1. Trang chủ >
  2. Công Nghệ Thông Tin >
  3. Kỹ thuật lập trình >

2 Cauchy's Problem for Second-Order PDEs

Bạn đang xem bản rút gọn của tài liệu. Xem và tải ngay bản đầy đủ của tài liệu tại đây (6.6 MB, 489 trang )


8.2 Cauchy’s Problem for Second-Order PDEs



199



n



n 1



N



n



-curves



-curves



Fig. 8.1 Cauchy surface



u.x/ D u0 .x/;



du

Á ru N D d0 .x/;

dn



x 2 †n 1 ;



(8.2)



where u0 .x/ and d0 .x/ are assigned functions on †n 1 corresponding to the values

of u.x/ and its normal derivative on this hypersurface.

A first step in discussing such a problem consists of introducing a convenient

coordinate system . 1 ; : : : ; n /, the Gauss coordinates, in the neighborhood of

†n 1 , in order to simplify its formulation.

Given a Cartesian coordinate system .O; ui / in
xi D ri . 1 ; : : : ;



n 1 /;



i D 1; : : : ; n;



be the parametric equations of the surface †n 1 . In a neighborhood of †n

the system of functions

xi D ri . 1 ; : : : ;



n 1/



C



n Ni



1



in
(8.3)



is considered, where n is the distance of an arbitrary point of the normal from †n 1

(see Fig. 8.1). Accordingly, the condition n D 0 again gives the equation of †n 1 .

The system (8.3) defines a coordinate transformation . 1 ; : : : ; n /

!

.x1 ; : : : ; xn / in a neighborhood of †n 1 , and the Jacobian of (8.3), evaluated

for n D 0, is

0 @x

1

@x1

1

N1

B@ 1

C

@ n 1

B

C

J DB

C:

@ @x

A

@xn

n

Nn

@ 1

@ n 1

We also remark that the first n

ei D



@x1

u1 C

@ i



1 column vectors of J ,

C



@xn

un ; i D 1; : : : ; n

@ i



are tangent to the n 1 coordinate curves on †n

is regular.



1



1;



and independent, since the surface



200



8 Wave Propagation



By noting that N is normal to †n 1 , so that it is independent of the vectors

.ei /, we see that the determinant of J is different from zero, so that the inverse

transformation

i



D



i .x1 ; : : : ; xn /



(8.4)



can be obtained from (8.3). Relative to the coordinates

hypersurface †n 1 becomes

n .x1 ; : : : ; xn /



i,



the equation of the



Á f .x1 ; : : : xn / D 0;



(8.5)



and Cauchy’s data (8.2) assume the simplified form

u. 1 ; : : : ;



n 1/



D u0 . 1 ; : : : ;



@u

D d0 . 1 ; : : : ;

@ n



n 1 /;



n 1 /:



(8.6)



Remark 8.1. It is worth noting that the knowledge of the directional derivative of

the function u.x/ along the normal to †n 1 gives the gradient of u.x/ in any point

of †n 1 . In fact, on †n 1 it holds that

ru D



n 1

X

@u i

@u

e C

N;

@

@

i

n

iD1



where .ei / are the vectors reciprocal to the .n



1/ tangent vectors to †n 1 .



The Cauchy–Kovalevskaya theorem introduced below is of basic relevance, since

it asserts the local existence of solutions of a system of PDEs, with initial conditions

on a noncharacteristic surface. Its proof is not straightforward and, for this reason,

only particular aspects related to the wave propagation will be addressed.

Theorem 8.1 (Cauchy–Kovalevskaya Theorem). If the coefficients aij , the function u, the Cauchy data (8.2) and the implicit representation (8.5) of †n 1 are

analytic functions of their arguments and, in addition, †n 1 satisfies the condition

aij .x; u; ru/



@f @f

Ô 0; x 2 n 1 ;

@xi @xj



(8.7)



then there exists a unique analytic solution of the Cauchy problem (8.1)–(8.2) in a

neighborhood of †n 1 .

Proof. Let us assume the coordinates . 1 ; : : : ;

of u in a neighborhood of †n 1 can be written:



n/



in


Â

u.x/ D u0 . 1 ; : : : ;



n



1 ; 0/ C



@u

@ n



Ã

n

n D0



8.2 Cauchy’s Problem for Second-Order PDEs



1

C

2



Â



@2 u

@ n2



201



Ã

n D0



2

n



C



:



(8.8)



To prove the theorem, we need to show that

1. all the derivatives of u appearing in the above expansion can be determined by

using (8.1) and Cauchy’s data (8.2);

2. the series (8.8) converges uniformly toward a solution of the Cauchy problem (8.1), (8.2);

Below we only prove that, under the assumption (8.7), all the derivatives of the

expansion (8.8) can be computed.

Let h and k be indices in the range 1; : : : ; n; then

@u

@u @ h

D

;

@xi

@ h @xi

Ä

@u @ h

@

@2 u

D

@xi @xj

@xi @ h @xj

Ä

@u @ h

@u @2 h

@

C

D

@xi @ h @xj

@ h @xi @xj

D



(8.9)



@2 u @ h @ k

@u @2 h

C

:

@ h @ k @xj @xi

@ h @xi @xj



(8.10)



By using the above expressions, we see that (8.1) becomes

Ã

Ã

Â

n Â

X

@2 u

@ h@ k

@u

;

aij

D g ; u;

@xj @xi @ h @ k

@



(8.11)



h;kD1



where on the left-hand side we have collected the second derivatives of u with

respect to h and the first derivatives of h with respect to xi . We note that the

latter are obtained by differentiating the known inverse transformation (8.4).

When the Cauchy data on †n 1

u. 1 ; : : : ;



n 1/



D u0 . 1 ; : : : ;



@u

D d0 . 1 ; : : : ;

@ n



n 1 /;



n 1 /;



(8.12)



are taken into account, we see that all the first derivatives of u at x0 2 †n

obtained by differentiating (8.12)1 or from (8.12)2 . The second derivatives

Â



@2 u

@ h@ k



Ã



Â

;

n D0



@2 u

@ h@ n



Ã



Ä

D

n D0



@

@ h



Â



@u

@ n



Ã

;

n D0



1



are



202



8 Wave Propagation



where at least one of the indices h and k changes from 1 to n 1, are evaluated from

Cauchy’s data (8.12).

Finally, about the second derivative

Â



@2 u

@ n2



Ã

;

n D0



we observe that the system (8.11), written at x0 2 †n 1 , when (8.5) is taken into

account, allows us to obtain

ÄÂ

à 2

@u

@f @f

aij

@xi @xj @ n2



D F;



(8.13)



n D0



where F is a quantity which can be derived from Cauchy’s data.

The relation (8.13) allows us to compute the desired derivative if





@f @f

aij

@xi @xj



Ô 0:

n D0



In addition, this condition allows us to compute the derivatives of higher order in

t

u

x0 2 †n 1 and to prove the remaining point (2).



8.3 Characteristics and Classification of PDEs

An .n



1/-dimensional hypersurface †n



1



f .x1 ; : : : ; xn / D 0

is said to be a characteristic surface with respect to the Eq. (8.1) and Cauchy’s

data (8.2) if z D f .x1 ; : : : ; xn / is a solution of the equation

aij .x; u; ru/



@f @f

D 0:

@xi @xj



(8.14)



In such a case, the Cauchy problem is ill posed, in the sense that there is no

uniqueness since there are several possibilities for computing the partial derivatives

@r u=@ nr ; r

starting from the same Cauchy’s data.



2



8.3 Characteristics and Classification of PDEs



203



It is relevant to observe that, if the coefficients aij of (8.1) depend only on

the coordinates xi , Eq. (8.1) is linear or semilinear and the characteristic surfaces

depend on the equation but not on Cauchy’s data.

On the other hand, in the quasi-linear case (8.14) allows us to define the function

f .x/ provided that the values of u and ru are known at any point. Since these

quantities are uniquely defined from the equation and Cauchy’s data, and are

therefore continuous across the characteristic surface †n 1 , (8.14) can be regarded

as an equation in the unknown f .x/ if the solution of (8.1) is known on at least

one side of †n 1 . This is not a severe requirement, since a solution of (8.1) is often

known in the form u0 D const.

Accordingly, when dealing with a quasi-linear equation, a solution u0 from which

we can compute the coefficients aij at any point x is supposed to be known.

However, Eq. (8.14) is a first-order nonlinear partial differential equation in

the unknown function f .x1 ; : : : ; xn /, so that its solution is not easy to find. As a

consequence, it is not easy to determine the characteristic surfaces of (8.1).

Suppose that, for a fixed point x 2 , there are characteristic surfaces †n 1

containing x. Since the components of the unit vector N, normal to a solution †n 1

of (8.14), are Ni D .@f =@xi /= jrf j, Eq. (8.14) can be written as

aij .x; u0 ; ru0 /Ni Nj Á aij0 .x/Ni Nj D 0;



(8.15)



where the vectors N are normal to the characteristic surfaces through the point x.

The collection of the vectors verifying (8.15) give rise to a cone Ax . In fact, N D 0

is a solution of (8.15). In addition, if N is a solution, then so is N provided that

is real.

These arguments allow us to classify equations (8.1) at any x 2
considering the eigenvalues of the symmetric matrix aij0 .x/.

1. The differential equation (8.1) is called elliptic at x (or with respect to a solution

u0 if it is quasi-linear) if all the eigenvalues 1 ; : : : ; n of aij0 .x/ are positive or,

equivalently, if the quadratic form

aij0 .x/Ni Nj

is positive definite.

In this case, there exists a transformation .xi / ! .xN i /, dependent on x,

which allows us to write at the point x the quadratic form aij0 .x/Ni Nj in the

canonical form

N2

1 N1



C



C



N2

n Nn ;



1; : : : ;



n



> 0:



We can also say that (8.1) is elliptic at x if there are no real vectors normal

to the characteristic surfaces passing through x; i.e., the cone Ax (8.15) at x

is imaginary. So if all the eigenvalues of aij are positive, then there is no real

solution of (8.14).



204



8 Wave Propagation



2. The differential equation (8.1) is called parabolic at x (or with respect to a

solution u0 if it is quasi-linear) if the matrix aij0 .x/ has at least one eigenvalue

equal to zero. In this case there is a transformation .xi / ! .xN i /, dependent on

x, which allows us to transform aij0 .x/Ni Nj into the canonical form

N2

1 N1



C



C



N2

m Nm ;



1; : : : ;



m



Ô 0;



mC1



D



D



n



D 0:



In this case the hyperplane NN 1 D

D NN m D 0 is contained in the cone Ax .

3. The differential equation (8.1) is called hyperbolic in x (or with respect to a

solution u0 if it is quasi-linear) if all but one of the eigenvalues of the matrix

aij0 .x/ have the same sign and the remaining one has the opposite sign. In this

case there is a transformation .xi / ! .xN i /, dependent on x, which allows us to

transform the quadratic form aij0 .x/Ni Nj into the canonical form

N2

1 N1



C



C



N2

n 1 Nn 1



N2

n Nn ;



where the eigenvalues 1 ; : : : ; n 1 have the same sign and

sign. In this case Ax is a real cone.



n



takes the opposite



8.4 Examples

1. Consider Laplace’s equation

u Á



@2 u

@2 u

C 2 D0

2

@x1

@x2



in the unknown function u.x1 ; x2 /. The matrix aij is given by

Â

aij D



10

01



Ã



with two coincident eigenvalues, equal to D 1. The equation is elliptic and the

cone Ax is imaginary, since it is represented by the equation

N12 C N22 D 0:

The characteristic curves are given by

Â



@f

@x1



Ã2



Â

C



@f

@x2



Ã2

D 0;



8.4 Examples



205



and this equation does not have real f .x1 ; x2 / solutions. Since Laplace’s equation

does not admit characteristic curves, the Cauchy problem is well posed for any

curve of the x1 ; x2 -plane.

2. As a second example, consider D’Alembert’s equation (also called the wave

equation)

@2 u

@x12



@2 u

D 0:

@x22



The matrix aij is

Â

aij D

its eigenvalues are 1 D 1 and

cone Ax is written as



2



Ã

1 0

;

0 1



D



N12



1, and the equation is hyperbolic. The

N22 D 0



and is formed by the two straight lines N1 N2 D 0 and N1 C N2 D 0. The

characteristic curves are the solutions of the equation

Â



@f

@x1



Ã2



Â



@f

@x2



Ã2

D 0;



which is equivalent to the following system:

Â



@f

@x1



@f

@x2



Ã



Â

D 0;



@f

@f

C

@x1

@x2



Ã

D 0:



With the introduction of the vector fields u1 D .1; 1/ and u2 D .1; 1/, the

previous system can be written as

rf u1 D 0; rf u2 D 0:

The above equations show that their solutions f .x1 ; x2 / are constant along

straight lines parallel to the vectors u1 and u2 , and, since this corresponds to

the definition of characteristic curves, we conclude that the characteristic curves

are represented in the plane Ox1 x2 by families of straight lines

x1



x2 D const; x1 C x2 D const.



In this case Cauchy problem is ill posed if Cauchy’s data are assigned on these

lines.



206



8 Wave Propagation



3. The heat equation

@2 u

@x12



@u

D0

@x2



is an example of a parabolic equation, since the matrix

Â

aij D

has eigenvalues



1



D 1 and



2



10

00



Ã



D 0. The cone Ax is defined at any point by

N12 D 0



and coincides with a line parallel to the x2 axis. The equation of the characteristic

curves reduces to

@f

D 0;

@x1

with solutions

f D g.x2 / D const,

where g is an arbitrary function. It follows that the characteristic curves are

represented by the lines

x2 D const.

4. Finally we consider Tricomi’s equation

@2 u

@2 u

C

x

D 0:

1

@x12

@x22

Since the matrix

Â

aij D



1 0

0 x1



Ã



has eigenvalues 1 D 1 and 2 D x1 , this equation is hyperbolic in those points

of the plane Ox1 x2 where x1 < 0, parabolic if x1 D 0, and elliptic if x1 > 0.

The equation of the cone Ax is given by

N12 C x1 N22 D 0;



8.5 Cauchy’s Problem for a Quasi-Linear First-Order System



207



and the characteristic equation

Â



@f

@x1



Ã2



Â

C x1



@f

@x2



Ã2

D0



does not admit real solutions if x1 > 0, whereas if x1 < 0 it assumes the form

Â

ÃÂ

Ã

p

p

@f

@f

@f

@f

D 0:

C jx1 j

jx1 j

@x1

@x2

@x1

@x2

This equation is equivalent to the following two conditions:

rf v1 D 0; rf v2 D 0;

where v1 D .1;



p



jx1 j/ and v2 D .1;



p



jxq

1 j/, and the characteristic curves are

the integrals of these fields, e.g., x2 D ˙ 23 jx1 j3 C c, where c is an arbitrary

constant.



8.5 Cauchy’s Problem for a Quasi-Linear First-Order

System

In the previous section we introduced the concept of the characteristic surface

related to a second-order PDE. Here we consider the more general case of a firstorder quasi-linear system of PDEs. It is clear that this case includes the previous

one, since a second-order PDE can be reduced to a system of two first-order PDEs.

Let x D .x1 ; : : : xn / be a point of a domain


.u1 .x/; : : : ; um .x// be a vector function of m components, each depending on n

variables .x1 ; : : : ; xn /. Assume that the vector function satisfies the m differential

equations

8

1 @u1

ˆ

ˆ

< A11 @x1 C



Á

m

C

C A11m @u

@x1



ˆ

ˆ

: A1 @u1 C

m1 @x1



Á

m

C

C A1mm @u

@x1



@u1

C An11 @x

C

n



Á

m

D c1 ;

C An1m @u

@xn



@u1

C Anm1 @x

C

n



Á

m

D cm ;

C Anmm @u

@xn

(8.16)



where

Aij h D Aij h .x; u/; ci D ci .x; u/

are continuous functions of their arguments.



(8.17)



208



8 Wave Propagation



It is necessary to clearly specify the meaning of indices in Aij h .x; u/:

i D 1; : : : ; n refers to the independent variable xi I

j D 1; : : : ; m refers to the equation;

h D 1; : : : ; m refers to the function uh :

If the matrices

0

Ai D @



Ai11



Ai1m



Aim1



Aimm



1



0



A;



cD@



c1



1

A;



(8.18)



cm



are introduced, system (8.16) can be written in the following concise way:

A1



@u

C

@x1



C An



@u

D c;

@xn



or in the still more compact form

Ai



@u

D c:

@xi



(8.19)



The Cauchy problem for the system (8.19) consists of finding a solution of (8.19)

which satisfies Cauchy’s data

u.x/ D u0 .x/



8x 2 †n 1 ;



(8.20)



where †n 1 is a regular surface of
By proceeding as in the previous section, we introduce the coordinates

. 1 ; : : : ; n /, so that †n 1 is represented by the equation n D f .x/ D 0. When

expressed in these coordinates, system (8.19) becomes

Ai



@u

@u @ h

D Ai

Dc

@xi

@ h @xi



or

Ã

Â

@ h @u

D c;

Ai

@xi @ h



(8.21)



whereas Cauchy’s data assume the form

u. 1 ; : : : ;



n 1 ; 0/



D uQ 0 . 1 ; : : : ;



n 1 /:



(8.22)



Xem Thêm
Tải bản đầy đủ (.pdf) (489 trang)

×