 − + t0 − 2 = 0
x x x
Using our assumptions about the behavior of t0 and t00 ,

1
−2t0 − ∼0
x
1
t0 ∼ −
2x
1
t ∼ − log x as x → ∞.
2

1554
This asymptotic behavior is consistent with our assumptions.
Thus the leading order behavior of the decaying solution is
1
y ∼ c e−x− 2 log x+u(x) = cx−1/2 e−x+u(x) as x → ∞,

where u(x)  log x as x → ∞.

By substituting t = − 12 log x + u(x) into the differential equation for t, you could show that u(x) → const as
x → ∞. Thus the full leading order behavior of the decaying solution is

y ∼ cx−1/2 e−x as x → ∞

where u(x) → 0 as x → ∞.
Asymptotic Series. Now we find the full asymptotic series for Kn (x) as x → ∞. We substitute

e−x
Kn (x) ∝ √ w(x)
x

into the modified Bessel equation, where w(x) is a Taylor series about x = ∞, i.e.,

∞
e−x X
Kn (x) ∝ √ ak x−k .
x k=0

First we differentiate the expression for Kn (x).

e−x
   
1
Kn0 (x)∝ √ 0
w − 1+ w
x 2x
e−x
     
00 00 1 0 1 3
Kn (x) ∝ √ w − 2+ w + 1+ + 2 w
x x x 4x

1555
We substitute these expressions into the modified Bessel equation.

x2 y 00 + xy 0 − (x2 + n2 )y = 0
   
2 00 2

0 2 3 0 1
x w − 2x + x w + x + x + w + xw − x + w − (x2 + n2 )w = 0
4 2
 
2 00 2 0 1 2
x w − 2x w + −n w =0
4

The derivatives of the Taylor series are

∞
X
0
w = (−k)ak x−k−1 ,
k=1
X∞
= (−k − 1)ak+1 x−k−2 ,
k=0
∞
X
w00 = (−k)(−k − 1)ak x−k−2 ,
k=1
∞
X
= (−k)(−k − 1)ak x−k−2 .
k=0

We substitute these expression into the differential equation.

∞ ∞  X ∞
X
−k−2
X
−k−2 1
x2
k(k + 1)ak x + 2x 2
(k + 1)ak+1 x + −n 2
ak x−k = 0
k=0 k=0
4 k=0
∞ ∞  X∞
X X 1
k(k + 1)ak x−k + 2 (k + 1)ak+1 x−k + − n2 ak x−k = 0
k=0 k=0
4 k=0

1556
We equation coefficients of x to obtain a recurrence relation for the coefficients.
 
1 2
k(k + 1)ak + 2(k + 1)ak+1 + − n ak = 0
4
n2 − 1/4 − k(k + 1)
ak+1 = ak
2(k + 1)
n2 − (k + 1/2)2
ak+1 = ak
2(k + 1)
4n2 − (2k + 1)2
ak+1 = ak
8(k + 1)
We choose a0 = 1 and use the recurrence relation to determine the rest of the coefficients.
k
Y
ak = (8(n + 1))−k (4n2 − (2j − 1)2 ).
j=1

The asymptotic expansion of the modified Bessel function of the second kind is
∞ k
!
e−x X Y
Kn (x) ∝ √ (8(n + 1))−k (4n2 − (2j − 1)2 ) z −k , as x → ∞.
x k=0 j=1

Convergence. We determine the domain of convergence of the series with the ratio test. The Taylor series
about infinity will converge outside of some circle.
 
 an+1 (x) 
lim  <1
k→∞  an (x) 
 an+1 x−k−1 
 
lim  <1
k→∞  an x−k 
 2
 4n − (2k + 1)2  −1

lim   |x| < 1
k→∞  8(k + 1) 
∞ < |x|

1557
The series does not converge for any x in the finite complex plane. However, if we take only a finite number of
terms in the series, it gives a good approximation of Kn (x) for large, positive x. At x = 10, the one, two and
three term approximations give relative errors of 0.01, 0.0006 and 0.00006, respectively.

1558
 Part V

Partial Differential Equations

1559
Chapter 37

Transforming Equations

Let {xi } denote rectangular coordinates. Let {ai } be unit basis vectors in the orthogonal coordinate system {ξi }.
The distance metric coefficients hi can be defined
s 2  2  2
∂x1 ∂x2 ∂x3
hi = + + .
∂ξi ∂ξi ∂ξi

The gradient, divergence, etc., follow.

a1 ∂u a2 ∂u a3 ∂u
∇u = + +
h1 ∂ξ1 h2 ∂ξ2 h3 ∂ξ3
 
1 ∂ ∂ ∂
∇·v = (h2 h3 v1 ) + (h3 h1 v2 ) + (h1 h2 v3 )
h1 h2 h3 ∂ξ1 ∂ξ2 ∂ξ3
      
2 1 ∂ h2 h3 ∂u ∂ h3 h1 ∂u ∂ h1 h2 ∂u
∇ u= + +
h1 h2 h3 ∂ξ1 h1 ∂ξ1 ∂ξ2 h2 ∂ξ2 ∂ξ3 h3 ∂ξ3

1560
37.1 Exercises
Exercise 37.1
Find the Laplacian in cylindrical coordinates (r, θ, z).

x = r cos θ, y = r sin θ, z

Exercise 37.2
Find the Laplacian in spherical coordinates (r, φ, θ).

x = r sin φ cos θ, y = r sin φ sin θ, z = r cos φ

1561
37.2 Hints

1562
37.3 Solutions
Solution 37.1

p
h1 = (cos θ)2 + (sin θ)2 + 0 = 1
p
h2 = (−r sin θ)2 + (r cos θ)2 + 0 = r
√
h3 = 0 + 0 + 12 = 1

      
2 1 ∂ ∂u ∂ 1 ∂u ∂ ∂u
∇ u= r + + r
r ∂r ∂r ∂θ r ∂θ ∂z ∂z
2 2
 
1 ∂ ∂u 1∂ u ∂ u
∇2 u = r + 2 2 + 2
r ∂r ∂r r ∂θ ∂z

Solution 37.2

p
h1 = (sin φ cos θ)2 + (sin φ sin θ)2 + (cos φ)2 = 1
p
h2 = (r cos φ cos θ)2 + (r cos φ sin θ)2 + (−r sin φ)2 = r
p
h3 = (−r sin φ sin θ)2 + (r sin φ cos θ)2 + 0 = r sin φ

      
2 1 ∂ 2 ∂u ∂ ∂u ∂ 1 ∂u
∇ u= 2 r sin φ + sin φ +
r sin φ ∂r ∂r ∂φ ∂φ ∂θ sin φ ∂θ
1 ∂2u
   
1 ∂ ∂u 1 ∂ ∂u
∇2 u = 2 r2 + 2 sin φ + 2
r ∂r ∂r r sin φ ∂φ ∂φ r sin φ ∂θ2

1563
Chapter 38

Classification of Partial Differential Equations

38.1 Classification of Second Order Quasi-Linear Equations

Consider the general second order quasi-linear partial differential equation in two variables.

a(x, y)uxx + 2b(x, y)uxy + c(x, y)uyy = F (x, y, u, ux , uy ) (38.1)

We classify the equation by the sign of the discriminant. At a given point x0 , y0 , the equation is classified as one
of the following types:

b2 − ac > 0 : hyperbolic
b2 − ac = 0 : parabolic
b2 − ac < 0 : elliptic

If an equation has a particular type for all points x, y in a domain then the equation is said to be of that type
in the domain. Each of these types has a canonical form that can be obtained through a change of independent
variables. The type of an equation indicates much about the nature of its solution.
We seek a change of independent variables, (a different coordinate system), such that Equation 38.1 has a
simpler form. We will find that a second order quasi-linear partial differential equation in two variables can be

1564
transformed to one of the canonical forms:
uξη = G(ξ, η, u, uξ , uη ), hyperbolic
uξξ = G(ξ, η, u, uξ , uη ), parabolic
uξξ + uηη = G(ξ, η, u, uξ , uη ), elliptic

Consider the change of independent variables

ξ = ξ(x, y), η = η(x, y).
The partial derivatives of u are
ux = ξx uξ + ηx uη
uy = ξy uξ + ηy uη
uxx = ξx2 uξξ + 2ξx ηx uξη + ηx2 uηη + ξxx uξ + ηxx uη
uxy = ξx ξy uξξ + (ξx ηy + ξy ηx )uξη + ηx ηy uηη + ξxy uξ + ηxy uη
uyy = ξy2 uξξ + 2ξy ηy uξη + ηy2 uηη + ξyy uξ + ηyy uη .
Substituting these into (??) yields an equation in ξ and η.

aξx2 + 2bξx ξy + cξy2 uξξ + 2 (aξx ηx + b(ξx ηy + ξy ηx ) + cξy ηy ) uξη

+ aηx2 + 2bηx ηy + cηy2 uηη = H(ξ, η, u, uξ , uη )

α(ξ, η)uξξ + β(ξ, η)uξη + γ(ξ, η)uηη = H(ξ, η, u, uξ , uη ) (38.2)

38.1.1 Hyperbolic Equations

We start with a hyperbolic equation, (b2 − ac > 0). We seek a change of independent variables that will put
Equation 38.1 in the form
uξη = G(ξ, η, u, uξ , uη ) (38.3)

1565
We require that the uξξ and uηη terms vanish. That is α = γ = 0 in Equation 38.2. This gives us two constraints
on ξ and η.

aξx2 + 2bξx ξy + cξy2 = 0, aηx2 + 2bηx ηy + cηy2 = 0 (38.4)

√ √
ξx −b + b2 − ac ηx −b − b2 − ac
= , = .
ξy a ηy a
Here we chose the signs in the quadratic formulas to get different solutions for ξ and η.
Consider ξ(x, y) = const as an implicit equation for y in terms of x. We differentiate ξ with respect to x.
dξ dy
= ξ x + ξy =0
dx dx
The derivative of y(x) is
√
dy ξx b− b2 − ac
=− = .
dx ξy a
Solving this ordinary differential equation for y(x) determines ξ(x, y). We just write the solution for y(x) in the
form F (x, y(x)) = const. We then have ξ = F (x, y). Upon solving for ξ and η we divide Equation 38.2 by β(ξ, η)
to obtain the canonical form.

Note that we could have solved for ξy /ξx in Equation 38.4.

√
dx ξy b − b2 − ac
=− =
dy ξx c
This form is useful if a vanishes.

Another canonical form for hyperbolic equations is

uσσ − uτ τ = K(σ, τ, u, uσ , uτ ). (38.5)

1566
We can transform Equation 38.3 to this form with the change of variables
σ = ξ + η, τ = ξ − η.
Equation 38.3 becomes
 
σ+τ σ−τ
uσσ − uτ τ = G , , u, uσ + uτ , uσ − uτ .
2 2

Example 38.1.1 Consider the wave equation with a source.

utt − c2 uxx = s(x, t)
Since 0 − (1)(−c2 ) > 0, the equation is hyperbolic. We find the new variables.
dx
= −c, x = −ct + const, ξ = x + ct
dt
dx
= c, x = ct + const, η = x − ct
dt
Then we determine t and x in terms of ξ and η.
ξ−η ξ+η
t= , x=
2c 2
We calculate the derivatives of ξ and η.
ξt = c ξx = 1
ηt = −c ηx = 1
Then we calculate the derivatives of u.
utt = c2 uξξ − 2c2 uξη + c2 uηη
uxx = uξξ + uηη

1567
Finally we transform the equation to canonical form.
 
2 ξ+η ξ−η
−2c uξη = s ,
2 2c
 
1 ξ+η ξ−η
uξη = − 2 s ,
2c 2 2c
If s(x, t) = 0, then the equation is uξη = 0 we can integrate with respect to ξ and η to obtain the solution,
u = f (ξ) + g(η). Here f and g are arbitrary C 2 functions. In terms of t and x, we have
u(x, t) = f (x + ct) + g(x − ct).
To put the wave equation in the form of Equation 38.5 we make the change of variables
σ = ξ + η = 2x, τ = ξ − η = 2ct.
utt − c2 uxx = s(x, t)
σ τ 
2 2
4c uτ τ − 4c uσσ = s ,
2 2c
1 σ τ 
uσσ − uτ τ = − 2 s ,
4c 2 2c

Example 38.1.2 Consider

y 2 uxx − x2 uyy = 0.
For x 6= 0 and y 6= 0 this equation is hyperbolic. We find the new variables.
p
dy y 2 x2 x y2 x2
=− = − , y dy = −x dx, = − + const, ξ = y 2 + x2
dx y2 y 2 2
p
dy y 2 x2 x y2 x2
= = , y dy = x dx, = + const, η = y 2 − x2
dx y2 y 2 2

1568
 We calculate the derivatives of ξ and η.
ξx = 2x ξy = 2y
ηx = −2x ηy = 2y
Then we calculate the derivatives of u.
ux = 2x(uξ − uη )
uy = 2y(uξ + uη )
uxx = 4x2 (uξξ − 2uξη + uηη ) + 2(uξ − uη )
uyy = 4y 2 (uξξ + 2uξη + uηη ) + 2(uξ + uη )
Finally we transform the equation to canonical form.
y 2 uxx − x2 uyy = 0
−8x2 y 2 uξη − 8x2 y 2 uξη + 2y 2 (uξ − uη ) + 2x2 (uξ + uη ) = 0
1 1
16 (ξ − η) (ξ + η)uξη = 2ξuξ − 2ηuη
2 2
ξuξ − ηuη
uξη =
2(ξ 2 − η 2 )

Example 38.1.3 Consider Laplace’s equation.

uxx + uyy = 0
Since 0 − (1)(1) < 0, the equation is elliptic. We will transform this equation to the canical form of Equation 38.3.
We find the new variables.
dy
= −i, y = −ix + const, ξ = x + iy
dx
dy
= i, y = ix + const, η = x − iy
dx

1569
 We calculate the derivatives of ξ and η.

ξx = 1 ξ y = i
ηx = 1 ηy = −i

Then we calculate the derivatives of u.

uxx = uξξ + 2uξη + uηη

uyy = −uξξ + 2uξη − uηη

Finally we transform the equation to canonical form.

4uξη = 0
uξη = 0

We integrate with respect to ξ and η to obtain the solution, u = f (ξ) + g(η). Here f and g are arbitrary C 2
functions. In terms of x and y, we have

u(x, y) = f (x + iy) + g(x − iy).

This solution makes a lot of sense, because the real and imaginary parts of an analytic function are harmonic.

38.1.2 Parabolic equations

38.1.3 Elliptic Equations
We start with an elliptic equation, (b2 − ac < 0). We seek a change of independent variables that will put
Equation 38.1 in the form

uσσ + uτ τ = G(σ, τ, u, uσ , uτ ) (38.6)

1570
If we make the change of variables determined by
√ √
ξx −b + i ac − b2 ηx −b − i ac − b2
= , =
ξy a ηy a
The equation will have the form

uξη = G(ξ, η, u, uξ , uη ).

ξ and η are complex-valued. If we then make the change of variables

ξ+η ξ−η
σ= , τ=
2 2i
we will obtain the canonical form of Equation 38.6. Note that since ξ and η are complex conjugate, σ and τ are
real-valued.

Example 38.1.4 Consider

y 2 uxx + x2 uyy = 0. (38.7)

For x 6= 0 and y 6= 0 this equation is elliptic. We find new variables that will put this equation in the form
uξη = G(·). From Example 38.1.2 we see that they are
p
dy y 2 x2 x y2 x2
= −i = −i , y dy = −ix dx, = −i + const, ξ = y 2 + ix2
dx y2 y 2 2
p
dy y 2 x2 x y2 x2
=i = i , y dy = ix dx, = i + const, η = y 2 − ix2
dx y2 y 2 2

The variables that will put Equation 38.7 in canonical form are
ξ+η ξ−η
σ= = y2, τ= = x2
2 2i

1571
We calculate the derivatives of σ and τ .

σx = 0 σy = 2y
τx = 2x τy = 0

Then we calculate the derivatives of u.

ux = 2xuτ
uy = 2yuσ
uxx = 4x2 uτ τ + 2uτ
uyy = 4y 2 uσσ + 2uσ

Finally we transform the equation to canonical form.

y 2 uxx + x2 uyy = 0
σ(4τ uτ τ + 2uτ ) + τ (4σuσσ + 2uσ ) = 0
1 1
uσσ + uτ τ = − uσ − uτ
2σ 2τ

38.2 Equilibrium Solutions

Example 38.2.1 Consider the equilibrium solution for the following problem.

ut = uxx , u(x, 0) = x, ux (0, t) = ux (1, t) = 0.

Setting ut = 0 we have the ordinary differential equation

d2 u
= 0.
dx2

1572
This equation has the solution

u = ax + b.

Applying the boundary conditions we see that

u = b.

To determine the constant, we note that the heat energy in the rod is constant in time.
Z 1 Z 1
u(x, t) dx = u(x, 0) dx
0 0
Z 1 Z 1
b dx = x dx
0 0

Thus the equilibrium solution is

1
u(x) = .
2

1573
38.3 Exercises
Exercise 38.1
Classify as hyperbolic, parabolic, or elliptic in a region R each of the equations:

(a) ut = (pux )x

(b) utt = c2 uxx − γu

(c) (qux )x + (qut )t = 0

where p(x), c(x, t), q(x, t), and γ(x) are given functions that take on only positive values in a region R of the
(x, t) plane.

Exercise 38.2
Transform each of the following equations for φ(x, y) into canonical form in appropriate regions

(a) φxx − y 2 φyy + φx − φ + x2 = 0

(b) φxx + xφyy = 0

The equation in part (b) is known as Tricomi’s equation and is a model for transonic fluid flow in which the flow
speed changes from supersonic to subsonic.

1574
38.4 Hints
Hint 38.1

Hint 38.2

1575
38.5 Solutions
Exercise 38.3
1.

ut = (pux )x
puxx + 0uxt + 0utt + px ux − ut = 0

Since 02 − p0 = 0, the equation is parabolic.

2.

utt = c2 uxx − γu
utt + 0utx − c2 uxx + γu = 0

Since 02 − (1)(−c2 ) > 0, the equation is hyperbolic.

3.

(qux )x + (qut )t = 0
quxx + 0uxt + qutt + qx ux + qt ut = 0

Since 02 − qq < 0, the equation is elliptic.

Exercise 38.4
1. For y 6= 0, the equation is hyperbolic. We find the new independent variables.
p
dy y2
= = y, y = c ex , e−x y = c, ξ = e−x y
dx p1
dy − y2
= = −y, y = c e−x , ex y = c, η = ex y
dx 1

1576
Next we determine x and y in terms of ξ and η.
p
ξη = y 2 , y = ξη
 
p p 1 η
η = ex ξη, ex = η/ξ, x = log
2 ξ
We calculate the derivatives of ξ and η.
ξx = − e−x y = −ξ
p
ξy = e−x = ξ/η
ηx = e x y = η
p
ηy = ex = η/ξ
Then we calculate the derivatives of φ.
s r
∂ ∂ ∂ ∂ ξ ∂ η ∂
= −ξ +η , = +
∂x ∂ξ ∂η ∂y η ∂ξ ξ ∂η
s r
ξ η
φx = −ξφξ + ηφη , φy = φξ + φη
η ξ
ξ η
φxx = ξ 2 φξξ − 2ξηφξη + η 2 φηη + ξφξ + ηφη , φyy = φξξ + 2φξη + φηη
η ξ
Finally we transform the equation to canonical form.
φxx − y 2 φyy + φx − φ + x2 = 0
 
η
−4ξηφξη + ξφξ + ηφη − ξφξ + ηφη − φ + log =0
ξ
 
1 η
φξη = φη + φ − log
2ξ ξ

1577
 For y = 0 we have the ordinary differential equation

φxx + φx − φ + x2 = 0.

2. For x < 0, the equation is hyperbolic. We find the new independent variables.
dy √ 2 √ 2 √
= −x, y = x −x + c, ξ = x −x − y
dx 3 3
dy √ 2 √ 2 √
= − −x, y = − x −x + c, η = x −x + y
dx 3 3
Next we determine x and y in terms of ξ and η.
 1/3
3 η−ξ
x=− (ξ + η) , y=
4 2
We calculate the derivatives of ξ and η.
1/6
√

3
ξx = −x = (ξ + η) , ξy = −1
4
 1/6
3
ηx = (ξ + η) , ηy = 1
4
Then we calculate the derivatives of φ.
 1/6
3
φx = (ξ + η) (φξ + φη )
4
φy = −φξ + φη
 1/3
3
φxx = (ξ + η) (φξξ + φηη ) + (6(ξ + η))1/3 φξη + (6(ξ + η))−2/3 (φξ + φη )
4
φyy = φξξ − 2φξη + φηη

1578
Finally we transform the equation to canonical form.

φxx + xφyy = 0
(6(ξ + η))1/3 φξη + (6(ξ + η))1/3 φξη + (6(ξ + η))−2/3 (φξ + φη ) = 0
φξ + φη
φξη = −
12(ξ + η)

For x > 0, the equation is elliptic. The variables we defined before are complex-valued.
2 2
ξ = i x3/2 − y, η = i x3/2 + y.
3 3
We choose the new real-valued variables.

α = ξ − η, β = −i(ξ + η)

We write the derivatives in terms of α and β.

φξ = φα − iφβ
φη = −φα − iφβ
φξη = −φαα − φββ

We transform the equation to canonical form.

φξ + φη
φξη = −
12(ξ + η)
−2iφβ
−φαα − φββ = −
12iβ
φβ
φαα + φββ = −
6β

1579
Chapter 39

Separation of Variables

39.1 Eigensolutions of Homogeneous Equations

39.2 Homogeneous Equations with Homogeneous Boundary Condi-

tions
The method of separation of variables is a useful technique for finding special solutions of partial differential
equations. We can combine these special solutions to solve certain problems. Consider the temperature of a
one-dimensional rod of length h 1 . The left end is held at zero temperature, the right end is insulated and the
initial temperature distribution is known at time t = 0. To find the temperature we solve the problem:

∂u ∂2u
= κ 2, 0 < x < h, t > 0
∂t ∂x
u(0, t) = ux (h, t) = 0
u(x, 0) = f (x)
1
Why h? Because l looks like 1 and we use L to denote linear operators

1580
We look for special solutions of the form, u(x, t) = X(x)T (t). Substituting this into the partial differential equation
yields
X(x)T 0 (t) = κX 00 (x)T (t)
T 0 (t) X 00 (x)
=
κT (t) X(x)
Since the left side is only dependent on t, the right side in only dependent on x, and the relation is valid for all t
and x, both sides of the equation must be constant.
T0 X 00
= = −λ
κT X
Here −λ is an arbitrary constant. (You’ll see later that this form is convenient.) u(x, t) = X(x)T (t) will satisfy
the partial differential equation if X(x) and T (t) satisfy the ordinary differential equations,
T 0 = −κλT and X 00 = −λX.
Now we see how lucky we are that this problem happens to have homogeneous boundary conditions 2 . If the left
boundary condition had been u(0, t) = 1, this would imply X(0)T (t) = 1 which tells us nothing very useful about
either X or T . However the boundary condition u(0, t) = X(0)T (t) = 0, tells us that either X(0) = 0 or T (t) = 0.
Since the latter case would give us the trivial solution, we must have X(0) = 0. Likewise by looking at the right
boundary condition we obtain X 0 (h) = 0.

We have a regular Sturm-Liouville problem for X(x).

X 00 + λX = 0, X(0) = X 0 (h) = 0
The eigenvalues and orthonormal eigenfunctions are
 2 r  
(2n − 1)π 2 (2n − 1)π
λn = , Xn = sin x , n ∈ Z+ .
2h h 2h
2
Actually luck has nothing to do with it. I planned it that way.

1581
 Now we solve the equation for T (t).

T 0 = −κλn T
T = c e−κλn t

The eigen-solutions of the partial differential equation that satisfy the homogeneous boundary conditions are
r
2 p 
un (x, t) = sin λn x e−κλn t .
h
We seek a solution of the problem that is a linear combination of these eigen-solutions.
∞ r
X 2 p 
u(x, t) = an sin λn x e−κλn t
n=1
h

We apply the initial condition to find the coefficients in the expansion.

∞ r
X 2 p 
u(x, 0) = an sin λn x = f (x)
n=1
h
r Z h
2 p 
an = sin λn x f (x) dx
h 0

39.3 Time-Independent Sources and Boundary Conditions

Consider the temperature in a one-dimensional rod of length h. The ends are held at temperatures α and β,
respectively, and the initial temperature is known at time t = 0. Additionally, there is a heat source, s(x), that
is independent of time. We find the temperature by solving the problem,

ut = κuxx + s(x), u(0, t) = α, u(h, t) = β, u(x, 0) = f (x). (39.1)

1582
Because of the source term, the equation is not separable, so we cannot directly apply separation of variables.
Furthermore, we have the added complication of inhomogeneous boundary conditions. Instead of attacking this
problem directly, we seek a transformation that will yield a homogeneous equation and homogeneous boundary
conditions.

Consider the equilibrium temperature, µ(x). It satisfies the problem,

s(x)
µ00 (x) = − = 0, µ(0) = α, µ(h) = β.
κ
The Green function for this problem is,
x< (x> − h)
G(x; ξ) = .
h
The equilibrium temperature distribution is
Z h
x−h x 1
µ(x) = α +β − x< (x> − h)s(ξ) dξ,
h h κh 0
 Z x Z h 
x 1
µ(x) = α + (β − α) − (x − h) ξs(ξ) dξ + x (ξ − h)s(ξ) dξ .
h κh 0 x

Now we substitute u(x, t) = v(x, t) + µ(x) into Equation 39.1.

∂ ∂2
(v + µ(x)) = κ 2 (v + µ(x)) + s(x)
∂t ∂x
vt = κvxx + κµ00 (x) + s(x)
vt = κvxx (39.2)
Since the equilibrium solution satisfies the inhomogeneous boundary conditions, v(x, t) satisfies homogeneous
boundary conditions.
v(0, t) = v(h, t) = 0.

1583
The initial value of v is

v(x, 0) = f (x) − µ(x).

We seek a solution for v(x, t) that is a linear combination of eigen-solutions of the heat equation. We substitute
the separation of variables, v(x, t) = X(x)T (t) into Equation 39.2
T0 X 00
= = −λ
κT X
This gives us two ordinary differential equations.

X 00 + λX = 0, X(0) = X(h) = 0
0
T = −κλT.

The Sturm-Liouville problem for X(x) has the eigenvalues and orthonormal eigenfunctions,
 nπ 2 r
2  nπx 
λn = , Xn = sin , n ∈ Z+ .
h h h
We solve for T (t).
2
Tn = c e−κ(nπ/h) t .

The eigen-solutions of the partial differential equation are

r
2  nπx  2
vn (x, t) = sin e−κ(nπ/h) t .
h h
The solution for v(x, t) is a linear combination of these.
∞ r
X 2  nπx  2
v(x, t) = an sin e−κ(nπ/h) t
n=1
h h

1584
We determine the coefficients in the series with the initial condition.
∞ r
X 2  nπx 
v(x, 0) = an sin = f (x) − µ(x)
n=1
h h
r Z h
2  nπx 
an = sin (f (x) − µ(x)) dx
h 0 h
The temperature of the rod is
∞ r
X 2  nπx  2
u(x, t) = µ(x) + an sin e−κ(nπ/h) t
n=1
h h

39.4 Inhomogeneous Equations with Homogeneous Boundary Con-

ditions
Now consider the heat equation with a time dependent source, s(x, t).
ut = κuxx + s(x, t), u(0, t) = u(h, t) = 0, u(x, 0) = f (x). (39.3)
In general we cannot transform the problem to one with a homogeneous differential equation. Thus we cannot
represent the solution in a series of the eigen-solutions of the partial differential equation. Instead, we will do the
next best thing and expand the solution in a series of eigenfunctions in Xn (x) where the coefficients depend on
time.
X∞
u(x, t) = un (t)Xn (x)
n=1

We will find these eigenfunctions with the separation of variables, u(x, t) = X(x)T (t) applied to the homogeneous
equation, ut = κuxx , which yields,
r
2  nπx 
Xn (x) = sin , n ∈ Z+ .
h h

1585
We expand the heat source in the eigenfunctions.
∞ r
X 2  nπx 
s(x, t) = sn (t) sin
n=1
h h
r Z h
2  nπx 
sn (t) = sin s(x, t) dx,
h 0 h
We substitute the series solution into Equation 39.3.
∞ r ∞  nπ 2 r 2 ∞ r
X 2  nπx  X  nπx  X 2  nπx 
u0n (t) sin = −κ un (t) sin + sn (t) sin
n=1
h h n=1
h h h n=1
h h
 nπ 2
u0n (t) + κ un (t) = sn (t)
h
Now we have a first order, ordinary differential equation for each of the un (t). We obtain initial conditions from
the initial condition for u(x, t).
∞ r
X 2  nπx 
u(x, 0) = un (0) sin = f (x)
n=1
h h
r Z h
2  nπx 
un (0) = sin f (x) dx ≡ fn
h 0 h
The temperature is given by
∞ r
X 2  nπx 
u(x, t) = un (t) sin ,
n=1
h h
Z t
−κ(nπ/h)2 t 2 (t−τ )
un (t) = fn e + e−κ(nπ/h) sn (τ ) dτ.
0

1586
39.5 Inhomogeneous Boundary Conditions
Consider the temperature of a one-dimensional rod of length h. The left end is held at the temperature α(t),
the heat flow at right end is specified, there is a time-dependent source and the initial temperature distribution
is known at time t = 0. To find the temperature we solve the problem:
ut = κuxx + s(x, t), 0 < x < h, t > 0 (39.4)
u(0, t) = α(t), ux (h, t) = β(t) u(x, 0) = f (x)

Transformation to a homogeneous equation. Because of the inhomogeneous boundary conditions, we

cannot directly apply the method of separation of variables. However we can transform the problem to an
inhomogeneous equation with homogeneous boundary conditions. To do this, we first find a function, µ(x, t)
which satisfies the boundary conditions. We note that
µ(x, t) = α(t) + xβ(t)
does the trick. We make the change of variables
u(x, t) = v(x, t) + µ(x, t)
in Equation 39.4.
vt + µt = κ (vxx + µxx ) + s(x, t)
vt = κvxx + s(x, t) − µt
The boundary and initial conditions become
v(0, t) = 0, vx (h, t) = 0, v(x, 0) = f (x) − µ(x, 0).
Thus we have a heat equation with the source s(x, t) − µt (x, t). We could apply separation of variables to find a
solution of the form
∞ r  
X 2 (2n − 1)πx
u(x, t) = µ(x, t) + un (t) sin .
n=1
h 2h

1587
Direct eigenfunction expansion. Alternatively we could seek a direct eigenfunction expansion of u(x, t).

∞ r  
X 2 (2n − 1)πx
u(x, t) = un (t) sin .
n=1
h 2h

Note that the eigenfunctions satisfy the homogeneous boundary conditions while u(x, t) does not. If we choose
any fixed time t = t0 and form the periodic extension of the function u(x, t0 ) to define it for x outside the range
(0, h), then this function will have jump discontinuities. This means that our eigenfunction expansion will not
converge uniformly. We are not allowed to differentiate the series with respect to x. We can’t just plug the series
into the partial differential equation to determine the coefficients. Instead, we will multiply Equation 39.4, by an
eigenfunction and integrate from x = 0 to x = h. To avoid differentiating the series with respect to x, we will use
2
(2n−1)π
integration by parts to move derivatives from u(x, t) to the eigenfunction. (We will denote λn = 2h
.)

r Z h r Z h
2 p 2 p
sin( λn x)(ut − κuxx ) dx = sin( λn x)s(x, t) dx
h 0 h 0
r h r Z h
0 2 p i h 2 p p
un (t) − κ ux sin( λn x) + κ λn ux cos( λn x) dx = sn (t)
h 0 h 0
r r r Z h
0 2 n 2 p h p ih 2 p
un (t) − κ(−1) ux (h, t) + κ λn u cos( λn x) + κλn u sin( λn x) dx = sn (t)
h h 0 h 0
r r
2 2 p
u0n (t) − κ(−1)n β(t) − κ λn u(0, t) + κλn un (t) = sn (t)
h h
r 
0 2 p n

un (t) + κλn un (t) = κ λn α(t) + (−1) β(t) + sn (t)
h

1588
Now we have an ordinary differential equation for each of the un (t). We obtain initial conditions for them using
the initial condition for u(x, t).
∞ r
X 2 p
u(x, 0) = un (0) sin( λn x) = f (x)
n=1
h
r Z h
2 p
un (0) = sin( λn x)f (x) dx ≡ fn
h 0

Thus the temperature is given by

r ∞
2X p
u(x, t) = un (t) sin( λn x),
h n=1
r Z t
−κλn t 2 p 
un (t) = fn e + κ e−κλn (t−τ ) λn α(τ ) + (−1)n β(τ ) dτ.
h 0

39.6 The Wave Equation

Consider an elastic string with a free end at x = 0 and attached to a massless spring at x = 1. The partial
differential equation that models this problem is

utt = uxx
ux (0, t) = 0, ux (1, t) = −u(1, t), u(x, 0) = f (x), ut (x, 0) = g(x).

We make the substitution u(x, t) = ψ(x)φ(t) to obtain

φ00 ψ 00
= = −λ.
φ ψ

1589
 First we consider the problem for ψ.

ψ 00 + λψ = 0, ψ 0 (0) = ψ(1) + ψ 0 (1) = 0.

To find the eigenvalues we consider the following three cases:

λ < 0. The general solution is

√ √
ψ = a cosh( −λx) + b sinh( −λx).

ψ 0 (0) = 0 ⇒ b = 0.
0
√ √ √
ψ(1) + ψ (1) = 0 ⇒ a cosh( −λ) + a −λ sinh( −λ) = 0
⇒ a = 0.

Since there is only the trivial solution, there are no negative eigenvalues.

λ = 0. The general solution is

ψ = ax + b.

ψ 0 (0) = 0 ⇒ a = 0.
ψ(1) + ψ 0 (1) = 0 ⇒ b + 0 = 0.

Thus λ = 0 is not an eigenvalue.

λ > 0. The general solution is

√ √
ψ = a cos( λx) + b sin( λx).

1590
 ψ 0 (0) ⇒ b = 0.
√ √ √
ψ(1) + ψ 0 (1) = 0 ⇒ a cos( λ) − a λ sin( λ) = 0
√ √ √
⇒ cos( λ) = λ sin( λ)
√ √
⇒ λ = cot( λ)

By looking at Figure 39.1, (the plot shows the functions f (x) = x, f (x) = cot x and has lines at x = nπ),
we see that there are an infinite number of positive eigenvalues and that

λn → (nπ)2 as n → ∞.

The eigenfunctions are

p
ψn = cos( λn x).

The solution for φ is

p p
φn = an cos( λn t) + bn sin( λn t).

Thus the solution to the differential equation is

∞
X p p p
u(x, t) = cos( λn x)[an cos( λn t) + bn sin( λn t)].
n=1

Let
∞
X p
f (x) = fn cos( λn x)
n=1
∞
X p
g(x) = gn cos( λn x).
n=1

1591
 10

2 4 6 8 10

-2

Figure 39.1: Plot of x and cot x.

From the initial value we have

∞
X p ∞
X p
cos( λn x)an = fn cos( λn x)
n=1 n=1
an = fn .

The initial velocity condition gives us

∞
X p p ∞
X p
cos( λn x) λn bn = gn cos( λn x)
n=1 n=1
gn
bn = √ .
λn

1592
 Thus the solution is
∞  
X p p gn p
u(x, t) = cos( λn x) fn cos( λn t) + √ sin( λn t) .
n=1
λn

39.7 General Method

Here is an outline detailing the method of separation of variables for a linear partial differential equation for
u(x, y, z, . . . ).

1. Substitute u(x, y, z, . . . ) = X(x)Y (y)Z(z) · · · into the partial differential equation. Separate the equation
into ordinary differential equations.

2. Translate the boundary conditions for u into boundary conditions for X, Y , Z, . . . . The continuity of u
may give additional boundary conditions and boundedness conditions.

3. Solve the differential equation(s) that determine the eigenvalues. Make sure to consider all cases. The
eigenfunctions will be determined up to a multiplicative constant.

4. Solve the rest of the differential equations subject to the homogeneous boundary conditions. The eigenvalues
will be a parameter in the solution. The solutions will be determined up to a multiplicative constant.

5. The eigen-solutions are the product of the solutions of the ordinary differential equations. φn = Xn Yn Zn · · · .
The solution of the partial differential equation is a linear combination of the eigen-solutions.
X
u(x, y, z, . . . ) = an φn

6. Solve for the coefficients, an using the inhomogeneous boundary conditions.

1593
39.8 Exercises
Exercise 39.1
Obtain Poisson’s formula to solve the Dirichlet problem for the circular region 0 ≤ r < R, 0 ≤ θ < 2π. That is,
determine a solution φ(r, θ) to Laplace’s equation
∇2 φ = 0
in polar coordinates given φ(R, θ). Show that
Z 2π
1 R2 − r 2
φ(r, θ) = φ(R, α) 2 dα
2π 0 R + r2 − 2Rr cos(θ − α)

Exercise 39.2
Consider the temperature of a ring of unit radius. Solve the problem
ut = κuθθ , u(θ, 0) = f (θ)
with separation of variables.

Exercise 39.3
Solve the Laplace’s equation by separation of variables.
∆u ≡ uxx + uyy = 0, 0 < x < 1, 0 < y < 1,
u(x, 0) = f (x), u(x, 1) = 0, u(0, y) = 0, u(1, y) = 0
Here f (x) is an arbitrary function which is known.

Exercise 39.4
Solve the following problem by separation of variables:
∂ 2 u 1 ∂u 1 ∂2u
+ + = 0, 0 < r < 1,
∂r2 r ∂r r2 ∂θ2

1594
 u(1, θ) = f (θ).
Thus, you must find the function u = u(r, θ) which satisfies the partial differential equation inside the unit circle
and which takes on the values of f (θ) on the circumference.

Exercise 39.5
Find the normal modes of oscillation of a drum head of unit radius. The drum head obeys the wave equation
with zero displacement on the boundary.
1 ∂2v 1 ∂2v
 
1 ∂ ∂v
∆v ≡ r + 2 2 = 2 2, v(1, θ, t) = 0
r ∂r ∂r r ∂θ c ∂t

Exercise 39.6
Solve the equation
φt = a2 φxx , 0 < x < l, t>0
with boundary conditions φ(0, t) = φ(l, t) = 0, and initial conditions
(
x, 0 ≤ x ≤ l/2,
φ(x, 0) =
l − x, l/2 < x ≤ l.
Comment on the differentiability ( that is the number of finite derivatives with respect to x ) at time t = 0 and
at time t = , where  > 0 and   1.

Exercise 39.7
Consider a one-dimensional rod of length L with initial temperature distribution f (x). The temperatures at the
left and right ends of the rod are held at T0 and T1 , respectively. To find the temperature of the rod for t > 0,
solve
ut = κuxx , 0 < x < L, t > 0
u(0, t) = T0 , u(L, t) = T1 , u(x, 0) = f (x),

1595
with separation of variables.

Exercise 39.8
For 0 < x < l solve the problem
φt = a2 φxx + w(x, t) (39.5)
φ(0, t) = 0, φx (l, t) = 0, φ(x, 0) = f (x)
by means of a series expansion involving the eigenfunctions of
d2 β(x)
+ λβ(x) = 0,
dx2
β(0) = β 0 (l) = 0.
Here w(x, t) and f (x) are prescribed functions.

Exercise 39.9
Solve the heat equation of Exercise 39.8 with the same initial conditions but with the boundary conditions
φ(0, t) = 0, cφ(l, t) + φx (l, t) = 0.
Here c > 0 is a constant. Although it is not possible to solve for the eigenvalues λ in closed form, show that the
eigenvalues assume a simple form for large values of λ.

Exercise 39.10
Use a series expansion technique to solve the problem
φt = a2 φxx + 1, t > 0, 0<x<l
with boundary and initial conditions given by
φ(x, 0) = 0, φ(0, t) = t, φx (l, t) = −cφ(l, t)
where c > 0 is a constant.

1596
Exercise 39.11
Let φ(x, t) satisfy the equation
φt = a2 φxx
for 0 < x < l, t > 0 with initial conditions φ(x, 0) = 0 for 0 < x < l, with boundary conditions φ(0, t) = 0 for
t > 0, and φ(l, t) + φx (l, t) = 1 for t > 0. Obtain two series solutions for this problem, one which is useful for
large t and the other useful for small t.

Exercise 39.12
A rod occupies the portion 1 < x < 2 of the x-axis. The thermal conductivity depends on x in such a manner
that the temperature φ(x, t) satisfies the equation
φt = A2 (x2 φx )x (39.6)
where A is a constant. For φ(1, t) = φ(2, t) = 0 for t > 0, with φ(x, 0) = f (x) for 1 < x < 2, show that the
appropriate series expansion involves the eigenfunctions
 
1 πn log x
βn (x) = √ sin .
x log 2
Work out the series expansion for the given boundary and initial conditions.

Exercise 39.13
Consider a string of length L with a fixed left end a free right end. Initially the string is at rest with displacement
f (x). Find the motion of the string by solving,
utt = c2 uxx , 0 < x < L, t > 0,
u(0, t) = 0, ux (L, t) = 0,
u(x, 0) = f (x), ut (x, 0) = 0,
with separation of variables.

1597
Exercise 39.14
Consider the equilibrium temperature distribution in a two-dimensional block of width a and height b. There
is a heat source given by the function f (x, y). The vertical sides of the block are held at zero temperature; the
horizontal sides are insulated. To find this equilibrium temperature distribution, solve the potential equation,

uxx + uyy = f (x, y), 0 < x < a, 0 < y < b,

u(0, y) = u(a, y) = 0, uy (x, 0) = uy (x, b) = 0,

with separation of variables.

Exercise 39.15
Consider the vibrations of a stiff beam of length L. More precisely, consider the transverse vibrations of an
unloaded beam, whose weight can be neglected compared to its stiffness. The beam is simply supported at
x = 0, L. (That is, it is resting on fulcrums there. u(0, t) = 0 means that the beam is resting on the fulcrum;
uxx (0, t) = 0 indicates that there is no bending force at that point.) The beam has initial displacement f (x) and
velocity g(x). To determine the motion of the beam, solve

utt + a2 uxxxx = 0, 0 < x < L, t > 0,

u(x, 0) = f (x), ut (x, 0) = g(x),
u(0, t) = uxx (0, t) = 0, u(L, t) = uxx (L, t) = 0,

with separation of variables.

Exercise 39.16
The temperature along a magnet winding of length L carrying a current I satisfies, (for some α > 0):

ut = κuxx + I 2 αu.

The ends of the winding are kept at zero, i.e.,

u(0, t) = u(L, t) = 0;

1598
and the initial temperature distribution is

u(x, 0) = g(x).

Find u(x, t) and determine the critical current ICR which is defined as the least current at which the winding
begins to heat up exponentially. Suppose that α < 0, so that the winding has a negative coefficient of resistance
with respect to temperature. What can you say about the critical current in this case?

Exercise 39.17
The ”e-folding” time of a decaying function of time is the time interval, ∆e , in which the magnitude of the function
is reduced by at least 1e . Thus if u(x, t) = e−αt f (x) + e−βt g(x) with α > β > 0 then ∆e = β1 . A body with heat
conductivity κ has its exterior surface maintained at temperature zero. Initially the interior of the body is at the
uniform temperature T > 0. Find the e-folding time of the body if it is:

a) An infinite slab of thickness a.

b) An infinite cylinder of radius a.

c) A sphere of radius a.

Note that in (a) the temperature varies only in the z direction and in time; in (b) and (c) the temperature varies
only in the radial direction and in time.

∂u
d) What are the e-folding times if the surfaces are perfectly insulated, (i.e., ∂n
= 0, where n is the exterior
normal at the surface)?

Exercise 39.18
Solve the heat equation with a time-dependent diffusivity in the rectangle 0 < x < a, 0 < y < b. The top and
bottom sides are held at temperature zero; the lateral sides are insulated. We have the initial-boundary value

1599
problem:

ut = κ(t) (uxx + uyy ) , 0 < x < a, 0 < y < b, t > 0,

u(x, 0, t) = u(x, b, t) = 0,
ux (0, y, t) = ux (a, y, t) = 0,
u(x, y, 0) = f (x, y).

The diffusivity, κ(t), is a known, positive function.

Exercise 39.19
A semi-circular rod of infinite extent is maintained at temperature T = 0 on the flat side and at T = 1 on the
curved surface:

x2 + y 2 = 1, y > 0.

Find the steady state temperature in a cross section of the rod using separation of variables.

Exercise 39.20
Use separation of variables to find the steady state temperature u(x, y) in a slab: x ≥ 0, 0 ≤ y ≤ 1, which has
zero temperature on the faces y = 0 and y = 1 and has a given distribution: u(y, 0) = f (y) on the edge x = 0,
0 ≤ y ≤ 1.

Exercise 39.21
Find u(r, θ) which satisfies:

∆u = 0, 0 < θ < α, a < r < b,

subject to the boundary conditions:

u(r, 0) = u(r, α) = 0, u(a, θ) = 0, u(b, θ) = f (θ).

1600
Exercise 39.22

a) A piano string of length L is struck, at time t = 0, by a flat hammer of width 2d centered at a point ξ,
having velocity v. Find the ensuing motion, u(x, t), of the string for which the wave speed is c.
b) Suppose the hammer is curved, rather than flat as above, so that the initial velocity distribution is
(  
π(x−ξ)
v cos 2d
, |x − ξ| < d
ut (x, 0) =
0 |x − ξ| > d.

Find the ensuing motion.

c) Compare the kinetic energies of each harmonic in the two solutions. Where should the string be struck in
order to maximize the energy in the nth harmonic in each case?

Exercise 39.23
If the striking hammer is not perfectly rigid, then its effect must be included as a time dependent forcing term of
the form:
(  
v cos π(x−ξ) πt


2d
sin δ
, for |x − ξ| < d, 0 < t < δ,
s(x, t) =
0 otherwise.

Find the motion of the string for t > δ. Discuss the effects of the width of the hammer and duration of the blow
with regard to the energy in overtones.

Exercise 39.24
Find the propagating modes in a square waveguide of side L for harmonic signals of frequency ω when the
propagation speed of the medium is c. That is, we seek those solutions of

utt − c2 ∆u = 0,

1601
where u = u(x, y, z, t) has the form u(x, y, z, t) = v(x, y, z) eiωt , which satisfy the conditions:

u(x, y, z, t) = 0 for x = 0, L, y = 0, L, z > 0,

lim |u| 6= ∞ and 6= 0.
z→∞

Indicate in terms of inequalities involving k = ω/c and appropriate eigenvalues, λn,m say, for which n and m the
solutions un,m satisfy the conditions.

Exercise 39.25
Find the modes of oscillation and their frequencies for a rectangular drum head of width a and height b. The
modes of oscillation are eigensolutions of

utt = c2 ∆u, 0 < x < a, 0 < y < b,

u(0, y) = u(a, y) = u(x, 0) = u(x, b) = 0.

Exercise 39.26
Using separation of variables solve the heat equation

φt = a2 (φxx + φyy )

in the rectangle 0 < x < lx , 0 < y < ly with initial conditions

φ(x, y, 0) = 1,

and boundary conditions

φ(0, y, t) = φ(lx , y, t) = 0, φy (x, 0, t) = φy (x, ly , t) = 0.

1602
Exercise 39.27
Using polar coordinates and separation of variables solve the heat equation
φt = a2 ∇2 φ
in the circle 0 < r < R0 with initial conditions
φ(r, θ, 0) = V
where V is a constant, and boundary conditions
φ(R0 , θ, t) = 0.
(a) Show that for t > 0,
∞  2 2 
X a j0,n J0 (j0,n r/R0 )
φ(r, θ, t) = 2V exp − 2 t ,
n=1
R0 j0,n J1 (j0,n )

where j0,n are the roots of J0 (x):

J0 (j0,n ) = 0, n = 1, 2, . . .
Hint: The following identities may be of some help:
Z R0
rJ0 (j0,n r/R0 ) J0 (j0,m r/R0 ) dr = 0, m 6= n,
0
Z R0
2 R02 2
rJ0 (j0,n r/R0 ) dr = J (j0,n ),
2 1
Z0 r
r
rJ0 (βr)dr = J1 (βr) for any β.
0 β

(b) For any fixed r, 0 < r < R0 , use the asymptotic approximation for the Jn Bessel functions for large argument
(this can be found in the notes for second quarter, AMa95b or in any standard math tables) to determine
the rate of decay of the terms of the series solution for φ at time t = 0.

1603
Exercise 39.28
Consider the solution of the diffusion equation in spherical coordinates given by

x = r sin θ cos φ,
y = r sin θ sin φ,
z = r cos θ,

where r is the radius, θ is the polar angle, and φ is the azimuthal angle. We wish to solve the equation on the
surface of the sphere given by r = R, 0 < θ < π, and 0 < φ < 2π. The diffusion equation for the solution
Ψ(θ, φ, t) in these coordinates on the surface of the sphere becomes

a2 1 ∂2Ψ
   
∂Ψ 1 ∂ ∂Ψ
= 2 sin θ + . (39.7)
∂t R sin θ ∂θ ∂θ sin2 θ ∂φ2
where a is a positive constant.
(a) Using separation of variables show that a solution Ψ can be found in the form

Ψ(θ, φ, t) = T (t)Θ(θ)Φ(φ),

where T ,Θ,Φ obey ordinary differential equations in t,θ, and φ respectively. Derive the ordinary differential
equations for T and Θ, and show that the differential equation obeyed by Φ is given by
d2 Φ
− cΦ = 0,
dφ2
where c is a constant.

(b) Assuming that Ψ(θ, φ, t) is determined over the full range of the azimuthal angle, 0 < φ < 2π, determine
the allowable values of the separation constant c and the corresponding allowable functions Φ. Using these
values of c and letting x = cos θ rewrite in terms of the variable x the differential equation satisfied by Θ.
What are appropriate boundary conditions for Θ? The resulting equation is known as the generalized or
associated Legendre equation.

1604
 (c) Assume next that the initial conditions for Ψ are chosen such that

Ψ(θ, φ, t = 0) = f (θ),

where f (θ) is a specified function which is regular at the north and south poles (that is θ = 0 and θ = π).
Note that the initial condition is independent of the azimuthal angle φ. Show that in this case the method
of separation of variables gives a series solution for Ψ of the form
∞
X
Ψ(θ, t) = Al exp(−λ2l t)Pl (cos θ),
l=0

where Pl (x) is the l’th Legendre polynomial, and determine the constants λl as a function of the index l.

(d) Solve for Ψ(θ, t), t > 0 given that f (θ) = 2 cos2 θ − 1.

Useful facts:
 
d 2 dPl (x)
(1 − x ) + l(l + 1)Pl (x) = 0
dx dx

P0 (x) = 1
P1 (x) = x
3 2 1
P2 (x) = x −
2 2

if l 6= m
(
Z 1 0
dxPl (x)Pm (x) = 2
−1 if l = m
2l + 1

1605
Exercise 39.29
Let φ(x, y) satisfy Laplace’s equation
φxx + φyy = 0
in the rectangle 0 < x < 1, 0 < y < 2, with φ(x, 2) = x(1 − x), and with φ = 0 on the other three sides. Use a
series solution to determine φ inside the rectangle. How many terms are required to give φ( 12 , 1) with about 1%
(also 0.1%) accuracy; how about φx ( 21 , 1)?

Exercise 39.30
Let ψ(r, θ, φ) satisfy Laplace’s equation in spherical coordinates in each of the two regions r < a, r > a, with
ψ → 0 as r → ∞. Let
lim ψ(r, θ, φ) − lim− ψ(r, θ, φ) = 0,
r→a+ r→a
lim ψr (r, θ, φ) − lim− ψr (r, θ, φ) = Pnm (cos θ) sin(mφ),
r→a+ r→a

where m and n ≥ m are integers. Find ψ in r < a and r > a. In electrostatics, this problem corresponds to that of
determining the potential of a spherical harmonic type charge distribution over the surface of the sphere. In this
way one can determine the potential due to an arbitrary surface charge distribution since any charge distribution
can be expressed as a series of spherical harmonics.

Exercise 39.31
Obtain a formula analogous to the Poisson formula to solve the Neumann problem for the circular region 0 ≤ r <
R, 0 ≤ θ < 2π. That is, determine a solution φ(r, θ) to Laplace’s equation
∇2 φ = 0
in polar coordinates given φr (R, θ). Show that
R 2π r2
Z  
2r
φ(r, θ) = − φr (R, α) ln 1 − cos(θ − α) + 2 dα
2π 0 R R
within an arbitrary additive constant.

1606
Exercise 39.32
Investigate solutions of

φt = a2 φxx

obtained by setting the separation constant C = (α + iβ)2 in the equations obtained by assuming φ = X(x)T (t):

T0 X 00 C
= C, = 2.
T X a

1607
39.9 Hints
Hint 39.1

Hint 39.2
Impose the boundary conditions

u(0, t) = u(2π, t), uθ (0, t) = uθ (2π, t).

Hint 39.3
Apply the separation of variables u(x, y) = X(x)Y (y). Solve an eigenvalue problem for X(x).

Hint 39.4

Hint 39.5

Hint 39.6

Hint 39.7
There are two ways to solve the problem. For the first method, expand the solution in a series of the form
∞
X  nπx 
u(x, t) = an (t) sin .
n=1
L

1608
Because of the inhomogeneous boundary conditions, the convergence of the series will not be uniform. You can
differentiate the series with respect to t, but not with respect to x. Multiply the partial differential equation by
the eigenfunction sin(nπx/L) and integrate from x = 0 to x = L. Use integration by parts to move derivatives in
x from u to the eigenfunctions. This process will yield a first order, ordinary differential equation for each of the
an ’s.
For the second method: Make the change of variables v(x, t) = u(x, t) − µ(x), where µ(x) is the equilibrium
temperature distribution to obtain a problem with homogeneous boundary conditions.

Hint 39.8

Hint 39.9

Hint 39.10

Hint 39.11

Hint 39.12

Hint 39.13
Use separation of variables to find eigen-solutions of the partial differential equation that satisfy the homogeneous
boundary conditions. There will be two eigen-solutions for each eigenvalue. Expand u(x, t) in a series of the

1609
eigen-solutions. Use the two initial conditions to determine the constants.

Hint 39.14
Expand the solution in a series of eigenfunctions in x. Determine these eigenfunctions by using separation of
variables on the homogeneous partial differential equation. You will find that the answer has the form,

∞
X  nπx 
u(x, y) = un (y) sin .
n=1
a

Substitute this series into the partial differential equation to determine ordinary differential equations for each of
the un ’s. The boundary conditions on u(x, y) will give you boundary conditions for the un ’s. Solve these ordinary
differential equations with Green functions.

Hint 39.15
Solve this problem by expanding the solution in a series of eigen-solutions that satisfy the partial differential
equation and the homogeneous boundary conditions. Use the initial conditions to determine the coefficients in
the expansion.

Hint 39.16
Use separation of variables to find eigen-solutions that satisfy the partial differential equation and the homogeneous
boundary conditions. The solution is a linear combination of the eigen-solutions. The whole solution will be
exponentially decaying if each of the eigen-solutions is exponentially decaying.

Hint 39.17
For parts (a), (b) and (c) use separation of variables. For part (b) the eigen-solutions will involve Bessel functions.
For part (c) the eigen-solutions will involve spherical Bessel functions. Part (d) is trivial.

1610
Hint 39.18
The solution is a linear combination of eigen-solutions of the partial differential equation that satisfy the homo-
geneous boundary conditions. Determine the coefficients in the expansion with the initial condition.

Hint 39.19
The problem is
1 1
urr + ur + 2 uθθ = 0, 0 < r < 1, 0 < θ < π
r r
u(r, 0) = u(r, π) = 0, u(0, θ) = 0, u(1, θ) = 1
The solution is a linear combination of eigen-solutions that satisfy the partial differential equation and the three
homogeneous boundary conditions.

Hint 39.20

Hint 39.21

Hint 39.22

Hint 39.23

Hint 39.24

1611
Hint 39.25

Hint 39.26

Hint 39.27

Hint 39.28

Hint 39.29

Hint 39.30

Hint 39.31

Hint 39.32

1612
39.10 Solutions
Solution 39.1
We expand the solution in a Fourier series.
∞ ∞
1 X X
φ = a0 (r) + an (r) cos(nθ) + bn (r) sin(nθ)
2 n=1 n=1

We substitute the series into the Laplace’s equation to determine ordinary differential equations for the coefficients.
1 ∂2φ
 
∂ ∂φ
r + 2 2 =0
∂r ∂r r ∂θ
1 1 1
a000 + a00 = 0, a00n + a0n − n2 an = 0, b00n + b0n − n2 bn = 0
r r r
The solutions that are bounded at r = 0 are, (to within multiplicative constants),

a0 (r) = 1, an (r) = rn , bn (r) = rn .

Thus φ(r, θ) has the form

∞ ∞
1 X
n
X
φ(r, θ) = c0 + cn r cos(nθ) + dn rn sin(nθ)
2 n=1 n=1

We apply the boundary condition at r = R.

∞ ∞
1 X X
φ(R, θ) = c0 + cn Rn cos(nθ) + dn Rn sin(nθ)
2 n=1 n=1

The coefficients are

1 2π 2π 2π
Z Z Z
1 1
c0 = φ(R, α) dα, cn = φ(R, α) cos(nα) dα, dn = φ(R, α) sin(nα) dα.
π 0 πRn 0 πRn 0

1613
We substitute the coefficients into our series solution.
Z 2π ∞
1 X  r n 2π
Z
1
φ(r, θ) = φ(R, α) dα + φ(R, α) cos(n(θ − α)) dα
2π 0 π n=1 R 0
∞  
Z 2π !
1 2π
Z
1 X r n in(θ−α)
φ(r, θ) = φ(R, α) dα + φ(R, α)< e dα
2π 0 π 0 n=1
R
!
Z 2π Z 2π r i(θ−α)
1 1 R
e
φ(r, θ) = φ(R, α) dα + φ(R, α)< dα
2π 0 π 0 1 − Rr ei(θ−α)
r 2
!
r i(θ−α)
Z 2π

1 2π e −
Z
1 R R
φ(r, θ) = φ(R, α) dα + φ(R, α)<
2 dα
2π 0 π 0 1 − 2 R cos(θ − α) + Rr
r
Z 2π
1 2π Rr cos(θ − α) − r2
Z
1
φ(r, θ) = φ(R, α) dα + φ(R, α) 2 dα
2π 0 π 0 R + r2 − 2Rr cos(θ − α)
Z 2π
1 R2 − r 2
φ(r, θ) = φ(R, α) 2 dα
2π 0 R + r2 − 2Rr cos(θ − α)

Solution 39.2
In order that the solution is continuously differentiable, (which it must be in order to satisfy the differential
equation), we impose the boundary conditions
u(0, t) = u(2π, t), uθ (0, t) = uθ (2π, t).
We apply the separation of variables u(θ, t) = Θ(θ)T (t).
ut = κuθθ
ΘT 0 = κΘ00 T
T0 Θ00
= = −λ
κT Θ

1614
We have the self-adjoint eigenvalue problem

Θ00 + λΘ = 0, Θ(0) = Θ(2π), Θ0 (0) = Θ0 (2π)

which has the eigenvalues and orthonormal eigenfunctions

1
λn = n 2 , Θn = √ einθ , n ∈ Z.
2π
Now we solve the problems for Tn (t) to obtain eigen-solutions of the heat equation.
Tn0 = −n2 κTn
2 κt
Tn = e−n
The solution is a linear combination of the eigen-solutions.
∞
X 1 2
u(θ, t) = un √ einθ e−n κt
n=−∞
2π

We use the initial conditions to determine the coefficients.

∞
X 1
u(θ, 0) = un √ einθ = f (θ)
n=−∞
2π
Z 2π
1
un = √ e−inθ f (θ) dθ
2π 0

Solution 39.3
Substituting u(x, y) = X(x)Y (y) into the partial differential equation yields
X 00 Y 00
=− = −λ.
X Y

1615
With the homogeneous boundary conditions, we have the two problems

X 00 + λX = 0, X(0) = X(1) = 0,

Y 00 − λY = 0, Y (1) = 0.

The eigenvalues and orthonormal eigenfunctions for X(x) are

√
λn = (nπ)2 , Xn = 2 sin(nπx).

The general solution for Y is

Yn = a cosh(nπy) + b sinh(nπy).

The solution for that satisfies the right homogeneous boundary condition, (up to a multiplicative constant), is

Yn = sinh(nπ(1 − y))

u(x, y) is a linear combination of the eigen-solutions.

∞
X √
u(x, y) = un 2 sin(nπx) sinh(nπ(1 − y))
n=1

We use the inhomogeneous boundary condition to determine coefficients.

∞
X √
u(x, 0) = un 2 sin(nπx) sinh(nπ) = f (x)
n=1

√ Z 1
un = 2 sin(nπξ)f (ξ) dξ
0

1616
Solution 39.4
Substituting u(r, θ) = R(r)Θ(θ) yields

1 1
R00 Θ + R0 Θ + 2 RΘ00 = 0
r r

R00 R0 Θ00
r2 +r =− =λ
R R Θ

r2 R00 + rR0 − λR = 0, Θ00 + λΘ = 0

We assume that u is a strong solution of the partial differential equation and is thus twice continuously differ-
entiable, (u ∈ C 2 ). In particular, this implies that R and Θ are bounded and that Θ is continuous and has a
continuous first derivative along θ = 0. This gives us the problems

Θ00 + λΘ = 0, Θ(0) = Θ(2π), Θ0 (0) = Θ0 (2π),

r2 R00 + rR0 − λR = 0, R is bounded

We consider negative, zero and positive values of λ in solving the equation for Θ.

λ < 0. The general solution for Θ is

√ √
Θ = a cosh( −λθ) + b sinh( −λθ).

Θ(0) = Θ(2π) ⇒ a = 0, Θ0 (0) = Θ0 (2π) ⇒ b = 0.

There are no negative eigenvalues.

1617
 λ = 0. The general solution for Θ is
Θ = a + bθ.

Θ(0) = Θ(2π) ⇒ b = 0, Θ0 (0) = Θ0 (2π) ⇒ a is arbitrary.

We have the eigenvalue and eigenfunction
1
λ0 = 0, Θ0 = .
2
λ > 0. The general solution is
√ √
Θ = a cos( λθ) + b sin( λθ).
Applying the boundary conditions to find the eigenvalues and eigenfunctions,
√ √
Θ(0) = Θ(2π) ⇒ a = a cos( λ2π) + b sin( λ2π)
√ √
⇒ cos( λ2π) = 1, sin( λ2π) = 0
√
⇒ λ = n, for n = 1, 2, 3, . . .
⇒ λ = n2 , for n = 1, 2, 3, . . .
The boundary condition, Θ0 (0) = Θ0 (2π) is satisfied for these values of λ. This gives us the eigenvalues and
eigenfunctions
λn = n2 , Θ(1)
n = cos(nθ), Θ(2)
n = sin(nθ), forn = 1, 2, 3, . . .

Now to find the bounded solutions of the equation for R. Substituting R = rα yields
α(α − 1) + α − λ = 0
√
α = ± λ.
There are two cases to consider.

1618
 λ0 = 0.

R = a + b log r

Boundedness demands that b = 0. Thus we have the solution

R=1

λn = n 2 > 0

R = arn + br−n

Boundedness demands that b = 0. Thus we have the solution

R = rn .

The general solution for u is

∞
a0 X
u(r, θ) = + [an cos(nθ) + bn sin(nθ)] rn .
2 n=1

The inhomogeneous boundary condition will determine the coefficients.

∞
a0 X
u(1, θ) = + [an cos(nθ) + bn sin(nθ)] = f (θ)
2 n=1

The coefficients are the Fourier coefficients of f (θ).

1
R 2π
an = π
f (θ) cos(nθ) dθ
1
R02π
bn = π 0
f (θ) sin(nθ) dθ

1619
Solution 39.5
A normal mode of frequency ω satisfies

v(r, θ, t) = u(r, θ) eiωt .

Substituting this into the partial differential equation and the boundary condition yields
1 ∂2u ω2
 
1 ∂ ∂u
r + 2 2 = − 2 u, u(1, θ) = 0,
r ∂r ∂r r ∂θ c
∂ 2 u 1 ∂u 1 ∂2u
+ + + k 2 u = 0, u(1, θ) = 0,
∂r2 r ∂r r2 ∂θ2
where k = ωc . Applying separation of variables to the partial differential equation for u, with u = R(r)Θ(θ),
r2 R00 Θ + rR0 Θ + RΘ00 + k 2 r2 RΘ = 0,
R00 R0 Θ00
r2 + r + k2 r2 = − = λ2 .
R R Θ
Now we have the two ordinary differential equations,
λ2
 
00 1 0 2
R + R + k − 2 R = 0, R(0) is bounded, R(1) = 0,
r r
Θ + λ Θ = 0, Θ(−π) = Θ(π), Θ0 (−π) = Θ0 (π).
00 2

The eigenvalues and eigenfunctions for Θ are

λn = n, n = 0, 1, 2, . . . ,

1
Θ0 = , Θ(1) n = cos(nθ), Θ(2)
n = sin(nθ), n = 1, 2, 3, . . .
2
The differential equation for R is then
n2
 
00 1 0 2
R + R + k − 2 R = 0, R(0) is bounded, R(1) = 0.
r r

1620
The general solution of the differential equation is a linear combination of Bessel functions of order n.

R(r) = c1 Jn (kr) + c2 Yn (kr)

Since Yn (kr) is unbounded at r = 0, the solution has the form

R(r) = cJn (kr).

Applying the second boundary condition yields

Jn (k) = 0.

Thus the eigenvalues and eigenfunctions for R are

knm = jnm , Rnm = Jn (jnm r),

where jnm is the mth positive root of Jn . Combining the above results, the normal modes of oscillation are

1
v0m = J0 (j0m r) eiωt , m = 1, 2, 3, . . . ,
2
vnm = cos(nθ + α)Jnm (jnm r) eiωt , n, m = 1, 2, 3, . . . .

u22 and u33 are plotted in Figure 39.2.

Solution 39.6
We will expand the solution in a complete, orthogonal set of functions {Xn (x)}, where the coefficients are functions
of t.
X
φ= Tn (t)Xn (x)
n

1621
 Figure 39.2: The Normal Modes u22 and u33

We will use separation of variables to determine a convenient set {Xn }. We substitite φ = T (t)X(x) into the
diffusion equation.

φt = a2 φxx
XT 0 = a2 X 00 T
T0 X 00
= = −λ
a2 T X
T 0 = −a2 λT, X 00 + λX = 0

Note that in order to satisfy φ(0, t) = φ(l, t) = 0, the Xn must satisfy the same homogeneous boundary conditions,

1622
Xn (0) = Xn (l) = 0. This gives us a Sturm-Liouville problem for X(x).

X 00 + λX = 0, X(0) = X(l) = 0
 nπ 2  nπx 
λn = , Xn = sin , n ∈ Z+
l l

Thus we seek a solution of the form

∞
X  nπx 
φ= Tn (t) sin . (39.8)
n=1
l

This solution automatically satisfies the boundary conditions. We will assume that we can differentiate it. We
will substitite this form into the diffusion equation and the initial condition to determine the coefficients in the
series, Tn (t). First we substitute Equation 39.8 into the partial differential equation for φ to determine ordinary
differential equations for the Tn .

φt = a2 φxx
∞  nπx  ∞ 
X
0 2
X nπ 2  nπx 
Tn (t) sin = −a Tn (t) sin
n=1
l n=1
l l
 anπ 2
Tn0 = − Tn
l

1623
Now we substitute Equation 39.8 into the initial condition for φ to determine initial conditions for the Tn .
X∞  nπx 
Tn (0) sin = φ(x, 0)
n=1
l
Rl
sin nπx


0 l
φ(x, 0) dx
Tn (0) = R l 2 nπx

0
sin l
dx
Z l
2  nπx 
Tn (0) = sin φ(x, 0) dx
l 0 l
2 l/2 2 l/2
Z  nπx  Z  nπx 
Tn (0) = sin x dx + sin (l − x) dx
l 0 l l 0 l
4l  nπ 
Tn (0) = 2 2 sin
nπ 2
4l
T2n−1 (0) = (−1)n , T2n (0) = 0, n ∈ Z+
(2n − 1)2 π 2
We solve the ordinary differential equations for Tn subject to the initial conditions.
 2 !
4l a(2n − 1)π
T2n−1 (t) = (−1)n exp − t , T2n (t) = 0, n ∈ Z+
(2n − 1)2 π 2 l
This determines the series representation of the solution.
∞  2  2 !  
4X l a(2n − 1)π (2n − 1)πx
φ= (−1)n exp − t sin
l n=1 (2n − 1)π l l

From the initial condition, we know that the the solution at t = 0 is C 0 . That is, it is continuous, but not
differentiable. The series representation of the solution at t = 0 is
∞  2  
4X n l (2n − 1)πx
φ= (−1) sin .
l n=1 (2n − 1)π l

1624
That the coefficients decay as 1/n2 corroborates that φ(x, 0) is C 0 .
The derivatives of φ with respect to x are
∞ 2m−3 2 !
∂ 2m−1 4(−1)m+1 X
   
n (2n − 1)π a(2n − 1)π (2n − 1)πx
φ= (−1) exp − t cos
∂x2m−1 l n=1
l l l
∞ 2m−2 2 !
∂ 2m 4(−1)m X
   
n (2n − 1)π a(2n − 1)π (2n − 1)πx
φ= (−1) exp − t sin
∂x2m l n=1
l l l

∂n
For any fixed t > 0, the coefficients in the series for ∂x
φ decay exponentially. These series are uniformly convergent
in x. Thus for any fixed t > 0, φ is C ∞ in x.

Solution 39.7

ut = κuxx , 0 < x < L, t > 0

u(0, t) = T0 , u(L, t) = T1 , u(x, 0) = f (x),

Method 1. We solve this problem with an eigenfunction expansion in x. To find an appropriate set of eigen-
functions, we apply the separation of variables, u(x, t) = X(x)T (t) to the partial differential equation with the
homogeneous boundary conditions, u(0, t) = u(L, t) = 0.

(XT )t = (XT )xx

XT 0 = X 00 T
T0 X 00
= = −λ2
T X
We have the eigenvalue problem,

X 00 + λ2 X = 0, X(0) = X(L) = 0,

1625
which has the solutions,
nπx  nπx 
λn = , Xn = sin , n ∈ N.
L L
We expand the solution of the partial differential equation in terms of these eigenfunctions.

∞
X  nπx 
u(x, t) = an (t) sin
n=1
L

Because of the inhomogeneous boundary conditions, the convergence of the series will not be uniform. We can
differentiate the series with respect to t, but not with respect to x. We multiply the partial differential equation
by an eigenfunction and integrate from x = 0 to x = L. We use integration by parts to move derivatives from u
to the eigenfunction.

ut − κuxx = 0
Z L  mπx 
(ut − κuxx ) sin dx = 0
0 L
∞
Z L X !
mπ L
 nπx   mπx  h  mπx iL Z  mπx 
0
an (t) sin sin dx − κ ux sin +κ ux cos dx = 0
0 n=1
L L L 0 L 0 L
L 0 mπ h  mπx iL  mπ 2 Z L  mπx 
am (t) + κ u cos +κ u sin dx = 0
2 L L 0 L 0 L
∞
!
L 0 mπ  mπ 2 Z L X  nπx   mπx 
am (t) + κ ((−1)m u(L, t) − u(0, t)) + κ an (t) sin sin dx = 0
2 L L 0 n=1
L L
L 0 mπ L  mπ 2
am (t) + κ ((−1)m T1 − T0 ) + κ am (t) = 0
2 L 2 L
 mπ 2 2mπ
a0m (t) + κ am (t) = κ 2 (T0 − (−1)m T1 )
L L

1626
Now we have a first order differential equation for each of the an ’s. We obtain initial conditions for each of the
an ’s from the initial condition for u(x, t).

u(x, 0) = f (x)
∞
X  nπx 
an (0) sin = f (x)
n=1
L
2 L
Z  nπx 
an (0) = f (x) sin dx ≡ fn
L 0 L
By solving the first order differential equation for an (t), we obtain

2(T0 − (−1)n T1 ) 2(T0 − (−1)n T1 )

 
−κ(nπ/L)2 t
an (t) = +e fn − .
nπ nπ

Note that the series does not converge uniformly due to the 1/n term.
Method 2. For our second method we transform the problem to one with homogeneous boundary conditions
so that we can use the partial differential equation to determine the time dependence of the eigen-solutions. We
make the change of variables v(x, t) = u(x, t) − µ(x) where µ(x) is some function that satisfies the inhomogeneous
boundary conditions. If possible, we want µ(x) to satisfy the partial differential equation as well. For this problem
we can choose µ(x) to be the equilibrium solution which satisfies

µ00 (x) = 0, µ(0)T0 , µ(L) = T1 .

This has the solution

T1 − T0
µ(x) = T0 + x.
L
With the change of variables,
 
T1 − T0
v(x, t) = u(x, t) − T0 + x ,
L

1627
we obtain the problem

vt = κvxx , 0 < x < L, t>0

 
T1 − T0
v(0, t) = 0, v(L, t) = 0, v(x, 0) = f (x) − T0 + x .
L
Now we substitute the separation of variables v(x, t) = X(x)T (t) into the partial differential equation.

(XT )t = κ(XT )xx

T0 X 00
= = −λ2
κT X
Utilizing the boundary conditions at x = 0, L we obtain the two ordinary differential equations,

T 0 = −κλ2 T,
X 00 = −λ2 X, X(0) = X(L) = 0.

The problem for X is a regular Sturm-Liouville problem and has the solutions
nπ  nπx 
λn = , Xn = sin , n ∈ N.
L L
The ordinary differential equation for T becomes,
 nπ 2
Tn0 = −κ Tn ,
L
which, (up to a multiplicative constant), has the solution,
2
Tn = e−κ(npi/L) t .

Thus the eigenvalues and eigen-solutions of the partial differential equation are,
nπ  nπx  2
λn = , vn = sin e−κ(nπ/L) t , n ∈ N.
L L

1628
Let v(x, t) have the series expansion,
∞  nπx  2
X
v(x, t) = an sin e−κ(nπ/L) t .
n=1
L

We determine the coefficients in the expansion from the initial condition,

∞  
X  nπx  T1 − T0
v(x, 0) = an sin = f (x) − T0 + x .
n=1
L L

The coefficients in the expansion are the Fourier sine coefficients of f (x) − T0 + T1 −T


L
0
x .

2 L
Z   
T1 − T0  nπx 
an = f (x) − T0 + x sin dx
L 0 L L
2(T0 − (−1)n T1 )
an = fn −
nπ
With the coefficients defined above, the solution for u(x, t) is
∞ 
2(T0 − (−1)n T1 )

T1 − T0 X  nπx  2
u(x, t) = T0 + x+ fn − sin e−κ(nπ/L) t .
L n=1
nπ L

Since the coefficients in the sum decay exponentially for t > 0, we see that the series is uniformly convergent for
positive t. It is clear that the two solutions we have obtained are equivalent.

Solution 39.8
First we solve the eigenvalue problem for β(x), which is the problem we would obtain if we applied separation of
variables to the partial differential equation, φt = φxx . We have the eigenvalues and orthonormal eigenfunctions
 2 r  
(2n − 1)π 2 (2n − 1)πx
λn = , βn (x) = sin , n ∈ Z+ .
2l l 2l

1629
We expand the solution and inhomogeneity in Equation 39.5 in a series of the eigenvalues.
∞
X
φ(x, t) = Tn (t)βn (x)
n=1
∞
X Z l
w(x, t) = wn (t)βn (x), wn (t) = βn (x)w(x, t) dx
n=1 0

Since φ satisfies the same homgeneous boundary conditions as β, we substitute the series into Equation 39.5 to
determine differential equations for the Tn (t).
∞
X ∞
X ∞
X
Tn0 (t)βn (x) =a 2
Tn (t)(−λn )βn (x) + wn (t)βn (x)
n=1 n=1 n=1
 2
(2n − 1)π
Tn0 (t) = −a 2
Tn (t) + wn (t)
2l

Now we substitute the series for φ into its initial condition to determine initial conditions for the Tn .
∞
X
φ(x, 0) = Tn (0)βn (x) = f (x)
n=1
Z l
Tn (0) = βn (x)f (x) dx
0

We solve for Tn (t) to determine the solution, φ(x, t).

 2 ! Z t  2 ! !
(2n − 1)aπ (2n − 1)aπ
Tn (t) = exp − t Tn (0) + wn (τ ) exp τ dτ
2l 0 2l

1630
Solution 39.9
Separation of variables leads to the eigenvalue problem
β 00 + λβ = 0, β(0) = 0, β(l) + cβ 0 (l) = 0.
First we consider the case λ = 0. A set of solutions of the differential equation is {1, x}. The solution that satisfies
the left boundary condition is β(x) = x. The right boundary condition imposes the constraint l + c = 0. Since c
is positive, this has no solutions. λ = 0 is not an eigenvalue. √ √
Now we consider λ 6= 0. A set of solutions of the√differential equation is {cos( λx), sin( λx)}. The solution
that satisfies the left boundary condition is β = sin( λx). The right boundary condition imposes the constraint
√  √ √ 
c sin λl + λ cos λl = 0
√  √
λ
tan λl = −
c
For large λ, the we can determine approximate solutions.
p (2n − 1)π
λn l ≈ , n ∈ Z+
2
 2
(2n − 1)π
λn ≈ , n ∈ Z+
2l
The eigenfunctions are
√

sin
λn x
βn (x) = qR √ , n ∈ Z+ .
l
sin2 λn x dx


0

We expand φ(x, t) and w(x, t) in series of the eigenfunctions.

∞
X
φ(x, t) = Tn (t)βn (x)
n=1
∞
X Z l
w(x, t) = wn (t)βn (x), wn (t) = βn (x)w(x, t) dx
n=1 0

1631
Since φ satisfies the same homgeneous boundary conditions as β, we substitute the series into Equation 39.5 to
determine differential equations for the Tn (t).
∞
X ∞
X ∞
X
Tn0 (t)βn (x) = a2 Tn (t)(−λn )βn (x) + wn (t)βn (x)
n=1 n=1 n=1
Tn0 (t) 2
= −a λn Tn (t) + wn (t)

Now we substitute the series for φ into its initial condition to determine initial conditions for the Tn .
∞
X
φ(x, 0) = Tn (0)βn (x) = f (x)
n=1
Z l
Tn (0) = βn (x)f (x) dx
0

We solve for Tn (t) to determine the solution, φ(x, t).

 Z t 
2 2



Tn (t) = exp −a λn t Tn (0) + wn (τ ) exp a λn τ dτ
0

Solution 39.10
First we seek a function u(x, t) that satisfies the boundary conditions u(0, t) = t, ux (l, t) = −cu(l, t). We try a
function of the form u = (ax + b)t. The left boundary condition imposes the constraint b = 1. We then apply the
right boundary condition no determine u.

at = −c(al + 1)t
c
a=−
 1 + cl 
cx
u(x, t) = 1 − t
1 + cl

1632
 Now we define ψ to be the difference of φ and u.

ψ(x, t) = φ(x, t) − u(x, t)

ψ satisfies an inhomogeneous diffusion equation with homogeneous boundary conditions.

(ψ + u)t = a2 (ψ + u)xx + 1
ψt = a2 ψxx + 1 + a2 uxx − ut
cx
ψt = a2 ψxx +
1 + cl
The initial and boundary conditions for ψ are

ψ(x, 0) = 0, ψ(0, t) = 0, ψx (l, t) = −cψ(l, t).

We solved this system in problem 2. Just take

cx
w(x, t) = , f (x) = 0.
1 + cl
The solution is
∞
X
ψ(x, t) = Tn (t)βn (x),
n=1
Z t
wn exp −a2 λn (t − τ ) dτ,


Tn (t) =
0
Z l
cx
wn (t) = βn (x) dx.
0 1 + cl

This determines the solution for φ.

1633
Solution 39.11
First we solve this problem with a series expansion. We transform the problem to one with homogeneous boundary
conditions. Note that

x
u(x) =
l+1

satisfies the boundary conditions. (It is the equilibrium solution.) We make the change of variables ψ = φ − u.
The problem for ψ is

ψt = a2 ψxx ,
x
ψ(0, t) = ψ(l, t) + ψx (l, t) = 0, ψ(x, 0) = .
l+1

This is a particular case of what we solved in Exercise 39.9. We apply the result of that problem. The solution
for φ(x, t) is

∞
x X
φ(x, t) = + Tn (t)βn (x)
l + 1 n=1
√

sin λn x
βn (x) = qR √ , n ∈ Z+
l
sin2 λn x dx


0
√  √
tan λl = − λ
Tn (t) = Tn (0) exp −a2 λn t


Z l
x
Tn (0) = βn (x) dx
0 l+1

This expansion is useful for large t because the coefficients decay exponentially with increasing t.

1634
 Now we solve this problem with the Laplace transform.

φt = a2 φxx , φ(0, t) = 0, φ(l, t) + φx (l, t) = 1, φ(x, 0) = 0

1
sφ̂ = a2 φ̂xx , φ̂(0, s) = 0, φ̂(l, s) + φ̂x (l, s) =
s
s 1
φ̂xx − 2 φ̂ = 0, φ̂(0, s) = 0, φ̂(l, s) + φ̂x (l, s) =
a s

The solution that satisfies the left boundary condition is

√ 
sx
φ̂ = c sinh .
a

We apply the right boundary condition to determine the constant.

√ 
sinh asx
φ̂ =  √  √  √ 
s sinh asl + as cosh asl

1635
We expand this in a series of simpler functions of s.
√ 
2 sinh asx
φ̂ =  √   √  √  √   √ 
s exp asl − exp − asl + as exp asl + exp − asl
√ 
2 sinh asx 1
φ̂ = √  √  √   √ 
s exp asl 1 + as − 1 − as exp − 2 asl
√   √ 
exp asx − exp − asx 1
φ̂ =  √  √   √   √ 
s sl 1− s/a
s 1 + a exp a 1 − 1+ s/a exp − 2 asl
√
√  √ 
exp s(x−l) − exp s(−x−l) ∞  √ n  √ 
a a X 1 − s/a 2 sln
φ̂ =  √  √ exp −
s 1+ s n=0
1 + s/a a
a

∞ √  √
(1 − s/a)n

1 X s((2n + 1)l − x)
φ̂ = √ exp −
s n=0
(1 + s/a)n+1 a
∞ √  √ !
X (1 − s/a)n s((2n + 1)l + x)
− √ exp −
n=0
(1 + s/a)n+1 a

By expanding
√
(1 − s/a)n
√
(1 + s/a)n+1
in binomial series all the terms would be of the form
 √ 
−m/2−3/2 s((2n ± 1)l ∓ x)
s exp − .
a

1636
Taking the first term in each series yields
  √   √ 
a s(l − x) s(l + x)
φ̂ ∼ 3/2 exp − − exp − , as s → ∞.
s a a
We take the inverse Laplace transform to obtain an appoximation of the solution for t  1.
   
(l−x)2 (l+x)2

√ exp − 4a2 t
exp − 4a2 t
φ(x, t) ∼ 2a2 πt  − 
l−x l+x
    
l−x l+x
− π erfc √ − erfc √ , for t  1
2a t 2a t

Solution 39.12
We apply the separation of variables φ(x, t) = X(x)T (t).
φt = A2 x2 φx


x
0 2 2 0 0


XT = T A x X
T0 (x2 X 0 )0
= = −λ
A2 T X
This gives us a regular Sturm-Liouville problem.

0
x2 X 0 + λX = 0, X(1) = X(2) = 0
This is an Euler equation. We make the substitution X = xα to find the solutions.
x2 X 00 + 2xX 0 + λX = 0 (39.9)
α(α − 1) + 2α + λ = 0
√
−1 ± 1 − 4λ
α=
2
1 p
α = − ± i λ − 1/4
2

1637
 First we consider the case of a double root when λ = 1/4. The solutions of Equation 39.9 are {x−1/2 , x−1/2 ln x}.
The solution that satisfies the left boundary condition is X = x−1/2 ln x. Since this does not satisfy the right
boundary condition, λ = 1/4 is not an eigenvalue.
Now we consider λ 6= 1/4. The solutions of Equation 39.9 are
 
1 p  1 p
√ cos λ − 1/4 ln x , √ sin λ − 1/4 ln x .
x x
The solution that satisfies the left boundary condition is
1 p 
√ sin λ − 1/4 ln x .
x
The right boundary condition imposes the constraint
p
λ − 1/4 ln 2 = nπ, n ∈ Z+ .
This gives us the eigenvalues and eigenfunctions.
 
1  nπ 2 1 nπ ln x
λn = + , Xn (x) = √ sin , n ∈ Z+ .
4 ln 2 x ln 2
We normalize the eigenfunctions.
Z 2   Z 1
1 nπ ln x ln 2
sin2 dx = ln 2 sin2 (nπξ) dξ =
1 x ln 2 0 2
r  
2 1 nπ ln x
Xn (x) = √ sin , n ∈ Z+ .
ln 2 x ln 2
From separation of variables, we have differential equations for the Tn .
 
0 2 1  nπ 2
Tn = −A + Tn
4 ln 2
   
2 1  nπ 2
Tn (t) = exp −A + t
4 ln 2

1638
We expand φ in a series of the eigensolutions.
∞
X
φ(x, t) = φn Xn (x)Tn (t)
n=1

We substitute the expansion for φ into the initial condition to determine the coefficients.
∞
X
φ(x, 0) = φn Xn (x) = f (x)
n=1
Z 2
φn = Xn (x)f (x) dx
1

Solution 39.13

utt = c2 uxx , 0 < x < L, t > 0,

u(0, t) = 0, ux (L, t) = 0,
u(x, 0) = f (x), ut (x, 0) = 0,

We substitute the separation of variables u(x, t) = X(x)T (t) into the partial differential equation.

(XT )tt = c2 (XT )xx

T 00 X 00
2
= = −λ2
cT X
With the boundary conditions at x = 0, L, we have the ordinary differential equations,

T 00 = −c2 λ2 T,
X 00 = −λ2 X, X(0) = X 0 (L) = 0.

1639
The problem for X is a regular Sturm-Liouville eigenvalue problem. From the Rayleigh quotient,
RL RL 0 2
2 − [φφ0 ]L0 + 0 (φ0 )2 dx (φ ) dx
λ = RL = R0 L
0
φ2 dx 0
φ2 dx
we see that there are only positive eigenvalues. For λ2 > 0 the general solution of the ordinary differential equation
is
X = a1 cos(λx) + a2 sin(λx).
The solution that satisfies the left boundary condition is
X = a sin(λx).
For non-trivial solutions, the right boundary condition imposes the constraint,
cos (λL) = 0,
 
π 1
λ= n− , n ∈ N.
L 2
The eigenvalues and eigenfunctions are
 
(2n − 1)π (2n − 1)πx
λn = , Xn = sin , n ∈ N.
2L 2L
The differential equation for T becomes
 2
00 2 (2n − 1)π
T = −c T,
2L
which has the two linearly independent solutions,
   
(1) (2n − 1)cπt (2n − 1)cπt
Tn = cos , Tn(2) = sin .
2L 2L

1640
The eigenvalues and eigen-solutions of the partial differential equation are,
(2n − 1)π
λn = , n ∈ N,
2L
       
(2n − 1)πx (2n − 1)cπt (2n − 1)πx (2n − 1)cπt
u(1)
n = sin cos (2)
, un = sin sin .
2L 2L 2L 2L
We expand u(x, t) in a series of the eigen-solutions.
∞      
X (2n − 1)πx (2n − 1)cπt (2n − 1)cπt
u(x, t) = sin an cos + bn sin .
n=1
2L 2L 2L

We impose the initial condition ut (x, 0) = 0,

∞  
X (2n − 1)cπ (2n − 1)πx
ut (x, 0) = bn sin = 0,
n=1
2L 2L

bn = 0.
The initial condition u(x, 0) = f (x) allows us to determine the remaining coefficients,
∞  
X (2n − 1)πx
u(x, 0) = an sin = f (x),
n=1
2L

L  
(2n − 1)πx
Z
2
an = f (x) sin dx.
L 0 2L
The series solution for u(x, t) is,
∞    
X (2n − 1)πx (2n − 1)cπt
u(x, t) = an sin cos .
n=1
2L 2L

1641
Solution 39.14

uxx + uyy = f (x, y), 0 < x < a, 0 < y < b,

u(0, y) = u(a, y) = 0, uy (x, 0) = uy (x, b) = 0,
We will solve this problem with an eigenfunction expansion in x. To determine a suitable set of eigenfunctions,
we substitute the separation of variables u(x, y) = X(x)Y (y) into the homogeneous partial differential equation.
uxx + uyy = 0
(XY )xx + (XY )yy = 0
X 00 Y 00
=− = −λ2
X Y
With the boundary conditions at x = 0, a, we have the regular Sturm-Liouville problem,
X 00 = −λ2 X, X(0) = X(a) = 0,
which has the solutions,
nπ  nπx 
λn = , Xn = sin , n ∈ N.
a a
We expand u(x, y) in a series of the eigenfunctions,
∞
X  nπx 
u(x, y) = un (y) sin .
n=1
a

Substituting this series into the partial differential equation and boundary conditions at y = 0, b, we obtain,
∞    nπx 
X nπ 2  nπx 
00
− un (y) sin + un (y) sin = f (x),
n=1
a a a
X∞  nπx  X ∞  nπx 
u0n (0) sin = u0n (b) sin = 0.
n=1
a n=1
a

1642
Expanding f (x, y) in the Fourier sine series,
∞
X  nπx 
f (x, y) = fn (y) sin ,
n=1
a

Z a
2  nπx 
fn (y) = f (x, y) sin dx,
a 0 a

we obtain the ordinary differential equations,

 nπ 2
00
un (y) − un (y) = fn (y), u0n (0) = u0n (b) = 0, n ∈ N.
a
We will solve these ordinary differential equations with Green functions. Consider the Green function problem,
 nπ 2
00
gn (y; η) − gn (y; η) = δ(y − η), gn0 (0; η) = gn0 (b; η) = 0.
a
The homogeneous solutions
 
 nπy  nπ(y − b)
cosh and cosh
a a
satisfy the left and right boundary conditions, respectively. The Wronskian of these two solutions is
 
 cosh(nπy/a) cosh(nπ(y − b)/a) 
W (y) =  nπ nπ

a
sinh(nπy/a) a
sinh(nπ(y − b)/a)
    
nπ  nπy  nπ(y − b)  nπy  nπ(y − b)
= cosh sinh − sinh cosh
a a a a a
 
nπ nπb
=− sinh .
a a

1643
Thus the Green function is

a cosh(nπy< /a) cosh(nπ(y> − b)/a)

gn (y; η) = − .
nπ sinh(nπb/a)

The solutions for the coefficients in the expansion are

Z b
un (y) = gn (y; η)fn (η) dη.
0

Solution 39.15

utt + a2 uxxxx = 0, 0 < x < L, t > 0,

u(x, 0) = f (x), ut (x, 0) = g(x),
u(0, t) = uxx (0, t) = 0, u(L, t) = uxx (L, t) = 0,

We will solve this problem by expanding the solution in a series of eigen-solutions that satisfy the partial differential
equation and the homogeneous boundary conditions. We will use the initial conditions to determine the coefficients
in the expansion. We substitute the separation of variables, u(x, t) = X(x)T (t) into the partial differential
equation.

(XT )tt + a2 (XT )xxxx = 0

T 00 X 0000
= − = −λ4
a2 T X
Here we make the assumption that 0 ≤ arg(λ) < π/2, i.e., λ lies in the first quadrant of the complex plane. Note
that λ4 covers the entire complex plane. We have the ordinary differential equation,

T 00 = −a2 λ4 T,

1644
and with the boundary conditions at x = 0, L, the eigenvalue problem,

X 0000 = λ4 X, X(0) = X 00 (0) = X(L) = X 00 (L) = 0.

For λ = 0, the general solution of the differential equation is

X = c1 + c2 x + c3 x2 + c4 x3 .

Only the trivial solution satisfies the boundary conditions. λ = 0 is not an eigenvalue. For λ 6= 0, a set of linearly
independent solutions is

{ eλx , eiλx , e−λx , e−iλx }.

Another linearly independent set, (which will be more useful for this problem), is

{cos(λx), sin(λx), cosh(λx), sinh(λx)}.

Both sin(λx) and sinh(λx) satisfy the left boundary conditions. Consider the linear combination c1 cos(λx) +
c2 cosh(λx). The left boundary conditions impose the two constraints c1 + c2 = 0, c1 − c2 = 0. Only the trivial
linear combination of cos(λx) and cosh(λx) can satisfy the left boundary condition. Thus the solution has the
form,

X = c1 sin(λx) + c2 sinh(λx).

The right boundary conditions impose the constraints,

(
c1 sin(λL) + c2 sinh(λL) = 0,
−c1 λ2 sin(λL) + c2 λ2 sinh(λL) = 0

(
c1 sin(λL) + c2 sinh(λL) = 0,
−c1 sin(λL) + c2 sinh(λL) = 0

1645
This set of equations has a nontrivial solution if and only if the determinant is zero,
 
 sin(λL) sinh(λL)
− sin(λL) sinh(λL) = 2 sin(λL) sinh(λL) = 0.
 

Since sinh(z) is nonzero in 0 ≤ arg(z) < π/2, z =

6 0, and sin(z) has the zeros z = nπ, n ∈ N in this domain, the
eigenvalues and eigenfunctions are,
nπ  nπx 
λn = , Xn = sin , n ∈ N.
L L
The differential equation for T becomes,
 nπ 4
T 00 = −a2 T,
L
which has the solutions,
        
nπ 2 nπ 2
cos a t , sin a t .
L L
The eigen-solutions of the partial differential equation are,
 nπx         
(1) nπ 2 (2)
 nπx  nπ 2
un = sin cos a t , un = sin sin a t , n ∈ N.
L L L L
We expand the solution of the partial differential equation in a series of the eigen-solutions.
∞  nπx          
X nπ 2 nπ 2
u(x, t) = sin cn cos a t + dn sin a t
n=1
L L L

The initial condition for u(x, t) and ut (x, t) allow us to determine the coefficients in the expansion.
X∞  nπx 
u(x, 0) = cn sin = f (x)
n=1
L
X∞  nπ 2  nπx 
ut (x, 0) = dn a sin = g(x)
n=1
L L

1646
cn and dn are coefficients in Fourier sine series.
Z L
2  nπx 
cn = f (x) sin dx
L 0 L

Z L
2L  nπx 
dn = 2 2 g(x) sin dx
aπ n 0 L

Solution 39.16

ut = κuxx + I 2 αu, 0 < x < L, t > 0,

u(0, t) = u(L, t) = 0, u(x, 0) = g(x).
We will solve this problem with an expansion in eigen-solutions of the partial differential equation. We substitute
the separation of variables u(x, t) = X(x)T (t) into the partial differential equation.
(XT )t = κ(XT )xx + I 2 αXT
T0 I 2α X 00
− = = −λ2
κT κ X
Now we have an ordinary differential equation for T and a Sturm-Liouville eigenvalue problem for X. (Note
that we have followed the rule of thumb that the problem will be easier if we move all the parameters out of the
eigenvalue problem.)
T 0 = − κλ2 − I 2 α T

X 00 = −λ2 X, X(0) = X(L) = 0

The eigenvalues and eigenfunctions for X are
nπ  nπx 
λn = , Xn = sin , n ∈ N.
L L

1647
The differential equation for T becomes,
   
nπ 2
Tn0 =− κ 2
− I α Tn ,
L
which has the solution,
     
nπ 2 2
Tn = c exp − κ −I α t .
L
From this solution, we see that the critical current is
r
κπ
ICR = .
αL
If I is greater that this, then the eigen-solution for n = 1 will be exponentially growing. This would make the
whole solution exponentially growing. For I < ICR , each of the Tn is exponentially decaying. The eigen-solutions
of the partial differential equation are,
     
nπ 2 2
 nπx 
un = exp − κ − I α t sin , n ∈ N.
L L
We expand u(x, t) in its eigen-solutions, un .
∞      
X nπ 2 2
 nπx 
u(x, t) = an exp − κ − I α t sin
n=1
L L

We determine the coefficients an from the initial condition.

X∞  nπx 
u(x, 0) = an sin = g(x)
n=1
L
Z L
2  nπx 
an = g(x) sin dx.
L 0 L
If α < 0, then the solution is exponentially decaying regardless of current. Thus there is no critical current.

1648
Solution 39.17

a) The problem is
ut (x, y, z, t) = κ∆u(x, y, z, t), −∞ < x < ∞, −∞ < y < ∞, 0 < z < a, t > 0,
u(x, y, z, 0) = T, u(x, y, 0, t) = u(x, y, a, t) = 0.
Because of symmetry, the partial differential equation in four variables is reduced to a problem in two
variables,
ut (z, t) = κuzz (z, t), 0 < z < a, t > 0,
u(z, 0) = T, u(0, t) = u(a, t) = 0.
We will solve this problem with an expansion in eigen-solutions of the partial differential equation that
satisfy the homogeneous boundary conditions. We substitute the separation of variables u(z, t) = Z(z)T (t)
into the partial differential equation.
ZT 0 = κZ 00 T
T0 Z 00
= = −λ2
κT Z
With the boundary conditions at z = 0, a we have the Sturm-Liouville eigenvalue problem,
Z 00 = −λ2 Z, Z(0) = Z(a) = 0,
which has the solutions,
nπ  nπz 
λn = , Zn = sin , n ∈ N.
a a
The problem for T becomes,
 nπ 2
Tn0 = −κ Tn ,
a

1649
 with the solution,
  nπ 2 
Tn = exp −κ t .
a
The eigen-solutions are
 nπz    nπ 2 
un (z, t) = sin exp −κ t .
a a
The solution for u is a linear combination of the eigen-solutions. The slowest decaying eigen-solution is
 πz    π 2 
u1 (z, t) = sin exp −κ t .
a a
Thus the e-folding time is

a2
∆e = .
κπ 2

b) The problem is
ut (r, θ, z, t) = κ∆u(r, θ, z, t), 0 < r < a, 0 < θ < 2π, −∞ < z < ∞, t > 0,
u(r, θ, z, 0) = T, u(0, θ, z, t) is bounded, u(a, θ, z, t) = 0.
The Laplacian in cylindrical coordinates is
1 1
∆u = urr + ur + 2 uθθ + uzz .
r r
Because of symmetry, the solution does not depend on θ or z.
 
1
ut (r, t) = κ urr (r, t) + ur (r, t) , 0 < r < a, t > 0,
r
u(r, 0) = T, u(0, t) is bounded, u(a, t) = 0.

1650
We will solve this problem with an expansion in eigen-solutions of the partial differential equation that
satisfy the homogeneous boundary conditions at r = 0 and r = a. We substitute the separation of variables
u(r, t) = R(r)T (t) into the partial differential equation.
 
0 00 1 0
RT = κ R T + R T
r
0 00 0
T R R
= + = −λ2
κT R rR
We have the eigenvalue problem,
1
R00 + R0 + λ2 R = 0, R(0) is bounded, R(a) = 0.
r
Recall that the Bessel equation,
ν2
 
1 0
00 2
y + y + λ − 2 y = 0,
x x
has the general solution y = c1 Jν (λx) + c2 Yν (λx). We discard the Bessel function of the second kind, Yν , as
it is unbounded at the origin. The solution for R(r) is

R(r) = J0 (λr).

Applying the boundary condition at r = a, we see that the eigenvalues and eigenfunctions are
 
βn βn r
λn = , Rn = J0 , n ∈ N,
a a
where {βn } are the positive roots of the Bessel function J0 .
The differential equation for T becomes,
 2
βn
Tn0 = −κ Tn ,
a

1651
 which has the solutions,
 2 !
βn
Tn = exp −κ t .
a

The eigen-solutions of the partial differential equation for u(r, t) are,

   2 !
βn r βn
un (r, t) = J0 exp −κ t .
a a

The solution u(r, t) is a linear combination of the eigen-solutions, un . The slowest decaying eigenfunction
is,
   2 !
β1 r β1
u1 (r, t) = J0 exp −κ t .
a a

Thus the e-folding time is

a2
∆e = .
κβ12

c) The problem is

ut (r, θ, φ, t) = κ∆u(r, θ, φ, t), 0 < r < a, 0 < θ < 2π, 0 < φ < π, t > 0,
u(r, θ, φ, 0) = T, u(0, θ, φ, t) is bounded, u(a, θ, φ, t) = 0.

The Laplacian in spherical coordinates is,

2 1 cos θ 1
∆u = urr + ur + 2 uθθ + 2 uθ + 2 2 uφφ .
r r r sin θ r sin θ

1652
Because of symmetry, the solution does not depend on θ or φ.
 
2
ut (r, t) = κ urr (r, t) + ur (r, t) , 0 < r < a, t > 0,
r
u(r, 0) = T, u(0, t) is bounded, u(a, t) = 0
We will solve this problem with an expansion in eigen-solutions of the partial differential equation that
satisfy the homogeneous boundary conditions at r = 0 and r = a. We substitute the separation of variables
u(r, t) = R(r)T (t) into the partial differential equation.
 
0 00 2 0
RT = κ R T + R T
r
0 00 0
T R 2R
= + = −λ2
κT R rR
We have the eigenvalue problem,
2
R00 + R0 + λ2 R = 0, R(0) is bounded, R(a) = 0.
r
Recall that the equation,
 
002 0 2 ν(ν + 1)
y + y + λ − y = 0,
x x2
has the general solution y = c1 jν (λx) + c2 yν (λx), where jν and yν are the spherical Bessel functions of the
first and second kind. We discard yν as it is unbounded at the origin. (The spherical Bessel functions are
related to the Bessel functions by
r
π
jν (x) = Jν+1/2 (x).)
2x
The solution for R(r) is
Rn = j0 (λr).

1653
 Applying the boundary condition at r = a, we see that the eigenvalues and eigenfunctions are
γn γ r
n
λn = , Rn = j0 , n ∈ N.
a a
The problem for T becomes
 γ 2
n
Tn0 = −κ Tn ,
a
which has the solutions,
  γ 2 
n
Tn = exp −κ t .
a

The eigen-solutions of the partial differential equation are,

γ r   γ 2 
n n
un (r, t) = j0 exp −κ t .
a a

The slowest decaying eigen-solution is,

γ r   γ 2 
1 1
u1 (r, t) = j0 exp −κ t .
a a

Thus the e-folding time is

a2
∆e = .
κγ12

d) If the edges are perfectly insulated, then no heat escapes through the boundary. The temperature is constant
for all time. There is no e-folding time.

1654
Solution 39.18
We will solve this problem with an eigenfunction expansion. Since the partial differential equation is homogeneous,
we will find eigenfunctions in both x and y. We substitute the separation of variables u(x, y, t) = X(x)Y (y)T (t)
into the partial differential equation.
XY T 0 = κ(t) (X 00 Y T + XY 00 T )
T0 X 00 Y 00
= + = −λ2
κ(t)T X Y
X 00 Y 00
=− − λ2 = −µ2
X Y
First we have a Sturm-Liouville eigenvalue problem for X,
X 00 = µ2 X, X 0 (0) = X 0 (a) = 0,
which has the solutions,
mπ  mπx 
µm = , Xm = cos , m = 0, 1, 2, . . . .
a a
Now we have a Sturm-Liouville eigenvalue problem for Y ,
  mπ 2 
00 2
Y =− λ − Y, Y (0) = Y (b) = 0,
a
which has the solutions,
r
mπ 2  nπ 2  nπy 
λmn = + , Yn = sin , m = 0, 1, 2, . . . , n = 1, 2, 3, . . . .
a b b
nπy
A few of the eigenfunctions, cos mπx



a
sin b
, are shown in Figure 39.3.
The differential equation for T becomes,
 
0 mπ 2  nπ 2
Tmn = − + κ(t)Tmn ,
a b

1655
 m=0, n=1 m=0, n=2 m=0, n=3

m=1, n=1 m=1, n=2 m=1, n=3

m=2, n=1 m=2, n=2 m=2, n=3

mπx nπy



Figure 39.3: The eigenfunctions cos a
sin b

which has the solutions,

  Z t 
mπ 2  nπ 2
Tmn = exp − + κ(τ ) dτ .
a b 0

The eigen-solutions of the partial differential equation are,

 mπx   nπy    Z t 
mπ 2  nπ 2
umn = cos sin exp − + κ(τ ) dτ .
a b a b 0

1656
The solution of the partial differential equation is,
∞ X
∞  mπx   nπy    Z t 
X mπ 2  nπ 2
u(x, y, t) = cmn cos sin exp − + κ(τ ) dτ .
m=0 n=1
a b a b 0

We determine the coefficients from the initial condition.

X∞ X ∞  mπx   nπy 
u(x, y, 0) = cmn cos sin = f (x, y)
m=0 n=1
a b

Z a Z b
2  nπ 
c0n = f (x, y) sin dy dx
ab 0 0 b

Z a Z b
4  mπ   nπ 
cmn = f (x, y) cos sin dy dx
ab 0 0 a b

Solution 39.19
The steady state temperature satisfies Laplace’s equation, ∆u = 0. The Laplacian in cylindrical coordinates is,
1 1
∆u(r, θ, z) = urr + ur + 2 uθθ + uzz .
r r
Because of the homogeneity in the z direction, we reduce the partial differential equation to,
1 1
urr + ur + 2 uθθ = 0, 0 < r < 1, 0 < θ < π.
r r
The boundary conditions are,

u(r, 0) = u(r, π) = 0, u(0, θ) = 0, u(1, θ) = 1.

1657
We will solve this problem with an eigenfunction expansion. We substitute the separation of variables u(r, θ) =
R(r)T (θ) into the partial differential equation.

1 1
R00 T + R0 T + 2 RT 00 = 0
r r
00 0
R R T 00
r2 +r =− = λ2
R R T
We have the regular Sturm-Liouville eigenvalue problem,

T 00 = −λ2 T, T (0) = T (π) = 0,

which has the solutions,

λn = n, Tn = sin(nθ), n ∈ N.

The problem for R becomes,

r2 R00 + rR0 − n2 R = 0, R(0) = 0.

This is an Euler equation. We substitute R = rα into the differential equation to obtain,

α(α − 1) + α − n2 = 0,
α = ±n.

The general solution of the differential equation for R is

Rn = c1 rn + c2 r−n .

The solution that vanishes at r = 0 is

Rn = crn .

1658
The eigen-solutions of the differential equation are,
un = rn sin(nθ).
The solution of the partial differential equation is
∞
X
u(r, θ) = an rn sin(nθ).
n=1

We determine the coefficients from the boundary condition at r = 1.

∞
X
u(1, θ) = an sin(nθ) = 1
n=1
Z π
2 2
an = sin(nθ) dθ = (1 − (−1)n )
π 0 πn
The solution of the partial differential equation is
∞
4 X n
u(r, θ) = r sin(nθ).
π n=1
odd n

Solution 39.20
The problem is
uxx + uyy = 0, 0 < x, 0 < y < 1,
u(x, 0) = u(x, 1) = 0, u(0, y) = f (y).
We substitute the separation of variables u(x, y) = X(x)Y (y) into the partial differential equation.
X 00 Y + XY 00 = 0
X 00 Y 00
=− = λ2
X Y

1659
We have the regular Sturm-Liouville problem,
Y 00 = −λ2 Y, Y (0) = Y (1) = 0,
which has the solutions,
λn = nπ, Yn = sin(nπy), n ∈ N.
The problem for X becomes,
Xn00 = (nπ)2 X,
which has the general solution,
Xn = c1 enπx + c2 e−nπx .
The solution that is bounded as x → ∞ is,
Xn = c e−nπx .
The eigen-solutions of the partial differential equation are,
un = e−nπx sin(nπy), n ∈ N.
The solution of the partial differential equation is,
∞
X
u(x, y) = an e−nπx sin(nπy).
n=1

We find the coefficients from the boundary condition at x = 0.

∞
X
u(0, y) = an sin(nπy) = f (y)
n=1

Z 1
an = 2 f (y) sin(nπy) dy
0

1660
Solution 39.21
The Laplacian in circular coordinates is

1 1
∆u = urr + ur + 2 uθθ .
r r
Since we have homogeneous boundary conditions at θ = 0 and θ = α, we will solve this problem with an
eigenfunction expansion. We substitute the separation of variables u(r, θ) = R(r)T (θ) into the partial differential
equation.
1 1
R00 T + R0 T + 2 RT 00 = 0
r r
00 0
R R T 00
r2 +r =− = λ2 .
R R T
We have the regular Sturm-Liouville eigenvalue problem,

T 00 = −λ2 T, T (0) = T (α) = 0,

which has the solutions,

nπ  nπx 
λn = , Tn = sin , n ∈ N.
α α
The differential equation for R becomes,
 nπ 2
2 00 0
r R + rR − R = 0, R(a) = 0.
α
This is an Euler equation. We make the substitution, R = rβ .
 nπ 2
β(β − 1) + β − =0
α
nπ
β=±
α

1661
The general solution of the equation for R is

R = c1 rnπ/α + c2 r−nπ/α .

The solution, (up to a multiplicative constant), that vanishes at r = a is

R = rnπ/α − a2nπ/α r−nπ/α .

Thus the series expansion of our solution is,

∞  
X
nπ/α 2nπ/α −nπ/α

nπθ
u(r, θ) = cn r −a r sin .
n=1
α

We determine the coefficients from the boundary condition at r = b.

∞  
X
nπ/α 2nπ/α −nπ/α

nπθ
u(b, θ) = cn b −a b sin = f (θ)
n=1
α

Z α  
2 nπθ
cn = f (θ) sin dθ
α (b nπ/α − a2nπ/α b−nπ/α ) 0 α

Solution 39.22

a) The mathematical statement of the problem is

utt = c2 uxx , 0 < x < L, t > 0,

u(0, t) = u(L, t) = 0,
(
v for |x − ξ| < d
u(x, 0) = 0, ut (x, 0) =
0 for |x − ξ| > d.

1662
Because we are interest in the harmonics of the motion, we will solve this problem with an eigenfunction expansion
in x. We substitute the separation of variables u(x, t) = X(x)T (t) into the wave equation.

XT 00 = c2 X 00 T
T 00 X 00
= = −λ2
c2 T X
The eigenvalue problem for X is,

X 00 = −λ2 X, X(0) = X(L) = 0,

which has the solutions,

nπ  nπx 
λn = , Xn = sin , n ∈ N.
L L
The ordinary differential equation for the Tn are,
 nπc 2
Tn00 =− Tn ,
L
which have the linearly independent solutions,
   
nπct nπct
cos , sin .
L L
The solution for u(x, t) is a linear combination of the eigen-solutions.
∞  nπx      
X nπct nπct
u(x, t) = sin an cos + bn sin
n=1
L L L

Since the string initially has zero displacement, each of the an are zero.
∞  nπx   
X nπct
u(x, t) = bn sin sin
n=1
L L

1663
Now we use the initial velocity to determine the coefficients in the expansion. Because the position is a continuous
function of x, and there is a jump discontinuity in the velocity as a function of x, the coefficients in the expansion
will decay as 1/n2 .
∞
(
X nπc  nπx  v for |x − ξ| < d
ut (x, 0) = bn sin =
n=1
L L 0 for |x − ξ| > d.
Z L
nπc 2  nπx 
bn = ut (x, 0) sin dx
L L 0 L
Z ξ+d
2  nπx 
bn = v sin dx
nπc ξ−d L
   
4Lv nπd nπξ
= 2 2 sin sin
nπ c L L
The solution for u(x, t) is,
∞      
4Lv X 1 nπd nπξ  nπx  nπct
u(x, t) = 2 sin sin sin sin .
π c n=1 n2 L L L L

b) The form of the solution is again,

∞  nπx   
X nπct
u(x, t) = bn sin sin
n=1
L L
We determine the coefficients in the expansion from the initial velocity.
 
∞
(
π(x−ξ)
X nπc  nπx  v cos 2d
for |x − ξ| < d
ut (x, 0) = bn sin =
n=1
L L 0 for |x − ξ| > d.
Z L
nπc 2  nπx 
bn = ut (x, 0) sin dx
L L 0 L

1664
 Z ξ+d  
2 π(x − ξ)  nπx 
bn = v cos sin dx
nπc ξ−d 2d L
(
8dL2 v nπd nπξ L



2 2 2 2 cos sin for d 6= ,
bn = nπv c(L −4d n ) L
2nπd


L nπξ
2n
L
n2 π 2 c
2nπd + L sin L sin L for d = 2n

The solution for u(x, t) is,

∞
8dL2 v X
     
1 nπd nπξ  nπx  nπct L
u(x, t) = cos sin sin sin for d 6= ,
π 2 c n=1 n(L2 − 4d2 n2 ) L L L L 2n
∞       
v X 1 2nπd nπξ  nπx  nπct L
u(x, t) = 2 2nπd + L sin sin sin sin for d = .
π c n=1 n2 L L L L 2n

c) The kinetic energy of the string is

Z L
1
E= ρ (ut (x, t))2 dx,
2 0

where ρ is the density of the string per unit length.

Flat Hammer. The nth harmonic is
     
4Lv nπd nπξ  nπx  nπct
un = 2 2 sin sin sin sin .
nπ c L L L L

The kinetic energy of the nth harmonic is

L 2
4Lv 2
Z       
ρ ∂un nπd nπξ nπct
En = dx = 2 2 sin2 sin 2
cos 2
.
2 0 ∂t nπ L L L

1665
This will be maximized if


nπξ
2
sin = 1,
L
nπξ π(2m − 1)
= , m = 1, . . . , n,
L 2
(2m − 1)L
ξ= , m = 1, . . . , n
2n

We note that the kinetic energies of the nth harmonic decay as 1/n2 .
L
Curved Hammer. We assume that d 6= 2n . The nth harmonic is

8dL2 v
     
nπd nπξ  nπx  nπct
un = cos sin sin sin .
nπ 2 c(L2 − 4d2 n2 ) L L L L

The kinetic energy of the nth harmonic is

L 2
16d2 L3 v 2
Z       
ρ ∂un nπd nπξ nπct
En = dx = 2 2 cos2 sin 2
cos 2
.
2 0 ∂t π (L − 4d2 n2 )2 L L L

This will be maximized if

 
2 nπξ
sin = 1,
L
(2m − 1)L
ξ= , m = 1, . . . , n
2n

We note that the kinetic energies of the nth harmonic decay as 1/n4 .

1666
Solution 39.23
In mathematical notation, the problem is
utt − c2 uxx = s(x, t), 0 < x < L, t > 0,
u(0, t) = u(L, t) = 0,
u(x, 0) = ut (x, 0) = 0.
Since this is an inhomogeneous partial differential equation, we will expand the solution in a series of eigenfunctions
in x for which the coefficients are functions of t. The solution for u has the form,
∞
X  nπx 
u(x, t) = un (t) sin .
n=1
L

Substituting this expression into the inhomogeneous partial differential equation will give us ordinary differential
equations for each of the un .
∞   2 
2 nπ
X  nπx 
00
un + c un sin = s(x, t).
n=1
L L

We expand the right side in a series of the eigenfunctions.

∞
X  nπx 
s(x, t) = sn (t) sin .
n=1
L

For 0 < t < δ we have

2 L
Z  nπx 
sn (t) = s(x, t) sin dx
L 0 L
2 L
   
π(x − ξ)
Z
πt  nπx 
= v cos sin sin dx
L 0 2d δ L
     
8dLv nπd nπξ πt
= 2 2 2
cos sin sin .
π(L − 4d n ) L L δ

1667
For t > δ, sn (t) = 0. Substituting this into the partial differential equation yields,
(
8dLv nπd nπξ πt




 nπc 2
2 2 2 cos sin sin , for t < δ,
u00n + un = π(L −4d n ) L L δ
L 0 for t > δ.

Since the initial position and velocity of the string is zero, we have

un (0) = u0n (0) = 0.

First we solve the differential equation on the range 0 < t < δ. The homogeneous solutions are
   
nπct nπct
cos , sin .
L L
Since the right side of the ordinary differential equation is a constant times sin(πt/δ), which is an eigenfunction
of the differential operator, we can guess the form of a particular solution, pn (t).
 
πt
pn (t) = d sin
δ
We substitute this into the ordinary differential equation to determine the multiplicative constant d.
8dδ 2 L3 v
     
nπd nπξ πt
pn (t) = − 3 2 2 2 2 2 2 2
cos sin sin
π (L − c δ n )(L − 4d n ) L L δ
The general solution for un (t) is
8dδ 2 L3 v
         
nπct nπct nπd nπξ πt
un (t) = a cos + b sin − 3 2 2 2 2 2 2 2
cos sin sin .
L L π (L − c δ n )(L − 4d n ) L L δ
We use the initial conditions to determine the constants a and b. The solution for 0 < t < δ is
8dδ 2 L3 v
       
nπd nπξ L nπct πt
un (t) = 3 2 2 2 2 2 2 2
cos sin sin − sin .
π (L − c δ n )(L − 4d n ) L L δcn L δ

1668
The solution for t > δ, the solution is a linear combination of the homogeneous solutions. This linear combination
is determined by the position and velocity at t = δ. We use the above solution to determine these quantities.
8dδ 2 L4 v
     
nπd nπξ nπcδ
un (δ) = 3 cos sin sin
π δcn(L2 − c2 δ 2 n2 )(L2 − 4d2 n2 ) L L L
2 3
     
0 8dδ L v nπd nπξ nπcδ
un (δ) = 2 cos sin 1 + cos
π δ(L2 − c2 δ 2 n2 )(L2 − 4d2 n2 ) L L L
The fundamental set of solutions at t = δ is
    
nπc(t − δ) L nπc(t − δ)
cos , sin
L nπc L
From the initial conditions at t = δ, we see that the solution for t > δ is

8dδ 2 L3 v
   
nπd nπξ
un (t) = 3 2 cos sin
π (L − c2 δ 2 n2 )(L2 − 4d2 n2 ) L L
         
L nπcδ nπc(t − δ) π nπcδ nπc(t − δ)
sin cos + 1 + cos sin .
δcn L L δ L L

Width of the Hammer. The nth harmonic has the width dependent factor,
 
d nπd
cos .
L2 − 4d2 n2 L
Differentiating this expression and trying to find zeros to determine extrema would give us an equation with both
algebraic and transcendental terms. Thus we don’t attempt to find the maxima exactly. We know that d < L.
The cosine factor is large when
nπd
≈ mπ, m = 1, 2, . . . , n − 1,
L
mL
d≈ , m = 1, 2, . . . , n − 1.
n

1669
Substituting d = mL/n into the width dependent factor gives us

d
(−1)m .
L2 (1 − 4m2 )

Thus we see that the amplitude of the nth harmonic and hence its kinetic energy will be maximized for

L
d≈
n

The cosine term in the width dependent factor vanishes when

(2m − 1)L
d= , m = 1, 2, . . . , n.
2n

The kinetic energy of the nth harmonic is minimized for these widths.
L L
For the lower harmonics, n  2d , the kinetic energy is proportional to d2 ; for the higher harmonics, n  2d
,
the kinetic energy is proportional to 1/d2 .
Duration of the Blow. The nth harmonic has the duration dependent factor,

δ2
         
L nπcδ nπc(t − δ) π nπcδ nπc(t − δ)
sin cos + 1 + cos sin .
L2 − n2 c2 δ 2 ncδ L L δ L L

If we assume that δ is small, then

 
L nπcδ
sin ≈ π.
ncδ L

and
  
π nπcδ 2π
1 + cos ≈ .
δ L δ

1670
Thus the duration dependent factor is about,
 
δ nπc(t − δ)
sin .
L − n2 c2 δ 2
2 L
L
Thus for the lower harmonics, (those satisfying n  cδ ), the amplitude is proportional to δ, which means that the
2 L
kinetic energy is proportional to δ . For the higher harmonics, (those with n  cδ ), the amplitude is proportional
to 1/δ, which means that the kinetic energy is proportional to 1/δ 2 .

Solution 39.24
Substituting u(x, y, z, t) = v(x, y, z) eiωt into the wave equation will give us a Helmholtz equation.

−ω 2 v eiωt − c2 (vxx + vyy + vzz ) eiωt = 0

vxx + vyy + vzz + k 2 v = 0.
We find the propagating modes with separation of variables. We substitute v = X(x)Y (y)Z(z) into the Helmholtz
equation.
X 00 Y Z + XY 00 Z + XY Z 00 + k 2 XY Z = 0
X 00 Y 00 Z 00
− = + + k2 = ν 2
X Y Z
The eigenvalue problem in x is
X 00 = −ν 2 X, X(0) = X(L) = 0,
which has the solutions,
nπ  nπx 
νn = , Xn = sin .
L L
We continue with the separation of variables.
Y 00 Z 00  nπ 2
− = + k2 − = µ2
Y Z L

1671
The eigenvalue problem in y is

Y 00 = −µ2 Y, Y (0) = Y (L) = 0,

which has the solutions,

mπ  mπy 
µn = , Ym = sin .
L L
Now we have an ordinary differential equation for Z,
  π 2 
00 2 2 2


Z + k − n +m Z = 0.
L
We define the eigenvalues,
 π 2
λ2n,m = k 2 − n2 + m2 .


L
π 2


If k 2 − L
(n2 + m2 ) < 0, then the solutions for Z are,
s  !
 π 2
exp ± (n2 + m2 ) − k 2 z .
L

We discard this

case, as the solutions are not bounded as z → ∞.
π 2
2
If k − L (n + m2 ) = 0, then the solutions for Z are,
2

{1, z}

The solution Z = 1 satisfies the boundedness and nonzero condition at infinity. This corresponds to a standing
wave.
2
If k 2 − Lπ (n2 + m2 ) > 0, then the solutions for Z are,

e±iλn,m z .

1672
 π 2


These satisfy the boundedness and nonzero conditions at infinity. For values of n, m satisfying k 2 − L
(n2 + m2 ) ≥
0, there are the propagating modes,
 nπx   mπy 
un,m = sin sin ei(ωt±λn,m z) .
L L

Solution 39.25

utt = c2 ∆u, 0 < x < a, 0 < y < b, (39.10)

u(0, y) = u(a, y) = u(x, 0) = u(x, b) = 0.
We substitute the separation of variables u(x, y, t) = X(x)Y (y)T (t) into Equation 39.10.
T 00 X 00 Y 00
= + = −ν
c2 T X Y
X 00 Y 00
=− − ν = −µ
X Y
This gives us differential equations for X(x), Y (y) and T (t).
X 00 = −µX, X(0) = X(a) = 0
Y 00 = −(ν − µ)Y, Y (0) = Y (b) = 0
T 00 = −c2 νT
First we solve the problem for X.
 mπ 2  mπx 
µm = , Xm = sin
a a
Then we solve the problem for Y .
 mπ 2  nπ 2  nπy 
νm,n = + , Ym,n = sin
a b b

1673
Finally we determine T .
r  !
cos m 2  n 2
Tm,n = cπ + t
sin a b

The modes of oscillation are

r  !
 mπx   nπy  cos m 2  n 2
um,n = sin sin cπ + t .
a b sin a b

The frequencies are

r
m 2  n 2
ωm,n = cπ + .
a b
Figure 39.4 shows a few of the modes of oscillation in surface and density plots.

Solution 39.26
We substitute the separation of variables φ = X(x)Y (y)T (t) into the differential equation.
φt = a2 (φxx + φyy ) (39.11)
XY T 0 = a2 (X 00 Y T + XY 00 T )
T0 X 00 Y 00
= + = −ν
a2 T X Y
T0 X 00 Y 00
= −ν, = −ν − = −µ
a2 T X Y
First we solve the eigenvalue problem for X.
X 00 + µX = 0, X(0) = X(lx ) = 0
 2  
mπ mπx
µm = , Xm (x) = sin , m ∈ Z+
lx lx

1674
 m=1, n=1 m=1, n=2 m=1, n=3 m=1, n=1 m=1, n=2 m=1, n=3

m=2, n=1 m=2, n=2 m=2, n=3 m=2, n=1 m=2, n=2 m=2, n=3

m=3, n=1 m=3, n=2 m=3, n=3 m=3, n=1 m=3, n=2 m=3, n=3

Figure 39.4: The modes of oscillation of a rectangular drum head.

Then we solve the eigenvalue problem for Y .

Y 00 + (ν − µm )Y = 0, Y 0 (0) = Y 0 (ly ) = 0
 2  
nπ nπy
νmn = µm + , Ymn (y) = cos , n ∈ Z0+
ly ly
Next we solve the differential equation for T , (up to a multiplicative constant).
T 0 = −a2 νmn T
T (t) = exp −a2 νmn t

The eigensolutions of Equation 39.11 are

sin(µm x) cos(νmn y) exp −a2 νmn t , m ∈ Z+ , n ∈ Z0+ .

1675
We choose the eigensolutions φmn to be orthonormal on the xy domain at t = 0.

s
2
sin(µm x) exp −a2 νmn t , m ∈ Z+


φm0 (x, y, t) =
lx ly
2
sin(µm x) cos(νmn y) exp −a2 νmn t , m ∈ Z+ , n ∈ Z+


φmn (x, y, t) = p
lx ly

The solution of Equation 39.11 is a linear combination of the eigensolutions.

∞
X
φ(x, y, t) = cmn φmn (x, y, t)
m=1
n=0

1676
We determine the coefficients from the initial condition.
φ(x, y, 0) = 1
∞
X
cmn φmn (x, y, 0) = 1
m=1
n=0
Z lx Z ly
cmn = φmn (x, y, 0) dy dx
0 0
s Z lx Z ly
2
cm0 = sin(µm x) dy dx
lx ly 0 0
p 1 − (−1)m
cm0 = 2lx ly , m ∈ Z+
mπ
Z lx Z ly
2
cmn =p sin(µm x) cos(νmn y) dy dx
lx ly 0 0
cmn = 0, m ∈ Z+ , n ∈ Z+
∞
X
φ(x, y, t) = cm0 φm0 (x, y, t)
m=1
∞ p
X 2 2lx ly
sin(µm x) exp −a2 νmn t


φ(x, y, t) =
m=1
mπ
odd m

Addendum. Note that an equivalent problem to the one specified is

φt = a2 (φxx + φyy ) , 0 < x < lx , −∞ < y < ∞,
φ(x, y, 0) = 1, φ(0, y, t) = φ(ly , y, t) = 0.
Here we have done an even periodic continuation of the problem in the y variable. Thus the boundary conditions
φy (x, 0, t) = φy (x, ly , t) = 0

1677
are automatically satisfied. Note that this problem does not depend on y. Thus we only had to solve

φt = a2 φxx , 0 < x < lx

φ(x, 0) = 1, φ(0, t) = φ(ly , t) = 0.

Solution 39.27
1. Since the initial and boundary conditions do not depend on θ, neither does φ. We apply the separation of
variables φ = u(r)T (t).

φt = a2 ∆φ (39.12)
1
φt = a2 (rφr )r (39.13)
r
T0 1
= (ru0 )0 = −λ (39.14)
a2 T r
We solve the eigenvalue problem for u(r).

(ru0 )0 + λu = 0, u(0) bounded, u(R) = 0

First we write the general solution.

√  √ 
u(r) = c1 J0 λr + c2 Y0 λr

The Bessel function of the second kind, Y0 , is not bounded at r = 0, so c2 = 0. We use the boundary
condition at r = R to determine the eigenvalues.
 2  
j0,n j0,n r
λn = , un (r) = cJ0
R R

1678
We choose the constant c so that the eigenfunctions are orthonormal with respect to the weighting function
r.
 
j0,n r
J0 R
un (r) = r  
RR j0,n r
0
rJ02 R
√  
2 j0,n r
= J0
RJ1 (j0,n ) R

Now we solve the differential equation for T .

T 0 = −a2 λn T
 2 !
aj0,n
Tn = exp − t
R2

The eigensolutions of Equation 39.12 are

√    2 !
2 j0,n r aj0,n
φn (r, t) = J0 exp − t
RJ1 (j0,n ) R R2

The solution is a linear combination of the eigensolutions.

∞ √    2 !
X 2 j0,n r aj0,n
φ= cn J0 exp − t
n=1
RJ1 (j0,n ) R R2

1679
 We determine the coefficients from the initial condition.
φ(r, θ, 0) = V
∞ √  
X 2 j0,n r
cn J0 =V
n=1
RJ 1 (j 0,n ) R
Z R √  
2 j0,n r
cn = Vr J0 dr
0 RJ1 (j0,n ) R
√
2 R
cn = V J1 (j0,n )
RJ1 (j0,n ) j0,n /R
√
2 VR
cn =
j0,n
 
∞ J j0,n r  2 !
X 0 R aj0,n
φ(r, θ, t) = 2V exp − 2
t
n=1
j 0,n J 1 (j 0,n ) R

2.
r
2  πν π 
Jν (r) ∼ cos r − − , r → +∞
πr  2 4

ν 1
jν,n ∼ n + − π
2 4
For large n, the terms in the series solution at t = 0 are
  q  
j0,n r
J0 j0,n
r 2R π
R πj0,n r
cos R
− 4
∼ q
j0,n J1 (j0,n ) j0,n πj20,n cos j0,n − 3π


4
 
(n−1/4)πr π
R cos R
− 4
∼ .
r(n − 1/4)π cos ((n − 1)π)

1680
 The coefficients decay as 1/n.

Solution 39.28
1. We substitute the separation of variables Ψ = T (t)Θ(θ)Φ(φ) into Equation 39.7
a2
 
0 1 ∂ 0 1 00
T ΘΦ = 2 (sin θ T Θ Φ) + T ΘΦ
R sin θ ∂θ sin2 θ
R2 T 0 1 Φ00
 
1 0 0
= (sin θ Θ ) + = −µ
a2 T sin θ Θ sin2 θ Φ
sin θ Φ00
(sin θ Θ0 )0 + µ sin2 θ = − =ν
Θ Φ
We have differential equations for each of T , Θ and Φ.

0 a2 1 0 0
 ν 
T = −µ 2 T, (sin θ Θ ) + µ − Θ = 0, Φ00 + νΦ = 0
R sin θ sin2 θ
2. In order that the solution be continuously differentiable, we need the periodic boundary conditions
Φ(0) = Φ(2π), Φ0 (0) = Φ0 (2π).
The eigenvalues and eigenfunctions for Φ are
1
νn = n2 , Φn = √ einφ , n ∈ Z.
2π
Now we deal with the equation for Θ.
d 1 d
x = cos θ, Θ(θ) = P (x), sin2 θ = 1 − x2 , =
dx sin θ dθ
1 2 1 0 0
 ν 
(sin θ Θ) + µ− Θ=0
sin θ sin θ  sin2θ


0 n2
1 − x2 P 0 + µ − P =0
1 − x2

1681
 P (x) should be bounded at the endpoints, x = −1 and x = 1.

3. If the solution
√ does not depend on θ, then the only one of the Φn that will appear in the solution is
Φ0 = 1/ 2π. The equations for T and P become


0
1 − x2 P 0 + µP = 0, P (±1) bounded,
a2
T 0 = −µ 2 T.
R

The solutions for P are the Legendre polynomials.

µl = l(l + 1), Pl (cos θ), l ∈ Z0+

We solve the differential equation for T .

a2
T 0 = −l(l + 1) 2 T
 2 R 
a l(l + 1)
Tl = exp − t
R2

The eigensolutions of the partial differential equation are

 2 
a l(l + 1)
Ψl = Pl (cos θ) exp − t .
R2

The solution is a linear combination of the eigensolutions.

∞  2 
X a l(l + 1)
Ψ= Al Pl (cos θ) exp − t
l=0
R2

1682
 4. We determine the coefficients in the expansion from the initial condition.
Ψ(θ, 0) = 2 cos2 θ − 1
∞
X
Al Pl (cos θ) = 2 cos2 θ − 1
l=0
 
3 2 1
A0 + A1 cos θ + A2 cos θ − + · · · = 2 cos2 θ − 1
2 2
1 4
A0 = − , A1 = 0, A2 = , A3 = A4 = · · · = 0
3 3
6a2
 
1 4
Ψ(θ, t) = − P0 (cos θ) + P2 (cos θ) exp − 2 t
3 3 R
6a2
   
1 2 2
Ψ(θ, t) = − + 2 cos θ − exp − 2 t
3 3 R

Solution 39.29
Since we have homogeneous boundary conditions at x = 0 and x = 1, we will expand the solution in a series of
eigenfunctions in x. We determine a suitable set of eigenfunctions with the separation of variables, φ = X(x)Y (y).
φxx + φyy = 0 (39.15)
X 00 Y 00
=− = −λ
X Y
We have differential equations for X and Y .
X 00 + λX = 0, X(0) = X(1) = 0
Y 00 − λY = 0, Y (0) = 0
The eigenvalues and orthonormal eigenfunctions for X are
√
λn = (nπ)2 , Xn (x) = 2 sin(nπx), n ∈ Z+ .

1683
The solutions for Y are, (up to a multiplicative constant),

Yn (y) = sinh(nπy).

The solution of Equation 39.15 is a linear combination of the eigensolutions.

∞
X √
φ(x, y) = an 2 sin(nπx) sinh(nπy)
n=1

We determine the coefficients from the boundary condition at y = 2.

∞
X √
x(1 − x) = an 2 sin(nπx) sinh(nπ2)
n=1
√ Z 1
an sinh(2nπ) = 2 x(1 − x) sin(nπx) dx
0
√
2 2(1 − (−1)n )
an = 3 3
n π sinh(2nπ)
∞
8 X 1 sinh(nπy)
φ(x, y) = 3 3
sin(nπx)
π n=1 n sinh(2nπ)
odd n

The solution at x = 1/2, y = 1 is

∞
8 X 1 sinh(nπ)
φ(1/2, 1) = − 3 .
π n=1 n3 sinh(2nπ)
odd n

1684
Let Rk be the relative error at that point incurred by taking k terms.

 − 83 P∞
 
1 sinh(nπ) 
 π n=k+2 3
n sinh(2nπ) 
Rk =  8 Podd ∞
n
1 sinh(nπ) 

 − π3 n=1 n3 sinh(2nπ) 
odd n
P∞ 1 sinh(nπ)
n=k+2 n3 sinh(2nπ)
Rk = Podd
∞
n
1 sinh(nπ)
n=1 3
odd n n sinh(2nπ)

Since R1 ≈ 0.0000693169 we see that one term is sufficient for 1% or 0.1% accuracy.
Now consider φx (1/2, 1).
∞
8 X 1 sinh(nπy)
φx (x, y) = cos(nπx)
π 2 n=1 n2 sinh(2nπ)
odd n
φx (1/2, 1) = 0

Since all the terms in the series are zero, accuracy is not an issue.

Solution 39.30
The solution has the form
(
αr −n−1 Pnm (cos θ) sin(mφ), r > a
ψ=
βrn Pnm (cos θ) sin(mφ), r < a.

The boundary condition on ψ at r = a gives us the constraint

αa−n−1 − βan = 0
β = αa−2n−1 .

1685
Then we apply the boundary condition on ψr at r = a.

−(n + 1)αa−n−2 − nαa−2n−1 an−1 = 1

an+2
α=−
2n + 1
( n+2
a
− 2n+1 r−n−1 Pnm (cos θ) sin(mφ), r > a
ψ= −n+1
− a2n+1 rn Pnm (cos θ) sin(mφ), r<a

Solution 39.31
We expand the solution in a Fourier series.
∞ ∞
1 X X
φ = a0 (r) + an (r) cos(nθ) + bn (r) sin(nθ)
2 n=1 n=1

We substitute the series into the Laplace’s equation to determine ordinary differential equations for the coefficients.

1 ∂2φ
 
∂ ∂φ
r + 2 2 =0
∂r ∂r r ∂θ
1 1 1
a000 + a00 = 0, a00n + a0n − n2 an = 0, b00n + b0n − n2 bn = 0
r r r
The solutions that are bounded at r = 0 are, (to within multiplicative constants),

a0 (r) = 1, an (r) = rn , bn (r) = rn .

Thus φ(r, θ) has the form

∞ ∞
1 X
n
X
φ(r, θ) = c0 + cn r cos(nθ) + dn rn sin(nθ)
2 n=1 n=1

1686
We apply the boundary condition at r = R.

∞
X ∞
X
n−1
φr (R, θ) = ncn R cos(nθ) + ndn Rn−1 sin(nθ)
n=1 n=1

In order that φr (R, θ) have a Fourier series of this form, it is necessary that

Z 2π
φr (R, θ) dθ = 0.
0

In that case c0 is arbitrary in our solution. The coefficients are

Z 2π Z 2π
1 1
cn = φr (R, α) cos(nα) dα, dn = φr (R, α) sin(nα) dα.
πnRn−1 0 πnRn−1 0

1687
We substitute the coefficients into our series solution to determine it up to the additive constant.

∞
R X 1  r n 2π
Z
φ(r, θ) = φr (R, α) cos(n(θ − α)) dα
π n=1 n R 0
∞
R 2π
Z X 1  r n
φ(r, θ) = φr (R, α) cos(n(θ − α)) dα
π 0 n=1
n R
Z 2π ∞ Z r n−1
R X ρ in(θ−α)


φ(r, θ) = φr (R, α) dρ< e dα
π 0 n=1 0
Rn
∞
!
R 2π
Z r X
ρn in(θ−α)
Z
1
φ(r, θ) = φr (R, α)< n
e dρ dα
π 0 0 ρ n=1 R
!
R 2π 1 Rρ ei(θ−α)
Z Z r
φ(r, θ) = φr (R, α)< ρ i(θ−α) dρ dα
π 0 0 ρ1− R e

R 2π
Z   r 
φ(r, θ) = − φr (R, α)< log 1 − ei(θ−α) dα
π 0 R
R 2π
Z  r 
φ(r, θ) = − φr (R, α) log 1 − ei(θ−α)  dα
 
π 0 R
R 2π r2
Z  
r
φ(r, θ) = − φr (R, α) log 1 − 2 cos(θ − α) + 2 dα
2π 0 R R

Solution 39.32
We will assume that both α and β are nonzero. The cases of real and pure imaginary have already been covered.
We solve the ordinary differential equations, (up to a multiplicative constant), to find special solutions of the

1688
diffusion equation.

T0 X 00 (α + iβ)2
= (α + iβ)2 , = 2
T X a 
α + iβ
T = exp (α + iβ)2 t ,


X = exp ± x
a
 
2 2


α β
T = exp α − β t + i2αβt , X = exp ± x ± i x
a a
  
α β
φ = exp α2 − β 2 t ± x + i 2αβt ± x


a a

We take the sum and difference of these solutions to obtain

 

2 2

α  cos β
φ = exp α −β t± x 2αβt ± x
a sin a

1689
Chapter 40

Finite Transforms

Example 40.0.1 Consider the problem

1 ∂2u
∆u − = δ(x − ξ)δ(y − η) e−iωt on − ∞ < x < ∞, 0 < y < b,
c2 ∂t2
with
uy (x, 0, t) = uy (x, b, t) = 0.
Substituting u(x, y, t) = v(x, y) e−iωt into the partial differential equation yields the problem
∆v + k 2 v = δ(x − ξ)δ(y − η) on − ∞ < x < ∞, 0 < y < b,
with
vy (x, 0) = vy (x, b) = 0.
We assume that the solution has the form
∞
1 X  nπy 
v(x, y) = c0 (x) + cn (x) cos , (40.1)
2 n=1
b

1690
and apply a finite cosine transform in the y direction. Integrating from 0 to b yields
Z b Z b
2
vxx + vyy + k v dy = δ(x − ξ)δ(y − η) dy,
0 0

 b
Z b
vy 0 + vxx + k 2 v dy = δ(x − ξ),
0

Z b
vxx + k 2 v dy = δ(x − ξ).
0

Substituting in Equation 40.1 and using the orthogonality of the cosines gives us

2
c000 (x) + k 2 c0 (x) = δ(x − ξ).
b
Multiplying by cos(nπy/b) and integrating form 0 to b yields
Z b  nπy  Z b  nπy 
2


vxx + vyy + k v cos dy = δ(x − ξ)δ(y − η) cos dy.
0 b 0 b
The vyy term becomes
Z b  nπy   nπy ib Z b
h nπ  nπy 
vyy cos dy = vy cos vy sin − dy −
0 b b 0 0 b b
h nπ  nπy ib Z b  nπ 2  nπy 
= v sin − v cos dy.
b b 0 0 b b
The right-hand-side becomes
Z b  nπy   nπη 
δ(x − ξ)δ(y − η) cos dy = δ(x − ξ) cos .
0 b b

1691
Thus the partial differential equation becomes
Z b  nπ 2   nπy   nπη 
2
vxx − v + k v cos dy = δ(x − ξ) cos .
0 b b b
Substituting in Equation 40.1 and using the orthogonality of the cosines gives us
  nπ 2 
00 2 2  nπη 
cn (x) + k − cn (x) = δ(x − ξ) cos .
b b b

Now we need to solve for the coefficients in the expansion of v(x, y). The homogeneous solutions for c0 (x) are
±ikx
e . The solution for u(x, y, t) must satisfy the radiation condition. The waves at x = −∞ travel to the left and
the waves at x = +∞ travel to the right. The two solutions of that will satisfy these conditions are, respectively,

y1 = e−ikx , y2 = eikx .

The Wronskian of these two solutions is 2ik. Thus the solution for c0 (x) is

e−ikx< eikx>
c0 (x) =
ibk
We need to consider three cases for the equation for cn .
p
k > nπ/b Let α = k 2 − (nπ/b)2 . The homogeneous solutions that satisfy the radiation condition are

y1 = e−iαx , y2 = eiαx .

The Wronskian of the two solutions is 2iα. Thus the solution is

e−iαx< eiαx>  nπη 

cn (x) = cos .
ibα b
nπη


In the case that cos b
= 0 this reduces to the trivial solution.

1692
k = nπ/b The homogeneous solutions that are bounded at infinity are

y1 = 1, y2 = 1.

If the right-hand-side is nonzero there is no way to combine these solutions to satisfy both the continuity
nπη
6 0 there is no bounded solution. If cos nπη



and the derivative jump conditions. Thus if cos b = b
=0
then the solution is not unique.

cn (x) = const.
p
k < nπ/b Let β = (nπ/b)2 − k 2 . The homogeneous solutions that are bounded at infinity are

y1 = eβx , y2 = e−βx .

The Wronskian of these solutions is −2β. Thus the solution is

eβx< e−βx>  nπη 

cn (x) = − cos
bβ b
nπη


In the case that cos b
= 0 this reduces to the trivial solution.

1693
40.1 Exercises
Exercise 40.1
A slab is perfectly insulated at the surface x = 0 and has a specified time varying temperature f (t) at the surface
x = L. Initially the temperature is zero. Find the temperature u(x, t) if the heat conductivity in the slab is κ = 1.

Exercise 40.2
Solve

uxx + uyy = 0, 0 < x < L, y > 0,

u(x, 0) = f (x), u(0, y) = g(y), u(L, y) = h(y),

with an eigenfunction expansion.

1694
40.2 Hints
Hint 40.1

Hint 40.2

1695
40.3 Solutions
Solution 40.1
The problem is

ut = uxx , 0 < x < L, t > 0,

ux (0, t) = 0, u(L, t) = f (t), u(x, 0) = 0.

We will solve this problem with an eigenfunction expansion. We find these eigenfunction by replacing the inho-
mogeneous boundary condition with the homogeneous one, u(L, t) = 0. We substitute the separation of variables
v(x, t) = X(x)T (t) into the homogeneous partial differential equation.

XT 0 = X 00 T
T0 X 00
= = −λ2 .
T X
This gives us the regular Sturm-Liouville eigenvalue problem,

X 00 = −λ2 X, X 0 (0) = X(L) = 0,

which has the solutions,

 
π(2n − 1) π(2n − 1)x
λn = , Xn = cos , n ∈ N.
2L 2L
Our solution for u(x, t) will be an eigenfunction expansion in these eigenfunctions. Since the inhomogeneous
boundary condition is a function of t, the coefficients will be functions of t.
∞
X
u(x, t) = an (t) cos(λn x)
n=1

Since u(x, t) does not satisfy the homogeneous boundary conditions of the eigenfunctions, the series is not u-
niformly convergent and we are not allowed to differentiate it with respect to x. We substitute the expansion

1696
into the partial differential equation, multiply by the eigenfunction and integrate from x = 0 to x = L. We use
integration by parts to move derivatives from u to the eigenfunctions.

ut = uxx
Z L Z L
ut cos(λm x) dx = uxx cos(λm x) dx
0 0
∞
!
Z L X Z L
a0n (t) cos(λn x) cos(λm x) dx = [ux cos(λm x)]L0 + ux λm sin(λm x) dx
0 n=1 0
Z L
L 0 L
a (t) = [uλm sin(λm x)]0 − uλ2m cos(λm x) dx
2 m 0
∞
Z L X !
L 0
a (t) = λm u(L, t) sin(λm L) − λ2m an (t) cos(λn x) cos(λm x) dx
2 m 0 n=1
L 0 L
am (t) = λm (−1)n f (t) − λ2m am (t)
2 2
0 2 n
am (t) + λm am (t) = (−1) λm f (t)

From the initial condition u(x, 0) = 0 we see that am (0) = 0. Thus we have a first order differential equation and
an initial condition for each of the am (t).

a0m (t) + λ2m am (t) = (−1)n λm f (t), am (0) = 0

This equation has the solution,

Z t
2
n
am (t) = (−1) λm e−λm (t−τ ) f (τ ) dτ.
0

1697
Solution 40.2

uxx + uyy = 0, 0 < x < L, y > 0,

u(x, 0) = f (x), u(0, y) = g(y), u(L, y) = h(y),

We seek a solution of the form,

∞
X  nπx 
u(x, y) = un (y) sin .
n=1
L

Since we have inhomogeneous boundary conditions at x = 0, L, we cannot differentiate the series representation
with respect to x. We multiply Laplace’s equation by the eigenfunction and integrate from x = 0 to x = L.
Z L  mπx 
(uxx + uyy ) sin dx = 0
0 L

We use integration by parts to move derivatives from u to the eigenfunctions.

mπ L
 mπx iL Z  mπx 
h L
ux sin − ux cos dx + u00m (y) = 0
L 0 L 0 L 2
h mπ  mπx iL  mπ 2 Z L  mπx  L
− u cos − u sin dx + u00m (y) = 0
L L 0 L 0 L 2
mπ mπ L mπ
 2 L
− h(y)(−1)m + g(y) − um (y) + u00m (y) = 0
L L 2 L 2
 mπ 2
00 m
um (y) − um (y) = 2mπ ((−1) h(y) − g(y))
L
Now we have an ordinary differential equation for the un (y). In order that the solution is bounded, we require
that each un (y) is bounded as y → ∞. We use the boundary condition u(x, 0) = f (x) to determine boundary

1698
conditions for the um (y) at y = 0.
∞
X  nπx 
u(x, 0) = un (0) sin = f (x)
n=1
L
Z L
2  nπx 
un (0) = fn ≡ f (x) sin dx
L 0 L

Thus we have the problems,

 nπ 2
u00n (y) − un (y) = 2nπ ((−1)n h(y) − g(y)) , un (0) = fn , un (+∞) bounded,
L
for the coefficients in the expansion. We will solve these with Green functions. Consider the associated Green
function problem
 nπ 2
G00n (y; η) − Gn (y; η) = δ(y − η), Gn (0; η) = 0, Gn (+∞; η) bounded.
L
The homogeneous solutions that satisfy the boundary conditions are
 nπy 
sinh and e−nπy/L ,
L
respectively. The Wronskian of these solutions is

 sinh nπy


−nπy/L 
e nπ −2nπy/L
 sinh nπy − nπ e−nπy/L  = − L e
 nπ L
 .
L L L

Thus the Green function is

L sinh nπy <

−nπy /L
L
e >

Gn (y; η) = − .
nπ e−2nπη/L

1699
Using the Green function we determine the un (y) and thus the solution of Laplace’s equation.
Z ∞
−nπy/L
un (y) = fn e + 2nπ Gn (y; η) ((−1)n h(η) − g(η)) dη
0
∞
X  nπx 
u(x, y) = un (y) sin .
n=1
L

1700
Chapter 41

Waves

1701
41.1 Exercises
Exercise 41.1
Sketch the solution to the wave equation:
Z x+ct
1 1
u(x, t) = (u(x + ct, 0) + u(x − ct, 0)) + ut (τ, 0) dτ,
2 2c x−ct

for various values of t corresponding to the initial conditions:

1. u(x, 0) = 0, ut (x, 0) = sin ωx where ω is a constant,

1
 for 0 < x < 1
2. u(x, 0) = 0, ut (x, 0) = −1 for − 1 < x < 0

0 for |x| > 1.


Exercise 41.2
1. Consider the solution of the wave equation for u(x, t):
utt = c2 uxx
on the infinite interval −∞ < x < ∞ with initial displacement of the form
(
h(x) for x > 0,
u(x, 0) =
−h(−x) for x < 0,
and with initial velocity
ut (x, 0) = 0.
Show that the solution of the wave equation satisfying these initial conditions also solves the following semi-
infinite problem: Find u(x, t) satisfying the wave equation utt = c2 uxx in 0 < x < ∞, t > 0, with initial
conditions u(x, 0) = h(x), ut (x, 0) = 0, and with the fixed end condition u(0, t) = 0. Here h(x) is any given
function with h(0) = 0.

1702
 2. Use a similar idea to explain how you could use the general solution of the wave equation to solve the finite
interval problem (0 < x < l) in which u(0, t) = u(l, t) = 0 for all t, with u(x, 0) = h(x) and ut (x, 0) = 0.
Take h(0) = h(l) = 0.

Exercise 41.3
The deflection u(x, T ) = φ(x) and velocity ut (x, T ) = ψ(x) for an infinite string (governed by utt = c2 uxx ) are
measured at time T , and we are asked to determine what the initial displacement and velocity profiles u(x, 0) and
ut (x, 0) must have been. An alert AMa95c student suggests that this problem is equivalent to that of determining
the solution of the wave equation at time T when initial conditions u(x, 0) = φ(x), ut (x, 0) = −ψ(x) are prescribed.
Is she correct? If not, can you rescue her idea?

Exercise 41.4
In obtaining the general solution of the wave equation the interval was chosen to be infinite in order to simplify
the evaluation of the functions α(ξ) and β(ξ) in the general solution

u(x, t) = α(x + ct) + β(x − ct).

But this general solution is in fact valid for any interval be it infinite or finite. We need only choose appropriate
functions α(ξ), β(ξ) to satisfy the appropriate initial and boundary conditions. This is not always convenient but
there are other situations besides the solution for u(x, t) in an infinite domain in which the general solution is of
use. Consider the “whip-cracking” problem (this is not meant to be a metaphor for AMa95c):

utt = c2 uxx ,

(with c a constant) in the domain x > 0, t > 0 with initial conditions

u(x, 0) = ut (x, 0) = 0 x > 0,

and boundary conditions

u(0, t) = γ(t)

1703
prescribed for all t > 0. Here γ(0) = 0. Find α and β so as to determine u for x > 0, t > 0.

Hint: (From physical considerations conclude that you can take α(ξ) = 0. Your solution will corroborate
this.) Use the initial conditions to determine α(ξ) and β(ξ) for ξ > 0. Then use the initial condition to determine
β(ξ) for ξ < 0.

Exercise 41.5
Let u(x, t) satisfy the equation

utt = c2 uxx ;

(with c a constant) in some region of the (x, t) plane.

1. Show that the quantity (ut − cux ) is constant along each straight line defined by x − ct = constant, and
that (ut + cux ) is constant along each straight line of the form x + ct = constant. These straight lines are
called characteristics; we will refer to typical members of the two families as C+ and C− characteristics,
respectively. Thus the line x − ct = constant is a C+ characteristic.

2. Let u(x, 0) and ut (x, 0) be prescribed for all values of x in −∞ < x < ∞, and let (x0 , t0 ) be some point in
the (x, t) plane, with t0 > 0. Draw the C+ and C− characteristics through (x0 , t0 ) and let them intersect
the x-axis at the points A,B. Use the properties of these curves derived in part (a) to determine ut (x0 , t0 )
in terms of initial data at points A and B. Using a similar technique to obtain ut (x0 , τ ) with 0 < τ < t,
determine u(x0 , t0 ) by integration with respect to τ , and compare this with the solution derived in class:
Z x+ct
1 1
u(x, t) = (u(x + ct, 0) + u(x − ct, 0)) + ut (τ, 0)dτ.
2 2c x−ct

Observe that this “method of characteristics” again shows that u(x0 , t0 ) depends only on that part of the
initial data between points A and B.

1704
Exercise 41.6
The temperature u(x, t) at a depth x below the Earth’s surface at time t satisfies

ut = κuxx .

The surface x = 0 is heated by the sun according to the periodic rule:

u(0, t) = T cos(ωt).

Seek a solution of the form

u(x, t) = < A eiωt−αx .

a) Find u(x, t) satisfying u → 0 as x → +∞, (i.e. deep into the Earth).

b) Find the temperature variation at a fixed depth, h, below the surface.

c) Find the phase lag δ(x) such that when the maximum temperature occurs at t0 on the surface, the maximum
at depth x occurs at t0 + δ(x).

d) Show that the seasonal, (i.e. yearly), temperature changes and daily temperature changes penetrate to
depths in the ratio:
xyear √
= 365,
xday

where xyear and xday are the depths of same temperature variation caused by the different periods of the
source.

Exercise 41.7
An infinite cylinder of radius a produces an external acoustic pressure field u satisfying:

utt = c2 δu,

1705
by a pure harmonic oscillation of its surface at r = a. That is, it moves so that

u(a, θ, t) = f (θ) eiωt

where f (θ) is a known function. Note that the waves must be outgoing at infinity, (radiation condition at infinity).
Find the solution, u(r, θ, t). We seek a periodic solution of the form,

u(r, θ, t) = v(r, θ) eiωt .

Exercise 41.8
Plane waves are incident on a “soft” cylinder of radius a whose axis is parallel to the plane of the waves. Find the
field scattered by the cylinder. In particular, examine the leading term of the solution when a is much smaller
than the wavelength of the incident waves. If v(x, y, t) is the scattered field it must satisfy:

Wave Equation: vtt = c2 ∆v, x2 + y 2 > a2 ;

Soft Cylinder: v(x, y, t) = − ei(ka cos θ−ωt , on r = a, 0 ≤ θ < 2π;
Scattered: v is outgoing as r → ∞.

Here k = ω/c. Use polar coordinates in the (x, y) plane.

Exercise 41.9
Consider the flow of electricity in a transmission line. The current, I(x, t), and the voltage, V (x, t), obey the
telegrapher’s system of equations:

−Ix = CVt + GV,

−Vx = LIt + RI,

where C is the capacitance, G is the conductance, L is the inductance and R is the resistance.
a) Show that both I and V satisfy a damped wave equation.

1706
 b) Find the relationship between the physical constants, C, G, L and R such that there exist damped traveling
wave solutions of the form:

V (x, t) = e−γt (f (x − at) + g(x + at)).

What is the wave speed?

1707
41.2 Hints
Hint 41.1

Hint 41.2

Hint 41.3

Hint 41.4
From physical considerations conclude that you can take α(ξ) = 0. Your solution will corroborate this. Use the
initial conditions to determine α(ξ) and β(ξ) for ξ > 0. Then use the initial condition to determine β(ξ) for ξ < 0.

Hint 41.5

Hint 41.6

a) Substitute u(x, t) = <(A eiωt−αx ) into the partial differential equation and solve for α. Assume that α has
positive real part so that the solution vanishes as x → +∞.

1708
Hint 41.7
Seek a periodic solution of the form,

u(r, θ, t) = v(r, θ) eiωt .

Solve the Helmholtz equation for v with a Fourier series expansion,

∞
X
v(r, θ) = vn (r) einθ .
n=−∞

You will find that the vn satisfy Bessel’s equation. Choose the vn so that u satisfies the boundary condition at
r = a and the radiation condition at infinity.
The Bessel functions have the asymptotic behavior,
r
2
Jn (ρ) ∼ cos(ρ − nπ/2 − π/4), as ρ → ∞,
πρ
r
2
Yn (ρ) ∼ sin(ρ − nπ/2 − π/4), as ρ → ∞,
πρ
r
(1) 2 i(ρ−nπ/2−π/4)
Hn (ρ) ∼ e , as ρ → ∞,
πρ
r
(2) 2 −i(ρ−nπ/2−π/4)
Hn (ρ) ∼ e , as ρ → ∞.
πρ

Hint 41.8

Hint 41.9

1709
41.3 Solutions
Solution 41.1
1.

1 x+ct
Z
1
u(x, t) = (u(x + ct, 0) + u(x − ct, 0)) + ut (τ, 0) dτ
2 2c x−ct
1 x+ct
Z
u(x, t) = sin(ωτ ) dτ
2c x−ct
sin(ωx) sin(ωct)
u(x, t) =
ωc
Figure 41.1 shows the solution for c = 1 and ω = 1/10.

2. We can write the initial velocity in terms of the Heaviside function.


1
 for 0 < x < 1
ut (x, 0) = −1 for − 1 < x < 0

0 for |x| > 1.


ut (x, 0) = −H(x + 1) + 2H(x) − H(x − 1)

We integrate the Heaviside function.


Z b 0
 for b < c
H(x − c) dx = b − a for a > c
a 
b − c otherwise


If a < b, we can express this as

Z b
H(x − c) dx = min(b − a, max(b − c, 0)).
a

1710
 10
5
u 60
0
-5
-10 40
-50 t
20
0
x
50 0

Figure 41.1: Solution of the wave equation.

Now we find an expression for the solution.

1 x+ct
Z
1
u(x, t) = (u(x + ct, 0) + u(x − ct, 0)) + ut (τ, 0) dτ
2 2c x−ct
1 x+ct
Z
u(x, t) = (−H(τ + 1) + 2H(τ ) − H(τ − 1)) dτ
2c x−ct
u(x, t) = − min(2ct, max(x + ct + 1, 0)) + 2 min(2ct, max(x + ct, 0)) − min(2ct, max(x + ct − 1, 0))

Figure 41.2 shows the solution for c = 1.

1711
 1
0.5 4
u
0
-0.5 3
-1
-4 2 t
-2
0 1
x 2
4 0

Figure 41.2: Solution of the wave equation.

Solution 41.2
1. The solution on the interval (−∞ . . . ∞) is
1
u(x, t) = (h(x + ct) + h(x − ct)).
2
Now we solve the problem on (0 . . . ∞). We define the odd extension of h(x).
(
h(x) for x > 0,
ĥ(x) = = sign (x)h(|x|)
−h(−x) for x < 0,
Note that
d
ĥ0 (0− ) = (−h(−x))x→0+ = h0 (0+ ) = ĥ0 (0+ ).

dx

1712
 Thus ĥ(x) is piecewise C 2 . Clearly

1
u(x, t) = (ĥ(x + ct) + ĥ(x − ct))
2

satisfies the differential equation on (0 . . . ∞). We verify that it satisfies the initial condition and boundary
condition.

1
u(x, 0) = (ĥ(x) + ĥ(x)) = h(x)
2
1 1
u(0, t) = (ĥ(ct) + ĥ(−ct)) = (h(ct) − h(ct)) = 0
2 2

2. First we define the odd extension of h(x) on the interval (−l . . . l).

ĥ(x) = sign (x)h(|x|), x ∈ (−l . . . l)

Then we form the odd periodic extension of h(x) defined on (−∞ . . . ∞).
     
x+l  x + l 
ĥ(x) = sign x − 2l h x − 2l  , x ∈ (−∞ . . . ∞)
2l 2l 

We note that ĥ(x) is piecewise C 2 . Also note that ĥ(x) is odd about the points x = nl, n ∈ Z. That is,
ĥ(nl − x) = −ĥ(nl + x). Clearly

1
u(x, t) = (ĥ(x + ct) + ĥ(x − ct))
2

satisfies the differential equation on (0 . . . l). We verify that it satisfies the initial condition and boundary

1713
 conditions.
1
u(x, 0) = (ĥ(x) + ĥ(x))
2
u(x, 0) = ĥ(x)
     
x+l  x + l 
u(x, 0) = sign x − 2l h x − 2l 
2l 2l 
u(x, 0) = h(x)
1 1
u(0, t) = (ĥ(ct) + ĥ(−ct)) = (ĥ(ct) − ĥ(ct)) = 0
2 2
1 1
u(l, t) = (ĥ(l + ct) + ĥ(l − ct)) = (ĥ(l + ct) − ĥ(l + ct)) = 0
2 2

Solution 41.3
Change of Variables. Let u(x, t) be the solution of the problem with deflection u(x, T ) = φ(x) and velocity
ut (x, T ) = ψ(x). Define
v(x, τ ) = u(x, T − τ ).
We note that u(x, 0) = v(x, T ). v(τ ) satisfies the wave equation.
vτ τ = c2 vxx
The initial conditions for v are
v(x, 0) = u(x, T ) = φ(x), vτ (x, 0) = −ut (x, T ) = −ψ(x).
Thus we see that the student was correct.
Direct Solution. D’Alembert’s solution is valid for all x and t. We formally substitute t − T for t in this
solution to solve the problem with deflection u(x, T ) = φ(x) and velocity ut (x, T ) = ψ(x).
1 x+c(t−T )
Z
1
u(x, t) = (φ(x + c(t − T )) + φ(x − c(t − T ))) + ψ(τ ) dτ
2 2c x−c(t−T )

1714
This satisfies the wave equation, because the equation is shift-invariant. It also satisfies the initial conditions.
1 x
Z
1
u(x, T ) = (φ(x) + φ(x)) + ψ(τ ) dτ = φ(x)
2 2c x
1 1
ut (x, t) = (cφ0 (x + c(t − T )) − cφ0 (x − c(t − T ))) + (ψ(x + c(t − T )) + ψ(x − c(t − T )))
2 2
1 1
ut (x, T ) = (cφ0 (x) − cφ0 (x)) + (ψ(x) + ψ(x)) = ψ(x)
2 2

Solution 41.4
Since the solution is a wave moving to the right, we conclude that we could take α(ξ) = 0. Our solution will
corroborate this.
The form of the solution is
u(x, t) = α(x + ct) + β(x − ct).
We substitute the solution into the initial conditions.
u(x, 0) = α(ξ) + β(ξ) = 0, ξ > 0
ut (x, 0) = cα0 (ξ) − cβ 0 (ξ) = 0, ξ > 0
We integrate the second equation to obtain the system
α(ξ) + β(ξ) = 0, ξ > 0,
α(ξ) − β(ξ) = 2k, ξ > 0,
which has the solution
α(ξ) = k, β(ξ) = −k, ξ > 0.
Now we substitute the solution into the initial condition.
u(0, t) = α(ct) + β(−ct) = γ(t), t > 0
α(ξ) + β(−ξ) = γ(ξ/c), ξ > 0
β(ξ) = γ(−ξ/c) − k, ξ < 0

1715
This determines u(x, t) for x > 0 as it depends on α(ξ) only for ξ > 0. The constant k is arbitrary. Changing k
does not change u(x, t). For simplicity, we take k = 0.

u(x, t) = β(x − ct)

(
0 for x − ct < 0
u(x, t) =
γ(t − x/c) for x − ct > 0
u(x, t) = γ(t − x/c)H(ct − x)

Solution 41.5
1. We write the value of u along the line x − ct = k as a function of t: u(k + ct, t). We differentiate ut − cux
with respect to t to see how the quantity varies.
d
(ut (k + ct, t) − cux (k + ct, t)) = cuxt + utt − c2 uxx − cuxt
dt
= utt − c2 uxx
=0

Thus ut − cux is constant along the line x − ct = k. Now we examine ut + cux along the line x + ct = k.
d
(ut (k − ct, t) + cux (k − ct, t)) = −cuxt + utt − c2 uxx + cuxt
dt
= utt − c2 uxx
=0

ut + cux is constant along the line x + ct = k.

2. From part (a) we know

ut (x0 , t0 ) − cux (x0 , t0 ) = ut (x0 − ct0 , 0) − cux (x0 − ct0 , 0)

ut (x0 , t0 ) + cux (x0 , t0 ) = ut (x0 + ct0 , 0) + cux (x0 + ct0 , 0).

1716
We add these equations to find ut (x0 , t0 ).

1
ut (x0 , t0 ) = (ut (x0 − ct0 , 0) − cux (x0 − ct0 , 0)ut (x0 + ct0 , 0) + cux (x0 + ct0 , 0))
2
Since t0 was arbitrary, we have

1
ut (x0 , τ ) = (ut (x0 − cτ, 0) − cux (x0 − cτ, 0)ut (x0 + cτ, 0) + cux (x0 + cτ, 0))
2
for 0 < τ < t0 . We integrate with respect to τ to determine u(x0 , t0 ).
Z t0
1
u(x0 , t0 ) = u(x0 , 0) + (ut (x0 − cτ, 0) − cux (x0 − cτ, 0)ut (x0 + cτ, 0) + cux (x0 + cτ, 0)) dτ
0 2
1 t0
Z
= u(x0 , 0) + (−cux (x0 − cτ, 0) + cux (x0 + cτ, 0)) dτ
2 0
1 t0
Z
+ (ut (x0 − cτ, 0) + ut (x0 + cτ, 0)) dτ
2 0
1
= u(x0 , 0) + (u(x0 − ct0 , 0) − u(x0 , 0) + u(x0 + ct0 , 0) − u(x0 , 0))
Z 2
1 x0 −ct0 1 x0 +ct0
Z
+ −ut (τ, 0) dτ + ut (τ, 0) dτ
2c x0 2c x0
1 x0 +ct0
Z
1
= (u(x0 − ct0 , 0) + u(x0 + ct0 , 0)) + ut (τ, 0) dτ
2 2c x0 −ct0

We have D’Alembert’s solution.

Z x+ct
1 1
u(x, t) = (u(x − ct, 0) + u(x + ct, 0)) + ut (τ, 0) dτ
2 2c x−ct

1717
Solution 41.6

a) We substitute u(x, t) = A eiωt−αx into the partial differential equation and take the real part as the solution.
We assume that α has positive real part so the solution vanishes as x → +∞.
iωA eiωt−αx = κα2 A eiωt−αx
iω = κα2
r
ω
α = (1 + i)
2κ
A solution of the partial differential equation is,
  r 
ω
u(x, t) = < A exp iωt − (1 + i) x ,
2κ
 r   r 
ω ω
u(x, t) = A exp − x cos ωt − x .
2κ 2κ
Applying the initial condition, u(0, t) = T cos(ωt), we obtain,
 r   r 
ω ω
u(x, t) = T exp − x cos ωt − x .
2κ 2κ

b) At a fixed depth x = h, the temperature is

 r   r 
ω ω
u(h, t) = T exp − h cos ωt − h .
2κ 2κ
Thus the temperature variation is
 r   r 
ω ω
−T exp − h ≤ u(h, t) ≤ T exp − h .
2κ 2κ

1718
c) The
p solution is√ an exponentially decaying, traveling wave that propagates into the Earth with speed
ω/ ω/(2κ) = 2κω. More generally, the wave

e−bt cos(ωt − ax)

travels in the positive direction with speed ω/a. Figure 41.3 shows such a wave for a sequence of times.

Figure 41.3: An Exponentially Decaying, Traveling Wave

The phase lag, δ(x) is the time that it takes for the wave to reach a depth of x. It satisfies,
r
ω
ωδ(x) − x = 0,
2κ

x
δ(x) = √ .
2κω

d) Let ωyear be the frequency for annual temperature variation, then ωday = 365ωyear . If xyear is the depth that
a particular yearly temperature variation reaches and xday is the depth that this same variation in daily

1719
 temperature reaches, then
 r   r 
ωyear ωday
exp − xyear = exp − xday ,
2κ 2κ

r r
ωyear ωday
xyear = xday ,
2κ 2κ

xyear √
= 365.
xday

Solution 41.7
We seek a periodic solution of the form,

u(r, θ, t) = v(r, θ) eiωt .

Substituting this into the wave equation will give us a Helmholtz equation for v.

−ω 2 v = c2 ∆v
1 1 ω2
vrr + vr + 2 vθθ + 2 v = 0
r r c
We have the boundary condition v(a, θ) = f (θ) and the radiation condition at infinity. We expand v in a Fourier
series in θ in which the coefficients are functions of r. You can check that einθ are the eigenfunctions obtained
with separation of variables.
∞
X
v(r, θ) = vn (r) einθ
n=−∞

1720
We substitute this expression into the Helmholtz equation to obtain ordinary differential equations for the coeffi-
cients vn .
∞   2
n2
 
X
00 1 0 ω
vn + vn + 2
− 2 vn einθ = 0
n=−∞
r c r

The differential equations for the vn are

ω 2 n2
 
1
vn00 + vn0 + − 2 vn = 0.
r c2 r

which has as linearly independent solutions the Bessel and Neumann functions,
 ωr   ωr 
Jn , Yn ,
c c
or the Hankel functions,
 ωr   ωr 
Hn(1) , Hn(2) .
c c
The functions have the asymptotic behavior,
r
2
Jn (ρ) ∼ cos(ρ − nπ/2 − π/4), as ρ → ∞,
πρ
r
2
Yn (ρ) ∼ sin(ρ − nπ/2 − π/4), as ρ → ∞,
πρ
r
(1) 2 i(ρ−nπ/2−π/4)
Hn (ρ) ∼ e , as ρ → ∞,
πρ
r
(2) 2 −i(ρ−nπ/2−π/4)
Hn (ρ) ∼ e , as ρ → ∞.
πρ

1721
u(r, θ, t) will be an outgoing wave at infinity if it is the sum of terms of the form ei(ωt−constr) . Thus the vn must
have the form
 ωr 
vn (r) = bn Hn(2)
c
for some constants, bn . The solution for v(r, θ) is
∞
X  ωr 
v(r, θ) = bn Hn(2) einθ .
n=−∞
c

We determine the constants bn from the boundary condition at r = a.

∞
(2) ωa
X  
v(a, θ) = b n Hn einθ = f (θ)
n=−∞
c
Z 2π
1
bn = (2)
f (θ) e−inθ dθ
2πHn (ωa/c) 0
∞
X  ωr 
iωt
u(r, θ, t) = e bn Hn(2) einθ
n=−∞
c

Solution 41.8
We substitute the form v(x, y, t) = u(r, θ) e−iωt into the wave equation to obtain a Helmholtz equation.
c2 ∆u + ω 2 u = 0
1 1
urr + ur + 2 uθθ + k 2 u = 0
r r
We solve the Helmholtz equation with separation of variables. We expand u in a Fourier series.
∞
X
u(r, θ) = un (r) einθ
n=−∞

1722
We substitute the sum into the Helmholtz equation to determine ordinary differential equations for the coefficients.
n2
 
00 1 0 2
un + un + k − 2 un = 0
r r
This is Bessel’s equation, which has as solutions the Bessel and Neumann functions, {Jn (kr), Yn (kr)} or the
(1) (2)
Hankel functions, {Hn (kr), Hn (kr)}.
Recall that the solutions of the Bessel equation have the asymptotic behavior,
r
2
Jn (ρ) ∼ cos(ρ − nπ/2 − π/4), as ρ → ∞,
πρ
r
2
Yn (ρ) ∼ sin(ρ − nπ/2 − π/4), as ρ → ∞,
πρ
r
(1) 2 i(ρ−nπ/2−π/4)
Hn (ρ) ∼ e , as ρ → ∞,
πρ
r
2 −i(ρ−nπ/2−π/4)
Hn(2) (ρ) ∼ e , as ρ → ∞.
πρ
From this we see that only the Hankel function of the first kink will give us outgoing waves as ρ → ∞. Our
solution for u becomes,
∞
X
u(r, θ) = bn Hn(1) (kr) einθ .
n=−∞

We determine the coefficients in the expansion from the boundary condition at r = a.

∞
X
u(a, θ) = bn Hn(1) (ka) einθ = − eika cos θ
n=−∞
Z 2π
1
bn = − (1)
eika cos θ e−inθ dθ
2πHn (ka) 0

1723
We evaluate the integral with the identities,
Z 2π
1
Jn (x) = eix cos θ einθ dθ,
2πin 0
J−n (x) = (−1)n Jn (x).

Thus we obtain,
∞
X (−i)n Jn (ka)
u(r, θ) = − (1)
Hn(1) (kr) einθ .
n=−∞ Hn (ka)

When a  1/k, i.e. ka  1, the Bessel function has the behavior,

(ka/2)n
Jn (ka) ∼ .
n!
In this case, the n 6= 0 terms in the sum are much smaller than the n = 0 term. The approximate solution is,
(1)
H0 (kr)
u(r, θ) ∼ − (1)
,
H0 (ka)
(1)
H0 (kr)
v(r, θ, t) ∼ − (1)
e−iωt .
H0 (ka)

Solution 41.9

a)
(
−Ix = CVt + GV,
−Vx = LIt + RI

1724
First we derive a single partial differential equation for I. We differentiate the two partial differential equations
with respect to x and t, respectively and then eliminate the Vxt terms.
(
−Ixx = CVtx + GVx ,
−Vxt = LItt + RIt
−Ixx + LCItt + RCIt = GVx

We use the initial set of equations to write Vx in terms of I.

−Ixx + LCItt + RCIt + G(LIt + RI) = 0

RC + GL GR 1
Itt + It + I− Ixx = 0
LC LC LC

Now we derive a single partial differential equation for V . We differentiate the two partial differential equations
with respect to t and x, respectively and then eliminate the Ixt terms.
(
−Ixt = CVtt + GVt ,
−Vxx = LItx + RIx
−Vxx = RIx − LCVtt − LGVt

We use the initial set of equations to write Ix in terms of V .

LCVtt + LGVt − Vxx + R(CVt + GV ) = 0

RC + LG RG 1
Vtt + Vt + V − Vxx = 0.
LC LC LC

Thus we see that I and V both satisfy the same damped wave equation.

1725
 b) We substitute V (x, t) = e−γt (f (x − at) + g(x + at)) into the damped wave equation for V .
   
RC + LG RG −γt RC + LG
2
γ − γ+ e (f + g) + −2γ + a e−γt (−f 0 + g 0 )
LC LC LC
1 −γt 00
+ a2 e−γt (f 00 + g 00 ) − e (f + g 00 ) = 0
LC
Since f and g are arbitrary functions, the coefficients of e−γt (f + g), e−γt (−f 0 + g 0 ) and e−γt (f 00 + g 00 ) must vanish.
This gives us three constraints.
1 RC + LG RC + LG RG
a2 − = 0, −2γ + = 0, γ 2 − γ+ =0
LC LC LC LC
√
The first equation determines the wave speed to be a = 1/ LC. We substitute the value of γ from the second
equation into the third equation.
RC + LG RG
γ= , −γ 2 + =0
2LC LC
In order for damped waves to propagate, the physical constants must satisfy,
 2
RG RC + LG
− = 0,
LC 2LC
4RGLC − (RC + LG)2 = 0,
(RC − LG)2 = 0,
RC = LG.

1726
Chapter 42

The Diffusion Equation

1727
42.1 Exercises
Exercise 42.1
Derive the heat equation for a general 3 dimensional body, with non-uniform density ρ(x), specific heat c(x), and
conductivity k(x). Show that

∂u(x, t) 1
= ∇ · (k∇u(x, t))
∂t cρ

where u is the temperature, and you may assume there are no internal sources or sinks.

Exercise 42.2
Verify Duhamel’s Principal: If u(x, t, τ ) is the solution of the initial value problem:

ut = κuxx , u(x, 0, τ ) = f (x, τ ),

then the solution of

wt = κwxx + f (x, t), w(x, 0) = 0

is
Z t
w(x, t) = u(x, t − τ, τ ) dτ.
0

Exercise 42.3
Modify the derivation of the diffusion equation

k
φt = a2 φxx , a2 = , (42.1)
cρ

1728
so that it is valid for diffusion in a non-homogeneous medium for which c and k are functions of x and φ and
so that it is valid for a geometry in which A is a function of x. Show that Equation (42.1) above is in this case
replaced by

cρAφt = (kAφx )x .

Recall that c is the specific heat, k is the thermal conductivity, ρ is the density, φ is the temperature and A is the
cross-sectional area.

1729
42.2 Hints
Hint 42.1

Hint 42.2
Check that the expression for w(x, t) satisfies the partial differential equation and initial condition. Recall that
Z x Z x
∂
h(x, ξ) dξ = hx (x, ξ) dξ + h(x, x).
∂x a a

Hint 42.3

1730
42.3 Solutions
Exercise 42.4
Consider a Region of material, R. Let u be the temperature and φ be the heat flux. The amount of heat energy
in the region is
Z
cρu dx.
R

We equate the rate of change of heat energy in the region with the heat flux across the boundary of the region.
Z Z
d
cρu dx = − φ · n ds
dt R ∂R

We apply the divergence theorem to change the surface integral to a volume integral.
Z Z
d
cρu dx = − ∇ · φ dx
dt R R
Z  
∂u
cρ + ∇ · φ dx = 0
R ∂t

Since the region is arbitrary, the integral must vanish identically.

∂u
cρ = −∇ · φ
∂t

We apply Fourier’s law of heat conduction, φ = −k∇u, to obtain the heat equation.

∂u 1
= ∇ · (k∇u)
∂t cρ

1731
Solution 42.1
We verify Duhamel’s principal by showing that the integral expression for w(x, t) satisfies the partial differential
equation and the initial condition. Clearly the initial condition is satisfied.

Z 0
w(x, 0) = u(x, 0 − τ, τ ) dτ = 0
0

Now we substitute the expression for w(x, t) into the partial differential equation.

t Z t
∂2
Z
∂
u(x, t − τ, τ ) dτ = κ 2 u(x, t − τ, τ ) dτ + f (x, t)
∂t 0 ∂x 0
Z t Z t
u(x, t − t, t) + ut (x, t − τ, τ ) dτ = κ uxx (x, t − τ, τ ) dτ + f (x, t)
0 0
Z t Z t
f (x, t) + ut (x, t − τ, τ ) dτ = κ uxx (x, t − τ, τ ) dτ + f (x, t)
0 0
Z t
(ut (x, t − τ, τ ) dτ − κuxx (x, t − τ, τ )) dτ
0

Since ut (x, t − τ, τ ) dτ − κuxx (x, t − τ, τ ) = 0, this equation is an identity.

Solution 42.2
We equate the rate of change of thermal energy in the segment (α . . . β) with the heat entering the segment

1732
through the endpoints.
Z β
φt cρA dx = k(β, φ(β))A(β)φx (β, t) − k(α, φ(α))A(α)φx (α, t)
α
Z β
φt cρA dx = [kAφx ]βα
α
Z β Z β
φt cρA dx = (kAφx )x dx
α α
Z β
cρAφt − (kAφx )x dx = 0
α

Since the domain is arbitrary, we conclude that

cρAφt = (kAφx )x .

1733
Chapter 43

Similarity Methods

Introduction. Consider the partial differential equation (not necessarily linear)

 
∂u ∂u
F , , u, t, x = 0.
∂t ∂x
Say the solution is
t1/2
 
x
u(x, t) = sin .
t x1/2
Making the change of variables ξ = x/t, f (ξ) = u(x, t), we could rewrite this equation as
f (ξ) = ξ sin ξ −1/2 .

We see now that if we had guessed that the solution of this partial differential equation was only dependent on
powers of x/t we could have changed variables to ξ and f and instead solved the ordinary differential equation
 
df
G , f, ξ = 0.
dξ
By using similarity methods one can reduce the number of independent variables in some PDE’s.

1734
Example 43.0.1 Consider the partial differential equation
∂u ∂u
x +t − u = 0.
∂t ∂x
One way to find a similarity variable is to introduce a transformation to the temporary variables u0 , t0 , x0 , and
the parameter λ.
u = u0 λ
t = t0 λm
x = x0 λ n
where n and m are unknown. Rewriting the partial differential equation in terms of the temporary variables,
∂u0 1−m 0
0 m ∂u 1−n
x0 λ n λ + t λ λ − u0 λ = 0
∂t0 ∂x0
∂u0 ∂u0
x0 0 λ−m+n + t0 0 λm−n − u0 = 0
∂t ∂x
There is a similarity variable if λ can be eliminated from the equation. Equating the coefficients of the powers of
λ in each term,
−m + n = m − n = 0.
This has the solution m = n. The similarity variable, ξ, will be unchanged under the transformation to the
temporary variables. One choice is
t t 0 λn t0
ξ= = 0 m = 0.
x xλ x
Writing the two partial derivative in terms of ξ,
∂ ∂ξ d 1 d
= =
∂t ∂t dξ x dξ
∂ ∂ξ d t d
= =− 2
∂x ∂x dξ x dξ

1735
The partial differential equation becomes
du du
− ξ2 −u=0
dξ dξ
du u
=
dξ 1 − ξ2
Thus we have reduced the partial differential equation to an ordinary differential equation that is much easier to
solve.
Z ξ 
dξ
u(ξ) = exp
1 − ξ2
Z ξ 
1/2 1/2
u(ξ) = exp + dξ
1−ξ 1+ξ
 
1 1
u(ξ) = exp − log(1 − ξ) + log(1 + ξ)
2 2
u(ξ) = (1 − ξ)−1/2 (1 + ξ)1/2
 1/2
1 + t/x
u(x, t) =
1 − t/x
Thus we have found a similarity solution to the partial differential equation. Note that the existence of a similarity
solution does not mean that all solutions of the differential equation are similarity solutions.

Another Method. Another method is to substitute ξ = xα t and determine if there is an α that makes ξ a
similarity variable. The partial derivatives become
∂ ∂ξ d d
= = xα
∂t ∂t dξ dξ
∂ ∂ξ d d
= = αxα−1 t
∂x ∂x dξ dξ

1736
The partial differential equation becomes
du du
xα+1 + αxα−1 t2 − u = 0.
dξ dξ
If there is a value of α such that we can write this equation in terms of ξ, then ξ = xα t is a similarity variable.
If α = −1 then the coefficient of the first term is trivially in terms of ξ. The coefficient of the second term then
becomes −x−2 t2 . Thus we see ξ = x−1 t is a similarity variable.

Example 43.0.2 To see another application of similarity variables, any partial differential equation of the form
 ut ux 
F tx, u, , =0
x t
is equivalent to the ODE
 
du du
F ξ, u, , =0
dξ dξ
where ξ = tx. Performing the change of variables,
1 ∂u 1 ∂ξ du 1 du du
= = x =
x ∂t x ∂t dξ x dξ dξ
1 ∂u 1 ∂ξ du 1 du du
= = t = .
t ∂x t ∂x dξ t dξ dξ
For example the partial differential equation
∂u x ∂u
u + + tx2 u = 0
∂t t ∂x
which can be rewritten
1 ∂u 1 ∂u
u + + txu = 0,
x ∂t t ∂x

1737
is equivalent to
du du
u + + ξu = 0
dξ dξ
where ξ = tx.

1738
43.1 Exercises
Exercise 43.1
With ξ = xα t, find α such that for some function f , φ = f (ξ) is a solution of

φt = a2 φxx .

Find f (ξ) as well.

1739
43.2 Hints
Hint 43.1

1740
43.3 Solutions

Solution 43.1
We write the derivatives of φ in terms of f .

∂ξ ∂
φt = f = xα f 0 = t−1 ξf 0
∂t ∂ξ
∂ξ ∂
φx = f = αxα−1 tf 0
∂x ∂ξ
∂ ∂
φxx = f 0 αxα−1 t + αxα−1 tαxα−1 t f 0


∂x ∂ξ
2 2α−2 2 00 α−2 0
φxx = α x t f + α(α − 1)x tf
−2
α2 ξ 2 f 00 + α(α − 1)ξf 0


φxx = x

We substitute these expressions into the diffusion equation.

ξf 0 = x−2 t α2 ξ 2 f 00 + α(α − 1)ξf 0

1741
In order for this equation to depend only on the variable ξ, we must have α = −2. For this choice we obtain an
ordinary differential equation for f (ξ).

f 0 = 4ξ 2 f 00 + 6ξf 0
f 00 1 3
= −
f0 4ξ 2 2ξ
1 3
log(f 0 ) = − − log ξ + c
4ξ 2
f = c1 ξ −3/2 e−1/(4ξ)
0
Z ξ
f (ξ) = c1 t−3/2 e−1/(4t) dt + c2
Z 1/(2√ξ)
2
f (ξ) = c1 e−t dt + c2
 
1
f (ξ) = c1 erf √ + c2
2 ξ

1742
Chapter 44

Method of Characteristics

44.1 The Method of Characteristics and the Wave Equation

Consider the one dimensional wave equation
utt = c2 uxx .
With the change of variables, v = ux , w = ut , we have the system of equations,
vt − wx = 0,
wt − c2 vx = 0.
We can write this as the matrix equation,
    
v 0 −1 v
+ 2 = 0.
w t −c 0 w x
The eigenvalues and eigenvectors of the matrix are
   
1 1
λ1 = −c, λ2 = c, φ1 = , φ2 = .
c −c

1743
The matrix is diagonalized by the similarity transformation,
   −1   
−c 0 1 1 0 −1 1 1
= .
0 c c −c −c2 0 c −c
We make the change of variables
    
v 1 1 ν
= .
w c −c ω
The partial differential equation becomes
      
1 1 ν 0 −1 1 1 ν
+ 2 = 0.
c −c ω t −c 0 c −c ω x
Now we left multiply by the inverse of the matrix of eigenvectors to obtain
    
ν −c 0 ν
+ = 0.
ω t 0 c ω x
This is two un-coupled partial differential equations of first order with solutions
ν(x, t) = p(x + ct), ω(x, t) = q(x − ct),
where p, q ∈ C 2 are arbitrary functions. Changing variables back to v and w,
v(x, t) = p(x + ct) + q(x − ct), w(x, t) = cp(x + ct) − cq(x − ct).
Since v = ux , w = ut , we have
u = f (x + ct) + g(x − ct).

where f, g ∈ C 2 are arbitrary functions. This is the general solution of the one-dimensional wave equation. Note
that for any given problem, f and g are only determined to whithin an additive constant. For any constant k,
adding k to f and subtracting it from g does not change the solution.
u = (f (x + ct) + k) + (g(x − ct) − k) .

1744
44.2 The Method of Characteristics for an Infinite Domain
Consider the problem
utt = c2 uxx , −∞ < x < ∞, t > 0
u(x, 0) = p(x), ut (x, 0) = q(x).
We know that the solution has the form
u(x, t) = f (x + ct) + g(x − ct). (44.1)
The initial conditions give us the two equations
f (x) + g(x) = p(x), cf 0 (x) − cg 0 (x) = q(x).
We integrate the second equation.
1
f (x) − g(x) = Q(x) + 2k
c
R
Here Q(x) = q(x) dx and k is an arbitrary constant. We solve the system of equations for f and g.
1 1 1 1
f (x) = p(x) + Q(x) + k, g(x) = p(x) − Q(x) − k
2 2c 2 2c
Note that the value of k does not affect the solution, u(x, t). For simplicity we take k = 0. We substitute f and
g into Equation 44.1 to determine the solution.
1 1
u(x, t) = (p(x + ct) + p(x − ct)) + (Q(x + ct) − Q(x − ct))
2 2c
1 x+ct
Z
1
u(x, t) = (p(x + ct) + p(x − ct)) + q(ξ) dξ
2 2c x−ct
1 x+ct
Z
1
u(x, t) = (u(x + ct, 0) + u(x − ct, 0)) + ut (ξ, 0) dξ
2 2c x−ct

1745
44.3 The Method of Characteristics for a Semi-Infinite Domain
Consider the problem
∂2u 2
2∂ u
= c , 0 ≤ x < ∞, t > 0
∂t2 ∂x2
∂u(x, 0)
u(x, 0) = f (x), = 0, u(0, t) = h(t).
∂t
We assume that f (0) = h(0) = u0 . Following the previous example we see that
1
G(ξ) = f (ξ) + k, for ξ > 0
2
1
F (ξ) = f (ξ) − k, for ξ > 0
2
The boundary condition yields
F (−ct) + G(ct) = h(t), for t > 0
F (ξ) + G(−ξ) = h(−ξ/c), for ξ < 0
1
F (ξ) = h(−ξ/c) − f (−ξ) − k, for ξ < 0.
2
Since u(0, 0) = F (0) + G(0) = f (0) = h(0) = u0 ,
1 1
F (ξ) = h(−ξ/c) − f (−ξ) + u0 , for ξ < 0.
2 2
Now F and G are
(
1
+ 12 u0 ,
2
f (ξ) for ξ > 0
F (ξ) =
h(−ξ/c) − 12 f (−ξ) + 12 u0 , for ξ < 0
1 1
G(ξ) = f (ξ) − u0 , for ξ > 0.
2 2

1746
Thus the solution is
(
1
2
[f (x − ct) + f (x + ct)] , for x − ct > 0
u(x, t) = .
h(t − x/c) − 12 f (ct − x) + 12 u0 , for x − ct < 0.

44.4 Envelopes of Curves

Consider the tangent lines to the parabola y = x2 . The slope of the tangent at the point (x, x2 ) is 2x. The set of
tangents form a one parameter family of lines,

f (x, t) = t2 + (x − t)2t = 2tx − t2 .

The parabola and some of its tangents are plotted in Figure 44.1.
1

-1 1

-1

Figure 44.1: A parabola and its tangents.

1747
 The parabola is the envelope of the family of tangent lines. Each point on the parabola is tangent to one of
the lines. Given a curve, we can generate a family of lines that envelope the curve. We can also do the opposite,
given a family of lines, we can determine the curve that they envelope. More generally, given a family of curves,
we can determine the curve that they envelope. Let the one parameter family of curves be given by the equation
F (x, y, t) = 0. For the example of the tangents to the parabola this equation would be y − 2tx + t2 = 0.
Let y(x) be the envelope of F (x, y, t) = 0. Then the points on y(x) must lie on the family of curves. Thus
y(x) must satisfy the equation F (x, y, t) = 0. The points that lie on the envelope have the property,
∂
F (x, y, t) = 0.
∂t
We can solve this equation for t in terms of x and y, t = t(x, y). The equation for the envelope is then

F (x, y, t(x, y)) = 0.

Consider the example of the tangents to the parabola. The equation of the one-parameter family of curves is

F (x, y, t) ≡ y − 2tx + t2 = 0.

The condition Ft (x, y, t) = 0 gives us the constraint,

−2x + 2t = 0.

Solving this for t gives us t(x, y) = x. The equation for the envelope is then,

y − 2xx + x2 = 0,

y = x2 .

Example 44.4.1 Consider the one parameter family of curves,

(x − t)2 + (y − t)2 − 1 = 0.

1748
These are circles of unit radius and center (t, t). To determine the envelope of the family, we first use the constraint
Ft (x, y, t) to solve for t(x, y).

Ft (x, y, t) = −2(x − t) − 2(y − t) = 0

x+y
t(x, y) =
2
Now we substitute this into the equation F (x, y, t) = 0 to determine the envelope.
   2  2
x+y x+y x+y
F x, y, = x− + y− −1=0
2 2 2

 2  2
x−y y−x
+ −1=0
2 2

(x − y)2 = 2

√
y =x± 2

The one parameter family of curves and its envelope is shown in Figure 44.2.

1749
 3

-3 -2 -1 1 2 3

-1

-2

-3

Figure 44.2: The envelope of (x − t)2 + (y − t)2 − 1 = 0.

44.5 Exercises
Exercise 44.1
Consider a semi-infinite string, x > 0. For all time the end of the string is displaced according to u(0, t) = f (t).
Find the motion of the string, u(x, t) with the method of characteristics and then with a Fourier transform in
time. The wave speed is c.

Exercise 44.2
Solve using characteristics:

 x
uux + uy = 1, ux=y = .
2

1750
Exercise 44.3
Solve using characteristics:

(y + u)ux + yuy = x − y, uy=1 = 1 + x.

1751
44.6 Hints
Hint 44.1

Hint 44.2

Hint 44.3

1752
44.7 Solutions
Solution 44.1

Method of characteristics. The problem is

utt − c2 uxx = 0, x > 0, −∞ < t < ∞,

u(0, t) = f (t).

By the method of characteristics, we know that the solution has the form,

u(x, t) = F (x − ct).

That is, it is a wave moving to the right with speed c. Substituting this into the boundary condition yields,

F (−ct) = f (t)
 
ξ
F (ξ) = f −
c
Now we can write the solution.

u(x, t) = f (t − x/c)

Fourier transform. We take the Fourier transform in time of the wave equation and the boundary condition.

utt = c2 uxx , u(0, t) = f (t)

−ω û = c ûxx , û(0, ω) = fˆ(ω)
2 2

ω2
ûxx + 2 û = 0, û(0, ω) = fˆ(ω)
c
The general solution of this ordinary differential equation is

û(x, ω) = a(ω) eiωx/c + b(ω) e−iωx/c .

1753
The radiation condition, (u(x, t) must be a wave traveling in the positive direction), and the boundary condition
at x = 0 will determine the constants a and b. u is the inverse Fourier transform of û.
Z ∞
a(ω) eiωx/c + b(ω) e−iωx/c eiωt dω


u(x, t) =
Z −∞
∞
a(ω) eiω(t+x/c) + b(ω) eiω(t−x/c) dω


u(x, t) =
−∞

The first and second terms in the integrand are left and right traveling waves, respectively. In order that u is a
right traveling wave, it must be a superposition of right traveling waves. We conclude that a(ω) = 0. Applying
the boundary condition at x = 0, we solve for û.

û(x, ω) = fˆ(ω) e−iωx/c

Finally we take the inverse Fourier transform.

Z ∞
u(x, t) = fˆ(ω) eiω(t−x/c) dω
−∞

u(x, t) = f (t − x/c)

Solution 44.2

 x
uux + uy = 1, ux=y = (44.2)
2
du
We form dy
.

du dx
= ux + uy
dy dy

1754
We compare this with Equation 44.2 to obtain differential equations for x and u.
dx du
= u, = 1. (44.3)
dy dy
The initial data is
α
x(y = α) = α, u(y = α) = . (44.4)
2
We solve the differenial equation for u (44.3) subject to the initial condition (44.4).
α
u(x(y), y) = y −
2
The differential equation for x becomes
dx α
=y− .
dy 2
We solve this subject to the initial condition (44.4).
1
x(y) = (y 2 + α(2 − y))
2
This defines the characteristic starting at the point (α, α). We solve for α.
y 2 − 2x
α=
y−2
We substitute this value for α into the solution for u.
y(y − 4) + 2x
u(x, y) =
2(y − 2)
This solution is defined for y 6= 2. This is because at (x, y) = (2, 2), the characteristic is parallel to the line x = y.
Figure 44.3 has a plot of the solution that shows the singularity at y = 2.

1755
 x -2
0
2
10
u 0
-10

-2
0
y 2

Figure 44.3: The solution u(x, y).

Solution 44.3


(y + u)ux + yuy = x − y, uy=1 = 1 + x (44.5)
We differentiate u with respect to s.
du dx dy
= ux + uy
ds ds ds
We compare this with Equation 44.5 to obtain differential equations for x, y and u.
dx dy du
= y + u, = y, =x−y
ds ds ds
We parametrize the initial data in terms of s.
x(s = 0) = α, y(s = 0) = 1, u(s = 0) = 1 + α

1756
We solve the equation for y subject to the inital condition.

y(s) = es

This gives us a coupled set of differential equations for x and u.

dx du
= es + u, = x − es
ds ds
The solutions subject to the initial conditions are

x(s) = (α + 1) es − e−s , u(s) = α es + e−s .

We substitute y(s) = es into these solutions.

1 1
x(s) = (α + 1)y − , u(s) = αy +
y y
We solve the first equation for α and substitute it into the second equation to obtain the solution.

2 + xy − y 2
u(x, y) =
y

This solution is valid for y > 0. The characteristic passing through (α, 1) is

x(s) = (α + 1) es − e−s , y(s) = es .

Hence we see that the characteristics satisfy y(s) ≥ 0 for all real s. Figure 44.4 shows some characteristics in the
(x, y) plane with starting points from (−5, 1) to (5, 1) and a plot of the solution.

1757
 2 -2
-1
1.75 x 0
1.5 1
1.25 2
15
1 10
0.75 u 5
0.5
0
0.25
0.5 1 1.5 2
-10-7.5 -5 -2.5 2.5 5 7.5 10 y

Figure 44.4: Some characteristics and the solution u(x, y).

1758
Chapter 45

Transform Methods

45.1 Fourier Transform for Partial Differential Equations

Solve Laplace’s equation in the upper half plane

∇2 u = 0 − ∞ < x < ∞, y > 0

u(x, 0) = f (x) −∞<x<∞

Taking the Fourier transform in the x variable of the equation and the boundary condition,
 2
∂ u ∂2u

F + = 0, F [u(x, 0)] = F [f (x)]
∂x2 ∂y 2
∂2
−ω 2 U (ω, y) + 2 U (ω, y) = 0, U (ω, 0) = F (ω).
∂y

The general solution to the equation is

U (ω, y) = a eωy + b e−ωy .

1759
Remember that in solving the differential equation here we consider ω to be a parameter. Requiring that the
solution be bounded for y ∈ [0, ∞) yields

U (ω, y) = a e−|ω|y .

Applying the boundary condition,

U (ω, y) = F (ω) e−|ω|y .

The inverse Fourier transform of e−|ω|y is

2y
F −1 e−|ω|y =
 
.
x2 + y2
Thus
 
2y
U (ω, y) = F (ω) F 2
x + y2
 
2y
F [u(x, y)] = F [f (x)] F 2 .
x + y2
Recall that the convolution theorem is
 Z ∞ 
1
F f (x − ξ)g(ξ) dξ = F (ω)G(ω).
2π −∞
Applying the convolution theorem to the equation for U ,
Z ∞
1 f (x − ξ)2y
u(x, y) = dξ
2π −∞ ξ 2 + y 2

∞
f (x − ξ)
Z
y
u(x, y) = dξ.
π −∞ ξ 2 + y2

1760
45.2 The Fourier Sine Transform
Consider the problem

ut = κuxx , x > 0, t > 0

u(0, t) = 0, u(x, 0) = f (x)

Since we are given the position at x = 0 we apply the Fourier sine transform.
 
2 2
ût = κ −ω û + ωu(0, t)
π
ût = −κω 2 û
2t
û(ω, t) = c(ω) e−κω

The initial condition is

û(ω, 0) = fˆ(ω).

We solve the first order differential equation to determine û.

2
û(ω, t) = fˆ(ω) e−κω t
 
ˆ 1 −x2 /(4κt)
û(ω, t) = f (ω)Fc √ e
4πκt

We take the inverse sine transform with the convolution theorem.

Z ∞
1  2 2

u(x, t) = √ f (ξ) e−|x−ξ| /(4κt) − e−(x+ξ) /(4κt) dξ
4π 3/2 κt 0

1761
45.3 Fourier Transform
Consider the problem
∂u ∂u
− + u = 0, −∞ < x < ∞, t > 0,
∂t ∂x

u(x, 0) = f (x).

Taking the Fourier Transform of the partial differential equation and the initial condition yields
∂U
− iωU + U = 0,
∂t
Z ∞
1
U (ω, 0) = F (ω) = f (x) e−iωx dx.
2π −∞

Now we have a first order differential equation for U (ω, t) with the solution

U (ω, t) = F (ω) e(−1+iω)t .

Now we apply the inverse Fourier transform.

Z ∞
u(x, t) = F (ω) e(−1+iω)t eiωx dω
−∞

Z ∞
−t
u(x, t) = e F (ω) eiω(x+t) dω
−∞

u(x, t) = e−t f (x + t)

1762
45.4 Exercises
Exercise 45.1
Find an integral representation of the solution u(x, y), of
uxx + uyy = 0 in − ∞ < x < ∞, 0 < y < ∞,
subject to the boundary conditions:
u(x, 0) = f (x), −∞ < x < ∞;
u(x, y) → 0 as x2 + y 2 → ∞.

Exercise 45.2
Solve the Cauchy problem for the one-dimensional heat equation in the domain −∞ < x < ∞, t > 0,
ut = κuxx , u(x, 0) = f (x),
with the Fourier transform.

Exercise 45.3
Let φ(x, t) satisfy the equation
φt = a2 φxx , (45.1)
for −∞ < x < ∞, t > 0 with initial conditions φ(x, 0) = f (x) in −∞ < x < ∞. Boundary conditions cannot
be given here because both endpoints are infinite. In this case all we can ask is that the solution be regular as
x → ±∞. Show that the Laplace transform of φ(x, t) is given by
Z ∞  √ 
1 s
Φ(x, s) = √ f (ξ) exp − |x − ξ| dξ,
2a s −∞ a
and hence deduce that
∞
(x − ξ)2
Z  
1
φ(x, t) = √ f (ξ) exp − dξ.
2a πt −∞ 4a2 t

1763
Exercise 45.4
1. In Exercise 45.3 above, let f (−x) = −f (x) for all x and verify that φ(x, t) so obtained is the solution, for
x > 0, of the following problem: find φ(x, t) satisfying

φt = a2 φxx

in 0 < x < ∞, t > 0, with boundary condition φ(0, t) = 0 and initial condition φ(x, 0) = f (x). This
technique, in which the solution for a semi-infinite interval is obtained from that for an infinite interval, is
an example of what is called the method of images.

2. How would you modify the result of part (a) if the boundary condition φ(0, t) = 0 was replaced by φx (0, t) =
0?

Exercise 45.5
Solve the Cauchy problem for the one-dimensional wave equation in the domain −∞ < x < ∞, t > 0,

utt = c2 uxx , u(x, 0) = f (x), ut (x, 0) = g(x),

with the Fourier transform.

Exercise 45.6
Solve the Cauchy problem for the one-dimensional wave equation in the domain −∞ < x < ∞, t > 0,

utt = c2 uxx , u(x, 0) = f (x), ut (x, 0) = g(x),

with the Laplace transform.

Exercise 45.7
Consider the problem of determining φ(x, t) in the region 0 < x < ∞, 0 < t < ∞, such that

φt = a2 φxx , (45.2)

1764
with initial and boundary conditions

φ(x, 0) = 0 for all x > 0,

φ(0, t) = f (t) for all t > 0,

where f (t) is a given function.

1. Obtain the formula for the Laplace transform of φ(x, t), Φ(x, s) and use the convolution theorem for Laplace
transforms to show that
Z t
x2
 
x 1
φ(x, t) = √ f (t − τ ) 3/2 exp − 2 dτ.
2a π 0 τ 4a τ

2. Discuss the special case obtained by setting f (t) = 1 and also that in which f (t) = 1 for 0 < t < T , with
f (t) = 0 for t > T . Here T is some positive constant.

Exercise 45.8
Solve the radiating half space problem:

ut = κuxx , x > 0, t > 0,

ux (0, t) − αu(0, t) = 0, u(x, 0) = f (x).

To do this, define

v(x, t) = ux (x, t) − αu(x, t)

and find the half space problem that v satisfies. Solve this problem and then show that
Z ∞
u(x, t) = − e−α(ξ−x) v(ξ, t) dξ.
x

1765
Exercise 45.9
Show that
Z ∞ √
−cω 2 x π −x2 /(4c)
ωe sin(ωx) dω = 3/2 e .
0 4c
Use the sine transform to solve:
ut = uxx , x > 0, t > 0,
u(0, t) = g(t), u(x, 0) = 0.

Exercise 45.10
Use the Fourier sine transform to find the steady state temperature u(x, y) in a slab: x ≥ 0, 0 ≤ y ≤ 1, which has
zero temperature on the faces y = 0 and y = 1 and has a given distribution: u(y, 0) = f (y) on the edge x = 0,
0 ≤ y ≤ 1.

Exercise 45.11
Find a harmonic function u(x, y) in the upper half plane which takes on the value g(x) on the x-axis. Assume
that u and ux vanish as |x| → ∞. Use the Fourier transform with respect to x. Express the solution as a single
integral by using the convolution formula.

Exercise 45.12
Find the bounded solution of
ut = κuxx − a2 u, 0 < x < ∞, t > 0,
−ux (0, t) = f (t), u(x, 0) = 0.

Exercise 45.13
The left end of a taut string of length L is displaced according to u(0, t) = f (t). The right end is fixed, u(L, t) = 0.
Initially the string is at rest with no displacement. If c is the wave speed for the string, find it’s motion for all
t > 0.

1766
Exercise 45.14
Let ∇2 φ = 0 in the (x, y)-plane region defined by 0 < y < l, −∞ < x < ∞, with φ(x, 0) = δ(x − ξ), φ(x, l) = 0,
and φ → 0 as |x| → ∞. Solve for φ using Fourier transforms. You may leave your answer in the form of an
integral but in fact it is possible to use techniques of contour integration to show that
 
1 sin(πy/l)
φ(x, y|ξ) = .
2l cosh[π(x − ξ)/l] − cos(πy/l)

Note that as l → ∞ we recover the result derived in class:

1 y
φ→ ,
π (x − ξ)2 + y 2

which clearly approaches δ(x − ξ) as y → 0.

1767
45.5 Hints
Hint 45.1 R∞
The desired solution form is: u(x, y) = −∞ K(x − ξ, y)f (ξ) dξ. You must find the correct K. Take the Fourier
transform with respect to x and solve for û(ω, y) recalling that ûxx = −ω 2 û. By ûxx we denote the Fourier
transform with respect to x of uxx (x, y).

Hint 45.2
Use the Fourier convolution theorem and the table of Fourier transforms in the appendix.

Hint 45.3

Hint 45.4

Hint 45.5
Use the Fourier convolution theorem. The transform pairs,

F[π(δ(x + τ ) + δ(x − τ ))] = cos(ωτ ),

sin(ωτ )
F[π(H(x + τ ) − H(x − τ ))] = ,
ω
will be useful.

Hint 45.6

1768
Hint 45.7

Hint 45.8
v(x, t) satisfies the same partial differential equation. You can solve the problem for v(x, t) with the Fourier sine
transform. Use the convolution theorem to invert the transform.
To show that
Z ∞
u(x, t) = − e−α(ξ−x) v(ξ, t) dξ,
x

find the solution of

ux − αu = v

that is bounded as x → ∞.

Hint 45.9
Note that
Z ∞ Z ∞
−cω 2 ∂ 2
ωe sin(ωx) dω = − e−cω cos(ωx) dω.
0 ∂x 0

Write the integral as a Fourier transform.

Take the Fourier sine transform of the heat equation to obtain a first order, ordinary differential equation for
û(ω, t). Solve the differential equation and do the inversion with the convolution theorem.

Hint 45.10

1769
Hint 45.11

Hint 45.12

Hint 45.13

Hint 45.14

1770
45.6 Solutions
Solution 45.1
1. We take the Fourier transform of the integral equation, noting that the left side is the convolution of u(x)
1
and x2 +a2.

  
1 1
2πû(ω)F 2 =F 2
x + a2 x + b2

We find the Fourier transform of f (x) = 1

x2 +c2
. Note that since f (x) is an even, real-valued function, fˆ(ω)
is an even, real-valued function.
  Z ∞
1 1 1
F 2 2
= e−iωx dx
x +c 2π −∞ x + c2
2

For x > 0 we close the path of integration in the upper half plane and apply Jordan’s Lemma to evaluate
the integral in terms of the residues.

e−iωx
 
1
= i2π Res , x = ic
2π (x − ic)(x + ic)
e−iωic
=i
2ic
1 −cω
= e
2c

Since fˆ(ω) is an even function, we have

 
1 1 −c|ω|
F 2 2
= e .
x +c 2c

1771
 Our equation for û(ω) becomes,
1 −a|ω| 1 −b|ω|
2πû(ω) e = e
2a 2b
a −(b−a)|omega|
û(ω) = e .
2πb
We take the inverse Fourier transform using the transform pair we derived above.
a 2(b − a)
u(x) =
2πb x2 + (b − a)2
a(b − a)
u(x) =
πb(x2 + (b − a)2 )

2. We take the Fourier transform of the partial differential equation and the boundary condtion.

uxx + uyy = 0, u(x, 0) = f (x)

−ω 2 û(ω, y) + ûyy (ω, y) = 0, û(ω, 0) = fˆ(ω)

This is an ordinary differential equation for û in which ω is a parameter. The general solution is

û = c1 eωy + c2 e−ωy .

We apply the boundary conditions that û(ω, 0) = fˆ(ω) and û → 0 and y → ∞.

û(ω, y) = fˆ(ω) e−ωy

We take the inverse transform using the convolution theorem.

Z ∞
1
u(x, y) = e−(x−ξ)y f (ξ) dξ
2π −∞

1772
Solution 45.2

ut = κuxx , u(x, 0) = f (x),

We take the Fourier transform of the heat equation and the initial condition.

ût = −κω 2 û, û(ω, 0) = fˆ(ω)

This is a first order ordinary differential equation which has the solution,
2
û(ω, t) = fˆ(ω) e−κω t .

Using a table of Fourier transforms we can write this in a form that is conducive to applying the convolution
theorem.
r 
π 2 /(4κt)
û(ω, t) = fˆ(ω)F e −x
κt
r Z ∞
π 2
u(x, t) = e−(x−ξ) /(4κt) f (ξ) dξ
κt −∞

Solution 45.3
We take the Laplace transform of Equation 45.1.

sφ̂ − φ(x, 0) = a2 φ̂xx

s f (x)
φ̂xx − 2 φ̂ = − 2 (45.3)
a a
The Green function problem for Equation 45.3 is
s
G00 − G = δ(x − ξ), G(±∞; ξ) is bounded.
a2

1773
The homogeneous solutions that satisfy the left and right boundary conditions are, respectively,
√   √ 
sa sa
exp , exp − .
x x

We compute the Wronskian of these solutions.

√   √  
√

s
 exp a x exp − as x
 
 2 s
W =  √s √ √ √
=− a
    
sa
 a exp x − as exp − xsa 

The Green function is

√   √ 
s s
exp a <
x exp − a x>
G(x; ξ) = √
−2as
 √ 
a s
G(x; ξ) = − √ exp − |x − ξ| .
2 s a

Now we solve Equation 45.3 using the Green function.

Z ∞
f (ξ)
φ̂(x, s) = − 2 G(x; ξ) dξ
−∞ a
Z ∞  √ 
1 s
φ̂(x, s) = √ f (ξ) exp − |x − ξ| dξ
2a s −∞ a

Finally we take the inverse Laplace transform to obtain the solution of Equation 45.1.
∞
(x − ξ)2
Z  
1
φ(x, t) = √ f (ξ) exp − dξ
2a πt −∞ 4a2 t

1774
Solution 45.4
1. Clearly the solution satisfies the differential equation. We must verify that it satisfies the boundary condition,
φ(0, t) = 0.
Z ∞
(x − ξ)2
 
1
φ(x, t) = √ f (ξ) exp − dξ
2a πt −∞ 4a2 t
Z 0 Z ∞
(x − ξ)2 (x − ξ)2
   
1 1
φ(x, t) = √ f (ξ) exp − dξ + √ f (ξ) exp − dξ
2a πt −∞ 4a2 t 2a πt 0 4a2 t
Z ∞ Z ∞
(x + ξ)2 (x − ξ)2
   
1 1
φ(x, t) = √ f (−ξ) exp − dξ + √ f (ξ) exp − dξ
2a πt 0 4a2 t 2a πt 0 4a2 t
Z ∞ Z ∞
(x + ξ)2 (x − ξ)2
   
1 1
φ(x, t) = − √ f (ξ) exp − dξ + √ f (ξ) exp − dξ
2a πt 0 4a2 t 2a πt 0 4a2 t
Z ∞
(x − ξ)2 (x + ξ)2
    
1
φ(x, t) = √ f (ξ) exp − exp − dξ
2a πt 0 4a2 t 4a2 t
Z ∞  2
x + ξ2
    
1 xξ xξ
φ(x, t) = √ f (ξ) exp − exp − exp − 2 dξ
2a πt 0 4a2 t 2a2 t 2a t
Z ∞  2
x + ξ2
  
1 xξ
φ(x, t) = √ f (ξ) exp − sinh dξ
a πt 0 4a2 t 2a2 t

Since the integrand is zero for x = 0, the solution satisfies the boundary condition there.
2. For the boundary condition φx (0, t) = 0 we would choose f (x) to be even. f (−x) = f (x). The solution is
Z ∞  2
x + ξ2
  
1 xξ
φ(x, t) = √ f (ξ) exp − cosh dξ
a πt 0 4a2 t 2a2 t

The derivative with respect to x is

Z ∞  2
x + ξ2
    
1 xξ xξ
φx (x, t) = 3 √ 3/2 f (ξ) exp − ξ sinh − x cosh dξ.
2a πt 0 4a2 t 2a2 t 2a2 t

1775
 Since the integrand is zero for x = 0, the solution satisfies the boundary condition there.

Solution 45.5
With the change of variables

∂ ∂t ∂ 1∂
τ = ct, = = , v(x, τ ) = u(x, t),
∂τ ∂τ ∂t c ∂t
the problem becomes
1
vτ τ = vxx , v(x, 0) = f (x), vτ (x, 0) = g(x).
c
(This change of variables isn’t necessary, it just gives us fewer constants to carry around.) We take the Fourier
transform in x of the equation and the initial conditions, (we consider τ to be a parameter),
1
v̂τ τ (ω, τ ) = −ω 2 v̂(ω, τ ), v̂(ω, τ ) = fˆ(ω), v̂τ (ω, τ ) = ĝ(ω).
c
Now we have an ordinary differential equation for v̂(ω, τ ), (now we consider ω to be a parameter). The general
solution of this constant coefficient differential equation is,

v̂(ω, τ ) = a(ω) cos(ωτ ) + b(ω) sin(ωτ ),

where a and b are constants that depend on the parameter ω. Applying the initial conditions, we see that
1
v̂(ω, τ ) = fˆ(ω) cos(ωτ ) + ĝ(ω) sin(ωτ ),
cω
With the Fourier transform pairs

F[π(δ(x + τ ) + δ(x − τ ))] = cos(ωτ ),

sin(ωτ )
F[π(H(x + τ ) − H(x − τ ))] = ,
ω

1776
we can write v̂(ω, τ ) in a form that is conducive to applying the Fourier convolution theorem.

1
v̂(ω, τ ) = F[f (x)]F[π(δ(x + τ ) + δ(x − τ ))] + F[g(x)]F[π(H(x + τ ) − H(x − τ ))]
c

Z ∞
1
v(x, τ ) = f (ξ)π(δ(x − ξ + τ ) + δ(x − ξ − τ )) dξ
2π −∞
Z ∞
1 1
+ g(ξ)π(H(x − ξ + τ ) − H(x − ξ − τ )) dξ
c 2π −∞

Z x+τ
1 1
v(x, τ ) = (f (x + τ ) + f (x − τ )) + g(ξ) dξ
2 2c x−τ

Finally we make the change of variables t = τ /c, u(x, t) = v(x, τ ) to obtain D’Alembert’s solution of the wave
equation,
Z x+ct
1 1
u(x, t) = (f (x − ct) + f (x + ct)) + g(ξ) dξ.
2 2c x−ct

Solution 45.6
With the change of variables

∂ ∂t ∂ 1∂
τ = ct, = = , v(x, τ ) = u(x, t),
∂τ ∂τ ∂t c ∂t
the problem becomes
1
vτ τ = vxx , v(x, 0) = f (x), vτ (x, 0) = g(x).
c

1777
We take the Laplace transform in τ of the equation, (we consider x to be a parameter),

s2 V (x, s) − sv(x, 0) − vτ (x, 0) = Vxx (x, s),

1
Vxx (x, s) − s2 V (x, s) = −sf (x) − g(x),
c
Now we have an ordinary differential equation for V (x, s), (now we consider s to be a parameter). We impose the
boundary conditions that the solution is bounded at x = ±∞. Consider the Green’s function problem

gxx (x; ξ) − s2 g(x; ξ) = δ(x − ξ), g(±∞; ξ) bounded.

esx is a homogeneous solution that is bounded at x = −∞. e−sx is a homogeneous solution that is bounded at
x = +∞. The Wronskian of these solutions is
 sx 
e e−sx 
W (x) =  sx
 = −2s.
se −s e−sx 

Thus the Green’s function is

(
1 sx −sξ
− 2s e e for x < ξ, 1
g(x; ξ) = 1 sξ −sx
= − e−s|x−ξ| .
− 2s e e for x > ξ, 2s

The solution for V (x, s) is

Z ∞
1 1
V (x, s) = − e−s|x−ξ| (−sf (ξ) − g(ξ)) dξ,
2s −∞ c

Z ∞ Z ∞
1 −s|x−ξ| 1
V (x, s) = e f (ξ) dξ + e−s|x−ξ| g(ξ)) dξ,
2 −∞ 2cs −∞

1778
 ∞ ∞
e−s|ξ|
Z Z
1 −s|ξ| 1
V (x, s) = e f (x − ξ) dξ + g(x − ξ)) dξ.
2 −∞ 2c −∞ s
Now we take the inverse Laplace transform and interchange the order of integration.
Z ∞  Z ∞ −s|ξ| 
1 −1 −s|ξ| 1 −1 e
v(x, τ ) = L e f (x − ξ) dξ + L g(x − ξ)) dξ
2 −∞ 2c −∞ s

∞ ∞
e−s|ξ|
Z Z  
1 −1
 −s|ξ|
 1 −1
v(x, τ ) = L e f (x − ξ) dξ + L g(x − ξ)) dξ
2 −∞ 2c −∞ s

Z ∞ Z ∞
1 1
v(x, τ ) = δ(τ − |ξ|)f (x − ξ) dξ + H(τ − |ξ|)g(x − ξ)) dξ
2 −∞ 2c −∞

Z τ
1 1
v(x, τ ) = (f (x − τ ) + f (x + τ )) + g(x − ξ) dξ
2 2c −τ

Z −x+τ
1 1
v(x, τ ) = (f (x − τ ) + f (x + τ )) + g(−ξ) dξ
2 2c −x−τ

Z x+τ
1 1
v(x, τ ) = (f (x − τ ) + f (x + τ )) + g(ξ) dξ
2 2c x−τ

Now we write make the change of variables t = τ /c, u(x, t) = v(x, τ ) to obtain D’Alembert’s solution of the wave
equation,
Z x+ct
1 1
u(x, t) = (f (x − ct) + f (x + ct)) + g(ξ) dξ.
2 2c x−ct

1779
Solution 45.7
1. We take the Laplace transform of Equation 45.2.

sφ̂ − φ(x, 0) = a2 φ̂xx

s
φ̂xx − 2 φ̂ = 0 (45.4)
a

We take the Laplace transform of the initial condition, φ(0, t) = f (t), and use that φ̂(x, s) vanishes as
x → ∞ to obtain boundary conditions for φ̂(x, s).

φ̂(0, s) = fˆ(s), φ̂(∞, s) = 0

The solutions of Equation 45.4 are

 √ 
s
exp ± x .
a

The solution that satisfies the boundary conditions is

 √ 
s
φ̂(x, s) = fˆ(s) exp − x .
a

We write this as the product of two Laplace transforms.

2
  
x x
φ̂(x, s) = fˆ(s)L √ exp − 2
2a πt3/2 4a t

We invert using the convolution theorem.

t
x2
Z  
x 1
φ(x, t) = √ f (t − τ ) exp − 2 dτ.
2a π 0 τ 3/2 4a τ

1780
 2. Consider the case f (t) = 1.
Z t
x2
 
x 1
φ(x, t) = √ exp − 2 dτ
2a π 0 τ 3/2 4a τ
x x
ξ = √ , dξ = −
2a τ 4aτ 3/2
√
Z x/(2a t)
2 2
φ(x, t) = − √ e−ξ dξ
π ∞
 
x
φ(x, t) = erfc √
2a t
Now consider the case in which f (t) = 1 for 0 < t < T , with f (t) = 0 for t > T . For t < T , φ is the same
as before.
 
x
φ(x, t) = erfc √ , for 0 < t < T
2a t
Consider t > T .
Z t
x2
 
x 1
φ(x, t) = √ exp − 2 dτ
2a π t−T τ 3/2 4a τ
Z x/(2a√t)
2 2
φ(x, t) = − √ √
e−ξ dξ
π x/(2a t−T )
   
x x
φ(x, t) = erf √ − erf √
2a t − T 2a t

Solution 45.8

ut = κuxx , x > 0, t > 0,

ux (0, t) − αu(0, t) = 0, u(x, 0) = f (x).

1781
First we find the partial differential equation that v satisfies. We start with the partial differential equation for u,

ut = κuxx .

Differentiating this equation with respect to x yields,

utx = κuxxx .

Subtracting α times the former equation from the latter yields,

utx − αut = κuxxx − ακuxx ,

∂ ∂2
(ux − αu) = κ 2 (ux − αu) ,
∂t ∂x
vt = κvxx .

Thus v satisfies the same partial differential equation as u. This is because the equation for u is linear and
homogeneous and v is a linear combination of u and its derivatives. The problem for v is,

vt = κvxx , x > 0, t > 0,

v(0, t) = 0, v(x, 0) = f 0 (x) − αf (x).

With this new boundary condition, we can solve the problem with the Fourier sine transform. We take the sine
transform of the partial differential equation and the initial condition.
 
2 1
v̂t (ω, t) = κ −ω v̂(ω, t) + ωv(0, t) ,
π
0
v̂(ω, 0) = Fs [f (x) − αf (x)]

v̂t (ω, t) = −κω 2 v̂(ω, t)

v̂(ω, 0) = Fs [f 0 (x) − αf (x)]

1782
Now we have a first order, ordinary differential equation for v̂. The general solution is,
2
v̂(ω, t) = c e−κω t .

The solution subject to the initial condition is,

2
v̂(ω, t) = Fs [f 0 (x) − αf (x)] e−κω t .

Now we take the inverse sine transform to find v. We utilize the Fourier cosine transform pair,
h 2
i rπ 2
Fc−1 e−κω t = e−x /(4κt) ,
κt

to write v̂ in a form that is suitable for the convolution theorem.

r 
0 π −x2 /(4κt)
v̂(ω, t) = Fs [f (x) − αf (x)] Fc e
κt

Recall that the Fourier sine convolution theorem is,

 Z ∞ 
1
Fs f (ξ) (g(|x − ξ|) − g(x + ξ)) dξ = Fs [f (x)]Fc [g(x)].
2π 0

Thus v(x, t) is
Z ∞
1 0

−|x−ξ|2 /(4κt) −(x+ξ)2 /(4κt)

v(x, t) = √ (f (ξ) − αf (ξ)) e −e dξ.
2 πκt 0

With v determined, we have a first order, ordinary differential equation for u,

ux − αu = v.

1783
We solve this equation by multiplying by the integrating factor and integrating.
∂
e−αx u = e−αx v


∂x Z
x
e−αx u = e−αξ v(x, t) dξ + c(t)
Z x
u= e−α(ξ−x) v(x, t) dξ + eαx c(t)

The solution that vanishes as x → ∞ is

Z ∞
u(x, t) = − e−α(ξ−x) v(ξ, t) dξ.
x

Solution 45.9

Z ∞ Z ∞
−cω 2 ∂ 2
ωe sin(ωx) dω = − e−cω cos(ωx) dω
0 ∂x 0
Z ∞
1 ∂ 2
=− e−cω cos(ωx) dω
2 ∂x −∞
Z ∞
1 ∂ 2
=− e−cω +iωx dω
2 ∂x −∞
Z ∞
1 ∂ 2 2
=− e−c(ω+ix/(2c)) e−x /(4c) dω
2 ∂x −∞
1 ∂ −x2 /(4c) ∞ −cω2
Z
=− e e dω
2 ∂x −∞
r
1 π ∂ −x2 /(4c)
=− e
2 c ∂x
√
x π −x2 /(4c)
= 3/2 e
4c

1784
 ut = uxx , x > 0, t > 0,
u(0, t) = g(t), u(x, 0) = 0.
We take the Fourier sine transform of the partial differential equation and the initial condition.
ω
ût (ω, t) = −ω 2 û(ω, t) + g(t), û(ω, 0) = 0
π
Now we have a first order, ordinary differential equation for û(ω, t).
∂  ω2 t  ω 2
e ût (ω, t) = g(t) eω t
∂t π
ω −ω2 t t
Z
2 2
û(ω, t) = e g(τ ) eω τ dτ + c(ω) e−ω t
π 0

The initial condition is satisfied for c(ω) = 0.

Z t
ω 2 (t−τ )
û(ω, t) = g(τ ) e−ω dτ
π 0

We take the inverse sine transform to find u.

 Z t 
ω −ω 2 (t−τ )
u(x, t) = Fs−1 g(τ ) e dτ
π 0

Z t hω i
2 (t−τ )
u(x, t) = g(τ )Fs−1 e−ω dτ
0 π
Z t
x 2
u(x, t) = g(τ ) √ 3/2
e−x /(4(t−τ )) dτ
0 2 π(t − τ )

t 2
e−x /(4(t−τ ))
Z
x
u(x, t) = √ g(τ ) dτ
2 π 0 (t − τ )3/2

1785
Solution 45.10
The problem is

uxx + uyy = 0, 0 < x, 0 < y < 1,

u(x, 0) = u(x, 1) = 0, u(0, y) = f (y).

We take the Fourier sine transform of the partial differential equation and the boundary conditions.
k
−ω 2 û(ω, y) + u(0, y) + ûyy (ω, y) = 0
π
k
ûyy (ω, y) − ω 2 û(ω, y) = − f (y), û(ω, 0) = û(ω, 1) = 0
π
This is an inhomogeneous, ordinary differential equation that we can solve with Green functions. The homogeneous
solutions are

{cosh(ωy), sinh(ωy)}.

The homogeneous solutions that satisfy the left and right boundary conditions are

y1 = sinh(ωy), y2 = sinh(ω(y − 1)).

The Wronskian of these two solutions is,

 
 sinh(ωy) sinh(ω(y − 1)) 
W (x) = 
ω cosh(ωy) ω cosh(ω(y − 1))
= ω (sinh(ωy) cosh(ω(y − 1)) − cosh(ωy) sinh(ω(y − 1)))
= ω sinh(ω).

The Green function is

sinh(ωy< ) sinh(ω(y> − 1))
G(y|η) = .
ω sinh(ω)

1786
The solution of the ordinary differential equation for û(ω, y) is
ω 1
Z
û(ω, y) = − f (η)G(y|η) dη
π 0
1 y sinh(ωη) sinh(ω(y − 1)) 1 1 sinh(ωy) sinh(ω(η − 1))
Z Z
=− f (η) dη − f (η) dη.
π 0 sinh(ω) π y sinh(ω)
With some uninteresting grunge, you can show that,
Z ∞
sinh(ωη) sinh(ω(y − 1)) sin(πη) sin(πy)
2 sin(ωx) dω = −2 .
0 sinh(ω) (cosh(πx) − cos(π(y − η)))(cosh(πx) − cos(π(y + η)))
Taking the inverse Fourier sine transform of û(ω, y) and interchanging the order of integration yields,

2 y
Z
sin(πη) sin(πy)
u(x, y) = f (η) dη
π 0 (cosh(πx) − cos(π(y − η)))(cosh(πx) − cos(π(y + η)))
2 1
Z
sin(πy) sin(πη)
+ f (η) dη.
π y (cosh(πx) − cos(π(η − y)))(cosh(πx) − cos(π(η + y)))

Z 1
2 sin(πη) sin(πy)
u(x, y) = f (η) dη
π 0 (cosh(πx) − cos(π(y − η)))(cosh(πx) − cos(π(y + η)))

Solution 45.11
The problem for u(x, y) is,
uxx + uyy = 0, −∞ < x < ∞, y > 0,
u(x, 0) = g(x).
We take the Fourier transform of the partial differential equation and the boundary condition.
−ω 2 û(ω, y) + ûyy (ω, y) = 0, û(ω, 0) = ĝ(ω).

1787
This is an ordinary differential equation for û(ω, y). So far we only have one boundary condition. In order that u
is bounded we impose the second boundary condition û(ω, y) is bounded as y → ∞. The general solution of the
differential equation is
(
c1 (ω) eωy + c2 (ω) e−ωy , for ω 6= 0,
û(ω, y) =
c1 (ω) + c2 (ω)y, for ω = 0.

Note that eωy is the bounded solution for ω < 0, 1 is the bounded solution for ω = 0 and e−ωy is the bounded
solution for ω > 0. Thus the bounded solution is

û(ω, y) = c(ω) e−|ω|y .

The boundary condition at y = 0 determines the constant of integration.

û(ω, y) = ĝ(ω) e−|ω|y

Now we take the inverse Fourier transform to obtain the solution for u(x, y). To do this we use the Fourier
transform pair,
 
2c
F 2 2
= e−c|ω| ,
x +c

and the convolution theorem,

 Z ∞ 
1
F f (ξ)g(x − ξ) dξ = fˆ(ω)ĝ(ω).
2π −∞

Z ∞
1 2y
u(x, y) = g(ξ) dξ.
2π −∞ (x − ξ)2 + y 2

1788
Solution 45.12
Since the derivative of u is specified at x = 0, we take the cosine transform of the partial differential equation and
the initial condition.
 
1
ût (ω, t) = κ −ω û(ω, t) − ux (0, t) − a2 û(ω, t), û(ω, 0) = 0
2
π
2 2

κ
ût + κω + a û = f (t), û(ω, 0) = 0
π
This first order, ordinary differential equation for û(ω, t) has the solution,
κ t −(κω2 +a2 )(t−τ )
Z
û(ω, t) = e f (τ ) dτ.
π 0
We take the inverse Fourier cosine transform to find the solution u(x, t).
Z t 
κ −1 −(κω 2 +a2 )(t−τ )
u(x, t) = Fc e f (τ ) dτ
π 0
κ t −1 h −κω2 (t−τ ) i −a2 (t−τ )
Z
u(x, t) = F e e f (τ ) dτ
π 0 c
κ t
Z r
π 2 2
u(x, t) = e−x /(4κ(t−τ )) e−a (t−τ ) f (τ ) dτ
π 0 κ(t − τ )
r Z t −x2 /(4κ(t−τ ))−a2 (t−τ )
κ e
u(x, t) = √ f (τ ) dτ
π 0 t−τ

Solution 45.13
Mathematically stated we have
utt = c2 uxx , 0 < x < L, t > 0,
u(x, 0) = ut (x, 0) = 0,
u(0, t) = f (t), u(L, t) = 0.

1789
We take the Laplace transform of the partial differential equation and the boundary conditions.

s2 û(x, s) − su(x, 0) − ut (x, 0) = c2 ûxx (x, s)

s2
ûxx = 2 û, û(0, s) = fˆ(s), û(L, s) = 0
c

Now we have an ordinary differential equation. A set of solutions is

n  sx   sx o
cosh , sinh .
c c

The solution that satisfies the right boundary condition is

 
s(L − x)
û = a sinh .
c

The left boundary condition determines the multiplicative constant.

sinh(s(L − x)/c)
û(x, s) = fˆ(s)
sinh(sL/c)

If we can find the inverse Laplace transform of

sinh(s(L − x)/c)
û(x, s) =
sinh(sL/c)

then we can use the convolution theorem to write u in terms of a single integral. We proceed by expanding this

1790
function in a sum.
sinh(s(L − x)/c) es(L−x)/c − e−s(L−x)/c
=
sinh(sL/c) esL/c − e−sL/c
e−sx/c − e−s(2L−x)/c
=
1 − e−2sL/c
∞
−sx/c −s(2L−x)/c

X
= e −e e−2nsL/c
n=0
∞
X ∞
X
= e−s(2nL+x)/c − e−s(2(n+1)L−x)/c
n=0 n=0
∞
X ∞
X
= e−s(2nL+x)/c − e−s(2nL−x)/c
n=0 n=1

Now we use the Laplace transform pair:

L[δ(x − a)] = e−sa .

  X ∞ ∞
−1 sinh(s(L − x)/c) X
L = δ(t − (2nL + x)/c) − δ(t − (2nL − x)/c)
sinh(sL/c) n=0 n=1

We write û in the form,

"∞ ∞
#
X X
û(x, s) = L[f (t)]L δ(t − (2nL + x)/c) − δ(t − (2nL − x)/c) .
n=0 n=1

By the convolution theorem we have

∞ ∞
Z t !
X X
u(x, t) = f (τ ) δ(t − τ − (2nL + x)/c) − δ(t − τ − (2nL − x)/c) dτ.
0 n=0 n=1

1791
We can simplify this a bit. First we determine which Dirac delta functions have their singularities in the range
τ ∈ (0..t). For the first sum, this condition is

0 < t − (2nL + x)/c < t.

The right inequality is always satisfied. The left inequality becomes

(2nL + x)/c < t,

ct − x
n< .
2L
For the second sum, the condition is

0 < t − (2nL − x)/c < t.

Again the right inequality is always satisfied. The left inequality becomes

ct + x
n< .
2L
We change the index range to reflect the nonzero contributions and do the integration.

b ct−x b ct+x

Z t 2L
c 2L
c
X X
u(x, t) = f (τ )  δ(t − τ − (2nL + x)/c) δ(t − τ − (2nL − x)/c) dτ.
0 n=0 n=1

b ct−x
2L
c b ct+x
2L
c
X X
u(x, t) = f (t − (2nL + x)/c) f (t − (2nL − x)/c)
n=0 n=1

1792
Solution 45.14
We take the Fourier transform of the partial differential equation and the boundary conditions.

1 −iωξ
−ω 2 φ̂ + φ̂yy = 0, φ̂(ω, 0) = e , φ̂(ω, l) = 0
2π
We solve this boundary value problem.

φ̂(ω, y) = c1 cosh(ω(l − y)) + c2 sinh(ω(l − y))

1 −iωξ sinh(ω(l − y))
φ̂(ω, y) = e
2π sinh(ωl)

We take the inverse Fourier transform to obtain an expression for the solution.
∞
sinh(ω(l − y))
Z
1
φ(x, y) = eiω(x−ξ) dω
2π −∞ sinh(ωl)

1793
Chapter 46

Green Functions

46.1 Inhomogeneous Equations and Homogeneous Boundary Con-

ditions
Consider a linear differential equation on the domain Ω subject to homogeneous boundary conditions.

L[u(x)] = f (x) for x ∈ Ω, B[u(x)] = 0 for x ∈ ∂Ω (46.1)

For example, L[u] might be

L[u] = ut − κ∆u, or L[u] = ut − c2 ∆u.

and B[u] might be u = 0, or ∇u · n̂ = 0.

If we find a Green function G(x; ξ) that satisfies

L[G(x; ξ)] = δ(x − ξ), B[G(x; ξ)] = 0

1794
then the solution to Equation 46.1 is
Z
u(x) = G(x; ξ)f (ξ) dξ.
Ω

We verify that this solution satisfies the equation and boundary condition.
Z
L[u(x)] = L[G(x; ξ)]f (ξ) dξ
Z Ω

= δ(x − ξ)f (ξ) dξ

Ω
= f (x)
Z
B[u(x)] = B[G(x; ξ)]f (ξ) dξ
Z Ω

= 0 f (ξ) dξ
Ω
=0

46.2 Homogeneous Equations and Inhomogeneous Boundary Con-

ditions
Consider a homogeneous linear differential equation on the domain Ω subject to inhomogeneous boundary
conditions,

L[u(x)] = 0 for x ∈ Ω, B[u(x)] = h(x) for x ∈ ∂Ω. (46.2)

If we find a Green function g(x; ξ) that satisfies

L[g(x; ξ)] = 0, B[g(x; ξ)] = δ(x − ξ)

1795
then the solution to Equation 46.2 is
Z
u(x) = g(x; ξ)h(ξ) dξ.
∂Ω

We verify that this solution satisfies the equation and boundary condition.
Z
L[u(x)] = L[g(x; ξ)]h(ξ) dξ
Z∂Ω
= 0 h(ξ) dξ
∂Ω
=0
Z
B[u(x)] = B[g(x; ξ)]h(ξ) dξ
Z∂Ω
= δ(x − ξ)h(ξ) dξ
∂Ω
= h(x)

Example 46.2.1 Consider the Cauchy problem for the homogeneous heat equation.
ut = κuxx , −∞ < x < ∞, t > 0
u(x, 0) = h(x), u(±∞, t) = 0
We find a Green function that satisfies
gt = κgxx , −∞ < x < ∞, t > 0
g(x, 0; ξ) = δ(x − ξ), g(±∞, t; ξ) = 0.
Then we write the solution
Z ∞
u(x, t) = g(x, t; ξ)h(ξ) dξ.
−∞

1796
 To find the Green function for this problem, we apply a Fourier transform to the equation and boundary
condition for g.
ĝt = −κω 2 ĝ, ĝ(ω, 0; ξ) = F[δ(x − ξ)]
2
ĝ(ω, t; ξ) = F[δ(x − ξ)] e−κω t
r
x2
 
π
ĝ(ω, t; ξ) = F[δ(x − ξ)]F exp −
κt 4κt
We invert using the convolution theorem.
∞
r
(x − η)2
Z  
1 π
g(x, t; ξ) = δ(η − ξ) exp − dη
2π −∞ κt 4κt
(x − ξ)2
 
1
=√ exp −
4πκt 4κt
The solution of the heat equation is
∞
(x − ξ)2
Z  
1
u(x, t) = √ exp − h(ξ) dξ.
4πκt −∞ 4κt

46.3 Eigenfunction Expansions for Elliptic Equations

Consider a Green function problem for an elliptic equation on a finite domain.
L[G] = δ(x − ξ), x ∈ Ω (46.3)
B[G] = 0, x ∈ ∂Ω
Let the set of functions {φn } be orthonormal and complete on Ω. (Here n is the multi-index n = n1 , . . . , nd .)
Z
φn (x)φm (x) dx = δnm
Ω

1797
In addition, let the φn be eigenfunctions of L subject to the homogeneous boundary conditions.

L [φn ] = λn φn , B [φn ] = 0

We expand the Green function in the eigenfunctions.

X
G= gn φn (x)
n

Then we expand the Dirac Delta function.

X
δ(x − ξ) = dn φn (x)
n
Z
dn = φn (x)δ(x − ξ) dx
Ω
dn = φn (ξ)

We substitute the series expansions for the Green function and the Dirac Delta function into Equation 46.3.
X X
gn λn φn (x) = φn (ξ)φn (x)
n n

We equate coefficients to solve for the gn and hence determine the Green function.

φn (ξ)
gn =
λn
X φn (ξ)φn (x)
G(x; ξ) =
n
λn

Example 46.3.1 Consider the Green function for the reduced wave equation, ∆u − k 2 u in the rectangle, 0 ≤
x ≤ a, 0 ≤ y ≤ b, and vanishing on the sides.

1798
 First we find the eigenfunctions of the operator L = ∆ − k 2 = 0. Note that φ = X(x)Y (y) is an eigenfunction
∂2 ∂2
of L if X is an eigenfunction of ∂x 2 and Y is an eigenfunction of ∂y 2 . Thus we consider the two regular Sturm-

Liouville eigenvalue problems:

X 00 = λX, X(0) = X(a) = 0
Y 00 = λY, Y (0) = Y (b) = 0
This leads us to the eigenfunctions
 mπx   nπy 
φmn = sin sin .
a b
We use the orthogonality relation
Z 2π  mπx   nπx  a
sin sin dx = δmn
0 a a 2
to make the eigenfunctions orthonormal.
2  mπx   nπy 
φmn = √ sin sin , m, n ∈ Z+
ab a b
The φmn are eigenfunctions of L.
 
mπ 2  nπ 2
2
L [φmn ] = − + +k φmn
a b
By expanding the Green function and the Dirac Delta function in the φmn and substituting into the differential
equation we obtain the solution.
∞ √2 sin mπξ sin nπη √2 sin mπx sin nπy





a b a b
X ab ab
G=  
mπ 2 nπ 2



− + b +k 2
m,n=1 a
∞ mπξ nπy nπη
sin mπx





X
a
sin a
sin b
sin b
G(x, y; ξ, η) = −4ab 2 + (nπa)2 + (kab)2
m,n=1
(mπb)

1799
Example 46.3.2 Consider the Green function for Laplace’s equation, ∆u = 0 in the disk, |r| < a, and vanishing
at r = a.
First we find the eigenfunctions of the operator
∂2 1 ∂ 1 ∂2
∆= + + .
∂r2 r ∂r r2 ∂θ2
d2
We will look for eigenfunctions of the form φ = Θ(θ)R(r). We choose the Θ to be eigenfunctions of dθ2
subject
to the periodic boundary conditions in θ.
Θ00 = λΘ, Θ(0) = Θ(2π), Θ0 (0) = Θ0 (2π)
Θn = einθ , n∈Z
We determine R(r) by requiring that φ be an eigenfunction of ∆.
∆φ = λφ
1 1
(Θn R)rr + (Θn R)r + 2 (Θn R)θθ = λΘn R
r r
1 1
Θn R00 + Θn R0 + 2 (−n2 )Θn R = λΘR
r r
For notational convenience, we denote λ = −µ2 .

n2
 
00 1 0 2
R + R + µ − 2 R = 0, R(0) bounded, R(a) = 0
r r
The general solution for R is
R = c1 Jn (µr) + c2 Yn (µr).
The left boundary condition demands that c2 = 0. The right boundary condition determines the eigenvalues.
 
jn,m r jn,m
Rnm = Jn , µnm =
a a

1800
Here jn,m is the mth positive root of Jn . This leads us to the eigenfunctions
 
inθ jn,m r
φnm = e Jn
a
We use the orthogonality relations
Z 2π
e−imθ einθ dθ = 2πδmn ,
0
Z 1
1 0 2
rJν (jν,m r)Jν (jν,n r) dr = (J ν (jν,n )) δmn
0 2
to make the eigenfunctions orthonormal.
 
1 inθ jn,m r
φnm = √ e Jn , n ∈ Z, m ∈ Z+
πa|J 0 n (jn,m )| a
The φnm are eigenfunctions of L.
 2
jn,m
∆φnm = − φnm
a
By expanding the Green function and the Dirac Delta function in the φnm and substituting into the differential
equation we obtain the solution.
   
∞ X ∞ √ 1 −inϑ jn,m ρ 1 inθ jn,m r
X πa|J 0 n (jn,m )|
e J n a
√
πa|J 0 n (jn,m )|
e Jn a
G=  2
n=−∞ m=1 − jn,m a
∞ X ∞    
X 1 in(θ−ϑ) jn,m ρ jn,m r
G(r, θ; ρ, ϑ) = − e Jn Jn
n=−∞ m=1
π(jn,m J 0 n (jn,m ))2 a a

1801
46.4 The Method of Images
Consider the problem

∇2 u = f (x, y), −∞ < x < ∞, y > 0

u(x, 0) = 0, u(x, y) → 0 as (x2 + y 2 ) → ∞.

The equations for the Green function are

∇2 g = δ(x − ξ)δ(y − η), −∞ < x < ∞, y > 0

g(x, 0; ξ, η) = 0, g(x, y; ξ, η) → 0 as (x2 + y 2 ) → ∞.

To solve this problem we will use the method of images. We expand the domain to include the lower half
plane and solve the problem

∇2 g = δ(x − ξ)δ(y − η) − δ(x − ξ)δ(y + η), −∞ < x, ξ, y < ∞, η>0

g(x, y; ξ, η) → 0 as (x2 + y 2 ) → ∞.

Because of symmetry, g is zero for y = 0.

We solve the differential equation using the infinite space Green functions.
1 1
log (x − ξ)2 + (y − η)2 − log (x − ξ)2 + (y + η)2
   
g=
4π  4π
2 2

1 (x − ξ) + (y − η)
= log
4π (x − ξ)2 + (y + η)2

Thus we can write the solution

∞ ∞
(x − ξ)2 + (y − η)2
Z Z  
1
u(x, y) = log f (ξ, η) dξdη.
0 −∞ 4π (x − ξ)2 + (y + η)2

1802
46.5 Exercises
Exercise 46.1
Derive the causal Green function for the one dimensional wave equation on (−∞..∞). That is, solve

Gtt − c2 Gxx = δ(x − ξ)δ(t − τ ),

G(x, t; ξ, τ ) = 0 for t < τ.

Exercise 46.2
By reducing the problem to a series of one dimensional Green function problems, determine G(x, ξ) if

∇2 G = δ(x − ξ)

(a) on the rectangle 0 < x < L, 0 < y < H and

G(0, y; ξ, η) = Gx (L, y; ξ, η) = Gy (x, 0; ξ, η) = Gy (x, H; ξ, η) = 0

(b) on the box 0 < x < L, 0 < y < H, 0 < z < W with G = 0 on the boundary.

(c) on the semi-circle 0 < r < a, 0 < θ < π with G = 0 on the boundary.

(d) on the quarter-circle 0 < r < a, 0 < θ < π/2 with G = 0 on the straight sides and Gr = 0 at r = a.

Exercise 46.3
Using the method of multi-dimensional eigenfunction expansions, determine G(x, x0 ) if

∇2 G = δ(x − x0 )

and

1803
 (a) on the rectangle (0 < x < L, 0 < y < H)

∂G
at x = 0, G=0 at y = 0, =0
∂y

∂G ∂G
at x = L, =0 at y = H, =0
∂x ∂y

(b) on the rectangular shaped box (0 < x < L, 0 < y < H, 0 < z < W ) with G = 0 on the six sides.

(c) on the semi-circle (0 < r < a, 0 < θ < π) with G = 0 on the entire boundary.

(d) on the quarter-circle (0 < r < a, 0 < θ < π/2) with G = 0 on the straight sides and ∂G/∂r = 0 at r = a.

Exercise 46.4
Using the method of images solve

∇2 G = δ(x − x0 )

in the first quadrant (x ≥ 0 and y ≥ 0) with G = 0 at x = 0 and ∂G/∂y = 0 at y = 0. Use the Green function to
solve in the first quadrant

∇2 u = 0
u(0, y) = g(y)
∂u
(x, 0) = h(x).
∂y

1804
Exercise 46.5
Consider the wave equation defined on the half-line x > 0:
∂2u 2
2∂ u
= c + Q(x, t),
∂t2 ∂x2
u(x, 0) = f (x)
∂u
(x, 0) = g(x)
∂t
u(0, t) = h(t)
(a) Determine the appropriate Green’s function using the method of images.
(b) Solve for u(x, t) if Q(x, t) = 0, f (x) = 0, and g(x) = 0.
(c) For what values of t does h(t) influence u(x1 , t1 ). Interpret this result physically.

Exercise 46.6
Derive the Green functions for the one dimensional wave equation on (−∞..∞) for non-homogeneous initial
conditions. Solve the two problems
gtt − c2 gxx = 0, g(x, 0; ξ, τ ) = δ(x − ξ), gt (x, 0; ξ, τ ) = 0,
γtt − c2 γxx = 0, γ(x, 0; ξ, τ ) = 0, γt (x, 0; ξ, τ ) = δ(x − ξ),
using the Fourier transform.

Exercise 46.7
Use the Green functions from Problem 46.1 and Problem 46.6 to solve
utt − c2 uxx = f (x, t), x > 0, −∞ < t < ∞
u(x, 0) = p(x), ut (x, 0) = q(x).
Use the solution to determine the domain of dependence of the solution.

1805
Exercise 46.8
Show that the Green function for the reduced wave equation, ∆u−k 2 u = 0 in the rectangle, 0 ≤ x ≤ a, 0 ≤ y ≤ b,
and vanishing on the sides is:
∞  
2 X sinh(σn y< ) sinh(σn (y> − b))  nπx  nπξ
G(x, y; ξ, η) = sin sin ,
a n=1 σn sinh(σn b) a a

where
r
n2 π 2
σn = k2 + .
a2

Exercise 46.9
Find the Green function for the reduced wave equation ∆u−k 2 u = 0, in the quarter plane: 0 < x < ∞, 0 < y < ∞
subject to the mixed boundary conditions:
u(x, 0) = 0, ux (0, y) = 0.
Find two distinct integral representations for G(x, y; ξ, η).

Exercise 46.10
Show that in polar coordinates the Green function for ∆u = 0 in the infinite sector, 0 < θ < α, 0 < r < ∞, and
vanishing on the sides is given by,
  

π r π
1 cosh α
log ρ
− cos α
(θ − ϑ)
G(r, θ, ρ, ϑ) = log   
 .
4π cosh π
log r π
− cos (θ + ϑ)
α ρ α

Use this to find the harmonic function u(r, θ) in the given sector which takes on the boundary values:
(
0 for r < c
u(r, θ) = u(r, α) =
1 for r > c.

1806
Exercise 46.11
The Green function for the initial value problem,
ut − κuxx = 0, u(x, 0) = f (x),
on −∞ < x < ∞ is
1 2
G(x, t; ξ) = √ e−(x−ξ) /(4κt) .
4πκt
Use the method of images to find the corresponding Green function for the mixed initial-boundary problems:
i) ut = κuxx , u(x, 0) = f (x) for x > 0, u(0, t) = 0,
i) ut = κuxx , u(x, 0) = f (x) for x > 0, ux (0, t) = 0.

Exercise 46.12
Find the Green function (expansion) for the one dimensional wave equation utt − c2 uxx = 0 on the interval
0 < x < L, subject to the boundary conditions:
a) u(0, t) = ux (L, t) = 0,
b) ux (0, t) = ux (L, t) = 0.
Write the final forms in terms showing the propagation properties of the wave equation, i.e., with arguments
((x ± ξ) ± (t − τ )).

Exercise 46.13
Solve, using the above determined Green function,
utt − c2 uxx = 0, 0 < x < 1, t > 0,
ux (0, t) = ux (1, t) = 0,
u(x, 0) = x2 (1 − x)2 , ut (x, 0) = 1.
For c = 1, find u(x, t) at x = 3/4, t = 7/2.

1807
46.6 Hints
Hint 46.1

Hint 46.2
Take a Fourier transform in x. This will give you an ordinary differential equation Green function problem for Ĝ.
Find the continuity and jump conditions at t = τ . After solving for Ĝ, do the inverse transform with the aid of a
table.

Hint 46.3

Hint 46.4

Hint 46.5

Hint 46.6

Hint 46.7

1808
Hint 46.8
Use Fourier sine and cosine transforms.

Hint 46.9
The the conformal mapping z = wπ/α to map the sector to the upper half plane. The new problem will be

Gxx + Gyy = δ(x − ξ)δ(y − η), −∞ < x < ∞, 0 < y < ∞,

G(x, 0, ξ, η) = 0,
G(x, y, ξ, η) → 0 as x, y → ∞.

Solve this problem with the image method.

Hint 46.10

Hint 46.11

Hint 46.12

1809
46.7 Solutions
Solution 46.1

Gtt − c2 Gxx = δ(x − ξ)δ(t − τ ),

G(x, t; ξ, τ ) = 0 for t < τ.
We take the Fourier transform in x.
Ĝtt + c2 ω 2 G = F[δ(x − ξ)]δ(t − τ ), Ĝ(ω, 0; ξ, τ ) = Ĝt (ω, 0; ξ, τ ) = 0
Now we have an ordinary differential equation Green function problem for Ĝ. We have written the causality
condition, the Green function is zero for t < τ , in terms of initial conditions. The homogeneous solutions of the
ordinary differential equation are
{cos(cωt), sin(cωt)}.
It will be handy to use the fundamental set of solutions at t = τ :
 
1
cos(cω(t − τ )), sin(cω(t − τ )) .
cω
We write the solution for Ĝ and invert using the convolution theorem.
1
Ĝ = F[δ(x − ξ)]H(t − τ ) sin(cω(t − τ ))
h cω
π i
Ĝ = H(t − τ )F[δ(x − ξ)]F H(c(t − τ ) − |x|)
Z ∞ c
π 1
G = H(t − τ ) δ(y − ξ)H(c(t − τ ) − |x − y|) dy
c 2π −∞
1
G = H(t − τ )H(c(t − τ ) − |x − ξ|)
2c
1
G = H(c(t − τ ) − |x − ξ|)
2c

1810
 The Green function for ξ = τ = 0 and c = 1 is plotted in Figure 46.1 on the domain x ∈ (−1..1), t ∈ (0..1).
1
The Green function is a displacement of height 2c that propagates out from the point x = ξ in both directions
with speed c. The Green function shows the range of influence of a disturbance at the point x = ξ and time
t = τ . The disturbance influences the solution for all ξ − ct < x < ξ + ct and t > τ .
1
0.8
0.6
t
0.4

0.2

0.4

0.2
-1 -0.5 0 0.5 0
1
x

Figure 46.1: Green function for the wave equation.

Solution 46.2
1. We expand the Green function in eigenfunctions in x.
∞  
X (2n − 1)πx
G(x; ξ) = an (y) sin
n=1
2L

We substitute the expansion into the differential equation.

∞ r  
2
X 2 (2n − 1)πx
∇ an (y) sin = δ(x − ξ)δ(y − η)
n=1
L 2L

1811
 ∞  2 !r  
X (2n − 1)π 2 (2n − 1)πx
a00n (y) − an (y) sin
n=1
2L L 2L
∞ r  r  
X 2 (2n − 1)πξ 2 (2n − 1)πx
= δ(y − η) sin sin
n=1
L 2L L 2L

 2 r  
(2n − 1)π 2 (2n − 1)πξ
a00n (y) − an (y) = sin δ(y − η)
2L L 2L
From the boundary conditions at y = 0 and y = H, we obtain boundary conditions for the an (y).
a0n (0) = a0n (H) = 0.
The solutions that satisfy the left and right boundary conditions are
   
(2n − 1)πy (2n − 1)π(H − y)
an1 = cosh , an2 = cosh .
2L 2L
The Wronskian of these solutions is
 
(2n − 1)π (2n − 1)π
W =− sinh .
2L 2
Thus the solution for an (y) is
   
(2n−1)πy<
 cosh cosh (2n−1)π(H−y >)
r 
2 (2n − 1)πξ 2L 2L
an (y) = sin  
L 2L (2n−1)π
− 2L sinh (2n−1)π
2

√    
2 2L (2n − 1)π (2n − 1)πy<
an (y) = − csch cosh
(2n − 1)π 2 2L
   
(2n − 1)π(H − y> ) (2n − 1)πξ
cosh sin .
2L 2L

1812
 This determines the Green function.
√ ∞    
2 2L X 1 (2n − 1)π (2n − 1)πy<
G(x; ξ) = − csch cosh
π n=1 2n − 1 2 2L
     
(2n − 1)π(H − y> ) (2n − 1)πξ (2n − 1)πx
cosh sin sin
2L 2L 2L

2. We seek a solution of the form

∞
X 2  mπx   nπy 
G(x; ξ) = amn (z) √ sin sin .
m=1 LH L H
n=1

We substitute this into the differential equation.

∞
2
X 2  mπx   nπy 
∇ amn (z) √ sin sin = δ(x − ξ)δ(y − η)δ(z − ζ)
m=1 LH L H
n=1

∞    
X
00 mπ 2  nπ 2 2  mπx   nπy 
amn (z) − + amn (z) √ sin sin
m=1
L H LH L H
n=1
∞  
X 2 mπξ  nπη  2  mπx   nπy 
= δ(z − ζ) √ sin sin √ sin sin
m=1 LH L H LH L H
n=1

   n 2   
m 2 2 mπξ  nπη 
a00mn (z) −π + amn (z) = √ sin sin δ(z − ζ)
L H LH L H
From the boundary conditions on G, we obtain boundary conditions for the amn .
amn (0) = amn (W ) = 0

1813
The solutions that satisfy the left and right boundary conditions are
r  ! r  !
m 2  n 2 m 2  n 2
amn1 = sinh + πz , amn2 = sinh + π(W − z) .
L H L H

The Wronskian of these solutions is

r  r  !
m 2  n 2 m 2  n 2
W =− + π sinh + πW .
L H L H

Thus the solution for amn (z) is

 
2 mπξ  nπη 
amn (z) = √ sin sin
LH L H
q  q 
m 2 n 2 m 2 n 2





sinh +
L
πz< sinh
H
+ π(W − z> ) L H
q
q 
m 2 n 2 m 2 n 2




− L
+ H π sinh L
+ H πW

 
2 mπξ  nπη 
amn (z) = − √ csch (λmn πW ) sin sin
πλmn LH L H
sinh (λmn πz< ) sinh (λmn π(W − z> )) ,

where
r 
m 2  n 2
λmn = + .
L H

1814
 This determines the Green function.
∞  
4 X 1 mπξ  mπx 
G(x; ξ) = − csch (λmn πW ) sin sin
πLH m=1 λmn L L
n=1
 nπη   nπy 
sin sin sinh (λmn πz< ) sinh (λmn π(W − z> ))
H H

3. First we write the problem in circular coordinates.

∇2 G = δ(x − ξ)
1 1 1
Grr + Gr + 2 Gθθ = δ(r − ρ)δ(θ − ϑ),
r r r
G(r, 0; ρ, ϑ) = G(r, π; ρ, ϑ) = G(0, θ; ρ, ϑ) = G(a, θ; ρ, ϑ) = 0

Because the Green function vanishes at θ = 0 and θ = π we expand it in a series of the form
∞
X
G= gn (r) sin(nθ).
n=1

We substitute the series into the differential equation.

∞  ∞
n2

X
00 1 0 1 X 2
gn (r) + gn (r) − 2 gn (r) sin(nθ) = δ(r − ρ) sin(nϑ) sin(nθ)
n=1
r r r n=1
π
1 n2 2
gn00 (r) + gn0 (r) − 2 gn (r) = sin(nϑ)δ(r − ρ)
r r πr
From the boundary conditions on G, we obtain boundary conditions for the gn .

gn (0) = gn (a) = 0

1815
 The solutions that satisfy the left and right boundary conditions are
 r n  a n
gn1 = rn , gn2 = − .
a r
The Wronskian of these solutions is
2nan
W = .
r
Thus the solution for gn (r) is
  n 
r> n a
n


2 r< a
− r>
gn (r) = sin(nϑ) 2nan
πρ ρ
   n 
1  r< n
 r> n a
gn (r) = sin(nϑ) − .
nπ a a r>
This determines the solution.
∞   n 
X 1  r< n  r> n a
G= − sin(nϑ) sin(nθ)
n=1
nπ a a r>

4. First we write the problem in circular coordinates.

1 1 1
Grr + Gr + 2 Gθθ = δ(r − ρ)δ(θ − ϑ),
r r r
G(r, 0; ρ, ϑ) = G(r, π/2; ρ, ϑ) = G(0, θ; ρ, ϑ) = Gr (a, θ; ρ, ϑ) = 0

Because the Green function vanishes at θ = 0 and θ = π/2 we expand it in a series of the form
∞
X
G= gn (r) sin(2nθ).
n=1

1816
We substitute the series into the differential equation.
∞  ∞
4n2

X
00 1 0 1 X 4
gn (r) + gn (r) − 2 gn (r) sin(2nθ) = δ(r − ρ) sin(2nϑ) sin(2nθ)
n=1
r r r n=1
π
1 4n2 4
gn00 (r) + gn0 (r) − 2 gn (r) = sin(2nϑ)δ(r − ρ)
r r πr
From the boundary conditions on G, we obtain boundary conditions for the gn .
gn (0) = gn0 (a) = 0
The solutions that satisfy the left and right boundary conditions are
 r 2n  a 2n
gn1 = r2n , gn2 = + .
a r
The Wronskian of these solutions is
4na2n
W =− .
r
Thus the solution for gn (r) is
  2n 
r> 2n a
2n


r< a
+ r>
4
gn (r) = sin(2nϑ) 2n
πρ − 4naρ
 2n !
1  r 2n
<
 r 2n
> a
gn (r) = − sin(2nϑ) +
πn a a r>

This determines the solution.

∞  2n !
X 1  r< 2n  r> 2n a
G=− + sin(2nϑ) sin(2nθ)
n=1
πn a a r>

1817
Solution 46.3
1. The set

  ∞
(2m − 1)πx
{Xn } = sin
2L m=1

are eigenfunctions of ∇2 and satisfy the boundary conditions Xn (0) = Xn0 (L) = 0. The set

n  nπy o∞
{Yn } = cos
H n=0

are eigenfunctions of ∇2 and satisfy the boundary conditions Yn0 (0) = Yn0 (H) = 0. The set

 
(2m − 1)πx
  nπy ∞
sin cos
2L H m=1,n=0

are eigenfunctions of ∇2 and satisfy the boundary conditions of this problem. We expand the Green function
in a series of these eigenfunctions.

∞ r   ∞  
X 2 (2m − 1)πx X 2 (2m − 1)πx  nπy 
G= gm0 sin + gmn √ sin cos
m=1
LH 2L m=1 LH 2L H
n=1

We substitute the series into the Green function differential equation.

∆G = δ(x − ξ)δ(y − η)

1818
 ∞  2 r  
X (2m − 1)π 2 (2m − 1)πx
− gm0 sin
m=1
2L LH 2L
∞  2  !  
X (2m − 1)π nπy 2 2 (2m − 1)πx  nπy 
− gmn + √ sin cos
m=1
2L H LH 2L H
n=1
∞ r  r  
X 2 (2m − 1)πξ 2 (2m − 1)πx
= sin sin
m=1
LH 2L LH 2L
∞    
X 2 (2m − 1)πξ  nπη  2 (2m − 1)πx  nπy 
+ √ sin cos √ sin cos
m=1 LH 2L H LH 2L H
n=1

We equate terms and solve for the coefficients gmn .

r  2  
2 2L (2m − 1)πξ
gm0 = − sin
LH (2m − 1)π 2L
 
2 1 (2m − 1)πξ  nπη 
gmn = − √   sin cos
LH π 2 2m−1 2 + n 2 2L H



2L H

This determines the Green function.

2. Note that
(r   )
8 kπx  mπy   nπz 
sin , sin , sin : k, m, n ∈ Z+
LHW L H W

is orthonormal and complete on (0 . . . L) × (0 . . . H) × (0 . . . W ). The functions are eigenfunctions of ∇2 .

We expand the Green function in a series of these eigenfunctions.
∞ r  
X 8 kπx  mπy   nπz 
G= gkmn sin sin sin
k,m,n=1
LHW L H W

1819
 We substitute the series into the Green function differential equation.

∆G = δ(x − ξ)δ(y − η)δ(z − ζ)

∞ 2 !
X 
kπ  mπ 2  nπ 2 r 8

kπx
  mπy   nπz 
− gkmn + + sin sin sin
k,m,n=1
L H W LHW L H W
∞ r    
X 8 kπξ  mπη  nπζ
= sin sin sin
k,m,n=1
LHW L H W
r  
8 kπx  mπy   nπz 
sin sin sin
LHW L H W
We equate terms and solve for the coefficients gkmn .
q
8 kπξ mπη nπζ




LHW
sin L
sin H
sin W
gkmn = − 

2 2 2

π 2 Lk + H m
+ Wn

This determines the Green function.

3. The Green function problem is
1 1 1
∆G ≡ Grr + Gr + 2 Gθθ = δ(r − ρ)δ(θ − ϑ).
r r r
We seek a set of functions {Θn (θ)Rnm (r)} which are orthogonal and complete on (0 . . . a) × (0 . . . π) and
∂2
which are eigenfunctions of the laplacian. For the Θn we choose eigenfunctions of ∂θ 2.

Θ00 = −ν 2 Θ, Θ(0) = Θ(π) = 0

νn = n, Θn = sin(nθ), n ∈ Z+

1820
Now we look for eigenfunctions of the laplacian.
1 1
(RΘn )rr + (RΘn )r + 2 (RΘn )θθ = −µ2 RΘn
r r
1 n2
R00 Θn + R0 Θn − 2 RΘn = −µ2 RΘn
 r 2
 r
1 n
R00 + R0 + µ2 − 2 R = 0, R(0) = R(a) = 0
r r
The general solution for R is

R = c1 Jn (µr) + c2 Yn (µr).

the solution that satisfies the left boundary condition is R = cJn (µr). We use the right boundary condition
to determine the eigenvalues.
 
jn,m jn,m r
µm = , Rnm = Jn , m, n ∈ Z+
a a

here jn,m is the mth root of Jn .

Note that
   
jn,m r +
sin(nθ)Jn : m, n ∈ Z
a
is orthogonal and complete on (r, θ) ∈ (0 . . . a) × (0 . . . π). We use the identities
Z π Z 1
2 π 1 2
sin (nθ) dθ = , rJn2 (jn,m r) dr = Jn+1 (jn,m )
0 2 0 2
to make the functions orthonormal.
   
2 jn,m r +
√ sin(nθ)Jn : m, n ∈ Z
π a|Jn+1 (jn,m )| a

1821
 We expand the Green function in a series of these eigenfunctions.
∞  
X 2 jn,m r
G= gnm √ Jn sin(nθ)
n,m=1
π a|Jn+1 (jn,m )| a

We substitute the series into the Green function differential equation.

1 1 1
Grr + Gr + 2 Gθθ = δ(r − ρ)δ(θ − ϑ)
r r r
∞  2  
X jn,m 2 jn,m r
− gnm √ Jn sin(nθ)
n,m=1
a π a|Jn+1 (jn,m )| a
∞    
X 2 jn,m ρ 2 jn,m r
= √ Jn sin(nϑ) √ Jn sin(nθ)
n,m=1
π a|J n+1 (jn,m )| a π a|Jn+1 (j n,m )| a

We equate terms and solve for the coefficients gmn .

 2  
a 2 jn,m ρ
gnm = − √ Jn sin(nϑ)
jn,m π a|Jn+1 (jn,m )| a
This determines the green function.
4. The Green function problem is
1 1 1
∆G ≡ Grr + Gr + 2 Gθθ = δ(r − ρ)δ(θ − ϑ).
r r r
We seek a set of functions {Θn (θ)Rnm (r)} which are orthogonal and complete on (0 . . . a) × (0 . . . π/2) and
∂2
which are eigenfunctions of the laplacian. For the Θn we choose eigenfunctions of ∂θ 2.

Θ00 = −ν 2 Θ, Θ(0) = Θ(π/2) = 0

νn = 2n, Θn = sin(2nθ), n ∈ Z+

1822
Now we look for eigenfunctions of the laplacian.
1 1
(RΘn )rr + (RΘn )r + 2 (RΘn )θθ = −µ2 RΘn
r r
2
1 (2n)
R00 Θn + R0 Θn − 2 RΘn = −µ2 RΘn
 r 2
r
1 (2n)
R00 + R0 + µ2 − 2 R = 0, R(0) = R(a) = 0
r r

The general solution for R is

R = c1 J2n (µr) + c2 Y2n (µr).

the solution that satisfies the left boundary condition is R = cJ2n (µr). We use the right boundary condition
to determine the eigenvalues.
j 0 2n,m
 0
j 2n,m r

µm = , Rnm = J2n , m, n ∈ Z+
a a

here j 0 n,m is the mth root of J 0 n .

Note that
j 0 2n,m r
   
0 +
sin(2nθ)J 2n : m, n ∈ Z
a

is orthogonal and complete on (r, θ) ∈ (0 . . . a) × (0 . . . π/2). We use the identities

Z π
π
sin(mθ) sin(nθ) dθ = δmn ,
0 2
02
j ν,n − ν 2
Z 1
0 0 0

2
rJν (j ν,m r)Jν (j ν,n r) dr = Jν (j ν,n ) δmn
0 2j 0 2ν,n

1823
to make the functions orthonormal.
 
2j 0 2n,m
 0
j 2n,m r
  
+
sin(2nθ)J2n : m, n ∈ Z
 √π a j 0 2
q
− 4n 2 |J (j 0 )| a 
2n,m 2n 2n,m

We expand the Green function in a series of these eigenfunctions.

∞
2j 0 2n,m j 0 2n,m r
X  
G= gnm √ q J2n sin(2nθ)
n,m=1
0 2 2 0
π a j 2n,m − 4n |J2n (j 2n,m )| a

We substitute the series into the Green function differential equation.

1 1 1
Grr + Gr + 2 Gθθ = δ(r − ρ)δ(θ − ϑ)
r r r

∞  0
j 2n,m 2 2j 0 2n,m j 0 2n,m r
X   
− gnm √ q J2n sin(2nθ)
a π a j 02 2 0 a
n,m=1 2n,m − 4n |J2n (j 2n,m )|
∞
2j 0 2n,m j 0 2n,m ρ
X  
= √ q 02 J2n sin(2nϑ)
n,m=1 π a j 2n,m − 4n2 |J2n (j 0 2n,m )| a

2j 0 2n,m j 0 2n,m r
 
√ q 02 J2n sin(2nθ)
π a j 2n,m − 4n2 |J2n (j 0 2n,m )| a

We equate terms and solve for the coefficients gmn .

2
2j 0 2n,m
  0
j 2n,m ρ

a
gnm = − 0 √ q 02 J2n sin(2nϑ)
j 2n,m 2 0
π a j 2n,m − 4n |J2n (j 2n,m )| a

This determines the green function.

1824
Solution 46.4
We start with the equation
∇2 G = δ(x − ξ)δ(y − η).
We do an odd reflection across the y axis so that G(0, y; ξ, η) = 0.
∇2 G = δ(x − ξ)δ(y − η) − δ(x + ξ)δ(y − η)
Then we do an even reflection across the x axis so that Gy (x, 0; ξ, η) = 0.
∇2 G = δ(x − ξ)δ(y − η) − δ(x + ξ)δ(y − η) + δ(x − ξ)δ(y + η) − δ(x + ξ)δ(y + η)
We solve this problem using the infinite space Green function.
1 1
log (x − ξ)2 + (y − η)2 − log (x + ξ)2 + (y − η)2



G=
4π 4π
1 1
log (x − ξ)2 + (y + η)2 − log (x + ξ)2 + (y + η)2



+
4π 4π
((x − ξ)2 + (y − η)2 ) ((x − ξ)2 + (y + η)2 )
 
1
G= log
4π ((x + ξ)2 + (y − η)2 ) ((x + ξ)2 + (y + η)2 )
Now we solve the boundary value problem.
Z   Z
∂G ∂u(x, y)
u(ξ, η) = u(x, y) −G dS + G∆u dV
S ∂n ∂n V
Z 0 Z ∞
u(ξ, η) = u(0, y)(−Gx (0, y; ξ, η)) dy + −G(x, 0; ξ, η)(−uy (x, 0)) dx
∞ 0
Z ∞ Z ∞
u(ξ, η) = g(y)Gx (0, y; ξ, η) dy + G(x, 0; ξ, η)h(x) dx
0 0
ξ ∞
Z ∞
(x − ξ)2 + η 2
Z    
1 1 1
u(ξ, η) = − + g(y) dy + log h(x) dx
π 0 ξ 2 + (y − η)2 ξ 2 + (y + η)2 2π 0 (x + ξ)2 + η 2
x ∞
Z ∞
(x − ξ)2 + y 2
Z    
1 1 1
u(x, y) = − + g(η) dη + log h(ξ) dξ
π 0 x2 + (y − η)2 x2 + (y + η)2 2π 0 (x + ξ)2 + y 2

1825
Solution 46.5
First we find the infinite space Green function.

Gtt − c2 Gxx = δ(x − ξ)δ(t − τ ), G = Gt = 0 for t < τ

We solve this problem with the Fourier transform.

Ĝtt + c2 ω 2 Ĝ = F[δ(x − ξ)]δ(t − τ )

1
Ĝ = F[δ(x − ξ)]H(t − τ ) sin(cω(t − τ ))
h cω
π i
Ĝ = H(t − τ )F[δ(x − ξ)]F H(c(t − τ ) − |x|)
Z ∞ c
π 1
G = H(t − τ ) δ(y − ξ)H(c(t − τ ) − |x − y|) dy
c 2π −∞
1
G = H(t − τ )H(c(t − τ ) − |x − ξ|)
2c
1
G = H(c(t − τ ) − |x − ξ|)
2c
1. So that the Green function vanishes at x = 0 we do an odd reflection about that point.

Gtt − c2 Gxx = δ(x − ξ)δ(t − τ ) − δ(x + ξ)δ(t − τ )

1 1
G = H(c(t − τ ) − |x − ξ|) − H(c(t − τ ) − |x + ξ|)
2c 2c

2. Note that the Green function satisfies the symmetry relation

G(x, t; ξ, τ ) = G(ξ, −τ ; x, −t).

This implies that

Gxx = Gξξ , Gtt = Gτ τ .

1826
We write the Green function problem and the inhomogeneous differential equation for u in terms of ξ and
τ.

Gτ τ − c2 Gξξ = δ(x − ξ)δ(t − τ ) (46.4)

uτ τ − c2 uξξ = Q(ξ, τ ) (46.5)

We take the difference of u times Equation 46.4 and G times Equation 46.5 and integrate this over the
domain (0, ∞) × (0, t+ ).
Z t+ Z ∞ Z t+ Z ∞
uGτ τ − uτ τ G − c2 (uGξξ − uξξ G) dξ dτ


(uδ(x − ξ)δ(t − τ ) − GQ) dξ dτ =
0 0 0 0
Z t+ Z ∞ Z t+ Z ∞  
∂ 2 ∂
u(x, t) = GQ dξ dτ + (uGτ − uτ G) − c (uGξ − uξ G) dξ dτ
0 0 0 0 ∂τ ∂ξ
Z t+ Z ∞ Z ∞ Z t+
u(x, t) = GQ dξ dτ + t+
[uGτ − uτ G]0 dξ − c 2
[uGξ − uξ G]∞
0 dτ
0 0 0 0
Z t+ Z ∞ Z ∞ Z t+
2
u(x, t) = GQ dξ dτ − [uGτ − uτ G]τ =0 dξ + c [uGξ ]ξ=0 dτ
0 0 0 0

We consider the case Q(x, t) = f (x) = g(x) = 0.

Z t+
2
u(x, t) = c h(τ )Gξ (x, t; 0, τ ) dτ
0

We calculate Gξ .
1
G= (H(c(t − τ ) − |x − ξ|) − H(c(t − τ ) − |x + ξ|))
2c
1
Gξ = (δ(c(t − τ ) − |x − ξ|)(−1) sign (x − ξ)(−1) − δ(c(t − τ ) − |x + ξ|)(−1) sign (x + ξ))
2c
1
Gξ (x, t; 0, η) = δ(c(t − τ ) − |x|) sign (x)
c

1827
 We are interested in x > 0.

1
Gξ (x, t; 0, η) = δ(c(t − τ ) − x)
c
Now we can calculate the solution u.
Z t+
2 1
u(x, t) = c h(τ ) δ(c(t − τ ) − x) dτ
0 c
Z t+  x
u(x, t) = h(τ )δ (t − τ ) − dτ
0 c
 x
u(x, t) = h t −
c

3. The boundary condition influences the solution u(x1 , t1 ) only at the point t = t1 − x1 /c. The contribution
from the boundary condition u(0, t) = h(t) is a wave moving to the right with speed c.

Solution 46.6

gtt − c2 gxx = 0, g(x, 0; ξ, τ ) = δ(x − ξ), gt (x, 0; ξ, τ ) = 0

ĝtt + c2 ω 2 ĝxx = 0, ĝ(x, 0; ξ, τ ) = F[δ(x − ξ)], ĝt (x, 0; ξ, τ ) = 0
ĝ = F[δ(x − ξ)] cos(cωt)
ĝ = F[δ(x − ξ)]F[π(δ(x + ct) + δ(x − ct))]
Z ∞
1
g= δ(η − ξ)π(δ(x − η + ct) + δ(x − η − ct)) dη
2π −∞
1
g(x, t; ξ) = (δ(x − ξ + ct) + δ(x − ξ − ct))
2

1828
 γtt − c2 γxx = 0, γ(x, 0; ξ, τ ) = 0, γt (x, 0; ξ, τ ) = δ(x − ξ)
γ̂tt + c2 ω 2 γ̂xx = 0, γ̂(x, 0; ξ, τ ) = 0, γ̂t (x, 0; ξ, τ ) = F[δ(x − ξ)]
1
γ̂ = F[δ(x − ξ)] sin(cωt)
hπ cω i
γ̂ = F[δ(x − ξ)]F (H(x + ct) + H(x − ct))
Z ∞ c
1 π
γ= δ(η − ξ) (H(x − η + ct) + H(x − η − ct)) dη
2π −∞ c
1
γ(x, t; ξ) = (H(x − ξ + ct) + H(x − ξ − ct))
2c

Solution 46.7

Z ∞ Z ∞ Z ∞ Z ∞
u(x, t) = G(x, t; ξ, τ )f (ξ, τ ) dξ dτ + g(x, t; ξ)p(ξ) dξ + γ(x, t; ξ)q(ξ) dξ
0 −∞ −∞ −∞

Z ∞ Z ∞
1
u(x, t) = H(t − τ )(H(x − ξ + c(t − τ )) − H(x − ξ − c(t − τ )))f (ξ, τ ) dξ dτ
2c 0 −∞
Z ∞
1 ∞
Z
1
+ (δ(x − ξ + ct) + δ(x − ξ − ct))p(ξ) dξ + (H(x − ξ + ct) + H(x − ξ − ct))q(ξ) dξ
2 −∞ 2c −∞

Z tZ ∞
1
u(x, t) = (H(x − ξ + c(t − τ )) − H(x − ξ − c(t − τ )))f (ξ, τ ) dξ dτ
2c 0 −∞
Z x+ct
1 1
+ (p(x + ct) + p(x − ct)) + q(ξ) dξ
2 2c x−ct

1829
 Z tZ x+c(t−τ ) Z x+ct
1 1 1
u(x, t) = f (ξ, τ ) dξ dτ + (p(x + ct) + p(x − ct)) + q(ξ) dξ
2c 0 x−c(t−τ ) 2 2c x−ct

This solution demonstrates the domain of dependence of the solution. The first term is an integral over the
triangle domain {(ξ, τ ) : 0 < τ < t, x − cτ < ξ < x + cτ }. The second term involves only the points (x ± ct, 0).
The third term is an integral on the line segment {(ξ, 0) : x − ct < ξ < x + ct}. In totallity, this is just the triangle
domain. This is shown graphically in Figure 46.2.

x,t

Domain of
Dependence

x-ct x+ct

Figure 46.2: Domain of dependence for the wave equation.

Solution 46.8
Single Sum Representation. First we find the eigenfunctions of the homogeneous problem ∆u − k 2 u = 0. We
substitute the separation of variables, u(x, y) = X(x)Y (y) into the partial differential equation.

X 00 Y + XY 00 − k 2 XY = 0
X 00 Y 00
= k2 − = −λ2
X Y

1830
We have the regular Sturm-Liouville eigenvalue problem,

X 00 = −λ2 X, X(0) = X(a) = 0,

which has the solutions,

nπ  nπx 
λn = , Xn = sin , n ∈ N.
a a
We expand the solution u in a series of these eigenfunctions.
∞
X  nπx 
G(x, y; ξ, η) = cn (y) sin
n=1
a

We substitute this series into the partial differential equation to find equations for the cn (y).
∞   
X nπ 2 00 2
 nπx 
− cn (y) + cn (y) − k cn (y) sin = δ(x − ξ)δ(y − η)
n=1
a a

The series expansion of the right side is,

∞
X  nπx 
δ(x − ξ)δ(y − η) = dn (y) sin
n=1
a
Z a
2  nπx 
dn (y) = δ(x − ξ)δ(y − η) sin dx
a 0 a
 
2 nπξ
dn (y) = sin δ(y − η).
a a

The the equations for the cn (y) are

  nπ 2   
00 2 2 nπξ
cn (y) − k + cn (y) = sin δ(y − η), cn (0) = cn (b) = 0.
a a a

1831
 p
The homogeneous solutions are {cosh(σn y), sinh(σn y)}, where σn = k 2 (nπ/a)2 . The solutions that satisfy the
boundary conditions at y = 0 and y = b are, sinh(σn y) and sinh(σn (y − b)), respectively. The Wronskian of these
solutions is,
 
 sinh(σn y) sinh(σ n (y − b)) 
W (y) =  
σn cosh(σn y) σn cosh(σn (y − b))
= σn (sinh(σn y) cosh(σn (y − b)) − sinh(σn (y − b)) cosh(σn y))
= σn sinh(σn b).

The solution for cn (y) is

 
2 nπξ sinh(σn y< ) sinh(σn (y> − b))
cn (y) = sin .
a a σn sinh(σn b)

The Green function for the partial differential equation is

∞  
2 X sinh(σn y< ) sinh(σn (y> − b))  nπx  nπξ
G(x, y; ξ, η) = sin sin .
a n=1 σn sinh(σn b) a a

Solution 46.9
We take the Fourier cosine transform in x of the partial differential equation and the boundary condition along
y = 0.

Gxx + Gyy − k 2 G = δ(x − ξ)δ(y − η)

1 1
−α2 Ĝ(α, y) − Ĝx (0, y) + Ĝyy (α, y) − k 2 Ĝ(α, y) = cos(αξ)δ(y − η)
π π
1
Ĝyy (α, y) − (k 2 + α2 )Ĝ(α, y) == cos(αξ)δ(y − η), Ĝ(α, 0) = 0
π

1832
Then we take the Fourier sine transform in y.

ˆ βˆ ˆ 1
−β 2 Ĝ(α, β) + Ĝ(α, 0) − (k 2 + α2 )Ĝ(α, β) = 2 cos(αξ) sin(βη)
π π
ˆ cos(αξ) sin(βη)
Ĝ = − 2 2
π (k + α2 + β 2 )

We take two inverse transforms to find the solution. For one integral representation of the Green function we take
the inverse sine transform followed by the inverse cosine transform.

ˆ sin(βη) 1
Ĝ = − cos(αξ)
π π(k + α2 + β 2 )
2

√
 
ˆ 1 − k2 +α2 y
Ĝ = − cos(αξ)Fs [δ(y − η)]Fc √ e
k 2 + α2
1
Z ∞
1   √   √ 
Ĝ(α, y) = − cos(αξ) δ(z − η) √ exp − k 2 + α2 |y − z| − exp − k 2 + α2 (y + z) dz
2π 0 k 2 + α2
cos(αξ)   √   √ 
Ĝ(α, y) = − √ exp − k 2 + α2 |y − η| − exp − k 2 + α2 (y + η)
2π k 2 + α2
1 ∞ cos(αξ) 
Z  √   √ 
G(x, y; ξ, η) = − √ exp − k 2 + α2 |y − η| − exp − k 2 + α2 (y + η) dα
π 0 k 2 + α2

For another integral representation of the Green function, we take the inverse cosine transform followed by the

1833
inverse sine transform.
ˆ cos(αξ) 1
Ĝ(α, β) = − sin(βη)
π π(k + α2 + β 2 )
2

√
" #
ˆ 1 − k2 +β 2 x
Ĝ(α, β) = − sin(βη)Fc [δ(x − ξ)]Fc p e
k2 + β 2
1
Z ∞
1  √ √ 
− k2 +β 2 |x−z| − k2 +β 2 (x+z)
Ĝ(x, β) = − sin(βη) δ(z − ξ) p e +e dz
2π 0 k2 + β 2
1 1  √ √ 
2 2 2 2
Ĝ(x, β) = − sin(βη) p e− k +β |x−ξ| + e− k +β (x+ξ)
2π k 2 + β 2
sin(βy) sin(βη)  −√k2 +β 2 |x−ξ| √2 2
Z ∞
1 
G(x, y; ξ, η) = − p e + e− k +β (x+ξ) dβ
π 0 k2 + β 2

Solution 46.10
The problem is:
1 1 δ(r − ρ)δ(θ − ϑ)
Grr + Gr + 2 Gθθ = , 0 < r < ∞, 0 < θ < α,
r r r
G(r, 0, ρ, ϑ) = G(r, α, ρ, ϑ) = 0,
G(0, θ, ρ, ϑ) = 0
G(r, θ, ρ, ϑ) → 0 as r → ∞.

Let w = r eiθ and z = x + iy. We use the conformal mapping, z = wπ/α to map the sector to the upper half z
plane. The problem is (x, y) space is
Gxx + Gyy = δ(x − ξ)δ(y − η), −∞ < x < ∞, 0 < y < ∞,
G(x, 0, ξ, η) = 0,
G(x, y, ξ, η) → 0 as x, y → ∞.

1834
We will solve this problem with the method of images. Note that the solution of,
Gxx + Gyy = δ(x − ξ)δ(y − η) − δ(x − ξ)δ(y + η), −∞ < x < ∞, −∞ < y < ∞,
G(x, y, ξ, η) → 0 as x, y → ∞,
satisfies the condition, G(x, 0, ξ, η) = 0. Since the infinite space Green function for the Laplacian in two dimensions
is
1
log (x − ξ)2 + (y − η)2 ,


4π
the solution of this problem is,
1 1
log (x − ξ)2 + (y − η)2 − log (x − ξ)2 + (y + η)2



G(x, y, ξ, η) =
4π  4π
(x − ξ)2 + (y − η)2

1
= log .
4π (x − ξ)2 + (y + η)2
Now we solve for x and y in the conformal mapping.
z = wπ/α = (r eiθ )π/α
x + iy = rπ/α (cos(θπ/α) + i sin(θπ/α))
x = rπ/α cos(θπ/α), y = rπ/α sin(θπ/α)
We substitute these expressions into G(x, y, ξ, η) to obtain G(r, θ, ρ, ϑ).
 π/α
(r cos(θπ/α) − ρπ/α cos(ϑπ/α))2 + (rπ/α sin(θπ/α) − ρπ/α sin(ϑπ/α))2

1
G(r, θ, ρ, ϑ) = log
4π (rπ/α cos(θπ/α) − ρπ/α cos(ϑπ/α))2 + (rπ/α sin(θπ/α) + ρπ/α sin(ϑπ/α))2
 2π/α
+ ρ2π/α − 2rπ/α ρπ/α cos(π(θ − ϑ)/α)

1 r
= log 2π/α
4π r + ρ2π/α − 2rπ/α ρπ/α cos(π(θ + ϑ)/α)
(r/ρ)π/α /2 + (ρ/r)π/α /2 − cos(π(θ − ϑ)/α)
 
1
= log
4π (r/ρ)π/α /2 + (ρ/r)π/α /2 − cos(π(θ + ϑ)/α)
 π log(r/ρ)/α
/2 + eπ log(ρ/r)/α /2 − cos(π(θ − ϑ)/α)

1 e
= log
4π eπ log(r/ρ)/α /2 + eπ log(ρ/r)/α /2 − cos(π(θ + ϑ)/α)

1835
    
π/α r
1 cosh log ρ
− cos(π(θ − ϑ)/α)
G(r, θ, ρ, ϑ) = log    
4π cosh π/α r
− cos(π(θ + ϑ)/α)
log ρ

Now recall that the solution of

∆u = f (x),

subject to the boundary condition,

u(x) = g(x),

is
Z Z I
u(x) = f (ξ)G(x; ξ) dAξ + g(ξ)∇ξ G(x; ξ) · n̂ dsξ .

The normal directions along the lower and upper edges of the sector are −θ̂ and θ̂, respectively. The gradient in
polar coordinates is

∂ ϑ̂ ∂
∇ξ = ρ̂ + .
∂ρ ρ ∂ϑ

We only need to compute the ϑ̂ component of the gradient of G. This is

1 ∂ sin(π(θ − ϑ)/α) sin(π(θ − ϑ)/α)
G=    +    
ρ ∂ρ π r
4αρ cosh α log ρ − cos(π(θ − ϑ)/α) π r
4αρ cosh α log ρ − cos(π(θ + ϑ)/α)

Along ϑ = 0, this is
1 sin(πθ/α)
Gϑ (r, θ, ρ, 0) =    .
ρ π r
2αρ cosh α log ρ − cos(πθ/α)

1836
Along ϑ = α, this is

1 sin(πθ/α)
Gϑ (r, θ, ρ, α) = −    .
ρ 2αρ cosh π log r + cos(πθ/α)
α ρ

The solution of our problem is

Z c Z ∞
sin(πθ/α) sin(πθ/α)
u(r, θ) = −     dρ + −     dρ
∞ 2αρ cosh απ log ρr + cos(πθ/α) c 2αρ cosh απ log ρr − cos(πθ/α)
Z ∞
− sin(πθ/α) sin(πθ/α)
u(r, θ) =    +     dρ
π r π r
c 2αρ cosh α log ρ − cos(πθ/α) 2αρ cosh α log ρ + cos(πθ/α)
   Z ∞
1 πθ πθ 1
u(r, θ) = − sin cos   
 dρ
α α α c ρ cosh2 απ log ρr − cos2 πθ α
   Z ∞
1 πθ πθ 1
u(r, θ) = − sin cos 2 πx


2 πθ

dx
α α α log(c/r) cosh α
− cos α
   Z ∞
2 πθ πθ 1
u(r, θ) = − sin cos 2πx

dx
− cos 2πθ


α α α log(c/r) cosh α α

Solution 46.11
First consider the Green function for

ut − κuxx = 0, u(x, 0) = f (x).

The differential equation and initial condition is

Gt = κGxx , G(x, 0; ξ) = δ(x − ξ).

1837
The Green function is a solution of the homogeneous heat equation for the initial condition of a unit amount of
heat concentrated at the point x = ξ. You can verify that the Green function is a solution of the heat equation
for t > 0 and that it has the property:
Z ∞
G(x, t; ξ) dx = 1, for t > 0.
−∞

This property demonstrates that the total amount of heat is the constant 1. At time t = 0 the heat is concentrated
at the point x = ξ. As time increases, the heat diffuses out from this point.
The solution for u(x, t) is the linear combination of the Green functions that satisfies the initial condition
u(x, 0) = f (x). This linear combination is
Z ∞
u(x, t) = G(x, t; ξ)f (ξ) dξ.
−∞

G(x, t; 1) and G(x, t; −1) are plotted in Figure 46.3 for the domain t ∈ [1/100..1/4], x ∈ [−2..2] and κ = 1.
Now we consider the problem

ut = κuxx , u(x, 0) = f (x) for x > 0, u(0, t) = 0.

Note that the solution of

Gt = κGxx , x > 0, t > 0,

G(x, 0; ξ) = δ(x − ξ) − δ(x + ξ),

satisfies the boundary condition G(0, t; ξ) = 0. We write the solution as the difference of infinite space Green
functions.
1 2 1 2
G(x, t; ξ) = √ e−(x−ξ) /(4κt) − √ e−(x+ξ) /(4κt)
4πκt 4πκt
1  −(x−ξ)2 /(4κt) 2

=√ e − e−(x+ξ) /(4κt)
4πκt

1838
 2 2
1
1
0
0

0.1
-1

0.2
-2

Figure 46.3: G(x, t; 1) and G(x, t; −1)

 
1 2 2 xξ
G(x, t; ξ) = √ e−(x +ξ )/(4κt) sinh
4πκt 2κt

Next we consider the problem

ut = κuxx , u(x, 0) = f (x) for x > 0, ux (0, t) = 0.

Note that the solution of

Gt = κGxx , x > 0, t > 0,

G(x, 0; ξ) = δ(x − ξ) + δ(x + ξ),

1839
satisfies the boundary condition Gx (0, t; ξ) = 0. We write the solution as the sum of infinite space Green functions.
1 2 1 2
G(x, t; ξ) = √ e−(x−ξ) /(4κt) + √ e−(x+ξ) /(4κt)
4πκt 4πκt
 
1 −(x2 +ξ 2 )/(4κt) xξ
G(x, t; ξ) = √ e cosh
4πκt 2κt
The Green functions for the two boundary conditions are shown in Figure 46.4.

2 1 2 1
1 0.8 1 0.8
0 0.6 0 0.6
0.05 0.4 0.05 0.4
0.1 0.2 0.1 0.2
0.15 0.15
0.2 0 0.2 0
0.25 0.25

Figure 46.4: Green functions for the boundary conditions u(0, t) = 0 and ux (0, t) = 0.

Solution 46.12

1840
 a) The Green function problem is
Gtt − c2 Gxx = δ(t − τ )δ(x − ξ), 0 < x < L, t > 0,
G(0, t; ξ, τ ) = Gx (L, t; ξ, τ ) = 0,
G(x, t; ξ, τ ) = 0 for t < τ.
The condition that G is zero for t < τ makes this a causal Green function. We solve this problem by expanding
G in a series of eigenfunctions of the x variable. The coefficients in the expansion will be functions of t. First we
find the eigenfunctions of x in the homogeneous problem. We substitute the separation of variables u = X(x)T (t)
into the homogeneous partial differential equation.
XT 00 = c2 X 00 T
T 00 X 00
2
= = −λ2
cT X
The eigenvalue problem is
X 00 = −λ2 X, X(0) = X 0 (L) = 0,
which has the solutions,
 
(2n − 1)π (2n − 1)πx
λn = , Xn = sin , n ∈ N.
2L 2L
The series expansion of the Green function has the form,
∞  
X (2n − 1)πx
G(x, t; ξ, τ ) = gn (t) sin .
n=1
2L

We determine the coefficients by substituting the expansion into the Green function differential equation.
Gtt − c2 Gxx = δ(x − ξ)δ(t − τ )
∞  2 !  
X
00 (2n − 1)πc (2n − 1)πx
gn (t) + gn (t) sin = δ(x − ξ)δ(t − τ )
n=1
2L 2L

1841
We need to expand the right side of the equation in the sine series
∞  
X (2n − 1)πx
δ(x − ξ)δ(t − τ ) = dn (t) sin
n=1
2L
Z L  
2 (2n − 1)πx
dn (t) = δ(x − ξ)δ(t − τ ) sin dx
L 0 2L
 
2 (2n − 1)πξ
dn (t) = sin δ(t − τ )
L 2L
By equating coefficients in the sine series, we obtain ordinary differential equation Green function problems for
the gn ’s.
 2  
00 (2n − 1)πc 2 (2n − 1)πξ
gn (t; τ ) + gn (t; τ ) = sin δ(t − τ )
2L L 2L
From the causality condition for G, we have the causality conditions for the gn ’s,
gn (t; τ ) = gn0 (t; τ ) = 0 for t < τ.
The continuity and jump conditions for the gn are
 
2 (2n − 1)πξ
+
gn (τ ; τ ) = 0, gn0 (τ + ; τ ) = sin .
L 2L
A set of homogeneous solutions of the ordinary differential equation are
    
(2n − 1)πct (2n − 1)πct
cos , sin
2L 2L
Since the continuity and jump conditions are given at the point t = τ , a handy set of solutions to use for this
problem is the fundamental set of solutions at that point:
    
(2n − 1)πc(t − τ ) 2L (2n − 1)πc(t − τ )
cos , sin
2L (2n − 1)πc 2L

1842
The solution that satisfies the causality condition and the continuity and jump conditions is,
   
4 (2n − 1)πξ (2n − 1)πc(t − τ )
gn (t; τ ) = sin sin H(t − τ ).
(2n − 1)πc 2L 2L

Substituting this into the sum yields,

∞      
4 X 1 (2n − 1)πξ (2n − 1)πc(t − τ ) (2n − 1)πx
G(x, t; ξ, τ ) = H(t − τ ) sin sin sin .
πc n=1
2n − 1 2L 2L 2L

We use trigonometric identities to write this in terms of traveling waves.

∞  
1 X 1 (2n − 1)π((x − ξ) − c(t − τ ))
G(x, t; ξ, τ ) = H(t − τ ) sin
πc n=1
2n − 1 2L
   
(2n − 1)π((x − ξ) + c(t − τ )) (2n − 1)π((x + ξ) − c(t − τ ))
+ sin − sin
2L 2L
 !
(2n − 1)π((x + ξ) + c(t − τ ))
− sin
2L

b) Now we consider the Green function with the boundary conditions,

ux (0, t) = ux (L, t) = 0.

First we find the eigenfunctions in x of the homogeneous problem. The eigenvalue problem is

X 00 = −λ2 X, X 0 (0) = X 0 (L) = 0,

1843
which has the solutions,

λ0 = 0, X0 = 1,
nπ  nπx 
λn = , Xn = cos , n = 1, 2, . . . .
L L
The series expansion of the Green function for t > τ has the form,
∞
1 X  nπx 
G(x, t; ξ, τ ) = g0 (t) + gn (t) cos .
2 n=1
L

(Note the factor of 1/2 in front of g0 (t). With this, the integral formulas for all the coefficients are the same.) We
determine the coefficients by substituting the expansion into the partial differential equation.

Gtt − c2 Gxx = δ(x − ξ)δ(t − τ )

∞  
1 00 X
00
 nπc 2  nπx 
g0 (t) + gn (t) + gn (t) cos = δ(x − ξ)δ(t − τ )
2 n=1
L L

We expand the right side of the equation in the cosine series.

∞
1 X  nπx 
δ(x − ξ)δ(t − τ ) = d0 (t) + dn (t) cos
2 n=1
L
Z L
2  nπx 
dn (t) = δ(x − ξ)δ(t − τ ) cos dx
L 0 L
 
2 nπξ
dn (t) = cos δ(t − τ )
L L
By equating coefficients in the cosine series, we obtain ordinary differential equations for the gn .
 nπc 2  
00 2 nπξ
gn (t; τ ) + gn (t; τ ) = cos δ(t − τ ), n = 0, 1, 2, . . .
L L L

1844
From the causality condition for G, we have the causality condiions for the gn ,

gn (t; τ ) = gn0 (t; τ ) = 0 for t < τ.

The continuity and jump conditions for the gn are

 
2 nπξ
+
gn (τ ; τ ) = 0, gn0 (τ + ; τ ) = cos .
L L
The homogeneous solutions of the ordinary differential equation for n = 0 and n > 0 are respectively,
    
nπct nπct
{1, t}, cos , sin .
L L
Since the continuity and jump conditions are given at the point t = τ , a handy set of solutions to use for this
problem is the fundamental set of solutions at that point:
    
nπc(t − τ ) L nπc(t − τ )
{1, t − τ }, cos , sin .
L nπc L
The solutions that satisfy the causality condition and the continuity and jump conditions are,
2
g0 (t) = (t − τ )H(t − τ ),
 L  
2 nπξ nπc(t − τ )
gn (t) = cos sin H(t − τ ).
nπc L L
Substituting this into the sum yields,
∞     !
t−τ 2 X1 nπξ nπc(t − τ )  nπx 
G(x, t; ξ, τ ) = H(t − τ ) + cos sin cos .
L πc n=1 n L L L

We can write this as the sum of traveling waves.

1845
 ∞  
t−τ 1 X 1 nπ((x − ξ) − c(t − τ ))
G(x, t; ξ, τ ) = H(t − τ ) + H(t − τ ) − sin
L 2πc n=1
n 2L
   
nπ((x − ξ) + c(t − τ )) nπ((x + ξ) − c(t − τ ))
+ sin − sin
2L 2L
 !
nπ((x + ξ) + c(t − τ ))
+ sin
2L

Solution 46.13
First we derive Green’s identity for this problem. We consider the integral of uL[v] − L[u]v on the domain
0 < x < 1, 0 < t < T .
Z T Z 1
(uL[v] − L[u]v) dx dt
0 0
Z T Z 1
u(vtt − c2 vxx − (utt − c2 uxx )v dx dt


0 0
Z T Z 1   
∂ ∂ 2


, · −c (uvx − ux v), uvt − ut v dx dt
0 0 ∂x ∂t

Now we can use the divergence theorem to write this as an integral along the boundary of the domain.
I
−c2 (uvx − ux v), uvt − ut v · n ds


∂Ω

The domain and the outward normal vectors are shown in Figure 46.5.

1846
 n=(0,1)

t=T

n=(-1,0) n=(1,0)

t=0
x=0 x=1
n=(0,-1)

Figure 46.5: Outward normal vectors of the domain.

Writing out the boundary integrals, Green’s identity for this problem is,

Z T Z 1 Z 1
2 2


u(vtt − c vxx ) − (utt − c uxx )v dx dt = − (uvt − ut v)t=0 dx
0 0 0
Z 0 Z T Z 1
2 2
+ (uvt − ut v)t=T dx − c (uvx − ux v)x=1 dt + c (uvx − ux v)x=0 dt
1 0 T

The Green function problem is

Gtt − c2 Gxx = δ(x − ξ)δ(t − τ ), 0 < x, ξ < 1, t, τ > 0,

Gx (0, t; ξ, τ ) = Gx (1, t; ξ, τ ) = 0, t > 0, G(x, t; ξ, τ ) = 0 for t < τ.

1847
If we consider G as a function of (ξ, τ ) with (x, t) as parameters, then it satisfies:

Gτ τ − c2 Gξξ = δ(x − ξ)δ(t − τ ),

Gξ (x, t; 0, τ ) = Gξ (x, t; 1, τ ) = 0, τ > 0, G(x, t; ξ, τ ) = 0 for τ > t.

Now we apply Green’s identity for u = u(ξ, τ ), (the solution of the wave equation), and v = G(x, t; ξ, τ ), (the
Green function), and integrate in the (ξ, τ ) variables. The left side of Green’s identity becomes:
Z T Z 1
u(Gτ τ − c2 Gξξ ) − (uτ τ − c2 uξξ )G dξ dτ


0 0
Z T Z 1
(u(δ(x − ξ)δ(t − τ )) − (0)G) dξ dτ
0 0
u(x, t).

Since the normal derivative of u and G vanish on the sides of the domain, the integrals along ξ = 0 and ξ = 1 in
Green’s identity vanish. If we take T > t, then G is zero for τ = T and the integral along τ = T vanishes. The
one remaining integral is
Z 1
− (u(ξ, 0)Gτ (x, t; ξ, 0) − uτ (ξ, 0)G(x, t; ξ, 0) dξ.
0

Thus Green’s identity allows us to write the solution of the inhomogeneous problem.
Z 1
u(x, t) = (uτ (ξ, 0)G(x, t; ξ, 0) − u(ξ, 0)Gτ (x, t; ξ, 0)) dξ.
0

With the specified initial conditions this becomes

Z 1
u(x, t) = (G(x, t; ξ, 0) − ξ 2 (1 − ξ)2 Gτ (x, t; ξ, 0)) dξ.
0

1848
Now we substitute in the Green function that we found in the previous exercise. The Green function and its
derivative are,
∞
X 2
G(x, t; ξ, 0) = t + cos(nπξ) sin(nπct) cos(nπx),
n=1
nπc
∞
X
Gτ (x, t; ξ, 0) = −1 − 2 cos(nπξ) cos(nπct) cos(nπx).
n=1

The integral of the first term is,

∞
!
Z 1 X 2
t+ cos(nπξ) sin(nπct) cos(nπx) dξ = t.
0 n=1
nπc

The integral of the second term is

∞
! ∞
Z 1 X 1 X 1
ξ 2 (1 − ξ)2 1 + 2 cos(nπξ) cos(nπct) cos(nπx) dξ = −3 4π4
cos(2nπx) cos(2nπct).
0 n=1
30 n=1
n

Thus the solution is

∞
1 X 1
u(x, t) = +t−3 4 4
cos(2nπx) cos(2nπct).
30 n=1
n π

For c = 1, the solution at x = 3/4, t = 7/2 is,

∞
1 7 X 1
u(3/4, 7/2) = + −3 cos(3nπ/2) cos(7nπ).
30 2 n=1
n π4
4

1849
Note that the summand is nonzero only for even terms.
∞
53 3 X 1
u(3/4, 7/2) = − cos(3nπ) cos(14nπ)
15 16pi4 n=1 n4
∞
53 3 X (−1)n
= −
15 16pi4 n=1 n4
53 3 −7π 4
= −
15 16pi4 720

12727
u(3/4, 7/2) =
3840

1850
Chapter 47

Conformal Mapping

1851
47.1 Exercises
Exercise 47.1
ζ = ξ + iη is an analytic function of z, ζ = ζ(z). We assume that ζ 0 (z) is nonzero on the domain of interest.
u(x, y) is an arbitrary smooth function of x and y. When expressed in terms of ξ and η, u(x, y) = υ(ξ, η). In
Exercise 10.13 we showed that
 −2  2
∂ 2 υ ∂ 2 υ  dζ  ∂ u ∂2u

+ 2 =  + .
∂ξ 2 ∂η dz ∂x2 ∂y 2
1. Show that if u satisfies Laplace’s equation in the z-plane,
uxx + uyy = 0,
then υ satisfies Laplace’s equation in the ζ-plane,
υξξ + υηη = 0,

2. Show that if u satisfies Helmholtz’s equation in the z-plane,

uxx + uyy = λu,
then in the ζ-plane υ satisfies
 2
 dz 
υξξ + υηη = λ   υ.
dζ

3. Show that if u satisfies Poisson’s equation in the z-plane,

uxx + uyy = f (x, y),
then υ satisfies Poisson’s equation in the ζ-plane,
 2
 dz 
υξξ + υηη =   φ(ξ, η),
dζ
where φ(ξ, η) = f (x, y).

1852
 4. Show that if in the z-plane, u satisfies the Green function problem,
uxx + uyy = δ(x − x0 )δ(y − y0 ),
then in the ζ-plane, υ satisfies the Green function problem,
υξξ + υηη = δ(ξ − ξ0 )δ(η − η0 ).

Exercise 47.2
A semi-circular rod of infinite extent is maintained at temperature T = 0 on the flat side and at T = 1 on the
curved surface:
x2 + y 2 = 1, y > 0.
Use the conformal mapping
1+z
w = ξ + iη = , z = x + iy,
1−z
to formulate the problem in terms of ξ and η. Solve the problem in terms of these variables. This problem is
solved with an eigenfunction expansion in Exercise ??. Verify that the two solutions agree.

Exercise 47.3
Consider Laplace’s equation on the domain −∞ < x < ∞, 0 < y < π, subject to the mixed boundary conditions,
u = 1 on y = 0, x > 0,
u = 0 on y = π, x > 0,
uy = 0 on y = 0, y = π, x < 0.
Because of the mixed boundary conditions, (u and uy are given on separate parts of the same boundary), this
problem cannot be solved with separation of variables. Verify that the conformal map,
ζ = cosh−1 ( ez ),

1853
with z = x + iy, ζ = ξ + iη maps the infinite interval into the semi-infinite interval, ξ > 0, 0 < η < π. Solve
Laplace’s equation with the appropriate boundary conditions in the ζ plane by inspection. Write the solution u
in terms of x and y.

1854
47.2 Hints
Hint 47.1

Hint 47.2
Show that w = (1 + z)/(1 − z) maps the semi-disc, 0 < r < 1, 0 < θ < π to the first quadrant of the w plane.
Solve the problem for v(ξ, η) by taking Fourier sine transforms in ξ and η.
To show that the solution for v(ξ, η) is equivalent to the series expression for u(r, θ), first find an analytic
function g(w) of which v(ξ, η) is the imaginary part. Change variables to z to obtain the analytic function
f (z) = g(w). Expand f (z) in a Taylor series and take the imaginary part to show the equivalence of the solutions.

Hint 47.3
To see how the boundary is mapped, consider the map,

z = log(cosh ζ).

The problem in the ζ plane is,

vξξ + vηη = 0, ξ > 0, 0 < η < π,

vξ (0, η) = 0, v(ξ, 0) = 1, v(ξ, π) = 0.

To solve this, find a plane that satisfies the boundary conditions.

1855
47.3 Solutions
Solution 47.1

 −2  2
∂ 2 υ ∂ 2 υ  dζ  ∂ u ∂2u

+ 2 =  + .
∂ξ 2 ∂η dz ∂x2 ∂y 2
1.
uxx + uyy = 0
 2
 dζ 
  (υξξ + υηη ) = 0
 dz 
υξξ + υηη = 0

2.
uxx + uyy = λu
 2
 dζ 
  (υξξ + υηη ) = λυ
 dz 
 2
 dz 
υξξ + υηη = λ   υ
dζ

3.
uxx + uyy = f (x, y)
 2
 dζ 
  (υξξ + υηη ) = φ(ξ, η)
 dz 
 2
 dz 
υξξ + υηη =   φ(ξ, η)
dζ

1856
 4. The Jacobian of the mapping is
 
 xξ y ξ 
J =   = xξ yη − xη yξ = x2ξ + yξ2 .
xη y η 
Thus the Dirac delta function on the right side gets mapped to
1
δ(ξ − ξ0 )δ(η − η0 ).
x2ξ + yξ2

Next we show that |dz/dζ|2 has the same value as the Jacobian.
 2
 dz 
  = (xξ + iyξ )(xξ − iyξ ) = x2ξ + yξ2
 dζ 

Now we transform the Green function problem.

uxx + uyy = δ(x − x0 )δ(y − y0 )
 2
 dζ  1
  (υξξ + υηη ) = δ(ξ − ξ0 )δ(η − η0 )
 dz  x2ξ + yξ2
υξξ + υηη = δ(ξ − ξ0 )δ(η − η0 )

Solution 47.2
The mapping,
1+z
w= ,
1−z
maps the unit semi-disc to the first quadrant of the complex plane.
We write the mapping in terms of r and θ.
1 + r eiθ 1 − r2 + i2r sin θ
ξ + iη = =
1 − r eiθ 1 + r2 − 2r cos θ

1857
 1 − r2
ξ=
1 + r2 − 2r cos θ
2r sin θ
η=
1 + r2 − 2r cos θ
2r
Consider a semi-circle of radius r. The image of this under the conformal mapping is a semi-circle of radius 1−r 2
1+r2 1−r 1+r
and center 1−r2 in the first quadrant of the w plane. This semi-circle intersects the ξ axis at 1+r and 1−r . As r
ranges from zero to one, these semi-circles cover the first quadrant of the w plane. (See Figure 47.1.)

1 5
4
3
2
1

-1 1 1 2 3 4 5

1+z
Figure 47.1: The conformal map, w = 1−z
.

We also note how the boundary of the semi-disc is mapped to the boundary of the first quadrant of the w
plane. The line segment θ = 0 is mapped to the real axis ξ > 1. The line segment θ = π is mapped to the real
axis 0 < ξ < 1. Finally, the semi-circle r = 1 is mapped to the positive imaginary axis.
The problem for v(ξ, η) is,

vξξ + vηη = 0, ξ > 0, η > 0,

v(ξ, 0) = 0, v(0, η) = 1.

We will solve this problem with the Fourier sine transform. We take the Fourier sine transform of the partial

1858
differential equation, first in ξ and then in η.
α
−α2 v̂(α, η) + v(0, η) + v̂(α, η) = 0, v̂(α, 0) = 0
π
α
−α2 v̂(α, η) + + v̂(α, η) = 0, v̂(α, 0) = 0
π
ˆ β) + α − β 2 v̂(α,
−α2 v̂(α, ˆ β) + β v̂(α, 0) = 0
2
π β π
ˆ β) = α
v̂(α,
π 2 β(α2 + β 2 )
Now we utilize the Fourier sine transform pair,
ω/π
Fs e−cx =
 
,
ω2+ c2
to take the inverse sine transform in α.
1 −βξ
v̂(ξ, β) = e
πβ
With the Fourier sine transform pair,
h  x i 1
Fs 2 arctan = e−cω ,
c ω
we take the inverse sine transform in β to obtain the solution.
 
2 η
v(ξ, η) = arctan
π ξ

Since v is harmonic, it is the imaginary part of an analytic function g(w). By inspection, we see that this function
is
2
g(w) = log(w).
π

1859
We change variables to z, f (z) = g(w).
 
2 1+z
f (z) = log
π 1−z
We expand f (z) in a Taylor series about z = 0,
∞
4 X zn
f (z) = ,
π n=1 n
oddn

and write the result in terms of r and θ, z = r eiθ .

∞
4 X rn eiθ
f (z) =
π n=1 n
oddn

u(r, θ) is the imaginary part of f (z).

∞
4X1 n
u(r, θ) = r sin(nθ)
π n=1 n
oddn

This demonstrates that the solutions obtained with conformal mapping and with an eigenfunction expansion in
Exercise ?? agree.

Solution 47.3
Instead of working with the conformal map from the z plane to the ζ plane,

ζ = cosh−1 ( ez ),
it will be more convenient to work with the inverse map,
z = log(cosh ζ),

1860
which maps the semi-infinite strip to the infinite one. We determine how the boundary of the domain is mapped
so that we know the appropriate boundary conditions for the semi-infinite strip domain.

A {ζ : ξ > 0, η = 0} 7→ {log(cosh(ξ)) : ξ > 0} = {z : x > 0, y = 0}

B {ζ : ξ > 0, η = π} 7→ {log(− cosh(ξ)) : ξ > 0} = {z : x > 0, y = π}
C {ζ : ξ = 0, 0 < η < π/2} 7→ {log(cos(η)) : 0 < η < π/2} = {z : x < 0, y = 0}
D {ζ : ξ = 0, π/2 < η < π} 7→ {log(cos(η)) : π/2 < η < π} = {z : x < 0, y = π}

From the mapping of the boundary, we see that the solution v(ξ, η) = u(x, y), is 1 on the bottom of the semi-infinite
strip, 0 on the top. The normal derivative of v vanishes on the vertical boundary. See Figure 47.2.

y
η
B D B
D z=log(cosh(ζ ))

C ξ x
A C A

y
η
v=0 uy=0 u=0

vξ=0
ξ x
v=1 uy=0 u=1

Figure 47.2: The mapping of the boundary conditions.

1861
 In the ζ plane, the problem is,

vξξ + vηη = 0, ξ > 0, 0 < η < π,

vξ (0, η) = 0, v(ξ, 0) = 1, v(ξ, π) = 0.

By inspection, we see that the solution of this problem is,

η
v(ξ, η) = 1 − .
π

The solution in the z plane is

1
= cosh−1 ( ez ) ,


u(x, y) = 1 −
π

where z = x + iy. We will find the imaginary part of cosh−1 ( ez ) in order to write this explicitly in terms of x and
y. Recall that we can write the cosh−1 in terms of the logarithm.
 √ 
cosh−1 (w) = log w + w2 − 1
 √ 
cosh−1 ( ez ) = log ez + e2z − 1
 
z
√ 
= log e 1 + 1 − e −2z
 √ 
= z + log 1 + 1 − e−2z

Now we need to find the imaginary part. We’ll work from the inside out. First recall,
r  y  p   y 
p p  i −1
x + iy = x2 + y 2 exp i tan−1 4
= x2 + y 2 exp tan ,
x 2 x

1862
so that we can write the innermost factor as,
√ p
1 − e−2z = 1 − e−2x cos(2y) + i e−2x sin(2y)
e−2x sin(2y)
  
p
4 −2x 2 −2x 2
i −1
= (1 − e cos(2y)) + ( e sin(2y)) exp tan
2 1 − e−2x cos(2y)
  
p
4 −2x −4x
i −1 sin(2y)
= 1 − 2e cos(2y) + e exp tan
2 e2x − cos(2y)

We substitute this into the logarithm.

√
   
  p i −1 sin(2y)
log 1 + 1 − e−2z = log 1 + 1 − 2 e−2x cos(2y) + e−4x exp
4
tan
2 e2x − cos(2y)

Now we can write η.

  √ 
η = = z + log 1 + 1 − e−2z
 p    
4
1 − 2 e−2x cos(2y) + e−4x sin 12 tan−1 e2xsin(2y)
−cos(2y)
η = y + tan−1     
sin(2y)
p 1
−2x −4x −1
1 + 1 − 2e
4
cos(2y) + e cos 2 tan e2x −cos(2y)

Finally we have the solution, u(x, y).

    
sin(2y)
p4
1 − 2 e−2x cos(2y) + e−4x sin 1 tan−1
y 1 2 e2x −cos(2y)
u(x, y) = 1 − − tan−1     
π π 1 + 4 1 − 2 e−2x cos(2y) + e−4x cos 12 tan−1 e2xsin(2y)
p
−cos(2y)

1863
Chapter 48

Non-Cartesian Coordinates

48.1 Spherical Coordinates

Writing rectangular coordinates in terms of spherical coordinates,

x = r cos θ sin φ
y = r sin θ sin φ
z = r cos φ.

1864
The Jacobian is
 
cos θ sin φ −r sin θ sin φ r cos θ cos φ
 
 sin θ sin φ r cos θ sin φ r sin θ cos φ 
 
 cos φ 0 −r sin φ 
 
cos θ sin φ − sin θ cos θ cos φ
 
= r2 sin φ  sin θ sin φ cos θ sin θ cos φ 
 cos φ 0 − sin φ 
 2 
= r sin φ(− cos2 θ sin2 φ − sin2 θ cos2 φ − cos2 θ cos2 φ − sin2 θ sin2 φ)
= r2 sin φ(sin2 φ + cos2 φ)
= r2 sin φ.

Thus we have that

ZZZ ZZZ
f (x, y, z) dx dy dz = f (r, θ, φ)r2 sin φ dr dθ dφ.
V V

48.2 Laplace’s Equation in a Disk

Consider Laplace’s equation in polar coordinates

1 ∂2u
 
1 ∂ ∂u
r + 2 2 = 0, 0≤r≤1
r ∂r ∂r r ∂θ

subject to the the boundary conditions

1. u(1, θ) = f (θ)

2. ur (1, θ) = g(θ).

1865
 We separate variables with u(r, θ) = R(r)T (θ).

1 0 1
(R T + rR00 T ) + 2 RT 00 = 0
r r
00 0
R R T 00
r2 +r =− =λ
R R T
Thus we have the two ordinary differential equations

T 00 + λT = 0, T (0) = T (2π), T 0 (0) = T 0 (2π)

r2 R00 + rR0 − λR = 0, R(0) < ∞.

The eigenvalues and eigenfunctions for the equation in T are

1
λ0 = 0, T0 =
2
λn = n 2 , Tn(1) = cos(nθ), Tn(2) = sin(nθ)

(I chose T0 = 1/2 so that all the eigenfunctions have the same norm.)

For λ = 0 the general solution for R is

R = c1 + c2 log r.

Requiring that the solution be bounded gives us

R0 = 1.

For λ = n2 > 0 the general solution for R is

R = c1 rn + c2 r−n .

1866
Requiring that the solution be bounded gives us
Rn = rn .
Thus the general solution for u is
∞
a0 X n
u(r, θ) = + r [an cos(nθ) + bn sin(nθ)] .
2 n=1

For the boundary condition u(1, θ) = f (θ) we have the equation

∞
a0 X
f (θ) = + [an cos(nθ) + bn sin(nθ)] .
2 n=1

If f (θ) has a Fourier series then the coefficients are

1 2π
Z
a0 = f (θ) dθ
π 0
1 2π
Z
an = f (θ) cos(nθ) dθ
π 0
1 2π
Z
bn = f (θ) sin(nθ) dθ.
π 0

For the boundary condition ur (1, θ) = g(θ) we have the equation

∞
X
g(θ) = n [an cos(nθ) + bn sin(nθ)] .
n=1

g(θ) has a series of this form only if

Z 2π
g(θ) dθ = 0.
0

1867
The coefficients are
Z 2π
1
an = g(θ) cos(nθ) dθ
nπ 0
Z 2π
1
bn = g(θ) sin(nθ) dθ.
nπ 0

48.3 Laplace’s Equation in an Annulus

Consider the problem

1 ∂2u
 
2 1 ∂ ∂u
∇ u= r + 2 2 = 0, 0 ≤ r < a, −π < θ ≤ π,
r ∂r ∂r r ∂θ

with the boundary condition

u(a, θ) = θ2 .

So far this problem only has one boundary condition. By requiring that the solution be finite, we get the
boundary condition

|u(0, θ)| < ∞.

By specifying that the solution be C 1 , (continuous and continuous first derivative) we obtain

∂u ∂u
u(r, −π) = u(r, π) and (r, −π) = (r, π).
∂θ ∂θ
We will use the method of separation of variables. We seek solutions of the form

u(r, θ) = R(r)Θ(θ).

1868
Substituting into the partial differential equation,
∂ 2 u 1 ∂u 1 ∂2u
+ + =0
∂r2 r ∂r r2 ∂θ2
1 1
R00 Θ + R0 Θ = − 2 RΘ00
r r
2 00 0
r R rR Θ00
+ =− =λ
R R Θ
Now we have the boundary value problem for Θ,
Θ00 (θ) + λΘ(θ) = 0, −π < θ ≤ π,

subject to

Θ(−π) = Θ(π) and Θ0 (−π) = Θ0 (π)

We consider the following three cases for the eigenvalue, λ,
√ √
λ < 0. No linear combination of the solutions, Θ = exp( −λθ), exp(− −λθ), can satisfy the boundary
conditions. Thus there are no negative eigenvalues.
λ = 0. The general solution solution is Θ = a + bθ. By applying the boundary conditions, we get Θ = a. Thus
we have the eigenvalue and eigenfunction,
λ0 = 0, A0 = 1.
√ √
λ > 0. The general solution is Θ = a cos( λθ) + b sin( λθ). Applying the boundary conditions yields the
eigenvalues
λn = n 2 , n = 1, 2, 3, . . .
with the associated eigenfunctions
An = cos(nθ) and Bn = sin(nθ).

1869
 The equation for R is
r2 R00 + rR0 − λn R = 0.
In the case λ0 = 0, this becomes
1
R00 = − R0
r
0 a
R =
r
R = a log r + b

Requiring that the solution be bounded at r = 0 yields (to within a constant multiple)

R0 = 1.
For λn = n2 , n ≥ 1, we have
r2 R00 + rR0 − n2 R = 0

Recognizing that this is an Euler equation and making the substitution R = rα ,

α(α − 1) + α − n2 = 0
α = ±n
R = arn + br−n .

requiring that the solution be bounded at r = 0 we obtain (to within a constant multiple)

Rn = rn
The general solution to the partial differential equation is a linear combination of the eigenfunctions
∞
X
u(r, θ) = c0 + [cn rn cos nθ + dn rn sin nθ] .
n=1

1870
We determine the coefficients of the expansion with the boundary condition
∞
X
2
u(a, θ) = θ = c0 + [cn an cos nθ + dn an sin nθ] .
n=1

We note that the eigenfunctions 1, cos nθ, and sin nθ are orthogonal on −π ≤ θ ≤ π. Integrating the boundary
condition from −π to π yields
Z π Z π
2
θ dθ = c0 dθ
−π −π
π2
c0 = .
3
Multiplying the boundary condition by cos mθ and integrating gives
Z π Z π
2 m
θ cos mθ dθ = cm a cos2 mθ dθ
−π −π
m
(−1) 8π
cm = .
m2 am
We multiply by sin mθ and integrate to get
Z π Z π
2 m
θ sin mθ dθ = dm a sin2 mθ dθ
−π −π
dm = 0

Thus the solution is

∞
π 2 X (−1)n 8π n
u(r, θ) = + r cos nθ.
3 n=1
n 2 an

1871
 Part VI

Calculus of Variations

1872
Chapter 49

Calculus of Variations

1873
49.1 Exercises
Exercise 49.1 Rα
Discuss the problem of minimizing 0 ((y 0 )4 − 6(y 0 )2 ) dx, y(0) = 0, y(α) = β. Consider both C 1 [0, α] and Cp1 [0, α],
and comment (with reasons) on whether your answers are weak or strong minima.

Exercise 49.2
Consider
Rx
1. x01 (a(y 0 )2 + byy 0 + cy 2 ) dx, y(x0 ) = y0 , y(x1 ) = y1 , a 6= 0,
Rx
2. x01 (y 0 )3 dx, y(x0 ) = y0 , y(x1 ) = y1 .

Can these functionals have broken extremals, and if so, find them.

Exercise 49.3
Discuss finding a weak extremum for the following:
R1
1. 0
((y 00 )2 − 2xy) dx, y(0) = y 0 (0) = 0, y(1) = 1
120
R1 1 0 2
+ yy 0 + y 0 + y dx


2. 0 2
(y )
Rb
3. a
(y 2 + 2xyy 0 ) dx, y(a) = A, y(b) = B
R1
4. 0
(xy + y 2 − 2y 2 y 0 ) dx, y(0) = 1, y(1) = 2

Exercise 49.4
Find the natural boundary conditions associated with the following functionals:
RR
1. D
F (x, y, u, ux , uy ) dx dy

1874
 RR
R
2. D
p(x, y)(u2x + u2y ) − q(x, y)u2 dx dy + Γ σ(x, y)u2 ds

Here D represents a closed boundary domain with boundary Γ, and ds is the arc-length differential. p and q are
known in D, and σ is known on Γ.

Exercise 49.5
The equations for water waves with free surface y = h(x, t) and bottom y = 0 are

φxx + φyy = 0 0 < y < h(x, t),

1 1
φt + φ2x + φ2y + gy = 0 on y = h(x, t),
2 2
ht + φx hx − φy = 0, on y = h(x, t),
φy = 0 on y = 0,

where the fluid motion is described by φ(x, y, t) and g is the acceleration of gravity. Show that all these equations
may be obtained by varying the functions φ(x, y, t) and h(x, t) in the variational principle
ZZ Z h(x,t)   !
1 2 1 2
δ φt + φx + φy + gy dy dx dt = 0,
R 0 2 2

where R is an arbitrary region in the (x, t) plane.

Exercise 49.6 Rb
Extremize the functional a F (x, y, y 0 ) dx, y(a) = A, y(b) = B given that the admissible curves can not penetrate
Rthe interior of a given region R in the (x, y) plane. Apply your results to find the curves which extremize
10 0 3
0
(y ) dx, y(0) = 0, y(10) = 0 given that the admissible curves can not penetrate the interior of the circle
(x − 5)2 + y 2 = 9.

1875
Exercise 49.7 R√ p
Consider the functional y ds where ds is the arc-length differential (ds = (dx)2 + (dy)2 ). Find the curve
or curves from a given vertical line to a given fixed point B = (x1 , y1 ) which minimize this functional. Consider
both the classes C 1 and Cp1 .

Exercise 49.8
A perfectly flexible uniform rope of length L hangs in equilibrium with one end fixed at (x1 , y1 ) so that it passes
over a frictionless pin at (x2 , y2 ). What is the position of the free end of the rope?

Exercise 49.9
The drag on a supersonic airfoil of chord c and shape y = y(x) is proportional to
Z c  2
dy
D= dx.
0 dx
Find the shape for minimum drag if the moment of inertia of the contour with respect to the x-axis is specified;
that is, find the shape for minimum drag if
Z c
y 2 dx = A, y(0) = y(c) = 0, (c, A given).
0

Exercise 49.10
The deflection y of a beam executing free (small) vibrations of frequency ω satisfies the differential equation
d2 d2 y
 
2
EI 2 − ρω 2 y = 0,
dx dx
where EI is the flexural rigidity and ρ is the linear mass density. Show that the deflection modes are extremals
of the problem
RL !
00 2
EI(y ) dx
δω 2 ≡ δ 0
RL = 0, (L = length of beam)
0
ρy 2 dx

1876
when appropriate homogeneous end conditions are prescribed. Show that stationary values of the ratio are the
squares of the natural frequencies.

Exercise 49.11
A boatman wishes to steer his boat so as to minimize the transit time required to cross a river of width l. The
path of the boat is given parametrically by
x = X(t), y = Y (t),
for 0 ≤ t ≤ T . The river has no cross currents, so the current velocity is directed downstream in the y-direction.
v0 is the constant boat speed relative to the surrounding water, and w = w(x, y, t) denotes the downstream river
current at point (x, y) at time t. Then,
Ẋ(t) = v0 cos α(t), Ẏ (t) = v0 sin α(t) + w,
where α(t) is the steering angle of the boat at time t. Find the steering control function α(t) and the final time
T that will transfer the boat from the initial state (X(0), Y (0)) = (0, 0) to the final state at X(t) = l in such a
way as to minimize T .

Exercise 49.12
Two particles of equal mass m are connected by an inextensible string which passes through a hole in a smooth
p
horizontal table. The first particle is on the table moving with angular velocity ω = g/α in a circular path, of
radius α, around the hole. The second particle is suspended vertically and is in equilibrium. At time t = 0, the
suspended mass is pulled downward a short distance and released while the first mass continues to rotate.
1. If x represents the distance of the second mass below its equilibrium at time t and θ represents the angular
position of the first particle at time t, show that the Lagrangian is given by
 
2 1 2 2
L = m ẋ + (α − x) θ̇ + gx
2
and obtain the equations of motion.

1877
 2. In the case where the displacement of the suspended mass from equilibrium is small, show that the suspended
mass performs small vertical oscillations and find the period of these oscillations.

Exercise 49.13
A rocket is propelled vertically upward so as to reach a prescribed height h in minimum time while using a given
fixed quantity of fuel. The vertical distance x(t) above the surface satisfies,

mẍ = −mg + mU (t), x(0) = 0, ˙

(x)(0) = 0,

where U (t) is the acceleration provided by engine thrust. We impose the terminal constraint x(T ) = h, and we
wish to find the particular thrust function U (t) which will minimize T assuming that the total thrust of the rocket
engine over the entire thrust time is limited by the condition,
Z T
U 2 (t) dt = k 2 .
0

Here k is a given positive constant which measures the total amount of fuel available.

Exercise 49.14
A space vehicle moves along a straight path in free space. x(t) is the distance to its docking pad, and a, b are its
position and speed at time t = 0. The equation of motion is

ẍ = M sin V, x(0) = a, ẋ(0) = b,

where the control function V (t) is related to the rocket acceleration U (t) by U = M sin V , M = const. We wish
to dock the vehicle in minimum time; that is, we seek a thrust function U (t) which will minimize the final time T
while bringing the vehicle to rest at the origin with x(T ) = 0, ẋ(T ) = 0. Find U (t), and in the (x, ẋ)-plane plot
the corresponding trajectory which transfers the state of the system from (a, b) to (0, 0). Account for all values
of a and b.

1878
Exercise 49.15 Rm√ p
Find a minimum for the functional I(y) = 0 y + h 1 + (y 0 )2 dx in which h > 0, y(0) = 0, y(m) = M > −h.
Discuss the nature of the minimum, (i.e., weak, strong, . . . ).

Exercise 49.16 R p
Show that for the functional n(x, y) 1 + (y 0 )2 dx, where n(x, y) ≥ 0 in some domain D, the Weierstrass E
function E(x, y, q, y 0 ) is non-negative for arbitrary finite p and y 0 at any point of D. What is the implication of
this for Fermat’s Principle?

Exercise 49.17
2
Consider the integral 1+y
R
(y 0 )2
dx between fixed limits. Find the extremals, (hyperbolic sines), and discuss the
Jacobi, Legendre, and Weierstrass conditions and their implications regarding weak and strong extrema. Also
consider the value of the integral on any extremal compared with its value on the illustrated strong variation.
Comment!
Pi Qi are vertical segments, and the lines Qi Pi+1 are tangent to the extremal at Pi+1 .

Exercise 49.18
Rx
Consider I = x01 y 0 (1 + x2 y 0 ) dx, y(x0 ) = y0 , y(x1 ) = y1 . Can you find continuous curves which will minimize I if

(i) x0 = −1, y0 = 1, x1 = 2, y1 = 4,
(ii) x0 = 1, y0 = 3, x1 = 2, y1 = 5,
(iii) x0 = −1, y0 = 1, x1 = 2, y1 = 1.

Exercise 49.19
Starting from
ZZ Z
(Qx − Py ) dx dy = (P dx + Q dy)
D Γ

1879
prove that
ZZ ZZ Z
(a) φψxx dx dy = ψφxx dx dy + (φψx − ψφx ) dy,
ZZ D ZZ D ZΓ
(b) φψyy dx dy = ψφyy dx dy − (φψy − ψφy ) dx,
ZZ D ZZ D ΓZ Z
1 1
(c) φψxy dx dy = ψφxy dx dy − (φψx − ψφx ) dx + (φψy − ψφy ) dy.
D D 2 Γ 2 Γ

Then, consider
Z t1 ZZ
−(uxx + uyy )2 + 2(1 − µ)(uxx uyy − u2xy ) dx dy dt.


I(u) =
t0 D

Show that
Z t1 ZZ Z t1 Z  
4 ∂δu
δI = (−∇ u)δu dx dy dt + P (u)δu + M (u) ds dt,
t0 D t0 Γ ∂n
where P and M are the expressions we derived in class for the problem of the vibrating plate.

Exercise 49.20
For the following functionals use the Rayleigh-Ritz method to find an approximate solution of the problem of
minimizing the functionals and compare your answers with the exact solutions.
•
Z 1
(y 0 )2 − y 2 − 2xy dx,


y(0) = 0 = y(1).
0

For this problem take an approximate solution of the form

y = x(1 − x) (a0 + a1 x + · · · + an xn ) ,
and carry out the solutions for n = 0 and n = 1.

1880
 •
Z 2
(y 0 )2 + y 2 + 2xy dx,


y(0) = 0 = y(2).
0

•
2
x2 − 1 2
Z  
0 2 2
x(y ) − y − 2x y dx, y(1) = 0 = y(2)
1 x

Exercise 49.21
Let K(x) belong to L1 (−∞, ∞) and define the operator T on L2 (−∞, ∞) by
Z ∞
T f (x) = K(x − y)f (y) dy.
−∞

1. Show that the spectrum of T consists of the range of the Fourier transform K̂ of K, (that is, the set of all
values K̂(y) with −∞ < y < ∞), plus 0 if this is not already in the range. (Note: From the assumption on
K it follows that K̂ is continuous and approaches zero at ±∞.)
2. For λ in the spectrum of T , show that λ is an eigenvalue if and only if K̂ takes on the value λ on at least
some interval of positive length and that every other λ in the spectrum belongs to the continuous spectrum.
3. Find an explicit representation for (T − λI)−1 f for λ not in the spectrum, and verify directly that this result
agrees with that givenby the Neumann series if λ is large enough.

Exercise 49.22
Let U be the space of twice continuously differentiable functions f on [−1, 1] satisfying f (−1) = f (1) = 0, and
d2
W = C[−1, 1]. Let L : U 7→ W be the operator dx 2 . Call λ in the spectrum of L if the following does not occur:

There is a bounded linear transformation T : W 7→ U such that (L−λI)T f = f for all f ∈ W and T (L−λI)f = f
for all f ∈ U . Determine the spectrum of L.

1881
Exercise 49.23
Solve the integral equations
Z 1
x2 y − y 2 φ(y) dy


1. φ(x) = x + λ
0
Z x
2. φ(x) = x + λ K(x, y)φ(y) dy
0

where
(
sin(xy) for x ≥ 1 and y ≤ 1,
K(x, y) =
0 otherwise

In both cases state for which values of λ the solution obtained is valid.

Exercise 49.24
1. Suppose that K = L1 L2 , where L1 L2 − L2 L1 = I. Show that if x is an eigenvector of K corresponding
to the eigenvalue λ, then L1 x is an eigenvector of K corresponding to the eigenvalue λ − 1, and L2 x is an
eigenvector corresponding to the eigenvalue λ + 1.
d t 2 2
2. Find the eigenvalues and eigenfunctions of the operator K ≡ − dt 2 + 4 in the space of functions u ∈
2
L2 (−∞, ∞). (Hint: L1 = 2t + dt
d
, L2 = 2t − dt
d
. e−t /4 is the eigenfunction corresponding to the eigenvalue
1/2.)

Exercise 49.25
Prove that if the value of λ = λ1 is in the residual spectrum of T , then λ1 is in the discrete spectrum of T ∗ .

Exercise 49.26
Solve

1882
 1.
Z 1
00
u (t) + sin(k(s − t))u(s) ds = f (t), u(0) = u0 (0) = 0.
0

2.
Z π
u(x) = λ K(x, s)u(s) ds
0

where

 ∞
 sin x+s 
1 2

X sin nx sin ns
K(x, s) = log  =

x−s 
2  sin 2 n=1
n

3.
2π
1 − h2
Z
1
φ(s) = λ φ(t) dt, |h| < 1
0 2π 1 − 2h cos(s − t) + h2

4.
Z π
φ(x) = λ cosn (x − ξ)φ(ξ) dξ
−π

Exercise 49.27
Let K(x, s) = 2π 2 − 6π|x − s| + 3(x − s)2 .
1. Find the eigenvalues and eigenfunctions of
Z 2π
φ(x) = λ K(x, s)φ(s) ds.
0

1883
 (Hint: Try to find an expansion of the form
∞
X
K(x, s) = cn ein(x−s) .)
n=−∞

2. Do the eigenfunctions form a complete set? If not, show that a complete set may be obtained by adding a
suitable set of solutions of
Z 2π
K(x, s)φ(s) ds = 0.
0

3. Find the resolvent kernel Γ(x, s, λ).

Exercise 49.28
Let K(x, s) be a bounded self-adjoint kernel on the finite interval (a, b), and let T be the integral operator on
L2 (a, b) with kernel K(x, s). For a polynomial p(t) = a0 + a1 t + · · · + an tn we define the operator p(T ) =
a0 I + a1 T + · · · + an T n . Prove that the eigenvalues of p(T ) are exactly the numbers p(λ) with λ an eigenvalue of
T.

Exercise 49.29
Show that if f (x) is continuous, the solution of
Z ∞
φ(x) = f (x) + λ cos(2xs)φ(s) ds
0

is
R∞
f (x) + λ 0
f (s) cos(2xs) ds
φ(x) = .
1 − πλ2 /4

1884
Exercise 49.30
Consider

Lu = 0 in D, u = f on C,

where

Lu ≡ uxx + uyy + aux + buy + cu.

Here a, b and c are continuous functions of (x, y) on D + C. Show that the adjoint L∗ is given by

L∗ v = vxx + vyy − avx − bvy + (c − ax − by )v

and that
Z Z
∗
(vLu − uL v) = H(u, v), (49.1)
D C

where

H(u, v) ≡ (vux − uvx + auv) dy − (vuy − uvy + buv) dx

 
∂u ∂v ∂x ∂y
= v −u + auv + buv ds.
∂n ∂n ∂n ∂n

Take v in (49.1) to be the harmonic Green function G given by

!
1 1
G(x, y; ξ, η) = log p + ··· ,
2π (x − ξ)2 + (y − η)2

and show formally, (use Delta functions), that (49.1) becomes

Z Z
∗
−u(ξ, η) − u(L − ∆)G dx dy = H(u, G) (49.2)
D C

1885
where u satisfies Lu = 0, (∆G = δ in D, G = 0 on C). Show that (49.2) can be put into the forms
Z


u+ (c − ax − by )G − aGx − bGy u dx dy = U (49.3)
D

and
Z
u+ (aux + buy + cu)G dx dy = U, (49.4)
D

where U is the known harmonic function in D with assumes the boundary values prescribed for u. Finally,
rigorously show that the integrodifferential equation (49.4) can be solved by successive approximations when the
domain D is small enough.

Exercise 49.31
Find the eigenvalues and eigenfunctions of the following kernels on the interval [0, 1].
1.

K(x, s) = min(x, s)

2.

K(x, s) = emin(x,s)

(Hint: φ00 + φ0 + λ ex φ = 0 can be solved in terms of Bessel functions.)

Exercise 49.32
Use Hilbert transforms to evaluate
Z ∞
sin(kx) sin(lx)
1. − dx
−∞ x2 − z 2

1886
 Z ∞
cos(px) − cos(qx)
2. − dx
−∞ x2
Z ∞
−(x2 − ab) sin x + (a + b)x cos x
3. − dx
−∞ x(x2 + a2 )(x2 + b2 )

Exercise 49.33
Show that
Z ∞
(1 − t2 )1/2 log(1 + t) 
2 1/2 π
 
− dt = π x log 2 − 1 + (1 − x ) − arcsin(x) .
−∞ t−x 2

Exercise 49.34
Let C be a simple closed contour. Let g(t) be a given function and consider
Z
1 f (t) dt
− = g(t0 ) (49.5)
iπ C t − t0
Note that the left side can be written as F + (t0 ) + F − (t0 ). Define a function W (z) such that W (z) = F (z) for z
inside C and W (z) = −F (z) for z outside C. Proceeding in this way, show that the solution of (49.5) is given by
Z
1 g(t) dt
f (t0 ) = − .
iπ C t − t0

Exercise 49.35
If C is an arc with endpoints α and β, evaluate
Z
1 1
(i) − dτ, where 0 < γ < 1
iπ C (τ − β) (τ − α)γ (τ − ζ)
1−γ
γ
τn
Z 
1 τ −β
(ii) − dτ, where 0 < γ < 1, integer n ≥ 0.
iπ C τ − α τ −ζ

1887
Exercise 49.36
Solve
Z 1
φ(y)
− dy = f (x).
−1 y2− x2

Exercise 49.37
Solve

1 1 f (t)
Z
− dt = λf (x), where − 1 < λ < 1.
iπ 0 t − x

Are there any solutions for λ > 1? (The operator on the left is self-adjoint. Its spectrum is −1 ≤ λ ≤ 1.)

Exercise 49.38
Show that the general solution of

tan(x) 1 f (t)
Z
− dt = f (x)
π 0 t−x

is
k sin(x)
f (x) = .
(1 − x)1−x/π xx/π

Exercise 49.39
Show that the general solution of
Z
0 f (t)
f (x) + λ − dt = 1
C t−x

1888
is given by
1
+ k e−iπλx ,
f (x) =
iπλ
(k is a constant). Here C is a simple closed contour, λ a constant and f (x) a differentiable function on C.
Generalize the result to the case of an arbitrary function g(x) on the right side, where g(x) is analytic inside C.

Exercise 49.40
Show that the solution of
Z  
1
− + P (t − x) f (t) dt = g(x)
C t−x
is given by
Z Z
1 g(τ ) 1
f (t) = − 2 − dτ − 2 g(τ )P (τ − t) dτ.
π C τ −t π C
Here C is a simple closed curve, and P (t) is a given entire function of t.

Exercise 49.41
Solve
Z 1 Z 3
f (t) f (t)
− dt + − dt = x
0 t−x 2 t−x

where this equation is to hold for x in either (0, 1) or (2, 3).

Exercise 49.42
Solve
Z x Z 1
f (t) f (t)
√ dt + A √ dt = 1
0 x−t x t−x

1889
where A is a real positive constant. Outline briefly the appropriate method of A is a function of x.

1890
49.2 Hints
Hint 49.1

Hint 49.2

Hint 49.3

Hint 49.4

Hint 49.5

Hint 49.6

Hint 49.7

Hint 49.8

1891
Hint 49.9

Hint 49.10

Hint 49.11

Hint 49.12

Hint 49.13

Hint 49.14

Hint 49.15

Hint 49.16

1892
Hint 49.17

Hint 49.18

Hint 49.19

Hint 49.20

Hint 49.21

Hint 49.22

Hint 49.23

Hint 49.24

1893
Hint 49.25

Hint 49.26

Hint 49.27

Hint 49.28

Hint 49.29

Hint 49.30

Hint 49.31

Hint 49.32

1894
Hint 49.33

Hint 49.34

Hint 49.35

Hint 49.36

Hint 49.37

Hint 49.38

Hint 49.39

Hint 49.40

1895
Hint 49.41

Hint 49.42

1896
49.3 Solutions
Solution 49.1
C 1 [0, α] Extremals
Admissible Extremal. First we consider continuously differentiable extremals. Because the Lagrangian
is a function of y 0 alone, we know that the extremals are straight lines. Thus the admissible extremal is
β
ŷ = x.
α
Legendre Condition.
F̂y0 y0 = 12(ŷ 0 )2 − 12
 2 !
β
= 12 −1
α

< 0 for |β/α| < 1

= 0 for |β/α| = 1

> 0 for |β/α| > 1


Thus we see that αβ x may be a minimum for |β/α| ≥ 1 and may be a maximum for |β/α| ≤ 1.
Jacobi Condition. Jacobi’s accessory equation for this problem is
(F̂,y0 y0 h0 )0 = 0
 2 ! !0
β
12 − 1 h0 = 0
α

h00 = 0
The problem h00 = 0, h(0) = 0, h(c) = 0 has only the trivial solution for c > 0. Thus we see that there are
no conjugate points and the admissible extremal satisfies the strengthened Legendre condition.

1897
 A Weak Minimum. For |β/α| > 1 the admissible extremal αβ x is a solution
of the Euler equation, and satisfies the strengthened Jacobi and Legendre
conditions. Thus it is a weak minima. (For |β/α| < 1 it is a weak maxima for
the same reasons.)

Weierstrass Excess Function. The Weierstrass excess function is

E(x, ŷ, ŷ 0 , w) = F (w) − F (ŷ 0 ) − (w − ŷ 0 )F,y0 (ŷ 0 )
= w4 − 6w2 − (ŷ 0 )4 + 6(ŷ 0 )2 − (w − ŷ 0 )(4(ŷ 0 )3 − 12ŷ 0 )
 4  2  3
4 2 β β β β β
= w − 6w − +6 − (w − )(4 − 12 )
α α α α α
 2 !  4  2
β β β β
= w4 − 6w2 − w 4 −3 +3 −6
α α α α

We can find the stationary points of the excess function by examining its derivative. (Let λ = β/α.)
E 0 (w) = 4w3 − 12w + 4λ (λ)2 − 3 = 0

1 √  1 √ 
w1 = λ, w2 =−λ − 4 − λ2 w3 = −λ + 4 − λ2
2 2
The excess function evaluated at these points is
E(w1 ) = 0,
3 4 √ 
E(w2 ) = 3λ − 6λ2 − 6 − 3λ(4 − λ2 )3/2 ,
2
3 4 √ 
E(w3 ) = 3λ − 6λ2 − 6 + 3λ(4 − λ2 )3/2 .
2
√ √
E(w2 ) is negative for −1 < λ < 3 and E(w3 ) is negative for √ − 3 < λ < 1. This implies that the weak
minimum ŷ = βx/α is not a strong local minimum for |λ| < 3|. Since E(w1 ) = 0, we√cannot use the
Weierstrass excess function to determine if ŷ = βx/α is a strong local minima for |β/α| > 3.

1898
Cp1 [0, α] Extremals

Erdmann’s Corner Conditions. Erdmann’s corner conditions require that

F̂,y0 = 4(ŷ 0 )3 − 12ŷ 0

and

F̂ − ŷ 0 F̂,y0 = (ŷ 0 )4 − 6(ŷ 0 )2 − ŷ 0 (4(ŷ 0 )3 − 12ŷ 0 )

are continuous at corners. Thus the quantities

(ŷ 0 )3 − 3ŷ 0 and (ŷ 0 )4 − 2(ŷ 0 )2

0 0
are continuous. Denoting p = ŷ− and q = ŷ+ , the first condition has the solutions

1 √ p 
p = q, p= −q ± 3 4 − q 2 .
2

The second condition has the solutions,

p
p = ±q, p=± 2 − q2

Combining these, we have

√ √ √ √
p = q, p= 3, q = − 3, p = − 3, q = 3.

0
√ 0
√
Thus we see that there can be a corner only when ŷ− = ± 3 and ŷ+ = ∓ 3.

1899
 √
Case 1,√ β = ± 3α. Notice the the Lagrangian is minimized point-wise if
y 0 = ± 3. For this case the unique, strong global minimum is
√
ŷ = 3 sign (β)x.
√
Case 2, |β| < 3|α|. For this case there are an infinite √ number of strong
√
0 0
minima. Any piecewise linear curve satisfying y− (x) = ± 3 and y+ (x) = ± 3
and y(0) = 0, y(α)
√ = β is a strong minima.
Case 3, |β| > 3|α|. First note that the extremal cannot have corners. Thus
the unique extremal is ŷ = αβ x. We know that this extremal is a weak local
minima.

Solution 49.2
1.
Z x1
(a(y 0 )2 + byy 0 + cy 2 ) dx, y(x0 ) = y0 , y(x1 ) = y1 , a 6= 0
x0

Erdmann’s First Corner Condition. F̂y0 = 2aŷ 0 + bŷ must be continuous at a corner. This implies that
ŷ must be continuous, i.e., there are no corners.

The functional cannot have broken extremals.

2.
Z x1
(y 0 )3 dx, y(x0 ) = y0 , y(x1 ) = y1
x0

Erdmann’s First Corner Condition. F̂y0 = 3(y 0 )2 must be continuous at a corner. This implies that
0 0
ŷ− = ŷ+ .

1900
 Erdmann’s Second Corner Condition. F̂ − ŷ 0 F̂y0 = (ŷ 0 )3 − ŷ 0 3(ŷ 0 )2 = −2(ŷ 0 )3 must be continuous at a
corner. This implies that ŷ is continuous at a corner, i.e. there are no corners.

The functional cannot have broken extremals.

Solution 49.3
1.
Z 1
1
(y 00 )2 − 2xy dx, y(0) = y 0 (0) = 0,


y(1) =
0 120

Euler’s Differential Equation. We will consider C 4 extremals which satisfy Euler’s DE,

(F̂,y00 )00 − (F̂,y0 )0 + F̂,y = 0.

For the given Lagrangian, this is,

(2ŷ 00 )00 − 2x = 0.

Natural Boundary Condition. The first variation of the performance index is

Z 1
δJ = (F̂,y δy + F̂,y0 δy 0 + F̂y00 δy 00 ) dx.
0

From the given boundary conditions we have δy(0) = δy 0 (0) = δy(1) = 0. Using Euler’s DE, we have,
Z 1
δJ = ((F̂y0 − (F̂,y00 )0 )0 δy + F̂,y0 δy 0 + F̂y00 δy 00 ) dx.
0

1901
Now we apply integration by parts.
h i1 Z 1
0
δJ = (F̂y0 − (F̂,y00 ) )δy + (−(F̂y0 − (F̂,y00 )0 )δy 0 + F̂,y0 δy 0 + F̂y00 δy 00 ) dx
0 0
Z 1
= ((F̂,y00 )0 δy 0 + F̂y00 δy 00 ) dx
0
h i1
0
= F̂,y δy
00
0
= F̂,y00 (1)δy 0 (1)

In order that the first variation vanish, we need the natural boundary condition F̂,y00 (1) = 0. For the given
Lagrangian, this condition is
ŷ 00 (1) = 0.

The Extremal BVP. The extremal boundary value problem is

1
y 0000 = x, y(0) = y 0 (0) = y 00 (1) = 0, y(1) = .
120
The general solution of the differential equation is
1 5
y = c 0 + c 1 x + c 2 x2 + c 3 x 3 + x.
120
Applying the boundary conditions, we see that the unique admissible extremal is
x2 3
ŷ = (x − 5x + 5).
120
This may be a weak extremum for the problem.
Legendre’s Condition. Since
F̂,y00 y00 = 2 > 0,

1902
 the strengthened Legendre condition is satisfied.
Jacobi’s Condition. The second variation for F (x, y, y 00 ) is
Z b
d2 J 
 
00 2 00 2
= F̂,y 00 y 00 (h ) + 2F̂,yy 00 hh + F̂,yy h dx
d2 =0 a

Jacobi’s accessory equation is,

(2F̂,y00 y00 h00 + 2F̂,yy00 h)00 + 2F̂,yy00 h00 + 2F̂,yy h = 0,

(h00 )00 = 0

Since the boundary value problem,

h0000 = 0, h(0) = h0 (0) = h(c) = h00 (c) = 0,

has only the trivial solution for all c > 0 the strengthened Jacobi condition is satisfied.

A Weak Minimum. Since the admissible extremal,

x2 3
ŷ = (x − 5x + 5),
120
satisfies the strengthened Legendre and Jacobi conditions, we conclude that it
is a weak minimum.

2.
Z 1  
1 0 2
(y ) + yy 0 + y 0 + y dx
0 2

1903
Boundary Conditions. Since no boundary conditions are specified, we have the Euler boundary condi-
tions,

F̂,y0 (0) = 0, F̂,y0 (1) = 0.

The derivatives of the integrand are,

F,y = y 0 + 1, F,y0 = y 0 + y + 1.

The Euler boundary conditions are then

ŷ 0 (0) + ŷ(0) + 1 = 0, ŷ 0 (1) + ŷ(1) + 1 = 0.

Erdmann’s Corner Conditions. Erdmann’s first corner condition specifies that

F̂y0 (x) = ŷ 0 (x) + ŷ(x) + 1

must be continuous at a corner. This implies that ŷ 0 (x) is continuous at corners, which means that there
are no corners.
Euler’s Differential Equation. Euler’s DE is

(F,y0 )0 = Fy ,

y 00 + y 0 = y 0 + 1,

y 00 = 1.

The general solution is

1
y = c0 + c1 x + x2 .
2

1904
The boundary conditions give us the constraints,
c0 + c1 + 1 = 0,
5
c0 + 2c1 + = 0.
2
The extremal that satisfies the Euler DE and the Euler BC’s is
1 2

ŷ = x − 3x + 1 .
2
Legendre’s Condition. Since the strengthened Legendre condition is satisfied,
F̂,y0 y0 (x) = 1 > 0,
we conclude that the extremal is a weak local minimum of the problem.
Jacobi’s Condition. Jacobi’s accessory equation for this problem is,
 0  
F̂,y0 y0 h0 − F̂,yy − (F̂,yy0 )0 h = 0, h(0) = h(c) = 0,

0
(h0 ) − (−(1)0 ) h = 0, h(0) = h(c) = 0,

h00 = 0, h(0) = h(c) = 0,

Since this has only trivial solutions for c > 0 we conclude that there are no conjugate points. The extremal
satisfies the strengthened Jacobi condition.

The only admissible extremal,

1 2

ŷ = x − 3x + 1 ,
2
satisfies the strengthened Legendre and Jacobi conditions and is thus a weak
extremum.

1905
3.
Z b
(y 2 + 2xyy 0 ) dx, y(a) = A, y(b) = B
a

Euler’s Differential Equation. Euler’s differential equation,

(F,y0 )0 = Fy ,

(2xy)0 = 2y + 2xy 0 ,

2y + 2xy 0 = 2y + 2xy 0 ,

is trivial. Every C 1 function satisfies the Euler DE.

Erdmann’s Corner Conditions. The expressions,

F̂,y0 = 2xy, F̂ − ŷ 0 F̂,y0 = ŷ 2 + 2xŷ ŷ 0 − ŷ 0 (2xĥ) = ŷ 2

are continuous at a corner. The conditions are trivial and do not restrict corners in the extremal.
Extremal. Any piecewise smooth function that satisfies the boundary conditions ŷ(a) = A, ŷ(b) = B is an
admissible extremal.
An Exact Derivative. At this point we note that
Z b Z b
2 0 d
(y + 2xyy ) dx = (xy 2 ) dx
a a dx
 2 b
= xy a
= bB 2 − aA2 .

The integral has the same value for all piecewise smooth functions y that satisfy the boundary conditions.

1906
 Since the integral has the same value for all piecewise smooth functions that
satisfy the boundary conditions, all such functions are weak extrema.

4.
Z 1
(xy + y 2 − 2y 2 y 0 ) dx, y(0) = 1, y(1) = 2
0

Erdmann’s Corner Conditions. Erdmann’s first corner condition requires F̂,y0 = −2ŷ 2 to be continuous,
which is trivial. Erdmann’s second corner condition requires that

F̂ − ŷ 0 F̂,y0 = xŷ + ŷ 2 − 2ŷ 2 ŷ 0 − ŷ 0 (−2ŷ 2 ) = xŷ + ŷ 2

is continuous. This condition is also trivial. Thus the extremal may have corners at any point.
Euler’s Differential Equation. Euler’s DE is

(F,y0 )0 = F,y ,

(−2y 2 )0 = x + 2y − 4yy 0

x
y=−
2

Extremal. There is no piecewise smooth function that satisfies Euler’s differ-

ential equation on its smooth segments and satisfies the boundary conditions
y(0) = 1, y(1) = 2. We conclude that there is no weak extremum.

1907
Solution 49.4
1. We require that the first variation vanishes
ZZ


Fu h + Fux hx + Fuy hy dx dy = 0.
D

We rewrite the integrand as

ZZ


Fu h + (Fux h)x + (Fuy h)y − (Fux )x h − (Fuy )y h dx dy = 0,
D

ZZ ZZ



Fu − (Fux )x − (Fuy )y h dx dy + (Fux h)x + (Fuy h)y dx dy = 0.
D D

Using the Divergence theorem, we obtain,

ZZ Z


Fu − (Fux )x − (Fuy )y h dx dy + (Fux , Fuy ) · n h ds = 0.
D Γ

In order that the line integral vanish we have the natural boundary condition,

(Fux , Fuy ) · n = 0 for (x, y) ∈ Γ.

We can also write this as

dy dx
Fux − Fuy = 0 for (x, y) ∈ Γ.
ds ds
The Euler differential equation for this problem is

Fu − (Fux )x − (Fuy )y = 0.

1908
 2. We consider the natural boundary conditions for
ZZ Z
F (x, y, u, ux , uy ) dx dy + G(x, y, u) ds.
D Γ

We require that the first variation vanishes.

ZZ Z Z


Fu − (Fux )x − (Fuy )y h dx dy + (Fux , Fuy ) · n h ds + Gu h ds = 0,
D Γ Γ

ZZ Z



Fu − (Fux )x − (Fuy )y h dx dy + (Fux , Fuy ) · n + Gu h ds = 0,
D Γ

In order that the line integral vanishes, we have the natural boundary conditions,
(Fux , Fuy ) · n + Gu = 0 for (x, y) ∈ Γ.
For the given integrand this is,
(2pux , 2puy ) · n + 2σu = 0 for (x, y) ∈ Γ,

p∇u · n + σu = 0 for (x, y) ∈ Γ.

We can also denote this as

∂u
p + σu = 0 for (x, y) ∈ Γ.
∂n

Solution 49.5
First we vary φ.
ZZ Z h(x,t)   !
1 1
ψ() = φt + ηt + (φx + ηx )2 + (φy + ηy )2 + gy dy dx dt
R 0 2 2

1909
 !
ZZ Z h(x,t)
ψ 0 (0) = (ηt + φx ηx + φy ηy ) dy dx dt = 0
R 0

ZZ  Z h(x,t) Z h(x,t) Z h(x,t)

0 ∂ ∂
ψ (0) = η dy − [ηht ]y=h(x,t) + φx η dy − [φx ηhx ]y=h(x,t) − φxx η dy
R ∂t0 ∂x 0 0
Z h(x,t) 
h(x,t)
+ [φy η]0 − φyy η dy dx dt = 0
0

Since η vanishes on the boundary of R, we have

ZZ  Z h(x,t) 
0
ψ (0) = − [(ht φx hx − φy )η]y=h(x,t) − [φy η]y=0 − (φxx + φyy )η dy dx dt = 0.
R 0

From the variations η which vanish on y = 0, h(x, t) we have

∇2 φ = 0.

This leaves us with

ZZ  
0
ψ (0) = − [(ht φx hx − φy )η]y=h(x,t) − [φy η]y=0 dx dt = 0.
R

By considering variations η which vanish on y = 0 we obtain,

ht φx hx − φy = 0 on y = h(x, t).

Finally we have

φy = 0 on y = 0.

1910
 Next we vary h(x, t).
ZZ Z h(x,t)+η(x,t)  
1 2 1 2
ψ() = φt + φx + φy + gy dx dt
R 0 2 2

ZZ  
0 1 2 1 2
ψ () = φt + φx + φy + gy η dx dt = 0
R 2 2 y=h(x,t)

This gives us the boundary condition,

1 1
φt + φ2x + φ2y + gy = 0 on y = h(x, t).
2 2

Solution 49.6
The parts of the extremizing curve which lie outside the boundary of the region R must be extremals, (i.e.,
solutions of Euler’s equation) since if we restrict our variations to admissible curves outside of R and its boundary,
we immediately obtain Euler’s equation. Therefore an extremum can be reached only on curves consisting of arcs
of extremals and parts of the boundary of region R.
Thus, our problem is to find the points of transition of the extremal to the boundary of R. Let the boundary
of R be given by φ(x). Consider an extremum that starts at the point (a, A), follows an extremal to the point
(x0 , φ(x0 )), follows the ∂R to (x1 , φ(x1 )) then follows an extremal to the point (b, B). We seek transversality
conditions for the points x0 and x1 . We will extremize the expression,
Z x0 Z x1 Z b
0 0
I(y) = F (x, y, y ) dx + F (x, φ, φ ) dx + F (x, y, y 0 ) dx.
a x0 x1

Let c be any point between x0 and x1 . Then extremizing I(y) is equivalent to extremizing the two functionals,
Z x0 Z c
0
I1 (y) = F (x, y, y ) dx + F (x, φ, φ0 ) dx,
a x0

1911
 Z x1 Z b
0
I2 (y) = F (x, φ, φ ) dx + F (x, y, y 0 ) dx,
c x1

δI = 0 ⇒ δI1 = δI2 = 0.

We will extremize I1 (y) and then use the derived transversality condition on all points where the extremals meet
∂R. The general variation of I1 is,
Z x0  
d x
δI1 (y) = Fy − Fy0 dx + [Fy0 δy]xa0 + [(F − y 0 Fy0 )δx]a0
a dx
c
+ [Fφ0 δφ(x)]cx0 + [(F − φ0 Fφ0 )δx]x0 = 0

Note that δx = δy = 0 at x = a, c. That is, x = x0 is the only point that varies. Also note that δφ(x) is not
independent of δx. δφ(x) → φ0 (x)δx. At the point x0 we have δy → φ0 (x)δx.
Z x0    
d 0
 0

δI1 (y) = Fy − Fy0 dx + (Fy0 φ δx)  + ((F − y Fy0 )δx) 

a dx
 x0 x0

− (Fφ0 φ0 δx)  − ((F − φ0 Fφ0 )δx)  = 0

 
x0 x0

x0
Z   
d 0 0 0 0

δI1 (y) = Fy − Fy0 dx + ((F (x, y, y ) − F (x, φ, φ ) + (φ − y )Fy0 )δx)  = 0
a dx x0

Since δI1 vanishes for those variations satisfying δx0 = 0 we obtain the Euler differential equation,
d
Fy − Fy0 = 0.
dx
Then we have

0 0 0 0

((F (x, y, y ) − F (x, φ, φ ) + (φ − y )Fy0 )δx)  =0
x0

1912
for all variations δx0 . This implies that

(F (x, y, y 0 ) − F (x, φ, φ0 ) + (φ0 − y 0 )Fy0 ) 

= 0.
x0

Two solutions of this equation are

y 0 (x0 ) = φ0 (x0 ) and Fy0 = 0.

Transversality condition. If Fy0 is not identically zero, the extremal must

be tangent to ∂R at the points of contact.

R 10
Now we apply this result to to find the curves which extremize 0 (y 0 )3 dx, y(0) = 0, y(10) = 0 given that the
admissible curves can not penetrate the interior of the circle (x − 5)2 + y 2 = 9. Since the Lagrangian is a function
of y 0 alone, the extremals are straight lines.
The Erdmann corner conditions require that

Fy0 = 3(y 0 )2 and F − y 0 Fy0 = (y 0 )3 − y 0 3(y 0 )2 = −2(y 0 )3

are continuous at corners. This implies that y 0 is continuous. There are no corners.
We see that the extrema are

3
±p
 4
x, for 0 ≤ x ≤ 16
5
,
2 16 34
y(x) = ± 9 − (x − 5) , for 5 ≤ x ≤ 5 ,

 3
∓ 4 x, for 34
5
≤ x ≤ 10.

Note that the extremizing curves neither minimize nor maximize the integral.

1913
Solution 49.7
C1 Extremals. Without loss of generality, we take the vertical line to be the y axis. We will consider x1 , y1 > 1.
p
With ds = 1 + (y 0 )2 dx we extremize the integral,
Z x1
√ p
y 1 + (y 0 )2 dx.
0

Since the Lagrangian is independent of x, we know that the Euler differential equation has a first integral.
d
Fy0 − Fy = 0
dx

y 0 Fy0 y + y 00 Fy0 y0 − Fy = 0

d 0
(y Fy0 − F ) = 0
dx

y 0 Fy0 − F = const

For the given Lagrangian, this is

√ y0 √ p
y0 y p − y 1 + (y 0 )2 = const,
1 + (y 0 )2

√ √ p
(y 0 )2 y − y(1 + (y 0 )2 ) = const 1 + (y 0 )2 ,

√ p
y = const 1 + (y 0 )2

y = const is one solution. To find the others we solve for y 0 and then solve the differential equation.

y = a(1 + (y 0 )2 )

1914
 r
y−a
y0 = ±
a
r
a
dx = dy
y−a

p
±x + b = 2 a(y − a)

x2 bx b2
y= ± + +a
4a 2a 4a
The natural boundary condition is
√ 0 
 yy
Fy0 x=0 = p  = 0,
1 + (y 0 )2 x=0

y 0 (0) = 0
The extremal that satisfies this boundary condition is
x2
y= + a.
4a
Now we apply y(x1 ) = y1 to obtain
 
1
q
2 2
a= y1 ± y1 − x1
2
for y1 ≥ x1 . The value of the integral is
Z x1 s 2
x1 (x21 + 12a2 )
  x 2 
x
+a 1+ dx = .
0 4a 2a 12a3/2

1915
By denoting y1 = cx1 , c ≥ 1 we have
1 √ 
a= 2
cx1 ± x1 c − 1
2
The values of the integral for these two values of a are
√
√ 2 2
3/2 −1 + 3c ± 3c c − 1
2(x1 ) √ .
3(c ± c2 − 1)3/2
√
The values are equal only when c = 1. These values, (divided by x1 ), are plotted in Figure 49.1 as a function
of c. The former and latter are fine and coarse dashed lines, respectively. The extremal with
 
1
q
2 2
a= y1 + y1 − x1
2

has the smaller performance index. The value of the integral is

p
x1 (x21 + 3(y1 + y12 − x21 )2
√ p .
3 2(y1 + y12 − x21 )3
√
The function y = y1 is an admissible extremal for all x1 . The value of the integral for this extremal is x1 y1
which is larger than the integral of the quadratic we analyzed before for y1 > x1 .
Thus we see that

x2
 
1
q
2 2
ŷ = + a, a= y 1 + y 1 − x1
4a 2

is the extremal with the smaller integral and is the minimizing curve in C 1 for y1 ≥ x1 . For y1 < x1 the C 1
extremum is,

ŷ = y1 .

1916
 4

3.5

2.5

1.2 1.4 1.6 1.8 2

Figure 49.1:

C1p Extremals. Consider the parametric form of the Lagrangian.

Z t1 p p
y(t) (x0 (t))2 + (y 0 (t))2 dt
t0

The Euler differential equations are

d d
fx0 − fx = 0 and fy0 − fy = 0.
dt dt
If one of the equations is satisfied, then the other is automatically satisfied, (or the extremal is straight). With
either of these equations we could derive the quadratic extremal and the y = const extremal that we found
previously. We will find one more extremal by considering the first parametric Euler differential equation.
d
fx0 − fx = 0
dt
p !
d y(t)x0 (t)
p =0
dt (x0 (t))2 + (y 0 (t))2

1917
 p
y(t)x0 (t)
p = const
(x0 (t))2 + (y 0 (t))2
Note that x(t) = const is a solution. Thus the extremals are of the three forms,
x = const,
y = const,
x2 bx b2
y= + + + a.
4a 2a 4a
The Erdmann corner conditions require that
√ 0
yy
Fy0 = p ,
1 + (y 0 )2
√ 0 2 √
0 √ p y(y ) y
F − y Fy0 = y 1 + (y 0 )2 − p =p
1 + (y 0 )2 1 + (y 0 )2
are continuous at corners. There can be corners only if y = 0.
Now we piece the three forms together to obtain Cp1 extremals that satisfy the Erdmann corner conditions.
The only possibility that is not C 1 is the extremal that is a horizontal line from (0, 0) to (x1 , 0) and then a vertical
line from (x1 , y1 ). The value of the integral for this extremal is
Z y1 √
2
t dt = (y1 )3/2 .
0 3
Equating the performance indices of the quadratic extremum and the piecewise smooth extremum,
p
x1 (x21 + 3(y1 + y12 − x21 )2 2
√ p = (y1 )3/2 ,
2
3 2(y1 + y1 − x1 ) 2 3 3

p √
3±2 3
y1 = ±x1 √ .
3

1918
The only real positive solution is
p √
3+2 3
y 1 = x1 √ ≈ 1.46789 x1 .
3

The piecewise smooth extremal has the smaller performance index for y1 smaller than this value and the quadratic
extremal has the smaller performance index for y1 greater than this value.

p √ √
The Cp1 extremum is the piecewise smooth extremal for y1 ≤ x1 3 + 2 3/ 3
p √ √
and is the quadratic extremal for y1 ≥ x1 3 + 2 3/ 3.

Solution 49.8
The shape of the rope will be a catenary between x1 and x2 and be a vertically hanging segment after that. Let
the length of the vertical segment be z. Without loss of generality we take x1 = y2 = 0. The potential energy,
(relative to y = 0), of a length of rope ds in 0 ≤ x ≤ x2 is mgy = ρgy ds. The total potential energy of the
vertically hanging rope is m(center of mass)g = ρz(−z/2)g. Thus we seek to minimize,
Z x2
1
ρg y ds − ρgz 2 , y(0) = y1 , y(x2 ) = 0,
0 2

subject to the isoperimetric constraint,

Z x2
ds − z = L.
0
p
Writing the arc-length differential as ds = 1 + (y 0 )2 dx we minimize
Z x2 p 1
ρg y 1 + (y 0 )2 ds − ρgz 2 , y(0) = y1 , y(x2 ) = 0,
0 2

1919
subject to,
Z x2 p
1 + (y 0 )2 dx − z = L.
0
Rb
Consider the more general problem of finding functions y(x) and numbers z which extremize I ≡ a F (x, y, y 0 ) dx+
Rb
f (z) subject to J ≡ a G(x, y, y 0 ) dx + g(z) = L.
Suppose y(x) and z are the desired solutions and form the comparison families, y(x) + 1 η1 (x) + 2 η2 (x),
z + 1 ζ1 + 2 ζ2 . Then, there exists a constant such that
∂ 
(I + λJ)1 ,2 =0 = 0
∂1
∂ 
(I + λJ)1 ,2 =0 = 0.
∂2
These equations are
Z b 
d
H,y0 − Hy η1 dx + h0 (z)ζ1 = 0,
a dx
and
Z b 
d
H,y0 − Hy η2 dx + h0 (z)ζ2 = 0,
a dx
where H = F + λG and h = f + λg. From this we conclude that
d
H,y0 − Hy = 0, h0 (z) = 0
dx
with λ determined by
Z b
J= G(x, y, y 0 ) dx + g(z) = L.
a

1920
 Now we apply these results to our problem. Since f (z) = − 21 ρgz 2 and g(z) = −z we have
−ρgz − λ = 0,

λ
z=− .
ρg

It was shown in class that the solution of the Euler differential equation is a family of catenaries,
 
λ x − c2
y = − + c1 cosh .
ρg c1
One can find c1 and c2 in terms of λ by applying the end conditions y(0) = y1 and y(x2 ) = 0. Then the expression
for y(x) and z = −λ/ρg are substituted into the isoperimetric constraint to determine λ.
Consider the special case that (x1 , y1 ) = (0, 0) and (x2 , y2 ) = (1, 0). In this case we can use the fact that
y(0) = y(1) to solve for c2 and write y in the form
 
λ x − 1/2
y = − + c1 cosh .
ρg c1
Applying the condition y(0) = 0 would give us the algebraic-transcendental equation,
 
λ 1
y(0) = − + c1 cosh = 0,
ρg 2c1
which we can’t solve in closed form. Since we ran into a dead end in applying the boundary condition, we turn
to the isoperimetric constraint.
Z 1p
1 + (y 0 )2 dx − z = L
0

1  
x − 1/2
Z
cosh dx − z = L
0 c1

1921
  
1
2c1 sinh −z =L
2c1
With the isoperimetric constraint, the algebraic-transcendental equation and z = −λ/ρg we now have
 
1
z = −c1 cosh ,
2c1
 
1
z = 2c1 sinh − L.
2c1
For any fixed L, we can numerically solve for c1 and thus obtain z. You can derive that there are no solutions
unless L is greater than about 1.9366. If L is smaller than this, the rope would slip off the pin. For L = 2, c1 has
the values 0.4265 and 0.7524. The larger value of c1 gives the smaller potential energy. The position of the end
of the rope is z = −0.9248.

Solution 49.9 Rc
Using the method of Lagrange multipliers, we look for stationary values of 0 ((y 0 )2 + λy 2 ) dx,
Z c
δ ((y 0 )2 + λy 2 ) dx = 0.
0

The Euler differential equation is

d
F( , y 0 ) − F,y = 0,
dx

d
(2y 0 ) − 2λy = 0.
dx
Together with the homogeneous boundary conditions, we have the problem

y 00 − λy = 0, y(0) = y(c) = 0,

1922
which has the solutions,
 nπ 2  nπx 
λn = − , yn = an sin , n ∈ Z+ .
c c
Now we determine the constants an with the moment of inertia constraint.
Z c  nπx  ca2
a2n sin2 dx = n = A
0 c 2

Thus we have the extremals,

r
2A  nπx 
yn = sin , n ∈ Z+ .
c c
The drag for these extremals is
Z c  2
2A nπ 2 nπx
  An2 π 2
D= cos dx = .
c 0 c c c2

We see that the drag is minimum for n = 1. The shape for minimum drag is
r
2A  nπx 
ŷ = sin .
c c

Solution 49.10
Consider the general problem of determining the stationary values of the quantity ω 2 given by
Rb
2 F (x, y, y 0 , y 00 ) dx I
ω = Rab ≡ .
G(x, y, y 0 , y 00 ) dx J
a

1923
The variation of ω 2 is
JδI − IδJ
δω 2 = 2
J 
1 I
= δI − δJ
J J
1
δI − ω 2 δJ .


=
J
The the values of y and y 0 are specified on the boundary, then the variations of I and J are
Z b 2  Z b 2 
d d d d
δI = F,y00 − F,y0 + F,y δy dx, δJ = G,y00 − G,y0 + G,y δy dx
a dx2 dx a dx2 dx

Thus δω 2 = 0 becomes
R b  d2 d

a
H
dx2 ,y
00 − H
dx ,y
0 + H ,y δy dx
Rb = 0,
a
G dx

where H = F − ω 2 G. A necessary condition for an extremum is

d2 d
H ,y 00 − H,y0 + H,y = 0 where H ≡ F − ω 2 G.
dx2 dx

For our problem we have F = EI(y 00 )2 and G = ρy so that the extremals are solutions of

d2 d2 y
 
EI 2 − ρω 2 y = 0,
dx2 dx

With homogeneous boundary conditions we have an eigenvalue problem with deflections modes yn (x) and corre-
sponding natural frequencies ωn .

1924
Solution 49.11
We assume that v0 > w(x, y, t) so that the problem has a solution for any end point. The crossing time is
Z l
1 l
−1 Z
T = Ẋ(t) dx = sec α(t) dx.
0 v0 0
Note that
dy w + v0 sin α
=
dx v0 cos α
w
= sec α + tan α
v0
w √
= sec α + sec2 α − 1.
v0
We solve this relation for sec α.
 2
0w
y − sec α = sec2 α − 1
v0

w 0 w2
(y 0 )2 − 2 y sec α + 2 sec2 α = sec2 α − 1
v0 v0

(v02 − w2 ) sec2 α + 2v0 wy 0 sec α − v02 ((y 0 )2 + 1) = 0

p
−2v0 wy 0 ± 4v02 w2 (y 0 )2 + 4(v02 − w2 )v02 ((y 0 )2 + 1)
sec α =
2(v02 − w2 )

p
−wy 0 ± v02 ((y 0 )2 + 1) − w2
sec α = v0
(v02 − w2 )

1925
Since the steering angle satisfies −π/2 ≤ α ≤ π/2 only the positive solution is relevant.
p
−wy 0 + v02 ((y 0 )2 + 1) − w2
sec α = v0
(v02 − w2 )
Time Independent Current. If we make the assumption that w = w(x, y) then we can write the crossing
time as an integral of a function of x and y.
Z l p
−wy 0 + v02 ((y 0 )2 + 1) − w2
T (y) = dx
0 (v02 − w2 )
A necessary condition for a minimum is δT = 0. The Euler differential equation for this problem is
d
F,y0 − F,y = 0
dx

!! !
d 1 v02 y 0 wy w(v 2 (1 + 2(y 0 )2 ) − w2 )
2
−w + p − 2 p − y 0 (v02 + w2 )
dx v0 − w2 v02 ((y 0 )2 + 1) − w2 (v0 − w2 )2 2 0 2
v0 ((y ) + 1) − w 2

By solving this second order differential equation subject to the boundary conditions y(0) = 0, y(l) = y1 we
obtain the path of minimum crossing time.
Current w = w(x). If the current is only a function of x, then the Euler differential equation can be integrated
to obtain,
!
1 v02 y 0
−w + p 2 = c0 .
v02 − w2 v0 ((y 0 )2 + 1) − w2

Solving for y 0 ,
w + c0 (v02 − w2 )
y0 = ± p .
v0 1 − 2c0 w − c20 (v02 − w2 )

1926
Since y(0) = 0, we have
x
w(ξ) + c0 (v02 − (w(ξ))2 )
Z
y(x) = ± p .
0 v0 1 − 2c0 w(ξ) − c20 (v02 − (w(ξ))2 )

For any given w(x) we can use the condition y(l) = y1 to solve for the constant c0 .
Constant Current. If the current is constant then the Lagrangian is a function of y 0 alone. The admissible
extremals are straight lines. The solution is then
y1 x
y(x) = .
l

Solution 49.12
1. The kinetic energy of the first particle is 12 m((α − x)θ̂)2 . Its potential energy, relative to the table top, is
zero. The kinetic energy of the second particle is 21 mx̂2 . Its potential energy, relative to its equilibrium
position is −mgx. The Lagrangian is the difference of kinetic and potential energy.
 
2 1 2 2
L = m ẋ + (α − x) θ̇ + gx
2

The Euler differential equations are the equations of motion.

d d
L,ẋ − Lx = 0, L − Lθ = 0
dt dt ,θ̇

d d  
(2mẋ) + m(α − x)θ̇2 − mg = 0, m(α − x)2 θ̇2 = 0
dt dt

2ẍ + (α − x)θ̇2 − g = 0, (α − x)2 θ̇2 = const

1927
 p
When x = 0, θ̇ = ω = g/α. This determines the constant in the equation of motion for θ.
√
α αg
θ̇ =
(α − x)2

Now we substitute the expression for θ̇ into the equation of motion for x.

α3 g
2ẍ + (α − x) −g =0
(α − x)4

α3
 
2ẍ + −1 g =0
(α − x)3

 
1
2ẍ + −1 g =0
(1 − x/α)3

 
2. For small oscillations,  αx   1. Recall the binomial expansion,
∞  
a
X a n
(1 + z) = z , for |z| < 1,
n=0
n

(1 + z)a ≈ 1 + az, for |z|  1.

We make the approximation,

1 x
3
≈1+3 ,
(1 − x/α) α

1928
 to obtain the linearized equation of motion,
3g
x = 0.
2ẍ +
α
This is the equation of a harmonic oscillator with solution
p 
x = a sin 3g2α(t − b) .

The period of oscillation is,

√
T = 2π 2α3g.

Solution 49.13
We write the equation of motion and boundary conditions,
ẍ = U (t) − g, x(0) = ẋ(0) = 0, x(T ) = h,
as the first order system,
ẋ = 0, x(0) = 0, x(T ) = h,
ẏ = U (t) − g, y(0) = 0.
We seek to minimize,
Z T
T = dt,
0

subject to the constraints,

ẋ − y = 0,
ẏ − U (t) + g = 0,
Z T
U 2 (t) dt = k 2 .
0

1929
Thus we seek extrema of
Z T Z T
1 + λ(t)(ẋ − y) + µ(t)(ẏ − U (t) + g) + νU 2 (t) dt.


H dt ≡
0 0

Since y is not specified at t = T , we have the natural boundary condition,


H,ẏ t=T = 0,

µ(T ) = 0.

The first Euler differential equation is

d
H,ẋ − H,x = 0,
dt

d
λ(t) = 0.
dt
We see that λ(t) = λ is constant. The next Euler DE is
d
H,ẏ − H,y = 0,
dt

d
µ(t) + λ = 0.
dt

µ(t) = −λt + const

With the natural boundary condition, µ(T ) = 0, we have

µ(t) = λ(T − t).

1930
The final Euler DE is,
d
H − H,U = 0,
dt ,U̇

µ(t) − 2νU (t) = 0.

Thus we have
λ(T − t)
U (t) = .
2ν
This is the required thrust function. We use the constraints to find λ, ν Rand T .
T
Substituting U (t) = λ(T − t)/(2ν) into the isoperimetric constraint, 0 U 2 (t) dt = k 2 yields
λ2 T 3
= k2,
12ν 2
√
3k
U (t) = (T − t).
T 3/2
The equation of motion for x is
√
3k
ẍ = U (t) − g = 3/2 (T − t).
T
Integrating and applying the initial conditions x(0) = ẋ(0) = 0 yields,
kt2 (3T − t) 1 2
x(t) = √ − gt .
2 3T 3/2 2
Applying the condition x(T ) = h gives us,
k 1
√ T 3/2 − gT 2 = h,
3 2

1931
 1 2 4 k 3
g T − T + ghT 2 + h2 = 0.
4 3
p √
If k ≥ 4 2/3g 3/2 h then this fourth degree polynomial has positive, real solutions
p for
√ T . With strict inequality,
3/2
the minimum time is the smaller of the two positive, real solutions. If k < 4 2/3g h then there is not enough
fuel to reach the target height.

Solution 49.14
We have ẍ = U (t) where U (t) is the acceleration furnished by the thrust of the vehicles engine. In practice, the
engine will be designed to operate within certain bounds, say −M ≤ U (t) ≤ M , where ±M is the maximum
forward/backward acceleration. To account for the inequality constraint we write U = M sin V (t) for some
suitable V (t). More generally, if we had φ(t) ≤ U (t) ≤ ψ(t), we could write this as U (t) = ψ+φ
2
+ ψ−φ
2
sin V (t).
We write the equation of motion as a first order system,

ẋ = y, x(0) = a, x(T ) = 0,
ẏ = M sin V, y(0) = b, y(T ) = 0.

Thus we minimize
Z T
T = dt
0

subject to the constraints,

ẋ − y = 0
ẏ − M sin V = 0.

Consider

H = 1 + λ(t)(ẋ − y) + µ(t)(ẏ − M sin V ).

1932
The Euler differential equations are
d d
H,ẋ − H,x = 0 ⇒ λ(t) = 0 ⇒ λ(t) = const
dt dt
d d
H,ẏ − H,y = 0 ⇒ µ(t) + λ = 0 ⇒ µ(t) = −λt + const
dt dt
d π
H,V̇ − H,V = 0 ⇒ µ(t)M cos V (t) = 0 ⇒ V (t) = + nπ.
dt 2
Thus we see that
π 
U (t) = M sin + nπ = ±M.
2
Therefore, if the rocket is to be transferred from its initial state to is specified final state in minimum time with
a limited source of thrust, (|U | ≤ M ), then the engine should operate at full power at all times except possibly
for a finite number of switching times. (Indeed, if some power were not being used, we would expect the transfer
would be speeded up by using the additional power suitably.)
To see how this ”bang-bang” process works, we’ll look at the phase plane. The problem

ẋ = y, x(0) = c,
ẏ = ±M, y(0) = d,

has the solution

t2
x(t) = c + dt ± M , y(t) = d ± M t.
2
We can eliminate t to get

y2 d2
x=± +c∓ .
2M 2M
These curves are plotted in Figure 49.2.

1933
 Figure 49.2:

There is only curve in each case which transfers the initial state to the origin. We will denote these curves
γ and Γ, respectively. Only if the initial point (a, b) lies on one of these two curves can we transfer the state of
b2
the system to the origin along an extremal without switching. If a = 2M and b < 0 then this is possible using
b2
U (t) = M . If a = − 2M and b > 0 then this is possible using U (t) = −M . Otherwise we follow an extremal that
intersects the initial position until this curve intersects γ or Γ. We then follow γ or Γ to the origin.

Solution 49.15
Since the integrand does not explicitly depend on x, the Euler differential equation has the first integral,
F − y 0 Fy0 = const.
0
√
p p 0 y y+h
y + h 1 + (y 0 )2 − y p = const
1 + (y 0 )2
√
y+h
p = const
1 + (y 0 )2

1934
 y + h = c21 (1 + (y 0 )2 )

q
y + h − c21 = c1 y 0

c1 dy
p = dx
y + h − c21

q
2c1 y + h − c21 = x − c2

4c21 (y + h − c21 ) = (x − c2 )2

Since the extremal passes through the origin, we have

4c21 (h − c21 ) = c22 .

4c21 y = x2 − 2c2 x (49.6)

Introduce as a parameter the slope of the extremal at the origin; that is, y 0 (0) = α. Then differentiating (49.6)
h 2αh
at x = 0 yields 4c21 α = −2c2 . Together with c22 = 4c21 (h − c21 ) we obtain c21 = 1+α 2 and c2 = − 1+α2 . Thus the

equation of the pencil (49.6) will have the form

1 + α2 2
y = αx + x. (49.7)
4h
α 2
To find the envelope of this family we differentiate ( 49.7) with respect to α to obtain 0 = x + 2h x and eliminate
α between this and ( 49.7) to obtain
x2
y = −h + .
4h

1935
 y

x
h 2h

-h

Figure 49.3: Some Extremals and the Envelope.

See Figure 49.3 for a plot of some extremals and the envelope.

All extremals (49.7) lie above the envelope which in ballistics is called the parabola of safety. If (m, M ) lies
2
outside the parabola, M < −h + m 4h
, then it cannot be joined to (0, 0) by an extremal. If (m, M ) is above the
envelope then there are two candidates. Clearly we rule out the one that touches the envelope because of the
occurrence of conjugate points. For the other extremal, problem 2 shows that E ≥ 0 for all y 0 . Clearly we can
embed this extremal in an extremal pencil, so Jacobi’s test is satisfied. Therefore the parabola that does not
touch the envelope is a strong minimum.

1936
Solution 49.16

E = F (x, y, y 0 ) − F (x, y, p) − (y 0 − p)Fy0 (x, y, p)

p p np
= n 1 + (y 0 )2 − n 1 + p2 − (y 0 − p) p
1 + p2
n  p p 
=p 1 + (y 0 )2 1 + p2 − (1 + p2 ) − (y 0 − p)p
1 + p2
n p 
=p 1 + (y 0 )2 + p2 + (y 0 )2 p2 − 2y 0 p + 2y 0 p − (1 + py 0 )
1 + p2
n p 
=p (1 + py 0 )2 + (y 0 − p)2 − (1 + py 0 )
1 + p2
≥0
ds 1
The speed of light in an inhomogeneous medium is dt
= n(x,y
. The time of transit is then
Z (b,B) Z b
dt p
T = ds = n(x, y) 1 + (y 0 )2 dx.
(a,A) ds a

Since E ≥ 0, light traveling on extremals follow the time optimal path as long as the extremals do not intersect.

Solution 49.17
Extremals. Since the integrand does not depend explicitly on x, the Euler differential equation has the first
integral,

F − y 0 F,y0 = const.

1 + y2 0 −2(1 + y )
2
− y = const
(y 0 )2 (y 0 )3

1937
 dy
p = const dx
1 + (y 0 )2

arcsinh (y) = c1 x + c2

y = sinh(c1 x + c2 )
Jacobi Test. We can see by inspection that no conjugate points exist. Consider the central field through
(0, 0), sinh(cx), (See Figure 49.4).
3

-3 -2 -1 1 2 3

-1

-2

-3

Figure 49.4: sinh(cx)

We can also easily arrive at this conclusion analytically as follows: Solutions u1 and u2 of the Jacobi equation
are given by
∂y
u1 = = cosh(c1 x + c2 ),
∂c2
∂y
u2 = = x cosh(c1 x + c2 ).
∂c1

1938
Since u2 /u1 = x is monotone for all x there are no conjugate points.
Weierstrass Test.
E = F (x, y, y 0 ) − F (x, y, p) − (y 0 − p)F,y0 (x, y, p)
1 + y2 1 + y2 0 −2(1 + y 2 )
= − − (y − p)
(y 0 )2 p2 p3
1 + y 2 p3 − p(y 0 )2 + 2(y 0 )3 − 2p(y 0 )2
 
= 0 2 2
(y ) p p
2
(p − y ) (p + 2y 0 )
0 2
 
1+y
= 0 2 2
(y ) p p
For p = p(x, y) bounded away from zero, E is one-signed for values of y 0 close to p. However, since the factor
(p + 2y 0 ) can have any sign for arbitrary values of y 0 , the conditions for a strong minimum are not satisfied.
Furthermore, since the extremals are y = sinh(c1 x + c2 ), the slope function p(x, y) will be of one sign only if
the range of integration is such that we are on a monotonic piece of the sinh. If we span both an increasing and
decreasing section, E changes sign even for weak variations.
Legendre Condition.
6(1 + y 2 )
F,y0 y0 = >0
(y 0 )4
Note that F cannot be represented in a Taylor series for arbitrary values of y 0 due to the presence of a discontinuity
in F when y 0 = 0. However, F,y0 y0 > 0 on an extremal implies a weak minimum is provided by the extremal.
2
Strong Variations. Consider 1+y
R
(y 0 )2
dx on both an extremal and on the special piecewise continuous variation
2
in the figure. On P Q we have y 0 = ∞ with implies that 1+y (y 0 )2
= 0 so that there is no contribution to the integral
from P Q.
On QR the value of y 0 is greater than its value along the extremal P R while the value of y on QR is less than
2
the value of y along P R. Thus on QR the quantity 1+y (y 0 )2
is less than it is on the extremal P R.

1 + y2 1 + y2
Z Z
0 2
dx < 0 2
dx
QR (y ) P R (y )

1939
Thus the weak minimum along the extremal can be weakened by a strong variation.

Solution 49.18
The Euler differential equation is
d
F,y0 − F,y = 0.
dx

d
(1 + 2x2 y 0 ) = 0
dx

1 + 2x2 y 0 = const

1
y 0 = const
x2

c1
y= + c2
x
(i) No continuous extremal exists in −1 ≤ x ≤ 2 that satisfies y(−1) = 1 and y(2) = 4.
(ii) The continuous extremal that satisfies the boundary conditions is y = 7 − x4 . Since F,y0 y0 = 2x2 ≥ 0 has a
Taylor series representation for all y 0 , this extremal provides a strong minimum.
(iii) The continuous extremal that satisfies the boundary conditions is y = 1. This is a strong minimum.

Solution 49.19
For identity (a) we take P = 0 and Q = φψx − ψφx . For identity (b) we take P = φψy − ψφy and Q = 0. For
identity (c) we take P = − 12 (φψx − ψφx ) and Q = 12 (φψy − ψφy ).
ZZ     Z  
1 1 1 1
(φψy − ψφy )x − − (φψx − ψφx )y dx dy = − (φψx − ψφx ) dx + (φψy − ψφy ) dy
D 2 2 Γ 2 2

1940
 ZZ  
1 1
(φx ψy + φψxy − ψx φy − ψφxy ) + (φy ψx φψxy − ψy φx − ψφxy ) dx dy
D 2 2
Z Z
1 1
=− (φψx − ψφx ) dx + (φψy − ψφy ) dy
2 Γ 2 Γ
ZZ ZZ Z Z
1 1
φψxy dx dy = ψφxy dx dy − (φψx − ψφx ) dx + (φψy − ψφy ) dy
D D 2 Γ 2 Γ
The variation of I is
Z t1 Z Z
δI = (−2(uxx + uyy )(δuxx + δuyy ) + 2(1 − µ)(uxx δuyy + uyy δuxx − 2uxy δuxy )) dx dy dt.
t0 D

From (a) we have

ZZ ZZ
−2(uxx + uyy )δuxx dx dy = −2(uxx + uyy )xx δu dx dy
D D Z
+ −2((uxx + uyy )δux − (uxx + uyy )x δu) dy.
Γ
From (b) we have
ZZ ZZ
−2(uxx + uyy )δuyy dx dy = −2(uxx + uyy )yy δu dx dy
D D Z
− −2((uxx + uyy )δuy − (uxx + uyy )y δu) dy.
Γ
From (a) and (b) we get
ZZ
2(1 − µ)(uxx δuyy + uyy δuxx ) dx dy
D
ZZ
= 2(1 − µ)(uxxyy + uyyxx )δu dx dy
D
Z
+ 2(1 − µ)(−(uxx δuy − uxxy δu) dx + (uyy δux − uyyx δu) dy).
Γ

1941
Using c gives us
ZZ ZZ
2(1 − µ)(−2uxy δuxy ) dx dy = 2(1 − µ)(−2uxyxy δu) dx dy
D D
Z
+ 2(1 − µ)(uxy δux − uxyx δu) dx
ZΓ
− 2(1 − µ)(uxy δuy − uxyy δu) dy.
Γ

Note that
∂u
ds = ux dy − uy dx.
∂n
Using the above results, we obtain
Z t1 Z Z Z t1 Z 
∂(∇2 u)

4 2 ∂(δu)
δI = 2 (−∇ u)δu dx dy dt + 2 δu + (∇ u) ds dt
t0 D t0 Γ ∂n ∂n
Z t1 Z 
+ 2(1 − µ) (uyy δux − uxy δuy ) dy + (uxy δux − uxx δuy ) dx dt.
t0 Γ

Solution 49.20
1. Exact Solution. The Euler differential equation is
d
F,y0 = F,y
dx
d
[2y 0 ] = −2y − 2x
dx
y 00 + y = −x.

The general solution is

y = c1 cos x + c2 sin x − x.

1942
Applying the boundary conditions we obtain,

sin x
y= − x.
sin 1

The value of the integral for this extremal is

 
sin x 2
J − x = cot(1) − ≈ −0.0245741.
sin 1 3

n = 0. We consider an approximate solution of the form y(x) = ax(1 − x). We substitute this into the
functional.
Z 1
3 1
(y 0 )2 − y 2 − 2xy dx = a2 − a


J(a) =
0 10 6

The only stationary point is

3 1
J 0 (a) = a − = 0
5 6
5
a= .
18
Since
 
00 5 3
J = > 0,
18 5

we see that this point is a minimum. The approximate solution is

5
y(x) = x(1 − x).
18

1943
 0.07

0.06

0.05

0.04

0.03

0.02

0.01

0.2 0.4 0.6 0.8 1

Figure 49.5: One Term Approximation and Exact Solution.

This one term approximation and the exact solution are plotted in Figure 49.5. The value of the functional
is
5
J =− ≈ −0.0231481.
216

n = 1. We consider an approximate solution of the form y(x) = x(1 − x)(a + bx). We substitute this into
the functional.
Z 1
1
(y 0 )2 − y 2 − 2xy dx = 63a2 + 63ab + 26b2 − 35a − 21b



J(a, b) =
0 210
We find the stationary points.
1
Ja = (18a + 9b − 5) = 0
30
1
Jb = (63a + 52b − 21) = 0
210

1944
 71 7
a= , b=
369 41
Since the Hessian matrix
  3 3

Jaa Jab
H= = 53 10
26 ,
Jba Jbb 10 105

is positive definite,

3 41
> 0, det(H) = ,
5 700
we see that this point is a minimum. The approximate solution is
 
71 7
y(x) = x(1 − x) + x .
369 41

This two term approximation and the exact solution are plotted in Figure 49.6. The value of the functional
is

136
J =− ≈ −0.0245709.
5535

2. Exact Solution. The Euler differential equation is

d
F,y0 = F,y
dx
d
[2y 0 ] = 2y + 2x
dx
y 00 − y = x.

1945
 0.07

0.06

0.05

0.04

0.03

0.02

0.01

0.2 0.4 0.6 0.8 1

Figure 49.6: Two Term Approximation and Exact Solution.

The general solution is

y = c1 cosh x + c2 sinh x − x.

Applying the boundary conditions, we obtain,

2 sinh x
y= − x.
sinh 2
The value of the integral for this extremal is

2( e4 − 13)
J =− ≈ −0.517408.
3( e4 − 1)

Polynomial Approximation. Consider an approximate solution of the form

y(x) = x(2 − x)(a0 + a1 x + · · · an xn ).

1946
The one term approximate solution is
5
y(x) = − x(2 − x).
14
This one term approximation and the exact solution are plotted in Figure 49.7. The value of the functional
is
10
J =− ≈ −0.47619.
21

0.5 1 1.5 2
-0.05

-0.1

-0.15

-0.2

-0.25

-0.3

-0.35

Figure 49.7: One Term Approximation and Exact Solution.

The two term approximate solution is
 
33 7
y(x) = x(2 − x) − − x .
161 46
This two term approximation and the exact solution are plotted in Figure 49.8. The value of the functional
is
416
J =− ≈ −0.51677.
805

1947
 0.5 1 1.5 2
-0.05

-0.1

-0.15

-0.2

-0.25

-0.3

-0.35

Figure 49.8: Two Term Approximation and Exact Solution.

Sine Series Approximation. Consider an approximate solution of the form

 πx   πx 
y(x) = a1 sin + a2 sin (πx) + · · · + an sin n .
2 2
The one term approximate solution is
16  πx 
y(x) = − sin .
π(π 2 + 4) 2
This one term approximation and the exact solution are plotted in Figure 49.9. The value of the functional
is
64
J =− ≈ −0.467537.
π 2 (π 2
+ 4)

The two term approximate solution is

16  πx  2
y(x) = − 2
sin + 2
sin(πx).
π(π + 4) 2 π(π + 1)

1948
 0.5 1 1.5 2
-0.05

-0.1

-0.15

-0.2

-0.25

-0.3

-0.35

Figure 49.9: One Term Sine Series Approximation and Exact Solution.

This two term approximation and the exact solution are plotted in Figure 49.10. The value of the functional
is
4(17π 2 + 20)
J =− 2 4 ≈ −0.504823.
π (π + 5π 2 + 4)

3. Exact Solution. The Euler differential equation is

d
F,y0 = F,y
dx
d x2 − 1
[2xy 0 ] = −2 y − 2x2
dx  x 
00 1 0 1
y + y + 1 − 2 y = −x
x x
The general solution is

y = c1 J1 (x) + c2 Y1 (x) − x

1949
 0.5 1 1.5 2

-0.1

-0.2

-0.3

Figure 49.10: Two Term Sine Series Approximation and Exact Solution.

Applying the boundary conditions we obtain,

(Y1 (2) − 2Y1 (1))J1 (x) + (2J1 (1) − J1 (2))Y1 (x)

y= −x
J1 (1)Y1 (2) − Y1 (1)J1 (2)

The value of the integral for this extremal is

J ≈ −0.310947

Polynomial Approximation. Consider an approximate solution of the form

y(x) = (x − 1)(2 − x)(a0 + a1 x + · · · an xn ).

The one term approximate solution is

23
y(x) = (x − 1)(2 − x)
6(40 log 2 − 23)

1950
 This one term approximation and the exact solution are plotted in Figure 49.11. The one term approximation
is a surprisingly close to the exact solution. The value of the functional is

529
J =− ≈ −0.310935.
360(40 log 2 − 23)

0.2

0.15

0.1

0.05

1.2 1.4 1.6 1.8 2

Figure 49.11: One Term Polynomial Approximation and Exact Solution.

Solution 49.21
1. The spectrum of T is the set,

{λ : (T − λI) is not invertible.}

1951
 (T − λI)f = g
Z ∞
K(x − y)f (y) dy − λf (x) = g
−∞

K̂(ω)fˆ(ω) − λfˆ(ω) = ĝ(ω)

 
K̂(ω) − λ fˆ(ω) = ĝ(ω)

We may not be able to solve for fˆ(ω), (and hence invert T − λI), if λ = K̂(ω). Thus all values of K̂(ω) are
in the spectrum. If K̂(ω) is everywhere nonzero we consider the case λ = 0. We have the equation,
Z ∞
K(x − y)f (y) dy = 0
−∞

Since there are an infinite number of L2 (−∞, ∞) functions which satisfy this, (those which are nonzero on
a set of measure zero), we cannot invert the equation. Thus λ = 0 is in the spectrum. The spectrum of T
is the range of K̂(ω) plus zero.

2. Let λ be a nonzero eigenvalue with eigenfunction φ.

(T − λI)φ = 0, ∀x
Z ∞
K(x − y)φ(y) dy − λφ(x) = 0, ∀x
−∞

Since K is continuous, T φ is continuous. This implies that the eigenfunction φ is continuous. We take the
Fourier transform of the above equation.

K̂(ω)φ̂(ω) − λφ̂(ω) = 0, ∀ω
 
K̂(ω) − λ φ̂(ω) = 0, ∀ω

1952
 ˆ is not
If φ(x) is absolutely integrable, then φ̂(ω) is continous. Since φ(x) is not identically zero, φ(ω)
ˆ is nonzero on some interval of positive length, (a, b). From
identically zero. Continuity implies that φ(ω)
the above equation we see that K̂(ω) = λ for ω ∈ (a, b).
Now assume that K̂(ω) = λ in some interval (a, b). Any function φ̂(ω) that is nonzero only for ω ∈ (a, b)
satisfies
 
K̂(ω) − λ φ̂(ω) = 0, ∀ω.

By taking the inverse Fourier transform we obtain an eigenfunction φ(x) of the eigenvalue λ.

3. First we use the Fourier transform to find an explicit representation of u = (T − λI)−1 f .

u = (T − λI)−1 f (T − λI)u = f
Z ∞
K(x − y)u(y) dy − λu = f
−∞

2π K̂ û − λû = fˆ
fˆ
û =
2π K̂ − λ
1 fˆ
û = −
λ 1 − 2π K̂/λ

For |λ| > |2π K̂| we can expand the denominator in a geometric series.

∞
!n
1 ˆ X 2π K̂
û = − f
λ n=0 λ
∞ Z ∞
1X 1
u=− Kn (x − y)f (y) dy
λ n=0 λn −∞

1953
 Here Kn is the nth iterated kernel. Now we form the Neumann series expansion.

u = (T − λI)−1 f
 −1
1 1
=− I− T f
λ λ
∞
1X 1 n
=− T f
λ n=0 λn
∞
1X 1 n
=− T f
λ n=0 λn
∞ Z ∞
1X 1
=− Kn (x − y)f (y) dy
λ n=0 λn −∞

The Neumann series is the same as the series we derived with the Fourier transform.

Solution 49.22
We seek a transformation T such that

(L − λI)T f = f.

We denote u = T f to obtain a boundary value problem,

u00 − λu = f, u(−1) = u(1) = 0.

This problem has a unique solution if and only if the homogeneous adjoint problem has only the trivial solution.

u00 − λu = 0, u(−1) = u(1) = 0.

This homogeneous problem has the eigenvalues and eigenfunctions,

 nπ 2  nπ 
λn = − , un = sin (x + 1) , n ∈ N.
2 2

1954
The inhomogeneous problem has the unique solution

Z 1
u(x) = G(x, ξ; λ)f (ξ) dξ
−1

where
√ √
sin( −λ(x< +1)) sin( −λ(1−x> ))



− √ √
−λ sin(2 −λ)
, λ < 0,

G(x, ξ; λ) = − 12 (x< + 1)(1 − x> ), λ = 0,
√ √
sinh( λ(x< +1)) sinh( λ(1−x> ))


−
 √ √ , λ > 0,
λ sinh(2 λ)

for λ 6= −(nπ/2)2 , n ∈ N. We set

Z 1
Tf = G(x, ξ; λ)f (ξ) dξ
−1

and note that since the kernel is continuous this is a bounded linear transformation. If f ∈ W , then

Z 1
(L − λI)T f = (L − λI) G(x, ξ; λ)f (ξ) dξ
−1
Z 1
= (L − λI)[G(x, ξ; λ)]f (ξ) dξ
−1
Z 1
= δ(x − ξ)f (ξ) dξ
−1
= f (x).

1955
If f ∈ U then
Z 1
G(x, ξ; λ) f 00 (ξ) − λf (ξ) dξ


T (L − λI)f =
−1
Z 1 Z 1
0 1 0 0
= [G(x, ξ; λ)f (ξ)]−1 − G (x, ξ; λ)f (ξ) dξ − λ G(x, ξ; λ)f (ξ) dξ
−1 −1
Z 1 Z 1
0 1 00
= [−G (x, ξ; λ)f (ξ)]−1 + G (x, ξ; λ)f (ξ) dξ − λ G(x, ξ; λ)f (ξ) dξ
−1 −1
Z 1
G00 (x, ξ; λ) − λG(x, ξ; λ) f (ξ) dξ


=
−1
Z 1
= δ(x − ξ)f (ξ) dξ
−1
= f (x).
L has the point spectrum λn = −(nπ/2)2 , n ∈ N.

Solution 49.23
1. We see that the solution is of the form φ(x) = a + x + bx2 for some constants a and b. We substitute this
into the integral equation.
Z 1
x2 y − y 2 φ(y) dy


φ(x) = x + λ
Z 01
a + x + bx2 = x + λ x2 y − y 2 (a + x + bx2 ) dy


0
λ
a + bx2 = −(15 + 20a + 12b) + (20 + 30a + 15b)x2


60
By equating the coefficients of x and x2 we solve for a and b.
0

λ(λ + 60) 5λ(λ − 60)

a=− 2
, b=−
4(λ + 5λ + 60) 6(λ2 + 5λ + 60)

1956
 Thus the solution of the integral equation is
 
λ 5(λ − 24) 2 λ + 60
φ(x) = x − 2 x + .
λ + 5λ + 60 6 4

2. For x < 1 the integral equation reduces to

φ(x) = x.

For x ≥ 1 the integral equation becomes,

Z 1
φ(x) = x + λ sin(xy)φ(y) dy.
0

We could solve this problem by writing down the Neumann series. Instead we will use an eigenfunction
expansion. Let {λn } and {φn } be the eigenvalues and orthonormal eigenfunctions of
Z 1
φ(x) = λ sin(xy)φ(y) dy.
0

We expand φ(x) and x in terms of the eigenfunctions.

∞
X
φ(x) = an φn (x)
n=1
X∞
x= bn φn (x), bn = hx, φn (x)i
n=1

1957
We determine the coefficients an by substituting the series expansions into the Fredholm equation and
equating coefficients of the eigenfunctions.
Z 1
φ(x) = x + λ sin(xy)φ(y) dy
0
∞
X ∞
X Z 1 ∞
X
an φn (x) = bn φn (x) + λ sin(xy) an φn (y) dy
n=1 n=1 0 n=1
∞ ∞ ∞
X X X 1
an φn (x) = bn φn (x) + λ an φn (x)
n=1 n=1 n=1
λn
 
λ
an 1 − = bn
λn

If λ is not an eigenvalue then we can solve for the an to obtain the unique solution.
bn λn b n λbn
an = = = bn +
1 − λ/λn λn − λ λn − λ

∞
X λbn
φ(x) = x + φn (x), for x ≥ 1.
λ −λ
n=1 n

If λ = λm , and hx, φm i = 0 then there is the one parameter family of solutions,

∞
X λbn
φ(x) = x + cφm (x) + φn (x), for x ≥ 1.
n=1
λn − λ
n6=m

If λ = λm , and hx, φm i 6= 0 then there is no solution.

1958
Solution 49.24
1.
Kx = L1 L2 x = λx

L1 L2 (L1 x) = L1 (L1 l2 − I)x

= L1 (λx − x)
= (λ − 1)(L1 x)

L1 L2 (L2 x) = (L2 L1 + I)L2 x

= L2 λx + L2 x
= (λ + 1)(L2 x)

2.
     
d t d t d t d t
L1 L2 − L2 L1 = + − + − − + +
dt 2 dt 2 dt 2 dt 2
d2 t2 d2 t2
 
t d 1 t d t d 1 t d
=− 2 + + I− + I− − 2− − I+ + I
dt 2 dt 2 2 dt 4 dt 2 dt 2 2 dt 4
=I

d2 1 t2 1
L1 L2 = − 2 + I + I = K + I
dt 2 4 2
2 2
We note that e−t /4 is an eigenfunction corresponding to the eigenvalue λ = 1/2. Since L1 e−t /4 = 0 the
2 2
result of this problem does not produce any negative eigenvalues. However, Ln2 e−t /4 is the product of e−t /4
and a polynomial of degree n in t. Since this function is square integrable it is and eigenfunction. Thus we
have the eigenvalues and eigenfunctions,
 n−1
1 t d 2
λn = n − , φn = − e−t /4 , for n ∈ N.
2 2 dt

1959
Solution 49.25
Since λ1 is in the residual spectrum of T , there exists a nonzero y such that

h(T − λ1 I)x, yi = 0

for all x. Now we apply the definition of the adjoint.

hx, (T − λ1 I)∗ yi = 0, ∀x
hx, (T ∗ − λ1 I)yi = 0, ∀x
(T ∗ − λ1 I)y = 0

y is an eigenfunction of T ∗ corresponding to the eigenvalue λ1 .

Solution 49.26
1.
Z 1
00
u (t) + sin(k(s − t))u(s) ds = f (t), u(0) = u0 (0) = 0
0
Z 1 Z 1
00
u (t) + cos(kt) sin(ks)u(s) ds − sin(kt) cos(ks)u(s) ds = f (t)
0 0
00
u (t) + c1 cos(kt) − c2 sin(kt) = f (t)
u00 (t) = f (t) − c1 cos(kt) + c2 sin(kt)

The solution of

u00 (t) = g(t), u(0) = u0 (0) = 0

using Green functions is

Z t
u(t) = (t − τ )g(τ ) dτ.
0

1960
 Thus the solution of our problem has the form,
Z t Z t Z t
u(t) = (t − τ )f (τ ) dτ − c1 (t − τ ) cos(kτ ) dτ + c2 (t − τ ) sin(kτ ) dτ
0 0 0
t
1 − cos(kt) kt − sin(kt)
Z
u(t) = (t − τ )f (τ ) dτ − c1 + c 2
0 k2 k2

We could determine the constants by multiplying in turn by cos(kt) and sin(kt) and integrating from 0 to
1. This would yields a set of two linear equations for c1 and c2 .
2.
Z ∞
πX
sin nx sin ns
u(x) = λ u(s) ds
0 n=1
n

We expand u(x) in a sine series.

∞ ∞
! ∞
!
Z π
X X sin nx sin ns X
an sin nx = λ am sin ms ds
n=1 0 n=1
n m=1
∞ ∞ ∞ Z
X X sin nx X π
an sin nx = λ am sin ns sin ms ds
n=1 n=1
n m=1 0
∞ ∞ ∞
X X sin nx X π
an sin nx = λ am δmn
n=1 n=1
n m=1 2
∞ ∞
X π X sin nx
an sin nx = λ an
n=1
2 n=1 n

The eigenvalues and eigenfunctions are

2n
λn = , un = sin nx, n ∈ N.
π

1961
3.
2π
1 − r2
Z
1
φ(θ) = λ φ(t) dt, |r| < 1
0 2π 1 − 2r cos(θ − t) + r2

We use Poisson’s formula.

φ(θ) = λu(r, θ),

where u(r, θ) is harmonic in the unit disk and satisfies, u(1, θ) = φ(θ). For a solution we need λ = 1 and
that u(r, θ) is independent of r. In this case u(θ) satisfies

u00 (θ) = 0, u(θ) = φ(θ).

The solution is φ(θ) = c1 + c2 θ. There is only one eigenvalue and corresponding eigenfunction,

λ = 1, φ = c1 + c2 θ.

4.
Z π
φ(x) = λ cosn (x − ξ)φ(ξ) dξ
−π

We expand the kernel in a Fourier series. We could find the expansion by integrating to find the Fourier
coefficients, but it is easier to expand cosn (x) directly.
 n
n 1 ix −ix
cos (x) = ( e + e )
2
        
1 n inx n i(n−2)x n −i(n−2)x n −inx
= n e + e + ··· + e + e
2 0 1 n−1 n

1962
If n is odd,

"   
1 n −inx n
cosn (x) = n inx
(e + e )+ ( ei(n−2)x + e−i(n−2)x ) + · · ·
2 0 1
  #
n
+ ( eix + e−ix )
(n − 1)/2
      
1 n n n
= n 2 cos(nx) + 2 cos((n − 2)x) + · · · + 2 cos(x)
2 0 1 (n − 1)/2
(n−1)/2  
1 X n
= n−1 cos((n − 2m)x)
2 m=0
m
n  
1 X n
= n−1 cos(kx).
2 k=1
(n − k)/2
odd k

1963
If n is even,
"   
1 n −inx n
n
cos (x) = n inx
(e + e )+ ( ei(n−2)x + e−i(n−2)x ) + · · ·
2 0 1
   #
n i2x −i2x n
+ (e + e )+
n/2 − 1 n/2
       
1 n n n n
= n 2 cos(nx) + 2 cos((n − 2)x) + · · · + 2 cos(2x) +
2 0 1 n/2 − 1 n/2
  (n−2)/2  
1 n 1 X n
= n + n−1 cos((n − 2m)x)
2 n/2 2 m=0
m
  n  
1 n 1 X n
= n + n−1 cos(kx).
2 n/2 2 k=2
(n − k)/2
even k

We will denote,
n
n a0 X
cos (x − ξ) = ak cos(k(x − ξ)),
2 k=1
where
1 + (−1)n−k 1
 
n
ak = .
2 2n−1 (n − k)/2
We substitute this into the integral equation.
Z π n
!
a0 X
φ(x) = λ ak cos(k(x − ξ)) φ(ξ) dξ
−π 2 k=1
n
a0 π
Z X  Z π Z π 
φ(x) = λ φ(ξ) dξ + λ ak cos(kx) cos(kξ)φ(ξ) dξ + sin(kx) sin(kξ)φ(ξ) dξ
2 −π k=1 −π −π

1964
 For even n, substituting φ(x) = 1 yields λ = πa1 0 . For n and m both even or odd, substituting φ(x) = cos(mx)
or φ(x) = sin(mx) yields λ = πa1m . For even n we have the eigenvalues and eigenvectors,

1
λ0 = , φ0 = 1,
πa0
1
λm = , φ(1)
m = cos(2mx), φ(2)
m = sin(2mx), m = 1, 2, . . . , n/2.
πa2m

For odd n we have the eigenvalues and eigenvectors,

1
λm = , φ(1)
m = cos((2m − 1)x), φ(2)
m = sin((2m − 1)x), m = 1, 2, . . . , (n + 1)/2.
πa2m−1

Solution 49.27
1. First we shift the range of integration to rewrite the kernel.
Z 2π
2π 2 − 6π|x − s| + 3(x − s)2 φ(s) ds


φ(x) = λ
Z0 −x+2π
2π 2 − 6π|y| + 3y 2 φ(x + y) dy


φ(x) = λ
−x

We expand the kernel in a Fourier series.

∞
X
2 2
K(y) = 2π − 6π|y| + 3y = cn einy
n=−∞

(
−x+2π 6
n 6= 0,
Z
1 ,
cn = K(y) e−iny dy = n2
2π −x 0, n=0

1965
 ∞ ∞
X 6 iny X 12
K(y) = e = cos(ny)
n=−∞
n2 n=1
n2
n6=0

∞ ∞
X 12 X 12

K(x, s) = cos(n(x − s)) = cos(nx) cos(nx) + sin(nx) sin(ns)
n=1
n2 n=1
n2

Now we substitute the Fourier series expression for the kernel into the eigenvalue problem.
∞
Z 2π X !
1

φ(x) = 12λ cos(nx) cos(ns) + sin(nx) sin(ns) φ(s) ds
0 n=1
n2

From this we obtain the eigenvalues and eigenfunctions,

n2 1 1
λn = , φ(1)
n = √ cos(nx), φ(2)
n = √ sin(nx), n ∈ N.
12π π π

2. The set of eigenfunctions do not form a complete set. Only those functions with a vanishing integral on
[0, 2π] can be represented. We consider the equation
Z 2π
K(x, s)φ(s) ds = 0
0
∞
Z 2π X !
12

2
cos(nx) cos(ns) + sin(nx) sin(ns) φ(s) ds = 0
0 n=1
n

This has the solutions φ = const. The set of eigenfunctions

1 1 1
φ0 = √ , φ(1)
n = √ cos(nx), φ(2)
n = √ sin(nx), n ∈ N,
2π π π

1966
 is a complete set. We can also write the eigenfunctions as

1
φn = √ einx , n ∈ Z.
2π

3. We consider the problem

u − λT u = f.

For λ 6= λ, (λ not an eigenvalue), we can obtain a unique solution for u.

Z 2π
u(x) = f (x) + Γ(x, s, λ)f (s) ds
0

Since K(x, s) is self-adjoint and L2 (0, 2π), we have

∞
X φn (x)φn (s)
Γ(x, s, λ) = λ
n=−∞
λn − λ
n6=0
∞
X 1
2π
einx e−ins
=λ n2
n=−∞ 12π
−λ
n6=0
∞
X ein(x−s)
= 6λ
n=−∞
n2 − 12πλ
n6=0

∞
X cos(n(x − s))
Γ(x, s, λ) = 12λ
n=1
n2 − 12πλ

1967
Solution 49.28
First assume that λ is an eigenvalue of T , T φ = λφ.
n
X
p(T )φ = an T n φ
k=0
n
X
= a n λn φ
k=0
= p(λ)φ

p(λ) is an eigenvalue of p(T ).

Now assume that µ is an eigenvalues of p(T ), p(T )φ = µφ. We assume that T has a complete, orthonormal
set of eigenfunctions, {φn } corresponding to the set of eigenvalues {λn }. We expand φ in these eigenfunctions.

p(T )φ = µφ
X X
p(T ) cn φn = µ cn φn
X X
cn p(λn )φn = cn µφn
p(λn ) = µ, ∀n such that cn 6= 0

Thus all eigenvalues of p(T ) are of the form p(λ) with λ an eigenvalue of T .

Solution 49.29
The Fourier cosine transform is defined,
Z ∞
1
fˆ(ω) = f (x) cos(ωx) dx,
π 0
Z ∞
f (x) = 2 fˆ(ω) cos(ωx) dω.
0

1968
We can write the integral equation in terms of the Fourier cosine transform.
Z ∞
φ(x) = f (x) + λ cos(2xs)φ(s) ds
0

φ(x) = f (x) + λπ φ̂(2x) (49.8)

1
We multiply the integral equation by π
cos(2xs) and integrate.
Z ∞ Z ∞ Z ∞
1 1
cos(2xs)φ(x) dx = cos(2xs)f (x) dx + λ cos(2xs)φ̂(2x) dx
π 0 π 0 0
λ ∞
Z
ˆ
φ̂(2s) = f (2s) + cos(xs)φ̂(x) dx
2 0
λ
φ̂(2s) = fˆ(2s) + φ(s)
4

4 4
φ(x) = − fˆ(2x) + φ̂(2x) (49.9)
λ λ

We eliminate φ̂ between (49.8) and (49.9).

πλ2
 
1− φ(x) = f (x) + λπ fˆ(2x)
4
R∞
f (x) + λ 0 f (s) cos(2xs) ds
φ(x) =
1 − πλ2 /4

1969
Solution 49.30

Z Z
vLu dx dy = v(uxx + uyy + aux + buy + cu) dx dy
D ZD
= (v∇2 u + avux + bvuy + cuv) dx dy
ZD Z
2
= (v∇u − u∇v) · n ds
(u∇ v + avux + bvuy + cuv) dx dy +
D C
Z Z   Z  
2 ∂x ∂y ∂u ∂v
= (u∇ v − auvx − buvy − uvax − uvby + cuv) dx dy + auv + buv ds + v −u
D C ∂n ∂n C ∂n ∂n

Thus we see that

Z Z
∗
(vLu − uL v) dx dy = H(u, v) ds,
D C

where

L∗ v = vxx + vyy − avx − bvy + (c − ax − by )v

and
 
∂u ∂v ∂x ∂y
H(u, v) = v −u + auv + buv .
∂n ∂n ∂n ∂n

Let G be the harmonic Green function, which satisfies,

∆G = δ in D, G = 0 on C.

1970
Let u satisfy Lu = 0.

Z Z
∗
(GLu − uL G) dx dy = H(u, G) ds
D Z Z C
∗
− uL G dx dy = H(u, G) ds
Z D Z C Z
∗
− u∆G dx dy − u(L − ∆)G dx dy = H(u, G) ds
Z D D Z C Z
∗
− uδ(x − ξ)δ(y − η) dx dy − u(L − ∆)G dx dy = H(u, G) ds
D Z D Z C
∗
−u(ξ, η) − u(L − ∆)G dx dy = H(u, G) ds
D C

We expand the operators to obtain the first form.

Z Z  
∂u ∂G ∂x ∂y
u+ u(−aGx − bGy + (c − ax − by )G) dx dy = − G −u + auG + buG ds
D C ∂n ∂n ∂n ∂n
Z Z
∂G
u + ((c − ax − by )G − aGx − bGy )u dx dy = u ds
D C ∂n
Z
u + ((c − ax − by )G − aGx − bGy )u dx dy = U
D

Here U is the harmonic function that satisfies U = f on C.

1971
 We use integration by parts to obtain the second form.
Z
u + (cuG − ax uG − by uG − auGx − buGy ) dx dy = U
D
Z Z  
∂y ∂x
u + (cuG − ax uG − by uG + (au)x G + (bu)y G) dx dy − auG + buG ds = U
D C ∂n ∂n
Z
u + (cuG − ax uG − by uG + ax uG + aux G + by uG + buy G) dx dy = U
D
Z
u+ (aux + buy + cu)G dx dy = U
D

Solution 49.31
1. First we differentiate to obtain a differential equation.
Z 1 Z x Z 1 
s x
φ(x) = λ min(x, s)φ(s) ds = λ e φ(s) ds + e φ(s) ds
0 0 x
 Z 1  Z 1
0
φ (x) = λ xφ(x) + φ(s) ds − xφ(x) = λ φ(s) ds
x x
00
φ (x) = −λφ(x)
We note that that φ(x) satisfies the constraints,
Z 1
φ(0) = λ 0 · φ(s) ds = 0,
0
Z 1
0
φ (1) = λ φ(s) ds = 0.
1

Thus we have the problem,

φ00 + λφ = 0, φ(0) = φ0 (1) = 0.

1972
 The general solution of the differential equation is

a + bx for λ = 0

 √  √ 
φ(x) = a cos λx + b sin λx for λ > 0

 √
√

a cosh −λx + b sinh −λx for λ < 0

We see that for λ = 0 and λ < 0 only the trivial solution satisfies the homogeneous boundary conditions.
For positive λ the left boundary condition demands that a = 0. The right boundary condition is then
√ √ 
b λ cos λ =0

The eigenvalues and eigenfunctions are

 2  
(2n − 1)π (2n − 1)π
λn = , φn (x) = sin x , n∈N
2 2

2. First we differentiate the integral equation.

Z x Z 1 
s x
φ(x) = λ e φ(s) ds + e φ(s) ds
0 x
 Z 1 
0 x x x
φ (x) = λ e φ(x) + e φ(s) ds − e φ(x)
x
Z 1
x
= λe φ(s) ds
x
 Z 1 
00 x x
φ (x) = λ e φ(s) ds − e φ(x)
x

φ(x) satisfies the differential equation

φ00 − φ0 + λ ex φ = 0.

1973
We note the boundary conditions,

φ(0) − φ0 (0) = 0, φ0 (1) = 0.

In self-adjoint form, the problem is

0
e−x φ0 + λφ = 0, φ(0) − φ0 (0) = 0, φ0 (1) = 0.

The Rayleigh quotient is

1 R1
[− e−x φφ0 ]0 + 0 e−x (φ0 )2 dx
λ= R1
0
φ2 dx
1
φ(0)φ0 (0) + 0 e−x (φ0 )2 dx
R
= R1
0
φ2 dx
R1
(φ(0))2 + 0 e−x (φ0 )2 dx
= R1
0
φ2 dx

Thus we see that there are only positive eigenvalues. The differential equation has the general solution
  √   √ 
φ(x) = ex/2 aJ1 2 λ ex/2 + bY1 2 λ ex/2

We define the functions,

 √   √ 
u(x; λ) = ex/2 J1 2 λ ex/2 , v(x; λ) = ex/2 Y1 2 λ ex/2 .

We write the solution to automatically satisfy the right boundary condition, φ0 (1) = 0,

φ(x) = v 0 (1; λ)u(x; λ) − u0 (1; λ)v(x; λ).

1974
 We determine the eigenvalues from the left boundary condition, φ(0) − φ0 (0) = 0. The first few are
λ1 ≈ 0.678298
λ2 ≈ 7.27931
λ3 ≈ 24.9302
λ4 ≈ 54.2593
λ5 ≈ 95.3057
The eigenfunctions are,
φn (x) = v 0 (1; λn )u(x; λn ) − u0 (1; λn )v(x; λn ).

Solution 49.32
1. First note that
sin(kx) sin(lx) = sign (kl) sin(ax) sin(bx)
where
a = max(|k|, |l|), b = min(|k|, |l|).
Consider the analytic function,
ei(a−b)x − ei(a+b)
= sin(ax) sin(bx) − i cos(ax) sin(bx).
2
Z ∞ Z ∞
sin(kx) sin(lx) sin(ax) sin(bx)
− 2 2
dx = sign (kl) − dx
−∞ x −z −∞ x2 − z 2
Z ∞ 
1 sin(ax) sin(bx) sin(ax) sin(bx)
= sign (kl) − − dx
2z −∞ x−z x+z
1
= −π sign (kl) (− cos(az) sin(bz) + cos(−az) sin(−bz))
2z

1975
 Z ∞
sin(kx) sin(lx) π
− 2 2
dx = sign (kl) cos(az) sin(bz)
−∞ x −z z

2. Consider the analytic function,

ei|p|x − ei|q|x cos(|p|x) − cos(|q|x) + i(sin(|p|x) − sin(|q|x))
= .
x x
Z ∞ Z ∞
cos(px) − cos(qx) cos(|p|x) − cos(|q|x)
− 2
dx = − dx
−∞ x −∞ x2
sin(|p|x) − sin(|q|x)
= −π lim
x→0 x

Z ∞
cos(px) − cos(qx)
− dx = π(|q| − |p|)
−∞ x2

3. We use the analytic function,

i(x − ia)(x − ib) eix −(x2 − ab) sin x + (a + b)x cos x + i((x2 − ab) cos x + (a + b)x sin x)
=
(x2 + a2 )(x2 + b2 ) (x2 + a2 )(x2 + b2 )

Z ∞
−(x2 − ab) sin x + (a + b)x cos x (x2 − ab) cos x + (a + b)x sin x
− = −π lim
−∞ x(x2 + a2 )(x2 + b2 ) x→0 (x2 + a2 )(x2 + b2 )
−ab
= −π 2 2
ab

Z ∞
−(x2 − ab) sin x + (a + b)x cos x π
− 2 2 2 2
=
−∞ (x + a )(x + b ) ab

1976
Solution 49.33
We consider the function

G(z) = (1 − z 2 )1/2 + iz log(1 + z).

For (1 − z 2 )1/2 = (1 − z)1/2 (1 + z)1/2 we choose the angles,

−π < arg(1 − z) < π, 0 < arg(1 + z) < 2π,

so that there is a branch cut on the interval (−1, 1). With this choice of branch, G(z) vanishes at infinity. For
the logarithm we choose the principal branch,

−π < arg(1 + z) < π.

For t ∈ (−1, 1),

√ 
G+ (t) = 1 − t2 + it log(1 + t),
 √ 
G− (t) = − 1 − t2 + it log(1 + t),

√
G+ (t) − G− (t) = 2 1 − t2 log(1 + t),
1 +
G (t) + G− (t) = it log(1 + t).


2
For t ∈ (−∞, −1),
√ 
G+ (t) = i 1 − t2 + t (log(−t − 1) + iπ) ,
 √ 
G− (t) = i − 1 − t2 + t (log(−t − 1) − iπ) ,

√ 
+ − 2
G (t) − G (t) = −2π t −1+t .

1977
For x ∈ (−1, 1) we have
1 +
G (x) + G− (x)


G(x) =
2
= ix log(1 + x)
√ Z 1 √
2 2 1 − t2 log(1 + t)
−1 −2π( t − 1 + t)
Z
1 1
= ∞ dt + dt
i2π − t−x i2π −1 t−x
From this we have
Z 1√
1 − t2 log(1 + t)
dt
−1 t−x
∞
√
t2 − 1
t−
Z
= −πx log(1 + x) + π dt
1 t+x
 π√ √ 
= π x log(1 + x) − 1 + 1 − x2 − 1 − x2 arcsin(x) + x log(2) + x log(1 + x)
2
√
Z 1
1 − t2 log(1 + t)  √ π 
dt = π x log x − 1 + 1 − x2 − arcsin(x)
−1 t−x 2

Solution 49.34
Let F (z) denote the value of the integral.
Z
1 f (t) dt
F (z) = −
iπ C t − z
From the Plemelj formula we have,
Z
+ − 1 f (t)
F (t0 ) + F (t0 ) = − dt,
iπ C t − t0
f (t0 ) = F + (t0 ) − F − (t0 ).

1978
With W (z) defined as above, we have

W + (t0 ) + W − (t0 ) = F + (t0 ) − F − (t0 ) = f (t0 ),

and also

W + (t) − W − (t)
Z
+ − 1
W (t0 ) + W (t0 ) = − dt
iπ C t − t0
F + (t) + F − (t)
Z
1
= − dt
iπ C t − t0
Z
1 g(t)
= − dt.
iπ C t − t0

Thus the solution of the integral equation is

Z
1 g(t)
f (t0 ) = − dt.
iπ C t − t0

1979
Solution 49.35
(i)

 γ
−1 τ −β
G(τ ) = (τ − β)
τ −α
 γ
+ −1 ζ −β
G (ζ) = (ζ − β)
ζ −α
− −i2πγ +
G (ζ) = e G (ζ)
 γ
+ − −i2πγ −1 ζ −β
G (ζ) − G (ζ) = (1 − e )(ζ − β)
ζ −α
 γ
+ − −i2πγ −1 ζ −β
G (ζ) + G (ζ) = (1 + e )(ζ − β)
ζ −α
−i2πγ
(1 − e
Z
1 ) dτ
G+ (ζ) + G− (ζ) = −
iπ C (τ − β) (τ − α)γ (τ − ζ)
1−γ

(ζ − β)γ−1
Z
1 dτ
− = −i cot(πγ)
iπ C (τ − β)1−γ (τ − α)γ (τ − ζ) (ζ − α)γ

(ii) Consider the branch of

 γ
z−β
z−α

1980
that tends to unity as z → ∞. We find a series expansion of this function about infinity.

 γ  γ 
z−β β α −γ
= 1− 1−
z−α z z
∞    j ! X ∞    !
X γ β −γ α k
= (−1)j (−1)k
j=0
j z k=0
k z
∞ j    !
X X γ −γ
= (−1)j β j−k αk z −j
j=0 k=0
j − k k

Define the polynomial

j
n    !
X X
j γ −γ j−k k
Q(z) = (−1) β α z n−j .
j=0 k=0
j−k k

Then the function

 γ
z−β
G(z) = z n − Q(z)
z−α

1981
 vanishes at infinity.
 γ
+ ζ −β
G (ζ) = ζ n − Q(ζ)
ζ −α
 γ
− −i2πγ ζ −β
G (ζ) = e ζ n − Q(ζ)
ζ −α
 γ
− ζ −β
+
ζ n 1 − e−i2πγ


G (ζ) − G (ζ) =
ζ −α
 γ
− ζ −β
+
ζ n 1 + e−i2πγ − 2Q(ζ)


G (ζ) + G (ζ) =
ζ −α
Z  γ  γ
1 τ −β −i2πγ 1 ζ −β
n
ζ n 1 + e−i2πγ − 2Q(ζ)



− τ 1− e dτ =
iπ C τ − α τ −ζ ζ −α
Z  γ n
 γ
1 τ −β τ ζ −β
− dτ = −i cot(πγ) ζ n − (1 − i cot(πγ))Q(ζ)
iπ C τ − α τ −ζ ζ −α
γ γ
τn
Z   
1 τ −β ζ −β n
− dτ = −i cot(πγ) ζ − Q(ζ) − Q(ζ)
iπ C τ − α τ −ζ ζ −α

Solution 49.36

Z 1 Z 1 Z 1
φ(y) 1 φ(y) 1 φ(y)
− dy = − dy − − dy
−1 y 2 − x2 2x −1 y−x 2x −1 y + x
Z 1 Z 1
1 φ(y) 1 φ(−y)
= − dy + − dy
2x −1 y−x 2x −1 y − x
Z 1
1 φ(y) + φ(−y)
= − dy
2x −1 y−x

1982
 Z 1
1 φ(y) + φ(−y)
− dy = f (x)
2x −1 y−x
1 1 φ(y) + φ(−y)
Z
2x
− dy = f (x)
iπ −1 y−x iπ
Z 1
1 2y p 1 k
φ(x) + φ(−x) = √ − f (y) 1 − y 2 dy + √
iπ 1 − x2 −1 iπ y−x 1 − x2
Z 1 p
1 2yf (y) 1 − y 2 k
φ(x) + φ(−x) = − √ − dy + √
π 2 1 − x2 −1 y−x 1 − x2
Z 1 p
1 yf (y) 1 − y 2 k
φ(x) = − √ − dy + √ + g(x)
2
π 1 − x −1 2 y−x 1 − x2

Here k is an arbitrary constant and g(x) is an arbitrary odd function.

Solution 49.37
We define
Z 1
1 f (t)
F (z) = − dt.
i2π 0 t − z
The Plemelj formulas and the integral equation give us,

F + (x) − F − (x) = f (x)

F + (x) + F − (x) = λf (x).

We solve for F + and F − .

F + (x) = (λ + 1)f (x)

F − (x) = (λ − 1)f (x)

1983
By writing

F + (x) λ+1
−
=
F (x) λ−1

we seek to determine F to within a multiplicative constant.


+ − λ+1
log F (x) − log F (x) = log
λ−1
 
+ − 1+λ
log F (x) − log F (x) = log + iπ
1−λ
log F + (x) − log F − (x) = γ + iπ

We have left off the additive term of i2πn in the above equation, which will introduce factors of z k and (z − 1)m
in F (z). We will choose these factors so that F (z) has integrable algebraic singularites and vanishes at infinity.
Note that we have defined γ to be the real parameter,
 
1+λ
γ = log .
1−λ

By the discontinuity theorem,

Z 1
1 γ + iπ
log F (z) = dz
i2π 0 t − z
   
1 γ 1−z
= −i log
2 2π −z
 1/2−iγ/(2π) !
z−1
= log
z

1984
  1/2−iγ/(2π)
z−1
F (z) = z k (z − 1)m
z
 −iγ/(2π)
1 z−1
F (z) = p
z(z − 1) z
−iγ/(2π)
e±iπ(−iγ/(2π)) 1 − x

±
F (x) = p
x(1 − x) x
−iγ/(2π)
e±γ/2

± 1−x
F (x) = p
x(1 − x) x

Define
 −iγ/(2π)
1 1−x
f (x) = p .
x(1 − x) x

We apply the Plemelj formulas.

1 1 γ/2
Z

f (t)
− e − e−γ/2 dt = eγ/2 + e−γ/2 f (x)


iπ 0 t−x
Z 1
1 f (t) γ 
− dt = tanh f (x)
iπ 0 t − x 2
Thus we see that the eigenfunctions are
 −i tanh−1 (λ)/π
1 1−x
φ(x) = p
x(1 − x) x

for −1 < λ < 1.

The method used in this problem cannot be used to construct eigenfunctions for λ > 1. For this case we
cannot find an F (z) that has integrable algebraic singularities and vanishes at infinity.

1985
Solution 49.38

1 1 f (t)
Z
i
− dt = − f (x)
iπ 0 t − x tan(x)
We define the function,
Z 1
1 f (t)
F (z) = − dt.
i2π 0 t − z
The Plemelj formula are,
F + (x) − F − (x) = f (x)
i
F + (x) + F − (x) = − f (x).
tan(x)
We solve for F + and F − .
 
± 1 i
F (x) = ±1 − f (x)
2 tan(x)
From this we see
F + (x) 1 − i/ tan(x)
−
= = ei2x .
F (x) −1 − i/ tan(x)
We seek to determine F (z) up to a multiplicative constant. Taking the logarithm of this equation yields
log F + (x) − log F − (x) = i2x + i2πn.
The i2πn term will give us the factors (z − 1)k and z m in the solution for F (z). We will choose the integers k and
m so that F (z) has only algebraic singularities and vanishes at infinity. We drop the i2πn term for now.
Z 1
1 i2t
log F (z) = dt
i2π 0 t − z
   z/π
1 z 1−z 1/π z−1
log F (z) = + log F (z) = e
π π −z z

1986
We replace e1/π by a multiplicative constant and multiply by (z − 1)1 to give F (z) the desired properties.
c
F (z) =
(z − 1)1−z/π z z/π

We evaluate F (z) above and below the branch cut.

c c e±ix
F ± (x) = =
e±(iπ−ix) (1 − x)1−x/π xx/π (1 − x)1−x/π xx/π

Finally we use the Plemelj formulas to determine f (x).

k sin(x)
f (x) = F + (x) − F − (x) =
(1 − x)1−x/π xx/π

Solution 49.39
Consider the equation,
Z
0 f (t)
f (z) + λ dt = 1.
C t−z

Since the integral is an analytic function of z off C we know that f (z) is analytic off C. We use Cauchy’s theorem
to evaluate the integral and obtain a differential equation for f (x).
Z
0 f (t)
f (x) + λ − dt = 1
C t−x
f 0 (x) + iλπf (x) = 1
1
f (x) = + c e−iλπx
iλπ

1987
 Consider the equation,
Z
0 f (t)
f (z) + λ dt = g(z).
C t−z
Since the integral and g(z) are analytic functions inside C we know that f (z) is analytic inside C. We use Cauchy’s
theorem to evaluate the integral and obtain a differential equation for f (x).
Z
0 f (t)
f (x) + λ − dt = g(x)
C t−x
f 0 (x) + iλπf (x) = g(x)
Z x
f (x) = e−iλπ(x−ξ) g(ξ) dξ + c e−iλπx
z0

Here z0 is any point inside C.

Solution 49.40

Z  
1
− + P (t − x) f (t) dt = g(x)
C t−x
Z Z
1 f (t) 1 1
− dt = g(x) − P (t − x)f (t) dt
iπ C t − x iπ iπ C
We know that if
Z
1 f (τ )
− dτ = g(ζ)
iπ C τ − ζ
then
Z
1 g(τ )
f (ζ) = − dτ.
iπ C τ − ζ

1988
We apply this theorem to the integral equation.
Z Z Z 
1 g(t) 1 1
f (x) = − 2 − dt + 2
− P (τ − t)f (τ ) dτ dt
π C t−x π C C t−x
Z Z 
P (τ − t)
Z
1 g(t) 1
=− 2 − dt + − dt f (τ ) dτ
π C t−x π2 C C t − x
Z Z
1 g(t) 1
=− 2 − dt − P (t − x)f (t) dt
π C t−x iπ C

Now we substitute the non-analytic part of f (t) into the integral. (The analytic part integrates to zero.)
Z Z  Z 
1 g(t) 1 1 g(τ )
=− 2 − dt − P (t − x) − 2 − dτ dt
π t−x iπ C π C τ −t
ZC Z  
P (t − x)
Z
1 g(t) 1 1
=− 2 − dt − − − dt g(τ ) dτ
π t−x π2 C iπ C τ − t
ZC Z
1 g(t) 1
=− 2 − dt − P (τ − x)g(τ ) dτ
π C t−x π2 C

Z Z
1 g(t) 1
f (x) = − 2 − dt − 2 P (t − x)g(t) dt
π C t−x π C

Solution 49.41

Solution 49.42

1989
 Part VII

Nonlinear Differential Equations

1990
Chapter 50

Nonlinear Ordinary Differential Equations

1991
50.1 Exercises
Exercise 50.1
A model set of equations to describe an epidemic, in which x(t) is the number infected, y(t) is the number
susceptible, is
dx dy
= rxy − γx, = −rxy + β,
dt dt
where r > 0, β ≥ 0, γ ≥ 0. Initially x = x0 , y = y0 at t = 0. Directly from the equations, without using the
phase plane:

1. Find the solution, x(t), y(t), in the case β = γ = 0.

2. Show for the case β = 0, γ 6= 0 that x(t) first decreases or increases according as ry0 < γ or ry0 > γ. Show
that x(t) → 0 as t → ∞ in both cases. Find x as a function of y.

3. In the phase plane: Find the position of the singular point and its type when β > 0, γ > 0.

Exercise 50.2
Find the singular points and their types for the system

du
= ru + v(1 − v)(p − v), r > 0, 0 < p < 1,
dx
dv
= u,
dx
which comes from one of our nonlinear diffusion problems. Note that there is a solution with

u = α(1 − v)

for special values of α and r. Find v(x) for this special case.

1992
Exercise 50.3
Check that r = 1 is a limit cycle for
dx
= −y + x(1 − r2 )
dt
dy
= x + y(1 − r2 )
dt
(r = x2 + y 2 ), and that all solution curves spiral into it.

Exercise 50.4
Consider
ẏ = f (y) − x
ẋ = y
Introduce new coordinates, R, θ given by
x = R cos θ
1
y = √ R sin θ

and obtain the exact differential equations for R(t), θ(t). Show that R(t) continually increases with t when R 6= 0.
Show that θ(t) continually decreases when R > 1.

Exercise 50.5
One choice of the Lorenz equations is
ẋ = −10x + 10y
ẏ = Rx − y − xz
8
ż = − z + xy
3

1993
Where R is a positive parameter.
1. Invistigate the nature of the sigular point at (0, 0, 0) by finding the eigenvalues and their behavior for all
0 < R < ∞.
2. Find the other singular points when R > 1.
3. Show that the appropriate eigenvalues for these other singular points satisfy the cubic

3λ3 + 41λ2 + 8(10 + R)λ + 160(R − 1) = 0.

4. There is a special value of R, call it Rc , for which the cubic has two pure imaginary roots, ±iµ say. Find
Rc and µ; then find the third root.

Exercise 50.6
In polar coordinates (r, φ), Einstein’s equations lead to the equation

d2 v 1
+ v = 1 + v 2 , v= ,
dφ2 r
for planetary orbits. For Mercury,  = 8 × 10−8 . When  = 0 (Newtonian theory) the orbit is given by

v = 1 + A cos φ, period 2π.

Introduce θ = ωφ and use perturbation expansions for v(θ) and ω in powers of  to find the corrections proportional
to .
[A is not small;  is the small parameter].

Exercise 50.7
Consider the problem

ẍ + ω02 x + αx2 = 0, x = a, ẋ = 0 at t = 0

1994
Use expansions
x = a cos θ + a2 x2 (θ) + a3 x3 (θ) + · · · , θ = ωt
ω = ω 0 + a2 ω 2 + · · · ,
to find a periodic solution and its natural frequency ω.
Note that, with the expansions given, there are no “secular term” troubles in the determination of x2 (θ), but
x2 (θ) is needed in the subsequent determination of x3 (θ) and ω.
Show that a term aω1 in the expansion for ω would have caused trouble, so ω1 would have to be taken equal
to zero.

Exercise 50.8
Consider the linearized traffic problem
dpn (t)
= α [pn−1 (t) − pn (t)] , n ≥ 1,
dt
pn (0) = 0, n ≥ 1,
p0 (t) = aeiωt , t > 0.
(We take the imaginary part of pn (t) in the final answers.)
1. Find p1 (t) directly from the equation for n = 1 and note the behavior as t → ∞.
2. Find the generating function
∞
X
G(s, t) = pn (t)sn .
n=1

3. Deduce that
pn (t) ∼ An eiωt , as t → ∞,
and find the expression for An . Find the imaginary part of this pn (t).

1995
Exercise 50.9
1. For the equation modified with a reaction time, namely

d
pn (t + τ ) = α[pn−1 (t) − pn (t)] n ≥ 1,
dt
find a solution of the form in 1(c) by direct substitution in the equation. Again take its imaginary part.

2. Find a condition that the disturbance is stable, i.e. pn (t) remains bounded as n → ∞.

3. In the stable case show that the disturbance is wave-like and find the wave velocity.

1996
50.2 Hints
Hint 50.1

Hint 50.2

Hint 50.3

Hint 50.4

Hint 50.5

Hint 50.6

Hint 50.7

Hint 50.8

1997
Hint 50.9

1998
50.3 Solutions
Solution 50.1
1. When β = γ = 0 the equations are

dx dy
= rxy, = −rxy.
dt dt

Adding these two equations we see that

dx dy
=− .
dt dt

Integrating and applying the initial conditions x(0) = x0 and y(0) = y0 we obtain

x = x0 + y 0 − y

Substituting this into the differential equation for y,

dy
= −r(x0 + y0 − y)y
dt
dy
= −r(x0 + y0 )y + ry 2 .
dt

1999
We recognize this as a Bernoulli equation and make the substitution u = y −1 .
dy
−y −2 = r(x0 + y0 )y −1 − r
dt
du
= r(x0 + y0 )u − r
dt
d −r(x0 +y0 )t

e u = −re−r(x0 +y0 )t
dt Z t
u=er(x0 +y0 )t
−re−r(x0 +y0 )t dt + cer(x0 +y0 )t
1
u= + cer(x0 +y0 )t
x0 + y0
 −1
1 r(x0 +y0 )t
y= + ce
x0 + y 0
Applying the initial condition for y,
 −1
1
+c = y0
x0 + y 0
1 1
c= − .
y0 x0 + y0
The solution for y is then
   −1
1 1 1 r(x0 +y0 )t
y= + − e
x0 + y 0 y0 x0 + y0
Since x = x0 + y0 − y, the solution to the system of differential equations is

−1
−1
   
1 1 r(x0 +y0 )t 1 1 r(x0 +y0 )t
x = x0 + y 0 − + 1−e , y= + 1−e .
y0 x0 + y0 y 0 x0 + y 0

2000
2. For β = 0, γ 6= 0, the equation for x is

ẋ = rxy − γx.

At t = 0,

ẋ(0) = x0 (ry0 − γ).

Thus we see that if ry0 < γ, x is initially decreasing. If ry0 > γ, x is initially increasing.
Now to show that x(t) → 0 as t → ∞. First note that if the initial conditions satisfy x0 , y0 > 0 then
x(t), y(t) > 0 for all t ≥ 0 because the axes are a seqaratrix. y(t) is is a strictly decreasing function of time.
Thus we see that at some time the quantity x(ry − γ) will become negative. Since y is decreasing, this
quantity will remain negative. Thus after some time, x will become a strictly decreasing quantity. Finally
we see that regardless of the initial conditions, (as long as they are positive), x(t) → 0 as t → ∞.
Taking the ratio of the two differential equations,
dx γ
= −1 + .
dy ry
γ
x = −y + ln y + c
r
Applying the intial condition,
γ
x0 = −y0 + ln y0 + c
r
γ
c = x0 + y0 − ln y0 .
r
Thus the solution for x is
 
γ y
x = x0 + (y0 − y) + ln .
r y0

2001
3. When β > 0 and γ > 0 the system of equations is
ẋ = rxy − γx
ẏ = −rxy + β.
The equilibrium solutions occur when
x(ry − γ) = 0
β − rxy = 0.
Thus the singular point is
β γ
x= , y= .
γ r

Now to classify the point. We make the substitution u = (x − βγ ), v = (y − γr ).

   
β  γ β
u̇ = r u + v+ −γ u+
γ r γ
 
β γ 
v̇ = −r u + v+ +β
γ r

rβ
u̇ = v + ruv
γ
rβ
v̇ = −γu − v − ruv
γ
The linearized system is
rβ
u̇ = v
γ
rβ
v̇ = −γu − v
γ

2002
 Finding the eigenvalues of the linearized system,
 
λ − rβ  rβ
γ  2
rβ  = λ + λ + rβ = 0

γ λ + γ  γ


q
− rβ
γ
± ( rβ
γ
)2 − 4rβ
λ=
2
Since both eigenvalues have negative real part, we see that the singular point is asymptotically stable. A
plot of the vector field for r = γ = β = 1 is attached. We note that there appears to be a stable singular
point at x = y = 1 which corroborates the previous results.

Solution 50.2
The singular points are

u = 0, v = 0, u = 0, v = 1, u = 0, v = p.

The point u = 0, v = 0. The linearized system about u = 0, v = 0 is

du
= ru
dx
dv
= u.
dx
The eigenvalues are
 
λ − r 0  2
 −1 λ = λ − rλ = 0.
 

λ = 0, r.

2003
Since there are positive eigenvalues, this point is a source. The critical point is unstable.
The point u = 0, v = 1. Linearizing the system about u = 0, v = 1, we make the substitution w = v − 1.
du
= ru + (w + 1)(−w)(p − 1 − w)
dx
dw
=u
dx

du
= ru + (1 − p)w
dx
dw
=u
dx
 
λ − r (p − 1)
  = λ2 − rλ + p − 1 = 0
 −1 λ 

p
r± r2 − 4(p − 1)
λ=
2
Thus we see that this point is a saddle point. The critical point is unstable.
The point u = 0, v = p. Linearizing the system about u = 0, v = p, we make the substitution w = v − p.
du
= ru + (w + p)(1 − p − w)(−w)
dx
dw
=u
dx

du
= ru + p(p − 1)w
dx
dw
=u
dx

2004
  
λ − r p(1 − p)
  = λ2 − rλ + p(1 − p) = 0
 −1 λ 
p
r ± r2 − 4p(1 − p)
λ=
2
Thus we see that this point is a source. The critical point is unstable.
The solution of for special values of α and r. Differentiating u = αv(1 − v),
du
= α − 2αv.
dv
Taking the ratio of the two differential equations,
du v(1 − v)(p − v)
=r+
dv u
v(1 − v)(p − v)
=r+
αv(1 − v)
(p − v)
=r+
α
Equating these two expressions,
p v
α − 2αv = r + − .
α α
1
Equating coefficients of v, we see that α = √2 .
1 √
√ = r + 2p
2
√
Thus we have the solution u = √12 v(1 − v) when r = √12 − 2p. In this case, the differential equation for v is
dv 1
= √ v(1 − v)
dx 2
dv 1 1
−v −2 = − √ v −1 + √
dx 2 2

2005
We make the change of variablles y = v −1 .
dy 1 1
= −√ y + √
dx 2 2
√
d  √
x/ 2
 e 2
x/
e y = √
dx 2
√
Z x x/√2 √
e
y = e−x/ 2 √ dx + ce−x/ 2
2
√
y = 1 + ce−x/ 2

The solution for v is

1
v(x) = √ .
1 + ce−x/ 2

Solution 50.3
We make the change of variables
x = r cos θ
y = r sin θ.
Differentiating these expressions with respect to time,
ẋ = ṙ cos θ − rθ̇ sin θ
ẏ = ṙ sin θ + rθ̇ cos θ.
Substituting the new variables into the pair of differential equations,
ṙ cos θ − rθ̇ sin θ = −r sin θ + r cos θ(1 − r2 )
ṙ sin θ + rθ̇ cos θ = r cos θ + r sin θ(1 − r2 ).

2006
Multiplying the equations by cos θ and sin θ and taking their sum and difference yields

ṙ = r(1 − r2 )
rθ̇ = r.

We can integrate the second equation.

ṙ = r(1 − r2 )
θ = t + θ0

At this point we could note that ṙ > 0 in (0, 1) and ṙ < 0 in (1, ∞). Thus if r is not initially zero, then the
solution tends to r = 1.
Alternatively, we can solve the equation for r exactly.

ṙ = r − r3
ṙ 1
3
= 2 −1
r r

We make the change of variables u = 1/r2 .

1
− u̇ = u − 1
2
u̇ + 2u = 2
Z t
−2t
u=e 2e2t dt + ce−2t

u = 1 + ce−2t
1
r=√
1 + ce−2t
Thus we see that if r is initiall nonzero, the solution tends to 1 as t → ∞.

2007
Solution 50.4
The set of differential equations is

ẏ = f (y) − x
ẋ = y.

We make the change of variables

x = R cos θ
1
y = √ R sin θ


Differentiating x and y,

ẋ = Ṙ cos θ − Rθ̇ sin θ

1 1
ẏ = √ Ṙ sin θ + √ Rθ̇ cos θ.
 

The pair of differential equations become

√ √
 
1
Ṙ sin θ + Rθ̇ cos θ = f √ R sin θ − R cos θ

1
Ṙ cos θ − Rθ̇ sin θ = √ R sin θ.


 
1 1 1
Ṙ sin θ + Rθ̇ cos θ = − √ R cos θ √ f √ R sin θ
  
1
Ṙ cos θ − Rθ̇ sin θ = √ R sin θ.


2008
Multiplying by cos θ and sin θ and taking the sum and difference of these differential equations yields
 
1 1
Ṙ = √ sin θf √ R sin θ
 
 
1 1 1
Rθ̇ = − √ R + √ cos θf √ R sin θ .
  
Dividing by R in the second equation,
 
1 1
Ṙ = √ sin θf √ R sin θ
 
 
1 1 cos θ 1
θ̇ = − √ + √ f √ R sin θ .
  R 
We make the assumptions that 0 <  < 1 and that f (y) is an odd function that is nonnegative for positive y
and satisfies |f (y)| ≤ 1 for all y.
Since sin θ is odd,
 
1
sin θf √ R sin θ

is nonnegative. Thus R(t) continually increases with t when R 6= 0.
If R > 1 then
     
 cos θ 1   1 

 R f √ R sin θ  ≤  f √ R sin θ 
    
≤ 1.

Thus the value of θ̇,

 
1 1 cos θ 1
−√ + √ f √ R sin θ ,
  R 
is always nonpositive. Thus θ(t) continually decreases with t.

2009
Solution 50.5
1. Linearizing the Lorentz equations about (0, 0, 0) yields
    
ẋ −10 10 0 x
ẏ  =  R −1 0  y 
ż 0 0 −8/3 z

The eigenvalues of the matrix are

8
λ1 = − ,
3 √
−11 − 81 + 40R
λ2 =
√2
−11 + 81 + 40R
λ3 = .
2

There are three cases for the eigenvalues of the linearized system.

R < 1. There are three negative, real eigenvalues. In the linearized and also the nonlinear system, the
origin is a stable, sink.
R = 1. There are two negative, real eigenvalues and one zero eigenvalue. In the linearized system the
origin is stable and has a center manifold plane. The linearized system does not tell us if the nonlinear
system is stable or unstable.
R > 1. There are two negative, real eigenvalues, and one positive, real eigenvalue. The origin is a saddle
point.

2. The other singular points when R > 1 are

r r !
8 8
± (R − 1), ± (R − 1), R − 1 .
3 3

2010
3. Linearizing about the point
r r !
8 8
(R − 1), (R − 1), R − 1
3 3

yields
 
  −10 10 0  
Ẋ q
 X
1 −1 − 83 (R − 1)  Y 

 Ẏ  = 
q q 
Ż 8
(R − 1) 8
(R − 1) − 8 Z
3 3 3

The characteristic polynomial of the matrix is

41 2 8(10 + R) 160
λ3 + λ + λ+ (R − 1).
3 3 3
Thus the eigenvalues of the matrix satisfy the polynomial,

3λ3 + 41λ2 + 8(10 + R)λ + 160(R − 1) = 0.

Linearizing about the point

r r !
8 8
− (R − 1), − (R − 1), R − 1
3 3

yields
 
  −10 10 0  
Ẋ q X
8
1 −1 (R − 1)
 
 Y
 Ẏ  =  3   
 q q
Ż − 83 (R − 1) − 83 (R − 1) − 83 Z

2011
 The characteristic polynomial of the matrix is
41 2 8(10 + R) 160
λ3 + λ + λ+ (R − 1).
3 3 3
Thus the eigenvalues of the matrix satisfy the polynomial,

3λ3 + 41λ2 + 8(10 + R)λ + 160(R − 1) = 0.

4. If the characteristic polynomial has two pure imaginary roots ±iµ and one real root, then it has the form

(λ − r)(λ2 + µ2 ) = λ3 − rλ2 + µ2 λ − rµ2 .

Equating the λ2 and the λ term with the characteristic polynomial yields
r
41 8
r=− , µ= (10 + R).
3 3
Equating the constant term gives us the equation
41 8 160
(10 + Rc ) = (Rc − 1)
3 3 3
which has the solution
470
Rc = .
19
For this critical value of R the characteristic polynomial has the roots
41
λ1 = −
3
4√
λ2 = 2090
19
4√
λ3 = − 2090.
19

2012
Solution 50.6
The form of the perturbation expansion is
v(θ) = 1 + A cos θ + u(θ) + O(2 )
θ = (1 + ω1 + O(2 ))φ.
Writing the derivatives in terms of θ,
d d
= (1 + ω1 + · · · )
dφ dθ
2
d d2
= (1 + 2ω 1 + · · · ) .
dφ2 dθ2
Substituting these expressions into the differential equation for v(φ),
1 + 2ω1 + O(2 ) −A cos θ + u00 + O(2 ) + 1 + A cos θ + u(θ) + O(2 )
  

= 1 +  1 + 2A cos θ + A2 cos2 θ + O()

 

u00 + u − 2ω1 A cos θ =  + 2A cos θ + A2 cos2 θ + O(2 ).

Equating the coefficient of ,

1
u00 + u = 1 + 2(1 + ω1 )A cos θ + A2 (cos 2θ + 1)
2
1 1
u00 + u = (1 + A2 ) + 2(1 + ω1 )A cos θ + A2 cos 2θ.
2 2
To avoid secular terms, we must have ω1 = −1. A particular solution for u is
1 1
u = 1 + A2 − A2 cos 2θ.
2 6
The the solution for v is
 
1 2 1 2
v(φ) = 1 + A cos((1 − )φ) +  1 + A − A cos(2(1 − )φ) + O(2 ).
2 6

2013
Solution 50.7
Substituting the expressions for x and ω into the differential equations yields
  2     2  
2 2 d x2 2 3 2 d x3
a ω0 + x2 + α cos θ + a ω0 + x3 − 2ω0 ω2 cos θ + 2αx2 cos θ + O(a4 ) = 0
dθ2 dθ2
Equating the coefficient of a2 gives us the differential equation
d 2 x2 α
+ x2 = − (1 + cos 2θ).
dθ2 2ω02
The solution subject to the initial conditions x2 (0) = x02 (0) = 0 is
α
x2 = (−3 + 2 cos θ + cos 2θ).
6ω02
Equating the coefficent of a3 gives us the differential equation
 2
α2 5α2 α2 α2
  
2 d x3
ω0 + x 3 + − 2ω ω
0 2 + cos θ + cos 2θ + cos 3θ = 0.
dθ2 3ω02 6ω02 3ω02 6ω02
To avoid secular terms we must have
5α2
ω2 = − .
12ω0
Solving the differential equation for x3 subject to the intial conditions x3 (0) = x03 (0) = 0,
α2
x3 = (−48 + 29 cos θ + 16 cos 2θ + 3 cos 3θ).
144ω04
Thus our solution for x(t) is
   2 
2 α 3 α
x(t) = a cos θ + a (−3 + 2 cos θ + cos 2θ) + a (−48 + 29 cos θ + 16 cos 2θ + 3 cos 3θ) + O(a4 )
6ω02 144ω04

2014
  
5α2
where θ = ω0 − a2 12ω0
t.
Now to see why we didn’t need an aω1 term. Assume that

x = a cos θ + a2 x2 (θ) + O(a3 ); θ = ωt

ω = ω0 + aω1 + O(a2 ).

Substituting these expressions into the differential equation for x yields

a2 ω02 (x002 + x2 ) − 2ω0 ω1 cos θ + α cos2 θ = O(a3 )

 

ω1 α
x002 + x2 = 2 cos θ − 2 (1 + cos 2θ).
ω0 2ω0
In order to eliminate secular terms, we need ω1 = 0.

Solution 50.8
1. The equation for p1 (t) is

dp1 (t)
= α[p0 (t) − p1 (t)].
dt
dp1 (t)
= α[aeiωt − p1 (t)]
dt
d αt
e p1 (t) = αaeαt eiωt


dt
αa iωt
p1 (t) = e + ce−αt
α + iω
Applying the initial condition, p1 (0) = 0,

αa
eiωt − e−αt


p1 (t) =
α + iω

2015
2. We start with the differential equation for pn (t).
dpn (t)
= α[pn−1 (t) − pn (t)]
dt
Multiply by sn and sum from n = 1 to ∞.
∞
X ∞
X
p0n (t)sn = α[pn−1 (t) − pn (t)]sn
n=1 n=1
∞
∂G(s, t) X
=α pn sn+1 − αG(s, t)
∂t n=0
∞
∂G(s, t) X
= αsp0 + α pn sn+1 − αG(s, t)
∂t n=1
∂G(s, t)
= αaseiωt + αsG(s, t) − αG(s, t)
∂t
∂G(s, t)
= αaseiωt + α(s − 1)G(s, t)
∂t
∂ α(1−s)t
G(s, t) = αaseα(1−s)t eiωt


e
∂t
αas
G(s, t) = eiωt + C(s)eα(s−1)t
α(1 − s) + iω
The initial condition is
∞
X
G(s, 0) = pn (0)sn = 0.
n=1

The generating function is then

αas
αeiωt − eα(s−1)t .


G(s, t) =
α(1 − s) + iω

2016
3. Assume that |s| < 1. In the limit t → ∞ we have
αas
G(s, t) ∼ eiωt
α(1 − s) + iω
as
G(s, t) ∼ eiωt
1 + iω/α − s
as/(1 + iω/α) iωt
G(s, t) ∼ e
1 − s/(1 + iω/α)
∞  n
aseiωt X s
G(s, t) ∼
1 + iω/α n=0 1 + iω/α
∞
X sn
G(s, t) ∼ aeiωt
n=1
(1 + iω/α)n

Thus we have
a
pn (t) ∼ eiωt as t → ∞.
(1 + iω/α)n

 
a iωt
=(pn (t)) ∼ = e
(1 + iω/α)n
 n
1 − iω/α
=a [cos(ωt) + i sin(ωt)]
1 + (ω/α)2
a
= [cos(ωt)=[(1 − iω/α)n ] + sin(ωt)<[(1 − iω/α)n ]]
(1 + (ω/α)2 )n
 
n n
a  X
(j+1)/2 ω
 j X
j/2 ω
 j 
=  cos(ωt) (−1) + sin(ωt) (−1) 
(1 + (ω/α)2 )n  j=1
α j=0
α 
odd j even j

2017
Solution 50.9
1. Substituting pn = An eiωt into the differential equation yields

An iωeiω(t+τ ) = α[An−1 eiωt − An eiωt ]

An (α + iωeiωτ ) = αAn−1

We make the substitution An = rn .

rn (α + iωeiωτ ) = αr n−1
α
r=
α + iωeiωτ

Thus we have

 n
1
pn (t) = eiωt .
1 + iωeiωτ /α

2018
Taking the imaginary part,

 n 
1 iωt
=(pn (t)) = = e
1 + i ωα eiωτ
n
1 − i ωα e−iωτ
 


== cos(ωt) + i sin(ωt)
1 + i ωα (eiωτ − e−iωτ ) + ( ωα )2
1 − ωα sin(ωτ ) − i ωα cos(ωτ ) n
  


== cos(ωt) + i sin(ωt)
1 − 2 ωα sin(ωτ ) + ( ωα )2
 n h
1 h ω ω n i
= cos(ωt)= 1 − sin(ωτ ) − i cos(ωτ )
1 − 2 ωα sin(ωτ ) + ( ωα )2 α α
h ω ω n i i
+ sin(ωt)< 1 − sin(ωτ ) − i cos(ωτ )
α  α
 n
1
=
1 − 2 α sin(ωτ ) + ( ωα )2
ω

n
(j+1)/2 ω ω
h X h ij h in−j
cos(ωt) (−1) cos(ωτ ) 1 − sin(ωτ )
j=1
α α
odd j
n hω
X ij h ω in−j i
+ sin(ωt) (−1)j/2 cos(ωτ ) 1 − sin(ωτ )
j=0
α α
even j

2019
2. pn (t) will remain bounded in time as n → ∞ if
 
 1 
 1 + i ω eiωτ  ≤ 1
 
α
 ω 2
1 + i eiωτ  ≥ 1
 
α
ω  ω 2
1 − 2 sin(ωτ ) + ≥1
α α
ω
≥ 2 sin(ωτ )
α

3.

2020
Chapter 51

Nonlinear Partial Differential Equations

2021
51.1 Exercises
Exercise 51.1
Solve the equation

φt + (1 + x)φx + φ = 0 in − ∞ < x < ∞, t > 0,

with initial condition φ(x, 0) = f (x).

Exercise 51.2
Solve the equation

αφ
φt + φx + =0
1+x
in the region 0 < x < ∞, t > 0 with initial condition φ(x, 0) = 0, and boundary condition φ(0, t) = g(t). [Here α
is a positive constant.]

Exercise 51.3
Solve the equation

φt + φx + φ2 = 0

in −∞ < x < ∞, t > 0 with initial condition φ(x, 0) = f (x). Note that the solution could become infinite in
finite time.

Exercise 51.4
Consider

ct + ccx + µc = 0, −∞ < x < ∞, t > 0.

2022
 1. Use the method of characteristics to solve the problem with

c = F (x) at t = 0.

(µ is a positive constant.)
2. Find equations for the envelope of characteristics in the case F 0 (x) < 0.
3. Deduce an inequality relating max |F 0 (x)| and µ which decides whether breaking does or does not occur.

Exercise 51.5
For water waves in a channel the so-called shallow water equations are

ht + (hv)x = 0 (51.1)
 
2 1 2
(hv)t + hv + gh = 0, g = constant. (51.2)
2 x

Investigate whether there are solutions with v = V (h), where V (h) is not posed in advance but is obtained from
requiring consistency between the h equation obtained from (1) and the h equation obtained from (2).
There will be two possible choices for V (h) depending on a choice of sign. Consider each case separately. In
each case fix the arbitrary constant that arises in V (h) by stipulating that before the waves arrive, h is equal to
the undisturbed depth h0 and V (h0 ) = 0.
Find the h equation and the wave speed c(h) in each case.

Exercise 51.6
After a change of variables, the chemical exchange equations can be put in the form
∂ρ ∂σ
+ =0 (51.3)
∂t ∂x
∂ρ
= ασ − βρ − γρσ; α, β, γ = positive constants. (51.4)
∂t

2023
 1. Investigate wave solutions in which ρ = ρ(X), σ = σ(X), X = x − U t, U = constant, and show that ρ(X)
must satisfy an ordinary differential equation of the form
dρ
= quadratic in ρ.
dX

2. Discuss ths “smooth shock” solution as we did for a different example in class. In particular find the
expression for U in terms of the values of ρ as X → ±∞, and find the sign of dρ/dX. Check that
σ 2 − σ1
U=
ρ2 − ρ1
in agreement with the “discontinuous theory.”

Exercise 51.7
Find solitary wave solutions for the following equations:

1. ηt + ηx + 6ηηx − ηxxt = 0. (Regularized long wave or B.B.M. equation)

2. utt − uxx − 23 u2 xx − uxxxx = 0. (“Boussinesq”)

3. φtt − φxx + 2φx φxt + φxx φt − φxxxx = 0. (The solitary wave form is for u = φx )

4. ut + 30u2 u1 + 20u1 u2 + 10uu3 + u5 = 0. (Here the subscripts denote x derivatives.)

2024
51.2 Hints
Hint 51.1

Hint 51.2

Hint 51.3

Hint 51.4

Hint 51.5

Hint 51.6

Hint 51.7

2025
51.3 Solutions
Solution 51.1
The method of characteristics gives us the differential equations

x0 (t) = (1 + x) x(0) = ξ
dφ
= −φ φ(ξ, 0) = f (ξ)
dt
Solving the first differential equation,

x(t) = cet − 1, x(0) = ξ

t
x(t) = (ξ + 1)e − 1

The second differential equation then becomes

φ(x(t), t) = ce−t , φ(ξ, 0) = f (ξ), ξ = (x + 1)e−t − 1

φ(x, t) = f ((x + 1)e−t − 1)e−t

Thus the solution to the partial differential equation is

φ(x, t) = f ((x + 1)e−t − 1)e−t .

Solution 51.2

dφ αφ
= φt + x0 (t)φx = −
dt 1+x
The characteristic curves x(t) satisfy x0 (t) = 1, so x(t) = t + c. The characteristic curve that separates the region
with domain of dependence on the x axis and domain of dependence on the t axis is x(t) = t. Thus we consider
the two cases x > t and x < t.

2026
• x > t. x(t) = t + ξ.

• x < t. x(t) = t − τ .

Now we solve the differential equation for φ in the two domains.

• x > t.
dφ αφ
=− , φ(ξ, 0) = 0, ξ =x−t
dt 1+x
dφ αφ
=−
dt 1+t+ξ
 Z t 
1
φ = c exp −α dt
t+ξ+1
φ = cexp (−α log(t + ξ + 1))
φ = c(t + ξ + 1)−α

applying the initial condition, we see that

φ=0

• x < t.
dφ αφ
=− , φ(0, τ ) = g(τ ), τ =t−x
dt 1+x
dφ αφ
=−
dt 1+t−τ
φ = c(t + 1 − τ )−α
φ = g(τ )(t + 1 − τ )−α
φ = g(t − x)(x + 1)−α

2027
Thus the solution to the partial differential equation is
(
0 for x > t
φ(x, t) =
g(t − x)(x + 1)−α for x < t.

Solution 51.3
The method of characteristics gives us the differential equations

x0 (t) = 1 x(0) = ξ
dφ
= −φ2 φ(ξ, 0) = f (ξ)
dt
Solving the first differential equation,

x(t) = t + ξ.

The second differential equation is then

dφ
= −φ2 , φ(ξ, 0) = f (ξ), ξ =x−t
dt
φ−2 dφ = −dt
−φ−1 = −t + c
1
φ=
t−c
1
φ=
t + 1/f (ξ)
1
φ= .
t + 1/f (x − t)

2028
Solution 51.4
1. Taking the total derivative of c with respect to t,
dc dx
= ct + cx .
dt dt
Equating terms with the partial differential equation, we have the system of differential equations
dx
=c
dt
dc
= −µc.
dt
subject to the initial conditions
x(0) = ξ, c(ξ, 0) = F (ξ).
We can solve the second ODE directly.
c(ξ, t) = c1 e−µt
c(ξ, t) = F (ξ)e−µt
Substituting this result and solving the first ODE,
dx
= F (ξ)e−µt
dt
F (ξ) −µt
x(t) = − e + c2
µ
F (ξ)
x(t) = (1 − e−µt ) + ξ.
µ

The solution to the problem at the point (x, t) is found by first solving
F (ξ)
x= (1 − e−µt ) + ξ
µ

2029
 for ξ and then using this value to compute
c(x, t) = F (ξ)e−µt .

2. The characteristic lines are given by the equation

F (ξ)
x(t) = (1 − e−µt ) + ξ.
µ
The points on the envelope of characteristics also satisfy
∂x(t)
= 0.
∂ξ
Thus the points on the envelope satisfy the system of equations
F (ξ)
x= (1 − e−µt ) + ξ
µ
F 0 (ξ)
0= (1 − e−µt ) + 1.
µ
By substituting
µ
1 − e−µt = −
F 0 (ξ)
into the first equation we can eliminate its t dependence.
F (ξ)
x=− +ξ
F 0 (ξ)
Now we can solve the second equation in the system for t.
µ
e−µt = 1 + 0
F (ξ)
 
1 µ
t = − log 1 + 0
µ F (ξ)

2030
 Thus the equations that describe the envelope are

F (ξ)
x=− +ξ
F 0 (ξ)
 
1 µ
t = − log 1 + 0 .
µ F (ξ)

3. The second equation for the envelope has a solution for positive t if there is some x that satisfies
µ
−1 < < 0.
F 0 (x)

This is equivalent to

−∞ < F 0 (x) < −µ.

So in the case that F 0 (x) < 0, there will be breaking iff

max |F 0 (x)| > µ.

Solution 51.5
With the substitution v = V (h), the two equations become

ht + (V + hV 0 )hx = 0
(V + hV 0 )ht + (V 2 + 2hV V 0 + gh)hx = 0.

We can rewrite the second equation as

V 2 + 2hV V 0 + gh
ht + hx = 0.
V + hV 0

2031
Requiring that the two equations be consistent gives us a differential equation for V .
V 2 + 2hV V 0 + gh
V + hV 0 =
V + hV 0
V 2 + 2hV V 0 + h2 (V 0 )2 = V 2 + 2hV V 0 + gh
g
(V 0 )2 = .
h
There are two choices depending on which sign we choose when taking the square root of the above equation.
Positive V0 .
r
0 g
V =
h
p
V = 2 gh + const

We apply the initial condition V (h0 ) = 0.

√ √ p
V = 2 g( h − h0 )
The partial differential equation for h is then
√ √ p
ht + (2 g( h − h0 )h)x = 0
√ √ p
ht + g(3 h − 2 h0 )hx = 0
The wave speed is
√ √ p
c(h) = g(3 h − 2 h0 ).
Negative V0 .
r
0 g
V =−
h
p
V = −2 gh + const

2032
We apply the initial condition V (h0 ) = 0.

√ p √
V = 2 g( h0 − h)

The partial differential equation for h is then

√ p √
ht + g(2 h0 − 3 h)hx = 0.

The wave speed is

√ p √
c(h) = g(2 h0 − 3 h).

Solution 51.6
1. Making the substitutions, ρ = ρ(X), σ = σ(X), X = x − U t, the system of partial differential equations
becomes

−U ρ0 + σ 0 = 0
−U ρ0 = ασ − βρ − γρσ.

Integrating the first equation yields

−U ρ + σ = c
σ = c + U ρ.

Now we substitute the expression for σ into the second partial differential equation.

−U ρ0 = α(c + U ρ) − βρ − γρ(c + U ρ)
 c β  c
ρ0 = −α ρ + + ρ + γρ ρ +
U U U

2033
 Thus ρ(X) satisfies the ordinary differential equation

 
0 2 γc β αc
ρ = γρ + + −α ρ− .
U U U

2. Assume that

ρ(X) → ρ1 as X → +∞
ρ(X) → ρ2 as X → −∞
ρ0 (X) → 0 as X → ±∞.

Integrating the ordinary differential equation for ρ,

Z ρ
dρ
X= γc β

αc
.
γρ2 + U
+ U
−α ρ− U

We see that the roots of the denominator of the integrand must be ρ1 and ρ2 . Thus we can write the
ordinary differential equation for ρ(X) as

ρ0 (X) = γ(ρ − ρ1 )(ρ − ρ2 ) = γρ2 − γ(ρ1 + ρ2 )ρ + γρ1 ρ2 .

Equating coefficients in the polynomial with the differential equation for part 1, we obtain the two equations

αc γc β
− = γρ1 ρ2 , + − α = −γ(ρ1 + ρ2 ).
U U U
Solving the first equation for c,

U γρ1 ρ2
c=− .
α

2034
Now we substitute the expression for c into the second equation.

γU γρ1 ρ2 β
− + − α = −γ(ρ1 + ρ2 )
αU U
β γ 2 ρ1 ρ2
=α+ − γ(ρ1 + ρ2 )
U α

Thus we see that U is

αβ
U= .
α2 + γ2ρ 1 ρ2 − −αγ(ρ1 + ρ2 )

Since the quadratic polynomial in the ordinary differential equation for ρ(X) is convex, it is negative valued
between its two roots. Thus we see that

dρ
< 0.
dX

Using the expression for σ that we obtained in part 1,

σ2 − σ1 c + U ρ2 − (c + U ρ1 )
=
ρ2 − ρ1 ρ2 − ρ1
ρ2 − ρ1
=U
ρ2 − ρ1
= U.

2035
 Now let’s return to the ordinary differential equation for ρ(X)

ρ0 (X) = γ(ρ − ρ1 )(ρ − ρ2 )

Z ρ
dρ
X=
γ(ρ − ρ1 )(ρ − ρ2 )
Z ρ 
1 1 1
X=− + dρ
γ(ρ2 − ρ1 ) ρ − ρ1 ρ2 − ρ
 
1 ρ − ρ1
X − X0 = − ln
γ(ρ2 − ρ1 ) ρ −ρ
 2 
ρ − ρ1
−γ(ρ2 − ρ1 )(X − X0 ) = ln
ρ2 − ρ
ρ − ρ1
= exp (−γ(ρ2 − ρ1 )(X − X0 ))
ρ2 − ρ
ρ − ρ1 = (ρ2 − ρ) exp (−γ(ρ2 − ρ1 )(X − X0 ))
ρ [1 + exp (−γ(ρ2 − ρ1 )(X − X0 ))] = ρ1 + ρ2 exp (−γ(ρ2 − ρ1 )(X − X0 ))

Thus we obtain a closed form solution for ρ

ρ1 + ρ2 exp (−γ(ρ2 − ρ1 )(X − X0 ))

ρ=
1 + exp (−γ(ρ2 − ρ1 )(X − X0 ))

Solution 51.7
1.

ηt + ηx + 6ηηx − ηxxt = 0

We make the substitution

η(x, t) = z(X), X = x − U t.

2036
 (1 − U )z 0 + 6zz 0 + U z 000 = 0
(1 − U )z + 3z 2 + U z 00 = 0
1 1
(1 − U )z 2 + z 3 + U (z 0 )2 = 0
2 2
U − 1 2
(z 0 )2 = z2 − z3
U rU !
U −1 1 U − 1
z(X) = sech 2 X
2 2 U
r !!
U −1 2 1 U −1 p
η(x, t) = sech x − (U − 1)U t
2 2 U

The linearized equation is

ηt + ηx − ηxxt = 0.

Substituting η = e−αx+βt into this equation yields

β − α − α2 β = 0
α
β= .
1 − α2

We set

U −1
α2 = .
U

2037
 β is then
α
β=
1p− α2
(U − 1)/U
=
1 − (U − 1)/U )
p
(U − 1)U
=
U − (U − 1)
p
= (U − 1)U .

The solution for η becomes

 
αβ αx − βt
sech 2
2 2
where
α
β= .
1 − α2
2.
 
3 2
utt − uxx − u − uxxxx = 0
2 xx
We make the substitution
u(x, t) = z(X), X = x − U t.
 00
00 3 2
2
(U − 1)z − z − z 0000 = 0
2
 0
0 3 2
2
(U − 1)z − z − z 000 = 0
2
3
(U 2 − 1)z − z 2 − z 00 = 0
2

2038
We multiply by z 0 and integrate.
1 2 1 1
(U − 1)z 2 − z 3 − (z 0 )2 = 0
2 2 2
(z 0 )2 = (U 2 − 1)z 2 − z 3
1√ 2
 
2 2
z = (U − 1) sech U − 1X
2
1 √ 2 √
  
2 2
u(x, t) = (U − 1) sech U − 1x − U U 2 − 1t
2

The linearized equation is

utt − uxx − uxxxx = 0.
Substituting u = e−αx+βt into this equation yields
β 2 − α2 − α4 = 0
β 2 = α2 (α2 + 1).

We set
√
α= U 2 − 1.
β is then
β 2 = α2 (α2 + 1)
= (U 2 − 1)U 2
√
β = U U 2 − 1.

The solution for u becomes

 
2 2 αx − βt
u(x, t) = α sech
2

2039
 where

β 2 = α2 (α2 + 1).

3.

φtt − φxx + 2φx φxt + φxx φt − φxxxx

We make the substitution

φ(x, t) = z(X), X = x − U t.

(U 2 − 1)z 00 − 2U z 0 z 00 − U z 00 z 0 − z 0000 = 0
(U 2 − 1)z 00 − 3U z 0 z 00 − z 0000 = 0
3
(U 2 − 1)z 0 − (z 0 )2 − z 000 = 0
2

Multiply by z 00 and integrate.

1 2 1 1
(U − 1)(z 0 )2 − (z 0 )3 − (z 00 )2 = 0
2 2 2
(z ) = (U − 1)(z ) − (z 0 )3
00 2 2 0 2

1√ 2
 
0 2 2
z = (U − 1) sech U − 1X
2
1 √ 2 √
  
2 2
φx (x, t) = (U − 1) sech U − 1x − U U 2 − 1t .
2

The linearized equation is

φtt − φxx − φxxxx

2040
 Substituting φ = e−αx+βt into this equation yields

β 2 = α2 (α2 + 1).

The solution for φx becomes

 
2 2 αx − βt
φx = α sech
2
where

β 2 = α2 (α2 + 1).

4.

ut + 30u2 u1 + 20u1 u2 + 10uu3 + u5 = 0

We make the substitution

u(x, t) = z(X), X = x − U t.

−U z 0 + 30z 2 z 0 + 20z 0 z 00 + 10zz 000 + z (5) = 0

Note that (zz 00 )0 = z 0 z 00 + zz 000 .

−U z 0 + 30z 2 z 0 + 10z 0 z 00 + 10(zz 00 )0 + z (5) = 0

−U z + 10z 3 + 5(z 0 )2 + 10zz 00 + z (4) = 0

Multiply by z 0 and integrate.

1 5 1
− U z 2 + z 4 + 5z(z 0 )2 − (z 00 )2 + z 0 z 000 = 0
2 2 2

2041
Assume that

(z 0 )2 = P (z).

Differentiating this relation,

2z 0 z 00 = P 0 (z)z 0
1
z 00 = P 0 (z)
2
1
z 000 = P 00 (z)z 0
2
1
z 000 z 0 = P 00 (z)P (z).
2

Substituting this expressions into the differential equation for z,

1 5 11 0 1
− U z 2 + z 4 + 5zP (z) − (P (z))2 + P 00 (z)P (z) = 0
2 2 24 2
4U z 2 + 20z 4 + 40zP (z) − (P 0 (z))2 + 4P 00 (z)P (z) = 0

Substituting P (z) = az 3 + bz 2 yields

(20 + 40a + 15a2 )z 4 + (40b + 20ab)z 3 + (4b2 + 4U )z 2 = 0

This equation is satisfied by b2 = U , a = −2. Thus we have

√
(z 0 )2 = U z 2 − 2z 3
√  
U 2 1 1/4
z= sech U X
2 2
√  
U 2 1 1/4 5/4
u(x, t) = sech (U x − U t)
2 2

2042
The linearized equation is

ut + u5 = 0.

Substituting u = e−αx+βt into this equation yields

β − α5 = 0.

We set

α = U 1/4 .

The solution for u(x, t) becomes

α2
 
αx − βt
sech 2
2 2

where

β = α5 .

2043
Part VIII

Appendices

2044
Appendix A

Greek Letters

Name Lower Upper

alpha α
beta β
gamma γ Γ
delta δ ∆
epsilon 
iota ι
kappa κ
lambda λ Λ
mu µ
nu ν
omicron o
pi π Π
rho ρ
sigma σ Σ
tau τ
theta θ Θ

2045
phi φ Φ
psi ψ Ψ
chi χ
omega ω Ω
upsilon υ Υ
xi ξ Ξ
eta η
zeta ζ

2046
Appendix B

Notation

C class of continuous functions

Cn class of n-times continuously differentiable functions
C set of complex numbers
δ(x) Dirac delta function
F[·] Fourier transform
Fc [·] Fourier cosine transform
Fs [·] Fourier sine transformR
∞
γ Euler’s constant, γ = 0 e−x Log x dx
Γ(ν) Gamma function
H(x) Heaviside function
(1)
Hν (x) Hankel function of the first kind and order ν
(2)
Hν (x) Hankel function of the second kind and order ν
Jν (x) Bessel function of the first kind and order ν
Kν (x) Modified Bessel function of the first kind and order ν
L[·] Laplace transform
N set of natural numbers, (positive integers)

2047
Nν (x) Modified Bessel function of the second kind and order ν
R set of real numbers
R+ set of positive real numbers
R− set of negative real numbers
o(z) terms smaller than z
O(z)
R terms no bigger than z
− principal value of the integral
d
ψ(ν) digamma function, ψ(ν) = dν log Γ(ν)
dn
ψ (n) (ν) (n)
polygamma function, ψ (ν) = dν n ψ(ν)
∂nu
u(n) (x) ∂xn
∂ n+m u
u(n,m) (x, y) ∂xn ∂y m
Yν (x) Bessel function of the second kind and order ν, Neumann function
Z set of integers
Z+ set of positive integers

2048
Appendix C

Formulas from Complex Variables

Analytic Functions. A function f (z) is analytic in a domain if the derivative f 0 (z) exists in that domain.
If f (z) = u(x, y) + iv(x, y) is defined in some neighborhood of z0 = x0 + iy0 and the partial derivatives of u
and v are continuous and satisfy the Cauchy-Riemann equations

u x = vy , uy = −vx ,

then f 0 (z0 ) exists.

Residues. If f (z) has the Laurent expansion

∞
X
f (z) = an z n ,
n=−∞

then the residue of f (z) at z = z0 is

Res (f (z), z0 ) = a−1 .

2049
Residue Theorem. Let C be a positively oriented, simple, closed contour. If f (z) is analytic in and on C
except for isolated singularities at z1 , z2 , . . . , zN inside C then
I XN
f (z) dz = 2πi Res (f (z), zn ).
C n=1

If in addition f (z) is analytic outside C in the finite complex plane then

I    
1 1
f (z) dz = 2πi Res f ,0 .
C z2 z

Residues of a pole of order n. If f (z) has a pole of order n at z = z0 then

dn−1 
 
1 n

Res (f (z), z0 ) = lim (z − z0 ) f (z) .
z→z0 (n − 1)! dz n−1

Jordan’s Lemma.
Z π
π
e−R sin θ dθ < .
0 R
Let a be a positive constant. If f (z) vanishes as |z| → ∞ then the integral
Z
f (z) eiaz dz
C

along the semi-circle of radius R in the upper half plane vanishes as R → ∞.

Taylor Series. Let f (z) be a function that is analytic and single valued in the disk |z − z0 | < R.
∞
X f (n) (z0 )
f (z) = (z − z0 )n .
n=0
n!

The series converges for |z − z0 | < R.

2050
Laurent Series. Let f (z) be a function that is analytic and single valued in the annulus r < |z − z0 | < R. In
this annulus f (z) has the convergent series,
∞
X
f (z) = cn (z − z0 )n ,
n=−∞

where
I
1 f (z)
cn = dz
2πi (z − z0 )n+1

and the path of integration is any simple, closed, positive contour around z0 and lying in the annulus. The path
of integration is shown in Figure C.1.

Im(z)

r
Re(z)

Figure C.1: The Path of Integration.

2051
Appendix D

Table of Derivatives

Note: c denotes a constant and 0 denotes differentiation.

d df dg
(f g) = g+f
dx dx dx
f 0g − f g0
 
d f
=
dx g g2

d
(f c ) = cf c−1 f 0
dx
d
[f (g)] = f 0 (g)g 0
dx

d2
[f (g)] = f 00 (g)(g 0 )2 + f 0 g 00
dx2

dn
  n   n−1   n−2 2   n
n d f n d f dg n d fdg n d g
(f g) = g + + + · · · + f
dxn 0 dxn 1 dxn−1 dx 2 dxn−2 dx2 n dxn

2052
 d 1
(log x) =
dx f

d x
(c ) = cx log c
dx
d df dg
(f g ) = gf g−1 + f g log f
dx dx dx
d
(sin x) = cos x
dx
d
(cos x) = − sin x
dx
d
(tan x) = sec2 x
dx
d
(csc x) = − csc x cot x
dx
d
(sec x) = sec x tan x
dx
d
(cot x) = − csc2 x
dx
d 1 π π
(arcsin x) = √ , − ≤ arcsin x ≤
dx 1 − x2 2 2

2053
 d 1
(arccos x) = − √ , 0 ≤ arccos x ≤ π
dx 1 − x2

d 1 π π
(arctan x) = , − ≤ arctan x ≤
dx 1 + x2 2 2
d
(sinh x) = cosh x
dx
d
(cosh x) = sinh x
dx
d
(tanh x) = sech 2 x
dx
d
( csch x) = − csch x coth x
dx
d
( sech x) = − sech x tanh x
dx
d
(coth x) = − csch 2 x
dx
d 1
( arcsinh x) = √
dx x2 + 1

d 1
( arccosh x) = √ , x > 1, arccosh x > 0
dx 2
x −1

d 1
( arctanh x) = , x2 < 1
dx 1 − x2

2054
 Z x
d
f (ξ) dξ = f (x)
dx c
Z c
d
f (ξ) dξ = −f (x)
dx x

Z h Z h
d ∂f (ξ, x)
f (ξ, x) dξ = dξ + f (h, x)h0 − f (g, x)g 0
dx g g ∂x

2055
Appendix E

Table of Integrals

Z Z
dv du
u dx = uv − v dx
dx dx

f 0 (x)
Z
dx = log f (x)
f (x)

f 0 (x)
Z p
p dx = f (x)
2 f (x)

xα+1
Z
xα dx = for α 6== −1
α+1
Z
1
dx = log x
x

eax
Z
eax dx =
a

2056
 abx
Z
bx
a dx = for a > 0
b log a
Z
log x dx = x log x − x

Z
1 1 x
dx = arctan
x2
+a 2 a a
(
1
log a−x for x2 < a2
Z
1 2a a+x
dx = 1 x−a
x2 − a 2 2a
log x+a
for x2 > a2
Z
1 x x
√ dx = arcsin = − arccos for x2 < a2
2
a −x 2 |a| |a|
Z
1 √
√ dx = log(x + x2 ± a2 )
x2 ± a2
Z
1 1 x
√ dx = sec−1
2
x x −a2 |a| a
 √ 
a2 ± x 2
Z
1 1 a+
√ dx = − log
x a 2 ± x2 a x
Z
1
sin(ax) dx = − cos(ax)
a
Z
1
cos(ax) dx = sin(ax)
a

2057
Z
1
tan(ax) dx = − log cos(ax)
a
Z
1 ax
csc(ax) dx = log tan
a 2
Z
1  π ax 
sec(ax) dx = log tan +
a 4 2
Z
1
cot(ax) dx = log sin(ax)
a
Z
1
sinh(ax) dx = cosh(ax)
a
Z
1
cosh(ax) dx = sinh(ax)
a
Z
1
tanh(ax) dx = log cosh(ax)
a
Z
1 ax
csch (ax) dx = log tanh
a 2
Z  
i iπ ax
sech (ax) dx = log tanh +
a 4 2
Z
1
coth(ax) dx = log sinh(ax)
a

2058
Z
1 x
x sin ax dx = 2
sin ax − cos ax
a a

a 2 x2 − 2
Z
2x
x2 sin ax dx = sin ax − cos ax
a2 a3
Z
1 x
x cos ax dx = 2
cos ax + sin ax
a a

2x cos ax a2 x2 − 2
Z
x2 cos ax dx = + sin ax
a2 a3

2059
Appendix F

Definite Integrals

Integrals from −∞ to ∞. Let f (z) be analytic except for isolated singularities, none of which lie on the real
axis. Let a1 , . . . , am be the singularities of f (z) in the upper half plane; and CR be the semi-circle from R to −R
in the upper half plane. If
 
lim R max |f (z)| = 0
R→∞ z∈CR

then
Z ∞ m
X
f (x) dx = i2π Res (f (z), aj ) .
−∞ j=1

Let b1 , . . . , bn be the singularities of f (z) in the lower half plane. Let CR be the semi-circle from R to −R in the
lower half plane. If
 
lim R max |f (z)| = 0
R→∞ z∈CR

2060
then
Z ∞ n
X
f (x) dx = −i2π Res (f (z), bj ) .
−∞ j=1

Integrals from 0 to ∞. Let f (z) be analytic except for isolated singularities, none of which lie on the positive
real axis, [0, ∞). Let z1 , . . . , zn be the singularities of f (z). If f (z)  z α as z → 0 for some α > −1 and f (z)  z β
as z → ∞ for some β < −1 then
Z ∞ Xn
f (x) dx = − Res (f (z) log z, zk ) .
0 k=1

Z ∞ n n
1X 2

X
f (x) log dx = − Res f (z) log z, zk + iπ Res (f (z) log z, zk )
0 2 k=1 k=1

Assume that a is not an integer. If z a f (z)  z α as z → 0 for some α > −1 and z a f (z)  z β as z → ∞ for some
β < −1 then
Z ∞ n
a i2π X
x f (x) dx = i2πa
Res (z a f (z), zk ) .
0 1 − e k=1

∞ n n
π2a X
Z
a i2π X a
x f (x) log x dx = i2πa
Res (z f (z) log z, zk ) , + 2 Res (z a f (z), zk )
0 1− e k=1
sin (πa) k=1

Fourier Integrals. Let f (z) be analytic except for isolated singularities, none of which lie on the real axis.
Suppose that f (z) vanishes as |z| → ∞. If ω is a positive real number then
Z ∞ X n
iωx
f (x) e dx = i2π Res (f (z) eiωz , zk ),
−∞ k=1

2061
where z1 , . . . , zn are the singularities of f (z) in the upper half plane. If ω is a negative real number then
Z ∞ n
X
iωx
f (x) e dx = −i2π Res (f (z) eiωz , zk ),
−∞ k=1

where z1 , . . . , zn are the singularities of f (z) in the lower half plane.

2062
Appendix G

Table of Sums

∞
X r
rn = , for |r| < 1
n=1
1−r

N
X r − rN +1
rn =
n=1
1−r

b
X (a + b)(b + 1 − a)
n=
n=a
2

N
X N (N + 1)
n=
n=1
2

b
X b(b + 1)(2b + 1) − a(a − 1)(2a − 1)
n2 =
n=a
6

2063
N
X N (N + 1)(2N + 1)
n2 =
n=1
6

∞
X (−1)n+1
= log(2)
n=1
n

∞
X 1 π2
=
n=1
n2 6

∞
X (−1)n+1 π2
=
n=1
n2 12

∞
X 1
3
= ζ(3)
n=1
n

∞
X (−1)n+1 3ζ(3)
=
n=1
n3 4

∞
X 1 π4
=
n=1
n4 90

∞
X (−1)n+1 7π 4
=
n=1
n4 720

2064
 ∞
X 1
= ζ(5)
n=1
n5

∞
X (−1)n+1 15ζ(5)
=
n=1
n5 16

∞
X 1 π6
=
n=1
n6 945

∞
X (−1)n+1 31π 6
=
n=1
n6 30240

2065
Appendix H

Table of Taylor Series

∞
X
(1 − z)−1 = zn |z| < 1
n=0

∞
X
−2
(1 − z) = (n + 1)z n |z| < 1
n=0

∞  
α
X α n
(1 + z) = z |z| < 1
n=0
n

∞
z
X zn
e = |z| < ∞
n=0
n!

∞
X zn
log(1 − z) = − |z| < 1
n=1
n

2066
 ∞
z 2n−1
 
1+z X
log =2 |z| < 1
1−z n=1
2n − 1

∞
X (−1)n z 2n
cos z = |z| < ∞
n=0
(2n)!

∞
X (−1)n z 2n+1
sin z = |z| < ∞
n=0
(2n + 1)!

z 3 2z 5 17z 7
tan z = z + + + + ··· |z| < π2
3 15 315

z3 1 · 3z 5 1 · 3 · 5z 7
 
−1 π
cos z = − z + + + + ··· |z| < 1
2 2·3 2·4·5 2·4·6·7

z3 1 · 3z 5 1 · 3 · 5z 7
sin−1 z = z + + + + ··· |z| < 1
2·3 2·4·5 2·4·6·7
∞
−1
X (−1)n+1 z 2n−1
tan z= |z| < 1
n=1
2n − 1

∞
X z 2n
cosh z = |z| < ∞
n=0
(2n)!

∞
X z 2n+1
sinh z = |z| < ∞
n=0
(2n + 1)!

2067
 z 3 2z 5 17z 7 π
tanh z = z − + − + ··· |z| < 2
3 15 315
∞
X (−1)n  z ν+2n
Jν (z) = |z| < ∞
n=0
n!Γ(ν + n + 1) 2

∞
X 1  z ν+2n
Iν (z) = |z| < ∞
n=0
n!Γ(ν + n + 1) 2

2068
Appendix I

Table of Laplace Transforms

Let f (t) be piecewise continuous and of exponential order α. Unless otherwise noted, the transform is defined for
s > 0.
Z ∞
f (t) e−st f (t) dt
0

Z c+i∞
1
ets F (s) ds F (s)
2πi c−i∞

af (t) + bg(t) aF (s) + bG(s)

ect f (t) F (s − c) s>c+α

f (t + c) F (s − c) s>c+α

d
tf (t) − [F (s)]
ds

2069
 dn
tn f (t) (−1)n [F (s)]
dsn
Z 1 Z ∞
f (t) f (t)
, dt exists F (t) dt
t 0 t s

Z t
F (s)
f (τ ) dτ
0 s
Z tZ τ
F (s)
f (s) ds dτ
0 0 s2

d
f (t) sF (s) − f (0)
dt

d2
f (t) s2 F (s) − sf (0) − f 0 (0)
dt2
dn
f (t) sn F (s) − sn−1 f (0)
dtn
−sn−2 f 0 (0) − · · · − f (n−1) (0)
Z t
f (τ )g(t − τ ) dτ, f, g ∈ C 0 F (s)G(s)
0

1
f (t/c), c>0 F (cs)
c
RT
0
e−st f (t) dt
f (t), f (t + T ) = f (t)
1 − e−sT

2070
 RT
0
e−st f (t) dt
f (t), f (t + T ) = −f (t)
1 + e−sT
1
H(t)
s
1
tH(t)
s2
n!
tn H(t), for n = 0, 1, 2, . . .
sn+1
√
π −3/2
t1/2 H(t) s
2
√ −1/2
t−1/2 H(t) πs
√
n−1/2 + (1)(3)(5) · · · (2n − 1) π −n−1/2
t H(t), n∈Z s
2n
Γ(ν + 1)
tν H(t), <(ν) > −1
sn+1
−γ − Log s
Log tH(t)
s
Γ(ν + 1)
tν Log tH(t), <(ν) > −1 (ψ(ν + 1) − Log s)
sn+1

δ(t) 1 s>0

δ (n) (t), n ∈ Z0+ sn s>0

2071
 1
ect H(t) s>c
s−c
1
t ect H(t) s>c
(s − c)2

tn−1 ect 1
H(t), n ∈ Z+ s>c
(n − 1)! (s − c)n
c
sin(ct)H(t)
s2 + c2
s
cos(ct)H(t)
s2 + c2
c
sinh(ct)H(t) s > |c|
s2 − c2
s
cosh(ct)H(t) s > |c|
s2 − c2
2cs
t sin(ct)H(t)
(s2 + c 2 )2

s2 − c2
t cos(ct)H(t)
(s2 + c2 )2

n!
tn ect H(t), n ∈ Z+
(s − c)n+1
c
edt sin(ct)H(t) s>d
(s − d)2 + c2

2072
 s−d
edt cos(ct)H(t) s>d
(s − d)2 + c2
(
0 for c < 0
δ(t − c) −sc
e for c > 0
(
0 for t < c 1 −cs
H(t − c) = e
1 for t > c s

cn
Jν (ct)H(t) √ √
ν ν > −1
s2 + c2 s + s2 + c2

cn
Iν (ct)H(t) √ √
ν <(s) > c, ν > −1
s2 − c2 s − s2 + c2

2073
Appendix J

Table of Fourier Transforms

Z ∞
1
f (x) f (x) e−iωx dx
2π −∞
Z ∞
F(ω) eiωx dω F(ω)
−∞

af (x) + bg(x) aF (ω) + bG(ω)

f (n) (x) (iω)n F (ω)

xn f (x) in F (n) (ω)

f (x + c) eiωc F (ω)

e−icx f (x) F (ω + c)

f (cx) |c|−1 F (ω/c)

2074
 Z ∞
f (x)g(x) F ∗ G(ω) = F (η)G(ω − η) dη
−∞
Z ∞
1 1
f ∗ g(x) = f (ξ)g(x − ξ) dξ F (ω)G(ω)
2π 2π −∞

2 1 2
e−cx , c>0 √ e−ω /4c
4πc

c/π
e−c|x| , c>0
ω2 + c2
2c
, c>0 e−c|ω|
x2 + c2
(
1 0 for ω > 0
, α>0
x − iα i eαω for ω < 0
(
1 i eαω for ω > 0
, α<0
x − iα 0 for ω < 0

1 i
− sign (ω)
x 2
(
0 for x < c 1 −icω
H(x − c) = e
1 for x > c 2πiω

2075
 1
e−cx H(x), <(c) > 0
2π(c + iω)

1
ecx H(−x), <(c) > 0
2π(c − iω)

1 δ(ω)

1 −iωξ
δ(x − ξ) e
2π

π(δ(x + ξ) + δ(x − ξ)) cos(ωξ)

−iπ(δ(x + ξ) − δ(x − ξ)) sin(ωξ)

(
1 for |x| < c sin(cω)
H(c − |x|) = ,c>0
0 for |x| > c πω

2076
Appendix K

Table of Fourier Transforms in n Dimensions

Z
1
f (x) f (x) e−iωx dx
(2π)n Rn
Z
F(ω) eiωx dω F(ω)
Rn

af (x) + bg(x) aF (ω) + bG(ω)

 π n/2 2 /4c 2
e−nx e−cω
c

2077
Appendix L

Table of Fourier Cosine Transforms

Z ∞
1
f (x) f (x) cos (ωx) dx
π 0
Z ∞
2 C(ω) cos (ωx) dω C(ω)
0

1
f 0 (x) ωS(ω) − f (0)
π
1 0
f 00 (x) −ω 2 C(ω) − f (0)
π
∂
xf (x) Fs [f (x)]
∂ω
1 ω 
f (cx), c>0 C
c c

2078
 2c
e−cω
x2 + c2

c/π
e−cx
ω2 + c2
2 1 2
e−cx √ e−ω /(4c)
4πc
r
π −x2 /(4c) 2
e e−cω
c

2079
Appendix M

Table of Fourier Sine Transforms

Z ∞
1
f (x) f (x) sin (ωx) dx
π 0
Z ∞
2 S(ω) sin (ωx) dω S(ω)
0

f 0 (x) −ωC(ω)

1
f 00 (x) −ω 2 S(ω) + ωf (0)
π
∂
xf (x) − Fc [f (x)]
∂ω
1 ω 
f (cx), c>0 S
c c

2080
 2x
e−cω
x2 + c2

ω/π
e−cx
ω2 + c2
x 1 −cω
2 arctan e
c ω
1 −cx 1 ω 
e arctan
x π c
1
1
πω
2
1
x
2 ω 2 /(4c)
x e−cx √ e−ω
4c3/2 π
√
πx −x2 /(4c) 2
e ω e−cω
2c3/2

2081
Appendix N

Table of Wronskians

W [x − a, x − b] b−a

W eax , ebx (b − a) e(a+b)x

 

W [cos(ax), sin(ax)] a

W [cosh(ax), sinh(ax)] a

W [ eax cos(bx), eax sin(bx)] b e2ax

W [ eax cosh(bx), eax sinh(bx)] b e2ax

W [sin(c(x − a)), sin(c(x − b))] c sin(c(b − a))

W [cos(c(x − a)), cos(c(x − b))] c sin(c(b − a))

W [sin(c(x − a)), cos(c(x − b))] −c cos(c(b − a))

2082
W [sinh(c(x − a)), sinh(c(x − b))] c sinh(c(b − a))

W [cosh(c(x − a)), cosh(c(x − b))] c cosh(c(b − a))

W [sinh(c(x − a)), cosh(c(x − b))] −c cosh(c(b − a))

W edx sin(c(x − a)), edx sin(c(x − b)) c e2dx sin(c(b − a))

 

W edx cos(c(x − a)), edx cos(c(x − b)) c e2dx sin(c(b − a))

 

W edx sin(c(x − a)), edx cos(c(x − b)) −c e2dx cos(c(b − a))

 

W edx sinh(c(x − a)), edx sinh(c(x − b)) c e2dx sinh(c(b − a))

 

W edx cosh(c(x − a)), edx cosh(c(x − b)) −c e2dx sinh(c(b − a))

 

W edx sinh(c(x − a)), edx cosh(c(x − b)) −c e2dx cosh(c(b − a))

 

W [(x − a) ecx , (x − b) ecx ] (b − a) e2cx

2083
Appendix O

Sturm-Liouville Eigenvalue Problems

• y 00 + λ2 y = 0, y(a) = y(b) = 0
 
nπ nπ(x − a)
λn = , yn = sin , n∈N
b−a b−a

b−a
hyn , yn i =
2

• y 00 + λ2 y = 0, y(a) = y 0 (b) = 0
 
(2n − 1)π (2n − 1)π(x − a)
λn = , yn = sin , n∈N
2(b − a) 2(b − a)

b−a
hyn , yn i =
2

2084
• y 00 + λ2 y = 0, y 0 (a) = y(b) = 0
 
(2n − 1)π (2n − 1)π(x − a)
λn = , yn = cos , n∈N
2(b − a) 2(b − a)

b−a
hyn , yn i =
2

• y 00 + λ2 y = 0, y 0 (a) = y 0 (b) = 0
 
nπ nπ(x − a)
λn = , yn = cos , n = 0, 1, 2, . . .
b−a b−a

b−a
hy0 , y0 i = b − a, hyn , yn i = for n ∈ N
2

2085
Appendix P

Green Functions for Ordinary Differential

Equations

• G0 + p(x)G = δ(x − ξ), G(ξ − : ξ) = 0

 Z x 
G(x|ξ) = exp − p(t) dt H(x − ξ)
ξ

• y 00 = 0, y(a) = y(b) = 0
(x< − a)(x> − b)
G(x|ξ) =
b−a

• y 00 = 0, y(a) = y 0 (b) = 0
G(x|ξ) = a − x<

• y 00 = 0, y 0 (a) = y(b) = 0
G(x|ξ) = x> − b

2086
• y 00 − c2 y = 0, y(a) = y(b) = 0

sinh(c(x< − a)) sinh(c(x> − b))

G(x|ξ) =
c sinh(c(b − a))

• y 00 − c2 y = 0, y(a) = y 0 (b) = 0

sinh(c(x< − a)) cosh(c(x> − b))

G(x|ξ) = −
c cosh(c(b − a))

• y 00 − c2 y = 0, y 0 (a) = y(b) = 0

cosh(c(x< − a)) sinh(c(x> − b))

G(x|ξ) =
c cosh(c(b − a))

npi
• y 00 + c2 y = 0, y(a) = y(b) = 0, c 6= b−a
, n∈N

sin(c(x< − a)) sin(c(x> − b))

G(x|ξ) =
c sin(c(b − a))

(2n−1)pi
• y 00 + c2 y = 0, y(a) = y 0 (b) = 0, c 6= 2(b−a)
, n∈N

sin(c(x< − a)) cos(c(x> − b))

G(x|ξ) = −
c cos(c(b − a))

(2n−1)pi
• y 00 + c2 y = 0, y 0 (a) = y(b) = 0, c 6= 2(b−a)
, n∈N

cos(c(x< − a)) sin(c(x> − b))

G(x|ξ) =
c cos(c(b − a))

2087
• y 00 + 2cy 0 + dy = 0, y(a) = y(b) = 0, c2 > d
√ √
e−cx< sinh( c2 − d(x< − a)) e−cx< sinh( c2 − d(x> − b))
G(x|ξ) = √ √
c2 − d e−2cξ sinh( c2 − d(b − a))
√
• y 00 + 2cy 0 + dy = 0, y(a) = y(b) = 0, c2 < d, d − c2 6= nπ
b−a
, n∈N
√ √
e−cx< sin( d − c2 (x< − a)) e−cx< sin( d − c2 (x> − b))
G(x|ξ) = √ √
d − c2 e−2cξ sin( d − c2 (b − a))

• y 00 + 2cy 0 + dy = 0, y(a) = y(b) = 0, c2 = d

(x< − a) e−cx< (x> − b) e−cx<

G(x|ξ) =
(b − a) e−2cξ

2088
Appendix Q

Trigonometric Identities

Q.1 Circular Functions

Pythagorean Identities

sin2 x + cos2 x = 1, 1 + tan2 x = sec2 x, 1 + cot2 x = csc2 x

Angle Sum and Difference Identities

sin(x + y) = sin x cos y + cos x sin y

sin(x − y) = sin x cos y − cos x sin y
cos(x + y) = cos x cos y − sin x sin y
cos(x − y) = cos x cos y + sin x sin y

2089
Function Sum and Difference Identities
1 1
sin x + sin y = 2 sin (x + y) cos (x − y)
2 2
1 1
sin x − sin y = 2 cos (x + y) sin (x − y)
2 2
1 1
cos x + cos y = 2 cos (x + y) cos (x − y)
2 2
1 1
cos x − cos y = −2 sin (x + y) sin (x − y)
2 2

Double Angle Identities

sin 2x = 2 sin x cos x, cos 2x = cos2 x − sin2 x

Half Angle Identities

x 1 − cos x x 1 + cos x
sin2 = , cos2 =
2 2 2 2

Function Product Identities

1 1
sin x sin y = cos(x − y) − cos(x + y)
2 2
1 1
cos x cos y = cos(x − y) + cos(x + y)
2 2
1 1
sin x cos y = sin(x + y) + sin(x − y)
2 2
1 1
cos x sin y = sin(x + y) − sin(x − y)
2 2

2090
Exponential Identities

eix − e−ix eix + e−ix

eix = cos x + i sin x, sin x = , cos x =
2i 2

Q.2 Hyperbolic Functions

Exponential Identities

ex − e−x ex + e−x
sinh x = , cosh x =
2 2

sinh x ex − e−x
tanh x = = x
cosh x e + e−x

Reciprocal Identities
1 1 1
csch x = , sech x = , coth x =
sinh x cosh x tanh x

Pythagorean Identities

cosh2 x − sinh2 x = 1, tanh2 x + sech 2 x = 1

Relation to Circular Functions

sinh ix = i sin x sinh x = −i sin ix

cosh ix = cos x cosh x = cos ix
tanh ix = i tan x tanh x = −i tan ix

2091
Angle Sum and Difference Identities
sinh(x ± y) = sinh x cosh y ± cosh x sinh y
cosh(x ± y) = cosh x cosh y ± sinh x sinh y
tanh x ± tanh y sinh 2x ± sinh 2y
tanh(x ± y) = =
1 ± tanh x tanh y cosh 2x ± cosh 2y
1 ± coth x coth y sinh 2x ∓ sinh 2y
coth(x ± y) = =
coth x ± coth y cosh 2x − cosh 2y

Function Sum and Difference Identities

1 1
sinh x ± sinh y = 2 sinh (x ± y) cosh (x ∓ y)
2 2
1 1
cosh x + cosh y = 2 cosh (x + y) cosh (x − y)
2 2
1 1
cosh x − cosh y = 2 sinh (x + y) sinh (x − y)
2 2
sinh(x ± y)
tanh x ± tanh y =
cosh x cosh y
sinh(x ± y)
coth x ± coth y =
sinh x sinh y

Double Angle Identities

sinh 2x = 2 sinh x cosh x, cosh 2x = cosh2 x + sinh2 x

Half Angle Identities

x cosh x − 1 x cosh x + 1
sinh2 = , cosh2 =
2 2 2 2

2092
Function Product Identities
1 1
sinh x sinh y = cosh(x + y) − cosh(x − y)
2 2
1 1
cosh x cosh y = cosh(x + y) + cosh(x − y)
2 2
1 1
sinh x cosh y = sinh(x + y) + sinh(x − y)
2 2

See Figure Q.1 for plots of the hyperbolic circular functions.

1
3
2 0.5
1
-2 -1 1 2 -2 -1 1 2
-1
-2 -0.5
-3 -1

Figure Q.1: cosh x, sinh x and then tanh x

2093
Appendix R

Bessel Functions

R.1 Definite Integrals

Let ν > −1.
Z 1
1 0 2
rJν (jν,m r)Jν (jν,n r) dr = (J ν (jν,n )) δmn
0 2
j 0 2ν,n − ν 2
Z 1
0 0

2
rJν (j ν,m r)Jν (j ν,n r) dr = 0 2 Jν (j 0 ν,n ) δmn
0 2j ν,n
Z 1  2 
1 a
rJν (αm r)Jν (αn r) dr = 2 2
+ αn − ν (Jν (αn ))2 δmn
2 2
0 2αn b

Here αn is the nth positive root of aJν (r) + brJ 0 ν (r), where a, b ∈ R.

2094
Appendix S

Formulas from Linear Algebra

Kramer’s Rule. Consider the matrix equation

A~x = ~b.

This equation has a unique solution if and only if det(A) 6= 0. If the determinant vanishes then there are either
no solutions or an infinite number of solutions. If the determinant is nonzero, the solution for each xj can be
written

det Aj
xj =
det A

where Aj is the matrix formed by replacing the j th column of A with b.

Example S.0.1 The matrix equation

    
1 2 x1 5
= ,
3 4 x2 6

2095
has the solution
   
5 2 1 5
 
6 4 8 3 6 −9 9
x1 =  = = −4, x2 =  = = .
1 2 −2 1 2 −2 2
3 4 3 4

2096
Appendix T

Vector Analysis

Rectangular Coordinates
f = f (x, y, z), ~g = gx i + gy j + gz k

∂f ∂f ∂f
∇f = i+ j+ k
∂x ∂y ∂z

∂gx ∂gy ∂gz

∇ · ~g = + +
∂x ∂y ∂z
 
i
 ∂ ∂j k 
∂ 
∇ × ~g =  ∂x ∂y ∂z 
 gx gy gz 

∂2f ∂2f ∂2f

∆f = ∇2 f = + +
∂x2 ∂y 2 ∂z 2

2097
Spherical Coordinates

x = r cos θ sin φ, y = r sin θ sin φ, z = r cos φ

f = f (r, θ, φ), ~g = gr r + gθ θ + gφ φ

Divergence Theorem.
ZZ I
∇ · u dx dy = u · n ds

Stoke’s Theorem.
ZZ I
(∇ × u) · ds = u · dr

2098
Appendix U

Partial Fractions

A proper rational function

p(x) p(x)
=
q(x) (x − a)n r(x)
Can be written in the form
 
p(x) a0 a1 an−1
= + + · · · + + (· · · )
(x − α)n r(x) (x − α)n (x − α)n−1 x−α
where the ak ’s are constants and the last ellipses represents the partial fractions expansion of the roots of r(x).
The coefficients are
1 dk p(x) 
 
ak = .
k! dxk r(x) x=α

Example U.0.2 Consider the partial fraction expansion of

1 + x + x2
.
(x − 1)3

2099
The expansion has the form

a0 a1 a2
3
+ 2
+ .
(x − 1) (x − 1) x−1

The coefficients are

1
a0 = (1 + x + x2 )|x=1 = 3,
0!
1 d
a1 = (1 + x + x2 )|x=1 = (1 + 2x)|x=1 = 3,
1! dx
1 d2 1
a2 = 2
(1 + x + x2 )|x=1 = (2)|x=1 = 1.
2! dx 2

Thus we have

1 + x + x2 3 3 1
3
= 3
+ 2
+ .
(x − 1) (x − 1) (x − 1) x−1

Example U.0.3 Consider the partial fraction expansion of

1 + x + x2
.
x2 (x − 1)2

The expansion has the form

a0 a1 b0 b1
2
+ + 2
+ .
x x (x − 1) x−1

2100
The coefficients are
1 1 + x + x2 
 
a0 = = 1,
0! (x − 1)2 
x=0
1 d 1 + x + x2  2(1 + x + x2 ) 
   
1 + 2x
a1 = = − = 3,
1! dx (x − 1)2 
x=0 (x − 1)2 (x − 1)3 
x=0
1 1 + x + x2 
 
b0 = = 3,
0! x2 
x=1
1 d 1 + x + x2  1 + 2x 2(1 + x + x2 ) 
   
b1 = = − = −3,
1! dx x2 
x=1 x2 x3 
x=1

Thus we have
1 + x + x2 1 3 3 3
2 2
= 2+ + 2
− .
x (x − 1) x x (x − 1) x−1
If the rational function has real coefficients and the denominator has complex roots, then you can reduce the
work in finding the partial fraction expansion with the following trick: Let α and α be complex conjugate pairs
of roots of the denominator.
 
p(x) a0 a1 an−1
= + + ··· +
(x − α)n (x − α)n r(x) (x − α)n (x − α)n−1 x−α
 
a0 a1 an−1
+ + + ··· + + (· · · )
(x − α)n (x − α)n−1 x−α
Thus we don’t have to calculate the coefficients for the root at α. We just take the complex conjugate of the
coefficients for α.

Example U.0.4 Consider the partial fraction expansion of

1+x
.
x2 + 1

2101
The expansion has the form
a0 a0
+
x−i x+i
The coefficients are
 
1 1 + x  1
a0 = = (1 − i),
0! x + i x=i 2

1 1
a0 = (1 − i) = (1 + i)
2 2
Thus we have
1+x 1−i 1+i
2
= + .
x +1 2(x − i) 2(x + i)

2102
Appendix V

Finite Math

Newton’s Binomial Formula.

n  
n
X k n−k k
(a + b) = a b
k=0
n
n(n − 1) n−2 2
= an + nan−1 b + a b + · · · + nabn−1 + bn ,
2
The binomial coefficients are,
 
k n!
= .
n k!(n − k)!

2103
Appendix W

Probability

W.1 Independent Events

Once upon a time I was talking with the father of one of my colleagues at Caltech. He was an educated man.
I think that he had studied Russian literature and language back when he was in college. We were discussing
gambling. He told me that he had a scheme for winning money at the game of 21. I was familiar with counting
cards. Being a mathematician, I was not interested in hearing about conditional probability from a literature
major, but I said nothing and prepared to hear about his particular technique. I was quite surprised with his
“method”: He said that when he was on a winning streak he would bet more and when he was on a losing streak
he would bet less. He conceded that he lost more hands than he won, but since he bet more when he was winning,
he made money in the end.
I respectfully and thoroughly explained to him the concept of an independent event. Also, if one is not counting
cards then each hand in 21 is essentially an independent event. The outcome of the previous hand has no bearing
on the current. Throughout the explanation he nodded his head and agreed with my reasoning. When I was
finished he replied, “Yes, that’s true. But you see, I have a method. When I’m on my winning streak I bet more
and when I’m on my losing streak I bet less.”
I pretended that I understood. I didn’t want to be rude. After all, he had taken the time to explain the
concept of a winning streak to me. And everyone knows that mathematicians often do not easily understand

2104
practical matters, particularly games of chance.
Never explain mathematics to the layperson.

W.2 Playing the Odds

Years ago in a classroom not so far away, your author was being subjected to a presentation of a lengthy proof.
About five minutes into the lecture, the entire class was hopelessly lost. At the forty-five minute mark the professor
had a combinatorial expression that covered most of a chalk board. From his previous queries the professor knew
that none of the students had a clue what was going on. This pleased him and he had became more animated
as the lecture had progressed. He gestured to the board with a smirk and asked for the value of the expression.
Without a moment’s hesitation, I nonchalantly replied, “zero”. The professor was taken aback. He was clearly
impressed that I was able to evaluate the expression, especially because I had done it in my head and so quickly.
He enquired as to my method. “Probability”, I replied. “Professors often present difficult problems that have
simple, elegant solutions. Zero is the most elegant of numerical answers and thus most likely to be the correct
answer. My second guess would have been one.” The professor was not amused.
Whenever a professor asks the class a question which has a numeric answer, immediately respond, “zero”. If
you are asked about your method, casually say something vague about symmetry. Speak with confidence and give
non-verbal cues that you consider the problem to be elementary. This tactic will usually suffice. It’s quite likely
that some kind of symmetry is involved. And if it isn’t your response will puzzle the professor. They may continue
with the next topic, not wanting to admit that they don’t see the “symmetry” in such an elementary problem. If
they press further, start mumbling to yourself. Pretend that you are lost in thought, perhaps considering some
generalization of the result. They may be a little irked that you are ignoring them, but it’s better than divulging
your true method.

2105
Appendix X

Economics

There are two important concepts in economics. The first is “Buy low, sell high”, which is self-explanitory. The
second is opportunity cost, the highest valued alternative that must be sacrificed to attain something or otherwise
satisfy a want. I discovered this concept as an undergraduate at Caltech. I was never very in to computer games,
but I found myself randomly playing tetris. Out of the blue I was struck by a revelation: “I could be having sex
right now.” I haven’t played a computer game since.

2106
Appendix Y

Glossary

Phrases often have different meanings in mathematics than in everyday usage. Here I have collected definitions
of some mathematical terms which might confuse the novice.

beyond the scope of this text Beyond the comprehension of the author.
difficult Essentially impossible. Note that mathematicians never refer to problems they have solved as being
difficult. This would either be boastful, (claiming that you can solve difficult problems), or self-deprecating,
(admitting that you found the problem to be difficult).
interesting This word is grossly overused in math and science. It is often used to describe any work that the
author has done, regardless of the work’s significance or novelty. It may also be used as a synonym for
difficult. It has a completely different meaning when used by the non-mathematician. When I tell people
that I am a mathematician they typically respond with, “That must be interesting.”, which means, “I don’t
know anything about math or what mathematicians do.” I typically answer, “No. Not really.”
non-obvious or non-trivial Real fuckin’ hard.
one can prove that . . . The “one” that proved it was a genius like Gauss. The phrase literally means “you
haven’t got a chance in hell of proving that . . . ”

2107
simple Mathematicians communicate their prowess to colleagues and students by referring to all problems as
simple or trivial. If you ever become a math professor, introduce every example as being “really quite
trivial.” 1

1
For even more fun say it in your best Elmer Fudd accent. “This next pwobwem is weawy quite twiviaw”.

2108
Index

a + i b form, 158 Bessel functions, 1501

Abel’s formula, 766 generating function, 1508
absolute convergence, 423 of the first kind, 1507
adjoint second kind, 1523
of a differential operator, 773 Bessel’s equation, 1501
of operators, 1178 Bessel’s Inequality, 1160
analytic, 304 Bessel’s inequality, 1206
Analytic continuation bilinear concomitant, 774
Fourier integrals, 1426 binomial coefficients, 2103
analytic continuation, 356 binomial formula, 2103
analytic functions, 2049 boundary value problems, 968
anti-derivative, 390 branch
Argand diagram, 154 principal, 6
argument branch point, 226
of a complex number, 155 branches, 6
argument theorem, 411
asymptotic expansions, 1113 calculus of variations, 1873
integration by parts, 1126 canonical forms
asymptotic relations, 1113 constant coefficient equation, 878
autonomous D.E., 850 of differential equations, 878
average value theorem, 410 cardinality
of a set, 3
Bernoulli equations, 842 Cartesian form, 158

2109
Cartesian product constant coefficient differential equations, 786
of sets, 3 continuity, 42
Cauchy convergence, 423 uniform, 44
Cauchy principal value, 498, 1424 continuous
Cauchy’s inequality, 407 piecewise, 44
Cauchy-Riemann equations, 310, 2049 continuous functions, 42, 432, 435
clockwise, 204 contour, 203
closed interval, 3 traversal of, 204
closure relation convergence
and Fourier transform, 1429 absolute, 423
discrete sets of functions, 1161 Cauchy, 423
codomain, 4 comparison test, 426
comparison test, 426 in the mean, 1160
completeness integral test, 427
of sets of functions, 1161 of integrals, 1342
sets of vectors, 24 ratio test, 428
complex conjugate, 152, 153 root test, 430
complex derivative, 303, 304 sequences, 422
complex line integral, 383 series, 423
complex number, 152 uniform, 432
magnitude, 154 convolution theorem
modulus, 154 and Fourier transform, 1431
complex numbers, 151 for Laplace transforms, 1363
arithmetic, 163 convolutions, 1363
set of, 3 counter-clockwise, 204
vectors, 163 curve, 203
complex plane, 154 closed, 203
first order differential equations, 661 continuous, 203
computer games, 2106 Jordan, 204
connected region, 203 piecewise smooth, 203

2110
 simple, 203 homogeneous, 636
smooth, 203 homogeneous coefficient, 645
inhomogeneous, 636
definite integral, 106 linear, 636
degree order, 635
of a differential equation, 636 ordinary, 635
del, 137 scale-invariant, 859
delta function separable, 643
Kronecker, 23 without explicit dep. on y, 802
derivative differential operator
complex, 304 linear, 758
determinant Dirac delta function, 904, 1162
derivative of, 762 direction
difference negative, 204
of sets, 4 positive, 204
difference equations directional derivative, 138
constant coefficient equations, 1035 discontinuous functions, 43, 1202
exact equations, 1029 discrete derivative, 1028
first order homogeneous, 1030 discrete integral, 1028
first order inhomogeneous, 1032 disjoint sets, 4
differential calculus, 37 domain, 4
differential equations
autonomous, 850 economics, 2106
constant coefficient, 786 eigenfunctions, 1195
degree, 636 eigenvalue problems, 1195
equidimensional-in-x, 854 eigenvalues, 1195
equidimensional-in-y, 856 elements
Euler, 795 of a set, 2
exact, 639, 801 empty set, 2
first order, 635, 649 entire, 304

2111
equidimensional differential equations, 795 alternate definitions, 1420
equidimensional-in-x D.E., 854 closure relation, 1429
equidimensional-in-y D.E., 856 convolution theorem, 1431
Euler differential equations, 795 of a derivative, 1430
Euler’s formula, 159 Parseval’s theorem, 1435
Euler’s theorem, 645 shift property, 1436
Euler-Mascheroni constant, 1490 table of, 2074, 2077
exact differential equations, 801 Fredholm alternative theorem, 968
exact equations, 639 Fredholm equations, 888
exchanging dep. and indep. var., 848 Frobenius series
extremum modulus theorem, 411 first order differential equation, 666
function
Fibonacci sequence, 1040 bijective, 5
formally self-adjoint operators, 1179 injective, 5
Fourier coefficients, 1156, 1200 inverse of, 6
behavior of, 1215 multi-valued, 6
Fourier convolution theorem, 1431 single-valued, 4
Fourier cosine series, 1210 surjective, 5
Fourier cosine transform, 1440 function elements, 356
of derivatives, 1442 functional equation, 327
table of, 2078 fundamental set of solutions
Fourier series, 1195 of a differential equation, 770
and Fourier transform, 1415 fundamental theorem of algebra, 409
uniform convergence, 1219 fundamental theorem of calculus, 109
Fourier Sine series, 1211
Fourier sine series, 1301 gambler’s ruin problem, 1027, 1036
Fourier sine transform, 1441 Gamma function, 1484
of derivatives, 1442 difference equation, 1484
table of, 2080 Euler’s formula, 1484
Fourier transform Gauss’ formula, 1488

2112
 Hankel’s formula, 1486 indicial equation, 1063
Weierstrass’ formula, 1490 infinity
generating function first order differential equation, 671
for Bessel functions, 1507 inhomogeneous differential equations, 636
geometric series, 425 initial conditions, 653
Gibb’s phenomenon, 1224 inner product
gradient, 137 of functions, 1152
Gramm-Schmidt orthogonalization, 1147 integers
greatest integer function, 5 set of, 2
Green’s formula, 775, 1179 integral bound
maximum modulus, 386
harmonic conjugate, 316 integral calculus, 100
harmonic series, 426, 457 integral equations, 888
Heaviside function, 655, 904 boundary value problems, 888
holomorphic, 304 initial value problems, 888
homogeneous coefficient equations, 645 integrals
homogeneous differential equations, 636 improper, 114
homogeneous functions, 645 integrating factor, 650
homogeneous solution, 651 integration
homogeneous solutions techniques of, 111
of differential equations, 759 intermediate value theorem, 44
identity map, 4 intersection
ill-posed problems, 659 of sets, 4
linear differential equations, 768 interval
image closed, 3
of a mapping, 5 open, 3
imaginary part, 152 inverse function, 6
improper integrals, 114 inverse image, 5
indefinite integral, 100, 390 irregular singular points, 1079
first order differential equations, 669

2113
Jordan curve, 204 linear space, 1141
Jordan’s lemma, 2050 Liouville’s theorem, 408

Kramer’s rule, 2095 magnitude, 154

Kronecker delta function, 23 maximum modulus integral bound, 386
maximum modulus theorem, 411
L’Hospital’s rule, 64 Mellin inversion formula, 1350
Lagrange’s identity, 774, 805, 1178 minimum modulus theorem, 411
Laplace transform modulus, 154
inverse, 1349 multi-valued function, 6
Laplace transform pairs, 1352
Laplace transforms, 1347 nabla, 137
convolution theorem, 1363 natural boundary, 356
of derivatives, 1363 Newton’s binomial formula, 2103
Laurent expansions, 491, 2049 norm
Laurent series, 452, 2051 of functions, 1152
first order differential equation, 665 normal form
leading order behavior of differential equations, 881
for differential equations, 1117 null vector, 12
least integer function, 5
least squares fit one-to-one mapping, 5
Fourier series, 1204 open interval, 3
Legendre polynomials, 1149 opportunity cost, 2106
limit optimal asymptotic approximations, 1132
left and right, 39 order
limits of functions, 37 of a differential equation, 635
line integral, 382 of a set, 3
complex, 383 ordinary points
linear differential equations, 636 first order differential equations, 661
linear differential operator, 758 of linear differential equations, 1046
orthogonal series, 1155

2114
orthogonality of a mapping, 5
weighting functions, 1154 ratio test, 428
orthonormal, 1152 rational numbers
set of, 2
Parseval’s equality, 1206 Rayleigh’s quotient, 1297
Parseval’s theorem minimum property, 1297
for Fourier transform, 1435 real numbers
partial derivative, 135 set of, 2
particular solution, 651 real part, 152
of an ODE, 915 rectangular unit vectors, 13
particular solutions reduction of order, 803
of differential equations, 759 and the adjoint equation, 804
periodic extension, 1201 difference equations, 1038
piecewise continuous, 44 region
point at infinity connected, 203
differential equations, 1079 multiply-connected, 203
polar form, 158 simply-connected, 203
power series regular, 304
definition of, 436 regular singular points
differentiation of, 443 first order differential equations, 664
integration of, 443 regular Sturm-Liouville problems, 1291
radius of convergence, 438 properties of, 1300
uniformly convergent, 436 residuals
principal argument, 155 of series, 424
principal branch, 6 residue theorem, 495, 2050
principal root, 168 principal values, 507
principal value, 498, 1424 residues, 491, 2049
pure imaginary number, 152 of a pole of order n, 491, 2050
range Riccati equations, 844
Riemann zeta function, 426

2115
Riemann-Lebesgue lemma, 1343 transformations
root test, 430 of differential equations, 878
Rouche’s theorem, 413 of independent variable, 885
to constant coefficient equation, 886
scalar field, 135 to integral equations, 888
scale-invariant D.E., 859 trigonometric identities, 2089
separable equations, 643
sequences uniform continuity, 44
convergence of, 422 uniform convergence, 432
series, 422 of Fourier series, 1219
comparison test, 426 of integrals, 1342
convergence of, 422, 423 union
geometric, 425 of sets, 3
integral test, 427
ratio test, 428 variation of parameters
residuals, 424 first order equation, 651
root test, 430 vector
tail of, 423 components of, 13
set, 2 rectangular unit, 13
similarity transformation, 1734 vector calculus, 134
single-valued function, 4 vector field, 135
singularity, 320 vector-valued functions, 134
branch point, 321 Volterra equations, 888
Stirling’s approximation, 1492 wave equation
subset, 3 D’Alembert’s solution, 1776
proper, 3 Fourier transform solution, 1776
Taylor series, 446, 2050 Laplace transform solution, 1777
first order differential equations, 662 Weber’s function, 1523
table of, 2066 Weierstrass M-test, 433
well-posed problems, 659

2116
 linear differential equations, 768
Wronskian, 763, 764

zero vector, 12

2117