Physics 215 Winter 2018

Integral representation of the Heavyside step function

The Heavyside step function is defined as,

(
1 , if k > 0 ,
Θ(k) = (1)
0 , if k < 0 .

Although the value of Θ(k) is not defined at k = 0, we shall nevertheless demand that

Θ(k) + Θ(−k) = 1 , (2)

should be satisfied for all real values of k, including the origin, k = 0. The Heavyside step
function is related to the Dirac delta function by differentiation,

dΘ(k)
δ(k) = . (3)
dk
The goal of these notes is to express the step function as a Fourier transform,
Z ∞
1
Θ(k) = eikx f (x) dx , (4)
2π −∞

where the function f (x) is to be determined.1

The function f (x) is determined by the inverse Fourier transform,
Z ∞
f (x) = e−ikx Θ(k) dk . (5)
−∞

This integral is not well defined. However it can be re-interpreted in the sense of distributions.
What this phrase really means is that quantities are treated as generalized functions (also
called distributions), which make sense only when integrated against test functions that are
smooth, regular, and vanish sufficiently fast at ±∞. We can evaluate the f (x) using the
following trick. Note that
Z ∞ Z ∞
−ikx dΘ(k)
1= e δ(k) dk = e−ikx dk . (6)
−∞ −∞ dk

We now integrate by parts. We can set the surface term to zero by employing

lim e−ikx = 0 , (7)

k→±∞

1
These notes are based on a derivation given on p. 151 of Ram P. Kanwal, Generalized Functions: Theory
and Applications, 3rd edition (Birkhäuser, Boston, 2004).

1
where the limit is interpreted in the sense of distributions (as mention in the class handout,
The Riemann-Lebesgue Lemma). It then follows that
Z ∞ Z ∞
d −ikx
1=− Θ(k) e dk = ix Θ(k)e−ikx dk = ixf (x) . (8)
−∞ dk −∞

To solve eq. (8), let us define h(x) ≡ if (x) and consider the equation

xh(x) = 1 . (9)

The solution to this equation for x 6= 0 is clearly h(x) = 1/x. But, how should we deal with
x = 0? The answer is again to appeal to generalized functions. That is, eq. (9) should be
interpreted as Z ∞ Z ∞
xh(x)g(x) dx = g(x) dx , (10)
−∞ −∞

for any smooth regular test function g(x) that vanishes sufficiently fast at ±∞.
The most general solution to the inhomogeneous equation, xh(x) = 1, must be of the
form,
h(x) = hp (x) + hh (x) , (11)
where hp (x) is a particular solution that satisfies xhp (x) = 1 and hh (x) is the solution to the
homogeneous equation, xhh (x) = 0. I claim that one choice for the particular solution to
eq. (9) is,
1
hp (x) = P , (12)
x
where P indicates the Cauchy principal value prescription when integrated against a test
function, g(x),
Z ∞ Z −δ Z ∞ 
g(x) dx g(x) dx g(x) dx
P ≡ lim + , (13)
−∞ x δ→0 −∞ x δ x

with δ > 0.
Let us check that h(x) = hp (x) given by eq. (12) provides a solution to eq. (10). It is
sufficient to observe that,
Z ∞ Z ∞
1
P x g(x) dx = g(x) dx , (14)
−∞ x −∞

where the P symbol can be dropped on the right hand side of eq. (14) since the corresponding
integral is now well defined. Hence, it follows that (in the sense of distributions),
1
xP = 1, (15)
x
and eq. (12) is verified.
We now turn to the most general solution to the homogeneous equation,

xhh (x) = 0 . (16)

2
We shall solve eq. (16) using a Fourier transform technique. We first write
Z ∞
1
hh (x) = eikx q(k) dk . (17)
2π −∞
Inverting the Fourier transform yields
Z ∞
q(k) = e−ikx hh (x) dx . (18)
−∞

We now take the derivative of q(k) with respect to k to obtain,

Z ∞
dq
= −i e−ikx xhh (x) dx = 0 . (19)
dk −∞

where we have used eq. (16) in the final step. The most general solution to the differential
equation dq/dk = 0 is q(k) = C, where C is an arbitrary constant.2 Inserting this solution
back into eq. (17), we end up with
C ∞ ikx
Z
hh (x) = e q(k) dk = C δ(x) , (20)
2π −∞
after employing the integral representation of the delta function,
Z ∞
1
δ(x) = e−ikx dk . (21)
2π −∞

It is simple to check the validity of eq. (20). In particular,

Z ∞
xδ(x)g(x) dx = xg(x) = 0, (22)
−∞ x=0

where again we have used the fact that g(x) is a smooth regular function. It then follows
that (in the sense of distributions),
xδ(x) = 0 . (23)
Hence, xhh (x) = Cxδ(x) = 0 as required. Combining the results of eqs. (12) and (20), one
obtains the most general solution of eq. (9),
1
h(x) = hp (x) + hh (x) = P + Cδ(x) . (24)
x
Returning to eq. (8), it follows in light of eq. (24) that
Z ∞
1
h(x) = if (x) = i Θ(k)e−ikx dk = P + Cδ(x) , (25)
−∞ x
2
This statement is trivial if solutions are restricted to the space of ordinary functions. Nevertheless,
q(k) = C is still the unique solution of dq/dk = 0 even if the solution space is expanded to included
generalized functions. A proof of this assertion can be found on pp. 39–41 of I.M. Gel’fand and G.E. Shilov,
Generalized Functions, Volume 1: Properties and Operations (Academic Press, New York, NY, 1964).

3
where the constant C is still yet to be determined. To fix the constant C we proceed as
follows. Replacing x → −x in eq. (25) yields,
Z ∞ Z ∞
1 ikx
−P + Cδ(x) = i Θ(k)e dk = i Θ(−k)e−ikx dk , (26)
x −∞ −∞

after noting that δ(−x) = δ(x) and changing the integration variable from k to −k. Adding
eqs. (25) and (26) and using eq. (2), we end up with
Z ∞ Z ∞
 −ikx
e−ikx dk .

2Cδ(x) = i Θ(k) + Θ(−k) e dk = i (27)
−∞ −∞

Finally, using the integral representation of the delta function [cf. eq. (21)], we conclude
that C = iπ. We now insert this value of C into eq. (25) and employ the Sokhotski-Plemelj
formula [cf. the class handout entitled, The Sokhotski-Plemelj Formula],
1 1
lim = P + iπδ(x) . (28)
ε→0 x − iǫ x
The end result is ∞
1
Z
if (x) = i Θ(k)e−ikx dk = lim . (29)
−∞ ε→0 x − iε

Returning to eq. (4), we conclude that3

∞
1 eikx
Z
Θ(k) = lim dx . (30)
2πi ε→0 −∞ x − iε
It is amusing to revisit eq. (25) with C = iπ, which yields the noteworthy result,
Z ∞
1
i Θ(k)e−ikx dk = P + iπδ(x) . (31)
−∞ x
In particular, using the definition of the Heavyside step function, it follows that
Z ∞
1
e−ikx dk = −i P + πδ(x) . (32)
0 x
Of course eq. (32) must be interpreted in the sense of distributions (since the integral does
not converge in the usual sense4 ). Moreover, taking real and imaginary parts of eq. (32)
yields,
Z ∞
cos kx dk = πδ(x) , (33)
0

∞
1
Z
sin kx dk = P . (34)
0 x
3
In problem 4(a) of Problem Set 1, you were given the right hand side of eq. (30) and asked to evaluate
the integral using the residue theorem of complex analysis (with suitably chosen closed contours for the two
cases of k > 0 and k < 0). The result of such a computation is, of course, Θ(k).
4
Compare this with the integral in eq. (21), which does not converge in the usual sense, but nevertheless
provides an integral representation of the delta function in the sense of distributions.

4
Once again, the integrals of eqs. (33) and (34) must be interpreted in the sense of distribu-
tions. For example, one can check that eq. (33) is a consequence of the integral representation
of the delta function,
Z ∞ Z 0 Z ∞ Z ∞ Z ∞
ikx ikx ikx
 ikx −ikx

2πδ(x) = e dk = e dk + e dk = e +e dk = 2 cos kx dk ,
−∞ −∞ 0 0 0
(35)
after changing the integration variable, k → −k, in the second integral above. Hence, we
have recovered eq. (33).
One additional consequence of eq. (31) can be extracted if we invert the Fourier transform,
Z ∞
1  1
eikx P + iπδ(x) dx = iΘ(k) .

(36)
2π −∞ x
Using the Cauchy principal value prescription and integrating over the delta function yields
Z ∞ ikx
1 e  
P dx = i Θ(k) − 21 . (37)
2π −∞ x
It is convenient to introduce the sign function,
(
+1 , for k > 0,
sgn(k) = Θ(k) − Θ(−k) = (38)
−1 , for k < 0,
which satisfies
d d 
sgn(k) = Θ(k) − Θ(−k) = 2δ(k) , (39)
dk dk
in light of eq. (3). Using eq. (2), it follows that
   
Θ(k) − 21 = Θ(k) − 21 Θ(k) + Θ(−k) = 12 Θ(k) − Θ(−k) = 1
2
sgn(k) . (40)
Inserting this result into eq. (37) yields,
Z ∞ ikx
e
P dx = iπ sgn(k) . (41)
−∞ x

Taking the real and imaginary parts of eq. (41) yields,5

Z ∞
cos kx
P dx = 0 , (42)
−∞ x
Z ∞
sin kx
dx = π sgn(k) . (43)
−∞ x
Note that the vanishing of the integral in eq. (42) is due to the fact that the integrand is an
odd function of x, which integrates to zero due to the Cauchy principal value prescription.
The P symbol is not needed in eq. (43) since the corresponding integrand has a finite limit
as k → 0.
As one final check, let us take the derivative of eq. (41) with respect to k and employ
eq. (39). This yields (once again) the integral representation of the delta function (where
the P symbol can be dropped as the resulting integrand is not singular at x = 0).
5
The result of eq. (43) is well-known and is often obtained using the residue theorem of complex analysis
(and a suitably chosen closed contour).