MIT 8.

044 Notes

Supplementary Notes for 8.044

©R. L. Jaffe 2015

February 16, 2023

Notes on δ-functions and related topics

©R. L. Jaffe MIT 8.044 Notes 2

1 Introduction
The (Dirac) δ-function is a mathematical representation of the physical concept
of the density associated with a point object.1 Consider charge density, for
example. The charge density ρ(x) associated with a point charge of unit strength
at a point x0 has the property that it vanishes everywhere except when x = x0 .
Its integral over any volume containing the point x0 is one, but its integral over
any volume not containing x0 is zero. We represent this behavior by a symbol
δ(x − x0 ). Summarizing its properties,

δ(x − x0 ) = 0, for x ̸= x0
Z
dx δ(x − x0 ) = 1, if x0 ∈ V
V
Z
dx δ(x − x0 ) = 0, if x0 ∈
/ V. (1)
V

an immediate extension is that for any suitably well-defined function f (x),

Z
dxf (x)δ(x − x0 ) = f (x0 ) (2)
V

as long as the region of integration V includes the point x0 . This follows because
the δ-function vanishes everywhere except at x0 where f takes the value f (x0 ).
The δ-function is not really a function as mathematicians define “function”.
Among other things, any function of the real numbers that is zero everywhere
except at one point must integrate to zero. Instead, the δ-function is an example
of a generalized function, another category of useful mathematical objects, which
have been studied in depth. We are only going to need very basic properties of
δ-functions, which we will introduce informally in these notes. 2
Physicists like to represent δ-functions as the limit of a sequence of sharply
peaked ordinary functions of unit strength of ever narrower profile and ever
greater peak height. It is also useful to think of δ-functions abstractly as a
rule for doing integrals as eq. (2) suggests. In these notes we first explore some
of the general properties of integrals involving δ-functions without introducing
any explicit representation for the δ-function. Then we introduce some of the
commonly used representations of δ-functions. Although the definition, eq. (1),
refers to any number of dimensions, we need only consider δ-functions of a single
variable.
1 The δ-function was introduced into physics by Paul Dirac in the 1920’s in his formulation

of quantum mechanics.
2 You can read about generalized functions in the classic book by M. J. Lighthill, An

introduction to fourier analysis and generalized functions.

©R. L. Jaffe MIT 8.044 Notes 3

2 Properties of integrals involving δ-functions

First we rewrite the defining properties of δ(x − a) for a single real variable x,
δ(x − a) = 0, for x ̸= a
Z x1
dx δ(x − a) = 1, if a ∈ (x0 , x1 )
x0
Z x1
dx δ(x − a) = 0, if a ∈/ (x0 , x1 )
x0
Z x1 
f (a), if a ∈ (x0 , x1 )
dxf (x)δ(x − a) = . (3)
x0 0, if a ∈
/ (x0 , x1 )
We are particularly interested in integrals involving δ-functions of more com-
plicated arguments. The rule for determining these integrals is to treat the δ-
function as an ordinary function with respect to changes of integration variables.
Here are some examples, (In all these examples, we assume that the range of
integration always includes the values of x where the argument of the δ-function
vanishes.)
• δ(cx) where c is a (real) constant:
R
To evaluate dxf (x)δ(cx), proceed as you would with ordinary functions
by defining y = cx and change the integration variable to y,
Z Z
1 1
dxf (x)δ(cx) = dy f (y/c)δ(y) = f (0) , (4)
|c| |c|
and the related result,
Z Z
1 1
dxf (x)δ(cx − a) = dy f (y/c)δ(y − a) = f (a/c). (5)
|c| |c|
You should remind yourself why it is the absolute value of c that appears
in these formulas.
• δ(g(x)), where g(x) is a function that has only isolated, simple zeros at a
set of points {xj , j = 1, 2, ...n}, i.e. g(xj ) = 0 and g ′ (xj ) ̸= 0:
R
The integral dx f (x)δ(g(x)) only gets contributions when x is near each
of the zeros at xj . For x ≈ xj we can approximate the argument of the
δ-function by the Taylor expansion of g(x) in the vicinity of xj ,
δ(g(x)) = δ(g(xj ) + (x − xj )g ′ (xj )) = δ(x − xj )g ′ (xj )) (6)
′
since g(xj ) = 0. Since g (xj ) is a constant, we can evaluate this con-
tribution to the integral using the previous result (5). Combining the
contributions of all the zeros of g(x), we obtain
Z n
X 1
dx f (x)δ(g(x)) = f (xj ) (7)
j=1
|g ′ (xj )|
©R. L. Jaffe MIT 8.044 Notes 4

Example

du p(u)δ(E − mu2 /2).

R
Evaluate f (E) =
p
We identify g(u) = E − mu2 /2. g(u) vanishes for u± = ± 2E/m.
Note that one must be careful to locate all the zeros of the δ-function;
in this case, taking both the
√ positive and negative square root. Since
g ′ (u) = −mu, |g ′ (u± )| = 2mE. Because of the absolute value, this
factor is the same at both zeros of g(u). Then substituting into eq. (7)
we obtain f (E),
1  p p 
f (E) = √ p( 2E/m) + p(− 2E/m)
2mE

• The indefinite integral of a δ-function: The properties of the δ-function

summarized in eq. (3) allow us to evaluate the indefinite integral of the
δ-function, Z x 
0 for x < 0
dx′ δ(x′ ) = (8)
−∞ 1 for x > 0
As it happens, this generalized function had already been defined in the
study of time dependent electric circuits. It is known as the step-function
or the(Heaviside) θ-function,3

0 for x < 0
θ(x) =
1 for x > 0

Note that formally, δ(x) = dθ/dx.

The boxed results are important in defining functions of a random variable.

3 Representations of the δ-function

It is often convenient to represent a δ-function as the limit of a sequence of
integrals over ordinary functions. One of the simplest examples in the Gaussian
representation:
1 2
Gaussian representation : δε (x) = √ e−x /2ε (9)
2πε
For any√ finite ε, δε (x) is an ordinary function with unit area and width
of
R order ε. In the limit ε → 0, δε → δ(x). For any finite ε, the integral
dx f (x)δε (x) can be evaluated using ordinary rules of calculus. When ε → 0,
the result is the same as would be obtained using the rules of eq. (3).
3 Named after Oliver Heaviside, who introduced the step-function in the study of signal

processing.
©R. L. Jaffe MIT 8.044 Notes 5

Other representations include:

1 ε
Lorentzian reprentation : δε (x) =
π x2 + ε2
1
Exponential representation : δε (x) = e−|x|/ε
2ε
1
for |x| < ε
“Box” representation : δε (x) = 2ε
0 for |x| > ε
Z
1
Fourier representation : δε (x) = dλeiλx−ε|λ| (10)
2π
The Lorentzian and Fourier representations need comments. First, they are
related. If you perform the integral over λ in the last line of eq. (10) you will
obtain the Lorentzian representation. The Fourier representation appears the
4
theory of Fourier transforms, where it embodies the inverse theorem.
R The
form of the Lorentzian representation suggest that the integral dx f (x)δϵ (x)
is only defined for functions that grow no more rapidly than |x|1−δ as x → ±∞
(otherwise the integral would diverge). This poses a constraint on the Lorentzian
representation of the δ-function, but it is not a constraint on the δ-function
itself, which is defined for any smooth function, no matter how quickly it grows
at infinity.

4 δ-functions and functions of a random variable

Suppose you are given probability distribution p(x) = dP/dx. As usual dP =
p(x)dx gives the probability that the random variable x takes a value between
x and x + dx. Suppose that y = g(x) is a function of the random variable. We
would like to know the probability distribution for y, i.e. dP/dy.
At first sight this looks like a simple change of variable accomplished by the
chain rule of differential calculus:
 −1
dP dP dx dP dy p(x)
= = = ′ .
dy dx dy dx dx g (x) x=g−1 (y)

This is almost, but not quite correct. It is missing an absolute value sign on
g ′ (x) and it failed to pick up the possibility that there may be several values of
x corresponding to a single value of y.
Both of these problems are cured, and the correct result is obtained if we use
δ-functions to define the probability distributions. Here are the steps to obtain
the correct result:
1. Write the identity
Z
dP
p(x) = = dx′ δ(x − x′ )p(x′ ) ,
dx
4 Although it is not part of 8.044 — the Fourier transform of a function f (x) is defined by
1
dxeikx f (x) and the inverse theorem says f (x) = dke−ikx g(k).
R R
g(k) = 2π
©R. L. Jaffe MIT 8.044 Notes 6

where the integration goes over the whole range of values allowed for x.
We can interpret this as an integral overall probabilities subject to the
constraint that the variable x′ takes on the value x.
2. Then define Z
dP
= dx′ δ(y − g(x′ ))p(x′ ) . (11)
dy
This equation has the interpretation of integrating over all probabilities
subject to the constraint that the function g(x′ ) takes on the value y.
3. Evaluate eq. (11) by the usual rules for δ-functions. First find all the zeros
of the argument of δ,

g(x′j ) = y → x′ = xj (y) for j = 1, ...n .

and then evaluate the integral using eq. (7),

n
dP X p(xj )
= , (12)
dy j=1
|g ′ (xj )| xj (y)

which is the correct result including both the absolute value and the sum over
possible multiple values of x that correspond to a given y.5
An example will illustrate the method.

Figure 1: Figure for example on following page

5 It agrees with the “find the cumulant and then differentiate” approach described in Prof.

Greytak’s notes.
©R. L. Jaffe MIT 8.044 Notes 7

Example

As shown in the figure on the previous page, a particle moves at constant

speed on a unit circle in the (x, y) plane, centered on the y-axis above the
origin. If its position is sampled randomly, it has equal probability to be
found at any angle θ with respect to the center of the circle.
dP 1
= p(θ) = for 0 < θ < 2π.
dθ 2π
We would like to find dP/dx, the probability to find the particle with a
given value of x, where x = cos θ.
• Step 1: Define
Z Z 2π
dP ′ ′ ′ 1
= dθ p(θ )δ(x − cos θ ) = dθ′ δ(x − cos θ′ ) .
dx 0 2π

• Step 2: Find zeros of x − cos θ. There are no solutions for |x| > 1,
so dP/dx = 0 for |x| > 1. For |x| < 1, it is clear from the figure
that there are two solutions for θ for every value of x: θ1 = arccos(x)
and θ2 = 2π − arccos(x). (Remember that the range of the arc-cosine
function usually taken to be 0 < θ < π.)
• Step 3: Compute |g ′ (xj )|.

d p
| (x − cos θ)| = | sin θj | = 1 − x2 ,
dθ θj

where we used some trigonometry.

• Step 4: Evaluate the integral using eq. (7). The two values of θ cor-
responding to a given x make equal contributions,
dP 1 1 1
=2× √ = √ for |x| < 1 .
dx 2π 1 − x2 π 1 − x2
Notice that although dP/dx diverges as x → ±1, its integral is finite
(and is equal to one as it must be).