Math 342 Partial Differential Equations «Viktor Grigoryan

27 Laplace’s equation: properties

We have already encountered Laplace’s equation in the context of stationary heat conduction and wave phenomena.
Recall that in two spatial dimensions, the heat equation is ut −k(uxx +uyy )=0, which describes the temperatures
of a two dimensional plate. Similarly, the vibrations of a two dimensional membrane are described by the wave
equation in two spatial dimensions, utt −c2(uxx +uyy )=0. If one considers stationary heat and wave states, i.e.
not changing with time, then ut =utt =0, and both the heat and wave equations reduce to the stationary equation
uxx +uyy =0.
This is the two dimensional Laplace equation. Analogously, in three dimension one has the equation
uxx +uyy +uzz =0.
Using the notation ∆=∇·∇, we can rewrite Laplace’s equation in any dimension as
∆u=0. (1)
The operator ∆ is called Laplace’s operator, or Laplacian. To distinguish the Laplacian in different dimensions,
we will use the subscript notation ∆n, where n stands for the dimensions. The solutions of the Laplace equation
(1) are called harmonic functions. The inhomogeneous Laplace’s equation
∆nu=f(x1,x2,...,xn),
is called Poisson’s equation.
Besides describing stationary heat and wave phenomena, Laplace’s and Poisson’s equations come up in the
study of electrostatics, incompressible fluid flow, analytic functions theory, Brownian motion, etc.
Notice that in one dimension Laplace’s equation is the ODE uxx =0, so the only harmonic functions in one
dimension are the linear functions u(x)=A+Bx.
An obvious distinction between Laplace’s equation and the heat and wave equations is that the processes
described by Laplace’s equation do not involve dynamics or evolution of the initial data in time. Hence the
natural problem to study for Laplace’s equation is the boundary value problem in some given domain D. We
will consider the equation
∆u=f inD,
with either of the following conditions on the boundary
 ∂D of the domain D. 
 ∂u  ∂u 
(D): u∂D =h; (N):  =h; (R): +au  =h,
∂n ∂D ∂n 
∂D
where n is the outer normal vector to this boundary.
When considering heat and wave boundary value problems, we saw that the boundary in one dimension
consists of the endpoints of the interval (a,b), in which the equation is being solved. In two dimensions the
boundary will be a curve, while in three dimensions it is a surface, and we expect that the geometry of the
boundary will play a role in solving Laplace’s equation.
We will restrict our study of Laplace’s equation to two and three dimensions, and will occasionally use the
vector notation x=(x,y), or x=(x,y,z) to denote a point in either two dimensional, or three dimensional space.
Lets us start by first discussing the properties of Laplace’s equation.
27.1 Maximum principle
It turns out that harmonic functions obey a maximum principle, which is similar to the maximum principle
for the heat equation. In what follows, an open bounded connected set D is a set that does not contain any
of its boundary points (open), entirely lies inside some ball centered at the origin (bounded), and consists of
one piece, i.e. any two points of the set can be connected by a curve that entirely lies inside the set (connected).
Maximum Principle. Let D be a connected bounded open set (in either two or three dimensional space),
and u be a harmonic function in D, which is also continuous on the closure of the set, D =D∪∂D. Then the
maximum and minimum values of u are attained on the boundary of D. Moreover, the maximum and minimum
values cannot be attained inside D, unless u≡constant.

1
 Mathematically, this means that if u is a non-constant harmonic function in an open bounded connected
set D, then
max{u}<max{u}=max{u}, and min{u}=min{u}<min{u}.
D D ∂D ∂D D D
If we think of harmonic functions as equilibrium states in the heat conduction, then the maximum principle
makes perfect sense, since if there was a maximum temperature at an interior point of the domain D, there
would have been a heat flux from this point to the points with lower temperature, which would consequently
decrease the temperature of this point, making the state unsteady.
The idea of the proof of the maximum principle is similar to the one used in proving the maximum principle
for the heat equation. We start by noting that if, say in two dimensions, (x0,y0) is a maximum point, then both
uxx(x0,y0)≤0 and uyy (x0,y0)≤0 by the second derivative test. But then ∆u(x0,y0)=uxx(x0,y0)+uyy (x0,y0)≤0.
This would be a contradiction, if the inequality was strict, however second order derivatives may be zero at
extrema points.
To eliminate this scenario, we modify the function u, and consider the new function v(x)=u(x)+|x|2, where
>0 is a small constant. But then we have
∆2v =∆2u+∆2(x2 +y2)=0+4>0 inD,
and similarly in three dimensions. But ∆2v = vxx +vyy ≤ 0 at an interior maximum point, therefore v(x) has
no interior maximum points in D. Since v(x) is continuous, it must attain its maximum value somewhere in the
closed set D, so the maximum must be attained at some point x0 ∈∂D. Then for every point x in D, we have
u(x)≤v(x)≤v(x0)=u(x0)+|x0|2 ≤max{u}+l2,
∂D
where l is the largest distance from the origin to the boundary of the (bounded) set D. Since  was arbitrary,
we can make it go to zero, yielding
u(x)≤max{u}, for everyx∈D.
∂D
So the maximum of u must be attained on the boundary. The proof for the minimum is similar. The stronger
statement that the maximum cannot be attained inside D will be proved later via the mean value property
of harmonic functions.
27.2 Uniqueness of the Dirichlet problem
As was the case for the heat equation, the maximum principle directly implies uniqueness of the Dirichlet problem
for Poisson’s equation. Indeed, suppose that  the Dirichlet problem
∆u=f inD,
u=h on∂D,
where D is open bounded and connected, has two solutions u1,u2. Then their difference, w =u1 −u2, is harmonic,
and has zero Dirichlet data on the boundary ∂D. But by the maximum/minimum principle we have for any
point x∈D,
0=min{w}≤w(x)≤max{w}=0,
∂D ∂D
so w(x)=u1(x)−u2(x)≡0.
We will give an alternative proof of the uniqueness using the energy method in a subsequent lecture, which
will also show that solutions to the Neumann problem are unique up to a constant.
27.3 Invariance
Laplace’s equation is invariant under rigid motions, which are the translations, and rotations. A translation
is a transformation x→x0, which is given by x0 =x+a for some vector a. In two dimensions this vector equation
is equivalent to
x0 =x+a, y0 =y+b,
and it is easy to see that uxx +uyy =ux0x0 +uy0y0 =0. So if a function is harmonic in the variables (x,y), it must
also be harmonic in the variables (x0,y0). This is the invariance under translations. Clearly this holds in higher
dimensions as well.
For the invariance under rotations, we need to show that
 Laplace’s
  equation remains the  same in the variables
x0 =xcosα+ysinα x0 cosα sinα x
or = ,
y0 =−xsinα+ycosα, y0 −sinα cosα y
2
where α is the angle of rotation.
Using the chain rule, one can compute ux,uy , and then uxx,uyy in terms of the partial derivatives of u with
respect to (x0,y0) variables, and show that
uxx +uyy =(ux0x0 +uy0y0 )(cos2α+sin2α)=ux0x0 +uy0y0 .
Thus, Laplace’s operator is invariant under rotations in two dimensions.
One can prove the invariance under rotations in any dimension n=2,3,... using the matrix notation as follows.
In any dimension n a rotation is given by
Xn
x0 =Bx, or x0k = bkixi,
i=1
where B ={bij } is an orthogonal matrix, that is
n
X
t t
BB =B B =I, or bkibli =δkl ,
i=1
where δkl =1, if k =l, and δkl =0, if k =
6 l is the Kronecker symbol. Using the chain rule, we can compute
n n
∂ X ∂x0k ∂ X ∂
= 0
= bki 0 .
∂xi k=1 ∂xi ∂xk k=1 ∂xk
To compute the second order derivatives, we multiply ! the first! order derivative operator by itself.
2 n n n
∂ X ∂ X ∂ X ∂2
= b ki 0
· bli 0
= b b
ki li 0 0
.
∂x2i k=1
∂x k l=1
∂x l k,l=1
∂xk ∂xl
But then !
n n n n n n
X ∂2 X X ∂2 X X ∂2 X
l ∂2
∆x = 2
= b b
ki li 0 0
= b b
ki li 0 0
= δk 0 0
=∆x0 .
i=1
∂xi i=1 k,l=1
∂xk ∂xl k,l=1 i=1
∂xk ∂xl k,l=1
∂xk ∂xl
So Laplace’s operator is indeed invariant under rotations.
The rotation invariance also implies that Laplace’s equation allows rotationally invariant solutions, that is,
solutions that depend only on the radial variable r =|x|. We will call such solutions radial.
27.4 Radial solutions of Laplace’s equation
In order to find radial solutions to Laplace’s equation, we make a change to polar variables in two dimensions,
and to spherical variables in three dimensions. Notice that in this case the radial solution simply means that
u(r,θ)=u(r), or u(r,θ,φ)=u(r), that is, the function depends on only one variable, and, as a consequence, the
PDE will reduce to an ODE.
We first make
 a change to polar variables in two dimensions, for which the
 transformation 
formulas are
∂y
 ∂x
x=rcosθ ∂(x,y) ∂r ∂r
cosθ sinθ
with Jacobian matrix = ∂x ∂y = .
y =rsinθ, ∂(r,θ) ∂θ ∂θ
−rsinθ rcosθ
p
Using r = x2 +y2, one can compute the partial derivatives
∂r 2x x ∂r y
= p = =cosθ, = =sinθ.
∂x 2 x2 +y2 r ∂y r
Also, differentiating both sides of x=rcosθ with respect to x, we get
∂r ∂θ ∂θ ∂θ 1−cos2θ sinθ
1= cosθ−rsinθ =cos2θ−rsinθ ⇒ =− =− .
∂x ∂x ∂x ∂x rsinθ r
∂θ
can be computed similarly. So the Jacobian matrix of the inverse transformation is
∂y  ∂r ∂θ   
∂(r,θ) ∂x ∂x cosθ −sinθ/r
= ∂r ∂θ = .
∂(x,y) ∂y ∂y sinθ cosθ/r
By the chain rule, we will have
∂ ∂ sinθ ∂ ∂ ∂ cosθ ∂
=cosθ − , and =sinθ + .
∂x ∂r r ∂θ ∂y ∂r r ∂θ

3
Using these, one can compute 2  2
∂2 1 ∂ 1 ∂2

∂ ∂
∆2 = + = 2+ + .
∂x ∂y ∂r r ∂r r2 ∂θ2
And the Laplace equation can be written in polar variables as
1 1
urr + ur + 2 uθθ =0.
r r
For a radial solution u(r,θ)=u(r), the last term in the above equation will vanish, yielding the equation
1
urr + ur =0, or rurr +ur =0,
r
which is an ODE, as expected. The last equation can be written as
´ (rur )r =0,
where we used the integrating factor exp( 1r dr). Integrating the last equation gives
1
rur =c1, or ur =c1 .
r
Integrating once more gives the solution
u(r)=c1logr+c2.
Disregarding the constant solution c2, we see that the function logr =log|x| is harmonic in two dimensions.

To find radial solutions in three dimensions, we need to make a change to spherical variables, which is given
by the transformations p √
r = x2 +y2 +z2 = s2 +z2
p
s= x2 +y2
x=scosφ z =rcosθ
y =ssinφ s=rsinθ.
Thus, the transformation to spherical variables can be thought of as the pair of successive transformations
(x,y,z)→(s,φ,z)→(r,θ,φ).
Using the above computation in two dimensions, we have that
1 1
uzz +uss =urr + ur + 2 uθθ , and
r r
1 1
uxx +uyy =uss + us + 2 uφφ.
s s
Adding these two identities, and canceling the term uss on both sides, we get
1 1 1 1
∆3u=urr + ur + us + 2 uθθ + 2 uφφ. (2)
r s r s
We can also compute
∂u ∂r ∂θ ∂φ s cosθ
us = =ur +uθ +uφ =ur +uθ .
∂s ∂s ∂s ∂s r r
Then replacing us in (2) by the above expression, and substituting s=rsinθ for all occurrences of s, we obtain
Laplace’s equation in the spherical variables inthree dimensions 
2 1 1
urr + ur + 2 uθθ +cotθuθ + 2 uφφ =0.
r r sin θ
For a radial solution u(r,θ,φ)=u(r), the entire square brackets term will vanish, so Laplace’s equation will
reduce to the ODE
2
urr + ur =0.
r
Multiplying this equation by r2, we can write it as
´2 (r2ur )r =0,
where we used the integrating factor exp( r dr). Integrating this equation gives
1
r2ur =c1, or ur =c1 2 .
r
4
Integrating yet again, we obtain the solution
1
u(r)=−c1 +c2.
r
So the function 1/r =1/|x| is harmonic in three dimensions.
Notice that both 1/r and logr functions are not defined at the origin r = 0, but they will be harmonic on
any domain which does not contain the origin. We will see in subsequent lectures that these functions in the
context of Laplace’s equation play a role similar to that of the heat kernel in the context of the heat equation.
27.5 Conclusion
In this lecture we studied the maximum principle for Laplace’s equation, which trivially implies the uniqueness of
solutions to the Dirichlet problem for Poisson’s equation. We also saw that Laplace’s equation is invariant under
translations and rotations. The last fact accounted for existence of radial solutions, which are solutions that
are invariant under rotations, and hence depend only on the radial variable r. Making a change to polar variables
in two dimensions, and spherical variables in three dimensions, we were able to find radial harmonic functions by
solving the ODEs satisfied by these functions. We will see in a later lecture that these radial harmonic functions
play a crucial role in finding the solution to the Dirichlet problem for Laplace’s and Poisson’s equations.

5
28 Harmonic functions in rectangles and cubes

In this lecture we demonstrate how boundary value problems for Laplace’s equation can be solved by separation
of variables in the case of rectangles in two dimensions and cubes in three dimensions.
Let us consider the two dimensional Laplace’s equation in a rectangle D = (0,a) × (0,b), with boundary
conditions prescribed on the four edges of the rectangle,
I1 =(0,a)×{0}, I2 =(0,a)×{b}; I3 ={0}×(0,b), I4 ={a}×(0,b).
We choose some particular boundary conditions, say Neumann conditions on the entire boundary, and study
the problem 
uxx +uyy =0 inD =(0,a)×(0,b),
(3)
uy (x,0)=h(x), uy (x,b)=g(x); ux(0,y)=j(y), ux(a,y)=k(y).
Other boundary conditions, Dirichlet on the entire boundary, or mixed – Dirichlet on some of the edges, and
Neumann on the others, can be handled in much the same way.
Notice that the solution to problem (3) can be written as a sum u = u1 +u2 +u3 +u4, where each of the
summands ui solves Laplace’s equation in the rectangle with vanishing boundary data on all edges but Ii, for
i=1,2,3,4. It is then enough to find only one of ui’s, since finding the others will be similar. Let us, for example,
find u3, i.e. we would like to solve the boundary value problem
uxx +uyy =0 inD =(0,a)×(0,b),
(4)
uy (x,0)=0, uy (x,b)=0; ux(0,y)=j(y), ux(a,y)=0.
We first look for separated solutions u(x,y)=X(x)Y (y). Plugging this separated solution into the equation gives
X 00Y +Y 00X =0.
Dividing by XY , and separating the variables on different sides of the equation, we obtain
X 00 Y 00
=− =λ.
X Y
Clearly λ is independent of both x and y, and hence is a constant. So we have the following ODEs for the
components X and Y .
X 00 =λX, and Y 00 =−λY.
We next observe that the first two boundary conditions of (4) imply
uy (x,0)=X(x)Y 0(0)=0, Y 0(0)=0,
0 ⇒
uy (x,b)=X(x)Y (b)=0, Y 0(b)=0,
since we are looking for solutions that do not vanish everywhere, hence X(x)6≡ 0. We thus have the following
eigenvalue problem for the Y component. 
Y 00 =−λY,
(5)
Y 0(0)=Y 0(b)=0.
This eigenvalue problem was solvedbefore. The eigenvalues and the eigenfunctions are
nπ 2 nπy
λn = , Yn(y)=cos , forn=0,1,....
b b
Plugging these values of λ into the equation for X, we have
A0 B0
Forλ0 =0, X 00 =0 ⇒ X0(x)= + x.
2 2
 nπ 2  nπ 2
Forλn = ,n=1,2,..., X 00 = X ⇒ Xn(x)=Anenπx/b +Bne−nπx/b.
b b
One can alternatively use the hyperbolic sine and cosine functions to write Xn(x) = A0n cosh(nπx/b) +
Bn0 sinh(nπx/b).
Notice that the forth boundary condition of (4) implies
ux(a,y)=X 0(a)Y (y)=0 ⇒ X 0(a)=0.
This gives the following conditions for the coefficients An,Bn.
B0
X00 (a)=0 ⇒ =0,
0
2
and Xn(a)=0 gives
nπ nπ
An enπa/b −Bn e−nπa/b =0 ⇒ Bn =Ane2nπa/b.
b b 6
Thus, Xn(x)=Anenπx/b +Ane2nπa/be−nπx/b =An enπx/b +e−nπ(x−2a)/b .


Now using Xn(x) and Yn(y), we can write the series solution as
∞
A0 X
nπy
u(x,y)= + An enπx/b +e−nπ(x−2a)/b cos (6)
2 n=1 b
Finally, the inhomogeneous boundary condition of (4) implies
∞
X nπ
nπy
ux(0,y)=j(y)= An 1−e2nπa/b cos ,
n=1
b b
hence,
nπ jn
1−e2nπa/b =jn, or An =


An ,
b (1−e2nπa/b)nπ/b
where ˆ
2 b nπy
jn = j(y)cos dy, n=1,2,...,
b 0 b
are the Fourier cosine coefficients of j(y). The zeroth coefficient of j(y) vanishes due to the condition that the
integral of the Neumann data over the boundary must be zero, for the problem (3) to be well posed.
Substituting the coefficients An into the series (6) gives the solution
∞
A0 X jn nπx/b −nπ(x−2a)/b

nπy
u(x,y)= + e +e cos .
2 n=1 (1−e2nπa/b)nπ/b b
Notice that the solution is determined up to the constant A0/2, which is expected for a Neumann problem.
The strategy of solving boundary value problems for Laplace’s equation in a rectangle is always the same,
which we summarize in the following scheme.
(i) Look for separated solutions
(ii) Solve the eigenvalue problem for the component with two homogeneous boundary conditions
(iii) Find the other component using the obtained eigenvalues, and the third homogeneous boundary condition
(iv) Form the series solution, and find the coefficients from the inhomogeneous boundary condition
28.1 Cubes in three dimensions
A similar strategy works for cubes in three dimensions as well. Indeed, let us consider the following Dirichlet
problem. 
 ∆3u=uxx +uyy +uzz =0 inD =(0,π)×(0,π)×(0,π),
u(π,y,z)=g(y,z),
 u(0,y,z)=u(x,0,z)=u(x,π,z)=u(x,y,0)=u(x,y,π)=0
For the separated solution u(x,y,z)=X(x)Y (y)Z(z), the equation will reduce to
X 00 Y 00 Z 00
+ + =0.
X Y Z
The homogeneous boundary conditions imply
X(0)=Y (0)=Y (π)=Z(0)=Z(π)=0.
So we have the following
 00ODEs with the associated
 00 boundary conditions 00
Y =−λY, Z =−δZ, X =(λ+δ)X,
Y (0)=Y (π)=0, Z(0)=Z(π)=0, X(0)=0.
The eigenvalue problems for Y and Z give
λm =m2, Ym(y)=sinmy; δn =n2, Zn(z)=sinnz.
Using these values of λ and δ, we can solve the X equation with the homogeneous boundary condition, which
will give √
Xmn(x)=Amnsinh( m2 +n2x).

7
 The series solution will then be ∞ X
∞
X √
u(x,y,z)= Amnsinh( m2 +n2x)sinmysinnz. (7)
m=1 n=1
For this solution the inhomogeneous boundary condition implies
∞ X ∞
X √
u(π,y,z)=g(y,z)= Amnsinh( m2 +n2π)sinmysinnz. (8)
m=1 n=1
This is the Fourier series of the function g(y,z) in two variables, and one can find its coefficients in the same
way as before. Notice that the elements of the set {sinmysinnz}∞ m,n=1 are pairwise orthogonal in the sense of
the dot product ˆ πˆ π
(f,g)= f(y,z)g(y,z)dydz,
0 0
and ˆ πˆ π
π π π2
sin2mysin2nzdydz = · = .
0 0 2 2 4
So the coefficients formula in this case will beˆ πˆ π
4
gmn = 2 g(y,z)sinmysinnzdydz.
π 0 0
Then, the coefficients in the series (7) can be found from (8) to ˆ be
πˆ π
gmn 4
Amn = √ = √ g(y,z)sinmysinnzdydz,
sinh( m2 +n2π) π2sinh( m2 +n2π) 0 0
and the series solution can be written in terms of the Fourier
√ coefficients of g(y,z) as follows.
∞ X ∞
X sinh( m +n2x)
2
u(x,y,z)= gmn √ sinmysinnz.
m=1 n=1
sinh( m2 +n2π)
28.2 Conclusion
In the case of rectangular domains, we demonstrated how the separation of variables can be used to solve
boundary value problems for Laplace’s equation in two dimensions. This is a consequence of the fact that the
rectangle itself can be separated into a product of two intervals R=(0,a)×(0,b). The main idea was to break
any boundary value problem to ones with homogeneous boundary conditions on all but one of the edges of
the rectangle. Then using these homogeneous boundary conditions one can solve the resulting eigenvalue problem
and obtain the solution in a series form, the coefficients of which will be determined by the inhomogeneous
boundary condition. The same goes for cubes in three dimensions, where one can find the solutions as a double
series. We will see next time that the method of separation of variables can be applied to Laplace’s equation
in a disk as well, which can be thought of as a rectangle in polar coordinates.

8
29 Poisson’s formula

Let us consider the Dirichlet problem  for the circle x2 +y2 =a2,
uxx +uyy =0 inx2 +y2 <a2,
(9)
u=h(θ) onx2 +y2 =a2.
Notice that we can write the disk as a rectangle in the polar coordinates
{x2 +y2 <a2}=[0,a)×[0,2π),
so it makes sense to solve the Dirichlet problem (9) by separation of variables in polar coordinates. Recall that
Laplace’s equation in polar variables has the form
1 1
urr + ur + 2 uθθ =0.
r r
So for the separated solution u(r,θ)=R(r)Θ(θ), the equation will reduce to
1 1
R00Θ+ R0Θ+ 2 RΘ00 =0.
r r
Dividing this equation by RΘ, and multiplying by r2, and separating the variables on different sides, we get
R00 R0 Θ00
r2 +r· =− =λ.
R R Θ
Thus, we have the following ODEs for R and Θ.
r2R00 +rR0 −λR=0, Θ00 =−λΘ.
Notice that the Θ component satisfies periodic boundary conditions Θ(θ+2π)=Θ(θ) due to the nature of polar
coordinates. This gives the eigenvalue problem  00
Θ =−λΘ,
Θ(θ+2π)=Θ(θ).
The eigenvalues and eigenfunctions of this problem are
λn =n2, Θn(θ)=Acosnθ+Bsinnθ, n=0,1,....
Using these values of λ, we can solve the R equation.
Forλ0 =0, r2R00 +rR0 =0 ⇒ R(r)=C0 +D0logr.
For the positive values of λ, the equation for R is of Euler’s type, so it has power solutions R(r)=rα. Substituting
this into the equation gives
2 α−2 α−1 α 2 α
√ r α(α−1)r2 +rαr −λr =0 ⇒ (α −λ)r =0,
so we must have α=± λ. For λ=n , we have
Rn(r)=Cnrn +Dnr−n.
We observe that the D0logr term of R0(r), and Dnr−n term of Rn(r) are not defined at the origin, so they
can not be parts of the solution in the entire disk r <a. Thus, we must have Dn =0 for all n=0,1,....
Using the components Rn(r) and Θn(θ), we can write the series solution to Laplace’s equation as
∞
A0 X n
u(r,θ)= + r (Ancosnθ+Bnsinnθ), (10)
2 n=1
where we combined the constants Cn into the coefficients An and Bn. The coefficients in the above series will
be determined by the (inhomogeneous) Dirichlet boundary conditions of (9). Indeed,
∞
A0 X
u(a,θ)=h(θ)= + an(Ancosnθ+Bnsinnθ),
2 n=1
so we have ˆ 2π ˆ 2π
1 1
An = n h(φ)cosnφdφ, and Bn = n h(φ)sinnφdφ. (11)
πa 0 πa 0
Thus, the solution of the Dirichlet problem (9) is given by (10), where the coefficients are determined from (11).
Amazingly, if the coefficients (11) are substituted into the solution (10), the resulting series can be summed

9
explicitly. Indeed, we have
ˆ 2π ∞ ˆ
dφ X rn 2π
u(r,θ)= h(φ) + h(φ)[cosnφcosnθ+sinnφsinnθ]dφ
0 2π n=1 πan 0
ˆ 2π " ∞  
#
X r n dφ
= h(φ) 1+2 cosn(θ−φ) .
0 n=1
a 2π
Using Euler’s formula to express cosα=(eiα +e−iα)/2, we can rewrite the term in the square brackets above as
∞   ∞   ∞
X r n X r n in(θ−φ) X r n −in(θ−φ)
1+2 cosn(θ−φ)=1+ e + e
n=1
a n=1
a n=1
a
 r n  r n
Noticing that the above series are geometric series with ratios ei(θ−φ)
, and e−i(θ−φ) respectively, both
a a
of which have absolute values less than 1, we can sum them to arrive at
∞  
X r n rei(θ−φ) re−i(θ−φ)
1+2 cosn(θ−φ)=1+ +
n=1
a a−rei(θ−φ) a−re−i(θ−φ)
a2 −ar(ei(θ−φ) +e−i(θ−φ))+r2 +ar(ei(θ−φ) +e−i(θ−φ))−2r2
=
a2 −ar(ei(θ−φ) +e−i(θ−φ))+r2
a2 −r2
= 2 ,
a −2arcos(θ−φ)+r2
where we used Euler’s formula again in the lastˆstep. The solution then can be written as
2π
2 2 h(φ) dφ
u(r,θ)=(a −r ) 2 2
. (12)
0 a −2arcos(θ−φ)+r 2π
This is Poisson’s formula. It expresses any harmonic function in a disk in terms of its values on the boundary circle.
If we use the Cartesian variables x0 for the point (a,φ) on the circle, and x for any point (r,θ) inside the disk,
then the denominator of the integrand in (12), according to the law of cosines, is exactly the square of the length
of the vector x−x0, i.e.
a2 −2arcos(θ−φ)+r2 =|x−x0|2.
But then we can rewrite Poisson’s formula (12) in terms ˆ of Cartesian coordinates as follows.
a2 −|x|2 u(x0)
u(x)= ds0, (13)
2πa |x0|=a |x−x0|2
where the integration is with respect to the arc length element of the circle |x0|=a, i.e. ds0 =adφ.
One can prove that the function u given by (13) is harmonic, differentiable to all orders inside the disk |x|<a,
continuous on the closed disk |x|≤a, and that limx→x0 u(x)=h(x0) for every point x0 on the boundary of the
disk |x0|=a.
29.1 Mean value property
Using Poisson’s formula (13), one can show the following important property of harmonic functions.
Mean value property. Let u be a harmonic function in a disk D, and continuous in its closure D, then
the value of u at the center of the disk is equal to the
ˆ average of u on its circumference. That is,
1
u(x0)= u(x0)ds0, (14)
2πa |x0−x0|=a
where x0 is the center of the disk D, and a is its radius.
To prove (14), we can shift the coordinate system so that the origin coincides with the center of the disk,
then (14) will take the form ˆ
1
u(0)= u(x0)ds0.
2πa |x0|=a
But this follows immediately from Poisson’s formula (13) by substituting x=0, and noticing that |0−x0|=|x0|=a.

10
 Using the mean value property, one can prove the strong maximum principle for harmonic functions, which states
that a nonconstant harmonic function in an open connected bounded domain D which is also continuous on the clo-
sure D =D∪∂D, must attain its maximum on the boundary ∂D, and not inside D. To see this, let us assume that
the harmonic function u attains its maximum value M at an internal point x0 ∈D. So u(x)≤u(x0)=M at any
point x∈D. Then for any circle C which entirely lies in D, and is centered at the point x0, the mean value property
implies that the average of u on the circle is equal to M. But since the values on the circle can not be greater than M,
the only way their average can be equal to M is for all the values to be exactly M. Thus u will be constant on any of
such concentric circles, and hence on the maximal disk that lies in D and is centered at x0. Repeating this logic for
all the points of this maximal disk, and then the points of the resulting maximal disks, and so on, we see that u is con-
tinuous on the entire set D. Thus, a nonconstant harmonic function can not attain its maximum at an internal point.
But then the maximum of the continuous function u on the closed set D must be attained on the boundary ∂D.
29.2 Conclusion
Using separation of variables in polar coordinates we found a series solution for the Dirichlet problem on the
circle. Using the Dirichlet conditions, we found the coefficients in the series in terms of the Dirichlet data. In
this case we were able to explicitly sum the series, arriving at Poisson’s formula (12).
Using Poisson’s formula, we also proved the mean value property of harmonic functions, as a corollary of
which we obtained the strong maximum principle for harmonic functions.
Separation of variables in polar coordinates can be also applied to any other polar rectangle, such as a wedge,
annulus, or the exterior of a circle, to solve boundary value problems for Laplace’s equation as a series. We
will do this in detail in the next lecture.

11
30 Wedges, annuli, exterior of a circle

Similar to the circle considered last time, we can apply separation of variables in polar coordinates for any polar
rectangle. Such examples are:
A wedge: {0<r <a,0<β},
An annulus: {0<a<r <b},
Exterior of a circle: {a<r <∞}.
We next treat each case separately, applying separation of variables to arrive at a series solution.
30.1 The wedge
Let us consider the following boundary problem  in a wedge D ={0<r <a,0<β}.
 uxx +uyy =0 inD,
u(r,0)=u(r,β)=0, (15)
 u(a,θ)=h(θ).
Proceeding as in the circle case, we look for separated solutions in terms of the polar coordinates, u(r,θ)=R(r)Θ(θ).
Laplace’s equation in polar coordinates will reduce to the following ODE’s for R(r) and Θ(θ)
r2R00 +rR0 −λR=0, Θ00 =−λΘ.
Noticing that the homogeneous Dirichlet boundary conditions on the lateral sides of the wedge imply
Θ(0)=Θ(β)=0, we will have the eigenvalue problem  00
Θ =−λΘ,
Θ(0)=Θ(β).
The eigenvalues and eigenfunctions ofthis problem are
2
nπ nπθ
λn = , Θn(θ)=sin , n=1,2,....
β β
Using these values of λ, we can find the solution to the R√equation. As we saw last time, the solution to this
Euler type equation has the form R(r)=rα, where α=± λ. So for λ=(nπ/β)2, we have
Rn(r)=Anrnπ/β +Bnr−nπ/β .
The Bnr−nπβ term is not defined at the origin, which is a boundary point of the wedge. Thus, this term is
discarded, since we are looking for a harmonic function which is also continuous on the boundary (this can
be thought of as a boundary condition at r =0 that the solution is finite). Now using the separated solutions
Rn(r)Θn(θ), we can write the series solution as
∞ ∞
X X nπθ
u(r,θ)= Rn(r)Θn(θ)= Anrnπ/β sin . (16)
n=1 n=1
β
The coefficients An are determined by the boundary condition on the boundary r = a. Indeed, checking this
condition gives
∞
X nπθ
u(a,θ)=h(θ)= Ananπ/β sin ,
n=1
β
which implies ˆ
nπ/β 2 β nπθ hn
Ana =hn = h(θ)sin dθ ⇒ An = nπ/β .
β 0 β a
So the solution to problem (15) is
X ∞  r nπ/β nπθ
u(r,θ)= hn sin ,
n=1
a β
with hn being the Fourier sine coefficients of the Dirichlet data h(θ).
Similarly, one can solve boundary value problems in a wedge with Neumann or Robin boundary conditions
on the boundary r =a. The series solution (16) will still be the same, but the coefficients An will be determined
by the new condition.

12
30.2 The annulus
We consider the following Dirichletproblem in the annulus,
 uxx +uyy =0 in0<a2 <x2 +y2 <b2,
u=g(θ), forx2 +y2 =a2, (17)
 u=h(θ), forx2 +y2 =a2.
In this case the boundary condition of the eigenvalue problem for Θ is the same as in the circle case, Θ(θ+2π)=Θ(θ),
which leads to both cosnθ and sinnθ being eigenfunctions corresponding to the eigenvalue λn = n2. Also, all
the components logr, rn and r−n of Rn(r) are allowed, since they are defined everywhere in the annulus. Hence,
the series solution in polar coordinates can be written as
∞
1 X
u(r,θ)= (C0 +D0logr)+ (Cnrn +Dnr−n)cosnθ+(Anrn +Bnr−n)sinnθ.
2 n=1
The coefficients are determined by the boundary conditions of (17). Indeed,
∞
1 X
u(a,θ)=g(θ)= (C0 +D0loga)+ (Cnan +Dna−n)cosnθ+(Anan +Bna−n)sinnθ,
2 n=1
and similarly  for the boundary conditionon r =b. But then we must have
C0 +D0loga=Ag0 Cnan +Dna−n =Agn Anan +Bna−n =Bng

and
C0 +D0logb=Ah0 , Cnbn +Dnb−n =Ahn, Anbn +Bnb−n =Bnh,
where Agn,Bng are the Fourier coefficients of g(θ), and Ahn,Bnh are the Fourier coefficients of h(θ). It is not hard
to see that the determinants of the coefficients matrices of the above linear systems of equations are nonzero,
provided a6= b, guaranteeing unique solutions for the coefficients Cn,Dn for n=0,1,... and the coefficients An,Bn
for n=1,2,.... Again, other types of boundary conditions can be handled in a similar way.
30.3 Exterior of a circle
We finally turn to the Dirichlet problem  for the exterior of a circle.
 uxx +uyy =0 inx2 +y2 >a2,
u=h(θ), forx2 +y2 =a2, (18)
 ubounded asx2 +y2 →∞.
We again look for a series solution in terms of the separated solutions. The eigenvalue problem for Θ(θ) will
be exactly the same as in the case of the interior of a circle and the annulus. Notice, however, that in this case
the terms logr and rn of Rn(r) must be discarded, since both of them are unbounded as r → ∞, which can
be thought of as a boundary point for the exterior of a circle. Thus, the series solution of (18) will be
∞
A0 X −n
u(r,θ)= + r (Ancosnθ+Bnsinnθ). (19)
2 n=1
The boundary condition then gives
∞
A0 X −n
u(a,θ)=h(θ)= + a (Ancosnθ+Bnsinnθ),
2 n=1
from which we get ˆ ˆ
n h an 2π n h an 2π
An =a An = h(φ)cosnφdφ, Bn =a Bn = h(φ)sinnφdφ.
π 0 π 0
We can then substitute these expressions for An, Bn into the series solution (19). Comparing this to the circle
case, we can see that the only difference in the solutions is that r is replaced by r−a, and a by a−1. So we can
follow the same procedure, and sum the series explicitly, since in this case the magnitude of the (complex) ratios
of the geometric series will be a/r, which is less than one in the exterior of the circle, thus leading to summable

13
series. Then Poisson’s formula in the exteriorˆof the circle will be
2π
h(θ)
u(r,θ)=(a−2 −r−2) −2 −1 −1 −2
dφ
0 a −2a r cos(θ−φ)+r
ˆ 2π
2 2 h(θ)
=(r −a ) 2 2
dφ.
0 a −2arcos(θ−φ)+r

30.4 Conclusion
Similar to the Dirichlet problem on the circle, we separated variables in polar coordinates to solve boundary
value problems for Laplace’s equation in several examples of polar rectangles. In general, Laplace’s equation
in any polar rectangle {a<r <b,α<θ <β} can be solved by separating variables in polar coordinates, just as
Cartesian rectangles were handled by separation of variables in Cartesian coordinates.
We also found that in the case of the exterior of a circle the series solution of the Dirichlet problem can be
explicitly summed, similar to the case of the interior of a circle discussed last time. This gave a representation
formula for the solution, using which one can find the values of the harmonic function in the domain D from
its values on the boundary of the domain ∂D. In the coming lectures we will use Green’s functions to derive
similar representation formulas for more general domains.

14