LECTURE 15

The Delta Function

1. Denition and Examples of Distributions
Let Co (Rn ) denote the space of all C complex-valued functions on Rn with compact support. We regard n Co (R ) as an innite dimensional vector space over over C.

The space D (Rn ) of distributions on Rn is the space of all linear functionals on Co (Rn ); that is to say, an element T D (Rn ) is a (continuous) map from Co (Rn ) to C such that (15.1) T (f + g ) = T (f ) + T (g ) , , C , f, g Co (Rn ) .

I should admit at this point that Im purposely avoiding the denition of the appropriate topology on Co (Rn ); nevertheless a standard (albeit technical) denition does exist. Example 15.1. The prototypical example of a distribution is as follows.
(Rn ) C dened (Rn ) then we can associate with g the linear map Tg : Co Let g be any element of Co by

(15.2)

Tg (f ) =
Rn

f (x)g (x) dn x

The integral on the right hand side is guaranteed to converge since the integrand is smooth and has only compact support. The linearity of Tg follows from the corresponding property of convergent integrals: g (x) (f1 (x) + f2 (x)) dn x =
Rn Rn

g (x)f1 (x) dn x +
Rn

g (x)f2 (x) dn x

In this way every element of Co (Rn ) can be regarded as an element of D (Rn ).

But D (Rn ) includes much more. If g is any piecewise continuous function (actually the analog of such objects in the multivariable case) then the right hand side of (??) still makes sense (all we need for convergence is for f to have compact support and this is guaranteed by the fact that we pick f from Co (Rn )).
Example 15.2. Let xo Rn . It is easy to see that the map Txo : Co (Rn ) C dened by

(15.3) is a linear functional on

Co

Txo (f ) = f (xo ) (R ).
n

Similarly, let xo be any point of Rn and let i {1, 2, 3, . . . , n}. One easily veries that the map Txo ,n : Co (Rn ) C dened by (15.4) Txo ,i (f ) = f xi
xo

is also a (continuous) linear functional on Co (Rn ) and hence a distribution.

2. THE DELTA FUNCTION

2. The Delta Function Consider the following 1-parameter family of functions a : R R. (15.5) a (x) =
1 2a

; ;

|x| < a |x| a .

So long as a = 0, this function is well dened and, in fact, piecewise continuous as a function of x for all x R. Now let f (x) be a continuous function on the real line and consider the integral
+ a

(15.6)

f (x) a (x)dx =
a

f ( x) dx 2a
+a

This again is well dened for all a = 0, so we can ask if the limit
+ a0

lim

f (x) a (x)dx = lim

a0

f (x) dx 2a

exists. By the Mean Value Theorem, for any nite a there exists an x1 (a, a) such that f (x1 ) = When one takes the limit a 0, x1 0, and so
+ +a

1 2a

f (x)dx
a

(15.7) Writing (15.8) we then have

a0

lim

f (x) a (x)dx = lim

a0

f (x) dx = f (0) 2a

(x) = lim a (x) ,

a0 +

(15.9)

f (x) (x)dx = f (0)

or, more generally,

+

(15.10)

f (x) (x xo )dx = f (xo ) .

(We simply made a change of variables x x xo in passing from (??) to (??).) The function (x xo ) so dened is called the Dirac delta function. Comparing (??) with (??) we see that the distribution dened by integrating (x xo ) against a function coincides with the distribution Txo on R. However, recalling the original denition (1) of a (x), it is clear that the denition (7.7) (15.11)
a0

lim a (x) =

if x = 0 if x = 0

does not really lead to any legitimate function. Of course, the limit (??) still makes sense. And so a more accurate denition of the delta function might be as follows: Let Ta (xxo ) be the distribution dened by (15.12) Ta (xxo ) (f ) = f (x) a (x xo ) dn x
Rn

then the Dirac delta distribution T(xxo ) is dened as T(xxo ) = lim Ta (xxo )
a0

Nevertheless, it is most common to regard the distributions, like the Dirac delta function, as some kind of generalized function (in fact, in the early literature, that was precisely what distributions were called),

3. OTHER REPRESENTATIONS OF THE DELTA FUNCTION

and to represent their action on functions via (pseudo-) integral notation (e.g., the right hand side of (??). rather than functional notation (resp., the left hand side of (??).. I should remark that like C functions, distributions can also be dierentiated; more precisely, if T D (Rn ) then i T is the distribution dened by (15.13) i T (f ) T f xi ,
f C0 ( Rn )

In the more common (and more abusive) notation in which one represents a distribution as integration against some generalized function g (x), the above denition takes the form (15.14)
Rn

g (x) f (x) dn x xi

g (x)
Rn

f xi

dn x

which can be (mis-) interpreted as the usual integration by parts formula (there is no boundary term since by denition f C0 (Rn ) has only compact support). I might also remark that it is possible to take the Laplace transform of the delta function: for the right hand side of
+

L [ (x xo )] (s)

(x xo ) esx dx = esxo

is perfectly well-dened.

3. Other Representations of the Delta Function Consider the expressions (15.15) (15.16) (15.17) (15.18)
2 2 n lim en x 1 n (x) = lim n 1 + n2 x2 sin(nx) (x) = lim n x

(x)

( x)

lim

eixt dt
n

Each of these expressions behaves like function (x) in (??) in the sense that (15.19) and
+

(x) =

if x = 0 if x = 0

(15.20)

f (x) (x)dx = f (0)

if f is continuous and the limits in (??) are taken only after the integration over x has been carried out. Here is yet another example of a representation of the delta function. Recall that the solution to (15.21) (15.22) 2 a2 2 t x (x, 0) = = 0 f (x)

was found (using Laplace transform techniques) to be

+

(15.23)

(x, t) =

f ( )g (x , t) d

3. OTHER REPRESENTATIONS OF THE DELTA FUNCTION

where x2 1 exp 2 . 4a t 2a t One can interprete the integral on the right hand side of (??) in the following way: g (x , t) represents the contribution to the total temperature at the point x and time t resulting from the propagation of the heat from the point at time t = 0. Indeed, integral kernels like g (x , t) that allow one to convert boundary conditions directly to solutions are often called propagators. (15.24) g (x, t) = Consider g (x, t) as a 1-parameter family of functions: gt (x) := g (x, t) | t R+ One can easily verify that
+

gt (x)dx = 1 ,

t R+ , , , x=0 x=0

t0

lim g (x, t)) =

+ t0

and that lim

f (x)g (x, t)dx = f (0)

for any continuous function f . In other words,

t0

lim g (x, t) = (x) .

+

This must be so since the original boundary condition on the solution (??) requires f (x) = (x, 0) = lim
t0

f ( )g (x , t) d

Here is yet another way of representing the delta function. Suppose { n (x)} is a complete orthonormal set of functions on the interval (a, b). Set

(x y ) =
n=1

n (x) n (y ) .

Then if f (x) =

am m (x)
m=1

we have
b

(15.25)
a

f (x) (x y )dx

=
m=1 n=1

am m (x) n (x) n (y )dx am m,n n (y )

m=1 n=1

(15.26) (15.27) (15.28)

= =
m=1

am m (y )

= f (y ) .

This representation will be very important later on when we relate it to Greens functions for Sturm-Liouville problems.