Beruflich Dokumente
Kultur Dokumente
Elliptic problems
An eective method for solving boundary value problems for the Laplace and
Helmoltz equations (in domains possessing a denite symmetry) is the method of
separation of variables. The general idea of this method is to nd a set of solutions
of the homogeneous partial dierential equation in question that satisfy certain
boundary conditions. These solutions then serve as atoms, from which, based
on the linear superposition principle, one constructs the general solution. Since
each of these atoms is a solution of the corresponding homogeneous equation,
their linear combination is also
a solution of the same equation. The solution of our
problem is given by a series n=1 cn un (x) (where un (x) are the atom solutions,
x = (x1 , . . . , xN ) is the current point of the domain of space under consideration,
and cn are arbitrary constants). It remains to nd constants cn such that the
boundary conditions are satised.
2 R + R
= . (1.2)
R
V.P. Pikulin and S.I. Pohozaev, Equations in Mathematical Physics: A practical course, 7
Modern Birkhuser Classics, DOI 10.1007/978-3-0348-0268-0_2,
Springer Basel AG 2001
8 CHAPTER 1. ELLIPTIC PROBLEMS
One says that in equation (1.2) the variables are separated, since the left-
[resp., right-] hand side of the equation depends only on [resp., ]. Since the
variables and do not depend of one another, each of the two sides of equation
(1.2) must be a constant. Let us denote this constant by . Then
2 R + R
= = . (1.3)
R
It is clear that when the angle varies by 2 the single-valued function u(, ) must
return to the initial value, i.e., u(, ) = u(, + 2). Consequently, R()() =
R()( + 2), whence () = ( + 2), i.e., the function () is 2-periodic.
From the equation + = 0 it follows that () = A cos( ) + B sin( )
(with A and B arbitrary constants), and in view of the periodicity of () we
necessarily have = n2 , where n 0 is an integer.
Indeed, the equality
A cos( ) + B sin( ) = A cos[ ( + 2)] + B sin[ ( + 2)]
implies that
sin( + ) = sin( + + 2 ),
where we denote
A B
sin = , cos = .
2
A +B 2 A + B2
2
Therefore, sin( ) cos( + + ) = 0, i.e., = n, or = n2 ,
where n 0 is an integer. Now equation (1.3) yields
2 R + R n2 R = 0. (1.4)
If n = 0, then we seek the solution of this equation in the form R() = .
Substituting this expression in equation (1.4) and simplifying by , we get
2 = n2 , or = n (n > 0).
For n = 0 equation (1.4) has two solutions: 1 and ln . Thus, we now have an
innite set of functions (atom solutions)
1, ln , n cos(n), n sin(n),
n cos(n), n sin(n), n = 1, 2 . . . ,
which satisfy the given partial dierential equation. Since a sum of such solutions
is also a solution, we conclude that in our case the general solution of the Laplace
equation has the form
u(, ) = a0 + b0 ln +
(1.5)
+ an n + bn n cos(n) + cn n + dn n sin(n) .
n=1
1.1. DIRICHLET PROBLEM FOR LAPLACE EQUATION IN ANNULUS 9
It remains only to nd all the coecients in the sum (1.5) so that the bound-
ary conditions u(R, ) = f1 (), u(R2 , ) = f2 () will be satised. Setting = R1
and then = R2 in (1.5) we obtain
u(R1 , ) = an R1n + bn R1n cos(n)+
n=1
+ cn R1n + dn R1n sin(n) + a0 + b0 ln R1 ,
u(R2 , ) = an R2n + bn R2n cos(n)+
n=1
+ cn R2n + dn R2n sin(n) + a0 + b0 ln R2 .
In this case we will seek the solution as a function that does not depend on ,
i.e., u() = a0 + b0 ln . Substituting this expression in the boundary conditions
we obtain the system of equations
a0 + b0 ln 1 = 2,
a0 + b0 ln 2 = 1,
One can verify that here all the coecients a0 , b0 , an , bn , cn , dn with n > 1
are equal to zero, while the coecients a1 , b1 , c1 , d1 are determined from the
systems of equations
a1 + b1 = 1, c1 + d1 = 0,
2a1 + b1 = 0, 2c1 + d1 = 1.
2 2
Solving these systems we obtain
1 4 2 2
a1 = , b1 = , c1 = , d1 = .
3 3 3 3
Thus, the solution of our problem is the function
1 4 2 1
u(, ) = + cos + sin .
3 3 3
Since the Dirichlet problem for the Laplace equation in a bounded domain
has a unique solution, in examples 13 there are no other solutions besides the
ones found.
is solved in exactly the same manner as the Dirichlet problem for the annulus,
with the only dierence that now we must discard the solution atoms that are
not bounded when approaches 0:
ln , n cos(n), n sin(n), n = 1, 2, . . .
Hence, the solution is given by the remaining terms, i.e.,
n
u(, ) = [an cos(n) + bn sin(n)] ,
n=0
R
n
and then multiply each term of the series by the factor R . For example, the
interior problem
u = 0, 0 < 1, 0 < 2,
u(1, ) = cos2 , 0 < 2,
is solved in much the same way as the preceding problem, with the dierence than
now we discard the solution atoms that are not bounded when goes to innity:
ln , n cos(n), n sin(n), n = 1, 2, . . .
where the coecients an and bn are calculated by means of formulas (1.7). For
example, the exterior problem
u = 0, 1 < , 0 < 2,
u(1, ) = sin3 , 0 2
0
=
u(
)
a,
)
,
u(
=
u = 0
0 u(, 0) = 0 a x
Figure 1.1.
Solution. Finding the steady temperature distribution reduces to solving the Di-
richlet problem
2
u + u + u = 0, 0 < a, 0 < < < 2,
u(, 0) = u(, ) = 0, 0 a,
u(a, ) = A, 0 .
n 2
We get n = and
n 2
( 1) + = 0,
whence
n
= .
14 CHAPTER 1. ELLIPTIC PROBLEMS
Using the fact that the function R() is bounded (according to the meaning
of the problem at hand), we write Rn () = n/ . The atoms from which our
solution is built are the functions
n
un (, ) = n/ sin , n = 1, 2, . . .
Thus, the solution itself is
n
u(, ) = cn n/ sin .
n=1
it follows that n
n/ 2
cn a = A sin d,
0
and so n
2A 2A
cn = sin d = (1)n+1 .
an/ 0 n
Finally, the solution of our problem is written in the form
n
2A n/ sin
u(, ) = (1)n+1 .
n=1 n
2
1 2 R2
u(, ) = f () d, > R.
2 0 R 2R cos( ) + 2
2
Let us show that these formulas are a consequence of the general superposition
method.
1.4. POISSON INTEGRAL. RATIONAL BOUNDARY CONDITION 15
For the sake of deniteness we shall consider the interior problem, and then
write the result for the exterior problem by analogy.
Substituting the expression for the Fourier coecients in the formula
n
u(, ) = [an cos(n) + bn sin(n)] ,
n=0
R
we obtain
1 n
2
1
u(, ) = f () + (cos(n) cos(n) + sin(n) sin(n)) =
0 2 n=0 R
1 n
2
1
= f () + cos(n( )) d.
0 2 n=0 R
Further, using the relation cos(n( )) = 12 ein() + ein() , the fact that
q = /R < 1 and the formula for the sum of an innite decreasing geometric
progresion, we get
1 1 n in()
1 n
+ q cos(n( )) = + q e + ein() =
2 n=1 2 2 n=1
1 n n
= 1+ qein() + qein() =
2 n=1
1 qein() qein()
= 1+ + =
2 1 qein() 1 qein()
1 1 q2 1 R2 2
= = .
2 1 2q cos( ) + q 2 2 R2 2R cos( ) + 2
Therefore,
2
1 R2 2
u(, ) = f () d, < R.
2 0 R2 2R cos( ) + 2
R2 2 R2 |z|2 Rei + z
= = Re i
R2 2R cos( ) + 2 |Re z|
i 2 Re z
16 CHAPTER 1. ELLIPTIC PROBLEMS
because
i
Rei + z Re + ei (Rei ei )
Re = Re
Rei z (Rei ei ) (Rei ei )
R2 |z|2 + R ei() ei() R2 |z|2
= Re 2
= .
|Re z|
i |Rei z|2
It follows that the Poisson integral can be written in the form
2
1 Rei + z
u(z) = Re f () d.
2 0 Rei z
If the boundary function f () is a rational function of sin and cos , then the
integral in formula (1.9) can be calculated by means of residues.
Example. Solve the Dirichlet problem
u = 0, |z| < 2,
2 sin
u||z|=2 = .
5 + 3 cos
1 2 4
2
2 sin 2i 2
u() = = =
5 + 3 cos 3 2
5+ +
2 2
2 2 4 2 2 4
= 2 = .
i 3 + 20 + 12 i 2
3( + 6) +
3
Let us compute the integral
1 2( 2 4)( + z)
J= d
2i ||=2 2
i 3( + 6)( + )( z)
3
1.4. POISSON INTEGRAL. RATIONAL BOUNDARY CONDITION 17
where the circle || = 2 is oriented counter- clockwise. In our case the integrand
F () has in the domain || > 2 only one nite singular point = 6 a pole
of order one and the removable singular point = . By the Cauchy residue
theorem,
J = res[F ()]=6 res[F ()]= .
2 32 z6 4 z6 2 6z
res[F ()]=6 =
16 = = .
3i 3 (z + 6) 6 i (z + 6) 6 3i 6 + z
4 z
1 1+
2 2 1 1 2 1
F () = z = + ... ,
3i 6 2 1 3i
1+ 1+
3
whence
2
res[F ()]= = .
3i
Therefore,
2 z6 2 2 2z 4z
J= + = = =
3i z + 6 3i 3i z + 6 3i(z + 6)
4 x + iy 4 (x + iy)(6 + x iy)
= = ,
3i 6 + x + iy 3i (6 + x)2 + y 2
which yields
8y
Re J = ,
36 + 12x + x2 + y 2
or
8 sin
Re J = .
36 + 12 cos + 2
We conclude that the solution of our Dirichlet problem is given by the ex-
pression
8 sin
u(, ) = .
36 + 12 cos + 2
18 CHAPTER 1. ELLIPTIC PROBLEMS
The coecients an and bn of this series are determined from the boundary condi-
tion u/|=R = f (), i.e., we have
2
R
an = f () cos(n) d,
n 0
2 n = 1, 2, . . . (1.10)
R
bn = f () sin(n) d,
n 0
Similarly, the solution of the exterior Neumann problem is sought in the form
of a series n
u(, ) = [an cos(n) + bn sin(n)] .
n=0
R
2 2
R R
= cos d + [cos(3) + cos ] d = 0.
2 0 4 0
1.6. POISSON EQUATION IN DISC AND ANNULUS 19
3 3
u(, ) = C + cos + cos(3),
4 12R2
where C is an arbitrary constant.
Remark. The Neumann problem can also be solved for an annulus. In this case
the boundary conditions specify the exterior normal derivative:
u u
(R1 , ) = f1 (), (R2 , ) = f2 ().
2u 2u
+ 2 = xy
x2 y
in the disc of radius R centered at the origin, under the condition u(R, ) = 0.
Solution. Passing to polar coordinates we obtain the problem
2 u + u + u = 1 4 sin(2), 0 < R, 0 < 2,
2 (1.11)
u(R, ) = 0, 0 2.
u1 (, ) = w() sin(2).
20 CHAPTER 1. ELLIPTIC PROBLEMS
1 v 1 2v
+ 2 = A2 cos(2) a < < b, 0 < 2,
2
(1.15)
v
v(a, ) = 0, (b, ) = 0, 0 2.
1 d 4
cos(2) (R ) 4 R cos(2) = A2 cos(2),
d
2 R + R 4R = A4 ,
R 4R = Ae4 t,
where the dot denotes dierentiation with respect to t. The general solution of this
1
last equation is R(t) = C1 e2t + C2 e2t + 12 Ae4t . Back to the variable we have
C2 1
R() = C1 2 + + Ae4t .
2 12
The constants C1 and C2 are found from the conditions R(a) = 0, R (b) = 0,
namely
A(a6 + 2b6 ) Aa4 b4 (2b2 a2 )
C1 = , C2 =
12(a4 + b4 ) 6(a4 + b4 )
A(a6 + 2b6 ) 2 Aa4 b4 (2b2 a2 ) 1 A 4
u(, ) = 1 + + + cos(2).
12(a4 + b4 ) 6(a4 + b4 ) 2 12
22 CHAPTER 1. ELLIPTIC PROBLEMS
u=0
B(0, b) C(a, b)
u=V u = 0 u=0
0 u=0 A(a, 0) x
Figure 1.2.
Solution. The problem reduces to that of solving the Laplace equation uxx +uyy = 0
in the interior of the rectangle with the boundary conditions
First we will seek nontrivial particular solutions of the Laplace equation which
satisfy only the boundary conditions
u(x, 0) = u(x, b) = b
in the form u(x, y) = X(x)Y (y). Substituting this expression in the equation
uxx + uyy = 0 we get X Y + XY = 0, which upon dividing by XY gives
X Y
= = 2 .
X Y
Using the fact that Y (0) = Y (b) = 0, we obtain the Sturm-Liouville problem
Y + 2 Y = 0, 0 < y < b,
Y (0) = Y (b) = 0,
n n n
u(x, y) = an cosh x + bn sinh x sin y . (1.16)
n=0
b b b
n n n
0= an cosh a + bn sinh a sin y ,
n=0
b b b
whence n n
an cosh a + bn sinh a = 0, n = 1, 2, . . .
b b
Next, setting x = 0 in (1.16) we obtain
n
V = an sin y ,
n=0
b
which gives
b n
2 0, if n is even,
an = V sin y dy, or an = 4V
b 0 b n , if n is odd.
Example 2 [18]. Suppose that two sides, AC and BC, of a rectangular homogeneous
plate (see Figure 1.2) are covered with a heat insulation, and the other two sides
are maintained at temperature zero. Find the stationary temeperature distribution
in the plate under the assumption that a quantity of heat Q = const is extracted it.
24 CHAPTER 1. ELLIPTIC PROBLEMS
Solution. We are dealing with a boundary value problem for the Poisson equation
with boundary conditions of mixed type;
Q
uxx + uyy = k ,
0 < x < a, 0 < y < b,
u(0, y) = 0, ux (a, y) = 0, 0 y b, (1.17)
u(x, 0) = 0, uy (x, b) = 0, 0xa
2
We get 2n = (2n+1)
2a and Xn (x) = sin (2n+1)
2a x , n = 0, 1, . . . We will seek
the solution of the above problem in the form of an expansion in eigenfunctions
(2n + 1)
u(x, y) = Yn (y) sin x ,
n=0
2a
where the functions Yn (y) are subject to determination. Substituting this expres-
sion of the solution in equation (1.17) we obtain
(2n + 1)2 2 (2n + 1) (2n + 1)
Yn (y) sin x + Y (y) sin x =
n=0
4a2 2a n=0
n
2a
(2n + 1)
= n sin x ,
n=0
2a
This yields the following boundary value problem for the determination of
the function Yn (y), n = 0, 1, 2, . . . :
2 2
Y (2n + 1) Y (y) = 4Q
n 2 n , 0 < y < b,
4a k(2n + 1)
Y (0) = 0,
n Yn (b) = 0.
1.7. LAPLACE AND POISSON EQUATIONS IN RECTANGLE 25
where
16Qa2
an = ,
k 3 (2n + 1)3
and
16Qa2 (2n + 1)b
bn = tanh y .
k 3 (2n + 1)3 2a
The nal expression of the solution is
2 cosh (2n+1)(by)
16Qa 1 1 2a (2n + 1)
u(x, y) = sin x .
k 3 n=0
(2n + 1)3 cosh (2n+1)b 2a
2a
Example 3 [18]. Find the solution of the Laplace equation in the strip 0 x a,
0 y < which satises the boundary conditions
x
u(x, 0) = 0, u(a, y) = 0, u(x, 0) = A 1 , u(x, ) = 0.
a
in the form v(x, y) = X(x)Y (y). We obtain two ordinary dierential equations:
(1) X + X = 0, and (2) Y Y = 0.
From the conditions v(0, y) = 0, v(a, y) = 0 it follows that X(0) = X(a) = 0.
Hence, the Sturm-Liouville problem
X + X = 0, 0 < x < a,
X(0) = X(a) = 0
26 CHAPTER 1. ELLIPTIC PROBLEMS
2
yields n = n a and Xn (x) = sin n a x , n = 1, 2, . . . . Then the corresponding
solutions of the equation Y Y = 0 are
n n
Yn (y) = An e a y
+ Bn e a y
.
We conclude that
n n n
vn (x, y) = An e a y + Bn e a y sin x .
a
Therefore, the solution of problem (1.18) is given by a series
n n n
u(x, y) = An e a y
+ Bn e a y
sin x . (1.19)
n=1
a
i.e.,
2 a x n 2A
An = A 1 sin x dx = .
a 0 a a n
We conclude that
2A 1 n y n
u(x, y) = e a sin x .
n=1 n a
Remark 1. The boundary value problem for the Laplace (Poisson) equation in a
rectangular parallelepiped is solved in a similar manner.
Remark 2. Let us assume that the mathematical model of a given physical phe-
nomenon is such that both the equation itself and the boundary conditions are
inhomogeneous. Then by using the superposition principle the original boundary
value problem can be decomposed into subproblems; one then solves the subprob-
lems and adds their solutions to obtain the solution of the original problem.
For example, the solution of the Dirichlet problem
u = f in the domain ,
u = on the boundary ,
z
u=0
u = 0 u = V0
l
a
0 y
u=0
x
Figure 1.3.
Solution. We need to solve the Laplace equation inside the cylinder with given
boundary conditions:
1 u 2u
+ 2 = 0, 0 < < a, 0 < z < l,
z
u(, 0) = u(, l) = 0, 0 a,
u(a, z) = V0 , 0zl
(the solution u(, z) does not depend on since the boundary values are indepen-
dent of ). Using the method of separation of variables, we represent the solution in
the form u(, z) = R()Z(z). Substituting this expression in the Laplace equation
1 u 2u
+ 2 =0
z
28 CHAPTER 1. ELLIPTIC PROBLEMS
we get
Z (R ) + RZ = 0
whence, upon dividing both sides by RZ,
(R )
Z
+ = 0,
R Z
or
(R )
Z
= = , (1.20)
R Z
where is the separation constant. Clearly, on physical grounds > 0: otherwise
the function Z(z), and together with it the potential, would not vanish on the
upper and bottom bases of the cylindrical wire.
Equation (1.20) yields two ordinary dierential equations:
(1) Z + Z = 0,
and
1 d
(2) (R ) R = 0.
d
Using the fact that Z(0) = Z(l) = 0, we obtain the standard Sturm-Liouville
problem:
Z + Z = 0, 0 < z < l,
Z(0) = Z(l) = 0.
This problem has the eigenfunctions Zn (z) = sin n z , corresponding to the
2 l
eigenvalues n = n
l , n = 1, 2, . . . . The function R() is determinded from the
equation
1 d n 2
(R ) R = 0, (1.21)
d l
which is recognized to be the Bessel equation of index zero and imaginary argu-
ment. Indeed, from equation (1.21) it follows that
n 2
2 R + R 2 R = 0.
l
Example 2 [18]. Consider a cylinder with base of radius R and height h. Assume
that the temperature of the lower base and of the lateral surface is equal to zero,
while the temperature of the upper base is a given function of . Find the steady
temperature distribution in the interior of the cylinder.
Solution. The mathematical formulation of the problems is as follows:
1 u 2u
+ 2 = 0, 0 < < R, 0 < z < h,
z
u(, 0) = 0, u(, h) = f (), 0 R,
u(R, z) = 0, 0 z h.
1 d
(1) (r ) + r = 0;
d (1.22)
(2) Z Z = 0.
We note that here > 0 (this will be clear once we nd the solution). The
boundary condition u(R, z) = 0 implies r(R) = 0. Equation (1.22) can be rewritten
as
2 r + r + 2 r = 0. (1.23)
Passing to the new independent variable x = we obtain the Bessel equation
of order zero
d2 r dr
x2 2 + x + x2 r = 0,
dx dx
whose general solution has the form
where J0 (x) and B0 (x) are the Bessel function of order zero of rst and second
kind, respectively, and C1 , C2 are arbitrary constants.
Returning to the old variable we have
r() = C1 J0 ( ) + C2 B0 ( ).
y
1
0.8
0.6
0.4 J0 (x)
0.2
2.4 5.5 8.7
2 4 6 8 10 x
0.2
0.4
Figure 1.4.
The constants Bn are found from the boundary condition u(, h) = f (). Indeed,
we have
n n
u(, h) = Bn J0 sinh h ,
n=1
R R
or
n n
f () = Bn J0 sinh h .
n=1
R R
32 CHAPTER 1. ELLIPTIC PROBLEMS
m
Multiplying both sides of this equality by J0 R and integrating the result
over the segment [0, R] we get
R R
J02
m m m
f ()J0 d = Bm sinh h d.
0 R R 0 R
But R R2 2
J02
m
d = J (m ),
0 R 2 1
where J1 (x) is the Bessel function of rst kind and order one. Therefore, the
solution of the problem has the form
n n R
2 sinh R z J0 R
u(, z) = 2 f ()J0
n
d.
R n=1 sinh n h J12 (n ) 0 R
R
Example 3. Find the potential in the interior points of a grounded cylinder of
height h and with base of radius R,
given that in the cylinder there is a charge
distribution with density = AzJ0 R3 (where A is a constant).
Solution. We must solve the Poisson equation with null boundary conditions:
1 u 2u 3
+ 2 = 4AzJ0 ,
z R
0 < < R, 0 < z < h, (1.24)
u(, 0) = u(, h) = 0, 0 R,
u(R, z) = 0, 0 z h.
Let us seek the solution in the form u(, z) = J0 R3 f (z), where the func-
tion f (z) is subject to determination. Substituting this expression of u(, z) in
equation (1.24) we get
1 d d 3 3 3
J0 f (z) + J0 f (z) = 4AzJ0 . (1.25)
d d R R R
Now let us observe that the function J0 R3 is an eigenfunction of the
Bessel equation, i.e.,
2
1 d d 3 3
J0 + 32 J0 = 0.
d d R R R
which in turn yields the following ordinary dierential equation for the determi-
nation of f (z):
3
3
f f = 4Az, 0 < z < h,
R
with f (0) = f (h) = 0. Solving this boundary value problem we nd that
3
4AR2 h sinh R z 4AR2
f (z) = 3 + z.
23 sinh h 23
R
Thus, the solution of our problem is given by the expression
3
4AR2 sinh z
3 h
u(, z) = J0 R z .
R 23 sinh
3
h
R
(m)
where Pn (x) are the so-called associated Legendre functions.
Example 1. Find the solution u(, , ) of the interior Dirichlet problem for the
Laplac equation with the boundary condition u(a, , ) = sin(3) cos .
Solution. In spherical coordinates the problem is written as follows:
1 2 u 1 u 1 2u
2 + 2
sin + 2 2 = 0,
sin sin 2
(1.26)
0 < < a, 0 < < , 0 < 2,
u(a, , ) = sin(3) cos , 0 , 0 2.
34 CHAPTER 1. ELLIPTIC PROBLEMS
1 d 1
(sin T ) + T + T = 0,
sin d sin2
whence
d
sin (sin T )
d + sin2 = = .
T
Thus, the function () is found by solving the problem
+ = 0,
() = ( + 2).
1.9. LAPLACE AND POISSON EQUATIONS IN A BALL 35
dT dx dT
T = = ( sin ),
dx d dx
d2 T dT
T = 2
sin2 cos ,
dx dx
equation (1.29) yields the following boundary value problem for eigenvalues and
eigenfunctions:
2
2
(1 x2 ) d T 2x dT + m T = 0, 1 < x < 1,
dx2 dx 1 x2
|T (1)| < , |T (+1)| < .
dm
Tn(m) (x) = Pn(m) (x) = (1 x2 )m/2 Pn (x),
dxm
are the associated Legendre functions. Hence, the solutions of equation (1.29) are
(m) (m)
the functions Tn (x) = Pn (cos ).
Combining the solutions of equation (1.29) with the solutions of the equation
+ = 0, we obtain the 2n + 1 spherical functions
Now let us return to the search for the function R(). Setting R() =
and substituting this expression in the equation 2 R + 2R R = 0, we obtain
36 CHAPTER 1. ELLIPTIC PROBLEMS
(m) (m)
However, the solutions (n+1) Pn (cos ) cos(m), (n+1) Pn (cos ) sin(m)
must be discarded because they are not bounded when 0. Hence, the so-
lution of our problem is given by a series
n
u(, , ) = n [Anm cos(m) + Bnm sin m] Pn(m) (cos ).
n=0 m=0
It remains to choose the constants Anm and Bnm so that the boundary con-
dition
u(a, , ) = sin(3) cos
n
sin(3) cos = an [Anm cos(m) + Bnm sin m] Pn(m) (cos ).
n=0 m=0
n
It follows that in the sum m=0 we must retain only the term corresponding
to m = 1. This yields
sin(3) = an An1 Pn(1) (cos ).
n=1
f () = bn Pn(1) (cos ),
n=1
then
2n + 1 (n 1)!
bn = f ()Pn(1) (cos ) sin d.
2 (n + 1)! 0
1.9. LAPLACE AND POISSON EQUATIONS IN A BALL 37
z
(3)
P5 (cos ) cos(3)
u = 0
R1 R2
O y
x
(1)
P2 (cos ) sin
Figure 1.5.
38 CHAPTER 1. ELLIPTIC PROBLEMS
n
n Bnm
u(, , ) = Anm + n+1 cos(m) +
n=0 m=0
Dnm
+ Cnm n + n+1 sin(m) Pn(m) (cos ),
where the numbers Anm , Bnm , Cnm and Dnm are subject to determination. The
boundary conditions yield the following systems of equations for the coecients
of the expansion:
2 D21 5 B53
C21 R1 + R3 = 1,
A53 R1 + R6 = 0,
1
1
B
D
2 21
5 53
A21 R1 + R3 = 0, C53 R1 + R6 = 0,
1 1
(1) (2)
2 D21
5 B53
C21 R2 + 3 = 0,
A53 R2 + 6 = 1,
R2
R2
B
D
A21 R22 + 21 = 0,
C53 R25 + 53 = 0,
R23 R26
All the remaining coecients are equal to zero. Solving the above systems
we obtain
R13
A21 = B21 = 0, C53 = D53 = 0, C21 = ,
R22 (R25 R15 )
We have
Bnm
u(, , ) = Anm n + cos(m)+
n=0 m=0
n+1
Dnm
+ Cnm + n+1 sin(m) Pn(m) (cos ).
n
From the boundary conditions it follows that in this sum we must retain only
the terms with the indices n = 2, m = 2 and n = 1, m = 0. In other words, it is
convenient to seek the solution in the form
b d
u(, , ) = a + 2 cos + c2 3 sin2 sin(2).
Using the boundary conditions we obtain the following system of equations for the
determination of the coecients a, b, c, d:
4a + b = 0,
5c = 5,
2a + b/4 == 1,
4c d/8 = 0.
Example 4 [4, Ch. IV, no. 125]. Find the solution of the Neumann problem for the
Laplace equation in the interior of the sphere of radius a with the condition
u
(a, , ) = A cos (A = const).
n
Solution. We are dealing with the case of an axially-symmetric solution of the
Neumann problem for the Laplace equation, since the boundary condition does not
depend on , and consequently the solution also does not depend of : u = u(, ).
40 CHAPTER 1. ELLIPTIC PROBLEMS
First of all, it is readily veried that the necessary condition for the solvability
of our problem is satised. Indeed
2 2
u
ds = 0, or d A cos sin a2 d = 0.
0 0 n 0 0
and
1 d
(sin T ) + T = 0. (1.31)
sin d
If in the equation (1.31) we pass to the new variable x = cos we arrive at
the Legendre equation
d dT
(1 x2 ) + T = 0, 1 < x < 1, (1.32)
dx dx
under the condition |T (1)| < . The bounded solutions of the Legendre equation
(1.32) on the interval (1, 1) are the Legendre polynomials Pn (x) for n = n(n+1).
Hence, the bounded solutions of equation (1.31) on the interval (0, ) are the
functions Pn (cos ). The bounded solutions of equation (1.30) are the functions
Rn () = n (n = 0, 1, 2, . . . ). It follows that
u(, ) = Cn n Pn (cos ),
n=0
or, setting = a,
A cos = nCn an1 Pn (cos )
n=0
u(, ) = C + A cos ,
Let us solve rst problem (1.33), seeking the solution in the form
(1)
v(, , ) = R()P2 (cos ) cos ,
(1)
where P2 (x) is the associated Legendre function with indices n = 2, m = 1.
Substituting this expression of v(, , ) in the equation of problem (1.33) and
(1) (1)
denoting P2 (cos ) cos = Y2 (, ) we get the equation
& '
(1) (1)
(1) d 2 1 Y2 1 2 Y2 4 (1)
Y2 ( R ) + R sin +R = Y (, ).
d sin 2
sin 2 6 2
(1)
But by the denition of the spherical function Y2 (, ) one has the identity
& '
(1) (1)
1 Y2 1 2 Y2 (1)
sin + 2 2
+ 6Y2 = 0, 0 < < , 0 < < 2.
sin sin
42 CHAPTER 1. ELLIPTIC PROBLEMS
Therefore,
d 2 (1) (1) 4 (1)
( R )Y2 6RY2 = Y ,
d 6 2
which yields the equation
d 2 4
( R ) 6R = , 0 < < a,
d 6
together with the boundary conditions |R(0)| < , R(a) = 0. Therefore, the
function R() is determined by solving the problem
2 4
R + 2R 6R = , 0 < < a,
6
|R(0)| < , R(a) = 0.
1 2 2
Its solution is R() = 84 ( a2 ). The solution of problem (1.34) is w() = 1.
We conclude that
1 2 2 (1)
w(, , ) = 1 + ( a2 )P2 (cos ) cos .
84
Remark 1. In the general case, when one solves the interior Dirichlet problem for
the Laplace equation with the condition u| = f (, ) (where is the ball of
radius a centered at the origin and is its boundary), one can write
n
f (, ) = an [Anm cos(m) + Bnm sin(m)] Pn(m) (cos ),
n=0 m=0
where the coecients Anm and Bnm are given by the formulas
2
f (, )Pn(m) (cos ) cos(m) sin d d
0 0
Anm = (m)
Yn 2 an
and
2
f (, )Pn(m) (cos ) sin(m) sin d d
0 0
Bnm = (m) 2 n
;
Yn a
also,
2m (n + m)! 2, if m = 0,
Yn(m) 2 = , where m =
2n + 1 (n m)! 1, if m > 0.
Remark 2. The solution of the aforementioned interior Dirichlet problem for the
Laplace equation at a point (0 , 0 , 0 ) admits the integral representation (Poisson
integral)
2
a a2 2
u(0 , 0 , 0 ) = f (, ) 2 sin d d,
4 0 0 (a 2a0 cos + 20 )3/2
where cos = cos cos 0 + sin sin 0 cos( 0 ).
1.10. PROBLEMS FOR THE HELMHOLTZ EQUATIONS 43
u(, ) = R()().
Inserting this expression in equation (1.35) and separating the variables we obtain
two ordinary dierential equations:
(1) + = 0,
and
d
(2) (R ) + (2 )R = 0.
d
To determine we have the Sturm-Liouville problem
+ = 0, 0 < < 0 ,
(0) = 0, (0 ) = 0.
2
This yields n = n
0 , n = 0, 1, . . . , and n () = cos n
0 . The function R()
is obtained from the following boundary value problem for the Bessel equation
d (R ) + (2 R) = 0, a < < b,
d (1.36)
R (a) = R (b) = 0.
44 CHAPTER 1. ELLIPTIC PROBLEMS
where C1 and C2 are arbitrary constants and N n 0
( ) is the Bessel function of
second kind. The values of are determined by means of the boundary conditions
in (1.36); namely, they provide the system of equations
C1 J n ( a) + C2 N n ( a) = 0,
0 0
C1 J n ( b) + C2 N n ( b) = 0.
0 0
(n) 2 (n)
is equal to zero. In other words, m,n = m , where m are the roots of the
equation
J n
0
( a) N n
0
( a)
= .
J n
( b) N n
( b)
0 0
Rm,n () = J n
((n)
m )N n
((n)
m a) J n
((n) (n)
m a)N (m ).
n
0 0 0 0
Thus, the natural oscillations of our plate are described by the functions
u 2 u = 0 ( > 0).
1.11. HELMOLTZ EQUATION IN A CYLINDER 45
1 d R
(R ) + 2 2 R = 0,
d
or
1 d
(R )
d
2 2 = = .
R
This yields two ordinary dierrential equations:
(1) + = 0,
and
d
(2) (R ) ( 2 2 + )R = 0.
d
From equation (1), by using the fact that () = ( + ), we obtain = n2
(n = 0, 1, 2, . . . ) and n () = An cos(n) + Bn sin(n), where An and Bn are
arbitrary constants.
Further, equation (2) yields
2 R + R ( 2 2 + n2 )R = 0.
d2 R dR
x2 +x (x2 + n2 )R = 0.
dx2 dx
where In (x) and Kn (x) are the cylindrical functions of imaginary argument of rst
and second kind, respectively. Clearly, we must put C2 = 0 because the solution
46 CHAPTER 1. ELLIPTIC PROBLEMS
is required to be bounded on the axis of the cylinder (Kn (x) has a logarithmic
singularity as x 0). Returning to the original variable we write
where the constants An are determined from the boundary condition. Namely, we
have
u(a, ) = [An cos(n) + Bn sin(n)] In (),
n=0
and since u(a, ) = u0 , we see that A0 = u0 /I0 (a), while all the remaining terms
of the series are equal to zero. Hence, the solution is
I0 ()
u(, ) = u0 .
I0 (a)
d
(R )
d
+ k2 2 = ,
R
d
(R ) + (k2 2 n2 )R = 0. (1.38)
d
d2 R dR
x2 +x + (x2 n2 )R = 0.
dx2 dx
This is the Bessel equation of order n and has the general solution
where Jn (x) and Yn (x) are the nth order Bessel functions of the rst and second
kind, respectively, and C1 , C2 are arbitrary constants.
Therefore, the solution of equation (1.38) has the form
The constants An and Bn are found from the boundary conditions. Setting
= a in (1.39), we obtain
f () = [An cos(n) + Bn sin(n)] Jn (ka),
n=0
whence 2
1
An = f () cos(n) d, n = 0, 1, . . . ,
2Jn (ka) 0
and 2
1
Bn = f () sin(n) d, n = 1, 2, . . . .
2Jn (ka) 0
48 CHAPTER 1. ELLIPTIC PROBLEMS
1 2 1 1
2
( R ) T + 2 R (sin T ) 2 RT = 0
sin
2 1
( R ) (sin T )
2 2
= sin = .
R T
d 2
(1) ( R ) ( 2 2 + )R = 0,
d
and
1
(2) (sin T ) + T = 0.
sin
Performing the change of variables x = cos in equation (2) (and using the
conditions |T (0)| < , |T ()| < ), we nd the eigenvalues and eigenfunctions
Equation (1) is readily reduced, via the substitution v() = R(), to the
form (for each n)
2
1
2 v + v ()2 + n + v = 0.
2
vn () = CIn+1/2 (),
where In+1/2 (x) are the Bessel functions of half-integer order and imaginary ar-
gument. Then
In+1/2 ()
Rn () = .
Therefore, the solution of our problem is given by the series
In+1/2 ()
u(, ) = Cn Pn (cos ),
n=0
where the constants Cn are determined from the boundary conditions. Specically,
In+1/2 (a)
u0 cos = Cn Pn (cos ).
n=0
a
This yields C1 = u0 a/I3/2 (a) (the remaining coecients are equal to zero).
Finally,
a I3/2 ()
u(, ) = u0 cos .
I3/2 (a)
Example 2 [6, Ch. IV, 18.51]. Solve the Neumann problem for the equation u +
k2 u = 0 in the interior as well as in the exterior of the sphere = R, under the
condition u/n|=R = A, where A is a constant.
Solution. (a) The interior Neumann problem can be written as follows:
1 u
2 + k2 u = 0, 0 < < R, 0 < < , 0 < 2, (1.41)
2
u
= A, 0 , 0 < 2. (1.42)
n =R
Since 12
2 u
= (u) , equation (1.41) can be recast as
v + k2 v = 0, v() = u().
1.13. HELMOLTZ EQUATION IN A BALL 51
and consequently
cos(k) sin(k)
u() = C1 + C2 .
Since the solution must be bounded at the center of the ball, we must put C1 = 0,
and so
sin(k)
u() = C .
Now let us calculate the normal derivative:
u u k cos(k) sin(k)
= =C .
n 2
Rk cos(kR) sin(kR)
C = A,
R2
whence
AR2
C= .
kR cos(kR) sin(kR)
We conclude that the solution of the interior problem has the form
AR2 sin(k)
u() = .
kR cos(kR) sin(kR)
v + k2 v = 0, v() = u().
Therefore,
eik eik
u() = C1 + C2 .
eik
Let us verify that the function u1 () = satises the Sommerfeld condition
u1
iku1 = o(1 ) ( ),
i.e., that
u
lim iku1 = 0.
Indeed,
ik
u 1 eik e 1
iku1 = iku1 u1 iku1 =
and .
ik
It follows that to pick a unique solution we must set C2 = 0, and then u() = C e .
Now let us calculate the normal derivative:
ik
u u e
= = C .
n
We have
u Ceik
= (1 ik),
n 2
and the boundary condition (1.44) yields
AR2
C= .
eikR (1 ikR)
Thus, the solution of the exterior Neumann problem is given by the formula
AR2 eik
u() = .
eikR (1 ikR)
Example 3 [6, Ch. V, 18.53]. Solve the Dirichlet problem for the equation u
k2 u = 0 in the interior and in the exterior of the sphere of radius = R with the
condition u|=R = A, where A is a constant.
Solution. (a) First let us solve the interior Dirichlet problem
u k2 u = 0, 0 < < R, 0 < < , 0 < 2,
u|=R = A, 0 , 0 < 2.
1.13. HELMOLTZ EQUATION IN A BALL 53
Therefore,
sinh(k) cosh(k)
u() = C1 + C2 .
where C1 and C2 are arbitrary constants.
Note that cosh(k)/ as 0. Hence, we must put C2 = 0, and the
solution has the expression
sinh(k)
u() = C .
In this case
ek ek
u() = C1 + C2 .
Since the solution of the exterior problem must satisfy u() 0 when , we
must put C1 = 0. Therefore,
ek
u() = C .
where Anm and Bnm are arbitrary constants. Each of these solutions corresponds to
a certain TM-wave, which can propagate without damping in the given waveguide.
Remark 1. The propagation of a TE-wave in an innitely long cylinder is associated
with the solvability of the Neumann problem for the Helmoltz equation:
1 H3 1 2 H3
+ 2 + 2 H3 = 0, 0 < < a, 0 < 2,
2
H3
(a, ) = 0, 0 2;
n
here n is the unit outer normal of the cylindrical waveguide.
By analogy with the preceding example, we obtain
where now nm = nm/a, with nm being the mth positive root of the equation
dJn (x)/dx = 0, n = 0, 1, 2, . . .
Remark. 2 If the component E3 (or H3 ) is known, then the other components of
the electric and magnetic eld vectors can be found by only one dierentiation
(this follows from the Maxwell equations for the electromagnetic eld).
Suppose one wants to solve the Laplace equation uxx + uyy = 0 with some
boundary condition in a domain of complicated shape in the plane of the variables
x, y. This boundary value problem can be transformed into a new boundary value
problem, in which one is required to solve the Laplace quation u ) = 0 in a
) + u
simpler domain of the variables , , and such that the second domain is obtained
from the rst one via a comformal mapping = f (z), where z = x + iy, = + i,
and u)() = u(z) for = f (z) (Figure 1.6).
y Original boundary
condition New bounary
condition
Domain
of complex
shape Domain
uxx + uyy = 0 Conformal of simple
mapping shape
u + u = 0
0 x 0
Figure 1.6.
Once the solution u )(, ) of the Laplace equation in a simple domain (disc,
half-space, rectangle) is found, it suces to substitute in that solution the expres-
sions = (x, y), = (x, y) in order to obtain the solution u(x, y) of the original
problem, expressed in the original variables.
Let us give several examples to show how to solve boundary value problems
for the Laplace equation (in the plane) by means of conformal mappings.
Example 1 [6, Ch. V, 17.13(4)]. Find the solution of the equation u = 0 in the rst
quadrant x > 0, y > 0, with the boundary conditions u|x=0 = 0, u|y=0 = (x 1),
where (x) = 1 if x > 0, (x) = 0 if x 0 is the Heaviside function.
z = x + iy = + i
4 = z2
4
3 2
2 3
6
1 x 3 0 1
Figure 1.7.
1.15. METHOD OF CONFORMAL MAPPINGS 57
Solution. Clearly, the function = z 2 , dened in the rst quadrant of the complex
z-plane, maps this domain into the entire half-plane > 0 of the complex -plane
(Figure 1.7), in such a manner that:
the positive x-semiaxis is mapped into the positive real -semiaxis;
the positive y-semiaxis is mapped into the negative real -semiaxis.
Thus, we arrive at the following conclusion:
1 1 y 2 x2 + 1
u(x, y) = arctan .
2 2xy
Example 2 [6, Ch. V, 17.14(2)]. Find the solution of the Dirichlet problem for
the equation u = 0 in the strip 0 < y < , with the boundary conditions
u|y=0 = (x), u|y= = 0.
Solution. The complex function = ez , dened in the strip 0 < y < , maps this
strip into the entire half-plane > 0 of the complex -plane (Figure 1.8), in such
a manner that:
the positive x-semiaxis is mapped into the positive -semiaxis [1, );
the negative x-semiaxis is mapper into the interval (0, 1);
the line y = is mapped into the negative -semiaxis.
58 CHAPTER 1. ELLIPTIC PROBLEMS
z = x + iy = + i
= ez
u = 0
u = 0
1
u = (x) 0 x u = 0 u = 1
Figure 1.8.
1 1 1
)(, ) =
u arctan ,
2
1 1 ex cos y
u(x, y) = arctan .
2 sin y
Example 3 [6, Ch. V, 17.18]. Find the solution of the Dirichlet problem
u = 0, Re z > 0, |z 5| > 3,
u|Re z=0 = 0, u||z5|=3 = 1.
Solution. First let us draw the domain D where we must solve the Dirichlet problem
(Figure 1.9). It can be regarded as a eccentric annulus (indeed, a line is a circle of
innite radius). Let us nd a conformal mapping of D onto a concentric annulus.
To this purpose let us nd two points that are simultaneously symmetric with
respect to the line Re z = 0 and with respect to the circle |z 5| = 3. Clearly, such
points must lie on the common perpendicular to the line and the circle, i.e., on
the real axis. From the symmetry with respect to the line Re z = 0 it follows that
these are precisely the points x1 = a and x2 = a with a > 0. The symmetry with
1.15. METHOD OF CONFORMAL MAPPINGS 59
1
(
u = 0, < || < 1, (1.48)
3
(|||=1 = 0,
u (|||= 13 = 1.
u (1.49)
60 CHAPTER 1. ELLIPTIC PROBLEMS
Let us solve the problem in the annulus 1/3 < || < 1 (in the plane (, )).
Since the boundary conditions (1.49) do not depend on the polar angle , it is
natural to assume that the solution u )() depends only on the variable (here
= cos, = sin ). To nd this solution, we rewrite the equation )
u in the
form u) = 0. The general solution of this equation is
)() = c1 + c2 ln ,
u
where c1 and c2 are arbitrary constants. Imposing the conditions (1.49), we obtain
c1 = 0, c2 = 1/ ln 3. Therefore,
1
)() =
u ln ||, because = ||.
ln 3
To nd the solution of the original problem it suces to return to the variable z,
using (1.47), which nally yields
1 z + 4
u(z) = ln .
ln 3 z 4
We shall assume that the function u(x) is continuous together with it rst-
order derivatives in the closed domain Rn , bounded by a suciently smooth
hypersurface , and has second-order derivatives that are square integrable in
. Here n is the outward unit normal to and 1 , 2 are given real numbers
satisfying 12 + 22 = 0; x = (x1 , . . . , xn ).
The Green function method for solving such problems consists in the follow-
ing. First we solve the auxiliary problem (see [1])
G = (x, x0 ), x0 ,
G (1.51)
1 G + 2 n = 0,
where = (x, x0 ) is the -function, which can formally be dened by the relations
0, if x = x0 , 1, if x0 ,
(x, x0 ) = , (x, x0 )dx =
, if x = x0 , 0, if x0
/ ,
1.16. THE GREEN FUNCTION METHOD 61
where x0 = (x01 , . . . , xn0 ) (the notation dx is obvious). The main property of the
-function is expressed by the equality
f (x0 ), if x0 ,
(x, x0 )f (x)dx =
0, if x0
/ ,
(where a is a vector eld and (a, n) denotes the scalar product of the vectors a
and n) if one puts successively a = vu and a = uv and subtract the results
from one another. Indeed, we have
v(u, n)ds = div(vu)dx, (1.53)
and
u(v, n)ds = div(uv)dx. (1.54)
Since (u, n) = u/n, (v, n) = v/n, div(vu) = (u, v) + vu and
div(uu) = (u, v) + uv, subtracting (1.54) from (1.53) we get the second
Green formula.
Now let us put v = G in (1.52). Then, since u = f (x) and G = (x, x0 ),
we obtain
u G
G(x, x0 )f (x)dx + u(x)(x, x0 )dx = G u ds.
n n
1 = 1, 2 = 0, G| = 0, u| = g
in the form
G
u(x0 ) = g ds G(x, x0 )f (x)dx;
n
in the form
u(x0 ) = Gg ds G(x, x0 )f (x)dx.
admits the following physical interpretation: the right-hand side of the equation is
regarded as an external action on the system and is decomposed into a continual
contribution of source distributed over the domain . Then one nds the response
of the system to each such source and one sums all these responses.
Construction of the Green function. One of the methods for constructing the Green
function is the reection method . For example, the Green function for the Poisson
equation in the case of the half-space (z > 0) has the form
1 1
G(M, M0 ) = ,
4RM M0 4RM M1
where RAB denotes the distance between the points A and B, M0 (x0 , y0 , z0 ) is a
point lying in the uper half-plane z > 0, M1 (x0 , y0 , z0 ) is the point symmetric
to M0 (x0 , y0 , z0 ) with respect to the plane z = 0, and M (x, y, z) is an arbitrary
point of the half-plane z > 0.
Physically the Green function can be interpreted as the potential of the eld
produced by point-like charges placed at the point M0 (over the grounded plane
z = 0) and the point M1 (Figure 1.11).
1.16. THE GREEN FUNCTION METHOD 63
M0 (x0 , y0 , z0 )
M (x, y, z)
RM M 0
0
y
RM M 1
x
M1 (x0 , y0 , z0 )
Notice that in the case of a half-plane (y > 0) the Green function has the
form (Figure 1.12)
1 1 1 1
G(M, M0 ) = ln ln .
2 RM M0 2 RM M1
M0 (x0 , y0 )
M (x, y)
0 x
M1 (x0 , y0 )
Example 1 [3, no. 244]. Find a function u(x, y), harmonic in the half-plane y > 0,
if it is known that
x
u(x, 0) = 2 .
x +1
Solution. We must calculate the integral
y s
u(x, y) = ds.
(1 + s2 )[(s x)2 + y 2 ]
Apparently, the easiest way to do this is to use the method of residues, namely,
the following formula:
s
2 2 2
ds = 2i[res[f (z)]z=i + res[f (z)]z=x+iy ,
(1 + s )[(s x) + y ]
where f (z) = z/ (1 + z 2 )[(z x)2 + y 2 ] .
Since
1 x + iy
res[f ]z=i = , res[f (z)]z=x+iy = ,
2[(i x)2 + y 2 ] 2iy[1 + (x + iy)2 ]
it follows that
y s iy x + iy
ds = + =
(1 + s2 )[(s x)2 + y 2 ] [(i x)2 + y 2 ] [1 + (x + iy)2 ]
iy x + iy
= + =
[(i x) + iy][(i x) iy] (x + iy 1)(x + iy + 1)
1.16. THE GREEN FUNCTION METHOD 65
1 1 1 1 1 1
= + + =
2 i x iy i x + iy 2 x + iy 1 x + iy + 1
1 1 1 1 1
= + + =
2 i(1 y) x i(1 + y) x i(y 1) + x i(1 + y) + x
1 1 1
= =
2 i(1 + y) + x i(1 + y) x
1 x i(1 + y) x + i(1 + y) x
= + = 2 .
2 x2 + (1 + y)2 x2 + (1 + y)2 x + (1 + y)2
Therefore, the solution of the problem is given by
x
u(x, y) = .
x2 + (1 + y)2
1
ln * = 0.
x + (y + a)2
2
1 1
ln * = 0, or ln * = 0,
x x2 + (y + a)2 x x2 + (y + a)2
*
i.e., (x/r2 ) = 0, where r = x2 + y 2 .
Thus, the function u = x/[x2 + (y + a)2 ] is harmonic in the upper half-plane.
Imposing the boundary condition, we conclude that the solution of our Dirichlet
problem is the function
x
u(x, y) = 2 .
x + (y + 1)2
Example 2 [6, Ch. V, 17.4(2)]. Find the solution of the Dirichlet problem
u = 0, < x, y < , z > 0,
u|z=0 = cos x cos y, < x, y < .
66 CHAPTER 1. ELLIPTIC PROBLEMS
z cos(u + x) cos(v + y) du dv
u(x, y, z) = =
2 (u2 + v 2 + z 2 )3/2
z (cos u cos x sin u sin x)(cos v cos y sin v sin y) du dv
= =
2 (u2 + v 2 + z 2 )3/2
z cos u cos v du dv
= cos x cos y
2 (u2 + v 2 + z 2 )3/2
because the other three integrals vanish thanks to the fact that their integrands
are odd functions.
Now let us calculate the integral
cos u cos v du dv
J= =
(u2 + v 2 + z 2 )3/2
[cos(u + v) + sin u sin v]du dv cos(u + v) du dv
= = ,
(u2 + v 2 + z 2 )3/2 (u2 + v 2 + z 2 )3/2
1 1
p = (u + v), q = (u v),
2 2
cos( 2p) dp dq dq
= cos( 2p) dp .
(p2 + q 2 + z 2 )3/2 (p2 + q 2 + z 2 )3/2
1.16. THE GREEN FUNCTION METHOD 67
*
But the substitution q = p2 + z 2 tan t transforms the integral
dq
J1 =
(p2 + q 2 + z 2 )3/2
into /2
| cos t| 2
J1 = 2 2
dt = 2 .
/2 p +z p + z2
Finally, the resulting integral
cos( 2p) dp
J =2
p2 + z 2
e 2z 2 2z
= 4i = e .
2zi z
Therefore the solution is
u(x, y, z) = e 2z
cos x cos y.
Remark 3. Since y/[(t x)2 + y 2 ] = Re[1/(i(t z))], where z = x + iy, the Poisson
formula (1.55) can be recast as
1 u(t) dt
u(z) = Re . (1.56)
i tz
Now let us consider the Dirichlet problem for the Laplace equation in the
half-plane Im z > 0 (i.e., for y > 0):
u = 0, < x < , y > 0,
u|y=0 = R(x), < x < ,
where the rational function R(z) is real, has no poles on the real axis, and R(z) 0
when z . By (1.56), the solution of this problem is the function
1 R(t) dt
u(z) = Re .
i tz
68 CHAPTER 1. ELLIPTIC PROBLEMS
where the residues are taken for all poles of the function R(z) in the lower half-
plane Im z < 0.
Example 3. Solve the Dirichlet problem
u = 0, < x < , y > 0,
k
u|y=0 = , k = const, < x < .
1 + x2
Solution. Using formula (1.57), we have
k k k(y + 1)
u(z) = 2 Re res = 2 Re = 2 .
(1 + 2 )( z) =i 2i(z + i) x + (y + 1)2
Biharmonic equation.
Example 1. Solve the following boundary value problem in the disc {(, ) : 0
a, 0 < 2}:
2
u = 0 in the disc,
u
u|=a = 0, = A cos on the boundary of the disc.
n
=a
But the solution of this problem is clearly v = a cos . Then, by the uniqueness of
the solution of the Dirichlet problem for the Laplace equation, we have the identity
2
1 (a2 2 ) cos d
= cos .
2 0 + a2 2a cos( )
2 a
Therefore, the solution of our problem is
A(2 a2 )
u(, ) = cos .
2a2
Example 2. Solve the following boundary value problem in the disc {(, ) : 0
a, 0 < 2}:
2
u = 1 in the disc,
u
=a
u| = 0, =0 on the boundary of the disc.
n =a
Solution. One can consider that the solution of the problem depends only on the
variable , i.e., u = u(). Next, let us remark that
4
2 2 3 1 2 1 u
u= + 2 u+ 3 ,
4 3 2
and so we obtain a boundary value problem for an ordinary dierential equation:
4u 2 3u 1 2u 1 u
4 + 3 2 2 + 3 = 1, 0 < < a, (1.61)
du
u|=a = 0, = 0. (1.62)
d =a
70 CHAPTER 1. ELLIPTIC PROBLEMS
du
Let us denote v = d . Then we obtain a third-order equation for the function
v = v():
3 v + 22 v v + v = 3 ,
which is recongnized to be the well-known Euler equation. Its general solution is
given by the function
1 3
v() = C1 1 + C2 ln + A + .
16
4u 4u 4u
4
+ 2 2 2 + 4 = e2y sin x. (1.65)
x x y y
We shall seek for u(x, y) in the form u(x, y) = f (y) sin x, where the function
f (y) is subject to determination. Substituting this expression in equation (1.65)
we obtain
whence
f (iv) 2f + f = e2y . (1.66)
The general solution of equation (1.66) has the form
1 2y
f (y) = C1 ey + C2 yey + C3 ey + C4 yey + e .
9
The constants C1 and C2 are equal to zero: otherwise, f (y) as y .
Hence,
1
f (y) = C3 ey + C4 yey + e2y .
9
The constants C3 and C4 are found from the boundary conditions (1.64), which
translate into f (0) = 0 and f (0) = 0. We have
1 1 1
f (y) = ey + yey + e2y .
9 9 9
Thus, the solution of our problem is
1 2y
u(x, y) = (e ey + yey ) sin x.
9
The Laplace and Poisson equations.
Example 4. Solve the following boundary value problem in the half-space {(x, y, z) :
< x, y < , z > 0}:
u = zez sin x sin y in the half-space,
u|z=0 = 0.
2u 2u 2u
u = + 2 + 2 = f sin x sin y f sin x sin y + f sin x sin y,
x2 y y
f 2f = zez .
72 CHAPTER 1. ELLIPTIC PROBLEMS
12. Find the steady temperature distribution inside a homogeneous and isotropic
ball if its surface is maintained at the temperature A sin2 (A = const).
13. Find the distribution of the potential in a spherical capacitor 1 < r < 2 if the
inner and outer plates have the potential V1 = cos2 and V2 = 18 (cos2 + 1),
respectively.
14. Find the temperature distribution inside a spherical layer 1 < r < 2 if the
inner sphere is maintained at the temperature T1 = sin sin , whereas the
outer sphere is maintained at the temperature of melting ice.
15. Solve the Dirichlet problem for the Poisson equation u = ey sin x in the
square 0 x , 0 y , with null boundary condition.
16. Solve the Dirichlet problem for the Poisson equation u = x4 y 4 in the
disc of radius one, with null boundary condition.
17. Solve the Dirichlet problem for the Poisson equation u = z in the ball of
radius one, with null boundary condition.
18. Solve the Dirichlet problem for the Poisson equation u = J0 R1 in a
cylinder of radius R and height h, with null boundary conditions.
19. Find the eigenoscillations of a rectangular membrane when two opposite edges
are clamped and the other two are free.
20. Find the eigenoscillations of a circular cylinder under null boundary condi-
tions of the rst kind.
21. Find the steady concentration distribution of an unstable gas inside a sphere
of radius a if a constant concentration u0 is maintained at the surface of the
sphere.
22. Solve the Dirichlet problem for the equation u + k2 u = 0 in the interior
and in the exterior of the sphere = R under the condition u|=R = A
(A = const).
23. Solve the Neumann problem for the equation u k2 u = 0 in the interior
and in the exterior of the sphere = R under the condition u/n|=R = A
(A = const).
24. Find the steady distribution of potential in the rst quadrant x > 0, y > 0 if
the half-line y = 0 is grounded while the half-line x = 0 is maintained at the
potential V .
25. Find the steady temperature distribution in the strip 0 < y < if the
temperature on the lower boundary y = 0 equals A cos x while the upper
boundary is kept at the temperature of melting ice (A = const).
26. Find the steady distribution of potential in the strip 0 < y < , x > 0 if
the horizontal sides of the strip are grounded and the vertical side has the
potential V .
1.18. PROBLEMS FOR INDEPENDENT STUDY 75
35. Solve the following Neumann problem for the Laplace equation in the spher-
ical layer 1 < < 2:
u = 0 inside the layer,
u
= P2 (cos ),
n =1
u
= P3 (cos ).
n =2
36. Solve the following boundary value problem in the disc {0 a, 0 <
2}: 2
u = x2 + y 2
in the disc,
u
u|=a = 0, n = 0.
=a
76 CHAPTER 1. ELLIPTIC PROBLEMS
37. Solve the following boundary value problem in the disc {0 a, 0 <
2}: 2
u = 0 in the disc,
u
=a
u| = 1, = sin3 .
n =a
38. Solve the following boundary value problem in the ball {0 a}:
2 2 2 2
u = x + y + z inside the ball,
u
u|=a = 0, n = 0.
=a
40. Solve the following boundary value problem in the lower half-plane (y < 0):
u = 0 < x < , y < 0,
2x
u|y=0 = .
(1 + x2 )2
1.19. Answers
ln b/
1. u1 = u2 + (u1 u2 ) .
ln b/a
1/r 1/b
2. u = u2 + (u1 u2 ) .
1/a 1/b
& ( '
m n n2 m2
3. u = Anm sin x sin y sinh + 2 z , where
n=1 m=1
a b a2 b
16V
+ , if n and m are odd,
2 m2
Anm = 2 nm sinh na2 + b2 c
0, if n or m is even,
1 1 2a sin
4. u = + arctan 2 , where a is the radius of the cylinder.
2 a 2
q
5. u = cos + c, where k is the heat conduction coecient of the cylinder.
k
q qb2
6. u = (2 a2 ) ln , where q = q0 /k, q0 = 0.24J 2 R, R is the re-
4 2 a
sistance per unit of length of the conductor, and k is the heat conduction
coecient.
2 a sin
7. u = V1 + (V2 V1 ) arctan 2 .
a2
(2m+1)
4qa sin a x sinh (2m+1)
a y
8. u = 2 .
k m=0 (2m + 1)2 cosh (2m+1) b
a
9. u = A cos + B sin , where a is the radius of the cylinder.
a a
(2n+1)
4V sin h z I0 (2n+1)
h
10. u = .
n=0 2n + 1 I (2n+1) R 0 h
sinh z)
m
a (l
J0 , where Am = k2 2aq
m
11. u = Am , k is the
m a m J1 (m )
m=0 cosh l z
heat conduction coecient, and m is the mth positive root of the equation
J0 (x) = 0.
2
2 r2 3 cos2 1
12. u = AA , where a is the radius of the ball.
3 a 3
1 3 cos2 1
13. u = + .
3r 3r3
1 8
14. u = r + 2 sin sin .
7 r
1
15. u = (yey sinh e sinh y) sin x.
2 sinh
1 2 4
16. u = ( 1) cos(2).
32
78 CHAPTER 1. ELLIPTIC PROBLEMS
1 3
17. u = (r r) cos .
10
, R2 R2 sin R1 z -
1 1 1
18. u = cosh z 1 + 2 1 cosh h J0 .
21 R 1 R sin 1 h R
R
2m2 n2
19. m,n = + 2 , m = 1, 2 . . . , n = 1, 2, . . . , where a and b are the side
a2 b m n
lengths of the membrane; um,n = sin x cos y .
a a
2 & (n) '2
k m (n)
20. m,n,k = + , n = 0, 1, . . . , m, k = 1, 2, . . . , where m is the
h a
mth positive root of the equation Jn (x) = 0, h is the height of the cylinder,
and a is its radius;
& (n) '
k m cos(n)
vn,m,k = sin z Jn .
h a sin(n)
a sinh(k)
21. u = u0 , where k is taken from the equation u k2 u = 0.
sinh(ka)
AR sin(k) AR eik
22. u = if R, and u = if R.
sin(kR) eikR
2V y
24. u = arctan .
x
A
25. u = cos x sinh( y).
sinh
V 2 sinh x sin y
26. u = arctan .
sinh2 x sin2 y
1 z2
27. u = ln 2 + ln .
ln(2/3) z 26
1 z + 4i
28. u = ln .
ln 3 z 4i
1.19. ANSWERS 79
z+1
31. u = .
[x2 + (z + 1)2 + y 2 ]3/2
32. u = (e 2z
ez ) sin x cos y.
A B C
33. u = cos(2) + 5 cos(5) + 10 cos(10).
2
sin
34. u = .
2 + 4 cos + 4
1 2 32 1 1 3
35. u = + 3 P2 (cos ) + 43 + 4 P3 (cos ) + C,
31 2 2 47
where C is an arbitrary constant.
a6 6 2
36. u = 3 +2 .
576 a a
a2 2 3 1 3
37. u = 1 sin sin(3) .
2a 4 a 4 a
a6 6 2
38. u = 3 +2 .
840 a a
2+ 2z 2 2z z
39. u = Ae + (1 A)e e sin x cos y, where
. .
1 1 1 1 1
A= 1+ + 1 .
2 2 2 2 2
2x(1 y)
40. u = .
[x2 + (1 y)2 ]2
http://www.springer.com/978-3-0348-0267-3