1.

15 Dirac Delta Function 83

1.14.4 Using Maxwell’s equations, show that for a system (steady current) the magnetic vector
potential A satisfies a vector Poisson equation,

∇ 2 A = −µ0 J,

provided we require ∇ · A = 0.

1.15 DIRAC DELTA FUNCTION

From Example 1.6.1 and the development of Gauss’ law in Section 1.14,
    
1 r̂ −4π
∇·∇ dτ = − ∇ · 2 dτ = (1.169)
r r 0,

depending on whether or not the integration includes the origin r = 0. This result may be
conveniently expressed by introducing the Dirac delta function,

 
1
2
∇ = −4πδ(r) ≡ −4πδ(x)δ(y)δ(z). (1.170)
r

This Dirac delta function is defined by its assigned properties

δ(x) = 0, x=0 (1.171a)

∞
f (0) = f (x)δ(x) dx, (1.171b)
−∞

where f (x) is any well-behaved function and the integration includes the origin. As a
special case of Eq. (1.171b),

∞
δ(x) dx = 1. (1.171c)
−∞

From Eq. (1.171b), δ(x) must be an infinitely high, infinitely thin spike at x = 0, as in the
description of an impulsive force (Section 15.9) or the charge density for a point charge.27
The problem is that no such function exists, in the usual sense of function. However, the
crucial property in Eq. (1.171b) can be developed rigorously as the limit of a sequence
of functions, a distribution. For example, the delta function may be approximated by the

27 The delta function is frequently invoked to describe very short-range forces, such as nuclear forces. It also appears in the
normalization of continuum wave functions of quantum mechanics. Compare Eq. (1.193c) for plane-wave eigenfunctions.
84 Chapter 1 Vector Analysis

FIGURE 1.37 δ-Sequence

function.

FIGURE 1.38 δ-Sequence

function.

sequences of functions, Eqs. (1.172) to (1.175) and Figs. 1.37 to 1.40:

 1
 0, x < − 2n
δn (x) = n, 1 1
− 2n < x < 2n (1.172)
 1
0, x > 2n
n

δn (x) = √ exp −n2 x 2 (1.173)
π
n 1
δn (x) = · (1.174)
π 1 + n2 x 2
sin nx 1 n
δn (x) = = eixt dt. (1.175)
πx 2π −n
 1.15 Dirac Delta Function 85

FIGURE 1.39 δ-Sequence function.

FIGURE 1.40 δ-Sequence function.

These approximations have varying degrees of usefulness. Equation (1.172) is useful in

providing a simple derivation of the integral property, Eq. (1.171b). Equation (1.173)
is convenient to differentiate. Its derivatives lead to the Hermite polynomials. Equa-
tion (1.175) is particularly useful in Fourier analysis and in its applications to quantum
mechanics. In the theory of Fourier series, Eq. (1.175) often appears (modified) as the
Dirichlet kernel:
1 sin[(n + 21 )x]
δn (x) = . (1.176)
2π sin( 21 x)
In using these approximations in Eq. (1.171b) and later, we assume that f (x) is well be-
haved — it offers no problems at large x.
86 Chapter 1 Vector Analysis

For most physical purposes such approximations are quite adequate. From a mathemat-
ical point of view the situation is still unsatisfactory: The limits
lim δn (x)
n→∞

do not exist.
A way out of this difficulty is provided by the theory of distributions. Recognizing that
Eq. (1.171b) is the fundamental property, we focus our attention on it rather than on δ(x)
itself. Equations (1.172) to (1.175) with n = 1, 2, 3, . . . may be interpreted as sequences of
normalized functions:
∞
δn (x) dx = 1. (1.177)
−∞

The sequence of integrals has the limit

∞
lim δn (x)f (x) dx = f (0). (1.178)
n→∞ −∞

Note that Eq. (1.178) is the limit of a sequence of integrals. Again, the limit of δn (x),
n → ∞, does not exist. (The limits for all four forms of δn (x) diverge at x = 0.)
We may treat δ(x) consistently in the form
∞ ∞
δ(x)f (x) dx = lim δn (x)f (x) dx. (1.179)
−∞ n→∞ −∞

δ(x) is labeled a distribution (not a function) defined by the sequences δn (x) as indicated
in Eq. (1.179). We might emphasize that the integral on the left-hand side of Eq. (1.179) is
not a Riemann integral.28 It is a limit.
This distribution δ(x) is only one of an infinity of possible distributions, but it is the one
we are interested in because of Eq. (1.171b).
From these sequences of functions we see that Dirac’s delta function must be even in x,
δ(−x) = δ(x).
The integral property, Eq. (1.171b), is useful in cases where the argument of the delta
function is a function g(x) with simple zeros on the real axis, which leads to the rules
1
δ(ax) = δ(x), a > 0, (1.180)
a

  δ(x − a)
δ g(x) = . (1.181a)
a, |g ′ (a)|
g(a)=0,
′
g (a)=0

Equation (1.180) may be written

∞ ∞  
1 y 1
f (x)δ(ax) dx = f δ(y) dy = f (0),
−∞ a −∞ a a

28 It can be treated as a Stieltjes integral if desired. δ(x) dx is replaced by du(x), where u(x) is the Heaviside step function
(compare Exercise 1.15.13).
 1.15 Dirac Delta Function 87
1
applying Eq. (1.171b). Equation (1.180) may be written as δ(ax) = |a| δ(x) for a < 0. To
prove Eq. (1.181a) we decompose the integral
∞
  a+ε

f (x)δ g(x) dx = f (x)δ (x − a)g ′ (a) dx (1.181b)
−∞ a a−ε

into a sum of integrals over small intervals containing the zeros of g(x). In these intervals,
g(x) ≈ g(a) + (x − a)g ′ (a) = (x − a)g ′ (a). Using Eq. (1.180) on the right-hand side of
Eq. (1.181b) we obtain the integral of Eq. (1.181a).
Using integration by parts we can also define the derivative δ ′ (x) of the Dirac delta
function by the relation
∞ ∞
f (x)δ ′ (x − x ′ ) dx = − f ′ (x)δ(x − x ′ ) dx = −f ′ (x ′ ). (1.182)
−∞ −∞

We use δ(x) frequently and call it the Dirac delta function29 — for historical reasons.
Remember that it is not really a function. It is essentially a shorthand notation, defined
implicitly as the limit of integrals in a sequence, δn (x), according to Eq. (1.179). It should
be understood that our Dirac delta function has significance only as part of an integrand.
In this spirit, the linear operator dx δ(x − x0 ) operates on f (x) and yields f (x0 ):
∞
L(x0 )f (x) ≡ δ(x − x0 )f (x) dx = f (x0 ). (1.183)
−∞

It may also be classified as a linear mapping or simply as a generalized function. Shift-

ing our singularity to the point x = x ′ , we write the Dirac delta function as δ(x − x ′ ).
Equation (1.171b) becomes
∞
f (x)δ(x − x ′ ) dx = f (x ′ ). (1.184)
−∞

As a description of a singularity at x = x ′ , the Dirac delta function may be written as

δ(x − x ′ ) or as δ(x ′ − x). Going to three dimensions and using spherical polar coordinates,
we obtain
2π π ∞ ∞
δ(r)r 2 dr sin θ dθ dϕ = δ(x)δ(y)δ(z) dx dy dz = 1. (1.185)
0 0 0 −∞

This corresponds to a singularity (or source) at the origin. Again, if our source is at r = r1 ,
Eq. (1.185) becomes

δ(r2 − r1 )r22 dr2 sin θ2 dθ2 dϕ2 = 1. (1.186)

29 Dirac introduced the delta function to quantum mechanics. Actually, the delta function can be traced back to Kirchhoff, 1882.
For further details see M. Jammer, The Conceptual Development of Quantum Mechanics. New York: McGraw–Hill (1966),
p. 301.
88 Chapter 1 Vector Analysis

Example 1.15.1 TOTAL CHARGE INSIDE A SPHERE


Consider the total electric flux E · dσ out of a sphere of radius R around the origin
surrounding n charges ej , located at the points rj with rj < R, that is, inside the sphere.
The electric field strength E = −∇ϕ(r), where the potential
n
 ej ρ(r′ ) 3 ′
ϕ= = d r
|r − rj | |r − r′ |
j =1

is the sum
of the Coulomb potentials generated by each charge and the total charge density
is ρ(r) = j ej δ(r − rj ). The delta function is used here as an abbreviation of a pointlike
density. Now we use Gauss’ theorem for
  
2 ρ(r) j ej
E · dσ = − ∇ϕ · dσ = − ∇ ϕ dτ = dτ =
ε0 ε0
in conjunction with the differential form of Gauss’s law, ∇ · E = −ρ/ε0 , and
 
ej δ(r − rj ) dτ = ej .
j j

Example 1.15.2 PHASE SPACE

In the scattering theory of relativistic particles using Feynman diagrams, we encounter the
following integral over energy of the scattered particle (we set the velocity of light c = 1):



d 4 pδ p 2 − m2 f (p) ≡ d 3 p dp0 δ p02 − p2 − m2 f (p)

d 3 p f (E, p) d 3 p f (E, p)
= + ,
E>0 2 m2 + p2 E<0 2 m2 + p2
where we have used Eq. (1.181a) at the zeros E = ± m2 + p2 of the argument of the
delta function. The physical meaning of δ(p 2 − m2 ) is that the particle of mass m and
four-momentum p µ = (p0 , p) is on its mass shell, because p 2 = m2 is equivalent to E =
± m2 + p2 . Thus, the on-mass-shell volume element in momentum space is the Lorentz
3
invariant d2Ep , in contrast to the nonrelativistic d 3 p of momentum space. The fact that
a negative energy occurs is a peculiarity of relativistic kinematics that is related to the
antiparticle.

Delta Function Representation by Orthogonal

Functions
Dirac’s delta function30 can be expanded in terms of any basis of real orthogonal functions
{ϕn (x), n = 0, 1, 2, . . .}. Such functions will occur in Chapter 10 as solutions of ordinary
differential equations of the Sturm–Liouville form.
30 This section is optional here. It is not needed until Chapter 10.
 1.15 Dirac Delta Function 89

They satisfy the orthogonality relations

b
ϕm (x)ϕn (x) dx = δmn , (1.187)
a

where the interval (a, b) may be infinite at either end or both ends. [For convenience we
assume that ϕn has been defined to include (w(x))1/2 if the orthogonality relations contain
an additional positive weight function w(x).] We use the ϕn to expand the delta function
as
∞

δ(x − t) = an (t)ϕn (x), (1.188)
n=0

where the coefficients an are functions of the variable t. Multiplying by ϕm (x) and inte-
grating over the orthogonality interval (Eq. (1.187)), we have

b
am (t) = δ(x − t)ϕm (x) dx = ϕm (t) (1.189)
a

or
∞

δ(x − t) = ϕn (t)ϕn (x) = δ(t − x). (1.190)
n=0

This series is assuredly not uniformly convergent (see Chapter 5), but it may be used as
part of an integrand in which the ensuing integration will make it convergent (compare
Section 5.5).
Suppose we form the integral F (t)δ(t − x) dx, where it is assumed that F (t) can be
expanded in a series of orthogonal functions ϕp (t), a property called completeness. We
then obtain
∞
 ∞

F (t)δ(t − x) dt = ap ϕp (t) ϕn (x)ϕn (t) dt
p=0 n=0
∞

= ap ϕp (x) = F (x), (1.191)
p=0

the cross products ϕp ϕn dt (n = p) vanishing by orthogonality (Eq. (1.187)). Referring

back to the definition of the Dirac delta function, Eq. (1.171b), we see that our series
representation, Eq. (1.190), satisfies the defining property of the Dirac delta function and
therefore is a representation of it. This representation of the Dirac delta function is called
closure. The assumption of completeness of a set of functions for expansion of δ(x − t)
yields the closure relation. The converse, that closure implies completeness, is the topic of
Exercise 1.15.16.
90 Chapter 1 Vector Analysis

Integral Representations for the Delta Function

Integral transforms, such as the Fourier integral
∞
F (ω) = f (t) exp(iωt) dt
−∞

of Chapter 15, lead to the corresponding integral representations of Dirac’s delta function.
For example, take

sin n(t − x) 1 n

δn (t − x) = = exp iω(t − x) dω, (1.192)
π(t − x) 2π −n

using Eq. (1.175). We have

∞
f (x) = lim f (t)δn (t − x) dt, (1.193a)
n→∞ −∞

where δn (t − x) is the sequence in Eq. (1.192) defining the distribution δ(t − x). Note that
Eq. (1.193a) assumes that f (t) is continuous at t = x. If we substitute Eq. (1.192) into
Eq. (1.193a) we obtain

1 ∞ n

f (x) = lim f (t) exp iω(t − x) dω dt. (1.193b)
n→∞ 2π −∞ −n

Interchanging the order of integration and then taking the limit as n → ∞, we have the
Fourier integral theorem, Eq. (15.20).
With the understanding that it belongs under an integral sign, as in Eq. (1.193a), the
identification

1 ∞

δ(t − x) = exp iω(t − x) dω (1.193c)
2π −∞

provides a very useful integral representation of the delta function.

When the Laplace transform (see Sections 15.1 and 15.9)
∞
Lδ (s) = exp(−st)δ(t − t0 ) = exp(−st0 ), t0 > 0 (1.194)
0

is inverted, we obtain the complex representation

1 γ +i∞

δ(t − t0 ) = exp s(t − t0 ) ds, (1.195)
2πi γ −i∞

which is essentially equivalent to the previous Fourier representation of Dirac’s delta func-
tion.
 1.15 Dirac Delta Function 91

Exercises
1.15.1 Let
 1
 0, x < − 2n ,
δn (x) = n, 1 1
− 2n < x < 2n ,
 1
0, 2n < x.

Show that
∞
lim f (x)δn (x) dx = f (0),
n→∞ −∞

assuming that f (x) is continuous at x = 0.

1.15.2 Verify that the sequence δn (x), based on the function

0, x < 0,
δn (x) =
ne−nx , x > 0,
is a delta sequence (satisfying Eq. (1.178)). Note that the singularity is at +0, the posi-
tive side of the origin.
Hint. Replace the upper limit (∞) by c/n, where c is large but finite, and use the mean
value theorem of integral calculus.
1.15.3 For
n 1
δn (x) = · ,
π 1 + n2 x 2
(Eq. (1.174)), show that
∞
δn (x) dx = 1.
−∞

1.15.4 Demonstrate that δn = sin nx/πx is a delta distribution by showing that

∞ sin nx
lim f (x) dx = f (0).
n→∞ −∞ πx
Assume that f (x) is continuous at x = 0 and vanishes as x → ±∞.
Hint. Replace x by y/n and take lim n → ∞ before integrating.
1.15.5 Fejer’s method of summing series is associated with the function

1 sin(nt/2) 2
δn (t) = .
2πn sin(t/2)
Show that δn (t) is a delta distribution, in the sense that

1 ∞ sin(nt/2) 2
lim f (t) dt = f (0).
n→∞ 2πn −∞ sin(t/2)
92 Chapter 1 Vector Analysis

1.15.6 Prove that

 1
δ a(x − x1 ) = δ(x − x1 ).
a
Note. If δ[a(x − x1 )] is considered even, relative to x1 , the relation holds for negative a
and 1/a may be replaced by 1/|a|.
1.15.7 Show that
 
δ (x − x1 )(x − x2 ) = δ(x − x1 ) + δ(x − x2 ) /|x1 − x2 |.
Hint. Try using Exercise 1.15.6.
2x2
1.15.8 Using the Gauss error curve delta sequence (δn = √n
π
e−n ), show that
d
δ(x) = −δ(x),
x
dx
treating δ(x) and its derivative as in Eq. (1.179).
1.15.9 Show that
∞
δ ′ (x)f (x) dx = −f ′ (0).
−∞
Here we assume that f ′ (x) is continuous at x = 0.
1.15.10 Prove that
 

  df (x) −1

δ f (x) =   δ(x − x0 ),
dx x=x0
where x0 is chosen so that f (x0 ) = 0.
Hint. Note that δ(f ) df = δ(x) dx.
1.15.11 Show that in spherical polar coordinates (r, cos θ, ϕ) the delta function δ(r1 − r2 ) be-
comes
1
δ(r1 − r2 )δ(cos θ1 − cos θ2 )δ(ϕ1 − ϕ2 ).
r12
Generalize this to the curvilinear coordinates (q1 , q2 , q3 ) of Section 2.1 with scale fac-
tors h1 , h2 , and h3 .
1.15.12 A rigorous development of Fourier transforms31 includes as a theorem the relations
2 x2 sin ax
lim f (u + x) dx
a→∞ π x x
1


 f (u + 0) + f (u − 0), x1 < 0 < x2

f (u + 0), x1 = 0 < x2
=

 f (u − 0), x1 < 0 = x2

0, x1 < x2 < 0 or 0 < x1 < x2 .
Verify these results using the Dirac delta function.
31 I. N. Sneddon, Fourier Transforms. New York: McGraw-Hill (1951).
 1.15 Dirac Delta Function 93

1
FIGURE 1.41 2 [1 + tanh nx] and the Heaviside unit step
function.

1.15.13 (a) If we define a sequence δn (x) = n/(2 cosh2 nx), show that
∞
δn (x) dx = 1, independent of n.
−∞

(b) Continuing this analysis, show that32

x 1
δn (x) dx = [1 + tanh nx] ≡ un (x),
−∞ 2


0, x < 0,
lim un (x) =
n→∞ 1, x > 0.
This is the Heaviside unit step function (Fig. 1.41).
1.15.14 Show that the unit step function u(x) may be represented by
1 1 ∞ dt
u(x) = + P eixt ,
2 2πi −∞ t
where P means Cauchy principal value (Section 7.1).
1.15.15 As a variation of Eq. (1.175), take
1 ∞
δn (x) = eixt−|t|/n dt.
2π −∞

Show that this reduces to (n/π)1/(1 + n2 x 2 ), Eq. (1.174), and that

∞
δn (x) dx = 1.
−∞

Note. In terms of integral transforms, the initial equation here may be interpreted as
either a Fourier exponential transform of e−|t|/n or a Laplace transform of eixt .

32 Many other symbols are used for this function. This is the AMS-55 (see footnote 4 on p. 330 for the reference) notation: u for
unit.
94 Chapter 1 Vector Analysis

1.15.16 (a) The Dirac delta function representation given by Eq. (1.190),
∞

δ(x − t) = ϕn (x)ϕn (t),
n=0
is often called the closure relation. For an orthonormal set of real functions,
ϕn , show that closure implies completeness, that is, Eq. (1.191) follows from
Eq. (1.190).
Hint. One can take

F (x) = F (t)δ(x − t) dt.

(b) Following the hint of part (a) you encounter the integral F (t)ϕn (t) dt. How do
you know that this integral is finite?
1.15.17 For the finite interval (−π, π) write the Dirac delta function δ(x − t) as a series of
sines and cosines: sin nx, cos nx, n = 0, 1, 2, . . . . Note that although these functions
are orthogonal, they are not normalized to unity.
1.15.18 In the interval (−π, π), δn (x) = √n exp(−n2 x 2 ).
π

(a) Write δn (x) as a Fourier cosine series.

(b) Show that your Fourier series agrees with a Fourier expansion of δ(x) in the limit
as n → ∞.
(c) Confirm the delta function nature of your Fourier series by showing that for any
f (x) that is finite in the interval [−π, π] and continuous at x = 0,
π 
f (x) Fourier expansion of δ∞ (x) dx = f (0).
−π

1.15.19 (a) Write δn (x) = √nπ exp(−n2 x 2 ) in the interval (−∞, ∞) as a Fourier integral and
compare the limit n → ∞ with Eq. (1.193c).
(b) Write δn (x) = n exp(−nx) as a Laplace transform and compare the limit n → ∞
with Eq. (1.195).
Hint. See Eqs. (15.22) and (15.23) for (a) and Eq. (15.212) for (b).
1.15.20 (a) Show that the Dirac delta function δ(x − a), expanded in a Fourier sine series in
the half-interval (0, L), (0 < a < L), is given by
∞    
2 nπa nπx
δ(x − a) = sin sin .
L L L
n=1
Note that this series actually describes
−δ(x + a) + δ(x − a) in the interval (−L, L).
(b) By integrating both sides of the preceding equation from 0 to x, show that the
cosine expansion of the square wave

0, 0x<a
f (x) =
1, a < x < L,
 1.16 Helmholtz’s Theorem 95

is, for 0  x < L,

∞   ∞    
21 nπa 21 nπa nπx
f (x) = sin − sin cos .
π n L π n L L
n=1 n=1

(c) Verify that the term

∞  
21 nπa   1 L
sin is f (x) ≡ f (x) dx.
π n L L 0
n=1

1.15.21 Verify the Fourier cosine expansion of the square wave, Exercise 1.15.20(b), by direct
calculation of the Fourier coefficients.
1.15.22 We may define a sequence

n, |x| < 1/2n,
δn (x) =
0, |x| > 1/2n.

(This is Eq. (1.172).) Express δn (x) as a Fourier integral (via the Fourier integral theo-
rem, inverse transform, etc.). Finally, show that we may write
1 ∞
δ(x) = lim δn (x) = e−ikx dk.
n→∞ 2π −∞

1.15.23 Using the sequence

n

δn (x) = √ exp −n2 x 2 ,
π
show that
1 ∞
δ(x) = e−ikx dk.
2π −∞

Note. Remember that δ(x) is defined in terms of its behavior as part of an integrand —
especially Eqs. (1.178) and (1.189).
1.15.24 Derive sine and cosine representations of δ(t − x) that are comparable to the exponential
representation, Eq. (1.193c).
2 ∞ 2 ∞
ANS. π 0 sin ωt sin ωx dω, π 0 cos ωt cos ωx dω.

1.16 HELMHOLTZ’S THEOREM

In Section 1.13 it was emphasized that the choice of a magnetic vector potential A was not
unique. The divergence of A was still undetermined. In this section two theorems about the
divergence and curl of a vector are developed. The first theorem is as follows:
A vector is uniquely specified by giving its divergence and its curl within a simply con-
nected region (without holes) and its normal component over the boundary.