GENERAL I ARTICLE

All about the Dirac Delta Function(?)

V Balakrishnan

Mouse Games
As any child of ten will tell you, to write an article on
the Dirac delta function (or on anything else, for that
matter), one must first log into 'Google' or 'Yahoo' or a
similar search engine. A judicious combination of click-
V Balakrishnan is in the ing, cutting and pasting - and voila, an article of any
Department of Physics, desired length is ready in an unbelievably short time!
Indian Institute of
Technology Madras, Now this is not as simple as it sounds. It does re-
Chennai. His research quire some finesse - possessed, no doubt, by the average
interests are in dynamical
ten-year old (but not necessarily by older, less capable
systems, stochastics and
statistical physics.
surfers). If one is naive enough to enter delta function
in Google and click on search, the magician takes only
0.16 seconds to produce a staggering 1,100,000 possible
references. Several lifetimes would not suffice to check
all of these out. To make the search more meaningful,
we enter "delta function" in quotes. This produces a less
stupendous 58,600 references. As even this is too much,
we try Dirac delta function, to get 52,500 references -
not much of an improvement. Once again, "Dirac delta
function" is much better, because Google then locates
only 12,100 references. "Dirac's delta function" brings
this down to 872, while "the delta function of Dirac"
yields a comfortable (but not uniformly helpful) 19 ref-
erences.
Motivated by a desire to include some interesting histor-
ical aspects in my article, I continued this fascinating
pastime by trying history of the Dirac delta function,
to be presented with 6,570 references to choose from.
Spotting my mistake, I promptly moved to "history of
the Dirac delta function' , to be told that there were just
Keywords
Delta function, distributions, 2 references, a most satisfying conclusion to the game.
generalized functions. .One of these was from M athematica, and was as short

--------~--------
48 RESONANCE I August 2003
 GENERAL I ARTICLE

and sweet as befits this impatient age. It said (in its en-
tirety): 0 Heaviside (1893-95), G Kirchhoff (1891), P A
M Dirac (1926), L Schwartz (1945). A true capsule his-
tory - provided you already knew the history! The other
reference I didn't pursue, asthe computer 'froze' at this
juncture. After all, the system had worked for nearly
forty-five minutes without a hitch, and some such event
was long overdue. The message was clear: it was time
to get down to real work by shutting down the system
and reverting to pencil and pad.
But what about the title of the article? Back to Google.
Brief experimentation showed that "All about the Dirac
delta functi~n" produced zero references, so this title
practically selected itself. Once this issue of Resonance
goes on-line, this article will be the sole reference, for
the time being, if you cared to search under "All about
the Dirac delta function" - but do note the all-important
question mark in my title, added for the sake of truth
and honesty!
What the Dirac Delta Function looks like
Suppose f (x) is a function that is defined, say, for all
values of the real variable x, and that it is finite every-
where. Can we construct some sort of filter or 'selector'
that, when operating on this function, singles out the
value of the function at any prescribed point xo?
A hint is provided by the discrete analogue of this ques-
tion. Suppose we have a sequence (aI, a2, .) = {aj Ii =
1, 2, .}. How do we select a particular inember ai from
the sequence? By summing over all members of (i.e.,
scanning!) the sequence with a selector called the Kro-
necker delta, denoted by 6ij and defined as 6ij = 1 if
i = j, and 8ij = O' if i =1= j. It follows immediately that

L8 ij aj = ai (1)
j=l

Further, we have the normalization Lj 8ij = 1 for each

-RE-S-O-N-A-NC-E--I-A-U9-U-st-2-0-03-----------~--------------------------4-9
 GENERAL I ARTICLE

value of i, and also the symmetry property 8ij = 8ji

Reverting to the continuous case, we must replace the
summation over j by an integration over x. The role
of the specified index i is played by the specified point
Xo The analogue of the Kronecker delta is written like
a function, retaining the same symbol 8 for it. (Presum-
ably, this was Dirac's reason for choosing this notation
for the delta function.) So we seek a 'function' 8(x - xo)
such that

f: dx5(x - xo) f(x) = f(xo) (2)

for continuous functions I (x).

Exactly as in the discrete case of the Kronecker delta,
we impose the normalization and symmetry properties

f: dx5(x-xo) = 1 and 5(x-xo) = 5(xo-x) (3)

The requirements in (2) and (3) may be taken to define

the Dirac delta function. The form of (2) suggests that
8(x - xo) is more like the kernel of an integral operator
than a conventional function. We will return to this
aspect subsequently.
What can 8(x - xo) possibly look like? A naive way of
answering this question is as follows. Take a rectangular
window of width 2c and height 1/(2c), so that the area
of the window is unity. Place it with its bottom edge
on the x axis and slide it along this axis, as shown in
Figure 1.
When the window is centred at the chosen point xo, the
integral of I (x) multiplied by this window function is
The form suggests
simply (1/2c) J:oo~: dx I(x) This does not quite se~ect
that t5 (x-xo) is more
I (xo) alone, of course. But it will do so if we take the
like the kernel of an limit c ~. O. In this limit, the width of the window
integral operator becomes vanishingly small. Simultaneously, its height
than a conventional becomes arbitrarily large, so as to 'capture' all of the
function. ordinate in the graph of I (x), no matter how large the

--------~--------
50 RESONANCE I August 2003
 GENERAL I ARTICLE

Figure 1. The window re-

duces to the delta function,
---------.- when E~ O.
window
I 1.....1 - - - - - -

i
1

value of f(xo) is. A possible explicit form for the Dirac

delta function 8(x - xo) is therefore given by

8(x-x)= {lime --+ o l/(2c), forxo-c<x<xo+c

° 0, for all other x .
(4)
This cannot be a stand-alone definition. If it is taken
literally, then, formally, 8(x - xo) must be zero for all
x -=f xo, while it must be infinite for x = Xo An ex-
plicit form such as (4) for the delta function must be
interpreted in the light of (2). The delta 'function' is
always to be understood as something that makes sense
when it occurs in an integral like (2), i.e., when it acts
on ordinary functions like f (x) and an integration is
carried out. It is immediately clear that the- so-called
Dirac delta 'function' cannot be a function in the con-
ventional sense. In particular, 8(x-xo) must be singular
(formally infinite) at x = Xo, that is, at the point where
its argument is zero.
The delta 'function'
Mathematically, an explicit form for the Dirac delta
is always to be
function is properly given in terms of a sequence or fam-
understood as
ily of conventional functions, rather than the 'window'
som~thing that
representation in (4). It can then be arranged that, in a
makes sense
suitable limit, the sequence approaches a quantity that
when it occurs in
has all the properties desired of the _delta function. An
an integral.
infinite number of such sequen~es may be constructed.

-RE-5-0-N-A-N-C-E-I-A-U-9-US-t-2-00-3------------~-----------------------------~
 GENERAL I ARTICLE

For instance, take a family of functions ¢e: (x - xo) para-

metrized by a positive constant c , and with the following
properties: each member of the family (i) has a peak at
xo, (ii) is symmetric about the point xo, and (iii) has
a definite integral from -00 to 00 whose value is unity.
Matters are arranged such that, as the parameter c is
made smaller and smaller, the height of the peak in ¢e: (x)
increases while its width simultaneously decreases, keep-
ing the total area under the curve equal to unity. Then
lime:~o ¢e: (x - xo) represents the delta function <5 (x - xo)
Let us now write down the simplest choices for such se-
quences. For ease of writing, let us set Xo = O. One of
the simplest possibilities is the family of 'Lorentzians',
given by
(5)

Then lime:~o ¢e: (x) is a representation of the Dirac delta

function <5(x), with the properties specified in (2) and
(3). Some other popular choices for ¢e:(x) are the fol-
lowing:
1 1 2
2c exp (-Ixl/c); 2fo exp (-x 14c);

sin(xlc) .
, (6)
1TX
It is instructive to sketch these functions schematically,
and to check out what happens as smaller and smaller
values of c are chosen. As an amusing exercise, think up
your own sequence of functions that leads to the delta
function as a limiting case.

As an amusing What's the point of all this? Before going on to an-

exercise, think up swer this question, it's helpful to re-write the last of the
your own sequence functions in (6) as follows. If we put c = IlL, we get
of functions that
leads to the delta
~() l' sin (Lx) = -1 l m
uX=lm
L~oo 1T X
l'
21T L~oo
lL dk
- L
eikx =
function as a limiting
case. ~ roo dke ikx (7)
27r i-oo

-52----------------------------~----------------------------­
RESONANCE I August 2003
 GENERAL I ARTICLE

This turns out to be perhaps the most useful way of

representing the delta function. Since Ieikx I = 1, it is
obvious that the last integral in (7) is not absolutely con-
vergent. Nor is the integral well-defined in the ordinary
sense, because sin kx and cos kx do not have definite
limits as k -4 ±oo. These are just further reminders
of the fact that the delta function is not a conventional
function, as we have already emphasized. Those young
readers who are familiar with Fourier transforms will
recognize that the· last equation above seems to suggest
that the Fourier transform of the Dirac delta function is
just unity. This is indeed so. It suggests, too, that one
way of defining 'singular' functions like the delta func-
tion might be via their Fourier transforms: for example,
we could define 8 (x) as the inverse Fourier transform of
a constant - in this case, just unity.
Some History
The Dirac delta function has quite a fascinating his-
tory. The book by Liitzen, cited at the end of this ar-
ticle, is an excellent source of information. The delta
function seems to have made its first appearance in the
early part of the 19th century, in the works of Pois-
son (1815), Fourier (1822), and Cauchy (1823, 1827).
Poisson and Cauchy essentially used arguments that im-
plied that the Lorentzian representation of the delta
function, (5), had the 'selector property' stated in (2).
Fourier, in his fundamental work Theone AnaZytique
de la Chaleur, showed (in connection with Fourier se-
The delta function
ries expansions of periodic functions) that the series
seems to have made
1/ (27r) + (1/ 7r) L~l cos n( x - xo) had precisely this sort
its first appearance
of selector property, i.e., was a representation of 8(x-xo)
in the early part of
in the fundamental interval (x - xo) E [-7r,7r). His
the 19th century in
I
arguments essentially amount to the last of the repre-
the works of Poisson
sentations in (6) for 8(x). These early works did not
(1815), Fourier
aim at mathematical rigour in the current sense of the
terr~. Subsequently, Kirchhoff (1882, 1891) and Heavi- (1822), and Cauchy
side (1893, 1899) gave the first mathematical definitions (1823,1827).

-RE-S-O-N-A-N-C-E-I-A-U-9-US-t-2-00-3----------~~-----------------------------53
 GENERAL I ARTICLE

(again non-rigorous, by modern standards) of the delta

function. Kirchhoff was concerned with the fundamental
solution of the three-dimensional wave equation, while
Heaviside introduced the function in his 'Operational
Calculus'. He pointed out that 8(x) could be regarded
as the derivative of the Heaviside or unit 'step function'
O(x), defined as unity for x > 0 and zero for x < o. After
Heaviside, the delta function was freely used - in partic-
ular, in connection with Laplace transforms, especially
by electrical engineers (e.g., Van der Pol, 1928). Dirac
(1926, 1930) introduced it in his classic and fundamental
work on quantum mechanics, essentially as the contin-
uous analogue of the Kronecker delta. He also wrote
down a list of its important properties - much the same
list that standard textbooks now carry. Over and above
(2) and (3), the delta function also satisfies

x8(x)=O, 8'(-x)=-8'(x), x8'(x)=-8(x),

8(ax) = (I/lal) 8(i), (8)

where a is any real number, and so on. Again, these
equationS are to be understood as valid when multi-
plied by suitable smooth functions and integrated over
x. Dirac also listed the useful but not immediately ob-
vious property

8( 2 _ 2) _ 8(x + xo) + 8(x - xo) (9)

x Xo - 21 x ol '

where Xo is any real number.

The use of the delta function became more and more
The use of the
common after the appearance of Dirac's work. Other
delta function
singular functions also made their appearance, as ~ar ly
became more and
versions of quantum field theory began to take shape in
more common
the works of physicists such as Jordan, Pauli and Heisen-
after the berg. Around the same time, mathematicians began at-
appearance of tempts to define such singular quantities in a rjgorous
Dirac's work. manner. The delta function and other such singular

--------~--------
54 RESONANCE I August 2003
 GENERAL I ARTICLE

objects were recognized to be what are called general-

ized functions or distributions, rather than functions in
the conve?tional sense. The first rigorous theory was
given by Bochner in 1932. Soon afterwards, Sobolev
(1935) gave the rigorous definition of distributions as
functionals, and the way had been paved for a definitive
mathematical theory. This was achieved by Schwartz
(1945-50), and comprehensively treated in his Theorie
des Distributions, Vol. 1 (1950) and Vol. 2 (1951). For
lack of space, we will not go further into these aspects,
other than to repeat that we now have a completely rig-
orous mathematical theory of distributions.
Why does the 8-Function appear in Physical Prob-
lems?
We can now turn to the question of why the delta func-
tion appears so naturally in physical problems. Con-
sider, for example, the basic problem of electrostatics:
given a static charge density per) in free space, what is
the corresponding electrostatic potential ¢(r) at any ar-
bitrary point r = (x, y, z)? From Maxwell's equations,
we know that ¢ satisfies Poisson's equation, namely,

(10)

where EO is the permittivity of the vacuum. What does

one do in the case of a point charge q located at some
point ro = (xo, Yo , zo)? A point charge is an idealiza- The first rigorous
tion in which a finite amount of charge q is supposed to theory was given by
be packed into zero volume. The charge density must Bochner in 1932.
therefore be infinite at the point ro, and zero elsewhere. Soon afterwards,
The delta function comes to our aid: we may write, in Sobolev (1935) gave
this case, the rigorous definition
of distributions as
per) = q 8(x - xo) 8(y - Yo) 8(z - zo) = q8(3)(r -
ro),
functionals, and the
(11)
way had been paved
where the three-dimensional delta function 8(3) is short-
for a definitive
hand for the product of the three delta functions in the
mathematical theory.
equation above. It is easy to verify that this expression

-RE-S-O-N-A-N-C-E-I-A-U-9-US-t-2-00-3------------~-----------------------------~
 GENERAL I ARTICLE

for p( r) has all the properties required of a point charge

at the point ro This illustrates how the delta function
frequently appears as-the right-hand side of fundamen-
tal equations of mathematical physics. It turns out that
it also appears as the singular part of fundamental so-
lutions to basic equations such as the wave equation.
It is worth noting that representations of 'higher-dimen-
sional' delta functions like 8(3) are easily written down.
For instance, the three-dimensional counterpart of (7)
above is just

(12)

The notation used in the final equation should be self-

explanatory.
We have mentioned earlier that Poisson was responsi-
ble for what was perhaps the first recognizable use of
the Dirac delta function. Poisson and Dirac seem to be
linked in more ways than one. Di~ac showed that Pois-
son brackets in classical dynamics essentially become the
commutators of the corresponding operators in quan-
tum mechanics. It is therefore appropriate to end this
short account with another fascinating link behyeen the
names of Dirac and Poisson. There is a very useful and
remarkable result in Fourier analysis called the Poisson
summation formula. In its simplest form, this says that
if J( k) is the Fourier transform of f (x), then
00 00

2: f(n) = L J(27rn) (13)

n=-oo n=-oo
Poisson and Dirac
seem to be linked
valid for suitably restricted functions f (x).
in more ways than A very elegant and simple way of deriving this formula
one. makes use of the so-called 'Dirac comb': an array of

-56----------------------------~------------RE-S-O-N-A-N-C-E-1-A-U-9-US-t-2-00-3
 GENERAL I ARTICLE

Box 1. Why did Dirac need the delta function?

The delta function appeared in Dirac's work on quantum mechanics in an avatar somewhat
different from the ones mentioned in the text.
Consider ordinary three-dimensional (Euclidean) space. This is a linear vector space (LVS).
Any vector in it can be expanded uniquely as a linear combination of the three unit vectors
i ,j and k. This is beca.use these three vectors are linearly independent of each other, and
they span the space: i.e., they form a basis in the LVS. Moreover, i . i = 1, i . j = 0,
etc. Using the superior notation el == i, e2 == j and e3 == k, all these relations can be
compressed into ei . ej = Oij (i,j = 1, 2 or 3) , in terms of the Kronecker delta. That is, the
set {ei Ii = 1,2, 3} is an orthonormal basis. Tllis can be generalized to any n-dimensional
LVS: an orthonormal basis {ei} satisfying ei . ej = Oij (i,j = 1, 2, ... ,n) can always be
chosen in it.
What happens if the dimensionality n ~ oo? Some subtleties arise. But the preceding
discussion goes through, provided care is taken to ensure that certain desirable properties
survive - e.g., the vectors in the LVS must have finite magnitudes, and the triangle inequality
must be satisfied by the magnitude of the resultant of two vectors. Function spaces provide
simple examples of such infinite-dimensional LVS's - for instance, the space of all square-
integrable functions of x in some interval [a, b]. Naturally, the basis is then an infinite set
of suitable functions. A common example is the set of Legendre polynomials {Pl(X)} where
x E [-1, 1] and f = 0, 1, . .. ad info The notion of the dot product of two vectors must also
be generalized appropriately. We do not go into this detail here.
But a new possibility arises when the dimensionality of an LVS is infinite. It may have a
basis that is itself uncountably infinite, i.e., a so-called continuous basis. Instead of a set {ed
e
where i is a discrete index, we have a set { e(~)} where is a continuous variable, taking values
in some range. (For simplicity we continue to use the symbol e for the basis elements of the
LVS, even when these may be functions or other objects.) Omitting several technical details,
the orthonormality condition for a continuous basis fonnally reads e(~) . e(~ ') = o(~ - ~ ') .
That is, the Dirac delta function replaces the Kronecker delta.
This is the context in which Dirac required the delta fUhction. In quantum mechanics, a
system is described by its so-called state vector. This is an element of a certain LVS called the
Hilbert space of the system. The classical dynamical observables of the system are replaced
by operators that act on the elements of its Hilbert space. It turns out to be convenient
to choose the eigenstates of (a subset of) these physical operators as the possible basis sets
in the Hilbert space. Moreover, the system gets into these eigenstates when measurements
of the corresponding physical observables are made. Certain fundamental operators such as
the position operator or linear momentum operator of a particle moving in a given force field
turn out to have continuous sets of eigenvalues. Their eigenstates then constitute continuous
basis sets, with the orthononnality condition as given by the last equation above.
An example is provided by the case of a .particle moving in one dimension under the influence
of a constant force. Its energy E can then be shown to have a continuous set of possible
values. The state e (E) of the particle corresponding to a definite value E of its total energy
is a member of a continuous basis set of states, satisfying e (E) . e (E ') = o(E - E ') .

-RE-S-O-N-A-N-C-E--I-A-U-9-U-s-t-2-00-3-------------~~------------------------------~-
 GENERAL I ARTICLE

Dirac delta functions located at the integers. It can be

shown that
00 00

L 8(x-n) = L (14)
n=-oo n=-oo

i.e., the Dirac comb is identically equal to a sum of ex-

ponentials! The latter can be reduced to the expression
1 + 2 Er' cos (27rnx). The cosines in the sum 'interfere
destructively' with· each other, leaving behind just the
sharp 8-function spikes at integer values of x. With the
help of (14), the Poisson summation formula is estab-
lished quite easily. We have thus come full circle, mov-
ing from Poisson to Dirac and returning to Poisson with
the help of Dirac.
Suggested Reading

[1] PAM Dirac, The physical interpretation of the [sic] quantum

mechanics,Proc. Roy. Soc. A Vol. 113, 621-641,1926. So heady was the
progress in that incredible period that the defmite article preceding
'quantum mechanics' was presumably no longer needed by 1930! P A
M Dirac, The Principles of Quantum Mechanics, Oxford University
Press, Oxford, 1930 (original edition). An inexpensive Indian reprint of
the latest edition is available.
[2] M J Lighthill, Introduction to Fourier Analysis and Generalized Func-
tions, Cambridge University Press, Cambridge. Originally published in
Address for Correspondence 1958, this little classic has been reprinted several times. The book's
V Balakrishnan dedication is as succinct as its text, and says: "To Paul Dirac, who saw
Department of Physics that it must be true, Laurent Schwartz, who proved it, and George
Indian Institute of Technology Temple, who showed how simple it could be made".
Chennai 600 036, India.
[3] J Liitzen, The Prehistory ofthe Theory of Distributions, Springer-Verlag,
New York, 1982.

~
"It was a sort ofact offaith with us that any equations
which describe fundamental laws of Nature must have
.1 I' great mathematical beauty in them. It was a very
profitable religion to hold and can be considered as the
basis of much of our success"

-Dirac on himself and Schrodinger

-58--------------------------------~~-------------------------------
RESONANCE I August 2003