January 26, 2018

Introduction to Green’s Functions: Lecture notes1

Edwin Langmann
Mathematical Physics, KTH Physics, AlbaNova, SE-106 91 Stockholm, Sweden

Abstract
In the present notes I try to give a better conceptual and intuitive under-
standing of what Green’s functions are. As I hope to convey, the concept of
Green’s functions is very close to physical intuition, and you know already many
important examples without (perhaps) being aware of it.

Aims (what I hope you will get out of these notes):

(i) know a few important examples of Green’s functions,
(ii) know if a given problem can be solved by Green’s functions,
(iii) write down the defining equations of a Green’s functions for such problems,
(iv) know how to use Green’s functions to solve certain problems.
(v) know how Green’s functions are related to Fourier’s method

WARNING: Beware of typos: I typed this in quickly. If you find mistakes please
let me know by email.

Prologue

Green’s functions provide a powerful tool to solve linear problems consisting of a

differential equation (partial or ordinary, with, possibly, an inhomogeneous term) and
enough initial- and/or boundary conditions (also possibly inhomogeneous) so that this
problem has a unique solution. The Green’s function is defined by a similar problem
where all initial- and/or boundary conditions are homogeneous and the inhomogeneous
term in the differential equation is a delta function. If one knows the Green’s function
of a problem one can write down its solution in closed form as linear combinations
of integrals involving the Green’s function and the functions appearing in the inhomo-
geneities. Green’s functions can often be found in an explicit way, and in these cases
it is very efficient to solve the problem in this way.
1
I thank Andreas Minne for helpful feedback.

1
 To give a specific example: Consider the problem to find the function u(x, t), x ∈ Ω
(= subset of RD , D = 1, 2, 3, with boundary ∂Ω) and t ≥ 0, satisfying the PDE (=
heat equation)
D∆u − ut = −h (P DE),
boundary condition
u|∂Ω = α (RV ),
and initial condition
u|t=0 = u0 (IC)
for given functions h = h(r, t), α = α(r, t) (which is defined for r ∈ ∂Ω), and u0 =
u0 (r). Then the Green’s function G is the solution of the similar problem

D∆G − Gt = −δr0 ,t0 (P DE 0 ),

G|∂Ω = 0 (RV 0 ),
G|t=0 = 0 (IC 0 )
where now all but the ODE are homogeneous, and δr0 ,t0 is the delta function localized
at the spacetime point r = r0 , t = t0 , i.e., δr0 ,t0 (r, t) = δ D (r − r0 )δ(t − t0 ). Note that the
Green’s function depends on twice as many variables as u: G = G(r, t; r0 , t0 ) (since it
depends on where the delta function is localized), and thus we should write the problem
determining G in more detail as follows,

D∆r G(r, t; r0 , t0 ) − Gt (r, t; r0 , t0 ) = −δ D (r − r0 )δ(t − t0 )

G(r, t; r0 , t0 )|r∈∂Ω = 0
G(r, 0; r0 , t0 ) = 0

where the ∆r means that the differentiation acts on the variable r. As shown in the
course book, given G we can write the solution of our problem above as follows,
Z t Z
u(r, t) = dt0 dD r0 G(r, t; r0 , t0 )h(r0 , t0 ) +
Z Z0 Ω

dD r0 G(r, t; r0 , 0)u0 (r0 ) − dn−1 S [n · ∇r0 G(r, t; r0 , t0 )]α(r0 )

Ω ∂Ω

where each term on the r.h.s. accounts for one inhomogeneity in our original problem
(the last integral is over the boundary of Ω, and the normal derivative n · ∇r0 only
acts on G; for n = 1 the boundary ∂Ω only consists of two points and the last integral
reduced to a sum of endpoints).
Below I give a more detailed discussion of various examples of Green’s functions
which you probably already know from other courses. I try to explain things in a way
that the generalization of the method to other cases should be obvious. A systematic
and complementary discussion can be found in our course book.

2
Examples you already know
I expect that most of what I discuss in the examples below is repetition for you.
However, it still should be worthwhile to go through these arguments in all detail since
I discuss things in a way which can be immediately adapted to other cases.
Example 1: I first recall that the Coulomb potential is an important example
1
of a Green’s function: as you know, the Coulomb potential 4π|r| corresponds to the
3
electric potential in three dimensional space R which is generated by a point charge
sitting in the origin r = 0. I now recall a mathematical characterization of the Coulomb
potential: Mathematically, this point charge can be described by the charge distribution

ρ(r) = δ 3 (r) = δ(x)δ(y)δ(z) (1)

where r = (x, y, z) ∈ R3 . Indeed, by definition of the delta function, δ 3 (r) = 0 for

r 6= 0 and δ(0) = +∞ such that the total charge equals 1,
Z
d3 r δ 3 (r) = 1.
R3

We know from electrostatics that the electrical potential V (r) generated by a charge
distribution ρ(r) obeys the Poisson equation,2

−∆V (r) = ρ(r) (2)

1
where ∆V := Vxx +Vyy +Vzz . We thus conclude that the Coulomb potential V (r) = 4π|r|
is a solution of −∆V (r) = δ 3 (r). Obviously, if we put the point charge not in the origin
but in another point r0 , then the Coulomb potential is V (r) = 4π|r−r 1
0 | , and it obeys

−∆V (r) = δ (r − r ). Now the potential depends on two arguments r and r0 , and to
3 0

indicate this we write V (r) = G(r, r0 ), i.e.,

1
G(r, r0 ) = ; (3)
4π|r − r0 |
we use the symbol G since this is an example of a Green’s function: the Coulomb
potential G(r, r0 ) above is the Green’s function of the Poisson equation (2) in R3 . The
equation determining this Green’s function is obtained from the Poisson equation in
(2) by choosing as inhomogeneous term a delta-function localized at an arbitrary point
r0 ,
−∆r G(r, r0 ) = δ 3 (r − r0 ); (4)
the subscript of the Laplacian is to indicate that the differentiations are to act on the
r-variable.
2
Strictly speaking we also should impose the boundary condition V (r) → 0 for |r| → ∞, but we
will ignore this in our discussion for simplicity.

3
 I now explain what this Green’s function is good for: In general we are interested
in the Poisson equation (2) for an arbitrary charge distribution ρ(r). For point charges
Qj sitting at the points rj , j = 1, 2, . . . , N ≤ ∞, this charge distribution is
N
X
ρ(r) = Qj δ 3 (r − rj ). (5)
j=1

Since the potential generated by the point change δ 3 (r − rj ) is G(r, rj ) and the Poisson
equation is linear, we can use the superposition principle to conclude that the potential
generated by the charge distribution in (5) is
N N
X X Qj
V (r) = Qj G(r, rj ) = . (6)
j=1 j=1
4π|r − rj |

Using the defining properties of the delta-function we can write this as

ρ(r0 )
Z Z
3 0 0 0
V (r) = d r G(r, r )ρ(r ) = d3 r0 . (7)
R3 R3 4π|r − r0 |

The latter equation holds true not only for charge distributions of the form as in (5) but
in general: V in (7) is the solution of (2) for (essentially) arbitrary charge distributions
ρ. To see this we write Z
ρ(r) = d3 r0 δ 3 (r − r0 )ρ(r0 )
R3

using a defining property of the delta function (recalling that an integral is just the
limit of a sum, this can be obtained as a limit from (5)). In this latter equation we
represent an arbitrary charge distribution as a linear superposition of point charges.
Since G(r, r0 ) is the potential generated by the point change δ 3 (r−r0 ), we can use again
the superposition principle and conclude that the potential generated by ρ(r) is as in
(7).
We can summarize the Green’ s function method to solve the problem in (2)
as follows: We first consider the simpler problem where ρ(r) is replaced by a delta-
functions δr0 (r) := δ 3 (r − r0 ) localized at the point r = r0 . The solution ur0 of this latter
problem is the Green’s function: G(r, r0 ) = ur0 (r). We then can write the solution (2)
in closed formal as an integral as in (7).
The advantage of the method is that it is often quite easy to find the Green’s
function of a given problem. Moreover, there are many different problems which have
the same Green’s functions.
As we will see, similar statements holds true for many linear differential equation
with suitable boundary and/or initial conditions.

4
Example 2: As a second example I consider a pendulum in the earth gravitational
field driven by an external time dependent force F (t). If the deviation y(t) from the
equilibrium position remains small we can model this system by the harmonic oscillator
equation
ÿ(t) + ω 2 y(t) = F (t) ∀t > 0; (8)
ẏ(t) = dy(t)/dt etc., and ω > 0 For simplicity we first consider the case with homoge-
neous initial conditions:
y(0) = ẏ(0) = 0; (9)
the dot means differentiation with respect to time t. I assume you know other methods
to solve this problem (e.g. Laplace transformation), but I now want to illustrate how
to solve it using a Green’s function. Similarly as in our first example above we define
the Green’s function as the solution yt0 (t) = G(t, t0 ) of this problem where the general
inhomogeneous term F (t) is replaced by a delta function δt0 (t) = δ(t − t0 ) localized at
t = t0 , i.e.,
Gtt (t, t0 ) + ω 2 G(t, t0 ) = δ(t − t0 ) ∀t, t0 > 0 (10)
together with the initial condition

G(0, t0 ) = Gt (0, t0 ) = 0. (11)

It is then easy to see that we can write the solution of our problem in (8) as an integral
as follows, Z ∞
y(t) = dt0 G(t, t0 )F (t0 ), (12)
0

similarly as in Example 1. Indeed, it is easy to see that the initial conditions in (11)
imply (9), and
Z ∞
2 2
(∂t + ω )y(t) = dt0 (∂t2 + ω 2 )G(t, t0 ) F (t0 ) = F (t)
0 | {z }
=δ(t−t0 ) due to (10)

proves (8); in the first identity we interchanged integration and differentiation, and in
the second we inserted (10) and used the defining property of the delta-function.
Below we first discuss the physical interpretation of the Green’s function. We then
present a simple method to compute it: we will find that
1
G(t, t0 ) = Θ(t − t0 ) sin(ω(t − t0 )) (13)
ω
where Θ is the Heaviside function. We finally show that the Green’s function not only
can be used to solve the problem above in (8) and (9) but even the more general one
with inhomogeneous initial conditions.

5
Physical Interpretation of (13):3 As discusses, the Green’s function G(t, t0 ) = yt0 (t)
is the solution of (8) and (9) with F (t) = δ(t − t0 ): the pendulum is in equilibrium
position and with zero velocity at time t = 0. In the time interval 0 < t < t0 it is left
to itself, but at time t = t0 > 0 it receives is hit by a force pulse, after which it is left
to itself again. We thus should expect that the pendulum remains in the equilibrium
position until it is hit: yt0 (t) = 0 for t < t0 . The force pulse, however, sets the pendulum
in motion, and it thus should oscillate freely for t > t0 : yt0 (t) = B sin(ω(t − t1 )) for
some constants B, t1 . Since the force pulse sets the pendulum in motion, right after
t = t0 is still should be in the equilibrium position: yt0 (t) = 0 for t → t0 . This
fixes the constant t1 : t1 = t0 . To find the constant B one can argue that the total
pulse, i.e. the integral of the external force over a tiny time interval including the pulse
time, should be equal R t2 to the abrupt velocity0 change of the pendulum due to the pulse:
0
ẏt0 (t2 ) − ẏt0 (t1 ) = t1 dt F (t), where t1 = t − ε and t2 = t + ε with ε > 0, ε → 0.
We will show below how to derive this condition mathematically, and that it gives
B = 1/ω. We thus have derived (13) by physical arguments: the Heaviside function
accounts for the pendulum being in equilibrium before the pulse, and otherwise it is
the usual motion of a free pendulum where the free constants are fixed by the pulse.
A physical interpretation of (12) is as follows: an arbitrary external force F (t)
acting on the pendulum can be thought of as a linear superposition of force pulses at
different times t = t0 of strength F (t0 ):
Z ∞
F (t) = dt0 δ(t − t0 )F (t0 )
0

(mathematically this latter equation is just one of the defining conditions for the delta-
function, of course). Since G(t, t0 ) is the response of the pendulum to the pulse δ(t − t0 )
and the system is linear, we should expect that the total response, i.e. the actual motion
of the pendulum, is just the corresponding linear superposition of the pulse responses:
this is exactly what (12) says.
It is also interesting that, due to the Heaviside function in (13), G(t, t0 ) = 0 for
0
t > t, and we therefore can write (12) as
Z t
y(t) = dt0 G(t, t0 )F (t0 ), (14)
0

(the upper integration limit is now t), and this has a natural physical interpretation as
causality: the position of the pendulum at any time t can only depend on the force
before that time.4

3
You might want to skip this at first reading.
4
More generally, causality means something like: what is can only depend on the past but not on
the future.

6
 We now derive (13) mathematically.
A method to solve (10) and (11):5
To simplify notation I write G(t, t0 ) = y(t) and suppress the t0 -dependence for now.
Thus we need to solve

ÿ(t) + ω 2 y(t) = δ(t − t0 ) ∀t > 0, y(0) = ẏ(0) = 0.

Since δ(t − t0 ) = 0 for t 6= t0 , y(t) solves the homogeneous oscillator equation for t 6= t0 ,
and we thus conclude
A sin(ω(t − t0 )) for 0 ≤ t < t0

y(t) = (15)
B sin(ω(t − t1 )) for t > t0
for some constants A, B, t0 , t1 to be determined. As discussed, the physical interpre-
tation of this is as follows: the pendulum oscillates freely, but the delta-force changes
this oscillation abruptly at time t = t0 (it gives a “kick”). Since y(0) = ẏ(0) we get
A = 0 and thus y(t) = 0 for t < t0 (the value of t0 is irrelevant now, of course); this just
corresponds to causality, but now we also have a mathematical proof of this important
property. To find the constants B and t1 we now show that the effect of the delta-force
the following conditions,

y(t0 + 0) = 0, ẏ(t0 + 0) = 1 (16)

where y(t0 + 0) = limε↓0 y(t0 + ε) etc.; by ε ↓ 0 we mean ε > 0 and ε → 0. To see this
we integrate our differential equation from t = t0 − ε to t > t0 in the limit ε ↓ 0:
Z t Z t
2
ds (yss (s) + ω y(s)) = ds δ(s − t0 ) = 1,
t0 −0 t0 −0

and since y(t) = ẏ(t) = 0 for t < t0 we get from this,

Z t
2
ẏ(t) = 1 − ω ds y(s) ∀t > 0.
t0

This proves the second condition in (16). To get the first one we integrate once more
and obtain Z t Z s
0
y(t) = (t − t ) + ds dr y(r)
t0 t0
implying the first condition in (16). Thus our solution for t > t0 can be found by
solving
ÿ(t) + ωy(t) = 0 ∀t > t0 , y(t0 ) = 0, ẏ(t0 ) = 1.
5
The method explained here is useful to remember for the quantum mechanics course since it
allows to solve the Schrödinger equation −ψ 00 (x) + V (x)ψ(x) = Eψ(x) for singular potentials V (x) =
gδ(x − x0 ).

7
This is not difficult, and we obtain:
1
y(t) = sin(ω(t − t0 )).
ω
We thus get y(t) = G(t, t0 ) as in (13). This concludes our computation. 
We now consider the more general problem in (8) with inhomogeneous initial con-
ditions,
y(0) = y0 , ẏ(0) = v0 y0 , v0 real, (17)
where y0 and v0 real parameters. From previous courses you might know some method
to derive the following solution of this problem:
Z t
1 1
y(t) = y0 cos(ωt) + v0 sin(ωt) + ds sin(ω(t − s))F (s). (18)
ω 0 ω
It is interesting to note that
Z t
y(t) = y0 Gt (t, 0) + v0 G(t, 0) + ds G(t, s)F (s), (19)
0

i.e., we can write the solution solely in terms of the Green’s function of the problem.
This has an important interpretation: the solution of our problem
ÿ(t) + ω 2 y(t) = F (t), y(0) = y0 , ẏ0 (0) = v0
for t > 0 and inhomogeneous initial conditions is identical with the solution of the
problem
ÿ(t) + ω 2 y(t) = F (t) + v0 δ(t) − y0 δ 0 (t), y(0) = 0, ẏ0 (0) = 0
for all t and y(t) = 0 for t < 0: it is possible to trade the inhomogeneous initial
conditions for inhomogeneous terms in the ODE.
This is no coincidence: non-trivial initial and/or inhomogeneous boundary con-
ditions can always be accounted for by terms involving the Green’s function of the
system. The physical interpretation of this is that, to account for non-trivial initial
conditions, we can start with the pendulum in equilibrium position, and give it an
appropriate pulse at t = 0.

Computation of Green’s functions using Fourier’s method I try explain this

using a simple representative example which is our well-studied model of a string:
Compute the function u = u(x, t), 0 < x < L, t > 0 such that
utt (x, t) − c2 uxx (x, t) = F (x, t)
u(0, t) = u(L, t) = 0
u(x, 0) = α(x), ut (x, 0) = β(x). (20)

8
We discussed how to solve this problem by expanding the solution in the eigenfunctions
defined by the corresponding eigenvalue problem f 00 (x) + λf (x) = 0, f (0) = f (L) = 0:
∞
X π
u(x, t) = an (t) sin(kn x), kn = n , (21)
n=1
L

and deriving and solving an ODE problem for the coefficients an (t). I now show how
to solve this model using the philosophy of Green’s functions.
The Green’s function for this model is defined as the solution u(x, t) = G(x, x0 , t, t0 )
of the equations above for the special case F (x, t) = δ(x − x0 )δ(t − t0 ), where t0 > 0
and 0 < x0 < L, and α(x) = β(x) = 0. We note that
∞
X
0 2
δ(x − x ) = sin(kn x) sin(kn x0 )
n=1
L

(check that!), and thus by the ansatz in (21) above we get

X X2
sin(kn x)[a00n (t) + (ckn )2 an (t)] = sin(kn x) sin(kn x0 )δ(t − t0 )
n n
L

etc., implying
2
a00n (t) + (ckn )2 an (t) = sin(kn x0 )δ(t − t0 ), an (0) = a0n (0) = 0
L
which has the solution (of you do not remember this you can derive it using the Green’s
function method)
2 1
an (t) = Θ(t − t0 ) sin(kn x0 ) sin(ckn (t − t0 )).
L ckn
We thus get the following formula for G(x, t, x0 , t0 ) = u(x, t):
∞
0 0
X 2 0 1
G(x, t, x , t ) = Θ(t − t ) sin(kn x) sin(kn x0 ) sin(ckn (t − t0 )). (22)
n=1
L ckn

Using that Green’s function we can write the solution of our problem in (20) as follows,
Z L
u(x, t) = G(x, t, x0 , 0)β(x0 )dx + Gt (x, t, x0 , 0)α(x0 )dx
0
Z ∞ Z L
+ dt0 dx G(x, t, x0 , t0 )F (x0 , t0 ), (23)
0 0

9
where the t0 -integral can be restricted to 0 < t0 < t due to causality. It is instructive
to convince oneselves that this answer is the same as the one one gets using Fourier’s
method: I recommend you do this!
The computation above can be immediately generalized to problems where the
interval Ω = [0, L] is replaced by some other bounded region Ω in D = 1, 2, 3 . . .
dimensions, e.g. a disc (D = 2) or a sphere of a cylinder (D = 3) or . . . . In this case
the functions sin(kn x) and kn above are to be replaced by the eigenfunctions defined
by the corresponding Helmholtz equation:

−∆un (x) = kn2 un (x) in Ω, plus boundary conditions,

RL 0
where
R D 0 now D can stand for several integers, and the integrals 0
dx are replaced by
Ω
d x . It is instructive to write everything out in one other example for Ω. The key
result needed is that X 1
δ D (x − x0 ) = u (x)un (x0 )
2 n
n
||u n ||
R
where ||un ||2 = Ω dD x |un (x)|2 — note that this is just a nice way of writing that
it is possible to expand any “nice” function as generalized Fourier series using these
functions un .

Green’s functions: generalities

Consider a problem

Lu = h, B1 (u) = α1 , B2 (u) = α2 , . . . BN (u) = αN (24)

for some function u = u(x), x in some subset Ω of RD for some D (where one of the
variables can be time), L = Lx some linear differential operator acting on the variables
x, and the Bj defining linear initial and/or boundary conditions.
Then the corresponding Green’s function G = G(x, x0 ) is defined as solution of the
following problem,

Lx G(x, x0 ) = δ D (x − x0 ), B1 (G) = B2 (G) = . . . = BN (G) = 0

(if the the Bj involve differentiations they are to act on the variables x!) where δ D (x −
x0 ) is the delta function localized at x0 ∈ Ω. If all initial and/or boundary conditions
are homogeneous: αj = 0 ∀j, the the solution of the problem is
Z
u(x) = dD x G(x, x0 )h(x0 ).
Ω

Otherwise one also has to add integrals involving G and αj , over the ‘boundary regions’
where αj is defined. To find the precise form of these boundary contributions in general

10
is somewhat tricky, but several important examples are derived in the course book (one
way for accounting for an inhomogeneity αj is to move it from the boundary/initial
condition to the differential equation: this is always possible).

11