# Partial Differential Equations Lecture Notes Erich Miersemann Department of Mathematics Leipzig University Version October, 2012 2 Contents

1 Introduction 9 1.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 1.2 Equations from variational problems . . . . . . . . . . . . . . 15 1.2.1 Ordinary differential equations . . . . . . . . . . . . . 15 1.2.2 Partial differential equations . . . . . . . . . . . . . . 16 1.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 2 Equations of first order 25 2.1 Linear equations . . . . . . . . . . . . . . . . . . . . . . . . . 25 2.2 Quasilinear equations . . . . . . . . . . . . . . . . . . . . . . 31 2.2.1 A linearization method . . . . . . . . . . . . . . . . . 32 2.2.2 Initial value problem of Cauchy . . . . . . . . . . . . . 33 2.3 Nonlinear equations in two variables . . . . . . . . . . . . . . 40 2.3.1 Initial value problem of Cauchy . . . . . . . . . . . . . 48 2.4 Nonlinear equations in Rn . . . . . . . . . . . . . . . . . . . . 51 2.5 Hamilton-Jacobi theory . . . . . . . . . . . . . . . . . . . . . 53 2.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 3 Classification 63 3.1 Linear equations of second order . . . . . . . . . . . . . . . . 63 3.1.1 Normal form in two variables . . . . . . . . . . . . . . 69 3.2 Quasilinear equations of second order . . . . . . . . . . . . . . 73 3.2.1 Quasilinear elliptic equations . . . . . . . . . . . . . . 73 3.3 Systems of first order . . . . . . . . . . . . . . . . . . . . . . . 74 3.3.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . 76 3.4 Systems of second order . . . . . . . . . . . . . . . . . . . . . 82 3.4.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . 83 3.5 Theorem of Cauchy-Kovalevskaya . . . . . . . . . . . . . . . . 84 3.5.1 Appendix: Real analytic functions . . . . . . . . . . . 90 3 4 CONTENTS 3.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 4 Hyperbolic equations 107 4.1 One-dimensional wave equation . . . . . . . . . . . . . . . . . 107 4.2 Higher dimensions . . . . . . . . . . . . . . . . . . . . . . . . 109 4.2.1 Case n=3 . . . . . . . . . . . . . . . . . . . . . . . . . 112 4.2.2 Case n = 2 . . . . . . . . . . . . . . . . . . . . . . . . 115 4.3 Inhomogeneous equation . . . . . . . . . . . . . . . . . . . . . 117 4.4 A method of Riemann . . . . . . . . . . . . . . . . . . . . . . 120 4.5 Initial-boundary value problems . . . . . . . . . . . . . . . . . 125 4.5.1 Oscillation of a string . . . . . . . . . . . . . . . . . . 125 4.5.2 Oscillation of a membrane . . . . . . . . . . . . . . . . 128 4.5.3 Inhomogeneous wave equations . . . . . . . . . . . . . 131 4.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136 5 Fourier transform 141 5.1 Definition, properties . . . . . . . . . . . . . . . . . . . . . . . 141 5.1.1 Pseudodifferential operators . . . . . . . . . . . . . . . 146 5.2 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149 6 Parabolic equations 151

6.1 Poisson’s formula . . . . . . . . . . . . . . . . . . . . . . . . . 152 6.2 Inhomogeneous heat equation . . . . . . . . . . . . . . . . . . 155 6.3 Maximum principle . . . . . . . . . . . . . . . . . . . . . . . . 156 6.4 Initial-boundary value problem . . . . . . . . . . . . . . . . . 162 6.4.1 Fourier’s method . . . . . . . . . . . . . . . . . . . . . 162 6.4.2 Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . 164 6.5 Black-Scholes equation . . . . . . . . . . . . . . . . . . . . . . 164 6.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170 7 Elliptic equations of second order 175 7.1 Fundamental solution . . . . . . . . . . . . . . . . . . . . . . 175 7.2 Representation formula . . . . . . . . . . . . . . . . . . . . . 177 7.2.1 Conclusions from the representation formula . . . . . 179 7.3 Boundary value problems . . . . . . . . . . . . . . . . . . . . 181 7.3.1 Dirichlet problem . . . . . . . . . . . . . . . . . . . . . 181 7.3.2 Neumann problem . . . . . . . . . . . . . . . . . . . . 182 7.3.3 Mixed boundary value problem . . . . . . . . . . . . . 183 7.4 Green’s function for 4 . . . . . . . . . . . . . . . . . . . . . . 183 7.4.1 Green’s function for a ball . . . . . . . . . . . . . . . . 186 CONTENTS 5 7.4.2 Green’s function and conformal mapping . . . . . . . 190 7.5 Inhomogeneous equation . . . . . . . . . . . . . . . . . . . . . 190 7.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195 6 CONTENTS Preface These lecture notes are intented as a straightforward introduction to partial differential equations which can serve as a textbook for undergraduate and beginning graduate students. For additional reading we recommend following books: W. I. Smirnov [21], I. G. Petrowski [17], P. R. Garabedian [8], W. A. Strauss [23], F. John [10], L. C. Evans [5] and R. Courant and D. Hilbert[4] and D. Gilbarg and N. S. Trudinger [9]. Some material of these lecture notes was taken from some of these books. 7 8 CONTENTS Chapter 1 Introduction Ordinary and partial differential equations occur in many applications. An ordinary differential equation is a special case of a partial differential equation but the behaviour of solutions is quite different in general. It is much more complicated in the case of partial differential equations caused by the fact that the functions for which we are looking at are functions of more than one independent variable. Equation F(x, y(x), y0(x), . . . , y(n)) = 0 is an ordinary differential equation of n-th order for the unknown function y(x), where F is given. An important problem for ordinary differential equations is the initial value problem y0(x) = f(x, y(x)) y(x0) = y0 , where f is a given real function of two variables x, y and x0, y0 are given real numbers. Picard-Lindel¨of Theorem. Suppose (i) f(x, y) is continuous in a rectangle Q = {(x, y) 2 R2 : |x − x0| < a, |y − y0| < b}. (ii) There is a constant K such that |f(x, y)| · K for all (x, y) 2 Q.

(ii) Lipschitz condition: There is a constant L such that |f(x, y2) − f(x, y1)| · L|y2 − y1| 9 10 CHAPTER 1. INTRODUCTION x y x y 0 0 Figure 1.1: Initial value problem for all (x, y1), (x, y2). Then there exists a unique solution y 2 C1(x0−®, x0+®) of the above initial value problem, where ® = min(b/K, a). The linear ordinary differential equation y(n) + an−1(x)y(n−1) + . . . a1(x)y0 + a0(x)y = 0, where aj are continuous functions, has exactly n linearly independent solutions. In contrast to this property the partial differential uxx+uyy = 0 in R2 has infinitely many linearly independent solutions in the linear space C2(R2). The ordinary differential equation of second order y00(x) = f(x, y(x), y0(x)) has in general a family of solutions with two free parameters. Thus, it is naturally to consider the associated initial value problem y00(x) = f(x, y(x), y0(x)) y(x0) = y0, y0(x0) = y1, where y0 and y1 are given, or to consider the boundary value problem y00(x) = f(x, y(x), y0(x)) y(x0) = y0, y(x1) = y1. Initial and boundary value problems play an important role also in the theory of partial differential equations. A partial differential equation for 1.1. EXAMPLES 11 y y0 xx y1 01x Figure 1.2: Boundary value problem the unknown function u(x, y) is for example F(x, y, u, ux, uy, uxx, uxy, uyy) = 0, where the function F is given. This equation is of second order. An equation is said to be of n-th order if the highest derivative which occurs is of order n. An equation is said to be linear if the unknown function and its derivatives are linear in F. For example, a(x, y)ux + b(x, y)uy + c(x, y)u = f(x, y), where the functions a, b, c and f are given, is a linear equation of first order. An equation is said to be quasilinear if it is linear in the highest derivatives. For example, a(x, y, u, ux, uy)uxx + b(x, y, u, ux, uy)uxy + c(x, y, u, ux, uy)uyy = 0 is a quasilinear equation of second order. 1.1 Examples 1. uy = 0, where u = u(x, y). All functions u = w(x) are solutions. 2. ux = uy, where u = u(x, y). A change of coordinates transforms this equation into an equation of the first example. Set » = x + y, ´ = x − y, then

i. thus v = w(») are solutions for arbitrary C1-functions w(»). Method of an integrating multiplier for an ordinary differential equation. then v´ = 0 and vice versa. where ©(x. y x W P P 0 1 Figure 1. Consider the ordinary differential equation M(x. »−´ 2 ¶ =: v(». 3. N = ©y. we have a large class of solutions of the original partial differential equation: u = w(x + y) with an arbitrary C1-function w.1. A large class of solutions is given by M = ©x. Consequently.u(x. A necessary and sufficient condition such that for given C1-functions M. N the integral Z P1 P0 M(x. y)dx + N(x. Two C1-functions u(x. N. 12 CHAPTER 1. 4. y) = u µ »+´ 2 . y)dx + N(x. ´). y) is an arbitrary C2-function. y) such that μMdx + μNdy is a total differential. 5. then v´ = 1 2 (ux − uy). This is a linear partial differential equation of first order for μ: Mμy − Nμx = μ(Nx −My). y)dy is independent of the curve which connects the points P0 with P1 in a simply connected domain ½ R2 is the partial differential equation (condition of integrability) My = Nx in . . INTRODUCTION Assume u 2 C1. y)dy = 0 1. y) are said to be functionally dependent if det µ ux uy vx vy ¶ = 0. Then we seek a C1-function μ(x. e. EXAMPLES 13 for given C1-functions M. that (μM)y = (μN)x is satisfied.3: Independence of the path This is one equation for two functions.. y) and v(x. If ux = uy. It follows from Gauss theorem that these are all C1-solutions of the above differential equation.

x 2 . 4v = 0. t) satisfies in × [0. t) = h(x. A large class of solutions is given by u = H(v(x. 0) = u0(x). u1 are given functions. 0) = u0(x). If h(x.. where h(x.t ) 2 lx Figure 1. The condition u(x. describes oscillations of membranes or of three dimensional domains. y). INTRODUCTION 8. t) = 0. The condition u(x. 0.which is a linear partial differential equation of first order for u if v is a given C1-function. Set f(z) = u(x. Let u(x. 7. 0). where z = x+iy and u. h is independent of t. Thus the initial position and the initial velocity are prescribed. If the string is finite one describes additionally boundary conditions. t) is given is a boundary condition for the heat equation. The Newton potential u= 1p x2 + y2 + z2 is a solution of the Laplace equation in R3 \ (0. It is known from the theory of functions of one complex variable that the real part u and the imaginary part v of a differentiable function f(z) are solutions of the Laplace equation 4u = 0. Then u(x. c is a positive constant. where 4u = ux1x1+ux2x2+ux3x3 and k is a positive constant. Cauchy-Riemann equations. 0) = u1(x). y)). t) tends to a function v(x) if t ! 1. where ½ R3 is a domain. e. 1.1) the heat equation ut = k4u. that is. Moreover. t) = 0 for all t ¸ 0. for example. t). 1. Here is a domain in R2. uy = −vx. In the one-dimensional case utt = c2uxx describes oscillations of a string. 6. u(l. Heat equation.4: Oscillating string utt = c24u. v satisfy the Cauchy-Riemann equations ux = vy. Wave equation. The wave equation y u(x. for example u(0. of uxx + uyy + uzz = 0. then one expects that the solution u(x. where u0(x) is given. where u0. If the function f(z) is differentiable with respect to the complex variable z then u. where 4u = uxx + uyy. t). EQUATIONS FROM VARIATIONAL PROBLEMS 15 Associated initial conditions are u(x.2 Equations from variational problems . t ¸ 0. x 2 @. t) = g(x). where u = u(x. 14 CHAPTER 1.2. where H is an arbitrary C1-function. 9.t 1 ) u(x. i. it turns out that v is the solution of the boundary value problem for the Laplace equation 4v = 0 in v = g(x) on @. ut(x. y)+iv(x. t) be the temperature of a point x 2 at time t. v are given C1()-functions. is an initial condition associated to the above heat equation.

then there is a ± > 0 such that (x0 − ±. Hints: For fixed Á 2 C2[a. v(b) = ub}. u(x). b). v(x). b] with Á(a) = Á(b) = 0 and real ². Since g(0) · g(²) it follows g0(0) = 0. 16 CHAPTER 1. x0 + ±] . Let u 2 V be a solution of (P). b) and Zb a h(x)Á(x) dx = 0 for all Á 2 C1 0 (a. u0(x)) = fu(x. set g(²) = E(u + ²Á).2 Partial differential equations The same procedure as above applied to the following multiple integral leads to a second-order quasilinear partial differential equation. b). Integration by parts in the formula for g0(0) and the following basic lemma in the calculus of variations imply Euler’s equation. Thus Á 2 C1 0 (a. b). |²| < ²0. b). Set Á(x) = ½¡ ±2 − |x − x0|2 ¢2 if x 2 (x0 − ±.2. ub 2 R V = {v 2 C2[a. x0 + ±) ½ (a.5: Admissible variations Basic lemma in the calculus of variations. b] : v(a) = ua. Exercise. where −1 < a < b < 1 and f is sufficiently regular. Set . v0(x)) dx and for given ua. which is a contradiction to the assumption of the lemma. Then h(x) ´ 0 on (a. 1. then d dx fu0(x. INTRODUCTION y y0 y1 abx Figure 1. Assume h(x0) > 0 for an x0 2 (a.1 Ordinary differential equations Set E(v) = Zb a f(x. b) \ [x0 − ±. One of the basic problems in the calculus of variation is (P) minv2V E(v). Proof. x0 + ±). x0 + ±) 0 if x 2 (a.2. u0(x)) in (a.A large class of ordinary and partial differential equations arise from variational problems. b) and h(x) ¸ h(x0)/2 on (x0 − ±. Proof. Euler equation. u(x). Let h 2 C(a. b) and Zb a h(x)Á(x) dx ¸ h(x0) 2 Z x0+± x0−± Á(x) dx > 0. 2 1.

. see Lions and Magenes [14].E(v) = Z F(x. . Euler equation. Proof. p. 1. x0+±) in the definition of Á must be replaced by a ball with center at x0 and radius ±. For a given function h. 155. 1. vxn). Let u 2 V be a solution of (P). v = v(x) : 7! R. set V = {v 2 C2() : v = h on @}. . The problems are equivalent. Exercise. EQUATIONS FROM VARIATIONAL PROBLEMS 17 where ½ Rn is a domain. INTRODUCTION Example: Minimal surface equation The non-parametric minimal surface problem in two dimensions is to find a minimizer u = u(x1. Vol. Example: Dirichlet integral In two dimensions the Dirichlet integral is given by D(v) = Z ¡ v2x + v2 y ¢ dxdy and the associated Euler equation is the Laplace equation 4u = 0 in . But these problems are not equivalent in general. Hint: Extend the above fundamental lemma of the calculus of variations to the case of multiple integrals. there is natural relationship between the boundary value problem 4u = 0 in . then Xn i=1 @ @xi Fuxi − Fu = 0 in . v. x2) of the problem min v2V Z q 1 + v2x 1 + v2x 2 dx. . The interval (x0−±. and rv = (vx1 . xn). . . provided the given boundary value function h is in the class H1/2(@). 18 CHAPTER 1. .2. Thus. u = h on @ and the variational problem min v2V D(v). Assume that the function F is sufficiently regular in its arguments. where for a given function h defined on the boundary of the domain V = {v 2 C1() : v = h on @}. see for an example Courant and Hilbert [4]. It can happen that the boundary value problem has a solution but the variational problem has no solution.rv) dx. x = (x1. . where h is a continuous function and the associated solution u of the boundary value problem has no finite Dirichlet integral. . defined on @.

u. This striking difference is caused by the strong nonlinearity of the minimal surface equation.S W Figure 1.1) In fact. see Simons [19]. see for example Gilbarg and Trudinger [9]. This is true also for higher dimensions 1. the additional assumption u 2 C2() is superfluous since it follows from regularity considerations for quasilinear elliptic equations of second order.7: Rotationally symmetric minimal surface Neumann type boundary value problems Set V = C1() and E(v) = Z F(x. that is u 2 V : E(u) · E(v) for all v 2 V. 1An experiment from Beutelspacher’s Mathematikum. Each linear function is a solution of the minimal surface equation (1. INTRODUCTION then Z ¡Xn i=1 Fuxi (x. v) ds. see Bombieri. v. Assume u is a minimizer of E(v) in V . then there exists also other solutions which define cones.ru)Áxi + Fu(x. See [18]. Figure 1. (1.2.rv) dx − Z @ g(x.6: Comparison surface Suppose that the minimizer satisfies the regularity assumption u 2 C2(). Let = R2. 62. An example is shown in Figure 1.ru)Á . u. In R2 linear functions are solutions but also many other functions in contrast to the minimal surface equation. where F and g are given sufficiently regular functions and ½ Rn is a bounded and sufficiently regular domain. EQUATIONS FROM VARIATIONAL PROBLEMS 19 n · 7. De Giorgi and Giusti [3]. More general minimal surfaces are described by using parametric representations.1). Wissenschaftsjahr 2008. pp. It was shown by Bernstein [2] that there are no other solutions of the minimal surface quation.71. The linearized minimal surface equation over u ´ 0 is the Laplace equation 4u = 0. Leipzig 20 CHAPTER 1. If n ¸ 8. for rotationally symmetric minimal surfaces. for example. then u is a solution of the minimal surface equation (Euler equation) in @ @x1 Ã p ux1 1 + |ru|2 ! + @ @x2 Ã p ux2 1 + |ru|2 ! = 0.

This follows after integration by parts from the basic lemma of the calculus of variations. u)Á ds = 0 for all Á 2 C1(). . Here · is a positive constant (capillarity constant) and ° is the (constant) boundary contact angle. . Then the related boundary value problem is div (Tu) = ·u in º · Tu = cos ° on @.¢ dx − Z @ gu(x. Example: Capillary equation Let ½ R2 and set E(v) = Z p 1 + |rv|2 dx + · 2 Z v2 dx − cos ° Z @ v ds. . then the associated boundary value problem is 4u = 0 in @u @º = h on @. ºn) is the exterior unit normal at the boundary @. x2).2. Example: Laplace equation Set E(v) = 1 2 Z |rv|2 dx − Z @ h(x)v ds. where we use the abbreviation . then u is a solution of the Neumann type boundary value problem Xn i=1 @ @xi Fuxi − Fu = 0 in Xn i=1 Fuxi ºi − gu = 0 on @. EQUATIONS FROM VARIATIONAL PROBLEMS 21 the capillary surface. Assume additionally u 2 C2(). where º = (º1.. defined by v = v(x1. e. i. . at the boundary. the angle between the container wall and 1.

Find nontrivial solutions u of uxy − uyx = 0 .rv) dx − Z @ g(v. x2). Hint: Real and imaginary part of holomorphic functions are solutions of the Laplace equation. Prove: In the linear space C2(R2) there are infinitely many linearly independent solutions of 4u = 0 in R2. v.8: Ascent of liquid in a wedge from [15]. Prove that a sufficiently regular minimizer in C1() of E(v) = Z F(x. is a solution of the boundary value problem Xn i=1 @ @xi Fuxi − Fu = 0 in Xn . A function u is said to be radially symmetric if u(x) = f(r). v) ds.1) and it is twice the mean curvature of the surface defined by z = u(x1. where r = ( Pn i x2i )1/2. div (Tu) is the left hand side of the minimal surface equation (1. 4.Tu = ru p 1 + |ru|2 . 22 CHAPTER 1. 3. in a vertical cylinder with cross section . Prove the basic lemma in the calculus of variations: Let ½ Rn be a domain and f 2 C() such that Z f(x)h(x) dx = 0 for all h 2 C2 0 (). Find all radially symmetric functions which satisfy the Laplace equation in Rn\{0} for n ¸ 2. Assume the gravity is directed downwards in the direction of the negative x3-axis. The above problem describes the ascent of a liquid. INTRODUCTION 1. Figure 1.3 Exercises 1.8 shows that liquid can rise along a vertical wedge which is a consequence of the strong nonlinearity of the underlying equations. see Finn [7]. Then f ´ 0 in . Hint: Show that a radially symmetric u satisfies 4u = r1−n ¡ rn−1f0 ¢ 0 by using ru(x) = f0(r) x r. 2. Write the minimal surface equation (1. 5. This photo was taken Figure 1. 6.1) as a quasilinear equation of second order. water for example. see an exercise.

. defined by z = u(x1. 9. Hint: Integrate the differential equation over .i=1 Fuxi ºi − gu = 0 on @. Let be bounded and assume u 2 C2() is a solution of div Tu = C in º · ru p 1 + |ru|2 = cos ° on @. Hint: From an exercise above it follows that C= 2 R cos °. Prove that º · Tu = cos ° on @. Hint: The angle between two surfaces is by definition the angle between the two associated normals at the intersection of the surfaces. EXERCISES 23 7. . where º = (º1. x2). Remark. Show that radially symmetric solutions u(x) = w(r). 1. It follows from a maximum principle of Concus and Finn [7] that a solution of the capillary equation over a disc must be radially symmetric. Prove that C = |@| || cos ° .3. º is the exterior normal at @. at the boundary of S. . which is here a cylinder. of the capillary boundary value problem are solutions of µ rw0 p1 + w02 ¶ 0 = ·rw in 0 < r < R w0 p1 + w02 = cos ° if r = R. 8. r = p x21 + x22. Show that div Tu is twice the mean curvature of the surface defined . INTRODUCTION 11. 10. and the surface S. 24 CHAPTER 1. Assume = BR(0) is a disc with radius R and the center at the origin. where C is a constant. . ºn) is the exterior unit normal at the boundary @. where ° is the angle between the container wall. Find all radially symmetric solutions of µ rw0 p1 + w02 ¶ 0 = Cr in 0 < r < R w0 p1 + w02 = cos ° if r = R.

(2. y. y). y). y)ux + a2(x. ¾ 2 R. y)q = 0 defines at each support (x.1. x. We assume L coincides with the envelope. z(¿ )) on S. y. y)) the normal N= 1p 1 + |ru|2 (−ux. 2. y)q(¸) = 0. z) support of the surface element. q) satisfying this equation. x. y) and z = u(x. q = uy(x. y). Chapter 2 Equations of first order For a given sufficiently regular function F the general equation of first order for the unknown function u(x) is F(x. u) = 0. q) is called surface element and the tuple (x.1) a1(x. y. y). The main tool for studying related problems is the theory of ordinary differential equations. For fixed (x. 0). Consider a curve x(¿ ) = (x(¿ ). c and f. y)(´ − y). y) is defined by a one parameter family of ascents p(¸) = p(¸. This is quite different for systems of partial differential of first order. y)uy = 0. z) a bundle of planes if we consider all (p. The general linear partial differential equation of first order can be written as Xn i=1 ai(x)uxi + c(x)u = f(x) for given functions ai. let Tx0 be the tangential plane at x0 = (x(¿0). p. y)p(¸) + a2(x. y).1 Linear equations Let us begin with the linear homogeneous equation a1(x. see Figure 2. At the intersection l(¾) we have (x − x0)p(¸0) + (y − y0)q(¸0) = (x − x0)p(¸0 + h) + (y − y0)q(¸0 + h). y. z(¿0)) of S and consider on Tx0 the line L : l(¾) = x0 + ¾x0(¿0). . q(¸) = q(¸. z). of the family of planes ¦(¸) at (x. 1) 25 26 CHAPTER 2. EQUATIONS OF FIRST ORDER and the tangential plane defined by ³ − z = ux(x. Thus. which is a line here. u(x. x0(¿0)p0(¸0) + y0(¿0)q0(¸0) = 0.ru) = 0 in 2 Rn. this family of planes ¦(¸) = ¦(¸. y.1) Assume there is a C1-solution z = u(x. y). which implies that the normal N(¸) on ¦(¸) is perpendicular on (a1. x2). On the other hand. y(¿0). The envelope of these planes is a line since a1(x. Set p = ux(x. y)(» − x) + uy(x. This function defines a surface S which has at P = (x. y)p + a2(x.by z = u(x1. u. u)uxi + c(x.−uy. x. The general quasilinear partial differential equation of first order is Xn i=1 ai(x. The tuple (x. The tangential plane is defined by the surface element. y(¿ ). differential equation (2. z. Assume that Tx0 = ¦(¸0) and consider two planes ¦(¸0) : z − z0 = (x − x0)p(¸0) + (y − y0)q(¸0) ¦(¸0 + h) : z − z0 = (x − x0)p(¸0 + h) + (y − y0)q(¸0 + h). a2.

The differential equations in (2. y(s)) ds. Assume Á 2 C1 is an integral. y(¿ )). y0(t) = a2(x. From this definition of the curves 28 CHAPTER 2. y) is said to be an integral of the characteristic system if Á(x(t). y0) the above initial value problem (2. y0)) defines the slope of these curves at (x(0). it follows Áxx0 + Áyy0 = 0 for |t| < t0. a2(x0. Since Á(x(t). y). for each characteristic curve. A function Á(x. a2 are in C1. Here we assume that x0(¿ ) 6= 0 and a1(x(¿ ).1. Proof. since ¿0 was an arbitrary parameter. If Á(x. y).1). where t(¿ ) = Z¿ ¿0 x0(s) a1(x(s). y0) = 0. y(¿0))p(¸) + a2(x(¿0). (2. Consider a1ux + a2uy = 0. EQUATIONS OF FIRST ORDER (x(t). y(¿0))q(¸) = 0 2. The constant depends on the characteristic curve considered. y0)a1(x0. Consequently (x0(¿ ).1. Examples 1. y) is a solution of equation (2. It follows x0(t) = a1(x. y) is a solution of (2. We denote x(¿ (t)) by x(t) again.1: Curve on a surface it follows a1p0(¸0) + a2q0(¸0) = 0. The system of characteristic equations is x0 = a1. 2 Remark. x(0) = x0. where a1. y(¿ )). y0) + Áy(x0. y(0) = y0. y(t)) = const. Consider for given (x0.2) From the theory of ordinary differential equations it follows (Theorem of Picard-Lindel¨of) that there is a unique solution in a neighbouhood of t = 0 provided the functions a1. y0 = a2. y). Now we consider the initial value problem x0(t) = a1(x. y(¿ )) (a1(x(¿ ). y(¿ )) 6= 0. t0 > 0 and sufficiently small. LINEAR EQUATIONS 27 y z x L S P( l 0) Figure 2.From the differential equation a1(x(¿0).2).2) are called characteristic equations or characteristic system and solutions of the associated initial value problem are called characteristic curves. where H(s) is a given C1-function. y(t)) = const. . y). Definition. y(0)). y0)a2(x0. y(t)) is follows that the field of directions (a1(x0.1) then also H(Á(x. a2 are constants. y0(¿ )) = x0(¿ ) a1(x(¿. y)). a2(x(¿ ). y0). then u = Á(x. Thus Áx(x0. y0(t) = a2(x. Definition. Then we introduce a new parameter t by the inverse of ¿ = ¿ (t). Proposition 2.

If follows x0x + yy0 = 0. 30 CHAPTER 2. (2. d dt (x2 + y2) = 0. y0 = bx. 4. 2.3 Figure 2. From these equations it follows that Á(x. or equivalently. Suppose that H0(s) > 0. Consequently. The associated characteristic system is x0 = y. then u defines right helicoids (in German: Wendelfl¨achen). see Figure 2. u = a2x − a1y is a solution of the differential equation. Italy) 3.1.2: Cylinder surfaces 2. x 6= 0. Here H(s) is an arbitrary C1-function. b are positive constants. If y/x = const. an integral is y/x. y = a2t + B. a2 < x2 + y2 < R2 (Museo Ideale Leonardo da Vinci. is also a solution. B are arbitrary constants. which define surfaces which have a hyperbola as the intersection with planes parallel to the (x. equivalently. y z x Figure 2. Consider the differential equation yux − xuy = 0. .2. where a. y)-plane. y)-plane. equation (2. The characteristic equations are x0 = x. where A. are solutions of the differential equation. Thus. It follows bxx0 − ayy0 = 0. Consider the differential equation xux + yuy = 0. d dt (bx2 − ay2) = 0. Since u is constant when a2x − a1y is constant. Solutions of the differential equation are u = H(bx2 − ay2).Thus the characteristic curves are parallel straight lines defined by x = a1t + A. y0 = −x. From an exercise it follows that u = H(a2x − a1y). The associated characteristic equations to ayux + bxuy = 0. QUASILINEAR EQUATIONS 31 or. for example. LINEAR EQUATIONS 29 where A.1. then u is constant. along each characteristic.3) defines cylinder surfaces which are generated by parallel straight lines which are parallel to the (x. and for a given C1-function the function u = H(x/y) is a solution of the differential equation. see Proposition 2. EQUATIONS OF FIRST ORDER and the characteristic curves are given by x = Aet. H0(s) 6= 0.3) where H(s) is an arbitrary C1-function.2. B are arbitrary constants. rotationally symmetric surfaces defined by u = H(x2 + y2). y0 = y. Thus. y) := a2x − a1y is constant along each characteristic curve.. which implies that x2 + y2 = const. see Figure 2. 2. are given by x0 = ay. where H0 6= 0.3: Right helicoid. y = Bet.

4). One arrives at characteristic equations x0 = a1. 32 CHAPTER 2. where the general case of nonlinear equations in two variables is considered. u. −1 < s1 < s2 < +1 be a regular curve in R3 and denote by C the orthogonal projection of ¡ . z0 = a3 from (2.4) The inhomogeneous linear equation a1(x. y0 = a2.2.4) and let (x(t). y. then Ã = w(x. y. z).5) where z := u.2. u(x. (2. it is constant along each fixed solution of (2. y. y) is a special case of (2. (2. (ii) The function z = u(x. y)). QUASILINEAR EQUATIONS 33 2. z) is a solution of (2. Assume Ãu 6= 0. then z(t) := u(x(t). One arrives at this system by the same arguments as in the two-dimensional case above.2. u)ux + a2(x. y. y) of (2.2. Exercise. We are looking for a function Ã(x. We consider the associated system of characteristic equations x0(t) = a1(x. y. is a solution of (2. y0(t) = a2(x. u)ux + a2(x. y)). y) be a solution of (2. Assume there is such a function Ã and let u be a solution of (2. is an integral.4) is defined implicitly by Ã(x. y(t)) be a solution of x0(t) = a1(x. y)uy = a3(x. then Ãx + Ãuux = 0. The additional equation z0 = a3 follows from z0(¿ ) = p(¸)x0(¿ ) + q(¸)y0(¿ ) = pa1 + qa2 = a3.5). z) = const.3.5). y = y0(s). From (2. y.2. y. implicitly defined through Ã(x. y. s1 · s · s2. y. u)uy = a3(x. y(t)) satisfies the third of the above characteristic equations. u). Ãy + Ãuuy = 0. y.4) we obtain a1(x. (i) Assume w 2 C1. EQUATIONS OF FIRST ORDER 2. y. Proof. (iii) Let z = u(x.4).2 Quasilinear equations Here we consider the equation a1(x. uy = − Ãy Ãu . provided that Ãz 6= 0. w = w(x. z)Ãy + a3(x.4) by the same arguments as in the case of homogeneous linear equations in two variables. y)ux + a2(x. z) z0(t) = a3(x. y. u(x. y..4). u). z)Ãx + a2(x. y. Let ¡ : x = x0(s). u) such that the solution u = u(x. Proposition 2.. 2.1 A linearization method We can transform the inhomogeneous equation (2. z). e. y. z) y0(t) = a2(x.2 Initial value problem of Cauchy Consider again the quasilinear equation (?) a1(x. u) = const. then ux = − Ãx Ãu . i. y). z = z0(s). y. y. y. z)Ãz = 0. u)uy = a3(x.4) into a homogeneous linear equation for an unknown function of three variables by the following trick. see also Section 2.

y0(s)) 6= 0. z) y0(t) = a2(x. y. y0(s)) = z0(s). a2. y)) is a solution of the initial problem of Cauchy. y) of (?) such that u(x0(s). z) with the initial conditions x(s. i. and the initial curve ¡ is noncharacteristic by assumption. y). y0(s)) − y00 (s)a1(x0(s). The curve ¡ is said to be noncharacteristic if x00(s)a2(x0(s). (i) Existence. y = y(s. t(x. a2 2 C1 in their arguments. Consider the following initial value problem for the system of characteristic equations to (?): x0(t) = a1(x. z = z(s. the initial data x0. t) and show that z(s(x.. y)-plane. The following calculation shows that u is also a solution of the differential equation (?). y) which contains the curve ¡. we consider the inverse functions s = s(x. s1 · s · s2. see Figure 2. then u satisfies the initial condition since u(x. t(x. t) be the solution. To show this. a1ux + a2uy = a1(zssx + zttx) + a2(zssy + ztty) = zs(a1sx + a2sy) + zt(a1tx + a2ty) = zs(sxxt + syyt) + zt(txxt + tyyt) = a3 . 0) = y0(s) z(s. y)). z0 2 C1[s1. y.4. Set u(x. s2] and ¡ is noncharacteristic. C : x = x0(s). Assume a1. t). Theorem 2. defines a surface. e. |t| < ´ for an ´ > 0. We will show that this set of strings sticked onto the curve ¡. z) z0(t) = a3(x. y) := z(s(x. we seek a surface S defined by z = u(x. y) of x = x(s.onto the (x. y)|t=0 = z(s. y = y0(s).4: Cauchy initial value problem Definition. Let x = x(s. i. t) ¯¯¯ t=0 = ¯¯¯¯ xs xt ys yt ¯¯¯¯ t=0 = x00(s)a2 − y00 (s)a1 6= 0. y = y(s. e. Initial value problem of Cauchy: Find a C1-solution u = u(x. y) @(s. y z x C G Figure 2. EQUATIONS OF FIRST ORDER Then there is a neighbourhood of C such that there exists exactly one solution u of the Cauchy initial value problem. The inverse functions s and t exist in a neighbourhood of t = 0 since det @(x. y0. 0) = x0(s) y(s. 0) = z0(s). t).. t = t(x. Proof. 0) = z0(s). y. y). y). 34 CHAPTER 2.1. t).

0) = z0(s) is given by x = t + x0(s). ² > 0 small. t). y = t + y0(s). y0(s) = 1. t0) = u(x0. 0) = z0(s0). t0). y0) = u(x0. According to its construction this curve is on the surface S defined by u = u(x. 36 CHAPTER 2. t). 0)) − z(s0. there is no uniqueness if the initial curve ¡ is a characteristic curve. t)) − z(s0. y0) in a neighbourhood of the curve (x0(s). z(t) := z(s0. t) be the solution of the above initial value problem for the characteristic differential equations with the initial data x(s0. In general. t) = 0. y0(t) = 1. s1 − ² · s · s2 + ². 0) = x0(s). 0) = y0(s). (ii) Uniqueness. 0) = x0(s0). y(t)) − z(t). then v(x0. t). y(s. 0). which shows that v(x0. y) and u(x0.2.6: Multiple solutions Examples 1. y0). y0) = z(s0. y) is a second solution. The inverse parameters are s0 = s(x0.. Set Ã(t) := v(x(t). y0). Thus. y(s0. 0) = 0 since v is a solution of the differential equation and satisfies the initial condition by assumption. 0) = y0(s0). y0) − z(s0. Ã(t) ´ 0. x = t + s. e. see Figure 2. y(s0. y(t) := y(s0.y’) Figure 2. t0 = t(x0. Consider a point (x0. y z x u Sv S Figure 2. y(s)).. y = t + 1. t0) = 0. e. EQUATIONS OF FIRST ORDER Set t = t0. z(s0.2. i. z = z0(s).6 which illustrates this case. 2 Remark. QUASILINEAR EQUATIONS 37 with the initial conditions x(s. x y (x 0 (s’). z0(s) is a given C1-function. QUASILINEAR EQUATIONS 35 since 0 = st = sxxt + syyt and 1 = tt = txxt + tyyt.y0 (s’)) (x’. u0(t) = 0 2. . Suppose that v(x. Consider the Cauchy initial value problem ux + uy = 0 with the initial data x0(s) = s. The solution of the associated system of characteristic equations x0(t) = 1. z = z0(s). y(s0. These initial data are noncharacteristic since y00 a1−x00a2 = −1. i. y0). then Ã0(t) = vxx0 + vyy0 − z0 = xxa1 + vya2 − a3 = 0 and Ã(0) = v(x(s0.2. y0) because of z(s0.5. v(x(s0. see an exercise and Figure 2. z(s.5: Uniqueness proof Let A : x(t) := x(s0.

y = t + c2 with constants cj . (i)The initial value problem in 1. The solution of this initial value problem is given by z(s. z(0) = 0. y)-plane are straight lines parallel to y = x. y0(t) = 1. x > 0. y0(s) = 0. In contrast to our previous assumptions. and u is the concentration of the chemical substance. x y x=y W W2 1 C C 1 2 Figure 2. t = y −1 and that u = z0(x−y +1) is the solution of the Cauchy initial value problem. separately. z0(0) = 0. in which the chemical reaction takes place. The variable x is the time and y is the hight of a tube.7: Domains to the chemical kinetics example 38 CHAPTER 2.7. 2. for example.7. y)-plane has a corner at the origin. y) = u0(y). t) = t + s. and the function u0(y) is given. see Figure 2. t) = t. y > 0. Consider for x ¸ 0. 0) = 0. It follows x = t + c1. these constants define the velocity of the reactions in consideration. The initial data are x0(s) = s. Thus the projection of the characteristic curves on the (x. We will solve the initial value problems in the domains 1 and 2. Thus z0(t) = (k0e−k1(t+s) + k2)(1 − z). t) = 1 − exp µ k0 k1 e−k1(s+t) − k2t − k0 k1 e−k1s ¶ . see Figure 2. z0(t) = ³ k0e−k1x + k2 ´ (1 − z). and u(0. Here the constants kj are positive. The projection C1 [ C2 of the initial curve onto the (x. A problem from kinetics in chemistry. It follows x = x(s. y ¸ 0 the problem ux + uy = ³ k0e−k1x + k2 ´ (1 − u) with initial data u(x. s ¸ 0. y = y(s.It follows s = x−y +1. EQUATIONS OF FIRST ORDER The associated system of characteristic equations is x0(t) = 1. the initial data are not in C1.

y) = 1 − (1 − u0(y − x)) exp µ k0 k1 e−k1x − k2x − k0 k1 ¶ is the solution in 2. It follows x = x(s. Consequently u2(x. t) = t + s. t) = 1 − (1 − u0(s)) exp µ k0 k1 e−k1t − k2t − k0 k1 ¶ . y = y(s. If time x tends to 1. y) = 1 − (1 − u0(0)) exp µ k0 k1 e−k1x − k2x − k0 k1 . then u1(x. y) = 1 − exp µ k0 k1 e−k1x − k2y − k0k1e−k1(x−y) ¶ is the solution of the Cauchy initial value problem in 1. y0(s) = s.2. The initial data are here x0(s) = 0. we get the limit lim x!1 u1(x. (ii) The initial value problem in 2. z0(0) = u0(s). y) = 1 − exp µ k0 k1 e−k1x − k2x − k0 k1 ¶ u2(x.. y) = 1 − e−k2y. Consequently u1(x. t) = t. 2. Thus z0(t) = (k0e−k1t + k2)(1 − z). s ¸ 0. z(0) = 0. QUASILINEAR EQUATIONS 39 The solution of this initial value problem is given by z(s. If x = y.

y0(s))) = u0(h(g)) = u0(s).9) p0(t) = −Fx − Fup (2. z(t). s1 < s < s2 be the initial data and let u = Á(x. and since u0 (h(Á(x0(s). y0(s)) = s − 1 and u0(Á + 1) = u0(x − y + 1) is the solution of the problem. Let x0(s). Figure 2. z0(s). A solution of the differential equation is Á(x.6). 2. q = uy(x. y)ux + a2(x. (2. then u1 < u2 if x = y.7) y0(t) = Fq (2. y0(s))x00(s) + Áy(x0(s).8: Gaspard Monge (Panth´eon.¶ .. i. y)uy = 0 is known. there is a jump of the concentration of the substrate along its burning front defined by x = y.6) and a solution (x(t). Together with (2. q) = 0. Remark. y0(s). 40 CHAPTER 2. The solution of the Cauchy initial problem is given by u0 (h(Á(x. If u0(0) > 0.3. q(t)) of the characteristic equations is called characteristic strip or Monge curve. see an exercise. The case that a solution of the equation is known Here we will see that we get immediately a solution of the Cauchy initial value problem if a solution of the homogeneous linear equation a1(x.8) z0(t) = pFp + qFq (2. this general nonlinear equation is more complicated. NONLINEAR EQUATIONS IN TWO VARIABLES 41 Definition. y). y0(s)) and let s = h(g) be the inverse function. p(t). y). Equations (2.6) where z = u(x. in particular from the assumption that there is a solution of (2. (2. p = ux(x. y) be a solution of the differential equation.10) q0(t) = −Fy − Fuq. y. y) = x − y.7)–(2. e. z. We assume that Áx(x0(s). y0(s))y00 (s) 6= 0 is satisfied. u0(s) is a given function. In contrast to the quasilinear case.11) 2. Paris) We will see. EQUATIONS OF FIRST ORDER This follows since in the problem considered a composition of a solution is a solution again. y(t). p. y))). Such a problem with discontinuous initial data is called Riemann problem. y0(s) = 1. that the strips defined by the characteristic . Set g(s) = Á(x0(s).6) we will consider the following system of ordinary equations which follow from considerations below as necessary conditions. See an exercise for another Riemann problem. y) and F 2 C2 is given such that F2 p + F2 q 6= 0.3 Nonlinear equations in two variables Here we consider equation F(x. Thus Á((x0(s).11) are said to be characteristic equations of equation (2. x0(t) = Fp (2. as in the quasilinear case. Example: Consider equation ux + uy = 0 with initial data x0(s) = s.

11. z0. z). Thus we assume that at each point of the surface S the associated tangent plane coincides with a plane from the family ¦(¸) at this point. we assume there exists.1. p. x.6). These (p(¸). e. The straight line l(¾) = x0 + ¾x0(¿0). z(¿ )) be a curve on the surface S which touches at each point its Monge cone. y0. y. q). z). and the z-coordinate of N is constant. Then the normal on the plane E directed downward is N= 1p 1 + |rv|2 (p.equations build the solution surface of the Cauchy initial value problem.6) defines a set of planes given by (x0.. q = vy(x0. More precisely. q(¸)) define a family ¦(¸) of planes. z) of solutions of (2. e. Let z = u(x. x.3. Let (x0. see Figure 2. 2. y) and which contains (x0. q(¸) = q(¸. y0). z0) be fixed. y0). Let E be a plane defined by z = v(x. (2.10. a one parameter C1-family p(¸) = p(¸. y(¿0). NONLINEAR EQUATIONS IN TWO VARIABLES 43 y z x Figure 2.9. q.12) that the normal N makes a constant angle with the z-axis. z0).10: Monge cones Let x(¿ ) = (x(¿ ). y0. y0.9: Monge cone in an example Thus the endpoints of the normals fixed at (x0. there is a cone which is the envelope of all these planes.6) defines such a Monge cone at each point in R3. vy = q at (x0.−1).. It follows from (2. see Figure 2.6). y0. z0) define a circle parallel to the (x. then equation (2. see Figure 2. z(¿0)). y0. In the case of quasilinear equations these set of planes is a bundle of planes which all contain a fixed straight line. y. y z x N P(l) (l) Figure 2. Consider the tangential plane Tx0 of the surface S at x0 = (x(¿0). It is defined through . see Section 2. y) be a solution of the general nonlinear differential equation (2.12) 42 CHAPTER 2. where p = vx(x0. is an apothem (in German: Mantellinie) of the cone by assumption and is contained in the tangential plane Tx0 as the tangent of a curve on the surface S. In the general case of this section the situation is more complicated. EQUATIONS OF FIRST ORDER where f is a given positive function. as in the above example. z0) and for which vx = p. y(¿ ). Consider the example p2 + q2 = f(x. i. −1 < ¾ < 1. y0). We assume that the general equation (2. planes given by z = v(x. Then we seek a surface S which touches at each point its Monge cone. i. y) which contain the point (x0. y)-plane. y.

It follows. (2. we have p0(¸0)l01 (¾) + q0(¸0)l02 (¾) = 0. y = l2(¾) in this limit position. (2. (2.13) that p0(¸0)x0(¿ ) + q0(¸0)y0(¿ ) = 0.16) Assume x0(¿0) 6= 0 and Fp 6= 0.13) The straight line l(¾) satisfies l3(¾) − z0 = (l1(¾) − x0)p(¸0) + (l2(¾) − y0)q(¸0). q(¸)) = 0 we see that Fpp0(¸) + Fqq0(¸) = 0. EQUATIONS OF FIRST ORDER y z x S Tx0 Figure 2.16) that z0(¿0) x0(¿0) =p+q Fq Fp .16) y0(¿0) x0(¿0) = Fq Fp . 44 CHAPTER 2. then we obtain from (2. since it is contained in the tangential plane Tx0 defined by the slope (p.13) we obtain z0(¿ ) = p(¸0)x0(¿ ) + q(¸0)y0(¿ ). On the intersection one has (x − x0)p(¸) + (y − y0)q(¸0) = (x − x0)p(¸0 + h) + (y − y0)q(¸0 + h). It follows l03(¾) = p(¸0)l01 (¾) + q(¸0)l02(¾).11: Monge cones along a curve on the surface Together with (2. x0(¿ ) µ p+q Fq Fp . y0(¿ ).15). z0. y0. since ¿0 was an arbitrary fixed parameter. Since x = l1(¾). Let h ! 0. x0(¿ ) = (x0(¿ ). q).15) 2. (2. z0(¿ )) = µ x0(¿ ). (2. p(¸).14) The above straight line l is the limit of the intersection line of two neighbouring planes which envelopes the Monge cone: z − z0 = (x − x0)p(¸0) + (y − y0)q(¸0) z − z0 = (x − x0)p(¸0 + h) + (y − y0)q(¸0 + h). NONLINEAR EQUATIONS IN TWO VARIABLES 45 From the differential equation F(x0. it follows (x − x0)p0(¸0) + (y − y0)q0(¸0) = 0.x0(¿0) = l0(¾). x0(¿ ) Fq Fp .3. and from (2.14) (2. and it follows from (2.

11) are satisfied in this quasilinear case automatically if there is a solution of the equation. i.3. e. The point is that the right hand sides are independent on p or q. let z = u(x. On the other hand. p(¸(¿ )). where F = F(x(¿ ). y) under consideration is in C2. (qx = py) Fx + Fzp + x0(t)px + y0(t)py = 0 Fx + Fzp + p0(t) = 0 46 CHAPTER 2. pFp + qFq).7)–(2. Proof. y(t).¶¶ = x0(¿ ) Fp (Fp. p. z). y) = p(x(t). 2.10) and (2. 1) such that z0(t) = p(t)x0(t) + q(t)y0(t). y. y(t)) on the curve x(t).11) is the following one. then z(t) := u(x(t). p(t). NONLINEAR EQUATIONS IN TWO VARIABLES 47 . z(t).1 that there exists a solution of the Cauchy initial value problem provided the initial data are noncharacteristic. y. In the previous quasilinear case F(x. Set a(¿ ) = x0(¿ ) Fp . pFp + qFq).3.11). we obtain the characteristic equations (2. This equation is called strip condition. we do not need the other remaining two characteristic equations. y(t). The geometric meaning of the first three characteristic differential equations (2. where t(¿ ) = Z¿ ¿0 a(s) ds. z).9).10) of the characteristic system is shown. The other two equations (2. That is. y0(t) = a2(x. F(x. y(¿ ). i.6) and from (2.6) with respect to y. z0(t) = a3(x. Here we denote x(¿ (t)) by x(t) again. Remark.−q. z)q − a3(x. y. see the above derivation of these equations. It follows from Theorem 2. y.11). y. Introducing the new parameter t by the inverse of ¿ = ¿ (t). From the differential equation (2. the tangential vector x0(¿ ) is proportional to (Fp.. e. q) is an integral. where p = ux and q = uy. y(t)) satisfies the strip condition. Proposition 2. EQUATIONS OF FIRST ORDER since p = p(x. z) the first three characteristic equations are the same: x0(t) = a1(x. q(¸(¿ ))). Differentiating the differential equation (2. z). z(t)) corresponds a tangential plane with the normal direction (−p. q(t)) = Fxx0 + Fyy0 + Fzz0 + Fpp0 + Fqq0 = FxFp + FyFq + pFzFp + qFzFq −Fpfx − FpFzp − FqFy − FqFzq = 0. d dt F(x(t).7)–(2. z)p + a2(x. z. we get finally equation (2.9) we get equations (2. p. Fq. y. Assume the surface z = u(x. q) = a1(x. z(¿ ). that is.10) and (2. Each point of the curve A : (x(t). the ”scales” defined by the normals fit together.7)–(2. y. then Fx + Fzp + Fppx + Fqpy = 0. z. y. Fq. y) defines a surface.. Thus equation (2. it is constant along each characteristic curve.

q) = 0 such that the surface S defined by z = u(x. p. then p0(t) = uxxFp + uxyFq q0(t) = uyxFp + uyyFq. F(x0(s). Let z = u(x. z(¿ ). y. z(t). z0. y0(s)) = z0(s). q0 2 C1 which satisfies the strip condition (2.11) with initial data (x0. y0. q0 = uy(x0. A strip (x(¿ ). p(¿ ). u(x. ux(x. ux(x(t). y)) y0(t) = Fq(x.17). We recall that the solution exists and is uniquely determined. p0. z. y. x0.4.7)–(2. z0 2 C2 and p0. Set p(t) = ux(x(t). uy(x. Consider the initial value problem x0(t) = Fp(x. y) of the Cauchy initial value problem in a neighbourhood of the initial curve (x0(s). see Figure 2. y). q0) Proof. q(¿ ))−y0(¿ )Fp(x(¿ ). q0(s)) = 0. ux(x. y. z(¿ ). u(x. q(¿ )).19) there exists exactly one solution z = u(x. then F = 0 along characteristic curves with the initial data (x0.18) and the differential equation (2. y0). y. u(x(t).2 Corollary. y(t)).1 Initial value problem of Cauchy Let x = x0(s). y) contains the above initial strip. is said to be noncharacteristic if x0(¿ )Fq(x(¿ ). we assume that the initial strip satisfies the nonlinear equation. that is. y). from the differential equation F(x. y) is the solution of the differential equation (2. fit together and define the surface for which we are looking at. Then the associated characteristic strip is in the surface S. z(¿ ). and z0(t) = uxx0(t) + uyy0(t) = uxFp + uyFq. y(t))) is a solution of the characteristic system. Thus z(t) = u(x(t). q(t)) is the solution of the characteristic system (2. p(¿ ). be a solution of the nonlinear equation (2. Set z(t) = u(x(t). ¿1 < ¿ < ¿2. y0(s)) = p0(s). z0. y). ux(x0(s). q(t) = uy(x(t). Moreover. y(¿ ).12. p(¿ ). z0. p = p0(s). y(t)) q(t) = uy(x(t). y(t)). Theorem 2. y(t). y(t)). y(¿ ). uy(x0(s). where (x(t). u(x. i. y0. Proposition 2.. y0). y(t)) p(t) = ux(x(t). y)) with the initial data x(0) = x0. u 2 C2. y). y). y(t)).3. q0) = 0. q0). p0. We will show that (x(t). uy(x. q = q0(s). uy(x. (2. For a given noncharacteristic initial strip (2. z = u(x.17) be a given initial strip such that the strip condition z00 (s) = p0(s)x00(s) + q0(s)y00 (s) (2. y = y0(s).7)–(2. z(t)) ½ S.2. y(0) = y0. Similar to the quasilinear case we will show that the set of strips defined by the characteristic system which are sticked at the initial strip. y). y(t). uy(x(t). z0(s). y(¿ ). defined by z = u(x. ux(x. y0. Consider the system (2. y0. z = z0(s). y)) = 0 it follows p0(t) = −Fx − Fup q0(t) = −Fy − Fuq. (2. y(t)). y0(s). y0(s)) = q0(s). y0(s). s1 < s < s2.18) is satisfied. y) of F(x. y). Proof. y0. EQUATIONS OF FIRST ORDER 2.11) with initial data . ) p0 = ux(x0. p0. 2 48 CHAPTER 2. y(t). p(t).19) Initial value problem of Cauchy: Find a C2-solution z = u(x. Definition. q(¿ )) 6= 0. Set z0 = u(x0. p0(s). then (x(t). z0(s)). y). y(t)). e.6).6) and u(x0(s). Assume F(x0. Finally.

Q(x.x(s. 0) = q0(s). One has h(s. t(x. y) = uy(x. y. z(s. since h(s. The first equation was shown above. t) is the surface defined by z = u(x. y). y) = ux(x. t). p(s. y) and Q(x. t) in a neigbourhood of t = 0. t). p(s. 0) = z0(s).3. To see this. where u is the solution of the Cauchy initial value 2.Q) = 0. From Proposition 2. Thus we have zs = pxs + qys zt = pxt + qyt zs = uxxs + uyys zt = uxyt + uyyt. y). 0) = p0(s). t). Hence h(s. y). EQUATIONS OF FIRST ORDER h(s. t). t) ¯¯¯ t=0 = x0Fq − y0Fq 6= 0. It follows P = ux and Q = uy.18. the second is a characteristic equation and the last two follow from z(s. t). t) and the characteristic equations we get ht(s. u. t) = u(x(s. 0) = z00(s) − p0(s)x00(s) − q0(s)y00 (s) = 0 since the initial strip satisfies the strip condition by assumption. This system implies (P − ux)xs + (Q − uy)ys = 0 (P − ux)xt + (Q − uy)yt = 0. P. t)). 0) = 0. It turns out that u(x. z(s. Since F(x(s. y) @(s. . t). q(s. t)) = 0.3 and Proposition 2.4 it follows F(x. We will show that P(x. t) = zs − pxs − qys. 0) = y0(s). t). we consider the function h(s. y). y). y)). Set P(x. y)).12: Construction of the solution problem. y)). t) ´ 0. y(s. q(s. t) = zst − ptxs − qtys − pxst − qyst = @ @s (zt − pxt − qyt) + psxt + qsyt − qtys − ptxs = (pxs + qys)Fz + Fxxs + Fyzs + Fpps + Fqqs. Thus the strip condition is also satisfied along strips transversally to the characteristic strips. t = t(x. y = y(s. Thaen the set of ”scales” fit together and define a surface like the scales of a fish. y). t(x. In the following we will find that for fixed s the function h satisfies a linear homogeneous ordininary differential equation of first order. it follows after differentiation of this equation with respect to s the differential equation ht = −Fzh. y) = p(s(x. t) = 0 in a neighbourhood of t = 0. We will show that the surface defined by x = x(s. Consequently. y) is the inverse of x = x(s. y) = z(s(x. y(s. NONLINEAR EQUATIONS IN TWO VARIABLES 49 y z x t=0 t>0 Figure 2. y(s. 0) = x0(s). y) = q(s(x. From the definition of h(s. t(x. This inverse exists since the initial strip is noncharacteristic by assumption: det @(x. where s = s(x. 50 CHAPTER 2. y(s. see Figure 2.

0).4. . y(t)) are the rays of light. . t(x. A differential equation which occurs in the geometrical optic is u2 x + u2y = f(x. z0(s). z. y. (2. y)) = q(s. t(x. . . sn−1). y) = const. Thus we have to extend the previous theory from R2 to Rn. Let x0(s) = (x01(s). 2. y)) = p(s. 52 CHAPTER 2. . p) = 0. t)) = p(s(x. e. p0n(s)) are given sufficiently regular functions. . y(s. The following system of 2n+1 ordinary differential equations is called characteristic system. x0(t) = rpF z0(t) = p · rpF p0(t) = −rxF − Fzp. t). In R3 the equation is u2 x + u2y + u2z = f(x. . t)) = q(s(x. The characteristic curves (x(t). y)) = z(s. p = ru. The level sets defined by u(x. NONLINEAR EQUATIONS IN RN 51 The uniqueness follows as in the proof of Theorem 2. t)) = z(s(x. . y(s. . y). n ¸ 3. y(s. 2 Example. EQUATIONS OF FIRST ORDER Assume z0(s) : D 7! R. t(x. . If n is a constant. Here s 2 D is a parameter from an (n − 1)-dimensional parameter domain D. . s = (s1.4 Nonlinear equations in Rn Here we consider the nonlinear differential equation F(x. . . 2. p0(s)) is called initial strip manifold and the condition @z0 @sl = nX−1 i=1 . xn).The initial conditions u(x(s. x = x0(s) defines in the three dimensional case a regular surface in R3. y). For example. t). y). t). y). 0)) = p0(s) uy(x(s. we assume rank @x0(s) @s = n − 1. . be a given regular (n-1)-dimensional C2-hypersurface in Rn. x0n(s)). 0)) = q0(s) are satisfied since u(x(s.1. t) uy(x(s. z). z = u(x) : ½ Rn 7! R. The (2n + 1)-vector (x0(s). y(s. y(s. p0(s) = (p01(s). 0). where the positive function f(x. i. then the rays of light are straight lines. . 0). t)..20) where x = (x1. y) is the index of refraction. y(s. . are called wave fronts. 0)) = z0(s) ux(x(s. t) ux(x(s.

. xn+1. .. Formally. xn. l = 1. . . (2. Sketch of proof. the equation is linear in pn+1 and does not depend on z explicitly. xn+1. . where Á is a function which is defined by a differential equation. . F(x0(s). u.... uxn) .5 Hamilton-Jacobi theory The nonlinear equation (2.5. . . . t(x1. p1. p = p(s. . . . .. z..20) such that the initial manifold is a subset of {(x. xn). . strip condition.. xn). xn.. .. .20). pn. . . xn.. Set xn+1 = u and seek u implicitely from Á(x1. @x01 @sn−1 @x02 @sn−1 · · · @x0n @sn−1 1 CCCA 6= 0. . Suppose the initial strip manifold is not characteristic and satisfies differential equation (2. . t) be the solution of the characteristic system and let s = s(x). .. t). . . t = t(x) 2. Remark. then 0 = F(x1. . .ru(x)) : x 2 }. . xn+1) = const. . . pn) = 0. The initial strip manifold is said to be noncharacteristic if det 0 BBB@ Fp1 Fp2 · · · Fpn @x01 @s1 @x02 @s1 · · · @x0n @s1 . where the argument of Fpj is the initial strip manifold.. The content of the Hamilton1-Jacobi2 theory is the theory of the special case F ´ pn+1 + H(x1. . t) which exists in a neighbourhood of t = 0. xn.. t).. sn−1(x1. xn. u(x).. . Then.. . . that is. z0(s)) such that there exists a unique solution of the Cauchy initial value problem. . .. xn)) is the solution of the problem. uxn) = 0 as an equation of type (2.. .21) i. . Then there is a neighbourhood of the initial manifold (x0(s).20) F(x1. . . . . . . . Let x = x(s. Initial value problem of Cauchy. ux1 . 2. z0(s). e.. p0(s)) = 0. . ...3.. HAMILTON-JACOBI THEORY 53 be the inverse of x = x(s. . . one can write equation (2.. n − 1. .p0i(s) @x0i @sl .. it turns out that z = u(x) := z(s1(x1. u. . . . Assume Áxn+1 6= 0.. . z = z(s. pn+1) = 0.21).20) of previous section in one more dimension is F(x1. Seek a solution z = u(x) of the differential equation (2. p1. . ux1 . . As in the two dimensional case we have under additional regularity assumptions Theorem 2.. . .

x. t. .= F(x1. . . p) is called Hamilton function.5. one obtains from solutions of the Hamilton-Jacobi equation also solutions of the canonical system of ordinary differential equations. . equation (2. . . p(t) be a solution of (2. . . . Definition. p = (p1. t) of the Hamilton-Jacobi equation. n. The function H(x. xn).21) are x0n+1(¿ ) = Fpn+1 = 1 x0k(¿ ) = Fpk = Hpk . . . Á1. p) (2. HAMILTON-JACOBI THEORY 55 canonical system. . On the other hand. . A solution Á(a. then it follows p0n+1(t) and z0(t) from the characteristic equations p0n+1(t) = −Ht z0(t) = p · rpH − H. . xn+1. .24) canonical system to H. Let x(t). . . . . . Áxn+1). .21) as ut + H(x. . (2. then Áxn+1 = H(x1. . . . . t. k = 1. . k = 1. . . . EQUATIONS OF FIRST ORDER The associated characteristic equations to (2.rxu) = 0 (2. Set t := xn+1. Áx1 . . . t. There is an interesting interplay between the Hamilton-Jacobi equation and the canonical system. . 1805–1851 54 CHAPTER 2.23) and (2. . xn. . . . According to the previous theory we can construct a solution of the Hamilton-Jacobi equation by using solutions of the 2. xn+1.21) Hamilton-Jacobi equation and the system (2. Áxn+1).22) and 2n of the characteristic equations are x0(t) = rpH(x. . William Rowan. then we can write partial differential equation (2. .− Áxn Áxn+1 ) = : G(x1. . p). Definition. . Suppose that GÁxn+1 6= 0. 1805–1865 2Jacobi.23).23) p0(t) = −rxH(x.24). 1Hamilton. an) is an n-tuple of real parameters.− Áx1 Áxn+1 . pn). . (2. . . xn. t. where a = (a1. n z0(¿ ) = nX+1 l=1 plFpl = Xn l=1 plHpl + pn+1 = Xn l=1 plHpl − H p0n+1(¿ ) = −Fxn+1 − Fzpn+1 = −Fxn+1 p0k(¿ ) = −Fxk − Fzpk = −Fxk . xn+1. Carl Gustav.24) Here is x = (x1. is called a complete integral of the Hamilton-Jacobi equation if det(Áxial)n .

b. 2 Example: Kepler problem The motion of a mass point in a central field takes place in a plane. . Then one obtains by solving of bi = Áai(a. t). 56 CHAPTER 2. e. t). Assume u = Á(a. . . x(a. be the solution of the above system. x. b are fixed. . and then by setting pk = Áxk (a. Now we differentiate pi(a. t. n are given real constants. t. t) with respect to t and Át + H(x. The solution exists since Á is a complete integral by assumption. x. then also u + const.rxÁ) = 0 with respect to ai. Thus x0k(t) = Hpk . c = const. it is @xk @t = x0k(t). . b. . . . and satisfies the system of ordinary .13. t) with respect to xl = xl(a. is a complete integral. If u is a solution of the Hamilton-Jacobi equation. t). . see Figure 2. . Remark. Differentiating Áai = bi with respect to t and the Hamilton-Jacobi equation Át+H(x. Á 2 C2 in its arguments.. Let xl(a. . Along a trajectory. We will show that x and p solves the canonical system. b. Set pk(a. . t) = Áxk (a.rxÁ) = 0 with respect to xi. x(a. . y)-plane. Proof. b. . . t) a 2n-parameter family of solutions of the canonical system. and obtain p0i(t) = Áxit + Xn k=1 Áxixkx0k(t) 0 = Áxit + Xn k=1 ÁxixkHpk + Hxi 0 = Áxit + Xn k=1 Áxixkx0k(t) + Hxi It follows finally that p0i(t) = −Hxi . . where a. k = 1. t). n. we obtain for i = 1. n. where bi i = 1.. t) + c. EQUATIONS OF FIRST ORDER Since Á is a complete integral it follows for k = 1. n @xk @t = Hpk . .i.4 (Jacobi). b. b. . t). say the (x. l = 1. .l=1 6= 0. . i. Theorem 2. n Átai + Xn k=1 Áxkai @xk @t =0 Átai + Xn k=1 ÁxkaiHpk = 0.

where U(x. q0(t) = −Hy. Set p = x0. for example. In the case that it is pushed one has to replace U by −U. for the related physics. Here we assume that k2 is a positive constant and that the mass point is attracted of the origin.U y ) q Figure 2.differential equations of second order x00(t) = Ux. HAMILTON-JACOBI THEORY 57 x y (x(t).26) and obtain from (2. 2. Vol 1. See Landau and Lifschitz [12]. Áµ = −¯ = const. y0(t) = Hq p0(t) = −Hx. The associated Hamilton-Jacobi equation is Át + 1 2 (Á2 x + Á2y )= k2 p x2 + y2 . y00(t) = Uy.13: Motion in a central field then x0(t) = Hp.y(t)) (Ux .25) that Á=± . (2. (2. q = y0 and H= 1 2 (p2 + q2) − U(x.25) by making the ansatz Át = −® = const. y) = k2 p x2 + y2 .25) Now we will seek a complete integral of (2. y). which is in polar coordinates (r. µ) Át + 1 2 (Á2r + 1 r2 Á2 µ)= k2 r .5.

Therefore we have a two parameter family of solutions Á = Á(®. Substitution ¿ = ½−1 yields µ − µ0 = −¯ Z 1/r 1/r0 d¿ p 2® + 2k2¿ − ¯2¿ 2 = −arcsin Ã ¯2 k2 1 r−1q 1 + 2®¯2 k4 ! + arcsin Ã ¯2 k2 1 . EQUATIONS OF FIRST ORDER of the Hamilton-Jacobi equation. r. Then t − t0 = − Zr r0 d½ q 2® + 2k2 ½ − ¯2 ½2 . From ansatz (2. is the r-coordinate depending on time t. t) 58 CHAPTER 2.Zr r0 s 2® + 2k2 ½− ¯2 ½2 d½ + c(t. The inverse function r = r(t). µ).26) it follows c(t. µ. ¯. and µ − µ0 = ¯ Zr r0 d½ ½2 q 2® + 2k2 ½ − ¯2 ½2 . According to the theorem of Jacobi set Á® = −t0. r(0) = r0. Á¯ = −µ0. This solution is a complete integral. see an exercise. µ) = −®t − ¯µ.

r0 − 1 q 1 + 2®¯2 k4 ! . Set µ1 = µ0 + arcsin Ã ¯2 k2 1 r0 − 1 q 1 + 2®¯2 k4 ! and p= ¯2 k2 , ²2 = r 1+ 2®¯2 k4 , then µ − µ1 = −arcsin µp r−1 ²2 ¶ . It follows r = r(µ) = p 1 − ²2 sin(µ − µ1) , which is the polar equation of conic sections. It defines an ellipse if 0 · ² < 1, a parabola if ² = 1 and a hyperbola if ² > 1, see Figure 2.14 for the case of an ellipse, where the origin of the coordinate system is one of the focal points of the ellipse. For another application of the Jacobi theorem see Courant and Hilbert [4], Vol. 2, pp. 94, where geodedics on an ellipsoid are studied. 2.6. EXERCISES 59 q1 p p 1+ 1−e e 2 2 p Figure 2.14: The case of an ellipse 2.6 Exercises 1. Suppose u : R2 7! R is a solution of a(x, y)ux + b(x, y)uy = 0.

Show that for arbitrary H 2 C1 also H(u) is a solution. 2. Find a solution u 6´ const. of ux + uy = 0 such that graph(u) := {(x, y, z) 2 R3 : z = u(x, y), (x, y) 2 R2} contains the straight line (0, 0, 1) + s(1, 1, 0), s 2 R. 3. Let Á(x, y) be a solution of a1(x, y)ux + a2(x, y)uy = 0 . Prove that level curves SC := {(x, y) : Á(x, y) = C = const.} are characteristic curves, provided that rÁ 6= 0 and (a1, a2) 6= (0, 0). 60 CHAPTER 2. EQUATIONS OF FIRST ORDER 4. Prove Proposition 2.2. 5. Find two different solutions of the initial value problem ux + uy = 1, where the initial data are x0(s) = s, y0(s) = s, z0(s) = s. Hint: (x0, y0) is a characteristic curve. 6. Solve the initial value problem xux + yuy = u with initial data x0(s) = s, y0(s) = 1, z0(s), where z0 is given. 7. Solve the initial value problem −xux + yuy = xu2, x0(s) = s, y0(s) = 1, z0(s) = e−s. 8. Solve the initial value problem uux + uy = 1, x0(s) = s, y0(s) = s, z0(s) = s/2 if 0 < s < 1. 9. Solve the initial value problem uux + uuy = 2, x0(s) = s, y0(s) = 1, z0(s) = 1 + s if 0 < s < 1. 10. Solve the initial value problem u2 x +u2y = 1+x with given initial data x0(s) = 0, y0(s) = s, u0(s) = 1, p0(s) = 1, q0(s) = 0, −1 < s < 1. 11. Find the solution ©(x, y) of (x − y)ux + 2yuy = 3x such that the surface defined by z = ©(x, y) contains the curve C : x0(s) = s, y0(s) = 1, z0(s) = 0, s 2 R. 2.6. EXERCISES 61 12. Solve the following initial problem of chemical kinetics. ux + uy = ³ k0e−k1x + k2 ´ (1 − u)2, x > 0, y > 0 with the initial data u(x, 0) = 0, u(0, y) = u0(y), where u0, 0 < u0 < 1, is given. 13. Solve the Riemann problem ux1 + ux2 = 0 u(x1, 0) = g(x1) in 1 = {(x1, x2) 2 R2 : x1 > x2} and in 2 = {(x1, x2) 2 R2 : x1 < x2}, where g(x1) = ½ ul : x1 < 0 ur : x1 > 0 with constants ul 6= ur. 14. Determine the opening angle of the Monge cone, i. e., the angle between

the axis and the apothem (in German: Mantellinie) of the cone, for equation u2 x + u2y = f(x, y, u), where f > 0. 15. Solve the initial value problem u2 x + u2y = 1, where x0(µ) = a cos µ, y0(µ) = a sin µ, z0(µ) = 1, p0(µ) = cos µ, q0(µ) = sin µ if 0 · µ < 2¼, a = const. > 0. 16. Show that the integral Á(®, ¯; µ, r, t), see the Kepler problem, is a complete integral. 17. a) Show that S = p® x + p1 − ® y + ¯ , ®, ¯ 2 R, 0 < ® < 1, is a complete integral of Sx − q 1 − S2 y = 0. b) Find the envelope of this family of solutions. 18. Determine the length of the half axis of the ellipse r= p 1 − "2 sin(µ − µ0) , 0 · " < 1. 62 CHAPTER 2. EQUATIONS OF FIRST ORDER 19. Find the Hamilton function H(x, p) of the Hamilton-Jacobi-Bellman differential equation if h = 0 and f = Ax + B®, where A, B are constant and real matrices, A : Rm 7! Rn, B is an orthogonal real n × n-Matrix and p 2 Rn is given. The set of admissible controls is given by U = {® 2 Rn : Xn i=1 ®2 i · 1} . Remark. The Hamilton-Jacobi-Bellman equation is formally the HamiltonJacobi equation ut + H(x,ru) = 0, where the Hamilton function is defined by H(x, p) := min ®2U (f(x, ®) · p + h(x, ®)) , f(x, ®) and h(x, ®) are given. See for example, Evans [5], Chapter 10. Chapter 3 Classification Different types of problems in physics, for example, correspond different types of partial differential equations. The methods how to solve these equations differ from type to type. The classification of differential equations follows from one single question: Can we calculate formally the solution if sufficiently many initial data are given? Consider the initial problem for an ordinary differential equation y0(x) = f(x, y(x)), y(x0) = y0. Then one can determine formally the solution, provided the function f(x, y) is sufficiently regular. The solution of the initial value problem is formally given by a power series. This formal solution is a solution of the problem if f(x, y) is real analytic according to a theorem of Cauchy. In the case of partial differential equations the related

(3. i = 1. where (aik)? = 1 2 (aik + aki). where x 2 Rn. see Figure 3. u : ½ Rn 7! R. . . Problem: Can we calculate all other derivatives of u on S by using differential equation (3. Concerning the classification the main part Xn i. xn). . n − 1. . since Xn i.D2u) = 0. . ¸n) @(x1.k=1 (aik)?uxixk . that aik = aki.1 Assume u and ru are given on S.Du. y(x). It is assumed that Â and ¸i are sufficiently regular.k=1 aik(x)uxixk + f(x. y0(x)) = 0 for a given function F. .1 Linear equations of second order The general nonlinear partial differential equation of second order is F(x. 63 64 CHAPTER 3. 3. i. LINEAR EQUATIONS OF SECOND ORDER 65 x Sx x3 . .. rÂ 6= 0. . . the differential equation is F(x. Even in the case of ordinary differential equations the situation is more complicated if y0 is implicitly defined. Consider a hypersurface S in Rn defined implicitly by Â(x) = 0.k=1 aikuxixk = Xn i. e.ru) = 0. .1) and the given data? We will find an answer if we map S onto a hyperplane S0 by a mapping ¸n = Â(x1. . Such a mapping ¸ = ¸(x) exists. . Suppose u 2 C2.1) The equation is linear if f= Xn i=1 bi(x)uxi + c(x)u + d(x). The function F is given and sufficiently regular with respect to its 2n + 1 + n2 arguments. .1. . without restriction of generality. xn) 6= 0 in ½ Rn. . then we can assume. see an exercise. . u. xn) ¸i = ¸i(x1. . u.theorem is the Theorem of Cauchy-Kowalevskaya. . CLASSIFICATION In this section we consider the case Xn i. . 3. .k=1 aik(x)uxixk plays the essential role. for functions ¸i such that det @(¸1. Du ´ ru and D2u stands for all second derivatives.

. n−1. ¸n−1. provided v is sufficiently regular. . .. ¸n). k = 1. . . ¸l+1. . . Thus. .k=1 ajk(x) @¸n @xj @¸n @xk . Thus the differential equation can be written as Xn j. we get uxj = v¸i @¸i @xj . . . ¸l. l = 1.2) uxjxk = v¸i¸l @¸i @xj @¸l @xk + v¸i @2¸i @xj@xk . Thus we know all second derivatives v¸i¸j on S0 with the only exception of v¸n¸n. it follows that v¸k¸l . . . are known on S0. i. . . see (3. ¸n−1. ¸l + h. ¸l+1. differential equation (3. . CLASSIFICATION 3 1 2 l l l S 0 Figure 3. . . . . (3. .1: Initial manifold S The above transform maps S onto a subset of the hyperplane defined by ¸n = 0. e. 0) − v¸k (¸1. . .2. we add terms with repeating indices. 66 CHAPTER 3. xn) and ¸ = (¸1. . . . n. . 0).2: Transformed flat manifold S0 We recall that. v¸k¸l(¸1. .2).1) in the new coordinates is given by ajk(x) @¸i @xj @¸l @xk v¸i¸l + terms known on S0 = 0. see Figure 3. . . . . . We will write the differential equation in these new coordinates. . . . . . Here we use Einstein’s convention. where x = (x1. 0) h as h ! 0. Since v¸k (¸1. are known. . .1 2 Figure 3. Since u(x) = u(x(¸)) =: v(¸) = v(¸(x)). ¸n−1. . ¸n−1. 0) is the limit of v¸k (¸1.

i.3) is satisfied for each Â with rÂ 6= 0 if the quadratic matrix (aij(x)) is positive or negative definite for each x 2 . 0) or of type (0. According to this definition there are other types aside from elliptic. which appears in the theory of transsonic flows. 2. In particular. provided u and ru are known on S. The Laplace equation in R3 is 4u = 0. then the surface defined by Â = 0 is called characteristic surface. y. where u = u(x. ¯ eigenvalues are negative and ° eigenvalues are zero (® + ¯ + ° = n).j=1 aij(x)ÂxiÂxj 6= 0 (3. 2. 68 CHAPTER 3. hyperbolic if it is of type (n − 1. This equation is elliptic if y > 0. . parabolic if it is of type (n−1. e. z) : Â(x. is a solution of the characteristic differential equation. An example is the Tricomi equation. CLASSIFICATION Remarks: 1. 1). parabolic if y = 0 and hyperbolic for y < 0. y. all derivatives of u are known on S. in the case that the matrix (aij) is poitive definite.. ¯. z) = 0}. all eigenvalues are different from zero and one eigenvalue has another sign than all the others. This follows since there is a ¸(x) > 0 such that. Xn i. Definition.1. is hyperbolic. °) at x 2 if ® eigenvalues of (aij)(x) are positive. where Â is an arbitrary sufficiently regular function such that rÂ 6= 0. LINEAR EQUATIONS OF SECOND ORDER 67 Definition. t).j=1 aij(x)³i³j ¸ ¸(x)|³|2 for all ³ 2 Rn. 0). e. Differential equation (3.v¸n¸n = terms known on S0. It follows that we can calculate v¸n¸n if Xn i. for example. The wave equation utt = uxx + uyy + uzz. Remark. z. Thus for each manifold S given by {(x.j=1 aij(x)ÂxiÂxj = 0 is called characteristic differential equation associated to the given differential equation (3. n−1. equation is called elliptic if it is of type (n. i.1). 3. e. This is a condition for the given equation and for the given surface S.1) is said to be of type (®. rÂ 6= 0. The condition (3. 0. n. Here and in the following we assume that the matrix (aij) is real and symmetric. 0) or of type (1. 0. The classification depends in general on x 2 . Examples: 1. i. which is equivalent to the property that all eigenvalues are different from zero and have the same sign.. parabolic or hyperbolic equations. Such a type describes oscillations of mechanical structures.3) on S. one eigenvalue is zero and all the others are different from zero and have the same sign. n − 1. The characterization of differential equation (3. 0). This equation is elliptic.. all eigenvalues are different from zero and have the same sign. The differential equation Xn i. yuxx + uyy = 0.1) follows from the signs of the eigenvalues of (aij(x)). If Â. 1) or of type (0. y. 1. where 4u := uxx + uyy + uzz.

b and c: μ1. .6). (3.. y. the transform to principle axis leads to a normal form from which the classification of the equation is obviously.. . e. n.2 = 1 a ³ b± p b2 − ac ´ . we can calculate μ := y0 from the (known) coefficients a.5) sometimes in such a way that we can solve the transformed equation explicitly. Set y = UT x and v(y) = u(Uy). Then there is a function y(x) defined in a neighbourhood of x0 such that Á(x.} and assume Áy 6= 0 at a point (x0.5) in ½ R2. 0 0 · · · ¸n 1 CCA ..1) are constant. Let U be the associated orthogonal matrix. The heat equation ut = uxx+uyy+uzz. y) = const.. y)uyy + terms of lower order = 0 (3. is parabolic.. y) be a solution of (3. see the characteristic equation (3. Let z = Á(x. then Xn i. In this case. (3. Consider the case that the (real) coefficients aij in equation (3. .. .1.. y)uxx + 2b(x. . 3.1. .. It follows y0(x) = − Áx Áy . AT = A.6).1 Normal form in two variables Consider the differential equation a(x.3. y)uxy + c(x..4) 3. y0) of the level set...6) We show that an appropriate coordinate transform will simplify equation (3. z. . zn).. y(x)) = const.. We recall that the matrix A = (aij) is symmetric. where u = u(x. i. which implies. the propagation of heat in a domain. It describes.. (3.. ay02 − 2by0 + c = 0. where zl.7) Then. The associated characteristic differential equation is aÂ2 x + 2bÂxÂy + cÂ2y = 0. 4. . l = 1. then UTAU = 0 BB@ ¸1 0 · · · 0 0 ¸2 · · · 0 .j=1 aijuxixj = Xn i=1 ¸ivyiyj . Consider the level sets {(x. LINEAR EQUATIONS OF SECOND ORDER 69 Here is U = (z1. y) : Á(x. t).. provided a 6= 0. is an orthonormal system of eigenvectors to the eigenvalues ¸l. for example.

3. (ii) ÁxÃy − ÁyÃx 6= 0. (ii): μ2 − μ1 = . This follows from an easy discussion of the eigenvalues of the matrix µ ab bc ¶ . (3.6) such that y01 ´ μ1 = − Áx Áy y02 ´ μ2 = − Ãx Ãy .5) is hyperbolic if ac − b2 < 0. y x (x. Equation (3. CLASSIFICATION Normal form of a hyperbolic equation Let Á and Ã are solutions of the characteristic equation (3.10).3: Level sets These level sets are characteristic curves of the partial differential equations (3. y)Ãy = 0. see an exercise. y) = ¯.9) and (3. (i): y02 − y01 ´ μ2 − μ1 = − 2 a p b2 − ac 6= 0.y)= (x. y)Áy = 0 (3.8). Thus Á and Ã are solutions of the linear homogeneous equations of first order Áx + μ1(x. where μ1 and μ2 are given by (3. 70 CHAPTER 3. respectively.1. y) are solutions such that rÁ 6= 0 and rÃ 6= 0. (3. Ã(x. (i) Curves from different families can not touch each other.y)= (x.10) Assume Á(x.8) These solutions are real if and only of ac − b2 · 0. LINEAR EQUATIONS OF SECOND ORDER 71 Proof. y) = ® and Ã(x. 3. see an exercise for the existence of such solutions. see Figure 3.y)= ja ja yb yb 1 2 1 2 Figure 3.y)= (x. Consider two families of level sets defined by Á(x. parabolic if ac − b2 = 0 and elliptic if ac−b2 > 0. y)..9) Ãx + μ2(x. see an exercise of the previous chapter. Lemma.

´ = Ã(x.o. The proof follows from a straightforward calculation. It follows v(». Proof. CLASSIFICATION The associated characteristic differential equation is Â2 x − Â2y = 0.5) into v»´ = lower order terms. y). The mapping » = Á(x. Thus the mapping » = x − y. Since ®° − ¯2 = (ac − b2)(ÁxÃy − ÁyÃx)2. ´) = Z» 0 f1(®) d® + g(´). y(». then v» = f1(») for an arbitrary C1 function f1(»). ´)). ´) = u(x(». the functions Á and Ã satisfy differential equations Áx + Áy = 0 Ãx − Ãy = 0. Solutions with rÁ 6= 0 and rÃ 6= 0 are Á = x − y. Since μ1 = −1 and μ2 = 1. 2 Proposition 3. ´) = 0. where ® : = aÁ2 x + 2bÁxÁy + cÁ2y ¯ : = aÁxÃx + b(ÁxÃy + ÁyÃx) + cÁyÃy ° : = aÃ2x + 2bÃxÃy + cÃ2 y. (3. ´ = x + y leads to the simple equation v»´(».11) where v(». it follows from the above lemma that the coefficient ¯ is different from zero.t. ..Áx Áy − Ãx Ãy . y) transforms equation (3.1. Thus auxx + 2buxy + cuyy = ®v»» + 2¯v»´ + °v´´ + l. Assume v 2 C2 is a solution. ux = v»Áx + v´Ãx uy = v»Áy + v´Ãy uxx = v»»Á2 x + 2v»´ÁxÃx + v´´Ã2x + lower order terms uxy = v»»ÁxÁy + v»´(ÁxÃy + ÁyÃx) + v´´ÃxÃy + lower order terms uyy = v»»Á2y + 2v»´ÁyÃy + v´´Ã2 y + lower order terms. 2 Example: Consider the differential equation uxx − uyy = 0. 72 CHAPTER 3. ´). The coefficients ® and ° are zero since Á and Ã are solutions of the characteristic equation. Ã = x + y.

2. The quasilinear equation (3. Assume equation (3.12) in a domain ½ Rn. solutions of the characteristic equation depend on the solution considered. z. u. 3. then 0 < ¸(x. p) be the minimum and ¤(x.13) 74 CHAPTER 3.2. p)³i³j · ¤(x. p) the maximum of the eigenvalues of (aij).j=1 aij(x. g the function (?) is a solution of the differential equation v»´ = 0. u. z. On the other hand. where u : 7! R. QUASILINEAR EQUATIONS OF SECOND ORDER 73 3. rÂ 6= 0. for arbitrary C2-functions f. z.j=1 aij(x. p)|³|2 · Xn i. z. Definition. p 2 Rn}. We assume that aij = aji. p) : x 2 .12) is called elliptic if the matrix (aij(x. Definition. 3. p) = ¡ 1 + |p|2¢ −1/2 µ . (3.where g is an arbitrary C2 function. p)|³|2 for all ³ 2 Rn. ´) = f(») + g(´). As in the previous section we can derive the characteristic equation Xn i. z.ru)uxixj + b(x. In contrast to linear equations. Consequently each C2-solution of the original equation uxx − uyy = 0 is given by u(x. z.12) is elliptic and let ¸(x. p)) is positive definite for each (x. y) = f(x − y) + g(x + y). z. p) 2 U. The coefficients aij are aij(x.2 Quasilinear equations of second order Here we consider the equation Xn i.12) is called uniformly elliptic if ¤/¸ is uniformly bounded in U. Equation (3. Thus each C2-solution of the differential equation can be written as (?) v(». g 2 C2.1 Quasilinear elliptic equations There is a large class of quasilinear equations such that the associated characteristic equation has no solution Â. where f. CLASSIFICATION The main part is the minimal surface operator (left hand side of the minimal surface equation). An important class of elliptic equations which are not uniformly elliptic (nonuniformly elliptic) is Xn i=1 @ @xi Ã p uxi 1 + |ru|2 ! + lower order terms = 0. g 2 C2. z. Set U = {(x.ru) = 0 (3. z. where f.j=1 aij(x.ru)ÂxiÂxj = 0. z 2 R. u.

.14) in new coordinates. ±ij denotes the Kronecker delta symbol.1 and write equation (3. u) = 0. Set v(¸) = u(x(¸))..13) is not uniformly elliptic... uxk = 0 B@ u1. We ask the same question as above: can we calculate all derivatives of u in a neighbourhood of a given hypersurface S in Rn defined by Â(x) = 0.xk 1 CA . It follows that ¸= 1 (1 + |p|2)3/2 .3 Systems of first order Consider the quasilinear system Xn k=1 Ak(x.. Thus equation (3.14) where Ak are m × m-matrices. sufficiently regular with respect to their arguments.¤= 1 (1 + |p|2)1/2 . rÂ 6= 0. provided u(x) is given on S? For an answer we map S onto a flat surface S0 by using the mapping ¸ = ¸(x) of Section 3. and u= 0 B@ u1 ..xk . Typical examples for nonuniformly elliptic equations are the minimal surface equation and the capillary equation. um 1 CA . um. bm 1 CA . 3. The behaviour of solutions of uniformly elliptic equations is similar to linear elliptic equations in contrast to the behaviour of solutions of nonuniformly elliptic equations. then .b= 0 B@ b1 . u)uuk + b(x.±ij − pipj 1 + |p|2 ¶ . (3.

u)Âxk ! 6= 0 on S. k = 1. . b. . ·). D depends on less than n parameters. ³) = 0 only if ³ = 0. Beltrami equations Wux − bvx − cvy = 0 (3. such that there exists m real roots ·k = ·k(x. . . u)Âxkv¸n = terms known on S0.14) is parabolic if there exists a regular linear mapping ³ = Q´ such that D is independent of ·. . u(x). Set C(x. (ii) System (3. . (i) The system (3. ´n−1). rÂ 6= 0.15) Wuy + avx + bvy = 0. e. u(x). ³) = det Ã Xn k=1 Ak(x.16) where W.Xn k=1 Ak(x. . u(x)) if there is a regular linear mapping ³ = Q´. . . u)³k ! for ³ 2 Rn. .. 76 CHAPTER 3. . . ´1. u)Âxk ! =0 is called characteristic equation associated to equation (3. . u(x). In the elliptic case all derivatives of the solution can be calculated from the given data and the given equation. a. i. c are given functions depending of (x. ·) = 0 for all (´1. ·) = C(x. provided that det Ã Xn k=1 Ak(x. The Beltrami system is a generalization of Cauchy-Riemann equations. (3. x. ´n−1). . ´n−1. of this characteristic equation is said to be characteristic surface. SYSTEMS OF FIRST ORDER 75 We can solve this system with respect to v¸n. ´n−1. . y).14) is hyperbolic at (x. where D(x. . . ´1. . Remark. u. .14) is elliptic if C(x. Definition. u. The function f(z) = u(x. . Equation det Ã Xn k=1 Ak(x. CLASSIFICATION 3.14) and a surface S: Â(x) = 0.1 Examples 1. defined by a solution Â. .m. W 6= 0 and the matrix µ ab bc ¶ is positive definite. is called a quasiconform . . u(x). of D(x. . ´1.Q´). where z = x + iy. Definition.3. ´n−1. 3.3. . (iii) System (3. where ´ = (´1. y). y) + iv(x.

y. H = (h1. C(x.18) 3. ² dielectricity constant.15). Thus. A2 = µ 0 −c Wb ¶ . for an application to partial differential equations. Set A1 = µ W −b 0a ¶ . c speed of light. Set p0 = Ât. ¸ specific conductivity.16) can be written as A1 µ ux vx ¶ + A2 µ uy vy ¶ = µ 0 0 ¶ .mapping. x = (x1. (3. (3. e3)T electric field strength. Here c. . e2. h3)T magnetic field strength. then the characteristic differential equation . Thus the Beltrami system is elliptic. ³) = ¯¯¯¯ W³1 −b³1 − c³2 W³2 a³1 + b³2 ¯¯¯¯ = W(a³2 1 + 2b³1³2 + c³2 2 ). i = 1. h2. pi = Âxi . which is different from zero if ³ 6= 0 according to the above assumptions. x2. 3. μ magnetic permeability. 2. ² and μ are positive constants. Then the system (3. .3. ei = ei(x. hi = hi(x. ¸. SYSTEMS OF FIRST ORDER 77 where E = (e1. . see for example [9].17) c rotx E = −μHt. t). t). x3). Maxwell equations The Maxwell equations in the isotropic case are c rotx H = ¸E + ²Et (3. Chapter 12.

t) = 0. here we assume that 0 is in he range of f : R 7! R. The evaluation of the above equation leads to q2(q + g2) = 0. see an exercise.is ¯¯¯¯¯¯¯¯¯¯¯¯ ²p0/c 0 0 0 p3 −p2 0 ²p0/c 0 −p3 0 p1 0 0 ²p0/c p2 −p1 0 0 −p3 p2 μp0/c 0 0 p3 0 −p1 0 μp0/c 0 −p2 p1 0 0 0 μp0/c ¯¯¯¯¯¯¯¯¯¯¯¯ = 0. 78 CHAPTER 3. t) ´ f(n · x − V t) = 0. CLASSIFICATION It follows immediately that Maxwell equations are a hyperbolic system. e. We obtain ¯¯¯¯¯¯ q + p21 p1p2 p1p3 p1p2 q + p22 p2p3 p1p3 p2p3 q + p23 ¯¯¯¯¯¯ = 0. (iv) multiply the 4th column with −p2 and the 5th column with p1 and add the sum to the 3th column. The first one are characteristic surfaces S(t). It follows that the planes S(t) . (iii) multiply the 4th column with p3 and the 6th column with −p1 and add the sum to the 2th column. Here is f(s) an arbitrary function with f0(s) 6= 0. Thus. S(t) is defined by n · x−V t = c. which satisfy Ât = 0. Â2t ³²μ c2 Â2t − |rxÂ|2 ´ = 0. (ii) multiply the 5th column with −p3 and the the 6th column with p2 and add the sum to the 1st column. i. where q := ²μ c2 p20 − g2 with g2 := p21 + p22 + p23. n is a unit vector and V = c/p²μ.. Functions defined by Â = f(n·x−V t) are solutions of this equation. where c is a fixed constant. The following manipulations simplifies this equation: (i) multiply the first three columns with μp0/c. These surfaces are called stationary waves. The second type of characteristic surfaces are defined by solutions of ²μ c2 Â2t = |rxÂ|2. 5th and 4th row. (v) expand the resulting determinant with respect to the elements of the 6th. defined by Â(x. The associated characteristic surfaces S(t) are defined by Â(x. There are two solutions of this characteristic equation.

19) ½t + ½ div v + v · r½ = 0. it is sufficient to assume that this assumption is satisfied at a fixed time t0 only. SYSTEMS OF FIRST ORDER 79 H are fields free of sources. CLASSIFICATION The characteristic differential equation is here ¯¯¯¯¯¯¯¯¯ . In fact. where p0(½) > 0 if ½ > 0. f2. Remark. v3) the vector of speed. e. t). Here is v = (v1. ½ density.20) where r ´ rx and div ´ divx. A special case of Maxwell equations are the telegraph equations. see an exercise. t). singularities of a solution of Maxwell equations are located at most on characteristic surfaces. 3. e. which follow from Maxwell equations if div E = 0 and div H = 0. E and 3. 80 CHAPTER 3.4: d0(t) is the speed of plane waves V is called speed of the plane wave S(t). f = (f1. Equations of gas dynamics Consider the following quasilinear equations of first order. v · rxv3))T . fi = fi(x. i. see Figure 3. i. x3). x2. it follows from Maxwell equations the uncoupled system 4xE = ²μ c2 Ett + ¸μ c2 Et 4xH = ²μ c2 Htt + ¸μ c2 Ht. vt + (v · rx) v + 1 ½rxp = f (Euler equations). Since rotx rotx A = gradx divx A −4xA for each C2-vector field A..4 x x2 n1 S(t) S(0) d(t) Figure 3. (v · rx)v ´ (v · rxv1. Then the above system of four equations is vt + (v · r)v + 1 ½ p0(½)r½ = f (3. x = (x1. According to the previous discussions. Assume the gas is compressible and that there is a function (state equation) p = p(½). p pressure.with normal n move with speed V in direction of n. vi = vi(x.3. f3) density of the external force. v2. (3. The second equation is ½t + v · rx½ + ½ divx v = 0 (conservation of mass). ½ = ½(x. v · rxv2. t).. these operators apply on the spatial variables only. p = (x. t).

One of the two normals on S(t) at a point of the surface S(t) is given by. As usually. where n is the normal (3. Consider a family S(t) of surfaces in R3 defined by Â(x.5: Definition of the speed of a surface 3. we get the characteristic differential equation µ dÂ dt ¶2 Ãµ dÂ dt ¶2 − p0(½)|rxÂ|2 ! = 0. (3. S(t 0) S(t 1) n Q Q1 0 Figure 3.21) This equation implies consequences for the speed of the characteristic surfaces as the following consideration shows.5. The limit P = lim t1!t0 |Q1 − Q0| t1 − t0 .dÂ dt 0 0 1 ½p0Âx1 0 dÂ dt 0 1 ½p0Âx2 0 0 dÂ dt 1 ½p0Âx3 ½Âx1 ½Âx2 ½Âx3 dÂ dt ¯¯¯¯¯¯¯¯¯ = 0.22 on S(t0) at Q0 and t0 < t1. see Figure 3. SYSTEMS OF FIRST ORDER 81 Definition. see an exercise. t1 − t0 small. we assume that rxÂ 6= 0.22) Let Q0 2 S(t0) and let Q1 2 S(t1) be a point on the line defined by Q0+sn.3. where x 2 R3 and c is a fixed constant. t) = c. Evaluating the determinant. where dÂ dt := Ât + (rxÂ) · v. n = rxÂ |rxÂ| . (3.

The proof follows from Â(Q0.24) where Akl are (m × m) matrices and u = (u1. t0) = 0 and Â(Q0 + dn. V := P −vn. um)T . p p0(½) is called speed of sound . u. (3. We assume Akl = Alk. t0 + 4t) = 0. 2 Set vn := v ·n which is the component of the velocity vector in direction n. see previous sections. the classification follows from the question whether or not we can calculate formally the solution from the differential equations. where d = |Q1 − Q0| and 4t = t1 − t0. (3.22) we get vn = 1 |rxÂ| v · rxÂ. Definition. The speed of the surface S(t) is P=− Ât |rxÂ| . From (3. Definition. the difference of the speed of the surface and the speed of liquid particles. . which is no restriction of generality provided u 2 C2 is satisfied.is called speed of the surface S(t). As in the previous sections. .21) that V 2|rxÂ|2 ¡ V 2|rxÂ|2 − p0(½)|rxÂ|2¢ = 0.4 Systems of second order Here we consider the system Xn k.23) Proof. CLASSIFICATION An interesting conclusion is that there are two relative speeds: V = 0 or V 2 = p0(½).2. Let the initial manifold S be given by Â(x) = 0 and assume that rÂ 6= 0.l=1 Akl(x.6: Definition of relative speed Using the above formulas for P and vn it follows V = P − vn = − Ât |rxÂ| − v · rxÂ |rxÂ| =− 1 |rxÂ| dÂ dt . 3. Then. if sufficiently many data are given on an initial manifold. The mapping x = x(¸). Proposition 3. we obtain from the characteristic equation (3. leads to Xn . n v S Figure 3.ru)uxkxl + lower order terms = 0. . 82 CHAPTER 3. is called relative speed. .

The system is called elliptic if det 0 @ Xn k.4.k. If there is a solution Â with rÂ 6= 0. The characteristic equation is here det 0 @ Xn k. u3(x. t)) displacement vector. t) = (u1(x. u2(x. 3.1 Examples 1. p = p(x. Definition. The problem is to find solutions v. where v(¸) = u(x(¸)). t) density of external force.l=1 AklÂxkÂxl 1 A = 0. ³ 6= 0. The characteristic equation is detC = 0. ¸. t). μ (positive) Lam´e constants. then it is possible that second derivatives are not continuous in a neighbourhood of S. where ½ is the (constant and positive) density of liquid. t) velocity vector of liquid particles.25) Here is. p of the above system. ° is the (constant and positive) viscosity of liquid. SYSTEMS OF SECOND ORDER 83 3. 2. ½ (constant) density. (3.4.l=1 Akl³k³l 1 A 6= 0 for all ³ 2 Rn. in the case of an elastic body in R3. The characteristic equation is ¡ . where the entries of the matrix C are given by cij = (¸ + μ)ÂxiÂxj + ±ij ¡ μ|rxÂ|2 − ½Â2t ¢ . Linear elasticity Consider the system ½ @2u @t2 = μ4xu + (¸ + μ)rx(divx u) + f. t) pressure. v = v(x. t). x 2 R3 or in R2. u(x. f(x.l=1 AklÂxkÂxlv¸n¸n = terms known on S. Navier-Stokes equations The Navier-Stokes system for a viscous incompressible liquid is vt + (v · rx)v = − 1 ½rxp + °4xv divx v = 0.

t) = const.26) and (3. . two questions arise: (i) Does the power series converge in a neighbourhood of 0 2 Rn? (ii) Is a convergent power series a solution of the initial value problem (3.14) can be written as uxn = nX−1 i=1 ai(x. and suppose that Â is not characteristic. . 84 CHAPTER 3. . Then. rÂ 6= 0. .3. Then. . u) (3. (3. From (3. see Section 3. xn−1. . We assume ai.1 (Cauchy-Kovalevskaya). We recall that P = −Ât/|rxÂ|. um)T .26) u(x1.3.. .14) of Section 3.5 Theorem of Cauchy-Kovalevskaya Consider the quasilinear system of first order (3. and P2 = r μ ½ . the system (3. the infinite sum P 1i=0 Á i(x) does not converge. . . b = (b1. defined by Â(x. u)uxi + b(x. xn−1) (3. as usually.(¸ + 2μ)|rxÂ|2 − ½Â2t ¢ ¡ μ|rxÂ|2 − ½Â2t ¢2 = 0. .26). .26). CLASSIFICATION 3.27)? Remark. Thus we have a formal power series of u(x) at x = 0: u(x) » X1 ®! D®u(0)x®. Theorem 3. There is a neighbourhood of 0 2 Rn such there is a real analytic solution of the initial value problem (3. . .27) it follows that we can calculate formally all derivatives D®u in a neigbourhood of the plane {x : xn = 0}. . . in contrast to the power series method of this section. . Here one is interested in a good approximation of an unknown solution of an equation by a finite sum PN i=0 Ái(x) of functions Ái. Quite different to this power series method is the method of asymptotic expansions. For notations and definitions used here and in the following see the appendix to this section.27) Here is u = (u1. in particular in a neighbourhood of 0 2 Rn. Assume an initial manifolds S is given by Â(x) = 0. . are possible. It follows that two different speeds P of characteristic surfaces S(t). bn)T and ai are (m×m)-matrices. See [15] for some asymptotic formulas in capillarity. 0) = f(x1. In general. b and f are in C1 with respect to their arguments. namely P1 = s ¸ + 2μ ½ .

(3. .29) . . xn−1.. . . .26). .27) is 0 BBB@ u1. . THEOREM OF CAUCHY-KOVALEVSKAYA 85 Proof..5. . um. The proof is taken from F.xn = nX−1 i=1 XN k=1 ai jk(z)uk. u1.. . xn−1. . u? xk = 0. . the new ai are ai(x1. 3. um)T . .. . The new u is u = (u1. . 0) = 0 and the extended system (3. . xn−1. . . bm 1 1 CCCA . . . .xi + bj(z). (3.. u?)T . . We introduce u−f as the new solution for which we are looking at and we add a new coordinate u? to the solution vector by setting u?(x) = xn. Thus we are led to an initial value problem of the type uj. . where the associated initial condition is u(x1. um. (3. John [10]. . u?) and the new b is b = (x1. . xn−1. um. u1.. . . . Then u? xn = 1. . n − 1. u?(x1. This solution is unique in the class of real analytic functions. . . . .N (3. .xn .xi u? xi 1 CCCA + 0 BBB@ b1 . um.28) uj(x) = 0 if xn = 0. .xn u? xn 1 CCCA = nX−1 i=1 µ ai 0 00 ¶ 0 BBB@ u1. 0) = 0. k = 1. j = 1.27).xi . . .

The point here is that ai jk and bj are independent of xn. see the appendix for the definition of real analytic functions. . . j = 1. Proof. (3. . defined by its formal power series are solutions of the initial value problem (3. Assume u 2 C1 in a neighbourhood of 0 2 Rn.28) is independent of xn. Inserting these functions into the left and into the right hand side of (3.. Then D®uj(0) = P® ³ d¯ai jk(0).where j = 1. and. .N and z = (x1.. uN).29) we can calculate formally all D¯uj . Moreover. d°bj(0) ´ .28). . they are real analytic functions. where |¯|.28) and (3. where c(j) ®= 1 ®! D®uj(0). u1.. . . . ± = (±1. xn−1.28) that DnD®uj(0) = P®(d¯ai jk(0). CLASSIFICATION right have the same coefficients caused by the calculation of the derivatives D®uj(0) from (3.30) Here is Dn = @/@xn and ®. Set d= µ @ @z1 . °.28) we obtain on the right and on the left hand side real analytic functions.29). (3. |°| · |®|.28) that the polynomials P® have integers as coefficients. it follows from (3. e. ± satisfy the inequalities |¯|.29) . From (3. This follows since compositions of real analytic functions are real analytic again. ¯. . It follows that uj(x). . The initial condition (3. which is essential in the proof. n. the last coordinates in the multi-indices ® = (®1. ±n) satisfy ±n · ®n since the right hand side of (3. . . . This fact simplifies the proof of the theorem. i. . . . . It follows from equation (3.. . . d°bj(0).28). ®n). We will show that these power series are (absolutely) convergent in a neighbourhood of 0 2 Rn. . . |°| · |®| and P® are polynomials in the indicated arguments with nonnegative integers as coefficients which are independent of ai and of b. . see Proposition A7 of the appendix to this section. . |±| · |®| + 1. @ @zN+n−1 ¶ Lemma A. The resulting power series on the left and on the 86 CHAPTER 3. Then we have formal power series for uj : uj(x) » X ® c(j) ® x®.D±uk(0)). ..

that is. ®n = 0. 1). . where D®uj(0) are defined in Lemma A. . Proof. F = (F1. (3. F 2 C1.5. where ® = (®1. . d°bj(0).N.31). ®n−1. ®n−1. . Then it follows from (3.30) that D®uj(0) = P®(d¯ai jk(0). The formal power series X ® 1 ®! D®uj(0)x®.32) Uj(x) = 0 if xn = 0. the proof is by induction with respect to ®n.xn = nX−1 i=1 XN k=1 Ai jk(z)Uk. . . Fm). The formal power series X ® 1 ®! D®uj(0)x®. . (3. .29) if ai jk << Ai jk and bj << Bj . 2 Definition.D±uk(0)). . Bj real analytic. fm). fi = fi(x). if f is majorized by F. It follows from Lemma A and from the assumption of Lemma B that |D®uj(0)| · P® ³ |d¯ai jk(0)|. . . Then. that is by zero. . We write f << F. ist called majorizing problem to (3. Lemma B. and f. (3.28).31) where ® = (®1. then we replace D±uk(0) in the right hand side of (3. Ai jk. |d°bj(0)| ´ · P® ³ |d¯Ai jk(0)|. and |D®uj(0)| · D®Uj(0). Fi = Fi(x). . . |d°Bj(0)| ´ ´ D®Uj(0). THEOREM OF CAUCHY-KOVALEVSKAYA 87 for all ®. . . We say f is majorized by F if |D®fk(0)| · D®Fk(0). . k = 1. .30) by (3.implies D®uj(0) = 0. The initial value problem Uj. . Definition. is convergent in a neighbourhood of 0 2 Rn if there exists a majorizing problem which has a real analytic solution U in x = 0. . The induction starts with ®n = 0. . . .33) j = 1. . . 0).m 3.xi + Bj(z) (3. Let f = (f1.

. . . .xn = Mr r − x1 − .N. t) = 1 Nn ³ r−s− p (r − s)2 − 2nMNrt ´ . j = 1. . . The solution of this problem is Uj(x1. Thus a majorizing problem is Uj.is convergent since X ® 1 ®! |D®uj(0)x®| · X ® 1 ®! D®Uj(0)|x®|. xn) = V (x1 + . bj(z) are real analytic in a neighbourhood of z = 0 it follows from Proposition A5 of the appendix to this section that there are positive constants M and r such that all these functions are majorized by Mr r − z1 − . . . CLASSIFICATION Proof. + xn−1. . . The right hand side is convergent in a neighbourhood of x 2 Rn by assumption. where V (s. . . . see an exercise. xn−1.N. − UN Ã 1+ nX−1 i=1 XN k=1 Uk. 2 Lemma C. t). + xn−1. . s = x1 + . 88 CHAPTER 3. . Thus the Cauchy-Kovalevskaya theorem is shown. Since ai ij(z). It follows that Uj(x) are also real analytic functions. − xn−1 − U1 − . . 2 . V (s. . . j = 1. 0) = 0. t) at (0. .xi ! Uj(x) = 0 if xn = 0. t = xn. which has the solution. There is a majorising problem which has a real analytic solution. is the solution of the Cauchy initial value problem Vt = Mr r − s − NV (1 + N(n − 1)Vs) . This function is real analytic in (s. . V (s. . − zN+n−1 . 0). . xn).

b = (q. q. f0(x) ¢ . Ã(0). uy.3. From the PicardLindel¨of theorem it follows that this analytic solution is unique even in the class of C1-functions. Á0(x). s. u. t)T . Á0(0). 3. r. q = uy. s = uxy. 0) = 0. y0) 2 R×Rn. Then the rewritten differential equation is the system Uy = AUx + b with the initial condition U(x. p. there is a real analytic solution in a neigbourhood of 0 2 R2 of the initial value problem 4u = 1 u(x. q. Partial differential equations of second order Consider the boundary value problem for two variables uyy = f(x. where x0 2 R and y0 2 Rn are given.5. 0. t. Á(x). There exists a real analytic solution in a neigbourhood of 0 2 R2 of the above initial value problem. 0. s). Á(0). uxx. 0) = Ã(x). 0. Ã0(x)). In particular. y) is real analytic in a neighbourhood of (x0. 0) = Á(x) uy(x. We assume that Á. p. r = uxx. Set p = ux. Ã(x). Then it follows from the above theorem that there exists an analytic solution y(x) of the initial value problem in a neighbourhood of x0. THEOREM OF CAUCHY-KOVALEVSKAYA 89 Examples: 1. r. 90 CHAPTER 3. Ã0(x). ux. Ã are analytic in a neighbourhood of x = 0 and that f is real analytic in a neighbourhood of (0. CLASSIFICATION The proof follows by writing the above problem as a system. 0) = 0 uy(x. Á00(x).1 Appendix: Real analytic functions Multi-index notation The following multi-index notation simplifies many presentations of formulas.5. . u. 0. Ã0(0)). uxy) u(x. y. y. Á0(x). This solution is unique in the class of analytic functions according to the theorem of Cauchy-Kovalevskaya. where f0(x) = f(x. y(x)) y(x0) = y0. 0) = ¡ Á(x). then t = f(x. Ã(x). fy + fuq + fqt)T and A= 0 BBBBBB@ 000000 001000 000000 000010 000001 0 0 fp 0 fr fs 1 CCCCCCA . t = uyy. Assume f(x. Á00(x). 0. 2. Set U = (u. Ordinary differential equations Consider the initial value problem y0(x) = f(x.

. . 0 . . .Dn) Du = (D1u. Using this multi-index notion. . . 3. . 1). . . . . . ° ¯+°=® ®! ¯!°! x¯y°. . 1 2 Rn. xn) and m ¸ 0 an integer. + xn)m = X |®|=m m! ®! . 1 . we have 1. . THEOREM OF CAUCHY-KOVALEVSKAYA 91 Define a partial order by ® ¸ ¯ if and only if ®i ¸ ¯i for all i. . .Let x = (x1. ¯. . . . . (x + y)® = X ¯. where 0. 2. · ®n! x® = x®1 1 x®2 2 · . y 2 Rn and ®. . . . . . . . . ° are multi-indices. Sometimes we use the notations 0 = (0. ®n) is called multi-index. then (x1 + . + ®n ®! = ®1!®2! · . . Let x = (x1. xn) and u : ½ Rn 7! R (or Rm for systems). . The n-tuple of nonnegative integers (including zero) ® = (®1. · x®n n (for a monom) Dk = @ @xk D = (D1. Taylor expansion for a polynomial f(x) of degree m: f(x) = X |®|·m 1 ®! (D®f(0)) x®. .Dnu) ´ ru ´ grad u D® = D®1 1 D®2 2 · . 1 = (1. . Set |®| = ®1 + . where x. 3.5. . . . . @x®n n . here is D®f(0) := (D®f(x)) |x=0. · D®n n´ @|®| @x®1 1 @x®2 2 . . 0). .

Directional derivative: dm dtmf(x + ty) = X |®|=m |®|! ®! (D®f(x + ty)) y®. ° ¯+°=® ®! ¯!°! (D¯f)(D°g). 92 CHAPTER 3. Leibniz’s rule: D®(fg) = X ¯. that Rm = (m + 1) X |®|=m+1 1 ®! µZ 1 . ®! · |®|! · n|®|®!. CLASSIFICATION 6. x. 5. 7. 0 < ± < 1. or Rm = 1 m! Z1 0 (1 − t)m©(m+1)(t) dt.x®. 8. It follows from 7. ± = ±(u. y). if x 2 N(y). where Rm = X |®|=m+1 1 ®! (D®u(y + ±(x − y))) x®.m. 4. D¯x® = ®! (® − ¯)! x®−¯ if ® ¸ ¯. y 2 Rn and t 2 R. then. D¯x® = 0 otherwise. u(x) = X |®|·m 1 ®! (D®u(y)) (x − y)® + Rm. Taylor’s theorem: Let u 2 Cm+1 in a neighbourhood N(y) of y. where x. where ©(t) = u(y + t(x − y)).

9.5. It follows that we can rearrange the order of summation. THEOREM OF CAUCHY-KOVALEVSKAYA 93 Power series Here we collect some definitions and results for power series in Rn. Let c® 2 R (or 2 Rm). Definition. we have 10. Let x 2 Rn. a convergent series is absolutely convergent. Using the above multi-index notation and keeping in mind that we can rearrange convergent series. the general linear partial differential equation of order m can be written as X |®|·m a®(x)D®u = f(x) in ½ Rn. Using multi-index notation. then X ® x® = Yn i=1 Ã 1X ®i=0 x®i i ! = . Remark. The series X ® c® ´ 1X m=0 0 @ X |®|=m c® 1 A is said to be convergent if X ® |c®| ´ 1X m=0 0 @ X |®|=m |c®| 1 A is convergent. According to the above definition.0 (1 − t)D®u(y + t(x − y)) dt ¶ (x − y)®. 3.

34) and assume this series is convergent for a z 2 Rn. Let x 2 Rn and |x1| + . 94 CHAPTER 3. + |xn| < 1. provided |xi| < 1 is satisfied for each i. . − xn = |¯|! (1 − x1 − . 13. . . . |xi| < 1 for all i. + xn)j = 1 1 − (x1 + . . . · (1 − xn) = 1 (1 − x)1 . . . by definition. and ¯ is a given multi-index. Assume x 2 Rn and |x1| + |x2| + . μ := X ® . . Then X ®¸¯ |®|! (® − ¯)! x®−¯ = D¯ 1 1 − x1 − . − xn)1+|¯| . then X ® |®|! ®! x® = 1X j=0 X |®|=j |®|! ®! x® = 1X j=0 (x1 + . Let x 2 Rn. Then X ®¸¯ ®! (® − ¯)! x®−¯ = D¯ 1 (1 − x)1 = ¯! (1 − x)1+¯ . . + |xn| < 1. Then. .1 (1 − x1)(1 − x2) · . 11. . . CLASSIFICATION 12. . Consider the power series X ® c®x® (3. + xn) .

where M= μ (1 − q)n . (ii) Assume f 2 C!(). CLASSIFICATION for all x 2 N(y). Then for each y 2 there exists a neighbourhood N(y) and positive constants M. . then f is said to be real analytic in y 2 if there are c® 2 R and if there is a neighbourhood N(y) of y such that f(x) = X ® c®(x − y)® 96 CHAPTER 3.34) is uniformly convergent for all x 2 Q(z). where Q(z) : |xi| · |zi| for all i. q). Hint: Use formula 12. Assume f is defined on a domain ½ Rn. Then f 2 C1(). Remark. . . r = (1 − q)min i |zi|. . . (i) In every compact subset D of Q(z) one has f 2 C1(D) and the formal differentiate series. p. See F. fm) is called real analytic if each coordinate is real analytic. (i) Let f 2 C!(). Thus the power series (3. is uniformly convergent on the closure of D and is equal to D¯f. Or an exercise. Set f(x) = X ® c®x®. 3. such that |xi| · q|zi| for all i.5. (ii) |D¯f(x)| · M|¯|!r−|¯| in D. A vector valued function f(x) = (f1(x).7. r such that f(x) = X . 64. For given x in a fixed compact subset D of Q(z) there is a q. 0 < q < 1. Proof.|c®||z®| < 1 and the series (3. . that is P ® D¯c®x®. where x is replaced by (q. The interior of Q(z) is not empty if and only if zi 6= 0 for all i.7: Definition of D 2 Q(z) Proposition A1. From the proposition above it follows c® = 1 ®! D®f(0). . see Figure 3. Definition. . according to a theorem of Weierstrass. and the series converges (absolutely) for each x 2 N(y). A function f is called real analytic in if it is real analytic for each y 2 . We will write f 2 C!() in the case that f is real analytic in the domain .34) defines a continuous function defined on Q(z). John [10]. THEOREM OF CAUCHY-KOVALEVSKAYA 95 z Q(z) D Figure 3. Proposition A2.

An open set 2 Rn is connected if and only if its path connected. that is. Proof.. Then f is said to be in the class CM.. Definition. Proposition A3. The proof follows from Proposition A1. 38. 65. We will prove the local version of the proposition. 1 6= . that a function f 2 C1 is real analytic if these estimates are satisfied. it follows 2 = . From the theory of one complex variable we know that a continuation of an analytic function is uniquely determined. Assume f 2 C!() and is connected. Set f = g − h and 1 = {x 2 : D®f(x) = 0 for all ®}. Proof. See F. e. John [10].5. 65-66. Proposition A4.r(y) if f 2 C1 in a neighbourhood of y and if |D¯f(y)| · M|¯|!r−|¯| for all ¯. p. The one-dimensional Taylor theorem says .® 1 ®! (D®f(y))(x − y)® for all x 2 N(y). r such that f 2 CM. with °(0) = x and °(1) = y. We say that is path connected if for any x. r positive constants. 0 · t · 1. Define a neighbourhood of y by Nd(y) = {x 2 : |x1 − y1| + . where d is a sufficiently small positive constant. we show it for each fixed y 2 . The set 1 is also open since f(x) = 0 in a neighbourhood of y 2 1. . see the next proposition. pp. Set ©(t) = f(y +t(x−y)). for example. On the other hand it follows.. pp. Let y 2 and M.r(y) for all y 2 S. Suppose g. John [10]. f 2 C!() if and only if f 2 C1() and for every compact subset S ½ there are positive constants M. Then f is uniquely determined if for one z 2 all D®f(z) are known. h 2 C!() and D®g(z) = D®h(z) for every ®. The same is true for real analytic functions. + |xn − yn| < d}. An open set 2 Rn is called connected if is not a union of two nonempty open sets with empty intersection. 2 It was shown in Proposition A2 that derivatives of a real analytic function satisfy estimates. The general version follows from Heine-Borel theorem. and |D¯f(x)| · M|¯|!r−|¯|. and the series converges (absolutely) for each x 2 N(y). Because of Proposition A3 it remains to show that the Taylor series X ® 1 ®! D®f(y)(x − y)® converges (absolutely) in a neighbourhood of y and that this series is equal to f(x). i. see [11]. See F. y 2 there is a continuous curve °(t) 2 . This follows from f(x) = X ® 1 ®! (D®f(y))(x − y)®. . 2 = {x 2 : D®f(x) 6= 0 for at least one ®}. The set 2 is open since D®f are continuous in . THEOREM OF CAUCHY-KOVALEVSKAYA 97 Since z 2 1. 3.

CLASSIFICATION From formula 7. then the Taylor series converges (absolutely) in Nd(y) and it is equal to f(x) since the remainder satisfies. . Choose d > 0 such that d < r. for directional derivatives it follows for x 2 Nd(y) that 1 j! dj dtj ©(t) = X |®|=j 1 ®! D®f(y + t(x − y))(x − y)®. |rj | = ¯¯¯¯ 1 (j − 1)! Z1 0 (1 − t)j−1©j(t) dt ¯¯¯¯ ·M . 98 CHAPTER 3. where rj = 1 (j − 1)! Z1 0 (1 − t)j−1©(j)(t) dt. + |xn − yn|)j ·M µ d r ¶j . we get for 0 · t · 1 ¯¯¯¯ 1 j! dj dtj ©(t) ¯¯¯¯ ·M X |®|=j |®|! ®! r−|®| |(x − y)®| = Mr−j (|x1 − y1| + .f(x) = ©(1) = Xj−1 k=0 1 k! ©(k)(0) + rj . From the assumption and the multinomial formula 3. see the above estimate. .

@/@um).r(0) and f(0) = 0 if and only if f << (© −M. Assume f(x) and g(u) are real analytic. |D®hk(0)| · P®(|±¯gl(0)|.µ d r ¶j . . . For each coordinate hk of h we have. 2 We recall that the notation f << F (f is majorized by F) was defined in the previous section. where ©(x) = Mr r − x1 − . P® are independent on g or f and ± := (@/@u1. (i) f = (f1. which characterizes real analytic functions. G(u) are in C1. .D°Fj(0)) = D®Hk(0). (ii) f 2 CM.½(0) . John [10]. See F. . then g(f(x)) is real analytic if f(x) is in the domain of definition of g. . .D°fj(0)). 68.r(y) and g 2 Cμ. D®©(0) = M|®|!r−|®|. Then f 2 CM. More precisely. fm) 2 CM. THEOREM OF CAUCHY-KOVALEVSKAYA 99 2 Remark. . . F(x). Assume that f maps a neighbourhood of y 2 Rn in Rm and g maps a neighbourhood of v = f(y) in Rm. Then g(f(x)) << G(F(x)). . H(x) = G(F(x)).© −M).½r/(mM+½)(y). that D®f << D®F. it follows that compositions of real analytic functions are real analytic functions again. 2 Using this result and Proposition A4. Proof. Thus. + xn) r − x1 − . Assume all functions f(x). G maps a neighbourhood of 0 2 Rm into Rp. . . D®hk(0) = P®(±¯gl(0). . . . The definition of f << F implies. the proposition follows from Proposition A4. Let f. . Set h(x) = g(f(x)). p. . 68. we have Proposition A6. 3. .½(v) implies h(x) := g(f(x)) 2 Cμ. where ©(x) = M(x1 + . − xn . F : Rn 7! Rm and g. f(0) = F(0) = 0. Proposition A7. The next proposition shows that compositions majorize if the involved functions majorize. .r(0) if and only if f << (©. Proof.5. − xn . according to the chain rule. 100 CHAPTER 3. See F. . f << F and g << G. where P® are polynomials with nonnegative integers as coefficients. . g(u). trivially. Proposition A5. we set h(y + x) := g(f(y + x)) ´ g(v + f(y + x) − f(x)) =: g¤(f¤(x)). p. John [10]. To show the inclusion. |D°fj(0)|) · P®(±¯Gl(0). . ©). . Once one has shown this inclusion. Proof. CLASSIFICATION where v = f(y) and g¤(u) : = g(v + u) 2 Cμ.

Show that in the hyperbolic case there exists a solution of Áx+μ1Áy = 0. Find the type of Lu := 2uxx + 2uxy + 2uyy = 0 and transform this equation into an equation with vanishing mixed derivatives by using the orthogonal mapping (transform to principal axis) x = Uy. . − xn ª(u) = μ½ ½ − x1 − x2 − . Show that the differential equation a(x. © : (x1. Hint: Transform to principal axis and stretching of axis lead to the wave equation. . . + xn) << μ½r ½r − (½ + mM)(x1 + . . . 5. where Â(x) = μ½ ½ − m(©(x) −M) = μ½(r − x1 − . 8. . xn) . . Hint: Consider an appropriate Cauchy initial value problem. y) = (1. Â(x)) ´ G(F). . − xn . + xn) = μ½r/(½ + mM) ½r/(½ + mM) − (x1 + . . . . . xn) 7! (¸1.r(0). Lu := xuxx + 2yuxy + 2xyuyy = 0. . Find all C2-solutions of uxx − 4uxy + uyy = 0. . See an exercise for the ”<<”-inequality. . see equation (3. such that rÁ 6= 0. 1/2). parabolic if ac − b2 = 0 and hyperbolic if ac − b2 < 0. From Proposition A5 it follows f¤(x) << (© −M. 4. 6. 3. Show equation (3. .6. Oscillations of a beam are described by wx − . . . . y)uxx + 2b(x. Show that for given x0 2 Rn there is in a neighbourhood of x0 a local diffeomorphism ¸ = ©(x). Let Â: Rn ! R in C1. U orthogonal. . . . . Determine the type of the following equation at (x. . . . . 2 3. 2. EXERCISES 101 3. ¸n). . where ©(x) = Mr r − x1 − x2 − . From Proposition A6 we get h(y + x) << (Â(x). 7.f¤(x) : = f(y + x) − f(y) 2 CM. y)uxy + c(x. y)uyy + lower order terms = 0 is elliptic if ac − b2 > 0.6 Exercises 1. rÂ 6= 0. such that ¸n = Â(x). − xn) ½r − (½ + mM)(x1 + . . . . ª) =: G. y are considered as fixed parameters.4).9). In the above formulas v.© −M) =: F g¤(u) << (ª. .

Transform equation x2uxx − y2uyy = 0. that is S(t) is independent of t. Prove formula (3.1 E ¾t = 0 ¾x − ½wt = 0. w. 17.6. Show that ¸= 1 (1 + |p|2)3/2 . Determine the type of uxx − xuyx + uyy + 3ux = 2x. into its normal form. . Assume a characteristic surface S(t) in R3 is defined by Â(x. w deflection of the beam and E. u = u(x. t) = const. 10. such that Ât = 0 and Âz 6= 0. Transform equation uxx + (1 − y2)uxy = 0. y). CLASSIFICATION where ¾ stresses. Hint. 14. where aij = ¡ 1 + |p|2¢ −1/2 µ ±ij − pipj 1 + |p|2 ¶ . c) Find non-zero solutions. z. Show that S(t) has a nonparametric representation z = u(x.A3). t) with ut = 0. where f is given. 15. ¾ occur only in one equation. into its normal form. div rot A = 0 for each C2-vector field A = (A1. Consider Maxwell equations and prove that div E = 0 and div H = 0 for all t if these equations are satisfied for a fixed time t0. u = u(x. Show that Maxwell equations are a hyperbolic system.23) for the speed of the surface S(t).22) for the normal on a surface.¤= 1 (1 + |p|2)1/2 . 19. 102 CHAPTER 3. that is. b) Transform the system into two uncoupled equations. y). y. y. 13.ru). Prove formula (3. 9. Transform the Tricomi-equation yuxx + uyy = 0. y. respectively. into its normal form. y). y). 12. are the minimum and maximum of eigenvalues of the matrix (aij). u. 11. EXERCISES 103 16. ½ are positive constants. Find nontrivial solutions (rÂ 6= 0) of the characteristic equation to x2uxx − uyy = f(x. a) Determine the type of the system. where u = u(x. where y < 0. u = u(x.A2. 3. 18.

½ density and a = p p0(½) the sound velocity.20. 25. f2) is given and ¸. 21. Write the Navier-Stokes system as a system of type (3.25) in the two dimensional case) is elliptic μ4u + (¸ + μ) grad(div u) + f = 0. y) = P(x. Let u be the solution of 4u = 1 u(x. Hint: 42u ´ 4(4u). 0). y) + O((x2 + y2)3/2) as (x. Find the polynomial P of degree 2 such that u(x. CLASSIFICATION where f is real analytic in a neighbourhood of (x0. Let y = y(x) be the solution of: y0(x) = f(x. Here are (u. Show formula 7. Discuss the type of the following system in stationary gas dynamics (isentrop flow) in R2. Hint: Multiply the differential equation with (r − s − NV ). Write 42u = −u as a system of first order. Solve the Cauchy initial value problem Vt = Mr r − s − NV (1 + N(n − 1)Vs) V (s. 23. 0) = 0. y(x)) y(x0) = y0. 28. 0) = uy(x. 24. The vector f = (f1. y0) 2 R2. u2). Show that the following system (linear elasticity. Hint: Induction with respect to m. stationary case of (3. Find the polynomial P of degree 2 such that y(x) = P(x − x0) + O(|x − x0|3) as x ! x0. (directional derivative).24). where u = (u1. v) velocity vector. 0) = 0. 22. 104 CHAPTER 3. ½uux + ½vuy + a2½x = 0 ½uvx + ½vvy + a2½y = 0 ½(ux + vy) + u½x + v½y = 0. μ are positive constants. y) ! (0. 27. 26. Write the minimal surface equation @ @x 0 @q ux 1 + u2 x + u2y 1 A+ @ @y 0 @q uy 1 + u2 x + u2y 1 .

t) = f(x + ct) + g(x − ct).1) where u = u(x. x0 + ²]-solution from the Picard-Lindel¨of theorem. Show that a real analytic solution in a neighbourhood of x0 of the problem y0(x) = f(x. HYPERBOLIC EQUATIONS Theorem 4. + xn) << μ½r ½r − (½ + mM)(x1 + . y0). t) = ®(x + ct) + ®(x − ct) 2 + 1 2c . all C2-solutions of the wave equation are u(x. . ² > 0 sufficiently small. 106 CHAPTER 3. . 0) = ®(x) ut(x. Hint: v1 := ux/ q 1 + u2 x + u2y . v2 := uy/ q 1 + u2 x + u2y .1) are given. Hint: Leibniz’s rule. .6. − xn) ½r − (½ + mM)(x1 + . 107 108 CHAPTER 4. (4. 0) = ¯(x). CLASSIFICATION Chapter 4 Hyperbolic equations Here we consider hyperbolic equations of second order. mainly wave equations.2) with arbitrary C2-functions f and g The Cauchy initial value problem for the wave equation is to find a C2solution of 1 c2 utt − uxx = 0 u(x. EXERCISES 105 30. . Show (see the proof of Proposition A7) μ½(r − x1 − . and this solution is given by d’Alembert’s1 formula u(x. y) y(x0) = y0 exists and is equal to the unique C1[x0 − ². According to previous considerations.1 One-dimensional wave equation The one-dimensional wave equation is given by 1 c2 utt − uxx = 0. ¯ 2 C2(−1. Let f : R × Rm ! Rm be real analytic in (x0.1. + xn) . 3. 4. . (4. . 29. t) is a scalar function of two variables and c is a positive constant. where ®. There exists a unique C2(R × R)-solution of the Cauchy initial value problem.A=0 as a system of first order.

Then f(x) = ®(x) 2 + 1 2c Zx 0 ¯(s) ds + C1 g(x) = ®(x) 2− 1 2c Zx 0 ¯(s) ds + C2. 1d’Alembert. t) defined by the right hand side of (4. (4. which implies. On the other hand. HIGHER DIMENSIONS 109 x+ct=const. then it follows from (4. t)? From the d’Alembert formula it follows that this domain is a cone. Let P be a point on the x-axis. C2 satisfy C1 + C2 = f(x) + g(x) − ®(x) = 0. Thus each C2-solution of the Cauchy initial value problem is given by d’Alembert’s formula.Z x+ct x−ct ¯(s) ds. t) of the Cauchy initial value problem.5) From (4. The constants C1. t x −ct x x +ct x (x .5).1. 0) = cf0(x) − cg0(x) = ¯(x). Jean Babtiste le Rond. 0) = f(x) + g(x) = ®(x) (4. Then we ask which points (x.4) we obtain f0(x) + g0(x) = ®0(x). Assume there is a solution u(x.4). (4.2.2. x + ct] is called domain of dependence. that f0(x) = ®0(x) + ¯(x)/c 2 g0(x) = ®0(x) − ¯(x)/c 2 .4) ut(x. The interval [x − ct.3) Proof. x+ct] and on the values of ¯ on this interval only. t) need values of ® or ¯ at P in order to calculate u(x.t ) 0 0 00000 Figure 4. see Figure 4.2) that u(x. t) of the initial value problem depends on the values of ® at the endpoints of the interval [x−ct. see Figure 4. see (4. . The solution u(x.3) is a solution of the initial value problem. 2 Corollaries. 1. This set is called domain of influence. the function u(x. 1717-1783 4.1: Interval of dependence 2. together with (4.

2: Domain of influence 4. Figure 4. t) for which we are looking at by u(x. HYPERBOLIC EQUATIONS and consider the initial value problem 2u = 0 in Rn × R (4. The next step is to derive a partial differential equation for the spherical mean. !nrn−1 is the area of a sphere with radius r. t) = lim r!0 M(r. (4. we have M(r.7) ut(x. t) = 1 !n .10) Using the initial data. 0) = g(x). where x and r are fixed.11) Mt(r. (4. .12) which are the spherical means of f and g.8) where f and g are given C2(R2)-functions. 4 ´ 4x = @2/@x21 + .6) u(x. (4.9) where !n = (2¼)n/2/¡(n/2) is the area of the n-dimensional sphere. M(r. 110 CHAPTER 4.t x P x−ct=const. By using spherical means and the above d’Alembert formula we will derive a formula for the solution of this initial value problem.9) of the spherical mean we obtain. t) of the initial value problem by M(r. 0) = 1 !nrn−1 Z @Br(x) g(y) dSy =: G(r). From the mean value theorem of the integral calculus we obtain the function u(x. t). . after the mapping » = (y − x)/r. (4. t) dSy. From definition (4. 0) = 1 !nrn−1 Z @Br(x) f(y) dSy =: F(r) (4. t) = 1 !nrn−1 Z @Br(x) u(y.2 Higher dimensions Set 2u = utt − c24u. Method of spherical means Define the spherical mean for a C2-solution u(x. + @2/@x2 n. 0) = f(x) (4.

t) dSydc.Z @B1(0) u(x + r». Integration by parts yields 1 !nrn−1 Z Br(x) Xn i=1 uyiyi(y.2. t) dSy = rn−1 c2 @2 . t)»i dS» = 1 !nrn−1 Z @Br(x) Xn i=1 uyi(y. t) dS». Consequently (rn−1Mr)r = 1 c2!n Z @Br(x) utt(y. t) dy since » ´ (y−x)/r is the exterior normal at @Br(x). t) dy = 1 c2!n Zr 0 Z @Bc(x) utt(y. 4. Assume u is a solution of the wave equation. HIGHER DIMENSIONS 111 It follows Mr(r. t) = 1 !n Z @B1(0) Xn i=1 uyi(x + r». The previous equation follows by using spherical coordinates. t)»i dSy. then rn−1Mr = 1 c2!n Z Br(x) utt(y.

x + ct] is a subset of (0. lim r!+0 G(r) = g(x). (4. which can be written as Mrr + n−1 r Mr = c−2Mtt. Lemma. Proof.1).1) such that rF0 and rG0 are C2(R)-functions as follows. . 0) = rF(r) (rM)t(r. rG0(r) 2 C2(R2). From definition of F(r) and G(r).2. We can extend F and G to F0 and G0 which are defined on (−1. t) dSy ! = rn−1 c2 Mtt.@t2 Ã 1 !nrn−1 Z @Br(x) u(y.13) This equation (4. (4.13) is called Euler-Poisson-Darboux equation. (4. respectively. Thus rM is the solution of the one-dimensional wave equation with initial data (rM)(r. t) = (r + ct)F(r + ct) + (r − ct)F(r − ct) 2r + 1 2cr Z r+ct r−ct »G(») d». 0) = rG(r). 112 CHAPTER 4. rF0(r). Thus we arrive at the differential equation (rn−1Mr)r = c−2rn−1Mtt.1 Case n=3 The Euler-Poisson-Darboux equation in this case is (rM)rr = c−2(rM)tt. it follows from the mean value theorem lim r!+0 F(r) = f(x). r > 0.14) From the d’Alembert formula we get formally M(r. HYPERBOLIC EQUATIONS 4. Set F0(r) = 8< : F(r) : r > 0 f(x) : r = 0 F(−r) : r < 0 The function G0(r) is given by the same definition where F and f are replaced by G and g.15) The right hand side of the previous formula is well defined if the domain of dependence [x − ct.

rF0(r) and rG0(r) are in C2(R). We have.Thus rF0(r) and rG0(r) are C(R)-functions. t) = (r + ct)F0(r + ct) + (r − ct)F0(r − ct) 2r + . provided F00 and G00 are bounded as r ! +0. This property follows from F00(r) = 1 !n Z @B1(0) Xn i. 2 The solution of the above initial value problem. is M0(r.2. This follows since F0 and G0 are in C1(R). Thus F00(+0) = 1 !n Xn i. We recall that f. respectively. These functions are also in 4. F0(r) = 1 !n Z @B1(0) Xn j=1 fyj (x + r»)»j dS» F0(+0) = 1 !n Z @B1(0) Xn j=1 fyj (x)»j dS» = 1 !n Xn j=1 fyj (x) Z @B1(0) nj dS» = 0. where F and G are replaced by F0 and G0. Then. g 2 C2(R2) by assumption. for example.j=1 fyiyj (x + r»)»i»j dS». HIGHER DIMENSIONS 113 C1(R).j=1 fyiyj (x) Z @B1(0) ninj dS».

t) is the solution of the initial value problem with given initially data (4.14) since F0(s) = F(s).6)-(4.1 2cr Z r+ct r−ct »G0(») d». G0(s) = G(s) if s > 0. Since for fixed t > 0 u(x. Since F0 and G0 are even functions. t) = lim r!0 M0(r.7). t) = (r + ct)F0(r + ct) − (ct − r)F0(ct − r) 2r + 1 2cr Z ct+r ct−r »G0(») d». Theorem 4. Thus M0(r. Then there exists a unique solution u 2 C2(R3 × [0.3. we have Z ct−r r−ct »G0(») d» = 0.17) .16) 114 CHAPTER 4. For fixed t > 0 and 0 < r < ct it follows that M0(r. HYPERBOLIC EQUATIONS r−ct ct−r ct+r Figure 4. (4. Assume f 2 C3(R3) and g 2 C2(R3) are given. where n = 3. (4. and the solution is given by the Poisson’s formula u(x.3: Changed domain of integration see Figure 4.1)) of the initial value problem (4. t) = ctF0(ct) + F(ct) + tG(ct) = d dt (tF(ct)) + tG(ct). t). it follows from d’Hospital’s rule that u(x. t) = 1 4¼c2 @ @t Ã 1 t Z @Bct(x) f(y) dSy ! + 1 4¼c2t Z @Bct(x) g(y) dSy.2.

±z. t) be a solution of (4.18)-(4. y.18)-(4. 0) = f(x. see F.2. ´) dS ! + 1 4¼c2t Z @Bct(x. y). Using the formula for the solution of the three-dimensional initial value problem we will derive a formula for the two-dimensional case. t) := v(x. y.2.18) v(x. g(x. case n = 3 4.20) where f 2 C3. since u satisfies (4.0) f(». t) = u(x.20). z.19) vt(x. and uz = 0. 0. The following consideration is called Hadamard’s method of decent. t) = u(x. g 2 C2. z. Above we have shown that a C2-solution is given by Poisson’s formula. then u(x. since u(x. t) 116 CHAPTER 4. y.21) n dS S _ + S r x . 2 Corollary. t) + uz(x. p. Let v(x. Poisson’s formula in the three-dimensional case implies v(x. see Figure 4. Under the additional assumption f 2 C3 it follows from Poisson’s 4. John [10]. t)z. we have v(x.y. t). t0) is the intersection of the cone defined by |y − x| = c|t − t0| with the hyperplane defined by t = 0. From Poisson’s formula we see that the domain of dependence for u(x. 0. HIGHER DIMENSIONS 115 formula that this formula defines a solution which is in C2.Proof.y. Hence. t) = 1 4¼c2 @ @t Ã 1 t Z @Bct(x.t ) 0 |y−x|=c| t−t | 0 Figure 4. (4. y. y. y) (4. y.2 Case n = 2 Consider the initial value problem vxx + vyy = c−2vtt (4. y.20). 0 < ± < 1. y). 0) = g(x. HYPERBOLIC EQUATIONS is a solution of the three-dimensional initial value problem with initial data f(x.0) g(». 129. (4. y. y. y. ´) dS. independent of z. y. y).4: Domain of dependence.4 t x (x.

25) . where u1 is a solution of problem (4.y) g(». y0) and not the boundary only. 0) = f(x) (4. see formula of Theorem 4.3.22)-(4. g have supports in a compact domain D ½ R2. t) for all time t > T. (4. 4. ´.22)-(4. the domain of dependence is here the disk Bcto(x0.3 Inhomogeneous equation Here we consider the initial value problem 2u = w(x.18)(4. t) on x 2 Rn. ³) := (» −x)2 +(´ −y)2 +³2 −c2t2 = 0.22) u(x. Corollary. Set u = u1 + u2. Therefore. t) on x 2 Rn. Since we have explicit solutions u1 in the cases n = 1. it remains to solve 2u = w(x.21) Theorem 4. Then it follows from (4. We have S = S+ [ S−.3.5: Domains of integration The integrands are independent on ³. 4. We assume f 2 C3. Set ½ = p (» − x)2 + (´ − y)2. INHOMOGENEOUS EQUATION 117 The positive sign applies on S+. where the third coordinate of n is n3 = ± p c2t2 − (» − x)2 − (´ − y)2 ct . t 2 R (4. where ³ > 0 and the sign is negative on S− where ³ < 0. ´) p c2t2 − ½2 d»d´.23) ut(x.5. 0) = g(x).h z dxdh Figure 4. t 2 R (4. T sufficiently large. if f. t) = 1 2¼c @ @t Z Bct(x.20) is given by v(x. n = 2 and n = 3. y. ´) p c2t2 − ½2 d»d´ + 1 2¼c Z Bct(x.3.y) f(». see Figure 4. which are given. then these functions have influence on the value v(x.24) where 2u := utt − c24u. In contrast to the three dimensional case. g 2 C2 and w 2 C1.24) with w := 0 and u2 is the solution where f = 0 and g = 0 in (4. y.24). The solution of the Cauchy initial value problem (4. The surface S is defined by Â(». Then the exterior normal n at S is n = rÂ/|rÂ| and the surface element is given by dS = (1/|n3|)d»d´.

26) ut(x.32). since u is an ansatz for (4. 0) = 0. s) = 0. Then . (4. The initial condition u(x. t. s) ds. v(x. 0) = 0 (4.31) and ansatz (4. s) ds = w(x.28). t). (4. see Theorem 4. s) be a solution of 2v = 0. s. t. = Zt 0 vt(x. v¤(x.28) where v is a function satisfying 2v = 0 for all s (4. s) = w(x. s. t. t) + Zt 0 vt(x.32) Let v¤(x. s).30) From ansatz (4. Thus necessarily vt(x. HYPERBOLIC EQUATIONS To solve this problem. s) 4.28) and assumption (4. s) = 0. 118 CHAPTER 4. s) ds. s) = 1 4¼c2t Z @Bct(x) w(». t. we make the ansatz u(x. s) dS».29). s) = w(x. utt − c24xu = vt(x. t) = w(x. s) := v¤(x. t. (4. t. We have seen that the ansatz provides a solution of (4. t. Therefore. t. t. (4. t. vt(x. In the case n = 3. t) = 0 is satisfied because of the ansatz (4. vt(x. t − s. INHOMOGENEOUS EQUATION 119 is a solution of (4.30) we get ut = v(x. see (4. t.27) if for all s 2v = 0. s). 0. (4. s) = 0.29) and v(x. s).u(x. t.3. t.28) we see that utt = vt(x.2. t) = Zt 0 v(x. where v¤ is given by.27). t) + Zt 0 (2v)(x. (4. s. v(x.33) then v(x.31) It follows ut(x.25)-(4. From (4. 0. t). 0) = 0. s) ds.25)-(4. 4xu = Zt 0 4xv(x.27) The following method is called Duhamel’s principle which can be considered as a generalization of the method of variations of constants in the theory of ordinary differential equations. t) + Zt 0 vtt(x. s) ds.

t). s) dS». u(x. t − ¿/c) ¿ dS»d¿ = 1 4¼c2 Z Bct(x) w(». t) = 1 4¼c2 Z ct 0 Z @B¿ (x) w(».v(x.28) it follows u(x. s) = 1 4¼c2(t − s) Z @Bc(t−s)(x) w(». t − r/c) r d». 0) = 0. Formulas for the cases n = 1 and n = 2 follow from formulas for the associated homogeneous equation with inhomogeneous initial values for these cases. s) t−s dS»ds. ut(x. t − r/c) r . t. t) = Zt 0 v(x. t − s. Theorem 4. 0) = 0. is given by: Case n = 3: u(x. where r = |x − »|. The solution of 2u = w(x. s) = v¤(x. s) ds = 1 4¼c2 Zt 0 Z @Bc(t−s)(x) w(». Changing variables by ¿ = c(t − s) yields u(x. t) = 1 4¼c2 Z Bct(x) w(». t. where w 2 C1.4. from ansatz (4.

37) where f 2 C in a neighbourhood of S and u0. 120 CHAPTER 4.34) is ÂxÂy = 0 which is satisfied if Âx(x. y = y(t). »2. where r = |x − »|.4. y(t)). »3). e. The integrand on the right in formula for n = 3 is called retarded potential. y)ux + b(x. y)u. One family of characteristics . » = (»1. t) = 1 2c Zt 0 ÃZ x+c(t−¿) x−c(t−¿) w(». x = (x1. (4. (ii) S is not a characteristic curve. x2. and y = const. x = (x1. i. y 2 C1[t1. HYPERBOLIC EQUATIONS Case n = 2: u(x. » = (»1.. which are lines here and are defined by x = const. We recall that the characteristic equation to (4. q0 2 C1 are given. A METHOD OF RIEMANN 121 (i) u00(t) = p0(t)x0(t) + q0(t)y0(t) (strip condition). tangents of the characteristic and S are different at this point. 4. y(t)) (4.d». Moreover assume that the characteristic curves. y) = 0. t) = 1 4¼c Zt 0 ÃZ Bc(t−¿)(x) w(». ¿ ) d» ! d¿. y(t)) (4. t1 · t · t2. we assume x. Consider the initial value problem Lu = f(x. y) = 0 or Ây(x. Case n = 1: u(x. and such a point is not a touching point. y)uy + c(x. that is. p0. be a regular curve in R2.36) q0(t) = uy(x(t). x2). »2).4 A method of Riemann Riemann’s method provides a formula for the solution of the following Cauchy initial value problem for a hyperbolic equation of second order in two variables. where a. The integrand is taken not at t. Set Lu := uxy + a(x. We assume: 4.. t2] and x02+y02 6= 0. ¿ ) p c2(t − ¿ )2 − r2 d» ! d¿. f 2 C in a neighbourhood of S. have at most one point of intersection with S.35) p0(t) = ux(x(t). it is taken at an earlier time t − r/c. y) (4. b 2 C1 and c. Remark. Let S : x = x(t). x3).34) u0(t) = u(x(t).

y ) 0 Figure 4. see the definition of P and Q. The previous equation follows from Gauss theorem or after integration by parts: Z (−Py + Qx) dxdy = Z @ (−Pn2 + Qn1) ds. (4. (4. HYPERBOLIC EQUATIONS A B S W x y 0 P=(x . y0(t) = 0.34)-(4. it follows from (4. From (4. if we integrate over a domain as shown in Figure 4.38) Set P = −(uxv − xxu + 2buv) Q = uyv − vyu + 2auv. see Chapter 2. where n = (dy/ds.6. 122 CHAPTER 4.40) The line integral from B to A is known from initial data. Since . y(s)) represents @.39) where integration in the line integral is anticlockwise.37) and suppose that v satisfies Mv = 0 in .−dx/ds). domain of integration Then. v 2 C1 and that uxy. We have 2(vLu − uMv) = (uxv − vxu + 2buv)y + (uyv − vyu + 2auv)x.39) that 2 Z vf dxdy = Z BA Pdx+Qdy+ Z AP Pdx+Qdy+ Z PB Pdx+Qdy. (4.38) it follows for a domain 2 R2 2 Z (vLu − uMv) dxdy = Z (−Py + Qx) dxdy = I Pdx + Qdy.associated to these first partial differential of first order is defined by x0(t) = 1.6: Riemann’s method. vxy exist and are continuous. Define the adjoint differential expression by Mv = vxy − (av)x − (bv)y + cv. Assume u is a solution of the initial value problem (4. Assume u. (x(s). s arc length.

(4. see Figure 4. (4.y ) 0 C C2 1 D(P) Figure 4. A METHOD OF RIEMANN 123 Combining these equations with (4.uxv − vxu + 2buv = (uv)x + 2u(bv − vx).43) av − vy = 0 on C2 (4.4.7: Definition of Riemann’s function Mv = 0 in D(P) (4.45).42)-(4.45) Assume v satisfies (4. it follows Z AP Pdx + Qdy = − Z AP ((uv)x + 2u(bv − vx)) dx = −(uv)(P) + (uv)(A) − Z AP 2u(bv − vx) dx. 4. By the same reasoning we obtain for the third line integral Z PB Pdx + Qdy = Z PB ((uv)y + 2u(av − vy)) dy = (uv)(B) − (uv)(P) + Z PB 2u(av − vy) dy.39). we get 2v(P)u(P) = Z BA (uxv − vx + 2buv) dx − (uyv − vyu + 2auv) dy +u(A)v(A) + u(B)v(B) + 2 Z AP u(bv − vx) dx +2 Z PB u(av − vy) dy − 2 Z fv dxdy.7 for the definition of domain D(P).41) Let v be a solution of the initial value problem. then 2u(P) = u(A)v(A) + u(B)v(B) − 2 Z fv dxdy .42) bv − vx = 0 on C1 (4.44) v(P) = 1. x y 0 P=(x .

= Z BA (uxv − vx + 2buv) dx − (uyv − vyu + 2auv) dy. for . Set w(x. y0) d¿ ¶ on C1.45) imply w(x. then a Riemann function is v(x. t)e·t. Bessel’s differential equation is x2y00(x) + xy0(x) + (x2 − n2)y(x) = 0. Stretching the axis and transform the equation to the normal form we get finally the following equation. We make the ansatz for a Riemann function v(x. ¿ ) d¿ ¶ on C2. then solutions are given by Bessel functions. where u stands for one coordinate of electric or magnetic field. Remark. y) ´ 1. where n 2 R. y. Remark. uxy + cu = 0.42)-(4. y. A function v = v(x. y. with a positive constant c. y) = exp µZ y y0 a(x0. w(x0. Introducing u = w(x.5. x0. y) = v(x. where · = −¸/(2"). This bounded solution is the Bessel function Jn(x) of first kind and of order n. A solution is J0(¾) = J0 ³p 4c(x − x0)(y − y0) ´ which defines a Riemann function since J0(0) = 1. see [1]. INITIAL-BOUNDARY VALUE PROBLEMS 125 Substitution ¾ = p4cs leads to Bessel’s differential equation ¾2z00(¾) + ¾z0(¾) + ¾2z(¾) = 0. where z(¾) = w(¾2/(4c)). we arrive at wtt = c2 "μ4xw − ¸2 4²2 . y). y0) = w(s). 2. Consider the telegraph equation of Chapter 3 "μutt = c24xu − ¸μut. x0. y0) satisfying (4. the new function is denoted by u and the new variables are denoted by x. y0) = exp µZ x x0 b(¿. s = (x − x0)(y − y0) and obtain sw00 + w0 + cw = 0. 4. 124 CHAPTER 4.42)-(4. y0) for fixed x0. HYPERBOLIC EQUATIONS where the right hand side is known from given data. y again. One of the two linearly independent solutions is bounded at 0.45) is called Riemann’s function. Then (4. x0. If n 2 N [ {0}. Examples 1. y0. uxy = f(x.

5. 4. if v(x)w(t) 6= 0. In this and in the following section we seek solutions u(x. n = 1. t) on (0. This problem is governed by the initial-boundary value problem utt(x. or. w00(t) w(t) = v00(x) v(x) .4 from Chapter 1. t) = u(l. l) (4. Assume the deflection occurs in the (x. 0) = f(x) (4. . l) (4.49). b].46) we obtain v(x)w00(t) = v00(x)w(t). provided v(x)w(t) is a solution of differential equation (4.1 Oscillation of a string Let u(x. g are sufficiently regular. (4. then v(x)w(t) satisfies the boundary condition (4.5 Initial-boundary value problems In previous sections we looked at solutions defined for all x 2 Rn and t 2 R. t) = uxx(x. HYPERBOLIC EQUATIONS Fourier’s method To find solutions of differential equation (4. t) = 0.46) and v(x)w(t) 6= 0. 2. t). t) defined in a bounded domain ½ Rn and for all t 2 R and which satisfy additional boundary conditions on @. t 2 R.47) ut(x. u)-plane.50) v(0) = v(l) = 0. (4. Assume v(0) = v(l) = 0. . 126 CHAPTER 4.51) which has the eigenvalues ¸n = ³¼ l n ´2 . 0) = g(x) (4. t) = v(x)w(t). 4. be the deflection of a string. Thus we look for solutions of the eigenvalue problem −v00(x) = ¸v(x) in (0. and associated eigenfunctions are vn = sin ³¼ l nx .example.46) we make the separation of variables ansatz u(x. .49) Assume the initial data f.46) u(x. x 2 [a. It follows. =: −¸ and v00(x) v(x) = −¸ since x. Inserting the ansatz into (4. This implies compatibility conditions f(0) = f(l) = 0 and g(0) = g(l). t are independent variables. . w00(t) w(t) = const. see Figure 1.48) u(0.

48) as follows. 0) = 1X n=1 ®n sin( p ¸nx) = f(x). Multiplying this equation by sin(p¸kx) and integrate over (0. Solutions of −w00(t) = ¸nw(t) are sin( p ¸nt). since (4. l).5. cos( p ¸nt).46).´ . ¯n 2 R. (4. t) = 1X n=1 ³ ®n cos( p ¸nt) + ¯n sin( p ¸nt) ´ sin ³p ¸nx ´ . Set wn(t) = ®n cos( p ¸nt) + ¯n sin( p ¸nt). We have u(x.46) and the boundary conditions (4. (4.47). also (principle of superposition) uN = XN n=1 wn(t)vn(x) which satisfies the differential equation (4. Formally. It is easily seen that wn(t)vn(x) is a solution of differential equation (4. we get ®n Zl 0 sin2( .49). 4.49) u(x. Consider the formal solution of (4. (4.46) is linear and homogeneous. the unknown coefficients can be calculated from initial conditions (4. INITIAL-BOUNDARY VALUE PROBLEMS 127 where ®n. and.46).52) ”Formal” means that we know here neither that the right hand side converges nor that it is a solution of the initial-boundary value problem.

54) Under additional assumptions f 2 C4 0 (0. ¯n are given by (4. l). Then ®k = 2 l Zl 0 f(x) sin µ ¼k l x ¶ dx.53) and (4.p ¸kx) dx = Zl 0 f(x) sin( p ¸kx) dx.54). We recall that Z l 0 sin( p ¸nx) sin( p ¸kx) dx = l 2 ±nk. (4. 0) = 1X n=1 ¯n p ¸n sin( p ¸nx) = g(x) that ¯k = 2 k¼ Zl 0 g(x) sin µ ¼k l x ¶ dx. where ®n.52). g 2 C3 0 (0. .53) By the same argument it follows from ut(x. l) it follows that the right hand side of (4. (4.

(4. t) = 1X n=1 ³ ®n cos( p ¸nt) + ¯n sin( p ¸nt) ´ vn(x). Then the formal solution of the above initial-boundary value problem is u(x. we make the ansatz (separation of variables) u(x. In general. We assume is sufficiently regular such that the eigenvalues are countable. t) = w(t)v(x) which leads to the eigenvalue problem −4v = ¸v in . Rectangle membrane. utt. t) = 0 on @ × R. 0 < T < 1 fixed.55) u(x. 4. We consider the initial-boundary value problem utt(x.59) are not known explicitly.60) which are given by ¸kl = r k2 a2 + . 0) = g(x).60) and vn a complete associated orthonormal system of eigenfunctions. uxx converges uniformly on 0 · x · l.57) u(x.56) ut(x. t) = 4xu in × R. INITIAL-BOUNDARY VALUE PROBLEMS 129 Examples 1. (4. we find all eigenvalues of (4. ux.2 Oscillation of a membrane Let ½ R2 be a bounded domain. a) × (0. HYPERBOLIC EQUATIONS respectively. where ®n = Z f(x)vn(x) dx ¯n = 1 p¸n Z g(x)vn(x) dx.59). see an exercise. x 2 . b).59).49) since under these assumptions the series for u and the formal differentiate series for ut.5. (4. Using the method of separation of variables.46)-(4. (4. see next examples. Let = (0.58) As in the previous subsection for the string. 0) = f(x). (4. eigenvalues of (4. (4.5. In some special cases.128 CHAPTER 4. 4. defines a classical solution of (4. Remark. x 2 . 0 · t · T.59). (4.59) v = 0 on @. (4. There are numerical methods to calculate these values. (4. which is satisfied in the following examples.60) Let ¸n are the eigenvalues of (4. these values are known.

k. we obtain − 1 r µ (rv0(r))0 v(r) + 1 . not normalized. Dividing by vq. . Set = {x 2 R2 : x21 + x22 < R2}. where v(R) = 0 and q(µ) is periodic with period 2¼ since u(r. In polar coordinates.l2 b2 . provided vq 6= 0. µ) is single valued. Disk membrane. This leads to − 1 r µ (rv0)0q + 1 r vq00 ¶ = ¸vq. µ) = v(r)q(µ). . l = 1. 2. the eigenvalue problem (4. are ukl(x) = sin µ ¼k a x1 ¶ sin µ ¼l b x2 ¶ .62) here is u = u(r. .60) is given by − 1 r µ (rur)r + 1 r uµµ ¶ = ¸u (4. and associated eigenfunctions. µ) = 0.61) u(R. (4. µ) := v(r cos µ. 2. (4. r sin µ).59). We will find eigenvalues and eigenfunctions by separation of variables u(r.

because we are looking for q not identically zero. 1. where A.63).R) (4. then. From the periodicity requirement Asin(pμµ) + B cos(pμµ) = Asin(pμ(µ + 2¼)) + B cos(pμ(µ + 2¼)) it follows2 sin(pμ¼) (Acos(pμµ + pμ¼) − B sin(pμµ + pμ¼)) = 0. B are not zero simultaneously.61)-(4. Consequently the eigenvalues are μn = n2. . .65) sup r2(0. (4. we arrive at the eigenvalue problem −q00(µ) = μq(µ) q(µ) = q(µ + 2¼). Solutions of this differential equations which are bounded at zero are Bessel functions of first kind and n-th order Jn(z). The eigenvalues follows from boundary condition (4. from Jn(p¸R) = 0.5. (4.64) v(R) = 0 (4. z2y00(z) + zy0(z) + (z2 − n2)y(z) = 0. =: −μ. . which implies. 130 CHAPTER 4. i. μ).. see (4.63) which implies q00(µ) q(µ) = const. we obtain the boundary value problem r2v00(r) + rv0(r) + (¸r2 − n2)v = 0 on (0.65). . since A. e. then the eigenvalues of (4. INITIAL-BOUNDARY VALUE PROBLEMS 131 where z > 0.64).r q00(µ) q(µ) ¶ = ¸. B are arbitrary real constants. It follows that eigenvalues μ are real and nonnegative.66) Set z = p¸r and v(r) = v(z/p¸) =: y(z).61) are ¸nk = ³¿nk R ´2 and the associated eigenfunctions are Jn( . All solutions of the differential equation are given by q(µ) = Asin(pμµ) + B cos(pμµ).R) |v(r)| < 1. sin(pμ¼) sin(pμµ + ±) = 0 for all µ and a ± = ±(A. Inserting q00(µ)/q(µ) = −n2 into (4.B. Denote by ¿nk the zeros of Jn(z). HYPERBOLIC EQUATIONS Thus. n = 0. 2 sin x − sin y = 2 cos x+y 2 sin x−y 2 cos x − cos y = −2 sin x+y 2 sin x−y 2 4.

HYPERBOLIC EQUATIONS 3. that is. 4. For tables with zeros of Jn(x) and for much more properties of Bessel functions see [25]. where aij = aji are given sufficiently regular functions defined on . .p ¸nkr) sin(nµ). the asymptotic formula Jn(x) = µ 2 ¼x ¶1/2 µ cos(x − n¼/2 − ¼/5) + O µ 1 x ¶¶ as x ! 1.j=1 aij³i³j ¸ º|³|2 for all x 2 and ³ 2 Rn. Lu = Xn i. n ¸ 1. um) and Lu = Xn i. . 2.j=1 @ @xj ¡ aij(x)uxi ¢ . In this section we consider the initial-boundary value problem utt = Lu + f(x. 0) = Á(x) x 2 (4. L = 4x. 1.3 Inhomogeneous wave equations Let ½ Rn be a bounded and sufficiently regular domain. Examples for L are: 1. . (4. . One has. . . oscillating membrane. t) = 0 for x 2 @ and t 2 Rn. 132 CHAPTER 4. n = 0. xn).68) ut(x. are double eigenvalues. Á.5. f. It follows from this formula that there are infinitely many zeros of Jn(x).j=1 @ @xj ¡ Aij(x)uxi . . Remark. . . 4. oscillating string. Thus the eigenvalues ¸0k are simple and ¸nk. x = (x1. We assume L is uniformly elliptic. t).69) u(x. Jn( p ¸nkr) cos(nµ). . there is a constant º > 0 such that Xn i.70) where u = u(x. Ã are given and L is an elliptic differential operator. 2. . Let u = (u1. . L = @2/@x2. t) in × R (4. 0) = Ã(x) x 2 (4. 2. . n = 1. . . in particular.67) u(x.

respectively. then Á(x) = 1X k=1 hÁ. An example for this case is the linear elasticity.¢ . which is complete and orthonormal in L2(). t) = 1X k=1 ck(t)vk(x) (4. .70) we make the ansatz u(x. vkivk(x) are Fourier’s decomposition of Á and Ã. vki = Z Á(x)vk(x) dx. Thus wk(t) = ak cos( p . which occurs in ansatz (4.67)-(4. . (4. from the requirement that u = vk(x)wk(t) is a solution of utt = Lu + ck(t)vk(x) and that the initial conditions wk(0) = hÁ.74) after multiplying with vl(x) and integrating over . From the above differential equation it follows w00 k(t) = −¸kwk(t) + ck(t). We have ck(t) = Z f(x. .73) with functions wk(t) which will be determined later. We assume that L defines an elliptic system. vki are satisfied. v2.75) which follows from (4. where Aij = Aji are given sufficiently regular (m × m)-matrices on . Set hÁ. vki. Consider the eigenvalue problem −Lv = ¸v in (4. vkivk(x) Ã(x) = 1X k=1 hÃ. t)vk(x) dx. . t) = 1X k=1 vk(x)wk(t).5.74) 4. . (4. In the following we will determine wk(t).73). (4. It is assumed that all series are convergent and that following calculations make sense.72) Assume there are infinitely many eigenvalues 0 < ¸1 · ¸2 · . ! 1 and a system of associated eigenfunctions v1. For the solution of (4.71) v = 0 on @. w0k (0) = hÃ. Let f(x. INITIAL-BOUNDARY VALUE PROBLEMS 133 be Fourier’s decomposition of f with respect to the eigenfunctions vk. This assumption is satisfied if is bounded and if @ is sufficiently regular.

76) + 1 p¸k Zt 0 ck(¿ ) sin( p ¸k(t − ¿ )) d¿.72) and the functions wk are defined by (4. HYPERBOLIC EQUATIONS Proposition 4.67)-(4.67)-(4. where vk is a complete orthonormal system of eigenfunctions of (4. The resonance phenomenon Set in (4. t) ! . t) = Avn(x) p¸n Zt 0 sin(!¿ ) sin( p ¸n(t − ¿ )) d¿ = Avn(x) 1 !2 − ¸n µ ! p¸n sin( p ¸kt) − sin(!t) ¶ . ! are real constants and vn is one of the eigenfunctions of (4. t) = Asin(!t)vn(x). It follows u(x.71). Summarizing.70) is u(x.72). where A.76). where ak = hÁ. t)vk(x) dx = A±nk sin(!t). Then the solution of the initial value problem (4. The (formal) solution of the initial-boundary value problem (4. (4.70) is given by u(x. It follows ck(t) = Z f(x. t) = 1X k=1 vk(x)wk(t). bk = 1 p¸k hÃ. (4. Ã = 0 and assume that the external force f is periodic and is given by f(x.2.67)-(4. we have 134 CHAPTER 4. vki.71). provided ! 6= p¸n.¸kt) + bk sin( p ¸kt) (4.70) Á = 0. vki.

t) = 0 on × R. since L is elliptic. t) 2 C2(R2) is a solution of the one-dimensional wave equation . HYPERBOLIC EQUATIONS 4. 0) = 0 and u(x. which implies. 0) = 0 we get E(0) = 0. the homogeneous initial and boundary value conditions lead to u(x. on × R. The right hand side is also the solution of the initial-boundary value problem if ! = p¸n. Finally. 2 136 CHAPTER 4.j=1 aij(x)uxiuxj + utut) dx. Let u1. t) = const. INITIAL-BOUNDARY VALUE PROBLEMS 135 Proof. The frequencies p¸n are called critical frequencies at which resonance occurs. Consequently |u| can be arbitrarily large at some points x and at some times t if ! = p¸n. It follows E(t) = const. As an example we consider Example 3 from above and set E(t) = Z ( Xn i. From ut(x. 0) = 0 x 2 u(x.5.6 Exercises 1.A 2p¸n vn(x) µ sin(p¸nt) p¸n − t cos( p ¸nt) ¶ if ! ! p¸n.70) is unique in the class C2( × R).j=1 aij(x)uxiutnj) dS +2 Z ut(−Lu + utt) dx = 0. t) = 0 for x 2 @ and t 2 Rn. 0) = 0 x 2 ut(x. u2 are two solutions. then u = u2 − u1 satisfies utt = Lu in × R u(x.67)-(4. that u(x. Consequently E(t) = 0 for all t. Then E0(t) = 2 Z ( Xn i. A uniqueness result The solution of of the initial-boundary value problem (4. 4. Show that u(x.j=1 aij(x)uxiuxj t + ututt) dx =2 Z @ ( Xn i.

0) = g(x0) . (ii) Set ®n := Zl 0 f(x) sin . t)-space. t) := 1X n=1 ®n cos ³¼n l t ´ sin ³¼n l x ´ is a C2-solution of the wave equation utt = uxx if |®n| · c/n4. Show that the Cauchy initial value problem 2u = 0 4. |°| < 1/c2) in (x. i. l). x03 = x3. see Figure 4. 0) = f(x) . x3).8: Figure to the exercise 2. t = °x} be spacelike. which are bounded by characteristic lines. x t B C A D Figure 4. Method of separation of variables: Let vk(x) be an eigenfunction to the eigenvalue of the eigenvalue problem −v00(x) = ¸v(x) in (0.. Prove that vk(x)wk(t) is a solution of the partial differential equation (wave equation) utt = uxx. x02 = x2. t0 = pt − °x1 1 − °2c2 into the initial value problem. t). 2u = 0 u(x0.8. 4. Solve for given f(x) and μ 2 R the initial value problem ut + ux + μuxxx = 0 in R × R+ u(x. (i) Show that u(x. 5. Here we denote the transformed function by u again. EXERCISES 137 with data for u on S can be transformed using the Lorentz-transform x1 = x1 − °c2t p 1 − °2c2 . 0) = f(x0) ut0(x0. v(0) = v(l) = 0 and let wk(t) be a solution of differential equation −w00(t) = ¸kw(t). t)-plane. e. Let S := {(x. x2.utt = c2uxx if and only if u(A) + u(C) = u(B) + u(D) holds for all parallelograms ABCD in the (x.6. in new coordinates. where the constant c is independent of n. 3. x = (x1.

0) = sin x ut(x. Remark. provided f 2 C4 0 (0. 9. t) = 0. e is here a positive constant. t) = f(x2 − c2t2) = f(s). 6. t) = 0 on @. 7. 0 < x3 < h}. Then eigenvalues are given by En = −me4/(2~2n2).1). and u(1. x3) 2 Rn : x21 + x22 < R2. t > 0. x 2 R u(x. t) = 0 for all t 2 [0. provided E 2 R is an (eigenvalue) of the elliptic equation 4u + 2m ~2 (E − V (x))u = 0 in Rn 138 CHAPTER 4. V (x) a given potential.6. 11. Find nonzero solutions by using separation of variables of utt = 4xu in × (0. ¸ = const. Prove |®n| · c/n4. 0) = x ut(x.³¼n l x ´ dx. Hint: Separation of variables. of utt − uxx + u = 0 . Solve the initial value problem utt − c2uxx = x2. u : Rn 7! C. u(x. which satisfies the side condition Zn R |Ã(x. HYPERBOLIC EQUATIONS under the side condition Rn R |u|2dx = 1. Find solutions of the equation utt − c2uxx = ¸2u. see [22]. 0 · t < 1. EXERCISES 139 with f(0) = K. 202. such that u(0. Find with the method of separation of variables nonzero solutions u(x. ~ Planck’s constant (a small positive constant). K a constant. 10. Here is Ã : Rn × R 7! C. pp. s := x2 − c2t2 4. Find a solution of Schr¨odinger’s equation i~Ãt = − ~2 2m4xÃ + V (x)Ã in Rn × R. and transform the above problem into an initial value problem with homogeneous differential equation by using this solution. Let be the rectangle (0. 12. u = 0 on @. Find all eigenvalues and associated eigenfunctions of −4u = ¸u in . t)|2dx = 1 . 0 · x · 1. t).1). a) × (0. 0) = 0 . . b). l). 0) = 1 . where is the circular cylinder = {(x1. In the case of a hydrogen atom the potential is V (x) = −e/|x|. Solve the initial value problem 3utt − 4uxx = 0 u(x. Hint: Find a solution of the differential equation independent on t. n 2 N. x2. 8. which can be written as u(x.

. Proof.Hint: Transform equation for f(s) by using the substitution s := z2/A with an appropriate constant A into Bessel’s differential equation z2f00(z) + zf0(z) + (z2 − n2)f = 0. then there is a constant M = M(n. Remark. . Find the formula for the solution of the following Cauchy initial value problem uxy = f(x.1 Definition.4). 14. ux = ®.2) is called inverse Fourier transform. and the initial conditions on S are given by u = ®x + ¯y + °. Let » = (»1. 5. uy = ¯. properties Definition. is called Fourier transform of f. s = 0. y). More precisely. where S: y = ax + b. HYPERBOLIC EQUATIONS Chapter 5 Fourier transform Fourier’s transform is an integral transform which can simplify investigations for linear differential or integral equations since it transforms a differential operator into an algebraic equation. Let f 2 Cs 0(Rn). ®.1) where » 2 Rn.1. s. From (5. the right hand side of (5. provided f 2 Cs 0(Rn). 140 CHAPTER 4. . we have Proposition 5. a > 0. a. In particular. z > 0 with n = 0.2. dD®f(») = i|®|»® b f(»). . Proposition 5. ° constants. The above differential equation for u is the transformed telegraph equation (see Section 4.2) exists for g := ˆ f if f 2 Cn+1 0 (Rn). (5. Find all eigenvalues μ of −q00(µ) = μq(µ) q(µ) = q(µ + 2¼) . FOURIER TRANSFORM The following proposition shows that the Fourier transform of f decreases rapidly for |»| ! 1. The function ˆ f defined by b f(») = (2¼)−n/2 Z Rn e−i»·xf(x) dx. provided the integrals on the right hand side exist. »n) be fixed and let j be an index such that |»j | = . .1) it follows by integration by parts that differentiation of a function is changed to multiplication of its Fourier transforms. . g) such that |bg(»)| · M (1 + |»|)s . and the function eg given by eg(x) = (2¼)−n/2 Z Rn ei»·xg(») d» (5. Assume g 2 Cs 0(Rn). b. 13. . 141 142 CHAPTER 5. where |®| · s. 1. ¯. or an analytical operation is converted into an algebraic operation. .

This inequality and Proposition 5. Proof. The proof of the other relation is left as an exercise.maxk |»k|. PROPERTIES 143 here. All integrals appearing in the following exist. see Proposition 5. DEFINITION.1. Then |»| = Ã Xn k=1 »2 k !1/2 · pn|»j | which implies (1 + |»|)s = Xs k=0 µ s k ¶ |»|k · 2s Xs k=0 nk/2|»j |k · 2sns/2 X |®|·s |»®|. We will prove the first assertion (2¼)−n/2 Z Rn ei»·x b f(») d» = f(x) (5.1 imply (1 + |»|)s|bg(»)| · 2sns/2 X |®|·s |(i»)®bg(»)| · 2sns/2 X |®|·s Z Rn |D®g(x)| dx =: M. eb f = f and be f = f.2) is justified by Theorem 5.2 and the special choice of g. See [27].1. 2 The notation inverse Fourier transform for (5.3) 5. (i) Formula Z Rn g(») b f(»)eix·» d» = Z Rn . for example.

then (5.5). (ii) Formula (2¼)−n/2 Z Rn e−iy·»g("») d» = "−nbg(y/") (5.1).5). Since.4) and (5. bG (y) = (2¼)−n/2 Z Rn e−iy·»g("») d» = "−nbg(y/"). Set G(») := g("»). .bg(y)f(x + y) dy (5.6) follows from (5. see (5. (iii) Equation Z Rn g("») b f(»)eix·» d» = Z Rn bg(y)f(x + "y) dy (5.4) implies Z Rn G(») b f(»)eix·» d» = Z Rn bG (y)f(x + y) dy.5) for each " > 0 follows after substitution z = "» in the left hand side of (5.4) follows by direct calculation: Z Rn g(») µ (2¼)−n/2 Z Rn e−ix·yf(y) dy ¶ eix·» d» = (2¼)−n/2 Z Rn µZ Rn g(»)e−i»·(y−x) d» ¶ f(y) dy = Z Rn bg(y − x)f(y) dy = Z Rn bg(y)f(x + y) dy.

(5.7) and (5.8).7) Set g(x) := e−|x|2/2.1 follows from (5. (iv) Proof of (5.8). we get g(0) Z Rn b f(»)eix·» d» = f(x) Z Rn bg(y) dy. FOURIER TRANSFORM we arrive at Z Rn g("») b f(») = Z Rn "−nbg(y/")f(x + y) dy = Z Rn bg(z)f(x + "z) dz. Since − µ x p2 +i y p2 ¶ · µ x p2 +i y p2 ¶ =− µ |x|2 . then Z Rn bg(y) dy = (2¼)n/2. Letting " ! 0. (5. We will show bg(y) : = (2¼)−n/2 Z Rn e−|x|2/2e−ix·x dx = e−|y|2/2.8) Since g(0) = 1. It remains to show (5.144 CHAPTER 5.8). the first assertion of Theorem 5. The proof of Z Rn e−|y|2/2 dy = (2¼)n/2 is left as an exercise.

where the integration is along the curve C which is the union of four curves as indicated in Figure 5. Consider first the one-dimensional case. Re Im C C C iC y 2 h h 4 3 1 2 −R R Figure 5.1.1: Proof of (5. PROPERTIES 145 where ´ := x p2 +i y p2 . According to Cauchy’s theorem we have I C e−´2 d´ = 0.8) Consequently Z .1.2 + ix · y − |y|2 2 ¶ it follows Z Rn e−|x|2/2e−ix·y dx = Z Rn e−´2 e−|y|2/2 dx = e−|y|2/2 Z Rn e−´2 dx = 2n/2e−|y|2/2 Z Rn e−´2 d´ 5. DEFINITION.

4. It follows lim R!1 Z C3 e−´2 d´ = p¼ since lim R!1 Z Ck e−´2 d´ = 0. .C3 e−´2 d´ = 1 p2 ZR −R e−x2/2 dx − Z C2 e−´2 d´ − Z C4 e−´2 d´. . 146 CHAPTER 5. where ´l = xl p2 +i yl p2 . FOURIER TRANSFORM From d´ = d´1 . . ´n). . Set ´= x p2 +i y p2 = (´1. . k = 2. d´l and e−´2 = e− Pn l=1 ´2 l= Yn l=1 e−´2 l it follows Z . . The case n > 1 can be reduced to the one-dimensional case as follows.

Rn e−´2 d´ = Yn l=1 Z ¡l e−´2 l d´l, where for fixed y ¡l = {z 2 C : z = xl p2 +i yl p2 ,−1 < xl < +1}. 2 There is a useful class of functions for which the integrals in the definition of b f and e f exist. For u 2 C1(Rn) we set qj,k(u) := max ®: |®|·k µ sup Rn ³ (1 + |x|2)j/2|D®u(x)| ´¶ . Definition. The Schwartz class of rapidly degreasing functions is S(Rn) = {u 2 C1(Rn) : qj,k(u) < 1 for any j, k 2 N [ {0}} . This space is a Frech´et space. Proposition 5.3. Assume u 2 S(Rn), then bu and eu 2 S(Rn). Proof. See [24], Chapter 1.2, for example, or an exercise. 5.1.1 Pseudodifferential operators The properties of Fourier transform lead to a general theory for linear partial differential or integral equations. In this subsection we define Dk = 1 i @ @xk , k = 1, . . . , n, and for each multi-index ® as in Subsection 3.5.1 D® = D®1 1 . . .D®n n. 5.1. DEFINITION, PROPERTIES 147 Thus D® = 1 i|®| @|®| @x®1 1 . . . @x®n n .

Let p(x,D) := X |®|·m a®(x)D®, be a linear partial differential of order m, where a® are given sufficiently regular functions. According to Theorem 5.1 and Proposition 5.3, we have, at least for u 2 S(Rn), u(x) = (2¼)−n/2 Z Rn eix·»bu(») d», which implies D®u(x) = (2¼)−n/2 Z Rn eix·»»®bu(») d». Consequently p(x,D)u(x) = (2¼)−n/2 Z Rn eix·»p(x, »)bu(») d», (5.9) where p(x, ») = X |®|·m a®(x)»®. The right hand side of (5.9) makes sense also for more general functions p(x, »), not only for polynomials. Definition. The function p(x, ») is called symbol and (Pu)(x) := (2¼)−n/2 Z Rn eix·»p(x, »)bu(») d» is said to be pseudodifferential operator. An important class of symbols for which the right hand side in this definition of a pseudodifferential operator is defined is Sm which is the subset of p(x, ») 2 C1( × Rn) such that |D¯ xD® » p(x, »)| · CK,®,¯(p) (1 + |»|)m−|®| for each compact K ½ . 148 CHAPTER 5. FOURIER TRANSFORM Above we have seen that linear differential operators define a class of pseudodifferential operators. Even integral operators can be written (formally) as pseudodifferential operators. Let (Pu)(x) = Z Rn K(x, y)u(y) dy be an integral operator. Then (Pu)(x) = (2¼)−n/2 Z Rn K(x, y)

Z Rn eix·»»®bu(») d» = (2¼)−n/2 Z Rn eix·» µZ Rn ei(y−x)·»K(x, y) dy ¶ bu(»). Then the symbol associated to the above integral operator is p(x, ») = Z Rn ei(y−x)·»K(x, y) dy. 5.2. EXERCISES 149 5.2 Exercises 1. Show Z Rn e−|y|2/2 dy = (2¼)n/2. 2. Show that u 2 S(Rn) implies ˆu, eu 2 S(Rn). 3. Give examples for functions p(x, ») which satisfy p(x, ») 2 Sm. 4. Find a formal solution of Cauchy’s initial value problem for the wave equation by using Fourier’s transform. 150 CHAPTER 5. FOURIER TRANSFORM Chapter 6 Parabolic equations Here we consider linear parabolic equations of second order. An example is the heat equation ut = a24u, where u = u(x, t), x 2 R3, t ¸ 0, and a2 is a positive constant called conductivity coefficient. The heat equation has its origin in physics where u(x, t) is the temperature at x at time t, see [20], p. 394, for example. Remark 1. After scaling of axis we can assume a = 1. Remark 2. By setting t := −t, the heat equation changes to an equation which is called backward equation. This is the reason for the fact that the heat equation describes irreversible processes in contrast to the wave equation 2u = 0 which is invariant with respect the mapping t 7! −t. Mathematically, it means that it is not possible, in general, to find the distribution of temperature at an earlier time t < t0 if the distribution is given at t0. Consider the initial value problem for u = u(x, t), u 2 C1(Rn × R+), ut = 4u in x 2 Rn, t ¸ 0, (6.1) u(x, 0) = Á(x), (6.2) where Á 2 C(Rn) is given and 4 ´ 4x. 151 152 CHAPTER 6. PARABOLIC EQUATIONS 6.1 Poisson’s formula Assume u is a solution of (6.1), then, since Fourier transform is a linear mapping, u\t −4u = ˆ0. From properties of the Fourier transform, see Proposition 5.1, we have d4u = Xn

k=1 d@2u @x2 k = Xn k=1 i2»2 kbu(»), provided the transforms exist. Thus we arrive at the ordinary differential equation for the Fourier transform of u dbu dt + |»|2bu = 0, where » is considered as a parameter. The solution is bu(», t) = bÁ(»)e−|»|2t since bu(», 0) = bÁ(»). From Theorem 5.1 it follows u(x, t) = (2¼)−n/2 Z Rn bÁ(»)e−|»|2tei»·x d» = (2¼)−n Z Rn Á(y) µZ Rn ei»·(x−y)−|»|2t d» ¶ dy. Set K(x, y, t) = (2¼)−n Z Rn ei»·(x−y)−|»|2t d». By the same calculations as in the proof of Theorem 5.1, step (vi), we find K(x, y, t) = (4¼t)−n/2e−|x−y|2/4t. (6.3) Thus we have u(x, t) = 1¡ 2p¼t ¢n Z Rn Á(z)e−|x−z|2/4t dz. (6.4) Definition. Formula (6.4) is called Poisson’s formula and the function K defined by (6.3) heat kernel or fundamental solution of the heat equation. 6.1. POISSON’S FORMULA 153 r K K(x,y,t ) K(x,y,t ) 1 2 Figure 6.1: Kernel K(x, y, t), ½ = |x − y|, t1 < t2 Proposition 6.1 The kernel K has following properties:

(i) K(x, y, t) 2 C1(Rn × Rn × R+), (ii) (@/@t − 4)K(x, y, t) = 0, t > 0, (iii) K(x, y, t) > 0, t > 0, (iv) R Rn K(x, y, t) dy = 1, x 2 Rn, t > 0, (v) For each fixed ± > 0: lim t!0 t>0 Z Rn\B±(x) K(x, y, t) dy = 0 uniformly for x 2 Rn. Proof. (i) and (iii) are obviously, and (ii) follows from the definition of K. Equations (iv) and (v) hold since Z Rn\B±(x) K(x, y, t) dy = Z Rn\B±(x) (4¼t)−n/2e−|x−y|2/4t dy = ¼−n/2 Z Rn\B±/p4t(0) e−|´|2 d´ 154 CHAPTER 6. PARABOLIC EQUATIONS by using the substitution y = x+(4t)1/2´. For fixed ± > 0 it follows (v) and for ± := 0 we obtain (iv). 2 Theorem 6.1. Assume Á 2 C(Rn) and supRn |Á(x)| < 1. Then u(x, t) given by Poisson’s formula (6.4) is in C1(Rn × R+), continuous on Rn × [0,1) and a solution of the initial value problem (6.1), (6.2). Proof. It remains to show lim x!» t!0 u(x, t) = Á(»). Since Á is continuous there exists for given " > 0 a ± = ±(") such that |Á(y)− x t x x+d x+2d Figure 6.2: Figure to the proof of Theorem 6.1 Á(»)| < " if |y − »| < 2±. Set M := supRn |Á(y)|. Then, see Proposition 6.1, u(x, t) − Á(») = Z Rn K(x, y, t) (Á(y) − Á(»)) dy. 6.2. INHOMOGENEOUS HEAT EQUATION 155 It follows, if |x − »| < ± and t > 0, that |u(x, t) − Á(»)| · Z B±(x) K(x, y, t) |Á(y) − Á(»)| dy + Z

Uniqueness follows under the additional growth assumption |u(x. 0) = Á(x). That means.Rn\B±(x) K(x. 2. u(x. t ¸ 0. t) = e−|»|2t bÁ(») + Zt 0 e−|»|2(t−¿) b f(». t). t) in x 2 Rn. t) = (2¼)−n/2 Z Rn eix·» ³ e−|»|2t bÁ(») + Zt . t) defined by Poisson’s formula depends on all values Á(y). step (vi). has influence to the value u(x. where M and a are positive constants. a perturbation of Á. y. t)| · Mea|x|2 in DT . this means that heat travels with infinite speed. y. t0 sufficiently small. u(x. In the one-dimensional case. ¿ ) d¿. t) bu(». t) |Á(y) − Á(»)| dy +2M Z Rn\B±(x) K(x. In physical terms. t) |Á(y) − Á(»)| dy · Z B2±(x) K(x. pp. see [10]. one has uniqueness in the class u(x. t) dy ·" Z Rn K(x. where Á and f are given. y. t) dy + 2M Z Rn\B±(x) K(x. 222. Applying the inverse Fourier transform and a calculation as in the proof of Theorem 5. we get u(x.1. t) ¸ 0 in DT . in contrast to the experience. t) 156 CHAPTER 6. even far from a fixed x. u 2 C1(Rn×R+). y.2 Inhomogeneous heat equation Here we consider the initial value problem for u = u(x. y 2 Rn. t). 2 Remarks. t) dy < 2" if 0 < t · t0. 0) = bÁ(»). ut −4u = f(x. 6.2 below. PARABOLIC EQUATIONS we obtain an initial value problem for an ordinary differential equation: dbu dt + |»|2bu = b f(». The solution is given by bu(». see Proposition 6. From u\t −4u =\f(x. y. 1.

there is an (x0. t) 2 × {0} or (x. that ut.0 e−|»|2(t−¿) b f(».1)). From the above calculation for the homogeneous problem and calculation as in the proof of Theorem 5. Proof. Assume initially ut − 4u < 0 in DT . ¿ ) d¿ ´ d». we obtain the formula u(x. uxixk exist and are continuous in DT . Let " > 0 be small and 0 < " < T. step (vi).2. t0) = max . Since u 2 C(DT−"). Then max DT u(x. ¿ )| < 1. and ut −4u · 0 in DT . T > 0. This function u(x. 6. t) 2 @ × [0. 6. t0) 2 DT−" such that u(x0. t) = 1 (2p¼t)n Z Rn Á(y)e−|y−x|2/(4t) dy + Zt 0 Z Rn f(y. ST = {(x. t) = max ST u.1. t) : (x.3 Theorem 6. Set DT = × (0.3 Maximum principle Let ½ Rn be a bounded domain. 0 · ¿ < 1. T]}.3: Notations to the maximum principle see Figure 6. Assume u 2 C(DT ). sup Rn |Á(x)| < 1 and if f 2 C(Rn × [0. T).3. ¿ ) 1³ 2 p ¼(t − ¿ ) ´n e−|y−x|2/(4(t−¿)) dy d¿. MAXIMUM PRINCIPLE 157 T D ST x t T dW dW T− e Figure 6. M(¿ ) := sup Rn |f(y. t) is a solution of the above inhomogeneous initial value problem provided Á 2 C(Rn).

It follows that there is (x. we obtain . . t0) 2 ×{T −"}. we have shown that max DT−" u(x. l = 1. t0) · 0 for each k. where k is a positive constant. 158 CHAPTER 6. t0) ¸ 0. t0) 2 DT−". PARABOLIC EQUATIONS Case (ii). Let (x0. The previous inequality implies that uxkxk (x0. . t) + kt) · max DT v(x. Then it follows as above 4u · 0 in (x0. Then vt −4v = ut −4u − k < 0. . t) since ST−" ½ ST and ST is compact. t) − kt. We arrived at a contradiction to ut −4u < 0 in DT again. t) = max DT (v(x. Letting k ! 0. t) + kT · max ST u(x. t") = max DT−" u(x. t) + kT. From above we have max DT u(x. Now assume ut −4u · 0 in DT . t) = max T−" u(x. Since u is continuous on DT . Thus we arrived at a contradiction to ut −4u < 0 in DT . t). uxl(x0. t). Thus there is an (x". ut(x0. t0) = 0. we have lim "!0 max DT−" u(x. t · t0. t0)³l³k · 0 for all ³ 2 Rn. Hence. Thus. t) = max DT u(x. t) := u(x. n and Xn l. t0) = 0. t) = max DT u(x. since DT−" is open. Case (i). t0) ¸ u(x0. t). .k=1 uxlxk (x0. t). and from u(x0. t) + kT = max ST v(x. t) 2 ST such that u(x. Assume (x0. Set v(x. one concludes that ut(x0. Summarizing. t") 2 ST−" such that u(x". t0). t). theorem is shown under the assumption ut −4u < 0 in DT .DT−" u(x.

0 < T < 1. PARABOLIC EQUATIONS for fixed y 2 Rn and for a constant μ > 0. t) = max ST u. Consider the comparison function vμ(x.3.2. uxixk exist and are continuous in DT . 0) for k¿ · t · (k + 1)¿ . iy. The initial value problem ut − 4u = 0 in DT .½ vμ. 6. k = 0. see Figure 6. the theorem is shown. x 2 Rn.½ = {(x. t) − μ (4¼(T + ² − t))−n/2 e|x−y|2/(4(T+²−t)) = u(x.max DT u(x. t)| · Mea|x|2 . We can assume that 4aT < 1. iy. Since the heat kernel K(ix. Then we conclude successively for k that u(x. . ut −4u · 0 in DT . 2 If we replace in the above theorem the bounded domain by Rn. and additionally that u satisfies the growth condition u(x. More precisely.2 with = B½(y).½ we have. that ut.3. t) satisfies Kt = 4Kx. since the finite interval can be divided into finite many intervals of equal length ¿ with 4a¿ < 1.2 that vμ(y. 0 < t < T}. 0) = f(x). Set for a fixed T. t) · Mea|x|2 . t) : x 2 Rn. . Then we obtain from Theorem 6. t) : |x − y| < ½. then the result remains true provided we assume an additional growth assumption for u. It follows immediately the Corollary. Set for a constant ½ > 0 DT. uxixk exist and are continuous in DT and |u(x. Then max DT u(x.½ ´ ST of Theorem 6. t). MAXIMUM PRINCIPLE 159 Since ST ½ DT . . On the bottom of ST. 0 < t < T}. has a unique solution in the class defined by u 2 C(DT ). Assume u 2 C(DT ). 217. k¿ ) · sup y2Rn u(y. t) : = u(x. t) · sup y2Rn u(y. t) · max ST u(x. ut. See [10]. where ST. DT = {(x. where M and a are positive constants. There is an ² > 0 such that 4a(T + ²) < 1.2. t) · max ST. Proposition 6. t) − μK(ix. . where N = T/¿ . T + ² − t) 160 CHAPTER 6. since μK > 0. . we obtain @ @t vμ −4vμ = ut −4u · 0. pp. we have the following result which is a corollary of the previous theorem. Proof of Proposition 6. u(x.N − 1.

t) · max ST. Theorem 6.½ it is vμ(x. 0 · t · T. (i) One proof is a consequence of the following lemma: Let A.½ vμ(x. MAXIMUM PRINCIPLE 161 Letting μ ! 0. Proof. Set Lu = Xn i. Assume u 2 C(DT ). t) · Mea|x|2 − μ (4¼(T + ² − t))−n/2 e½2/(4(T+²−t)) · Mea(|y|+½)2 − μ (4¼(T + ²))−n/2 e½2/(4(T+²)) · sup z2Rn f(z) for all ½ > ½0(μ). 2 The above maximum principle of Theorem 6. B real. that ut. Summarizing. t) − μ (4¼(T + ² − t))−n/2 · sup z2Rn f(z). 6.½ vμ(x. t)³i³j ¸ 0 for all ³ 2 Rn. t) 2 DT . Xn i. and ut − Lu · 0 in DT . we obtain the assertion of the proposition. 0) · sup z2Rn f(z). t) = max ST u. t) · sup z2Rn f(z) if ½ > ½0(μ). t) − μ (4¼(T + ² − t))−n/2 it follows u(y. 0) · u(x. where aij 2 C(DT ) are real. Nonnegative means that all eigenvalues are nonnegative.3.3. aij = aji. Then max DT u(x. we have max ST. uxixk exist and are continuous in DT .vμ(x. and (x. Since vμ(y.2 holds for a large class of parabolic differential operators. symmetric and nonnegative matrices. We recall that 4a(T + ²) < 1. t) · sup z2Rn f(z) if ½ > ½0(μ). that is.j=1 aij(x. Then trace (AB) ´ . Thus vμ(y. of ST. On the cylinder part |x − y| = ½. t) = u(y. t)uxixj . ½0 sufficiently large. and the matrix (aij) is nonnegative. even for degenerate equations.j=1 aij(x.

0) = c0(x) x 2 (6.1). . PARABOLIC EQUATIONS 6. x = UT (x−x0)y and v(y) = u(x0 + Uy. (6. Consequently Xn i. .j=1 uxixj ³i³j = Xn i=1 vyiyi´2 i. (6. t0)). Remark. then 0· Xn i.Pn i.5. t0).8) @v @n = 0 on @. c(x.7). Set ³ = U´. 6.4. (6. t0)uxixj (x0.6) @c @n = 0 on @ × (0.5) c(x.10) Assume is bounded and @ sufficiently regular. t) is the concentration of a substance in a solution. 2 162 CHAPTER 6.7) Here is ½ Rn.4 Initial-boundary value problem Consider the initial-boundary value problem for c = c(x.j=1 aij(x0.j=1 aijbij ¸ 0.1) (6. . . n the exterior unit normal at the smooth parts of @. t) = v(x)w(t) leads to the eigenvalue problem. then the eigenvalues . see an exercise. see the arguments of Section 4. zn).j=1 aij(x0. we assume that there is no flow through the boundary.j(x0. First Fick’s rule says that w = D@c/@n. It follows ¸i ¸ 0 and vyiyi · 0 for all i. Thus according to the Neumann boundary condition (6. D a positive constant and c0(x) a given function. In application to diffusion problems. c0(x) its initial concentration and D the coefficient of diffusion.1 Fourier’s method Separation of variables ansatz c(x. where w is the flow of the substance through the boundary @. where zl is an orthonormal system of eigenvectors to the eigenvalues ¸l of the matrix A = (ai.9) and to the ordinary differential equation w0(t) + ¸Dw(t) = 0. t0) = Xn i=1 ¸ivyiyi · 0. t) ct = D4c in × (0. t0)³i³j = Xn i=1 ¸i´2 i 0¸ Xn i. −4v = ¸v in (6. (ii) Another proof exploits transform to principle axis directly: Let U = (z1.

Let vj(x) be a complete system of orthonormal (in L2()) eigenfunctions. (6.5)-(6.4. ! 1. x3. t) = 1X n=0 Cne−D( ¼ l n)2 t cos ³¼ l nz ´ . t) is independent of x1 and x2. where Cj are arbitrary constants. with Cj = Z c0(x)vj(x) dx. see Figure 6. 6. 0) = c0(z) cz = 0. where C0 = .of (6.8) and c(x. 164 CHAPTER 6. cN(x. is a formal solution of the initial-boundary value problem (6.8).4: Diffusion in a tube tion c0(x1. x2. Solutions of (6. t) := 1X j=0 Cje−D¸j tvj(x).10) are wj(t) = Cje−D¸j t.7). z = 0.4. t) := XN j=0 Cje−D¸j tvj(x) is a solution of the differential equation (6. x2.9) are countable and 0 = ¸0 < ¸1 · ¸2 · . then the above initial-boundary value problem reduces to ct = Dczz c(z. x3) of the substrate in a solution is constant if x3 = const. . Assume the initial concentraW l Q x x x 3 1 2 Figure 6. The (formal) solution is c(z. . PARABOLIC EQUATIONS Set z = x3. It follows from a uniqueness result below that the solution of the initialboundary value problem c(x1. Diffusion in a tube Consider a solution in a tube. INITIAL-BOUNDARY VALUE PROBLEM 163 According to the superposition principle. z = l.

6. for example of a call or put option. 0) = 0 @c @n = 0 on @ × (0. for example.4.7) are uniquely determined since from ct = D4c in × (0.5.5)(6. it follows that for each ¿ > 0 0= Z¿ 0 Z (ctc − D(4c)c) dxdt = Z Z¿ 0 1 2 @ @t (c2) dtdx + D Z Z¿ 0 |rxc|2 dxdt = 1 2 Z c2(x. 6.5 Black-Scholes equation Solutions of the Black-Scholes equation define the value of a derivative. An asset 6. which is based on an asset. In principle. BLACK-SCHOLES EQUATION 165 can be a stock or a derivative again. n ¸ 1.1 l Zl 0 c0(z) dz Cn = 2 l Zl 0 c0(z) cos ³¼ l nz ´ dz.2 Uniqueness Sufficiently regular solutions of the initial-boundary value problem (6. ¿ ) dx + D Z Z¿ 0 |rxc|2 dxdt.1) c(x. for example n-th derivatives. there are infinitely many such products. The Black- .1).

d1 = ln(S/E) + (r + ¾2/2)(T − t) ¾pT − t . 166 CHAPTER 6. T) = max{S − E. 0} (6. see [26]. (6.14) is explicitly known and is given by C(S. see below. condition (6. d2 = ln(S/E) + (r − ¾2/2)(T − t) ¾pT − t . uniformly in t.4 (Black-Scholes formula for European call options). E is the exercise price and T the expiry. where V (S. PARABOLIC EQUATIONS S t T C=max{S−E. t) = 0 (6.14) means that we buy assets if its value becomes large. 0 · S < 1. r are positive constants. t). More precisely. 0 · t · T.5: Side conditions for a call option where N(x) = 1 p2¼ Zx −1 e−y2/2 dy. = {(S.5. condition (6. Proof.14) where E and T are positive constants. t). t) := C(S. of the initial-boundary value problem (6. see Figure 6.12) C(0. t) of a derivative is Vt + 1 2 ¾2S2VSS + rSVS − rV = 0 in . where C(S. then V (S(t).11)-(6.11): C(S.13) C(S. 0 < T < 1.Scholes equation for the value V (S. t) is the value of the (European) call option. The solution C(S. and ¾. for example. Side condition (6. t) is the solution of an appropriate initial-boundary value problem for the Black-Scholes equation. t) = SN(d1) − Ee−r(T−t)N(d2). Theorem 6. Substitutions . If S(t) is the value of an asset at time t. where the side conditions are indicated. The Black-Scholes equation follows from Ito’s Lemma under some assumptions on the random function associated to S(t). ¾ is the volatility of the underlying asset S. 0 < t < T}. t) 2 R2 : 0 < S < 1.0} C=0 C~S Figure 6. t) = S + o(S) as S ! 1. (6.11) where for a fixed T.13) says that it makes no sense to buy assets if the value of the asset is zero. r is the guaranteed interest rate of a risk-free investment.12) means that the value of the option has no value at time T if S(T) · E. t) is the value of the derivative at time t. Call option Here is V (S. In this case we have following side conditions to (6.

(6.16) For a solution of (6. (6.15) we make the ansatz v = e®x+¯¿u(x. with u0(x) = max n e(k+1)x/2 − e(k−1)x/2.15) where k= r ¾2/2 .11) to v¿ = vxx + (k − 1)vx − kv. A solution of this initial value problem is given by Poisson’s formula u(x. t = T − ¿ ¾2/2 .15). −1 < x < 1. then u¿ = uxx. where I1 = 1 p2¼ Z 1 −x/(p2¿) e(k+1)(x+qp2¿)e−q2/2 dq I2 = 1 . ¿ ) = e−(k−1)x/2−(k+1)2¿/4u(x. we get u(x. Changing variable by q = (s − x)/(p2¿ ). Set ¯ = ®2 + (k − 1)® − k and choose ® such that 0 = 2® + (k − 1). 6. (6.S = Eex. 0 o . Thus v(x. ¿ > 0 u(x. ¿ ). 0) = u0(x). we get ¯u + u¿ = ®2u + 2®ux + uxx + (k − 1)(®u + ux) − ku. Initial condition (6.17) where u(x. BLACK-SCHOLES EQUATION 167 where ® and ¯ are constants which we will determine as follows.15) implies v(x. ¿ ) is a solution of the initial value problem u¿ = uxx. ¿ ).5. 0}. Inserting the ansatz into differential equation (6. ¿ ) = 1 2p¼¿ Z +1 −1 u0(s)e−(x−s)2/(4¿) ds. C = Ev(x. 0) = max{ex − 1. ¿ ) = 1 p2¼ Z +1 −1 u0(qp2¿ + x)e−q2/2 dq = I1 − I2. ¿ ) change equation (6.

in −1 < x < 1. Assume there are two solutions of (6. Then the solution of u¿ = uxx.p2¼ Z 1 −x/(p2¿) e(k−1)(x+qp2¿)e−q2/2 dq. t = T − 2¿/(¾2). t) = 0. Thus we have the growth property |u(x. Uniqueness.18). see an exercise.17) of v(x. see for example [10]. t). ¿ = ¾2(T − t)/2 and C = Ev(x. it follows that |W(S. pp.2 of this chapter.14).11) and the side conditions W(S.4. Combining the formula for u(x. W(0.18) with positive constants M and a. 0 · ¿ · ¾2T/2. ¿ ).14).5. T) = 0. 2. where P(S. PARABOLIC EQUATIONS where d1 = x p2¿ + 1 2 (k + 1)p2¿ d2 = x p2¿ + 1 2 (k − 1)p2¿ N(di) = 1 p2¼ Z di −1 e−s2/2 ds. 0 · t · T. t) := P(S.11). That is. where S = Eex. with the initial condition u(x. t) = O(S) as S ! 1 uniformly in 0 · t · T. t) satisfies the differential equation (6. From the definition of u we see that u(x. see Proposition 6. 206. ¿ ) ´ 0.12)-(6. In general. The constant c is independent on S and t. From a maximum principle consideration. (6. (6. ¿ ) and the previous settings x = ln(S/E). W(S. the solution u of the initial value problem for the heat equation is not uniquely defined. ¿ )| · Mea|x|. u(x. An elementary calculation shows that I1 = e(k+1)x/2+(k+1)2¿/4N(d1) I2 = e(k−1)x/2+(k−1)2¿/4N(d2). t) is the value of the (European) put . ¿ ). x 2 R. 0) = 0 is uniquely defined in the class of functions satisfying the growth condition (6. t). we get finally the formula of of Theorem 6. The uniqueness follows from the growth assumption (6. ¿ ) = 1 E e−®x−¯¿W(S. definition (6. t)| · cS on S ¸ 0. i = 1. 168 CHAPTER 6. then the difference W(S. 2 6. BLACK-SCHOLES EQUATION 169 Put option Here is V (S.

3.21) is explicitly known and is given by P(S. symmetric and nonnegative matrices. Side condition (6. where μ is a constant.4. Solve for given f(x) and μ 2 R the initial value problem ut + ux + μuxxx = 0 in R × R+ u(x.20) P(S. 4. 6. t < T of the initial-boundary value problem (6. Concerning the put-call parity see an exercise.option. t) · sup z2Rn '(z) . (6. t). t) given by Poisson’s formula satisfies inf z2Rn '(z) · u(x. P Prove that trace (AB) ´ n i. . In this case we have following side conditions to (6.19)-(6. un(s. d2 are the same as in Theorem 6.1). B are real.19) means that the value of the option has no value at time T if S(T) ¸ E. The formula for the put option follows by the same calculations as in the case of a call option or from the put-call parity C(S. Hint: Concerning (ii). 2 170 CHAPTER 6.19) P(0. t) be a positive solution of ut = μuxx. 2. d1. t) = Ee−r(T−t) (6. PARABOLIC EQUATIONS 6. uniformly in 0 · t · T. t > 0. b] . for a heuristic argument which leads to the formula for the put-call parity. t) is a solution of the heat equation ut −4u = 0 in Rn × (0.11). pp. Proof.21) means that it makes no sense to sell assets if its value becomes large. t) = Ee−r(T−t)N(−d2) − SN(−d1) where N(x). t) − P(S. (6. Nonnegative means that all eigenvalues are nonnegative. condition (6. Theorem 6. t > 0. t) are solutions of ut = uss. provided '(x) is continuous and bounded on Rn. 0) = f(x) . Show by using Poisson’s formula: (i) Each function f 2 C([a. Let u(x.j=1 aijbij ¸ 0. 40. Show that Qn k=1 uk(xk. See also [26].. Let A. 5. Show that µ := −2μux/u is a solution of Burger’s equation µt + µµx = μµxx. 0 < S < 1. T) = max{E − S. The solution P(S. t).5 (Black-Scholes formula for European put options).20) says that it makes no sense to sell assets if the value of the asset is zero. condition (6. (ii) In (i) we can choose polynomials fn (Weierstrass’s approximation theorem).6 Exercises 1. 0} (6. t) = o(S) as S ! 1. replace the kernel K = exp(−|y−x|2 4t ) by a sequence of Taylor polynomials in the variable z = −|y−x|2 4t . Show that the solution u(x. Assume u1(s.. t) = S − Ee−r(T−t) and from N(x) + N(−x) = 1.11): P(S.. .21) Here E is the exercise price and T the expiry. b]) can be approximated uniformly by a sequence fn 2 C1[a.

Then trace C = Pn l=1 μl(C). Show that the comparison principle implies the maximum principle.2). t) on DT .. μn(C) are the eigenvalues of a real symmetric n×nmatrix.. v2 satisfy (v1)t −4v1 · f(x... Then X=U 0 BB@ p¸1 0 · · · 0 0 p¸2 · · · 0 .. ... v2 2 C2(DT ) \ C(DT ).. 0 0 · · · ¸n 1 CCA ... . u is a solution of the heat equation and u satisfies the regularity assumptions of the maximum principle (Theorem 6. .. Then ut −4u · vt −4v in DT u · v on ST imply that u · v in DT . Assume is bounded. We recall that UTBU = 0 BB@ ¸1 0 · · · 0 0 ¸2 · · · 0 . . zn). (iii) Let μ1(C)... (ii) trace (QR) =trace (RQ)... . 9.. 11. where zl is an orthonormal system of eigenvectors to the eigenvalues ¸l of the matrix B. t) on ST . EXERCISES 171 is a square root of B. Assume u.. . Prove the following comparison principle: Assume is bounded and u. . . 12.. 0 0 · · · p¸n 1 CCA UT 6. 10. . Á are given.. + cnn)¸n−1 + . + μn)¸n+1 + . .. and u is a solution of the previous boundary-initial value problem and v1. v satisfy the regularity assumptions of the maximum principle.6. . Prove uniqueness in the class u... .. ut. .... . .. where f. t) · v2(x. which follows from the fundamental lemma of algebra: det (¸I − C) = ¸n − (c11 + . t) · (v2)t −4v2 in DT v1 · Á · v2 on ST ... ´ (¸ − μ1) · . Show by using the comparison principle: let u be a sufficiently regular solution of . 7. . 172 CHAPTER 6. ... uxixk 2 C(DT ). PARABOLIC EQUATIONS Show that (inclusion theorem) v1(x. · (¸ − μn) = ¸n − (μ1 + ..Hint: (i) Let U = (z1. t) · u(x.. 8.. Consider the boundary-initial value problem ut −4u = f(x. t) in DT u(x. v1... t) = Á(x. Show that u achieves its maximum and its minimum on ST .

0 · t · T. 17. uniformly on 0 · t · T. (ii) V (0. 6. x 2 (0.ut − 4u = 1 in DT u = 0 on ST . t)uxixj + Xn i bi(x. Prove that a solution V (S. Prove the put-call parity for European options C(S. T) = 0. (i) Find the solution of the diffusion problem ct = Dczz in 0 · z · l.11) in under the side conditions (i) V (S. Prove the Black-Scholes Formel for an European put option. t) = S − Ee−r(T−t) by using the following uniqueness result: Assume W is a solution of (6. Then W(S. 15. under the boundary conditions cz(z. T) = 0. EXERCISES 173 16.3 for the case Lu = Xn i. t) = S+o(S) as S ! 1. then 0 · u(x. 0 · t · T. t). is a solution of the initial-boundary value problem ut = uxx. D = const. (iii) V (S. > 0. W(0. 19. 13. 174 CHAPTER 6. Discuss the result of Theorem 6. t) = 1X n=1 cne−n2t sin(nx). 0) = f(x). if f 2 C4(R) is odd with respect to 0 and 2¼-periodic. t) = 0. T) = 0. t) − P(S. t) of the initial-boundary value problem (6. u(x. Show that u(x. if 0 · z · h 0 if h < z · l. Prove that a solution V (S.11) in .6.j=1 aij(x. t > 0. t) = 0. t) = 0 if z = 0 and z = l and with the given initial concentration c(z. t)uxi + c(x. t) = 0 uniformly in 0 · t · T. 0 · t < 1. 18.11) under the side conditions W(S. t). t) = 0. (ii) Calculate limt!1 c(z. uniformly on 0 · t · T. t)| · cS for all S ¸ 0 and 0 · t · T. t) = 0 and W(S. S ¸ 0. t) = O(S) as S ! 1. t)u(x. (iii) limS!1 V (S. u(¼. PARABOLIC EQUATIONS Chapter 7 . (ii) V (0. is uniquely determined in the class C2() \ C(). ¼). under the side conditions (i) V (S. 0) = c0(z) := ½ c0 = const. satisfies |V (S. Hint: Put-call parity. 14. u(0. t) = 0. S ¸ 0. t) · t in DT . t) ´ 0. t) of the initial-boundary value problem (6. where cn = 2 ¼ Z¼ 0 f(x) sin(nx) dx.

Set r = |x − y|. The function s(r) := ( . T stands for the above composition of mappings. c2. Solutions of the Laplace equation are called potential functions or harmonic functions.1 Fundamental solution Here we consider particular solutions of the Laplace equation in Rn of the type u(x) = f(|x − y|).j=1 aijuxixj = 4v. where the matrix A = (aij) is real.j=1 aij(x)uxixj + Xn j=1 bj(x)uxj + c(x)u = f(x). symmetric and positive definite. ELLIPTIC EQUATIONS OF SECOND ORDER where y 2 Rn is fixed and f is a function which we will determine such that u defines a solution if the Laplace equation. then uxi = f0(r) xi − yi r uxixi = f00(r) (xi − yi)2 r2 + f0(r) µ 1 r− (xi − yi)2 r3 ¶ 4u = f00(r) + n−1 r f0(r). is Lu := Xn i. mainly the Laplace equation 4u = 0. Thus a solution of 4u = 0 is given by f(r) = ½ c1 ln r + c2 : n = 2 c1r2−n + c2 : n ¸ 3 with constants c1. Definition. The Laplace equation is called also potential equation.Elliptic equations of second order Here we consider linear elliptic equations of second order. If A is a constant matrix. 7. Set r = |x − y|. x 2 ½ Rn. where v(y) := u(Ty). The general elliptic equation for a scalar function u(x). then a transform to principal axis and stretching of axis leads to Xn i. 175 176 CHAPTER 7.

which is called ±-function. This formula follows from considerations similar to the next section.1: Notations to Green’s identity Theorem 7. where ± is the Dirac distribution. t > 0. see [6] for generalizations. then for each fixed y 2 u(y) = Z @ µ °(x. . The fundamental solution ° satisfies for each fixed y 2 the relation − Z °(x.1.−1 2¼ ln r : n = 2 r2−n (n−2)!n :n¸3 is called singularity function associated to the Laplace equation. y) @u(x) @nx − u(x) @°(x. y) = s(r) + Á(x. Remark. This is the case if is bounded. Here is !n the area of the n-dimensional unit sphere which is given by !n = 2¼n/2/¡(n/2). @ is in C1. this relation can be written by definition as −4x°(x. x y nx W dW Figure 7. is the Gamma function. the function Á which appears in the definition of the fundamental solution and the potential function u considered are sufficiently regular such that the following calculations make sense. A function °(x. Á 2 C2() for each fixed y 2 and u 2 C2(). y) is called fundamental solution associated to the Laplace equation if Á 2 C2() and 4xÁ = 0 for each fixed y 2 . where ¡(t) := Z 1 0 e−½½t−1 d½. REPRESENTATION FORMULA 177 see an exercise. 7. Definition. y)4x©(x) dx = ©(y) for all © 2 C2 0 (). In the language of distribution. Let u be a potential function and ° a fundamental solution. y) @nx ¶ dSx. y) = ±(x − y).2. 7.2 Representation formula In the following we assume that .

See Figure 7.2 for notations. Thus we have to consider Z @½(y) v @u @n dS = Z @ v @u @n dS + Z @B½(y) v @u @n dS Z @½(y) u . Z @½(y) µ v @u @n − u @v @n ¶ dS = 0.Proof. if v is a fundamental solution and u a potential function. Z ½(y) (v4u − u4v) dx = Z @½(y) µ v @u @n − u @v @n ¶ dS 178 CHAPTER 7. v 2 C2(). From Green’s formula. ELLIPTIC EQUATIONS OF SECOND ORDER y n W n r(y) r Figure 7. Set ½(y) = \ B½(y).2: Notations to Theorem 7. for u.1 we obtain. Let B½(y) ½ be a ball.

@v @n dS = Z @ u @v @n dS + Z @B½(y) u @v @n dS.1) (ii) Consider the case n ¸ 3. y)|. then Z @B½(y) u @v . ½ · ½0. (7. C = C(y) = sup x2B½0 (y) |Á(x.2. REPRESENTATION FORMULA 179 From the definition of s(½) we get the estimate as ½ ! 0 Z @B½(y) v @u @n dS = ½ O(½| ln ½|) : n = 2 O(½) : n ¸ 3. 7. We estimate the integrals over @B½(y): (i) ¯¯¯¯¯ Z @B½(y) v @u @n dS ¯¯¯¯¯ ·M Z @B½(y) |v| dS ·M ÃZ @B½(y) s(½) dS + C!n½n−1 ! . where M = M(y) = sup B½0 (y) |@u/@n|.

n = 2. (7.@n dS = 1 !n Z @B½(y) u 1 ½n−1 dS + Z @B½(y) u @Á @n dS = 1 !n½n−1 Z @B½(y) u dS + O(½n−1) = 1 !n½n−1 u(x0) Z @B½(y) dS + O(½n−1). for an x0 2 @B½(y). 2 Corollary. we obtain the representation formula of the theorem.2) u(y) = 1 (n − 2)!n Z @ µ 1 rn−2 @u @nx − u . Set Á ´ 0 and r = |x − y| in the representation formula of Theorem 7. = u(x0) + O(½n−1). then u(y) = 1 2¼ Z @ µ ln r @u @nx − u @(ln r) @nx ¶ dSx.1). Combining this estimate and (7.1.

@(r2−n) @nx ¶ dSx. in . In fact.. 2 We recall that a domain 2 Rn is called connected if is not the union of two nonempty open subsets 1. 2 such that 1 \ 2 = .2).3 (Maximum principle). that Dlu(y) exist and 180 CHAPTER 7.3) 7. (7.1. It follows from representation formulas (7. in particular from (7. Then u ´ const.2 (Mean value formula for harmonic functions). Proposition 7.3) where := B½(x) since r = ½ and Z @B½(x) 1 rn−2 @u @ny dSy = 1 ½n−2 Z @B½(x) @u @ny dSy = 1 ½n−2 Z B½(x) 4u dy = 0. Proof. Assume u is harmonic in .3). see an exercise. (7. Proposition 7. Assume u is harmonic in . (7. Proposition 7. A domain in Rn is connected if and only if its path connected. where := 0. Proof.3). Then u 2 C1(). Assume u is harmonic in a connected domain and achieves its supremum or infimum in .1 Conclusions from the representation formula Similar to the theory of functions of one complex variable. Consider the case of the supremum. n ¸ 3.2). ELLIPTIC EQUATIONS OF SECOND ORDER are continuous for all l since one can change differentiation with integration in right hand sides of the representation formulae.2. Then for each B½(x) ½½ u(x) = 1 !n½n−1 Z @B½(x) u(y) dSy. 2 Remark. a function which is harmonic in is even real analytic in . Consider the case n ¸ 3. Let x0 2 such that u(x0) = sup . Proof. we obtain here results for harmonic functions from the representation formula. The assertion follows from (7. We recall that a function u is called harmonic if u 2 C2() and 4u = 0 in . Let 0 ½½ be a domain such that y 2 0.

0 < ½ < ½0 and u(bx) < M.4. Set 1 := {x 2 : u(x) = M} and 2 := {x 2 : u(x) < M}. From the mean value formula. Another example see an exercise. Proposition 7. the set 2 is empty since 1 is open.3: Counterexample Assume u 2 C2() \ C( \ {0}) is a solution of 4u = 0 in u = 0 on @ \ {0}. The set 2 is open since u 2 C2(). BOUNDARY VALUE PROBLEMS 181 If not.3 Boundary value problems Assume ½ Rn is a connected domain. Proof.2. In contrast to this behaviour of the Laplace equation.5) where © is given and continuous on @. Assume is bounded. ELLIPTIC EQUATIONS OF SECOND ORDER x2 1 x1 Figure 7. see Proposition 7.1 Dirichlet problem The Dirichlet problem (first boundary value problem) is to find a solution u 2 C2() \ C() of 4u = 0 in (7. Then u achieves its minimum and its maximum on the boundary @. one has uniqueness if 4u = 0 is replaced by the minimal surface equation @ @x1 Ã p ux1 1 + |ru|2 . it follows M= 1 !n½n−1 Z @B½(x) u(x) dS < M !n½n−1 Z @B½(x) dS = M. which is a contradiction. Assume is connected and bounded.u(x) =: M.3. 7. This problem has solutions u ´ 0 and u = Im(z +z−1). (7.5) one point from the the boundary as the following example shows. 2 Corollary. The previous corollary fails if is not bounded as simple counterexamples show. The set 1 is not empty since x0 2 1. Remark.4) u = © on @. where z = x1 +ix2. then a solution to the Dirichlet problem is uniquely determined. then there is a ½0 > 0 such that B½0(x) ½ and u(x) = M for all x 2 B½0(x). 7. Maximum principle. 7. then there exists ½ > 0 and bx such that |bx − x| = ½. 2 is empty if we can show that 1 is open. Consequently. Let x 2 1. and u 2 C2() \ C() is harmonic in . Remark. Let ½ R2 be the domain = {x 2 B1(0) : x2 > 0}.3. The previous result fails if we take away in the boundary condition (7. Thus. 182 CHAPTER 7.

7.2 Neumann problem The Neumann problem (second boundary value problem) is to find a solution u 2 C2() \ C1() of 4u = 0 in (7. (7. Proposition 7.e © and h are given and continuous on @.2. Hint: Multiply the differential equation 4w = 0 by w and integrate the result over . Since ° = s+Á. GREEN’S FUNCTION FOR 4 183 Another proof under the weaker assumption u 2 C1() \ C2() follows from the Hopf boundary point lemma. Proof.8) @u @n + hu = © on @.! + @ @x2 Ã p ux2 1 + |ru|2 ! = 0.6. Assume is bounded and sufficiently regular.10) where °(y. Proof. we try to find a Á such that u satisfies also the boundary condition. Exercise. (7. Proposition 7.4 Green’s function for 4 Theorem 7. Hint: Multiply the differential equation 4w = 0 by w and integrate the result over . x) is a fundamental solution.4. u does not satisfies the boundary condition in the above boundary value problems. x) @u(y) @ny − u(y) @°(y. x) @ny ¶ dSy.1 says that each harmonic function satisfies u(x) = Z @ µ °(y.5. Assume is bounded. Exercise.6) @u @n = © on @. see Section 7. 7. where Á is an arbitrary harmonic function for each fixed x. for example.3. . 7.9) where © and h are given and continuous on @. see Lecture Notes: Linear Elliptic Equations of Second Order. In general. then a solution to the mixed problem is uniquely determined in the class u 2 C2() provided h(x) ¸ 0 on @ and h(x) > 0 for at least one point x 2 @. then a solution to the Dirichlet problem is in the class u 2 C2() uniquely determined up to a constant.3 Mixed boundary value problem The Mixed boundary value problem (third boundary value problem) is to find a solution u 2 C2() \ C1() of 4u = 0 in (7.3.7) where © is given and continuous on @. 7. (7.

and some additional assumptions. is called Green function associated to and to the Dirichlet problem (7. (A) G(x.5). x 6= y. y) < s(|x − y|). ELLIPTIC EQUATIONS OF SECOND ORDER Suppose that u achieves its boundary values © of the Dirichlet problem. 4yG(y. x(2)) are harmonic in \ ¡ B1 [ B2 ¢ we obtain from Green’s identity. x) = 0 if y 2 @.7. Definition. (7. we define a Green function as follows.5) if for fixed x 2 . y. see Figure 7.4). x) @ny u(y) dSy. Proof. x 2 .Consider the Dirichlet problem. x 6= y. 2. that is we consider G(y. Remark. x) which satisfies (7. 7. x 2 . x 6= y. x) = 0.7 0= Z @(\(B1[B2)) µ . then u(x) = − Z @ @°(y. x 6= y. x) @ny ©(y) dSy. x(1)) and G(y.4. GREEN’S FUNCTION FOR 4 185 n nn x x(1) (2) Figure 7. x) as a function of y. Concerning the existence of a Green function for more general domains see [13]. A Green function has the following properties. (7. (B) 0 < G(x.4: Proof of Proposition 7. x) − s(|x − y|) 2 C2() \ C().12) We claim that this function solves the Dirichlet problem (7. A function °(y. A function G(y. More precisely.. x) 2 C2( \ {x}) \ C( \ {x}). the following properties hold: (i) G(y. (7. (iii) G(y. x) (symmetry). Set Bi = B½(x(i)). is called Green’s function. x). (7. It is an interesting fact that we get from (i)-(iii) of the above definition two further important properties.11) Then u(x) = − Z @ @°(y.4 for notations. i = 1. x. We assume Bi ½ and B1 \ B2 = . We will see in the next section that a Green function exists at least for some domains of simple geometry. We assume that is bounded. Proposition 7. (A) Let x(1). y 2 .4). In the case n = 2 we assume diam < 1. y) = G(y. then we look for a Á such that °(y. sufficiently regular and connected. x(2) 2 .11). 184 CHAPTER 7. (ii) G(y. y 2 @. x) = 0. x 2 . Since G(y.

x(2)) @ @ny G(y. x(1)) ¶ dSy. and Z @B1 µ G(y. x(2)) @ @ny G(y. x(2)) − G(y. x(1)) ¶ dSy + Z @B1 µ G(y. x(2)) −G(y. x(1)) @ . x(1)) ¶ dSy + Z @B2 µ G(y. x(1)) @ @ny G(y. The integral over @ is zero because of property (iii) of a Green function. x(2)) − G(y. x(1)) @ @ny G(y. x(2)) @ @ny G(y.G(y. x(1)) @ @ny G(y. x(1)) @ @ny G(y. x(2)) − G(y. x(1)) ¶ dSy = Z @ µ G(y. x(2)) @ @ny G(y.

x(1)) as ½ ! 0. There is a sufficiently small ½0 > 0 such that for each ½. x(1)) ¶ dSy ! −G(x(2). x) > 0 on \ B½(x). x(1)) @ @ny G(y. It remains to show that G(y.1. Let = BR(0) be a ball in Rn with radius R and the center at the origin. ELLIPTIC EQUATIONS OF SECOND ORDER and G(y. y) = s(r) − s ³½ R r1 ´ .@ny G(y. see Figure 7. Z @B2 µ G(y. x) < s(|x − y|). y 2 . 2 7. x(2)) − G(y. x) = 0 if y 2 @ it follows from the maximum-minimum principle that G(y. Set G(x. y 2 BR(0) and let y0 the reflected point of y on the sphere @BR(0). in particular |y||y0| = R2. then Green’s function is explicitly known. x(1)) ¶ dSy ! G(x(1). Let x. x 6= y.1. 0 < ½ < ½0.4. x) < 0 if y 2 @. Since 4yG(y. r = |x − y| and ½2 = Xn . G(y. x(2)) @ @ny G(y. x) 186 CHAPTER 7. Thus. x) > 0.1 Green’s function for a ball If = BR(0) is a ball. x 6= y. see property (iii) of a Green function. From the definition of s(|x − y|) it follows that Á(y. x) < 0 for all y. x 2 . x) > 0 for all y 2 B½(x). x) > 0 if y 2 @B½(x) G(y. x 6= y. x. x(2)). x) = 0 if y 2 @ and x 2 we have for y 2 @ Á(y. x) = −s(|x − y|). since 4yÁ = 0 in . x(2)) − G(y. Consequently G(y. x(2)) @ @ny G(y.5 for notations. x. Fix x 2 and let B½(x) be a ball such that B½(x) ½ for all 0 < ½ < ½0. where s is the singularity function of Section 7. x) = 0 in \ B½(x) G(y. y 2 . that is. x) = s(|x − y|) + Á(y. This follows by considerations as in the proof of Theorem 7. the maximum-minimum principle implies that Á(y. (B) Since G(y.

y) = 1 R!n R2 − |x|2 |y − x|n . see [13].4). y)© dSy is a solution of the Dirichlet problem (7. . − @ @ny G(x.i=1 y2 i . (7.4).4. Then u(x) = Z @BR(0) H(x.5). r1 = Xn i=1 µ xi − R2 ½2 yi ¶2 .5: Reflection on @BR(0) This function G(x. (7. 7.5) in the class C2() \ C().2. This formula is also true for a large class of domains ½ Rn. Assume © 2 C(@). y) ¯¯¯¯ |y|=R = 1 R!n R2 − |x|2 |y − x|n . Exercise. y) satisfies (i)-(iii) of the definition of a Green function. We claim that u(x) = − Z @BR(0) @ @ny G(x. (7. Proof. y)©(y) dSy is the solution of the first boundary value problem (7.13) which is called Poisson’s kernel. Theorem 7. Lemma. GREEN’S FUNCTION FOR 4 187 0y r R y’ r1 x Figure 7. Set H(x.

Fix ³ 2 @BR(0) and let x 2 BR(0). ELLIPTIC EQUATIONS OF SECOND ORDER Proof. |y| = R.6 for notations. y) dSy = 1. Then u(x) − ©(³) = Z @BR(0) H(x.13) or from H=− @G(x. y) > 0. y) = G(y. The proof follows from following properties of H(x. |y−³| > ±. for each harmonic function u. (iii) R @BR(0) H(x. |y| = R. y) < ² 2M!nRn−1 if x and y satisfy |x−³| < ±0. We obtain (ii) if we set u ´ 1. then limx!³. Property (iii) is a consequence of formula u(x) = Z @BR(0) H(x. y) = 0. (iv) and (v) follow from the definition (7. y) 2 C1. Thus R x . |y| = R. is in C(BR(0)) and that u achieves the prescribed boundary values. |x| < R. |x| < R.4. x 6= y. x). |x| < R. |y−³|¸± H(x. (v) Fix ³ 2 @BR(0) and ± > 0.13) of H and (ii) from (7. see calculations to the representation formula above. |y−³|<± H(x.188 CHAPTER 7. |y − ³| > ±. 7. |x| < R. y): (i) H(x.|x|<R H(x. It follows |I1| · ² because of (iii) and (iv). (i). y) (©(y) − ©(³)) dSy I2 = Z @BR(0). y)u(y) dSy. y) = 0 uniformly in y 2 @BR(0). GREEN’S FUNCTION FOR 4 189 For given (small) ² > 0 there is a ± = ±(²) > 0 such that |©(y) − ©(³)| < ² for all y 2 @BR(0) with |y − ³| < ±. see Figure 7. (iv) H(x. Set M = max@BR(0) |Á|. y) @ny ¯¯¯¯ y2@BR(0) . G harmonic and G(x. It remains to show that u. where I1 = Z @BR(0). y) (©(y) − ©(³)) dSy. From (v) we conclude that there is a ±0 > 0 such that H(x. (ii) 4xH(x. given by Poisson’s formula. y) (©(y) − ©(³)) dSy = I1 + I2.

pp.2 as u(x) = R2 − |x|2 !nR Z @BR(0) ©(y) 1 (|x|2 + R2 − 2|x|Rcos ±)n/2 dSy. 2 Remark. 246.4.7. G(z. 190 CHAPTER 7. II. E Figure 7. ¼] through cos ± = x · y/(|x||y|). or [4]. Then the Green function of is. Define ± 2 [0. see [16].2 Green’s function and conformal mapping For two-dimensional domains there is a beautiful connection between conformal mapping and Green’s function.6: Proof of Theorem 7. Let w = f(z) be a conformal mapping from a sufficiently regular connected domain in R2 onto the interior of the unit circle. Vol. where z = x1 + ix2. see for examw= f(z) z w x y 1.2 |I2| < ² and the inequality |u(x) − ©(³)| < 2² for x 2 BR(0) such that |x − ³| < ±0 is shown. p.dz d’ Figure 7. . then we write Poisson’s formula of Theorem 7. W. z0) = 1 2¼ ln ¯¯¯¯¯ 1 − f(z)f(z0) f(z) − f(z0) ¯¯¯¯¯ . 18 for an easy proof of this formula. z0 = y1 + iy2.7: Conformal mapping ple [16] or other text books about the theory of functions of one complex variable. 7. since R2 − |x|2 |x| + R2 − 2|x|Rcos ± = 1 − ½2 ½2 − 2½ cos ± + 1 =1+2 1X n=1 ½n cos(n±). see Figure 7. ELLIPTIC EQUATIONS OF SECOND ORDER In the case n = 2 we can expand this integral in a power series with respect to ½ := |x|/R if |x| < R.

n = 2. n ¸ 3.14) u = 0 on @. We recall that !n = |@B1(0)|. We need the following lemma concerning volume potentials. (i) The first assertion follows since we can change differentiation with integration since the differentiate integrand is weakly singular. To simplify the presentation. Let B½ be a fixed ball such that x 2 B½. V (x) = Z f(y) 1 |x − y|n−2 dy and set in the two-dimensional case V (x) = Z f(y) ln µ 1 |x − y| ¶ dy. (7.7. (i) Assume f 2 C(). Proof. Let for x 2 Rn. x 2 . see an exercise. Then V 2 C1(Rn) and Vxi(x) = Z f(y) @ @xi µ 1 |x − y|n−2 ¶ dy.5. . Vxi(x) = Z f(y) @ @xi µ ln µ 1 |x − y| ¶¶ dy if n = 2. n ¸ 3 4V = −2¼f(x). if n ¸ 3. we consider the case n = 3. We assume that is bounded and sufficiently regular such that all the following integrals exist. (ii) If f 2 C1(). then V 2 C2() and 4V = −(n − 2)!nf(x).5 Inhomogeneous equation Here we consider solutions u 2 C2() \ C() of −4u = f(x) in (7. (ii) We will differentiate at x 2 . See [6] for generalizations concerning these assumptions.15) 7. INHOMOGENEOUS EQUATION 191 where f is given. x 2 . Lemma.

according to (i) and since we have the identity @ @xi µ 1 |x − y| ¶ =− @ @yi µ 1 |x − y| ¶ which implies that f(y) @ @xi µ 1 |x − y| ¶ =− @ @yi µ f(y) 1 |x − y| ¶ + fyi(y) 1 |x − y| .½ sufficiently small such that B½ ½ . 192 CHAPTER 7. Then. ELLIPTIC EQUATIONS OF SECOND ORDER we obtain Vxi(x) = Z f(y) @ @xi µ 1 |x − y| ¶ dy = Z \B½ f(y) @ @xi µ 1 |x − y| .

¶ dy + Z B½ f(y) @ @xi µ 1 |x − y| ¶ dy = Z \B½ f(y) @ @xi µ 1 |x − y| ¶ dy + Z B½ µ − @ @yi µ f(y) 1 |x − y| ¶ + fyi(y) 1 |x − y| ¶ dy = Z \B½ f(y) @ @xi µ 1 |x − y| ¶ dy + Z B½ fyi(y) 1 .

x 6= y. Now we choose for B½ a ball with the center at x.|x − y| dy − Z @B½ f(y) 1 |x − y| ni dSy. It is I1 = O(½) as ½ ! 0 and . Because of 4x(|x − y|−1) = 0. It follows that the first and second integral is in C1(). The second integral is also in C1() according to (i) and since f 2 C1() by assumption. then 4V = I1 + I2. where I1 = Z B½(x) Xn i=1 fyi(y) yi − xi |x − y|3 dy I2 = − Z @B½(x) f(y) 1 ½2 dSy. it follows 4V = Z B½ Xn i=1 fyi(y) @ @xi µ 1 |x − y| ¶ dy − Z @B½ f(y) Xn i=1 @ @xi µ 1 |x − y| ¶ ni dSy. We recall that n · (y − x) = ½ if y 2 @B½(x). where n is the exterior unit normal at @B½.

INHOMOGENEOUS EQUATION 193 for a y 2 @B½(x) I2 = − 1 ½2 f(y) Z @B½(x) dSy = −!nf(y). Then u(x) = Z G(x. where x 2 . y).4).3. y) = 1 4¼|x − y| + Á(x. ELLIPTIC EQUATIONS OF SECOND ORDER where I1(x) = Z \B½(x0) G(x. Since G(x.15). we obtain from the above lemma that 4u = 1 4¼4 Z f(y) 1 |x − y| dy + Z 4xÁ(x. which implies that lim½!0 I2 = −!nf(x). Assume f 2 C1() \ C(). where Á is a potential function with respect to x or y. y)f(y) dy = −f(x). Set u(x) = I1 + I2. (7. which is the case if is a ball. I2(x) = Z \B½(x0) . 2 In the following we assume that Green’s function exists for the domain . for fixed x0 2 @ we will prove that lim x!x0.5). Proof. y)f(y) dy. 194 CHAPTER 7. (7.5. y)f(y) dy is the solution of the inhomogeneous problem (7. It remains to show that u achieves its boundary values. We will show that u(x) := Z G(x. For simplicity of the presentation let n = 3. y)f(y) dy is a solution of (7.for I2 we obtain from the mean value theorem of the integral calculus that 7. That is. x2 u(x) = 0. Theorem 7.14).

Hint: See the proof of the representation formula. Let M = max |f(x)|. if x 2 B½(x0) \ . let f 2 C¸(). x2 I1(x) = 0 since G(x0. For each fixed ½. |I2| · M 4¼ Z \B½(x0) dy |x − y| + O(½2) · M 4¼ Z B2½(x) dy |x − y| + O(½2) = O(½2) as ½ ! 0. y) is uniformly continuous in x 2 B½/2(x0) \ and y 2 \ B½(x0). then V 2 C2. More precisely.¸().8: Proof of Theorem 7.6. y) = 0 if y 2 \ B½(x0) and G(x. y) be a fundamental solution to 4. y). 3. we obtain. 0 < ½ · ½0. Show that − Z °(x. 2. y) = 1 4¼ 1 |x − y| + Á(x. For the proof of (ii) in the above lemma it is sufficient to assume that f is H¨older continuous. y)f(y) dy. bounded and sufficiently regular. Show that |x|−1 sin(k|x|) is a solution of the Helmholtz equation 4u + k2u = 0 in Rn \ {0}. see for example [9]. see Figure 7. Let °(x. is a solution . EXERCISES 195 y x x r0 Figure 7. Consequently for given ² there is a ½0 = ½0(²) > 0 such that |I2| < ² 2 for all 0 < ½ · ½0. 0 < ¸ < 1.8.3 7. y 2 .6 Exercises 1.G(x. Assume u 2 C2(). 7. 2 Remark. y) 4©(x) dx = ©(y) for all © 2 C2 0 () . we have lim x!x0. Since G(x.

0)}. Let u 2 C2(Rn) be a solution of 4u = 0 in Rn satisfying u 2 L2(Rn). Show that u ´ 0 in Rn.. R Rn u2(x) dx < 1. Assume ½ Rn is bounded and u. 4. y) = u(1. 1) satisfying the boundary conditions u(0. multiply the differential equation by ³ := ´2u. ´ ´ 0 in the exterior of B2R(0). e. ELLIPTIC EQUATIONS OF SECOND ORDER 5. R2 Z B2R(0) |u|2 dx. provided k is an integer. 1] and uy(x. Show that a bounded harmonic function defined on Rn must be a constant (a theorem of Liouville). Let u 2 C2() be a solution of 4u = 0 on the quadrangle = (0. 9. Show that u = 0 in . Show that there are at least two solutions. ² > 0. 10. Assume u 2 C2(B1(0)) \ C(B1(0) \ {(1. . Let ® = {x 2 R2 : x1 > 0. 0 < ® · ¼. Prove that u ´ 0 in . 1) × (0. where ´ 2 C1 is a cut-off function with properties: ´ ´ 1 in BR(0). v 2 C2()\C() satisfy 4u = 4v and max@ |u − v| · ² for given ² > 0. Show that u(x) = r ¼ ®k sin ³¼ ® kµ ´ is a harmonic function in ® satisfying u = 0 on @®. 196 CHAPTER 7. Hint: Consider u(x. i. µ) are polar coordinates with the center at (0. Hint: Prove Z BR(0) |ru|2 dx · const. Integrate the product. 0). 0)}) is a solution of 4u = 0 in B1(0) u = 0 on @B1(0) \ {(1. Set = Rn \ B1(0) and let u 2 C2() be a harmonic function in satisfying lim|x|!1 u(x) = 0. Here (r. 0 · ´ · 1. Prove that max |u| = max @ |u| . Show that max |u − v| · ². apply integration by parts and use the formula 2ab · ²a2 + 1 ² b2. 0 < x2 < x1 tan ®}. 0) = uy(x. 8.of 4u = u3 in u = 0 on @. 1) = 0 for all x 2 [0. |r´| · C/R. where c is a constant independent of R. To show this inequality. y) = 0 for all y 2 [0. 7. 1]. 6. y) = 1 − (x2 + y2) (1 − x)2 + y2 .

6. Prove (Harnack’s inequality): an−2(a − |³|) (a + |³|)n−1 u(0) · u(³) · an−2(a + |³|) (a − |³|)n−1 u(0) .2) u(y) = a2 − |y|2 a!n Z |x|=a u(x) |x − y|n dSx for y = ³ and y = 0. 1) and u 2 C2() \ C() a solution of the boundary value problem −4u = 1 in .R. ELLIPTIC EQUATIONS OF SECOND ORDER 15. Let ® = {x 2 R2 : x1 > 0. 1) × (−1.R) of 4u = f in ®. Show that for each solution u 2 C1(®. Assume u 2 C2() satisfies 4u = 0 in . ´ appropriate constants. 0 < x2 < x1 tan ®}. ®. 198 CHAPTER 7. EXERCISES 197 11. Let Á(µ) be a 2¼-periodic C4-function with the Fourier series Á(µ) = 1X n=0 (an cos(nµ) + bn sin(nµ)) . 4u = 0 in Ba(0). Hint: Consider the comparison function v = A(x2 + y2). Find a lower and an upper bound for u(0. u = 0 on @. Let u 2 C2(Ba(0)) \ C(Ba(0)) satisfying u ¸ 0. (b) An appropriate comparison function is v = Ar ¼ ®−² sin(B(µ + ´)) . A. B. u. Let Ba(³) be a ball such that its closure is in . 14. Show that |D®u(³)| · M µ |®|°n a ¶ . R large. 7. B. Show that u= 1X n=0 (an cos(nµ) + bn sin(nµ)) rn solves the Dirichlet problem in B1(0). Then u · v in . Hint: (a) Comparison principle (a consequence from the maximum principle): Assume is bounded. ´ positive. A = const. 12. 0).Hint: Apply the maximum principle to \ BR(0). v 2 C2() \ C() satisfying −4u · −4v in and u · v on @.R) \ C2(®. holds: For given ² > 0 there is a constant C(²) such that |u(x)| · C(²) |x| ¼ ®−² in ®.R \ BR(0). Let be the quadrangle (−1. 13. 0 < ® · ¼. and assume f is given and bounded on ®.R satisfying u = 0 on @®.R = ® \ BR(0). Hint: Use the formula (see Theorem 7.R.

which is a half-space of R3. u ´ 0 in if u ´ 0 in an open subset of .2. EXERCISES 199 20. Show that G(x.r(³). Hint: Use the formula of Theorem 7. . Use the result of the previous exercise to show that u 2 C2() satisfying 4u = 0 in is real analytic in .−y3). Prove that u ´ 0 in if D®u(³) = 0 for all ®.|®| . which is a wedge in R3. Show that G(x. The constant c is the constant in the estimate nn · cenn! which follows from Stirling’s formula. where y = (y1. x3 > 0}. where M = supx2Ba(³) |u(x)| and °n = 2n!n−1/((n − 1)!n).6. y) = 1 4¼|x − y| − 1 4¼|x − y| . 7.−y3). x2. is the Green function to . Assume is connected and u 2 C2() is a solution of 4u = 0 in . 17. where K = cM and r = a/(e°n). Show that G(x. which is half of a ball in R3. Let = {(x1. . . 1. y) = 1 4¼|x − y| − . for a point ³ 2 . is the Green function to . Let = {(x1. where y = (y1. . x2. k = o. y2. y? = R2y/(|y|2) and y? = R2y/(|y|2). y2. y) = 1 4¼|x − y| − R 4¼|y||x − y?| − 1 4¼|x − y| + R 4¼|y||x − y?| . successively to the k th derivatives in balls with radius a(|®| − k)/m. In particular. Let = {(x1. to show that u is in the class CK. 18. x2. See Section 3. x3) 2 R3 : x2 > 0.5 for the definition of a real analytic function. 16. x3) 2 R3 : x21 +x22 +x23 < R2. x3 > 0}.m − 1. x3) 2 R3 : x3 > 0}. 19. Hint: Use Stirling’s formula n! = nne−n µ p2¼n + O µ 1 pn ¶¶ as n ! 1.

23. 22. y)| · c |x − y|® with a constants c and ®. y) is continuous in × where x 6= y. i.−y3). u = Á on @. ® < n. 0 < ® < ¼.. where ´² = ´(|x − y|/²). 26. a ball or a quadrangle for example. Transform this problem into a Dirichlet problem for the Laplace equation. Prove (i) of the lemma of Section 7. 21. Hint: Consider the case n ¸ 3. 0 · ´0 · 2. where y = (y1. Show that F(x) is continuous on . ELLIPTIC EQUATIONS OF SECOND ORDER as ² ! 0. Let for a sufficiently regular domain 2 Rn. is the Green function to . y0 = (y1. Find Green’s function for the exterior of a disk. of the domain = {x 2 R2 : |x| > R}. F(x) = Z K(x. 25. Find Green’s function for the slit domain = {z 2 C : 0 < arg z < 2¼}. ´(t) = 0 for t · 1.5. and which satisfies |K(x. where K(x. e. Find Green’s function for the angle domain = {z 2 C : 0 < arg z < ®¼}. where w(x) := R .−y3). and that @V²(x)/@xi converges uniformly on compact subsets of Rn to Z f(y) @ @xi µ 1 |x − y|n−2 ¶ dy 200 CHAPTER 7.−y2.−y2. ´(t) = 1 for t ¸ 2 and consider for ² > 0 the regularized integral V²(x) := Z f(y)´² dy |x − y|n−2 . Show that V² converges uniformly to V on compact subsets of Rn as ² ! 0. y3) and y0 = (y1. y2. 24.1 4¼|x − y| − 1 4¼|x − y0| + 1 4¼|x − y0| . Hint: Set u = w + v. Consider the inhomogeneous Dirichlet problem −4u = f in . Fix a function ´ 2 C1(R) satisfying 0 · ´ · 1. y) dy.

Leis. Lehrbuch der Theoretischen Physik. I. Translation from Russian. Dover. Leipzig. of Math(2) 88 (1968). Garabedian. Leipzig. Elsevier. Bombieri. Trudinger. [6] L. Berlin. 1964. M. Giusti. 1968. [19] J. Teil IV. New York. Leipzig. Translation from Russian. German translation from Russian. Equilibrium Capillary Surfaces. [16] Z. German translation: Math. Oxford. Kapillarfl¨achen. John. Studies in Advanced Mathematics. New York. Math. Graduate Studies in Mathematics . Miersemann. Ann. [7] R. Z. de Kharkov (2)15. Courant und D. Hirzel. Pergamon Press. Vol. 26 (1927). National Bureau of Standards Applied Mathematics Series.S. Abramowitz and I. Finn. Soc. Berlin. 4. Petrowski. 1967. S¨achs. English translation: Course of Theoretical Physics. and Mathematical tables. Vol. Vol. 1955. Partial Differential Equations. New York. E. New York. Minimal cones and the Bernstein problem. Graphs. 1982. Partial Differential Equations. Verh. 1964. 19. [2] S. 1986. Lions and E. Wiss. 1975.-Natur. G. 1962. Reprinted by Dover. Vorlesungen ¨uber Partielle Differentialgleichungen. 1983. [21] W.. 243–268. AMS. Englisch translation: Lectures on Partial Differential Equations. S. Elliptic Partial Differential Equations of Second Order. English translation: Course of Higher Mathematics. Methoden der Mathematischen Physik. 1. Springer-Verlag. Springer-Verlag. . 1976. 2. Evans. Vol. 224. Chelsia Publishing Company. 1975. 201 202 BIBLIOGRAPHY [10] F. I. Vol. Stegun. Teubner. Simons. Vol. [13] R. 1954. Vol. De Giorgi and E. Inv. Vol. Akad. Vol. 1992. 2. Akademie-Verlag. 1964. A. 7 (1969). [5] L. 1968. Springer-Verlag. [9] D. Springer-Verlag. Paris. Handbook of Mathematical Functions with Formulas. Elsevier. 2008. 1986. 1972. K¨onigsberger. Berlin. F. Lehrgang der H¨oheren Mathematik. Smirnow.. Comm. 130 (2008). 62–105. English translation: Course of Higher Mathematics. Probl´emes aux limites non homog´enes et applications. Vorlesungen ¨uber partielle Differentialgleichungen zweiter Ordnung.. Teil II. Lifschitz. S. Nehari. Introduction to the Calculus of Variations. Sagan. Bibliography [1] M. Heft 4.(1915–1917). 1991. [8] P. New York. Gariepy. Bernstein. Grundlehren. Analysis 2. CRC Press. Math. DC. D. [11] K. 1. Berlin. Grundlehren. Providence. Translation from Russian. U. 1 and Vol. R. Minimal varieties in riemannian manifolds. B. Dunod. Partial Differential Equations. [18] H. Ber. [14] J.-L. Magenes. Smirnow. Wiley-Interscience. Evans and R. 284. Mannheim. [17] I. Wiley-Interscience. 55. 1992.. New York.-Hochschultaschenb¨ucher 165/165a. I. 1993. Berlin. Measure Theory and Fine Properties of Functions.. Math. Reprinted by Dover. Gilbarg and N. Berlin. English translation: Methods of Mathematical Physics. VEB Verlag der Wiss. [15] E. Conformal Mapping. VEB Verlag der Wiss. 551–558. Hilbert. 38–45. 1975. [3] E. Government Printing Office.Washington. C.s(|x − y|)f(y) dy. Band 1 und Band 2.. Boca Raton. Landau and E. Lehrgang der H¨oheren Mathematik. Sur un th´eor`eme de g´eom´etrie et son application aux d´eriv´ees partielles du type elliptique.. Springer-Verlag. 1964. [12] L. Kl. C. [4] R. [20] W.

1996. 1981. 169 boundary condition 14. Cambridge. A Student Introduction. N. Vol. 33. 2008. Functional Analysis. A. 17 first order equations 25 two variables 40 Rn 51 Fourier transform 141 inverse Fourier transform 142 Fourier’s method 126. Dewynne. Partielle Differentialgleichungen. 47. 162 functionally dependent 13 . German translation: Partielle Differentialgleichungen. Cambridge University Press. BIBLIOGRAPHY 203 [24] M. [23] W. A treatise on the Theory of Bessel Functions. 74 characteristic curve 28 characteristic strip 47 classification linear equations second order 63 quasilinear equations second order 73 systems first order 74 cylinder surface 29 diffusion 163 Dirac distribution 177 Dirichlet integral 17 Dirichlet problem 181 domain of dependence 108.[22] A. 15 capillary equation 21 Cauchy-Kowalevskaya theorem 63. 82 uniformly elliptic 73 Euler-Poisson-Darboux equation 111 Euler equation 15. Index d’Alembert formula 108 asymptotic expansion 84 basic lemma 16 Beltrami equations Black-Scholes equation 164 Black-Scholes formulae 165. Grundlehren. 75 nonuniformly elliptic 73 second order 175 system 75. Second edition. E. S. Yosida. Strauss. Berlin. Pseudodifferential operators. Princeton. 41. 123. Vieweg. Sommerfeld. Taylor. [26] P.Watson. An Introduction. The Mathematics of Financial Derivatives. Leipzig. 84 Cauchy-Riemann equations 13 characteristic equation 28. Partial Differential equations.Wilmott. [25] G. Geest & Portig. 1952. 1954. Howison and J. Springer-Verlag. 115 domain of influence 109 elliptic 73. Wiley-Interscience. 1995. [27] K. New Jersey. 1965.

48 integral of a system 28 Jacobi theorem 55 Kepler 56 Laplace equation 13. 129 initial value problem Cauchy 33. 182 Newton potential 13 normal form 69 noncharacteristic curve 33 option call 165 put 169 parabolic equation 151 system 75 . 43 multi-index 90 multiplier 12 Navier Stokes 83 Neumann problem 20. 25 second order 63 maximum principle heat equation 156 parabolic 161 harmonic function 180 Maxwell equations 76 mean value formula 180 minimal surface equation 18 Monge cone 42. 176 Gamma function 176 gas dynamics 79 Green’s function 183 ball 186 conformal mapping 190 Hamilton function 54 Hamilton-Jacobi theory 53 harmonic function 179 heat equation 14 inhomogeneous 155 heat kernel 152.fundamental solution 175. 20 linear elasticity 83 linear equation 11. 153 helicoid 30 hyperbolic equation 107 inhomogeneous equation 117 one dimensional 107 higher dimension 109 system 74 initial conditions 15 initial-boundary value problem uniqueness 134 204 INDEX 205 string 125 membrane 128.

131 wave front 51 volume potential 191 . 147 quasiconform mapping 76 quasilinear equation 11. 83 sound 82 spherical mean 110 strip condition 46 telegraph equation 78 wave equation 14.Picard-Lindel¨of theorem 9 Poisson’s formula 152 Poisson’s kernel 187 pseudodifferential operators 146. 31 real analytic function 90 resonance 134 Riemann’s method 120 Riemann problem 61 Schr¨odinger equation 137 separation of variables 126 singularity function 176 speed plane 78 relative 81 surface 81. 107.