Dirichlet & Heat Problems in Polar Coordinates

Section 13.1
1

Steady State Temperature in a circular Plate

Consider the problem

uxx (x, y) + uyy (x, y) = 0, x2 + y 2 < R2
u(x, y) = f (x, y), x2 + y 2 = R2 ,
u(x, y) bounded on x2 + y 2 R2 .

(1.1)
(1.2)

u=f
R

It turns out that we cannot solve this problem using separation of variables as it is written.
But, as you will see, if we change coordinates to polar coordinates then separation of variables
works fine.
To this end recall that polar coordinates are given by
x = r cos(), y = r sin()
or
r 2 = x2 + y 2 ,

= tan1 (y/x).

So we need to translate
uxx (x, y) + uyy (x, y) = 0
into the variables r and . First we use r2 = x2 + y 2 and implicit differentiation to compute
2rrx = 2x rx =

x
y
, 2rry = 2y ry =
r
r

so we have

x
y
= cos() and ry = = sin().
r
r
Similarly, differentiating y = r sin() with respect to x and using (1.3) we have
rx =

0 = rx sin() + r cos()x
= cos() sin() + r cos()x
1

(1.3)

(1.4)
(1.5)

which implies
x =

sin()
.
r

(1.6)

Differentiating x = r cos() with respect to y and using (1.3) we have

0 = ry cos() r sin()y
= cos() sin() r sin()y
and we obtain
y =

(1.7)
(1.8)

cos()
.
r

(1.9)

Next, using the chain rule and (1.3) we compute

ux = ur rx + u x = ur cos() u
and
uy = ur ry + u y ur sin() + u

sin()
r

(1.10)

cos()
.
r

(1.11)

Using the formulas (1.10) and (1.11) we now compute the second derivatives:
sin()
uxx = (urr rx + ur x ) cos() (ur rx + u x )
r


sin()
+ ur (cos())x u
r

x

sin()
= urr cos() + ur
cos()
r



sin()
sin()
ur cos() + u
r
r


(r cos()x sin()rx )
+ ur ( sin())x u
r2


 2 
sin() cos()
sin ()
2
= urr cos () 2ur
+ u
r
r2


sin2 ()
cos() sin() sin() cos()
u
.
+ ur
r
r2
Finally we arrive at
uxx


 2 
sin() cos()
sin ()
= urr cos () 2ur
+ u
r
r2


2
sin ()
(2 sin() cos())
+ ur
+ u
r
r2
2

Exactly the same type of calculation which begins with


uy = ur ry + u y = ur sin() + u
2

cos()
r

(1.12)

leads to
uyy


 2 
cos ()
sin() cos()
+ u
= urr sin () + 2ur
r
r2


cos2 ()
(2 sin() cos())
+ ur
u
r
r2
2

(1.13)

Now adding (1.12) and (1.13) leads to

1
1
uxx + uyy = urr + ur + 2 u .
r
r

(1.14)

With this we can rewrite the problem (1.1) as

1
1
urr + ur + 2 u = 0, 0 < r < R, 0 2
r
r
u(R, ) = f (), 0 2
u(r, ) bounded.

(1.15)
(1.16)

Remark 1.1. We could just as easily have considered the following problem.
1
1
urr + ur + 2 u = 0, 0 < r < R,
r
r
u(R, ) = g(),
u(r, ) bounded.

(1.17)
(1.18)

In other words we could write the problem on any 2 interval. The only difference is that
the function f (x) and g(x) would be different since we are assuming that the given function
is 2 periodic. So if we intend to use the interval < < instead of 0 < < 2 and
we intend to use the same periodic function then we need to make some adjustments. For
example, if
(
(
f1 () , 0 < <
f2 ( 2) , < < 0
f () =
g() =
.
f2 () , < < 2
f1 ()
,0 < <
As a specific example consider
(

,0 < <
f () =
2 , < < 2

g() =

, < < 0
.
, < < 2

The advantage of the interval [, ] is that in the integrations you are integrating over a
symmetric interval about zero, so you can use even and odd to your advantage.
Separation of variables proceeds as follows. We seek simple solutions to (1.15) in the form
u = ()(r).
3

Substituting this into the equation in (1.15) we have

1
1
(()(r))r + 2 (()(r)) = 0
r
r
1
1
() 00 (r) + () 0 (r) + 2 00 ()(r) = 0.
r
r

(()(r))rr +

Next we divide by ()(r) and multiply by r2 to obtain

r2 ( 00 + (1/r) 0 )
00
=

which as usual in separation of variables must equal a constant .

Recall the solutions for so-called Euler-Cauchy equations:
r2 00 + ar 0 + b = 0
Consider the change of variables s = ln(r) or r = es . By the chain rule
d ds
1 d
d
=
=
dr
ds dr
r ds
and

d
d2
=
dr2
ds

d 1
dr r


=

1 dR
1 d 2

r2 ds2
r2 ds

So the equation becomes

r

1 d 2
1 d
2
2
2
r ds
r ds


+ ar

1 d
+ b = 0
r ds

which simplifies to
d
d 2
+ (a 1)
+ b = 0.
2
ds
ds
This is a constant coefficient equation and we recall from ODEs that there are three possibilities for the solutions depending on the roots of the characteristic equation. In the present
case we have a = 1 and b = .
Thus we obtain the pair of BVPs for ODEs:
00 () + () = 0,

() = (),

0 () = 0 ()

For = 0 we have () = 1. For nonzero we have = 2 and the general solution

is () = A cos() + B sin().
The first boundary condition implies B sin() = B sin() or sin() = 0. This
implies = n an integer. The second boundary condition implies A sin() =
A sin() which again gives = n so that = n2 and both A and B are
arbitrary. Thus we have = n2 and
() = Am cos(n) + Bm sin(n)

With = n2 with n = 0, 1, 2, as above

r2 00 (r) + r 0 (r) n2 (r) = 0
for which we seek solutions in the form (r) = rc . When n = 0 we get
r2 00 (r) + r 0 (r) = 0
which is an Euler-Cauchy problem with general solution (r) = a + b ln(r).
But in order for the solution to be bounded we need b = 0 so is an arbitrary
constant, say = 1.
For m 6= 0 we have again an Euler-Cauchy problem with general solution
= arn + brn .
Once again, for bounded we need b = 0, so we take
= rn .

Combining these results we seek our general solution in the form

u(r, ) = a0 +

rn [an cos(n) + bn sin(n)]

(1.19)

n=1

In the case of (1.15) where the boundary function is given on [0, 2]

f () = u(R, ) = a0 +

Rm [an cos(n) + bn sin(n)] .

n=1

This is a general Fourier series and we have

Z 2
1
a0 =
f () d
2 0
Z 2
1
f () cos(n) d
an =
Rm 0
Z 2
1
bn =
f () sin(n) d
Rm 0
In the case of (1.17) where the boundary function is given on [, ] (see Remark 1.1)
g() = u(R, ) = a0 +

Rm [an cos(n) + bn sin(n)] .

n=1

This is a general Fourier series and we have

1
a0 =
2

g() d

1
an =
Rm
bn =

1
Rm

g() cos(n) d

g() sin(n) d

Example 1.1.
1
1
urr + ur + 2 u = 0, 0 < r < 1,
r
r
u(1, ) = cos2 (),
u(r, ) bounded.
Using the trig identity
1 1
+ cos(2)
2 2
and our usual orthogonality conditions the solution is given by
cos2 () =

u(r, ) =

1 1 2
+ r cos(2)
2 2

Example 1.2.
1
1
urr + ur + 2 u = 0, 0 < r < 1,
r
r
u(1, ) = sin(3),
u(r, ) bounded.
u(r, ) = a0 +

rn [an cos(n) + bn sin(n)]

n=1

Now we need
sin(3) = u(1, ) = a0 +

[an cos(n) + bn sin(n)] .

n=1

But we can argue (using our knowledge of orthogonality) that the solution is given by
u(r, ) = r3 sin(3).

Notice that this solution can be transformed back into rectangular coordinates but it would
be a mess.
Example 1.3 (Integral Formula for Dirichlet Problem in a Disk). We recall that the Dirichlet
problem for for circular disk can be written in polar coordinates with 0 r R,
as
1
1
u = urr + ur + 2 u = 0
r
r
u(R, ) = f ().
6

As we have seen, we can obtain the solution u in the form

u(r, ) = a0 +

(r)n (an cos(n) + bn sin(n)) .

n=1

We need

f () = u(R, ) = a0 +

(an cos(n) + bn sin(n)) ,

n=1

and
1
a0 =
2

f () d
0

1
an =
Rn

1
bn =
Rn

f () cos(n) d,
0
2

f () sin(n) d.
0

Lemma 1.1. The series solution can thus be written as

Z 2
1
(R2 r2 ) f ()
u(r, ) =
d.
2 0 R2 + r2 2rR cos( )
Proof. To prove this result we need only insert the formulas for an and bn into the infinite
sum representation for the solution, interchange the sum and integral and sum a resulting
geometric series:
Z 2
Z

1
1 X  r n 2
u(r, ) =
f () d +
f () [cos(n) cos(n) + sin(n) sin(n)] d
2 0
n=1 R
0
1
=
2

1
=
2

1
=
2

1
=
2

1
=
2

0
2

1 X  r n 2
f () d +
f () cos[n( )] d
n=1 R
0
"
f () 1 + 2

n=1
2

"
f () 1 +

0
2


f () 1 +

 
X
r n
n=1

 
X
r n

#
cos[n( )] d
#
ein() + e

in()

rei()
rei()
+
R rei() R rei()

(R2 r2 )
f () d.
(R2 + r2 2rR cos( ))


d

Heat Equation in a Disk

Next we consider the corresponding heat equation in a two dimensional wedge of a circular
plate. So we write the heat equation with the Laplace operator in polar coordinates.
Example 2.1.



1
1
ut = k urr + ur + 2 u , 0 < r < 1, 0 /2
r
r
u(1, , t) = 0, 0 /2
u(r, 0, t) = 0, u (r, /2, t) = 0, 0 < r < 1,
u(r, , t) bounded,
u(r, , 0) = f (r, ) = (r r3 ) sin().

(2.1)

Separation of variables proceeds as follows. We seek simple solutions to (2.1) in the form
u = T (t)()R(r).
Substituting this into the equation in (2.1) we have


1 00
1
0
00
()R (r) + ()R (r) + 2 ()R(r)
T 0 (t)
r
r
=
= 2 .
kT (t)
()R(r)
But from the right hand side we can further see that there must be another constant which
we denote by 2


1 0
2
2
00
r R (r) + R (r) + R(r)
00 ()
r
=
= 2 .
R(r)
()
This leads to the following equations

T 0 (t) + k2 T (t) = 0
r2 R00 + rR0 + (2 r2 2 )R = 0
00 + 2 = 0
with boundary conditions
|R(0)| < ,

R(1) = 0
0

(0) = 0, (/2) = 0.
First we consider the equation for :
00 + 2 = 0, (0) = 0, 0 (/2) = 0
8

which gives
m = (2m 1),

2
Tm (t) = m () = sin((2m 1)).

The eigenfunction m () satisfy the orthogonality relations

Z

/2

m ()k () d = m,k .
0

Then for each m we have

r2 R00 + rR0 + (2 r2 (2m 1)2 )R = 0, |R(0)| < ,

R(1) = 0.

(2.2)

For each m the differential equation in (2.2) is a Bessel equation which has a regular singular
point at r = 0. The solutions are called Bessel functions of the first kind of order . The
problem (2.2) (i.e., the equation and boundary conditions) is a singular Sturm-Liouville
problem and the associated eigenvalues and eigenfunctions
m,n , Rm,n , m, n = 0, 1, 2, .
The eigenfunctions are orthogonal with the integral
Z 1
Rm,n (r)Rm,p (r) r dr = n,p .
0

We will use standard notation for the normalized eigenfunctions

Rm,n (r) = m,n BesselJ((2m 1), m,n r)
where
Z
m,n =

1

BesselJ ((2m 1), m,n r) r dr

The eigenvalues are the zeros of the Bessel function obtained from the boundary condition
R(1) = 0 by
BesselJ((2m 1), m,n ) = 0.
(2.3)
It is well known that for each m this equation has infinitely many solutions m,n which tend
to infinity with n.
The solution is given by
!

2 X X
2
m,n Cm,n ekm,n t BesselJ((2m 1), m,n r) sin((2m 1)) .
u(r, , t) =
m=1 n=1
For this example the special form of the initial condition f (r, ) = (r r3 ) sin() together
with the orthogonality relations in show that the solution reduces to
!

X
2
2
u(r, , t) =
1,n C1,n ek1,n t BesselJ(1, 1,n r) sin().
n=1
9

Also, in this case we can express 1,n by

1,n =

2
.
BesselJ(0, 1,n )

Thus we only need the eigenvalues 1,n for n = 1, 2, . We cannot solve for this values
exacly but we can solve numerically and, for example, we find
1,1 = 3.831705970, 1,2 = 7.015586670, 1,2 = 10.17346814
and a plot of y = BesselJ(1, ) for from 0 to 20 is given below

We obtain the corresponding Fourier coefficients C1,n from

Z
Z
1,n 2 /2 1
C1,n =
(r r3 )BesselJ(1, 1,n r) dr d
0
0
from which, using Maple, we obtain
C1,n
and

64 X
u(r, , t) =
n=1

16 2
= 3
1,n
2

ek1,n t BesselJ(1, 1,n r) sin()

31,n BesselJ(0, 1,n )

u(r, , 0)

u(r, , 3)
10

!
.