10 Greens Function
10 Greens Function
10 Greens Function
10 Green’s Functions
A Green’s function is a solution to an inhomogenous differential equation with a “driving
term” that is a delta function (see Section 9.7). It provides a convenient method for solv-
ing more complicated inhomogenous differential equations. In physics, Green’s functions
methods are used to describe a wide range of physical phenomena, such as the response of
mechanical systems to impacts or the emission of sound waves from acoustic sources.
Here, m is the mass of the particle, γ is the damping coefficient, and ω0 is the natural
frequency of the oscillator. The left side of the equation is the same as in the damped
harmonic oscillator equation (see Chapter 4). On the right side, we introduce a time-
dependent driving force f (t), which acts alongside the pre-existing spring and damping
forces. Given an arbitrarily complicated f (t), our goal is to determine x(t).
The function G(t, t0 ), which depends on the two variables t and t0 , is called the Green’s
function. Note that the differential operator on the left side involves only derivatives in t.
The Green’s function describes the motion of a damped harmonic oscillator subjected to
a particular driving force that is a delta function, describing an infinitesimally sharp pulse
centered at t = t0 :
f (t)
= δ(t − t0 ). (3)
m
Here’s the neat thing about G(t, t0 ): once we know it, we can find a specific solution to the
driven harmonic oscillator equation for any f (t). The solution has the form
Z ∞
f (t0 )
x(t) = dt0 G(t, t0 ) . (4)
−∞ m
To show that this is indeed a solution, plug it into the equation of motion:
2 Z ∞ 2 0
d d 2 0 ∂ ∂ 2 0 f (t )
2
+ 2γ + ω0 x(t) = dt 2
+ 2γ + ω 0 G(t, t ) (5)
dt dt −∞ ∂t ∂t m
Z ∞ 0
f (t )
= dt0 δ(t − t0 ) (6)
−∞ m
f (t)
= . (7)
m
Note that we can move the differential operator inside the integral because t and t0 are
independent variables.
86
Y. D. Chong (2020) MH2801: Complex Methods for the Sciences
The Green’s function concept is based on the principle of superposition. The motion of
the oscillator is induced by the driving force, but the value of x(t) at time t does not depend
only on the instantaneous value of f (t) at time t; instead, it depends on the values of f (t0 )
over all past times t0 < t. We can thus decompose f into a superposition of pulses described
by delta functions at different times. Then x(t) is a superposition of the oscillations produced
by the individual pulses.
Here, we have used the sign convention for time-domain Fourier transforms (see Section 9.3).
Applying the Fourier transform to both sides of the Green’s function equation, and making
use of how derivatives behave under Fourier transformation (see Section 9.4), gives
Z ∞
0
0
dt eiωt δ(t − t0 ) = eiωt .
2 2
−ω − 2iγω + ω0 G(ω, t ) = (9)
−∞
0
The differential equation for G(t, t ) has thus been converted into an algebraic equation for
G(ω, t0 ), whose solution is
0
0 eiωt
G(ω, t ) = − 2 . (10)
ω + 2iγω − ω02
Finally, we retrieve the time-domain solution by using the inverse Fourier transform:
Z ∞
dω −iωt
G(t, t0 ) = e G(ω, t0 ) (11)
−∞ 2π
Z ∞ 0
dω e−iω(t−t )
=− 2 2. (12)
−∞ 2π ω + 2iγω − ω0
The denominator of the integral is a quadratic expression, so this can be re-written as:
Z ∞ 0
dω e−iω(t−t )
q
0
G(t, t ) = − where ω± = −iγ ± ω02 − γ 2 . (13)
−∞ 2π (ω − ω+ )(ω − ω− )
This can be evaluated by contour integration. The integrand has two poles, which are
precisely the complex frequencies of the damped harmonic oscillator; both lie in the negative
complex plane. For t < t0 , Jordan’s lemma requires us to close the contour in the upper
half-plane, enclosing neither pole, so the integral is zero. For t > t0 , we must close the
contour in the lower half-plane, enclosing both poles, so the result is
" #
−iω+ (t−t0 ) −iω− (t−t0 )
e e
G(t, t0 ) = iΘ(t − t0 ) + (14)
ω+ − ω− ω− − ω+
hp i
√ 1
sin ω 2 − γ 2 (t − t0 ) , γ < ω0 ,
0
ω02 −γ 2 0
= Θ(t − t0 ) e−γ(t−t ) × hp i (15)
√ 21 2 sinh
γ 2 − ω02 (t − t0 ) , γ > ω0 .
γ −ω0
87
Y. D. Chong (2020) MH2801: Complex Methods for the Sciences
1.0 0.4
0.5 0.3
0.2
0.0
0.1
-0.5
0.0
-1.0
-20 0 20 40 60 -10 0 10 20
On the last line, the expression on the left-hand side represents the difference between the
velocities just after and before the pulse. Evidently, the pulse imparts one unit of velocity
at t = t0 . Looking at the solutions obtained in Section 10.1.2, we can verify that ∂G/∂t = 0
right before the pulse, and ∂G/∂t = 1 right after it.
For t − t0 > 0, the applied force goes back to zero, and the system behaves like the
undriven harmonic oscillator. If the oscillator is under-damped (γ < ω0 ), it undergoes a
decaying oscillation around the origin. If the oscillator is over-damped (γ > ω0 ), it moves
ahead for a distance, then settles exponentially back to the origin.
88
Y. D. Chong (2020) MH2801: Complex Methods for the Sciences
10.1.4 Causality
We have seen that the motion x(t) ought to depend on the driving force f (t0 ) at all past
times t0 < t, but should not depend on the force at future times. Because of the relation
Z ∞
f (t0 )
x(t) = dt0 G(t, t0 ) , (21)
−∞ m
This condition is referred to as causality, because it is equivalent to saying that cause must
precede effect. A Green’s function with this feature is called a causal Green’s function.
For the driven harmonic oscillator, the time-domain Green’s function satisfies a second-
order differential equation, so its general solution must contain two free parameters. The
specific solution that we derived above, Eq. (15), turns out to be the only causal solution.
There are a couple of ways to see why.
The first way is to observe that for t > t0 , the Green’s function satisfies the differential
equation for the undriven harmonic oscillator. But based on the discussion in Section 10.1.3,
the causal Green’s function needs to obey two conditions at t = t0 + 0+ : (i) G = 0, and
(ii) ∂G/∂t = 1. These act as two boundary conditions for the undriven harmonic oscillator
equation, giving rise to the specific solution that we found.
The other way to see that the causal Green’s function is unique is to imagine adding
to our specific solution any solution x1 (t) for the undriven harmonic oscillator. It is easily
verified that the resulting G(t, t0 ) is also a solution to Eq. (2). Since the general solution for
x1 (t) contains two free parameters, we have thus found the general solution for G(t, t0 ). But
the solutions for x1 (t) are all infinite in the t → −∞ limit, except for the trivial solution
x1 (t) = 0. That choice corresponds to the causal Green’s function (15).
where ψ(x, t) is a complex wavefunction and c is the wave speed. Henceforth, to simplify the
equations, we will set c = 1. (You can reverse this simplification by replacing all instances
of t with ct, and ω with ω/c, in the subsequent formulas.)
The wave equation describes how waves propagate after they have already been created.
To describe how the waves are generated in the first place, we must modify the wave equation
by introducing a term on the right-hand side, called a source:
2
∂2
∂
− 2 ψ(x, t) = f (x, t). (24)
∂x2 ∂t
The source term turns the wave equation into an inhomogenous partial differential equation,
similar to the driving force for the driven harmonic oscillator.
89
Y. D. Chong (2020) MH2801: Complex Methods for the Sciences
As can be seen, G is a function of two spatial variables, x and x0 , as well as two temporal
variables t and t0 . It corresponds to the wave generated by a pulse
The differential operator in the Green’s function equation only involves x and t, so we can
regard x0 and t0 as parameters specifying where the pulse is localized in space and time. This
Green’s function ought to depend on the time variables only in the combination t − t0 , as we
saw in our earlier discussion of the harmonic oscillator Green’s function (see Section 10.1.3).
To emphasize this, we have written it as G(x, x0 ; t − t0 ).
The Green’s function describes how a source localized at a space-time point influences
the wavefunction at other positions and times. Once we have found the Green’s function, it
can be used to construct solutions for arbitrary sources:
Z Z ∞
ψ(x, t) = dx0 dt0 G(x, x0 ; t − t0 ) f (x0 , t0 ). (27)
−∞
Just as we can write the time-domain solution to the wave equation in terms of the time-
domain Green’s function, we can do the same for the frequency-domain solution:
Z
Ψ(x, ω) = dx0 G(x, x0 ; ω) F (x0 , ω), (30)
where Z ∞ Z ∞
iωt
Ψ(x, ω) = dt e ψ(x, t), F (x, ω) = dt eiωt f (x, t). (31)
−∞ −∞
90
Y. D. Chong (2020) MH2801: Complex Methods for the Sciences
The Green’s function should correspond to a left-moving wave for x to the left of the source,
and to a right-moving wave for x to the right of the source.
We can guess the form of the Green’s function obeying these boundary conditions:
0
A e−iω(x−x ) , x ≤ x0 ,
0
G(x, x ; ω) = 0 for some A, B ∈ C. (32)
B eiω(x−x ) , x ≥ x0
It is straightforward to verify that this formula for G(x, x0 , ω) satisfies the wave equation in
both the regions x < x0 and x > x0 , as well as satisfying outgoing boundary conditions. To
determine the A and B coefficients, note that G(x, x0 ) should be continuous at x = x0 , so
A = B. Then, integrating the Green’s function equation across x0 gives
x0 + Z x0 +
∂2
Z
0
lim 2
+ ω G(x − x ) = lim δ(x − x0 ) (33)
→0 x0 − ∂x2 →0 x0 −
( )
∂G 0
∂G 0
= lim (x, x ) − (x, x ) = iω(B + A) = 1. (34)
→0 dx x=x0 + ∂x x=x0 −
There is a problem on the last line: the integral runs over the real-ω line, yet the integrand
has a pole at ω = 0, on the real axis, making the integral ill-defined.
To resolve this, we redefine G(x, x0 ; ω) as an integral over a deformed contour Γ:
0 0
eiω[|x−x |−(t−t )]
Z
G(x, x0 ; t − t0 ) ≡ dω . (37)
Γ 4πiω
We will choose the deformed contour in a very specific way, which turns out to be the choice
that satisfies causality. As shown in the left subplot of the figure below, it runs along the
real axis, but skips above the pole at the origin.
The integral can be solved by either closing the contour in the upper half-plane, or in
the lower half-plane. If we close the contour above, then the loop contour does not enclose
the pole, and hence G(x, x0 ; t − t0 ) = 0. According to Jordan’s lemma, we must do this if
the exponent in the integrand obeys
This inequality is satisfied in two cases: either (i) t < t0 (in which case the inequality is
satisfied for all x, x0 because |x − x0 | is strictly non-negative), or (ii) t > t0 but the value of
t − t0 is smaller than |x − x0 |. To understand the physical meaning of these two cases, recall
that G(x, x0 ; t − t0 ) represents the field at position x and time t resulting from a pulse at
the space-time point (x0 , t0 ). Thus, case (i) corresponds to times occurring before the pulse,
91
Y. D. Chong (2020) MH2801: Complex Methods for the Sciences
and case (ii) corresponds to times occurring after the pulse but too far away from the pulse
location for a wave to reach in time.
For the other case, |x − x0 | − (t − t0 ) < 0, the residue theorem gives
G(x, x0 ; t − t0 ) = −1/2. (39)
The space-time diagram below summarizes the above results:
92
Y. D. Chong (2020) MH2801: Complex Methods for the Sciences
10.5 Exercises
1. Find the time-domain Green’s function of the critically-damped harmonic oscillator
(γ = ω0 ).
where h· · · i denotes an average taken over many realizations of the random force and
A is some constant. Using the causal Green’s
function, find the correlation function
hx(t1 ) x(t2 )i and the mean squared deviation [x(t + ∆t) − x(t)]2 .
[solution available]
93