This action might not be possible to undo. Are you sure you want to continue?

**PARTIAL DIFFERENTIAL EQUATIONS
**

(Second Edition)

**An Introduction with Mathernatica
**

and MAPLE

This page intentionally left blank

PARTIAL DIFFERENTIAL

**EQUATIONS (Scond Edition)
**

An Introduction with Mathematica and MAPLE

Ioannis P Stavroulakis

University of Ioannina, Greece

Stepan A Tersian

University ofRozousse, Bulgaria

WeWorld

N E W JERSEY

6

Scientific

SHANGHAI * HONG KONG * TAIPEI CHENNAI

LONDON

*

SINGAPORE

*

BElJlNG

Published by World Scientific Publishing Co. Re. Ltd. 5 Toh Tuck Link, Singapore 596224 USA ofice: Suite 202, 1060 Main Street, River Edge, NJ 07661 UK ofice: 57 Shelton Street, Covent Garden, London WC2H 9HE

British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library.

PARTIAL DIFFERENTIAL EQUATIONS An Introductionwith Mathematica and Maple (Second Edition) Copyright 0 2004 by World Scientific Publishing Co. Re. Ltd. All rights reserved. This book or parts thereoj may not be reproduced in anyform or by any means, electronic or mechanical, includingphotocopying, 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 981-238-815-X

Printed in Singapore.

To our wives Georgia and Mariam and our children Petros, Maria-Christina and Ioannis and Takuhi and Lusina

This page intentionally left blank .

T. Special thanks are due to Dr J. Stepan A. The second author would like to thank National Research Fund in Bulgaria for the support by the Grant MM 904/99. Also many thanks to our graduate students over several semesters who worked through the text and the exercises making useful suggestions. 2003 Ioannis P.Preface In this second edition the section “Weak Derivatives and Weak Solutions” was removed to Chapter 5 to be together with advanced concepts such as discontinuous solutions of nonlinear conservation laws. Stavroulakis. many points in the text were improved and the errors in the first edit ion were corrected. for the continuous support. Tersian vii . Barbatis for his comments. Lu. advice and active interest in the development of the second edition. Scientific Editor of WSPC. Many thanks are due to G. The figures were rearranged. September.

This page intentionally left blank .

Preface to the First Edit ion This textbook is a self-contained introduction to Partial Differential Equations (PDEs). the treatment here is reduced to the following features: 0 To consider first and second order linear classical PDEs. based on classical calculus and ordinary differential equations. physics. each one divided into several sections. To consider the one dimensional spatial case before going on to two and three dimensional cases. half-line and the mixed problem using the reflection method. 0 The book consists of eight Chapters. Advanced concepts such as weak solutions and discontinuous solutions of nonlinear conservation laws are also considered. linear. quasilinear. To illustrate the effects for different problems with model examples: To use Mathematics software products as Mathematzca and MAPLE in ScientifiCWorkPlacE in both graphical and computational aspects. as well as to present some ideas for nonlinear equations. 0 To give explicit formulae and derive properties of solutions for problems with homogeneous and inhomogeneous equations. the method of characteristics and the Cauchy problem. Although much of the material contained in this book can be found in standard textbooks. In Chapter I1 we give the classification of second-order PDEs in two variables based on the method of characteristics. without boundaries and with boundaries. In Chapter I we present the theory of first-order PDEs. It is designed for undergraduate and first year graduate students who are mathematics. engineering or. A classification of almost-linear second-order PDEs in n-variables is also given. in general. Chapter I11 is concerned with the one dimensional wave equation on the whole line. To give a number of exercises completing the explanation to some advanced problems. The goal is to give an introduction to the basic equations of mathematical physics and the properties of their solutions. The inhomogeneous equation as well as weak derivatives i x . nonlinear. science majors.

Kiguradze. diffusion and Laplace equations are also considered. Graef.H. Kyventidis. G. We also present the programs in Mathematica for those examples. Tersian . Shen for their extensive help with the proofreading of the material. Stepan A. Th. K. In Chapter IV the one dimensional diffusion equation is presented. Th. Kyriaki. Kon. Kvinikadze. Georgiou. Ifantis. Chapter V contains an introduction to the theory of shock waves and conservation laws. Green’s identities and the representation formulae are given. discontionuous solutions and Rankine-Hugoniot condition are considered. The Maximum-minimum principle.H. Stavroulakis. N. E. K. In Chapter VII some basic theorems on Fourier series and orthogonal systems are given. Biltchev. Riemann problem. Popivanov. Gopalsamy. For further details in Muthemutica the reader is referred to Wolfram [49].R. M. Karakostas. Sficas and P. Many thanks are also due to G. Ladas.G. 1999 Ioannis P. The help of S. Editor of WSPC. Raptis. Y. as well as Harnack’s inequalities and theorems. P. Finally in Chapter VIII two and three dimensional wave and diffusion equations are considered. Vidalis for their comments and to T. A special word of gratitude goes to N. Fourier methods for the wave. the Poisson formula with applications and the reflection method are given. Our deep appreciation to Calouste Gulbenkian Foundation and to the Greek Ministry of National Economy. Grammatikopoulos. Gan. for her continuous support. Chaparova and M. Special thanks are due to Ms S. Kirchoff’s formula and Huygens’ principle as well as Fourier method are presented . the mean value property. Green’s functions for the half-space and sphere are discussed. advice and active interest in the development of this project. Karaivanova is gratefully acknowledged.K. Maximum principles. In Chapter VI the Laplace equation on the plane and space is considered.I. Burgers’ equation and Hopf-Cole transformation are discussed.X Preface to the First Edition and weak solutions of the wave equation are also discussed. Model examples are given illustrated by software products as Muthematicu and MAPLE in ScientifiCWorkPlacE. Dassios. Grossinho. A. G. M. G. June. Artemiadis. J. Ross [34] and Vvedensky [47]. Siafarikas who reviewed the book and offered helpful comments and valuable suggestions for its improvement. The notion of weak solutions. J. M.R. Popivanov. J. G.

Mixed Problem for the Wave Equation 3. The Wave Equation on the Whole Line.1. General Solutions of Quasi-linear Equations 1. Introduction 1.2. Second-order Partial Differential Equations 2.1. The Diffusion Equation on the Whole Line 4.3. Classification and Canonical Forms of Equations in Two Independent Variables 2. The Cauchy Problem for First-order Quasi-linear Equations 1.4. Shock Waves and Conservation Laws 5.2.Contents 1.3.5. Inhomogeneous Diffusion Equation on the Whole Line 97 97 103 115 118 123 123 5. First-order Partial Differential Equations 1.2. Weak Solutions. Linear Equations 2.4. Diffusion on the Half-line 4. Inhomogeneous Wave Equation 3.4.1. Classification of Almost-linear Equations in R" 3. The Wave Equation on the Half-line. D'Alembert Formula 3. Maximum-minimum Principle for the Diffusion Equation 4. Linear First-order Equations 1. Reflection Method 3. Fully-nonlinear First-order Equations 2.1. One Dimensional Diffusion Equation 4.3.1. Weak Derivatives and Weak Solutions xi . One Dimensional Wave Equation 3.3. Conservation of the Energy 1 1 4 11 23 28 39 39 46 59 67 67 78 84 87 92 4.5.2.

4. Weak Solutions. Green’s Identities 6. Fourier Method for the Diffusion Equation 7. Burgers’ Equation 5. Fourier Method for the Diffusion Equation in Higher Dimensions 262 8.2. Rankine-Hugoniot Condition 6.1. Green’s Functions 6. Fourier Method for the Wave Equation 7.5. Fourier Method for the Wave Equation on the Plane.5.3. Fourier Series and Fourier Method for PDEs 7. Kirchoff’s Formula for the Wave Equation.2. General Fourier Series 7.1. The Diffusion Equation in Three Dimensional Space 255 8. Huygens’ Principle 269 8.3.5. Harmonic Functions. Diffusion and Wave Equations in Higher Dimensions 255 8.2. The Laplace Equation 6. Harnack’s Inequalities and Theorems 130 140 153 162 169 169 173 182 185 193 199 199 217 229 238 243 7.2. Conservation Laws 5.3. Nodal Sets 276 References Answers and Hints to Exercises Index 287 29 1 301 . Green’s Functions for a Half-space and Sphere 6. Discontinuous Solutions of Conservation Laws.xii Contents 5.3.1.4. Orthonormal Systems. Maximum-minimum Principle 6. Fourier Method for the Laplace Equation 8.4. Fourier Series 7. Riemann Problem 5.4.

. By a solution of the equation (1..... a z j .Chapter 1 First-order Partial Differential Equations 1... 1 . We denote by C ( R ) the space of continuous functions in R and by Ck(a) the space of continuously differentiable functions up to the order k in R. ..1)is a PDE of order m.. X n . The order of a PDE is the order of the highest derivative which appears in the equation. x n ) E R. . j = = 1. perhaps. 2.. . A set R in the n-dimensional Euclidean space Rn is called a domain if it is an open and connected set.. .) be a function of n independent variables z1.. uxCixj a 2 U / a X . Partial Differential Equation (PDE for short) is an equation that contains the independent variables q . the dependent variable or the unknown function u and its partial derivatives up to some order.1) we mean a function u E C" ($2) such that the substitution of u and its derivatives up to the order m in (1.. A region is a set consisting of a domain plus. i .2 A .n are the partial derivatives of u.. Suppose (1. It has the form where F is a given function and u X j= a u / a X j .1)makes it an identity in ( X I .1 Introduction Let u = u ( q ... some or all of its boundary points. .

8 . u )= 0. ut ... + uuy = 0 ( shock waves ) 3. The correspondence defines an operator L. = 0 ( heat or diffusion equation ) + u y y= 0 ( Laplace equation ) 7.. ) is a nonlinear function u 2 . 8.y or z . ut + cuu. Equations 1. + 2uxxYy u y y y y 0 ( biharmonic equation ) + = 8.. 7 is of fourth-order.. the equation 2 is nonlinear because (u1 u2)(ul u ~ = ) ~ ululy ~ 2 is not ~ satisfied for any functions u1 and u2. Nonlinearity may be of various types. ut + m u . t . For instance. 9.. u.2) for any functions u1. c2 E R.t = f (z.uyy ( Monge-Amphre equation ) .. The operator L is nonlinear if (1. Examples 2. 9.. 11 and 12 are distinguished from the others in that they are not "linear".2 Partial Differential Equations Some examples of PDEs ( all of which occur in Physics ) are: + u y = 0 ( transport equation ) 2. 10 is of third-order. 4.u. + 2u. 11 and 12 are of second-order. u.y) . Each one of these equations has two independent variables denoted either by z. ui + ut = 1 ( eikonal equation ) 1. Linearity means the following. 6.. = EU. The operator L is said to be linear iff (if and only if ) L (ClUl + c2u2) = ClLUl + c2Lu2 + + + (1..2) is not satisfied. + u3 = 0 ( wave with interaction ) 9.(1 + u:) utt = 0 ( Born-Infeld equation ) 12.u.u. 2 and 3 are of first-order. = 0 ( wave equation ) 5. utt . 11. u. 3.. (1 . 10. utt . An equation is said to be almostlinear if it is of the form Lu+f f ( x .utu.. + u. where f (2. u:y .u. = 0 ( Korteweg-de Vries equation ) 6 . u. ( Burgers' equation ) 10. 5.u2 and any constants c1. Equations numbered as 4.u:) u.

For instance. The three most important kinds of boundary conditions are: (i) Dirichlet conditions or boundary conditions of the first kind are the values of u prescribed at each point of the boundary d D ... x. 10 and 11 are quasi-linear. are the values of the unknown function u and of an appropriate number of its derivatives at the initial point. 9.. particle mechanics.First-order Partial Differential Equations 3 with respect to u. ) is where F is a given function and u x j = a u / d x j . The philosophy of treatment of first-order PDEs is in many ways different from that of the more commonly encountered secondorder PDEs appearing in physics and science. geometrical optics. . also known as Cauchy conditions. In the case of two independent variables x . First-order PDEs may always be reduced to a system of Ordinary Differential Equations (ODES for short ). . while the equations 3 and 12 are fully-nonlinear. . is called a linear inhomogeneous equation. The general form of a first-order PDE for a function u = u ( x 1 . 5. while Lu= f. while the boundary conditions are the values on the boundary d D of the domain D under consideration. The initial conditions..x.. the equations 2. 6 and 7 are linear homogeneous equations. j = 1.. the equation 8 is almostlinear.. y the above form is Equations of this type occur in the calculus of variations. 4. etc. An equation is said to be quasi-linear if it is linear with respect to highest order derivatives and fully-nonlinear if it is nonlinear with respect to highest order derivatives...n are the partial derivatives of the unknown function u. where f # 0. A partial differential equation subject to certain conditions in the form of initial or boundary conditions is known as an initial value problem (IVP for short) or boundary value problem (BVP for short)..) of n independent variables ( 2 1 . If the operator L is linear then the equation Lu = 0 is called a linear homogeneous equation. It is clear that Examples 1.

ay then equation (1. In this textbook we concentrate on problems for first-order PDEs (linear. Y) . E C1 (a). If U h denotes the general solution of the homogeneous equation and up a particular solution of the inhomogeneous equation (1. We first consider one spatial dimension before going on two and three dimensions.3).4 Partial Differential Equations (ii) Neumann conditions or boundary conditions of the second kind are the values of the normal derivative of u prescribed at each point of the boundary dD.4) we mean a relation involving an arbitrary function such that for any choice of the arbitrary function we derive a solution of equation (1.R c R2 and a2 + b2 # 0. 1. (iii) Robin conditions or mixed boundary conditions or boundary conditions of the third kind are the values of a linear combination of u and its normal derivative prescribed at each point of the boundary d D . b. Y)U. y and the dependent variable u has the form (1.3) is Lu = 0. If we consider the differential operator L := a- ax a + b. + 4 2 . d coefficients a or b does not vanish on R. that is. c.4) By a general solution of (1.4). then the general solution of (1. heat or diffusion and Laplace equations) as well as the Burgers’ equation. (1.3) 4 2 . Y)UZ + b ( z . the three classical linear second-order PDEs (wave.3) is .a + C. at least one of the where a . quasi-linear and fully-nonlinear ) . homogeneous equations before inhomogeneous equations.Y)U = d (z. problems without boundaries before problems with boundary conditions.3) is written as LU = d .2 Linear First-order Equations A linear first-order PDE in two independent variables z . while the homogeneous equation corresponding to (1.

Thus w = U h f up. then we will show that it is of the form (1.3).u p )= Lv .Then L (v . we derive the solution . Take the function w . Thus.d = 0. where f is an arbitrary continuously differentiable function.5) is a solution of equation (1.x c . Thus the general solution of the inhomogeneous equation (1. if w is a solution of (1.up for some choice of the arbitrary function which appears in uh. The corresponding homogeneous equation is Integrating with respect to x (holding y as a constant).6) is up = x e .Lu?. v . Find the general solution of the equation dU -+ u= e-xc. (considering y as a constant).6).6) is u (x. we have Conversely. = e-”f (y) y) +m P . that is. Uh Example 1. we have where f is an arbitrary continuously differentiable function. dX Solution. (1.4) and therefore = w .up is a solution of the homogeneous equation (1.First-order Partial Differential Equations 5 Indeed.3).up. Observe that a particular solution of (1. We could also work as in the case of ordinary differential equations.7). = d .5). This is the general solution of (1. from (1.1. since by the linearity property of the operator L .

s2 c R2.r). Assume. Y)U = 0. the same result. Next we will derive the form of the general solution of the linear first-order homogeneous equation a ( z . (1.8) is transformed into the following equation (atx + Ky) + ( ~ +7bTy) + cu = 0. by choosing r) such that t .Y ) U X + b(x. that a (z. ~ UE U. + bqy = 0.6 Partial Differential Equations u ( x . Observe here that f is an arbitrary continuously differentiable function of y (instead of an arbitrary constant C that we have in the case of ODES). c E C1(a). (1.9) where the coefficients are now expressed in terms of the new variables aim is to simplify equation (1. 3 I that is. ~ ) = e-Jdx f (9)+ = = e-x f (y) [ [ + 1 J e-xeSdxdx e-"eXdz e-" f (y)-+ xe-". without loss of generality. y) # 0 and consider the ordinary differential equation .8) where a. b. Consider the transformation with Jacobian Since the equation (1.9).10) This is accomplished as follows. (1. Our ar).y)uy + c(x.

u<+ c (<.13) called the canonical form for the linear equation (1. rl) (1. Now choose Then and the transformation constructed in this manner. = K is the general solution of the ODE (l. and it can be solved as an ODE (cf.is invertible. in view of (l.14) .ll).10) is satisfied.11) are called characteristic curves of the differential equation (1. =K (1.ll). that is y) where 7 (2. Example 1. The one-parameter family of curves (1.rl) u = 0. equation (1.12) where qy # 0 and K is an arbitrary constant.8).11) Let the general solution of equation (1.3) we derive the following form (1.1).11) be r) (w).First-order Partial Differential Equations 7 (1. The equation (1. Then.9) reduces to the following simple form a (t.8). In the case of the inhomogeneous equation (1. y) and.12) defined by equation (1. for this function rl (x.

16). d=y2. b=-y. An appropriate transformation is since the Jacobian The coefficients a and c with respect to < and 7 become and therefore equation (1. Find the general solution of the linear equation xux-yuy+y 2 u = y 2.13) yields This is the canonical form of the homogeneous equation (1.16) The equation (1.15) Solution. The general solution of the last equation is .y#O.11)is and its general solution (which gives the family of the characteristic curves) is xy =K.yuy + y2u = 0.8 Partial Differential Equations Example 1. . (1. K a constant. c=y2. x. Consider the homogeneous equation xu. (1. The coefficients are a=x.2.

Note that we could consider from the beginning the inhomogeneous equation (1-15) and using the same transformation.16) is Y2 Observe that the constant 1 is a particular solution of equation (1.15) is given by (1.First-order Partial Differential Equations 9 and therefore the general solution to equation (1.15) is given by the function (1.17) where f is an arbitrary continuously differentiable function. .14) yields with general solution U Thus the general solution of equation (1. equation (1.17).15) and therefore the general solution of the inhomogeneous equation (1.

xn In the case of the linear equation n -1 =n-1. c are constants and ai for some i = 1 . . = 0.u n ... Find the general solutions of the following equations: (a) X U . (Extension of the linear equation in n-variables). ( ~ 1 . + + + + #0 . ( x . y u .Z n ) ...- . = xn (c) au.1 . where a l . = 0. a 3 . e. 2 . un-1) .y) u. Find the general solutions of the equations (a) (y ..2... where a . X ... ( z . If u~(zI.9) U .. c.. 3 . y u . buy c21= d .1 are n ..) = c n . a 2 .X) u p z ( X . . then the general solution is given by u=f The functions if u 1 (XI. .xn rank [ Ul. Un-l. U1.x ) u.. Consider the equation n z a j j=1 (21. 2. + a 2 u y + a 3 u .Xl * * . +cu = 0 . (b) x (9 .. d l dxn -X...un-l ( 5 1 .. 3. The characteristic system a1 an describes the family of the characteristic curves. ) U X j = 0... .xl * * * Un-l. b. the general solution is given by where v is a particular solution. x n ) are functionally independent .. (XI.10 Partial Differential Equations Exercises 1..) = ~ 1. . . d constants and u2 + b2 + + + + # 0... . = nu (Euler’s relation) (b) X U . x.2) U X y ( z .z ) u.. (c) q u .1 functionally independent solutions of the characteristic system. .

u) Then the equation (1. For inis nonlinear.y.b(z.u). The normal to this surface at the point P ( x .18) defines an integral surface S : u = u (x.3 The Cauchy Problem for First-order Quasilinear Equations We consider the case of the quasi-linear equation (1.y. c2.First-order Partial Different i d Equations 1 1 1. u) is the y. in the Euy) clidean (x.-l) andleti$ bethevector (a(z. but it is linear with respect to the derivatives ( u z . where R is a domain in the (x.y. u) space.19) The last system can be rewritten shortly as .c(x.y.u). space and consider the vector field V = : P E R}.18) Quasi-linearity means that the operator uy). vector ??p (uz. We define as characteristic curwes G {G the integral curves in R of the characteristic system (1. 212 and any constants c1. u) y. Suppose that P f R. but it is quasi-linear because 211. stance.18) can be interpreted as the condition that at each point P of the integral surface S the vector is tangent to the surface S. A solution of (1. the equation 2 in the Introduction is nonlinear. for any functions u.uy.

yo ( t ) ) As PO(ZO.08.1789-23. uo) E R.1857. where u ( t ) = u (20 ( t ). Theorem 1.19) and the curves do not change by translating the independent variable t. .u (xo( t ).b]such that UO) E S n ro. 21.1. uo = uo ( t o ) and We have u ( t o ) = 0. condition U ( t o ) = 0 and by the uniqueness theorem for the Cauchyl problem for ODEs it follows Augustin Louis Cauchy. there exists t o E [a. Yo ( t ) ) a F t F b.yo ( t ) ) The equation (1. then S is an integral y) surface and conversely every integral surface S is a union of characteristic curves.12 Partial Differential Equations which is an autonomous system of ODEs. Note that if a surface S : u = u (x. T h e n roc S yo.YO. b and c to be of class C1 (R) by the existence and uniqueness theorem for ODEs it follows that through each point Po (xo. is a union of characteristic curves. . Assuming a . > Proof.05.20) is an ODE with initial . Let U ( t ) = uo ( t ) .b]) intersect the integral surface S at the point PO(xo. which means uo ( t )= u (xo ( t ). r 0* There is a 2-parameter family of characteristic curves in R of (1. Let the characteristic curve r : o { x = xo(t) Y=Yo(t) u = uo(t) t E [U. xo = xo ( t o ) >YO = Yo (to). U O ) E S-2 passes exactly one characteristic curve yo.

t and D f c R& be domains @ : D + D' be of class C' ( D ).Firs t-order Partial Differential Eq ua tions 13 As a consequence of Theorem 1.2. This is formulated as Find a solution u = u ( x . containing a prescribed curve r constitutes the Cauchy problem for (1. the existence of an integral surface in a neighborhood of the curve r. i.t) Y =Y W > and T h e n there exist neighborhoods U of POE D and U' of QOE D fand a mapping @-' E C1(U') such that @-' ( 7 ) U and 2 '= J@-' ( Q o )= ( J @(Po))-'. y) of (1. yo) @:{ x = x(s. We shall consider the local solvability of the Cauchy problem. Theorem 1.y) among all integral surfaces. The main tool for solving the local problem is the well known Inverse Mapping Theorem (IMT).e.18). then they intersect along the characteristic curve ro through PO.3.PO( S O .18) f o r which where x = 20 (s) u = uo(s) is a n initial curve. which also has a local character. Let D c RS. t o ) E D . (IMT). (Existence and Uniqueness Theorem) Consider the firstorder quasi-linear PDE in the domain s1 c R3 . Now we prove a local existence theorem for the Cauchy problem Theorem 1. Q o ( 2 0 . The selection of an individual surface S : u = u ( x . @ (Po)= Qo.1we have that if two integral surfaces S1 and Sz have a common point PO . E D'.

t) Y =Y W ) . p (s)] . is an initial smooth curve in R and Then there exists one and only one solution u = u ( x .t) ) u =u (s.Yo > 1. b and c are of class C1 (R) Partial Differential Equations r:{ x = xo(s) y=yo(s) u = uo(s) OLsL1. t ) defined for t : a ( s ) I t 5 continuous functions and ) y =y (s. Let us consider the Cauchy problem for the ODEs system with initial conditions jF'rom the existence and uniqueness theorem for ODEs the problem has a unique solution x = J: ( s .14 where a . a (s) and p (s) are According to (1.y) defined in a neighborhood N of the initial curve l?.18) and the initial condition uo(s) = u ( 2 0 (4 (4)0 L s I .21) for the mapping @: { x = x(s.t) [a(s) ) p ( s ) where 0 E . Proof. which satisfies the equation (1.

t ) .) ( w z +Ytsy) + (ztt.) + (at.t ) ). 2 be y) the corresponding integral surfaces. and (p2 (x.Ut = c and Moreover cp (2. Consider now We find that acp. = = a (Us% - us + utt.V j (s.l .yj (s. Therefore by the uniqueness theorem for ODEs coincide in the common domain of definition . Then ( z j (s.. t ) . + &tY) Ut Ut = u.t ) ) are soz lutions of system (C) .) + b (u. + utt. + bs. 15 defined in a neighborhood N' of I?' = Pro. . Considering the systems of ODEs with initial conditions we find solutions ( j(s. + b'p.t ) .O+ut. Indeed let cp1 (2. is a unique solution. It follows that the characteristics rl and r2 starting from the point P (xo(s) . ..) us (as.s.uo ( s ) )also coincide. r.First-order Partial Differential Equations By IMT there exists a unique inverse mapping @-' : D' --+D .yj (s.'yo ( s ).j = 1 . y) y) y) be two solutions satisfying the initial condition and Sj = pj (x. + bt.

It is easy to see that the characteristic curves are given by x2 + y 2 = k and the general solution is u= f (x2 y2) ..3.kb.%.21) is violated. (2.kc. 2) or dX0 . In the following example it is shown that when (1.b. d -b ds ds which contradicts (1.-a ~ o= kab . c) is not tangent to the initial curve I at the point ( 2 0 .. Consider the following three cases: (i) The initial curve is given by the parametric equations + ..21) implies that the vector ( a . Solution. if then for the Cauchy problem there may not exist a solution or there may exist infinitely many distinct solutions.. where f is an arbitrary function. uo) . = k(a.ka. For if it were ' yo.kba = 0 . then the Cauchy problem has n o solution or there exist infinitely m a n y distinct solutions. . Note that condition (1.xuy = 0. ds ds for some const ant k .21).c) ds . Show that there exist initial curves such that when (1.21) holds with the equality sign. . either there is no existence of a solution or there is no uniqueness.16 Partial Differential Equations Remark. Example 1. In other words.e.b. Consider the equation yu. Thus dxo . i.

y) plane of a characteristic curve. u = u (s) = sins. y = o . + .sin s . y) plane coincides with the projection on the ( x. o that is. f ( t )= t . 2 y = yo (s) = sins.uo)-a(zo.yo. but r 2 itself is non-characteristic.yo. We have and by Theorem 1.u ) plane. u=y. Indeed the tangent vector (. = f (s2) = s2. This is incompatible with the requirement u = y and therefore no solution exists. cos s.uo) (-sins) (-coss) .First -order Partial Differential Eq ua tions 17 This curve is the parabola u=x2.(coss) (sins) = 0. This surface is a circular paraboloid. (ii) The initial curve is given by r : x = xo ( s ) = coss.cos s. which leads to the unique solution u = x2 + y 2. then on the circle x2 y2 = 1 one has u = f (1) a constant. 0) along r2. The condition that such a surface contains rl is f (xi+y ) . that is.3 there exists a unique solution. cos s ) to r is nowhere parallel to the characteristic vector 2 (sins. which lies in the ( x . Here dX0 dYo --b(xo. . Note that the given curve I72 is such that its projection on the ( x . = ds ds If u = f ( x 2 + y2) is a solution. r 2 is the ellipse x 2 + y 2 =1. Indeed the integral surfaces u = f (x2+ y2) are surfaces of revolution about the u axis.

that is. x2+y2=1. Solution. 0 5 s 5 1. 0) = 4 4 . Clearly there are infinitely many solutions in this case. . Observe that the initial curve r is now a characteristic curve. Indeed the tangent vector (. u = f (x2 y2) is an integral surface which contains r3.L (s.# 0 ds ds 4 for s # 4. f (w) = wn). 3 + + Example 1 4 Solve the PDE uu. 0) 3 to r is parallel to the characteristic vector (sins. The initial curve + uy = 1/2. u = UO ( s ) = 1.{ where 0 5 s 5 1 satisfies (1.sins. . . cos s. 0) = s .e. Here again In order for u = f (x2 y2) to be a solution it should satisfy f (1) = 1 which is possible for any function f such that f (1) = 1(i. For such a function f . u ( s .21) x=s y=s u =~ / 4 S dx0 do Y -b . u=l. The characteristic system f dx with initial conditions 2 (s.cos s .0) = s. with initial condition r.-a = 1 .. r is the circle 3 y = yo ( s ) = sins. y ( s .. 7. s ) = 4 4 .18 Partial Differential Equations (iii) The initial curve is given by r3: x = xo (s) = COSS. 0) along r3.

Y> for y = s # 4.PlotPoints. (2t+s)/4}) { s.-I.t in terms of 2.y2 4-Y 4 (Y . Graph of the function . The integral surface S through the initial curve I? is plotted in the Figure 1. + Solving with respect to s.s/4}.f2.> 101 f2=Par ametricPlot 3D [{ s .5.42 . y we obtain and the unique solution of the problem is U= i 42 .-0 .I}.5}] Show[fl.First-order Partial Differential Equations has a solution 19 x = s + st/4 + t2/4 y=s+t u = s/4 t / 2 .O.y2 4 (4 . Shading->False.1. { s.>')Integral surface through initial curve"] I Integral surface through i n i t i a l curve Figure 1.4 t= 4-Y S = 8y .1.1 using the Mathernatica program fl=ParametricPlot3D [ { s+ (t"2+st)/4)t +s. Plo tlabel.1) .{ t .s.

In this case for the solution . u becomes infinite at the positive time .5. where h is a given function. h' 1 (4 we have U. At the time . Find the integral surface satisfying the initial condition u ( s .0 ) = 0. # .e. y = t xs zt Ys Yt 1 =1 + h' ( s )t # 0 . = h' ( s ) S.0 ) = h (5) .20 Partial Differential Equations Example 1. = h'(4 1 h' (s) t ' + Hence for h' (s) < 0 . (s.t ) can be expressed in terms of (z. = 0 can be interpreted as a vector field on the x-axis varying with the time y . As before. y ( s . The characteristic system f dx with initial conditions z ( s . 0) = s . Solution. The solution of the equation uy+uu.0 ) = h ( s ) has the solution x = s + h(s)t y=t u = h(s). u ( s .. y) when I i. T=-- 1 h' ( s ) ' The smallest y for which this happens corresponds to the value s = SO at which h' (s) has a minimum.

>” y =O.>” y=O. [ PlotRange.h3}.4 to demonstrate the effect of blow up with the Mathernatica program u[s-]:=sA3-3sA2+4 x [s-.4}.>None] .PlotLabel->”y=0.0.2}. SO = Then h (s) has a minimum at ‘ We plot the curves ct 1 and TO= 1/3. 0.0. 0. Ct : {x = s t(s3 .0.Firs t-order Partial Differential Eq u ations 21 the solution has a gradient catastrophe or blow up.333.u[s] { s .h5}}].2] .O.>{ 0.>” y =O .4}. + + + in the Figure 1. t -1 :=s+t u [s] hO=ParametricPlot [Evaluate[x[s. F’rame->True.Plot Label. 0 < s < 2 = 3(2s .{s.4}.hl}.>” y=O. 0. 3331ru[s]] . .2}.> { 0.4}.33.u[s]] s.{ PlotRange-> { 0.>{ 0. 4”] Show[GraphicsArray[{{ hO.2}.u[s]] s.2 for the instants t = 0.33”] h4=Par ametricPlot [Evaluat e[x[s.h (s) = 6. consider uo(s) = = s3-3s2+4.2”] h2=ParametricPlot [Evaluate[x[s.2) = 0 for s = 1. { h4.O .2. 331. There can not exist a smooth solution beyond the time TO.3. 09352.3”] 3.3s2 4).PlotLabel->” y=O”] hl=ParametricPlot [Evaluate[x[s.4]. PlotRange.0. .O.4) .Fr ameTicks. Plot Range-> { 0.2}.O . 111 hl ( s ) hI1(s) 3(s2 .PlotLabel.PlotLabel. u = s3 .PlotLabel.2s) < 0.{ PlotRange. h3=ParametricPlot [Evaluate[x[s.2}.O] .0.0. As an example.0.u[s]] s. .3] .3 2 4.4}. 333”] h5= Par ametr icP lot [Evaluate[xs . { h2. 0.u [s]] .2}. 0.{ PlotRange-> { 0.{s .

0. .8 1.4 1.5 3 2. + . . 0) = s2. .0. U ( 2 ) s) = s . yu.6 1. + 3.8 1. . 2.5 2 1. + (y + 2 ) U Y = u . y) Show that the solution becomes singular for some positive y unless a ( h ( s ) )is a nondecreasing function. Curves ct at the instants t = 0.9 Figure 1.2. . + uY = u2.6 1.5 1. 0) = h ( s ) is given implicitly by u = h (z .33.333 y=0. .3 4 3.0. . . .6 1. Prove that if two integral surfaces S1 and S of the equation (1.2 > \ y=0. = 0 with the initial condition u ( s . Exercises 1. .4 \ 0.5 1 1.0.2.5 1. = 2u. . (b) U . U ( S .8 1. Show that the solution of the quasi-linear PDE uy a(u)u.0.a (u) .7 1.22 Partial Differential Eq uations y=0.4. Solve the following initial value problems: = (a) u .0.5 1 0.7 1. ~ ( 1 . then I is a ' characteristic curve. .9 y=0.9 4 3. (c) xu. .18) 2 intersect transversally along a curve I?) which means that at each point P of I? the normal vectors n i i and n i i are linearly independent.5 J .333.4 1.5 1. .7 1.3. s ) s.4.5 1.

u ) and w (x. dx where k = . u ) and w (2. Definition 1. c ) # (O. 0) . u) is said to be a first integral of (1.c = 0.22) for which the normal vectors 6 (vz.18) are y y functionally independent i f (1. Suppose 21 (x. u ) are functionally independent first intey. y. wxa+Wyb+wuC = 0.y. r: { x = x(t) Y =y(t) u = u(t) can be represented as the intersection of two surfaces (1. The first integrals v (x. which means that (1. The characteristic curve ( a .b.22) holds.First-order Partial Differential Equations 23 1. . wxx+wy$j+wuiL = 0.a+vyb+v.1.23) is fulfilled. independent at each point P .y. .and dt v.O. w. v. From it follows v x x + v y y + v u ~ = 0.4 General Solutions of Quasi-linear Equations Suppose that for P ( x . z ) E 0. .18) if it is a constant on characteristic curves. u ) of (1. wy.) are linearly 1 vy.) and 62(wx. grals and (1.23) A continuously differentiable function v ( x .

be functionally independent first integrals of (1. by (1. Proof. # (0. y . Conversely let u = u (x. ) ) and W = 20 (2. be a solution of (1.uy + VUUX wx W U U X wy i.24 Partial Differential Equations From (1. Theorem 1 4 Let v (x.y) be a function for which Differentiating (1.V u W x ) ~ . Then the general solution of (1. u (2.w. buy .25) with respect to x . u (x.c) = 0. F.18). (vxwu . it follows (1. y. u ) y) y. u ) be functionally independent .uy + + . .wy . . + (1. first integrals of (1. 0.26). we have FdVX F (VY w Assuming (F.26) jF'rom (1.vuwx)uy= vywx . We have for the functions V = v (x.18).) + vuux) + Fw(wx + wuux) + vuuy ) + F w (wy + wuuy) = = 0.18)..vywu)ux (vxwu .24) . v (x.vxwy. y)) y. Then .24) it follows aux buy = c. Let u = u(x. u ) and 20 (x. 1 21%- vy + + v.vxwy) y X (au.26) and (1.18) is where F is a n arbitrary continuously difierentiable function of two variables. y + vx IVY w x WYl = = = (vuwy .23) it follows that v and w are functionally independent first integrals iff a b C (1. u)and w (z.O) it follows or ('u. y.'uywu)u. y y.(vywx .24) wy wu wu wx WX which geometricaly means that the vector 6 x n'a is a tangent vector to I? at 1 P.

f((x Y + y)2 .. y .e. i.First-order Partial Differential Equations 25 jF'rom the rank theorem of Calculus it follows that one of the functions V and W can be expressed as a function of the other. + s.u2)- (1. (u . x + y + u # 0 we get the relation (1.y)ux + yuy = x + y. The characteristic system is -U-Y dx - -= (4 (iii) Using the proportion property ( i ) (iii)= (ii)we have + Then v=-is a first integral.c1 + (ii)= (iii) it follows that w=(z+y)2 -u 2 =c2 is a second first integral. d Y Y (ii) du Z+Y' and solve the initial value problem u ( s .1) = 2 Solution. We have [ and for y or wz vx wy vy wu vu ] [ = l/Y 2(x+y) .U(X.Y)> = f ( w b .24).Y.(x 2 (x + Y) +4 l Y 2 -2u # 0.27) . there exists a function f such that V(X. From ( i ) u t x Y . u ( z . The general solution is -- u + x . y ) ) ) * Example 1 6 Find the general solution of the equation .

3.O.PlotPoints-> 101 hl=PlotSD[Sqrt [(x+y) ^2+5].Shading->False] Show[GraphicsArray [{ g l .O.O. Two families of characteristics in Example 1.{y.O. {x.>False] $2. 10 for 0 5 x 5 1 .{y.1 and the solution is u + z +(x+y)2-u2+1=O.O.l}.O.h2. {x. {y.27) 2 f((s 1)2.l}. hO=Plot3D[x+y.hl . 5.f 1 Shading.O.O.l}.>101 g2=Show[hO.l}) PlotPoints.{x. 5.l}.g2}]] 0.26 Partial Differential Equations We plot the surfaces v = 0.O. }. PlotPoints-> 101 h2=Plot3D[Sqrt [ (x+y) ^2+10].O. Y .{x.l}.{y. +s +s = + + 2)2)) so f ( t ) = -t .PlotPoints-> 101 1 f2=Plot3D[ 1Oy-x.l}.O.PlotPoints-> 10) g l =S how [fo.{x. {y.3 ) .{x.(s 2 ( s + 1) = f(-2s .3 using the Mathernatica program fO=PlotSD[-x.3. Y f 0.l}.l}. 0 5 y 5 1 in Figure 1.l}. 10 and w = 0.75 1 " I 7 0 I Figure 1. To solve the initial value problem we substitute the initial conditions in (1.PlotPoints-> 101 fl=PlotSD [5y-x.O)l}){y.

X) U Y = z .y) x 2 u y . .U ) u. y (u . uy = 2.y.: + yzu.z2)uY -x.u) . Exercises 1.y) y 2 u . Find the general solutions of the equations (a) ( x . + y u y + zu. = (e) (I d m )u.2 ) u y = ( x .y) u.First-order Partial Differential Equations 27 We now indicate (cf. 1) = xy.( x . (u2.: (U . z ) is xu. (c) 2 (y .y. Exercise 2. + + + + + + 2. (b) (9 . u ( X ) y. The Euler PDE for a homogeneous function u (z. = au. For the quasi-linear equation the characteristic system is If are n functionally independent first integrals of the characteristic system then the general solution is implicitly given by where F is an arbitrary continuously differentiable function. (b) 3. Section 1. = 0 . .( x 2 y2) u = 0. u (d) U U .2) how to proceed in the case of more than two variables. Solve the initial value problems (a> xu.

q = uy .) X (1. u ) on an integral surface S : u = u (5.y. 4. F is a twice continuously differentiable function with respect to its arguments x. We assume now that the operator + is nonlinear with respect to ( p .y.28). has a solution u = y. (b) The function 00 + satisfies the equation (1. y) and the direction y. z # 0 and u (Ax. q ) . 1) = h (x. p . z ) . q and F: F: # 0. Verify that: (a) The general solution of the differential equation uy = (5. For instance the so called eikonal equation u: ut = 1 arising in geometric optics is nonlinear because there exist u1 and u2 such that + The equation (1. (z.5 filly-nonlinear First-order Equations The general first-order equation for a function u = u (x. y) z"h 5 ) . .29) can be viewed as a relation between the coordinates of the point P (x. u .29) is a fullynonlinear first-order equation.28 Partial Differential Equations Show that the initial value problem u (5.Xy.. In this case we say that (1. (c) The following identity is satisfied 00 00 1. has the form y) where p = u. Xz) = X"u (z.28) is u = xf (x2 y 2 ) .

q .1746-28.29) at PO. Recall that the envelope of a family of smooth surfaces Sx : u = G (2.20) + (Y .Yo) . y depending on a parameter A E [a. The tangent plane TS (PO) the point at PO( a : ~ .yo. ( where (1. different values of p therein will in yield different values of Q and hence a one-parameter family of tangent planes.07. parametrized by p .30). is a surface C for which at each point P E C there exists A0 E [u. U O ) for the equation ug + u: = 1. It is easy to see that the Monge cone is a cone with vertex PO.32) 2Gaspard Monge. -1) at P.30) Given the values (zo. Find the equation of the Monge cone at Po(x0.31) In the case of the Monge cone.b] . . 10. for every X E R.32) Recall that a set K C R3is said to be a cone with a vertex PO if for every P E K the point XP + (1 . The envelope of these tangent planes is called the Monge2 cone for (1. .A) POE K . . By p 2 + q2 = 1 we have q = has the form k d m and the system (1. y ~ .First-order Partial Differential Equations 29 of the normal vector n'p ( p .uo) (1. Example 1.yo.7. Solution.b]such that The equation of C is implicitly given by the system (1.A) . u E )S is given by o u .uo = p .31) is ) (1. assuming 4 = ~ ( pthe system (1.05.1818.

. 1.fl. Assuming xo = yo = uo = 0 the cone.210)2.1.1.. {x.{x.1} .4.30 Partial Differential Equations Taking squares of both equations and adding we get (x .>False] L I Figure 1.uSin[v].u}. Shading.{v. Monge cone . using the Muthematicu program fO=ParametricPlotSD[ { uCos[v].f2 $3.yo)2 = (u .1} .1} . .1 . { u.d / 2 is given in Figure 1. Plot Points-> 201 f l =Plot3D [y.-.4.-1.PlotPoints-> 101 Show[fO .-l.>101 } f3=Plot3D[ (-x-y)/Sqrt [2].1} .O. represented as an envelope of planes for which p = 0.xo)2 + (9.{ y.l}.{ y.PlotPoints.{x.2Pi}.PlotPoints.1} .> 101 1 f2=P lot 3D [x.{y.O.-1.

y ( t ) )and q = q ( x ( t ).y ( t ) )we have (1.29).= PFp + 4Fq dt or D (1. we have so that .34) .29) iff it remains tangent to the cone M ( P ) at each point P E S. A surface S in R3 solves the equation (1. Namely.uo.y . by (1. q.Y o ) = u . u ) E R3. The completion of these cones { M ( P ) : P E R3} is the cone field. The reason is that there are three equations only for the five unknown functions x .33) are not sufficient to determine the characteristic curves comprising the integral surface.p . y . u .33) It is clear that the three equations (1.First -order Partial Differ tial Eq uations en 31 The equation (1.2 0 ) + 4(Y . . let us consider the Monge cone M ( P ) at each point P ( x .= Fp . However for p = p ( x ( t ).29) defines a cone field.32) and the equations describing the Monge dP cone are dq P(" . Assuming q = q ( p ) .may be eliminated in (1.= Fq dt du .Y-Yo x---0 F P or -- 4 ' u-UO x-xo F P --- y-yo Fq - PFP +qFq' The characteristic curves are determined as integral curves of the ODE system dx .

36) follows from (1.u ( t ).P ( t ).q ( t ) ) const.u ( t ).33) give a system of five ODES for the five functions z.q (z.P ( t )> ( t ) -Fx . p ( t ).37) for it is a first integral of (1.+ Fqdt dt 'dt dt which means that F ( x ( t ).y . p ( t ). q ( t ) is a solution of (1.Y ( t ).34) may be written as (1. The equation (1.+ Fu. Indeed if z ( t ). y) .= -Fy . = . q depending on t .u ( t ).-.y ( t ).35) . p (z.37).29). u .35) associated with (1.37) : dF d .PFU. y ) ) = 0 it follows Since Py = u x y = u y x = qx equations (1.32 Partial Differential Equations and from F (z.--F ( x (-4 . however (1. u .p .qFu.+ F . dt Equations (1.29) together with the characteristic system provides a system of six equations for the unknown functions z ( t )> 3 ( t )> u ( t ). This system is called a charucteristic system related to the equation (1. y .y ( t ). This system is overdetermined. !I ( t ) ) dt dt dx dy du dp dq = F x X + Fy.

Firs t-order Partial Differential Equations 33 If F = 0 is satisfied at an “initial point” (50.y ( t ).q ( t ) )is a solution of (1. then the . 40) for t = 0 .q ( t o ) = qo then u ( t ) = u (z ( t ). u ( t ) )the tangent plane T ( P ) with the normal vector np ( p ( t ).1 and 1. We consider the structure of integral surfaces and the initial value problem for (1. then it lies completely on the surface.6. Note that not any five functions define a strip. P ( t ) = ux (x ( t )7 Y ( t ) ) 9 4 ( t ) = uy (x ( t )> Y ( t ) ) * Theorem 1.37) and there exists t o such that x ( t o ) = X O .. = solution of (1.3 of the quasilinear case..5.qF4 = P ( t )dt + Q (t)X’ dt We call the strips which are solutions of (1. We formulate without proofs theorems which correspond to Theorems 1. uo.y ( t ).u ( t ). q and suppose that r: { x = 20 (s) Y =Yo(s) u = uo (s) 0 5 s 51 . we require that the planes be tangent to the curve I? which means that (1.37) can be interpreted as a strip . . Namely. If a characteristic strip has an element (50. Theorem 1.37) satisfies F (x ( t ). q ( t ) ) 0 for every t.. and their corresponding curves characteristic curves.y). A solution of (1.p ( t ).37) because du dx dY . This means a space curve r:{ x = 2 (t) Y=YN u = u (t) and along its point P (x( t ). p ( t ). The strip condition is guaranteed by the system (1. Consider the PDE (1. q ( t ). qo) in common with an integral surface u = u(x. 3.291. yo... u ( t ). which means that i f (x( t )....38) called the strip condition.29)..Y ( t ) ) . -I) ..= PF. where F has continuous secondorder derivatives with respect to its variables x.PO.37) characteristic strips .y ( t ).

a01 - Z F . the proofs are based on the existence and uniqueness theorem for ODES and IMT. dX0 ( 5 0 . dY0 (1. + u.34 Partial Differential Equations is an initial curve.Po.Po. u = 1.. (xo. through the initial curve I? : x = coss. aa) # 0... = 1 0 5 s 5 27r. Solution.39) is fulfilled duo . y = sins. Functions po ( s ) and qo (s) such that Po" (4 ao" + (4= 1.Y o . uo.39) Then there exasts a unique solution u = u ( x . uo. Example 1 8 Find the solution of the eikonal equation . u2.0 = P O (s) (-sins) + qo (s) (coss) .y) of (1. which contains the initial strip. ? As before.e.29) in a neighborhood N' of I' = ProzyI? . For these functions the condition (1. i. qo ( s ) are two differentiable functions satisfying and Z F . po ( s ) . Yo. ds are po (s) = cos s and qo ( s ) = sin s.

O) P(S.I}.{v . 1 Show[fO. PlotPoints-> 151 f2=ParametricPlot3D[{u Cos[v]-l. sin s.O 2Pi} PlotPoints.1} {v .O.l { u. 1) is given in Figure 1. f l . O . Y (0) sin s.Shading->False] ) ) .l . { u.uSin[v] +u} . = = 1.0.2}.O) we get x = (2t 1)COSS y = (2t+I)sins u = (2t 1). O ) = cos s . sins) = cos s.2} ] fl=ParametricPlot3D [{ 1+uCos[v]. = sins. + + Then x2 + y2 = u2 is the integral surface for which u(coss. uz (cos s.O . .u Sin[v]. { u.First-order Partial Differential Equations Integrating the system 35 with initial conditions i u(s.f2.O . sins) = The surface with Monge cones at the points (1.O .> 15. = cos s.1) and ( . uSin[v].> 15 .5 using the Muthernatica program: fO=ParametricPlot3D[{uCos[v]. uy (cos s.sins) = 1. q(s.O) x = 2p y = 24 u=2 p=0 q=0 x ( s .u}.l+u}.2Pi}.O $Pi}.PlotRange-> {0. PlotPoints. {v .

o ~ s g . { r:x=s. (4 (4 (4 uxuy= U ) y = i . u. y=s. { r:z=o.u. u=3s. = 1. { u. Y=S.2u = 0 . u=s2. (1. uy + + { r : z = s . o ~ s g .40) .= 2 s f0i. Solve the following initial value problems: (4 (b) { r:x=s. y = o . u=s. u = o .y) .uy = 2. r : x = s . Integral surface with Monge cones. + .5. Exercises 1. 2.. u. o s s g . u = o i s g . . o ~ s i i . Consider the differential equation u2.z) (UY . u. 2 (ux. y=o. + x u y = 0.36 Partial Differential Equations Figure 1. u.

First-order Partial Differential Equations 37 Making the so called Legendre3 transformation where p = u z .1752-10.09. .40) can be expressed in parametric form as where f is an arbitrary continuously differentiable function.show that v satisfies the equation p2 + qv.q = uY. 18.1833. = 0. Adrien Marie Legendre.01. Show that the solution of (1.

This page intentionally left blank .

y is where a . 39 .Chapter 2 Second-order Partial Differential Equations 2. If we consider then the equation (2. e. c. d.1) is written as Lu = g. while the homogeneous equation corresponding to (2. f .1) is Lu = 0.g E C2(0) R C R2 and a2+b2+c2 the partial differential operator ) # 0 in R. b.1 Linear Equations The general form of a linear second-order equation in two independent variables x .

We classify linear differential operators L we shall study separately. (2.. if up is a particular solution of Eq. (2.3) results. i..cn the function is also a solution of (2. Then in some cases we can obtain the general solution of Eq. then Thus is also a solution of Eq. Fkom the linearity of the operator it follows that if are solutions of the homogeneous equation (2. then is termed the general solution of the inhomogeneous equation.&) is irreducible or non-factorable if it cannot be so written. .u2 E C2(R) and any constants c1.40 Part i d Different i d Equations The operator L is linear since the condition (1.2) is satisfied for every pair of functions u1. Assume also that the given function g is a real-valued analytic function in R.3). .3) and up is any particular solution of the inhomogeneous equation (2.. c2 E R..e. Furthermore. (2. We say that: (i) L A.1). then for every choice of constants c1..3). 6 0 into two types which (g.. (2.1).1) for every choice of constants c1. 6) reducible or factorable if it can be written as a product of is + b& + c. If uh denotes the general solution of the homogeneous equation (2. We shall now consider the simplest case when the coefficients in Eq. linear first-order factors of the form a& (ii) L (&.1) are real constants. (2.. a relation involving two arbitrary C2(0)functions such that for every choice of the arbitrary functions a solution of Eq. .c.3).

aly) + 1c) (blx .6) The operator L is always reducible when it is a homogeneous operator.3). that is u2 > +e --X c2 1c) (b22 . Similarly if u2 is a solution of L2u = 0. Since L is a linear operator. X2 a - d2 + 2b-dxdy + 8x2 d2 C- dy2’ are the roots of the quadratic equation aX2 + 2bX + c = 0. If u1 is a solution of the linear first-order equation Llu = 0.5) where cp and $ are arbitrary twice continuously differentiable functions.Second-order Partial Differential Equations 41 (i) Reducible Equations In this case the general solution can be found with the aid of results of Section 1. u1 is a solution of (2.U 2 Y ) L=L1L1= ( a l .2. Accordingly. that is.+ cd ) + l ax dy then the general solution is --X C1 2 . . of the form L= If a # 0 and XI.e.d + c l ) dy ax ( a 2 -d dx + b2-a + c 2 ) .3) is given by + _-c1 x uh = e a1 p (blx . if a = a1a2 # 0 and the factors L1. Suppose L is such that = (a1. d2 (2. that is. i. L1L2 = L2L1. then u 2 is a solution of (2. then the general solution of (2. L2 are distinct..a b l .a i y ) ) . then d2 d2 axay LUl = (LlL2)u1 = (L2L1)u1 = L2 ( L l U l ) = L2 (0) = 0.d + b1. then u = u 1 u is also 2 a solution.dydx’ L1. uh = e a1 (xp (blx .3).aly) (2. dy the operators Since the coefficients are constants and . If L1 = L2. L2 commute.

.ac 2 0. A linear second-order equation in n independent variables the form .X I .Y) ..2 ) ( -d .42 Partial Differential Eq ua tions then L = ax:( ..$ and h are arbitrary functions. = dy d (2bz + c. X2 are real iff b2 . This equation is written as Lu = 0 . 'p (x XI.+ Id + ) ax dy ax dy and according to (2. Example 2.- y :) (&$)* a If a = 0 . where L is the operator The operator reduces to L=L&2= ( E .xn has .) .Y) 'p (x + y) + e-Yh (x .5) the general solution is which may also be written in the form u = = + y) + e-xex-yh (x .1.. where 'p. then L Note that the roots XI. Find the general solution of the equation Solution.

.. Accordingly.7) is written as LU = G and the corresponding homogeneous equation is Lu = 0. where up is a particular solution. Bi. If either a1 or bl is zero..8) Assume that the coefficients Aij. the general solution of Eq. but it is possible to construct solutions which contain as many arbitrary constants as we wish. When L is reducible L = LlLZ then we can work as in the case of two independent variables. C in L are real numbers and Aij = Aji.7) is u = uh +up. (ii) Irreducible Equations When the operator L is irreducible it is not always possible to find the general solution. (2. (2. j = 1. The general solution of the inhomogeneous equation (2. the form of the general solution is modified appropriately. i.8) is where cp.$ are arbitrary functions.6) .Second-order Partial Differential Equations 43 If we consider the operator then Eq. n.. This is achieved by attempting exponential type solutions of the form (&. (2.

8) when the coefficients are constants.. As an example. and differentiation within the summation sign or within the integral sign is legitimate. let us consider the heat equation uzZ. More generally the superpositions are solutions whenever they define C2(52) functions. The operator . when L (aJ) = 0. Then the function /3 so as to obtain a functional is a solution of (2.-ut = 0. Since . Also u = cp ( a ) e az+h(a)y 1 for arbitrary choice of the function cp is a solution. dY dU it is easy to see that and therefore u = ecrz+PY is a solution of the homogeneous equation (2. Suppose that the last relation is solved for relationship /3= h ( a ). (2.3).3). The preceding ideas extend to Eq. dX dU -=pu. 1 k k > 0 constant.au.44 Partial Differential Equations u = ecrz+py where a and p are constants to be determined.

k > 0. c o s n x + Bnsinnx)e-kn2t.XUt = 0. a . we obtain Thus p = k a 2 .c2uxx= 0. (d) utt = au. Find solutions of the exponential type eas+pY for the equations 1 (a) uxx . (2. and for any value of a the function is a solution.p u t t = 0.Second-order Partial Differential Equations 45 is irreducible. 2buxy cuyy.9). Check whether the operators in the following equations are reducible and in the case they are find the general solution (a) utt . uyy = 0. (c) u. Looking for solutions of the form u = euxSpt. If we take a = in. b. then the function is a solution and also superpositions of the form co = C ( A . + .. 1 (b) u X X .ac = 0. c positive constants and b2 .. 1 (b) uxx . (c) 3uxx louxy 3uyy= 0.p t = 0. n=l are solutions of Eq. + + + + 2. Exercises 1.

Ut = Ax2 B x t C. (b) u.. d. U. C R2 is a domain and ( a . b. Find the general solution of the equation u.b. # (0.10) and the almost-linear equation in two variables auxx (2. dux + 2buxy+ cuYy+ F (x. c) (2.O) . . . . 0. + 2buXy+ cuyyi. c. u Y )= 0. 4. (a) Using the change of independent variables < = lnx.. e. . (d) u.U t = A cos (ax P t ) .utt = t Partial Differential Equations + e2.. u.. (c) u X X ..2 Classification and Canonical Forms of Equat ions in Two Independent Variables Consider the linear equation au..+ euy -k fu= g . + + + 5 .11) where a .ut = e2x+4t. where a ..46 3. . R in R.g .. (b) Find the general solution of the equation x2uxx + 2xyux:y+ y2uyy = 0. is transformed into an equation with constant coefficients.ut = e3x+2t.g are of class C2 (R) .. show that the equation q = lny. 2. Show that Then find a particular solution of the equations (a) u.. f are constants.

Proof. The function A defined by is called the discriminant of Eq. The sign of the discriminant is invariant under invertible transformations of variables.10).1. Since the principal part mainly determines the properties of solutions we shall classify the more general form (2.11) instead of (2.11).12) where: .11) we obtain the equation (2. for which and equation (2.Second-order Partial Differential Equations The expression auxx 47 + 2buxy + cuyy is called the principal part of each of these equations. (2.11) is tmnsfomed into (2. Theorem 2. Let be a smooth change of variables. Making the change of variables we have: Substituting in (2.12) Then the sign of the discriminant at Q = @ ( P ) is the same as at P.

15) . H jFrom the above it is clear that we can classify Eq. y) The equation is hyperbolic (parabolic.2. (ii) if PO(XO. Definition 2. as follows: (i) if PO( X O . i f A (x. i f A (x. y) (ii) parabolic.11) i s hyperbolic. Since JQ>( P ) # 0 the proof is complete. = 0 . (2. parabolic or elliptic in a neighborhood of a point PO(XO. E R y) is: ( i ) hyperbolic.ac) + ca) J2@(P)A. such that the equation (2. yo). Next we will show that we can find new coordinates and 7 so that in terms of the new coordinates the form of Eq.1. > 0. Then we say that the equation is in canonical fomn.11) at a point P (x. yo) i s a parabolic point (2. elliptic) in a subset G c R i f it i s hyperbolic (parabolic. (2.AC = . (2.11) is such that its principal part is particularly simple. < 0.11) according to the sign of the discriminant. y) ( i i i ) elliptic. elliptic) at every point of G.48 Partial Differential Equations Applying the MAPLE procedure Simplify we have A' = = = B~ .14) (iii) i f P (20. W e say that the equation (2.17zEd2 (b2 . < Theorem 2. T h e n there exists a n invertible change of variables yo) defined in a neighborhood of the point PO( X O . i s a hyperbolic point yo) (first canonical f o r m f o r hyperbolic equations). i f A (x.(-czvar v z ~ y )(-b2 + 2 (5ZllY .yo) i s a n elliptic point o (2.11) can be reduced to one of the three forms. Assume that Eq.

upp + 0 (a.+ 2b(p.yo) # 0 we can determine y = y (x) as an implicit function in a neighborhood of the point xo and By (2.5 (2. 2 + 2bqxqy + cqy = 0. a . = 2 (up2 + 2bpq + cq2) = 0. U called the second canonical form for hyperbolic equations. We choose yo) and q in order to have A = a<: C = aq2 + 2b<. (i) Let PO( X O .2by' + c = 0.<. < Proof of Theorem 2.13) to uaa . u. (2.cpy + ccp2y = 0. + <: = 0. y) so along the characteristics of (2.Second-order Partial Differential Equations In the case of hyperbolic equations the transformation 49 reduces (2. be a hyperbolic point.16) (2.16) dt =pFp + qF.2.u p ) = 0. By the theory of Section 1. So < and q are solutions of the first-order nonlinear equation a(p2. = const. p. cp (x.17) If we suppose cpy ( X O .16) the function y (x) satisfies the ODE ayf2 .18) If we suppose cpS (20. # 0 we can determine x = x (y) as an implicit yo) function in a neighborhood of the point yo and .

19) Both equations (2. A = b2 . because by it follows The case c (20.yo) # 0 or c ( z ~ . yo) = 0.b I Suppose t ( x . yo) we obtain the form (2.19) can be presented in the differential form a ( d ~-) 2bdzdy ~ +~ ( d x=~ ) 0. a are respectively their general solutions The change of variables reduces (2. yo) # 0 and dividing (2. (2. y) # 0 in a neighborhood N of the point (x0. It is invertible. (2. Since b2 .20) Without loss of generality we can suppose a (z0. Let us suppose a (zo. yo) # 0 and a (z.13). then b(x0. Next we describe the parabolic and elliptic cases.yo).18) reduces to two ODES Y: = b i .ac.11) by b(x0.13).11) to the form (2. Then . We should choose t and q such that yo) A = B = 0.a (2.18) and (2. ) because if a (x0.21) . y o# 0 .ac = 0 it follows that one of the two coefficients a or c is .50 Then the function z(y) satisfies the ODE Partial Differential Equations cxf2. y) = C 1 2 defined in a domain N1 c N .2bx' + a = 0. be a parabolic point. y) = C and q(x. (ii) Let PO( 2 0 .yo) # 0 is treated similarly. 7y2=.yo) = c(x0.a . The equation (2.

c ) # (O. (iii) Let PO(XO. . Then the change of variables + reduces (2.11) admits one family of real characteristics.ac < 0 it follows that a # 0 and (2.11) to the canonical form (2. which intersect transversally. 0 ) . be an elliptic point.11) admits two families of real characteristics. In the hyperbolic domain the equation (2.y) = [(z.Second-order Partial Differential Equations 51 not zero.20) reduces to Suppose its general solution is Take = (x. then c # 0 and we follow a similar procedure. H The equation (2.14). Since b2 . We should choose [ and q such yo) that A = C and B = 0. In the parabolic domain the equation (2. By (2. y) iq(s. In the case that a = 0. The change of variables reduces (2.b. If a # 0 equation (2. = K be the general solution of the first equation.15). Otherwise b should also be zero which contradicts ( a .11)) while its solutions are characteristics.O. while in the elliptic domain it has no real characteristics.20) is called the characteristic equation of (2.16) it follows that A = C and B = 0. a simple function such that y) Then A = B = 0 and C # 0.20) reduces to ODES of the complex form y) Let cp(x.11) to the form (2.

The discriminant is b2-ac c = x2y2 and the equation is hyperbolic in R2\{(x. Determine the type of the equation x2uxx. Replacing y’ by X in (2. dx x exact solution is xy = ~ 2 . We apply MAPLE procedures of ScientifiCWorkPlacE to realize the algorithm of canonization. The new variables are 3. Namely 1. Solution of characteristic equation. y = 0).y) : x = 0. applying Solve to the last equation we get the solutions 2. 3. Applying Solve O D E we find solutions of equations -= dy dx Y. For < and 7 we apply VectorCalculus+Jacobian. b = 0 . On the lines x = 0 and y = 0 the equation is parabolic.y 2u y y . Solution . c=-y 2 For u = x 2. 3.2. reduce it to the canonical form in the hyperbolic domain and find the general solution.18) we derive the equation aX2 .2yuy = 0.-Y. Let us consider the hyperbolic damain.52 Partial Differential Equations Example 2.1.2. x Y exact solution is . For ( and 7 we apply VectorCalculus+Hessian .2bX + c = 0.= CI X and 3 .

Second-order Partial Differential Equations 53 [ 2 y Hessian is Xi 1-7 xy.u<x uqx3 23 + . a3 = 0. Hessian is -$I7 0 [ y=y.3. 22 ' p3 P1 =o. s=x. 3. Denote 1 a=-sy. 1 ]. . 2 a =3 3 . 1 p=-. Applying Simplify we compute: - -ut2 y + UT2 yx4 . p = 2 =o. X 1 a2= -1.

q = xy it follows y2 = Eq. ..2quv + because from [ = 2 . we have * u (07) = P Therefore the general solution is 2 v(7) + .3.$J ( 0 * Example 2. For .2yuy = = = -2y (2u.54 Partial Differential Equations Then x2uxz . 2J = 0. 2 + 2xyu. Solution. Determine the type of the equation y u.Y2Uyy . . + 2X2UYY yuy = 0 + and reduce it to the canonical f o r m in the elliptic domain.2u. As A = -x2y2. y # 0.cvy uvx) -4utvy2 . the equation is elliptic if x # 0.xy -4Jqut. X The canonical form of the equation in the hyperbolic domain is Utv 1 + -u. The substitution uv = v reduces the last equation to the first-order equation 1 with general solution v ( rl) = P 2(59 0 f Integrating with respect to q.

c = 2 x 2 the equation aX2 . b = x y . = u t a i urly = - -2utx . 1 2Y2 dx 2y2 (2xy+2ixy) yields the solution zy2 1 (x) . p=2y.x2.2UVX. for -x2. 7 = -x2.s = 2uty. . New variables are { For y2 . a 2 =o.52. uy = u t p 4. Hessian is Denote: a=-2x.2bX +c =0 A2 has solutions 1 X I = -(2x9 2Y2 Now Solve ODE -=dy + 2ixy) . = -(2xy . -2.x2.-x2 1 2 . : . and compute: u. For (y2 .Second-order Partial Differential Equations 55 a = y 2. -x2). . 6=0. Hessian is -2. a1 = = y2 .u.2ixy) . Jacobian is [ 1 21. y = -22. a 3 =2.-iX2 1 2 = c1.

(e) = 0. -Ut The canonical form in the elliptic domain is Utt 1 + + -uq + 5 -17 2rl uq q = 0.-u 2x2 y2Ut) + + + + '= 4y2x2 (ut2 +uq2 + -uq + 27 1 J . + yuyy + .56 Partial Differential Equations Then y2uxz 2xyuzy 2x2uyy yu = 4y2u52x2 + 4y2uq2x2 . Exercises 1. + + . Using Muthematica plot the two families of real characteristics in hyperbolic domains.uy = 0. Determine the type of the following equations and reduce them to the canonical form.3uyy uY= 0. e2xuxx 2ex+yuXy e2yuyy + (e2y .ex+Y)uy= 0..2y2uq + 4X2Ut 1 1 = 4y2s2 (uc2 +uq2 .2uxy .r l Ut). + ( c ) u. (4 (1 + x2) uxx + (1 + y2) uyy + xux + yu. (a) uxx .

1897-21.05..y) Z ~ ferentiable functions.cos xuy = 0. (d) x2uXx y2uYy 2yuY= 4x9. - 3.. (b) 3u.J) = l / (p (X .2uXy.Second-order Partial Differential Equations 57 2..2 sin zuXy. . . + uyy= 0. Find PDEs. (c) u. Show that this equation is: (a) elliptic for y > 0 and with the change of variables it reduces to UEE 1 + uqq + 37 -Uv = 01 (b) hyperbolic for y < 0 and with the change of variables lF'rancesco Jacopo Tricomi. Consider the Tricomi' equation YU.where p and @ are arbitrary dif- 4.cos2 xuyy . (a) u. - + + uy = 2. ..3uyy= 0. . 05. + + (x + y)) . Find the general solutions of the following equations in the domains of constant type.1978.11.5UXY 2uyy 3u. whose general solutions are of the form (d) u (x.

24) and show that the characteristics in ( u . and their envelope is the hyperbola 1+ 4uv = 0.v) the system (2. = (2. 6 (t. Leopold Infeld.uv) = 0.23) is equivalent to the system (z or to the equation v qv + (1+ 2uv) <v + 5v .(1+ V2)cptt = 0. (2.1898-15. q = x+t.v = c. v = 0 .23) the equation (2. vv(tJ7) (2.01.58 it reduces to Partial Different ial Equations (us .v< = 0 v2ut .(1 2uv) uv + u2vv = 0.77) Plot the picture of characteristics in the hyperbolic domain.22) is equivalent to the system uv . Max Born.0.1968.v ) is the inverse mapping then t v = qu.01.24) u2tuu (1 2uv)t u v + + + v 2 t v v + 2 (utu+ v t v ) = 0. v )plane are the lines + c1.d.1882-05. The Born-Infeld2 equation is (1. 20. 11. usv 5.12. (d) Determine the hyperbolic domain of the equation (2.08.‘I = 77 (u. u<tu= vqrlv * (c) In the new variables (u.1970. + (b) If uq = 21s and < = < (u.u + C2.u = 0 .22) Show that: (a) Introducing new variables x-t.v) . u = = v cpt(t.cp?)cpZZ 1 + 2cpZcptcpZt . .zv istics using Mathernatica. Plot the picture of characteru = c.qu = 0 U2tU .

As before.. aij (x) = aji (x). we make a classification locally. Almost-linearity means that the equation is (2.. A function u (2) E C2 ( D ) is a solution of the equation (2. 0 = 0.j=l c n aij (4u x i x j + F (x.25).uZn) the gradient of u.. u (x) is an unknown function and vu = (uxl.3 Classification of Almost-linear Equations in Rn Let D be a domain in the n-dimensional Euclidean space R". Let and be nonsingular mappings defined in neighborhoods N and N' of xo and yo = 4 (xo) respectively.. ) (2. > the usual scalar product in R". if the substitution of u and its derivatives in (2. for a fixed point xo E D .25) where the coefficients aij (x) are assumed to be continuously differentiable functions in x..e. u.Second-order Partial Differential Equations 59 2.25) in D . A main requirement for a classification of the equation (2. erator L := i.. Denote by x = ( X I . such that .25) results an identity in x E D ..25) is linear with respect to second-order derivatives u x i x jThe linear op. i... . . xn) a point of R" and by < . An almost-lanear second-order equation in R" has the form z.25) is to be invariant under nonsingular changes of independent variables.j=l C n aij d2 (x) dXidXj is called the principal part of the operator appearing in equation (2.

> ) > ( .j=l n aij ( x o )tiEj. (2.26) where (2. t E R".t) = C i.25) in D and '(3)= u ($ (9)) > Y E ". Then u> = ' ( 4 ( .25) leads to or n (2.28) Let A be the symmetric matrix .27) The classification of the equation (2.25) at the point xo is based on the classification of the characteristic fomn Q (xo. and x E N A substitution in (2.60 Partial Differential Equations Let u (x)E C2 ( D ) be a solution of the equation (2.

r] > . . Let (2.Second-order Pax tial Differential Eq uations We have 61 and if where A is a n x n matrix.30) = ATx.30) we obtain the transformed coefficients as which coincide by (2. then Q (' [) =< ARq. Here AT denotes the transpose matrix of A. i.e. Denote by B the matrix with elements bkl. hr]>=< RTAhq. x. (2.27). Note that If we make the linear change of variables (2.29) A = y . B= It is symmetric and B = ATAh.

.31) at the point xo to the form P i=l 4 i=l where .31).31) is called the positive index.28). and either p = 0 or q = 0.25) at the point zo is said to be: (1) elliptic.. reduces the equation (2. The equation (2. q 2 0.28). (2) hyperbolic. The important statement is that these numbers are invariant with respect to nonsingular linear transformations of the variables and x .nullity of the characteristic form (2.62 Partial Differential Equations By (2. if v = 0. Let + < be the nonsingular linear transformation reducing the characteristic form (2.29) the characteristic forms of (2. + rlp .7 p2+ l .1 and q = 1.V P + 4 2 (2. or p = 1 and q = n ..28) to its canonical form (2..1. The number p of positive terms in (2. r = p q . and either p = n .rl) = 7: 2 + .rank and u = n . Therefore the classification of (2.(d=u((A T -1 ) Y). p + q 5 n. the number q of negative terms the negative index. .q = 0..T .25) is made regardless of the canonical form of the characteristic form (2. . Then the transformation y = A TX .p .25) and the transformed equation at xo and y* = ATxo are equal: By the basic theorem for quadratic forms there exists a nonsingular matrix A such that Q (5') () reduces to the canonical f o m (2 (Y*.31) where p 2 0. if u = n .

ultrahyperbolic.. The equation (2... if it is elliptic (hyperbolic. The classification can also be made with respect to the eigenvalues of the coefficient matrix A . Moreover the number of positive. An be the eigenvalues of the matrix A of the principal part of Eq.. (2. if XI. i. if v > 0. .. ultrahyperbolic...25).25) is said to be elliptic (hyperbolic. if AI.25) at the point zo is said to be: (1) elliptic. .. positive and two negative. Let XI..e. parabolic) at every point of D . zero and negative eigenvalues of the matrix A remains invariant under nonsingular changes of independent variables.An are nonzero and have the same sign.An are nonzero and all except one have the same sign. For instance.. = while the heat or diffusion equation where c is a constant. . The equation (2.1.. the roots of the equation jF'rom linear algebra it is known that since the matrix A is symmetric its eigenvalues are all real. (3) ultrahyperbolic. is hyperbolic in Rn+'. (2) hyperbolic. is elliptic in Rn.. parabolic) in D.An is zero. (4)parabolic.. (4) parabolic.. .Second-order Partial Differential Equations 63 (3) ultrahyperbolic. if u = 0 and 1 < p < n .the wave equation + + uXnxn 0 ...A are nonzero and at least two of them are . if XI. the Laplace equation Au := uxIxl . if any one of X I .. . .

where a is a constant.Z U X ~ -I. -4 Jz 2 1 -1 -1 -?] 0 reduces the original equation to the canonical form . Solution.~ u Z ~ X ~ 1 X ~ X Z . Reduce the equation 2Ux1x1 -I.~ U X Z.2 U ~ 2 ~ 30 = to the canonical form.r 3 ) 2 ) l 2 = . Determine the type and change of variables. The change of variables 1 has an inverse 0 -Jz Then the linear transformation 2 -11 1 0 [ 511. is ultrahyperbolic in R5.4.I?+$ -.((1 . The characteristic form of the equation is = ( a 1 + fi(. Example 2.a2Au = 0.64 Partial Differential Equations ut . is parabolic in Rn+'.I$. z ( 3+ Kl + C2I2 .u x 5 x 5 = 0.ux3x3 . The equation UXlXl + uxzx2 .ux4x4 .

Suppose that ai. Note that and Exercises. reduces the equation n n to the form n i=l 3.vy3y3 = 0. 1. .n and c are constants and ai # 0. Reduce the following equations to the canonical form. bi.. i = 1..Xz of the coefficent matrix Compare with Definition 2..11) with respect to the eigenvalues X1. So the equation is hyperbolic on the whole space. Find a function w such that the change of the dependent variable u = wz.Second-order Partial Differential Equations 65 VYlYl + V y z y z . . 2.1. Classify the equation (2.

This page intentionally left blank .

d /2c (e .t ) represents the “value” of the normal displacement of a particle at position x and time t . = 0 . t the ‘(time” variable.Chapter 3 One Dimensional Wave Equation 3. D’Alembert Formula .1 The Wave Equation on the Whole Line. t ) the unknown function and c is a given positive constant. Physically u ( 2 . The wave equation describes vibrations of a string. u = u ( x . Using the theory of Section 2.1) is ( d x ) 2.c2uz. Introducing the new variables a: and the function { r]=x-ct ( = X + d ’ w1 { : 67 x = + rl) /2 t = (t . The simplest hyperbolic second-order equation is the wave equation (34 where x signifies the spatial variable or (‘position”.2 the characteristic equation of (3.c2 ( d q 2 = 0 and Utt ct { xx + ct - = c1 = c2 are two families of real characteristics.

Theorem 3.11. ) is of the form t u (x.10.3.68 Partial Differential Equations the equation (3. (3. E C1(R) the problem (CW)has a unique solution u E C2(R x R+) given by the formula x+ct u ( 5 . . U t (x.1) reduces to Therefore u (<. Consider the Cauchy (initial value) problem for (3.ct) . u (x. ) = $9 (x) 0 x E R.4) 2-ct Proof.1).3) known as the general solution of (3. If cp E C 2 (R) and 1c. (3. ) = f (x ct) t + + g (x . = 1cI (4 0) Utt where cp and @ are arbitrary functions of x.another shape.1) .3) satisfying the initial conditions at t = 0 'Jean Le Rond D'Alembert.1783. 16.1717-29.t ) = 5 (cp (x ct) 1 + + cp (x .c2uXz= 0 x E R.Further we denote R+ = {t : t 2 0). (' D'Alembert' formula ).c t ) )+ 2c 1 / $ (s)ds.7) = / F (t)4 + 9 (7)= f (5)+ 9 (7)> and in the original variables u (x. t > 0 . traveling to the left with speed c.c t ) which presents a shape traveling without change to the right with speed c and the function f (x + ct) . It is the sum of the function g (x . x E R. We are looking for a solution of the problem in the form (3.

Note that if cp = $ = 0.cg/(x)= $ (x) ' Differentiating (3.5) f (0) + g (0) = cp (0).c t ) ) ( 2 xfct x-ct 0 0 x+ct x-ct Conversely it is easy to see that for cp E C2(R) and I) E C1(R) this I formula gives the solution u E C2(R x R+) of (CW). then it follows u = 0. Therefore u (z. t ) = = f (z + C t ) + 9 (x .C t > 1 -(cp (z + ct) + cp .One Dimensional Wave Equation 69 cf ' (x). W Some corollaries from D' Alembert formula are as follows: .6) and (3.7) from 0 to x weget X By (3. we obtain f'and Integrating (3.5) with respect to z and solving the linear system for g'. .

tl) is determined by the values of cp and I) on [x1 . The point (x0. (iii) The solution is stable.70 Partial Differential Equations I . Well-posedness. (ii) The solution is unique. ) in the wedgeshaped region t + For any 3. This is also referred to as continuous 2Jacques Hadamard.1963.ctl.cto. 18. z1 ctl]. Domain of influence. . Domain of dependence.12.O)on the x-axis influences the value of u at (x. through the point (x0. o ) is determined by the t restriction of initial functions cp and I) in the interval [ZO . The value of u at (xo.1865-17. I 2. Statement (iii) means that small variations of the initial data yield small variations on the corresponding solutions. xo cto] on the I z-axis.to). The characteristic triangle A (20.10. The problem (CW)is well-posed in the sense of Hadamard2 if the following three requirements are satisfied: (i) There exists a solution. whose end-points are cut out by the characteristics: + x--0 =fc(t-to). is defined as the triangle in R x R+ to) with vertices and u (21.

.4) it follows x+ct 2-ct > 0 there exists 6 E (0.PlotLabel->” Wave at t=l”] h2=Plot [Evaluate[u[x. Using Mathematica the profile of u (x.8}. The meaning of small variation is made precise in terms of the topology suggested by the problem.1 at successive t instants t = 0. 1. 5 .l}.{x.True. 3.m)) introduce uniform norms and For a given T > 0 by (3.8}.O]] .T E . ) is presented t in Figure 3. We use the Mathernatica program f[x-]:=Which[-Pi/2< =x<=Pi/2.-8. Note that at t = 0 the amplitude is 1.1}. 1. @ = 0 and Solution.t-] :=( f[x+t]+f[x-t])/2 hO=Plot [Evaluate[u[x.t ) ).PlotLabel->” Wave at t=O”] hl=Plot [Evaluate[u [x. After the instant t = $ the profile breaks up into two traveling waves moving in opposite directions with speed 1 and amplitude The surface u = u (x. The solution of the problem is u (z. Solve the problem (CW) with c = 1. (~ < Then for any E &) Example 3.t ) = (cp (z t ) cp (z .2. $. {x. PlotRange-> { 0. A problem that does not satisfy any one of these conditions is called ill-posed. such that if llqllw < S and I l q ! ~ l < S it follows I{ulloO.-8. For v (2) E C (R) and w (x.One Dimensional Wave Equation 71 dependence upon the initial data. t ) E C (Rx [0. PlotRange->{O.1 1{x. ) is presented in Figure 3.3.Cos[x] ^S.-8.2]].8}. which proves the continuous dependence.O] u[x-. a + + a. 4.

8}. 4.8} .{ PlotRange-> {O.PlotPoints->40 .x.5 5 7.5 -5 -2. .5 2.PlotLabel->”Waveat t=Pi/2”] { h3=Plot [Evaluate[u [x.8}.5 5 7. PlotRange->{ 0.l}. 5. 1..t].5 -7.h5}}] . .h l } .-8.5 2. .5]] x .”Value”.{ h2.1} .{t .l}.5 -5 -2.”Time”. { h4 . .PlotLabel->”Wave at t=3”] h4=Plot [Evaluate[u[x. .8}. .5 5 7.4]].-8.6 . -7.6 -7. . 4. 1.-8.5 2.5 -5 -2. PlotRange->{O.3]].5 Mve at t = P i / 2 Wve at t=l 0.5 Figure 3.6 Wve a t t = 5 II 0.5 -7.5 5 7. . { PlotRange-> { 0 .5 -5 -2. . . .5} .. .F’rameTicks->None] Plot3D[u[x.A. .3.{x . 0.5 5 7.5 2.PlotLabel->” Wave at t=4”] h5=Plot [Evaluate[u[x.PlotLabel->”Wave at t=5”] O Show [Gr aphicsArr ay [{ { h .5 -5 -2.5 Flave at t = 3 o. .6 } -7. AxesLabel->”Position”.0.8 1 1 0. x.5 Wave a t t = 4 0. The wave at instants t = 0.5 2. Frame->True.5 -5 -2.Shading->False] -7.1.:j 0.5 0.l) . .h3}.6 .l} .-8.q 0. 2 Flave at t = O 2.5 5 7.h:.72 Partial Differential Equations PlotRange-> {O.

The surface u = u (x. Graph of the function u = u (x.4.3. 4 Using Mathernatica the profile of u (x. = 0 and cp e-x2 $ J (x)= Solution. t Example 3.One Dimensional Wave Equation 73 Value Figure 3. The solution can be expressed in terms of erf as 1 . 3. Let erf (x)= 2 . ) in Example 3. Solve the problem ( C W ) with c = fi. ) is presented in Figure 3.3 at the suct cessive instants t = 0. [ e-”ds be the error function used in statistics.2.erf (x . 2. 1.f i t ) ) . We use the following program = . Note that at t = 1 the amplitude is 1/2 and it remains the same for all next instants.(erf (x + f i t ) . ) is plotted in t Figure 3.4.

0. 1.74 Partial Differential Eq uations u [x-.5 -7.x.5}.5}.PlotPoints->20.>False] Wve at t = O 0. PlotRange.1 25 5 7. PlotRange->{ 0. 3. . .O]]{x.O. { PlotRange-> {0.8}.’)Value” .{ h2 .>’)Wave at t =0” ] hl=Plot [Evaluate[u[x.5 2.5 2.{ AxesLabel->” Position” . .l]] .”Time”.5 -5 -2.-8.5 Wavg-at t = 2 Wave at t = 3 b -7.5 ) .3]]{x. 2. 4 ) .{x.PlotLabel->” Wave at t=3”] Show[GraphicsArray[{ { hO.5 5 7.PlotLabel.3. PlotRange-> { 0. 5} .3 0.8} .5}. .O.5 -5 -2.PlotLabel->”Wave at t=2”] h3=Plot [Evaluate[u[x.8}. 0 .0.2 0.8}.5 1 0.5 -5 -2. .h3}}]. PlotRange-> { 0.Shading.5 5 7.FrameTicks->None] plot3~[u[x. Wave at instants t = 0.> { 0.5 Figure 3.t -1 :=(Erf[x+Sqrt [Pi] t]-Erf [x-Sqrt[Pi] )/4 t] hO=Plot [Evaluate[u[x.PlotLabel->”Wave at t=l”] h2=Plot [Evaluate[u[x.Z]]. Frame->True.2 : -7.-8.-8.0.8 .0.ti x .h l } .5 1 I 04! .-8. ~ .

One Dimensional Wave Equation

75

Figure 3.4. Graph of the function u =

(erf (x

+ fit)

- erf (x - f i t ) )

Exercises

1. Prove the formula for the general solution of the wave equation (3.1) reducing it to the system of first order equations:

{

2. Suppose

vt - cvx = 0 Ut cux = v.

+

A (x, ) , q x cs, t s ) , t C ( x c ( s - T ),t s + T ) , D ( x - CT,t+ T )

+

+

+

+

are vertices of a characteristic parallelogram, where s,r are positive parameters. Prove that if u E C’ (R’) a solution of the wave equation (3.1) then is

u (A) u (C) u (B) u ( D ). =

+

+

(3-8)

76

Partial Differential Equations

Conversely, prove that if u is of class C2 (R2) and satisfies (3.8) for every ( s , t ) E R2, then u is a solution of the equation (3.1).

3. (a) Prove that if u (x, z ) = u ( p ) ,p = y,

Au = u X X

dx2+ y2 + z2, then

2 +P

-up.

+

uyy

-I- u z p = u p p

**(b) Making the change of variables v ( p , t ) = pu ( p , t ) show that the general solution of the threedimensional wave equation
**

Utt

2 - c2(upp ,up) = 0

+

is

( c ) Prove that the initial problem for the spherical wave equation with conditions

has a solution

p- c t

Note that this solution exists provided p 2 ct.

**4. Show that for $J E C1 (R) the function u (x, ) = & t
**

the problem:

x+ct

$J (s)ds verifies

2-ct

(

Utt

- c2uxx= 0

Check it also using Mathernatica.

One Dimensional Wave Equation

77

5 . (a) Prove that if p(s) is a continuous function, then c p ( z 5 c t ) are “weak” solutions of the equation utt - c2uz, = 0 in the sense

R R

JJ

cp (z f ct) (vtt - c2vz,) dzdt = 0 ,

for every test function w (z, t ) of the space

**Co” (R2) = {f E C” (R2) : suppf is compact},
**

where suppf = { ( UE R2 : f ( z , t )# O } . ) (b) Prove that if cp (5) is a continuous function, then the problem

t has a weak solution u (x, ) = $ (cp (z + c t )

+ cp (z - c t ) ) in the sense

1/

R R

u (z, t ) (wtt - c2v,,) dxdt = 0 ,

for every test function w (2, t ) E C r (R2) . (c) Using Muthematicu draw the profile of the solution of the problem (3.1) with

at each of the instants t = 0, 0.2, 0.6, 0.8, 1.2.

78

Partial Differential Equations

3.2

The Wave Equation on the Half-line. Reflection Method

Let us consider the problem (CW) on the half-line ( 0 , ~ with Dirichlet3 ) boundary condition at the endpoint x = 0. This is the problem:

(CDW):

{

U(X,O)

Utt - c2uxx = 0 = c p ( 2 ) , U t ( 2 , O ) = .II) (x)

u ( 0 , t )= 0

0 < 2 < 00,t > 0, 0 < x < 00, t 2 0.

**It can be interpreted as vibrations of a very long string with a clamped
**

one end. We are looking for a solution of (CDW) given by an explicit formula. In fact we shall reduce the problem (CDW) to a problem (CW) by the odd re*jIection method. It consists in considering the odd extensions of the initial functions cpo (z) and Qo(x) where

cp(x)

if x

> 0,

The problem

vtt - c2vXx 0 =

II: E

R, > 0 , t

R,

(x,O) = cpo (x)

vt (x, ) = Qo(x) 0

zE xE

R.

has the solution

v (4 - ( y o(x + c t ) + cpo (x - ct>> =2

Its restriction

1

5

1

x+ct

x-ct

/

$o

(s) ds.

**is the unique solution of the problem ( C D W ). If 0 < x < ct, then
**

3Lejeune Peter Gustav Dirichlet, 13.02.1805-05.05.1859.

One Dimensional Wave Equation

79

x+ct

x+ct

0

x-ct

0 x+ct

x-ct 0

x+ct

0

ct-x

c t Lx

Therefore we have

a (cp (x + ct) + cp (x - c t ) ) + 5 1 $

1 u ( x , t )=

x-ct ct+x

x+ct

(s) ds

if x

> ct,

1 3 (cp (x + ct ) - cp (ct - x)) + 2c

1+

(s) ds

if 0 < z

< ct .

(3.9)

ct-x

Note that u (z,t ) is a continuous function if the Compatibility condition cp ( 0 ) = 0 is satisfied. Otherwise u (z,t ) is a discontinuous solution and the jump of u (z, t ) on the characteristic x = ct is

u (ct

+ 0 , t ) - u (ct - 0 , t ) = cp (0).

We have

**Theorem 3.4. Let cp (z) E C2 (R+),$J (z) E C1(R+) and the following compatibility conditions be satisfied:
**

cp (0) = cp" (0) = $J (0) = 0.

(3.10)

Then the function u (2, t ) defined b y (3.9) is the unique solution of the problem ( C D W ) of class C2(R+ x R+). Proof. The function u (x, ) is of class C2 in domains { (x, ) : x > ct > 0) t t and ((2, t ) : 0 < x < c t } . We shall prove that the derivatives of u (z,t ) up to order two are continuous along the line z = ct. We have

12

( x , t )=

{

(9' (x

+ ct) +

+ ct) +

$Dl

(x - c t ) )

(ct - z))

1 + 2c (@(x + ct) - 1c) (x - c t ) ) ,

1 + 5 ($ (z + + 1c) (ct - x)) ,

Ct)

. 3 ct,

;

($0' (z

'p'

0 < z < ct.

80 Therefore by (3.10)

Partial Differential Equations

u, (ct

By the same way

UXX

1 + 0 ,t ) - u, (ct - 0 ,t ) = --q C

(0) = 0.

(ct

+ 0 ,t ) - u,,

(ct - 0 , t ) = cp" (0) = 0,

Ut

(ct + 0 , t ) - U t (ct - 0 , t ) = q (0) = 0,

utx

(ct (ct

+0,t)-

Ut,

(ct - 0 , t ) = -c(pII (0) = 0, (ct - 0 , t ) = c2# (0) = 0.

Utt

+ 0 ,t ) -

Utt

Moreover the function u (z, t ) satisfies the equation, boundary and initial conditions of the problem (CDW).

=

We can do the same for the problem with the Neumann4 boundary condition, considering even extensions of initial data. Let us consider the problem

( C N W ):

{

U t t - C2UX, = 0 u (z, 0 ) = cp (2), U t (z, 0 ) = 1c) (z) u, ( 0 , t )= 0

0 < z < 00,t > 0,

0<z

< 00,

t 2 0.

v e

In this case we reduce the problem ( C N W )to (CW)with initial functions (z) and 1c)e (.> 7 where

As before we can show that the problem ( C N W ) has a unique solution

u (x, ) = t

i

(cp (z

+ c t ) + cp (z - c t ) ) + s1( P (z + ct) - P (z - c t ) ) ,

z

> ct,

1 f (cp(a: + ct) + cp (ct - 2))+ s(!v + ct) + !v (ct - x)), 0 < 5 < c , . (a:

4Karl Gottfried Neumann, 07.05.1832-27.03.1925

**One Dimensional Wave Equation
**

t

81

where

X ( t )= .Ic) (s) ds. D

0

Example 3.5. Solve the problem (CDW) with c = l,$ = 0 and

’(”)

=

{

c0s3z x E (3n/2,5~/2), 0 z E R+\ (3~/2,57r/2).

Solution. The solution of the odd extended problem

**The original problem has the solution
**

u(x4)=

{

q’P(“+t)+cp(Z-t)) !((D(z+t)-cp(t-r))

x > t, O<z<t.

The profile of u (z, t ) is presented in Figure 3.5 at successive instants t = 0, 27r, using the Mathematica program

9, F , 9

5,

f[x-]:=Which[SPi/2<=~<=5Pi/2,Cos[x] A3,T!rue,0] g[x-]:=Which[x<O,-f[-x],True,f[x]] u[x-,t-]:=(f[x+t] +g[x-t])/2 hO=Plot [Evaluate[u[x,O]] ,{x,O,8Pi}, PlotRange- >{- 1,1} ,PlotLabel- >” Wave at t =O” ] hl=Plot [Evaluate[u[x,Pi/2]] ,{x,O,8Pi}, Plot Range-> {- 1,1} ,PlotLabel->”Wave at t =Pi/2”] h2=Plot [Evaluate[u[x,3Pi/2]], {x,O,8Pi}, PlotRange-> {-1 1} ,PlotLabel->” Wave at t=3Pi/2”] h3=Plot [Evaluate[u [x,2Pi]], x,O,8Pi} { Plot Range- >{-1,1} ,PlotLabel-> ’’Wave at t =2Pi”] h4=Plot[Evaluate[u[x,5Pi/2]], {x,O,8Pi}, PlotRange-> {-l,l},PlotLabel->”Wave at t=5Pi/2”] h5=Plot [Evaluate[u[x,7Pi/2]] ,{x,O,8Pi}, ,PlotLabel->” Wave at t =7Pi/ 2”J Plot Range- > { - 1,1} Show[GraphicsArray[ { { hO,h l } ,{ h2,h3}, { h4,h5}}]

)

)

82

Partial Differential Equations Frame-> True,FrameTicks- >None] Plot3D [u[x,t],{ x,O$Pi}, { t ,0,4Pi}, AxesLabel- >’’ Position” ,”Time” ,”Value” ,PlotPoints- >40, Plot Range-> { - 1,l},Shading- >False]

Wave a t t = O

0.75 i

-0.25

I 2.5 5 7.5 1012.51517.5

:

-0.75 :

-0.25 : -0.5 : -0.75 i

I 2.5 5 7.5 1012.5 1517.5

Wave at t = 3 P i / 2

1 0.75

1

Wve a t t 2 P i

-0.25 t -0.5 1 -0.75 i

5

10

15

20

25

-0.25 t -0.5 i -0.75 i

5

10

15

20

25

Wave a t t = 5 P i / 2

0.5 I 0.25 f

1: 0.75 ! 0.5 ; 0.25 1 -0.25 i -0.5 i -0.75 j - : 1

10

15

20

25

-0.75 i

Figure 3.5. Wave at the instants t = 0,

5, 2 ,27r ,% , 9.

One Dimensional Wave Equation

83

Figure 3.6. Graph of the function u = u (x, ) in Example 3.5 t Note that the initial profile splits into two profiles with amplitude up to the instant 27r, when the left one turns to zero and after this instant it changes its direction. This fact is known as a “swimmer efeci!?. The graph of the function u ( x , t ) on the rectangle R1 = { ( z , t ): 0 5 x 5 8n,0 5 t 5 47r} is plotted in Figure 3.6.

Exercises

a

1. Prove that for a function f (x) E C2(R+) its odd extension C2 (R) if and only if f (0) = f“ (0) = 0.

fo

(x) E

2. Solve the problem

{

U t t - u,, = 0 u (x, = sin3 x, ut (x, = 0) 0) u (0, t ) = 0

0

o

< x < O0,t> 0, o < x < m,

t 2 0.

Prove that the solution u (x, ) E C2((0,00) x R). t

84

Partial Differential Equations

3.3

Mixed Problem for the Wave Equation

Let us consider the problem (CW) on a finite interval [O,Z] with Dirichlet boundary conditions at the end-points z = 0 and z = 1. This is the problem

( M D W ):

{

U t t - c2uxx = 0 , 0 < z < z,t > 0, 0 5 z 5 I, u (z, 0 ) = $9 (x) U t (z, 0 ) = 1c, (2), , u (0, t ) = u (1, t ) = 0, t 2 0.

It can be interpreted as vibrations of a string with clamped ends, for instance vibrations of a guitar string. We can get the solution of the problem ( M D W ) again using the method of reflection in this case through both ends. We extend the initial data cp(z) and (z) given on the interval (0,Z) to the whole line using “odd”extensions q e o (2) and q e o (2)with respect to both sides z = 0 and z = I, where

qeo

(x):=

o < z < I, -1 < 12: < 0, extended to be of period 21.

cp(x) -cp(-z)

Consider the problem (CWeo):

By Section .3.1 it has a solution

x-ct

Its restriction

gives the unique solution of the problem ( M D W ) . Note that the solution formula is characterized by a number of reflections at each end z = 0 and II: = I along characteristics through reflecting points. They divide the domain R = { (2, ) : 0 < x < 1, t > 0} into diamond-shaped domains with sides parallel t to characteristics and within each diamond the solution u (z,t ) is given by a different formula. On the data cp and 1c) we impose the compatibility condition

One Dimensional Wave Equation

85

cp(0) = $9 (1) = 1c) (0) = 1c) ( 1 ) = 0.

(3.11)

In this case the solution u ( x , t ) is a continuous function on R. Note that u ( x , t )E C 2 ( ~ ) if

cp(0) = cp (1) = cp"(0) = cp" (1) = @ (0) = $ ( I ) = 0.

(3.12)

We can do the same for the problem with the Neumann boundary condition, considering even extensions of initial data. Namely, let us consider the problem

0 < x < I , t > 0) U t t - c2uxx= 0 0 < x < 1) u (X) ) = cp (x), U t (2) ) = $ (x) 0 0 t 2 0. u, ( 0 ) t )= u, (1,t)= 0

In this case we reduce the problem ( M N W ) to tions p e e (z) and 1c)ee (z) ,where

(CWee)

with initial func-

(Pee

(x):=

**(P(Z), O<x<l, y(-x) -1 < x < 0) extended to be of period 21.
**

)

As before the problem (MNW) admits the unique solution

where w (x,t) the solution of the problem is

Example 3 6 Solve the problem ( M D W ) with c = l,$ = 0 and ..

p(x) =

{

COS~ z

0

zE [(3~/2,5~/2)], x E [ 0 , 3 ~ / 2U ( 5 ~ / 2 , 4 ~ ] . )

Solution. The solution of the odd extended problem

x E R, t > 0, x E R, (x) = p e o (x) 0) x E R, V t ( a ,0 ) = 0

vtt

- v,,

=0

7 using the Mathernatica program f[x-]:=Which[3Pi/2<=~<=5Pi/2.f[x]] u[x-.0.True. a.{ t .t-]:=(81[x+t]+go [x-t])/2 Plot3D [u[x.1} .4Pi}.True.True. AxesLabel->” Position” .”Value”.”Time”.7.Shading->False] Vl.86 Partial Differential Equations The original problem has a solution The graph of the function u (x. PlotPoints >40.f[x]] g l [x-]:=Which[x>4Pi. .O] gO[x-]:=Which[x<O.> { -1. . ) on the rectangle t is presented in Figure 3. Reflection of a wave. PlotRange. Figure 3.-f[-x] .4Pi}.Cos[x] ^3.t]{ x.O.-f [x-4Pi] .

problem 3) { 0 < x < 1.t > 0. Find the values u ($. where u ( x .O)= p(x) x E R.c2uXx f = It can be split into two problems .c2u.O < t < 4n) using Mathematica.which we solve. ) = 0 0 t 2 0. = 0 0 < 12: < 1. Utt .t) on the rectangle 4n.t > 0. ut (x.t) : 0 <x < 3. (3. (0. x E R.. t > 0.t ) is the solution of the . u (z. x E R. when both turn to zero and after this instant they change their direction up to the instant 4n. 0 ) = z2(1 .one homogeneous with nonzero initial data (CW).t) = u x (1.O) = + ( x ) .13) .One Dimensional Wave Equation 87 Note that the initial profile splits into two profiles with amplitude up to the instant 27r. Solve the problem Plot the graph of the function u ( x . u. x E R. and one inhomogeneous with zero initial data Utt . = f u(x.t ) = 0 2.l) u (j. 3 Exercises 1.z). ut(x.O) = 0 u&O) = 0 x E R . u(z.u x . R = 3 { ( x .4 Inhomogeneous Wave Equation Let f E C'(R2) and consider the inhomogeneous Cauchy problem Utt x E R.

in view of (3. 7 7 7) where (3.14) with respect to 7 we have which.t) ul(z.t) uz(z.16) Let us make change of variables s = z a .15).13) into 4 = -gF(E. = u+cr. Making change of variables = x +ct. yields Integrating the last equation with respect to < (3.13) respectively.cr.15) Integrating (3. 1 we transform (3. and 212(x. = + is Let us consider (3.88 Partial Differential Equations If u1 (x.13).t) are solutions of (CW)and (3. 7 = x .ct. .14) (3. then t) u(x. .t) a solution of ( I C W ).

:~ ) c(t . z ) : rl 5 s < z. 0 5 T _< t}.One Dimensional Wave Equation 89 a(s 4 = which has Jacobian J = .t ) = 1 -(cp(X 2 + ct) + cp(X . t) denotes the characteristic triangle. xIndeed by.cr 5 u + cr < x -k ct. 7 < s 5 z 5 5.18) Note that from (3.ct)) (3. we have x . Then it follows 0 5 2cr and < 2ct e 0 < r < t . <a x-c(t-7) The solution of (3. for 0 5 t 5 T . 4 D into = { ( s .T ) 5 0 _< x + ~ (-t T ) . as in Section 3. The last change transforms the domain of integration w. rl< z I t} D‘ = ( ( 0 .ct 5 u .1. We prove that problem (ICW)has a solution given by the exact formula u ( z .13) is < z+c(t-7).2c. Indeed. we have . where A(x.18) it follows the well-posedness of ( I C W ) .

90 because Partial Differential Equations JJ A d a d r = S(A) 5 - 2cT. x E R.O) = cosx ut(x.u. = xt x E R. x E R..t > 0 . t > 0.O) = 0 x E R.. = sin x x E R.T 2 Example 3. ut(x. u(x.t > 0 . . The solution is t r(t . u(x.7.O)= x x E R. x E R.O)= 0 utt .u. x E R. Solve the problems (4 { (b) utt . x ER.u.r ) d 7 = x ( : - %> Exercises 1.O) = 0 Solution.O) = 0 ut(x. = ex-t u(x. Solve the problem Utt ..

w . = f ( X ) t ) . = t U ( X . George Green.~] . (a) Prove the formula where f ( x . u(2. ut(x.w.17) applying Green’s5 identity c2vdt L A + udx = JJ A (ut .c2uzX= f ( x .1793-31. One program is g [w :=Integrate [f [v .0 ) = 0) U t ( X )0 ) = 0 ) where f ( x .t ) and a(t) are differentiable functions.where dA is the oriented boundary of the characteristic triangle A = A ( x .t).u. Prove the formula (3. 3.v .03.t ) .t .07. :=Integrate [g[w]. . (b) Verify that the function satisfies the problem Utt . > 0 .x. x E R.c2wX) dxdt.. 14.One Dimensional Wave Equation 91 (4 Utt .2] Simplify[ut t-WM] Give another program.t ) E C1(R2). .1841. 0 ) = cos x x E R.w] .x. Check the problem using a Mathemutica program.O) = 0 2.2] t] UXX=D[U[X. to the equation utt .t] t /2 -1 -1 + + ut t =D [u [x.u X x x2 x E R.x t -w] t u [x-.O .

= 0 u (z.c2ux. Proof.5.(U&) OX d . The sum of the kinetic and potential energy E ( t ) = K E ( t )+ PE ( t )= 0 (u: ( z . 0) = u (0. Ut (z. the quantity 1 PE ( t )= 2 / c 2 u 2 ( z . Theorem 3.92 Part id Differentid Eq uations 3. t ) = u ( 1 . 0<x t 2 0. and u E C2 ( R )be a solution of the problem (MDW): { . For the above mentioned problem ( M D W ) we show that the total energy E ( t ) is a constant independent o f t .t > 0 . t ) = 0. t ) ) dz + is the total energy of the system at the instant t.. Utt + (x) 0 <x < 1. t ) c2u2 ( z . using the identities UtUtt = -- I d 2 at (UP) 1 UtUXZ = . t )dx 0 1 is the potential energy. Multiplying the equation by ut . < 1. The conservation of energy is one of the most basic facts about the wave equation.) d z t 0 is known as the kinetic energy. The quantity 1 K E ( t )= 2 / 1 u. 0 ) = cp (z) . t ): 0 < x < I .5 Let Conservation of the Energy R = { ( x . If u E C2( R ) is a solution of the problem ( M D W ) .-2 at 18 (4) . then the energy E ( t ) is a constant E ( t )= E ( 0 ) . 0 < t < OO}. (a. This is the law of conservation of energy.

t ) + c2u2(z. t )U t ( I .- (0.19) = 1 1(q2(z)+ ~ ~ c p ' ~ ( adz. ( 1 (x.dE . t ) dt ' Therefore for t > 0 E ( t ) = E (0) = 0 1 (u: (x. H Exercises 1. jFkom (3. ) t + u: (x. 0 ) ) d~ (3.19) it follows that if cp = $J = 0 then u = 0 on R. ) 0 + C ~ U :(a:.) :) 2 0 so the energy is conserved. t ) )dx + C2UX 0 -c2ux (1. Consider the problem Prove that the energy E ( t )= f 1I . t )U t (0. t ) .t ) )da: 0 ..c2u.One Dimensional Wave Equation and integrating by parts we get 93 0 = 1 1 (utt .) utdx 0 1 - Id 2 dt / 1 (up ( x .

(c2ux2Lt).c2uxx= 0 .(ul 2 + c”u2). q) dx..t) : a .94 Partial Differential Equations is a decreasing function.) = . Let u E C2 (R2) be a solution of the wave equation utt .O 5 t 5 T}.O 5 t 5 T } .2’) dx 5 (uf + c2u2)(x.)~(x. (a) Using Ut 1 (utt . a-cT (b) From the last identity it follows b+cT (c) Applying Green’s identity in DT prove that (up atcT + c2uE)(x.O)dx b+ cT (ut .t)U + ct 5 z 5 b : ct.)~ (x.C2UX.ct 5 x 5 b + ct. y)+ 1 dx b b (ut + CU. )dx u T a-cT = 1 s b a (ul + c2ui)(x. and DT = {(x. 2.CU. ) dx. 0 . and Green’s identity in DT prove that b+cT 1 ( :+ c2u2)(x. DT = {(~.

) t -ca + U: (z. Consider the problem and suppose that f E C1 (R x R+) n L2 (R x and $ E C1 (R) n L2 (R) . t Prove that t +m . 0 -ca .One Dimensional Wave Equation 95 3.) )dx.'p' 'p E L2 (R) (u: (2. E C2 (R). Let Rs).

This page intentionally left blank .

t)the temperature of the bar at the point x at the time t . The constant k = a2 is known as the themnal conductivity.1). Water 0.Chapter 4 One Dimensional Diffusion Equation 4. In order to determine the temperature in the bar at any time t we need to know: (1) initial temperature distribution where cp(z) is a given function. Aluminium 0. we consider a thin metal bar of length l whose sides are insulated.z = 0.86. The units of k are (length)2/time.1 Maximum-minimum Principle for the Diffusion Equation In this section we consider the homogeneous one-dimensional diffusion (heat) equation Ut -a 2 u.0014. For instance.14. Denote by u(x.71. Copper 1. we assume that the temperatures at the ends are fixed 97 . (2) boundary conditions at the ends of the bar. Some values of k are as follows: Silver 1. (44 which appears in the study of heat conduction and other diffusion processes. As a model for equation (4. The parameter k depends only on the material from which the bar is made.

so that no heat can pass through them. (4. We can also assume that the ends of the bar are insulated.3) u(0. 0 < x < I .98 Partial DifferentialEquations u(0. 0 and the boundary conditions <x <I (4. ) : 0 5 x 5 I.(Z. u(Z.t ) =min u(x. while (4.t)= u(Z.t ) .7) min u ( x .4) u. (4.(Z. ) satisfies the initial condition t u(x. tR Theorem 4.O) = cp(x).3). t > 0. = 0.(O.5) The problem (4. Let u(x. known as the maximum-minimum principle.. (4. However it turns out that it suffices to consider the case 7’1 = 7’2 = 0 only. t > 0.t) = 0 .(O. Then max u(x.5) as the Neumann problem. t Let R = { (x. ) be a continut ous function in R which satisfies equation (4. A “well posed” problem for a diffusion process is ut . 0 5 t 5 T } be a closed rectangle and I? ={ ( ~ .4) is known as the Dirichlet problem for the diffusion equation.t) = u. ) =max u ( x . At first we discuss a property of the diffusion equation. t > 0.t)= 0 . (4.) .t)= TI.2).t)= T2. (4. (Maximum-minimum principle).2) in R\r.1. (4. t > 0 or (4.t) = u. r t R .6) (4.t) = 0 . which implies u.ku. t R r (4.2) where u (x. t > 0. E ) : t = 0 or x = 0 or z = I } .2).3).

t ) attains its maximum at a point of d R = I U y.t l ) 2 0.t ) 5 R M which implies (4.where E is a positive constant.x2).(Z. wzz(iE.7).on R for every E > 0.T) = 0.y = { ( z .T) =lim 640 v(z.T)5 0. t ) . Consider the function v(z. Therefore v ( z . 0 < Z < 1.1. Proof of Theorem 4. T )E y.6). As v(z. L w ( q . H we By the maximum-minimum principle it follows the uniqueness of the solution of the Dirichlet problem for the diffusion equation . which implies u ( z .2 k = .T) 2 0. which contradicts (4. which contradicts (4.t) get (4.S) .One Dimensional Diffusion Equation 99 By Theorem 4. Hence = m a x v ( z . ) = u ( z . Suppose v(x. t ) R a 5M+EZ~. -6 Therefore Lw(z. Letting E 4 0. t ) E R : t = T } . 0 < S < T .t) E Rr \ L v ( z .t )+m2. T ) 2 0.t ) . t ) = mr x v ( z .S). T . T ) 1 w(z. t ) =maxu(z. we obtain u(z. t We have for (z.8) If w(z.8).1 the maximum (minimum) of u(x.t). t ) cannot be assumed anywhere inside the rectangle but only on the bottom or lateral sides (unless u is a constant).v(z. Denote M = m a u ( z .t ) 5 M on R which means that m a x u ( z .t ) attains its maximum at an interior point ( q t l ) it follows that .t) has a ' maximum at a point ( 5 . t ) 5 M &(Z2 . R + r Considering the function w ( z . r We shall show that max u(x. we have wt(%.8).T .2 ~ k< 0 ~ and (4. t ) = -u(z.t ) = Lu(z. Then v.

u(2. egg.t ) . w(0.t ) be a solution of (HDD) with initial data Y j ( x ) . The problem (IDD) has no more than one solution. j = 1. t )E R.o < t 5 T . Let uj(x. By Theorem 4. Then { wt . Then . As a Corollary of Theorem 4.t)= 0 OSzSl. t ) =min w(z.100 Partial DifferentialEquations Suppose By a solution we mean a function u E C(R) which is differentiable inside R and satisfies the equation along with the initial and the boundary conditions of ( I D D ) .1. Proof. = 0 0 < 2 < 1. R R Therefore w(z.kw. so that u l ( z .t)for every ( z . t ) = 0 .=O O<x<l.1 it follows max w(z.t) = w(1. Suppose u1 (z. Theorem 4. t )E uz(z.t ) .O)= p(z) u(0.2.2. t ) = 0.1 the continuous dependence of solutions of (HDD)with respect to initial data follows. ) are two solutions of ( I D D ) .O) = 0 OSz<l. Corollary 4.O<t<T.t)= u(1. Consider the problem ( I D D ) with f = g = h = 0..that is Ut-kU. w(z.. t ) = u1(z. t ) and u2 (2. Let t w(z.u2(z. t) = 0 OLtST.

l ( 4 -(P2M w ( 0 . In contrast to the wave equation where the energy is a constant. u2(x.One Dimensional Diffusion Equation for 101 every t E [0.which satisfies z [ and W ( Z . Let u be a solution of the problem ( H D D ) . ) t be a solution of (HDD) corresponding to the initial data ( ~ j ( x ) . O )= V . Then j . J The uniqueness and stability of solutions to ( H D D ) can be derived by another approach. Wt By Theorem 4. The quantity H ( t ) = J.u ~ ( t ).t)dz is referred to as the thermal energy at the instant t. t )= w ( l .2. known as the energy method. tbe a solution of (HOD). OLX<l. t ) = 0 OLtST. ( a ) Let u ( ~ . .1 it follows that 2 .TI.kwxx = 0 0 < 2 < l . Then ) H(t1) 2 H ( t z ) .5 for the wave equation. Pl Theorem 4.V2(Z)l which imply (4. (b) Let uj (x. if 0 5 tl 5 t 2 _< T.O S X < l I V l ( 4 . We have already used this method in Section 3. Proof Consider the function w ( x .t ) = q ( x .t ) .3.9). o < t 5 T. we shall show that H ( t ) is a decreasing function. = 1.

so if 0 5 tl 5 t 2 5 T . t ) .t)= 0 o < x < 2 .O)=x(2-x) u ( 0 .=O u(x.) t ) ) (0.t ) satisfies (HDD) with cp(x) = 9 (2) .t ) . t ) = u(2 . dX and integrating.t ) d x ) . 2 --(t). using UUt = 1 8 --(u 2 at 2 ).102 Partial Differential Equations Proof.k((uu.u2(x. t ) = u(2. d uu.. . Therefore for t 2 0 by (a) 1 1 dH 2 dt Exercises 1. Show that: (a) 0 < u ( x . t ) = u1(x. (a) Multiplying the equation by u. we obtain = ( i g ( u 2 ) k-(uuz) . 0 5 t..d + ku: - 2 dt (Jdl dX ) dx u 2 ( x . OIx52. = -(uuz) -u2. (b) u ( x . O<t.x.) ( 1 . Therefore H ( t ) is a decreasing function. ( c ) u 2 ( x .cpz(x). (b) The function w(x.t ) d x 5 for every t 2 0. Consider the mixed problem for the diffusion equation { Ut-U. t ) < 1 for every t > 0 and 0 < x < 2.(uu. then H ( t 1 ) 2 H(t2).t)for every t 2 0 and 0 I x 5 2.

.t)= -2xt . Conversely.t ) = u(-1.(x. t) (b) Using the Cauchy-Schwarz inequality derive that H’2(t) 5 H(t)H”(t). 0 5 t 5 1) has a solution u(x..2 The Diffusion Equation on the Whole Line In this section we give an explicit formula for the solution of the Cauchy problem for the diffusion equation on the whole line We shall prove that the solution of ( C D ) is given by the Poisson’ formula (4. 4.10) that the value of u (x.d z . (c) Show that for every 0 _< tl < t < t2 5 T t2-t the inequality t-tl H ( t ) _< H(t1)t 2 .One Dimensional Diffusion Equation 103 2.x2 and max u(x. the value of cp at a point 20 has < Simeon Denis Poisson. = 0 in the rectangle R = {(x.1781-25.t)dx and P ( t )= 4 1 1 Uf(II:. 1) = 1.04. holds.t): -2 5 x 5 2. 2 1. (a) Show that 1 H’(t) = .06. The maximum principle is not valid for parabolic equations with variable coefficients. Verify that the equation ut. R 3.10) assuming that p(z) is continuous and bounded on R.1840.2 k i u2. Notice from (4. Consider the thermal energy H ( t ) of the problem ( H D D ) . ) depends on the values of the t initial data cp(J) for all E R.t l H(t:!)t 2 .xu.t l . known as logarithmic convexity of H ( t ) .

then the function I(y) is continuous in [a.b].b]. be a continuous function in (5.b]. b ) and I'(Y) =49). I(y) is convergent for every y E [a.104 Partial Differential Equations an immediate effect everywhere for t > 0. Theorem 4.10) is the Fourier transform. .5. Theorem 4. y) E R x [a. If the integral (4. Recall some definitions and proper ties. This effect is known as infinite speed of propagation which is in contrast to the wave equation. Suppose the y) integral I(Y) = is convergent for every y E [a. for y E [a. i f for every E > 0 there exists A0 = A o ( E )such that if A > A0 . we would not expect well-posedness of the backward-in-time problem for the diffusion equation. then for every y E [a. Let f (x.10) is infinitely differentiable for t > 0.b].y)dx (4.b].11) is uniformly convergent . Srn -W f(x. Then I ( y ) i s a differentiable function in ( a . and g ( x . A natural way to derive (4.1. W e say that the integral (4. It is known that the diffusion is a smoothing process going forward in time.b].10) satisfies the problem (CD) we need some preliminaries on improper integrals. Moreover the solution given by (4.b] and is uniformly convergent for y E [a.y) are continuous functions in y) R x [a.b]. Going backward (antidiffusion) the process becomes chaotic. but we do not consider it in our text.11) Definition 4.11) is uniformly convergent f o r y E [a. Suppose f (x.4. A criterion on uniform convergence of integrals is the following.b]. In order to prove that (4. Therefore.

31. Definition 4. We have 1 u ( x .b].14) > 0 be fixed.15) 2Karl Theodor Wilhelm Weierstrass. A differentiable function u ( x . y)I 5 g ( x ) f o r every y E [a.b] and the integral 00 is convergent. = 0 in R x (0. As p ( x ) is continuous.10) defines an infinitely differentiable function u ( x . Then Poisson fomnula (4. Proof.c (cp(x .11) is uniformly convergent f o r y E [a. t ) is a solution of the problem ( C D ) i f it satisfies the equation ut .12) is fulfilled.One Dimensional Diffusion Equation 105 Theorem 4.13) we obtain Let us show that (4. 2 f i -00 By Poisson identity J-00 m we have (4.t ) which is a solution of the problem (CD) and lu(x.7. t ) = -J r ne-gcp(x . t ) = cp(x) tlo (4. Then the integral (4. Note that the formula (4.y ( x ) = -/ 0 0. T ) and lim u ( x .10) has a meaning for t > 0 and the initial condition is satisfied in the limit sense..t)l 5 M. .02.6. From (4. t ) .1897.10.ku. there exists S > 0 such that or (4.cp(x))dp.p&) e 2 f i -00 Let E .p&)dp. (4.1815-19. Let cp(x) E C (R ) and Icp(x)I 5 M . making the change of variables 5 = x . (Weierstras2 criterion) Suppose there exists a function g ( x ) such that If (2.101.2.12) Theorem 4.p u ( x .

. A] x [S. As & & it sufficies to show that Suppose that (x.12) is fulfilled.>& if 0 -2 4dp<- 2fi P I ‘S P k e 4dp<4M ’ & -z (4. where A > 0 and 0 < S < T are fixed.106 Partial Differential Equations As the integral s-”.16) for t E (0. (4.ku. (4.. 2 which means that (4.17) we need to prove that the last integral is uniformly convergent for We have (X.Let Lu = ut .t) E s.to) we have < .5.G d p e is convergent there exists a sufficiently small t o such that &Lp. t )satisfies the equation ut . By Theorem 4.16) < t < t o ..T).18) .15) and (4. = 0 in R x (0. It remains to show that u ( z . ) E S := [-A.14).+ -2= = . t TI. by (4. in order to show that (4. Then.e .ku.

2x2 8kt . T .6) is a constant.2A2 8kT (4.18)) (4. Following the same way we show that . Note that the integral t2e-t2dC is convergent and s-". we have where C = C ( M .19) Then.6.A .19) and Theorem 4. so it follows that the integral (4.13) is uniformly convergent.One Dimensional Diffusion Equation 107 I- x2+p 2kfl 1 +-2@ I jF'rom the elementary inequality A2 -+-+ lt 2 2kf 1 2@' it follows ( x .<I2 4kt 'p tz ' . by (4. k .

H The initial data p(x) in (CD) a continuous function in Theorem 4. As A > 0 and 0 < 6 < T ut . which is a solution of the equation Ut . .8. Theorem 4. It is can be supposed p(x) to have a jump discontinuity.4. A function p(x) is said to have a jump at 20.T]. t )E [-A. Then formula (4. = 0 .. are arbitrary = 0 .ku. t ) .0)) for all x E R. X. T ) t and 1 lim u ( z ..108 Partial Differential Equations and the last integrals are uniformly convergent. Let p(x) be a bounded piecewise continuous function. t which completes the proof.A] x [S. TI.t ) = -(p(x tl0 2 + 0) + cp(x . Then ut . ) E R x (0.10) defines an infinitely dinerentiable function u ( x . if ( z .kuxx = 0. ) E R x (0.0) = lim p(x).ku. if both left and right limits of p(z) exist p(x0 .X Q 2 <"O p(z0 + 0) = lim p(x) x-xo ">"O and The function cp(z) is said to be piecewise continuous if in each finite interval it has a finite number of jumps and is continuous at all other points. (x. if (x.

21).One Dimensional Diffusion Equation 109 Proof. .t>O x E R.20). i f O < t < t o .o)( < E . Then < . Suppose E > 0 and 6 > 0 are such that I'p(z0 .(1 h + 2 M . Let us prove (4. i f 0 < p <- 6 &' As e . t > 0. Solve the problem -uxx =0) u(z. such that e -2 d p < .G d p = fi there exists t o > 0. 4 J. .O) = e-".20) (4. as before.As in the proof of Theorem 4.1. Example 4.Analogously it can be proved (4. which completes the proof.p(z0 .7 we show that (4.) q 2 f i &fi& ) = 2M which proves (4.. 2M ' where. Ut z€R.21) as t -+ 0.p a > .20). Let xo be a point of jump discontinuity of cp(x). lcp(z)I 5 M for every z.

2.t ) = e-" for every x E R.4. (4.2tx + 6 2 + 4<t) + t2+ 4t2 . already used in Section 3. . tl0 Example 4. 1 " e-p2dp.t and making change of variable Observe that lim u ( x .22) erf (-x) = -erf (x). we have (t + 2t . Note that erf(O) = O . Solve the problem Solution.2 t x + 4 t t . We express the solution in terms of the error function of statistics erf(z) = 2 J.x)2 + x . 2++w lim erf(z) = 1.4xt + 4xt . By the Poisson formula we have Using (x4t +< = = - -(x2 1 4t 1 -(x2 4t .4t2) 4t (+2t-x 2& = p .110 Partial Differential Equations Solution. Example 3..1.

= -(u(l 2 2 1 1 + 0.8 and (4.. Example 4.2&p we have .O 2 e- 4t dJ .O)). Solution. Making the change of variable < = x . = 0 with initial data Show that u ( x .0.One Dimensional Diffusion Equation 11 1 =x -2 By the Poisson formula and the change of variable u(x.2& J'7 pe-p2dp+2&/:pe-p2dp 2dT I-1 2 t .22) we have - .= 1 (erf 2 (s) (2)) -erf e-p2 dp . Solve the Cauchy problem for the diflusion equation ut u. t )-+ 0 as t -+ +oo for every x .O) + u ( l . By Theorem 4.t) = 4p -/ 1 2fi 1 -1 ( X .3.

Graph of the function u = u(x. Note that for every x E R as t --+ +oo.3.t ) in Example 4.1.1 using MAPLE in ScientificWor kPlacE. J Figure 4.112 Partial Differential Eq ua tions The graph of this function is given in Figure 4. Indeed for x 2 0 .

The same way & J" that pe-P2dp -+ %+ o as t + +oo. 2 4& (x+ 1)2 4& as t + +m. Then by erf(0) = 0 it follows that lim u(x. .2. 1 x2 (-( x + 1)2 .t) 0 for every x E R.t)= erf (L)i 2f +2 6 (e-h .1) . If -1 <z <0 = 1 x2 -(-+-)+o. The graph of this function is given in Figure 4. Note = t++m u(0.-) 4 4 4 4 + 0.One Dimensional Diffusion Equation 113 pdp = 2 as t -+ +oo. Finally for x 5 -1 as t -+ +oo.

Graph of the function u = erf (h) 2 + 8 (e-h . { + 4. t+m Compute lim u(0.t ) . Solve the Cauchy problem for the diffusion equation with convection U t .1) Exercises.t)is also an odd (even) function of x. Solve the Cauchy problem for the diffusion equation with constant dissipation U t . (b) u ( x . 0 ) = c p ( ~ ) . 2.O) = 4. + .O) = ~ ( x ) . ) E R x (0.114 Partial Differential Eq uations 0 1 2 t 3 4 5 Figure 4.2. tl0 2. (b) and show that lim u(0. (2. (4 +. if x > 0. (x.t) = 0. then the solution u(x.x E R.0) = e-lzl. = 0. OO). if x < 0. Compute u ( 0 . Show that if (p(x)is an odd (even) function. = 0 . Solve the Cauchy problem for the diffusion equation ut . u(x. t > 0 with initial data (a) u ( x .x 2 . t ) in the cases (a).u. 3.kuxx WU. Consider the Cauchy problem for the diffusion equation with the initial condition u ( x . 1.. t U ( X . t ) E R x (0. x E R.0) = cp(x).kuxx bu = 0 .0 ) = e . OO). x E R.

Note that u.23) We are looking for a solution formula for (4.which easily . Let us consider the problem ( C D ) with initial data p. Using the reflection method considered in Section 3.One Dimensional Diffusion Equation 115 4. t is The restriction is the unique solution of the problem (4.2 for the wave equation we shall treat the problem Ut +. +w).23) analogous to the Poisson formula.24) where Let u o ( x .(z. t > 0.24) which.kuxx = 0 = P(Z) u(0. -too).co) and take the Dirichlet boundary condition at the end-point x = 0. which is the odd extension of p(x) on the whole line (4..O) .23). t ) . E (0.(x. t ) = -u. (4. by the Poisson formula. ) be the unique solution of (4.t) satisfies the diffusion equation and is an odd function uo(-x.3 Diffusion on the Half-line Let us consider the diffusion equation on the half-line (0. 0 5 t.t)= 0 x E 2 (0.

we have The restriction u(x.t) 0 = Ut (4.t) ax aU = -(O.t > 0.kuxx = 0 x E (0.t) = 0 and u(a.25).0) = c ( p4 t > 0.t)satisfies the diffusion equation.t) d X due As before.26) In this case we use the even reflection of P ( X ) Let Ue(x. u(z.t) satisfies the initial condition for x > 0. . -too). Moveover. t ) satisfies the diffusion equation and the initial condition.(O.t)be the solution of the problem As before. u ( z .is the solution of (4. . Let us consider now the Neumann boundary condition at the end point x = 0 for the diffusion equation on the half-line. UX(0. Namely.26).t)= u. +m). z E (0. Then u(0. let us consider the problem .t ) = ue(x. U ( X . Note that t)lx>O and -(O.116 Partial Differential Equations follows from (4.

)2 +t ) Note that u ( 0 . By the solution formula for (4.et (erf (4) erf (-4)) + = 0. Solution. t ) = ..)2 +x-t -(x (X+tl2 +t (t + + .5. Solve (4. Solve (4.26) with p ( x ) = e-“ and k = 1.4.23) 117 Using (.26) and previous calculations we obtain that the solution for x 2 0 i s .One Dimensional Diffusion Equation Example 4.23) with cp(x)= e-” andk = 1. By the solution formula for (4.-El2 4t 4t we obtain +[ = = (< + 2t 4t 2t 4t . Solution. Example 4.

.118 Partial Differential Equations As erf'(x) = L.26) show that e-"b@(a .(O. x E R. then the solution u ( x . 2 2.t) and w(x. t ) such that { Ut U(X.e . x 2 O .T 2 5 @(x)5 1 .27) known as Cauchy problem for the inhomogeneous diffusion equation.. (4.t ) is the u.2 d p be the density function of the standard normal distribution.the solution of (4. 1. Show that @(+m) 1 and @(-x) = 1 .28) .hXx =0 v(x. By linear properties of the operator Lu G ut . (4. b > 0. 4.O) . . t ) = 44 x E R.26). = ) (a) Derive that the problem (4. (b) Using the maximum principle for (4.k .t) of the problems wt . If cp(x) is a bounded continuous function. Let @(x) = e. t ) of (4. a E R.26) with the initial condition ~ ( x = e-x has a solution s-. sum of the solutions w(x.x 2 we have Je .ku.27) u(z. t > 0. Prove the following maximum principle for the problem (4..b) + eab@(-(a + b ) ) 5 e-g.4 Inhomogeneous Diffusion Equation on the Whole Line Consider the problem of finding a function u(x. Exercises.@(x). u. = f (2. x E R. t > 0.O) = v(x) x E R. 1 2 322 1+-e--'i.t) = 0.26) satisfies I 4 X ) t>I<SUP XZO lv(x)I * 1 .

oo))be a bounded function and t T h e n the function w(x. E)R X( o .kw. (4.t-~))=o. t) J: E =0 x R.we prove that the function w(z. By the maximum principle for the diffusion equation on the whole line it follows that the function ~ ( x ). t ) given by (4.given by (4.27) is Assuming that f(x. t jFrom (4. t . = f W(Z'0) (IL'.t f ~ .31) it follows that the solution of (4. t Theorem 4.33) is the unique solution of (4. t ) = is a solution of the problem (4. Let f (x. ) E C(R x (0. E R. ~ ) .29).30) and will show that Denote known as Green's function or fundamental solution of the diffusion operator L.27).30) and (4.29).t > 0.9. oo).One Dimensional Diffusion Equation and 119 wt .29) respectively.7 ) d T . ( ~ . We have that (4.t ) is bounded and continuous on R x (0. Ju t w(x..31) satisfies the problem (4. It is clear that L(G(x-t.

x (4.7 ) d 7 + lim v(z. t .34) u(z. . t ) = 0.120 Partial Differential Equations Proof. t ) u(0.kuzz = f ( ~ . By virtue of the estimates of Section 4. t)I 5 I’ 142.O)= ( P ( 4 > 0. t .t .t. t > 0. t .4 d7 1 and therefore lim w(z. t. tl0 Consider now the inhomogeneous problem on the half-line with the Dirichlet boundary condition Ut .T)dT= d t-& = 10 vt(x.t . t > 0. Jd t u&.E ) &+O To show that the initial condition is satisfied observe tiiat M X .7 ) d T I”Ut(X.t)= h(t) x > 0. are uniformly convergent over bounded and closed intervals of R. t .2 and the assumption of the theorem the integrals It and we have v(z.d d 7 . By the formula for differentiation of integrals depending on parameters aW at - Lv(z.

35) is the sum v(z.37) The solution of (4. where vl(z. Namely. we have .One Dimensional Diffusion Equation 121 We reduce (4.t) = 0 for t > 0.t ) satisfies the problem The solution v(z.t ) . Note that (4.t). let Fo(x. t.36) and ~ 2 .t ) = f ( x . z E R.34) to a simpler problem letting v ( x .kv2Xz = f t ( ~ .t)of (4.t ) be the odd extension of F ( x .37) can be solved again by the reflection method using the odd extension of the source function F ( x .t) = 0 t > 0. t ) .kwZx = F o ( x . x > 0. t ) be the solution of the problem wt .t) +v2(z. x E R.h(t).h’(t).O) = 0 .h ( t ) x > 0.O) = 0 (4.36) is found by the reflection method of the previous section.t ) = u ( x . t ) with respect to x and w ( x .t ). )’ w2(0. t > 0. given by Then is the solution of (4. t >0 w(s.Then v ( x . t ) and v2(xjt ) are solutions of the problems (4. v2(x. Let us show that v2(0.t)= vl(z.37).

Solve the inhomogeneous Neumann problem on the half-line .122 Partial Differential Equations because G ( z . t )is an even function with respect to x. Exercise 1.

The other is the sequential approach. It was noted in Section 3. Let L2 (R) be the usual Lebesgue space of square integrable functions u :R R. Consider approximating problems with smooth data ( c p k .1 that to have a solution u E C2(R x R+) (CW)we require cp E C2 (R) and $ E C1 (R) . where R is a measurable domain in R". We prove that in some sense these two approaches are equivalent. + k ) E C2(R) C1 (R) . It is possible to define x a weak (' generalized ) solution of the problem by passing to the limit in L2 spaces of corresponding solutions U k . all functions p (x) E 123 . How to justify the meaning of a solution in this case? There exist two main approaches. Shock Waves and Conservation Laws 5.1 Weak Derivatives and Weak Solutions Consider the Cauchy problem (CW). of If the last assumptions are not satisfied then the solution given by D'Alembert formula is not a classical solution. i. One is to introduce the so called weak derivatives so that the wave equation is satisfied in a form of integral identity.e.Chapter 5 Weak Solutions.L2 (0)is a Banach space with a norm --+ Denote by CF (0) the space of test functions.

8Xj W2>l is a Banach space with the norm (R) By W..01. 1902-1982. 3Kurt Otto Friedrichs.l(0)we denote the space of functions u : R + R such that for every compact subset K c R it holds u l E W2i1( K ). 2Paul Adrien Maurice Dirac..oc(R) is said to be the weak a d xj derivative of a given function u E Lf?oc i f ($2) for every test function p ( x ) E C r (0) If u aU I E C1 (R) it is easy to see that the weak & derivative of u ( x ) is equal m to -. 08.E L2 (52) .1989.. such that for every compact subset K c 52 it holds U ~ K L2( K ). E Definition 5..1908-03. j = 1. A function v E L. Next (R) is the space of functions u : R -+ R.2d. Consider the so-called Dirac? kernels or mol lifying kernels or the fiedrichs3 mollijiers.10. dzj The Sobolevl space W2>l is introduced as the space of L2 (R) functions (0) OU u : R -+ R for which there exist weak derivatives .124 Partial Differential Equations C (52) with compact support " suppp = { X E R" : p ( x ) # 0).n.1. . 1901-1982. Every function of W29l(R) ~ can be approximated by a sequence of smooth functions with respect to W2y1 norm on compact subdomains of R. Let E > 0 and po (x) E C r (0) 'Sergej Lvovich Sobolev.

01. p. E Here 0 = 3 dlkl ax. is a partial derivative of order Ik(. Rn for which p.u tend to u in W2i1 ( K )for every compact K c R. (0) and as JEU Mollifications or Regularizations J.11. i.e. (x)dz = 1. 25. By the change of variables y = 2 .One Dimensional Diffusion Equation where the constant co is such that Rn 125 po (2) = 1.80) ..u of a function u E Lioc(0)are defined (4 = 1 n P (x . For a given function u E W?: (a) the regularizations J.Y) (Y) dY.184330. Proof.ax$ ' Ik( = kl + . + k.. Define the sequence of dx mollifiers 1 P (4 = E"P0 E ( ) .1.E Z and the Cauchy-Schwarz4 inequality we have *Hermann Amandus Schwarz..1921 . Theorem 5.. (z) E C r (0). suppp. (x)= B. Let E < dist ( K ..1 .

126 Partial Differential Equations or where K.) such that IIu . as €1 -+0.4IC(K) ~1 + 0.V . (a) it follows IIJ&.v l ) L z ( K E )< 6By Exercise 2. There exist v E C (K. Then for sufficiently small <E Moreover and then . = {z E 0 : dist (z. Let S < 5. K ) 5 E } .

Definition 5.2(R) -Exercise 4.U l ( W 2 . Definition 5.2 i s equivalent to Definition 5.c2uxx = 0 iff u ( p t t . 1 ( K ) -+ 0 as k 00.' (R2) i s said to be a weak solution of the wave equation utt . suppp C K and ( U k ) be a sequence of C2 smooth solutions of the wave equation such that .2. we have + .ld.c 2 p x z )dxdt = 0 J R 2 for every test function p E C r (R2).2. t ) E c2(R2) of the wave equation such that for every compact set K c R2.3.c2uxx = 0 iff there exists a sequence of smooth solutions u k ( x . A function u E W 2 . llUk . I ( K ) 3 0 as k -+ 00.3 3 Definition 5. Integrating by parts.u ) l l L 2 ( K ) ~1 -+ 0 -+ 0 as and for u E W.One Dimensional Diffusion Equation 127 A similar procedure is used to prove that IlDj (J.3. H Definition 5. A function u E W"d. Let p E C r ( R 2 ) .2. (R2) i s said to be a weak solution 1 dl of the wave equation utt ..u .u I I w ~ . Theorem 5. (a) Definition 5. Proof.

d X = dxdt.3 and the Cauchy-Schwarz inequality it follows ask-+oo.lluE U \ I W ~ . Denote for simplicity X = ( x . be Let u E W (R2) a weak solution of the wave equation in the sense of Definition 5.3. We use regularizations to construct approximating sequences of solutions. Therefore / K u (ptt .c2pzx)d x d t = 1 u (ptt .. By integration by parts and h b i n i theorem we have .2 and uE = J E u .c2p. t ) .-r). I ( K ) 0 as E -+ 0. R 2 (b) Definition 5. It remains to prove that uEis a smooth solution of --+ the wave equation. d Y = dyd-r. For every compact set K C R2.) d x d t = 0.128 Partial Differential Eq ua tions By Definition 5. Y = (y.2 + Definition 5.

14)" E L. < 1) for m < f. .4 I C ( K ) = zy I ( J & I + 0. (b) Let u E C" (0) and K c R be a compact set. Prove that .IXI)" I 1 (1 . Using .(X) 0 which completes the = proof. Show that the function 1 (1 . c be a compact set.One Dimensional Diffusion Equation 129 = Jm Jp€(X-Y)LxP(X)dXdY R$ R2.u) (x) 2.oc (1x1< 1) for every m. = JP(X)Su(Y)Lup. Rx 2 Ry 2 As p ( X ) is arbitrary it follows that Lxu. but E L~ (1. (a) Let u E C (n) and K IIJE. as E -+ 0. H Exercises 1.(X-Y)dYdX=O.

0 - .. t ) = 1x1. weak derivatives = { ( x . a weak derivative dU = 1 -1 ifx>O ifx<O du & B = { ( x . Prove that if u E W.= sgnx dX in B . prove that 5 Partial Differential Equations 3.2 Conservation Laws We consider solutions of hyperbolic systems of conservation laws. Verify that u has ' -.in B .: (R) . 8X dU 4. (b) Let u ( x . t ) = sgnx.130 for Jk) m. t ) : x2 + t 2 < l} . (a) Let u ( x . These are systems of PDEs of the form where u : R x R + Rm is a vector function and f : Rm -+ R" is a mapping . then 5. t ): x2 +t2 < l}.Show that u has not .

0. The mapping f ( u ) is called a flux function. t ) is the density. (pv).1707-18. The equations (5.1783.09. In one space dimension these equations are where p = p ( x . = 0. = = conservation of mass. i. Et + (v ( E + p ) ) . The system (5. 15-04. v is the velocity.e. there exists a complete set of m linearly independent eigenvectors. energy in fluid dynamical problems. pv is the momentum. The Euler5 system in gas dynamics is a system of conservation laws. conservation of momentum. + (pv2 + p ) .1) is hyperbolic iff the Jacobian matrix A=Jf(u)= 1 i I’ has only real eigenvalues and is diagonalizable. p is the pressure and E is the energy. . Introducing new variables the system can be written in the form (5.2) are known as pt + (pv). 0. momentum. conservation of energy.1) with Leonard Euler.Shock Waves and Conservation Laws 131 The function u describes physical quantities as mass.

t > 0 and .2) is hyperbolic in R3\{u : = 0).132 Partial Differential Equations Example 5. x3=-u2 U1 1. Then the system (5. Assume p (u) = u1. 212 U1 x2=- u2 U1 +1. The Jacobian of is the matrix A= with eigenvalues XI=-. because the determinant I0 0 1 1 1 1 -2. The simple initial value problem for the system (5.1) holds for 2 E R. Corresponding eigenvectors are which are linearly independent. Solution.1) is the Cauchy problem in which (5.1.

t ) = uo (x . 4. Solve the problem +u. Cos[x]-3.2] . PlotRange-> { 0.True. Ut = uo(x) = { c0s3 x 0 if x E [ 3 ~ / 2 . t ) at the instants t = 0. Example 5.O) = uo (4. It can be found by the method of characteristics of Chapter I. The solution is u ( x .1 using the Muthematica program f [x-] :=Which [3Pi/2 <=x <=5Pi/2. Let us consider the Cauchy problem for the simplest equation of the form (5.Shock Waves and Conservation Laws 133 U(X.t ) = uo ( x . 2. because Then u ( z . The problem (5.5) if the initial function uo ( x ) E C1(R) .m)). t )= u o ( x .1) in one dimension where a is a constant.PlotLabel->” Wave at t=O”] hl=Plot [Evaluate[u[x.{x. 5 ~ /. If U Q E Ck (R) then u ( x .t ) E C1(R x [O.5Pi}. The solution presents a right moving profile (graph) of the function uo (x) with speed a. known as the linear advection equation or one-side wave equation. Note that uo E C1(R). (5. The graphs of u ( x .5Pi}.O] u[x-.t ) E C k ( Rx [O.O] .5Pi/4. .4) has the unique solution u ( x .2}. (5.m)). 6 are plotted in Figure 5.{x.3) where uo (x)is a prescribed function. 0.at) .t ) .2 ] if x $ [ 3 ~ / 2 . 5 ~ / 2 ] .2.t-]:=f [x-t] hO=Plot [Evaluate[u[x.5Pi/4.

5 1.25 .25 1 0. PlotRange-> {0.4) in a weak (generalized) notion.5t 0.PlotLabel->” Wave at t=6”] Show[Gr aphicsArray [ { { ho. Suppose now that uo (2) is not a smooth function.25 : t 0.75 i 0.2}. FrameTics->None] 1.75 .hl} . PlotRange-> {0.{x75Pi/4. 2 1.75 . The wave u ( z .2}.75 1.75 2: 1.25 .5Pi/4.75 f 1.5! 1.{ h2.5) is not smooth and does not satisfy (5. 4.75 t 1. . It satisfies (5. 2. Then the function (5.5. A 6 1 1 0.2}.{x.PlotLabel->” Wave at t=4”] h3=Plot[Evaluate[u[x76] .5Pi).25 ! 0.5 ! 1.h3}}] .25 6 8 1 12 14 0 16 6 8 10 12 14 16 Figure 5.4]. Frame->True.1.75 0. An approach to generalize the notion of solution is to satisfy an integral identity. t ) at the instants t = 0. 6. 0.5Pi}.4) in the usual sense. 1. 1.25 t 8 1 1 14 0 2 16 \20 5* 1.5 0.5 i 0.PlotLabel->” Wave at t=2”] h2=Plot[Evaluate[u[x.134 Part id Differential Equations PlotRange-> {0.

) = uno (x . m)). integrating in [-T.oo)) t t and suppp [-T.T] on the lines t = T. t ) E . t . Assume that uo (x) E Lie.4) is defined as a L1 limit of the sequence un (2. ) i s a weak solution of the problem.4) by p. ) E Lioc(Rx [0.d t d x - ]]u T T (pt + up.e. i.4). ) E C.) dxdt Another approach to generalize the notion of solution of (5.4) with initial data uno (x).T]x [O. t Proof. T ] and using p (fT. T ] x [O. Let u (x. Then a generalized solution of (5.4) is to approximate the nonsmooth initial function uo (x) with a sequence of smooth functions U n o (2) . ) be a smooth solution of the problem (5. Ltoc(R x [O. that p = 0 outside of [-T.oo))i s a weak solution of (5. Let p (x. there exists T > 0 such that suppp C_ [-T. Obviously u (x.TI.m)) the space of C1 functions p vanishing outside of a compact set in t 2 0 . Multiplying (5.4. ) . A Proposition 5.4) zfl 0000 00 0 -00 -00 f o r every test function p E C (Rx [O. TI x [0. t T h e n u (x. ) = 0. we obtain t) T 0 = 77 0 -T T T (Ut +au. The function un (x.T] x [O. (R) A function u ( z .1. (Rx [O.at) is the solution of the t problem (5.oo)).x = -T so and and x = T .Shock Waves and Conservation Laws 135 Denote by Ci (R x [O. Definition 5.T]. = p (x.)pdxdt = 1J T T (up).

) . = 0 . u(x. where is the matrix of right eigenvectors (5. The system (5.at)l dxdt t = 0.uno (X .uo () . by R-lut + AR-lu. Unfortunately the sequential approach is not appropriate for nonlinear differential equations.7) Changing the variables v = R-lu.6) is is strictly hyperbolic iff the matrix A is diagonalizable and has m distinct real eigenvalues.O) = u o ( 2 ) . As an extension of one-side wave equations we consider linear strictly hyperbolic systems ut+Aux = 0.5. Let A = RAR-'. The function u = u (z.4) ifl 000r r 0 n-00 lim J J IU 0 -00 (2. we obtain . t ) E E L1 (Rx [O. b) that a strong solution is a weak solution. -00 It can be proved (Exercise 2. Here u : R x R -+ R".I d x = 0. f o r any sequence (un0)of smooth functions such that lim n-+w J Iuno ( x ) .oo))is a strong solution of (5. Assume that uo(z) L1(R) .136 Partial Differential Equations Definition 5. A E RmX" a constant matrix.

1 4 y3] 1 -2 fi(4 = {-0 sin3x x E [-ST.9). .8) are (5.Shock Waves and Conservation Laws 137 or componentwise The initial conditions to (5.sin3 x z E [r. 3 A=[. 01 .A k t ) t (5.-2T] ' .10) is the solution of (5. Solve the problem where u : R x R -+R3.3. 27r] .6) is m Example 5.sin3 x x E [-n.9) Then vk (x) ) = VkO (x .2 ~ 1 Ic # [-37r. The solution of (5.8)' (5.

t -1 :=f [x-2t ] .{ t . 1 and corresponding eigenvectors Then V = The graph of the function Mathematica program u1 (x.t].0.u] f[x-]:=Which[-3Pi <=x<= -2Pi.O] h[x-]:=Which[Pi <=x<= 2Pi.>False] . -Sin[x]^3. A3 = = 2."Value u1" . Shading.8Pi}.True.2 plotted by the Clear [f . t ) is given in Figure 5.1}.True. -Sin[x]"3.138 Partial Differential Equations Solution. -Sin[x]"3.h.{x. Plot Points.)'Time".6Pi} AxesLabel->"Position" .True. The matrix A= has eigenvalues A1 2 1 4 1 -3 A2 B I'].-4Pi. PlotRange-> {-4. = -1.O] u [x-.>40.7g [x+t]/2 +h [x-] t Plot3D[u[x.O] g[x-]:=Which[-Pi <=x<= 0.g.

Consider the problem with Show that the system is hyperbolic. Diagonalize it and solve the Cauchy problem with initial conditions .3 are made by MAPLE in Scient ifiCWorkPlacE. Graph of the function u = ul(x.1 and 5. All calculations in Examples 5.2. Exercises. but not strictly hyperbolic. 1.t ) .Shock Waves and Conservation Laws 139 Value ul Figure 5.

11) was studied at first in a physical context by Bateman (1915).t ) is a weak solution of the Cauchy problem t in the sense of Definition 5.11) The equation (5. 7E. Burgers. The partial differential equation ut uux = p u x x . ) satisfies the identity t uo (x . (R) be any locally integrable function.11) could be found by using the transformation 6J. 3(1950).Hopf. 'J.140 Partial Differential Equations where f i (x) f2 (z) f3 (x) are given in Example 3. Q. Around 1950. v 0) for any v E C (R x [O. + . 45 (1948).D. A mathematical model illustrating the theory of turbulence. Pure Appl. (b) Prove that if the function u = u (x.4. Prove that the function u (x) ) = uo (Z . Math 9 (1951). u (x. On a quasilinear parabolic equation occuring in aerodynamics. Hopf7 and independently Cole'. Mech.. ) E L1 (Rx [0.4). Plot the graph of the . Subsequently. t) ) 2. function u2 (x. Comm.4) then it is a weak solution of (5. 171-199. = EU. Cole. Adv.) d x d t + -00 7 uo (z) (z. (a) Let uo (x)E Lie. Burgers (1948) rederived it as a model equation in the theory of turbulence. 0 0 ) ) is a strong t solution of (5.. 5. dx = 0. Appl. Math. showed that the exact solution of (5. Appl.M.3 Burgers' Equation The simplest equation combining both nonlinear propagation and diffusion effects is the Burgers" equation Ut + uu. i. 225-236. (5. 201-230.e. ..t ) (vt 0 -00 + v.oo)) h .

Shock Waves and Conservation Laws 141 u = q X .13).11) as a conservation law Ut - (EU.11) with initial condit ion u(x.15) Now introducing it is easy to show that (5.12) This. Let us rewrite (5.. (5.16) has the unique solution .16) By the Poisson formula the problem (5.O) = 2Lo ( 3 2 ) .12) is as follows.11) to the diffusion equation 'Pt = &'Pxx* (5. known now as Hopf-Cole transformation. $t (5.p. From (5.14) Then Q X t = $tx implies (5.13).15) is equivalent to the diffusion equation (5. reduces (5. Let us consider the Cauchy problem for equation (5.14) it follows (5. (5. 1 = EUX .42. Under the transformation (5.1u2) 1 5 =0 and try to find $ E C2 such that $x = u.)ds 1 2E q(x. .13) The motivation of (5.O)=e 0 (5.(.12) the initial condition reduces to --S. $ = -2~lncp.11).

The exact solution of (5.17) as E -+ 0.17) -00 We consider the behavior of the solution (5. Let us recall an asymptotic formula derived by the so called steepest descent method (5.t .uo ( x ) ) is fixed.12.11) is (5.09.1731. while ( x .-( x - 4&t r )2 -00 -00 where t + Juo 0 ( s ) ds. g“ ( s ) > 0. The estimate is motivated by Taylor’sg formula and the Poisson integral -00 -00 Brook Taylor. 18.18) where s is a strong local minimum point of g ( x ) : g/ ( s ) = 0.142 Partial Differential Equations 0 0 - 2& 1 ] u 0 ( s ) d s . .1685-29.

Shock Waves and Conservation Laws 143 In order to apply the asymptotic formula (5.20) is smooth for small t if uo (x) is a smooth function.U t ) . The last function is exactly the solution of the problem (5.Chapter I.[) with respect to [. ) N -.19) jFrom (5. Suppose that there is only one strong local minimum E (2. t ) which satisfies (5. Differentiating (5.20) with respect to x we have ) or .18). The solution (5.18) to (5.17))in view of (5.17) we need to study t critical points of the function g (x) . it follows x-E u (x.21) found by the method of characteristics .uo t t ' The asymptotic solution may be rewritten as or in implicit form u = uo (x .20) (5.

) = . instant TOwhen u = 00. (so) ’ ub’ (so) = 0) Ub” (so) > 0. The first .t>O.1 b.max 1 &cosh2 1 E min cosh2 1 because cosh2 . Solution.tanh ( ) : 1 &cosh2 u’o (x)= - z which has a minimum at x = 0 minub(x) = .4. Example 5. The solution of the problem in implicit form is u (x. Find the instant of gradient catastrophe for the problem U~+UU.= & .144 if 1 u‘o (<) t Partial Differential Equations + # 0. = 0. 00 Suppose ub (x) < 0 for every x. O ) = -tanh(:). u ( x . Then u. x€R. corresponds to a so where ub (x)has a minimum To = -. = if t = -ub:t).tanh t For the function uo (x)= . known as gradient catastrophe.. ~ E R .

O] .tanh 5. PlotRange. i.3.1.01} .01.l] {s.tanh (:)) H(x)= = 1 .-5. The gradient catastrophe is demonstrated in Figure 5.3 .1.> {.> { . 5. where H (2)is the Heaviside function { ::x < 0. .3. PlotRange.PlotLabel. E The graphs of the functions . x>o.01 .tanh (E) for E = 1.u [s]] .-5.Shock Waves and Conservation Laws Then To = E -+ 0 145 as E --+ 0. 1 .>” t =1”] h2=Parametr icPlot [Evaluate[Evaluate [x[s. l .> {. Graphs of the functions .5}.5}.2H ) . Note that E-+O lim (.{ s .01 . i.PlotLab el.4 by the Muthemutica program u[s-I:=-Tanh[s] x[s-. a.5}.{ s .2].01} .t-] :=s+tu[s] hO=Par ametricPlot [Evaluate[Evaluate [x[s .u[s]3.l.-5.>” t =2”] h3=Par ametr icPlot [Evaluate[Evaluate[x [s. ( z ) for = 1.>” t =O” ] hl=ParametricPlot [Evaluate[Evaluate[x[s.1. {s . $.PlotLab el. PlotRange. l .u[s]].5}. ( . $ are given in Figure Figure 5.O 1} .-5.u[s]] .

Gradient catastrophe.146 Partial Differential Equations Plot Range. 1} .{ h2. In Figure 5. FrameTics->None] I 4 .PlotLabel. Frame->True.>’’t =3”] 0 Show[GraphicsArr ay [{ { hO..1.4.ttanhs.5 it is given the surface in R3 s: { x = s .h3}}].h 1} . t = t.01. .2 1 2 4 1 I L- ”h_ 1 .l. u = -tanhs.> {.1 2 1 3 3 Figure 5.

17) as E -+ 0. The surface S (x (s.a and Figure 5. For a given uo (x) and fixed (x. Let us return to the asymptotic behavior of solution (5. t ) .Shock Waves and Conservation Laws 147 Figure 5.b while (5.22) is demonstrated in Figure 5.7.23) in Figure 5.22) or For the function uo (x) = eVz2 the situation (5. ) it is possible to have three solutions of equation t (5.5.6.a and Figure 5. such that and J2 are points of local minima of uo (x). while is a local maximum.t ) .19) J1 < ( 0 < J2.u (s. .b. t ) . It is possible to have (5.6.7.

Graph of .6. Graphs of functions y = e-”2and y = -4 -2 0 2 x 4 Figure 5.148 Partial Differential Equations Figure 5.b.6 a.

b. Graphs of functions y = e-”’and y = 32.7.7 a.x -4 -2 0 2 x 4 Figure 5..Shock Waves and Conservation Laws 149 Figure 5. 5 . Graph of .

51 .3. = = EU.51 (5. The inequality g ( x .24) it follows uo (() and 9 uo (51) + uo (52) -LJ uo (s)ds.&] have equal areas. For a fixed x the changeover from (1 to (2 occurs at an instant 7 such that g ( x . t . & ) or its opposite determines the behavior of u ( x )t ) as & + 0 at a given ( x . Both 51 and & depend on ( x ) t ) .22)) by virtue of (5. ~ ) ) g ( x . Similarly in the case (5.uo (El) t 7 as E + 0. From (5.24) The last equation means that the regions in (E. &( x . t ) be a solution of the problem Uti-UU. 2 (2 t 2 .. u(z.t > 0. T . ~ ) ) ( = which implies (5.23) as E + 0. . ( x ..18))we have N 4 x-.u) plane between the graphs of for ( E [ r l . ( ~ x .501 and 5 E [<o. J 1 ) < g ( x .25) El We summarize these observations in Theorem 5. Let u. T . UO(Z)) x E R.O) x E R. t .t ) .150 Partial Differentid Equations In the case (5.

1882. x E R. Exercises 1. u ) y. . u ' ) such that y'. Show that any solution of the problem U~+UU. = = u ~ .eu dxdy (5. u(x. 24. n = 2.25) is satisfied.x E R.t > 0 .09. u = etuO (x . U= u o (x .In (1 . there exist and J1 < 6 2 such that and (5. Consider Liouville 'slo equation -- d2U .03.u + ue-t) .tuo (x . Liouville.1809-08.Shock Waves and Conservation Laws 151 T Suppose that for a given x. T h e n as E --+ 0.O) satisfies the functional relation UO(Z). n = 1.t u ) ) ' 2.tu)) 1 .26) Show that : (a) Making the change of variables (x. loJosef - (x'.In (1 .

the equation (5. t) it . Making a change of variables w (x.26) is where f and g are functions. t > 0.26) reduces to is u(x) = 21n e-? e 3 7q(s)ds "0 2 + a Jp(t)e 20 3 s q(s)ds "0 t dt (c) The general solution of the equation (5. xo and yo are constants. = u (x t) reduces to Cauchy problem for the diffusion equation + at. (a) Consider the Cauchy problem for the advection-diffusion equation ut +au. 3.. u(x. = EU. (5.O) x E R .27) with uo(x) E L1 (R) . x E R.152 Partial Differential Equations I where (b) The solution of the problem 2 ' =x Y =Y ' p is a constant. = UO(X)..

.2.O) which is a limit case of Burgers' equation as E -+ 0. (b) Show that the solution u.(x. Let p ( x . If uo (x)is nonsmooth we introduce. ) E C" t (Rx (0. 00)) . .at) as E .Shock Waves and Conservation Laws 153 x E R .(R).O) = U O ( Z ) . + Proof. Riemann Problem X E R > 0~ . o o ) ) is a weak solution of (5.. Let u E C1(R x [0. = 0 and a weak solution of the problem (5.4 Weak Solutions. If uo (x)is continuous at a point xo . x E R.) Lio. Definition 5. (5. Assume u~(a.6.28)..27) is and u ( ~ . 5. As in Proposition 5.28) u(z. t )uo (X .1 we are led to . 0 0 ) ) of (5. Consider the Cauchy problem for the quasilinear equation U~+UU.t ) E Ci (R x [0.2. x E R. = U O ( Z ) .0 0 ) ) be a smooth solution of the equation ut uu. as in Section 5. Vt = EV. t > 0. = 0. 0 0 ) ) .28) iff for every test function p E C A (Rx [0. N -+0. a notion of weak solution. v(x. then u (x0.O)= uo (XO) .(R x [ O . We have Proposition 5. A function E L.2.

oo))such that t Then (u 0) . Then The problem (5. 0) Suppose u (xo. 1 Case I.07. which is a contradiction. 17.30) where u and ur are constants. while if u < U. dx (x. such that u ( x . O ) > uo (x) ) x E u.09. Let us consider the initial data (5. 1 The two cases u > U r and u < Ur are quite different with respect to the 1 1 solvability of problem (5. then the weak 1 solution is unique.28).O) < uo(SO)is impossible.1826-20. It can be proved that if u > up . . > uo (20) By continuity there exists a neighborhood U 0) .28) with discontinuous initial data is known as a Riemann" problem. . P 0) > 0.. then there exist infinitely many solutions.uo (x))p (x. Similarly u (x0. u > ur 1 Consider the problem llGeorg Friedrich Bernhard Riemann.uo (4) (x. dx = 0. ) E Ci (R x [O. Take p (x.154 Partial Differential Equations 7 -00 (u 0) (x.1866.

30).3 r=-.t > 0 . The function + u ( x .28) is known as a shock wave.O) Let x x E R. the speed at which the discontinuity of the solution travels. = = EU.3 is determined by the slope k of the straight line through the points ( q 0 ) and (0. 2 > st. If u > U.. we are in a situation to apply Theorem 5.28) with initial data (5.3. while s = ( 1 u T ) is a shock speed. (5.s) Then and by Theorem 5. S X u /2 The unique solution of (5. where Proof. 1 > 0 be fixed and The instant r of Theorem 5. UO(X). x E R. Denote for simplicity t .. u(x. t )= { UI UT x < st.1. o o ) ) . Proposition 5.31) is a weak solution of the problem (5. Let p (x) ) E Ci (Rx [ O .Shock Waves and Conservation Laws 155 ut+uu.

dz) + 2 J (7 -00 &dz) dt.-p(St)t)S -00 and it follows = - 1 p(x.t)d. 0) p (z.O)dx.. -00 . --oo A1 : = 1/ 00 st (ptui +pz$) dt dxdt 0 -00 - ui/6 (7 -00 . 0) dz. By st J p t ( z .156 Partial Different ial Eq uations B := We have /m u (z. t ) d z = dt d --oo jP(..

Shock Waves and Conservation Laws 157 Then 11 00 st pt(x. )d2 t it + sp ( s t .t ) . 0) da: +s f p ( s t .t)dt. 0 0 -00 -00 A1 = -u1 Similarly (] -m p(a:. t )d t ) + $T 0 p ( s t .O)dz-s~p(st.t)dzdt= - / 0 00 p(z. t )d t . because dt st p (2. 00 P Then .

O)dx-/u(z. 0) dz = 0 / -00 0 p (z. u < u. 1 In this case there exist more than one weak solutions. 0 -00 / 00 u (x. Proof. 0) dz. 4 Case 11. 0) d z .4. For simplicity we take ur = 1 and denote UI = -1 and c := D := - /1 0000 (upt + $Iz) dxdt. Let p (z. One is (5.O)p(z.30).T p (z.m)) . The function x < Ult is a weak solution of the problem (5.31).158 On the other hand Partial DifferentialEquations = - / 0 00 u(z. ) p (z.O)dr 0 -00 0 00 -00 0 Finally and since we obtain A = B.28) with initial data (5.O)p(z. We show Proposition 5. 0 -00 .t ) E Ci (Rx [0.

1 0 O t Because C2 has a singularity at 0 C2 = lim C Z .Pz)dxdt= p p ( x .Px) 00 00 d x d t = -/p(x. where (71 :=// 00 -t (-Pt+. 0 -t and c := 3 11 0000 (Pt + .Pt. o ) d x . x x . lsa -00 0 --oo 0 c 2 := 11 0 0 t (:pt 1 + 5 (x7 )2 p.. ~ E-0 where We have (-P> X t t = -pP'.t)dt.> dxdt.Shock Waves and Conservation Laws The function x / t for t 159 We have # 0 satisfies the equation ut + uu2 = 0.O)dx+ 0 +.2 p(-t.t)dt.

p (-t.E) . t ).p(-t.t))dt./ . 7 2 1 x 2 t t -t -t Then c .160 Partial Differential Equations .)2P2 f = (TP).E ) dx = 0. E it follows (-E. = 2& 7. Because the function p is bounded and lim x E -+0 as E -+ 0 &+O / & E p ( x .t))d t .Pt + (. 0 . By the mean value theorem where x .pX x . + ( 5 ( ) P) . & ) d z ( & -& 2 's & 00 (p(t. (1 & & ( F p ) dx) dt - f J(p(t. & -& Then C2 = Finally -1 2 1 00 ( p ( t .t)) dt -t = .t) & .t) -p(-t.

E [ lu. = EU. There exist infinitely many weak solutions of the problem (5. Exercises 1. u..a t ) . Determine the behavior of this solution as E -+0.t < 2. where w satisfies the equation EW' + 1 (9) aw (y) = s w 2 (y) + + c.a weak solution of the problem. Show that the equation ut uu. is uz-I- 2 . ~ a2 2E 2' satisfies the equation (5. has a traveling wave solution of the form u.30) in the case u < u. ( z .. t )= w (z .28) with initial data (5.~ ~ t a n h ~C S . (5.~ y . 2.] and s = .Shock Waves and Conservation Laws 161 0 0 -00 0 which completes the proof.32) Verify that the function w ( y ) = a . Show that every function 1 u.32). where u.

x E R.36) Along the characteristics u is a constant because By (5. The characteristics of (5. ) = uo (z) 0 . x E R.1) m = 1.t > 0.36) it follows that a characterictic through the point straight line (x0.35) in (x. then Ut +g (u) U. Rankine-Hugoniot Condition.162 Partial Differential Equations 5.33) For the sake of simplicity we assume that in (5. x E R.O)is the with slope Assume that there exist two points 21 < 2 2 such that .34) If u is a classical solution of (5.33). = 0. t ) ) dt (5. (5. (5. ) t plane are the curves c: { x t =xt( t ) ' = such that d x ( t ) = g (u (x( t ).5 Discontinuous Solutions of Conservation Laws. u (x. Let us consider the Cauchy problem for a general conservation law Ut + (f( u ) ) ~= 0.35) which is a quasilinear first order equation. Suppose f is a C1 function and let (5.t > 0.

the tangent and normal vectors to I? at ( x . A function is a weak solution of (5. otherwise it can't be defined for all t > 0. j = 1. t ) are ?(+(t). Hence the solution can not be continuous at P.00) if there exist a finite number of smooth curves rj c R x [0.k outside of which u is a C1 function and across rj it has a jump discontinuity. Let I' be the curve of discontinuities r: { x ~ ~ t ( t . which is impossible. and G ( 1 .. So. Assume that U O ( X ) E L.O)intersect at some point P.. 0) and (x2. . We show that not every discontinuity is admissible.33) i f u E Lt. where q ( t )= 1) Denote 2.oc(R) .7. 0 0 ) ) and . At this point u ( P ) = uo (XI) = uo (x2).. f o r every test function p E C (Rx [O. the limits of u on each side of I?.33).Shock Waves and Conservation Laws 163 Then the characteristics c1 and c2 through (XI. the existence of a classical solution depends on the intersection of the characteristics of equation (5. 00). We say that u is piecewise smooth in R x [0.35) and is independent of smoothness of the functions uo (x) and f (u) If the function . .33) which are piecewise smooth only... It can be shown that the solution u is smooth up to the instant m 1 The above considerations lead us to introduce a weak solution of the Cauchy problem (5. Assume that dx for some x.oo)) f (u) E Ltoc(Rx [0. A We consider now weak solutions of (5. g (UO (x)) is monotone increasing then the classical solution exists for t > 0.+ ( t ) ) . (Rx [O.) Assume that r is a smooth curve. 0 0 ) ) . Definition 5.

160(1870)' 277-288.2 u is a classical solution in domains where u is a C1 function.4. H. ( 2 ) u satisfies the j u m p condition along every discontinuity curve I? : x = y ( t ). Hugoniot. which does not intersect is other curves of discontinuity.oo) a small ball centered at P . By Green's identity l2 William John Macqorn Rankine 1820-1872. Suppose u is a piecewise C1 function. we have 0 --oo where B* are the two open components of B on each side of I?. Let u : R x [ O . 1851-1887. J. 1'Ecole Polytech. .164 Partial Differential Equations Theorem 5. Assume that I : x = y ( t )is a discontinuity curve.M. P E I ' ' and B c R x (0. As in the proof of Proposition 5. 58(1889). Pierre Henri Hugoniot. The jump condition is known as Rankine-Hugoniot'2 n condition. Let p E Ci ( B ). For the case of Burgers' equation f ( u )= -.33)iff (1) u is a classical solution in domains where u is a C' function.Rankine. W. which is a weak solution of (5. 1-125. T h e n u is a weak solution of (5. Trans. Sur la propagation du mouvement dans les corps et specialement dans les gaz parfaits. As u is a weak solution and suppp c B . Phil.4.33). 2 UL it reduces to If I? : x = xo + kt is a straight line the last equation means Proof of Theorem 5.J. On the thermodynamic theory of waves of finite longitudial disturbance. o o ) --+ R be a piecewise C1 function.

33). then it is a weak solution of (5. 1-x . . The characteristic through the point (x0.O) is co tuo (xo) . = 0. 20 > 1. O L x < 1.O)= { : 2 L 0.8. x>1. so that 20 +t zo+t(l-xo) 50 : x = xo + co:x= { xo 5 0 . Solution.6.(x. I Example 5. Consider the Cauchy problem Ut + uu. The picture of characteristics is given in Figure 5.Shock Waves and Conservation Laws 165 Since p E C ( B ) is arbitrary we obtain the jump relation A Conversely. Determine the time of existence of a continuous weak solution and find a discontinuous weak solution. O<zo<_l. it is easy to check that if u is a piecewise Cf function which satisfies (1) and (2).

l ) . By Theorem 5. a:< 1 / 2 ( t + 1 ) . 1.6. x > 1/2 (t 1) . Characteristics in Example 5. No pair of characteristics intersect for t < 1. ) = 1) x < l + k ( t . t u (x.t) = Characteristics intersect for t 2 1. x > l + k ( t . u(x.166 Partial Differential Equations X Figure 5. for t >_ 1 the function u (x.. .t<x<l. x 2 1.. A continuous weak solution for t < 1 is 1 x 5 t.1 ) . 0. + is a weak solution of the problem. In this case we are looking for a weak solution of the form { .1 2 2’ So.t) = { 0.2 and the Rankine-Hugoniot condition we should have k= u++u.8.. .

(a) Find the characteristics and the solution of the problem Ut + uux = 0 . . a (b) Show that: if a 2 0 the solution is differentiable for t 0. where k 2 2 is an integer. (a) Find the characteristics of the problem and show that their envelope is the curve x=1+in (2.Shock Waves and Conservation Laws Exercises 1. = 0 . ) plane.1 (b) Verify that the continuous solution exists for 2k and does not exist for t 2 k2-1. 2. Consider the Cauchy problem Ut 167 + uu. 1 if a < 0 the solution is continuous for 0 5 t < --. t k2 . (c) Plot the picture of characteristics and their envelope with Mathematicu in the case Ic = 2 .

Show that the characteristics of the problem are Discuss the question of breaking of solutions. u(x. = 0 . Consider the Cauchy problem Ut u ( x .0 ) = (a) Determine the characteristics and show that they have an envelope of two branches. . (b) Plot the picture of chracteristics and their envelope with Mathernatica.+au = 0. Consider the problem ut +uu. 0 5 x 5 7q2. {: cos2 x + uu. x 2T / 2 .168 Partial Differential Equations 3. (c) Find a weak solution. 4. x 50.O) = u o ( x ) .

03. V2u= u.1749-05. 169 . 2. A function u E C2 (R) which satisfies the Laplace equation is called a harmonic function. + uyy+ u.. Steady-state incompressible fluid flow.. It appears in physical phenomena such as 1.Chapter 6 The Laplace Equation 6.1) The Laplacel equation or potential equation is where Au is the Laplacian of the function u Au Au = = V2u = u. Maximum-minimum Principle Au = 0. + uyy in two dimensions. The inhomogeneous Laplace equation n u = f.03. in three dimensions. 3. (6. where f is a given function is known as the Poisson equation. The Laplace equation is very important in applications.1 Harmonic Functions. Pierre Simon Laplace.1827. 23.. Steady-state heat conduction in a homogeneous body with constant heat capacity and constant conductivity. Electrical potential of a stationary electrical field in a region without charge.

The Robin or third BVP is = V u . = n2 uZz uyy = 0 + 1 Show that u n ( x .) but n3 . The solution is unique. 3.1.y) = . The Cauchy problem for Laplace equation is ill-posed. A boundary value problem for the Laplace equation is well posed in the sense of Hadamard with respect to a class of boundary data if 1. o u + ~ = + onan. Small variations of the boundary data yield small variations on the corresponding solutions. is called the Dirichlet or first boundary value problem (BVP) for the Laplace equation. 2. cos nx U ( Z ) O ) = 0) UY(Z)O) -. Example 6.sinhny cosnx is a solution of the problem (CL. the name boundary value problem was attributed to only problems for which the PDE was of the elliptic type. Historically. A modification of Hadamard’s example follows. A solution of the problem exists. where n’ denotes the outward unit normal to 00 and derivative. The problem of finding a I function u E C2(s2)n C(a) such that Au=O u=’p inn. Let ‘p and I) be continuous functions on 82. ondn. The Neumann or second BVP is I Au=O i n n . where cr is a continuous function on 22.170 Partial Differential Equations The basic mathematical problem is to solve the Laplace or Poisson equation in a given domain possibly with a condition on its boundary dR = n \ R . Consider the problem in R x ( O . 6 is the normal (RL): { Au=O ins. o o ) . Today we use this term in a much wider sense.

This follows from the fact that if x is an irrational multiple of T . Then m-ax u =max u . In contrast to the Cauchy problem for Laplace equation. Suppose that u E C2(R)n C(a) is a harmonic function in a bounded domain R.lim 2n-m Then enY . then the set of points { (cos nz. Let X E (0. such that cosnkxo -+ X as k -+ 00.3) (6. (Maximum-minimumprinciple). For every y > 0. sinnz) : n E N} is a dense set in the unit circle S = { (z. 171 Solution. Consider the case N = 2.= . Let R c RN. ifN=2 or Theorem 6. y) : x2 + y2 = 1). Denote by P a point of a. It can be easily seen that Un(z. R r (6.lim n+oo n3 2 n+oo 1 = . nk for y > 0.y)= sinhny cosnx is a solution of (CLn). N = 2 or 3 be a bounded domain and r = dR be its boundary.1. R r mjn u =min u. we have sinh n y 1 lim . l).Then = + E >0 and consider the modified AV = AU+ €A(x2 y2) = 4~ > 0. the Dirichlet problem is well posed.2) is not true. Let function v(P) u(P) €IPl2. There exist x~ and nk -+ +00.4) Proof. which implies that (6. eny n3 k--tW 1 lim 7 sinh nky cos n ' k q = +00. This follows by the maximum-minimum principle for harmonic functions. + .e-nY 5 n3 -=+00.Laplace Equation is not fu@lled.

it should attain its maximum on OR with + . f E C(n) and cp E C(r). + . Then u E C ( 0 ) is a harmonic function and u = 0 on I?. being a continuous function. r n Then for P E n u ( P ) < v ( P ) 5 .(PI) =max v ( P ) =m-ax u ( P ) .m-ax (-u) (6. 1 1 + -up + P2 P -Uw. Let R be a bounded domain.1.3) follows. Prove that u(p)= c1 l n p c2 if u is rotationaly invariant. By the Maximumminimum principle it follows that u = 0 on R.(PI) where R is such that R + &IP1I25 m F u + &R2.5) it follows u(P) Lmax u5m-ax u.u2. (6. r n As P E is arbitrary by the last inequality (6. (a) Show that in polar coordinates x = p cos 0. y = p sin 0.(PI) = . tive test in Calculus. 0) depends only on p. the two dimensional Laplacian is A u ( x . y) is rotationaly invariant if u(p.Since E is an arbitrary by (6. Since v ( P ) has no interior maximum in 0. T h e n the Dirichlet problem has n o more than one solution.172 Partial Differenti d Eq uations vyy 5 0 at an interior maximum point by the second derivawhile Av = v. 1.. y ) = up. n R n Corollary 6.4) also follows.5) c BR(O).6) and u = u1 . (b) A harmonic function u(x.. Proof. Suppose u j ( P ) . Because -u is also a harmonic function and mjn u = . j = 1 . 2 are two solutions of (6. = Exercises.

f. y) using polar coordinates..p < 1 y) and show that M m = 1.2 Green's Identities Let u .u. 8. z ) is spherically symmetric if u(p. 2Karl Friedrich Gauss.02. + is harmonic in R2\ 0 .1861 .09.23. 2. . l ) . be a domain with smooth boundary 80.1)2 Find the maximum M and minimum m of u (x. We have the divergence theorem or the Gauss-Ostrogradskii2 formula JJJ div@dV = JJ @ . C2 if u is spherically symmetric.O) y).04.hx. y = psinosincp. 'p) depends only on p..O) . in the disk Bp(0. fz . 12.1777. h ) is a vector field. Recall the following notations of field theory gradu div? = Vu = (ux.z = p cos cp. Michail Vasilievich Ostrogradskii.u E Z . where (x.uy. + h. Prove that the function u (2. .12. ( x 2 (y . .x2 .g. E Bo.y2 + 6. rot@ = V x @ = ( h . the three dimensional Laplacian is expressed as (d) A harmonic function u(z. dSp a surface element and dsp an arc length element at P on dQ. Show that u ( p ) = (71.g .)) Au = div(Vu) = V2u = uzz iuyy-I.Lap1ace Equation (c) In spherical coordinates 173 x = p cos 8 sin cp. n' be the outward 0 unit normal vector to 80.1801. y. fidSp.20.1855. Plot the graphic of u (x. Denote d V = dzdydz. = V - = fx + g. where p ( f .). 30. y) = 1 .g.w E C2(o).9 (0.

we have JJJ A u d V = JJ E d s P .174 Partial Differential Equations =V If F u .Include R2 c R3 by (2. by (6. an (6. y. (6.y) --+ (a.v g ) d S p . v E C 2 ( D )n C ( D ) . we obtain Green’s second identity in R2 .10) n Subtracting (6. Let D c R2 be a bounded domain with smooth closed oriented boundary C. u .9) from (6. As u and v do not depend on z (UAV V A Udxdy. ) (UAV v A u ) d x d y d z = - JJJ K JJ D JJ’ ( u g . VvdV = JJ E an u---dSp.11) known as Green’s second identity. V u d V JJ v g d S p . n n an known as Green’s first identity. Changing the role of v and u we have JJJ U A v d V + JJJ n vU. n an known as Gauss formula.10) we obtain J J J ( u ~ v vAu)dV = n JJ ( u g . Consider the two dimensional case.11). + vAu = JJJ v A u d V + JJJ t 7 v .O) and consider the cylinder K c R3 with base D and altitude 1.7). By the product rule it follows div(vVu) = V v V u and by (6.v g ) d S p = dK ( u g - v2) dsp. Then.

Theorem 6. Let u ( P ) be a harmonic function on B . = By (6.-/.13) For the sphere S the unit normal vector at P E S is Let us make the change of variables x = xo + pcosOsincp. Mean value property.PO( a } .Pol 5 a } ) S = S. S (6.. we have x s z . + sin 0 sin 'puy -/. + pcoscp.xo uz + .ZO . Proof.(Po) = ( P E R3 : IP .Z Y -Yo uy U a = cos 9 sin pu. The average value of any harmonic function over any sphere as equal to its value at the center.2. y Then for = yo z = 20 + psindsinp.9) it follows o = JJJAUdv B = JJ E d s p .cos 'puZ a a . where B = B.Laplace Equation 175 (6. 1.(Po) = { P E R3 : IP.12) Consider some applications of Green's identities.

13) becomes Partial Differential Equations and as a >0 > 0. we get = = 1 ' u( PO) cpdcpd8 sin 4TU(Po). J(12s iT -+ U ( T . Then it follows or . cp) sin cpdCpd8 Letting r 0 .176 Therefore (6. so that we can consider a as a variable The last identity is valid for every a r and we have Then I ( T )= is independent of T. 8.

We 1 shall prove that 0 is an open set.(Pl). Po(z0. as R is connected. Let 0 = N n Sr(P1). which completes the proof. Therefore u ( P ) = M if P E S. PO E R and 0 = { P E R : u ( P ) = M } . Let u ( P ) be a harmonic function in the domain R and u be bounded f r o m above.y) is a harmonic function in R2. By the continuity of u there is a 2 neighborhood N of P2 such that u ( P ) < M if P E N . By the mean value property it follows the maximum-minimum principle as well as uniqueness for solutions of Dirichlet problem in domains of R3. n ! . Theorem 6.Laplace Equation 177 Note that the mean value property is also valid in the two dimensional case. As M =sup u. E R2 and yo) is a disk.3. ~ ) . We shall 1 prove that Br(P1)c 0 . 1 1 we have u _< M on the boundary Sr(P1) { P : IP = = r } . 1 1 Let PI E 0 and & ( P I ) c R. T h e n u attains supu a C aff u is a constant. Finally u ( P ) = M in & ( P I ) and this means that 0 is an open set. Namely. By the same way u ( P ) = M if P E Sp(P1)for every p E ( 0 . which means that 0 is an open set. As u 1 R is a continuous function 0 is relatively closed and 0 2 = R\O1 is open. 1 . Then. if u ( z . As 2 is a connected set it can not be represented as a union of two ! nonempty open subsets 0 and 0 2 whose intersection is empty. Suppose there is a point P E Sr(P1)such that u(P2) < M . Ca = aKa then which is the mean value formula for the two dimensional equation. 1 Let M =sup u = PO). we have 0 2 = 8 1 because 0 # 8 and R = 0 which means that u is a constant in R. where & ( P I ) = { P : (P-P1(< r } .By the mean value property 52 which is a contradiction. Proof.

Then m-axu R = ail = max u.14). + .2. Let R be a bounded domain with smooth boundary and u E C2(R) n C(n> be an harmonic function. f ( P )E C(s2). = where p(P) E C(r). the lowest energy (6. r mjnu R = minu.w. Let R be a bounded domain. Let R c R3 be a domain with boundary r = dR. By the Green's first identity E(w) = E(u . Then the Dirichlet problem has no more than one solution u E C2(R) n C(a).178 Part id Differential Eq uations As a direct consequence we have Corollary 6. p(P) E C(r).2VuVw + IVwI2)d V = E(u) E ( w ) 2 E ( u ) . such that u(P) = p ( P ) on I?.we have E(42E(4 We can represent w = u . We prove that if u is the unique harmonic function. Proof. r Corollary 6. where w ( P ) = 0 on I?. Dirichlet principle Theorem 6.3. Among a11 functions w ( P )E C2(R) nC( that satisfy the Dirichlet boundary condition a) w ( P ) p(P) on r. 2.w) = //i (IVu12 .14) is attained b y a harmonic function satisfying (6. then for every w E C2(R) n C(0) with v(P) p(P) on = r.4.

Denote for simplicity On S. we have (6. Fix P E R and let E be sufficiently small such that B.15) It follows by the Green’s second identity that (6. Representation formula 179 Theorem 6.which is harmonic I& . Let us apply the Green’s second identity 1 in O\B. Then for every P E a.15) .16) By (6.(P) for the functions u(&) and v ( Q ) = . H 3.PI ’ for Q # P.(P) c R. zf N = 3.5. Let u E C2(n) n C1(Q) be such that Au E L1(R). and af N = 2 Proof.Laplace Equation which completes the proof. Consider the three dimensional case.

In the case of an harmonic function u we get the following conclusions. (-)QI (u$ +47ru(P). --) IQ 1 du . by (6. letting E + 0 . we obtain as E 0 1 .PI - or JJL v m Au(Q)dVQ = JL. Let u E C2(R) n C ' ( 0 ) be a harmonic function in R. Then.16). As u E C ' ( o ) .PI dSQ Motivated by the representation formula we set 1 if N = 3.(u) denotes the mean value of u over S. The function F ( Q .. and ISe[the area of S. by we have 1. .180 Partial Differential Equations where M. for every P E R . = 47ru(P). P ) is called a fundamental solution of the Laplacian with pole at P.4. -+ &+O lim A. Corollary 6.

Let u E C2(R) be harmonic in R. W (a) Exercises 1. R is a domain with smooth boundary I?.Laplace Equation 181 Corollary 6. If u E C2(0) C1 n the conclusion follows by (6. iff u is a constant. Prove the vector form of the Green's second identity n as2 where G(P) and G(P) are smooth vector-valued functions.18). n' is the outward normal vector to I? ( ii x v' means the vector product of vectors u' and 5.18) to any subdomain R' c 52 with smooth boundary. Among all functions w E C4($2) C3(n) satisfying the boundary conditions n the where p(P) and $ ( P ) E C(B$2). If u E C2($2) we can apply (6. then u attains supu in s2. (a) Prove the Green's first identity for the biharmonic operator A2 where u .) 3. Proof.5. v E C4(R) nC3(n). A function u E C(R) is called subharmonic if for every P exists a ball B. if u is subharmonic and bounded from above. there Prove that.(P) c R such that for every p < r E R. Then u E P(Q) and every partial derivative of u is a harmonic function in Q . lowest energy . (b) Prove Dirichlet principle for biharmonic functions. 2.

19). then the u(Q)dS+ u(P0) a~ E 4 0. which satisfies (6. dU . Consider the problem of finding a function Q ( Q . i.e.(PO)= 4rE2 (b) Show that the last statement is not true if u is a discontinuous function. a function satisfying A2u = 0 and (6. (a) Prove that if u E 1 mean value IM.P ) E C2(R)n C1(Q)such that where P E R is fixed and . (6. 6. and POE R is an interior point. Among all functions w E C2(R) n C1(n) satisfying the boundary condition d --. (a) Show that if u is a solution of the Neumann problem Au = f in R. 5 .(P) dn = +(P). 4.20). .20) is attained by a harmonic function u . JLe . dn then (b) Prove Dirichlet principle for the Neumann boundary condition. P E dR.182 Partial Differential Equations is attained by a biharmonic function u .= +ondR.3 Green’s Functions Now we use Green’s indentities to study the Dirichlet problem.

. A main property of the Green's function is its symmetry. The functions u ( P ) = G(P. P E an.. P ) = F ( Q . where cp(P) E C(d52).22) and using (6. .Q ( Q . Applying the Green's second identity to u ( P ) and v ( P ) in 52.Pl) and u ( P ) = G(P.25) is an integral representation of any solution of the Dirichlet problem (6.21). Formula (6.23).. By the Green's second identity we have Consider the Dirichlet problem of finding a function u ( P ) E C2(R) nC1(fi) such that Au(P) = 0.21) has a solution and u E C2(52)n C1(a) is a harmonic function.e.27) because u ( P ) = v ( P ) = 0 on dR. 2 and & ( P I ) n B. i = 1 . Lemma 6 1 The Green's function for the Laplacian in R is symmetric. u ( P ) = cp(P).(P2) = la.24) from (6.(P2)). P ) .Laplace Equation 183 Suppose that (6. Let E > 0 be small enough such that &(Pi) c R.25) where G ( Q . we obtain (6. i.23) Substracting (6. P ) is known as Green's function for the Laplacian in 52.By the representation formula (6.P2) are harmonic in RE = S2\(BE(P1)U B. f o r every PI and P E R 2 Proof. P E 52. we have (6.

we obtain which means that G(Pl. As Ap@(P. P ) = 0 for Q E dR.Pl).P ) = @(P.6.Q) = 0 by the definition it follows that Q).Q)= F ( Q .27) lim &+O // v&dS =lim an E-+O // udndS.184 Letting E +0 Partial Differential Equations we observe that Therefore by (6.P) dnQ Proof.(P) aF(P)P2) dS. d -G(Q. dn S (P1) E S (P2) E Calculating the limits in the last expression. as in (6.Q) and F(P. Moreover we have . P ) = G(P.P2) = G(P2.17). As G ( Q . P ) it follows that @(Q. dV Then lim E+O J/ aF(P1P1)dS =lim dn E-+O // . P E R. which implies that ApG(Q. The functions P I+ - G(Q.P) and P I----+ are harmonic in R for every Q E dR. Corollary 6.

P * ) . By symmetry of P and P* with respect to d D I& .PJ IQ . 6. z ) E D has a reflected point P*(x.23) is given by formula (6.4. We solve the problem (6.q. Suppose Q(<.28) reduces to solving the Dirichlet problem (6.G ( Q . = (6.y. that is the function and its derivatives are tending to 0 as [PI + 00. i.O)E d D = { z = 0).-z) 4 D . z > 0. Each point P ( x .6 the function u ( P ) is harmonic since Apu(P)= - JJ BR d ( ~ ( Q ) A ~ .4 Green’s finctions for a Half-space and Sphere 6.1 Half-space Let D C R3 be the half-space of points P ( x .y. u ( P ) = p(P) on 352. The Green’s function also allows us to solve Dirichlet problem for Poisson equation.25).21) for some regions 52 with simple geometry. is given by (6.30) As D is an infinite region.21). if the Green’s function for the Laplacian in 52 is determined. By Corollary 6.y.Laplace Equation 185 The solution of the Dirichlet problem (6. z ) .e. We assert that the Green’s function for D is . P ) ~ S= 0. Namely. all the properties on Green’s function are still valid if we impose the so-called boundary condition at infinity.28) Solving the Dirichlet problem (6.21) has a solution. Q dnQ It remains to show the boundary condition in the sense This can be shown if the problem (6. the solution of the problem A u ( P ) = f(P)in R.

2. As for Q E d o . y) E R2 . P ) = 0 in D . @ ( Q . <) and observe that Then. in view of (6. 4 4 -~ I P on dD.31) Denote P ( z .P* I it follows that We find an explicit formula for the Dirichlet problem for the Laplace equation in the half-space (6. p )= As the function 1 is harmonic in D . IQ .y. z ) E D .31) is (6.7. For all z > 0 and for all (z.P ) = 1 Partial Differential Equations is a solution of the problem 4 ~ l Q P*I 1 A.25). because P* 4 D it remains to 4 ~ l Q P*I I prove that the boundary conditions are satisfied. Q = (<.186 Indeed we have to show that @(Q.P = IQ .@(Q. the solution of the problem (6.32) Lemma 6.

\&I = R and .4.33) If Q E SR by (6.Laplace Equation 187 Proof. SR = ~ B R .IS1 IPI IQI ..y.--+ IPl2 which implies IPI P*I = R2. = 1. By Lemma 6.7. (6. Let P E BR and P* be the inverse point of P with respect to SR defined as R2 opj = -OP.33) it follows that the two triangles QOP* an( POQ are similar. z ) given b y (6.34) .Ip*(’ Then it follows (6. Making the change of variables have for z > 0 < = + pcos8.2 Sphere Let BR be the open ball of radius R centered at the origin 0.2 there follows a maximum principle for harmonic functions in a half-space. Ifcp(Q)E C(R2)nLw(R2). r] = y + psin8 we J.32) is harmonic and Iu(P)I 5 M . Icp(Q)l u ( x . 5 M then the function 6. Corollary 6. because they have a common angle <QOP = <QOP*.

Thus the Green's function for the sphere SR n R3 is R ) In order to solve the Dirichlet problem according to the formula (6.P*l 1 ary condition @(Q. The function (a(Q. P E BR.. By (6.PI' In the two dimensional case we' have if P = 0. 1 ifP=O. T h e r e h e for Q E SR.P) = Q E SR is satisfied. P).PI and I& . which for the sphere SR is Applying the cosine theorem to the triangles O Q P and OQP* we have where y = uQOP. 4 r ) Q . y fixed .P) is harmonic with respect to P in BR\{P} because 1 1 are harmonic.188 Partial Differential Equations if Q E SR. For each P E BR\{O} the solution of (6.25) let us compute the normal derivative &G(Q.21) is given by I.34) it follows that the boundI& .

35) in (6.35) Substituting (6. Then (6.37) and (6.38) are known as Poisson formulae.36) u ( P ) = q ( P ) on SR.38) Note that (6. Let u E C 2 ( B ~ )C'(&) be a solution of the Dirichlet f~ problem Au = 0 in B R .8. In the two dimensional case we have respectively (6.25) we obtain Theorem 6.37) Setting u ( P ) = 1 in we obtain Corollary 6.40) .39) (6. For every P E BR (6.6. (6.Laplace Equation 189 (6.

O) = u(pcosO. For every cp(P) E ~ ( S R the function ).psinO). p sin 8).37) and (6.190 Introducing polar coordinates for Partial Differential Equations P ( p cos 8.41) is the unique solution of the Dirichlet problem. To prove the boundary condition in the sense let us fix a QO E SR and E > 0.42) .T ) + p2 dr -- 27r R2 .pi3 SR (6. As cp E C ( S R )there exists S > 0 such that Moreover let M > 0 be such that Icp(P)I I M .40) we obtain 1 27r R2 .\PI2 /J c p ( Q ) d S ~ 47r R IQ . cp(r)= cp(Rcosr. u(p.38) we have (6.6.41) is a harmonic function in BR.37) gives the unique solution of the Dirichlet problem for the sphere. Proof. The formula (6. By Corollary 6. u given by (6.2Rpcos(8 . Q(R cos r.P2 ’ known also as Poisson formulae.39)) (6.R sin r ) . which is the statement of the following Theorem 6. u ( P ) = R2 . P E SR.R s i n r ) by (6. By (6.7.

(a) Find the Green’s function for the half-plane G = { (x.61)2 2A4 47rR ( . Taking 6 E (0. > 0.6 1 ) 3 IS2I 6 The last estimate implies P+Qo lim J ( P ) = O . : y y) > 0) c R2.(P(Qo)I< E * By the maximum-minimum principle it follows that u ( P ) is the unique solution of the Dirichlet problem. . ~ = 0.42) we obtain that for JP.Laplace Equation where 191 and < - R2 . by (6.s) such that J ( P ) < 1 61. ) u ( z . Exercises 1. x E R. (b) If u E C2(G) C1(@ is a solution of the Dirichlet problem n A u ( z .QoI < b(P).P E BR it follows It means lim u ( P ) = cp(Q0).0) = C p ( 4 z E R. P+Qo 5 .( R .

192 then Partial Differential Equations (6.p i 3 (6. (b) if u E C2(R)n C'(n> is a solution of the Robin problem + au = C ~ ( P ) .43) (c) Prove that and if cp(x) E C(R) n Lm(R) then the function (6. P E r. IPI-fm 3. = E = E r. (a) Find the Green's function for the exterior sphere B& = { P E R3 : > R) (b) If u E C2(B&)n C 1 ( E ) is a solution of the Dirichlet problem A u ( P ) = 0.44) is harmonic. P E 0. ) . &J'(Q. E R. Q P )R. P I 2. Let R be a bounded domain with smooth boundary F = dR. d = (P(z0). Pl # P2. P ) P where ) { A. The function R(Q. a # 0 is a constant. P E B&. .P ) .P dn Prove that: (a) R(P1. Pp)+0.R 2 P2 47rR /J ~ Q P S Qp B k .43) satisfies lim (W)--+(~O>O) u ( z . SR IQ. U ( P ) = Cp(P). is called Robin function. P 2 E R. AU = f(P). P ) + aJ'(Q. P2) = R(P2. a+(Q. lim u(P) = (p(&). P ) = F(Q.P E SR. if Q E SR P+Q and lim u(P) = O .44) (c) Prove that if p(P) E C ( S R )the function (6. Q &$(Q.+(&. then u ( P )= I 1 . Pl) if Pl.+(Q.

37) we derive. Let u be a nonnegative harmonic function in 0. Leipzig. Grundlagen des logaritmischen Potentiales. A. Using a translation of the argument if it is necessary we may assume PO= 0. Harnack. . 1851-1888. Theorem 6.8.Laplace Equation 193 6. 1887. We have by Poisson formula and the mean value property By the same way 3Alex Harnack. the so called 3 Harnack’s inequa2ities for harmonic functions. Then for every PO E 0.R).5 Harnack’s Inequalities and Theorems As an application of Poisson formula (6. &(PO) C s2 and every p E (O.45) Proof.(Po) (6.P E S.

Fix Po E R3 and p > 0. It is easy to see that the function u ( z . IP . By (6. Then w = 21 . By Liouville's theorem it must be a constant. By Corollary 6. given by (6. H As a consequence of Liouville's theorem we derive the unique continuation property for harmonic functions on a half-space. The function = is bounded and harmonic on the whole R3. As wJ.194 Partial Differential Equations Corollary 6. The existence part shows that (6.46) is the unique harmonic extension of cp to R2 x { z > 0).212 is harmonic on R2 x { z > 0} 1 and wIzz=o 0.46) Proof. y . A nonnegative harmonic function in R3 is a constant.45) letting R -+ +oo we obtain .9. then w = 0 on R3. (Liouville 's theorem).46)) is harmonic for z > 0.y) E C(R2)f l LO"(R2) has a unique bounded n harmonic extension a R2 x { z > 0 } . Theorem 6.7 it is a bounded function. which is a harmonic function By Theorem 6. ZL should be a constant in R3. Proof.7 is a harmonic function in BR. Every cp(x.(Po) = u ( P ) for every P. As POand p are arbitrary. given by (6. z ) .=o = 0. . Denote for simplicity by rI(Q)P ) the Poisson kernel in R3.Pol = p .9. Suppose 211 and v2 are two bounded harmonic functions on R2 x { z > 0) such that 211 lzZ=o = 212lZ=o = p.

. Let 0 be the set of points P of s2 where the sequence { u n ( P ) } 1 is convergent. (Harnack's first theorem) Let { u n ( P ) } be a sequence of harmonic functions in a domain 0. T h e n { u n } is uniformly Convergent o n every compact subset K c R and the limit function u i s a harmonic in 0. J P. Let B be an open ball. Proof. It is clear that the limit function u ( P ) is a continuous function. u ( P ) is harmonic in R. Let Q E 0 and 2R = 1 1 dist(Q.Laplace Equation 195 Theorem 6. uniformly convergent o n every compact K c R.. Let u l ( P ) 5 u2(P) 5 . which is convergent in a point Q E s2. (Harnack's second theorem). Let for n > N ( E ) m > 0 .QI = p 5 -f.be a monotone increasing sequence of harmonic functions in R. by it follows that u ( P ) is harmonic in B. E > 0 be arbitrary and N ( E )be such that E un+m(Q) .10. As B is arbitrary. Theorem 6. T h e n the limit function u ( P ) i s harmonic in R. 5 un(P) 5 . Proof. P ) is a harmonic function in B .. We show that the sequence {un(P))is convergent in the ball BR/3(Q). Passing to the limit in we obtain As II(Q.By the monotone property and Harnack's inequality we have and (6.. S = aB and P E B .11. aa).47) for every P. We shall show that 0 is an open set.u n ( Q ) < 3- .

As R is a connected set and 0 # 0 it implies that 0 2 = 0 and 0 = 01.dR). 2.47) we have Partial Differential Equations (6. if u ( P ) is a harmonic function in R3 and for every Q E R3 . IP .= dist(Q. we obtain that {un} is uniformly convergent on K . 3 (a) Prove that if u ( P ) is a harmonic function in R3. Let now 0 2 = R\01.Q < I Thus 0 is an open set.196 jF’rom (6. 5 Exercises 1 (a) Prove the Harnack’s inequalities in the two dimensional case. KR(Po)c R and every p E ( O . so { u n ( P ) }is convergent at every point of 0. By Harnack’s first theorem the limit function is a harmonic function. then dU -(P) dX = 3 - 4rR3 / l R ( p l u(Q)nxdS’ (b) Prove that.(Po) --(PO) R+P 5 u ( P ) 5 -+(PO). 4.48) it follows that {un} is uniformly convergent on closed balls. If there exists a point QO E 01. We prove that 0 2 is also 1 open. bounded from above (below). R > 0 . As we have noted.P E C. 1 Let Q E 0 2 be arbitrary and 47. Prove that Liouville’s theorem holds for harmonic functions in R3.Qol < it follows by previous observation that Q E 01. R ) . this implies that R = 0 .48) which means that { u n ( P ) }is convergent for every P. which shows that 0 2 is open. c R2. IQ . For every POE R. By 1 (6. If K c R is a compact subset covering K by a finite number of balls with appropriate radius. R.P where u ( P ) is a nonnegative harmonic function on R (b) Prove Liouville’s theorem in R2. Therefore Bg(Q) c 0 2 .

z ) = -cp(x. -+. Prove that if cp E C(BB) and p(x. 197 4 (a) Let B be the unit ball centered at the origin and u be the unique solution of Au = 0 in B . y. 9. . > 0} vanishing for z = 0. y. (x.Laplace Equation then u is a constant.z ) then u ( x . u ( a B = q. z ) = (b) Let u be a harmonic function in B+ = B n{ z Extend u to a harmonic function on B . .y.-2).

This page intentionally left blank .

Chapter 7 Fourier Series and Fourier Method for PDEs 7. Let us begin with the Fourier sine series.Z]and f(x) be expressed as f ( ~= ) nrx C bn sin 1 ' 00 n=l The problem is. x E [O. (7. 21. presented to the Academy of Sciences in Paris in 1807 and 1811. 'Jean Baptiste Joseph Fourier.03. Let f(x) be a piecewise continuous function.1768-16. Convergence of Fourier series In this chapter we consider Fourier series and the Fourier method in order to solve boundary value problems for linear PDEs in terms of series.05.1830. His results on the representation of functions by trigonometric series.2) (7. who had developed his ideas on trigonometric series studying heat conduction. were criticized (most strongly by Lagrange) for a lack of rigor and were not published until 1822.1.3) 2 n E N.1 Fourier Series 7. 199 . how to find the coefficients bn if f(x) is a given function? Observe that I' sin nrx mr 7 -dxx sin 1 1 =0 if m # n. This approach was used by Joseph Fourier' .1 Fourier coefficients.

m # n. integrating from 0 to 1 and using (7. we obtain that the coefficients am are expressed as A full Fourier series.4) 1 h Similarly suppose that f ( x ) is expanded in Fourier cosine series nrx Using orthogonality of the trigonometric system cos . or simply Fourier series.5) 1. ' nrx = 0.2) and I (7.where x (-1. n # m.">.1) term by term.l) is defined as E f (4 Observe that = -a0 1 2 + C (a.3). of a function f ( x ) . rnrx Multiplying (7. I cos nrx rnrx cos -dx 1 rnrx sin -dx 1 = 0.: n E N U (0)) 1 { cos -dx cos2 y 1 1 = 0. . cos -+ bn sin nrx 1 n=l 03 ". d x =2' n E N.1) by sin -. n # m. (7. Suppose that we can integrate (7.200 Partial Different ial Eq u ations known as orthogonality property of the trigonometric system {sin-n r x : 1 EN}. we obtain 1 rnrx bm = f ( x )sin -dx. mEN (7.

x 1 m E N U (0) (7.10) is pointwise convergent to a function u ( x ) on I iff for every 20 E I the series n=l N jU(x0) . n E N. then using orthogonality we obtain the coefficients a.. u n ( x ) : I a function.x u n ( x 0 ) < n=l I n=l I . 'dn.. known as orthogonality of the trigonometric system 7rx r { 1.. (7.cos d . The series -+ R. be (7.cos -. Let I = [a. nrx 1 .9) Recall some facts for series of functions.} If we can integrate (7.Fourier Series and Fourier Method for PDEs 201 ' and nrx sin -dx 1 mrx = 0.5) term by term. ' f ( x )cos -dx.1x .b] be a closed and bounded interval. and bm as am = and 1 mr [.'dmE N U (0) ' sin2 -dxx nr 1 = 1..7) dx = 21. s i n -..sin nrx 1 -.

n E N such that Iun(z)I I cn.11) is uniformly convergent on R. differentiability and integrability of the sum are as follows: . Some basic statements for uniformly convergent series with respect to continuity. vx E I and the series n=l be convergent.11) 00 C."=l U n ( x ) is uniformly For instance. then the series 00 Cn=l ( x ) v ( x )is uniformly convergent.1.1. If v ( x ) is a bounded function on I .202 Partial Differential Equations for every x E I .10) converges in the mean-square (or in L2) sense to u ( x ) on I iff for every E > 0 there exists NO = No(€) such that if N > NO Note that uniform convergence is stronger than both pointwise and meansquare convergence. for every x E I and un(x) is uniformly convergent on I .10) is absolutely convergent if the series CrZ1 Iun(x)l is convergent. vn. A criterion for uniform convergence is c. (7."=l Theorem 7. Remark 7. The series (7. 21. (Weierstrass criterion) Let there exist constants cn. if the coefficients an and bn of the trigonometric series 5 + C ( a n cosnx + bn s i n w ) 2 are such that the series n=l 00 n=l is convergent. then (7. The series (7. T h e n the series c convergent o n I .

"=."=l n ( x ) . the orthogonality properties (7.x E I be a continuous function and C. (7.x E I be an integrable function and C.6) and (7.4.8). u be uniformly convergent on I . By Remark 7.13) the Fourier coefficients (7. and f ( x ) is absolutely integrable on I1 By (7.9). .."=l n ( x ) is a continuu ous function on I .I. i. n ( x ) is u pointwise convergent and u ( x ) is its sum."=l() .e.Fourier Series and Fourier Method for PDEs 203 Theorem 7.1. be uniformly convergent on I . Let u n ( x ) . Suppose f ( x ) : R -+ R is a periodic function with period 21. known as Fourier coeficients of the function f ( x ) . If C.12) (7.9) are well-defined. Then the sum u ( x ) = C.1] and suppose the series (7.2.7) we obtain the coefficients formulae (7. then u ( x ) is differentiable on I and u ' ( z )= Cuh(z).8) and (7..x E I be a continuously differentiable function and the series C.3."=l un(x) be uniformly convergent on I . Theorem 7 3 Let u n ( x ) . Theorem 7. n=l 00 Denote Il := [-1.5) is uniformly convergent on I. Then Theorem 7. Let u n ( x ) .

j = 1.1 and f ( x ) has a jump discontinuity at each point x j .xj+1).O ) ) . The function f (2) is said to be piecewise continuous on I .6. such that f ( x ) is continuous on each open interval ( x j . The function f ( x ) has a j u m p discontinuity at a point xo E I if the one side limits f (xo + 0) = %o .8) and (7. Theorem 7. f (-1) = f (I) and f‘(x) be piecewise continuous o n Il. a 5 x1 < 2 2 < .204 Partial Differentia1 Eq u ations Let us associate to the function f its Fourier series The natural question is: Does f ( x ) coincide with the sum of its Fourier series and what kind of convergence appears? There are answers to this question for some classes of functions f(x). T h e n the Fourier series o f f converges uniformly to f (x) o n 1 . .5.0) 2 1 + f (-1 + 0)).. < xn 5 b.6 is based on a result on the mean-square convergence of the Fourier series. where the coeficients an and bn are given by (7.0) = ). At x = z t l . 1 Theorem 7.. Theorem 7.. 2 if f has a jump discontinuity at XO.n..n . j = 1. f(zo .f(x0 .-go f(41 x<zo exist but are not equal. z>zo f ( 4 . ..9).... The value of the jump discontinuity is the number J f ( 2 0 ) := f ( x 0 + 0) . the series converges t o 1 -(f (1 . if there exist a finite number of points x j ..0).l) where f is continuous and is equal to -(f (xo + 0) + f (xo .. T h e n the s u m of the Fourier series s(x) is equal to f ( x ) at each point x E (-1. Let f (x) be a continuous function o n Ii. Suppose f (x)and f ‘ ( x ) are piecewise continuous functions on11 and 00 nrx f (x) :a0 (an cos 1 bn sin - +C n=l + is the Fourier series of f .

Fourier Series and Fourier Method for PDEs 205 Theorem 7. Let f be a 21-periodic piecewise continuous function o n Il. it is absolutely integrable and square integrable. = :/ 1 1 -1 nrx f(x)sin-dx 1 -1 (7. Denote by a/.16) b.8) and (7. Proof of Theorem 7.l ) . 17’55-1 833.7 ah2 I f’(z)I2dx = .9) by parts and using f ( 1 ) = f ( . the Fourier ‘ coefficients of f .+ C (a$’ + b:) 2 n=l 00 < 00 (7.17) - nn [ nxx f’(x) cos -dx I Using the elementary inequality -(la1 1 n 1 + lbl) L a2 + b2 + 2132’ Mark-Antoine Parseval des Chimes.6. . By Theorem 7.15) Integrating (7. As f’(x) is piecewise continuous on Il.14) n=l The equation (7.14) is known as ParseVal’s2 equality. Then the Fourier series f o r f converges to f (x) a the mean-square sense and n 00 (7. we obtain an = I-1 J 1 1 nnx f(x)cos-dx 1 (7.7. and b/.

.206 by (7.17). . an and bn be the Fourier coeficients. suppose that f (x) is diferentiable and f ‘ ( x ) is absolutely integrable on I.. Then.19) where If f ’ ( x ) is continuous and f “ ( x ) is absolutely integrable o n Il.15) we obtain that the series Partial Differential Equations n=l n is convergent. Then (7. By the Weierstrass criterion the Fourier series for f is uniformly convergent and the sum is a continuous function which coincides with f ( 2 ) on Il.16) and (7. absolutely integrable on .18) where M =- 1 I s_. Ii. Then (7. Using integration by parts we have the following estimates for Fourier coefficients. lf(x)ldx* Moreover. it follows that the series n=l is convergent. then (7.20) where . Lemma 7 1 Let f ( x ) be a 21-periodic function. by (7.

1 Then lf'(x)ldx = -.-t nrx 1 nr I'.19) we integrate by parts an = n -S_lf(x)cos-dx r x l 1 1 1 1 -f(x)sin nr (7. we have l bn = 7 and 1 nr J_.21) and (7. The complex form of the full Fourier series A function f defined on R or on an interval I1 is said to be even if for every x .21) = - l -G ll 1 . which follows by continuity.Fourier Series and Fourier Method for PDEs 207 Proof.22). n Ml Similarly.13).20) we integrate by parts in (7. in view of f ( 1 ) = f ( . It is easy to see that (7.2 Even and odd functions. By f'(1) = f'(-1) we have Then it follows 7.22) In order to obtain (7.18) follows from (7. In order to obtain (7.1.l ) . 1 1 -1 nrx sin -cix 1 nrx f'(x)sin -dx. f ( ~sin -dxx ) 1 =- (7.12) and (7.

nrx 1 n E NU{O}. (2) The product of an even and an odd function is an odd function. l ) .208 The function f is called odd if for every x Partial Differential Equations f ( .23) If f( x ) is an odd function. l ) . Then for the Fourier coefficients we have an = I1 1 ~(X)COS -dx. bn = 0.I .I . 1 (7.1 .4= . I ) . 21periodic and absolutely integrable on ( . The even extension is not necessarily defined at 0. 21-periodic and absolutely integrable on ( . If f is an integrable odd function on I1 It is easy to see that: (I) The sum of two even (odd) functions is an even (odd) function.24) nEN. then a. x = 0.It can be extended to ( . Let f ( z ) be a function defined on the interval ( 0 . Suppose f ( x ) is an even function. The function f ( x ) can be extended to (-1. I ) as an even function by fe(x) = { f(x) f(-x) -1 0 <x <I.f ( x ) The graph of an even function is symmetric with respect to axis Oy. while the product of two odd (even) functions is an even function. If f is an integrable even function on I1 then If f is an odd function then f(0) = 0. Let us return to Fourier series. -I < x < 0. The graph of an odd function is symmetric with respect to origin 0. < x < 0. (7. = :i 1 €NU{O} nrx f(~)sin-dx. .I) as an odd function by f o ( x )= { f(x) -f(-x) O 0 < x < 1.

given a function f ( x ) on (0. nrx Z n E N. As the given function is odd in (-r. Let us consider some examples of Fourier series.24).JuTcosnxdx) L .Fourier Series and Fourier Method for PDEs 209 Finally. Similarly. .nn..Z) we get the Fourier sine series of f ( x ) n=l where b. taking the 21-periodic extension of the odd extension fo(x) we obtain a Fourier series with coefficients (7. -+ R.Z) we obtain the Fourier cosine series of f ( x ) M n=l where nrx f (x) cos -dx.23). = 1 1 f (x) sin -dx.Z) it can be expanded both in Fourier cosine or Fourier sine series. Namely. let us consider the 21-periodic extension of the even extension fe(x) and calculate the Fourier series with coefficients (7. Restricting to the interval (0. we have r) bn = :JdTxsinnxdx=-- nIr JdTxdcosnx - . an=$1 1 1 n E NU{O}..(z c o s n x ~ . Example 7 1 Expand the 21r-periodic function f : R . in Fourier series. Solution. Restricting to the interval (0.

an Fourier series.2 c n=l (-l)n+l n sin nx. o an 2 JT~2cosnxdx. then 1 . Solution.7r) 0. Example 7. xE (-7r. Expand the 27r-periodic function f : R f(x) = x2. we have bn = = --+ R. n E N. As the function is even.5 we have O0 (-1)"+' n sinnx. If x = 7r . By Theorem 7.2.210 The Fourier series is O0 Partial Differential Equations x . - 4 Jd x xdcosnx x cos n x l i - LT cos nxdx) 4(-1)" n2 ' .(f(7r . -7r x = 2 c n=l < x < 7r.0) + f(-T + 0)) 2 1 $"-") = = sin n7r n=l 0. T n€N.

7T-X Olz<7r. f ( x )= { .X bn d cos n x cos n x l i + IT cos n x d z ) . As the function is odd in an (-T.. The Fourier series is The coefficients and the partial sums can be calculated by the Muthematica program as follows . R -+ R. we have = 0.3. Expand the 2~-periodic function f .-. in Fourier series.2+.1 n = 1. -7T -- 2 < x < 0.7r] n=l Taking x = 7~ and x = 0 we obtain the identities Example 7. .2.Fourier Series and Fourier Method for PDEs The Fourier series of the function is 211 n=l As the function is continuous in [-7r.. sin n x d x T . n . r)\{O} . Solution. = n € NU{O}.

5 1 0. 1211. + 1.5 8 Figure 7. 0. {x. n. 12}] g l = Plot[Evaluate[f[x]].x)/2] fI[x-] := (Pi . Pi}] g l l = Plot[Evaluate[f[x . Pi}] g22 = Plot[Evaluate[fs[x. {x. 121 := Sum[a[k] Sin[k x]. x. -Pi. $21 The graphs of f(x) and the partial sum S12(2) are plotted in Figure 7.2Pi]]. f. k. Graphs of Consider the partial sum 5’12 ( x ) and f (x) in Example 7. g l l .3. {x. We have the trigonometric identity . (Pi . Pi.1.x)/2 a[k-] := (2/Pi)Integrate[fl[x] Sin[k x].2Pi. {x. 1211. Pi}] fs[x-. fs] f[x-] := Which[x <= 0. {k. g2. True. Pi.1. -(x Pi)/2.212 Partial Differential Equations Clear[a.{x. 1. 3Pi)l Show[gl. -Pi. 3Pi}] g2 = Plot[Evaluate[fs[x.

04.26) It is easy to see that lim t-0 (- 1 2sin$ :) = 0. so that the function gN(t) = (- 1 Zsin $ i) sin (N + i) t is bounded in a neighborhood of 0. 11. For this and other contributions Gibbs has been honored by a prize with his name by the American Mathematical Society. In a letter to Nature 59 (1899) he described the Gibbs phenomenon. Let us illustrate the Gibbs phenomenon by Example 7. Therefore Josiah Willard Gibbs. ) t d t . .5X (7. Gibbs showed that the limit deviation of SN(Z) in a neighborhood of a jump point xo is greater than the jump j(x0) at xo by an amount of 18%.02.1839-28.1903.Fourier Series and Fourier Method for PDEs 213 1+ &osnt 2 where n=l = sin (N + 2sini a> t ’ D N ( t ) := is known as the Dirichlet kernel. Then sin ( N 2sing + 4)t ’ Let us consider now the so called G b s phenomenon3 describing the difib ference between the partial sums of the Fourier series and the value of the function near a jump point. We have /Z sN(x)= sin ( N + a) t d t + l Z Jo t (& t) - sin ( N +.3.

n] such that Taking x~ = --+ 0 as N --+ 00.27) we get N+1 N-oo to lim SN(zN) . The graph of the function Si is plotted in Figure 7.2. Si(n)] there exists a unique t o = Si-l(p0) E [-n.Si(n).8519. for every po E [. Making the change of variable s = ( N integral of (7.26) we obtain + 3) t in the first (7. As the function Si(t) is monotone increasing on [-n. it is invertible and n]. Si(n) M 1.214 Partial Differential Equations uniformly in N . Figure 7.2. by (7.27) Let us consider the so called Sine integral Si(t) = -ds. Graph of the function Si (x). The maximum of Si(t) is attained at n. S sins It is easy to see that Si(t) is an odd and bounded function.

29) eix + e .1. l l Z n7rx f ( x ) cos -dx. n EN. e-ix ..ix . .cosx .28) (7. 1 .Fourier Series and Fourier Method for PDEs 215 N + 112 = 1 to -ds S sins = po E [.8519. while Note that J f ( 0 ) = f(0 N+w + 0) . = = bn Ll 71.Si(7r).17902 M 18%.30) a.Si(7r)I. 1 n E NU{O}.f(0 .i s i n x .8519 . 1 n7rx f(x)sin-dx.0) = 7r and the Gibbs amount is lim (sN (*) -sN (-*)) - Jf (0) N 1. If po = Si(7r) then to = 7r and we get N+W lim S N (m) Si(7r) 7r = M 1.-ix (7.i x 2 Let us consider a Fourier series cosx = sinx = 2i * (7.5707 = 0. Finally in this Section let us note that there exists a complex form of Fourier series based on Euler's formulae eix = cos x + i sin x.5707 1.

n E N.216 Substituting (7.32) (7. (7. n E Z. that is .31) with coefficients (7.30).29) in (7.i T f ( x ) d x .isin -)r x ( nrx 21 1 1 f(x)dx = e . Simply we have cn = 1 21 / 1 e . we obtain Partial Differential Equations or more briefly 00 (7. and where is the complex conjugate of Cn. Let us note that the system of complex exponentials has the orthogonality property.31) n=--00 where = n L J1 1cos -.i y f ( x ) d x .33) -1 The series (7.32) is known as the complex form of the Fourier series.

Fourier Series and Fourier Method for PDEs

217

if n # m and

7.2

Orthonormal Systems. General Fourier Series

The system of complex-valued functions {cpn(x)}defined on the interval I = [a,b] is said to be orthonormal iff

Ja

for every m , nE N. For example, the system

{ h e i n z } is orthonormal on

rb

[-T,

TI.

Let {cpn(z)}be an orthonormal system and f(z)be an absolutely integrable function on [a,b]. Then the numbers

are well-defined and are called the Fourier coeficients of f with respect to (9,). We write as before

n=l

where the series is called the Fourier series of f with respect to {cpn}. The partial sums of the Fourier series have a minimal property in L2-sense.

218

Partial Differential Equations

**Theorem 7.8. Let f E L2[a,b]and ( p n ) be an orthonormal system in L2[a, Then the function b].
**

rb

I

n

2

dx attains its minimum at the point (c1, ...,C n ) , where ck is the k-th Fourier coefficient o f f . Proof. By the orthonormality of the system { p n } we have

n

rb

n

k=l

n

k=l

(7.34) By (7.34) it follows that

@(?I,

...,y n ) is minimal at

rb

(q, ..

Cn)

and (7.35)

n

**Corollary 7.1. Let f E L2[u,b]and ( p n ) be an orthonormal system in L2[a,b]. Then
**

rb

00

(7.36)

Proof. The result follows from (7.35). As the series C,"=lI C n l 2 is convergent it follows that lcnl + 0 or C n 0 as n + 00. W

**The inequality (7.36) is known as Bessel inequulity4.
**

4F'riedrich Wilhelm Bessel, 22.07.1784-17.03.1846.

Fourier Series and Fourier Method for PDEs

219

Let us consider some special examples of orthonormal systems generated by boundary value problems for linear second-order differential equations.

**7.2.1 The Bessel functions
**

The Bessel equation of order p is

x2y"

or

+ xy' + (x2 -p2)y

=0

(7.37)

Y" + ;d+ (1 -

1

5)

y = 0,

z

# 0,

where p is a nonnegative constant. As the equation (7.37) is a linear secondorder differential equation its general solution is of the form

Y = ClYl

+ C2Y2,

where y1 and y2 are two linearly independent solutions of (7.37), and c1 and c2 are arbitrary constants. The Bessel function of the first kind of order p is defined as

(7.38)

where

qp) =

lw

+

CZJ-,(Z).

t p - ~ d t

is known as the Gamma function. If p is not an integer, then J-p(x) is a second linearly independent solution and the general solution is

Y = ClJp(Z)

If p is an integer, then

and so J'Z -() function

is not a second linearly independent solution. In this case the

Yp(x)=lim

4+P

Jq(x)cosq7r - J-,(x)

sin q7r

7

220

Partial Differential Equations

known, as the Bessel function of the second kind of order p , is a second linearly independent solution. Let us recall some properties of the Gamma function r ( p ) .

3. 4.

5.

r(n + 1) = n! if n is a positive integer.

Ir(-n)l

= 00, n = 0 , 1 , 2 ,...

qp) =

r(P+ k) if p P ( P + 1)...(P+ k - 1)

# 0, -1, ..., -k

+ 1.

The solution (7.38) is derived by the power series method for solving second-order differential equations. For simplicity let us consider it for the case p = 0. Suppose the solution of the equation

xy” + yt + xy = 0

is presented in the form of power series

(7.39)

Differentiating (7.40) twice and substituting in (7.39) we obtain

( 2 ~ 2 3 . 2 ~ 3 4 . 3 ~ 4 +~ ~ ~ ~ 3 ...) +(cl 3- 2c22

+

+ 3c3x2 + ...) + (cox + c1x2 + c2x3 + ...) = 0.

+

(7.41)

**Equating the coefficients of the powers x k , k = 0,1,2, ... in (7.41) to zero we obtain c1 = 0,
**

(n

+ 2 ) 2 ~ , + 2 + cn = 0,

n E N.

As c1 = 0 it follows

and

**Fourier Series and Fourier Method for PDEs
**

Then a solution of (7.39) has a form

221

(7.42)

Note that, by D'Alembert criterion, (7.42) is absolutely convergent for every x . If Q = 1 we get the Bessel function of the zeroth-order

It is a solution of (7.39) with initial data

y(0)

=

1, y'(0) = 0.

Similarly, the Bessel function of the first-order

is the solution of the equation

x2y"

with initial data

+ xy' + (x2 - 1)y = 0

y'(0) = 2.

y(0) = 0,

1

Let us note that the Gamma function and the Bessel functions of the first and the second kind of order p are denoted in Mathematica by Gammab], BesselJb, x] , BesselY b,x] respectively. The graphs of the functions Jo(x),Jl(x) and J2(x) are plotted in Figure 7.3. Note that the functions Jh(x), k = 0,1,2,, have a countable number of positive zeros. The first three of them are computed by the Mathematicu program

a[l, 1 = FindRoot[BesselJ[O, x] == 0, 1 a[1, 2 = FindRoot[BesselJ[O, x] == 0, 1 a[l, 3 = FindRoot[BesselJ[O, x] == 0, 1 a[2, 1 = FindRoot[BesselJ[l, x] == 0, 1 a[2, 2 = FindRoot[BesselJ[l, x] == 0, 1 a[2, 3 = FindRoot[BesselJ[l, x] == 0, 1 a[3, 1 = FindRoot[BesselJ[2, x] == 0, 1

{x, l}] {x, 5}] {x, lo}] {x, l}] {x, 5}] {x, lo}] {x, l}]

222

Partial Differential Equations a[3, 2 = FindRoot[BesselJ[2,x] == 0, {x, 5}] 1 a[3, 3 = FindRoot[BesselJ[2,x] == 0, {x, 7}] 1 Table[a[i,jl, ti, 1, 3), {j, 1, 3)l Do[g[n]= Plot[BesselJ[n,XI, {x,0, 20}, Plotstyle -> {GrayLevel[O.S],{}}I, {n, 0, 211 Show[g[O],g[l], g[2], PlotLabel -> "Bessel's functions J[n,x], n=0,1,2"] The first three zeros are as follows:

JO(x):x -> 2.40483, x -> 5.52008, x -> 8.65373, J1 (x):x -> 1.32349 10-20 , x -> 3.83171, x -> 10.1735, 52 (x):x -> 0.00086538, x -> 5.13562, x -> 11.6198.

Bessel ' s f n tm u ci

J[n,x], n=0,1,2

Figure 7.3. Graphs of Jo(x),J l ( z ) and & ( x ) Let us consider the orthogonality of the functions { f i & ( p k Z ) } , where are distinct positive zeros of J o ( z ) . Note that the function & ( A x ) satisfies the differential equation

pk

**Lemma 7.2. T h e system of the functions { & J O ( p k x ) }
**

[O, 1 1

is orthogonal o n

)Zyly2 = 0. we have .X. .A ) . 1 + . J 1 0 Wl92da: = = -9.b + &2 Y =0 by zyl the equation Y2 I1 and substracting we have or (4dY2 .43) 1) * JA (A 1)Jo (X2) + A2 JA (X2) Jo (A If X i = p k and A2 = p n are distinct positive zeros of Jo(z) follows it By (7.(l)yl(l) (7. Then (A? .Fourier Series and Fourier Method for PDEs 223 and Proof.(l)Y2(1) -A1 + y.Yl))' + (A: . Multiplying + y: + X?y1 = 0. Let yi(x) = Jo(X1x) and by zy2 the equation 1 y' : y2(z) = Jo(X2z) for A1 # X2.43) Letting X -+p k .Y.

1752-10. XI. For n 2 0 an integer. 5}] Adrien Marie Legendre.1)"). OH. (-1 P (x)= .x2)y'(x))' + n(n + l)y(a:) = 0.44) where n 2 0 is an integer. {x. bounded solutions of (7.1833.2 Legendre polynomials The Legendre dzfierential equation5 is ((1. -1. 5}] Do[g[n] = Plot[LegendreP[n. . ~ ) ( x )= ( 1 5 . XI.2.4 by the program Table[LegendreP[n. 18. defined as 1 d" Pn(z)= -. 0.31. + 322) . {n. l}.44) are polynomials. b. Plotstyle -> {Gra~Level[0.70x3 ~ +6 3 ~. 0. known as Legendre polynomials.01.44) is bounded on the interval [-1. I].224 Partial Differential Equations 7. 2"n! dxn Using Mathematicu we find that the first six Legendre polynomials are as follows .( ( x 2. The points x = f l are singular points and not every solution of (7. (7. P2 (x) = P 4 P 5 (x)= ( 3 .09. ~ ) They are plotted in Figure 7. 3 (-3a: + 5x3) .30x2 + 3 5 ~ .

5".2.(z)I 5 1 if 1 1 5 1.2.(-I) = (-l). (x)are orthogonal on (. P. 7".l).P.2(x)dx = 2n+1' Let us prove 5" using the procedure considered in Lemma 7.(x)as follows: 1". P.(x) has n real simple zeros on the interval (-1.l) for n 2 1. 2". 2 P.. Multiplying by P ( ) the equation .4 .Fourier Series and Fourier Method for PDEs 225 Figure 7. P. IP. 3'.1. Graphs of the first six Legendre polynomials We list some properties of P.(1) = 1.(z) is an even function if n is an even integer. P. and Pn(x) is an odd function if n is an odd integer.1.x .. . x 4". for every n = O.

I ( x ) d z n = 1 1 ~:-I(x)dx. . ..~ ( x ) ~ . .(x)dx = (2n - valid for n = 1 . . = -1 Therefore for m #n fl P. multiplying by [-I.~:(x)dx. ( x ) =xn d S_.(z)dx 0.( -..1 in 6". 11.1]we obtain (m(m = (m(m 1) .(z)P.P: . 2 .n(n + 1)) / 1 ~~(x)~. ( x ) ~ .n(n + + I))P. and (2n + 1 ) S_. x ~ .P.226 by Pn(x) equation the Partial DifferentialEquations and subtracting we have or ( ( 1 . Let us prove 7" using 6". 1 Therefore by (7.P~. Hence n .(x)and integrating over ( 2 n . + 1) . we get P. Multiplying the identity 6" by integrating in [--1.1]by 5" we have &I) .45) we obtain the identity (272 + 1) 1' P.X~)(P.45) Replacing n by n . Ll (7.1) JI: -1 x ~ .P.))' Integrating over [ -1.(x)dx = 0.

.. . 0 -7r<x<o.".sin32 2 2 3 = +-- . then In (2cos ) . + (-1y -+ . -7r (b) sin3 x.2 prove that if -7r z2 z3 z4 +-. Expand the following functions in Fourier series: (a) eaz.') (21 < x < 7r . <x <T... (a) Expand in Fourier series the function y = Si(x). 2. -7r < x < 7r. + (-1y -+ '".** * (b) Find the sums of the series cos2x --3 sin22 --3 if -7r cos nx + . (4 f (4 = (dl g ( x ) = { { e" sinz 0 0 5 x < 7r/2.-+ 2 3 X sin22 . 8 n -1 2 cos3z < x < 7r.. 3. where a = const # 0. T/2 < x L 7r.Fourier Series and Fourier Method for PDEs 227 Exersices 1..sinx . cos2x cos3x cosz . (a) Using the Euler formula and the expansion ln(1+ z) = z . OLx~7r.-+ 3 4 '. 8 n2 .1 sin32 sin nx + . -7r < z < 7r (b) Show that: + 2Si7r).

Find the general solution of the differential equation y" 5 + -y/ + y = 0.1 n=O n! Prove that: C Bn (t + 1) .01.1 n=O C Prove that: Bzn-l = 0 if n 2 2.= -n 2 7 r n=O Compute first seven Bernoulli polynomials and plot their graphs with Mathemutica. (a) The Bernoulli numbers Bn6 are the coefficients of the Taylor series of the function X -= " 1 B n x n . 29. (b) The Bernoulli polynomials Bn ( t )are defined as the coefficients of xn in the expansion xext .1782. (c) Prove the identity 5. X using the change of variables u = x2y. . z ex .03.1700-17. 'Daniel I Bernoulli.Bn ( t )= ntn-') B1 sin 2n1rt ( t )= t .. 1 -ex .228 Partial Differential Equations 4.O0 -Bn ( t )xn. Compute the first seven Bernoulli numbers.

O) = p(x) o<x<z.a 2 X " ( x ) T ( t )= 0 or T'(t) Xll(x) -a2T(t) . u(x.t ) = 0 t 2 0. t > 0.3. t > 0.t)= 0 0 < x < 1. Our goal is to find the solution of ( M D H ) using the method of separation of variables or Fourier method. we get X ( x ) T ' ( t ). Q2T(t) X ( x ) .3 Fourier Method for the Diffusion Equation 7. t 2 0.t ) = u(1. Plugging this into the diffusion equation.X ( z ) * In order for the last relation to be an equality each side must be identically equal to a constant: T'(t) X"(x) -= -= A.Q 2 U Z X = 0 0 < x < 1.t ) = X ( Z ) T ( t ) to the problem (sD) { u(0. u(0.1 Homogeneous equation and boundary conditions Consider the boundary-value problem ( M D H ): { ut .t) Ut ct2uzx= 0 = u(l. A separable solution is a solution of the form u ( z .Fourier Series and Fourier Method for PDEs 229 6. Prove that 1 7.

X Q 2 T ( t ) = 0. then the equation (7. Consider the following three cases: (i) Let X = solution p2 > 0 . X ( x ) has the form X ( z ) = c1x + c2. (iii) If X = solution -p2 < 0. while T ( t )satisfies the equation (7.46) has the general By the boundary conditions it follows and c1 = c2 = 0 because (ii) If X = 0.46) T'(t). So in the first two cases the problem (7. X ( 0 ) = X(1) = 0. It follows again c1 = c2 = 0. . p > 0.46) has the general X ( z ) = c1 cospz + c2 sinpz. p > 0. u(1.P ~ .47) We are looking for the values of X which lead to nontrivial solutions.46) admits the trivial solution only.230 By the boundary conditions Partial Differential Equations u(0. (7.t ) = X(O)T(t)= 0. Then the equation X ( Z ) = clePZ + ~ 2 e .t ) = X(Z)T(t)= 0 it follows X ( 0 ) = X(Z) = 0 so that X ( x ) satisfies the following eigenvalue (Sturn-Liouville) problem X / / ( z )= X X ( z ) 0 < z < 1. (7.

The last identity means that An are the Fourier sine coefficients of cp(x).( y ) 2 tsin (y).x I 1 n = 1 . i. Solving (7..t ) = A n .50) . . p ( x )sin -dx. the function (7.e. In order to find a solution of ( M D H ) we take a superposition of un(z. (7. Namely. n E 2.49) n=l is the solution of ( M D H ) provided that cp(z) = n=l nnx C An sin -’ 1 00 OLX<Z. . Then = nr.48) are solutions of the problem ( S D ). 1 The above values An are called eigenvalues and the functions X n ( x ) ezgenfunctions.8l = 0.t ) . (7. we obtain and has the form Therefore functions of the form un(z.. . n E N.47) with X = An . An = nn f J. 2 . EN. So the only nontrivial solution of (7.Fourier Series and Fourier Method for PDEs 231 By the boundary conditions it follows that for a nontivial solution c1 = 0 and sin .46) appears when nrx X n ( x ) = an sin -.

l ] . cos I ’ O ~ x g ) o<x<z.( (&)2 +A?) 7 CA?5 n=l 00 1 1 cp12(x)dx. T h e n the problem ( M D H ) has a unique solution given by (7. Proof.49) is uniformly convergent and u(x. In order to justify the method we prove Theorem 7 9 Let cp E C [ O . we have and the series CrX1 is convergent because lAnl IAnl and by Bessel inequality I . . Let us note that the function p(z) satisfies the assumptions of Theorem 7.cp‘(x) be a piecewise continuous function .232 Partial Differential Equations So the formal solution of ( M D H ) is the function (7.49) with coefficients An determined by (7. and cp(0) = cp(Z) = 0.6.49) and (7. we have cp(z) and cp’(z) -2 -C 00 nrx An sin I ’ n=l nrx A.50).t ) is a continuous function. n=l As in Theorem 7.. By uniform convergence it follows U ( Z .6. Since by the Weierstrass criterion the series (7. 0) = C A n sin nrx 7 n=l cp(X)) 0 5 x 5 1.50) . Taking the Fourier expansions of the odd extension of cp(x) on [-l) 11.

Formal differentiation yields (7.. By the maximum-minimum principle for the diffusion equation it follows that the solution obtained is unique. As S and T are TI.[Ant is convergent. we obtain (7. I) x ( 0 . we obtain (7. Suppose 6 5 t 5 T and 0 < x < 1. t ) satisfies the heat equation. Similarly.53) we get ut = a2u. 7.2 Inhomogeneous equation and boundary conditions Let us consider now the boundary-value problem for the inhomogeneous diffusion equation .Fourier Series and Fourier Method for PDEs 233 Since the boundary conditions are satisfied let us prove that u ( x ."=.52) and (7.51) and (7.3. I) x [S. arbitrary we have ut = a2u. by the Weierstrass criterion and Theorem 7.51) observe that there exists no such that for n 2 n o because for n 2 no. Combining (7.53). in (0.51). As the series C.53) In order to prove convergence of (7.52) n=l (7.. in (0.4. Let us fix 0 < 6 < 7'. ~ ) .

a2uz2 = nrx C fn(t>sin -’ I 00 we obtain = (7.54)’ we obtain or In order to calculate An(0) observe that u ( x . 1 where .t ) into the equation Ut .54) n=l where 00 n=l U Then equating coefficients in (7.234 Partial Differential Equations To solve ( M D I ) we apply the method of variation of constants (parameters) looking for a solution of the form Substituting formally u ( z .0) = and therefore nrx C An(()) 7 P ( X ) sin = n=l 00 A J O )= Then the solution of ( M D I ) is Ju 1 nrx p ( x )sin -dx.

Z] and f ( x ) E C[O$]. 1 n E N. Finally in this section we consider the case of inhomogeneous boundary conditions and the method of shzfting the data. Consider the diffusion equation with sources at both endpoints The last problem can be reduced to a problem ( M D I ) subtracting from u any known function satisfying the boundary conditions (BCs) .9 it is left to the reader as an exercise.Fourier Series and Fourier Method for PDEs 235 an = nr f 1 cp(x)sin -dx.x 1 1 In order to justify the solution we consider the case when f(z.) = f ( x ) .cp'(z) and f( x ) are piecewise continuous functions and ' cp(0) = cp(1) = f(0) = f ( 1 ) = 0. Since the proof is similar to the proof of Theorem 7.10. Suppose p(z) E C[O. Then the problem ( M D I ) has the unique solution where 1 fn = n7rx f(Z)sin-dx. t Theorem 7.

s ( 2 . = w(I.t) satisfies the problem ( M D I ) { wt . t ) = u(2.t)= (1 satisfies the BCs. + .T ) P ( 0 ) w(0. 0 II 2 I w ( 2 .-q’(t) X I 0 < x < I ) t > 0) O<a:<l. If we have an inhomogeneous equation and inhomogeneous BCs we can split it into two problems and Solving ( P I )and (P2) by previous procedures we obtain that u(x. 2 2 ) . t Then. p ’ ( t ) . which has been considered before. = . 42. 0) = .236 The linear combination Partial Differential Equations s(x.t) =0 t 2 0. Consider now X p(t) 2 + 74(t). t ) is a solution of ( P ) .t ) .Q w.t)= w(z. as the function w(z.t) 0.(1 .(1 . ) .t ) w(z.

u ( 1 .Fourier Series and Fourier Method for PDEs 237 Example 7. The function s(z) = 2(1 . t ).2 os < 1. v(0.1 . t ) = 2. t > 0 .1 .1 .4. 1 4 = 0.2) s i n n m d x 1 The solution of the original problem is u(2. s(1) = 6. t ) = u ( z .s(z) we have vt .0) = s i n 2 r z .x) solution satisfies St + 6 x = 2 + 42. t >_ 0. known as a steady state . 1 4 = 0 ~ if 0 < x ~ ~ w(x. t > 0 . where cn = 2 J/. zs1. Solve the heat conduction problem for the copper rod of length one ut . ) = sin2nz + 42 0 OlxL1.56r2t . t ) = 0 The last problem has the solution t 2 0. 1 4 = 0 ~ O < x < l . ~ ~ ~ ~ ( 0 = 2.t) = v ( 1 . t ) = 6 Solution.t ) = 2 + 4 x + sin 2 m e-4. ~ ~ u(x. u ( 0 . (sin2xx . ) Then for v ( x .

(z) = a. =0 O<a:<Z.c%.t) = X ( Z ) T ( t )= 0 it follows X ( 0 ) = X(Z) = 0. sin .. Our goal is to find the solution of ( M W H ) using the Fourier method.t) X(x)T(t) = to the problem ( S W ): { u(0... 0 < 2 < 1.xy2) .2. A separable solution is a solution of the form u(x. Plugging the form into the wave equation. which describes the motion of the vibrating string. t 2 0.. .. = 0 = u(1. As before the problem ( P ) has nontrivial solutions nrx X.56) (PI : XX(2) { X”(Z) -X(1) 0. I ’ . t>0) ( u(0.4.12 = 1.t ) = 0 O<z<Z.238 Partial Differential Equations 7.55) By the boundary conditions U ( 0 ) t ) = X ( O ) T ( t )= 0 ) u(Z.t>O. we get X(z)T”(t) C2X”(Z)T(t)= 0 or T“(t) -_-. So X(z)satisfies the problem (7.t ) utt c2u. X(0) = = = 0.1 Homogeneous equation and boundary conditions Let us consider the Dirichlet boundary value problem (BVP) for the homogeneous wave equation ( utt .4 Fourier Method for the Wave Equation 7.A.t)= U ( Z ) t ) = 0 t > 0. C2T(t) X(5) (7.

t ) = An cos -t ( nm 1 + Bn sin n E N. t ) .57) Plugging (7.55). known as norrnal modes of vibration. 1 In order to justify the formal solution we prove l nrx B n Z nm -1 Theorem 7.2. are solutions of the problem (SW).Fourier Series and Fourier Method for PDEs corresponding to 239 x =A . (7. In order to find a solution of ( M W H ) we take a superposition of u n ( x .Namely. Suppose cp E C2[0.11.. 1 n E N. n E N.”(x) is piecewise continuous.(z. (7..59) (7.’p’”(z) is piecewise continuous $ E C1[0. we obtain the ODE TI’@) + with general solution nm 2 (I)= 0 T(t) Tn(t)= bn cos -t nm 1 nm + cn sin -t.Z]. $. 1 n E N. t)= n=l nm n n1rx C (An cos -t 1 + Bn sin -t 1n c ) sin -. we are looking for a solution of the form U(X.58) Formally the last function satisfies the initial conditions if n= 1 m Using the Fourier-sine series for ~ ( xand +(z) we obtain that ) An = 2 1 1 nrx cp(x) sin -dx. Therefore functions of the form u.60) +(x) sin -dx. 00 (7. =- (7) .57) into (7. and . .Z]. 2 n = 1.

Proof.61) Then the function (7. Q(0) = +(Z) = 0. sin -t I nTx sin 1 are uniformly convergent in ( 0 . l ) x (0. Since nm nm (An cos T t Bn sin -t sin 1 I(y)’ + ) * nyxl I (T) nr 2 + IBnl) it suffices to show that the series (7.9. is the unique solution of the problem ( M W H ) . the main tool to justify differentiation of the series (7.58) is Theorem 7.60). m).58).59) and (7. where the coeficients A n and B n are determined by (’7.63) is convergent. As in Theorem 7.64) n=l u Similarly for the Fourier sine series of $ ( x ) and $”(x) (7. We show that (7.65) n=l U . (7. Let us take the Fourier sine series of cp(x) and the Fourier cosine series of (~”’(x) x E [O.6.58) and the series uzz(x.t ) = - n=l c (T) nm 2 (An + B.240 Partial Differential Equations cp(0) = Cp”(0) = cp(Z) = cp”(Z) = 0.Z] for (7.t) = -C (y)2 00 nm nm cos I t nrx n=l 00 u t t ( x .

because by (7.2.65) are uniformly convergent in [0.64) and (7.66) By the Bessel inequality. 1 ($) B:.61).63) is convergent.1] and the initial conditions are satisfied. 3 (7. we have An = - (&) A: 00 3 and B. we obtain 1 Ar2 5 n=l (p1'12(x)dx and Then the series (7.63) it follows also that the series (7. . = .66) By the convergence of the series (7.60) and using conditions (7.59) and (7. The uniqueness of the solution follows by the energy method Section 3.Fourier Series and Fourier Method for PDEs 241 Integrating by parts (7.

W ( 0 . substituting (7.t) in the Fourier sine series i vtt .68) n=l where vn(0) = vI(0) = 0.70) . n=l where (7.69) Formally. we get The last linear second-order ODE with initial conditions (7.242 Partial DifferentialEquations 7.t) = 0 0 < x < I . ( W I ): The last problem can be solved by reducing it to the Cauchy problem for the inhomogeneous wave equation.t ) u(z.t) of the form (7. (7.t) 0 < z < 1 . t ) = w(1.t)= u(Z.O) = cp(4 ut(z.c2vxz = f(z.O) = 0 OLXLZ.O>= $(XI u(0. t v(z. Another approach is to expand f(x. O<z<Z. by odd reflection of f(z.t) = 0 > 0.t ) with respect to z = 0 and x = 1. t > 0.4.67) Let us try to find a solution w(x. Consider now the mixed BVP for the inhomogeneous wave equation ( M W I ): The solution of ( M W I ) can be constructed by superposing the unique solution of ( M W H ) with the unique solution of the problem { utt OLXLZ) t > 0.69) has the unique solution wn(t) = n m f f n ( r sin (y(t ) - T)) dr.C 2 U X Z = f(x.O) = 0 o < z g vt(z. (7.68) into the wave equation.2 Inhomogeneous equation and boundary conditions . t > 0.

1 BVPs for the Laplace equation in a rectangle We consider now the Laplace equation u x x u y y 0 in D .Fourier Series and Fourier Method for PDEs 243 The solution of ( W I ) is (7. considering W ( x . y) : 0 < z < a .5 wtt . t ) = &)) w(Z.71) where D = { (2.68). . t ) = h(t) t > 0.c2w. t ) = 0 t + + + 0 < x < 1. where f n ( t ) and vn(t) are determined by (7. t > 0.x)g(O) xh(0)) 0) Wt(2. OLxLZ. 0 < y < b} is a rectangle in a plane. ( W B ): In order to solve the last problem. ) = W(Z. t > 0. On each side of D we assume that either Dirichlet or Neumann boundary conditions are prescribed.5. wt(x.70). These problems can be solved by the method of separation of variables. Namely.x)g”(t) xh”(t)) W ( x .67) and (7. Fourier Method for the Laplace Equation 7. = 0 O<S<Z) t > 0 ) w(x.t ) .-((Z = .x ) g ( t ) x h ( t ) ) t we reduce ( W B ) again to a problem of the type ( M W I ) : + { 7.((I .x)g’(O) X h ’ ( 0 ) ) W(0. we use the method of shifting the data. ) = l w ( x .c2wx. o<x<z. = -((Z . = + (7.O)= -((1 .O) = 0 O<X<Z. w(0. Let us consider finally the case of inhomogeneous boundary conditions The solution of the last problem can be found by superposing the solution of the problem ( M W I ) with the solution W of the problem with zero initial data and source wtt ..O) = 0 o < x g .

The solution of the problem has a form u = u 1 and u 2 + u2. Nontrivial solutions of (7. 1 for a constant A. 0<y (7. (7. Y ” ( y ) .244 Partial Differential Equations Example 7.71) with the boundary conditions Solution.AY(y) = 0. y) 1 We find each one of u1 and u2 by the Fourier method.75) X’(a) = 0. 0 < x < a .74) are Yn(y) sin. Solve (’7.b = nTY .71) respectively with the boundary conditions u1 ( q 0 )= u ( q b ) = 0 1 O<xLa. Separating variables = X (x)Y (y) we have This implies that X ” ( 2 )+ X X ( x ) = 0.5.73).74) (7.73) < b. Since the function u satisfies (BC1) we should have Y ( 0 )= Y(b)= 0 (7. and for u (z.72) (7. where u1 satisfy (7.

71) and (BC2).= .) = 9 n=l nrx nra Can (cash 7 tanh -sinh . y ) = X ( s ) Y ( y )satisfies (7. .77) Suppose now u 2 ( z . we have the equations (7.73) for X ( x ) and Y ( y ) with the boundary conditions Y ( 0 )= Y(b)= 0 and X ( 0 ) = 0.tanh -sinh b b We are looking for a solution 00 u1 of the form u ~ ( x .E N . n 2 The differential equation for X ( x ) X”(X)implies that (Y) 2 X ( x )= 0 nrx nrx 2 X ( x ) = C1 cosh -+ C sinh b b c 2 nra .rry (7.sin 7.tanh -.Fourier Series and Fourier Method for PDEs corresponding to 245 X=X..76) b b It satisfies the boundary condition when which implies that (7. As before.rrx) n.75) is satisfied if Then X ( z ) has the form nrx nra X n ( x ) = U.72) and (7. (cosh . C1 b The condition (7.=-(y). n.

Let us consider the problem + uyy= O in x 2 + y 2 < 2. which yields (7.246 Partial Differential Equations Then corresponding to nrY Y n ( y )= sin b For X (2) X ” ( x )- (Y) 2 X ( x ) = 0. Y ) .77) and (7. In polar coordinates the problem reduces to upp 1 1 + -up + --t/. X ( 0 ) = 0.Oe P P2 = 0 if 0 5 p < a.Y) =q Y ) .) = y the condition n=l nrx nry C bn sinh -sin b b U2x(a. which implies nrx X n (Z ) = bn sinh b ’ Looking for u 2 ( x .79) .78) Finally the solution is ‘LL(X.2 Dirichlet problem for Laplace equation in a disk u. 7.y) in the form 00 nEN u ~ ( x .Y) = 2 L 1 ( X . . (7. where an and bn are determined by (7. should be satisfied.78). Y ) + U Z ( X .5.

(7.XR = 0 + for some constant A For the function O(8) it is natural to require the periodic . Making change of variables p = et it reduces to with general solution Rn(t) = cnent Then + dneVnt. Plugging into equation (7. 8 ) = u ( p cos 8. we obtain 0'' + XO = 0 (7.81) and (7. # 0 and multiplying by p2 . Note that if X = 0.83) Then (7.82) is an Euler equation.81). is a nontrivial solution of (7. The separation of variables means seeking a solution of the form u ( p . psin 8) and h(8) = g ( a cos 8. n E N . then the equation (7. (8) = an cos n8 + bn sin n8 is the corresponding solution. 8 ) = R ( p ) 0 ( 8 ) .82) and p2R" + pR' .1 Pn and we have a separable solution of the form . nEN R n ( p ) = &pn + dn-.81) (7. boundary condition o(e)= o(e 2 4 .79) we find that R"O + -R'@ + -R@" P P2 Dividing by RO 1 1 = 0.Fourier Series and Fourier Method for PDEs with boundary condition 247 where u ( p .83) imply that X = n2 and 0. a sin 8 ) . If X = n2.

8) = + C pn(An cosn8 + Bn sinno).r 2 =xrncosnt= 1 .86) The solution of the problem (7. 00 n=l By the Fourier expansion of h(8) we observe h ( ~ ) nTdT.79). (7.82) reduces to Partial Differential Equations pR" with general solution + R' = 0 + do In p.2 r c o s t + r 2 ' : rcost .248 If X = 0 the equation (7. Then 00 s c : = Crnsinnt= n=l 00 r sin t 1. It is surprising that this series is summable explicitly and the result coinsides with the Poisson integral formula for the Dirichlet problem in a disk. 1). (7. At p = 0 some of these solutions are infinite if dn # 0 and bo # 0.86).84) with coefficients (7. We are looking for a solution of the form u ( p . Proposition 7. The functions un and uo are harmonic in D.84) which satisfies the boundary condition if h(8)= A0 2 + C Un(An cosne + Bn sinno). n E N. cos n E N U {0}. (7.80) is the series (7. Ro(P) = co So we have a separable solution for some constants a0 and bo.1. Let r E [O.%cost n=l + r2' . 00 (7. We reject these terms in order to have bounded solutions.85) and (7.85) and h ( r )sinnrdr.

A0 5+ n=l pn(Anc o s d + Bn sinn0) .I ) t .rn sinnt + rn+l sin(n .r cost)' + ( r sint)2 .rn cosnt + rn+l C O S ( ~. Bn = r sint . Cn + isn n-1 = C rk k=l (COS kt + i sin kt) n-1 = Crkeikt (1 .r cost .1.:: r k cos kt and Sn = By the Euler formula we have r k sin kt.r 2 1 .Fourier Series and Fourier Method for PDEs 249 Proof. Then An Cn = 1 .r e v i t ) (1 .1)t. We have rn -+ 0 as n + 00 because 0 < r < 1 and 1. Theorem 7. Consider the partial sums Cn = C.12. 1-2rcost+r2 - where An = 1 .2rcost +r 2 .. Then the sum of the series ~ ( p0) = .%cost r2' r sin t S = lim Sn = 12-00 1-2rcost+r2 - + .1 1 (An +iBn) .rneint) (1 .1. Suppose that the function h(0) is continuous and 2nperiodic.r c o s t C = lim Cn = -1 72-00 1 .%cost + r2 rcost .

Solve the heat conduction problems for an aluminium bar of length 2 Ut .0 . u (z. t ) = 0.2 u p ~ ~. u (0. We have By Proposition 7.p2 2 (a2 .t) = 0.1 as 5 < 1 we have - apcos(8 -1+ a2 . 0) = 10 cos 2. 0) 7TX .87) Proof.t)= u (2.2 ~ cos(8 .-)p2 p2 p r + 2 T) - a2 . Exercises 1.T () e p2) ~ + . = sin -. t ) = u (2.86) is (7.85) and (7.. 8 6 = 0. t > 0 ~ ~ with initial and boundary conditions as follows: (4 (b) 2 u(0.250 Partial Differential Equations with coeficients (7. u (x. ~ 0 < x < 2.

u ( 0 )t). 0 ) = 1 0 5 ~ 1 3 .x 3. ) = 0 0 0 5 ~ 5 3 . 3 U t (z.= u (2)t ) = O# O < X < l . (z. U t -u.ku. (d) u (X) ) = 0 { t2)0 20x. 6 . t ) = U ( 3 ) t )= 0 u (x.t ) = 0 t > 0.z(2)t )1 (2 ) u. = 0 0 < x < 1 ) t> 0) O<z<l.O) = 0 0 5 ~ 5 3 . x =0) 0 < 2 < 3)t> 0 : TX (4 u (z)0) = 1 .O) = 0 u ( 0 . t )= u ( 3 . t ) = u. 2 5 x 5 3. t > 0. Ut + 2 u = 0 0 < 2 < 1 .cos . <z 52) = 0. (3. 40 l < z L 2 . u ( 0 )t ) = u ( 1 .O < z < 3 .O)= c o s z O<z<l. t > 0) u(z. 2. Ut u (z)0 ) = {: oLz<l) 15x52. u ( 0 . Solve the following mixed problems for the wave equation utt .22 Ut (2. Solve the BVPs for the diffusion equation: (4 { .Fourier Series and Fourier Method for PDEs (c) 251 u (z)0 ) = 0 O<z<l. u (Z) ) = To 0 u ( 0 .u .: (0) = u. 0 5 ~ 5 3 .. t > 0. t )= 0 . t ) = 0 . t > 0. t ) = T I ) U ( 1 ) t )= T2 t > 0. u (0.

O) = ( P ( 4 U t ( Z ) O )= 0 U ( 0 ) t ) = U ( Z ) t )= 0 + 0 < x < I. O < y < T } : : (a) uy (2. Solve the mixed problem for the wave equation utt .4uxx = sin3x U ( Z ) O )= 0 ut(x. y ) = uy (x..t > 0.252 Partial Differential Equations u ( x . 5.4u.rr) s i n % . u(x.. 6.O)= 0 o < x g . > 0. . u ( 0 .rr.t)= U ( 7 r ) t ) = 0 utt utt .t)= 1) u(1.O) = 1 o < x g . = 2 sin x 0 < x < T ) t > 0. ux(r... t>0. 0 < x < Z. where a is a constant.0) = uy (qT ) = ux(0. u(x. For which w does the resonance (growth in time) occur ? 8. t O<z<l.O)+ u ( x . O<x<Z.O)= cosx u(0. Prove Theorem 7. 7. Ut (2.10. t > 0) O<x<l. with zero initial and boundary conditions. O ) = xcos7l-x 0 < x < 1. Solve the boundary value problems for the Laplace equation in the square K = {(x. t )= U ( 1 ) t )= 0 t > 0. Solve the mixed problem { U t t .~)O < x < T .c2ux. 9) = (b) u ( 0 . y ) = 0.O)= 0 O < Z < l ) u(0. = cos3y.c2 u . O<Z<T) OLx<.t)= 1) t > 0. y) = 0. 4. Solve the following mixed problems for the inhomogeneous wave equations: (4 . = Asinwt. . ut(x. O )= u ( n . t > 0. . au = 0 u(x.

u ( x c . Let Un(z.Fourier Series and Fourier Method for PDEs 253 9.y3 on x2 + y 2 = 1. y) + Prove that: (4 . u (x) = y n on cc2 y2 = 1. uyy= O .. u . y ) x4 . (a) Using Euler's formula prove the identities: (b) Solve the Dirichlet problem in x2 + y2 < 1. = + 10.y) the solution of the problem be u. + u y y = O in x 2 + y 2 < 1.

y) (d) The series C m=l 00 is divergent for x2 + y 2 < 1. U3(x. is convergent for o 5 q < I.3) using Mathy). 00 ( c ) The series m=l & (2. x2 + y2 < 1. y) ematica. y) .while the series m=l C 00 qmUzm+l (x.254 Partial Differential Equations Plot the graphs of the functions U1(x. U2. (z. and u5(x.") is divergent.

z .t)u3(z. Then u ( x .t) a solution of ut .t)u2(y. ) . t ) = ~1(x. We have 255 . Suppose u1 (x. y . t ) and u3(z. y.kuSs = 0 . u2(y. . where s E {x.kAu = 0 in R3. z ) . t ) are solutions of t the one-dimensional dinusion equation ut .1. z ) E R3 and #(P) is a given function. is Proof.1 Diffusion Equation in Three Dimensional Space Let us consider the Cauchy problem for the diffusion equation in R3 where P = (x.Chapter 8 Diffusion and Wave Equations in Higher Dimensions 8. y At first observe Proposition 8.

. $ and 0 are bounded and continuous functions..1). = 0.1) G3 ( P .3)) where cp.t > 0.3) Proposition 8. Then u ( P jt ) = S. Suppose that $ ( P ) is a function with separable variables (8.2. (8.t ) is again called the Green's function or fundamental solution of (8.tM(Q>dQ.~ A = 0.~)= e-4lct 2 m is a fundamental solution of the diffusion equation ut .256 Partial Differential Equations The function X2 G(x. By Proposition 8. P u E R3. G3P- Q .. Observe that Po0 P o 0 We consider the case when the initial data $ ( P )is a function with separable variables 4(P>= cp(4lCl(y)W* (8.1 the function is a solution of Ut .ku.

q s> E R3 is a solution of (CD3).1. (Bernstein). .(z) = 2 (.7 and Proposition 8.kAu = 0 for ( P . Proof. 1 T h e n Bn(z) -+ f(z) 1.Diffusion and Wave Equation in H.1] .z y k . Let f(z)E C[O. J: E [O. t ) E R3x ( 0 . unzformly for Theorem 8. ~ ) and By linearity Proposition 8. . 06.2 can be extended for any initial data which is a finite linear combination of functions with separable variables of the form n ($71 (p) = k=l c k q k (z)'$k b ) O k ( 2 ) (8.03.1968.1] as n + +w. 257 with Q = (t.1 it follows that ut .1880-26.10. given by B. Recall Bernstein's' polynomial for a bounded function on the interval [0.5) on bounded domains.D. 'Sergej Natanovich Bernstein.5) Let us show that any continuous and bounded function on R3 can be uniformly approximated by functions of type (8. Separating integration we have By Theorem 4.)f (!) k=O z"1 .

1 there exists n such that < -' 2 for ( x .ll3) . n nk k=O m O = We have that 4 n ( x . Let E > 0.y . 5 There exists a function with separable variables & ( P ) E C ([O.z)x'(l-z)n-"ym(l-y)"'m. 112. Let $ ( P ) E C ([0.m. ( z . Let $ n ( x . z ) E [O.258 Partial Differential Equations Proposition 8. & & 2 2 By the construction of ~ $ ~ y. we have I&(x. y. z ) = e g (L)(2)$(b. lI3). By Theorem 8.y. By the same way there exists nk & such that for (y. z ) E [0. z)I 5 M . x ). [q5(P)[ M and E > 0. y . such that I4n(P)I I and M Proof. z ) E C ([0. iI3. lI3)is a function with separable variables and & k=O = -+-=&.3.

Let us show that (8.8).t)= +(P) tl0 uniformly on bounded sets of R3. z ) can be uniformly approximated by a function &(z. RI3) with separable variables. 8M I 2 & I$(P) . R]') with separable variables such 1 e-IP I dP' < . Let I+(P)I 5 M .10). (8. Suppose E > 0 and B c R3 is a bounded set. y.&(P)I < if P E [-R. By Proposition 8.2&P')dP'. (8.6).9) and (8. (. (8.2&P' we have 1 u(P. z ) I---+ we can prove that every function $(z.6) .RI3. By rescaling the variables (z. RI3. z ) E C ([-I?.2 it follows that u(P.1).6) holds. q .Diffusion and Wave Equation in H.10) Finally for P E B c K R and t E (0.P E R3 and denote K R = [-R.1) on R3 and lim u(P. There exist R that: > 0 and & ( P ) E C ([-R. Making the change of variables Q = P . by (8.and B c [-R. Proof. (8. Then the function is a solution of the diffusion equation (8.2.-' -' 2R 2R 2R E C ([-R 'q3) x+R y f R z+R Theorem 8.t ) satisfies (8. Let R 259 > 0. RI3.t)= - e-IP'I2$(P.y . then P'EKR & max l$(P . k~= R 3 \ K ~ .D.7) where P' = ( p . we have .$(P)l < 4 for P E KR. Suppose 4 ( P )E C(R3)n L"(R3). r ) E R3.Z a p ' ) . y .9) By continuity of $ ( P ) there exists 6 > 0 such that if t E (0.

W The same arguments hold as well for the diffusion equation in higher space dimensions. Gn(P .kAu = 0. The fundamental solution for the diffusion equation in R" is given by G"(P. t c R" be a . where P = ( X I . u(P. t = T } . E R" UP. t)4(Q)dQ The solution of the inhomogeneous problem .) = t ~t S.O)= 0 P E R".t) : P E R . t )E R" x (0.-+-+-+-=E. I = aRT\{(P.. u(P. 8 M 4 4 & & & & 4 which completes the proof. Following previous steps one can prove that the solution of the Cauchy problem Ut .t ) = 1 -- PI2 4kt 2nJ(nkt)ne . is given by The maximum principle holds in higher dimensions as well.kAu = f ( P .. . .Q ..xn) and /PI =-/ .O) 4 ( P ) P = is given by ( P . OO).t ) E R" x (O. Let R domain RT = R x (O.260 Partial Differential Equations < 2M.t ) ( P .OO).T).

oo).oo) x R.zyz? ( u k cos k x bk sin kz) is called a 2. (b) If f(z) E C[O. t > 0)) % ( w . t ) E R x R+. Find solutions of the problems (4 (b) R3 { ut Au 0 )0 z2yz u(2. .y.z > 0 .t) 0 ( y . = b. o o ) .kAu = 0 in { ( z .kAu = 0 in {(z.(x) = a0 trigonometric polynomial of n-th order if u i b i # 0. y . then there exists an even trigonometric polynomial T.y. = 0. 0) = z2yz . Y.. t > 0).4 E R2x ( 0 7 4 .D. t ) : (z.r . z .(z)is even if bl = bz = . Find a formula for the solution of the BVP for the diffusion equation in half-space ut .Au = 0 in R3 x ( O .(z) such that for every z E [0. 4 (z. Y ) E R2. nT n Exercises 1. z .y E R . Suppose that u E C2(S2. = u ( w .) ut = kAu in S Then 2 .y) E R2. - = in x (O. t ) = 0 ( X .r] and E > 0 .Diffusion and Wave Equation in H. u ( z . ut . = ut .t) : z > 0. u(O. T. . Prove that: ( a ) The function cosk x can be represented as an even trigonometric polynomial of k-th order. O .y. 261 n C ( f i ~is a solution of the diffusion equation ) max U max u. 4 x 7 Y. I( X 4. . z . Using the reflection method find a formula for the solution of the BVP for the diffusion equation in half-plane. y. n nT minu = min u.7r] + ct=l + + 3.. The function T. 0) = d h 3 . O ) = + ! )> Y ) E (0.

t) = @(Z.262 Partial Differential Equations 8. (8. Such BVPs are as follows: ut = kAu in R x (O.12) .y. t ) = 0 on dR x [O. where R c R2 is a bounded domain with standard initial and boundary conditions on do.n ~ ~ t ) + a u ( ~=~ .11) we see that @ and T must satisfy where X is constant. y.00).oo). d( Oot ) Separating variables u(z.Y)T(t) and substituting into (8. 00) or -(x. 00) dn or dU dU . This leads to the eigenvalue problem for the Laplacian -A@==@ inR with boundary condition @=O or ondR (8.oo).2 Fourier Method for the Diffusion Equation in Higher Dimensions ut = kAu In this Section we shall apply the Fourier method to the diffusion equation in R x (0.y. t ) = O on 8 R x [O.11) u(2. n d R x [o.

consider the heat transfer problem on a circular plate . An = * As an example. 263 -=0 dn or -+aQ>=O dn ondR (8.y.y) as t 4 0 in the mean-square sense in R.14) It can be shown that for each one of the boundary conditions (8.12)-(8. Denote by <pn the eigenfunction corresponding to A n with the understanding that not all of An are distinct. (8.15) which satisfies the initial condition if n=l By the orthogonality of (Qn) it follows that JJa @(z. t ) -+4 ( x .D.~ ) d z d y Y (8. Y)dZdY If we suppose 4(x.13) dQ> ondo.16) JJ* @?&.Diffusion and Wave Equation in H. ) Q n ( Z .14) there is an infinite sequence of eigenvalues and an infinite set of orthogonal eigenfunctions which is complete.15) is convergent for t > 0 and u(x. Solving the ODE fof T ( t ) we find ~ ( t ) = ane-kXnt. We are looking for a solution of the form (8.y) E L2(R) it can be shown that the series (8.

-orc ler U(P>= JO(XP) (8. We are looking for a solution of the form (8.17) The first solution of (8.t ) = u( d m .17) is Y (Xp).19) where /-&4 +oo as k 4 oo are the zeros of the Bessel function J.17). Then the eigenvalues of the problem (8.264 Partial Different i d Eq uations Using polar coordinates for u(P. (8.-J(z).( a A ) 2 t while U(p) satisfies the Bessel equation Uyp) 1 + -PU / ( p ) + X2U(p)= 0.19) are (8. (8.we have Separating variables we have As before T ( t )= C k e . The boundary condition U l p = = 0 is satisfied if ~ Jo(XR) = 0.. t ) . which we do o not take into account because it is infinite at zero.17) is the Bessel function of zerot.20) k=l which satisfies the initial condition if .t ) = u ( p .18) while a second linearly independent solution of (8.

In polar coordinates the problem is The solution is where By the orthogonality of {fiJo(fLkr)} it follows . it follows ck = - (8. Example 8.21) The solution of the problem is (8.20) with coefficients given by (8.D.2. Solve the problem Solution.1. 265 For r = 5 E [0.1] we have k=l By the properties of the Bessel function.Diffusion and Wave Equation in H.21).1. Subsection 7.

v.1}] .O]] .{v.v] .1 at the instants t = 0.b r] hO=ParametricPlot3D[Evaluate[x[r. {r . t ) at the instants t = 0. The surface ( P c o s ~ . -0 V.1.hl .O .a r]+Exp [-b" 2 t] BesselJ [0. 0.4 . h2=ParametricPlot3D[Evaluate[x[r . { r .4 using the Mathematzca program Clear [a.y .1.1.t -1 :=Exp [.v].PlotRange-> {.1}.1.1. Temperature u ( p .O .u[r.v] .v]. P s i n w P .52 x[r-.1} {v .y [r.266 The solution is u(p.l } .v.lr] 1 is plotted in Figure 8.h2}].PlotRange-> {-1.x.y [r .v-]:=r Sin[v] u [r-.1}] Show[GraphicsArray[{ h . 0.u[r. 0.2 1 Figure 8.0.2Pi}. Shading->False. 0. O Frame->True.O .2}] [r h 1=Paramet r icPlot 3D[Evaluate[x .4]].u [r.2.v].a"2 t] BesselJ [0.u] a=2.11] .t ) = e-P%7o(plp) Partial DifferentialEquations + e-pgtJo(pzp).v.2Pi}.O.v] [r. where p1 = 2.O .b . Shading-> False. {r . 1 x [0.PlotRange->{.l?rameTicks->None] 0.52. Shading-> False.v-]:=rCos[v] y [ [r-.O) E [O. t )> )(P.{v .4 b=5.y.2Pi}.v-.0.O.4 and p2 = 5.

Looking for a solution of the (8. Using polar coordinates for u(P.22) k=l by we find where . 0. 267 Finally in this Section we consider the problem with the Neumann condit ion ut = a2Au 0. we have Separating variables for U ( p ) we find U(p) = Jo(Xp) and U‘(R) = 0.t ) .D.t ) = u(p.Diffusion and Wave Equation in H. Then JA(XR)= -Jl(XR) = 0 and the eigenvalues are where vk are the zeros of the Bessel function 51.

0 p t 2 0.up u(p.2. t>0.268 Working as in Lemma 7. we have Partial Differential Equations Therefore the solution of the problem is (8. Solve the problem .22).t) = 0 osp<1.p2 u(1. 2. Solve the problem { The solution is Show that ut=upp+.O) = 1 . < 1. where n rl Exercises 1.

Diffusion and Wave Equation in H. Assume that ( P ) E C3 (R3) and 2c. Huygens’ Principle 8. t ) are as follows: lo. cp. t ) = G(P.24) as in the D’Alembert formula.23) with initial conditions (8.23). 269 8. ( P ) E C2(R3) . t ) = 4n where 1 ’ u ( r .t ) r 0 2Gustav Robert Kirchoff. .T .03.1 Kirchoff’s formula. G(P.3 Kirchoff’s Formula for the Wave Equation.3. Let us denote to be the mean value of u ( P .(P) = { Q : IQ . If u ( P . Then there exists a unique solution to the problem (8.24) given by the formula which is due to Poisson but known as Kirchofl’s formula2 To derive it we shall use the so called spherical means introduced by Poisson.r . (8. We are looking for the solution of y.c 2 A u = 0. 2) E R3. (8.1887.t ) and P = (2. (8. t )sin cpdcpde. (8. then r-+o lim fi(P.26) 2O. . t ) = u ( P .23) where u = u(P. The linear wave equation in R3 is Utt .10.1824-17. 12.PI = T } with center P and radius T .D. Some properties of G(P.8. t )is a continuous function. t ) over the sphere S. . Spherical means.

00) and ~ ( r E C2(0. From 7 ) 2 1 8 1 1 + -u. ) . 9. 2 ~ and cp E [ O . If u(P.q P ) r . Let us change to spherical coordinates for Q(<. If u(P)t ) is differentiable in t then ) d .rx). z ) then y. + r2 sin1cp dcp (sin quV)+ -uee r r2 sin2 cp (8. q = y+rsingsincp c where 8 E [ 0 . then P +--+ i l ( P ) r . t ) = u. T ) ] Au(P. we have Proposition 8. t ) is twice differentiable in (x. t )E C" (R4) . cp) t ) with respect to 9.4.r = . t ) = K(P) )t). The ) .q. T at (8.26)) the F'ubini theorem and the periodicity of u(r.270 Paxtial Differential Eq uations 3'. If u(P.t) E C" (R4)rn ' 5 ) E N .27) 4'. Let 4(r) E C3(0. C ) { = z+rcosgsincp z+rcoscp. Proof of 4'.

t ) for T 2 0 we easily find that This problem for the wave equation on the half-line has a solution i Wtt . .)..O).-. if 0 5 r 5 ct. Introducing W ( T . if r > ct.(t$(ct)) 2c dt = + 2ct %U(Ct) . ( r . w t ( q 0 ) = rW(r). T E (0.r ..(t+(ct))+ tw(ct). ~ x (0. oo) w(0. In order to find . dt d . t ) = rw(r.c2w1-r = 0. 2c d .$(. t E (O.O). t ) = 0.(Opt) observe that where sr E [ct . c t Then + r ] .Diffusion and Wave Equation in H. t ) E ( 0 . r E (O. ) w ( r .0) = . 271 Proof. w). sr + ct as r + 0.D.

P E R2.O) = q5(P).2 Wave equation on R2. t > 0.3. P E R2.t ) d = . We can derive the solution formula for the wave equation on R2 x R+ Utt (CW2) . = . P E R2.t ) = E(P.Met hod of descent. u(P.4. In general if 4 E Cm+' (R3) and $ E C" (R3) m 2 2.O) $ ( P ) . Huygens' principle.272 Partial Different ial Eq u ations Derivation of Kirchoff's formula Applying the mean-sphere operator to (8. 0 . However.(@(P. and 2' we obtain U ( P )t ) = E ( P . the solution is less regular than the data 4 and $. If q5 and $ are of class C2 (R3) then the second derivatives of u might blow up at some point even though the second derivatives of 4 and $ are bounded.23). In the case of the oncdimensional wave equation the solution given by D'Alembert formula is as regular as the initial data.r )t ) . because of the t -derivative in the Kirchoff's formula. u~(P.we have Then by Proposition 8.24) by properties 1" .c 2 A u = 0 . t ) = w(0.5" for w(r. 8. (8. t ) ) + t q P )C t ) c at which is Kirchoff's formula. in the threedimensinal case. then u E C" (R3 x R+) . This is known as the focusing effect.

y. c ) .O) and solution of (CW2) is = (Q. q .c) = ( < . 273 Assume that 4 E C3 (R2) and $ J C2 (R2).C) = ( c . where = (x. We where are the upper and lower hemisphere.Let us transform the integral where have = (P.O) E R2 x (0) and = (Q.Diffusion and Wave Equation in H. ( ) E R3.y.O) = (z. q .D. For both hemispheres and In order to solve (CW2) we can consider it as a problem in R2 x (0) By Kirchoff's formula c R3.By previous calculation the .

t)# O if t E ( t o . . As an application of Kirchoff's formula we consider the so called Huygens' princzpZe3. until . > 0 in the interior of K . with @(P. t o +S).&I = ct does not intersect K for these values of t. This means that C C the sound passes through P for C. it is heard t z(P K ) at the point P 4 K from the instant z(P' K .Pol = ct only. the values of $ and q at the point POE R3influence the solution on the sphere IP .274 Partid Differential Eq u ations which is Kirchoff's formula for the wave equation on R2.&I : Q E K } .K ) . K ) max{lP .07. If u(P.) K or ct > d " ( ~K ) . Kirchoff 's formula implies U ( P . @ has a compact support K = { P : + ( P )# 0) and I.Z(P. .QI : Q E K ) . in the three dimensional space . ~ " ( P .t)= 0 if t < t o u(P.K)= cto} 3Christian Huygens.t ) # 0 if t E ( t o .o ) . According to it. J(P.6.1629 08. Denote for P 4 K I z(P.t)= 0 if t > t o u(P. a compact set. Suppose for simplicity $ = 0. t ) = o if ct < d ( ~ . t In the case under consideration $ = 0 and $ IK 2 0 implies and and { P : d"(P.t ) is a sound produced on K .K ) C time and moves with speed Let us introduce the forward and backward wave fronts at the instant t o as u(P.K ) = min{ IP . = which exist because the function g(Q) = ) P.04. because the sphere J P.) as an initial speed. 14.1695. u(P. - c @(to).QI is continuous on K .

275 so w ( t 0 ) and @(to) are non empty sets.D. ) C The sound produced on K will be heard infinitely. For instance. Verify that Kirchoff’s formula gives the solution of the problem (8. This means that Huygens’ principle does not occur in R2. This phenomenon does not occur in the two dimensional world (Flatland) where sounds are heard forever. t j --+ 00 such that u(tj)# 0.PI 5 ’(p:K) C ct and contains K for t 4 K > -.t) intersects K for t E > 0 if t > -because the disk I& . p . . Namely. this is the Huygens’princzple.Diffusion and Wave Equation in H. It does not hold in R2 where *(to) = 0. (8.This means that there exists { t j } . We are lucky to live in a three dimensional space because we are able to hear every sound for a finite interval of time. in the case 4 = 0 and $J IK 2 0 in R2by Kirchoff’s formula it follows that u(P.23). for $ E C2(R3)show that = c2Au. Exercises 1. Namely.24) in the case 4 = 0.

y.t) dU an + au(z.35) (8. Nodal Sets In this section we shall apply the Fourier method to the wave equation utt = c2Au (8.29) we see that @ and T must satisfy where X is constant.oo).Y. t> (8.29) in 0 x ( O .d $J(Z.33) Separating variables u ( z .O) = b.y.Y.30) Ut(z. o o ) (8. = @(x. (Z)Y>E f4 (8. where R c R2is a bounded domain with standard initial End boundary conditions on dR as follows: U(Z. t ) = 0 on dR x [0.Y) E 0.00) or -(z.I.36) -=0 dn a@ onda .31) dn =O on dR x [ O .t)= O on dR x [O. This leads to the eigenvalue problem for the Laplacian A@=A@ i n n with boundary condition <p=O or ondR (8.y.y. 3 ) W and substituting into (8.4 Fourier Method for the Wave Equation on the Plane. o o ) . (.34) (8.276 Partial Differential Eq uations 8.O) = $.32) or -(z.Y).Y.t) dU (8. u ( z .

derstanding that not all of An are distinct. Solving the ODE for T ( t ) we find T ( t )= An cos c a t + Bn sin c a t .D.Diffusionand Wave Equation in H.y ) : x2 y 2 5 1) satisfying the problem + (dm-. We are looking for a solution of the form which satisfies the initial conditions if n=l By the orthogonality of (an)it follows (8. Determine the radial vibrations u t ) = u ( p . or 277 (8. .2.39) Example 8.t ) of the circular drum D = { ( x .37) Denote again by @n the eigenfunction corresponding to A with the un.

The solution is where = 2.4 and p 2 = 5. In polar coordinates the problem is Separating variables as in Section 8.2. 2 . The surface . ifk>3. we find the solution where n r1 In our case By the orthogonality of { f i J O ( P k f ) } it follows ak= { 01 if k = 1 .278 Partial Differential Equations Solution.52.

u] a=2.y [r .0 ] . 1 x [O. x/pl .2 at the instants t = 0.>{.u[r . O 1 { r .1} .u [r .v].1. -0 Figure 8. Frame-> True.v-.2}] h l =Par ametr icPlot3D [Evaluate[x[r .v].{v.b. O .y. Shading. 2x1 1 is plotted in Figure 8.x.{v. x/p2.Pi/ a]]. (P. t>) . x/p2.{ v.O .0.4 b=5. .a r]+Cos[b t] BesselJ[O.v.hl .1}] h2=ParametricPlot3D [Evaluate[x[r .>{.1}] Show[GraphicsArray [{ h0.52 x[r-.PlotRange >{.2Pi}.y [r .v-]:=rCos [v] y[[r-.u[r .Diffusion and Wave Equation in H.h2}]. P sin 0.D.2. 0) E [O.2Pi}.1.v .1. -0 0. The surface Dt at the instants t = 0.0. x / p l using the Mathernatica program Clear[a.>False. l } .FkameTicks->None] 0.O . { r .v].v.b r] h =Par amet r icPlot 3D [Evaluate[x [r .(P. 279 Dt : ( P cos 0. { r .v].v].1} . Shading->False.v-]:=r Sin[v] u[r-.PlotRange.t-]:=Cos[a t] BesselJ[O.O.2Pi}. Shading-> False. Plot Range.Pi/ b]1.v].y [r .

Let us consider nodal sets of some eigenfunctions of (8. E rI. uly=o . y) E n. Note that boundary points of II do not belong to Nu. y) (8.41).=0 - 4.UlY.1827 . a ) x (0. (8. Both eigenfunctions K = (0. There is a physical experiment due to Chladny4 which allows one to visualize the nodal set in two dimensions.t > 0.39) in the square and vnm have the eigenvalue (0. These sets are known also as Chladny's figures. curves and surfaces in one.!I) 0) u : 'U(Z. Covering the membrane II with fine sand and vibrating it with a given frequency the sand particles take the place of the nodal set.39). 30. with the usual initial condition and the Dirichlet boundary condition 4. because it presents the set where the rectangular membrane IT does not move at all. b) be a rectangle and consider the wave equation utt = c2 (uzx + Uyy) . two and three dimensions respectively.40) (8. (2.Let II = (0.11. The eigenfunctions of the problem -vxx . The nodal set has a physical meaning. The nodal sets consist of points.Y) = of an eigenfunction v of the problem (8.280 Partial Differential Equations Next we consider the wave equation on rectangular domains in R2. (x.=.v y y = xu. T ) x *Ernst Florens F'riedrich Chladny.04.1756-03. T ) .41) are corresponding to the eigenvalues We shall discuss the nodal set N = {@.b - = 0.

y) = Vn-l (x.Y) is a mode of vibration with frequency w.y) E Nn .1) y) . .D. ) = sin 2nx sin y y + sin x sin 2ny. and divides the square K into several different regions which vibrate independently. then the points also belong to Nn .1) x + cos (2n. = aurnn(x.a)Unm(x. = Xnm = m2 + n2 and vibrate with angular frequency w =C J X .1 1 21. It is clear that if (x. = O} Y Y) of the function Vn(x.Diffusion and Wave Equation in H. For each Q! E [O. Using mathematical induction one can prove: Claim 1'. Vn(z. ) E K : Vn (x.. to the nodal set Nu. We consider now in detail the nodal sets Nn = { (x. 281 x = Am.The nodal set Nva is the curve It varies from the nodal set Nu. .Y) + (1 . ) y + 2 sin x siny (cos (2n ..

where g (5.y) = sinxsiny (cosx +cosy) (8cos2 z .l 2n) The first four functions V. = sin z sin y (cos z + cos y) z (224 (cos2(. fi (x.C O S ~ + 24 (C O S ~ cos z + cos y C O S ~ y X) X) + C O S X) .1).3.2coszcosy . y) are expanded as follows: & (x.C O S ~ Y )6 and v4(z. = sinz sin y (cos x + cosy) f (z.cosy c0s5 z) (8. ~ -24 C O S ~ C O S x + 10 ~ (COS2 Computations are made by MAPLE in Scientific WorkPlacE using Expund+Fuctor.. + ( .cos II: C O S ~ y) O ~ x + + + C O S ~ X C O S ~ Y + C O S X C O S ~ . Y). y) = sin z sin y (cos + cos y) g (z. y) for n 2 2. (z.4 + 8cos2 y) .-1) + cos2(n--1)y) + .42) +16 (c0s4y cos2z + 6 cos2y c0s4x) . or n = 1 .C O S ~ COSY . a ) = sin2xsiny +asinxsin2y = 2sinxsiny (cosx + acosy) the curves c. For the function Vl ( z .10 cOS y COS Z .282 Claim 2O.5 are presented in Figure 8. y .. 4 are presented using Plot2D+ Implicit. Partial Differential Equations V. . : cos x + a cos y = 0 for a = 0. = ~ ~ ( c2o sC ~ S y . The nodal sets h/. y) where f (z. 2 . Y) .5.16 (c0s5 zcosy . y) = 8(16 (cos62 + cos6 y) .1 y . 1) 1.C O S ~ X .16 c0s3y c0s3 z -24 (C O S ~ . (z.

Diffusion and Wave Equation in H.D.

283

Figure 8.3. Curves c, : cosx

+ acosy = 0 for a = 0.5, 1, 1.5

+

& is presented

Using the expansion of V (x, = sin 4x sin y sin II: sin 4y the nodal set 2 y) in Figure 8.4. It divides K into 4 regions.

0'

0.5

1

1.3

2

2.5

3

Figure 8.4. Nodal set

N 2

284

Partial Differential Equations

The nodal set Ng , where g is given by (8.42),is presented in Figure 8.5. The nodal set Ng divides K into 8 regions.

2.5::

2: :

i!5:1::

0.5::

I . . . . , . . . . , . .

_ , . _ _ . , _ . . _ , ~ ,_ + ,

0’

0.5

1

1.;

2

2.5

3

Figure 8.5. Nodal set

Ng

Claim 3’. The nodal set Nn divides K into 2n regions. Exercises 1. Solve the problem

using the method of separation of variables. Consider separately the “nonresonance” case w # wmn = 7 / and the “resonance” w = w rm q , for some (mo,no) .

Diffusion and Wave Equation in H.D.

285

2. Consider the axis symmetric bounded solutions u = u(p,cp) of the Laplace equation in a ball BR

(a) Separating variables u (P, cp) = R (PI @ (9) show that R ( p ) satisfies the Euler equation

while

(cp) =

(x) satisfies the Legendre equation

-( ( l - X 2 ) 2 ) dx

where @ ( 9 )= @ (arccosz) = 6 (z) . (b) The solution of the problem

d

+v(v+l)6=0,

(P2) is

**where Pn is the n-th Legendre polynomial and
**

Cn

= - h (cp) 2n

2Rn

+

1

0

Pn (COS q)sin cpdcp.

**(c) Solve the problem (P2)with R = 1 and h (cp) = cos 29. Plot the surface
**

p cos 8 sin cp

psinosin9 u ( P , cp) cos cp

,

with Mathematica for ( 8 , q ) E [0,2n] x [O,n] and p = 0.25, 0.5, 1.

3. Prove Claims lo-3O.

286

Partial DifferentialEquations

4. Plot the nodal set h / 3 using the expansion

**v3(2, = siny sinz (cos z + cosy) f (z, y) 9)
**

where

f (x, = ~ ~ ( cxo+sC ~ S ~ - C O S ~ cos y - cos x C O S ~ y) O y x y

+

COS~ COS~ z y

+ cos z

)

COSY

- C O S ~ - C O S ~ ) +6. z 3

Show that

**Nfintersects the diagonal II: = y of K
**

)

at the points 5.rr/6) .

( d 6 )~ / 6 )(7d3)~ / 3 ) ( 2 ~ ' 3 , 2 ~ / 3 )5, (@

)

287

References

1. M.L. Abell and J.P. Braselton, Differential Equations with Mathematica, (Academic Press, Boston, San Diego, New York, 1993). 2. L.C. Andrews, Elementary Partial Differential Equations with Boundary Value Pro blerns, (Academic Press College Division, Orlando, San Diego, New York, 1986). 3. I.G. Aramanovich and V.I. Levin, Equations of Mathematical Physics, (Nauka, Moscow, 1969), (in Russian). 4. L. Bers, F. John and M. Schechter, Partial Differential Equations, (Interscience Publishers, Inc., New York, 1964.) 5. P.L. Bhatnagar, Nonlinear Waves in One-dimensional dispersive systems, (Clarendon Press Oxford, 1989). 6. W. Boyce and R. Di Prima, Elementary Differential Equations and Boundary Value Problems, (Wiley, New York, 1992). 7. M. Braun, Differential Equations and Their Applications, (SpringerVerlag, New York, Berlin, 1992). 8. B.M. Budak, A.A. Samarskii and A.N. Tichonov, Problems on Mathematical Physics, (GITL, Moscow, 1956), (in Russian). 9. G.F. Carrier and C.E. Pearson. Partial Diflerential Equations. Theory and Techniques, (Academic Press, 1976). 10. J . Chaparova, Change of variables in some differential expressions via MAPLE, in Proceedings of XXVI-th Spring Conference of U.B.M., (Sofia, 1998), p. 389-394. 11. J. Cooper, Introduction to Partial Differential Equations with MATL A B , (Birkhauser, 1998). 12. R. Courant and D. Hilbert, Methods of Mathematical Physics, Vol. II., (Wiley-Interscience, New York, 1962). 13. G. Dassios and K. Kyriaki, Partial Differential Equations, (Athens, 1994), (in Greek). 14. R. Dennemeyer, Introduction to Partial Differential Equations and Boundary Value Problems, (Mc Graw-Hill Book Company, NY, 1968).

288

Partial Differential Equations

15. E. DiBenedetto, Partial Differential Equations (Birkhauser. Boston, 1995). 16. G.M. Fichtengoltz. Course on Diflerential and Integral Calculus, vol. 1 1 (GIFML, Moscow, 1963), (in Russian). 1 17. D.G. Figueredo, Analise de Fourier e EquaC6es Diferencias Parciais. Projeto Euclides. Rio de Janeiro, Instituto de Matematice Pure e Aplicada, (CNPa, 1977), (in Portuguese). 18. N. Finizio and G. Ladas, An Introduction to Diflerential Equations with Diflerence Equations, Fourier Series, and Partial Differential Equations, (Wadsworth Publishing Company, Belmont, Cal., 1982). 19. P.R. Garabedian, Partial Diflerential Equations, (Wiley, New York, 1964). 20. T. Genchev, Partial Differential Equations, (Nauka and Izkustvo, Sofia, 1982), (in Bulgarian). 21, T. Genchev, Ordinary Differential Equations, (Sofia University Press, Sofia, 1991), (in Bulgarian). 22. E. Godlewski and P-A. Raviart, Hyperbolic Systems of Convervation Laws, (Mathematiques & Applications N 3/4, Ellipses, ~1991). 23. A. Gray, M. Mezzino and M.A.Pinsky, Introduction to Ordinary Differential Equations with Mathematica, (Springer-Verlag, New York, 1997). 24. F. John, Partial Differential Equations, (4th edition, Springer-Verlag, New York, 1982). 25. P.Y. Kythe, P. Puri and M.R. Schaferkotter. Partial Differential Equations and Mathematica, (CRC Press, Boca Raton, NY, London, Tokyo, 1997). 26. Th. Kyventidis, Partial Differential Equations, (Thessaloniki, 1988), (in Greek). 27. R. LeVeque, Numerical Methods for Conservation Laws, (Birkhauser, Basel, 1992). 28. V.P. Michajlov, Differential Equations with Partial Derivatives, (Nauka, Moscow, 1986), (in Russian). 29. I.P. Natanson, Theory of Functions of Real Variable, (GITL, Moscow, 1958), (in Russian). 30. Ch. Philos, An Introduction to Differential Equations, (University of Ioannina, Ioannina, 1989), (in Greek). 31. P. Popivanov , N. Popivanov and J. Jordanov, Guidance on Partial Differential Equations, (Sofia University Press, Sofia, 1991), (in Bulgarian). 32. M. Proter and H. Weinberger, Maximum Principles in Differential Equations, (Prentice-Hall, Englewood Cliffs, NJ, 1967).

Special Functions. New York. 1991). 38. D. Tersian. Whitham. Inc. England. (Moscow. (Addison-Wesley. G.LXflerentiul Equations. Wiley and Sons. Wolfram. 1960)) (in Russian). S. An Introduction. (in Greek). .L. 40. Nodal sets or Chladny's figures.C. Diflerential Equations. Rozdestvenskii and N. Wokingham.A. Mathematica: A System for Doing Mathematics by Computer. Proceedings of XXVI-th Spring Conference of U. 46.M.D. 2nd edition. (University of Ioannina.New York. Fourier Series.. 1998)) p. Smoller. An Introduction with Mathematica. 1992). 1992). J. (Springer-Verlag. New York. (Dover Publications. (Patras. 35. Shock Waves and Reaction-Diflusion Equations.B. 47. New York. Trachanas. Sneddon. (in Greek). 1974). Strauss. Nauka. Tolstov.D. S.. Partial Diflerential Equations with Mathematica.W. (Moscow. a 36. (J. 1997)) (in Greek). Linear and Nonlinear Waves. Zachmanoglou and D. 1992)) (in Greek). Notes on Partial Diflerential Equations. 37. 369385. (Addison-Wesley.(to a p Pe4. 44. Wiley and Sons. 43. P. S. 41. 34.B. Siafarikas. Tersian. 45. Partial Differential Equations. ROSS. 1998). Propagation of one dimensional waves via Mathematica. New York. (Springer-Verlag. New York. 1986). S. Yanenko. Introduction to Partial Differential Equations with Applications. Diflerential Equations with Partial Derivatives (Athens. I. 39. (Athens. 42. 1995). Ioannina. Mathematica+. Elements of Partial Diflerential Equations. G.C. 49. B. (McGraw-Hill Book Company.References 289 33. .P. (SpringerVerlag. Stavrakakis.N. P. 1989). Stavroulakis. Systems of Quasilinear Equations and Applications to G s Dynamics. J. (in Greek). C. S. (J. 48. I. Berlin. (Sofia. 1992). Reading. N. 1983). 1988)) (in Russian). Inc. 1957). Tersenov. 1991). Vvedensky. 50. Rauch. Ordinary Diflerential Equations. Partial Dzflewntial Equations. MA. W. E. (Iraklio-Crete. Thoe.

This page intentionally left blank .

Answers and Hints to Exercises Chapter 1 Section 1. (c) F (Z+ y + u. (a21 .m y .3 + y + z . (b) u = f ( f ) + $)(c) u = e-cs’af(aY -bX) 5- 3. Section 1. (d) F ( x 2 + u2.29) (e) F (u .y + 2/-). Section 1. (a) It is a linear equation in three variables and the characteristic system is dX dY dz -=-=y-z z-x x-y The general solution is u = f (x (b) u = f (zyz.( r + u)2 . (b) F ( ~ + y + ~ . x 2 + y2 + z 2 ) . y ) =0. a35 +y + z) ) (c) u = u = e-c2/a1 f .2y. zyu) = 0. (a) u = xnf (f) . = 0. 291 .4 1.aiz) .2 1. (a) F ( r 3 + y 3 . ~ ~ + y ~ + = 0. x where a1 # 0.

(b) Irreducible. (a) autc + 2but.. ( + zg (:I > : Section 2.c) a.+ f u = 0. (b) u = zy. u = f (x + ct) + g (x . (c) u = 2xy (e) u = x + - 3 . (a) u = f (x + ct) + g (x .2 ....ct) + [a2 ( a x cos 3 -4 .$t3. (a) up = 3e3z+2t.a ) ut + (e .5 Partial Differential Equations 1.2 Y2 9 m' 2x& 1 22' + Y2 Chapter 2 Section 2. + * (b) 24 = f .O (d) up = &z4 . (c) up = - + ~ t+ ) sin ( a x + P t ) ] .d ) . (d . (a) u = y (d) u = + 22.f z t 2 . . 5.+ cu.292 Section 1.(b) up = ixe2x+4t. 3.. e2x 4.1 1.

3. (e) uqV= 0. 2. (4VYlY.W Y 3 Y 3 . . . q = 2 8 .Answers and Hints to Exercises 293 ify<O. Substitute ' 1 ~ wzl in the equation and choose w such that = 2 a i ~ . . if 9( ) : (b) u. 1 + Vy2y2 = 0. .cos x (d) u (5..3x2 + x 3 . y) = -229 + x) + g (y . y1 = g a x 1 + . Section 2. (c) u (x. + biw = 0. / 3 .1 2 22 x3. y) = f (y . 7 = x. XY + f (XY)fl+. ( = e-. . 7 7 = x .e-9. Y2 = -21 2x2 2x3 2x4. + + + + + + y3 = -5x1 (4 %y.2uXy+ uyy = 0.tv=O ( = x + 2 .x) . (c)u. y2 = -x2.3 Y1 = fix1 .fix*.fix:!.uyy= 0.fix3 . (a) 2uxz + uZy.J Z x 2 . u" + uqq = 0 ( = x. if y > 0.2 6 . Y3 = -21 5 2 . < 0.cos x . y4 = . 1 .vy4y4 = 0. + VYZIiZ .

Hint. 1+4 (b) ~ ( z .2 = 1.f i ) )+ e x (1 .3 1. Hint. . Section 3. = i. $4) .t) = e-btv(x. ( c ) The result follows from (b) if H ( t ) > 0 for t E [tl. t ) = $ (6 . + (c) u ( z .e Z F t (1 + 2t)). t ) .4 1. Chapter 4 Section 4. u (z. Make the change of dependent variable u(x. Then let E --+ 0. t ) % (e-x (1 = + erf (s. 0. 3. 1) u (x. replacing H ( t )by H E ( t )= H ( t ) one can prove the inequality for H E . y) = zt + cos z cost + sin z (1 . y) = (ez+t .erf (s A))). l). 2 ] because t If H ( t ) >.1 3 . y ) = Q ( 3 x 2 9 + + cosxcost.5 ) . e --2 t . (b) u (z. (a) u (z.32 Section 3.3 ' 3 1 .294 Partial Differential Equations Chapter 3 Section 3. y> = + (sin3 (z + t>+ sin3 (z .2erf (&)) . u (i.2 2.t) m 1 + E then for some E E (0.t ) ). (a) u(z. Section 4.cost) (c) u ( z .

2 1. The eigenvalues of A are A1 = 8 and X2. The solution is Section 5. Chapter 5 Section 5.vt. Section 5.4 1. Make the change of independent variable y = z . (a) The envelope is . Hint. Consider the function v(x. Section 4.4 2. Hint. (a) The charactersistics are The solution is 3. Use the characteristic method.3 = -1.Answers and Hints to Exercises 295 4.t) = u(x.5 2.xh(t) and use the even extension of the source function.t).

P ) and in the region { P : R 5 IQI I T.rotv'. Hint.2 2. Letting T -+ 00 and E -+ 0 obtain that fi if IPI > R.PI 2 E } . P-Q Section 6. 1 if IQI = R and lim u ( P ) = cp(Q). (a) Use the mean value property of harmonic functions (b) Show that the first partial derivatives are equal to zero. Use that div(ii x 5') = v' .I& .ii .4 (b) Show that (b) Use that (c) To show the boundary condition apply the Green's second identity to G ( Q .296 Partial Differential Equations Chapter 6 Section 6.5 3. .8. Section 6.follows as in Theorem 6.rotii .

Answers and Hints to Exercises 297 Chapter 7 Section 7. . f(4= 7+ ex .sin3x) a -7r 5 x 5 7r. 4.2 -51r n=l ZFi sin2nx) (-1)nn 0 < x 5 7r. (cos nx .2 (b) sin3 x = sinx .n sin nx) .sinx . (4 . (a) Use that x In (1 + e i x ) = 1n2cos 2 (b) Use that - + 2- x 2' 1 and results of (a). Hint. (a) Equate the coefficients of xn in both sides of the identity (c) Evaluate the coefficient to tn in the product (5 &&(s)zk) (5hBl(t)z') and use that k=O 1=0 .I ex -51r n=l c O0 (-I)~ I 1+ 1 4 O0 g(x) = . -7r < x < 7r# 3.

pJ.t) = r(wcosnnt + 9 sinnnx.To)) (To n sin n r x (b) u ( z . Y(4 = $(ClJ2(4 + C2Y2(4).t) = T 1 + (T2 .Tl + (-qn(T2 . = ~ sinnnt) (-& cosnt + 3 sinnt) s i n r x . (4 u(x."=1 .+l(x). 6. Use the indentity zJL(x) . 2. Hint. t )= t +p ~ (d) u(x.298 Partial Differential Equations 5.T1)x e-kn2n2t + : c.(x) = -xJ.

y.2 t ) (Z. z .cos 6 t ) sin 3 2 + C. t ) = x2yz + 2tyz.~) = (b) 4 4 = cosh 3x cos 3y 3 sinh 371- 3 cosh . u ( p .1 1.Z) . Z.Answers and Hints to Exercises 299 5. t ) = T + q R 2--- (. y.t :(1-2g)) -c 00 2e-(a~. (a) U (x. Section 8.2 2. (4 ~ ( 2 )= t.t ) = 9 ( X Z .2 sinh 3J: sin 3cosh% .where p n are positive zeros of J1 ( p ) = 0. 9 (1. (b) u (2."=. (a) u(x. -= sin 2nt sin n x 8.1/R)2t PZJO (Pn) n=l Jo (y) .a.2 s i n h F 2 % 2 Chapter 8 Section 8. .

This page intentionally left blank .

M. 46 Autonomous system 12 C Canonical form 7. 228 Bernoulli number 228 polynomial 228 Bernstein S. 59 Compatibility condition 84 Complex form Fourier Series 215 Conservation law 130 Conservation of mass 131 momentum 131 energy 92. 78 Neumann 4.Index A Absolutely integrable function 203 Almost-linear equation 2. 12 Cauchy problem first order equation 13 wave equation 68 diffusion equation 103 Characteristic curves 7. 257 Bernstein’s polynomial 257 Bessel F. 46 second-order equations n var. 131 Convergence absolute 202 mean-square 202 pointwise 201 uniform 201 B Bernoulli D. 218 Bessel equation 218 inequality 218 function first kind 219 function second kind 220 Blow up 21 Boundary conditions Dirichet 3. 11 form 60 strip 33 system 11 triangle 70 Chladny E. I . 62 Cauchy A. 280 Chladny’s figures 280 Classification 46 Classification second-order equations two var. 80 Robin 4 Boundary value problem 3 Burgers J.N. 140 301 .W.

34 Euler 247 fully-nonlinear 3 homogeneous 3.G. 68 D’Alembert formula 68 Dirac P. 199 Fourier cosine series 200 series 199 . 39 Liouville 151 Monge-Ampike 2 nonlinear 2 parabolic 48 Poisson 169 quasi-linear 3.P. 140 diffusion or heat 2 diffusion one-dim 97 elliptic 48 eikonal 2. 39 hyperbolic 48 inhomogeneous 3. 169 linear 2.B. 110 Euler L. 58 Burgers’ 2. 11 shock waves 2 transport 2 Tricomi 57 ultrahyperbolic 63 wave 2. 131 F First-order PDEs 3 First integral 23 Fourier J. 40 Korteweg-de Vries 2 Laplace 2. 67 Error function 73.J. 78 Dirichlet boundary condition 78 Dirichlet kernel 213 Dirichlet principle 178 Discriminant 47 Domain 1 Domain of dependence 70 influence 70 E Eigenfunction 231 Eigenvalue 231 Eigenvalue problem 230 Energy kinetic 92 potential 92 total 92 thermal 101 Equation almost-linear 2 biharmonic 2 Born-Infeld 2. 124 Dirac kernel 124 Dirichlet L.302 Partial Differential Equations D D’Alembert J.

182 Ill-posed problem 71 Improper integrals 104 Index positive 62 negative 62 Inhomogeneous 3 diffusion equation 118 wave equation 87 Initial value problem 3 Integral surface 12 Inverse Mapping Theorem 13 .Index sine series 199 identity 173 Fourier method second identity 174 diffusion equation 229 Laplace equation 243 wave equation 238 H Fully-nonlinear equation 28 Function Hadamard J.W. 70 biharmonic 181 Harmonic function 169 even 207 Harnack A. G I Gamma function 220 Gauss K. 193 Gamma 220 Harnack ’s harmonic 169 first theorem 195 odd 208 inequality 193 second theorem 195 piecewise continuous 204 Heaviside function 145 Robin 4. 170 Hopf-Cole transformation 141 subharmonic 181 Functionally independent functions 10 HuYgens 274 Fundament a1 solution 180 Huygens’ principle 274 Hyperbolic system 131 303 c. 91 Green’s first identity 174 function 7. 173 Gauss formula 174 Gauss-Ostrogradskii formula 173 General solution 4. 144 Green G. 213 Gibbs phenomenon 213 Gradient catastrophe 21. 40 Gibbs J.

304 Irreducible equation 43 Partial Differential Equations M Maximum-minimum principle diffusion equation 98 Laplace equation 171 Mean value property 175 Method of shifting the data 235 Method of steepest descent 142 Mixed problem wave equation 84 Mollifying kernels 124 Monge G. 37. 40 Linear equation 4. 46 Liouville J. 151 Liouville’s equation 151 theorem 194 Negative index 62 Neumann K.R. 80 Nodal set 280 Nonlinear operator 2 Non-factorable equation 40 0 Operator linear 2 nonlinear 2 Order 1 Orthogonal system 200 Orthonormal system 217 . 29 Monge cone 29 Mollifiers 124 J Jump discontinuity 204 K Kinetic energy 92 Kirchoff G.G. 169 Legendre A.S. 224 Legendre differential equation 224 polynomial 224 transformation 37 Linear operator 2. 80 Neumann boundary condition 4. 269 Kirchoff’s formula 269 N L Laplace P.

154 Riemann problem 154 Taylor B.Index 305 S P Parseval M-A. 142 Thermal conductivity 97 Thermal energy 101 Transpose matrix 61 U Ultrahyperbolic equation 63 .D. 63 parabolic 48. 103 Poisson formula diffusion equation 103 Laplace equation 189. 155 Sobolev S.L. 63 ultrahyperbolic 63 Poisson S. 63 hyperbolic 48. 190 Potential energy 92 Principal part 47 Schwarz H. 124 Sobolev space 124 Solution 1 Spherical means 269 Spherical wave equation 76 Steepest descent 142 Strictly hyperbolic system 136 Strip condition 33 St urm-Liouville problem 230 Superposition 231 Swimmer effect 83 Symmetry of Green’s function 183 Q Quasi-linear equation 3. 125 Second-order PDEs 39 Shock speed 155 Shock wave 123.Ch. 11 T R Rank 10 Rankine-Hugonio t condition 164 Reducible equation 40 Reflection method 78 Region 1 Regularizations 125 Riemann G. 205 Parseval’s equality 205 Periodic function 203 Point elliptic 48.F.A.

W. 105 Weierstrass criterion 105. 170 . 135 Weierstrass K. 202 Well-posedness 70.306 Partial Differential Eq ua tions V Variation of constants (param.) 234 W Wave equation 67 Weak derivative 123 solution 123.

Sign up to vote on this title

UsefulNot useful- Partial Differential Equations an Introduction 2nd Ed Straussby Melkzedekue Moraes De Alcântara Calabrese Moreira
- Differential Equations - Linear, Nonlinear, Ordinary, Partialby vladanea_alina
- Nonlinear Ordinary Differential Equationsby Gabo Greenwood
- [Mark S. Gockenbach] Partial Differential Equationby Rohit Mundat

- Partial Differential Equations With Fourier Series and Boundary Value Problems -- Asmarby Ryan Barrage
- Basic Theory of Ordinary Differential Equations Universitextby sfilop
- Elementary Partial Differential Equations with Boundary Value Problems - Andrewsby geoffscheid1094
- 0521194245 Differential Equationsby Jagdish Singh Mehta

- Partial Differential Equations and Fourierby api-3706602
- Introduction to Numerical Ordinary and Partial Differential Equations Using MATLABby Özlem Yılmaz
- 048646248XEquationsBby Doug Allen
- Vrabie I. - Differential Equations - An Intro to Basic Concepts, Results and Applications (World, 2004)by Alaa Bani Bakr

- Partial Differential Equations I
- partial differential equation
- Partial-Differential-Equations
- partial differential equations
- Partial Differential Equations an Introduction 2nd Ed Strauss
- Differential Equations - Linear, Nonlinear, Ordinary, Partial
- Nonlinear Ordinary Differential Equations
- [Mark S. Gockenbach] Partial Differential Equation
- Asmar - Partial Differential Equations
- Partial Differential Equations 2
- Partial Differential Equations 3
- Differential Equations
- Partial Differential Equations With Fourier Series and Boundary Value Problems -- Asmar
- Basic Theory of Ordinary Differential Equations Universitext
- Elementary Partial Differential Equations with Boundary Value Problems - Andrews
- 0521194245 Differential Equations
- Evans EDP.pdf
- Differential Equations in Applications
- Fourier and Laplace Transforms
- Haberman - Elementary Applied Partial Differential Equations 2e
- Advanced Engineering Mathematics With Matlabb 2nd Ed
- Mikhailov Partial Differential Equations
- Numerical Mathematics With MATLAB
- Pinchover-Rubinstein - An Introduction to Partial Differential Equations (Cambridge, 2005)
- Partial Differential Equations and Fourier
- Introduction to Numerical Ordinary and Partial Differential Equations Using MATLAB
- 048646248XEquationsB
- Vrabie I. - Differential Equations - An Intro to Basic Concepts, Results and Applications (World, 2004)
- Ordinary Differential Equations
- Differential Equations
- Partial Differential Equations - An Introduction With Ma Thematic A and Maple