GREENS FUNCTIONS AND FUNCTION SPACES

DEVANG BAJPAI

Abstract. In this note we provide the General Introduction to Green’s Func-

tion, that is, most important concept in both physics and mathematics. I also
try to explain further the statement above that the Green’s function is the
inverse of the associated operator. This discussion is somewhat abstract, but
it is actually useful to work at this level of abstraction, not only for Elec-
trodynamics or quantum mechanics but also for the general study of systems
governed by partial differential equations. I hope that the discussion here will
make the abstractions introduced in the course more clear to you.

Contents
1. Introduction to Green’s Function 1
2. Poisson’s Equation 2
3. Harmonic Oscillator Equation 2
4. Harmonic Oscillator Equation- 2nd Method 3
5. Function Space 4
6. Quadratic forms and Gaussian correlations 5

1. Introduction to Green’s Function

Consider a linear system for a set of fields ϕ(x) governed by the differential
equation
(1.1) Oϕ(x) = 0,
where O is a linear differential operator. We might also consider the inhomogeneous
version of this equation,
(1.2) Oϕ(x) = ρ(x),
To analyze these equations, it is very useful to consider the Green’s function of O,
defined as the solution to the differential equation
(1.3) OG(x, y) = δ(x − y)
where δ(x) is the Dirac delta function. The Green’s function gives the behavior of
the system, measured at x, due to a perturbation placed at the point y. Thus, this
function encodes the complete mechanical response of the linear system. Since the
system is linear, we can superpose the responses to perturbations placed at different
points to evaluate any general mechanical motion of the system.

Date: September 29, 2024.

Lectures organized by Theoretical Nexus under Bose Lecture Series.
1
2 DEVANG BAJPAI

In particular, the solution to eq. 1.2 is

Z
(1.4) ϕ(x) = dyG(x, y)ρ(y)

To show this, just act the linear operator O on both sides of this equation. You
will find
Z
(1.5) Oϕ(x) = dyδ(x − y)ρ(y) = ρ(x)

2. Poisson’s Equation
Let me give you some simple examples: First, consider the Laplace equation in
3-dimensional electrostatics. We would like to find the electrostatic potential φ(x)
that results from a charge distribution ρ(x). This is the solution to the equation
(2.1) (−∇2 )φ(x) = ρ(x)
Let G(x, y) be the solution to the associated Green’s function equation
(2.2) (−∇2x )G(x, y) = δ(x − y)
Here and below, the subscript x means that the derivatives in the operator act on
x. You know that the solution to this equation in unbounded space is the Coulomb
potential
1 1
(2.3) G(x, y) =
4π |x − y|
Then the solution to the inhomogeneous equation 2.1 is
Z
(2.4) φ(x) = d3 yG(x, y)ρ(y)

or, more explicitly

Z
1 1
(2.5) φ(x) = d3 y ρ(y)
4π |x − y|
In complete generality, if we can solve for the Green’s function with the given
boundary conditions, the solution 2.4 will solve the inhomogeneous equation 2.1
and will automatically have the correct boundary conditions.

3. Harmonic Oscillator Equation

Here is another example that illustrates some different features. Consider a
damped harmonic oscillator obeying the equation
d2
 
2 d
(3.1) m 2 + mΩ + mγ x(t) = 0
dt dt
The Green’s function for this equation is the solution to the equation
d2
 
2 d
(3.2) m 2 + mΩ + mγ G(t, t0 ) = δ(t − t0 )
dt dt
However, probably I should not say “the” Green’s function, because the form of
the Green’s function will depend on the boundary conditions imposed in time. For
this problem, the corresponding inhomogeneous equation is
d2
 
2 d
(3.3) m 2 + mΩ + mγ x(t) = F (t)
dt dt
 GREENS FUNCTIONS AND FUNCTION SPACES 3

where F (t) is a force applied as a function of time. Physically, we would always

like the response to occur after the force is applied. This is simple causality. We
can implement the requirement of causality by choosing the Green’s function in 3.2
to satisfy
(3.4) G(t, t0 ) = 0 for all t < t0
For γ = 0 consider the function,

0 t < t0 1
(3.5) G(t, t0 ) = 1 = sin Ω(t − t0 )θ(t − t0 )
mΩ sin Ω(t − t 0 ) t > t0 mΩ
where θ(t) equals 1 for t > 0 and 0 for t < 0. It is easy to see that this equation
satisfies 3.4 and satisfies 3.2 for any t ̸= 0. Just near t0 , we need one more piece of
analysis. Integrate 3.2 over a small region around t = t0 :
Z t0 +ϵ   Z t0 +ϵ
d2
(3.6) dt m 2 + . . . = dtδ(t − t0 )
t0 −ϵ dt t0 −ϵ

Then
d d d
(3.7) m G(t0 + ϵ, t0 ) − m G(t0 − ϵ, t0 ) = m G(t0 + ϵ, t0 ) − 0 = 1
dt dt dt
which is satisfied by 3.5. Then the general solution to the homogeneous problem
taking into account the requirement of causality is
Z ∞
F
(3.8) x(t) = dt0 sin Ω(t − t0 )θ(t − t0 )
−∞ mΩ
Z t
F
(3.9) = dt0 sin Ω(t − t0 )
−∞ mΩ
This solution is the superposition of the solutions generated from small increments
of force F (t0 ) at all times earlier than t. The function 3.5 is called the retarded
Green’s function. The opposite case, in which the response occurs entirely before
the perturbation, is called the advanced Green’s function.

4. Harmonic Oscillator Equation- 2nd Method

It is not so hard to construct the Green’s function and the general solution for any
positive value of γ by straightforwardly solving the equation 3.2. However, I would
like to use a different method which illustrates yet more features of the Green’s
function. Here I will construct the Green’s function by Fourier transformation,
Z Z
dω −ιωt
(4.1) x(t) = e x
e(t) x
e(t) = dωeιωt x(t)
2π
By translation invariance, G(t, t0 ) = G(t − t0 ), so it suffices to find the Green’s
function for t = 0. For t0 = 0, the right-hand side of 3.2 is δ(t). Note that
Z
(4.2) dteιωt δ(t) = 1

so the Fourier transform of 3.2 for t0 = 0 is

(4.3) m(−ω 2 + Ω2 − ιγω) = 1
4 DEVANG BAJPAI

and the solution to the equation 3.2 is found by solving this and inverting the
Fourier transform
−1/m
Z
dω −ιωt
(4.4) G(t) = e
2π (ω − Ω2 + ιγω)
2

This is a tricky but very compelling integral. It is useful to think of ω as a complex

variable.

5. Function Space
Now I would like to introduce the other property of the Green’s function, that
it is the inverse of its defining operator on the space of functions. By “the space
of functions”, I mean an infinite dimensional space that is controlled in a well-
defined mathematical sense. To do 1-particle quantum mechanics, we restrict the
Schrödinger wavefunction ψ(x) to be a square-integrable function of x. Here I will
assume that the function space is properly defined to be a Hilbert space on which
appropriate linear operators can act.
In quantum mechanics, we write transformation of wavefunctions by linear op-
erators in the form
(5.1) |h⟩ = F |k⟩
Writing this out in terms of the Schrödinger wavefunctions describing the states,
this equation would be
Z
(5.2) h(x) = dyF (x, y)k(y)

I would like you to think of this linear operation as matrix multiplication in the
function space. Think of x and y as indices of the matrix F (x, y) or of the vectors
h(x) and k(y). The matrix multiplication is accomplished by summation over one
index of the matrix. You can make this correspondence precise by approximating
space by a lattice of points or considering h(x) and k(y) to be piecewise constant
functions. Eventually, we will take the limit in which h(x) and k(y) become arbi-
trarily varying (e.g., square-integrable) functions on a continuum.
The composition of two linear operators F and G gives
Z
(5.3) h(x) = dωdyF (x, ω)G(ω, x)k(y)
or
Z
(5.4) (F G)(x, y) = dωF (x, ω)G(ω, x)

This is the equation for the multiplication or composition of two matrices on the
function space.
We can include differential operators into this picture. For example, for the
Laplace operator, let
(5.5) L(x, y) = −∇2 δ(x − y)
Then
Z
(5.6) dyL(x, y)k(y) = −∇2x k(x)

is the linear action of the Laplace operator on the function (or vector) k(x).
 GREENS FUNCTIONS AND FUNCTION SPACES 5

The identity operator on this space is

(5.7) I(x, y) = δ(x − y)
To see that this is true, consider that I|k⟩ is represented by
Z Z
(5.8) dyI(x, y)k(y) = dyδ(x − y)k(y)

so, finally
(5.9) I|k⟩ = |k⟩
From this point of view, we can see that the equation
(5.10) −∇2 G(x, y) = δ(x − y)
takes the form in the the function space as
Z
(5.11) dyL(x, ω)G(ω, x) = I(x, y)

that is
(5.12) LG = I
Thus, the Green’s function of the Laplace operator is indeed the inverse of the
Laplace operator as a linear operator on the function space.

6. Quadratic forms and Gaussian correlations

A quadratic form involving the Laplace operator has the form
Z Z
2

(6.1) dx m(x)(−∇ )m(x) = ⟨m(x)|L|m(x)⟩ = dxdy {m(x)(L(x, y)m(y)}

I hope that it is clear in this notation that this equation is the direct general-
ization to an infinite-dimensional function space of the quadratic form in a finite
dimensional space
X
(6.2) ma Mab mb
a,b

Let’s explore this correspondence a little more. I would like to compute the Gauss-
ian integral
Z  
a 1
(6.3) I = dm exp − ma Mab mb
2
where the sum over repeated indices now understood. We can solve this integral
by going to a basis of eigenvectors of the matrix M . These are given by
X
(6.4) Mab ξbj = λj ξbj
b
We can represent the ma in this basis as a linear combination of normalized eigen-
vectors,
(6.5) ma = Cj ξaj
The quadratic form becomes
X X
(6.6) ma Mab mb = λj c2j
a,b j
6 DEVANG BAJPAI

and we can carry out the integral over all ma by integrating over all values of cj .
The result is
Y  2π 1/2
(6.7) I=
j
λj

The expectation value ⟨ma mb ⟩ can be evaluated similarly. First

Z  
1 1
(6.8) dcj ck cℓ exp − λj c2j = δjk
2 λj
Thus,
Z 
1

1 Y  2π 1/2
(6.9) dma mc md exp − ma Mab mb = ξcj ξdk · δjk
2 λj j
λj
 1/2
1 Y 2π
(6.10) = ξcj ξdj
λj j
λj

We can recognize the first term as a representation of the inverse matrix to the
matrix Mab . Dividing by 6.7,
(6.11) ⟨mc md ⟩ = ⟨M −1 ⟩cd
By analogy, we can evaluate the integral of
 Z 
(6.12) exp − dx{m(x)(−∇2 )m(x)}

over the whole space of functions m(x). We have seen that the expression in the
exponent is
Z
(6.13) dxdy {m(x)(L(x, y)m(y)}

The eigenvalues of the linear operator, in this case, the Laplacian, satisfy
Z
(6.14) −∇2 ξ j (x) = dyL(x, y)ξ j (y) = λj ξ j (x)

We can expand any function m(x) in a basis of normalized eigenfunctions

X
(6.15) m(x) = cj ξ j (x)
j

Then the quadratric form becomes

Z X
(6.16) ⟨m|L|m⟩ = dxdy {m(x)(L(x, y)m(y)} = λj c2j
j

where now the sum over j runs over an infinite number of eigenfunctions with in-
creasing eigenvalues. Typically, though, we would regularize this system by ignoring
eigenvalues with λj > Λ, giving a finite-dimensional problem, and then, at the end
of the argument, taking the limit Λ → ∞. In the regulated case, for any finite
number of eigenvalues, the calculation of the correlation function would proceed
just as in the finite-dimension problem with the quadratic form Mab . We then find
X 1
(6.17) ⟨m(x)m(y)⟩ = ξ j (x) ξ j (y)
j
λj
 GREENS FUNCTIONS AND FUNCTION SPACES 7

The differential equation for the Green’s function is

Z
(6.18) −∇2x G(x, y) = dωL(x, ω)G(ω, y) = δ(x − y)

Expand G(x, y) in a basis of the eigenfunctions of the operator L(x, y)

X
(6.19) G(x, y) = ξ j (x)g j (y)
j

and use the completeness relation

X
(6.20) δ(x − y) = ξ j (x)ξ j (y)
Then the equation 6.18 becomes
(6.21) λj ξ j (x)g j (y) = ξ j (x)ξ j (y)
and so,
X 1 j
(6.22) G(x, y) = ξ j (x) ξ (y)
j
λj
In this derivation, actually, I did not use any property of the Laplace operator
other than the fact that it is a Hermitian operator. So, actually, 6.22 holds for any
Hermitian operator L(x, y).
This analysis suggests another definition. For F a linear operator, we can define
the functional determinant of F as
Y
(6.23) det F = λj
j

where λj are the eigenvalues of F . Typically det F is infinite, but it can be regulated,
and that regulated function is often useful in quantum-mechanical problems. The
idea of the Green’s function as the inverse of a differential operator comes up often
in quantum mechanics, particularly in scattering theory. There is it common to
represent the solution of equations
 2 
P
(6.24) + V |ψ⟩ = |ϕ⟩
2m
as
 2 −1
(6.25) P /2m + V |ϕ⟩
It doesn’t matter that the computation of the inverse Schrödinger operator involves
the solution of a complicated partial differential equation. This notation still clari-
fies one’s thinking.
Department of Physics, CSJM University, kanpur, 208024 UP, India.
Email address: [email protected]
URL: https://sites.google.com/view/devangbajpai/home