Greens Function of the Wave Equation

The Fourier transform technique allows one to obtain Greens functions for a
spatially homogeneous infinite-space linear PDEs on a quite general basis
even if the Greens function is actually a generalized function. Here we apply
this approach to the wave equation.

The wave equation reads (the sound velocity is absorbed in the re-scaled t)

utt = u . (1)

Equation (1) is the second-order differential equation with respect to the

time derivative. Correspondingly, now we have two initial conditions:

u(r, t = 0) = u0 (r) , (2)

ut (r, t = 0) = v0 (r) , (3)

and have to deal with two Greens functions:
Z Z
(0) 0 0 0
u(r, t) = G (r r , t) u0 (r ) dr + G(1) (r r0 , t) v0 (r0 ) dr0 . (4)

Both functions satisfy the equation

Gtt = G , (5)

but with different initial conditions:

(0)
G(0) (r, 0) = (r) , Gt (r, 0) = 0 , (6)
(1)
G(1) (r, 0) = 0 , Gt (r, 0) = (r) , (7)
Looking for the solution of (5) in the form
Z
G(r, t) = g(k, t) eikr dk/(2)d , (8)

we get
g = k 2 g . (9)
That is
g(k, t) = A(k) cos(kt) + B(k) sin(kt) , (10)

1
where the functions A(k) and B(k) are defined by the initial conditions
(6)-(7). Plugging (11) into (6)-(7) and taking into account that the Fourier
transform of the -function is unity:
Z
(r) = eikr dk/(2)d , (11)

we get
(0)
g (0) (k, 0) = 1 , gt (k, 0) = 0 , (12)
(1)
g (1) (k, 0) = 0 , gt (k, 0) = 1 , (13)
and readily find
g (0) (k, t) = cos(kt) , (14)
(1)
g (k, t) = (1/k) sin(kt) . (15)
(1)
Comparing Eqs. (14)-(15) one notices that g (0) (k, t) = gt (k, t), and thus
(1)
G(0) (r, t) = Gt (r, t) . (16)

Hence, it is sufficient to evaluate only G(1) , and then find G(0) by differen-
tiating G(1) with respect to t.
From (15) we obtain
Z
(1) cos(kr) sin(kt) dk
G (r, t) = , (17)
k (2)d
where we took into account the k k symmetry.
Performing the integral (17) essentially depends on the dimension, and
we need to consider separately three different cases: d = 1, 2, 3.

1D case. Z
cos(kx) sin(kt) dk
G(1) (x, t) = . (18)
k 2
[Note that despite the fact that in (17) the symbol k stands for the absolute
value of vector k, there is no contradiction between (17) and (18) because
the integrand of (18) remains the same when k k.]
Using
sin cos = [sin( + ) + sin( )]/2 , (19)
we rewrite our integral as
Z
(1) sin k(t + x) + sin k(t x) dk
G (x, t) = , (20)
k 4

2
and recall that Z
sin ky
dk = sgn(y) , (21)
k
where
1,
y>0,
sgn(y) = 1 , y<0, (22)

0, y=0.
This yields the final answer:
(
1/2 , x [t, t] ,
G(1) (x, t) = [sgn(t + x) + sgn(t x)]/4 = (23)
0, x/ [t, t] .

2D case. In polar coordinates:

k = (k cos , k sin ) , dk = k dk d , (24)

with being the angle between k and r, we have

Z 2 Z
1
G(1) (r, t) = d cos[kr cos ] sin(kt) dk . (25)
(2)2 0 0

First, we integrate over k. Once again we use (19) and see that we need to
perform Z
I(y) = sin(ky) dk , (26)
0
in terms of which we then would have
Z 2
1
G(1) (r, t) = [I(t r cos ) + I(t + r cos )] d, (27)
8 2 0

or simply Z 2
1
G(1) (r, t) = I(t + r cos ) d , (28)
4 2 0
because of the symmetry of the cosine function: cos( + ) = cos .
However, the integral (26) is divergent and we should introduce a regu-
larization. With an infinitesimally small positive we can write
Z
1
I(y) = Im e(iy)k dk = Re . (29)
0 y + i

3
It is too early here to take the limit of 0: The integral over also
needs a regularization which is easily done by just keeping the term i in
I(y) while doing the integral (28). We thus have
Z 2
(1) 1 d
G (r, t) = 2 Re . (30)
4 0 t + r cos + i
By a standard trick,

z = ei d = idz/z , cos = (z + 1/z)/2 , (31)

this integral is reduced to a contour integral along a unity-radius origin-

centered circle in a complex plane:
Z 2 I
d dz
I2 = = 2i . (32)
0 t + r cos + i rz 2 + 2(t + i)z + r

Doing the complex integral by residues, we get

2
I2 = p . (33)
(t + i)2 r2

Finally, taking the real part of this integral in the limit of 0, we obtain

1 (t r)
G(1) (r, t) = , (34)
2 t2 r2

where (
1, x0,
(x) = (35)
0, x<0.

3D case. In spherical coordinates,

k = (k sin cos , k sin sin , k cos ) , dk = k 2 dk d d(cos ) , (36)

with the z-axis along the r vector, the integrals over and are readily
done, since the integrand is -independent, and the only place where the
-dependence comes from is kr = kr cos . The result is
Z
(1) 1
G (r, t) = 2 sin(kr) sin(kt) dk . (37)
2 r 0

4
Recalling that

sin sin = [cos( ) cos( + )]/2 , (38)

we write it as
Z
(1) 1
G (r, t) = 2 [cos k(r t) cos k(r + t)] dk , (39)
4 r 0

and see that we need to do the integral

Z
I3 (y) = cos(ky) dk . (40)
0

This integral is similar to I(y). It is also divergent and is regularized and

calculated the same way:
Z
i
I3 (y) = Re e(iy)k dk = Re = 2 = (y) . (41)
0 y + i y + 2
We thus have
1
G(1) (r, t) = (t r) . (42)
4r

Constructing the solution

(1)
The function G(0) = Gt turns out to be a generalized function in any
dimensions (note that in 2D the integral with G(0) is divergent). And in 3D
even the function G(1) is a generalized function. So we have to establish the
final form of the solution free of the generalized functions. In principle, it is
sufficient to take care of the function G(1) only, since in view of the relation
(1)
G(0) = Gt we can always write

(1) (t) |v0 i + G

|u(t)i = G (1) (t) |u0 i . (43)
t
That is we act on the function u0 with the same operator G (1) (t) producing
thus some smoothby the nature of the operator G (t)time-dependent
(1)

function, and then differentiate this function with respect to t. As we will

(0) (t) without resorting to
see, it is also possible to express the operator G
the time-differentiation. However, in 2D and 3D this will lead to a spatial
derivative of the function u0 .

5
1D case. Writing Eq. (43) with G(1) of Eq. (23), we have
Z x+t Z x+t

u(x, t) = (1/2) v0 (x0 ) dx0 + (1/2) u0 (x0 ) dx0 . (44)
xt t xt

The differentiating in the second term can be done explicitly, so that finally
we get
Z x+t
u(x, t) = (1/2) [u0 (x + t) + u0 (x t)] + (1/2) v0 (x0 ) dx0 . (45)
xt

3D case. Here it is convenient to introduce a shifted variable for integration,

r1 = r0 r, and to take into account that G(1) (r1 ) G(1) (r1 ):
Z Z
(1)
G (r r0 ) v0 (r0 ) dr0 = G(1) (r1 ) v0 (r1 + r) dr1 . (46)

We see that without loss of generality we may set r = 0, since the solution
at any finite r is obtained by just translating the initial conditions by the
vector r. Writing the integrals with the Greens function (42) in the spherical
coordinates (and omitting the subscript 1), we get
2 Z Z
t
u(r = 0, t) = d d sin v0 (r = t, , ) +
4 0 0
Z 2 Z
t
+ d d sin u0 (r = t, , ) . (47)
t 4 0 0

Differentiating with respect to time in the second term, we get

Z 2 Z
t
u(r = 0, t) = d d sin v0 (r = t, , ) +
4 0 0
Z 2 Z
1 u0
+ d d sin [u0 (r = t, , ) + t (r = t, , )] . (48)
4 0 0 r
The meaning of the angular integrals is the averaging over the solid angle:
Z 2 Z
1
h. . .i = d d sin (. . .) . (49)
4 0 0

Correspondingly, our final result can be written as

u0
u(r = 0, t) = t hv0 i|r=t + hu0 i|r=t + t . (50)
r r=t

6
2D case. In two dimensions the function G(1) (34) is a regular function so
that we can simply write
Z
1 v0 (r0 ) dr0
u(r, t) = p +
2 |rr0 |t t2 |r r0 |2
Z
1 u0 (r0 ) dr0
+ p . (51)
2 t |rr0 |t t2 |r r0 |2

However, if we want to eliminate the time-derivative in the second term by

differentiating under the sign of the integral, we face a problem: The integral
becomes divergent. This means that if we differentiate under the sign of the
integral, we get a generalized function and need to properly process it. The
trick is to replace the time-derivative with a spatial derivative. To this end
it is convenient to write the Greens function in such a way that its self-
similarity is explicitly seen, and then take advantage of the self-similarity
in relating the temporal and spatial derivatives. As is seen, for example,
from the dimensional analysis of the wave equation, a proper dimensionless
variable is = t/r. Correspondingly, we rewrite Eq. (34) in the self-similar
form as
1
G(1) (r, t) = Q() , (52)
2r
where

()
Q() = p 2 , (53)
1
and (
= 1, 1,
() (54)
0, <1.
Now we have
G(1) 1 1
= Q0 () = Q0 () . (55)
t 2r t 2r2
On the other hand,
Q t
= Q0 () = 2 Q0 () . (56)
r r r
That is
r2 Q
Q0 () = , (57)
t r

7
and we have

G(1) 1 Q 1
(t/r)
= = p . (58)
t 2t r 2t r (t/r)2 1

This is a generalized function. To arrive at an ordinary function, we just

need to do the integral by parts. Using the representation (46) and setting
without loss of generality r = 0, in polar coordinates we have
Z Z
(t r)r 2 d
u(r, t) = dr v0
0 t2 r2 0 2
Z 2 Z
d (t/r)
drr u0 p . (59)
0 2t 0 r (t/r)2 1

Doing the integral in the second term by parts,

Z Z
(t/r) (t/r) u0
drr u0 p = dr p r + u0 , (60)
0 r (t/r)2 1 0 (t/r)2 1 r

we arrive at a regular integral. Taking into account that the -functions just
fix the upper limit of integration over r, we finally get
Z t Z 2
dr d r u0 u0
u(r, t) = p v0 + + . (61)
0 (t/r)2 1 0 2 t r t