Lecture notes on

Elliptic Equations
Prof. K. Sreenadh
IIT Delhi

1 Green’s Identities
In this section we recall the divergence theorem and derive some consequences which are
important for studying the boundary value problems with Dirichlet and Neumann bound-
ary conditions. Let us consider an open domain Ω ⊂ Rn , n ≥ 2 and a vector field F⃗ defined
on the closure of Ω and differentiable in Ω. Then the divergence theorem states
Z Z
(∇ · F⃗ ) dx = (F⃗ · n̂) dS
Ω ∂Ω

where ∂Ω stands for the boundary of Ω and n̂ is the unit outward normal vector to this
boundary, and the integral on the right hand side is the surface integral. Taking F⃗ = v∇u
and using the identity
∇ · (v∇u) = ∇u · ∇v + v∆u.

and the above divergence theorem, we get

Z Z Z
∂u
v dS = ∇u · ∇vdx + v∆udx (1.1)
∂Ω ∂n Ω Ω

∂w
Now taking ∇u = (0, ...w, 0, ..., 0) with w at it h place, we get ∆u = ∂xi
and the integra-
tion by parts formula:
Z Z Z
∂w ∂v
v dx = − w dx + vwni dS (1.2)
Ω ∂xi Ω ∂xi ∂Ω

where ni is the ith component of the unit outward normal. From (1.1), we can also obtain
the Green’s identity
Z Z
∂u ∂v
(v − u )dS = (v∆u − u∆v)dx. (1.3)
∂Ω ∂n ∂n Ω

If we choose v = 1 we get an important identity

Z Z
∂u
dS = ∆udx (1.4)
∂Ω ∂n Ω

1
If we take u = v we get
Z Z
∂u
u dS = (u∆u + |∇u|2 )dx. (1.5)
∂Ω ∂n Ω

Theorem 1.1. Uniqueness: Let u ∈ C 2 (Ω) ∩ C 1 (Ω) satisfies ∆u = 0 in Ω and either

∂u
u = 0 or ∂n = 0 on ∂Ω. Then from (1.5) we get that u is constant in Ω.

Also let us recall the equation of continuity from the flow through a uniform medium.
Suppose u(x, t) measures the temperature at the point x and at the point t, then we have
the equation
∂ ∂
(uA) + (qA) = f (x, t)
∂t ∂x
where q(x, t) is the flux. In case of heat conduction it observed that the flux follows the
Darcy’s law which says that flux is proportional to ∂u
∂x

∂u
q(x, t) = −β , β>0
∂x
Hence we obtain the heat conduction equation

∂u ∂ 2u
− β 2 = f (x, t)
∂t ∂ x
where f (x, t) represent the internal sources. There will be infinitely many functions sat-
isfying the above equation. But the temperature distribution is unique. So we look for
boundary and initial conditions from the physical model. The initial temperature distri-
bution is u(x, 0) which can be measured. For the boundary conditions:
Forced boundary conditions: If we assume that the temperature at the both end points
x = a and x = b is maintained at fixed. Say u(a, t) = T1 and u(b, t) = T2 .
Natural boundary conditions: Suppose we do not maintain any fixed temperature,
then the natural diffusive property of heat implies the flux at these ends are known. That
is ∂u
∂x
(a, t) = T0 (t) and ∂u
∂x
(a, t) = T1 (t).
Steady state temperature: This the state where the temperature does not change with
time. That is when ∂u ∂t
= 0. In this case the steady state temperature for forced boundary
conditions satisfies the following:

2
Dirichlet Problem: Given the functions f and g, find the function u that satisfies

−∆u(x) = f (x), in Ω
u(x) = g(x) on ∂Ω.

In case of natural boundary conditions, the steady state temperatures satisfy the problem
problem:
Neumann Problem: Given the functions f and g, find the function u that satisfies

−∆u(x) = f (x), in Ω
∂u
(x) = g(x) on ∂Ω.
∂n
∂u
where ∂n is the normal derivative of u on the boundary.
The following equivalent formulation of optimization problem. In case of Rn it is easy to
show such optimization problems has solution using the Lagrange-multiplier methods. In
case of PDEs the problems is posed on infinite dimensional space. These formulations are
often used for computing solutions using numerical methods known as Galarkin methods.

Dirichlet Principle:

Consider the Dirichlet problem

−∆u = f (x) in Ω u = 0 on ∂Ω (1.6)

and the functional Z Z

1 2
I(u) = |∇u| dx − f udx
2 Ω Ω

over A = {u ∈ C 2 (Ω) ∩ C(Ω), u = 0 on ∂Ω}. We consider the minimization problem

I(u) = min I(w) (1.7)

w∈A

Then we have the following

Theorem 1.2. The problem (1.6) has a solution in A if and only if u solves the problem
in (1.7).

3
Proof. =⇒ : If u solves the problem in (1.6). Then for any w ∈ A, then multiplying the
equation in (1.6) with u − w and integrating by parts, we get
Z Z
∇u · (∇u − ∇w) = f (u − w)dx
Ω Ω

from this we obtain

Z Z Z Z Z Z
2 1 2 2
|∇u| dx − f udx = ∇u · ∇w − f wdx ≤ (|∇u| + |∇w| )dx − f wdx
Ω Ω Ω Ω Ω 2 Ω

Therefore, I(u) ≤ I(w) for all w ∈ A.

To prove the converse, if u solves the minimization problem (1.7). Then t = 0 is a critical
point of the real valued function i : R → R given as

i(t) = I(u + tw)

Z Z
′
i (t) = (∇u + t∇w)∇wdx − f wdx
Ω Ω

taking t = 0 we get Z Z
0= ∇u · ∇w − f wdx, ∀w ∈ A.
Ω Ω

Therefore u satisfies (1.6).

Now consider the Neumann problem

∂u
−∆u = 0 in Ω = h(x) ∂Ω. (1.8)
∂n
R
Then it is necessary that ∂Ω
h(x)dS = 0 (see exercise 16). Consider the functional I over
C 2 (Ω) as Z Z
1 2
E(u) = |∇u| dx − hudS.
2 Ω ∂Ω

and the minimization problem

E(u) = min E(v). (1.9)

v∈C 2 (Ω)

Theorem 1.3. The problem (1.8) has a solution in C 2 (Ω) if and only if u solves the
problem in (1.9).

4
Proof. Let u be a solution and let w be any other function. By taking v = u − w, we can
show that
Z Z Z
1 2 ∂u
E(w) = E(u) + |∇v| dx − ∇u · ∇vdx + v dS
2 Ω ∂Ω ∂n
Z ZΩ
1
= E(u) + |∇v|2 dx + v∆udx
2 Ω Ω

≥ E(u).

The conclusion follows from this.

R
If we equip the space A with norm ( Ω |∇u|2 dx)1/2 , then the functional I(u) = ∥u∥2 −⟨f, u⟩
is bounded below. In case of Rn , if we take a functional

I(x) = ∥x∥2 − a · x, a ∈ Rn

Then it is easy to see that this is bounded below and any minimizing sequence (xn ) that
converges to inf I(x) is bounded. By compactness, such there exists a convergent subse-
quence xnk → x0 . Now it is easy to see that x0 is the required minimizer. But this is
not possible on A as it is infinite dimensional space. However taking a clue from above
discussion we can try to find solution un in any finite dimensional subspace and then pass
through the limit n → ∞. Such methods leads to the so called Galerkin methods.

Definition 1.1. A function u that satisfy the equation

∆u = 0 in Ω

is called harmonic function in Ω.

There are some special domains for which we can find harmonic functions by the separating
the variables. For example a rectangle [0, a] × [0, b]:

Example 1.1. Solve the Dirichlet problem

uxx + uyy = 0, u(0, y) = u(a, y) = u(x, 0) = 0, u(x, b) = f (x).

By assuming that the solution is in the form of X(x)Y (y), we get

X ′′ Y ′′
=− = k (say)
X Y

5
where is k is a constant. This leads to solving two ODEs

X ′′ = kX, Y ′′ = −kY

Also the boundary conditions

u(0, y) = u(a, y) = 0 =⇒ X(0) = 0, X(a) = 0.

Therefore solving the SLP:

X ′′ = kX, X(0) = X(a) = 0, (1.10)

we get the solution

nπ n2 π 2
X(x) = sin x, k = − 2 , n = 1, 2, 3...
a a
The second equation now becomes

n2 π 2
′′
Y = 2 Y, Y (0) = 0
a
The general solution of this is
nπ
Y (y) = A sinh y
a
where A is an arbitrary constant. Since these are solutions for each n ∈ N , we take
∞
X nπ nπ
u(x, y) = An sin x sinh y
n=1
a a

Now the unknown constants An can be found by substituting the final boundary condition
u(x, b) = f (x),
∞
X nπ nπ
f (x) = An sin x sinh b
n=1
a a
Now using the orthogonality of the functions sin nx, n = 1, 2, 3.... So An can be obtained
from the relation:
nπb a 2 nπx
Z Z a
nπx
An sinh sin dx = f (x) sin dx.
a 0 a 0 a

6
In case of Neumann problem, the boundary condition in SLP in (1.10), becomes

X ′ (0) = 0, X ′ (a) = 0.

Then the corresponding eigenvalues and eigen functions be taken in the infinite sum.
In the domains like disc Br (0), again one can write u(x, y) = R(r)Θ(θ) where r, θ are poloar
coordinates to solve the problems. These methods have limitation that the domain has to
be a rectanlge or disc. In general it is not always possible to find the solutions exactly. To
understand this we need to study qualitative properties of solutions.

2 Maximum Principles
If we want to solve the Dirichlet problem in a general domain, we need first understant
some qualitative properties of harmonic functions. In this direction we have the following
maximum principle:

Theorem 2.1. Weak form: Let Ω be a bounded domain in Rn and let u ∈ C 2 (Ω) ∩ C(Ω)
satisfies ∆u ≥ 0 in Ω. Then
max u(x) = max u(x)
x∈Ω x∈∂Ω

Proof. Case 1: ∆u > 0 in Ω.

∂2u
If x0 ∈ Ω is a point of interior maximum of u. Then by second derivative test ∂x2i
(x0 ) ≤ 0.
Therefore, ∆u(x0 ) ≤ 0. Contradiction to ∆u > 0 in Ω.
Case 2: ∆u ≥ 0 in Ω.
Now consider the function v(x) = u(x) + ϵ|x|2 . Then

∆v(x) = ∆u(x) + 2nϵ > 0.

Then by case 1,
max v(x) = max v(x)
Ω ∂Ω

max u + ϵ min |x|2 ≤ max(u(x) + ϵ|x|2 ) = max v(x) = max v(x) ≤ max u(x) + ϵ max |x|2
Ω Ω Ω Ω ∂Ω ∂Ω ∂Ω

Taking ϵ → 0, conclusion follows.

Corollary 2.1. If u ∈ C 2 (Ω) ∩ C(Ω) satisfies −∆u ≥ 0 in Ω, u ≥ 0 on ∂Ω. Then u ≥ 0

in Ω.

7
Remark 2.1. 1. In case of Harmonic functions, i.e., ∆u = 0, above theorem holds for
−u as well. Therefore using minu(x) = − max(−u(x)), we obtain

min u = min u.
Ω ∂Ω

2. If u ∈ C 2 (Ω) ∩ C(Ω) satisfies ∆u = 0 in Ω, then

max |u| = max |u|.

Ω ∂Ω

(follows from |a| = max(a, −a))

3. If u ∈ C 2 (Ω) ∩ C(Ω) satisfies ∆u = 0 in Ω and u = 0 on ∂Ω. Then u ≡ 0 in Ω.

Theorem 2.2. Uniqueness: The Dirichlet problem −∆u = f in Ω and u = g on ∂Ω

has at most one solution.

Proof. Let u1 and u2 be two solutions, then by considering u = u1 −u2 we see that −∆u = 0
in Ω and u = 0 on ∂Ω. Therefore by the the (3) of Remark 1.1, we get u = u1 − u2 ≡ 0 in
Ω.

Mean values: For a continuous function h(x) on Rn , let us introduce its spherical mean
or average on a sphere of radius r and center x:
Z
1
Mh (x, r) = h(x + rξ) dSξ ,
wn |ξ|=1

where wn denotes the area of the unit sphere S n−1 = {ξ ∈ Rn : |ξ| = 1} and dSξ denotes
surface measure.

Definition 2.1. Mean Value property: A function u ∈ C(Ω) is said to satisfy Mean
Value property if u(ξ) = Mu (ξ, r) for all r > 0 such that Br (ξ) ⊂⊂ Ω.

Theorem 2.3. Mean Value theorem (Gauss): If u ∈ C 2 (Ω) satisfies ∆u = 0 in Ω

then u satisfies mean value property.

Proof. From (1.4), we get for Br (ξ) ⊂ Ω,

Z Z Z
∂u ∂u
0= ∆udx = dS = (x)dSx
Br (ξ) ∂Br (ξ) ∂n |x−ξ|=r ∂n

8
Taking x = ξ + ry with |y| = 1, dSx = rn−1 dSy we have from above equation
Z Z
n−1 ∂u ∂ ∂
0=r (ξ + ry)dSy = rn−1 u(ξ + ry)dSy = rn−1 wn Mu (ξ, r)
|y|=1 ∂r ∂r |y|=1 ∂r

That is Mu (ξ, r) is independent of r. Therefore,

Mu (ξ, r) = lim Mu (ξ, r) = u(ξ),

r→0

thanks to Lebesgue’s theorem.

Theorem 2.4. If u ∈ C 2 (Ω) satisfies ∆u = 0 in Ω. Then

Z
n
u(ξ) = u(ξ + rx)dx.
wn |x|≤1

Proof. From the above theorem, we have

Z
1
u(ξ) = u(ξ + rx)dSx
wn |x|=1
Z
1
= u(ξ + rρx)dSx
wn |x|=1

since the formula is true for any r. Multiplying both sides by ρn−1 and integrating from 0
to 1, we get
Z 1 Z 1 Z
u(ξ) n−1 1 n−1
= ρ u(ξ)dρ = ρ u(ξ + rρx)dSx dρ
n 0 wn 0 |x|=1
Z
1
= u(ξ + rx)dx.
wn |x|≤1

Corollary 2.2. If u ∈ C 2 (Ω) satisfies ∆u ≥ 0 in Ω. Then u(ξ) ≤ Mu (ξ, r).

Proof. This follows from the previous theorem by replacing = by ≤

Z Z
∂u ∂
0≤ ∆udx = dS = ... = Mu
Br ∂Br ∂n ∂r

9
That is Mu (x, r) is increasing in r. Therefore

Mu (x, r) ≥ Mu (x, 0) = u(ξ).

Corollary 2.3. If u ∈ C 2 (Ω) and ∆u ≥ 0 in Ω. Then

Z
n
u(ξ) ≤ u(ξ + rx)dx.
wn |x|≤1

Next we have the converse of Theorem 2.3:

Theorem 2.5. If u ∈ C 2 (Ω) satisfies meanvalue property then u is harmonic in Ω.

Proof. Assume by contradiction that ∆u(ξ) > 0 for some point ξ. Then there exists a ball
R
Bϵ around ξ such that ∆u > 0 in Bϵ (ξ) and hence Bϵ (ξ) ∆udx > 0. But on the other hand,
From the Theorem 1.1, we have
Z
∂
∆udx = ϵn−1 wn Mu (ξ, r)
Bϵ (ξ) ∂r

Now since u satisfies mean value property, then Mu (ξ, r) = u(ξ). Therefore from the above
equation we get Z
∆udx = 0.
Bϵ (ξ)

This is a contraction to ∆u > 0 in Bϵ (ξ).

Next we have the following

Theorem 2.6. Strong form of Maximum principle

Let Ω be a connected domain and let u ∈ C 2 (Ω) satisfies ∆u ≥ 0 in Ω, then either u is
constant or u(ξ) < sup u(x) for all ξ ∈ Ω.
x∈Ω

Proof. Let A = supx∈Ω u(x) < ∞. By continuity of u, we know that the set

M = {x ∈ Ω : u(x) = A}

is relatively closed in Ω. Now we claim that this is also open in Ω.

10
If u(ξ) = A. Then by above corollary, taking ξ + rx = y
Z Z
1 1
u(ξ) ≤ u(ξ + rx)dSx = u(y)dSy .
wn |x|=1 rn−1 wn |ξ−y|=r

Therefore, Z Z
n−1
0≤ u(y)dSy − wn r A= (u(y) − A)dSy .
|ξ−y|=r |ξ−y|=r

But u(y) − A ≤ 0 and u is continuous. Therefore u(y) = A for all small ρ and y ∈ ∂Bρ (ξ).
That is y ∈ M for all y in a neighbourhood of ξ. Hence M is relatively open in Ω. Since
Ω is connected, we get a contradiction. Therefore ξ ̸∈ Ω o .

Corollary 2.4. If Ω is connected domain and u ∈ C 2 (Ω) ∩ C(Ω) satisfy ∆u ≥ 0 in Ω.

Then either u is constant OR u(ξ) < max∂Ω u for all ξ ∈ Ω.

Corollary 2.5. Stability: If u, v ∈ C 2 (Ω)∩C(Ω) satisfy ∆u = f in Ω and |u(x)−v(x)| ≤

ϵ for all x ∈ ∂Ω. Then
|u(x) − v(x)| ≤ ϵ for all x ∈ Ω.

Moreover, one can show that if u ∈ C(Ω) is enough. In fact we have the following

Theorem 2.7. Suppose that u ∈ C(Ω) has the mean value property. Then u ∈ C ∞ (Ω)
and is harmonic.

Proof. Let η ∈ Cc∞ (Ω) be radially symmetric function with support in Bϵ (x) ⊂⊂ Ω and
R
η = 1. Then we claim that (η ∗ u)(x) = u(x) and Dα (η ∗ u) = Dα η ∗ u.

Z Z
(η ∗ u)(x) = η(x − y)u(y)dy = η(z)u(x − z)dz
|x−y|<ϵ |z|<ϵ
Z ϵ Z 
= η(r|z|)u(x − rz)dS(z) rn−1 dr
0 ∂B (0)
Z ϵ Z 1 
= u(x − rz)dS(z) rn−1 η(r)dr
0 ∂B1 (0)
Z ϵ
= wn u(x) η(r)rn−1 dr
Z 0
= u(x) η(y)dy = u(x).
Bϵ (x)

11
Next we will show that Dα (η ∗ u) = Dα ∗ u:

[η(x − y − hei ) − η(x − y)] u(y)

Z
∂
(η ∗ u)(x) = lim dy
∂xi h→0 Rn h

Now since η is C ∞ function with compact support, we have

η(x − y − hei ) − η(x − y) ∂η

→ |(x − y) uniformly in y.
h ∂xi

Hence, Z
∂ ∂η ∂η
(η ∗ u)(x) = (x − y)u(y)dy = ( ∗ u)(x).
∂xi Rn ∂xi ∂xi
This concludes the proof.

Remark 2.2. The above theorems imply that if u satisfies mean value property then it
is harmonic and u ∈ C ∞ (Ω). If u is harmonic then it satisfies mean value property and
hence is C ∞ (Ω).

More generally, the following theorem is due to Weyl.

Theorem 2.8. (Weyl):

Let u : Ω → R be measurable and locally integrable in Ω. If u satisfies ∆u = 0 in D′ , in
the sense of distributions,. Then u is harmonic and C ∞ (Ω).

Next we have the following Hopf maximum principle.

Definition 2.2. Interior ball condition The boudary ∂Ω satisfies interior ball condition
at x0 if there is a ball Bϵ (x1 ) ⊂ Ω such that ∂Ω ∩ Bϵ (x1 ) = {x0 }

Theorem 2.9. (Hopf, Oleinik): Let u ∈ C 2 (Ω) ∩ C 1 (Ω), ∆u ≥ 0 in Ω and let Ω be a

domain with its boundary ∂Ω satisfying interior ball condition at x0 where x0 is the point
of maximum of u such that u(x0 ) > u(x) for all x ∈ Ω. Then u is either constant OR
∂u
∂n
(x0 ) > 0.

Proof. Let us assume that there is a ball Br ⊂ Ω such that Br ∩ ∂Ω = {x0 }. Without loss
2 2
of generality, let Br = Br (0). Consider the function v = e−λ|x| − e−λr , x ∈ Br (0). Then
we can check that
2
vxi = e−λ|x| (−2λxi )

12
Let R = Br (0)\Br/2 (0) and let λ be fixed as large so that

2
X
∆v = e−λ|x| (4λ2 x2i − 2λ)
2
= e−λ|x| (4λ2 |x|2 − 2nλ) > 0 in R.

Now ∆(u + ϵv − u(x0 )) ≥ 0 in R and u + ϵv − u(x0 ) ≤ 0 on ∂R for ϵ small. By weak

maximum principle,
u + ϵv − u(x0 ) ≤ 0 in R

and since v(x0 ) = 0

u(x0 ) + ϵv(x0 ) − u(x0 ) = 0.
∂
Therefore ∂n
(u + ϵv) x0 ≥ 0. Hence

∂u 0 ∂v ∂v xi 0 2
(x ) ≥ −ϵ (x0 ) = − = 2λ|x0 |2 e−λ|x | > 0.
∂n ∂n ∂xi r

3 Fundamental solution
A fundamental solution K(x) for the Laplace operator is a ”distribution” satisfying the
relation
∆K(x) = δ(x) in D′ (Rn ),

where δ is the delta distribution supported at x = 0. That is, K(x) is a ”locally integrable”
function that satisfies,
Z
K(x)∆ϕ(x)dx = ϕ(0) for all ϕ ∈ Cc∞ (Rn ).
Rn

To obtain such solutions, we first assume that u(x) = v(r), r = |x|. By chain rule,

xi n−1 ′
uxi = v ′ (r) , ∆u = v ′′ (r) + v (r)
r r
Therefore solving for ∆u = 0 we get

b log r + c n = 2,
v(r) =
b
 n−2 + c, n ≥ 3,
r

13
where b, c are constants of integration.

Definition 3.1. For x ̸≡ 0, the function


 1 log |x| n=2
Φ(x) = 2π
 1 |x|2−n n ≥ 3.
(2−n)wn

where wn is the surface area of the unit ball in Rn is called the fundamental solution of
Laplacian operator.

Poisson’s equation: The equation

−∆u = f (x) in Rn .

is known as Poisson equation. Here the problem is to find u for a given f . From the theory
of distributions, we know that (Φ ∗ f )(x) is a distributional solution of Poisson equation if
f is a regular distribution with compact support. Now we will show that this is actually
classical solution if we assume more regularity on the function f (x).

From the construction above we see that x 7→ Φ(x) is harmonic for all x ̸= 0. If we shift
the orgin to another point y the PDE remains the same. Also the function x 7→ Φ(x − y)
is also harmonic as a function of x, x ̸= y. Now for a function f (x) the mapping x 7→
Φ(x − y)f (y), (x ̸= y) is harmonic for each point y ∈ Rn .

We will show that the convolution

Z
(Φ ∗ f )(x) = Φ(x − y)f (y)dy
Rn

solves the problem. The important observation here is ”differentiation under integral sign”
R
is not allowed. Otherwise ∆(Φ ∗ f ) = ∆x Φ(x − y)f (y) = 0. The reason is the second
derivative of Φ is not integrable near 0.

Theorem 3.1. Assume that f ∈ Cc2 (Rn ) and let u(x) = (Φ ∗ f )(x). Then u ∈ C 2 (Rn ) and
−∆u(x) = f (x) in Rn .
R
Proof. First note that (Φ ∗ f )(x) = (f ∗ Φ)(x) = y∈supp(f )
Φ(x − y)f (y)dy and so
 
u(x + hei ) − u(x) f (x − y + hei ) − f (x − y)
Z
= Φ(y) dy
h Rn h

14
 f (x−y+hei )−f (x−y) ∂f
Also since f ∈ Cc2 (Rn ), h
→ ∂xi
(x − y) uniformly. Therefore,

∂ 2u ∂ 2f
Z Z
∂u ∂f
= Φ(y) (x − y)dy, and = Φ(y) (x − y)dy.
∂xi Rn ∂xi ∂xi ∂xj Rn ∂xi ∂xj

Therefore,
Z Z
∆u(x) = Φ(y)∆x f (x − y)dy = Φ(y)∆y f (x − y)dy
n Rn
ZR Z
= + (Φ(y)∆y f (x − y)dy)
Bϵ (0) Rn \Bϵ (0)

= Iϵ + Jϵ

We can estimate Iϵ and Jϵ as follows

Z
|Iϵ | ≤ ∥∆y f ∥L∞ Φ(y)dy
Bϵ (0)
 Z ϵZ
C∥∆f ∥L∞ |y|2−n rn−1 rdrdθ n≥3


n−1
≤ Z0 ϵ ZS2π
C∥∆f ∥L∞ (log r)rdrdθ n=2


0 0

o(ϵ2 ) n≥3
=
o(ϵ2 log ϵ) n = 2

→ 0 as ϵ → 0.

Integration by parts on Jϵ , we get

n Z
X ∂ 2f
Jϵ = Φ(y) (x − y)dy
n
i=1 R \Bϵ (0)
∂yi2
n Z n Z
X ∂f X ∂Φ ∂f
= Φ(y) (x − y)ni dSy − (x − y)dy
i=1 ∂Bϵ (0) ∂yi i=1 Rn \Bϵ (0) ∂yi ∂yi
= Lϵ + Kϵ

15
As above, we can estimate Lϵ as follows
Z
|Lϵ | ≤ ∥∇f ∥L∞ |Φ(y)|dSy
∂Bϵ (0)
 Z
C

 ϵ2−n dθ = Cϵ n≥3
∂B (0)
≤ Z ϵ

C

 log |ϵ|dθ = C(log |ϵ|)(2πϵ) n = 2
∂Bϵ (0)

→ 0 as ϵ → 0.

To estimate Kϵ , we again use integration by parts and using the fact that ∆Φ = 0 away
from 0, we get
n Z
X ∂Φ ∂f
Kϵ = − (x − y)dy
n
i=1 R \Bϵ (0)
∂yi ∂yi
n Z Z
X ∂Φ
=− f (x − y)ni dSy − ∆Φ(y)f (x − y)dy
i=1 ∂Bϵ (0) ∂yi Rn \Bϵ (0)
n Z
X ∂Φ
=− f (x − y)ni dSy
i=1 ∂Bϵ (0) ∂yi

y
and using the definition of Φ and n = − |y| on ∂Bϵ ,

∂Φ ∂Φ (−yi ) 1 yi yi −1 1−n ϵ1−n

ni = = (2 − n)|y|1−n = |y| =−
∂yi ∂yi r (n − 2)wn |y| r wn wn

Therefore,
−1
Z
Kϵ = f (x − y)dy → −f (x), as ϵ → 0.
wn ϵn−1 ∂Bϵ (0)

Therefore ∆u = lim (o(ϵ log ϵ) + Kϵ ) = −f (x).

ϵ→0

From the above theorem we infer that

Z
u(x) = Φ(x − y)∆u(y)dy whenever u ∈ C0∞ (Rn ).
Rn

The following result generalizes this formula allowing boundary terms.

16
Theorem 3.2. If Ω is a smooth bounded domain in Rn , u ∈ C 2 (Ω), and x ∈ Ω, then
 
∂Φ(x − y)
Z Z
∂u(y)
u(x) = Φ(x − y)∆u(y)dy + u(y) − Φ(x − y) dSy . (3.11)
Ω ∂Ω ∂ny ∂n

Moregenerally, (3.11) holds if u ∈ C 2 (Ω) ∩ C 1 (Ω) and the integral over Ω converges.

Proof. Recall the Green identity (1.3):

Z Z
∂u ∂v
(v − u )dS = (v∆u − u∆v)dx.
∂Ω ∂n ∂n Ω

For u ∈ C 2 (Ω) and x ∈ Ω , 0 < ϵ < dist(x, ∂Ω), let Ωϵ = Ω\Bϵ (x). Applying the above
identity on Ωϵ we get
Z 
Z  Z
∂u ∂Φ
Φ(x − y)∆u(y)dy = Φ(x − y) − u(y) (x − y) dSy + u∆Φ(x − y)dy
Ωϵ ∂Ωe ∂n ∂n Ωϵ
(3.12)
1
Now note that ∆Φ = 0 in Ωϵ and using the fact that Φ(x − y) ∈ L (Ω), we get
Z Z
lim Φ(x − y)∆u(y)dy = Φ(x − y)∆u(y)dy
ϵ→0 Ωϵ Ω

The boundary of Ωϵ consists of ∂Ω and ∂Bϵ (0):

Z Z Z
= +
∂Ωϵ ∂Ω ∂Bϵ (x)

 R
∥u∥ 2−n
C1 |x−y|=ϵ ϵ dS n≥3
Z
∂u
Φ(x − y) dS ≤ −→ 0 as ϵ → 0.
|x−y|=ϵ ∂n ∥u∥C 1 2πϵ(| log ϵ|) n=2

17
  n
xi − y i
Z
1 X ∂
u(y) (|x − y|2−n ) · − n ≥ 3,


(n − 2)wn |x−y|=ϵ ∂yi ϵ
Z 
∂Φ 
i=1
u(y) (x − y)dSy = n
xi − yi
Z
|x−y|=ϵ ∂ν  1 X ∂

 u(y) (log |x − y|) · − , n = 2,
2π |x−y|=ϵ ∂y ϵ

i=1 i
n

(xi − yi )2
Z
 1 1−n
X

 u(y)|x − y| n≥3
wn |x−y|=ϵ ϵ

= i=1
1 X (xi − yi )2
Z
 1

 u(y) n = 2.
2π |x−y|=ϵ |x − y| ϵ


−→ u(x) as ϵ → 0.

3.1 Dirichlet Problem

From (3.11), if u is a harmonic function, we get

Theorem 3.3. If Ω is a smooth bounded domain in Rn , u ∈ C 2 (Ω) is a harmonic function,

and x ∈ Ω, then
 
∂Φ(x − y)
Z
∂u(y)
u(x) = u(y) − Φ(x − y) dSy . (3.13)
∂Ω ∂ny ∂n

Dirichlet Problem: Given f and g find u satisfying:

−∆u(x) = f (x) in Ω, u(x) = g(x) on ∂Ω. (3.14)

So we need to modify the fundamental solution Φ. We do this by adding ”corrector” to Φ

so that one of the term on the Right hand side of (3.13) is zero. To achieve this we take
a harmonic function w(x) and consider the function G(x, y) = Φ(x − y) + wx (y). Then
taking v(y) = wx (y) in the Green’s identity (1.3), we get
Z  Z 
∂w ∂u
w∆udy + u −w dS = 0.
Ω ∂Ω ∂n ∂n

Also from (3.11), we have

Z Z  
∂Φ ∂u
Φ(x − y)∆u(y)dy + u(y) (x − y) − Φ(x − y) dSy = u(x)
Ω ∂Ω ∂ny ∂ny

18
Adding these two equations, we get
Z  Z 
∂G ∂u
u(x) = G(x, y)∆u(y) + u(y) − G(x, y) dSy
Ω ∂Ω ∂ny ∂ny

where G(x, y) = Φ(x − y) + wx (y) Now if we can choose wx (y) satisfying G(x, y) =
0 for all y ∈ ∂Ω. That is,

∆wx (y) = 0, in Ω, wx (y) = −Φ(x − y) for all x ∈ Ω, y ∈ ∂Ω.

Then Z Z
∂G
u(x) = G(x, y)∆u(y)dy + u(y) (x, y)dSy . (3.15)
Ω ∂Ω ∂n
Therefore we have

Theorem 3.4. If u(x) solves the problem (3.14). Then

Z Z
∂G
u(x) = G(x, y)f (y)dy + g(y) (x, y)dSy . (3.16)
Ω ∂Ω ∂n

Proof. follows from (3.15).

To prove the converse of the above theorem, we investigate the regularity of the integral in
Z
u(x) = Φ(x − y)f (y)dy.
Ω

As mentioned in the beginning of the section, if f is locally integrable, by extending

the function f (x) to be zero outside Ω, we get a compactly supported function and is
a distributional solution of the equation −∆u = f (x) in D′ (Ω). The following theorem
expresses ways in which additional regularity of f improves the regularity of u.

Theorem 3.5. For a bounded domain Ω and f ∈ L1 (Ω) define u(x) by

Z
u(x) = Φ(x − y)f (y)dy.
Ω

Then

1. u is harmonic and C ∞ in Rn \Ω.

2. if f is bounded on Ω, then u ∈ C 1 (Rn ).

19
 3. if f ∈ C 1 (Ω), then u ∈ C 2 (Ω).

Proof. (1): For any x ∈ Rn \Ω, Φ(x − y) is a smooth function as x − y ̸= 0 for any y ∈ Ω.
So we can differentite under integral sign and since Φ is harmonic we get ∆u(x) = 0.

(2): for j = 1, 2, ...n we define

∂Φ(x − y)
Z
uj (x) = f (y)dy,
Ω ∂xj

which is well defined because ∂Φ/∂xj is O(|x − y|1−n ) as |x − y| → 0. We first approximate

u by smooth functions as follows: Let η ∈ C ∞ (Rn ) satisfy η(t) = 0 for t < 1, η(t) = 1 for
t > 2, and 0 ≤ η(t) ≤ 1 and 0 ≤ η ′ (t) ≤ 2 for all t, we define
 
|z|
Z
uϵ (x) = Φϵ (x − y)f (y)dy, Φϵ (x) = Φ(z)η .
Ω ϵ

Then uϵ is smooth function as Φϵ = 0 for |x − y| < ϵ. It is not difficult to check that uϵ → u

uniformly on compact subsets of Rn . Indeed, if x ∈ K,
Z    
|x − y|
|uϵ (x) − u(x)| = Φ(x − y)η − Φ(x − y) f (y)dy
Ω ϵ
  
|x − y|
Z
≤ ∥f ∥∞ Φ(x − y) 1 − η dy
|x−y|<2ϵ ϵ
→ 0 as ϵ → 0.

Here we used the fact that η( |x−y|

ϵ
) = 1 for |x − y| > 2ϵ. Next we will show that ∂uϵ
∂xj
→ uj .
For this,
  
|x − y|
Z
∂uϵ ∂
uj (x) − (x) = 1 − η( ) Φ(x − y) f (y)dy (3.17)
∂xj |x−y|<2ϵ ∂xj ϵ
 ′   
|z| ∂Φ(z)
Z
η (|z|/ϵ) zj
= Φ(z) − 1 − η( ) f (x − z)dz (3.18)
|z|<2ϵ ϵ |z| ϵ ∂zj

Now using the facts:

1. |η ′ (t)| ≤ 1
Z
1
2. Φ(z)dz = o(ϵ)
ϵ |z|<2ϵ

20
 Z
∂Φ(z)
3. | | ≤ 2ϵ,
|z|<2ϵ ∂zj
∂uϵ ∂u
we see that ∂xj
converges uniformly to ∂x j
. Therefore u ∈ C 1 (Rn ).
∂u
(3): We let again v = ∂xj
, then introduce

∂2
Z
vk (x) = Φ(x − y)f (y)dy
Ω ∂xk ∂xj
∂2 ∂2
Z Z
= Φ(x − y)(f (y) − f (x))dy + f (x) Φ(x − y)dy (3.19)
Ω ∂xk ∂xj Ω ∂xk xj

By divergence theorem

∂2
Z Z
∂
Φ(x − y)dy = − Φ(x − y)nk dSy .
Ω ∂xk xj ∂Ω ∂xj

Again we use the smoothing trick

|x − y|
Z
∂
vϵ (x) = Φ(x − y)η( )f (y)dy
Ω ∂xj ϵ

and calculate,
   
|x − y| |x − y|
Z Z
∂vϵ ∂ ∂ ∂ ∂
= Φ(x − y)η( ) (f (y)−f (x))dy+f (x) Φ(x − y)η( ) dy
∂xk Ω ∂xk ∂xj ϵ Ω ∂xk ∂xj ϵ

Again using divergence theorem on the second term, we obtain

   
|x − y| |x − y|
Z Z
∂ ∂ ∂
Φ(x − y)η( ) dy = − Φ(x − y)η nk dSy
Ω ∂xk ∂xj ϵ ∂Ω ∂xj ϵ
Z
∂
=− Φ(x − y)nk dSy . (3.20)
∂Ω ∂xj

provided ϵ > 0 small enough such that 2ϵ < dist(x, ∂Ω), and hence η(|x − y|/ϵ) = 1 for all
y ∈ ∂Ω. Thus we have from (3.19) and (3.20)
   
|x − y|
Z
∂ ∂ ∂
vk (x) − vϵ (x) = Φ(x − y) 1 − η (f (y) − f (x))dy,
∂xk |x−y|<2ϵ ∂xk ∂xj ϵ

since the boundary integrals cancel each other. Estimating as earlier and noting that

21
 ∂
∂zi
Φ(z) = o(|z|1−n ), we get

∂2
Z
∂
|vk (x) − vϵ (x)| ≤ o(ϵ) + Φ(x − y) |f (x) − f (y)|dy
∂xk |x−y|≤2ϵ ∂xk ∂xj
|f (x) − f (y)|
Z
≤ C sup |x − y|1−n dy = o(ϵ)
y∈Ω |x − y| |x−y|<2ϵ

where C(ϵ) → 0 as ϵ → 0. But f ∈ C 1 (Ω) ensures that the supremum is finite, so we

conclude that v ∈ C 1 (Ω) and hence u ∈ C 2 (Ω).

Remark 3.1. A close observation at the proof suggests that f (x) ∈ C 0,α (Ω) is enough to
get the C 2 regularity.

The domain Green’s function is of the form Φ(x − y) + wx (y) where w is

Z a harmonic func-
∂G
tion. So now we need to understand the regularity boundary intergral g(y)dSy .
∂Ω ∂n

Poisson Integral formula: We want to solve the problem of finding harmonic functions
with prescribed boundary values. Let Ω be a domain in Rn with boundary ∂Ω: Given g,
find u satisfyng
(P ) : ∆u = 0 in Ω, u(y) = g(y) on ∂Ω.

Theorem 3.6. Consider the domain Ω = Rn+ = {x ∈ Rn , xn > 0} and let g is continuous
and bounded. Then the solution of problem

(P ) : ∆u = 0, in Rn+ , u(y) = g(y) on ∂Rn+

is given by the Poission Integral formula:

Z
u(x) = H(x, y)g(y)dy, (3.21)
∂Rn
+

2xn
where H(x, y) = n
, x ∈ Rn , y ∈ ∂Rn+ is called Poisson Kernel.
wn |y − x|

Proof. The proof is by constructing Green’s function for the domain Rn+ . We need to define
the harmonic function wx (y) satisfying

∆wx (y) = 0 in Ω, wx (y) = −Φ(x − y), y ∈ ∂Rn .

22
We use the notation

Rn+ = {(x′ , xn ) = (x1 , x2 , ..., xn ), xn > 0}.

For x = (x′ , xn ) ∈ Rn+ , Define its reflextion x∗ = (x1 , x2 , ..., −xn ) ̸∈ Rn+ . Then the function
Φ(y − x∗ ) is harmonic for all y ∈ Rn+ . Moreover, for y = (y1 , ..., yn−1 , 0) ∈ ∂Rn+ , we have

|y − x| = |y − x∗ |

Therefore, Φ(y − x) = Φ(y − x∗ ). Hence taking wx (y) = −Φ(y − x)

G(x, y) = Φ(y − x) − Φ(y − x∗ )

is the Green’s function for Rn+ . Now to solve the problem (P ) we use the formula (3.16),
we notice that for y ∈ ∂Rn+ , x ∈ Rn+

∂Φ y n − xn
(y − x) = |y − x|−n ,
∂yn wn

yn − x∗n
   
∂G(y, x) ∂Φ ∂Φ ∗ yn − xn −n ∗ −n
=− (y − x) − (y − x ) = − |y − x| − |y − x |
∂n ∂yn ∂yn wn wn

For y ∈ ∂Rn+ , we have |y − x| = |y − x∗ | and x∗n = −xn . Therefore,

∂G(y, x) |y − x|−n
=− (yn − xn − yn − xn )
∂n wn
2xn
= |y − x|−n := H(x, y)
wn

Hence by (3.21) we get Z

2xn g(y)
u(x) = dy
wn ∂Rn |x − y|n
for y ∈ ∂Rn+ . Now since x 7→ G(y, x) is harmonic for x ̸= y, we have x 7→ H(y, x) =
∂G
− ∂y n
(y, x) is harmonic because x ∈ Rn+ and y ∈ ∂Rn+ . Hence defining u(x) as (3.21) and
applying Laplacian on u(x), for x ∈ Rn+ ,
Z
∆u(x) = ∆x H(y, x)g(y)dSy = 0.
∂Rn
+

23
Claim: lim0 u(x) = g(x0 ) for x0 ∈ ∂Rn+ .
x→x
to prove the claim, choose x0 ∈ ∂Rn+ , and let ϵ > 0. Choose δ > 0 such that |g(y)−g(x0 )| <
ϵ whenever |y − x0 | < δ, y ∈ ∂Rn+ . Then if x ∈ Rn+ with |x − x0 | < 2δ , we have
Z
0
|u(x) − g(x )| = H(y, x)(g(y) − g(x0 ))dSy
∂Rn
+
Z Z
≤ + := J + I
∂Rn 0
+ \Bδ (x ) ∂Rn 0
+ ∩Bδ (x )

Now since H is integrable, we estimate I as

Z
|I| ≤ ϵ H(y, x)dy → 0 as ϵ → 0.
∂Rn
+

δ
For J, as x → x0 , we have |x − x0 | < 2
and |y − x0 | ≥ δ. Therefore,

δ 1
|y − x0 | ≤ |y − x| + |x − x0 | ≤ |y − x| + ≤ |y − x| + |y − x0 |.
2 2
δ
from this we see that 2
≤ 12 |y − x0 | ≤ |y − x|. Therefore, we can estimate J as
Z
|J| ≤ 2∥g∥L∞ H(y, x)dy
∂Rn
+
Z Z ∞
0 −n
≤ 2xn ∥g∥L∞ 2 n
|y − x | dy = 2xn ∥g∥L∞ 2 n
r−n rn−2 dr → 0 as xn → 0.
∂R+
n δ

Note that in the above proof we have taken y ∈ ∂Rn+ .

Green’s function for Disc Ba (0):

Here again we use the reflection/inverse point to define the correction function w(x). For
a2 x
x ∈ Ba (0), define x∗ = |x| 2 . This point is called inverse point with respect to the boundary

|y| = a. Note that

|y − x∗ |2 = (y − x∗ ) · (y − x∗ ) = |y|2 − 2y · x∗ + |x∗ |2
2 ∗ a2 x
∗ 24 |x|
2
2
= a − 2y · x + |x | = a − 2y · 2 + a
|x| |x|4
a2 a2
= 2 |x|2 − 2x · y + |y|2 = 2 |x − y|2


|x| |x|

24
 |y−x∗ | a2
Therefore, |y−x|
= |x|2
for all y such that |y| = a. Now define
 
|x| ∗
wx (y) = −Φ |y − x |
a

This is an anlytic function for y ∈ Ba (0) and moreover wx (y) = −Φ(y − x) for y ∈ ∂Ba (0).
The unit normal in this case is n̂ = ay . So ∂G
∂n
= n̂ · ∇G.
 
∂G ∂Φ ∂ |x|
(x, y) = (y − x) − Φ (y − x∗ )
∂yi ∂yi ∂yi a

∂Φ 1 y i − xi
(y − x) =
∂yi wn |y − x|n

   2−n
∂Φ |x| 1 |x|
∗
(y − x ) = (yi − x∗i ) |y − x∗ |−n
∂yi a wn a
 2  −n
1 |x| |x| ∗
= |y − x | (yi − x∗i )
wn a a
 2
yi |x2 − a2 xi
 
|x| 1 −n
= |y − x|
a wn |x|2
1 1
|y − x|−n yi |x2 − a2 xi


= 2
a wn

Therefore,
 
∂G 1 1 2 2
= yi − xi − 2 (yi |x| − a xi )
∂yi wn |y − x|n a
yi
= 2 (a2 − |x|2 )
a wn |y − x|n

Hence
n
∂G X ∂G yi
= ·
∂n i=1
∂y i a
X y2 n
1 2 2 i
= 2 (a − |x| )
a wn |y − x|n i=1
a
1
= n
(a2 − |x|2 ) since |y| = a.
awn |y − x|

25
So from the formula (3.21) we obtain

a2 − |x|2
Z
g(y)
u(x) = dSy
wn ∂Ba (0) |x − y|n

Now following the similar steps as in the previous case one can show

lim u(x) = g(x0 ), for any point x0 ∈ ∂Ba (0).

x→x0 ,x∈Ba (0)

In case of general domains the following is useful.

Theorem 3.7. If u is harmonic function and O is an orthogonal n × n matrix. Then

v(x) = u(Ox) is also harmonic function.

Proof. Let O = [oij ]. We can compute

n
X
Di v(x) = Dk u(Ox)oki ,
k=1

n X
X n
Dij v(x) = Dkl u(Ox)oki olj .
l=1 k=1

Since OOT = I. We have for all k, l = 1, ...n


n
X 1 k = l
oki oli = .
0 k ̸= l
i=1

Thus
n X
X n X
n
∆v(x) = Dkl u(Ox)oki oli
i=1 l=1 k=1
n X
n n
!
X X
= Dkl u(Ox) oki oli
l=1 k=1 i=1
n X
X n
= Dkl u(Ox)δkl = ∆u(Ox) = 0.
l=1 k=1

26
From the above discussions we established the following:

Theorem 3.8. Let Ω be a domain in Rn such that the Green’s function G(x, y) = Φ(x, y)+
wx (y) exists. Then u(x) defined in (3.16) is the unique solution to the Dirichlet Problem
(3.15) when f ∈ C 1 (Ω) with all its first oder partial derivatives are in L∞ (Ω) and g ∈
C(∂Ω) ∩ L∞ (∂Ω).

Proof. Proof now follows from the Theorem 3.5 and the above Poisson integral formula.
This as left as an exercise.

3.2 Properties of harmonic functions

Theorem 3.9. (Harnack’s inequality): Suppose u ∈ C 2 (Ω) is a non-negative harmonic
function and let Ω1 ⊂⊂ Ω be bouned sub domain. Then there exists C1 depending only on
Ω1 such that
sup u ≤ C1 inf u.
Ω1 Ω1

Proof. From Poison integral formula for ball Ba (0), we have

a2 − |x|2
Z
u(y)
u(x) = dSy (3.22)
awn ∂Ba (0) |x − y|n

Now using the fact that

a − |x| ≤ |x − y| ≤ a + |x|, for |y| = a, x ∈ Ba (0),

and Z
1
u(0) = u(y)dSy
wn an−1 ∂Ba (0)

from (3.22), we get

a + |x|
Z
u(x) ≤ u(y)dSy
awn (a − |x|)n−1 ∂Ba (0)
an−2 (a + |x|)
= u(0)
(a − |x|)n−1

Similarily, we get
an−2 (a − |x|)
u(x) ≥ u(0)
(a + |x|)

27
From this we can show that by taking supremum and infimum over B a2 we get

sup u(x) ≤ C inf u(x).

B a (0) B a (0)
2 2

Now for any relatively compact set Ω1 use a finite cover of a/2 balls.

Theorem 3.10. Liouville’s Theorem:

If u(x) is bounded and harmonic in Rn . Then u(x) is a constant.

Proof. From (3.22), differentiating H(x, y) with respect to xj , we get

 
1 −2xj y j − xj
Hxj (x, y) = n
+ n(a2 − |x|2 )
awn |x − y| |x − y|n+2

Therefore
na yj
Hxj (0, y) =
wn an+2
nyj
=
wn an+1

Therefore from (3.22), we get

Z
∂u
| (0)| = Hxj (0, y)u(y)dSy
∂xj ∂Ba (0)
Z
n
= yj u(y)dSy
wn an+1 |y|=a
n n
≤ n+1 an sup |u(y)| ≤ sup |u(y)|
a |y|=a a Ba

This can be proved for any ball around a point x. That is, we get ak
n
|uxj (x)| ≤ ∥u∥L∞ (Ω) (3.23)
ak

This shows that u is infinitely differentiable. Now since u is harmonic in the whole of Rn ,
taking ak → ∞ we get u is constant.

We can extend (3.23) to compact subset Ω1 ⊂ Ω in which u is harmonic. Let d > 0 be the
distance of Ω1 to ∂Ω, pick an increasing sequence dk → d. We can apply (3.23) to each of

28
Bdk (ξ) for every ξΩ1 :

∂u n n
| (ξ)| ≤ max |u(x)| ≤ ∥u∥L∞ (Ω)
∂ξj dk ∂Bdk (ξ) dk

Now taking maximum over ξ and taking limit in dk we get

∂u n
max | (ξ)| ≤ ∥u∥L∞ (3.24)
Ω1 ∂ξj d

Theorem 3.11. If uk ∈ C 2 (Ω) is a uniformly bounded sequence of harmonic function

in Ω and let Ω1 ⊂⊂ Ω, then uk contains a subsequence that converges uniformly on Ω1 .
Moreover the limit is also harmonic function.

Proof. Using (3.23) on uxi we get replacing a by a/2

 2
2n
uxi xj (x) ≤ sup |u(y)|.
a Ba/2

Taking supremum over Ba we get

 2
2n
sup |uxi xj (x)| ≤ sup |u(x)|
Ba a Ω

following the argument as in (3.24), we get for any Ω1 , a compact subset of Ω:

 2
2n
sup |uxi xj (x)| ≤ sup |u(x)|
Ω1 d Ω

where d is the distance from Ω1 to ∂Ω Then by Ascoli-Arzela theorem there exists a

subsequence that converges uniformly. Then applying the above inequality to uk −ul shows
that the second order derivatives of the uk also converge uniformly on Ba (x). Then applying
the above inequality to um − un we get that for this sub sequence we have ∆uk → ∆u.
But ∆uk = 0 implies ∆u = 0. Moreover, u is C ∞ .

Theorem 3.12. If u ∈ C 2 (Ω) is harmonic, then u ∈ C ∞ (Ω).

Proof. For ξ ∈ Ω, choose a ball Ba (ξ) ⊂⊂ Ω and u ∂Ba (ξ) := g ∈ C(∂Ba (ξ)). Then by the

29
existence of Green’s function, we can show that there exists u1 ∈ C ∞ (Ω) such that

∆u1 = 0 in Ba (ξ), u1 = g on ∂Ba (ξ).

But by uniqueness we have u = u1 in Ba (ξ). Therefore u ∈ C ∞ (Ω).

Theorem 3.13. If u ∈ C 2 (Ω) is harmonic, then u is real analytic in Ω.

Proof. Pick a point x0 ∈ Ω and a > 0 so that Ba (x0 ) ⊂ Ω. For simplicity x0 = 0. We

want to show that for a1 small, the Taylor series for u converges to u on the ball Ba1 (0).
for small a1 . Let a1 = ϵa. Then the Remainder term in the Taylor’s theorem
X Dα u(tx)
RN (x) = xα , for some 0 ≤ t ≤ 1.
α!
|α|=N

Let M = supBa |u(x)|. Then following (3.24),

 N
α Nn
|D u(tx)| ≤ M
(1 − ϵ)a

Now using the estimate |α||α| ≤ Cα!(en)|α| and |xα | ≤ |x||α| , we get
N
e N nN

1 α α nN
|D u(tx)|x ≤ CM N (ϵa)N
α! n (1 − ϵ)a
n2N eN aN ϵN
=C M
(1 − ϵ)N aN
 2 N
n eϵ
=C M
1−ϵ

ϵen2
Now we can choose ϵ small so that 1−ϵ
< 12 . So
X
|RN (x)| ≤ CM 2−N ≤ CM 2−N (N + 1)n → 0 as N → ∞.
|α|=N

That is, the Taylor’s series converges. Hence u is real analytic.

30
4 Existence Theory (Perron’s method)
In this section we will prove the existence of harmonic function in general bounded domains
(where Green’s function is difficult to evaluate explicitely) with prescribed boundary values
u = g on the boundary. Motivated from the Corollary 2.2, we have the following

Definition 4.1. A function u ∈ C(Ω) is called subharmonic if

u(ξ) ≤ Mu (ξ, r)

for all ξ ∈ Ω and Br (ξ) ⊂⊂ Ω.

We note that the theorem 2.6, is indeed holds for any subharmonic functions:

Theorem 4.1. Let Ω be a connected domain and let u be a subharmonic function. Then
either u is constant or u(ξ) < sup u(x) for all ξ ∈ Ω.
x∈Ω

Theorem 4.2. Comparison theorem: Let u be a harmonic function with u = g on ∂Ω

and let v is subharmonic function in Ω with v ≤ g on ∂Ω. Then v(x) ≤ u(x) in Ω.

Proof. By Mean value theorem 2.3 we have

u(ξ) = Mu (ξ, r), and v(ξ) ≤ Mv (ξ, r)

Therefore, (v − u)(ξ) ≤ Mv−u (ξ, r). Hence v − u is subharmonic function and (v − u)(x) ≤
g(x) − g(x) = 0 on the boundary. Therefore by the above theorem 4.1 we get v ≤ u in
Ω.

Therefore, we consider the class of functions


Sg = u ∈ C(Ω) : u is subharmonic in Ω and u ≤ g on ∂Ω

Then Sg is nonempty for g bounded and continuous function in ∂Ω. Indeed, let k be a
constant such that k ≤ min∂Ω g(x), then k ∈ Sg . Also, if M = max∂Ω |g(y)|, then for any
u ∈ Sg , u(x) ≤ M . Moreover, u − M ∈ S0 . For g ∈ C(∂Ω), define

wg (x) = sup{u(x) : u ∈ Sg }.

Then we have the following existence result

31
Theorem 4.3. wg is harmonic in Ω and lim u(x) = g(x0 ) for all x0 ∈ ∂Ω.
x→x0

Before we prove this, we need the following results:

For a subharmonic function u, define its harmonic representative as ũξ,r (x)

∆ũξ,r = 0 in Br (ξ), ũξ,r = u on ∂Br (ξ), Br (ξ) ⊂⊂ Ω

and its continuous extension uξ,r (x)


u(x) x ∈ Ω\B (ξ)
r
uξ,r (x) =
ũξ,r x ∈ Br (ξ)

Lemma 4.1. We have

(a) u(x) ≤ uξ,r (x) for all x ∈ Ω and

(b) uξ,r is subharmonic in Ω.

Proof. (a) u − uξ,r is subharmonic in Br (ξ) and u − uξ,r = 0 on ∂Br (ξ). Then by theorem
4.2, u(x) ≤ uξ,r (x) in Br (ξ). Also u(x) = uξ,r (x) for all x ∈ Ω\Br (ξ).

(b) Claim: uξ,r (x) ≤ Muξ,r (x, ρ) for all x ∈ Ω and Bρ (x) ⊂⊂ Ω.

(i) If x ∈ Br (ξ), then for small ρ such that, Bρ (x) ⊂ Br (ξ), we have uξ,r is harmonic
in Br (ξ). Therefore claim holds.

(ii) If x ∈ Ω\Br (ξ), then choose Bρ ⊂ Ω\Br (ξ) and u is subharmonic in Ω proves
the claim.

(iii) If x ∈ ∂Br (ξ). Then for x ∈ Ω we have u(x) ≤ uξ,r (x). Therefore

Mu (x, ρ) ≤ Muξ,r (x, ρ) (4.25)

Now x ∈ ∂Br (ξ) and u is subharmonic in Ω implies

uξ,r (x) = u(x) ≤ Mu (x, ρ) (4.26)

From (4.25) and (4.26), we get

uξ,r (x) = u(x) ≤ Mu (x, ρ) ≤ Muξ,r (x, ρ)

32
Lemma 4.2. If u1 , u2 , ..., uk are subharmonic in Ω, then v = max{u1 , u2 , ...uk } is subhar-
monic in Ω.

Proof.
uj is subharmonic =⇒ uj (ξ) ≤ Muj (ξ, r)

uj ≤ v =⇒ Muj (ξ, r) ≤ Mv (ξ, r)

Therefore,

v(ξ) = max{u1 (xi), u2 (ξ), ...uk (ξ)} ≤ max {Mu1 (ξ, r), Mu2 (ξ, r), ..., Muk (ξ, r)} ≤ Mv (ξ, r)}

Hence v is subharmonic in Ω.

Lemma 4.3. wg is harmonic in Ω.

Proof. It is enough to show that wg is harmonic in B r2 (ξ) ⊂ Br (ξ) ⊂ Ω. The proof is

divided into several steps:

1. By the definition of wg , we can find functions ujk ∈ Sg such that

wg (ξ) = lim uj (ξ)

j→∞

2. Let m = min{g(x) : x ∈ ∂Ω} and M = max{g(x) : x ∈ ∂Ω}. Then

m ≤ uj (x) ≤ M, for all x ∈ Ω

3. We can replace uj by ujξ,r in the limit in step 1 as uj ≤ ujξ,r . Still the limit will be
same and so with out of loss of generality we may assume uj in the limit in step 1 is
harmonic in Br (ξ).

4. By step 2, {uj } is uniformly bounded and by step 3 {uj } is harmonic in Br . Therefore

uj converges to a harmonic function w(x) in Br/2 (ξ).

5. w(x) = wg (x) for all x in Br/2 (ξ). Since wg is supremum, w(x) ≤ wg (x) for all
x ∈ Br/2 . Now if there exists a point x′ ∈ Br/2 such that w(x′ ) < wg (x′ ). That

33
 means there exists v ∈ Sg such that w(x′ ) < v(x′ ) < wg (x′ ). Now replacing uj by
max{uj , v} and its harmonic replacement, we obtain the limit η such that

w(x) ≤ η(x) ≤ wg (x), f or all x ∈ Br/2 and w(ξ) = η(ξ)

On the other hand we also have w(ξ) = wg (ξ). Therefore

Z Z Z
1 1 1
η(y)dy ≥ w(y)dy = w(ξ) = η(ξ) = η(y)dy
|Br/2 | Br/2 |Br/2 | Br/2 |Br/2 | Br/2

The only possibility is w ≡ η in Br/2 . But this is a contradiction as

η(x′ ) ≥ v(x′ ) > w(x′ ).

Therefore w ≡ wg in Br/2 .

To proceed further we need the following

Definition 4.2. Barrier function: Qz ∈ C(Ω) is called barrier function at z ∈ ∂Ω if

1. Qz is subharmonic in Ω

2. Qz (z) = 0

3. Qz (x) < 0 for any x ∈ ∂Ω\{z}.

Definition 4.3. Regular boundary:

A point z ∈ ∂Ω is regular if there exists a barrier function at z and ∂Ω is regular if all
z ∈ ∂Ω are regular.

Example 4.1. If ∂Ω satisfies exterior sphere condition at z ∈ ∂Ω, then z is a regular

point. In this case if Bϵ (ξ) is exterior sphere at z ∈ ∂Ω, then we can take

−( |x−ξ| )2−n , n ≥ 3
ϵ
Qz (x) =
− log( |x−ξ| ) n = 2
ϵ

Lemma 4.4. If g ∈ C(∂Ω) and z ∈ ∂Ω is regular then

lim wg (x) = g(z)

x→z

34
Proof. Consider the function u− (x) = g(z) − ϵ + kQz (x) (where ϵ, k will be chosen later).
Then u− is subharmonic in Ω.
Claim 1: u− (x) ≤ g(x) for all x ∈ ∂Ω.
By continuity of g, there exists δ > 0 such that

|x − z| < δ =⇒ |g(x) − g(z) < ϵ.

If x ∈ ∂Ω and |x − z| < δ. Then

u− (x) − g(x) = g(z) − g(x) − ϵ + kQz (x) < 0

If x ∈ ∂Ω, |x − z| ≥ δ. Then choose k such that kQz (x) ≤ −2M where M = max∂Ω |g(x)|.
Then again,
u− (x) − g(x) ≤ 2M − ϵ + kQz (x) < 0.

Therefore, u− (x) ≤ g(x) on ∂Ω and u− is subharmonic. Hence u− ≤ g and

u− (x) ≤ wg (x) for allx ∈ ∂Ω.

Similarly, defining
u+ (x) = g(z) + ϵ − kQz (x)

Then u+ ∈ C(Ω) and −u+ is subharmonic.

Claim 2: −u+ ≤ −g on ∂Ω.

|x − z| < δ =⇒ −u+ (x) + g(x) = −g(z) + g(x) − ϵ + kQz (x) < 0

If |x − z| ≥ δ as earlier choose k such that kQz (x) ≤ −2M .

−u+ (x) + g(x) = −g(z) + g(x) − ϵ + kQz (x) ≤ 2M − ϵ − 2M < 0.

Therefore, −u+ (x) ≤ −g(x) on ∂Ω.

Therefore for any u ∈ Sg , u − u+ is subharmonic and u − u+ ≤ 0 on ∂Ω. Hence by
Maximum principle, u < u+ in Ω. Again by the definition of wg , we get

wg (x) ≤ u+ (x) in Ω.

35
Hence from the cliams, we have

u− (x) ≤ wg (x) ≤ u+ (x) in Ω.

Now using the fact that Qz (x) → 0 as x → z we get |wg (x) − g(z)| → 0.

5 Problems
1. Let Ω be a bounded domain in Rn . Suppose u ∈ C 2 (Ω) ∩ C(Ω) satisfies
n
X ∂u
∆u + − u ≥ 0, in Ω, u = 0 on ∂Ω.
i=1
∂x i

Then show that u ≤ 0 in Ω.

2. Prove the weak form of maximum principle for general second order elliptic opeator
n
X ∂ ∂u X
Lu = (aij ), aij ξi ξj ≥ C|ξ|2 , ξ ∈ Rn
i=1
∂xj ∂xi

3. Suppose q(x) ≥ 0 for x ∈ Ω and consider solutions u ∈ C 2 (Ω)∩C 1 (Ω) of ∆u−q(x)u =

0 in Ω. Establish uniqueness theorem for Dirichlet problem.

4. If Ω is a bounded domain and u ∈ C 2 (Ω) ∩ C(Ω) is harmonic in Ω, then show that

max |u(x)| = max |u(x)|.

Ω ∂Ω

5. Construct the Green’s function for the domain

(a) {(x, y) ∈ R2 : x > 0, y > 0} (b) {(x, y) : x > 0, x2 + y 2 < 1}.

6. Show that Φ(x) is a distributional solutoin of −∆u = δ in Rn .

7. Suppose u ∈ C(Ω) satisfies mean value property in Ω. Then show that boundary
values on Br (ξ) ⊂ Ω uniquely determine u.

8. Show that the bounded solution of the Dirichlet problem in a half-space is unique.
What can you say about unbounded solutions?

36
 9. Let Ω = Ba (0), Ω+ = Ω ∩ Rn+ and Ω0 = {x ∈ Ω, xn = 0}. If u ∈ C 2 (Ω) ∩ C(Ω+ ∪ Ω0 )
is harmonic in Ω+ , and u = 0 on Ω0 . Prove that u may be extended to a harmonic
function on all of Ω (use reflection).

10. Show that u ∈ C 2 (Ω) is subharmonic then ∆u ≥ 0 in Ω.

11. Let f, g be continuous and bounded functions and let u is a smooth solution of −∆u =
f in B(0, 1), u = g on ∂B(0, 1). Then Prove the following stability estimate: There
exists a constant C, depending only on n, such that

max |u| ≤ C( max |g| + max |f |)

B(0,1) ∂B(0,1) B(0,1)

12. Prove the stability estimate in problem 11 above for any bounded domain Ω instead
of B(0, 1).

13. Let Ω = Rn+ and let u ∈ C 2 (Ω) ∩ C(Ω) such that ∆u = 0 in Ω and u = 0 on the
boundary. If u(x) is bounded then show that u ≡ 0. If u is not bounded, then show
that there can be more than one such functions.

14. Let u be a solution of Neumann Problem

∂u
∆u = 0 in Ω = h(x) on ∂Ω.
∂n
R
Then show that ∂Ω
h(x)dS = 0.

37