Advanced Mathematics for
Engineering Science
_ .
and
_g ^^ViBE9Mttfl£VflHflHflHflHflHflHflHflHflHflHf
 •' J^ I^BHHHHHKHHHHHHHHHHHHHHHHHHHB^ t
\v
V\' 2 * *
Hi
HH
P^^I
Advanced Mathematics for
Engineering Science
/•^ and
by
C F Chan Man Fong D De Kee
Tulane University, USA
P N Kaloni
University of Windsor, Canada
V f e World Scientific
wb New Jersey • London • Singapore • Hong Kong
Published by World Scientific Publishing Co. Pte. Ltd. 5 Toh Tuck Link, Singapore 596224 USA office: Suite 202, 1060 Main Street, River Edge, NJ 07661 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
British Library CataloguinginPublication Data A catalogue record for this book is available from the British Library.
ADVANCED MATHEMATICS FOR ENGINEERING AND SCIENCE Copyright © 2003 by World Scientific Publishing Co. Pte. Ltd. All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the Publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher.
ISBN ISBN
9812382917 9812382925 (pbk)
Printed in Singapore.
PREFACE
The objective of this book is to provide a mathematical text at the third year level and beyond, appropriate for students of engineering and sciences. It is a book of applicable mathematics. We have avoided the approach of listing only the techniques followed by a few examples, without explaining why the techniques work. Thus we have provided not only the know how but also the know why. Equally it is not written as a book of pure mathematics with a list of theorems followed by their proofs. Our emphasis is to help students develop an understanding of mathematics and its applications. We have refrained from using cliches like "it is obvious" and "it can be shown", which might be true only to a mature mathematician. In general, we have been generous in writing down all the steps in solving the example problems. Contrary to the opinion of the publisher of S. Hawking's book, A Short History of Time, we believe that, for students, every additional equation in the worked examples will double the readership. Many engineering schools offer little mathematics beyond the second year level. This is not a desirable situation as junior and senior year courses have to be watereddown accordingly. For graduate work, many students are handicapped by a lack of preparation in mathematics. Practicing engineers reading the technical literature, are more likely to get stuck because of a lack of mathematical skills. Language is seldom a problem. Further selfstudy of mathematics is easier said than done. It demands not only a good book but also an enormous amount of selfdiscipline. The present book is an appropriate one for selfstudy. We hope to have provided enough motivation, however we cannot provide the discipline! The advent of computers does not imply that engineers need less mathematics. On the contrary, it requires more maturity in mathematics. Mathematical modelling can be more sophisticated and the degree of realism can be improved by using computers. That is to say, engineers benefit greatly from more advanced mathematical training. As Von Karman said: "There is nothing more practical than a good theory". The black box approach to numerical simulation, in our opinion, should be avoided. Manipulating sophisticated software, written by others, may give the illusion of doing advanced work, but does not necessarily develop one's creativity in solving real problems. A careful analysis of the problem should precede any numerical simulation and this demands mathematical dexterity. The book contains ten chapters. In Chapter one, we review freshman and sophomore calculus and ordinary differential equations. Chapter two deals with series solutions of differential equations. The concept of orthogonal sets of functions, Bessel functions, Legendre polynomials, and the Sturm Liouville problem are introduced in this chapter. Chapter three covers complex variables: analytic functions, conformal mapping, and integration by the method of residues. Chapter four is devoted to vector and tensor calculus. Topics covered include the divergence and Stokes' theorem, covariant and contravariant components, covariant differentiation, isotropic and objective tensors. Chapters five and six consider partial differential equations, namely Laplace, wave, diffusion and Schrodinger equations. Various analytical methods, such as separation of variables, integral transforms, Green's functions, and similarity solutions are discussed. The next two chapters are devoted to numerical methods. Chapter seven describes methods of solving algebraic and ordinary differential equations. Numerical integration and interpolation are also included in this chapter. Chapter eight deals with
id
ADVANCED MATHEMATICS
numerical solutions of partial differential equations: both finite difference and finite element techniques are introduced. Chapter nine considers calculus of variations. The EulerLagrange equations are derived and the transversality and subsidiary conditions are discussed. Finally, Chapter ten, which is entitled Special Topics, briefly discusses phase space, Hamiltonian mechanics, probability theory, statistical thermodynamics and Brownian motion. Each chapter contains several solved problems clarifying the introduced concepts. Some of the examples are taken from the recent literature and serve to illustrate the applications in various fields of engineering and science. At the end of each chapter, there are assignment problems labeled a or b. The ones labeled b are the more difficult ones. There is more material in this book than can be covered in a one semester course. An example of a typical undergraduate course could cover Chapter two, parts of Chapters four, five and six, and Chapter seven. A list of references is provided at the end of the book. The book is a product of close collaboration between two mathematicians and an engineer. The engineer has been helpful in pinpointing the problems engineering students encounter in books written by mathematicians. We are indebted to many of our former professors, colleagues, and students who indirectly contributed to this work. Drs. K. Morrison and D. Rodrigue helped with the programming associated with Chapters seven and eight. Ms. S. Boily deserves our warmest thanks for expertly typing the bulk of the manuscript several times. We very much appreciate the help and contribution of Drs. D. Cartin , Q. Ye and their staff at World Scientific. New Orleans December 2002 C.F. Chan Man Fong D. De Kee P.N. Kaloni
CONTENTS
Chapter 1 1.1 1.2 Review of Calculus and Ordinary Differential Equations 1 1 3 4 4 5 5 6 7 8 9 10 13 13 16 19 21 21 22 22 25 26 26 27 31 36 39 41 46 49 57 65 74 89 89
Functions of One Real Variable Derivatives Mean Value Theorem Cauchy Mean Value Theorem L'Hopital's Rule Taylor's Theorem Maximum and Minimum 1.3 Integrals Integration by Parts Integration by Substitution Integration of Rational Functions 1.4 Functions of Several Variables 1.5 Derivatives Total Derivatives 1.6 Implicit Functions 1.7 Some Theorems Euler's Theorem Taylor's Theorem 1.8 Integral of a Function Depending on a Parameter 1.9 Ordinary Differential Equations (O.D.E.)  Definitions 1.10 FirstOrder Differential Equations 1.11 Separable FirstOrder Differential Equations 1.12 Homogeneous FirstOrder Differential Equations 1.13 Total or Exact FirstOrder Differential Equations 1.14 Linear FirstOrder Differential Equations 1.15 Bernoulli's Equation 1.16 SecondOrder Linear Differential Equations with Constant Coefficients 1.17 S olutions by Laplace Transform Heaviside Step Function and Dirac Delta Function 1.18 Solutions Using Green's Functions 1.19 Modelling of Physical Systems Problems Chapter 2 2.1 Series Solutions and Special Functions Definitions
Xiii
ADVANCED MATHEMATICS
Power Series Ordinary Points Regular Singular Points and the Method of Frobenius Method of Variation of Parameters Sturm Liouville Problem Special Functions Legendre's Functions Bessel Functions Modified Bessel's Equation 2.8 Fourier Series Fourier Integral 2.9 Asymptotic Solutions Parameter Expansion Problems Chapter 3 3.1 3.2 3.3 3.4 3.5 Complex Variables Introduction Basic Properties of Complex Numbers Complex Functions Elementary Functions Complex Integration Cauchy's Theorem Cauchy's Integral Formula Integral Formulae for Derivatives Morera's Theorem Maximum Modulus Principle Series Representations of Analytic Functions Taylor Series Laurent Series Residue Theory Cauchy's Residue Theorem Some Methods of Evaluating the Residues Triginometric Integrals Improper Integrals of Rational Functions Evaluating Integrals Using Jordan's Lemma Conformal Mapping Linear Transformation Reciprocal Transformation Bilinear Transformation SchwarzChristoffel Transformation
2.2 2.3 2.4 2.5 2.6 2.7
92 95 99 118 120 126 126 134 141 144 158 162 170 179 191 191 192 198 213 220 227 233 233 237 237 238 242 246 253 257 258 263 265 269 275 279 282 284 287
3.6
3.7
3.8
CONTENTS
fe
Joukowski Transformation Problems Chapter 4 4.1 4.2 4.3 4.4 Vector and Tensor Analysis
289 293 301 301 301 305 327 327 333 336 336 338 340 341 342 342 347 354 357 361 366 368 379 382 391 401 401 402 402 411 414 420 420 427 427 431 435 437
Introduction Vectors Line, Surface and Volume Integrals Relations Between Line, Surface and Volume Integrals Gauss' (Divergence) Theorem Stokes' Theorem 4.5 Applications Conservation of Mass Solution of Poisson's Equation NonExistence of Periodic Solutions Maxwell's Equations 4.6 General Curvilinear Coordinate Systems and Higher Order Tensors Cartesian Vectors and Summation Convention General Curvilinear Coordinate Systems Tensors of Arbitrary Order Metric and Permutation Tensors Covariant, Contravariant and Physical Components 4.7 Covariant Differentiation Properties of Christoffel Symbols 4.8 Integral Transforms 4.9 Isotropic, Objective Tensors and TensorValued Functions Problems Chapter 5 5.1 5.2 Partial Differential Equations I Introduction First Order Equations Method of Characteristics Lagrange' s Method Transformation Method Second Order Linear Equations Classification Method of Separation of Variables Wave Equation D'Alembert's Solution Diffusion Equation Laplace's Equation
5.3 5.4
x
ADVANCED MA THEMA TICS
5.5 Cylindrical and Spherical Polar Coordinate Systems 5.6 Boundary and Initial Conditions 5.7 NonHomogeneous Problems 5.8 Laplace Transforms 5.9 Fourier Transforms 5.10 Hankel and Mellin Transforms 5.11 Summary Problems Chapter 6 6.1 6.2 6.3 6.4 Partial Differential Equations II
439 447 450 460 474 483 494 496 511 511 511 523 532 532 540 545 547 548 551 551 552 562 569 569 570 571 572 574 576 580 580 584 585 595 600 601 608 611
Introduction Method of Characteristics Similarity Solutions Green's Functions Dirichlet Problems Neumann Problems Mixed Problems (Robin's Problems) Conformal Mapping 6.5 Green's Functions for General Linear Operators 6.6 Quantum Mechanics Limitations of Newtonian Mechanics Schrodinger Equation Problems Chapter 7 7.1 7.2 Numerical Methods Introduction Solutions of Equations in One Variable Bisection Method (Internal Halving Method) Secant Method Newton's Method Fixed Point Iteration Method Polynomial Equations Newton's Method Simultaneous Linear Equations Gaussian Elimination Method Iterative Method Eigenvalue Problems Householder Algorithm The QR Algorithm Interpolation
7.3 7.4
7.5
7.6
CONTENTS
Xi
Lagrange Interpolation Newton's Divided Difference Representation Spline Functions Least Squares Approximation 7.7 Numerical Differentiation and Integration Numerical Differentiation Numerical Integration 7.8 Numerical Solution of Ordinary Differential Equations, Initial Value Problems First Order Equations Euler's method Taylor's method Heun's method RungeKutta methods AdamsBashforth method Higher Order or Systems of First Order Equations 7.9 Boundary Value Problems Shooting Method Finite Difference Method 7.10 Stability Problems Chapter 8 8.1 8.2 8.3 Numerical Solution of Partial Differential Equations Introduction Finite Differences Parabolic Equations Explicit Method CrankNicolson Implicit Method Derivative Boundary Conditions Elliptic Equations Dirichlet Problem Neumann Problem Poisson's and Helmholtz's Equations Hyperbolic Equations Difference Equations Method of Characteristics Irregular Boundaries and Higher Dimensions NonLinear Equations Finite Elements OneDimensional Problems Variational method
611 614 619 623 625 625 629 637 638 639 640 643 643 646 648 659 659 665 671 673 681 681 681 683 684 686 690 696 697 701 703 706 706 710 715 717 722 722 723
8.4
8.5
8.6 8.7 8.8
xij
ADVA NCED MA THEM A TICS
Galerkin method TwoDimensional Problems Problems Chapter 9 9.1 9.2 9.3 9.4 9.5 9.6 Calculus of Variations
724 727 734 739 739 740 742 745 747 749 749 751 752 754 756 757 760 765 772 786 790 803 809 809 809 814 817 819 823 826 830 840 842 845 851 852 853 856
Introduction Function of One Variable Function of Several Variables Constrained Extrema and Lagrange Multipliers EulerLagrange Equations Special Cases Function f Does Not Depend on y' Explicitly Function f Does Not Depend on y Explicitly Function f Does Not Depend on x Explicitly Function f Is a Linear Function of y' 9.7 Extension to Higher Derivatives 9.8 Transversality (Moving Boundary) Conditions 9.9 Constraints 9.10 Several Dependent Variables 9.11 Several Independent Variables 9.12 Transversality Conditions Where the Functional Depends on Several Functions 9.13 Subsidiary Conditions Where the Functional Depends on Several Functions Problems Chapter 10 Special Topics
10.1 Introduction 10.2 Phase Space 10.3 Hamilton's Equations of Motion 10.4 Poisson Brackets 10.5 Canonical Transformations 10.6 Liouville's Theorem 10.7 Discrete Probability Theory 10.8 Binomial, Poisson and Normal Distribution 10.9 Scope of Statistical Mechanics 10.10 Basic Assumptions 10.11 Statistical Thermodynamics 10.12 The Equipartition Theorem 10.13 Maxwell Velocity Distribution 10.14 Brownian Motion Problems
CONTENTS
xiii
References Appendices Appendix I  The equation of continuity in several coordinate systems Appendix II  The equation of motion in several coordinate systems Appendix III  The equation of energy in terms of the transport properties in several coordinate systems Appendix IV  The equation of continuity of species A in several coordinate systems Author Index Subject Index
861 867 867 868 871 873 875 877
CHAPTER 1
REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS
1.1 FUNCTIONS OF ONE REAL VARIABLE
The search for functional relationships between variables is one of the aims of science. For simplicity, we shall start by considering two real variables. These variables can be quantified, that is to say, to each of the two variables we can associate a set of real numbers. The rule which assigns to each real number of one set a number of the other set is called a function. It is customary to denote a function by f. Thus if the rule is to square, we write
y
= f(x) = x 2
(l.lla,b)
The variable y is known as the dependent variable and x is the independent variable. It is important not to confuse the function (rule) f with the value f(x) of that function at a point x. A function does not always need to be expressed as an algebraic expression as in Equation (1.11). For example, the price of a litre of gas is a function of the geographical location of the gas station. It is not obvious that we can express this function as an algebraic expression. But we can draw up a table listing the geographical positions of all gas stations and the price charged at each gas station. Each gas station can be numbered and thus to each number of this set there exists another number in the set of prices charged at the corresponding gas station. Thus, the definition of a function as given above is general enough to include most of the functional relationships between two variables encountered in science and engineering. The function f might not be applicable (defined) over all real numbers. The set of numbers for which f is applicable is called the domain of f. Thus if f is extracting the square root of a real number, f is not applicable to negative numbers. The domain of f in this case is the set of nonnegative numbers. The range of f is the set of values that f can acquire over its domain. Figure 1.11 illustrates the concept of domain and range. The function f is said to be even if f (x) = f (x) and odd if f(x) = f(x). Thus, f(x) = x 2 is even since f(x) = (x) 2 = x 2 = f(x) while the function f(x) = x3 is odd because f(x) = (x) 3 =  x 3 = f(x) (l.l3a,b,c) (l.l2a,b,c)
2
ADVANCF.n
MATHEMATICS
A function is periodic and of period T if f(x + T) = f(x) (1.14)
y ;,
RANGE
OF f
I
\
i \
I
y= f (X )
>w
"I"""
I L _
DOMAIN OF f
FIGURE 1.11
•
j ^
X
Domain and range of a function
An example of a periodic function is sinx and its period is 2K. A function f is continuous at the point x 0 if f(x) tends to the same limit as x tends to x 0 from both sides of x 0 and the limit is f(x 0 ). This is expressed as lim
X—^XQ_J_
f(x) = f(x 0 ) = lim
X—^XQ_
f(x)
(l.l5a,b) lim
x—>xo_
The notation lim
x—>x 0+
means approaching x 0 from the right side of x 0 (or from above) and
the limit as x 0 is approached from the left (or from below). An alternative equivalent definition of continuity of f (x) at x = x 0 is, given e > 0, there exists a number 8 (which can be a function of 8) such that whenever I x  xol < 8, then  f ( x )  f ( x o )  <£ This is illustrated in Figure 1.12. If the function f (x) is continuous in a closed interval [a, b], it is continuous at every point x in the interval a < x < b. (1.16)
REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 3 ^o+€ y^ ijl *oa! i !««*> ^ X FIGURE 1. . Likewise higher dx 2 derivatives can be defined and the n th derivative is written either as f ^ or dx n Geometrically. . For example. one might want to know. waiting at the corner of "walk and don't walk". x 0 ] is defined as Af Ax = f (xQ + Ax) . The derivative of f with respect to x is also denoted as 4*. is defined as f( } = *L Ax>0 Ax lim = f(x o + Ax)f(x o ) Ax*O AX lim (i 22ah) We have assumed that the limit in Equations (1.2 DERIVATIVES Continuity of a function We might be interested not only in the values of a function at various points x but also at its rate of change. The average rate of change of f (x) in an interval [x0 + Ax. which is the derivative of f at x 0 .22a. not only the position of a car but also its speed before crossing the road.12 1. dx d2f The second derivative of f is the derivative of f and is denoted by f" or . f'(x 0 ) is the tangent to the curve f(x) at the point x = x 0 .b) exists and f is thus differentiable at x = x 0 .f (x0) Ax ( i 2 i ) The rate of change of f at x 0 .
b). Cauchy Mean Value Theorem If f and g are continuous in [a. 1] is the factorial of n.28) reduces to Equation (1. such that f'(c)M b—a (1 .1). and f(x) is differentiable in the open interval a < x < b. b) such that f(b)f(a) g(b)g(a) = f'(c) ( 1 2 8 ) g'(c) If g(x) = x. that is to say.. b). (nr)!r! Rule (iv) is known as Leibnitz rule (one of them!). we deduce that if f'(c) = 0 for every c in (a. n! [= n (n .25) + ( n ) dx d% + n dx d u d^_v + dx dx"" 1 " d^ud^v r + + d%_ y d2"6) dx 11 d x r dx n " r ' " dxn where ( n ) = 1 &L— .23) (ii) _udv _dx_^x vdu_ (L2. then f is a constant.27). . as x increases f(x) increases. Conversely.4) f(uv) = u^v+vdu dx d > ) dx11 = u (1. If f'(c) > 0 for every c in (a. f(x) is a decreasing function of x.4 ADVANCED MATHEMATICS Rules for Differentiation (i) If y is a function of z and z is a function of x ^ = ^ ^ (chain rule) dz dx dx If u and v are differentiable functions of x jL(ii) (iii) (iv) = (1. Mean Value Theorem If f(x) is continuous in the closed interval a < x < b .27).27) From Equation (1. then Equation (1. b] and differentiable in (a. then f(x) is an increasing function.b). b). if f'(c)<0 for every c in(a. there exists a point c in (a. b). there exist a number c in (a..
Taylor's Theorem If f(x) is continuous and differentiable f (x) = f (x0) + (x .. the remainder term for each point x in the interval will generally be different.x 0 ) f'(x 0 ) + (X " 2XQ) where R n is the remainder term. But if the lim —— exists x>x0 x>x0 x>x0 g(x) x>xo g'( x ) lim ^ x^x 0 g(x) = lim I M x^x0 (1. = ..29) can be If lim ——1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 5 L'Hopital's Rule lim f(x) = 0 and lim g(x) = 0.is indeterminate.29) g'( x ) f'(x) f(n) (x) — exists for some value of n. the lim ~~. There are various ways of expressing the remainder term R n . R n is the result of a summation of the remaining terms. and represents the error made by truncating the series at the n * term.does not exist but lim x>x0 g(n) ^ x^x 0 g (x) replaced by to M. ta V®. Other indeterminate forms. x»x o g(n)( x ) (1.211) The simplest one is probably \n+^ ° (n+1)! f ( n + 1 ) [*o + 6(*xo)] 0212) where 0 < 6 < 1. must first be reduced to one of the indeterminate forms discussed here before applying the rule.2. + (X " J ° ) n f(n) (x 0 ) + R n (1. such as the difference of two quantities tending to infinity. Lagrange's expression which may be written as ( R f "(X 0 ) + . the lim y^r is indeterminate. The . x^x o g(x. Equation (1. Note that we are expanding about a point x 0 which belongs to an interval (x 0 .10) ffx) The same rule applies if the lim f (x) = °° and lim g(x) = °o. x). The x»x 0 x ^ x o x~*xo S( x ) rule holds for x —> °° or x —> °°. Therefore.
21 Extremum of a function f .211).x 0 . y=f(x) /\ j ^ Af\^l h h j j I h h x FIGURE 1.. if the function f has an extremum at xQ Af = f ( x o + h ) . If Af is positive. In Equation (1.f " ( x 0 ) + . where h = x .211). Figure 1. associated with the evaluation of a function.214) y< MAX. From Equation (1. f has a minimum at x 0 and if Af is negative. and if x 0 is the origin.213) (1. determines the value of 0 in Equation (1. f has a maximum at x 0 .f ( x o ) must have the same sign irrespective of the sign of h. The Taylor series expansion is widely used as a method of approximating a function by a polynomial.6 ADVANCED MATHEMATICS maximum truncation error.212).21 defines such extrema.. we are dealing with Maclaurin series. h2 (1. we have expanded f(x) about the point xQ. we see that Equation (1. Thus.213) can be written as Af = hf(x o ) + ^ . Maximum and Minimum We might need to know the extreme (maximum or minimum) values of a function and this can be obtained by finding the derivatives of the function. at different values of x within the considered interval.
c) From Equation (1. we have f'(x) = 3x 2 . Thus.. We need to consider higher derivatives until we obtain a f (n) (x 0 ) which is nonzero. f(x) does not have an extreme value at x = x 0 .3 INTEGRALS An integral can be considered to be an antiderivative. Since the derivative of a constant is zero.21. From the criteria given earlier we deduce that f does not have an extreme value at the origin.b). we see that f(O) = f"(O) = O (1.217) (1. 1. and if f ^ ( x 0 ) exists and is nonzero. as can be seen by drawing the curve given by f (x) = x 3 .219a.218 a. f has a maximum at x = x o and if f (n) (x0) > 0.215) But if f "( x o) = 0> we cannot deduce that f has an extreme value at xQ. the derivative of ^x 3 is x 2 .b. we write . f (x) has an extreme value at x 0 if n is even.b) Thus we need to consider higher derivatives and the next one f"'(0) happens to be nonzero. and an integral of x 2 is ^. f has a minimum at x = x Q .214) we deduce that the condition for f to have an extremum at xQ is f'(x o )=O The conditions for f to have a maximum or minimum at x = x 0 are f"(x o )>O. b). The origin is neither a maximum nor a minimum. Example 1. If f( n )(x 0 )<0. the general criteria for extreme values are if f(x) is defined in [a. Find the extreme values of f (x) = x3. it is a point of inflection. if they exist. an integral of f(x) is F(x). The integral we have defined is known as an indefinite integral which is usually denoted by the symbol I . = f ^~^ (x 0 ) = 0.. For example. if we know that the derivative of F (x) is f(x) [=F'(x)]. If n is odd.216) (1. Thus. F (x) is arbitrary to the extent of an arbitrary constant.x 3 . Note that we have used the article an. fhasaminimum f hasamaximum (1. but f'(x 0 ) = f"(x 0 ) = . Thus.218a.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS _J_ From Equation (1. On differentiating. f"(x o )<O. f"(x) = 6x. f'"(x) = 6 (1.b] and x 0 is an interior point of (a.
For the area to be definite.34) that (1.32) Equation (1.33a. There are tables of integrals.32) defines a definite integral.b) Thus to evaluate a definite integral analytically.31) to a definite integral if x = b. Below we list some of the general methods of integration. Integration by Parts If f and g are functions of x £(fg) = fg + fg' It follows from Equation (1.31. x = a and x = b. x = b A= /a f(x)dx (1. Carrying Q.b) where a is an arbitrary constant of integration. we usually write F(b)F(a) = Ja f(x)dx = [F(x)J a (1. In this case. the limits x = a and x = b are given. where the indefinite integrals of standard functions are given.b) define a function of x in terms of a dummy variable t.35) Example 1. we first need to find an indefinite integral. if A is the area bounded by the curve y = f(x).8 ADVANCED MATHEMATICS F(x)= I f(x)dx= I f(t)dt (1.3la.3la. Thus.X out the integration. identifying from Equation (1. Integrate I e ax sin bx dx.34) fdx = [fg]f gdx (1. Equations (1.35) e ax as f(x) and sinbx as §. The integral may be interpreted as the area enclosed by the curve y = f(x) and the xaxis. We can convert the indefinite integral in Equation (1. we have . such as. the xaxis and the ordinates x = a. We integrate by parts. we need to fix the ordinates.
36.310a. Integrate I V a 2 .a2 sin2 z ) a cos z dz (1.x 2 d x . Since x = <>(z).b) that I f(x) dx = I f [<)(z)] Q (z) dz (1.310a.x 2 dx = I (Va2 . we obtain (1. where a is a constant.38) (1.^ I e a x sinbxdx (1.32.b cos bx] J a2 + b 2 Integration by Substitution Certain integrals I f(x) dx can be easily evaluated if we substitute x by a function (j) (z) say.312) I Va 2 .311) Example 1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 9 I e ^ sin bx dx = [ " eaX ^ OSbx + J I e ax cos bx dx On integrating I e a x cos bx dx by parts again. Substitute x by asinz.b) I e ^ sin bx dx = .313a) . 7) yields 1+— Hence ( I e^ sin bx dx = — [a sin bx . so that dx = acoszdz (1.je ax sin bx] (1.39) It follows from Equations (1.36) I e ^ c o s b x d x = \~ sinbx .37) Combining Equations (1.1 [e ax cos bx] + 3. dx = 4i'(z)dz (1.
it is often simpler to express the limits of integration in terms of the new variable z. the key is to find a substitution such that the integral is reduced to a standard form. x38 5* + 2 x38 = 5x + 2 (x2)(x 2 + 2x + 4) =_A__+ Bx + C x2 + 2x + 4 (1.b) x2 where A.313b.c) =  .c) J [a 2 (1. we have z = arcsin() sin2z = 2 sin z cos z = a V 1 ~ ( f ) Thus r 2 I Va 2 .W ADVANCED MATHEMATICS = a2 I cos2z dz = y I (cos 2z + 1) dz (1.*. Integrate I 5x + 2 dx.315) a • In evaluating finite integrals. we obtain . In the method of substitution. where f(x) and g(x) are polynomials in x.x 2 + arc sin (*) 2 (1.Va 2 .316a. B and C are constants.33. J x 3 8 The function 5x + 2 can be expressed as a sum of partial fractions as follows. By comparing powers of x.313d) (1. The rational function can be expressed as a sum of partial fractions and can thus be integrated.314a) (1.x 2 dx = §.314b. Integration of Rational Functions A rational function of x is a function of the form f(x)/g(x). Example 1.[ i s i n 2 z + z] Returning to the original variable x.
c) To evaluate Ij.b.d) I 5x + 2 dx = i n (x .324a.319) The second integral on the right side of Equation (1.321a.b) ( (x + l)dx J x2 + 2x + 4 To evaluate I 2 . we obtain i Q ^ arc tan [ x + 1 \ VI VJ ' VI ' (1.b) I dx _ [ V3 sec 2 9d6 _ t J3 sec 2 9d9 j ( x + l ) 2 + 3 j 3(tan 2 9 + l) j 3 sec 2 9 Combining Equations (1.i n (x2 + 2x + 4) + 2= arc tan ( ^ l j (1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS ]± A=l. dx = V3" sec2 9 d9 = (1.x ) d x Jx38 The integral I j (1.322a.b. we let = 2 1 [ dj J Z = l i n ( x 2 + 2x + 4) 2 (1.c) f 5x+^ dx = f _dx_ + [ d .325) .318) can be evaluated as follows f (lx)dx J x2 + 2x + 4 = J (x+l2)dx j x2 + 2x + 4 = ( (x+l)dx ( 2 I" dx =_r j x2 + 2x + 4 J (x+l) 2 + 3 +2I (1.b.^.c.320a.317a. Thus B=l. dz = (2x + 2)dx (1. we make the following substitution z = x 2 + 2x + 4 . C=l (1.b) (x + 1) = V3~ tan 0 .323a.319 to 24d).2) .318) J x2 "x Zt j x2 + 2x + 4 is standard and x I ^ =*n(x2) (1.
. if f(x) is odd I f(x)dx=j f ( x ) d x +  f (x) dx Ja Ja Jb a (iv) I f(x)dx= I f(ax)dx JO Jo (1.33Oa.a (i) Ja f ( x ) d x = .326) r (ii) (b r (1.b m g(x)dx < I f ( x ) g ( x ) d x < M g(x) dx Ja Ja Ja (1.a) Ja (1.b .12 ADVANCED MATHEMATICS In the past.b) m(b . b) .b) 0. f(x)dx= \ if f(x) is even (1. Some Theorems .b .33 la.b . in the theory of complex analysis (Chapter 3) can be used to evaluate real integrals.b) (vi) Generalized first mean value theorem Under the conditions on f(x) given in (v). Currently. Contour integration. a popular approach is to resort to numerical methods (Chapter 7). considerable efforts were devoted to finding methods to express integrals in closed form and in terms of elementary functions. and for g ( x ) > 0 every where in (a.329) (v) First mean value theorem If M and m are the upper and lower bounds respectively of f(x) in (a.327) 2 I f(x)dx.a) < I f(x) dx < M(b .b) The above mean value theorems provide bounds on integrals and can be useful in error analysis.328a. f(x)dx Jb (1.
The dependent variable u is said to be a function of the independent variables x and y if to every pair of values of (x. x n is obvious.5 DERIVATIVES (1. we often encounter one variable which depends on several other independent variables. there exists a 8. the volume of gas depends on the temperature and the pressure. y0) if given e > 0.41) The domain of f is the set of values of (x. x2. y) over which f is applicable.y) (1. Thus J dx dx x x df=]im ox AX>0 f(x + Ax.y) Ay (1 5 2 ) 8f The computation of ^— is the same as in the case of one independent variable. Similarly to compute jr. y) one can assign a value of u.. their partial derivatives with respect to x and y may exist. ^— is defined as dy 9f oy = lim f(x.43) Since u is a function of two variables x and y.y + Ay40 Ay)f(x. y o )  <£ 1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 13. Here. In this case we write u = f(x. In science and engineering. The range of f is the set of values that u may have over the domain of f. 1.4 FUNCTIONS OF SEVERAL VARIABLES So far. such that whenever V(xxo)2 + (yyo)2 < 5  f ( x . we consider x to be a constant. y is treated as a constant. Since fx and f are functions of x and y. we may calculate the rate of change of u with respect to x. the extension to n variables Xj. . holding y fixed. y ) . For example. For simplicity. we shall consider functions of two independent variables x and y. we have considered functions of one variable only.y) Ax (1 5 1 ) Similarly.42) (1. f or u Y . Other notations are: ^—.y)f(x. y) is continuous at (x 0 .. The function f (x. This is the partial derivative of u with respect to x and is denoted by ^r—. .f ( x 0 . In most cases. They are defined as .
5. In an xycoordinate system.56a.b) (1. where the vector n is parallel to the line along which we wish to determine the rate of change. whereas for f the order of differentiation is reversed. We can also define and compute the rate of change of f along any arbitrary line in the xyplane.b) We note that f means taking the partial derivative of f with respect to x first and then with respect to y. Such a rate of change is known as a directional derivative and is denoted by 5—. In practice.(x.58a. If the partial derivatives are continuous. y0) along a line that makes an angle 0 with the xaxis is defined as <*L = Urn dn p_>o L f ( X ° + P C0S 6 ' Yo + P Sl" 6 ) ~f ( X ° ' y o ) J (1510) P .14 ADVANCED MATHEMATICS j(£U dx \ dx / l i m f.59) Likewise higher partial derivatives can be defined and computed.(^Ax.y) AX>0 Ax (1. the first order partial derivatives fx and f may be regarded as the rate of change of f along the x and yaxis respectively. If f is continuous then the order is not important.4) The secondorder partial derivatives are denoted as _(L] = ^Uf dx \dx ) 5x2 xx (l55a. One may wonder if the order of differentiation is important.y)f. this is generally the case and fxy = fyx (1. Thus 5— at a point (x0.=fx y (L5"7a'b) _fl)=^l=f (i.b) f ff)=^=f dy \ dx / dydx ^x d7(d7) = axay. then the order of differentiation is not important.53) a [*) dy \ dx / = lm A y>0 y^±Mzifcy> Ay (1.
we have kept y 0 + p sin 6 = y } constant. y 0 ) an p>o L p (1. y 0 ) (1 5llb) dn p>0 L P J .f(x 0 .f (x 0 . y i ) + f (xQ. y o +psin6) .51. This situation is illustrated in Figure 1. in the first two terms on the right side of Equation (1. 3f = ]im f (xQ + p cos 6. y0).51 la) Note that.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 75 where p is the distance of any point on the line from (x0.51 i i ^ x x0 Directional derivative along a line parallel to n We can rewrite Equation (1.510) as 0 II 9f _ f (x o +pcos6. y{) .51 la). y o +p sin6).f ( x Q . y i ) . y o +psin9) + f (x 0 .f (xQ. y * i i I FIGURE 1.
514a) (1. yj)+ ^—p cos 0. y + Ay) + f (x.513) y an Total Derivatives We now determine the change in u. =^sin0 dy (1.511) are applicable to Equation (1. Thus .f ( x o .514b).f (x. when both x and y change simultaneously to x + Ax and to y + Ay respectively. =^cos0 J °x (1. y!) + ^ p c o s 0 . resulting in: f (xQ. y. y0) + ^— p sin 0.16 ADVANCED MATHEMATICS This amounts to considering f to be a function of x only in the first two terms. yo) ^ p J . A (1. We deduce that f (x 0 . .514b) The observations made following Equation (1. but we also require f to be continuous. and on taking the limits Ax —> 0. Higher differentials d2f. . we obtain du = df = ^ dx + ^ dy dx dy (1. dnf can also be defined.512b) Therefore ~ =~ cos0 + ^ sin0 dn dx dy Thus if fv and fv are known. lim p>0 L y o ) + ~ p s i n 0 ..f ( x o . d3f. y + Ay) . Au. dx and dy. Then Au = f (x + Ax.f (x. y + Ay) . The differential df may be regarded as a function of 4 independent variables x. y) = f (x + Ax.b) The existence of df guarantees the existence of fx and f . Similarly.. y2) lim p>0 L P .f (x. we can compute ^r. Expanding this function of x in a Taylor series yields: f(xQ. For df to exist we require not only the existence of fx and f .512a) and similarly f (x 0 . y + Ay) .515a. one can consider f to be 9f a function of y only in the last two terms. y) (1. but the converse is not true. Ay —> 0.
. d (df A\A d f = T .—dy dx + ^ .d y dy dxldx J dy\dx j J dx\dy Jj dy \ dy J 0 0 0 „ * .d x dy + ^ . (L521) ar+d7 ^ ) d r + [dx.^ . \ . then from Equation (1. dxnrdyr y ayn (dy)» y' (1.516d) = — (dx)2 + 2 ^ ^ d x d y + — (dy)2 3x2 dxdy 9y2 It can be shown by induction that dnf = ^ (dx)n + f " ) .d s dr ds dy = ^ d r + ^ .U .516c) (1.^ L _ (dx)11"1 dy + .521) that .515).516a. then dx = ^ d r + ^ .a^ + d7 i)ds It follows from Equation (1. + ( n ) ^*..(dx)nr (dy) r+ . + ^L ax n I ! / ax^^y l r .d s 3r ds Substituting dx and dy in Equation (1.515) we have ( 1 5 18) *L = 3£dx £& (1. I df dx df dy \ . \ .520a) (1..520b) 8f 3y \ . one can rewrite it as d n f = (s d x + iH" f In Equation (1. If both x and y are functions of another variable t.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS IJ_ d 2 f = d(df) = d ( ^ dx) + d ( ^ dy) (1..517). we obtain / df 8x du = (d7 ' (1.518).5.U — d x dx + ^ . the right side can be expanded formally as a binomial expansion. d [df A . ^ (1.b) Substituting d by ^— dx + ^— dy yields dx dy A2* ^ /df .517) In order to remember Equation (1. d /df .19) v dt 3x dt dy dt If x and y are functions of another set of independent variables r and s.
3f Calculate ^ — and — . b. .522a. 3f . dr 30 (1. b. The transformation equations are x = rcos0. n . b).18 ADVANCED MATHEMATICS * dr 3u 9s = * dr **tLl dx dr dy dr (1 . In rectangular Cartesian coordinates system.= 2x cos 0 + 2y sin 0 = 2r — = 2x (r sin 0) + 2y (r cos 0) = 0 30 (1.r sin 0 . .526e. d).527c.d) .b) (1. c. we have 3 f _ 3 f 3x 3f 3y 3r " 3 x 3 r + 3 y 3r 3f=3f3x 30 3x 30 + (1. dx ~ = cos 0 .524a. c.523) y = rsin0 (1.b) 3f = 3£3x + 3f 3y ds 3x 3s 3y 3s (1522cd) Equations (1.22a . d) again express the chain rule. we obtain 3f j . 30 ?=2y dy ^ = sin 0 dr ^ = r cos 0 30 (1.527a.525a) (1525b) 3f3y 3y 30 Computing the partial derivatives yields ^=2x. Example 1.526c.d) (1. f is given by f = x 2 + y2 We change to polar coordinates (r.51.b) From Equations (1. 9).5.b) (1. dr — = .526a to f) into Equations (1.522a.f) Substituting Equation (1.525a.526a.
62).61) [Vnb]nRT = 0 (1. We now consider an implicit function written as f(x. But as can be seen from Equation (1. y and u. In theory.c) (1. y and u are not all independent. we can obtain ^ and other partial derivatives. There are examples where it is more convenient to express u implicitly as a function of x and y. in thermodynamics an equation of state which could be given as T = f (P. V is the volume and T is the temperature. We are free to choose any one of them as the dependent variable. y = V.62) partially oT with respect to P and then deduce 5= . c. The twoparameter Redlich and Kwong (1949) equation is expressed as P+ — T 1 / 2 V(V + nb) where a and b are two parameters. d) can be obtained by substituting Equations (1.524) into Equation (1. Since f = 0.62) it is not easy to solve for T.6 IMPLICIT FUNCTIONS So far we have considered u as an explicit function of x and y [u = f (x. Then rfT (1.62) and express T as a function of P and V. u=T (1. we can solve for T in Equation (1.b. implicitly as f(P. It is simpler to differentiate the expression in Equation (1.62) by partial differentiation.515). We shall show how this is done by reverting to the variables x. the resulting function will be even more complicated than Equation (1. df = 0 and from Equation (1.523) and thus f is expressed explicitly as a function of r and 0 and the partial differentiation can be carried out.u) = 0 In terms of the thermodynamic example.64a. and finding ^ r or (say) will be time consuming. In many cases. we can think of x = P. For example.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS _£9 Equations (1. R is the gas constant and n the number of moles.VT) = 0 where P is the pressure.y. it implies solving a cubic equation. we obtain .527a. the substitution can be very complicated.63) The variables x. Even if we solve for T. y)]. V) is usually written. b. 1.
g(x. the variables x. Differentiating f and g with respect to x.66) I~(S)'(I) (1.610) we essentially have two equations involving three variables.65) is still true for f = constant.63) and using the chain rule (Equations 1.b) du The partial derivative ^ c — is obtained from Equation (1./ 5 ox I \ du (1.y. since df = 0 in this case also.67) 3u du We could equally obtain 5—. we have (1. (1.65) by putting dy = 0 (since y is kept as a constant). b).6lla) (16llb) . So 5=ox Similarly 5 .69) If. y and u are related by another equation written as (1.68) It then follows that id)/(l) Similarly ^— and higher derivatives such as can be computed. in addition to Equation (1. Thus taking u as the dependent variable and differentiating partially with respect to x. u) = 0 (1. We choose the only independent variable to be x.522a. v~ by differentiating f from Equation (1.63).65a.20 ADVANCED MATHEMATICS df = ^ d x + ^ d y + ^ d u = O 3x dy du Equation (1. yields <* + «* %L+* ^ = o dx dy dx du dx 3 g + 9 g 3 y + a g du 3x dy dx 3u dx = 0 (1.
u) Differentiating Equation (1. (174) Choosing a = 1. y.b.612b) J = fy g y fu (1.73) . y. b) form a system of two algebraic equations involving two unknowns ^^—.u*) = a n f(x.61 la.74) becomes xH+y+udl=nf(x'y'u) (L75) . au) = a n f ( x .y.c) (171) Equation (1. 3x dx 1.612c) v o &u ' 3y 9u Thus for ^— and ^— to exist. we write x* = ax.J—Jr— By* da du* da =na n l c( (1.73) with respect to the parameter a yields df dx dx* da df dy df du + — .612a) fx fy (1. u* = au (1.71) becomes f(x*. which we indicate by a star (*).+ n .72a. u) Defining new variables. u) .7 SOME THEOREMS Euler's Theorem A function f (x.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 27 Equations (1.y*. The solutions are dx and (1. Equation (1.. ay. the Jacobian J must not vanish. f(x. u) is a homogenous function of degree n if f(ax. y*=ay. y.
. u) (1. y. y 0 + Ok) and 0 < 9 < l .77) The remainder term Rn is given by R = n __L(n+1)! \ h n+1 ^— I + hn k 1JL + 9x n+l + k n + 1 ^—^ dyn+l (1 78) f ^ } 3xnay The derivatives in Equation (1.75) is x ^ + y ^ +u^f (x.75) is known as Euler's theorem. 1. Equation (1. y.y)dy (1.78) at the point (x 0 + 0h.. otherwise we would be trying to differentiate by a constant.22 \ * ADVANCED MATHEMATICS Note that we substitute (= x) etc. + _L (h» ^L + (n\ h nl k __lL_ + + (n) hnr kr _ 3 ^ _ + .. into Equation (1.76) Taylor's Theorem Taylor's theorem for functions of one variable can be extended to functions of several variables.81) . + k" ll\ + R (1..r + 1) f (x..74) before setting a = 1. ( n . u) = n ( n . y) is a function of two variables x and y and we may integrate the function with respect to y holding x fixed. We then obtain an integral which is a function of x and we may consider x as a parameter.8 INTEGRAL OF A FUNCTION DEPENDING ON A PARAMETER The function f (x.y0) + {h^ + k  } + l j h 2 g + 2 h k ^ + k 2 g j + .1) .. y 0 ) and those in Equation (1. The generalization of Equation (1. For simplicity we give the formula for two independent variables. Thus if we integrate f(x.. .y0+k) = f(x0. f(x0+h.77) are to be evaluated at the point (x 0 . we have IW = Ja f(x. y) between two fixed points y = a and y = b.
Using the results in Section 1.5.83) Note that if we were integrating at another point on the same curve. I may be treated as a function of three variables x.82) Ja If the limits of integration are not fixed but are functions of x.87a. identifying y with t and v (x) with x.c.86a. v(x) and u(x). via Equation (1.y)dy = f[x. we have 1 =   I Ju (x) f(x. Equation (1.31) yields dv f x ^ p =)a f(t)dt = ^[F(x)F(a)] =F'(x) = f(x) (l. v(x)] (l.85) yu(x) Note that u(x) and v(x) are not fixed! To evaluate the partial derivative —. the limits of integration would read u(xi) and u(x2) which is equivalent to integrating from a to b. dL is given by dx di_ dx = a i _ + a ] [ d v + ai_du dx dv dx du dx (1.b.U(X) (1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 21 Differentiating I with respect to x results in (1. we fix the variables x and u(x). 5— and v~.d) Therefore. y)dy . we integrate with respect to y from a point on a curve given by y = u (x) to a point on another curve given by y = v (x) /•v(x) I(x)= f(x. we have.84) 31 dl 91 From the definitions of the partial derivatives 5—.82) /•v(x) (1.b) Similarly .
24 ADVANCED MATHEMATICS !?.81.^  (1.812b) that .=f[x.88) After appropriate substitution. Equation (1.(x) (x) (1.81 shows a projection in the xyplane of the integration path.89) is known as Leibnitz rule.810) From Equation (1.81 lb) =~ . u(x)] oil (1.8llc) On integrating directly 2 1 = Mr L = ^T JX (1.812a.81 la) = \~\ +2x 4 x 2 (1.b) / z z It follows from Equation (1. Let I be given by /•y=x2 1 =1 xydy / y=x Calculate dL by dx (a) (b) using Equation (1.89) r2 A =I JX y dy + (x) (x2) (2x) . Example 1. (1.84) becomes (1. integrate and obtain I explicitly as a function of x and then differentiate.89) /u(x) Equation (1. Figure 1.89).
The highest order derivative occurring in the differential equation determines the order of the differential equation.D. economics. engineering. Otherwise it is nonlinear. The degree of a differential equation is determined by the power to which the highest derivative is raised. There are instances when it is dx not possible to evaluate I explicitly and one has to use Leibnitz's rule.D.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 25 dL dx = 5xl_3x^ 2 2 (1. it is an ordinary differential equation (O. yf y = x2 x FIGURE 1.81 1.E. it is a partial differential equation (P. A differential equation is linear only if the dependent variable and its derivatives occur to the first degree.E. and other fields of applied science are expressed as differential equations.E. Many laws and relations in science. If only one independent variable is involved.810) ORDINARY DIFFERENTIAL EQUATIONS (O. .DEFINITIONS A differential equation is an equation involving one dependent variable and its derivatives with respect to one or more independent variables.) and if more than one independent variable is involved.812C) Both methods result in the same expression for dL ? as they should.9 Integration described by Equation (1.) .D.).
Examples of firstorder differential equations are separable equations. standardized. For each of these types. Starting at Section 1. 1. we can easily verify that f (x) = e ax is a solution of the equation y' = ay. (y is a function of x only) of order one (y' is the highest order derivative) and of degree one (y' is raised to the power one) and is linear.111. a straightforward integration will yield a result. f'(x) = a e a x = y' and the right side (ay) is of course af (x) = ae ax . Solve y dx .10.E.26 ADVANCED MATHEMATICS For example.D. linear differential equations.19. there exists a known. we look at the modeling problem.11 SEPARABLE FIRSTORDER DIFFERENTIAL EQUATIONS For the case where M is a function of x only and N is a function of y only. as follows I M(x)dx = .102) Equations of this type occur in problems dealing with orthogonal trajectories. Indeed. of degree three and is nonlinear. 1. and chemical reactions.^ = 0 * y (1. A function f(x) is a solution of a given differential equation on some interval.111) (1.I N(y)dy Example 1. There are several types of ordinary differential equations.112) (1. leading to the solution of several of the types of ordinary differential equations encountered in practice.10 FIRSTORDER DIFFERENTIAL EQUATIONS The standard form of a firstorder differential equation is as follows M(x. y) (1. Dividing both sides of the equation by x 2 y results in the appropriate form ^ . procedure to arrive at a solution. y' = 5y is an O.x 2 dy = 0. In Section 1.101) or y'=^H N(x. homogeneous linear equations. we summarize the approach. y) dy = 0 (1. (y " ) 3 + y = x is of order two.113) . etc. decay. if f(x) is defined and differentiable on that interval and if the equation becomes an identity when y and y ^ are replaced by f(x) and f^(x) respectively. For example. exact differential equations. y) dx + N(x. growth.
\ that is (1.1111) (1.12 HOMOGENEOUS FIRSTORDER DIFFERENTIAL EQUATIONS (1.3 xy + Y' is an example of a homogeneous polynomial of degree two. y) in Equation (1.117) y = ce"1/x Example 1. . In a constant volume batch reactor.1110) (1. then the substitution y = ux or x = vy will generate a separable firstorder differential equation.^ =kf(cA) Solve Equation (1.116) (1. the rate of disappearance of reactant A can be given by . x 2 .101) are homogeneous polynomials of the same degree. This can be written as inyinc = . y) and N(x.1 is not a homogeneous polynomial. dcA .REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 27 Integration yields iny = i n c .112.114) in£ or =\ (1.^ =kdt i n cA =kt + i n c in^=kt c A = ce~ kt 1. x + y .118) If M(x.118) for the case where f (cA) = c A .119) (1.1112) (1.115) (1.Y where c is the constant of integration.
23) Ml^l x(l+u) (L12 .u 2 x y Replacing u by — .i n ( l ..8) 1+u (1. 4 .u 2 ) ^r = l .2 u .^ where P = 1 .1213) (1. Solve (x + y ) d y .2u .2 u .122) g.u 2 ) i n ex"2 = i n ( l .1211) (1. Therefore incx = .1212) (1.5) (U2_6) vdx/ x (du) + u = x + ux £ ^ (1+u) \dx/ *[^) = ^"=l?Uu2 Idx/ ^ = x (1+u) .^ .( x .u2.y ) d x = O dy (xy) dx ~ (x+y) The substitution y = ux defines y' as (1.1210) .(dj) x + 0 ) I 1 We now have x(du)+u=^iH = a. we finally obtain (1.2 u .dn l_2uu2 (1+U) d127.121) (1.28 ADVANCED MATHEMATICS Example 1.121.129) Integration yields inx + inc = I .
1219. we manage to do away with the constants Cj and c 2 .REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 22 ^=l2y_y! x x x or x22xyy2 = c (1.1216) yields d(Y+p) d Y _ a1(X + a ) + b 1 ( Y + p ) + c 1 d(X + a) = dX = a 2 (X + a) + b2(Y + p) + c 2 .1221a.1214) (1. We are left with the homogeneous equation .1215) NOTE: An equation such as dy a1x + b 1 y + c 1 (1.b) dX = a2X + b 2 Y + a 2 a + b 2 p + c 2 (1.18) with solutions x = a and y = p.1217.18) we then obtain a situation where the directions of the lines are preserved (aj andfyremain unchanged) and the coefficients Cj vanish.Y ajX+bjY+aja + bjP + Cj (1. we perform the following change of variables JX=xa {Y=yp Substitutions in Equation (1. Since the coordinates of this point of intersection are (a.1222) where the underlined terms add up to zero by virtue of system (1.1217. If we translate the origin of the coordinate system to their point of intersection. b j . Think of the numerator and the denominator as representing two intersecting straight lines. c 1? a^ b 2 and c 2 are constants can be reduced to a homogeneous equation if through a change of variables. P). p) of the system { ajX +bjy +Cj =0 a 2 x+b 2 y + c2 = 0 (1.1216) dx ~ a 2 x + b 2 y + c 2 where a}.20) (1. that is to the solution (a.
2 x y . £ .1227) A problem arises if the determinant (ajb 2 .p respectively. The final solution is thus given by x 2 .1224) dY = ——— X — Y using the substitution Y = u X.1223) determining the solution (a and (3) of system (1.30 ADVANCED MATHEMATICS dY a l X + b l Y d X ~ a 2 X + b2Y The solution of such a reducible equation is thus obtained by i) ii) solving equation (1. the coefficients of x and y in the linear equations are multiples of one another. A.18).y 2 .^4 dx i) x+ y. iii) We now substitute X and Y in the solution of part (i) by (x .1223) by x . that is. Introducing a variable z = ajx + bjy yields a relation of the form — — = f and f is a function of z only. (1. That is to say.bja 2 ) = 0.122. This requires the determinant al bl *0 a2 b2 and iii) replacing X and Y in the solution of (1.7 = constant (1. . ii) Next we determine the values a and (3 by solving the system [xy = 3  x + y = 1 (1.26) We obtain a = 2 and P = .2) and (y + 1).a and y . This leads to Equation We first solve rrr (1. the two lines are parallel and do not intersect.1223) [a firstorder homogeneous equation]. Solve Example 1.1217. a separable firstorder equation. + Y d.l .1 QA.1215).6 x + 2y = c .1225.
2 8 i n ( z + 9 ) = .1234) This equation can be solved by separating the variables to yield where c' is a constant.1228) (1. For example.13 (1.= J Let z = 2x .1236) (1. Solve dy _ 2 x .1232) = 2 " 7 (^^r) or dJ = f^x 3 z .123.3 y .2 1 y .1235) (1.7 y + 9) = c' x . Replacing z by 2x .1230) ^2x7y + n =2 7Ux2iyiJ a1231) (1.x + c' (112"33) (1.7 y + 1 dx~ " 6 x .2 1 y . could have been obtained by differentiating an implicit function.7 y + 9) = c TOTAL OR EXACT FIRSTORDER DIFFERENTIAL EQUATIONS A given differential equation.1229a.1237) 7 x .7y then dz dx = (1.4 i n ( 2 x .REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 37 Example 1.2 8 i n ( 2 x .7y yields the solution to the problem 6x21y28/en(2x7y + 9) = x + c' or or 1.b) 27^ dx (1.131) .l Note that 2 7 1 6 = I2}. one can determine by inspection that the equation xdy + ydx = 0 results from expanding (1.
9) The following test allows one to determine if the equation Mdx + Ndy = 0 (1. Equation (1.134) An equation such as x 2 dy + 2 x d x = 0 (1.12) So M and N are partial derivatives of the same function F. of at least order two.= .136) (1.133) or y = f (1. assuming that F and its partial derivatives.101) could be represented by dF (x.138. That is.2 i n ex x2 (1.135) can be solved by inspection. Furthermore. after multiplication by an appropriate "integrating" factor. This can be written as f Therefore dx+^dy=0 (U3_1O) M= and N = J £ (1. are continuous in the region of interest. multiplication by e^ results in the following total differential equation d(x2ey) = 0 thus and x 2 ey = c y = i n .1311. We suspect that the equation is exact.132) (1.101) is a total (or exact) differential equation.° .22 ADVANCED MATHEMATICS d(xy) = O Integration yields xy = constant (1.137) (1. y) = 0.1313) . In this particular case. we note that dxdy ~ dydx That is to say d F _ 3 F 2 2 (1.
= .4xsiny .4 s i n y Therefore J £ = 3x 2 6xy +4cosy and F = I (3x2 .3x2y + 4x cos y + f(y) (1. as illustrated next.1319) (1. That is y .[ x 3 .131. a "partial" integration with respect to x (keep y constant) yields F. y) = constant.y (1.1318) (1.1321) or 3x24xsin y+f(y) = 2y3x24xsin y .1314). Equating that function to a constant yields the solution F(x. which will disappear on differentiating with respect to x.3 x 2 .= M.6xy + 4cosy)dx + (2y3x 2 4xsinyJ)dy = 0 ^=6x4siny 4g. Solve (1.1317) (3x 2 .3x 2 y + 4 x c o s y + f(y)] = 2y .6 x .1320) r)F f (y) is now determined by substitution of the previous equation for F in 3— = N . That is to say F =  M d x + f(y) 3F f (y) can then be determined from 3— = N . One way of determining F is as follows: since •?.1315) Example 1. (1.y (1. The "constant" of integration will in general be an arbitrary function of y. we know that we are dealing with the total derivative of a function of two variables.1316) (1.1322) .REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 31 If our starting equation satisfies relation (1.6xy + 4 cos y) dx = x3 .
Then 5 — = 0 .3 x2y + 4 x cos y + y2 .J f(y) = y 2 .101) becomes IMdx + INdy = 0 which is exact if j . Therefore. and Equation (1.1325) If Equation (1. the integrating factor I can be determined and Equation (1.i n y + c Hence. (i) I is a function of x only.1329) becomes 1 dl _ dM/dydN/dx I dx N ^ (1.1328) dy or I \ dx dy dy j ~ dy dx (1.(IM) = ^ That is T3M A/r ai T (1.1327) 3 N XT ai dx dx (1. we can try to make it exact by multiplying with an integrating factor I.101) is not exact. Equation (1. the solution is x3 .3d ADVANCED MATHEMATICS f'(y) = 2 y . .1330) is also a function of x only.1329) In genera] it is not easy to find I but there are special cases when there is a standard procedure to obtain I.1324) (1.1326) is exact and can be solved.1330) The assumption that I is a function of x only implies that the right side of Equation (1.1323) (1.Jin y = constant • (1.1326) (IN) (1.
leading to a solution.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS £5 (ii) I is a function of y only.2xydx = 0 by finding an appropriate integrating factor. we have ( 3 x V 4 .1338) is exact.y2) dy .131. (1. we can proceed as in Example 1. ^=2x dy 3N »x In this case the equation is not exact.1331) The right side is a function of y only and allows I to be determined.1334) • 3 .± \dx dy J 2xy y (1.b) is a function of y only. we note that .1333) fiN  ! ] / M = &L = .1335a.1339) (1.1332) (1. we obtain liL=_4 ^ dl = _ 4 1 d dy y I y (1.1332) by the integrating factor y~ . we have _ 3N/9x3M/3y i" dy ~ M { dI (1.2xy~3 dx = 0 Equation (1.= 6x (1.1336a. Then as in (i). ^ = 2xy3 3x F = x 2 y~ 3 + f(y) (1. From Equation (1.1331).b) (1.1340) .132.1338) (1. Example 1.1337) I = y" 4 Multiplying Equation (1.y~2) dy . Solve (3x2 .
141.146. Solve y'2xy = x I(x) = exp I 2xdx = e x p .36 ADVANCED MATHEMATICS ^ = 3x 2 y. The integrating factor I (x) is given by I(x) = exp I P(x)dx (1.[ y l ] = Q(x)I(x) and (1.145) (1. 7) .b) _y2 (1.4 + f = 3x 2 y" 4 y.2 dy dy Therefore df dy = (1.1342) (1.144) Example 1.141) by I(x) yields a left side which is equal to — .14 LINEAR FIRSTORDER DIFFERENTIAL EQUATIONS The standard form is given by ^ + P(x)y = Q ( x ) (1.1343) f = y!+c The solution is F = constant => .143) r /x y = I"1 Q®I($)d$ + c (1.b) 1.141) Note that P and Q are not functions of y.142) d(Iy) Multiplying both sides of Equation (1.1344a. Direct integration produces the solution as follows ^ .x V ^ + y"1 = c (1.^ (1.1341a.
2 y = cex • Another procedure which will generate a solution for Equation (1.y e x + c (1.1411) to yield the homogeneous solution yh. We then propose the solution of Equation (1.2 x e .141) to be of the form of y h . B and C involved in the following firstorder reactions ki k2 A > B > C are written as dc A dt ~ dc B ~df " (1. where we replace the constant of integration by a function of x.143. This method is known as the method of variation of parameters.149) Integrating yields I .x 2 y = x e~x2 or .142.148) (1. The rate equations for components A.( y e " x 2 ) d x = I x e ~ x dx x2 ye and finally 1 x2 .1414) (1. Example 1.1415) ICA klcA"k2cB .x V .^ .1413) x (1. as illustrated in Example 1.1410) 1 x2 = .141) is as follows: first one solves the homogeneous equation y' + P(x)y = 0 (1.p (ye" x2 ) = xe~ x 2 (1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 37 Multiplying both sides of the equation by e~x yields e ..
y'^T X "T 1 (U4"22) Solve = (X+1)3 (1.1414) has been treated in Example 1.1416) We wish to solve for the time evolution of the concentration c B .1417) Combining Equations (1. The integrating factor I(t) = e k 2 t .A° .143. 17) leads to ^+k2c B = k lC A ( e. Multiplying Equation (1. Equation (1.1420) The constant c is evaluated from the condition c B = 0 at t = 0.1418) by I (t) leads to ^tae^kiC^efe*)* Integration yields k c e(kr~kl)t . +c k2k1 (1.1423) The homogeneous equation . c A = c A and c B = c c = 0. Note that the independent variable is now t.1421) Finally we obtain / e kit_ k 2 t\ c^k^hpir) Example 1.1415. That is 0 = k1 CA k2kj l f° +c (1.k ' t (114_18) which is of the form of Equation (1.3& ADVANCED MATHEMATICS dt 2 B (1.141) with P(t) = k 2 and Q(t) = kx c A e~ klt .112 and yields CA = %e"klt (1. given that at t = 0.1419) c B e k2t = (1.
1428) (1.1426) We now propose the solution of the given problem to be of the form y = u (x + I) 2 where u is a function of x which is to be determined. The general solution y = y h + y . Substitution of this solution.1429) = U ( x + l ) 2 + c](x+l)2 (1.151) .15 BERNOULLI'S EQUATION The standard form of this equation is given by y ' + P ( x ) y = Q(x)y n (1.1431) The second term on the right side (that is choosing c = 0) is a particular solution y . in the given problem results in the following expression u'(x+l)2 + 2 u ( x + l ) .1424) v' 2 i .1425) (1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 22.1427) (1. y'^y = 0 J\.2 u / X + 1 ) (x + lj Note that since (x+1) 2 is the homogeneous solution.1430) or y = c ( x + l) 2 + Y ( x + 1)4 (1. we can write u as u=y(x+l)2 + c The solution to the problem is thus given by y = (x+1) 3 (1. the terms in u have to cancel. We then obtain u'=x + l and on integrating.= —=— v x+ 1 results in the solution or yh = c ( x + l ) 2 (1. 1. "l 1 (1.
153) reduces now to the following linear firstorder equation (1.n i y dx dx dx Equation (1. ( y 1 " n ) = ( l .1510.152) ^ S " + P(X) y l " = QW (1.40 ADVANCED MATHEMATICS This equation can be reduced to the form of a linear firstorder equation through the following substitution y 1 "" = v First one divides both sides of the equation by yn. we obtain ~idx~~ V = X °r dx" + 4 v = ~4x (1.159) Let v = y.151. This yields the following equation (1.5) (r^r)i.158) Dividing by y5 yields y"5^y"4 = x Q.13) This equation can now be solved with the following integrating factor I(x) = e 4 ^ d x = e 4 x (1.153) Then let v = y1"". Solve = xy 5 (L156> (1. Example 1.159 to 11).1512.y (1.X (1.n ) P(x)v = ( l .+ ( 1 .1514.n ) y . 15) . This means that dv = A. 11) Combining Equations (1.n ) Q(x) which is linear in v.+ P(x)v=Q(x) or ^ .4 => 4^ = 4y" 5 pdx dx (1.154.157) ^ .
161) If yj and y 2 are two linearly independent solutions of the homogeneous equation L(y) = 0. That is L (cl yx + c 2 y2) = Cj L(yj) + c 2 L(y 2 ) = 0 As mentioned earlier. the general solution to L (y) = Q (x) is given by y = yh + y p (1.16 SECONDORDER LINEAR DIFFERENTIAL EQUATIONS WITH CONSTANT COEFFICIENTS The standard form is given by y " + A y ' + By = Q(x) where A and B are constants. C j y j + c 2 y 2 is also a solution where Cj and c 2 are constants.166) (1.164) (1.162) (1.= _x+ \ y4 4 c (L15"17) + ce4x (1. then by the principle of superposition.1516) Substituting y~ for v produces + y4e4x = .161) is L(y) = Q(x) where the linear differential operator L (y) is defined by L(y) = y " + A y ' + B y The homogeneous differential equation is L(y) = 0 (1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 4± to yield ve 4x = _ x e 4x + 1 e4x + c (1. An alternative form of Equation (1.1518) 1.b) .163) (1.x e 4 x + l e 4 x and finally i.165a.
ii) two equal real roots (the case where D = 0).1610). Since we are dealing with a second order equation.1611) iii) two complex conjugate roots (the case where D 2 < 0) y h is now given by yh = Cj e(a+ib>x + c 2 e(aib>x (1. We propose y h to be of the form e a x . In this case. In this case. (Xj and 0C2. we have that a 2 + Aot + B = 0 (1. y h is given by y h = c1e0ClX + c 2 xe aiX (1. using the Euler formula to yield .1612) where (Xj 2 = y ± i z = a ± ib This complex solution can be transformed into a real one.1610) The second order characteristic equation could have 2 2 i) two real and distinct roots (the case where D = A . Substitution of this function and its derivatives into the homogeneous equation yields e ax (oc 2 + A a + B) = 0 Since e a x * 0 .169) (1.168) (1. we also have two constants of integration and the solution y h is given by yh = Cleaix + c 2 e a2X (1.4B > 0).42 ADVANCED MATHEMATICS Solving the homogeneous equation y"+Ay'+By = 0 will generate y h . y h is given by equation (1.167) This is known as the characteristic or auxiliary equation. This equation has two solutions.
3 y ' + 2y = x 2 + 1 The characteristic equation a2 .1621) .1622) (1.1623) (1.3 b + 2c = x 2 + l Equating the coefficients of x 2 yields 2a = 1 and a= y (1. This solution technique is referred to as the method of undetermined coefficients.3 ( 2 a x + b) + 2(ax 2 + bx + c) = x 2 + 1 2a . Solve (1.3b + 2ax2 + 2bx + 2c = x 2 + l 2ax2 + x ( 2 b .1616) y " . (1. Example 1. we propose for y the quadratic polynomial ax 2 + bx + c.1617) + c2e2x (1.1620) (1.1618) Since the right side of the equation is of order two.1614. This can be done by proposing a solution.6 a ) + 2 a .161.1624) (1. Substitution of the proposed form and of its derivatives into the equation to be solved followed by equating the coefficients of the like terms allows one to determine the values of the introduced constants as illustrated next. b and c. the right side of the equation.b) (1. and proceed via substitution in the given equation to determine the values of the constants a.3 a + 2 = 0 has the following roots: a 1 = 1 and a 2 = 2 so that yh = Clex (1. based on the form of Q(x). y p = ax 2 + bx + c y p = 2ax + b yp'= 2a Substitution yields 2 a .1619) (1.1625a.1613) (1. 15) Next we have to determine the particular solution y .REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 43 y h = e M (c 3 cos bx + c 4 sin bx) where c 3 and c 4 are constants and real.6ax .
..1635) (1.1629) or c= (1. 28) Equating the coefficients of x° yields 2a .1636) ... B * 0. (b) Q (x) = sin qx or cos qx or a combination of both. (1.. A * 0. which in turn depends on L(y).1626) (1. Q(x) = q0 + q lX + q 2 x 2 + . + a n x n ) (1.44 ADVANCED MATHEMATICS Equating the coefficients of x yields 2b6a = 0 2b = 6a = 3 and b =  (1. Examples are (a) Q (x) is a polynomial. If B = 0..b. try y p = x (a 0 + axx + .1627a. + a n x n if A * 0. + q n x n Try y p = a 0 + a lX + .1630a.1633) If both B and A are equal to zero.1634) (1..1631) and the solution to the problem is y = y h + yp = Cl e x + c 2 e 2x + j (2x2 + 6x + 9) (1.3b + 2c = 1 l+2c y is thus given by v2 y (1.1632) The form of the y to be chosen depends on Q (x) and on the homogeneous solution y^. then the solution can be obtained by direct integration.b) = l 3 + Q + P = 2 2X 4 (1...
A mass balance on A leads to the following differential equation which we wish to solve J9 ^AB CA dc A 2 I / v \ CA I i{C .1643) in standard form as follows . then y is y p = x(a o e ( l x ) If xe qx is also part of the homogeneous solution.431 U .i o is) We can write Equation (1. The concentration of A is given at the inlet (z = 0) of the reactor by c A and at the outlet (z = L) by c ^ . Levenspiel (1972) describes a flow problem with diffusion.1637) If cos qx (or/and sin qx) is not a solution of the homogeneous equation.o \ z ' dz A ~ (] 16. The average velocity in the tubular reactor is (v z ). The particular integral will be of the form y p = aoe«lx if e^* is not a term in y h .1642) (1.1639) We can also determine the particular solution for a combination of the cases examined in (a to c) by intelligent guess work. Example 1. We need to find the number of undetermined coefficients that can be determined to fit the differential equation. 181).162. If e1 x occurs in y h . involving a firstorder chemical reaction.143.1641) (1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 41 In this case an appropriate y is y p = a 0 cosqx + b 0 s i n q x (1.1638) (1. The particular solution can also be obtained by the method of variation of parameters (see Examples 1. then y p = x 2 (a 0 e1x) (1. If cos qx (or/and sin qx) is present in y h then we need to try y p = x (a 0 cos qx + b 0 sin qx) (c) Q(x) = e ^ .1640) (1.k c A . The reaction is characterized by the rate expression — ^ = .
1647) (1.1644) JJ^ We now propose a solution of the form cA = e a z .st dt (1. The Laplace transform is linear. substitution leads to the characteristic (or auxiliary) equation a2^ a^=0 (1.1645) The result in terms of the concentration cA is given by cA = c 1 e a i z + c 2 e a 2 z (1. The Laplace transform L[f(t)] of a function f(t) is defined as L[f(t)] = F(s) = I f(t)e. That is to say .1648) This leads to the following final result /cALcA eaiMl cAIcA eaiL ^ A° eaiz+ ^ Ao ea2z ea2LeaiL ea2LeaiL c A  cA A° (1.46 ADVANCED MATHEMATICS ^_KideA_^_ dz ^ ^ az C A = 0 (1.17 SOLUTIONS BY LAPLACE TRANSFORM Linear differential equations can sometimes be reduced to algebraic equations. which are easier to solve.171) Jo where the integral is assumed to converge. A way of achieving this is by performing a so called Laplace transform. As before. 2 = .^  1 ±A / I + ^ ® A B The constants Cj and c 2 are evaluated from the conditions cA = CAQ CA = CAL at z = 0 at z = L (1.1646) where a.1649) 1.
the procedure to follow is i) ii) iii) take the Laplace transform of the equation. For a given differential equation.. (1.. to obtain the solution f (t).at f(t)] = F(s + a) (1. solve the resulting algebraic equation. This is usually done by consulting tables of Laplace transforms.. that is to say.. c 2 are constants. Next.173) ii) Finalvalue theorem lim f(t) =lim s F(s) t»oo s>0 (1.6) (1. where Cj.177) _ Jo Convolution is used when E (s) does not represent the Laplace transform of a known function but is the product of the Laplace transform of two functions whose transforms are known. The Laplace transform of a variety of functions are tabulated in most mathematical tables. Table 1.171 gives some useful transforms.] = CjLffjCOj + ^Lf^Ct)] + .. obtain F(s).174) iii) Translation of a function L[e. we state some theorems without proof.172) invert the transform.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 47 Lfcjf^O + c / ^ t ) + .17. i) Initialvalue theorem lim f(t) = lim s F(s) t—»0 S—>«> (1. ..175) iv) Derivatives of transforms L[tnf(t)] = ( _ l ) n ^ Z M ds n v) Convolution E(s) = F(s)G(s) =L fl I f(tu)g(u)du (1.
2.48 ADVANCED MATHEMATICS TABLE 1. 0<t<k e~x^ s _a e" k s /s^.^ ^ V n ^ i ) .r s 2 + a2 snF(s)_ysnk sin at cos at ^ d^f(O) d t tl dt" rt F(S) j£ F(s) I f(t)dt s j= exp (x2/4t) —— exp (x 2 /4t) 2t 3/2 erfc (x/2VT) o Y(fj" exp (x V7) Vft exp (x Vs ) L exp (x Vs~) x V~s~ (t + ^ ..) erfc (x/2Vt~) . t>k a"J n ( a t ) (Vs2 + a 2 ...s ) n —. +l _1_ sa — r s 2 + a2 — .x i/L exp (x 2 /4t) 2 V 7t 5 S2 leat(e"xVierfc[^__^T)l+e^erfcr^^+f^ylj 2 I L2 VT J L2VT Jl (0. }X>0 ( ( t . n>l 4J7^ .171 Laplace Transforms Function I t" eat Transform 1 s JlL sn n =l.
17lld) An arbitrary function f(t) can be shifted over a distance a. t>0 H (t) is discontinuous at t = 0. This will result in the quantity f (t .171 illustrates this translation graphically. This can be expressed as [ 0.179a) Jo = [e. H(ta) = { t<a (1.a). .s t ]°° = Je~sa (1.a ) .REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 49 Heaviside Step Function and Dirac Delta Function The Heaviside step function is a function which is equal to zero for °o < t < 0 and is equal to 1 for all positive t.1710a.179b) (1. by multiplying f(t) by H ( t .b) I 1.b) I 1.st ]°° L S (1.171 la) L[H(ta)] = I e~ st H(ta)dt Jo = f e~stdt •/a (1. t>a (1.179c) J Q = \ We can generalize the Heaviside step function for the case where the discontinuity occurs at t = a. The Laplace transform of H (t) is given by r L[H(t)] = I e~stH(t)dt (1. H(t) = °°<t<0 (1.a) H (t . In this case.171 lb) = [e. Figure 1.178a.17llc) (1. we have I 0.
1712d) Jo = e. we obtain r L[f(ta)H(ta)] = I e"stf(ta) H(ta)dt JO (1.50 ADVANCED MATHEMATICS fit) > /fit) I _ / / /f(ta) H(ta) o a t FIGURE 1.s a F(s) where F is the Laplace transform of f.1712C) (1.171 Translation of a function f (t) over a distance a Taking the Laplace transform.1712a) = I e~ st f(ta)dt Ja (1.1713) (1.s ( t l + a ) f(t')dt' JO = esa I e.1712b) r°° = I e.1712e) .a (1.st1 f(t')dt' (1.1712c) we have used the transformation t' = t . In Equation (1.
1716b) (1. if x < 0 (1. c) need not to be (oo.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 57 The Dirac delta function is defined as 8(t) = 0. via Equation (1. L[8(t)l = L J e~ st 6(t)dt (1.1714a) I J —oo 5(t) dt = 1 (1.1717) indicates that 8(x) is not an ordinary function.1715c) The limits in Equation (1.1715a) Jo . = H(x) ifx>0 (1.31) 8(x) = && (1. they could be (Ej.1714b) I J — oo f(t)5(t)dt =f(0) (1.1716c) where H (x) is the Heaviside step function.1717) Equation (1.1716a) = 1. everywhere except at t = 0 (1.1714c) The Laplace transform of 8(t) can be obtained using Equation (1. e2) for any positive e l> e 2  Thus I J oo 8(t)dt = 0.oo = I J oo e~ st S(t)dt (1.1714c).1716c) we obtain.1714b. By formally differentiating both sides of Equation (1.1715b) = 1 (1. oo). The derivative dK is zero in any interval that dx . because H(x) is not continuous at x = 0 and therefore does not have a derivative at x = 0.
a) dt = = I /aEj /aEj f (t) 8(t .a) to be an impulse at t = a and this will be considered in the next section.d) L[8(ta)l = I L[8(ta)l JO e~ e~stst8(t 8(ta)dt .1718e. We will illustrate the usefulness of the Heaviside and Dirac functions in Example 1.171.a) dt = e"sa sa (1. s 2 + a2 Example 1. If the discontinuity is at t = a.52 ADVANCED MATHEMATICS does not include the origin.1719) Referring to Equation (1.f) (1.1718b) f °° f a+e2 I /—oo /—oo f (t) 5(t .174 and in Problems 35a and 36b. Applying the definition.171.a cos at] 9 9 s 2 + a2 0 Evaluation of this term for t = °° and t = 0 yields the solution — .— as given in Table 1. Example 1.a) = = 0 . we may regard the derivative of the Heaviside step function to be the Dirac delta function.1718e. Compute the Laplace transform of sin (at).1718g) (1.1718a) I J _oo oo = 1 1 5(ta)dt = (1.a ) = $& (ta) 8(ta) ^ (ta) We may interpret 8 (t . everywhere except at t = a (1. we write f°° L[sin (at)] = I e~ st sinatdt JO (1.1718c. For formal computational purposes.172. Solve the following second order differential equation .a) dt = = f (a) (1.f) (1.1718g) 8 ( t .39) we observe that the integral on the right side is given by . oo e st [s sin at . then 8 (t .
171.^ . F(s) equals .h ' ( 0 ) = s 2 H(s) (1. we obtain TT/ .1721c.1720) subject to the conditions h = 0 at t = 0 and ^ = 0 at t = 0.s + X2s2) (1. Using Table 1.b) f dt h > sH(s)h(0) = sH(s) (1.1722b) f(t) > Substituting into Equation (1.1724) can then be written as H(s) = ^ 2 = ^ Q 5(85^(882) s (1 + aX.d) » H(s) F(s) (1. H(s) = X F(s) ^— .1720) yields the following algebraic equation s 2 H(s) + a§.1726) This can be further expressed in terms of partial fractions.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 21 ^ dt 2 +» * X dt 1 X2 h = f(t) (1.1725a.b) where s{ 2 = laX±X V(a 2 4J \/2X (1. s2 1 +a>s + X (1.s h ( 0 ) .1724) To invert H(s) requires knowledge of F(s) and therefore of f(t).1723) X2 Solving for H(s). which are easier to invert. say c 0 .H(s) + ^ = F(s) X (1. That is .1721a. For the simplified case where c f(t) is assumed to be a constant. Equation (1. we determine the following Laplace transform ^4 dt 2 » s 2 H ( s ) .1722a) (1.
subject to the condition c B = 0 at t = 0. In particular.16 iii). a critically damped response is obtained.1730a. Sj and s 2 are complex and one has an oscillatory response (see Section 1. Invert using convolution.s 2 ) H(s)  i = S2 C° (1.1729a.s 2 ).1726)].1727) becomes »2 2 + (1. s = s 1 and s = 0.S) and s and setting in turn s = s 2 . by Laplace transform. Example 1.1727) respectively by (s .173.b) 2 H(s) = A ^ L S Sj S 2 .1733) .S j ) S2 ( S 2 . For a = 2. B and C by multiplying Equation (1. This leads to C = (s . The Laplace transform of the equation is sC B (s) + k 2 C B (s) = k j c ^ j ^ ^ ) (1.1418).S 2 ) ( S . Solve the kinetic expression given by Equation (1.b) vs2 "" s\) B = (s. if a > 2.S j ) ( S .1732) ( s i ~ S2^ S 2 (S2 ~SF The behavior of h(t) depends on the values of Sj and s 2 and therefore on the values of a and X [see Equation (1. (s .1728a. If a < 2. A 'o + .b) A = s H(s)  s = 0 = ^ Equation (1.1731) Sj (Sj . ^ ^ (1.S l )H(s) = s C° l * • 1 "" 2^ (1. Sj and s 2 are real and h changes with t in a nonoscillatory fashion.S 2 ) Inversion results in h (t) which is given by 2 h (t) 2 A si 2 = ^ o + s l S2 °o e5'^ ^ C° es^ (1.1727) We determine the constants A.51 ADVANCED MATHEMATICS H(s) = A + _ B _ + _ O _ S S — Si S — So (1.
^ d u (l.tu(k9ki) 1 2 v du (1. we write cB(t) = klCAo f ' e ^ ^ e . Determine the shear stress in the material as a function of time.1734) Using convolution (Equation 1. We assume the relation between the shear stress (T ) and the shear rate ( y v x ) to be given by x +Xx =uV (1.1420) Example 1. 1WA ° . The shear is given by Yyx = Y 0 H(t) Using Equation (1.1735b) Jo = k1cA()ekit ( e ^ ^ d u (1.Q are the viscosity and relaxation time respectively.l735a) Jo = klCA() I e k. and the dot over the quantities denote differentiation with respect to time.1717).1736) x yx +A 0 T y x .^i) (l1735d) o yk2~KV (1.17350 Jo k c e~ k l l e "" (k2 " kl) l = rl \ \ (k2 . A slab of a viscoelastic material was at rest and at time t = 0.174.n yyx (I. as illustrated in Figure 1. it is suddenly sheared by an amount y 0 . we have Yyx = Tb»(t) (1 (1. N (s + k 1 )(s + k 2 ) (1. X.177).REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 55 k c C B (s) = B W .1737) 1M8) .172.
1738) into Equation (1. . is given from Equation (1.1740) Taking the Laplace transform of Equation (1.1739) and assuming that t T yx (s) + V T y x ( s ) = ^Yo T (s). we get (1. the Laplace transform of x .1736).56 ADVANCED MATHEMATICS I  7 1 7 yf 1_ X 0V / V / FIGURE 1.1742) Thus the shear stress decays to zero exponentially.172 Shear deformation of a slab Substituting Equation (1.171 to be t y x (t) = ^ ^ Xo e~t/X° (1.1739) = 0 at t = 0. we obtain Tyx + A < o V = JLt y0 6(0 (1.1741) The inverse of T (s) can be seen from Table 1.1740) by T (s) = =^2 X0(s + l/X0) (1.
except at t = x.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 57 1. t) is the Green's function for this problem. Once a Green's function has been obtained for a given operator and a set of boundary conditions. Thus the Green's function is equivalent to the integrating factor in the firstorder equation. as a differential equation in t. G is continuous at t = x. That is G = G t= x_ t= x + (1. subject to the conditions y = 0 at x = xQ and x = Xj can be written as y(x) = P G(x. This requires that during the process of constructing G.164). x and t. The Green's function is a function of two variables. In Equation (1.t)f(t)dt (1.182) G satisfies the boundary conditions. The solution of Equation (1. The Green's function for any x has to satisfy conditions such as i) G(x. we regard the differential equation which G has to satisfy. The solution of a nonhomogeneous boundary value problem can be obtained if the Green's function G is known for the homogeneous equation. The boundary conditions will be applied at t = x0 and t = Xj . Our purpose in introducing the Green's function now is to show the generality of the method.181). then the solution to the boundary value problem can be written as an integral.181) where G (x.18 SOLUTIONS USING GREEN'S FUNCTIONS The Green's function method is a powerful method to solve boundary value problems and can be used not only for ordinary differential equations but also for partial differential equations and integral equations.161) with Q(x) replaced by f(x). t) is a solution of the homogeneous Equation (1. That is to say d_G_ + A dG_ + BG dt 2 ii) iii) dt = o (1. we integrate with respect to t so as to obtain a solution y at point x.183) . That is G = 0 at t = xQ and t = x 1 .
t) = G(t. x) Example 1.1.182). Having obtained the solution to Equation (1.189) (1. We try a solution of the form y p = a (x) cos x + b (x) sin x where we replace the constants Cj and c 2 by functions of x.186) for any f (x). We can then show that the Green's function satisfies conditions (i to v). We then calculate yp = a(x) sin x + b(x) cos x + a'(x) cos x + b'(x) sinx (1.184) More generally.187) (1. dG dt t=x + dG dt =1 t=x_ (1.58 ADVANCED MATHEMATICS iv) The first derivative is discontinuous at t = x.186). The solution to the differential equation is given.185) (1. in general. by Equation (1.181.186) The particular solution y p can be obtained via the method of variation of parameters. Having dx 2 obtained the Green's function allows us to write down the solution to Equation (1. The solution to the homogeneous equation is y h = Cj cos x + c 2 sin x (1. v) G is symmetric. it will be possible to deduce the Green's function for the operator d2 L = \. G(x.188) . as follows. Solve the equation d2v H r + Y = f(x) dx 2 subject to y (0) = y (1) = 0 and identify the Green's function. this difference equals the reciprocal of the coefficient of the highest derivative in Equation (1.166).
The obvious condition is a'(x) cos x + b'(x) sin x = 0 Next.a ' (x) sin x + b1 (x) cos x = f (x) (1. we have = a'Wcosx sin x (1.1810) We can now use Equation (1.1815) As in Equation (1. It becomes therefore necessary to impose one additional condition. from Equation (1. Also.186) one generates only one equation to determine the functions a(x) and b(x).31). a(x) and b(x). Replacing b1 (x).1810).186). t is a dummy variable. y p contains two unknown (arbitrary) functions. we obtain . However.186) yields .1814) r a(x) = f(t)sintdt (1. Such a condition should be chosen to simplify the problem.b ( x ) sin x Substitution into Equation (1.1810) to replace a'(x) or b'(x).1811) (1.1816) On integrating.1812) becomes 2 .a ' ( x ) ^ _ ^ = f(x) sin x which reduces to a'(x) = f(x)sinx Integration yields (1. we compute y p y p = a'(x)sinx + b ' ( x ) c o s x .a ' ( x ) s i n x . Equation (1.1813) (1. One should realize that on substituting y p and y p in Equation (1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 59 Note that y p has to satisfy Equation (1.1812) (1.a ( x ) c o s x .
1817) The constants Cj andc 2 can be determined from the boundary conditions.o Jc2 This may be written as (1. a' = b'=O.c o s ( l ) l f(t)sintdt + sin(l) I f(t)costdt .1821b) = . A similar situation was encountered in Example 1. In other words. We note that when f(x) = 0.c o s ( l )  f(t)sintdt + sin(l) f ( t ) c o s t d t +  f(t)costdt JO Jc2 JO C C 1° (1. Imposing the boundary condition at x = 1 yields 0 = .c o s (1)1 f(t)sintdt + sin(l) f(t) cost dt +sin (1)1 f(t)costdt JO JO Jc2 . a and b are constants (say Cj and c 2 respectively) and we obtain y h . introducing Cj and c 2 as limits of integration in Equations (1. 17) will generate a solution y p which in fact includes y h .1819) which leads to Cj = 0.60 ADVANCED MATHEMATICS [x b(x) = f(t)costdt (1.1818) )cx h2 The boundary condition y = 0 at x = 0 implies r 0 = I f(t)sintdt (1. Thus the general solution of Equation (1.1821a) 0 = .1820) I1 \l° I1 (1.1815.143. That is.186) may be written as y = cosxj r f(t)sintdt + sinx I r f(t)costdt (1.
1 .1824a) r f(t)sin(xt)dtS u ^ c f(t)sin(lt)dt (1.1823) Combining Equations (1.1821c) Jo Thus Jc 2 f(t)costdt = rJ— f(t)sin(lt)dt (1.0 61 f(t) sin (1t) dt + sin (1) I f(t)costdt (1. Thus we write Equation (1. we obtain r y (x) = cos x r f (t) sint dt + sin x il f (t) cost dt .1 ) ~ s i n x s i n ( 1 — 0 "I d t L sin(l) J Jx[ I [ f(t) sinx sin(lt) 1 d lg sin(l) j . in order to write the solution as in Equation (1.181).1818) as y (x) = . 22).1824c) Jo sm(1) Jo r • \r fl = \ f(t)sin(xt)dt4mjV f(t)sin(lt)dt+ Jo Sm(1) [Jo Jx c Jo Jo f (t) sin(xt) dt .1824b) f(t)sin(lt)dt (1.c o s x I f(t) sint dt + sin x I f(t) cost dt + I f(t) cost dt Jcx )c2 JO (1. we wish to determine G in two regions: t < x and t > x .1824d) = 1 f (t) [ sin(1 ^ s" 1 ^.MUL r S i n ( 1 ) Jo • tl f (t) sin( 1 t) dt ..1822) Jc2 Sm(1) Jo Recall that we are in the process of determining G.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS . To achieve this. because G has several properties to satisfy at the points t = x_ and t = x + .1819.^Z Jo Jo s i n ( 1 ) Jo f (t) sin (1 t) dt (1.^ ^ S i n ( 1 ) Jx f(t) sin(1 t) dt (1.
c o s x sin ( x .182 and in Problem 37b.181. b) we can deduce that G satisfies conditions (iv. t) should be chosen from the right side column in Table 1.1825a. b). . t) is the Green's function and is given by /sirU^inCx^) sin(l) G(x. The Green's function can of course also be constructed from the condition given by Equations (1. b) From Equations (1.1825a. via Equations (1.62 ADVANCED MATHEMATICS = (Xf(t)[sintsin(xl)dtldt Jo L sin(l) J + fl ) x rsinxsin(tl)ldt [ sm(l) J ^ y(x) = I f(t) G(x. is given by Equation (1.l ) t=x _" = n is o&»\ (L18'26a) sinTI) sinx cos(x1) sm(l) t=x+ and G' t= x + .1824f) where the appropriate function G(x.l ) sin(l) Y < t < 1 Q ^ x (1. which is independent of f (t). That is to say. t) = sinx s i n ( t .G1 = sinx cos ( x . the solution to a given problem is to be written as in Equation (1.181) where G(x.1 ) .1825a.1825a. As mentioned earlier. The Green's functions for several frequently used operators and boundary conditions are given in the following table.182 to 5). b). we deduce that r1 G G. v) given earlier. _ cosx s i n ( x . t).1 ) t=x_ sin(l) = { ^ The solution to the problem therefore is given by Equation (1. t) dt (1. The boundaries in Table 1. The boundary conditions will be satisfied automatically. In particular. This exercise is addressed in Example 1.186) can now be determined for any f (x). the solution of Equation (1.181) where G (x.181 are all chosen to be x = 0 and x = 1.
x < t t.=f(x) dx 2 where T is the tension in the string.y" 2.b) .1829a.1). y " . x = 0 and x = 1.1830a.A 2 y y (0) = y(l) = 0 Example 1. then y satisfies .5 dx 2 (1.b) We now construct the Green's function for the operator .T ^ .2 L(y) = y " = . as a result of a force distribution f(x). y" Boundary Conditions y(O) = y(l) = O y(O) = y'(l) = O G (x.182.181 Green's Functions Operator L(y) l. Consider the transverse displacement of a string of unit length fixed at its two ends.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 61 TABLE 1. t) t(xl). If y is the displacement of the string from its equilibrium position. The boundary conditions are y (0) (1. y" + A2y y (0) = y(l) = 0 4.1) A sin A sinh At sinh A (x .1) A sinh A sinh Ax sinh A (t . x < t sin At sin A (x .1) A sin A sin Ax sin A (t . x>t x ( t .1828) = y(l) = 0 (1. x>t x.1) A sinh A ^ ^ 3.
181.b.x . the force acts at one point only. t) is given by G(x. c3 = . Thus f(x) will be zero every where except at the point x = £.1831) G (x.1835) (1. That is to say.1838) t<x t>x (1. = c3(x) + c4(x) t .1836a. we can consider f (x) to be an impulsive force. t<x t>x (1.1837a) (1.1832a) (1. t) is given by G(x.181) as /•I y(x)= .1833a) (1. The displacement y at any point x is then given by Equation (1.d) (1. = x(tl).l ) .1834) (1.t) = t ( x .1832b) Applying the boundary conditions on t yields Cj = 0 c3 + c4 = 0 The continuity condition implies xc 2 = c 3 + xc 4 The jump discontinuity can be expressed as c4c2 = 1 From Equations (1. which is given in Table 1. t) = C!(x) + c 2 (x)t . say at x = £.c. We write f(x) as .t) ^ dt (1. The result are Cj=O.l ) .1833 to 35) we can solve for c1 toc 4 .1833b) Thus G (x.1837b) Jo As an example. c2 = ( x . Mathematically f(x) can be represented by the Dirac delta function 8. G(x.64 ADVANCED MATHEMATICS Since G has to satisfy ft d2G ^=° (1. c4 = x (1. where it is not defined.
. 5.17. a Reynolds number. one can compare flow situations in pipes of small to very large diameters in terms of a dimensionless variable £. The main properties of the Dirac delta function are given in Section 1. For example. Draw a sketch of the problem. Jo (. it is useful to proceed according to the following steps. Formulate a model in mathematical terms. etc. That is to say.U e x p . is to consider the Dirac delta function as a functional or distribution. 2. G represents the displacement at x due to a point force applied at "t. This may reduce the number of variables and it allows for the identification of controlling variables such as.f ' G ( * .1841b) provides a physical interpretation of the Green's function. 3.^ d . we obtain y(x) = ." T 5 ( .19 MODELLING OF PHYSICAL SYSTEMS In order to generate the equation(s) representing the physical situation of interest. An alternative approach. 1.84^) (U84.b) 5%i2 Equation (1. Combining Equations (1.1838. = £ which will always vary from 0 to 1. introduced by Schwartz (1957).REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 65. for example.1840) e.1839) The delta function is not an ordinary function but it can be regarded as a limit of a sequence of functions. given by Equations (1. determine the equation(s) as well as the boundary and/or initial condition(s).5 ) (1. Indicate information such as pressures. . R Determine the limiting forms (asymptotic solutions). temperatures.1714c). It also allows one to easily compare situations which are dimensionally quite different. An example of such a sequence is lim . 4. Nondimensionalize the equation(s). Make sure you understand the physical process(es) involved. f(x) = 5 ( x . 1. 39).( M 2 e>0 VTCE L v = 5(x) (1. Both approaches produce equivalent results..
191 illustrates the problem which is to be modelled subject to the following assumptions steadystate flow. the fluid is incompressible: p = constant.191 Flow in a pipe with associated velocity profile v z (r) and shear stress profile x r z .66 ADVANCED MATHEMATICS 6. < i FIGURE 1. there are no end effects. Figure 1. Solve the equation(s). We can imagine thin concentric cylindrical sheets of liquid sliding past each other. the piece of pipe of length L which we are considering is located far from either end of the pipe which is very long. 7. that is. laminar flow: Re = — ^ — — — < 2100 and vz is a function of r only. we consider the flow of a Newtonian fluid in a pipe. Verify your solution(s)! Do they make sense physically? Do all terms have the same dimensions? As a first example.
r. The forces acting on the system (cylindrical shell of thickness Ar and length L) are: (i) the pressure force in the zdirection. and (ii) by viscous (molecular) transport. c) zmomentum in by viscous transport at r. Since the flow is laminar. and (ii) the zcomponent of the gravity force.191) is zero because of the steadystate assumption (no accumulation!).REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 67 Cylindrical coordinates are used since the geometry of the problem is cylindrical. That is zmomentum in T rz (27crL) r by viscous transport at r L o . The differential equation representing this physical situation is obtained by applying a momentum (or force) balance on a shell of thickness Ar. n. This makes it mathematically easier (less calculations) to apply the boundary conditions. That is zmomentum in (v z 27crAr)pv z  z = by convection at z = 0 b) Similarly we write zmomentum out (v z 27trAr)pv z  z = by convection at z = L Note that the quantities 2. The individual contributions making up the force balance are a) zmomentum in by convection at z = 0. with pvz. Analyzing the problem. v z is not going to change with z so that the net contribution to the force balance of the zmomentum associated with convection is zero. Ar and p are not changing. This contribution is obtained by multiplying the shear stress x rz (force per area) by the appropriate area (2 TtrL). one has to realize that z momentum is transported according to two mechanisms: (i) by convection due to bulk flow. This contribution is obtained by multiplying the volumetric flowrate associated with the appropriate area (27trAr). The balance is given as follows / rate of z \_l rate of z \ + [net sum of forces] _ Q vmomentumin/ Imomentum out/ I in zdirection / Q 191) " The right side of Equation (1.
That is pressure force at z = 0 f) Similarly we write pressure force at z = L g) P L (27rrAr) PQ (27trAr) The zcontribution to the force balance due to the gravity acting on the system (shell) is given by the weight of the fluid in the shell.192) has to be a function of r and we will have a net contribution from the zmomentum by viscous transport to the force balance. . That is. e).192) dv vz is a function of r. We established earlier that Trz is a function of r.Jg^)r Ar L + rpgKz!!L) = 0 ' ^ L ' (1.191.19. The Newtonian viscosity I is constant and therefore the left side of Equation (1.h L . f) and g) into the force balance yields t r z (27crL) r x r z (27 t rL) r + A r +(P 0 P L )27crAr + 27crpg(h 0 h L )Ar = 0 We can divide Equation (1.191) by the appropriate area (27trAr). d). e) The contribution from the pressure force is obtained by multiplying the pressure (force per area) at z = 0 (Po in Figure 1. the difference in height of the positions at z = 0 and z = L in Figure 1.193) [^^UA.68 ADVANCED MATHEMATICS d) Similarly we write zmomentum in T r z (27irL) r + A r by viscous transport at r + Ar Note that since we are dealing with a Newtonian fluid.h L ) Ar. the shear stress t r z is given by dv^ ^ dr XfZ (1. the volume (2rcrArL) multiplied by pg and by cos a (zcontribution) gravity force (27trArL) pg cos a Note that L cos a = h 0 . Substitution of the terms in c).4) Note that we did not divide by r. The gravity force can therefore be written as: 2nr pg (h 0 .193) by 2ftLAr to obtain (1. otherwise —^ would be zero.
Note that the potential drop per length I—2——I becomes the pressure drop per length when h 0 = h L (horizontal flow).1911) (1.197) is solved as follows.f ( x o ) Ay df — = lim ^.1912) V^f^L)^ . so that we are dealing with A(rx r z ).= (1.194).195) .197) Equation (1. to yield the linear shear stress profile shown in Figure 1. Taking the limit for Ar —> 0 of Equation (1.197) is the desired differential equation with the potential F defined by IP = P + pgh (1. rx r z initial lim Ax>0 (at r) .rx r z final (at r + Ar).191. we have a term involving the quantity r Trz  r .r t r z  r + A r .^£(^) + Cl (1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 69 We are in the process of setting up a differential equation relating xrz to r! Recall that the derivative of a function y = f(x) with respect to x at the point x = x 0 is defined by f(x o + A x ) .b) Ax Ax>0 AX dx In Equation (1. /F .195a.tP \ £<riy = r(^L) (1.196) r(IP^PL) = 0 (1.198) So the potential at z = 0 is given by: F o = Po + pgh0 and the potential at z = L is: P L = P L + pgh L .^ ( ^ J + P )+Pg(h0hL)] = 0 (1.194) yields Urn ( r T r z l r . that is.199) I d(ny = I r(^i)dr .r T r Z l r + A r  +j_ [(p Ar>0 \ Ar / L or via Equation (1.1910) d. The differential Equation (1.
\z = § Px^) (1. We cannot accept an infinitely large force (per area). vz = 0 at r = R.1916) allows us to evaluate the integration constant C 2 . we look at the physical situation for r = 0. Therefore Q = 0 and Trz is a linear function of r. That is.1913) dv The velocity profile is obtained by replacing the left side of Equation (1.19) H^MR2 Substitution into Equation (1.70 _ _ ADVANCED MATHEMATICS To evaluate Cx.191). (1960). Applying this to Equation (1. v z is related to the square of r (a parabola as in Figure 1. The relation for v z is usually written as (1.17 > (ii9i8) (L19. The minus sign is as a result of following the convention by Bird et al.1914) This equation is solved as follows \dv*" " F i ^ ) Irdr ^(^MT^ ( 1 1 9 15) (U9I6) The physics of the situation allows us to assume that the velocity is zero at the wall.1913) by fi — £ . We now have a differential equation relating vz to r given by (1.1920) .1916) yields Vi = ?ci) R 'Jti r > That is. ° = (^)f+c^ So " 19 .
as it should.1923) Note also that the limits of integration have to be adjusted according to the newly introduced variable £. The lower limit remains unchanged. That is. Indeed. Equation (1. as follows e2n /R I vrdrdG <vz) = *>» I r dr d6 Jo Jo (1. as it should.1920. The integral can then be written as R I (l^ 2 )^d^ = l Jo The constant R2 necessary to nondimensionalize r dr is obtained from the denominator. we consider the flow of a nonNewtonian fluid in a pipe. goes to one. at r = 0. As a second example.1921) The denominator yields n R2. We will assume the fluid to obey the powerlaw. 21) yields = \ 4. Combining Equations (1.192.b) Example 1. That is to say \z = . = £•. the new variable £. (1. ) h 1 lR>] TCR2 (1 19 _ 22) We introduce a dimensionless variable £.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 77 Note that the maximum velocity occurs at the center of the tube.^  n a1925) .0. Note that this integral is now independent of R and is valid for pipes of any radius.1922) reduces to (1. Integrating this product yields a flow rate. The numerator is the product of a point velocity (vz) and an infinitesimal area (r dr d0).1924a. The average velocity (vz) is obtained by dividing the flow rate by the crosssectional area. as the upper limit goes to R.
1926) /dv \ and the shear rate — — is given by (1. Equation (1. one should realize that Equation (1. vz = i^f^f11 1/n  r1/n dr r /n+ (1.121.1928a) p _p \ _ l l\ (fen TTT +C2 < 1 1 9 "28b) n / The constant C2 is obtained.1929) reduces to Equation (1. it is valid for any fluid. In fact.1913) now becomes (1.1930b) . We now proceed by computing the velocity distribution.1930a) 1 +3 n 2\iL J (1. using the same (no slip) boundary condition as in Example 1.1921). and the velocity profile is (1.1929) Note that for n = 1. the powerlaw fluid reduces to the Newtonian fluid. (1. The average velocity is obtained by substituting Equation (1. Equation (1.1913) is still valid.191. following the procedure given in Example 1. since no assumption as to the type of fluid has to be made to derive the shear stress distribution.1920).72 ADVANCED MATHEMATICS At this point. That is to say.1929) into Equation (1.1927) Integration yields the velocity profile.
. note that the dimensions of ji depend on n and jx has the same dimension in both examples when n = 1.1930c) yields i = 2a±jJi_(i.192. In this example. 0<JL< 1 0 1 l(if = \ R R 2 ['(it) 2 We note that for n = 1.) 1/n+1 ] n + 1 [ lR) J (vj (L1931) v ' Different values of n yield different profiles. However. For n = 0. The asymptotic solutions (linear profiles) are obtained for n = 0 and n = <x>. Vz (l.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIA!. For intermediate values of n. as n tends to zero and to infinity. the slope is zero and for n = °°. we obtain the Newtonian result (parabolic profile). To achieve this.191 and are shown in Figure 1. we continued to use parameter \JL as in Example 1. In practice. TABLE 1. These profiles are illustrated in Figure 1.1929) by Equation (1. we plot j^r versus —.192.191. \ v z) R Dividing Equation (1. EQUATIONS 73 We are now in a position to compare the velocity profiles in dimensionless form and to evaluate y asymptotic behavior.191 Velocity profiles n r . the slope must lie between 0 and 3. n is usually between 0 and 1 and the observed velocity profiles are indeed more blunt than the Newtonian one. we obtain a maximum slope of 3. The ones of interest are tabulated in Table 1.
x 2 (iii) (1 + 2 / e n x ) / x i n x 2a. Answer: r Use the chain rule to compute ^ 3b.x)°\ where a is a constant. If y = e " a x / ( l . A = 7tr2 . show that (1 x)p.74 ADVANCED MATHEMATICS — <v"z> 2 I / Vn=i / /An = oo l/^<& n=o \ 0 _r R I FIGURE 1.= ocxy . The circumference C and the area A of a circle of radius r are given respectively by C = 2rcr. Dimensionless velocity profiles Differentiate the following functions with respect to x (i) (ii) (iii) arc sin x arc cos Vl x 2 i n (x2 i n x) Answer: (i) and (ii) l/V 1 .192 PROBLEMS 1 a.
cos e x . + ax)yW_nay(nl) = 0 w ^ 1 sinhx v (iv) ' xx (Hint: in (iv). Find the limit as x —> 0 of the following functions (i) S i ^ w x (ii) xcosjc sinx (iii) v . (ii) ^[cos ( i n x) + sin ( i n x)j . Find the remainder term if only the first three terms are retained. take in) 5a. Integrate the following (i) I xe x dx (ii) I cos(inx)dx (iii) I xVx 2 +l dx (iv) I e x sine x dx J (v) f 2x2 + x + 1 . using Maclaurin's expansion. the xaxis. x = 0. If it is known that r\ is given by the rational function a + bY 2 Tl = — 6b.dx 1 1/ i\ 3/2 ) ( x .3) . and c. b. 7 a. . Answer (i) to (iv): 1 Use Taylor's theorem to obtain an expression for cos x about x = n/2. (iii) Ml + x z j (iv) . Sketch the curve y = cos x in the range 0 < X < T C . If 7 is small. such that "y4 and higher powers of 7 can be neglected.l ) 2 ( x + 3) r Answer: (i) e x (x . 1 +c72 write down the coefficients of the expansion you have obtained in terms of a.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 75 Use Leibnitz rule to show that (l_x)y("+l)_(n where y ^ = — —.I)"1 8a. and x = K. (v) i n (x2 + 2x .1).(x . What is the maximum error if we use the expansion you have derived to compute the value of cos x in the interval n/4 < x < 3n/4 ? The viscosity r\ of a nonNewtonian fluid is an even function of the shear rate 7. Find the area of the region enclosed by y = cos x. dx n 4a. express TJ as a polynomial in 7.
the wave equation . y) = x cos y + y sin x lib. Tl = XCt. The area of the surface of revolution S obtained by rotating the curve y = f(x) about the xaxis from x = a to x = b is given by 8=2 ) 1 y V"+If d x Ja Compute S for y = e~x. b. 0 < x < ° ° . c. b. In addition. The pressure P. and temperature T for 1 mole of an ideal gas are related by the equation PV = RT where R is a constant. . and T o (> 0) are constants. fy. Find the rate of change of T along x.y) = i n ( X 2 + y 2) Answer: n (V2 + sintr 1 1) (iii) f (x. The temperature T in a body is given by T = T 0 ( l + a x + by)e c z where a. 12b. Find the change in volume if both the temperature and pressure are increased by 1%. ADVANCED MATHEMATICS 9b. 10a. find the direction in which the temperature changes most rapidly at the origin. fxy for the following functions (i) f(x. y.x 2 y 2 + y 3 (ii) f(x. (a.76. volume V. and zaxes at the origin. c) Answer: n = — Va 2 + b 2 + c 2 13 a. Show that if the independent variables x and t are changed to t. Find fx. y) = x 3 . = X + Ct.
The volumetric flow rate Q of a nonNewtonian fluid in a circular tube of radius R is given by Q(R. 15 a.xR) = ^ [ R Y r z t r 2 z d t r z TR JO where t R and t r z are the shear stress at the wall and at any point in the tube respectively.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 7J_ a2u = c 3t 2 2a2u . The pressure P. can be written as . c is a constant.and ^ . 16b. (3. „ 3T . which is a function of the shear stress. By differentiating with respect to i R . 3x 2 is transformed to 35 an 14a. y ^ is the shear rate. show that the viscosity r\ (y R ). show that the shear rate at the wall YR is given by y = JM3Q + TtR3 k dQl dtJ Also. which is defined as the ratio of the shear stress to the shear rate. the volume V and the temperature T are given by p + 0LJ(Vp) = RT where a. dy Compute —^ if dx y 2 .9V Compute ^p. This is the Van der Waals' equation.sin xy + x 2 = 4 . and R are constants.
Assuming the rate constants kj and k 2 to be related as follows: kj = —. y = ^1 cos y \ 2' Answer: — = . 17 a.x 2 d y = 0 y' = sinx (given that at x = n.cos x In a constantvolume batch reactor.^ y 4 + c Answer: y = c e ' ' " x Answer: sin y = . (ii) ( y ' ) 2 . the amount of component A. x 2 dx + 3y3 dy = 0 xydx + V l . Solve the following separable firstorder differential equations 3 (i) (ii) (iii) 20b. In the reversible chemical reaction A ^_ B + C .78 ADVANCED MATHEMATICS (Q/TCR 3 ) L d(inT R ) The above result is known as the WeissenbergRabinowitsch equation. the rate of disappearance of reactant A can be given by ^ kC n "~kCA dt Solve for C A given that C A = C A o at t = 0. Determine the order and degree of 2 (i) ^ + 3 ^ = cosx dx 2 dx 18a. Discuss the obtained result for the cases n >1 and n < 1. show that .k 2 . broken down at time t is represented by %. ^ = k1(A0X)k2X2 with A o representing the initial concentration of chemical A. ^ i 21b. 19a.x y ' = 0 Verify that e~2x (cj cos x + c 2 sin x) is a solution of y " + 4 y' + 5y = 0.
The volumes of the components are V A (t) and V B (t). Solve the following linear firstorder differential equations . At t = 0. — ^ .REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 79 given the initial condition that % = 0 at t = 0.S) dt + t dS = 0 Answer: x 2 + 2 x y = c Answer: y + Vx 2 + y 2 = c x 2 Answer: t exp i / ^ " = c A mixture of liquids A and B is boiling in a vessel. That is to say. (x + y) dx + x dy = 0 x dy . dt Show that dt H BW p i n vA(t) V A(°) = a i n [«VA(t) + (P«)v B ( t )' [a VA(0) + (p .8y . Solve the following exact firstorder differential equations (i) (ii) x dx + y dy = (x2 + y2) dy y exy dx + x exy dy = 0 Answer: i n (x 2 + y 2 ) = 2y +c Answer: exy = c 26a. 22a.6x + 1) = c Answer: 4x .3x + 2) dx = 0 (2y + x + 1) dx = (2x + 4y + 3) dy Answer: 2 (y + x) .). the initial volumes are V A (0) and V B (0).i n (2y . 1M> _ pvB(.i n (4x + 8y +5) = c 25a.a) VB (0) 24a.= a VA(t) dt The evaporation of component B is related to the evaporation of component A as follows £M> . Solve the following homogeneous firstorder differential equations (i) (ii) (iii) 23b. The evaporation of component A is proportional to the volume V A (t) of A.3x) dy + (y .y dx = Vx 2 + y 2 dx (2 VST . Solve the following reducible firstorder differential equations (i) (ii) (y .
28a.OAol y / Oo (t')dt where the dummy integration variable t' is interpreted as a time in the past. y " .— e x 6 Viscoelastic behavior can be described via a Maxwell model given by where T yx is the shear stress.+ (6x + 21) y = 3 (x + 8) 2 y 5/3 Answer: y~ 2/3 = (x + 8) [1 + c ( x . Show. y ' + ^ y = x.(x~2 + x 2 ) 4 y' .^ + 2xy = . and Xo is the relaxation time. Solve the following Bernouilli equations (i) (ii) . Solve the following secondorder differential equations with constant coefficients . Yyx is the shear rate.80 ADVANCED MATHEMATICS 0) (ii) 27b. t' < t. That is.5 y ( t ) = 0 (subject to initial condition y (0) = 2) y"(t) + y (t) = 2 (y (0) = 0 and y1 (0) = 3) Answer: y (t) = 2 e 5t Answer: y (t) = 2 + 3 sin t .— g 2x 29b.x y 4 2 (x2+ 7x .2 cos t . Discuss the particular cases where k2 » k^ and k 2 « k i .142 by computing the time evolution of c^.4 y ' + 7y = e 2x y " + 4y = 4 cos 2x Answer: y = e 2x (cjcos V3 x + C2sin V3 x)+ — Answer: y = (c^ + x) sin 2x + c 2 cos 2x Solve the following differential equations via Laplace transform (i) (ii) y ' ( t ) . Complete Example 1.l + c e 3 x (i) (ii) 31a.— . \i is the viscosity.7y = ex Answer: y = c e 7x . Answer: y~ 3 = . A y(l) = O Answer: y = ]. using the appropriate integrating factor that the Maxwell model can be written as x = _ \]L e ( l .8) p.1)] 30a.
P b is given by d2k dr where k = 2h 6i dk R 2 p dt A ^\Pg/ 3gk _ 2L p is the density of the fluid. Q is the volumetric flow rate and p is the density. the mass balance reduces to Wi s w O s = 0 At time t = 0. is given by the following unsteadystate macroscopic mass balance ^ t o t = Aw where m tot = V p 0 (t) is the total mass. Solve the following system of differential equations via Laplace transform y i = yi + 3y 2 y"2 = 4y14et for the following initial conditions: y^O) = 2. dt 34a. Answer: w o (t) = (w i s + A W i )+ ^ e ~ 5 l (w Os . and where w = p Q. Determine w o (t). describing the motion of a manometer fluid. the mass flow rate at the outlet.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 8± 32b. L is the length of the manometer fluid.P34. Answer: yj = e l + e 2t y 2 = e2t 33a. at t = 0. R is the manometer tube radius. [i is the fluid viscosity. The differential equation. y\ (0)= 3. y 2 (0) = 2. illustrated by Figure 1 .Awj) . Determine the relation between k and t. subject to a sudden pressure difference AP = P a . At steady state. The conditions on h at t = 0 are: h = 0 and ^h = 0. a step change AWJ is imposed. y2(0) = 1 . for a step change in AP.w i s . and 2h is identified in Figure l. The differential equation governing the mixing process.P33. using the Laplace transform method.
P33 Motion of a manometer fluid 35a. . Consider separately the cases when t > a and t < a. a shear of magnitude Yo w a s suddenly imposed. it was deduced that the constitutive equation of a Maxwell fluid can be written as v . Using the properties of the Heaviside and Dirac functions. deduce the shear stress Tyx at time t. In Problem 27b. at time t = a.82 ADVANCED MATHEMATICS P I FIGURE l.f f<=(tt')A»v<t'>d'1 A Maxwell fluid has been at rest and.
a > 0 . Construct the Green's function.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS £2 wL I t ) 1 •Wo It) V FIGURE l. y = e~l + e a . That is to say (i) (ii) determine the solution to the homogeneous equation for the cases t < x and t >x. 37b. Answer: t > a. by applying the conditions given by Equations (1.182 to 5). (iii) . Mixing process at constant volumetric flow rate Solve the differential equation £ + y = «(. (i) (ii) by the Laplace transform method.a) with y(0)= 1.186). y = e.P34 36b.1 Consider the cases t > a and t < a separately. use the jump discontinuity condition of Equation (1.184) to determine the remaining equation needed to solve for the constants that were generated in step (i). needed to solve Equation (1. use the boundary and continuity conditions to determine the relations which have to be satisfied by the constants. by finding the integrating factor.t t < a.
(1960). f(x) = a o 8 ( x .181. 0<X< 1 y(0) = y ( l ) = 0 by the Green's function method. A firstorder reaction A —> products takes place in a tubular reactor of radius R and length L in the z direction. Its two ends x = 0 and x = l are fixed. Assume that y is of the form y (x.1.181).84 ADVANCED MATHEMATICS Then solve Equation (1. If y is the vertical displacement of the string. co is a constant and t is the time.x cos x 2 sin 1 2 38 a.t) = 0. Answer: L c o s } sin x . Choose the appropriate Green's function from Table 1. The tension is T. A uniform string of unit length and of mass m lies along the xaxis. Answer : y = — e x . then y satisfies the following equations oa y cz — 3x 2 C2 dy 9t 2 = f (x) cos cot = T/m y(O.1644). aQ is a constant and 8 is the Dirac delta function. to generate the .t) = y (l. Perform a mass balance over a differential element of thickness Az. Solve for h using the appropriate Green's function. Solve the boundaryvalue problem y"_y = ex.186) for f (x) = sin x. Hence determine y for the following f (i) (ii) 40b. t) = h (x) cos cot and determine the differential equation and the boundary conditions that h has to satisfy. using Equation (1.e s m n x J 2 2 sinh 1 39b.l / 2 j . appropriate expression for the flux N A z defined by Bird et al. Differentiate N A z with respect to z to generate Equation (1. Verify that the Green's function you have chosen satisfies conditions (i) to (v). f(x) = x. A periodic force f (x) cos cot is applied to the string.
Show that this can be done in the following way. As the air penetration in the liquid film depends on the film thickness. we wish to establish a relation between the film thickness 8 and the velocity V. up to a wall C as illustrated. In order to accelerate the aeration of a Newtonian fluid. The belt has a width W and a velocity V. (i) Perform a momentum (or force) balance over a length L to obtain the following shear stress distribution txz = pgxcosp List all assumptions made.P41 Aeration of a fluid . (ii) Combine Newton's law with this equation and show the velocity profile to be Vz = _v+pgcosP(g2_x2) (iii) Calculate the flow rate r f8 Q = v z dxdy Jo Jo and deduce from the physics of the situation that 62 = _3MV_ p g cos P FIGURE l. The belt transports a laminar film of liquid.P41.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS §A 41b. a continuous belt has been introduced in the fluid reservoir. as shown in Figure l.
the sensible heat required to cool the droplet from To to TM is negligible compared to the latent heat of fusion. The constant k is the thermal conductivity. note that the heat flux qr is linearly related to the temperature gradient ^J. In the design of spray dryers.86 ADVANCED MATHEMATICS 42b. one is interested in knowing the time required to solidify the liquid droplets. (ii) (iii) Perform an energy balance on the solidified spherical portion and show that the temperature profile is given by TTo T —T ls l0 = R'f1!1 1 1 R^R1 In performing this calculation. show that the time required to solidify the droplet is given by tf = f \6 k /lRJi + l[p R A f i f S/^ToTj where AHf is the latent heat of solidification per unit mass.P42. as follows dr 4r dr This is Fourier's law. Estimate the time tf required to solidify a droplet of radius R if (i) the droplet is initially at the melt temperature T o and the surrounding air is at T^ as shown in Figure 1 . the heat transfer coefficient h at the solid gas interface is constant. . Show that the heat loss to the surrounding air is given by 9 47tR2h(TnTj Given that the heat loss at the liquidsolid interface (Rf) is pAH f 47tRf —j£.
. T° \ \ \ " T °° / /LIQUID FIGURE l.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 87 T .P42 Solidification of a droplet .
.
The form of this series depends on the nature of the point xQ.'s.D. If the coefficients of the differential equation are not constants.E.11) can be written in standard form as y " + p 1 y 1 + p2y =b where pj = aj/a 2 .E.12a) is homogeneous. Since the derivatives of exponential functions are also exponential functions.168. Higher order O.12a) can be written as y = yh + yP ( 2 i3) where y^ is the homogeneous solution and y p is the particular integral. The point x 0 is an ordinary point if both pj and p 2 are analytic at the point xQ.x 0 ). If b is zero. we consider second order O. Equation (2.181 and Section 2.E.c. The solution of Equation (2. We recall that a . The particular integral (or particular solution) is usually obtained by the method of variation of parameters which is described in Example 1.'s).d) (2. we have seen that second order linear differential equations with constant coefficients admit a solution in the form of an exponential function. p 2 = a Q /a 2 . we develop methods of solving O. 9)].12a) (2.D. the differential equation reduces to an algebraic equation [see Equations (1. The solution of the homogeneous equation in the neighborhood of a point x 0 is assumed to be given by a power series in (x .12b.E. then the solution is not of an exponential form.CHAPTER 2 SERIES SOLUTIONS AND SPECIAL FUNCTIONS 2.D. we reviewed several of the standard techniques used to solve ordinary differential equations (O. we have only the homogeneous solution.E. As in Chapter 1.11) We recall that if b is zero.'s with variable coefficients. b = a/a2 (2.D. In particular. Equation (2.D. can be written as a 2 (x)y" + a 1 (x)y' + a o (x)y =a(x) where a2 (x) * 0 and ' denotes differentiation with respect to x. In this chapter. A second order O.5.'s can be solved by the same method.1 DEFINITIONS In Chapter 1.
iii) The standard form of this equation is given by (2.b.15b) (2. i) This equation is in standard form and P! = x . P2. ii) The standard form of this equation is given by ""•TV+M^T)^0 p=^T^23^).15a) (2. xQ is a singular point. i) ii) iii) y"+xy' + (x24)y =0 (xl)y" + x y ' + i y =0 x 2 ( x .14a) (2. c). From Equations (2. «"••» (217a) <2'™.cd) Pl is not analytic at x = 1 and P2 is not analytic at x = 0 and at x = 1.12b. Analyze the following equations for ordinary and singular points. The singular points are therefore at x = 0 and at x = 1 and all other points are ordinary points. then x 0 is a singular point. That is to say.11.P2) is not analytic at x 0 . and b and by determining the presence or absence of singular points.15c) We proceed by evaluating the functions p^.2 ) 2 y " + 2 ( x . f(x) can be represented by the series oo f(x) = X a n (xx 0 ) n n=0 (2. Example 2. it can be seen that if a2(x 0 ) is zero and either a^ (x 0 ) or a o ( x o ) is nonzero.16a. and b = 0 (2.14b) an = f ( n ) ( x 0 ) / n ! If one or the other (or both) coefficients (pi. P2 = x 2 .2Q ADVANCED MATHEMATICS function f(x) is analytic at x 0 if its Taylor series about x 0 exists.c) These functions are analytic everywhere and all points are ordinary points.4 .2 ) y ' + (x + l ) y = 0 (2.18a) x 2 (x2) 2 x 2 (x2r .
x o ) p j as well as ( x . assuming x 0 = 0.x0) pi and (x . x2(x2)2 and b = 0 (2. we would be dealing with (x2)Pl=^ x and (x2)2p2= £ i i x (2.c.19a. x2(x2) p2 = — ^ ± J — .SERIES SOLUTIONS AND SPECIAL FUNCTIONS £/ Pj = ^ .11 la.b) L) X (A x + 1 (2. .d) is a (xxo)2p2 = x2p2= ^ y Both (x .110a. In this chapter. If not. (2.110c. x 0 is an irregular singular point. x 0 is a regular singular point of the differential equation. the singular points are at x = 0 and at x = 2.d) Here.b) (2.11 are regular singular points. Determine which of the singular points in Example 2.12.d) (x2)1 xpi is not analytic at x = 0 and therefore xo = O is an irregular singular point. i) ii) All points are ordinary points.b) which are analytic at x 0 = 2 and therefore x 0 = 2 would be a regular singular point.18b. we use power series extensively and we next summarize their properties. If x 0 is a singular point of the differential equation and ( x . hi) (xxo)Pl = xPl= (xxo)2p2 = x2p2= 2 (2. Note that had we chosen x 0 = 2 in part (iii). Example 2. x— 2 ( x .x o ) 2 p 2 are both analytic at x 0 .19c.x 0 ) 2 p 2 are analytic at x = 0 and therefore x 0 = 0 regular singular point.x o ) P l = x P l = ? /v J.
92 ADVANCED MATHEMATICS 2. The third series diverges for all values of x (* 0). It is usual to state that R is zero for series that converge only at x = 0 and R is infinite for series that converge for all values of x. No conclusion can be drawn if L = 1. There are three possibilities: (a) the series converges for all values of x.. and (c) the series diverges for all nonzero values of x.2lc) The first series represents exp (x) and is convergent for all values of x. The series converges at a point x if the lim m>0° 2L an x " n=0 the sum of the series is the value of this limit. We illustrate this situation by the following examples. If a series converges for xj < R and diverges for I x I > R. (2. (2. we can write the series as ^ m n=0 e x i s t s and an x n . (2.2lb) (c) £ n=0 n ! x n = l + x + 2 ! x 2 + 3 ! x 3 + .. In many cases.14a) is a power series in (x ... Thus all power series have a radius of convergence. If L > 1.Not all power series converge for all nonzero values of x.. A series may or may not converge at its radius of convergence. Every power series converges at x = 0 and its sum is a0. (b) the series converges for some values of x and diverges for other values of x. 00 Applying this test to the series ^ n=0 an x n .22) is less than one (L < 1). 00 (a) ]£ n=o 00 x "/( n ! ) = 1+x + x 2 /2!+x 3 /3! + . By translating the oo origin to x 0 .21 a) (b) 2L x " n=0 00 = l + x + x 2 + x 3 + .2 POWER SERIES The series on the right side of Equation (2. we have (x * 0) . The second series is a geometric series and its sum is 1 /(I .. R is the radius of convergence.x 0 ).x) and is convergent if x < 1 and diverges if  x > 1. We recall that the 00 series X n=0 un is convergent if the ratio Hfl±i _ u n lim n —> 00 L (2. the radius of convergence R can be determined by the ratio test. the series is divergent.
If the ratio v n /u n tends to a finite nonzero limit as n—> °°. if ^ n=0 u n is divergent.b) ^ O .> . (1+2/n)2 .c) . Discuss the convergence of the following power series.SERIES SOLUTIONS AND SPECIAL FUNCTIONS a xn+1 a ££ l j m tn±l anx" n—>o° Thus if Urn n>o° = lim _JI±1 x = L n—>o° an (2.b.b) the series is convergent.^ n — an+1 (2.24a.^ L _ n>~ a n + 1 (2.26) (2. (2. R = km — = lim — =1 n>o° ( 1 + n ) 2 n ~ >o ° (1 + 1/n)2 . OO n=0 The series ^ n=l 1 /n is divergent and the series j ^ 1 /n n=l is convergent.an+1 the series is divergent. _ _ . If ^ n=0 u n is convergent. (2 + n) 2 . the series ^ n=0 v n is divergent if v n > K u n for all n and any positive constant K. Example 2. the series ^ n=0 OO v n is convergent if v n < K u n ..x < 1 an => x < lim .21. then oo oo 2_j v n converges or diverges according as ^ n=0 OO u n converges or diverges.27a. . This test can also be stated as follows.23a...^ " . If x > lim .25) The comparison test is another simple method of determining whether a series is convergent or oo co oo divergent. It follows that the radius of convergence R is given by R = lim . oo oo oo a) ]Tx n /(l+n) 2 n=0 b) £xn/(l+n) n=0 c) £ (x2)2n/(l + 2n) n=0 a) .
that is to say .2 < 1 and is divergent if x  > 1 or x . the series oo oo oo becomes ^T 1/(1 + n) and since ^ n=0 n=l 1/n is divergent. For oo x = 1. we translate the origin and write x* = x .28) As in a). oo If the series ^ n=0 a n x n converges for  x  < R (R •£• 0) and its sum is denoted by f (x). At both of these oo values of x. if u n > u n + 1 > 0 n=0 and lim u n—>~ is zero. the series is convergent.b) n=0 The series becomes £ R = lim 3 + 2n _ j n —>~ l + 2n The series is convergent if x  < 1 or x .l . oo the series 1/n .2 oo (2. which states that for an alternating series j ^ ( . If x = . the series ^ ( .1 .29) (x*) 2n /(l +2n). the series is convergent if x = .2 > 1. the series is convergent if  x < 1 and divergent if  x > 1. We apply n=0 oo Leibnitz's test. In the present example.l ) n u n . the series behaves as ^ n=0 1/n and is divergent. We make use of the following properties of power series. (2.l . it follows that ^ n=0 OO 1/(1 + n) is also divergent.94 ADVANCED MATHEMATICS The series is convergent if x < 1 and is divergent if x > 1. To x = 1 corresponds x = 1 or x = 3.l ) n / ( l + n ) is an alternating series. For x = 1. b) R = Urn n —> oo f±Jl 1 + n = 1 (2. In this example x* is raised to an even power and there is no need to separately consider the cases x = 1 and x = .210a. c) In this case. we becomes jT 1 / ( I + n ) n=0 and by comparison with the convergent series ^ n=l deduce that the series is also convergent if x = 1.
d) where p l n = pf°(O) and p 2 n = p f (0) . If we are required to find the solution at points near infinity. Without loss of generality. we can translate the origin and write x* = x .32c. 2. The differentiated series have the same radius of convergence R and converge to the corresponding derivatives of f (x). The functions pj and P2 are analytic at x 0 (= 0) and their Taylor series are oo (2. as given by Equation (2.3lb) Pi 00 = X Pln X " n=0 oo <23"2a) P200 = Z n=0 P2nx" (2.E.SERIFS SOLUTIONS AND SPECIAL FUNCTIONS oo 91 f(x) = X n=0 anx" (2.32b) (2. the two series ^ n=0 a n x n and ^ n=0 b n x n can be added and multiplied in the same way as polynomials. If x 0 is nonzero. we can set x 0 to be zero. If two power series converge to the same sum throughout their interval of convergence. The series can also be integrated term by term and the resulting series represents the integral of f (x).x 0 In the new variable x .3la) (2.D. oo oo If ^ n=0 OO b n x n is another power series with radius of convergence R. x = x 0 corresponds to x = 0 . their coefficients are equal.3 ORDINARY POINTS Consider the standard form of the second order O.211) the series can be differentiated term by term as many times as is required. We seek a solution in the neighborhood of XQ.12a). we change the independent variable from x to Xj and write X! = 1/x To points near x at infinity correspond to points near Xj at the origin. Further discussions on power series are given in Chapter 3.
.36) y' 1+x2 2—y 1+x2 = o (2.36) is written as y"+ 2x (2.96 ADVANCED MATHEMATICS The homogeneous form of Equation (2. = 0 where kj are known expressions in terms of p^:.34a to c) into Equation (2. P2k ...2 n=2 Substituting Equations (2. ) is zero. In standard form.37) .. Equation (2.31. and c^ . 1...34a) where c n (n = 0.35) is true for all x and this implies that each of the kj (i = 0. The recurrence equation (kj = O) allows us to determine c n in terms of pjj and P2j . Example 2.33) results in an equation of the form k 0 + kj x + k 2 x 2 + .. Differentiating term by term yields oo / = X ^n^"1 n= l oo (23"4b) (2.35) (2..12a) can be written as y"+ X Pm x " yf+ X P 2 « 4 \n=0 / Vn=O / = ° (23"3) The solution y also has a Taylor series near the origin and y can be expressed as y = X cnxn n=0 (2. Obtain the power series solution of (l+x2)y"+2xy'2y = 0 in the neighborhood of the origin. Equation (2.34c) y" = X n(nl)c n x n . 1. + k n xn + . The next example illustrates the method of obtaining a series solution.. .) are constants.
These equations are now written as oo y = X r=0 OO CrxI" (2. b. we obtain oo X [(r + 2)(r+l)c r + 2 x r + (r + 2)(r+l)c r + 2 x r + 2 + 2 ( r + l ) c r + 1 x r + 1 ..SERIES SOLUTIONS AND SPECIAL FUNCTIONS 97 The functions pj [ = 2 x / ( l + x 2 ) ] and p 2 [ = .. . we write Equations (2. We seek a solution given by Equation (2.l ) c s + 2 s c s .36).h) We note from Equation (2.l ) c s / ( s + 1) (2. C5...37).310g. we write (n = r. eg.310a. cj. we obtain c 2 . C4.f) 6C3 + 2CJ2CJ = 0 12c4 + 2c 2 + 4 c 2 . Substituting Equations (2. . For ease of computation.310c. in terms of Cj. n = r + l .34a to c) such that the summation index starts from zero.2 c r x r ] = 0 r=0 (2. . we work with Equation (2. c) respectively.310h) that we have a formula relating c s + 2 and c s and this implies that we can separate the solution into an even and an odd solution.2 / ( 1 + x2)] are analytic at the origin.310e. It follows that the solution can be written as y = c o [ l + x 2 .d) (2. In the present example.39) Comparing powers of x.34a) with y' and y" given by Equations (2.. we have x°: x1: x2: 2c22c0 = 0 => c 2 = c0 => c3 = 0 => c 4 = c 2 /3 (2.36) rather than with Equation (2. ] +CjX (2.311) .b) (2.38c) To avoid having to expand (1 + x 2 )" 1 .34a. To achieve this.( s . .34b.38a to c) into Equation (2.x 4 / 3 + x 6 / 5 .2 c 2 = 0 xs: (s + 2 ) ( s + l ) c s + 2 + s ( s . c) respectively. n = r + 2) in Equations (2. in terms of c 0 and C3.38a) y1 = X ( r + l ) c r + 1 x r r=0 OO (2.2 c s = 0 => c s + 2 = . C3 is zero and consequently all the coefficients with an odd index greater than or equal to three are zero. That is to say.38b) y" = ]T (r + 2)(r+l)c r + 2 x r r=0 (2.
.. On expanding p^ (x) and p 2 (x) in powers of x. b) are convergent for  x < 1 and pj (x) and p 2 (x) are analytic in the interval  x < 1.31.. it can be seen that yj can be written as yx = 1 + x (x ..x 3 /3 + x 5 /5 . In Example 2.D.315a. In Example 2. we obtain Pj (x) = 2x (1 .2 (1 . ) p 2 (x) = .x 4 /3 + x 6 /5 . the series is an alternating series and by Leibnitz's test. ) (2...98 ADVANCED MATHEMATICS where c 0 andcj are arbitrary constants. It must be pointed out that it does not follow that there is no analytic solution that goes beyond the critical point of the differential equation. the solution y 2 (= x) is valid for all values of x.x 2 + x 4 .. There is no loss of generality in setting c 0 and Cj to be equal to 1. The general solution is given by the linear combination of yj and y 2 .x 2 + x 4 . We observe that the series solution about an ordinary point is valid in an interval that extends at least up to the nearest singular point of the differential equation.65). The general solution of a second order equation involves two arbitrary constants (A and B) and they are determined by the initial conditions [y (0) and y'(0)] or by the boundary conditions.312a.315b) We note that the series in Equations (2.314) (2. From Equation (2. The solution yi is in the form of an infinite series and is valid as long as the series is convergent. The general solution y is expressed as y = Ay1+By2 where A and B are arbitrary constants.x 6 + . At x = 1.x 6 + .31.b) The infinite series is convergent for x < 1. . The .E and has two linearly independent solutions yj and y 2 . it is convergent..31 and can be generalized to all power series solutions about an ordinary point. The fundamental solutions y^ and y 2 can be written as yj = 1 + x 2 . • Note that Equation (2. The solution yj is valid for  x  < l . y 2 (= x) has only one term and is a valid solution for all values of x. The point ( x = l ) is a singular point (see Example 3. This observation is not restricted to Example 2.315a) (2.311)..36) is a second order O.313) y2 = x (2..) (2.
b) into Equation (2..SERIES SOLUTIONS AND SPECIAL FUNCTIONS 99 presence of singular points in a differential equation does not imply that all the solutions are singular at the singular points.45) (2.43b). Alternatively Equation (2.) y = 0 (2.43b) can be transformed to an equation with constant coefficients by writing x = el Using the chain rule..4lb) x 2 p 2 (x) = b o + b 1 x + b 2 x 2 + .12a) and considering the homogeneous case.43a) (2.46) (2.42) We seek a solution in the neighborhood of the origin and the leading terms of Equation (2.43b) Equation (2. we obtain (2. we require r ( r .47) .4 REGULAR SINGULAR POINTS AND THE METHOD OF FROBENIUS We recall that x 0 (=0) is a regular point if xpj(x) and x 2 p2(x) are analytic. 2. Further examples will follow to illustrate this statement..l ) + aor + b o = 0 (2.. It admits a solution of the form y = xr On differentiating and substituting into Equation (2.43b) is Euler's (Cauchy's) equidimensional equation... xpi (x) and x 2 p2 (x) can be expanded as xpj(x) = aQ + ajX + a 2 x + . we obtain y" + (a o /x + aj + a 2 x + . That is to say.4la) (2..46) is a quadratic in r and has two roots rj and r 2 . we obtain x r [r (r 1) + a o r + b 0 ] = 0 To obtain a nontrivial solution (xr ^ 0).) y' + (b Q /x 2 + bj/x + b 2 + . Substituting Equations (2. The two linearly independent solutions are x r ' and xr2.. (2.42) are y" + (a o /x)y 1 + (b o /x 2 )y = 0 or x2y"+a0xy'+boy =0 (2.44) Equation (2.4la.
b) The general solution is a linear combination given by y = Clx1/2 + c2x1/2 (2. Solve the following Euler equations.49) The coefficients in Equation (2. a) b) a) 4 x 2 y " + 4 x y ' . Equation (2.1 / 2 (2. Equation (2.l ) ^ + boy = 0 dr dt (2.414b) We have a double root (r = 1) and we have only one solution .43b.410b) The two linearly independent solutions are 1/9 — 1/9 Yl = * . 8a. y2 = x 1/z (2.n MATHEMATICS dy _ t dy (2. Example 2. + ( a o .49) are constants and can be solved by the methods described in Chapter 1.b) = 0 (2.y =0 x2y"xy'+y=0 In this example.46) is 4r(rl) + 4rl The two roots are vx = 111 and r2 = .413c) b) In this case. b) yields i ^ .100 ADVANCF.413a.412a.410a) (2.411) (2.414a) (2.48b) (2.46) becomes r(rl)r+l=0 or (r1)2 = 0 (2.41.48a) ^ = et(e^ + et^) dx 2 \ dt dt 2 / Combining Equations (2.
b) This means that if Equation (2. x r and ^r.416a. f (r0) = 0 (2.420a.421) where c 0 is not zero and r is any real or complex number.(x r ) = j .4la.[exp (r i n x)] = ( i n x) exp (r i n x) = xrinx The other linearly independent solution is y2 = x i n x (2. b). by changing the independent variable x to t [Equation (2.417c) Alternatively. To differentiate with respect to r.415) (2. . That is to say j .(x r ) are solutions of the or differential equation.611)] Yi = el = x y2 = t e l = x i n x • Note that the exponent r is not necessarily an integer. we note that if f (r) has a double root at r 0 .418) (2.b) (2.SERIES SOLUTIONS AND SPECIAL FUNCTIONS ]QJ Yl = x To obtain the other solution. then f(r 0 ) = 0. we write x r as exp (r i n x). This suggests that if we retain all the terms on the right side of Equation (2.410b) can be written as *JL _ 2 ^ + y = 0 dr dt The solutions are [see Equation (1.419) (2.46) has a double root. we should try a solution of the form CO (2.420c.d) y = x r X cnxn n=0 (2.417b) (2. Equation (2.417a) (2.47)].
422a) yi = X n=0 cnx"+r2 (2. xpj(x) = ( x + l ) / 2 . Obtain a power series solution to the following equation 2xy" + (x + 1) y' + 3y = 0 in the neighborhood of the origin.25 ) On differentiating. On comparing powers of x. the two roots are coincident.b) (2.102 ADVANCED MATHEMATICS The series is differentiated term by term and substituted in the differential equation. p 2 (x) = 3/2x x 2 p 2 (x) = 3x/2 (2. The other equations are the recurrence formulae and are used to determine the coefficients c n [see Equations (2. In the examples that follow. we consider the three possible cases: the roots of the indicial equation are distinct and do not differ by an integer. we deduce that x = 0 is a regular singular point.46)] and is the indicial equation. We seek a solution of the form oo y = Z c n* n + r n=0 ( 2 . In this example p : (x) = ( x + l ) / 2 x .423) From Equations (2.42. This method of solving a differential equation is called the method of Frobenius. yj and y2 are not linearly independent and we have to modify our method.424c. The two linearly independent solutions are oo yi = S n=0 Cnx"+Fl (2.424a to d).39atoh)]. The equation associated with the lowest power of x is a quadratic equation in r [see Equation (2. and the two roots differ by an integer.424a.4lb) or the two roots differ by an integer.422D) If the two roots coincide (see Example 2. we obtain oo y1 = X n=0 (n + r)cnxn+T1 (2. In general.d) (2.4 . the indicial equation yields two distinct values of r which are denoted rj and r2. Example 2. we obtain a set of algebraic equations.426a) .
C3. 26a.SERIES SOLUTIONS AND SPECIAL FUNCTIONS DO 103 y" = £ n=0 (n + r)(n + r . C2.l ) c 0 + rc 0 = 0 => c 0 [r (2r.1 : 2 r ( r .428e. The lowest power of x is x r .Z * + 21*'jGL*» + .425. 7 c 0 (Z. 3 • 2 .428ad) 2 ( l + r ) ( r ) c 1 + r c 0 + (l+r)c 1 + 3c 0 = 0 => cx =[(3+r)c o ]/[(r+1) (2r+1)] (2.426b) Substituting Equations (2.f) : 2(r+s + l)(r+s)c s + 1 +(s+r)c s + (s+r+l)c s + 1 + 3cs = 0 => c s+ 1 = .2 (2.<t3Z) Cs+1 2 S + 1 (2s + 3) (2s + 1)..429.1)] =0 => T1 = 1/2 and r 2 = 0 xr: (2.430a) (2. and they are c 1 =7c 0 /[2«3] c 2 = .428h) leads to c s + 1 = .429) y .7 » 9 « l l c o / [ 2 3 » 7 « 5 « 3 » 2 .[(s+r+3) c s ]/[(s+r+1) (2s+2r+1)] (2..h) x Substituting the value of r = 1/2 into Equation (2.1) c n xn+T~l + (n + r) c n x n + r + (n + r) c n x n + r " 1 + 3c n x n + r ] = 0 (2.l l c 2 / [ 2 « 7 » 3 ] = ..423) yields oo J n=0 [2 (n + r) (n + r .—  + 7)(2s + 5 ) ..1 x r .c) (2.428g.430b.e) (2. ..l l .43!) (l) s + 1 (2 S . b) in Equation (2.9 . 7 • 5 • 3 • (s + 1) (s).3 ] We denote the solution corresponding to r = 1/2 by yj and it can be written as (2.l ) c n x n + r .430d.. we deduce that c (2.9c2 /[2 • 5 • 2] = 7 • 9 c o /[2 2 • 5 • 3 • 2] c3 = .] From Equations (2...[(2s + 7) c s ]/[2 (2s + 3) (s + 1)] We can compute C\. 30a to e). = v / T [ l .427) We compare powers of x.
.. . .3 ] = 4 .438b..4 b 1 / [ 2 . b) yields _ ( ..3 b o / [ 2 .428h) becomes bs + 1 = .32(2s+l)(2sl). 33a.3 ] b 3 = .438d.434) becomes Cn (2.437) (2..433b) 2 S + I ( s + 1)! Combining Equations (2. . We substitute this value of r2 in Equation (2..5 • 4 • 3 b o /[2 • 3 • 3 • 5] From Equations (2. are bl = . b3. 38a to e).! ) » • (2s + 7)1 ca 2 s 5! (s + 3)! (2s+ 3)! Setting n = s + 1.3 b 0 b2 = . 53 V ' . b 2 . 5 .=c0V7i (D°(2n + 5 ) ! x ° _ (24_36) n=o 2 n " 1 5 ! (n + 2)! ( 2 n + l ) ! To obtain the other solution y 2 .c) (2. we deduce that b s+1 (2.l ) S + 1 ( s + 3)(s + 2 ) . Equation (2. 4 .( s + 3 ) b s / [ ( s + l ) ( 2 s + l ) ] The coefficients bj.434) .104 ADVANCED MATHEMATICS The right side can be simplified by the following identities (2s + 7) .428h) and to avoid confusion we denote c s by b s for the case r 2 = 0. Equation (2. 3 = ( 2 s + 3)! (2.5b 2 /[3 • 5] = .e)  ( . 7 = (2s + 7 ) ! 3 ! (2..437. 3 b 0 (s+l)s.432.. we consider the case r 2 = 0.433a) 6! 2 s ( s + 3)! (2s + 3) .438a) (2. _ ^ ( 2 n + 5V..ca ( 2 4 3 5 ) 2 n " j 5 ! (n + 2)! (2n+l)! The solution y^ can be written as y..
= oo (2. we deduce that the radii of convergence of the two series are r> r Cs R = km —§— s —> oo c r 2(2S + 3 ) ( S + 1 ) = lim —~s —> ~ (2 s + 7) = oo . . In this example..443a.' < " + 2>!*n n=o (2.445a..b.429. xpj = 1. . ..44.43. .444) The origin is a regular singular point and the series solution is of the form . (2n)! The arbitrary constants c 0 and b 0 are determined from the initial conditions or the boundary conditions.442a. From Equations (2.c) R = bm The solutions yj and y2 are valid for all values of x. Example 2. Ar. we have p x = 1/x. Obtain a series solution to the equation x 2 y " + x y ' + x2y = 0 in the neighborhood of the origin.i r 1 2 s ( s + 3)!b0 (2s+ 2)! U4^c. 37).l ) s + 1 (s + 3)! 2 S + I ( s + 1)! b n = ~ 2 ( .= lim S (s+l)(2s + l) / >o° v (S + 3) .26. Again setting s + 1 = n. we find that the solution y2 can be written as (2.b) (2.SERIES SOLUTIONS AND SPECIAL FUNCTIONS 105 ( . >oo ^ b — s .445c..440) n=0 V '' The general solution y is a linear combination of y^ and y2 and can be written as y = C o V7 £ n=o <'>n'2" + 5 > U — + b 0 2 n " 1 5 ! ( n + 2)! ( 2 n + l ) ! £ ^ ° 2 n .c) _ S .._ . . + l)l (2s + 2)f A = (24"39b) ( . p2 = 1 x 2 p 2 = x2 (2.b. (2.d) (2.
f) x r + 1 : cx [(1 + r ) r + (l +r)] = 0 x r + 2 : c 2 [(2 + r ) ( l + r ) + (2 + r)]+c 0 = 0 xr+s: c s [(s + r) (s + r . c 4 . c s is zero.450a) (2..b) (2.448h) becomes C2m = c2m2/(2m)2 ( 2 . From Equation (2.c) (2.c s _ 2 /(s + r) 2 (2.449.c o / ( 2 + r) 2 (2. we deduce that if s is odd. 6 2 . starting with the lowest power (n = 0).448h)..4 ' 4 9) The coefficients c 2 .450d.l ) m c 0 / [ 2 2 « 4 2 .e) 0/2 c 4 = . we deduce that C2m = ( .448a.1) + (s + r)] + c s _ 2 = 0 => c s = . we deduce that Cj is zero.448e. eg .444) yields oo £ n=0 [(n + r) (n + r . 50 a to e).450b..448g.c 2 / 4 2 = c o /[2 2 «4 2 ] c6 = .4 2 » 6 2 ] From Equations (2.451b) One series solution can be written as . we have xr: c o [ r ( r .446) Differentiating and substituting in Equation (2.451a) (2.1) c n x n + r + (n + r) c n x n + r + c n x n + r + 2 ] = 0 (2..h) Note that with r = 0.c 4 / 6 2 = .d) (2.447) Comparing the powers of x.c o / [ 2 2 . are given by c C2 c/22 c (2.l ) + r] = 0 => r = 0 (double root) =» Cj = 0 (r = 0) => c 2 = . We consider the case where s is even and we write s = 2m and Equation (2. (2m)2] = (l) m c 0 /[(2 2 m )(m!) 2 ] (2.448c.106 CO ADVANCED MATHEMATICS y = X n=0 Cnxn+I (2.
SERIES SOLUTIONS AND SPECIAL FUNCTIONS ~ m=o r ££Z _nm 2m XiW = c o Z V T (2452) 22m(m!)2 Equation (2.452) becomes ~ m=o /_ n m 2m W . b) in Equation (2.454. From Equation (2.4lb suggests that we look for a solution of the form oo y = i n x J 0 ( x ) + X bnxn n= l (2.456b) Substituting Equations (2. oo oo oo inx[x2Jo+xj. the form of the series solution is oo y = (in x) y! (x) + X n=0 bn x n + r ° (2.456a) y" = i n x Jo (x) + 2JQ (X)/X .453).J 0 (x)/x 2 + ^ n=2 n (n .+x2Joj + 2xj. the indicial equation [Equation (2.448a)] has a double root and the second linearly independent solution is not readily available.I ^r~I 2 (m!) (2453) In this example.457) The function Jo is a solution of Equation (2.454) Note that if the double root of the indicial equation is r0 (*= 0).444) and the terms inside the square bracket equal zero. Equation (2. It is customary to set c 0 = 1 and to denote yi (x) by J 0 (x). 56a. Example 2.+ X n(nl)b n x n +X nb n x n +X b n x n+2 = 0 n=2 n=l n=l (2. we obtain .1) b n x n " 2 (2. we obtain oo y1 = i n x JQ(X) + J0(x)/x + X n ^ x " " 1 n=l OO (2.444) yields .455) Differentiating y.444) is Bessel's equation of order zero and the equation of arbitrary order will be considered later.
.d) = . l + 1 ^ . we compare powers of x. is denoted by y2 and is given by y2=inxJ0(x) + ^ .459) To obtain the coefficients b s .463a. 22m ' + 2m (2ml) b 2 m + 2m b 2 m + b 2 m _ 2 = 0 m!(ml)! (2. we can obtain b4.l + 2 b 2 + 2b 2 = 0 6b 3 + 3b 3 + b 1 = 0 => => b 2 = 1/4 b3 = . . which is linearly independent of Jo.462.460d.462). . x 2m : 9 .10& ADVANCED MATHEMATICS (2.b) (2. 63a to d). .f) One observes that b s is zero if s is odd and we need to consider the even powers of x only.458) m=l 2 (m!) Substituting Equation (2.461) The recurrence formula is \ b2m (l)m 2^7^ + b 2rn2 1/ (2m)2 (2 462) ~ L22m m!(ml)! J/Um) ^.460b. we obtain £ s=i — ^ .2 4 3!2! j / 3 6 = m24 The second solution.c) (2. we can verify that the general term b 2 m can be written as .b : / 9 = 0 (2.463c.464) From Equations (2.31)..460c)] and from Equation (2. as follows b4 = .458) in Equation (2.4 sz) The coefficient b 2 is known [Equation (2.460a) (2.^ + s(s+l)bs+1xs+1+sbsxs + bsxs+2 .e.[ 4 ^ + b 2 ] / > " = . bg. x1: x2: x3: b] = 0 .457) and changing the indices appropriately such that the summation index s starts from one (see Example 2.2 2 s ~ 2 s! (s1)! =0 (2.T  8 b6 (2. (2.
SERIES SOLUTIONS AND SPECIAL FUNCTIONS
1J&
(2.465)
The solution y2 is
oo
y2 = i n x J0(x) + X b 2m x 2m
m=l
(2.466)
The general solution of Equation (2.444) is the linear combination of Jo and y2 and is y = AJ 0 + By 2 where A and B are constants. Note the presence of Jtn x in y 2 . This implies that y2 is singular at the origin. If the physics of the problem require y to be finite at the origin, we require B to be zero. Example 2.44. Solve the equation x 2 (x2  1) y"  (x2 + 1) x y' + (x2 + 1) y = 0 in the neighborhood of the origin. In this example, pi and p 2 are given respectively by (2.468) (2.467)
p ,    ^  .
x(x2l)
P2 = f ^
x2(x2l)
(2.469a,b)
The singular points are at x = 0, x = 1, and x = —1. We further note that x = 0 is a regular singular point and we assume a solution of the form
oo
y =X
n=0
cnxF+n
(2.470)
Differentiating and substituting the resulting expressions in Equation (2.468) yields
oo
£
n=0
[(n+r) (n+r1) cn x r + n + 2  (n+r) (n+r1) cn x r + n  (n+r) cn x r + n + 2  (n+r) cn x r + n + c n x r + n + 2 + cnxr+n] = 0 (2.471)
Comparing powers of x, we obtain
110
ADVANCED MATHEMATICS
xr:
c o [  r ( r  l )  r + 1 ] = 0 => r
2
l=0
=> r = ±1
(2.472a,b,c) (2.472d,e,f)
x r + 1 : C j t  O + r J r  r + l ] = 0 => c t (r 2 + 2 r  1 ) = 0 => c 1 = 0 ( r = +l)
x r + 2 : c o [ r ( r  l )  r + l ] + c 2 [  ( 2 + r ) ( l + r )  ( 2 + r ) + l ] = O => c 2 = c 0 ( r  l ) 2 / [ ( 2 + r ) 2  1 ] (2.472g,h) xr+s: cs_2[(s+r2)(s+r3)(s+r2)+l] + cs[(s+r)(s+rl)(s+r)+l] = 0 => S = c s _ 2 [(s+r2) (s+r4)]/[(s+r) 2  1] (2.4721 j)
We deduce from Equations (2.472f, j) that c^= c 3 = C5 = ... = 0. Substituting the value of r = 1 in Equation (2.472h) leads to c 2 = 0 and Equation (2.472J) implies that c 4 = c 6 = ... = 0. The solution yi corresponding to r = 1 is a polynomial (one term only) and is yi = c o x (2.473)
To determine y 2 , the solution corresponding to r =  1 , we substitute this value of r in Equation (2.472h) and we observe that the numerator ( r  1 ) 2 is nonzero and the denominator [(2 + r) 2  1] is zero. This implies that c 2 is infinity and we must seek a solution of the form
00
y = Yl i n x + £
n=0
OO
b n x"" 1
(b0 * 0)
(2.474a)
= c o x i n x + J bn x""1
n=0
(2.474b)
On differentiating term by term, we obtain
00
y1 = c 0 ( i n x + 1) + ]T (n  1) b n x11"2
n=0
OO
(2.475a)
y" = co/x + £
n=0
(n " 1) (n  2) b n x11"3
(2.475b)
Substituting y, y', and y" in Equation (2.468) yields
00
X [(nl)(n2)b n x n+1 (nl)(n2)b n x n  1 (nl)b n x n+1 (nl)b n x 11  1
n=0
+ b n x n + 1 + b n x" 1 ]  2c 0 x = 0
(2.476)
SERIES SOLUTIONS AND SPECIAL FUNCTIONS
777
We now compare powers of x. x"1: x°: x1: x3: b 0 (2 + 1 + 1) = 0 b! = 0 2b 0 + b 0  b 2 + b 0 + b 2  2 c 0 = 0  ( 3 ) ( 2 ) b 4  b 2  3 b 4 + b2 + b4 = 0 => => b 0 = cQ/2 b4 = 0 (2.477a) (2.477b) (2.477c,d) (2.477e,f)
From Equations (2.476, 77a to f), we deduce that b 1 = b 3 = ... = O, bo*O, b 2 = b 4 = ... = 0 (2.478a,b,c)
The solution y2, corresponding to r =  l , is given by y2 = c 0 x i n x + bo/(2x) (2.479)
Without loss of generality, we can set c 0 = b Q = 1 and the two linearly independent solutions y \ and y 2 can be written as
yi
= x,
y 2 = x i n x + 1/(2x)
(2.480a,b)
The general solution of Equation (2.468) is y = Ayi+By2 where A and B are arbitrary constants. Note that y 2 is singular at the origin and if the general solution y is finite at the origin, B has to be zero. Example 2.45. Find a series solution to the equation x 2 y" + x y ' + ( x 2  l / 4 ) y = 0 valid near the origin. The origin is a regular singular point [xpj(x) = 1, x 2 p 2 (x) = (x2  1/4)] and we seek a solution in the form of Equation (2.470). Proceeding as in the previous example, we obtain
oo
(2.481)
(2.482)
] £ [(n +r) (n + r 1) cn x n + r + (n + r) cn x n + r + cn x n + r + 2  (1/4) cn x n+r ] = 0
n=0
(2.483)
112
ADVANCED
MATHEMATICS
Comparing powers of x yields xr: c o [ r ( r  l ) + r  l / 4 ] = 0 => r = ±1/2 (2.484a,b) 2.4.84c) (2.484d)
x r + 1 : cY [r(l + r) + ( l + r )  l / 4 ] = 0 => c{ = 0 if r= 1/2 => Cj is arbitrary if r = 1/2
x r+s : c s [(s+r)(s+rl) + (s+r)l/4]+c s _ 2 = 0 => cs = c s _ 2 /[(s+r) 2 1/4] We now substitute the value of r = 1/2 in Equation (2.484f) and we obtain
c s=
(2.484e,f)
c s _ 2 /[s (s + 1)]
(2.485)
From Equations (2.484c, 85), we note that all the coefficients with an odd index are zero. Writing s = 2m, we obtain
C2m =
 C 2(ml)/[( 2m ) ( 2 m
+
Dl
(2.486)
From Equation (2.486), we compute the first few coefficients and they are c2 = c o /[23] c4 = _c 2 /[4«5] = c o /5! c6 =  c 4 / [ 6  7 ] = c o /7! From Equations (2.486, 87a to e), we deduce that
C2m
(2.487a) (2.487b,c) (2.487d,e)
= (Dmc0/(2m+l)!
(2.488)
The solution yj is given by
yi  co L
m=0
f
( l) m x 2 m + ^ 2
(2m+1)!
_
 cox
1/2
f
2,
m=0
(l) m x 2m+1
(2m+l)!
(2.489a,b)
In Equation (2.489b), the sum is sin x and yj can be written in closed form as yj = c 0 x 1 / 2 sinx To obtain y2, we substitute the value of r = 1/2 in Equation (2.484f) and we obtain (2.490)
SERIES SOLUTIONS AND SPECIAL FUNCTIONS
cs
LU.
= cs_2/[s(sl)]
(2.491)
The denominator is zero if s = 1 and it is not possible to determine c^. From Equation (2.484c), we deduce that C] is arbitrary and we can compute cj, c 5 , ... All these coefficients are nonzero. The coefficients with even indices can be written as multiples of c 0 and those with odd indices are multiples of q . We compute the first few coefficients as follows c2 =  c 0 / ( 2 ) , c3 =  c 1 / ( 3 » 2 ) , c 4 =  c 2 / ( 4 « 3 ) = c 0 /4! , c 5 =  c 3 / ( 5  4 ) = c 1 /5! , c 6 =  c 4 / ( 6 • 5) = c Q /6!, c 7 = c 5 /C76) =  c 1 / 7 ! , ... (2.492a)
... (2.492b)
From Equations (2.491, 92a, b), we verify that
C2m
= (Dmc0/(2m)! ,
c2m+1 = (l)mCl/(2m+1)!
(2.493a,b)
The solution y 2 can be written as
y2
2^[
(2m)!
+
(2m+l)r
J
(2A 94)
Note that the second term on the right side of Equation (2.494) is y^ (with Cj replacing c 0 ). The summation ^
m=0
(  l ) m x 2m /(2m)! is cos x and the two linearly independent solutions y^ and y 2
are, to the extent of a multiplicative constant, yx = x ^ 1 / 2 s i n x , y 2 = x" 1 / 2 cosx (2.495a,b)
The general solution of Equation (2.482) is y = Ay!+By2 where A and B are arbitrary constants. Observe that y 2 is singular at the origin and y^ is finite at the origin. For solutions which remain finite at the origin, we require B to be zero. Example 2.46. Gupta and Douglas (1967) considered a steady state diffusion problem, associated with a first order irreversible reaction involving isobutylene and spherical cation exchange resin particles. Equation (1) of their paper can be written as (see Equation A.IV3) (2.496)
Ul
ADVANCED MATHEMATICS
(2.497a) _r where R A =  k c A dr v dr / (2.497b)
Assuming c A to be a function of r only (symmetry!), Equation (2.497a) reduces to an O.D.E. given by ^AB r \
2
i  ^ +2 r ^ A kr2cA = 0 dr dr
(2.498a)
The boundary conditions are cA = c A . ^ dr = 0 at r = R at r = 0 (2.498b) (2.498c)
We now solve Equations (2.498ac) for c A via the method of Frobenius. Note that the origin (r = 0) is a regular singular point. We seek a solution of the form
oo
C A=
X
n=0
Cnr"+P
(2499)
In Equation (2.499), the exponent involves p, so as to avoid confusion with the radial variable r. Differentiating Equation (2.499) and substituting the resulting expressions in Equation (2.498a) yields
oo
£
n=0
[c n (n+p)(n+pl)r n+ P + 2(n+p)c n r n+ PKc n r n+ P +2 ] = 0
(2.4100a) (2.4100b)
where K = k / , 0 ^ Comparing powers of r, we have rp: c o (p) ( p  l ) + 2pc 0 = 0 =» p ( p + l) = O => p = 0 a n d p =  l
(2.4101a,b,c,d)
r 1 + P: c 1 ( l + p)(p) + 2 ( l + p ) c 1 = 0 => cx (1+p) (2+p) = 0 (Cl=0, =>  Cj is arbitrary , if p =  1 if p = 0 (2.4101e,f,g,h)
SERIES SOLUTIONS AND SPECIAL FUNCTIONS
m
r s + P: c s ( s + p ) ( s + p  l )
+
2 ( s + p ) c s  K c s _ 2 = 0 => c . =
(
g
J ^
+ 1)
(2.4101ij)
To determine c s , we substitute the values of p in Equation (2.4lOlj)
c^T^ry
forp=0
<24102)
Equation (2.4101h) states that q = 0 if p = 0. Therefore, c s = 0 if s is odd. We proceed by replacing s by 2m and Equation (2.4102) becomes
C
^
2Om+l)
(24"103)
We compute the first few coefficients as follows
r C
 Kc0 2~2.3'
c
_ K c 2 _ K Co 4 " 4 ^ 5 " " 5!
f2 4 104a hi U.4iU4a,Dj
and deduce that Kmc
c ^ = ^TTjT
<24105>
One solution is
~ m=0
Km
2m
(2.4106) Substituting the other value of p in Equation (2.4101 i), we obtain Kc.
cs=7
s 2
wh
(2.4107)
(s  1) (s) In this case, we have two arbitrary constants, c 0 and C}. This implies an even as well as an odd solution. To compute the even solution, we replace s by 2m and Equation (2.4107) becomes (2.4108)
116
ADVANCED MATHEMATICS
As before, we obtain Kmc c2m=^/ For the case where s is odd, we write (24109)
^
" ( 2 m ) ( 2 ^ 1)
(24U0)
In terms of c^, we obtain Kmc
C*=(2^7TJ! <2  4  ul)
Another solution to Equation (2.498a) is
' A , " c. 2 £ ^  + 0 , 5 ) ^ ^
m=0 m=0
(2.4112)
Note that the second term on the right side of Equation (2.4112) is equal to the right side of Equation (2.4106). Therefore, Equation (2.4112) is the general solution of Equation (2.498a). The solution can be written in a closed form by observing that
~
m=0
v
2m
'
"
m=0
v
2m+l
 h x = X (f^ •
Equation (2.4112) can be written as
CA
 h x = X (tTJv
(24H3a,b)
= T"
coshr^K~
+ 7 sinhrVKT
(2.4114)
Differentiation yields ^A dr
=
_ £Q. c o s h r ^ r2
+ c
X ^ s i n h r V ^  ^  s i n h r V ^ + C j ^ ^ cosh TVKT r rz r (2.4115)
and applying the boundary conditions requires cA = ^  c o s h R V ^ + ^  s i n h R V F Ai R R (2.4116a)
SERIES SOLUTIONS AND SPECIAL FUNCTIONS
1JJ
0 = [c 0 cosh r VK + c 0 r VK sinh r V K  c j sinh r VK + c l r VK cosh r V K ~ ] / r 2 (2.4116b) On evaluating Equation (2.4116b), as r—> 0, we obtain c 0 = 0. We now have c A . = ^sinhRVKf and substituting for cj in Equation (2.4114) yields the final result c A R sinh r VK _i r sinh R VK (2.5117)
cA
=
(2.4118)
which is Equation (2) in Gupta and Douglas (1967). Before summarizing the Frobenius method, we note that Equation (2.498a) can be transformed to a much simpler equation involving constant coefficients. Replacing c A by —— transforms the equation to
2
—Ku = 0 dr 2 with solution u = c 0 cosh rv/K~ + c 1 sinh rVfT and with c A correctly given by Equation (2.4114). •
(2.4119)
(2.4120)
We summarize Frobenius's method of finding a solution near a regular singular point as follows. We assume that the solution is of the form
oo
y = Zcnxn+r'
n=0
c 0*°
(2.4121)
We differentiate the series term by term and substitute the resulting expressions in the differential equation. On setting the coefficient of the lowest power of x to zero, we obtain a quadratic (for a
118
ADVANCED MATHEMATICS
second order equation) equation in r. We denote the two solutions by rj and r2. We now consider the following three cases. (a) rj and r2 are distinct and do not differ by an integer. By comparing powers of x, we obtain a recurrence formula that allows us to obtain cj, c 2 , c 3 ,... in terms of c 0 and r\ (or r 2 ). The two linearly independent solutions are
oo
yi = X ^ ( r ^ x 1 1 ^
n=0
(2.4122a)
yi = X
n=0
cn(r2)x"+r2
(2.4122b)
(b)
?i = r2. In this case, one solution yj is given by Equation (2.4122a). To obtain the other solution, we assume a solution of the form
oo
y = y i i n x + £ bnxn+r>
n=0
(2.4123)
We proceed as in case (a) to obtain b n . (c) rj and r 2 differ by an integer. Let us assume that r t > r 2 . The solution y 1 can be obtained as in (a). We try to compute the coefficients c s with the value of r = r 2 . If all the coefficients can be computed as in Examples 2.45 and 6, the second linearly independent solution y 2 is obtained. If, in the computation of c s , we have to divide by zero as in Example 2.44, we assume a solution in the form of Equation (2.4123) and proceed to calculate b n . Note that in case (b) one of the solutions always has a i n x term whereas in case (c) this is not the case (see Examples 2.45 and 6). If one solution is known, we can obtain the second solution by the method of variation of parameters and this is explained in the next section. 2.5 METHOD OF VARIATION OF PARAMETERS
We use the method of variation of parameters to find a second linearly independent solution of Equation (2.12a) if one solution is known. Let yj be a solution of Equation (2.12a) and we assume a second solution to be given by y(x) = u ( x ) y i ( x ) On differentiating and substituting y, y1, and y" in Equation (2.12a), we obtain (2.51)
SERIES SOLUTIONS AND SPECIAL FUNCTIONS
7/g
u(yi' + Piyj + p 2 y i ) + u 1 (2y; + p 1 y 1 ) + u"y 1 = 0 Since yj is a solution of Equation (2.12a), Equation (2.52) simplifies to u ' ( 2 y ; + p 1 y 1 ) + u"y 1 = 0 We substitute u1 by v and Equation (2.53) can be written as
(2.52)
(2.53)
j^=j(^H<fx
On integrating, we obtain i n v =  2 i n y 1  I pjdx
(2.54)
(2.55a)
or
v = (l/y 2) exp  J P l dx
(2.55b)
One further integration yields
u= I (l/yf) exp   p ^ x dx
The function u is generally not a constant and the second linearly independent solution y 2 is y2 = u ( x ) Y l ( x )
(2.56)
(2.57)
Example 2.51. Obtain a second linearly independent solution of Equation (2.468) given that one solution yi is x. We assume that the second solution is y = x u (x) We substitute y, y', and y" from Equation (2.58) in Equation (2.468) and we obtain u" [x 3 (x 2  1)] + u1 [2x 2 (x 2  1)  x 2 (x 2 + 1)] = 0 We denote u' by v and Equation (2.59) becomes x (x 2  1) p + (x 2  3) v = 0 dx (2.510) (2.59) (2.58)
UQ
ADVANCED MATHEMATICS
Equation (2.510) can be written as ^
V
=^ > d *
X(X 2 1)
(2.51 la)
= [—^ + —Lr  2 dx \x 1 x+1 x J On integrating, we obtain v = cl(x2l)/\3 where C\ is a constant. The function u is obtained by integrating v and is found to be u = cj [ i n x + l/(2x 2 )] + K where K is a constant. Combining Equations (2.58,13) yields y = c j x [in x + l/(2x 2 )] + K x The second linearly independent solution can now be identified to be y2 = x [ i n x + l / ( 2 x 2 ) ]
v
(2.51 lb) '
(2.512)
(2.513)
(2.514)
(2.515)
There is no loss of generality in setting cj = 1 and Kx is y j . Equation (2.515) is exactly Equation (2.480b). 2.6 STURM LIOUVILLE PROBLEM
In the previous section, it is stated that the general solution of a second order O.D.E. is a linear combination of two linearly independent solutions. To determine the two constants, we need to impose two conditions. In many physical problems, the conditions are imposed at the boundaries and these problems are boundary value problems. Many of the second order boundary value problems can be stated as follows (r y')'+ (q + A,p) y = 0, subject to a i y ( a ) + a 2 y'(a) = 0 b 1 y(b) + b 2 y'(b) = 0 (2.6lb) (2.6lc) a<x<b (2.6la)
SERIES SOLUTIONS AND SPECIAL FUNCTIONS
727
where r, p, and q are continuous real functions of x, A , is a constant parameter (possibly complex), aj, a2, bj, and b2 are constants. The system defined by Equations (2.6la to c) is the Sturm Liouville problem. Equation (2.6la) can be written as r y " + r ' y ' + (q + Ap)y = 0 (2.6ld)
Many of the equations in mathematical physics are special cases of Equations (2.6la or d) and some of them are given next. (a) Simple harmonic equation y"+ ly = 0 (In this case, r = p = 1, q = 0.) (b) Legendre equation (1  x 2 ) y"  2xy' + Z (Z + 1) y = 0 or [ ( l  x 2 ) y ' ] ' + i ( i + l)y = 0 (2.62b) (2.62c) (2.62a)
[In this case, r = ( l  x 2 ) , q = 0, p = l , X= Z (Z + 1).] (c) Bessel equation We denote the independent variable by x and write Bessel's equation as
2
 2<^J_+ x ^ I dx2 dx
+
(x2v2)y = 0
(2.62d)
On setting x = x VX,, Equation (2.62d) becomes x2y"+xy' + (Xx2v2)y = 0 or (xy')' + (v 2 /x + ?ix)y = 0
2
(2.62e) (2.62f)
(In this case, r = x, q =  v /x, p = x.)
121
ADVANCED MATHEMATICS
(d)
Hermite equation y "  2 x y ' + ny = 0
2 2
(2.62g) (2.62h)
or
(e~ x y')' + )J,e" x y = 0
2
(In this case, r = p = e ~ x , q = 0, X = Li.) In general, the nontrivial solution (y # 0) of Equation (2.6la) depends on X and it is only for some values of X that the boundary conditions [Equations (2.6lb, c)] can be satisfied. These values of X are the eigenvalues (characteristic values) and the corresponding functions y (x, X) are the eigenfunctions (characteristic functions). Example 2.61. Solve the equation y" + Xy = 0 subject to the conditions y(0) = 0, y'(7t) = 0 (2.63b,c) (2.63a)
We assume X to be real and it can be positive, zero, or negative. We consider these three cases separately. (a) X,<0. For convenience, we set A. =  m 2 and the solution of Equation (2.63a) is y = c1emx + c2emx where Cj and c 2 are constants. To satisfy Equations (2.63b, c), we require 0 = C!+c2 0 = m (c, emnc2 e~mn) (2.65a) (2.65b) (2.64)
The only solution of Equations (2.65a, b) is cl = c 2 = 0 This leads to the trivial solution (y = 0). (2.66a,b)
SERIES SOLUTIONS AND SPECIAL FUNCTIONS
J23.
(b)
X = 0. The solution is now given by y = c3 + c4 x where C3 and C4 are constants. The boundary conditions imply c3 = c4 = 0 Again the only possible solution is the trivial solution. (2.68a,b) (2.67)
(c)
X,>0. We set X = n2. The solution is y = c 5 sin nx + c 6 cos nx where C5 and eg are constants. Applying Equations (2.63b, c) yields 0 = c6 0 = c 5 ncosn7c (2.610a) (2.610b) (2.69)
Equation (2.610b) implies that either C5 is zero which leads to a trivial solution or that cos nn is zero which provides the nontrivial solution. The cos function has multiple zeros and cos n7t is zero if n = (2s+l)/2, s = 0, 1,2, ... (2.611)
The system defined by Equations (2.63a to c) has an infinite number of eigenvalues and they are given by
Xs
= f2^1) '
s = 0' 1 ' 2
'
(2.612)
The corresponding eigenfunctions are
ys = sinf^s^Jjx
(2.613)
Note that the eigenvalues are real, positive, and discrete. Such properties are associated for example with discrete energy levels in quantum mechanics. • Next we discuss the general properties of the Sturm Liouville problem.
124
ADVANCED
MATHEMATICS
The Eigenvalues Are Real Suppose X is complex and this implies that y is also complex. Taking the complex conjugate of Equations (2.6la to c) and noting that r, p, q, aj, a2, bj and b 2 are real, we obtain (ry 1 )' + ( q + I p ) y = 0 a 1 y(a) + a 2 y ' ( a ) = 0 b1y(b)+b2y'(b) = 0 where y is the complex conjugate of y. We multiply Equation (2.6la) by y and Equation (2.614a) by y and subtract one from the other to yield (XX)pyy = y(ry')'y(ry')' On integrating, we obtain (2.615) (2.614a) (2.614b) (2.614c)
r b ( h (XX)l pyydx= [yfry^'yCry'Vldx
/a /a
(2.616a)
= [yry'yry']aI (y'r y '  y ' r y') dx Ja = 0
(2.616b)
(2.616c)
To obtain Equation (2.616c), we have used boundary conditions [Equations (2.6lb, c, 14b, c)]. Since the eigenfunctions are nontrivial, Equation (2.616c) implies that X = X, that is to say, X is real. The Eigenfunctions Are Orthogonal Let Xn and Xm (Xn ^ Xm) be two eigenvalues and their corresponding functions are y n and y m . The functions y n and y m satisfy ( r y ; ) I + (q + X n p ) y n = 0 a i y n ( a ) + a 2 y;(a) = 0 , b1 y n (b) + b 2 y'nQo) = 0 (2.617a) (2.617b,c)
SERIES SOLUTIONS AND SPECIAL FUNCTIONS
725
( r y ^ ) ' + (q + ^ m p ) y m = 0 a1ym(a) + a2y^(a) = 0, bl y m (b) + b 2 y ^ ( b ) = 0
(2.618a) (2.618b,c) y m and yn
We proceed as in (i), that is to say, we multiply Equations (2.617a, 18a) by respectively, subtract one from the other and integrate the resulting expression to yield
(anam)l p y n y m d x
Ja
= 0
(2.619)
We have assumed that Xn * Xm and it follows that
rb
Pynymdx = ° Ja
The functions yn and y m are orthogonal with respect to the weight p(x). If Xn = Xm, Equation (2.620) is no longer true and
(2620)
py n 2 dx = In2 Ja
(^0)
(2.621)
The eigenfunctions can be normalized and we define the normalized eigenfunction y n to be yn* = y n ^ n Equations (2.620, 21) can be expressed as (2622)
f P yn* ym dx = f ^ V
Ja Ja n m
dx = 5nm
(2.623a,b)
where 8 n m is the Kronecker delta and is defined by 0 , 5nm = 1 , if n = m if n ^ m (2.624a,b)
Legendre's Functions Legendre's equation can be written as (1 . In this section. polynomial (A.71).2 xy1 + ky = 0 where k is a constant. many such functions have been considered and their properties are listed in Erdelyi et al.75a. it is shown that if Laplace's equation in spherical coordinates is solved by the method of separation of variables and if axial symmetry is assumed. and y" in Equation (2.(k/2) c 0 (2. = 0). one obtains oo oo oo oo X n ( n . which is an ordinary point. In Chapter 5. 1955) and in Abramowitz and Stegun (1970).73) Following the procedure described in Example 2.l ) c n x n . Since the beginning of the eighteenth century.l ) c n x n . Equation (2. We seek a solution near the origin.63a) are hyperbolic functions (k < 0).31. the equation in the 9direction is Legendre's equation with x = cos 0. e. we make a change in the indices so that in all cases the summation starts from r = 0 to °°. g) are referred to as special functions and their properties have been investigated and recorded. The properties of these functions are well known.126 ADVANCED MATHEMATICS 2. Thus the poles (0 = 0.7 SPECIAL FUNCTIONS The solutions of Equation (2. y'. we consider Legendre polynomials and Bessel functions.£ n ( n .74) r=0 We now compare powers of x x°: 2c 2 + kc 0 = 0 => c 2 = .7t) correspond to x = ±1.2 ^ nc n x n + k £ c n x n = 0 n=2 n=2 n=l n=0 (2.71) Y= X n=0 cnx" (272) On differentiating term by term and substituting y.2 . and trigonometric functions (k > 0).x 2 ) y" . The solutions of Equations (2.b) . (1953.73) becomes oo ]T [(r+2)(r+l)c r + 2 x r (r+2)(r + l)c r + 2 x r + 2 2(r+l)c r + 1 x r + 1 +kc r x r ] = 0 (2.62b. and propose y to be of the form oo (2.
SERIES SOLUTIONS AND SPECIAL FUNCTIONS 727 x1: x2: 3«2 c 3 . . ) (2..75g...75h)].d) (2. We note that x = ±1 are singular points and.. . c 3 . c 4 . From Equation (2. c 5 . C5.2 c ! + k c j = 0 => c3 = c 1 ( 2 . in terms of Cj.2 c 2 + kc 2 = 0 xs: ( s + 2 ) ( s + l ) c s + 2 .s ( s . From Equation (2..75h). For convenience. The odd function y 2 is in terms of Cj .k ) / 1 2 (2. c 4 .h) The solution can be separated into an even function and an odd function.710) For values of x = ±l. we determine that the series diverges by comparing it with ^ 1/n.75b) that in this case c 2 is zero and this implies c 4 = c 6 = . From the recurrence formula [Equation (2. The even function involves c 0 . the even solution is a constant and is valid for all values of x including x = ±1.2 . This result can be extended to general values of k.k ) / ( s + 2 ) ( s + l ) (2. .75h) becomes . we deduce that the radius of convergence R is 2 R = (2.b.77a. we consider the special case of k = 0...78) We note from Equation (2. . The general solution can be written as y = c0yi+ciy2 where c 0 and cj are arbitrary constants. = 0 and Yl is given by yi = c 0 (2.f) 4 « 3 c 4 ..76) Um n _ >O o ^±2 cn = 1^ n _ >TO n n +n ~k = 1 (2.79) In this case. can be expressed in terms of c 0 and C3.78).2 s c s + kc s = 0 =} c s + 2 = cs ( s 2 + s .c) 2+3n+1 To examine the validity of the solution at x = 1. and we denote this function by yj.. c 2 .75c. we set k = Z (Z + 1) and Equation (2. one solution is valid at the singular points and the other is not.75h) now becomes c s+2 = s c s / ( s + 2) (2.l ) c s . Equation (2. for this special value of k. we deduce that y 2 is given by y 2 = Cj x (1 + x 2 / 3 + x 4 /5 + x 6 /7 + .75e.2 c 2 . c 2 .k ) / 6 => c4 = c2 ( 6 ....
if i is odd. These polynomials are the Legendre polynomials and are denoted by P^ (x). The constants c 0 (or q ) are chosen such that P^ (1) is unity. In general.1).71 Legendre polynomials A relatively easy method of computing P^ (x) is to use Rodrigues' formula which can be written as P^ (X) = .d) P.3x) (2.i ) ( s + i + l)/(s+2)(s+l) (2.711) From Equation (2. The first few Legendre polynomials are shown in Figure 2.J — A _ ( x 2 . for any integer i . The case we considered earlier is k = 0 {Jt = 0) and the even solution is a constant. we observe that if Z is a nonnegative integer. the even solution (a polynomial) is valid for all values of x and.U) PpU) * i\ ' P*(XX P(x J // yf l \ / \ /// .712a.l( FIGURE 2. P2 (X) = 1 (3x2 .713) . P^x) = x P3 (x) = 1 (5x2 . if i is even. c^+2 is zero and so are CZ+A' ci+6> •• ^nus tnc mfiniie series becomes a polynomial and the solution is valid for all values of x.i 2 . the odd solution is a polynomial.128 ADVANCED MATHEMATICS c s + 2 = c s ( s 2 + s .712c. In particular.l / 2ZJt\ dxZ (2.i ) / ( s + 2 ) ( s + l ) = c s ( s . the degree of the polynomial is i .711).71 and are P 0 (x) = 1 .b) (2.
The orthogonal property can be written as ( P i (x) Pm(x) dx = — 1 — 8e m Jl 2i + l where 8^ m is the Kronecker delta.71) can be obtained by the method of variation of parameters. and the integrating factor IF. 20.720) (2.2xP/J + i ( i + 1) Pj] + (1x 2 ) (u"P i + 2u'P<J) . we can calculate ?2.715) ( i + l)Pj + 1 (x) = (2i + l ) x P i ( x ) .2xu'P i = 0 Since P^ is a solution of Legendre' s equation. The second linearly independent solution of Equation (2.62c. From Equations (2. we obtain u [(1x 2 ) PJ . Equation (2.E. is (2. P3.718) simplifies to (lx2)Piu"+uI[2(lx2)P/J2xPi] = 0 On writing u' = v.717) (2...718) . l = X l * pi i=o to (2714) V l2xt+t 2 Another method of determining P^ (x) is to use the recurrence formula which can be written as (2.720) is a first order O. Equation (2.SERIFS SOLUTIONS AND SPECIAL FUNCTIONS 129 The Legendre polynomials can also be obtained by using the generating function (l2xt+t ) Expanding this function in powers of t yields oo .i P i _ 1 ( x ) Knowing Po and Pj. we deduce that the Legendre polynomials are orthogonal with respect to weight one. 21).716) On differentiating Q^ (x) twice and substituting the resulting expressions in Equation (2.)x/(lx2)] = 0 dx A * Equation (2. An important property of the Legendre polynomials is the orthogonal property..719) can be written as 4 x + 2v[(Pl/P. We denote the solution by Q^ (x) and assume that Q i (x) = u(x)P i (x) (2.719) (2. .71) with k = i ( i + 1).D.
725a) = %} [T^ + T7i] d 5 = fin[j^] It is usual to choose C to be one and Q o (x) is given by (2'725b) (2.723) Q^ (x) = C P .)2x/(lx 2 )]dx (2. It follows that Q^ (x) is given by /•x (2.720) can be written as jL [V (1x 2 ) P / ] = 0 The solution is v = C/[(lx 2 ) P / ] where C is a constant.721b) (2.721c) = exp [IJlnPjt + i n (1x 2 )] = (1x 2 ) P / Equation (2. Example 2. we have /•x Qo (x) = C I . 24).724) ] (i$)p?($) The functions P^ (x) and Q^ (x) are Legendre's polynomials of the first and second kind. (x) I ^ (2. = exp [(2P / J/P /( .722) (2.7250) . From Equations (2.72 la) (2.UO ADVANCED MATHEMATICS IF. /•x ^ (1$) (2. Q o (x) and Qj (x).71. Calculate the Legendre polynomials of the second kind.712a.
x 4 / 4 .726.730) From Equations (2. The infinite series given by Equations (2.7 " 28c » (2. the odd solution Pj(x) (= x) is valid at  x  = 1 and the even solution Qj (x) is not valid at  x  = 1. 29). we note that Q o and Qj have singularities at x = ±1...x 2 / 2 + x 3 / 3 .x .728a) = Cx J p + * (T^+TT$ p ( 2 728b) = Cx [M in (T^)] As in the case of Q o (x).712b. From Equations (2. For I = 0 ( i is even).)] = x + x 3 /3 + .727a) (2. we obtain (2.x 2 / 2 .x 4 / 4 + .. we choose C to be one and Q^ (x) is given by < 2 .( . 24). the even solution P 0 (x) (=1) is valid at  x  = 1 and the odd solution Q o (x) is not valid at  x  = 1.726) (2..729) Qx(x) = 2 L i n J j i i ) . + 1 = x (1 + 2X) Q .731) . ) (2. . • The other Legendre functions of the second kind can be computed from the recurrence formula ( i + 1) Qj.x 3 / 3 .727b. Likewise for i = 1 ( i is odd). we obtain Qx (x) = ..1 + 1 (x 2 + x 4 /3 + .SERIES SOLUTIONS AND SPECIAL FUNCTIONS 131 Qo (x) = 1 in [ 1 ± A ] Expanding Qo (x) about the origin yields Q 0 (x) = l [ x . 30) are the infinite series solutions of Legendre's equation and are valid for  x  < 1.727b) Q1(x) = C x  ^ (2....l Expanding the Jin function in powers of x. ..X Q ^ j (2.
The Legendre polynomials Pg (x) satisfy the equation ( l . Equation (2.x2) represents the nonsymmetric contribution. (x) where A and B are constants and Z is a nonnegative integer. We start by considering the simplest case (m = 1) and Equation (2.2x ^ + [X ( i + 1) .£ .x 2 )] y = 0 dx z dx where Z and m are integers.71)]. The general solution of Legendre's equation can be written as y = A Pj. The additional term m 2 y / (1 .736) (2.732) One would be tempted to introduce a series solution. The function Q^ (x) is singular at  x  = 1 and if we require the solution y to be finite at  x  = 1. B has to be zero.733) (2.733) reduces to the standard Legendre equation [Equation (2. the following procedure provides us with an ingenious way of solving the problem.(1 .x 2 ) P 7 4xPJ' + i ( i + l)Pj = 0 A new function w (x) is defined by w(x) = ( l . If m is zero.m 2 /(l .2 x P j +Z (Z + l)Yz Differentiating with respect to x yields =0 (2.x 2 ) P j .x 2 ) ^ .734) (2. Equation (2.735) ( l .132 ADVANCED MATHEMATICS Note that the recurrence formulae for both P^ [Equation (2. The associated Legendre equation can be written as (1 .2x ^ + [Z (Z + l). (x) + B Q^.733) becomes (1 .715)] and Q^ [Equation (2.737a) . however.537b)].x 2 ) ^ .733) is derived from Laplace's equation in spherical coordinates [see Equation (5.x 2 ) 1 7 ^ (2.x 2 )" 1 ] y = 0 dx dx z (2.731)] are identical.
736) yields ( l . P^ 1 (x) is zero.737b.740) Similarly the associated Legendre function of the second kind can be computed from the formula dmO ^ Q?(x) = ( l .x 2 ) m / 2 (2. By convention.740).2 x w ' + w [ i (i + l ) . Note that due to the term (1 .737a). P™ is a polynomial iff m is even. From Equation (2. the solution of the associated Legendre equation is denoted by P^ (x).733b) is y = AP™(x) + BQ™(x) where A and B are constants. If y is finite at  x  = 1. b) in Equation (2.741) The functions Q™ (x) are singular at  x  = 1. we deduce that PJ(x) = (lx 2 ) 1 / 2 P](x) In the general case. The function P™ is defined for Z > m.x 2 ) m / 2 *• (2.742) .738a) (2.738b) 2x (1 . 38a. (x) is given by P^(x) = ( l . The polynomial P .x 2 )~ 3/2 w'+ w (1 + 2x 2 ) (1 .x 2 r 5 / 2 Substituting Equations (2.737b) On differentiating.740) that if m > Z.x 2 r ! / 2 w ( x ) (2.x 2 ) w " . we obtain p]' = (l _ x 2 r 1 / 2 w' + x (1 .x 2 )~ 3/2 w P^" = (l X 2 ) .739) dmP dxm (2.SERIES SOLUTIONS AND SPECIAL FUNCTIONS Jj£ or P^ = ( l .734) with y = w.734) This is Equation (2. P. is of degree Z and we note from Equation (2.( l . in Equation (2.I / 2 w "+ (2. B has to be zero.x 2 ) " 1 ] = 0 (2. The general solution of Equation (2.x 2 ) m (2.
v 2 ] = 0 xr+s: c s [(s + r)(s + r .745) Comparing powers of x.l ) + (s + r ) .v 2 ] (2. Equation (2. C4..c s _ 2 / [ s ( s + 2v)] (2.58a.743) yields oo £ n=0 [(n+r)(n+rl)c n x n+r + (n+r)c n x n+r + c n x n + r + 2 . Writing s = 2p.f) For r = v. lb)].. Bessel's equation occurs in the solution of Laplace's equation in cylindrical coordinates [see Equation (5.b) (2. we obtain xr: co[r(rl) +rv2] = 0 => r = ±v (c o *O) => cl = 0 (2.746e.v 2 ) y = 0 where v is a constant.746f) becomes c s = . It can be written as x 2 y" + xy' + (x 2 .d) x r + 1 : cx [ r ( r + l) + (r+ l ) . The origin is a regular singular point and we seek a solution of the form oo (2. Equation (2. eg. we can compute the even terms C2.746c.v 2 ] + c s _ 2 = 0 => c s = . in the discussion of a problem in astronomy.746a.v 2 c n x n + r ] = 0 (2.747) From Equation (2.747) can now be written as C2p = ~ c 2p2 / P 2 P (P + v )l (2..134 ADVANCED MATHEMATICS Properties of the Legendre and the associated Legendre functions of both kinds are listed in the references cited earlier.c s _ 2 / [ ( s + r) 2 .743) y=X n=0 cnx"+r (2.747). Bessel Functions Bessel functions were introduced by Bessel in 1824.748) .744) Differentiating term by term and substituting the resulting expressions in Equation (2.
b) .748).l ) P c 0 / [ 2 2 p ( p ! ) ( l + v ) ( 2 + v ) .750) satisfies Equation (2.753) r (v + 1) = [tv e~l] o + v I t v . (p + v)] p=0 (2.c 0 /[2 2 (1 + v)] c 4 = .752) Jo The condition that v is positive is necessary so as to ensure the convergence of the integral.c 2 / [ 2 2 ( 2 ) ( 2 + v)] = c o / [ 2 4 ( 2 ) ( l + v ) ( 2 + v)] c 6 = .749b.750) It can be verified that Equation (2.754b) f°° F(l) = JO e" l dt = 1 (2. To give a meaning tov! when v is not an integer. .749d.754a) = vT(v) From Equation (2. ( p + v)] (2. we define the gamma function F(v) by r(v) = I t ^ e ^ d t .. One solution of Equation (2.751) If v is a positive integer.c) (2..749a) (2. we deduce that (2.1 e~l dt JO (2. (p + v) can be written as (p + v)! /v! . the product (1 + v) (2 + v) . r (v + 1) = I tv e~l dt Jo On integrating by parts.755a. v>0 (2..753).c 4 /[2 2 (3) (3 + v)] = .D P x 2 p + v / [ 2 2 p ( p ! ) ( l + v ) ( 2 + v).e) = ( .743) can be written as oo y = c o X ( .c 0 /[2 6 (2) (3) (1 + v) (2 + v) (3 + v)] Equations (2. ..SERIES SOLUTIONS AND SPECIAL FUNCTIONS 135 The first few coefficients are c 2 = .749a to e) suggest that C2P (2. we obtain (2.
.Ite ADVANCED MATHEMATICS Combining Equations (2. for all positive v. Problem 9b) F(l/2) = VTC Equation (2. in terms of gamma functions. that F(v) is ±°° for all negative integers. we obtain r ( v + l ) = v! for v > 0.758).b) If .754b.l < v < 0 .759a. we define F(v) as F(v) = [ r ( v + l)]/v (2.2 < v < .b.758).)• That is to say. Similarly. The graph of F(v) is shown in Figure 2. we deduce that F(0) can be defined as F(0) = lim F(v) = Iim r ( v + 1 ) = ± ~ v—>0 v—>0 V (2. .l ..754b) can be used to define F(v) for all values of v (* 0. That is to say oo J v (x) = X p=0 oo (l)Px2p+v/[22p+v(p!)F(p + v+l)] (2.1 and 0.758) and 0! = 1 (2. T(3) = 2 F(2) = 2 • I. r(4) = 3 r(3) = 3 • 2 • 1 (2. the function F(v) is defined for all negative nonintegers.760) c (2.2 . and F(v + 1) has just been defined. Similarly if . as y = o X (l)Px2p+vr(v+l)/[22p(p!)F(p + v+l)] p=0 (2.762a) = £ p=0 (l)p(x/2)2p+v/[(p!)F(p + v+l)] (2. then v + 1 lies between . We note that (see Chapter 4.756ae) Generalizing Equations (2. then 0 < v + l < l and T(v + l) is defined.756a to e).762b) .1 ..761) By choosing c 0 to be 1/[2V F(v+1)]. 55b) yields T (2) = 1.758) that F(v) is defined.757a.72. we obtain the Bessel function of the first kind of order v and it is denoted by J v (x). It follows from Equation (2. From Equation (2.751) can be written.c) It follows from Equation (2. Equation (2. .
7'43) is J v (x). We consider the two cases where 2v is an odd or an even integer separately. If v is not an integer or zero. r 2 = . (2. \ / / ^ / I I \/TT 3* 2* I* W*\ I 3 r 'i ! * ! ! ! . For v (* 0 or an integer). the other solution is obtained by considering the other root of the indicial equation [Equation (2. B must be zero.72 Gamma function Thus one solution of Equation (2. the general solution of Equation (2. If v is zero.763) (2.v ) is an integer. In Example . this implies that 2v (rl=v. J_v (x) is singular at the origin.* • i i n FIGURE 2.v /[(p!)r(pv+l)] Note that whereas J v (x) has no singularity at the origin.743) is y = AJ v (x) + BJ_ v (x) where A and B are constants. That is to say.43. the two linearly independent solutions are not obtained in a straight forward manner as described earlier for nonintegral values of v. If the solution is finite at the origin. the two roots are coincident and this case has been considered in Example 2. the other solution is J_v (x) and is written as oo J_v(x) = X (D P (x/2) 2 p . If the two roots differ by an integer.746b)].SERIES SOLUTIONS AND SPECIAL FUNCTIONS ZiZ ! U *\ ] I I ' r(u)l I I 2.764) We recall that if the two roots of the indicial equation are coincident or differ by an integer.
J v (x) and J_v (x) are not linearly independent.J_v (x)] / sin vn (2.v + l) is ±°° for p < v .45.743) is given by Equation (2.768) are zero.762b.484d)] and we have one solution starting with c 0 and the other starting with cj. (2.763) can be written as oo L V W = X (D q+V (x/2) 2q+V /[(q + v)!(q!)] q=0 (2. the Bessel function of the second kind Y v (x) is defined as Y v (x) = [Jv (x) cos v% . We recall that if 2v is one. Instead.764 or 69). J v (x) and J_v (x) are obtained in the case where 2v is an odd integer. we obtain J 1 / 2 ( x ) a n ^ L1/2 00.769) . B is zero. we deduce that Y v (x) exists in the limit as v tends to an integer. If v is not an integer. we deduce that J_ V W = ( . we solved the case where 2v is one. This method yields (2. we conclude that both the numerator and the denominator on the right side of Equation (2. the other linearly independent solution can be obtained by the method of variation of parameters.765) Comparing Equations (2. J x[J v (x)] 2 (2767) The solution y2 is usually not considered.138 ADVANCED MATHEMATICS 2.v . v is an integer and from Equation (2. By applying l'Hopital's rule. Writing q = p . the general solution is y = AJ v (x) + BY v (x) where A and B are constants. Equation (2.768) From Equation (2. Thus. The function Y v (x) has a in(x/2) term (see Problem 19b) and is singular at the origin. Since J v (x) is known. If y is finite at the origin.Similarly.D V J v ( x ) If v is an integer. the solution of Equation (2. q is not necessarily zero [see Equation (2. If 2v is an even integer. but if v is an integer only Equation (2. By assigning to c 0 and Cj the values given earlier.769) is valid.763) we note that the series starts from p = v since F ( p . 65).766).766) yi = J v 00 I d]L—.
b).771a) (2.74.73 and Yo and Yl in Figure 2.743) instead of J v (x) and Y v (x). JuU) FIGURE 2. we obtain Jv(x) = i(Ht1) + Hf) Yv(x) = i ( H f . We recall that there are circumstances when it is preferable to work with exp (± ix) rather than with sin x and cos x.77 lb) The functions Jo and Jj are shown in Figure 2.770b) are linearly independent and the general solution of Equation (2. From Equations (2. Equally. there are circumstances when it is preferable to choose Hankel functions (Bessel functions of the third kind) of order V as solutions of Equation (2.743) is a linear combination of H^.770a. and H^ . Table 2. are defined by H ^ } = J v (x) + iY v (x) H f = Jv(x)iYv(x) The functions H^ and H^ (2.71 lists some properties of Bessel functions. These Hankel functions H^ and H*.< } ) (2.770a) (2.73 Bessel functions of the first kind .SERIES SOLUTIONS AND SPECIAL FUNCTIONS US.
x . Yv (x).v J v + 1 (x) J ^ W + J^Cx) = ^ J v ( x ) J v _ i W .74 Bessel functions of the second kind The functions J_v (x). TABLE 2.v J v (x)]' = .71 Properties of Bessel functions (i) (ii) (iii) (iv) [x v J v (x)]' = x v J v _!(x) [x. H^ (x) and H^ (x) have the same recurrence relations as the function J v (x).J v + 1 ( x ) = 2J^(x) I x v J v _ 1 (x) dx = x v Jv (x) + constant I x" v Jv + j (x) dx = x~v Jv (x) + constant ^v) (vi) .140 ADVANCED MATHEMATICS Y v (x) I Y 0 (x) / o —/—\i / \ \ '—\/ /^~xr • y» » ' \ — \ / iL / V^O^ x FIGURE 2.
45.774) (2.. (2. we have solved the Bessel equation for the case v = +1/2 and the two linearly I/O 1 fj independent solutions were x sin x and x cos x [Equations (2. n = l . Compute J3/2(x). . b)]..SERIFS SOLUTIONS AND SPECIAL FUNCTIONS Mi The function J v (x) has an infinite number of zeros and we denote the zeros by Xn. we obtain J 3/2 00 = k J1/2 (x) " J 1/2 = \H^ V ir.. We recall that in the definition of J v (x). we note that they differ only in the coefficient of y. In Example 2.776a) (2. we deduce that J 1 / 2 (x) and J_ I / 2 (x) are given by JmW = {£*>* J " 2(X) = (2.(x 2 + v 2 ) y = 0 (2. n = 1.772) The orthogonal property of J v (x) can be written as (l I x J v (Imx) J° 0 J v (Xnx) dx = i[Jv + i ( ^ n ) ] 2 ifm = n if m * n (2.b) Example 2.776b) (siSJL _ c o s x ) v x / Modified Bessel's Equation The modified Bessel's equation is of frequent occurrence in applied mathematics [see Equation (5. By writing z = ix. 77).777) becomes . the multiplicative constant (c 0 ) is 1/[2V F ( v + 1 ) ] .72. Using the properties of the gamma function. 2. 2.743.775) V^C°SX From Table 2.773a.x to (2.45..495a.777) Comparing Equations (2.71. using the results of Example 2.58a)] and can be written as x 2 y" + X y' . Equation (2.. That is to say J v ( ^ n ) = 0.
the two linearly independent solutions are I v (x) and I_ v (x). We seek a series solution as described previously for Bessel's equation. p. 1963.781) The function K v (x) is linearly independent of I v (x) and the limit as v tends to an integer is defined. 524) d4 Y d2 Y —*A.i 7 l v / 2 J v (ix) (2. is real.777) for all values of v (including integral values) are I v (x) and K v (x). The function I v (x). That is to say. defined by Equation (2. A simplified form of the equation governing the linear stability of a Newtonian fluid flowing between two parallel walls can be written as (Rosenhead. Kamke (1959) and Murphy (1960) list several such possibilities.779) Comparing Equations (2. we define a new function K v as (2.73. the solution needs to be given in the form of real variables and J v (ix) and J_ v (ix) are not in a suitable form.778) Equation (2. If v is an integer.777) are J v (ix) and J_ v (ix).779).778) is exactly Equation (2.b) Note that we have already established that the solution of Equation (2. Example 2. The properties of all Bessel functions are given in Watson (1966). Usually.777) is J v (ix) and to obtain the real part of the solution we multiply J v (ix) by a complex constant (i~ V ).782) . The solution we obtain can be denoted by I v (x) as follows I v (x) = ]T (x/2) 2 p + v /[(p!) r ( p + v + 1)] p=0 (2.iri — ^ . we deduce that I v (x) = T v J v (ix) = e .v 2 ) y = 0 dz dz (2.142 2 ADVANCED MATHEMATICS z2 ^ + z ^ + (z 2 . they are Bessel functions with purely imaginary argument.762b. 79). If v is not an integer. The two linearly independent solutions of Equation (2.743) with z being replaced by x and the linearly independent solutions of Equation (2.= 0 dr 4 dr 2 (2. Many second order linear differential equations can be transformed to Bessel's equation (or to another standard equation) by a suitable substitution.780a.
2 . we obtain n ^ + T T 1 / 2 <*v .786) yields z 2 . Equation (2.785) 1 / 2 (2786) We further transform the independent variable TI by introducing z as z = (iTl)3/2 The chain rule yields (2. Since we are dealing with complex functions.788c) Substituting Equations (2.784) (2.SERIES SOLUTIONS AND SPECIAL FUNCTIONS IJ2_ where % 0 is related to the stream function and r\ is related to the distance from the centre of the channel. c) in Equation (2. The two linearly independent solutions of Equation .787) (2.784).789) Equation (2.782) can be reduced to a second order equation by writing d2Yn u = —*&• dri 2 Equation (2. — + z+(z2l/9)v = 0 dz 2 dz (2.i T ^ v = 0 dT)2 dri 4 (2.783) —iTiu = 0 dTi2 We now write u = ri 1 / 2 v Differentiating and substituting into Equation (2.b) (2.782) becomes 2 (2.l T T ^ v .788a.789) is Bessel's equation of order 1/3.782). Both quantities are complex and we wish to solve Equation (2.788b. we write the solution in terms of Hankel functions.
••• . We recall that the eigenfunctions y n (x) and y m (x) are orthogonal with respect to weight p (x) if b / 0 if m ^ n (2. we can choose n vectors g j .784) are »l=V/2H«[f(in)3/2] u2 = V C H ( »[(i 1 1 ) 3 / 2 ] On integrating twice.b) The product g j ' g . which are presented in more detail in Chapter 4. g 2 . the dimension of the space is infinite. ( o r / g j . T h a t is t o s a y .789) are H ^3 (z) and H 1 / 3 (z). 2.790a) (2. If u is any vector in the space. Legendre polynomials P^ (x) and Bessel functions J v (^ m x) are among the eigenfunctions we have encountered. It follows that the fundamental (linearly independent) solutions of Equation (2. g : \ ) is the scalar (dot or inner) product and Ij is the magnitude of the vector g j.8la. the dimension of the space is finite and if the dimension is n. These square integrable functions generate a space ( L 2 space) and since n can be infinity..8 FOURIER SERIES (2. we have seen that the SturmLiouville system generates orthogonal eigenfunctions. g n which are linearly independent as bases. Usually. These bases are orthogonal (not orthonormal) if 0 gigj = (gi.6.82a. u can be expressed as u = X ci l i i= l (28"3) w h e r e q are constants a n d are obtained b y forming the d o t p r o d u c t o f u w i t h g .144 ADVANCED MATHEMATICS (2.b) if m = n I p(x)y n (x)y m (x)dx = Ja \ In We assume that the integral exists and this means that the functions are square integrable.gj) = I. 2 (*0) ifi=j if i * j (2. cos nx.790b) In Section 2. Trigonometric functions sin nx. The orthogonal property of y n reminds us of the property of orthogonal vectors and we state some of the properties of vectors. we obtain two solutions for % 0.
J cnyn < e n=l (2. That is to say ly n ll 2 = ( y n .b. y n ) = p(x)f(x)yn(x)dx (2. Forming the inner product. we find that c n is given by cn = { f > y n ) /   y n H 2 (2. The magnitude yn.8la. g j ) / ( g j . The sum converges to f (x) if f ( x ) .89) .s6a. gj) = X i=l c i(li'Ij) =c jIj2 (2. g j ) / I j 2 = ( H . we now continue with function spaces and we regard the functions yn as the basis of the infinite space. we consider pointwise approximation and the difference between the sum and the function f (x) is small for all values of x in the interval. b) as the inner product.88a) (f. we make use of the orthogonal property of y n . b]. y n ) = p(x)y n 2 (x)dx = i 2 (2.SERIES SOLUTIONS AND SPECIAL FUNCTIONS /£> (U. and the integral in Equations (2.88a).88c) We need to examine in what sense the sum (letting N —> «>) approximates f (x) in the interval [a.88b) Ja llynll 2 = (b /a p(x)[y n (x)] 2 dx (2. g j ) (2. which is denoted by   y n   is the norm of y n .85a. Usually.b) Having been inspired by these properties of vector spaces. with c n defined by Equation (2. C ) We now represent a function f(x) by a linear combination of yn and write N f(x)« X n=l cnyn(x) (287) To determine the constants c n .b) We deduce that Cj = ( u .84a.
146 ADVANCED MATHEMATICS for each e > 0 whenever N > N e and for all x in the interval. we obtain an j ~ = s (b N p(x)[2fy s + 2 y s X c n y n ] d x Ja n=l and using the orthogonal property of yn. to the mean value of f(x) as x approaches x 0 from the left and from the right. If D —> 0 as N —> °°. The function y n can be continuous though f (x) can be discontinuous. the set of functions (y l5 y2. s = l .) is complete and the space they span is a Hilbert space.[f(x 0 + ) + f (x o _)].X c n y n ] dx = Ja n=l Ja X c n y n +(X cnyn) ]dx n=l n=l (2.81 lb) ( p(x)[2fy s + 2c s y s 2 ]dx Ja (2. = 0. we require that the integral I p (x) [f (x) ~2J c n y n ] dx be a minimum. If for rb every f (x) for which j •'* 2 £ converges m p (x) f dx is finite.88a).810a. 2 that is to say. . the sum ^ c n y n converges in the mean to n=l f (x). Let Ja (b D= N 2 (h P(x)[f22f N N 2 P « [ f .81 la) On differentiating. . 2. Note that in this case the requirement is that the integrals exist and the function can have a jump discontinuity in the interval. X c n v n n = l the mean to f (x). the sum ^ n=l cnvn c o n v erges to ^.. we write f^= acs (2.8llc) Minimizing D implies that c s is given by Equation (2.b) D is a function of c n and D is a minimum if  5 . The approximation in this case is in the N least square sense. N If f(x) has a jump discontinuity at x 0 . Another approximation which is widely used in the treatment of experimental data is the least square a p p r o x i m a t i o n . Using the least square criterion.. (2.. The series is the Fourier series and the coefficient c n is the Fourier coefficient.. We demonstrate that this is the present case..
n a =1 71 I f(x)cosnxdx Jn .814c) The coefficients a n and b n given by Equations (2.SERIES SOLUTIONS AND SPECIAL FUNCTIONS MZ Trigonometric Fourier Series The trigonometric functions sin x and cos x are periodic with period 2TC. b] to [n .aQ and not a 0 so that the formula for a n [Equation (2. c) are special cases of Equation (2. If f (x) is defined in the interval [a.a0 + 2^ [a n cosnx + b n sinnx] 2 (2. the Fourier series [ p ( x ) = l ] . That is to say oo f(x) = i. The functions sin x and cos x form an orthogonal basis in the interval [71. The orthogonal property of cos x and sin x can be written as (2.c.813a) r (2.812) I sinnxcosmxdx = 0.a ) transforms the interval [a.7t] with unit weight fK f 2 dx is bounded.88a).814a) n=l . b]. Jn r r for all m and n (2. If f(x) is periodic and of period 2TC and I Jn converges in the mean to f (x).814b) b n = JJn f(x)sinnxdx (2. A periodic function f(x) of period T is defined by Equation (1.114).814b.d) I sinnxsinmxdx = I cosnxcosmxdx = n 8 n m Jn Jn where 8 n m is the Kronecker delta. Note that we have written j . For simplicity. 7t].814b)] is . we assume f (x) to be defined in the interval [%. then the transformation x* = 27t[x(a + b ) / 2 ] / ( b . n].813b.n (2.
we obtain and at x = n ? a0 = 1 Jn iTt x dx = 1 [ "K =0 (2.148 ADVANCED MATHEMATICS applicable to a 0 . The function is continuous at x = 7c/2 and the series converges to the function.816g.2 ^ n=l °° .sin (im/2) (2.815) and periodic with period 2TE.81.81.h.sinnx n= l (2. Determine the Fourier series for the function f (x) defined in the interval (TT.c) i71 an = l j xcosnxdx = l[xsinnxj n _ 1 I shmx Jn ~n JK dx = 0 (2. n2s+l ^—.816d.817) The periodic function f(x) is sketched in Figure 2.814b. 7t<x<7i (2.f) J " 71 e 71 xsinnxdx = IFXcosnxj" + 1 /7C cosnx dx = .2 ]T t ^ . %) by f(x) = x.b. Recall that if f(x) has a jump discontinuity at x 0 . the series converges to ^ [ f ( x o + ) + f(x o _)].814a).i) 71 Substituting a n and b n in Equation (2.816a. That is to say 7X/2 = . To what value does the series converge at xn/2 From Equations (2.e. 2 ( ^ 1 (2. Example 2.818b) (28"18c) = 2 i(5TTT) s=0 .818a) = 2 ^ s=0 (^TlTSin[(2s+1)7C/2] (2. we obtain x . c).
SERIES SOLUTIONS AND SPECIAL FUNCTIONS 74£ We note that sin (nrc/2) is zero when n is even and if n is odd. . That is to say ).820) are zero. Both sides of Equation (2. f(7i0) is n.818c). .0) + f (JI + 0)] = .2..[f (K .81 From Equation (2.. and sinn7C is zero. f(7C + 0) is it. 1. f(x).0) + f (% + 0)]. and sin [(2s + 1 ) T C / 2 ] = ( . the function is discontinuous and the sum converges to j . 37T ZTT/ IT / IT ZTy 37r FIGURE 2. • Even and odd functions are defined in Chapter 1.[f (7t .2 ] £ ^ n=l sin n7t (2.l ) s .81. we write n = (2s + 1 ) with s = 0. we deduce that Periodic function of period In XST S=0 =7 C / 4 ( 2 8"19) At x = 71.820) From Figure 2.
821) The product of two even (or odd) functions is an even function and the product of an even and an odd function is an odd function.823b) (2. we obtain oo (2. I n f (X) sin nx dx (2.824a) fn b n = 2.823a) an = 2.825a. If f (x) is an even function.150 ADVANCED MATHEMATICS The functions cos x. x 2n (n is an integer) are even functions and the functions sin x. x^ 2n + 1^ are odd functions.814atoe) reduce to oo f (x) * 1 a0 + £ n=l an cos nx (2.824b) Jo Note that we have made use of the fact that fn 0. If f(x) is an even function f'(0) = 0 and if f(x) is odd f(0) = 0 (2.823c) f(x) » £ n=l b n sinnx (2. it is possible to define f (x) to be even or odd. Mathematically.b) if f(x) is even In many situations.822) (2. I f(x)cosnxdx n Jo bn = 0 Similarly. all coefficients b n are zero and Equations (2. it is known that f (x) has a period of 2% but is defined only in the interval 0 < x < 7C. if f (x) is odd I Jn f(x)dx = I 21 f(x)dx. if f (x) is an odd function. Jo r7 C (2. but usually the physics of the problem dictates whether f (x) is even or odd (see Chapter 5). .
Example 2. 23b). we obtain an = 2.827d) Jo ~.tj (2s+l)2 (2.82..82. it implies that at x = 0.T o we obtain f (0) = 0.SERIES SOLUTIONS AND SPECIAL FUNCTIONS £££ Consider a rod of length K in the interval 0 < x < 71 with temperature distribution T(x). we deduce that T is an odd function. The function f in case (b) is sketched in Figure 2.821).. by writing f(x) = T .827a) (2. If the temperature at x = 0 is kept at a constant temperature T o . 1. 0<x<rc (2. The function f (x) is periodic and of period 2K. From Equation (2. and (b) f is even.2. (2. If the end x = 0 is insulated.822). Case (a) is considered in Example 2.823a) yields a 2 _ 4 y cos(2s+l)x «. f.827c) s = 0.n = 4 n (2.81. . It is defined by f(x) = x. From Equation (2.828) .\QQ^L] n\.816e.826) Obtain the Fourier series of f (x) if: (a) f is odd.827e) 71 Jo Substituting a n in Equation (2. 7t(2s+ l ) z a0 = 2 x dx = 71 (2. dT/dx is zero. I x n h cos nx dx (2. we deduce that T is even.827b) = 2. From Equations (2.
we can write . that is to say. This means that t x e x oo f x I /x 0 f(x)dx = i . If f(x) has a Fourier series. Under the conditions stated earlier. ADVANCED MATHEMATICS f(x). but not in the interval K < x < 0. /x 0 aodx+^T I n = l /x 0 (a n cosnx + b n sinnx)dx (2. 27T 7T 0 7T 27T x FIGURE 2.U2. we deduce that y l—  sTo(2s+l)2 = — 8 (2. f (x) can be represented by Equations (2.828).829) If f (x) has a Fourier series.814a to c). if f'(x) is piecewise continuous and f (—TC) = f (n).82 Even function Note that both series on the right side of Equations (2.830) where x 0 is arbitrary. At x = 0.817. the series can be differentiated term by term. the series can be integrated term by term. 28) represent the same function in the interval 0 < x < K. the function is continuous and from Equation (2.
833d. Equations (2. TU<X<7C (2.i n x )] n=l OO (2.i sin nx] dx = ±.834) and with period 2n.\ ao + \ Z P™ K . .e ~ i n x ] Substituting Equations (2.f) Note that Equation (2.832a) (2. c).e.814a) yields oo (2. Calculate the sum of the series at X = 71.83.l a o + l j [an(einx + e.SERIES SOLUTIONS AND SPECIAL FUNCTIONS oo 755 i[f (x .833b)  X n=oo cneinX (2.814b.832b) f(x) .814a to c) can be written in an alternative form.833C) where cn = ^(anibn) = ^ I f (x) [cos nx .i n x )ib n (e i n x e.I f(x)einxdx (2. b) in Equation (2.83.831) where a n and b n are defined by Equations (2.833a) .0) + f' (x + 0)] = ]T [n(a n sinnx + b n cosnx)] n=l (2. This function is shown in Figure 2.832a. Example 2. Determine the Fourier series for f(x) defined by f(x) = e x .833f) is valid for both positive and negative values of n. Recall that cosnx = 1 [e i n x + e" i n x ] sinnx = JL [ e i n x .ibn> + e ~ i n X (an + ibn)] n=l oo (2.
835a) (2.833c.833f)] .835d) 2TI(1+n2) = (1+in) [(_i) n sinh 7 C ] 7C(l+n 2 ) Combining Equations (2.n e x e " i n x dx cn = .835c) (2.836a) .i sin nic) .i n J ^1+in) [e71 (cos nit .e~n (cos nre + i sin nn)] 2n (2.835a) FIGURE 2.835b) J_K " 2n[ l . U+n2) (2.835b) (2.^ \ (2.154 ADVANCED MATHEMATICS f(x) T 0 V 3 IT x (2.83 Periodic function ex The coefficients c n are given by [Equation (2. 35d) yields ex „ sinhjt 71 f (Dn(l+in)ein* »r.
840a) . we deduce oo (1) cosn7t (2.838b) = 7 m=l (mcos mxsin mx) + V^ (ncos nx+sin nx) n=l (2.83.0) + f(71 + 0)] = Sinhjrc 1 + 2 Y 2 * L n=i From Figure 2.837a) « sinhTT j + 2 V (l) n (cosnxnsinnx) (2. At x = 7t. we obtain ex s (2 836b) sinhrc y (l) n (cosnxnsinnx) (2. ^ n (cos nx .839) .n sin nx) + i (n cos nx + sin nx) 11 „=— d+n2) Taking the real part. 1+n2 L[en + e~n] = sinh_7t 1+2^ —1— L n=i 1 + n _ (2. That is to say oo « 1 [f(71 .838a) 1 m=oo °° n= l = 2^ l(~m)cos (~mx) + s i n (mx)j + 2^ (ncos nx + sin nx) (2.838b) and we made use of the fact that cos is an even function and sin is an odd function.837b) The imaginary part can be shown to be zero in the following way.838c) (2. oo — 1 oo ^ n=—oo (n cos nx + sin nx) = ^ n=— oo (n cos nx + sin nx) + jT (n cos nx + sin nx) n= l (2.838d) = 0 We replaced n by .m in the first sum on the right side of Equation (2. the function is discontinuous and the series converges in the mean.SERIES SOLUTIONS AND SPECIAL FUNCTIONS 755 sinhrc y .
b) at x = 0 and x = JL. The deflection w of a uniform beam of length Z with an elastic support under a given external load p is given by E I ^ + k w = p(x) dx 4 where E I is the flexural rigidity and k is the modulus of the elastic support. W mini FIGURE 2. Assume that the load is constant and is applied on the interval Jt /3 < x < 2Z13.840b) Example 2.841) d_w _ dx 2 0 (2.U6 ADVANCED MATHEMATICS cosh n = sinhJL l + 2 V — 1 — (2. Beam on elastic foundation under an external load .84 .842a. If the beam is supported at the ends.841) is given in von Karman and Biot(1940).84. The derivation of Equation (2. Obtain the deflection w (x). as shown in Figure 2. the boundary conditions are 2 w = (2.84.
Its Fourier series is given by oo p(x) = £ n=l tZ bn bjjSinfi^ ^ .846a) ^ * n=l We deduce that n 4_4 w EI E_ZL_ + k = b (2. p(x) = pQ.846b) [ or X4 J (2.b. Differentiating w four times and substituting the resulting expression in Equation (2.846c) w n = b n ^ /(EIn 4 7c 4 + k^ ) . nTi L 3 3 J nn We seek a solution of the form oo DJL s i n DJL 2 6 (2.c) = _ ?£Q.2ZI3 (2. 0 < x < X13 i/3<x<2i/3 2X /3 < x < X (2.841) yields El ] £ 5_5_WnSinlfflx + _n = l X * J k ^ n=l WnSinmcx = ^ bnSinmcx (2.845) * Note that w automatically satisfies the boundary conditions.c) where p0 is the constant external load.844d.843a. [cos ^2L _ Cos nffl1 = iPfi. 0.e) v w = ]T w n s i n n J ^ n=l (2.844a) = ~ I P(x)sin n ^ x dx = (2p o /i) I * JO * Ji/3 sin sin B^Lx d x ^ (2.844b. The function p (x) can be considered to be an odd function of period IX.SERIES SOLUTIONS AND SPECIAL FUNCTIONS 1JZ The function p (x) can be written as / 0.
Equation (2.845).848a).848a) where a n = nrc/L Let A a n = (n+l)7t/Ln7C/L = 7U/L (2.b) Substituting Equation (2.^ I f(^)ex P [ia n (x*^)]d^ (2. 47) yields oo f ( x *) (2. wn can be calculated from Equation (2. On substituting w n in Equation (2. we obtain f(x*> ~ X n =oo ^ r ) f(^)exp(ia n ^)d^exp(ia n x*) /—L eL (2.849b) in Equation (2. we obtain w.848b) (2.850b) On letting L —> °°.850b) becomes f°° f ( ^ ) e i o t ( x * . Fourier Integral The transformation from a function f(x) having a period 2n to a function f(x*) having a period 2L is obtained by writing x = x*7t/L Combining Equations (2.U8 ADVANCED MATHEMATICS The coefficients b n are given by Equation (2.833c.o o L" /co (2.850a) oo 18 S .^ d ^ d a .847) r L \ ~ Z ^ \ f(x*)exp(ia n x*)dx* exp(ia n x*) (2.85lb) — oo J —oo n Jo I /  I f(^)cosa(x*E)d^ da J (2.844e).851a) J J « .849a.851c) . f.I /» oo ^ oo I f (^)[cosa(x*^) + isina(x*^)]d^da (2.846c).
the integral converges to j . A (a) = 2 I f (£) cos a^ d£ = 2 I cos a^d£ = 2 sin a (2. If f (x ) has a jump discontinuity at a point x 0 . IxI> 1 0<x<l l<x<0 (2.SERIES SOLUTIONS AND SPECIAL FUNCTIONS * 00 159 ~ — I K I f (£) (cos ax* cos at.851f.b. f°° B(a)=l f©sina^d^ (2. + sin ax* sin a£) dt. I x I > 1 Figure 2. (a) f(x) is even and B(a) is zero. da (2. A (a) is zero. / (b) 1.1 .85 la to e) is the Fourier integral of f (x ).853a.852a. If f(^) is even. B(a) is zero and if f(^) is odd. (a) f(x) = x<l (2. The quantities A (a) and B(a) exist if J f(^)d^ converges.b) 10.c) Jo Jo a .c) f(x) = / . Example 2.85.854a. The right side of Equations (2.g) Note that sin a is an odd function so j J00 sin a d a is zero. ( 0.85Id) Jo LJ°° (2. Determine the Fourier integral of the following functions 11.b.[f (x 0 + ) + f (x o _)].85 illustrates these two functions.85le) f°° ~— I [A(a) cos ax* + B (a) sin ax*J da 71 Jo where A(a)= f(£)cosa£d£.
I sin_a d a _ n/2 (2. (b) Odd function B(a) = 21 sina£d£ = ^ .c o s a ) a (2.b) Jo .856a.85 (b) f(x) is odd and A (a) is zero. I sin a cos a x a d a = n/^ 0<x<l x = ±1 (2.160 ADVANCED MATHEMATICS f (X) = 1 I 2 sin a cos a x d a (2.( 1 .b.857) f (x) i (o) 1 I 1 I • X FIGURE 2. we obtain x>l r .858a.855) The function is continuous everywhere except at x = ±l. On setting x = 0. We deduce that In/2.c) Jo \0. (a) Even function.
lxl<a (2. We deduce that / 0.860ad) x=l x>1 0.b) x> a where p 0 and a are constants.SERIES SOLUTIONS AND SPECIAL FUNCTIONS 767 f(x)«2.j K (1 ~cos a a) sin ax da (2.863) We deduce that . Assume that p (x) is given by Po> p(x) = 0.I [A(a) cos ax + B(a) sin ax] da n)o (2.862a) k [C (a) cos ax + D (a) sin ax] d a (2. Extend the problem considered in Example 2.86. The appropriate boundary condition is that w vanishes at infinity.841) yields A t 0 0 t°° n Jo I [C(a)cos ax + D(a) sin ax] da = J. We represent p (x) and w (x) by their Fourier integrals and write r p(x) ~ i.84 to an infinite beam extending from °° to oo. (2. x=0 0 < x < 1 )0 (1cos a) sin ax « d / JC/4. so 7t/2.862b) w (x) « 1 nh Differentiating w (x) four times and substituting the resulting expression together with the expression for p in Equation (2.859) Jo The function is continuous everywhere except at x = 0 and at x = ±1.86 la.I [A (a) cos ax + B (a) sin ax] da n (2. Example 2.
38a).865d. 61a.866b) w(x) = ^ 71 I Jo sin aa cos ax a ( a 4 E I + k) da (2.91) yields . we obtain (2.91. b) in Equation (2. Example 2. that is to say.865a.91) The origin (x = 0) is an irregular singular point. This is illustrated in the next example.864a) (2.851f. we can obtain a formal series solution which is a good approximation for small values of x.f) Joo Combining Equations (2. 2.c) Joa B(a) = I p(x)sinaxdx = p 0 I sin ax dx = 0 Ja r (2. Find a power series that satisfies the equation x 3 y" + (x2 + x ) y ' .864b) A(a) = I r r p(x)cosaxdx = 2p 0 I cos ax dx = (2p0 sin cca)/a JO r (2.9 ASYMPTOTIC SOLUTIONS We have obtained convergent series solution in the neighborhood of x 0 (= 0) if x 0 is an ordinary or regular singular point. In some cases.e.867) Examples of LegendreFourier and BesselFourier series are given in Chapter 5.866a) (2. x (x +x)/x and x (l)/x do not have a Taylor series about x = 0.38a to c) in Equation (2. there is no method of generating a convergent series solution.864a to 65f) yields C(a) = 2 p 0 s i n a a / [ a ( a 4 E I + k)] D(a) = 0 Substituting Equations (2.862b) yields /•oo (2.1SZ ADVANCED MATHEMATICS (a 4 EI + k)C(oc) = A (a) (a 4 EI + k)D(a) = B(a) From Equations (2.b. g.y = 0 (2. b).866a. We seek a formal series solution as given by Equation (2. Substituting Equations (2. If x 0 is an irregular singular point.
93c. We can increase the value of n as long as the value of the n th term is less than the value of the (n .l ) s + 1 (s1)! x s (2.2 c 2 = 2c 2 xs: (sl)(s2)cs_1 + (sl)cs_1 + s c s . we find that y (0.SERIES SOLUTIONS AND SPECIAL FUNCTIONS oo /££ ]T [(r + 2)(r+l)c r + 2 x r + 3 + (r+l)c r + 1 x r + 2 + ( r + l ) c r + 1 x r + 1 .c r x r ] = 0 r=0 (2. The power series s n (x) represents f(x) asymptotically as x—>0 if x ~ n [ f ( x ) .( s . Since the asymptotic series is divergent.94) Can this divergent series [Equation (2. From Equation (2.f) CJ+2C2C2 = 0 => c 2 = . the first two terms are sufficient and y(0. (s1) xs + .93g.Cj [10"1 .]= cl £ s=l ( .. Using the divergent series.s n ] —> 0..c s = 0 => c s = .1) to be accurate to two decimal places.h) The formal series that satisfies the differential equation is oo c 1 [ x .93e.6 x 10" 4 + 24 x 10~4 + . . n The divergent series s n (x) [= X r=0 crxF] is an asymptotic series expansion of a function f(x).1) = 0.l ) t h term if the value of the n* term is greater than that of the (n .96) The error in approximating a function by a convergent series decreases as the number of terms increases.10" 2 + 2 x 10~3 .92) Comparing powers of x.1) c s _! (2.95) We note that the magnitude of the terms decreases as the order increases and if we require y (0. for asymptotic series.d) (2..x 2 + 2x 3 + ( .99 Cj.. it is seen that the approximation improves as the magnitude of x decreases for a fixed value of n.1) .93b) (2.96).l) t h term.l ) s + 1 234 . for all n > 0 as x—>0 (2. the error may increase as n increases. We are certain of attaining the required accuracy by taking a sufficient number of terms. we obtain x°: x1: x2: x3: c0 = 0 cjcj = 0 => cj is arbitrary (2. ] (2. Usually. only a few terms are required...94)] be used to calculate the values of y for small values of x ? The answer is yes.93a) (2.l) t h term.c x 2c 2 + 2c2 + 3c 3 C3 = 0 => c 3 = .1) can be written as y (0. We stop at the (n .
we obtain oo oo u' = ie ix X c n x. In applied mathematics. On differentiating.913a) .444)] for large values of x. Example 2.n e i x £ n c ^ . Even if a convergent series is available. A second order differential equation is in its normal form if the first derivative (dy/dx) is not present. We first transform Equation (2.99) For large values of x.444) yields x 2 (u" v + 2u' v1 + u v") + x (u1 v + u v') + x 2 u v = 0 or x2vu"+ u'(2x2v'+xv) + u(x2v"+xv'+x2v) = 0 (2.444) to its normal form. one often neither knows nor cares if the series converges or not [Van Dyke (1975)]. we look for a solution as x —> °°.98a) (2. To achieve this. in this case. we write y = u(x)v(x) Differentiating and substituting the resulting expressions in Equation (2.63a)] and the solution is e lx .164 ADVANCED MATHEMATICS Asymptotic series are widely used in the solution of differential equations and for evaluating integrals. This suggests that we seek a solution of the form oo u = e ix £ c n x" n n=0 (2.98b) becomes u" + u ( l + l/4x 2 ) = 0 (2. therefore we expand in reciprocal powers of x." " 1 n=0 n=l (2.911) (2.912) Note that.92. Equation (2.97) We now impose the condition 2x 2 v'+ xv = 0 The solution is v = x~ 1/2 Equation (2. it is sometimes profitable to consider an asymptotic series as shown in the next example.98b) (2.910) (2.911) is approximately the simple harmonic equation [Equation (2. Solve Bessel's equation of order zero [Equation (2.
2) (s .c s — 2i (s — 1) cs_j + (s .d) (2.913b) = e i x ]T [ic r x.i) We note that the coefficients c s are alternately real and imaginary.i c 0 / 4 » 2 =* c 2 = .r .915c.l ) c s _ ! = cs_2 (s 2 .2 +(r+ 1) (r + 2)c r+1 x.913a.2 4 0 ^ 0 2 + ^0(3 = 0 . to allow for the tabulation of Bessel functions. ^n.915a) (2.914) Comparing powers of x yields x°: x"1: x~2: x~3: .r (r+l)c r + 1 x.915g. we obtain oo £ r=0 [c r x~ r 2i(r+ l)c r + 1 x~ r .915b) (2.f) . As usual.c 3 ..4 i c 2 + 2 c 1 + c 3 + ic1 = 0 x~s: .3s + 9/4) => cs_j = .916a) .r .0 2 .3 / 2 ) 2 / 2 ( s . d) in Equation (2.i c s _ 2 ( s .911) and multiplying by e~ix.2 ] r=0 (2. we set c 0 to be equal to one and u can be written as ixL 1 i 4«2x 32 + 42«22«2x2 i»32«52 + 43«233!x3 325272 4424»4!x4 u = e 1 + ..913d) Substituting Equations (2.913c) The second derivative u " is given by oo u " = eix J r=0 [crxr2i(r+l)cr+1xr2 + (r+l)(r + 2)cr+1xr3] (2.c o + co = 0 Ci+cj = 0 => c 0 is arbitrary => Cj is arbitrary =» c 1 = .3 +c r x~ r + (1/4) c r x~ r ~ 2 ] = 0 (2.h.SERIFS SOLUTIONS AND SPECIAL FUNCTIONS oo oo 165 = ieiX X r=0 oo Cr x ~ r ~ elX X r=0 (r + ^ Cr+ 1 x ~ r ~ 2 (2.1 ) (2.£s (2.1) cs_2 + c s + 1/4 cs_2 = 0 => 2 i ( s .915e.i ( 9 0 ^ 4 ) / 4 (2.
we deduce that lim w = 1 . 20a..3 2 /[4 2 » 2 2 .. sin x) (2. 18b.916c. b). is given by the linear combination of yj and y 2 .918a.i w i ) = x [wf cos x + Wj sin x + i (wr sin xWj cos x)] (2. as x —> °o J 0 (x) . are . The solution of Equation (2.917c) The two real fundamental solutions y^ and y 2 are given by yj = ( y + y)/2 = x~ 1 / 2 (w r cosx + w ..444). we deduce that the leading terms of y'j and y 2 .916c. 10.921) From Equations (2. as x —> °°.916b) (2.97. wj = l / [ 4 » 2 x ] . as x —> °°.919) yields J^(x) « A y ' 1 + B y 2 (2. d). From Equations (2.A y 1 + B y 2 where A and B are arbitrary constants. Combining Equations (2.918c.166 ADVANCED MATHEMATICS = eix(wriWi) where w r = 1 .W j cos x) where y is the complex conjugate of y.920a.919) Differentiating both sides of Equation (2. d.3 2 « 5 2 / [ 4 3 « 2 3 » 3 ! x 3 ] + ..917b) (2.916c) (2. d.444) is the Bessel function of order zero.b) (2. 16b) yields y (2. The general solution of Equation (2.d) y 2 = ( y .y ) / 2 i = x ~ 1 / 2 ( w r s i n x .916d) = x~1/2eix(wriwi) = x~ 1 / 2 (cosx + i s i n x ) ( w r . lim W: = 0 (2. That is to say.917a) (2.2x 2 ] + 32» 52« 7 2 /[4 4 « 24« 4! x 4 ] + .b) (2.
(x) cos x] The integral representation of Jo (x) is (see Problem 22b) r J n (x) = Jcos (x cos 9) dG (2.— I [(lcos9)cos(x+xcos9)+(l+cos9)cos(xxcos9)]d9 x>~ 271 j o (2.922a.924b) B = lim x m [ J 0 (x) sin x + jj.925.I n [sin (x cos 9)] cos 9 d9 (2.924a) yields xi/2 r A = lim I [cos x cos (x cos 9) + sin x cos 9 sin (x cos 9)] d9 x>~ 7 C JQ (2.xT 1 / 2 sinx y'2 « x~ 1 / 2 cosx Combining Equations (2.925) It follows that Jjj(x) = .b) (2.923b) The solution is A = Urn x m [ J o (x) cos x .927a) xi/2 r = lim I [cos (x + x cos 9) + cos (x .927b) xi/2 r = lim .cos (x + x cos 9)}] d9 (2.923a) (2.J.A s i n x + Bcosx X >°° (2.927c) .922c) (2.924a) (2. 26) in Equation (2.926) Jo Substituting Equations (2.x cos 9) .919 to 22c) yields Urn x 1 / 2 J n (x) = Acosx + Bsinx lim x 1 / 2 j ' ( x ) = .SERIES SOLUTIONS AND SPECIAL FUNCTIONS 167 y\ « x" 1 / 2 (w r sinx) .J' (x) sin x] X >~ (2.x cos 9) x _ > ~ In ) Q + cos 9 {cos (x .
19.168 ADVANCED MATHEMATICS = lim . we obtain d<> = .! • (V2x~) sin 8/2 d9 It follows from Equations (2.nr~ 1/2 x1/2m1/2f = lim—Mx >~ K x / 621 11 2 (cos<))d() (2.928.— sin 2 9/2 cos(2xcos 2 8/2)d0 I 2x / 2 1/2 1 .930a) Jo JV27 r^/ = yT I Jo Therefore x i/2 r 2 2 lim . It is found that A = B = jL Combining Equations (2. 32a.— ( cos f) d(t) \ (2. we write (j) = {4lx) cos 6/2 (2.927d) To evaluate the first integral on the right side of Equation (2.I cos^d^) K Jo = TV Similarly. b.93 la) Jo I 2x/ (2.918a. B and the associated integrals can be evaluated.930b) V 2x/ x>~ n jQ .\T^) h _ 1 _ d<) (2.b) .932a.929) f f / V 2\ I sin2 9/2 cos (2xcos 2 0/2) d9 = (cos <^) f.— x l/2 (K x>~ n j Q [(sin 2 0/2)cos(2xcos 2 0/2)+(cos 2 0/2)cos(2xsm 8/2)]d9 7 0 2 (2.93 lb) (2.931c) = i^. b) yields (2.928) On differentiating. 29) that n 1/2 (2.927d).
From the two examples we have considered. In Example 2. r = l/2 (2. that the series representations of w r and wj [Equations (2.936) as x —> °°.n/4)] (2.453)] and.b) Further details on asymptotic expansions can be found in Cesari (1963).92. we need to consider only the first seven terms of the series (four for w r and three for Wj) and we obtain J o (6) = 0. . d)] are divergent. where X(x) = ix. using the ratio test. However. we deduce.cos x)] [wr cos (x .15064 (2.916c.7t/4) + wf sin (x . Since Poincare's pioneering work.915i). X = r = 0 and it is not common for an asymptotic solution to be simply a power series.SERIES SOLUTIONS AND SPECIAL FUNCTIONS 1/2 762 Jo ~ V JL [wr (cos x + sin x) + Wj (sin x .935) c n x" n (2.937a. and Wasow (1965). • An asymptotic series is often more useful than a convergent series.933a) (2. we can postulate that the asymptotic solution of a differential equation can be of the form y . Nayfeh (1973).' From Equation (2. to determine J 0 (6) accurate to five places of decimals.933b) * MWv A. we obtain J o (6) = 0. considerable progress has been made in the understanding of the asymptotic series.91.{exp [X(x)]} x r £ n=0 (2.934) The function Jo (x) can also be represented by a convergent series [Equation (2.15067 and this value is accurate to four decimal places only. Usually. after adding the first twenty one terms. Van Dyke (1964). exponential functions are involved as in Example 2.
939a. Example 2. = . we have the Reynolds number which can be small (Stokes flow) or large (boundarylayer flow).. t m = tQ + etj +. 39a. A particle of unit mass is thrown vertically upwards (ydirection) with an initial velocity v 0 .941a) (2... The equation of motion (Newton's second law of motion) is y = [g + ey2] (2. If the air resistance at speed v is assumed to be ev 2 .. as y (t m ) = y 0 (t0 + etj + . We take the origin at the surface of the earth.. = v 0 At time t m .b) We have identified e to be a small parameter and we start by expanding all functions of interest in powers of e..940a) in Equations (2.940b) ..) + .93. That is y (t) = y o (t) + eyi(t) + ..938.938) where g is the gravitational acceleration and the dot denotes differentiation with respect to time.941c) (2. In fluid dynamics. we consider its reciprocal.. the particle is at rest and this is expressed.. (2.940a) (2.) + ey l (t0 + ztl + . we obtain respectively y ' o + e y j + .170 ADVANCED MATHEMATICS Parameter Expansion Many physical problems involve a parameter which can be small or large. The next example illustrates the method of parameter expansion. We seek a solution in powers of £ and often the series solution is an asymptotic series... Substituting Equation (2.)] yo(O) + e yi (O) + . determine the maximum height y m reached by the particle at time t m .[g + e ( y 2 + 2ey 0 y j +.. where e is a small positive constant. The initial conditions are y(0) = 0.. y(0) = v0 (2...942a) (2.. using Taylor's expansion.941b) (2. We consider the case when the parameter is small and we denote it by 8. If the parameter is large. b). = 0 y o (O) + ey !«)) + .
.949) .942c. f) can be written as y l =l\ g 2 t 3 .^ .e [ i ( v o 3 / g 2 ) ] + .940b) yields t m = v o / g . we deduce y o = ..948a. 47) yields ti = .947) (2. = y o ( t o) + e ttiy 0 (to) + (2.b) (2.g t + vo Yo= ~ ^ g t 2 + v 0 t Substituting Equation (2.946a) (2.e.b.943d.946b) (2.^ v o g t 3 + l v 2 t 2 ] From Equations (2.^ ( v 0 3 / g 2 ) Substituting Equations (2.943a to c). we obtain t0 = v o /g Combining Equations (2..f) From Equations (2.g 2 t 4 .944b) Yl = ..[ .947.944a) (2.c) (2. + ey x (t0) + e 2 t i y x (t0) + .v 0 gt 2 + v 2 t ] (2.d) yi( t o)] + . (2.. yo(°) = v o y x(0) = Q (2.[ g 2 t 2 . 46a. y'i=yo2..942c.= ° Comparing powers of e.944a) in Equation (2.943d) yields y'l = . The effect of the air resistance is to reduce t m .2 v o g t + vo2] The solution subject to Equations (2.SERIES SOLUTIONS AND SPECIAL FUNCTIONS JJ1 = y 0 (t0) + e t ^ (t0) + . 44a).y i ^ / y ' o C o ) = . yO(°) = °' Yi(0) = 0.942b) (2. 48b) in Equation (2.943e. 43a.945) (2.943a. we obtain from Equations (2.94la to c) e°: e1: yo=g.942d.
) 4 .952) ^fc Lg + ev _ The maximum height ym is given by ym = ^ n ( l + e v o 2 / g ) = ^ [ev o 2 /ge 2 v4/2g 2 + .95 la) (2.953b) (2. the nonlinear system reduces to a linear system.. The nonlinear terms become the nonhomogeneous terms [Equation (2. it is often impossible to obtain an exact solution.. 49) and can be written as Ym = .. Equation (2. b) is y = J_ i n ili^O. .938) becomes v ^ = [g + ev 2 ] dy or vdv (2. and unlike the present example..... Replacing y by (v dv/dy).) 3 + \ v 2 (v o /g + ..953c). 2? 2 (2.. as in Example 2..944b. the air resistance lowers the maximum height.e v 4 / 4 g 2 + ..e [ ^ g2 (v o /g + .939a. The equations can then be solved and an approximate solution is obtained..950b) is exactly Equation (2.] .953c) It is not uncommon for the small parameter to be associated with nonlinear terms.950a) (2. 46b.] 2 + v 0 [v 0 /g(e/3)(v 0 3 /g 2 ) + .. The present problem can be solved exactly..943d)].95 lb) subject to Equations (2.953a) (2. Equation (2. By the perturbation method.^ g [ v 0 / g .172 ADVANCED MATHEMATICS The maximum height ym is obtained from Equations (2.950b) (2.)2] = (v 0 /2g)e(v 0 4 /4g 2 ) As expected.L v o g (vo/g + .( 8 / 3 ) ( v 0 3 / g 2 ) + .93.95 lb) = dy g + ev 2 The solution of Equation (2. • (2.] = v 2 / 2 g .
B.4 and Example 2. If x < (1/e). If we are required to evaluate exp (ex) for large values of x. the rate equations for components A.. Thus the approximation of exp ( . i) The function exp (ex) can be approximated as oo exp (ex) = ^ n=0 (l) n (ex) n /n! = 1ex + e 2 x 2 + . (2. (2. ( e « l ) by the first three terms. it is not possible to obtain the analytic solution and the quasisteady state approximation is introduced.1414 to 16) in dimensionless form. we deduce from Equation (1. is valid for x < ( l / e ) but not for x » ( l / e ) . as an asymptotic expansion. the solution in powers of e is often singular (see Section 2.b) Several methods have been developed to extend the validity of the solution. we need to include more terms as the value of x increases.b) The infinite series is convergent and.142. but if x » (1/e). the approximation is no longer valid and the magnitude of the first three terms can exceed one while exp (ex) < 1 for all x > 0.954a)] is not useful (see Example 2. Follow Bowen et al. In Example 1. Discuss the validity of this approximation.) Except for the first term.1415) is set to zero. We next introduce several examples involving a singular perturbation.SERIES SOLUTIONS AND SPECIAL FUNCTIONS 775 The regular perturbation (straight forward expansion) method as described in the preceding example may generate a solution which is not valid throughout the region of interest (usually at infinity or at the origin). Example 2.954a.94. (1963) and write Equations (1. ii) The function Vx + e can be approximated as VxT~e = Vx" (1 + e / x ) m = VxT (1 + e/2x . Show that the solution by the regular perturbation is singular. we consider the first three terms as shown in Equation (2. In the case of a convergent series. the approximation is valid. This is discussed in the next example. This implies that the left side in Equation (1.94). In this case. iii) If the coefficient of the highest derivative in a differential equation is e.956) . (e = 0)..92). There is a limit to the number of terms that can be included.955a.1415) that cB = ( k j / k 2 ) c A (2. In more complicated cases. and C involved in first order reactions are solved. Setting dB/dt = 0..e x ) .e 2 /8x 2 + . the convergent series [Equation (2. The basic idea is to use more than one scale.954b). we have a singular perturbation.. all the other terms are singular at the origin.
we find that the approximate solution [Equation (2.1415 to 16) becomes ^A = c* * (2. Even if k 2 » k j .962c) .958a.960b) dt* e^ f = c* dt* The initial conditions are <£(0) = 1 .b.b.c) On solving Equations (2. exp(kjk2)t « 0 (2.956)] is valid if kj/kj « 0. we obtain _ MAtlexpCl^k^t] " k^lk^) Cfi (2'9 } Comparing Equations (2. b.959d. We introduce the following dimensionless quantities CA = cA/cA0' CB = cB/cA0 ' CC = cC/cA0 (2. Equations (1. t* = kjt Using Equations (2.e)" 1 [e1* .956. the approximation is not valid for small values of t and this restriction will be discussed later.174 ADVANCED MATHEMATICS From the exact solutions [Equations (1. c).ee" 1 * 78 ] .960a to c) subject to Equations (2.c) (2. 57).e)" 1 [e"1* .1417.96 la. 22)].960a) i\ dt e^B_ = _c^ + ec^ (2.1 (2. Cg(0) = cJ(O) = 0 (2.96la.e) e = kj/kj.959a.b) The approximation is valid if k2 » ^ and t—> «>. we obtain c*A = e~l* c* = e (1 .959a to e).962b) (2.e"1*76] c* = (1 .962a) (2.960c) (2.
we obtain eo.960a.^ = cA2 dt %^ 2 +% dt (2. b) in Equations (2. + 8 2 c A 2 + " = CBO + ecB. (2.e.964a to 65d). we obtain C.f) CBO = CAO' CB2 .b.= .963a. b) and comparing powers of e.c) = ° (2.964e) e2: dc A * .+cI0 dt o (29"64d> (2.964b) e1: —^.SERIES SOLUTIONS AND SPECIAL FUNCTIONS /Z2 For simplicity.= 0 (2.963b) CB + e2cB2 +  Substituting Equations (2.965a. b) and solve them by the regular perturbation method. % dt = c^ (2.f) = C B V ° ) = C B 2 (°) = • = ° Solving Equations (2.965d.963a) (2.^ = cB.964c) .960a.e.o = e"1*. We express c A and Cg as cI =cI0 + e c i .964a) 0 = c*Q (2.964f) The initial conditions are C I O (°) CBO(°) =L C A.(°) = CA2(°> = .966a.b.% dt* dCr> * * (2.966d. we consider only Equations (2. = °' c^ = c^ = 0 C B.c) (2. (2.
The coefficient of dc B /dt is e. The solutions given by Equations (2. Expanding c B in powers of 8. Note that in Equations (2.967b. We use two different time scales. the limit of e lim e>0 t*>0 * is one. We further note that the zeroth approximation CIO = e~l*> CBO =° (2. one for the inner solution and one for . We obtain a solution near t* = 0. This zeroth order solution cannot be improved because c B is not uniformly valid. t * 0. we note that Equation (2.968) Comparing Equations (2. On reversing the taking the limit as e —> 0 first and then as t* —> 0. c B and c B are known quantities and are defined in Equations (2.) [e" 1 . the limit of e~l t*/F limiting process.962b) can be written as c B .964d. d) respectively. This implies that lim e" l * /£ (2.b) Equation (2. This implies that the quasisteady state method is valid for the zeroth approximation and not for higher approximations. f).l * / e * t*>0 e>0 We further note that e~l changes rapidly near t* = 0. The Stokes solution of the flow past a sphere is uniformly valid but the higher approximation is not valid and is known as the Whitehead paradox.. A thorough discussion on the Whitehead paradox is given in Van Dyke (1975). At time t ~ 0. f)] and no arbitrary constant can be introduced so as to satisfy the initial conditions. 68). On is zero. It is the presence of the term e t*/f in Equation (2. d. 67b) are not uniformly valid and do not satisfy the initial condition [cB(0) = 0].e(1 + e + . This suggests that to obtain a uniformly valid solution we need to seek the solution in two separate regions.967b) is a valid approximation in the case 8 —> 0. which is the inner region or boundary layer where 8 is important.176 ADVANCED MATHEMATICS The approximate solutions for c A and c B are c*A « e~l .967a. c B « ee~ l (2.956.970) e. which is the outer region.960b)] reduces to a system of algebraic equations [Equations (2. and a solution for t* » 0.969a..962b) that contributes to the singularity.e" 1 /£] (2. A similar situation exists in fluid mechanics.b) is uniformly valid.964b. the differential equation [Equation (2.964b. the approximation is not valid.
E°: T£..973. by our choice of T..<* dc®* (2974a) The initial conditions can be applied in the inner region and they are 4i}o*(O) = c®*(0) = 0 The solution of Equation (2..974c. the coefficient of dc B /dT is one and not e. The time scale for the inner region is T = t*/e Using Equation (2.967a)] is uniformly valid and we need to solve only Equation (2.972) Note that. Equation (2. we obtain .971).974a) that satisfies Equation (2.972) and comparing powers of e.974c) is <$* = 0 Combining Equations (2. c A [Equation (2. This perturbation method is the matched asymptotic expansion method (boundary layer method) and was introduced by Prandtl to resolve d'Alembert's paradox in fluid mechanics (see Chapter 3).973) in Equation (2.SERIES SOLUTIONS AND SPECIAL FUNCTIONS 177 the outer solution. 74b) yields (2. 71.960b) by the method of matched asymptotic expansion. The independent variable for the inner region is chosen such that the coefficient of the highest derivative is not e. We denote the inner solution by ci and it is expanded as of = c < » *+ <+.966a. Substituting Equation (2. In the present example.d) .2.975) (2.960b) becomes ^ = CB + (2.971) EC A (2.
t * (o) * (2. .980a. the arbitrary constant is determined by using a matching principle.979) is identically satisfied. 78.c) On solving Equation (2.e ~ T The inner solution c^J can be written as (2. The solution which is uniformly valid is the composite solution and is denoted by CA .b. we obtain c£j* = l . The simplest one was proposed by Prandtl and can be expressed as Km 4 ° * = Jim c£o)* T—>oo B t*_>0 (2. Equation (2. The outer solution Cg' is given by Equation (2.976c) subject to Equation (2.b..967b) is valid for t » 0. As mentioned earlier. no condition can be imposed on the outer solution.. +"i o r / o \*~] i ci j and [ci J denote the outer limit (T —> °°) of the inner solution and the inner limit (t* —> 0) of the outer solution respectively.e T T—>~ l) = lim ee~ l t*—>0 t* = E (2.967b. For the outer region. This solution is given by (2. Equation (2. If the outer solution is obtained by solving a differential equation.t * .983a) .976a.178 ADVANCED MATHEMATICS dc®* —^+41J dT ci = e " e l = 1ET + . the outer solution is determined by an algebraic equation and there is no arbitrary constant..977) c£°* = E(le" T ) + .974d). Note that since all conditions are at time t = 0.982) Combining Equations (2. 81a) yields c£c)* = £ ( l .981a.b) [ /•«. There exists a region where both the outer and the inner solutions are valid and are equal.c) The matching principle can be interpreted as follows.1 (2.979) B In the present example. 80c.978) is the time scale and the solution given by Equation (2.T ) + £ e .£ + O ( £ 2 ) (2.e . That is to say lim £ ( l .979) implies (2.967b)..
( x + 2)y = 0 Compare the series solution with the exact solution. Determine the singular points of the following equations... Answer: (1 + x ) e x 4b.+ sin x — + y cos x = 0 dx dx 9d y dy ...x 3 ) ^ +x 2 ^ dx 2 dx + x4y = 0 . x3y = 0 Determine the radius of convergence of the following series. PROBLEMS la. (in) dy .e " 1 Note that Cg /e] (2. Are the singular points regular or irregular? (i ) ( l . (n) .983b) satisfies condition c B = 0. „ x —— + cos x — + y sin x = 0 dx 2 dx x ^ +exd^ dx 2 dx + (iv) 2a. dy _ x —^. one can consult Nayfeh (1973) or Van Dyke (1975). Compute a series solution of the equation (1 + x ) y ' .e[e" 1 . For further details on singular perturbation. Do they converge on the circle of convergence? oo oo n=0 2 n=0 (2n+ 1)Z oo £ ^ v av) x < = ^ v ' 4i n=l n ' *> n=0 n 3a.. Hermite's equation can be written as ..SERIES SOLUTIONS AND SPECIAL FUNCTIONS 779 ..
If n is a nonnegative integer. note that the coefficient of x n in H n is 2 n and deduce the n th /•oo 2 derivative of H n . the constant a 0 (or aj) is chosen such that the coefficient of the highest power of x (x n ) is 2 n . Calculate the first four polynomials.xy = 0 dx 2 d2v —— + xy + y = 0 dx2 (i) 6a.l ) n e x 2 ^^ dx Use Rodrigues's formula to deduce that (°° j00 2 e.180 ADVANCED MATHEMATICS ^ .x [Hn(x)]2dx = (2 n n!V^) [Hint: Integrate by parts. (ii) Find the general solutions of the following equations 2 (i) x 2 ——+5x — . 2n 2 22n(n2) 4 [ 2(nl) 3 1 y = ao[1"2TX + 4! X J + a H x  3! X + J Determine the radius of convergence of the series. To specify H n completely. and H 3 . e~ x dx = in/2] Find a series solution of the following equations d2v —— .2 x ^ dx 2 dx + 2ny = 0 Show that its general solution can be written as the sum of two infinite series and are of the form . H 0 .5 y = 0 dx 2 dx 2 Answer: Ax~ 5 + Bx (ii) x2—^3x^+4y = 0 dx 2 dx Answer: Ax 2 + Bx2Jinx . one of the series terminates and the polynomial is denoted by H n (x). H2. H i . Verify that the polynomials Ho to H3 are given by Rodrigues's formula 2 Hn(x) = ( . that is. I 5 a.
e (1 ~ x) Answer: x + x 10a. Show that r = 0 is a double root of the indicial equation of 2 x^+(lx 2 )^+4xy = 0 dx2 dx Find the solution that is finite at the origin.y = x subject to y(l) = y'(l) = 0 Answer: ( 1 ~XI + inx 2 4x 8 a. Answer: 1 .1 ) .I) 2 ^Z.x 2 + x 4 /8 . . Determine the solution that satisfies the conditions y ( l ) = l. Solve the following equations ( i ) 4 X 1_I+3^Z + 3y = o dx2 d x x2^3x^+4(x+l)y = 0 dx2 d x ( i i ) ( i i i ) x^+2^+xy =0 dx2 d x 9b. y ' ( l ) = .l ) + B (x1)" 1 Solve the equation x 2 y" + x y ' . ( x .SERIES SOLUTIONS AND SPECIAL FUNCTIONS 787 (iii) 7b.1) — . Deduce a series solution for the equation X2d!y +(X2 + X ) dy_y = Q dx 2 dx Express the series in a closed form.l .y = 0 dx2 dx Answer: A ( x .+ (x . (x .
k is the thermal conductivity of the fin. Sheppard and Eisenklam (1983) have considered the dispersion of an ideal gas through porous masses of solid particles. Deduce that the indicial equation of 2d y . If T denotes the fin temperature and T A denotes the air temperature. 2^ dy „ x z — '. where C is the Laplace transform of the concentration. obtain C in a power series of x up to x5.+ (x . show from an energy balance that y (= T . 13b. at x = b .x z ) — . as shown in Figure 2. Npo is the dispersion number. y remains finite y = TBTA .0 (b . and p is the Laplace transform variable. Obtain one series solution which is nonsingular at the origin and verify that it is e x . Use the method of variation of parameters to show that the second solution is I (e~ x /x) dx. find the second series solution.x) ^ + (b . p = h / k s i n a . By expanding e~x in powers of x. and a is the half angle of the vertex of the triangular fin. The gas is assumed to be compressible and after some simplifications (details are given in the paper). The outer radius of the fin is b. The appropriate boundary conditions are at x = 0. x0 is a residence time. the equation to be solved is N D O (l + Bx)^(l+Bx+N D O ) d ^ + [S(l+Bx)3/2(T0)(p)]c = 0 \X A.T A ) satisfies the equation x (b . On expanding all quantities in powers of x up to x5.182 ADVANCED MATHEMATICS 1 lb.x) y = 0 dx2 dx where x is the distance from the rim of the fin. B is related to the permeability of the porous mass. Show that the origin is an ordinary point. h is the heat transfer coefficient.xy = 0 dx dx 2 is r 2 = 0. 12a. Jenson and Jeffreys (1963) have considered the problem of the temperature distribution in a transverse fin of triangular crosssection.a .2x) ^ .P13b.
Obtain a series solution for y if a = 5 cm. and T B = 373 K. The potential u at any point P is given by u = e/rj . where a positive charge e is placed at point A.e / r 2 ^ y(7c) + p 2 y1 (TC) = 0 .P15b. a distance a from the origin. of the SturmLiouville problem dx 2 a i y ( O ) + oc2y'(O) = 0 .P13b 14a. as shown in Figure 2.SERIES SOLUTIONS AND SPECIAL FUNCTIONS J83. A negative charge . Show that the origin (x = 0) is a regular singular point. a 2 P 2 ) t a n 7 t *^ ~ " ^ ( P i a 2 ~ a i P2) 15b. —/TCL __!___ j° FIGURE 2. (3 = 380 W/mK. a distance .a from the origin. are given by (al Pj + A . Transverse cooling fin Show that the eigenvalues A.e is placed at B. Tg is the temperature of the pipe and a is the radius of the pipe. Consider the following electrostatic problem. T A = 288 K. b = 15 cm.
Charges (±e) at A and B Use Rodrigues's formula [Equation (2. ] . P j .713)] to compute P o . P y r/ \ n B a 0 a A FIGURE 2. we obtain r~ ! = r . show that in terms of the Legendre polynomials. Show that I P 2 dx = 2/5 [Hint: Use Rodrigues's formula.P15b 16a.714) to express r^" and r 2 Hence. integrate by parts. 17b. " = ¥i(f) 2s+lp 2 S+1 <cose) s=0 Find the limiting value of u in the case a —> 0. r2 = BP.1 [ l . and P 2 .184 ADVANCED MATHEMATICS where i^ = AP. and note that P 2 = 3. This limiting case corresponds to a dipole. Using the cosine rule.( 2 a / r ) c o s 0 + a 2 / r 2 ] ~ 1 / 2 r" 1 = r" 1 [1 + (2a/r) cos 9 + a 2 / r 2 ] " 1 / 2 Use Equation (2. e a —> \i (* 0).
( . Apply l'Hopital's rule to Equation (2. deduce that (i) (ii) (iii) 21b. They assumed the existence of a steady temperature and. Verify that they satisfy the associated Legendre equation. 20a. is a where R is the dimensionless radial distance.1 ) " ^ n K v>n [ 3V dv where n is an integer. R] is the dependent variable.740).x 2 ) 19b. ^  dR R=1 BiR l ( l) where Bi is the surface Biot number. J3/2W = . The boundary conditions [their equation (14)] are *L dR R=o =0.72 and Table 2. show that Y n (x) has a i n (x/2) term.SERIES SOLUTIONS AND SPECIAL FUNCTIONS 1 2 18$ 18a.71. to obtain the steady temperature.768) and deduce that Yn(x) = l lim [ ^ . n ) . they had to solve the equation [their equation (12)] . Assuming that J v (x) and J_ v (x) are given by Equations (2. 63).762b. and A constant. From Example 2.J j ( x ) V ^£>7L X Choudhury and Jaluria (1994) obtained an analytical solution for the transient temperature distribution in a moving rod. Answer: 3 x V l .x 2 . Calculate the associated Legendre functions P 2 and P  from Equation (2.f ^ ( e f J L + sinx) J5/2(x) = AT2T f3mx _ 3cosi _ s i n x \ ' 7tX  Xz X j Jo' (X) = VT~— (COS Xsin X) = . Obtain Rj in terms of Bessel functions and show that the eigenvalues A^ are given by M i ( X B ) = BiJ 0 (*. 3 ( l .
S = I r=0 s=0 n=<*> t"Jn(x) Show that on substituting t = e*e and equating the real and imaginary parts.186 ADVANCED MATHEMATICS 22b. The generating function for Bessel's functions of integral order is exp [(x/2)(t. Expanding the generating function yields exp[(x/2)(tl/t)] = £ t±f(f) r+ V.444). 23a.2x + 1 defined in the interval . Answer: P o .813a to d) to deduce that j (x) = J% Jo cos (n0 .I cos(xsin9)d9 71 Jo satisfies Equation (2.2 P 3 The function f(x) is defined by 1 f(x) = i0 24b. we obtain oo cos (x sin 0) = Jo (x) + 2 ^ n=l J 2 n (x) cos 2n9 sin (x sin 9) = 2 ]jT J 2 n + 1 (x) sin [(2n + 1)9] n=0 Multiply the first equation by cosm9. Verify that J (x) = J.1/t)]. and use Equations (2.2 Pj + 1. where m is an integer.0. Represent the function f (x) = 3x . the second equation by sin m9.x sin9) d9 This is the integral formula for Bessel's function. 0<x<l/2 x=l/2 1/2<X<1 .1 < x < 1 by a linear combination of Legendre polynomials.
833c)]. It is defined by 1 TC<X<0 f (x) = 1 0<X<7t Sketch f(x) over the interval 37tto37i. The function f(x) is periodic and of period 2n. Represent the function 0 f(x) = 1 0 < x <% by a complex Fourier series [Equation (2. Show that the coefficients cn are given by cn = J i ( V 2 ) / [ ^ n J i ( M 2 25a.i f ) ~ n ± s=0 sin (2s + 1) x (2s+1) Deduce that oo c V ^ s=0 (D (2s+1) = £ 4 26a. where the ^ n are the zeros of JQ(X). . . Show that its Fourier series is f^ H x .SERIES SOLUTIONS AND SPECIAL FUNCTIONS oo [£Z and is represented by ^ n=l c n JQ (kn x ) . ~ e i(2s+l)x n < x < 0 Answer: A—— > 2 % *i S=oo (2s + 1) 27b. The equation governing the motion of a damped harmonic oscillator under the influence of an external periodic force f(t) is dVkdy+n2y dt 2 dt = f(t) where k and n are constants.
A (a) = 1 / ( 1 + a 2 ) Jo 1 + a 2 2 (ii) I Jo cos a d a 1+a2 = S2e Answer: B(a) = a / ( l + a 2 ) 29a. Differentiate the series term by term and substitute the resulting expression in the differential equation. . Calculate the radius of \r=0 j convergence of the series and comment on the validity of the series as a solution of the equation. the cos terms. Show that the equation X 2di dx 2 + ( 3 x _1)dy+y dx = o has an irregular singular point at the origin. Answer: a0 = c/n 2 as = ks7tc 2 T 3 /[a 2 + k 2 s 2 7C 2 T 2 ] bs = c T 2 a s / [ a 2 + k 2 s V T 2 ] a* = sVn 2 T 2 28 a.188. and the sin terms. Assume that y (t) is also periodic of period 2T and write its Fourier series expansion. if f(t) is periodic of period 2T and is equal to ct/2T in the interval (0. 2T). Determine the Fourier integral representation of e" x f(x) = 0 Deduce that x<0 x>0 r (i) I doc = 2E. ADVANCED MATHFMATICS Find the Fourier series of f(t). I~ \ Find a power series I ^ c r x I t n a t satisfies the equation. Answer. Determine the Fourier coefficients by comparing the coefficients of the constant term.
Substituting the expansion of sin y in the equation of motion yields 2 ^ + C 0 2 ( y . the equation governing the motion of a pendulum can be written as 2 — .x .. assuming e to be small. pd x =0 x=0 by the method of perturbation.e . [Hint: Expand all quantities in powers of £.y 3 / 6 + .Eyz = 0 dx 2 dx subject to initial conditions y (0) = 1 . Find a two term expansion valid for small £ for the solution of ^ dx 2 Ey2 = 0 y(l+e) = 1 Answer: x . Solve the equation 2 d y dy 2 n — : .) = 0 dt 2 Assuming co2/6 to be small.+ co2yey3 = 0 dt 2 Suppose that the initial conditions are .+ C O sin y = 0 dt 2 where y is the angle of inclination and co is the frequency.( l l + x 3 ) + y(0) = 0 ..+ — . Calculate the first two terms of the expansion.SERIES SOLUTIONS AND SPECIAL FUNCTIONS 189 30a.l + x + e~ x ) 31b. The equation of a simple pendulum is —.] 32b. Answer: l + e ( .
Determine the uniformly valid solution to order e. We introduce two time scales To and T] and write t = T o + 8 Tj Substitute t in the solution and determine Tj such that the secular term is eliminated.190 ADVANCED MATHEMATICS y (0) = 1 . To obtain a uniformly valid solution. Answer: Tl = 3T o /8co 2 . ^ d t =0 t=0 Show that the solution to order e is y = cos cot+ e [(cos cotcos 3 cot)/32 co + (3tsin cot)/8co] The presence of the term 3t sin cot (secular term) renders the solution invalid for large values of t. we can use the method of multiple scales.
The theory of differential equations has been extended within the domain of complex variables. However. therefore. Applied mathematicians. Wessel.13) (3.1 INTRODUCTION The inadequacy of the real number system (rational and irrational numbers) in solving algebraic equations was known to mathematicians in the past. and asymptotic expansions. The concept of a function was subsequently extended to complex functions of the type w = f(z) where z (= x + i y) is the independent variable.CHAPTER 3 COMPLEX VARIABLES 3. The concept of complex variables is a powerful and a widely used tool in mathematical analysis. were solved by accepting VT as a possible number. Euler was the first to introduce the symbol i for i^A with the basic property i2 = . when d is zero. It was not until around 1800 that sound footing was given to the complex number system by Gauss. called an "imaginary" (as opposed to real) number. Complex integral calculus has found a wide variety of applications in evaluating integrals. which could not be solved in the domain of real numbers. no actual meaning could be assigned to the expression V^T. Real roots are special cases.) He also established the relationships between complex numbers and trigonometric functions. Gauss proved that every algebraic equation with real coefficients has complex roots of the form c + i d. so as to obtain meaningful solutions to simple equations such as x2+ 1 = 0 (3. it appears that equations. This notation. has had its own shortcomings. This usage still prevails in the literature.11) For quite sometime. inverting power series. however.12) .1 (Electrical engineers use j to denote i^T. It therefore became necessary to extend the real number system. It was. Argand proposed a graphical representation of complex numbers. forming infinite products. and Argand. (3. in those times.
As in vector algebra. ^ X 2 = xLy FIGURE 3.21. a complex number can be represented by a point in a plane. This representation is the Argand diagram and is shown in Figure 3. it implies . Just as a real number can be represented by a point on a line.192 ADVANCED MATHEMATICS physicists. We can also regard z as an ordered pair of real numbers.2 BASIC PROPERTIES OF COMPLEX NUMBERS We can write a complex number z as z = x + iy where x and y are real numbers. If z^ (=Xj + i y i ) and Z2 (=X2 + iy2) are equal. It is indispensable for students in mathematical. and engineering sciences to have some knowledge of the theory of complex analysis. and engineers make extensive use of complex variables. (3.21 Argand diagram Two complex numbers are equal if and only if (iff) their real and imaginary parts are equal. we write z as (x. physical. The numbers x and y are the real and imaginary parts of z respectively and are denoted by Re (z) and Im(z). y).> z=x+Ly A N.21) y . 3.
and denote it by Arg z. The addition and multiplication can be handled in the same way as for real numbers and. we obtain r = Vx 2 + y 2 tan9 = y / x (3.yj y2. we find x = r cos 0. The polar representation is useful for computational purposes.d) (3 2 3 e f } x 2 + iy 2 (xi+iyi)(x2iy2) (x2 + i y 2 ) ( x 2 . y } . associative. 9).27) (3.b) Inverting Equations (3.b) Thus a complex equation is equivalent to two real equations.i y 2 ) = [x 1 x 2 + y 1 y 2 + i ( x 2 y 1 . it is replaced by .22a.24) In the Argand diagram.b) Since any multiple of 2K radians may be added to 0 without changing the value of z. y = r sin 6 (3.28a. as shown next. yi + y2) z2 .1 . whenever i 2 appears.25a.23c. It can be regarded as the length of the vector represented by z.23b) (3.26a) (3.21.y 2 ) •LX • z 2 = (xT x 2 .21. it is the reflection of z about the xaxis and it is shown in Figure 3. yi = y 2 (3. The absolute value of z is also given by z = VzT 0 is the argument or amplitude of z. The commutative.z 2 = (x2 . b).26b) The number r is called the modulus or the absolute value of z and is denoted by I z I.25a.23a) (3. From Figure 3. So far we used only the rectangular Cartesian system. We list some of the results zj + z 2 = (x t + x2.x 2 . and distributive laws hold. xj y 2 + x 2 y{) = z 2 • zx zl Z2 = xl + iyi = (3.COMPLEX VARIABLES 193. It is written as arg z = 0. Hence we can write z = x + iy = r (cos 0 + i sin 0) (3. we specify K < 0 < n as the principal value of arg z. . We can also use the polar coordinate system (r. M = x2.x 1 y 2 ) ] X2 + y 2 The complex conjugate z of the complex number z (= x + i y) is defined as z = xiy (3.
+ 9 n ) + i sin (0! + 0 2 + . = z n = z = r (cos 0 + i sin 0) in Equation (3. in polar form.. To perform a division.b) t c o s ®1 c o s ^2 ~ s m ®1 s m ®2 + i ( c o s ^1 s m ^2 + cos ®2 s m ^l)] (3.212).213) This is known as De Moivre's theorem (formula). + 0 n )] (3.210b).i sin 0 2 j r^ (cos 0^ cos 0 2 + sin 0j sin 02) + i (cos 0 2 sin 0^ ..... Generalizing Equation (3.sin 0 2 cos Q^) r2 cos 2 0 2 + sin2 0 2 (3. .210b) = n r2 [cos (0j + 62) + i sin (0 t + 0 2 )] From this equation. we obtain z n = r n (cos n0 + i sin n0) (3.194 ADVANCED MATHEMATICS First we calculate the product of two complex numbers. z n = ri (3. ±1. but it is true for all rational values of n. ±2. Then z l Z2 = r l r2 z 2 = r2 (cos 0 2 + i sin 02) (3.21 la) n = 0.. Let zj = rj (cos 0j + i sin 0j).2llb) r 2 r3 . rn [cos (Ql + 0 2 + . we write zx z 2 z 3 . We have deduced it for positive integral exponents.29a. Z2 = rjcoset + isine!) r2 (cos 0 2 + i sin 02) h.212) Setting Z = z 2 = . (3.210a) (3. we write z.214) (3. we note that I zj z 2 I = rj r 2 Arg(z 2 z 2 ) = Argz 1 + Argz 2 + 2n7t.. ( c o s 9i (3 2 1 5 a ) + i s i n 9 i) ( c o s 9 2 ~ *sin 9 2) (rt 9 1 CM r 2 (cos 0 2 + i sin 0 2 j (cos 0 2 .....215c) ..
217a) (3.COMPLEX VARIABLES 795 = J.219c) To find the n lil root of a complex number.6 2 ) + i sin (BJ .10. (3.[cos 6 2 . r.rf^l 2 cos— .. Determine (1 + i) 10 .217b) (3.i sin 62] Example 3. By definition z£ = zx Writing z 0 and z\ as z 0 = r0 (cos 9 0 + i sin 60) (3.= J.A — K \ 4 10.2. If we set z\ to be one in Equation (3. We use the polar form and write (1+i) = V2~(cosJ + i s i n j ) Therefore n ( 3 .219a) = 2 5 ' 0 0 0 (cos 2.000 .500 TC) = 25>00° • (3.22la) (3.. ±2. Suppose we want to find then* root of zj.218) (l+i) .220) .92)] Thus we note that z. Let z 0 be the n * root. n = 0.500 it + i sin 2.000 \ (3.2 . . 10. we obtain ±.216b) (3.[cos (©i .16a ) (3. .219b) (3.000 +isin——7t 4 / . we use the polar representation.[cos(02) + isin(e 2 )] = J. ±1.000 ' =2 .215d).15d) r2 = r2 A r g ( ^ ) = Arg Z l Argz 2 + 2n7t.21.
227b) (3. Example 3. .. p n . 1/n .220) becomes IQ (cos n6 0 + i sin n0 o ) = rl (cos 0j + i sin 0j) Equating the real and imaginary parts. 1. Any other value of k would yield a value of 0O which differed from the one obtained earlier by a multiple of 2n radians.. with the first argument being —. . we obtain r t cos 0j = r" cos n0 o rjsinOj = rj}sinnG0 The solution is r0 = r. n .k) + i sin (2j£ k) Denoting the roots by p 0 .22. we obtain from Equation (3. Thus the number of n^1 roots of a complex number is n. p i . .222) (3.225) where m is a remainder. Determine all the n th roots of unity. 2.227a) (3. Since 1 = cosO + isinO we have 1 1 / n = cos (0 + 2ffifc) + i sin (0 + ^ k ) = cos (^p.223a) (3. j .b) (3. Geometrically these roots lie on a circle of radius r^ whose arguments differ by ^. 0<m<n 90 = ^ + ^ (3.196 ADVANCED MATHEMATICS zx = rl (cos Ql + i sin 0j) Equation (3. .226) . we obtain all the values of 0O which produce the n distinct values of z 0 by choosing k = 0.224a. where k is an integer.1. Since. 3. .227b) (3.223b) (3. .221b) (3. for any integer k given by k = pn + m.
COMPLEX VARIABLES 797 p 0 = cos 0 + i sin 0 = 1 p! = cos ( ^ ) + i sin (^t) p 2 = cos (4ffi) + i sin (±&) (3.230e) (3.28d) We list some of the properties of the conjugate of z 1.228c) Pn.z2 = Zj • z 2 (3. 4.z =  z  2 = (Rez) 2 +(Imz) 2 Rez<lzl. 2. zx ± z 2 = z2 ± z 2 z j . 5.230f) I zx z21 = Izil lz 2 l T zl =rL. z. We have  Z l + z 2  2 = (zj + z 2 )(z 1 + z 2 ) = (zj + z 2 ) (z^ + z2) (3.2. z = z z . Show that I z x + z 2 I ^ I Zj I + lz 2 l. (1) = ^ ( z j = Zj ^l^0) (3.230b) (3. 6. (zl^°) I z 1 + z 2 l < \zx\ + \z2\ Example 3.229d) and some of the properties of the absolute value of z 1. l l m z l < Izl (3.1 = cos[fc^] + isin [(HzM] (3.229a) (3. 2.23 la.228b) (3.230c) (3.b) .229c) (3.228a) (3. This is the triangle inequality.230a) (3.230d) (3. 3.229b) 3.23. 4.
we first define curves. z 2 and Z3 are all real.234c) (3.198 ADVANCED MATHEMATICS = ZJZJ+Z2Z2 + z 2 Z j + z 2 z 2 (3. limits. so that the equation z = x(t) + iy(t) (3.232) Hence (zjZ2 + ZjZ2) = 2 Re (zjz2) It follows that (3. defined for a < t < b. Before considering these concepts. Expressions such as Z j > z 2 or zj < Z3 have no meaning unless z\.234a) (3.3 COMPLEX FUNCTIONS The concept of functions.234b) (3. 3.231c) Observe that ZJ+ZJ = 2Re(zj) (3. regions and domains in the complex plane.235a) (3.234(1) (3.31) . continuity and derivatives discussed in the calculus of real variables can be extended to complex variable calculus.(  Z l  + z2)] [z1+z2+(z1 + z2)] < 0 Hence z1+z2 < (Zl + z2) (3. Let x (t) and y (t) be two continuous functions of a real parameter t.235b) • Warning: We emphasize that complex numbers are not ordered.233) I zx + z21 2 = I zx\ 2 + 2 Re ( Zl z 2 ) +1 z 2  2 < I Zj 2 + 21 ZjZ 2  + I z 2  2 < z12 + 2z1 z2 + z22 < (  Z l  1. z 2  ) 2 This implies [  Z l + z 2  .
.31.z o  > 8 represents the exterior of the circle. The curve C is simple if it does not cross itself. the curve C is a closed curve. It can also be expressed as z = 1 The set of all points z which satisfy the inequality zzo < § (3.32b) (3. It consists of all points z lying inside but not on the circle of radius 5 with the center at zQ.31 6 neighborhood of z 0 Similarly.32a) If z (a) and z (b) are equal.33b) is called the 5 neighborhood of the point zQ. *~ X FIGURE 3. It is in fact the unit circle.COMPLEX VARIABLES 799 defines a curve C that joins the point z(a) = x(a) + iy(a) to the point z(b) = x(b) + iy(b) (3.33a) is a simple closed curve. For example z = cost + i sin t (0<t<27i) (3.34) (3. the inequality  z . This is illustrated in Figure 3.
The conditions x > 0 and x < 0 define the right half plane and the left half plane respectively. S is an open set. (3.36b). none or all of its boundary points is a region. it is called an exterior point of S if there exists a neighborhood of z 0 which does not contain points of S. Similarly. Example 3. Let S be a set of complex numbers and let z. the set of all points z (= x + i y) such that y > 0 is the upper half plane.b) The real functions u and v denote the real and imaginary parts of w. The set of all points in the plane that do not lie on  z  = 1 is an open set which is not connected. be a complex variable. The open set  z  < 1 is connected and so is the annulus 1 <  z  < 2. The unit circle does not belong to the set. Then find the value of f (1 + i). S is closed. Express the function w = f(z) = z 2 + 2 z . Just as the variable z is decomposed into real and imaginary parts.2Q0 ADVANCED MATHEMATICS A point z 0 is said to be an interior point of a set S whenever there is some neighborhood of z 0 that contains only points of S . A set that is formed by taking the union of a domain and its boundary is a closed region. Clearly an open set contains none of its boundary points. We write w = f(z) (3. An open set that is connected is a domain.3 z in the form of Equation (3. This is because we cannot join a point inside the unit circle to a point outside the unit circle without crossing the unit circle. we have the lower half plane.b) . A domain together with some.31. which varies in S. Finally. If z 0 is neither an interior point nor an exterior point of S. y) (3. A function f defined on a set S of complex numbers assigns to each z in S a unique complex number w. We now define functions of a complex variable. We write w = f (z) = u (x. If every point of a set S is an interior point of S. w can be decomposed into a real and an imaginary part. The set S is the domain of definition of f(z) and the set of all images f (z) is the range of f (z). We note that a complex valued function of a complex variable is a pair of real valued functions of two real variables.37a. z 2 in S can be joined by a polygonal line that lies entirely in S. An open set S is a connected set if every pair of points z i . The closure S of a set S is the closed set consisting of all points in S as well as the boundary of S.35) The number w is the image of z under f. If a set S contains all its boundary points . y) + i v (x. for y < 0.36a. it is a boundary point of S.
(3.COMPLEX VARIABLES 201 w = (x + i y) 2 + 2 (x + i y) .z o  < 6 We adopt the same notation as in the case of a real variable and write lim f(z) = Z z>z0 or f(z)— >Z as z — > z 0 (3.3 (x . z may approach z 0 from any direction in the complex plane and the limit is independent of the direction. y) and v (x. y)> (xo.yo) v(x. we have lim (x.310a.Z  < e whenever 0 <  z .312a) (3. The concept of limit is illustrated in Figure 3.b) f ( l +i) = u ( l .32. 8c). y) = i .yo) u(x.39a.3x] + i [2xy + 2y + 3y] = (x2 .38c) (3. We define the limit of f (z) as z approaches z 0 to be a number Z. y) = Z1 (3.311) We note that in the present case. we have u = x2xy2. l) + i v ( l . v = 2xy + 5y (3.38b) (3. y)> (xo.312b) (3.l + 7 i • Let f (z) be defined in some neighborhood of z 0 with the possible exception of the point zQ itself. . 1) = .y2) + i (2xy + 5y) Comparing Equations (3. y) is Urn f(z) = Z = Zx + iZ2 Z> (3.b) (3. it is unique.314b) If a limit exists.38a) (3.i y) = [(x2 . An equivalent definition in terms of u (x.36b.313) Zo Equating the real and imaginary parts. if for any e > 0. there is a positive number 8 such that  f (z) .x .y 2 ) + 2x .314a) lim (x.
315b) (3.316) z provided the limit exists.317b.c) .315a) ii) iii) f(z 0 ) exists lim f(z) = f(z 0 ) Z>ZQ (3.317a) (3.32 Limit of f (z) defined by Equations (3. Az = z .202 ADVANCED MATHEMATICS y A V i. An alternative definition is to let w = f(z) Aw = f ( z ) .314a. if the following three conditions are satisfied i) lim f(z) exists z>z 0 (3.f ( z 0 ) . b) Let f(z) be a function of the complex variable z defined for all values of z in some neighborhood of zQ.315c) The derivative of f(z) at zQ written as f'(zQ) is defined as f ( z 0 ) = lim z>z0 f(z) f(Zo} ~zo (3. • ' • X U FIGURE 3. The function f(z) is continuous at z 0 .z 0 (3.
320b) .317c. Example 3. the converse is not true.319a) (3. Show that the continuous function f(z) = Izl 2 is not differentiable everywhere..318) f'(z 0 ) = = (3.32la) (3.317d) we have I2 I I2I lim ^ ^oL Az>0 Az lim A z > 0 = Um A z > 0 Mm ZZ (3. dz If f(z) is differentiable at z 0 .. Let z 0 be any point in the complex plane. then using Equation (3. If zQ is not the origin. f(z0) = .319d). (cos0isin9) km ZQ + ZQ) f Ar>0L (cos 9 + i sin0) .i sin 9) Substituting Equations (3.320a.COMPLEX VARIABLES 203.. it is continuous at z 0 . we obtain . (3. n O1 .32. As in the case of real variables.319c) is obtained by using Equations (3. we express Az in polar form and write Az = Ar (cos 0 + i sin 6) Az = Ar (cos 0 .e) Az>0_ We also write f'(z) as ^—. z 0 is also zero and f'(0) exists and is zero. b) into Equation (3. 229a). f'(z) = ta Az>0 A w ^z = fa [AuO^O ^z + j AvOcoOJ AZ _ (3. .3190 ^ z^ + Z o ^ z (3.319b) "ZoZo ^z ( z o + A z ) fzQ + A z ) ~ ZQ Z o (3.317d..319d) Az>0[° °AzJ Equation (3.320a) (3. If z 0 is the origin.
. such as the rules for differentiating a constant.c .322b) The function f is not differentiable every where except at the origin where f'(0) is zero irrespective of 9.322a) (3.32lc) f'(z 0 ) = z 0 + z 0 f(z0) = z o . =— and ^— exist and satisfy the equations dx 3y dx dy 9u _ 3v du _ 9v H 3 23aM Proof: Since f(z) is differentiable at any point z 0 in D. integer power of z. If f (z) is analytic for all finite values of z. Theorem 1 A necessary condition for the function f (z) to be analytic in a domain D is that the four partial j • <. e > 2 l cos 9 + sin 9 . • All the familiar rules.i S i n . f(z) is an entire function.321b) (3. We thus define a broad class of functions. we obtain respectively from Equation (3. sum. Functions which are differentiable at a single point are not of great interest. if z approaches z 0 along the real axis (9 = 0). The basic criterion for analyticity of a complex function f(z) is given by the CauchyRiemann conditions. then f'( z o) must exist. derivatives ^r— . and along the imaginary axis (0 = n/2).du du dv . ^— . product and quotient of differentiable functions as well as the chain rule of differential calculus of real variables hold in the case of complex variables. Synonyms for analytic are regular and holomorphic. The function f(z) is analytic at z 0 if its derivative exists at each point z in some neighborhood of z 0 .2Q4 ADVANCED MATHEMATICS • = ^ + Zo ( c ° S 2 e . difference. Also the limit does not exist.z o (3. Points where f (z) ceases to be analytic are singular points. if f (z) is defined and has a derivative at every point in D.. dv . For example. z 0 + z 0 (cos 29 . . A function f (z) is analytic in a domain D.i sin 29) (3.321c) Thus f '(z 0 ) depends on 9 and is not unique.
u ( x o .33.z 0 = X . We choose to approach zQ along a line parallel to the real axis (y = constant). The increment Az is simplified to Az = z .y 0 ) = i A y (3. Equation (3.33. They are in fact the partial derivatives of u and v with respect to x. shown as path 2 in Figure 3. irrespective of the path taken.f(z 0 )] / (z .v (x 0 .324a. the two limits on the right side of Equation (3. f '(z 0 ) is given by f(z 0 ) = Jim Ax>0 u ( x o + A x . y0) ^ X + . y 0 ) .b) .325) exist.b.327a.c) y n 2 1 yo • +z<r i t 1 Xo ^ X FIGURE 3.z 0 ) must tend to a definite limit as z —> z 0 .325) Since f '(z 0 ) exists.33 Two paths approaching z 0 Using Equation (3.COMPLEX VARIABLES 205 This implies that the ratio [f(z) . In this case Az = i ( y .325) can be written as f (zo} =K + l Bi (33"26) Next we approach z 0 along a line parallel to the imaginary axis (x = constant). y 0 ) Ax (3. y o ) .X 0 = AX (3. shown as path 1 in Figure 3.317e). ^ Ax>0 v (x Q + Ax.
y3 — +1 J— . 0) = lim ""'• ( ""< 0 .329a) 1± = Pdy dx (3. 28b). This is shown in Example 3." ' ° ' 0 ) = U m l y 3 ^ .0 > = K m ("3/f° = 1 dx x >0 x (3.0) = l dy (3.329b) These are the CauchyRiemann conditions. we have  i (0. z=0 (3. They are the necessary but not sufficient conditions for a function to be analytic. z^ 0 7 7 12 2 x +y \x + y f(z) = 0 .b) satisfies the CauchyRiemann conditions at the origin.33.328a) = .i ^ dy + ^ ay (3.0) = 1 . dx ^(0. yQ) Ay>0 j Ay (3. x 3 + y 3 _ x3 .326 or 28b).0 = 1 5y y>o (3.326.331d.c) x>0 x »i (0.v (xQ.331a.f) y y>o y Similarly  ^ (0.325.206 ADVANCED MATHEMATICS The analogues of Equations (3. Show that the function .h) . 0) = lim u ' 0 .y > .328b) Comparing the real and imaginary parts of Equations (3. Note that f'(z 0 ) is given by Equations (3. ^ v (xQ> yQ + Ay) . From the definition of partial derivatives.33.331g.330a. 26) are f(z ) ° = iim u ( x o .b. we obtain ^ = ^ dx dy (3. Example 3. y o + Ay)u(x Q > y o ) Ay>0 j^y + .e. but f (0) does not exist.
From Equations (3.b) P._ _ _„ . Theorem 2 A function f (z) is analytic in a domain D. we can write using Equation (3.336c.26a. 0) = l + i Since the two limits are different. 0). satisfy the CauchyRiemann equations at each point of D and are continuous. y) and v (x.326) f'(0) =  ^ (0.333a.337a.= sin 9 dy 9=co^ 3y r (3.= cos 6. (3.336a. b) we obtain (3.b) to _UL_ x + ix = ^ L = i(li) l + i (l+i)(li) = 1+ l j 2 2 (3 . The CauchyRiemann conditions can also be written in polar form. in theorem 2.(0. we have du du dr du 30 „ du — = 1 = cos 0 dx dr dx 30 dx 3r ^ = sin 0 *L + c o s i 9u oy 3r r dd sin 0 du r 90 (3 3_37c) . 0).COMPLEX VARIABLES 207 Hence.d) Using the chain rule. we require the partial derivatives to be continuous. 0) + i ^. if the origin is approached along the xaxis (y = 0). provided the four first partial derivatives of u (x. y) exist. Let z vary along the line y =x f(z) = u + i v = 0 + ix f'(0) = lim M l M z>0 z . f'(0) does not exist.335a. dx ae = _sin^ dx r I 1 . the CauchyRiemann conditions are satisfied at (0.b) (3. .b) .334ae) ' However.0 = x>0 (3. We now show that f'(z) does not exist at (0.332) (3. Note that.
~ = e x siny.341c. Show that the function ex (cos y + i sin y) is analytic and determine its derivative. eliminating — and — from Equations (3.339a) 3r r 30 Similarly.337e) Combining Equations (3. and adding the resulting expressions yields (3.341a. b. b).d) (3. 39a.341e.339a.323a.326. d. We have f(z) = u + iv = e x cosy + i ex sin y 9u v ^— = e x cosy.338a. 37b.b) Example 3.337d) ^ l ^ r 99 (3. av r 30 (3338b) Multiplying Equations (3. Combining Equations (3.b) (3. 9u v — = exsiny ^ = ex cos y (3. b) by cos 0 and sin 0 respectively.340a.33*) (3.338a.338a) sin9 ^ + «*«.34. we obtain 3r 90 £ = i p3r r 30 Equations (3. 37b to e) yields c o s e ^ . au 3r r 30 _coseav 3r + sine.208 ADVANCED MATHEMATICS ^ = cose^^i^ 3* 9r r 90 ^ = sin0^ 3y 9r + (3. b) are the CauchyRiemann conditions in polar form.M 3r r *L = s i n 9 ^ + ™ ^ ^ 90 3r r 80 = (3. b) yields f'(z) = ^ dz = (cos0isin0) ^ or (3.f) .
The functions u and v of the analytic function are also conjugate functions (harmonic conjugates). Hence f(z) is analytic. If a harmonic function u (x. • Let f (z) be an analytic function. The solution of Laplace's equation is a harmonic function. (3. we deduce that both u and v satisfy the equations ^ + ^ = 0 (3. y) through the CauchyRiemann relations. The analytic function f (z) can then be determined.35.3 x y 2 (3. and their partial derivatives satisfy the CauchyRiemann equations and are continuous functions of x and y. v.342a. We will see in Section 3. y) = x 3 .b) (3. b).345) .343a.344a.b.d) ay2 axz From Equations (3. b) are Laplace's equations which will be solved in Chapter 5. Assume that the mixed second derivatives of u and v exist and are equal.4 that this function is ez. y) is given in some domain D.343b.329a.COMPLEX VARIABLES 202 The functions u. From Equations (3. we have fiax3 y ax ay ax2 ^ =^ =^ ax2 ay z ay2 + ^ = 0 (3.344a) (3. d). we can determine the harmonic conjugate v (x. Both the real and imaginary parts of an analytic function are harmonic functions. Show that u(x. Example 3.344b) a x2 a y 2 Equations (3. The derivative f'(z) = Y~ + i^= e x cosy + i e x s i n y (3.342c) = e x (cos y + i sin y) which is identical to the given function.343c.
c) (3.348) From Equation (3. y) and the analytic function f(z). Find the harmonic conjugate function v (x.347a) = 3x 2 y .350a) (3.351a. The function v also has to satisfy Equation (3.210 ADVANCED MATHEMATICS is a harmonic function.d) From Equations (3. Taking the partial derivatives of u.b) . we obtain ^ = 3x23y2. H. Therefore y = (^r) (3.348).329a).6 x y = . and u is harmonic.3 y 2 ) d y + g(x) (3. we deduce x = (*±i). we deduce that u satisfies Laplace's equation. From Equation (3. d).21).y 3 + g(x) where g (x) is an arbitrary function.346a.346c.349) Laplace's equation can be written in terms of z and z.350b. we obtain v = I ( 3 x 2 .347b) (3. a — dy2 = 6x (3.346c.[ 6 x y + g'(x)] (3.b) a — 3x2 = 6x . The function v can then be written as v = 3x2yy3 +c The function f is given by f = (x33xy2) + i(3x2yy3) + c = (x + i y) 3 + c = z 3 + c (3. From Equation (3. = _6xy (3. we deduce that g'(x) is zero and g is a constant c.329b) and this implies .
y) satisfies Laplace's equation.b) (3.354) dy) 3 . 3 .358) Example 3.= 0 (3. it follows that 92(P —^.360) d vv dv (3.^ dx dy = 0 (3. (3.COMPLEX VARIABLES 221 di 9z = a¥ a7 + a l a7 dz dx dz dy = 2 (a7 ~ * ay) 2 \dx + i (3.352a.—^ = 0 9y dx From Appendix I.36.355) (ax2 dy2! That is ^ p 9x 2 + 3 ^ = 4 ^ ( 3 3 j 6 ) dy2 3z9z If (p (x. This can be expressed as _ * .357) dzdz and the general solution is < p = f(z) + g ( z ) This solution is often used in two dimensional physical problems. vy) be the velocity components. I ( 9 .b) (3. . i 3 ) ( 3 9 ) dz dz 4 \dx 3y/ \dx dyj 4f4.M az az (3. The flow is irrotational and this implies that the vorticity (curlv) is zero. we obtain the equation of continuity for an incompressible flow as follows d vY dv ^ + .353a. Obtain the complex potential O(z) for a two dimensional irrotational flow of an incompressible fluid. We choose the rectangular Cartesian coordinate system and let (vx.359) .
we note that < >  and \/ satisfy the CauchyRiemann relations and are therefore harmonic functions.363) * • = _ .361a. 62a to d). The derivative of O yields both velocity components. and will be discussed further in this chapter.b) vy .328b.326. • The concept of a complex potential is widely used in hydrodynamics. we deduce dv 3v (3. Applications of complex potential in electrostatics are given in Ferraro (1956). f (3. we have ^=^+i^=vxivv dz dx dx x y Similarly using Equation (3. . ! • + ! ? . (3.361a) 3 vY 3v it =i? We introduce two functions (j) (x.211 ADVANCED MATHEMATICS From Equations (3.364a.362a to d). They are conjugate functions. we have (3.329a.359.362c.  £ = . 60). we can obtain both the potential (Re O) and the stream function (Im O).365a.362a.b) Using the complex function 3>.d) From Equations (3. Equations (3.3 . 60.b) (3. The combinationty+ i \/ is an analytic function and the complex potential is given by O(z) = < >  + iY Differentiating O(z) with respect to z and using Equations (3. v x iv y dz 8y dy x y (3. b). 62a to d). 62a to d).61b > Substituting v x for u and vy for (v). vx = £ = J f . the components v x and vy satisfy the CauchyRiemann relations. y) such that ( 3 . b) are identical to Equations (3. y) and \/ (x. The function < )  is the potential and \/ is the stream function. This can also be verified by combining Equations (3. That is.359.
. However.COMPLEX VARIABLES 2/. Similarly.? 3.. + a j z + a o where a n * 0. a 0 are all complex constants. One such property is the possibility of multiple values mentioned in connection with Equation (3. The exponential function can be written as ez _ ex + i y _ ex ( c o s y + j Sin y) (3. defines a rational algebraic function. + 4 + 2! oo (33'3a> n! " nto n! Setting the real part of z to zero. The exponential function is denoted by e z [or exp (z)] and is defined as (3.42) (3.45a. 1 z n . an_j.44a.b) . . exponential and logarithmic. The moduli  e iy  =  cos y + i siny  = 1 (3..b) For real z (y = 0). e z reduces to e x . such as polynomial.41) ez = l + z + £ + . we have (343b) n=o n! = m=o (2m)! cos y + m=o (2m+1)! i sin y (3. can be extended to complex variables.. b).. the functions become identical to the functions considered in the calculus of real variables. It is usually defined such that for real values of the independent variable z.b) (3.1 + . A polynomial function P n (z) of degree n is defined as P n (z) = a n z n + a n .44c) Equation (3.4 ELEMENTARY FUNCTIONS The definition of many elementary functions. the complex functions may have properties which the real functions do not possess. a function w(z) = P(z)/Q(z) in which P (z) and Q (z) are polynomials..46a. .44c) is Euler's formula.28a.
= . That is to say. ez=l implies z = 2mti (n is an integer) (e z *0) (3.f) (3.e x sin y ^. the values of sin y and cos y remain unchanged and e z is periodic with period 2%. Bx v = .b) (3.b. ez = e z + 2ikJi (34_7) Because of the periodicity.= e x cosy.2M ADVANCED MATHEMATICS e z  =  e x   e i y  = ex The modulus of e z is e x and the argument of e z is y. we need to consider only the strip 7i<y<7t Some of the properties of e z are: (a) (b) (c) (d) (e) e z is analytic.49h) Example 3.49a.z 2 = 2n7ti (n is an integer) If w is an analytic function of z _d_ (ew) dz = ew iw dz (3. We have e z = e x ~ iy = e x » e ~ i y = e x (cos y .49f. ^.48) e" z = L ez d(e z ) = e z dz If z\ (= x^ + i y^) and Z2 (= X2 + i y2) are two complex numbers e z i • eZ2 = e Z l + Z 2 (3.46c.49e) (3.49c) (3.410g.41.h) .49d) (3.e x s i n y ay (3.410e. Show that e z is not an analytic function of z in any domain.g) (f) (g) e Zl = eZ2 implies z j .d) (3.c.410a.d) If the value of y is increased by 2krc (k is an integer).i siny) = u + i v u = e x cos y. (3.
b) for any finite values of x.329a.412h) (3.412b) cos z is an even function [cos (z) = cos z] and sin z is an odd function [sin (z) = sin z] e lz = cos z + i sinz Euler's formula holds for complex variables (3. We list some of them: (a) (b) (c) (d) (e) sin z and cos z are periodic with period 2% 4~ (sin z) = cos z dz •£.41 la.412i) .f) Since e z is analytic for all z.412e) (3.COMPLEX VARIABLES 211 ^=exsiny.412f) (3. cosz = cotz = CO"sm z C (3.41 le. The complex trigonometric functions have the same properties as the real functions.412g) (3.412d) (3.410g to j) do not satisfy Equations (3.d) tanz = sinz.410ij) It can be seen that Equations (3.(cos z) = .41 lc. sin z and cos z are also analytic for all z. ? cos z secz=5olT' °SeCZ=i (3.sin z dz (3.z = cosz (n is an integer) sin (x + i y) = sin x cosh y + i cos x sinh y cos (x + iy) = c o s x c o s h y . • The trigonometric functions are defined as e iz_ e iz eiz + e iz sinz = — .b) (3. ^=excosy (3.412c) (f) (g) (h) (i) (j) (k) sin 2 z + cos 2 z = 1 sin(zj±Z2) = sin z^ cos Z2 ± cos zj sin Z2 cos (zi ± Z2) = cos Z\ cos z 2 + sin Zj sin z 2 sin (2n+ 1 ) ^ . Thus the CauchyRiemann relations are not satisfied and e z is not analytic.412a) (3.i sinxsinhy (3.
414k.414d) (3.sinh2 z = 1 sinh (x + i y) = cos y sinh x + i sin y cosh x cosh (x + i y) = cos y cosh x + i sin y sinh x sinh(iz) = isinz.414e) (3.414j) (3. if x is any positive real number and .414b) We recall that for real variables. z = (2n+l)^i (n is an integer) (3. cosh (iz) = cosz.414a) (3.i) (3.g) (3..f) sinhz cosechz = r\— sinhz sechz = — \ — .d) (3.— .216 ADVANCED MATHEMATICS (1) The zeros of sin z and cos z are respectively z = nn.414c) (3. sinh z is an odd function cosh2 z .b) sinlLz ? coshz cothz=cQsh_z (3. coshz The complex hyperbolic functions have the same properties as the real functions.413a.e~ z PZ u p~z + e sinhz = — .z ) = icoshz The zeros of sinh z and cosh z are respectively z = nrci. z = (2n+l)^(n is an integer) (3.412 j.413e. The other hyperbolic functions are tanh z = coshz= e (3.l) i sinhz = sin(iz) coshz = cos (iz) (3. sinh(iS.414f.413c.k) Hyperbolic sine (sinh z) and cosine (cosh z) of z are defined as e z .414h. Some of these are: (a) (b) (c) (d) (e) (f) (g) (h) (i) (j) (k) They are analytic 4~ (sinhz) = coshz dz {L (coshz) = sinhz cosh z is an even function.
420b) (3. That is to say. then Arg z is zero and the definition of Ln z is identical to the /8n in the theory of real variables.COMPLEX VARIABLES 2JZ x = e" (3. we can write i n z = Lnz±2n7ti (n is an integer) (3.421) is a multiplevalued function and since the argz can differ by multiples of 2K.422a) (3. Some of the properties of Jin z (Ln z) are: . b) yields i n z = i n l z l + iargz (3.418) (3. we obtain Izl = eu argz = v Combining Equations (3. Ln z is defined as Lnz = i n l z l + iargz 7t<argz<7i.423) If z is a real positive number.416) Since arg z can differ by multiples of 2n.417) (3. We then have the principal value of Jin z and we denote it by Ln z.419) (3.417) becomes z = eu (cos v + i sin v) From Equation (3. 20a.419).422b) The function i n z as defined by Equation (3.418.420a) (3.421) (3. we restrict the definition to the principal values of arg z (Arg z). Equation (3.415) then u = inx We now extend the definition of Jin to the complex variables and write z = ew It follows that we can define Jin z as w = inz Separating w into its real and imaginary parts (w = u + iv). (3.
einz Ln (ez) = z = z = inz r inz 2 inW Note that Ln z is not defined at the origin (I z I = 0).425a) D x FIGURE 3.g) (3.41. The negative real axis is a line of discontinuity since the imaginary part of Ln z has a jump discontinuity of 2K on crossing that line.i 2Jt i n (zj z2) = i n zj + i n z 2 i n (ez) = z + 2n7ti.424f.424c) (3.424d) (3.i) = 1 Ln 2 . We make a cut.424b) (3.424h) (3.l ) = (2n+l)rci (n is an integer) (3.424a) (3. In the resulting domain D.213. The derivative of Ln z is given by dCLnz) = 1 dz z (3. Ln z is analytic. as shown in Figure 3. ADVANCED MATHEMATICS (a) i n 1 = 2njci i n ( .424e) (3.424i) (b) (c) (d) (e) (f) (g) Ln i = i  Ln (1 . in the complex plane to remove the origin and the negative real axis.41 Domain D in which Ln z is analytic .
is l a always equal to 1? From Equation (3.i) = l i n 2 .427) (3.427) and noting that 1 can be written as e2nm. The principal values of z a and ocz are obtained by substituting i n z by Ln z in Equations (3.429a) (3. Since i n z and Ln z differ only by an arbitrary multiple by 27ti.430a) (3. (iii) (i) 1 .428b) (3.43.c) (i) i n e. If a is a complex number.43la) (3.^ i (iii) Taking the principal value. Find the values of (i) i n e = i n I eI + i arg e = 1 + i (0 + 2nK) = 1 + 2n7ti (ii) Ln (1 .426) (3.i ) .431d) (3.i I + i Arg (1 .426. ^(inz) =1 If a is any complex number. we have (i)i = e i L n ^ _ e i (iic/2) _ e n/2 Example 3.431b) (3. we define z a to be z a = exp (a i n z) Similarly the function ocz is defined as a z = exp (z i n a) (3.428a) (3.430b. (ii) L n ( l .429b) . (3. Example 3. 27). we have l a = exp [a i n e 2raii ] = exp [a (27tni)] = exp [27inp + 27tni y] = exp [2n.525b) In general z a and ocz are multiplevalued functions since i n z is a multiplevalued function.np] {cos 27Cny+ i sin 27tn'y} (3.COMPLEX VARIABLES 219.i) = i n 11 .431 c) (3.42.
To begin with. Note that if n is zero (consider only the principal value) l a is equal to 1.b (3. then I [f(t) + g(t)] dt = I f ( t ) d t + I g(t)dt Ja Ja Ja .220 ADVANCED MATHEMATICS To obtain Equation (3.2. A definite integral of a complex function of a complex variable is defined on the curve joining the two endpoints of the integral in the complex plane. This two dimensional aspect has an effect on complex integration. we stated that a complex plane is required to display complex numbers. Ja Ja Several properties of real integrals are carried over to complex integrals. Let f(t) = u(t) + iv(t). if f (t) and g(t) are complex functions.53c) where a (= p + i y) is a complex constant.424f) and we have expressed the complex number a as y + i p .52) u(t)dt + i  v(t)dt Ja la fb fb Note that both I u(t)dt and I v(t) dt are real. For example. a<t<b (3.c .51) and we assume that u (t) and v (t) are continuous functions of t. We define r f(t)dt = Ja fb tb (3. . Thus in genera] l a is not equal to 1.43lb). P and y are real. we consider the definite integral of a complex valued function of a real variable t over a given interval a < t < b.5 COMPLEX INTEGRATION In Section 3.53b) af(t)dt = a Ja Ja f (t) dt (3.53a) f(t) dt = Ja Ja f(t)dt + Jc f(t)dt (3. 3.b . we have made use of Equation (3.
52) and write /•Ti/4 rn/4 tnIA I te:tdt = I tcostdt + i  tsintdt (3. we use Equation (3. a<t <b (3.58) .54b) (3. 0<t<2n (3.55) is differentiable if both x (t) and y (t) are differentiable for a < t < b .u (3. The complex function z (t) in Equation (3.55) If z (a) = z (b) is the only point of intersection. a curve C joining the points z(a) to z(b) is given by z(t) = x(t) + i y ( 0 .56) is a simple closed curve oriented in the counterclockwise direction.57) is a unit circle oriented in the clockwise direction.3.51.54c) In Section 3. C is a simple closed curve.53d) A Jb rnIA Example 3. Evaluate I Jo t e dt.51. Since the integrand is a complex valued function of a real variable t. The derivative z'(t) is given by z (t) = x1 (t) + i y' (t). 0<t<2. a<t<b (3. This is illustrated in Figure 3.54a) Jo Jo = (t sin t +cost) 0 Jo nlA JI/4 + i (t cos t + sin t) 0 (3. But z = ek. The unit circle z = e il .f(t)dt =  f(t)dt (3. The orientation is defined by moving from z (a) to z(b) as t increases.
0<t < 1 (3.55) and if we let (C) be the curve that traces the same set of points in the reverse order.b) L = I z'(t)dt Ja (3.510) If C is given by Equation (3.51 la) A curve C that is constructed by joining a finite number of smooth curves end to end is called a contour (or path). Arrows indicate orientation Curve C is smooth if z (t) is continuous and nonzero in the interval. A formula for representing the line segment joining two points z t and z 2 in a complex plane is z = zx + t (z2 . b<t<a (3.C) is given by z(t) = x(t) + iy(t).b (3.59a.zx). curve (.y(a)] V ^ I ^ z = ( a ) = z ( b ) c2 FIGURE 3.222 ADVANCED MATHEMATICS y " z(b)=[x(b). the differential arc length is given by ds = V[x'(t)]2 + [y'(t)f dt = z'(t)dt and the length L of the curve C is given by .y(b)] / ( (  z(a) = [x(o).51 lb) .51 Simple (Cj) and simple closed (C2) curves in the complex plane. If C is smooth.
In the theory of definite integrals of real functions. For complex integrals. we have taken the moduli of all complex quantities and we are dealing with real quantities.514b) is replaced by the length of the interval of integration (= b . b).52.512) Differentiating z(t) from Equation (3. Contour integration in the real twodimensional plane will be considered in Chapter 4. It was stressed in Section 3.514a) Ja (ii) I f(z)dz c Ja < Jf(z)dz < ML c (3. yields I f(z)dz = I (u + iv) (dx + idy) c c = I (udxvdy) + i I (v dx + u dy) c c (3.2 that complex numbers are not ordered and inequalities have meaning only when associated with real numbers.513a) (3.53a to d).a.512).55). we can replace the real variable t by the complex variable z in Equations (3. we have the following inequalities: (i) f (t) dt < f(t)dt (3. the length L in Equation (3. where a and b are the limits of integration).514b) where M is the upper bound of lf(z)l and L is the length of the contour C.514a. and decomposing f into its real and imaginary parts. That is to say. Evaluate the integral I ( z .513b) The contour integral has similar properties to those of integrals of a complex function of a real variable. Note that in Equations (3. Example 3.z o ) n d z c .COMPLEX VARIABLES 22J The integral of f (z) along a curve C joining the points z(a) to z(b) is a contour (line) integral and is written as I f(z)dz = I f(z)z'(t)dt c (3. substituting the resulting expression in Equation (3.
The equation of a circle of radius r with centre at z 0 is given by z = zo + r e i t .517d) Since cos and sin are periodic.Lsin(n+l)ticos(n + l ) t J 0 (3.518a) (3.224 ADVANCED MATHEMATICS where n is an integer and C is a circle of radius r with centre at z 0 and is described in the counterclockwise direction.2K Jo r . Using Equation (3.517c) . In this case. . i) to (1.1 . show that i z4 c I dz < 4V2" (3. .515.519) where C is the segment joining the point (0. we obtain dz = i r e J t dt Using Equations (3.2K = ^ — .516) 0<t<27t (3. (3.517c) reduces to I {zzoyl dz = i c = 27Ci dt (3. Equation (3.53. 16).517a) rn e i n t i r e11 dt J = irn+1 :rn+l i271 e i ( n + 1 ) t dt (3.518b) Example 3.515) ( l(zzo)ndz= c f2n I .52. On differentiating.514b). the integral becomes (3. the integral is zero except when n is . 0) as shown in Figure 3.517b) o [cos (n + 1) t + i sin (n + 1) t] dt (3.
0) to the origin is the midpoint of the segment. It follows from Equation (3.b) I di c • < 4V2" (3. . as shown in Figure 3. As z varies along the segment.COMPLEX VARIABLES 225 y^ (O+L) ^_ 0 ^ (I+LO) »~ X FIGURE 3.519) We note that a simple closed contour C divides the complex plane into two domains.514b) that (3.520a. and the other domain is unbounded and is the exterior of C.52 Path of integration The closest point on the segment joining the points (0. The upper bound M of — is given by z4 M = lV2 j = 4 The length L of the segment is v2 . One domain is bounded and is referred to as the interior of C. i) and (1. its distance from the origin I z I varies and its minimum value is l / V z . as shown in Figure 3. The distance from the origin to that point is 1 /V2 .53. The maximum value of — is l / ( l / V 2 j and is V2 .52.
^ _ ^v . A domain that is not simply connected is multiply connected. there are no holes in a simply connected domain. Figure 3. — F^ ^^ F^ \=\ (a) (b) (c) FIGURE 3.226 ADVANCED MATHEMATICS 'f m FIGURE 3.54 shows examples of simply connected and multiply connected domains.53 ^ — ^ — — . J» n_ >.54 Simply connected (a. In other words. M X ' Interior (shaded) and exterior domains Domain D is simply connected if every simple closed curve C in D encloses only points of D. b) and multiply connected domains .
A consequence of Cauchy's theorem is the concept of path independence.329a.jp) dx dy c s (3.523a) I (v dx + u dy) = \\ (^ . We have  (u dx . That is to say I f(z)dz = I ( u d x .4). Proof: The integral can be written as in Equation (3.523a. b) are zero. Let z\ and Z2 be two arbitrary points on C and let them divide C into two arcs Cj and C2 as shown in Figure 3.H) dx dy c s (3.521) holds.55. then I f(z)dz = 0 c The integral round a closed contour is also denoted by (j) f (z) dz.513b). Since f(z) is analytic. It follows that Equation (3.522) (3.523b) where S is the domain enclosed by C. b) and the right side of Equations (3. Consider the integral round a closed contour C.j £ .521) We transform each of the integrals on the right side of Equation (3.b) .524a.COMPLEX VARIABLES 227 Cauchy's Theorem If f (z) is analytic in a simply connected domain D and if C is a simple closed contour that lies in D. We have I f(z)dz= I f(z)dz+ I f(z)dz = 0 c C! c2 (3.522) to a double integral using the twodimensional Stokes (Green's) theorem (see Section 4.v d y ) + i  c c c (vdx + udy) (3.v dy) = II (. u and v satisfy the CauchyRiemann relations (Equations 3.
Evaluate I z 2 dz along each of the straight lines OA.55 We deduce Path independence I f(z)dz = . Thus the value of the integral is independent of the path and depends only on the end points. x(t) = t. x (t) = 1. y (t) = t. OB and AB as illustrated in Figure 3.^ C.56.C 2 .51 lb) are along OA: along OB: along AB: x (t) = t. c2 f(z)te = I f(z)dz c2 (3.526c.525b) implies that the integral of f(z) from zj to z 2 is the same whether we integrate along Cj or .l c.b) (3. FIGURE 3. The parametric equations of the lines (Equation 3. 0<t< 1 0<t< 1 0<t< 1 (3.525a. y (t) = t.f) .526a.526e. Example 3.228 ADVANCED MATHEMATICS 22 I ^ .54. Equation (3. y (t) = 0.d) (3.b) Note that .C 2 is the arc obtained by changing the orientation of C2 and is the curve joining z^ and z 2 .
we have [ z 2 dz =  t 2 dt = 1 OA (3.513b).y 2 ) + 2 ixy = u + i v with dz = dx + idy (3. + 21 (3. then .56 The function z 2 can be written as Contour integration z 2 = (x + i y) 2 = (x2 .f) I z 2 d z = I (2t)dt + i I ( l .528c. b.1 + 21 J Jo Jo AB If C is the closed contour OABO.t 2 ) d t = .c) (3.528e.528a.527d) Using Equation (3. 0 »» i Ad+Lo) ^ X FIGURE 3.527a.d) J OB Jo Jo (3.b) f z 2 dz = ( (2t2)dt + i [ (2t2)dt = .COMPLEX VARIABLES 229 y • B(i + i) c.
the doubly connected domain has been cut by Lj and L2 and the resulting domain is simply connected. • A multiply connected domain can be transformed to a simply connected domain by making suitable cuts.529a) = 3 + (" l + ¥ ) " (~f + ¥ ) = ° (3. one pair of cuts is sufficient and for a triply connected region two pairs of cuts are needed so as to obtain a simply connected region. the area enclosed by them are on the left.57 Transformation of a doubly connected domain to a simply connected one .529b.57. As we move along the curves.230 ADVANCED MATHEMATICS I z2dz = I C OA z2dz+ AB z2dz+ BO z2dz (3.c) verifying Cauchy's theorem. For a doubly connected region. The curve enclosing the simply connected domain D is C L j C ^ I ^ C . By Cauchy's theorem I f(z)dz = I f ( z ) d z +  C L! f(z)dz+ C[ f(z)dz+ L2 f(z)dz = 0 (3.b) CLjQL^C FIGURE 3. Note the orientation of the curves in Figure 3. In Figure 3.530a.57.
do not intersect each other. n) inside another simple closed contour C.531) then becomes I f(z)dz = I f(z)dz (3. Let the function f(z) be analytic in domain D which contains all contours and the region between them.COMPLEX VARIABLES 231 We note that Lj and L 2 are in opposite direction (that is.532) Equation (3.55. Suppose we have n simple closed contours which we denote as C. (3. Via Cauchy's z—a theorem J 2"a c I Az_ = 0 (3. Equation (3. by convention. Evaluate I ^z— where C is a simple closed curve. their limits of integration are interchanged) and so their contributions will cancel. have no common points. the point z = a is inside C. Example 3. the function —±— is analytic everywhere. the contours C. Since the point z = a is outside C. are oriented in the same direction.I f(z)dz c c. We then write I f(z)dz = X I f(z)dz J j=i j C (3. (j = 1.533) Cj All the contours C and C.532) can be generalized to the case where there is more than one curve inside C..530b) becomes I f(z)dz = . usually in the anticlockwise direction which. is the positive direction. That is to say.531) We note that C is oriented anticlockwise and Cj is oriented clockwise.. Consider the following two J Z — 3.. Equation (3. .534) . c cases: (i) (ii) (i) the point z = a is outside C. The regions interior to each contour C. 2. We reverse the orientation of C\ and denote (Cj) by Cj.
we have I dZ. f (z) is analytic and using Equation (3. we deduce . 37c).c) 0<6<27C (3.533). Jo £e lfl Jo From Equations (3.537a.535) FIGURE 3.I dz_ I za c I c.232 ADVANCED MATHEMATICS (ii) In this case. On C and Cj and the region enclosed by the two curves. za (3. The integral then becomes _ _dz_ _ i_e±_ i 2TI i (3. we write z = a + eei6.58. the function is singular at z = a and we enclose it by a circle Q of radius e with centre at the point of singularity as shown in Figure 3. b.58 Contour integral with point z = a inside C To evaluate the integral along Cj.535.= .536) d6 = de = J z~a c.
we have verified Equation (3.539) J z~zo c In Example 3.COMPLEX VARIABLES 221 I &.538) Cauchy's Integral Formula Let f(z) be analytic in a simply connected domain D and let C be a simple closed positively oriented contour that lies in D. Example 3.= 2%\ I z —a c (3. Integral Formulae for Derivatives If f (z) is analytic in D. The values of these derivatives at a point z 0 in D are given by the formulae f'(Zo) = ^/(7vdz .540C) In Equations (3.56.55. we have 1 M _ dz = 2 n i f ( z n ) (3. all derivatives exist. it has derivatives of all orders in D which are also analytic functions in D. C is any simple closed path in D which encloses zQ. For any point zQ which lies interior to C. Find the value of the integral I e coshz dz c .540b) f (n) (z 0 ) = ^ ( f(Z) . We omit the proofs of these results but consider their applications.540a toe). We also note that if one derivative exists. (zz0) < 3 '5"40a) (3. dz (3. This is only true for complex variables.539) for the case f (z) equals to one.
234
ADVANCED
MATHEMATICS
where contour C is a square whose sides are: x = ±4 and y = ±4, described in the positive direction. The region D, the contour C and the point z 0 are shown in Figure 3.59. From Equation (3.540b), we identify f(z) = e z coshz z0 = n (3.54 la) (3.541b)
y '•
D
'C
7T~4
x
FIGURE 3.59
Integration around a square
The conditions for Equation (3.540b) are satisfied and it follows that
f
eZcoshz
dz = in  ^  (ezcoshz)
(3.542)
1 (znf
dz2
c Carrying out the differentiation, we obtain
2
— (ez cosh z) = 2 e z (sinh z + cosh z) dz 2 = 2e 27r atz = 7 C
(3.543a)
(3.543b,c)
COMPLEX VARIABLES
231
Combining Equations (3.542, 43b) yields
[
eZ c o s h z
dz = 2irce271
(3.544)
c
{z~n)3
Example 3.57. Let f (z) be analytic inside and on the circle of radius R with its centre at the origin. Let z 0 (= r 0 ei0°) be any point inside C as shown in Figure 3.510. Show that
27C
JQ R 2 2r 0 cos(e 0  ¥ ) + r5
p (R^)f(R^)
/
(3.545)
y
L/^
ro
V z'
o
FIGURE 3.510
Point z 0 and its inverse point z1
Obtain the real and imaginary parts of f (z 0 ). Since z 0 is inside C, using the Cauchy's integral formula (Equation 3.539), we have
f(z } = v °;
^ f
2TII J zz0
ffldz
(3.5.46)
c
We define the inverse point Z\ of z 0 with respect to C to be
226 2 2
ADVANCED MATHEMATICS
zj =   = f ei9° zo o
(3.447a,b)
(3.548) The point z\ is outside C and from Cauchy's theorem (Equation 3.521), we have
(3.549a) Subtracting Equation (3.548) from Equation (3.546), we obtain
f(z ° } =
MIt V — T ~
c
2™J
f(z)dz
( 3 5"49a)
= ^ f
c
(zzo)(zR2/z0)
k°~ R2/ *°] , f(z)dz
(3.54%)
(3.550a) (3.550a) (3.550b) (3.550c)
We change to the polar form and write z = Re iv t / dz = i R e ^ d \ ) / z0 = r o e i e " Equation (3.549b) becomes
f(Zo) =
^ C I
27C1
e^o(ro R X ) f ( R ^ )
ei(60+V)(r2_R2)f(RiV) ® \r0 K j t V K e /
) [R^^
(3,.51a)
°
Jo \ [R e^  r0 eieo] [R j v _ ( R ^ / J e^o] /
Z2* I" (3.551b)
= JL
27C
Jo [ (R e*V  r 0 ei6o) (r0 j *  R e ie o)J
. JL r
2» J o
=
7
<R a f/fe^
(R2r02)f(Re^) ^
d¥
(3,.51c)
a5 . 51d)
(R eiV _ f() ei60) ( R e iV _ r o eiOo)
^ p
271
Jo R 2 2Rr 0 cos(e 0 \ l /) + r2
COMPLEX VARIABLES
237
To obtain the real and imaginary parts of f (z 0 ), we write f (z 0 ) = u (r0, 60) + i v (r0, 80) f ( R e i v ) = u (R, \\f) + i v (R, \/) (3.552a) (3.552b)
Substituting Equations (3.552a, b) in Equation (3.55Id) and equating the real and imaginary parts, we obtain
u(r0,90) = £ f
Z71
Jo
2 ^ ° ^ \ d¥ R  2 R r0 cos (6 0  \/j + r^
(3,53a,
v(r0,9j = JL I'"
111
te4)^*>
Rz  2 R r0 cos (6 0  \/j + r£
d¥
(3.5.53b)
Jo
Equations (3.55Id, 53a, b) are Poisson's integral formulae and are important in potential theory. We note that the values of f at any point inside C can be determined from the values of f on C. The functions u and v are harmonic functions, that is to say, are solutions of Laplace's equation, and we deduce that the solution of Laplace's equation is determined by the values of the function at the boundary only.
Morera's Theorem
If f is a continuous function in a simply connected domain D and if
I f(z)dz = 0 c
for every closed contour C in D, f (z) is analytic in D.
(3.554)
Goursat proved Cauchy's theorem (Equation 3.521) requiring only the existence and not the continuity of f'(z). Morera's theorem implies that f(z) is analytic as a consequence of Cauchy's theorem. Maximum Modulus Principle If f is a continuous analytic function and is not a constant in a closed bounded domain D, then I f (z) I has its maximum value on the boundary C and not at an interior point. If M is the maximum value of lf(z)l on C,
2i£
ADVANCED MATHEMATICS
I f(z)l < M for all z in D If f (z) is constant, lf(z)l < M for all z in D
(3.555a)
(3.555b)
The maximum modulus principle is true for harmonic functions but not for any smooth real valued functions of two real variables. 3.6 SERIES REPRESENTATIONS OF ANALYTIC FUNCTIONS
In this section, the equivalence between analytic functions and power series is explored. The concept of sequences, series and power series of complex numbers are in many cases similar to those of real variables. Sequences and Series Let Zj, Z2,..., zn be a sequence of complex numbers. A sequence {zn} converges to zQ if lim z n = z0
n> oo
(3.61)
Alternatively Equation (3.61) is stated as z n —> z 0 as n —> °°. Equation (3.61) implies that for every e > 0, there corresponds a positive integer N, such that znz0 < e N depends on e. If the limit does not exist, then the sequence {zn} diverges. Separating z n and z 0 into their real and imaginary parts, Equation (3.61) becomes lim x n = x0
n> °°
foralln>N
(3.62)
(3.63a) (3.63b)
lim yn = y 0
n> oo
If the sequence {zn} converges, then the sequences {xn} and {yn} also converge. Example 3.61. Discuss the convergence of {(1 + i) n }. We write z n [= (1 + i)n] in its polar form
COMPLEX VARIABLES
229
zn = \4l
(cos 3 L + i sin EJ]"
(3.64a)
(3  6 " 4b)
= 2n/2[coSnJt + i " n Y ]
The sequence 2 n
cos * ^ oscillates and does not converge. Since the real part of the sequence {zn}
does not converge, the sequence {zn} diverges. The same thing can be said for the imaginary part.
Let {zn} be a sequence and let S], s2, s 3 ,..., sn be the partial sum defined as follows
sl
= zl >
S2 = z l + Z2
'
S3
= z l+ Z2 + Z3 '
sn = z l + Z2
+ ••• + z n
(3.65ad)
If n—> °°, we have an infinite series. If the sequence of the partial sums {sn} is convergent, that is to say lim s n = s
n> oo
(3.66)
exists, the series X zn is convergent and the complex number s is the sum of the series. If {sn} diverges, the series diverges.
oo
A necessary (but insufficient) condition for a complex series X z n to be convergent is that the
n=l
lim z n vanishes. That is to say, s n  s n . j in Equation (3.65d) tends to zero for large n.
n —> oo
The sum of a convergent series of complex numbers can be found by computing the sum of its real and
oo
imaginary parts. A series is absolutely convergent if X  z  is convergent.
n=l
Example 3.62. Discuss the convergence of the geometric series
oo
X zn = 1 +z + z 2 + ... + z n + ...
n=0
n
(3.67)
240
ADVANCED
MATHEMATICS
The partial sum s n is given by lzn+1
1  z
sn
=
(3.68)
If I z I < 1, we deduce
lim
n —> oo
sn
= 7T7
i
z,
(36"9)
The series £ z n converges to —^—. 1 z The function —— is analytic inside the circle I z I < 1 and has the representation
1 Z
oo
j
1

= X zn
n=0 "
(3.610)
1Z
If I z I > 1 , the series diverges.
Comparison Test
oo
Let X M be a convergent series with real nonnegative terms. If, for all n greater than N,
n=0
lz n l < M n the series X z n also converges absolutely.
(3.61 la)
Ratio Test
oo
Let X z be a complex series and
n=0
n
lim J^fL = L
n > oo  Zn I
(3.612)
If L < 1, the series converges absolutely and if L > 1, the series diverges. No conclusion can be drawn if L is one. Note that in both tests we use the absolute value of zn since complex numbers are not ordered.
COMPLEX VARIABLES
241
Example 3.63. Determine the convergence of the series
X
n=l
—.
2n
Using the ratio test, we have
J z ^
l z nl
=
 n + l+ i  2 _ ^ _ . i t ( n + l ) 2 + l ]
2 n + 1 n + i  2
2
6 _, 3
n2+l
Taking the limit as n tends to infinity yields ^ Since the limit is less than one, the series converges.
A series of the form
oo
Z
n=0
c n ( z  z / = co + C l (zz o )+ ... +c n (zz 0 ) n + ...
(3.614)
where z is a complex variable, z 0 , c 0 , Cj, ... are complex numbers is a power series. By a change of origin, we can set z 0 to be zero. In Chapter 2, we have shown that every power series of a real variable has a radius of convergence R. This result can be extended to complex variables. Every series has a radius of convergence R (0 < R < «>) and the series converges absolutely if I z  z 0 I < R and diverges if Iz  zQ I > R. On the circle of convergence (Iz  z 0 I = R), the series may converge at some points and may diverge at other points. When R is zero, the series converges only at z 0 and when R is infinity, the series converges for all values of z. The radius of convergence depends on the absolute values of l c n l . converges to the limit L, the radius of convergence R is given by If the sequence y cnl
R =f
An alternative equation for R is 1 = lim  ^ ± 1 .
(3.615)
(3.616)
if the limit exists. A power series represents a continuous function at every point inside its circle of convergence.
242
ADVANCED MATHEMATICS
~
(l)nZ2n
Example 3.64. Determine the radius of convergence of the series the function it represents. From Equation (3.614), we identify 0,
c =
2 J —
n=0
22n(n)2
and determine
if n is odd (3.617a,b) .f . , if n is even
(l)n/2 —— 2 n [(n/2)!] 2 For odd values of n
V l c nl = 0
For even values of n lim V ' c n l
n>oc
(3.618a)
= lim
1
=0
(3.618b,c)
n >~ 2 [ ( n / 2 ) ! ] 2 / n
Thus R is infinity and the series converges absolutely for all values of z. In Chapter 2, we have defined Bessel functions of a real variable. Replacing x by z, we find that the series represents the complex Bessel function of order zero.
Taylor Series
We now consider the expansion of an analytic function f (z) as a power series. Let f (z) be analytic everywhere inside the circle C with centre at z 0 and radius R. At each point z inside C,
f(z)= X ^  f M z  z o ) n
n=0 n!
(3.619)
That is, the power series converges to f (z) when I z  z 0 I < R. We first prove the theorem when z 0 is the origin. Let z be any point inside the circle C of radius R, as shown in Figure 3.61. Let C\ be a circle with radius Rj < R, and let £, denote a point lying on Cj. Since z is interior to Cj and f(z) is analytic, we have, using the Cauchy's integral formula
COMPLEX VARIABLES
243
f(z) = JL f ^ J
27C1 ] ^_z
(3.6.20)
c,
Note that Cj has to be positively oriented.
y '•
FIGURE 3.61
Illustration for the proof of Taylor series
We expand —— in powers of — (< 1), as
L
= Ifl ^r1
(3.62la)
=
1
i + i + i \ 2 + ... + /z\nl
+
( z/ ^)
(3.621b)
=
J  +  L z +  L z 2 + ... t  L z ^ ' + z"
1
1
(3.621c)
Multiplying each term by ——r and integrating around the circle, we have
ZJll
244
ADVANCED
MATHEMATICS
2*1 J ^_z
27C! J
<:
J
2
J
(
)n
(3.622)
Using Equations (3.539,40a to c), we obtain f(z) = f(0) + ^ z 1! where ^ z 2 2! . . . +^^zn(n —1): Rn(z) (3.623a)
+
+
1 +
(3.623b)
Ci
To evaluate the remainder term Rn, we note that
ISZI
>^z = Rjr
(3.624)
If M denotes the maximum value of If (^) I on C\, we write the absolute value of the remainder as follows
rn M2TCRI MR, it \n
Rn(z)
< I
L =  — L
—
(3.625)
2TC ( R i  r j R ;
( R l " r ) lRl/
Since (r/Ri) is less than one, it follows from Equation (3.625) that lim Rn = 0
n> °°
(3.626)
We proved that f(z) has a power series representation given by Equation (3.623a) with R n tending to zero as n tends to infinity. That is to say, f (z) has an infinite series representation. The proof was restricted to the case where z 0 is the origin. This infinite series is the Maclaurin series which is a special case of the Taylor series. To extend the proof to the case where z 0 is not the origin, we need to shift the origin to z 0 . We define a function g(z) to be g(z) = f(z + z 0 ) (3.627)
Since f (z) is analytic in I z  z 0 I < R, g (z) is analytic in I (z + z 0 )  z 0 I < R, which is I z I < R. Thus g (z) has a Maclaurin series expansion which is written as
COMPLEX VARIABLES
241
g(z) = E ^  r ^  z " ,
n=0 n!
lzl<R
(3.628)
Replacing z by ( z  z 0 ) in Equations (3.627, 28) yields g ( z  z o ) = f(z) (3.629)
g(zz0) = Z o ^  ^ ( z  z o ) n , Combining Equations (3.629, 30), we obtain
lzzol<R
(3.630)
f(z) = Z  — ^ ( z  z o ) n ,
n=0 n!
lzzol<R
(3.631)
Note that in the case of complex variables, if f (z) is analytic in I z  z 0 I < R, the Taylor series represents the function, the remainder term Rn tends to zero as n tends to infinity. In the calculus of real variables, the remainder term R n (Equation 1.212) does not have this property. In the theory of complex variables, an analytic function has a power series representation and the power series is analytic. If f (z) is analytic in a domain containing z 0 , and zj is the nearest singular point to z 0 , the Taylor series (Equation 3.631) is convergent in the domain I z  z 0 I < I Zj  z 0 I. Since the Taylor series is convergent for an analytic function, term by term differentiation and integration are permissible. The radii of convergence of the differentiated and integrated series are the same as that of the original series. Example 3.65. Expand 1 / ( I + z 2 ) in a Taylor (Maclaurin) series about the origin. Determine its radius of convergence. The singularities of —  — are 1+z2 z = i, z = i (3.632)
The function —  — is analytic in the domain 1+z2 Izl < 1 Using Equation (3.623a), we obtain — 1 — = l  z 2 + z 4  z 6 + .... 1+z2 (3.634) (3.633)
246
ADVANCED MATHEMATICS
Since —  — is analytic in the domain I z I < 1, the radius of convergence of the series is one. 1 + z2 The radius of convergence can also be deduced from Equation (3.615) and is found to be one, as expected. Note that, in the case of real variables, the radius of convergence can be deduced from the series expansion only. The function l / ( l + x2) has no singularity along the real line. Laurent Series If the function f(z) is not analytic at a point zQ, it does not have a Taylor series about z 0 . Instead it can be represented by a power series with both positive and negative powers of (z  z 0 ). This series is a Laurent series. Laurent's theorem can be stated as follows. Let D be the annular region bounded by two concentric circles C o and C^ with centre z 0 and radii Rj and R.2 respectively, as shown in Figure 3.62. Let f(z) be analytic within D and on C o and C^. At every point z inside D, f(z) can be represented by a Laurent series which can be written as
f(z) = X a n (zz 0 ) n + X
n=0
n—
(3.635a)
n = l ( z _ Z ( ) )n
where
an
= T^
I
f(
^ ^ , .
n = 0, 1, 2, ...
(3.635b)
bn =  L (
f &
^
,
n = 1, 2, ...
(3.6350
The integrals around Co and C^ are to be taken in the anticlockwise direction. The proof of this theorem is similar to that of Taylor's theorem. We surround the point z by a circle y as shown in Figure 3.62. Let £ be any point on the curves C o , Q or y. From Equation (3.533), we have
[tJL^ilGLK (1^=0 J Sz J Sz J Sz
C, Co Y
(3.636)
where C\, C o and y are considered to be in the anticlockwise direction.
COMPLEX VARIABLES
247
yt
\.
D
y
•
X
FIGURE 3.62
Illustration for the proof of convergence of Laurent series
Using Equation (3.539), we write the last integral as
27U f (z) = [ ^ ^
Y
(3.637)
Combining Equations (3.636, 37) yields
f(z) =
^
[ m«k _ J , ( £ ^ i
^_z 2711 J c0 ^_z
0.638)
27C1 J c,
On C], 11, I > I z I and we have the same situation as in the case of the Taylor series. Equations (3.621 to 26) can be carried over and we have
1
I f (5) d^ _ y
c,
,
,n
r 3 6 _ 39 v
27Ti J ^ 7
"ntt)^12"^
(
}
where the an are given by Equation (3.635b).
248
ADVANCED
MATHEMATICS
On Co, I z I > 11,1 and, in this case, we write
^ =
£z
=
L
(zz o )^z o )
= _JL_ L M " 1
(zz0) L
+
(3 . 6 . 4 oa,b)
(zz0)
_ i _ + j^zj
(zz0) (zz0)2
+ fe z z Q T +
(zzo)n
fez/
(zzo)n(z^)
(36.4Oc)
Multiplying Equation (3.640c) by ~
f(^) and integrating around C o , we obtain
(3.641)
where b n are given by Equation (3.635c). The remainder Tn is given by
Tn = L ^ _ (
27ti (zz o ) n j
Co
fefff^
(z^)
(3 . 6 .42)
Let K be the maximum value of I f(^)  on Co. For any £ on Co, we have
z$ = (zz o )fez o ) > I zz 0 1  I^z 0 1 = r  ^
It follows that
(3.643a,b,c)
(3.644a,b) Since Rj < r, I T n I —> 0 as n —> °°. Hence
2^7/£f  .  , M ^ Co
0.645)
Combining Equations (3.638, 39, 45), we obtain Equations (3.635a to c). The series in Equations (3.639, 45) converges in the domain Iz  z 0 I < R2 and Iz  z 0 I > R} respectively. Consequently the series in Equation (3.635a) converges in the annulus Rj < Iz  z 0 I < R 2 .
COMPLEX VARIABLES
249
Comments 1. If f(z) is analytic in a region l z  z o l < R , the coefficients b n in the Laurent series are zero. The Laurent series reduces to the Taylor series about the point z 0 . The Laurent series in a specified annulus Rj < Iz  z 0 I < R/> is unique. That is to say, if, by any method, we have obtained a series representation for an analytic function in the given annulus, that series is the Laurent series. In the examples that follow, we shall derive a representation of f (z) by methods other than via Equations (3.635a to c). Since f(z) is analytic in the annulus, the contours Cj and C o in Equations (3.639, 45) can be replaced by any circle C lying in the annulus. That is to say, the contour integrals can be taken around the same curve C, as long as C lies in the annulus Rj < Iz  z 0 I < R2.
2.
3.
Example 3.66. Expand f(z) =
z+3
(3.646)
(z2z2jz in powers of z in the following regions (i) (ii) within the unit circle about the origin, within the annular region between concentric circles about the origin having radii 1 and 2 respectively, exterior to the circle of radius 2.
(iii)
The function f (z) can be decomposed in partial fractions as
f(z)
• ,(, z 2)(!+n
(3  6  47a)
= 'h + eltl) * 3(£i)
(i) When 0 < I z I < 1, we write f (z) as
f(2) =
(3647b)
~h ~ VI I1 " f)" 1 + 3 (1 + z)1
(3.648a) (3.648D)
= f^£(f)n+£(l)nzn
2z 12 n =0 V2/ 3 n =0
250
ADVANCED
MATHEMATICS
(ii)
When 1 < I z I < 2, we write
f(z) =
 ^  i # l ) + i(riij
(3649a)
= " i  4 nf0(t)"+ £„?„<')>)"
(iii) When I z I > 2, we have
(364%)
(3.650a)
= "£• £j.©"+&.I H)^
<3.6 5Ob ,
Note that the series in Equations (3.648b, 49b, 50b) converge in the region indicated. In (i), (ii) and (iii), we have written f (z) in a form such that when we expand the appropriate expression as a binomial series, the expansion is valid. The series we have obtained are the Laurent series expanded about the origin which is a singular point of f (z). The other two singular points of f (z) are at z = 1 and z = 2. The domain is divided into regions such that in each region the function is analytic. Example 3.67. Prove that
oo
cosh(z+ i ) = a where
,2n
o
+ I anzn+L,
Iz I> 0
(3.65 la)
an =  L I
%
cos n6 cosh (2 cose) d0
(3.65 lb)
JO
The function cosh (z + — ) is analytic for all nonzero finite values of z. Therefore it can be expanded in a Laurent series at any point z about the origin in the region 0 < I z I < °°. Equation (3.635a) becomes
00 00
cosh (z + 1) = X o a n z n + ^
b n (I)"
(3.652)
where an and b n are given by Equations (3.635b, c).
the unit circle centered at the origin. the function remains unchanged and a. we divide the region of integration and write J f2n rn rln cosh(2cos0) sinn0 d0 = I cosh (2 cosG) sinn0 d0 + I cosh(2cos0J sinnG d0 0 JO in (3.655b) = L I 71 Jo cosh (2 cos 0) [cos n0 . = b n On C. we choose a common circle C.654b) (3. We note also that on interchanging z and —.e.656c) 0 = 0 Jo (3. the unit circle.655c) To evaluate the second term on the right side of Equation (3.653) J^f n cosh(e'° + e">)..ed9 e i(n+i)e v 2ni I ' Jo tin = L I 2% I JO tin cosh (2 cos 0J e" i n 9 d0 (3.656a) J fn rO cosh (2 cos©) sin n0 d0 + I cosh [2 cos(2n§)] sin [n(27t())] (d()) h 0 (3. we have z = eie dz = i e i e d 8 The coefficients a^ are given by (3.i sin ne] d9 (3.656b) cosh (2 cosG) sin n0 d0 + I cosh (2 cos ((>) [sin n§\ d<) (3.655c).656d) .COMPLEX VARIABLES 251 To evaluate ^ and b n .554a) (3.
658b) (3.^ .) I xz + y z / (3. for all values of y and. we obtain < >  + i \  / = a 1 (x + iy) + b l ( x " 1 ^ (3. 56d) yields an = . In Example 3. incompressible flow past an infinite stationary circular cylinder of radius a.660) .659a) x2 + y2 ¥ = Y fa! . we deduce $ = atx+ blX + i y fal .68. there is no singularity and the complex potential O(z) is analytic and can be represented by a Laurent series. incompressible flow. The centre of the cylinder is taken to be the origin.657) We have shown in Example 3. We start by choosing the simplest Laurent series given by O(z) = a l Z + ^L (3. irrotational.. Deduce the complex potential for a twodimensional.36 that the potential < j ) and the stream function \/ are given by the real and imaginary parts of <E>(z) respectively.252 ADVANCED MATHEMATICS Combining Equations (3. There is no flow in the region I z I < a and singularities may be present in this region. Separating O(z) into its real and imaginary parts.659b) The cylinder is a streamline and we can assume that \/ is zero on the cylinder.L cosh (2 cos e) cos n9 d6 (3.659b). we have shown that an analytic function can be a suitable complex potential for an irrotational..65 lb) Jo Example 3.658b). we obtain 0=L .658a) U +y) = a l x + ~~ xz + yz From Equation (3. from Equation (3. This implies that \j/ = 0 for x 2 + y 2 = a2.M (3.36. In the region a < I z I < °o.655c.^ .r ) \ x z + y 1} (3.
we obtain as (x2 + y2) tends to infinity vM = &\ From Equations (3. 659a).COMPLEX VARIABLES 251 Far away from the cylinder. z 0 is an isolated singular point. If z 0 is a singular point but f(z) is analytic in the region 0 < Iz .660.6.664) (3.21 < R. The point z = 2 is an isolated singular point.41. 3.663) (3. The origin is not an isolated singular point. 62). The function 1 / (z .z / + X b n (zz o r n n=0 n=l (3.362a.657) can be written as ^ ) = v0O(z +  . The function i n z is singular at the origin but also along the negative part of the real axis as illustrated in Figure 3.) (3. There are many other singular points near the origin. That is to say.635b.zQ I < R for some positive R. 0) From Equations (3. If z 0 is an isolated singular point in the annulus 0 < Iz .2) has a singular point at z = 2 and is analytic in the region 0 < I z . b n are given by Equations (3. the flow is undisturbed by the cylinder and we assume the velocity distribution to be v = (v^.662) (3. . f (z) is analytic in a region in which the point z = 2 has been removed.35a) where the coefficients an.661) The function O is analytic in the annulus a < I z I < « and satisfies all the boundary conditions and is the complex potential for the present flow. c).7 RESIDUE THEORY We have defined the singular point (singularity) zQ of f(z) to be the point at which f(z) ceases to be analytic. f (z) has a Laurent series representation which can be written as oo oo f(z) = X a n ( z .z 0 I < R. we deduce t>! = a 2 v 00 Equation (3.
Discuss the nature of the singularities of (i) cosz1 . We can also obtain this result using 1'Hopital's rule. (3.. the expansion of (cosz . The coefficient bj is the residue and is denoted as Res (z 0 ) or Res [f (z).254 ADVANCED MATHEMATICS We consider three types of singularities (i) If all the coefficients b n are zero.. the function &&%.l ) / z 2 can be deduced to be cosz ~1 z2 = _ l +^ _ ^ + 2 4! 6! ( 373) { } We define the value of (cos z . the Laurent series reduces to the Taylor series. We define lim f(z) = aQ z>z 0 (3. z 0 ] (iii) If all the coefficients b n do not vanish. .71. If m is one. The singular point is a removable singular point. j s n o t defined at the origin.71) For example. b m ) are nonzero and the coefficients b n vanish for all n > m.72) The series in Equation (3. The singular point z 0 is a pole of order m if (bj. . (ii) If all but a finite number of the coefficients b n are zero. . the function is analytic everywhere... We can also use l'Hopital's rule to deduce that its value at the origin is (1/2). (iii) M ^ z5 at the origin. z2 (ii) e 1/z2 .72) is convergent and we define f (0) to be one..^ +  ^ . (i) The expansion of cos z in powers of z is known. The origin is a removable singular point.l ) / z 2 at the origin to be (1 IT). z 0 is a pole. z 0 is a simple pole. Note that the function is not defined at the origin. The origin is a removable singularity. In so doing. z 0 is an essential singular point. The function has a series representation f(z) = s i  z = 1 . Example 3.
76b) If a0 is zero and the other coefficients an do not vanish.77d) g(z) = X a n( z . Equation (3.] = (zzo)mI n=m an(zzo)n"m (3.74) (iii) The expansion of (sinh z) / z 5 is obtained by dividing the expansion of sinh z by z 5 to obtain sinhjz z5 = J_ + _ J _ + JL+A + . the first ( m .z o ) m + * + = (z .1 ) derivatives of f(z) at zQ vanish.z o) n " m = (zz o r m f(z) n =m (3... (3. f (z0) is zero and z 0 is a simple zero.f) . z 4 3!z 2 5! 7! ( 3 7_5) The origin is a pole of order four. e 1 / z = 1 +iH—1—h . am_1 are all zero and an (n > m) are nonzero.COMPLEX VARIABLES 255 (ii) The expansion of the exponential function is 1 + ... • The zero of a function f (z) is the point at which the value of f is zero.z o) m [am+ a m + 1 ( z .76a) an = LJM (3. the origin is an essential singular point.. al5 .zo) + .. .77e. f (z) has a zero of order m at z 0 . If a0. That is to say.z o ) m + a m + j (z . + n!(z 2 n ) z2 2 z 4 We have an infinite series.77a) (377b) (377C) = (zz o ) m g(z) where (3..76a) becomes f(z) = am (z . The Taylor series expansion of f (z) about any point z 0 (not a singular point) can be written as f(z) = X a n (zz 0 ) n n=0 f(«)/ 7 ) (3..
) .z o ) n + m + bx (z .78a) (ZZ O ) (zzo) = (z .78f.24+\ ( 3 79d) .. .78b) = (zzormi(z) where oo (3. Expanding the function f (z).b) [2z4 2z6 I2 (3.z / 1 . i (z) If h(z) has a pole of order m at z 0 ..h) h(z) = £ an (z .g) i(z 0 ) = b m ^ 0 If f(z) has a zero of order m at z Q ..+ ^ .^ . + — ^ n=0 OO (3. we have f(z) = ( 2 c o s z .77g.z 0 r m I > n ( z ~ z o ) n + m + b l ( z . Determine the order of the pole of (2 cos z . its Laurent series can be written as (3.e) (3.79c) = z*\h. + b m = (z .256 ADVANCED MATHEMATICS g(z0) = a m * 0 If h (z) has a pole of order m.+ .z / + ..1 + . n (z) If f(z) and g(z) have poles of order k and m respectively at z 0 . .zo) m " ! + .+ b m n=0 (3.^ + .2 + z 2 ) 2 = [ 2 ( l . their product fg has a pole of order k + m at z 0 .78c) X (z) = X an (z . it is easier to consider the zeros of (2 cos z .72. Instead of looking at the poles of (2 cos z .2 + z2j (3.2 + z 2 )" 2 . Example 3.2 + z 2 )~ 2 at the origin. j— has a pole of order m at z 0 .79a.78d.\ .z o ) m h(z) n=0 (3.2 + z 2 ) 2 . —j— has a removable singularity at z 0 .
we have I f(z)dz = X I f ( z ) d z Jc j=i JCJ (3. 12b).712a) Jq (3.. .711) Consider the integral I f(z) dz for a fixed value of i... positively oriented. Cauchy's residue theorem states that k f(z)dz = 2ni £ Res[f(z). .711. z n contained in the interior of C. Each circle Cj (j = 1.710) Proof. Let C\. The residue bj as defined by Equation JCi (3.635c) is given by bj = . within and on which a function f (z) is analytic except at a finite number of singular points z l5 Z2.. with their centers at the isolated singular points z 1. . the function l/f(z) has a pole of order 8.. • If all the singularities of f(z) in the finite complex plane are poles.. as shown in Figure 3.L 27tl I f(z)dz (3. . z 2 .COMPLEX VARTARLES 257 The function f(z) has a zero of order 8.zJ Jc n=l [ (3. .. Cauchy's Residue Theorem Let C be a simple closed contour. f(z) is a meromorphic function.. k) lies inside C and exterior to the other circles.710) is obtained by combining Equations (3. From Equation (3. C2.533).. Therefore. .71. be k circles each positively oriented..712b) 27uib1 = I f(z)dz Equation (3.. z^ respectively. Q.
f(z) has a simple pole at z.713) From Equation (3.z i ) n n=0 + . which implies that h(z) has a simple zero at zj [h(Zj) = O].713).715. that is to say f(z) = S j l (3. X FIGURE 3. we deduce that bj is given by bj = lim (zZj)f(z) Z—>Zj (3.71 Illustration for the proof of Cauchy's residue theorem Some Methods of Evaluating the Residues (i) Simple pole at n The Laurent series of f(z) about Zj can be written as f(*)= £ a n ( z .228 ADVANCED MATHEMATICS y i _ ^.714) If the function f (z) is given in the form of a rational function.714) can be written as . Equation (3.^ (z~zi) 0.
717b) (zzj)2 = (zZirmg(z) where g (z) is analytic and nonzero at Zj (see Equations 3.717a).ta J^L kzj£ld dz 1 "" 1 If f (z) is as given by Equation (3. it is possible to decompose f (z) into its partial fractions. By comparing coefficients of powers of (z .716c) f(z) = £ a n (zz i ) n + . we can expand g(z) and h(z) in their Taylor series.+ .717a. it has a Taylor series which can be written as g(z)= I^r^Zi)" n=0 n (3718) Comparing Equations (3. Since g (z) is analytic.715).^ n=0 (ZZi) + .. . .16a) (3.. 18). 19b ) (m1)! . we find that the residue b } is given by (ml)( \ b> = isdyr z ^^i ("19a) (3 .A .716b) = lim / i^\ *>* \[h(z)h(z i )]/(zz i )/ = ^ L h'(Zi) The derivative h'(zj) is nonzero because h(z) has only a simple zero at zy (ii) Pole of order m > 2 The Laurent series of f (z) about Zj is (3. 7 . In many cases.K m 1 z ^ Zi fef^M [h(z)h(zi)J (3. b.COMPLEX VARIABLES 259 b.78a to g).717a) (3. we can determine bj.zj) with Equation (3. .7. +*»(ZZi)m (3.
I ^ Z 7 2 ) ^ Z = 2ni{Res [f(z). C2 and C 3 are circles with center at the origin and of radii 2. We illustrate the method of calculating residues by evaluating contour integrals. we obtain I (z2)dz /c (ii) z^z = 27ci[2l] = 27ti (3.714) b t = lim Zif~2\ = 2 1 z_>0 z ( z .b) ' We rewrite the function as —1 1 —vT = 1 z(2 + z . (i) (z2) The function ^ f.260 ADVANCED MATHEMATICS (Hi) Essential singular point In this case. we have to expand f (z) in powers of (z . 1/2 and 1 respectively. b). Use the residue theorem to evaluate the following integrals where Cj. b. 0] + Res[f(z). including negative powers. 1]} (3. the residue at z = 1 is given by b = lim z»i (Z~1)(Z"2) (3. both are z(zl) inside Cj.b) z(zl) Using Equations (3.720) The residue at the origin is obtained by using Equation (3.724) .73.z 2 ) z(2 + z ) ( l .722a.has two simple poles.b) = 1 (3.l ) Similarly.721a.). 22a. Example 3.z ) (3. 21a.723a. one at the origin and the other at z = 1. and obtain bj.720.z.
b) I (iii) / 5dz ^ = 27ci fJ] = 5 ™ L2J (3.b) ic 3 Example 3.+ ^ .z ) 2 The integral is (3. Evaluate the following r\ I I —e ^ (1) .!.725a.COMPLEX VARIABLES 2&L The function has singularities at the origin. 2z dz.£(£)• (3 .z 2 ) The function z e 2/z has an essential singularity at the origin. . we find b2 = 2 I ze 2 / z dz = 27ci(2) = 4ni (3. .i (3.27a » (3.+ . of which only the origin is interior to C2. JcCOSh7lZ r\ I I 7tcot(7tz) (11) — .7 .b) .730a. The residue at the origin is given by bi = lim — — \ r = I1 2 ^ 0 z(2 + z ) ( l .726a. z 3z 2 From Equation (3. (i) The zeros of cosh rcz which lie interior to C are z1=i.729a.727c) (3.b) ic 2 z(2 + z . z2 = .727c).74.728) (3.727b) = z+2 + ^ .dz Jc Z2 where C is a unit circle with center at the origin. at z = 2 and at z = 1. We expand the exponential function and obtain "*•*.
737a.262 ADVANCED MATHEMATICS Using Equation (3. The function 7tcot(7tz)/ z 2 has a pole of order three at the origin.sin TTZI z^° (3735c) [ sin3 TCZ To evaluate the limit in Equation (3.b.b) Example 3.Hm A .b) lim [ ^ cos TCZ .716c).c) J c cosh7tz (ii) We write the function as rccotfoz) z2 [ 7i n J L J ' _ n cos (TCZ) z 2 sin(7iz) 7 3 The only zero of sin TCZ which is inside C is the origin. the residue at i / 2 and . = — r .b.c) (3.b.i / 2 are respectively given by p2z i z=i/2 : i b. The residue bj can be determined using Equation (3.733a.736a.c) z=_i/2 dz = 2.735c).r 1 n sinh TCZ e 2z bj = — ~ 1 n sinh TCZ The integral is given by e.731a. Blasius (1908) has shown that if F x and F y are the x and y components of F . = + ^— rc [i sin (—7C/2)] T C (3.735a. The force F per unit length exerted on a cylinder of infinite length in a steady irrotational flow can be determined by complex variable methods.732a.75. we use l'Hopital's rule and we obtain bj = TC2 lim [ .— + 1^ = 2[e i e~ i l = 4i sin 1 (3.719b) which gives b .ui .^ z s i ^ z z>0 j = 3 rt2Lll (3. [Z3TCCOS71Z1 = 7 L i i m ^ i Tzcosjizl ^ o d z 2 [ 2 ! Z2 s i n TCZJ 2 z ^ 0 d z 2 L sin TCZ J = TC2 (3.b) _3TC sin2TCz COSTCZJ The integral is given by [ ™f& dz = 2TCi[^] = *f Jc z (3. Z = —TTf—^ = — n (I sin TC/2) T C pi i e" 1 = e .
741) where F (cos 9. sin 9) is a rational function of cos 9 and sin 9.742) .739) and using Equation (3. and C is the curve defining the surface of the cylinder. Differentiating Equation (3. sin e) d6 o (3.V o o (z + a 2 /z) (3.740a) (3. we write z = eie The trigonometric functions cos 0 and sin 9 can be written as (3. We consider a few examples. we have shown that Equation (3. Calculate the force if O = .739) In Example 3. This surprising result is known as d'Alembert's paradox and is discussed at length by Batchelor (1967).740b) The residue is zero and this means that there is no force acting on the cylinder.68.iF y = T[ Jc tefdz ( 3 7"38) where p is the density of the fluid. we obtain F X~ ' ^= ^ l V ~ I' " ^+ ^ ^ v2 = i p ^ 2 7 U i SR.739) represents the complex potential for a flow past a circular cylinder of radius a with center at the origin. Triginometric Integrals To evaluate integrals of the form J f27t F (cos 6.738).COMPLEX VARIABLES 263 F *. • The integral of real variables can be evaluated using the residue theorem. O is the complex potential.es (3.
^ Jc b z 2 + 2az + b (3.b) Substituting Equations (3.2M ADVANCED MATHEMATICS cosG = 1 (z + l ) .747a. I z 2 I < 1 and the only pole inside C is z 2 . we obtain 1 = I f(z)dz = 27ciERes(f(z). = _ti^Z Since a > b > 0.746a. 43a) into Equation (3. zn) Jc where C is the unit circle.748a) z.745a.76.742) the limits of the integration 0 to 2n correspond to the unit circle and the integral is taken in the anticlockwise direction. We evaluate the residue bj using Equation (3.b) (3. 43a.742.7.b) a + bcose Va 2 b 2 Substituting Equations (3. Show that (3.749a.741).745a).1 ) (3.714) and we have bj = lim Z ^ Z2 (Z~Z2) T (3.b) Jc a + l l z + z"1) The poles of the function are the zeros of f(z) = bz 2 + 2az + b = 0 They are (3.48b) = l (3. b) into Equation (3.b) b(zz 1 )(zz 2 ) 2^1 a2~b2 .744a. sine = ±{ ( z .b) J l»2rc ^ 0 = 2n a>b>0 (3. zn are the poles of f (z) inside the unit circle. Note that from Equation (3.743a. we obtain I= f "'f*' = 2i f .742. Example 3.
752a) (3. but the limits on the right side of Equation (3. we write.COMPLEX VARIABLES 265 Combining Equations (3.°°. °°) is defined by r Jo f(x)dx = lim a *°° r Jo f(x)dx (3. 49b) results in I = i ^ M = E = TM= 2 Va 2 b 2 Va 2 b 2 Improper Integrals of Rational Functions (3. for example. the improper integral of f over [0.754) . we have f(x)dx = lim I f(x)dx (3.b) We recall from the theory of functions of real variables that if f (x) is a continuous function on 0 < x < oo.75 lb) J— a ^°°A When both limits exist.751a) provided the limit exists.753) Ja It may happen that the limit in Equation (3.753) exists. 0]. the function f(x) defined by f(x) = x The limit (3.744b. Consider. for an integrable function on the whole real line (—°°.75Oa.752b) JO = I Joo f(x)dx + The value of the improper integral is computed as r Jc f(x)dx = lim a ^°° r f(x)dx (3. Similarly if f(x) is continuous on (. °°) r f(x) dx = Joo a f lim ^ / a r f(x) dx + lim f(x) dx a ^°° Jo f(x)dx (3.752a) may not exist. 46b.
we have . f°° Consider the integral I J— oo f (x) dx where (3. The contour associated with the complex integration must therefore also include the real line. This is defined by Equation (3.755) does not exist. From now on. We denote the semicircle by F and we choose R to be large enough so that the semicircle encloses all the poles of f (z).72..) of an integral over the interval (oo. Note that the integral on the left side involves the real line.753) if the limit exist.v. The contour we choose is a semicircle.> ~ 2 (3. The integral is then given by J r°° k [ f(x)dx = 2ni nX Res[f(z).756) This leads us to define the Cauchy principal value (p. Q (x) has no real zeros.R . the principal value of I Joo f (x) dx is implied whenever the integral appears. But lim I xdx = 0 /a r (3. The degree of P (x) is at least two less than that of Q (x). zj JR Jr n= 1 (3.757) (i) (ii) (iii) (iv) f(x) = P ( x ) / Q ( x ) P (x) and Q(x) are polynomials.266 ADVANCED MATHEMATICS T lim a—>°o/0 a2 xdx = lim =a .. z ] = 1 oo (3. k f(x)dx+ f(z)dz = 27ti X Res[f(z). centered at the origin with radius R in the upper half plane as shown in Figure 3..759) . z^ are the poles of f (z) that lie in the upper half plane.758) where z l5 z 2 .. By the residue theorem. c»).
761) Contour integral for evaluating infinite integrals From condition (iv).761) (3. e—>0 (3.760) I f(z)dz = I f(Re ie )iRe ie d6 Jr Jo (3.758). we deduce that for sufficiently large R I z f (z) I < 8 for all points of F.759) reduces to Equation .760). Equation (3. 0 > 1 R (3.758) implies the integral along F to be zero and this will be established next.762b) Jo < en (3. we obtain (3. and theby integral around F (3.762c) Equation is obtained using inequality As R—>(3.72 .759).760) (3.762b) oo. By substituting z = Reie into the second integral on the left side of Equation (3.COMPLEX VARIABLES 267 Equation (3.762a) r < £ d9 (3. is zero. I R FIGURE 3.
Show that . z 2 = ae 3i7t/4 .757) and using Equation (3. z4 = a e W 4 (3.~ x 4 + a4 a>0 (3. we obtain . We denote any one of them by a.x4 + a4 2a3 The function f(x) = — 1 .767ad) (3.769d) into Equation (3.758).769b) (3. J_«.268 ADVANCED MATHEMATICS Example 3. we write z4 = .764) is in the form of Equation (3.769a) z»az4_0c4 = lim ^ ^ 1 z ^ a (za)(z + a)(z 2 +a 2 ) = J4a3 = S4a 4 (3.769c.765) To determine the zeros of z4 + a4.765). we have (3. z 3 = ae 5 i l t / 4 .763) (3.77.766).a 4 The residue at z = a is given by lim ( z .a 4 = a 4 e i 7 t ( 1 + 2 n ) Using De Moivre's theorem.d) Substituting Equation (3.768) ^Z~a^ (3.a ) f ( z ) = lim z > a (3. we find that a4 = . From Equations (3.« d*— = &f. we find that the zeros are zt = ae i7t/4 .766) Of the four zeros only Zj and z 2 are in the upper half plane.
772) reduces to f°° k I Joo e imx f(x)dx = 27ii £ Res[eimzf(z).2 E i [ae i7c/4 + ae 3i7t/4 ] J_oo x 4 + a4 4a 4 = _ _i2L [ c o s & + c o s 3iL + i (Sin 2L + s i n 3TL)1 2a3 L 4 4 V 4 4n = ^JL 2 a3 Evaluating Integrals Using Jordan's Lemma Let f(x) be of the form given by Equation (3. f°° • To evaluate the integral I Joo e i m x f(x)dx. z j n=l (3.757).770c) ir where m is a positive integer.COMPLEX VARIABLES 269 f°° &— = . If F is the semicircle shown in Figure 3.773) into its real and imaginary parts. z j (3. Jordan's lemma states I e i m z f(z)dz > 0 as R > °° (3. Example 3.72. Condition (iv) is replaced by the condition that the degree of P (x) is at least one less than that of Q(x).72 and we have fR e imx f(x)dx + iR f ir k e imz f(z)dz = 27ii £ n=i Res[eimzf(z). we integrate around the contour shown in Figure 3.771) (3. Show that . P(x) and Q(x) satisfy conditions (ii) and (iii).772) Using Jordan's lemma and letting R —> «>. we can evaluate I J—oo f (x) cos mx dx and I J—oo f (x) sin mx dx.770b) (3.770a) (3.78.773) By separating Equation (3. Equation (3.
777b. we obtain I cosxdx J . bi 1 V (Z~ia) e l Z /O 7 T7 \ = ^n 1 ? • \( • \t—^w ^rr zMa (zia)(z + ia)(zib)(z + ib) (3. we have assumed that the poles of f (z) were not on the real axis. we indent the contour shown in Figure 3. Z3 = i b. since we chose Q (x) ^ 0.777a) = —f^TA 2ia(b2a2) Similarly the residue b } at z 3 is b.b2)j / 5 ?_!.776ad) Of the four poles.774) J .72 by making small .779a) (3.(x2 + a2) (x2 + b2) From Equation (3.779b) (a2b2) Lb Poles on the Real Axis a J So far. only zj and z 3 are in the upper half plane.270 ADVANCED MATHEMATICS cosxdx = _n__ e^_e^_ a >b>0 (3. Z4 = i b (3. (3.(x2 + a2) (x2 + b2) = R e l2jci 1 = 7—&— e^ + e^ I _2ia (b2 .778) Substituting Equations (3. J 1 L ( x 2 + a2)(x2 + b2) \ nTi [(z2 + a2)(z2 + b2) J / The four poles of the function are zj = i a.773). The residue b\ at Zj is given by . f y r . If some of the poles are on the real axis.a2) 2ib (a2 . = Re 27ci X Res . = e ^ 2ib(a2b2) (37"77b) (3. 78) into Equation (3.775) (3.775). we have a2b2 b a) I 1 COS^dx „. Z2 = i a.
The contour is indented by making a semicircle of radius e. The contour integral given in Equation (3. zn] n= 1 (3.73 Contour integral with a pole on the real axis It was shown earlier that in the limit as R —> °°. we can write (3. We denote the small semicircle by y. we have z = a + eeie and the contour integral becomes (3. I f (z) dz —> 0. where a is real. Suppose f (z) has a pole at z = a. We now consider the integral around y.759) is now modified as follows J f ( x ) d x +  f(z)dz+ I f(x)dx+( f(z)dz = 2ni X Res[f(z).73.781) I f(z)dz = I f(a + e e i e ) e i e i e d 0 Jy Jn Since f(z) has a simple pole at z = a. On y.780) R Jy A+e Jr R 0 R FIGURE 3.782) .COMPLEX VARIABLES 271 semicircles in the upper half plane to remove the poles from the real axis. centered at z = a in the upper half plane as shown in Figure 3.
272 ADVANCED MATHEMATICS f(z) = g ( z ) / ( z . In the limit. ..74. Example 3.780) becomes J f oo k ! f(x)dx = 2ni 2 Res[f(z). one at z = a denoted by Ji.772).786) If there is more than one simple pole on the real axis and if we denote these poles by aj.782). We illustrate this by considering an example.783) where g(z) is analytic and can be expanded as a Taylor series about the point a. 84) into Equation (3. at a and a. where C is the indented contour to be defined later. 83) that g(a) is the residue of f(z) at a and is denoted by Ra. as R —> °o and e —> 0.79. that is to say. The Jc a z z z function f(z) [ = l / ( a 2 . The function g(z) can be approximated by g(z) = g(a) + 0(e) Substituting Equations (3.z 2 ) ] has two poles on the real axis. a2.z] + 7tiRa oo n=l (3. .785a) (3. the sum of the residues at aj. We indent the contour by making two small semicircles of radius e. The contour C is shown in Figure 3.784) I f(z) dz > i I g(a) de Jy Jn > iicg(a) (3.~ a2x2 TLjsrna^ a (3. . The same modification can be applied to Equation (3.785b) We note from Equations (3. we obtain. and the other at z = a denoted by y2. (3.783. as £ —> 0.. ^ ••• > a i > we replace R a in Equation (3. Equation (3.787) = a is real I We consider the integral I 1 2 eiz 2 dz.714. Show that cosjLdx ) . a^ .786) by the sum of R .a ) (3.
791) . zn r. we write z = a + e e i e (3.2 7 r " —^ l_d —z Using Jordan's lemma.74 Equation (3. r fR I I J_R e lx I e lz / e lx / e lz I e lx — dx+ — dz+ — dx+l — dz+l — dx a2x2 ]yi a 2 . f Contour integral with two poles on the real axis .COMPLEX VARIABLES 273 r R o 0 a R FIGURE 3.772) is modified to /•ae .789) h a2z2 To evaluate the integral around yj.x 2 }y2 a 2 .788) r.z 2 J_ a+e a 2 .i a + i £ e i 9  i £ e i e d0 JYl a 2 z 2 1 e e ie (2aee ie ) (3.2 (3.x 2 / + I /„ iz e'Z O2 k r O2 dz = 2 Tci X Res n= 1 iz ^ .790) f ¥. we have as R —> °° t e iz — dz = 0 (3. ra.dz = [° e x P ( .z 2 Ja+e a 2 .E .
^ .785b).792b. The function e i z / (a 2 . Equation (3. so the right side of Equation (3.791) simplifies to I Jyi a .x z = i7c[R a + R_J (3.d x .796) The residue R a is given by (3.793b) f eiz lim ^rdz =i7rR_ a (3.788) is zero. 93b).a and a. = lim a z _^_ a e~ia = ^ From Equations (3. (az)(a + z) (3.4r i p~la / de j n (3J92a) (3.792b) » " ^ ^ Using Equation (3.z fziz eiz . Equation (3.788) simplifies to . Similarly we have I eiz Urn ^ 0 J Y2 .274 ADVANCED MATHEMATICS In the limit as e —> 0.795) where R a is the residue at z = a.794) ^o jYl a2z2 Equation (3.793a) (3.794) is a special case of Equation (3. we find that the residue R_a at z = . we obtain /Z+f)£lZ.d z a2_z2 =i7tRa (3.797a) . e—> 0.z 2 ) has no other poles except at .a is given by R . a .714)._<*. In the limiting case R—> < * > .^ .
the mapping is one to one. we obtain (3.798b) into its real and imaginary parts. we regard one complex plane to be z [= (x. From Equations (3.798b) = (3.8 _sinjL_ dx = 0 (3.82a. y)] and the other to be w [= (u. b).b) (3. v) plane there corresponds one and only one point in the (x.799a) .797b) . If to each point in the (u.^ d x J_ooa 2 x 2 = i^[eiaeia] 2a (3. The equation of T is written as w(t) = f(z) = u(t) + iv(t) (3._ooa 2 x 2 I fl2 Yx2 y_oo a — 3.36a. we have I _cos_x_ d x ILsin_a a (3.796). We examine the geometric properties of this mapping. we consider the function f defined by Equation (3.35) to be a mapping from one subset of a complex plane to another. The tangent to C at P is given by . y) plane and vice versa.798a) = J (sin a) Separating Equation (3. The equation of a curve C in the (x. v) plane. we have defined a curve in the (x.3. v) plane. y) plane by introducing a parameter t.COMPLEX VARIABLES 275 eia = ~^ Substituting the values of R a and R_a into Equation (3. In Section 3. v)] and f establishes the correspondence between the (x.799b) CONFORMAL MAPPING In this section.81) Suppose the curve C passes through a point P at which t takes the value tQ. y) plane is written as z(t) = x(t) + iy(t) The curve C is transformed to a curve T in the (u. y) plane and the (u.
83) t lo The tangent to the curve T in the (u. Using the CauchyRiemann conditions [Equations (3.276 ADVANCED MATHEMATICS d* at t lo = <£ dt t lo +i % at (3.85) 3v dy dy J L dt  Equation (3.87d) ax ^ dz The condition that the determinant be nonzero is equivalent to the condition that — be nonzero.85) expresses the transformation of tangent vectors from the zplane to the wplane. it ensures that nonzero tangent vectors in one plane are transformed to nonzero tangent vectors in the other plane. as at at "dul dt dv.84) £o to [o Using the chain rule "^ and ^ can be written. v) plane at the point [u (tQ). v (t 0 )] is (3.87c) (3. If it is nonzero.dt J r^L 3x 3v \_dx ^L 1 [ dx 1 9y dt (3. The condition for a unique solution is that the determinant du dx J = 9v dx 3v 9y du dy # 0 (3. in matrix form.87b) = ^ = + i  ^ ax 2 (3. dz . b)]. .329a. J can be written as a_u a_v J a u a_v ~ ax a y _ /3u\ 2 ay ax /av) 2 (38"7a) (3.86) The determinant J is the Jacobian.
Further discussions are given in Chapter 4. the mapping is conformal. but the sense is not the same. y0). The Jacobian J is defined as . in a conformal mapping. we briefly review the case of two variables. v 0 ). U FIGURE 3. The curves Fj and F 2 intersect at (u0 . both the magnitude and sense of angles are preserved. measured from Fj and F 2 . dz y A VJ uo. Suppose Cj and C 2 are mapped by f to Fj and F 2 respectively in the wplane. Let Uj and u 2 be two differentiable functions of two variables xj and x2. as shown in Figure 3. the mapping is isogonal. (c) 4^ i s nonzero. y 0 ). f is single valued.COMPLEX VARIABLES 2ZZ The curves Cj and C 2 in the zplane intersect at (x 0 .yo) ^ V La r.81.81 Conformal mapping The Jacobian J introduced earlier plays an important role in the theory of transformations. Here. measured from Cj to C 2 as illustrated in Figure 3. Let A9 be the angle between the tangent vectors to Cj and C 2 at (x 0 . v0) and let Aa be the angle between the tangent vectors to Fi and F 2 at (u 0 . The conditions that f represents a conformal mapping are (a) (b) f is analytic. If the magnitude of A0 is equal to the magnitude Aa. 1 ^ X 1 +.81. If A0 is equal to Aa. Thus.
we obtain a Uj dxl _ 1 9u2 dUj J dx 2 9x l _ du 2 1 9u l J dx2 n o 1OaM pl = .1 . b2) are linearly dependent if the determinant . For the inverse function to exist. By differentiating Equations (3. That is to say. that is to say.Lpl.u2) = 0 (3.810cd) We can also write 1/J as 3xj 3uj y = 3x 2 dUj 3x 1 5u 2 (3.b) we require that the Jacobian to be nonzero.x 2 ).812).b) du2 3xj du2 3x2 J is also known as the functional determinant.89a. The points at which J is zero are singular points. This is analogous to the case in linear algebra where two vectors a (a1? a2) and b (bj. duj J dxj pi du2 = L*± J dxj (3.89a.278 ADVANCED MATHEMATICS J = 3(u. uj and u 2 are not independent.811) 3x 2 9u 2 If the Jacobian vanishes. u 2 = u 2 (x 1 ( x 2 ) (3. b) partially with respect to Xj and x 2 . and solving for the dxpartial derivatives v. Xj and x 2 do not occur explicitly. u2) d(x x d(x 1? x 2 j 3xj = 3x2 (3.88a. for the possibility to write u : = U! (x!. there exists a relationship between uj and u 2 and f(Ul.812) In Equation (3.
818) (3. v) in the wplane is mapped to a point (x + Bj. ^2 = 0 bl b2 (3. .813) If Xi and x 2 are also functions of yj and y2. the double integral of any function (j) over R is given by I I (J)(x 1 .82. B2) as shown in Figure 3.U2) = 5 d(yi>Y2) K U 2 ) a(xl. x 2 (u!.y + B2) (3.817) into its real and imaginary parts. Equation (3. If we take A to be unity.817) (3. Separating Equation (3. we have a(Ul. we obtain (u.u 2 )]  J duj du 2 R R (3.COMPLEX VARIABLES al 279. The Jacobian J gives the magnification of the area due to the transformation. u 2 ) plane.815) If < ] ) is unity. y) by (Bj. We now consider some transformations in the complex plane.v) = (x + B 1 . Linear Transformation Consider the transformation w = Az + B where A and B are complex constants.816) Equation (3.x 2 )dx 1 dx 2 = I I ^ [ x j f u j . we obtain the area of the closed region. This corresponds to a displacement of (x.818) shows that a point (u.816) becomes u + i v = (x + i y) + (Bj + i B2) where B is written as B j + i B 2 .x2) 3 ( X 1' X 2) ^(yi'Y2) 8 j If a closed region R in the (xl5 x2) plane is mapped into a closed region R' in the (uj. ^ ) . y + B 2 ) in the zplane. B^ and B 2 are real constants.
82 Translation If we take B to be zero and we write w. . 1 6 ~ I 0 1 8 1 » • I 2 I 4 I 6 I 8 ^ FIGURE 3.b) (3. Equation (3.c) Equation (3.819a. A and z in their polar form as w = pe'*.821a) represents a contraction (a < 1) or an expansion (a > 1) of the radius vector r by the factor a. Equation (3. The shape of all the figures is preserved.821a. Combining these two cases. z = reie (3.816) represents a displacement.816) becomes p e j * = are**0""9* We deduce p = ar. < >  = a +9 (3. and a rotation.280 ADVANCED MATHEMATICS V I v " w 8  ' 8  z 2 1 0 2 4 i. we deduce that the linear transformation given by Equation (3.83.b. a magnification. This transformation is illustrated in Figure 3.821b) is a rotation through an angle a.820) A = aeia.
1) and (1. . 2) in the zplane are mapped into ( 1 . .822ad) Determine the region R' of the wplane into which R is mapped under the transformation w = (l + i)z + ( l .2 i ) Equation (3. v = x+y2 (3. x=l. X It—1—1 O ». (2.825a. 2) and (0.2) It follows that u = xy+l. x=l. (0.1).2 ) .y + 1) + i (x + y .824b) (3.826ad) in the zplane are mapped into . b). 0).b) (3.824a) (3. y = 2. The lines y = 0.81.825a.0) respectively in the wplane. (1. 0).COMPLEX VARIABLES 281 y4 z v> w k:—1 0 ». U FIGURE 3.823) From Equation (3. x=0 (3. A rectangular region R in the zplane is bounded by x=0.83 Rotation and magnification Example 3. y=2 (3. y=0. we find that the points (0.2i) = (x . (1.823) can be written as u + i v = (1 + i) (x + i y) + (1 .
84. as shown in Figure 3.v = 3.816. the distance between two points has been magnified by the factor V2" and all the lines have undergone a rotation of n/4. Reciprocal Transformation Consider the transformation w = 1 (3. 23).828a. u + v=l.828c) . uv =. .B y\ c1/ R \ ^ \ R1 >v ^ 0 A X Nv / VY/4 o1 FIGURE 3.829) (3. We identify A and B to be A = (1+i) = n e i7tM B = (l2i) The transformation is as described earlier. v i * B' c .827ad) respectively in the wplane. We note that all the points in the zplane are displaced by (1.2) in the wplane.b) (3. u+v=l (3.282 ADVANCED MATHEMATICS u .84 Mapping of a rectangle in the zplane to another rectangle in the wplane The transformation can also be deduced by comparing Equations (3. The rectangle R in the zplane is mapped into rectangle R' in the wplane.l .
the inverse mapping is also a reciprocal transformation.830) From Equation (3.COMPLEX VARIABLES 283.85. we obtain the identity transformation. The equation .85 Reciprocal transformation On reflecting about the real axis. r FIGURE 3.832) On applying the transformation twice in succession.b) y z. That is to say.829). is given by z.e" i e (3. We recall that if z 0 is a point in the complex plane.57. we obtain z = 1 (3.= ± e i e zo r (3.831a. in this case. / 1%/ \ I Xe \ r ^7 1 z. This is illustrated in Figure 3. The reciprocal transformation maps circles or straight lines into straight lines and circles. The process of inversion was introduced in Example 3. its inverse point Zj. with respect to the unit circle. from Equation (3. we deduce that the reciprocal transformation consists of an inversion with respect to the unit circle and a reflection about the real axis. z\ is transformed to w. Writing z in its polar form.830). = J . We note that.829) becomes w = 1. Equation (3.
c ± 0) is mapped into a circle passing through the origin in the wplane.834.833) can also be written as azz + bz + bz + c = 0 where a and c are real constants and b is a complex constant. b. But the points inside the circle in the zplane are mapped to points outside the circle in the wplane. the circle not passing through the origin in the zplane (a ^ 0. (3. d are complex constants and (ad . the straight line passing through the origin in the zplane (a = c = 0) is mapped into a straight line passing through the origin in the wplane.833) (3. Bilinear Transformation The bilinear (Mobius) transformation is given by w = ^±4 cz + d where a. we deduce the following (i) the straight line not passing through the origin in the zplane (a = 0. Equation (3.284 ADVANCED MATHEMATICS a (x2+ y 2 ) + bxx + b2y + c = 0 represents a circle if a is nonzero and a straight line if a is zero. center at the origin in the wplane.be) is nonzero. c = 0) is mapped into a straight line not passing through the origin in the wplane.836) can be written as (3. 35).829) into Equation (3. Substituting Equation (3. center at the origin in the zplane. c. (3. the circle passing through the origin in the zplane (a ^ 0.835) (ii) (iii) (iv) Note that the unit circle.836) . we obtain cww + b w + bw + a = 0 From Equations (3. Equation (3. if a is nonzero and c is zero. is mapped into a unit circle.834). c * 0) is mapped into a circle not passing through the origin in the wplane. we have a circle passing through the origin. we have a straight line.834) If a is zero.
z 4 ) ( z 3 . z2. .a which is again a bilinear transformation. Z3) in the zplane are known to map to three points in the wplane. a rotation. Let zj. we obtain via Equation (3. Z2. a magnification.1 w* w* = cz + d (3.w2) (w3 .836) reduces to Equation (3. and an inversion.z2) (z3 .838) (3. we can obtain the inverse transformation z =+^ ^ cw . Z3 and Z4 be any four points in the zplane and wj. Simplifying Equation (3.837a) v ' =  + (^MZ* z* = . W3 and W4 be the corresponding points in the wplane.a z .841) a unique relationship between w (= W4) and z (=z 4 ). If c is zero. W2.838) is bilinear in z and w (linear in both z andw).836) yields czw + d w . Equation (3.COMPLEX VARIABLES 285. From Equation (3.8_37b) (3.840b).840a) b) (3.837C) (3.w4) (zj . w = a + i^M c c(cz + d) (3. If three points (zj.838).839) .z 2 ) (wj .837d) The bilinear transformation is a combination of a translation.w_ = — E — s—j 1 s cz r + d cz s + d = (adbc)(ZrZ) (cz r + d) (czs + d) From Equation (3.b = 0 Equation (3. we have az_ + b az_+ b w. it follows that (3 841) K ~ w 4 ) (W3 ~ W2) = ( z i .841) is the cross ratio of the four points and the cross ratio is an invariant under a bilinear transformation. Using Equation (3.836).816).z4) The ratio in Equation (3.„ „ arK x (3.
we obtain w = eie(zzn) .83.847) (3.i + iz) (53i)z(l+i) P84JJ Example 3. v _ .i.845a.z 0 ) (3.b/a by z 0 and . with the condition that Im (zQ) is positive.i) in the zplane into (0.b) . 45a.0> (zz0) The real axis of the zplane is mapped into w = e ' 6 (X I ^ (x . The bilinear transformation can be written as (z .b/a corresponds to w = 0. . We denote . we obtain 2 ( l .82. 1. we have Iwl = loci = 1 Combining Equations (3. the point z = . The point zQ is then mapped to the origin which is inside the unit circle.844). and the lower plane (z 0 ) is mapped into the outside of the unit circle.d/c by z 0 . From Equation (3. 2 .844) (z~zo) The origin in the zplane is mapped into a point on the unit circle in the wplane.zn) w = a )—=&• (3. Find the bilinear transformation that maps the upper half plane Im (z) > 0 into the unit circle I w I < 1. i) in the wplane. we deduce (i) (ii) the point z = . Using Equation (3. we obtain hzHln!) _ (l)(iw) (i + iz)(2) (l+2i)(2iz) ('*42J On simplifying Equation (3. From Equation (3. Determine the bilinear transformation that maps the points (1 + i.846) (3. b).286 ADVANCED MATHEMATICS Example 3.842).841).d/c corresponds to w = °o.844.836).
The SchwarzChristoffel transformation is the required transformation and is given by iw.86.847). Determine the complex The motion of the ideal fluid is due to a source of strength m at potential <&.849) may be considered to be a constant which can be absorbed in K.°o.848) Thus the real axis of the zplane is mapped into the unit circle in the wplane. (3.851) .36. which would allow the determination of the velocity and of the streamlines. its interior angle is zero and we may take the corresponding point a in the wplane to be infinity. Equation (3. and we wish to map its boundary into the real axis (v = 0) of the wplane. 0) respectively in the wplane. We map the vertices A. If one vertex of the polygon is at infinity.. C.= K ( w . and (<*>... 0).ES 287 From Equation (3. B. (1. and D to the points (.l ) 1 / 2 (3. SchwarzChristoffel Transformation Suppose we have a polygon with interior angles o^. The semirectangular channel ABCD is shown in Figure 3. (X2 .a 1 ) 1 . In this example. as shown in Example 3.a 2 ) 1 . An ideal fluid is bounded by a semiinfinite rectangle given by x>0. l>y>0 (3.COMPLEX VARIARI. It has already been shown that a point on the upper zplane is mapped into a point inside the unit circle in the wplane.84. (1.0)..849) becomes &£.c t 2 / 7 C . we obtain _ eie(xz0) e~ie(xz0) 1 (3.a^) in Equation (3.a i / 7 t ( w . a2 are the real values of w (= u) corresponding to the vertices of the polygon. Equation (3.846) is the required transformation.= K ( w + l ) 1 / 2 ( w .850) . Example 3.849) where K is a constant and a1. 0). The factor (w .
K where Kj is a constant.86 Mapping a rectangle into the upper half plane Integrating Equation (3. Inverting Equation (3. d w K J v^n (3.852a) = 1.cosh"1 w + K.855a. we obtain Ki=0.1 = cosh(iKKK!) 1 = cosh(KK!) From Equations (3.851).b) .854b) (3. b).H .852b) yields w = cosh(KzKK!) The mapping of the point B (=i) to B' (=1) and C (=0) toC' (=1) implies .854a) (3.288 ADVANCED MATHEMATICS B S • C A A' g. we have z = . B1 c' » I » 0 D' ^ u D X FIGURE 3.852b) (3.854a. K=T C (3.853) (3.
we have w = ae i 9 (3. ) .860) +— a.m i n  w + i sinh%\ = . we find that the singular points are z = ±l Consider a circle of radius a with center at the origin in the zplane. (3.m i n (w 2 + sinh 2 1) = .855a.m i n [w .m i n (cosh2rcz + sinh2 £•) ' (3.857b) (3. as shown in Figure 3.857a) (3.861) . p.858a) (3.i s i n h 2 .858c) Joukowski Transformation The Joukowski transformation plays an important role in aerodynamics. we have w = cosh (TCZ) The source in the zplane at ^——.857c) = i sinh & in the wplane.859). b) into Equation (3.859) is zero.858b) (3.853). It transforms an aerofoil into a circle.86.856) w = cosh[(l +i)l = cosh^ c o s h ^ + sinh^ s i n h ^ = isinh The source at S = ^ ' in the zplane is mapped into a source at S' (3. 1965. 210] 0> = . In the wplane. The complex potential < E > due to a source of strength m at i sinh ~ and a wall along the real axis is [MilneThomson. The transformation can be written as w = z+ 1 The singular points of the transformation are the points at which ^(3. From Equation (3.is mapped into a source at (3.COMPLEX VARIABLES 289 Substituting Equations (3.
which is equivalent to saying that the circle does not pass through the singular points.87 Mapping of a unit circle into a segment of the real line Example 3.862a. v=0 (3. b) become u = 2cos6. The center of Cj and C 3 is at the .864a.87. A circle passing through one singular point and enclosing the other singular point in the zplane is mapped into an aerofoil in the wplane. r TT 71 "x 2 2 *u FIGURE 3. we obtain u = 'a +1 ' cos9.290 ADVANCED MATHEMATICS Separating Equation (3.b) The circle passing through both singular points in the zplane is mapped into a segment of the real axis in the wplane. b) yields j^r +j ^ = 1 (a2+l) (a2l) (3. If the radius of the circle is one (a = 1). Equations (3.863) A circle in the zplane is mapped into an ellipse in the wplane.862a. as shown in Figure 3.859)]. and C3 in the zplane into the wplane under the Joukowski transformation [Equation (3.85. as long as the radius of the circle is not one. v = l l J l i l sine a (3.862a.b) a Eliminating B from Equations (3. Discuss the transformation of the circles Cj.861) into its real and imaginary parts. C2.
865a. It also intersects C^ at the singular point A and C 3 at the point B [=(2. can be written as u2 v2 1 h+ V =1 The equation for C2 is given by z = \ + eie (3866) (3. an ellipse [Equation (3. C2.0)]. b)] and can be written as 2 < u < 2. in this example.b) The circle C3 is mapped into F 3 . v=0 (3.88 Mapping of the circles Cj. The circle C2 passes through one singular point (z = 1) and encloses the other singular point (z = 1). The circles C\. y i v i. 0) and its radius is 3/2.88.COMPLEX VARIABLES 297 origin and their radii are one and two respectively. FIGURE 3. and C 3 from the zplane to the wplane The unit circle Cj is mapped into Fj. The center of Ci is on the real axis at (112. and C3 are shown in Figure 3.864a.867) . a segment of the real axis in the wplane [Equation (3. C 2 .863)] and.
869a) (3. 0 < Q < J Strip 0 < y < n half plane v > 0 w = w = ez ^~A upper half plane v > 0 . The points A and B in the zplane are mapped into the points A1 and B' in the wplane.[(1 +3cosej + 3isinej + —^ '2 w= (3.859) yields 1 I7i o a\ T • ft I 2 [ ( l + 3 c o s e ) . b).88. From Equations (3.3 i sine] J. The mapping of the three circles from the zplane to the wplane is shown in Figure 3.867) into Equation (3. 0<X<TC wplane unit circle I w I < 1 unit circle I w I < 1 Transformation j 2 w = i+z iz w = 1 + i e— lie1Z Region outside the ellipse unit circle I w I < 1 z _ 1 [/a _ ^\ w 2 + ( a + b)j * i + y— = 1 a2 b2 u PP e r Sector I z I < 1. • In Table 3. A more extensive table is given in Kober (1952). we list some useful transformations which map a domain in the zplane conformally to a domain in the wplane.868) 10+ 6 cos6 (3.81.869a.81 Some useful transformations zplane Upper half plane y > 0 Infinite strip of finite width _ o o < y <oo.292 ADVANCED MATHEMATICS Substituting Equation (3. TABLE 3. it is seen that F 2 is symmetrical about the real axis. we obtain u = (l + 3 cose) ( 1 + 1 ) 5 + 3cose/ V2 v = 3 sine ( 1 \2 1 ) 5 + 3 cos6/ The circle C2 in the zplane is transformed into an aerofoil F 2 in the wplane.869b) Separating w into its real and imaginary parts.
57. C O . and s are real. If we need to solve Laplace's equation subject to a boundary condition in a complicated domain D. Determine all three values of z 1/3 and plot them in the Argand diagram. we can map the complicated region into a simpler region. there exists a transformation that can map it into a unit disk (I z I < 1). Unfortunately it is not shown how the transformation can be obtained. for any simply connected domain which is not the whole complex plane.87Oa. Riemann has shown that. Write z = 4 Y 2 + 4 V 2 i in polar form. we can transform the domain to a simpler one D1 and solve the problem in D'. Using a conformal transformation to the wplane. If the geometry of the region of interest is complicated. such as the unit circle. v)]} is also harmonic satisfying the same boundary conditions [Equations (3. b)] at the transformed boundary. we have mentioned the importance of Laplace's equation and have shown that an analytic solution is a solution of Laplace's equation. the solution of Laplace's equation is known for any domain. PROBLEMS la. TI* = I f(s)e. We can then invert back to D.3. we have shown that if the value of a harmonic function is given on a circle of radius R. the transformed function K (u. In Example 3. v) {= § [x (u. In Section 3. In all physical problems.870a. By using a conformal transformation. We have applied this technique in Example 3. The complex viscosity T] of a linear viscoelastic liquid is given by 2a. since we can map the domain into a unit circle. In theory. its value at any other point is known. y (u. . the problem of imposing the boundary conditions can be demanding. Let <) be a harmonic function satisfying conditions at the boundary. the solution has to satisfy certain boundary conditions. v).84.i ( O s ds Jo where f.b) d<b where §0 is a real constant and — is the derivative of § with respect to the normal to the boundary curve. The main problem is to determine the suitable transformation from D to D'.COMPLEX VARIABLES 293.= 0 dn (3. such as (j) = <>0 or dd> p.
what are the stream lines in this case? Do they represent the flow due to a source at the origin? Answer: arctan (y / x) 7b. If f (z) is an analytic function. If f (z) is the complex potential O. Show that the equation 6b. Obtain the harmonic conjugate v (x.326). The stream lines are given by \}/ = constant. y) = inVx 2 + y2 is a harmonic function.294 ADVANCED MATHEMATICS If we write n* = Ti'iri" f(s) = Ke~s/X where K and X are constants. identify the potential < j ) and the stream function \/. u(x. calculate r' and r". Discuss the continuity and differentiability of the following functions (i) (ii) 5a. use the CauchyRiemann conditions to obtain two other expressions for f'(z 0 ). Determine and plot the curve given by z. f(z) = z 2 f(z) = 1 (iii) (iv) f(z) = z 2 +  f(z)=z l+X2®2 Starting form the definition of f'(z 0 ) given by Equation (3.1 4a. Answer: ^K ^ (oKX2 l+X2(a2 3b. show that —7 + — r If (z) I = 4 f (z) 3x z dyz . The quantities r\ and r\" are the dynamic viscosity and dynamic rigidity respectively. y) and the function f (z). one in terms of u only and the other in terms of v only.
dz Calculate lim z>0 sinz z along the real and imaginary axes. If C is the closed curve (Cj . lib. This is shown in Figure 3. c what condition(s) in the Cauchy's theorem is (are) not satisfied? y " I t x FIGURE 3. Is the origin a singular point? 9a. 10a. show that "4 = 0. z ) is an analytic function. Cj is the upper semicircle of unit radius.1 . Evaluate I zdz along the semicircles Cj and C2. centered at the origin traversed in the clockwise direction from ( . If f ( z . 0) to (1.COMPLEX VARIABLES 295 8a. Determine the real and imaginary parts of f(z) = ^ Li Are the CauchyRiemann conditions satisfied? Compute f'(z).Pllb Integration around semicircles . what is the value of I z dz ? Is the integral zero? If not. 0) and C 2 is the image of Cj about the real axis.C2).Pllb.
What is the value of T if the radius of C is changed from 1 to 3/2 ? 14a.J — = l + z + .296 ADVANCED MATHEMATICS 12a. . for which of the following < E > is T zero? (i) (ii) O = z O = i log z (iii) * = ^ If F is not zero. z 0 is not enclosed by C z 0 is enclosed by C Answer: 0 Answer: 27iicosz 0 The circulation T round a closed curve C in a twodimensional flow is given by r = Re I ^ dz I dz C where O is the complex potential.. . 1 z z<l \62ni . If C is the unit circle with center at the origin.. . evaluate T. Evaluate the integral I —sin z AZ ^ w here C is a simple closed curve for the following 1 (zz 0 ) 2 C cases (i) (ii) 13b. + z n + . Evaluate the integral I (cos 2 z + 2cosh 4 z) 1 C l z where C is any simple closed curve that encloses the origin. Show that the expansion of —^— about the origin is . Answer: 15b.
Locate the singularities of the following functions and discuss their nature (i) (ii) e1/z ez/z3 (iii) (iv) zsin(l/z) (cos z)/(z7c) 2 22a. Find the Laurent series of (zl)(z3) (i) (ii) in powers of z. Locate the poles of the following functions and specify their order rs z(z2) ( z + i r ( z 2 + 4) . in powers of ( z . are there points at which the series about the point (0. 17a.. Determine the radius of convergence of the new series. ez cos z z 21a. Expand Ln ( \ for  z  > 1. Sketch the circles of convergence of the two series. z 19a. Find the Taylor series of sin z about the point Z = K/2. ... What is its radius of 18a. Is the origin an isolated singular point? To describe the behavior of a function f(z) at infinity. Do they coincide? If not. e 2z Obtain the Laurent series of (z1)3 convergence? about the point z = 1. we make the transformation £ = l / z and the point at infinity in the zplane is mapped to the origin in the ^plane.Expand the same function about the point (0. The function cosec (1 /z) has singularities at the origin and at points on the real axis given by z = 1 /nn . 1/2) is convergent while the series about the origin is not convergent? This process of extending the region of convergence is called the analytic continuation of the function. Discuss the behavior of the following functions at infinity (i) (ii) ez z1/2 (iii) (iv) zsin(l/z) z/[(zl)(z3)] 23b. 16a. Note that this point lies inside the circle of convergence and is not on the line joining the origin to the singular point.1 ) . 20a. 1/2).
/2 I ( Z z — 1) c (ii) I — — d z . 2 n p v^ k 27a. by the method of contour integration. Calculate the components F x and F y of the force acting on C.298 ADVANCED MATHEMATICS 24a.2 ) ] Answer: (i) .1 / 2 (iii) 1 (iv) 1 (v) . k. The complex potential < J > of a twodimensional motion of a fluid is given by < E > = Voo (z + a 2 /z) + i k i n (z/a) where v ^ . C is any closed contour in the upper plane ) l2e"iz Answer: 0 c 26b. C is the circle  z  = 2 Answer: . that (i) d9 j ^ = Jo 3 . C is the unit circle centered at the origin Answer: in. Answer: 0 . Determine the velocity components and show that there are two stagnation points (v = 0 ) on C. Find the residues at the poles of the following functions (i) (ii) (iii) l/(lz) l/(lz2) cos z / sin z (iv) (v) (vi) ez/sinz cosz/z3 1 / [z 2 ( z .1 / 2 .1 . Evaluate the following integrals by means of the Cauchy residue theorem (i) I z d z .1 ) ( z . . Show. . the circle of radius a centered at the origin if v^ > k/(2a). and a are constants (real). 1/4 25a.1 (ii) .1 / 2 (vi) 3/4.3 5 i n / 6 J z(zl) 2 (z3) c (iii) I dz .cos 0 Vz . Verify that the stream function is zero on C.
using a semicircle in the upper halfplane. Use Jordan's lemma to calculate f e ictx dx J_oc (x2 + a2) Deduce f OO A OO fi) ( cos a x dx I 22 yoo x +a /jj\ sin a dx 122 y_oo x +a Answer: (i) ne~aa/a.cos 6 =7I(4 _2^") (iii) I Jo cos3ede l2pcos9 + p2 = ^(ip 1  + p2)^ Q<p<1 P 28a. and c are not satisfied? 30a. and c is the result valid? What modifications need to be made to the contour if the conditions on a. Answer: (i)7cVT72.COMPLEX VARIABLES 299 (ii) sm 2 9d9 Jo 2 . 29b. b. Show that I dx _ 2n V74acb 2 ) J_oo ax 2 + bx + c For what values of a. b. the following integrals a) r ^ x J—l+x4 (ii) r—x^dx— J_co (x 2 + 4) (x 2 + 9) (iii) r x2dx (iii) n/2 JooCl+x 2 ) 2 (ii) n/5 . Evaluate. (ii) 0 I .
.
7. i. y. V . a and q are vectors. j and k are unit vectors along the x. A vector is also defined as a tensor of order one. It is a tensor of order zero and temperature is an example of a scalar.. T_Y. TABLE 4.. . is represented by nine numbers (components).A. The extra stress tensor in fluid mechanics is an example of a second order tensor and can be denoted by T. z axes respectively. T__. Third and fourth order tensors will also be introduced.. ^y Aj£ J x. we briefly review some of the known and useful results of vector algebra and of vector calculus..23) (4.. velocity is a vector and.21. in a threedimensional space. or time t. when referred to rectangular Cartesian axes..22) (4. In Table 4. we will be dealing mainly with vectors and second order tensors. x v .. . x.2 VECTORS In this section. A..CHAPTER 4 VECTOR AND TENSOR ANALYSIS 4. v.24) (4.25) (4.26) .21) (4. y. A vector is defined as a quantity that has both magnitude and direction. it is specified by its three components (v x .21 Results of vector analysis a+v=v+a v + (a + q) = (v + a) + q c1v = vc1 (Cj + c 2 ) Y = c 1 v + c 2 v a « v = v « a (dot or scalar product) a • v =  a   v  cos 0 (0 is the angle between a and v) (4. and % „ . A scalar has only magnitude and is completely characterized by one number. For example. v . it is described by three numbers (components). Cj and c 2 are scalars and cp and \)/ are differentiable scalar functions of position x. z... In a threedimensional space. A second order tensor. v z ). v y . x. 4. t v .1 INTRODUCTION In this chapter.
216)  .221) (4.302 Y # ( a +q) =v » a +y » q Cj (v • a) = (cj v) • a = v • (Cj a) ADVANCED MATHEMATICS (4.+ j§2.b) x y z Y«(axq)= ax a y az qx qy qz (4.223a.222) i V = i A + j  _+ j 3 dz dx .b) a x v = .214a.212) C j (vxa) = (cj v)xa = vx(cja) Y'(axq) = a»(qxv)=q«(vxa) V V V (4.213a.oy Vcp = i ^ .+ k ^ .220) (4.210) a x v = ax a y a z V V V (4.219) A(vxil = v x  t  x a A ( r i x q ) =v .28a.211) x y z v x ( a + q) = v x a + v x q (4. a i(v.=grad(p dz dx .27) (4.217) (4.dy (4.b) . n is a unit vector orthogonal to v and a i j k (4.v x a (cross or vector product) a x v =  a   v  (sin 0) n.  (4.29) (4.a) = v .b) (4.a x q dt dt dt .a x ^ + v 4 x q +^ .215) ife+a) =f +  + (4.218) (4.dt [vx(axq)] = v x ( a x 4 )+ v x (  xq) +  x ( a x q ) (4.
+vz3x y dy z dz (4.227) (4.230a.Y = ^ .2 _ is the Laplacian operator (4.+ ^ dx dy dz On the other hand v • Y is a scalar operator and is given by YY = v x x (4.(v • Y) a + v (Y • a) Y • (Y 9 ) = V2cp = ^ > dx2 7)2 ?2 rl2 + (4.232) ~ + v v ^ .225a.229) (4.v • Y. If r = (x. Show that Y * v ^ v « Y Y • v is a scalar quantity and is given by dv dvw 3v V.+ ^. Example 4.228) (4.b) = curlv or (rotv) V(cp + \/) = V(p + V\/ Y«(v + a) = V .b) = divergence v or (div v) i v _ d j d k d ~X~ 37 37 d7 vx vy vz (4.226) (4.b) dz2 ^ dy2 + ^ V2 = — + i i _ + .a (Y • v) .233) Thus Y • v t.VECTOR AND TENSOR ANALYSIS 303 dv x dx dv dy dv 7 dz (4.224a. a Y x ( v + a) = Y x v + Y x a Y x (v x a) = (a • Y) v .231) ax 2 dy2 az 2 Example 4. y. r = I r I and © is a constant vector.v + V .21. z) = i x + j_ y + k z.22. show that .
 2 .239a) (4.38a) ( l + y 2 + l J r (4 = ( x2 + y 2 + z 2 ) = j x (4.xcoy) =0 (iii) From Equation (4.r V.(zcoy .zco^) + ^ .242a) (4.242b.G))(co. we have Yx(Gjxr) = (rY)G3r(Y.yooz) + ^ .(©xr) = O Vx(fflxr) = 2ro Ycp(r) = (4.(xcoz . (co.and = .241) The first two terms on the right side of Equation (4.239b) (4.241) are zero since ro is a constant.240b) Y • (co x r) = ^ .240a) (4.Y)r = ( c o ^ + o o ^ + c o ^ j r = ( i cox + j^ (Oy + k ODZ) = C Q (4.c) .234) (4..304 ADVANCED MATHEMATICS 0) (ii) (iii) (i) Y(p(r) = l ^ . on substituting these expressions into Equation (4.236) (4. we obtain Ycp(r) = ^ = (ii) T (fi + j l + Yk) ^ I (4.235) (4.V)r + ra(Yr) (4.237b)  .238b) (4.and.238c) We have similar expressions for ^ .(ycox .237b).229).237a) i ^ + j  P + kP ax ay dz (4.
+ j ^ .31. y. •B A FIGURE 4.VECTOR AND TENSOR ANALYSIS 2Q1 Y«r = ( i ^ . z) along a curve C from the point A to the point B as illustrated in Figure 4.243c) (4. 42c.3 LINE. 43c) yields Yx(coxr) = co+3co = 2Gi 4. the integration is along the xaxis.oy oz } — (4.244a) (4.243b) (4.243a) = 1 +1+1 = 3 Combining Equations (4. We now consider the integral of a scalar function (p (x.31 Integral along a curve C .+ k^)»(xi+yj+zk) \ ox . SURFACE AND VOLUME INTEGRALS (4.32).244b) Line Integral of a Scalar Function In the definite integral defined by Equation (1.241.
We shall consider only smooth curves.32b) (4. The line or curvilinear integral of (p along the curve C from s = a to s = b is defined as I = I <p[x(s). z(s)] ds /a (4. y(s). the region enclosed is to our left.f <p(s) ds Ja Jb (4.31) where s is the arc length of C and the points A and B correspond to s = a and s = b respectively. then A coincides with B and the line integral around a closed curve C is usually denoted by I = (f <p(s) ds where a modified integral sign is introduced. Then in Equation (4.32c) = f ©(s) ds Jc If the curve C is closed. Unless otherwise specified. This means that C has a unique tangent at each point and such a curve is referred to as a smooth curve. The representation of the curve C by the arc length s is not always simple for integration.34a) (4. Thus (4. The evaluation of the line integrals is done by writing them as ordinary integrals.306 ADVANCED MATHEMATICS Let the curve C be represented parametrically by r (s) = x(s) i + y(s)j_ + z(s) k (4. The properties of ordinary definite integrals are equally valid for line integrals. The positive direction along a closed curve is the direction such that as we move around the curve. the integral is taken along the positive direction. It might be more convenient to use a new parameter t instead of s.31).33) f (p(s) ds = . The line element ds is then given by . the variable is t and not s.32a) = [ (p(s) ds JAB (4.34b) f JAB q)(s) ds = f (D(s) ds + f q)(s) ds JAP JPB where P is a point on the curve between A and B. We assume that x (s) is continuous and has continuous first derivatives for all values of s under consideration.
32.36) where (p = x 2 + y 2 .35a) = Vt 7 ! * • V(¥) 2 + (!) 2 + (f) 2 * Example 4. t2 X FIGURE 4. 2. and C is the triangle OAB with O the origin. 0.32.35c) I = [ <p [x(s). z(s)] ds (4.32 Integral of q > around the closed curve C From Figure 4. A the point (1. 0) as shown in Figure 4. y(s). Evaluate the line integral ( 4 35b) (4. it can be seen that the integral I may be written as I = f Jok cp ds + f ( p d s + f JAB /BO (pds (4. so .VECTOR AND TENSOR ANALYSIS 307 ds = Ydr»dr (4. 0) and B the point (0.31. y = z = 0.37) Along OA.
310c) (4.308 ADVANCED MATHEMATICS f JOA <pds = ( JO x 2 dx = U  =1 3 (4.312a) (pds = f JO [ ( l .310a) (4. then (4.j lit.312b) (A T>\2C\ y+. ^ Along BO.31 lb) (4.31 lc) (4.1) z =0 Equations (4. If s is the arc length of AB.c) L 3 Jo The line AB is given by y = . b) may be written in parametric form as x=lt y = 2t z =0 where the points A and B correspond to t = 0 and t = 1 respectively. .b.38a.t ) 2 + 4 t 2 ] ^ dt dt = V ? f [l2t + 5t2]dt Jo — — (4.313a) (4.310b) (4.39b) (£)V(£) 2 + (£f + (£) 2 = V(l) 2 + (2)2 = VT f JAB <43lla) (4.313b) = .2 (x . x = z = 0. so f JBO (pds = f y 2 dy JI (4.39a.39a) (4.
If v is a velocity and C is a closed curve then I.314b) I = + = 5 41> 1 3 Line Integral of a Vector Function Let v [x (s). The tangent to C at the point P is by definition the line joining P to a neighboring point P 1 as P' approaches P. Evaluate I = j JA (4. then di / I = dt/ dr E(S + A S ) . y (s). Thus X is by definition given by T = lim As>0 = ^ ds If C is defined in terms of t instead of s. 13b). is known as the circulation around C. z (s)] be a vector field defined at all points of a smooth curve C and T be the unit tangent to C.£ ( S ) (4. Example 4.317a) (4. 12c. we obtain 3 3~~3 (4. If F ( = v ) is a force.32.37.316) /" d T The scalar line integral of v along C is then defined as I = f Y'Tds Jc = f Y»dr Jc where dr = T d s . then I represents the work done by F in moving a particle along C.317b) Y»dr (4. c.314a) (4.VECTOR AND TENSOR ANALYSIS $02.315b) As (4. b. Combining Equations (4.315a) (4. 8a. in fluids mechanics.318) .
z = 0 joining A = (1. 0.33.z . 2y) in displacing a particle along the parabola y = 2x2. .310 ADVANCED MATHEMATICS where v = (2x. 2). 0.320a) = I . The path of integration is shown in Figure 4. . 2) to (1.320d) Example 4. 0) to B = (0. 1.33. 0.2 cos 0 sin 0 dG (4.s i n 8 i + cos8j ] d 8 (4. The vector position r of any point on C which is part of a circle can be given by l = c o s 0 i + s i n e ^ + Ok (4.319) B X FIGURE 4. and along the arc of the circle x 2 + y 2 = 1.320C) (4. 2. Calculate the work done by the force F = (x. z = 2 from the point (0.320b) Jo = i[cos2e][J /2 = 1 (4. 0).33 /• B rn/2 Line integral along curve C I JA vdr= JO fit/2 [ 2 c o s 8 i + ( 2 c o s 8 + 2 s i n 8 ) k ] « [ . 2x + 2y + 2z) = i (2x) + k (2x + 2y + 2z).
It then follows that y = 2t 2 since y = 2x2.322b. The more practice one has. 2.VECTOR AND TENSOR ANALYSIS Ml The path of integration is shown in Figure 4. 2).34 Line integral along a parabola The vector position r of any point on the parabola may be written as r = t i + 2 t 2 2 + 2k (4. fBFdx = f J A J Q [ t i .c) The evaluation of vector line integrals.34 with point A = (0. such as . the luckier one gets. The point A corresponds to t = 0 and the point B corresponds to t = 1.2 j + 4 t 2 k ] » [ i + 4tj]dt  (4.322a) = f1 [t8t]dt =l JO 2 (4. Such parametrization is usually done via educated guessing. 2) and point B = (1.321) This has been obtained by choosing x = t. AZ ) • y / FIGURE 4. 0.
323b) can be done by integrating each component separately.o. 0. y.323a) (4. This is illustrated in the next example.2TTb)l (t=27T) ' I I /o(N x y FIGURE 4. 0) to the point (a.34. 2rcb) Jc Jc on the circular helix illustrated in Figure 4. a sin t. bt) (4. z) = (a cos t. Example 4.35 Line integral along a helix .m ADVANCED MATHEMATICS I = f Y ds Jc or I = f vxdr Jc (4.35.324) (a. Evaluate I £ ds and I £ x d£ from the point (a. 0. given by X = (x.
by ds = 7 ( .328b) The above integral is an example of a repeated integral and is evaluated by integrating in the order given in Equation (4.328a).326c) (4.1 cos t) i . we have I I(x)dx= I f(x.325b) (4.328a) = 1 I f (x. 0.326a) = Va 2 + b 2 L (a sin t) i . y)dy dx Ja J a. 2Kb) corresponds to t = 2K. The line element ds is given.327b) r 9 i2n = [ab (t sin t .327c) (4.(a cos t) j + ^ .327d) In Equation (1.a sin t) 2 + (a cos t) 2 + b 2 dt = Va 2 + b 2 dt f L ds = f Jc Jo [(a cos t) i + (a sin t) j + (bt) k ] Va 2 + b 2 dt (4.327a) = I [ab (sin t .83).326b) J0 = Va 2 + b 2 [ 2 K2 b k ] I £ x djr = I [a cos t i + a sin t j + bt k ] x [a sin t i + a cos tj^ + b k ] dt (4. via Equation (4. 0.325a) (4.ab (t cos t + 2 sin t) j + az t k J 0 = [2n 3bi + 2K a2 k ] Repeated Integrals (4.VECTOR AND TENSOR ANALYSIS 311 We note that the point (a.k — \ 2 ' (4. 0) corresponds to t = 0 and the point (a. y) with respect to y between y = u (x) to y = v (x). then the order . If we now integrate I (x) with respect to x between the limits x = a and x = b.35c). y) dy dx Ja Ju(x) (4. yJ\x{x) /•b /v(x) (4. we have defined a function I(x) by integrating a function f (x.ab (t sin t + cos t) j + a2 k J dt (4. If u (x) = c and v (x) = d where c and d are constants.2 cos t) i .
If further f (x. y) is 1/(1 + x). y. we integrate f(x. x = a and x = b.331).1 = l(xl)2 (4. y) = 1.x 2 + 2x . z) dz J dy dx ) (4. A thin plate is bounded by the parabola y = 2x . Thus. In a single integral the integration is taken along a curve. y. The equation of the parabola may be written as y = 1 . The area of the base of the cylinder is A. Thus a triple integral can be defined as * rh fv(x) / /•q(x.y)dydx (4. then f f f(x. the area A over which the integration is to be performed is the area bounded by the curves y = u(x).329a) Ja Jc J a J c rb _ J a.331) Ja. Thus an alternative notation is /•b iv(x) re f(x. In a double integral the integration is taken over an area. [yu(x) (. in Equation (4.328b). If z = f (x. y) dy dx = f f <p(x) V/ (y) dy dx (4. y = v(x).314 ADVANCED MATHEMATICS of integration is not important. y) to z = q (x.x 2 and y = 0. We can extend the process of double integration to triple or higher integration.330) If f (x. y) then the double integral is the volume of the cylinder formed by lines parallel to the zaxis and bounded by z = 0 and z = f. y). y) is separable and may be written as a product of a function of x and a function of y.y) In Equation (4. y) resulting in a function of x and y./p(x.332a) (4. Determine its mass if the density at any point (x.y) \ I = I I f(x. (p(x)dx l[f d J L.35.: y(y)dy (4.332b) . z) with respect to z between the limits z = p(x.y)dydx= J a Ju(x) JJ A f(x.329b) Thus in this case the repeated integral becomes the product of two single integrals. Example 4.330) is the area A. say g (x. then the double integral in Equation (4. The triple integral is thus reduced to a double integral and can be evaluated as shown before.
334) y=2xx 2 y FIGURE 4.3.VECTOR AND TF.NSOR ANALYSIS 575 The parabola intersects y = 0 at x=0 x=2 (4.o) ** Shape of the flat plate.J — dy dx Jo [Jy=O 1 +x J Note that the double integral results from the fact that we consider a plate of unit thickness.36.333b) The shape of the plate is shown in Figure 4.36 y^o \(2. we integrate with respect to y first from y = 0 to y = 2x .35a) .334). Since the density is independent of y. (4. Then we integrate with respect to x from x = 0 to x = 2. The integral with respect to y is indicated by the arrow In Equation (4. The mass M of the plate of unit thickness is given by f2 f2xx2 1 M =I I . we obtain on integrating with respect to y M= r2r^q2xx2dx JO L 1 + x J o (4.x 2 as can be seen from Figure 4.36.333a) (4.
d) / / _ \ j ^—L{ FIGURE 4.36.y + 3x3in(l+x)j = 4 .335d) (4. x = 0 and x = 1.2 x . ADVANCED MATHEMATICS = f 2 ^x) d x JO (1+x) (4 .335c) v r 2 i2 = [ .37 illustrates the area.335e) Figure 4.W. y = 0. 3 35b) v } = ( f .c. then the area A is given by A =J J dy dx (4.l dx l+x J 7o L (4. (4. If we integrate with respect to y first. Find the area bounded by the curves y = x 2 .37 •.3in 3 Example 4.336a) ^O^MTJO^ (4336b. Sketch of the area .x + .
= _ d(u. The limits of integration can be inferred from the diagram.339) where A is the area of integration in the xyplane and A' is the corresponding area in the uvplane. then y = y (u. y(u. It is always useful to sketch the region of integration. z(u.b) Similarly for a triple integral. z) dx dy dz =jj j [f(x(u. w). v. v. w). y 3 / 2 ] 0 (4. The Jacobian J is given by dx 3 C V) 3u J = g ^ .337b) =\ = 1 (lVy)dy Jo = [y . v)] IJ I du dv A A' (4.VECTOR AND TENSOR ANALYSIS 3JJ We can interchange the order of integration. w are defined by . If u and v are two new variables such that x = x (u. v).341) V V where V is the region of integration in the xyzspace and V' is the corresponding region in the uvwspace. v. then A is given by A=/o1[/^dTy /•I (43"3?a) (4. y. y) dx dy = jj f [x(u. • A change of variables may sometimes make the evaluation of a double integral easier. v) 3y_ 3u dx dv _ 3y 3v (4. v) (4.b) jj f(x. y(u. v) . we have Iff f(x.337c. v.d) In this example we have seen that it is easier to integrate with respect to y first and then with respect to x. w)) I J  du dv dw] (4. The variables u.340a.338a.
w). O<0<7C.b. ( >  is 0<r<l. It can be seen from Figure 4.344) can be written in a parametric form as x = ar sin 0 cos § y = br sin 0 sin § x = cr cos 0 The relations between (x. TheJacobian J is y = y (u.37. z = z (u.= l a2 b2 c2 (4. w) (4.38. §) are illustrated in Figure 4.+ z.c) dx du dx dv dv dz 8v dx dw (4 3 . w) = *L y y du dz du dw dz 3w Example 4. v.347b) (4. v. Evaluate the triple integral. v.38 that the range of r.318 ADVANCED MATHEMATICS x = x (u. 0<(J)<2TI (4.345b) (4. 0. by a suitable change of variables. z) and (r. 0.b. w). / / / [x2 + y2+z2]dxdydz V where V is the region bounded by the ellipsoid x2 V2 z2 ^. 43 b) j = ^ 4 d(u.342a.347a) b sin 0 sin § c cos 0 = abcr2sin0 .346a.345a) (4. y. v.344) The ellipsoid given by Equation (4.c) The Jacobian J is given by a sin 0 cos § J = ar cos 0 cos § br cos 0 sin (() cr sin 0 ar sin 0 sin § br sin 0 cos (j) 0 (4.+ y.345c) (4.
348c) = aM j 0 j s i n 3 0 ^2n + b27t) + (c2C0S2e s i n e ) (2jc)j d e (4. y. z) and (r.348b) 5 \0 J it / = ] r I n27t r ]2TC J 0=0 <>=0 abc 0  a 2 sin 3 e  + isin2(j) I + c 2 cos 2 0 sinO [ J Q +b 2 sin 3 9 ^Isin2() \ d6 (4.VECTOR AND TENSOR ANALYSIS 319 c # ^ 1 FIGURE 4.348d) .38 Relations between (x. 9.1 n 2TI = abc — I I [a2sin30cos2<>fb2sin3esin2(l>fc2cos2e sine] d())de (4.348a) 6=0 ()=0 r=0 . <j)) for an ellipsoid i f f [x 2 +y 2 +z 2 ]dxdydz v 71 27C 1 = I I [r 2 (a 2 sin 2 6cos 2 ()+b 2 sin 2 esin 2 (()+c 2 cos 2 e)abcr 2 sinedrd(j)de] (4.
349c. If we keep the value of v fixed and let u vary.x 2 . y = y (u.y 2 x = a sin 9 cos < > . then P will trace out a curve C u as shown in Figure 4.350c.320 ADVANCED MATHEMATICS = &&£. y. v) (4.350a) (4. in the xyzspace is mapped into a rectangle.349a) Equation (4.34%) (4. Similarly by fixing the value of u and letting v vary. z = a cos 0 (4. in the 0<)plane. y) or in parametric form as x = x (u. (a 2 + b 2 } [ _ 1 ( c o s J \ l * = 4rcabc ( a 2 + h2 + C 2 } e s i n 2 e + 2 CQS e ) l " + 2c 21 J0 L cose 3 I Joj (4. e) defines a mapping (projection) of a region A in the xyzspace into a region A' in the uvplane.e) (4. 0 < < Let P be a point with vector position r on a general surface S given by Equations (4.39. z) = constant or z = f(x. Thus the equation of the surface of a sphere of radius a and center at the origin may be written as x2 + y 2 + z 2 = a2 z = ± Va 2 .d. y = a sin 0 sin (j).348e) (4.349c. d. v).349c to e). >  < 2n. z = z (u. a curve C v will be traced out.35!) .210) by n=f^4 (4. 0 < 0 < K. The tangent to Cu is given by ~~ ( = ! „ ) and the tangent to C v is — ( = I V ) .e) We note that the surface of a sphere of radius a. A unit normal n to the surface S is a vector which is perpendicular to the tangents to C u and C v and is given via Equation (4.348f) Surfaces The equation of a surface S may be written as < p (x. v).350b) (4.d.
z(s)] satisfy Equation (4. n is also given by a= (4. Similarly we can deduce that grad(p»T v =O where T v is the unit tangent to C v .353. y(s). [x(s). From Equations (4.349a).354) _grad(L I grad < p (43 . we deduce that grad cp is perpendicular to both T_u and T v and thus is a normal to the surface S. 54).352) From the definition of grad (p and the tangent to C u .39 Curves C u and C v on surface S The curve Cu can also be given by Equation (4.352) can be written as gradcp«T u =O where T u is the unit tangent to C u .55) . On differentiating with respect to s and using the chain rule we obtain cpdx+3«pdy+3cpdz=0 dx ds dy ds dz ds (4. we find that Equation (4. That is to say.31) and since C u lies on the surface S.353) (4.VECTOR AND TENSOR ANALYSIS 321 Z i \ >/ oJ^ / x r • y FIGURE 4.
The surfaces we shall consider are twosided surfaces.351).310 is an example of a onesided surface.350c to e) —^.356a) . Find the unit normal n to the surface of a sphere of radius a. 55). It is obtained by twisting a strip of paper once and gluing the ends together.a sin 0 k 30 (4.310 Mobius strip A surface is said to be smooth if its unit normal exists and is continuous everywhere on the surface. A Mobius strip illustrated in Figure 4. An insect can crawl on that strip and reach all points on the strip without ever having to cross an edge! Thus the strip has only one side. there are surfaces which have one side only.322 ADVANCED MATHEMATICS If r u and r v are interchanged in Equation (4. The equation of the surface of the sphere is given by Equations (4. Example 4.351. Therefore. If S is a closed surface. In such a case we cannot designate a positive side.38.= a cos 0 cos §i + a cos 0 sin <)j . The union of a finite number of smooth surfaces forms a simple surface. However. using Equations (4. centered at the origin. the sign of n is reversed. n is chosen to be positive if it points outwards. we obtain one circle. we need to establish a convention to label one side of S to be positive. If we cut the strip along the centre line. FIGURE 4.
z)dS s (4. sin 0 sin §.356b) a= f^4 _ (a 2 sin 2 9 cos <).JJ/UJ a^sin 0 = (sin 0 cos ((). then the surface element dS is given by dS=  r u x r v l d u d v (4.361).361) In Equation (4.357c) =§ (4.350a) and cp is given by (p = x 2 + y 2 + z2 = a2 = constant grad (p = 2x i + 2y j + 2z k = 2r From Equation (4. it has to satisfy the equation of the surface. z is not an independent variable. a 2 sin 2 0 sin §. z) is a scalar function.VECTOR AND TENSOR ANALYSIS 321 — = .3. cos 0) (4.a sin 9 sin §i + a sin 6 cos (j)j (4. a2sin 0 cos 0) — — (4. That is to say.360b) (4.57a) {H. Surface and Volume Integrals The surface integral is an extension of the double integral to an integration over a surface S.355) we have n = p^j= I (4.357d) The equation of the surface of the sphere is also given by Equation (4.359b) since I r I = a.359a) (4.358) (4.y.360a) (4. If S is given in parametric form. If cp (x. it is given by Equation (4. y.349b).362) . then the surface integral of cp over S is denoted by I =J(p(x.
324 ADVANCED MATHEMATICS The surface integral I given by Equation (4.363) where S' is the region in the uvplane corresponding to S in the xyzspace. z) and S is the plane S 2x + y + 2z = 6 in the positive octant. Evaluate I I v • n dS where v is the vector (4x.v)(ruxrv)dudv S S' (4. is shown in Figure 4.361) becomes I = ff cp[x(u. 62).368c) S'.367) that z=3uv/2 (4.367) The original inclined plane in the xyzspace is replaced by its projection in the uvplane and we write u = x. On S z = 3xy/2 (4.31 la. then from Equations (4.369) .u . z(u.351. y=v (4.31 lb.39.b) It follows from Equation (4. we deduce the following relations ffv* ndS =J*Jv(u. If v is a vector field.365a) (4. The vector position r of any point on S is X = u i + vj^ + ( 3 . the corresponding region of S in the uvplane. v). v)] I r u x r v  du dv S' (4.366) Example 4. v) (r u x r v ) du dv s s1 (4.368a. y(u. y.364) (4.365b) [JvdS = rj(v x i + vyj^ + vzk)dS S S = ff (vx i + vy j^ + v z k ) I r u x r J du dv S' ff (p n dS = ff cp(u. v). The region S over which the integration is carried out is shown in Figure 4.v/2) k (4.
k Xv = i .2u2 + 4u + 6] du = 54 (4.^ .v.37 lc) = 3 f [.311 (a) Region S in the xyzspace.3u)»(l.370a) (4.371a) =[ f 3(u+l)dvdu (4. we have j7v«ndS =JY(4u.VECTOR AND TENSOR ANALYSTS ^ ru = i .370c) z A v. 1 2x+y+2z =6 X A^~~z^^—t J^S—— io) ' — * ^ (b) * • u FIGURE 4.365)..37 lb) u=0 v=0 =3 ( JO [uv + v] 6 " 2u du 0 (4.37ld.370b) (4.e) .k l u x i v = i + ii + k (4. (b) Region S' in the uvplane Using Equation (4. ljdvdu S S' 3 62u (4.±.
375) where V' is the volume in uvwspace corresponding to V in xyzspace. Thus fff V Y dV = Iff V [i vx + j vy + k vz ] dx dy dz [ i v x + j vy + k v z ]  ( r u x r v ) « r j du dv dw (4. Evaluation of volume integrals can sometimes be simplified by a transformation of variables. one defines the tangents r_v and r w . v. y. the curve generated by allowing u to vary while keeping v and w fixed. Similarly.341). z) dV) . y.374) iff cp(x. Equation (4. w). w)] I (r u x r v ) • r w l du dv dw (4. is r u . v.375) is identical to Equation (4. z(u. v.226 ADVANCED MATHEMATICS If cp (x. y. w).372) Some authors use only a single integral sign ( I cp(x. If v (x.376b) = fff V Examples of volume integrals are given in the next section. then the volume integral of cp over V is a triple integral of cp and is written as JJJ V Cp(^X' Y ' ^ dx dy dz = JJJ V (P(X ' y ' ^ dV (4. for the volume V integral as written on the right side of Equation (4. w)j + z (u. z) dx dy dz =jjj V V cp[x(u. y. instead of three. z) is a scalar function defined throughout a volume V. w) k (4.376a) (4. then the vector position r of any point can be written as r = x (u. . v.372). w) i + y (u.372) becomes (4. Thus if we make the transformation from the xyzspace to the uvwspace. then the volume integral of v is evaluated by decomposing v into its components and evaluating each component separately. z) is a vector function.373) The tangent to the C u curve. v. v. Thus the volume element dV is given by dV = I (r u x r v ) • r w  du dv dw The integral given in Equation (4. y(u.
then (fv'ndS S = f ff divvdV = (ft V. (j)) and these two sets of variables are related by an equation similar to Equations (4.44a) (4.c) The volume V is enclosed by two spheres. SURFACE AND VOLUME INTEGRALS Two important theorems in vector analysis are Gauss' theorem and Stokes' theorem.b. y. .b) y = r sin 0 sin ().4la.4 RELATIONS BETWEEN LINE. y. The tangent vectors are £ r = sin 0 cos § i + sin 0 sin § j + cos 0 k LQ = r cos 0 cos (() i + r cos 0 sin (() j .y_dV V V (4. z = r cos 0 (4. z)/r 2 and V is the volume enclosed by the spheres x 2 + y 2 + z 2 = 1 and x 2 + y 2 + z 2 = £ 2 .44c) (4. Example 4.c) We make a change of variables from (x. Gauss' (divergence) Theorem If a vector field v and its divergence are defined throughout a volume V bounded by a simple closed surface S. e < 1. as shown in Figure 4. z) to (r. diw = A/M+A(M+A/M ( 4 . S t with radius 1 and S £ with radius e.b) where n is the unit outward normal to S. They are also known by other names and the origins of these theorems were summarized by Ericksen (1960).41.350c to e). We shall state these theorems without proof.r sin 0 k XA = . 0.44b) (4. Verify Gauss' theorem if v = (x.43a.4_ 2 a ) 3x \r2) dy U 2 ] 3z Ir 2 j 1 ?x2 1 2v2 1 ?7 2 1 rz r4 rz r4 rz r4 rz (4.42b.r sin 0 sin () i + r sin 0 cos 0 j dV =  (j:r x x e ) • r^  dr d0 d<J) = r 2 sin 0 dr d0 d<) (4.VECTOR AND TENSOR ANALYSTS 327 4.45a.41. which is x = r sin 0 cos (j).
46b) e = o L <t> =o = f 2% (18) sine d6 0 = 4JC(1E) (4.364). we obtain .328 ADVANCED MATHEMATICS FIGURE 4.360b).r2sin 6 dr # e=o L 4 > =o \r=e r / v 7C I" 27C d6 (4.46c) (4.e ) sin 6 d<) d9 (4.41 7i r Shaded volume V enclosed by sphere Sj and S e 2TI / 1 \ f([divvdV = f f J 1. We also note that y may be written as v = L r2 (447) Using Equation (4.46d) The unit normal to a sphere is given by Equation (4.46a) = f n f (1 .
49) If Y • n dS = f Se ji r f 27i JL . . On S £ .410a) 9=0 [ _ <(»=0 =J f £sin0d(()de (4.48b) (4. We circumvent this problem by enclosing the origin with a sphere of > 0. 10c).48b). To obtain the volume integral of div v in a region that includes the origin. In this case n = = % r 2n (4. we let 8 — >0 In this example.VECTOR AND TENSOR ANALYSIS u r 2i m JJ v»ndS =J S! 7t I r»rsin6d<> d6 (4.410b) e=oL<()=o = 4TI e From Equations (4.48a) 9=0 L <>=0 = J In sin 9 de 0 = 471 Note that in Equation (4. it can be seen that (4. In this example. we obtain from Equation (4.41 lb) Thus the divergence theorem is verified.L E2 si n e ^ d9 £ (4.46d) with e — .410c) f f f divvdV = f f v « n d S + f f v . 8c.46d. div v (= 1 /r ) is not defined at the origin and so we cannot apply Gauss' theorem in a region that includes the origin.48c) (4.41.41 la) S£ = (fvndS S (4.n d S V Sl (4. radius e. the outward normal is towards the origin as shown in Figure 4. r » r = 1 since S 1 has radius 1.
415a) = J ff q • Yep dV V (4.415b) may be written as q.415b) Equation (4. S V (4. it is not necessarily perpendicular to F. it follows that F = 0.i2Q _____ ADVANCED MATHEMATICS I I I div v dV = 4TC V • (4.414c) (4. since the right side of Equation (4.b) Since q is arbitrary. we can deduce that . as usual.416a) is zero. x = q * I + qy 87 + ^ ~ i = q«Y(p From Gauss' theorem. b).J f f Y ( p d V = q « F = O .ff <pndS . we have f f ( p q » n d S = ((( div (<p q) dV S V (44"l4b) (4.413).412) We can deduce the following from Gauss' theorem.414a) .4l6a. verifying Equation (4.214a.413) where. (ii) JJ(nxv)dS =j"JJ(Yxv)dV S V (4. 1A. (i) ff <p n dS = fff S V Y 9 dV (4. Y • (<p q) = _ (<pqx) + — (cpqy) + — (<pq_) 3(p 3cp dtp (4.A .417) From Equations (4. S is the surface that encloses volume V. Let q be an arbitrary constant vector. However.
Gauss' theorem yields (4.VECTOR AND TENSOR ANALYSIS .421) can be derived as in cases (i) and (ii) by considering a vector cp Y \j/ and applying Gauss' theorem.419) becomes q • jj (n X v) dS = q • [ [ [ (V x v) dV S V (4.421). (v x q) = q • (V x v) where q is an arbitrary constant vector. (Hi) (( cp ^ S dS = ((( V (<p V2 v + V cp • V \/) dV (4.421) where ^ — is the directional derivative and is defined in Equation (1.(vxq)dV S V (4.417) is true. as shown in Figure 4. The vector field v is a function of x and y only and v z = 0. 418). on Equation (4. Equation (4. ^ Y . .421) is the mathematical statement of Green's first theorem.423) is the statement of Gauss' theorem in twodimensions.214a.^ +^.419) Using Equations (4.420) It follows as in (i) that Equation (4. Apply Gauss' theorem to a right circular cylinder bounded by the planes z = 0 and z = h.42.510). Green's second theorem is obtained by interchanging cp and \j/ in Equation (4.423) (iv) jf A . Example 4. The result is // ( 9 ^ ~ ¥ l ? ) d s =SSJ{(?V^' v v2( p )dv S V ( 4 422) (4.418) JJ (vxq)ndS =JJJv. Equation (4. b.421) and subtracting the resulting expression from Equation (4. Equation (4.42.dxdy =i> (v x dyv y dx) C where C is the curve enclosing the area A in the xyplane.
The unit outward normal to S j . dV = h dS 2 .n j dS + ( ( v » n 2 dS + ff v » n dS S Sj S2 Sc (4.^2 _ ADVANCED MATHEMATICS y / FIGURE 4.426a) Since we are considering a right circular cylinder of height h.424) Since n j and n 2 are of opposite sign and v is independent of z. the first two integrals on the right side of Equation (4.n d S = [ [ v . Applying Gauss' theorem yields [ f Y»n dS = f f f d i v v d V Sc V (4. S 2 (the plane z = 0) and the curved surface S c .42 Right circular cylinder bounded by planes S 1 ? S 2 .424) cancel each other. and curved surface S c The three surfaces that enclose the volume V are Sj (the plane z = h). f f v . nj is in the positive zdirection whereas n 2 . the unit outward normal to S 2 is in the opposite direction.425a) ///fe^H V ( 4 425b) (4.
43. Substituting Equations (4.423)] for this simple geometry.430) where T is a unit tangent to C and n a unit outward normal to the surface S.426b) h J vnds =h jf (^f+ ^ c s2 The normal n to the curve C is given by dxdy (4.429) = //& + ^l d x d y A ( 4 423) where we have replaced S2 by A. We have thus deduced Gauss' theorem in two dimensions [Equation (4. we obtain (4.426a. the line integrals along AB will cancel. The direction of AB from C to Cj is opposite to the direction from Ci to C. 28) and dividing by h. we obtain f (vx dy . then <f v • T ds = f f curl v • n dS C S (4. Thus we only need to evaluate the integrals along C and C\.vy dx) = ff ( ^ + ^ ) dxdy s2 c (4.427. Therefore.425b).427) (4. Stokes' Theorem If the vector field v and curl v are defined everywhere on a simple open surface S bounded by a curve C. b) into Equation (4. We then go around Ci in the positive direction once and leave C\ at B to rejoin C at A and proceed along C in the positive direction until completion of the circuit. This is illustrated in Figure 4.428) Combining Equations (4.VECTOR AND TENSOR ANALYSIS  i i i dS c = h ds where ds is the line element along the circle that encloses the surface S2. then at a point A on C we make a cut and draw a curve from A to a point B on Ci. We have to ensure that the direction of integration is chosen . If the surface S is enclosed by two curves C and Ci.
b) define curl v as i curlv = ox y3 j ay x3 k az 0 (4. 1) (4.433a) . The same technique of making appropriate cuts can be extended to surfaces enclosed by more than two simple curves. 3 (x 2 + y 2 )] n = (0.43 la) = [0.43. x 3 . Equations (4. 0. 0) and S is the circular disk x2+ y 2 = 1 with z = 0.43lb) (4.334 .43 Surface S enclosed by two curves C and Cx Example 4. FIGURE 4.225a. ADVANCED MATHEMATICS properly. 0. Verify Stokes' theorem in the case where v = (y 3 .432) fj s curl v • n dS = [ f 3 (x2 + y2) dS s (4.
430) is verified.436) Let q be an arbitrary constant vector.437a) . (ii) (( S ndS = i(f C rxdr (4. cos 9) £ v • T ds = f C (. The following can be deduced from Stokes' theorem. that curl v = 0.r (Y»qJ(q • V)r + q(Y«r) (4. 0) d0 = f Jo (sin4 0 + cos 4 0) d0 Q + ^2 (4. Since the line integral is zero. v is then said to be irrotational. Such a vector v is called a conservative vector. cos 0.s i n e .435d) 4 L^o = ^ Thus Equation (4.433b) (4.433c) The curve C enclosing S is the unit circle (r = 1).VECTOR AND TENSOR ANALYSIS 335 = 3 = ~ Jo f2K r 1 9 r z r dr d0 Jo (4. from Stokes' theorem.434b) (4. we have curl(qxr) = (r • Y ) q . (i) If the line integral I v • T ds is independent of the path of integration and depends only on its C end points. From Equation (4.*] (4.sin 0. then the line integral of v around a closed curve C (a curve with coincident end points) is zero. 0) • (.434a) (4. the unit tangent T is given by 1 = — . cos 3 0.sin3 0.435b) = f _ cos 0 sin 3 0 + sin 0 cos3 0] '* + 3 [ & (^ L 4 4 J 0 ^ J J ^ ^ ^ (4. it follows. In parametric form.229).435a) Ue de/ = ( .
5 APPLICATIONS Conservation of Mass Consider a fixed volume V of a fluid of density p enclosed by a surface S./// l d V V (452a'b) (4. Equation (4.437c) curl (q x r) • n dS = j> (q x r) • dr C (4..438) From Equations (4.£& ADVANCED MATHFMATICS = .440) where A is the area in the xyplane enclosed by curve C.439) Since q is an arbitrary vector.437c. (iii) Stokes' theorem in twodimensions is given by <f (vx dx + vy dy) = If .^ . 4.438) may be written as 2 q «  * J n d S = q<£ r x dr S C (4. 214).436) is satisfied.q + 3q = 2q From Stokes' theorem \\ S (4.437b) (4.51) The rate of influx of mass Q • t . we deduce that Equation (4. The mass m of the fluid is m = ((( V pdV (4.53) where Q = .  p Y « n d S S .^ dx dy C A (4.
f(x) must be zero). The minus sign indicates that n is an outward normal.58a) (4. Combining Equations (4.52a. 7c) yields J.58b) .(p vx) + A.54a) to transform the surface integral to the volume integral and as a result we obtain Equation (4. ^v7] 3p = p div v + v • grad p Combining Equations (4.55) is valid for any arbitrary volume V it must true at every point.57a) 9p 9p . (This is similar to I f (x) dx = 0 for any limits a and b.54a) (4. Therefore Ja ^.56.VECTOR AND TENSOR ANALYSIS _ £57 and v is the velocity of the fluid.57c) (4.54b) which can be written as /// [ df+ diV (P ~i dV = ° V (45'5) Since Equation (4. (p Vy) + ^ (p v j dv ^vy (4.+ div(pv) = 0 Expanding div (p v).56) rb div (p v) = J . we have (4.54b) = .A * ™.+ v • grad p + p div v = 0 at which can be written as Dp —^ + p d i v v = 0 (4.((( V div(p v)dV We have applied Gauss' theorem to the right side of Equation (4. 3) yields / / / lf~dV V = . b./ / Px#ndS S (4.
We assume the boundary condition to be (p = f(x. everywhere except at P. We want to determine the value of (p at a point P.511) which gives the value of (p on S is known as the Dirichlet condition.18 that we need to construct a Green's function G that satisfies the homogeneous equation. y.18 for ordinary differential equations.G ! ) d S = / / / U2G^ S V V 2 < p ) d V (4. the potential (p satisfies the equation V2 9 = p (x. we have Laplace's equation. inside a volume V enclosed by a surface S. which is sufficient to ensure a unique solution to Equation (4. In some problems.512) Replacing \/ by G in Equation (4. V G = 0. such as electrostatics. that is to say G=0 on S (4. ~.513) .510) using the method of Green's functions which was introduced in Section 1.510) is known as Poisson's equation and if p = 0. which for incompressible fluids (p = constant) simplifies to div v = V • v = 0 (4.is given at the boundary and this condition is the Neumann condition.510) Equation (4.422) we have / / ( ^ . We choose P to be the origin. in the present case. We propose to solve the inhomogeneous Equation (4.58b) is the equation of continuity in fluid dynamics.510).59a. z) on S (4. We recall from Section 1. ADVANCED MATHFMATICS where £• = — + v • grad is the material or substantial derivative.£M . which is a homogeneous equation. y. We also assume that G satisfies the homogeneous boundary condition. z) (4.511) The boundary condition given by Equation (4. L X at Equation (4.b) Solution of Poisson's Equation In many applications.
515b) (4.1 (4.514) as e — » 0.515e) (4.1 £ (4. Equation (4.513) becomes (4. G is singular at the origin. As in Example 4.517a) .41.> 0 To evaluate the right side of Equation (4.512) and cp satisfies Equations (4. we need to isolate the origin by enclosing it with a sphere of radius e. In V .VECTOR AND TENSOR ANALYSIS 339 Since G and V G are not defined at the origin. the region enclosed by S and S e .515c) dr //[^<#s=f ff M^lkH d e se L J (4.515d) = J I j 9 + £ ^ 1 sin 0 d(j) d0 (4.514) s se vve Near the origin we assume ~ to be the dominant term of G. To evaluate I £ we note that p is finite everywhere and let its upper bound in V e be M.516) where I £ is the contribution from V e . G satisfies Laplace's equation.V e . On S e we have r=e (4. Then in V e we have Ip<M (4. we work in terms of spherical coordinates. surface S e and volume V e as in Example 4. a s e . 11).515f) = 47U(p(0).510.41.515a)  . Equation (4.= fdn G . we write f f f G p dV = ((( VVe V G p dV .
we generate a relationship between x and y.517c) (4. y) (4.517b) (4. we obtain < p < 0 ) = if/// G p d V + // f ^ d S ] .52 la) (4. b). 15f.518) G Ie<M7ie3l ^ 7tMe2 Thus as e ^ 0.517d) (4. On eliminating t between them.514. Combining Equations (4. then x(t + T) = x(t) y(t + T) = y(t) (4.520a) (4. y) and g(x.521b) . following Chapters 5 and 6 on partial differential equations. If we can solve Equations (4. NonExistence of Periodic Solutions Many dynamical systems to be covered in Chapter 10 are governed by a nonlinear autonomous system 4 r = f (x. y) are continuous functions with continuous partial derivatives.517e) (4. We can plot x versus y and the obtained curve is the path or trajectory.519) gives the value of 9 at P in the form of an integral involving the Green's function G. Ie > 0. The system is said to be autonomous because f and g do not depend explicitly on time t. V S ( 4 5 19) Equation (4.520b) where f (x. 18).520a. If x and y are periodic and of period T. We will be in a position to construct G.^C Ve ADVANCED MATHEMATICS = ^rte3 =^ (4. we obtain x and y as functions of t. y) dt ^ = g (x. 16. The xyplane is known as the phase plane.
g t) d t =0 ( 4 '522b) (4. then the left side of Equation (4. The closed curve will be traversed once as t increases from t 0 to t 0 + T.522c) cannot be zero. Thus if 5— + ^ p does not change sign in the phase space.423) we set vx = f and vy = g. (ii) Thus laws (i) and (ii) can be expressed as .522c) follows from using Equations (4. In other words there is no periodic solution.m. This experimental observation is usually stated as Neumann's law and Lenz's law. If ^— + ~ dx dy is of one sign only in the phase plane. So this leads to a contradiction which implies that there is no closed curve.520a. b). then an additional electromotive force (e.) is set up in the circuit and is of magnitude ~. Equation (4. that is • = —+~ dx dy is either positive or negative throughout the phase plane.522b) is obtained by introducing the variable t. Thus a periodic solution corresponds to a closed path in the phase plane.f. (i) Neumann's law: if the magnetic flux N through a closed circuit varies with time. Maxwell's Equations Faraday discovered that if a closed circuit is being moved across a magnetic field or if the circuit is placed in a varying magnetic field. Other criteria for determining the existence or nonexistence of periodic solutions are given in Cesari (1971). This is Bendixson's negative criterion. b) does not have a periodic solution.VECTOR AND TENSOR ANALYSIS m The path in the phase plane will then be a closed curve.520a. then we can write // (fx~ + If) dx dy = §(f dy gdx) A C ( 4 522a) = f T ( f *. Lenz's law: the current induced in the circuit opposes the change in N. If in Equation (4.522c) Equation (4.520a. for every t 0 . Integrating once around the closed curve C corresponds to integrating from t = t 0 to t = t 0 + T. then the system given by Equations OX 0y (4. It should be pointed out that the nonlinear Equations (4. b) are usually difficult to solve exactly and only approximate solutions can be obtained. a current is generated in the loop. Thus it is of interest to have analytical criteria to determine the existence of periodic solutions.
jf ?= «n dS s (4.525) where B is the magnetic induction.M2 ADVANCED MATHEMATICS em.6 GENERAL CURVILINEAR COORDINATE SYSTEMS AND HIGHER ORDER TENSORS (4.f.f.528) Cartesian Vectors and Summation Convention A vector y is represented by its components which depend on the choice of the coordinate system. around a closed circuit is equal to the change in potential in going around the circuit once. and N are measured in the same system of units.527) holds for every S. we have e.m.d N (4.2 = at Equation (4. If we change the coordinate system. = (t E » T d s (4.523 to 25) £ E • T ds = . we obtain by combining Equations (4.m. Since S is a fixed surface. it follows that curlE = . we write (4.[ [ ?= • n dS s c Applying Stokes' theorem.f. the components will generally change. Since the e. = . Thus there is a relationship between the components of v in .527) Since Equation (4.m. The vector v however is independent of the coordinate system.f.528) is one of Maxwell's equations in electromagnetic field theory. The total normal magnetic induction N across any surface S is N = [ [ B»ndS S (4.523) provided both the e. 4.524) c where E is the electric field.526) jf s n»curlEdS = .
z) system as shown in Figure 4. y.62a) and k = ^x i * I + Vy I * I + ^z £ * I (4. It is the object of this section to establish such relationships. z) obtained by rotating the (x. z) system.VECTOR AND TENSOR ANALYSIS Ml one coordinate system and the components of the same vector v in another coordinate system. Vector v has components vx. v = vxi + vyj + v z k = vxi + vyj + v z k (4. v y and vz are obtained through forming dot products with unit vectors j_ respectively. z) system and components v x . y. vy. b) with the unit vector i . v z in the (x. vz in the (x. y.62c) v z = v x I • k + vy J • k + vz k • k . y.61. vx= v x i » i + v y j / i + v z k • i Similarly.61 Vectors v and r in two coordinate systems The component v x is obtained by forming the dot product of Equations (4.6la. We wish to relate those components.b) z i ^ X. 7 / PCx.6la. vy (4.y. and this can be achieved as follows. z) to another Cartesian coordinate system (x.z) /I / I : h FIGURE 4. v . y. We start by considering a transformation from one Cartesian coordinate system (x.62b) (4.
The three equations given in matrix form in Equation (4. y. the indices in £mn notation has the following advantages (i) are written as numbers instead of as x.j = k«k = 1 Inverting the set of Equations (4. x 2 . v 2 . j . Note that the indices in (x 1 .k = i .b) (4.66b) = % inm . i ) cos (i. 3 ) and (5. we can extend it to an infinite dimensional space.e. z) will be relabelled as (x 1 .63d. j ..344 ADVANCED MATHEMATICS Note that in the rectangular Cartesian system i»j_ = j _ . In this notation. are the direction cosines and can be represented in a more compact form by £mn. v v z ) will be denoted by (vj. x 3 ) and (x *. i ) £2l = J ' i = 1  l  c o s ( J . 1? 8.62a to c). z and this in the case of an extension to an n dimensional space. (ii) Similarly a vector v with components (v x .65c. 2 > < 3 3) respectively. 6.b. x 2 . k) will be denoted by (6. where m takes the values 1. 2 and 3. The nine quantities £ \ \. v z ) can be represented by v m . . The coordinates (x.k = O i .64). i) is the cosine of the angle between the unit vectors i and i . k) and (i.63a. we would run out of letters. x 3 ) respectively. In Equation (4.d) where £n = I « i = 1 111 cos ( I . v 2 = vy and v 3 = vz. the notation is more compact. where both indices m and n take the values 1.f) K\ vy = I hi hi hi \ /M £2l £22 £23 vy (4. vy. 2 and 3.i = j . j .c) (4. the unit vectors (i.65a.. z) and (x.66a) (4. v t = vx.64) can now be written as 3 vm = X n=l ^n ^nm (4.64) \vj \ £3l £32 £33 I \v z / (4. Thus x 2 is not x squared and we shall denote x 2 squared as (x 2 ) 2 with a bracket round x2. x 2 . we obtain (4. v 3 ). The components (v x . where n can be greater than 26. 5. Indeed. y. 2. £21. y. x 3 ) are written as superscripts and the reason for this notation will be explained later.
b).610) . 2 or 3. v 2 . From the definition of £mn.67) can be replaced by any other letter except n (or p). 5r (4. Thus the three equations represented by Equation (4. Equations (4. In our case. and Equation (4.66a) are obtained by assigning the value of m = 1. Thus in Equation (4.66a). r (we cannot use 5.66b). The index n is called the dummy (repeated) index and the right side of Equation (4. x ^ x 3 ) coordinate system is an orthonormal coordinate system §nI r = 8nr (4. we must apply the same value to m wherever m occurs.67).67) The right side of Equation (4. m can take the values of 1. which is the same as the summation over all the possible values of n as implied by Equation (4.b) can be written as ^=vm£m = %^n (4. Similarly the free index m in Equation (4.66a) is a summation over all the possible values of n as indicated by the £ sign.VECTOR AND TENSOR ANALYSIS 345 In Equation (4. and is equal to 0 if n ^ r. whenever an index occurs twice and twice only in an expression it implies summation over all the possible values of that index.68) Since the (x '.69) where 8nr is the Kronecker delta and is equal to 1 if n = r. we can replace it by other letter. v 3 ) can be expressed in terms of (v l 5 v 2 . we should not replace the dummy index n (or p) by the free index m. making use of the properties of £pm.6la. The components (vj. known as the Einstein summation convention.67). unless stated otherwise. or by repeating the process used in obtaining Equation (4. 2 and 3 in turn.5r= v n 6 n . According to this convention. p say. Once a value of m is chosen.66b).65a. we obtain vm5m. n or 5_ m due to the summation convention). we have & m S r = ^rm (4. We shall adopt the second procedure to demonstrate the elegance and conciseness of the summation convention. the index n is a dummy index and the expression on the right side of Equation (4.67) is a summation over all the possible values of p.6la. Since n is a dummy suffix.66b) can equally well be written as vm= Vpm (4.66b) implies summation over all possible values of n. Equations (4. v 3 ) by inverting Equation (4. the index m is known as the free index and it can take any of its possible values. To obey the summation convention. The summation sign occurs so frequently that it is useful to adopt a convention.b) Forming the dot product with 5.
b). it is not generally permissible to do so in a general coordinate transformation.612b) ^rmxm = xr Here we have freely written the indices as superscripts and subscripts. is associated with 6. (4.613a) Note that £Tm can also be obtained via Equation (4. from Ox*x 2 x 3 to Ox 1 x 2 x 3 is a linear transformation. Although this is permissible in our present coordinate transformation. That is to say xm = xPpm (4. b) imply that dxT dxm = 8x m dlr tAr^ (46'13b) . the transformation.61 lb) On summing the right side of Equation (4. m. r and the second index. Substituting Equations (4. which is r.612a) (4.613a. is associated with 8 m of the unbarred coordinate system.69.612a) by replacing the dummy index p by r. then the components of the vector position OP (= r) will transform from the barred to the unbarred system or viceversa.612a. Because the direction cosines £rm are constants. we obtain ^rmvm = % 5nr = vr (461 la) (4. and the coordinates of P are (x1. 1 lb).67.613c) . as we shall discuss in the next section. to yield r)xm 'nn==7 Equations (4. 10) into Equation (4. as defined by Equations (4. x 3) relative to the Ox 1 x 2 x 3 and Ox 1 x 2 x 3 coordinate systems respectively. due to the definition of 5 nr . If P is any point in space.612b) that (4. x2. that the only surviving term is v r .68). x3) and (x *. we find.61 la) for all possible values of n.346 ADVANCED MATHEMATICS Note that the first index in £rm. Further we note from Equation (4. x 2. according to Equations (4.
the index m "cancels" and we have the superscript r on both sides of the equation. we obtain fdxm axr vr = 3xr dxm V m (4.61 lb). Similarly the spherical polar coordinate system is chosen for solving flow past a sphere. we are left with a subscript r and on the left side we also have only a subscript r. Similarly in Equation (4. General Curvilinear Coordinate Systems The rectangular Cartesian coordinate system is not always the most convenient coordinate system to use in solving problems. q3) as shown in Figure 4.614a). We now consider a general curvilinear coordinate system (q1.614b) should be written as vr = — dxm vm (4. The laminar flow of a fluid in a circular pipe is solved using a cylindrical polar coordinate system. b)] and thus there is no need to make a distinction between subscripts (covariant components) and superscripts (contravariant components). components that transform according to Equation (4. In Equation (4. The superscript m is associated with Ox 1 x 2 x 3 and the subscript r with Ox 1 x 2 x 3 . m is a dummy index. both laws of transformation are equivalent [Equations (4. We can consider them as "cancelling" each other. we need to distinguish between covariant and contravariant components. The velocity is then a function of the radial position r only and not of two variables x1 and x2.b) Vm Equations (4. b) describe two types of transformation laws. Components that transform according to Equation (4. b) into Equation (4. we regard m as a superscript and r as a subscript) and dxT dx r once as a subscript in vm.614a. Substituting Equations (4. The choice of coordinate system depends on the geometry of the problem.614b) are known as contravariant components and the indices are written as superscripts. q2. But for Cartesian components.614a) are called covariant components (vm) and the indices are written as subscripts.614a.62. it occurs once as a dxm Bxm superscript in —=— (in the expression—=—. Equation (4.615). Thus in proper tensor notation.VECTOR AND TENSOR ANALYSIS 341 This is only true in the case of a Cartesian transformation.613a.615) Note the symmetry in the notation.614a. On the right side. . Thus in a general coordinate transformation.
616b) (4. we define a set of base vectors g m of the generalized curvilinear coordinate system as being tangent to the coordinate axes q m .q 2 .62 Generalized coordinate system This coordinate system is related to the rectangular Cartesian coordinate system (x 1 .616c) (4.63. q 3 ) x 2 = x 2 (q 1 . q 3 ) x 3 = x 3 (q!. The vector connecting the origin to point P is r. as illustrated in Figure 4.q 2 . Likewise. x 3 ) to (q 1 .616a) (4. q 3 ) or in concise notation xm = x m( q n) (4. q2.617) We assume that the transformation from (x 1 . x 2 . x 3 ) by x 1 = x 1 (q 1 . x 2 . q 3 ) is onetoone and it can be inverted as q n = q n( x m) (4. . q2.618) The base vectors 8 m of the orthonormal coordinate system can be defined as being tangent to the coordinate axes x m .348 ADVANCED MATHEMATICS q2 Q / q1 FIGURE 4.
the tangent and the normal coincide. . In the particular case of the rectangular Cartesian coordinate system.355) and. which explains why Equation (4.b) Base vectors defined as tangents to coordinate curves are covariant base vectors.63. They are given by Equation (4.61. in the present notation. For example the base vector 5 3 is normal to the surface x^ = constant.63 Generalized coordinate system and base vectors The base vectors g m are given by (4.c) Note that both g m and g n are not necessarily unit vectors.b.619a. Example 4. The base vectors 8 m can also be considered as being normal to the planes x m = constant.613c) is valid. Base vectors defined as normal to coordinate surfaces q n = constant are contravariant base vectors and are denoted by g n as illustrated in Figure 4.VECTOR AND TENSOR ANALYSIS 349 V3 q1 FIGURE 4.620a. that is to say. they are written as g n = V q n = gradq" = ^ 6^ (4. This is not generally the case. it is parallel to the x3axis. Obtain the covariant and contravariant base vectors for the spherical polar coordinate system.
61%) ^x 1 gr = ^ 8 ?)x2 ^ £ 2 ?\3 + ^ 8 1 + 3 (4. q1 = r. q2 = 0. x 3 = r cos 0 (4.r sin 0 8 3 ( 4 6"22c) (4. 6 _24a) = sin 0 cos <) 8 l + sin 0 sinty8 2 + cos 0 8 3 (4. 0.350 ADVANCED MATHEMATICS The transformation from the rectangular Cartesian coordinate system (x 1 .622f) To obtain the contravariant base vectors g n we have to invert Equations (4.623b) ' ( X 2) 2 l V (xf § = arc tan — We calculate g n from Equation (4.620c) (4. c).621a.623a) ' (4.b.ty)is x1 = r sin 9 cos ().622b) = sin 0 cos § 8 ! + sin 0 sin 0 8 2 + cos 0 8 3 ?xl 7)\2 ?x3 8° = lr a > + tr^ + lr £ 3 = r cos 0 cos c f > 8 j + r cos 0 sin § 8 2 .623c) gl dx M i8 dx 2 + A83 dx (4 .622a) (4.c) (In our notation. b. x 2 . q3 = ()).r sin 0 sin (j) 8 j + r sin 0 cos () 8 2 (4622e) (4.622d) i*%&'+W^+%^ = . x 2 = r sin 0 sin <>. The covariant base vectors are given by Equation (4.624b) .62la. x 3 ) to the spherical polar coordinates system (r. We then obtain r = V f x f + f x ^ + fx3)2 0 = arc tan A / /(X1)2 l + (4.
626) (4. but they are not generally equal in magnitude. the indices m and n can take the values 1.628) and S ^ V q " gn = ^ 3x £ (4.VECTOR AND TENSOR ANALYSIS 351 g8 = rTsi dx + T7s2 + rTS3 dx + (4. g 3 = cos J^cos 6 g ^ sin 6 c A cos < b ~ 2 g* = ~ Note that gr = g r lQ =~ xl r^i^—^8 r sin 0 r sm 0 (4.g+ r2sin20 Thus g Q . .. where as usual. g A are parallel to g e .b) .625c) g• = —J—.630a. . The relation between these coordinate systems is qm =qm(qn) (4. Any vector v may be written as v = v m g m = v n g n (4.b) We will now establish the relationship between the components of a vector y in the two generalized curvilinear coordinate systems q m and q n . g^ respectively. 2 and 3. _ .625a) Ie (4.625b) (4.624d) .629a.627) or q n = q n (q m ) The base vectors will be given by (4. • We now consider the transformation from one generalized coordinate system q m to another generalized coordinate system q n .624e) (4. dx sin ^cos 6 g 2 _ si^e.624c) (4.
the components v s are defined as .634a) (4. 20)] and the chain rule.352 ADVANCED MATHEMATICS Since in the general case.J At>o At (4.633) are contravariant components. covariant and contravariant base vectors are different. we can no longer take dot products. b) yields ^ = v m (4Mla) (4. to establish the relation between components v m and v m . as follows «. in the q m coordinate system. See also Equation (4. are defined as vm Aa m _ jjm _ J _ At»o At 8q m =  (4. As far as the summation convention is concerned.632a.635a) .615).633) Components of vectors which transform according to Equation (4.= a ? ^ S. General coordinate systems are not necessarily orthogonal and g j • g 2 is not necessarily zero.619. Obtain the law of transformation of the velocity components v\ The velocity components v m .630a..68).62.b) It now follows that _ da n v n = ^— v m dq m (4. repeated indices should appear as subscriptsuperscript pairs.634b) On transforming to the q s coordinate system. Those relationships are obtained via the definitions [Equations (4. as in Equation (4. one has to pay special attention to the position of the indices. In the general case.s . Example 4. Aq s v = lim .63!b) ^ I n = vng n (4.fg Substitution in Equation (4.
638).637a) = ^ 1 vm dqm Thus the components v s transform as contravariant components.636) (4. we obtain v^lim ^ ^ At>o dqm At (4.637b) We further note from Equation (4.636) into (4. In Equations (4. using the chain rule.640a) (4. That is v = vmgm = vngn The relation between g n and g m can be deduced. as follows gra = Vqm = ^ Y qn (4.638) = ^ . 27).VECTOR AND TENSOR ANALYSIS 353 = ^ According to the chain rule Aq s = ^5_L Aq m dqm Substituting Equation (4.635a). we have expressed the vector v in terms of the covariant base vectors g m . b). • (4.630a. We could equally have expressed v in terms of the contravariant base vectors g n .636) that the components Aqm also transform as contravariant components.626. we obtain ^ v m (4. but we still write the indices as superscripts because Aqm are contravariant components.639) (4. as can be seen from Equation (4.635b) (4.640b) into Equation (4.gn dq n ~ Substituting Equation (4. Aqm are contravariant components and the indices are written as superscripts.640b) g n = vn g « (4. Although the coordinates q m are not components of a vector. The transformation from q m to q n is arbitrary.641) .
q2 and q3. the components u n are given by (4. we have seen that V the quantity V ^ v is a tensor of order two and is known as the velocity gradient. In Example 4. It is a _ (p is a vector but (p is a scalar.63.643b) (4. A vector has both a magnitude and a direction and its components transform according to Equations (4. In fluid mechanics.614a)]. One index is sufficient to specify its components.63.642) are known as covariant components [see.645a) (4. where cp is a scalar function of q1. also Equation (4.633. transform as covariant components of a vector. A scalar is a tensor of order zero and its numerical value at a point remains invariant when the coordinate system is transformed. . Example 4. we have 9cp 9q m (4. q 3) coordinate system. Thus tensor of order one.644) Applying the chain rule to Equation (4.354 ADVANCED MATHEMATICS We then deduce % = f ^ vm (4. Tensors of Arbitrary Order (4645b) So far we have considered only scalars and vectors. Show that the components of V(p. 42) when the coordinate system is transformed.644). q 2 . Let u = V cp then we define u m as um =  ^ In the (q 1 .642) Components of vectors which transform according to Equation (4.643a) 8q m = ^ »m Thus the components u n transform as covariant components.
633. One recognizes the 3qP 3 q m 9qP quantity — — (in the unbarred coordinate system) as representing the Kronecker delta.646). We need two indices to specify the components of a second order tensor. It thus follows that 9q l 9q T™  r1^vS =8 pu'= U p (465Oab) In Equation (4. 42). we are at liberty to change the free index m to p and the dummy index n to s.646). which is another tensor of order two. The stress tensor. Equation (4. be written as —^ — .647) l £ ». is equal to the sum of the velocity gradient and its transpose.VECTOR AND TENSOR ANALYSIS 355 the rate of deformation tensor. while a quantity such as —~— is not in general 8 \. q 2 . This can be written as u m = T m n v" (4. maps linearly the surface force on the surface of a deformable continuum to the unit normal on the surface. we obtain 3qP dqm 9q r = 3 q m dqn s (4. Similarly. " ^  ^ V da m Multiplying both sides of Equation (4. In a threedimensional space. Equation (4. according to the chain rule. 3q p r a dV ^u' q = T a^ a^ v ( 4 6"49) —5— would represent 5 s t . 2 and 3 and T has nine components.648. q 3) coordinate system. The quantities T m n are the components of a second order tensor which we denote as T.647) becomes (4.646) In Equation (4. The components of u are written as covariant components and the components of v are written in contravariant form.can.646) becomes um = Tmn v n Using Equations (4.648) by ——. In the (q *. we then obtain up = T ps vs (4. which is a tensor of order two. m and n can take the values 1. dar d a m Note that —^. Thus a tensor of order two transforms a vector u linearly to another vector v. we have adopted the summation convention by writing the dummy index as a subscriptsuperscript pair.651) .
. y Contravariant components T m n transform according to T mn = 3<L^ <*iil T rs dqr dqs (4. The components Cjjk^ need four indices and are the components of a fourth order tensor.642).652) are known as covariant components. q3) and (q l.S T m n f4 653^ ^ } dq p dqs Components that transform according to Equation (4. q 3 )in Equation (4. The law of transformation for a fourth order tensor can be obtained by generalizing Equations (4. q 2 .655) Tensors of order higher than two also exist and are frequently used. Note the similarity to Equation (4. An example of a tensor of order three is the permutation tensor. the infinitesimal strain tensor and the elastic tensor respectively. The constitutive equation of a linear elastic material may be written as tij = Ciju yu (4.657a) 3qP 3q r 3q s 3q> "" . y k ^ a n d cyk^ are the stress tensor.654) For second order tensors. which will be defined in the next section. in addition to covariant and contravariant components. we can have mixed components T ™ which transform according to (4. A tensor that maps linearly a second order tensor to another second order tensor is a tensor of order four. it becomes T 1 P 3a = S m ZH .653 to 55) as follows T SiSiH!5lT i w prSt (4.656) where tjj. q2.356 ADVANCED MATHEMATICS Comparing Equations (4. _ D da n °4 A. we deduce dq m Ml T (4 652) Interchanging (ql.652). 51).650b.
Metric and Permutation Tensors Equation (4. The commutative law does not hold and g j g • is in general not equal to g : g j . ds 2 .658) can also be written as I = uigjvJ gj = um g m v n g n = = ui VJI (4. If P and Q are two neighboring points with vector positions r and r + dr with coordinates (q1) and (qJ + dqi) respectively. T = uv = u®v Equation (4.f) i I j = um v n I m I " Tij l i I j = Tmn I m I " A second order tensor T is also known as a dyad and the notation adopted in Equations (4.658a.660C) (4.660a) dqmdqn = T*k ' TT dqm dqn (4. The second order tensor T may be defined as the dyadic product of vectors u and v.57c) A second order tensor may also be defined as the juxtaposition of two vectors u and v. between P and Q is ds 2 = dr«dr (4. If they are equal.b) (4.660d) = gm'gndqmdqn = gmn dq m dq n .VECTOR AND TENSOR ANALYSIS 357_ Tprst = 3 £ M_ _Bql M_ T ijk^ dq[ 9q j dqk dq£ (4.659c. Tensors of higher order can likewise be defined. The component Tjj is not necessarily equal to the component Tjj. The juxtaposition of the two vectors u and v is also known as the outer product of the two vectors.b) (4. then the square of the distance.657b) v rSt *tmLWL%!_Ti a q 1 3q r 3q S 3q l jW (46.659a to f) is called the dyadic notation.d) (4.659e.619a) defines the covariant base vector g m .660b) (4. T is a symmetric tensor.659a.
we have 9x p s (4. . The dual (conjugate or associate) of g mn which is denoted by grs is defined as §rS = I r ' I s = Vqr • Vqs From Equation (4.662a) (4. .663c) Equation (4. we can deduce an expression for gJn in terms of (x*) as follows *j • * " . the metric tensor simplifies to the Kronecker delta.662b) 3x^ 5 . hence the subscript notation.619b).358 ADVANCED MATHEMATICS The second order tensor g mn is known as the metric tensor and in general is a function of (q1). .663b) shows that the components of the metric tensor gmj transform as covariant components. . In an orthonormal coordinate system. .„ .. (4.$ * ' • ! £ * ' = (4664a) (4. In an orthogonal coordinate system ( =0 if m ^ n < 4 M l a 'b) Since the dot product is commutative.620c).664b) 3qi_ dx* 3^ 3x' ir = B3L ^ dxr dxr (4. g mn is symmetric.663b) . fi»"£JBa^*''a7*' =  ^ 3q m 3x p = 9q m 3qJ ^ 8p.664C) . 9qJ P£ dxP (4'663a) (4.. Starting from Equation (4. .
667c. 3 appear in the clockwise direction .668) .b.c) ° l (4.665d) (4.665b) (4.665e) In Equation (4.b) (4. The relationship between gmj and gJn can be established by combining Equations (4. and is defined as 0.663c. if the indices 1.d) (4.1 . The convention is framed so as to be compatible with the rules of transformation.f) = e312 = = e213 = .VECTOR AND TENSOR ANALYSIS 359 It can be seen from Equation (4.665c) — — hnr 3x r p r ^ 9x r (4.665e). we have written the Kronecker delta as a mixed tensor so as to conform to the convention that an index appearing as a subscript (superscript) on one side of the equation should also appear as a subscript (superscript) on the other side of the equation.664b) that gJn components transform as contravariant components. if any two indices are equal ejjk = 1.667e. The permutation tensor is an example of a third order tensor which in an orthonormal coordinate system is denoted by ejjk. e ll2 = e l22 = e123 e321 (4.667a. The fundamental reason for writing the Kronecker delta as a mixed tensor is because it transforms as a mixed second order tensor.! Let w be the vector product of two vectors u and v. 2. 3 appear in the anticlockwise direction Thus. Wj is given by wi = e ijk u j v k (4. 64c) and the result is in dxP dxP dq) dqn tAtzcs Smjg a^^a^^ =M ^ dqm = dqm = ^1 dqm = K 9x r (46"65a) *£L 3x r (4. 2. if the indices 1.666a. Then in an orthonormal coordinate system. for example.
671b) = Bta^ff«'v' Multiplying both sides of Equation (4. so as to avoid confusion with the Cartesian components.671a. we deduce that (46"73a) (4. We put a bar over the components of the vectors in the present coordinate system. Equation (4. u m and v n ) to the orthonormal coordinate system using Equations (4.670) Transforming (w^.671b) by ——.673b) e pst = *«» ivit^ = ^ eAnn dq£ dqm dqn (46"74) which is the law of transformation of covariant components. then the determinant of D can be written as Igl =eijkdildj2dk3 (4669) We extend the definition of the permutation tensor to a general curvilinear coordinate system and we denote it by e^ mn . 42). Equation (4.26Q ADVANCED MATHEMATICS If D is a (3 x 3) matrix.673a.672) becomes dqe dqm dqn wp = e ^ a x V ^ ^ U s V t = epst u s v t From Equations (4. Alternatively e^mn may be written as eAnn (4. identifying the coordinates qm as x m and q n as qn. Thus w"^ is expressed in a general barred coordinate system as W£ = e£mn " *" ^ " (4. (4. we do not distinguish between covariant and contravariant components.mn f £ »• flf V dqe 3x s 3x l (4.675) .633. b).670) becomes f 7 » r = e. with elements dy. we obtain 9xP dqe 3xr r dqe dqm dqn £^mn s t v (4672) ^ H w j V 1T7 TT u dxP 3q £ dxP dx s 3 x l Using the chain rule on the left side of the equation and noting that in an orthonormal coordinate system.
The components of the metric tensor are given by grr = g r • g r = 1 See = I e * I e = g«x> = g<t>g<t> = f2 (4. its dual g rs and its determinant g for the spherical polar coordinate system.64.b) X r2 (4.f) r2sin2e All the other gjj are zero since the spherical polar coordinate system is orthogonal.VECTOR AND TENSOR ANALYSIS 361 where g is the determinant of the metric tensor g m n .g r = l gee = (4. Contravariant and Physical Components (4.677e. grr = g r .b) (4. The contravariant form of the permutation tensor is E1* = j= e ijk (4. we have (4.61. Calculate the metric tensor gjj.d) gG. Via the metric tensor it is possible to establish a relationship between these two types of components and this is done as follows v = vm g m = vn g n Forming the dot product with g r .679) It has been noted that the covariant and contravariant base vectors do not in general have the same dimensions and it is not surprising that the covariant and contravariant components of a vector also do not have the same dimensions.678e.d) (4.677c.b) .678a. =^TLVY ~* r2sin29 The determinant g is given by g = r4sin20 Covariant.677a. The base vectors have been obtained in Example 4.680a.676) Example 4.678c.f) (4.gG = g** = g * ' g < .
686a) .682a) (4. This process is known as lowering and raising indices.362 Vm ADVANCED MATHEMATICS § m ' §r = v" In * Ir (4.683a. Similarly for higher order tensors we have Trs = gnr gms T n m Tnm = gnr gms T f s (4. Equation (4. The normalized covariant base vectors are given by Kn^lfVvf^  S nl °nn (4.682C) T™ = g nr g ms g pk gqi TJ* In a space in which a metric is defined. The physical components of a vector are expressed in terms of normalized base vectors.682b) (4. it is possible to transform covariant components to contravariant components and vice versa.68lc) shows the transformation from contravariant components to covariant components. This makes it necessary to define the so called physical components.685) implies that for each n. b) there is no summation and the index n is contained in brackets to designate physical components.684a. yields (vn^l)gn = O (4. vn _ ^n^ &nn (4. c). noting that m is a dummy index.b) Note that in Equations (4. It was pointed out in Example 4.684C) =>nn Combining Equations (4.683a.68la) v m 8™ = v n gm vr = gnr v n (4.68 lc) Equation (4. A vector v may be written as v = v = v m g m = v (n) g ( n ) V^M (4.61 that g r and g Q do not have the same magnitude and they have different dimensions.681b) (4.684a.685) Since the vectors g n are base vectors and are linearly independent.b) (4.
g n m in Equation (4.687) into Equation (4. we obtain Vg™~ v n = v (n) (4. b) there is no summation over n. The index n occurs three times! The contravariant component vn can be written as vn = gnm Vm (4. dqm In Example 4. in terms of normalized covariant base vectors.62. In Equation (4.688). In the framework of an orthogonal coordinate system. Obtain the contravariant. the physical components V(n) can be obtained via the metric tensor. at contravariant components of v are vr = ^ = f The (4.689).690b) = V g 1 ^ Vg^ T m = Vg^Vg^T^ (4.65.689) Similarly the physical components of higher order tensors are defined. For an orthogonal coordinate system we have V) = ^ ^ ^ ^ (4. we obtain v(n) = ^ g" m v m (4688) In an orthogonal coordinate system. they may not be identical since Equation (4. we have shown that ^— transforms as contravariant components.687) Substituting Equation (4. We could equally have chosen to define physical components via normalized contravariant components.69 la.689) could not be obtained from Equation (4. Indeed.b) . b.VECTOR AND TENSOR ANALYSIS v(n) 363 or = ^ n v" (4.686b) That is to say.690c) So far. there is no summation over n.690a) (4. Note that in Equation (4. both are identical. for a nonorthogonal system.686b).686a.688) represents three nonzero components for each value of n. combining Equations (4. In the case of a nonorthogonal coordinate system. Example 4. we have defined the physical components.66la. 88). covariant and physical components of the velocity vector v of a particle in the spherical polar coordinate system.
) (4.f) = ^7 V < 1 > = r (sin 6) i Example 4. then 3qm ur = ^ (4.d) (4. contravariant and physical components of V f in the spherical polar coordinate system. In Example 4.693a.692a. we have vr = g r r v r = f ve = g e e v e = r 2 0 v<(.691c.b) .364 ADVANCED MATHEMATICS ve = 4s. we have shown that —— is a covariant component. If we denote V f by u.d) (4.686b) v(r) = V g ^ v r = r >) = ^geev6 = f9 v«j.694b) (4.692c.693c.68 lc).= e at v* = .d) e) i (4.694a) ue = H u. Calculate the covariant.^ = 4 at Using Equation (4.692e.f) = &w v* = f2 (sin 2 The physical components are given by Equation (4.691e.63.695a.b) (4.6940 (4.693e.66.f) (4. (4.b) (4. = ^ The contravariant components are u r = g rr u r = ~ (4.
contravariant and physical components.695e. In a space in which a metric tensor exists.d) » = ^»* = 7±eH ( 4 6"96e'f) In Examples 4. contravariant or physical components.d) u«> = g<* u^ = — . then on the left side we must also have a mixed component. the components of a tensor refer to physical components. whereas the r. a tensor can be represented in terms of covariant. there is no distinction between covariant. as shown in Examples 4. That is to say. Each expression in the equation should be a tensor component of the same kind and order. 0 components.65 and 6. all have the same dimension. In the rectangular Cartesian coordinate system.695c. 9 and (j) physical components.65 and 6. • The laws of physics are independent of the coordinate system and they should be written in tensorial form.b) (4.696c. .1 — ^ v r 2 sin 2 0 5(t> The physical components are (4. The metric tensor is the Kronecker delta. They can be transformed from one to the other by the process of raising or lowering the indices.VECTOR AND TENSOR ANALYSIS 365 ue = g e e u e = L ^ (4. if on the right side of the equation we have a mixed component which is covariant of order m and contravariant of order n. Many authors omit the word physical when referring to physical components of a tensor. Finally the components of a tensor have to be measured in terms of certain units and it is desirable to express all the components in terms of the same physical dimensions. The quantities that enter into the equations expressing these laws should be in covariant or contravariant components. we note that the r covariant and contravariant components do not have the same dimension as the 9. Thus it is safe to assume that unless otherwise stated. covariant of order m and contravariant of order n.f) u(r) = Vg" ur = jjp > ) = V g ^ u0 = 1 H (4.696a.
but functions of (q1. q 2 .73) Note that 5 p .619b). we obtain h~JTs™ + vmjf dqJ dqJ dqJ Using Equation (4.7 COVARIANT DIFFERENTIATION We have shown in Example 4.626) is a linear transformation.630a).655. we have dvJ 3qj = 3 3qs (dql i —*. We shall presently show that if v1 is the contravariant component of a vector v.71b) 3 q s dq£ 3qJ 9q s 3q ' 3qi 3v i Comparing Equations (4. — — is a covariant component of a dq* vector. On transforming from (q 1 .626.63 that if cp is a scalar function. q 3 ).366 ADVANCED MATHEMATICS 4.7la) (4. q 2. q 3 ) to (q *. .^ S dqJ dq m 3qJ p P (4J"2) (4. q2. dv1 aqj does Consider the transformation given by Equations (4. That is to say. In general the partial deviative of the components of a vector is not a tensor. transforms as a component of a tensor only if the transformation dqi 3q s dq £ given by Equation (4.v [dq1 dqs L3qJ (An . 27). q 3 ) .63 lb) . the base vector in the rectangular Cartesian coordinate system is a constant. 7lb). we note that transforms as a mixed component if 9qj 32qi dvi — = 0. not transform as tensor. The transformation of the covariant base vector 5 p to g ^ is given by Equation (4. From Equation (4. (4. This is because the base vectors are in general not constants. we find ^ = . On taking the partial deviative of a vector there is a contribution from the base vectors and this has to be taken into account.
79) The covariant derivative of higher order tensors are .75a) = 3v£ _3qJ +v m m a V dq e] 4 g (4.75c. allowing us to factor out g £. it is given by . = — + Jv m dq) lm jj 'J (4. The covariant derivative of v^ with respect to qi is denoted by v^ • or v^ •. denoted by [rs.3 .76) m j{ 3 q m 3qJ ^? is known as the Christoffel symbol of the second kind and is also denoted by T£ •. d). we have replaced the dummy index m in the first term on the right side of the equation by I. From Equations (4.75a to b).77) The notation correctly suggests that \e ^ represents a mixed component of a second order tensor.g£ (4.78) The covariant derivative of the covariant component v^ is given by w£ d\e i = —• 'J aqj Imj \i j ! v (4.75c) = [v'Jg/ (4. t] (or r r s t ). The quantity (4.75b) 3q m 9qJ 3x p J ~ ' (4. is defined as I K t] = Su r s (4. 3qj and { z .VECTOR AND TENSOR ANALYSIS 367 dv dwm m 32xP =: = — r g m + v m : dq£ .75d) In going from Equation (4. do not transform as tensors. lm J) The Christoffel symbol of the first kind. £ I I \ vl .
Properties of Christoffel Symbols (i) The Christoffel symbol of the second kind can be calculated in terms of the metric tensor and is given by (4.713a) )= 1 g// *SlL (4.710a) T t a .710d) T ~J 3qJ . .m = ADVANCED MATHEMATICS 'J 9T^ 3qJ + t s j TSm+ m Ttt t j (4.710a to e). t j T '' Tlm (4710b) (4.712) (iii) If the coordinate system is orthogonal / l \ ( ) = 0.4 J J 2 dq£ .713c) . = y g" .jJT"mjT» nil™ (4J"10e) The covariant derivative of tensors of arbitrary order can be written down by observing the pattern shown in Equations (4. if t.i = ^U T ™"(mM T « T '.368 T.711) m j (ii) 2 ^qm 9qJ 3qkJ / M = (/ M lm jj V J mj (symmetry) (4.7. m.i= ^ + U lJ "l".13b) (4.710c) 3 T \ + I TSm mlft ul T < m (4. and j are all different lm j) i £ (4.
(iv) On performing the transformation given by Equations (4. does the Christoffel symbol \3q m 3qJ / transform as a mixed third order tensor. (iii) The metric tensors gy. no summation is implied. £ is a dummy index and the dot product of the two second order tensors is another second order tensor.k = Tij>k + S ij(k (ii) (4. Rules of Covariant Differentiation (i) The covariant derivative of the sum (or difference) of two tensors is the sum (or difference) of their covariant derivatives (Tij + Sij). 27).713d) In Equations (4.716a) (4. g^m and the Kronecker delta 8j are constants with respect to covariant differentiation Sij.716b).626.715) The covariant derivative of a dot (or outer) product of two tensors is equal to the sum of the two terms obtained by the dot (or outer) product of each tensor with the covariant derivative of the other tensor (Ti£ S^j).c) A consequence of this result is that the metric and Kronecker tensors can be put outside the covariant differentiation sign.713a to d). k = °rs. since the metric tensors are constants.j 3 (4.k = 0 (4.k) S mj (4.k + (Ti£. the Christoffel symbol transforms as ti\ __ j M a q ^ a q ^ ^ ^ L \m j) Is t/ 3q r 3 q m 3qj 9 q t 9q m 3 . In the rectangular Cartesian coordinate system.b.714) Note that only for the linear transformation = 0 .k = g ^ .VECTOR AND TENSOR ANALYSIS 369 1 = U"^? (4.k = Ti£ Smj.716a). That is to say . we have formed the outer product of two second order tensors resulting in a fourth order tensor.k = T i i s \ k + (TiAk) S ^j ( T i^ Smj). In Equation (4. all the Christoffel symbols are zero.717a.716b) Note that in Equation (4.
m is a dummy index in the third term and it can be replaced by s. The fourth order tensor R 1 1 1 ^ is the RiemannChristoffel tensor.719b) ~ aq k 8qj ~ b q k ^ j/J ^ (^ j / 3q k (^ k) 3qJ j k 3q s ^/JlohHMr. The dot product Rm^jk vm is a covariant component of a third order tensor.719c.720b) is a covariant component of a third order tensor. Rm^  k is a mixed fourth order tensor. ^n ^ "<*"<*=[rt t J . Thus the terms on both sides of Equation (4. Equation (4.h1"^ If we interchange the order of differentiation.^ L j + U L j } . Noting that the Christoffel symbol is symmetric (Equation 4.712).(  s S ) vt s f3vs  m \ 1  s \ \dv£ I (4.! / j { .720b) = R ° V vm The left side of Equation (4.270 ADVANCED MATHEMATICS (gijTJ^k = (TW)gij (iv) (4718) The covariant derivative of v^ with respect to qJ is a second order tensor and we can take its covariant derivative with respect to qk.79. we obtain K i) k = ^ T (vt\) d ldve = . 10b).I S vs j . .719a) m  m  1 \ d^W~\t jim\~U k [aqT "(s j Vm j~(j kHa^"U s)Vm (4. e) that ' d m  d  m Is I in I Is ( m il //t ^. d. Jj v ™ < 4 720a) (4. likewise in the fourth term s can be replaced by m. Combining Equations (4.719c) becomes ( 4 7"19C d i) ^ (4.719e).720b) are covariant components of third order tensors. contravariant of order one and covariant of order three. and assuming that v^ is continuous with continuous second partial derivatives. it follows from Equations (4.719e) v lM = + ^yi r 3 ( 3qJ 3qk / s  m \ m iiv i m i3vm f s ^/s m [dqj U k/J s vm+ U k 3qJ  m vm U j 3q k \k j 3q s U j U k/ m k j U s m On the right side of Equation (4.
From Equations (4.71. Example 4.b) {/J4« I I !(«**> = "in2e ( e 8 H «"!«*»>•* i 9 j = " 2 g e e !<«••> = .657c) and is thus zero in all coordinate systems. we can always set up a rectangular Cartesian coordinate system and the Christoffel symbols are zero. 0. Since the coordinate system is orthogonal. R " 1 ^ is a fourth order tensor and transforms according to Equation (4. The only nonzero Christoffel symbols are (9re! = 2«rrrteee) = r (4.VECTOR AND TENSOR ANALYSIS 321 We can interchange the order of covariant differentiation if R " 1 ^ = 0. calculate v r r + v e expression in physical components.S i n 9 c ° S e {/ j " 2 g++ I M = t ( 4 721cd) «™°*> (47"21g'h) <4™iJ) < 4 721M) 0 j/ 6 )2«**lr( 8 «). we make use of Equations (4.77. Let v* be the contravariant components of the velocity vector v.72. consequently the Riemann Christoffel tensor is zero. Obtain v' j in + v^ x . we have Example 4. In a Euclidean space. R 1 1 1 ^ is zero and the order of covariant differentiation is not important as long as the components of the tensor have continuous second partial derivatives. Rewrite this . We conclude that in a Euclidean space.c < * e spherical polar coordinate system.721a. 21a to 1).677a to f. Calculate the Christoffel symbols of the second kind for the spherical polar coordinate system (r. §). The metric tensors are given by Equations (4. The Riemann Christoffel tensor is a property of the space and is independent of the vector v. 78a to f). That is to say.713a to d).
372 ADVANCED MATHEMATICS r dr (m rj + (4. we have vr = v(r).J — v w r sin 6 (4.vr le el foe/ (4.722a) = ^ ( r )v' + ( f ve + /Mv* (4.686b). which is shown in the next example. + e * + a / x 9 n \ a / i \ 2 v (r) cot © '% v*.724b) vVv% + v V ^ + g + ^ + lv'+co.727) expresses the divergence of the velocity vector v and is a scalar.727) every term has the same physical dimension. we obtain r v'. Show that v1 . is a scalar..24a) = ^ . Rewriting in physical components. .723a) (4.722b) =~ dr (4.c) Substituting Equations (4.727) We note that on the right side of Equation (4.* • ^ (v«) + ^ (}v(e. using Equation (4.eve <4.b.^ vw) + ^U + _ v(6) (4. Equation (4.73. namely the reciprocal of time. v^ = . v9 = 1 v(e).+ ^ v r + cotev 6 a^ r (4.725).) + ^ ( .726a.722c) 'e ae \r e/ = — + i.7.725.723b) v * ae r 3v* + Mvr + Mve + M v t (4.726a to c) into Equation (4. Example 4.
72%) (4. The sum v* i is independent of the coordinate system and is a scalar or an invariant.28) 3qr 3qJ Setting j = i and summing Vi '' ^ ! M aq'dq4 = 8 sr v r>s = vss = v^j vr (4. and Curl The operator V in a general curvilinear coordinate system can be defined as V=griL dqr If cp is a scalar function gradcp = Vcp = g r ^SL . Grad. Div. we have shown that — J — transforms as a covariant tensor. c) are obtained by using the usual chain rule.a qr In Example 4.7.655) v i 'J V 3ql v r 'S (4 7.r (4.v l g t ~ ~ dqr ~ ~ dqr = gr!S Vs.b) (4.7. That is to say. we obtain the same value in the barred and in the unbarred coordinate system.63. The indices i and s are dummy and can be interchanged freely.732c.30) (4.73 la.732a.29a) 'S (4. and the property of the Kronecker delta respectively.729d) Equations (4.vs g s = gr  .b) (4.729c) (4.d) = I' It yt .VECTOR AND TENSOR ANALYSIS 373_ The component v* • is a mixed tensor and will transform according to Equation (4. 3qr If v is a vector 1 v = g r f .r .729b.
734c) (4.1 .734a) Ir#Is It (4. This confirms the statement made earlier that the operation of taking the grad of a tensor raises the order of the tensor. which in the V notation is written as V .733a) (4. The divergence of a tensor of order n is a tensor of order n .733b) (4. y = gr — »vsg = v s r gr • g s = v s r 8rs = vss The divergence of a second order tensor is defined as Z " l = gr~7Tstgs gt .b) (4.733c) (4. since t is the only free index.733d) (4. Thus the divergence of a vector is a scalar and the divergence of a tensor of order two is a vector. In Example 4.dq r ~ = T S t . then Vi = ^ dql V 2 cp can be written as (4. we have calculated the divergence of a vector.214 ADVANCED MATHEMATICS The components vS) r and v 1 r are the covariant and mixed components respectively of a second order tensor.736) .734d) = Tst>r 8rs g t = T st>s g t The component T s t s is the contravariant component of a vector.734b) (4.735a. The Laplacian of a scalar function cp is given by p = div (grad <p) = V • (V cp) V2 < If we denote grad cp by v.r (4.73.
737c).739d) may be written as u'= [^" ( s P r ) Vp] ff ( 4 '7"40) If the coordinate system is orthogonal..= I r (4. we have used Equation (4.739b) (4.7370 Note that V • v is defined as the covariant derivative of contravariant components.739a) = vs.68 lc) to transform the covariant components VJ to contravariant components v s .738) T 7x dq r vs §S (4. Equation (4. then { I is zero if all three suffixes are different but if r = s.737d) = g 4 . the permutation tensor erst is zero.b) (4.r I r x I S (4. In Equation (4.737c).737a.VECTOR AND TENSOR ANALYSTS 375 V2(p = Vv = \ = gsi — s s (4.. The metric tensor is a constant with respect to the covariant derivative and Equation (4. If the coordinate system is orthogonal. If u is curlv u.739c) 1 e rct = ^!sPrKJ#g_t = v s>r e r s t g t 9v s (n \ (4. The only possibility of nonzero contributions from the Christoffel symbol arises when p = s and p = r. ) ^ ] W aq1 ll l] aqj.737c) = g si — (4.M^l (4. The Christoffel symbol is symmetric [Equation .737d) follows from Equation (4.^ .739d) Note that u is expressed in terms of contravariant components and Equation (4.737e) simplifies to v2cp = g i i U ! ! _ _ / J .
678a to f.7.742a. s = 3 and when r = 3.740) becomes U Vg [dq2 13 2/ V 3 »3 2/ V 2 J + fg [dq3 \2 3 r 2 12 3 / ^ ] (4>/ ^ From the definition of the permutation tensor e231 = *' e321 = ~l (4. ADVANCED MATHEMATICS (4. e rst = . Thus Equation (4.745) becomes V2f a2f 1 \d2f vf = + — J I *L $ )j Bqj (4. If f is a scalar function of position.44) Example 4.43) In an orthogonal coordinate system. we find that the contributions from the Christoffel symbols cancel out and Equation (4.b) Using Equations (4. 72la to 1). Since the spherical polar coordinate system is orthogonal. s = 2. we can then use Equation (4.741) reduces to Ul = JL fe _ *2. ^g [8q2 aq3j (4.e srt . and the contributions from the Christoffel symbol cancel out. (()). 2n 3f • o o 3f + r sin 6 — + sin 0 cos 0 — dr2 r 2 [a©2 9rJ r 2 sin2 0 a^2 9r 30 (4.742a. Equation (4. write down V2f in the spherical polar coordinate system (r. (4.74.7.740) simplifies to ut = ^ M <!ls. consider the case t = 1. b).712)]. To clarify this point further.738) and we have v2f = g rr[^ _ ( J ) K] + g e e f e _ ( J I K] + g n [ ^ l Lar2 V r< aqjj [ a8 2 \Q Bl dqi\ [^2 Using Equations (4. The only nonzero contribution from erst arises when r = 2.745) ^ 3fl +r — + 1 \d2f . 0. Equation (4.746a) (47 _ 46b) = ^f 3r2 + l A r 2 + _ ^ _ 3 ? f i^si^B + 2 3 f + QatAaf 3 9 3i>2 r 3 r ^ 3 6 .m.
Assume that p and %_ are functions of r and 9 only..7470 Transforming all the components to physical components via Equations (4.7. we obtain uw = ^ [ l < r s i n 9 v < « > . the equation can be written in component form as . The equation of motion for slow flows may be written as gradp = d i v i 1 \ 3V/QX 3vM where p is a scalar (pressure) and 1 is the stress tensor.679.^  r2sin6 \ 3 r 99/ (4.^ r 2 sin 9 39 d < >  _^i) dr I =—L_ ^ r 2 sin9 ^ (4.VECTOR AND TENSOR ANALYSIS 377 Example 4. we have from Equations (4. and that x is symmetric.47a) ue (4.749) is written in the socalled coordinate free form.^ H < 4 748a) (47'48b) = ^e[sine^L+cosev»»lf] Um = ^u[^ ¥ »"s(" i l l e v »))] = < 4 748c) (4.75.747b) u* = — 1 — f^ft .t*. 89). 744) Ur = — L .748d) r ± ? [ l f ^e^»sin8vwj u <«^& (rv «>^M ( 4 748e) ( 4 7"48f) (4. If we denote curl v by u.749) in component form for the spherical polar coordinate system. q 3 ). Write down Equation (4.76.749) = H r "^" + v «/J Example 4. Equation (4. q 2 . Calculate the physical components of curl v in the spherical polar coordinate system. For a general curvilinear coordinate system (q 1 .686b.
750).V + s 2) V .2 2 V (4.1 2) * t 3q z (4751b) ^13 TV + / 3  s M l 3 U 3 ^1 " 1 3 V ^ = V .2 + t 2 3 .3 (4751a) + M \ s s l V  t \ i 3T. 2 2 I s I t \ 2 l i V + .752) Making use of Equations (4.62.3 dq3 (4. as is the right side dq1 of Equation (4.750). q3 = < >  (4. q 2 = 6.75Id) 7 T = V .3 (4.T + Is 1/ T 2 " 2 l } \ + 7 7 + 3q 9q 8x23 M i s I t 1 3 + T ~ + s 3h2 " 2 3/ xt ^q M i s I t \ 2 s 2 / V.j dq1 (4750) From Example 4.2 + T33. we have = V.is a covariant component of order one.2 / 2 1 3qz s / t \ 2 + ^33 / 3 1 s / t 1 3 .3 3 V ^q (4. f) become .75 lc) ^2! / 1 \ s / t \ 1 d%22 = . it is known that —*.751e) W dq M l s I t I 1 dx.751f) For the spherical polar coordinate system we identify ql = r . l +^2 2 .2 + V. Expanding Equation (4.75 lb. d.V + s 3 T3 .378 ADVANCED MATHEMATICS fr = *ij.72la to 1).l ~ dq 1 W = 7T dq + + *12. Equations (4. l + ^32.
754C) " ^ 5r" (f S l n 9 ^ r ) ) + ^ ( Sln 9 X(*0)) + o9 r sin 6 2 s i n 6 X«.VECTOR AND TENSOR ANALYSIS 272 dp 3Trr 3t r 9 1? tB9 tJ „ „ ij (4. In this section.j dV = fuinj dS V S (4.754a to c) every term has the same dimension. 4.n dS V S (4.r) + c o s 6 x(6$) Note that in Equations (4.753a) i =^T+lt + ^+"r6^ + COte (^V) W. we state the divergence theorem and Stokes' theorem for a first and second order tensor in a generalized coordinate system.8 INTEGRAL TRANSFORMS The divergence theorem which transforms a volume integral to a surface integral and Stokes' theorem which transforms a surface integral to a line integral can be extended to higher order tensors and higher dimensional spaces.x«4 ° = ^^ ( 4 J "54b) (4. We recall that the divergence theorem for a vector u is J divudV = J u . we obtain (4.753c) I = I M + h ^ + ( 2T(rr) " T r"^) + XJ^ ±$e =l*%(rt(er))+ ^ +2 T(er>+cote (T(0e) .753W + r sin 2 0 Tr* + sin 6 cos 9 T ^ Transforming all the covariant and mixed components to physical components.8lb) .8la) or [ uJ.
Example 4. we obtain f T^j dV = f T'Jnj dS V S (4.2 and 3).j dV = j e[i V S k uk n j dS (4. C is the closed curve bounding the surface S. 2 and 3.8la) and.84) Again.82) is a vector equation. For a second order tensor T.Ml ADVANCED MATHEMATICS where V is the volume enclosed by the surface S whose outward unit normal is n . Show that I curl udV = J n x u dS V S (4.86) On substituting Equation (4. we obtain the following f (e1 J k uk) . Stokes' theorem may be written as /(eijkuk.87a) . as usual.86) into Equation (4. To extend the theorem to a second order tensor T. contain one free index (i) and it represents three equations for i = 1. Both sides.84) represents three equations {£ = 1. containing no free index.8lb) is one scalar equation.85) Let T»i = e i J k u k (4.82) We note that Equation (4.81.82) and noting that e 1 ^ is a constant with respect to covariant differentiation. Equation (4. in component form. Equation (4.83) where.83) becomes /(£iJkTkJni S dS = f T^dqk C (4. we may replace u by T in Equation (4. Equation (4.j)ni S dS = lujdq1 C (4.
810) holds at every point and we obtain j£ (p v1) + (p vJ v1) j = ni'1 .j + p g1 (4.88) where v is the velocity. Equation (4.810) Since V is an arbitrary volume.p vi v1 dS . (1987) have obtained the equation A J p v d V = .VECTOR AND TENSOR ANALYSIS m f e[ikukj V dV = f EiikniukdS S (4. Example 4. Write down the equation in component form and hence deduce the equation of motion at each point in space. Since the volume V is fixed in space.88) becomes jL J p v1 dV = j n.89) To obtain the equation of motion at each point.85).811) Expanding the left side of Equation (4. We choose to write the components as contravariant components and hence Equation (4. Bird et al.[ 71 J1 j dV + j p g1 dV V V V V (4.87b) is the component form of Equation (4.J a • TC dS + J p g d V V S S V (4.J [ n • p v v ] dS .82. we obtain p !r + v i !!r + p v J V Vp v V i + pj v J v i ^^ (4. p is the density. Applying the law of conservation of momentum to a volume of a continuous medium in motion. we obtain the equation of continuity which may be written as ^ +(pvJ).87b) Equation (4.812) From the mass balance.f n^ n'i'1 dS + f p g1 dV V S S V (4.813) .j = 0 (4. 7t is the stress tensor and g is the gravity force. Equation (4. we need to use the divergence theorem to transform the surface integrals to volume integrals.89) now becomes f i (p v1) dV = f (p VJ v'Jj dV .811). we may include the time derivative inside the volume V.
b. 2a to d) yields ul = . Since tensors of order zero are scalars and are independent of direction.9 ISOTROPIC. 4. To find out if a tensor is isotropic or not. x 3 ) and rotate the axes to obtain a new coordinate system (x 1 . the components of these tensors are identical in all rectangular Cartesian coordinate systems. we can write "m = ^mn u n (491) ii) Let [x 1) be the coordinate axes obtained by rotating the (x*) system through n rad about the x3axis. and (u ') are the components of the same vector in the (x 1) coordinate system.U j . u3 = u3 (4. u2 = u2 = 0 (4.93a.c. i) All tensors of order zero are isotropic. their properties are independent of direction. 0 is the only isotropic tensor of order one. b). u2 = . that is to say. u3) are the components of a tensor of order one (a vector) in the (x1) coordinate system.814) The left side of Equation (4. we deduce that for u to be isotropic Uj = Uj = 0. (4. 13).u 2 .c) From Equations (4. x 3 ).b. ^22 = ~ 1 ' f 33=1' the other 4 j = 0.92a.c.812. the tensor is isotropic. x 2 .d) .811) becomes p ^ + vJ v1 ti = TCJ1 j + p g1 (4.93a. Below we list the isotropic tensors of order zero to four.b. u 2 . x2.91.814) is often written as —— ——I where 2_ is known as the substantial (material) derivative. OBJECTIVE TENSORS AND TENSORVALUED FUNCTIONS Isotropic Tensors Many materials are isotropic. then •*H=1.d) Combining Equations (4.94a. If in the new coordinate system the components are identical. Thus if these properties are described by tensors. we express its components in a rectangular Cartesian coordinate system (x1. It is the time derivative following a material element. If (Uj. Equation (4. they are isotropic.382 ADVANCED MATHEMATICS Combining Equations (4. Here we consider only Cartesian coordinate systems.
iv) The permutation tensor e} • k is an isotropic tensor of third order. iii) The Kronecker delta 8j: is isotropic.is a component of a second order tensor in the (x*) coordinate system and T r s is a component of the same tensor in the (x J) coordinate system.98) ijkersk = 5 ir 5 js 5 is 5 jr ( 4 9"9) v) Any fourth order isotropic tensor c} • k £ may be expressed as cHk£ = ^5ij5k^ + ^8ik5j^ + v5 i^5jk (4. incompressible Newtonian fluid. Any second order isotropic tensor can be expressed in terms of the Kronecker delta. = ' r i ' s j S i j = ^ri^si = S r s Thus the Kronecker delta transforms into itself and is thus an isotropic tensor. there is no nonzero isotropic vector. A useful relation between e^ j k and 8 r s is e (4'9"7a) (4.95) Letting Tj: be the Kronecker delta.c) (4. For a Newtonian fluid. .96) becomes Tr.910) where X. Example 4. In fluid mechanics. the isotropic part of the stress tensor 7U• can be written as rc[j0) = . we obtain u3 = 0 Thus 0 is the only isotropic tensor. that is to say.91. Obtain the constitutive equation of an isotropic. then T = I • I • T(A 96\ (4. If T.97b. Equation (4.VECTOR AND TENSOR ANALYSIS 383 By rotating the axes about the x^axis through n radians. \i and v are scalars. the deviatoric stress tensor x_ depends linearly on the rateofdeformation y .p S j j where p is a scalar.
Consider two observers. Since the second observer is both translating and rotating relative to the first one. Quantities which are indifferent to the motion of the observers are known as objective quantities.912a) and others (Bird et al. Q is orthogonal at all times. The coefficient (v + i) is known as the viscosity of the fluid. Objective Tensors The constitutive equation of a material should be independent of the motion of the material. one at rest (coordinate system x = x1).913) The vector c (t) in Equation (4. Some authors adopt the positive sign in Equation (4.. The matrix Q (t) denotes the rotation of the second observer relative to the first one and the elements of Q are the £.is symmetric. Such transformation is known as a transformation of frames of reference.913) denotes the translation of the second observer relative to the first observer.912c) (4.)Vij Equation (4.8j. and the other in relative motion (coordinate system x*1 = x*). both c and Q are functions of time t. . VYki5jk] = ± [ X y k k 6 i j + ^Y i j + VY ji ] = ± ( v + M. Also j . Note that in the present transformation.384 ADVANCED MATHEMATICS The constitutive equation of a Newtonian fluid may be written as (4.912b) (4. Alternatively we may state that the constitutive equation should be the same for all observers. 1987) adopt the negative sign.912d) ^yi. That is to say. we can relate these two systems by x* = c(t) + Q(t)«x (4. irrespective whether they are at rest or in motion. Y kk is zero.911) Since the fluid is isotropic.910) *ij = i ^ j S ^ = ±[XYkk5iJ +^k^ + + vfii/^jYw + (4912a) (4. the direction cosines of the axes x** relative to x*.912d) follows from Equation (4.912c) since the fluid is incompressible. we obtain using Equation (4.
Note that due to the rotation of the axes. i) Velocity vector v Differentiating Equation (4. Equally a second order tensor T is an objective tensor if T* = Q « T » Q t We now examine the objectivity of some tensors. but to an observer standing on the road we are travelling at a finite velocity. v is not an objective vector.917) (4. we seem to be at rest. If we denote the components of a vector u relative to the x1 coordinate system as u1 and the components of the same vector u relative to the x** coordinate system as u*'.913). This observation is a common experience. ii) The line element (ds)2 From Equation (4. we obtain v* = c (t) + Q (t) • v + Q (t) • x (4.e) .914) Since c (t) and Q (t) do not vanish at all times.Qdx = dxt dx = (ds)2 (4.918c) (4. we have dx* = Q ' d x (ds*)2 = dx*t dx* = dx*«dx* = dxtQt.913).VECTOR AND TENSOR ANALYSIS 381 Objective tensors are thus tensors which are invariant under a change of frame of reference.918a.b) (4.918d.915) (4. the components u*1 transform to components uJ under the usual tensor transformation laws. Sitting in a moving bus and watching the passenger sitting opposite to us.916) (4. then if u* = Q«u u is an objective tensor (vector).
L h * = I .923c) = Q ' L ' Q 1 ^ Q » Q f + Q « L t » Q t + Q»Qf Q*(L + L t ) .922) L is not an objective tensor since Q • Q is defined as is not zero at all times. iii) The rateofdeformation y Let L*=^\.[£ (t) + Q * Y + Q • x ] — d2L (4. 3x* = Combining Equations (4.Q f « c = x Note that ^ .921) (4.Qf+Q. 3x* L = ^ "^ becomes (4.e.913). 21) yields L * = Q ' L " ^ ^ Q'Q 1 ^ * * + (4.920.923b) (4. The rateofdeformation y Y = L +L T (4.923a) (4. we obtain Qf«x* .b) Using Equation (4.918d) follows since Q is orthogonal (i.919a.916).= Q t.Qt = + . Thus (ds) 2 is an objective quantity.920) = 9x* Inverting Equation (4. The length of an object in nonrelativistic mechanics does not depend on the motion of the observer.Q + Q.386 ADVANCED MATHEMATICS Equation (4. Q ' Q = 1 ) .
I 3 = det T .926) is its characteristic equation.924) The trace of tensor T (tr T ) is the sum of the diagonal elements.VECTOR AND TENSOR ANALYSIS 287 = Q . Thus y is an objective tensor and is an admissible quantity in a constitutive equation. The scalar product of u and u (u1 Uj) is an invariant. The equation of motion holds only relative to an inertial frame of reference.is a scalar and is an eigenvalue of T . Q t + i_JQ. The equation of motion is not objective. y .1 2 X + 1 3 = 0 where I2 = tr T . This can be expressed as Tu = Xu where A . I2 = l [ ( t r I ) 2 . These invariants arise naturally when we consider the eigenvalues and eigenvectors of a second order tensor. The functions Ij. TensorValued Functions As indicated earlier.923d) (4.t r ( T 2 ) ] . For second order tensors. A nonzero vector u is said to be an eigenvector of a second order tensor T if the product T • u is parallel to u.] = 0 On expanding the determinant. . They have an important role in tensorvalued functions. we have three principal invariants.926) (4. the invariants of a tensor are scalar quantities which remain unchanged when the coordinate system is transformed.924) is det[T^I.Qtj (4.923e) = QY'Q1" Equation (4. Not all equations in physics are objective.923e) follows since Q is orthogonal.X3 + I x X2 . I 2 and I 3 are known as the principal invariants of T . The condition for the existence of a nonzero solution to Equation (4. (4. and Equation (4. we obtain .925) (4. because velocity and acceleration are not objective quantities. If the motion of the earth can be neglected then a frame of reference fixed on the surface of the earth can be considered to be an inertial frame.
.928c) Further if cp is a scalar function of T . . which in turn is a function of A.b) (4.2 I 2 III = t r T 3 = l"(6I 3 + 2 l J .930) . + a n T n where a 0 . Then I = A. We also need to consider tensorvalued functions of T and we write S = F (T) or Sjj = Fjj ( T k e ) (4.j.927e.b) If S can be expressed as a polynomial in T .. we have S = ao: + a1T + a2T2+a3T3+.. (4. X2 and Xg as its diagonal elements.. . That is to say. a 1 ? .f) If T is a symmetric tensor. That is to say.927c. X2 anc* ^3 are its eigenvalues. if Xy. T satisfies Equation (4. Every matrix satisfies its own characteristic equation.931) (4. T can be transformed to a diagonal matrix with Xy.d) (4. then its eigenvalues are real and it is diagonalisable.I 2 T +I3J = 0 Expanding F as a polynomial in T .928a) 1 2 = XXX2 + Xfa + X2X?) 13 = \{kfa (4. The CayleyHamilton theorem can be stated as follows.j + X2 + X3 (4. then (p is a function of the invariants of T .388 ADVANCED MATHEMATICS Another set of invariants is defined as I = tr T = Ij II = tr T 2 = I 2 .T 3 + I 1 T 2 .928b) (4.6 I 1 I 2 ) (4. a n are constants. X^ and X3 in the case of a symmetric tensor T .926) and substituting T for X yields . then the CayleyHamilton theorem can be used to simplify the representation of S .927a.929a.
that is to say. \_ . we have S = Q 5 Q t = QF(T)Qf (4.  = F(I) (4. all powers of T higher than two in Equation (4. We note that T 2 may be written in terms of T .930) by T .913). I_ and I 1? I 2 and I 3 . Thus Equation (4.b) (4. Similarly.932) is a representation of S without requiring that it can be expressed as a polynomial in T . Since F is the same in both coordinate systems.930)]. T . on rotating the axes. we find that T can be replaced by T . and the three invariants of T .932) In continuum mechanics.936) and the inverse of T I T "M if it exists. (4. I premultiplying Equation (4. F is the same in all Cartesian coordinate systems.VECTOR AND TENSOR ANALYSIS 289 Using the CayleyHamilton theorem [Equation (4. and S and T are symmetric. T . 3j. On . we obtain (4. Thus.935) If F is isotropic.T 2 + Ii T . and P 2 are functions of the invariants of T .931) simplifies to § = P0I+PiT + P2T2 where Po.931) can be replaced by T 2 .933c) T =QTQt where Q is the orthogonal matrix defined in Equation (4.1 2 I + I 3 T 4 = 0 Thus an alternative representation of S is  = Tol+ Yll +Y11'1 (4. Equation (4.935) defines an isotropic second order tensorvalued function F .937) .934) Combining the two sets of equations. it is not uncommon to restrict attention to isotropic materials and F is then an isotropic function.933a. (4. we deduce Q F ( T ) Q1" = F I Q T QtJ Equation (4.
.1 has been dropped since we are considering the deviatoric stress. x is a tensorvalued function of y and if we assume that x can be expanded as a power series of y we obtain Equation (4.938a) + ¥ 6 ( T ? T 2 + T2T]) + V7(T1II+ T^T1) + V8 (T5T1+ lllfj (4. A Stokesian fluid is a fluid whose deviatoric stress tensor t depends on the rateofdeformation y . T 1 and T 2. A Stokesian fluid is isotropic. t r ( j j ) . I 2 and I 3 . Equation (4. t r (  ? T 2 ) .939) representation of t . Obtain a representation for T ..938b) The quantities \/ 0 ..932) which. t r f T i j j ) .390 ADVANCED MATHEMATICS where y 0 .92.939) is an exact (4. The results for a function of an arbitrary number of tensors can be found in Truesdell and Noll (1965). T2) = V 0 I + V i T 1 + V2T2 + V3T? + V 4 T2 + V 5 ( T 1 T 2 + T2i) is an isotropic (4. Yj and y j are functions of 1^. The ten principal invariants are tr(Ti). S = F j T j . the first invariant of y is zero because the . If the fluid is incompressible. Example 4. we have increased the number of functions from three to eight. (i = 1. T 2 and their products. in this case. is written as I = P1Y + P2Y 2 The term P o . The number of arguments of each function has increased from three to ten.2) tr(T?jj) tr(TiT2). In the case where F function of two symmetric tensors. t r f j j ) . In extending from one tensor to two tensors.. \/g are functions of the invariants of T j . Since both x and y are symmetric. The function F can be a function of more than one tensor.
Rivlin obtained Equation (4. Thus. v is a conservative field. deduce that curl v = 0. Are they better prophets than Lord Kelvin? PROBLEMS 1 a. Ifv = 2yzj_xyj+xzk (i) (v«V)cp. (iv) (v»V)V<p y«(V<p). cp = ax + by + cz cp = ax2 + 2bxy + 2cz2 (p = f(r). Determine the acceleration potential cp.939). If v x a = q x a. It is thus not surprising that Equation (4. If the magnitude of a vector a (t) is a constant. 5 a. 1967) did not believe vectors would be of the slightest use to any creature. 3 a. determine the relation between v and q. Determine V cp in each of the following cases (i) (ii) (iii) 4b. 2a. the third invariant is also zero. in shear flows. such as shear flows. we observe that Stokes proposed his constitutive equation in 1845. calculate the following (iii) (vxVJcp. (ii) and cp=xyz. It is perhaps appropriate to close this chapter by observing that Lord Kelvin (Crowe.939) is known as the constitutive equation of a ReinerRivlin fluid. show that a (t) is perpendicular to da ——. r2 = x 2 + y 2 + z 2 Answer: r f'(r)/r Show that Y (v • v) = 2 (v • V) v + 2 v x curl v If v = V cp. Hamilton stated the CayleyHamilton theorem in 1853 for a special class of matrices. As a matter of historical interest. that is to say.VECTOR AND TENSOR ANALYSIS 391 equation of continuity has to be satisfied. but it was only in 1945 that Reiner combined both results to obtain Equation (4. If v is a conservative field. show that the acceleration defined by a = == + (v • V) v at is also a conservative field. . In some flows.939) two years later without requiring the polynomial approximation. Can we imagine a presentday physicist not using vectors at all? Many engineers believe tensors are of no use. generalized by Cayley in 1858. Pj and P2 can be functions of the second invariant of y .
z = 0. 1. Answer: 0 7a.292 ADVANCED MATHEMATICS 6a. (c) y = I. y > 0. 4. evaluate y = r sin 0 .1 (ii) dxdy Answer: 7i/12 [Hint: sketch the area of integration. if necessary. along the 4 lines given by (a)y = 0. evaluate the following /•A (i) I v • dr Answer: 42 Jo /•A (ii) I v x dr Jo Answer: ^r. By inverting the order of integration. z).i . z = t 3 from the origin 0 to point A (2. 1). By means of the transformation x = r cos 0. (d) x = 0. y = t2. Answer: K (iii) i) xy ds along the sides of the square of unit length.8 j + ^. 0 < x < l .k 5 . If v = (x. 0 < x < 1. 8). Evaluate the following (i) M I (x + y + z)ds Jo along the line joining the origin 0 to the point A (1. That is to say.3 along the curve given in parametric form by x = t. ( b ) x = l . Answer: 3V3~/2 (ii) f (x2 + y 2 )ds Jc along the semicircle x 2 + y 2 = 1. split the area into two parts. 8b. 0 < y < 1. 0 < y < l . y. evaluate (i) I I Jo Jx r1 r1 I I Jo )y v dydx y 3 (x2 + y 2 ) Answer: 2 i n 2 .] 9b.
Hence.P10b. Jo where a is a constant. y > 0) of the xyplane. 384) tZ FIGURE 4. y =0. Evaluate .VECTOR AND TENSOR ANALYSIS i££ jf e( x2 + y2)dxdy A where A is the positive quadrant (x > 0.x j +y 2 k and V is the region bounded by the surfaces x = 0.] Answer: (128. 24. y = 6. z = 4. deduce the value of (i) I" e^dx. [Hint: V is shown in Figure 4. z = x 2 . Evaluate ///ydV (ii) f°° Jo e^dx Answer: (i) ±&. Volume of integration S is the complete surface of a cube whose sides are of length 2a. (ii) irV — v where v = 2xzi . 10b.P10b lla.
(1960) obtained the equation of continuity for a binary mixture by considering a volume element Ax Ay Az fixed in space.1 j k . The volume V is enclosed by a surface S. 12a.a d S s O O T where v = x i + y j + z k and S is the surface of a sphere of radius a. (i) v = xi + 4yj + zk . Using the Jv mass balance and the divergence theorem to transform the surface integral to a volume integral.440) is equivalent to Equation (4. Bird et al. The output of A across the surface at J\ is I n A • n dS. v = xi + y j + (z . 13b. the volume V is bounded by the coordinate planes and the plane 2x + y + 2z = 6 in the positive octant. Answer: * ^ a Verify the divergence theorem in each of the following cases. by evaluating both the volume and surface integrals.375) is identical to Equation (4. dV. V is the region occupied by the circular cylinder x 2 + y 2 < 1. The time rate of change of mass of A in V is ^ — I pA dV. deduce the equation . n A (= p A v A) is the mass flux vector and n is the unit outward normal to S. We can equally consider an arbitrary volume V fixed in space.394 ADVANCED MATHEMATICS (i) (ii)  J ( x 2 + y2)dS s J I (xi + y j + z k ) » n dS s where n is the unit normal. The rate of production of A by chemical reaction in volume V is I r.423). where pA is the density of A. Use the divergence theorem to evaluate Answer: (i) * * Oa t (ii) 24 a3 ff v . (ii) 14a. 15a.341). Show that Equation (4. Show that Equation (4. 16b. z = ±l. n is the unit outward normal to S.
Write down the set of direction cosines which corresponds to this rotation. Calculate the covariant and contravariant base vectors of the elliptical coordinate system. 0 < t. x 2 . 6) when referred to Oxyz. y. find x. Explain your result. The set of axes Oxyz.653) and the fact that in a rectangular Cartesian coordinate system the metric tensor is the Kronecker delta 5ij. by using the results obtained in Problem 19a. Using the summation convention. T. Let Tjj be a covariant tensor of order two. z = 3. o 1 [x~ 2 _s 2 0 i 2 0 o 1 y z If the coordinates of a point P are x =1. z. 5. z) is x 1 = cosh 4 cos r). 3. x3 = z . Calculate the sum of the products ujVi and UjVj. 4) and (4. n <r\ <K. obtain their components when referred to O x y z . . x 2 = sinh \ sin r). 18a. Calculate Vx 2 + y 2 + z 2 and V x 2 + y 2 + z 2 . <«. Verify that uiv i = u i v i . v<j> is a quantity whose tensorial properties are not known. This process of setting a superscript equal to a subscript is known as the process of contraction. But the inner product Tij v<j> is known to be a covariant vector (tensor of order 22a. If the vectors u and v have components (2.655). z) coordinate system considered in Problem 19a. The transformation from the rectangular Cartesian coordinate system (x J . Calculate the metric tensor gij and its dual g rs for the (£. Let m = n in Equation (4. 21a. r\. the direction of rotation being from the xaxis to the yaxis. Hence.VECTOR AND TENSOR ANALYSIS 231 17a. Is it an orthogonal coordinate system? 20b. °°<Z<°°. The set of axes Ox y z is obtained by rotating the set of axes Oxyz through an angle of 60° about the zaxis. 19a. show that "xi y z = r i 2 n. x 3 ) to the elliptical coordinate system (£. Ox y z are as defined in Problem 17a. Determine the geometrical shapes of the ^ and r\ coordinate curves. Also obtain gij by using the transformation given by Equation (4. y = 2. show that T nn is a scalar.
0) to the Cartesian coordinate system is given by x 1 = r cos 9.P23a Fourroll mill . Obtain the physical components of the velocity in the polar coordinate system. V 2 = CX 2 where c is a constant. That is to say. Show that v<j> is a contravariant vector. The Taylor (1934) fourroll mill consists of four infinitely long cylinders immersed in a fluid. the fluid is drawn in at A and C and is expelled at B and D.P23a. cr sin 0 cos 0] o  f 'o —A 2 c —„.C X 1 . as shown in Figure 4. Each cylinder rotates in the direction opposite to the two adjacent cylinders.396 ADVANCED MATHEMATICS one). The axes of the cylinders pass through the corners of a square as shown in Figure 4. The velocity components are Vl = .cos 2 0). FIGURE 4. The stagnation point O is taken to be the origin of a rectangular Cartesian coordinate system Ox 1 x 2 . v<j> transforms as given by Equation (4. 23 a.P23a. x 2 = r sin 0.633). Answer: [cr (sin2 0 . This method of ascertaining whether a quantity is a tensor or not is known as the quotient law. The resulting flow field is a point of stagnation at O. The transformation from the polar coordinate system (r.
the volume integral may be written as /•I rn fin I I I u u r z sin 6 dr d0 d<b h k Jo Show that the volume integral is equal to ^JL j . In modelling a vibrating jet. may be written as V« = v(0) 2 dJ+Ul(r'z)COS0 = ^(rui)sin0 v(z) = ~ w ( z ) The covariant component of the rateofstrain tensor y. where u is a unit vector. Appendix E. x 2 = r sin 0. (1987). in the cylindrical polar coordinate system (r. In the sperical polar coordinate system (see Example 4. 0. le el Ir e l The transformation from the Cartesian coordinate system (x 1 . 0. z) system is orthogonal and the metric tensors are gn = g zz = 1.i + Show that the physical components of j (denoted by A j in the original paper) are . 26b. (1993) assumed that the physical components of the velocity field. geg = r2.is defined as Yij = v i.VECTOR AND TENSOR ANALYSIS 397 24b. x 3 ) to the cylindrical coordinate system (r. it is required to evaluate the volume integral of the dyad u u. The (r.j v j. z) is x 1 = r cos 0. z). Chan Man Fong et al. Determine them.41). 25 a. x 2 . In Bird et al. z) are / \ and / \. x 3 = z. Show that the only nonzero Christoffel symbols of the second kind corresponding to the cylindrical polar coordinate system (r. 0. over a unit sphere. 0.
cosG dz dr . show that for an orthogonal coordinate system.(r L + V . div v may be written as fh. j = ° Show that .4 9vj (l) f dq dqJ (ii) 29a. (iii) Tij vJ 30a. d u 1 r.=^ + =i cos6 2 dz2 dz .+ —L) sine 3r3z 3zy 2dw dz i ^ 2 dz2 27a.V.733d).( r ^ I + ^I)Sine 3r3z 3 z y r d2W du.o^ divv = —L— (h 9 hoV n ^ + ~ h l h 2 h 3 [ a q l U 3 W) 3 q 2 l l 3 (2)' 8 q 3 U 2 0)) where gi{ = h2. e1J (v: : . Show that for any vector v 5vi Vj • . ^cose 3z Starting from the definition of div v given by Equation (4.(r .) sinG ^r2 dr + / d U1 du.V: . = . The symmetric stress tensor Ttjj satisfies the equation K=ij.^ J . 28 a. If Ty is a symmetric tensor. show that EijkTy = 0.h 9 v.hc.vns) + (h.N . Calculate the physical components of grad cp in the cylindrical polar coordinate system (see Problem 25a). 31a. c o s e dz 3r . j . . (ii) gij VJ . 33b. Is it a covariant or a contravariant vector? Write the covariant derivatives with respect to qk of each of the following (i) gi j vJ. 32a.(r L + ^L) sinG 3r2 dry aw _ 2 _ J . :) is a vector. ^ + 2 ^r1.398 ADVANCED MATHEMATICS dw . 2 Find V (p in the cylindrical polar coordinate system.
I' 34b. I2) is a function of the first two principal invariants of a second order tensor T.• i.• J is objective.• + V. Working with Cartesian components. Verify that the Christoffel symbols cancel. v* are the velocity components and Y.T rs) .+ ai^^ 6 rs.J J. By considering the transformation of coordinates given in Problem 17a. (n) •ij 3Y ij v r?i = —— + vss3Y —ij (2) dv 1 y . show that dty dty = dty 1 \ 3T^ a!7 6 . The derivative defined in (ii) is known as the Oldroyd contravariant upper convected derivative.] 36a. 37b. Show that 35b. The rateofdeformation tensor y has been shown to be objective. (ii) 6 E Compute the principal invariants of y for the two flows. show that if Ty is an isotropic tensor.• = rU V.S j SJ 3v j v ii • u. The rateofdeformation tensor y for (i) a simple shear flow and (ii) an uniaxial elongational flow are given respectively by (i) TO 1 0" Y= T 1 0 0 L0 0 0y = e (ii) . The scalar function ty (I\. at ax s dxs dxs [Note that in (ii) we have ordinary partial derivatives and not covariant derivatives as in (i).VECTOR AND TENSOR ANALYSIS 222 1 j " TCij yij dV = j v^ijnJdS V S where V is the volume enclosing the surface S. (i) j— = ^— + v s y1J s is not objective.H °° 010 [0 0 2 Answer: (i) 2 "f. 8jj. it is necessarily of the form A.
.
E. is one in which there is only one independent variable and all derivatives are ordinary derivatives.). which we denote by x 1 and x 2 .E.CHAPTER 5 PARTIAL DIFFERENTIAL EQUATIONS I 5. we have considered ordinary differential equations (O.E. x 2 ).D. A P. A partial differential equation (P.1 INTRODUCTION In the first two chapters.D. we consider only two independent variables . c is the concentration. Most processes that are of interest to engineers and scientists take place in a two or threedimensional space. Other examples of P.E. we shall see that. The equation describing the vibrations of a string can be written as —T = c ^ (51D at 2 ax2 where y is the displacement of the string from its equilibrium position. we have seen that to determine the velocity field of an incompressible and irrotational flow. The diffusion of a material in a homogeneous medium is governed by the equation ^ = « ^ (512) 2 at ax2 where a is a constant.) is one in which there are two or more independent variables and the derivatives that occur in it are partial derivatives. and their solutions will be given in this and the next chapter. and c is a constant. We denote the unknown function by u (Xj. in mechanics. The number of independent variables is usually more than one and the equations governing these processes are partial differential equations. can be written as . In Chapter 10. In Chapter 3. x is the coordinate of a point on the string.D. we have to solve Laplace's equation. An O.E. which is a partial differential equation. t and x are time and position as in Equation (5. For simplicity. t is the time. time may also be involved.11). Hamilton's equations of motion are given by a set of partial differential equations. Frequently.D.D.
the coefficients of u and its derivatives are only functions of the independent variables Xj and x 2 ..E. that is to say. .13) where f is a function. . 2) are of second order and are linear.+ a 3 (x 1? x 2 )u = g ( x 1 > x 2 ) If g is zero.402 ADVANCED MATHEMATICS f( Xl . Equations (5.^. The order of the equation is defined to be the order of the highest order partial derivative in the equation. the equation is linear.u.^._ . is an equation in which there is no term which is a function of the independent variables only.11.= ° C52"1) where &i and a2 are given functions. ..+ u r . (5.) = 0 dXj dx 2 9xj (5.15) Just as for ordinary differential equations.D. A first order linear n o n homogeneous partial differential equation can be written as a ^ x j . „ (5.=0 dXj dx 2 . An example of a nonlinear equation is 3u (du \ n v— + 3— = 0 dxj dx 2 / .16) In this section. x 2 ) ^ .x 2 . Method of Characteristics Consider the linear homogeneous equation which is written as al(xl'x2)^~+ a 2(xl'x2)^. it is nonlinear.^. a homogeneous P. An example is 3u 3u _ j . If f is linear only in the highest order partial derivatives. 5.14) If the equation is neither linear nor quasilinear. we present several methods of solving first order partial differential equations. the equation is quasilinear.2 FIRST ORDER EQUATIONS (5.. If f is linear in u and its derivatives. .+ a 2 (x l 5 x 2 ) ^ . the equation is homogeneous.
31.x2) (5. I".22a) (5.x 2 ) a2(x1. . u is a function of one variable s and we can write du ds 3u dx. = a 1 (x 1 . .25) Along Fj. Geometrically. b. Solve the equation g. 3xj ds 3u dx9 3x2 ds (5.21.PARTIAL DIFFERENTIAL EQUATIONS I 403 The solution of Equation (5. can be written as ^ al . we need to know the value of the function at a given point.24) is u = C where C is a constant. additional conditions need to be prescribed. defined in parametric form by [see Equations (4. we prescribe the value of the function u along a curve.= 0 9xj given that l (5.^ a2 (526) The solution has an arbitrary constant and the characteristics form a oneparameter family of curves. u is a constant and Fj is a characteristic curve.27) dx2 . Example 5.+ x. Let Fj be a curve in the (x1? x 2 ) plane. we regard u as a surface in the (xj. F 1? F 2 . For a first order ordinary differential equation. b) which.22b) = Along Fj. 3) yields 411 = 0 ds The solution of Equation (5.23) Combining Equations (5.22a. x2) plane and the surface is generated by a family of curves. j± dx ^2. that is to say.24) (5. For a partial differential equation.. The characteristics are obtained by solving Equations (5. 15)] dxi .22a.21) in the (x ls x2) plane. (5. To obtain a solution. we need to impose an initial condition. by eliminating s.21) is a function u that satisfies Equation (5..
The values of u in the region bounded by the characteristics passing through the origin and (0. x 2 ) is any point in R.29a. We denote this region by R which is shaded in Figure 5.404 ADVANCED MATHEMATICS u (0. b)] £ =' • 517 = x> The solution is % .211). u is a constant and we write u = C The parabola will intersect y at the point (5. which is x j . the value of u is constant. Let (xj. x 2 ) = 1 + x 2 . On each of the parabolas.215) u = 1 + 1 (2x 2 xf) (5.10 > Combining Equations (5. x 2 ) is given by Equation (5.214a. 0 < x2 < 2 (5. 13. 2) can be determined.2 . X2 =2 fe^l2) (5.22a.x> ( 5 29b) (5 . 14b) yields C = 1 + 1 (2x 2 xj 2 ) Since (x"j.28.28) and combining Equations (5.b) The value of u along y is given by Equation (5.21. (5.28) The equations for the characteristics are [Equations (5. The equation of the characteristic that passes through (xj. we can replace it by (x 1? x 2 ) and the solution in R is (5.213) (5.21.21.212) xx = 0.2 x 2 = Xj 2 2x 2 = K On this parabola. The values of u are given along the x2axis in the interval 0 < x 2 < 2 and we have denoted this line segment by y in Figure 5.211) The characteristics are parabolas and are shown in Figure 5. b) yields xf2x 2 = K where K is a constant. x 2 ) be any point in R.216) .
+ f 2 (x 1 . we obtain 3O 9xj + (5. the conditions must be consistent. FIGURE 5.u) = 0 Differentiating with respect to xx and x2.PARTIAL DIFFERENTIAL EQUATIONS I 401 0 X.21 Characteristics F of the differential equation and the line y along which initial data are given • Note that y is not a segment of a characteristic F.x 2 . u ) j ^ . otherwise there is no solution.x2. That is to say. u ) Suppose the solution can be written in an implicit form as O(Xl.21%) . since u is constant on F.217) (5.218) 3O 8u_ 9u 3x2 = Q (5. u must also be constant on y. If y is part of F.219a) i * + ^ ^L 3x2 3u 8x2 = o (5. This method can be applied to a quasilinear equation which can be written as f 1 ( x 1 .= g ( x l 5 x 2 . x 2 . u ) ^ .
22a.220a.220b) du ^dx 2 = jrL / jr dx 2 du Substituting Equations (5. b) can be written as dx ^J.! * / £ dXj dXj (5. and u. in this case.221) is now a linear equation in O.x 2 . f and f2 are also functions of u and that the characteristics depend on u. The characteristics can also be given in parametric form and Equations (5.217).222b) yields the values of u along the characteristics and Equation (5.= f j ( x . b) into Equation (5. Note that. The characteristics of Equation (5. x 2 . However.223a) g =g(Xl.220a) (5. x 2 . If g is zero. we obtain 3cb f 1 3o z 3o =° (5.406 ADVANCED MATHEMATICS Assuming that ^— is not identically zero.u) (5.2. unless g is zero. . only Xj and x 2 are independent variables and u is the dependent variable).x2. .221) are given by ^ fl = ^1 f2 = dL § (5. j s not zero. We consider all three as independent variables (in the original problem.221) du l^+f2 5^ + g ^ dXj dx 2 Equation (5.b) The characteristics of Equation (5.217) is solved. Solving Equation (5.223b) (5. The initial values of u are given along the curve y and the solution (u) is defined along the characteristics that pass through y.= .u) (5.223c) Note that in this case ^U. it now involves three variables x t .222a). we deduce all p. as shown in Figure 5.222a. u ) dxo ^= f 2 (x 1 .217) are obtained by solving Equation (5.22. u is a constant along the ds characteristics.
I1) implies that ^ + A(pvx) = O (5. .224) Our driving experience leads us to believe that vx is a decreasing function of p. Model the movement of cars at a traffic light We model the automobile traffic on the road by considering the flow to be onedimensional and the road to be the xaxis. we have a traffic jam and v x reduces to zero.226a.22 Characteristics Tj of the differential equation and the initial data curve y Example 5.PARTIAL DIFFERENTIAL EQUATIONS I 401 U / r ' FIGURE 5.d) where L is a typical length. A simple function that describes this situation is vx = v m d . If p is zero.224. a block).p / p c ) We nondimensionalize Equations (5. we can travel at the maximum allowable speed v m and if p reaches a critical value p c .b. 25) by introducing (5.225) u=X vm < J = ^.e. which can be the distance between two traffic lights (i. We denote the local density (number of automobiles per unit length) by p (x) and the velocity by v x .c. The principle of the conservation of the number of cars (Equation A. Pc t =V ' L ^= T L (5.22.
227a) (5. we deduce (5. The road in front of the traffic light is free of vehicles and behind the traffic light the automobiles are bumpertobumper.228) We now assume that at the origin (£.b) The speed of propagation c of the state c is given by c = ji From Equations (5.2 .232) . = 0) there is a traffic light which turns green at T = 0. The density a initially can be written as a (£. This can be shown by differentiating c along a characteristic curve.3Oa > (5.230b) dt _J = i _ 2 o We note that the slopes of the characteristics change sign at a equals to a half.227b)  + (l2o)g0 (5. b) yields (5.829a. £ > 0 The characteristics of Equation (5.23la. The factor (1 . a is a constant (g = 0).2a) can be interpreted as the speed of the propagation of the state a. Also along each characteristic. b).23la. We have da = 0 = ~ dx +   d£ (5.227a.224.228) are given by f or = T^ ( 5 .0) = { ' ^< (5.4Q8 ADVANCED MATHEMATICS Equations (5.b) 1 0. 25) now become a7 + ^ M ) = 0 u = (lo) Combining Equations (5.
0 0 £ FIGURE 5.23. In II. the slopes of the characteristics vary from one to minus one and this implies (Equation 5. and III as shown in Figure 5. For negative values of £. in I and in. That is to say. the values of a are zero and one respectively.228).23 Characteristics and path of vehicles (. The xaxis corresponds to a value of o equals to one half. T) upper half plane into three regions which we denote by I.) for the continuum traffic model . (5. a is zero and we deduce from Equation (5. The characteristics Fj and F2 divide the (£...227b. u > c.233b) is obtained from Equation (5. For positive values of %. the slopes are 1 and we denote by T2 the characteristic that passes through the origin. 33b). We now draw the characteristics which are shown in Figure 5.233a) = (l2o) Equation (5.23. we find that the velocity of the vehicles is greater than the speed of propagation of o. II.233b) Comparing Equations (5.230b) that the slopes of the characteristics are one.PARTIAL DIFFERENTIAL EQUATIONS I 409 (5. On the characteristics. The initial conditions are given along the ^axis. Let Fj be the characteristic that passes through the origin. the value of o is a constant.23Ob) that the values of a vary from zero to one.
410
ADVANCED MATHEMATICS
Let D be any point on a characteristic in II, as illustrated in Figure 5.23. Integrating Equation (5.230b) and imposing the condition that the characteristic passes through the origin yields £ = (1 2a) x From Equation (5.234), we deduce that the value of a on the characteristic through D is o= ^(l) (5.235) (5.234)
The equations for Fj (a = 0) and F 2 (a = 1) are respectively £ = T S = x (5.236a) (5.236b)
Having obtained a description of a, we now consider the path of an individual automobile. In this model, the car at the origin takes off at maximum speed since the state a associated with it is always zero (along Fj). The car initially at A (£ =  a ) in Figure 5.23 is associated with the condition a = 1. That is to say, it cannot move. It has to wait a finite time for the car density a to become less than one, before it can move. This is associated with point B in the figure. The time it has to wait can be obtained from Equation (5.236b) and is x = a (5.237)
From now on as it moves forward in space and time, the value of a is given by Equation (5.235). The speed u of the car is given by Equation (5.227b) and is written as
(5.238a,b)
Note that, in Equation (5.230b), — refers to the slope of the characteristics which are defined by the dx J P.D.E. [Equation (5.228)]. In Equation (5.238b),  ^ refers to the speed of the car and is defined dx by the trajectory of the car, as shown by the broken line in Figure 5.23. Note also that, in the case of the first car, a = 0 and u = c. That is to say, the slope of the characteristic is also the velocity of the car. Integrating Equation (5.238b) yields
£ =
X+KVIT
(5.239)
where K is an arbitrary constant.
PARTIAL DIFFERENTIAL EQUATIONS I
411
Imposing the condition that the vehicle passes through B [= (a, a)], Equation (5.239) becomes
£ = t2Va7
(5.240)
Equation (5.239) is the equation of a parabola and is the path of the vehicle shown as a broken line in Figure 5.23. The time required for the vehicle to reach the traffic light (£ = 0) from B can be obtained from Equation (5.240) and is given by x = 4a (5.241)
The last vehicle in the block (£, = 1) crosses the traffic light after five units of time (5a) and if the light changes in less than five units of time, the car has to wait for the next change. The typical path of a vehicle starting at point A can be summarized as follows. After the light turns green, it remains stationary (AB) for a finite time (a), after which it starts to accelerate to cross the traffic light after a further 4a units of time at point C. It continues to accelerate until it reaches its maximum speed asymptotically. • The quasilinear equation considered in this example can be used to describe the movement of glaciers and floods. Further details are given in Rhee et al. (1986); see Problem 3b.
Lagrange's Method
Lagrange's method is similar to the method of characteristics. Equation (5.217) is now also solved via Equations (5.222a, b). The latter equations are now referred to as the auxiliary (or subsidiary) equations. They can also be written as (5.242a,b)
dxj
fj
dxj
fj
Instead of defining the solution along the characteristics, in Lagrange's method the solutions of Equations (5.242a, b) are written in implicit form as O 1 ( x 1 , x 2 , u ) = Cj O 2 ( x b x 2 , u) = C 2 where C^ and C 2 are the arbitrary constants of integration. The functions Oj and O 2 are integrals of the auxiliary equations and are solutions of Equation (5.217). This can be shown by differentiating ®i to yield (5.243a) (5.243b)
412
ADVANCED
MATHEMATICS
0 = 9xj
dx t + —  dx2 + 9x 2 3u
du
(5.244)
Using Equations (5.242a, b), Equation (5.244) becomes  BO,
flT1
aXj
3$,
dx 2
9$,
(5245)
ou
+ ^ 7  ^ + 8 ^ =°
From Equations (5.220a, b), we deduce that Oj is a solution of Equation (5.217). Similarly, O 2 is a solution and a general solution can be written as F(ObO2) = 0 (5.246) where F is an arbitrary function. The system of Equations (5.222 a , b) might not be easy to integrate. A possible method of solving the equations involves writing Equations (5.222 a, b) as follows dx 9 du X,dx,+Lidx ? + vdu dx, 7+ = j2 = — = l * 2 where we used the following relation
(5.247a,b,c)
r = J =
b d
a
r
Xa + lie
Xb + id
—
(5.248a,b)
We now choose the functions X, \x, and v such that Xf 1 +^if 2 + v g = 0 Since we are now dividing by zero, we have to require that A,dxj+Lidx 2 +vdu = 0 Comparing the exact differentials [Equations (5.244, 50)], we identify
X =
(5.249)
(5.250)
™±, dxj
n=^ I , 3x2
v  ^ I du
(5.251a,b,c)
Since X, L i and v are known functions of x l5 x 2 and u, we can integrate Equations (5.25 la toe) to obtain O j . A different choice for the set {X, Li, v} in Equation (5.249) will provide a different function O 2 .
PARTIAL DIFFERENTIAL EQUATIONS I
47i
The general solution will be given by Equation (5.246) which implicitly determines u as a function of Xj and x2. The arbitrary function F is determined by the initial conditions. Different choices of 3>j will generate different functions F, but the initial conditions will ensure that the solution is unique. Example 5.23. Chromatography is widely used as a method of separation of chemical species and references are given in Rhee et al. (1986). The simplest case is the chromatography of a single solute. Suppose that the solution flows along the zdirection through a bed with void fraction e at a constant superficial velocity v0. If c is the concentration in the fluid phase and n is the concentration in the stationary phase, a mass balance [Rhee et al. (1986) or Bird et al. (I960)] yields (5.252) We assume a linear relation between n and c and write n = Kc where K is a constant. Combining Equations (5.252, 53) yields (5.253)
vo^+[e + K ( l  e ) ]  ^ = 0
The initial and boundary conditions are assumed to be c (z, 0) = 0 , c (0, t) = c 0 H(t)
(5.254)
(5.255a,b)
where c 0 is a constant and H(t) is the Heaviside function. [Equation (1.178a, b)]. The auxiliary equations are dz _ dt _ dc v0 a 0 (5.256a,b) (5.256c)
where < x= e + K(le) From Equation (5.256a), we have d i = a. = p On integrating, we obtain t  p z = Cj
(5.257a,b)
(5.258)
414
ADVANCED MATHEMATICS
where C\ is a constant. From Equation (5.256b) , we deduce that i£ _ o => c = C2 (5.259a,b)
The general solution is F ( t  p z , c) = 0 or c = f(tpz) (5.260a) (5.260b)
Our aim is now to determine the form of the function f. That is to say, we want to know if f is an exponential function, a trigonometric function, etc. To determine the form of the function, we use Equations (5.255a, b). From Equation (5.255b), we determine that the unknown function f has to be a Heaviside function. The solution is given by Equation (5.261). The reader should verify that this equation satisfies the other boundary condition. From Equations (5.255a, b), we deduce that c = c0H(tpz) Transformation Method We illustrate this method by considering a first order partial differential equation which can be written as
al
(5.261)
( x l' X2> 57~ + a 2 (xi» X2) a~
+ a 3 ( x l' X2) u =
f
(xl> X2>
(5.262)
We make a change of independent variables and write £ = ^(x1,x2), Using the chain rule, we obtain r = r((x 1 ,x 2 ) (5.263a,b)
3L
9xj
=
f£ 1 1
9q dxl
+
£p .
dr\ 8x1
(5.264a)
3L .  »  L
dx2 dc, dx 2
+
 i !L
dT) 3x 2
(5.264b)
Substituting Equations (5.264a, b) into Equation (5.262) yields
PARTIAL DIFFERENTIAL EQUATIONS I
415
(5.265) We now choose £ and r\ such that we reduce the P.D.E. to an equation which involves only one derivative. One possibility is to choose % = xx and to require the coefficient of 3— to be zero. That is to say (5.266a)
a'!^7 + a
^=0
(52"66b)
This allows us to obtain r\ by solving
P = ^
dx 2 a2 Along each of the characteristics given by Equation (5.267), T is constant. Equation (5.265) simplifies to
&1fe,n)
(5.267)
  + a 3fe,TI) u = ffe,TI)
u =
(5.268a)
(5.268b)
or
^L
3^
+
£i
aj
1
ax
Equation (5.268b) can be solved by introducing an integrating factor I defined as
I = exp I 52 d^
Equation (5.268b) can now be written as
(5.269)
4(uI) = —I
3^ aj
(5.270)
On integrating, we obtain the solution to the problem. u = I
1
— Id£ + F(r) I) a l
(5.271)
416
ADVANCED MATHEMATICS
where F (r\) is an arbitrary function of r), which is determined from the initial (boundary) conditions. The solution in terms of x{ and x 2 is obtained via Equations (5.266a, b). The transformation from (xj, x2) to (£, r) is valid as long as the Jacobian
ii
8xj J = 9xj is nonzero.
ii
9x 2 (5.272) dx2
Example 5.24. Solve the problem considered in Example 5.23 by the present method. Equations (5.254) can be written as
0 1
§7 + v° If = °
<5273)
<«74>
(5.275a,b,c)
We identify aj to be a, a 2 to be v 0 , ^ to be t, and T is obtained by solving
£  VQ
Solving Equation (5.274), we find that r) is given by • H = f t " ^ 1 = t  p z = constant v vo / Equation (5.273), in terms of £ and TJ, becomes   0
(5.276)
The solution is c = F(TI) where F is an arbitrary function. Combining Equations (5.255a, b, 75b, 77) yields c = c0H(tPz) which is the solution obtained by Lagrange's method. (5.278) (5.277)
PARTIAL DIFFERENTIAL EQUATIONS I
417
Example 5.25. Solve the following equation using the three methods described in this section (5.279) subject to the condition u=2 on x 2 = xf (5.280)
In this example, Equations (5.222a, b) are dx, dx 9
=
du
•at= ~H
x 2 = Kxj
2^2
(5281a'b)
From Equation (5.28la), we deduce that the characteristics are (5.282)
where K is a constant. The characteristics are straight lines through the origin, with slope K. Combining Equations (5.28 lb, 82), we obtain
*± = —
X! 2Kxj 2
(5.283)
Solving Equation (5.283), we obtain along each characteristic u = KxJ + C where C is a constant. The constant C depends on the characteristic; that is to say, it is a function of K (= X2/X1). Using Equation (5.282), Equation (5.284) can be written as u = x 1 x 2 + f(x 2 /x 1 ) Imposing the condition given by Equation (5.280) yields 2 = xf + fCXi) We deduce
f(xj) = 2Xj 3 (5.287)
(5.284)
(5.285)
(5.286)
418
ADVANCED MATHEMATICS
Substituting f into Equation (5.285), we obtain u = x1x2 + 2(x2/xl)3 We now solve Equation (5.279) by Lagrange's method. From Equations (5.247a to c, 81a, b), we note that by choosing X = x2, we obtain x 2 dx : + Xj dx2  du = 0 Equation (5.290) is an exact differential and can be written as d(x1x2u) = 0 Equation (5.291) implies that x^u = constant (5.292) (5.291) (5.290) M  = xi> andv =  l (5.289a,b,c) (5.288)
The two solutions of the auxiliary equations are given by Equations (5.282, 92). The general solution [Equation 5.246)] can be written as
Xlx2u
= g(x2/Xl)
(5.293)
where g is an arbitrary function. Using Equation (5.280), we find, as in the previous method, that g is given by g(x 2 /xj) = ( x 2 / X l ) 3  2 It follows that Equation (5.288) is the solution of Equations (5.279, 80). We now use the transformation method. We introduce two new variables £, and r\ and choose ^ to be Xj. The variable T is given by Equation (5.266b) and in this example, it is (5.295) The characteristics of Equation (5.295) are ^ = ^ x2 x2 (5.296) (5.294)
PARTIAL DIFFERENTIAL EQUATIONS I
479
The solution of Equation (5.296) is
x9
 ^ = constant
xl
(5.297)
The variable r\ is given by Tl = xj/xj Equation (5.279) becomes, in terms of \ and T, „ du Jl \— = 2£ Tl Solving Equation (5.299), we obtain u = £2r+h(r) where h is an arbitrary function of r\. Equation (5.280) now becomes u=2 on 4 = TI (5.2101a, b) (5.2100) (5.298)
(5.299)
Combining Equations (5.2100, 101a, b) yields h (TI) = 2  r  3 It follows that the solution is u = 2 + £2rr3 (5.2103) (5.2102)
Reverting to the original variables (replacing £ by xj and r\ by x 2 /x!), Equation (5.2103) becomes Equation (5.271). As expected, all three methods yield the same solution.
420
ADVANCED MATHEMATICS
5.3
SECOND ORDER LINEAR EQUATIONS
Second order linear partial differential equations are frequently encountered in the applications of mathematics. They are classified into three types: hyperbolic, parabolic, and elliptic. Each type describes a different physical phenomenon. Hyperbolic equations describe wave phenomena, parabolic equations describe diffusion processes, and elliptic equations describe equilibrium conditions. We next deduce the canonical form of each type via a transformation of independent variables.
Classification
The general second order linear partial differential equation in two independent variables (xp x2) can be written as d u n^T
dXj
0
a
+
d u ai2, ^
dXjdx 2
0
3 u , 3 u , 9 u ., , + a 2 2 ^ T + b j  — + b 2 ^ — + c u = f( X l ,x 2 )
ox 2 dXj dx 2
0
,,<, ,^ (5.31)
where a^, a12, a22, bj, b 2 , and c are functions of Xj and x2. We simplify Equation (5.31) by changing the variables (xj, x2) to (£, T), in an attempt to reduce the number of higher derivatives. This transformation is given by Equations (5.263a, b). From Equation (5.264a), we have
9jj_
=
a ^ du_
_a_§_ T _ a _ /au\ d^_ _a_ (du\ dr\]
+ 5xi
B\
du_
ax?
=
ax? K
+
k Naxi
+9T1
\^)dxi\
+
ax? ^
3
a ^ r^_ /3uj d$_ + d_ id_u\ a ^ j
axj ^& ^arijax!
ari \arijaxj
a2u
= fli.) 2 ^JL + 2^i ^ [dxJ ^2
a X l a X l d^dr]
+ f i n f ^u_  a2^ du ( a2Ti a u
+
[dxJ
dy]i
9x2 a ^ + 9x2
an
(5.32b)
9 9
Similarly, we can compute 3xj dx2
and dx\
. Equation (5.31) now becomes
PARTIAL DIFFERENTIAL EQUATIONS I
421
d2u\a
(3M2+a ^ ^ + a fdM2l + ^H_[2a ^ ^ +a ( ^ ^ ^ L M a ^ J 123*i 3x2 + a 2 2 f e J j % ^ J 2 a ^ a ^ + M a ^ a ^
3x 2 3xj/
2 2 3x 2
3x 2
i 2
^^Xj/
1 2 3xj
3x 2
2 2 \3x 2 /
a u I" d2^
+—
a
d2^
d2^ , dt, , a^l 9u [ a2ri
+ a22l + b
a2n
—
l l  ^ +a 1 2 — T ~
lT~
+ b
2 — +^T
a
l l ^
+ a 127
+ a 2 2  4 + b 1 — + b 2 — +cu = f
9x 2 dxl dx2
(5.33)
The transformation (Xj, x 2 ) to (^, r) is not yet defined. As mentioned earlier, the aim of the transformation is to simplify Equation (5.31). We examine the possibility of defining a d2u transformation such that the coefficient of vanishes. This does not affect the generality of the
a2u
equation. That is to say, Equation (5.33) with the coefficient of = 0 will be equivalent to
a^2
Equation (5.31). An example is provided by Equations (5.46b, 21). That is to say, we seek the conditions under which it is possible to have
J&t
\dx 1 /
a12ii ii + ajiif  0
axj ax2 \ax 2 /
(5.34,
The solution of Equation (5.34) provides the desired transformation for £. Assuming that 3x 2 is nonzero, Equation (5.34) can be written as
a n (l / iiy + a i 2 fii / ii) + a 2 2 =
\oxi The solution is 0x2/ \dxi 3x 2 /
O
(5.35,
C i l / i l ^ = ~ai2± V a14a11a22
\a X l ax2/
2an
(5 3 6 )
422
ADVANCED MATHEMATICS
Consider the following cases, (i) ax > 4 a n a 2 2 In this case, we have two distinct roots and this is the hyperbolic case. Consider the curve E, = constant On differentiating, we obtain ^  dx, + ^  dx 9 = 0 3xj l dx 2 l It follows that (5.38) (5.37)
ii/il
9x] 3x 2
=
_d^
dxj
(5.39)
Substituting Equation (5.39) into Equation (5.36) yields dx^ d Xl
=
a12 + V a 1 2  4 a n a 2 2 2an
(5.310)
We have two possible solutions; that is to say, we have two possible curves (t, = constant), one corresponding to the positive root and the other to the negative root. Noting that the coefficient of is identical to that of —  if we replace S, by r, we choose one root for
the curve % (= constant) and the other root for r\ (= constant), since r\ is also a function of Xj and x 2 . These two curves are the characteristics of the partial differential equation. For different values of the constants, these characteristics span a twodimensional space. For a hyperbolic equation, we have two real characteristics (^ = constant, r\ = constant) and they are obtained by solving the equations dxo_ dxl dx^
dxj
=
a12 + V a 1 2  4 a 1 1 a 2 2 2all a12Va1224ana22
2all
(5.31 la)
=
(5 3llb)
respectively.
PARTIAL DIFFERENTIAL EQUATIONS I
423.
With the coefficients of —  and
as
^2
equal to zero, we obtain the canonical form of a
hyperbolic equation which is written as
J ^
+ P l
^
+
P 2 ^ + Yu  o
(5.312)
where P l 5 P 2 , Y > a n d o are functions of t, and TJ. Note that Equation (5.11) is hyperbolic. In Equation (5.36), we have assumed ajj to be nonzero; but if a ^ is zero, we can still solve for = — / = — from Equation (5.35) and finally obtain Equation (5.312). If both ajj and a22 are zero, the equation is already in the canonical form. (ii) a12 = 4 a n a 2 2 This is the parabolic case and the two characteristics are coincident. We choose £ (= constant) to be a characteristic and the coefficient of 32u to be zero. The coefficient of
, which we denote by B, can be written as
B = (2.n *L (i/l) + a , 2 fe fi/l)+Sl +2a22^}  t
(
X1
3xj v9xj 9 x 2 /
L^X2 R
3x2/
3x x J
3x 2 j 5x 2 (5.313)
Substituting Equation (5.36) into Equation (5.313), we have
B= 2aU11 3L(^2) + aU 1 2 fl(^2) + Hl + 2a 2 2 lL)i§dxl\2an) [dx2 \ 2au) dxj 22 dx2 j dx2
(5.314a)
= n(.A
9x2 \ 2a n
= 0
+2 J l  i
22/J
(5.314b)
(5.314c)
3x2
The canonical form of a parabolic equation is
424
ADVANCED
MATHEMATICS
^ + P
3
! + P 4 ^ + Yl»=°1
and
(5315)
where P3, P4, Yi>
a j are functions of £ and TJ.
Equation (5.12) is a parabolic equation, (iii) a 12 < 4 a u a 2 2 This is the elliptic case and the characteristics are complex. Since the coefficients a;; are real, the characteristics are complex conjugates [Equations (5.31 la, b)] and we write £ = a + if* ri = a  i p where a and P are real. Writing Equation (5.34) in terms of a and P and equating the real and imaginary parts to zero yields (5.316a) (5.316b)
3a\2
all + a12
8a 3a
+ a22 dXj dx 2
(3a\2 (d$\2 3p 3p /3p\2
= all + a12 + a 22
dXj/
\"X2/
\dxi/
" x i "X2
\3 X 2/
(5.317a)
2ail^L^+ai2(^^+*L
iL
+2
a 2 2 ^ ^ ,o
8x 2 9x 2
(5.3Hb)
3xj 3xj
\8xj 3x 2 3x 2 8xj
Rewriting Equations (5.33) in terms of a and P and using Equations (5.317a, b), we find
2
9 9
that the coefficient of
is zero and the coefficients of
and
are equal. The
3a 3P
canonical form of an elliptic equation is ^  3a2
+
3a 2
3p2
^ 
+
P
5
^ 5a
+
P
6
^
+
y2u=a
2
(5.318)
3p2
3p
where P 5 , P^, y 2 , and a 2 are functions of a and p. Laplace's equation is an elliptic equation.
PARTIAL DIFFERENTIAL EQUATIONS I
425
As usual, we require the transformations from (x l5 x2) to (£, r\) to be nonsingular; that is to say, the Jacobian J, given by Equation (5.272), is nonzero. In the elliptic case, we replace the complex functions £ and r\ by the real quantities a and (3 and this is implied in all subsequent discussions. Let the coefficients of—, , and be denoted by a 11? a 1 2 , and a 2 2 respectively. By
direct computation from Equations (5.272, 33), we obtain
a 12
4aila22
=
j 2 ( a 12
 4ana22)
(5.319)
The Jacobian J is real, its square is positive, and Equation (5.319) implies that the sign of a l2 ~ 4 a l l a 2 2 is preserved on a transformation of coordinates. That is to say, the type (hyperbolic, parabolic, or elliptic) of a P.D.E. is independent of the coordinate system. The quantity a l2 ~ 4a 1 1 a 2 2 can be a function of (x1? x 2 ), its sign can depend on the region of the (xj, x 2 ) plane. The type of an equation may change on moving from one region to another. Example 5.31. Reduce the equation ^ 3xj x2^=0 3x 2 (5.320)
+
to its canonical form. Equation (5.320) is elliptic in the region x 2 > 0, hyperbolic in the region x 2 < 0, and parabolic along the Xjaxis. In this example, Equations (5.31 la, b) become (5.321a) dx / —2 =  V  x 2
QXi
(5.321b)
We consider the regions (x2 > 0 and x 2 < 0) separately. In the upper half plane (x 2 > 0), the characteristics are complex. (5.32la, b), we obtain
o.
On integrating Equations
1/2
xj = 2ix 2
. . + constant
(5.322a) '
426
ADVANCED
MATHEMATICS
(53"22b)
x 1 =  2 i x ^ 2 + constant The solutions are ^
= X]2ix^/2
= constant
(5.323a,b) (5.323c,d)
Ti = x 1 + 2ix; / 2 = constant We deduce that
i/T
a = Xj ,
P = 2x2
(5.324a,b)
Using Equations (5.324a, b), Equation (5.320) becomes
^JL 3a 2
+
^L__I^=0 3(32 PW
(5.325)
In the region x 2 < 0, the characteristics are real and the solutions of Equations (5.321a, b) are ^ = n
= Xl2x^/2
= constant
(5.326a,b) (5.326c,d)
x 1 + 2 x ; / 2 = constant
On transforming to ^ and x\, Equation (5.320) becomes
^H_
+
* _ ( *L _ M = o
(5.327)
Equations (5.326a, c) show that the characteristics are parabolas. On the x^axis (x2 = 0), the equation is parabolic and Equation (5.320) reduces to the canonical form ~ 3xj = 0 (5.328)
We note that Equations (5.325, 27) have singularities at p = 0, ri = % (5.329a,b)
In both cases, the singularities correspond to the xjaxis (x2 = 0). That is to say, both hyperbolic and elliptic equations have singularities on the Xjaxis and it is across this axis that Equation (5.320)
4 METHOD OF SEPARATION OF VARIABLES This method relies on the possibility of writing the unknown function u (xj. We can extend this method to more than two variables. the partial differential equation reduces to two ordinary differential equations for fl and f2. The y component of the equation of motion is . 5. the boundary and initial conditions will have to change according to the region in the (Xj. x2) plane. Wave Equation An elastic string is tied at its ends (x = 0. as shown in Figure 5. it is at P 0 Q 0 .41.41. In a given problem. This will be addressed later. x 2 ) as the product of two functions f1 (Xj) f 2 (x 2 ). At equilibrium. We illustrate this method by solving the following equations. x = L).PARTIAL DIFFERENTIAL EQUATIONS I 427 changes from elliptic to hyperbolic. In many cases. The tension T acts along the tangents at P and Q. y Jk \s I iS/v^ + S^ 0 Po Qo L x FIGURE 5. We assume that the tension T in the string is uniform and the density of the string is p.41 Vibrating string The angles these tangents make with the xaxis are \/ and y + 8y. as shown in Figure 5. Consider a small element PQ of the string of length 5s. The displacement from the equilibrium position is y.
4lb) (5. (5.4lc) = Tsin(\f+5\/)Tsin\/ (5.= tan \> On differentiating.46b) is identical to Equation (5.T sin \/ = T5\/COS\/ + 0(5\J?>2 By definition (5.b) .44b) (5.11).42) (5. <• . (5.43c) 3y Equation (5.43a) = l+(i)2]5v .4lc) yields T 92y 9 32y d2y —].1 8s = 8x 1 + R ^ \dx) .43c.= 3. Similarly 2 1l/2 cos y = 1 + ll^l .44a) (5.45b) /c . .428 ADVANCED MATHEMATICS a2y p8s^at2 = T(sin\/cos 8\j/+cos\/ sin 8\/) . Its square and higher powers can then be neglected.45a) (5. we obtain — ¥ .4la) p . 5b) into Equation (5.46a.43b) (5.= c 2 — 3t 2 P 3x 2 3x 2 Note that Equation (5.Sx Substituting Equations (5.43c) is obtained by assuming that •— is small. 4b.8y (5.8x = sec 2 \f8\/ 3x 2 (5.
n2.2Y (5. This condition is written as y (0. so there is no displacement at these two points. The conditions given by Equations (5. we need to impose additional conditions. where x and t are independent variables.410) by XT yields 1 d 2 T = c 2 d2X = constant t t T dt 2 X dx 2 . £ dt = g(x) t=0 (5. Equation (5. We choose the constant to be . we obtain (5. 0) = f(x).411) Since the left side of Equation (5. b) are the initial (Cauchy) conditions.411) represents two ordinary linear equations which can be written as follows d2T 9 5_L+n2T = 0 dt 2 . t) = y (L.410) The function XT is not identically zero. These are expressed as y(x.48a.49) X^ dt 2 = c 2 ^fT dx 2 (5.412a) (5. The string is tied at its ends. b) are the boundary conditions and those given by Equations (5.* A *„ (5.b) The vibrations depend on the initial shape of the string and the speed at which the string is released.b) where f (x) and g (x) are given functions of x.412b) 2 ^A+n_^x = 0 c2 dx2 . so as to satisfy the boundary conditions. We now seek a solution of the form y = X(x)T(t) Differentiating and substituting into Equation (5.411) is a function of t only and the right side a function of x only.48a.PARTIAL DIFFERENTIAL EQUATIONS I 429 To complete the problem. each side is equal to a constant.46b). dividing Equation (5. t) = 0 (5.47a.47a.
413b) (5. b)] yields oo f(x) = ^ A s s i n ( ^ ) s=l oo (5. B n .414b) Note that we do not impose both C n and D n to be zero because we are not interested in the trivial solution (y = 0)..416) Assuming termbyterm differentiation to be permissible.E.430 ADVANCED MATHEMATICS The P.'s. Equations (5.) (5. b) form a SturmLiouville problem which is discussed in Chapter 2.417) Imposing the initial conditions [Equations (5. .47a.414a) (5.412a. b) depend on the constant n and can be written as Tn = A n c o s ( n t ) + B n s i n ( n t ) (5. C n . 7a.412b.415) where s is an integer.D. b) imply Cn = 0 sin(DjL) = 0 (5.48a.418a) g(x) = J s=l Sffi£ B s s i n ( ^ ) (5.418b) . in two independent variables has been transformed to a pair of O. The solutions of Equations (5.D. Using the principle of superposition. we express the solution as oo y = Y  A S COS fsS£l) + B s sin fsitct1)] s i n (SEX) ^•^ L VL / V L /J vL / s=l (5.E. and D n are arbitrary constants.413a) X n = Cn cos (If) + Dn sin (*f) where A n . Equations (5. we have h. 2 . From Equation (5.414b). we deduce (ILL) = s n ( S = l . = V sice [_As s=l sin (sEctj + B s c o s (s^ctjl sin to] (5..
ri = x .PARTIAL DIFFERENTIAL EQUATIONS I 431 The right sides of Equations (5. In this case.E.312)]. We note that we first impose the boundary conditions and then the initial conditions.'s and on integrating.46b) to its canonical form [Equation (5.421) . If the string is released from rest [g(x) = 0].c t) where F and G are arbitrary functions. b)] £ = x + ct. Its period in x is (2L) and in t.420a. x 2 (= t).422a) (5. It is necessary to use the superposition principle.46b) becomes 82v ^JL.D. the eigenfunctions are the trigonometric functions and we obtain a Fourier series.b) From Equation (5. is decomposed in two O. On differentiating y partially with respect to t. The P. The coefficients of the expansion are then determined.31 la.419a) Jo L Bs = . This procedure is necessary because the initial conditions are given as functions and not as constants. the coefficients B s are zero for all s.2 S7CC J (L g (X) Sin (Sffii) VL / dx (5. we obtain (5. it is I—).33) with xl (= x).c t (5.This is accomplished by introducing the characteristics.418a.= 0 The solution is y = F ( 0 + G Cn) = F (x + c t) + G (x . b) are the Fourier series of f (x) and g(x) and the Fourier coefficients are given by As = f L f(x)sinfesx) dx (5. that is to say. which are [Equations (5.E. to expand the functions given in the initial conditions in terms of appropriate eigenfunctions.422b) (5. we find that Equation (5.419b) The solution y is periodic both in x and t. we obtain constants and not arbitrary functions. D'Alembert's Solution D'Alembert proposed a solution to the wave equation by transforming Equation (5.D.
[f(x + ct) + f(xct) + l I Jxct g(Q<] (5.425a.c t)].426a) /•x+ct rxo = X [f( x + ct) + f(xct) + l I /•x+ct g(QdC + l j g(OdC] (5. we obtain F(x) = j .c G ' ( x .426c) .424b) yields 1 I g(QdC = F(x)G(x) 7x0 where x 0 is arbitrary. c).[f(x)^ I g(Qd£] Jxo (5.[f(x) + l J g(QdC] JX0 (5. From Equations (5.48a.425b) Substituting Equations (5.[f(x + ct) + f(xct) + ^ JXO g(OdCl •'XO g(OdC] (5.426b) = j .424a.425a) G(x) = i .424b) (5.424c) (5.422b) yields /•X + Ct iXCt y = j . b) into Equation (5.424a) (5. b)] imply f(x) = F(x) + G(x) g(x) = cF'(x)cG'(x) Integrating Equation (5.c t ) (5.422 ADVANCED MATHFMATICS  = cF'(x + c t ) .423) where the prime ( ' ) denotes differentiation with respect to the argument [(x + c t) or (x . The initial conditions [Equations (5.
on integrating Equation (5. we obtain arbitrary functions and these functions are determined by using the initial conditions.428b) The value of t is not fixed and Equations (5. for any a.c t) is negative. we observe that initially the shape of the string is given by f (x). they are reflected. then at time t the point x that corresponds to the same value of TI is Xj + ct. F is also periodic with period 2L.429) (5.47a. We note in this case that.428b) becomes F (a) + G (2L . From Equation (5. To do this.427) We need to extend the range of the interval when (x + c t) exceeds L and when (x . Thus f (x .ct) is a pattern moving to the right at a speed c without changing shape.c t ) = 0 F(L + ct) + G ( L .PARTIAL DIFFERENTIAL EQUATIONS I .428a. we consider the simpler case where the string is released from rest. The functions f (x) and g (x) are defined only in the interval 0<x<L (5. if initially (t = 0) the point xj corresponds to a constant value r. it is seen that this implies that g is zero. . b)] and from Equation (5. one moving to the right and the other to the left. The transformation rj [= (x . The solution can now be extended for all values of (x . and at the end points (x = 0 and x = L).430) shows that G is periodic with period 2L.431) Consider a function f(r). We deduce that. From Equation (5.c t ) = 0 If we replace L + ct by o. Similarly. we need to impose the boundary conditions [Equations (5. Equation (5.c t)] represents a displacement to the right at a speed c.422b).416).430) Equation (5. f(r) is constant.a) = 0 (5.431). G(a) = G(2La) (5. the original pattern breaks up into two similar patterns but only half the size (amplitude) of the initial one.428a) (5.c t) and (x + c t).421). For a fixed value of TJ.48b). Afterwards (t > 0). 4J3 This solution is d'Alembert's solution and is equivalent to Equation (5.431) represents the superposition of two waves. They are periodic and of period 2L. To examine further the physical significance of the solution. Equation (5. 29) are valid for any t and a.42. Equation (5. That is to say.426c) is then written as y = X[f(x + ct) + f(xct)] (5. we deduce F(ct) + G ( .)] to the left as shown in Figure 5. One pattern moves to the right and the other [f (£.
47a.426c) simplifies to yx+ct y(x.43. t) depends only on the interval (x . This is the domain of dependence and it is shown in Figure 5. 8a. . Motivated by physical considerations. It can be seen from Equation (5.426c) that the domain of dependence is the same in the case f ^ 0. boundary and initial conditions [Equations (5.c t) and (x + c t).t) = ± I Jxct g(C)dC (5.432) The value of y at (x. Equation (5. These conditions are sufficient and necessary to determine a unique solution. b. b)] have been imposed.42 Displacement of an initial pattern Next we consider the case where f is zero and g is nonzero.434 ADVANCED MATHEMATICS y" t =o 1 1 1 ^~ X y J t >o X FIGURE 5.
In this case. c is the temperature and a 2 is the thermal diffusivity.IV1) simplifies to ^A at = jc.12) also describes the conduction of heat in an isotropic medium.12) which is reproduced here for convenience i£ = a 2 — 3x 2 9t (512) Equation (5.PARTIAL DIFFERENTIAL EQUATIONS I 431 lx. &p& is the diffusivity. For simplicity. we write Equation (5. t is the time and x is the position. Under appropriate conditions.^ ^ A 3x2 (5. Equation (A. .t) L A • xct FIGURE 5.433) where c A is the concentration of A. Consider the diffusion of chemical species A in a binary system of A and B.43 Diffusion Equation x+ct Domaine of influence x Diffusion is a process by which matter is transported from one part of a system to another.433) as Equation (5.
The surfaces are kept at zero concentration at all times.12) subject to Equations (5.439) Bsexp(s27i2a2t/L2]sin^^ (5.434b. s = l . c) constitute an eigenvalue problem.. n = ^ a . c(O. we find that the partial differential equation leads to two ordinary differential equations.436a) is T n = C n e"11'1 where C n is a constant. we deduce An = 0 ..436 ADVANCED MATHEMATICS We consider the diffusion out of a plane sheet of finite thickness L. 34b. 2. Using the principle of superposition. t) = X(x) T(t) (5.434a) (5. The initial and boundary conditions can be written as c(x. The solution is Xn =Ancos~ + Bnsin^f (5437) Imposing the boundary conditions.0) = f(x).b. the general solution of Equation (5. c) is c = ^ s=l (5.436b) — ' + —X = 0 dx 2 a2 where n2 is a positive constant which is determined by the boundary conditions. They are d j + n2T = 0 2 9 (5. Initially the distribution of the diffusing substance is f(x).438a. Equations (5.t) = 0 . (5. c(L.436b.c) We seek a solution of the form c (x. .436a) (5.440) The Fourier coefficients B s are obtained from the initial conditions and they are given by . 0<x<L t>0 (5.435) Proceeding as in the case of the wave equation.c) The solution of Equation (5. t) = 0 .434b.
y) = T (L. x = L) are kept at zero temperature and the wall (y = 0) is kept at a temperature f (x). It decays more rapidly with increasing a in agreement with physical expectation.443) (5.445a. decaying exponentially in time.PARTIAL DIFFERENTIAL EQUATIONS I 437 Bs = f Wo f (x) sin S ? 2 L d x L (5. Equation (5. As y tends to infinity. y) = 0 T (x. y = 0. this is the first problem considered in detail by Fourier. T —> 0 as y —> 00 (5. y—>oo (5.444). as in the case of a vibrating string. and dividing by XY (* 0) yields .445c.d) The two dimensional heat equation is given by (see Appendix HI) Kma*lil + tl) where T is the temperature and a2 is the thermal diffusivity. x = L. as pointed out in Carslaw and Jaeger (1973). the temperature tends to zero.c. unlike the case of a vibrating string. At steady state.441) The solution is sinusoidal in x. Laplace's Equation We consider the steady temperature in a semiinfinite rectangular solid. substituting the resulting expression into Equation (5.443) reduces to ^ 1 +^ 1 = 0 dx2 dy2 (5.444) The walls (x = 0.442a. The boundary conditions are T (0. 0) = f (x).d) We employ the method of separation of variables and write T = X(x)Y(y) (5. Historically.446) Differentiating.b) (5. The solid is bounded by the planes x = 0.b.
we deduce fL B = ^ f (x) sin ^ ^ dx Lj0 L Substituting Equation (5.445c.450) Combining Equations (5. Using the principle of superposition.438 ADVANCED MATHEMATICS ^ .451). b) is.n2Y = 0 dy 2 (5. .447a) satisfying Equations (5. where s is an integer.448a. we require A n to be zero. as in the previous two cases.447b) The solution of Equation (5.445a.447b) is Yn = A n e n y + B n eny (5.445d).4 + n2X = 0 dx 2 ^ .452) L stt JO L I (5.447a) (5.b) To satisfy Equation (5.451) From Equation (5. T is given by oo T = X s=l Bs e" s7ty/L sin SfL (5. 50) yields oo f(x) = X B s s i n ^ s=l (5.450).453) L The temperature is sinusoidal in x and decays exponentially to zero with increasing y as required by the boundary conditions. we obtain (5.449) n = ^ (5. The solution of Equation (5.452) into Equation (5. Xn = sinnx.
52) . 0. the diffusion equation. the wave equation. 5. G)Z(z) (5. .5la. .PARTIAL DIFFERENTIAL EQUATIONS I m We have so far considered the dependent variable to be a function of two independent variables. we have defined the cylindrical polar coordinate system (r.= 0 dr2 r dr r 2 dQ2 dz2 We assume that u is of the form u = F(r. and dividing by FZ {it 0). (5.454c) Cylindrical Polar Coordinate System In Chapter 4 (Problem 25a). z).. and the Laplace equation are written respectively as v2 V u l 32u = — —7 cl 3t (5.5lb). The rectangular Cartesian coordinate system is not always the most suitable coordinate system to refer to. .. it is more appropriate to choose the cylindrical polar or the spherical polar coordinate system.b) Differentiating.454a) V u = — — (5. Using Equation (4.231)] and u is the dependent variable.454b) or at V u = 0 where V is the Laplacian operator [Equation 4. Problem 31a) that Laplace's equation can be written as J23 2 U 1 9u 1 92u 32u _ V u = — . we deduce (Chapter 4.5 CYLINDRICAL AND SPHERICAL POLAR COORDINATE SYSTEMS 2 (5. . We next solve Laplace's equation in these two coordinate systems. we obtain X ^ F 9r 2 + J_3F+J_^F rF dr r 2 F dQ s _J_d?Z = n 2 (55_3ajb) Z dz 2 where n 2 is a constant. We note that at steady state. both the wave and diffusion equations reduce to Laplace's equation. substituting the resulting expression into Equation (5.+ + — — + — . In many cases. In a threedimensional space.738).
7ab) Equations (5. b).y + m20 = 0 d9 (5.5lb) implies solving Equations (5. 0.ga) ^ . 8a.58b) Solving Equation (5.55) We now write F as F = R(r)0(6) Proceeding in the usual manner.510) . b) can be written as r 2d*R + r dr 2 2 dR_(nV dr + m2)R = 0 (55. which we denote by n 2 . dR _ R dr n V = _d 2 © d6 = m2 (5 5 . and z are independent variables.57a. We consider the simpler case where u is not a function of z. The constant n2 is zero and Equation (5. b).440 ADVANCED MATHEMATICS Note that the left side is a function of r and 8 and the right side is a function of z.58b) remains unchanged and its solution is 0 = AcosmG + B sinmG. That is to say. we consider the twodimensional case. and since r. where A and B are constants. From Equations (5.56) r.53a. we obtain two equations which can be written as — + n2Z = 0 dz 2 (5. that is to say m*0 (5. each side must be equal to a constant.58a) simplifies to r 2d2R1+rdR_m2R dr 2 dr = 0 (55_9) Equation (5.54) ^ 3r 2 + l^+l^n^F = 0 r dr r2 39 (5. we obtain r ! d^R R dr 2 + (5.54. The function 0 must be single valued.
Co. and Do are constants. 0 and R are given by 0 = Ao0 + B o R = C o i n r + Do where Ao.54. If the function has no singularity and is periodic in 0. Equation (5. F is given by F m*0 (5. and B m are constants. The solution /En r is the fundamental solution and is singular at the origin. Am. 9). Using the principle of superposition.513a) (5. If m is zero.PARTIAL DIFFERENTIAL EQUATIONS I 447 cosmB = cos m (9 + 2 s %) where s is an integer. we make a change of variable and write a = inr (i = / r T ) (5. Ao.515) Equation (5.59) is the Euler (unidimensional) equation and its solution is R = Cr m + Dr~ m .516) (5.513b) = Eo + X LAm cos m6 + B m sin mBj r m m=l (5.511) This implies that m must be an integer. The solution of Equation (5.58a).514) where E o . Bo. 8. where C and D are constants. To solve Equation (5.58a) becomes a 2d^R+adR+(a2_m2)R da 2 da = 0 (55_1?) . we have to solve Equations (5. (5. and Co are zero.512) (5. the constants D.54) is Z = Hcosnz + Gsinnz where H and G are constants. In the threedimensional case (m # 0).
z) = i. and n are to be determined from the boundary conditions.442 ADVANCED MATHEMATICS Equation (5. Its solution is R = LIm(nr) + MKm(nr) (5. K.520a. The constant M must then be zero.518) becomes (5. The coefficients A n m and B n m are given by . G.52 la) =\ X An0 sin n z f^ + £ Km cos m 6 + Bnm s i n **] ^ nz ^  (5.522) The right side of Equation (5. The general solution is (5.521b) yields CO OO f (0. n) = 0 u(a.517) is the Bessel equation with complex argument. L and M are constants. we assume the boundary conditions to be u (r.520c) to Equation (5. 6. there is no loss of generality in assuming that R(a) is one and Equation (5. z) = f(8.518) where I m and K m are the modified Bessel functions of the first and second kind respectively.n = l [A nm cos m6 + B n m sin me] sin nz (5. 6.519) Im(na) The constants A. b) imply that H is zero and that n is an integer. If the radius of the cylinder is a.521b) Applying Equation (5. To complete the problem. The function K m has a singularity at the origin. 0) = u (r.b) (5. z) Equations (5. H.522) is the (double) Fourier series of f (6. z).520a. We assume that the function remains finite at the origin.] £ A n0 sin nz + ^ n=l m.520c) u = X G n s i n n z ^  + X £ Gn sin nz[Amcosm9 + Bm sin me] n=l 0 n=l m=l m ^  (5. B.
and L is a constant. There are an infinite number of them and we denote them as Xnm (knl. 0) (5. If the height is L.523b) n Jo Jo We have assumed the height of the cylinder to be n for convenience. z)sinnz cosmG d0 dz (5.524a) d*R dr 2 + r dR dr + ( n 2 r 2 _ m 2) R = o (5 .2n f (6.524b) Equation (5.PARTIAL DIFFERENT!AT. nm ) = 0 (5.527a.. and u tends to zero as z tends to infinity. EQUATIONS I 44? A n m = 2.527a) implies that n a are the zeros of J m . 0.528a.53a.523a) K2 JO JO = \ I I f (0' z> s i n n z sin m e d e d z (5.I rn Bnm I .58b) remains unchanged. The remaining conditions are assumed to be u (a. 8) become 5^n2Z = 0 dz 2 r2 (5. 9. b) has to be changed. u (r.526) (5. the solution cannot be sinusoidal in z as given here. Xn2. . Equations (5.54. 0) = f (r. The sign of n 2 in Equations (5.) J m (na) = J m (A. Equation (5. We have already assumed that u is finite as r tends to zero. it can be scaled to n by introducing z 1= ^j..b) . If the cylinder is semiinfinite. and u tends to zero as z tends to infinity. The solution of Equation (5.525) Equation (5.524b) is the Bessel equation and the solution that remains finite at the origin is R = LJm(nr) where J m is the Bessel function of the first kind.b) (5.524a) that satisfies the condition Z tends to zero as z tends to infinity is Z = Ke~ n z where K is a constant. z) = 0 .
0) = 1 £ AOn Jo [hjQL) + £ n=l m.^f\ Jm (hf) [Am cos mO + Bm sin me] (5.n = l [Anm c ^ me + Bnm sin me] exp [. 0.527b) implies that f (r.530) is the (double) FourierBessel series expression of f (r.530) Equation (5.2u K n e x Pl—i~/ J ol^r/ n=l +X I n = l m=l Kn exp (. < >  ) was deduced in Example 4. \ / + I m.n=l [Anm cos m0 + B nm sin me] Jm pf) (5. The coefficients A n m and B n m are given by /•a f2n U \ 6 r dr d e Anm = ——2 f ( r .— 2 ^2a2^+i(^nm) )o )o f (r> 0 ) j jmi_ sin m e r d r d 0 (5.V F pvn ( \ nOz T \ i\ %0r \ n .^) Jm ( W ) (5.531a) Bnm = — .444 ADVANCED MATHEMATICS The general solution is given by 1 1 . 0 ) j m ^m!_J c o s m (5.529a) 1Y A exn ( ^a^ /) T K°r ~ 2 ^ On p\~ ol a / n=l / .53lb) Spherical Polar Coordinate System Laplace's equation in the spherical polar coordinate system (r.74 and can be written as a\ 3r 2 + 29u+i_^u r dr r 2 30 +Cote au+_L_a!u dQ r 2 sin 2 6 d <  > =Q (5532) r2 We assume that u can be written as . 6).529b) Equation (5. We recall that the Bessel functions are orthogonal and their properties are discussed in Chapter 2.
534a.b) — + m2O = 0 d<)2 r2sin20 (5.532) becomes 1 92F cot 9 9 F ) + + — —y"1" —. b) can be written as 2 (5.537c) into the standard form of a Legendre equation by writing x as cos 9 and we obtain .536) d_R dr2 + 2r dR _ ^ R dr = Q (5.534a. Equations (5.535b) We now write F as F = R(r)0(9) Equation (5. (5. Equation (5. u is independent of < >  . and we first consider this case.537a) and dy + cot 9 — + ^ 0 = 9 d9 d9 (5.535b) can be written as two ordinary differential equations.533) r 2 sin 2 9 (d2F 2 9F 1 d2O 2 O dd> 2 = m .537b) In many physical situations..537c) We transform Equation (5. This implies that m is zero and we need to solve Equation (5.„ u. 9) O (<>) In terms of F and < & .PARTIAL DIFFERENTIAL EQUATIONS I 445 u = F (r.535a) o n fl5Lil F+ A 2 ^9F 9_F cot 9 ^ 9F T L + _1 L^ L + £°1?.537a) 2 I + cote 2 \ = 0 ) d_ d9 d e _ _m__XQ d9 \sin29 (5. = F 9r 2 r 9r r 2 99 r 2 99 v / where m2 is a constant. £ _m2 F r 9r r 2 99 r 2 99 9r 2 / / 9 0 \ = 0 (5. They are 2 r2 (5.c < .
f . Equation (5. The singular points of the equation are at x (= ± 1) which corresponds to 0 (= 0. If the solution is finite for all values of 0.x 2 ) ^ .537d) By writing X as n (n + 1).539) Note that we have replaced X by n ( n + 1) and n is an integer.. The solution is 0 = AP n (x) + BQ n (x) (5. 0) = f (0) Combining Equations (5.537a) is the Euler equation and its solution is R = Cr n + ^ n+l (5. Equation (5.7t). 41) yields oo (5.542) shows that f(0) is expressed as a FourierLegendre series and the coefficients A n can be determined using the orthogonal properties of Legendre polynomials. D must be zero.540. B must be zero.543) .541) f(6) = £ n=0 A nanPn(cos0) (5.540) Suppose the boundary condition on the sphere of radius a is specified as u (a.538) where A and B are constants and P n and Qn are the Legendre functions of the first and second kind respectively. The coefficients A n are given by An = (2n+l) I 2an Jo f(e) pn(cos 0 ) sin e d e (5. If R is finite at the origin.537d) becomes the standard Legendre equation discussed in Chapter 2.2 x — + XQ = 0 dx dx (5.542) Equation (5. The general solution is oo u = X A n r n P n (cos0) n=0 (5.446 ADVANCED MATHEMATICS ( l . and n is an integer. P n are the Legendre polynomials.
the differential equations form eigenvalue problems which generate eigenfunctions. 37a. The functions to be chosen depend on the coordinate system. The solution of Equation (5.57. Each type of equation requires different conditions. Therefore. The initial conditions are expanded in terms of these eigenfunctions in the form of Fourier series. In the general case.535a.6 BOUNDARY AND INITIAL CONDITIONS The three types of second order linear partial differential equations have been derived in a physical context and the prescribed boundary and initial conditions have been based on physical considerations. hyperbolic functions. and we have to solve Equations (5. In general. from the initial conditions. both u and ^— are at given initially and at the boundaries only u is given. b). The initial value of =— can be determined from Equation (5. In Example 3. The method of separation of variables consists of writing the unknown function of two or more independent variables as a product of two or more functions. and orthogonal functions considered in Chapter 2. In addition. In the method of separation of variables. the undetermined functions are expressed as Fourier series.544) The coefficients C n and D n are determined from the boundary conditions. Together with the boundary conditions. The solution is then given as u = X [Cnrn + ^j") P n( c o s e ) rn+ / n=o I (5. the solution is — and it is the fundamental solution. These two types of equations describe evolutionary processes and. When n is zero.537b) is the associated Legendre function. the solution of a partial differential equation involves undetermined functions rather than undetermined constants. only u needs to be known initially. The partial differential equation is transformed to two or more ordinary differential equations. The solution of Equation (5. For convenience. In the case of a parabolic equation. The region can be enclosed by surfaces of spheres of radii a and b with the possibility that b tends to infinity. we have shown that the solution of Laplace's .61. the condition at the origin cannot be imposed and D is not necessarily zero. The constants generated by the solutions of the ordinary differential equations are identified as the coefficients of the Fourier series and they can be determined. we denote the dependent variable by u. u is a function of all three variables. Laplace's equation characterizes the steady state and u is prescribed on the surfaces that enclose the region of interest. For a hyperbolic equation.12). 5.PARTIAL DIFFERENTIAL EQUATIONS I 441 We have now obtained the solution inside the sphere. Possible base functions for the series are trigonometric functions. If the region of interest is outside the sphere. the values of u are given at the boundaries.535a) involves sin m§ and cos m(j) and we require the function to be single valued. These are shown in Figure 5. such that each of these functions is a function of one variable only. m has to be an integer. The problem can be completed by imposing appropriate boundary conditions. we can integrate forward to determine the values of u at a later stage.
.61 Prescribed conditions for (a) hyperbolic. + t l 3x 2 subject to u (x. 0) = 0 . lf to) M (b) u u u u o u ow) jjj[ dt L x o u L x y l_.62a. Solve the equation ^H. We elaborate the importance of imposing the appropriate boundary and initial conditions by considering two examples.U. (b) parabolic.448 ADVANCED MATHEMATICS equation in twodimensions is determined by the values of the function on the boundary curve.b) . and (c) elliptic equations Example 5. The conditions to be specified have to be compatible with the particular type of equation.. (C) u o u u L x FIGURE 5. 3y 2 =o (5.61..61) prfy = Sia^ y =o n (5.
64) (5. Solve the equation ^ = 0 dxdt subject to u (x. This example violates this rule and the problem is not well posed.66a._ ^ AS (5. t) = g 2 (t) (5.61) is transformed to X"+n2X = 0 Y"n2Y = 0 The solution that satisfies Equations (5.62.b) (5. conditions at y = 0. b) is u = sinhny sin nx 3— n . as a result of having imposed the wrong type of boundary conditions.PARTIAL DIFFERENTIAL EQUATIONS I 449 This example was first considered by Hadamard.65) is u = F(x) + G(t) Imposing Equations (5. u (0.63b) As n tends to infinity.62a. In most cases. The solution of Equation (5. 1) = f2(x) u (1. we have imposed initial conditions.65) is a hyperbolic equation and we have prescribed boundary conditions. 0) = f2 (x) . we expect that a small change in the initial conditions leads to a small change to the corresponding solution.68a) (5.66a to d) yields F(x) + G(0) = fj(x) (5.63a) (5. both u and = — tend to zero but the solution u oscillates with increasing dy amplitude.67) . Instead.66c. The solution does not approach zero as the initial conditions tend to zero. as opposed to the required initial conditions. Equation (5. t) = gl (5. that is to say.d) (t) . Note that we have to solve an elliptic equation and that we did not impose conditions on the curve enclosing the region required. Equation (5. Using the method of separation of variables and writing u as X(x)Y(y). Example 5.65) u (x.
68b) (5. 5. Example 5. In Section 4. that is to say. and are therefore not arbitrary. b). Chan Man Fong et al.71a) . the equations governing the motion are IT1i! v 3t 3f (5.71.. This is another example of an ill posed problem. (1993) considered the transient flow of a thin layer of a Maxwell fluid on a rotating disk. We illustrate this method by considering a few examples. If u is given on part of the boundary and ^— on the remaining part of the boundary.68a.68c) (5._ „ . small changes in initial and boundary conditions as well as in the coefficients of the equations lead to small changes in the solution. Hadamard has proposed the following conditions for a well posed problem: (i) (ii) (iii) the existence of a solution.68d) From Equations (5..4 satisfy Hadamard's conditions.5. In most physical problems. the solution is unique. we deduce that both f j (x) . For some nonhomogeneous problems. the boundary conditions can be imposed arbitrarily.450 ADVANCED MATHEMATICS F(x) + G ( l ) = f2(x) F(O) + G(t) = g^t) F(l) + G(t) = g2(t) (5.7 NONHOMOGENEOUS PROBLEMS The method of separation of variables was used to solve homogeneous partial differential equations with homogeneous boundary conditions. The equations solved in Section 5. we have defined Dirichlet's problem (u is given on the boundary) and Neumann's problem (^— is given on the on boundary). the condition imposed at one end (x = 0) is independent of the condition at the other end (x = 1). we on have Robin's problem.F(x) are constants and this implies that fj(x) and f2(x) can differ only by a constant. N .F(x) and f2(x) . we can introduce an auxiliary function and the problem can be reduced to a homogeneous problem. In dimensionless form.
d) Equation (5.75) Note that this choice introduces one extra degree of freedom.72a. (5. ^ ) (5. Substituting Equation (5.tj) = f^zj + g ^ ) (5. Equations (5.73a. Xj is a dimensionless number characterizing the relaxation time of the fluid.76a) so as to remove the inhomogeneity.76b) As a result of having introduced an extra degree of freedom. tj). The initial and boundary conditions are f (z .76a) dt] at! dz2 (5.7la.b) where rj is the dimensionless radial distance.c.b. 7). The functions f and x are related to the dimensionless radial velocity u 0 and the dimensionless shear stress T(rz) by u 0 = rj f ( z . 3d) imply ff(l. t l ) = O We assume that fj satisfies boundary conditions (5.4) (5.74) and separating the resulting equation into an equation for f j and another one for g. 0) = f (0. we are allowed to choose Equation (5.73c. namely (5. That is to say. b) yields ^ 3tf • * . We introduce another function and write ffz. tj) = T(1. we obtain A —r dz = 1 (5. 3z2 . tj) = 0 Combining Equations (5.75) into Equation (5.77) .7lb. 3tj * .74) is nonhomogeneous and the method of separation of variables cannot be applied directly.PARTIAL DIFFERENTIAL EQUATIONS I 451 where tj and z are the dimensionless time and height respectively.7. one function on the left side is replaced by two functions on the right side. T (rz) = r ^ l z . 0) = T(Z .
76b.79) The method of separation of variables can now be used to solve the boundary value problem defined by Equations (5.a t l + K 2 e .76a) subject to Equations (5.711) into Equation (5. b) can be written as X1T"+T' + a2T = 0 Z"+a2Z = 0 The solutions of Equations (5.4 ^ ! a 2 ] /2Xl Imposing Equations (5.b) The solution of Equation (5.b) Equations (5. b) are respectively T = Kje.711) Substituting Equation (5.452 ADVANCED MATHEMATICS fjCO) = ^ a z =0 z=0 (5.714d) .b t l Z = K3 cos a z + K4 sin a z where the Kj are constants and where a and b are given by (5. we obtain rp II mt ry II O ^ l ^ +Y = _ = _ a 2 (5. t j =   ( 1 .+ z The boundary conditions that g has to satisfy are deduced from Equations (5. ta) = 0 (5.713a. b) yields (5. 10a. 7) and are g(0.713b) (5.714b) a = [l . tj) = ZC^TCtj) (5.710a. .712a.714a) (5.b) (5.713a) (5.78a. b) is 2 fj = .710a.Vl4A 1 a 2 ]/2^ 1 b = [l + V l .714c) (5.712a.78a.76b) and separating the resulting expression into functions of z and tj. b) and we write g(z.73c.
PARTIAL DIFFERENTIAL EQUATIONS I 451 K3 = 0 a = ——. n is an integer (5.717) [An + B n ] s i n a n z (5.b) KKK) We can now substitute An and B n into Equations (5.^ .716) where A n and B n are constants (Fourier coefficients). and bnSubstituting Equation (5. 7 . The coefficients A n and B n are found to be An . we obtain x = l . a n is given by Equation (5.715a) (5.716. b) by sin a n z and cos a n z and integrating with respect to z between 0 and 1.79.715b).73a.7lb) and solving the resulting equation subject to Equation (5.716) into Equation (5. by multiplying both sides of Equations (5. 11.718b) "n The Fourier coefficients An and B n are determined in the usual manner.718a.b " tl ]sina n z n=0 (5.73d). I^i! (5 .z = £ n=0 (5. 2&Z&.ai > tl +B n e. b) on f and x yields A_ . and an and b n are given by Equations (5. a n . 14a to d. that is to say.19 a. 15a.+ z + ]T [A n e. d) with a.715b) From Equations (5. and b being replaced by ocn.718a) Z_l = £ n=0 anAn + b nBncosa^ J (5.714c. We next consider an example where the boundary condition is nonhomogeneous. we find that f can be written as f = . 17) and f and t are given in the form of an infinite series. b) and using the principle of superposition.z + X M " e ~ a " t + b " B " e ~ b n 1 cos a n z n=0 L "n J Applying the initial conditions given by Equations (5. . a. aA\K) B n .
= 0 . The diffusion of oxygen into blood is important in surgery.454 ADVANCED MATHEMATICS Example 5.71 Cross section of a wettedwall column The diffusion equation can be written as [Equation (A.71. Hershey et al. the following assumptions are made: (i) the system is at steady state 3 .IV1)] (5. A film of blood of thickness L flows down along the zaxis and is in direct contact with a rising stream of oxygen. It is illustrated in Figure 5.72. 0 L y / x ^ / f / f f Oxygen Blood i*—Column ^ wall / / / z? FIGURE 5. c is the concentration of oxygen in the blood. It is given by TT = ^ + v«gradc Dt at where y is the velocity of the blood. and — is the substantial (material) derivative. In this problem. (1967) have modelled this process as a wettedwall column. (5.720b) .720a) where «0OB is the diffusion coefficient which is assumed to be constant.
724b) ^f = 0 dx Again. ^ ax =0 x=L (5. 0) = q .722a. z) = c f . 23.724b) subject to Equations (5. no change in concentration in the ydirection. b) can be simplified to yield 3c m 3 c = ^OB^ vz7 (5721) The appropriate boundary and initial conditions are c (0. we write c as c(x. q is the concentration at the inlet. Note that Equations (5.724b) and g also satisfies g (0) = c f . we write vz To reduce Equation (5.b) The solution of Equation (5.722a to c.723) f" = <®0B f i (5724a) (5.b. we are allowed to impose an extra condition. homogeneous equation. 25a.723) into Equation (5.725c) Combining Equations (5.722a. b) is g = cf (5.722a) to a (5.720a. b) are nonhomogeneous.z) = f(x.725a. ^ dx =0 x=L (5. since we have introduced an extra function g.PARTIAL DIFFERENTIAL EQUATIONS I 455 (ii) (iii) (iv) negligible diffusion in the zdirection.725a. b) yields the boundary conditions on f and they are given by .z) + g(x) Substituting Equation (5. Equations (5. vz is constant. This condition is given by Equation (5. c (x.c) where cf is the concentration at the blood /oxygen interface.721).
726atoc).x (5. we obtain .730a) / V / V X = Bcosn/\/ —z— x + C s i n n / \ / —^.O) = CjCf .456 ADVANCED MATHEMATICS f(O. 27.724a) and separating the variables. b. c) yields V <© 0B B=0.b.723.z) = O. b) and applying the principle of superposition. .724a) subject to Equations (5.b) where s (= 0.729a) (5. Imposing conditions (5.729b) + ^^ X = 0 The solutions of Equations (5. » vz (5.727) into Equation (5. We write f as f = X(x)Z(z) Substituting Equation (5.731a. 31a. f(x. B and C are constants.727) I = ^f = "' (5.) is an integer. ££dx =0 x=L (5.b) where n 2 is the separation constant.726a. b) are respectively 2 Z = A e~n z (5. we obtain (5..728a. 25c. Equations (5. 30a. 2. n =^ ^ J ^ ZL.729a.726a.730b) V <O 0B where A. b) can be written as Z' + n 2 Z = 0 X" (5..c) We can now use the method of separation of variables to solve Equation (5.728a. Combining Equations (5. 1.
*^n™ + cf = X As exP — ~z m~ " ^T (5.736a) rL I (c) = c(x.PARTIAL DIFFERENTIAL EQUATIONS I 457 V c A (2s+l)2rc2J90Bz . (2S+1)JCX sin . the nonhomogeneous terms are constants.336b) is obtained by integrating Equation (5.734b) Substituting Equation (5. • In the two examples so far considered. z) dx 8 (c: . The method can be extended to the case where the nonhomogeneous terms are functions of one of the independent variables.^ — 2 sin i —^ dx JQ 2L L = 4(ci"^f) (2S+1)7C (5.>2A1rt. .M5i s=0 (5.734b) into Equation (5.c.736b) > s=0 * exp — ^ — — (2s + I ) 2 L 4L2vz Equation (5.335) and is identical to Equation (6) of Hershey et al.722b) implies c. except for a difference in notation.732) S I 4L2vz J 2L The boundary condition (5. We illustrate this method in the following example.736a) (5. (1967).cf) v = —Ki K2 fl 1 f( 2s + i ) ^ 2 ^OR z l 2e— + c f (5.735) The average concentration across the film at a point z is defined as (5. we obtain (5.734a) (5.732).736b) (5.733) The Fourier coefficients A s are given by 2(cjc f ) f (2s + l)nx As = .
738a. We express f(0) as . We simplify the problem by considering the steady state case. The equation of energy [Equation (A.740b) generates Legendre polynomials and it is appropriate. Wilson (1979) solved the problem of film boiling on a sphere in forced convection.540)]. as discussed in Section 6. we obtain r (5.III3)] can be simplified to yield a!l + 23I + ^ A 3r 2 r 9r s i n e 3I 30 = 0 (5. The general solution which is finite at the origin is oo T = A0 + X n= l A n r" P n( c o s 0 ) (57"41) where P n (cos 0) are the Legendre polynomials. 0) is the temperature. to expand f(0) as a series expansion in Legendre polynomials.739) 2dR dr 2 2 +2r dR _ dr n ( n + 1 ) R = 0 (5. Note that there is a typographical error in the expression for f(0) in Wilson (1979).737) r 2 sin0 30 where T (r.740b) The solution of Equations (5.458 ADVANCED MATHEMATICS Example 5. We note that Equation (5.73. c o s 0 + icos 3 0) T is finite at the origin.737).740a) M  + cot0 — + n ( n + l ) 0 = 0 d0 d0 (5. b) are given in Section 5 [Equations (5. The boundary conditions are: on the surface of the sphere (r = 1)  1 = f(9) =  s i n 2 9 ( l .740a. We use the method of separation of variables and write T as T = R(r)0(0) Substituting T and its derivatives into Equation (5.b) (5.
744a) + .b) — 2 — if m = n V 2n+ 1 Multiplying both sides of Equation (5. We know that Po is one.742) by Pm(cos 0) and using Equations (5. We recall that [l Pm(cos 6) Pn(cos 0) d(cos 0) = ( ll 0 if m ^ n (5.7450 (5.742) The coefficients B n are determined using the orthogonal properties of Pn (cos 0).PARTIAL DIFFERENTIAL EQUATIONS I oo 459 f(6) = X n=l B n P n( c o s e ) (5. I f(e)P n ( C os0)(sin0)d0 Jn = ( 2 n + 1 I I f(0)P n (cos0)(sin0) d0 (5.744c) Jo The function f (0) is given.( 2 n + 1 ) .744b) (5.745d) ^ Jo 3V^ = ±2 .738b). f(0) is given by Equation (5. b).745b) (5. Pn (cos 0) are known in terms of powers of cos 0 and the integral can be evaluated.743a. we deduce Bn = ^ " ^ 1 f(0)P n (cos0)d(cos0) (5.743a. We calculate B o in detail. and Bo can be written as B o B lf 3(sin^0)sin0d0 _ Jo 4Ylcos0+lcos 3 0 y(lx2)dX (5 ? 4 5 a ) =f f = \[ 8 (5.
It is equally possible to use Laplace transforms to reduce a partial differential equation in two independent variables to an ordinary differential equation. B2. we define the Laplace transform of u with respect to t by .00 L[u(x. x + £x 3 (5. We recall that the Laplace transform of a function f (t) was defined by Equation (1. t)] = U(s. Likewise.741) and using Equation (5. if u (x. This is not surprising because.t)dt JO (5.171). we have used Laplace transforms to simplify an ordinary differential equation to an algebraic equation.747).8 LAPLACE TRANSFORMS In Chapter 1. c) are obtained by making the following substitutions respectively x = cos 6 (5.746a) y = l . t) is a function of two independent variables x and t. including nonhomogeneous problems. the solution is arbitrary to the extent of a constant.746b) Similarly Bj. ••• can be calculated. for a Neumann problem.81) . The method of integral transforms. The constant A o is arbitrary. can be used to solve partial differential equations. The boundary condition given by Equation (5. The coefficients B n are known and so the An (n > 1) can be determined. (5.748) We deduce that An = ^ for all n > 1.738a) can now be written as 00 3f r=1 =ZBnPn(cos9) n=0 (5747) Differentiating T from Equation (5.745b. we obtain 00 00 £ n=0 n A n P n (cos 9) = £ n=0 B n Pn (cos 0) (5.460 ADVANCED MATHEMATICS Equations (5. x) = I e" st u(x. introduced next.749) 5.
.85a. Wimmers et al. t) = c 0 . Combining Equations (5.Soa. We illustrate the method of Laplace transforms by solving a few partial differential equations. 2 dC AB — 7 + [dr2 r dr .S + KJL.IV3).k c A where k is the rate constant. 3) yields (5.85b)]. (1984) studied the diffusion of a gas from a bubble into a liquid in which a chemical reaction occurs. Example 5. and r is the radial distance.b. t) dt). the concentration of the absorbed gas in the liquid phase is found."^AB T 7 r T ~ ~ [r z dr \ dr }\ We make the substitution C = £ and Equation (5.83) We denote the Laplace transform of c A by C (= I e~st c A (r.PARTIAL DIFFERENTIAL EQUATIONS I 461 Other properties of Laplace transforms are listed in Chapter 1. by simplifying Equation (A. to be ^^JLffr2^]*^ (5.87) ro d 2 C .82) where c A is the concentration of gas A in the liquid.86b) reduces to the following equation with constant coefficients (5.c) (5. .b) (* R . c A (r. p.J9 I (. 0) = 0 ( r > R ) . We assume that the rate equation is first order and RA is given by RA = . Taking the Laplace transform of Equation (5. If the bubble is at rest and has a constant radius R. RA is the production rate of A..82. c A — > 0 as r — > °° for all t (5. t is the time.81. J S A B is the diffusion coefficient of solute A in solvent B.84) and using the initial conditions [Equation (5. we obtain " 1 d (r2 dCW _ (s + k) C .84) The boundary and initial conditions are c A (R.
85a).812b) = ^(^nr) exp[^ T ^ l Equation (5.or C = 0 dr 2 a2 = (s + k ) / . Substituting this into Equation (5.89) C = f<TaT (5.R) (5.810) The constant A is obtained by using Equation (5.171.85c) is C = A e~ a r From Equation (5.87). t) is c o / s .88a) (5.88b) (5.81 lb) K = ^0e«R The solution can now be written as A C = ^oe«(rR) (5.813) cA = 5^0 ekt /exVk erfcr_^_VkTJ + e xVF e r f c [ ^ . we obtain (5. The Laplace transform of c A (R.810) yields ^4e"aR s (58iia) (5.[^+^ 1/2 (r .812a) = ^ exp .812c) is in a standard form and is obtained by substituting a = s+k From the table of Laplace transforms given in Table 1.88a) satisfying Equation (5. © ^ The solution of Equation (5.+ V k t i l (5.814a) x = (rRj/Vj^B (5. we find that c A is given by (58"12c) (5.814b) .462 ADVANCED MATHEMATICS ?± .
k is zero and Equation (5. and evaluating the integral using the Cauchy residue theorem [Equation (3.J . then letting the contour tend to infinity. There is a general inversion formula which can be written as f(t) = L.710)]. This is illustrated in Figure 5. These functions are defined by erf x = 3= e~u du (5.817) • In using the method of Laplace transforms.81. The method we have used is to read the inversion from the table of transforms. we obtain cA = ^ erfc ( '~R ) (5. The integration is usually done by taking L to be a finite line. closing the contour. we need to invert the transform F (s) so as to obtain the solution f (t).82. see Problem 9b in Chapter 4. The constant y is chosen so that all singularities of F(s) are to the left of L. One method of obtaining a closed contour C is by joining the line L to an arc of a circle centered at the origin and of radius R.812b) reduces to C = ^ o exp . as shown in Figure 5.818) where s (= x + iy) is a complex variable and the path of integration is along a straight line L in the complex splane.1 L ^ «®AB (rR) (5816) From the table of Laplace transforms.PARTIAL DIFFERENTIAL EQUATIONS I 463 where erfc is the complement of the error function (erf).815a) ^ Jo erfcx = 1erfx = L in I Jx e~u du (5.815b) (To show that — ( Vif Jo e~u du is one. .1 [F(s)] = ±r Jyioo estF(s)ds (5.) If no chemical reaction occurs.
464 ADVANCED MATHEMATICS yt L : r r FIGURE 5.82 Closed contour in the inversion of Laplace transforms .81 Path of integration (L) in the inversion of Laplace transforms. The dots (•) are singularities of F(s) I R/ ^ ~ ^ P Q FIGURE 5.
the integral tends to zero as R tends to infinity. we deduce that. 23). On the arc MN. because cos 0 is negative.b) I /NP e st F(s)ds = 1 Jn/2 r 371/2 exp[Rt(cosG + isine)] F(Re i 0 )iRe i e d0 (5. b. F (s) —> 0. F satisfies the inequality F(s)<^Rm (5. That is to say. We assume that.823) where a and m are positive constants.710). s j (5.819) where sn are the singularities of e st F(s). f f < 0 < ^ /•3JC/2 (5.821a.822a) exp(Rtcos0) exp(iRtsin0) F (R elG) i R e10 d0 Jn/2 r 37t/2 eRtcos0 F (5. c.822C) Jn/2 since e i a = 1 for all real a.822b) (Reie)Rde (5. we have s = Reie.822a. when R —> <».PARTIAL DIFFERENTIAL EQUATIONS I 461 From Equation (3. we have ^r 1 estF(s)ds = £ JC n Res[estF(s).820) On the semicircle NP. on the arc of the circle. The integral around C can be decomposed as I e s t F(s)ds = I e s t F(s)ds + estF(s)ds+ estF(s)ds+ e s t F(s)ds JL JMN JNP JPQ Jc (5. From Equations (5. we have the following inequalities .
823). + . s j (5.824e) and Equation (5.B.! ho <^(0 0 ) where e 0 = cos'^y/R) (5. If F(s) satisfies inequality (5.824b) rn/2 < _QL_ eRtcose d0 (5.466 ADVANCED MATHEMATICS I JMN e st F(s)ds < I JMN e s t F(s)ds (5..826) The integral I JMN e s t F(s) ds tends to zero as R tends to infinity.827) .824f) (.825).819) becomes ±r I estF(s)ds = ^ JL n Res[estF(s). Equation (5. (5.825) From inequality (5. we deduce e s t F(s)ds JMN < ^— K y (5. the I JPQ estF(s)ds tends to zero as R tends to infinity.824a) fn/2 < J% eRtcose^Rd9 R (5.. Similarly.2 4e) If we let R tend to infinity.824c) rn/2 < _o_ eRt^/R d9 (5.824d) R m . (y/R) will tend to zero. Expanding c o s ^ y / R ) about zero in a Taylor series yields cos" 1 (y/R) =  .
the appropriate cuts should be made as discussed in Chapter 3. The solution of Equation (5. (5. We consider the heat conduction problem of a semiinfinite region x > 0.831) is T = A exp[(Vs/k* ) x] + B exp[(Vs/k* ) x] where A and B are constants.III1).833a) Jo .831) where T is the Laplace transform of T.c) where c 0 .829) where k The initial and boundary conditions are T(x. and e are constants.832) The condition that T —> 0 as x—> «> implies that T —> 0 as x—> «>.818.828) We have assumed that e st F(s) has no branch points inside the region enclosed by C.830a). 27) yields f(t) = £ n Res[estF(s). C O .t) = c 0 sin (cot + e ) . is ^ dt = k* t l 3x 2 (= k/pC ) is the thermal diffusivity and T is the temperature. Example 5. Taking the Laplace transform of Equation (5.83Oa.PARTIAL DIFFERENTIAL EQUATIONS I 467 Combining Equations (5. s j (5. The constant A has to be zero.829) and using Equation (5.b. If e s t F(s) has branch points. T(O. We illustrate this point by the following example. 0) = 0 . The governing equation. we obtain 2— ST = k*~ dx 2 (5. obtained by simplifying Equation (A. T — >0 as x — > °° (5.82. The condition at the origin (x = 0) implies that T at the origin is given by r T= e~st c 0 sin (cot + e) dt (5. (5.
83.b) where p is real and positive. Contour C consists of the line L.833b) Applying the boundary conditions on T.835) along Lj and L 2 . We have to remove the origin and the negative part of the real axis. s = p e" i7t (5. The resulting contour is illustrated in Figure 5. the line L2. t) = K 711 I Jyioo e st — ^ — (s sine + co cos e) exp [x Vs/k* ] ds S2 +C0 (5.b) The contour shown in Figure 5. it is 7t. the integrals along the arcs MN and PQ are zero. We denote the integrals by Ij and I 2 . Similarly.836a.b) We now evaluate the integrand of Equation (5.82 is no longer appropriate. The square roots of s on Lj and L 2 are given respectively by Vs~ = iVp~. and finally the arc PQ. b. From Equations (5. We note that the integrand has a branch point at the origin (s = 0) and simple poles at Sj = ico.835) To evaluate the integral.834) The inversion formula [Equation (5. On letting the radius R tend to infinity. b). the line Lj.838a. the integral around Cl tends to zero.818)] gives /•y + ioo T (x. On Lj and L 2 . we transform it to a closed contour integral as explained earlier.835. we write s = p ei7t . the arc MN. 38a. as e tends to zero. 37a. we obtain I: = I J°° e"pt C° (p sine + cocos e) exp[i(Vp/k*j xj (dp) (5. Vs~ = iVp~ (5. the argument of s is n and on L 2 . we obtain Q r i i T= 2 — ( s sine + cocos e) exp[xVs/k* J s 2 + co2 (5. as shown previously. the small circle Cj. s 2 = ico (5.837a.839a) p 2 + co2 .468 ADVANCED MATHEMATICS = Q — (s sine + cocos e) s 2 + co2 (5. On Lj. The integrand is a singlevalued function of s on and inside contour C.
^ ^ * ] •'O p2 + co 2 dp (5.— TM 1 L. c) yields I = l 1 + i 2 = I eP1—^— (p sine+cocos E ) ^ ^ * .840c) ^ . ^ Q FIGURE 5. = lim T±—^r.b) = j ^0 e pt —^o— (_p s i n e p2+ o 2 + (acos e^ (_2j) s i n [(Vp/k* ) xj dp (5.PARTIAL DIFFERENTIAL EQUATIONS J 469 = I JO e"pt—^—(p sine + cocos e)exp[i(Vp/k*)xj dp (5.714).841a) .83 Contour integral with a branch point at the origin From Equation (3.840a. we find that the residue Rj at s 1 is given by (s — ico)e s *c f / I *"\ 1 R. ^ ( s sine + co cos e)exp[\Vs/k j xj 1 s—>iro (sico)(s + ico)v (5.839c) (o2 Combining Equations (5.839b.839b) p 2 + a)2 I2 = I JO e~pt—^—(p p2+ sine + cocos e) exp[i(Vp/k jx] (dp) (5.
the residue R 2 at s 2 is found to be R2 = _eo£^! e ie e (V^72k^)x exp[i(Vco/2k*)x] (5.710].842a.b) = c0 sin (cot + 8 . (5. we deduce that T is given by 2ni JL T = c0 sin (cot + e . The function T given in Equation (5.840c.xVco/2k*) e~x Vco/2k* Using Cauchy's residue theorem [Equation (3.x V co/2k*) e"x Vco/2k * + ^o 71 _e_^—(_ p sine + co cose) sin (x Vp/k* ) dp Jo P 2 + C O 2 (5. 43b).843a) JCX JL2 JVQ = 27tic0 sin (cot + e . From Equations (5. f .835) is J— I .842c) 1 = 1 + 1 Jc JL /MN +  + I +1 + j JL.843b) We have shown that f .— (5. we can write (5.xVco/2k*) e~x Vco/2k* (5.470 ADVANCED MATHEMATICS c eicot r i /—^\ i = Q.841e) The sum of residues R is given by R = R1+R2c0 [expi (cot + exVco/2k*)expi (cot + exVco/2k*)] _xVco72k*' — e (5.841(1) Similarly.844) .841b) (5.841c) (cos e + i sine) exp .Q±J± (Vco/k*) x L V2 = £ o  p eie e(VS/2F)x e x p [_ i (Vco/2k* ) x] (5. and f are equal to zero. (ico sine + cocos e)exp_vvico/k ] xj = C^r.
Then Equation (5. Model the smoke dispersion from a high chimney and solve it by the method of Laplace transforms.845) simplifies to .844) tends to zero.84.84 Smoke dispersion from a chimney We further assume that the derivative with respect to x is small compared to the derivative with respect to z. where vx = V. and the diffusion coefficient «©AB = D (a constant). Smoke diffuses from a chimney of height h into the atmosphere.PARTIAL DIFFERENTIAL EQUATIONS I 471 We note that as t tends to infinity. Example 5. Suppose the prevailing wind is in one direction only (xdirection) and has a constant velocity V. the solution is given approximately by the first term on the right side of Equation (5. is V ^ = D[(^) + ( ^ 1 3x L\ax2/ \3z2 .844). R A = 0. z).845) Z wind ^ velocity h smoke >» x FIGURE 5. as shown in Figure 5. obtained by simplifying Equation (A. The average crosssectional concentration c (x. (5. the integrand on the right side of Equation (5. The vertical axis is the zaxis and the base of the chimney is at the origin. For sufficiently large t.83.IV1).
852) Jo Equation (5. z. We introduce the dimensionless quantities x. The solutions are given in terms of exponential functions.853) This is a second order equation which we solve for z < 1 and for z > 1 separately because the delta function introduces a singularity at z = 1.^ = 0 dz at x —> « .847a. at x = 0.h) .b) =0 where Q is the source strength (flowrate) emanating from the chimney and 8 is the Dirac delta function.846 to 48b) become ^ 8x = *°dz (5.850)  _ = °' f _ = °' 5 lx=0 = S ^ " 1 ) (5. z ) d x (5. ^.848a. 8 ( z .850) transforms to 5^sC = 8(zl) dz2 (5. the boundary conditions are at z = 0.c) The Laplace transform C is defined by r C = e~sx c(x. Under these assumptions.b. . at z—>oo.849a. c (5. h 5 = 7? Q (5. keeping in mind that a second order equation is associated with two arbitrary constants. and c by writing x = m.b.l ) = 0.b) (5. For both z < l and z > l . dz c = (Q/V) 5(z .= 0 .851a.5 Vh 2 z = U". .4Z2 ADVANCED MATHEMATICS V ^ = D ^ 3x dz2 (5.c) Equations (5.846) The inversion height (where there is no transport of smoke) is at infinity and the rate of deposit of smoke at ground level is negligible.
s) =e^/Vs~ Using Equation (5.i V i ).857) Note that the second integral on the left side of Equation (5. The values of A and B are found to be A = —±= ( e ^ + e " ^ ) .PARTIAL DIFFERENTIAL EQUATIONS I 421 The solutions satisfying Equations (5. 2 Vs Vs B = ~= 2 Vs (5.85la to c) are C = A e~zVi. we deduce that Vs~ [Ae"^ + B C e ^ .860) Jy_ioo V T We note that the integrand has a branch point at s = 0 and we proceed to evaluate the integral as in Example 5. we invert C and obtain 1 1 c (x. we determine C (0. where A and B are constants.856) is zero because C is continuous at z= 1.859) e s x e~"^ e %.856) Combining Equations (5. 56) and letting e —> 0. we integrate Equation (5. b. = B(e i V i + e.i v 2111 (5. s) and this is given by C(0.854a) (5. On Lj and hj.854a. ds (5.853) and obtain [iC1 si Cdz = 8(zl)dz =1 (5.83).855) z>1 z<l (5.82.818).858a. the square roots of s can respectively be written as . The only nonzero contribution to the integral is along Lj and L2 (Figure 5. We further assume that C is continuous at z = 1 and this implies that A = B(e 2 V 5 +l) (5. 0) = .854b) The derivative of C at z = 1 is not continuous and to evaluate it.e " ^ ) ] = 1 (5.b) To determine the concentration at ground level (z = 0).
8. From Equations (5.l / 4 x ] d p "1Mx) 71 (5. its Fourier series can be written as oo f(x) = 2 bn 5511111 ^ n=l L (5.\ J —oo exp[(pV^" + i / 2 V ^ ) 2 .862a) = ^  J oo exp[(p 2 x + ip)]dp (5.861a.862d) = eXp( exp[(pV^ + i/2V^) 2 ]dp Joo = «P(1^> VTCX ( 5. Vs~ = ip (5. c max = J±1/2 (5.862c) (5.474 ADVANCED MATHEMATICS V 7 = ip.860) yields c(x.J .862b) /./co n Jo /• oo exp [(p 2 xip)] dp (5.b) Substituting Equations (5.I n . 5. b).862e) The maximum concentration c m a x is given by (5.861a. 0) = J.864a.9 FOURIER TRANSFORMS From Section 2.863a. the maximum concentration is (Q/V) (2/TU e) at a distance Vh /2D from the chimney.b) 2 In dimensional variables. it is seen that the assumption 9 / B z » 3 / 3 x requires that Vh/D»l.849a.I exp[(p 2 x + i p ) ] d p .9la) .b) ^li V *v A. The solution is x = 1/2.oo =. we deduce that if f (x) is an odd periodic function of period 2L. b) into Equation (5.
and y = 0 as shown in Figure 5.. If u is the potential.96) .. In the semiinfinite space bounded by the planes x = 0. Example 5.94) Its inverse is given by oo f(x) = ^ F C O + X F c n ^ 8 1 1 ^ Z <59"5) n=l * The Fourier sine and cosine transforms can be used to solve ordinary and partial differential equations. u satisfies the equation 9 u + 3x 2 3y 2 9 u = _p . The procedure is similar to that employed when Laplace transforms are adopted. (5..91. We solve an electrostatic problem using the method of Fourier transforms.91.. we can define its finite Fourier cosine transform as Fcn = ? I f00cosli?L Jo dx (5. we deduce that the inverse is CO (5. x = %. there is a uniform distribution of charge of density (p/47i).93) n= l Similarly if f (x) is an even periodic function of period 2L.9lb) We can regard b n as the finite Fourier sine transform of f (x). In keeping with the notation used in Laplace transforms.9la). we write Fsn = f L fOOsin^dx Jo L (5.PARTIAL DIFFERENTIAL EQUATIONS I 475 bn = f I fOOsinQp dx L Jo L (5.92) From Equation (5.
2. 0) = 0 (5. y) = 0 . y) = 1 .99b) .c) We assume u to be bounded throughout the region.96) by — sin nx and integrate from 0 to n.91 The boundary conditions are u (0.99a) * Jo dx2 * L3x Jo J o 3x  I 7 1I n (* \ = — \n [u cosnx]"n 2 I u sinnxdx} Jo J (5.I sinnxdx = — U—sinnx n I ^ r — cosnxdx} (5. u (7t. The region is bounded in x and we use the finite sine transform method to solve Equation (5.98) 3x2 * Jo 9y 2 « Jo The first integral can be integrated by parts and we obtain 2.I Kh sin nx dx + — I sin nx dx = —— I sin nx dx (5.96).97a. We multiply both sides of Equation (5.476 ADVANCED MATHEMATICS Vt I 0 1 IT »> x FIGURE 5.b. Semiinfinite region u (x.
9.n y .b) [(I)'l]+2fcl)! Ten3 nK <5." sin nx (5. Substituting Equation (5. 1 } n / (5.9c) where U sn is the finite sine transform of u.^ = ^ 7i n 3 nn (5. 12a. we obtain B = 2 p (5. From Equation (5.14) \ n3 Using Equation (5. the region can be considered to be infinite. In many physical problems. The Fourier series is then extended to a Fourier integral.911) simplifies to u . b). The function f (x) can be represented in complex form as .PARTIAL DIFFERENTIAL EQUATIONS I = 4TL ( 5 9 _2IL (_!)"_ n 2 U s n .2 p [ ( ~ 1 ) n ~ 1 ] .912a.93). ii = 2(PlH^ll K + H):\( e . we express u as oo u(x.y) = X u s n s i n n x n=l (5.911) The condition that u is bounded implies that A is zero.910) is U s n = Ae n y + B e .10) K n T U The solution of Equation (5.9lSb) So far we have assumed that the function is defined in a finite interval.913) Equation (5.99c) into Equation (5.' ) " .98) and carrying out the remaining integration yields ( _l)i2n _ n 2u ^ k ^y2 = 2pKVnV (59.911. y .97c). we deduce y =0.915a) +LJl!{e^l} = 2. U sn = 0 From Equations (5.  ( P " .
919a) Jo F s (a) = I f(x)sinaxdx r (5.918a) r B(a) = f(x)sinaxdx (5.918b) Jo We define B(a) as the Fourier sine transform of f(x) and Equations (5.I Fe (a) sin ax da % (5.917a.916b) We define F(a) as the Fourier transform of f (x) and its inverse is given by Equation (5.917b) r B(a) = J oo f(x)sinaxdx (5. b) can be written as f(x) = 2.917a) Jo r A(a) = I Joo f(x)cosaxdx (5.919b) Jo .916a) I J — oo r F(a) = f(x)e~ i a x dx (5.478 ADVANCED MATHEMATICS f(x) = L. An alternative form of writing Equation (5.917c) If f(x) is odd.916a). A (a) is zero and Equations (5.916a) is f (x) = L I n [A (a) cos ax + B(a) sin ax] da (5.918a. I 2K F(a)e iax doc (5. c) become f (x) = 2L I B ( a ) s i n a x d a n Jo (5.
Similarly the Fourier cosine transform and its inverse are defined by Fc(cc) = I f(x)cosaxdx Jo f(x) = 2.921a) f(x) = J ^ I Fc (a) cos ax da Jo Note that both sets of Equations (5. Sedahmed (1986) studied the rate of mass transfer at a cathode. some authors define the Fourier cosine transform and its inverse as f°° Fc(cc) = 4^ 11 I f(x)cosaxdx Jo (5. The mass diffusion equation is determined by simplifying Equation (A.921b) ^ oo f(x) = — I I f(x)cosaxdx cos ax da n Jo Jo (5. For the sake of symmetry.924a) .PARTIAL DIFFERENTIAL EQUATIONS I 479 The inverse of the Fourier sine transform Fs(oc) is given by Equation (5.IV1) with R^ = 0 to yield ^ dt = J9 *?dy2 (5.I F c (a)cosaxda Jo r (5. b) lead to p oo (5.920a) (5. V 7E Example 5. b.919b)] can be defined with the factor v — in front of the integral.919a). 21a. The initial and boundary conditions are c = Cj for y = 0 at t > 0 (5.920a.920b) The definitions given by some authors differ from those given here by having different factors in front of the integral.916b)] and the Fourier sine transform [Equation (5.92.923) where c is the concentration and J9 is the diffusivity.922) Similarly the Fourier transform [Equation (5.
930). c* (y.226a) (5. t) = 1 .226b. dyz c* — > 0 as y — > °° .925) c* (0.I a^dy cosaydy) (5.927c) = « © la [c* cos ay] ~ .927b) (5.927a) o ay 2 py Jo Jo j c* sin a y dy > (5.927c).^. we deduce that at t = 0.929) (5. that is to say. From Equation (5. we find that C s is given by (5928) (5.sinocydy = « © (U—sin ay .926a) becomes dC o —^ + a <®CS = a<0 The solution is C s = J.930) . Cs = 0 Applying Equation (5. 0) = 0 for y > 0 (5 . we write c* = ( c b .480 ADVANCED MATHEMATICS c—>c b c = cb as y — > ° o a t t > 0 for y > 0 at t = 0 (5.926a). Using Equation (5.( a 2 t J9)] where A is an arbitrary constant. we multiply both sides of the equation by sin ay and integrate with respect to y from 0 to infinity.C i ) Equations (5.924c) For simplicity.d) We take the Fourier sine transform of Equation (5. Equation (5.c. 24a to c) become ^ dt = JS.c ) / ( c b .+ A e x p [ . The right side is now —%.a 2 I = JS)[aa2Cs] where C s is the sine transform of c .924b) (5. (5.923.926d).
t) = 2. we obtain c* (y. Using the inversion formula.916a.919a)]. b)] is used. 0<x<a \ Fs(ct) Jflcos(aa)] (5. we find that the inverse of Cs is c* = erfc(y/2fjS)T) which is the solution quoted by Sedahmed (1986).91 and 2.931) To obtain c .PARTIAL DIFFERENTIAL EQUATIONS I 481 r s = i.91 Fourier sine transform f(x) 1.( l .a a tJ9) (5.933) 0.91). the Fourier transform [Equations (5. a < x < °° j e" x xe" x erfc ax x~a.sin ay da (5.e . If the region of interest is the whole infinite space («> to <»).932) Jo From the table of Fourier sine transforms (see Table 5. we have used the sine transform because the values of u and c are given at the origin. TABLE 5. • In Examples 5.I 71 ^— a . r is the gamma function ala . 0 < a < l a/(l+a2) 2a/(l + a2)2 X [1 _ eXp (a 2 /4a 2 )] ^cosNr(la) V2 . We choose the cosine transform if the first derivative is given at the origin. This is shown in the next example. we can either look up in a table of sine transforms or use the inversion formula [Equation (5.
34) subject to the initial conditions u(x.i c a t (5. The solution is U = Ae iccct + B e . as before.9.936) where U (= I Joo u (x.935a) (5. we obtain 2 Ht dt 2 +c 2 a 2 U = 0 r°° (5.939b) F(a) = I f(x)e.935b) =0 t=0 Taking the Fourier transform of Equation (5. Solve the wave equation ^JL_ c 2 3jL 3t2 3x2 = o.939c) .934) and integrating by parts.939a) (5.482 ADVANCED MATHEMATICS Example 5. Differentiating U with respect to t yields ^ = iac(AeicatBeicat) (5.O) = f(x) ^ dt (5. t) e~ m x dx) is the Fourier transform of u.937) where A and B are constants.i a x dx Joo (5. oo< x <oo (5.93.938) The initial conditions imply F(a) = A + B 0 where = AB (5.
c t )]da (5. 5.941) (5.916a)].102) . the derivatives with respect to r can be removed by the application of the Hankel transform. 1 I 271 1 I J _oo F(oc) [ e i c a t + e .940a.942a) = _i_ I 4TC F(a)[e icc ( x+ct ) + e i a ( x .10 HANKEL AND MELLIN TRANSFORMS (5.943) We have seen that the solution of Laplace's equation in cylindrical polar coordinates involves Bessel functions.i c a t ] e i a x d a (5.937) becomes U = i .939a. The inverse is given by f(r) = r Jo aF n (a) J n (ar)da (5. we obtain A = B=lF(oc) Equation (5.942b) as u = 1 [f(x + ct) + f(xct)] Equation (5.b) u = J_ . We define the Hankel transform of f (r) by F n (a) = f(r)rJn(ar)dr (5.101) Jo where Jn (ar) is the Bessel function of the first kind of order r. Solving Equations (5.PARTIAL DIFFERENTIAL EQUATIONS I 483.943) is d'Alembert's solution. we obtain (5. b). In problems formulated in cylindrical polar coordinates and if the radial distance r is from 0 to infinity.icoct ] Using the inversion formula [Equation (5.F ( a ) [ e i c a t + e.916a) allows one to write Equation (5.942b) I J <x> Equation (5.
r> 1 dz (5. = O (5. z) J 0 (ar) dr = 0 (5. As an example of the use of the Hankel transform. we consider the problem of a charged disk.m F(m)dm (5. Multiplying every term of Equation (5.104) is similar to the contour used in the inversion formula for the Laplace transform [Equation (5. The contour integral in Equation (5.103) is chosen such that the integral exists.101. Example 5. The potential u due to a flat circular disk of unit radius satisfies Laplace's equation and the following boundary conditions ^+I^L+^J1 r dr dz2 dr2 o n z = 0.0<r<l.1O5a) u=un.484 ADVANCED MATHEMATICS The Mellin transform is defined by F(m) = I x m .103) Jo Its inverse is given by /•Y+i~ f(x) = L I x.1 f(x)dx (5.105a) by rJ o (otr) and integrating with respect to r from zero to infinity.105a) suggests that we take the Bessel function of order zero.104) Z7C1 I /yioo The exponent m in Equation (5.105b.106) Jo W r dx) dz 2 j o The first integral can be written as f r (& + ^) J » (ar)dr = f^( r ^ J ° (ar)dr (5.107a) . we obtain f°° / 2 \ 2 f°° r — +— J 0 (ar)dr + — I r u (r. u ^ =0.818)].c) (5.1O5d) u—>0 as z—><» Equation (5.
1010) u = I aA(a)eazJ0(ar)dcc (5.a dz 2 2 U =0 (5. we use Equation (5.I a 2 r u J 0 (ar)dr (5.1012) 9z Jo Imposing the boundary condition on z = 0.102) which leads to (5.107c) = . we have .107d) Jo = a2U (5.109) The solution that satisfies Equation (5.I a2A(a)eazJo(ar)dcc (5. To obtain u.107b) 0 Jo = .106) becomes 2 (5.107d) is obtained by recognizing that J0(cxr) satisfies the following equation a r JQ + J o + a r J o = 0 Equation (5.[ a r u Jo(ar)J Q + I a u [J 0 (ar) + r a JQ (ar)] dr Jo (5.1011) Jo Differentiating with respect to z yields — = .107e) The terms inside the square brackets are zero because of the boundary conditions.PARTIAL DIFFERENTIAL EQUATIONS 1 485 r^J 0 (ar)  ar^j:(ar)dr dr (5.108) ^ . Equation (5.105d) is U = Ae~az where A is a constant.
b) become (5. we deduce that B(a) = sin a From Equations (5.1011. Jo 0<r<l (5.1013a. oz . b).1014) I a~lB(a)JQ{ai)da Jo = ^. 0<r<l (5.1018) Note that in this example. r > l r (5. 17). b.1016a) I J 0 ( a r ) s i n a d a = 0. the boundary conditions are of the Robin type (u is given on a section of the boundary and = — on the remaining part of the boundary) and we determine A (a) after the inversion.486 ADVANCED MATHEMATICS I aA(a)J0(ar)doc = u0 .1O13b) Jo We make the substitution a 2 A ( a ) = ^u f t B(a) Equations (5.1O13a) (5. 16a. r > l Jo From the integral properties of Bessel functions [see Watson (1966).1015a. 14. we obtain u = ^21 a~1sinae"azJo(ar)da r (5. r > l Jo Comparing Equations (5. we note that (5.1017) Kh (5. 405]. p.1015a) r B ( a ) J 0 ( a r ) d a = 0. 0 < r < l Jo I a2A(a)J0(ar)da = 0.1016b) (5.1015b) I a"!J0(ar)sinada =  .
101. we obtain r ml r2 3_T + r 3lU+A_ rm1Tdr = 0 (5.1021) becomes . T = 0.IH2) r 2^T 3r 2 + r 3T 9r +dh = Q (51O19) 362 where T is the temperature.PARTIAL DIFFERENTIAL EQUATIONS I 487 If the boundary conditions are of the Dirichlet or Neumann type.101 Heat flow in a wedge Taking the Mellin transform of Equation (5. The steady heat flow problem in a wedge.102.1O2Oa. as shown in Figure 5.1019). T=l. we determine A (a) and then carry out the inversion. consists of solving Laplace's equation. subject to on e = ± 9 0 .1021) Jo 3r2 dtf d0 2 J o We integrate the first integral by parts and Equation (5. r > a (5. Example 5.b) y" 1/ \T=o FIGURE 5. which can be obtained from Equation (A. 0<r<a.
1023) yields T = a^£°sM mcosm0 o Using the inversion formula [Equation 5. T is expressed as rmamcoSm0 y_i<x. b) we deduce that T and T are even functions of 8.1023) I 1 1 r ia nm T = Jo r ^ d r = J r [r m ] 0 = ^  (5.1025) (5.1026) mcosm0 o The integrand has a simple pole at the origin.104)].1020a. From Equations (5. as shown in Figure 5. The indenture is in the form of a semicircle of radius £.= Acosm0 o Substituting the value of A into Equation (5. Equation (5.1022) is T = A cos m0 On 0 = ± 0 O . T is given by — (5.1028) .1024a to c) into Equation (5. we obtain ^ . d m (5 1 0 2 ? ) (5.488 d2T 9— ADVANCED MATHEMATICS ^ +m2T = 0 d0 2 where T is the Mellin transform of T. The appropriate solution of Equation (5.1027) can be written as T = . We choose the line integral to be along the imaginary axis of m with an indenture at the origin.L [ " ( y cosmO Jioo mcosm8 0 d m + / " £ ( y cosmO J_ie mcosm0o dm+ <" (a)™ cosrnO Jie mcosm0 o dm (5.b.c) Substituting Equations (5.1024a.1022) (5.1023).102.
1030b) = in The first and last integrals are evaluated by substituting m = iy j . we write m = ee i e /•is Um ( (5.1030a) £>0Jie " mcosme0 e^° J_.1031) [ (a)iy cos(iye) .1032a) = 1 Joo exp [iy In (a/r)] coshy9 dy (5. d y ] _ „ ' * iycos(iye 0 ) (5.> 0 J_ioo mcosme 0 dm = (5.PARTIAL DIFFERENTIAL EQUATIONS I 489 I m (m)< > i1 f) Z/ Re(m) FIGURE 5.1029) rn/2 a p cos m 9 dm = ljm i8e^_de (5. m I " ( a p cos m9 e ./2 eeie (5.1032b) y cosh y 6 0 .102 Contour integral for the evaluation of the inverse of Mellin transform To evaluate the middle integral.
490 ADVANCED MATHEMATICS f°° = Jo exp [iy In (a/r)] COsh(y9) dy (5. 32c. • So far. This process is illustrated in the next example.1032c) ycosh(y0o) Equation (5. and t. In general. we have considered partial differential equations in two independent variables. r.e"^ 111 ^]dy! 2%1 \ Jo ycoshy6 0 j (5. if we start with a partial differential equation involving three independent variables. Stastna et al.427) noting that cosh is an even function.1034b) ycoshy90 ye [Sin ( y i n (a/r)] dy (5. we find lim f £—>0 } i e (f)m cosm9 dm = f Jo exp [iy In (a/r)] COshy9 dy (5.103.1030b.1034c) 1 1 + 1 j n cosh Jo ycoshy90 We have the solution in the form of an integral.1032b) is obtained by using Equation (3. The diffusivities are assumed to be constant and we denote them by . Example 5.i . Similarly.1) transforms to reduce it to an ordinary differential equation. we need to apply two transforms to reduce it to an ordinary differential equation.1028) yields T = .1034a) ( = 0 OO in+ I Jo y 9 [2i sin (y In (a/r)] dy (5. we have reduced a partial differential equation to an ordinary differential equation.103. We need to invert (n . Similarly. an equation in n independent variables needs (n . Symmetry is assumed and there is no dependence on 0.1033) mcosm80 ycoshy0o Substituting Equations (5. (1991) considered the diffusion of a cylindrical drop through a membrane. the concentration is c 2 and it is a function of z. In the membrane K2. Applying the integral transform method.[in + f" C O S h y 9 [e*ln<a'r> . A circular cylindrical drop Kj of radius a and height h diffuses into a circular membrane K2 of infinite radius and of thickness Z as shown in Figure 5. 33) into Equation (5. It is assumed that Kj is filled with a penetrant of concentration c 1 which is a function of z and t only.1) times so as to obtain the solution.
we have 9c2 1 3 / 3c2\ 32c2 — L = JS 2 r — .1036b) 3t c2 t=0 L r 3r ' = 0 9r ' 9z2  .d) z=h where c^ is a constant.o H . we have ^ . ^ l dz ^1 3 z = Kf(t) z=0 (5. K (= J9 2 /J9i) is the ratio of the diffusivities.1035a) at Clt=0 dz2 = cj = (5. * I a —»H ' l T h I C« .= ^ ^ 1 Cylinder K t and membrane K 2 (5.+ (5.1035b) o ._ f FIGURE 5.1036a) (5. In K2.1035c.PARTIAL DIFFERENTIAL EQUATIONS I 491 <©! and <0 2 in Ki and K2 respectively.103 In K^. The relevant equations are obtained via Equation (A.IV2).
c) where C^ ( = 1 transform of f (t). . = { z=0 (0. J9 2 _ c + 2. = ° 0<r<a (5. z=h /•oo dC.1 (5.1O35a to d).d) r>a (5.1037a) = 0. / . we obtain d2C s C j .1037b. t) dt ) is the Laplace transform of c^ Likewise F(s) is the Laplace The solution of Equation (5.1 dC. e" st Cj (z.1036a).101. .1036c.+ . we obtain Ci = K p J a 7 c o s h ( ^ V s ^ 7 + c f (51M9) sinhChVs/JS)! ) From the table of the Laplace transforms.492 a ADVANCED MATHEMATICS ^ dz (f(t).1037a) is . rL dz (5.1036e) C2z=i Taking the Laplace transform with respect to t of Equations (5. Proceeding as in Example 5.+Bsinh lz A &.1041) 3t [ . we obtain —2. 3z2 = a / 2 (5. ^L az = KF(s) z=0 (5.c f = JS)Y — . we first take its Hankel transform. we find that cj is given by °1 = irf0 m 1+2f (^"^{(if f (t^JSiJcosHILi2^ dx + cf (5.1040) To solve Equation (5. co C. ) s where A and B are constants. .1037b. = Acosh Z . c).1038) I V «©! ) I V JS>. Imposing conditions (5.S .
1044a.b) where F (s) is the Laplace transform of f (t). Using Equations (5.1042e) = f <W> Equation (5. (5. .1042b. „ . . C2 n n z=i = 0 ' d^2 AT a z z=0 = F(s)aJ1(qa) zr a . we take its Laplace transform. Equations (5. The Hankel transforms of Equations (5.1043) is C 2 = A cosh pz + B sinh pz P = V a 2 + s/«©2 where A and B are constants.1043) where C 2 (= I e ~ s t C 2 d t ) is the Laplace transform of C 2 .1042d) <5. (5.1044a. To remove the time derivative in Equation (5.1046) . on taking their Laplace transforms.1042c) = f(t) I rJ 0 (ar)dr JO (5.1042a.1042e) is obtained by using the properties of Bessel functions.1036b to e) are given by C2 t=0 = 0. we obtain (5. . ..1045) (5. The solution of Equation (5. Note that C2 is a function of z and t.b) —2dz = I rJ0(ar)f(t)dr z=0 Jo (5. C2z=i = 0 (5. b). The resulting equation is s C 2 = JE)2 cc 2 C 2 + L j» 00 2d z 2 .1041).PARTIAL DIFFERENTIAL EQUATIONS I 493 f°° where C 2 (= I r c 2 Jo(otr) dr) is the Hankel transform of c 2 . e) become.
1048) r c2 = Jo = aC 2 J 0 (ar)da 2 a ^ f^^j^f^jr^nl * '° '° nTl exp{.1040. 5. We need to introduce an auxiliary function and the method can be complicated. The solution is often expressed as an infinite series.. s)f(t)dt (5.1049b) .. (2n .. then c 1 and c 2 can be determined from Equations (5. the method of separation of variables cannot be applied directly. 49) respectively. we find that (5. we obtain (5.  2i which is the solution given by Stastna et al. If the equations are nonhomogeneous.X) n . _ \ .. .e ) ap cosh 3i From the table of Laplace transforms. (1991).494 ADVANCED MATHEMATICS C2 = F ( s ) a J ' ( a a ) s i n h P ( Z . The method of integral transforms can be applied to linear differential equations (ordinary or partial) with arbitrary boundary and initial conditions.1049a) f [2 I L 2 /2n .111) . \2i / J ( t .n(2nl)(ze)ttdx 2i Using Equation (5.. If f(t) is specified.1 \2 . .a +7T (5. .1) (z .11 SUMMARY The method of separation of variables is suitable for homogeneous equations with homogeneous boundary conditions. — '— dx da (5. The problems solved by the method of separation of variables can be solved by the method of integral transforms..102).T ) <O7> sm . . The various transforms F(s) of a function f(t) introduced in this chapter can be written as rb F(s) = Ja K(t.1047) s.
In the next chapter. s) e~ st e~ ist sin (st) cos (st) sin £2Ei cos^i J—/ t J n (st) ts~l . can be complicated. that is to say.PARTIAL DIFFERENTIAL EQUATIONS I 495 where K (t. Other kernels than those discussed in this chapter have been considered by Latta in the handbook edited by Pearson (1974). finding f(t) given F(s). TABLE 5. The methods of separation of variables and of integral transforms are applicable to linear equations only. it might involve evaluating an integral which cannot be written in a closed form.111 lists the kernel corresponding to various transformations. s) is the kernel. Numerical methods can be used to evaluate the integral. Extensive tables of integral transforms are given by Erdelyi et al. the solution is obtained immediately. Otherwise. Table 5. (1954). we shall consider methods which can be used to solve nonlinear equations. If f(t) can be read from a table. The inverse process.111 Kernels of various transforms Transforms Laplace Fourier Fourier sine Fourier cosine Fourier finite sine Fourier finite cosine Hankel Mellin a 0 <*> 0 0 0 0 0 0 b °° oo °° °° L L oo °° K (t.
. As a first approximation..x 2 ) Answer: 5 x i / x 2 .= 2x l U given that u(0. They can be written. 1 given that u(xj. p and vx are given by P = Po(1+X).0) = Xj du du given that u = 5 on x 2 = x^ 2 Answer: ( x j . Solve the following equations (i) H..496 ADVANCED MATHEMATICS PROBLEMS 1 a. The direction of flow is taken to be the xaxis and the flow velocity vx is assumed to be a function of x and t. in the absence of a pressure gradient. x 2 ) = (sinhx 2 )/x 2 Answer: L sinh (X 2 + x 2 ) 2b. Obtain p and v x if initially (t = 0)..••. v x = v o ( l +x) Answer: v o (l + x ) / (1 + tv Q ) po(l+x)/(l+tvo)2 3b. as 3vv 8vY —— + v —— = 0 at x dx where p is the density and vx is the velocity. du n du „ (m) x 2 ^ .2 X l x 2 ^ . The equations governing onedimensional compressible flow are the equations of continuity and motion given in Appendix I and II.+ H = 0 . the shear stress x between the glacier and its bed is given by t = pgha where p 0 and v0 are constants.. We model the downhill movement of a glacier as the sliding of a block of height h down a plane surface of constant slope a.
Il)] can be written as dh d .. .^— + x 2 3 — + v = 0 if we assume u = v + i. n Assuming that h is a function of x and t.(X!2 + X22) Obtain u if on x = 1. + x f ) + — — — . and elliptic. Show that the nonhomogeneous equation 8u 3u 2 2 reduces to 3v dv _ x.(1 + x 2 ) i 2 2N e x p ( . (I) d 2u .(x. 4a.PARTIAL DIFFERENTIAL EQUATIONS I 491 where p is the density and g is the gravitational acceleration. parabolic. 82u _ +4 = 0 dxf (n) 32u dxldx2 + xj 32u ax2 + x2 82u _ = 0 3Xl2 a X l ax 2 ax . Determine the regions where the following equations are hyperbolic. +4 d 2u . u is given by u = e * 2 cosx 2 + ^. It is further assumed that the relationship between t and v x is v x = kT n where k and n are constants.— ^ ^—lL 3 l z xj 5a. The equation of continuity [Equation(A.x 2 2 / x 1 2 ) c o s ( x 2 / x l ) Answer: J. obtain h.
0) = Xj2 9a. 5x 2 9 0 Find the characteristic coordinates £. What condition(s) has (have) to be imposed to satisfy u(xj.c o s x 2 . 0) = x (n .498 ADVANCED MATHEMATICS 92u (ill) X.Xj Rewrite the equation in terms of L. dxf 6a. X7 o 1/. and r\.x 1 sin x 2 + cos x 2 . . ^dt 3v =0 t=o Answer: sin x cos c t 8 a. Solve the equation ^ L + ^H_ = o 9x Bx2 by the method of separation of variables. Compare the solution with the solution obtained in Problem la (i).46b)] given that the initial conditions are y (x.sin Xo) + 2 cos x 7 2 2 3x 9 u v 3 u 3x _ 1 u = 0 2 23Xl3x2 2 Answer: sin x 2 . 0) = sin x . t=0 u (x.x) Answer: & Y K s=o cos (2s + l ) t s i n ( 2 s + l ) x (2s + I ) 3 . subject to the given conditions (i) ^ = ^ 0<X<7l. 7a. t) = 0 . = 0. Use d' Alembert's method and the method of separation of variables to solve the wave equation [Equation (5. i . l _ o + Xo Z a2u 9x22 o 3u +U = 0 h X. rj for . 2 x ^ u (cos X. t) = u (rc. t > 0 9t 2 3x 2 ~ 9t u (0. Solve the following equations.
. at ax2 0 < x < n. „ „ 0 < x < 7t. t) = y (1. k is a constant u(x.y ) sin(2sl)x (4s2l)(2s3)sinh(2sl)7t 10b. we have assumed that the frictional resistance is negligible. y) = u (71.. u (x. t > 0. 0<y<7t. t) = 0. f dt = v0 5 ( x .l ) t=0 l where v0 is a constant and 8 is the Dirac function. t) is the temperature and a is the thermal diffusivity. the wave equation becomes ^y + k ^y = c 2^y at2 at ax2 where k is a constant. The onedimensional heat equation can be written as (see Appendix III) dT 32T — = a 3t 3x 2 where T (x..k 2 / 4 lla. ~l sinh(2s~ ^ f r .. If the resistance is assumed to be proportional to the velocity. t) = y (x. 0) = sin2 x 1 n u (0. In deriving the wave equation. t) = U(TI. 0) = sin 3x Answer: e~ 9 k t sin3x u (0. The boundary and initial conditions are . 0) = 0 . oo Answer: Y s=0 vo e~kt/2 sinq 2 s + 1 t sin(2s + l)7cx q2s+1 = c 2 ( 2 s + l ) 2 7 C 2 . rc) = 0 . (m) 3 u 3x 2 9 + a u 3y 2 9 „ =0. Solve the heat equation for a rod of unit length. Answer:  Y. y) = u (x. Obtain the solution if y satisfies the following conditions y (0. .PARTIAL DIFFERENTIAL EQUATIONS I 499 (ii) — = k .
Deduce the boundary conditions. The equation governing the diffusion process is (see Appendix IV) 9cA i — * ^ ATI IT. 0 < y < a is given by . By writing c = c A~ c f cocf obtain the partial differential equation governing the process of diffusion. The blood is in direct contact with a rising stream of oxygen.500. Hershey et al. An^en i ^ i t l f exp [(2s + 1)2K2 J3ABz/4L2vz] ^ O i + i l " +Cf K 13a. 0) = 2(lx). c A (x. y) in a square 0 < x < a . which is the maximum velocity. Note that the boundary conditions are now homogeneous at x = 0 and at x = L. ADVANCED MATHEMATICS 2x. t) = T (1. as shown in Figure (5. 0) = c 0 and at the wall x = L . The appropriate boundary conditions are c A (0. (1967) considered the transport of oxygen into blood down a wetted wall column. d2°A z Ac 9 dz dx z where c A is the concentration of oxygen in the blood and «©AB *S the diffusion coefficient. T (0. the velocity v z may be approximated by the velocity at the interface which we denote by v m . ~^ dx = 0 Assuming that the oxygen does not penetrate very far into the blood. t) = 0 . T (x. O<X<1 2 ^x<l Q V (I) s e^ 2 s + 1 ) 2 a 7 l 2 t sin(2s+l)7cx Answer: 8 > —s=0 7C2(2s+l)2 12a. tZ (2s + 1) 2L The steady temperature T (x. z) = c f .71). A film of blood of uniform thickness falls with velocity vz.
y) = T (a. In Chapter 3. x=a T(x.68. is written as r 2 3r \ 3r / r 2 sin 8 36 \ 36 / .0)=^ d x °° sin ( 2 s + l)7tx/a sinh (2s + l)7ty/a (2s + I ) 3 sinh (2s + 1)71 s=0 = ^I x=0 d x =0.PARTIAL DIFFERENTIAL EQUATIONS I 507 ^1 + ^1 = 0 3x 2 dy2 Obtain the temperature T for the following boundary conditions: (i) T (0. Laplace's equation is written as 9r 2 r 3r r 2 dQ2 The appropriate boundary conditions are r = a.x ) z °° A 8a "V cos (2s + l)7tx/a sinh (2s + l)7ty/a Answer: =— > ^ —l— 7t2 s = 0 (2s + I ) 2 sinh (2s + 1)K 14b. Determine the potential cp for the flow of an inviscid. In the polar coordinate system. Consider the flow past a circular cylinder of radius a with its centre at the origin.a) = x ( a . it is shown that the potential cp associated with the flow of an incompressible inviscid fluid satisfies Laplace's equation. Answer: v x (r + a 2 /r) cos 0 15a. y) = T (x. assuming spherical symmetry. 0) = 0 . A T (x. dr r —> °°. incompressible fluid past a sphere of radius a.x) •* * f * _ V* K3 (ii) T(x. Obtain (p and compare your answer with that given in Example 3. cp —> r vx cos 9 where vx is constant (velocity far from the cylinder). a) = x ( a . Laplace's equation in the spherical polar coordinate system. 9q> ^— = 0 .
on the surface (r = 1). we consider a very simple model. show that the equation is reduced to one with constant coefficients. the boundary conditions are =— 3r =0. «©AB *S ^ ne diffusivity. Determine c A . The molecular diffusion of species A inside a spherical drop is described by (see Appendix IV) at " ^[r^drl dr ) _ where c A (r. By writing u = r c A . There is no source of heat. E s=l ( 1 \S+1 2 2 e~s n irii sTtr ^^l sin srcr + cc s . c A is finite at the origin.502 ADVANCED MATHEMATICS Choosing the centre of the sphere to be the origin of the coordinate system. The quantities cQ and c s are constants. We assume the earth to be a sphere with constant thermal properties. For such a model. Answer: 2/3 17b. T = sin2 0 on the surface of the earth (r = 1). c A = c s . o aT _ =0 ^— r — + = sin 0 — r 2 dr\ drl r 2 sin 0 dQ \ 90 T is finite at the centre of the earth (r = 0). The temperature distribution on the surface of the earth has a maximum at the equator and a minimum at the poles. cA = c0 . The temperature is at steady state. To find the temperature inside the earth. the temperature T satisfies the following equations i a / 2 dT\ i_ a /. The initial and boundary conditions are t = 0. t) is the concentration. r=a (p —> v^ r cos 6 as r —> °° Answer: vTO (r . Determine the temperature at the centre of the earth. Apply the appropriate substitution such that the boundary condition at r = 1 is homogeneous.a3 / 2r 2 ) cos 0 16a. Determine the initial and boundary conditions on u.
at the surface. «©AB *S the diffusivity. T(0.0) = 0 . cA = 0 .*«* then c A is given by cA = k j fe^'dt'+fe"1"1 Jo Show that c A and f satisfy the same initial and boundary conditions. t ) = Tj Solve the differential equation. If heat is produced at a rate proportional to the temperature T. T ( l . 18b. and kj is the rate constant for the reaction. The equation governing the diffusion of species A in a medium B and reacting with it irreversibly according to a first order reaction is given by (see Appendix IV) ^CA _ ra v 2 c k c ^j~~ ~ d J AB v C A~ K 1 C A where c A is the concentration. The initial and boundary conditions are t = 0. Then. for the case kj = 0. Can you use the method of separation of variables to solve the equation .t) = 0 . use the result given in Problem 18b to obtain the solution for nonzero values of k^. then T satisfies the equation where a and kj are constants. Assume that the initial and boundary conditions are T(x. Verify that the solution satisfies the differential equation. the initial and the boundary conditions. 20b. c A = c s Verify that if f is the solution of I . subject to the given initial and boundary conditions. 19b.PARTIAL DIFFERENTIAL EQUATIONS I 502.
The velocity component vz of a Newtonian fluid flowing in a long circular pipe of unit radius under a pressure gradient is given by [see Equation (1. vzt(l. If the region of interest is 0 < r < 0. Answer: 4r 2 (1 . then the transient velocity v zt is given by 3t r ar \ dr j where v is the kinematic viscosity.5 and given that u (0.+ — . v zt is finite at the origin . 0) = v m ( l . Obtain a series solution for R (see Chapter 2). 6) = sin 20 obtain an approximate solution for u.+ (xz + yl) u = 0 ax 2 a y 2 Transform to the polar coordinate system and show that the equation can be written as 4 1 d I du\ 1 d2u r— + +r u = 0 r 9r * dr > r 2 90 2 By writing u = R (r) 0 (0).0 2\2 ~ ADVANCED MATHEMATICS —.r 2 ) . The initial and boundary conditions are v zt (r.r 2 ) where v m is the maximum velocity (velocity along the centre line). deduce that F and 0 are solutions of r f (rdR) dr I dr ) + (r 6_n2)R = 0> ^0 d02 + n 2 0 = 0 where n 2 is the separation constant.1920)] vz = v m ( l .r 6 / 6 0 ) sin20 21b. If the pressure gradient is stopped.t) = 0.504 d U d U .5.
4v m J 2 (2. Chiappetta and Sobel (1987) have modeled the combustiongas sampling probe as a hemisphere whose flat surface is kept at a constant temperature T c . 0)] dr u where k is the thermal conductivity and h is the heat transfer coefficient. ^ ) = T (r. The system is steady and the temperature T at any point inside the hemisphere satisfies Laplace's equation. By considering only the first term of the infinite series. estimate the time required for the velocity at the centre line to be reduced by 90%. which in spherical polar coordinate. 23b. Note that the boundary conditions are not homogeneous and we introduce u defined by u = TTc Solve for u and then.4 ) 2 v t J 0 (0) y Answer: — n L ^ [2. T is finite everywhere in the hemisphere. as a model for a turbulent flow in a channel.4)] 2 22b. Show that the transformation (ColeHopf transformation) .(a. 0) = h [T r .T (a. Burgers (1948) has proposed the equation 9v 3v d v —+ v— = v—3t dx 3x z where v is a positive constant.PARTIAL DIFFERENTIAL EQUATIONS I 505 Solve for vzt. obtain T. Gas at constant temperature TQ > TQ is flowing over the probe.4)e( 2 . can be written as (see Appendix HI) dr I dr / sin 0 30 36 J 2 2 where a is the radius of the hemisphere.f ) = T c . The boundary conditions are on the flat surface: T (r. dT on the curved surface: k ^.4J 1 (2. .
Obtain v (x. t>x2/2 . t) = v (a. 0) = 0 . t) = 0 . U (7t.. By introducing a suitable auxiliary function.506 ADVANCED MATHEMATICS V (X. 0<t<x2/2 tx2/2.t) = 0. t) if the initial and boundary conditions are v (x. x(lx2) + V Answer: — > (~l) S 2 cos sTCct sm srcx 6c 2 s=1 (sn)3 (n) . 0) = X/7T d u d u  t o Answer: — + ^ (e~l . subject to the given conditions: /•x (l) 5 u = c at2 ax2 u(x. t) = A ~ [in 9 (X.l ) V a v < > / 2 v ] 24b. solve the following equations by the method of separation of variables.. v (0.0) = ^ dt 2d u +x.t) = t Answer: 0.. U (X.( . 0 < x < a. u (0. t) = 0 ~ 2 2 Ase s 2 Answer: 2 J s=0 " vt/a cosrcsx/a As = ^ [ l / ( v 2 / 4 v 2 + s V / a 2 ) ] [ l . Use the method of Laplace transforms to solve the following equations: (i) —• + x  ^ = 0 .+ e sin 3 x at ax2 U (0. n 1 0<x<I =0 t=0 u(0.. Determine A and solve for 9 by the method of separation of variables. u (x. — = — . 0) = v0. t) = 1 .e~9t) sin 3x 25a.t) = u(l. t)] dx transforms Burger's equation to the diffusion equation for a suitable choice of constant A.
u(x. c and g are constants. t) = 0 x ~ >o ° T (a .g . t>0 T (x. . The problem can be expressed as (see Appendix III) 9T 32T — =a. Layer Kj (0 < x < a) is initially kept at a temperature To and is in perfect thermal contact with layer K2 (a < x °°) which is initially kept at zero temperature. 0) = T o . 0<t<x/c t>x/c . t>x2 (in) ^ = c 2 ^ . t=0 x>0 x—>oo d x lim ^ r — = 0. t) (continuity of temperature at the interface) k j 3— dx = k 2 3— x=a_ dx (continuity of heat flux at the interface) x=a + Solve for T using the method of Laplace transform. 0) = 0 . a j and k2. t > 0 . [t2(tx/c)2]. .x l .+ 2 x ^ . t) = 0.X ) v .O) = u ( O . 0<x<a. _ Answer: T n 0 T 0 ( l . lim T(x. Answer: . x>a. ^dx x=o =0. a 2 . . ox at u(x. at dT 3x 2 32T — = ou . 0<x<a 2 La \ 2VaTT 2 VaTt s=0 1 J L 1 J. 0<t<x2 (t+l)(tx2). [(2s + l ) a + x l  .O) = ^ dt =0. t) = T (a + 0. dt 2 dx2 u (0. x > a at ax2 T(x. 26b.PARTIAL DIFFERENTIAL EQUATIONS I 507 (ii) ^ . t ) = l Answer: t + 1 . n { ^ (2s + l ) a .0. A composite solid is made of two layers Kj and K2. ^—> X erfc 1—— +erfc . The thermal conductivity and diffusivity in Kj and K2 are respectively ki. 0 < x < a . t>0 .g t /2.= 2x. The face x = 0 is insulated.
( e ~ s y .508 ADVANCED MATHEMATICS lolLt^V s=0 Xn e r f c [ 2 s a + L H(xa)l_erfcr(2s 1 J L + 2)a + H ( x . Answer: . y>0 ax 2 a y 2 u (0. =_p. + ^H. 0) = 0 u is bounded throughout the region.. 7U ** Bs B .a ) l ) 1 J/ x>& 27b. u (re. t>0 at2 subject to u = ^at dx2 ax 2 at 2 = 0. x>0. Show that the solution of 32u = 32u + 34u . t = 0. u (x.] 28b. can be written as u — > 0 as x—>c» for all t > 0 J Deduce that dx rY+i°° Jexp(sx/Vl+s2 FV s +st)ds Y_loo x=0 [Hint: Use the table of Laplace transforms. y) = 0 . t) = 1. x>0 u(0. Use the method of Fourier transform to solve Poisson's equation ^H. y) = 1 .1) sin sx s = ^{(l)Ss + p[l(l)s]/s) s . 0<X<TT.
2 £09 29b.u = 0. u (x.e~ a ' . u (x. 0 < y < l x 3x2 subject to ~ 3y2 = 0.] Solve by the method of Fourier transform the equation — . y) —> 0 as x —> ± < * > °y y=o Leave your answer in integral form. . 1) = e" x . [Hint: Complete the square in the exponential term and use the appropriate contour.PARTIAL DIFFERENTIAL EQUATIONS I 2 . Show that the Fourier transform of e" ax is y j . oo< <oo.+ — — .
.
we have used the method of characteristics to solve first order partial differential equations and the wave equation (d'Alembert's solution). we have used the method of separation of variables and transform methods to solve partial differential equations.b) . a method which has been encountered in Chapter 5. We now describe Riemann's method of solving hyperbolic equations. y) (6. Discontinuous initial data are carried along the characteristics.E. y ) .CHAPTER 6 PARTIAL DIFFERENTIAL EQUATIONS II 6.22a.D. and parabolic. We recall that the characteristics are the curves along which information is propagated.D.2la. and for each type we discuss one method. and a similarity variable is introduced to solve parabolic equations.D.1 INTRODUCTION In Chapter 5.'s.E.'s. hyperbolic.b) +a(x'y)a^~ + b<Xy)^p We assume that the prescribed conditions are on a curve y and that they can be written as u = g(x. y) (6. and Laplace's equation are all important equations in classical physics. the hyperbolic equations are solved by the method of characteristics. y)u = p(x. Here. elliptic. It is also shown that hyperbolic equations have real characteristics.2 METHOD OF CHARACTERISTICS In Chapter 5.E.E. The canonical form of a linear hyperbolic equation is 32 L(u) = ^ d r) 3 +c(x. the diffusion (heat) equation. The method of Green's function which was employed in Chapter 1 to solve nonhomogeneous O.D. We consider also nonlinear equations.'s can be classified in three types.'s is now extended to solve elliptic P. In this chapter. In this chapter. 6. The equations we considered in Chapter 5. we seek the solution of Schrodinger's equation which is one of the most important equations in quantum mechanics. P. ^ = h(x. we introduce additional methods which can be used to solve P. namely the wave equation.
512
ADVANCED MATHEMATICS
where ^— is the normal derivative of u on y. on We also assume that y is not a characteristic of Equations (6.2la, b), as shown in Figure 6.21.
y
l
y0 •
°V—•*—
1
^
x0
*
FIGURE 6.21
Characteristics Tx and T 2 and curve y along which conditions are prescribed
The characteristics are given by [Equations (5.31 la, b)] Fj : x = constant, F2 : y = constant (6.23a,b)
We now determine the value of u at a point P (x 0 , y0) which is a point of intersection of two characteristics. The solution at P is determined by the domain of dependence A (see also Figure 5.43). The adjoint operator L* of the operator L defined by Equations (6.2la, b) is [see Equation (6.51) for the general case]
L* W
= iNr  T~(av) " T" ( b v ) + cv
dxdy ox ay
(62"4)
If L* = L, L is selfadjoint.
PARTIAL DIFFERENTIAL EQUATIONS II
5/3
From Equations (6.2la, b, 4), we obtain
v L (u)  u L* (v) = ~ (auv  u p) + ^ (buv + v ~)
dx dy dy dx Applying Gauss' twodimensional theorem [Equation (4.423)] to Equation (6.25), we obtain  J [vL(u)uL*(v)]dxdy = < f [ ( a u v  u ^ ) dy (buv + v—) dx]
A C
(6.25)
(6.26)
where C is the closed curve formed by y, T2, and F j , as shown in Figure 6.21. Noting that on Ti, dx = 0 and on F2, dy = 0, it follows that
/ r/ dvN , . , du. , , / r, 3v. , , , 3u. , , d) [(auvUr) dy (buv + v^—) dx] = [(auvu^r—) dy(buv + v^—) dx] I dy dx I dy dx C Y
+ 1 (auvu^)dy  I (buv + v^)dx
(6.27)
JR
°y
h
dx
Integrating I
lQ
v 3— dx by parts yields
dX
du
Jp
(6.28)
Combining Equations (6.26 to 8), we obtain \l
A
[vL(u)uL*(v)]dxdy = I [ ( a u v  u  ^ ) dy (buv + Vy) dx]
Y
+ 1 ( a u v  u ^ ) d y  v(Q)u(Q) + v(P) u(P)  I
(buv  u  ^ ) dx (6.29)
Equation (6.29) expresses the value of u at P in terms of v and the sum of integrals. By choosing v appropriately, we can reduce some of the integrals to zero. Following Riemann, we impose on v (the Riemann function) the following conditions
514
ADVANCED
MATHEMATICS
(a)
v satisfies the adjoint equation, that is to say L*(v) = 0 (6.210a)
(b)
along x = constant (Fj) j p = av (6.210b)
(c)
along y = constant (F 2 )  ^ = bv (6.210c)
(d)
at the point P v(xo,yo) = 1 (6.2lOd)
Substituting Equations (6.210a to d) into Equation (6.29) yields
u(P) = v(Q)u(Q)J [(auvu^)dy(buv + v^)dx]+j j vpdxdy
Y A
(6.211)
Equation (6.211) is not symmetrical in the sense that it only involves the point Q and the derivative 3— but not R and ^—. To obtain a more symmetric expression, we consider the identity
? ? fR [ /  (uv) dx + ^ (uv) dy] = d(uv) = u(R) v(R)  u(Q) v(Q) ;Q dx dy JQ
fR
(6.212a,b)
Substituting u(Q) v(Q) from Equations (6.212a, b) into Equation (6.29) and using conditions (6.210a to d), we obtain u(P) = u(R)v(R) I [ ( a u v  u ^  ) d y  ( b u v + v ^  ) d x ] J dy dx
Y
  f^(uv)dx + ^(uv)dy]+(I vpdxdy
Y A
(6.213a)
= u(R)v(R) [(auv + v ^  ) d y  ( b u v  u ^ ) d x ] + I I vpdxdy
Y A
(6.213b)
PARTIAL DIFFERENTIAL EQUATIONS II
511
Adding Equations (6.211, 13b) yields u(P) = i[u(R)v(R) + u(Q)v(Q)]+
A
I vpdxdy
if
Y
[(2auvu^ + v^)dy(2buv + v ^   u  ^ ) d x ]
(6.214)
The function u and its first partial derivatives on y are given by Equations (6.22a, b), R and Q are on y; that is, u(R) and u(Q) are known, p is given in A, and if the Riemann function v is known, u(P) can be evaluated. The problem of obtaining a solution to Equations (6.2la, b, 2a, b) has now been transformed to finding the Riemann function v satisfying Equations (6.210a to d). Once v is found, u(P) can be obtained from Equation (6.214). Note that Equation (6.214) can be written in a general form as
u(P) = J K£,P)p($)dS
(6.215)
where the kernel K depends on the operator and the boundary conditions, and is independent of p. The Riemann method is similar to the Green's function method [see Equation (1.181)]. Note that we have assumed that each characteristic intersects y at no more than one point. This condition is necessary, otherwise the problem might have no solution. Example 6.21. Solve the following equation using Riemann's method
JH_J*i = O
dxdy subject to u (x, y) = f (x, y ) ,  ^ = h(x, y) 2x 3y
(6.216)
(6.217a,b)
on the line y =  x + c, where c is a positive constant. The characteristics are given by Equations (6.23a, b) and the adjoint operator [Equation (6.24)] is
J^
dxdy
+
±?I=0
2x dy
(6.218)
516
ADVANCED MATHEMATICS
On integrating with respect to y, we obtain
£ + £ = FW
where F (x) is an arbitrary function. The integrating factor of Equation (6.219) is Vx and the solution is Vx~ v = I Vx~ F (x) dx + G (y) or v = F1(x) + G(y)/Vx"
(6.219)
(6.220a) (6.220b)
where Fj and G are arbitrary functions. The function v has to satisfy Equations (6.210b to d) and this implies that on x = constant, on y = constant, at the point P(x 0 , y 0 ) , G'(y)/Vx~ = 0 r 1  r G ( y ) / x 3 / 2 = v/(2 x)
\X A. £*
(6.22la) (6.221b) (6.221c)
v(x 0 , y o ) = l
From Equations (6.22la, b), we deduce that G is a constant and dF, F, r1 + TT = ° dx 2x The solution is Fj = K o / V 7 where KQ is a constant. Combining Equations (6.220b, 21c, 23) yields v = Vxo/x The solution u (x 0 , y 0 ) is given by Equation (6.214) and, in this example, it simplifies to u(x 0 >y 0 ) = ^[u(R)v(R) + u(Q)v(Q)] + l  7 ( v  ^ + ~  + v  ^   u — ) dx (6.225) (6.224) (6.222) v ' (6.223)
PARTIAL DIFFERENTIAL EQUATIONS 11
511
where y is the line y =  x + c and on y, dx = dy. From, Figure 6.21, we deduce that Q and R have coordinates ( c  y 0 , y 0 ) and (x 0 , c  x 0 ) respectively. From Equations (6.217a, 24), we obtain u(R) = f ( x 0 , c  x 0 ) , v(R) = 1, u(Q) = f ( c  y o , y o ) (6.226a,b) (6.226c,d)
v(Q) = V x o / ( c  y o )
Using Equation (1.513), Equation (6.217b) can be written as ^  + ^  = V2"h(x,y) ox ay Combining Equations (6.217a,b, 24, 25, 26a to d, 27) yields (6.227)
u ( x o , y o ) = i f ( x o , c  x o ) + V x o / ( c  y o ) f ( c  y 0 , y o ) + I
•/cy0
[(V2x Q /x) h(x, cx)
+  (Vx o /x 3 ) f (x, cx)] dx The functions f and h are given and the value of u at any point (x 0 , y 0 ) can be calculated. Example 6.22. Solve the telegraph equation ^  c at 2 subject to u (x, 0) = f! (x), by Riemann's method. ^
dt
(6.228)
2
^ ax 2
+
a ^ at
+
Pu = 0
(6.229)
= f 2 (x)
t=0
(6.230a,b)
Equation (6.229) arises in the transmission of electrical impulses in a long cable with distributed capacitance, inductance, and resistance. Note that if a = P = 0 (no loss of current and no resistance), Equation (6.226) reduces to the wave equation [Equation (5.1.1)]. We can simplify Equation (6.229) by assuming that u is of the form
518
ADVANCED MATHEMATICS
u(x, t) = 0(t) w(x, t) In terms of 6 and w, Equation (6.229) becomes
(6.231)
(6.232)
U2
d*2
at ' at
'
at2
dt
/
Replacing u (x, t) by two functions 9 and w in Equation (6.231) allows us to impose an additional condition on say 9 so as to simplify Equation (6.232). We choose to set the coefficient of 3w 3t 2d0~ + a 9 = o dt The solution of Equation (6.233) is G = C0eat/2 where C o is a constant. Substituting Equation (6.234) into Equation (6.232) yields
a w 2a w „! ,
c
to zero. That is to say,
(6.233)
(6.234)
.
c at 2 where K' = p  a 2 / 4 ax 2
+ Kw = 0
(6.235a)
(6.2~35b)
The characteristics of Equation (6.235a) are [Equations 5.420a, b)] ^= x + ct = constant, r = x  ct = constant (6.236a,b,c,d)
Using £ and r\ as the new variables, Equations (6.230a, b, 35a) become —  + Kw = 0 , K = K'/4c 2 (6.237a,b) (6.238a)
a^dn
w
U=n = f l ^ )
PARTIAL DIFFERENTIAL EQUATIONS II
Ml
C[lT
~^ )
= f2& +
2"^
= Sl^
(6.238b,c)
We note from Equations (6.236a, c) that t = 0 corresponds to £, = TJ. We evaluate the value of w at a point P(£ 0 ,r 0 ). The point P, the characteristics, and the curve y on which w and its derivatives are prescribed by Equations (6.238a to c) are shown in Figure 6.22.
V
% p
I
•—y*
/
r, /
/
FIGURE 6.22
Characteristics r x and T 2 of the telegraph equation
The Riemann function v has to satisfy Equations (6.210a to d) and the operator L defined by Equation (6.237) is self adjoint. The function v is a function of £, and r) and depends also on t,0 and r\ 0 . We write v as v (£, rj; £ 0 , r\ 0 ). The characteristics Fj and T2 are given by J^: ^ = ^0 ; T2:
TI=TI0
(6.239a,b)
The conditions that v has to satisfy can be written as J^L. + Kv = 0 (6.240a)
dfcdTi [v(^,Tio;^,Tio)] = 0 9^ (6.240b)
520
ADVANCED
MATHEMATICS
^[v(£0,Ti; $0,TI0)] = 0
V(^0,TI0; ^0,TI0)
(6.240C) (6.240d)
= 1
Rather than solving Equation (6.240a) right away, we first examine the form that v might take. We expand v about (^0, r\ 0) in a Taylor series and we obtain
Be,
Ol
z
d£,
+ ^ ( T 1  T 1 O ) 2 ^  + ( ^  ^ ) ( T 1  T V ) )  ^  + ...
(6.241)
All the derivatives are evaluated at the point (^0,f\0). and 3Tin a^ n  ^ = KvU _ = K
From Equations (6.240b, c), we deduce that
are zero at (%0> T)n) From Equations (6.240a, d), we obtain
(6.242a,b)
We deduce that
82nv
are the only nonzero terms and, from Equation (6241), we deduce
that v is of the form v = f [ ( $  £ 0 ) ( T I  T I 0 ) ] = f (s) On differentiating, we have 3v df 9s , . df — = = (r\  rj0) — 3£ ds 3^ ds .,_ .. , . (6.244a,b) (6.243a,b)
Ibu.«+
9^3TI ds
(n _ „„) ft_ 4o) 4
ds2
= « + . d!i
ds ds2
(62.44c4)
Substituting Equations (6.244a to d) into Equation (6.240a), we obtain the following ordinary differential equation
PARTIAL DIFFERENTIAL EQUATIONS II
521
s
d_l+df d s 2 ds
+ Kf
= O
(6.245)
We make a change of variable and write X = (4Ks) 1 / 2 Using the chain rule, we have dl ds
2
=
(6.246)
df dX dX ds
= 2K
(4Ks)1/2^ dX
2
(6.247a,b)
— = 4K2(4Ks)"3/2— +4K2(4Ks)~!^4 ds 2 dX dX Equation (6.245) becomes d2f+ldf+f
dX2
= Q
(6.247c)
(6.248)
X dX
Equation (6.248) is Bessel's equation of order zero and the solution which is regular at the origin (X = 0) is Jo (A,). The Riemann function v is v = 3Q(k) = J o [ V 4 K ( £  £ 0 ) ( n  T i 0 ) ] (6.249a,b)
We use Equation (6.214) to determine w(P). We note that p, a, and b are zero. On y (t, = r\), we find that w (P) is given by
w(P) = 1 [w(R) v(R) + w(Q) v(Q)]  1
2 2
JQI
[vff 5  — U w f^ ^ 1 d^
\dr^
5
W l^
3TVJ
(6.250)
The points Q and R are respectively (^ 0 , 4 0 ) and (T)0, T O ). The functions v and w required in Equation (6.250) are given respectively by Equations (6.238a to c, 49a, b). The required values of v, w, and their derivatives are v(Q) = v(R) = J 0 (0) = 1 w(Q) = f i ^ o ) , w(R) = fjCrio) (6.25 la,b,c) (6.252a,b)
522
ADVANCED
MATHEMATICS
(6.253)
3v
=
[VT(ni,) = , ; ( J
J ;
(6 2.54)
^
•a»_3i]
La^
_ r vTgg.)
3TU
J
,,»]
Jn=4
(6 2_55)
(6.2.56)
. [ViMW
^ Kft5o)(TiTi 0 )
Substituting Equations (6.25 la to 56) into Equation (6.250) and replacing the integration variable ^ by z, to avoid confusion, yields w(P) = i[f 1 (^ 0 ) + f 1 ( r i 0 ) ] + ^ I " "jQ^g^dz
•So
.ipf.w^Mg)!;^),,
2
(62.57a)
^0
V(Z^O)(ZT1O)
where ?4 = V4K(z^ 0 )(zT! 0 ) We revert to the x and t variables, £ 0 ^ 0 = x + ct, To = x  c t
an^
(6.257b) " H o a r e replaced by (6.258a,b)
The quantity Xj in Equation (6.257b) is now written as X.J = Y4K(zxct)(zx+ct) = / V4K[(zx) 2 c 2 t 2 ] Substituting Equations (6.258a to 59b) into Equation (6.257a), we obtain
/•XCt
(6.259a,b)
w (P) = A [f j (x + ct) + f! (x  ct)] + \ \ where
F (x, t, z) dz
(6.260a)
PARTIAL DIFFERENTIAL EQUATIONS II
523
F(x,t,z) = l  J o ( ? i 1 ) g 1 ( z )  [ V K " ( 2 c t ) f 1 ( z ) J o a i ) ] / [ ( Z  x ) 2  c 2 t 2 ] The solution u is obtained from Equations (6.231, 34) and is u = e"at/2 w
(6.260b)
(6.261)
Comparing Equations (6.261, 5.431, 32), we realize that, in the present example, u is damped. • An extension of the method described here to higher order problems is given in Courant and Hilbert (1966). Abbott (1966) has described and solved several problems of engineering interest by the method of characteristics. In Chapter 8, a numerical method for solving hyperbolic equations based on characteristics is presented. 6.3 SIMILARITY SOLUTIONS
In many physical problems, the solution does not depend on the independent variables separately but on a combination of the independent variables. In this case, a similarity solution exists and the existence of a similarity solution is usually associated with the process of diffusion. We describe the method of obtaining a similarity solution by considering a few examples. Example 6.31. Look for a similarity solution for the diffusion problem obtained by considering a generalized form of Equation (A.IV1)
^ = A[D(c)^,
dt dx L dxj c(0, t) = c 0 ,
X>0, t>0
c(x, 0) = c 2
(6.31)
(6.32a,b,c)
lim c (x, t) = c, ,
We introduce a similarity variable by writing
ri = xat^
where a and p are constants. We assume c to be a function of r\ only. Its derivatives are given by ^ dt
=
(6.33)
^ = d c _ P x « dr\ dt dr)
t
P  l
63
.4
b
'
(6 .3 4c)
f ( D  ° ) = a ( a  l ) x «  2 t P D ^ + a 2 x 2 « 2 t 2 PAf D ^\
3x \ dx! dr dr) \ drj
524
ADVANCED MATHEMATICS
Substituting Equations (6.34a to c) into Equation (6.31) yields
Px«tP1^= a(al)xa2t0D^ + a2x2a2t2eA(Dli]
dt dr dri \ dTjj
(6.35)
v
'
We now substitute, wherever possible, the combination of x and t by T and Equation (6.35) becomes (6.36) Both sides of Equation (6.36) are functions of Tj only if (x 2 /t) is a function of TJ. We choose r to be
r\ = x/VT
This choice implies that a = 1, p = 1
(6.37)
(6.38a,b)
Substituting Equations (6.38a, b) into Equation (6.36) yields 1qifi. = J L ( D ^  1 2 dt dr \ dr\J (6.39) '
Note that the value of a has been chosen so that the first term on the right side of Equation (6.36) vanishes. We now check that the initial and boundary conditions are consistent with the choice of TJ. From Equation (6.37), we deduce x = 0=»T = 0, x—>OO=>T—>oo, t = O=>T—>«> (6.3lOatof)
Equations (6.32a to c, 10a to f) imply that a similarity solution exists if ci = c 2 (6.311)
If D is a function of c, Equation (6.39) is a nonlinear equation and we can solve it numerically. It is generally easier to solve an O.D.E. than a P.D.E. Thus, looking for a similarity solution can simplify the problem. On the curves (TJ = x/Vt = constant), c(r) is a constant, hence the name
similarity solution.
To simplify the problem, we assume D to be constant and
PARTIAL DIFFERENTIAL EQUATIONS II
521
c\ = c 2 = 0 Equation (6.39) becomes
2
(6.312a,b)
^4+lTi^ = 0 dr ,2 2D dr The solution is
(6.313) ^ J
P
c = c3
2
exp(£ /4D)d^ + c 4 (6.314)
Jo
where C3 and C4 are constants. Note that the lower limit of integration in Equation (6.314) is arbitrary. That is to say, had we integrated between the limits 1 and T, the constants C3 and C4 would have been adjusted accordingly. However, since the boundary conditions refer to r\ = 0, the obvious choice for the lower limit is zero. Imposing the initial and boundary conditions [Equations (6.32a to c, 10a to f, 12a, b)] yields
exp(£ /4D)d£ =  c o i /  W
0
v
(6.315a,b,c)
7tu
Equation (6.314) can now be written as rx/it
c 2
= co
1
 \ H K \
exp(^ /4D)d^
0
(6.316)
v nu j
Example 6.32. Obtain similarity solutions for the twodimensional boundary layer flow over a flat plate of a Newtonian fluid. The twodimensional boundary layer equation of a Newtonian fluid can be written via Equations (A.I1, II1) as 3v 3v 3^ + ^ =0 dy dx dv
V
(6.317) dv^
= v
dv
+V
9 v
^
dx
y^
dy
T
+ V
7T
dy
(63"18)
dx
where v x and v y are the x and y components of the velocity, v^ is the free stream velocity, and v is the kinematic viscosity (see also Rosenhead, 1963). We define a stream function \/(x, y) by
526
vx
x
ADVANCED MATHEMATICS
= ir'
dy
v
v=?
y dx
(6.319a,b)
Equation (6.317) is satisfied automatically and Equation (6.318) becomes
3v_aY_8vKa^
dy dxdy dx dy2
= v
^
dx
+ vaV
(6320)
3y 3
The appropriate boundary conditions are y = 0, ^ = ^
dx dy
= 0,
lim ^
y—>°° dy
= VTO
(6.321a,b,c,d)
We assume T to be of a more general form than Equation (6.33) and write Tl = y / $ 0 0 where t, is to be specified later. Equation (6.320) suggests that \/ can be written as V = v 00 (x)^(x)f(Ti) Computing the derivatives of \/, we have (6.323) (6.322)
^
dx
=
l^^f+v
dx ^
^
"dx
f + v
4
« ^
=
^
f
+v
^ f  v ^ ^ f
°° dx °° dx drj
(6.324a,b)
(6.324c,d)
°°^ dr 3x = ^ ^ dx dr] ^
dx ^
^ = v ^ , 3y °°dr '
^ dxdy
^ ^ 1 £ dx dr 2
?!?.^i!f,
3y 2 ^ dri2
^
3y 3
=^4
£ &x\
(6.324e,f,
Substituting Equations (6.324a to f) into Equation (6.320), we obtain after some simplification
i!l +fejL(gv ) f ^ I + ( ^ ^ ) h  f f f] = 0
dr)3
(6.325)
L v dx
vs
°°;
dr2
V v dx /[
Urj J
For the existence of similarity solutions, we require that ^(^vj = a (6.326a)
PARTIAL DIFFERENTIAL EQUATIONS II
527
V IT=P
where a and P are constants. Equation (6.325) can now be written as
<6'3'26b)
i!l + af4+P[l(£fl = °
dr(3 dT]2 L ldTl/J
(6327)
Equation (6.327) is the Falker Skan equation. Various values of a and (3 are associated with various outer flow situations. The following cases have been examined. Blasius Flow In this case, we choose a =1 p = 0 (6.328a,b)
From Equations (6.326a, b, 28 a, b), we deduce vx = constant, £ = (vx/vj (6.329a,b)
This flow corresponds to the flow over a flat plate and Equation (6.327) simplifies to
£+lf£0
The boundary conditions [Equations (6.32la to d)] are f(0) = f'(0) = 0 , f — > 1 as r—>oo
(6.330)
(6.331a,b,c)
where the prime denotes differentiation with respect to T. Equation (6.330) is a third order nonlinear O.D.E. and its solution, subject to the three boundary conditions, can be obtained numerically. Stagnation Flow Substituting the values of a = 1, p = 1 (6.332a,b)
528
ADVANCED MATHEMATICS
into Equations (6.326a, b) leads to
(6.333a,b) The solution can be written as
£ = VT'
where c is a constant.
voo = cx
(6.334a,b)
Equation (6.327) and the boundary conditions are now
^f+f^Hff+l=O
dr^ and d t  2 Idty f' —> 1 as r— >~ f(0) = f'(0) = 0 ,
(6.335)
(6.336a,b,c)
The system defined by Equations (6.335, 36a to c) can be solved numerically. The solution of various flows is discussed in Rosenhead (1963). Example 6.33. Stefan problems are concerned with the phase change across a moving boundary. We consider the phase change from a solid to a liquid. The boundary between the solid [x < s (t)] and the liquid [x > s (t)] is given by x = s(t) and is shown in Figure 6.31. We assume that there is no motion and that the temperature T, in both phases, satisfies a simplified form of Equation (A.III1) (6.337)
pc? I = k^ 1
dt 9xz
(6.338)
where p is the density, c is the specific heat, and k is the thermal conductivity. We assume that they are constants and have the same values in both phases. At the interface, the temperature is at the melting point T m .
We use Birkhoff s method to obtain the similarity variables.PARTIAL DIFFERENTIAL EQUATIONS 11 529 t " X =S(t) SOLID / LIQUID 0 I ^ — »• x FIGURE 6. dt 8xJ x = s _ where L is the latent heat. k ^ = L 4s.339) (6.31 An energy balance yields the Stefan condition ^ T Solid and liquid phase i X=S.340a.a3 (6.b. In terms of the new variables. x = a"2 x . T = a"3 T (6.a 3 = 2 a 2 .341) The equation should be independent of a and this implies that a2 .c) where a\. and 0 C 3 are constants. C C 2 .338) becomes ^ p c a ai«3 <*I = 9t k a2a 2«3 LI 3x 2 (6. A parameter a is introduced and all variables are written as t = a" 1 1 . Equation (6.342) .
346a. For the interface. 45a. and is written as ^ + J . b).347) (6.E..b) Note that the same TJ was obtained in Example 6.343a.348) dr 2 2D dr) where D = k/pc.D. £2.342). t a 2 /ai T = —3— t a 3 /a] (6.338) is reduced to an O. We introduce a set of new variables defined by Ti = —^—.343a.344) One possible solution is ^ = 0. = 1 (6. The interface (x = s) is now given by r\ = s 0 The conditions at the interface are (6.530 ADVANCED MATHEMATICS We now combine the variables x. and T in such a way as to reduce the number of variables.b) The ratios OC2/CC1 and aL^/CLy can be obtained from Equation (6. leading to 1_^= a1 2^21 a{ a j (6. we assume that s is given by s = s o ff where s0 is a constant.T I ^ = 0 (6.T m Equation (6.345a. We define f(Tl) = T .350) (6.349) . x= T (6. b. t. we obtain T\ = X.31.b) Combining Equations (6.
52a) and f becomes (6. These methods can be extended to more than two independent variables.b) Equations (6.355). b) imply that.55) The interface (s0) is obtained by solving Equation (6.35 lb) yields r > s0 (6. we have the liquid phase and the temperature is at the melting point (f = 0).354a) JoSo e"^ /4D d$ f = 0.PARTIAL DIFFERENTIAL EQUATIONS II 531 f(s») = o' ffl!.353) P e^/4Dd^ f=_l+l2 . The solution of Equation (6. we have introduced three methods of obtaining similarity variables. at x = 0. 0<"n<s 0 (6. (6. the temperature is one degree below the melting point and beyond the interface. . Crank (1984) has discussed Stefan problems in detail. To complete the problem. In this section.252a.354b) L!oesJ/4D[S°e^/4D^= j 2k Jo (6>3.Imposing Equation (6.35la.348) is f = CJ + CQ f e^ / 4 D d^ JO The constants C o and Cj are determined from Equations (6.1 . we need to define the domain which we assume to be semiinfinite (0 < x < oo). Further details can be found in Ames (1972) and in Bluman and Cole (1974). The additional boundary conditions we impose are f(0) = ./1^ lim fCn) = 0 T) >oo ( 6 35iab) where sQ is in the solid phase and s 0 is in the liquid phase.352a.
it is stated that G has to satisfy the following conditions (a) V G = 0. The conditions which G has to satisfy in the case of ordinary differential equations are stated in Chapter 1. Note that some authors include the factors (1/471. that the solution of Poisson's equation is known if the Green's function for that problem is available.4la) (6. z ) . We now have to compute G. y). In particular. u = f (x. if V 2 u = p (x. 4).4lb) where S is the surface enclosing V. 1/2TC) in the definition of G. y).45a) . it is shown. In Chapter 4.43a) (6.532 ADVANCED MATHEMATICS 6. y. the Dirichlet problem defined by V 2 u = p (x. as an application of Green's theorem.42. These conditions can be extended to partial differential equations. in a plane.43b) where C is the curve enclosing the domain A. in V on S (6. U(P) = 2K \\ G P d A + I f  f A C ds (64"4) G is the twodimensional Green's function. u = f(x. These factors are then absent in Equations (6. This method is widely used to solve elliptic partial differential equations.4 GREEN'S FUNCTIONS Dirichlet Problems Green's functions were used to solve boundary value problems in Chapter 1. Similarly. z) . In Chapter 4. in A on C (6. everywhere in V except at P (6. y. The value of u at a point P is given by u < P )= ^[///GpdV+//flM _V S (642> where G is the Green's function for the problem.
45b) (6. we have for any differentiable function f (x.48b) If P is the point (x 0 .y 0 ) dx dy —oo —oo (6. y) 8(y . 8(xx0.46) Using the definition of the delta function [Equation (1.y0) G = 0. y .1718d)]. on C near P (6.45a) in terms of the Dirac delta function.48a) i l l f(x.x 0 ) 8(y . in the threedimensional case. We can rewrite Equation (6. y . z 0 ).x 0 .47b.49a) (6. y. z) 8 ( x . A simple physical interpretation of G is that it is the potential due to placing a unit charge at P and earthing the surface S. y . yy0.x 0 ) 8(y .y0) is written as a product of two delta functions in the following manner 8 (x .45a) can be written as (a') V 2 G = 4JC 8(x .x 0 .z 0 ) dx dy dz = f(x 0 . zz0) = 8(xx0) 8(yy0) 8(zz0) oo (6. G = in r. y) 5(x . z 0 ) — oo (6.49b) (6.45c) where r is the radial distance from P.y 0 ) dy = f (x 0 .PARTIAL DIFFERENTIAL EQUATIONS II 5ii (b) (c) G = 0. y 0 ) (6.c) Similarly. G « 1/r.y0) (6.x 0 ) 8 ( y .x 0 ) 5 (y .47a) = ( /oo f(x 0 . The twodimensional delta function 8 (x .x 0 ) 5 (y . Equation (6. y) 8(x .y0) = 8 (x . y 0 . y) oo oo \\ f (x. on S near P (6.x0.y 0 ) dx dy = \\ f (x.z 0 ) (6. y0.49c) .y 0 ) 8 ( z .y0) 8(z .45a') In the twodimensional case. G satisfies the following conditions (a) (b) (c) V 2 G = 2K 8(X .
x o ) 2 + ( y . G is symmetric. y 0 . G behaves as the fundamental solution of Laplace's equation [Equations (5. except for simple domains such as a half space or a circle. The function G is the potential of a unit charge placed at (x0. The Green's function G (x. y. Upper half plane Consider the upper half plane z > 0.dS = An se I y^. x.41 lb) (6. The potential at (x. x 0 . To maintain the plane z = 0 at zero potential. z) due to a unit charge at (x 0 .y o ) 2 + ( z + z 0 ) 2 In the twodimensional case (y > 0). ZQ) and the plane z = 0 is kept at zero potential. y0. y 0 . y0. It is generally not easy to calculate G. Since the Laplacian is selfadjoint. that is G (x.410) This can be deduced from physical considerations.ds = 27t (6.41 lc) .45a' or 9a). z) (6. Each is of radius e with e —> 0. .z 0 ) . y. y0. z 0 . zQ) is given by G = i1 r r r = V(xxo)2+(yyo)2+(zzo)2 r' = V ( x .45c') (6.49c') where S e and C e are respectively the spherical surface and circumference that enclose the point P. y. y0. z 0 ) = G (x 0 . We next deduce G for simple geometries by the method of images.534 ADVANCED MATHEMATICS where r is the radial distance from P. z). 39)] and G satisfies Equations (6. z. Condition (c) for the three and twodimensional cases can respectively be replaced by I I ^p. x0. y. we place a unit charge of opposite sign at (x0. z. y.513b. y 0 .41 la) (6. G is given by (6. Note that near P. z 0 ) due to a unit charge at (x. zQ) is the same as the potential at (xQ.
and P' can be written respectively as OQ = r[sin9 cos()). P. On introducing spherical polar coordinates (r.41 If Q is any point. The inverse point P' is defined such that OP* OP' = a2 (6. the vector positions of Q. as shown in Figure 6.412) Consider a sphere of radius a with a unit negative charge placed at P. 0. ())).41. cos0] (5.415a) .414) r2 where rj and r 2 are the distances from Q to P' and P respectively. The point P is at a distance b from O. sinG sin(J). the centre of the sphere. G is given by G = ± Image system for a sphere and a circle + v^brj (6.PARTIAL DIFFERENTIAL EQUATIONS II 5£5_ G = ^in[(xx o ) 2 +(yy o ) 2 ]iin[(xx o ) 2 +(y + yo)2] Sphere and circle (6.413) o p 1 P1 FIGURE 6. The image system that produces a zero potential on the surface of the sphere is a charge a/b placed at the inverse point P' of P.
2 r b c o s ( 9 . Determine the Green's function for the Dirichlet problem in the unit circle. 9).415d.49a to c).e) (6.415b) (6.h) (6.O P ') = r 2 + a 4 / b 2 .415C) O P ' = ^[sin6 0 cos())0. sin9 o sin())0.2 ( a 2 r / b ) c o s ( 0 . cos0oJ 2 (6. Example 6. The method of images is simple and the image system for other simple boundaries can be found in Ferraro (1956).41. the centre of the circle. We can expand G in terms of eigenfunctions. r^ and r2 are given by r 2 = r2 + a 4 / b 2 .416) . Again. The appropriate G at any point Q which satisfies G = 0 on the circle is G = ina^/Cb^) where r^ and r2 are the distances from Q to P 1 and P respectively. cos6 0 j It follows that r2 = ( O Q . the point P is at a distance b from O.0 o ) r  = r2 + b 2 .417b) (6. the inverse of P.415i) r 2 = (OQOP)« (OQOP) = r 2 + b 2 .O P ' ) » ( O Q . We need to find a function G that satisfies Equations (6. There are other techniques of obtaining G.2 r b c o s G cos© = sin0 sin0 o cos (()()0)+ cos9 cos0 o In the twodimensional case.0 o ) where 0 and 0O are the polar angles of OQ and OP (OP1) respectively.415g.2 (a2r/b) cosG (6. sin0 o sin<)0.413) defines the point P ' . Equation (6. We write G as a sum of two functions G = Gj + G 2 (6.536 ADVANCED MATHEMATICS OP = b[sin0 o cos()0. In polar coordinates (r. Among them are splitting G into the sum of two functions Gj and G2. The image system is a charge of opposite sign placed at P'.415f) (6. We illustrate these methods by considering the following examples.417a) (6. we suppose the circle to be of radius a with a unit charge placed at P.418) (6.2 O Q » 6 P ' = r 2 + a 4 /b 2 .
G ! (6.PARTIAL DIFFERENTIAL EQUATIONS II 537 Let Gj be the fundamental solution.1 i n [(1 .422c) = 1 ^ ^ Lm=l (e im(0e o ) + gimCBeo)) (6. the unit circle. we denote (xQ. on C. i n [1 + b 2 . In Chapter 5.9 0 )] m=i 2 (6.419a. y0) by (b cos0 o . b sin9 0 ) and rj is given by rj = V (r cosG . 21a.42 lb) Eo + X tAm c °sm0 + B m sinm0] = .[ i n ( l . 00 G2 = E o + X [Am cosmG + B m sinm0] r m m=l (5. for convenience.x o ) 2 + ( y .514) Using the polar coordinate system.419b) (6.y o ) 2 (6. We choose G2 such that the boundary condition is satisfied. using the method of separation of variables.422d) ~ .b e " i ( e " 0 o ) ) ] (6.422b) = .i .419a) The function Gj does not satisfy the boundary condition.b cose o ) 2 + (r sin0 .2b cos(0 . it is reproduced here. If P is the point (x0.421a) (6.be1 (0eo)) (l _ be"1 (e0o))] (6.b sin0 o ) 2 = Vr2 + b22rbcos(00o) Combining Equations (6. y 0 ) . Gj is given by Gl = i n r : where ri = V ( x . b) yields 00 (6.m = X ^cosm(00 o ) m=l (6.514) and. It satisfies Equations (6. G 2 has to satisfy Laplace's equation and should not have a singularity inside the unit circle. 20.420) In addition.b e i ( 9 " e o ) ) + i n ( l .±.422a) = .49a.422e) . c). we have found that G 2 is given by Equation (5. That is to say. we have G2 = .
44) enables us to write the value of u at a point P as follows r\ U (P) tin .424a) = ^ m=l ^—^— [cosmG0 cosmG + sinmG0 sinmG] (6.422f) Comparing the coefficients of cos mG and sin mG. 19a.423a to c) yields G 2 = X IJf [ c o s m 9 o m=l c o s m e + sinm6o sinm9l rm (6.422c) has to be modified.427) Jo Jo Jo dr From Equation (6.514) is no longer appropriate and the expansion in Equation (6.425)] is identical to the one obtained by the method of images [Equations (6. In this case.416.c) Combining Equations (5. we obtain E o = 0.42. B m = (b m sinmG 0 )/m (6.6 0 )] From Equations (6. we obtain .538 °° hm ADVANCED MATHEMATICS = 2^ — (cosmG cosm0o + sinmG sinm0 o ) m=l (6.425).b.424c) G = 1 i .i n [1 + b 2 r 2 .2br cos(G . [r 2 + b 2 2rbcos ( 9e 0 )l Li + b 2 r 2 .426b) Equation (6.423a.i. 21b. 6.6 0 ) (6 4_25) We have obtained G inside the unit circle. 17a. Example 6.514. Solve Poisson's equation 2 V u = p u = f inside the unit circle on the unit circle (6. and we note that G obtained by this method [Equation (6. = L271 pGrdGdr+ f~ dG (6. Equation (5.2 b r c o s ( G . We can equally derive G for the region outside the unit circle. b)].2% .418.424b) = .426a) (6. Am = (bmcosm60)/m. we deduce that G is given by (6. 24c).
y) = V m n (a.d. 0 < y < b for the Dirichlet problem.43lb.429) Jo l2bcos(0e o ) + b 2 Equation (6. b sin9 0 ).429) is Poisson's integral which is the solution of Laplace's equation satisfying Equation (6. . 27. 28) yields the value of u at (bcos0 o .4.PARTIAL DIFFERENTIAL EQUATIONS II 539 *° 3r r=1 = ^ l . y0) as oo oo G (x. 0) = V m n (x.3 °) where V m n are the eigenfunctions defined by V2Vmn = ^mnVmn V m n (0.e) V m n and the V^CsinfEflsinjSM) (6.428) The combination of Equations (6. b) = 0 By the method of separation of variables or by inspection.L 271 f(lb ) d0 (6.425. m and n are integers.c.4 .32a) (6. y 0 ) = £ m=l n=l cmn(*0' YQ) V mn( x .y) ( 6 .426b).432b) where C is a constant.6 0 ) + b2 (6. If p is zero. y. the eigenfunctions eigenvalues A. u reduces to u(bcose o .2 b c o s ( 9 . Example 6. y. We expand G (x.bsin6 0 ) = . y) = V m n (x. Obtain the Green's function G inside the rectangle 0 < x < a. x 0 .mn are given by (6.43. x 0 .431a) (6.
we obtain V2G = 27C 5 (x .n The Fourier coefficients c m n are given by a b (6. by simply changing the boundary condition G = 0 to 3G/9n = 0. From Equation (4.540 ADVANCED MATHEMATICS We normalize the eigenfunctions. that is to say. We examine some of the causes.y 0 ) = 2= £ c m n Xmn sin ( ™ p ) sin i^\ Vab m.433a. 30. it is the normal derivative 3u/9n which is given on the surface.438) . 31a.x 0 ) § (y .4O I) __ ^ 7 ^ aD (m7c/ar + (nre/br Neumann Problems In this case.y0) dx dy \nnVab J J 0 0 (6. This is not as straightforward as it may seem. we obtain 8K (j = "V m. we deduce by setting cp = 1 and \\i = G I I ~ d S = I I I V2G dV S V (6.435a.c) The normalized eigenfunctions are then V^^sinpfjsinj^) Vab d (6. n sm ( m 7 t x ^ a ) sm (m7rxo/a) sin (n7ty/b) sin (nKy o /b) (O. 34).b.x0) 5 (y .422). It is tempting to extend the theory developed for Dirichlet problems to this case.434) From Equations (6. we require a b ab I I Vmn dy dx = 1 = C 0 0 I I sin 0 0 sin —^ dy dx = C ^ (6.b) cmn = — ^ — = I I sin p « ) sin fe) 5 (x .430).436b) Substituting c m n into Equation (6.436a) = 4B= sin ( ^ sin fe) (6 .49a.
everywhere in V except at P (6. At steady state. the temperature satisfies Equation (5. modifying the boundary condition is not sufficient and we need to impose additional conditions.440). By definition V G = 0 everywhere except at P. it is stated that the solution to a Neumann problem is arbitrary to the extent of a constant. We impose the following conditions on the Green's function for the Neumann problem (N) 2 (a) V N = A (constant).73.440) Equation (6. One possible physical interpretation is to consider the problem of heat conduction. the boundary condition must satisfy the left side of Equation (6.438). denote the surface of this sphere by S e and the volume by V e . n = 0) leading to the impossibility of determining c m n . In Example 5.41 to determine G. in Example 643. the function G2 will not be uniquely defined. It follows that. One possibility is to modify condition (a) [Equation (6.44lc) (6.45a or 9a)]. implies J J ^ . Enclose P by a sphere of radius e. Thus.PARTIAL DIFFERENTIAL EQUATIONS II 547 //^ dS = /// v 2 ° d V S V 2 <S438) Let P be the singular point. for a Neumann problem the eigenfunctions V mn are of the form cosnmx cosnfty and one of the eigenvalues 'kmn can be zero (m = 0. the heat flux across the boundary must be zero and this is expressed by the left side of Equation (6.441b) (6.438) where G is the temperature. if we use the method adopted in Example 6. from Equation (6. Also.438). V G is zero and. we obtain (6.45c') contradicts Equation (6.nearP . on S on Nl/r.441d) = A + 4TT8(X£0) (b) (c) ^ = 0 . In the volume V .444) which is Laplace's equation. To maintain a steady temperature.V e .dS = 0 (6. Note that if G satisfies Laplace's equation everywhere in V.44la) (6.439) 2 s se Imposing the condition 3G/8n = 0 on S.
near P (6. c).x o ) .542 ADVANCED 0 MATHEMATICS In Equation (6.443a) (6.444a) (6.b) v The twodimensional Neumann function satisfies the following conditions (a) (b) (c) V N = A1+27i8(xr0) ^ .443b) (6. on 2 in V (or S in a plane) on S (or C in a plane) (6. we deduce that I I I A dV = A V = 4K is the vector position of P. Poisson's equation along with the Neumann condition are written as V u = p (x. b) into Equation (4.445b).442a.445a) s v = A i J j u d V + 4?cu(P).= 0.N p ] dV (6. 44a. ^— = h (x. z) .onC on N .441 b).443c) where Aj is a constant.444b) Substituting Equations (6.445b) From Equation (6.441b.V S (6. 41a. y. (6.422) and setting \/ = N and cp = u yields  Nh dS = I I I [Au + 4 7 t u 8 ( x . y.i n r .45c\ 38. we obtain u(P) = _L 11 I Np d V . x is the vector position of any point and L From Equations (6.446) . c.[ ( I Np dV v v (6. z) .I  Nh dS + constant .
PARTIAL DIFFERENTIAL EQUATIONS II 543. In order for the fluid not to cross the line y = 0. in hydrodynamics.x 0 ) 2 + (y .447) Neumann problems are usually associated with hydrodynamics where the usual boundary condition is that the normal velocity (3u /dn.x 0 ) 2 + (y + y 0 ) 2 ] On differentiating partially with respect to y. Example 6.446) in twodimensions is u(P) = L I I Np dS . u is the potential) at the boundary is zero.I Nh ds _S C (6.44. where the constant term is A Jjj V u d V. . This is the difference between hydrodynamics and electrostatics. This term can be ignored since the solution to a Neumann problem is arbitrary to the extent of a constant. 1967) N = 1 i n [(x .4_49) =0 y=0 (6. If N is the potential.446) and can be calculated once N is known. Further discussions on the image system in hydrodynamics can be found in MilneThomson (1967). We illustrate this method in the next example.x 0 ) 2 + (y . we obtain (6. The analog of Equation (6. b) is given by Equation (6.444a. Find the Neumann function N for the plane y > 0.y 0 ) 2 + (y±zo) (x . the image of a source is another source and not a sink. Neumann's functions can also be constructed by the method of eigenfunction expansion as explained in the next example. A source is placed at (x 0 .448b) ^L = dy It follows that ^ dy <±zi& (x .x 0 ) 2 + (y + y 0 ) 2 ] = 1 i n [(x .448a) (6. N is given by (MilneThomson. The solution of Equations (6.x 0 ) 2 + (y + y 0 ) 2 (6. We can determine N by the method of images. y 0 ).x 0 ) 2 + (y .y 0 ) 2 ] + 1 i n [(x . we need to place another source of equal strength at (x 0 .450) • Note that.y 0 ).y 0 ) 2 ] [(x .
. n*0 (6. n = 0. Find Neumann's function N inside the rectangle 0 < x < a .451a.44. the eigenfunctions w mn have to satisfy ^ L y=b = ^ X =a =0 (6. We use the same method to determine N.2 — cos mJIX cos ^ Vab wmft = \[^r cos ^ 2 ^ . 1. . n = 0 m = 0. . m*0. . In this example.b) The appropriate eigenfunctions are w m n = C cos " ^ 11111 a cos ^ o . The normalized eigenfunctions are w m n = . 1.45. .446 or 47). In Example 643. we obtain CXJ .544 ADVANCED MATHEMATICS Example 6. 0 < y < b.452) Note that.c o s 1 ^ . Once N is known. Green's function for the Dirichlet problem was derived.454) a b n =i N is defined to the extent of an arbitrary constant.— • / ab m =i (m7t/a)2 + (n7t/b) 2 cos(m7ix/a)cos(mKx 0 /a) (mjc/a)2 ^ cos (nrcy/b) cos (n7iyo/b) (nrt/b) 2 n=l 4jt ab ^ m =i 47C (6. the solution of Poisson's (or Laplace's) equation is found by substituting N into Equation (6. m and n can be zero. n * 0 (6.453a) m * 0. unlike in Example 643. in this case. (6.453b) (6..453c) g^ ^ i cos (mjcx/a) cos (m7txo/a) cos (n7ty/b) cos (n7cyo/b) IN —.. UH V ab b Proceeding as in Example 6. m = 0.. V ab a w nn = \[\.
d) by the method of Green's functions.455) dy2 (6.i n [(x + x 0 ) 2 + (y .42.b) p.= 0 .y 0 ) 2 ] + ±. The extension is straightforward and is illustrated in the next example. x>0 y>0 (6.1 i n [(x . Tp. y. .458a) + x0)2 + ( y .b) (6. (u is given in parts of the boundary and 3u/3n in the remaining parts of the boundary). dx2 subject to u = f (x) .y 0 ) 2 ] .456c.x 0 ) 2 + (y + y 0 ) 2 ] . Let P be the point (x 0 .y 0 ) . o n x = 0.456a.1 i n [(x + x 0 ) 2 + (y + y 0 ) 2 ] . y0) as shown in Figure 6. on y = 0 y >o (6.= h (y) . we combine the techniques explained previously. The function G (x.x 0 ) 2 + (y + y 0 ) 2 ] = 1 in t(xx0)2 + (yy0)2][(x 2 (6.457a. y 0 ) and S ( x 0 . y 0 ).x 0 ) 2 + (y . we have a Neumann problem and the image system is another source at Q (x 0 . We first determine the Green's function G for this mixed boundary value problem by the method of images.PARTIAL DIFFERENTIAL EQUATIONS II 545 Mixed Problems (Robin's Problems) For mixed problems. y0) is given by G = 1 i n [(x . x 0 . Example 6. The function G has to satisfy G = 0. The boundary y = 0 defines a Dirichlet problem and the image system are charges of opposite sign at R ( .46. on x = 0 (6.d) Relative to x = 0.x 0 . + iLn = o. Solve the equation iLn.y 0 ) 2 ] (6458b) [(x + x 0 ) 2 + (y + y 0 ) 2 ] [(x . OX o n y = 0.457c. x > o.
61) yields (6. ADVANCED MATHEMATICS Q P y0 —i 1 ^ ~XQ XQ X R y° s FIGURE 6. That is to say.460) Joo Note that the contour integral is taken in the counter clockwise sense.461) . so that the integral along the yaxis is from °° to 0. On the xaxis f = ~ an ay Combining Equations (6.42 Image system of a mixed boundary value problem From Equations (6.457a to d).546 .460.459) where C is the contour enclosing the positive quadrant. we deduce that u (P) is given by u (P) = ^ [ (f H . along the positive x and y axes. From Equations (6.459) can be written as u(P) = J 2K f^dxl JO dn Ghdy (6.Gh) ds c (6. we deduce that Equation (6.47).44.
Example 6.463a.462). 271 2 y \/y0 [(x + x 0 ) 2 + ( x . The method of transforms discussed in Chapter 5 can also be used to determine G (see Problems 12b.47.PARTIAL DIFFERENTIAL EQUATIONS II 54Z u(P) = i [f H Jo y d x + f G x=o hdy y=° •'o (6.x 0 ) 2 + 2 y 0 2 ] rax [(x + x o ) 2 +y o 2 ][(xx o ) 2 +y 2 ] \ + Jo f in Jo [x o 2+ (y. we can use conformal mapping to find G. 13b). given that G is known for a unit circle.x 0 ) 2 + 2 y 0 2 ] [(x + x o ) 2 + y o 2 ] [ ( x . If G. is known in a certain domain R' and if R' can be mapped into R.x o ) 2 + y 2 ] [ x (6.x0. for the Laplacian operator.463a) G =in o+(yyo)2] [x 2 + (y + y 0 ) 2 ] (6.462) From Equation (6. 4 . we obtain p oo u(x v ) . We illustrate this method by an example. Conformal Mapping In twodimensional cases.2 y 0 [ ( x + x0)2 + ( x . we have obtained Green's (or Neumann's) functions by the method of images. b) into Equation (6. G is known in R.l u(. by expansion in eigenfunctions. Determine G for the Dirichlet problem in the case y > 0. 64) [x02 + (y + y 0 ) 2 ] • So far. y0) which we denote by zQ. Let P be the point (x 0 . In Example 3. and by separation of variables.83.y o) 2 ] h d y (6 . we have shown that the mapping .463b) Substituting Equations (6. y 0 .458b). we obtain 3°_ 3y = y=0 x O .
412). Thus. The Green's function for a unit circle with the singular point at the origin (b = 0) is given by Equation (6...( a n v ) + — (a12v) + . .5 GREEN'S FUNCTIONS FOR GENERAL LINEAR OPERATORS (64.( a 2 2 v ) . as expected..^ .x o + iy o (6.67b) = in i**o)2 + (yyo)2 ( x .465) maps the upper half plane into a unit circle.— ( b 1 v ) .467c) is Equation (6.4.x o ) 2 + (y + y o ) 2 Equation (6.466) * + y*oiyO x + i y . The point zQ is mapped to the centre of the unit circle and the complex conjugate zQ is mapped to the region outside the circle.51) 9xj 3xj3x 2 9x 2 dx^ 3x2 We can verify that 3 u allv — 3xj o . 66) yields G = i n ^ zz0 = Zn (6.^ . J —J ( a ll v ) = r— a n v — .(a n v) 3xj 3xj [ dx2 dxj . 3u L(u) = a n — r + a 12 + a22—2+bi 3xj 3xj 3xj3x2 3x 2 The adjoint operator L is defined by T 0 0 0 . . .31)..67c) The method involving Green's functions can be extended to the wave and the diffusion equations as well. _ _. we reproduce it here.548 ADVANCED MATHEMATICS w = 71ZJ^. 6.425) and can be written as G = i n r = inlwl Combining Equations (6. y0) is mapped to the origin. X2) which is given by Equation (5.31) 3x 2 L*(v) = . We now discuss the case of the general second order linear differential operator in two variables (xj. x 3 u 3 u 3 u . (5.52) u .u — . + b2 3u + cu £( = f(xl'x2) . (6. 3 \ 3u 3 . the singular point (x0.465. 0 0 0 . zz0 (6.( b 2 v ) + cv (6. For convenience.^ ._ .
x 10 ) 8(x2 . rj and r 2 simplify to _* / 3u a12 du \ //ICON f i = G [a"d^+f BT2) (6"58a) . b) become tt * A I (6.53a) is div x and the two components of £ (rj.423). we deduce that v L (u) . Equations (6. If u satisfies homogeneous boundary condition (u = 0 on C).53a. Equation (6. c).PARTIAL DIFFERENTIAL EQUATIONS II 549 Similarly.+ ^dXj dx 2 + (6. 4.53b) f2 = ~ 2 3x7 ( 12V) + ^ " ^ 7 + 822V ^ "U ^ ^ ^ + 2UV ( } Note that the right side of Equation (6.53b. Replacing v by G* and using Equations (5.53b. x 2 0 ) = I (rj dx 2 .u 5 _ ( a 1 1 v ) b1uv (6.+ ^ . r 2 ) are given by Equations (6.u L * ( v ) = divr The operator L is selfadjoint iff L = L* From Equations (6.. 6.53a) ^v^^(a12v) + ri = a 1 1 v ^ .r 2 dx x ) (6.b) A A C The adjoint Green's function is defined as L* (G*) = 2K 8(XJ .53a') I I [v L (u) .56) The function G also satisfies certain homogeneous boundary conditions to be defined in conjunction with the boundary conditions associated with Equation (5.54) (6.u L*(v)] dx{ dx 2 = I I U±. J dxj dx 2 = I (rx dx 2 .53a) can be written as v L ( u ) . we obtain (6. c) with G replacing v and are to be evaluated on the boundary C.r 2 dxj) The functions rj and r 2 are given by Equations (6.31).56).57) C I I G f dx^ dx 2 27r u (XJQ.x 20 ) (6.55a.55a.u L* (v) = 1^.31.
That is to say. we deduce that the integral in Equation (6. dx 2 A (6.x 1 ) 5 ( x 2 .G ( x 1 0 . we have .t[G*(x ] . The function G (Xj.57) to be zero. X2Q .512) Multiply Equation (6. they are homogeneous. We now show that G (xj. x 2 . b) and the homogeneous boundary conditions imposed on G and G . then G* (or 3G /3n) is zero on C.X 1 0 ) A 8(x2 . x 2 ) satisfies the equation L(G) = 2 7 u 5 ( x 1 .513) From Equation (6.550 ADVANCED MATHEMATICS (6.513) is zero.58b) We impose the condition that G vanishes on the boundary.511).X 2 ) . we obtain Equation (6.I (r 1 dx 2 r 2 dx 1 ) _A C (6. X 2 . X 2 Q) = G(XJQ.X J ) 8 ( X 2 . x 2 0 . x 20 ) f dx.G ( X 1 . Xj. x j . XJQ. x 2 0 ) = ^ L I I G * ( X l .55a. x 2 . if u (or3u/3n) is given on C. subtract the products. x 2 . x 2 . It follows that G (x^. XJQ. resulting in the line integral in Equation (6. X2Q) = G (XJQ. x 2 )] (6.59) If the boundary conditions on u are nonhomogeneous. X J .56) by G. X 1 0 . to obtain 2K I I [G*(X 1 ? X 2 . x 1 0 . x 2 ) (6.512) by G* and Equation (6. In the selfadjoint case.x 20 )] dxj dx 2 = 2. x 2 ). u (x10. The value of u at (x 10 . x 20 ) is then given by u (x 10 . for G*. x 2 0 ) . x 20 ) = ^ I I G*fdx!dx 2 .x 2 ) (6. ^ ) by (x 1 .514) Replacing ( X j . X 2 ) 8(XJ . x 2 ) (6. Xj. x 2 . X 2 0 ) 8(XJ . x 10 .510) The boundary conditions to be imposed on G are of the same type as those imposed on u. However.511) where G is the original Green's function. Xj. X2Q . and integrate the difference over A.
x 20 ) = G(x 1 0 . At about the same time. It was the mathematician Jordan who saw the connection between the theory proposed by Heisenberg and the theory of matrices. x 2 . x 2 0 . x 1 0 . It has led to a number of applications. 10) can be replaced by G via Equation (6.59. His theory is based on the possibility that matter can have particle and wave properties. electron microscopes. and elliptic equations but the method involving Riemann's functions is restricted to hyperbolic equations. To formulate the laws of quantum mechanics on an acceptable basis.515) The function G in Equations (6. have to be applied. from an apparently different starting point and using the classical continuous formalism. both G and G* are equal. New devices based on the principles of quantum mechanics are still being developed. such as lasers. G is symmetric in (xi. de Broglie suggested that . The method of Green's functions is applicable to hyperbolic. The importance of quantum mechanics needs no stressing. Newton's laws have to be replaced by the laws of quantum mechanics. They fail to describe the motion of bodies traveling at high speeds. 6. silicon chips. at speeds approaching the velocity of light. Schrodinger.511) and.PARTIAL DIFFERENTIAL EQUATIONS II 557 G ( x 1 . For micro particles. in selfadjoint cases.6 QUANTUM MECHANICS Limitations of Newtonian Mechanics Newton's law of motion are adequate to describe the motion of bodies on a macroscopic scale usually encountered in everyday life. Note that the heat (diffusion) equation is not selfadjoint whereas the Laplace and wave equations are selfadjoint. parabolic.x 2 ) and (xio>x2o) (6. he was not aware of the existence of the matrices. provided an alternative foundation for quantum mechanics. This led to the publication of the famous BornHeisenbergJordan (1926) paper. the laws of relativity. such as atoms and their constituents. Heisenberg proposed the theory of matrix mechanics. x 2 ) that is to say. x 1 . Greenberg (1971). that is to say. In 1924. and Zauderer (1983). Only hyperbolic equations have two real characteristics. though at that time. and nonlinear optics. two approaches were adopted in the 1920's. It provides explanations for a variety of physical phenomena ranging from the structure of the atoms to the beginning of the universe. Courant and Hilbert (1966). Further details on Green's functions can be found in Morse and Feshbach (1953). as developed by Einstein at the turn of the twentieth century. In this case.
63) Schrodinger Equation It was stated earlier that Schrodinger based his theory on the possibility of matter having both particle and wave properties. < & is the potential energy. the wave properties have to be considered for microscopic particles only. The function \j/ is normalized such that .b) (ii) Heisenberg's uncertainty principle It is not possible to simultaneously measure the momentum p and the position x to arbitrary precision. t )  the probability density of the particle being at x at time t.61) (6. has to be small and the wave properties can be neglected. The wave is function \/ is complex and the observable quantity is not \/ but \/ .T T ~ V2V/ + Oi)/ = i h . (i) Electromagnetic energy occurs in discrete quantities. t) obeys a diffusion equation which can be written as . He proposed that the wave function V/(x. for macroscopic particles (p is large). the matrix and the wave mechanics were thought to be unrelated until Schrodinger showed the equivalence of the two formalisms.^ 2m at 2 (6. Thus.62a. The relationship between the momentum p of the particle and the wavelength A. we can deduce that. h = h/2n. For a short period. ApAx = h (6.64) where m is the mass of the particle. The quantity  \  / ( x .552 ADVANCED MATHEMATICS atomic particles might behave as waves. the wavelength A. If Ap and Ax are the errors in the measurements of p and x respectively. The energy E is related to the frequency v by E = h v = h co O = 2TZV. From Equation (6. We state some of the basic assumptions of quantum mechanics. G (6. of the "particle wave" is p = h/X where h is Planck's constant.61). where h is the Planck's constant. and V is the Laplacian.
2 2 U+ ] u ["25T V ° ( 2 ° U' = ^ dt = E (6. We write \/(x. the solution is of the form given by (6. That is to say.66) \t.t)2dx = 1 V (6.— V u + Ou = Eu 2m f = C e[Ei/n where C is a constant. The wave function \/(x.64) and assuming a static potential O (x). Equally. does not have a visible orbit as a particle but its presence can be detected when it hits a target and acts as a particle. Its position is given by a probability density and it is this probability density which is associated with a wave.68a) is the timeindependent Schrodinger equation.b) (6.t) depends on the potential <P and if O is time independent. Equation (6.69) (6.67a. Equation (6. The property which is manifested depends on the situation under consideration.68b) . which is an electromagnetic wave.68a) (6. b) lead to h2 2 . a photon. we have (6. Note that it is not claimed that the particle has become a wave but rather that its position is not known with certainty. Equations (6.610a. \}/ can be considered to be a wave of probabilities.b) where E is a constant representing the total energy (kinetic + potential).t) = u ( x ) f ( t ) Substituting \/ into Equation (6.67a.65) where V is the whole space. t) can be written as V/ = u (x) eiEt/n The probability density P r is given by P r = v/ 2 =  u ( x )  2 and is independent of time.PARTIAL DIFFERENTIAL EQUATIONS II 553  v/(x. Thus. we have a coexistence of wave and particle properties. These waves are abstract waves in the same sense that crime waves are abstract waves. The solution \/(x.64) can be solved by the method of separation of variables.
611) _  l ( £ i + 2^]_eiu =Eu 2m [dr 2 r dry r To simplify Equation (6. Obtain the wave function in the case of an electron moving radially in a potential < E > given by C > = e2/r Angular variables can be neglected. Since the potential is time independent.617) . ij/ is given by Equation (6.612) becomes dV2du+^_lu dp 2 p dp \p 4/ = 0 ( 6 6 H ) We require that u —> 0 as p —> °° and this suggests that we look for a solution of the form u = e~ p p v(p) where p1 is a positive constant. 5.614.554 ADVANCED MATHEMATICS Equation (6.616) The number of terms in Equation (6. X = 2me2/ah2 (6.b.68a.69) and stationary states exist. we deduce that u satisfies (6.612).69). Combining Equations (6.616) will be reduced. The function u depends on r only. we consider the case of the hydrogen atom. we introduce the following variables p = ccr.616) now becomes (6.615) d!x + 2 (I_p)dX + ( p 2 _ l + ^ _ ? l ) v dp 2 lp I dp \ 4 p p/ = O (6.612) (6. Example 6. In the next two examples. 11. if we choose (3=1/2 Equation (6. 15) yields (6.c) Equation (6. a 2 = 8mE//i2.532).61. From Equations (6.613a.
On differentiating term by term and substituting the result into Equation (6.1)] p^" 1 " 1 = 0 m=0 m=0 (6. = 0 and the solution is given by v = constant = v0 (6.2 . If k is zero (the ground state). If k is an integer (k = 0. The valid solution is the polynomial solution.). . we expand v in an inverse power series of p and write oo v = X m=0 cmPk~m (6619) where k is a constant.621) From Equation (6. we obtain c s + 1 = . Cj = C2 = . we note that the series solution does not converge. since we are interested in the region p —> <».s .s ) ( k . the infinite series becomes a polynomial.622) (6.618).623b) (6.623b).X cm [(km) (A.m . 1.619) in the form of a polynomial of degree k.m .l)] Substituting Equation (6.620) On comparing powers of p. we obtain oo oo X c m [ ( k .623b) and the solution is given by Equation (6..623a).s ) c s / ( s + l) (6.m ) ( k .l ) + 2(ks)] = cs+1[(ksl)(A. we can calculate the coefficients cs from Equation (6.s ) ( k + l . we deduce that co[k(A.624a..( k .l)] = 0 For a nontrivial solution cQ ^ 0 and this implies that X = 1 +k Equating terms of p k—s—2 leads to c s [ ( k ..m ) ] p k .PARTIAL DIFFERENTIAL EQUATIONS 11 555 dp 2 lp i dp I p / (6.. .623a) (6.618) We seek a series solution and..b) For any value of k.l ) + 2 ( k .622) into Equation (6. We recall a similar situation in Chapter 2 concerning the Legendre equation.
613b. 24b) and can be written as V/o = v 0 e . Combining Equations (6.627a.626) = \%\2 = v ? e ~ P (6.65)] and noting that V is the volume of a sphere of infinite radius.P / 2 e .b) .631a. 29) yields Pr = ^o O7C eaor (6.b) (6. we deduce that E = E k = .m e 4 / 2 / i 2 ( k + I) 2 (6.69. we obtain f°° 4TI I r 2 v 2 e~ a r dr = 1 (6.628) Jo On integrating by parts. The lowest energy level E o is given by Eo = me4/2h2 (6.i E ° t / S The probability density P r is given by Pr (6.b) Applying the normalization condition [Equation (6.625a) gives the energy levels and.629c) is obtained from Equations (6.613b. 25b). There are an infinite number of energy levels and they get closer and closer as k —> °°. we find that v0 is given by v 2 = a 3 /87t = C6O/8TC cc0 = 2 me 2 //? 2 (6.627b.625a) Equation (6.629c) Equation (6.630) The probability P^ of finding an electron in the spherical shell between r and r + Ar is given by P A = 4 7 t r 2 P r A r = r 2 a3Q e " a ° r Ar 12 The maximum value of P A is given by (6.556 ADVANCED MATHEMATICS From Equations (6. in quantum mechanics. 15. the energy levels are discrete and not continuous as in classical mechanics. c.629a.625b) The corresponding wave function \/0 can be obtained from Equations (6. 22).
PARTIAL DIFFERENTIAL EQUATIONS II 557 ^ = 0 (6632) On differentiating P^. < >  ) = G(0)H(4>) Proceeding in the usual manner. Equation (6. Solve the Schrodinger equation for the potential given in Example 6. The function F (0. . (>).633) On differentiating u.68a) can be written as _ 1 3 / 2 3u\ 1 3 / . we find that the maximum value of P^ is at r = 2/a0 The most probable radial distance of the electron from the nucleus is 2 / (XQ .638a) . In spherical polar coordinates (r.636b) (6. and separating the functions of r.61 without assuming radial symmetry. „ 3u\ 1 3u 2m /T. 0. .634) (6.634).62. r — + smG— + + (EO)u = 0 r23r^ 3 r ' r 2 sine 30 I 3 8 / r 2 sin 2 0 3<>2 h2 We seek a solution of the form u = R(r)F(0. Example 6. 0. (j)) (6. we obtain 1 A (r2 dR) + 2m r 2 dr V (E _ O) R = i(i + 1)R (6 6 3 6 a ) dr. _ . _.n 2 ] G = 0 d0 \ d0/ (6. h2 r2 _ ! _ A. 0) is assumed to be given by F(8. we find that G and H satisfy sin0 .635) . substituting the results into Equation (6. and (()..i . Sin0 ^L + _ J _ ^_Z = _ x (i + i) F sin0 30 30 sin20 3 r where Z (Jt + 1) is the separation constant.637) (6. (6.(sin0 &G] + [ i ( i + 1) sin 2 0 .
538).643) (iii) If n > i .t i l l d .42) (ii) P7n = ( l) n H=Jl21p" * (i+n)! is a multiple of PJ? and we need to consider only positive n. For H to be single valued. Equation (6. If n is nonzero.x2)] G = 0 dx L dxJ (6. n has to be an integer.711)].638b) where n is another separation constant.640) is the associated Legendre equation [Equation (2.2 AJ^_[ix2_l)*] 2* (i!) We note the following properties of PJ? (i) n and Jt are integers. The function F can now be written as .640) If n is zero (axial symmetry).638b) is H = Aein* + Bein* where A and B are arbitrary constants.71)] and the solution is given by Equation (5.639) f. [Note that the term inside the square bracket is of order 2/2 and its (/6 + n) derivative is zero if n > i ] . dxi +n (66.640) is the standard Legendre equation [Equation (2. The solution of Equation (6.733)] and the solution can be written as G(x) = CP{(x) where C is a constant.[d .n 2 /(l .] + [i (i + 1) .x2) iS. This shows that P» (6. Z is positive [see Equation (2.x 2 ) " . The associated Legendre functions PJ? (x) are defined by (6.558 2 ADVANCED MATHEMATICS ^ + n dC 2 H = 0 (6.641) Pj?(x) = .638a) is transformed to the standard form by setting x = cos 0 and it becomes (6. Equation (6. Equation (6.642) that PJ? is zero. we deduce from Equation (6.
649a) (6. The solution \/ is obtained by combining Equations (6.646) into Equation (6. ar) [A ein(f + B e" 1 "*] P^ (cosG) (6. 46.648) (6. The splitting of e(ar/2) (6.PARTIAL DIFFERENTIAL EQUATIONS II 559 F = (A e i n * + B e" i n *) P^ (cos9) Combining Equations (6. p) The properties of confluent hypergeometric functions are given in Slater (1960).646) d!vt2_.644) d!+2dR+^_l)R dpz p dp Ip 4/ = AM±l)R pz (6. c.647) can be reduced to a standard form by writing v = p /)! w(p) Substituting v into Equation (6.645) Equation (6.69.649a) is the confluent hypergeometric differential equation and its solution can be written as w = J F J (a. 44.47) Equation (6. 35.645) is similar to Equation (6. we obtain (6. 13a.525a) which are observed in spectroscopic measurements of moderate resolution. 50) and is Y/ = eiEt/h • One important result of Schrodinger's theory is the prediction of the energy levels given by Equation (6.645). 36a) yields (6.c) Equation (6.)ivt[ft^l)_iii+i)l¥ dpz \p /dp p pz =0 (66.X.651) . we obtain p ^ 4 + (cp)^aw = 0 dp 2 dp where c = 2 ( i + 1).614) and we assume that R is of the form R = e " p / 2 v (p) Substituting Equation (6. 2 i + 2.611. b. c.650) ( a r ) i 1 F 1 ( i + 1 . a = (Jt + lX) (6.647).649b.
Since O is time independent and is discontinuous. Equation (6. We next consider an example which leads to an application.560 ADVANCED MATHEMATICS the energy level observed in high resolution measurements can be accounted for by including the spin effects.655b) (6.68a) in the regions x < 0 and x > 0 separately. the solution of Equation (6.E) Ih (6. x<0 x>0 (6.652a.2 <\2 = E u ! x < 0 (6.653b) can be written as u = Ce^ 2 X + D e ^ 2 X where C and D are constants and \^ is given by %2 = V2m (O0 .63. In this example.654a) where A and B are constants and Aj is given by Xj = (V2mE )lh Since O 0 > 0.^ — ^ + %u = E\x.68a) can be written for the given O as _«_il 2m dx 2 ft2 t. 2m dx 2 The solution of Equation (6. The conditions requiring u and du / dx to be continuous at the origin yield the following .654b) The condition that u tends to zero as x tends to infinity implies that C is zero. the potential has a jump discontinuity at the origin. we need to solve Equation (6.653b) (6. Solve the onedimensional Schrodinger equation for a potential given by O(x) = f0.653a) H2 .b) where O 0 (constant) is greater than the total energy E. Example 6.655a) (6. This potential function can represent the motion of a charged particle between two electrodes kept at different voltages. \%. Interested readers can consult Eisberg and Resnick (1985).653a) is u = AeiXl" + B e " a i X x>0 (6.
This phenomenon can be observed experimentally and has found an application in the tunnel diode which is used in modern electronics.656a.b) In quantum mechanics.b) (6. we obtain A = £. there is a finite probability that the particle may penetrate the classically excluded region and this phenomenon is called penetration (tunneling).V 2 m ( O 0 .661) (6. it is not possible to state that O 0 is definitely greater than E and that the particle cannot enter the region x > 0. if the potential energy is greater that E. .2 « / i / V 2 m ( O 0 .iA.660a. For small values of x. b).1/A.656a) (6. For details.b) The probability P r decreases rapidly with increasing x.655b. we deduce from Equations (6. If the depth of penetration is Ax.O 0 . The probability P r of finding the particle in the region x > 0 is Pr =  u  2 =  D  2 e " 2 ^ x (6. The particle does not enter the region x > 0.E (6. 58a.658a.656b) (6. there is a finite probability of finding a particle in this region.PARTIAL DIFFERENTIAL EQUATIONS II 561 A+B = D i?4(AB) = . That is to say.? i 2 D On solving Equations (6.b) The constant D can be obtained by applying the normalization condition.63) and is Ap = h/Ax . Note that the statement O 0 is greater than E should be subjected to the uncertainty principle. the kinetic energy must be negative which is physically impossible.(Xl + i X2).E ) The uncertainty in the momentum Ap can be obtained from Equation (6.657a. see Eisberg and Resnick (1985). b) that Ax .E ) The total energy E is proportional to p 2 /2m and it follows that AE . This is not predicted in classical mechanics where the total energy E is constant. In quantum mechanics. 2Xl B = E(A..659a. 2 ) 2XX (6.
Show that on introducing a new variable r) = x a t . t 2 ) = f(x).v_\ o Integrate the partial differential equation and show that v is given by v = Voyn) Obtain u by Riemann's method if at time t = tj (#0). 0<x<oo. u and du/at are given by u(x. the partial differential equation 9 x ^L_^_ = o. ^ dt = g(x) t=0 Answer: v = 1 2b. = t / x . _ t>0 at 2 can be written as t.0) = f(x). = g(x) 3a. ^"'• t=t. oo<X<°o. ax 2 rj = xt Deduce that the Riemann function v satisfies the equation J ^ + A ( JL\ = A [3v + . t >0 at ax 2 . Show that the characteristics of the equation 23 u t = X 23 u . °o<X<°o. t > 0 at2 dx2 Obtain the solution u that satisfies the following conditions using Riemann's method u(x.562 ADVANCED MATHEMATICS PROBLEMS 1 a. Determine the Riemann function v for the wave equation r>2 =C v2 .
1997). v = v z = 0]. £. The viscoelastic model they chose was a modified form of the WhiteMetzner model (Carreau et al. Tyy = Xzz where y = dv x /dy. lim u (x.C t) I T(l/3)] JO exp (£ ) d£. N R e (Reynolds number) = p U 2 ~ n L n /  i . x xy = 0 . 0) = C.. N = p U 2 ~ s L s / ^ . State the conditions that C j .PARTIAL DIFFERENTIAL EQUATIONS II 563 can be transformed to ^ f drf + 3Ti2^ = 0 dri The boundary and initial conditions are u(0. y). at y = 0 at y = R(x) On introducing a stream function \j/ (x. Assuming that these conditions are satisfied. From the symmetry of the problem. find u and express it in the form of an integral. . the walls are given by y = ±R (x). C 2 . 4b. . t) = C 9 . y) defined by v x = 3\//3y . we need to consider only the region 0 < y < R (x). It is assumed that the velocity component in the ydirection (v y ) is small. t) = C. In a viscometric flow [v x = v x (y). v = d\\f/dx . The boundary conditions are vx = 0 . p is the density. the stress components x^ are given by Xxy = y '^Re' T xx~ T yy = Y'N. U and L are a characteristic velocity and length respectively. and J L X are constants.*/t"3 3 Answer: C x + [3 (C 2 . u(x. vy = g(x). and C 3 are constants. Balmer and Kauzlarich (1971) considered a twodimensional steady flow of a viscoelastic fluid between two nonparallel walls with suction or injection at the walls. and C3 have to satisfy in order that a similarity solution can exist. Referred to a Cartesian coordinate system (x. C2. where C\.
Verify that if R = kj x + k 2 .a 2 / a i 6b.= V. g = k 3 . R.= 0 y—>oo V. The following similarity transformation was proposed \/ = g ( x ) F ( r  ) .E. \\f was found to satisfy dy dxdy dx dyz [ dyz J dy3 \ [ dy2 dy2dx where f(x) = — ^—[p (x.N R e R 3 f ( x ) / g . 0)] and p is the pressure. and k 3 are constants.E. and f have to satisfy so that the P.E.D. r\ = y/R(x) Determine the conditions that g. y = a a 2 y . the equation for \(/ becomes (lk1TiNws)Hm2Nwsk1H" + NRek1k3a(H1)2 = 1 where H = F / a .T (x. where k l 5 k 2 . 4>Cn) = \(//(x a 3 / a i )] = Answer: l . n. dy Transform the P. to an O.D. The nonlinear O.E. has to be solved numerically.D. s. and obtain the associated boundary conditions. Deduce the boundary conditions in terms of F. 0) . The heat equation in a cylindrical polar coordinate system can be written as oc 3 /a 1 . N w s = N R e / N .= 0 dx dy at y = 0 .D. can be transformed to an O.E.564 ADVANCED MATHEMATICS and making appropriate assumption. a = . Use Birkhoff's method to obtain the similarity variables for the boundary layer system defined by dy dxdy dx 3y 2 By3 lim^. [Hint: write x = a a i x . y = a a3 \/. r\ = y / ( x a z / a i ) . Answer: F(0) = F'(l) = F"(0) = 0 5a. n = s= 1.D.
t) = t v (r. that the solution of the boundary value problem d2u 3x z 32u dyz on y = 0 .2 b c o s e ) 3 / 2 8a.o jo ( i + b 2 . 0) = 0 The time dependent condition at r = 0 can be transformed to a constant boundary condition by writing u (r.+ r 3r The initial and boundary conditions are u (0. y > 0 — . Use Equations (6. „. t) = t . A ^ o A ^ ( b 2 . <))).x o ) 2 + y 2 47U ^0 Joo [(xxo)2+(y + y o ) 2 9a.14) to obtain u in the form of an integral (Poisson's integral).(j))sin9d9d(l) — Answer: u (b. r—> oo u (r._ t>0 — = —. 0<r<oo. On the surface of the sphere. . u = f (x) is K 3— ( x . 7a. t) Determine the partial differential equation. Show that by introducing the variable Ti = r / V t . v(r) has to satisfy d!l + dv(i n + i)_ v dt z dr \2 IV = 0 Obtain the boundary conditions on v(r). u = f (0.42. 6 0 . Show. t) = 0 . . the boundary and initial conditions that v has to satisfy. using the Green's function given by Equation (6.D (2n (K f(9. lim u (r. °o < x < oo. )Q) = ^ 47t .= p (x. Show that the solution of the boundary value problem . The function u satisfies Laplace's equation inside and on a unit sphere. y ) .412).PARTIAL DIFFERENTIAL EQUATIONS II 361 3u at 3 u 3r2 1 3u .+ — .
166 ADVANCED MATHEMATICS — . 60) = . Obtain u (x.I 2% h(e){in[l+b 2 2bcos(9e o )]}d0 Jo Verify that u satisfies Laplace's equation. c). 3xz dyz 3 . Show that the Neumann function for the threedimensional upper half plane z > 0 is Ir r/ where r and r' are defined by Equations (6.41 lb. Use the method of images to show that Neumann's function (N) for the interior of the unit circle is N = i n (b r : r 2 ) in the notation of Equation (6. y > 0 on y = 0 = ^ l l p(' g n[(xxo) 2 +(yyo) 2 ][( x . x) if u satisfies the following conditions .y .+ — . y) dy can be written as u ( x o'y o ) oo < x < °°.416). y) .= p (x. y.x o) 2+ (y f°° + yo) 2] ) dxd y ~ \ J00 h{jn[(xxo)2+yjj]}dx 10b.= h (x. If u satisfies Laplace's equation inside and on the unit circle and if on the unit circle show that u is given by u(b. lib.
64)] .  0. x —— =e st°8(xx0) Next. the Fourier transform of V/. The Green function G associated with the wave equation satisfies the equation 32G 1 92G x .t 0 ) ax 2 c2 at 2 Solve for G by first taking the Laplace transform of G with respect to t and verify that G (the Laplace transform of G) satisfies the equation 3 G S G _stn S .= 8 ( x .s t <>e.jr.t) = F(a)e~ i Y a l where F is the Fourier transform of f and y=h/2m.x o/c)/2s Deduce that G can be written as (c/2..t o ) >  x .o o < X < oo . aw Cf X > 0 as  x  —> °° Take the Fourier transform with respect to x [Equations (5. — . t 0 ) = ( \ 0.x 0 . .x o  ] otherwise Is G finite at (x o . s. The onedimensional Schrodinger equation for a free particle of mass m is [Equation (6.s ! x . G(x. t >0 at 2m ax 2 V/ and — y (x.t t ._ aw lfc— %2 a 2 \i/ = . oo<X<oo. is *F(a.. .— —r. . t . t). if (x 2 + y 2 ) < l otherwise " z z=0 12b. 13b. if [ c ( t . oo < y < oo 9u = (l.x o ) 8 ( t . 0) = f ( x ) .t Q )? Discuss the discontinuity of G. take the Laplace transform of G with respect to x and verify that G is given by G = (ce. 567 Z>0 .PARTIAL DIFFERENTIAL EQUATIONS II 2 V U= 0 .916a. b)] and show that *F (a.
.i) exp (ix 2 /4yt)] / 2 V2rcyt Deduce \\f if f(x) = 5(x . Show that G is given by G(x) = [(1 . i/ can be written as [ oo * oo ¥ = 2^ I F(a)e~iYa2teiaXd« = 1 f(^)G(x^)d^ J—oo J —oo The function G is Green's function.268 ADVANCED MATHEMATICS On inverting and using the convolution theorem.x Q ).
11) In Equation (7. mathematical modelling is now more realistic and a new field of numerical simulation has been opened up.150) is correct to at most one significant figure.1 INTRODUCTION Many engineering problems can not be solved exactly by analytical techniques.149 = 0. We will look. f (0. This type of error is known as loss of significance. We can now use iterative methods with much greater speed. for example. at Newton's method of approximating the solution of an algebraic or transcendental equation.150 radians. at the Gaussian elimination method for solving linear systems of equations and at the Euler and RungeKutta methods for solving initialvalue problems. in one form or another. Since the arrival of computers.CHAPTER 7 NUMERICAL METHODS 7. Many nonlinear equations which were intractable in the past can now be solved approximately by numerical methods. . In some cases.1500. the potential of these methods has been realized and the entire character of numerical methods has changed. This type of error is known as round off error and is unavoidable. the number of significant digits will be reduced. the value of sin (0. We first briefly examine the possible sources of error that may occur in the solutions obtained by numerical methods. Losing all significant figures during machine computations is not unheard of.150) is correct to three significant figures but f (0. Numerical methods. In this chapter. If we subtract two almost equal numbers. As a result of these possibilities.s i n x We now evaluate f(x) when x is equal to 0. We can also solve large systems of equations numerically. The knowledge of numerical methodology is essential for determining approximate solutions. we present different methods which will be useful to applied science students.12a) (7.001 (7. we generate errors. have been studied for several centuries. It is therefore necessary to check the numbers during the calculations. Most numbers have an infinite decimal representation and for computational purposes they have to be rounded to a finite number of decimal places.12b) (7. During the process of calculation. Successful simulations of complex processes in science and engineering are now being achieved. we consider the function f (x) given by f(x) = x . As an example.12a).150) = 0.
13b) The value of sin (0. 7. We need to establish that the sequence converges to the solution. In computational mathematics.000562 (7. In this sense.14) gives a better approximation to f (0.^x3 o Using Equation (7. numerical analysis is an art and not a science.13a) (7. It is often possible to provide an estimate of such errors..14944 and it can be seen that Equation (7. Alternatively we will also say that x is a zero of the function f. we can consider only finite sums. As mentioned earlier.21) for a given f. we consider infinite processes.I X 3 + . iterative methods are used to solve equations numerically. This equation could be algebraic or transcendental.21). For example.2 SOLUTIONS OF EQUATIONS IN ONE VARIABLE We begin our discussion by considering equations of the form f(x) = 0 (7.21) where x and f (x) are real or complex.x .( x . If x is a real or complex number satisfying Equation (7. In classical analysis. . This results in an error called the truncation error. We need to find values of the variable x that satisfy Equation (7.150) correct to five significant figures is 0. we can approximate sin x by its Taylor series and f (x) is approximated as f(x) . Ideally we should always perform an error analysis before accepting a particular solution which was generated numerically. Next we present four different methods of solving Equation (7. we can rely on intuition and past experience and hope that the results obtained are reliable and useful. In such a case.14) (7.12b). ) o .150) than Equation (7.21).13a).570 ADVANCED MATHEMATICS it is possible to rewrite the function to be evaluated in such a way that the loss of significant figures does not occur.150) = 0. such as infinite sums. we say that x is a root of the equation. A more detailed discussion on error analysis can be found in Hildebrand (1956) and Elden and WittmeyerKoch (1990). Often this is not possible as the problem is too complicated. These methods generate a sequence of numbers. we find that f (0..
with f(a) and f(b) having opposite signs. the required root lies in the interval (a. If f(xj) and f(a) have the same sign. One starts by letting Xj be the midpoint of [a. f(a) ^ r ^ f*x. b). The method consists of a repeated halving of the interval [a. there exists at least one number x such that a < x < b for which f (x) = 0. b] with f(a) and f(b) having opposite signs. b]. defined on the closed interval [a. as illustrated in Figure 7.b). We then proceed to find the midpoint of (xj.21 Identification of the root (•) of the equation via the bisection method . xj) and in this case x2 is the midpoint of (a. If xj satisfies Equation (7. b].21). until the difference between two consecutive values of xj is within the required accuracy. f(xj) has the same sign as either f(a) or f(b). the required root lies in the interval (xj. If not. We repeat this process of halving until a satisfactory value of the root is obtained.)TV 0 J L \ a x.NUMERICAL METHODS £77 Bisection Method (Internal Halving Method) This method is based upon the use of the intermediate meanvalue theorem which states that for a continuous function f. Similarly if f (xj) and f (b) have the same sign. xj is the required root. xj). That is to say. f(x) i.21. \ nr* jb x FIGURE 7.
1 0 = 0 (7. f (xj)].365. TABLE 7.5. Note that as k increases the difference x k .3651 f(xk) 2. The equation of the secant line is .21 Values of x^ and f(x^) k 1 2 3 4 5 6 7 8 9 10 11 12 13 xk 1. The values of x k and f (x k ) are given in Table 7. In this method. at least one root lies in the interval (1.37 1. we see that the root accurate to three decimal places is 1.3437 1.3652 1. It is not necessary that x 0 < Xj or that f(xQ) and f(xj) have different signs.2) and x1? the midpoint of (1.3655 1.162 0.3671 1. x j .3632 1.b) We note that f (1) is negative (5) and that f (2) is positive (14).25 1.22a.029 0. f (X])] which will intersect the xaxis at some point X2. Secant Method Consider the graph of the function f as shown in Figure 7.848 0.21. That is to say. Solve the following equation via the bisection method f(x) = x 3 + 4 x 2 .048 0. we consider two points [xQ.5.797 0.0038 0.x k _ 1 decreases and f(x k ) approaches zero.102 0. f (x 0 )] and [xj.3125 1.5) which is found to be positive (2.2).0127 0.21.25.5 1. We draw the secant line. is 1.0045 0.37) and the root lies between 1 and 1.375 1.572 ADVANCED MATHEMATICS Example 7.351 0. it is not necessary that the root lies in the interval [xQ.0003 0. The second approximation X2 is 1.3661 1.0017 From Table 7.3593 1. We continue this process.3651 1. We calculate f (1. Thus. that is the line passing through points [x 0 .21 for various values of k. f (x 0 )] and [xj.22.
.26) f (X k ) . we draw the secant line joining the points [xj. f (xj)] and [x2.21.NUMERICAL METHODS 573 f(x)f(x1) = f ( ' ° ) " ^ l ) ( x . we find in general f(x v ) (x k . \( ki.f (X k _!) The root is determined to the desired degree of accuracy.x 1 ) (7.23) f(x).X2 " fCx^fC^) Proceeding in this manner.22 Identification of the root (•) of the equation via the secant method The point x2 for which f (x2) is zero is given from Equation (7. f (x2)] and determine X3 as f (x2) (x2 . (k > i) (7.22. Use the secant method to solve the equation given in Example 7.: k+1 k k .23) by f( X l )( X l x 0 ) X 2 " X 1 fO^f^) (724) If f(x 2 )*0.. Example 7.x t 1) (?2 5) x k+1 = x k .x t ) X3 . X #„ \ Xp %*/?Y I f (XQ) FIGURE 7.
1.> f (xk) (k*<>) (7.26).365.7^7 f (x 0 ) (72"l0) Generalizing.26 Via Equation (7. we start with only one point.574 ADVANCED MATHEMATICS Taking x 0 to be 1 and x2 to be 2.29) = x o ._ .443678 . we obtain from Equation (7. In this method..14(5) (72"7a) = 1.64914 (7. . We assume xj to be an approximate root and set f (x^) to be zero in Equation (7.3652.(044368) (1.x 0 ) + .3667 .024291+0. Thus the root accurate to three decimal places is 1. Expanding f(x) in a Taylor series about the point x 0 yields f (x) = f (x0) + f' (x0) (x .2H) .262) 1..338 x . Newton's Method This is a popular and widely used method.338 = 1.28d) (0. ^ ^ (7. We note that we achieve this accuracy in five iterations whereas in the bisection method we needed twelve iterations.28b) ( 3 (7.024291) (1.1.24) X2 _ " 2 2 14(21) .649) (1.26) _o. We assume that f (x) is differentiable in the interval containing xQ and the root.^ .27b) X3 L26 (72"8a) = 1. we obtain *k+l = x k .649 x 4 .„_ X5 = L366? (7.1 338 . _ .443678 + 1. we generate " (1.338) 0.28f) = 1. which we denote by x 0 .29) yielding xl (7.3381.3667 .3652 We can calculate xg and it is found that x6 is also 1.
22) using Newton's method.214a) (7. That is xn = A or f(x) = x n . Suppose that the n* root of a number A is x. we deduce (7. we obtain f'(x) = 3x2 + 8x We start the iteration with x0 = 1 From Equations (7.210.216) .365 accurate to three decimal places after three iterations.45450.0037 = 1.c) Taking the derivative.212) Using Newton's method.214f) (7. Example 7.0856 = 1.22). we have f'(x) = nx"" 1 From Equation (7.4545 x 2 = 1. • Newton's method is also used to extract the n^1 root of real numbers.36890.3652 (7.213) (7.215a) (7. 11). Solve Equation (7.215b. we obtain the root 1.214b) (7.214e) (7. we have xl = 1 + ^ = 1.211).214d) (7.214c) (7. the root can be determined to the desired degree of accuracy.NUMERICAL METHODS 575 Again. On differentiating Equation (7.23.A = 0 (7.3689 x 3 = 1.
7320508 t ?33214) which approximates the actual root to six decimal places.21) in the following form x = g(x) If Equation (7.219d. The conditions on g which will ensure converge are (i) there is a closed interval I = [a. (7..221) (7.. . Fixed Point Iteration Method In this method.5 Xl = (7.73214 = 1. we have n = 2 and Equation (7.c) (7.e) (7.1.217) becomes xk+l =X k . k = 0.18b) Example 7.H h r ^ zxk = l(xk+A) (7.576 ADVANCED MATHEMATICS xk+i = xk^~r nxk (7217) In the case of a square root.24. we choose a starting point x 0 and then compute g(x 0 ).2. and we start with x 0 = 1. We now label g(x 0 ) as xj and compute g(xj).75 + j^) x 3 = 1 (1. x is said to be a fixed point of g(x). As in Newton's method. b] such that for x e I.220) until the root to the desired degree of accuracy is obtained.. g (x) is defined and g (x) e I.218a) (7.220) holds.2.219a) = L75 2 (L5+ fj) (7. We continue this process by writing *k+l = g(Xk).g) x2 = 1 (l.219f. Find the square root of 3. Here A = 3. we rewrite Equation (7.219b.73214+ = 1.
NUMERICAL METHODS 571 (ii) (iii) g (x) is continuous on I.725 0. Starting with the initial value TC/4. Example 7.733 0. there is exactly one solution in [0. TABLE 7.220) has exactly one solution.22. K/2].743 0. generate the values of x^ as given in Table 7.b) In this example. we can. g(x) satisfies all the conditions stated earlier.c o s x = 0 We rewrite Equation (7.760 0.725 0.738 71/4 0. • .74.740 0.741 0. which is achieved after six iterations.743 0. The mapping g is said to be a contraction mapping in I.25. Thus.740 From Table 7.736 0.22.221). solve f(x) = x . g' (x)  < k for all x e I and k is a nonnegative constant less than one. Equation (7. via Equation (7. g(x) is cosx. In the interval [0.760 0.748 0.748 0.707 0.222a.733 0.741 0.738 0. and g (x) is differentiable on I. we deduce that the root correct to two decimal places is 0.22 Values of x^ and g ( x ^ ) k 0 1 2 3 4 5 6 7 8 9 10 xk § ( x k) = c o s x k 0.738 0.736 0. If g satisfies conditions (i to iii).707 0.223) (7. TC/2].222b) as x = cosx (7. Using the fixed point iteration method.
227) = ^VlOx3.22) as x = iVlOx3 The iteration function g (x) is g(x) = i .360 x 7 = 1.365 . x f(x t ) g(xk) = x k .224) In the case of Newton's method (Equation 7. We note that for the secant method (Equation 7.220) is not unique for any given f.345 x 5 = 1.229c) (7.226) (7.375 x 6 = 1.5Z£ ADVANCED MATHEMATICS The transformation from Equation (7.228) Starting with the initial value x 0 = 1.229f) (7.368 x 8 = 1.229J) x 2 = 1. We can rewrite Equation (7. we generate.366 x 1 0 = 1.228).5 (7. g(xy) is . (7.229b) (7.229i) (7.21) to Equation (7.229d) (7.229g) (7.26).287 x 3 = 1. g (x^) is given by (7. in order to ensure that the iterations converge.T T ^ f (xk) Example 7.229h) (7.V l O . We should choose g so that all the conditions on g stated earlier.364 x 9 = 1. Solve Equation (7.229a) (7. are satisfied. the following xx =1.22) by the fixed point iteration method.x 3 The iteration formula is x k+l (7225) (7.229e) (7.26. via Equation (7.211).402 x 4 = 1.
231) do not converge to the root.232c) (7.365 in ten iterations. However.2 .484 x 4 = 0. the iteration formula is (7. is greater than one and condition (iii) is violated.26 ten iterations are needed to achieve the same accuracy. computed via Equation (7.365) but they are oscillating. Other forms of g(x) can be chosen.30 ) In this case.232a to d). g(x).621 (7. We now examine the properties of the iteration function g (x). we obtain xi = 2 x 2 = 0. For Newton's method. in the interval (1.228) converges to the solution 1.234). An alternative form for Equation (7.231) Again.225) is given by g(x) = 2xi ± ixi ± I 0 3x 2 + 8x (? 2 3 5 ) It is shown in Example 7. • . it is faster to compute g (x) as given in Equation (7.23 that the root is obtained after three iterations whereas in Example 7. The sequences generated by Equation (7. we find that  g'(x).232a) (7. we note that the x^ are not converging to the solution (1.22) is X = ^xT4) < 7 .227) compared to that in Equation (7. 1. We can verify that g (x) satisfies the conditions for the convergence of the iteration.5).235). starting with an initial value equal to 1.NUMERICAL METHODS _ _ _ _ 579 We note that the iteration given by Equation (7. The function g (x) and g'(x) are given by g(x) g00 = roTT4) = ~^±^ (x2 + 4x) 2 <7233> (7 .232b) (7. 2 _34) From Equation (7.833 x 3 = 2.232d) From Equations (7.
.3la. 1) 2) 3) 4) There are n (not necessarily distinct) real or complex roots. 4) are compatible provided V i = an (7.35a) (7.! x""1 + . then necessarily f(x) = ( x .b) with real or complex coefficients.. We note the following facts regarding Equations (7.580 ADVANCED MATHEMATICS For human computers. Newton's method is used for faster convergence.3la. n > 0.x o ) q ( x ) + R where q (x) = bn_! x""1 + bn_2 xn"2 + ...b) Newton's Method To use Newton's method for computing the root of an equation [Equation (7. 3. they occur as conjugate pairs. for calculating f (xfc) and f' (x k ) where f is a polynomial. Equations (7. + aj x + a0 = 0 where a n * 0. If xQ is a root of Equation (7.3la. (7. known as Horner's method.x o ) g ( x ) where g (x) is a polynomial of degree (n . 7.b).33) .34) (7. If all the coefficients are real and complex roots exist. Once a reasonably good approximation is obtained.1). b. Newton's method should be avoided if f' (x) is close to zero near the root.b) we can write f(x) = ( x . To avoid dividing by zero. + t>! x + b 0 and R is a constant. We want to find the roots (real or complex) of the polynomial equation f (x) = a n xn + a n .32) (7.3la. Recall that for any polynomial as given in Equations (7. If n is odd and all coefficients are real. We describe a method.211)].3 POLYNOMIAL EQUATIONS We now apply the discussion of the previous section to the special case where f (x) is a polynomial of degree n.31 a. we need to evaluate f(xjc) and f'(x k ). there is at least one real root. it is recommended to start with the bisection or secant method.
36a.35d) (7.e) (7.35c) (7.3 + 2 ( 5 ) = 7 and f'(2) = b2Xo + b i x o + b o = 21 (7.x 0 ) is not a factor of f (x). To compute f' (x 0 ).35a).35a to c) b2 = 3 bj = .j) (7.x 0 ) q' (x) Thus f'(x 0 ) = q(x 0 ) (7.3 = 0 find f (2) and f' (2).36a. .310a) (7. If f (x) = 3x3 .33) f(x 0 ) = R = a o + x o b o (7. Therefore. b). we have from Equation (7. we first obtain the constants bj using Equations (7. j . For x 0 = 2. we first differentiate Equation (7.31. .37) Example 7.310i.4 + 2(3) = 2 b 0 = 1+2(2) = 5 Hence f(2) = R = .310d.g. the bj are given in terms of a n and x Q .310b.b) In Equations (7. starting from Equation (7.33) and obtain f' (x) = q(x) + (x .35b) b 0 = a\ + x o b 1 R = a0 + x 0 b 0 Thus if (x . b 0 .c) (7.35a to d).310f.39a. . .38) (7. We can then determine f(x 0 ) from Equation (7.4x 2 + x .h) (7.b) . we can successively compute b n .NUMERICAL METHODS 581 bn2 = an_i + x o b n _i (7.
533 (7.533) = 4. .316a.g.3 .5 b 0 = l+2.3 + 2.5 = 0..313f.5) = R = a o + x o b o = 0. Equations (7.= ! ^ = 2.53.533) and f'(2. we obtain as a first approximation for the root Xj = 2 .3131 j ) We now compute f (2.e) (7.312a to d) We first need to calculate f(2.314a. ao = 3 (7.5(0.0227 Hence the second approximation for the root is x 2 = 2.211).32.008423 f (2.5) = 1.313a) (7. X4.25 f(2.315b) = 2532098 (7.b) (7. Solve x33x2 + 3 = 0 given that a root is near x = 2. accurate to two decimal places is 2.5) = b 2 x2 + b i x 0 + b 0 = 3.5). We note that a3 = (7.c) (7.b) We find that the root.582 ADVANCED MATHEMATICS Example 7. a2 = .533) = 0.315a) (7. Use Newton's method to evaluate it.h) (7.533 04°Q?273 (7. We can improve the accuracy of the root to any desired number of decimal places by proceeding to compute X3.5.533) in a similar manner to obtain f (2. 5 .313d..5) and f'(2.125 f (2.75 According to Equation (7.311) l. • . ai = 0.313b.35a to d) yield b2 = 1 b! = .
we consider the reciprocal variable — .321a.b) As previously noted. i Xj Example 7. and consequently Equation (7. we find f (0.b) (7. f(0) = 6 f (0) = 16 (7. 1].NUMERICAL METHODS 583 We now briefly outline a method for locating intervals containing the zeros of Equation (7. It thus follows that to approximately determine all the real zeros of Equation (7. it suffices to search in the intervals [1.°°). For example. Furthermore if x is a zero of f(x).0).1].317c) = ^k+anlx = Lf(x) x  + a l x n " 1 + a O^] We note that changing the variable x to 1/x allows us to look for solutions on the intervals [1. x ^ 0.322a) .l ) = 47.0) and (0.319a.318 to 21b). »). and x. From Equations (7. Similarly if f (xp f (xA < 0. 1/x is a zero of f. the zeros of f are in the interval (0. f (_1) = 47. 1 e (oo.33.1] evaluating both •s f and f to determine sign changes.31) has a zero between L and —.b) [ 1. f has a zero between x. Also f (i) = a n(i) n + Vl(xr1 + . in order to consider the finite intervals [1.317a) (7.320a. and if x e [1.317b) (7.125 (7. The converse is also true.e [1. Locate the roots of the polynomial equation f(x) = 16x 4 40x 3 + 5x2 + 20x + 6 = 0 (7.0) and (0. we deduce that there is no change of sign in the interval — Consequently. Recall that if f(xj) f (xp < 0.b) (7.31) where f(x) is a polynomial of degree greater than 2.l).318a.31).^ 0 + (7. to identify sign changes.1]. 0).0) and (0.5) = 0. f has a zero between Xj and Xj.40x + 16 We note that f ( .1] instead of on the original interval (<». To determine them we evaluate f at several points. f (1) = f (X) = 6x4 + 20x3 + 5x2 . J. We note that if x € (0.
.5M ADVANCED MATHEMATICS f (0. 2 .42b) Equation where A = (ay) is the coefficient matrix.1). 7.5. —L) = ( — .4la. 3 .42a) or Ax = b (7. + a l n x n = bj a21 x l + a 22 X2 +  + a2n x n = b 2 (7.8516 f (1) = 7 (7. The quartic can be written as a product of two quadratic expressions...b.75) = 0.75) and at least one other in the interval (0. ^ ) .c) is n Xaijxj i=i =b i ' i = l . (7.We have identified two approximate roots. 0. .b. The rows of A are linearly independent.42b) has a unique solution if the inverse of A exists.322c) Thus f has at least one zero in the interval (0. 2) and at least one other in the interval ( l . n (7. The rank of A is n.c) anlxl+an2x2 + + a nn x n = bn A more compact form of writing Equations (7.75.4 SIMULTANEOUS LINEAR EQUATIONS We now describe methods of solving systems of linear equations which are written as a n x 1 + a 1 2 x 2 + . 1) 2) 3) 4) 5) The determinant of A is not zero.4la. . One of them is known and the other is obtained by division. The homogenous system (A x = 0) has only the trivial solution (x = 0). The columns of A are linearly independent. It then follows that f has at least one zero in the interval l—^— . The conditions for the existence of A"1 can be stated in the following alternate forms..322b) (7. . and x and b are column matrices.
. . x2. We then choose the new coefficient of x 2 in the second row as pivot and reduce the coefficient of x 2 to zero in row three as well as in the subsequent rows. Gaussian Elimination Method The basic idea behind this method is to convert the n x n coefficient matrix of the given system to an upper triangular matrix. multiplying a row by a nonzero constant. We then operate on the augmented matrix A g .. > v n } are linearly independent. Vectors {v 2 . v n } is a set of vectors in Rn. The row operations described earlier are to be applied to column b as well.. v 2 .41. If A" exists. adding a multiple of one row to another row. where b is the (n+l) th column of A g . Similarly we reduce the coefficients of xj in all subsequent rows to zero. We note that the following operations on a system of linear equations do not change the solution of the system 1) 2) 3) interchanging two rows. We continue this process until we have only one unknown xn in the n th row. . v n } span Rn. We reduce the matrix A to a triangular form by the following elementary operations.. Vectors {vj. v 2 . It is economical to consider the matrix A and the column b simultaneously.j. we can determine the solutions for x n . otherwise A is said to be a singular matrix. We now discuss a standard method of solving the set of Equations (7. . Recall also that if {vj. We choose a n as a pivot and keep the first row unchanged. We multiply the first row by an appropriate factor and add it to the second row so as to reduce the coefficient of xi in the second row to zero. The matrix A g is now a rectangular matrix with n rows and (n+1) columns.. v 2 .41). xj. •• . we say that the matrix A is nonsingular. By backward substitution.42b) has a unique solution iff A is nonsingular. Solve the following system of equations via the Gaussian elimination method. v n } form a basis for Rn. . which is obtained by combining A and b . . Thus the linear system (7. Example 7.NUMERICAL METHODS 585. . We can solve for xn.. the following statements are equivalent. 1) 2) 3) Vectors {vj. .. . v 2 ..
we did not need to carry out further manipulations.43d) 1 2 2 0 0 3 (7.2 and adding it to row 3. we obtain (7.43e) 1 is the required triangular form. From row 3 of Equation (7.1 5" (7.43a) in matrix form as follows (7. as well as multiplying row 1 by .43b) L J L X 3J L J To carry out the required operations. Since a3 2 was already zero. we write the augmented matrix " 1 Ag = 1 2 2 3 4 3 3 7 5" 4 12 (7.43d) we can identify x3 = 2 From row 2. we choose a^ = 1 as pivot and we do not change the first row. yields the modified augmented matrix A — © " 1 A(g1} = 0 0 The matrix "1 A (1) = 0 0 1 2 1 0 3 0 .1 and adding it to row 2. Multiplying row 1 by .44a) .43c) For the row operations on the augmented matrix.586 Xj + 2 x 2 + 3x 3 = 5 ADVANCED MATHEMATICS x 1 + 3x 2 + 3x 3 = 4 2xj + 4 x 2 + 7x 3 = 12 We rewrite Equations (7.43a) 1 1 3 2 2 4 3 3 7 1 P 1 ] [5 x2 = 4 12 (7.
9 and this allows us to determine X3. a pivot row must be selected.9 . However.45b) If. we would not be able to proceed.44b) (7. That is. since the coefficient a 22 is zero.4.44c) (7. x 2 and Xj provided the vector b is known.9 9 10 4 " (row 2 .45b) as "1 A(2) = 2 4 " (7.5 x row 1) (7. interchanging rows 2 and 3 allows us to eliminate the coefficient of x 2 in such a way as to proceed towards the computation of the required triangular matrix.44d) The following observations concerning this method are important. we were to choose the second row as the pivot row.45a) Choosing the first row as a pivot row yields " A(1) = 1 0 0 2 0 .r o w 1) (row 3 . At each stage in the Gaussian elimination procedure.1 0 0 0 . Appropriate multiples of the pivot row are added to other rows to perform the desired elimination. we write Equation (7.5c) 0 . . We illustrate a problem which could arise by considering the following nonsingular matrix 1 A = 1 5 2 2 .NUMERICAL METHODS 587 X2 = . in the usual manner. The selection of this pivot row is very important.1 Substituting the values of x 2 and X3 in row 1 yields X ! + 2 ( .l ) + 3(2) = 5 and the value of Xj is xx = 1 • (7.5 1 10 4 " (7.
We interchange the third row with the first row and write the augmented matrix as 5 Ag = 2 1 2 5 2 . this strategy will always work. After the first column of zeros has been determined. .46a. as described.588 ADVANCED MATHEMATICS This example illustrates the need for adopting an appropriate pivoting strategy. If the matrix is nonsingular.42. Example 7. (i) (ii) (iii) Scan the first column to find the largest absolute value  a j j .3 0 l 6 (7.c) (7. scan the second column below the first row to find the largest absolute value  a i 2 1. i * 1.b. One such strategy is as follows. Solve the following system of equations using the Gaussian elimination method.46d) 4 23 Choosing the first row as the pivot row. The row in which this element is found now becomes the second row.3 x 3 = 0 Here the largest absolute value in the first column occurs in the third row. Xj + 5x 2 + 4x 3 = 23 2xj + ix 2 + x 3 = 6 5x x + 2 x 2 . Exchange this row with the first row. we proceed to generate zeros in the first column. (iv) This process is continued until the matrix has been transformed into an upper triangular matrix.
47a.5 (7. yielding Xl = 1.6 0 4. x 2 = 100 (7.5 7.b) is X! = . Performing the elimination.410a. When the matrix is singular.1 r o w 1) J (row 3 .*.b) (7. we obtain X2 and finally.b) is .NUMERICAL METHODS 589 5 2 .48a.01x1 + x 2 = 2 The exact solution of Equations (7. we do not have a unique solution.49a.48a.4_6f) 0 0 The last row identifies X3 as 7.46e) 46 46 ° 10 10 23 We now look at the second column and note that ~ > .6 23 2.3 0 3 22 A(g1} = ° " T o To ft 6 (row 2 . We then interchange the second and third rows leaving the first row unchanged. x3 = 3 (7.5/2.. as stated earlier.b. Substituting this value of X3 in the second row. x2 = 2 .b) The exact solution of Equations (7. we find the new matrix to be " 5 A<2) = 2 . We examine the case of illconditioning by considering the following two systems of equations x l + X2 = 1 (7. we obtain x^ .49a.b) (i) Xj + 1.9 9 .5.c) The second kind of problem that can arise in the solution of a system of linear equations deals with singular or illconditioned (nearly singular) matrices.1 row 1) (7.01x2 = 2 and x t + x2 = 1 (ii) 1.3 0 " 4. from the first row.
11x1 = 0 iff x = 0. . we assign a real number which we call a norm and denote it by II x II. We define the closeness of two matrices by the condition number which in turn is expressed in terms of the norm of the matrices.41 la. we find that Equations (7. The determinant of each of the coefficient L l iJ matrices is nearly zero.001 1 ri which are.590 ADVANCED MATHEMATICS X! = 100. A norm shares certain properties with absolute values. the solution (99. A norm satisfies the following properties.414a. 100) is changed to (100.01 ' r I I 1 1. x2 * 2 (7. 99). in a sense.412a.b) can have an incorrect approximate solution.b) The equations are also satisfied approximately by xx « 0 . The coefficient matrices in the three previous cases were 'i 1. close to the singular matrix ii . 1) 2) 3) llxll > 0 for all x .b) Note that for a small change in the value of the coefficient in the second equation.413a.001 The exact solution here is X! = 1 .01 ii 1 ' r I li 1.412a. I IcxI I = I c III x I I for any real constant c. To each vector (or matrix) x. x2 = .9 9 (7. from one quadrant to another quadrant! Consider the system of equations Xj + x 2 = 2 1. II x + y II < II x II + II y II (triangle inequality). x2 = 1 (7. That is to say. Illconditioning is thus in a sense a measure of the closeness of the coefficient matrix to a singular matrix. Vector and matrix norms are defined as follows.b) Here.001 Xj+ x 2 = 2.b) (7. The above examples are cases of illconditioning.
415 a.43. irrespective of the direction. The familiar Euclidean norm. The L p norm is defined as Lp = I I x lip = ( X l x i l P ) 1 / P for p > l where Xi are the components of x.. + xl The I IxI I x norm is defined as 11x11^= max Xj ~ l<i<n (7. The maximum value of the column sum is IIAII1. the Lj norm is the sum of the absolute values of the components of x . as defined by Equation (7..419a. The Lj and L^ norms of a matrix are defined as n Lj = HAH. II j norm. is obtained by setting p = 2. For p = 1. To calculate II A. is obtained by adding the absolute values of the elements of each column of A. In this case. I I ^ we interchange the roles of columns and rows and repeat the process described earlier in the determination of IIAII j . Calculate IIAII j and IIAII^ if . 591 The first condition states that all nonzero vectors have a positive norm.418a.b) L^ = HAIL = max X ~ l<i<n j = 1 a ii J (7. The second property implies that the norm of a vector is invariant.b) The II A. = max V l<j<n i = 1 aH J (7. which is the length of the vector x. as a measure of x.NUMERICAL METHODS .416) (7. It is easier to compute I I x II ^ than I I x II2 and it is not surprising that I IxI I ^ is preferred to I I x II2 in numerical analysis. The third property is the usual triangle inequality.415b) becomes llxll 2 = Vx} + x^ + .b) (7.418b). Example 7.417) The I IxI I ^ chooses the maximum absolute value of x. Equation (7.
426a) II A"1 II II 5b II = HAlFimF  iibi. we have IIA x II = libII and lib II < HAH llxll (7.421) In terms of norms.42b). The maximum is 8 and HAIIj = 8 Similarly.424a) (7.3 4 2 2 ' 2 (7. We now investigate the effect of slightly changing the vectors x and b in Equation (7. Suppose x is perturbed to x + 5 x and b to b + 5 b.425a) (7.424b) (7.422) (7.4 " 26b) .423) yields A5x = 5 b and 5 x = A' 1 8 b (7.425b) We can then write 115x11 < II A"1 II II 8b II II 5x II or (7.420) 3 2 The sum of the absolute values of the columns is 5. and 6 respectively. ' (7 .423) (7. that is IIAIIM = 9 • A norm of a vector or a matrix may be considered to be a measure of its size.592 ADVANCED MATHEMATICS 1 A = 0 .42b) and (7. Then A ( x + 5 x ) = b + 8b Combining Equations (7. the rows yield a maximum of 9. 8.
If K ( A ) is large we say that the matrix is illconditioned.oi i ' ] The Lj and L ^ norms are n < 7 4"27) Li = HAH.1 1 " " (7. We can then expect that a small change in b will produce a large change in x resulting in loss of accuracy.01. Note that K ( A ) is undefined for a singular matrix. The resulting n x 2 n matrix is in the case of the matrix A given by Equation (7. Example 7.c) n ^ = UAH = °° = max T 1 ^ 2 j = 1 aJ = max(2.b.428a. We separate the matrices A and 1 by a dotted line.426c) Equation (7. we use the GaussJordan method which proceeds as follows.. The other elements in the last row are zeros. 2) = 2. We then perform the Gaussian elimination process from the n t h row upwards until A has been transformed to I and I is then transformed to A" . Determine K ( A ) using the Lj and L ^ norms where ^ .01 (7. r • i T 1 I_ • A"1 . 2. To the matrix A .NUMERICAL METHODS 593 or 118x11 i II 5b II Jff * HA II II A' 1 I I . This quantity IIAII II A" II is called the condition number or COND ( A ) and is denoted by K ( A ) . we adjoin the identity matrix I and we obtain a n x 2n matrix which we write as y A : I_ j .01 J (7.[ i. We use the Gaussian elimination method to reduce A to the usual triangular form.b. = max Y i^j<2 i= 1 a.44.oi i : o i ] (74"3Oa) ." II with the relative change of vector b. We divide the resulting last row by an appropriate number so that the element of the last row and last column is one.426c) shows that a relative change in the solution vector is given by the product of IIAII II A.427).c) To obtain the inverse of A.= max(2.01) = 2. performing simultaneously the same operations on I. Below we illustrate the method 1 : 1 0 " [ A ^ ] = [i.429a. The choice of norm is not important.
01 : : 1 0 " 1. we find that K(A) is given by K ( A ) = IIAIIj II A"1 llj = 2.432) II A" IIOQ is given by the maximum of the row sum (200.01 x 201 = 404. K(A) is K ( A ) = 2.c) (7. 200) WA'l\\l = 201 (7431) (7.433) .01 1 J (row 2 .b) (7. 0 1 row 1) (7.01) I * 1 0 : 100 100 1 (row 1row 2) o The inverse of A is T100 100" i : io.434b. (201.435a.430b) I " 1 0 1 1 : • : 1 0 rj 4_30c) 101 100 J (divide row 2 by0. loo J (7430d> t'1 • [ 1Oi 100J The norm II A llj is given by the maximum of the column sum. 201) II A'1 IL = 201 Using the Lj norm.434a) (7.01 Similarly using the L M norm.01x201 = 404.01 (7.1 .594 ADVANCED MATHEMATICS i " 1 0 1 0.
438) IIABII = 0. the coefficient matrix A is written as the difference of two matrices B o and C o . when the number of variables is large and when the coefficient matrix is sparse (has many zero entries). Equation (7. Sparse matrices of high order are of frequent occurrence. iterative techniques are preferable in terms of computer storage and time requirements. Iterative Method When the dimension of the system of linear equations to be solved is small.440) . When using an iterative method.437) (7.NUMERICAL METHODS . $95 We conclude this section by stating the nearest singular matrix theorem which helps to circumvent the cumbersome problem of finding A" . However.440) by the inverse of B o we obtain (7. An example is when partial differential equations are solved numerically (see Chapter 8). we have K(A) > 2. we choose B to be "1 B = i 1 i (7. If A and B are a nonsingular and a singular matrix respectively ^BT £ "^'" (7'4"36> Using this result in the above example.42b) is now written as Box = b + Cox Premultiplying Equation (7.01.} — = 102 (7.439a.01 So that II A"1 II > .01 x 102 which is of the same order of magnitude as that obtained earlier. the Gauss elimination method is useful.b) With II A II = 2.
with the largest coefficient xl = l ~ IX2 + IX3 = 7 + 7 x l + f X3 ( lSt r o w ) ( 3rd r o w ) (7.b. Jacobi's method In Jacobi's method. That is to say. Example 7. Substituting these values into the right side of the set of equations generates new approximations that are closer to the required value. to solve for the variable with the largest coefficient.c) ~^X2 + ^X3 = "^ We first write Equations (7.45. we can generate a sequence of values of x from Equation (7. Each component of x on the right side of Equation (7. We first consider the method of Jacobi and then the GaussSeidel method. choosing whenever possible. c) for one variable in each row. b.442) A judicious choice of BQ and CQ will lead to a sequence of x k which will converge to the required solution x. § 0 is chosen to be a diagonal matrix.44 lb) as follows x k+l = ixk + £ (7. We begin with some initial approximate values for each variable.443a. These new values are then substituted in the right side of each equation to generate successive approximations until the difference of two successive approximations of each variable is within the desired degree of accuracy. solve the system of equations o X i 4" X<2 — Xo — O 2 x 1 + x 2 + 9x3 = 12 xl (7. we solve each equation for only one of the variables.444a) (7. Using Jacobi's iteration method.442) may be taken equal to zero if no information is available.44 lb) Starting from an initial guess x 0 .443a.596 ADVANCED MATHEMATICS x = BQ1 b + B o! (: O X = Bx + £ (7. as shown in the next example.444b) X2 .44la) (7.
446a) Equations (7.442)] are x{ k + 1 ) = 1 .1 Co = 1 2 and A = B o .445) The matrices B o and C o are chosen to be ' 8 Bo = 0 0 .444a to c) or Equation (7.441a.x(k)_ix(k) ( 7 4 .446c) (7.47c) .NUMERICAL METHODS ££Z X3 = f .1 ^ + 1 ^ X(k+1) = (7.444C) We interchange the second and third equations of Equations (7. The matrix A is "8 A = 1 1 .Co ° 1 0 0" 0 9 1 " ~2 0 (7. We denote the i* component of x k by x V and the iterative formulae [Equations (7.\ X2 (2 nd row) (7.444a to c) are obtained by substituting Equations (7.7 2 1 1 " 2 9 (7. b) into Equations (7. b). x l .447a) ( 7 4 A lx(k)+9 (k) _4?b) x ( k + l) = 4__2.443) so that the diagonal elements of the coefficient matrix A are the elements that have the maximum absolute value for each row.446a.446b) (7.7 0 " 0 .
we have made use of the values of x.453) is the GaussSeidel iteration method._i r1 ** ii J (7452) Equation (7. . xf = I (7. in computing x^ the values of x{ k + 1 ) and x£ k+1) in Equation (7. we obtain x f = 1.452) instead of x k ) and x^ k) .c) *!k+I> = ^ .449a.^ .453) j=l n j=i+l n Equation (7.450a.45. we eventually obtain the exact solution which is Xj = x 2 = x 3 = 1 • With the choice of B o as described in Example 7. although the .t ii .b.i ^ *r 11 <7. Equation (7.448a to c) into Equations (7.b. we take xf» = x f = x f = 0 Substituting Equations (7. Equation (7.095 . For example.452) to compute x> values of x.095 .c) xf = 1 .048 (7.598 ADVANCED MATHEMATICS As a first approximation. By so doing. we hope to be able to accelerate the convergence. we can substitute ^ (j < i) are already known.447a to c). + only. GaussSeidel method (k+1 ^ (k*) In Equation (7.42a).452) is modified to *sk+i> .448a.b.452) is the Jacobi's iteration formula for solving the system of Equations (7. x f = 1.b.c) Substituting the values of x[ ' into Equations (7.c) Repeating this procedure. xf = i . x f = 1. + .442) can be written in component form as (7.1 £ *r° .45 la.447a to c) generates (7.
000 .447a to c. Solve the system of Equations (7.001 1.b.443) by the GaussSeidel method.571 1. c) are modified to x (k + l) = i + lx(k+l)+lx(k) (74_54a) x(k+i) = ^2x(k+i)_ix(k+i) (74_54b) Using the same initial values (x 0 = 0).000 0 1.455a.000 0 1.993 1.000 1.003 1. GaussSeidel (k) Ao Y (k) Ai Y (k) A/x Y (k) A« (k) A^ (k) Ao 0 1 2 3 4 5 6 7 0 1.968 0.967 (7. Equation (7. 1.005 0. Using Equations (7. the exact solution (1.967 1.994 0.45.46.c) Continuing the process. 54a. we obtain xj 2) = 1.027 0.000 0 0. generated by the methods of Jacobi's and GaussSeidel Jacobi .447a) remains unchanged.000 0 1.001 1.000 1.109.998 1.000 1.990 0.001 1.002 1. the first iteration gives the same values for x j .NUMERICAL METHODS 599 Example 7. 1) whereas Jacobi's method requires seven iterations. we obtain. Table 7.41 Values of x.095 1. after five iterations.999 1.095.447a. TABLE 7.333 1.333 0.095 0. Equations (7.988 1. xf] = 0.095 0.048 0. b).998 1. x® = 1.002 1.571 1. K. 47a.982 1. The coefficient matrix is rewritten as in Example 7.109 0. 54a.41 gives the values of x ® generated by Equations (7.000 0 0. b).447b.
is an eigenvalue of A and x is the corresponding eigenvector Ax^ = lx_ (7. If A. In this section. A real matrix Q is orthogonal iff Q"1 = Q f If A is a real symmetric matrix.54) (7. eigenvalues arise in connection with vibration problems in mechanical engineering. We say that a matrix A is similar to another matrix B if there exists a nonsingular matrix P such that A = P"'BP If A. in this case. we briefly summarize the basic results and definitions. Most of the algorithms for estimating eigenvalues of a symmetric matrix A make use of similarity transformations and are therefore carried out in two stages. If A is triangular or diagonal.5 EIGENVALUE PROBLEMS Problems involving the determination of eigenvalues and eigenvectors often arise in science and engineering. The general problem of finding all eigenvalues of a nonsymmetric matrix is much more difficult as it easily leads to stability problems with respect to perturbations. there exists an orthogonal matrix Q such that Q"1 A Q = C (7. the matrix is reduced to a suitable form and.600 ADVANCED MATHEMATICS 7. in the second stage. is an eigenvalue of A with associated eigenvector x.51) Suppose that matrix A is similar to B .52) (7. the diagonal entries of A are the eigenvalues of A. Thus the diagonal entries of C. then A is also an eigenvalue of B with P x as the associated eigenvector of B . Before proceeding with the method. For example. the method of determining the eigenvalues is executed. In the first stage. are the eigenvalues of A .53) where C is a diagonal matrix. we study the problems of calculating the eigenvalues of a square matrix. in discussing the stability of an aircraft in aeronautical engineering and in quantum mechanics.
58) . the Hermitian transpose of A is the transpose of the complex conjugate of A and is denoted by A .57) (7. If A = AH (7. 0 4 5J 0 3 4 2 0 0 0 2 15 _ 0 0 0 5l_ We next describe a method of transforming a real symmetric matrix to a tridiagonal matrix. there exists a real orthogonal matrix Q such that Q A Q is diagonal. The diagonal elements of A are then its eigenvalues.55) the eigenvalues of A are all real and A is known as a Hermitian matrix.56) The only entries in A that could be nonzero are the elements along the diagonal. A matrix U is a unitary matrix iff U" 1 = U H A matrix A (= a^) is tridiagonal if a^ = 0 whenever I i . Let v be a unit vector (column matrix). Householder Algorithm It was stated earlier that for any real symmetric matrix A.j I > 1 (7. Consider the matrix Q = I . the subdiagonal and the super diagonal. If A is a matrix with complex entries. Examples of tridiagonal matrices are 3 10] .NUMERICAL METHODS 601 The eigenvalues of a real symmetric matrix are real numbers and their corresponding eigenvectors are mutually orthogonal.1 2 0 0 0" 2 13 0 0 124 . The methods proposed so far do not always converge. Attempts at diagonalizing A have not always been successful. Instead we transform A to a tridiagonal form via the method of Householder which is economical and reliable.2 v vf (7. and the columns of Q are the eigenvectors.
Example 7.59a) (7.59b.510b) (7.2 v k v J (7.2 v v f ) = I4vv + 4 v v t vyt = I (7.59b). the matrix A n 2 will be a tridiagonal matrix. Reduce the matrix A to a tridiagonal matrix by the method of Householder.510d) We choose v k such that its first k elements are zero.511) We choose v j to be 0" v} = v2 _V3_ V2 + V3 = 1 (7.2) operations.59c) follows from Equation (7.2 v y t ) ( I . After (n .c) t Equation (7.512b) (7.510a) (7. since v is a unit vector. we perform the following sequence of transformations Ao = A Ak = Q k A k _ ! Q £ Qk = I . The product v ' v is a scalar.51. Note that v v' is a symmetric matrix and its transpose is v v ' . if A is given by all a 12 a 13 A = a12 a13 &22 a23 a23 a33 (7. we note QQ1" = ( ^ .512a) . To transform a matrix A to a tridiagonal form. We illustrate the method by the following example.5lOc) XkXk = l (7.602 ADVANCED MATHEMATICS To verify that Q is orthogonal.
we have " 1 0 01 Ql = 0 1 0 0 0 0 2v\ 0 2v 2 v 3 (7. 2v 2 v 3 ^2 ^3 0 " a (1) all 2v 2 v 3 a (1) a 12 a U) a l3 " 12v J L a13 ^ 3 ^ 3 J [ 0 2v 2 v 3 l2v 2 (7.510c). we obtain 1 Ai = 0 0 0 I" a n a l2 a i2 a13 1 1 0 0 12 v^ 0 2v 2 v 3 12V2.515c) (7.515f) .513a) .513b) Applying Equation (7.515d) (7.515b) (7.2 $ (2v 2 v 3 ) a23 + a 22 (2v 2 v 3 ) 2 For A(i) to be tridiagonal.NUMERICAL METHODS 601 From Equation (7. we have from Equation (7. 2v 2 v 3 2v 2 v 3 l2v 3 (7. 0 0 1J 1 0 0 0 [ 0 0 2v 2 v 3 2vf 12 v2.510b).515e) (7.2 $ (2v 2 v 3 ) (a 22 + a33) + a23 (2v 2 v 3 ) 2 Ql = a33 (12v 2 ) 2 + 2 ( 1 .514a) = a^ a (1) _ a 13 a^ a (1) a 23 aS a (1) a 33 _ (7.515a) (7.514b) where a™ = a n a(12} = a 12 (12 v\) + a 13 (2v 2 v 3 ) a(13} = a 12 (2v 2 v 3 ) + a 13 (12v^) ^ 2 = a22(l2v2)2 + 2a23(l2v2)(2v2v3) + a33(2v2v3)2 a g = a 23 (12v 2 ) 2 + ( 1 .15c) (7.5.
we choose the negative sign and if aj 2 is positive.2 v 3 (a 12 v 2 + a 13 v 3 ) = 0 Since Q is an orthogonal matrix. we deduce v 3 = ~+ ^  ( 7 '521) (7.520) * = k[l * If) From Equations (7.522) Since we have v 2 in the denominator. 20).518a) (7.518b) Multiplying Equations (7.520) into Equation (7. we obtain a 1 2 v 2 + a 1 3 v 3 = +SjV 2 Substituting Equation (7. we choose the positive sign.17) becomes [ a 1 2 .519) (7. we choose the sign in Equation (7.519) yields (7. it follows that (7. Equation (7. b). adding the resulting expressions and using Equation (7.512b).521).2 v 2 ( a 1 2 v 2 + a13v3) = ±Sl (7. we obtain v2 = ± / l [i i a l2 ( si g" a l2)j (7.523) .521) such that the absolute value of v 2 is the larger of the two possible values to achieve the best accuracy.2 v 2 ( a 1 2 v 2 + a13v3)] =a 1 2 2 + a123 = s\ Equation (7. 19) by v 3 and v 2 respectively.5. if aj2 is negative.604 ADVANCED MATHEMATICS a 1 3 .516) a ^ + an^nf +^l2 (7517) Note that a J3 is zero Using Equations (7.518b) implies a 1 2 .516.516. From Equation (7. That is to say.515a.
The orthogonal matrix Qj is now constructed (Equation 7.526) Then v. + a i 2 n (7..1 [l + *•»<*•»")] ah (sign a12) vij = J2S v (7. If the matrix A is a (n x n) matrix (n > 2).515a to f).. The choice of v 1 2 ls different from Vjj (j > 3) because we want to reduce the matrix A to a tridiagonal form [see also Equations (7..522. we have considered a ( 3 x 3 ) matrix.NUMERICAL METHODS 605 We choose the positive sign in Equation (7.522). we start with v j given by l\ = (°'vl2'v13''vln) (7524) Then from Equation (7. Let Sj be given by Sl = a122 + a123 + .22 . A j will then be of the form . Substituting v 2 and v3 into Equation (7.510c)..523) and it follows that we take the negative sign in Equation (7. 0 x x (7.525) 0 x x x where the x denote nonzero entries. .527a) (7..51.527b) is valid only for j > 3.510c) and the operations defined by Equations (7. 23)]. Qj will be of the form 1 0 0 Ql = 0 x x 0 x x .. • In Example 7. the matrix A (1) is tridiagonal.510a.527b) > J^3 Note that Equation (7.. b) can be performed..
... We write v 2 as a row vector y j = (0.." A . . the first row and column are of the desired form.d) v13 = ..530) .531a.527a...f) .. 0 X Aj = 0 x x . . Use the Householder method to find a tridiagonal matrix similar to the matrix "1 A = 2 2 . We now proceed to reduce the second row and column to the form of a tridiagonal matrix. 6(V273 ) From Equation (7..1 2 S as defined by Equation (7.1 1 3 2 2 1 1 2 2 1 . b) (7...b) V12 = VH1+1] = Vl (7.532c..J L = .526) is given by S^ = (l) 2 + 2 2 + 2 2 = 9 The Vj: are given by Equations (7. as shown in the next example. x (7. v2n) (7.529) We then proceed to construct Q 2 and obtain A 2 as described earlier.1 (7..b) (7. Example 7. The process is continued until we obtain A n 2 which will be of the tridiagonal form.52. Ql is = ^f 6 (7.^ 6(V2/3) 6 v14 = . v 23 .510c).532e. 0.606 ADVANCED MATHEMATICS X X X X X 0 .532a.528) 0 x x x In Aj.
we can proceed as before by writing v 2 as a row vector y j = (0.535) 105 The matrix B can be reduced to a tridiagonal form as in Example 7.b)C) = V 0.536) S ^ ( a £ ) 2 + (a<' 4 >) 2 =(!f + (if=5a The elements v 22 and v 23 are (7.1 2 . 0.5.538b) (7.37a. 1 7 25 1 10 (7.1 2 27 34 7 1 0 7 0 1 (7. We can partition Aj into a (3x3) matrix B.538e) . v23) S 2 is now given by 2 (7. v22.9975 V 23 (7. We now proceed to reduce to zero the element a 2 4 .538c) = ^/[2(^)V22] = 0.99497 = 0. choosing B to be 34 B = 1 7 .NUMERICAL METHODS 607 ~ 3 t 0 0 0 0 .1 2 2 (7533) 0 0 9 2 2 .534) 0 0 25 10 105 Ql = T = 3 ! Aj = 1 9 27 The first row and column in Aj are in the desired form. Alternatively.51.0709 (7538d) (7.
7778 0. the elements which have been reduced to zero in the first column and row at the first iteration remain zero at the second and subsequent iterations. Once the matrix has been reduced to a tridiagonal matrix.6Q8 ADVANCED MATHEMATICS The orthogonal matrix Q 2 is given by ?2 =1 " 2l2l\ (7. the most efficient method of obtaining all the eigenvalues is the QR method.6000 0 0 (7.7857 0 0 0.8000 The tridiagonal matrix A 2 is obtained by the similarity transformation A2 = Q 2 A j Q  (7.539a) 0 1 0 0 0 0. = 0 0 3 3. The main idea is based upon the following result of linear algebra. In general. This property made Householder's method very reliable. The QR Algorithm We now discuss the powerful QR algorithm for computing eigenvalues of a symmetric matrix.9899 1 0 0 0 A 2 is then calculated and is 1 3 A.6000 0.1414 0 0 0 (7.541 a) = Q2QIAQ1QJ (7.539b) 0. .0222 0.540) 0. all the zeros created in previous iterations are not destroyed in subsequent iterations.1414 0.7857 3.541c) = P A Pf • We note that in the Householder method.9899 0.541b) (7.
545).542) Since rotation matrices are orthogonal matrices. we deduce that c = an (7.545) (7.543d. there exists an upper (right) triangular matrix R and an orthogonal matrix Q such that A = QR (7. P2 = r i 0 o c o i s .546b) V a?i + a ? 2 With this choice of c and s and denoting Pj A by Aj .544b) We set the element in the second row and first column of P l A to be zero.543a. we use rotation matrices to transform A to the triangular matrix R.NUMERICAL METHODS 609 If A is an (n x n) real matrix. P3 = r c 0 o 1 s 0 _ 0 0 1 J [ 0 s c J [ s 0 c _ (7.c) where c = cos 0 and s = sin 0 (7. that is sajjca12 = 0 From Equation (7.b.546a) V Al + a?2 s = ai2 (7. a rotation matrix may be one of the following c Pj = s s c o I 0 .e) Let A be a (3 x 3) tridiagonal matrix and form the product of A with the rotation matrix P j c PjA= s s c 0 1 f an 0 a i2 a 12 a22 0 a23 (7. In the case of a (3 x 3) matrix.544a) 0 0 1 J _  0 a 23 a 33 _ = sanca12 0 sa 1 2 +ca 2 2 a 23 ca 2 3 a33 (7. we write .
b) . we require (1) c2 = a22 (7. (1).550a) J . If we denote the product P 2 A j by A 2 .548b) C D (l) 0 0 (i) (l) c2a22~S2a32 C2a23~S2a33 (1) (1) ^2 a 22"*" ^ 2 a 3 2 (1) (1) ^2 a 23"^"''2 a 33 where c 2 and s 2 are cos 9 2 and sin 0 2 .549). we have to set s 2 a 2 2 ) + c 2 a (32) = 0 (2) (7. (1).547) (l) C D a 33 _ We now form the product P 2 A } as follows ~ (i) (i) (i) " 1 ?2Al = 0 C2 0 1 ~S2 c7 11 0 12 (1) a 22 13 (1) a 23 ° 0 (7.549) for the element a 3 2 to be zero. To satisfy Equation (7.670 " a ADVANCED CD ll (1) a12 MATHEMATICS (1) " a13 Aj = . 0 ° a 22 a 23 a32 (1) (1) (7.548a) s9 (l) (l) L 2 (l) all 2 J L 0 a 3 2 a33 _ (1) al2 (i) a 13 (7.55Ob) +(a32) Thus A 2 is now of an upper triangular form and we denote it by R. Summarizing we have R = P2A! = P2PjA Generalizing to a (n x n) matrix. 2 . 2 M (a 22 ) + (a 32 ) (i) s2 = / V(a 2 2 ) 32 (7. we state that (7.551a.
given a collection of experimental data points [x k .NUMERICAL METHODS _677 R = gnl A n2 = P n _! P n . . n . Similarly we choose P r so as to reduce the element aj. we address the following problem. many of the elements are already zero.553a) (7.. of any matrix A r (r = 0. 7. b.1) will be of the form given in Equation (7.552a... Lagrange. Lagrange Interpolation This method is based on choosing a polynomial p n (x) of degree n for which p n ( x k ) = f(x k ). and Newton. The terms c and s are given by c = . f (x k )]. the c is along the diagonal elements and the s is off diagonal. (r) . In a tridiagonal matrix... 2 . . .. . k = 0. b.is . The elements of P r are zero every where except (i) . a[[ s = (7.. we can successively reduce the nondiagonal elements in the lower triangular section of the matrix to zero.s ..b) The matrices P r (r = 1.2. n . A standard text on the eigenvalues of a matrix was written by Wilkinson (1965).543a. and the reduction to an upper triangular matrix is fast. Pj is s and Pj. (11) along the diagonal where they are one with the exception of P^ where they are c . x n ). xn) or a curve that passes through the data points. An upper triangular matrix R is then obtained and the eigenvalues are the diagonal elements. c).1. l . We note that in Equations (7. n. _(r) . . . c) except that they will be (n x n) matrices.2 ) to zero.553b) jl Vafi + afi By this process..543a. The points x k are sometimes called the nodes. This problem has been studied by well known mathematicians such as Gauss. . 1. The earliest method requires one to fit a polynomial (an interpolation function) that approximates the function f over (x 0 .61) ..6 INTERPOLATION In this section. .g i A (7. (7. We wish to estimate the value of f (x) for some x in the interval (x 0 . k = 0.
5) This equation can also be written as M^fof^+fl^ x0~xl xl~x0 C7. ijCx.68c.b) (7.) = 0 . fj) is p ( x ) = x l f o~ x o f l Xl~x0 +x lozlL X 0~ X l (7. therefore. x n } in the domain of a function f with n > 1 and x 0 < Xj. x n are distinct points on the domain of f. x j .b) Thus. XJXQ a = iozlL xo~xl (7. x^. < x n . f0) and (x^.^ XQ (7.66b) = f o i o (x)+f 1 i 1 (x) where io(x) = .67a. there exists at least one polynomial p n (x) of degree less than or equal to n such that Equation (7..612 ADVANCED MATHEMATICS An existence theorem for polynomial interpolation states that given (n + 1) points {x0.68a. .i .63b) ao = ^iloZ^L. ...66a) (7. Moreover. .64a.6. f0) and (x1? fj) will define a 0 and aj as follows Pl(x 0 ) = a o + a i x o = f0 Pl(xi) = aQ + ajXi = fi Solving for aj and a 0 .d) Generalizing.^ L and i^x) = ^ . we find (7. .63a) (7.62) This equation has two coefficients and. . we can write . two points (x 0 .b) X1 X J XQ Observe that i o (x o ) = 1 .61) holds. A polynomial of degree one (linear polynomial) has the form Pj(x) = a o + a lX (7. if x 0 .(xj) = 1 (7... . i o (x 1 ) = 0 i.. p n (x) exists for any function f.. the linear polynomial for the points (x 0 .
we can construct a polynomial of degree n. (xx^. (§. Construct a Lagrange polynomial that passes through the points 12. c 1 0 1 Here the nodes are 2. ^ and 4 andf(x 0 ). which passes through (n + 1) points.(x2.611a) ( x j. . A natural choice is X.610) is the Lagrange interpolating polynomial of degree n with i j (x) given by Equation (7.  ) and (4. (x n . (x l5 fj).613c) 033x21.612)  (x~xi)(xx2) .x j . }). .5 x + 1 ° (7 .6 6 7 i 2 (x) = ^ 12) [4 I 25) and = (7.69a. Since we have three points. b) and are of degree n.61 lb) i=o ( x j~ x i) Equation (7.x n) = 11/ 'I for each j = 0. Example 7.x l)( x o.x o)( x j.66b)...NUMERICAL METHODS 6/5 (0 *j(x k ) = J if j * k (7.. jr and jr respectively. (x) are quotients which have to satisfy Equations (7. .5x+1.2.x 2) " (22..x j + l)( x j. 1.1 .5) (24) " X " 6 . which passes through (n + 1) points [(x 0 . f(x^ andf(x2) are ~.1).1 0 .61... + f(x n ) i n (x) (7. p 2 (x) = f ( x o ) i o ( x ) + f(x 1 )i 1 (x) + f ( x 2 ) i 2 ( x ) Now » M (7.3 3 3 x 2 + 8 x . fn)] of the form p n (x) = f(x0) i 0 + f( Xl ) ij(x) + . .(x) = ( ^ ( x .5)(x4) _ 2 6 S x 10 (16 n .X jl)( x j.667 .b) \\ if j = k Similarly generalizing the concepts involved in Equation (7.Hxx^O (xxj J (7. f0).x l)( x j. ' ° ( X ) " rx o .69a.6 " 13a) ^ W = (2(5Z2)(2 5}4) = .61 lb). n (7. the polynomial will be of degree 2.610) The i .613b) (7.
05 (9)0.425 x + 1.0.b) (7.. b) and Equations (7. we have from Equation (7.15 An approximation of f (3) [= p 2 (3)] is p 2 (3) = 0. Xj] and is defined as f xi'xi L JJ = t T^= .x i) Similarly the nth divided difference is f rx I L x 0' x l ' x 1.x i) ( x j.617a.616) (7.618) The first divided difference of f with respect to Xj and Xj is denoted by f [XJ. f[Xi] = f(Xi) (7.15 = 0. T^ (7.425 (3)+1.05 x 2 . Newton's Divided Difference Representation Before we discuss this method we shall state some definitions and introduce the accompanying notation.' x nlJ  f t X l' "• ' X nl] ~ f l X 0' X 1 ' "• ' X nJ ix „\ n fi 2 0^ W.ozu.614 ADVANCED MATHEMATICS p 2 (x) = 0.617a.615a.620) f[*o xll = f{^[iXo) xi (7. The zeroth divided difference of the function f with respect to Xj is denoted by f [xj and is simply the value of f at Xj.325 We note that the three points lie on the curve f(x) = 1 and hence f(3) = 1 = 0.3333 (7. For specific values of Xj and x.619a.b) ( x j.614) (7.62U) ~xo .615a.b) shows that the value obtained by interpolation is correct to the first two decimal places.b) A comparison between Equations (7.
Xj. the divided difference can be tabulated as in Table 7. X k ] (zeroth difference) xo f Oo> (first difference) f( (second difference) . x 3 ] .f [ x 3 .7. (7. In Section 7. we shall approximate the derivative by the finite difference. f [Xj] f [Xj. We write . f(x 1 )f(x Q ) l Xl f( X l ) ° f^xjf^. x t ] f(x 3 )f(x 2 ) x 3 x 2 x3 f(x 3 ) f(x4)f(x3) x4x3 x4 f(x 4 ) X 3~ x l f[x 4 . f[x 0 . 22) l~x0 In the limit as xj tends to x 0 .xp] X2~xo f(x2)f(xl) X2~xl x2 f(x 2 ) f[x 3 . TABLE 7.61 Divided difference X. x 2 ] . x2] X 4~ X 2 This divided difference scheme lends itself to the construction of a polynomial through the points {XJ.621b) We recall that the definition of the derivative of f (x) at x 0 is f(Xo) = Um f N . f(Xi)}.62 lb) is the difference of the first difference and can be associated with the second derivative of f.9 1 1 V . Noting that the n th difference is the difference of the (n . The derivative can be discretized in terms of finite differences.61. Xj] f [Xi. xj] is the derivative of f at x Q .l) t h difference.f [ x 2 .f K ) xi > x0 x (76 .NUMERICAL METHODS f 615 n rY Y YI tLx0'xl'x2J _ f [ x l 'x 2 ] ~ f [ x 0 ' x l ] 7Z 7 1 v x 2~ x 0' f. f( . The second difference (Equation 7.
f (XJ)} as (0. .OZ8a. f (xj)}. (1.x0) (x2 . x 1? .624 to 28) can be generalized and an = f[x 0 . We note that using Equation (7. + f [ x 0 .x0) + a2 (x2 .D) (7.b) (7. an in Equation (7.625a. x 1 .. ..623) ai=^"f Similarly o ) =f[ x o' x l] (7. construct a divided difference table and then use formula (7.x1](x2xQ) a2 7Z VYhc V~\ vx2 ~xti> v x 2~ x l' ) Equations (7. x j ( x .)...x l ) + .630) + . (xxn.1)(xxn) and require this polynomial to pass through the points {xj. x n ] _ Ux0'Xl'X2J (7. Example 762..61) p o (x o ) = a0 = f(x 0 ) = f[x 0 ] Vx{\l) = ao + a 1 ( x 1 .. (2.x o ) ( x ..626a.624a. 1).x .x nl) + an(xx0)(xx. . . 2) and (4. we obtain Newton's forward divided difference formula which can be written as PnW = f [ x o] + f [ x O ' X l K x .b. ..b) r7 6?8 a h^ (/.+ a nl( x . ( x . we find a _ f(x2)f(xo)f[xo..x o ) + a 2 ( x .616 ADVANCED MATHEMATICS PnW = a0 + a l ( x ..x o ) ( x .624b).x o ) = f(xj) and via Equation (7.( x . 5). we get (7.. . a 1.x o)( x .x o ) + f[xO'xl'x2] (xxo)(xx1) (7...x j ) .c) (7.b) Xj .623). ! ) Note that it is easier to construct the polynomial by first constructing a table of divided differences. .624 to 27).629) Substituting the values of a0. Given the four values of {XJ.1) .xj) = f (x2) With the use of Equations (7.x l) .630) to find the third degree polynomial through the given points.X Q p 2 (x2) = a0 + ax (x2 .627a.
14017).l ) .05. 1.11803) and (1. a2 = l a 3=~^ (7. we construct the table of differences.62 Zeroth to third difference Zeroth difference 1 First difference Second difference Third difference Xj 0 ¥•» 1 1 21 21 ~ 2 2 52_3 42 2 4 5 2 I"0 ^ 1 ~° 2 1/6 . (1.623) yields p3(x) = l + l ( x .632b) .NUMERICAL METHODS 617 TABLE 7.631atod) Equation (7. (1.0 ) ( x .15.0 ) ( x . 1.62.25.07238). 1. (1.^ .12 .2 ) = X (_X3 + 9x 2 _ gx + 12) We next give an example to show the difference between the two methods discussed so far.1/2 i 40 ~~12 3/21 = ! 4~l " 6 Note that the points Xj are not equidistant. (7. given the following: {xj. ai=0. (1. Example 7.02470). (1.1.l ) ( x .09544).30. To apply Newton's divided difference formula.632a) (7. Use Newton's and Lagrange's formulas to evaluate V 1. 1. From Table 7. we obtain ao=l.20. 1. (1.04881). 1). f(Xj)} values where f(x) = Vx~. 1. ( x .63.
12) .15) (x .12 (7.10) (x .03) .02470 2411 54 5 1 1 4 4 1.03) (.20) (x1.630).02) .1.1) (x .07) + 5 (.04881 2357 50 2 3 2 1.02) (.05) (x .08) + 4 (.1 .12) (.1.13) (7.1.10) (x .30 Note: 1.1 (.05) (x .1 (x .09544 2259 45 1.1.07238 2307 48 1 3 1.20 1.10 1.634b) = 1.05) + 5 ( x .15) .05 1.1.11803 2214 1.12) (.1.07) (.05) (x . This is customary in writing tables of differences.12) .634a) = 1 + 2470 (.1) (x .618 ADVANCED MATHEMATICS TABLE 7.07) (. Using Equation (7.07) (.12) (.l ) ( x .25 1.1 (x .634c) .15 1.05735 (7.10) .15) (x .12) (.Vl.63 Divided differences for Vx X f(x) 1st 2«d 3rd 4th 5th gth 1 1 2470 59 1.1) (x . we have p 6 (x) = 1 + 2470 (x .1.1.1.1.07) (.1.20) + 4 (x .02) (.633) p 6 (1.1) (x . 0 5 ) (x1.08) (.03) (.1.14017 We have omitted the decimal point and the leading zeros.59 (x .1) .12) (.1 (.25) (7.10) (x .59 (.02) (.
20) = 1. 1.w> {1 ^ ^ (7. we have to decide at the outset of the degree of the polynomial to be considered. (xx Q )(x. We could consider all the given points but the calculations become lengthy.20. we only have to calculate the higher differences. The above conditions of continuity. The points closest to 1.051.121.051. 1.12 than Newton's formula. The points at which the low degree polynomials are joined are termed knots.15 and 1. .10)(U21.15)(1. (xxo)(xx2)(xx3) (x o xi)(x o *2)(*o. Using Lagrange's method. In the recent past. Using Newton's method. Spline Functions A high degree polynomial interpolation method is not satisfactory when the number of data points becomes too large. slope and curvature at the knots can be easily satisfied. piecewise polynomial approximations have become prominent. or when the data points are associated with a function whose derivatives are large or do not exist.121.10. we have to perform the calculations again. Such piecewise low degree polynomials are called splines.20) = (1.05830 n n _. We write n M = P 3 W (XXl)(xX2)(x~x3) f/x ^ . If subsequently we wish to improve the approximation by using a higher degree polynomial._nw (L°2470) + n .636b) We note that Lagrange' s formula produces a more accurate value for V1.x 3) (x^x^x^x/^ + (XXQ)(XXI)(XX3) f(x } + 2> (x 2 x 0 )(x 2 x 1 )(x 2 x 3 ) Thus n P 1O . although we have used fewer points in Lagrange's method.10)(1. it is more convenient to approximate the function by a piecewise polynomial function in a small interval.051.05.NUMERICAL METHODS 612 For Lagrange's formula.15)(1.12 are: 1. First degree polynomials (straight lines) are continuous but not smooth functions.635) 3(L12) (1. Instead of trying to approximate a function over the entire interval by a single polynomial of a high degree. we approximate Vx by a third degree polynomial involving four points. Cubic splines are the most commonly used splines. Second degree polynomials (parabolas) allow us to impose continuity and slope conditions at the knots but the curvature changes abruptly at the knots. A spline function is a function consisting of polynomial pieces joined together with certain smoothness conditions. since we cannot make use of the values already computed.X l )(xx 2 ) (x 3 x 0 )(x3x 1 )(x 3 x 2 ) 3l (7. Their first derivatives are discontinuous at the knots.
. i = 0..638a) (7. .. Thus over the whole interval (x 0 .638b) = xi+l~xi (7. ! i i i i i i . fj). Instead of using a high degree polynomial to pass through the (n + 1) points.61 We write Si(x) as Si(x) = aA + bi (x Xi ) Approximating f (x) by a set of cubic splines + q (x . 1. slope and curvature conditions. Each Sj(x) joins only two points Xj andx i + 1 . we choose a set of cubic splines Sj(x).1) (7. This is illustrated in Figure 7. x n ).. x^. (x n . VI y f (x) f^r^C.637.(x i+1 ) = S i + 1 (x i + 1 ) Combining Equations (7.. (xj. bj. f0). fn) be the given data points.. Continuity at the knots requires S.. the function is approximated by n cubic splines. 38a) yields aj + bihi + Cih?+ dihf = a i + 1 where hi (7. x i + 2 xn X FIGURE 7.61.638c) .62Q ADVANCED MATHEMATICS Let (x 0 . ! I i ' • I i i i i * • ! _ • l i i l i 0 X.637) where the coefficients aj.Xj)2 + di(x Xi )3. cj and dj are to be determined from the definition of cubic splines and the following continuity. (n .
639a) becomes ai = fj The continuity of the slopes at the knots imposes S(x i + 1 ) = S. 43a.NUMERICAL METHODS 62/ At the knots. Equation (7.. b) yields c0 = 0 Cni+Sd^h. Cj and dj).640a. . 121 and 144.642b) Equations (7.640b) (7. Use cubic splines to determine an interpolation formula for the function Vx . bj. 39b. Using Equation (7. the values of S.. Si(xi) = f. the conditions (7.639b) (7. We need to impose two additional conditions which are S"Q (x0) = 0 S.637. + 1 (x i + 1 ) On differentiating Sj and Sj + 1 . b) are applied at the interior points and the two end points (x 0 .640a) bj + 2 c i h i + 3djhf = b i + 1 Similarly the continuity of the curvature at the knots leads to s"(x i + 1 ) = S" +I (x i + 1) Ci + Sdjhj = c i + 1 (7. x n ) have to be excluded. 42a.643b) (7.643a) (7. 40b._i(x n ) = 0 Combining Equations (7.638b. (x) are equal to f (x).637). b) form a set of 4n equations to be solved for the 4n unknowns (aj.! = 0 (7.642a) (7. The values of the function are 10.64. We choose the three knots to be 100. we obtain from Equation (7. 11 and 12.641b) (7.640a) (7. Consider three points (knots).639a) The derivatives of f are not known. Example 7.641a) (7. 41b.
637) and are written as S 0 (x) = a 0 + b o (x .64%) (7.640b.c) (7.647b) The six unknowns (b 0 . a 2 = 12 (7.647a) (7.b) (7. b). b.643a. we have h 0 = 121 . we obtain 10 + 21b0 + (21)2 c 0 + (21)3 d0 = 11 11 + 23b! + (23)2 cx + (23)3 dx = 12 Similarly Equations (7. The solution is b 0 = 0.648b) (7. b. bj = 0.644b) From Equation (7. c o = O.647a. 48a.100) + co(x .100)3 SjCx) = zl+bx{x\2\) + cl{xl2\)2 + &x{x\2\f (7.100 = 21 hi = 144121 = 23 Equation (7. c 0 .622 ADVANCED MATHEMATICS S 0 (x) and Sj(x) are given by Equation (7. b) give respectively c0 = 0 cj + 63 dT = 0 (7.b) (7. cj. d0.0000022 (7.639b) yields a 0 = 10. 41b) yield respectively b 0 + 42 c 0 + 3 (21)2 d 0 = bj c o + 63d o = C! Finally Equations (7.b.644a) (7.645c.638b).0001 d! = 0.65 la.0486 . d0 = .0. 49a.0456 C] = 0.b) .d) From Equation (7.646a.638c).100)2 + do(x .650a.0000022 .645a.648a) (7.652a.649a) (7. dj) are to be determined from Equations (7.b) (7. a! = 11 . bj.
we use Sj. we use S o . For the interval 10 < x < 11. yj}. b.656a) (7.646a. b. b) into Equations (7. Suppose we have a set of n data points {XJ.654b) = Z(a i=l + bxiyi)2wi The best values of a and b imply that we choose a and b such that f(a. = 3b = ° Combining Equations (7. for each i). and a set of weights {Wj}. We want to obtain the best values of a and b to fit the data.b) = X efwj i=l (7. The deviation at each point is 8j = a + b x j .654a) (7.644a. The weights express our confidence in the accuracy of the points. If we think that they are all equally accurate.657b) If all the points are equally weighted (w. b. That is of of „ .653) f(a.656a.y i The weighted sum of the squares of the deviation is (7. b) can be rewritten respectively as n n n (7655a'b) (7. Substituting Equations (7. __ . 52a.656b) aX i=i n wi + b X xiwi = X viwi i=i n (7. we have . 55a. b) is a minimum. b. Least Squares Approximation The concept of least squares approximation was introduced in Chapter 2. we obtain S o and Sj. 51a. we set Wj = 1 for all i.NUMERICAL METHODS 623. b) yields 2 X ( a + bxiyi)wi = 0 2 X ( a + b x i . and for the interval 11 < x < 12.657a) i=i n a X xiwi i=i + b X x?wi i=i = X x iyi w i i=i (7. we start with the simplest case of a linear polynomial (y = ax + b).yi) x i w i = ° Equations (7. Here. = 1. b).654a. where n is greater than 2. We assume that the x's are exact. 50a._ . N r.
658a) 2 i=i xi + b Z x? = X xiYi i=i i=i (7. b)..66 la. we obtain 4a + 10b = 6.659a. .c) (7. ••• . 2..25 The solution of Equations (7.17..65.325 (7. x 2 .5).75 + 0.663) (7. .25 10a + 30b = 17. y is a linear function of r variables x 1. S x i = 10 ' X vi = 6 . i = 1. Find the best fitting regression line through the set of points (1. 1.e) X x? = 30 . 1. b) is a = 0...624 n n ADVANCED MATHEMATICS na+b^xj i=l n a = £ Yj i=i n n (7.. b).b) (7. which represents the best fitting regression line. 2). x~J . .658a. b = 0. .660a.25 (7. We have n=4 .75) and (4.658a. and obtain .660b) The equation for the regression is y = 0. xr and is written as y = b 0 + bi Xj + . n.. Example 7. 1).. (2. In this case.659d.25 Substituting these values in Equations (7.325x • The above method can be extended to multiple regression.662) We can then form the square of the deviations at each of the n points [xj .660a) (7.658b) Solving for a and b from Equations (7. X2. £ x{ y . (3. we obtain the desired linear expression.75. + b r x r (7.b.
. f (x 0 )] and [xj. 1. b) can be written as (7. + br X x® x® = i y. from which we can obtain the (r + 1) constants.7. b r such that e 2 is a minimum with respect to the constants b 0 ..b) b0 1 *? + b. .. b1... 7. we obtain p[(x) = (767) jj^ + <j^ (7. We refer the readers to Chapter 2 and Hildebrand (1956) for further details. x» i=l i=l i=l i=l (7..yj i= l (7. b r . . r. ..665a.. Recall that a polynomial of first degree passing through the points [x 0 . + brx(r° . .NUMERICAL METHODS 621 i (b0..665a.666) represent (r + 1) equations... bj. Equations (7.. we choose b 0 . . We approximate the function by a polynomial and we assume that the derivatives of the function are approximately equal to the derivatives of the polynomial..666) where Jt = 0. 1.. .. This implies that § £ = 2 l [ b 0 + b1x(. . b 1? . b 0 ...) + . f (xj)] is ^^rr^^TT^ Differentiating p}(x) with respect to x. The least squares approximation is not restricted to linear functions only. We can choose polynomials of degree higher than one... br) = £ [b0 + bjx^ + . X x» x® + . . r and x ^ = 1 for all i. .. we describe methods of calculating the derivatives of a function f using only its values at (n + 1) distinct points. b j . b r .. Equations (7.664) As in the case of one variable. orthogonal functions such as Legendre polynomials and trigonometric functions.7 NUMERICAL DIFFERENTIATION AND INTEGRATION Numerical Differentiation In this section.la) .+brx®yi][x®] = 0 where X = 0.
we obtain f'(x0)  f(X° + h)2hf(X0"h) (7 . 12).x 0 ) is the distance between the points xj and x 0 .211. Since Similarly if we approximate f(x) by a polynomial of second degree p2(x) and differentiate p 2 (x). x 0 + h).73b) behaves well in the interval. 3b). we find that the error is of the order h. The error can be reduced by taking h to be smaller.73a) may be written as (7.7 " 4) Equation (7. It is also referred to as the central difference formula.74) is known as the threepoint formula as it involves the three points (x 0 .h) and at (x 0 + h ) . provided f f is generally not known. (Xo) . since the derivative at x 0 depends on the value of f at (x 0 .7lb) where h (= xj . ffronoffro) (77. It is found to be (see Problem 16a) .72. From Equations (1. A fourpoint formula is obtained by considering a third degree polynomial.h.f " ( x o + 0h) where 0 < 9 < 1.^^^tfV+eh) By comparing Equations (7. Equation (7. A formula for f' (x 0 ) is then given approximately by f . as the derivative at x 0 depends on the forward point xj. (7.73a) f'^.72) is known as the twopoint formula since it involves two points XQ and x^.)f(x 0 ) h (7. It is also known as the forward difference formula.2) The formula given by Equation (7.626 ADVANCED MATHEMATICS = f(x. x 0 . we cannot estimate the error. we have 2 f(x o + h) = f(x o ) + hf'(x o ) +  .
We use Equation (7.3 f ( x o ) . • In Table 7.8 (7. t) relation for a moving body is given by (4. 1. we obtain f. we list the formulas giving the first and second derivatives of f.01).o) h2 + f(x 0 h) (? 7 6 ) Example 7.. The velocity v is given by ~ .f ( x o + 2h) + 6 f ( x o + h ) .99). 7e>0 5 004 .74).4 905 = 9. (5.72) to find v at various times t.j) v(1.03).9 (7.00) = v(1QI) = 49050014807 =9.77i.01) = 51O3O'Q2905 Comparing Equations (7.h) (7. h is 0. Find the velocity at each instant of time.9 o0 5i5pO4 = w Using the threepoint formula (Equation 7. v(0.00) = 5004024807 =9. we obtain v ( 1.71.77a. 1.7'7c to j).77g. The distancetime (s. (5.02) and (5. 1. we find that in this case the difference between using two or threepoint formulas is small.d) (77 .905. f(x0+h)2f(. . 0. Thus by differentiating p2(x) twice.h ) (7.85 = 9. (4. .b) (7.2 f ( x o .103.77c.75) Formulas for approximating higher derivatives of f (x) can be obtained in a similar way.204. We use f+k to denote f(x0 ±kh). In this case.01.00).71. 1.807.004.99) = v(1.NUMERICAL METHODS f'fT 627 ) .
1. Using the formulas in the order given in Table 7.78a.f) (7.7.686) and (2.5.166 + 8 x 6. (1. find the derivative of f at 1.b) (7.474 and not surprisingly the fourpoint formula gives the most accurate result. (1.1 .7) ..474) = 6.166).L.8 x 4.78c.686 . Given the values of [x k .(6. 5.7) . (1.3 x 5.628 ADVANCED MATHEMATICS TABLE 7.7) « J j (6.71 Formulas for f' and f f'(x 0 ) twopoint hzk h + 0 (h) f2 f"(x 0 ) ~ 2 f 1 ~ fo h2 + 0 (h) threepoint (central) threepoint fl"fl 2n + 0 (h2) fl2f0 + fl + 0 (h2) h2 + 0 h + 4fl~3fo 2h f2 (h2) V .490 f'(1. 8.6865.489 + 3.489) = 5. .166 + 4 x 6. f(x k )] to be (1. we obtain f'(1.06 f'(1.(8.489).3.669).7) is 5.d) (7.78e.475 (7. 6.h) We note that f(x) is e x and its derivative is also ex.J j (8.7. 0 ( h 4) f2+16^30^+16^^ 9 12h 2 19h i/n Example 7.669) = 5.474) = 5.72. ~h +4f2 ~5fl h2 +2fo + 0 U\ V ' / 4) fourpoint + 8fl8fl+f2 .474).7) . 3.390 f'(1.78g..71.9.686 . Thus f'(1.6864. 4.
As usual we denote the distance between (XJ. Unlike numerical differentiation.710a) . We need to resort to numerical integration when the integral cannot be evaluated exactly or f(x) is given only at a finite number of points. (7. we have I f(x)dx P 2 (x)dx Ja Ja (7. we approximate the function f (x) by a polynomial of second degree. These methods are called Ja numerical quadratures. numerical integration is a smooth operation and many adequate formulas exist. b by x 2 and we let xj be the midpoint of (x 0 .£ [ f ( x o ) + f(*i)] Equation (7.9a) »~nr ( Vxi)dx + ^ir ( ( x .65)]. . We denote a by x 0 .x o) dx (?79b> (7. The polynomial p 2 (x) may be written as [Equation (7.612) [ °' Let the limits of integration be a and b. A common practice in numerical methods is to approximate the function by a polynomial and then carry out the mathematical operation.79c) x0 x0 .612)] .79d) To improve on the Trapezoidal rule. we have [Xl I f (x) dx « p . _ (xXl)(xx2) P2W " v (Z 7XIZ TT x O~ x lM x o~ x 2^ (xxo)(xx2) (x X UX x \ *• U lxl ~ x oM x l ~ X2i (xxo)(xx1) (x Ux x 1 ^ ^ V x 2~x x oA x 2~ x lJ (7. x 0 ) and (x2.79d) is known as the Trapezoidal rule. xj) by h.f (x0) + ^ o Jx0 Jxo fXl \ f (Xi) dx (7 7.^ . If we approximate f (x) by a polynomial of first degree [Equation (7.NUMERICAL METHODS 629 Numerical Integration /•b We consider numerical methods of evaluating the integral I f (x) dx. x2). Approximating the integral of f (x) by the integral of p 2 (x).
We apply the formulas given earlier to each subinterval and the desired result is the sum of the integrals of each subinterval. If the interval [a. .711) In Equation (7.712b) rb The integral I f (x) dx can be written as Ja .710d) f (x) dx « ^ •'Xo [f (x0) + 3 f (Xl) + 3 f (x2) + f (x3)] (7. We denote the breakpoints by xj.630 ADVANCED MATHEMATICS J ' xo I(xx1)(xx9) (xxn)(xx9) „. This should be resisted. the formulas given so far will not generate good result. x 3 ).710d) is the well known Simpson's rule.10b) ~ ^ [ 2(xx1)(xx2)dx . . b] is large. we subdivide the interval [a. If we approximate f(x) by a third degree polynomial p 3 (x).. N (7. we write xj = a + ih. . 1.712a) (7.. The points Xj and x 2 are in the interval (x0.) f(X°)+ .^ f Vx o )(xx 2 )dx + ^ ) [ 2(x_x )(x_x ) d x 2h 2 Jx0 «  [ f ( x o ) + 4f(x1) + f(x2)] Formula (7. in the present section. b] into N subintervals of equal length h. Equally. . h = ^ i = 0. the limits of integration are (xQ. As pointed out in the previous section.711). we obtain the 3/8 rule which is given by 2 h 2 J Xo (7. for example) over the subintervals. x l5 x 2 and x 3 are equidistant points and the distance between two consecutive points is h. (xxft)(xx.1 f I (h)(2h) (h)('h) h 2 JXo f(x ' ) + (2h)(h) HdX <77 . We may be tempted to approximate f(x) by a high degree polynomial. it is better to divide the interval into subintervals and approximate the function by a low degree polynomial (cubic splines. that is to say. X3) such that the points x 0 .7lOc) (7.
79d)] to each subinterval [XJ.715a) can then be written as rb r NI N f(x)dx .715b) 5 L j=i j=i J Example 7.b. we need to have an odd number of points and an even number of intervals.h.714c) i=i r Ni « h.713) becomes N f(x)dx = X y i=J A fh Equation [f(*ii) + f(*i)] (7. x j . Similarly Simpson's composite rule can be deduced and is given by J J rb r f(x)dx .f(xo) + f(xn) + 2 X f ( x i ) z L i=i r NI . The values of f(x^) for various values of x k is given in Table 7.3 Trapezoidal rule and Simpson's rule to evaluate I f(x) dx. 1/2) L i=l i=l (7. .710d)].I NI N f(a) + f(b) + 2 X f(Xj) + 4 X f(Xi.714a) »frZ[f(Xi.J. (7. b] into 2N subintervals and each subinterval is of length h.f(a) + f(b) + 2 £ f(x 2j ) + 4 £ f(x2_l) (7.72.714b) (7.i) + f(Xi)] z (7. f(a) + f(b) + 2 X f(Xj) z (7. Equation (7.713) We now apply the Trapezoidal rule [Equation (7.715a) by Xj 1/2 • To apply Simpson's rule .714d) L i=i Equation (7. we need to evaluate the function at the midpoints which we have denoted in Equation (7. Use the rl. Thus we divide the interval [a.NUMERICAL METHODS 631 [b N fx' A f(x)dx = X f(x)dx i=l AM (7.715a) a We note that in Simpson's rule [Equation (7.714d) is referred to as the Trapezoidal composite rule.73.
30 1.05) + f(1.3 .{ f ( l ) + f(1.20)] + 4[f(1.04881 1.09544 1.718a) = f[x 3 / 2 ].05 1.72 is Vx and by direct integration.20) + f(1.717a) = 0.32149 (7..02470 1.{ f ( l ) + f(1. which is an odd number and we can directly apply Simpson's rule.32149 The function given in Table 7.30) + 2[f(1.1. h is 0.10) + f(1.72 Values of f(x k ) for various x^ xk f(xk) 1 1 1. Using Equation (7. 18c).716b.3 f(x)dx = ^ p .718b) (7.717b) I )\ f(x)dx = I J\ Vx~dx (7.714d).33147 (7. we have .3 = 0.718c) By comparing Equations (7.15) + f(1. we find that Simpson's rule yields the exact solution whereas the Trapezoidal rule is incorrect in the last decimal place.11803 1.10) + f(1. 17b.15) + f(1.15 1.25 1.25)]} (7.632 ADVANCED MATHEMATICS TABLE 7. • .10 1.716a) = 0.1.07238 1.715b).20 1.716b) We note that the number of points is seven.1. we obtain .3 (7. we have .25)]} (7.05.14017 In this case.30) + 2[f(1. From Equation (7.3 I f(x)dx = ^ .05) + f(1.1.
610) fh f(x) dx can then be approximated by The integral I /a .719) can then be written as I = Th + C h 2 where C is a constant and is assumed to be independent of h. we obtain an improved value of the integral which we denote by T ( 2 ) .. We denote the value obtained with a gap h/2 by T^^. h/2.NUMERICAL METHODS 633 The error involved in evaluating the integral depends on h.721) = ^(Th/2Th) 3h 2 (7722) Substituting C into Equation (7. This method of obtaining an improved value of the integral from two approximate values is known as the Romberg method (extrapolation method). we may write I = T h + O(h 2 ) (7.723) We can repeat this process of halving the interval.Thus I is given by I = T h / 2 + Ch 2 /4 From Equations (7. We recall that the Lagrange interpolation formula for the polynomial p n (x) can be written as PnW = X f ( x i ) 4 ( x ) i=0 (7. We now recalculate the integral using the same Trapezoidal rule but halving h. we obtain C (7. .719) We now assume that the error which is of O(h2) is proportional to h2..720. Thus T ( 2 ) (=1) is given by T(2) = T h/2 + ^ ( T h / 2 .721).. using intervals of h. h/4. It is widely used in computer programs.T h ) (7.720) (7.714d)] by T h . If we denote the rb exact value of I f(x) dx by I and the value obtained using the Trapezoidal formula [Equation (7. the length of the gap. Equation (7. that is to say. h/2 n successively until two improved values of the integral are within the desired degree of accuracy. 21).
= A 0 [a 0 + a 1 z 0 + a 2 z2 + a3z3] + A 1 [a 0 + a 1 z 1 + a 2 z2 + a3z3] (7. we transform the interval [a. b] to [. We choose Xj such that Equation (7.726) We consider the case where f(z) is approximated by p 3 (z).l ^) We now have to determine the points Zj which lie in the interval [1. 1].724C) (7.724d) Note that f(Xj) are the values of f at the points Xj and are constants. It can be written as p 3 (z) = ao + a 1 z + a 2 z 2 + a 3 z 3 We now choose z 0 .724d) is "exact" or "best" in a sense to be defined later. That is to say p 3 (z)dz = p 3 (z o )A o + p 3 (z 1 )A 1 (7. A o and Aj such that Equation (7.1. We now assume that the Xj are not given.724a) fb ~ Sf(xi) ^iWdx i=0 Ja « Z i=0 f ( x i) A i (7. For simplicity. The Aj 1= I /8j(x) dx are known as the weights. z 1.729) .634 ADVANCED MATHEMATICS C f(x)dx Ja f p n (x)dx Ja n [b = X f(x i )i i (x)dx i=0 Ja (7.727) Carrying out the required integration.724b) (7. They can therefore be taken I rb \ Ja. So far the points Xj are given and are equally spaced.726) is "exact".i f(z)dz . This can be achieved by writing _ _2x(a + b) (ba) {l. 1] such that c J. \ j out of the integral sign.728) (7.728) becomes 2ao+^2. Equation (7.X f ( z i ) A i i=o (7.
33).d) The first solution (Ao = 0) implies. z0 = °> zi=zo' zl=zo (7. 30c. This is illustrated in the next example. a1( a2 and a3 and.730c) (7.b.730b or d). A1 = l . Form Equation (7.735) Equation (7.730b) (7. a ls a2 and a3. z0.729) holds for all a0.730d)  = Aozo + A l z l 0 = A o zJ + AjzJ The set of Equations (7. we deduce A0 = A! Combining Equations (7.732a. from Equation (7.L r ) Ji \ V3 / W3 j (7.b. Of the four possible solutions. V3 z\=}= V3 (7.730a) (7.735) is the Gaussian two point formula and it involves evaluating the function at two points.726). the only valid solution is the last one.730b). Thus the number of evaluations of the integrand is the same as in the Trapezoidal rule. A1. that either A1 or z1 is zero and Equation (7. Zj) and we have exactly four equations to solve.734a .NUMERICAL METHODS 635 Equation (7.731) are A0 = 0. we obtain A0=l.733) Substituting Equations (7.730d) yields Aozo(zl"zo) = ° (7. .c. by comparing the coefficients of a0.731) The possible solutions of Equation (7. we have approximated the function f (x) by a third degree polynomial and we expect better accuracy. we obtain respectively 2 = A o +A! 0 = AQZQ+AJZJ (7.c.730c) cannot be satisfied.730a.d) (7. zo=L. Multiplying Equation (7.730a to d) involves four unknowns (Ao.d) into Equation (7. but in this case.734a.730b) by z \ and subtracting the resulting expression from Equation (7. we obtain 1 f(z)dZ = f( ^ r ) + f / .
32103 We now halve the gap and the interval is now 0.741b) The value we obtain for the integral using only two points is as accurate as Simpson's rule using seven points.b) Comparing T ' ' with the values obtained in Example 7.y + 1.3 I V7dx = 0.(1 + 1.07238 + 1.74la) (7. To use the Gaussian quadrature. we have To 15 (7.723) T ( 2 ) = 0.736a. Using Equation (7.l5z+ 1.73.735.79d) and the values given in Table 7. • . 1.14017) = 0.15 (7.40) yields /•I.72. we have x = O.74.15 dz (7.738a. though the number of points considered is less.y + 1.32149 / o . We use the trapezoidal rule to evaluate Tjj using h = 0.72. l 5 / . 1].b) = 0^5.15 I Vo.636 ADVANCED MATHEMATICS rl. From Equation (7.14017) = 0. we find the Romberg method gives an answer nearer to the exact values.b) The improved value T121 is given from Equation (7. From Equation (7. we have (7..7.32137 + 1(0.15 = 0.15 + .(l + 2 x 1.15z+1.739) I Vx~dx = 0.15.3] to [—1. Evaluate I Vx dx using the Romberg method and the Gaussian method. l 5 / . we have T 0 3 = 03.725).14d) and Table 7.740) Combining Equations (7.737a.32137 (7.32148 (7.3 r— Example 7.15 By direct substitution.3.32103) = 0. / o .321370. we have to change the interval from [1.
.82c.D.. . z (0) = 0.E.b) (7. whereas y" = f ( x . see Hilderbrand (1956) or Stroud and Secrest (1966).1).'s) with all conditions specified at one value of the independent variable is called an initial value problem. with the conditions specified at different points represents a boundary value problem.84a) .E. In this case. INITIAL VALUE PROBLEMS The order and the degree of an ordinary differential equation (O.D.726) is "exact" if f (x) is a polynomial of degree (2n .81) is a two point boundary value problem whereas if the conditions to be satisfied were y(a) = Y 0 . we have 2n unknowns (z 0 .d) it would be an initial value problem.. However if the conditions are specified at more than one point.82a.8 NUMERICAL SOLUTION OF ORDINARY DIFFERENTIAL EQUATIONS.E. y' (0) = 2. . y (b) = Y : (7.) have been discussed in Chapter 1.D. z" = z + 1 with y (0) = 1. . An O. which are discussed in Chapter 2.83a) (7.83c to f) is an initial value problem of a second order system of differential equations. z (0) = 1 0<x<l (7. For example. A o .D.1) has 2n constants which exactly corresponds to the number of unknowns. An_i) and we have to choose them such that Equation (7. For further details. In the case where time is the independent variable. Similarly y" = ( z 2 + l ) y . 1 . . The determination of Zj and Aj is simplified by considering Legendre polynomials..NUMERICAL METHODS 637 We can generalize the twopoint formula to an npoint formula. all conditions may be specified at t = 0. (or a system of O. y ' ) . a<x<b (7. 7.83b) (7. the differential equation a 2 y" + a 1 y l + a o y = g(x) subject to y (a) = Y o . This is possible because a polynomial of degree (2n . y . z n . the O.E. y'(a) = Y 2 (7.
c) is a boundary value problem (SturmLiouville problem)..85a)] into an integral equation. we obtain yOO = y o + I f(t.86) is known as an integral equation.88) .85a) subject to the initial condition given by Equation (7.85a) (7. b] I f (x.. z) I < L I y ...f (x. 2. In this section. we can determine y^ and subsequently y 2 .85b).86) by considering yk+1 to = y0 + I f & yfc) d t ' Ja k = 0. we seek a solution of Equation (7. (7. First Order Equations A first order initial value problem can be written as ^ = f(x.85b) y(a) = y 0 Formally integrating Equation (7. Equation (7.6M ADVANCED MATHEMATICS a 1 y(a) + a 2 y'(a) = c : .87) Since y 0 is given. The iteration will converge if (a) (b) f(x. bj y(b)+ b 2 y'(b) = c 2 (7.. only initial value problems will be considered. oo<y<oo there is a constant L.86) and we cannot evaluate such an integral.86) Note that the unknown variable y is inside the integral sign in Equation (7. . 220). such that for any two numbers y.221).84b. y) is continuous in the interval a < x < b . 1.z I Condition (7. .86. As suggested by the iteration given by Equation (7.y)dt Ja (7. y) (7. y) . There is a similarity between Equations (7. (7.88) is the Lipschitz condition. z and any x e [a. We have transformed the differential equation [Equation (7.
we obtain Yi+l = yi + hf(Xi.89) is Euler's formula. Equation (7. it is not suitable for numerical computation because it requires the evaluation of many integrals. which is of length h. b] into subintervals [xv xj +1 ] and assume that in each subinterval.89) yiL /yU) i 1—i 0 X. Figure 7. b) using Equation (7.85a).85a. y) is constant. Euler's method We subdivide the interval [a. Integrating Equation (7. f (x.. y i ) where y(x. • : approximate value of yi+l . Although it ensures that if f satisfies the conditions stated earlier a solution is generated. X FIGURE 7.81 Euler's method.87). Its accuracy is not very good.NUMERICAL METHODS 639 Picard's method involves solving Equations (7. ! ^ X+.81 illustrates this method. x : exact value of yj+x. Other numerical methods are available and are considered next. (7.) is denoted by y.
is given in the last column of Table 7. x 0 (= a) equals to 0.3 0. we generate Table 7.812) Differentiating Equation (7.1 0. accurate to three decimal places. . dx y(0)=l (7.811) The exact solution.0611 f(Xi. yj i 0 1 2 3 4 xj 0 0. y0) (7.81 Values of x { .x From Equation (7.81.640 ADVANCED MATHEMATICS Example 7.005 1.2 0.1. ^ = xy. Taylor's method We expand the function y (x) in f [x.046 1. yields . b) is y = exp(x 2 /2) (7.810a.01 1.b) dx dy dx .b) o ) 2 + . It can be seen that the numerical solution differs significantly from the exact solution. y (x)] in a Taylor series about the point x 0 and we obtain y(x) = y(x o ) + y I ( x o ) ( x ..85a). denoting ih by Xj and using Equation (7.0611 Exact y{ 1 1.813a.020 1.0302 1.4 yj 1 1 1.89).8.30906 y i+ i 1 1.81.1 0.81.b) We assume h to be 0.810a.813a) with respect to x.x o ) + ^ ^ ( x . y) and y' (x 0 ) = f (x 0 .yi) 0 0.4a.. TABLE 7. W = » t » * = * 3 r dx dy (7. Starting with y(0) (= y0) equals to 1. Solve the initial value problem using Euler's method.083 The exact solution of Equations (7.202 0. (7. we have y' (= ^ = f (X.0302 1.01 1.
821a.. + ^ r f ( p . y0) f(x 0 .y.822c.816f) (7.818a. y) We define T p (x. . we have f (x.d) .h) Example 7.816a to f).b) Using the notation given by Equations (7.NUMERICAL METHODS 641 y"(x 0 ) = f x (x 0 . we obtain from Taylor's formula y(xj) = yo + hT p (x o . h) In general Yn+1 = yn + hT p (x n . y) = y"(x) = 2y (1 + y2) (7.d) dx dy (7.h) = f ( 0 ) (x. y) f (1)(x. Use Taylor's method to solve y' = l + y 2 .1 ) ( x .b) (7.y) +  r f ( 2 ) ( x .y n ..819) (7.815) (7.b) (7.822a. y) y"(x) = — + f — = f(Vy) (7. y0) + fy (x0.816c. y) y (k+1) (x) = f°°(x. y ) + .y o . y ) Since y( X l ) = y(x o +h) = yo + hy'(x o ) +  r y " ( x o ) +  y y m ( x o ) + .y)+^f ( 1 ) (x.817) (7. y0) We introduce the following notation y'(x) = f (x.b) (7. y) = f (0) (x.816e) y m (x) = f (2) (x. .816a. y(0)=l (7.820) (7. y) = 1 + y2 = f (0)(x.82.
.n) Next we compute y2 = y 1 + h T 4 ( x 1 .y.822i) With h = 0. we have T4 (x0.22304 + 0. • Taylor's method has the disadvantage of having to differentiate f (x.1.22304 + 0. y) = y"' (x) = 2 ( 1 + y2) (1 + 3y2) f (3) (x.22304.2.1050218) + 4 ^ (27.23042 (7.( 1 5 8 .8221) (7.822p) + ^ .822e. For computational purposes. y0. y).k) = y0 + h T 4 (x 0 .22304.1(2.1. h) = 2 + . x 0 = 0.1) Since T 4 (0.h) = ( l + y 2 ) + (7. y 1 .05 (2) (2) + ^ and yi 2 (16) + 4 g L (80) . y) = y"" (x) = 8y (1+ y2) (2 + 3y2) T 4 (x.r) (7.t) (7. 1.1.822j. h) = 1+0.h) + ^2y(l+y2) + ?2(l+y2)(l+3y2)+^8y(lfy2)(2 3 y2 ) (7. y0.8543 y 2 . y o = l .391174) 6 (7. 0.1) = 2.822s. h ) = 1.4958 + .8543) « 1.f) (7.822m.642 ADVANCED MATHEMATICS f (2) (x. 1.2.822u) and the values accurate to five decimal places are y (0.05 (6.50847 Note that the exact solution of the initial value problem is y(x) = t a n ( x + j ) (7. 4 2 5 3 2 ) .22304 (7.1 (2.822o) (7. Heun's method is more economical. 0.1.50848.2) = 1.22305 and y (0.822g.1 T 4 (0.23042) « 1.822q.1) = 1.
823).083 0 0.yi) + f ( x i + 1 . y) (= y'.82 Values of xj and improved values of yj i 0 1 2 3 Xi f(xj. are given in Table 7. we obtain an improved value of yj + 1 .823) We determine y i+ j using Equation (7. We assume that yl+\ is given by y i+1 = y i + a k 1 + b k 2 (7. we find that Heun's method gives the exact solution to three decimal places.823). We derive the formula for the RungeKutta method of order two by generalizing Equation (7.81 and 2.3 Comparing the last column of Tables 7. yj) by the average value of f (x i? yt) and f (xj + i. Solve the initial value problem of Example 7. They are more accurate than Euler's method and do not involve finding derivatives as in Taylor's method. We can then calculate f (x. TABLE 7. Use the same value of h.82. Equation (7. It represents an improvement over Euler's method. to determine the value of y at x i + 1 . we only used the value of f (x.1 0.89) now becomes Yi+l = yi + [f(Xi.020 1. We can improve the accuracy of Euler's method by replacing f (xj.1 0.824) .89). yj) 0 0.3091 f(xi+1. yl+\). The values of yj.823).046 1. y i + 1 ) ] (7. Combining Table 7. RungeKutta methods The RungeKutta methods are extensions of Heun's method. we generate Table 7. y) at selected points on each subinterval.NUMERICAL METHODS 645 Heun's method In Euler's method [Equation (7.yi+1) 0.1 0.89)].81 via Heun's method. We only need to evaluate the function f (x.202 0. Vj+i).202 0. and finally using Equation (7. the slope of y) at the point X.81.4244 yi+1 1. obtained by using Euler's method.005 1.3091 0. +1 .81 and Equation (7.83.2 0. Example 7.
825c) = yi + hff4f«4ff> = yi + hfi + 1 + .825e) We have used the notation introduced in Equations (7. yj) yields f(xi + ah.824) yields .816a to d)..825c to e) are to be evaluated at (XJ.825a) (7.. yj + pkj) with a. yj). Expanding y i+1 [= y (x i+1 )] in a Taylor series about x i. ay/i 6 [ 8x 2 3x 9y 3yj J . ^ ax9yj.8. + a h . All the functions on the right side of Equations (7.825d) ^ ! L + f ^ +  f t + f (f] 2 ] ax3y ^ (  I + f I) 2\dx + ^^l + .826b) into Equation (7. Expanding f(x.. we have .. a. (3 as constants..yi) k2 = hf(xi + a h . (7. Substituting Equation (7. fi + ah(I) + Pk. +2 f 9y2 (7. 2 \dy2 j (7..825b) v i+1 = Yi + h Vi + y y" + \ y'i +  (7.. (I) + ^ ( ^ f ) +aphkl(^) + —L — +. (7.yl + pkl) .644 ADVANCED MATHEMATICS where k1=hf(xi.26b) P^ff^+.. We have substituted kj by hfj. (7. yj + (3 kj) about (XJ.826a) Wi + I 3y)i 2 ^X2J. b.
We can then determine y2.. y.yi) k 2 = h f (Xi + h/2 . yj + k3) (7. 2 3y 2 J i (7.829c. and is written as y i+1 = Yi + k k j + 2k2 + 2k3 + k4) o where k1=hf(xi. The most widely used formula is not of order two.825e.828a to c)] to solve for four unknowns. + kj IT) k3 = h f (Xj + h/2 . b.8.. we obtain Euler's formula. Other sets of values of a. then k2 and successively k3.827) i _  2 9x 2 dxdy We choose the constants a.828b) (7. yj + k 2 /2) k 4 = h f (xj + h . but of order four.8.30d) (7. Thus there is an infinite number of solutions to Equations (7.8. We may choose.825e) matches exactly Equation (7. a and P can be chosen.30c) (7.823) which is the formula for Heun's method.8. By comparing powers of h. we can calculate kj.830a) Knowing the initial values (xQ.829a to d) into Equation (7. we obtain (a + b) = 1 ba = 1 bp = 1 (7..b) (7.828a to c).30e) (7. y3..NUMERICAL METHODS 645 yi+1 = yi + (a + b)hfi + bh2(afl + e f  + b h 3 k ^ + a p f ^ 1 dx dy + ^f2^]+. .. for example.827) up to the terms of order h .829a.828a) (7. a and P such that Equation (7. It is clear that it is impossible to match Equations (7.30b) (7.824) results in Equation (7. y0). a = b = 1 a = p = 1 (7. yn. 27) for all f.828c) We have three equations [Equations (7. k 4 andyj. Note also that when a and b are set to one and zero respectively. b.d) Substituting Equations (7. .
00033 (7. k 4 and then y2 [= y (0.h) (7.83 lb) k 3 = 0.f) (7. assuming that somehow we can determine the values of f at some points in the interval.9f i3] ( 7 .y)dx (7.. we have yi = y(0. b] into subintervals [xj.0025)2] = 0.8 .831c.1.830a). AdamsBashforth method We divide the interval [a.83 le. which can be written as (see Problem 17b) yi+i = y i + ^ [ 55f i  5 9 f i .1) 2 + (0. XJ + J]. Equation (7.1 (^) +(°°Q 25 ) = 0. Take h to be 0.831g.00025 (7.86) becomes yi+l=yi+ hi f(x.i + 3 7 f i2 .831ij) k 4 = 0.831k) We now use (xj.0010 From Equation (7.1 (^J) 2 = 0. We can then integrate the interpolation polynomial and obtain the AdamsBashforth formula.0005 + 0.0005 + 0. The values of kj . Solve the initial value problem y' = x 2 + y 2 y(0) = 0 using the fourth order RungeKutta method. yj) to determine k 1. .001) = 0. .2)] and so on until we have values of y for the required interval.d) (7.646 ADVANCED MATHEMATICS Example 7.l[(0.00025 (7..IC4 are given by ki = x 2 + y2 = 0 k2 = 0.1) = 1(0.6. y) by an interpolation formula as discussed in Section 7.832) We now approximate f (x.83 la) (7.33 ) .84.
204990 1.148125 0.85.0 0. once the computation can be started. f2 and f3 so as to be able to calculate y 4 .J.1155 1.10000 0. we need to know f0. for example. The methods considered previously are selfstarting methods as they require only the values of (XJ. yj. This is the disadvantage of this method. Several multistep formulas have been proposed [see. The AdamsBashforth method requires only one computation of f per step. This is illustrated in Examples 7. yj using the RungeKutta method i xj kx k2 k3 k4 yi+1 (7. We use the RungeKutta method of order four to calculate y^ .11575 0.This type of formula is known as a multistep formula.20496 0.14896 0. Example 7. such as RungeKutta methods.1 0. TABLE 7. A common practice is to start with a RungeKutta method and once the values of f at a sufficient number of points have been generated. Gerald and Wheatley (1994)]. In Problem 17b.83 Values of Xj. f\. The calculations are given in Table 7.833). yj) to determine yj +1 . prior to applying Equation (7.166450 0.85 and 7. (xi2> Yi2) a n d ( x i3' Yi3).j). yi3).184740  0.4496 . we indicate the derivation of the AdamsMoulton formula. To get the method started.834a) (7.131575 0.16642 0.NUMERICAL METHODS 647 Note that in Equation (7.3 0. Vj_2. Take h to be 0. y^) to calculate y i + 1 but also (XJ.18566 0. to determine ( y ^ . as can be seen from Equation (7.2 0. Solve the initial value problem y' = 2x + y y(0) = 1 using the AdamsBashforth method. we need to use other methods. but it involves less computation than the RungeKutta methods of order four. a multistep methods is used.115000 0.V3.13155 0.833). To use the AdamsBashforth method. we not only need to know (xj.833).1.834b) 0 1 2 3 0.2642 1.83. whereas the RungeKutta method of order four requires four evaluations of f at each step.
9f0] (7. we have Y5 = Y4 + ^ [55f4 .39) .838a) (7.8395] = 1.1878 + 48. we have Y4 = y 3 + ^ [55f3 ..67359] (7.59f3 + 37f2 .835) becomes y 4 = 1.837a) (7.11. yi) = kj/h (7. Higher Order or Systems of First Order Equations Any r t h order differential equation can be written as a system of equations. f3 have previously been evaluated. since f ^ .836) Equation (7.9fj] (7.83. the third order differential equation r first order ^+y^i+h)2+y dx 3 can be written as dx 2 \dx / =g(x) (7.648 ADVANCED MATHEMATICS From Equation (7.. • We next extend the methods employed to solve first order equations to higher order equations.72898.891 +61. we need to perform only one calculation.4496 + ^ = 1.833). As stated earlier.f3. For example.838c) [112...5754. we obtain f0 . we need to evaluate only f4. at each step.837b) = 1.8..9462 We can similarly proceed to calculate yg. From Equation (7.835) From Table 7.147120.6754 Continuing.838b) (7.6754 +QL [136..830b). we have f(xj. Note that to determine y 5 .59f2 + 37fx .
. y j .844a) where k i = f (x . y { + k 2 / 2 ) k4 = h f ( x i + h.= fj(x. The RungeKutta method of order four [Equations (7. ) 1  v l (7. yj + k j / 2 ) k 3 = h f (xA + h/2. we consider systems of first order equations which are written as y. y l 5 .844e) . .c.844b) (7.840b to d) form a system of three first order equations. y . .841) can be written in vector form as y' = f (x.d) yi= y3 (=y") Y 3 = yiYsyfyi + s w Equations (7. .843) All the formulas considered earlier can be extended to the present case. ) . . (7.841) Equation (7. all conditions are given at x = x 0 and they can be written as y>o) = yo (7. y)  (7.b. 2 .NUMERICAL METHODS 649 Yi = y yi = y2(=y<) (7.840a.844c) (7.yi + k3) (7. Instead of considering higher order differential equations..842) Since we are dealing with initial value problems only.. j = l ..844d) k 2 = h f (xj + h/2. .830a to e)] becomes y i+1 = Y i + ^ (k1+2k2 — — O — — + 2k3+k4) — — (7. .
y u + ^ .846f.i b r k 1 1 + 2 k 1 2 + 2k 13 + k 1 4 " (7. we consider a second order differential equation or equivalently a system of two first order equations.86.650 ADVANCED MATHEMATICS To clarify the notation in Equations (7. The system of equations [Equation (7. y2i.(1 .i + k 23) (7.i + k 1 3 ' V2. Equations (7. then kj 2 .y^j. k21 = h f2(x{.i+l (7.ylii +1 " I y 2. kl2 = h h(xi+.i + k13' y2. kj 4 and finally Yj.846a) [ k 2 1 + 2k 2 2 + 2k 2 3 + k 2 4 where k n = hfjxj.846b.i+k23). i I + \ y2. y u .y^).845b) Ty 1 .842)] is yl = fj(x.844a to e) in component form become .846d. y'(0) = 0 (7.g) k14 =nfl(xi + h'yi.847b. k24 = h f 2 ( x i + h ' yi. + (7.847a to c) as a system of first order equations . kj 3 . y 1 > i + ^ . i+i • The index j in k j m is associated with the j in f. + ^ l ) (7.847a) (7. y l s y 2 ) y 2 = f2(x. y 1 ? y 2 ) We denote yi(xj) and y2(xj) by y H and y 2 i respectively.2). Solve the initial value problem y " . Example 7. y2>i+^l). and the m is associated with the k m in Equations (7.845a) (7. y u + ^ .y2) y' + y = 0 y(0) = 1 .c) We rewrite Equations (7. k 23 = h f2(x1+^.e) k yu+klL k n = h f!(xi + ^.c) .^) (7.i) The order of computation is to calculate kjj (j = 1.846h.844a to e). y2>i+^).830a to e). y^. y 2i ) k 2 2 = h f2 ( x .
846a) yields ~yi.98 0. They studied the phenomenon of electric signaling by individual nerve cells and were awarded a Nobel prize for their work.848c) (7.850a.02.h) k 1 4 = 0.0.0. identifying fi = y2' k n = 0 .98.0.982) ( .f) k 2 4 = 0.87. k 1 2 = 0.0.1) .2[(ll)(^]l] = 0.O) = 1 y 2 ( o )(=y 2) o) = ° We choose h to be 0.1) = 0.0.199 (7.992) ( .2 and we compute kj m . 2 ( 0 ) = 0.2 [(10.99] = .H968" (7.1984) = . k 1 3 = 0.852a.850g.848b) (7. Substituting Equations (7.i] V 2.851b) The solution is y(0. f2 (7.1976 (7. y'(0..0.848a) (7. Solve the set of equations proposed by Hodgkin and Huxley (1952).2 (0.848d) (7. When activated by a .2 ( .y ? ) y 2 .850c.199 r0.2) = 0.b) (7.°^) = .850e.s49a.2 k 2 3 = 0.2[(l)]=0.b) (7.2 [(10.850a to h) into Equation (7.2) = 0.2 (0.848e) = ( 1 yf)y 2 yi (7.b) Example 7. In a state of rest.0.02 .98] = .1984 (7.03968 .d) k 21 =0.NUMERICAL METHODS 651 yi = y yj = y 2 y'2 = ( 1 .1984) . a nerve cell has a resting potential difference and this potential difference is attributed to the unequal distribution of ions across its membrane.1944 (7.85 la) 1.l rii 0 + 1 6 0.y i yi(O)(=y1.0.2 k 2 2 = 0.
855a.853c) (7. 2 . The permeabilities Pj^a a n d P K are assumed to be given by PNa = g N a Y 2 Y 3 .1] .d) Equations (7. b 2 = 4exp (Y 4 /18) . Y3 represent the mole fractions of unspecified chemicals that regulate ionic permeability.^ a^lY^b^! = a2(lY2)b2Y2 a3(lY3)b3Y3 = ~ P N a <Y4 .652 ADVANCED MATHEMATICS current. Y 2 = 0. For computational purposes. a ^ O .b) (7.1 0 .507 (7. ( i = l .853d) ^ . V K . PK = gKYl (7. b . Y 4 is the potential difference across the membrane.1Y 4 +l)1].318 . V K = 12.1 (Y 4 +25)/[exp (0.07exp(Y 4 /20). V L are the permeability coefficients and resting potentials for Na + . = 0.596 .V Na ) . Y 3 = 0. P N a . a 2 = 0.855c.6.P K (Y4 .= C^ where C is the capacitance.V K ) .1Y 4 + 8) + l]. Hodgkin and Huxley (1952) assumed that the mechanism that determines ionic fluxes across the membrane can be described by a set of first order equations where the rate constants depend on the voltage.V L ) (7.5) . This change in potential can be measured. g K = 36. A multistep formula (AdamsMoulton) is then applied to obtain . The coefficients aj and b. The equations they proposed are ^L= . VL = .853a to d) are solved numerically. Based on experimental observations. VNa = .854a. K + . P L . 3) are empirical constants and are functions of Y 4 .1 1 5 .125 exp (Y 4 /80).853a) (7. O l (Y 4 +10)/[exp(0. a 3 = 0. V N a . =0. Y 2 . The initial conditions are assumed to be Y.P L (Y4 . 6 . P L = 0.553b) (7.b) where g N a and g K are the conductance for Na+ and K+ respectively.053 Y 4 = 6. A RungeKutta formula is used to generate values of Yj at a sufficient number of points.1 Y 4 +2. the permeability properties of the membrane change resulting in an alteration of the potential difference across the membrane. and another unspecified ion respectively. Y j . b 3 = l/[exp(0. P K . we have set gNa=120.
Y2. Y4+D14/2) D24=TAU*F4(Yl+Dll/2.Y3PZ.0 WRITE(6.Y4PZ REALD11.Y4+D34) D43=TAU*F3(Y3+D33.0005 T=0.D14.TAU INTEGER I.Y4+D24/2) D33=TAU*F3(Y3+D23/2.Y4+D14/2) D22=TAU*F2(Y2+D12/2.82.Y4+D24/2) D32=TAU*F2(Y2+D22/2.87 REAL Y1.D42.Y4P REAL YIPZ.D33.Y4) D14=TAU*F4(Y1.Y4 1=1 C STARTUP WITH RUNGEKUTTA 44 D11=TAU*F1(Y1.Y4+D34) D44=TAU*F4(Y1+D31 .Y4) D12=TAU*F2(Y2.Y4+D14/2) D23=T AU*F3( Y3+D13/2.Y4+D34) .Y3+D13/2.Y1P.D41. The results are shown in Figure 7.Y3P.Y4K.Y2+D12/2.Y2P.D34.NUMERICAL METHODS .J.Y4+D24/2) D41=TAU*F1(Y1+D31 .Y3.R.Y4+D14/2) D3 l=TAU*Fl(Yl+D21/2.Y34+D34) D42=TAU*F2(Y2+D32.Y3+D33. The following FORTRAN program is used to compute Yj.Y3+D23/2.STATUS='NEW) C INITIALIZATION OF THE VARIABLES Yl=0.D43.D44 REAL T.D23.D13.053 Y3=0.Y2K.D21.Y2+D32.*)T.D22.D24.Y2+D22/2.Y4) D13=TAU*F3(Y3. C PROGRAM TO SOLVE EXAMPLE 7. 653 values of Y.318 Y2=0.Y3K.Y4+D24/2) D34=TAU*F4(Yl+D21/2.D31.Y3. up to t = 20.D32.K OPEN(UNIT=1 .D12.DAT' .Y4.FILE='A:/RESULT.Y2PZ.596 Y4=6.Y4) D21=TAU*F1(Y1+D1 l/2.507 TAU=0.Y1K.Y2.
Y2P.GT.Y2K.Y4K)F3(Y3.20.TAU Y1P=Y1K Y2P=Y2K Y3P=Y3K Y4P=Y4K R=1.*)T.Y3PZ.3X.Y4P FORMAT(F9.Y2PZ.Y4P)+8*F4(Y1K.Y4K)EF4(Y1.EQ.Y4PZ WRITE(6.Y4P)+8*F1(Y1K.Y3P.Y4P)+8*F3(Y3K.6.Y4)) Y2PZ=Y2K+TAU/12*(5*F2(Y2P.654 ADVANCED MATHEMATICS Y1K=Y1+(D1 l+2*D21+2*D31+D41)/6 Y2K=Y2+(D 12+2*D22+2*D32+D42)/6 Y3K=Y3+(D13+2*D23+2*D33+D43)/6 Y4K=Y4+(D 14+2*D24+2*D34+D44)/6 C ADAMSMOULTON U=2*TAU DO 10 T=U.F9.Y4K)F1(Y1.F9.Y4)) Y3PZ=Y3K+TAU/12*(5*F3(Y3P.lE5) Y1PZ=Y1K+TAU/12*(5*F1(Y1P.1OOO)Y1P.6.6) Y1P=Y1PZ Y2P=Y2PZ Y3P=Y3PZ Y4P=Y4PZ ENDWHILE J=I/10 IF(J.3X.1OOO)Y1PZ.6.Y2P.0 WHILE(R.Y3P.Y4P)+8*F2(Y2K.1)THEN WRITE(1.Y4K)F2(Y2.Y4)) R=(Y1PZY1P)**2+(Y2PZY2P)**2+(Y3PZY3P)**2+(Y4PZY4P)**2 R=SQRT(R) C C 1000 WRITE(6.Y4)) Y4PZ=Y4K+TAU/12*(5*F4(Y1P.3X.F9.Y4P 1=0 ENDIF 1=1+1 .Y2.Y3.Y3K.
Y A2=0.1*Y+l)l) Bl=0.01*(Y+10)/(EXP(0.B3.B2.07*EXP(Y/20) B3=l/(EXP(0. Y.Z) REAL W.125*EXP(Y/80) F1=A1*(1X)B1*X RETURN END FUNCTION F2(X.NUMERICAL METHODS 6.Y Al=0.B1.1*Y+3)+l) F3=A3*(1X)B3*X RETURN END FUNCTION F4(W.3*(Z+10.Y) REAL A2.X. F4=36*W**4*(Z12)0.5)l) B2=4*EXP(Y/18) F2=A2*(1X)B2*X RETURN END FUNCTION F3(X.1 *Y+2. X.X.Y A3=0.Z.6)120*X**3*Y*(Z+115) RETURN END .1 *(Y+25)/(EXP(0.Y) REAL A3.Y) REALA1.X.55 Y1=Y1K Y2=Y2K Y3=Y3K Y4=Y4K Y1K=Y1P Y2K=Y2P Y3K=Y3P Y4K=Y4P 10 CONTINUE STOP END FUNCTION F1(X.X.Y.
0 6 0.656 ADVANCED MATHEMATICS Y.5 . k is the surface curvature and Ca is the capillary number. For small deformations.7  Y2 Y 3 ^ . > .82 Plot of Yj . (1996) studied the deformation of nylon drops in an extensional flow of a polyethylene melt. Y 2 . {(>). r) d are the viscosities of the matrix and dispersed phase respectively. k is approximately given by k= !_2£_J[_J—iLe^i)] R R2 R 2 LsinB d 8 V d6/J (7._5 0 . Compute the shape of a drop in an extensional flow of a viscoelastic fluid. 0.. 0. The drop is assumed to be spherical of radius R prior to entering the extensional flow.~ ^ ^ ^ . nd are the stress fields associated with the matrix (polyethylene) and the dispersed phase (nylon) respectively. 10 Time 15 ___10 20"15 FIGURE 7...K . Choosing the centre of the drop as the origin of a spherical polar coordinate system (r. j .\ . 2^ °0 \ 5 . GonzalezNunez et al.858) . < ^ >" > 0. 10 > • .n ^ ^ 0. the radial distance r of any point on the surface of the deformed drop in the extensional flow can be written as r = R+^ We assume the velocity field to be continuous across the boundary and a force balance yields (7. n is the unit outward normal to the surface.i " " . T)m.J o. Y 3 and Y 4 versus time Example 7..\ \ _ r^xr~~~—5 / / ^ ^ .4 "•. L*.8 " 57) where nm..856) fem(W)«d]*5 = kCaE (7 . A ' .88.3 "•"" "I .
Z2(5000).R.R0 COMMON R. + cot e — + 2b. and k.859) Equations (7.XK.57D0 (7. C PROGRAM TO SOLVE EXAMPLE 7.88 USING RK22 IMPLICIT NONE INTEGER N.r\d/r\m) T o exp [(TQ/L) (R+^) cosG] = 0 d0 d0 Assuming the deformed drop is symmetrical.TEST 1 .TEST2.NUMERICAL METHODS 657 The stress fields are calculated by assuming that the polymer melts can be modeled as upper convected Maxwell fluids (Carreau et al.TAU2. — =0.TO C DATA R=3. b) are solved numerically using a RungeKutta formula. . we obtain the following equation for ^ — .FUNC2.TEMP.2R .D0/(4.J REAL*8PHIl(5000).XK.Zl(500O).RR(50OO).R 2 Ca (1 .B ' ) / ( L ' + B') Figure 7.D.X. Substituting the values of nm.10 CA=0.I. the two appropriate boundary conditions are % = Ro .860a.01D0*CA*TO R0=l.D0*CA) C INITIAL VALUES PHI1(1)=ROR PHI2(l)=0.00157 TAU2=TAU/2.1D0*J TO=2..TO.83 illustrates D' as a function of Ca for various values of R. Kd. 60a. d0 at 0 = 71/2 (7.TE(50OO) REAL*8 TAU.D0 TAU=0.967D0 XK=R*R*0.D0 N=1000 X=1. 1997). The deformation D1 is given by D1 = ( L ' .b) (7.859.PHI2(5000).861) . Having found £.CA. the major (L1) and the minor (B1) axes of the drop can be calculated.35D0 DO 1 J=l.
*)R.100 D=(Z1 (I)R0)/(Zl(I)+R0) CONTINUE WRITE(6.Y2.*) CA.X DO 110I=l.1E.1.TEST2.N+l.X) IMPLICIT NONE RERAL*8 Y1.658 ADVANCED MATHEMATICS C START INTEGRATION DO99I=1.C2.D0 FUNC2=2*C1(COS(X)/SIN(X))*Y22*Y1C2*EXP(C3*(C1+Y1)* 1 COS(X)/24.C3 COMMON C1.C2.X)) IF(ABS(PHI1(I+1)).N+1 TEST 1 =PHI 1 (I)+TAU*PHI2(I) TEMP=FUNC2(PHI1(I).PHI2(I).Y2.D CONTINUE STOP END FUNCTION REAL*8 FUNCTION FUNC2(Y1.0D4)GO TO 100 RR(I)=R+CA*PHI1(I) TE(I)=X Z1(I)=RR(I)*COS(X) Z2(I)=RR(I)*SIN(X) X=XTAU CONTINUE RESULTS WRITE(6.C3 FUNC2=0.X) TEST2=PHI2(I)+TAU*TEMP PHI1(I+1)=PHI1(I)+TAU2*(PHI2(I)+TEST2) PHI2(I+1)=PHI2(I)+TAU2*(TEMP+FUNC2(TEST1.C1.D0) RETURN END 99 C 100 110 1 C .X.
NUMERICAL METHODS 659 0. dx 2 dx y(a) = A .8 1 FIGURE 7. the conditions here are given at more than one point and the formulas given in Section 7. Shooting Method This method is widely used.9 BOUNDARY VALUE PROBLEMS Boundary value problems have been discussed in Chapter 2.9lb.8 cannot be applied. We assume that . We described two methods which can be used to solve boundary value problems.9lc) and replacing it by imposing a condition at x = a.2 0.c) We convert the boundary value problem to an initial value problem by dropping the boundary condition given by Equation (7.83 Deformation of D' versus capillary number Ca for various values of radius R 7.8 • * * • * # • * R=5 • R:7 a 0.+ p ( x ) ^ +q(x)y = r(x).9la) (7. We consider a second order linear equation ^ .4 Ca 0. It consists of transforming the boundary value problem to an initial value problem.4 * ^ .6 0. y(b) = B a<x<b (7. I—— °0 0. Unlike initial value problems.
it is unlikely that the boundary condition given by Equation (7.9la to c). we can compute yj and y 2 as in the linear case and we use Equation (7.y 2 ( b ) (7.9lc) will be satisfied. Since oci is a guessed value of y'(a). With y'(a) given by Equation (7.94.9la). and the value of y'(a) that will satisfy Equation (7.9lc) will still not be satisfied.y > ( a ) ] < 7 97) In Equation (7. ( a ) + S^l k < a ) . all the terms on the right side are known.93) where c 1 and c 2 are constants.9lb) and y must also satisfy this condition. from these two values of a (ocj and a 2) it is possible to obtain the exact value of y'(a). we obtain y ' w = y .91. 2) form an initial value problem and we can integrate the differential equation using one of the formulas given in the previous section.95) (7. 3) yields 1 = cx + c 2 Imposing condition (7. The above method is applicable only for linear equations. Let y 1 be the solution obtained by using (Xj and y 2 be the solution with a 2 . Combining Equations (7.9lc). Equation (7. That is to say y = c1y1+c2y2 (7. we can compute y 3 .93) is not valid. is a solution of Equation (7. 5).9lc) is known.9la.97).92) Equations (7. a linear combination of yj and y 2 is also a solution of Equation (7. we cannot apply the principle of superposition and Equation (7. Unless we are lucky. In the linear case.97).93) and using Equations (7. For nonlinear equations. we compute the solution y that satisfies Equations (7.660 ADVANCED MATHEMATICS y'(a) = (*! (7.9la).94) Cl (79"6a'b) Differentiating y in Equation (7.96a.91 b. Both y 1 andy 2 satisfy Equation (7. However. we have B = c i y i ( b ) + c 2 y 2 (b) From Equations (7. b). Since the system is linear. b.97) to obtain an improved value of y'(a) which we denote as y3(a). We now guess another value of y'(a) (say a2) and we proceed with the integration. we obtain _ " By2(b) yi(b)y2(b) ' _ Yi(b)B °2 " y l ( b ) . as illustrated in Figure 7. Note that y3(a) is not the exact value of y'( a ).Using y 3 (a).
. Equation (7.97)..x / 5 ) y y(l) = 2.910b. We need to determine a such that Equation (7.98) can also be obtained as follows. y ' V2lb)" v*V ' V(b) " / / ^ * 1 i i i i [ J i . b) can be solved by the secant method [Equation (7.98) Equation (7.l (7.91.99a.99a. we obtain Shooting method yi+2(a) = y.b) Equations (7.98). Example 7. The solution y is a function of y'(a) (= a) and the solution can be written as y(oc. x).NUMERICAL METHODS 667 in the same way as yj and y2 were computed. (7. I 0 1 o 1 b x ^ FIGURE 7. y(3) = . We repeat this process until successive solutions yj and y i+1 are within the required degree of accuracy.91 c) is satisfied. Solve the boundary value problem y" = x + ( l .26)] and the result is Equation (7.c) .B = 0 (7.9lc) can be written as g(cc) = y(cc. ^ ^ l b ) k i ( a ) y[ (1 i = L 2.91 On generalizing Equation (7.910a) (7. 1 ) . (a) + y ..
Next.000 1.912) (7911) y(1. Equation (7. x) for various values of a and x are given in Table 7.99a).000 1.908 1. x) 2.811. b).348 0.328 0.8 2.99a.305 0.6 1.561 1. b) can also be solved by Newton's method.662 ADVANCED MATHEMATICS We start the process of integration with the guessed value of y'(l) (= o^) to be . 3) which is 0.0 and we obtain y^ (3.803 2.V ^ g (ock) where the prime denotes differentiation with respect to a. from Equation (7.751 1.910c).5.8 3.128 3. The resulting yj (1.1 . 5 .0 • Equations (7.5. we take y'(l) (= a 2 ) to be 3.97).787 0. we obtain «k+l = « k .175 0.013 0. TABLE 7.4 1.91. Then using Equation (7. Values of y (a.2 2.453.911).712 0.6 2.0.5.91 Values of x 1. 3) satisfies Equation (7. Combining Equations (7. we obtain the value of y' (1) which is . 3) is 4. g'(ock) is given.625 1.0 2.045 0.619 0.453 y(3.910a) is written as a set of first order equations and the integration formula used is that of Heun [Equation (7.880 4.811 y ( a . 99a.605 1. x) y(3.104 0.026 1.5.991 0. x) 2.3.007 0.118 0. we need to determine g'(ock) and since B is constant.443 0. and y (3.211. 1) (7. To apply Equation (7.000 . x) 2.5.542 3.000 1.4 2.0 1.2 1.000 1.105 2.823)].499 0. by g'(cck) = y ' ( a k .
we need to solve an initial value problem.911).y.^) dx 2 x dx/ (7.913c) by the initial condition Pdx a =a (7. b.915b. We need to obtain a good approximation to a via Equation (7.NUMERICAL METHODS 661 To obtain g'(ock). b) and then g'(a) is obtained.911) to obtain an improved value of a. we obtain £'(a. a) = 0 £ ' ( a . d) with respect to a yields £(cc. The iteration is repeated until two iterates yield values within the desired degree of accuracy. a) = 1 (7. 16a. b) can be solved and ^ ( a . d) define an initial value problem. b. which in turn implies finding g'(a k ).916a) (7. From Equation (7.913a.916b) The initial value system defined by Equations (7. d). Then on differentiating £.914) The prime denotes differentiation with respect to x and this notation will be kept from now on.914) yields ^'(a. we note that g'(a) is £(a.913d) Equations (7.c) We replace Equation (7.x) = ~[y'(a. b) is obtained. Equations (7. y(b) = B (7.by ^(oc. This can be substituted in Equation (7. We demonstrate this by considering the boundary value problem given by 4=f(x.x) = ^ ^ dy da = ^ dy + + i L 3Z1 dy' da ' (7.913a) with respect to a and using Equation (7. We denote ^ .915b.913.915a) ^ dy (7. with respect to x and interchanging the order of differentiation. Differentiating Equation (7. x). b) are the associated variational equations which need to be solved simultaneously with Equations (7. 16a.913b.912).913a.) y(a) = A . .x)] (7.915b) Similarly differentiating Equations (7.913b.
18c.664 ADVANCED MATHEMATICS Example 7. Use the shooting method to solve the nonlinear boundary value problem .918d) We denote y by y\ and ^ by y3.y \ y'3 = Y 4 (7.918c) (7. 1<X<1 (7.917c) has to be replaced by the initial condition y'(l) = a Equations (7.2 y" = ^  y..920) .l ) = (e+e1) (7.918b) (7.l ) = (e+e"1) y(l) = (e+e1) The associated variational equations are = .917c) y ( .917b) (7. 18a) can now be written as a system of four first order equations as follows yl = Y2 yi = ^ .917a. 18d.917c.917a) (7.92.Equations (7.2 ^ ) £(1) = 0 £'(1) = 1 +1 ^ + ^ ' (7.921a) (7.919c) y4 = ~r(yl) + 1 J y 3 + 4 (yr) y 4 (78~19d) Equation (7.919a) (7. 20) are written as y i ( .919b) (7.
9la). b) Y xj = (7923) Substituting Equations (7. We denote a by xQ and b by x n + 1 .917c). 23) into Equation (7.919a to d. Further. The value of a 0 chosen is 0. 22a . b] into (n + 1) equal intervals.922b.y j . We again consider the boundary value problem given by Equations (7. the number of initial conditions to be guessed is also high.9la to c). Finite Difference Method In transforming a twopoint boundary value problem to an initial value problem. We use the central difference scheme and the derivatives at x. each of length h. (= x0 + jh) are given by d y  dx d*y dx2 xj = ^lHM=aiCil 2h 2h VjHl^j + yjl h2 (7 . • A method widely used to solve boundary value problems is the method of finite differences which is considered next.324. The shooting method can then be very laborious. We divide the interval [a. if we did not make a good guess of the initial conditions. If the order of the differential equation is high. In this method. which is close enough to the value given by Equation (7.2 and after six iterations the value of y (1) was found to converge to 0. An alternative numerical method of solving boundary value problems is the method of finite differences.921c) (7.921b) (7. we have replaced the boundary condition(s) at one end by guessed initial condition(s) at the other end. 1 )] h + q(xj)yj = r(Xj) (7. 9 .2 y j + yj.1] + 2t[p(x j )(y j + 1 . we obtain ^ [ y j + l .92 Id) The initial value problem given by Equations (7. we replace the derivatives by finite differences. the convergence might be slow. 21 a to d) can be solved using the RungeKutta method with an initial value of a (= a 0 ).924) .NUMERICAL METHODS ^ 5 y2(l) = « y3(0 = 0 y4(!) = 1 (7.
925b) (7.928a) Yll y = ! s = f s ll : = h2 [ r ( x l)l [ a i A 0  0 (7.926b) Substituting Equations (7.9lb.666 ADVANCED MATHEMATICS On multiplying throughout by h 2 and collecting like terms together. Equation (7.d) _ynj LSnJ Lr(Xn)J LcnB^ .925d) aj = 1 ~2P(XJ) bj = . we have a1A + b 1 y 1 + c 1 y 2 = h 2 r(x x ) a 2 yj + b 2 y 2 + c 2 y 3 = h r(x 2 ) (7. b) into Equation (7.2 + h2q(Xj) Cj = l +  p ( X j ) The boundary conditions (Equations 7. and writing out the first two and the last two lines of the latter.926a. c) are now written as y0 = A y n+ l = B (7.925c) (7.925a).927a) (7.928b.927b) VlYn2 + b niy n l + C niy n = h^nl) {1.925a) (7.924) becomes a j yjl + b j y j + c j yj+l = h 2 r ( x P (7.921d) W l + V n + Cn6 = *»2r(xj Equation (7.927a to d) can be written in matrix form as Ay = s (7.921c) (J.926a) (7.c.
The exact solution of the boundary value problem is y = sinhx The finite difference equation [Equation (7.4). bo A = • 0 • • : . x1? x2. Equation (7.(2 + h2) y j + y j+1 = 0 (7. Solve the boundary value problem d2y . we divide the interval into four equal intervals and label the points x 0 (= 0)...930) Since h is 0.93.931) can be written as .! .NUMERICAL METHODS t>l Cj b2 ^ _ ^ 0 C2 6& 0 .. ••• 0 0 0 0 0 0 a.931) (7.928a)].c) y(0) = 0 . by the method of finite difference.929a) (7.. 0 0 0 h 0 0 Co .925a)] is yjl . .928a) can be solved by the method of Gaussian elimination (discussed in Section 7. The unknown values of y at intermediate points (y 1 ..929b. Equation (7. In matrix form [Equation (7.y dx 2 = 0 y ( l ) = sinh 1 (7. 0 an bn 0 The matrix A is a tridiagonal matrix whose elements are known...^ .25. : : : (7.. x 3 and x 4 (= 1). yn) can thus be determined.. Example 7..928e) o 0 o 0 o 0 •• V i V i cn.. Choose h to be 0.25.
2. Equation (7. Catalytic monoliths are widely used in the automobile industry to control the emissions from the vehicle exhaust systems.94.2528 0. n) resulting in a set of nonlinear algebraic (or transcendental) equations.922b. ) (7. Thus by discretizing Equation (7. This set of equations are solved by iteration techniques. b).2 and 7. ADVANCED MATHEMATICS (2+h2) 1 1 (2+h2) 0 1 ?1 Y2 = ° 0 (7. Numerical solution 0.668 . ^.5211 0.92 Values of y .932). Example 7. Nowadays numerical software is available which implements the formulas given in this chapter. However.75 0. This is illustrated in the next example. y 2 and y3 and their values are given in Table 7. ^ ^ . ^ . .933) can be written out for each value of j (j = 1.933) The boundary conditions are given by Equations (7.928a) will be nonlinear.50 0. (1993) modeled a catalytic combustion in a monolith reactor.932) o l (2+h 2 ) J _y3j Lsinhl_ Solving Equation (7.25 0.5214 0.12^ + ^1 = ^ 4 . For a given f. we still have to use our judgment in choosing the appropriate method. . we obtain yj. The monolith usually consists of a number of cells through which the exhaust gases flow...926a. Spence et al.4. using Equations (7.2526 0.8223 For nonlinear equations. Equation (7.8226 Exact solution 0. TABLE 7. 23).92. we obtain. . some of which are described in Sections 7.913a). We consider only a single cell and a schematic diagram of such a cell is shown in Figure 7.92 together with the values of the exact solution.
936b) .92 j / Schematic diagram of a cell The reaction considered is the combustion of propane and is given by C 3 H 8 + 5O 2 —> 3CO 2 + 4H 2 O The heterogeneous reaction rate rc at the catalytic surface is assumed to be rc = Z c e x p [ . R is the ideal gas constant.935) (7. and Tj is the reacting wall temperature as shown in Figure 7. For simplicity. (1993) deduced that the equations to be solved are 2 . a onedimensional model is considered.92.934) where Z c is a preexponential factor.NUMERICAL METHODS 669 8 r ' "I 1 M ^x ^ !  . C w (x) is the fuel concentration at the wall.w 6) .936a) ^ 2 _ J H [w(©X2©] (7. Using a series of energy and mass balances.J _ A i _• 2y 6) .60 G). This implies that all variables are functions of x only and only a crosssection of a single cell is considered for deriving the basic equations. Spence et al.E c / R T i ( x ) ] C w ( x ) (7. E c is the activation energy. 0 <£< 1 (7.* / / • L L FIGURE 7.
though .b) (7. D c . as shown in Figure 7. x 2 = T B (x)/T B0 .939) The shooting method is found to work only for P e < 1 and P e > 500. b) form a standard boundary value problem for a known w(x). Pe is the Peclet number.938b) The boundary value problem defined by Equations (7.937a. The reasons for the failure are as follows. For large values of P e . = PeJk{2yiwe0) d£ (7. w = T i (x)/T B 0 .939) is almost exact and the method works.936d) (7. T m . Tj are the temperatures at the outer wall. and yc are dimensionless parameters. X J = C B ( X ) / C O . 37a. L is the length of the cell. 0Q = T O / T B O . y = T m (x)/T B0 . F.936a) is written as a system of first order equation as follows ^L d^ = y (7.936a.d) = x2(0) = 1 Note that Equation (7.936a) is almost zero and the starting value of yj(O) given by Equation (7. TB(x) is the bulk gas temperature and T B0 is TB(0). As an initial guess. J D . For small values of P e .=0 d £ ^=1 Xl (0) (7.92.1) /w]}" where £ = x/L. J H . The outside wall temperature (0O) is given and Equations (7. J k . The associated boundary conditions are ^ = ^ =0 d £ i. The complementary function of Equation r / 1 (7. the mean solid wall temperature. and the temperature at the inner wall.6ZQ ADVANCED MATHEMATICS ~ = JD[xl(^)aw^)] (7.936b to 38b) can be solved by the shooting method.937c. the errors remain small and the method works.938a) ^2. CB(x) is the fuel concentration in the bulk gas and Co is CB(0). Equation (7. y j (0) is assumed to be Y l (0) = [w(O) + e o (O)]/2 (7. TQ.936e) °=JH[x2^)w(^)]+Jk[y^)w^)]rDcexp[yc(w1)/w]aw a w = xj {1 + (DC/JD) exp [yc (w .936d) is an algebraic equation and allows us to determine w(0).936a) is a linear combination of exp [± ^V2P e Jk J and. round off errors propagate as multiples of these exponentials. in the shooting method.936c) (7. the left side of Equation (7.
Using Equation (7. We illustrate this situation further by considering the following example.101.103) numerically.l ] + C2a2 1 (7. (1993) have used alternative methods to overcome this instability problem. The exact solution is y = e~2x We note that y is a decreasing function of x. The solution is of the form y.l = 0 The values of a j 2 (7. we have seen that due to roundoff errors.102) (7103) Instead of solving Equation (7. The roundoff errors (or truncation errors) amplify as the integration proceeds and the magnitude of the errors exceeds the solution.10lb) (7.105) From Equation (7. we deduce that a j 2 are the roots of a2 + 2 h a .103). the shooting method does not work when the Peclet number is in the range of 1 and 500.2 h y j (7.105).104) [«2 + 2 h a 2 . we discuss the problem of instability in more details. the finite difference equation is yj+iyji = .10la) (7.94. In the next section.= 2y dx y(0) = 1 Examine the stability of the system.10 STABILITY In Example 7.104) into Equation (7.NUMERICAL METHODS ] 577 unreliable. Spence et al.= C 1 a J 1 + C 2 a ^ Substituting Equation (7.106) are .922b). 7. we seek an analytic solution. This is an example of instability. Example 7.l ] = 0 (7. Write the finite difference equation for the system P. we obtain Cjoc^"1 [ 0 4 + 2 1 1 0 4 .
101 b). If Equation (7. When x. is a decreasing function of X J In an actual computation.1011) becomes the dominant term and the solution yj oscillates. choose another one. .1010a) (7..( 1 +2h) The solution is given approximately by yj = C 1 ( l . It is multiplied by an exponentially increasing function of x. and noting that lim (1 2h) 2 x J / 2 h = e"2xJ h>0 lim (1+2h) 2 x J / 2 h = e2xJ h>0 The solution y.£Z2 ____ ADVANCED MATHEMATICS al2 = 2h ± V l + 4 h 2 (7.108a) (7. the solution would be oscillatory for some values of X.exceeds a certain critical value. Instability depends on the differential equation and the method used.2 h a2 = . If one method is unstable. there is roundoff error and the value of C 2 will not be exactly zero.2 h ) j + C 2 ( . but will be small.109) (7. can be written as ^ = qe" 2 *) +C 2 (l) j e 2 x J (7.107) Since h is assumed to be small.1010b) (7.1011) (7. This shows that the difference equation is unstable. The roots are then approximately given by «1 = l . we find C 2 to be zero and y..103) were solved numerically.l ) j ( l + 2 h ) j Replacing j by Xj (Xj=jh). greater than a certain value. This arises because of the error in the initial condition and of replacing the first order differential equation by a twopoint difference equation.108b) Applying boundary condition (7. the second term (which is the error term) on the right side of Equation (7. we expand V1 + 4h in powers of h.
7148 The cubic x 3 .111.4ac . Answer: 0. Answer: 1. calculate xj a and xjt. Obtain the three roots of x 3 . b.8608 . One root x^ of a quadratic equation a x 2 + bx + c = 0 is xl = [b + V b 2 . show that xj can be approximated by xla = . b = 111 and c = 0. By expanding V b .4 a c ] / 2 a . Use (i) the secant method and (ii) Newton's method to find the root. Calculate xx if a = 1. x = x l a .2541.4 a c ) ] For the given values of a. 2]. Is the answer reliable? Note that Xj is given by the difference of two almost equal numbers and this could result in loss of significant figures. 1.c / b By multiplying the numerator and denominator by b + V b .. and x = x^.618 Deduce that an iterative formula for finding the cube root of a real number A is x k+l 3a..4ac . show that x^ can be written as xlb = [2c/(b + V b 2 . PROBLEMS la. Answer: 2. Test the convergence in each case. Evaluate a x ^ + b x + c for x = Xj.11.1 has a root in the interval [1.2x .NUMERICAL METHODS 673. 0.4 x = 0 has a root in the interval [0. 2a. Comment on the results.1149. 5a.107 Use the formula to calculate the cube root of 30. 4a = xk(x^A)/3x£ Answer: 3. The equation e x .4x + 1 by the fixed point iteration method.1]. Obtain the root (i) by the method of bisection and (ii) by Newton's method. c.
A = 0.1 = 0. 7b. Try to solve the equation x 3 . using Newton's method with starting values of x 0 = 0 and x 0 = 1. Show that the following set of equations has no solution x1 + x2 + x3 = 0 Xj + 2 x 2 + 3x 3 = 0 3x1 + 5 x 2 + 7x 3 = 1 Is the coefficient matrix A singular? What is the rank of A? 1 Ob. Are their solutions almost identical? Compute the determinant of the coefficient matrix A. Did you expect this value of IA I? Compute the condition number K (A).1 and explain why Newton's method does not work with x 0 = 0.08.x . T = 486. and R = 1. The RedlichKwong equation [Equation (1. i. V = 12.98. Plot the function f (x) = x 3 . Answer: 2491 Answer: — jr. . Are the other two roots real or complex? Compute them. Solve the following two sets of equations (i) 2 8 x j + 2 5 x 2 = 30 19xj + 17x 2 = 20 (ii) 28x! + 25x 2 = 30 19x! + 17x 2 = 19 Note that the two sets are almost identical.62)] can be written as P = RT Vb A V(V + b) Calculate b if P = 87. Answer: 1. 0 8 a.3247 Write the following set of equations in matrix form and solve by the method of elimination Xj + x 2 + x 3 = 0 Xj + x 2 + 3x 3 = 0 3 x j + 5 x 2 + 7x 3 = 1 9a.6ZA ADVANCED MATHEMATICS 6b.x .
(b) rounding off each number to three figures. In structural mechanics.3636 14a. Calculate its condition number K (A). by the method of elimination in (a) exact computation. 0. Compute the stiffness matrix K.3636.x 2 + 4x3 .x 4 = 1 4xj . which is the inverse of F .NUMERICAL METHODS j ^ lib. 12a.l ) . 2 ± il . Solve the system of equations Xj + 4 x 2 . 13 a.4545.x 2 =1 x3 + 4x4 = 1 by (i) Jacobi and (ii) GaussSeidel iteration methods. Use the QR method to find the eigenvalues of the tridiagonal matrix "2 1 0 " 1 2 1 0 1 2J Answer: 2. 0. Compute the inverse of 1 A = 1/2 _ 1/3 1/2 1/3 1/4 1/3 " 1/4 1/5 . the flexibility matrix F of a cantilever is given by L/(EA) F = 0 0 0 L 3 /(3EI) L 2 /(2EI) 0 L 2 /(2EI) L/(EI) where L is the length of the cantilever. Compare their rates of convergence. 0. E is Young's modulus. and I is the second moment.4545. The matrix A is a Hilbert matrix and its elements a. A is the crosssectional area. Answer: 0.x 3 = 1 .= l / ( i + j .
h. 6).610). x2 = x 0 + h. The solution of the differential equation subject to the condition y(*i) = yj /* x i+i is y ( x i + 1 ) = yi + i = y . y^) by fj. Calculate • it i j ( x 0 ) and i j (x 0 ) and verify Equations (7.. Xj and Xj_l5 y i+1 is given by yi+i = y i + ^ ( 5 f i + i + 8fifi_1) 18 a...676 ADVANCED MATHEMATICS 15b. In the derivation of the AdamsBashforth formulae. Can we replace the QR method by the Gaussian elimination in the calculation of eigenvalues? 16a. show that Vj+1 is given respectively by yi+i = y1+^(23fii6fI_1 yi+i = y i + ^ ( 5 5 f i . If the point XJ+J is included. x ^ .75.i + 5f i _ 2 ) 37fi29fi_3) + These formulae are the AdamsBashforth formulae [Equation (7. By approximating f by a polynomial of degree 2 (through 3 points) and of degree 3 (through 4 points).833)]. 17b. and x3 = x 0 + 2h. .5 9 f i . Xj +1 . + I f(x. Denote f (XJ. y) by a polynomial p n (x). deduce the expressions P2( x ) and P3(x) for equidistant points x 0 . . From Equation (7.y)dx Approximate f(x. x1? x 2 . and x 3 . Choose x^ = x 0 . Show that if f is approximated by a quadratic expression passing through xj + 1 . The Gaussian elimination method transforms a matrix to an upper diagonal matrix. we obtain the AdamsMoulton formulae.51) have the same eigenvalues. we have interpolated f through the points Xj. and not through the point Xj+i. . with a constant stepsize h. Show that the following functions are not cubic splines . in the form given by Equation (7.610). taking the interpolation points to be Xj. Show that the matrices A and B defined by Equation (7.
and a2 such that I [f (x) .y (x)] dx is a minimum (least squares approximation). 21a.. The sales of the product for three different prices are given in the following table Price ($) 1. The measured value has been reported to be 0.8 Compute the cubic splines that pass through the three given points.NUMERICAL METHODS 677 Il24x+18x24x3. The function f(x) = e x .50 330 Find the least squares regression line and estimate the demand when the price is $1.25 375 1. /•I 2 Obtain a0.1 0 I n _ i Use this formula to calculate I4 retaining (i) three significant and (ii) five significant figures. (i) f(x) = 54 + 72x30x2 + 4 x 3 . Answer: 349 22b. . 0 < x < l is approximated by y(x) = a o + a 1 x + a 2 x 2 .40. Show that the recurrence formula for evaluating the integral I n defined by * In = n . (ii) f(x) = 35 + 51x22x2 + 3x3. 19a.00 Demand  450 1. Expand f(x) in a Taylor series about x = 0.5 up to the term x 2 . l<x<2 2<x<3 l<x<2 2<x<3 The viscosity r of water at various temperatures T is given in the following table T("C) 10 20 30 Ti(cp) I 1. &\.8904 cp. 1331x + 2 3 x 2 .3 L0 0. 20b. The demand for a certain product is a linear function of the price. Compare this series with y (x). Evaluate the viscosity at 25°C.5 x 3 .
1 26a.1 to find y(0. (b) by the Simpson's rule. Answer: 2e x . Find conditions on h such that yy tends to the proper limit as i tends to infinity.0185 23a. Compare the two methods and proposed a modified recurrence formula so as to avoid loss of significant figures. Solve the equation dv f.1 0 / ( 1 0 + x ) ] x n . Use Euler's and Heun's methods to solve the equation ^ = x+y.a h Z + b/a is a solution of the difference equation. Evaluate I e __ 2 x dx (a) by the trapezoidal rule. that the initial value problem dv _^.678 ADVANCED MATHEMATICS Use Simpson's rule to evaluate I4. 24a. 25a. Xj = ih. Verify that yj = ( l . and (d) by the Gaussian method. dx y (0) = 1 + b/a a>0. Solve the equation analytically and compare the values of y at x = 0.7468 Romberg's method.X j ) . Show.= ay + b .( l . h is the constant interval ( x i + 1 . using Euler's method.x . 0<x<°o leads to the difference equation y i + 1 .1. (c) by the Answer: 0.1 dx] JO JO t\ Answer: 0. y(0) = 1 with h = 0.a h ) y i = bh where yj~y(xj). [Hint: I [x n /(10+x)] dx = I [ l .1).025 and (ii) the fourth RungeKutta method with h = 0. y (0) = 1 dx using (i) Euler's method with h = 0.= x + y + xy . .01.
Answer: 0. + 1 . Show that. under certain conditions which should be stated. and (b) the finite difference method. 27 a. In Problem 17b. analytically and numerically.l . y(4) = 3 0<x<4 by the method of finite differences. we obtain a multistep formula which can be written as . using the RungeKutta method. accurate to four decimal places. Use the AdamsBashforth method to obtain y at x = 1 with h = 0.NUMERICAL METHODS 679 Work to five decimal places. 0 1 (T20) = 0 dx2 The boundary conditions are T(0) = 4 0 . T(10) = 200 Find the analytical solution and solve the problem numerically using (a) the shooting method. The steady onedimensional heat equation can be written as 2 — .5). we can integrate from Xj_j to x i+1 instead of from Xj to x.1.11587. dx 2 dx y(0) = . Solve the boundary value problem ^ .0 . is 1. Choose an appropriate value of h.+2(2x)^=2(2x). given that y(0.1). 29a. Compare the results obtained by the three methods. compare the accuracy and amount of computations required associated with the two methods.3784 28b. Is h = 1 appropriate? 30b. A vibrating system with a periodic forcing term is given by ^ + 64y = 16cos8t dt 2 y(0) = y'(0) = 0 Find y (0.
Which of the two methods gives the correct solution? Discuss the stability of the two methods. y(O) = l by the formula obtained in this problem and by Euler's formula.680 ADVANCED MATHEMATICS y i + i = y i _i + 2hfi where h = Xj+j . . Solve the initial value problem ^ = 2y+l.Xj.
the value of u(x i? yj+k) by u i j + 1 . We shall also consider the method of finite elements.'s. Vj ) = u i + l i j = u i. Many equations cannot be solved exactly and we have to be satisfied with approximate solutions. yj)by u^. Using computers.. We divide the xyplane into a grid consisting of (n x m) rectangles with sides Ax = h and Ay = k as shown in Figure 8. we have u( X i +h. we consider the unknown function u to be a function of two variables x and y.'s. The solution is then an approximate solution. .CHAPTER 8 NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 8.D.E. we can sum only a finite number of terms. nonhomogeneity can be transformed to homogeneity via auxiliary functions as shown in Section 5. the method of separation of variables can generally be applied only to linear homogeneous equations with homogeneous boundary conditions. u i + s j + t denotes the value of u(xj+sh. we have discussed several analytical methods for solving P. In this chapter. 8. Using Taylor's theorem. and we have to integrate them numerically. the infinite Fourier series solution obtained in Chapter 5 has the appearance of an exact solution. in many cases. numerical methods are the most appropriate methods of solving some P. y^+tk). These methods cannot be used to solve all P.'s.E. and that of u(x i +h.2la.D. Similarly. the value of u(Xj+h. For instance. the Fourier coefficients are expressed as integrals and some of them cannot be evaluated analytically.'s.D. For example. yj+k) by u i + 1 j + 1 .D.E.j + h d ^.j where Rn is the remainder term.b) i. + 1>J h 2 d2 Y a ^ + R " (8.7. Also. but. we shall extend the method of finite differences described in Chapter 7 to P.2 FINITE DIFFERENCES For simplicity.y) by u i + 1 j .21. Even some of the exact solutions given in Chapter 5 are in reality approximate solutions. We denote the value of u(x i . where s and t are integers.1 INTRODUCTION In Chapters 5 and 6.E. Under favorable conditions.
j x +Rn jj (822a.22a.21 Grid system Similarly.u i _ l j ) / 2 h + O(h2) > J (8. 1 1 [ I I I I I o x. b).23a. b. i i i i „ . y j ) = Ui_1.j P3x i.b).b) by !r (UijUiu)/h dx i. we find that the difference of Equations (8.h ^  .j = u i ._.j <ui+lijuitj)/h (82"3a) The partial derivative 3u /9x can also be approximated from Equations (8.2la. j . ^ J x. + y 0 i. i — y. x i+1 xn x IZZIZIZIIIIZIZZ:}* h FIGURE 8. u(x r h. To include term of O (h ). b) yields — ax i j = ( u i + l j .23c) ..j (823b> In deriving Equations (8.. we have neglected terms of O (h2). we deduce that — can be approximated as 9x i. x.2la.682 ____ ADVANCED MATHEMATICS y •> y m— i yH y..b) From Equations (8. 2a.
and central difference forms of respectively.(ui + l. elliptic.j + l .u i .2la. j . — and — . We suppose the region of interest is . 2a. 8. backward. b) yields du/dx 0 i. Adding Equations (8.2 u 1 .j ~ (u1 + l.l .23a.l + u i .25a.can be approximated as dy dy ( u i.j + u il. We now consider the parabolic.3 PARABOLIC EQUATIONS The canonical form of the parabolic equation is ^ =« ^ (8.j + l . b.l > / 4 h k <827) The partial derivatives are replaced by the finite differences and a partial differential equation becomes a finite difference equation.26) Using the central difference form.c) dy ij (ui. j + l ~ u i + l . J ) / h 2 (824) Similarly.j (8. the mixed second order derivative is given by 82u ^ y . c) are the forward. j . b. Formulae (8.^/k 2 i. and hyperbolic types of equations.b.l . j ) / k — « (UijUijJ/k (8. J .u i .j+iuiji)/2k ~ ^ y ~ ("ij + lZUij+Uij.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 683.31) 2 at dx2 where a is a constant.
34a) (83'4b) where r = a 2 k / h 2 .684 ADVANCED MATHEMATICS 0<x<a. Formula (8. In this case. the values of u along the (j + l) t h row can be determined if the values of u along the j t h row are known.b. u J + 1 •. andu i _ 1 •• that is to say.33a.34b) can be written as u i. t) = g l (t).34c) +l = (12r)ui.c) We subdivide the region of interest into rectangles as shown in Figure 8. 5a).c) Example 8.24. Equation (8. t) = g 2 (t). t>0 (8. we have a semifinite region with t m tending to infinity and x n ( = n h ) = a. j = ^ ( » i + l j . anduj j j is an explicit formula. u n j = g 2 (kj). Equation (8. we can compute u along the second row. Note that Equations (8. The initial condition gives the values of u along the zeroth row. so the values of u along the first row can be computed. + 1 in terms of known values of M^ • u i + 1 :.t) = u ( l .35) gives the value of Uj : + 1 in terms of Uj • .j2ui.b) The boundary and initial conditions are assumed to be given by u (0.21 with the yaxis being replaced by the taxis.37a. Solve Equation (8.35) which gives the value of Uj .0) = f(x) (8. This process can be repeated for the subsequent rows.31) becomes «ij + l . u i 0 = f(ih) (8.b.j+r(ui+l.j+uilj) (835) Equation (8.2 u i j + u i _ 1 > j ) h = r(ui+l.j) (8.b) .« i . u (a. 0<x<l (8.31) with a = 1 by the method of finite differences. u(x. b.31. The boundary and initial conditions are u(0. t ) = 0 .j (8. Once the values of u along the first row are known.36a. Explicit Method Using Equations (8.32a. c) can be written respectively as u o j = g 1 (kj).33a.j + u il.
311) From Equations (8.0 0.4 (a) (b) 0. (b) r = 1.35) can be written as u i.8 0.200 — 1. b).400 x 4 = 0.563 0.31.b) ^ x < l (a) h = 0.484 1.0 0.800 0.344 0.005 0. k = 0.O) = 2(1x).600 . u(x.311) yields the values of U j .310a.600 — 0.800 0.344 0.0 0.400 0.6 — 0.200 — 0.200 — 2.200 0. u is symmetrical about the midpoint (x = 0.200 x 2 = 0.187 0.2 (a) (b) 0.5 Equation (8.2 0.0 0.200 — 0.4. By symmetry.2 — 0.700 0.015 0.550 0.375 0.01 (8.625 0.172 0.800 — 0. u (0.b.516 0.625 0.400 0. t) is equal to u (0.600 0.0 0.1.025 0.550 0.020 0.8 — 0.005.0 .200 x 5 = 0.625 0. we can calculate the values of Uj 0 and substituting these values into Equation (8.38a.375 0.0 0.d) (8.500 0.700 0.531 0.4 0.4 — 0.0 0. The values of Uj j then generate the values of Uj 2 .040 X!=0.400 — 0.030 0.035 0.400 — 0.0 0.400 0.172 0.0 0.b) l =0. In case (a) r = k/h 2 = 0.200 0.c. 5 .200 0.0 0.1.500 0.0 0.5) and so we need to compute the values of u in the interval 0 < x < 0 .0 0.000 — 0. that is to say.312 0. and this process is repeated to obtain the values of Uj m . t).430 0. O xo = O (a) (b) 0.600 — 1.5 (a) (b) 1. The computed values of u are given in Table 8.700 0.1 (a) (b) 0.600 — 0.200 0.0 0.0 0.000 x 3 = 0.187 0.400 — 1.0 0.400 — 0.453 0.600 0. Choose 0<x<i(8.700 0.5(ui+1J + ui_lj) (8.200 — 0. (a) r = 0.39a.0 0. TABLE 8.3 (a) (b) 0.6.31 Values of u for two values of r.010 0. k = 0.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 681 2x.800 0.6 0.531 0.200 — 1.j + (b) h = 0. the values of u at time t m .0 tj 0.0 0.38a.0 0.5.492 0.0 0. in this example.0 — 0. We note that.
b) and the temperature is positive.31. This example can be associated with the heat transfer in a rod of unit length with its two ends kept at zero degrees. j + l + u i .j + ( u i+l. Such an analysis can be found in a more advanced text. the explicit method is valid only when 0 < r < 1/2. Chapter 5] is u (x. CrankNicolson Implicit Method Crank and Nicolson (1947) proposed that 3x2 be approximated by the mean value of i. b.313) and these values are also tabulated in Table 8.313) Proceeding as in case (a).j il.314) Equation (8.315) .38a.(2n + I ) 2 7C2t] 7i2n=0 (2n+l)2 (8. such as the one by Richtmyer and Morton (1967).35) becomes u i.j i.31. stability. b).j (8. it can be seen that the values of u obtained in case (b) cannot be the solution of Equations (8.2 u i . A convergence.Y — ^ sin [(2n + 1) Ttx] exp [. the temperature can never be negative and we conclude that the values of u in column (b) are unacceptable. This example shows the importance of the choice of the value of r and the explicit method is not valid for all values of r. The values of u obtained in case (a) are in qualitative agreement with the analytical solution whereas the values of u obtained in case (b) are not.j2ui. we can compute the values of Uj l .j + l . 7a. t) = £.31. The analytical solution [see Problem 1 la.686 ADVANCED MATHEMATICS In case (b) r=l Equation (8.j ui. and compatibility analysis is necessary.j = f (ui+l. the temperature distribution is given by Equations (8.j + u il. Uj 2 . ••• from Equation (8. Initially.j) (8. In the present example.34b) is now replaced by dx2 i. Equation (8.U +l) (8. From the physics of the problem.j + i + u +lui. We next consider an implicit method which is valid for a wider range of values of r. From Table 8.312) + l = ui.314) predicts that u is an exponentially decreasing function of t.j 32u and 3 x 2 d2u .j + u i + l. 8a.
316) can be written as . v n _j. we introduce the following notation u i.317a.r v n _ 1 = cn_2 rvn_2 + 2(l+r)vn_1rvn = c ^ (8. we have only the values of u at an earlier time t. From the initial conditions. In the implicit method [Equation (8.319d) (8. .r v t + 2 ( 1 + r ) v 2 .320a.r v i + 1 = rw i _ 1 + 2 ( l . To simplify the writing of Equation (8. and v J + 1 only.318b) and we obtain .319c) .315).319b) (8. Vj. We write the first few as well as the last few rows of Equation (8. we obtain ruilj + l+2(1+r)ui. Equations (8.r v o + 2 ( l + r ) v j .b) (8.b) Equation (8.320c) (8.316) In the explicit method.r ^ + 2 ( 1 + r ) V j . the left side is in terms of V J J .35) involves three known terms..r ) w i + r w i + 1 = c{ (8. Writing all the known quantities on the right side. we can calculate the initial values of c j .b) In Equations (8. the right side of Equation (8.r v 4 = d3 (8. Xj.r ) U i j + ru i + 1 j (8. which are values of u at time (j + 1) at points XJ_J. the right side also involves three known terms.+ 1 ) and v n (= u n : + 1 ) are known and the unknowns are vj.j+lrui+l.316).r v 3 = d2 .316)].rv 4 = c 3 (8. v2. b).318a.r v 2 + 2 ( l + r ) v 3 .j +l = vi and u i.j = wi (8.319a to e) become 2 ( l + r ) V j . while the left side contains one unknown term. On the right side..r v 1 + 2 ( l + r ) v 2 .j + l = r u i _ 1 j + 2(1 .319e) If the boundary conditions are given in terms of u..NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 687 On simplifying Equation (8.318a.319a) (8.r v 2 + 2 (1 + r) v 3 . then v0 (= uQ .r v 3 = c2 .320d) .r v 2 = CJ+TVQ = dj . but the left side contains three unknown terms. and x i + 1 . .r v 2 = cj .r v n _ 3 + 2 ( l + r ) v n _ 2 .
Among the most widely used methods are the Gaussian elimination method and the iterative GaussSeidel method. d = i d n2 (8. v and d are column vectors.320f.320e) (8.1) tridiagonal matrix. the (k + 1) iterate of Vj is given by (8.l + r v n = dnl Equations (8. Equation (8.322) . v n2 .321c.1) elements.r v n _ 3 + 2(l + r ) v n _ 2 .l = c n .1) x (n .453)]. If the GaussSeidel method is chosen [Equation (7.321a) r 2(l+r) r 0 r 2(l+r) 0 0 r 0 0 0    0 0 0 l V2 d2 v = '.g) ( 1 + r ) y n .320 a to g) can be written in matrix form as Av = d where 2 (1 + r) r 0 A = 0 0 0 0 0 0 0 0 0 0 0 0 r 0 2 ( 1+ r) 2 (1 + r) r r (8.d) _ v nlj Ld»1_ The matrix A is a (n .r v n _ 1 = d n _ 2 rvn2 + 2 (8. each having (n .321b) vl d (8.32la) can be solved by one of the methods discussed in Chapter 7.688 ADVANCED MATHEMATICS .
1 4 . we can solve. Solve Equation (8.32.4 x 5 = 0.01 and solving Equation (8.37a.6 0.5.1 0 0 .6921 1. and d [Equations (8. A can be written as " 4 . the value of r is unity. In this case.4 0.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 689 From the initial and boundary conditions.1. we can obtain the values of Uj at time t 2 .01 0.3 N x 4 = 0.3956 0.5 only. and the analytical solution at the point x = 0. at time t j . From the symmetry of the problem. b. 20a tog).0 0.02.2 1.32 Numerical (N) and analytical (A) values of u N tj xo = o N Xi=0. Assume a to be unity and choose h = 0. Example 8.1 0 0 1 4 .8 d = 1. we can calculate dj at time t o (=O) and solving Equation (8. we find that initially the dj are " 0.1 0 A = 0 . These values can be used to compute dj at time tj and from Equation (8.6  (8. TABLE 8. We can now calculate dj at time t = 0.1 and k = 0.324) For the given A.1989 0. b).32la).321a).6461 1.7381 0.7691 0.8 0.2 N x 3 = 0.01.01. 24)].4 " 0.3789 0.1 4 0 0 0 .1936 0..323. This process can be repeated to obtain the values of uj at time t3.321a).0 0.2 0.1 N x 2 = 0.6 .0 0.318b.5834 0. we obtain the values of u at time t = 0. 8a. t m .5400 0.323) From Equations (8. We can repeat this process and obtain the values of u at subsequent times. we need to calculate up to x = 0.. using the CrankNicolson formula.7743 0.321a) to obtain Vj which are the values of u at points Xj at time t = 0. using the symmetry property (v6 = v 4 ). we obtain vj which are the values of u. Table 8. With the given choice of h and k. Equation (8.02 0..5 N A 0.6809 . by the elimination method.32 gives the values of u at various times t..2 0 0 0 1 4 (8.0 0 0 0.31) subject to Equations (8.
[Equation (8.325) (8.2.35)] u o..327) We note that u^ • is outside our region and can be eliminated via Equation (8. but the implicit method does.31) as shown later. Derivative Boundary Conditions So far we have assumed that u is given on the boundary.j) (8.33a) by ^ = Pu. From Tables 8. there is no restriction on r.5. it is seen that the CrankNicolson formula provides more accurate values of u than the explicit method for r = 0. in the heat conduction problem.690 ADVANCED MATHEMATICS We noted that for r = 1.j + i = (12r)uo. we use the central difference form of du/dx Equation (8. • Compared to the explicit method [Equation (8. t>0 d x x=0 where (3 is a constant.326) becomes (UlJu_1. The CrankNicolson formula is convergent for all values of r. if the material is insulated.326) (8. in the CrankNicolson method. the explicit method did not generate a meaningful solution. though the smaller the value of r the better the accuracy. the time interval can be larger. however. To obtain better accuracy. j = 0.23c)] and = 0 (8. the difference equation at the origin (x 0 = 0) is given by [Equation (8. . If we employ the explicit formula. Equation (8.31 and 2.j + r(ul.328) .325) corresponds to the case J3 = 0. 1.318a)] involves more computation. and this generally compensates for the increase in computational effort at each grid point. the CrankNicolson method [Equation (8. We now replace the boundary condition (8.j) = 2hpu0j .. it could be that the derivative of u is given on the boundary. In some cases.j + ul. there is no heat flow across the surface and the boundary condition is ^ on where n is the unit outward normal to the surface. For example.35)].
330) where y is a constant.j + ul.j + 2r[Unl.r v j = (1 .320a to g). b) yields ( 1 + r + r h p ) v o . the values of u at subsequent times can be computed.j2hPuo.329a) (8.r .334a.j] (8.w _ 1 = 2hpw 0 Combining Equations (8.331) and using the initial condition. We have two equations to determine v0 (= u 0 .331) From Equation (8. Uj 0 are known and from Equation (8. we deduce that u n. we eliminate u_j : to obtain u o.jO+hy)un.327) is true for all times and we have VJV^J (8. If the CrankNicolson formula is used.j + r(ul.333a) (8.333b) w 1 . 28).332) = 2hpv 0 (8. the vector v has two new elements (v0 and vn) .318a) . Using the same technique as at the origin. we obtain from Equation (8.+ l) and vn (= u n .j + l = u n.( l + h p ) u O J ] From the initial condition u 0 0 . (8.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 697 Combining Equations (8.j + l = (12r)uo.r ) wQ + rw.r v j = rw_j + 2 (1 . the values of u at later times at the origin can be computed.+ 1) .j) (8. 33a. Suppose that at the other end (xn = 1 ) the boundary condition is given by ^ d x = yu.329b) = uOJ + 2 r [ u 1 J . x=l t>0 (8.332.329b).hPr) w 0 + rwj = c 0 The boundary condition at the other end (xn =1) can be treated similarly. Equation (8. . Note that v0 and vn are not given and we work with Equations (8. and similarly vector c is increased by two elements (c0 and c n ).327. b) The matrix A has been augmented by two columns and two rows.r v _ j + 2 ( 1 + r ) v o .319a to e) instead of with Equations (8.
1/2 < x < 1 ^ =§5.1 0 0 0 1 0 0 0 0 1 rvol vi [co~ ci 4 . .335) (8.1 4 .01). Solve the equation 9H.338) with the given values of Cj [Equation (8.337a. We repeat the process and we obtain the values of u at times t 3 . 1. b)] and Equation (8. (i = 0.b) " 2 .6.338). Table 8. As discussed before.336a. = i J i > 9t 3x 2 subject to the conditions /2x.339)]. we can compute Cj at time tj and.334a.8. Choose h = 0.01..1 4 . t m .1 v2 v3 V4 _ " c2 c3 C4 <8338) 0 0 0 0 .8]. we obtain vj which are the values of u^ at time tj (= 0. Writing Cj as a row vector. we obtain the values of Uj at time t 2 (= 0..339) Solving Equation (8.337a.x). x=0 = 0 <x<l (8.336c.02). 19a to e) can be written as (8.33 gives the values of u.b) o. 0) = ( \ 2 (1 . . 0 < x < l / 2 u(x. The value of r is one and the value of P is zero. on solving Equation (8..5. the problem is symmetrical about x = 0.336a. 0. . 0..5 and we need to compute the values of u between 0 < x < 0. we have Ci = [0.4. 0.33.692 ADVANCED MATHEMATICS Example 8. 5) (8.1 0 0 0 1 4 1 0 0 0 .d) ax ax x=1 Use the CrankNicolson formula and the central difference scheme for the boundary conditions. With these values of Vj. .1 0 0 0 . t^. for all t (8..1 and k = 0. at various times.337b).2. Equations (8. b) become 2 v o .V l = wl = c 0 Equations (8. b.2 4 J L V 5J LC5 The initial values of Cj can be computed from the initial conditions [Equations (8. 0.0.
390 0.214 0.341a.GAMM .T H E RIGHT SIDE OF THE EQUATIONS YNEW THIS VECTOR WILL CONTAIN THE NEW SOLUTION VECTOR BETA.C .01 0.31.214 x 5 = 0.c) 9x 2 .t) = u ( l .4 0.189 x4 = 0. t ) = 1 (8. 0 <x< 1 (8. This is confirmed in Figure 8.342) which is adopted in the following program.3 0.WORK SPACE FOR THE EQUATION SOLVER . we deduce that u—>1 as t—> °o.0 0. The results are illustrated in Figure 8.2 0.b. 43) ' From Equation (8.34 SOLUTION METHOD IS CRANKNICOLSON Y .343). 0) = 0 .212 0.02 xo = o x1=0.6 0.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 693 TABLE 8. u(0. C C C C C C C MAIN PROGRAM TO SOLVE EXAMPLE 8.4 0.342) It is Equation (8.234 0.8 0.0 0.340) Using the CrankNicolson method.316) which on dividing by r can be written as Uil)j + l + 2 ( l + l/r)uiJ + 1 .140 x2 = 0.0 0.33 Values of u for r = 1 tj 0.454 0. Equation (8.THE COEFFICIENTS OF THE TRIDIAGONAL SYSTEM OF EQUATIONS D .u i + 1 ) J + 1 = u i _ l f j . Solve the heat equation — = — at subject to u(x.5 1.340) is replaced by the finite difference Equation (8.34.1 0.31.l / r ) u i > j + ui+1>j (8.274 0.B .THE SOLUTION VECTOR A.427 0.176 Example 8. The analytical solution is u=1_A£""P[^On')2t]sin(2ll+1)lIX n=0 ^ (83 .151 x3 = 0.2 0.2 ( l . t > 0.
D0 CONTINUE BOUNDARY CONDITIONS AND INTEGRATION PARAMETERS YNEW(1)=1.Y(6).BETA(21).D(21) REAL*8 DT.694 ADVANCED MATHEMATICS C C FOR ILLUSTRATIVE PURPOSES ALL VECTORS HAVE THE SAME DIMENSIONS ALTHOUGH SOME ELEMENTS ARE NOT USED REAL*8Y(21).D0 CONTINUE INITIAL OUTPUT WRITE(6.YNEW(21).A(21).Y(16).D0*(DX**2)) CFl=2.DO B(I)=CF1 C(I)=1.D0 DT=.C(21). NITER=25 ALPHA=DT/(2.D0 YNEW(I)=O.21 A(I)=1.B(21).DO GAMM(I)=0.D0/ALPHA) CF2=2.D0+(l.DO D(I)=O.ALPHA.DX.002D0 DX=.Y(21) 2 C C START INTEGRATION DO3 L=l.DO/ALPHA) 1 C C INITIALIZE THE VECTORS FOR THE TRIDIAGONAL SYSTEM DO2 1=1.999)T.21 Y(I)=0.DO YNEW(21)=l.GAMM(21).CF2 C INITIALIZE THE SOLUTION VECTORS DO1 1=1.CF1.10 DO 31 K=l.Y(1).Y(11).DO BETA(I)=O.NITER .05D0 T=0.998) WRITE(6.D0( 1 .
GAMM) T=T+DT 311 C C UPDATE THE SOLUTION VECTOR DO 312 J=l.C(2). * 'X=1.Y(11).GAMM YNEW .A.Y(16).YNEW(N).'X=0.'X=0'.4X.4X.D(N).BETA(N).D.7X.07) 999 FORMAT(1X.WORK SPACE VECTORS .19 D(J)=Y(J1 )+Y( J+1 )CF2* Y(J) CONTINUE SOLVE THE EQUATIONS AND UPDATE TIME CALLTRDIAG(19.GAMM(N) .THE SOLUTION VECTOR C C C C C SUBROUTINE TRDIAG(N.GAMM) REAL*8 A(N).C(N).A(2).7X.B(2).YNEW.Y(6).BETA.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 695 C THE RIGHT SIDES FOR THE FIRST AND LAST EQUATIONS D(2)=Y( 1)+Y(3)CF2* Y(2)+YNEW( 1) D(20)=Y( 19)+Y(21 )CF2* Y(20)+YNEW(21) C THE RIGHT SIDES FOR THE OTHER EQUATIONS DO 311 J=3.5'.T H E NUMBER OF EQUATIONS .C D BETA.21 Y(J)=YNEW(J) 312 31 C CONTINUE CONTINUE OUTPUT RESULTS WRITE(6.B.'X=0.'T'.B.5X.6F10.BETA.75'.Y(21) 3 CONTINUE STOP 998 FORMAT(7X.THE VECTORS OF COEFFICIENTS OF THE LEFT SIDE THE RIGHT SIDE VECTOR .5) END C SUBPROGRAM TRDIAG SOLVES A TRIDIAGONAL SYSTEM OF EQUATIONS N A.Y(1).'X=0.YNEW(2).25'.B(N).999) T.D(2).C.
15 *••• t=0.10 * t=0.4 ELLIPTIC EQUATIONS An example of an elliptic equation is Laplace's equation which can be written in twodimensions as _ _ + _JL _ 0 dy2 3x 2 (8.30 »•• t=0. * • * A A * * • . 2 I 1 \ x \ 1 1 1 < s / / 1 1 1 * 0 0.m ADVANCED MATHEMATICS 1 2 BETA(1)=B(1) GAMM(1)=D((1)/BETA(1) DO 1 1=2.35 FIGURE 8. A .8  \m ^ N.05 *1=0. . A  * A / 0.  • / / / / \ \ * _ . TK: "^ _i'' :«^ : ' ^ =r / ^ 3 p { 0 .41) .20 « t=0.4 0. _ . .2 0.25 »1=0.N1 I=NJ YNEW(I)=GAMM(I)(C(I)*YNEW(I+1)/BETA(I)) CONTINUE RETURN END \ \ •  .6 0.8 1 X *1=0.N BETA(I)=B(I)(A(I)*C(I1)/BETA(I1)) GAMM(I)=(D(I)(A(I)*GAMM(I1)))/BETA(I) CONTINUE YNEW(N)=GAMM(N) DO 2 J=1.31 Temperature distribution at various times 8.
b. For simplicity. . Equation (8. ! . a combination of u and its normal derivative is given on the boundary (Robin problem).c. Section 6): (i) (ii) (iii) values of u given on the boundary (Dirichlet problem). maximum modulus principle). Dirichlet Problem We assume the region to be rectangular and given by 0 < x < a.d) The boundary conditions can be written as u 0.j + u i J + 1 + u i J _ 1 .0 ' u i. such that xn = a> ym = b ' xo = yo = ° (8.d) where (Xj : are given.43a) = 4 ( u i+l.j = an.0 = a i. We subdivide the region into small squares each of length h (= k).m (8. we obtain the difference form of Laplace's equation and it is U. 6) into Equation (8. as shown in Figure 8.24.42) h2 k2 It is usual to take h = k and Equation (8.l ) ( 8 " 4 " 3b ) Equation (8.4 u i j = 0 or u i.43a) is known as a fivepoint formula as it involves five points. j + u i .j + u i .2 U : • +U. The appropriate boundary conditions associated with elliptic equations are (see Chapter 5. 0 < y < b .C. .43b) shows that the value of u at point (i.43a) now becomes.21. normal derivative of u given on the boundary (Neumann problem).j ' Ui.m = ai.l . This implies that the value of u at an interior point cannot be a maximum or a minimum (see Chapter 3.45a.44a. i .j (8. Equation (8.j = «0. we consider a coarse mesh and assume m = n = 3. using the rowbyrow ordering.42) simplifies to Uj+ij + u j .b. j + I + u i .2 U j : + U : i: Uj • . . (8.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 697 Substituting Equations (8.• .j ' Un. . j) is the average value of u computed from its four nearest surrounding points. j .41).
An alternative way of writing Equation (8.4 v 4 = a3.46d) a n d U2 .1) (m .l + u 0 .47a to h) can be written in matrix form as "4i 1 4 .3+ul.4 u 2 2 + u 3 2 + u l 2 + U2 3 + U 2 1 = ^ We label the four unknowns (u j j .48b) is " B 7 I ' " v = r (8.49) L _ where ]_ is the (2 x 2) identity matrix and B is another (2 x 2) matrix.2+ul. For any value of n and m.47e. A is a (n .l ) x ( r i .48a). 2 + ul.d) (8.46b) (8.47g.b) (8. 1 .47c.0~a3.3 Equations (8.4 J [_V4J L r 4_ (8. 6a to d).1 ) denoted by B .2 + u 2 2 + U2 o ^ + u0.48a) 1 0 . 3 (8.2~a2.698 ADVANCED MATHEMATICS 4ul.l+u2.48b) Note that the matrix A can be partitioned into four matrices as shown in Equation (8.1) rows * (n . u^ 2> Equations (8.  4 1 3 0 or Av = x 1 . l + u l . A and B are of the form .1) identity matrices each of size ( n .4 u 2 i +u3 2 +u1 4ul)2+u2.45a to d. 0 i 1 o i rvi] V2 r r i" r2 = Vo r.f) (8.h) l~4v2+V4 4v3+v4 ~a2.l ) and (m . .47a.0 1 = ° = (8.1) columns matrix.l 1 3 Vl = a a 0 2 = r3 = r4 V2+ v 3 .46c) (8.1) square matrices each of size ( n .1 ) x ( n .l = ° .1) (m .46a) (8. It can be partitioned into (m . u 2 j . we obtain 4VJ+V2+V3 V 2) a s (vi> V2> V 3 ' anc * V4^ Combining =ao>1aljO = = rx = r2 (8.
411) Note that. j . .j = (k) i U i.410a) o 4 1 B = i l l i 1 4 1 '.4). Section 4.13 ) .410b) Q " o l 4 l 1 4 If the size of A is small. Section 7. and given that it is sparse. .4 u i .j+l+ui.jl 4ui. From Equation (8.j3 < 8 . In this chapter.R. j + u i.43b). it is more economical to adopt the method of iteration. <8412) To improve convergence.j = (k) „ U i. j ] . j + U i . we see that a possible iteration formula is (8.R.48b) can be solved by the method of Gaussian elimination.). We use the same notation as in Chapter 7 and the k th iterate is written as u.j +f (k) (k+l) (k) (k+l) (k) [ui+l.j1" Equation (8.O.o ..j +ui. But if A is large. and Ui. as in the GaussSeidel method. . method is essentially a modified version of the GaussSeidel method (see.412) is modified to (k+l) u i. we have used the new values (k + 1 iterate) of u : i . 0 " (8.l .j + 4 (k) (k+i) (k) (k+i) (k) [ u i + l . we describe the method of successive over relaxation (S. .j + l + u i .NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS $99 "1 i I A = B I '.O. Equation (8.411) can also be written as (k+l) U i. (8.4 . Equation (8.j+uil.l . This method is discussed in Chapter 7. The S.
d) 9y 2 0<x<3. .420a) .75 (8. For the Dirichlet O (coop) is given by problem in a rectangular region.419) Taking into account the boundary conditions.8 0 . u(0.O. u 2 j . .416a.41.700 ADVANCED MATHEMATICS The optimum value of G O depends on the mesh size and lies between 1 and 2. we deduce that G0Op is given approximately by co op = 1.25 .25.9 0 ] Solving Equation (8. Example 8. the optimum value of G coop = 4 (8. we obtain V][ (8.418a. 3) = 9 0 .75 v 4 = u 2>2 = 43.48a) directly (Gaussian elimination).1 6 0 .y) = 7 0 . see Hageman and Young (1981).415) We divide the region into 9 unit squares.418c.1 0 . y) = 0 (8.47a to h). we need to assume the values of the zeroth iterative of the interior points.R.4_14) 2 + ^4(cos J + cosJ^r For details on co op . From Equation (8.b. method. Equation (8. We now solve the same set of equations by the S.0)=10.417) =u u =41. Uj 2 ' U<U U = U and U2 2) c a n ^ e written as [u 2 k i + 7 0 + u i k 2 + 10 U + °'289 4uU ] (8. u(x. The values of u at the four interior points are given by Equations (8.c. One possibility is to assume that the value of all the interior points is the same and equal to the average value of the boundary points. Solve the equation U i + ^ J i = 0. .414). 0<y<3 (8. v 2 = u 2 1 = 23.b) (8. 3x 2 subject to u(x.156 (8. u (3.d) v 3 = Uj 2 = 61. To start the iterative process.413) for the four unknowns (Uj j . The transpose of the column vector x is given by LT = [ .
0<x<l.5 42.331 43.41.4 u £ ] (8.42.039 24. we assume that the value of u at the interior points is equal to the average value at the boundary points.750 Neumann Problem In this case.420b) (8.422a. This will introduce points which are outside the region of interest and these fictitious outside points can be eliminated as previously shown in the parabolic case.240 43.420d) To start the iteration.4 u ® ] (8.407 63.752 k=3 41.252 23. Example 8.289 [ u ^ + 70 + 90 + u £ * 1 } . Solve the equation — + — = 0. 0<y<l (8.2 u2(2 k=0 42.761 61. The average value in this case is 42. the values of u at the boundary points are not given and have to be determined at the same time as the interior points.5.277 23.4u(jk)2] S^ =u£ + 0.5 k=l 41.055 20.289 [u(^+1) + 90 + u * + 1 ) . The next example shows how this is done. The values of Ujj at subsequent iterations are given in Table 8.5 42.420c) u ( j k2+1} = u ^ + 0. We can proceed in the same way as in the case of parabolic equations and write the normal derivative in a finite difference form.689 61.41 Values of u j : using the S. TABLE 8.R.784 43.703 k=2 41.b) .O. method ug uu u2)i "1.289 [u ( £ \ u £ + 10 .757 43. It can be seen that the convergence is rapid.421) ax 2 dy2 subject to the conditions ^ = ^=0 dx x=0 3 y y=0 (8.755 k=4 41.197 60.5 42.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 70/ u £ 1 } = u™ + 0.
1) l±*l il3ll) I (2.d) We subdivide the region of interest.41.2). The first and last few difference equations [Equation (8.422c.  <0.1)1 1 Q. On writing the difference equation for points on the boundary.424b) l)l+ul. That is to say h = k = 1/2 (8.43a)] are ul.3) I (1.424C) (8.3)_ J_5. into four squares. .1) (2.41 Grid for solving Laplace's equation in the Neumann case .423a.0) 0.2) (1.2) (1.2) (2.0 = ° u 2. t 0 '" (0.l . Note that only the single interior point (1.5.43a).0) I (3. l + u 0 .5 .702 ADVANCED MATHEMATICS P5 x = 1. Oj3)_ 0.4 u 0 .0 + ul.2 + u l ) 3 + u U 32 + U 12 + U2 l!l4ul)2 = ° = (8.b) Equation (8. 0 u2)0 + u0)0 + u = ° (8.0) 0 .1) (1. each of size (1/2 * 1/2). as shown in Figure 8.2) ' 1(3.5 x I i (3.0) i  I I 1(1. which is a unit square.l4ul.1) FIGURE 8.424d) 3 + U2 1~ 4 u 2 2 ° y H.2 + u 0.424a) (8. 10.O) t 1. we will introduce fictitious outside points which lie on the dotted lines as shown in Figure 8. (1.421) in difference form is given by Equation (8.41.3) (2i3)_ (3.1) has all its surrounding points in the region of interest. x=l ~ ^ y=l = 1 (8.0 + u 0 .
428) can be written as ui + l. u j 0 . .l 2.j + u il. Uj _ j . u3.0 0..j + l+ui. Uj 1 .l = 0 1 0 0 1 1 1 L0_ (8. u 3 :.1 + Uj j) respectively..g (x.julJ ujjuj^j = 0 ui.425a. This set of 9 equations is ..4 2 1 . the finite difference form for Equation (8.u _ l j = 0. Using the central difference form for the derivatives [Equations (8. the boundary conditions [Equations (8.426a) 0 0 1 2 0 0 1 0 0 0 0 1 0. o l [ ~ 0 ~ 0 0 0 1 0 2 4 4 0 0 0 1 0 1 J 0 u u u u U u u l.j.426b) (8.d) = 1.0 2.4 u i. and Uj 3) can be replaced by (Uj :. 1 + Uj •.b) (8.429a) .2j Au = b Equation (8. =1 The values of u at the outside points (u_j •. Since the size of A is manageable. We now have a set of 9 equations to solve for the 9 unknowns (u0 0 .l l. u Q _ j . u 2 2 ). u 3 2 . and u 2 3 in Equations (8.422a to d)] can be written respectively as u l j .jlh2gij = 0 (8.426a) can be solved either by a direct method or an iterative method.424a to d) via the boundary conditions. y) = 0 (84"27) (8.j + u i.4 0 2 . we have obtained the solution by a direct method and it is ST = [ftM£*2f'°] Poisson's and Helmholtz's Equations Poisson's equation can be written as V u .3ui.l (8.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 701 We eliminate u_j 0 .2 Lu2.2 l.4 1 0 0 0 0 0 _ or 0 0 0 0 1 0 0 0 1 2 0 0 0 2 0 4 2 2 0 2 0 2 4 2 0 0 0 0 1 0 4 4 0 0 1 2 0 0 00 0 ^ 1 r u o .23c. . and .428) Proceeding as in the case of Laplace's equation.425c. 5c)].
The program used to solve the present problem is next listed.OUT) C SET BOUNDARY CONDITIONS .y) = 1 . 31a)] can be solved by the methods discussed earlier. j + u il. The results are shown in Figure 8.433a. y) = 0 Its finite difference form is u i + l .j The value of C O is chosen to be 1.43 la) (8. j .OMEGA.y) = .435) MAIN PROGRAM FOR SOLUTION OF EXAMPLE 8.) (8.( 4 .Y.b) (8.L i j + u i / i + u i A i > / 4 + < x ^°> u i. u(x.11).433c.h 2 g i j (8. 1) = x 2 . The analytical solution is u = x2y2 C C (8.j .432) subject to the following boundary conditions u(x.704 ADVANCED gij MATHEMATICS where = g (xi. y) u . 3y2 3x2 0 < x < 1.43 USING SIMULTANEOUS OVERRELAXATION TO SOLVE THE FINITE DIFFERENCE EQUATION REAL*8U(11.FILE='MATRIX3. that is (k+1) u!.430) = ° (8.43lb) where fj = f(x i? yf) The difference equations [Equations (8. (k) (k) (k) (k) s / (k) (8434) = « > ( » L u + . j .X. Example 8.42.j + u i .l u(l.y 2 (8. u(0. 0<y< 1 (8.5.432) can be written in the form of Equation (8.y 2 .h 2 f i j ) u i .d) The finite difference form of Equation (8.l .0) = x 2 . y.429b) Helmholtz's equation is V 2 u + f (x.g (x.43.229a. j + l + u i .TEMP OPEN(UNIT=3. Solve the equation — +— =0.413).
2.999)U STOP FORMAT(11E12.3) END SUBROUTINE SOR FOR SIMULTANEOUS OVERRELAXATION THE CALCULATIONS ARE DONE IN PLACE RATHER THAN COPYING TO ANOTHER MATRIX SUBROUTINE SOR(U. J+1 ))/4.40 CALL SOR(U.I)=1.M) REAL*8 U(N.M). J)+U(I.D2 U(1.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 705 1 2 C DO 1 1=2. J)+U(I+1.11.1)=TEMP U(11.J) CONTINUE RETURN END 1 .MSTART.10 DO1 J=2.OMEGA 3 C 999 C C C C LOOP THROUGH THE MATRIX DO 1 I=NSTART.OMEGA.11) CONTINUE OUTPUT THE FINAL RESULTS WRITE(3.N.11 TEMP=(I1)**2/1.M1 U(I.N1 DO 1 J=MSTART.I)=TEMP U(I. DOTEMP U(I.2.NSTART.5D0 DO 3 NITER=l. J1 )+U(I.D0OMEGA)*U(I.D0 DO 2 1=1.D0 * +(l.11)=TEMP1.DO CONTINUE SET OVERRELAXATION PARAMETER AND ITERATE THE SOLUTION OMEGA=1.I)=0.10 U(J.OMEGA. J)=OMEG A*(U(I1.
9t 2 3x2 t>0. 0). 0<x<J! (8.42 8.5lb.5ld. ^ dt (8..11)] is an example of a hyperbolic equation.jl = c 2( U i + l. is a set of boundary and initial conditions.j + U il.5la) is Ui.706 ADVANCED MATHEMATICS 1 ^ ^ ^  ^ " \ .=c2^.D.2 u i.e) where c and Jt are constants.j.5 Values of u as a function of x and y HYPERBOLIC EQUATIONS Difference Equations The wave equation [Equation (5. The finite difference form of Equation (8.t) = u ( i . t ) = 0 u (x. = f (x). Associated with this P.j + l2"ij+ui.. We write the whole system as ^.i) (g5 2) k2 I h2 j .E.c) =g(x) t=0 (8. 5 y x FIGURE 8.5la) u(0. 05 ^ ~ " \ _ ^ ^ ^ 0 .
.l + 2kgi) (857) Note that uQ j and u n j are given by Equations (8.t) = u ( l . t 3 . f) involve the fictitious point of t _ j .55b) = 2ui>0ui.j u i. Solve the equation ^ =^ (8. As usual. xn = J .c) Equations (8. 5a) yields u i. t ) = 0 (8.5lb to e) are now written as u o.54c to f. t) is subdivided into rectangles of size (h x k).c) . b) and need not be computed.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS ZGZ where the first subscript denotes position (x) and the second subscript denotes time (t).56b) Equation (8.b) (8. Equation (8.56b) simplifies to u i.53a. we can similarly determine the values of u at subsequent times t 2 .0 (8. Having determined the value of u at time ti.f) "^^k1'"" g(Xi) = gi Note that Equations (8.58b.54a.54e. Example 8. to = O (8.l+ui+l)02ui.b.0 = fi+i+fi_iui>1+2kgi + u il.l = ^ ( f i + l + f i . Recall that x o = O.54c. the space of interest (x.j2uij+uilj) (85"5a) where r 2 = c 2 k 2 / h 2 We choose r to be unity and combining Equations (8.54e.51.j + l = 2u ijui)jl+r2(ui+l.54a.l (8.52) can be written as u i.56a) (8.58a) at 2 ax 2 subject to u(0..o = V j = °' = f ( x i ) = fi V J^° (8..d) (8.
2 ^ dt =0 t=0 (8.060 0.045 0.3 0.040 0. TABLE 8.jl (8.02 0 0.51.080 0.b) u =i(*i+i + *iixi2+i*ii) < 8 .x).1 0.58d)] provides u i 0 .5 Equation (8.5 . In this example.511).100 0.105 0.511) The initial condition [Equation (8.5 0. (*O12) .j + l = ui+l.510) determines Uj j .080 0.030 0.020 0. can be solved analytically (see also Chapter 5.59a.2 0. we need to compute for the interval 0 < x < 0. then from Equation (8.120 0.3 0.115 0.e) We choose h = k = 0.04 0 0.j + ui_l. From the symmetry of the problem. the values of u at all positions and subsequent times can be computed.58b to e).512).> 7T s = 0 (2s+ I ) 3 From Equation (8.075 0.57) simplifies to u (8.040 0.045 0 x t 0 0.03 0 0. g is zero and Equation (8.085 0.5 only.4 0.55a) simplifies to u i.2 0.120 0.1 The value of r is then unity.708 ADVANCED MATHEMATICS u (x.105 0. we can deduce that at .58a).060 0.51 Values of u at various positions and times 0 0 0 0 0 0 0 0.5 are tabulated in Table 8. The values of u from time t = 0 to t = 0.10 ) Equation (8. Problem 9a) and the solution is 4 V cos (2 s + 1) 7tt sin (2s + 1) %x u = —.100 0.jui.075 0. 0) = 1 x (1 .4 0.125 0.58d. Equation (8.01 0 0.1 0. subject to Equations (8.
43 that the domain of dependence of Equation (8.511) is exactly that of Equation (8.b) From Table 8. we observe that there is complete agreement between the analytical and numerical results. This concept is discussed in Chapter 5. Another way of showing that Equation (8.511) provides an exact solution of the problem is to consider the domain of dependence of the solution.h + tp .5la).t ) + f (x + t) Using Equation (8. we have used the method of characteristics to solve hyperbolic equations.f ( x i .jui.51.515b) (8. Section 4) of Equations (8.jl = f ( x i + i . In Chapters 5 and 6. In Figure 8.15d) is exactly the left side of Equation (8.513a. Equation (8.511).5.51 and 5.51.55a) is a stable algorithm for solving Equation (8. u=0 (8.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 709 t = 0. Thus.511) is ui + (8. Noting that the characteristics in this case have gradients ± 1. This method has the advantage that discontinuities in the initial data are propagated along the characteristics. D'Alembert's solution (Chapter 5.t j ) + f(x i + 1 +t j ) + f(x i _ 1 t j ) + f(x i _ 1 +t j ) .515d) We note that the right side of Equation (8. The parameter r is the Courant parameter and it has been shown by Courant et al. (1928) that if r is less than or equal to one.58a). Section 4.515a) = f (xj + h .t j _ 1 ) .tp + f (xj + h + tp + f (xj . . • We now show that Equation (8.511) is an exact difference equation of the problem.t j + h)f(xi + tjh) = f(xi + h + tj) + f ( x i .j+uil. we observe by comparing Figures 8.515c) (8.f ( x i .f ( x i + tj_1) (8.5la to e) with g (x) = 0 can be written as u = f ( x . we have marked the pivotal triangular domains which determine the value of Uj .h .t j ) = ui>j + 1 (8. the right side of Equation (8. This result is due to r being unity (h = k).h .514) l.514).5.+ 1. this method is generally more suitable for cases of discontinuities in the given data.tp + f (xj .
_ _ . d x + ^ d y dx dy = 3ji d x + J j L d y dx dxdy (8. which can be written as ll—7 dxz + a i2 a — + a22—7 = dxdy dyz f (8.E.D.51 Pivotal points determining U j j + 1 Method of Characteristics We consider a quasilinear P. ^ r — and ^ — .517a.518b) .516) where a. we denote 3u P = 3u q = . (8.: and f are functions of x.b) d7' 37 Writing out the differentials of p and q yields dp = ^ ._ . . J dx dy For simplicity. u.710 ADVANCED MATHEMATICS t j+l • • • • • • • • • • • • • • • • • i x FIGURE 8. y.518a) (8.
52.a 1 2 + a22— = f 11 dx ll dy 3x9y\ dx 1Z z z dy/ Equation (8.^ T " a l l 7 " .ir^ + a ^ . Section 3.518d) d2u dy dxdy dx (8.— ^ — a 1 1 . . b) into Equation (8.b. m_ and F_ denote the root and the curve corresponding to the negative sign.518c) (8. dy dx dy = ^ _ d x +^ d y 8x dy dy From Equations (8.518b. dy d2u [ (dy\ 2 (dy\ 1 n . such that the slope of the tangent at every point of F is a root of the equation a l1 (8.522) has two roots dy dx = a12±Va224alia22_ 2ajj = m (8.519b) Substituting Equations (8.^ .520) dp dy dq dy .O1 .d) Since the equation is hyperbolic..520) can be written as .^ .o _ __. 7 ^ ^ + a 2 2 T ^ ^ . we deduce a2u _ dp 3x 2 ^u 3y 2 = (8.523a.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 7U_ dq = ^L d x + p. (8. F + are the characteristics of Equation (8. the two roots are real.522) Equation (8.p 1 .521) 11 dx dx ll dy dx dx 3x3y L Vdx/ lz\dx/ ZZJ We now define a curve F in the xyplane. Similarly. a.f ^ " .o .516) and are discussed in Chapter 5. d).519a) dx dq_J2JLdx dy 9x9y dy (8.a 1 2 r + a 22 = ° (8. We denote the root given by the positive sign by m + and the curve corresponding to m + by F + .516) yields dp dq 9 u / dy dx\ .c. as shown in Figure 8.519a.
b). C is not a characteristic. + a 2 2 ll ^f^ = 0 dx dx (8.525a) (8.m .712 ADVANCED MATHEMATICS ^—P ' ^ \ c FIGURE 8.525a.524a) (8. and q are known.524b) a n T^dp + a22 dq .x Q ) xR = —^ r ^ R (m + m_) /««<v^ (8. and R denote that the quantities are evaluated at P. Q. The characteristic F+ through P intersects F_ through Q at R.f dy = 0 Let C be a curve along which initial values for u. As a first approximation.525b) = mQ(xR~xQ) where the suffixes P.52. The equation of these lines are YRYP YRYQ = m+p(xRxp) (8.52 Method of characteristics It follows from Equation (8. y^) and they are (YQYp) + ( m + x p . Q. and R respectively. Let P and Q be two neighboring points on C. we assume PR and QR as straight lines with slopes m + and m_. we can determine (XR. p. as shown in Figure 8.521) that along a characteristic a n ^ ^ 11 dx dx or LI A. From Equations (8.526a) .
This process can be repeated until the required accuracy is achieved. Similarly.f (y R .525a.530b) or uR = u p + i(pR + p p ) ( x R .y<2 = 2 ( m . To determine u R we use the relation du = ^ dx + p. and qq are known.529b) is then approximated by UR . .527a.x p ) + l ( q R + q p ) ( y R . b) by the average values of m + and m_ evaluated at P and R.527b) Combining Equations (8.y p ) = 0 a j ! m_ (p R .524b) along F+ and F are respectively given by a n m + dp + a 2 2 dq . b) yields values of p R and q R . Equation (8. q p . Equation (8.528b) (8. b. 28 a.m _ y p ) (m+m_) (8.530a) (8.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 713 yR  m+ m _ ( x p .x Q ) + ( m + y Q .f (y R . The values of p p .R } (xR ~ XQ} The value of UR is obtained as described earlier.f dy = 0 aj 1 m _ d p + a 22 dq .525a. b) now read yRyp = ^(m+p + m+R)(xRxp) (8. by considering other points along C.q p ) .dy dx dy = p dx + q dy (8.527a) (8. we replace the values of m + and m_ evaluated at P in Equations (8. p q .53 la) (8. Equations (8. the solution u in the xyplane can be determined.526b) where m + and m_ denote m + p and m_ p .UP = 2 (pR + ^ *XR ~ XP* + 2 ^ R + q p ^ (YR ~ Yp') (8.526a. b) can be approximated by a! i m + (p R .p p ) + a 22 (q R .531b) yR .y Q ) = 0 (8.q Q ) .529a) (8.y p ) To improve the accuracy.f dy = 0 Equations (8.529b) The values of p and q along PR are assumed to be given by the average values evaluated at P and R.Q + m .p Q ) + a 22 (q R .528a) (8.
b) and is found to be x R = 0. qp = qQ = 0 p p = 0.53.532c.2.534a. The characteristics are given by r.b) (8. In this problem.045.0) respectively as shown in Figure 8.2. = 1 dx + 41 = m_ = .532a.: + #L = m . We choose the points P and Q to be (0. Solve the problem stated in Example 8. y R = 0.2 .4 (8.4 ^ x FIGURE 8.52.53 Numerical solution of the wave equation by the method of characteristics The point (x R . e) and are Up = 0. y R ) is determined from Equations (8.3.d) T : The initial values of u are given along the xaxis which is not a characteristic.i  r_ /W* / * * iZ_ 0 .0) and (0.58d.53.2 _v< .b.1 dx (8. p Q = 0. u Q = 0.714 ADVANCED MATHEMATICS Example 8.1 (8.f) t .e.51 by the method of characteristics.3 Q 1 . and q at P and Q can be computed from Equations (8. .1.533a.1 P 1 .105.533d.526a.b) This can also be seen from Figure 8. The values u. y is replaced by t.c) (8. p .
1 (8. 8. we obtain p R = 0.61 shows an example of such a situation. This is usually the case where the geometry is simple.537) From Table 8. q R = 0.6 IRREGULAR BOUNDARIES AND HIGHER DIMENSIONS So far.536a.b) (8.535b) The approximate value of u R is given approximately by Equation (8. y • S P Q^S __\c X FIGURE 8.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 7L5 Equations (8. we have considered only the case where the grid points lie on the boundaries.535a) (8. we note that this is exactly the value obtained by the finite difference method.51. The values of u at other points in the xtplane can similarly be obtained. it is not always possible to arrange for the grid points to be on the boundaries.3.53Ob) and is found to be u R = 0.61 Irregular boundary .528a. For curved boundaries.535a.075 (8. Figure 8. b). b) reduce to PR°4qR = 0 °2~PRqR = ° On solving Equations (8.
For compactness. 3b).b) /o .k ^ + ^ . yp) = u p h—.6la. h and k refer to the grid size and where 9j and 62 satisfy the inequality 0<91<l.+ — —^ ox 2 ox (8.63a.. dy 2 9y Similarly. y r k) = u p . Let the length of PQ and PR be 0jh and 62k respectively.+ 1 ^ ox 2 dx . we use the suffixes.c . the grid is not rectangular and we cannot use Equation (8.( l + 9 1 ) u p ] (8. we have 2 2 2 u R = u (Xp. x .. to denote that the quantities are evaluated at P. 0<62<l (8.( l + 0 2 ) u p ] k202(l+02) (8 6 6 ) 2[uQ + 0 1 u s . . and u s are given by (8. where.62a.62b.b) 2 u Q = u (Xp + Bjh. UQ . we have 9^u = = 2[u R + 9 2 u T ..b) For points next to the boundary.8u h 3u us = u(xph. ..716 ADVANCED MATHEMATICS We consider Laplace's equation with Dirichlet boundary conditions. Instead we need to deduce the difference form of Laplace's equation at the point P involving the given values of u at R and Q. The value of u is given on the boundary and let Q and R be two points on C. Expanding UR in a Taylor series about P.62b. the partial derivatives are evaluated at P.65a..67) dx2 h 2 e 1 ( i + e1) The difference form of Laplace's equation at P is . we deduce that 8^u dy2 Similarly. as usual. The points P. 5b). Q.61.42).. u T ..b) uT = u (xp. P..64a. (8. Q.~ dy 2 dyz 2 2 (8. y p + 9 2 k) = Up + 9 2 k — + 2 j + . S and T are grid points as shown in Figure 8. From Equations (8. yp) = Up + 0 ^ 5.b) In Equations (8.
t j .k . Some of the methods described in this chapter can be used for nonlinear equations. Laplace's equation in three dimensions for a cubical grid of size h 3 is u i+l.2u i.610) As in the twodimensional case. k + u i.j.k + l. 8.2 u i.( i + e2)Up] = Q (g 6 _g) k 2 e 2 ( i + e2) The difference equation for the other points closest to the boundary are deduced in the same way. Other methods are available and are described in more advanced books.q + e^up] h2el(i + el) + [uR + e 2 u T .k { u i.610) can be solved by the methods described earlier.7 NONLINEAR EQUATIONS One of the advantages of numerical methods is that many of the methods based on linear equations with constant coefficients can be carried over directly to nonlinear equations.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 711 [iiQ + e ^ s .2u i. we deduce that u i.j. Equation (8. j . k . however it involves a larger number of points and the direct method is not recommended.j. the difference equation is complicated and it is preferable to use the finite elements method which will be described later. k + u i.kl = ( ) h2 h2 h2 (8.k.k + u i.69) From Equation (8. For Neumann conditions.j.l .72a) k h2 J .2 u i > i J > JL +U iu] (8 . The difference form of the quasilinear equation 3T = srl D ( x . j .k + u i. j . k) is the average of the values of u at the six neighboring points that surround it.k = ^[ui+l.j.l ] (8. j . U i ) h + l J . ui.k + u i .j.j + l.j.j + l.j.l .k.j.ji.69).k + u il.t i U ) d can be written as Ui 'J + 1 ~ U i ' J (87l) = D( X i .k+l + u i .j. the value of u at the point (i. The derivation of the difference equation for three and higher dimensional problems is straightforward.k + u i .
The equation of continuity and the equation of motion.73) We have evaluated D (x.^ + v — ^ = v — f (8.j)(ui+lo2ui. assuming no pressure gradient and making the usual boundary layer approximations. 1963) . The appropriate boundary conditions are y = 0.71) by ^ = D(x.71. II1) and are 3vY 3v v 21 + ^ =0 dx dy 3vY 3vY 9 vY vx . 34b). y —> oo ? x =0 . Note that Equation (8.76c) (8.71.72a. are obtained form Equations (A. vy = V (a constant).tj. b) are the simplest form of the explicit difference for Equation (8.b) (8.74) Experience has shown that this expectation is justified.j + l = uij+J^D(xi.75a) /n (8.718 ADVANCED MATHEMATICS or u i.76a. v y ).76d) v x —> v^ (free stream velocity). v x = 0 .ui. we expect the numerical scheme to be stable if 4D(x i . Ames (1969) has considered the flow past a solid plate in the region Z < x < 0 and over a porous plate in the region x > 0.72b) is linear. In fact. Comparing Equations (8.j + u il.t j . we have approximated Equation (8. Using the initial and boundary conditions. for x > 0 for x > 0 (8. u) at the grid point (i. The velocity components along the x and y axes are respectively (vx. Equation (8.71).t. Example 8.u)% at dx2 (8.I1.72b) generates values of u at all points for subsequent times. j). t. as shown in Figure 8.75b) dx y dy dy2 where v (=fl/p) is the kinematic viscosity.u i j ) < 1 h z (8. v x = Blasius solution (Rosenhead.j) (8"7"2b) h Equations (8.72b.
79) can be written as 3vY 3vv 3vY 3JL = V L . The components v x and vy can be functions of x and y or functions of x and \/. y). 7a.71 Flow over a solid plate and a porous plate We introduce a stream function \/ (x. b) yields d\\f d2\\f d\J 9 2 \/ 3 3 \i/ = v __v ¥ ^ _ ^ ^ _ (87_8) dy dx dy dx dy dy Equation (8. Equation (8.75a.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 779 y <• l 0 X FIGURE 8. defined by 3\/ d\if (8. y) to (x. Using the chain rule. we can write 3vx Q' y) dx = 5 v x (x> v) + 3vx & v) <?j_ d\\f dx ( 8 7 _ 9) dx Using Equation (8. b. \/).77a.77b).b) Combining Equations (8.78) is a third order equation and it can be reduced to a second order equation by the von Mises transformation which makes a change of variables from (x.710) .v v ^ A dx dx y 3v/ (8.
715) The conditions at y = 0 and y —> °° correspond to \/ = Vx and \/ —> °° respectively and the boundary conditions [Equations (8.713a) (8.720 Similarly 3v Y 3v v 3vi/ = ADVANCED MATHEMATICS 3v Y Ti dy 32vx TiT=vxTi x 3\/ d\]f dy 3 / 3v x \ (8.79 to lie). we obtain respectively fy V= JO v x (x.y)d$ + g(y) (8.£)d^Vx (8.76a.714b) Since \/ is defined to the extent of an arbitrary constant. y) is given by \f = r JO v x (x. ^x dx = VA(V d\/ \ x ^A 3\/j (8712> The boundary conditions also have to be transformed to (x. we obtain (8Jllc) Note that in Equations (8.714a) (8. b). \/) coordinates.76a. b). Imposing the conditions given by Equations (8. c)] are transformed to .713b) ¥ =  Jo where f(x) and g(y) are arbitrary functions.£)d£ + f(x) v (5. b.77a.f Vd^ + g(0) Jo = .b) 7? = W'£) Substituting Equations (8.V x + g(0) (8. Integrating Equations (8.71 la.79 to lie) into Equation (8. we can choose g (0) to be zero and \/ (x. we obtain f(x) = .75b). vx is a function of x and y on the left side and a function of x and \/ on the right side.
719d) (8.718) can be approximated as u i. u = 1.b.76d) remains unchanged. *. r = 0 . vx = 0 v x —> vM (8. we can compute the values of u j .NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 727 V =Vx.716b) Equation (8.716a) (8.c) (8. ^ h2 < 1 2 (8.™ (8 . u = 0 P = V" ^ T / (8. Further details can be found in Ames (1969). Note that Equations (8.721) we have computational stability. As expected. .718). we have divided the region of interest into rectangles each of size (h x k).j + l = U i j + ^ V l .U j j [ui+1J2uiij+ui_1>j] h (8.719a to 20). From Equations (8. y _ > oo.720) In discretizing Equation (8.719a to c) imply that we have irregular (triangular) boundaries and we need to make the modifications proposed in Section 6. 18) *L = /TTIT A The boundary conditions become \l/2 /6 = _ pn . refers to the value of u {/&v x\).712) can now be written as • " = 'fe)2 (3.719a. 7 .719e) u is given by the Blasius solution Equation (8. it is found that if k VlUj • . . The quantity Uj. A —> » .£&) Equation (8. We now introduce the following dimensionless variables 1/2 nJ.
Implicit formulae which are unconditionally stable have also been derived. 0<x<l (8. More recently. it has been widely used to obtain approximate solutions to continuum problems. 1].71). It must be pointed out that one does not usually solve onedimensional problems by this method. We introduce the basic concepts of the finite element method by considering onedimensional problems. Zienkiewicz and Morgan (1983) have shown that a generalized finite element method can be defined which includes all the numerical methods and it is left to the user to choose the optimum method. These formulae can be found in more advanced books on numerical solutions of P.8 FINITE ELEMENTS The finite element method was developed to study the stresses in complex discrete structures. we have only one independent variable and the equation we need to solve is an ordinary differential equation.b) By an appropriate scaling. . The battle between finite differences and finite elements is over. We consider the differential equation .82a. For regular boundaries and simple equations. the finite element method is the obvious choice. Several formulae have been proposed which are more complicated than Equation (8. OneDimensional Problems In a onedimensional problem. the method of finite differences is preferable because finite difference equations are easier to set up.2 ^ L = _g(x).722 ADVANCED MATHEMATICS The accuracy of Equation (8. t:. 8 .D.72b) can be improved by replacing D (Xj. The reason for this choice will become apparent after we have introduced the method. dx 2 subject to u (0) = A.E.81) u (1) = B (8.72b) but which give results closer to the solution of Equation (8. we can always transform any arbitrary interval into an interval of [0. Uj :) by an average value of D over the points considered. The onedimensional case is used to ease the way to the understanding of two and higher dimensional problems. It was pointed out earlier that if the boundary is irregular.
we obtain an expression of the form I = F(a1)a2 g (8. In Chapters 9 and 10. They can be polynomials.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 721 Variational method In the finite difference method. we give examples of such variational formulations. it is shown that the function y (x) that yields the extreme values of the integral I. we look for a function that extremizes an integral.81). u .5) (886) srHnr) .83) ^1 _ A.84) yields ( °. i l 9y dx 9y' Consider the integral =0 (8. They are usually chosen such that the boundary conditions are satisfied. or other functions whose properties are known.g r ^ ^ ( i u H]° We approximate u by m Simplifying Equation (8. Instead of looking for a function that satisfies the differential equation. are constants and the Nj(x) are the shape (trial) functions.85).X i= l aiNiW (88"7) where a.s u ] d x Applying the condition given by Equation (8.86). we replace the derivative by a finite difference. defined by I = I f (x. In Chapter 9. we reformulate the problem into a variational problem. y')dx Ja is (8. This means that the solution of Equation (8.88) . y.85). we obtain Equation (8.81) is the function that extremizes the integral I defined by Equation (8. On substituting Equation (8. trigonometric functions.84) i = / o te) 2 .87) into Equation (8. In the finite element method.
1] into a number of subintervals and each subinterval is called an element.2. we choose the weight function to be the approximate solution (w = u). Galerkin method Not all problems can be formulated as extremum problems. Equation (8.b) Jo In the Galerkin method.81). . On the first interval. then u is an exact solution of Equation (8. If R is identically zero.89). In the method of finite elements.81 la) becomes /"xo I N : [ajN^' + a 2 N 2 ' + g] dx = 0 JO (8. we obtain the coefficients a. We approximate the solution u in the interval 0 < x < l by u(x).812a) . We now describe a method which provides a weak solution to Equation (8. In each element.. and on substituting them into Equation (8. we subdivide the interval [0. we approximate u by Equation (8..87) provide sufficient accuracy. and the extreme values of I are given by dF ^ . The function u generally does not satisfy Equation (8. This method is known as the RayleighRitz method (see Example 9.112).81). 1] into subintervals (elements) and on each element.87). . In the finite element method. We illustrate this method in Example 8. Suppose we subdivide the interval [0. we obtain an approximate solution for Equation (8.81). The function u is a weak solution if the integral of R with respect to a weight function w (x) is zero.1] into [0. x 0 ) and (x 0 .89) From Equation (8.724 ADVANCED MATHEMATICS It can now be considered as a function of a. i = 1.87).81) and we write ~ + g(x) = R dx 2 (8. Since each element is shorter than the whole interval.810) where R is the residual. That is to say C Jo Td2~ 1 w(x) —^ + g(x) dx = Ldx2 J (l w(x)R(x)dx = 0 (8.m (8. 1] and assume that two terms in Equation (8.87).87). we can expect that a better accuracy can be obtained for the same number of terms in Equation (8. u is approximated by Equation (8.81 la.81. we subdivide the interval [0.= 0.
81.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS Z25 I N j ^ N ' i + a2N2' +g] dx = 0 JO Integrating Equations (8. have been obtained. Gj = gNjdx (8.8 . the approximate solution u is given by Equation (8. b) by parts yields a l i N l N l J o +a2LNiN2J0 (8.15a ) Kjj = I NjNjdx.87).814a.813a) (8. we obtain another set of equations..c) gNjdx G2 = gN2dx (8. subject to the boundary and continuity conditions. m) and we obtain iKijlNiNjJxk where aj = G J i ( 8 . we have avoided having to evaluate the second derivative of the shape functions. . The coefficients aj can be determined by assembling Equation (8.812b) aiKlla2K12 alK12a2K22 + Gl = ° = ° (8. b) by parts. The shape functions Nj are usually chosen such that K can be evaluated easily (analytically or numerically).812a.815a) for all the elements. and well conditioned. This implies that we can choose linear functions as shape functions.814d.e) 0 Jo Similarly.813a. 1].b. Note that.815b. Solve the equation . We can generalize Equations (8. on integrating Equations (8. b) for any interval x^ < x < x^+i and for any number of shape functions Nj (i = 1.. . is sparse.ixo + G2 a U N 2 N l J o + a 2L N 2 N 2Jo Kn = JO Nj dx 'Xf) K 12 = I N x N 2 d x Jo [X0 K 22 = Jo N 2 dx (8.813b) r where jxo r .c) Ak J\k The quantities Kjj are the elements of the stiffness matrix K and Gj are the components of the force vector G . for the interval (x 0 . Example 8.812a. Once a.
u is approximated respectively by u « 3ujX u » U1(23x) + u 2 ( 3 x .c) by the method of finite elements. u(l) = 1 (8. Equation (8.817b) (8.817c) Note that in Equations (8. (1/3.817a to c).818) yields I = I j + ^ + Ig (8.818) Substituting Equations (8. We approximate the function u by a piecewise linear function. and E 3 . E2. 2/3). we obtain <•«»> (8. We denote the elements by Ej.l ) u « 3 u 2 ( l .816a) is equivalent to finding the extremum of I = I l f d u ) 2 _ x u dx (8.> 2 £^ The extremum of I is given by ^= j ^ .b) . and E 3 and the values of u at the nodal points Xj by Uj. E2. We solve the problem by the variational method.816b.x ) + (3x2) (8.817a to c) into Equation (8.726 2 ADVANCED MATHEMATICS — + x = 0 dx 2 subject to the conditions u(0) = 0.816a) (8.819a) •(¥£Mi<^££Mi<>. 1/3).820a.= 0 On carrying out the differentiation. 1] into three elements of equal length [0. We subdivide the interval [0. In E 1 . u satisfies the boundary and the continuity conditions at the nodal points (Problem 14a shows how these functions are derived). and (2/3. 1].817a) (8.
824a) (8. We can also consider rectangular or other geometrical shapes.822a. j . it is zero. the nodal values are the exact values.728 The exact solution is u = *(7x2) o (8. We choose the shape function Nj such that at node i. Nj (Xj.821b) (8.824a to d) yields 1 xi yj 1 T a i l yj I" i " = 0 (8825) 1 1 Xj x bi k ykJLciJ L o  Solving Equation (8. yj) = 0 .822a) (8.d) Combining Equations (8. Nj (x. We propose linear shape functions. we subdivide the domain into elements and we consider triangular elements.c. TwoDimensional Problems In twodimensional problems.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 72Z 6uj3u 2 = 1/9 3uj + 6u2 = 29/9 The solution is u2 = 62/162 = 0. the two coincide. yk) = 0 (8. Triangular elements allow for an easier fit to nonrectangular domains. y) has the following properties Nj(x.822b) (8. and at the nodal points. y) = aj + biX + q y Ni (xj.82 la) (8. it is one and at nodes j and k. Consider a typical element E with nodes i. Strang and Fix (1973) have observed that in general. we obtain . we note that the solution we have obtained by the finite element method is exact! The exact solution is cubic. yi) = 1 .823) By comparing Equations (8. the approximate solution is linear. if the shape functions satisfy exactly the homogeneous differential equation.81. 23).824b.383 u 2 = 59/81 = 0. This is not generally true. That is.81.825). This is shown in Figure 8. b. N{ (xk. and k as shown in Figure 8.
y k ) / D .827) Domain subdivided into triangular elements. y) dx2 dy2 u = 0 on the boundary (8..728 ADVANCED MATHEMATICS a{ = (x j y k x k yj)/D. we generalize the variational and the Galerkin methods to twodimensional problems.827) i where Uj is the value of u at node i.f) y o FIGURE 8 81 x (8.828a) (8.828a) We consider Poisson's equation which can be written as + = .c) D = l 1 yi y?k = 2 (area of element E) = 2 A (8. (8. E is a typical element with nodes i.e.828b) . then we set Nj to be zero. Next. j .824a to e) and if i does not belong to E. we define a shape function Nj given by Equations (8.826d.b. and k For each element E. ci = (x k x j )/D (8. 1 xi x xk bj = ty . we can approximate u as u = £ Uj Nj (8. With this definition of N.826a.g ( x .
831c) (8.83 lb) I =uTKuuTG Gj = I I gNj dxdy The extreme values of I are given by ^~ = 0 Combining Equations (8.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 722 The function u that satisfies Equation (8.831a) (8.832) (8.830a) /3NI i 3 N I5NI) 3N2 3Ni 8N 2 \ + 2uM^^ + ^F^yL)"(UlgNl + " + U m g N m ) ] d x d y (O"30b) Equation (8. + u m N m ) ] dx dy (( fl 2 P N i 3N (8..828a) is the function that extremizes the integral [see Equation (9.g (u1Nl + .833a) .83la.1113)] (8.831d) (8.827) into Equation (8..830b) can be written in a compact form as I =XKijuiujXuiGi or where (8.829) Substituting Equation (8.829) yields . 32) yields ^KjjUjGi j = 0 (8.
w s (t\l 9N. This is true for all symmetric problems. the matrix K is no longer symmetric. 24a). d).836b) In the usual compact form.837) where Kjj and Gj are defined by Equations (8.833b) To obtain a weak solution u (x.g N i d x d y =0 (8.838a) . such that w(x. By combining Equations (8.834) On integrating by parts and assuming that w vanishes on the boundary.G = 0 (8.827) and using Galerkin's method (setting w = Nj in turn). we find that for each element E.833b. 3Nj 3N. aNj\ /3N m 9N3N_ dNA XT 3N m \8Nj / 3N. Equation (8. y).+ U m (^r^r + ^V). we define a weight function w (x.+u ^)^ + hV" + " +Um ^f)^" sNi J dxdy = 0 (8. y) dxdy = 0 (8. 5N m \aN: XT1J J . y ) p 4 + ^ .836b) can be written as XKJJUJGJ = j o (8. Equation (8.83lc.4 + g(x. y).834) becomes It Idw du dw du .831c. Kjj is given by Kij = J / [ b i b j + c i c j] d x dy (8. JJ (ara^ + 3757. In this case. the variational method may not be applicable but the Galerkin method can still be used. .835) can be written as JJ [( u ^ + ". For nonsymmetric problems. 37)]. A or )) u i ( ^ ^ + ^ V ) + . Equation (8. Note that in this case both the variational and the Galerkin methods yield the same discretized equations [Equations (8.836a) (( /3N.730 ADVANCED MATHEMATICS or Ku . dxdy = 0 (88'35) The function u is chosen to be given by Equation (8.
839) which simplifies the computation of Gj.g (c ) [aj + bjx + cjy ] A «^g(c)A (8. 42c).842b). Similarly.g ( c ) [aj + bjx + Cjy ] I I dx dy (8.840a. we can obtain a sparse matrix K. x and y can be approximated as the average values at the nodes. then g can be approximated by its values at the centre of mass of the triangular element and we denote it by g(c).b) . Example 8.82. Thus using Equations (8.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 731 = [bjbj + c j Cj] (area of the element) = Afbjbj + CjCj] Using Equation (8. and G . x and y are approximated as x « x = i.(xj + Xj + x k ) y « y = j (Yi + yj + y k ) The approximate value of Gj is thus G . K is a (3 x 3) matrix for each element.842a) (8.84ia. is a vector with m components. Equation (8. y We note that since each triangular element has three nodes.838c.826a to c) into Equation (8.824a).842b) (8.842c) is obtained by substituting Equations (8. . That is to say. We then have to assemble them to obtain the global K. then the global matrix K. If the domain has m nodes.838b) (8. ^ is a (m * m) matrix and the global G / > . Solve the equation .838c) = I I g ^ j + bjx + cjy] dxdy (8.b) (8. we can compute Kjj and gj for each element E.831d) can be written as Gi (8.842c) Equation (8. and G is a vector with three components. By numbering the nodes appropriately. •>. If the triangular element is small enough.
2 ( 1 . The area of each triangular element is 0. and 6(2.843) subject to the condition u = 0 on the boundary (8. We subdivide this reduced domain in four triangles and label the elements and nodes as shown in Figure 8. 5 ( 2 . a3 = 0 .5. 0 ) .0) A (2.b. 1 ) .h. 4 ( 2 . b3 = 0 .i) . 2<y<2 (8. 1 ) . = . 0 ) .f) (8.2) y ' J2. we may limit our calculations to one octant only.844) From the symmetry of the problem.1 .e. q =0 c2 = . W = 1 .845d.712 ADVANCED MATHEMATICS ^Jl+iH. (0. Node 1 is the origin of the coordinate system.c) (8. b2=l.0) X FIGURE 8.l c3=l (8.845g. Ej: aj = 1 .82.2) 3/ 5 (2.1) E3 / . 3x 2 dy2 2<x<2.0) / (1. 42c) to calculate K and G for each element starting with element Ej. 38c.826a to c. a2 = 0 . The coordinates of the nodes are 1(0.82 Division of a domain in triangular elements We now use Equations (8.845a.0). 3 ( 1 .
b4=l.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 733.f) (8.b.The result is ~ 1 2 1 0 _1 2 2 1 0 1 2 1 I I 0 0 1 1 0 0 0 0 l 0 fuil u2 u3 1 3 3 _1 6 0 0 1 0 0 1 I I 1 1 1 2 0 1 u4 u5 1 3 (8.c) G 9 = G.c) G 9 = G 4 = G.1 J [u 6 j _ 1_ .c) G3 = G5 = G6 = 1 We assemble the K and G from each element so as to obtain the global K/ ^ and G( g ).l .852a. = G.849a to f) (8.l . K36 = 0 . we have E3: K22 = i .853atof) (8.b. K35 = .855) 0 0 0 I 0 .c) Gl = G2 = G3 = ^ E^ a2 = 2 . K55 = l (8. K25 = .854a.851atof) (8.l . K 22 b2 = . for E 3 and E4.b. a5 = 0 .l c5 = 0 = ^> K (8.e. K 35 = . K 24 c2 = 0 c4 = .848d. b 5 = 1 .b.85Oa.l . K56 = .l .l .848a.b.847a. = L ^ J J 6 E4: K33 = l .h. = 1 J 6 Similarly.1 .848g.i) K 45 = ^> =. K66 = l (8.^ ' K 25 44=l' = " £• K 55 =^ (8. K ll=2' K 12 = ~i' K13=0> K 22=1' K 23=^' K 33 =^ (8.c) (8. K55 = 1 . K33 = l .846a to f) (8. a4 = . K23=l.
855) now reduces to "i 2 1 0 _i 2 2 1 o i fuji 1 2 u2 + u3 To 1 0 o 0 1 o i ru4i 0 0 u5 u6 =1 ri 3 3 (8. K( x and G /gx are being assembled as the calculation for each element proceeds. they are transferred to K/ ^ and G / y. 0<x<l u(0. PROBLEMS 1 a. and u 3 which can be determined by partitioning the matrix K. and u 3 .856) that "uji u2 = r i 2 . The only unknowns are U p u 2 .b) ui\ I ° 1 2 J L i l [%_ Note that the matrix in Equation (8. x into four (3 * 3) matrices. and u 6 are the values of u on the boundary and are known. Equation (8. u 2 .0) = x 2 . K and G are calculated for each element. in Equation (8.1 _i 2 2 o 1 T 1 rii 6 1 = [ i~ 4 11 (8. since we require only three equations to solve for U j . . u and G ^ <. Strang and Fix (1973) have discussed the computational aspects of finite element methods.t) = 0 . We choose to consider the upper half of K.857 a) is nonsingular (its inverse exists) and U j . u(l. . Use the explicit method to solve the initialboundary value problem defined by ^ = ^ 4 . The vectors u and G . u 5 . we deduce from Equation (8.857a. The components u 4 . s are each divided into two vectors each with three components as shown by the dotted lines in Equation (8. . u 2 . at ax 2 u(x.855). and u 3 can be determined.05 so as to ensure the validity of the method.1 and k = 0.856) Using the boundary conditions (u 4 = u 5 = u 6 = 0).855) is singular and this implies that the uj cannot be determined. Thus K and G for each element need not be stored. that is to say.234 ADVANCED MATHEMATICS Note that the matrix K. t > 0 .t) = 1 Choose h = 0.
5. t) = I . ^ 5x =0 x=l Obtain c for D = 10 and calculate the average concentration (c) defined by Equation (5.721. 0) = I . By introducing the following dimensionless variables c = (ccf)/(cjcf).^  . 0<x<5. 0) = I . 0 < x < 5 .NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 735 2b.(u + cos u ) . Use the CrankNicolson method to solve the problem examined in Example 5. 3z dx c (0. z) = 0 . z = z/L show that Equations (5. 0<x<l.. Use the finite difference method to solve the following elliptic equation d u + 9 u y 3u x 9u u = 0. D = J90B/(Lvz) c (x. at ax2 u(x.t) = I . 22 a to c) become . at 9x 2 subject to the following conditions u(x. 0<t<I u(0. 4b. t>o u(l. Solve the nonlinear parabolic partial differential equation 9U 3 U .72. 3a. ^ 5x =0 x=0 by the method of CrankNicolson. 0<y<l ax 2 ay2 ax ay u = x . .0) = 0 .736a) by numerical integration. x n e rx 1 — = — .= D . t) = u(5. x = x/L.y on the boundary of the square. Choose h = k = 0. 5a. 0 < t < I Solve the following initialboundary problem ^ = ^JL. 0 <x<i.
! . Obtain a central difference approximation to the biharmonic equation :J4 }4 . 7b. an implicit formula for Equation (8.l . Use the finite difference method to solve the hyperbolic equation ^ = % 0<x<l.l j . j + 2 [1 + (h/ck) 2 ] U J J . The Laplacian is 32u 3r2  1 du r 3r 1 32u r 2 dQ2 9a.^ i . Solve Poisson's equation ^4 dx2 + ^ i =2. Is it possible to choose k > h? 10b. j) and (i.O. u(x.5la) is ui_1.j + 1 .t) = O . u(l.O) = x . t>0 at2 subject to ax2 u(O. Choose h = k = 2." i + i j _ i . 0<y<8 u = 0 on the boundary (i) by direct elimination and (ii) by the S.4 (h/ck) 2 u i .2 [ l + ( h / c k ) 2 ] u i j + 1 + u i + l j + 1 = . 3u + JJL =o dx2dy2 3y4 Derive a finite difference approximation to Laplace's equation in the polar coordinate system.4 di + 2 dx4 8b.t)=l. Show that by approximating (9 u / 3 x ) at (i. j + 1 ) .R. methods. and G O = 1. dy2 0<x<6.2. ^ a t = 0 t=0 Choose h = k = 0.736 ADVANCED MATHEMATICS 6a. j) by its mean value at (i.383.
4.+ a 0 (x)u = r(x) dx 2 u (a) = a . t ) = t ( l . t) satisfies the following conditions 92u  92u 92u = 1 3x2 dxdt dt2 ^dt u(x. k = 1/8. .5. I fdi) .a n u 2 + 2ru dx L^dx/ u Solve the boundary value problem ^ = 3x+l dx 2 u(0) = u(l) = 0 (i) by the RayleighRitz method and (ii) by the Galerkin method. (0. t>0 at2 ax 2 u(x. Answer: 0. 0.O) = x ( l .NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 737 Use the implicit formula to solve ^ = % 0<x<l. u (b) = p is given by the extremum of the integral I which is defined by /•b r I = 13 a.02 12a. 0). Ja.t ) .01.0) = 0. 0. 0). lib. =1 t=o The function u (x.t) = u ( l . ^ dt u(0.01. =x(lx) t=0 Use the method of characteristics to find u at the intersection of the characteristics through (0. Show that the solution to the following boundary value problem 2 d ^.6. 0) and (0.x ) . Choose h = 1/4 .
y) on a square plate of size 2m x 2m has reached a steady state and satisfies the equation ^ . one deduces that pj (x) can be written as Pi 00 = Ui + Cuj+jUjXxXjJ/h in the interval [XJ. The sides y = 0 and y = 2 are insulated (3u/3y = 0). Note that u is approximated by a linear function pi (x) and. d!u +0 . from Section 7. Verify that 1 . Answer: 5 x ( x . + u2 = lf o<x<rc/2 u(7i/2) = 0 by the method of finite elements. Compare the approximate solutions with the exact solution. Impose the boundary conditions and obtain the required results. Compare the values of u (1 /3)) and u (2/3) with those obtained in Problem 13a. + ^ = 0 . 17a. 0<y<2 The sides x = 0 and x = 2 are kept at 0°C and 100°C respectively. 15a.2 ) / 2 14a. Use the method of finite elements to find u at the points (1. Deduce Equations (8.817a. x(x 2 + x . Compare the analytical solution with the numerical one. Consider three elements and approximate u by a piecewise linear function. (1. Answer: 50°C . Solve the nonlinear system 16b. b. 8y2 9x2 0<x<2. x ^ ] .738 ADVANCED MATHEMATICS Use a quadratic in x as the approximating function.sin x is a solution of the boundary value problem.2).0). The temperature u (x. c). 5 [(djM 2 Hdx' dx2 u(0) = 1 . Solve the boundary value problem defined in Problem 13a by the method of finite elements.6.l ) / 4 .1) and (1. Divide the domain into five elements and choose linear shape functions.
y (b)) respectively. where its value is 0. and for computing approximate solutions of differential equations.11. In differential calculus we learned to determine the extreme values (maxima or minima) of a function. the function f (x) = x 2 has a minimum at x = 0. If the coordinates of A and B are (a. Calculus of variations is an extension of the above concept. (9'» y(b) y(a) ^ ^ . y (a)) and (b.11 Curves joining two points A and B . a J *s/Uy b x FIGURE 9. For example.1 INTRODUCTION Calculus of variations was introduced by Johann Bernouilli two centuries ago and is now widely used in optimization. control theory. f (x) is greater than zero everywhere else.CHAPTER 9 CALCULUS OF VARIATIONS 9. Suppose we wish to determine the shortest plane curve joining two points A and B in the xyplane as shown in Figure 9. then the length s of the curve y = y (x) joining A and B is given by s=f J a V r + (l) 2 d x y .
11) is referred to as a functional. where ( ' ) denotes differentiation with respect to x and Ax = x . In variational calculus we need to determine a function..11 shows two such curves. we assign a number s to each function y (x) and the rule refers to an integration with respect to x from a to b. irrespective of the sign of Ax is f'(x o ) = O (9. Hence Q represents an extremum.x 0 . we review the theory of extreme values in differential calculus. Figure 9.. Figure 9. We need to determine the function r B that will maximize the integral. The reaction rates rA. we wish to find the extreme values of a functional.22a.21) retains the same sign for all small values of Ax. 9. Expanding the right side of Equation (9. It can be seen from Equation (9. we can obtain the corresponding value of s and we have to determine the curve y(x) which yields the minimum value of s.21 illustrates this for the situations where x 0 is at P (both Af and Ax change sign) or at Q (only Ax changes sign). In kinetics we can consider the reactions »B — > C.21) about x 0 yields f (x) . rB and r c may be functions of temperature and concentration. In this example. irrespective of the sign of Ax. The integral on the right side of Equation (9.b) . tj) is given by P 1= JO r B dt (9.2 FUNCTION OF ONE VARIABLE In Chapter 1.11).f (x0) = Af = f' (x 0 ) Ax + 1 f "(x 0 ) (Ax)2 + .23) (9. where a functional can loosely be defined as a rule which assigns a real number to a function. In engineering we encounter similar variational problems..12) Equation (9. A— We may wish to determine the reaction temperature that will produce the maximum amount of B in a fixed time interval from 0 to tj.12) is similar to Equation (9. such that an integral is an extremum. The production I of B in the time interval (0. Prior to considering variational calculus.22) that the condition for Af to preserve its sign. y 0 and y.740 ADVANCED MATHEMATICS For each curve y (x). if it exists. That is to say. it is stated that a function of one variable f(x) has an extreme value in the neighborhood of x 0 if Af given by Af = f(x o + Ax)f(x o ) (9.
21 1 Q 1 P X Extremum for a function of one variable Thus the condition for point x 0 to give an extremum of f is given by Equation (9.l ) n + 1 (9..23).. (9. Example 9.d) n = 0.2. On differentiating.b. Q represents a minimum since Af is always positive. It follows from Equation (9.27a. 1. cos x = 0.22) that f(x 0 ) is a maximum if f"(x 0 ) is negative and f(x 0 ) is a minimum if f "(x 0 ) is positive.26) These extreme values are maxima when n is even and minima when n is odd. so that x n = (2n + 1) 7C/2 . that is.24. Find the extreme values of f (x) = sin x. Thus x o = 7 t / 2 is maximum (sin %I2 = 1) and Xj = 3rc/2 is a minimum (sin 37t/2 = 1).21. At these points f "(x n ) = sin x n = sin (2n + 1) 7t/2 = sin %I2 cos tin = (.21. . . 5) Thus the extreme values of sin x are at f' (x) = 0.CALCULUS OF VARIATIONS 741 y  V \ »«/!*y ayl/jax i FIGURE 9.c. In Figure 9. we have f'(x) = cos x and f "(x) = sin x (9.
.742 ADVANCED MATHEMATICS 9.35) .. we obtain (9. i = 1.x 2 + Ax2.. x2..32). ... n n n ^2rdxidxj (9. 1=1 J = l 1 J In Equation (9.31). xn + Ax n )f(x 1 . .. we deduce the condition for f to have an extremum at x^ = Xj is p _ = 0 ..2.. x 2 . xn if Af defined as Af = f (Xj +Ax 1 . The conditions for this are D1=^>0 0X1 (9...32.31) Af=S^Ax1 + l i S ^ A X i A x j 1=1 l + . . xn then f is said to have an extreme value in the neighborhood of Xj. xn) preserves its sign for all Axi. i = 1..33) The function f will be a minimum if the quadratic expression ^ ^ 5—^— Axj AXJ is i=l j=l positive definite. (9. for all i= 1. ...3 FUNCTION OF SEVERAL VARIABLES If f is a function of several variables x^.. . From Equation (9. n. x 2 . all the derivatives are to be evaluated at Xj = Xj.32). .. Expanding the right side of Equation (9... .34) A D2= 1 32f Bx^xj _^f_ \ _^f 3x2 2 >0 (9. n.
The conditions for f to have a maximum at x. we obtain i!=2x1+x2. The solutions are x^x^O. is that the quadratic expression must be negative definite.37a.^ .CALCULUS OF VARIATIONS 743 3^f 32f 92f a2f 3x^x2 Dn= .b) So the extreme value of f is at the origin.b. x 2 = x2 = 0 (9. When the quadratic expression is not definite we have to use ad hoc methods or consider higher order terms to determine the nature of the extrema. 3x2 ^ = 2 . D2 > 0 . that is Dj < 0 . Example 9.c) and so on with the sign of the determinant alternating. . At the origin.310a.. a2^ ax2 a2f ax2axn >0 (9.38) — =2.b. (9. dx\ .d) 2x2 + x 1 = 0 (9.c. D3 < 0 . Find the extreme values of f (x 1> x 2 ) = x 2 + x 2 + x 1 x 2 On differentiating..= 1 dxxdx2 (9.31. 3xj ^=2x2 3x2 + Xl. (9.31 la. both Dj and D 2 are positive and f has a minimum at the origin.39atoe) The extreme values of f are at 2x 1 + x 2 = 0 .36) a2f axjaxn a2! ax2 All the derivatives are to be evaluated at Xj = Xj.
32.d) The solutions are x1=0.b) L 2 = 0 => 2x2 + xf = 0 (9. dxj dx2 = 2x 2 + x 2 . Af is given approximately by Af = (Ax2)2 + (AXl)2 (Ax2) + ( A x / = [ Ax2 + 1 ( A x / ] 2 +  ( A x i ) 4 and Af will always be positive and the origin is a minimum.244 ADVANCED MATHEMATICS Example 9.314c. and so we need to investigate further to determine if the origin is a maximum. 3X2 ^L.316a.315a.33.x 2 ) = x^ + xfx 2 + xj The partial derivatives of f are = 2xjx 2 + 4x].= l 3x^x2 (9.314a. we obtain (9. x2 = 0 (9.x2) = x1x2 On differentiating. a minimum or neither. Example 9.312) ^4=2. Both Dj = D 2 = 0 at the origin.313d.b) f(x1.3lSatoe) .b. Find the extreme values of f(x 1 .e) The extreme values of f are given by — L = 0 => 2 X l x 2 + 4xf = 0 OXi (9.b) Thus f has its extreme value at the origin.317) ^=x 3xj 2 . 9x 2 ^ = 0 . Find the extreme values of (9. 3x ^f=2Xl 3x^x2 (9.c) (9.313a. 9X2 = 2x2 + 12x2 (9. 9X2 ^ = 0 . ^=x1.
.. x 2 . X) = f (x1? x 2 .n (9. m (9.43b) dX From the (n + 1) equations [Equations (9.4 CONSTRAINED EXTREMA AND LAGRANGE MULTIPLIERS We have assumed earlier that Xj. . xn) subject to the condition g ( X l . . . x2. and the constraints are given by g j ( x 1 .. If we have m constraints. Thus if Xj represents the length of a rod and x 2 is the temperature. In practice.. There might be a relationship between the Xj.43a. then x^ is a function of x2. it is a maximum along one line and a minimum along another line. . instead of one.. x 2 .. .. The Lagrangian becomes m L = f (xj. Xj and x 2 are not completely independent...44) (9. 2 .. x n ) j=l j = l .x n ) = 0 We now consider the function L. x n . if Xj is the cost of a product. It is preferable to use the method of Lagrange multipliers. 2 . . The extremum of L is then given by ^=0. we can solve for Xj = Xj and X.. x 2 . . in the neighborhood of the origin. . .. .CALCULUS OF VARIATIONS 745 The extreme value of f is at the origin where Dj and f are zero and the quadratic expression is not definite. .... sometimes called a Lagrangian. .. In theory it is possible to eliminate all dependent variables and retain only the independent variables. . .. . .. xn are all independent and can take any arbitrary value..42) (9. xn) + Xg (xj. this can be very cumbersome and the process of elimination can be very time consuming. The origin is a saddle point. For example. . .41) — =0 (9. and is given by L (xj.. x 2 . x 2 .43a) (9. . Suppose we wish to find the extremum of f (xj. This might not necessarily be the case.b)]. then Xj cannot be negative. xn) + ]T. 9..x n ) = 0 . .. i = l . Xj gj (x lf x 2 . f is greater than zero if Xj and x 2 are of the same sign and less than zero if Xj and x 2 are of opposite sign.45) ... x n ) where X is the Lagrange multiplier. x 2 .
V2 f = . The equation of constraint is g(x 1 . .[l (9.b.46b) From the (n + m) equations. . dx2 (9..49) (9..2. lla..b) . we are required to find the maximum or minimum of the rectangle inscribed in a circle of unit radius. x 2 .46a) — =0.48) In this example. we obtain X!=^r.x 2 ) = xJ + x  .. xn) are identically equal to zero.2. the extremum of L is the extremum of f and the Lagrangian has been introduced to ensure that the conditions of constraints are automatically satisfied.41 lc). . .x 2 ) = x 1 x 2 subject to (9.41 la.412) in Equation (9..413a.410) — =Xj + x 2 .746 ADVANCED MATHEMATICS The extremum of L is then given by ^ =0. Example 9.c) From Equations (9.46a.412) (9. Since the g: (xj.l ) =— = x 2 + 2?ix1 . Find the extremum of f (x 1 .47) xf + x\ = 1 (9. .l = 0 dX (9. j = l.41.. b).m (9.l = 0 The Lagrangian and its derivative are L = x 1 x 2 + X(x^ + x ^ . dxl ^— =xl+ 2Xx2 .:. . we can solve for Xj and A. we deduce that x1 = x 2 Substituting Equation (9. i = 1.n (9.
Substituting Equation (9.5 EULERLAGRANGE EQUATIONS We now deduce the conditions for the functional (integral) I given by 1= I" f(x. the point (x1? x2) has to be on the circle of unit radius with the centre located at the origin.CALCULUS OF VARIATIONS 747 Note that.53) Integrating the second term by parts yields /•b . On expanding the function f [x.55). in this example. (9.54) into Equation (9.y' + eTi(x)]dxl f[x. (Xj. x 2 ) can be any point and thus f = XjX2 has a saddle point at the origin. The points a and b are fixed and we also assume y (a) and y (b) to be fixed.54) Since y (a) and y (b) are fixed. y')dx (9. according to the Taylor expansion [Equation (1. we obtain ib AI = I eri ^.dx (9. The increment AI is given by ib rb AI = I f[x. y^(x)]dx Ja Ja (9.52) where e is a small quantity and r\ is an arbitrary continuous function of x.77)] and substituting into Equation (9. we deduce that AI preserves its sign for all arbitrary functions erj (x) if . y^ + er'(x)] in powers of e. r\ (a) = r\ (b) = 0.y o + ETi(x). we obtain . (Xj.33. fb en' — dx = [ETI^1  erijU^. We can consider a neighboring curve yj = yQ + er (x) as shown in Figure 9.51) to have an extreme value.53).52). /a y. Let the function y 0 (x) represent an extreme value for I.b AI= [eT] K^.a L ^y dx \dy' ]dx + O(82) . yo(x).+ eif . x2) cannot be the origin.55) From Equation (9.11.±IK . 9. y o + er(x). In the present problem.^ + O(e2) dx (9. In Example 9.
The extremum of the integral (h Ja is given by 41 de 1= f(x. We recall that y^ = y 0 + £T) and.56) by considering I to be a function of e.5. we obtain yj = y 0 • We now denote y 0 by y.b) .748. y o + eri.5.510a. We can also deduce Equation (9. on setting e = 0.57a) =0 e=0 (9. Equation (9.8b) = fb[iL^i+iL?yi]dx J a [By! 9e \nJ*L ia [ a yi dyj 3e (9.9) On again integrating I dx by parts. we obtain A dl = f J*L _ A f i L \ l dx = 0 de e=o Ja [a y dx (ay1 jj (9.56) is the EulerLagrange equation and gives the condition for I to have an extremum.58c) becomes rb il e=0 = ^^1 + ^iLldx Ja [ ^y a y ' T)1 a y' (9.56) Equation (9.58c) = + T 1 '_§Ll d x 3y.yo + eTi')dx (9.57b) S = /a d t [ f (x> y ° + eT1' y ° + e T 1 ' ) ] dx (95'8a) (9. ADVANCED MATHEMATICS £± dy dx ^ dy' t = 0 (9.
If y does not satisfy Equations (9.511) reduces to  = 0 (9. y = y (b) (9. b)]. y is not an admissible solution and the functional I does not have an extreme value. dx =0 dxdy' dy' a y 8y' Hint: Note that can be written as g [x.fte 1^ (9.. Example 9. y' (x)] and compute Thus to determine the extremum of the functional I. at x = b.512a. y' of f are present and Equation (9. we need to solve a secondorder differential equation [Equation (9. a2f y .CALCULUS OF VARIATIONS 749 Thus the condition for ^ d£ e=o = 0 for all r\ (x) is also given by Equation (9.56).62). b). We now consider these special cases. f is a function of x and y only and it follows % =0 (9.61. 9. y. subject to the boundary conditions at x = a.511)]. The variation ~ d£ is also known as the Gateaux variation. a2f Oy') 2 n .56).511) is then simplified. y (x).512a.6 SPECIAL CASES Explicitly Function f Does Not Depend on y' In this case. Find the extreme values of . y = y (a). not all the arguments x. The solution y = y (x) has to satisfy the boundary conditions [Equations (9.62) From Equation (9. we obtain the relation y = y (x).511) ^. we have af By a2f y .512a. Expanding Equation (9.61) ay Equation (9. b) In practice.
it satisfies Equation (9.1719 Comparing Equations (9. suppose we have a straight line joining the points (1.63) The boundary conditions are (a) (b) y(l) = l.67e. which yields 2xyl=0 (9.64a) (9.66) [y = (l/2x)] does not satisfy Equation (9.2 1 = 1= [XJ Jl I 4x2 Udx = 2x) Ji L dx 4x (9. The extremum of I is then given by Equation (9.e) = . The equation of such a line is y = l(3x) Substituting this y into I.x ) 2 .b. 1/4). 9d).62). y(l)= 1/2. I has no extremum.67d.64a) and. However. it is seen that ^ is a minimum.66) (9.750 ADVANCED MATHEMATICS I" 1=1 2 (xy z y)dx (9. in this case. y(2)=l y(2) = l/4 (9.1733 For the purpose of comparison.64b) and y = (l/2x) gives the extreme value of I.67a.2 .c) (9.68) f2 I = I2 = 1 { ^ ( 3 . (9.c.1 i n 2 = 0.b. In this case .l (3x) jdx = .d) .1 1 =0.65) The solution of Equation (9. 1/2) to (2.64b) In this problem f = xy2y which is not an explicit function of y1. we obtain (9.69a.
the function y = (l/2x) minimizes I.b) I /T^2? where c 2 is a constant.615a.c ? x 2 + c 2 (9.512a. f = dx (9.12) y. Of all the functions that satisfy Equation (9. Equation (9.J .611) dy' From Equation (9.614) Vlc 2 x 2 Integrating yields y = I C]X dx = . we obtain y' = hi (9.CALCULUS OF VARIATIONS 757 This means that there is no function (curve) y (x) that extremizes I and satisfies Equation (9. here we always have a solution. Cl .b) ^ x Solving for y'.611). we introduce two arbitrary constants which are determined by the end conditions [Equations (9. we can solve for y' and y' can be further integrated to obtain y.62.613a.V1 . b)]. Unlike the previous case. Since we have integrated twice. Function f Does Not Depend on y Explicitly In this case.611) yields \ V 1 + ^—^ (v')2— 1 is not an explicit function of xVl+(y')2 (9.64b).64a).6.56) simplifies to f(!M = o dx \dy J (96io) which on integrating yields = constant = Cj (9. Equation (9. Example 9. Find the extremum of i= C H ^ L In this example.
As usual.622). If the curve passes through the origin (0.618) (9.752 ADVANCED MATHEMATICS We may rewrite Equation (9.c 2 x 2 ) (9.y l ^ r l = o L dy The solution of the above equation is f .= constant = Cj dy' (9. Function f Does Not Depend on x Explicitly Since f is not an explicit function of x (9.621) (9.616) Cj=c2=l Substituting the values of Cj and c 2 into Equation (9. 1). 0) and the point (1. 1).b) I"0 Equation (9.620) by y1. we find that the resulting expression is an exact differential which may be written as f. The constants c1 and c 2 are determined by the boundary conditions.[ f .616). a2f .c 2 ) 2 = ( l .y —T7 = ° 8ydy 0y')2 ( 9 6 19) .620) Multiplying Equation (9. we can solve for y' and this can be integrated so as to obtain the required y as a function of x.616) and this represents a family of circles with centres along the yaxis.622) From Equation (9.615b) as c2 ( y .511) simplifies to df By .^ . we obtain from Equation (9. we obtain (yl)2 + x2=l which is the equation of unit circle with centre at (0.y ' .617a.. the two integration constants are determined from the boundary conditions. . a2f _ y T ^ T .n£*ns (9.
Find the extremum of . while falling under gravity.26) ^(l+(y')2)1/2yl/2 V2g il. Integrating y' yields (9.625a.630) .b) Jo Since f = v ^ Jo V 7 (9.(y1)2 0 + (y')2)~m y~m = Cl (9.6. 624) t = fa ds = J L f3 Vl+(y') 2 dx (9. we have to determine the path that a particle will take. so that the time taken is a minimum. It is usually accepted that this problem marks the beginning of calculus of variations. If we take one of the points to be the origin of the Cartesian coordinate system.11. Equation (9. x the horizontal axis. using Equations (9. b) is. Then the time t needed for the particle to fall from the origin (0. we can then solve for y'. ^ f Vl + (y')2 dx (9. we obtain I =ciy1/2(l+(y')2)1/2 Squaring both sides of the above expression. 8y' y T " 2 ( i + (/>2)'/2 V2g (9 .CALCULUS OF VARIATIONS 7J1 Example 9. and y the vertical axis. 6 .622) becomes (i + (y') 2 ) 1 / 2 y'm . then the velocity v at height y is computed through the conservation of energy (potential + kinetic = constant) to be v = /2gy (9. It was first proposed by Johann Bernouilli in 1696. 0) to a point (a.623) This example is known as the brachistochrone problem.624) where g is the gravity. In this example.63. 27) Substituting f and dy' and absorbing V2g into the constant c 1 .628) On simplifying the above expression.629) I A / ^— dy = I dx ] J V lc2y (9.
I ^isS. 0).y) + y ' q ( x .631) I / A / ClY dy = 2.154 ADVANCED MATHEMATICS To evaluate the integral on the left side of the above equation. y ) Then (9.J .cos 29] Cl (9.633) c2 Z where c 2 is a constant.638) l+y'l^y^fM/ (9.sin 2 0 (9.639a. The curve has to pass through the origin. (0. y) dy Equation (9.637) (9. 9 = 0.[l .s in 2 9 = . we use the substitution y = J .636) ^T = q (x. y = b.634) y = J .y sin 29J = x + c 2 (9. Thus the curve of quickest descent is a cycloid. Therefore.sinG cos9 d9 = ±. Function f Is a Linear Function of y ' Let us write f as f = p(x. .b) V1cfy Equation (9.l sin 29] (9. The parametric equation of the curve is X= X f e .630) can now be written as M© .[e . and c 2 = 0. The parametric equations of the curve as given above is that of a cycloid.b) ^ C l The constant Cj is to be determined from the condition x = a.56) becomes (9.b.635a.1 sin 20] c 2 J cosG c2 L 2 J (9.632a.
that the work done by a conservative force is independent of the path taken. If we take (p.644) (9. y') dx = I ^ dx = ())b .645) Then the work done by a from one point to another is independent of the path and is only a function of the difference of potential between the two points. Find the extremum of .512a. then (9.c) «. The above result is equivalent to the statement in mechanics.640) is identically satisfied.(££) If Equation (9.640) Equation (9. then the expression given by Equation (9.640) is identically satisfied.b)].636) is an exact differential.642a.643) (9.640) gives y as a function of x and there is no integration constant.64.CALCULUS OF VARIATIONS 755 The condition for I to be an extremum is dp da (9. Example 9.641) The integral becomes I =  f (x. q) to be the components of a vector a in a twodimensional space. and there exists a function < ] > such that dd> (9. such that >  a = ± grad < (9.<j)a Ja Ja ax Thus I is independent of the path taken and all curves yield the same value of I.b. y. This is similar to the case where f is not an explicit function of y'. Thus an extremum of I will exist only if y satisfies the end conditions [Equations (9. If Equation (9. then curl a = 0 So a is a conservative force and there exists a potential <).
648a.640) is identically satisfied. y. there is no extremum. In general. . 0) and (1. .b) Thus the condition is 2y = 2x or xy = 0 (9. f1 1= (y + xy')dx= fl [ydx + xdy] = fl d (xy) = [xy]^ = y (1) (9. I does not depend on which curve y (x) is taken. So in this case there is no extremum.65.649) In this example.650atoe) Jo • Jo Jo Therefore.o (9. 9.256 ADVANCED MATHEMATICS 1=1 (y2 + x 2 y')dx (9. then we have to solve the second order differential Equation (9. y ^n\ Thus we find the extremum of . 1). then the first end condition is satisfied. q = x. . such as y ".640) with p = y2 and q = x2 (9. . and y'. then the curve y = x will give the extremum of I.7 EXTENSION TO HIGHER DERIVATIVES We extend the theory to f being an explicit function of higher derivatives of y. so Equation (9. y ' " . Find the extremum of r1 1=1 (y + xy')d* .647a. if f is an explicit function of all three variables x. Example 9. If the second end condition is modified to (1.646) Jo The requirement for an extremum of I is given by Equation (9. p = y.b) If the end conditions are represented by the points (0. In this case.511) and the two end conditions will determine the two arbitrary constants.2). but the second one is not.
81) L dy Ja Using Equation (9. y". The ycoordinate of the curve will have its ends at any point along the ordinates x = a and x = b as shown in Figure 9. If the value of y is fixed at one of the end points (say at x = a) and is free at the other end. since y (a) and y (b) are not fixed.y. We obtain Equations (9.56. y " . . are given at x = a and x = b.73) dy dx U y ' dxMay" dx3 (ay111 "" dxn \ 3y (n) j This equation is essentially a generalization of Equation (9. y1. 4). In deducing Equation (9..81) simplifies to ^ =0 (9. are assumed to be satisfied. but the values of y and its derivatives at the end points not specified. but we can now no longer impose the boundary conditions r (a) = r\ (b) = 0.56).8 TRANSVERSALITY (MOVING BOUNDARY) CONDITIONS We reconsider the problem of finding the extremum of I [see Equation (9. then the condition given by Equation (9.72) where e is small and r\ (x) is an arbitrary function of x and proceeding as in Section 9. we can show that the condition for I to be an extremum is (9. we have assumed that y. y " \ . Thus the conditions for I to be an extremum are now given by Equations (9. then by considering another function y = y o + eii(x) (9. We proceed as before by considering two curves y 0 and y 0 + £T. as required. . we obtain Equations (9. Instead we need to deduce two new conditions.5. y ( n ) ) d x (9.82) V x=b . 9.81). y'.. The alternating signs are a result of several integrations by parts of the derivatives of T. 81).73). .CALCULUS OF VARIATIONS 757 I = /a f(x.53.55. The extremum curve has a greater degree of freedom and we cannot determine the two arbitrary constants in the solution of the EulerLagrange equation from the end conditions.71) If y = y o ( x ) is the function that will make I an extremum. We impose a new transversality condition Tl27 =0 (9.51)] with the end points x = a and x = b fixed. 6).81. The existence and continuity of y and its derivatives.
Find the extremum of 1=1 [(y') 2 +l] 2 dx (9. Equation (9. Since f = [(y') +1] is not an explicit function of y. the transversality condition imposes i l =0 9y' It follows that y'= 0 .85a.b.83) Jo given y(O)=l but y(l) is not prescribed. ADVANCED MATHEMATICS y ^J a FIGURE 9.611) implies Cl 4y'[(y')2+l] = (9.752.b) .86a.c) => cj = 0 (9.84) At x = 1.81. y = constant = c 2 The solution that satisfies the prescribed condition [y (0) = 1] is (9.81 L_^ b x Variable values of y and its derivatives at constant x Example 9.
814a) (9. Imposing the end conditions y(0)=l. k1 = ( a . we can assume that y' is a constant.2 a + 2] [2a2] = 0 (9.813a. We suppose y (1) = a. we find that ko=l.81 la. At least one of the roots must be real and since c1 is an arbitrary constant.89) The function y and the integral I are given by y=(al)x+l (9.87) I = 70 dx = 1 (9.2 a + 2] Jo (9.86).814b) The solution is a = 1 and I = 1 which is the solution given by Equation (9. The condition for I to be an extremum is still given by Equation (9. That is ^ = 0 or 2 [ a 2 .b) We regard a as a variable and determine the values of a that will make I a minimum.88) Let us determine I when the values of y are specified at both ends. we obtain automatically the value of y(l) which renders I an extremum. .b) y(l) = a (9. Thus y = k 1 x + k0 where kj and kQ are arbitrary.l ) (9.CALCULUS OF VARIATIONS 759 y=1 Thus the extremum of I is given by (9.812) I = [a2 .2 a + 2] I dx = [ a 2 .b) (9.84).810a. which is a cubic equation in y'. Using the transversality condition.
260 ADVANCED MATHEMATICS 9.9 CONSTRAINTS So far we have been concerned with finding the extremum of (b Ja with y satisfying certain end conditions.93) J* = I g(x. y')dx (9. y o + e 1 r  1 + 8 2 r i 2 . y.51) J = I g (x. algebraic equations or differential equations. Consider a neighboring curve to y 0 given by y = y o (x) + e 1 r 1 (x) + e 2 r 2 (x) (9. y o + e 1 r ) j + e 2 r i 2 ) d x (9. We define I* and J* as /•b I =1 /a f(x. We need to find the extremum of I Equation (9. in many problems y has to satisfy additional conditions and these conditions may take the form of integral relations.94b).94a. y. In extremizing I we can allow £j to be completely arbitrary as before and then adjust e2 so that the condition given by Equation (9. y')dx = constant = C (9. The functions Tj^x) and T2(x) are arbitrary functions. we have introduced two parameters £l and £ 2 . = C (9. y'Q+elr\[+e2ri'2)dx Ja.9lb) is satisfied. subject to the constraint 1= f(x.b) I* and J* can be considered as functions of £j and £ 2 and the extremum of I* is given by subject to the constraint given by £j = e 2 = 0. then they are known as isoperimetric constraints which we shall consider next.9la.92) In this case. Suppose we have to find the extremum of I as given by Equation (9. This problem is equivalent to the case of finding the extremum of a function of two . If the supplementary conditions are of an integral type.51). However.b) We assume that y o (x) is the function that extremizes I. y o + EJTIJ+ e 2 ri 2 .
95a. is a constant.2) (9. yo+£ 1 rij+8 2 Ti2)dx Ja where A.CALCULUS OF VARIATIONS 761 variables subject to a constraint as discussed in Section 9. we obtain JT. We therefore define a function H* as H* = I * + A J * = I (f+X.b) ^ k _ A 3L =O=d(f 9y dx 3y 3y + Xg)±\lT(f dx [ 3y + Xg)\ (9.96) — =C (9.95c) =0 e.b) = 1 L(x.58).4. y o +8 1 ri 1 +e 2 Ti 2 .910a.98a) = fTli[^^(^]jdx Ja [3y 0 dx \dy'o) fb dx by parts and using the boundary conditions (9.98b) as a result of integrating I rij ia 3yo •ni(a) = Tli(b) = 0 Thus the extremum of I is given by (9. The extremum of I* is given ^ aei (9.99a.b) .= Ui — + T\\— dx (9.g)dx Ja (9. The method of Lagrange multipliers can thus be used.=0 (i = l.97) Following the development leading to Equation (9.
912) Jo and the length s is given by Equation (9.911) Example 9.M y + W l + t y 1 ) 2 ) ] ^ (9. we obtain .91.11).91 The area A = I ydx i * Area (shaded) enclosed by a curve of length s (9.91).910). y o FIGURE 9.762 ADVANCED MATHEMATICS and gdx = C (9. Given a curve of fixed length s joining the points (0.913) dy dx [ dy Taking the partial derivatives. 0) as shown in Figure (9. remembering that y and y' are to be treated as independent variables.^ f . Using Equation (9. the extremum is given by My + x V l + t y f l . find the extremum of the shaded area enclosed by the curve and the xaxis. 0) and (1.
921) .)2 To integrate the above expression. we finally obtain ( y .cose we obtain y = + I X cos6 d6 = + X sin 6 + c 2 (9.915) Vl+(y')2 X where Cj is an arbitrary constant.916) ^(xc.920a. we use the substitution z = (xc 1 ) Then the integral becomes (9. The end conditions implies c 2 + c 2 = X2 (9.c 2 ) 2 + ( x .zdz On further substituting z = ?. On squaring both sides of Equation (9.C l ) 2 = X2 Thus the curve is an arc of a circle.CALCULUS OF VARIATIONS 763.914) dx L V l + ( y ' ) 2 _ This can be integrated to yield y' = ^ 1 > (9.b) On squaring both sides of Equations (9.920a.915).922) (9.919) (9.917) y =± I . b) and returning to the original variables.918) (9. we obtain m \ l d J ( ".C ' ) 2 (9. \\4 Y =0 (9.
925a) I \2X> X J o = 2A.and c 2 = ± V A (9.927a) (9. y ) (9. Find the extremum of .926) Thus the required curve is the semicircle with centre (1/2.92. then we have to choose the Lagrange multiplier as a function of x. We can now express X in terms of (9.11).11) yields s = XI dx — jo y?i 2 (xi/2) 2 This integral is of a standard form and its value is (9. X .0) and radius 1/2. y. j .927b) Example 9. y') = constant or y1 = g ( x .1/4) ) and radius X. with solutions Cj = }. ±V X2.1/2 and the equation of the curve is given by y2+(xl/2)2= 1 (9. instead of a constant. Differentiating Equation (9.924) s^fsin1^^]1 (9.923) To obtain X. If s = K/2.C l ) 2 = l 2 . • If the constraint is of algebraic type or in the form of a differential equation.siiT1( J) [since sin"1 (x) = sin" 1 (x)]. The constraints can be of the form f (x.925b) s and the required curve is a circle with centre (1/2.921) with respect to x and substituting y' into Equation (9.764 ADVANCED MATHEMATICS cf + ( l . we impose the isoperimetric constraint given by Equation (9. This result can also be obtained from geometrical considerations.
930) Subject to the constraint yiy' = o The Lagrangian L now is given by L = (l+(yi)2) + ?i(y1y') = 0 (9. yi' yi» . y 2 (b) = b 2 . y 1 (b) = b 1 .. y 2 (a) = a 2 . y n (a) = an y n (b) = b n (9.102a... y 2 .103a.CALCULUS OF VARIATIONS 265 1 = 1 [l+(y") 2 ]dx (9. ... and then we shall be able to complete the above example. We write yi = y" Equation (9.932) (9.10 SEVERAL DEPENDENT VARIABLES We generalize to the case of several dependent variables....• y i i ) d x (9.928) becomes (9. . . . . y n . 9. y n .c) (9.c) . . We need to determine the n functions y 1. yn that extremize the functional I given by i= }a f(x.b. y2. .929) 1 = 1 [l+(yi)2]dx JO (9.928) Jo We reduce the integrand to the usual form in terms of y'. The function f depends on n dependent variables y J .. . y2.931) Thus the problem of having a higher derivative in the integrand has been transformed to one involving two dependent variables y and y^ subject to a constraint.101) We assume the end points are fixed and the conditions are y 1 (a) = a 1 . . yi.b. We first extend the theory of calculus of variations to the case of more than one dependent variables.
.59). yoi+ejTi!. . .106) The extremum of I is then given by ft* aT aei = (9. as Tli — + ^ T L dx = 0 ( 9 .92. yon+EnV yii+eiTll.10 " 8 ) Ja [ ^i hi The conditions for I to be an extremum are obtained by integrating the second term in Equation (9.)dx (9...io7) £i=0 Equation (9. again following the development leading to Equation (9. Suppose y 01 . y o2 . 11) become . .766 ADVANCED MATHEMATICS We proceed as in the case of one variable.105a...b) !*(%) = /a f (x. Equations (9. yOn are the functions that extremize I and consider the functions y i = yoi + e i r l i ( x ) ' i=1 n (9. dy{ dx \ dy[ j i=l. Complete Example 9.n (9.1010) (9. y.108) by parts and the result is iL__l(_*U0. The extremum of I is now given by *_A(ik\ = 8y dx 1 dy' I ^±(^) 0 (9.932). .1011) =0 3Yl dx [ dy[ J Using Equation (9..104) where the r^ are arbitrary functions.2. satisfying the conditions Tii(a) = Tii(b) = 0 The functional I* (£j) is given by (9..101.109) Example 9.1010.. n +e n Ti.. .107) can be written. ..
y'(l)=l (9. ^T=2y2 (9.1022.1014) then the constants a0 to a3 will be given by a o = a2 = a3 = O and al = \ (9. subject to the end conditions given by Equation (9. a3 are constants.  L = 2yJ (9.d) Thus. Yl (Ti/2) = y2(7c/2) = 1 (9.930)].1013) Equation (9. is y =x Example 9.c.1016a.1018) subject to the following conditions yi (0) = y 2 (0) = 0 .1019a. Remember that y ^ y ' .109).CALCULUS OF VARIATIONS 767 ^(?l) = 0 ?i^(2y) = 0 (9.c. the function that will extremize I [Equation (9. y'(0)=l.1012) implies that X is a constant and we can integrate Equation (9.23) Thus the extremum of I are given by .102.1015a. 21) f=2yi. If the end conditions were y(0) = 0 .1017) I = Jo [(y. we compute  f = 2y2 .b.b.b. a^. y(l)=l.d) Via Equation (9.d) (9.1020. a2.1013) to obtain y = a 3 x 3 + a 2 x 2 + a1x + a 0 where a 0 . Find the extremum of lTt/2 (9.)2 + (yi) 2 + 2 y i y 2 ]dx (9.1015a to d).1012) (9.c.
768 ADVANCED MATHEMATICS 2 y 2 . = 0 yt . In elementary mechanics.^ (2y 2 ) = 0 =» =» y22y.^ (2yi) = 0 2yi .1019a to d) yields ai (9.d) Thus yj and y 2 are given by Yi = y? = 1 l s i u h K. which may be stated as follows. It is also possible to develop a theory of mechanics starting from a variational principle and this type of mechanics is known as analytical mechanics.y 2 = 0 (9.a1 cos x . In advanced mechanics. we obtain y2"y2 = 0 The solution of Equation (9.c.1027) (9. Derive Lagrange's equations. we deduce that y2 = .b) Example 9.1030a. the variational approach has certain advantages [Lanczos (1966)].1029a.1025).b.1031a.b) (9. one usually starts with Newton's laws of motion.n (9.1025a.b) v J smhrc/2 I is given by fK/2 j = 2 )o 2 2 cosh x + sinh x sinh2K/2 dx = 2 cosh nil Slnh7t/2 (9.1O24a.1028) = a2 = a3 = 0 . The path that a dynamical system takes in moving from one point at time tj to another point at time t 2 is such that the line integral .b) Eliminating yj from this system. One of the possible variational principles we may start with is Hamilton's principle.1026) (9.1026) is y 2 = &i cos x + a2 sin x + a3 cosh x + a4 sinh x where a1 to a4 are arbitrary constants.a2 sin x + a3 cosh x + a4 sinh x Imposing the boundary conditions given by Equations (9. From Equation (9.103. a4 = s . n h 1 7 t / 2 (9.
.. A particle P of mass m lies on a smooth horizontal table and is connected to another particle Q of mass 3m by an inextensible string which passes through a smooth hole in the table as shown in Figure 9..101.1033) *n velocities are q . 0) as shown.2. The components of the velocity of P are (r. q2. be the origin.1035) . The polar coordinates of P are (r. If the particle P is projected from rest with a speed V8ag along the table at right angle to the string when it is at a distance a from the hole. q .er) (9. The generalized coordinates of the dynamical system are q1.1036) (9.1037) (9. r 0). determine the ensuing motion of P and Q.CALCULUS OF VARIATIONS J69 1 =I L d t (9.r). . qn and the generalized *1 *2 *j (9. q .1034) are known as Lagrange's equations. Taking the level of the table to be the zerolevel of potential energy.. Thus the condition for I to be an extremum is ^LA/^M dt\9q'j Bq1 o.. L is then a function of q1 and q . i=l.1032) in an extremum..104. The dot denotes differentiation with respect to time. . Let O. the position of the hole. Since the string is inextensible. The kinetic energy of the system is K = i. the potential energy of particle Q is . Example 9. L is the Lagrangian and is given by L = Kq> where K is the kinetic energy and c p is the generalized potential energy. where k is the length of string.n (9. ....m ( r 2 + r 2 0 2 ) + 3 m ( f 2 ) The potential energy cp is q >= 3mg(ir) TheLagangian L becomes L= \ m(f2 + r202) + 2lllr2+3mg(. Q will move up or down with speed r.1034) The n equations given by Equation (9.3 mg(/8 .
we find that 9 is given by (9.3 m g .1040) . so C^afs^ Combining Equations (9.. a 0 = V8ag.101 Particles P and Q connected by a string of constant length.f (4mr) = 0 dt (9. passing through a hole in the table The Lagrange equations are therefore given by — — ( ^ ) = 0 dx dt 3r => m r e 2 . 41).1040.720 ADVANCED MATHEMATICS FIGURE 9.1038a.b) ^k±(^k)=0 ae dt ae = > f(mr2e) = 0 dt (9.b) Integrating Equation (9.1039b) yields r 2 6 = constant = C^ Initially r = a.1041) (9.1O39a.
1043) may now be written as 2dr^2) = ^3^~ 3 g (9.1045) which can be integrated to yield 2 r 2 •9 = _ ^ . we obtain 4 r'=8a_g_3g r3 (9.1042) Substituting the value of 0 into Equation (9.1046.3 g r + C2 r2 4a p (9. r* is negative (2g) [see Equation (9. 47). this does not imply that the radial acceleration at t = 0 is zero. we find that r r. r* is positive. r = 0 . C 2 is calculated to be C 2 = 7ag From Equations (9.1046) where C 2 is a constant.2=7ag_2*jg_3gr (9. That is to say.1047) is given by (9104g) 2 r2 2 We can deduce from Equation (9. we deduce that r = 0.1043) The term r can be written as " =i <h = i < f >$ ' ^ • 2t (f2) (9.1048). r" will be zero when r3 = 8a 3 /3. At r = 2a.1043).1043)] and .1044a. the particle Q will initially more up and r will increase. Consequently.& .1038b). when r = 2a and Q will come momentarily to rest. From the initial conditions r = a.d) Equation (9. From Equation (9. that initially since r = a. although the radial velocity is zero initially (at time t = 0).c.b.CALCULUS OF VARIATIONS 721 6 = ^ (9.
Geometrically it means that we are considering a surface instead of a curve.2a). That is to say. because of the lack of friction.772 ADVANCED MATHEMATICS so Q will fall. u X i (x!. The value of u (xj.x 2 ). 2). = =— (i = 1. let us consider a function u (xj. The particle P aV^gas^ has a radial velocity given by Equation (9.111 ^ Region R bounded by curve C We need to find the function u which extremizes I over the region R and satisfies the boundary condition .1048) and a tangential velocity — ~ . For simplicity. The radial position of P lies therefore between a and 2a.a) and (Z .' Figure 9. we allow the cycle to repeat when r is again equal to a.x 2 )] dxj dx2 (9.. Q FIGURE 9. 9. Thus Q will oscillate between the heights ( i . u X2 (Xj. x 2 ) which is a function of two independent variables Xj and x 2 .x 2 ).11 SEVERAL INDEPENDENT VARIABLES The integrand might be a function of two or more independent variables. R u(Xj. x2) is given on C.111) where u x .111. and R is a region in the XjX2plane bounded by a curve C as shown in 1 O A. We now have to find the extremum of I = l  f [xv\2.
Note that df 9f 9u.119) we finally obtain .116) now becomes fI(ui) = fr^. We consider another function (a neighboring surface) given by ul(xl.115) Substituting Equation (9.117) Equation (9.x 2 ) = 0 on C Then the extremum of I is given by A l ( u o + £Tl) ae (9. i=l.duv d I df .111).113) into Equation (9. ^ df — ==: T^ + X ^ 1 1=1 d"i x ^—i IXj au. we can introduce the operator — inside the integration. x 2 ) + eTi (x l 7 x 2 ) where e is a constant and r\ is an arbitrary function satisfying the condition T)(x 1 .118) Making use of the fact that ri xi 5 — = 5— Tj^ 3uY dX: duY 3f d df ^5—5— dx.CALCULUS OF VARIATIONS 773 u(x 1 ) x 2 ) = g ( x 1 . x 2 ) is the function that renders I an extremum.116) Since e is independent of Xj and x 2 . with^J=r) /ft 1 1 ^ (9.114) =0 e=0 (9. (9. _ (9.x 2 ) = u 0 ( x 1 .x2) u 1 (x 1 .+ riXi 4J^—+ ri ^ .2 . x 2 ) o n C where g is a given function.] d X l d x 2 de ^e^o // [ x 2 ^ux R (9. .112) Suppose u o (x 1 . we obtain ^ I K + e i )  e _o= ^JJ f ( X l' X 2' U 0 + e i l' U Ox 1 + e nx 1 ' U Ox 2 + e T lx 2 ) d x l d x 2 e = 0 R (9.113) (9.
1112) is zero. . .114) implies that the integral in Equation (9. .. u X2 .1113) We can extend to m independent variables x^.114). x m . . x 2 ) bounded by a curve C as shown in Figure 9.T l ^ ldxldx2 8x2 3uX2 j ] (9. we deduce that the condition for I to be an extremum is 9u 3x.15) 3 x m 3u Y \ X2 where. Green's theorem may be stated as follows Hi^lk)*1***=/<pdXl R C + Q d X 2 ) ( 9 "n) The double integral becomes //[sr("^K^)]^/[^^"tH (9""12) R C The boundary condition given by Equation (9. u. . .. x 2 .1114) 3f _J_ du 8x. x m .1110).. did many experiments with soap films to determine minimal surfaces. uY = ^—. From Equation (9. by Equation (9. I f (x1? x 2 . u Xj . .. a Belgian physicist (18011883).. u x j dxj. Plateau.. 1 3uv X] 3x7 2 9uv I x2 / (9... We are interested in converting the double integral to a line integral.112.. This problem is known as Plateau's problem.. Then we consider the functional I = j I . because the condition on T] is given on the curve C.. Find the minimum surface area of the surface x 3 = u (xj.774 ADVANCED MATHEMATICS dl(ut) =((\ df 8 8f \ 8 8f \ dE e=0 J) [^ 3u 8 Xl 8u x J 9x2 8u x J R + ^"l1!^3x^ 3uXj + ^ . as before.111. dx m R The condition for I to be an extremum is (9. xi dXj Example 9.1110) We can write the terms inside the second brace of the double integral as a line integral using Green's theorem. 1 i! 3u Y xl d_idf_ 9 x 9 19u Y z _J_ m _a^ xm = 0 (911..
112 Further let £ 0 =_r (Xj. That is to say. we consider four neighboring points Po.r 0 ) = r (x 1 ? x 2 + A x 2 ) . Let the vector positions of OP0.1 l18a.r (x l5 x 2 ) « ^ — Ax 2 according to Taylor's expansion.b) (9. enclosed by the points PQ. (9. Pj.£ (x 1. x2) . OP3 be r 0 . OP2. x 2 ) « ^=. r j .1117a. x 2 ) . (9.Ax : • dr . Pj. r 2 and r 3 respectively. OP 1.1116a.b.1116c.c) .£ 0 ) = £ (xj + Axj.112. The projections of these points on the XjX2plane are denoted by primes.b. Surface S bounded by curve C £ i = r_(xj + A x ^ x ^ r 3 = r (Xj + Axj. r 2 = r (xj.CALCULUS OF VARIATIONS 775 We shall first derive the formula for the surface element AS. P 2 and P 3 . S FIGURE 9. x 2 + Ax2) (9.c) P 0 P 2 = ( r 2 . x 2 + Ax 2 ).d) The vectors P 0 P^ and P 0 P 2 are given by dr P Q P J = (£j .P 2 and P 3 on the surface as shown in Figure 9.
1120) oxj  U2 = L+t^ < 9 1 1 22) Substituting Equations (9.776 ADVANCED MATHEMATICS The surface element AS. 3r 5r = j — A 5 — AxjAx2 OXi (9. we obtain ^ . the vector r can be written as r = Xj i_ + x 2 j + u ( x 1 > x 2 ) k where i. we find (9. From Equation (9. AS = P Q P J A P ^ . and k are the unit vectors along the coordinate curves. we find that Equation (9.1121) (9. we deduce the condition for the extremum of S to be (9.1123) The surface area S is given by S = jf V 1 + u^ + u^2 dxj dx2 From Equation (9.1120).1126) .1119a.1125) l X2 3X2 Vl+U2x+U2x V / / Noting that V l + u 2 + u 2 is a function of u x .1119) and carrying out the cross product.= i +^dx^ k (9.b) O X^ On the surface. only and differentiating the last two terms of Equation (9.1124) lVl+u2x 9U +u2x J3Xl gf Vl+U2x+U2x xl X2 # ^ x —  = n (9.1125) simplifies to ".1113). neglecting curvature. + ( ^ f ] ilJt + (i 2 j ^ _2L * L 3^ =o (9.1121. 22) into Equation (9. j .1125). is .
CALCULUS OF VARIATIONS 2ZZ In general the nonlinear Equation (9. We have to _ du * 3xj dr dXj dx2 r dr = = a (9.1128 to 32) into Equation (9.1133) can be transformed to a first order equation.1132) 92u 8xj 3x 2 x t x2 d 2 u r 2 dr 2 Substituting Equations (9.b) A dxj = xld\ + xdu (9n31) r 2 dr 2 = r 3 dr X! x 2 du r 3 dr (9. we obtain rdJL + dr 2 di + ( d i ) 3 = 0 dr V dr .1133) becomes r = v (9.1135) .1126) and simplifying.1127) du r dr (9.1130a.1134) dv dr + v + v3 = 0 (9. In this case the equation of the surface is given by x 3 = u(r) where r = Vx 2 + x 2 . (9. by substituting ^ dr Equation (9.1128a.1126) is difficult to solve. We shall consider a special case when the surface is a surface of revolution with the x3axis as the axis of symmetry.1133) Equation (9.b) (9.1129) t± = A £ u + 1 ^u _^1 du dx\ r 2 dr2 r dr r 3 dr xf d^u + x  du r 2 dr2 r 3 dr (9.
=f i ft (9. u n ) of several independent variables (xj. .dx m The conditions for I to be an extremum are then . x 2 . 5 i d Xl .42) j m .1140) (9. ^ .1137) .1135) is a separable equation and we have ( dv . the functional is given by I 1 ri 9ui 3ui dun dun (9.= £ (9. . .i .. j f ( x 1 . s i . In this case. X m . u m .1141) The extension to several functions (uj5 u 2 .i n r + i n c where c is an arbitrary constant. .. (9... . . . Thus the minimum surface is a catenoid.1138) Vl+v 2 r We can solve for v and "0.1139) r = ccoshf1^) This is the equation of a catenary. .. . .. u 1 . s X . This can also be written as v (9. .778 ADVANCED MATHEMATICS Equation (9. Thus the generating curve is (9. .. .1136) 1 vd+v 2 ) which on integrating yields i n v .. .i n ( l +v 2 ) = . x m ) is straight forward.is then given by ^ = c y r 2 _ C2 dr Integration gives u as u = c cosh"1 (r/c) + k where k is a constant..
The system given by Equations (9.dv: \ where y.1. The equation of continuity is given by ^i. Note that in this example. as g + ^_ p f l = 0.1144.= v H = 0 dXj M (9.are the velocity components. 1 = 1. Show that the equations governing the slow flow of a generalized Newtonian fluid are equivalent to extremizing a functional. Remember the convention. x^ is the deviatoric stress tensor.1146) m e rate °f deformation tensor. = 1 / 1 y.44) where p is the pressure. p is the density and fj are the external applied forces (usually only gravity). neglecting inertia. ni is the component of the unit normal to the surface S t .. requiring summation over repeated indices. y.\ = ^rL +^is I I . we consider different boundary conditions on different parts of the enclosing surface. v. / 3v. (9. The equations of motion. can be written. ) ^ .1 1 ( 1 1 . 46) subject to the boundary conditions stated above will be shown to be equivalent to finding the extremum of .1145a..CALCULUS OF VARIATIONS 779 (9. when referred to a rectangular Cartesian coordinate system O x ^ X g .2.. i = = — * .b) The constitutive equation of a generalized Newtonian fluid may be written a Tij = .(p ns + x. Let V be the flow region (volume) bounded by two nonoverlapping surfaces S v and S t . and is the second invariant of y.1143) where V . j Example 9.3 (9.n:)] is given on S t . The velocity is given on S v and the traction (surface) force tj [= .112.
From the definition of I I .1149) Jv The extremum of I is given by ^=0 de e=o In the computation of this extremum.1148a. The quantities fj and tj are given.b) where Vj and p extremize I. Next we illustrate these 3II. Substituting Equations (9.o .1148a. so 7 . we will have to determine Jst (9.1150) :——.Y o'ij L 2 Y'J Y J l J f9 1151abc) ( " 2 L 2 Yl J Y JJ YlJ " 211..l t^Vj + ew^dS (9. p* = p + eq (9.Y oiij " . Since II. .780 ADVANCED MATHEMATICS I [gaiOpfiVipVjiJdVl t iVi dS Jv Jst where g (II.l 1 / 2 .1 r l v.J L r l .1.. is defined in de y * ds.r 1/2 Y.1147) u T (u) du . i d V . Y manipulations. dg terms of Vj .1147) yield I*=l g ( I i : ) . To find the extremum of I.p f i ( v i + ew i )(p + eq)(vi + ew i ).) = I Y (9. Y.tO. Jo The functions inside the integral are functions of the velocity components Vj and the pressure p. Y ' ' } Since g (II ) = I ur)(u)du. b) into Equation (9. we consider neighboring functions v* and p* which are given by v ^ V j + ewj. we will be required to calculate —^_ an( j hence j. we have 1 . v.
iqvifi]dV Recognizing that (XjjWj)^ = XjjjWi + XjjWjj or T i j W i j = (TijW^. the velocity is prescribed. is symmetric.b.= — 1^~ = JTl(n. .1154) (9. On the surface S v .^ .1 Ty (wifj + w jfi ) = xy wifj 9e 3Yij 3e since TJ.1155a) (9. j d V = I [(wi^ij)'j'cijJwi]dV Jw Jv = 1 ( w i t i j ) n j d S .1156b) Jv Similarly the integral /st Jv .d) — = ~.1157) (9.( Tjj )jWi dV ys t +s v Jv where we used the divergence theorem on the first integral on the right side.^ i j aYij dIU 3Yij 211^ 2 Y J 2 — = .j (Tijj)widV (9.b.1156a) (9. and so Wj = 0. we obtain I [xijwi.c.jPfiwiPwi.jTijjWi (9.c) t i W i dS = O (9.Ti(ii. . (9.)T§i = .1150.1152a. Combining Equations (9.) Tji .1155b) It follows that I x i j w i .1156b) can then be written as xijWi j d V = I ( w i T i j ) n j d S .1153a. n. Equation (9. 53).CALCULUS OF VARIATIONS 781 ag ag an.^ = .
1147).1158a) (9.1144.1162) The boundary conditions are r = kR.1158b) yst Jv Substituting Equations (9. .1145) (9. Note that the boundary conditions are satisfied automatically.1157.1154).b) r = R. i d V = I ( p w ^ ^ d V .1144) (9.jWjdV Jv ' Jv Jv = I pWjnjdSl p. 0<9<27t. as illustrated by Figure 9. for all 6 and z (9.c) The radii of the inner and outer cylinders are k R and R.113.782 ADVANCED MATHEMATICS I p w i .( i j j + p S i p n j (9. vz = v z (r) (9. vz = 0 vz = 0 .1 l64a.113. The flow region to be considered is defined by kR<r<R. we can instead extremize the integral I given by Equation (9.1161a. The velocity distribution is vr = 0 . 58b) into Equation (9.1147). since we used them in formulating the integral given by Equation (9.b) (9. Example 9. ve = 0 .J p.b. we obtain Jv I [{^ijJ + PfiP'iJwi + q v d d V + l Jst [{Cijj + pSijJnj + tiJJwidS = 0 (9.p f i + Pu = 0 vi.jWjdV (9.1159) Since w^ and q are arbitrary it follows that t i j j .1163a. We assume the flow to be in the zdirection.1160) Thus instead of solving Equations (9. 46). 0<z<L (9.i = 0 tj = . Use the variational method to solve the axial annular flow of a power law fluid between two coaxial cylinders.
(l+n)/n. v z = a ( l . Thus an appropriate v z can be chosen to be I .113 Annular flow bounded by cylindrical surfaces S v and end surfaces S t On the shaded surfaces at z = 0 and z = L. This is the usual procedure in variational calculus.1168a) . —* —kR —R FIGURE 9.. the pressure is given.1165) ri(y) = mlyl"" 1 We nondimensionalize r by writing r =  (9.\o\ ) (9.1167) We select a trial solution which satisfies automatically the boundary conditions [Equations (9. Suppose the potential drop from z = 0 to z = L to be A P . 1997) (9.1164.65)]. where the potential IP combines the effects of pressure and gravity according to F = p + pgh The viscosity function r\ for a powerlaw fluid may be written as (Carreau et al.CALCULUS OF VARIATIONS 783.1166) (9.
From Equation (9.169a.( l +k) (1_k) (9.A F R ^ 1 v z [ ( i ^ a + <i±i>] ( l ^ ) d o (9.1170a. 2 dv7 2a (1 + n) i 11/n W *"df = *R Iff = ±^1^10 to = 'nRTTTky (9.1172C) .1172a) h JO JkRn + 1 = Jo L rv z dr Jk lnTir Jk Y n+1 FdrAF27tR 2 (9.1147) and where the boundary conditions on the surface S t result in a potential drop. We now rewrite I as follows rL I = flit fR tin rR l I I 2 f l JBLT Y n+lr drd9dzAF I fl I v7rdrd0 Z (9.284 ADVANCED MATHEMATICS q = 2 r . dv7 T= 1 dv7 .a .d) where the upper sign is for a < 0 and the lower one for a > 0.1166).1172b) . Then y is given by .c. we have n g(Y) = uti(u)du = ^ E y y n + 1 (9. APvzdS (9.1171) iv )st where the mass balance for an incompressible fluid (vj j = 0) eliminates the pressure term in the first integral of Equation (9.b..b) /o Thus the integral I [Equation (9.1147)] we need to extremize is now given by 1= I ^ j Y n n + l d V .1168b) where a is a constant to be determined.
This integral has to be minimized and we need to compute ^ .1 l68a. (1987).r k) J KR (9.CALCULUS OF VARIATIONS Z21 Since we have defined 7 and v z as functions of \a\.q a where C o = ° (9. da Differentiating yields 4 1 = C n (n+l)[2a(n+l)] n »2(n + l ) . 75). b.C 1 = 0 da Hence RAF ) 1/n R(lk)< l4 *> /n n 2^7/ 2(n+l) .b) using vz. given by Equations (9.1173) .1174a. (9H75) (9.Q 1 . For smaller values of k and n.1172) will be zero when integrated from a = .176a. This technique is known as the RayleighRitz method.5 and n > 0. the first term in both integrals of Equation (9. Cl rcLRm(lk)n (n+l)(l + 2 2 = £ ± £ KAP R2(l k 2 ) .b) a = We can now calculate the volumetric flow rate Q.l to a = 1. In Bird et al. Integration yields I = CopaCn + l f l ^ . . <2n+1) 2n)[nR(lk)] n + 1 ' The integral I is now a function of a. the value of Q for various values of n and k are compared with the exact solution. since they are odd functions of a. the approximate solution gets worse and could probably be improved by considering more terms in the trial solution. „ . which is given by ^/ R ^«fer (i c. This is followed by expressing the unknown functions as a series of simple functions that satisfy the boundary conditions. The above approximate solution is within 2% of the exact solution for k > 0. The unknown coefficients are determined by satisfying the variational principle. • The method we have used to solve the present boundaryvalue problem consists of rewriting the governing differential equations and boundary conditions as a variational problem.25.
. only along the straight lines x = a. x = b). yj.8 to the case where the functional depends on n functions. y 2 . The quantity 3— is given by d8j df <*ei J.. [ df dy: + j=1 df dy': ] (9.122a.y 2 ) (9.124a) dy. i = l. y n . Using Leibnitz's rule.786 ADVANCED MATHEMATICS Although simple problems.121) Jp The end points P and Q are not fixed nor are the values of y 1.107) becomes * /Q L= l^dx +f ^ . We assume that P and Q lie on two surfaces respectively given by x = <l>(yi>y2)' x = \/(y 1 .10.12 TRANSVERSALITY CONDITIONS WHERE THE FUNCTIONAL DEPENDS ON SEVERAL FUNCTIONS We extend the theory considered in Section 9.f  * = 0 . but now the end points are free to move on the surfaces given by Equation (9. more complex problems require the use of an approximate solution.. y . We further relax the condition that the end points can move only along the ordinates (that is.aej ^ = 2 3 y j 9£j =2 j=1 ^^+r^^ :r dy.b) where P and Q are functions of 8j.b) We proceed as in Section 9. b). such as this flow problem. Equation (9. can be solved exactly via the equations of continuity and motion.y 2 . We allow the end points to move along given curves or surfaces and we label the end points as P and Q.123a.2 (9. For simplicity we derive the transversality conditions for the case of two functions y 1 and y 2 . 9.dB{ 9yj d^ v (912"4b) dx ' . y2)dx (9.122a. Variational methods are widely used to obtain approximate solutions to boundary value problems in continuum mechanics. y 2 at these points. We aim to find the extremum of rQ 1= f(x. . y i ..
124c) Substituting Equation (9. we obtain 0= ^ I dx + li + f ^ f^ (9.2 (9.125). That is to sa y . Substituting Equation (9. Therefore.104)] are related for example b y an expression such as y ^ + y2 = c.CALCULUS OF VARIATIONS 787 & F 3f ty d df dy: i+ _ ^ _ i = 2.125) J P a£i J9 Yj dx a y .a £ i j p ^ Q a^ P using the convention of summation over repeated indices.d / df 1_ 9 ^ dx ^ 3 y j (9.127) Next.126) where d^ is the Kronecker delta. we deduce the extremum conditions to be K_±(K\ 3Yi dx [ dy[ J = 0. to be functions of x and e.122a. we differentiate to obtain .128) (9. dx a y j 3e. we deduce the transversality conditions. if equations y\ and y 2 [see Equation (9. It can be seen from Equation (9.122a. The transversality conditions are to be imposed at the points P and Q which are on the surfaces given b y Equation (9.129) Since y^ is a function of x and e. then a change in £j automatically induces a change in e 2 and we can indeed consider the y . i=l. b). except at P and Q. The transversality conditions may now be written as 2 2 "iL^L + iL^ +f 3Ql = 0 dy\ de dy'2 de 5e _ '^^L+^_^l+faP] dy[ 9e 3y 2 9e de = 0 (9.  j [a yj . y 2 and x are related through Equation (9. derivatives with respect to £j are now replaced b y derivatives with respect to e.104) that (9. b) and we can regard y j and y2 to be functions of x and e instead of considering them individual^ as functions of £j and e 2 .126) into Equation (9. Thus y i .123) and simplifying. j=i p Y j ^ dy.124c) into Equation (9. At an y point in space. y i and y2 are independent.
1212) becomes .128). 9y2 3y i y2jjp . .^ is given by 2 de = dx = Mi ^ de dy{ de +  V <^2 dy2 de (9. dyi . de \ayl a y i l ayj dE ! ayi 2 J ) (9.j5 r)l 3y2 3y 2 \ 3yi ay2jjQ =0 (9m5) Similarly. lla.fy.ZM dy. ] ! _3f_ I^Xl _ v ' ( ^V dYi ay dY2 \ 1 3 y .122b) and .1210.12lla.1210) We can also do this for y 2 . the transversality conditions at P are given by rr+:r4yi]F " y ^)i =° ayi 9y 2 (912"l6) (9. b) into Equation (9. Equation (9.b) 7 Substituting Equations (9. 3f \ l ^ L f_y _y n = 0 .1217) ayi\ ayi y7 ay2jjp =0 a +—. 1 de y Mdy 1 de dy2 de // a y ' \ de y 2 \ 3 y i de dy2 de / / + fj^fl \dyi de + ^^)l 8y 2 d e  J Q =0 (9. the transversality conditions at Q are given by " df + dy /.1213) \ay2 ay21 a yi ^ 2 j / J Q Thus. we obtain ' #  d y i _ v ' / _ ? V dll + i V ^ 2 .1212) Collecting similar terms.n^ 1/1N (9. dx ADVANCED MATHEMATICS (9.1214) 3yi 8Yl ^ 3yi 3y 2 j J Q r + rh^. The point Q is given by Equation (9. 3f .
b) The extremum condition will be given by Equation (9.1222) S =I JO Vl+(y') 2 dx (9.1224a. the transversality conditions at Q are "iL 3 ( f _£ . y n . i=1 „ (9.127) with y^ = y.. The transversality conditions then reduce to "iL+?l(fy'it] .1225a.b) With f given by Equation (9. y 2 .122!) Example 9. Thus if f is a function of yj..„ L3y dy \ 3y' j Q (912. then the end points P and Q will be free to move on the curves.218) Similar expression for the condition at P can be established. 0). \ j=i J3yj j J Q =0 . ] ] _3yi 3y..1223) Since y does not appear explicitly in the integrand.1219a. we have ? y = Co => y' = k 1 (constant) (9. Equation (9. The equation of the circle is (x4)2 + y 2 = 9 The distance from the origin to any point Q on the circle is (9. .127) simplifies to — dx — dy' = 0 => ^ = constant 3y' (9. x = \j/(y) (9.b) Vl+y'2 . £ . Find the shortest distance from the origin to a point on the circle of radius 3 and centre (4.CALCULUS OF VARIATIONS 789 The above derivations can be extended to more than two functions.121..1223).20) "jL+?lf_y'iLl 3y' dy dy' J P =o (9. If f is a function of only one dependent variable y. respectively given by x = 9(y)..
.. we obtain 2 kl \ *1J V l + kf  kl =0 (9. . In this section.k. y n . consider two types of constraints: an integral type and an algebraic type. y = 0) yields y = kjx (9.9.c) Substituting the value of y into Equation (9.1222).1227a. We shall. Q lies on the circle given by Equation (9. y = 0 and the point Q is on the xaxis.0).. A. as in Section 9. The solution is kj = 0. ..1220). The condition for the extremum of I subject to the isoperimetric constraint is .. From Equation (9. . Thus f dy .131) subject to Jj = I gj (x.. ..1222) as well as on the line given by Equation (9. in addition to the end points conditions.1228) V1 +kf (x4) Vl+kf . yj.790 ADVANCED MATHEMATICS Integrating Equation (9. we established the conditions for the extremum of a functional which depends on only one function.b. y ^ d x (9. * L =^L dy (x4) = dil± (x4) (9. .13 SUBSIDIARY CONDITIONS WHERE THE FUNCTIONAL DEPENDS ON SEVERAL FUNCTIONS In Section 9.. y n .132a.. Then the minimum distance is 1.. y^) dx = constant = Cj . The function was subjected to certain supplementary conditions.1226).1220). X2.. we consider the case of a functional which depends on several functions. j = i. we make use of Equation (9. yj. .. y l 5 y 2 ..1225b) and using the end condition at the origin (x = 0. 9..b) We use the method of Lagrange multipliers and introduce ?ij.1226) To determine the slope k j ... . We now look for an extremum of the functional 1= /a f(x. Ja The end conditions are prescribed. y } . k (9..9. we deduce that Q is the point (1.
we write d.138) . The k Lagrange multipliers can be determined from Equations (9.x] = 0 4 + .139.133) We have to solve Equations (9.[ A { ( y i 2 + y224xyi4y2) + ^( y .d) y 1 (0) = y 2 (0) = 0 .b).b.CALCULUS OF VARIATIONS Z2Z Mf4Mi[iHM}°' Example 9. Equation (9.137) (9. b. 2 xy.y 22 )}l = 0 dx 9y2 Carrying out the differentiations in Equations (9.134.133) is in general a second order differential equation.i r { ( y .y 2 2 ] d x = 2 Jo The end point conditions are ( 9 1 3 .A [ . 5) into Equation (9.139) (9. 2 + y 22 4xy 2 4y 2 ) + X(y. Substituting f and g from Equations (9.5a > b ) yi(l) = y2(l)=l (9.A.131.d [2y 2 4x2?iy 2 ] = 0 OX (9. I = k 2 + y224xy24y2]dx (9.[2yJ + 2^yJ .132a.132a. 2 x y .c.y' 2 )}j = 0 dx 3yj (9. 10) yields .x yI . The 2n arbitrary constants resulting from the solution of the n differential equations are to be determined from the end point conditions.133) and noting that f and g are not explicit functions of yj.134) Jo subject to J = I [y 1 2 .1310) Integrating Equations (9.137. 8) and simplifying. we have for i = 1 and 2 . We have one constraint and so we need to introduce only one Lagrange multiplier X.136a. Find the extremum of i1 i"in (9.4 . 3).
1311.136atod)andare cl (9.1315a. The constants Cj to c 4 can be determined from the end point conditions given by Equations (9. we introduce Lagrange multipliers which are functions of x rather than being constants as in the case of a functional with only one function. The value of A. we find that yj and y 2 are given by Yl = ^ 4(1+X.1313) (9.1317) ~2 x 2 + 2X' y2 = X (9.b) (9. 12) respectively.1316a.1312) where Cj and c 2 are arbitrary constants.1318a. y2=1 (9.c. we obtain 2 y 1 ( l + X ) = ^ ~ + c l X + c3 2 y 2 a . 12).b) If the constraint is not of an integral form.X ) .b.X ) = c 2 x + c4 where c 3 and c 4 are constants.d) Substituting the values of C and c 2 into Equations (9.1314) = ( 4 +23X) » C2 = 2 ( l . Suppose we need to find the extremum of .) The Lagrange multiplier can now be determined from Equation (9. is X = 10/11 Thus the required functions yj and y 2 are Yl = . On further integrating Equations (9.1311.1311) (9. + 2XyJXx = 2 y 2 .135) since y J and y 2 are known in terms of X.2 ? i y 2 = c2 Cl (9. c3=c4 = 0 (9.792 ADVANCED MATHEMATICS 2y.
.1324) Finding of the extremum of I now consists of solving Equation (9. y n . y^) = 0 . the equation of the sphere is x2 + y2 + z 2 = l If the curve joins the points (x 0 .1325) pi s= V l + ( y ' ) 2 +(z') 2 dx (9.2.1326) The Lagrangian function L is then L = V l + ( y ' ) 2 +(z') 2 +^(x)(x 2 + y 2 + z 2 . y ^ d x (9.. .1319) Subject to the constraints gj (x. zj).. y n (a) = an (b) = b.. .1323) The conditions for the extremum can be shown to be ?L±l*L) = 0t 3yj dx \dy' J i=l. the distance s is given by (9... 22). .y2. . Determine the shortest curve between two fixed points on the surface of a sphere of unit radius and centre at the origin. ..1322a.1327) .. ..b) (9.132... .l ) (9. Example 9. . 2. . y n (b) = b n (9. z 0 ) and (XJ.yj..1320.1320) The end point conditions are given as follows yi (a) = a{ . y i . j = 1.. m (9. .. y].1321a. .CALCULUS OF VARIATIONS JSA 1= Ja f(x. yj. y n . . In a rectangular Cartesian coordinate system. y 0 . .1324) subject to Equations (9... yj.b) Yl We now introduce a Lagragian function L given by m L = f + £ Ajtogj j=l (9....n (9..
1330a. z = z(t) (9.1324) identifies the extremum conditions as (9. x(t 1 ) = t 1 (9.e. r i i dlioi 2x = d _ rii dtLoi 2y = A ri] dtLgJ.794 ADVANCED MATHEMATICS It is useful to introduce the parametric representation of the curves as x = x(t). z are the dependent variables.1334a.1335a.1327) can be written as L = V x 2 + y 2 + z 2 + n ( t ) ( x 2 + y 2 + z2) where [i (t) is the tdependent Lagrange multiplier. 2^izf [ i ] = 0 dt L a J (9. 2 n y .~ (^) dy dt 8y =0 i.1331) f.b) Equation (9. Noting that d ^ = dxj^ & E q u a t i o n (9.b) On differentiating [ — J . we obtain . Equation (9.e.1326) becomes dXj dt dXj 4 ' s = I V x 2 + y 2 + z 2 dt = I adt it 0 /to where the dot denotes differentiation with respect to t.1328a.b) and — . The fixed points are x(t o ) = t o .— (—) = 0 3z dt V3z i.± ( ^ ) = 0 3x dt dx i.c) where t is now the independent variable and x. 2z (9. y = y(t).1329a.1333a.A U] at a =0 (9.1332a.b) ^ . y.b.b) (9.b) Eliminating l(t) yields d. 2nxA [j] = 0 at CT (9.e.
1337). dt L o J dt L a J Equation (9.a y ) or or a (xyxy) = a (xyxy) £ =& ^ a (xyxy) (9. we have y (ox .x y .b) (9.CALCULUS OF VARIATIONS 795 I [i]  ^ (9.1337).1340).1336) We have similar terms for ".\ — 1 and — \ — 1.b) becomes •• • • oxox 2xa = •• • • ayay 2ya = •• • • ozoz 2ZO2 (9.1337a.1338C) Similarly.1338. we obtain ^n (xyxy) = i n (zyzy) + i n Cj (9. from Equation (9.o x ) = x ( a y . from the last 2 sets of terms in Equation (9.1342a) .1340) .b) From the first two sets of terms in Equation (9.x y = x y . we find £ o = (zjr^ (zyzy) (9 1 3 3 9 ) Also. we have made use of the following expressions ~ (xyxy) = xy + x v . = dLi .1340). 39).1341a. we deduce £ (xyxy) ~ J (zyzy) (9.x y dt Integrating Equation (9. (xyxy) (zyzy) Note that in deriving Equation (9.1335a.1338b) (9.1338a) (9.
1144b) (9. z0) and (xj.796 ADVANCED MATHEMATICS or xyxy = c^zyzy) (9. Silastic implants are widely used in plastic surgery (Atkinson. The balloon is then filled with saline over a period of several weeks until the skin has been sufficiently stretched so as to cover an adjacent defect area.131.1144a) (9. y0. Zj) are given. .1343b) can be integrated to yield inCxCjZ) = i n y + /8nc 2 or or xCjZ = c 2 y xc2yc1z = 0 (9. The arbitrary constants Cj and c 2 can be determined if the fixed points (x0.1344c) Equation (9. Formulate the problem of silastic implant and determine the optimum design of the implant.1342b) where Cj is a constant. 1988).1344c) is the equation of a plane passing through the origin. The intersection of the plane given by Equation (9. The plastic surgeon needs to know the optimum design of the implant so as to cover the damaged area for a given volume of saline.133. Let A be the area of the base of the implant which is in the xyplane. yj. Example 9.1325) is the arc of the great circle. This problem is similar to the Plateau problem considered in Section 9.11. Equation (9. as shown in Figure 9.1343a) • Z • V or l— = 5y (9.1343b) xCjZ Equation (9.1344c) and the sphere given by Equation (9. The technique consists of inserting a deflated silastic balloon underneath the skin.1342b) may be written as yCxCjz) = y(xcjz) • XCi (9.
1124) is given by S=j A [ l + h 2 + h2] 1/2 dxdy (9..1347a) . I) = SIV (9. dy The volume V of saline is V = I I h (x.1345) . the surgeon will perform the operation and advance the skin to cover the defect area. Once the skin has been sufficiently expanded.131 Geometry of an implant The base of the implant is PQRP and the defect area to be covered is PTRP. then the surface area S according to Equation (9. y) dx dy = Vo A (9. h " = 5 dh . • .* /! \ t FIGURE 9. . x dx .1346a. 9h where h =^.CALCULUS OF VARIATIONS 797 y R IMPLANT 1 y HEALTHY «MN A KCALTHV SKIN ^tm^ V^TS^ I I . y) be the height of the expanded skin above the base. This isoperimetric problem can be solved by introducing the Lagrange multiplier and is equivalent to finding the extremum of L(h.b) To ensure that the defect area will be covered we need to know the minimum surface S for a given volume V. Let h (x.
8 7 (as. We choose the trial function for h to be h = (Ai+Bix + ciy) (P 2 x 2 + a V  a2 P 2 ) (ai . Further we assume that the yaxis is the axis of symmetry for PSRTP and thus the equation of the curve PTR is given by Equation (9. 9 h . We now use the RayleighRitz method to solve for h.a ?y 2 .1347b) into Equation (9.3 ^ ( 3 ^ ) .1348) yields ^ [ ( 1 + h x + h y r 2 h x ] +  7 [ ( 1 + h x + h yr 1 / 2 h y] = ^ P13^9) We assume that the implant is made up of sections of ellipses and the curves PQR and PSR are given respectively by 2 PQR: y 2 = 1  (X ~d) (9. This nonlinear minimization problem is solved numerically.a j < d < a + a j ...(x .x replacing x.d>2) ( 9 1 3 _ 5 2) 2 ( a 2 y 2 + (32x2) The boundary conditions h = 0 on the perimeters PQR and PSR are automatically satisfied by our choice of h.1349).1347b) = 11 Fdxdy A (9. (913"48) Substituting F from Equation (9.(9. . The constants Aj.1113).1351) a2 where a. p\ otj and d are constants and a .1347C) The condition for L to be an extremum is given by Equation (9.1351) with . Bj and C^ are to be determined by satisfying the variational condition of Equation (9.1350) a2 2 R2x2 PSR: y 2 = (3 ._ .798 ADVANCED MATHEMATICS = l A (l+h^ + hy) Xhjdxdy (9. .. that is 3F 3 / 3F \ 9 / 3F = 0 .
and V o can be obtained. We consider an isotropic body of volume V with nonhomogeneous thermal conductivity k(Xj) which is unknown. y) respectively. X = 2. Bj = 0. it can be substituted into Equation (9. For other values of the geometric parameters and X. large values of (3 imply that more importance is given to achieving the .4632. Bj and Q : A{ = 1. This problem has been considered by Meric (1985).CALCULUS OF VARIATIONS 799 The parameter X can be determined in terms of V o . d. For the values of a = oil = d = 1.8186.2192.1346) and the relation between A. Also the values of E and the required values of E for various conditions are tabulated in Atkinson's paper. Thus varying the values of X is equivalent to varying the values of V o .1355) Small values of (3 imply that more weight is given to attaining the desired temperature distribution T d . We need to determine the optimal distribution of k(Xj) such that it satisfies a desired distribution kd(Xj) and a desired temperature distribution Td(Xj) in the body. ocj and X. Atkinson has found the following values for A 1 .131) is given by It L = l ? \ (1+h^j m dx (9. Bj and Cj are obtained numerically for various values of a. p\ d and a j . (3. The length of the curve L on the inflated balloon corresponding to a horizontal (y = constant) section PiP 2 (Figure 9. Cj = 0.5. At the other extreme. (3 = 1.134. This is a problem in the manufacture of composite insulating materials. Mathematically we are required to minimize the socalled performance index J defined by (9. The extension E produced by the inflation of the balloon in the xdirection for a given y is E = L(x!x2) The balloon must be inflated to a volume such that E is sufficient to cover the defect.1354) J =  [(TT d ) 2 + p(kk d ) 2 ]dV where T is the temperature and (3 is a weighting factor.1353) where the coordinates of Pj and P 2 are (x1? y) and (x2. (9. The shape of the implant is determined by the values of a.85. Solve the problem of optimization of thermal conductivities of isotropic solids. since once h has been obtained in term of X. The values of Ai. the reader is referred to the original paper. Example 9.
*(Y.c) I*= I h j(T*Td)2 + P(k*k d ) 2 } + A. (k' Y T + k Y T ' ) ] \ dV = 0 (9.1359) where X is the Lagrange multiplier.1356) is equivalent to finding the extremum of I = J+ X [ V .k d ) + X'(Y • k YT + Q) + X [Y .1362) .b.1356) where Ts is the prescribed temperature on Sj.(k*VT*) + Q)]dV (9.1361) Then the extremum condition. q is the boundary heat flux.1358) (9. We assume that the body (V) is enclosed by two surfaces Sj and S 2 and the boundary conditions on Sj and S 2 are given by onSj: onS 2 : T = TS k ^ .800 ADVANCED MATHEMATICS desired thermal conductivity distribution k d . leads to e=0 I (T'(T . a is the heat transfer coefficient and TM is the ambient temperature.112 and consider the neighboring functions T* = T + e T \ k* = k + e k \ X* = l + el' (9. the equation governing the temperature throughout the body of volume V is Y(kYT) + Q = 0 where Q is the energy.T d ) + 3k'(k . associated with the heat source.+q + a(TToo) = 0 on (9.1360a.1357) (9. Finding the minimum of J subject to the constraint given by Equation (9. ( k Y T ) + Q]dV (9. To show this equivalence we proceed as in Example 9. At steady state. de = 0. ^— is the directional derivative along the outward normal to S 2 .
^r dn \ dn T ^r— dn dS = 0 (9. we have I { T ' ( T . we have I A. r.(wVu)]dV = I w ( u ^ . the boundary condition given by Equation (9.1363) iv Jv Using Equations (9.i^ oT .1364) dnj k'YA. we deduce from Equation (9.i oA.1366) Substituting Equations (9.1368) . 64).YTdV (9. (k V T) + Q]J dV V I .CALCULUS OF VARIATIONS 801 We now apply Green's generalized first and second identities to transform part of the volume integral to a surface integral because the conditions are specified on the surface.(T + eT') + q + a ( T + e T ' .T ' y 1 dS+l V T'Y*(kYA. 66) into Equation (9. JO „ + 1 I Si k A. T is prescribed (= Ts) and T* = 0. L aT _.(kVT')dV = I V S\ + S2 k L ^ .+ k ^ .(wVv)vV. ^ dn on T ^— dn dS + I I S2 k A.V.„ I .(k'YT)dV = I V S 1 +S 2 is is k'X^dSl dn I [uV.1367) On Sj.T J = 0 dn Expanding and comparing powers of e.T d + Y • (k Y X)) + k' [p (k .) d S l dn (9. On S2.1369) (9. ^— + k A.+ a T ' dn dn = 0 (9. i^ a T .1358) must also be satisfied by k* and T*.Y X • Y T] + X' [Y . The Green's identities may be written as follows I [w V U «Yv + uV . o T „ ! oh .k d ) .v ^ .V.(wYv)] dV = I w u ^ d S (9.1362).)dV (9.1368) that on S 2 k'^.1365) V I A.1365. This implies that (k + e k ' ) ^ .1363. = r —+k A .
1374) We note that Equations (9. q = T^ for various values of a. They are similar to Equations (9. in addition to an extra set of equations which may be written as inV: Y«(kYX) + ( T .1370) J Si dn J S2 dn Since T'. 74) are the equations governing A.Y X• VT] + X'[Y • (k YT) + Q]) dV V + 1 k ' ? i ^ dS .I T' Xa + k ^ dS = 0 (9..1371) (9.1372) is the socalled gradient condition.01.1371. In the case of the infinite plate.1370) implies that Equations (9.1372) (9. (3 and for several layers of material are discussed.1356. k' and X' are arbitrary. .T 0 ) = 0 (3(kkd)YX«YT = 0 onS^ onS 2 : X= 0 Xcc + k ^ . an infinite plate and an infinite cylinder with T d = T s = k d = 1.1373) (9.= 0 dn (9. 58) have to be satisfied. the average optimal thermal conductivity decreases slowly with increasing number of layers and increases with increasing heat transfer coefficient.1367) reduces to I JT'(T .SQ2 ADVANCED MATHEMATICS Equation (9. 58). Equation (9. Q = 0. Numerical results for two geometries. Equation (9.T o + Y . (k V X. This nonlinear system has been solved numerically and details are given in the original paper by Meric (1985).)) + k' [(3(kk d ) .1356.
..2. it costs (T .. Then determine the minimum surface area. We are required to find the straight line V = mT + c which will "best" approximate the data. 4a. A processing operation produces twenty cups per minute at 420 K.. if we denote the error at the point Vj by e. Find the radius r and the height h so that the surface area of the can is a minimum. i = 1.. determine the profit as a function of T. use the method of Lagrange multipliers. By considering E 2 as a i=l (ii) function of m and c. then ej = (V(Tj) . Use the following methods (i) determine the surface area of the can in terms of r and h and substitute r or h in terms of V and hence write the surface area as a function of one variable only. Tj). show that for E equations n n to be a minimum. Answer: h = 2r 3a. If the cups are sold at one dollar each. n. The constraint is the fixed volume V. A cylindrical can holds Vm 3 of soup. A volume temperature profile is represented by n experimental points (Vj.Vj).6). m and c must satisfy the m^Tj i=l + nc = Xvi i=l mlT^cXTj^TjVj i=l i=l i=l The above equations are known as normal equations (see Section 7. That is. What is the most profitable temperature to run the machine? Answer: T = 450K 2a.CALCULUS OF VARIATIONS $01 PROBLEMS la. An extra two cups per minute can be produced for each degree above 420 K. At a given temperature T.420) 2 15 + ———— dollars per minute to operate the machine. Evaluate the integral I = I ( y2 + x 2 y')dx Jo .Vj) = (mTj + c . n The "best" line can be defined such that E = ^ e 2 is a minimum.
0) to (1. Answer: cot (() = a cos (0 + c2) 7a. Answer: y = x 5a. 9.804 ADVANCED MATHEMATICS along the following curves joining the points (0. z = a cos () Let 0 = F((>) be a curve lying on the surface of the sphere joining two points A and B. The length of the curve is given by /•B rB I ds = I Vdx 2 + dy 2 + dz 2 JA JA Show that in the spherical polar coordinate system /•B . d) and is rotated about the xaxis. 1): (i) (ii) (iii) y=x y = x2 y = 1/2 (x + x 2 ) Answer: 0. 6b. Find the curve y (x) which extremizes the functional . y = a sin § sin 0 .667 Answer: 0.700 Answer: 0. y. cosh Any point (x. Answer: y = c.B I ds = a I V 1 + (F'(<(>))2 sin 2 ^ d^ JA JA By solving the EulerLagrange equation.675 Find the curve that will extremize I. determine the function F (())) that minimizes the length of the curve. deduce if I is a maximum or a minimum. The surface area S thus generated is given by /•b S = 27C I y V l + ( y ' ) 2 Ja dx (x — c ) — cl Find the curve y (x) which will extremize S. z) on the surface of a sphere of radius a can be expressed in a spherical polar coordinate system (r. c). (b. (])) as x = a sin § cos 0 . From the values of I determined using (i to iii). The curve y = y (x) passes through the points (a.
y ' 2 + y ' y + y ' +y)dx The boundary conditions are (i) (ii) 9a.73).10.. Answer: y = ^ . Answer: y = 2 x 3 . 8 a.^ . Find the extremum of /•I y'(0) = y ' ( l ) = l . y(O) = y(l) = O is equivalent to finding the extremum of I = I [y' 2 +y 2 + 2xy]dx Jo An approximate solution ya that satisfies the boundary conditions is given by ya = c o x ( l .x ) Substitute ya into I.9 and 9. y (1) is unspecified. I can now be considered as a function of cQ. Answer: y a = .3 x 2 + x 1 = 1 (^.CALCULUS OF VARIATIONS 805 I =( [y' + (y")2]dx Jo by the following methods (i) (ii) using Equation (9. + 1 2 7* Answer: y = x — — y+1 Show that the boundaryvalue problem y " . Determine c 0 by extremizing I.x ( l . using the method of constraints explained in Sections 9. Compare the approximate solution ya with the exact solution.y = x.x ) . y (0) = y (1) = 1 y (0) =1 . The boundary conditions are y(O) = y(l) = 0 .
V 1 + (y ) dx. The differential equation is known as the dxf V i=l I Schrodinger equation (see Section 6.4 .806 ADVANCED MATHEMATICS 10a. If C A is the concentration of A. (p is the potential.T)dx Jo . then the relation between C A and T is given by a kinetic equation of the form dC A _ e(C T) ~dT " g ( C A ' T) The production of A is given by 1= f(CA. Consider the irreversible reaction A — > B in a tubular reactor of length Z. Show that extremizing the functional 1 = 111 R I In" K + ^2 + ^3) + ^ dx l dx2 dx3 subject to the condition (II 2 I I I \/ dxl dx 2 dx 3 = 1 R leads to k2 2 V ~2nT V+9¥ = . 12b. Find the shortest distance from the origin (0.6). /•P / —^ Hint: the distance s = I x = 4 + 2y = <>(y). The constraint is known as the normalization condition. k and m are constants. y (0) = 0 and the point P is on the line Answer: ^V~5 lib. 0) to any point P on the line 2y = x .^ / 3 d2 \ The operator V 2 is the Laplacian =^ . We wish to determine the best operating temperature T as a function of the axial position x along the tube so as to maximize the production of B.
In Example 9.V 3 C A and the boundary conditions are: C A = 1 at x = 0 and T = T o at x = i . the dependent variables are C A and T. then vz = vz ^ ^ / ( a Ap) has to satisfy the following equations of motion V vz = . we have to minimize the production of A.113. Answer: T = 2 sinh 2x + V T cosh 2x + C (2 cosh 2x + V T sinh 2x) C = [ T 0 . we have to extremize I subject to the kinetic equation.CALCULUS OF VARIATIONS 807 Thus to maximize the production of B. that is to say. The flow is still an axial flow but the interface between the two liquids is not circular and the axial velocity v z is a function of r and 0. such as crude oil. The boundary conditions are . The equation of constraint is in the form of a differential equation and has to be satisfied at each point along the tube. The Lagrange multiplier is thus a function of x and is not a constant. Write down the Lagrange function L. = 9r 2 1 r 3r 1 3 ^ r 2 dQ2 1 7pis the where r\: is the viscosity of liquid i (i = 1. g = T . If v z is nondimensionalized and Ap is the pressure drop. V Laplacian. we used a variational method to solve the axial flow of a powerlaw fluid between two coaxial circular cylinders. In this example. In the present problem we consider the axial flow of two immiscible fluids in a circular pipe of radius a and length Z. Show that the EulerLagrange equations are 3C A 9C A 2 dx 2 / Determine T if f = C A + T . Bentwich has shown that the power required to transport a required volume of liquid.( 2 s i n h 2 i +V3~ cosh 2Jt)] /(2 cosh 21 +V3~ sinh 2 i ) 13b.1 2_ in phase (liquid) 1 r^/r^V vz = 1 2 _ in phase (liquid) 2 2 Tp. This flow situation was investigated by Bentwich (1976). 2). is reduced if a small volume of less viscous liquid is added to the crude.
Note that for one phase flow v z is given by: v z = 1 .808 ADVANCED MATHEMATICS v = 0 on the surface of the pipe which is taken to be at r = — = 1. (j)nm = Jn (A. ^ a At the interface Vzl 1 = V Z  2 9v z ni^rT 3vz j^^rT 2 where n is the outward normal to the surface. . the above expansion of vz corresponds to a Fourier series expansion. Show that the above boundary value problem corresponds to extremizing the functional Rj Rj+R. have to satisfy. On substituting the expression for v z in the double integral.2 where Rj denote the surface occupied by liquid i. We now require to extremize I with respect to Ao and A n m (properties of Jn are given in Section 2. Find the equations that extremize I and write down the algebraic equations that A o and A ^ .7 2 For a fixed 7. 1 and 2 indicate the evaluation is to be made at the interface in liquid 1 and 2 respectively. An approximate solution can be expressed as N Vz M = A0 % + X n=0 m=l X Anm V where ())0 = 1 .7 2 . Jn is the Bessel function of the first kind of order n.nm7) cosn0. A n m are constants which are to be determined.7). we obtain I as a function of Ao and A n m . ^ n m are the roots of the equations Jn (knm) = 0 and the coefficients A o .
2la. Let the vector position of particle i be denoted by r j relative to an origin O. Students used to deterministic concepts find it hard to think in terms of probability. The present chapter does not intend to provide such a course. each particle is defined by three coordinates. unlike the Cartesian coordinates need not all have the same physical dimension. The choice of the generalized coordinates depends on the geometry of the problem under consideration. then the two generalized coordinates are. a topic which is gaining widespread popularity. ql=r. as shown in Figure 10. q 2 .b) If the particle is constrained to move on a flat surface of the cylinder. It will be at an elementary level. q2 = 6 (10. we shall extend the analytical mechanics introduced in Chapter 9. Next. Our aim is more modest.1 INTRODUCTION It is not uncommon nowadays to require students in engineering to take a course in statistical mechanics. the only two generalized coordinates are ql=z. we shall consider probability. If there are m constraints. the number of generalized coordinates will be reduced to 3N .CHAPTER 10 SPECIAL TOPICS 10. say the surface z = h.21. If there is no constraint between the particles. We may regard the system to be in a 3N dimensional space and to have 3N degrees of freedom. Hamiltonian mechanics will be introduced.m.b) . This can also be used to pave the way for a course in chaos. The q1 are independent variables and are known as generalized coordinates.22a. We denote these 3N quantities by q1. In a threedimensional space. q2 = 6 (10...21. In the first few sections. We plan to provide the necessary background for the students to be able to follow with ease a more demanding course in statistical mechanics. . Finally. we need 3N quantities to describe the configuration of the system. some topics in thermodynamics and Brownian motion will be examined. but we shall emphasize the interpretation of probability.. The generalized coordinates. The idea of an ensemble average will be discussed.2 PHASE SPACE Consider a system consisting of N particles. q 3N . as shown in Figure 10. 10. Thus if a particle is forced to move on the curved surface of a cylinder of radius a.
. ..m) degrees of freedom and we have n generalized coordinates which are related to the N vector positions r l 5 r 2 . .. t) (10.23b) .23a) I N = rj^Cq^q 2 . r N .q n . .21 Motion of a particle on the curved and flat surface of a circular cylinder We now consider the system to have n (= 3N . . t) where t is the time. — ..6) r j = r ^ q ^ q 2 . The relationship can be expressed as (see Section 4..q n .810 ADVANCED MATHEMATICS z I I I ' . The velocity of particle i can be obtained by differentiating r j and we obtain (10. _1_* y X FIGURE 10..
i l _ 9r. as they refer to a particle i. (10. L is given by L = im(x2 + y2 + z2)(p It follows that ^ = mx = px dx In Equation (10.b) (10.25b) In Example 9. dr. I (10. z).k dr.25a) _. then in Cartesian coordinates (x. the indices denoting generalized coordinates are written as superscripts. Following the notation introduced in Chapter 4. p x is the xcomponent of the momentum. Consider a single particle of mass m moving in a potential field which depends on the position only.27b).24b). adopt a different notation.27a. we have deduced Lagrange's equation which is ^A/iM = aq1 dt \aq J j 0 (9. y.1034) where L = K . dr.b) 3qJ 9t In Equation (10. The kinetic energy of the system is given by K = 2 £ 1 m i . there is less of a compelling need to distinguish between subscripts and superscripts.103. Note that the indices on the vectors Vj and r i appear as subscripts. we have adopted the summation convention.26) . •.SPECIAL TOPICS 811 Vj = £ j = — q ] + ^ (10.i ' . Some authors dealing with analytical mechanics. They employ subscripts throughout. This suggests that a generalized momentum associated with the q1 coordinate can be defined as (10. As they consider mostly scalar quantities such as kinetic and potential energies.24a.cp is the Lagrangian and (p is the generalized potential.
p n . .. Let q be the displacement of the particle from its equilibrium position at any time t. .28) dq If the coordinate qj does not appear explicitly in L (qJ may be present in L). A particle of mass m is attached to the end of a linear spring of natural length /Co and of negligible mass.29) . q 1 . The Lagrangian is given by L = \ [m(q)2G(q)2] The equation of motion is (10. The dynamical state of the system is thus defined by the 2n quantities. qn and the momenta (mass x velocity) by the n generalized momenta p 1 . The mass is displaced from its equilibrium by a distance q 0 along the length of the spring and is then released from rest. which defines a space of 2n dimensions.. qn. The kinetic energy K = ±m(q) 2 The potential energy is 9 = iG(q) 2 where G is the modulus of the spring.. Determine the subsequent motion.21..812 P1 = .212) (10.7 7 ADVANCED MATHEMATICS (10. qi is known as a cyclic (or ignorable) coordinate. The configuration of the system is specified by the n generalized coordinates q1. The motion of the system is thus described by a path (trajectory or orbit) in this 2ndimensional space....210) (10. p 1 . .1034) simplifies to = pJ = constant aqj Thus the generalized momentum of a cyclic coordinate is a constant... Equation (9. Example 10. . . .211) (10. The spring is fixed at the other end and lies on a smooth horizontal table. known as the phase space. p n . .
1 ^[m(q)2G(q)2] .216a.G q . the position and momentum of the particle is given by a point on the ellipse.215a) (10. m Equation (10.^ [ ^ .213c) where co2 = —.213a) or or .216b.217) Thus the trajectory in the phase space is an ellipse (closed curve).213b) (10. as shown in Figure 10.G ( q ) 2 ]  = 0 (10. The solution of Equation (10.213c) is the simple harmonic equation and is the equation governing the motion of all linear oscillators.SPECIAL TOPICS 813. Thus the ellipse describes the dynamical state of the particle (or oscillator).216a) (10.213c) is q = A cos cot + B sin cot O sin cot + B co cos cot q = A C where A and B are constants.214a) (10. At any time t.214b) f—) +(—— \ = 1 \mq0co/ \q 0 / (10.215b) (10.d) (10. All oscillators with the same total energy .A(mq) = 0 q+co2q = 0 (10. Using the initial conditions q = q0 q = 0 we obtain q = q 0 cos cot p = — = mq = mcoq 0 sincot 3q Eliminating the time t between Equations (10.c.[ m ( 4 ) 2 .22. d) yields (10.
. q1. p n ). (qn. In the phase space.. If we have n oscillators not necessarily all with the same energy. f FIGURE 10. the variables are (q1. t) system.. p 1 ).. It is desirable to transform from the (q1. the variables are (q1. .32) Differentiating Equation (10. t).22 The trajectory of an oscillator 1 0 . 3 HAMILTON'S EQUATIONS OF MOTION In Lagrangian dynamics.31). pJ).L ( q J .. we have . q J . pi) corresponds to an oscillator.814 ADVANCED MATHEMATICS describe the same ellipse. This can be accomplished by the use of the Legendre transformation.. we have 2n variables (q 1 . qj. The differential of H is given by ffi = j = l.31) 9H dt + 3H d q j 3qJ + 3H d p J 9pJ (10. t) system to the (q1.n (10. p 1 .t) = p m q m . t ) . and we sum over m. q1.. Each set of (q1.2. The starting point on the ellipse depends on the initial conditions. A new function H known as the Hamiltonian is defined as H(pJ.
we deduce that K is a quadratic function in q J .28).^ d q j . .35) (10. N .33C) = qmdpm^dqJ^ aqj at Equation (10. Then from Equation (10.— aqj 3L 3H (10. .4 1 d q j . we deduce qj = — aPJ pj = .^\ dqj + q m dpm .34b.ar = ^ (1036c) Equations (10. We also assume that the system is conservative. From Equations (9.36a to c) are known as Hamilton's equations of motion.28) may be written as .p j dqj .23).36a) (10.^ (10.33c) is obtained using Equation (10.^ dqJ . They are a set of first order equations.36b) „ .25b). 5).c) (10.28). then < p is a function of qj but not of q J .^ 9qJ 3q J 9t (10. 10.34a) (10.34c) into Equation (10. We now consider the case where t does not occur explicitly in Equation (10. It then follows that Equation (10.33b) \ aqJj aqj at (10.^ at Comparing Equations (10. we have *h=*m aqJ dt[aqJ = ft (Pj) = P j Substituting Equation (10. whereas Lagrange's equations are second order equations. Note also the symmetry in pj and qj.32.SPECIAL TOPICS 8H dH = pmdqm + q m d p m .33a) = fpi .33c) yields dH = q j dpj .1034. Such a situation may arise if the constraints are time independent.
212). The generalized momentum p is given by p = — = mq dq Thus H is given by H = p q .37a.b) (10. Using Euler's theorem (Section 1. The Lagrangian L is given by Equation (10.b) (10.7).3lla.39a) (10. b).816 ADVANCED MATHEMATICS Pm = A ( K ~ < P ) = : ~ dq dq K is a quadratic in q m .31. Obtain the Hamiltonian H from Equation (10.31 lc) (10.312b) .38a.b) = \ ^ T + G(q)2 From Equations (10.310a.L = m(q)2±[m(q)2G(q)2] = ±[m(q) 2 + G(q) 2 ] (10.39b) Thus the Hamiltonian H represents the total energy of the system and is a constant for the system under consideration.213c).b) q m p m = q m 4  .31 Id) q=£ p = Gq (10. Example 10. Write down Hamilton's equations of motion and show that they reduced to Equation (10.38b) into Equation (10. we have (10.31) for the oscillator described in Example 10.21.312a) (10.= 2K 3q Substituting Equation (10.36a.31) yields H = 2K(K(p) = K + (p (10. we have (10.
q').312a. df . b.4.i rD + rO .. f] dt where [H. 14c). q1. 10. q 2 . Equation (10. p 2 .314a.314c) (10. . p n ..2a) dq1 dp1 3p! dql (10.41) dt 3t dp1 9qJ Combining Equations (10. we observe as expected that H is a .4 POISSON BRACKETS Let f be a function of p 1 . and Co = G q 0 2 q = q0 (10. If f does not depend explicitly on t and the Poisson b