Electrodynamics

Lecture 3.4
Greens Functions
Pankaj Sharan
Physics Department, Jamia Millia Islamia
New Delhi
1981-2011
Lecture 3.4
Greens Functions
1
Superposition Principle
The static electric eld due to a charge density (r) is
governed by the Poisson eqution

2
E =

0
This equation is linear in E. If the charge density can be
written as a sum (or superposition) of two densities (r) =

1
(r) +
2
(r) with electric elds E
1
and E
2
respectively then
using

2
E
1
=

1
(r)

0
, and
2
E
2
=

2
(r)

0
we can infer that

2
(E
1
+E
2
) =

1
(r) +
2
(r)

0
.
This shows that the eld solution for a sum of charge distri-
butions can be obtained by simply adding the solutions for
individual distributions.
We can continue the process of dividing the charge dis-
tributions indenitely till the limit of point charges. A point
charge density is given by a Dirac delta function. Therefore
if we can solve the problem of the eld of a point charge, we
can solve it for any other charge distribution.
1
The solutions of linear partial dierential equations (PDE)
of this kind where the right hand side of the equation cor-
responding to the source of the eld is a delta function are
called Greens functions or fundamental solutions.
2
Three dimensional Dirac-delta
First we dene the three-dimensional Dirac delta function

3
(r) as

3
(r) = (x)(y)(z)
=
_
1
(2)
_
e
ik
x
x
dk
x
_
(similar integral for k
y
) (for k
z
)
=
1
(2)
3
_
e
ik.r
d
3
k
The function has value zero at all points except the origin
r = 0, and is such that for any smooth function of r, the
integral
_
d
3
r(r)(r) = (0)
In particular we can write (choosing for a smooth function
which is equal to unity around the origin)
_
V
d
3
r(r) = 1
where V is a region including the origin. The integral is zero
when the region V excludes the origin r = 0. This shows,
for example, that q(r) represents a charge density of a point
particle located at the origin with charge q.
2
3
Greens function for the Poisson Equation
The method of Greens function involves solving the equa-
tion

2
G(r r

) =
3
(r r

)
Suppose we are able to solve this equation. Then the solution
to our original Poisson equation can be written as
(r) =
1

0
_
d
3
r

G(r r

)(r

)
because

2
_
d
3
r

G(r r

)(r

) =
_
d
3
r

2
G(r r

)(r

)
=
_
d
3
r

3
(r r

)(r

)
= (r)
Before the calculation of the Greens function we give the
answer and verify that it actually is the solution. The solution
is

2
_

1
4r
_
=
3
(r)
and you should commit it to memory. This is a very important
formula.
By expressing
2
in polar coordinates

2
=
1
r

2
r
2
r +
1
r
2
sin

sin

+
1
r
2
sin
2

2
3
we see that
2
(1/r) = 0 for any r = 0. Therefore we must
check
_
V
d
3
r
2
_

1
4r
_
= 1
when the volume V includes the origin.
Let us take the region V to be a sphere of radius a centered
at the origin. Then using the Gauss theorem
_
V
d
3
r .
_

1
4r
__
=
1
4
_
S

_
1
r
_
.ndS
where n is the unit radial vector on spherical surface S. Using
the fact that at the boundary S of V where r = a = 0,

_
1
r
_
=
r
r
3
=
n
r
2
=
n
a
2
we get
1
4a
2
_
S
n.ndS =
1
4a
2
_
S
dS = 1
which proves the result.
4
Calculation of the Greens Function
The greens functions are solutions of PDEs which involve
generalized function like the Dirac delta. Therefore they are
generalized functions themselves.
Normally, one begins the calculation of Greens function
by assuming a Fourier transform
G(r) =
1
(2)
3
_
d
3
k

G
(k) e
ik.r
4
Then, applying
2

2
G(r) =
1
(2)
3
_
d
3
k(k
2
)

G
(k) e
ik.r
and equating it to the fourier expansion of the delta function
(r) =
1
(2)
3
_
d
3
ke
ik.r
we get

G
(k) =
1
k
2
So the Greens function is
G(r) =
1
(2)
3
_
d
3
k
1
k
2
e
ik.r
provided we can integrate the right hand side. But we cannot.
The right hand side is singular because of the 1/k
2
factor.
By a method which should, by now, be familiar to you,
we re-interpret the integral by introducing a small positive
number and equating it as the limit 0 of
G

(r)
1
(2)
3
_
d
3
k
1
k
2
+
2
e
ik.r
There are a number of tricks used at this stage :
1. The integral on the right hand side in independent of
what directions are chosen for the axes in the k-space.
For the xed value of r we chose the direction of k
z
axis along r. Then calling the polar coordinates in the
k-space by k, , we get k.r = kr cos ,
2. Integrating gives a factor of 2.
5
3. Calling cos = u
G

(r) =
1
(2)
2
_

0
k
2
dk
_
1
1
du
e
ikru
k
2
+
2
=
1
ir(2)
2
_

0
k dk
e
ikr
e
ikr
k
2
+
2
=
1
ir(2)
2
_

k dk
e
ikr
k
2
+
2
where the two integrals in k over half range (0, ) are
combined into a single integral over (, +).
4. The integral over k can be interpreted as a complex k-
integration along a closed contour running along the real
axis from to and along an innite semicircle in
the upper half plane. Why upper half plane? Because
on this part of the contour the integrand is zero :
exp(i(k + i)r) = exp(r + ikr) 0 for large . (In
the lower half plane a similar contour will spell disaster
: the integrand blows up exponentially!)
5. There is one pole in the upper half plane, at k = i,
k
e
ikr
k
2
+
2
=
e
ikr
2
_
1
k + i
+
1
k i
_
and the residue theorem of complex integration gives
G

(r) =
1
ir(2)
2
_

k dk
e
ikr
k
2
+
2
=
1
ir(2)
2
(2i)
e
r
2
=
e
r
4r
6
Remark 1 The result given above is itself of great importance.
We have not used the smallness of anywhere. Thus the Greens
function G
m
of the static Klein-Gordon equation
(
2
m
2
)G(r) = (r)
is
G
m
(r) =
e
mr
4r
.
It is called the Yukawa potential.
The Greens function for the Poisson equation is, nally,
G(r) = lim
0
G

(r)
= lim
0

e
r
4r
=
1
4r
5
Greens function for the Helmholtz equation
The Helmholtz equation is
(
2
+
2
) = source term, = a real number.
Its Greens function satises
(
2
+
2
)G(r) = (r)
The Fourier transform

G
(k) of G(r) is
G(r) =
1
(2)
3
_
d
3
k

G
(k) e
ik.r
7
where

G
=
1
|k|
2

2
The integral is problematic because the integrand is singular
at k = |k| = .
G(r) =
1
(2)
3
_
d
3
k
1
k
2

2
e
ik.r
We trace the same steps as before : as far as possible.
1. Choose the z-axis in k-space along r so that k.r =
kr cos .
2. d
3
r = k
2
dkd(cos )d with ranges k = (0, ), cos =
(1, 1), = (0, 2).
3. d integration gives a factor of 2.
4. d(cos ) integration of exp(ikr cos ) gives
1
ikr
(e
ikr
e
ikr
)
5. Combine the two exponentials (with k = 0, ) into a
single integral of exp(ikr) running from to .
Thus we have
G(r) =
1
4ir
2
_

e
ikr
kdk
k
2

2
=
1
8ir
2
_
_

e
ikr
dk
k
+
_

e
ikr
dk
k +
_
Again, we encounter a function dened through a singular
integral. The cure is to redene the integral as a generalized
8
function with an innitesimal parameter so that the original
function is obtained in the limit of 0.
The two integrals can be redened by changing i.
G

(r) =
1
8ir
2
_
_

e
ikr
dk
k i
+
_

e
ikr
dk
k + i
_
As r > 0, the integral can only be closed in the uhp (upper
half plane of k) because then the contribution of the semi-
circular contour is zero and the real line integral from
to can be converted into a contour integral. The pole in
the upper half plane for G
+
is at k = + i integral can be
calculated by residue theorem :
G
+
(r) =
1
8ir
2
(2i)e
ir
=
e
ir
4r
.
Similarly, for G

the pole falling in uhp is at + i and

contributes
G

(r) =
e
ir
4r
.
These two solutions G

are called the outgoing and in-

coming wave solutions.
Remark 2 Why is there one solution of the Poisson equation
and two for the Helmholtz? Actually, for any inhomogeneous equa-
tion the general solution is always of the form
general solution = a particular solution of the inhomogeneous
equation
+ a general solution of the homogeneous equation.
We can think of either G
+
or G

as the particular solution,

then the other one is obtained by adding a solution of the homo-
geneous solution to it. For example,
G

= G
+
+ (G

G
+
),
9
and the dierence G

G
+
satises the homogeneous equation.
6
Greens function for the Wave equation
Helmholtz equation simplies the derivation of the Greens
function for the wave equation. The wave equation is
_

1
c
2

2
t
2
_
= source term,
and its Greens function G(t, r) satises
_

1
c
2

2
t
2
_
G =
3
(r)(t).
Assume the Fourier transform of four variables,
G(r, t) =
1
(2)
4
_
d
3
k
_
d

G
(k, ) e
ik.rit
.
The minus sign in the Fourier exponential for time variable t
is conventional.
(It has a deep reason though. That has something to do
with the structure of space-time being Minkowski and not
Euclidean. Or, with the energy of a physical system being
bounded from below while there is no such restriction on the
momenta. Or, with why in quantum mechanics the energy is
i h/t but momenta are i h. But we do not go into that
here because that is irrelevant. If it makes you feel better,
you can substitute for everywhere with absolutely no
change in anything but the notation.)
We solve this problem in two steps. First, we dene the
intermediate Greens function
g(r, ) =
1
(2)
3
_
d
3
k

G
(k, ) e
ik.r
.
10
The Greens function is thus determined by,
G(r, t) =
1
(2)
_
d g(r, ) e
it
,
and the function g satises the Helmholts equation
_

2
+

2
c
2
_
g(r, )
_

2
+
2
_
g(r, ) = (r)
where = c. The tw0 solutions for the Helmholtz equations
are
g

=
e
ir
4r
,
which gives
G

(r, t) =
1
(2)
_
d g

(r, ) e
it
=
1
4r
1
(2)
_
d e
irit
=
1
4r
1
(2)
_
d e
ir/cit
=
1
4r

_
t
r
c
_
7
Greens function for the heat equation
[Not really required for the Electrodynamics course.]
In this case we have a function on both r and t.
The equation to be solved is
G(r, t)
t
= D
2
G(r, t) +
3
(r)(t)
11
We introduce the fourier transform of G(r, t) with re-
spect to r :
G(r, t) =
1
(2)
3
_
d
3
k

G(k, t) e
ik.r
and write the delta function
3
(r) by its usual formula
(r) =
1
(2)
3
_
d
3
ke
ik.r
we nd after substituting in the heat equation and
comparing both sides

G(k, t)
t
= Dk
2

G(k, t) + (t)
which can be solved as follows :
Consider the equation
_
d
dt
+ a
_
F = (t)
We already know that

(t) = (t). Therfore we try a solution of

the type F(t) = (t)f(t) where f is an normal unknown function.
Substituting we get
(t)f(t) + (t)f

(t) + a(t)f(t) = (t)f(0) + (t)f

(t) + a(t)f(t) = (t)

This implies that f(0) = 1 and f

(t) = af(t) for t > 0. Therefore

the solution is
F(t) = (t)e
at
Coming back :

G(k, t) = (t) exp[Dk

2
t]
therefore
G(r, t) =
(t)
(2)
3
_
d
3
k exp[Dk
2
t] e
ik.r
12
We can perform the three dimensional integral on k
by noting that there are actually three independent
gaussian integrals
_
d
3
k exp[Dk
2
t] e
ik.r
=
_
dk
x
exp[Dk
2
x
t + ik
x
x]
_
dk
y
exp[Dk
2
y
t + ik
y
y]

_
dk
z
exp[Dk
2
z
t + ik
z
z]
The rst integral gives for example
_
dk
x
exp[Dk
2
x
t + ik
x
x] =
_

Dt
exp[x
2
/(4Dt)]
and similarly for the other two factors.
The result is
G(r, t) =
(t)
(2

Dt)
3
exp[r
2
/(4Dt)]
13