You are on page 1of 33

Stokes Theorem and Harmonic Functions

Semester Project Report

by

SHUBHAM GIRDHAR
1311041

to the

School of Mathematical Sciences


National Institute of Science Education and Research
Bhubaneswar
16 November 2016

Abstract
The Harmonic functions and Stokes Theorem on Rn are studied. The complex
analysis has been employed to study Harmonic functions on R2 . Greens identity has
been mainly used in the study of harmonic functions on Rn which gives the motivation
to study general Stokes Theorem on Rn .

Contents
1 Introduction

2 Harmonic functions in C
2.1 Basic Properties . . . . . .
2.2 Poissons Formula . . . . .
2.3 Functions with mean value
2.4 Harnacks Principle . . . .

. . . . .
. . . . .
property
. . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

2
2
3
5
6

3 Stokes theorem on Rn
3.1 Algebraic preliminaries .
3.2 Fields and Forms . . . .
3.3 Geometric preliminaries
3.4 Stokes theorem on Rn .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

8
8
12
16
18

4 Harmonic functions on Rn
4.1 Definitions and Basic Properties . .
4.2 Mean value property . . . . . . . .
4.3 Poisson kernel . . . . . . . . . . . .
4.4 Dirichlet problem . . . . . . . . . .
4.5 Functions with mean value property

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

21
21
22
25
26
27

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

References

29

Appendix A

30

Chapter 1
Introduction
This report is concerned with the study of harmonic functions and Stokes theorem
on Rn . The approach taken for harmonic functions on R2 uses complex analysis.
To study harmonic functions on higher dimensions, Greens identities are required.
This gives the motivation to study Stokes theorem in great detail on Rn . All the
properties of harmonic functions are then generalised on Rn .
Harmonic functions are solutions to Laplaces equation and play an important role in
many areas of mathematics, physics, and engineering. Chapter 2 introduces the the
harmonic functions and explores their properties using techniques of complex analysis. This gives us an understanding of what to expect in higher dimensions and make
the learning more focussed.
Chapter 3 discusses the Stokes theorem on Rn in every detail. The proof of the theorem involves many prerequisites like differential forms, chains, and so on and even
the statement of the theorem cannot be understood without all the definitions in this
chapter.
Chapter 4 is devoted to study of the properties of harmonic functions on higher dimensions, generalizing the properties that were introduced in Chapter 2. This is
necessary as harmonic functions on R2 are always real part of some holomorphic
functions on a convex set which is not the case in higher dimensions.

Chapter 2
Harmonic functions in C
2.1

Basic Properties

Definition 1 A real-valued function u(z) or u(x, y), defined and single valued in a
region , is said to be harmonic in , if it is continuous together with its partial
derivatives of first two orders and satisfies Laplaces equation
0 = 4u =

2u 2u
+
.
x2 y 2

The sum of two harmonic functions and a constant multiple of a harmonic function
is again a harmonic due to the linear character of Laplace equation.
Theorem 1 Let u be a real-valued harmonic function in the convex open subset G of
C. Then there is a holomorphic function g in G such that u = Re g. The function g
is unique to within addition of an imaginary constant.
Proof: Consider the function f =

u
x

i u
, which is holomorphic in G (by Cauchyy

Riemanns equations). The function f has a primitive in G that is there is a function


g in G such that g 0 = f . Fix a point z0 G. By adding constant to g, we may assume
g(z0 ) = u(z0 ). Let u1 = Re g and v1 = Im g. We have
g0 =

u1
v1
v1
u1
+i
=
i
,
x
y
x
y

and also
g0 = f =

u
u
i .
x
y

2 Harmonic functions in C

Hence,

u
x

u1
x

and

u
y

u1
.
y

The first partial derivatives of the function u u1

thus vanish throughout G, implying that u u1 is constant. Since u(z0 ) = u1 (z0 ),


the constant is 0, and u = Re g.
Theorem 2 Harmonic functions are of class C .
Proof: As differentiability is a local property, it will suffice to prove this for a harmonic function u in an open disk D. Then, by theorem 1, we can write u = Re g
with g holomorphic in D. The desired conclusion now follows by the indefinite differentiability of holomorphic functions.
Theorem 3 (Mean Value Property): Let u be a harmonic function in an open subset
G of C, and let z0 G. Then
1
u(z0 ) =
2

u(z0 + reit )dt, 0 < r < dist(z0 , C \ G)

(2.1)

Proof: Follows from mean value property for holomorphic functions.


Theorem 4 (Maximum principle) Let u be a nonconstant real-valued harmonic function in the connected open subset G of C. Then u does not attain a local maximum
in G.
Proof: Follows from corresponding property for holomorphic functions.
Remark: By applying theorem 4 to u, we see that u does not attain a local
minimum in G.

2.2

Poissons Formula

The maximum principle has the important consequence: If u(z) is continuous on a


closed bounded set E and harmonic on the interior of E, then it is uniquely determined
by its values on the boundary of E. If u1 and u2 are two such functions with same
3

2 Harmonic functions in C

boundary values, then u1 u2 is harmonic with boundary values 0. By maximum


and minimum principle the difference u1 u2 must be identically zero on E. Hence,
there arises the problem of finding u when its boundary values are given. We shall
solve the problem only for the simplest case, that is for a closed disk.
Formula 2.1 determines the value of u at the center of the disk. But there exist a
linear transformation which carries any point to the center. Suppose that u(z) is
harmonic in the closed disk |z| R. The linear transformation
R(R + a)
R + a

z = S() =

maps || 1 onto |z| R with = 0 corresponding to z = a. The function u(S())


is harmonic in || 1,and by equation 2.1 we obtain
1
u(a) =
2
From =

R(za)
,
R2 az

Z
u(S())d arg.
=1

we compute

d
d arg = i = i

1
a
+ 2
z a R az


dz =

z
az
+ 2
z a R az


d

On substituting R2 = zz the coefficient of d in the last expression, it can be rewritten


as
z
a
R2 |a|2
+
=
.
za za
|z a|2
We obtain the form
1
u(a) =
2

Z
|z|=R

R2 |a|2
u(z)d
|z a|2

(2.2)

of Poissons formula.
Remark: In the derivation we have assumed u(z) is harmonic in the closed disk.
However, the result remains true under the weaker condition that u(z) is harmonic
in the open disk and continuous in the closed disk. Indeed, if 0 < r < 1, then u(rz)

2 Harmonic functions in C

is harmonic in the closed disk, and we obtain


1
u(ra) =
2

Z
|z|=R

R2 |a|2
u(rz)d.
|z a|2

Now let r tend to 1, because u(z) is uniformly continuous on |z| R it is true that
u(rz) u(z) uniformly for |z| = R, and conclude that equation 2.2 remains valid.
Formulating the result as a theorem:
Theorem 5 Suppose that u(z) is harmonic for |z| < R, continuous for |z| R.
Then
1
u(a) =
2

Z
|z|=R

R2 |a|2
u(z)d
|z a|2

for all |a| < R.

2.3

Functions with mean value property

Let u(z) be a real valued continuous function in a region . We say that u satisfies
the mean-value property if
1
u(z0 ) =
2

u(z0 + rei )d

(2.3)

when the disk |zz0 | r is contained . It is already shown that mean value property
implies the maximum principle. It is already shown that every harmonic functions
satisfies the mean value condition, the following theorem proves that converse is also
true.
Theorem 6 A continuous function u(z) which satisfies mean value condition 2.3 is
necessarily harmonic.
Proof: The condition need to be satisfied only for a sufficiently small r. If u satisfies
2.3, so does the difference between u and any harmonic function. Suppose that the
5

2 Harmonic functions in C

disk |z z0 | is contained in , the region where u is defined. By using Poissons


formula we can construct v(z) which is harmonic for |z z0 | < , continuous and
equal to u(z) on |z z0 | = . The maximum and minimum principle, applied to u v,
implies that u(z) = v(z) in the whole disk, and consequently u(z) is harmonic.
The implication of the theorem 6 is that we may define harmonic function to be a
continuous function with the mean-value property. Such a function has automatically
continuous derivatives of all orders, and it satisfies Laplaces equation.

2.4

Harnacks Principle

Let u be a harmonic function in |z| and it can be expressed through its values
on the the circle using Poissons formula as
1
u(z) =
2

Z
0

2 r 2
u(ei )d
|ei z|2

(2.4)

where |z| = r < . Also, the elementary inequalities


2 r2
+r
r

i
2
+r
|e z|
r

(2.5)

together with the equation 2.4 yields the estimate


1 +r
|u(z)|
2 r

|u(ei |d

If it is known that u(ei ) 0 we can use the first inequality 2.5 as well, and obtain
a double estimate
1 r
2 + r

Z
0

1 +r
ud u(z)
2 r

ud
0

but the arithmetic mean of u(ei ) equals u(0), and end up with the following upper
and lower bounds:
r
+r
u(0) u(z)
u(0)
+r
r
6

(2.6)

2 Harmonic functions in C

This is Harnacks inequality and is valid only for positive harmonic functions. The
inequality 2.6 when applied to increasing sequences of harmonic functions leads to a
powerful, yet simple theorem known as Harnacks principle.
From left hand inequality it follows that if {un (z)} is a sequence of positive harmonic
functions on the disk |z z0 | < R such that un (z0 ) +, then un (z) + for
any subdisk |z z0 | < , < R. From right hand inequality it follows that if {un (z)}
is bounded, then un (z) is uniformly bounded on any subdisk |z z0 | < , < R.
Theorem 7 Let {un (z)} be an increasing sequence of harmonic functions on a domain . Then {un (z)} converges uniformly on compact subsets of , either to harmonic function or to +.
proof: Let U be the set of z such that un (z) +. The remarks preceding
the theorem applied to vn = un u1 show that both U and \ U are open. Since
is domain and hence connected implies U = or U is empty. If U = , then u is
identically infinite and uniformity follows by usual compactness argument.
In other case the limit function u(z) is finite everywhere. Then un+p (z) un (z)
3(un+p (z0 ) un (z0 )) for |z z0 | R/2 and n + p n 1. Hence, convergence at
z0 implies uniform convergence in a neighbourhood of z0 , therefore, convergence is
uniform on every compact set. The harmonicity of the limit function can be inferred
from the fact that u(z) can be represented by Poissons formula.

Chapter 3
Stokes theorem on Rn
3.1

Algebraic preliminaries

If V is a vector space (over R), we will denote the k-fold product V V by V k .


A function T : V k R is called multilinear if for each i with 1 i k we have
T (v1 , . . . , vi + vi0 , . . . , vk ) = T (v1 , . . . , vi , . . . , vk ) + T (v1 , . . . , vi0 , . . . , vk ),
T (v1 , . . . , avi , . . . , vk ) = aT (v1 , . . . , vi , . . . , vk ).
Definition 2 A multilinear function T : V k R is called k-tensor on V and is the
set of all k-tensors is denoted by Jk (V ).
If S, T Jk (V ) and a R, we define
(S + T )(v1 , ..., vk ) = S(v1 , ..., vk ) + T (v1 , ..., vk ),
(aS)(v1 , ..., vk ) = a S(v1 , ..., vk ).
Then Jk (V ) becomes a vector space over R. There is also an operation connecting
various spaces Jk (V ). If S Jk (V ) and If T Jl (V ), we define the tensor product
S T Jk+l (V ) by
S T (v1 , ..., vk , vk+1 , ..., vk+l ) = S(v1 , ..., vk ) T (vk+1 , ..., vk+l ).
Note that the order of the factors S and T is crucial since S T and T S are not
equal. The following are properties of :
1. (S1 + S2 ) T = S1 T + S2 T,
8

3 Stokes theorem on Rn

2. S (T1 + T2 ) = S T1 + s t2 ,
3. (aS) T = S (aT ) = a(S T ),
4. (S T ) U = S (T U ) = S T U.
The proof is by simple expansion and equating. The higher-order products T1 Tr
are defined similarly.
Remark: Notice that J1 (V ) is just dual space V . The operation allows us to
express other vector spaces Jk (V ) in terms of J1 (V ).
Theorem 8 Let v1 , ..., vn be the basis for V , and let 1 , ..., n be the dual basis,
i (vj ) = ij . Then the set of all k-fold tensor products
i1 ik 1 i1 , ...ik n
is the basis for Jk (V ), which therefore has dimension nk .
Proof: Note that
i1 ik (vj1 , ..., vjk ) = i1 ,j1 . . . ik ,jk
If w1 , ..., wk are k vectors with wi =
T (w1 , ..., wk ) =

n
X

Pn

j=1

aij vj and T is in Jk (V ), then

a1,j1 . . .ak,jk T (vj1 , ..., vjk ) =

j1 ,...,jk =1

Thus T =

Pn

i1 ,...,ik =1

n
X

T (vj1 , ..., vjk )i1 ik (w1 , ..., wk ).

i1 ,...,ik =1

T (vj1 , ..., vjk ) i1 ik . Consequently the i1 ik

span Jk (V ). Suppose now that there are numbers ai1 ,...,ik such that
n
X

ai1 ,...,ik i1 ik = 0.

i1 ,...,ik =1

Applying both sides of this equation to (vj1 , ..., vjk ) yields ai1 ,...,ik = 0. Thus the
i1 ik are linearly independent.
9

3 Stokes theorem on Rn
One example of tensor products is inner product h, i J2 (Rn ). Therefore, we generalise inner product on V by defining it to be a 2-tensor T such that T is symmetric,
that is T (v, w) = T (w, v) for v, w V and such that T is positive definite, that is
T (v, v) > 0 if v 6= 0.
Another familiar tensor is det Jk (Rn ). To generalize it, we first recall that interchanging any two rows of a matrix changes the sign of its determinant. This suggests
the following definition.
Definition 3 A k-tensor Jk (V ) is called alternating if
(v1 , ..., vi , ..., vj , ..., vk ) = (v1 , ..., vj , ..., vi , ..., vk )
for all v1 , ..., vk V .
The set of all alternating k-tensors is clearly a subspace k (V ) of Jk (V ). Recall that
sign of a permutation , denoted sgn , is +1 if is even and -1 if is odd. If
T Jk (V ), we define Alt(T ) by
Alt(T )(v1 , ..., vk ) =

1 X
sgn T (v(1) , ..., v(k) ),
k! S
k

where Sk is the set of all permutations of the numbers 1 to k.


Now, our main aim is to determine the dimension of k (V ), we want to have theorem
analogous to Theorem 8. If k (V ) and l (V ), then is usually not in
k+l (V ). We therefore define a new product, the wedge product k+l (V ) by
=

(k + l)!
Alt( ).
k!l!

The following are the properties of :


1. (1 + 2 ) = 1 + 2 ,
2. (1 + 2 ) = 1 + 2 ,
10

3 Stokes theorem on Rn

3. a = a = a( ),
4. = (1)kl ,
Theorem 9

1. If S Jk (V ) and T Jl (V ) and Alt(S) = 0, then


Alt(S T ) = Alt(T S) = 0.

2. Alt(Alt( ) ) = Alt( ) = Alt( Alt( )).


3. If k (V ), l (V ) and m (V ), then
( ) = ( ) =

(k + l + m)!
Alt( ) = .
k!l!m!

Proof: We only prove (3). Now,


( ) =

(k + l + m)! (k + l)!
(k + l + m)!
Alt(( ) ) =
Alt( ).
k!l!m!
k!l!m!
k!l!

The other equality is proved similarly.


The higher order products 1 r are defined similarly. If v1 , ..., vn is a basis for
V and 1 , ..., n is the dual basis, we can now construct basis for k (V ).
Theorem 10 The set of all
i1 ik

1 i1 < < ik n

is a basis for k (V ), which therefore has dimension


 
n
n!
=
.
k
k!(n k)!
Proof: If k (V ) Jk (V ), then we can write
=

ai1 ,...,ik i1 ik .

i1 ,...,ik

11

3 Stokes theorem on Rn

Thus
= Alt() =

ai1 ,...,ik Alt(i1 ik ).

i1 ,...,ik

Since each Alt(i1 ik ) is a constant times one of the i1 ik , these


elements span k (V ). Linear independence is proved as in Theorem 8.
If V is of dimension n, it follows from Theorem 10 that k (V ) has dimension 1.
Thus all alternating n-tensors on V are multiples of any non-zero one. Since the
determinant is an example of such a member of n (Rn ), we have
Theorem 11 Let v1 , ..., vn be a basis for V, and let n (V ). If wi =

Pn

j=1

aij vj

are n vectors in V , then


(w1 , ..., wn ) = det(aij ) (v1 , ..., vn ).
Proof: Define Jn (Rn ) by
((a11 , ..., a1n ), . . . , (an1 , ..., ann )) = (

a1j vj , ...,

anj vj ).

Clearly n (Rn ) so = det for some R and = (e1 , ..., en ) = (v1 , ..., vn ).
Theorem 11 shows that a non zero n (V ) splits the bases of V into two disjoint
groups, those with (v1 , ..., vn ) > 0 and those for which (v1 , ..., vn ) < 0. If (v1 , ..., vn )
Pn
and (w1 , ..., wn ) are two bases and A = (aij ) is defined by wi =
j=1 aij vj , then
v1 , ..., vn and w1 , ..., wn are in the same group if and only if det A > 0. This criterion
is independent of . Either of these two groups is called orientation for V .

3.2

Fields and Forms

If p Rn , set of all pairs (p, v), v Rn is denoted by Rnp is called the tangent space
of Rn at p. This set is made into a vector space by defining
(p, v) + (p, w) = (p, v + w)
12

3 Stokes theorem on Rn

a (p, v) = (p, av)


We will denote (p, v) as vp .The vector space Rnp is so closely allied with Rn that many
structures of Rn are analogues. In particular, the usual inner product for Rnp , h, ip is
defined as hvp , wp i = hv, wi and the usual orientation for Rnp is [(e1 )p , . . . , (en )p ].
Definition 4 A vector field is a function F such that F (p) Rnp for each p Rn .
For each p there are numbers F 1 (p), . . . , F n (p), such that
F (p) = F 1 (p) (e1 )p + + F n (p) (en )p
We thus obtain n-components functions F i : Rn R. Also, F is said to be continuous,differentiable, etc., if each F i is. Similar considerations may be applied to such
that (p) k (Rnp ) and such functions are called k-form on Rn or differential forms
on Rn . If 1 (p), . . . , n (p) is the dual basis to (e1 )p , . . . , (en )p , then
(p) =

i1 ,...,ik (p) [i1 (p) ik (p)]

i1 <<ik

for certain functions i1 ,...,ik ; the form is called continuous, differentiable, etc.,
if these functions are. From now on, we shall assume that differentiable means
C , this allow us to eliminate the need for for counting how many times a function
is differentiable in a proof. The sum + , product f , and wedge product are
in obvious way. A function f is considered to be a 0-form and f is also written as
f .
Let f : Rn R is differentiable, then Df (p) 1 (Rn ). By a minor modification we
can obtain a 1-form df by
df (p)(vp ) = Df (p)(v).
Let us consider in particular the 1-forms dxi . Since dxi (p)(vp ) = Dxi (p)(v) = v i ,
which implies that dx1 (p), ..., dxn (p) is just the dual basis to (e1 )p , . . . , (en )p . Thus
13

3 Stokes theorem on Rn

every k-form can be written


=

i1 ,...,ik dxi1 dxik .

i1 <<ik

The expression for df is of particular interest.


Theorem 12 If f : Rn R is differentiable, then
df = D1 f dx1 + + Dn f dxn .
In classical notation,
f n
f 1
dx
+

+
dx .
x1
xn
P
P
Proof: df (p)(vp ) = Df (p)(v) = ni=1 v i Di f (p) = ni=1 dxi (p)(vp ) Di f (p).
df =

An important construction associated with forms is a generalization of the operator


d which changes 0-forms into 1-forms. If
=

i1 ,...,ik dxi1 dxik ,

i1 <<ik

we define a (k+1)-form d, the differential of , by


d =

i1

ik

di1 ,...,ik dx dx =

D (i1 ,...,ik )dx dxi1 dxik .

i1 <<ik =1

i1 <<ik

Theorem 13

n
X X

1. d( + ) = d + d

2. If is a k-form and is an l-form, then


s( ) = d + (1)k d
3. d(d) = 0. Briefly, d2 = 0.
Proof: For (1), expand the terms and equate.
For (2), note that the formula is true for if = dxi1 dxik and = dxj1 dxjl ,
14

3 Stokes theorem on Rn

since all the terms vanish. The is easily checked when is a 0-form. The general
formula may be derived from above part and these two observations.
Since
n
X X

d =

D (i1 ,...,ik ) dx dxi1 dxik ,

i1 <<ik =1

we have
n X
n
X X

d(d) =

D, (i1 ,...,ik )dx dx dxi1 dxik .

i1 <<ik =1 =1

In this sum the terms


D, (i1 ,...,ik )dx dx dxi1 dxik
and
D, (i1 ,...,ik )dx dx dxi1 dxik
cancel in pairs. This concludes the proof of (3).
Definition 5 A form is called closed if d = 0 and exact if = d, for some .
Theorem 13 shows us that every exact form is closed, and it is natural to ask whether,
P
conversely, every closed form is exact. Suppose that = ni=1 i dxi is a 1-form on
P
Rn and happens to equal df = i=1 f dxi . We can clearly assume that f (0) = 0.
We have,
Z
f (x) =
0

d
f (tx)dt =
dt

Z
0

n
X

Di f (tx) x dt =
0

i=1

n
X

i (tx) xi dt.

i=1

This suggests that in order to find f , given , we have to consider the function I,
defined by
Z
I(x) =
0

n
X

i (tx) xi dt

i=1

Note: The definition of I makes sense if is defined on an open set A Rn with


the property that whenever x A, the line segment from 0 to x is contained in A.
Such an open set is called star-shaped with respect to 0.
15

3 Stokes theorem on Rn
Theorem 14 (Poincare Lemma). If A Rn is an open set star-shaped with respect
to 0, then every closed form on A is exact.
Proof: We will define a function I from k-forms to (k-1)-forms (for each k), such
that I(0) = 0 and = I(d) + d(I) for any form . It follows that = d(I) if
d = 0. Let
X

i1 ,...,ik dxi1 dxik .

i1 <<ik

Since, A is a star-shaped we can define


Z
l
X X
1
I(x) =
(1)
i1 <<il =1

1
l1


i1 ,...,il (tx)dt xxi dxi1 ((dxi )) dxil .

The symbol ((.)) on dxi indicates that it is omitted. The proof that = I(d) +
d(I) is an elaborate computation which we skip here.

3.3

Geometric preliminaries

Definition 6 A singular n-cube in A Rn is a continuous function c : [0, 1]n


A.
Example: A simple but important example of a singular n-cube in Rn is standard
n-cube, I n : [0, 1]n Rn defined by I n (x) = x, x [0, 1]n . A singular 1-cube is
often called as curve.
Definition 7 A finite sum of singular n-cubes with integer coefficients is called an
n-chain in A.
In particular a singular n-cube c is also considered as an n-chain 1 c. It is clear
how n-chains can be added, and multiplied by integers. For example, 2(c1 + 3c2 ) +
(2)(c1 + c3 ) = 6c2 2c3 .
For each n-chain c in A, an (n-1)-chain will be defined in A, called boundary of A
16

3 Stokes theorem on Rn
and denoted by c. The boundary of I 2 may be defined as the sum of four singular
1-cubes arranges in counter-clockwise direction around boundary of [0, 1]2 .
n
n
,
and I(i,1)
Now, for each i with 1 i n, we define two singular (n-1)-cubes I(i,0)

called (i, 0) face of I n and (i, 1) face of I n respectively as:


If x [0, 1]n1 , then:
n
(x) = I n (x1 , . . . , xi1 , 0, xi , ..., xn1 ) = (x1 , . . . , xi1 , 0, xi , ..., xn1 )
I(i,0)

n
(x) = I n (x1 , . . . , xi1 , 1, xi , ..., xn1 ) = (x1 , . . . , xi1 , 1, xi , ..., xn1 )
I(i,1)

Therefore, the boundary of I n is defined as:


n

I =

n X
X

n
(1)i+ I(i,)
.

i=1 =0,1

For a general singular n-cube c : [0, 1]n A, we define (i, ) as:


n
c(i,) = c (I(i,)
)

Definition 8 Boundary of a general singular n-cube c is defined as:


c =

n X
X

(1)i+ c(i,)

i=1 =0,1

Boundary of a n-chain

ai ci is defined as:
X
X
(
ai c i ) =
ai (ci ).

We include a special property of here.


Theorem 15 If c is an n-chain in A, then (c) = 0. Briefly, 2 = 0.
n
Proof: Let i j and consider (I(i,)
)(j,) . If x [0, 1]n2 then we have,

n1
n
n
(I(i,)
)(j,) = I(i,)
(I(j,)
(x)) = I n (x1 , ..., xi1 , , xi , ..., xj1 , , xj , ..., xn2 )

17

3 Stokes theorem on Rn

Similarly,
n
(I(j+1,)
)(i,) = I n (x1 , ..., xi1 , , xi , ..., xj1 , , xj , ..., xn2 )
n
n
Thus, (I(i,)
)(j,) =(I(j+1,)
)(i,) i j. It follows that (c(i,) )(j,) = (c(j+1,) )(i,) i j.

Now,
(c) = (

n X
X

(1) c(i,) ) =

i=1 =0,1

n X X
n1 X
X

(1)i+++j (c(i,) )(j,) .

i=1 =0,1 j=1 =0,1

In this sum, (c(i,) )(j,) and (c(j+1,) )(i,) occur with opposite signs. Therefore, all the
terms cancel in pairs. Since, theorem is true for any singular n-cube, it is also true
for singular n-chains.

3.4

Stokes theorem on Rn

The fact that d2 = 0 and 2 = 0, suggest some connection between chains and forms.
This connection is established by integrating forms over chains. Henceforth, only
differentiable singular n-cubes will be considered.
Let be a k-form on [0, 1]k then = f dx1 dxk for a unique function f . We
define:
Z

Z
=

It can also be written as,


Z
Z
1
k
f dx dx =
[0,1]k

(3.1)

[0,1]k

[0,1]k

f (x1 , . . . , xk )dx1 . . . dxk .

[0,1]k

If is a k-form in A and c is a singular k-cube in A, we define:


Z
Z
=
c .
[0,1]k

A special definition must be made for k = 0, if c : 0 A is a singular 0-cube in A


the define,
Z
= (c(0))
c

18

3 Stokes theorem on Rn

Definition 9 The integral of over a k-chain c =


Z
:=

ai ci is defined to be

Z
ai

ci

Example:The integral of a 1-form over a 1-chain is often called a line integration. If


P dx + Qdy is a 1-form on R2 and c : [0, 1] R2 is a singular 1-cube(a curve) then
Z
P dx + Qdy = lim
c

n
X

[c1 (ti ) c1 (ti1 )]P (c(ti )) + [c2 (ti ) c2 (ti1 )]Q(c(ti ))

i=1

where t0 , . . . , tn is a partition of [0,1]. The limit is taken over all partitions such that
max |ti ti1 | 0. The right hand side is generally taken to be the definition of
R
P dx + Qdy and it is also natural as ordinary integrals involves the sum and limit.
c
However, this definition is difficult to work with and is also not valid in general case.
The relation between d and is neatly summed up in Stokes theorem, sometimes
also called fundamental theorem of calculus in higher dimensions.
Theorem 16 (Stokes Theorem): If is a (k-1)-form on an open set A Rn and c
is a k-chain in A, then
Z

Z
d =

Proof: Suppose first that c = I k and is a (k-1)-form on [0, 1]k . Then is of the
type
f dx1 ((dxi )) dxk
(The symbol ((.)) around dxi indicates that it is omitted.) It is sufficient to prove
theorem for each of these. This simply involves a computation:
Note that
Z
[0,1]k1

k
I(j,)
(f dx1 ((dxi )) dxk ) =

Therefore
19

 R

[0,1]k

f (x1 , ..., , ..., xk )dx1 dxk if i=j


o/w

3 Stokes theorem on Rn

f dx1 ((dxi )) dxk

I k
k X
X

j+

= (1)

k
I(j,)
(f dx1 ((dxi )) dxk )

(1)

[0,1]k1

j=1 =0,1
i+1

f (x , ..., 1, ..., x )dx dx +(1)

[0,1]k

f (x1 , ..., 0, ..., xk )dx1 dxk .

[0,1]k

on the other hand,


R
d(f dx1 ((dxi )) dxk )
Ik
R
R
= [0,1]k Di f dxi dx1 ((dxi )) dxk = (1)i1 [0,1]k Di f.
By Fubinis theorem and the fundamental theorem of calculus (in one dimension) we
have
R
d(f dx1 ((dxi )) dxk )
Ik

R
R1
1
= (1)i1 0 0 Di f (x1 , ..., xk )dxi dx1 ((dxi )) dxk
R1
R1
= (1)i1 0 0 [f (x1 , ..., xk ) f (x1 , ..., 0, ..., xk )]dx1 ((dxi )) dxk
R
R
= (1)i1 [0,1]k f (x1 , ..., 1, ..., xk )dx1 dxk + (1)i [0,1]k f (x1 , ..., 0, ..., xk )dx1 dxk .
Thus
Z

Z
d =

.
I k

Ik

If c is an arbitrary singular k-cube, working through the definitions will show


Z

Z
c .

=
I k

Therefore,
Z

Z
c (d) =

d =
Ik

Finally, if c is a k-chain

d(c ) =
Ik

c =
I k

d =

Z
ai

d =
ci

Z
ai

20

Z
=

ci

.
c

ai ci , we have

Z
c

.
c

Chapter 4
Harmonic functions on Rn
4.1

Definitions and Basic Properties

Harmonic functions considered will be on open subsets of real Euclidean spaces. Also,
n will denote a fixed positive integer greater than 1 and will denote open, non-empty
subset of Rn .
Definition 10 A twice continuously differentiable, complex valued function u defined
on is harmonic on if it follows Laplaces equation,
4u = 0
where 4 = D12 + + Dn2 and Dj2 denotes the second partial derivative with respect
to j th coordinate variable.
Let x = (x1 , ..., xn ) denote a typical point in Rn and let |x| = (x21 + + x2n )1/2 denote
the euclidean norm of x.
Example: The non-constant harmonic functions are coordinate functions, u(x) = x1 .
A slightly more complex example is the function on R3 defined by
u(x) = x21 + x22 2x23 + ix2 .
Also, the function u(x) = |x|2n is harmonic on Rn \ 0 for n > 2.
Because the Laplacian is linear on C 2 (), sums and scalar multiples of harmonic
functions are harmonic. For y Rn and u a function on , the y-translate of u is
the function on + y whose value at x is u(x y). Clearly, translations of harmonic
21

4 Harmonic functions on Rn

functions are harmonic. For a positive number r and u a function on , the r-dilate
of u, denoted by ur , is of the function
(ur )(x) = u(rx)
defined for x in (1/r) = {(1/r)w | w }. If u C2 (), then 4(ur ) = r2 (4ur )r
on (1/r). Hence, dilates of harmonic functions are harmonic.

4.2

Mean value property

The formal similarity between Laplacian 4 = D12 + + Dn2 and the function |x|2 =
x21 + + x2n , whose level sets are spheres centred at the origin. The connection
between harmonic functions and spheres is central to harmonic function theory. The
mean-value property, which we discuss in this section, best illustrates this connection.
Another connection involves linear transformations on Rn called orthogonal, that
preserve the unit sphere. A linear map T : Rn Rn is orthogonal if and only if
|T x| = |x| x Rn .
Theorem 17 If T is orthogonal and u C 2 (), then
4(u T ) = (4u) T
on T 1 ().
Proof: Let [tjk ] denote the matrix of T relative to standard basis of Rn . then
Dm (u T ) =

n
X

tjm (Dj u) T

j=1

, Differentiating once more and summing over m yields


4(u T ) =

n X
n
X

tkm tjm (Dk Dj U ) T

m=1 j,k=1

22

4 Harmonic functions on Rn

n
X

n
X

j,k=1

m=1

n
X

!
tkm tjm (Dk Dj u) T

(Dj Dj u) T = (4u) T,

j=1

as desired. The function u T is called rotation of u. Hence, rotations of harmonic


functions are harmonic.
We will need following Greens identity, as it is key to the proof of mean value property:
Z

Z
(u 4 v v 4 u) dV =

(uDn v vDn u) ds.

(4.1)

Here is a bounded open set of Rn with smooth boundary. The measure V = Vn


is Lebesgue volume measure on Rn , and s denotes surface are measure on . The
symbol Dn denotes the differentiation with respect to outward unit normal n. The
Greens identity (4.1) follows easily from divergence theorem of calculus:
Z
Z
div w dV =
w n ds.

(4.2)

Here, w = (w1 , ..., wn ) is the smooth vector field on a neighbourhood of . To obtain


equation 4.1, simply let w = u 4 v v 4 u and compute. Following is a useful form
of Greens identity when u is harmonic and v = 1:
Z
Dn u ds = 0.

(4.3)

Notation: B(a, r) = {x Rn : |x a| < r} is the open ball centred at a of radius


r; its closure is the closed ball b(a, r). The unit ball B(0, 1) is denoted by B. The
boundary of unit sphere B is denotes by S. The normalized surface area measure on
S is denoted by (so that (S) = 1).
Theorem 18 (Mean Value Property): If u is harmonic on B(a, r), then u equals the
average of u over B(a, r). More precisely,
Z
u(a) =
u(a + r) d().
S

23

4 Harmonic functions on Rn

Proof: Assume that n > 2. WLOG, we may assume that B(a, r) = B. Fix (0, 1).
Apply Greens identity (4.1) with = {x Rn | < x < 1} and v(x) = |x|2n to
obtain
Z
0 = (2 n)

u ds (2 n)

1n

Z
u ds

Dn u ds
S

2n

Z
Dn u ds.
S

By equation 4.3, the last two terms are 0, thus


Z
u ds =

1n

Z
u ds,
S

which is same as
Z

Z
u d =

u() d().
S

Letting 0 and using the continuity of u at 0, we obtain the desired result.

Theorem 19 (Maximum Principle): Suppose is connected, u is real valued and


harmonic on , and u has a maximum or a minimum in . Then u is constant.
Proof: Suppose u attains a maximum at a . Choose r > 0 such that B(a, r) .
If u were less than u(a) at some point of B(a, r), then the continuity of u would show
that average of u over B(a, r) is less than u(a), contradicting mean value property.
Therefore, u is constant on B(a, r), proving that the set where u attains its maximum
is open in . Because this set is also closed in (again by continuity of u), it must
be whole of (by connectivity). Thus u is constant on . If u attains in , we can
apply same argument to u.
We immediately get the following useful corollary, whose proof follows directly.
Corollary 1 Suppose is bounded and u is a continuous real valued function on
that is harmonic on . Then u attains its maximum and minimum values over on
.
24

4 Harmonic functions on Rn

4.3

Poisson kernel

In this section we show that for every x B, u(x) is a weighted average of u over S.
More precisely, we show that there exists a function P on B S such that
Z
u()P (x, )d()
u(x) =
S

for every x B and every u harmonic on B.


We already know the answer for special case n = 2. Therefore, to find P (x, ) when
n > 2, we start with symmetry lemma.
Lemma 1 For all non-zero x and y in Rn ,


y
x


|y| |y|x = |x| |x|y

Proof: Square both sides and expand using inner product.


Now, let n > 2 and suppose that u is harmonic on B. Fix x B. It is natural to try
v(y) = |y x|2n . This function is harmonic on B \ {x}, has a singularity at x, but
is not constant on S. However, by symmetry lemma (1) shows that for y S,
2n


x
2n
2n

|y x|
= |x|
.
y |x|2
Notice that the right side of the equation is harmonic( as function of y) on B. Thus
the difference of the left and right sides has all the properties we seek.
So set v(y) = L(y) R(y), where
2n

L(y) = |y x|

2n

, R(y) = |x|


2n


y x
,

|x|2

and choose small enough so that B(x, ) B. now, apply Greens identity (4.1) as
in proof of mean-value property, with = B \ B(x, ). We obtain
Z
Z
Z
0=
uDn v ds (2 n)s(S)u(x)
uDn R ds +
S

B(x,)

25

B(x,)

Dn u ds.

4 Harmonic functions on Rn

Because uDn R and RDn u are bounded on B, the last two terms approach 0 as 0.
Hence
1
u(x) =
2n

Z
uDn v d.
S

Setting P (x, ) = (2 n)1 (Dn v)(), we have the desired formula:


Z
u(x) =

u()P (x, ) d().

(4.4)

1 |x|2
.
|x |n

(4.5)

A computation o fDn v, yields


P (x, ) =

The function derived above is called the Poisson kernel for the ball.

4.4

Dirichlet problem

We now answer a famous question that is given a continuous function f on S, does


there exist a continuous function u on B, with u harmonic B, such that u = f on S.
If there exists such a function, then do we find u. This Dirichlet problem for the ball.
Also, if solution exists, then it is unique by maximum principle.
Let f be a continuous function on S and define extension of f into B by defining
Poisson integral of f :
Z
P [f ](x) =

f ()P (x, )d().


S

Theorem 20 Suppose f is continuous on S. Define u on B by



u(x) =

P [f ](x) if x B
f (x)
if x S

Then u is continuous on B and harmonic on B.


The proof of Theorem 20 depend on the following lemma and properties of Poisson
integral.
26

4 Harmonic functions on Rn
Lemma 2 Let S. Then P (, ) is harmonic on Rn \ {}.
Proof: Calculate the Laplacian using product rule.
Lemma 3 The Poisson kernel has the following properties:
1. P (x, ) > 0 for all x B and all S;
2.

R
S

P (x, ) d() = 1 x B;

3. for every S and every > 0,


Z
P (x, ) d() 0 as x .
||>

Proof: Properties (1) and (2) follow immediately from the formula of Poisson kernel
(4.5). To prove (2), take u to be identically 1 in equation 4.4.
Proof of Theorem 20: To prove that u is continuous on B, fix S and > 0.
Choose > 0 such that |f () f ()| < whenever | | < . For x B, (1) and
(2) of Lemma 3 imply that
R
|u(x) u()| = | S (f () f ())P (x, )d()|
R
R
|| |f () f ()|P (x, )d() + ||> |f () f ()|P (x, )d()
R
+ 2kf k ||> P (x, )d(),
where kf k denote the supremum of |f | on S. The last term above is less than for
x sufficiently close to (by Lemma 3 (3)), proving that u is continuous at .

4.5

Functions with mean value property

We have seen that every harmonic function has the mean-value property. In this
section, we use the solvability of the Dirichlet problem for the ball to prove that
harmonic functions are the only continuous functions having the mean-value property.

27

4 Harmonic functions on Rn

Theorem 21 Suppose u is a continuous function on . If for reach x there is


a sequence of positive numbers rj 0 such that
Z
u(x) =

u(x + rj )d()
S

for all j, then u is harmonic on .


Proof: WLOG, we can assume that u is real valued. Suppose that B(a, r) . Let
v solve the Dirichlet problem for B(a, r) with boundary data u on B(a, r).
Suppose that v u is positive at some point of B(a, r). Let E be subset of B(a, r)
where vu attains its maximum. Because E is compact, E contains a point x farthest
from a. Clearly, x B(a, r), so there exist a ball B(x, r0 ) B(a, r) such that u(x)
equals the average of u over B(x, r0 ). Because v is harmonic, we have
Z
(v u)(x) =

(v u)(x + r0 ())d().

But (v u)(x + r0 ()) < (v u)(x) S, on a non-empty open subset of S,


contradicting equation above. Thus, v u 0 on B(a, r). Similarly, v u 0 on
B(a, r).

28

References
[1] Ahlfors, L., Complex Analysis, 3 ed. (McGraw-Hill,New York, 1979).
[2] T. W. Gamelin, Complex Analysis, (Springer-Verlag, New York, 2001).
[3] M. Spivak, Calculus on Manifolds, (Addison-Wesley,New York, 1965).
[4] S. Axler, P. Bourden and W.Ramey, Harmonic Function Theory, 2 ed. (Springerverlag,New York, 2001).

29

Appendix A
Theorem 22 (Fubinis Theorem) Let f be bounded on R = [a, b] [c, d] and assume
that set of discontinuities of f on R has zero area. If every line parallel to the axes
meet S in atmost finitely many points then
Z bZ

Z Z
f dA =
R

f (x, y)dx dy =
a

f (x, y)dy dx
c

30

You might also like