MASSACHUSETTS INSTITUTE OF TECHNOLOGY 8.323 LECTURE NOTES 4, SPRING 2008: Dirac Delta Function as a Distribution p.

2
Physics Department
8.323: Relativistic Quantum Field Theory I the area under g(x) must be less than the total area of the boxes. But the total
area of the boxes is given by a geometric series,
Prof. Alan Guth March 13, 2008
1 1
A =
+
+
+ . . . ≤ 2
. (4.3)
INFORMAL NOTES 2 4
DIRAC DELTA FUNCTION AS A DISTRIBUTION Since
can be chosen as small as one likes, the area under the limit function g(x)
must be zero.
Why the Dirac Delta Function is not a Function: This result does not contradict the statement that the area under gσ (x) is 1
The Dirac delta function δ(x) is often described by considering a function that for any σ > 0. Rather, this is a case where the limit of an integral is not the same
has a narrow peak at x = 0, with unit total area under the peak. In the limit as the as the integral of the limit of the integrand. The integral has the value 1 for every
σ > 0, so the limit of the integral as σ → 0 is 1. However, if one takes the limit of
peak becomes infinitely narrow, keeping fixed the area under the peak, the function
the integrand first, and then integrates, the answer is zero.
is sometimes said to approach a Dirac delta function. One example of such a limit
is Dirac Delta Function as a Distribution:
g(x) ≡ lim gσ (x) , (4.1) A Dirac delta function is defined to have the property that
σ→0

where
∞
1 1 2 2
dx ϕ(x) δ(x − a) ≡ ϕ(a) .
gσ (x) ≡ √ e− 2 x /σ . (4.2) (4.4)
2π σ −∞

The area under gσ (x) is 1, for any value of σ > 0, and gσ (x) approaches 0 as σ → 0
for any x other than x = 0. But we have just seen that there is no function δ(x) which has this property, as long
as integration is defined by the area under a curve. However, there is no problem
However, it was pointed out long ago that the delta function cannot be rigor- defining a distribution that behaves this way. Using the notation we defined earlier
ously defined this way. The function g(x) is equal to zero for any x = 0, and is for distributions, we define
infinite at x = 0; it can be shown that any such function integrates to zero. To see
this, define the integral as the area under the curve, and consider the construction: Tδ(x−a) [ϕ] ≡ ϕ(a) . (4.5)

Here Tδ(x−a) is the name that we will use for the distribution that acts like integra-
tion over the delta function δ(x − a), and ϕ(x) is the test function. This is certainly
a linear functional that is defined on all Schwartz functions, and therefore is a tem-
pered distribution. We interpret Eqs. (4.4) and (4.5) as meaning the same thing,
where the second form emphasizes the definition as a distribution, and the first form
emphasizes that distributions can be viewed as a generalized kind of function, with
a notation that makes them look like functions. That is, we define

∞
dx ϕ(x) δ(x − a) ≡ Tδ(x−a) [ϕ] ≡ ϕ(a) . (4.6)
−∞

In this picture the vertical axis is entirely encased in rectangles, each of which has To put it differently, we must remember that an integral over a delta function, such
height 1. The width of the rectangles vary, with the lowest rectangle having width as Eq. (4.4), is not defined as a standard integral— instead it is symbolic integral,

, for some
> 0, and each successive rectangle has half the width of the rectangle which is defined as a distribution which maps the function that multiplies the delta
below. Note that the outline of the boxes is everywhere above the curve g(x), so function to its value at the point where the argument of the delta function vanishes.
8.323 LECTURE NOTES 4, SPRING 2008: Dirac Delta Function as a Distribution p. 3 8.323 LECTURE NOTES 4, SPRING 2008: Dirac Delta Function as a Distribution p. 4


What about dp e−ip(x−a) ? If we write these distributions as symbolic integrals, then the above equation
becomes
∞
∞
If we interpret this integral in the sense of Riemann or Lebesgue, it simply does
not exist — it diverges. As a distribution, however, we can interpret dx I(x)ϕ(x) = 2π dx δ(x − a)ϕ(x) , (4.15)
−∞ −∞

where we remembered from Eq. (4.9) that T̃f [ϕ] is equal to the distribution corre-
∞
sponding to I(x). Since Eq. (4.15) holds for an arbitrary test function ϕ(x), we can
I(x) ≡ dp e−ip(x−a) (4.7) equate the distributions without showing their arguments:
−∞

∞

as the Fourier transform of I(x) = dp e−ip(x−a) = 2πδ(x − a) . (4.16)

ipa −∞
f (p) = e . (4.8)
Since I(x) is the Fourier transform of the distribution Tf [ϕ], it must itself be a The Derivative of a Delta Function:
well-defined distribution. Using the notation of distributions, the distribution cor- If a Dirac delta function is a distribution, then the derivative of a Dirac delta
responding to I is then given by function is, not surprisingly, the derivative of a distribution. We have not yet
defined the derivative of a distribution, but it is defined in the obvious way. We
TI [ϕ] = T̃f [ϕ] . (4.9) first consider a distribution corresponding to a function, and ask what would be the
distribution corresponding to the derivative of the function.
To evaluate the distribution TI [ϕ], we first write the distribution corresponding Starting with a well-behaved (i.e., piecewise continuous and bounded by some
to function f (p) in the usual way: power of t) function f (t), we defined the corresponding distribution by

∞

∞
∞ Tf [ϕ] ≡ dt f (t)ϕ(t) . (4.17)
Tf [ϕ] = dp f (p) ϕ(p) = dp eipa ϕ(p) . (4.10) −∞
−∞ −∞
Then if we write the distribution corresponding to df /dt, we get

∞
The Fourier transform of this distribution is then defined by applying the same df
Tdf /dt [ϕ] = dt ϕ(t) (4.18)
distribution to the Fourier transform of the test function, so −∞ dt

∞ Since f (t) is bounded for large |t| by a power of t, and ϕ(t) falls off faster than any
T̃f [ϕ] ≡ Tf [ϕ̃] = dp eipa ϕ̃(p) . (4.11) power, we can integrate by parts without encountering a surface term:
−∞
∞  
dϕ dϕ
Tdf /dt [ϕ] = − dt f (t) = −Tf . (4.19)
But the inverse Fourier transform is given by −∞ dt dt

∞ This result can then be taken as the general definition of the derivative of a distri-
1 bution:
ϕ(x) = dp eipx ϕ̃(p) , (4.12)
2π −∞  
dϕ
T  [ϕ] ≡ −T . (4.20)
so by comparing the two formulas above, one sees that dt

T̃f [ϕ] = 2πϕ(a) . (4.13) Applying this result to a delta function in the notation of Eq. (4.6), it looks exactly
like integration by parts:
But this is exactly the definition of the delta function distribution, as given in


Eq. (4.6), so ∞
dδ(t − a) df 
dt f (t) =− . (4.21)
−∞ dt dt t=a
T̃f [ϕ] = 2π Tδ(x−a) [ϕ] . (4.14)