Greens functions.

Peeter Joot
Dec 19, 2008. Last Revision: Date : 2008/12/2005 : 31 : 11

Contents
1 Motivation. 1

2 Greens functions. 2

3 Solving linear differential equations. 3

3.1 Simplest case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
3.1.1 With a sampling impulse function. . . . . . . . . . . . . 4

1 Motivation.
For electromagnetism the vector potential equation to solve is

J
( A) =
ce0
Imposing a guage condition A on the solutions A, we have

J
2 A =
ce0
Which gives us our four scalar potential equations

!
1 2 2
c2 (t)2
k 2
A0 =
k ( x )
e0
!
1 2 2 Ji
2 2
k 2
Ai =
c (t) k ( x )
ce0

One form of solution, found for example in [Feynman et al.(1963)Feynman,

Leighton, and Sands] for these equations is the retarded time potentials. For

1
example, with = A0 , the retarded solution for at some time t, is expressed
as an integral over all space

1 (x0 , t |x x0 |/c) 0
Z
(x, t) = dV
4e0 |x x0 |
The first time I saw this equation, I was struck by the relativistic nature of
the solution. Only the charges with speed of light separation from the point of
interest has an effect on the field at that point.
Unfortunately understanding the origin of those solutions, as covered in
[Fitzpatrick()], appears to require Greens functions, Fourier transforms and a
whole mess of complex seeming stuff.
Intuition tells me that relativistic treatment of the electrostatics potential
solution (ie: Lorentz boosting the Coulomb statics potential) can be used to
develop the retarded time solutions, and that will be worth a try. However, it
also appears that direct study of the Greens function tools will be worthwhile.
These notes will examine topics related to Greens functions, with the even-
tual goal of building towards these retarded time potential solutions.

2 Greens functions.
The basic idea is given a linear differential operator L, and the dirac delta func-
tion , a Greens function solution to the operator equation is

L G ( x, s) = ( x s)

Using convolution the impulse (delta) function can be used to form arbi-
trary driving functions, and the general solution, by superposition then only
requires an equivalent convolution integration.
Examples of possible differential operators are

d
L= +a
dx
2 2
L=a 2
+b
(x ) (y)2
1 2
L = 2
v2 (t)2
L = 2

The last of which is the desired operator for the electromagnetism case.

2
3 Solving linear differential equations.
We dont neccessarily need Greens functions and transform theory explicitly
to solve linear equations. Lets review some simple cases to see how to directly
solve some of these.

3.1 Simplest case.

The simplest case is a homogeneous linear first order equation of one variable.

f0 + af = 0

The usual exponential charectaristic solution method solves this one

f = erx
=
(r + a)erx = 0

So a homogeneous solution is

f = eax

The next level of complexity is the inhomogenous case, with a constant

forcing function

f0 + af = b

This is separable and the solution follows from direct integration

df
Z Z
= dx
b af
1 1
ln(b a f ) = x ln( B)
a a

So our complete solution is

1
f = Aeax + b + Beax


a

3
3.1.1 With a sampling impulse function.
Now consider a unit area rectangular spike driving function of width e, and
height 1/e. Will this be a reasonable approximation of a delta function when e
is let approach zero?


1/e if x [e/2, e/2]
b( x ) =
0 otherwise

With a degree of freedom for the homogeneous solution in each of the three
ranges, and one degree of freedom for the inhomogeneous part in the impulse
interval we have the following general solution

a( x +e/2)

A e
  x < e/2
f (x) = Beax + 1a 1e + Ceax x [e/2, e/2]

A+ ea( xe/2) x > e/2

This isnt neccessarily a continuous function, and there is in fact a degree

of freedom too many in the middle term above. Suppose that one wanted this
term to take values B , and B+ at the boundaries of the impulse interval, one
is then left with the following set of simulaneous equations



1 1
B = B + C e ae/2 +
ae ae
 
1 1
B+ = B + C eae/2 +
ae ae

As a vector equation this is

   ae/2   
B e 1
ae = ( aeB + C ) ae/2 +
B+ e 1

and we can wedge with (1, 1) to eliminate it, leaving

ae( B B+ ) = ( aeB + C ) (e ae/2 eae/2 )

So we have

1 B B+
B+ C=
ae 2 sinh( ae/2)

and the solution to our equation is therefore

4

a( x +e/2) x < e/2
A e

B B+
f (x) = 2 sinh( ae/2)
eax + 1
ae x [e/2, e/2]
A+ ea( xe/2)

x > e/2

However, with the middle term being a function of only the difference be-
tween the endpoints, means that we must have an additional constraint on the
endpoints. In general one could make this match the a desired value at only
one of the endpoints, and then the other is then determined. How about pick-
ing the middle term so that its value at x = 0 is the midpoint between A and
A+ . If one writes f = eax + 1/ae we have

1 1
+ = ( A + A+ )
ae 2
=
1 1
f = ( A + A+ ) eax + 1 eax


2 ae
Now, this was for only the impulse interval. If however, we have A = A =
A+ , then writing u( x ) for the unit step function, we have for the complete
interval the following general solution

1
f = Aeax + 1 eax (u(e/2 + x ) + u(e/2 x ))


2ae
This finally has a desirable form, with the sum of a regular old homogenous
solution, plus a specific solution. The specific solution here also has the role of
the Greens function for our delta like impulse function. Doing a convolution
sum with this impulse function has a sampling like action, but it isnt clear how
to take the limit of the Greens like function as e 0.
Seeing the delta like function here suggests that the proper Greens function
for this operator may in fact be a weighted delta function. Specifically if we had

f 0 + a f = ( x )

Then is the corresponding general solution in terms of homogeneous solu-

tion and a Greens function like so:
1
f = Aeax + 1 eax ( x )


a

How would one even see if this makes sense (ie: how does one take the
derivative of a delta function)? Probably need the convolution before this
would make sense. Lets try this.

5
References
[Feynman et al.(1963)Feynman, Leighton, and Sands] R.P. Feynman, R.B.
Leighton, and M.L. Sands. feynman lectures on physics.[Lectures on
physics]. 1963.

[Fitzpatrick()] Richard Fitzpatrick. Relativity and electromagetism. http:

//farside.ph.utexas.edu/teaching/jk1/relativity.pdf.