CUST 2023-2024

Maths for Physics 2

Session 5

The Dirac δ-function

Summary

Here we discuss the so-called Dirac δ-function, a mathematical object that is of

fundamental importance in a broad range of domains of science and technology.
We first heuristically construct the δ-function from the orthogonality relation of
the complex exponentials (relation that was derived in the context of the complex
Fourier series). We then give the definition of the δ-function. While the latter is
by construction not a function, we see how it can still be represented by functions
in a certain limit. Finally, we also discuss the integral and the derivatives of the
δ-function.

i
Contents

1 Definition of the Dirac δ-function 1

1.1 Heuristic construction of the Dirac δ-function from the orthogonality
of the complex exponentials . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Mathematical definition of the Dirac δ-function . . . . . . . . . . . . 3
1.3 Some properties of the Dirac δ-function . . . . . . . . . . . . . . . . . 4

2 Representations of the Dirac δ-function 7

2.1 Rectangular representation . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Gaussian representation . . . . . . . . . . . . . . . . . . . . . . . . . 8

3 Integral and derivatives of the Dirac δ-function 11

3.1 Integral of the Dirac δ-function . . . . . . . . . . . . . . . . . . . . . 11
3.2 Derivatives of the Dirac δ-function . . . . . . . . . . . . . . . . . . . 13

ii
Chapter 1

Definition of the Dirac δ-function

Here we introduce this very important mathematical object known as the Dirac
δ-function , or simply δ-function. Roughly speaking, the Dirac δ-function can some-
how be viewed as a generalization of the Kronecker delta to the case of continuous
variables. We can for instance heuristically see this from the orthogonality condi-
tion satisfied by the complex exponentials: this will be the starting point of our
discussion.

1.1 Heuristic construction of the Dirac δ-function

from the orthogonality of the complex exponen-
tials
In the context of Fourier series, we saw that the complex exponentials e2niπt/T are
said to be orthogonal in the sense that they satisfy the relation
 T /2
∗ ′
dt e2niπt/T e2n iπt/T = T δn,n′ , ∀n, n′ ∈ Z , (1.1)
−T /2

∗
where the asterisk denotes complex conjugation. Introducing the quantities

2nπ 2n′ π
kn ≡ and kn′ ≡ , (1.2)
T T

we can thus rewrite (1.1) as

 T /2
dt e−ikn t eikn′ t = T δkn ,kn′ . (1.3)
−T /2

Then we saw that the Fourier transform could be (heuristically) constructed

from the Fourier series by considering the limit as T → ∞ of the Fourier series.
Therefore, it seems legitimate at this point to ask the question: what happens to

1
CHAPTER 1. DERIVATION OF THE DIRAC. . . 2

the orthogonality condition (1.3) if we take this limit T → ∞? Following a similar

line of reasoning as the one used in our construction of the Fourier transform from
the Fourier series, we can see that the discrete quantities kn and kn′ in (1.3) go
to continuous quantities k and k ′ in the limit T → ∞. This hence yields the
′
exponentials e−ikt and eik t of continuous quantities, and we get for the left-hand
side of (1.3) in this limit
 T /2  ∞
−ikn t ikn′ t ′
lim dt e e = dt e−ikt eik t ,
T →∞ −T /2 −∞

so that we get for (1.3) in this limit

 ∞
′
dt ei(k −k)t = lim T δk,k′ . (1.4)
−∞ T →∞

Let’s denote by δ(k ′ − k) the right-hand side of (1.4), that is we define

δ(k ′ − k) ≡ lim T δk,k′ , (1.5)

T →∞

and now let’s ask the question: what is this object that we called δ(k ′ − k)? Let’s
first try to get some feeling about what it looks like. If we just use the definition of
the Kronecker delta, we know that δk,k′ = 1 if k = k ′ , and δk,k′ = 0 if k ̸= k ′ . Of
course, the Kronecker delta is only defined for discrete variables, but here let’s just
naively apply its definition to the continuous variables k and k ′ that we have here.
So if we agree to do this, what we naively get from (1.5) is thus the following:

 Tlim
 T ×1=∞ , if k = k′
→∞
δ(k ′ − k) = . (1.6)
 lim T × 0 = 0
 , if k ̸= k ′
T →∞

Wait, this certainly can not be a proper mathematical object: a function that returns
an infinite value for a specific value of its variables (namely for k ′ − k = 0 here) and
zero otherwise, that is some kind of infinite zero-width spike, this cannot make any
sense!
Well, for sure, δ(k ′ − k) as given by (1.6) cannot define a function, because
a function is by definition an object that maps a number onto another number.
While δ(k ′ − k) could be viewed as mapping any nonzero number k ′ − k onto the
number 0, the problem of course arises from the case where k ′ − k = 0: in this
case, δ(k ′ − k) maps 0 onto ∞, which is not a proper number. But, while it’s not a
function, it nevertheless turns out that the object δ(k ′ − k) can actually be shown
to be a proper mathematical object: it is a so-called generalized function, which
is also called a distribution. The term generalized function indeed indicates how
CHAPTER 1. DERIVATION OF THE DIRAC. . . 3

such an object should be seen: it is an object that extends the notion of a function.
While the theory of generalized functions is a rigorous mathematical theory, to
discuss it is beyond the scope of this course. Only the main feature of a generalized
function will be important for us: roughly speaking, a generalized function has a
well-defined mathematical meaning only when it acts on a function (typically called
a test function in this context) inside an integral. With this in mind, in practice
we’ll mostly manipulate such generalized functions as if they were usual functions.
The mathematical object that we constructed in (1.6) is actually so ubiquitous
in science and technology that it has been given a special name: it is called the Dirac
δ-function 1 . Let’s now give a more proper mathematical definition of this object.

1.2 Mathematical definition of the Dirac δ-function

REMARK: While we conventionally use the name “Dirac δ-function”, we must keep
in mind that this object is not a true function!

The Dirac δ-function, which we typically denote by δ(x), is this mathematical

object that simultaneously satisfies the two following conditions:

i) it is zero for any x ̸= 0;

ii) it is infinite at x = 0 such as to have a finite, unit area under its “curve”.

That is, the mathematical definition of δ(x) is that it must satisfy both

δ(x) = 0 , ∀x ̸= 0 , (1.7)

and
 ∞
dx δ(x) = 1 . (1.8)
−∞

We can already note that, because δ(x) = 0 for any x ̸= 0, the integration range
in (1.8) does not need to be the whole R: it can actually be any interval (−ϵ0 , ϵ1 ),
with ϵ0 , ϵ1 > 0, that contains the point x = 0 (where δ is, by construction, infinite),
that is
 ϵ1
dx δ(x) = 1 , ∀ϵ0 , ϵ1 > 0 . (1.9)
−ϵ0

1
Because Dirac was the first one to use it for algebraic manipulations, as a mere continuous
analog of the Kronecker delta, as can be e.g. seen in his celebrated textbook on quantum mechanics,
The Principles of Quantum Mechanics.
CHAPTER 1. DERIVATION OF THE DIRAC. . . 4

Furthermore, note that we also have

 −ϵ0  ϵ1
dx δ(x) = dx δ(x) = 0 , ∀ϵ0 , ϵ1 > 0 . (1.10)
−ϵ1 ϵ0

We also agree that, even when we deal with this special mathematical object
δ(x), the usual rule for reversing the limits of integration upon changing the sign
still applies, that is
 −∞  ∞
dx δ(x) = − dx δ(x) = −1 . (1.11)
∞ −∞

1.3 Some properties of the Dirac δ-function

Let’s first prove the following (very!) important property of the δ-function: for a
function f (x), we have
 ∞
dx f (x)δ(x) = f (0) . (1.12)
−∞

Such a function f is typically called a test function.

Proof: From (1.7) we know that δ(x) = 0 for any x ̸= 0, which immediately
allows us to write

f (x)δ(x) = f (0)δ(x) ,

so that we get
 ∞  ∞  ∞
dx f (x)δ(x) = dx f (0)δ(x) = f (0) dx δ(x) ,
−∞ −∞ −∞

and thus, in view of (1.8),

 ∞
dx f (x)δ(x) = f (0) ,
−∞

which is indeed precisely (1.12). ■

Let’s then show that the Dirac δ-function is even, in the sense that

δ(−x) = δ(x) . (1.13)

Proof: First, it is clear from (1.7) that δ(−x) = 0 for any x ̸= 0: because if
CHAPTER 1. DERIVATION OF THE DIRAC. . . 5

x < 0, then −x > 0 and thus δ(−x) = 0 in view of (1.7), and similarly if x > 0,
then −x < 0 and thus δ(−x) = 0 in view of (1.7).
Then, for the integral of δ(−x), let’s make the change of variable

x → x′ = −x ,

with

dx = −dx′ ,

so that, also using (1.11), we have

 ∞  −∞  ∞
′ ′
dx δ(−x) = (−dx ) δ(x ) = dx′ δ(x′ ) = 1 ,
−∞ ∞ −∞

so that δ(−x) satisfies the exact same two conditions (1.7) and (1.8) as δ(x). This
allows us to conclude that

δ(−x) = δ(x) ,

that is indeed (1.13). ■

We can then define the translation of δ(x) as follows: for any a ∈ R, δ(x − a) is
such that

δ(x − a) = 0 , ∀x ̸= a , (1.14)

and
 ∞
dx δ(x − a) = 1 . (1.15)
−∞

Proof: First, it is clear from (1.7) that δ(x − a) = 0 for any x ̸= a, since to say
that x ̸= a is nothing but saying that x − a ̸= 0.
Then, for the integral of δ(x − a), let’s make the change of variable

x → x′ = x − a ,

with

dx = dx′ ,
CHAPTER 1. DERIVATION OF THE DIRAC. . . 6

so that we have
 ∞  ∞
dx δ(x − a) = dx′ δ(x′ ) = 1 .
−∞ −∞

Therefore, δ(x−a) indeed satisfies the two conditions (1.14) and (1.15). ■

This then allows to generalize (1.12) as follows: for a function f (x), we have
 ∞
dx f (x)δ(x − a) = f (a) . (1.16)
−∞

Proof: From (1.14) we know that δ(x−a) = 0 for any x ̸= a, which immediately
allows us to write

f (x)δ(x − a) = f (a)δ(x − a) ,

so that we get
 ∞  ∞  ∞
dx f (x)δ(x − a) = dx f (a)δ(x − a) = f (a) dx δ(x − a) ,
−∞ −∞ −∞

and thus, in view of (1.15),

 ∞
dx f (x)δ(x − a) = f (a) ,
−∞

which is indeed precisely (1.16). ■

Chapter 2

Representations of the Dirac

δ-function

As we discussed above, the Dirac δ-function is not a function: however, it can be

represented by means of functions. A possible approach to do this is to view the
δ-function as the limit of some function.
Intuitively, the δ-function can indeed be viewed as the limit of a function with
a very narrow width and a very large height, with the height varying inversely with
the width so as to make the area under the curve finite and equal to 1. While
there are (infinitely) many possibilities to construct such functions, here we’ll ex-
plicitly consider two different representations of the δ-function: first, as the limit of
a rectangular window function, and then as the limit of a Gaussian function.

2.1 Rectangular representation

Let’s consider the function fϵ (x) defined by

 1/ϵ , if x ∈ (−ϵ/2, ϵ/2)
fϵ (x) ≡ , (2.1)
0 , otherwise


where ϵ > 0 is an arbitrary positive real number. This function fϵ (x) is thus a mere
rectangular window function, of width ϵ and height 1/ϵ.
Let’s first compute the integral of fϵ on R: clearly, we have
 ∞  ϵ/2
1 1 ϵ/2 1 ϵ ϵ
dx fϵ (x) = dx = [x]−ϵ/2 = + ,
−∞ −ϵ/2 ϵ ϵ ϵ 2 2

7
CHAPTER 2. REPRESENTATIONS OF THE DIRAC. . . 8

that is
 ∞
dx fϵ (x) = 1 , ∀ϵ > 0 . (2.2)
−∞

Then, let’s quickly analyze the behavior of fϵ (x) as we decrease the value of the
parameter ϵ: by doing this, we make fϵ (x) take nonzero values only within a narrower
and narrower interval (−ϵ/2, ϵ/2), while the amplitude 1/ϵ of fϵ (x) becomes larger
and larger. Therefore, intuitively, we have in the limit ϵ → 0

 ∞ , if x=0
lim fϵ (x) = . (2.3)
ϵ→0
0 , if x ̸= 0


Comparing (2.2)-(2.3) with the definition (1.7)-(1.8) of the Dirac δ-function hence
allows to write the following rectangular representation of δ(x):

δ(x) = lim fϵ (x) , (2.4)

ϵ→0

with fϵ (x) the rectangular window function (2.1).

2.2 Gaussian representation

Let’s now consider the function gϵ (x) defined by

1 2 2
gϵ (x) ≡ √ e−x /ϵ , (2.5)
ϵ π

where ϵ > 0 is, here again, an arbitrary positive real number. This function gϵ (x) is
a Gaussian function centered at position x = 0, whose width1 is characterized by ϵ.
Let’s first compute the integral of gϵ on R. To do this, we recall the following
standard result regarding the integral of a Gaussian function:
 ∞
2 √
dx e−x = π. (2.6)
−∞

We then use the change of variable

x
x → x′ = ,
ϵ
1
Namely,
√ the full width at half maximum of gϵ (x) is proportional to ϵ, more precisely it is given
by 2 ln2ϵ.
CHAPTER 2. REPRESENTATIONS OF THE DIRAC. . . 9

such that

dx = ϵdx′ ,

to get for the integral of gϵ (x) (also remembering that ϵ > 0 by assumption)
 ∞  ∞  ∞  ∞
1 −x2 /ϵ2 1 ′ −x′2 1 2
dx gϵ (x) = √ dx e = √ ϵdx e =√ dx e−x
−∞ ϵ π −∞ ϵ π −∞ π −∞

that is in view of (2.6)

 ∞
dx gϵ (x) = 1 , ∀ϵ > 0 . (2.7)
−∞

Then, let’s here again analyze the behavior of gϵ (x) as we decrease the parameter
ϵ. For a given nonzero value of x, say x = x0 ̸= 0, the value gϵ (x0 ) gets smaller
and smaller and eventually becomes indistinguishable from zero. For instance, we
indeed have for x0 = 1 and ϵ = 1, 0.5, 0.2, 0.1

g1 (1) ≈ 0.21 , g0.5 (1) ≈ 0.02 , g0.2 (1) ≈ 3.9 × 10−11 , g0.1 (1) ≈ 2.1 × 10−43 .

2 2
This behavior comes from the exponential part e−x /ϵ in (2.5), which rapidly decays
as ϵ decreases and completely dominates the corresponding increase of the 1/ϵ factor
2 2
in (2.5). Only at x = 0 does this decaying term e−x /ϵ disappear, since e0 = 1.
Therefore, we have

lim gϵ (x) = 0 , ∀x ̸= 0 . (2.8)

ϵ→0

Furthermore, precisely at x = 0 we immediately get from (2.5) that

1
gϵ (0) = √ ,
ϵ π

so that we readily see that

1
lim gϵ (0) = lim √ = ∞ . (2.9)
ϵ→0 ϵ→0 ϵ π

Therefore, combining (2.8) with (2.9) allows us to write


 ∞ , if x=0
lim gϵ (x) = . (2.10)
ϵ→0
0 , if x ̸= 0


Comparing (2.7) and (2.10) with the definition (1.7)-(1.8) of the Dirac δ-function
CHAPTER 2. REPRESENTATIONS OF THE DIRAC. . . 10

hence allows to write the following Gaussian representation of δ(x):

δ(x) = lim gϵ (x) , (2.11)

ϵ→0

with gϵ (x) the Gaussian function (2.5).

Chapter 3

Integral and derivatives of the Dirac

δ-function

We first compute the integral of the Dirac δ-function, which is obtained from the
definition (1.7)-(1.8) of δ(x). We then construct the derivatives (of any order n ∈ N)
of the δ-function.

3.1 Integral of the Dirac δ-function

Let’s first define the so-called Heaviside function H(x), also called the step-function,
given by

 0 , if x<0
H(x) ≡ , (3.1)
1 , if x>0


as well as the sign function (also called signum function 1 ) sgn(x) given by

 −1 , if x<0
sgn(x) ≡ . (3.2)
1 , if x>0


Note that in view of (3.1) and (3.2) we can thus in particular write

H(x) − H(−x) = sgn(x) . (3.3)

Let’s now consider the integral of the δ-function δ(x). Clearly, in view of its
1
This name may be clearer in spoken English, so as to avoid any confusion between the sign
and the sine functions!

11
CHAPTER 3. INTEGRALS AND DERIVATIVES OF. . . 12

definition (1.7)-(1.8), we have [see also (1.9)-(1.10)]



x  0 , if x<0
′ ′
dx δ(x ) = . (3.4)
−∞
1 , if x>0


Comparing (3.4) with (3.1) hence readily shows that

 x
dx′ δ(x′ ) = H(x) . (3.5)
−∞

Furthermore, we readily get from (3.4) that



x  −1 , if x<0
−1 + 2 dx′ δ(x′ ) = ,
−∞
1 , if x>0


so that we get from (3.2)

 x
sgn(x) = −1 + 2 dx′ δ(x′ ) . (3.6)
−∞

It is thus in particular clear from (3.4) that the δ-function becomes less singular
after integration: while δ(x) is an infinite spike, its integral is simply a function with
a finite discontinuity.
Note also that differentiating2 (3.5) or (3.6) readily shows that δ(x) can thus also
be expressed as the derivative of a function that has a finite discontinuity, namely

d 1 d
δ(x) = H(x) = sgn(x) . (3.7)
dx 2 dx

Intuitively, this makes sense. Indeed, both the Heaviside and the sign functions are
constant for any x ̸= 0, so differentiating them should yield zero for any x ̸= 0,
which is indeed what δ(x) must satisfy. Furthermore, both the Heaviside and the
sign functions are discontinuous at x = 0: the tangent to their curves should thus be
vertical at x = 0, i.e. this tangent should have an infinite slope, i.e. the derivatives
of both the Heaviside and of the sign functions at x = 0 should be infinite, which is
indeed the behavior exhibited by δ(x) at x = 0.
2
Remember the Leibniz integral rule for differentiation under the integral sign: namely,
 b(y)  b(y)
d db da ∂f
dx f (x, y) = f [b(y), y] − f [a(y), y] + dx ,
dy a(y) dy dy a(y) ∂y

where the integration limits a(y) and b(y) are two functions that depend on the differentiation
variable y.
CHAPTER 3. INTEGRALS AND DERIVATIVES OF. . . 13

3.2 Derivatives of the Dirac δ-function

Contrary to the integral of δ(x), its derivative δ ′ (x) can not be constructed from δ
itself: indeed, what would it mean to differentiate the infinite spike exhibited by δ
at x = 0? So we must actually define what the derivative of the δ-function should
be. We must construct a definition that is as reasonable as possible in view of what
we expect from a derivative.
First of all, since δ(x) is only singular at x = 0 (and constant, equal to zero,
for x ̸= 0), we can already expect that its derivative δ ′ (x) should exhibit a similar
behavior. Therefore, it seems reasonable to require that δ ′ (x) should be zero for any
x ̸= 0,

δ ′ (x) = 0 , ∀x ̸= 0 . (3.8)

Then the question is: what do we do at x = 0? Well, for this, let’s remember
the property (1.12) of the δ-function, namely
 ∞
dx f (x)δ(x) = f (0) , (3.9)
−∞

for some function f (x). Can we now write something similar but for the derivative
δ ′ (x), i.e. can we evaluate
 ∞
dx f (x)δ ′ (x) (3.10)
−∞

given the known action (3.9) of δ(x) on any function? To do this, we actually
merely require that integration by parts should also apply to integrals that involve
the δ-function and its derivative(s), a requirement that indeed seems to be rather
reasonable. That is, we rewrite the integral (3.10) by means of an integration by
parts, and we get
 ∞  ∞
′
dx f (x)δ (x) = [f (x)δ(x)]∞
−∞ − dx f ′ (x)δ(x) . (3.11)
−∞ −∞

Let’s separately compute the two terms in the right-hand side of (3.11). First,
because δ(x) = 0 for any x ̸= 0, we have in particular

lim δ(x) = 0 ,
x→±∞

so that the first term in the right-hand side of (3.11) vanishes, i.e.

[f (x)δ(x)]∞
−∞ = 0 . (3.12)
CHAPTER 3. INTEGRALS AND DERIVATIVES OF. . . 14

Then for the second term in the right-hand side of (3.11), we merely apply the
known property (3.9) of the δ-function, that is we have
 ∞
dx f ′ (x)δ(x) = f ′ (0) . (3.13)
−∞

Finally, substituting (3.12) and (3.13) into (3.11) yields

 ∞
dx f (x)δ ′ (x) = −f ′ (0) . (3.14)
−∞

Therefore, combining (3.8) with (3.14), we see that the derivative δ ′ (x) of the Dirac
δ-function is defined by the two conditions

δ ′ (x) = 0 , ∀x ̸= 0 , (3.15)

and
 ∞
dx f (x)δ ′ (x) = −f ′ (0) . (3.16)
−∞

We can then proceed in the exact same way, i.e. by successive integrations by
parts, to define the derivative δ (n) (x) of an arbitary order n ∈ N. Requiring here
again the validity of the successive integration by parts, we define δ (n) (x) such that
it satisfies the two conditions

δ (n) (x) = 0 , ∀x ̸= 0 , (3.17)

and
 ∞
dx f (x)δ (n) (x) = (−1)n f (n) (0) , (3.18)
−∞

where f (n) (0) denotes the n-th derivative of the function f evaluated at x = 0.