This action might not be possible to undo. Are you sure you want to continue?
5.0
Views: 647Likes: 27https://www.scribd.com/doc/62146766/AdvancedMathematicsforEngineering
05/28/2014
text
original
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
12 Homogeneous FirstOrder Differential Equations 1.18 Solutions Using Green's Functions 1.11 Separable FirstOrder Differential Equations 1.10 FirstOrder Differential Equations 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 1.9 Ordinary Differential Equations (O.15 Bernoulli's Equation 1.E.1 1.CONTENTS Chapter 1 1.6 Implicit Functions 1.3 Integrals Integration by Parts Integration by Substitution Integration of Rational Functions 1.) .5 Derivatives Total Derivatives 1.19 Modelling of Physical Systems Problems Chapter 2 2.Definitions 1.16 SecondOrder Linear Differential Equations with Constant Coefficients 1.13 Total or Exact FirstOrder Differential Equations 1.7 Some Theorems Euler's Theorem Taylor's Theorem 1.1 Series Solutions and Special Functions Definitions .8 Integral of a Function Depending on a Parameter 1.14 Linear FirstOrder Differential Equations 1.17 S olutions by Laplace Transform Heaviside Step Function and Dirac Delta Function 1.D.4 Functions of Several Variables 1.
2 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 2.5 2.4 3.8 Fourier Series Fourier Integral 2.2 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 3.7 3.8 .1 3.9 Asymptotic Solutions Parameter Expansion Problems Chapter 3 3.6 2.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 2.3 3.6 3.3 2.4 2.
Contravariant and Physical Components 4.7 Covariant Differentiation Properties of Christoffel Symbols 4.5 Applications Conservation of Mass Solution of Poisson's Equation NonExistence of Periodic Solutions Maxwell's Equations 4.3 5. Objective Tensors and TensorValued Functions Problems Chapter 5 5. Surface and Volume Integrals Gauss' (Divergence) Theorem Stokes' Theorem 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.1 5.9 Isotropic.CONTENTS fe Joukowski Transformation Problems Chapter 4 4.4 .3 4.1 4. Surface and Volume Integrals Relations Between Line.8 Integral Transforms 4.2 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 5.2 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.
3 7.9 Fourier Transforms 5.11 Summary Problems Chapter 6 6.1 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 7.10 Hankel and Mellin Transforms 5.4 7.2 6.5 Green's Functions for General Linear Operators 6.6 .3 6.5 Cylindrical and Spherical Polar Coordinate Systems 5.6 Boundary and Initial Conditions 5.6 Quantum Mechanics Limitations of Newtonian Mechanics Schrodinger Equation Problems Chapter 7 7.8 Laplace Transforms 5.4 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.x ADVANCED MA THEMA TICS 5.7 NonHomogeneous Problems 5.5 7.1 6.
7 8.2 8.7 Numerical Differentiation and Integration Numerical Differentiation Numerical Integration 7.10 Stability Problems Chapter 8 8.8 . 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 7.6 8.9 Boundary Value Problems Shooting Method Finite Difference Method 7.5 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 8.1 8.8 Numerical Solution of Ordinary Differential Equations.CONTENTS Xi Lagrange Interpolation Newton's Divided Difference Representation Spline Functions Least Squares Approximation 7.4 8.
13 Maxwell Velocity Distribution 10.12 The Equipartition Theorem 10.11 Several Independent Variables 9.2 Phase Space 10.1 Introduction 10.13 Subsidiary Conditions Where the Functional Depends on Several Functions Problems Chapter 10 Special Topics 10.4 Poisson Brackets 10.10 Basic Assumptions 10.3 9.9 Constraints 9.9 Scope of Statistical Mechanics 10.8 Transversality (Moving Boundary) Conditions 9. Poisson and Normal Distribution 10.xij ADVA NCED MA THEM A TICS Galerkin method TwoDimensional Problems Problems Chapter 9 9.7 Extension to Higher Derivatives 9.1 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' 9.8 Binomial.12 Transversality Conditions Where the Functional Depends on Several Functions 9.4 9.3 Hamilton's Equations of Motion 10.5 Canonical Transformations 10.11 Statistical Thermodynamics 10.10 Several Dependent Variables 9.2 9.6 Liouville's Theorem 10.14 Brownian Motion Problems .7 Discrete Probability Theory 10.5 9.
The equation of continuity of species A in several coordinate systems Author Index Subject Index 861 867 867 868 871 873 875 877 .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 .CONTENTS xiii References Appendices Appendix I .The equation of continuity in several coordinate systems Appendix II .
.
But we can draw up a table listing the geographical positions of all gas stations and the price charged at each gas station. It is customary to denote a function by f.b. The range of f is the set of values that f can acquire over its domain. Thus.1 FUNCTIONS OF ONE REAL VARIABLE The search for functional relationships between variables is one of the aims of science.11). we shall start by considering two real variables.c) (l. 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. to each of the two variables we can associate a set of real numbers. A function does not always need to be expressed as an algebraic expression as in Equation (1.CHAPTER 1 REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 1.11 illustrates the concept of domain and range. Thus if the rule is to square.l3a. For simplicity. The rule which assigns to each real number of one set a number of the other set is called a function.c) .b. f is not applicable to negative numbers. It is important not to confuse the function (rule) f with the value f(x) of that function at a point x. Figure 1. These variables can be quantified.lla. The domain of f in this case is the set of nonnegative numbers. 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 = . The function f might not be applicable (defined) over all real numbers. the price of a litre of gas is a function of the geographical location of the gas station.b) The variable y is known as the dependent variable and x is the independent variable. The set of numbers for which f is applicable is called the domain of f. The function f is said to be even if f (x) = f (x) and odd if f(x) = f(x). Thus. It is not obvious that we can express this function as an algebraic expression. Thus if f is extracting the square root of a real number.l2a.x 3 = f(x) (l. 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. that is to say. we write y = f(x) = x 2 (l. For example.
it is continuous at every point x in the interval a < x < b.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).11 • j ^ X Domain and range of a function An example of a periodic function is sinx and its period is 2K.xol < 8.l5a.n MATHEMATICS A function is periodic and of period T if f(x + T) = f(x) (1.. An alternative equivalent definition of continuity of f (x) at x = x 0 is. given e > 0. b]. there exists a number 8 (which can be a function of 8) such that whenever I x .14) y . then  f ( x ) .16) . If the function f (x) is continuous in a closed interval [a. (1.f ( x o )  <£ This is illustrated in Figure 1.12.2 ADVANCF. RANGE OF f I \ i \ I y= f (X ) >w "I""" I L _ DOMAIN OF f FIGURE 1. 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.
.f (x0) Ax ( i 2 i ) The rate of change of f at x 0 . Likewise higher dx 2 derivatives can be defined and the n th derivative is written either as f ^ or dx n Geometrically. . The average rate of change of f (x) in an interval [x0 + Ax.22a. dx d2f The second derivative of f is the derivative of f and is denoted by f" or . not only the position of a car but also its speed before crossing the road. For example. which is the derivative of f at x 0 . 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.b) exists and f is thus differentiable at x = x 0 . x 0 ] is defined as Af Ax = f (xQ + Ax) . f'(x 0 ) is the tangent to the curve f(x) at the point x = x 0 .12 1. one might want to know.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 3 ^o+€ y^ ijl *oa! i !««*> ^ X FIGURE 1. The derivative of f with respect to x is also denoted as 4*. waiting at the corner of "walk and don't walk".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.
b).28) reduces to Equation (1. Conversely. Mean Value Theorem If f(x) is continuous in the closed interval a < x < b . then f(x) is an increasing function. n! [= n (n . b) such that f(b)f(a) g(b)g(a) = f'(c) ( 1 2 8 ) g'(c) If g(x) = x. b). then f is a constant.4) f(uv) = u^v+vdu dx d > ) dx11 = u (1. then Equation (1. 1] is the factorial of n.27). (nr)!r! Rule (iv) is known as Leibnitz rule (one of them!). such that f'(c)M b—a (1 .1). f(x) is a decreasing function of x. as x increases f(x) increases. b).b). that is to say. and f(x) is differentiable in the open interval a < x < b. If f'(c) > 0 for every c in (a.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. there exists a point c in (a. Cauchy Mean Value Theorem If f and g are continuous in [a..27) From Equation (1.27).23) (ii) _udv _dx_^x vdu_ (L2.. b). there exist a number c in (a.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— . . if f'(c)<0 for every c in(a. we deduce that if f'(c) = 0 for every c in (a. b] and differentiable in (a.
The x»x 0 x ^ x o x~*xo S( x ) rule holds for x —> °° or x —> °°..x 0 ) f'(x 0 ) + (X " 2XQ) where R n is the remainder term. Equation (1.is indeterminate. = . x^x o g(x. x).2.does not exist but lim x>x0 g(n) ^ x^x 0 g (x) replaced by to M. Note that we are expanding about a point x 0 which belongs to an interval (x 0 . Other indeterminate forms.29) can be If lim ——1.29) g'( x ) f'(x) f(n) (x) — exists for some value of n. The . must first be reduced to one of the indeterminate forms discussed here before applying the rule. Lagrange's expression which may be written as ( R f "(X 0 ) + . 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. Therefore.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 5 L'Hopital's Rule lim f(x) = 0 and lim g(x) = 0.10) ffx) The same rule applies if the lim f (x) = °° and lim g(x) = °o. Taylor's Theorem If f(x) is continuous and differentiable f (x) = f (x0) + (x . + (X " J ° ) n f(n) (x 0 ) + R n (1.211) The simplest one is probably \n+^ ° (n+1)! f ( n + 1 ) [*o + 6(*xo)] 0212) where 0 < 6 < 1. and represents the error made by truncating the series at the n * term. There are various ways of expressing the remainder term R n . the lim y^r is indeterminate. ta V®. the lim ~~.. x»x o g(n)( x ) (1. such as the difference of two quantities tending to infinity. the remainder term for each point x in the interval will generally be different. R n is the result of a summation of the remaining terms.
f has a minimum at x 0 and if Af is negative. If Af is positive. we see that Equation (1.f " ( x 0 ) + . The Taylor series expansion is widely used as a method of approximating a function by a polynomial..211).21 Extremum of a function f . 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. Figure 1. h2 (1. Thus. f has a maximum at x 0 .214) y< MAX.211).f ( x o ) must have the same sign irrespective of the sign of h. From Equation (1.21 defines such extrema. determines the value of 0 in Equation (1. y=f(x) /\ j ^ Af\^l h h j j I h h x FIGURE 1. and if x 0 is the origin. at different values of x within the considered interval. we have expanded f(x) about the point xQ.213) can be written as Af = hf(x o ) + ^ .212).213) (1.6 ADVANCED MATHEMATICS maximum truncation error. if the function f has an extremum at xQ Af = f ( x o + h ) . In Equation (1. where h = x ..x 0 . we are dealing with Maclaurin series. associated with the evaluation of a function.
x 3 .215) But if f "( x o) = 0> we cannot deduce that f has an extreme value at xQ.21. Thus. F (x) is arbitrary to the extent of an arbitrary constant. it is a point of inflection. if they exist. f (x) has an extreme value at x 0 if n is even. On differentiating. = f ^~^ (x 0 ) = 0.217) (1. as can be seen by drawing the curve given by f (x) = x 3 . f has a maximum at x = x o and if f (n) (x0) > 0. The integral we have defined is known as an indefinite integral which is usually denoted by the symbol I . we see that f(O) = f"(O) = O (1. Since the derivative of a constant is zero. 1. the derivative of ^x 3 is x 2 .216) (1. Example 1. f"(x o )<O. we have f'(x) = 3x 2 . Thus.b). fhasaminimum f hasamaximum (1.b.. From the criteria given earlier we deduce that f does not have an extreme value at the origin.218a. an integral of f(x) is F(x).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.3 INTEGRALS An integral can be considered to be an antiderivative. If f( n )(x 0 )<0. and if f ^ ( x 0 ) exists and is nonzero. but f'(x 0 ) = f"(x 0 ) = . Thus. Find the extreme values of f (x) = x3.b) Thus we need to consider higher derivatives and the next one f"'(0) happens to be nonzero. We need to consider higher derivatives until we obtain a f (n) (x 0 ) which is nonzero. if we know that the derivative of F (x) is f(x) [=F'(x)]. For example.218 a.219a. Note that we have used the article an. the general criteria for extreme values are if f(x) is defined in [a. If n is odd.. f(x) does not have an extreme value at x = x 0 . we write .b] and x 0 is an interior point of (a. f'"(x) = 6 (1. b).c) From Equation (1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS _J_ From Equation (1. f has a minimum at x = x Q . and an integral of x 2 is ^. The origin is neither a maximum nor a minimum. f"(x) = 6x.
3la.32) defines a definite integral. we need to fix the ordinates. We integrate by parts.31) to a definite integral if x = b.b) Thus to evaluate a definite integral analytically. There are tables of integrals. where the indefinite integrals of standard functions are given. identifying from Equation (1. we have . Carrying Q. Below we list some of the general methods of integration.b) define a function of x in terms of a dummy variable t. Integrate I e ax sin bx dx.X out the integration. x = a and x = b. x = b A= /a f(x)dx (1. Thus. if A is the area bounded by the curve y = f(x).b) where a is an arbitrary constant of integration. the limits x = a and x = b are given.8 ADVANCED MATHEMATICS F(x)= I f(x)dx= I f(t)dt (1.34) fdx = [fg]f gdx (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.31. such as.35) Example 1. Integration by Parts If f and g are functions of x £(fg) = fg + fg' It follows from Equation (1.33a. In this case. For the area to be definite. Equations (1. we first need to find an indefinite integral.3la. We can convert the indefinite integral in Equation (1.32) Equation (1. the xaxis and the ordinates x = a. we usually write F(b)F(a) = Ja f(x)dx = [F(x)J a (1.34) that (1.
38) (1.32. we obtain (1.313a) . 7) yields 1+— Hence ( I e^ sin bx dx = — [a sin bx .^ I e a x sinbxdx (1.36) I e ^ c o s b x d x = \~ sinbx .39) It follows from Equations (1.310a. dx = 4i'(z)dz (1.1 [e ax cos bx] + 3.37) Combining Equations (1.311) Example 1.310a. Integrate I V a 2 .b) that I f(x) dx = I f [<)(z)] Q (z) dz (1.a2 sin2 z ) a cos z dz (1.312) I Va 2 .36.b) I e ^ sin bx dx = . Since x = <>(z).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. where a is a constant.x 2 dx = I (Va2 .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.je ax sin bx] (1.x 2 d x . so that dx = acoszdz (1.
c) =  . J x 3 8 The function 5x + 2 can be expressed as a sum of partial fractions as follows.315) a • In evaluating finite integrals. the key is to find a substitution such that the integral is reduced to a standard form. it is often simpler to express the limits of integration in terms of the new variable z. B and C are constants. Integrate I 5x + 2 dx.x 2 + arc sin (*) 2 (1.314b.33. Example 1.*.Va 2 . Integration of Rational Functions A rational function of x is a function of the form f(x)/g(x).c) J [a 2 (1.W ADVANCED MATHEMATICS = a2 I cos2z dz = y I (cos 2z + 1) dz (1. we obtain .313b. we have z = arcsin() sin2z = 2 sin z cos z = a V 1 ~ ( f ) Thus r 2 I Va 2 .313d) (1. By comparing powers of x.314a) (1.[ i s i n 2 z + z] Returning to the original variable x.b) x2 where A. In the method of substitution. The rational function can be expressed as a sum of partial fractions and can thus be integrated.316a. x38 5* + 2 x38 = 5x + 2 (x2)(x 2 + 2x + 4) =_A__+ Bx + C x2 + 2x + 4 (1. where f(x) and g(x) are polynomials in x.x 2 dx = §.
321a. dx = V3" sec2 9 d9 = (1.b. C=l (1.c.2) .325) .b) (x + 1) = V3~ tan 0 .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) 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 (1.320a.319) The second integral on the right side of Equation (1.c) f 5x+^ dx = f _dx_ + [ d .b.i n (x2 + 2x + 4) + 2= arc tan ( ^ l j (1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS ]± A=l.317a. dz = (2x + 2)dx (1. we obtain i Q ^ arc tan [ x + 1 \ VI VJ ' VI ' (1.c) To evaluate Ij.318) J x2 "x Zt j x2 + 2x + 4 is standard and x I ^ =*n(x2) (1. we let = 2 1 [ dj J Z = l i n ( x 2 + 2x + 4) 2 (1.x ) d x Jx38 The integral I j (1.324a. Thus B=l.319 to 24d).d) I 5x + 2 dx = i n (x .323a.b) ( (x + l)dx J x2 + 2x + 4 To evaluate I 2 . we make the following substitution z = x 2 + 2x + 4 .322a.b.^.
b) m(b .a) < I f(x) dx < M(b .b) The above mean value theorems provide bounds on integrals and can be useful in error analysis.b . and for g ( x ) > 0 every where in (a. 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. Contour integration.b .b) 0. Some Theorems . Currently. f(x)dx Jb (1. f(x)dx= \ if f(x) is even (1.b .b) (vi) Generalized first mean value theorem Under the conditions on f(x) given in (v).328a. in the theory of complex analysis (Chapter 3) can be used to evaluate real integrals. b) .33Oa.326) r (ii) (b r (1.a (i) Ja f ( x ) d x = .b m g(x)dx < I f ( x ) g ( x ) d x < M g(x) dx Ja Ja Ja (1.12 ADVANCED MATHEMATICS In the past.33 la. considerable efforts were devoted to finding methods to express integrals in closed form and in terms of elementary functions. a popular approach is to resort to numerical methods (Chapter 7).327) 2 I f(x)dx.a) Ja (1.329) (v) First mean value theorem If M and m are the upper and lower bounds respectively of f(x) in (a. .
y0) if given e > 0. we often encounter one variable which depends on several other independent variables. The function f (x. holding y fixed. 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. y) one can assign a value of u. 1. They are defined as . . f or u Y . y is treated as a constant. y o )  <£ 1.41) The domain of f is the set of values of (x. For simplicity. Since fx and f are functions of x and y. such that whenever V(xxo)2 + (yyo)2 < 5  f ( x . This is the partial derivative of u with respect to x and is denoted by ^r—. Similarly to compute jr. there exists a 8. The range of f is the set of values that u may have over the domain of f. In most cases. y ) .42) (1. In this case we write u = f(x. x n is obvious.4 FUNCTIONS OF SEVERAL VARIABLES So far. the volume of gas depends on the temperature and the pressure. the extension to n variables Xj.y)f(x...y) (1. In science and engineering. we may calculate the rate of change of u with respect to x. we have considered functions of one variable only.y) Ay (1 5 2 ) 8f The computation of ^— is the same as in the case of one independent variable. y) over which f is applicable. ^— is defined as dy 9f oy = lim f(x. Here.y) Ax (1 5 1 ) Similarly. Thus J dx dx x x df=]im ox AX>0 f(x + Ax. For example. y) is continuous at (x 0 .5 DERIVATIVES (1. .REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 13. we shall consider functions of two independent variables x and y.f ( x 0 .43) Since u is a function of two variables x and y. we consider x to be a constant.y + Ay40 Ay)f(x. Other notations are: ^—. x2. their partial derivatives with respect to x and y may exist.
53) a [*) dy \ dx / = lm A y>0 y^±Mzifcy> Ay (1. where the vector n is parallel to the line along which we wish to determine the rate of change.4) The secondorder partial derivatives are denoted as _(L] = ^Uf dx \dx ) 5x2 xx (l55a. this is generally the case and fxy = fyx (1. Such a rate of change is known as a directional derivative and is denoted by 5—.5. If the partial derivatives are continuous.y) i AX>0 Ax (1. whereas for f the order of differentiation is reversed.(x.59) Likewise higher partial derivatives can be defined and computed. In an xycoordinate system. the first order partial derivatives fx and f may be regarded as the rate of change of f along the x and yaxis respectively. then the order of differentiation is not important. One may wonder if the order of differentiation is important.(^Ax.b) (1.y)f.56a. In practice. If f is continuous then the order is not important.b) We note that f means taking the partial derivative of f with respect to x first and then with respect to y.14 ADVANCED MATHEMATICS j(£U dx \ dx / lm f.=fx y (L5"7a'b) _fl)=^l=f (i. Thus 5— at a point (x0. 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 .58a. We can also define and compute the rate of change of f along any arbitrary line in the xyplane.b) f ff)=^=f dy \ dx / dydx ^x d7(d7) = axay.
f (xQ.51 la) Note that. y o +psin6) . This situation is illustrated in Figure 1. we have kept y 0 + p sin 6 = y } constant. y o +psin9) + f (x 0 . y 0 ) an p>o L p (1.51 la). y0). in the first two terms on the right side of Equation (1.51 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. 3f = ]im f (xQ + p cos 6.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 75 where p is the distance of any point on the line from (x0. y * i i I FIGURE 1.f(x 0 . y o +p sin6). y{) . y i ) . y 0 ) (1 5llb) dn p>0 L P J .f (x 0 .51.f ( x Q . y i ) + f (xQ.
b) The existence of df guarantees the existence of fx and f .514a) (1.f (x.f ( x o .514b). =^sin0 dy (1.512b) Therefore ~ =~ cos0 + ^ sin0 dn dx dy Thus if fv and fv are known. we can compute ^r.16 ADVANCED MATHEMATICS This amounts to considering f to be a function of x only in the first two terms. one can consider f to be 9f a function of y only in the last two terms. y + Ay) . Au. Then Au = f (x + Ax. Ay —> 0.. For df to exist we require not only the existence of fx and f . dnf can also be defined. Thus . Expanding this function of x in a Taylor series yields: f(xQ. y) (1. Higher differentials d2f.. but the converse is not true.515a.f (x. lim p>0 L y o ) + ~ p s i n 0 . d3f.f (x. The differential df may be regarded as a function of 4 independent variables x. resulting in: f (xQ. y2) lim p>0 L P .512a) and similarly f (x 0 .f ( x o .514b) The observations made following Equation (1. A (1. and on taking the limits Ax —> 0. when both x and y change simultaneously to x + Ax and to y + Ay respectively. we obtain du = df = ^ dx + ^ dy dx dy (1. =^cos0 J °x (1. but we also require f to be continuous. yj)+ ^—p cos 0. y) = f (x + Ax. . y + Ay) . We deduce that f (x 0 . y + Ay) . y + Ay) + f (x. y!) + ^ p c o s 0 .511) are applicable to Equation (1. Similarly.513) y an Total Derivatives We now determine the change in u. y. dx and dy. y0) + ^— p sin 0. . yo) ^ p J .
U — d x dx + ^ .U .518). dxnrdyr y ayn (dy)» y' (1. + ^L ax n I ! / ax^^y l r .^ L _ (dx)11"1 dy + ..—dy dx + ^ .520a) (1. d (df A\A d f = T .19) v dt 3x dt dy dt If x and y are functions of another set of independent variables r and s. d [df A .(dx)nr (dy) r+ . the right side can be expanded formally as a binomial expansion. ^ (1..d y dy dxldx J dy\dx j J dx\dy Jj dy \ dy J 0 0 0 „ * .b) Substituting d by ^— dx + ^— dy yields dx dy A2* ^ /df .515).d x dy + ^ ..516d) = — (dx)2 + 2 ^ ^ d x d y + — (dy)2 3x2 dxdy 9y2 It can be shown by induction that dnf = ^ (dx)n + f " ) . we obtain / df 8x du = (d7 ' (1.521) that .516a.d s 3r ds Substituting dx and dy in Equation (1. If both x and y are functions of another variable t. (L521) ar+d7 ^ ) d r + [dx. then dx = ^ d r + ^ . I df dx df dy \ . one can rewrite it as d n f = (s d x + iH" f In Equation (1.d s dr ds dy = ^ d r + ^ . + ( n ) ^*.5. then from Equation (1. d /df ..a^ + d7 i)ds It follows from Equation (1.516c) (1. \ . \ .517) In order to remember Equation (1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS IJ_ d 2 f = d(df) = d ( ^ dx) + d ( ^ dy) (1.520b) 8f 3y \ .515) we have (51) 18 *L = 3£dx £& (1.^ .517).
f is given by f = x 2 + y2 We change to polar coordinates (r.5. we have 3 f _ 3 f 3x 3f 3y 3r " 3 x 3 r + 3 y 3r 3f=3f3x 30 3x 30 + (1. c.525a.r sin 0 .527c.524a. n .522a. dr — = . d) again express the chain rule. 9).526e.f) Substituting Equation (1.525a) (1525b) 3f3y 3y 30 Computing the partial derivatives yields ^=2x. In rectangular Cartesian coordinates system.523) y = rsin0 (1.b) 3f = 3£3x + 3f 3y ds 3x 3s 3y 3s (1522cd) Equations (1.d) . 3f . b. we obtain 3f j . .b) (1.526a. . b).522a.= 2x cos 0 + 2y sin 0 = 2r — = 2x (r sin 0) + 2y (r cos 0) = 0 30 (1.51.527a. c.b) From Equations (1.18 ADVANCED MATHEMATICS * dr 3u 9s = * dr **tLl dx dr dy dr (1 .22a .526a to f) into Equations (1. Example 1. b. 30 ?=2y dy ^ = sin 0 dr ^ = r cos 0 30 (1.526c. The transformation equations are x = rcos0. — dr 30 (1. d). dx ~ = cos 0 .b) (1. 3f Calculate ^ and — .d) (1.
61) [Vnb]nRT = 0 (1.523) and thus f is expressed explicitly as a function of r and 0 and the partial differentiation can be carried out.VT) = 0 where P is the pressure.62) partially oT with respect to P and then deduce 5= . c.62) and express T as a function of P and V. It is simpler to differentiate the expression in Equation (1. But as can be seen from Equation (1. the resulting function will be even more complicated than Equation (1. we can think of x = P.64a. V) is usually written. d) can be obtained by substituting Equations (1. R is the gas constant and n the number of moles. we can obtain ^ and other partial derivatives. it implies solving a cubic equation. V is the volume and T is the temperature. y and u.62) it is not easy to solve for T.63) The variables x. 1. For example.515).62) by partial differentiation. y and u are not all independent. In many cases. y)].u) = 0 In terms of the thermodynamic example.b. u=T (1. implicitly as f(P. We shall show how this is done by reverting to the variables x. y = V. The twoparameter Redlich and Kwong (1949) equation is expressed as P+ — T 1 / 2 V(V + nb) where a and b are two parameters. Even if we solve for T. the substitution can be very complicated. Since f = 0.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS _£9 Equations (1. we can solve for T in Equation (1.6 IMPLICIT FUNCTIONS So far we have considered u as an explicit function of x and y [u = f (x.y. and finding ^ r or (say) will be time consuming. Then rfT (1.524) into Equation (1. df = 0 and from Equation (1. we obtain . We now consider an implicit function written as f(x. There are examples where it is more convenient to express u implicitly as a function of x and y.c) (1. We are free to choose any one of them as the dependent variable.62). In theory. in thermodynamics an equation of state which could be given as T = f (P.527a. b.
67) 3u du We could equally obtain 5—. y and u are related by another equation written as (1.b) du The partial derivative ^— is obtained from Equation (1.63). We choose the only independent variable to be x.610) we essentially have two equations involving three variables.20 ADVANCED MATHEMATICS df = ^ d x + ^ d y + ^ d u = O 3x dy du Equation (1.65) by putting dy = 0 (since y is kept as a c constant). Differentiating f and g with respect to x.6lla) (16llb) . Thus taking u as the dependent variable and differentiating partially with respect to x.y. (1. v~ by differentiating f from Equation (1. b). we have (1.63) and using the chain rule (Equations 1.66) I~(S)'(I) (1.68) It then follows that id)/(l) Similarly ^— and higher derivatives such as can be computed./ 5 ox I \ du (1.65) is still true for f = constant.69) If. g(x. since df = 0 in this case also. the variables x. in addition to 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 (1.522a. u) = 0 (1. So 5=ox Similarly 5 .65a.
(174) Choosing a = 1.74) becomes xH+y+udl=nf(x'y'u) (L75) . y.b.71) becomes f(x*. b) form a system of two algebraic equations involving two unknowns ^^—. 3x dx 1. The solutions are dx and (1. u* = au (1.J—Jr— By* da du* da =na n l c( (1. the Jacobian J must not vanish. we write x* = ax. u) is a homogenous function of degree n if f(ax. y..REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 27 Equations (1.c) (171) Equation (1.+ n . ay.612b) J = fy g y fu (1. u) Defining new variables. Equation (1.y. which we indicate by a star (*).72a.u) Differentiating Equation (1. y. y*=ay.612a) fx fy (1. f(x.73) .61 la.u*) = a n f(x.612c) v o &u ' 3y 9u Thus for ^— and ^— to exist. u) .7 SOME THEOREMS Euler's Theorem A function f (x.73) with respect to the parameter a yields df dx dx* da df dy df du + — .y*. au) = a n f ( x .
y 0 + Ok) and 0 < 9 < l . Thus if we integrate f(x. y.y0) + {h^ + k  } + l j h 2 g + 2 h k ^ + k 2 g j + ....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. into Equation (1.y)dy (1. we have IW = Ja f(x. y) between two fixed points y = a and y = b. ( n . otherwise we would be trying to differentiate by a constant.76) Taylor's Theorem Taylor's theorem for functions of one variable can be extended to functions of several variables.8 INTEGRAL OF A FUNCTION DEPENDING ON A PARAMETER The function f (x. 1.. + k" ll\ + R (1.. We then obtain an integral which is a function of x and we may consider x as a parameter.22 \ * ADVANCED MATHEMATICS Note that we substitute (= x) etc.78) at the point (x 0 + 0h. . f(x0+h. y. For simplicity we give the formula for two independent variables. u) (1. u) = n ( n . The generalization of Equation (1.1) .y0+k) = f(x0.74) before setting a = 1. y) is a function of two variables x and y and we may integrate the function with respect to y holding x fixed.77) are to be evaluated at the point (x 0 . + _L (h» ^L + (n\ h nl k __lL_ + + (n) hnr kr _ 3 ^ _ + .75) is known as Euler's theorem..r + 1) f (x. y 0 ) and those in Equation (1..81) . Equation (1.
identifying y with t and v (x) with x.85) yu(x) Note that u(x) and v(x) are not fixed! To evaluate the partial derivative —. Using the results in Section 1.b.b) Similarly .REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 21 Differentiating I with respect to x results in (1.83) Note that if we were integrating at another point on the same curve. y)dy .86a. we have 1 =   I Ju (x) f(x.84) 31 dl 91 From the definitions of the partial derivatives 5—.31) yields dv f x ^ p =)a f(t)dt = ^[F(x)F(a)] =F'(x) = f(x) (l.82) /•v(x) (1. we have.c.82) Ja If the limits of integration are not fixed but are functions of x.U(X) (1. dL is given by dx di_ dx = a i _ + a ] [ d v + ai_du dx dv dx du dx (1. v(x)] (l. 5— and v~.5. the limits of integration would read u(xi) and u(x2) which is equivalent to integrating from a to b. v(x) and u(x). 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 fix the variables x and u(x). I may be treated as a function of three variables x.d) Therefore. Equation (1. via Equation (1.y)dy = f[x.87a.
89).^  (1.81. Equation (1. Example 1.812b) that . (1.89) /u(x) Equation (1.81 la) = \~\ +2x 4 x 2 (1. integrate and obtain I explicitly as a function of x and then differentiate.81 shows a projection in the xyplane of the integration path.81 lb) =~ .812a.84) becomes (1.(x) (x) (1. Let I be given by /•y=x2 1 =1 xydy / y=x Calculate dL by dx (a) (b) using Equation (1.810) From Equation (1.88) After appropriate substitution.=f[x.89) r2 A =I JX y dy + (x) (x2) (2x) . u(x)] oil (1.8llc) On integrating directly 2 1 = Mr L = ^T JX (1. Figure 1.24 ADVANCED MATHEMATICS !?.89) is known as Leibnitz rule.b) / z z It follows from Equation (1.
There are instances when it is dx not possible to evaluate I explicitly and one has to use Leibnitz's rule.812C) Both methods result in the same expression for dL ? as they should.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 25 dL dx = 5xl_3x^ 2 2 (1.) and if more than one independent variable is involved.810) ORDINARY DIFFERENTIAL EQUATIONS (O. it is an ordinary differential equation (O. . and other fields of applied science are expressed as differential equations.D.DEFINITIONS A differential equation is an equation involving one dependent variable and its derivatives with respect to one or more independent variables.E.9 Integration described by Equation (1. economics.81 1.). Otherwise it is nonlinear.D. The highest order derivative occurring in the differential equation determines the order of the differential equation.E. A differential equation is linear only if the dependent variable and its derivatives occur to the first degree. If only one independent variable is involved.D. engineering.) . it is a partial differential equation (P. yf y = x2 x FIGURE 1.E. Many laws and relations in science. The degree of a differential equation is determined by the power to which the highest derivative is raised.
y) dy = 0 (1. Starting at Section 1. 1. For example. Indeed. we can easily verify that f (x) = e ax is a solution of the equation y' = ay. 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. linear differential equations.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 = . procedure to arrive at a solution.10 FIRSTORDER DIFFERENTIAL EQUATIONS The standard form of a firstorder differential equation is as follows M(x. A function f(x) is a solution of a given differential equation on some interval. f'(x) = a e a x = y' and the right side (ay) is of course af (x) = ae ax .111. (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. y) (1.113) . and chemical reactions. leading to the solution of several of the types of ordinary differential equations encountered in practice. There are several types of ordinary differential equations. exact differential equations.10. growth. y) dx + N(x.111) (1. homogeneous linear equations. there exists a known. decay. we summarize the approach. y' = 5y is an O.26 ADVANCED MATHEMATICS For example. Dividing both sides of the equation by x 2 y results in the appropriate form ^ . a straightforward integration will yield a result. standardized. (y " ) 3 + y = x is of order two. Solve y dx . Examples of firstorder differential equations are separable equations. of degree three and is nonlinear.D. In Section 1.112) (1. 1. For each of these types.19.101) or y'=^H N(x.E.x 2 dy = 0. we look at the modeling problem.^ = 0 * y (1.102) Equations of this type occur in problems dealing with orthogonal trajectories. etc.I N(y)dy Example 1.
Y where c is the constant of integration.101) are homogeneous polynomials of the same degree.^ =kf(cA) Solve Equation (1. then the substitution y = ux or x = vy will generate a separable firstorder differential equation. x + y .REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 27 Integration yields iny = i n c .112. x 2 .1110) (1.1111) (1.118) If M(x.117) y = ce"1/x Example 1.1 is not a homogeneous polynomial.12 HOMOGENEOUS FIRSTORDER DIFFERENTIAL EQUATIONS (1.119) (1. y) in Equation (1.\ that is (1.115) (1. the rate of disappearance of reactant A can be given by .3 xy + Y' is an example of a homogeneous polynomial of degree two.116) (1. This can be written as inyinc = .114) in£ or =\ (1.118) for the case where f (cA) = c A .1112) (1. y) and N(x.^ =kdt i n cA =kt + i n c in^=kt c A = ce~ kt 1. dcA . In a constant volume batch reactor. .
122) g.^ .121.1212) (1.8) 1+u (1.y ) d x = O dy (xy) dx ~ (x+y) The substitution y = ux defines y' as (1.2 u .dn l_2uu2 (1+U) d127.. Therefore incx = .121) (1.( x .u 2 ) i n ex"2 = i n ( l .^ where P = 1 .u 2 x y Replacing u by — .u2.u 2 ) ^r = l .2 u . 4 .23) Ml^l x(l+u) (L12 .5) (U2_6) vdx/ x (du) + u = x + ux £ ^ (1+u) \dx/ *[^) = ^"=l?Uu2 Idx/ ^ = x (1+u) . we finally obtain (1.1211) (1.2u . Solve (x + y ) d y .i n ( l .2 u .129) Integration yields inx + inc = I .1210) .(dj) x + 0 ) I 1 We now have x(du)+u=^iH = a.1213) (1.28 ADVANCED MATHEMATICS Example 1.
b) dX = a2X + b 2 Y + a 2 a + b 2 p + c 2 (1.1216) dx ~ a 2 x + b 2 y + c 2 where a}.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 . Think of the numerator and the denominator as representing two intersecting straight lines. we manage to do away with the constants Cj and c 2 . P).Y ajX+bjY+aja + bjP + Cj (1. We are left with the homogeneous equation .REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 22 ^=l2y_y! x x x or x22xyy2 = c (1.1214) (1.1219.20) (1.1217. Since the coordinates of this point of intersection are (a. p) of the system { ajX +bjy +Cj =0 a 2 x+b 2 y + c2 = 0 (1.1215) NOTE: An equation such as dy a1x + b 1 y + c 1 (1.1221a. If we translate the origin of the coordinate system to their point of intersection. b j .1222) where the underlined terms add up to zero by virtue of system (1.18) we then obtain a situation where the directions of the lines are preserved (aj andfyremain unchanged) and the coefficients Cj vanish. we perform the following change of variables JX=xa {Y=yp Substitutions in Equation (1. that is to the solution (a. c 1? a^ b 2 and c 2 are constants can be reduced to a homogeneous equation if through a change of variables.1217.18) with solutions x = a and y = p.
a separable firstorder equation. (1.l . Solve Example 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. .1223) determining the solution (a and (3) of system (1.1217.1223) [a firstorder homogeneous equation]. + Y d. This leads to Equation (1. the two lines are parallel and do not intersect.p respectively. iii) We now substitute X and Y in the solution of part (i) by (x .26) We obtain a = 2 and P = . That is to say. £ .^4 dx i) x+ y. Introducing a variable z = ajx + bjy yields a relation of the form — = f and f is a function of z — only.7 = constant (1. ii) Next we determine the values a and (3 by solving the system [xy = 3  x + y = 1 (1.bja 2 ) = 0.1225.a and y .1223) by x .1215).2 x y .1227) A problem arises if the determinant (ajb 2 .y 2 .2) and (y + 1). that is.122. This requires the determinant al bl *0 a2 b2 and iii) replacing X and Y in the solution of (1.1 QA.1224) dY X— Y We first solve rrr = ——— using the substitution Y = u X.18).6 x + 2y = c . The final solution is thus given by x 2 . the coefficients of x and y in the linear equations are multiples of one another. A.
For example.2 8 i n ( 2 x .2 1 y .7y yields the solution to the problem 6x21y28/en(2x7y + 9) = x + c' or or 1.1236) (1.123.1232) = 2 " 7 (^^r) or dJ = f^x 3 z . Solve dy _ 2 x .1234) This equation can be solved by separating the variables to yield where c' is a constant.7 y + 9) = c TOTAL OR EXACT FIRSTORDER DIFFERENTIAL EQUATIONS A given differential equation.3 y . Replacing z by 2x .2 1 y . one can determine by inspection that the equation xdy + ydx = 0 results from expanding (1.1235) (1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 37 Example 1.7 y + 1 dx~ " 6 x .1237) 7 x .2 8 i n ( z + 9 ) = .7 y + 9) = c' x .= J Let z = 2x . could have been obtained by differentiating an implicit function.l Note that 2 7 1 6 = I2}.1228) (1.x + c' (112"33) (1.1230) ^2x7y + n =2 7Ux2iyiJ a1231) (1.13 (1.1229a.131) .b) 27^ dx (1.4 i n ( 2 x .7y then dz dx = (1.
That is. We suspect that the equation is exact.133) or y = f (1.12) So M and N are partial derivatives of the same function F.138. multiplication by e^ results in the following total differential equation d(x2ey) = 0 thus and x 2 ey = c y = i n .2 i n ex x2 (1.° .135) can be solved by inspection.1311. In this particular case.101) could be represented by dF (x. assuming that F and its partial derivatives.9) The following test allows one to determine if the equation Mdx + Ndy = 0 (1.136) (1.22 ADVANCED MATHEMATICS d(xy) = O Integration yields xy = constant (1.1313) . y) = 0.137) (1. after multiplication by an appropriate "integrating" factor. Equation (1. Furthermore.101) is a total (or exact) differential equation.= .132) (1. we note that dxdy ~ dydx That is to say d F _ 3 F 2 2 (1. of at least order two.134) An equation such as x 2 dy + 2 x d x = 0 (1. This can be written as f Therefore dx+^dy=0 (U3_1O) M= and N = J £ (1. are continuous in the region of interest.
That is y .4xsiny .6xy + 4cosy)dx + (2y3x 2 4xsinyJ)dy = 0 ^=6x4siny 4g. a "partial" integration with respect to x (keep y constant) yields F.1321) or 3x24xsin y+f(y) = 2y3x24xsin y .6 x . The "constant" of integration will in general be an arbitrary function of y.3x 2 y + 4 x c o s y + f(y)] = 2y .1320) r)F f (y) is now determined by substitution of the previous equation for F in 3— = N . (1.3x2y + 4x cos y + f(y) (1.y (1.1314).[ x 3 .3 x 2 .1318) (1.1315) Example 1.1319) (1.4 s i n y Therefore J £ = 3x 2 6xy +4cosy and F = I (3x2 . Equating that function to a constant yields the solution F(x. One way of determining F is as follows: since •?.1316) (1.1317) (3x 2 . we know that we are dealing with the total derivative of a function of two variables.1322) . Solve (1.= . That is to say F =  M d x + f(y) 3F f (y) can then be determined from 3— = N .6xy + 4 cos y) dx = x3 . y) = constant. as illustrated next. which will disappear on differentiating with respect to x.131.y (1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 31 If our starting equation satisfies relation (1.= M.
1326) is exact and can be solved.3 x2y + 4 x cos y + y2 . the integrating factor I can be determined and Equation (1.101) is not exact. .1326) (IN) (1.1330) is also a function of x only. Therefore.(IM) = ^ That is T3M A/r ai T (1. and Equation (1.1327) 3 N XT ai dx dx (1.1329) becomes — 1 dl _ dM/dydN/dx I dx N ^ (1. Equation (1. (i) I is a function of x only.3d ADVANCED MATHEMATICS f'(y) = 2 y .i n y + c Hence.1330) The assumption that I is a function of x only implies that the right side of Equation (1. we can try to make it exact by multiplying with an integrating factor I. the solution is x3 .Jin y = constant • (1.1329) In genera] it is not easy to find I but there are special cases when there is a standard procedure to obtain I.1328) dy or I \ dx dy dy j ~ dy dx (1.J f(y) = y 2 .1323) (1.1325) If Equation (1.101) becomes IMdx + INdy = 0 which is exact if j . Then 5 = 0 .1324) (1.
1332) (1. (1.1340) . we can proceed as in Example 1.1339) (1.b) is a function of y only.132.y~2) dy .1338) is exact. ^=2x dy 3N »x In this case the equation is not exact. we note that .131. we obtain liL=_4 ^ dl = _ 4 1 d dy y I y (1.1333) fiN  ! ] / M = &L = . From Equation (1.1331). leading to a solution.1334) • 3 . Then as in (i).2xy~3 dx = 0 Equation (1.= 6x (1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS £5 (ii) I is a function of y only. Example 1.y2) dy .1337) I = y" 4 Multiplying Equation (1.1336a. Solve (3x2 .2xydx = 0 by finding an appropriate integrating factor. we have ( 3 x V 4 .1331) The right side is a function of y only and allows I to be determined.1338) (1. ^ = 2xy3 3x F = x 2 y~ 3 + f(y) (1.1332) by the integrating factor y~ .b) (1. we have _ 3N/9x3M/3y i" dy ~ M { dI (1.1335a.± \dx dy J 2xy y (1.
^ (1.1341a.14 LINEAR FIRSTORDER DIFFERENTIAL EQUATIONS The standard form is given by ^ + P(x)y = Q ( x ) (1.2 dy dy Therefore df dy = (1. Direct integration produces the solution as follows ^ .4 + f = 3x 2 y" 4 y. 7) .b) 1.144) Example 1. The integrating factor I (x) is given by I(x) = exp I P(x)dx (1.36 ADVANCED MATHEMATICS ^ = 3x 2 y.b) _y2 (1. Solve y'2xy = x I(x) = exp I 2xdx = e x p .143) r /x y = I"1 Q®I($)d$ + c (1.141.146.145) (1.1342) (1.[ y l ] = Q(x)I(x) and (1.1343) f = y!+c The solution is F = constant => .142) d(Iy) Multiplying both sides of Equation (1.141) Note that P and Q are not functions of y.x V ^ + y"1 = c (1.1344a.141) by I(x) yields a left side which is equal to — .
REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 37 Multiplying both sides of the equation by e~x yields e .1415) ICA klcA"k2cB .1414) (1. This method is known as the method of variation of parameters.( y e " x 2 ) d x = I x e ~ x dx x2 ye and finally 1 x y = cex 2 2 • Another procedure which will generate a solution for Equation (1.148) (1.x V .142. We then propose the solution of Equation (1.141) is as follows: first one solves the homogeneous equation y' + P(x)y = 0 (1.149) Integrating yields I .y e x + c (1.p (ye" x2 ) = xe~ x 2 (1.^ . where we replace the constant of integration by a function of x.1411) to yield the homogeneous solution yh.1410) 1 x2 = . The rate equations for components A. Example 1.x 2 y = x e~x2 or .2 x e . B and C involved in the following firstorder reactions ki k2 A > B > C are written as dc A dt ~ dc B ~df " (1.1413) x (1. as illustrated in Example 1.141) to be of the form of y h .143.
The integrating factor I(t) = e k 2 t .1421) Finally we obtain / e kit_ k 2 t\ c^k^hpir) Example 1. That is 0 = k1 CA k2kj l f° +c (1.1415. Equation (1.1417) Combining Equations (1.141) with P(t) = k 2 and Q(t) = kx c A e~ klt .1418) by I (t) leads to ^tae^kiC^efe*)* Integration yields k c e(kr~kl)t . y'^T X "T 1 (U4"22) Solve = (X+1)3 (1.1414) has been treated in Example 1. +c k2k1 (1. c A = c A and c B = c c = 0.1419) c B e k2t = (1.1416) We wish to solve for the time evolution of the concentration c B .112 and yields CA = %e"klt (1.3& ADVANCED MATHEMATICS dt 2 B (1. 17) leads to ^+k2c B = k lC A ( e. Multiplying Equation (1. given that at t = 0.k ' t (114_18) which is of the form of Equation (1.1420) The constant c is evaluated from the condition c B = 0 at t = 0.A° . Note that the independent variable is now t.143.1423) The homogeneous equation .
1425) (1. The general solution y = y h + y . We then obtain u'=x + l and on integrating.1424) v' 2 i . we can write u as u=y(x+l)2 + c The solution to the problem is thus given by y = (x+1) 3 (1.1428) (1. Substitution of this solution.1427) (1.1430) or y = c ( x + l) 2 + Y ( x + 1)4 (1.2 u / X + 1 ) (x + lj Note that since (x+1) 2 is the homogeneous solution. y'^y = 0 J\. 1.1429) = U ( x + l ) 2 + c](x+l)2 (1.15 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 .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 terms in u have to cancel.151) .= —=— v x+ 1 results in the solution or yh = c ( x + l ) 2 (1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 22. "l 1 (1. in the given problem results in the following expression u'(x+l)2 + 2 u ( x + l ) .
5) (r^r)i.1510.n ) y .+ P(x)v=Q(x) or ^ .154. 11) Combining Equations (1. we obtain ~idx~~ V = X °r dx" + 4 v = ~4x (1. Example 1. ( y 1 " n ) = ( l .+ ( 1 .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 (1.4 => 4^ = 4y" 5 pdx dx (1.y (1.n ) Q(x) which is linear in v.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. This yields the following equation (1.153) Then let v = y1"".157) ^ . 15) .1514.159) Let v = y.1512.n ) P(x)v = ( l .159 to 11).n i y dx dx dx Equation (1. This means that dv = A.153) reduces now to the following linear firstorder equation (1.X (1.151.152) ^ S " + P(X) y l " = QW (1. Solve = xy 5 (L156> (1.
b) .1518) 1. the general solution to L (y) = Q (x) is given by y = yh + y p (1. An alternative form of Equation (1.162) (1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 4± to yield ve 4x = _ x e 4x + 1 e4x + c (1.161) If yj and y 2 are two linearly independent solutions of the homogeneous equation L(y) = 0.16 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.166) (1.165a. then by the principle of superposition.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. C j y j + c 2 y 2 is also a solution where Cj and c 2 are constants. That is L (cl yx + c 2 y2) = Cj L(yj) + c 2 L(y 2 ) = 0 As mentioned earlier.1516) Substituting y~ for v produces + y4e4x = .x e 4 x + l e 4 x and finally i.164) (1.= _x+ \ y4 4 c (L15"17) + ce4x (1.163) (1.
168) (1. This equation has two solutions.1610).1612) where (Xj 2 = y ± i z = a ± ib This complex solution can be transformed into a real one. 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 .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 (1.169) (1. ii) two equal real roots (the case where D = 0). In this case. we also have two constants of integration and the solution y h is given by yh = Cleaix + c 2 e a2X (1.42 ADVANCED MATHEMATICS Solving the homogeneous equation y"+Ay'+By = 0 will generate y h . (Xj and 0C2.167) This is known as the characteristic or auxiliary equation. We propose y h to be of the form e a x . using the Euler formula to yield . Since we are dealing with a second order equation. y h is given by equation (1. In this case.1610) The second order characteristic equation could have 2 2 i) two real and distinct roots (the case where D = A .4B > 0). we have that a 2 + Aot + B = 0 (1. y h is given by y h = c1e0ClX + c 2 xe aiX (1.
3 b + 2c = x 2 + l Equating the coefficients of x 2 yields 2a = 1 and a= y (1.1618) Since the right side of the equation is of order two.b) (1.1614.3 a + 2 = 0 has the following roots: a 1 = 1 and a 2 = 2 so that yh = Clex (1.3 ( 2 a x + b) + 2(ax 2 + bx + c) = x 2 + 1 2a . we propose for y the quadratic polynomial ax 2 + bx + c. Example 1.3 y ' + 2y = x 2 + 1 The characteristic equation a2 . Solve (1. This can be done by proposing a solution.1623) (1.1617) + c2e2x (1.6 a ) + 2 a .1619) (1. based on the form of Q(x).1625a.1620) (1. (1.1622) (1.1613) (1. b and c.1624) (1.1621) .1616) 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.3b + 2ax2 + 2bx + 2c = x 2 + l 2ax2 + x ( 2 b . 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.6ax .161. and proceed via substitution in the given equation to determine the values of the constants a. 15) Next we have to determine the particular solution y . This solution technique is referred to as the method of undetermined coefficients. y p = ax 2 + bx + c y p = 2ax + b yp'= 2a Substitution yields 2 a . the right side of the equation.
1632) The form of the y to be chosen depends on Q (x) and on the homogeneous solution y^. Examples are (a) Q (x) is a polynomial..1630a. A * 0. 28) Equating the coefficients of x° yields 2a . + q n x n Try y p = a 0 + a lX + . Q(x) = q0 + q lX + q 2 x 2 + ..1626) (1. + a n x n ) (1. (b) Q (x) = sin qx or cos qx or a combination of both.1627a. (1.b.. which in turn depends on L(y).b) = l 3 + Q + P = 2 2X 4 (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...1629) or c= (1. B * 0. + a n x n if A * 0.. then the solution can be obtained by direct integration. try y p = x (a 0 + axx + .1633) If both B and A are equal to zero.1634) (1.1635) (1..3b + 2c = 1 l+2c y is thus given by v2 y (1..44 ADVANCED MATHEMATICS Equating the coefficients of x yields 2b6a = 0 2b = 6a = 3 and b =  (1. If B = 0.1636) ..
1641) (1.1640) (1. then y is y p = x(a o e ( l x ) If xe qx is also part of the homogeneous solution. If e1 x occurs in y h . 181). then y p = x 2 (a 0 e1x) (1. We need to find the number of undetermined coefficients that can be determined to fit the differential equation.1638) (1. The average velocity in the tubular reactor is (v z ). 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 .431 U . Example 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 ^ .1639) We can also determine the particular solution for a combination of the cases examined in (a to c) by intelligent guess work.1643) in standard form as follows .o \ z ' dz A ~ (] 16. Levenspiel (1972) describes a flow problem with diffusion.162. The particular solution can also be obtained by the method of variation of parameters (see Examples 1. The particular integral will be of the form y p = aoe«lx if e^* is not a term in y h .k c A .1637) If cos qx (or/and sin qx) is not a solution of the homogeneous equation. The reaction is characterized by the rate expression — ^ = .143.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 (1.1642) (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 ^ .i o is) We can write Equation (1. involving a firstorder chemical reaction.
A way of achieving this is by performing a so called Laplace transform.1644) JJ^ We now propose a solution of the form cA = e a z . 2 = .1647) (1. The Laplace transform is linear.1646) where a.1649) 1. The Laplace transform L[f(t)] of a function f(t) is defined as L[f(t)] = F(s) = I f(t)e.171) Jo where the integral is assumed to converge. As before. which are easier to solve.st dt (1. That is to say . substitution leads to the characteristic (or auxiliary) equation a2^ a^=0 (1.17 SOLUTIONS BY LAPLACE TRANSFORM Linear differential equations can sometimes be reduced to algebraic equations.46 ADVANCED MATHEMATICS ^_KideA_^_ dz ^ ^ az C 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 (1.1648) This leads to the following final result /cALcA eaiMl cAIcA eaiL ^ A° eaiz+ ^ Ao ea2z ea2LeaiL ea2LeaiL c A  cA A° (1.^  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.
.174) iii) Translation of a function L[e. c 2 are constants. that is to say.171 gives some useful transforms.17. (1. where Cj.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 (1.173) ii) Finalvalue theorem lim f(t) =lim s F(s) t»oo s>0 (1.177) _ 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..6) (1. the procedure to follow is i) ii) iii) take the Laplace transform of the equation. solve the resulting algebraic equation.] = CjLffjCOj + ^Lf^Ct)] + . This is usually done by consulting tables of Laplace transforms.. ...172) invert the transform. Table 1. For a given differential equation.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 47 Lfcjf^O + c / ^ t ) + . Next.. we state some theorems without proof. The Laplace transform of a variety of functions are tabulated in most mathematical tables.at f(t)] = F(s + a) (1. i) Initialvalue theorem lim f(t) = lim s F(s) t—»0 S—>«> (1. to obtain the solution f (t). obtain F(s).
0<t<k e~x^ s _a e" k s /s^.2. n>l 4J7^ ..48 ADVANCED MATHEMATICS TABLE 1.s ) n —..^ ^ V n ^ i ) .) erfc (x/2Vt~) . }X>0 ( ( t . +l _1_ sa — r s 2 + a2 — 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 + ^ .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 (0.171 Laplace Transforms Function I t" eat Transform 1 s JlL sn n =l.. t>k a"J n ( a t ) (Vs2 + a 2 .
This can be expressed as [ 0. t>a (1.1710a. by multiplying f(t) by H ( t .171 lb) = [e. we have I 0.a ) . t>0 H (t) is discontinuous at t = 0. In this case.17llc) (1.a). The Laplace transform of H (t) is given by r L[H(t)] = I e~stH(t)dt (1. Figure 1.171 la) L[H(ta)] = I e~ st H(ta)dt Jo = f e~stdt •/a (1.s t ]°° = Je~sa (1.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.17lld) An arbitrary function f(t) can be shifted over a distance a.179b) (1.b) I 1. .a) H (t .179c) J Q = \ We can generalize the Heaviside step function for the case where the discontinuity occurs at t = a.b) I 1.st ]°° L S (1.178a.179a) Jo = [e. H(t) = °°<t<0 (1. This will result in the quantity f (t . H(ta) = { t<a (1.171 illustrates this translation graphically.
1712b) r°° = I e. we obtain r L[f(ta)H(ta)] = I e"stf(ta) H(ta)dt JO (1.1713) (1.s a F(s) where F is the Laplace transform of f.1712e) .171 Translation of a function f (t) over a distance a Taking the Laplace transform.s ( t l + a ) f(t')dt' JO = esa I e.1712c) we have used the transformation t' = t .1712C) (1.1712a) = I e~ st f(ta)dt Ja (1.50 ADVANCED MATHEMATICS fit) > /fit) I _ / / /f(ta) H(ta) o a t FIGURE 1.st1 f(t')dt' (1. In Equation (1.a (1.1712d) Jo = e.
oo). The derivative dK is zero in any interval that dx .31) 8(x) = && (1.1716b) (1.1716c) we obtain.1715b) = 1 (1.1717) indicates that 8(x) is not an ordinary function.1715c) The limits in Equation (1. via Equation (1.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 57 The Dirac delta function is defined as 8(t) = 0. if x < 0 (1. everywhere except at t = 0 (1. By formally differentiating both sides of Equation (1. = H(x) ifx>0 (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).1716a) = 1.1714a) I J —oo 5(t) dt = 1 (1.1715a) Jo .1717) Equation (1.1714b. L[8(t)l = L J e~ st 6(t)dt (1. they could be (Ej. c) need not to be (oo.1716c) where H (x) is the Heaviside step function.1714c) The Laplace transform of 8(t) can be obtained using Equation (1. because H(x) is not continuous at x = 0 and therefore does not have a derivative at x = 0.1714b) I J — oo f(t)5(t)dt =f(0) (1.
1718e.1718a) I J _oo oo = 1 5(ta)dt = 1 (1.d) L[8(ta)l L[8(ta)l = I JO e~stst8(t .— as given in Table 1.a) to be an impulse at t = a and this will be considered in the next section. Solve the following second order differential equation . Compute the Laplace transform of sin (at). For formal computational purposes.174 and in Problems 35a and 36b. then 8 (t .52 ADVANCED MATHEMATICS does not include the origin. We will illustrate the usefulness of the Heaviside and Dirac functions in Example 1. we write f°° L[sin (at)] = I e~ st sinatdt JO (1.1718e.1719) Referring to Equation (1.a) = 0 . If the discontinuity is at t = a.171. Applying the definition.a cos at] 9 9 s 2 + a2 0 Evaluation of this term for t = °° and t = 0 yields the solution — .172.171. oo e st [s sin at .a) dt = e"sa e~ 8(ta)dt sa (1. Example 1. everywhere except at t = a = (1. s 2 + a2 Example 1.f) (1.a) dt = I = /aEj /aEj f (t) 8(t .1718c.f) (1.1718g) (1.39) we observe that the integral on the right side is given by .a) dt = f (a) = (1. we may regard the derivative of the Heaviside step function to be the Dirac delta function.1718b) f °° f a+e2 I /—oo /—oo f (t) 5(t .1718g) 8(ta) 8(ta) = $& (ta) ^ (ta) We may interpret 8 (t .
s + X2s2) (1.^ .1721a.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 21 ^ dt 2 +» * X dt 1 X2 h = f(t) (1.1722b) f(t) > Substituting into Equation (1.171.1722a) (1. That is . Equation (1.1726) This can be further expressed in terms of partial fractions. H(s) = X F(s) ^— . For the simplified case where c f(t) is assumed to be a constant. Using Table 1.1720) subject to the conditions h = 0 at t = 0 and ^ = 0 at t = 0.d) » H(s) F(s) (1. 1 +a>s + X s 2 (1.1724) can then be written as H(s) = ^ 2 = ^ Q 5(85^(882) s (1 + aX.b) f dt h > sH(s)h(0) = sH(s) (1. F(s) equals .1723) X2 Solving for H(s).h ' ( 0 ) = s 2 H(s) (1.s h ( 0 ) .1720) yields the following algebraic equation s 2 H(s) + a§. which are easier to invert. we determine the following Laplace transform ^4 dt 2 » s 2 H ( s ) .b) where s{ 2 = laX±X V(a 2 4J \/2X (1.1724) To invert H(s) requires knowledge of F(s) and therefore of f(t).H(s) + ^ = F(s) X (1.1725a.1721c. say c 0 . we obtain TT/ .
173.S 2 ) ( S .S l )H(s) = s C° l * 1 "" 2^ • (1.1726)]. a critically damped response is obtained.1418).1733) .S) and s and setting in turn s = s 2 .1730a. If a < 2. (s . B and C by multiplying Equation (1.1731) Sj (Sj .b) vs2 "" s\) B = (s. The Laplace transform of the equation is sC B (s) + k 2 C B (s) = k j c ^ j ^ ^ ) (1. if a > 2. Sj and s 2 are complex and one has an oscillatory response (see Section 1. Solve the kinetic expression given by Equation (1. A 'o + . Sj and s 2 are real and h changes with t in a nonoscillatory fashion.1727) becomes »2 2 + (1.b) 2 H(s) = A ^ L S Sj S 2 .1729a. subject to the condition c B = 0 at t = 0.16 iii). by Laplace transform.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. For a = 2.s 2 ) H(s)  i = S2 C° (1.1727) We determine the constants A. s = s 1 and s = 0.1728a. ^ ^ (1.b) A = s H(s)  s = 0 = ^ Equation (1.S j ) ( S .S j ) S2 ( S 2 . This leads to C = (s . Example 1. Invert using convolution.1727) respectively by (s .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. In particular.51 ADVANCED MATHEMATICS H(s) = A + _ B _ + _ O _ S S — Si S — So (1.s 2 ).
1737) 1M8) .1420) Example 1. Determine the shear stress in the material as a function of time. 1WA ° .tu(k9ki) 1 2 v du (1.174. We assume the relation between the shear stress (T ) and the shear rate ( y v x ) to be given by x +Xx =uV (1.^i) (l1735d) o yk2~KV (1.1734) Using convolution (Equation 1. X. as illustrated in Figure 1.Q are the viscosity and relaxation time respectively.1735b) Jo = k1cA()ekit ( e ^ ^ d u (1. N (s + k 1 )(s + k 2 ) (1.^ d u (l. A slab of a viscoelastic material was at rest and at time t = 0.17350 Jo k c e~ k l l e "" (k2 " kl) l = rl \ \ (k2 .1736) x yx +A 0 T y x . The shear is given by Yyx = Y 0 H(t) Using Equation (1.n yyx (I.172. and the dot over the quantities denote differentiation with respect to time. we have Yyx = Tb»(t) (1 (1.177).REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 55 k c C B (s) = B W . we write cB(t) = klCAo f ' e ^ ^ e .l735a) Jo = klCA() I e k. it is suddenly sheared by an amount y 0 .1717).
1739) and assuming that t T yx (s) + V T y x ( s ) = ^Yo T (s).171 to be t y x (t) = ^ ^ Xo e~t/X° (1.1739) = 0 at t = 0.172 Shear deformation of a slab Substituting Equation (1.1741) The inverse of T (s) can be seen from Table 1. is given from Equation (1.1738) into Equation (1. we obtain Tyx + A < o V = JLt y0 6(0 (1.1742) Thus the shear stress decays to zero exponentially. . we get (1.56 ADVANCED MATHEMATICS I  7 1 7 yf 1_ X 0V / V / FIGURE 1.1740) Taking the Laplace transform of Equation (1.1736).1740) by T (s) = =^2 X0(s + l/X0) (1. the Laplace transform of x .
181). G is continuous at t = x.164). as a differential equation in t. The boundary conditions will be applied at t = x0 and t = Xj . subject to the conditions y = 0 at x = xQ and x = Xj can be written as y(x) = P G(x.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 57 1. That is G = 0 at t = xQ and t = x 1 . The Green's function is a function of two variables. The solution of a nonhomogeneous boundary value problem can be obtained if the Green's function G is known for the homogeneous equation. In Equation (1. we regard the differential equation which G has to satisfy. we integrate with respect to t so as to obtain a solution y at point x. x and t. t) is a solution of the homogeneous Equation (1. then the solution to the boundary value problem can be written as an integral.183) .t)f(t)dt (1. except at t = 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. This requires that during the process of constructing G.182) G satisfies the boundary conditions. Once a Green's function has been obtained for a given operator and a set of boundary conditions. Our purpose in introducing the Green's function now is to show the generality of the method. That is to say d_G_ + A dG_ + BG dt 2 ii) iii) dt = o (1. The solution of Equation (1.181) where G (x. The Green's function for any x has to satisfy conditions such as i) G(x. t) is the Green's function for this problem. That is G = G t= x_ t= x + (1.161) with Q(x) replaced by f(x). Thus the Green's function is equivalent to the integrating factor in the firstorder equation.
We then calculate yp = a(x) sin x + b(x) cos x + a'(x) cos x + b'(x) sinx (1.187) (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. G(x. v) G is symmetric. The solution to the homogeneous equation is y h = Cj cos x + c 2 sin x (1.186) for any f (x). Having obtained the solution to Equation (1. x) Example 1. in general.181. dG dt t=x + dG dt =1 t=x_ (1. this difference equals the reciprocal of the coefficient of the highest derivative in Equation (1.58 ADVANCED MATHEMATICS iv) The first derivative is discontinuous at t = x. The solution to the differential equation is given.188) .186).166). t) = G(t. Having dx 2 obtained the Green's function allows us to write down the solution to Equation (1. 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. it will be possible to deduce the Green's function for the operator d2 L = \.185) (1.184) More generally.189) (1.186) The particular solution y p can be obtained via the method of variation of parameters.182). by Equation (1.1. We can then show that the Green's function satisfies conditions (i to v).
a ' ( x ) s i n x .1814) r a(x) = f(t)sintdt (1.1810) to replace a'(x) or b'(x).1812) (1. we obtain . The obvious condition is a'(x) cos x + b'(x) sin x = 0 Next.b ( x ) sin x Substitution into Equation (1.1813) (1.1812) becomes 2 . a(x) and b(x). One should realize that on substituting y p and y p in Equation (1. Such a condition should be chosen to simplify the problem.1811) (1. Replacing b1 (x).a ( x ) c o s x .1815) As in Equation (1. we have = a'Wcosx sin x (1.186) one generates only one equation to determine the functions a(x) and b(x).186). However.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 59 Note that y p has to satisfy Equation (1.31). Equation (1.a ' (x) sin x + b1 (x) cos x = f (x) (1. from Equation (1.1816) On integrating. Also. we compute y p y p = a'(x)sinx + b ' ( x ) c o s x .a ' ( x ) ^ _ ^ = f(x) sin x which reduces to a'(x) = f(x)sinx Integration yields (1. It becomes therefore necessary to impose one additional condition. t is a dummy variable.186) yields . y p contains two unknown (arbitrary) functions.1810) We can now use Equation (1.1810).
Thus the general solution of Equation (1.1821a) 0 = .1818) )cx h2 The boundary condition y = 0 at x = 0 implies r 0 = I f(t)sintdt (1. 17) will generate a solution y p which in fact includes y h .c o s (1)1 f(t)sintdt + sin(l) f(t) cost dt +sin (1)1 f(t)costdt JO JO Jc2 .143.1820) I1 \l° I1 (1. In other words. That is. a' = b'=O. Imposing the boundary condition at x = 1 yields 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° (1.1817) The constants Cj andc 2 can be determined from the boundary conditions.1821b) = .c o s ( l ) l f(t)sintdt + sin(l) I f(t)costdt .1815. A similar situation was encountered in Example 1. We note that when f(x) = 0.60 ADVANCED MATHEMATICS [x b(x) = f(t)costdt (1. introducing Cj and c 2 as limits of integration in Equations (1. a and b are constants (say Cj and c 2 respectively) and we obtain y h .1819) which leads to Cj = 0.186) may be written as y = cosxj r f(t)sintdt + sinx I r f(t)costdt (1.o Jc2 This may be written as (1.
.MUL r S i n ( 1 ) Jo • tl f (t) sin( 1 t) dt . To achieve this. Thus we write Equation (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 (1.181).1819. in order to write the solution as in Equation (1.^ ^ S i n ( 1 ) Jx f(t) sin(1 t) dt (1.0 61 f(t) sin (1t) dt + sin (1) I f(t)costdt (1. we obtain r y (x) = cos x r f (t) sint dt + sin x il f (t) cost dt .1 .REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS .1823) Combining Equations (1.1821c) Jo Thus Jc 2 f(t)costdt = rJ— f(t)sin(lt)dt (1.1824a) r f(t)sin(xt)dtS u ^ c f(t)sin(lt)dt (1.^Z Jo Jo s i n ( 1 ) Jo f (t) sin (1 t) dt (1.1824d) = 1 f (t) [ sin(1 ^ s" 1 ^. 22). because G has several properties to satisfy at the points t = x_ and t = x + .1818) as y (x) = .1824b) f(t)sin(lt)dt (1. we wish to determine G in two regions: t < x and t > x .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 .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 .1822) Jc2 Sm(1) Jo Recall that we are in the process of determining G.
The Green's function can of course also be constructed from the condition given by Equations (1. The boundary conditions will be satisfied automatically.181 are all chosen to be x = 0 and x = 1.181) where G (x. t).181) where G(x.1825a.G1 = sinx cos ( x . In particular. This exercise is addressed in Example 1. v) given earlier. we deduce that r1 G G. the solution to a given problem is to be written as in Equation (1.l ) t=x _" = n is o&»\ (L18'26a) sinTI) sinx cos(x1) sm(l) t=x+ and G' t= x + .186) can now be determined for any f (x). t) dt (1.1825a. As mentioned earlier. the solution of Equation (1. via Equations (1. b). .181. b). The Green's functions for several frequently used operators and boundary conditions are given in the following table.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.1 ) t=x_ sin(l) = { ^ The solution to the problem therefore is given by Equation (1.1825a.l ) sin(l) Y < t < 1 Q ^ x (1. b) we can deduce that G satisfies conditions (iv.c o s x sin ( x .1824f) where the appropriate function G(x.1 ) . That is to say. is given by Equation (1. b) From Equations (1. _ cosx s i n ( x . t) is the Green's function and is given by /sirU^inCx^) sin(l) G(x.182 and in Problem 37b. The boundaries in Table 1. t) should be chosen from the right side column in Table 1.182 to 5). t) = sinx s i n ( t . which is independent of f (t).1825a.
y " . If y is the displacement of the string from its equilibrium position. as a result of a force distribution f(x). x < t t.T ^ .REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 61 TABLE 1. t) t(xl).1) A sin A sinh At sinh A (x .A 2 y y (0) = y(l) = 0 Example 1.1) A sin A sin Ax sin A (t . Consider the transverse displacement of a string of unit length fixed at its two ends.1829a. y" + A2y y (0) = y(l) = 0 4.b) We now construct the Green's function for the operator . x>t x.1) A sinh A sinh Ax sinh A (t .181 Green's Functions Operator L(y) l. y" Boundary Conditions y(O) = y(l) = O y(O) = y'(l) = O G (x. x = 0 and x = 1.1) A sinh A ^ ^ 3.5 dx 2 (1. x>t x ( t . x < t sin At sin A (x .y" 2.182.b) . then y satisfies .2 L(y) = y " = . The boundary conditions are y (0) (1.1828) = y(l) = 0 (1.1830a.1).=f(x) dx 2 where T is the tension in the string.
the force acts at one point only.1837a) (1.t) ^ dt (1.d) (1. c2 = ( x .1833 to 35) we can solve for c1 toc 4 .c. = x(tl). G(x. where it is not defined.181) as /•I y(x)= .1834) (1.l ) . The displacement y at any point x is then given by Equation (1.64 ADVANCED MATHEMATICS Since G has to satisfy ft d2G ^=° (1. we can consider f (x) to be an impulsive force. That is to say. c4 = x (1.x .1838) t<x t>x (1. say at x = £.1831) G (x.l ) . which is given in Table 1. = c3(x) + c4(x) t .1832a) (1. We write f(x) as .t) = t ( x .1835) (1.1832b) 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. t) is given by G(x. t) = C!(x) + c 2 (x)t .b. t) is given by G(x. Mathematically f(x) can be represented by the Dirac delta function 8.1836a. Thus f(x) will be zero every where except at the point x = £. c3 = .181.1833a) (1.1837b) Jo As an example.1833b) Thus G (x. The result are Cj=O. t<x t>x (1.
b) 5%i2 Equation (1.1841b) provides a physical interpretation of the Green's function. f(x) = 5 ( x ..U e x p . An example of such a sequence is lim . a Reynolds number. is to consider the Dirac delta function as a functional or distribution. 39). . 1. It also allows one to easily compare situations which are dimensionally quite different. given by Equations (1. we obtain y(x) = .. Formulate a model in mathematical terms.17. one can compare flow situations in pipes of small to very large diameters in terms of a dimensionless variable £.5 ) (1.f ' G ( * . Indicate information such as pressures. G represents the displacement at x due to a point force applied at "t. Draw a sketch of the problem. Combining Equations (1. 1." T 5 ( .1840) e. 2.84^) (U84. introduced by Schwartz (1957).19 MODELLING OF PHYSICAL SYSTEMS In order to generate the equation(s) representing the physical situation of interest. That is to say. etc. 4.1714c). = £ which will always vary from 0 to 1. Both approaches produce equivalent results. Make sure you understand the physical process(es) involved. temperatures. Jo (. determine the equation(s) as well as the boundary and/or initial condition(s). An alternative approach. The main properties of the Dirac delta function are given in Section 1. This may reduce the number of variables and it allows for the identification of controlling variables such as.^ d .1839) The delta function is not an ordinary function but it can be regarded as a limit of a sequence of functions. R Determine the limiting forms (asymptotic solutions). 3.1838. For example.( M 2 e>0 VTCE L v = 5(x) (1. it is useful to proceed according to the following steps.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 65. 5. Nondimensionalize the equation(s). for example.
191 Flow in a pipe with associated velocity profile v z (r) and shear stress profile x r z .191 illustrates the problem which is to be modelled subject to the following assumptions steadystate flow. < i FIGURE 1. 7. Solve the equation(s). laminar flow: Re = — — — < 2100 and vz is a function of r only. We can imagine thin concentric cylindrical sheets of liquid sliding past each other. there are no end effects.66 ADVANCED MATHEMATICS 6. the piece of pipe of length L which we are considering is located far from either end of the pipe which is very long. 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. Figure 1. that is. ^ — the fluid is incompressible: p = constant.
The forces acting on the system (cylindrical shell of thickness Ar and length L) are: (i) the pressure force in the zdirection. one has to realize that z momentum is transported according to two mechanisms: (i) by convection due to bulk flow. The differential equation representing this physical situation is obtained by applying a momentum (or force) balance on a shell of thickness Ar. Analyzing the problem. This contribution is obtained by multiplying the volumetric flowrate associated with the appropriate area (27trAr). This contribution is obtained by multiplying the shear stress x rz (force per area) by the appropriate area (2 TtrL).191) is zero because of the steadystate assumption (no accumulation!). 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. n. Since the flow is laminar. c) zmomentum in by viscous transport at r. with pvz. The individual contributions making up the force balance are a) zmomentum in by convection at z = 0. This makes it mathematically easier (less calculations) to apply the boundary conditions. r. Ar and p are not changing.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 67 Cylindrical coordinates are used since the geometry of the problem is cylindrical. and (ii) by viscous (molecular) transport. That is zmomentum in T rz (27crL) r by viscous transport at r L o . 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. and (ii) the zcomponent of the gravity force. 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.
the shear stress t r z is given by dv^ ^ dr XfZ (1.191. e). The gravity force can therefore be written as: 2nr pg (h 0 . We established earlier that Trz is a function of r. 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).193) by 2ftLAr to obtain (1. .h L ) Ar. d).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. the volume (2rcrArL) multiplied by pg and by cos a (zcontribution) gravity force (27trArL) pg cos a Note that L cos a = h 0 . 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. e) The contribution from the pressure force is obtained by multiplying the pressure (force per area) at z = 0 (Po in Figure 1.Jg^)r Ar L + rpgKz!!L) = 0 ' ^ L ' (1. That is.193) [^^UA. The Newtonian viscosity I is constant and therefore the left side of Equation (1. the difference in height of the positions at z = 0 and z = L in Figure 1. otherwise —^ would be zero.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.4) Note that we did not divide by r. Substitution of the terms in c).h L .192) dv vz is a function of r.19.
195) .1912) V^f^L)^ .197) Equation (1. rx r z initial lim Ax>0 (at r) .191. so that we are dealing with A(rx r z ).^ ( ^ J + P )+Pg(h0hL)] = 0 (1.1910) d.f ( x o ) Ay df — = lim ^.rx r z final (at r + Ar).r T r Z l r + A r  +j_ [(p Ar>0 \ Ar / L or via Equation (1. we have a term involving the quantity r Trz  r .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 ) . Taking the limit for Ar —> 0 of Equation (1.195a. to yield the linear shear stress profile shown in Figure 1. /F .199) I d(ny = I r(^i)dr . The differential Equation (1.196) r(IP^PL) = 0 (1.^£(^) + Cl (1.197) is the desired differential equation with the potential F defined by IP = P + pgh (1. that is.1911) (1.197) is solved as follows.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 .194).tP \ £<riy = r(^L) (1.r t r z  r + A r . Note that the potential drop per length I—2——I becomes the pressure drop per length when h 0 = h L (horizontal flow).= (1.b) Ax Ax>0 AX dx In Equation (1.194) yields Urn ( r T r z l r .
We now have a differential equation relating vz to r given by (1.1913) by fi — £ .1920) .1914) This equation is solved as follows \dv*" " F i ^ ) Irdr ^(^MT^ (15 191) (U9I6) The physics of the situation allows us to assume that the velocity is zero at the wall. we look at the physical situation for r = 0.19) H^MR2 Substitution into Equation (1.17 > (ii9i8) (L19.1916) yields Vi = ?ci) R 'Jti r > That is. (1960). The minus sign is as a result of following the convention by Bird et al.1916) allows us to evaluate the integration constant C 2 .70 _ _ ADVANCED MATHEMATICS To evaluate Cx. Therefore Q = 0 and Trz is a linear function of r. vz = 0 at r = R.191). v z is related to the square of r (a parabola as in Figure 1. Applying this to Equation (1.1913) dv The velocity profile is obtained by replacing the left side of Equation (1. We cannot accept an infinitely large force (per area). The relation for v z is usually written as (1. \z = § Px^) (1. ° = (^)f+c^ So " 19 . That is.
as the upper limit goes to R.^  n a1925) . That is. We will assume the fluid to obey the powerlaw. (1. as it should. The lower limit remains unchanged. at r = 0. The average velocity (vz) is obtained by dividing the flow rate by the crosssectional area. Integrating this product yields a flow rate. we consider the flow of a nonNewtonian fluid in a pipe.1921) The denominator yields n R2.1923) Note also that the limits of integration have to be adjusted according to the newly introduced variable £. 21) yields = \ 4. Indeed. 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.b) Example 1. As a second example. = £•. Combining Equations (1. as it should.1924a.1922) reduces to (1.192. Equation (1. The numerator is the product of a point velocity (vz) and an infinitesimal area (r dr d0). as follows e2n /R I vrdrdG <vz) = *>» I r dr d6 Jo Jo (1. That is to say \z = .0. goes to one. the new variable £. Note that this integral is now independent of R and is valid for pipes of any radius.1920. ) 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.
1929) Note that for n = 1.1920).1921). (1. it is valid for any fluid.121.1926) /dv \ and the shear rate — is given by — (1. We now proceed by computing the velocity distribution. That is to say. Equation (1.1929) into Equation (1. one should realize that Equation (1. and the velocity profile is (1. vz = i^f^f11 1/n  r1/n dr r /n+ (1.1927) Integration yields the velocity profile. In fact.1913) now becomes (1.72 ADVANCED MATHEMATICS At this point. Equation (1.1930a) 1 +3 n 2\iL J (1.1930b) . since no assumption as to the type of fluid has to be made to derive the shear stress distribution. using the same (no slip) boundary condition as in Example 1.191. the powerlaw fluid reduces to the Newtonian fluid. following the procedure given in Example 1.1913) is still valid. The average velocity is obtained by substituting Equation (1.1929) reduces to Equation (1.1928a) p _p _ l (fen\ TTTl \ +C2 <1"8 ) 19 2b n / The constant C2 is obtained.
These profiles are illustrated in Figure 1. the slope must lie between 0 and 3. we continued to use parameter \JL as in Example 1.) 1/n+1 ] n + 1 [ lR) J (vj (L1931) v ' Different values of n yield different profiles. Vz (l. we obtain the Newtonian result (parabolic profile). In this example. However. The ones of interest are tabulated in Table 1. For intermediate values of n. In practice.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIA!. \ v z) R Dividing Equation (1. the slope is zero and for n = °°. TABLE 1. EQUATIONS 73 We are now in a position to compare the velocity profiles in dimensionless form and to evaluate y asymptotic behavior. To achieve this. . we plot j^r versus —. note that the dimensions of ji depend on n and jx has the same dimension in both examples when n = 1. we obtain a maximum slope of 3.191. 0<JL< 1 0 1 l(if = \ R R 2 ['(it) 2 We note that for n = 1.1929) by Equation (1. as n tends to zero and to infinity.1930c) yields i = 2a±jJi_(i. For n = 0.191 and are shown in Figure 1.191 Velocity profiles n r .192. The asymptotic solutions (linear profiles) are obtained for n = 0 and n = <x>. n is usually between 0 and 1 and the observed velocity profiles are indeed more blunt than the Newtonian one.192.
Answer: r Use the chain rule to compute ^ 3b.74 ADVANCED MATHEMATICS — <v"z> 2 I / Vn=i / /An = oo l/^<& n=o \ 0 _r R I FIGURE 1.x)°\ where a is a constant. The circumference C and the area A of a circle of radius r are given respectively by C = 2rcr.= ocxy .x 2 (iii) (1 + 2 / e n x ) / x i n x 2a. If y = e " a x / ( l . 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 . A = 7tr2 . show that (1 x)p.192 PROBLEMS 1 a.
7 a. take in) 5a. (iii) Ml + x z j (iv) .dx 1 1/ i\ 3/2 ) ( x .REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 75 Use Leibnitz rule to show that (l_x)y("+l)_(n where y ^ = — . — dx n 4a. + ax)yW_nay(nl) = 0 w ^ 1 sinhx v (iv) ' xx (Hint: in (iv). Find the area of the region enclosed by y = cos x.(x .I)"1 8a. (v) i n (x2 + 2x . x = 0. Find the remainder term if only the first three terms are retained. using Maclaurin's expansion. (ii) ^[cos ( i n x) + sin ( i n x)j . and x = K. 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 limit as x —> 0 of the following functions (i) S i ^ w x (ii) xcosjc sinx (iii) v . 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 .l ) 2 ( x + 3) r Answer: (i) e x (x .1). Answer (i) to (iv): 1 Use Taylor's theorem to obtain an expression for cos x about x = n/2. such that "y4 and higher powers of 7 can be neglected. 1 +c72 write down the coefficients of the expansion you have obtained in terms of a. .cos e x . If it is known that r\ is given by the rational function a + bY 2 Tl = — 6b.3) . express TJ as a polynomial in 7. Sketch the curve y = cos x in the range 0 < X < T C . b. the xaxis. If 7 is small. and c.
Show that if the independent variables x and t are changed to t. volume V. 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 = 21 y V"+If d x ) Ja Compute S for y = e~x. b. and T o (> 0) are constants. the wave equation . c.x 2 y 2 + y 3 (ii) f(x. . c) Answer: n = — Va 2 + b 2 + c 2 13 a. and temperature T for 1 mole of an ideal gas are related by the equation PV = RT where R is a constant. The temperature T in a body is given by T = T 0 ( l + a x + by)e c z where a. b. y) = x cos y + y sin x lib. In addition. Find fx. = X + Ct. Find the rate of change of T along x. find the direction in which the temperature changes most rapidly at the origin. The pressure P. y. 12b. Find the change in volume if both the temperature and pressure are increased by 1%. 0 < x < ° ° . Tl = XCt.76. (a.y) = i n ( X 2 + y 2) Answer: n (V2 + sintr 1 1) (iii) f (x. y) = x 3 . and zaxes at the origin. fy. 10a. fxy for the following functions (i) f(x. ADVANCED MATHEMATICS 9b.
15 a. can be written as .and ^ . y ^ is the shear rate. The volumetric flow rate Q of a nonNewtonian fluid in a circular tube of radius R is given by Q(R. the volume V and the temperature T are given by p + 0LJ(Vp) = RT where a. show that the viscosity r\ (y R ). „ 3T . which is a function of the shear stress. By differentiating with respect to i R .REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 7J_ a2u = c 3t 2 2a2u . dy Compute —^ if dx y 2 .sin xy + x 2 = 4 . 3x 2 is transformed to 35 an 14a. The pressure P. c is a constant. This is the Van der Waals' equation. 16b. (3.9V Compute ^p. show that the shear rate at the wall YR is given by y = JM3Q + TtR3 k dQl dtJ Also. and R are constants. which is defined as the ratio of the shear stress to the shear rate.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.
In the reversible chemical reaction A ^_ B + C . show that . x 2 dx + 3y3 dy = 0 xydx + V l . 17 a.k 2 . (ii) ( y ' ) 2 .78 ADVANCED MATHEMATICS (Q/TCR 3 ) L d(inT R ) The above result is known as the WeissenbergRabinowitsch equation. Solve the following separable firstorder differential equations 3 (i) (ii) (iii) 20b. y = ^1 cos y \ 2' Answer: — = . ^ i 21b. ^ = k1(A0X)k2X2 with A o representing the initial concentration of chemical A.cos x In a constantvolume batch reactor. 19a. Assuming the rate constants kj and k 2 to be related as follows: kj = —.^ y 4 + c Answer: y = c e ' ' " x Answer: sin y = . 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. Discuss the obtained result for the cases n >1 and n < 1.x y ' = 0 Verify that e~2x (cj cos x + c 2 sin x) is a solution of y " + 4 y' + 5y = 0. Determine the order and degree of 2 (i) ^ + 3 ^ = cosx dx 2 dx 18a.x 2 d y = 0 y' = sinx (given that at x = n. broken down at time t is represented by %. the amount of component A.
y dx = Vx 2 + y 2 dx (2 VST .i n (2y . 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 . (x + y) dx + x dy = 0 x dy . Solve the following reducible firstorder differential equations (i) (ii) (y .i n (4x + 8y +5) = c 25a. At t = 0. 1M> _ pvB(. 22a.8y . Solve the following homogeneous firstorder differential equations (i) (ii) (iii) 23b.3x + 2) dx = 0 (2y + x + 1) dx = (2x + 4y + 3) dy Answer: 2 (y + x) . The evaporation of component A is proportional to the volume V A (t) of A. The volumes of the components are V A (t) and V B (t).= a VA(t) dt The evaporation of component B is related to the evaporation of component A as follows £M> .a) VB (0) 24a.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 79 given the initial condition that % = 0 at t = 0. That is to say.). 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. the initial volumes are V A (0) and V B (0).3x) dy + (y .6x + 1) = c Answer: 4x .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. — ^ . Solve the following linear firstorder differential equations .
t' < t. using the appropriate integrating factor that the Maxwell model can be written as x = _ \]L e ( l . \i is the viscosity.2 cos t .— g 2x 29b. That is. Solve the following Bernouilli equations (i) (ii) .OAol y / Oo (t')dt where the dummy integration variable t' is interpreted as a time in the past.l + c e 3 x (i) (ii) 31a.— . Show. Complete Example 1.7y = ex Answer: y = c e 7x . and Xo is the relaxation time. Answer: y~ 3 = . Yyx is the shear rate.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 ) . A y(l) = O Answer: y = ].x y 4 2 (x2+ 7x . 28a.(x~2 + x 2 ) 4 y' .+ (6x + 21) y = 3 (x + 8) 2 y 5/3 Answer: y~ 2/3 = (x + 8) [1 + c ( x .142 by computing the time evolution of c^. Solve the following secondorder differential equations with constant coefficients .— e x 6 Viscoelastic behavior can be described via a Maxwell model given by where T yx is the shear stress. y " . y ' + ^ y = x.1)] 30a.80 ADVANCED MATHEMATICS 0) (ii) 27b.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 .8) p.^ + 2xy = . Discuss the particular cases where k2 » k^ and k 2 « k i .
L is the length of the manometer fluid. describing the motion of a manometer fluid. [i is the fluid viscosity. R is the manometer tube radius. the mass flow rate at the outlet.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 8± 32b.P34. Q is the volumetric flow rate and p is the density.P33. y2(0) = 1 . Determine w o (t). y 2 (0) = 2. The differential equation. using the Laplace transform method.Awj) . for a step change in AP. At steady state. is given by the following unsteadystate macroscopic mass balance ^ t o t = Aw where m tot = V p 0 (t) is the total mass. The conditions on h at t = 0 are: h = 0 and ^h = 0. 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. the mass balance reduces to Wi s w O s = 0 At time t = 0. subject to a sudden pressure difference AP = P a . Answer: w o (t) = (w i s + A W i )+ ^ e ~ 5 l (w Os . and where w = p Q. y\ (0)= 3. The differential equation governing the mixing process. and 2h is identified in Figure l. illustrated by Figure 1 .w i s . at t = 0.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. Determine the relation between k and t. Answer: yj = e l + e 2t y 2 = e2t 33a. a step change AWJ is imposed. dt 34a.
Consider separately the cases when t > a and t < a. it was deduced that the constitutive equation of a Maxwell fluid can be written as v . deduce the shear stress Tyx at time t.f f<=(tt')A»v<t'>d'1 A Maxwell fluid has been at rest and.82 ADVANCED MATHEMATICS P I FIGURE l.P33 Motion of a manometer fluid 35a. In Problem 27b. a shear of magnitude Yo w a s suddenly imposed. . Using the properties of the Heaviside and Dirac functions. at time t = a.
Mixing process at constant volumetric flow rate Solve the differential equation £ + y = «(. 37b.1 Consider the cases t > a and t < a separately.182 to 5). by applying the conditions given by Equations (1.184) to determine the remaining equation needed to solve for the constants that were generated in step (i). use the boundary and continuity conditions to determine the relations which have to be satisfied by the constants.186). needed to solve Equation (1. That is to say (i) (ii) determine the solution to the homogeneous equation for the cases t < x and t >x. (i) (ii) by the Laplace transform method. use the jump discontinuity condition of Equation (1. a > 0 .P34 36b. Construct the Green's function. y = e~l + e a . Answer: t > a. y = e. by finding the integrating factor.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS £2 wL I t ) 1 •Wo It) V FIGURE l. (iii) .a) with y(0)= 1.t t < a.
0<X< 1 y(0) = y ( l ) = 0 by the Green's function method.84 ADVANCED MATHEMATICS Then solve Equation (1.t) = 0. A uniform string of unit length and of mass m lies along the xaxis. t) = h (x) cos cot and determine the differential equation and the boundary conditions that h has to satisfy. Assume that y is of the form y (x. to generate the . Verify that the Green's function you have chosen satisfies conditions (i) to (v). (1960).186) for f (x) = sin x. f(x) = x. Answer: L c o s } sin x . Solve for h using the appropriate Green's function. If y is the vertical displacement of the string.t) = y (l.l / 2 j .1644). Answer : y = — e x . Solve the boundaryvalue problem y"_y = ex. appropriate expression for the flux N A z defined by Bird et al.x cos x 2 sin 1 2 38 a.181. using Equation (1.e s m n x J 2 2 sinh 1 39b.181). Perform a mass balance over a differential element of thickness Az. Choose the appropriate Green's function from Table 1. aQ is a constant and 8 is the Dirac delta function. Hence determine y for the following f (i) (ii) 40b. Differentiate N A z with respect to z to generate Equation (1. A periodic force f (x) cos cot is applied to the string. Its two ends x = 0 and x = l are fixed. co is a constant and t is the time.1. A firstorder reaction A —> products takes place in a tubular reactor of radius R and length L in the z direction. then y satisfies the following equations oa y cz — 3x 2 C2 dy 9t 2 = f (x) cos cot = T/m y(O. f(x) = a o 8 ( x . The tension is T.
The belt transports a laminar film of liquid. The belt has a width W and a velocity V. a continuous belt has been introduced in the fluid reservoir.P41 Aeration of a fluid . (i) Perform a momentum (or force) balance over a length L to obtain the following shear stress distribution txz = pgxcosp List all assumptions made. In order to accelerate the aeration of a Newtonian fluid. as shown in Figure l. Show that this can be done in the following way. up to a wall C as illustrated. (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.P41. As the air penetration in the liquid film depends on the film thickness.REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS §A 41b. we wish to establish a relation between the film thickness 8 and the velocity V.
(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. 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£.P42. In the design of spray dryers. note that the heat flux qr is linearly related to the temperature gradient ^J. as follows dr 4r dr This is Fourier's law. the heat transfer coefficient h at the solid gas interface is constant. The constant k is the thermal conductivity.86 ADVANCED MATHEMATICS 42b. the sensible heat required to cool the droplet from To to TM is negligible compared to the latent heat of fusion. 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 . . one is interested in knowing the time required to solidify the liquid droplets.
T° \ \ \ " T °° / /LIQUID FIGURE l..P42 Solidification of a droplet .REVIEW OF CALCULUS AND ORDINARY DIFFERENTIAL EQUATIONS 87 T .
.
d) (2. Equation (2. the differential equation reduces to an algebraic equation [see Equations (1.D.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. 9)].E.12a) (2. 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. then the solution is not of an exponential form. If b is zero. we reviewed several of the standard techniques used to solve ordinary differential equations (O.12a) is homogeneous.1 DEFINITIONS In Chapter 1.11) We recall that if b is zero.'s). b = a/a2 (2. 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 .168. We recall that a .5. we develop methods of solving O. p 2 = a Q /a 2 . The solution of Equation (2.x 0 ).'s can be solved by the same method. we have only the homogeneous solution.E. Equation (2. A second order O. The particular integral (or particular solution) is usually obtained by the method of variation of parameters which is described in Example 1.D.D.CHAPTER 2 SERIES SOLUTIONS AND SPECIAL FUNCTIONS 2. we consider second order O.c. As in Chapter 1. The form of this series depends on the nature of the point xQ.D.D. In this chapter. If the coefficients of the differential equation are not constants. Higher order O.11) can be written in standard form as y " + p 1 y 1 + p2y =b where pj = aj/a 2 .'s with variable coefficients. we have seen that second order linear differential equations with constant coefficients admit a solution in the form of an exponential function.181 and Section 2.'s. In particular.E.E. Since the derivatives of exponential functions are also exponential functions.E.12b.
Analyze the following equations for ordinary and singular points.18a) x 2 (x2) 2 x 2 (x2r .2 ) 2 y " + 2 ( x .15c) We proceed by evaluating the functions p^.P2) is not analytic at x 0 .12b.16a.15b) (2. ii) The standard form of this equation is given by ""•TV+M^T)^0 p=^T^23^). f(x) can be represented by the series oo f(x) = X a n (xx 0 ) n n=0 (2. c).2 ) y ' + (x + l ) y = 0 (2. «"••» (217a) <2'™. From Equations (2.14a) (2. xQ is a singular point.2Q ADVANCED MATHEMATICS function f(x) is analytic at x 0 if its Taylor series about x 0 exists. it can be seen that if a2(x 0 ) is zero and either a^ (x 0 ) or a o ( x o ) is nonzero.c) These functions are analytic everywhere and all points are ordinary points. then x 0 is a singular point.4 . P2. i) This equation is in standard form and P! = x . P2 = x 2 . That is to say.11.b. The singular points are therefore at x = 0 and at x = 1 and all other points are ordinary points.cd) Pl is not analytic at x = 1 and P2 is not analytic at x = 0 and at x = 1. and b = 0 (2. Example 2.14b) an = f ( n ) ( x 0 ) / n ! If one or the other (or both) coefficients (pi. iii) The standard form of this equation is given by (2. and b and by determining the presence or absence of singular points.15a) (2. i) ii) iii) y"+xy' + (x24)y =0 (xl)y" + x y ' + i y =0 x 2 ( x .
11 la. x2 ( x . assuming x 0 = 0. the singular points are at x = 0 and at x = 2. x2(x2)2 and b = 0 (2.SERIES SOLUTIONS AND SPECIAL FUNCTIONS £/ Pj = ^ . Determine which of the singular points in Example 2. Example 2.12.x o ) 2 p 2 are both analytic at x 0 . If x 0 is a singular point of the differential equation and ( x .x o ) P l = x P l = ?— /v J.b) (2. If not. x2(x2) p2 = — ^ ± J — .d) is a (xxo)2p2 = x2p2= ^ y Both (x .110c. x 0 is a regular singular point of the differential equation.18b.19c. we use power series extensively and we next summarize their properties.x o ) p j as well as ( x .11 are regular singular points. Note that had we chosen x 0 = 2 in part (iii).d) Here. i) ii) All points are ordinary points. we would be dealing with (x2)Pl=^ x and (x2)2p2= £ i i x (2. x 0 is an irregular singular point.b) L) X (A x + 1 (2.19a.c. .110a. In this chapter.b) which are analytic at x 0 = 2 and therefore x 0 = 2 would be a regular singular point.x 0 ) 2 p 2 are analytic at x = 0 and therefore x 0 = 0 regular singular point. (2.d) (x2)1 xpi is not analytic at x = 0 and therefore xo = O is an irregular singular point.x0) pi and (x . hi) (xxo)Pl = xPl= (xxo)2p2 = x2p2= 2 (2.
R is the radius of convergence... No conclusion can be drawn if L = 1. (2. we can write the series as ^ m n=0 e x i s t s and an x n .21 a) (b) 2L x " n=0 00 = l + x + x 2 + x 3 + . By translating the oo origin to x 0 . and (c) the series diverges for all nonzero values of x.. 00 (a) ]£ n=o 00 x "/( n ! ) = 1+x + x 2 /2!+x 3 /3! + .2 POWER SERIES The series on the right side of Equation (2.. we have (x * 0) .2lb) (c) £ n=0 n ! x n = l + x + 2 ! x 2 + 3 ! x 3 + .2lc) The first series represents exp (x) and is convergent for all values of x.. (2. the series is divergent. (2. If L > 1.14a) is a power series in (x .x) and is convergent if x < 1 and diverges if  x > 1. We illustrate this situation by the following examples. If a series converges for xj < R and diverges for I x I > R. In many cases. 00 Applying this test to the series ^ n=0 an x n . Every power series converges at x = 0 and its sum is a0. A series may or may not converge at its radius of convergence..22) is less than one (L < 1).92 ADVANCED MATHEMATICS 2.Not all power series converge for all nonzero values of x. (b) the series converges for some values of x and diverges for other values of x.x 0 ). The third series diverges for all values of x (* 0). We recall that the 00 series X n=0 un is convergent if the ratio Hfl±i _ u n lim n —> 00 L (2. There are three possibilities: (a) the series converges 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. 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. The second series is a geometric series and its sum is 1 /(I . Thus all power series have a radius of convergence. the radius of convergence R can be determined by the ratio test.
. then oo oo 2_j v n converges or diverges according as ^ n=0 OO u n converges or diverges. If x > lim . the series ^ n=0 v n is divergent if v n > K u n for all n and any positive constant K. if ^ n=0 u n is divergent. If the ratio v n /u n tends to a finite nonzero limit as n—> °°. R = km — = lim — =1 n>o° ( 1 + n ) 2 n ~ >o ° (1 + 1/n)2 .25) The comparison test is another simple method of determining whether a series is convergent or oo co oo divergent.. _ _ .^ n — an+1 (2.24a.x < 1 an => x < lim . .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 (2.26) (2.b. Example 2. OO n=0 The series ^ n=l 1 /n is divergent and the series j ^ 1 /n n=l is convergent. (1+2/n)2 .b) the series is convergent.^ " .27a. (2.an+1 the series is divergent.c) .> .. (2 + n) 2 . Discuss the convergence of the following power series. the series ^ n=0 OO v n is convergent if v n < K u n .21. This test can also be stated as follows. If ^ n=0 u n is convergent. 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) .^ L _ n>~ a n + 1 (2.b) ^ O .23a.
For x = 1. b) R = Urn n —> oo f±Jl 1 + n = 1 (2. the series behaves as ^ n=0 1/n and is divergent. We make use of the following properties of power series. oo the series 1/n . the series oo oo oo becomes ^T 1/(1 + n) and since ^ n=0 n=l 1/n is divergent. 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. At both of these oo values of x.1 .2 < 1 and is divergent if x  > 1 or x . if u n > u n + 1 > 0 n=0 and lim u n—>~ is zero. In the present example. the series is convergent. the series is convergent if x = . which states that for an alternating series j ^ ( .b) n=0 The series becomes £ R = lim 3 + 2n _ j n —>~ l + 2n The series is convergent if x  < 1 or x . We apply n=0 oo Leibnitz's test. To x = 1 corresponds x = 1 or x = 3.l . c) In this case. that is to say .29) (x*) 2n /(l +2n). it follows that ^ n=0 OO 1/(1 + n) is also divergent.210a. the series ^ ( . oo If the series ^ n=0 a n x n converges for  x  < R (R •£• 0) and its sum is denoted by f (x).28) As in a).l . the series is convergent if  x < 1 and divergent if  x > 1.2 oo (2.2 > 1. In this example x* is raised to an even power and there is no need to separately consider the cases x = 1 and x = . For oo x = 1.l ) n u n .l ) n / ( l + n ) is an alternating series. (2.94 ADVANCED MATHEMATICS The series is convergent if x < 1 and is divergent if x > 1. we translate the origin and write x* = x . If x = .
we can set x 0 to be zero. We seek a solution in the neighborhood of XQ. The functions pj and P2 are analytic at x 0 (= 0) and their Taylor series are oo (2.3la) (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. as given by Equation (2. 2.3lb) Pi 00 = X Pln X " n=0 oo <23"2a) P200 = Z n=0 P2nx" (2.SERIFS SOLUTIONS AND SPECIAL FUNCTIONS oo 91 f(x) = X n=0 anx" (2. Further discussions on power series are given in Chapter 3. 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 .32b) (2.3 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. If x 0 is nonzero. their coefficients are equal.32c. If two power series converge to the same sum throughout their interval of convergence.12a). If we are required to find the solution at points near infinity. 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.D. x = x 0 corresponds to x = 0 .d) where p l n = pf°(O) and p 2 n = p f (0) .E. Without loss of generality. we can translate the origin and write x* = x . The differentiated series have the same radius of convergence R and converge to the corresponding derivatives of f (x). oo oo If ^ n=0 OO b n x n is another power series with radius of convergence R.
In standard form. . and c^ . + k n xn + .34a) where c n (n = 0..96 ADVANCED MATHEMATICS The homogeneous form of Equation (2.. Obtain the power series solution of (l+x2)y"+2xy'2y = 0 in the neighborhood of the origin. 1.) are constants.35) (2...37) .34a to c) into Equation (2.2 n=2 Substituting Equations (2..33) results in an equation of the form k 0 + kj x + k 2 x 2 + . 1. ) is zero.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.. Differentiating term by term yields oo / = X ^n^"1 n= l oo (23"4b) (2.35) is true for all x and this implies that each of the kj (i = 0. Equation (2. The recurrence equation (kj = O) allows us to determine c n in terms of pjj and P2j . Equation (2.31.34c) y" = X n(nl)c n x n . = 0 where kj are known expressions in terms of p^:.36) is written as y"+ 2x (2. P2k .. Example 2.. The next example illustrates the method of obtaining a series solution.36) y' 1+x2 2—y 1+x2 = o (2..
( s .34a to c) such that the summation index starts from zero.38c) To avoid having to expand (1 + x 2 )" 1 .2 c r x r ] = 0 r=0 (2. ] +CjX (2. C4. we write Equations (2.38a) y1 = X ( r + l ) c r + 1 x r r=0 OO (2. c) respectively.l ) c s / ( s + 1) (2.38a to c) into Equation (2. eg. in terms of Cj.f) 6C3 + 2CJ2CJ = 0 12c4 + 2c 2 + 4 c 2 .l ) c s + 2 s c s . we have x°: x1: x2: 2c22c0 = 0 => c 2 = c0 => c3 = 0 => c 4 = c 2 /3 (2. For ease of computation. These equations are now written as oo y = X r=0 OO CrxI" (2.38b) y" = ]T (r + 2)(r+l)c r + 2 x r r=0 (2. in terms of c 0 and C3.310h) 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.36) rather than with Equation (2. n = r + 2) in Equations (2. b.37).310e. To achieve this.b) (2. we write (n = r. ..d) (2.h) We note from Equation (2.. It follows that the solution can be written as y = c o [ l + x 2 . . That is to say. C3 is zero and consequently all the coefficients with an odd index greater than or equal to three are zero. . we work with Equation (2.34a) with y' and y" given by Equations (2.34a. .310g. cj..310a.34b.SERIES SOLUTIONS AND SPECIAL FUNCTIONS 97 The functions pj [ = 2 x / ( l + x 2 ) ] and p 2 [ = . c) respectively.39) Comparing powers of x.2 / ( 1 + x2)] are analytic at the origin.x 4 / 3 + x 6 / 5 .. We seek a solution given by Equation (2. Substituting Equations (2.2 c 2 = 0 xs: (s + 2 ) ( s + l ) c s + 2 + s ( s . we obtain c 2 . In the present example.2 c s = 0 => c s + 2 = .311) . C5.310c. n = r + l . 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 .36)..
31. The solution yi is in the form of an infinite series and is valid as long as the series is convergent..313) y2 = x (2. The point ( x = l ) is a singular point (see Example 3..315a.31.x 4 /3 + x 6 /5 . . b) are convergent for  x < 1 and pj (x) and p 2 (x) are analytic in the interval  x < 1. The solution yj is valid for  x  < l . At x = 1... it is convergent.315b) We note that the series in Equations (2.315a) (2.311). In Example 2.314) (2. y 2 (= x) has only one term and is a valid solution for all values of x.x 2 + x 4 . The general solution y is expressed as y = Ay1+By2 where A and B are arbitrary constants. • Note that Equation (2. There is no loss of generality in setting c 0 and Cj to be equal to 1.312a. 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.x 2 + x 4 .b) The infinite series is convergent for x < 1. From Equation (2..x 6 + .65). In Example 2. On expanding p^ (x) and p 2 (x) in powers of x..D.x 3 /3 + x 5 /5 . ) (2.x 6 + .31 and can be generalized to all power series solutions about an ordinary point. The general solution is given by the linear combination of yj and y 2 .E and has two linearly independent solutions yj and y 2 .. The fundamental solutions y^ and y 2 can be written as yj = 1 + x 2 . The .. 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. 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.) (2.2 (1 ... it can be seen that yj can be written as yx = 1 + x (x . ) p 2 (x) = . the solution y 2 (= x) is valid for all values of x.98 ADVANCED MATHEMATICS where c 0 andcj are arbitrary constants.36) is a second order O. we obtain Pj (x) = 2x (1 . This observation is not restricted to Example 2. the series is an alternating series and by Leibnitz's test.
we obtain (2.. b) into Equation (2. Alternatively Equation (2.43a) (2.4lb) x 2 p 2 (x) = b o + b 1 x + b 2 x 2 + . Substituting Equations (2. (2.) y = 0 (2.4la) (2.4 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.12a) and considering the homogeneous case. we require r ( r .46) is a quadratic in r and has two roots rj and r 2 . Further examples will follow to illustrate this statement.45) (2. we obtain y" + (a o /x + aj + a 2 x + . It admits a solution of the form y = xr On differentiating and substituting into Equation (2.. The two linearly independent solutions are x r ' and xr2.44) Equation (2.43b). 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.43b) can be transformed to an equation with constant coefficients by writing x = el Using the chain rule.) y' + (b Q /x 2 + bj/x + b 2 + ..46) (2. xpi (x) and x 2 p2 (x) can be expanded as xpj(x) = aQ + ajX + a 2 x + .43b) Equation (2.42) We seek a solution in the neighborhood of the origin and the leading terms of Equation (2..4la. That is to say... we obtain x r [r (r 1) + a o r + b 0 ] = 0 To obtain a nontrivial solution (xr ^ 0)..42) are y" + (a o /x)y 1 + (b o /x 2 )y = 0 or x2y"+a0xy'+boy =0 (2.47) .43b) is Euler's (Cauchy's) equidimensional equation.l ) + aor + b o = 0 (2..
n MATHEMATICS dy _ t dy (2.b) = 0 (2. 8a.43b. y2 = x 1/z (2.48b) (2.46) becomes r(rl)r+l=0 or (r1)2 = 0 (2. Example 2.49) The coefficients in Equation (2.410b) The two linearly independent solutions are 1/9 — 1/9 Yl = * . a) b) a) 4 x 2 y " + 4 x y ' .1 / 2 (2.100 ADVANCF.46) is 4r(rl) + 4rl The two roots are vx = 111 and r2 = .412a.411) (2.b) The general solution is a linear combination given by y = Clx1/2 + c2x1/2 (2. Equation (2.y =0 x2y"xy'+y=0 In this example.49) are constants and can be solved by the methods described in Chapter 1.413a.413c) b) In this case. + ( a o .414b) We have a double root (r = 1) and we have only one solution .41. Solve the following Euler equations.414a) (2. b) yields i ^ .410a) (2.l ) ^ + boy = 0 dr dt (2. Equation (2.48a) ^ = et(e^ + et^) dx 2 \ dt dt 2 / Combining Equations (2.
b) (2.421) where c 0 is not zero and r is any real or complex number.415) (2.46) has a double root. Equation (2. f (r0) = 0 (2.[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. . we note that if f (r) has a double root at r 0 .418) (2.417a) (2. x r and ^r.419) (2. we should try a solution of the form CO (2.(x r ) are solutions of the or differential equation.417c) Alternatively.611)] Yi = el = x y2 = t e l = x i n x • Note that the exponent r is not necessarily an integer.d) y = x r X cnxn n=0 (2.417b) (2.410b) can be written as *JL _ 2 ^ + y = 0 dr dt The solutions are [see Equation (1.(x r ) = j .420c. To differentiate with respect to r.47)].420a. This suggests that if we retain all the terms on the right side of Equation (2. b). b) This means that if Equation (2.4la. then f(r 0 ) = 0.SERIES SOLUTIONS AND SPECIAL FUNCTIONS ]QJ Yl = x To obtain the other solution. That is to say j .416a. by changing the independent variable x to t [Equation (2. we write x r as exp (r i n x).
423) From Equations (2.46)] and is the indicial equation. In this example p : (x) = ( x + l ) / 2 x .424a to d). and the two roots differ by an integer. the two roots are coincident.b) (2. The equation associated with the lowest power of x is a quadratic equation in r [see Equation (2.422D) If the two roots coincide (see Example 2.102 ADVANCED MATHEMATICS The series is differentiated term by term and substituted in the differential equation. we obtain a set of algebraic equations.d) (2.4lb) or the two roots differ by an integer.39atoh)]. p 2 (x) = 3/2x x 2 p 2 (x) = 3x/2 (2.4 .424a. The other equations are the recurrence formulae and are used to determine the coefficients c n [see Equations (2. the indicial equation yields two distinct values of r which are denoted rj and r2. Obtain a power series solution to the following equation 2xy" + (x + 1) y' + 3y = 0 in the neighborhood of the origin. We seek a solution of the form oo y = Z c n* n + r n=0 ( 2 .424c.422a) yi = X n=0 cnx"+r2 (2. In the examples that follow. xpj(x) = ( x + l ) / 2 .426a) . yj and y2 are not linearly independent and we have to modify our method. On comparing powers of x.25 ) On differentiating.42. In general. Example 2. we consider the three possible cases: the roots of the indicial equation are distinct and do not differ by an integer. The two linearly independent solutions are oo yi = S n=0 Cnx"+Fl (2. we deduce that x = 0 is a regular singular point. This method of solving a differential equation is called the method of Frobenius. we obtain oo y1 = X n=0 (n + r)cnxn+T1 (2.
7 c 0 (Z.2 (2.430d...7 » 9 « l l c o / [ 2 3 » 7 « 5 « 3 » 2 .430a) (2. 30a to e). and they are c 1 =7c 0 /[2«3] c 2 = .423) yields oo J n=0 [2 (n + r) (n + r ..—  + 7)(2s + 5 ) .c) (2.e) (2.. 26a.l ) c n x n + r .. .428g.1 x r .[(s+r+3) c s ]/[(s+r+1) (2s+2r+1)] (2.430b. C2.Z * + 21*'jGL*» + . b) in Equation (2.3 ] We denote the solution corresponding to r = 1/2 by yj and it can be written as (2.l l c 2 / [ 2 « 7 » 3 ] = ..426b) Substituting Equations (2.9c2 /[2 • 5 • 2] = 7 • 9 c o /[2 2 • 5 • 3 • 2] c3 = ..<t3Z) Cs+1 2 S + 1 (2s + 3) (2s + 1).425.1)] =0 => T1 = 1/2 and r 2 = 0 xr: (2.427) We compare powers of x.h) x Substituting the value of r = 1/2 into Equation (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 = .428e. = v / T [ l .1 : 2 r ( r .429) y .l l . The lowest power of x is x r .] From Equations (2.43!) (l) s + 1 (2 S . C3.429.428h) leads to c s + 1 = .SERIES SOLUTIONS AND SPECIAL FUNCTIONS DO 103 y" = £ n=0 (n + r)(n + r . 7 • 5 • 3 • (s + 1) (s). we deduce that c (2.[(2s + 7) c s ]/[2 (2s + 3) (s + 1)] We can compute C\. 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)] (2.1) 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..9 .l ) c 0 + rc 0 = 0 => c 0 [r (2r.
e)  ( .ca ( 2 4 3 5 ) 2 n " j 5 ! (n + 2)! (2n+l)! The solution y^ can be written as y.. .3 b o / [ 2 .3 ] b 3 = .434) ..104 ADVANCED MATHEMATICS The right side can be simplified by the following identities (2s + 7) . Equation (2.432. ..428h) becomes bs + 1 = .4 b 1 / [ 2 ..5 • 4 • 3 b o /[2 • 3 • 3 • 5] From Equations (2.5b 2 /[3 • 5] = .434) becomes Cn (2. we deduce that b s+1 (2. 3 b 0 (s+l)s. are bl = .=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 . Equation (2.3 b 0 b2 = .l ) S + 1 ( s + 3)(s + 2 ) . 3 = ( 2 s + 3)! (2. 53 V ' ..32(2s+l)(2sl). b3. 5 ..3 ] = 4 . we consider the case r 2 = 0.433a) 6! 2 s ( s + 3)! (2s + 3) . 7 = (2s + 7 ) ! 3 ! (2. 33a. 38a to e).( s + 3 ) b s / [ ( s + l ) ( 2 s + l ) ] The coefficients bj. We substitute this value of r2 in Equation (2.! ) » • (2s + 7)1 ca 2 s 5! (s + 3)! (2s+ 3)! Setting n = s + 1. b 2 . .433b) 2 S + I ( s + 1)! Combining Equations (2.438a) (2. b) yields _ ( .438b. 4 ...428h) and to avoid confusion we denote c s by b s for the case r 2 = 0.c) (2..437. _ ^ ( 2 n + 5V.438d..437) (2.
Example 2. Again setting s + 1 = n.429. >oo ^ b — s .= lim S (s+l)(2s + l) / >o° v (S + 3) . Obtain a series solution to the equation x 2 y " + x y ' + x2y = 0 in the neighborhood of the origin.b) (2.443a._ . we find that the solution y2 can be written as (2... (2n)! The arbitrary constants c 0 and b 0 are determined from the initial conditions or the boundary conditions. + l)l (2s + 2)f A = (24"39b) ( .442a..SERIES SOLUTIONS AND SPECIAL FUNCTIONS 105 ( . 37). Ar. p2 = 1 x 2 p 2 = x2 (2. we have p x = 1/x.445a. xpj = 1. . = oo (2.26..d) (2.l ) s + 1 (s + 3)! 2 S + I ( s + 1)! b n = ~ 2 ( .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 .i r 1 2 s ( s + 3)!b0 (2s+ 2)! U4^c. 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 .44. (2.' < " + 2>!*n n=o (2. From Equations (2.43.c) R = bm The solutions yj and y2 are valid for all values of x.b.b. .444) The origin is a regular singular point and the series solution is of the form . In this example.c) _ S . . . ..
450d.444) yields oo £ n=0 [(n + r) (n + r ..c 4 / 6 2 = .1) c n x n + r + (n + r) c n x n + r + c n x n + r + 2 ] = 0 (2. starting with the lowest power (n = 0)..106 CO ADVANCED MATHEMATICS y = X n=0 Cnxn+I (2.1) + (s + r)] + c s _ 2 = 0 => c s = .b) (2.448h) becomes C2m = c2m2/(2m)2 ( 2 .e) 0/2 c 4 = .c o / ( 2 + r) 2 (2. c 4 .446) Differentiating and substituting in Equation (2. we deduce that if s is odd. are given by c C2 c/22 c (2. we have xr: c o [ r ( r .450a) (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 . we deduce that C2m = ( .450b.d) (2.448e.449. (2m)2] = (l) m c 0 /[(2 2 m )(m!) 2 ] (2. 50 a to e).448h). eg .4 2 » 6 2 ] From Equations (2.448c.448g.h) Note that with r = 0.c 2 / 4 2 = c o /[2 2 «4 2 ] c6 = .l ) m c 0 / [ 2 2 « 4 2 .. c s is zero. we deduce that Cj is zero.447) Comparing the powers of x.c s _ 2 /(s + r) 2 (2.l ) + r] = 0 => r = 0 (double root) =» Cj = 0 (r = 0) => c 2 = .. From Equation (2.451b) One series solution can be written as .448a.451a) (2. 6 2 .c) (2.4 ' 4 9) The coefficients c 2 .c o / [ 2 2 . We consider the case where s is even and we write s = 2m and Equation (2.
SERIES SOLUTIONS AND SPECIAL FUNCTIONS ~ m=o r ££Z _nm 2m XiW = c o Z V T (2452) 22m(m!)2 Equation (2.I ^r~I 2 (m!) (2453) In this example.455) Differentiating y. b) in Equation (2. we obtain oo y1 = i n x JQ(X) + J0(x)/x + X n ^ x " " 1 n=l OO (2. Equation (2. the form of the series solution is oo y = (in x) y! (x) + X n=0 bn x n + r ° (2. we obtain .444) is Bessel's equation of order zero and the equation of arbitrary order will be considered later. From Equation (2.456a) y" = i n x Jo (x) + 2JQ (X)/X .444) and the terms inside the square bracket equal zero.+x2Joj + 2xj. Example 2. oo oo oo inx[x2Jo+xj.453).452) becomes ~ m=o /_ n m 2m W .454) Note that if the double root of the indicial equation is r0 (*= 0).448a)] has a double root and the second linearly independent solution is not readily available. It is customary to set c 0 = 1 and to denote yi (x) by J 0 (x).454.457) The function Jo is a solution of Equation (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 (2.444) yields .1) b n x n " 2 (2. the indicial equation [Equation (2. 56a.456b) Substituting Equations (2.J 0 (x)/x 2 + ^ n=2 n (n .4lb suggests that we look for a solution of the form oo y = i n x J 0 ( x ) + X bnxn n= l (2.
as follows b4 = . is denoted by y2 and is given by y2=inxJ0(x) + ^ .460b.b) (2. bg. x1: x2: x3: b] = 0 .460c)] and from Equation (2.e.2 2 s ~ 2 s! (s1)! =0 (2.462). (2.462. ..d) = . 63a to d).l + 2 b 2 + 2b 2 = 0 6b 3 + 3b 3 + b 1 = 0 => => b 2 = 1/4 b3 = .458) m=l 2 (m!) Substituting Equation (2.31).458) in Equation (2.459) To obtain the coefficients b s .461) The recurrence formula is \ b2m (l)m 2^7^ + b 2rn2 1/ (2m)2 (2 462) ~ L22m m!(ml)! J/Um) ^..2 4 3!2! j / 3 6 = m24 The second solution. .[ 4 ^ + b 2 ] / > " = . x 2m : 9 .f) One observes that b s is zero if s is odd and we need to consider the even powers of x only.^ + s(s+l)bs+1xs+1+sbsxs + bsxs+2 . which is linearly independent of Jo.463c.b : / 9 = 0 (2. .464) From Equations (2.463a. we can obtain b4. we can verify that the general term b 2 m can be written as .c) (2.460a) (2. we obtain £ s=i — ^ .10& ADVANCED MATHEMATICS (2. l + 1 ^ . 22m ' + 2m (2ml) b 2 m + 2m b 2 m + b 2 m _ 2 = 0 m!(ml)! (2.457) and changing the indices appropriately such that the summation index s starts from one (see Example 2.T  8 b6 (2.4 sz) The coefficient b 2 is known [Equation (2.460d. we compare powers of x.
x = 1.465) The solution y2 is oo y2 = i n x J0(x) + X b 2m x 2m m=l (2. In this example.468) yields oo £ n=0 [(n+r) (n+r1) cn x r + n + 2 . x(x2l) P2 = f ^ x2(x2l) (2.b) The singular points are at x = 0.469a. we obtain .. and x = —1.1) y" .470) Differentiating and substituting the resulting expressions in Equation (2. .444) is the linear combination of Jo and y2 and is y = AJ 0 + By 2 where A and B are constants. we require B to be zero.471) Comparing powers of x. If the physics of the problem require y to be finite at the origin.44.(n+r) (n+r1) cn x r + n . Solve the equation x 2 (x2 .^ .467) p .466) The general solution of Equation (2.. This implies that y2 is singular at the origin.468) (2.SERIES SOLUTIONS AND SPECIAL FUNCTIONS 1J& (2. pi and p 2 are given respectively by (2.(x2 + 1) x y' + (x2 + 1) y = 0 in the neighborhood of the origin.(n+r) cn x r + n + c n x r + n + 2 + cnxr+n] = 0 (2.. 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. Example 2.(n+r) cn x r + n + 2 . Note the presence of Jtn x in y 2 .
472g. Substituting the value of r = 1 in Equation (2.472J) implies that c 4 = c 6 = .474a) = c o x i n x + J bn x""1 n=0 (2.r + l ] + c 2 [ .O + r J r .f) x r + 1 : C j t .l ) .r + l ] = 0 => c t (r 2 + 2 r .( 2 + r ) + l ] = O => c 2 = c 0 ( r .476) .l ) 2 / [ ( 2 + r ) 2 .1] (2.468) yields 00 X [(nl)(n2)b n x n+1 (nl)(n2)b n x n .e.475b) Substituting y.1 ) = 0 => c 1 = 0 ( r = +l) x r + 2 : c o [ r ( r . the solution corresponding to r = ... = 0.473) To determine y 2 .2) b n x11"3 (2.c) (2.472h) and we observe that the numerator ( r .b.4721 j) We deduce from Equations (2. 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) (2.1 n=0 + b n x n + 1 + b n x".1) b n x11"2 n=0 OO (2.472d.474b) On differentiating term by term.r ( r .472h) leads to c 2 = 0 and Equation (2.1 ) 2 is nonzero and the denominator [(2 + r) 2 . we obtain 00 y1 = c 0 ( i n x + 1) + ]T (n . = 0. The solution yi corresponding to r = 1 is a polynomial (one term only) and is yi = c o x (2. y'.472f. we substitute this value of r in Equation (2.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 .472a.r + 1 ] = 0 => r 2 l=0 => r = ±1 (2.1 ] .2c 0 x = 0 (2.475a) y" = co/x + £ n=0 (n " 1) (n . j) that c^= c 3 = C5 = .1 ] (2.( 2 + r ) ( l + r ) .110 ADVANCED MATHEMATICS xr: c o [ ..1 .1 (nl)b n x n+1 (nl)b n x 11 . and y" in Equation (2..1] is zero.l ) .
b.45.477b) (2.3 b 4 + b2 + b4 = 0 => => b 0 = cQ/2 b4 = 0 (2.l .. = 0 (2. B has to be zero. = O.480a.468) is y = Ayi+By2 where A and B are arbitrary constants. corresponding to r = .b 2 .. we can set c 0 = b Q = 1 and the two linearly independent solutions y\ and y 2 can be written as yi = x.b) The general solution of Equation (2..l / 4 ) y = 0 valid near the origin.470). bo*O.483) . x"1: x°: x1: x3: b 0 (2 + 1 + 1) = 0 b! = 0 2b 0 + b 0 .( 3 ) ( 2 ) b 4 . y 2 = x i n x + 1/(2x) (2.b 2 + b 0 + b 2 . we obtain oo (2. Proceeding as in the previous example.f) From Equations (2. Example 2. is given by y2 = c 0 x i n x + bo/(2x) (2.477e. b 2 = b 4 = .477a) (2.. The origin is a regular singular point [xpj(x) = 1. 77a to f).479) Without loss of generality.482) ] £ [(n +r) (n + r 1) cn x n + r + (n + r) cn x n + r + cn x n + r + 2 .477c.(1/4) cn x n+r ] = 0 n=0 (2.478a. Find a series solution to the equation x 2 y" + x y ' + ( x 2 .c) The solution y2. Note that y 2 is singular at the origin and if the general solution y is finite at the origin.SERIES SOLUTIONS AND SPECIAL FUNCTIONS 777 We now compare powers of x. we deduce that b 1 = b 3 = . x 2 p 2 (x) = (x2 .d) (2.481) (2.476.1/4)] and we seek a solution in the form of Equation (2.2 c 0 = 0 .
b) In Equation (2.b) 2.l / 4 ] = 0 => r = ±1/2 (2.l ) + r . m=0 (l) m x 2m+1 (2m+l)! (2. we deduce that C2m (2.486.485) From Equations (2.486).e) = (Dmc0/(2m+l)! (2.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. the sum is sin x and yj can be written in closed form as yj = c 0 x.487a) (2.co L m=0 f ( l) m x 2 m + ^ 2 (2m+1)! _ .c) (2. 85).484f) and we obtain c s= (2.1 / 2 sinx To obtain y2.4. we substitute the value of r = 1/2 in Equation (2.490) .484f) and we obtain (2.484c.484a.484d) x r + 1 : cY [r(l + r) + ( l + r ) .f) c s _ 2 /[s (s + 1)] (2.84c) (2.C 2(ml)/[( 2m ) ( 2 m + Dl (2.112 ADVANCED MATHEMATICS Comparing powers of x yields xr: c o [ r ( r .cox 1/2 f 2.489a. we compute the first few coefficients and they are c2 = c o /[23] c4 = _c 2 /[4«5] = c o /5! c6 = . 87a to e).488) The solution yj is given by yi .7 ] = c o /7! From Equations (2. we note that all the coefficients with an odd index are zero.489b). we obtain C2m = .487b.484e. Writing s = 2m.486) From Equation (2.487d.c 4 / [ 6 .
. to the extent of a multiplicative constant. y 2 = x" 1 / 2 cosx (2. We compute the first few coefficients as follows c2 = . we deduce that C] is arbitrary and we can compute cj.492b) From Equations (2.IV3) (2. The coefficients with even indices can be written as multiples of c 0 and those with odd indices are multiples of q .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 (2.4 ) = c 1 /5! .c 1 / 7 ! . (2.. yx = x ^ 1 / 2 s i n x .496) . Example 2. c2m+1 = (l)mCl/(2m+1)! (2.l ) m x 2m /(2m)! is cos x and the two linearly independent solutions y^ and y 2 are. we verify that C2m = (Dmc0/(2m)! . All these coefficients are nonzero. we require B to be zero.c 1 / ( 3 » 2 ) . Observe that y 2 is singular at the origin and y^ is finite at the origin.491) The denominator is zero if s = 1 and it is not possible to determine c^.c 4 / ( 6 • 5) = c Q /6!. c 4 = . c3 = .. (2.482) is y = Ay!+By2 where A and B are arbitrary constants.494) is y^ (with Cj replacing c 0 ).. For solutions which remain finite at the origin. From Equation (2.46.c 3 / ( 5 .495a. 92a. c 5 = .484c).c 0 / ( 2 ) .492a) . associated with a first order irreversible reaction involving isobutylene and spherical cation exchange resin particles.b) The general solution of Equation (2.. c 7 = c 5 /C76) = . c 6 = . b). c 5 .491. Equation (1) of their paper can be written as (see Equation A. . The summation ^ m=0 ( .SERIES SOLUTIONS AND SPECIAL FUNCTIONS cs LU. = cs_2/[s(sl)] (2. Gupta and Douglas (1967) considered a steady state diffusion problem.c 2 / ( 4 « 3 ) = c 0 /4! ..493a..
498a) The boundary conditions are cA = c A . so as to avoid confusion with the radial variable r.g.d) r 1 + P: c 1 ( l + p)(p) + 2 ( l + p ) c 1 = 0 => cx (1+p) (2+p) = 0 (Cl=0.498ac) for c A via the method of Frobenius.^ +2 r ^ A kr2cA = 0 dr dr (2.Ul ADVANCED MATHEMATICS (2.k c A dr v dr / (2.4101e.D. Differentiating Equation (2. if p = . We seek a solution of the form oo C A= X n=0 Cnr"+P (2499) In Equation (2.f.b.1 if p = 0 (2.499).l ) + 2pc 0 = 0 =» p ( p + l) = O => p = 0 a n d p = . Equation (2.497b) Assuming c A to be a function of r only (symmetry!).497a) _r where R A = .498b) (2. given by ^AB r \ 2 i . ^ dr = 0 at r = R at r = 0 (2.499) and substituting the resulting expressions in Equation (2. =>  Cj is arbitrary .4100b) where K = k / .4101a.c. the exponent involves p.l (2. 0 ^ Comparing powers of r.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. Note that the origin (r = 0) is a regular singular point.498c) We now solve Equations (2.E.h) .4100a) (2.497a) reduces to an O. we have rp: c o (p) ( p .
To compute the even solution. we replace s by 2m and Equation (2.4107) becomes (2.Dj and deduce that Kmc c ^ = ^TTjT <24105> One solution is ~ m=0 Km 2m (2. = ( g J ^ + 1) (2. We proceed by replacing s by 2m and Equation (2.4101ij) To determine c s . c 0 and C}.SERIES SOLUTIONS AND SPECIAL FUNCTIONS m r s + P: c s ( s + p ) ( s + p .Kc0 2~2.4101h) states that q = 0 if p = 0.3' c _ K c 2 _ K Co 4 " 4 ^ 5 " " 5! f2 4 104a hi U. c s = 0 if s is odd.K c s _ 2 = 0 => c .4106) Substituting the other value of p in Equation (2.4102) becomes C ^ 2Om+l) (24"103) We compute the first few coefficients as follows r C .l ) + 2 ( s + p ) c s .4iU4a.1) (s) In this case.4107) (s . cs=7 s 2 wh (2. we have two arbitrary constants. we substitute the values of p in Equation (2.4108) .4101 i). Therefore.4lOlj) c^T^ry forp=0 <24102) Equation (2. we obtain Kc. This implies an even as well as an odd solution.
we obtain Kmc c2m=^/ For the case where s is odd. 2 £ ^ .4112) can be written as CA .ul) Another solution to Equation (2. we write (24109) ^ " ( 2 m ) ( 2 ^ 1) (24U0) In terms of c^.4106).4112) Note that the second term on the right side of Equation (2. 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. we obtain Kmc C*=(2^7TJ! <2 . Therefore.4115) and applying the boundary conditions requires cA = ^ .h x = X (f^ • Equation (2.4112) is the general solution of Equation (2. The solution can be written in a closed form by observing that ~ m=0 v 2m ' " m=0 v 2m+l .4116a) .h x = X (tTJv (24H3a.4112) is equal to the right side of Equation (2. 5 ) ^ ^ m=0 m=0 (2.498a) is ' A .sinhrVKT (2.498a).c o s h R V ^ + ^ . " c.s i n h R V F Ai R R (2.116 ADVANCED MATHEMATICS As before.4 .^ .+ 0 . Equation (2.4114) Differentiation yields ^A dr = _ £Q.b) = T" coshr^K~ + 7.
we obtain c 0 = 0. we obtain a quadratic (for a .4114) yields the final result c A R sinh r VK _i r sinh R VK (2.4116b) On evaluating Equation (2. On setting the coefficient of the lowest power of x to zero.SERIES SOLUTIONS AND SPECIAL FUNCTIONS 1JJ 0 = [c 0 cosh r VK + c 0 r VK sinh r V K . We now have c A . We assume that the solution is of the form oo y = Zcnxn+r' n=0 c 0*° (2.c j sinh r VK + c l r VK cosh r V K ~ ] / r 2 (2. • (2. we note that Equation (2.4116b).5117) cA = (2. = ^sinhRVKf and substituting for cj in Equation (2. Before summarizing the Frobenius method. as r—> 0.4121) We differentiate the series term by term and substitute the resulting expressions in the differential equation. 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).498a) can be transformed to a much simpler equation involving constant coefficients.4118) which is Equation (2) in Gupta and Douglas (1967).4120) We summarize Frobenius's method of finding a solution near a regular singular point as follows.4119) (2.
2. we assume a solution in the form of Equation (2. The two linearly independent solutions are oo yi = X ^ ( r ^ x 1 1 ^ n=0 (2. we assume a solution of the form oo y = y i i n x + £ bnxn+r> n=0 (2.45 and 6). we can obtain the second solution by the method of variation of parameters and this is explained in the next section. c 3 .12a) if one solution is known. In this case. and y" in Equation (2.. (c) rj and r 2 differ by an integer. c 2 .. If. If one solution is known. To obtain the other solution. We try to compute the coefficients c s with the value of r = r 2 . the second linearly independent solution y 2 is obtained. one solution yj is given by Equation (2. in the computation of c s .4122a) yi = X n=0 cn(r2)x"+r2 (2. in terms of c 0 and r\ (or r 2 ).4123) and proceed to calculate b n ..4122b) (b) ?i = r2.4123) We proceed as in case (a) to obtain b n .44. By comparing powers of x.5 METHOD OF VARIATION OF PARAMETERS We use the method of variation of parameters to find a second linearly independent solution of Equation (2. Let us assume that r t > r 2 .12a). The solution y 1 can be obtained as in (a).51) .4122a). If all the coefficients can be computed as in Examples 2. Let yj be a solution of Equation (2. We now consider the following three cases. we have to divide by zero as in Example 2.118 ADVANCED MATHEMATICS second order equation) equation in r.12a) and we assume a second solution to be given by y(x) = u ( x ) y i ( x ) On differentiating and substituting y. 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. y1. we obtain (2. We denote the two solutions by rj and r2. (a) rj and r2 are distinct and do not differ by an integer. we obtain a recurrence formula that allows us to obtain cj.
59) becomes x (x 2 .3) v = 0 dx (2.52) simplifies to u ' ( 2 y . + p 1 y 1 ) + u"y 1 = 0 Since yj is a solution of Equation (2. + p 1 y 1 ) + u"y 1 = 0 We substitute u1 by v and Equation (2.58) in Equation (2.12a). we obtain i n v = .53) can be written as (2.58) .I pjdx (2.J P l dx (2.x 2 (x 2 + 1)] = 0 We denote u' by v and Equation (2.51.468) and we obtain u" [x 3 (x 2 .+ (x 2 .53) j^=j(^H<fx On integrating.1) p. 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 ) (2.510) (2.59) (2.57) Example 2. We assume that the second solution is y = x u (x) We substitute y.2 i n y 1 . Obtain a second linearly independent solution of Equation (2.55b) One further integration yields u= I (l/yf) exp .468) given that one solution yi is x.54) (2.56) (2.1)] + u1 [2x 2 (x 2 .55a) or v = (l/y 2) exp . Equation (2.SERIES SOLUTIONS AND SPECIAL FUNCTIONS 7/g u(yi' + Piyj + p 2 y i ) + u 1 (2y. y'.52) (2. and y" from Equation (2.1) .
Many of the second order boundary value problems can be stated as follows (r y')'+ (q + A.6la) .p) y = 0.513) (2.51 lb) ' (2. 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.515) There is no loss of generality in setting cj = 1 and Kx is y j .dx \x. the conditions are imposed at the boundaries and these problems are boundary value problems. Equation (2.6lc) a<x<b (2. we need to impose two conditions. 2.514) (2.58.2. it is stated that the general solution of a second order O. we obtain v = cl(x2l)/\3 where C\ is a constant.510) can be written as ^ V =^ > d * X(X 2 1) (2.480b).UQ ADVANCED MATHEMATICS Equation (2. subject to a i y ( a ) + a 2 y'(a) = 0 b 1 y(b) + b 2 y'(b) = 0 (2.D.51 la) = [—^ + —Lr .515) is exactly Equation (2.E. In many physical problems.512) (2. is a linear combination of two linearly independent solutions.6 STURM LIOUVILLE PROBLEM In the previous section. Combining Equations (2.1 x+1 x J On integrating.6lb) (2. To determine the two constants.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.
p = x.SERIES SOLUTIONS AND SPECIAL FUNCTIONS 727 where r. a2.2<^J_+ x ^ I dx2 dx + (x2v2)y = 0 (2. and q are continuous real functions of x.6la or d) and some of them are given next. Equation (2. X= Z (Z + 1).62d) becomes x2y"+xy' + (Xx2v2)y = 0 or (xy')' + (v 2 /x + ?ix)y = 0 2 (2. q = .. r = ( l .62f) (In this case. A is a constant parameter (possibly . q = 0.6ld) Many of the equations in mathematical physics are special cases of Equations (2.) (b) Legendre equation (1 .62a) [In this case. p = l .x 2 ) y ' ] ' + i ( i + l)y = 0 (2. r = x. q = 0.x 2 ) y" .2xy' + Z (Z + 1) y = 0 or [ ( l . bj.] (c) Bessel equation We denote the independent variable by x and write Bessel's equation as 2 . The system defined by Equations (2.6la to c) is the Sturm Liouville problem. r = p = 1.62b) (2. p.x 2 ) .) . complex).62e) (2.v /x.62c) (2. and b2 are constants.62d) On setting x = x VX. (a) Simple harmonic equation y"+ ly = 0 (In this case.6la) can be written as r y " + r ' y ' + (q + Ap)y = 0 (2. aj. Equation (2.
r = p = e ~ x . y'(7t) = 0 (2.63b. b) is cl = c 2 = 0 This leads to the trivial solution (y = 0). X = Li.65b) (2.65a. we require 0 = C!+c2 0 = m (c.62h) or (e~ x y')' + )J. Solve the equation y" + Xy = 0 subject to the conditions y(0) = 0.m 2 and the solution of Equation (2. Example 2.66a. the nontrivial solution (y # 0) of Equation (2. = .6lb.65a) (2. c).64) The only solution of Equations (2. For convenience. c)] can be satisfied.b) .c) (2.2 x y ' + ny = 0 2 2 (2. We consider these three cases separately. or negative. emnc2 e~mn) (2.) In general. (2.121 ADVANCED MATHEMATICS (d) Hermite equation y " .61.63a) We assume X to be real and it can be positive. q = 0. To satisfy Equations (2.6la) depends on X and it is only for some values of X that the boundary conditions [Equations (2.62g) (2.<0.63b. These values of X are the eigenvalues (characteristic values) and the corresponding functions y (x. we set A. (a) X.e" x y = 0 2 (In this case. zero. X) are the eigenfunctions (characteristic functions).63a) is y = c1emx + c2emx where Cj and c 2 are constants.
The cos function has multiple zeros and cos n7t is zero if n = (2s+l)/2. The solution is y = c 5 sin nx + c 6 cos nx where C5 and eg are constants. (b) X = 0.68a.69) Equation (2.63b. (2.610b) (2. (2. The boundary conditions imply c3 = c4 = 0 Again the only possible solution is the trivial solution. The solution is now given by y = c3 + c4 x where C3 and C4 are constants.>0. and discrete. .. .SERIES SOLUTIONS AND SPECIAL FUNCTIONS J23. 1. Such properties are associated for example with discrete energy levels in quantum mechanics. • Next we discuss the general properties of the Sturm Liouville problem. positive.613) Note that the eigenvalues are real.610b) implies that either C5 is zero which leads to a trivial solution or that cos nn is zero which provides the nontrivial solution.612) The corresponding eigenfunctions are ys = sinf^s^Jjx (2.. We set X = n2.2. c) yields 0 = c6 0 = c 5 ncosn7c (2.b) (2.610a) (2. s = 0.63a to c) has an infinite number of eigenvalues and they are given by Xs = f2^1) ' s = 0' 1 ' 2 ' (2. Applying Equations (2.611) The system defined by Equations (2.67) (c) X.
bj and b 2 are real.616c) implies that X = X.616c) To obtain Equation (2.6la) by y and Equation (2.6lb. ) I + (q + X n p ) y n = 0 a i y n ( a ) + a 2 y. X is real. We multiply Equation (2.c) . Equation (2.y C r y ' V l d x (XX)l /a /a (2.r b p y y d x = (h [ y f r y ^ ' .614c) . Taking the complex conjugate of Equations (2.615) (2.614b) (2.(a) = 0 . we have used boundary conditions [Equations (2.y ' r y') dx Ja = 0 (2.614a) by y and subtract one from the other to yield (XX)pyy = y(ry')'y(ry')' On integrating. The functions y n and y m satisfy ( r y . b1 y n (b) + b 2 y'nQo) = 0 (2.616a) = [yry'yry']aI (y'r y ' .617b. c.614a) (2. q. p. we obtain (2. a2. aj. Since the eigenfunctions are nontrivial.616b) (2.6la to c) and noting that r.617a) (2. 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. c)]. that is to say.124 ADVANCED MATHEMATICS The Eigenvalues Are Real Suppose X is complex and this implies that y is also complex. 14b. The Eigenfunctions Are Orthogonal Let Xn and Xm (Xn ^ Xm) be two eigenvalues and their corresponding functions are y n and y m .616c).
if n = m if n ^ m (2. 5nm = 1 .623a. subtract one from the other and integrate the resulting expression to yield (anam)l p y n y m d x Ja = 0 (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. 18a) by respectively.SERIES SOLUTIONS AND SPECIAL FUNCTIONS 725 ( r y ^ ) ' + (q + ^ m p ) y m = 0 a1ym(a) + a2y^(a) = 0.c) y m and yn We proceed as in (i).b) . we multiply Equations (2.b) where 8 n m is the Kronecker delta and is defined by 0 . If Xn = Xm.617a. that is to say.620.619) 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).624a. Equation (2.618a) (2. 21) can be expressed as (2622) f P yn* ym dx = f ^ V Ja Ja n m dx = 5nm (2.618b. bl y m (b) + b 2 y ^ ( b ) = 0 (2.
.
s ( s .. we set k = Z (Z + 1) and Equation (2.710) For values of x = ±l.f) 4 « 3 c 4 .... . c 4 .. in terms of Cj... we consider the special case of k = 0..k ) / 1 2 (2.b. From the recurrence formula [Equation (2.78). This result can be extended to general values of k.76) Um n _ >O o ^±2 cn = 1^ n _ >TO n n +n ~k = 1 (2.75h) becomes ...c) 2+3n+1 To examine the validity of the solution at x = 1.78) We note from Equation (2.2 .75e. C5. c 3 . we deduce that y 2 is given by y 2 = Cj x (1 + x 2 / 3 + x 4 /5 + x 6 /7 + .75h)].2 c ! + k c j = 0 => c3 = c 1 ( 2 . From Equation (2.d) (2.75h) now becomes c s+2 = s c s / ( s + 2) (2.. From Equation (2. we determine that the series diverges by comparing it with ^ 1/n. we deduce that the radius of convergence R is 2 R = (2. The general solution can be written as y = c0yi+ciy2 where c 0 and cj are arbitrary constants. c 2 . = 0 and Yl is given by yi = c 0 (2. . c 4 ..h) The solution can be separated into an even function and an odd function.75g. We note that x = ±1 are singular points and.l ) c s . the even solution is a constant and is valid for all values of x including x = ±1.k ) / 6 => c4 = c2 ( 6 .75h). For convenience.75c.2 c 2 + kc 2 = 0 xs: ( s + 2 ) ( s + l ) c s + 2 . for this special value of k. . c 5 .SERIES SOLUTIONS AND SPECIAL FUNCTIONS 727 x1: x2: 3«2 c 3 .75b) that in this case c 2 is zero and this implies c 4 = c 6 = . ) (2. and we denote this function by yj. The even function involves c 0 . . one solution is valid at the singular points and the other is not.2 s c s + kc s = 0 =} c s + 2 = cs ( s 2 + s .79) In this case. The odd function y 2 is in terms of Cj .2 c 2 . Equation (2. can be expressed in terms of c 0 and C3. c 2 .77a.k ) / ( s + 2 ) ( s + l ) (2.
711).712c.128 ADVANCED MATHEMATICS c s + 2 = c s ( s 2 + s . The constants c 0 (or q ) are chosen such that P^ (1) is unity.i ) / ( s + 2 ) ( s + l ) = c s ( s .d) P.1).71 and are P 0 (x) = 1 .3x) (2. if i is odd. P2 (X) = 1 (3x2 .711) From Equation (2. P^x) = x P3 (x) = 1 (5x2 . the odd solution is a polynomial.J — A _ ( x 2 .i 2 . the degree of the polynomial is i .l( FIGURE 2. The first few Legendre polynomials are shown in Figure 2. The case we considered earlier is k = 0 {Jt = 0) and the even solution is a constant. In particular.712a. 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. we observe that if Z is a nonnegative integer. In general. the even solution (a polynomial) is valid for all values of x and.713) .i ) ( s + i + l)/(s+2)(s+l) (2. These polynomials are the Legendre polynomials and are denoted by P^ (x).U) PpU) * i\ ' P*(XX P(x J // yf l \ / \ /// .b) (2. if i is even.71 Legendre polynomials A relatively easy method of computing P^ (x) is to use Rodrigues' formula which can be written as P^ (X) = .l / 2ZJt\ dxZ (2. for any integer i .
2xu'P i = 0 Since P^ is a solution of Legendre' s equation.71) with k = i ( i + 1).)x/(lx2)] = 0 dx A * Equation (2.i P i _ 1 ( x ) Knowing Po and Pj.715) ( i + l)Pj + 1 (x) = (2i + l ) x P i ( x ) . The second linearly independent solution of Equation (2. we can calculate ?2.718) simplifies to (lx2)Piu"+uI[2(lx2)P/J2xPi] = 0 On writing u' = v.2xP/J + i ( i + 1) Pj] + (1x 2 ) (u"P i + 2u'P<J) .. . 20..716) On differentiating Q^ (x) twice and substituting the resulting expressions in Equation (2. is (2. From Equations (2.719) can be written as 4 x + 2v[(Pl/P. and the integrating factor IF.720) (2.62c. An important property of the Legendre polynomials is the orthogonal property.D.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 .. Equation (2. we deduce that the Legendre polynomials are orthogonal with respect to weight one.717) (2. 21). 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.718) . We denote the solution by Q^ (x) and assume that Q i (x) = u(x)P i (x) (2. we obtain u [(1x 2 ) PJ .719) (2. 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. Equation (2.71) can be obtained by the method of variation of parameters. P3.720) is a first order O.E.
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) (2.722) (2. Calculate the Legendre polynomials of the second kind.71.724) ] (i$)p?($) The functions P^ (x) and Q^ (x) are Legendre's polynomials of the first and second kind.720) can be written as jL [V (1x 2 ) P / ] = 0 The solution is v = C/[(lx 2 ) P / ] where C is a constant.723) Q^ (x) = C P .)2x/(lx 2 )]dx (2. (x) I ^ (2. From Equations (2. Example 2. 24).7250) . = exp [(2P / J/P /( . It follows that Q^ (x) is given by /•x (2. /•x ^ (1$) (2. Q o (x) and Qj (x).72 la) (2.UO ADVANCED MATHEMATICS IF.712a. we have /•x Qo (x) = C I .721c) = exp [IJlnPjt + i n (1x 2 )] = (1x 2 ) P / Equation (2.721b) (2.
. we obtain Qx (x) = . the even solution P 0 (x) (=1) is valid at  x  = 1 and the odd solution Q o (x) is not valid at  x  = 1.( . ) (2. + 1 = x (1 + 2X) Q . • The other Legendre functions of the second kind can be computed from the recurrence formula ( i + 1) Qj. we choose C to be one and Q^ (x) is given by < 2 . .x 4 / 4 .727b..726.. 29)..x 4 / 4 + .x 3 / 3 .726) (2.727a) (2..7 " 28c » (2. we obtain (2..1 + 1 (x 2 + x 4 /3 + . we note that Q o and Qj have singularities at x = ±1. the odd solution Pj(x) (= x) is valid at  x  = 1 and the even solution Qj (x) is not valid at  x  = 1.728a) = Cx J p + * (T^+TT$ p (8 ) 272b = Cx [M in (T^)] As in the case of Q o (x).SERIES SOLUTIONS AND SPECIAL FUNCTIONS 131 Qo (x) = 1 in [ 1 ± A ] Expanding Qo (x) about the origin yields Q 0 (x) = l [ x . . For I = 0 ( i is even). 24).731) . The infinite series given by Equations (2.730) From Equations (2.X Q ^ j (2.729) Qx(x) = 2 L i n J j i i ) ..)] = x + x 3 /3 + .712b..x 2 / 2 . Likewise for i = 1 ( i is odd). From Equations (2. 30) are the infinite series solutions of Legendre's equation and are valid for  x  < 1.727b) Q1(x) = C x  ^ (2.x 2 / 2 + x 3 / 3 ..l Expanding the Jin function in powers of x.x .
x 2 ) 1 7 ^ (2. Equation (2. Equation (2. We start by considering the simplest case (m = 1) and Equation (2.737a) .735) ( l .x 2 ) P 7 4xPJ' + i ( i + l)Pj = 0 A new function w (x) is defined by w(x) = ( l .732) One would be tempted to introduce a series solution. The general solution of Legendre's equation can be written as y = A Pj.736) (2.x 2 ) ^ .x 2 ) ^ .x 2 )" 1 ] y = 0 dx dx z (2.734) (2.715)] and Q^ [Equation (2.731)] are identical.132 ADVANCED MATHEMATICS Note that the recurrence formulae for both P^ [Equation (2.2 x P j +Z (Z + l)Yz Differentiating with respect to x yields =0 (2. however.2x ^ + [X ( i + 1) . If m is zero. B has to be zero. (x) + B Q^.m 2 /(l .733) reduces to the standard Legendre equation [Equation (2.x 2 )] y = 0 dx z dx where Z and m are integers.2x ^ + [Z (Z + l).£ . (x) where A and B are constants and Z is a nonnegative integer.(1 . The associated Legendre equation can be written as (1 .733) (2. The additional term m 2 y / (1 . The Legendre polynomials Pg (x) satisfy the equation ( l . The function Q^ (x) is singular at  x  = 1 and if we require the solution y to be finite at  x  = 1.x 2 ) P j .71)].537b)]. the following procedure provides us with an ingenious way of solving the problem.x2) represents the nonsymmetric contribution.733) becomes (1 .733) is derived from Laplace's equation in spherical coordinates [see Equation (5.
737b.x 2 )~ 3/2 w'+ w (1 + 2x 2 ) (1 .734) This is Equation (2.x 2 ) w " .2 x w ' + w [ i (i + l ) . From Equation (2. P^ 1 (x) is zero. P.740) that if m > Z.734) with y = w. is of degree Z and we note from Equation (2.x 2 ) m / 2 (2.( l .SERIES SOLUTIONS AND SPECIAL FUNCTIONS Jj£ or P^ = ( l .740) Similarly the associated Legendre function of the second kind can be computed from the formula dmO ^ Q?(x) = ( l .742) . (x) is given by P^(x) = ( l . the solution of the associated Legendre equation is denoted by P^ (x). The general solution of Equation (2. B has to be zero. we obtain p]' = (l _ x 2 r 1 / 2 w' + x (1 .738a) (2.738b) 2x (1 .741) The functions Q™ (x) are singular at  x  = 1.736) yields ( l . P™ is a polynomial iff m is even.737a). By convention.x 2 ) m / 2 *• (2.x 2 )~ 3/2 w P^" = (l X 2 ) .733b) is y = AP™(x) + BQ™(x) where A and B are constants.x 2 ) " 1 ] = 0 (2.x 2 r 5 / 2 Substituting Equations (2.I / 2 w "+ (2.737b) On differentiating.x 2 r ! / 2 w ( x ) (2. we deduce that PJ(x) = (lx 2 ) 1 / 2 P](x) In the general case.x 2 ) m (2. b) in Equation (2. 38a.739) dmP dxm (2. The function P™ is defined for Z > m. If y is finite at  x  = 1.740). The polynomial P . Note that due to the term (1 . in Equation (2.
v 2 c n x n + r ] = 0 (2..745) Comparing powers of x.. eg.743) y=X n=0 cnx"+r (2.v 2 ] (2. lb)]. we can compute the even terms C2.58a.v 2 ] = 0 xr+s: c s [(s + r)(s + r . Bessel's equation occurs in the solution of Laplace's equation in cylindrical coordinates [see Equation (5. Equation (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 .v 2 ) y = 0 where v is a constant. Equation (2.743) 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 .748) .v 2 ] + c s _ 2 = 0 => c s = .746e.746a.746f) becomes c s = .747) From Equation (2. in the discussion of a problem in astronomy.f) For r = v.747). The origin is a regular singular point and we seek a solution of the form oo (2.d) x r + 1 : cx [ r ( r + l) + (r+ l ) . It can be written as x 2 y" + xy' + (x 2 . Bessel Functions Bessel functions were introduced by Bessel in 1824.b) (2.747) can now be written as C2p = ~ c 2p2 / P 2 P (P + v )l (2.l ) + (s + r ) .c s _ 2 / [ s ( s + 2v)] (2.746c. Writing s = 2p. C4. we obtain xr: co[r(rl) +rv2] = 0 => r = ±v (c o *O) => cl = 0 (2.744) Differentiating term by term and substituting the resulting expressions in Equation (2..
754b) f°° F(l) = JO e" l dt = 1 (2. (p + v)] p=0 (2. the product (1 + v) (2 + v) . we define the gamma function F(v) by r(v) = I t ^ e ^ d t .748). To give a meaning tov! when v is not an integer.750) It can be verified that Equation (2.e) = ( . we deduce that (2.753).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. ..749b.1 e~l dt JO (2..750) satisfies Equation (2.749a) (2. (p + v) can be written as (p + v)! /v! . v>0 (2. we obtain (2.c) (2.c 4 /[2 2 (3) (3 + v)] = .753) r (v + 1) = [tv e~l] o + v I t v .749d. ( p + v)] (2. r (v + 1) = I tv e~l dt Jo On integrating by parts.755a. One solution of Equation (2.l ) P c 0 / [ 2 2 p ( p ! ) ( l + v ) ( 2 + v ) .b) .c 2 / [ 2 2 ( 2 ) ( 2 + v)] = c o / [ 2 4 ( 2 ) ( l + v ) ( 2 + v)] c 6 = ..c 0 /[2 6 (2) (3) (1 + v) (2 + v) (3 + v)] Equations (2.751) If v is a positive integer.D P x 2 p + v / [ 2 2 p ( p ! ) ( l + v ) ( 2 + v).. .SERIES SOLUTIONS AND SPECIAL FUNCTIONS 135 The first few coefficients are c 2 = .743) can be written as oo y = c o X ( .754a) = vT(v) From Equation (2.749a to e) suggest that C2P (2.
2 < v < . Similarly. Similarly if . in terms of gamma functions. Problem 9b) F(l/2) = VTC Equation (2.l < v < 0 .754b. the function F(v) is defined for all negative nonintegers. .758). then v + 1 lies between .c) It follows from Equation (2. that F(v) is ±°° for all negative integers. It follows from Equation (2... and F(v + 1) has just been defined. 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 (2.1 and 0.754b) can be used to define F(v) for all values of v (* 0.758).751) can be written. We note that (see Chapter 4.l . 55b) yields T (2) = 1. we obtain the Bessel function of the first kind of order v and it is denoted by J v (x).72. . then 0 < v + l < l and T(v + l) is defined.759a.757a.758) that F(v) is defined.Ite ADVANCED MATHEMATICS Combining Equations (2.758) and 0! = 1 (2. we define F(v) as F(v) = [ r ( v + l)]/v (2. T(3) = 2 F(2) = 2 • I. we obtain r ( v + l ) = v! for v > 0.762a) = £ p=0 (l)p(x/2)2p+v/[(p!)F(p + v+l)] (2. r(4) = 3 r(3) = 3 • 2 • 1 (2.756a to e).756ae) Generalizing Equations (2.761) By choosing c 0 to be 1/[2V F(v+1)]. That is to say oo J v (x) = X p=0 oo (l)Px2p+v/[22p+v(p!)F(p + v+l)] (2.760) c (2. .)• That is to say. Equation (2.b) If .762b) . From Equation (2. for all positive v.1 . as y = o X (l)Px2p+vr(v+l)/[22p(p!)F(p + v+l)] p=0 (2.b.2 .
the two linearly independent solutions are not obtained in a straight forward manner as described earlier for nonintegral values of v.743) is y = AJ v (x) + BJ_ v (x) where A and B are constants. If the two roots differ by an integer. the other solution is J_v (x) and is written as oo J_v(x) = X (D P (x/2) 2 p .7'43) is J v (x).763) (2.72 Gamma function Thus one solution of Equation (2. J_v (x) is singular at the origin. (2. this implies that 2v (rl=v. the two roots are coincident and this case has been considered in Example 2. We consider the two cases where 2v is an odd or an even integer separately. the general solution of Equation (2.746b)]. That is to say.SERIES SOLUTIONS AND SPECIAL FUNCTIONS ZiZ ! U *\ ] I I ' r(u)l I I 2. In Example . If v is not an integer or zero. If v is zero.* • i i n FIGURE 2.43. the other solution is obtained by considering the other root of the indicial equation [Equation (2. B must be zero. For v (* 0 or an integer). \ / / ^ / I I \/TT 3* 2* I* W*\ I 3 r 'i ! * ! ! ! .v ) is an integer. If the solution is finite at the origin. r 2 = .v /[(p!)r(pv+l)] Note that whereas J v (x) has no singularity at the origin.764) We recall that if the two roots of the indicial equation are coincident or differ by an integer.
Thus.766). This method yields (2. (2. we deduce that J_ V W = ( .766) yi = J v 00 I d]L—. the other linearly independent solution can be obtained by the method of variation of parameters.762b. q is not necessarily zero [see Equation (2.Similarly. the general solution is y = AJ v (x) + BY v (x) where A and B are constants.45. Since J v (x) is known. We recall that if 2v is one. The function Y v (x) has a in(x/2) term (see Problem 19b) and is singular at the origin.138 ADVANCED MATHEMATICS 2.768) From Equation (2. we obtain J 1 / 2 ( x ) a n ^ L1/2 00. the Bessel function of the second kind Y v (x) is defined as Y v (x) = [Jv (x) cos v% . we conclude that both the numerator and the denominator on the right side of Equation (2. If y is finite at the origin. Equation (2.v + l) is ±°° for p < v . If v is not an integer. but if v is an integer only Equation (2.768) are zero. If 2v is an even integer. Writing q = p . we deduce that Y v (x) exists in the limit as v tends to an integer.484d)] and we have one solution starting with c 0 and the other starting with cj.769) is valid. 65). By applying l'Hopital's rule. we solved the case where 2v is one. J v (x) and J_v (x) are not linearly independent.765) Comparing Equations (2.763) can be written as oo L V W = X (D q+V (x/2) 2q+V /[(q + v)!(q!)] q=0 (2.769) .D V J v ( x ) If v is an integer.763) we note that the series starts from p = v since F ( p .v . By assigning to c 0 and Cj the values given earlier. v is an integer and from Equation (2. the solution of Equation (2.J_v (x)] / sin vn (2.764 or 69).743) is given by Equation (2. J x[J v (x)] 2 (2767) The solution y2 is usually not considered. B is zero. Instead. J v (x) and J_v (x) are obtained in the case where 2v is an odd integer.
73 and Yo and Yl in Figure 2. We recall that there are circumstances when it is preferable to work with exp (± ix) rather than with sin x and cos x.71 lists some properties of Bessel functions. Equally.< } ) (2. These Hankel functions H^ and H*.743) is a linear combination of H^. b). From Equations (2.771a) (2.73 Bessel functions of the first kind .770a.770a) (2.SERIES SOLUTIONS AND SPECIAL FUNCTIONS US.74. JuU) FIGURE 2. and H^ . are defined by H ^ } = J v (x) + iY v (x) H f = Jv(x)iYv(x) The functions H^ and H^ (2.743) instead of J v (x) and Y v (x). Table 2.77 lb) The functions Jo and Jj are shown in Figure 2. we obtain Jv(x) = i(Ht1) + Hf) Yv(x) = i ( H f . there are circumstances when it is preferable to choose Hankel functions (Bessel functions of the third kind) of order V as solutions of Equation (2.770b) are linearly independent and the general solution of Equation (2.
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) . TABLE 2.140 ADVANCED MATHEMATICS Y v (x) I Y 0 (x) / o —/—\i / \ \ '—\/ /^~xr • y» » ' \ — \ / iL / V^O^ x FIGURE 2.x .v J v (x)]' = .71 Properties of Bessel functions (i) (ii) (iii) (iv) [x v J v (x)]' = x v J v _!(x) [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). H^ (x) and H^ (x) have the same recurrence relations as the function J v (x). Yv (x).
58a)] and can be written as x 2 y" + X y' .773a.743.775) V^C°SX From Table 2.72. (2.772) 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. we deduce that J 1 / 2 (x) and J_ I / 2 (x) are given by JmW = {£*>* J " 2(X) = (2. n = l .SERIFS SOLUTIONS AND SPECIAL FUNCTIONS Mi The function J v (x) has an infinite number of zeros and we denote the zeros by Xn. 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. b)]. .45. using the results of Example 2.45. Equation (2.b) Example 2. n = 1. we note that they differ only in the coefficient of y.x to (2.(x 2 + v 2 ) y = 0 (2.777) Comparing Equations (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 (5.71..776a) (2. the multiplicative constant (c 0 ) is 1/[2V F ( v + 1 ) ] .777) becomes . In Example 2. 2.. That is to say J v ( ^ n ) = 0. 77). Compute J3/2(x).. By writing z = ix. we obtain J 3/2 00 = k J1/2 (x) " J 1/2 = \H^ V ir.495a.774) (2.. We recall that in the definition of J v (x). Using the properties of the gamma function.
779) Comparing Equations (2.= 0 dr 4 dr 2 (2.778) Equation (2. the two linearly independent solutions are I v (x) and I_ v (x). we deduce that I v (x) = T v J v (ix) = e . 524) d4 Y d2 Y —*A.778) is exactly Equation (2. The function I v (x).743) with z being replaced by x and the linearly independent solutions of Equation (2. Usually. If v is an integer. 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. they are Bessel functions with purely imaginary argument. Example 2.782) .762b.i 7 l v / 2 J v (ix) (2. is real. If v is not an integer.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 ).73. we define a new function K v as (2.142 2 ADVANCED MATHEMATICS z2 ^ + z ^ + (z 2 . defined by Equation (2.777) are J v (ix) and J_ v (ix). Many second order linear differential equations can be transformed to Bessel's equation (or to another standard equation) by a suitable substitution. 1963. A simplified form of the equation governing the linear stability of a Newtonian fluid flowing between two parallel walls can be written as (Rosenhead.781) The function K v (x) is linearly independent of I v (x) and the limit as v tends to an integer is defined. The two linearly independent solutions of Equation (2. The properties of all Bessel functions are given in Watson (1966).780a.v 2 ) y = 0 dz dz (2.b) Note that we have already established that the solution of Equation (2.779). That is to say.iri — ^ .777) for all values of v (including integral values) are I v (x) and K v (x). 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. p. 79). We seek a series solution as described previously for Bessel's equation. Kamke (1959) and Murphy (1960) list several such possibilities.
788a. Both quantities are complex and we wish to solve 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. c) in Equation (2.788b.784) (2.782) becomes 2 (2. we obtain n ^ + T T 1 / 2 <*v .782) can be reduced to a second order equation by writing d2Yn u = —*&• dri 2 Equation (2.782).787) (2.l T T ^ v .788c) Substituting Equations (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.b) (2.784).789) is Bessel's equation of order 1/3. The two linearly independent solutions of Equation . we write the solution in terms of Hankel functions.2 .786) yields z 2 . Since we are dealing with complex functions.783) —iTiu = 0 dTi2 We now write u = ri 1 / 2 v Differentiating and substituting into Equation (2. — + z+(z2l/9)v = 0 dz 2 dz (2.789) Equation (2.i T ^ v = 0 dT)2 dri 4 (2. Equation (2.
It follows that the fundamental (linearly independent) solutions of Equation (2.8la. 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. Usually.790b) In Section 2. Legendre polynomials P^ (x) and Bessel functions J v (^ m x) are among the eigenfunctions we have encountered.8 FOURIER SERIES (2. 2 (*0) ifi=j if i * j (2.. we can choose n vectors g j . cos nx. which are presented in more detail in Chapter 4.b) The product g j ' g .gj) = I. g 2 .790a) (2. 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 . we have seen that the SturmLiouville system generates orthogonal eigenfunctions. ( o r / g j . The orthogonal property of y n reminds us of the property of orthogonal vectors and we state some of the properties of vectors. g : \ ) is the scalar (dot or inner) product and Ij is the magnitude of the vector g j. the dimension of the space is infinite. T h a t is t o s a y .82a.144 ADVANCED MATHEMATICS (2.789) are H ^3 (z) and H 1 / 3 (z). 2.784) are »l=V/2H«[f(in)3/2] u2 = V C H ( »[(i 1 1 ) 3 / 2 ] On integrating twice. Trigonometric functions sin nx. These square integrable functions generate a space ( L 2 space) and since n can be infinity.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. These bases are orthogonal (not orthonormal) if 0 gigj = (gi. If u is any vector in the space. g n which are linearly independent as bases.6. ••• . we obtain two solutions for % 0. the dimension of the space is finite and if the dimension is n.
which is denoted by   y n   is the norm of y n . The sum converges to f (x) if f ( x ) .8la. and the integral in Equations (2.84a. b) as the inner product. with c n defined by Equation (2. That is to say ly n ll 2 = ( y n . The magnitude yn. gj) = X i=l c i(li'Ij) =c jIj2 (2.88c) We need to examine in what sense the sum (letting N —> «>) approximates f (x) in the interval [a.85a. y n ) = p(x)f(x)yn(x)dx (2.88a).89) .b) Having been inspired by these properties of vector spaces. g j ) / ( g j .88a) (f.b) We deduce that Cj = ( u . 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 . g j ) / I j 2 = ( H . y n ) = p(x)y n 2 (x)dx = i 2 (2. Forming the inner product. Usually. we find that c n is given by cn = { f > y n ) /   y n H 2 (2.b. b].J cnyn < e n=l (2. we make use of the orthogonal property of y n . 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.s6a.SERIES SOLUTIONS AND SPECIAL FUNCTIONS /£> (U. g j ) (2.88b) Ja llynll 2 = (b /a p(x)[y n (x)] 2 dx (2.
Note that in this case the requirement is that the integrals exist and the function can have a jump discontinuity in the interval.) is complete and the space they span is a Hilbert space.. the set of functions (y l5 y2. .81 lb) ( p(x)[2fy s + 2c s y s 2 ]dx Ja (2.. = 0.810a. 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 .. the sum ^ c n y n converges in the mean to n=l f (x). 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. . Using the least square criterion.146 ADVANCED MATHEMATICS for each e > 0 whenever N > N e and for all x in the interval. we require that the integral I p (x) [f (x) ~2J c n y n ] dx be a minimum. to the mean value of f(x) as x approaches x 0 from the left and from the right.88a).X c n y n ] dx = Ja n=l Ja X c n y n +(X cnyn) ]dx n=l n=l (2.81 la) On differentiating. If for rb every f (x) for which j •'* 2 £ converges m p (x) f dx is finite.8llc) Minimizing D implies that c s is given by Equation (2... The approximation in this case is in the N least square sense.b) D is a function of c n and D is a minimum if  5 . the sum ^ n=l cnvn c o n v erges to ^. we write f^= acs (2. s = l . 2. If D —> 0 as N —> °°. 2 that is to say. X c n v n n = l the mean to f (x). N If f(x) has a jump discontinuity at x 0 .[f(x 0 + ) + f (x o _)]. The series is the Fourier series and the coefficient c n is the Fourier coefficient. The function y n can be continuous though f (x) can be discontinuous. We demonstrate that this is the present case. Let Ja (b D= N 2 (h P(x)[f22f N N 2 P « [ f . (2.
812) I sinnxcosmxdx = 0. b].SERIES SOLUTIONS AND SPECIAL FUNCTIONS MZ Trigonometric Fourier Series The trigonometric functions sin x and cos x are periodic with period 2TC.814c) The coefficients a n and b n given by Equations (2.a0 + 2^ [a n cosnx + b n sinnx] 2 (2. The functions sin x and cos x form an orthogonal basis in the interval [71.114).813a) r (2.814b)] is .813b.814b) b n = JJn f(x)sinnxdx (2. b] to [n .d) I sinnxsinmxdx = I cosnxcosmxdx = n 8 n m Jn Jn where 8 n m is the Kronecker delta. For simplicity.a ) transforms the interval [a.814a) n=l . the Fourier series Jn converges in the mean to f (x). then the transformation x* = 27t[x(a + b ) / 2 ] / ( b . That is to say oo f(x) = i.c. Jn r r for all m and n (2. 7t].7t] with unit weight f 2 [ p ( x ) = l ] .814b. n]. If f (x) is defined in the interval [a.n a =1 71 I f(x)cosnxdx Jn . c) are special cases of Equation (2.aQ and not a 0 so that the formula for a n [Equation (2.88a).n (2. we assume f (x) to be defined in the interval [%. The orthogonal property of cos x and sin x can be written as (2. If f(x) is periodic and of period 2TC and IK f dx is bounded. Note that we have written j . A periodic function f(x) of period T is defined by Equation (1.
we obtain and at x = n ? a0 = 1 Jn iTt x dx = 1 [ "K =0 (2.81.148 ADVANCED MATHEMATICS applicable to a 0 .f) J " 71 e 71 xsinnxdx = IFXcosnxj" + 1 /7C cosnx dx = .815) and periodic with period 2TE. the series converges to ^ [ f ( x o + ) + f(x o _)].817) The periodic function f(x) is sketched in Figure 2.81. %) by f(x) = x.2 ]T t ^ .sinnx n= l (2. Example 2. To what value does the series converge at xn/2 From Equations (2. 7t<x<7i (2.816g.e. Determine the Fourier series for the function f (x) defined in the interval (TT.814b.b.h. 2 ( ^ 1 (2. The function is continuous at x = 7c/2 and the series converges to the function. c).sin (im/2) (2. we obtain x .814a).2 ^ n=l °° .i) 71 Substituting a n and b n in Equation (2.c) i71 an = l j xcosnxdx = l[xsinnxj n _ 1 I shmx Jn ~n JK dx = 0 (2.818a) = 2 ^ s=0 (^TlTSin[(2s+1)7C/2] (2.816a. That is to say 7X/2 = .818b) (28"18c) = 2 i(5TTT) s=0 . n2s+l ^—.816d. Recall that if f(x) has a jump discontinuity at x 0 .
820) are zero.[f (K . f(7i0) is n. the function is discontinuous and the sum converges to j .81 From Equation (2. and sinn7C is zero. 37T ZTT/ IT / IT ZTy 37r FIGURE 2.0) + f (% + 0)].SERIES SOLUTIONS AND SPECIAL FUNCTIONS 74£ We note that sin (nrc/2) is zero when n is even and if n is odd. and sin [(2s + 1 ) T C / 2 ] = ( .. f(7C + 0) is it. • Even and odd functions are defined in Chapter 1. Both sides of Equation (2. 1. f(x).l ) s . That is to say ).2.0) + f (JI + 0)] = ..[f (7t . . we deduce that Periodic function of period In XST S=0 =7/ C4 ("9 28 1) At x = 71.81.2 ] £ ^ n=l sin n7t (2.818c). .820) From Figure 2. we write n = (2s + 1 ) with s = 0.
823a) an = 2.823b) (2.1108578 0 Td( w)Tj0 Tc 0e8j898 184C 9. I f(x)cosnxdx n Jo bn = 0 Similarly. we obtain oo (2. if f (x) is odd I Jn f(x)dx = I 21 f(x)dx.52 Tmw 09 03 00. all coefficients b n are zero and Equations (2.074 Tw 119 0 T25a. If f(x) is an even function f'(0) = 0 and if f(x) is odd f(0) = 0 (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.52 .998 43649.8799 10.825a.814atoe) reduce to oo f (x) * 1 a0 + £ n=l an cos nx (2.824a) fn b n = 2. if f (x) is an odd function.2516 m.822) (2.72 262.823c) f(x) » £ n=l b n sinnx (2. Jo r7 C (2. x^ 2n + 1^ are odd functions.150 ADVANCED MATHEMATICS The functions cos x. x 2n (n is an integer) are even functions and the functions sin x. If f (x) is an even function. I n f (X) sin nx dx (2.824b) Jo Note that we have made use of the fact that fn 0.394.07 0e8j898 00129 0494ansn T12 78.b) if f(x) is even In man 9 149.
and (b) f is even.T o we obtain f (0) = 0. we deduce that T is an odd function.. dT/dx is zero.82.828) . The function f in case (b) is sketched in Figure 2.827e) 71 Jo Substituting a n in Equation (2.82.822).821). by writing f(x) = T . I x n h cos nx dx (2. we deduce that T is even. .SERIES SOLUTIONS AND SPECIAL FUNCTIONS £££ Consider a rod of length K in the interval 0 < x < 71 with temperature distribution T(x). Example 2. If the end x = 0 is insulated. From Equation (2. 23b).n = 4 n (2. Case (a) is considered in Example 2. we obtain an = 2.823a) yields a 2 _ 4 y cos(2s+l)x «. 1. it implies that at x = 0. 0<x<rc (2.81. (2.816e. From Equations (2. It is defined by f(x) = x..tj (2s+l)2 (2.827d) Jo ~. From Equation (2.827c) s = 0. f. 7t(2s+ l ) z a0 = 2 x dx = 71 (2.827b) = 2. If the temperature at x = 0 is kept at a constant temperature T o .2.827a) (2.\QQ^L] n\.826) Obtain the Fourier series of f (x) if: (a) f is odd. The function f (x) is periodic and of period 2K.
f (x) can be represented by Equations (2. At x = 0.U2. but not in the interval K < x < 0. 28) represent the same function in the interval 0 < x < K. Under the conditions stated earlier.828). the series can be integrated term by term. we can write . 27T 7T 0 7T 27T x FIGURE 2.829) If f (x) has a Fourier series. This means that t x e x oo f x I /x 0 f(x)dx = i . if f'(x) is piecewise continuous and f (—TC) = f (n).817.814a to c). the series can be differentiated term by term.82 Even function Note that both series on the right side of Equations (2. we deduce that y l— sTo(2s+l)2 = — 8 (2.830) where x 0 is arbitrary. the function is continuous and from Equation (2. /x 0 aodx+^T I n = l /x 0 (a n cosnx + b n sinnx)dx (2. that is to say. ADVANCED MATHEMATICS f(x). If f(x) has a Fourier series.
TU<X<7C (2. .\ ao + \ Z P™ K .0) + f' (x + 0)] = ]T [n(a n sinnx + b n cosnx)] n=l (2. c).834) and with period 2n.833f) is valid for both positive and negative values of n.832a.i sin nx] dx = ±. b) in Equation (2.e ~ i n x ] Substituting Equations (2.814a) yields oo (2.833d.ibn> + e ~ i n X (an + ibn)] n=l oo (2.832b) f(x) .SERIES SOLUTIONS AND SPECIAL FUNCTIONS oo 755 i[f (x .e.i n x )ib n (e i n x e.814a to c) can be written in an alternative form.l a o + l j [an(einx + e.83.I f(x)einxdx (2. Equations (2. Example 2. This function is shown in Figure 2.83. Determine the Fourier series for f(x) defined by f(x) = e x . Recall that cosnx = 1 [e i n x + e" i n x ] sinnx = JL [ e i n x .i n x )] n=l OO (2.831) where a n and b n are defined by Equations (2.832a) (2.833C) where cn = ^(anibn) = ^ I f (x) [cos nx . Calculate the sum of the series at X = 71.833a) .814b.833b)  X n=oo cneinX (2.f) Note that Equation (2.
154 ADVANCED MATHEMATICS f(x) T 0 V 3 IT x (2.835a) FIGURE 2.835b) (2.833c.835c) (2.836a) .n e x e " i n x dx cn = . U+n2) (2.i n J ^1+in) [e71 (cos nit .835d) 2TI(1+n2) = (1+in) [(_i) n sinh 7 ] C 7C(l+n 2 ) Combining Equations (2.i sin nic) .835b) J_K " 2n[ l .83 Periodic function ex The coefficients c n are given by [Equation (2.835a) (2.^ \ (2.e~n (cos nre + i sin nn)] 2n (2. 35d) yields ex „ sinhjt 71 f (Dn(l+in)ein* »r.833f)] .
^ n (cos nx .83. the function is discontinuous and the series converges in the mean.838d) = 0 We replaced n by .840a) .837b) The imaginary part can be shown to be zero in the following way. we deduce oo (1) cosn7t (2.m in the first sum on the right side of Equation (2.n sin nx) + i (n cos nx + sin nx) 11 „=— d+n2) Taking the real part. 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 (2.838b) and we made use of the fact that cos is an even function and sin is an odd function. That is to say oo « 1 [f(71 .839) .837a) « sinhTT j + 2 V (l) n (cosnxnsinnx) (2. At x = 7t. 1+n2 L[en + e~n] = sinh_7t 1+2^ —1— L n=i 1 + n _ (2.SERIES SOLUTIONS AND SPECIAL FUNCTIONS 755 sinhrc y .0) + f(71 + 0)] = Sinhjrc 1 + 2 Y 2 * L n=i From Figure 2. we obtain ex s (2 836b) sinhrc y (l) n (cosnxnsinnx) (2.838c) (2.838a) 1 m=oo °° n= l = 2^ l(~m)cos (~mx) + s i n (mx)j + 2^ (ncos nx + sin nx) (2.838b) = 7 m=l (mcos mxsin mx) + V^ (ncos nx+sin nx) n=l (2.
842a.841) is given in von Karman and Biot(1940). If the beam is supported at the ends. the boundary conditions are 2 w = (2. Assume that the load is constant and is applied on the interval Jt /3 < x < 2Z13. The derivation of Equation (2. Obtain the deflection w (x). Beam on elastic foundation under an external load .U6 ADVANCED MATHEMATICS cosh n = sinhJL l + 2 V — 1 — (2. W mini FIGURE 2. as shown in Figure 2.840b) Example 2. 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.84.b) at x = 0 and x = JL.84 .84.841) d_w _ dx 2 0 (2.
SERIES SOLUTIONS AND SPECIAL FUNCTIONS 1JZ The function p (x) can be written as / 0.b. 0 < x < X13 i/3<x<2i/3 2X /3 < x < X (2.c) = _ ?£Q.845) * Note that w automatically satisfies the boundary conditions. 0. Its Fourier series is given by oo p(x) = £ n=l tZ bn bjjSinfi^ ^ . [cos ^2L _ Cos nffl1 = iPfi.846c) w n = b n ^ /(EIn 4 7c 4 + k^ ) .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 ^ (2.844d.841) yields El ] £ 5_5_WnSinlfflx + _n = l X * J k ^ n=l WnSinmcx = ^ bnSinmcx (2. Differentiating w four times and substituting the resulting expression in Equation (2.2ZI3 (2. p(x) = pQ.844b. The function p (x) can be considered to be an odd function of period IX.846b) [ or X4 J (2.846a) ^ * n=l We deduce that n 4_4 w EI E_ZL_ + k = b (2.c) where p0 is the constant external load.843a. nTi L 3 3 J nn We seek a solution of the form oo DJL s i n DJL 2 6 (2.
848a).848b) (2.^ d ^ d a .b) Substituting Equation (2. 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 (2.850a) oo 18 S .o o L" /co (2. On substituting w n in Equation (2.844e). f.849b) in Equation (2. we obtain f(x*> ~ X n =oo ^ r ) f(^)exp(ia n ^)d^exp(ia n x*) /—L eL (2.850b) becomes f°° f ( ^ ) e i o t ( x * .846c).^ I f(^)ex P [ia n (x*^)]d^ (2. Equation (2.845).U8 ADVANCED MATHEMATICS The coefficients b n are given by Equation (2.850b) On letting L —> °°.849a.847) r L \ ~ Z ^ \ f(x*)exp(ia n x*)dx* exp(ia n x*) (2.833c. we obtain w.851a) J J « . wn can be calculated from Equation (2.I /» oo ^ oo I f (^)[cosa(x*^) + isina(x*^)]d^da (2.851c) .85lb) — oo J —oo n Jo I /  I f(^)cosa(x*E)d^ da J (2.848a) where a n = nrc/L Let A a n = (n+l)7t/Ln7C/L = 7U/L (2. 47) yields oo f ( x *) (2.
85.85 illustrates these two functions. B(a) is zero and if f(^) is odd.853a.c) f(x) = / .854a. The right side of Equations (2.[f (x 0 + ) + f (x o _)].851f. the integral converges to j .c) Jo Jo a . ( 0. A (a) is zero. da (2. I x I > 1 Figure 2. (a) f(x) = x<l (2.g) Note that sin a is an odd function so j J00 sin a d a is zero. f°° B(a)=l f©sina^d^ (2.85le) f°° ~— I [A(a) cos ax* + B (a) sin ax*J da 71 Jo where A(a)= f(£)cosa£d£.852a.SERIES SOLUTIONS AND SPECIAL FUNCTIONS * 00 159 ~ — I K I f (£) (cos ax* cos at. The quantities A (a) and B(a) exist if J f(^)d^ converges. Determine the Fourier integral of the following functions 11.b. A (a) = 2 I f (£) cos a^ d£ = 2 I cos a^d£ = 2 sin a (2. (a) f(x) is even and B(a) is zero.b.b) 10.85 la to e) is the Fourier integral of f (x ).1 . IxI> 1 0<x<l l<x<0 (2. Example 2.85Id) Jo LJ°° (2. + sin ax* sin a£) dt. If f (x ) has a jump discontinuity at a point x 0 . / (b) 1. If f(^) is even.
b.855) The function is continuous everywhere except at x = ±l. On setting x = 0.c) Jo \0.b) Jo .160 ADVANCED MATHEMATICS f (X) = 1 I 2 sin a cos a x d a (2. I sin_a d a _ n/2 (2. I sin a cos a x a d a = n/^ 0<x<l x = ±1 (2. we obtain x>l r . We deduce that In/2.858a.c o s a ) a (2.856a.85 (b) f(x) is odd and A (a) is zero. (a) Even function. (b) Odd function B(a) = 21 sina£d£ = ^ .( 1 .857) f (x) i (o) 1 I 1 I • X FIGURE 2.
86.86 la. lxl<a (2. Example 2.j K (1 ~cos a a) sin ax da (2. The appropriate boundary condition is that w vanishes at infinity.b) x> a where p 0 and a are constants. Extend the problem considered in Example 2.860ad) x=l x>1 0.859) Jo The function is continuous everywhere except at x = 0 and at x = ±1.863) We deduce that .862b) w (x) « 1 nh Differentiating w (x) four times and substituting the resulting expression together with the expression for p in Equation (2.I [A (a) cos ax + B (a) sin ax] da n (2.I [A(a) cos ax + B(a) sin ax] da n)o (2. x=0 0 < x < 1 )0 (1cos a) sin ax « d / JC/4.SERIES SOLUTIONS AND SPECIAL FUNCTIONS 767 f(x)«2. Assume that p (x) is given by Po> p(x) = 0. so 7t/2. We represent p (x) and w (x) by their Fourier integrals and write r p(x) ~ i. (2. We deduce that / 0.862a) k [C (a) cos ax + D (a) sin ax] d a (2.841) yields A t 0 0 t°° n Jo I [C(a)cos ax + D(a) sin ax] da = J.84 to an infinite beam extending from °° to oo.
91.864a 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. b) in Equation (2.38a to c) in Equation (2. Find a power series that satisfies the equation x 3 y" + (x2 + x ) y ' . Example 2. b).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.b.38a).y = 0 (2. there is no method of generating a convergent series solution. that is to say. g. we obtain (2.867) Examples of LegendreFourier and BesselFourier series are given in Chapter 5.864b) A(a) = I r r p(x)cosaxdx = 2p 0 I cos ax dx = (2p0 sin cca)/a JO r (2.866b) w(x) = ^ 71 I Jo sin aa cos ax a ( a 4 E I + k) da (2. We seek a formal series solution as given by Equation (2.862b) yields /•oo (2. If x 0 is an irregular singular point.f) Joo Combining Equations (2. Substituting Equations (2. This is illustrated in the next example.91) The origin (x = 0) is an irregular singular point.851f. 2. we can obtain a formal series solution which is a good approximation for small values of x.865d.866a. 61a.1SZ ADVANCED MATHEMATICS (a 4 EI + k)C(oc) = A (a) (a 4 EI + k)D(a) = B(a) From Equations (2.864a) (2.91) yields .c) Joa B(a) = I p(x)sinaxdx = p 0 I sin ax dx = 0 Ja r (2.866a) (2. x (x +x)/x and x (l)/x do not have a Taylor series about x = 0.e.865a.
1) can be written as y (0. .6 x 10" 4 + 24 x 10~4 + .92) Comparing powers of x. 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.s n ] —> 0.1) = 0. for all n > 0 as x—>0 (2.2 c 2 = 2c 2 xs: (sl)(s2)cs_1 + (sl)cs_1 + s c s .10" 2 + 2 x 10~3 .95) We note that the magnitude of the terms decreases as the order increases and if we require y (0.94) Can this divergent series [Equation (2.99 Cj.h) The formal series that satisfies the differential equation is oo c 1 [ x . Using the divergent series.c x 2c 2 + 2c2 + 3c 3 C3 = 0 => c 3 = . Since the asymptotic series is divergent.]= cl £ s=l ( .96).c r x r ] = 0 r=0 (2. ] (2. (s1) xs + .x 2 + 2x 3 + ( . the error may increase as n increases... we find that y (0.l) t h term..93e.l ) s + 1 234 .( s .. only a few terms are required.c s = 0 => c s = .1) . it is seen that the approximation improves as the magnitude of x decreases for a fixed value of n.96) The error in approximating a function by a convergent series decreases as the number of terms increases. From Equation (2.93b) (2.93g.. Usually.1) c s _! (2.l) t h term.. for asymptotic series. We stop at the (n .l ) s + 1 (s1)! x s (2. the first two terms are sufficient and y(0.94)] be used to calculate the values of y for small values of x ? The answer is yes.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 .d) (2.f) CJ+2C2C2 = 0 => c 2 = .93a) (2. n The divergent series s n (x) [= X r=0 crxF] is an asymptotic series expansion of a function f(x). 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 ) t h term if the value of the n* term is greater than that of the (n .93c.Cj [10"1 . The power series s n (x) represents f(x) asymptotically as x—>0 if x ~ n [ f ( x ) .1) to be accurate to two decimal places.
63a)] and the solution is e lx . A second order differential equation is in its normal form if the first derivative (dy/dx) is not present.n e i x £ n c ^ .913a) . Equation (2.99) For large values of x.164 ADVANCED MATHEMATICS Asymptotic series are widely used in the solution of differential equations and for evaluating integrals. we look for a solution as 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.98a) (2.911) (2. Solve Bessel's equation of order zero [Equation (2. we write y = u(x)v(x) Differentiating and substituting the resulting expressions in Equation (2.97) We now impose the condition 2x 2 v'+ xv = 0 The solution is v = x~ 1/2 Equation (2. To achieve this. in this case. We first transform Equation (2.92.912) Note that. we obtain oo oo u' = ie ix X c n x. In applied mathematics. On differentiating. Even if a convergent series is available.444)] for large values of x.98b) (2.910) (2.98b) becomes u" + u ( l + l/4x 2 ) = 0 (2. one often neither knows nor cares if the series converges or not [Van Dyke (1975)]. Example 2. This suggests that we seek a solution of the form oo u = e ix £ c n x" n n=0 (2. therefore we expand in reciprocal powers of x." " 1 n=0 n=l (2. it is sometimes profitable to consider an asymptotic series as shown in the next example.911) is approximately the simple harmonic equation [Equation (2.
to allow for the tabulation of Bessel functions.r .h.i ( 9 0 ^ 4 ) / 4 (2.4 i c 2 + 2 c 1 + c 3 + ic1 = 0 x~s: .i c 0 / 4 » 2 =* c 2 = .c 3 .913d) Substituting Equations (2.d) (2.914) Comparing powers of x yields x°: x"1: x~2: x~3: .0 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] (2.2) (s .l ) c s _ ! = cs_2 (s 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 + .f) .. As usual.916a) .3 / 2 ) 2 / 2 ( s .1) cs_2 + c s + 1/4 cs_2 = 0 => 2 i ( s .915g. ^n.r (r+l)c r + 1 x.c o + co = 0 Ci+cj = 0 => c 0 is arbitrary => Cj is arbitrary =» c 1 = .i c s _ 2 ( s .913a.1 ) (2.913b) = e i x ]T [ic r x. we obtain oo £ r=0 [c r x~ r 2i(r+ l)c r + 1 x~ r .2 4 0 ^ 0 2 + ^0(3 = 0 .911) and multiplying by e~ix.915e.3s + 9/4) => cs_j = .915b) (2.915c.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.i) We note that the coefficients c s are alternately real and imaginary.3 +c r x~ r + (1/4) c r x~ r ~ 2 ] = 0 (2. d) in Equation (2.£s (2.915a) (2.2 ] r=0 (2..2 +(r+ 1) (r + 2)c r+1 x.r .c s — 2i (s — 1) cs_j + (s .
j + ui_l.080 0. we need to compute for the interval 0 < x < 0.100 0. In this example.075 0. From the symmetry of the problem.511) The initial condition [Equation (8.045 0 x t 0 0.57) simplifies to u (8. TABLE 8. Problem 9a) and the solution is 4 V cos (2 s + 1) 7tt sin (2s + 1) %x u = —. The values of u from time t = 0 to t = 0.040 0. g is zero and Equation (8.jui.59a.55a) simplifies to u i.b) u =i(*i+i + *iixi2+i*ii) < 8 .58a).02 0 0.3 0.120 0. can be solved analytically (see also Chapter 5.105 0.075 0.510) determines Uj j . the values of u at all positions and subsequent times can be computed.5 are tabulated in Table 8. 2 ^ dt =0 t=0 (8.060 0.4 0.115 0.512).080 0.511).10 ) Equation (8. then from Equation (8.4 0.e) We choose h = k = 0.58b to e). 0) = 1 x (1 .jl (8.2 0.105 0. (*O12) .100 0.04 0 0.x).085 0.045 0.040 0.58d.5 only.060 0.120 0.3 0.708 ADVANCED MATHEMATICS u (x.j + l = ui+l.58d)] provides u i 0 . subject to Equations (8.2 0.030 0.> 7T s = 0 (2s+ I ) 3 From Equation (8.020 0.51.01 0 0.5 .51 Values of u at various positions and times 0 0 0 0 0 0 0 0. Equation (8.5 0.1 0.125 0.5 Equation (8.03 0 0.1 0.1 The value of r is then unity. we can deduce that at .
we have used the method of characteristics to solve hyperbolic equations.5la).511) is ui + (8.515a) = f (xj + h .h + tp .511) is an exact difference equation of the problem. Noting that the characteristics in this case have gradients ± 1. (1928) that if r is less than or equal to one.b) From Table 8.515d) We note that the right side of Equation (8. we have marked the pivotal triangular domains which determine the value of Uj . This result is due to r being unity (h = k).t j ) = ui>j + 1 (8.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 709 t = 0.513a. Thus.515b) (8. we observe that there is complete agreement between the analytical and numerical results.51.jui.h .514) l.5la to e) with g (x) = 0 can be written as u = f ( x .t ) + f (x + t) Using Equation (8. The parameter r is the Courant parameter and it has been shown by Courant et al.t j + h)f(xi + tjh) = f(xi + h + tj) + f ( x i . Section 4) of Equations (8. we observe by comparing Figures 8. This concept is discussed in Chapter 5. this method is generally more suitable for cases of discontinuities in the given data. Section 4. u=0 (8.tp + f (xj + h + tp + f (xj .51.515c) (8.f ( x i .f ( x i + tj_1) (8.t j ) + f(x i + 1 +t j ) + f(x i _ 1 t j ) + f(x i _ 1 +t j ) . In Figure 8.43 that the domain of dependence of Equation (8. Equation (8. D'Alembert's solution (Chapter 5. • We now show that Equation (8.55a) is a stable algorithm for solving Equation (8.514).58a).5.5.+ 1.t j _ 1 ) . the right side of Equation (8.jl = f ( x i + i . .f ( x i . In Chapters 5 and 6.511).15d) is exactly the left side of Equation (8.511) is exactly that of Equation (8.511) provides an exact solution of the problem is to consider the domain of dependence of the solution.tp + f (xj . This method has the advantage that discontinuities in the initial data are propagated along the characteristics. Another way of showing that Equation (8.h .51 and 5.j+uil.
517a.710 ADVANCED MATHEMATICS t j+l • • • • • • • • • • • • • • • • • i x FIGURE 8.D.E. u. y. ^— and ^— . which can be written as ll—7 dxz + a i2 a — + a22—7 = dxdy dyz f (8.518b) .516) where a.51 Pivotal points determining U j j + 1 Method of Characteristics We consider a quasilinear P. (8. we denote 3u P = 3u q = . .518a) (8._ . r J dx dy For simplicity._ _ . d x + ^ d y dx dy = 3ji d x + J j L d y dx dxdy (8.: and f are functions of x.b) d7' 37 Writing out the differentials of p and q yields dp = ^ .
^ .516) yields dp dq 9 u / dy dx\ . the two roots are real.. a.521) 11 dx dx ll dy dx dx 3x3y L Vdx/ lz\dx/ ZZJ We now define a curve F in the xyplane.52.d) Since the equation is hyperbolic.519b) Substituting Equations (8.a 1 2 r + a 22 = ° (8.520) can be written as .518c) (8.522) has two roots dy dx = a12±Va224alia22_ 2ajj = m (8.p 1 . m_ and F_ denote the root and the curve corresponding to the negative sign.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 7U_ dq = ^L d x + p.518b. b) into Equation (8. dy d2u [ (dy\ 2 (dy\ 1 n . d). we deduce a2u _ dp 3x 2 ^u 3y 2 = (8.— ^ — a 1 1 .o . 7 ^ ^ + a 2 2 T ^ ^ .ir^ + a ^ .523a. . such that the slope of the tangent at every point of F is a root of the equation a l1 (8.f ^ " .a 1 2 + a22— = f 11 dx ll dy 3x9y\ dx 1Z z z dy/ Equation (8. as shown in Figure 8.^ T " a l l 7 " .520) dp dy dq dy .518d) d2u dy dxdy dx (8.b. Section 3.516) and are discussed in Chapter 5.519a.O1 . We denote the root given by the positive sign by m + and the curve corresponding to m + by F + . F + are the characteristics of Equation (8.522) Equation (8.c. (8.^ .519a) dx dq_J2JLdx dy 9x9y dy (8.o _ __. Similarly. dy dx dy = ^ _ d x +^ d y 8x dy dy From Equations (8.
x Q ) xR = —^ r ^ R (m + m_) /««<v^ (8.525b) = mQ(xR~xQ) where the suffixes P. b). y^) and they are (YQYp) + ( m + x p . we can determine (XR. The characteristic F+ through P intersects F_ through Q at R.525a. The equation of these lines are YRYP YRYQ = m+p(xRxp) (8.521) that along a characteristic a n ^ ^ 11 dx dx or LI A. we assume PR and QR as straight lines with slopes m + and m_. C is not a characteristic.52 Method of characteristics It follows from Equation (8.524b) a n T^dp + a22 dq .525a) (8.526a) .524a) (8.f dy = 0 Let C be a curve along which initial values for u. Q. As a first approximation.712 ADVANCED MATHEMATICS ^—P ' ^ \ c FIGURE 8. and q are known. p. From Equations (8. + a 2 2 ll ^f^ = 0 dx dx (8. Let P and Q be two neighboring points on C. and R denote that the quantities are evaluated at P. and R respectively. Q.52.m . as shown in Figure 8.
NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 713 E58367( 9 .
p .i  r_ /W* / * * iZ_ 0 .0) and (0.534a.714 ADVANCED MATHEMATICS Example 8. e) and are Up = 0.58d.53.1 (8.b) (8.: + #L = m .2. Solve the problem stated in Example 8.1 P 1 . The characteristics are given by r. u Q = 0.4 (8.52. = 1 dx + 41 = m_ = .2 . p Q = 0. The values u.1.2.53 Numerical solution of the wave equation by the method of characteristics The point (x R .532a. b) and is found to be x R = 0.b.1 dx (8.045. We choose the points P and Q to be (0.526a.532c.53.533d.2 _v< . y R ) is determined from Equations (8. and q at P and Q can be computed from Equations (8. .d) T : The initial values of u are given along the xaxis which is not a characteristic.533a.105.b) This can also be seen from Figure 8.c) (8.3 Q 1 .51 by the method of characteristics.f) t .3.0) respectively as shown in Figure 8. In this problem. y R = 0. y is replaced by t.e.4 ^ x FIGURE 8. qp = qQ = 0 p p = 0.
3. 8. we note that this is exactly the value obtained by the finite difference method.075 (8. we have considered only the case where the grid points lie on the boundaries.b) (8. The values of u at other points in the xtplane can similarly be obtained.528a.536a.1 (8.537) From Table 8.61 shows an example of such a situation. Figure 8. q R = 0.535a) (8.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 7L5 Equations (8. For curved boundaries.535a.6 IRREGULAR BOUNDARIES AND HIGHER DIMENSIONS So far. y • S P Q^S __\c X FIGURE 8. we obtain p R = 0.53Ob) and is found to be u R = 0. This is usually the case where the geometry is simple.535b) The approximate value of u R is given approximately by Equation (8. b). it is not always possible to arrange for the grid points to be on the boundaries. b) reduce to PR°4qR = 0 °2~PRqR = ° On solving Equations (8.51.61 Irregular boundary .
S and T are grid points as shown in Figure 8.63a.716 ADVANCED MATHEMATICS We consider Laplace's equation with Dirichlet boundary conditions. The points P. u T .b) /o .. P. x . From Equations (8.( l + 9 1 ) u p ] (8.65a.6la. to denote that the quantities are evaluated at P.k ^ + ^ .~ dy 2 dyz 2 2 (8. the partial derivatives are evaluated at P. (8. For compactness. 3b).64a. dy 2 9y Similarly. 5b).. Expanding UR in a Taylor series about P. as usual. we use the suffixes.42)..b) uT = u (xp...67) dx2 h 2 e 1 ( i + e1) The difference form of Laplace's equation at P is . The value of u is given on the boundary and let Q and R be two points on C. .+ — —^ ox 2 ox (8. and u s are given by (8. 0<62<l (8.b) In Equations (8.62b. yp) = u p h—.62b..( l + 0 2 ) u p ] k202(l+02) (8 6 6 ) 2[uQ + 0 1 u s . h and k refer to the grid size and where 9j and 62 satisfy the inequality 0<91<l. y r k) = u p .8u h 3u us = u(xph. we deduce that 8^u dy2 Similarly.c . Let the length of PQ and PR be 0jh and 62k respectively. 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. y p + 9 2 k) = Up + 9 2 k — + 2 j + . .+ 1 ^ ox 2 dx .. we have 9^u = = 2[u R + 9 2 u T . where.b) For points next to the boundary. we have 2 2 2 u R = u (Xp. yp) = Up + 0 ^ 5. the grid is not rectangular and we cannot use Equation (8. Q.62a. UQ .. Q.b) 2 u Q = u (Xp + Bjh.61.
k { u i.j. j .l . we deduce that u i. j .2u i.j.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.69) From Equation (8.l .k.ji.l ] (8. The difference form of the quasilinear equation 3T 80.kl = ( ) h2 h2 h2 (8.j. ui. The derivation of the difference equation for three and higher dimensional problems is straightforward.q + e^up] h2el(i + el) + [uR + e 2 u T . the difference equation is complicated and it is preferable to use the finite elements method which will be described later.k + u i.8799 092163Tw 0. j . Other methods are available and are described in more advanced books.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 711 [iiQ + e ^ s .j.j + l.2u i. For Neumann conditions. j .j. however it involves a larger number of points and the direct method is not recommended.610) can be solved by the methods described earlier.j.k + u i .k = ^[ui+l.k + l.69).k+l + u i .k + u i . k .10066 Tws 11.7 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.j.j. Some of the methods described in this chapter can be used for nonlinear equations. the value of u at the point (i.k .j + l.k + u i.5 0 T0220X9ON equati9J1 0 Td(s)T . k + u i. k) is the average of the values of u at the six neighboring points that surround it.k + u il.j. 8.j.610) As in the twodimensional case. Laplace's equation in three dimensions for a cubical grid of size h 3 is u i+l.k. Equation (8. k + u i.
72a.j + l = uij+J^D(xi.71.b) (8. The equation of continuity and the equation of motion.j)(ui+lo2ui. II1) and are 3vY 3v v 21 + ^ =0 dx dy 3vY 3vY 9 vY vx . for x > 0 for x > 0 (8. j). as shown in Figure 8. v x = 0 . Comparing Equations (8.j) (8"7"2b) h Equations (8. t.I1.tj.74) Experience has shown that this expectation is justified. assuming no pressure gradient and making the usual boundary layer approximations. 34b). 1963) . The appropriate boundary conditions are y = 0.76d) v x —> v^ (free stream velocity). In fact.76a.^ + v — ^ = v — f (8.72b) is linear. 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. Equation (8.73) We have evaluated D (x.j + u il. b) are the simplest form of the explicit difference for Equation (8.718 ADVANCED MATHEMATICS or u i.75b) dx y dy dy2 where v (=fl/p) is the kinematic viscosity.75a) /n (8. u) at the grid point (i.t j .76c) (8.u i j ) < 1 h z (8. Example 8. v y ). vy = V (a constant).u)% at dx2 (8.ui.71. are obtained form Equations (A. v x = Blasius solution (Rosenhead.71) by ^ = D(x. The velocity components along the x and y axes are respectively (vx. Using the initial and boundary conditions.72b) generates values of u at all points for subsequent times.t. Note that Equation (8.71). we expect the numerical scheme to be stable if 4D(x i . we have approximated Equation (8. y —> oo ? x =0 .
v v ^ A dx dx y 3v/ (8. The components v x and vy can be functions of x and y or functions of x and \/. y).77a. 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. Equation (8.b) Combining Equations (8. 7a. defined by 3\/ d\if (8.79) can be written as 3vY 3vv 3vY 3JL = V L .NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 779 y <• l 0 X FIGURE 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.71 Flow over a solid plate and a porous plate We introduce a stream function \/ (x.710) . y) to (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. Using the chain rule. \/). b.75a.77b).
£)d^Vx (8. c)] are transformed to . we obtain f(x) = . y) is given by \f = r JO v x (x.f Vd^ + g(0) Jo = . Imposing the conditions given by Equations (8. Integrating Equations (8.71 la.713a) (8.75b).713b) ¥ =  Jo where f(x) and g(y) are arbitrary functions. b.76a.77a.£)d£ + f(x) v (5.76a. \/) coordinates.79 to lie). we can choose g (0) to be zero and \/ (x.y)d$ + g(y) (8. b).715) The conditions at y = 0 and y —> °° correspond to \/ = Vx and \/ —> °° respectively and the boundary conditions [Equations (8. ^x dx = VA(V d\/ \ x ^A 3\/j (8712> The boundary conditions also have to be transformed to (x.V x + g(0) (8.714a) (8. b). vx is a function of x and y on the left side and a function of x and \/ on the right side.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.b) 7? = W'£) Substituting Equations (8.714b) Since \/ is defined to the extent of an arbitrary constant. we obtain respectively fy V= JO v x (x.79 to lie) into Equation (8. we obtain (8Jllc) Note that in Equations (8.
18) *L = /TTIT A The boundary conditions become \l/2 /6 = _ pn . Further details can be found in Ames (1969). We now introduce the following dimensionless variables 1/2 nJ.U j j [ui+1J2uiij+ui_1>j] h (8.£&) Equation (8. r = 0 . vx = 0 v x —> vM (8. .716b) Equation (8.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 727 V =Vx.720) In discretizing Equation (8. u = 0 P = V" ^ T / (8. . refers to the value of u {/&v x\).™ (8 . A —> » .j + l = U i j + ^ V l . it is found that if k VlUj • .719d) (8.719a to c) imply that we have irregular (triangular) boundaries and we need to make the modifications proposed in Section 6. 7 . *.718). y _ > oo.719e) u is given by the Blasius solution Equation (8. Note that Equations (8.718) can be approximated as u i.721) we have computational stability. we have divided the region of interest into rectangles each of size (h x k). we can compute the values of u j .716a) (8. As expected.c) (8. ^ h2 < 1 2 (8. The quantity Uj.712) can now be written as • " = 'fe)2 (3.719a.719a to 20). u = 1.b.76d) remains unchanged. From Equations (8.
We introduce the basic concepts of the finite element method by considering onedimensional problems. dx 2 subject to u (0) = A. 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. 8 . we can always transform any arbitrary interval into an interval of [0. The battle between finite differences and finite elements is over. The reason for this choice will become apparent after we have introduced the method. It must be pointed out that one does not usually solve onedimensional problems by this method.72b) but which give results closer to the solution of Equation (8.71).72b) can be improved by replacing D (Xj.82a. we have only one independent variable and the equation we need to solve is an ordinary differential equation. t:.81) u (1) = B (8. Several formulae have been proposed which are more complicated than Equation (8. OneDimensional Problems In a onedimensional problem. We consider the differential equation .2 ^ L = _g(x). . the method of finite differences is preferable because finite difference equations are easier to set up.722 ADVANCED MATHEMATICS The accuracy of Equation (8. Implicit formulae which are unconditionally stable have also been derived. These formulae can be found in more advanced books on numerical solutions of P. The onedimensional case is used to ease the way to the understanding of two and higher dimensional problems. More recently. For regular boundaries and simple equations.E.8 FINITE ELEMENTS The finite element method was developed to study the stresses in complex discrete structures. 0<x<l (8.b) By an appropriate scaling. the finite element method is the obvious choice.D. 1]. it has been widely used to obtain approximate solutions to continuum problems. It was pointed out earlier that if the boundary is irregular. Uj :) by an average value of D over the points considered.
84) i = / o te) 2 . we reformulate the problem into a variational problem.g r ^ ^ ( i u H]° We approximate u by m Simplifying Equation (8. we replace the derivative by a finite difference.85). They are usually chosen such that the boundary conditions are satisfied.86).X i= l aiNiW (88"7) where a. or other functions whose properties are known. In Chapter 9. They can be polynomials. i l 9y dx 9y' Consider the integral =0 (8. In Chapters 9 and 10. we obtain an expression of the form I = F(a1)a2 g (8.s u ] d x Applying the condition given by Equation (8. we obtain Equation (8.81) is the function that extremizes the integral I defined by Equation (8. u . In the finite element method.88) . Instead of looking for a function that satisfies the differential equation.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 721 Variational method In the finite difference method. On substituting Equation (8. it is shown that the function y (x) that yields the extreme values of the integral I. we give examples of such variational formulations. y')dx Ja is (8. y.85). are constants and the Nj(x) are the shape (trial) functions. we look for a function that extremizes an integral.81).87) into Equation (8.84) yields ( °.5) (886) srHnr) . This means that the solution of Equation (8. defined by I = I f (x. trigonometric functions.83) ^1 _ A.
2.810) where R is the residual. we obtain an approximate solution for Equation (8.87).= 0.81). .81). In the method of finite elements.. and on substituting them into Equation (8. We illustrate this method in Example 8. This method is known as the RayleighRitz method (see Example 9.81 la. we subdivide the interval [0. We now describe a method which provides a weak solution to Equation (8. then u is an exact solution of Equation (8. we obtain the coefficients a.81). we can expect that a better accuracy can be obtained for the same number of terms in Equation (8. . 1] into a number of subintervals and each subinterval is called an element.89). The function u is a weak solution if the integral of R with respect to a weight function w (x) is zero.b) Jo In the Galerkin method.812a) .1] into [0. we choose the weight function to be the approximate solution (w = u). Suppose we subdivide the interval [0.. we subdivide the interval [0.81) and we write ~ + g(x) = R dx 2 (8.87).87). If R is identically zero. we approximate u by Equation (8.87). 1] into subintervals (elements) and on each element.81 la) becomes /"xo I N : [ajN^' + a 2 N 2 ' + g] dx = 0 JO (8.89) From Equation (8.m (8. In each element. We approximate the solution u in the interval 0 < x < l by u(x). That is to say C Jo Td2~ 1 w(x) —^ + g(x) dx = Ldx2 J (l w(x)R(x)dx = 0 (8. x 0 ) and (x 0 . On the first interval.81. Equation (8. Since each element is shorter than the whole interval. and the extreme values of I are given by dF ^ .112). The function u generally does not satisfy Equation (8. i = 1. In the finite element method.724 ADVANCED MATHEMATICS It can now be considered as a function of a. Galerkin method Not all problems can be formulated as extremum problems.87) provide sufficient accuracy. u is approximated by Equation (8. 1] and assume that two terms in Equation (8.
we obtain another set of equations.15a ) Kjj = I NjNjdx. Solve the equation . . We can generalize Equations (8.813b) r where jxo r . subject to the boundary and continuity conditions.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.. This implies that we can choose linear functions as shape functions.813a) (8. and well conditioned. Gj = gNjdx (8. b) by parts.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS Z25 I N j ^ N ' i + a2N2' +g] dx = 0 JO Integrating Equations (8. m) and we obtain iKijlNiNjJxk where aj = G J i ( 8 .814d. b) for any interval x^ < x < x^+i and for any number of shape functions Nj (i = 1. Once a.c) gNjdx G2 = gN2dx (8. is sparse.814a.812b) aiKlla2K12 alK12a2K22 + Gl = ° = ° (8.813a..812a. 1]. Note that. we have avoided having to evaluate the second derivative of the shape functions. on integrating Equations (8. the approximate solution u is given by Equation (8. for the interval (x 0 . b) by parts yields a l i N l N l J o +a2LNiN2J0 (8.87).815a) for all the elements. have been obtained.8 .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 . The shape functions Nj are usually chosen such that K can be evaluated easily (analytically or numerically).e) 0 Jo Similarly.81.815b. Example 8. .b.812a. The coefficients aj can be determined by assembling Equation (8.
818) yields I = I j + ^ + Ig (8. (1/3. u satisfies the boundary and the continuity conditions at the nodal points (Problem 14a shows how these functions are derived). 2/3). and E 3 and the values of u at the nodal points Xj by Uj. We subdivide the interval [0.b) . E2. Equation (8.816a) is equivalent to finding the extremum of I = I l f d u ) 2 _ x u dx (8.816a) (8.816b.x ) + (3x2) (8. u(l) = 1 (8. and (2/3. 1] into three elements of equal length [0. and E 3 .= 0 On carrying out the differentiation.817a to c) into Equation (8.820a. 1/3). E2.c) by the method of finite elements. We approximate the function u by a piecewise linear function.817a) (8. In E 1 .l ) u « 3 u 2 ( l .818) Substituting Equations (8. u is approximated respectively by u « 3ujX u » U1(23x) + u 2 ( 3 x .817a to c).817b) (8.> 2 £^ The extremum of I is given by ^= j ^ . We solve the problem by the variational method.819a) •(¥£Mi<^££Mi<>. 1].726 2 ADVANCED MATHEMATICS — + x = 0 dx 2 subject to the conditions u(0) = 0. we obtain <•«»> (8. We denote the elements by Ej.817c) Note that in Equations (8.
NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 72Z 6uj3u 2 = 1/9 3uj + 6u2 = 29/9 The solution is u2 = 62/162 = 0. y) has the following properties Nj(x. yk) = 0 (8.821b) (8. we obtain . we note that the solution we have obtained by the finite element method is exact! The exact solution is cubic.d) Combining Equations (8. yj) = 0 .822a.824a) (8. yi) = 1 . TwoDimensional Problems In twodimensional problems. if the shape functions satisfy exactly the homogeneous differential equation. it is zero. it is one and at nodes j and k.81. the two coincide. We choose the shape function Nj such that at node i. Nj (x. We propose linear shape functions. This is shown in Figure 8.825). Nj (Xj. 23). Consider a typical element E with nodes i. This is not generally true.822b) (8. the approximate solution is linear. j .822a) (8.823) By comparing Equations (8. and k as shown in Figure 8. y) = aj + biX + q y Ni (xj.383 u 2 = 59/81 = 0. b.82 la) (8.824b. we subdivide the domain into elements and we consider triangular elements. That is.81. Triangular elements allow for an easier fit to nonrectangular domains. and at the nodal points. We can also consider rectangular or other geometrical shapes. the nodal values are the exact values. N{ (xk.c.728 The exact solution is u = *(7x2) o (8. Strang and Fix (1973) have observed that in general.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.
c) D = l 1 yi y?k = 2 (area of element E) = 2 A (8.826a. we can approximate u as u = £ Uj Nj (8.f) y o FIGURE 8 81 x (8. we generalize the variational and the Galerkin methods to twodimensional problems.728 ADVANCED MATHEMATICS a{ = (x j y k x k yj)/D. Next.827) i where Uj is the value of u at node i. ci = (x k x j )/D (8. we define a shape function Nj given by Equations (8.b.y k ) / D .e.g ( x .827) Domain subdivided into triangular elements.826d. then we set Nj to be zero.. (8.824a to e) and if i does not belong to E. and k For each element E.828a) (8.828a) We consider Poisson's equation which can be written as + = . With this definition of N. j . y) dx2 dy2 u = 0 on the boundary (8. 1 xi x xk bj = ty . E is a typical element with nodes i.828b) .
832) (8.830b) can be written in a compact form as I =XKijuiujXuiGi or where (8.831a) (8.827) into 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 (9.831c) (8.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 722 The function u that satisfies Equation (8.833a) .831d) (8.g (u1Nl + ..829) yields .1113)] (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 (8.829) Substituting Equation (8. 32) yields ^KjjUjGi j = 0 (8.83 lb) I =uTKuuTG Gj = I I gNj dxdy The extreme values of I are given by ^~ = 0 Combining Equations (8.83la..
+u ^)^ + hV" + " +Um ^f)^" sNi J dxdy = 0 (8.838a) .4 + g(x. aNj\ /3N m 9N3N_ dNA XT 3N m \8Nj / 3N.833b.836a) (( /3N. By combining Equations (8. 24a). This is true for all symmetric problems.836b) can be written as XKJJUJGJ = j o (8. dxdy = 0 (88'35) The function u is chosen to be given by Equation (8. y). . y ) p 4 + ^ . For nonsymmetric problems.834) On integrating by parts and assuming that w vanishes on the boundary. the variational method may not be applicable but the Galerkin method can still be used.w s (t\l 9N. the matrix K is no longer symmetric. we define a weight function w (x. 37)]. 3Nj 3N.730 ADVANCED MATHEMATICS or Ku . y). Kjj is given by Kij = J / [ b i b j + c i c j] d x dy (8.83lc. Note that in this case both the variational and the Galerkin methods yield the same discretized equations [Equations (8.G = 0 (8. such that w(x. 5N m \aN: XT1J J .831c. d).834) becomes It Idw du dw du .+ U m (^r^r + ^V). Equation (8.g N i d x d y =0 (8.835) can be written as JJ [( u ^ + ". we find that for each element E. JJ (ara^ + 3757. Equation (8.833b) To obtain a weak solution u (x.837) where Kjj and Gj are defined by Equations (8. y) dxdy = 0 (8.827) and using Galerkin's method (setting w = Nj in turn). A or )) u i ( ^ ^ + ^ V ) + . Equation (8. In this case.836b) In the usual compact form.
Equation (8.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 731 = [bjbj + c j Cj] (area of the element) = Afbjbj + CjCj] Using Equation (8.(xj + Xj + x k ) y « y = j (Yi + yj + y k ) The approximate value of Gj is thus G .824a). 42c). ^ is a (m * m) matrix and the global G / > is a vector with m components. then the global matrix K.84ia. Thus using Equations (8. x and y can be approximated as the average values at the nodes. •>. Solve the equation .842a) (8. x and y are approximated as x « x = i.838c) = I I g ^ j + bjx + cjy] dxdy (8. If the domain has m nodes. That is to say.842b).842c) Equation (8. K is a (3 x 3) matrix for each element. a sparse matrix K. and G is a vector with three components.831d) can be written as Gi (8.842b) (8. By numbering the nodes appropriately. y We note that since each triangular element has three 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).826a to c) into Equation (8.842c) is obtained by substituting Equations (8.g (c ) [aj + bjx + cjy ] A «^g(c)A (8.838b) (8. If the triangular element is small enough. we can obtain .b) (8.b) .82.g ( c ) [aj + bjx + Cjy ] I I dx dy (8. Example 8.839) which simplifies the computation of Gj. we can compute Kjj and gj for each element E. and G .840a. Similarly.838c. We then have to assemble them to obtain the global K. .
4 ( 2 . 38c.c) (8.2) y ' J2. = .844) From the symmetry of the problem.826a to c.5. 0 ) .e. 1 ) .2) 3/ 5 (2. 1 ) . and 6(2. a2 = 0 .0) X FIGURE 8.0) / (1.845d. The area of each triangular element is 0. Node 1 is the origin of the coordinate system.1 .82.f) (8.i) . We subdivide this reduced domain in four triangles and label the elements and nodes as shown in Figure 8. a3 = 0 . Ej: aj = 1 . b3 = 0 .712 ADVANCED MATHEMATICS ^Jl+iH.l c3=l (8. 2 ( 1 . 2<y<2 (8.0). W = 1 . q =0 c2 = . b2=l.845a. 3 ( 1 . 5 ( 2 . we may limit our calculations to one octant only.82 Division of a domain in triangular elements We now use Equations (8.0) A (2. 42c) to calculate K and G for each element starting with element Ej.843) subject to the condition u = 0 on the boundary (8. 0 ) .b.1) E3 / . The coordinates of the nodes are 1(0. 3x 2 dy2 2<x<2.h.845g. (0.
b. K 24 c2 = 0 c4 = .854a.846a to f) (8.c) (8. K 22 b2 = .h.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 .c) Gl = G2 = G3 = ^ E^ a2 = 2 .l .847a.855) 0 0 0 I 0 . a4 = .b.c) G 9 = G 4 = G.f) (8.i) K 45 = ^> =. K23=l. we have E3: K22 = i .l . = 1 J 6 Similarly. K56 = .^ ' K 25 44=l' = " £• K 55 =^ (8. K25 = .b.c) G 9 = G.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 733. K55 = l (8. K35 = .848g.l c5 = 0 = ^> K (8. K66 = l (8.l . = G. b4=l. K 35 = .848d.852a. K33 = l .851atof) (8.e. b 5 = 1 .849a to f) (8.848a. K36 = 0 . a5 = 0 .853atof) (8.b. K55 = 1 . for E 3 and E4. K ll=2' K 12 = ~i' K13=0> K 22=1' K 23=^' K 33 =^ (8.l .l .1 J [u 6 j _ 1_ .85Oa. = L ^ J J 6 E4: K33 = l .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.
856) that "uji u2 = r i 2 . x into four (3 * 3) matrices. they are transferred to K/ ^ and G / y.857a. . . Equation (8. and u 3 . u 5 . u and G ^ <. in Equation (8.t) = 1 Choose h = 0. The vectors u and G .855) is singular and this implies that the uj cannot be determined. .1 _i 2 2 o 1 T 1 rii 6 1 = [ i~ 4 11 (8. u(l. We choose to consider the upper half of K. Thus K and G for each element need not be stored.857 a) is nonsingular (its inverse exists) and U j . The only unknowns are U p u 2 . u 2 . t > 0 . at ax 2 u(x.0) = x 2 . that is to say.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 and G are calculated for each element. Strang and Fix (1973) have discussed the computational aspects of finite element methods. K( x and G /gx are being assembled as the calculation for each element proceeds. u 2 . and u 3 which can be determined by partitioning the matrix K. since we require only three equations to solve for U j . Use the explicit method to solve the initialboundary value problem defined by ^ = ^ 4 .t) = 0 .856) Using the boundary conditions (u 4 = u 5 = u 6 = 0). 0<x<l u(0. and u 6 are the values of u on the boundary and are known.855).1 and k = 0. and u 3 can be determined. s are each divided into two vectors each with three components as shown by the dotted lines in Equation (8. The components u 4 . PROBLEMS 1 a. we deduce from Equation (8.b) ui\ I ° 1 2 J L i l [%_ Note that the matrix in Equation (8.05 so as to ensure the validity of the method.234 ADVANCED MATHEMATICS Note that the matrix K.
0) = 0 . ^ 5x =0 x=0 by the method of CrankNicolson. t) = u(5. 0<t<I u(0. 3z dx c (0. Solve the nonlinear parabolic partial differential equation 9U 3 U . 0 < t < I Solve the following initialboundary problem ^ = ^JL.= D .(u + cos u ) .721. Use the finite difference method to solve the following elliptic equation d u + 9 u y 3u x 9u u = 0. ^ 5x =0 x=l Obtain c for D = 10 and calculate the average concentration (c) defined by Equation (5. By introducing the following dimensionless variables c = (ccf)/(cjcf).t) = I . 0 <x<i. x n e rx 1 — = — . t) = I . Use the CrankNicolson method to solve the problem examined in Example 5. 5a. 0) = I . at ax2 u(x. . z) = 0 . t>o u(l. 0<x<5. D = J90B/(Lvz) c (x.^  .736a) by numerical integration. 22 a to c) become ..y on the boundary of the square. x = x/L. 0<x<l. 0<y<l ax 2 ay2 ax ay u = x .NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 735 2b.5. at 9x 2 subject to the following conditions u(x.72. Choose h = k = 0. 0 < x < 5 . 0) = I . 3a. 4b. z = z/L show that Equations (5.
j) and (i.O) = x . j) by its mean value at (i.R. Use the finite difference method to solve the hyperbolic equation ^ = % 0<x<l." i + i j _ i .t)=l. ! . 0<y<8 u = 0 on the boundary (i) by direct elimination and (ii) by the S. methods. and G = 1. Is it possible to choose k > h? 10b. u(x.^ i .l . an implicit formula for Equation (8. t>0 at2 subject to ax2 u(O.2.t) = O .l j .5la) is ui_1.j + 1 . Solve Poisson's equation ^4 dx2 + ^ i =2. Choose h = k = 2. The Laplacian is 32u 3r2  1 du r 3r 1 32u r 2 dQ2 9a. dy2 0<x<6.O. Obtain a central difference approximation to the biharmonic equation :J4 }4 . ^ a t = 0 t=0 Choose h = k = 0. O 7b.383.2 [ 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.736 ADVANCED MATHEMATICS 6a.4 (h/ck) 2 u i .4 di + 2 dx4 8b. j + 2 [1 + (h/ck) 2 ] U J J . u(l. Show that by approximating (9 u / 3 x ) at (i. j + 1 ) .
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. =1 t=o The function u (x.+ a 0 (x)u = r(x) dx 2 u (a) = a . t>0 at2 ax 2 u(x. I fdi) .5. 0).6. u (b) = p is given by the extremum of the integral I which is defined by /•b r I = 13 a. 0. Show that the solution to the following boundary value problem 2 d ^.01. 0) and (0.t) = u ( l .t ) . k = 1/8. Ja.NUMERICAL SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS 737 Use the implicit formula to solve ^ = % 0<x<l. lib.02 12a. Answer: 0. Choose h = 1/4 . =x(lx) t=0 Use the method of characteristics to find u at the intersection of the characteristics through (0. 0).4.01.O) = x ( l . (0. 0. . t ) = t ( l .0) = 0. ^ dt u(0.x ) . t) satisfies the following conditions 92u  92u 92u = 1 3x2 dxdt dt2 ^dt u(x.
17a. y) on a square plate of size 2m x 2m has reached a steady state and satisfies the equation ^ . Solve the boundary value problem defined in Problem 13a by the method of finite elements.2 ) / 2 14a. 15a.l ) / 4 .6. 0<y<2 The sides x = 0 and x = 2 are kept at 0°C and 100°C respectively. Compare the values of u (1 /3)) and u (2/3) with those obtained in Problem 13a. Divide the domain into five elements and choose linear shape functions. Note that u is approximated by a linear function pi (x) and. b. Deduce Equations (8. Impose the boundary conditions and obtain the required results. from Section 7. Answer: 5 x ( x . Verify that 1 .2). Solve the nonlinear system 16b. + ^ = 0 .1) and (1. d!u +0 . x ^ ] . one deduces that pj (x) can be written as Pi 00 = Ui + Cuj+jUjXxXjJ/h in the interval [XJ. Answer: 50°C . Compare the approximate solutions with the exact solution.738 ADVANCED MATHEMATICS Use a quadratic in x as the approximating function. + u2 = lf o<x<rc/2 u(7i/2) = 0 by the method of finite elements. The temperature u (x. x(x 2 + x .sin x is a solution of the boundary value problem. Use the method of finite elements to find u at the points (1.817a. The sides y = 0 and y = 2 are insulated (3u/3y = 0). c). (1. 8y2 9x2 0<x<2.0). Compare the analytical solution with the numerical one. 5 [(djM 2 Hdx' dx2 u(0) = 1 . Consider three elements and approximate u by a piecewise linear function.
Suppose we wish to determine the shortest plane curve joining two points A and B in the xyplane as shown in Figure 9. y (a)) and (b.1 INTRODUCTION Calculus of variations was introduced by Johann Bernouilli two centuries ago and is now widely used in optimization. where its value is 0.11 Curves joining two points A and B . and for computing approximate solutions of differential equations. y (b)) respectively. For example.11. a J *s/Uy b x 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 . Calculus of variations is an extension of the above concept. control theory. If the coordinates of A and B are (a. the function f (x) = x 2 has a minimum at x = 0.CHAPTER 9 CALCULUS OF VARIATIONS 9. (9'» y(b) y(a) ^ ^ . f (x) is greater than zero everywhere else. In differential calculus we learned to determine the extreme values (maxima or minima) of a function.
if it exists. In engineering we encounter similar variational problems. where ( ' ) denotes differentiation with respect to x and Ax = x .11). y 0 and y. Expanding the right side of Equation (9.f (x0) = Af = f' (x 0 ) Ax + 1 f "(x 0 ) (Ax)2 + . where a functional can loosely be defined as a rule which assigns a real number to a function. Hence Q represents an extremum. we can obtain the corresponding value of s and we have to determine the curve y(x) which yields the minimum value of s.23) (9.12) is similar to Equation (9. irrespective of the sign of Ax is f'(x o ) = O (9. Prior to considering variational calculus.22a.. such that an integral is an extremum.. Figure 9. rB and r c may be functions of temperature and concentration. 9. tj) is given by P 1= JO r B dt (9. It can be seen from Equation (9. 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. That is to say.22) that the condition for Af to preserve its sign. Figure 9. The reaction rates rA.x 0 . In kinetics we can consider the reactions » > A — B — C.12) Equation (9.b) . we wish to find the extreme values of a functional. In variational calculus we need to determine a function. irrespective of the sign of Ax.11) is referred to as a functional.21) retains the same sign for all small values of Ax.11 shows two such curves.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).21) about x 0 yields f (x) . We need to determine the function r B that will maximize the integral.2 FUNCTION OF ONE VARIABLE In Chapter 1.740 ADVANCED MATHEMATICS For each curve y (x).. The production I of B in the time interval (0. we review the theory of extreme values in differential calculus. 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. In this example. 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. The integral on the right side of Equation (9.
. that is. Thus x o = 7 t / 2 is maximum (sin %I2 = 1) and Xj = 3rc/2 is a minimum (sin 37t/2 = 1). Q represents a minimum since Af is always positive.23). 5) Thus the extreme values of sin x are at f' (x) = 0. so that x n = (2n + 1) 7C/2 . At these points f "(x n ) = sin x n = sin (2n + 1) 7t/2 = sin %I2 cos tin = (.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. we have f'(x) = cos x and f "(x) = sin x (9. Example 9.CALCULUS OF VARIATIONS 741 y  V \ »«/!*y ayl/jax i FIGURE 9. . On differentiating.21.. In Figure 9.26) These extreme values are maxima when n is even and minima when n is odd.21. 1.22) 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.24.c. Find the extreme values of f (x) = sin x.27a.b.. (9.l ) n + 1 (9. cos x = 0.d) n = 0.2. It follows from Equation (9.
33) The function f will be a minimum if the quadratic expression ^ ^ 5—^— Axj AXJ is i=l j=l positive definite. 1=1 J = l 1 J In Equation (9. i = 1. Expanding the right side of Equation (9.. xn then f is said to have an extreme value in the neighborhood of Xj.. we deduce the condition for f to have an extremum at x^ = Xj is p _ = 0 .. x2.32). . The conditions for this are D1=^>0 0X1 (9. n. ... (9. x 2 .. . n.. i = 1. n n n ^2rdxidxj (9..31).x 2 + Ax2.742 ADVANCED MATHEMATICS 9. xn if Af defined as Af = f (Xj +Ax 1 ... . all the derivatives are to be evaluated at Xj = Xj.31) Af=S^Ax1 + l i S ^ A X i A x j 1=1 l + ...35) .34) A D2= 1 32f Bx^xj _^f_ \ _^f 3x2 2 >0 (9. . .32). we obtain (9.. xn) preserves its sign for all Axi..32. x 2 .. . xn + Ax n )f(x 1 . From Equation (9....2. for all i= 1.3 FUNCTION OF SEVERAL VARIABLES If f is a function of several variables x^.
The conditions for f to have a maximum at x.31 la.31.39atoe) The extreme values of f are at 2x 1 + x 2 = 0 . 3x2 ^ = 2 .d) 2x2 + x 1 = 0 (9. 3xj ^=2x2 3x2 + Xl. we obtain i!=2x1+x2.= 1 dxxdx2 (9.c. Example 9.38) — =2.. both Dj and D 2 are positive and f has a minimum at the origin.b.310a.c) and so on with the sign of the determinant alternating.37a. D2 > 0 . x 2 = x2 = 0 (9. (9.b) So the extreme value of f is at the origin. At the origin. .CALCULUS OF VARIATIONS 743 3^f 32f 92f a2f 3x^x2 Dn= . Find the extreme values of f (x 1> x 2 ) = x 2 + x 2 + x 1 x 2 On differentiating. that is Dj < 0 .b.^ . is that the quadratic expression must be negative definite. a2^ ax2 a2f ax2axn >0 (9.36) a2f axjaxn a2! ax2 All the derivatives are to be evaluated at Xj = Xj. 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. (9. dx\ .. D3 < 0 . The solutions are x^x^O.
316a.315a.e) The extreme values of f are given by — L = 0 => 2 X l x 2 + 4xf = 0 OXi (9.312) ^4=2. 3X2 ^L.b) L 2 = 0 => 2x2 + xf = 0 (9.c) (9.3lSatoe) .b. Example 9. 3x ^f=2Xl 3x^x2 (9. 9X2 = 2x2 + 12x2 (9.33. 9X2 ^ = 0 . Find the extreme values of (9. a minimum or neither. 9x 2 ^ = 0 .= l 3x^x2 (9.b) f(x1.313a.32. ^=x1.d) The solutions are x1=0.244 ADVANCED MATHEMATICS Example 9.314c. Both Dj = D 2 = 0 at the origin.x2) = x1x2 On differentiating.313d. Find the extreme values of f(x 1 . we obtain (9. 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. x2 = 0 (9.x 2 ) = x^ + xfx 2 + xj The partial derivatives of f are = 2xjx 2 + 4x].314a.b) Thus f has its extreme value at the origin. dxj dx2 = 2x 2 + x 2 .317) ^=x 3xj 2 . and so we need to investigate further to determine if the origin is a maximum.
. x2... . .. . .x n ) = 0 We now consider the function L. x 2 . . sometimes called a Lagrangian.45) .. X) = f (x1? x 2 .. The extremum of L is then given by ^=0. instead of one. . x n ) j=l j = l .. . . Xj and x 2 are not completely independent.. In practice.. It is preferable to use the method of Lagrange multipliers. .. in the neighborhood of the origin.41) — =0 (9.43a. i = l . xn) + ]T... . x 2 . .43a) (9. x 2 . The origin is a saddle point.. and is given by L (xj. 9.42) (9.4 CONSTRAINED EXTREMA AND LAGRANGE MULTIPLIERS We have assumed earlier that Xj. xn) + Xg (xj. xn are all independent and can take any arbitrary value.. For example. x 2 . There might be a relationship between the Xj. Xj gj (x lf x 2 . then x^ is a function of x2. it is a maximum along one line and a minimum along another line. x 2 .n (9. . 2 . and the constraints are given by g j ( x 1 . In theory it is possible to eliminate all dependent variables and retain only the independent variables. . .. m (9.. . x n ) where X is the Lagrange multiplier.. . . this can be very cumbersome and the process of elimination can be very time consuming.b)]. x 2 . .. Suppose we wish to find the extremum of f (xj.. If we have m constraints. then Xj cannot be negative. The Lagrangian becomes m L = f (xj. we can solve for Xj = Xj and X. 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.x n ) = 0 .44) (9.. 2 . xn) subject to the condition g ( X l .. ... if Xj is the cost of a product. This might not necessarily be the case.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..43b) dX From the (n + 1) equations [Equations (9. Thus if Xj represents the length of a rod and x 2 is the temperature.. x n .
. ..n (9. .49) (9.m (9.x 2 ) = xJ + x  .. lla.b. Example 9. we deduce that x1 = x 2 Substituting Equation (9.46a) — =0.413a.48) In this example.746 ADVANCED MATHEMATICS The extremum of L is then given by ^ =0. dxl ^— =xl+ 2Xx2 .:.46a.l = 0 The Lagrangian and its derivative are L = x 1 x 2 + X(x^ + x ^ .[l (9.412) (9. we can solve for Xj and A.2. we are required to find the maximum or minimum of the rectangle inscribed in a circle of unit radius. x 2 .46b) From the (n + m) equations. dx2 (9.l ) =— = x 2 + 2?ix1 ..41 lc).c) From Equations (9.410) — =Xj + x 2 . b). 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. we obtain X!=^r. V2 f = ..b) .41. . Find the extremum of f (x 1 .2.x 2 ) = x 1 x 2 subject to (9. j = l.47) xf + x\ = 1 (9. i = 1. . .l = 0 dX (9. The equation of constraint is g(x 1 .41 la. . xn) are identically equal to zero.. Since the g: (xj.412) in Equation (9.
11. On expanding the function f [x.55) From Equation (9. We can consider a neighboring curve yj = yQ + er (x) as shown in Figure 9. x2) cannot be the origin. y o + er(x). fb en' — dx = [ETI^1  erijU^. we obtain ib AI = I eri ^. (Xj. x 2 ) can be any point and thus f = XjX2 has a saddle point at the origin. In Example 9. we deduce that AI preserves its sign for all arbitrary functions erj (x) if .54) Since y (a) and y (b) are fixed.55).a L ^y dx \dy' ]dx + O(82) . yo(x).y' + eTi(x)]dxl f[x. y^ + er'(x)] in powers of e.CALCULUS OF VARIATIONS 747 Note that.53). (9. 9. Let the function y 0 (x) represent an extreme value for I.^ + O(e2) dx (9.77)] and substituting into Equation (9. r\ (a) = r\ (b) = 0. (Xj.52). The points a and b are fixed and we also assume y (a) and y (b) to be fixed. Substituting Equation (9.±IK . y^(x)]dx Ja Ja (9.51) to have an extreme value.5 EULERLAGRANGE EQUATIONS We now deduce the conditions for the functional (integral) I given by 1= I" f(x. in this example. the point (x1? x2) has to be on the circle of unit radius with the centre located at the origin. we obtain .33.dx (9. y')dx (9.y o + ETi(x).52) where e is a small quantity and r\ is an arbitrary continuous function of x. /a y. according to the Taylor expansion [Equation (1. The increment AI is given by ib rb AI = I f[x.53) Integrating the second term by parts yields /•b .54) into Equation (9.+ eif . In the present problem.b AI= [eT] K^.
we obtain yj = y 0 • We now denote y 0 by y.b) .58c) = + T 1 '_§Ll d x 3y.5.56) is the EulerLagrange equation and gives the condition for I to have an extremum.57b) S = /a d t [ f (x> y ° + eT1' y ° + e T 1 ' ) ] dx (95'8a) (9.58c) becomes rb il e=0 = ^^1 + ^iLldx Ja [ ^y a y ' T)1 a y' (9.57a) =0 e=0 (9.56) Equation (9.yo + eTi')dx (9. on setting e = 0.9) On again integrating I dx by parts.748.5.510a.8b) = fb[iL^i+iL?yi]dx J a [By! 9e \nJ*L ia [ a yi dyj 3e (9. Equation (9. ADVANCED MATHEMATICS £± dy dx ^ dy' t = 0 (9. y o + eri. we obtain A dl = f J*L _ A f i L \ l dx = 0 de e=o Ja [a y dx (ay1 jj (9. The extremum of the integral (h Ja is given by 41 de 1= f(x.56) by considering I to be a function of e. We recall that y^ = y 0 + £T) and. We can also deduce Equation (9.
b).. a2f Oy') 2 n . not all the arguments x.62) From Equation (9. f is a function of x and y only and it follows % =0 (9.511) is then simplified. y is not an admissible solution and the functional I does not have an extreme value.fte 1^ (9. y = y (b) (9.61. y' (x)] and compute Thus to determine the extremum of the functional I. we obtain the relation y = y (x).512a. y (x).512a.511)]. The variation ~ d£ is also known as the Gateaux variation. a2f y . y. y' of f are present and Equation (9.511) reduces to  = 0 (9. subject to the boundary conditions at x = a.512a. at x = b. Expanding Equation (9. we need to solve a secondorder differential equation [Equation (9. we have af By a2f y . If y does not satisfy Equations (9. We now consider these special cases.CALCULUS OF VARIATIONS 749 Thus the condition for ^ d£ e=o = 0 for all r\ (x) is also given by Equation (9. Example 9. b)].62).56).61) ay Equation (9.511) ^. 9. dx =0 dxdy' dy' a y 8y' Hint: Note that can be written as g [x. Find the extreme values of . The solution y = y (x) has to satisfy the boundary conditions [Equations (9.6 SPECIAL CASES Explicitly Function f Does Not Depend on y' In this case. b) In practice. y = y (a).56).
66) [y = (l/2x)] does not satisfy Equation (9.e) = .c.l (3x) jdx = .64a) and.68) f2 I = I2 = 1 { ^ ( 3 . y(l)= 1/2. The extremum of I is then given by Equation (9.b. 1/4). I has no extremum.2 .1 i n 2 = 0.d) . In this case . The equation of such a line is y = l(3x) Substituting this y into I.b.67a.c) (9. 1/2) to (2. y(2)=l y(2) = l/4 (9.67e. in this case.64a) (9. it is seen that ^ is a minimum. 9d).64b) and y = (l/2x) gives the extreme value of I.69a.1719 Comparing Equations (9.63) The boundary conditions are (a) (b) y(l) = l.1733 For the purpose of comparison. (9. However.66) (9. it satisfies Equation (9.62).750 ADVANCED MATHEMATICS I" 1=1 2 (xy z y)dx (9. suppose we have a straight line joining the points (1.2 1 = 1= [XJ Jl I 4x2 Udx = 2x) Ji L dx 4x (9.x ) 2 .1 1 =0. which yields 2xyl=0 (9. we obtain (9.64b) In this problem f = xy2y which is not an explicit function of y1.67d.65) The solution of Equation (9.
Example 9. Find the extremum of i= C H ^ L In this example. we can solve for y' and y' can be further integrated to obtain y. b)].64a). we obtain y' = hi (9.614) Vlc 2 x 2 Integrating yields y = I C]X dx = . Equation (9.CALCULUS OF VARIATIONS 757 This means that there is no function (curve) y (x) that extremizes I and satisfies Equation (9.64b).611) dy' From Equation (9. f = dx (9. Of all the functions that satisfy Equation (9. the function y = (l/2x) minimizes I.V1 .b) ^ x Solving for y'.512a.12) y.615a. Since we have integrated twice. here we always have a solution. Equation (9.62. we introduce two arbitrary constants which are determined by the end conditions [Equations (9. Unlike the previous case.6.J .611) yields \ V 1 + ^—^— is not an explicit function of (v')2 1 xVl+(y')2 (9.c ? x 2 + c 2 (9.611). Cl .613a. Function f Does Not Depend on y Explicitly In this case.56) simplifies to f(!M = o dx \dy J (96io) which on integrating yields = constant = Cj (9.b) I /T^2? where c 2 is a constant.
Function f Does Not Depend on x Explicitly Since f is not an explicit function of x (9.511) simplifies to df By .b) I"0 Equation (9. the two integration constants are determined from the boundary conditions.[ f . 0) and the point (1. .c 2 ) 2 = ( l . As usual. 1). a2f .617a. The constants c1 and c 2 are determined by the boundary conditions.622) From Equation (9.616).621) (9.. If the curve passes through the origin (0.622).y —T7 = ° 8ydy 0y')2 (61) 99 .c 2 x 2 ) (9.620) Multiplying Equation (9.^ . a2f _ y T ^ T .618) (9. we obtain (yl)2 + x2=l which is the equation of unit circle with centre at (0.616) Cj=c2=l Substituting the values of Cj and c 2 into Equation (9.616) and this represents a family of circles with centres along the yaxis.752 ADVANCED MATHEMATICS We may rewrite Equation (9.615b) as c2 ( y .n£*ns (9. we can solve for y' and this can be integrated so as to obtain the required y as a function of x. we find that the resulting expression is an exact differential which may be written as f.y l ^ r l = o L dy The solution of the above equation is f .y ' .620) by y1. 1).= constant = Cj dy' (9. we obtain from Equation (9.
Integrating y' yields (9. 8y' y T " 2 ( i + (/>2)'/2 V2g (9 .628) On simplifying the above expression. we can then solve for y'. In this example.6.624) where g is the gravity.623) This example is known as the brachistochrone problem.625a. It was first proposed by Johann Bernouilli in 1696. so that the time taken is a minimum.63.CALCULUS OF VARIATIONS 7J1 Example 9. and y the vertical axis. It is usually accepted that this problem marks the beginning of calculus of variations.630) . If we take one of the points to be the origin of the Cartesian coordinate system. b) is.11. 27) Substituting f and dy' and absorbing V2g into the constant c 1 .b) Jo Since f = v ^ Jo V 7 (9. ^ f Vl + (y')2 dx (9.26) ^(l+(y')2)1/2yl/2 V2g il. Then the time t needed for the particle to fall from the origin (0.(y1)2 0 + (y')2)~m y~m = Cl (9. while falling under gravity. Equation (9. 6 . 624) t = fa ds = J L f3 Vl+(y') 2 dx (9. 0) to a point (a. x the horizontal axis. Find the extremum of .629) I A / ^— dy = I dx ] J V lc2y (9. we have to determine the path that a particle will take. then the velocity v at height y is computed through the conservation of energy (potential + kinetic = constant) to be v = /2gy (9. using Equations (9. we obtain I =ciy1/2(l+(y')2)1/2 Squaring both sides of the above expression.622) becomes (i + (y') 2 ) 1 / 2 y'm .
cos 29] Cl (9.633) c2 Z where c 2 is a constant.630) can now be written as M© .y) + y ' q ( x .J .154 ADVANCED MATHEMATICS To evaluate the integral on the left side of the above equation.634) y = J . y = b. Therefore.l sin 29] (9.y sin 29J = x + c 2 (9.637) (9. Function f Is a Linear Function of y ' Let us write f as f = p(x.[e .b) ^ C l The constant Cj is to be determined from the condition x = a.[l .636) ^T = q (x. we use the substitution y = J . . y ) Then (9.b.639a. Thus the curve of quickest descent is a cycloid. (0.631) I / A / ClY dy = 2. The parametric equations of the curve as given above is that of a cycloid.sinG cos9 d9 = ±.632a.sin 2 0 (9. The parametric equation of the curve is X= X f e .635a. 9 = 0. 0). and c 2 = 0.b) V1cfy Equation (9.56) becomes (9.I ^isS.s in 2 9 = .638) l+y'l^y^fM/ (9. y) dy Equation (9. The curve has to pass through the origin.1 sin 20] c 2 J cosG c2 L 2 J (9.
such that >  a = ± grad < (9. If Equation (9.640) is identically satisfied.640) is identically satisfied.b.642a.CALCULUS OF VARIATIONS 755 The condition for I to be an extremum is dp da (9. Find the extremum of .644) (9.64. Thus an extremum of I will exist only if y satisfies the end conditions [Equations (9. then the expression given by Equation (9. then curl a = 0 So a is a conservative force and there exists a potential <).643) (9. This is similar to the case where f is not an explicit function of y'. q) to be the components of a vector a in a twodimensional space.636) is an exact differential.(££) If Equation (9.640) Equation (9.c) «. that the work done by a conservative force is independent of the path taken.640) gives y as a function of x and there is no integration constant. then (9. The above result is equivalent to the statement in mechanics. Example 9.512a.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.<j)a Ja Ja ax Thus I is independent of the path taken and all curves yield the same value of I. and there exists a function < such that ] > dd> (9. y') dx = I ^ dx = ())b .b)].641) The integral becomes I =  f (x. y. If we take (p.
If the second end condition is modified to (1. . So in this case there is no extremum.2). y. q = x. f1 1= (y + xy')dx= fl [ydx + xdy] = fl d (xy) = [xy]^ = y (1) (9. then we have to solve the second order differential Equation (9. 0) and (1.640) is identically satisfied. there is no extremum. p = y.b) Thus the condition is 2y = 2x or xy = 0 (9.646) Jo The requirement for an extremum of I is given by Equation (9. so Equation (9.649) In this example. y ^n\ Thus we find the extremum of . if f is an explicit function of all three variables x.65. Find the extremum of r1 1=1 (y + xy')d* .b) If the end conditions are represented by the points (0. . but the second one is not. then the curve y = x will give the extremum of I. . y ' " . 1). In general. .647a. then the first end condition is satisfied.511) and the two end conditions will determine the two arbitrary constants. such as y ".648a. 9. and y'.650atoe) Jo • Jo Jo Therefore.640) with p = y2 and q = x2 (9. Example 9.256 ADVANCED MATHEMATICS 1=1 (y2 + x 2 y')dx (9. In this case.7 EXTENSION TO HIGHER DERIVATIVES We extend the theory to f being an explicit function of higher derivatives of y.o (9. I does not depend on which curve y (x) is taken.
81). We obtain Equations (9.81) L dy Ja Using Equation (9.51)] with the end points x = a and x = b fixed.55. . are given at x = a and x = b.. Thus the conditions for I to be an extremum are now given by Equations (9. Instead we need to deduce two new conditions. but the values of y and its derivatives at the end points not specified. 6). but we can now no longer impose the boundary conditions r (a) = r\ (b) = 0.81) simplifies to ^ =0 (9.CALCULUS OF VARIATIONS 757 I = /a f(x.82) V x=b . If the value of y is fixed at one of the end points (say at x = a) and is free at the other end. 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. as required. since y (a) and y (b) are not fixed.81). y'. The alternating signs are a result of several integrations by parts of the derivatives of T. are assumed to be satisfied. . The ycoordinate of the curve will have its ends at any point along the ordinates x = a and x = b as shown in Figure 9.81. . then by considering another function y = y o + eii(x) (9.5.56). y " \ . The existence and continuity of y and its derivatives. we can show that the condition for I to be an extremum is (9.y. We impose a new transversality condition Tl27 =0 (9. y " . y ( n ) ) d x (9.56. y".73).. In deducing Equation (9. we obtain Equations (9. we have assumed that y. 9.53.71) If y = y o ( x ) is the function that will make I an extremum. We proceed as before by considering two curves y 0 and y 0 + £T.73) dy dx U y ' dxMay" dx3 (ay111 "" dxn \ 3y (n) j This equation is essentially a generalization of Equation (9.72) where e is small and r\ (x) is an arbitrary function of x and proceeding as in Section 9. then the condition given by Equation (9. y1. 4).8 TRANSVERSALITY (MOVING BOUNDARY) CONDITIONS We reconsider the problem of finding the extremum of I [see Equation (9.
b.81.c) => cj = 0 (9. y = constant = c 2 The solution that satisfies the prescribed condition [y (0) = 1] is (9.84) At x = 1.752. ADVANCED MATHEMATICS y ^J a FIGURE 9.611) implies Cl 4y'[(y')2+l] = (9.b) . Equation (9. Find the extremum of 1=1 [(y') 2 +l] 2 dx (9. the transversality condition imposes i l =0 9y' It follows that y'= 0 .85a. Since f = [(y') +1] is not an explicit function of y.83) Jo given y(O)=l but y(l) is not prescribed.81 L_^ b x Variable values of y and its derivatives at constant x Example 9.86a.
2 a + 2] [2a2] = 0 (9.81 la. we obtain automatically the value of y(l) which renders I an extremum. We suppose y (1) = a.814b) The solution is a = 1 and I = 1 which is the solution given by Equation (9.89) The function y and the integral I are given by y=(al)x+l (9.b) We regard a as a variable and determine the values of a that will make I a minimum.b) y(l) = a (9.86).814a) (9. . Thus y = k 1 x + k0 where kj and kQ are arbitrary.812) I = [a2 .87) I = 70 dx = 1 (9. which is a cubic equation in y'.810a. Imposing the end conditions y(0)=l. That is ^ = 0 or 2 [ a 2 .813a.88) Let us determine I when the values of y are specified at both ends. Using the transversality condition.84). The condition for I to be an extremum is still given by Equation (9.b) (9.2 a + 2] I dx = [ a 2 .2 a + 2] Jo (9.CALCULUS OF VARIATIONS 759 y=1 Thus the extremum of I is given by (9. At least one of the roots must be real and since c1 is an arbitrary constant. we find that ko=l. k1 = ( a .l ) (9. we can assume that y' is a constant.
94b). y')dx = constant = C (9. y o + EJTIJ+ e 2 ri 2 . y o + e 1 r ) j + e 2 r i 2 ) d x (9.9lb) is satisfied.92) In this case. subject to the constraint 1= f(x.9la.94a. However.260 ADVANCED MATHEMATICS 9. y.93) J* = I g(x.b) We assume that y o (x) is the function that extremizes I. Suppose we have to find the extremum of I as given by Equation (9. y')dx (9. then they are known as isoperimetric constraints which we shall consider next. This problem is equivalent to the case of finding the extremum of a function of two . Consider a neighboring curve to y 0 given by y = y o (x) + e 1 r 1 (x) + e 2 r 2 (x) (9. The functions Tj^x) and T2(x) are arbitrary functions. We need to find the extremum of I Equation (9.9 CONSTRAINTS So far we have been concerned with finding the extremum of (b Ja with y satisfying certain end conditions.51) J = I g (x. = C (9. If the supplementary conditions are of an integral type. algebraic equations or differential equations. We define I* and J* as /•b I =1 /a f(x. y. y'Q+elr\[+e2ri'2)dx Ja. 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.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. we have introduced two parameters £l and £ 2 . y o + e 1 r  1 + 8 2 r i 2 .51). in many problems y has to satisfy additional conditions and these conditions may take the form of integral relations.
96) — =C (9.99a.4.98b) as a result of integrating I rij ia 3yo •ni(a) = Tli(b) = 0 Thus the extremum of I is given by (9.910a. y o +8 1 ri 1 +e 2 Ti 2 .CALCULUS OF VARIATIONS 761 variables subject to a constraint as discussed in Section 9.b) = 1 L(x.b) ^ k _ A 3L =O=d(f 9y dx 3y 3y + Xg)±\lT(f dx [ 3y + Xg)\ (9. yo+£ 1 rij+8 2 Ti2)dx Ja where A.95a. We therefore define a function H* as H* = I * + A J * = I (f+X. we obtain JT.=0 (i = l.2) (9.98a) = fTli[^^(^]jdx Ja [3y 0 dx \dy'o) fb dx by parts and using the boundary conditions (9. The extremum of I* is given ^ aei (9.97) Following the development leading to Equation (9. The method of Lagrange multipliers can thus be used. is a constant.58).95c) =0 e.g)dx Ja (9.b) .= Ui — + T\\— dx (9.
the extremum is given by My + x V l + t y f l . Using Equation (9.91.911) Example 9. Given a curve of fixed length s joining the points (0.762 ADVANCED MATHEMATICS and gdx = C (9.913) dy dx [ dy Taking the partial derivatives.912) Jo and the length s is given by Equation (9. we obtain .91 The area A = I ydx i * Area (shaded) enclosed by a curve of length s (9. 0) and (1. 0) as shown in Figure (9. remembering that y and y' are to be treated as independent variables.910).91). find the extremum of the shaded area enclosed by the curve and the xaxis.M y + W l + t y 1 ) 2 ) ] ^ (9.^ f . y o FIGURE 9.11).
918) (9. The end conditions implies c 2 + c 2 = X2 (9.916) ^(xc.915).921) .914) dx L V l + ( y ' ) 2 _ This can be integrated to yield y' = ^ 1 > (9. we finally obtain ( y . On squaring both sides of Equation (9.zdz On further substituting z = ?.922) (9. we obtain m \ l d J ( ".cose we obtain y = + I X cos6 d6 = + X sin 6 + c 2 (9. \\4 Y =0 (9.920a. b) and returning to the original variables.920a.b) On squaring both sides of Equations (9.CALCULUS OF VARIATIONS 763.917) y =± I .C l ) 2 = X2 Thus the curve is an arc of a circle.C ' ) 2 (9.)2 To integrate the above expression.915) Vl+(y')2 X where Cj is an arbitrary constant.c 2 ) 2 + ( x .919) (9. we use the substitution z = (xc 1 ) Then the integral becomes (9.
926) Thus the required curve is the semicircle with centre (1/2. instead of a constant.921) with respect to x and substituting y' into Equation (9.92.j .11).1/2 and the equation of the curve is given by y2+(xl/2)2= 1 (9. • If the constraint is of algebraic type or in the form of a differential equation.and c 2 = ± V A .11) yields s = XI dx — jo y?i 2 (xi/2) 2 This integral is of a standard form and its value is (9. Find the extremum of . We can now express X in terms of (9.927a) (9.siiT1( J) [since sin"1 (x) = sin" 1 (x)]. we impose the isoperimetric constraint given by Equation (9. (9. y ) (9.927b) Example 9. The constraints can be of the form f (x.925b) s and the required curve is a circle with centre (1/2. Differentiating Equation (9. with solutions Cj = }. ±V X2.1/4) ) and radius X. X . This result can also be obtained from geometrical considerations. If s = K/2.925a) I \2X> X J o = 2A.C l ) 2 = l 2 .924) s^fsin1^^]1 (9. y.923) To obtain X. then we have to choose the Lagrange multiplier as a function of x. y') = constant or y1 = g ( x .764 ADVANCED MATHEMATICS cf + ( l .0) and radius 1/2.
102a. yi' yi» ..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... ..928) Jo We reduce the integrand to the usual form in terms of y'. We write yi = y" Equation (9.10 SEVERAL DEPENDENT VARIABLES We generalize to the case of several dependent variables.101) We assume the end points are fixed and the conditions are y 1 (a) = a 1 . .929) 1 = 1 [l+(yi)2]dx JO (9.c) . We first extend the theory of calculus of variations to the case of more than one dependent variables. y n (a) = an y n (b) = b n (9. y n . yi. y2.103a..932) (9. . .CALCULUS OF VARIATIONS 265 1 = 1 [l+(y") 2 ]dx (9. We need to determine the n functions y 1. The function f depends on n dependent variables y J .b.c) (9. y 2 ..• y i i ) d x (9. 9.b. y 2 (a) = a 2 . . y n .. and then we shall be able to complete the above example. y 2 (b) = b 2 .. . y 1 (b) = b 1 . .928) becomes (9. y2. . yn that extremize the functional I given by i= }a f(x. .930) Subject to the constraint yiy' = o The Lagrangian L now is given by L = (l+(yi)2) + ?i(y1y') = 0 (9..
10 " 8 ) Ja [ ^i hi The conditions for I to be an extremum are obtained by integrating the second term in Equation (9. 11) become .n (9..105a.1010. yOn are the functions that extremize I and consider the functions y i = yoi + e i r l i ( x ) ' i=1 n (9.109) Example 9. . ..106) The extremum of I is then given by ft* aT aei = (9.b) !*(%) = /a f (x. Complete Example 9. Equations (9.)dx (9. satisfying the conditions Tii(a) = Tii(b) = 0 The functional I* (£j) is given by (9.766 ADVANCED MATHEMATICS We proceed as in the case of one variable. dy{ dx \ dy[ j i=l.59).104) where the r^ are arbitrary functions.io7) £i=0 Equation (9.932).108) by parts and the result is iL__l(_*U0. yon+EnV yii+eiTll. . y.1011) =0 3Yl dx [ dy[ J Using Equation (9.107) can be written. .. again following the development leading to Equation (9.. . as Tli — + ^ T L dx = 0 ( 9 ..101.1010) (9.92. The extremum of I is now given by *_A(ik\ = 8y dx 1 dy' I ^±(^) 0 (9...2. n +e n Ti.. . yoi+ejTi!. Suppose y 01 .. y o2 ..
21) f=2yi. Find the extremum of lTt/2 (9.23) Thus the extremum of I are given by . subject to the end conditions given by Equation (9.b.)2 + (yi) 2 + 2 y i y 2 ]dx (9. we compute  f = 2y2 .1022.1013) to obtain y = a 3 x 3 + a 2 x 2 + a1x + a 0 where a 0 .b.1017) I = Jo [(y.d) Thus.1018) subject to the following conditions yi (0) = y 2 (0) = 0 . is y =x Example 9.102.1012) implies that X is a constant and we can integrate Equation (9.c.109). ^T=2y2 (9.c. If the end conditions were y(0) = 0 .1013) Equation (9.1014) then the constants a0 to a3 will be given by a o = a2 = a3 = O and al = \ (9.1019a. a2. a3 are constants. the function that will extremize I [Equation (9. y'(0)=l.  L = 2yJ (9.d) Via Equation (9. y'(l)=l (9.1012) (9.b.1020.1016a.930)].d) (9.c. Yl (Ti/2) = y2(7c/2) = 1 (9. Remember that y ^ y ' . y(l)=l. a^.CALCULUS OF VARIATIONS 767 ^(?l) = 0 ?i^(2y) = 0 (9.1015a.1015a to d).
1029a. the variational approach has certain advantages [Lanczos (1966)].1025). One of the possible variational principles we may start with is Hamilton's principle.1030a. From Equation (9. we obtain y2"y2 = 0 The solution of Equation (9.^ (2yi) = 0 2yi . Derive Lagrange's equations.1026) (9.1019a to d) yields ai (9. a4 = s .103. one usually starts with Newton's laws of motion.d) Thus yj and y 2 are given by Yi = y? = 1 l s i u h K.1027) (9.y 2 = 0 (9.a2 sin x + a3 cosh x + a4 sinh x Imposing the boundary conditions given by Equations (9. In elementary mechanics.768 ADVANCED MATHEMATICS 2 y 2 .b) Example 9. = 0 yt .1026) is y 2 = &i cos x + a2 sin x + a3 cosh x + a4 sinh x where a1 to a4 are arbitrary constants.^ (2y 2 ) = 0 =» =» y22y.b) Eliminating yj from this system. 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.1031a.1025a. 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 .a1 cos x . In advanced mechanics.b. we deduce that y2 = .b) (9.n (9.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. n h 1 7 t / 2 (9.1O24a.c.1028) = a2 = a3 = 0 .
The dot denotes differentiation with respect to time.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(. p The generalized coordinates of the dynamical system are q1.. The kinetic energy of the system is K = i. Example 9. L is then a function of q1 and q . q2.1032) in an extremum.er) (9. qn and the generalized *1 *2 *j (9. be the origin. Since the string is inextensible.. the potential energy of particle Q is . . 0) as shown.CALCULUS OF VARIATIONS J69 1 =I L d t (9. Q will move up or down with speed r.. 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 9.104. L is the Lagrangian and is given by L = Kq> where K is the kinetic energy and c is the generalized potential energy.1033) *n velocities are q . ... Let O. q .1034) The n equations given by Equation (9.1037) (9.1034) are known as Lagrange's equations.2..3 mg(/8 . The components of the velocity of P are (r...n (9.1035) . where k is the length of string.101. q . 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. the position of the hole.1036) (9. The polar coordinates of P are (r. Thus the condition for I to be an extremum is ^LA/^M dt\9q'j Bq1 o. determine the ensuing motion of P and Q. . Taking the level of the table to be the zerolevel of potential energy. r 0).r). i=l. .
so C^afs^ Combining Equations (9.1038a.b) ^k±(^k)=0 ae dt ae = f(mr2e) = 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).720 ADVANCED MATHEMATICS FIGURE 9.3 m g ..f (4mr) = 0 dt (9. we find that 9 is given by (9.1040) .b) Integrating Equation (9.1O39a.1040.1039b) yields r 2 6 = constant = C^ Initially r = a.1041) (9.101 Particles P and Q connected by a string of constant length. a 0 = V8ag.
although the radial velocity is zero initially (at time t = 0). we obtain 4 r'=8a_g_3g r3 (9. From the initial conditions r = a.d) Equation (9.1046) where C 2 is a constant. r* is negative (2g) [see Equation (9.2=7ag_2*jg_3gr (9. That is to say.1042) Substituting the value of 0 into Equation (9. the particle Q will initially more up and r will increase.1045) which can be integrated to yield 2 r 2 •9 = _ ^ .1043)] and .1048).1043) The term r can be written as " =i <h = i < f >$ ' ^ • 2t (f2) (9. C 2 is calculated to be C 2 = 7ag From Equations (9. r" will be zero when r3 = 8a 3 /3. this does not imply that the radial acceleration at t = 0 is zero. From Equation (9. r* is positive.1043).3 g r + C2 r2 4a p (9. At r = 2a. we find that r r.1046.1047) is given by (9104g) 2 r2 2 We can deduce from Equation (9. that initially since r = a. 47).b. when r = 2a and Q will come momentarily to rest.c.1043) may now be written as 2dr^2) = ^3^~ 3 g (9.1044a. r = 0 . Consequently.& .1038b). we deduce that r = 0.CALCULUS OF VARIATIONS 721 6 = ^ (9.
That is to say. 2). 9.111. we allow the cycle to repeat when r is again equal to a. Thus Q will oscillate between the heights ( i .x 2 ). Geometrically it means that we are considering a surface instead of a curve.' Figure 9. and R is a region in the XjX2plane bounded by a curve C as shown in 1 O A. u X2 (Xj. let us consider a function u (xj.x 2 )] dxj dx2 (9. u X i (x!. Q FIGURE 9.2a). For simplicity..772 ADVANCED MATHEMATICS so Q will fall. x2) is given on C. R 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 .x 2 ). We now have to find the extremum of I = l  f [xv\2. = =— (i = 1.111) where u x . x 2 ) which is a function of two independent variables Xj and x 2 .11 SEVERAL INDEPENDENT VARIABLES The integrand might be a function of two or more independent variables. The radial position of P lies therefore between a and 2a. because of the lack of friction.a) and (Z .1048) and a tangential velocity — ~ . The value of u (xj.
112) Suppose u o (x 1 .116) now becomes fI(ui) = fr^. i=l. Note that df 9f 9u. (9.116) Since e is independent of Xj and x 2 . x 2 ) o n C where g is a given function. we can introduce the operator — inside the integration.+ riXi 4J^—+ ri ^ . 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) (9.113) into Equation (9. x 2 ) is the function that renders I an extremum. .x 2 ) = u 0 ( x 1 .119) we finally obtain . _ (9.CALCULUS OF VARIATIONS 773 u(x 1 ) x 2 ) = g ( x 1 .x 2 ) = 0 on C Then the extremum of I is given by A l ( u o + £Tl) ae (9.118) Making use of the fact that ri xi 5 — = 5— Tj^ 3uY dX: duY 3f d df ^5—5— dx.x2) u 1 (x 1 . with^J=r) /ft 1 1 ^ (9.114) =0 e=0 (9.duv d I df .115) Substituting Equation (9.2 . ^ df — ==: T^ + X ^ 1 1=1 d"i x ^—i IXj au. 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 .] d X l d x 2 de ^e^o // [ x 2 ^ux R (9.117) Equation (9. We consider another function (a neighboring surface) given by ul(xl.111).
as before. u x j dxj. Plateau. .1114) 3f _J_ du 8x. x m . . . u X2 . I f (x1? x 2 ..112.. we deduce that the condition for I to be an extremum is 9u 3x.. xi dXj Example 9.114) implies that the integral in Equation (9.. x 2 ) bounded by a curve C as shown in Figure 9. . a Belgian physicist (18011883). From Equation (9.111..T l ^ ldxldx2 8x2 3uX2 j ] (9. 1 i! 3u Y xl d_idf_ 9 x 9 19u Y z _J_ m _a^ xm = 0 (911. Green's theorem may be stated as follows Hi^lk)*1***=/<pdXl R C + d2 QX) ( ". . u Xj . uY = ^—.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 + ^ ... dx m R The condition for I to be an extremum is (9. did many experiments with soap films to determine minimal surfaces. .. by Equation (9. We are interested in converting the double integral to a line integral. ..1110).. u.) 9 n The double integral becomes //[sr("^K^)]^/[^^"tH (9""12) R C The boundary condition given by Equation (9..114). Then we consider the functional I = j I . x 2 .1110) We can write the terms inside the second brace of the double integral as a line integral using Green's theorem.1112) is zero.1113) We can extend to m independent variables x^. because the condition on T] is given on the curve C. Find the minimum surface area of the surface x 3 = u (xj. x m .. This problem is known as Plateau's problem.15) 3 x m 3u Y \ X2 where. 1 3uv X] 3x7 2 9uv I x2 / (9.
r 0 ) = r (x 1 ? x 2 + A x 2 ) .1117a. Let the vector positions of OP0. x 2 ) « ^=.c) P 0 P 2 = ( r 2 . P 2 and P 3 .d) The vectors P 0 P^ and P 0 P 2 are given by dr P Q P J = (£j . x2) .CALCULUS OF VARIATIONS 775 We shall first derive the formula for the surface element AS. Surface S bounded by curve C £ i = r_(xj + A x ^ x ^ r 3 = r (Xj + Axj.£ 0 ) = £ (xj + Axj. Pj.1116c.r (x l5 x 2 ) « ^ — Ax 2 according to Taylor's expansion.Ax : • dr . x 2 ) .b.1116a. r j . x 2 + Ax 2 ). (9.1 l18a. S FIGURE 9. Pj.112. OP2.b) (9.b.112 Further let £ 0 =_r (Xj. (9.£ (x 1.P 2 and P 3 on the surface as shown in Figure 9. OP3 be r 0 .c) . we consider four neighboring points Po. r 2 and r 3 respectively. enclosed by the points PQ. OP 1. x 2 + Ax2) (9. The projections of these points on the XjX2plane are denoted by primes. r 2 = r (xj. That is to say.
we find that Equation (9. only and differentiating the last two terms of Equation (9.= i +^dx^ k (9. 22) into Equation (9.1119) and carrying out the cross product. neglecting curvature.1124) lVl+u2x 9U +u2x J3Xl gf Vl+U2x+U2x xl X2 # ^ x —  = n (9. is .1123) The surface area S is given by S = jf V 1 + u^ + u^2 dxj dx2 From Equation (9.1125).1121. we find (9.b) O X^ On the surface.1126) .1113). AS = P Q P J A P ^ . From Equation (9.1120) oxj  U2 = L+t^ <12 91 2) Substituting Equations (9. and k are the unit vectors along the coordinate curves. we obtain ^ . the vector r can be written as r = Xj i_ + x 2 j + u ( x 1 > x 2 ) k where i.1119a.1125) l X2 3X2 Vl+U2x+U2x V / / Noting that V l + u 2 + u 2 is a function of u x . we deduce the condition for the extremum of S to be (9.1120).1125) simplifies to ". 3r 5r = j — A 5 — AxjAx2 OXi (9.776 ADVANCED MATHEMATICS The surface element AS. j . + ( ^ f ] ilJt + (i 2 j ^ _2L * 3 ^ = o L (9.1121) (9.
b) A dxj = xld\ + xdu (9n31) r 2 dr 2 = r 3 dr X! x 2 du r 3 dr (9.1134) dv dr + v + v3 = 0 (9.1126) is difficult to solve.CALCULUS OF VARIATIONS 2ZZ In general the nonlinear Equation (9. by substituting ^ dr Equation (9. (9.1133) Equation (9. we obtain rdJL + dr 2 di + ( d i ) 3 = 0 dr V dr .1128a.1132) 92u 8xj 3x 2 x t x2 d 2 u r 2 dr 2 Substituting Equations (9.1128 to 32) into Equation (9.1127) du r dr (9.1133) can be transformed to a first order equation.1135) . We have to _ du * 3xj dr dXj dx2 r dr = = a (9.b) (9.1126) and simplifying.1130a.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. In this case the equation of the surface is given by x 3 = u(r) where r = Vx 2 + x 2 .1133) becomes r = v (9. We shall consider a special case when the surface is a surface of revolution with the x3axis as the axis of symmetry.
. X m . .42) j m ..1138) Vl+v 2 r We can solve for v and "0.. This can also be written as v (9.1136) 1 vd+v 2 ) which on integrating yields i n v .i n r + i n c where c is an arbitrary constant.. s i . the functional is given by I 1 ri 9ui 3ui dun dun (9. . . Thus the minimum surface is a catenoid.1135) is a separable equation and we have ( dv . . . .1137) ..1139) r = ccoshf1^) This is the equation of a catenary.1140) (9.=f i ft (9. . j f ( x 1 . u m . . .i n ( l +v 2 ) = . u n ) of several independent variables (xj.i . x 2 .778 ADVANCED MATHEMATICS Equation (9..dx m The conditions for I to be an extremum are then .. (9. . . u 1 .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.. s X . . In this case. x m ) is straight forward. Thus the generating curve is (9. .1141) The extension to several functions (uj5 u 2 .. .= £ (9. 5 i d Xl . ^ .
CALCULUS OF VARIATIONS 779 (9.112. 1 = 1. Let V be the flow region (volume) bounded by two nonoverlapping surfaces S v and S t .1143) where V . neglecting inertia. x^ is the deviatoric stress tensor.1145a.1146) m e rate °f deformation tensor. / 3v. y.(p ns + x.= v H = 0 dXj M (9. 46) subject to the boundary conditions stated above will be shown to be equivalent to finding the extremum of . ) ^ . requiring summation over repeated indices. = 1 / 1 y. The equation of continuity is given by ^i. can be written. —j Example 9.1. The equations of motion. ni is the component of the unit normal to the surface S t . as g + ^_ p f l = 0. and is the second invariant of y.\ = ^rL +^is I I . The velocity is given on S v and the traction (surface) force tj [= . Note that in this example..b) The constitutive equation of a generalized Newtonian fluid may be written a Tij = .2. The system given by Equations (9..dv: \ where y. i = = * . v. p is the density and fj are the external applied forces (usually only gravity). we consider different boundary conditions on different parts of the enclosing surface.1144. when referred to a rectangular Cartesian coordinate system O x ^ X g .n:)] is given on S t .3 (9.are the velocity components.1 1 ( 1 1 . Show that the equations governing the slow flow of a generalized Newtonian fluid are equivalent to extremizing a functional. (9.44) where p is the pressure. Remember the convention.
1150) :——.1148a.l 1 / 2 .1149) Jv The extremum of I is given by ^=0 de e=o In the computation of this extremum.p f i ( v i + ew i )(p + eq)(vi + ew i ).r 1/2 Y. dg terms of Vj . Jo The functions inside the integral are functions of the velocity components Vj and the pressure p.1148a..) = I Y (9. so 7 .o .780 ADVANCED MATHEMATICS I [gaiOpfiVipVjiJdVl t iVi dS Jv Jst where g (II.1 r l v. Substituting Equations (9. From the definition of I I . The quantities fj and tj are given.Y o'ij L 2 Y'J Y J l J f9 1151abc) ( " 2 L 2 Yl J Y JJ YlJ " 211. b) into Equation (9.1. we have 1 . Y. Y manipulations.. i d V . we consider neighboring functions v* and p* which are given by v ^ V j + ewj.tO. To find the extremum of I.Y oiij " .b) where Vj and p extremize I. p* = p + eq (9.1147) yield I*=l g ( I i : ) . v. we will be required to calculate —^_ an( j hence j.l t^Vj + ew^dS (9. . Since II. we will have to determine Jst (9. Next we illustrate these 3II.1147) u T (u) du .J L r l . Y ' ' } Since g (II ) = I ur)(u)du. is defined in de y * ds.
On the surface S v .1156b) can then be written as xijWi j d V = I ( w i T i j ) n j d S .b.1155b) It follows that I x i j w i .( Tjj )jWi dV ys t +s v Jv where we used the divergence theorem on the first integral on the right side. n.1157) (9.iqvifi]dV Recognizing that (XjjWj)^ = XjjjWi + XjjWjj or T i j W i j = (TijW^.jTijjWi (9.c. . we obtain I [xijwi.1 Ty (wifj + w jfi ) = xy wifj 9e 3Yij 3e since TJ.1154) (9. is symmetric.j (Tijj)widV (9.1150.= — 1^~ = JTl(n. and so Wj = 0.) Tji . . Equation (9. Combining Equations (9.^ = .^ .1156b) Jv Similarly the integral /st Jv .)T§i = .Ti(ii. j d V = I [(wi^ij)'j'cijJwi]dV Jw Jv = 1 ( w i t i j ) n j d S .^ i j aYij dIU 3Yij 211^ 2 Y J 2 — = .1152a.CALCULUS OF VARIATIONS 781 ag ag an. the velocity is prescribed.b.d) — = ~.1153a.jPfiwiPwi. (9. 53).c) t i W i dS = O (9.1155a) (9.1156a) (9.
c) The radii of the inner and outer cylinders are k R and R.1145) (9. since we used them in formulating the integral given by Equation (9. Use the variational method to solve the axial annular flow of a power law fluid between two coaxial cylinders.b) r = R. The velocity distribution is vr = 0 .1157.1158b) yst Jv Substituting Equations (9.1162) The boundary conditions are r = kR. as illustrated by Figure 9.1144) (9.113.p f i + Pu = 0 vi. we can instead extremize the integral I given by Equation (9.J p.1159) Since w^ and q are arbitrary it follows that t i j j . 0<9<27t. . 46). for all 6 and z (9.113.1161a.jWjdV (9. 58b) into Equation (9.1160) Thus instead of solving Equations (9.i = 0 tj = .jWjdV Jv ' Jv Jv = I pWjnjdSl p.( i j j + p S i p n j (9.b) (9. vz = v z (r) (9.782 ADVANCED MATHEMATICS I p w i .b.1144.1147). vz = 0 vz = 0 . ve = 0 . The flow region to be considered is defined by kR<r<R. Example 9.1163a.1158a) (9. we obtain Jv I [{^ijJ + PfiP'iJwi + q v d d V + l Jst [{Cijj + pSijJnj + tiJJwidS = 0 (9. 0<z<L (9.1154). i d V = I ( p w ^ ^ d V . Note that the boundary conditions are satisfied automatically.1147).1 l64a. We assume the flow to be in the zdirection.
Thus an appropriate v z can be chosen to be I . This is the usual procedure in variational calculus.1168a) . 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. the pressure is given.1167) We select a trial solution which satisfies automatically the boundary conditions [Equations (9.. —* —kR —R FIGURE 9. 1997) (9.1164.65)]. v z = a ( l . Suppose the potential drop from z = 0 to z = L to be A P .CALCULUS OF VARIATIONS 783.1166) (9.1165) ri(y) = mlyl"" 1 We nondimensionalize r by writing r =  (9.\o\ ) (9.(l+n)/n.113 Annular flow bounded by cylindrical surfaces S v and end surfaces S t On the shaded surfaces at z = 0 and z = L.
284 ADVANCED MATHEMATICS q = 2 r .1168b) where a is a constant to be determined..1147)] we need to extremize is now given by 1= I ^ j Y n n + l d V .169a. APvzdS (9. we have n g(Y) = uti(u)du = ^ E y y n + 1 (9.d) where the upper sign is for a < 0 and the lower one for a > 0. Then y is given by .( l +k) (1_k) (9. From Equation (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 (9.1172C) .a .1170a.1172b) .1166).A F R ^ 1 v z [ ( i ^ a + <i±i>] ( l ^ ) d o (9.1172a) h JO JkRn + 1 = Jo L rv z dr Jk lnTir Jk Y n+1 FdrAF27tR 2 (9. 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.b) /o Thus the integral I [Equation (9. 2 dv7 2a (1 + n) i 11/n W *"df = *R Iff = ±^1^10 to = 'nRTTTky (9.c.1147) and where the boundary conditions on the surface S t result in a potential drop.b. dv7 T= 1 dv7 .
C 1 = 0 da Hence RAF ) 1/n R(lk)< l4 *> /n n 2^7/ 2(n+l) . <2n+1) 2n)[nR(lk)] n + 1 ' The integral I is now a function of a.b) using vz. • 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. In Bird et al. since they are odd functions of a.1 l68a.1172) will be zero when integrated from a = . The unknown coefficients are determined by satisfying the variational principle.176a.CALCULUS OF VARIATIONS Z21 Since we have defined 7 and v z as functions of \a\.r k) J KR (9. (9H75) (9. Integration yields I = CopaCn + l f l ^ . the first term in both integrals of Equation (9.q a where C o = ° (9. which is given by ^/ R ^«fer (i c. da Differentiating yields 4 1 = C n (n+l)[2a(n+l)] n »2(n + l ) .25. 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.5 and n > 0. . b. „ .1174a. (1987). This integral has to be minimized and we need to compute ^ . This technique is known as the RayleighRitz method. 75). the approximate solution gets worse and could probably be improved by considering more terms in the trial solution.Q 1 . This is followed by expressing the unknown functions as a series of simple functions that satisfy the boundary conditions.1173) .l to a = 1. For smaller values of k and n. given by Equations (9. the value of Q for various values of n and k are compared with the exact solution. The above approximate solution is within 2% of the exact solution for k > 0.
but now the end points are free to move on the surfaces given by Equation (9.12 TRANSVERSALITY CONDITIONS WHERE THE FUNCTIONAL DEPENDS ON SEVERAL FUNCTIONS We extend the theory considered in Section 9. We further relax the condition that the end points can move only along the ordinates (that is.dB{ 9yj d^ v (912"4b) dx ' . x = b). only along the straight lines x = a. [ df dy: + j=1 df dy': ] (9. .121) Jp The end points P and Q are not fixed nor are the values of y 1.. Equation (9. y2)dx (9.107) becomes * /Q L= l^dx +f ^ . 9.y 2 ) (9.10..y 2 .2 (9. can be solved exactly via the equations of continuity and motion.123a.786 ADVANCED MATHEMATICS Although simple problems. We assume that P and Q lie on two surfaces respectively given by x = <l>(yi>y2)' x = \/(y 1 . y n .122a.f  * = 0 . The quantity 3— is given by d8j df <*ei J. Using Leibnitz's rule.122a. y 2 at these points. y i .8 to the case where the functional depends on n functions.b) We proceed as in Section 9. For simplicity we derive the transversality conditions for the case of two functions y 1 and y 2 . y 2 . such as this flow problem.b) where P and Q are functions of 8j. y . . Variational methods are widely used to obtain approximate solutions to boundary value problems in continuum mechanics.aej ^ = 2 3 y j 9£j =2 j=1 ^^+r^^ :r dy. yj. We aim to find the extremum of rQ 1= f(x. b). i = l.. We allow the end points to move along given curves or surfaces and we label the end points as P and Q. more complex problems require the use of an approximate solution.124a) dy.
y 2 and x are related through Equation (9.  j [a yj . y i and y2 are independent.2 (9.a £ i j p ^ Q a^ P using the convention of summation over repeated indices. if equations y\ and y 2 [see Equation (9.124c) into Equation (9. we deduce the transversality conditions. Therefore.104)] are related for example b y an expression such as y ^ + y2 = c. The transversality conditions are to be imposed at the points P and Q which are on the surfaces given b y Equation (9. At an y point in space. to be functions of x and e. dx a y j 3e. then a change in £j automatically induces a change in e 2 and we can indeed consider the y .122a.104) that (9. derivatives with respect to £j are now replaced b y derivatives with respect to e.124c) Substituting Equation (9. That is to sa y .126) where d^ is the Kronecker delta. j=i p Y j ^ dy.125) J P a£i J9 Yj dx a y . Thus y i . we differentiate to obtain .125). 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. we deduce the extremum conditions to be K_±(K\ 3Yi dx [ dy[ J = 0.128) (9.CALCULUS OF VARIATIONS 787 & F 3f ty d df dy: i+ _ ^ _ i = 2. Substituting Equation (9. i=l. we obtain 0= ^ I dx + li + f ^ f^ (9.126) into Equation (9.129) Since y^ is a function of x and e.127) Next.122a.d / df 1_ 9 ^ dx ^ 3 y j (9.123) and simplifying. except at P and Q. b). It can be seen from Equation (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 .
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 (9.1214) 3yi 8Yl ^ 3yi 3y 2 j J Q r + rh^. The point Q is given by Equation (9.1212) becomes . lla.128).1210) We can also do this for y 2 . 3f \ l ^ L f_y _y n = 0 .1213) \ay2 ay21 a yi ^ 2 j / J Q Thus. ] ! _3f_ I^Xl _ v ' ( ^V dYi ay dY2 \ 1 3 y .1217) ayi\ ayi y7 ay2jjp =0 a +—.1212) Collecting similar terms. the transversality conditions at Q are given by " df + dy /. we obtain ' #  d y i _ v ' / _ ? V dll + i V ^ 2 . .12lla.j5 r)l 3y2 3y 2 \ 3yi ay2jjQ =0 (9m5) Similarly.1210.n^ 1/1N (9.122b) and . the transversality conditions at P are given by rr+:r4yi]F " y ^)i =° ayi 9y 2 (912"l6) (9.b) 7 Substituting Equations (9. Equation (9. dx ADVANCED MATHEMATICS (9. b) into Equation (9.^ is given by 2 de = dx = Mi ^ de dy{ de +  V <^2 dy2 de (9. 3f . de \ayl a y i l ayj dE ! ayi 2 J ) (9.fy.ZM dy. dyi . 9y2 3y i y2jjp .
127) with y^ = y.. \ j=i J3yj j J Q =0 .b) Vl+y'2 . the transversality conditions at Q are "iL 3 ( f _£ .20) "jL+?lf_y'iLl 3y' dy dy' J P =o (9. then the end points P and Q will be free to move on the curves. If f is a function of only one dependent variable y.1222) S =I JO Vl+(y') 2 dx (9.127) simplifies to — dx — dy' = 0 => ^ = constant 3y' (9. £ .218) Similar expression for the condition at P can be established.. 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. 0). Thus if f is a function of yj. y 2 . The transversality conditions then reduce to "iL+?l(fy'it] .1223) Since y does not appear explicitly in the integrand. .1219a.. ] ] _3yi 3y. we have ? y = Co => y' = k 1 (constant) (9. i=1 „ (9.. y n .b) With f given by Equation (9.122!) Example 9.1225a.„ L3y dy \ 3y' j Q (912. x = \j/(y) (9.1223). Find the shortest distance from the origin to a point on the circle of radius 3 and centre (4. respectively given by x = 9(y).CALCULUS OF VARIATIONS 789 The above derivations can be extended to more than two functions.b) The extremum condition will be given by Equation (9.121..1224a. Equation (9.
y l 5 y 2 .1225b) and using the end condition at the origin (x = 0... The function was subjected to certain supplementary conditions. Then the minimum distance is 1. y^) dx = constant = Cj . we obtain 2 kl \ *1J V l + kf  kl =0 (9. A.b) We use the method of Lagrange multipliers and introduce ?ij. consider two types of constraints: an integral type and an algebraic type. 9. .0).. . The condition for the extremum of I subject to the isoperimetric constraint is ..790 ADVANCED MATHEMATICS Integrating Equation (9. .. . we deduce that Q is the point (1.1222). Q lies on the circle given by Equation (9.. We shall. in addition to the end points conditions.1220). y = 0) yields y = kjx (9.9. .. yj.. y = 0 and the point Q is on the xaxis.c) Substituting the value of y into Equation (9. we make use of Equation (9. y n . X2.. we consider the case of a functional which depends on several functions. we established the conditions for the extremum of a functional which depends on only one function.1220)..9. We now look for an extremum of the functional 1= /a f(x. In this section.1222) as well as on the line given by Equation (9.. .131) subject to Jj = I gj (x. Thus f dy . j = i. The solution is kj = 0.1227a. as in Section 9.b. yj..1226).1226) To determine the slope k j .k.. * =^L L dy (x4) = dil± (x4) (9. y ^ d x (9. k (9.13 SUBSIDIARY CONDITIONS WHERE THE FUNCTIONAL DEPENDS ON SEVERAL FUNCTIONS In Section 9..1228) V1 +kf (x4) Vl+kf . y n ...132a. Ja The end conditions are prescribed. y } . .. From Equation (9.
5) into Equation (9.A.133) and noting that f and g are not explicit functions of yj. We have one constraint and so we need to introduce only one Lagrange multiplier X.132a.b).y' 2 )}j = 0 dx 3yj (9. b. 2 xy.b. we have for i = 1 and 2 .A [ .d [2y 2 4x2?iy 2 ] = 0 OX (9.139. I = k 2 + y224xy24y2]dx (9. 3).136a.y 2 2 ] d x = 2 Jo The end point conditions are ( 9 1 3 . Substituting f and g from Equations (9. Equation (9. 10) yields .138) .139) (9.d) y 1 (0) = y 2 (0) = 0 .137) (9.131.134) Jo subject to J = I [y 1 2 .y 22 )}l = 0 dx 9y2 Carrying out the differentiations in Equations (9. 2 + y 22 4xy 2 4y 2 ) + X(y.137.1310) Integrating Equations (9. The k Lagrange multipliers can be determined from Equations (9.x] = 0 4 + . we write d.[ A { ( y i 2 + y224xyi4y2) + ^( y .CALCULUS OF VARIATIONS Z2Z Mf4Mi[iHM}°' Example 9.c. 8) and simplifying.133) We have to solve Equations (9.133) is in general a second order differential equation. The 2n arbitrary constants resulting from the solution of the n differential equations are to be determined from the end point conditions.4 .i r { ( y . 2 x y .132a.134.5a > b ) yi(l) = y2(l)=l (9.x yI .[2yJ + 2^yJ . Find the extremum of i1 i"in (9.
Suppose we need to find the extremum of .1313) (9. The constants Cj to c 4 can be determined from the end point conditions given by Equations (9.792 ADVANCED MATHEMATICS 2y.1311.1317) ~2 x 2 + 2X' y2 = X (9.b) If the constraint is not of an integral form.X ) = c 2 x + c4 where c 3 and c 4 are constants. we obtain 2 y 1 ( l + X ) = ^ ~ + c l X + c3 2 y 2 a .X ) .135) since y J and y 2 are known in terms of X.1314) = ( 4 +23X) » C2 = 2 ( l .1311) (9.2 ? i y 2 = c2 Cl (9.) The Lagrange multiplier can now be determined from Equation (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.136atod)andare cl (9. y2=1 (9. The value of A. is X = 10/11 Thus the required functions yj and y 2 are Yl = .1316a. we find that yj and y 2 are given by Yl = ^ 4(1+X.b) (9.1312) where Cj and c 2 are arbitrary constants. 12) respectively. c3=c4 = 0 (9.1318a. On further integrating Equations (9. 12).d) Substituting the values of C and c 2 into Equations (9.b.1311.c. + 2XyJXx = 2 y 2 .1315a.
... .1320.1322a..2.1320) The end point conditions are given as follows yi (a) = a{ . z 0 ) and (XJ. . zj)...b) Yl We now introduce a Lagragian function L given by m L = f + £ Ajtogj j=l (9.1321a. ... y n (a) = an (b) = b.. y n . 2. . . m (9. yj. 22).1327) .1326) The Lagrangian function L is then L = V l + ( y ' ) 2 +(z') 2 +^(x)(x 2 + y 2 + z 2 ..l ) (9. . y^) = 0 .... y n (b) = b n (9. Determine the shortest curve between two fixed points on the surface of a sphere of unit radius and centre at the origin.1325) pi s= V l + ( y ' ) 2 +(z') 2 dx (9.n (9. .b) (9..yj. y n ....y2. ..132. .1319) Subject to the constraints gj (x. . y]. the equation of the sphere is x2 + y2 + z 2 = l If the curve joins the points (x 0 . Example 9.1324) subject to Equations (9. y i . y ^ d x (9. j = 1. . y 0 .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. In a rectangular Cartesian coordinate system. .CALCULUS OF VARIATIONS JSA 1= Ja f(x. yj..1324) Finding of the extremum of I now consists of solving Equation (9..
± ( ^ ) = 0 3x dt dx i. 2 n y .1335a. z are the dependent variables.e. Noting that d ^ = dxj^ & E q u a t i o n (9.e.b) On differentiating [ — J . The fixed points are x(t o ) = t o .1328a. Equation (9.1334a.1324) identifies the extremum conditions as (9.794 ADVANCED MATHEMATICS It is useful to introduce the parametric representation of the curves as x = x(t).1330a.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. r i i dlioi 2x = d rii _ dtLoi 2y = A ri] dtLgJ.b. y.b) Eliminating l(t) yields d.A U] at a =0 (9. 2z (9.1331) f.1332a.c) where t is now the independent variable and x.~ (^) dy dt 8y =0 i.b) ^ .e. 2nxA [j] = 0 at CT (9. y = y(t). x(t 1 ) = t 1 (9.1333a.— (—) = 0 3z dt V3z i. we obtain . z = z(t) (9.b) and — .b) Equation (9.b) (9.1329a.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.
x y . dt L o J dt L a J Equation (9. we have made use of the following expressions ~ (xyxy) = xy + x v .1337).1342a) .1340). = dLi .1340). 39).b) becomes •• • • oxox 2xa = •• • • ayay 2ya = •• • • ozoz 2ZO2 (9.1338.\ — 1 and — \ — 1.x y dt Integrating Equation (9.1337).1336) We have similar terms for ".1338C) Similarly. (xyxy) (zyzy) Note that in deriving Equation (9. from the last 2 sets of terms in Equation (9.1338b) (9. we have y (ox . from Equation (9.o x ) = x ( a y .1335a.1341a.CALCULUS OF VARIATIONS 795 I [i]  ^ (9.1338a) (9.a y ) or or a (xyxy) = a (xyxy) £ =& ^ a (xyxy) (9.b) (9. we deduce £ (xyxy) ~ J (zyzy) (9.1340) .b) From the first two sets of terms in Equation (9.1337a. we obtain ^n (xyxy) = i n (zyzy) + i n Cj (9.x y = x y . we find £ o = (zjr^ (zyzy) (9 1 3 3 9 ) Also.
This problem is similar to the Plateau problem considered in Section 9. The intersection of the plane given by Equation (9. Silastic implants are widely used in plastic surgery (Atkinson. Example 9.131. yj. Equation (9.1342b) may be written as yCxCjz) = y(xcjz) • XCi (9.1343a) • Z • V or l— = 5y (9.1342b) where Cj is a constant.1144a) (9.1344c) Equation (9.1343b) xCjZ Equation (9.1344c) is the equation of a plane passing through the origin.11.796 ADVANCED MATHEMATICS or xyxy = c^zyzy) (9. as shown in Figure 9. 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. Let A be the area of the base of the implant which is in the xyplane. Zj) are given. Formulate the problem of silastic implant and determine the optimum design of the implant. 1988).133. The arbitrary constants Cj and c 2 can be determined if the fixed points (x0. z0) and (xj.1144b) (9. The technique consists of inserting a deflated silastic balloon underneath the skin.1325) is the arc of the great circle. y0. 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.1343b) can be integrated to yield inCxCjZ) = i n y + /8nc 2 or or xCjZ = c 2 y xc2yc1z = 0 (9.1344c) and the sphere given by Equation (9. .
Let h (x.1124) is given by S=j A [ l + h 2 + h2] 1/2 dxdy (9. y) be the height of the expanded skin above the base. x dx . . h " = 5 dh .1346a. This isoperimetric problem can be solved by introducing the Lagrange multiplier and is equivalent to finding the extremum of L(h. then the surface area S according to Equation (9. dy The volume V of saline is V = I I h (x.. • . Once the skin has been sufficiently expanded. the surgeon will perform the operation and advance the skin to cover the defect area. 9h where h =^.1347a) .131 Geometry of an implant The base of the implant is PQRP and the defect area to be covered is PTRP.b) To ensure that the defect area will be covered we need to know the minimum surface S for a given volume V. I) = SIV (9.CALCULUS OF VARIATIONS 797 y R IMPLANT 1 y HEALTHY «MN A KCALTHV SKIN ^tm^ V^TS^ I I . y) dx dy = Vo A (9.1345) .* /! \ t FIGURE 9.
.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. This nonlinear minimization problem is solved numerically.(x .798 ADVANCED MATHEMATICS = l A (l+h^ + hy) Xhjdxdy (9. 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 (9.1347b) = 11 Fdxdy A (9. We now use the RayleighRitz method to solve for h.1351) a2 where a. The constants Aj. Bj and C^ are to be determined by satisfying the variational condition of Equation (9. 9 h . p\ otj and d are constants and a .1349).a j < d < a + a j .. We choose the trial function for h to be h = (Ai+Bix + ciy) (P 2 x 2 + a V  a2 P 2 ) (ai .1347b) into Equation (9.1348) 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. ..a ?y 2 ._ .. that is 3F 3 / 3F \ 9 / 3F = 0 .1347C) The condition for L to be an extremum is given by Equation (9.(9.8 7 (as.1351) with .1350) a2 2 R2x2 PSR: y 2 = (3 .3 ^ ( 3 ^ ) . (913"48) Substituting F from Equation (9.1113).
X = 2. For other values of the geometric parameters and X.1355) Small values of (3 imply that more weight is given to attaining the desired temperature distribution T d . it can be substituted into Equation (9. The shape of the implant is determined by the values of a.CALCULUS OF VARIATIONS 799 The parameter X can be determined in terms of V o . 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. large values of (3 imply that more importance is given to achieving the .2192.4632. Mathematically we are required to minimize the socalled performance index J defined by (9. (3. 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.1353) where the coordinates of Pj and P 2 are (x1? y) and (x2. since once h has been obtained in term of X. and V o can be obtained.1354) J =  [(TT d ) 2 + p(kk d ) 2 ]dV where T is the temperature and (3 is a weighting factor. We consider an isotropic body of volume V with nonhomogeneous thermal conductivity k(Xj) which is unknown. Bj and Cj are obtained numerically for various values of a. y) respectively. Bj and Q : A{ = 1. Also the values of E and the required values of E for various conditions are tabulated in Atkinson's paper. (9.85.1346) and the relation between A. At the other extreme. Cj = 0. (3 = 1.5. Atkinson has found the following values for A 1 . d. For the values of a = oil = d = 1.134. the reader is referred to the original paper. Thus varying the values of X is equivalent to varying the values of V o . This problem has been considered by Meric (1985). Solve the problem of optimization of thermal conductivities of isotropic solids. ocj and X. Example 9. This is a problem in the manufacture of composite insulating materials. p\ d and a j .131) is given by It L = l ? \ (1+h^j m dx (9. The values of Ai.8186. The length of the curve L on the inflated balloon corresponding to a horizontal (y = constant) section PiP 2 (Figure 9. Bj = 0.
+q + a(TToo) = 0 on (9.112 and consider the neighboring functions T* = T + e T \ k* = k + e k \ X* = l + el' (9. Finding the minimum of J subject to the constraint given by Equation (9.1358) (9. leads to e=0 I (T'(T .1356) is equivalent to finding the extremum of I = J+ X [ V . ( k Y T ) + Q]dV (9. a is the heat transfer coefficient and TM is the ambient temperature. 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 ^ . To show this equivalence we proceed as in Example 9.1362) . de = 0. ^— is the directional derivative along the outward normal to S 2 .b.1360a.1357) (9.1359) where X is the Lagrange multiplier. the equation governing the temperature throughout the body of volume V is Y(kYT) + Q = 0 where Q is the energy. q is the boundary heat flux.800 ADVANCED MATHEMATICS desired thermal conductivity distribution k d .(k*VT*) + Q)]dV (9.k d ) + X'(Y • k YT + Q) + X [Y .c) I*= I h j(T*Td)2 + P(k*k d ) 2 } + A.T d ) + 3k'(k .1361) Then the extremum condition.1356) where Ts is the prescribed temperature on Sj. (k' Y T + k Y T ' ) ] \ dV = 0 (9.*(Y. At steady state. associated with the heat source.
we have I { T ' ( T .1367) On Sj. ^— + k A.1366) Substituting Equations (9.(wVv)vV.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. This implies that (k + e k ' ) ^ .1368) that on S 2 k'^. JO „ + 1 I Si k A. we have I A.+ a T ' dn dn = 0 (9. 64). L aT _.1363.(T + eT') + q + a ( T + e T ' .(wVu)]dV = I w ( u ^ .1365. the boundary condition given by Equation (9.) d S l dn (9.T J = 0 dn Expanding and comparing powers of e.(k'YT)dV = I V S 1 +S 2 is is k'X^dSl dn I [uV.„ I . we deduce from Equation (9.i^ oT . On S2.1364) dnj k'YA.V.1363) iv Jv Using Equations (9. r. ^ dn on T ^— dn dS + I I S2 k A.)dV (9.1358) must also be satisfied by k* and T*.i oA.1365) V I A.1362). (k V T) + Q]J dV V I . i^ a T .V.(wYv)] dV = I w u ^ d S (9.Y X • Y T] + X' [Y . T is prescribed (= Ts) and T* = 0.1368) .k d ) .1369) (9.T d + Y • (k Y X)) + k' [p (k . 66) into Equation (9. .T ' y 1 dS+l V T'Y*(kYA.YTdV (9. The Green's identities may be written as follows I [w V U «Yv + uV .r + k A ^r =— dn \ dn T ^r— dn dS = 0 (9.+ k ^ .(kVT')dV = I V S\ + S2 k L ^ . o T „ ! oh .v ^ .
an infinite plate and an infinite cylinder with T d = T s = k d = 1.I T' Xa + k ^ dS = 0 (9.SQ2 ADVANCED MATHEMATICS Equation (9. . This nonlinear system has been solved numerically and details are given in the original paper by Meric (1985).01. (k V X.1356. in addition to an extra set of equations which may be written as inV: Y«(kYX) + ( T . Equation (9. (3 and for several layers of material are discussed.1367) reduces to I JT'(T .= 0 dn (9.1372) is the socalled gradient condition. the average optimal thermal conductivity decreases slowly with increasing number of layers and increases with increasing heat transfer coefficient.1370) implies that Equations (9.1373) (9. Q = 0. Numerical results for two geometries.1374) We note that Equations (9. 74) are the equations governing A.1371. q = T^ for various values of a.T o + Y . 58)..1371) (9. In the case of the infinite plate.)) + k' [(3(kk d ) .Y X• VT] + X'[Y • (k YT) + Q]) dV V + 1 k ' ? i ^ dS . 58) have to be satisfied. Equation (9. k' and X' are arbitrary.T 0 ) = 0 (3(kkd)YX«YT = 0 onS^ onS 2 : X= 0 Xcc + k ^ .1370) J Si dn J S2 dn Since T'.1372) (9.1356. They are similar to Equations (9.
. We are required to find the straight line V = mT + c which will "best" approximate the data.. n. then ej = (V(Tj) . A processing operation produces twenty cups per minute at 420 K. 4a. A volume temperature profile is represented by n experimental points (Vj. By considering E 2 as a i=l (ii) function of m and c. Answer: h = 2r 3a.. 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. If the cups are sold at one dollar each. At a given temperature T. That is.CALCULUS OF VARIATIONS $01 PROBLEMS la. A cylindrical can holds Vm 3 of soup. use the method of Lagrange multipliers.2.. Tj).6). Evaluate the integral I = I ( y2 + x 2 y')dx Jo .Vj). it costs (T . 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. The constraint is the fixed volume V. i = 1.420) 2 15 + ———— dollars per minute to operate the machine. if we denote the error at the point Vj by e. What is the most profitable temperature to run the machine? Answer: T = 450K 2a. Find the radius r and the height h so that the surface area of the can is a minimum. determine the profit as a function of T.Vj) = (mTj + c . n The "best" line can be defined such that E = ^ e 2 is a minimum. show that for E equations n n to be a minimum. An extra two cups per minute can be produced for each degree above 420 K.. Then determine the minimum surface area.
The curve y = y (x) passes through the points (a. From the values of I determined using (i to iii).667 Answer: 0. deduce if I is a maximum or a minimum. (])) as x = a sin § cos 0 . 1): (i) (ii) (iii) y=x y = x2 y = 1/2 (x + x 2 ) Answer: 0. 0) to (1. 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 . 6b. c). y = a sin § sin 0 . Answer: y = x 5a. Find the curve y (x) which extremizes the functional .B I ds = a I V 1 + (F'(<(>))2 sin 2 ^ d^ JA JA By solving the EulerLagrange equation.700 Answer: 0.804 ADVANCED MATHEMATICS along the following curves joining the points (0. (b. d) and is rotated about the xaxis. z = a cos () Let 0 = F((>) be a curve lying on the surface of the sphere joining two points A and B. y. 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. determine the function F (())) that minimizes the length of the curve. Answer: cot (() = a cos (0 + c2) 7a. 9. cosh Any point (x.675 Find the curve that will extremize I. Answer: y = c.
3 x 2 + x 1 = 1 (^. The boundary conditions are y(O) = y(l) = 0 . 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 . Determine c 0 by extremizing I..x ( l .9 and 9. using the method of constraints explained in Sections 9.^ . Compare the approximate solution ya with the exact solution.y ' 2 + y ' y + y ' +y)dx The boundary conditions are (i) (ii) 9a. y (0) = y (1) = 1 y (0) =1 .y = x. Answer: y = 2 x 3 . y (1) is unspecified. Answer: y a = .10. + 1 Answer: y = x 2 — y + 1 — 7* Show that the boundaryvalue problem y " . 8 a.CALCULUS OF VARIATIONS 805 I =( [y' + (y")2]dx Jo by the following methods (i) (ii) using Equation (9. Find the extremum of /•I y'(0) = y ' ( l ) = l . Answer: y = ^ .73). I can now be considered as a function of cQ.x ) .x ) Substitute ya into I.
806 ADVANCED MATHEMATICS 10a. /•P / —^ Hint: the distance s = I x = 4 + 2y = <>(y). 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. y (0) = 0 and the point P is on the line Answer: ^V~5 lib. The constraint is known as the normalization condition. If C A is the concentration of A.^ / 3 d2 \ The operator V 2 is the Laplacian =^ .4 .6). Consider the irreversible reaction A — B in a tubular reactor of length Z. k and m are constants. 12b. 0) to any point P on the line 2y = x . 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¥ = . Find the shortest distance from the origin (0.T)dx Jo . V 1 + (y ) dx. The differential equation is known as the dxf V i=l I Schrodinger equation (see Section 6. (p is the potential. 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.
then vz = vz ^ ^ / ( a Ap) has to satisfy the following equations of motion V vz = . In the present problem we consider the axial flow of two immiscible fluids in a circular pipe of radius a and length Z. such as crude oil. Show that the EulerLagrange equations are 3C A 9C A 2 dx 2 / Determine T if f = C A + T .1 2_ in phase (liquid) 1 r^/r^V vz = 1 2 _ in phase (liquid) 2 2 Tp. we have to extremize I subject to the kinetic equation. = 9r 2 1 r 3r 1 3 ^ r 2 dQ2 1 7pis the where r\: is the viscosity of liquid i (i = 1. the dependent variables are C A and T. The equation of constraint is in the form of a differential equation and has to be satisfied at each point along the tube. This flow situation was investigated by Bentwich (1976). The boundary conditions are . Bentwich has shown that the power required to transport a required volume of liquid. In Example 9. g = T . that is to say.V 3 C A and the boundary conditions are: C A = 1 at x = 0 and T = T o at x = i . Write down the Lagrange function L. 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.( 2 s i n h 2 i +V3~ cosh 2Jt)] /(2 cosh 21 +V3~ sinh 2 i ) 13b. is reduced if a small volume of less viscous liquid is added to the crude. If v z is nondimensionalized and Ap is the pressure drop. V Laplacian. we have to minimize the production of A.CALCULUS OF VARIATIONS 807 Thus to maximize the production of B. Answer: T = 2 sinh 2x + V T cosh 2x + C (2 cosh 2x + V T sinh 2x) C = [ T 0 . In this example. 2). we used a variational method to solve the axial flow of a powerlaw fluid between two coaxial circular cylinders. The Lagrange multiplier is thus a function of x and is not a constant.113.
Find the equations that extremize I and write down the algebraic equations that A o and A ^ . Jn is the Bessel function of the first kind of order n. we obtain I as a function of Ao and A n m . On substituting the expression for v z in the double integral. have to satisfy.7). ^ n m are the roots of the equations Jn (knm) = 0 and the coefficients A o . the above expansion of vz corresponds to a Fourier series expansion.2 where Rj denote the surface occupied by liquid i. An approximate solution can be expressed as N Vz M = A0 % + X n=0 m=l X Anm V where ())0 = 1 . A n m are constants which are to be determined. We now require to extremize I with respect to Ao and A n m (properties of Jn are given in Section 2. Show that the above boundary value problem corresponds to extremizing the functional Rj Rj+R. . 1 and 2 indicate the evaluation is to be made at the interface in liquid 1 and 2 respectively. ^ 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.7 2 For a fixed 7. Note that for one phase flow v z is given by: v z = 1 .nm7) cosn0.7 2 .808 ADVANCED MATHEMATICS v = 0 on the surface of the pipe which is taken to be at r = — = 1. (j)nm = Jn (A.
In a threedimensional space. Our aim is more modest.22a. The present chapter does not intend to provide such a course. Next. q 2 . ql=r. q2 = 6 (10. then the two generalized coordinates are.1 INTRODUCTION It is not uncommon nowadays to require students in engineering to take a course in statistical mechanics. q2 = 6 (10..21. each particle is defined by three coordinates. the number of generalized coordinates will be reduced to 3N . The idea of an ensemble average will be discussed. This can also be used to pave the way for a course in chaos. a topic which is gaining widespread popularity.21. some topics in thermodynamics and Brownian motion will be examined. The q1 are independent variables and are known as generalized coordinates. we shall extend the analytical mechanics introduced in Chapter 9. Hamiltonian mechanics will be introduced. In the first few sections. If there are m constraints. unlike the Cartesian coordinates need not all have the same physical dimension. we shall consider probability. the only two generalized coordinates are ql=z.2la. Students used to deterministic concepts find it hard to think in terms of probability. We denote these 3N quantities by q1. If there is no constraint between the particles.b) .m.b) If the particle is constrained to move on a flat surface of the cylinder.. Finally. The choice of the generalized coordinates depends on the geometry of the problem under consideration. Let the vector position of particle i be denoted by r j relative to an origin O.2 PHASE SPACE Consider a system consisting of N particles. We plan to provide the necessary background for the students to be able to follow with ease a more demanding course in statistical mechanics. say the surface z = h. Thus if a particle is forced to move on the curved surface of a cylinder of radius a. It will be at an elementary level. 10. we need 3N quantities to describe the configuration of the system. but we shall emphasize the interpretation of probability.CHAPTER 10 SPECIAL TOPICS 10. as shown in Figure 10. . We may regard the system to be in a 3N dimensional space and to have 3N degrees of freedom. The generalized coordinates. as shown in Figure 10. q 3N ..
.m) degrees of freedom and we have n generalized coordinates which are related to the N vector positions r l 5 r 2 . . . The relationship can be expressed as (see Section 4. . r N .810 ADVANCED MATHEMATICS z I I I ' . t) (10. _1_* y X FIGURE 10. — .q n .. t) where t is the time.23a) I N = rj^Cq^q 2 ..21 Motion of a particle on the curved and flat surface of a circular cylinder We now consider the system to have n (= 3N .6) r j = r ^ q ^ q 2 ..q n . The velocity of particle i can be obtained by differentiating r j and we obtain (10...23b) ..
26) . we have deduced Lagrange's equation which is ^A/iM = aq1 dt \aq J j 0 (9. Note that the indices on the vectors Vj and r i appear as subscripts.24b). Some authors dealing with analytical mechanics. (10. The kinetic energy of the system is given by K = 2 £ 1 m i . dr.cp is the Lagrangian and (p is the generalized potential. dr.i ' . Following the notation introduced in Chapter 4. I (10. the indices denoting generalized coordinates are written as superscripts.1034) where L = K .27b).25b) In Example 9.SPECIAL TOPICS 811 Vj = £ j = — q ] + ^ (10.25a) _.i l _ 9r.b) (10. y. Consider a single particle of mass m moving in a potential field which depends on the position only. This suggests that a generalized momentum associated with the q1 coordinate can be defined as (10. L is given by L = im(x2 + y2 + z2)(p It follows that ^ = mx = px dx In Equation (10. we have adopted the summation convention.103. •. adopt a different notation. there is less of a compelling need to distinguish between subscripts and superscripts. as they refer to a particle i. then in Cartesian coordinates (x. They employ subscripts throughout. p x is the xcomponent of the momentum. z).24a.27a.k dr.b) 3qJ 9t In Equation (10. As they consider mostly scalar quantities such as kinetic and potential energies.
qi is known as a cyclic (or ignorable) coordinate. The spring is fixed at the other end and lies on a smooth horizontal table..212) (10.. q 1 . . . .. p 1 . 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. The configuration of the system is specified by the n generalized coordinates q1.. qn and the momenta (mass x velocity) by the n generalized momenta p 1 .28) dq If the coordinate qj does not appear explicitly in L (qJ may be present in L). known as the phase space. . The dynamical state of the system is thus defined by the 2n quantities.. Let q be the displacement of the particle from its equilibrium position at any time t. . The motion of the system is thus described by a path (trajectory or orbit) in this 2ndimensional space. qn. p n . Equation (9.. The Lagrangian is given by L = \ [m(q)2G(q)2] The equation of motion is (10.210) (10. p n . . The mass is displaced from its equilibrium by a distance q 0 along the length of the spring and is then released from rest.812 P1 = . Example 10. ..211) (10..1034) simplifies to = pJ = constant aqj Thus the generalized momentum of a cyclic coordinate is a constant.29) .. which defines a space of 2n dimensions.7 7 ADVANCED MATHEMATICS (10.21. A particle of mass m is attached to the end of a linear spring of natural length /Co and of negligible mass.
213c) is the simple harmonic equation and is the equation governing the motion of all linear oscillators.215b) (10.216a) (10.SPECIAL TOPICS 813. The solution of Equation (10.216b. All oscillators with the same total energy .216a.G ( q ) 2 ]  = 0 (10. Thus the ellipse describes the dynamical state of the particle (or oscillator).215a) (10.213c) is q = A cos cot + B sin cot O q = A C sin cot + B co cos cot where A and B are constants.22.[ m ( 4 ) 2 . as shown in Figure 10.A(mq) = 0 q+co2q = 0 (10.G q .c.d) (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. d) yields (10.213b) (10.214b) f—) +(——\ = 1 \mq0co/ \q 0 / (10. 1 ^[m(q)2G(q)2] . the position and momentum of the particle is given by a point on the ellipse.213c) where co2 = —.217) Thus the trajectory in the phase space is an ellipse (closed curve).^ [ ^ .214a) (10.213a) or or . m Equation (10. At any time t.
814
ADVANCED
MATHEMATICS
describe the same ellipse. The starting point on the ellipse depends on the initial conditions. If we have n oscillators not necessarily all with the same energy, we have 2n variables (q 1 , p 1 ), ... , (qn, p n ). Each set of (q1, pi) corresponds to an oscillator.
f
FIGURE 10.22
The trajectory of an oscillator
1 0 . 3 HAMILTON'S EQUATIONS OF MOTION In Lagrangian dynamics, the variables are (q1, q1, t). In the phase space, the variables are (q1, pJ). It is desirable to transform from the (q1, q1, t) system to the (q1, p 1 , t) system. This can be accomplished by the use of the Legendre transformation. A new function H known as the Hamiltonian is defined as H(pJ, qj,t) = p m q m  L ( q J , q J , t ) , and we sum over m. The differential of H is given by
ffi =
j = l,2,...,n
(10.31)
9H dt
+
3H d q j 3qJ
+
3H d p J 9pJ
(10.32)
Differentiating Equation (10.31), we have
SPECIAL TOPICS
8H
dH = pmdqm + q m d p m  ^ d q j  4 1 d q j  ^ 9qJ 3q J 9t
(10.33a)
= fpi  ^\ dqj + q m dpm  ^
dqJ  ^
(10.33b)
\
aqJj
aqj
at
(10.33C)
= qmdpm^dqJ^ aqj at Equation (10.33c) is obtained using Equation (10.28). From Equations (9.1034, 10.28), we have
*h=*m aqJ dt[aqJ
= ft (Pj) = P j
Substituting Equation (10.34c) into Equation (10.33c) yields dH = q j dpj  p j dqj  ^ at Comparing Equations (10.32, 5), we deduce qj = — aPJ pj =  — aqj 3L 3H
(10.34a)
(10.34b,c)
(10.35)
(10.36a)
(10.36b) „ . . ,
N
 ar = ^
(1036c)
Equations (10.36a to c) are known as Hamilton's equations of motion. They are a set of first order equations, whereas Lagrange's equations are second order equations. Note also the symmetry in pj and qj. We now consider the case where t does not occur explicitly in Equation (10.23). Such a situation may arise if the constraints are time independent. Then from Equation (10.25b), we deduce that K is a quadratic function in q J . We also assume that the system is conservative, then < is a p function of qj but not of q J . It then follows that Equation (10.28) may be written as
816
ADVANCED
MATHEMATICS
Pm = A ( K ~ < P ) = : ~
dq dq K is a quadratic in q m . Using Euler's theorem (Section 1.7), we have
(10.37a,b)
q m p m = q m 4   = 2K 3q
Substituting Equation (10.38b) into Equation (10.31) yields H = 2K(K(p) = K + (p
(10.38a,b)
(10.39a) (10.39b)
Thus the Hamiltonian H represents the total energy of the system and is a constant for the system under consideration. Example 10.31. Obtain the Hamiltonian H from Equation (10.31) for the oscillator described in Example 10.21. Write down Hamilton's equations of motion and show that they reduced to Equation (10.213c). The Lagrangian L is given by Equation (10.212). The generalized momentum p is given by p = — = mq dq Thus H is given by H = p q  L = m(q)2±[m(q)2G(q)2] = ±[m(q) 2 + G(q) 2 ] (10.3lla,b) (10.31 lc) (10.310a,b)
= \ ^ T + G(q)2 From Equations (10.36a, b), we have
(10.31 Id)
q=£
p = Gq
(10.312a)
(10.312b)
SPECIAL TOPICS
811
Eliminating p between Equations (10.312a, b), we obtain the result. Equation (10.217) can also be deduced from Equation (10.31 Id). Since H is a constant, Equation (10.31 Id) may be written as Gq 2 + ^  = C o where C o is a constant depending on the initial conditions. At time t = 0, we have P = 0, and Co = G q 0 2 q = q0 (10.314a,b) (10.314c) (10.313)
Substituting Equation (10.314c) into Equation (10.313) and dividing both sides of the resulting equation by Gq 0 2 , we obtain Equation (10.217). From Equations (10.313, 14c), we observe that the level of energy determines the ellipse. 10.4 POISSON BRACKETS Let f be a function of p 1 , p 2 ,... , p n , q1, q 2 ,..., qn and t. Then
df
=
df
+
df  i , df  i
rD + rO
,1A .
1N
(10.41)
dt
3t
dp1
9qJ
Combining Equations (10.36a, b, 41) yields
df
dt
=
3f + ^3H_iLM
8t
(1 0.4.2a)
dq1 dp1
3p! dql
(10.42b)
= — + [H, f] dt where
[H.f
=iLM4^
9qi dp1 9pi dq1
(10.42C)
is the Poisson bracket of f with H. We note that the Poisson bracket is the Jacobian of the transformation (H, f) to (pi, q'). If f does not depend explicitly on t and the Poisson bracket [H, f] is zero, we deduce from Equation (10.42b) that f is a constant of motion. The Hamiltonian and the total energy are examples of constants of motion. Setting f = H into Equation (10.42c), we observe as expected that H is a
818
ADVANCED
MATHEMATICS
constant of motion. The converse is also true. The Poisson brackets of all constants of motion with H vanish. This provides a test for constants of motion. Hamilton's equations of motion, Equations (10.36a, b), may be written in terms of Poisson brackets. From Equation (10.42c), we obtain
[H,PJ] =  5 ^ =  ^
J
(10.43a,b)
aq 1
= ^
aqJ
(10.43c,d)
[H, qj] =
8ij
^
J
ap 1
aPJ
Combining Equations (10.36a, b, 43a to d) yields
qj = [H, qj] p j = [H, pj]
The Poisson bracket, for any pair of dynamical quantities f and g, is defined as [g,f] = «^^r^ dp 1 dq 1 dq 1 dp 1
(10.44a) (10.44b)
(10.45)
From Equation (10.45), we deduce that, for any three quantities f, g and h, we have [g,f] = [f, g] [f + g,h] = [f, h] + [g, h] [fg, h] = f [ g , h] + g[f, h] [f, [g, h] ] + [g, [h, f] ] + [h, [f, g] ] = 0 [pj, p k ] = 0 [qj, qk] = 0 [P j ,q k ] = S j k (10.46a) (10.46b) (10.46c) (10.46d) (10.46e) (10.46f) (10.46g)
Equation (10.46d) is known as Jacobi's identity. In theory it is possible to construct from this identity a constant of motion, if two constants of motion are known. Thus if in Equation (10.46d), we put h = H (Hamiltonian), f and g are two constants of motion and their Poisson brackets with H vanish. Equation (10.46d) reduces to
SPECIAL TOPICS
819
[H, [f,g]] = 0
(10.47)
Equation (10.47) shows that [f, g], the Poisson bracket of f with g, is a constant of motion. Thus the Poisson bracket of a constant of motion with another constant of motion yields another constant of motion. However, often the constants of motion so obtained are trivial functions of the original constants of motion. The properties of Poisson brackets as given by Equations (10.46a to g) are the properties of Lie algebra [see Lipkin (1985)]. The Poisson bracket corresponds to the commutator in quantum mechanics.
10.5 CANONICAL TRANSFORMATIONS
In Section 10.2, we have shown that if a generalized coordinate is cyclic, its corresponding momentum is a constant. Thus it will be desirable to transform all the generalized coordinates to cyclic coordinates. Normally the most obvious generalized coordinates will not be cyclic. We should then look for a set of generalized coordinates which contains the largest subset of cyclic coordinates. In the Hamiltonian formulation, both the generalized coordinates and the generalized momenta have the same status. Thus we shall consider the transformation from the set (p1, q1) to a new set (P\ Q1) given by P 1 = PMq 1 ,... , q n , p 1 , . . . , p n , t ) Q 1 = Q 1 (q1, ... , qn, p 1 , ... , p n , t) (10.5la) (10.5lb)
We also require that the new P'and Q* satisfy Hamilton's equations of motion. That is to say, there exists a function M (P1, QJ) such that Q1 = ^ M 3P 1 P1 =  ^
( 10. 5 2a)
(10.52b)
Transformations given by Equations (10.52a, b) are known as canonical transformations. If now all the Q* are cyclic coordinates, then M is a function of the P 1 only and the P 1 are constants. Equation (10.52a) can then be integrated to yield Q 1 = f i ( P 1 , . . . , P n ) t + ci where fj = 3P 1 (10.53)
and c; is a set of arbitrary constants to be determined from the initial conditions.
820
ADVANCED MATHEMATICS
Thus if we can find the momenta P1, the coordinates Q1 as well as M, we can integrate the equations of motion. Those quantities can be obtained via the socalled generating function F. Since the sets {pJ, qJ} and {P1, Q'} both satisfy Hamilton's equations, we expect them to satisfy Hamilton's principle which is given in Example 9.103. Using Equation (10.31), Hamilton's principle can be expressed as
SI = 8 I
(p m q m  H (pj, qJ, t) dt = 0
(10.54a)
81 = 8 1
 P m Q m  M ( P j , Q j ,t)jdt = 0
(10.54b)
The two integrals given in Equations (10.54a, b) can differ at most by an arbitrary function of the form " £ . This can be seen by observing that on integrating ^E with respect to t between the limits dt dt we obtain F(t 2 )  F(tj) and since the variation at the end points is zero, the variation of the integral — is zero. Thus we have dt p m q m  H = PmQmM+<iE where F is the generating function. The generating function F can be a function of (4n + 1) variables, the old set {p\ q1}, the new set {P1, Q'} and the time t. Equations (10.5la, b) impose 2n constraints, so that we only have (2n +1) independent variables. The generating function can be one of the following forms F = Fl (qj, Q j , t) , F = F 3 (pj, QJ, t) , F = F 2 (qJ, pj, t) , F = F 4 (pj, PJ, t) (10.56a,b) (10.56c,d) (10.55)
The form of F to be chosen depends on the problem. If the form Fj is chosen, then (10.57) dt 3t dqi dQ]
Substituting Equation (10.57) into Equation (10.55) yields
51. Since we consider the q m and Q m to be independent.Q 9Q m m (10.510) Example 10. we obtain p (10.59c) H = M^L dt The other generating functions can be obtained by using the Legendre transformation and is described in Goldstein (1972).58).SPECIAL TOPICS £27 pmqmH = P m Q m .q at 9q m m +^ i .59a) P m = f^T (105"9b) (10. We choose a generating function of type F. the coefficients of q m and Q m in Equation (10.d) p = mco(qf 2sin 2 Q .M + ^ .58) On the right side of the Equation (10.59a to c).213c).b) (l0. Obtain a cyclic coordinate for the problem considered in Example 10.58) must be identically zero.513a.512) = _ L = mcoqcotQ dq _9FL 3Q = f)F (10.513c. From Equations (10. and write Fj = nuo ( q )2 c o t Q where co is defined in Equation (10.+^ . We thus have (10. We note that if F is not an explicit function of time. we have changed the dummy index from j to m. then H = M (10.511) (10. The Hamiltonian H is given by H = l[^+G(q) 2 ] Note that q is not a cyclic coordinate.21.
514.511. From Equations (10. b) yields M = !(2o)Pcos 2 Q + 2coPsin 2 Q) = (DP Thus Q does not appear explicitly in M and is a cyclic coordinate. From Equation (10.517) (10.516a.518) yields q = V—&• sin (cot + c) (10.515b) (10.0 (10.b) Integrating Equations (10.513e) dO.52b).515a) (10. we then deduce .513c.? 25 n sin2 Q q f = —Q i™ ma Substituting Equation (10.822 ADVANCED MATHEMATICS H = M Equations (10.516a.518) v (10.214a) with (10.517) into Equation (10.52a).519b) . .513a to d) may be rewritten as «» 2 = ^ (p)2 = 2mcoPcos 2 Q Combining Equations (10.519b) is identical to Equation (10.b) yields Q = cot + c where c is an arbitrary constant to be determined from initial condition. 13e. From Equation (10.519a) = \\—2. 14 a. d). we obtain Q = ~ = co (10. we deduce that P is a constant and we denote it by 5 0 .[sin cot cose + cos cot sine] V mco Equation (10.514b) (10.
it is very unlikely that we can accurately determine the 2n initial conditions.. Thus we do not have a precise knowledge of the system. reversing the motion from a point of intersection would lead to the unacceptable possibility of more than one motion for the subsystems. We consider the system we are investigating to be made up of a large number of subsystems. dp n dq 1 . dq n (10. we had to solve a differential equation. . . that is to say. Such a transformation obviates solving a differential equation. n). The integral in Equation (10.. 10. We also need to solve 2n equations of motion so as to obtain the trajectory in the phase space. That is to say.qn. If N is the number of particles in the actual system.21). cose V mco (10. We have assumed that the systems in the ensemble are numerous enough so that p can be considered to be a continuous function. . Similarly if the system consists of N particles and has n degrees of freedom. No trajectories may intersect since this would violate the uniqueness of the system.520a) v ' (10. Thus if the position and momentum of a particle are known at a given instant of time. momenta p 1 . p n and time t. If n is very large. 2.520b) ' Note that in Example (10. we write N = f pdp 1 . the motion of that particle is determined.. which can be taken to be the initial time. .. namely: p'(0) and q1 (0) (i = 1.SPECIAL TOPICS 821 A = V — 2 ... which is a 2ndimensional space.61) is a volume integral in the 2nspace. .. we need 2n initial conditions. we consider an ensemble of subsystems such that each subsystem of the ensemble has the same structure as the original system.. Further it will be time consuming to solve the 2n first order equations of motion.61) The density p is a function of the generalized coordinates q1. .. This leads us to the concept of an ensemble introduced by Gibbs. We can associate a trajectory in the phase space to the motion of a particular subsystem of the ensemble..sine V m© B = \ —2. as opposed to using a canonical transformation in the present example.6 LIOUVILLE'S THEOREM The dynamical evolution of a particle is given by a trajectory in the phase space and any point on that trajectory corresponds to the state of that particle at a particular time. We may also define a density p in the phase space which describes the condition of the ensemble. .
... . . Equation (10. dr2n = 0 (10. .. . p 1 . . Replacing r } . qn.. .64).62) a s where n is the unit outward normal to the surface S. qn.64) becomes ^ . r2n by q1. 6 ..824 ADVANCED MATHEMATICS Let Q be a volume fixed in the phase space and S be the surface enclosing Q. .. Equation (10. _r is a vector in the 2nspace... 6 a. Let r be a vector whose components are q1.36a. . that is to say.64) In Equation (10...b) a^aq1 ^ + ^ = 0 (10. .+ — ( p q V — (PP 1 ) = 0 (10.65b) dq1 From Equation (10. p 1 . we have used the summation convention and we have sum from i = 1 to i = 2n. . p n ...b) we have *ii = J!^ aq1 It follows that = J± aPj (10 . Applying the divergence theorem to Equation (10.. The law of conservation of subsystems in the ensemble leads to ^ \ pdri. ..65a) at or dq1 + q l dp1 ^ dt ^ dq1 + p i^_+p^i dp1 + ^i dp1 = 0 (10.63) Since O is fixed in space and is arbitrary.. dr2n + I div (pr) dr. we obtain ^ j p drj.dr2n = I prndS (10.67) aq1 ap1 . p n ..62).63) implies (10.
Using Equations (10. which form a set of first order equations. and time. The total energy or the Hamiltonian H are constants of motion. The reaction can be described in the (n .+ [H. Functions of these quantities are possible forms for p. The ideas and methods used to solve problems in Hamiltonian mechanics can be extended to other fields of science and technology. Equation (10.610b) reduces to [H. p is not an explicit function of t and Equation (10.1. . are modeled by kinetic rate equations. A n and the mass of Aj by m A .SPECIAL TOPICS 825 Note the similarity of Equation (10. . with time as the independent variable. p] = 0 (10. for example.69) Thus along a trajectory.65b. p is a constant of motion.: dp .68) can be written as 9p dt + 3H 3p_ _ 3H 3p_ dp1 dq1 dq1 dp1 = Q m ^ .610b) If the system is in equilibrium. If we consider n chemical species which we denote by A j . b. Equation (10. Chemical reactions.67) to the equation of continuity in fluid dynamics. . It is the analogue of the conservation of mass in continuum mechanics and is a fundamental theorem in statistical mechanics. There is no equivalent theorem in the configuration or momentum space.l)dimensional space. p* and t.36a. 7) yields dp . They are not unlike Hamilton's equations of motion. p is a constant.68) can be written as (10. we have n equations involving n. 42c). It should be pointed out that Liouville's theorem is valid only in the phase space. p] = 0 3t (10. known as the reaction space. Combining Equations (10.68) expresses Liouville's theorem. (10. m A .: dp — + q 1 ~ + p 1 — =0 ap1 dt aq1 Since p is a function of q1. we have one constraint and the number of degrees of freedom is n .68) §f = 0 where ~ is the material derivative (total or substantial derivative).611) Thus in the state of equilibrium. Since the total mass is conserved. . Equation (10. This suggests that the Hamiltonian formalism is the most appropriate one in statistical mechanics. with .
the equations are coupled and are difficult to solve. the sum of two halves. It is now applied in many fields of science. m A and the time has been eliminated. Further. In our coin experiment. We have excluded the possibility that the coin may roll away.. . Thus the event A U B is certain to occur. In recent years. We therefore look for equivalent generalized coordinates. We denote by A U B the event that either A or B or both occur. We shall start from the simplest experiment which is tossing a coin. there is a growing acceptance of oscillating chemical reactions and the occurrence of chaos in nonequilibrium chemical reactions [Epstein (1983). The two events cannot occur simultaneously. In this case we choose a transformation that will transform m* to b. Thus the first toss and the second toss are independent. then it is as likely to fall on its head as on its tail. If the coin is unbiased.. We assign the value zero to the probability of an event which cannot happen. then the probability of A U B is the sum of the probabilities of A and B. which we denote by AB. B are said to be mutually exclusive if when event A occurs. the outcome of the first toss does not influence the outcome of the second toss. Pascal. 10. The reaction space corresponds to the phase space. If A and B are mutually exclusive. Leibnitz and Bernouilli and was associated with gambling. that is to say. the probability of A U B is one. The probability of certainty is one. the required transformation is the transformation that will diagonalise the rateconstant matrix. event B cannot occur. Wei and Prater (1962) have eloquently shown how this can be done.7 DISCRETE PROBABILITY THEORY The theory of probability was started in the 17th century by Fermat. is the product p (A) p (B).so that the equations will be decoupled. It must fall either on its head or on its tail. The two events are equally likely and we assign to each the probability of 112.826 ADVANCED MATHEMATICS coordinates m A . . Two events A. . If the equations are linear. if A is the event of landing on its head and B on its tail. The outcome of this experiment is either head or tail. event AB cannot occur. If we denote two events by A and B and if the probability that event A occurs is p (A) and that event B occurs is p(B) then the probability that both events occur. Roux (1983)]. A knowledge of Hamiltonian mechanics can be helpful in the understanding of these processes. If A and B are two mutually exclusive events then the probability of AB is zero. In the experiment of tossing a coin. The probability of having two successive heads on tossing a coin twice is 11 A. If we toss a coin twice. Two events are said to be independent if they do not interact. the events of the coin landing on its head and landing on its tail are two mutually exclusive events. technology and commerce.
then the first element can be chosen in n ways. event A does not occur. if A and B are not mutually exclusive.r + 1) and this product is denoted by P n r . (h) PCAIB) = P(B) ^§ If A and B are independent. (n . p (A) = 0. we can determine all the elements of the space. This is the definition in practice. In this case.2 ways and continuing along the same reasoning we find that the number of choices is n (n . . there are n ways of choosing the second element. it is returned and is available to be drawn again. then p(AIB) = p(A). The conditional probability of A given B is defined as the probability of A occurring given that B has happened and is denoted by p (A IB). .p (AB). it is no longer available to be selected. replacement is not allowed. Thus the total number of possible samples of size r is nr.1 ways. if A and B are mutually exclusive. We have thus used the relative frequency of event A as its probability. p(A)>0. Suppose we have n elements a j . If the sampling is random.1. since the number of elements available is now n . Since the sample space is discrete. p (A U B) = p (A) + p (B) . . The third element can be chosen in n . once an element has been selected. that is to say . The probability of A is the ratio of the number of elements of A to the total number of elements of the space. Since there is replacement and once again the sampling is random.. p(AB) = p(A) p(B). To facilitate this computation we state briefly the rules of combination and permutation. The total number of ways for choosing the first two elements is thus n 2 . p (A U B) = p (A) + p (B). To compute the probability of an event A we have to define the sample space.. p (A) = 1. In this case. that is to say. an and out of that set of n elements we want to draw a sample of r elements. each element is equally likely to be selected. event A is certain to occur. In theory we need the number of elements in the sample space to tend to infinity.1). There are two procedures of selecting those elements. There is the sampling with replacement. Thus the first element can be chosen in n ways and the second element in n .. In the second procedure. if A and B are independent.SPECIAL TOPICS §21 The basic rules can be summarized as follows (a) (b) (c) (d) (e) (0 (g) p (A) denotes the probability that event A occurs. once an element has been chosen. .
828 ADVANCED MATHEMATICS Pn r = n (n .. we need to find the number of combinations of n elements taken r at a time. CD. In case (a).. If n = r. . we make no distinction between AB and BA. CB. A change in the order of the r elements will result in a new selection.71. the order of selection is important. Equation (10.73a) = Tu%k> (10'73b) Example 10. the order is not important (similarly for the other pairs AC..1) . DA.71 b) becomes Pn. C. C n r is given by Cn r = . Such a situation would occur if all r elements are alike. DB.72) Thus there are n! ways of permuting n elements.7la) (107lb) = OTO! where n! = n (n .r + 1) (10.. If a change of order in the r elements does not make a difference to the final result. Pn r denotes the number of permutations of n elements taken r at a time. (a) The number of permutations is P 4 .. CA. we can list the elements in each case. 1. 2 = 12 (b) The number of combinations is C4>2 = Y = 6 (10.. D in sets of two.M (10. CA.1) . BA.75) As the sample is small. we have: AB.n = n! (10. BC.74) (10. This is denoted by C n r . Since the number of permutations of r elements is r! and the number of permutations of n elements taken r at a time is P n r . n! ways of ordering n elements. Find (a) the number of permutations. (n . that is to say.. In case (b). AD.. B. (b) the number of combinations of the letters A. BD.. DC. In the determination of the number of permutations. AC.) and the number of elements in this case is half that of (a).
Thus the probability p of having one ball in each cell is P = ni nn (10. the order of the r j elements is not important.rk! Note that r 1 + r 2 + .77a) = r^ r ri!r2!. Within the subgroup. What is the probability of having one ball in each cell? The first ball can placed in any of the n cells. The number of ways of accomplishing this r ways.. p turns out to be 0. Using Equation (10. The first subgroup contains rj elements. Determine the number of ways the partition can be done.77b) (10.r . The number of elements in the sample is now reduced to (n . The total number of possibilities is nn of which n! are favorable.76) If n = 6...r 2 ) ! "" rk!(nr1. Thus the number of permutations is P n n which is given by Equation (10. r i  nr r r 2 . 1 ( r k n! (nr})! (n .r2 . A sample of n elements is partitioned into k subgroups.r ^ ) ! ^ i ( n ..SPECIAL TOPICS 829 Example 10.015. To have only one ball in each cell is equivalent to sampling without replacement: once a cell is filled it cannot be used again.0154.72. we have = ^n. the k th subgroup contains r k elements.73... The number of ways of forming such a subgroup is C n r . the second r2 elements.. there are n possibilities. we find that the total number of ways of randomly distributing the n balls in n cells is n n .^ . Similarly the k th subgroup can be obtained in Cn_r _r _ _r partition can be achieved in C n r Cn_r r .72).r k . . Reasoning along the same line. Thus the is Cn_r r . Similarly the second ball can be placed in any of the n cells.73b). . That is to say.rj) out of which we have to choose r2 elements to form the second subgroup.. we have to select r^ elements from a sample of n elements..rk)! (10. + rk = n (10.. a small probability! Example 10.tx .. the probability that each cell will contain a ball is 0..78) .. Cn_r _r _ _r r ways.. ) ! r2! ( n .. which shows that if we have six balls and six cells. To form the first subgroup. . We have n undistinguishable balls to be distributed randomly in n cells.
We can now calculate the probability b (k. The number of ways of doing this is given by Equation (10. We now consider the case of having six balls distributed in six cells with three of the cells containing two balls. If we throw a dice we can have a six or not. Since there are only two possible outcomes p+q = 1 (10. so that the k successes can be arranged in C n ^ ways. The order of the successes is not important. Thus the probability p of obtaining the desired distribution is p = N.n. POISSON AND NORMAL DISTRIBUTION Binomial Distribution We now consider the case where the trials can have only two outcomes.710a. we deduced the probability of having one ball in each cell. the placing of balls in cells give rise to either empty cells or cells containing one or more balls. n.82a) . Also. p) of k successes in n trials. such as tossing a coin. Thus the total number of ways (N) of obtaining a distribution of three cells containing two balls each is N = £Lfii 3! 3! 2! 2! 2! (10 79) ' K The six balls can be put randomly in the six cells in 6 6 ways. The cells are divided into two groups.830 ADVANCED MATHEMATICS Example 10. Thus b (k. though intuition might suggest otherwise.k) failures with probability q. In Example 10. one is a success with probability p and the other a failure with probability q. = 0. Determine the probability of such a partition. p) is given by b(k.0386 66 (10.k) failures is p k qn~k.b) Comparing with Example 10. we have k successes with probability p and (n .k (10.74. n. Of the two outcomes. The events are independent and the probability of k successes and (n .81) We assume that p remains constant in n trials and that the trials are independent.72. and each group has three elements. 10.77b) and is 6! / (3! 3!). In the n trials.72. The partition of the six balls into three subsets of two elements each and three subsets that are empty can be achieved in 6! /(2! 2! 2!) ways.8 BINOMIAL. those that contain balls and those that are empty.p) = C n k p k q n . we find that the distribution considered in the present example is more than twice as likely to happen as the distribution considered in the previous example.
n. Thus this distribution is known as the binomial distribution. b (k. p) = 1 k=0 (10. 2a.82b).86c) (n±l)pk kq From Equation (10.. + p n . Then from Equation (10. we note that as k increases from 0 to n... we obtain JT b(k. if the dice is perfect.1 < m < ( n + l ) p (10. + Cn k q". n.83a.J p + Cn>2 q n ~ 2 p 2 + . n. Compute the probability of obtaining one six in twelve throws of a dice. (10.269 (10. + "foD. 1/6) = ^ j (1) (I) 11 = 0. Using Equations (10. + p" (10. we have b(l.k + l ) ! y H n ! p kl q nk+l ( 1 0 8 6a) = P(n~k+1) kq = 1+ (10. The trials are independent and the probability of obtaining a six in each throw remains the same. p) is the kth term of the expansion of (q + p) n .1 successes.("~k+1) qnkpk + .. .k p k + . we obtain < I0 .b) To compare the probability of having k successes with k ..82b) P> b(kl. is 1/6..l ) ! ( n .85a.87) .83a) = qn + C n J q".86c). The probability of obtaining a six in one throw. The value of m is given by (n+ l ) p . n.86b) (10. 2a). we find that b (k. we have from Equation (10.8 .p) b f e n> = n] k!(nk)! k qnk ( k . 12. p) at first increases until k is equal to m and then decreases.83b) Comparing Equations (10.84) Example 10.81.2b > (q + p)" = q" + nqnlp+ "i^dl qn2p2 + . 3b).81.SPECIAL TOPICS 831 = Fu^P k ^ k If we expand (p + q)n as a binomial expansion.
p is small. 1/2) = ^ ° ^ .(27tn)1/2 nn e~n (10. p) (10. especially when n is greater than ten.812b) . For 100! the percentage error reduces to 0.811) The percentage error decreases dramatically as n increases.l . In such a case.n.^f° ^f° = 0.82. If (n + l)p is an integer and is equal to m. (a) We assume that the coin is perfect and that each toss is independent. A coin is tossed 40 times. though greater than for 25 heads.832 ADVANCED MATHEMATICS The number m is the most probable number of successes.40.82b.n. 1) b(0.b) The probability of obtaining 20 heads is b(20. p) = b ( m .8%. The exact value of 10! is 3 628 800.89a. it is desirable to find an approximation to the binomial distribution.1254 (10. p) = Of p0q n n! = (lp)n (10. Poisson Distribution In many cases. n is large and successes are rare. Calculate the probabilities of obtaining (a) 25 heads.810a.0366 (10. Stirling's formula yields 3 598 600 resulting in a percentage error of 0. 1/2) = ^ (b) ^ ( y f 5 ( ^ ) 1 5 = 0. n . is not small The probability of failure is. that is to say. We can then use Stirling's formula which is n! . • To calculate n! can be tedious. we deduce from Equation (10. Only in one out of eight experiments can we expect to have twenty heads. We assume that A. (b) 20 heads.08%.88) Example 10. using Equations (10.86c) b(m.812a) (10.b) We note that 20 is the most probable number of heads but its probability is still small. The probability of obtaining 25 heads is b(25.40. The product n p is the average or expected value and is denoted by X.
814) (  (1°815a) (10.— .e^ This can be shown by induction as follows.n.p) .n. b(l.n.812c) Taking the logarithms of both sides of Equation (10.^ Jv / (10. we obtain b(k.~ e~^ Generalizing. we then have (10. from Equations (10.815b) since p is small. 15b).n..815b.86b) may be written as b(k.818) (10.k ) " k(lp) .n.813a) (10.816) .814.p) b(kl. Equation (10. we obtain inb(0. (10.818) is true for k = s.p) = n i n ( l .U n ) = — .813b) implies b(0. p) .n.p) _ ?i + p ( l .813b) n I n 2n 2 as n—> °o.p) « e~x Equation (10.e~xX Similarly b(2.812c) and expanding the resulting right side.n..p) . Assume that Equation (10. From Equations (10. 18).817b) (10.SPECIAL TOPICS 831 = (1 X/n)n (10.n.p) .^ . we obtain b(l.817a) (10.
We now derive an approximation of b (k.19c) Thus Equation (10. X) is one. p) relaxing the condition that p has to be small. The number of atoms disintegrating per second in a quantity of radioactive substance. by induction.) (10. the Poisson distribution is also a distribution of random rare events and is not merely an approximation of the binomial distribution.819a) ~ r ^ T V K e~X (s + 1) s! T.821) X P(k..820.) = 1 k=0 (10. X) = A .816). the number of hits on a target. Normal Distribution Let u denote the deviation from the mean value np. are all examples of the Poisson distribution.834 ADVANCED MATHEMATICS b(s+l.820) is known as the Poisson distribution. A.818) is true for (s + 1) and from Equation (10. the number of wrong telephonic connections. the number of misprints in a book. p) (s + >. n. .. X). S + l (10. we know that it is true for s = 1 and thus. X) given by Equation (10. +£+ . it is true for all s. The expansion of e^1 is e^ = i + A + . Since the sum of p (k.818) is now written as p(k. u = knp (10.. p) under the conditions p—> 0. the number of reported cases of a rare disease .822) for any fixed X. p) . and the number of automobile accidents in a fixed time interval at a particular spot. n.823) .e* (10. n. Combining Equations (10. n—> °o and np—> finite X by p (k.819b) " (TTT)! e~" (la8 .820) The distribution p (k.. n. Equation (10.r^k— b(s. 21) yields (10. We denote the approximation of b (k.
824c) (10.u) + np (l+ ^\ i n p + nq ( l .iin[27U(np + u)] + n p ( l + n ^ ) i n n p + ^L L (10. we have i n n! « L i n (2nn) + n i n n .829a) npj[ P np 2 n 2 p 2 J l nqj^ 4 nq 2^2q2 .825) Taking the logarithms on both sides gives Znb = Zn n! + (np + u) i n p + (nq .i n (np + u)! .i n (nq .SPECIAL TOPICS 831 If i denotes the number of failures.n + (np + u) + (nq .827) (10.824b) (10.826) (10. p) = ^ fl (u + np)! (nqu)! p n P + u qn^u F M (10.n Similarly i n (np + u)! « 1 i n [2K (np + u)] + (np + u) i n (np + u) .82b) yields b(k.828a) (10.828b) (10.J L ) i n n q .823.i n 2% (nq .828d) i n (nqu)! . 28d.(np + u) .811)].u)] + n i n n _ np (l+_u_\ i n n p + U Pl ^ nq(lM innqU 4 ^~ (10. n.826) yields 7 i n b = ^. 28e) in Equation (10.^ ^ (nqu) (10.i n 2U (np + u) .lin[27t(np+u)] + n p ( l + ^ ) [ i n n p + i n ( l + ^ ) ] .(np + u) » 1 i n [2rc (np + u)] + np (l+ JL) i n [np (l+ ^ ) ] . then i is given by i = nk = nunp = nqu Substituting Equations (10.828e) Substitution of Equations (10.I in[27i (nqu)] + n q ( l .824a) (10. 24c) into Equation (10.827.[in (27tn) .( n p + u ) .u)! Using Stirling's formula [Equation (10.828c) _ ( n p + u) (10.^\ i n q .u) i n q .
830).!>8n27cnpqy— 2 2npq (10.832b) . The variable x is a measure of the departure from the mean value np [see Equation (10.Jin27Cnpq^— ^— 2 2np 2nq 2 (10.831b) (10.831a) (10.4 = e" x 2 / 2 a V2rc « ^r <p(x) In Equation (10.831a) (10. p) « . n. Problem 9b. Using the result given in Chapter 4. x can be on either side of the mean value np . (p(x) is given by cp(x) = k= ex2/2 (10. we introduce a = Vnpq x = i a Equation (10.830) V27tnpq To simplify the expression on the right side of Equation (10. In the binomial distribution.832a) (10. n. so that its value can range from °o to °°.829c) .823)].833) V2lu The function cp (x) is known as the standard normal density function.832b).u ) 27tn J ADVANCED MATHEMATICS 1 2 L 0 B /1 \ u\r« ^V) n 1 ( 1+ ^)( u 4).836 2 4TU (np + u ) ( n q . we obtain b (k. we calculated the probability of k successes in a discrete sample space. The function cp (x) is now a continuous function of a continuous variable x. p) « 1 exp (u 2 / 2npq) (10.830) now becomes b (k. The quantity a is the standard deviation of the binomial distribution. we obtain (10.829d) Taking the antilogarithm of Equation (10.nq ( 1 ^) [innq " nq]+ ( 1 ^)( u+ ^ J ) (ia829b) 2 2 .829d).
33) are for continuous variables.SPECIAL TOPICS 837 I (p(x)dx = 1 (10.unpq \ 2npq l (c) (d) (10. 33) can be written as b(k.838a) c fc°E) = JL.837) (10.836) (10. Equation (10.834) Thus (p (x) satisfies the condition that the sum (integral) of all probabilities is a certainty and is equal to one.837) is known as Laplace's limit theorem. that is.t ^ l p ^[fnpqj i2n 2npq / (10. The cumulative distribution function of a standard normal random variable is denoted by O (x) and is defined as J (a) (b) fx I 2 e~y / 2 dy (10. is p ( a < x < b ) = O(b)O(a) Equation (10. D We list some of the properties of (p (x) and O (x) (p (x) is symmetrical about x = 0. p) « exp(k~np)2 V2.838b) Equations (10. O(x) = 1 . In terms of the original variables.835) <£ is also known as the normal distribution function . exp . the most probable number of successes. O (x) increases steadily from 0 to 1.832b. The values of < (x) for positive values of x are tabulated.832b. n. (p (x) has its maximum at x = 0. The probability that x lies between a and b.838a. when k = np. . b) can be used in the case of discrete variables whereas Equations (10. a < b.O(x).
Getting a full house is a rare event which can be assumed to follow the Poisson distribution. l ) = e ^ + e"1 = 0.840a) (10. Calculate the probability that out of a sample of twenty items.841a.p = 0. The probability of being dealt a full house in a hand of poker is 0.0014 x 1000 = 1. X) . In this case.d) (10. 4 e . there will be at most one defective item.83. From Equation (10.839a.736 The probability of having at least one defective item is q = 1 .838 ADVANCED MATHEMATICS Example 10. A machine is known to produce on an average one faulty item out of twenty. l) + p ( l .4 e"14 = l . we obtain . which is not small.4 The probability of having at least two houses is p = 1 . the distribution is normal and not Poisson since the value of p associated with each toss is 0.838a).1.p (1.2 .p (0.840b) (10.c) Thus it is almost twice more likely to find at most one faulty item than to find more than one faulty item.842b.0014.85.408 (10. The probability of getting a head is 1/2. A perfect coin is tossed 40 times.b) The probability of having at most one faulty item is p = p(0. We assume that the distribution is Poisson and the parameter X is given by 1 = 2°X2O = l (10.b) (10. Compute the probability of being dealt at least two full houses in one thousand hands.b) Example 10. is I = 0. Example 10.843a. The parameter A.L 4 = 0.84. Determine the probability of having twenty heads.840c.e"14 .5.842a) (10. X) = 1 .264 (10.
We first calculate the probability of the number of men whose heights are 6 feet (= 72 inches) and below. 1 6 ) .837) to determine the probability of twenty heads.0.885 The expected number of men who are at most 6 feet tall is N t = 10 000x0. b) b = ^ = ^ .6 ^ < x < ? 2 ~ 6 9 ) = O(1. The number 19.0. x has to be transformed into a continuous variable. we have p f . the expected number of men to be exactly 6 feet tall.8 850 = 1 150 (10.b) .( 2 °~ 0 2 Q ) 2 ) .848a.16 (10.16)4> (0.885 = 8 850 The number of men that are at least 6 feet tall is N 2 = 1 0 0 0 0 .b) We could also use Equation (10. 31a. as expected.846a) (10.b) (10.846b. give the same answer.b) (10.13 (10.844a..823.5 and this leads us to calculate aandb via Equations (10. Both methods.836) and a Table of O. From Equation (10.845c.837). The heights of 10 000 young men follow closely a normal distribution with a mean of 69 inches and a standard deviation of 2.847a. In this case.5 span the range closest to the integer 20.5 inches.d) a = (19^220) « _0.5 and 20.849a.0.SPECIAL TOPICS 839 p = ^ L = exp(.l .b) (10.13 We have used Equation (10. Find (a) (b) (c) (a) the expected number of men N2 to be at least 6 feet tall.16) = 2 O ( 0 .86. The number of heads should then be between 19.5 and 20.845a. (10.2)O(28) .l6 The probability of having twenty heads is thus p = 0(0.0.c) Example 10. the expected number of men N to be between 70 inches and 74 inches tall.
statistical methods are most appropriate. Though we cannot predict its coordinate q and its momentum p. In statistical mechanics.b) In Section 10. E cannot be determined exactly. we predict the macroscopic properties of the system from its microscopic properties. For example.322x10 000 = 3 200 1 0 .2.655 « 0. Since it is isolated. We illustrate the idea of an ensemble by considering a simple example. we established that the state of a dynamical system is defined if the generalized coordinates and momenta are known.21 is isolated and is in equilibrium. (c) P f . a = b = 1. it is impossible to know the initial conditions of each particle and it is prohibitively impossible to solve the equations of motion.850b. Its energy is therefore constant. the trajectory for a constant E is an ellipse. In quantum mechanics. In our example.91. In classical mechanics.^ T ^ X1^TT') = *( 2 )*( 0 . since we cannot measure to that level of accuracy. We do not expect anybody to be exactly 6 feet tall.322 The number of men N whose heights are between 70 and 74 inches is N = 0. we can solve the equations of motion and predict the coordinate q and momentum p of the oscillator at any time. These same methods are used to solve problems in mechanics. 9 SCOPE OF STATISTICAL MECHANICS (10. There are many oscillators which could satisfy the conditions of being isolated and having energy in the specified range and they form an ensemble. as shown in Example 10.8. In Section 10. it does not exchange energy with any other system. we have seen that statistical methods can be employed to solve real problems.850a) (10.2 and the probability (area of a line) is zero.0. the pressure on the wall of a gas container is due to the impacts of the gas molecules on the wall.9770.840 ADVANCED MATHEMATICS (b) In this case. Since we are dealing with a system of a large number of particles. we know they must be in the area between the two ellipses in the phase plane.6. one corresponding to E and the other to E + AE as shown in Figure 10. that we do not know the initial conditions with sufficient precision and we are unable to solve the equations of motion. . But suppose as stated earlier. E can theoretically be determined exactly but in practice there is bound to be an error AE in its determination. in equilibrium and its energy lies between E and E + AE. Consequently the idea of an ensemble was introduced in Section 10.851a. Suppose the oscillator in Example 10. Since the number of impacts per second per square centimeter is of the order of a few billion. In the phase plane. Let its energy be in the range E and E + AE.4 ) (10. we have two ellipses. A pressure gauge is not able to measure the pressure exerted by each individual molecule but it measures only an average force due to a large number of impacts.c) . All we know is that the oscillator is isolated.21. If we know the initial conditions.
.SPECIAL TOPICS ML They could have distinct initial conditions but they all will be confined to the area between the two ellipses in the phase plane.. We have n generalized coordinates q^.. qn and p ] . and in this case.. p) is prob. p).. .. . We introduce a probability distribution f (q. Apn and Aq 1? . p n respectively.. . Apn respectively. we write q and p for q1... . The integral of f (q. .. (10. Aqn and the probability distribution function is f ( q b . . . Then f (q. p n . . the total number of particles. . and the dimension of the phase space is 2n. Similarly Aq and Ap stand for Aq1? . qn and n generalized momenta pj. each of size Aq Ap.91) f>Ap FIGURE 10.p)AqAp If there is one oscillator then the integral I I f (q... f is normalized such that its integral over the complete phase space is one. p) dq dp over the relevant area is one. We can now introduce statistical methods. The phase space is divided into cells of size Ap1. For brevity... p) Aq Ap is the probable number of particles in the cell of size Aq Ap at (q.. p) is known as the phase space distribution function. We subdivide the area between the ellipses into small cells.. . . Aqn and Apj. f is the probability density.. . . p).91 Phase diagram of an oscillator at two different energy levels We can extend to the case of N particles with n degrees of freedom. as shown in Figure 10. which is a continuous function of q and p... The probability of an oscillator occupying the cell of size Aq Ap at (q. qn. The function f(q. p) over the whole phase space is N. p n ). In many cases. = f(q. . p b . . .91...
842
ADVANCED
MATHEMATICS
If B (q, p) is any quantity associated with the system, then the average value of B, denoted by <B>, is
([ <B> = —
B(q,p)f(q,p)dqdp (10.92)
fj
f(q,p)dqdp
Based on some plausible physical assumptions, it is possible to determine f. The average value of the physical quantities can be calculated.
10.10
BASIC ASSUMPTIONS
One of the basic assumptions is that all microstates having the same energy have the same probability of existence. Let us consider an isolated system consisting of N particles in a volume V. Since the system is isolated, its energy is constant. Suppose it lies in the range E and E + AE. All systems having N particles occupying volume V with energy in the range E and E + AE form an ensemble. The probability assumption for such an ensemble in equilibrium is that each of its accessible states is equally likely. Thus the probability P r of finding a system with energy E r is given by (C Pr = / \0 i f E < E r < E + AE (10.10la.b) otherwise
where C is a constant. Such an ensemble is called a microcanonical ensemble. If the system is in thermal contact with other systems or heat reservoirs then there will be an exchange of energy and its energy will not be constant. An ensemble of such systems which are in thermal equilibrium is known as a canonical ensemble. Consider two systems A and B which are in thermal equilibrium with the same heat reservoir. We denote the probability of finding the system A with energy E A by P A (E A ). Similarly the probability of finding B with energy E B is P B (EB). The systems A and B can be regarded as one system with energy E A + B and the probability of the composite system with energy E A + B is denoted by PA+B (EA+B) Since A and B are macroscopic system, we can assume that the interaction energy is negligible compared to E A and E B , so that
EA+B
= EA + E B
d0 1 0 2)
Using Equation (10.102), we have
PA+B(EA+B) = PA+B(EA + EB)
O0.103)
SPECIAL TOPICS
843
The right side of Equation (10.103) can be interpreted as expressing the probability of system A having energy E A while, at the same time, system B has energy E B . We assume systems A and B to be independent. Using relation (e) of Section 10.7, we have
PA+B(EA + EB)
= PA(EA) W
(10104)
Since we have assumed systems A and B to be independent, E A and E B are independent. Differentiating the left side of Equation (10.104) with respect to E A , we have ^
[PA + B(EA + B B ) ]
= ^
^
[^(BA
+
EB)]d^AlM
(lo.lO.5a)
= PA+B(EA + EB) where the prime refers to differentiation with respect to (EA + EB). Similarly ^ [PA+B (EA + E B)1 = PA+B (EA + EB)
(10.105b)
(10.106)
Differentiating both sides of Equation (10.104) and using Equations (10.105b, 6), we have PA(EA) P B (E B ) = PA(EA) PglEg) where the prime refers to differentiation with respect to the argument. Equation (10.107) implies
PA(EA) PA(EA)
(10.107)
_
^(BB) _
PB(EB)
a
P
nn
m
SaM
"
""
(10.108a,b)
Since systems A and B are independent, P is a constant, that is to say, it is a quantity independent of systems A and B. Integrating Equations (10.108a,b), we have PA(EA) = CAePEA P B (E B ) = CBePEB (10.109a) (10.109b)
where C A and CB are constants and depend on the systems. The quantity J3 is independent of A and B but may depend on the heat reservoir. The temperature T is the quantity that characterizes a reservoir, thus p can depend on T.
844
ADVANCED
MATHEMATICS
Generalizing Equations (10.109a, b), we state that the probability Pr of a system in thermal equilibrium in a particular state with energy Er is Pr = Ce pE r To determine C we sum over all possible microstates, and we obtain JCe"^
r
(10.1010)
= 1
(10.1011)
Thus C is given by C =
r
1
(10.1012)
Equation (10.1010) may be written as Pr = Z ^ e ^ r (10.1013)
where Z = £
r
e"^.
If the energy can vary continuously, the distribution function f (E) is given by f(E) = DeP E (10.1014)
where E is a continuous function of the coordinates q, and momenta pj. The constant D can be obtained by integrating over the whole phase space. Thus, using the notation of Equation (10.92), we have D e" p E dqdp = 1 (10.1015)
Using Equation (10.1015), Equation (10.1014) can be written as f (E) = Z"1 DN e~pE where Z = DN (( e~ p E dqdp. (10.1016)
The quantity DN has been introduced so that Z becomes dimensionless. Note that in Equation (10.1013), Z is dimensionless. In classical mechanics, it is usual to assume DN to be unity. The quantity Z (which stands for Zustandssumme, state sum) is called the partition function. The exponential factor e~PE is known as the Boltzmann factor. The probability distribution given by
SPECIAL TOPICS
845
Equations (10.1013, 16) is known as the canonical distribution, and DN is assumed to be unity. The factor P is given by p = 1/kT where k is Boltzmann's constant and T is the temperature. 10.11 STATISTICAL THERMODYNAMICS (10.1017)
The internal energy U of a system is identified as the statistical average energy <E>. Thus combining Equations (10.92, 1013), we have
U = <E> = z r ' J ^ e " ^
r
(10.11la,b)
On differentiating the partition function Z with respect to p, we obtain
— =  1 ^ " ^ dp r
Comparing Equations (10.11la, b, 2), we find that U =Z!^ = — (inZ)
(10.112)
(10.1 l3a) (10.1 l3b)
ap
The partition function Z is a function of the temperature T and the volume V. Thus d(inZ) = ^  ( i n Z ) d T + ^  ( i n Z ) d V
aI aV
(10.1 l4a)
= — (inZ)dT^ + ^(inZ)dV
(10.1 l4b)
= — (inZ)dp+^(inZ)dV
(10.114c)
ap
ov
av
= Udp+^linZJdV = d(up)+pdU+^UnZ)dV
aV
(10.114d) (10.114e)
846
ADVANCED
MATHEMATICS
Using Equation (10.1017), Equation (10.1 l4e) can be written as dU = k T d ^ n Z + u p l  k T ^ r U n Z ) d V (10.115)
From the laws of thermodynamics, we can relate the quantities in Equation (10.115) to other thermodynamic variables. We recall that the first law of thermodynamics may be written as dU = dW + dQ (10.116)
where dW is the infinitesimal work done by the system and dQ is the infinitesimal heat absorbed by the system. The second law of thermodynamics introduces the entropy S and dS is related to dQ by dS = ^ For an ideal gas, the work done dW is given by dW = pdV where p is the pressure. Substituting Equations (10.117, 8) into Equation (10.116) yields dU =  p d V + TdS Comparing Equations (10.115, 9), we deduce p = kT^yUnZ) aV dS = k d ( i n Z + Up) On integrating Equation (10.1 l10b), we obtain S = k i n Z + kU(3 + constant (10.1111) (10.1110a) (10.1110b) (10.119) (10.118) (10.117)
The constant term in Equation (10.1111) has no physical significance since only a change in S can be measured. We now introduce the Helmholtz free energy A which is defined as A = UTS (10.1112)
SPECIAL TOPICS
847
Differentiating A yields dA = d U  T d S  S d T Combining Equations (10.119, 13) results in dA=pdVSdT We consider A to be a function of two independent variables T and V, that is to say A = A(T,V) It then follows that dA = l  £
\dl/v
(10.1113)
(10.1114)
(10.1115)
dT+ ~
\CV/ T
dV
(101116)
Comparing Equations (10.1114, 16) yields
(5v) T = "P
I?A\ ~ = S Wl/y
(1011  17a)
(10.1117b)
Since A is a continuous function of T and V with continuous partial derivatives, we deduce from Equations (10.1 l17a,b)
IU) = [W\
A function H (S, p) is now introduced. H is known as the enthalpy and is defined as H = U + pV On differentiating, we have dH = dU + pdV + Vdp Since H is a function of S and p, we can write
(iani8)
The four thermodynamic variables p, V, S and T are not all independent. They are related via Equation (10.119). We can consider any two of them as independent variables. In the definition of A, we have assumed T and V to be the independent variables.
(10.1119)
(10.1120)
848
ADVANCED
MATHEMATICS
dH =  £  dS+ ^ dp \dsjv \dp!s
Combining Equations (10.119, 20, 21), we deduce r)H\ T = ^
(10.1121)
(10.1122a)
V = ~ Wp/ S Cross differentiation of Equations (10.1 l22a, b) yields
(10.1122b)
(II • i l
G = HTS or G = A + pV
(10.1123)
The Gibbs free energy G is usually assumed to be a function of T and p and is defined as (10.1124a) (10.1124b)
Proceeding as in the case of A and H, we deduce that S =  Kzr (10.1125a)
V = ~)
(10.1125b)
(10.1125c) In an isothermal process, dT = 0 and Equation (10.1113) becomes dU = dA + TdS From Equations (10.116, 7, 26), we deduce dA =  d W (10.1127) (101126)
Substituting Equation (10.1111) into Equation (10.1112), choosing the constant term to be zero and p = 1/kT, we obtain A = kTinZ (10.1128)
...1129a) (10. e .1132) Z = V N I . Since the energy does not depend on the position of the atoms. Let vj be the velocity of atom i. the E i are independent of the £ j .1129b) (10..1016) with D N = 1 Z = I I e"pEi e"pE2 .. each of mass m. for a perfect gas. I e~PEi .e^ E N dvj . (10.1130) = t = mXi * Xi + v2 + v2). and we can carry out the integral with respect to d£.. (10.1133) is given by (10.1013) can now be written as Pr = e ^ " ^ This form of P r is quite common in statistical mechanics.p E n dr dv where d. There are N atoms and since each integral with respect to d r j yields the volume V.. Ej be the kinetic energy. The potential energy is zero since. d y n respectively... Z can be written as Z = e^1" = ePA The probability P as given by Equation (10.1131a) (10. d v n . d £ n and dv 1 .111.. the interatomic forces are negligible. Each of the integrals in Equation (10. Example 10..SPECIAL TOPICS 849 From Equation (10. N .r and dv are d£j . Calculate the partition function Z of a perfect monoatomic gas containing N atoms..1133) . Ei (10.1128).1131b) lm(v2 The partition function Z is given by Equation (10.
we obtain p = k T ^ [NinV + ^ aV (10. is given by U = .1136) Ipm/ i n Z = N/enV + ^K inf^L\ (10.1137) (10.1139c) in^] (10. v z .1 l3b).1139a) ap [ 2 \pmjj (10.1133) yields Z = VNM3N/2 (10. Equation (10. we find I = I2x)3/2 Ipm/ Substituting I into Equation (10.1 l10a).1135) Using the results given in Chapter 4.1138). Problem 9b.M P E L 2K\ 2 2TU [ m(3 j = ~ The pressure p is given by Equation (10.1140a) (10. Using Equation (10.1134) can be written as I = I y°° e ( 1 / 2 >P m v 2 dv / (10.— [ N i n V + ^ in/^L\l (10.850 oo oo oo ADVANCED MATHEMATICS 1 = f [ [ e.P(1/2)m ( v2x + V' + V ' ) d v x d v y d v z OO OO OO (10. so that Equation (10.1138) 2 Ipmj The internal energy U.1140b) [ 2 U3m)J = ™i . v y .113%) = .1134) The limits of the integral are the same for each of the velocity components v x .
1016)...102c) \PcJ where Z' stands for the remaining multiple integrals.123c) (10. In many actual situations. and the n generalized momenta pj.111..121) Z = I .SPECIAL TOPICS 851 Equation (10. In Example 10.122b) = 4' 2 Z' (10. I e" p W + E ') d P l dq dp 2 ... In the liquid state. E is a quadratic function of the momenta. where there are strong interactions. . The internal energy U is given by U = . all the other thermodynamic variables can be deduced. can be written as (10.12 THE EQUIPARTITION THEOREM The energy E of a system is a function of the n generalized coordinates q.pE 'dq dp 2 .123d) .. Thus if the partition function Z can be calculated.. Equation (10. dpn (10. the calculation of Z is difficult and the integral in Equation (10.. qn. 10. Suppose that E can be written as the sum of two parts E = cpJ + E ' ( q i ... . . .— [inZ] (10.1016) cannot be evaluated.. p 2 .122a) = I e"Pcpi dp! I .1 l40b) is the equation of state for a perfect gas. for example. dpn (10.123a) 3(3 = — [ J .i n W + inZ'l 3(3 [2 ((3cj = JA inZ1 2(3 3(3 = ^kT + U' (10. p n ) where c is a positive constant. I e.. The partition function Z.123b) (10. it was easy to calculate Z since the atoms did not interact.
Thus the contribution to U of each quadratic term in E is ~ kT. they too will contribute — kT each to U.210. It is seen that. dp) about (q. 11). The probability of a molecule lying in a cell of size Aq Ap at (p. then its energy E. The probability of finding a molecule in a region (dq. from Equations (10.125) (10.852 ADVANCED MATHEMATICS Thus the internal energy (average energy) is the sum of the two terms y k T and U'.124) Consider a monoatomic gas in a container of volume V.132) . The contribution to U from cp^ is y k T and is independent of c. we do not consider a density gradient. Calculate the internal energy U of a harmonic oscillator. This result is known as the equipartition theorem.13lb) = ^ P 2 where p is the momentum. so U = kT 10. Interactions between molecules can be neglected. which is purely kinetic.kT to U.13la) (10. The total energy of the system is the sum of the energies of the individual molecules. the energy E is E = 1 m q 2 + l Gq 2 From the partition theorem each of the quadratic terms will contribute }. The problem of an oscillator was introduced in Example 10.12~l. If the mass of a molecule is m and its velocity is v. = Cexp(p 2 /2mkT) dq dp where C is a constant.21. p) is given by Prob.13 MAXWELL VELOCITY DISTRIBUTION (10. kept a temperature T. that is to say. Example 10. We assume that there is no preferred location in the container. (10. q) is independent of the generalized coordinate q . is E = \ mX2 (10. Similarly if other variables in Z appear as quadratic forms in E.
134) = 1 ^7 V(27tmkTr2 (10. We note that the Equation (10. be they organic or inorganic. dp) around (q. It is also known as a Gaussian distribution.133a) (10.137a) XM = ( v ) (^cmkT I 3 '' eXP (E 2 / 2 m k T ) (10. The integration of Equation (10. a biologist. . Problem 9b. drew attention to the movements of pollen suspended in a fluid. Some of his contemporaries attributed the motion of the pollen to the fact that they were alive. we deduce that C is given by c (10. Such a motion became known as Brownian motion. will move when suspended in a fluid. we use the normalization condition.SPECIAL TOPICS 853.134) into Equation (10. Robert Brown.133b).133b) (10.136) The probable number of molecules N p in a region (dq.132) can now be written as PrOb . we have I exp (p 2 /2mkTJ dp = (27tmkT) 3/2 Substituting Equation (10.833) and %M is also referred to as a normal distribution. the total number of molecules in V. It is certain that there is a molecule in the whole phase space.= V (2lif)3/2 eXp (~£2/ m k T ) d a dE (10.135) Equation (10.14 BROWNIAN MOTION In 1828.132) yields 1 = II Cexp(p 2 /2mkT) d q d p = C V l exp(p 2 /2mkT) dp where V is the generalized volume.137b) is similar to Equation (10.137b) The distribution function %M is known as the Maxwell velocity distribution and it is normalized to N. From Chapter 4. p) is given by NP = X M d a dE (10. It was later found that all sufficiently small particles. To obtain C. 10.
The mean value of K (t) over a suitable time interval is zero. Its present range of application extends from diffusion processes to the modelling of stock prices.142) into Equation (10. The probability of such an event is not zero. The equation of motion is m ^ = K(t)_F (10.144) . The particle is acted on by a rapid fluctuating force K (t) caused by the bombardment of the particle by the fluid molecules. but the combined contribution of the collision of several fluid molecules on the same particle at the same time could displace the particle. We shall consider only the Langevin approach to Brownian diffusion. The fluid molecules are in random motion and thus will occasionally hit the particles. In addition. the application of Brownian motion is no longer confined to pollen dispersal. Suppose that the mass of the suspended particle is m.= = K(t)6rcrav (10.143) dt — ~ ~ Forming the scalar product of Equation (10. It is usual to assume that the particle is spherical with radius a. As a result. yields dv m r » — = r »K 67trar • v ~ dt ~ —— ~ ~ (10. Exploring all the avenues that it has opened up is beyond the scope of this book.141). and the resistive force F_ is given by Stake's formula I? = 67tray where r\ is the viscosity of the fluid. the vector position of the particle.854 ADVANCED MATHEMATICS It is now known that the motion of these minute particles is due to the collision of the fluid molecules with these particles.141) where y is the velocity. The mathematics developed to provide an understanding of Brownian motion has led to the opening of a new field in mathematics known as stochastic processes. Brownian motion can be explained in probabilistic terms.143) with T_ . we obtain dv m .142) (10. One molecule by itself may not have the momentum to displace the particle. a resistive force F acts on the particle. Substituting Equation (10.
SPECIAL TOPICS The following results will be needed
851
f (r«r) = 2 r  / at ~ ~ ~ at
=2rv ~ ~ ~
(10.145a,b)
dv d2 ^(1'r) = 2r«^ +2vv Substituting Equations (10.145a to c) into Equation (10.144) yields r 2 1 — A r ( r * r )  2 w = r »K 37UTia— ( r  r )
(10.145c)
(10.146)
2 [dr
~
J
"~
dt
The time average of r • K is zero since r and K are uncorrelated and the average value of K is zero. From the equipartition theorem (Section 10.12), we deduce in the present case, where v has three components, that the average < y • v > is m < v « v > = 3kT Taking the average of all the terms in Equation (10.146) leads to (10.147)
4 < r . r > + ^ l i A < r . r > = 6M:
dt m dt
m
(1014 . 8)
The solution of Equation (10.148) is < ! • ! > = ^  ( « t  l ) + ^  +C 2 e ( x t ma a where a = (10.149)
—, Cj andC 2 are arbitrary constants to be determined from initial conditions.
In the limit as t —> °°, <£ • r > tends to lim<r.r>
t>oo ~ ~
=
6kTi
ma
(10.1410a) (10.1410b)
= i^1
We note from Equation (10.1410b) that <£•£> is proportional to the temperature T and time t and inversely proportional to the radius a of the particle and the viscosity r. Thus the larger the
856
ADVANCED
MATHEMATICS
particle the smaller the contribution < r • r > , that is to say, the distance the particle travels is reduced. That <£ • jr> is proportional to t has been observed experimentally and is also a result of the diffusion process. We also note the absence of the density in Equation (10.1410b). A modern survey on the development of Brownian motion in mechanics is given by Russel (1983). Lavenda (1985) has written an article popularizing the importance of Brownian motion. PROBLEMS la. In Example 10.21, it was shown that the trajectory of the system q + co2q = 0 in the phase plane is a closed curve. Sketch the trajectory of the system
•• 2
q 0) q = 0
in the phase plane. 2b. For a conservative system, the total energy is constant. This is expressed as l m ( q ) 2 + (Kq) = C By differentiating with respect to time, show that the energy equation can be written as mq = f(q) Express f (q) in terms of < (q). >  The points of equilibrium of the system are the points at which the acceleration is zero (q = 0). What can you deduce about the potential <) at the points of equilibrium? If m = 1 and C = 1, ( show that the equation of the trajectories is p = ± V2 — 2 j <> What happens when < > 1 ? > 
SPECIAL TOPICS
857
Sketch the function § and the trajectories for the following cases (i) < = (q)2, >  3 a. (ii) < = 1  i (q) 2 , >  (iii) < = (q)3, >  (iv) <) = I j
A particle of unit mass moves under a force F given by
F = &• r
where r_ is the vector position of the particle relative to an origin and K is a constant. If the motion is twodimensional and we choose a polar coordinate system (r, 9), show that the Hamiltonian H is given by
HI[W 2 + (£) 2 ]£
Obtain the r and 0 components of the equation of motion. 4a. Show that the transformation Q = q tan p , is a canonical transformation. 5a. Using the definition of the Poisson bracket given by Equation (10.45), verify Equations (10.46e to g). Liouville's theorem is given, in Bird et al. (1987), by Equation (17.2.2) and is written as P = i n sin p
6b.
at "
f dx{
(x t }
where the xj are the generalized coordinates and momenta. Show that in the case of a single particle, the equation given by Bird et al. (1987) is the same as Equation (10.610b) with p = f. Verify that a steady solution of Equation (10.610b) for a single harmonic oscillator is given by P
= PoH"
where p 0 and P are constants and H is the Hamiltonian.
858
ADVANCED
MATHEMATICS
7a.
Twelve identical balls are randomly distributed in a box. Calculate the probability of (i) (ii) (iii) having three balls in the first quarter of the box; having six balls in the first quarter of the box; having all twelve balls in the first quarter of the box.
8a.
A dilute suspension contains ten spheres per liter. A sample (one tenth of that solution) is removed. What is the probability that there is (a) at least one sphere, (b) at most one sphere in the sample. You may assume that the particles are distributed according to a Poisson distribution.
9b.
In Section 10.9, we have defined the mean value for a continuous distribution. The mean value (x) of a random variable which assumes values xj, X2, ... with corresponding finite probabilities f (xj), f (X2), ••• is given by
(x) = X x i f ( x i)
i
Show that, for a Poisson distribution, the mean value (x) is A,. 10a. There are 3 x 103 spheres in a box. Calculate the number of different combinations for the spheres in order to have 1100 spheres in the front third of the box. Assuming that the distribution is normal, calculate the probability of having exactly 1100 spheres in the front third of the box. Answer: 1O858 ; 8.8 x 10~6 lla. A polymer chain may be idealized by n + 1 beads, joined by n links, each of length a. r_ is the endtoend vector of the chain and we assume spherical symmetry. The probability density of the chain distribution f (n, r) can be shown to be
/,
?
f(n,r) = 7ia 2 n
X2
\3/2
exp
_3r2\
^\
'
2a 2 n/
where r = I r I. Calculate (r ), the mean value of r2. Answer: a 2 n
SPECIAL TOPICS
859
12b.
Assuming a Maxwell distribution, calculate the average values of (a) the xcomponent of the velocity v x ; (b) the speed v (= V vx2 + v 2 + v z 2 ); (c) the square of the velocity v 2 of a molecule in a gas at temperature T. Why is (v x ) not equal to ^ (v) ? Answer: 0 ; v ^^X ; ^kT
13b.
In Bird et al. (1977), the partition function Z for a macromolecular solid is given by
Z = /exp  £ 5£E(rvr )(rvr )
J eq
v, n ^ JV1
where J eq , H v ^ are constants, _rv and £„ are positions vectors of the junctions. Combining Equations (10.1111, 12), show that the Helmholtz free energy A is given by
A = kT^nfL) + 1 £ Hv^ ( £ v  r )  ( r v  r )
\ J eq/
z
v,(i
V<(X
Under what condition(s) is the change in A equal to the work done?
REFERENCES
ABBOT, M.B., An Introduction to the Method of Characteristics (Thames and Hudson, London, 1966). ABRAMOWITZ, M. and STEGUN, I.A., Handbook of Mathematical Functions (Dover, New York, 1964). AMES, W.F., NonLinear Partial Differential Equations in Engineering. Vol. I and II (Academic Press, New York, 1972). AMES, W.F., Numerical Methods for Partial Differential Equations (Nelson, London, 1969). ATKINSON, C , Quart. J. Mech. Appl. Math.. 41, 301 (1988). BALMER, R.T. and KAUZLARICH, J.J., AIChE J.. 17, 1181 (1971). BATCHELOR, G.K., An Introduction to Fluid Mechanics (Cambridge University Press, New York, 1967). BENTWICH, M., Chem. Eng. Sci.. 31, 71 (1976). BIRD, R.B., ARMSTRONG, R.C., and HASSAGER, O., Dynamics of Polymeric Liquids. Vol. 1 Fluid Mechanics. 2nd Ed. (John Wiley, New York, 1987). BIRD, R.B., HASSAGER, O., ARMSTRONG, R.C., and CURTISS, C.F., Dynamics of Polymeric Liquids. Vol. 2  Kinetic Theory (John Wiley, New York, 1977). BIRD, R.B., HASSAGER, O., ARMSTRONG, R.C., and CURTISS, C.F., Dynamics of Polymeric Liquids. Vol. 2  Kinetic Theory. 2nd Ed. (John Wiley, New York, 1987). BIRD, R.B., STEWART, W.E., and LIGHTFOOT, E.N., Transport Phenomena (John Wiley, New York, 1960). BLASIUS, H., Z. Math. Phvs.. 56, 1 (1908). BLUMAN, G.W. and COLE, J.D., Similarity Methods for Differential Equation (Springer Verlag, New York, 1974). BORN, M., HEISENBERG, W., and JORDAN, P., Z'Phvsik. 35, 557 (1926). BOWEN, J.R., ACRIVOS, A., and OPPENHEIM, A.K., Chem. Eng. Sci.. 18, 177 (1963).
7D. CARSLAW..R. D. L. EISBERG. R. Int. R. Higher Transcendental Functions.. F. 2nd Ed. Conduction of Heat in Solids. CRANK.J. II (Interscience Pub. Phil. Ann.. A. 46. CROWE. (ed. R. DE KEE. W. Oxford. EPSTEIN. 1193 (1994). F. Soc...G. 1953.F.. Tables of Integral Transforms. "The Temperature Distribution Within a Hemisphere Exposed to a Hot Gas Stream". Applied Phvs. L. and GRYTE. Free and Moving Boundaries (Clarendon Press.R.. 3 Vols. CHOUDHURY. Camb. MAGNUS. Vol. L. and CHHABRA.. D. Appl.. and NICOLSON. J. Mech. 1967). CHAN MAN FONG. and SOB EL. 47 (1983). Klamkin (Siam... CARREAU. J..J.F.. 1990). R.). . CHIAPPETTA. J.. ERDELYI. and JALURIA. 1984).S. Vol.. 1973). J. DE KEE. Philadelphia. J. Indiana. 43. 171 (1948). I. D. Physica. 3rd Ed. Berlin. S. COURANT. CESARI. 1987). OBERHETTINGER. M.S. edited by M. I and II (McGraw Hill... Polymer Rheologv: Principles and Applications (Hanser. Y. Heat and Mass Transf. P. Methods of Mathematical Physics. New York. and MARCOS.. 2nd Ed. C ... J. and TRICOMI. H. COURANT.. P. 111 (1993).M. 1966). 1955).. New York. Numerical Analysis . New York. and LEWY. A. DE KEE. 32 (1928).. New York.. J. Proc. Oxford. 1. (John Wiley. Mat. 1971). A History of Vector Analysis (University of Notre Dame Press. 40 (1993). Adv. R.C. C... 100. 1997). CHAN MAN FONG. 50 (1947).M. (McGraw Hill. and JAEGER. ELDEN. 37.. 60. B. 1954). 74.R. and WITTMEYERKOCH. (Clarendon Press. K.862 ADVANCED MATHEMATICS BURGERS.P. Asymptotic Behavior and Stability Problems in Ordinary Differential Equations.An Introduction (Academic Press. in Mathematical Modelling. CRANK. 1985). L.. Quantum Physics. D. (Springer Verlag. NonNewtonian Fluid Mech. New York. and HILBERT. New York. D. C.. and RESNICK. ERDELYI.. H.... FRIEDRICHS.
LEVENSPIEL. Lie Groups for Pedestrians (John Wiley. 252/2. in Chemical Engineering in Medicine and Biology. 70 (1985). H. Introduction to Numerical Analysis (McGraw Hill.. and DOUGLAS.. F. KAMKE. and YOUNG. GUPTA. 13. L.M. Sci. and DE KEE. and WHEATLEY. FERRARO. Electromagnetic Theory (Athlone Press.REFERENCES 863 ERICKSEN.C. 1971). 1994).. E. Applications of Green's Functions in Science and Engineering (Prentice Hall. 1956). 1972). and HESSELBERTH. MENKE.. R. New York. GOLDSTEIN. J. Physiol. LANCZOS. New York. Losungmethoden und Losungen (Chelsea Pub. GREENBERG. 1956). C. Fliigge (Springer Verlag. GONZALESNUNEZ. Hershey (Plenum Press. W. Ill/1.. GERALD.F. Reading. New York. 1959). A. University of London.. Applied Polvm. Classical Mechanics (AddisonWesley. 1981). edited by J. O. HAGEMAN. 1952).J.C. FA VIS. V.. 1972). HODGKIN. 62.. edited by S. New York... Chemical Reaction Engineering (John Wiley. 883 (1967). . V. J. HILDEBRAND...D. LIPKIN..J. D.H..M. "Tensor Fields". C... "Oxygen Diffusion Coefficients for Blood Flowing Down a WettedWall Column". H. and JEFFREYS.G. LAVENDA. Toronto. D. 117. JENSON. 1963). C. D.. in Handbuch der Phvsik.V. H.P.L.F.... MILLER. Co. Scientific American.F. Applied Numerical Analysis.D.F.A. J... 1960). New York.. London. M. Dictionary of Conformal Representations (Dover. 5th Ed. 1965). and HUXLEY. Differentialgleichungen. R. CHAN MAN FONG. Reading. New Jersey. AIChE J.. New York. G. J.J. 1967). KOBER. Massachusetts..L.. A. HERSHEY.O. B. Masschusetts. (AddisonWesley. C . New York. Mathematical Methods in Chemical Engineering (Academic Press. The Variational Principles of Mechanics (University of Toronto Press. Applied Iterative Methods (Academic Press. Vol. P.. New York. 1966). 1627 (1996). V. Berlin. 117.A. 500 (1952). B.
and MORTON... G.S.. Chem. ROUX. J... New York 1953). RUSSEL. PEARSON. N. C. MORSE.M. A.J.. R. Computers Chem. H. New York.. 169 (1983). Theorie des distributions (Hermann. J. P.. P. W. L. 44. (ed. (Interscience Pub. KOLACZKOWSKI. and CRUMPTON. Fluid Mech. and EISENKLAM... R. Theoretical Hydrodynamics. Eng. Sci. and AMUNDSON. H. P. RHEE. J. 1963). STASTNA. Perturbation Methods (John Wiley.). Commun. L. L.. . Eng. 1965). 508 (1985).K. 107. 17. SEDAHMED. SPENCE.. 5th Printing (Macmillan. Rev..... B. 111 (1991). 64. Chem. 7D.. Paris. and FIX. Annual Rev..C. Vol. Eng. L. G.H. DE KEE. Oxford. D. and KWONG. I and II (Prentice Hall. Chem. 57 (1983). and FESHBACH. Ordinary Differential Equations and their Solutions (van Nostrand. ROSENHEAD.M...J.T. Difference Methods for Initial Value Problems. SLATER.D. Chem. 2nd Ed. First Order Partial Differential Equations. 1974). Can. G. 1973). REDLICH. D. 1960).M. WORTH.864 ADVANCED MATHEMATICS MERIC.. 38. Handbook of Applied Mathematics (van Nostrand. Heat Transfer. Eng. A..W.. G. New Jersey.. 13. 1986). RICHTMYER. 1957).. A. ARIS. 233 (1949).... 110. SCHWARTZ. New York. 425 (1981).E. O. Methods of Theoretical Phvsics (McGraw Hill. NAYFEH. Phvsica.. 1967). SHEPPARD.N. 1960).R. Cambridge. STRANG.I. MURPHY. New Jersey. 75 (1986).. S.R. MILNETHOMSON. K. Laminar Boundary Layers (Oxford University Press. New York... An Analysis of the Finite Element Method (Prentice Hall. Confluent Hypergeometric Functions (Cambridge University Press. New York. 1973). J. R. J.H. New York. 1057 (1993).J. and HARRISON.
W. Advances in Catalysis. J. TAYLOR.. and FORTUIN. 501 (1934). New York. 1975)..D. (Cambridge University Press.R. G. and PRATER... Stanford. J. D. in Handbuch der Phvsik. ZIENKIEWICZ. WILKINSON. Gaussian Quadrature Formulas (PrenticeHall. K. and BIOT..N. A146.. M. O.H. J. Vol. New York. III/3. PAULUSSEN. Perturbation Methods in Fluid Mechanics (Parabolic Press. New Jersey. O.. J. WILSON. T.. Chem.. Soc. The Algebraic Eigenvalue Problem (Oxford University Press. Cambridge. W... edited by S.. A. 13. Partial Differential Equations of Applied Mathematics (John Wiley. 1983). and MORGAN. Roy.J. Berlin. 1965). ScL 39. R. Fliigge (Springer Verlag. VAN DYKE. S. New York. Oxford. ZAUDERER. Eng. "The NonLinear Field Theories of Mechanics".REFERENCES 865 STROUD. Mathematical Methods in Engineering (McGraw Hill. 1966). 207 (1979). A Treatise on the Theory of Bessel Functions. 2nd Ed. WIMMERS.. TRUESDELL.P. Proc. 1965). 203 (1962)..M.C. 1966). C. WASOW. CD. WEI. 22. Heat and Mass Transf. and SECREST. G. Int. WATSON. and NOLL.... 1415 (1984). . E. 1965). 1940).H. M.I. VERMEULEN.. New York. 1983). VON KARMAN. Finite Elements and Approximation (John Wiley. Asymptotic Expansions for Ordinary Differential Equations (Interscience Pub. D.
.
A.I3) .APPENDIX I THE EQUATION OF CONTINUITY IN SEVERAL COORDINATE SYSTEMS RECTANGULAR COORDINATES (A.(pr2vr) + .J —  r (Pv^) = 0 dt r z dr rsin6 30 r sin9 d() v (A.I2) SPHERICAL COORDINATES ^.+ \  .+I:(prVr) + I_(pv9)+_(pVz) = 0 (A.I1) CYLINDRICAL COORDINATES .(pve sinG) + .? .
868
ADVANCED
MATHEMATICS
APPENDIX II THE EQUATION OF MOTION IN SEVERAL COORDINATE SYSTEMS
RECTANGULAR COORDINATES In terms of velocity gradients for a Newtonian fluid with constant p (density) and n (viscosity) xcomponent
/3v Y dvY 3v Y 3vx\ p h r 1 + v Y  v 1 + v v =rJL+ v_ TT^I dp +U 3 vY 3 vY 3 vY *• + ^ + ^ +PgY
/ A TT
_ (A.II1)
\dt
x 5x
y a
y
zaz/
ax
ax2 ay2 az2
x
ycomponent
9 9 9
/av v \at
zcomponent
avv
x ax
avv
y dy
3vv
Z5z
ap ay
3 v v 3 v v ? vv
+ + + pg,,
y
P \=r + v v ^r1 + v v  ^ + v_ =* = — ^  + 1 1
(A.II2)
\ a x 2 a y 2 az2
/3v_ dv7 3v7 3v_\ ap d v d y d \ + + + pg v p h r +vY ^r^ + vv ^z+ v7 ^r^ = —  +U
9
9
9
(A.II3)
lat
x
^
y a
y
z d z
l
dz
ax2 ay2 az2
z
APPENDICES
869
CYLINDRICAL COORDINATES In terms of velocity gradients for a Newtonian fluid with constant p and \i rcomponent 3vr
9vr
r
vfl 3v_
VQ
3vr
v —L
— L + v — L + — —t —2.+
at
dr aP
ar
r ae a /i a
ar \r dr
r
z
az j 2 aVfi a 2 v
r2 39 3z 2
(A.II4)
\ 1 a2Vr
j r2 a e 2
9component
KI
at
r
3r
r 30
r
z
3z I
i a vfl
2 avr
a vo
+pg e
+—f+T +~f r 2 a e 2 r 2 ae a z 2
zcomponent
(dvv dv7 v fl 3 v 7 3v7\ n — + v — H—— —— + v —— I — — p(at r ar r ae z azj
<An5)
dp
= ^+H
i a / 3v 1 i a 2 v a 2 v
r— +  + +Pg 7 (A.II6)
3z
[ r a r l dr) r 2 dQ2
aZ2j
*
8ZQ
ADVANCED
MATHEMATICS
SPHERICAL COORDINATES
In terms of velocity gradients for a Newtonian fluid with constant p and fi rcomponent 9v r P\ dt 9v r dr vA 9v r r 50 9v r vo + v? 1 rsine d<> r I
VA
r
=
JjL+ii v V ^  T ^ — v e c o c e  ^  ^ + pg,
8r r2 r 2 90 r2 r 2 s i n 0 d(f> /
( .) An 7
9component
(
^e
3t
+ v
r
2
\
^Ze . Xe?Ze ,
9r 1 9p r 99 r 99 1^2 \
v > 9ve, d
Ve
r
v d) c o t e
rsine 9<)
r / 2cos9 9v A r 2 s i n 2 9 9^ (A.II8
2 9v r vfl r 2 99 r 2 s i n 2 9
(^component
I dt
r
9r 1
r 90
r sin 9 9(j)
r
r
I
/ATTm g
=
9p I 2  + WV v rsin9 9(() \
r
VA 2 9v 2cos9 9v fl \ f + 1+ _ ^+ p r 2 sin 2 9 r 2 sin9 d<\> r 2 sin 2 9 dty)
(A.II9)
Vs = i i r ^ 2
r 2 9r
+
J— ±
r 2 sin0 99
Sn i9
^ +
90
!
^
(A.II10)
9r
r 2 s i n 2 9 9<j>2
APPENDICES
871
APPENDIX III THE EQUATION OF ENERGY IN TERMS OF THE TRANSPORT PROPERTIES IN SEVERAL COORDINATE SYSTEMS
For Newtonian fluids of constant p (density) and k (thermal conductivity)
RECTANGULAR COORDINATES
 (dT dT oT dT\ . o 2T d2T d2T p C n 3— + vY ^r— + v v 3  + v 7 ^— = k + +
plat
*3x
L\ox/
ya
\oy/
y
dzj
\azj J
d x
i
9y2
dzij
n 1^VY\2 (^ V V\ /dv7\2 (3v v 3v v +2 ^ 1 ^ + 3 ^ + 3 ^ +\l —i +—i
l\dy
dx
/8v Y 9v_\ 2 3v v + Ui +3^ + ^i
\ dz dx / dz
3v 7 +3^
ay
(A.III1)
CYLINDRICAL COORDINATES
r

3T P at
r
3T dr
v e 3T r 80
dT
Z3z
1 3 f 3T ^ r 9 r \ dr
r2 9 0
1 32T 2
92T ^2
2
n i(d\\2 l\or/
~W 3v fi [r\o0
\ f PV7\2\ r IJ \ d z M
/f dV fl H 3z
! 3 v z f ( dV 7 d V r r 36 J \dr dz
(A.III2)
872
ADVANCED
MATHEMATICS
SPHERICAL COORDINATES  (dT 9T v e 9T VA 9T\ . [ 1 9 25T\ 1 pC n U— +v r ^— + — ^77 + — r ^ r ^rr = k —— r z — +— 9 . . 9T sm6
1
32T
.
/9v_ L\3r
n2r
/l 9v e \r 90
v r
/
1
3VA
v r
v e cot8 r
2\
r 2 s i n 2 9 3()2
r
\rsin6 dty
]2 r
9 /v e \
1 9v_
1
9v
9 /v<a
sinG 9 / VA \
\
9vfi
[ 9rvr;
r 36 J
[rsin6 d<\> 9 r v r ;J [ r 9 6 \ s i n e i rsin9 3<)J 
(A.III3)
APPENDICES
871
APPENDIX IV THE EQUATION OF CONTINUITY OF SPECIES A IN SEVERAL COORDINATE SYSTEMS
For constant p (density) and J9^g (binary diffusivity)
RECTANGULAR COORDINATES
^ A + f V x ^ + v ^A+
^A)=JS)
^
+
^£A
+
^
+R
(A.IV1)
CYLINDRICAL COORDINATES dcA I dc/, 1 9cA
e
9c A \
z
1 d
9c A \
1 3cA
f2
3 cA
3t
[
r
a r
r36
dz J
A B ^ ^ j
^
^
j
A
(A.IV2)
SPHERICAL COORDINATES 3c A / 3cA 1 dcA 1 9c A \
„
f i 3  2 3c A \
r2arl
i
a /.
oacA\
I
a2cA]
dr I r 2 sin9 3 0 \
3 9 / r 2 s i n 2 6 3() 2 (A.IV3)
L. 865 Von Karman. 865 Wheatley. 863 Levenspiel. 865 Lightfoot. 826. 182. 863 Kauzlarich. A. 864 Nicolson. C. 863 Jenson.... 861 Acrivos. R. 863 Gryte. 862 Jeffreys.G. J. 727.... 864 Fix. 861 Biot. 864 Resnick.K. P..M.. 413. 686.T.. 156. 45. 611. H. 551. 864 Stroud. S.. C. J. 862 Spence. W. W. 861 Koberi H 292. 570.. 856.. 865 Watson. 861 Ames.. 864 Slater. 185. P. P. G. CD. 490. 262. 390.W.P. P. 863 Kolaczkowski.. 861 Stewart.. 69. M .S.. 481.R. 862 Hildebrand.. 458. 647. J. 861 Strang.M. 864 Curtiss. 864 Stastna. 561.. B.. 864 Sheppard. 126. 861 Amundson. 861 Kamke. 863 Eisberg... O. D. 700.A. 861 Courant. 863 Jaeger. R.R. 686. 647. 718... 437. 551. W. LA.. 861 Born. 637. V.I.P. W.. C. 857. 768. 865 Bird. 327. 864 Morton. 560. 865 Sedahmed. 142. 19.. D.. 819. 394. G. 397.. 390. 863 Hilbert.L. P.B.. 864 Hassager. 289. 563. W. O. 651. G. 862 Crowe. R. 862 Choudhury. 864 Stegun. L. C... 861 Balmer... 864 Douglas. W. L. V. 718.. 531.... 862 Rhee. 863 Wilkinson..H. 862 Erdelyi.E.C. B. 19. 563. P. 861 Abramowitz... 865 Wimmers. 863.. 864 Young.. J. 182. 722. 450.. 156.. 861 Bowen. 142. 864 Aris.M. J.M. 864 Richtmyer. H. 862 Crumpton.F. D. 862 Ericksen. H. 863 Harrison.F... 863 Feshbach. 674. R. 862 Carreau. 862 Chan Man Fong.W... G. 668. 863 Magnus. M. W.. 863 Greenberg. M. 862 Cole. 169... I. 179.L.. 528. 865 Zienkiewicz.. 551. 863 Eisenklam.. 863 Jordan. 864 Roux. J. J. O.. 212. G. 864 Russel.F.. G. 437. 454.J.O. J. 864 Schwartz. E. J. 169. A.. W. 862 Gupta. 85. 142. 864 Kwong. 126. A..C. 656. 505. L.M. 863 Hesselberth. 865 Wilson. J. 495. 531.. ST.M. L. M. 176.R. 863 M e r i C . 561.D. 479.. 861 Bluman. 391... C . B. 142. 862 Truesdell...J.J.C. F. R> 7 9 9 .. 864 Elden.P. 863 Hageman.E.N. M. 687.. 864 Prater..C. 865 Vermeulen.. 173..K. K.. 861 Hershey.. 560. 686. 709..S. 551. 179. J. F. M. 570.M. F. 865 . A. 864 Nayfeh.. 113. L. H. 863 Lewy. 551. P.. 862 Menke. 863 Hodgkin. 863 Goldstein.. C. 862 Gerald.J.R. 531.. 862 JaJuria. 864 Sobel. 863 MilneThomson. 865 Oberhettinger..I. 865 Tricomi. 865 Wei. 531. 863 Lavenda.M. E... 413. L. R..H. 865 Redlich.H. 865 Van Dyke.N. 864 Armstrong. R.A. 861 Lipkin. 674. 826. 826.F. K..H. K.... 862 Oppenheim. 734.J. V.. 864 Fortuin. R. 563. A.C. 182. 495. 685. 523. 182... 523.R. 865 WittmeyerKoch. J. R. R. C .D... 411. A. 862 Cesari. 551. Y... 861 Batchelor. 113. 461.. 536.D.J. A. G... 570.K. 861 De Kee.H. R. 864 Secrest. A. 864 Lanczos. 126.AUTHOR INDEX Friedrichs.. 559. T. S. 637.D. 863 Ferraro. 8 6 4 Miller. 862 Epstein. 821. D. 722.. B. L. 341. 500... 657.. 551. 64. 862. 700. 543. 865 Taylor. 861 Atkinson..R. 796. O.... 865 Morse. 862 Noll.... 859. 861 Bentwich. 862 Chiappetta. G. G.. 862 Crank.R.B.. D. 185. 734.J. 864 Morgan. H 862 Abbot.. 862 Marcos. 396. 861 Paulussen. C .N. 525. O. 551... 863 Favis. J.V. C. H. A. 863 Zauderer. D.J. 863 GonzalesNunez. 785. J. 381. 864 Rosenhead. 783.... 395.F. H.J. 551.. 865 Wasow. M.. R. R.C. 864 Murphy. 862 > 8 6 3 Chhabra. 722. M. 505.F. D. D. 505.. 169. 523. 863 Huxley.. 625.. 861 Burgers. E. 826. L...J..D. 862 Worth.G. 807. 169. 861 Heisenberg. N. 862 Carslaw. 263. 861 Blasius. 856. 865 Pearson. 727. 164... 651.
.
191 variables. 13 total. 347 generalized. 226 Dummy index. 431. 624 Binomial distribution. 220 number. 42. 16 substantial. 527 Boltzmann constant. 362 Composite solution. 529 Bisection method. 331. 193 spherical polar. 812 Cycloid. 303. 844 Boundary layer.809 polar. 512. 571 Blasius flow. 665. 830 canonical. 739 Argand diagram. 90. 676 Adjoint operator. 686 in the mean. 403 equation. 211 roots.SUBJECT INDEX AdamsBashforth method. 626. 192 Argument. 697 Distribution binomial. 297 function. 769. 222 integral. 511 Christoffel's symbol. 167. 51. 429 integral formula. 853 Maxwell. 683 divided. 233 mean value problem. 39 Bessel's equation. 3 covariant. 837. 349 Courant parameter. 845 Gaussian. 275. 121 modified. 367 Coefficient matrix. 193 Fourier series. 199 smooth. 525 Boundary value problem. 547 Contour. 853 Catenary.411 Average value. 344 Dirichlet's problem. 227 CayleyHamilton theorem. 266 residue theorem. 646 AdamsMoulton method. 200 simply connected.387 method. 194 Derivative. 303. 341 Bernouilli's equation. 683 Diffusion equation. 170 Augmented matrix. 422 curve. 202 partial. 93. 347. 357. 637. 709 CrankNicolson method. 584 Combination. 828 Comparison test. 827 Conformal mapping. 357 product. 585 Autonomous system. 754 D'Alembert's paradox. 178 Condition number. 659 Brachistochrone problem. 435 Dirac delta function. 852 normal. 198 integration. 614 finite. 176. 379 Domain dependence. 395 general curvilinear. 340 Auxiliary equation. 683 central.374 theorem. 356 physical. 306 Cyclic coordinate. 347. 204 theorem. 221 simple. 345 Dyad. 4. 842 Bendixson's negative criterion. 139 Best fit regression line. 141 properties. 306. 686 Complex conjugate. 778 Cauchy condition. 686 Cross ratio. 338. 434 multiply connected. 375 Curve closed. 450. 177. 191 Components. 357 . 192 potential. 626. 577 Convergence. 347 contravariant. 830 Birkhoffs method. 204 Arc length. 681 forward. 834 Divergence. 14 of a complex function. 138 third kind. 483 De Moivre's theorem. 92. 548 Analytic continuation. 853 Poisson. 354 mixed. 191. 590 Conditional probability. 4 principal value. 454 Difference backward. 845 factor. 222. 162. 199 orientation. 136 generating function. 285 Curl. 153 function. 240 Compatibilty. 193 Associated variational equation. 134 first kind. 42. 226 of a function. 812 cylindrical. 18 Characteristic. 120. 141 Bessel's functions. 92 Convolution. 140 second kind. 367 directional. 146 radius. 163 solution. 327. 753 Brownian motion. 397 elliptical. 186 integral form. 663 Asymptotic series. 64 Direction cosine. 186 orthogonality. 223 Contraction mapping. 532. 257 Riemann conditions. 1 of an open set. 388 Chain rule. 47 Coordinates cyclic. 263 solution. 352 covariant. 382.
769. 276. 144 solution. 305 principal value. 6. 146 finite cosine transform. 102 Function. 576 iteration method.441. 191 Indicial equation. 338. 396 Fourier Bessel series. 580 Householder's algorithm. 601 Hyperbolic functions. 596 Jacobian. 217 meromorphic. 415 Inverse point. 158. 99 formula. 595 Jacobi's identity. 811 velocities. 811 interpolation. 191. 331. 303. 638 improper. 202 entire. 837 transform. 600 Eigenvector. 474 transform. 475 finite sine transform. 124 Eigenvalues. 740 determinant.374 Laurent's series. 265 indefinite. 820 harmonic. 211. 740. 7 definite. 437 limit theorem. 176 Integral. 704 free energy. 447 Galerkin's method. 8 equation. 317 Jordan's lemma. 13 shape. 424. 748 method. 200 confluent hypergeometric. 326 Integrating factor. 842 microcanonical. 2 range. 122. 814 Hankel's functions. 639 theorem. 278 Fundamental solution.532 Ensemble. 19 in. 289 Jump discontinuity. 323 triple. 842 Enthalpy. 209 homogeneous. 3g9 Iterative method. 724 Gamma function. 338. 213 Lagrange equation. 36. 635 Generalized coordinates. 379 two point formula. 327. 269 Joukowski's transformation. 21 hyperbolic. 1 rational. 387 trigonometric. 463 eve"' ! explicit 19 exponential.511 Illconditioned matrix. 31 Extreme values. 811 Laplace's equation. 527 Finite difference. 585 Seidel method. 387. 847 Entropy. 532 theorems.559 conjugate. 742 Falkner Skan equation. 1 analytic. 204 complex. 338 Equipartition theorem. 818 method. 122. 600 Einstein summation convention.477 Leiendre series 446 series 144 146' 442. 99. 611 method. 598 theorem. 846 Equation of continuity. 589 Imaginary number. Laplacian. 212 tensor. 769 Generating function B e s s e . 21 Exact differential equation. . 569 truncation. 681. 745 Lagrangian. 313 surface. 411 multiplier. 852 Error roundoff. 57. 723 stream. 422. 246 . 98. functions> 1g6 for Legendre polynomial. 387. 216 implicit. 129 G l b b s f r e e ener gy> 8 4 8 Grad > 3 0 2 > 3 7 3 ADVANCED MATHEMATICS Hermitian matrix. 139 transform. 180 Kronecker delta. 815 principle. 32. 749 Gauss elimination method. 122. 266 repeated. 7 line. 601 transpose.681 elements. 146 Kerne1' 4 9 5 for Green's functions. 331 Hamilton's equations. 552 Helmholtz's equation. 216 partial differential equation. 144 partition. 209 continuous. 478 ' ' Frobenius method. 760 Isotropic tensorvalued function. 460 .809 momenta. 643 Homer's method. 2. 722 Fixed point. 768 Hamiltonian. 102 Initial value problem. 21. 576 Four roll mill. 314 volume. 846 bJ Hermite's equation. 89. 124. 10 several variables. 823 canonical. 46. 257 odd> j orthogonal. 637 Inner product. 345 Elliptic partial differential equation.878 Eigenfunctions. 665. 483 Heaviside step function. 444 coefficients. 475 integral. 215 Functional. 535 Isoperimetric constraints. 198. 769. 49 Heisenberg's uncertainty principle. 601 Heun's method. 135 Gateaux variation. 179 polynomial. 125 Lagrange's equations. „„„ „ „ . 844 periodic. 511. 769. 213 gamma 135 generating. 570 Euler equation. 204 erfc'463 error. 125.
450 quasilinear.. . 631 Singular matrix. 125 Ordinary differential equation. / ' . 4 1 . 2 5 4 R i e m a n n ' s function. 254 ™i = . 241 Ratio test. 238 divergence.. .. . Neumann s function. 1 3 2 . or. 239 divergence. . . 89. 287 „ . 511 R o b i n ' s problem. 238 convergence. 2 4 0 RayleighRitz method. 671. 527 Stefan problem. 423. t.™ Shooting method. 541. 128 second kind.. .„ _ ._„ Lipschitz s condition. . „ „_ Saddle point. 121 associated. ..511. nn.SUBJECT INDEX Normalized functions. 25. 19 i H 19 generalized. 819 . 190 Selfadjoint. 1125. . 686 Stagnation flow. Melhn s transform. . 580 !>„*„*• T o n Potential.„„ . . 580 N 145 590 or ™' ' P' L . /4i Mixed problems. . ^4n R19 p. 638 F . . I. . 659 Similar matrices. 239 Set 2QQ Least square approximation. ' T. 95 „ . 25. Liouville s theorem. A~ orv particular solution. . s i z Picard's method. 254 Spline. 551 Quasisteady state approximation. 561 Permutation. . 89.^ x. 396 . 144. 275 conformal. . 38. 523 variable. 173 Singular point. . 648 homogeneous. 401 normal f lfi4 879 Radius of convergence. . _„_ Secant method. 4. 826 J first order. ' .. . Parabolic partial differential equation. 173 ReinerRivlin fluid. 178 Maximum. 24 „. • „ . 89 .^ cubic. m oc/i order m.301 _ . 25 first order. ' Rombergs method. . 25 . Orthogonal bases. 128. 129 orthogonality.806 > . 541 ' „_ . 129. . . Mutually exclusive. . J4U. 7 2 4 . 239 absolute convergence. . A r. . 131 Rodrigues formula. 92. „_. 144 . 832 Stochastic process. series solution. 828 . . 619 ^ ^ ^ ^ ^ U4 Stability. T •. . 341 f. . * T° RungeKutta methods. 572 Secular term. functions. 91.36 nonlinear. n R QR a gonthm 608 Quantum mechanics. ^ 91 isolated. \ * . 725 Stirling's formula. . 130 transformation. 614 method. 129 Regular perturbation. 545 ™0h™ S t "P' 3 2 2 Modulus 193 Moreras theorem. 6. 825 . 402. Newton's divided difference.539 pole . 144 .277 isogonal. . 814 . 549 Sequence. removable.no Matching principle. . 5 4 5 Rodrigues formula. 237 . . V . 7 8 5 Recurrence equation. ™.n T Loss of significance. Plateau's problem. 600 Similarity solution. 639 Planck's constant. „ . 1 8 0 _. 9 2 . 552. 126 Legendre's polynomial.. 774 Pointwise approximation) 145 Poisson's bracket. . . 43.' 591 • . . 98 higher order. 90. 745 n . 523 Simpson's rule. 638 general solution. nAA Maclaunn series.. . 341 i « T >u.. T.. 741 Maximum modulus principle. . . 6. Lenz slaw.1?1. 212 Power law fluid. . . 239 convergence. 96 Ordinary point. 484 Minimum b. 630. 129 recurrence formula.523 Parameter expansion. 238 Series. 146. 254 Polynomial. SchwarzChnstoffel o_ r transformation. „. 512. 176 „ . . 96. 420 path independence> 2 2 7 D ?£. 450.. 253 . o_.. 619 . 585 perturbation. 402 nonhomogeneous. 551 Residue... 89 lineari 25. 173 Quotient law.. . . 633 _. Schrodmgers equation. . 5 .. 92.. Outer solution. 402 homogeneous. 26. Scalar product.„ 0<.. 4 f.. . t l u l ^ ' 8 " ' 8 Z Mean value theorem. . „.99 . 4 1 .. 817 ^ ^ ^ ?03 integral formula. 212. L Hopital s rule. 5 5 8 functions. __. 76' I . ^^.. . 237.. H on eo „.„„ matrix. 71 p o w e r serieS) ^ 241 . 391 Relativity. 552 t t „. 128 associated. Penetration. Leibnitz s rule.™. 170 P a r t i a l d i f f e r e n t i a l equation. 450. 278 ^ P .. 574./l simple. . '. • r. 623 Legendre's equation. problems. 401 degree. 643 _. 402 second order. . 102. . 277_ Matched asymptotic expansion.. . 338. 619 functions. rnase plane. . •. 528 Stiffness matrix. . 569 5 . . f° O .•• i. 238 Maxwell's equations. 558 generating function. 244 Mapping. 854 irre gular. Lie algebra. . regular. 5 1 3 method. _ . 600 ~ .
335 contravariant base. 370 symmetric. 22. 322 smooth. 358 objective. 118. 279 reciprocal. 384 Joukowski. 301 arbitrary order. 570 simple. 323 one sided. 37. 322 simple. 354 isotropic. 255 order m.££Q Stake's flow. 601 Trigonometric functions. 439. 5. 724 WeissenbergRabinowitsch equation. 357 Transformation bilinear. 427. 384 permutation. 359 RiemannChristoffel. 255 ADVANCED MATHEMATICS . 322 Taylor's method. 629. 699 Surface. 640 theorem. 212 SturmLiouville problem. 723 Vector. 211 Wave equation. 819 ColeHopf. 505 frame of reference. 333. 242 Tensor. 301 conservative. 585 Van der Waals' equation. 120 Successive over relaxation. 77 Variation of parameters. 390 Stream function. 287 von Mises. 320 Unitary matrix. 170 theorem. 129 Variational method. 552 Weak solution. 349 covariant base. 302 Vorticity. 282 SchwarzChristoffel. 450 function. 380 Stokesian fluid. 176 Zero. 78 Well posed problem. 449 Whitehead's paradox. 382 metric. 601 Upper triangular matrix. 631 Tridiagonal matrix. 757 Trapezoidal rule. 719 Transversality condition. 289 linear. 349 irrotational. 215 Unit normal. 335 product. 320 integral. 58. 284 canonical.
dx } f ) cosh x dx = sinh x J ' [cos(p + g)x 2[ P+q t n n J cos(pq)x pq JL (cos x) = sin x jL (tan x) = sec2 x f J f CQspx CQsqx dx = X [ S i n ( p .q ) x _ sin(p + q ) x  2[ pq P+q J + J .e " ^ [arc sinh (x/a)] = I /\/a 2 +x 2 .[arc tanh (x/a)] = a / ( a 2 .cos x dx f sin x dx = cos x I sec2 x d x = tan x sinpx sinqx dx = L J f sin px cos qr..12 nt dx = —"^r Vax dx 2an +l f°° _ (a x 2 + p/x 2 ).sech 2 x ' A [arc sin (x/a)] = l / V a 2 .' > f sin"" 2 xdx n J d.a r c cos (x/a) f00 Jo = J. d. r— 47^ & 2 Jo = .STANDARD DERIVATIVES AND INTEGRALS = inx n ""' x sin x cos (n 7<*>*) = 7 dx x ( c o s x d x = sinx ( cos n x dx = + ~ 1} f cos"" 2 x dx fldx j x f sin"xdx = .(sin x) . — (cosh x) = sinh x ' j" s e c h 2 x dx = tanh x " cos (px + q) dx = f% 2 [a cos (px +q) + p sin (px + q)] ' a +p f e a s sin (px + q) dx = ^ ! 1 _ [ a s in (px +q) .x 2 ) dx J x 2 = 1 arc tanh (x/a) a .i [arc cosh (x/a)] = ± I / V x : . e dx .yj.x 2 f = arc sin (x/a) dx dx J f°° _ax2 .a2 j & = ± arc cosh (x/a) 2 2 dx Vx a  dx 2 a d.(sinh x) = cosh x I l [sin(p + q)x sin(pq)x dX i sinh x dx ) _ cosh x e 2 [ p +q P  q .[arc tan (x/a)] = a/\/a 2 +x 2 i nr 2I^B" dx I dx J ^T^T a = arc sinh (x/a) Jo e dx = i.p cos (px + q)] J a + p JL (tanh x) .i [arc cos (x/a)] = 1 /Va 2 .x2 [ J .arc tan (x/a) C = 2n + i x 2 V« ax2.c o s j "~' x sin x + ( " .
the equation can be made to be exact after multiplication by an integrating factor I(x) = eteOOdx. First. p) of the system a. y) are homogeneous polynomials of me same qegree.— — . y) dx + N(x. The equation will reduce to a linear firstorder equation. \ay ox I we take 'M = c t3M dm Jx0 ( . y) = a.. . b) If j — " =£ but  3 y . are real and distinct: z ' a x a x y h . e a i x + c 2 x e a i x iii) a. based on the.y) dy = 0 dM _ 3N dy ~ dx X Secondly. That is: yP = c 1 ( x ) e a i x + c 2 ( x ) e a 2 * Substitution of the proposed y P and its derivatives into the equation to be solved. b. . .x + b]y.y) dx + N(x. M(x. .y + c. °y 3M dN /dM d N W . y) and N(x..TYPE OF EQUATION Separable FirstOrder The integrating factor I(x) = e ^ x ) dx dv TYPE OF EQUATION Linear FirstOrder j. . . M(x. which involves: .X + b. introduced by the proposed y p . The solution is then given by: y(x) = G x Similarly. G has to satisfy a number of conditions outlined in §1.inverting the transform.= N to compute f(y). + fiy = Q ( x ) M(x. one proposes a solution y p . = 0 a2x + b 2 y + c 2 = 0 iii) replace X and Y in the solution of (i) by xa and y~p.taking the Laplace transform of all terms in the equation .y) = c can be determined as follows: dF .. ./ M = h(y) (a function of y only). . =ot2 yh = c . . solve y" + Ay1 + By = 0 to obtain yh. n) determine the solution (a. The characteristic equation or + Aa + B = 0 has two ^ so)utions: i) ai and a.cie ' +C2e 2 ii)ot.x + b. Reducible FirstOrder generates a separablefirstorderequation.= u + x^= dx dx M(x. The Homogeneous FirstOrder ^ _ _ _ _ _ Bernouilli's Equation ^ + P(x)y = Q(x) y n dx SecondOrder Linear wi(h „+ A y . This generates afirstorderseparable equation of the form dz = f(z) dx Total or Exact a) The solution F(x. y P can also be obtained by the method of variation of parameters. .form Q(x). P^ ^ P(X)y = g ( X ) left side s n o u i d I METHOD OF SOLUTION M(x) dx + N(y) dy = 0 dx + ^ d(Iy) dx I METHOD OF SOLUTION ^ Straightforward term by term integration f r J M ( x ) d x + J N<y> d y = ° ru \ Multiply the differential equation by l(x) and integrate. b) I "2 b 2 Introduce a variable z = a. are the boundaries at which y is given.Y dX aOC + b2Y . if U . The solution of the problem is y = yh + yp.solving the resulting algebraic equation . . „ .y) = a2x + bjy + c 2 0 solve the homogeneous firstorder equation ^ = a.«)fw* where x o andx.x + b. d ^. That is: d _ d ^(IMJ^dN) Green's function method for nonhomogeneous equations involves finding the Green's function G. .18.^ ) / N = 8 ( x ) Other methods of solution are: i) the Laplace transform method. and o^ are complex conjugate roots. allows for the determination of c ( and c 2 . F = J M dx + f(y) substitute into =r. ij) the (a function of x only) then. associated with the differential equation L(y) = f(x). y) dy = 0 The substitution y = ux and consequently . . a) ll b\ * ° N(x.y + c.. Let v = y'". First divide both sides of the equation by y". (a + ib) and (aib): yh = eax(c3cosbx + c4sinbx) I a.
This action might not be possible to undo. Are you sure you want to continue?