arXiv:funct-an/9510004v1 4 Oct 1995

ON DIRACS DELTA CALCULUS

S ergio F. Cortizo

Instituto de Matem atica e Estat stica, Universidade de S ao Paulo Cidade Universit aria, Rua do Mat ao, 1010 05508-900, S ao Paulo, SP, Brasil cortizo@ime.usp.br

Abstract It is shown that theories already presented as rigorous mathematical formalizations of widespread manipulations of Diracs delta function are all unsatisfactory, and a new alternative is proposed. PACS 02.90.+p

I. Introduction There is a myth in Mathematical Physics according to which: The widespread manipulations of Diracs delta function can be rigorously formalized by the Theory of Distributions. It is easy to verify, however, that this statement can be taken as true only by someone who does not know the nature of such manipulations or by someone who does not know the available theories about the concept of distribution. The most elementary manipulations of the delta function, and also the most often found in literature, involve the composition of and a real function. For example: ( x2 a 2 ) = or, more generally:
n

1 [ (x a) + (x + a)] 2a

(a > 0),

[g (x)] =
i=1

1 (x ai ), |g (ai )|

where a1 , a2 , . . . , an are the simple roots of a real function g . Considering the delta function as a distribution according to the best known denition of this concept 1 linear functionals on some test functions space the above formulae could not be written as such, since the composite of a distribution and a generic real function simply cannot be done. Using the sequential approach 2 we can dene the composition of a distribution and them. Therefore, also in this context, not even the rst formula above could have been written. It is clear that we can use the above equality to dene the symbol [g (x)] , but in any way could this be considered a formalization of the relatively simple idea which one has in mind when writing it down. It is obvious that the formal meaning ascribed to the basic and elementary manipulations of Delta Calculus must be related to the mental process usually understood in these manipulations. In this sense, the sequential approaches to the concept of distribution are much superior to that of Schwartz-Sobolev, which systematically uses the theorems of Delta Calculus as denitions of the involved concepts. For instance: the distributions are dened as linear functionals on some test functions space so we can characterize the delta 2 some very simple real functions, but the function which sends x to (x2 a2 ) is not among

function by its sifting property:

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

That procedure, considered elegant by many mathematicians, merely dismisses the fact that the sifting property itself is a basic result of the Delta Calculus to be formally proved. Dirac has used a simple argument, based on the integration by parts formula, to get the sifting property of the derivative of the delta function:

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

The theory of distributions as linear functionals, instead of dening the integral of a distribution and so proving that it satises some kind of integration by parts formula, just uses the formula deduced by Dirac for the delta function:

(x)f (x) dx =

(x)f (x) dx

as distributions derivative denition. Proceeding systematically in this fashion, it is not surprising that some unnecessary constructions must be ignored after certain symbols and operations have been dened, because we know Delta Calculus has a nontrivial mathematical content. The real surprise is the consideration of this Theory of Distributions as a satisfactory formalization of Delta Calculus, despite its ignoring the basic operational rules of this Calculus. All proposed formalizations of Delta Calculus fail because they try to generalize the real functions without having generalized its variables. This will become clear if we compare those formalizations to the one proposed in this work. In Ref. 4 we presented an extension process that can be applied to any set. This process was applied 5 to the ordered eld of real numbers, formally dening what we call virtual numbers. Among those numbers we have innitesimal and innite quantities, which were so used in an reorganization of Dierential and Integral Calculus. The same extension process was applied 6 to the set of real functions, introducing what we call virtual functions. So, the basic techniques of Innitesimal Calculus were generalized to these mathematical objects. The concept of virtual function is very close to the original idea of improper function created by Dirac: the limit of a sequence of real functions 3 . The essential dierence 3

is that Dirac did not have access to improper numbers which could have been used as argument of these functions, whereas the virtual functions can be evaluated at (previously dened) virtual numbers, exactly as we do with real functions and numbers. We will here use many concepts and results presented in Refs. 4, 5 and 6, including the notational conventions adopted then (like the notation for reduced integrals stated in Ref. 6). This paper begins by dening, in Sec. II, the class of virtual functions having the characteristics supposed by Dirac for his delta function. In Sec. III we prove the Diracs functions sifting property. Section IV is dedicated to discussing the meaning of the operational rules which make up Delta Calculus. In Sec. V we prove the basic rule for compositions involving delta function (above). Other well known results of Delta Calculus are demonstrated in the last two sections. II. Diracs Functions The aim of this section is to characterize the virtual functions which possess the basic properties assumed by Dirac for his delta function, and to show that there exist virtual functions in F (R) F ( R ) with those properties. We will say that a virtual function F (R) is a Diracs function when: (i) is dened at any R and is non-negative: ( ) 0, (ii) is integrable with

for every R ;

(x) dx = 1 ; and

(iii) there exists a positive innitesimal R such that | | ( ) = 0. Before proceeding, it is convenient to verify that there exist virtual functions under innitely derivable which vanishes outside the interval (1, 1) , increasing between 1
1

those conditions. This is not so dicult: we know there exists a real function f : R R

and 0, where it takes the value f (0) = 1 , decreasing between 0 e 1, and with unitary integral: f (x) dx = 1.
1

From such a function f , we can dene a virtual function 0 : R R by: 0 ( ) = f ( ), i.e., 0 F (R) is the virtual function represented by the sequence (f1 , f2 , . . .) [F (R)] fn (x) = nf (nx). It is easy to see that this virtual function satises the three conditions above: 0 ( ) 0,

given by:

for every R , 0 (x) dx = 1,

and | | 0 ( ) = 0. Therefore 0 is a Diracs function. That virtual function 0 perfectly corresponds to the image given by Dirac 3 : To get a picture of (x) , take a function of the real variable x which vanishes everywhere except inside a small domain, of length say, surrounding the origin x = 0 , and which is so large inside this domain that its integral over this domain is unity. The exact shape of the function inside this domain does not matter, provided there are no unnecessarily wild variations (for example provided the function is always of order 1 ). Then in the limit 0 this function will go over into (x) . continuous), normalized in the strong sense:

Moreover, the function 0 : R R constructed above is innitely derivable (so

0 ( ) d = 1,

and, evaluated at real values of its argument, provides: 0 (x) = 0 , if x = 0; , if x = 0.

that (0) = , or even that (0) . But the condition (x) = 0 , for every real x = 0 , 5

In order for a virtual function : R R to be a Diracs function, it is not necessary

immediately follows from the existence of an innitesimal such that | | > ( ) = 0 .

For instance, the virtual function : R R given by: ( ) = is such that (x) 0 for every x = 0 : ( x) = 1 0. + x2 1 + 2 2

But ( ) = 0 , for any R , and so is not a Diracs function. (That function is the derivative of ( ) = arctan( ) , and it was used in various examples in Ref. 5.) It should be clear that there exist many distinct Diracs functions in F (R) . For

example, if 0 : R R is the function above constructed then the virtual functions + : R R and : R R given by:

+ ( ) = 0 ( 2 ), and ( ) = 0 ( + 2 ) are two other innitely dierentiable Diracs functions distinct from 0 . The function + is the class of the sequence (g1 , g2 , . . .) [F (R)] dened by: g n ( x) = f n ( x 2 ), n

class of the sequence (h1 , h2 , . . .) [F (R)] dened by: hn (x) = fn (x + 2 ). n

where fn are the above functions which represent 0 F (R) . Analogously, is the

A Diracs function like + can be convenient if we intend, for instance, to apply the Laplace transformation, since it vanishes for every < 0 , but is normalized in the sense:

+ (x) dx = 1.

Furthermore, it is easy to verify that: by: If 1 and 2 are two Diracs functions then the virtual function 3 : R R dened 3 ( ) = 1 [1 ( ) + 2 ( )] 2 6

is also a Diracs function. We have not required a Diracs function to be always continuous or derivable, so we can take square pulses like the function F (R) represented by the sequence: g n ( x) = n/2 , if |x| < 1/n; 0 , if |x| 1/n.

specic situations. It is such that:

This Diracs function, which is discontinuous at = and = , can also be useful in /2 , if | | < ; 0 , if | | .

( ) = III. Sifting Property

Our aim in this section is to demonstrate the sifting property of Diracs functions. We will say that a real function is continuous around x R when there exists a real open interval containing x in which it is dened and continuous. If : R R is a Diracs functions, and f : R R is a real function continuous

around the origin, then:

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

Proof: First, we note that there exists a positive innitesimal such that: | | ( ) = 0, for is a Diracs function. By the additivity with respect to the virtual integration interval 6 , we have:

( )f ( ) d =

( )f ( ) d +

( )f ( ) d +

( )f ( ) d.

On the right-hand side of the above equation, the integrands of the rst and third integrals vanish, for any function f . So those integrals exist and are equal to zero. In the second integral (right-hand side) we have the product of two integrable functions, so the left-hand side integral exists and:

( )f ( ) d =

( )f ( ) d.

It is clear that, in R , the following statement holds: if g and h are two real functions dened between a and a , where a is a positive real number, with g integrable and 7

non-negative and h continuous, then there exist two real numbers c1 and c2 between a and a such that:
a a a

h(c1 )
a

g (x) dx

g (x)h(x) dx h(c2 )

g (x) dx.
a

So we have, by the Virtual Extension Theorem (VET, Ref. 4), that if and are two virtual functions dened between and , where is a positive virtual number, with integrable and non-negative and continuous, then there exist two virtual numbers 1 and 2 between and such that:

( 1 )

( ) d

( ) ( ) d (2 )

( ) d.

Now making = , = and = f , we conclude that there exist two virtual numbers 1 and 2 between and such that:

f ( 1 )

( ) d

( )f ( ) d f (2 )

( ) d.

themselves. Then it follows from the continuity of f at the origin that: f (1 ) f (0) f (2 ), and therefore:

Since 1 and 2 are between the innitesimals and , they are also innitesimals

( )f ( ) d f (0).

It is important to note that the sifting property demonstrated above holds for all Diracs functions. If we are working with a particular Diracs function in a specic context then we can deduce variants of this property by analogous methods. For instance, we can use the function + from the previous section to unilaterally sift a real function f dened only for positive values of x . IV. Diracs Equivalence Delta Calculus is based on a set of operational rules, whose meaning has been established clearly by Dirac himself 3 : There are a number of elementary equations which one can write down about functions. These equations are essentially rules of manipulation for algebraic work involving 8

 functions. The meaning of any of these equations is that its two sides give equivalent results as factors in an integrand. equivalent when, for every continuous real function f : R R , we have:

We will say that two integrable virtual functions : R R and : R R are

(x)f (x) dx =

(x)f (x) dx.

It is important to note that, according to this denition, both integrals must be reducible in order for the two functions to be equivalent. That means, if : R R is such that there exists a continuous real function f : R R which makes the virtual integral

( )f ( ) d

irreducible, then is equivalent only to itself. For example, if is a Diracs function and since the integral k R a non-zero real constant, then the sum function ( + k ) is equivalent only to itself,

[ ( ) + k ] d =

( ) d +

k d 1 + 2k

is not reducible. We will indicate that two virtual functions and are equivalent by writing: (x) (x). For instance: (2x) 1 ( x) , 2

for every Diracs function (this is a simple consequence from the composition rule which will be stated and proved in the next section). We use a real variable in this notation (Latin, not Greek letter) to reinforce that the two expressions are interchangeable only in reduced integrals. virtual functions dened at any R . This relation will be called Diracs equivalence. It is not dicult to verify that is an equivalence relation on the set of integrable The above example can be used to gain a ash of intuition about the nature of this

equivalence: the graph of function (2x) is the same height as one of the function , but half of its width; whereas the graph of function (1/2) (x) is the same width as the 9

one of function , but half of its height. Thus, although fairly dierent, those functions are equivalent as factors in an integrand: 1 1 (x)f (x) dx = f (0), 2 2 since both delimit the same area innitely concentrated around the origin, whereas any (2x)f (x) dx = continuous real function f varies slowly. To discuss the equivalence of Diracs functions, let us rst consider the following denition: an integrable virtual function : R R is sifting when:

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

for any continuous real function f : R R . according to the Diracs relation.

It is not dicult to see that the set of all sifting functions is an equivalence class Every Diracs function is sifting (as we saw in the previous section), but not every

sifting virtual function is a Diracs function. As a counterexample, we can take the sequence (f1 , f2 , . . .) [F (R)] used in Sec. I to dene 0 , and alter all fn at one and only one point: g n ( x) = fn (x) , if x = 7; 3 , if x = 7.

This sequence (g1 , g2 , . . .) [F (R)] represents an integrable virtual function : R R which is clearly sifting, but not a Diracs function, for (7) = 3 . Even so, we can arm: Any two Diracs functions are always equivalent to each other. This fact makes two conventions which are part of the traditional language of Delta Calculus compatible: (i) All Diracs functions are represented by the same symbol: . (ii) The Diracs equivalence is indicated simply by the equality symbol: = . Those conventions drastically simplify notation, and certainly do not jeopardize the rigour of a scientic work if used properly. Nevertheless, since they might generate some confusion, we will continue to explicitly distinguish equality ( = ) from equivalence ( ), and to represent distinct Diracs functions by distinct symbols. The traditional language of Delta Calculus also does not establish a clear distinction between sifting functions and Diracs functions, since the latter are generally handled using the sifting property. That is another possible source of confusion, which we should keep in mind, mainly when the above conventions are used. 10

V. Composition Rule Diracs functions, as dened in Sec. II, are virtual functions F (R) with certain

specic characteristics. Therefore, it is clear that they can be composed with any other in the domain of , since Diracs functions are dened at any virtual number R . virtual function . Besides, the composite virtual function ( ) is dened at every point

Several of the operational rules of Delta Calculus involve the composition of a Diracs

function and a real function g . We looked at the meaning of those rules in the previous discuss these equivalences, we will suppose that the real function g is dened on the whole virtual extension R of the real line. Diracs functions do not vanish only around the origin, so: exists a positive r R with |g (x)| > r for every x R , then the composite function ( g ): R R is identically null: If : R R is a Diracs function and g : R R a real function for which there section: they establish Diracs equivalences for the composite virtual function ( g ) . To

[g ( )] = 0,

for every R .

In this case, it is cleat that [g ( )] 0 . For instance: (x2 + 1) 0 and (sin x + 2) 0.

Let us consider now a real function whose image approximates the origin: g (x) = x2 , for example. If we take the Diracs function dened in Sec. II: ( ) = 0 ( + 2 ), then: ( 2 ) = 0, for every R ,

since ( ) = 0 for any 0 . So it is clear that: (x2 ) 0. Sec. II) and that same function g (x) = x2 , we get a non-vanishing composite: | | < ( 2 ) = 11 . 2 On the other hand, composing the square Diracs function (also dened in

Furthermore, it easy to see that: ( 2 ) 0, for:

( 2 ) d =

d = 2 = . 2 2

So, we have: (x2 ) (x2 ), which shows that the equivalence class of this composite depends on the particular Diracs function chosen. That means, there is not a virtual function such that (x2 ) (x) for any Diracs function . Therefore, there is not an operational rule for the composite (x2 ) in the traditional language of Delta Calculus, which deals only with generic equivalences that hold for all Diracs functions. This dependence of the class of the composite with respect to the selection of the Diracs function might also occur if the graph of the real function g : R R is asymptotic of the x -axis. As an example, for g (x) = ex we have: (ex ) = 0 whereas: + (ex ) 0. However, the class of the composite function will not depend on the chosen Diracs number of roots, not approximating it in any other region: function if the graph of the real function g : R R only crosses the x -axis in a nite Let : R R be any Diracs function and g : R R a real function with a nite

for every R ,

number of roots a1 , a2 , . . . , an ; and such that there exist n + 1 positive real numbers r, r1 , r2 , . . . , rn having the following properties: (i) the n intervals [ai ri , ai + ri ] are disjoint in pairs and |g (x)| > r, for any x R outside those intervals ; and 12

derivative does not vanish. we have:

(ii) inside the n intervals [ai ri , ai + ri ] , the function g is dierentiable and its

Then, for any real function f : R R continuous on the n intervals [ai ri , ai + ri ] ,

n

[g (x)]f (x) dx =
i=1

f (ai ) . |g (ai )|

Proof: Since the composition [g ( )] vanishes outside the n intervals [ai r, ai + r ] , we have:
n ai +ri

[g ( )]f ( ) d =
i=1 ai ri

[g ( )]f ( ) d.

We will calculate the integrals in the above sum separately. First, we note that, for each inverse hi : i = 1, 2, . . . , n , the restriction of the function g to the interval [ai ri , ai + ri ] admits an hi [g (x)] = x, for every x [ai ri , ai + ri ] ,

and this inverse hi is monotonic, dierentiable and its derivative does not vanish between g (ai + ri ) and g (ai ri ) . Thus, changing variables: = g ( ), we get:
ai +ri ai +ri

= hi ()

and

d = g ( )d,

[g ( )]f ( ) d =
ai ri ai ri

[g ( )]

f ( ) g ( ) d = g ( )

g (ai +ri )

()
g (ai ri )

f [hi ()] d. g [hi ()]

If g (ai ) > 0 then g (ai ri ) < 0 < g (ai + ri ) , so the sifting property guarantees that:
g (ai +ri )

()
g (ai ri )

f [hi (0)] f (ai ) f [hi ()] d = . g [hi ()] g [hi (0)] g (ai )

On the other hand, if g (ai ) < 0 then g (ai + ri ) < 0 < g (ai ri ) , so the sifting property provides:
g (ai +ri )

()
g (ai ri )

f [hi ()] f [hi (0)] f (ai ) d = . g [hi ()] g [hi (0)] g (ai )

Thus, in any of both cases we have:

ai +ri ai ri

[g ( )]f ( ) d 13

f (ai ) . |g (ai )|

above result, so we have the well known sifting property of translated Diracs functions:

For any number a R , the function g (x) = x a satises the hypothesis of the (x a)f (x) dx = f (a),

for any f : R R continuous around a R . Diracs function : [g (x)]

n

So, we conclude that, for any real function g under the above conditions, and for any 1 |g (a
i )|

i=1

(x ai ).

This formula will be called composition rule. The sifting property of translated Diracs functions implies: f ( x) ( x a ) f ( a ) ( x a ) , for any function f : R R continuous around a R . The composition rule, on its turn, shows that: (ax) and ( x2 a 2 ) for any Diracs function . VI. Contraction of Diracs Functions Diracs functions, as dened in Sec. II, are integrable virtual functions : R R . 1 ( x) |a| (a = 0),

1 [ (x a) + (x + a)] 2a

(a > 0),

Thus, it is clear that, for any R , the virtual function which assigns ( ) to is also integrable. The result below will be used as a lemma in the following demonstration: If : R R is a Diracs function, and R a nite virtual number, then:

( ) d =

( ) d,

Proof: Changing the variables = we get:

( ) d = =

() d () d + 14

() d +

() d,

but the integrands of the above rst and third integrals vanish, for is nite and is a Diracs function. R dened by: If 1 and 2 are two continuous Diracs functions then the virtual function 3 : R 3 ( ) =

1 ( )2 ( ) d

is also a Diracs function. Besides, we have:

1 ( )2 ( ) d = 3 ( ),

for any nite R .

is non-negative. To show that 3 ( ) vanishes far from the origin, we rst note that there exist positive innitesimals 1 and 2 such that: | | 1 1 ( ) = 0 Taking 3 = 1 + 2 we get: | | 3 3 ( ) = 0. To demonstrate that 3 is integrable and normalized:

Proof: It is quite easy to see that 3 ( ) 0 , for every R , since the above integrand

and

| | 2 2 ( ) = 0.

3 ( ) d =

1 ( )2 ( ) d d = 1,

let us consider the following armation in R : if a1 and a2 are two positive real numbers; b between a2 and a2 : g1 is a real function dened and continuous between a1 and a1 , and such that for every
a1 a1 a1

g1 (x b) dx =

g1 (x) dx ;
a1

and g2 is a real function dened and continuous between a2 and a2 ; then:

a1 a1 a2 a2 a1 a2

g1 (x b)g2 (b) db dx =

g1 (x) dx
a1 a2

g2 (b) db .

This armation can easily be proved in R by Fubinis Theorem, and so extended to R by the VET: 15

If 1 and 2 are two positive virtual numbers; 1 is a virtual function dened and continuous between 1 and 1 , and such that for every between 2 and 2 :
1 1 1

1 ( ) d =

1 ( ) d ;
1

and 2 is a virtual function dened and continuous between 2 and 2 ; then:

1 1 2 2 1 2

1 ( )2 ( ) d d =

1 ( ) d
1 2

2 ( ) d .

Now, making

3 ( ) d =

2 2

1 ( )2 ( ) d d,

get:

the above lemma guarantees that taking 1 = , 2 = 2 , 1 = 1 , and 2 = 2 , we

3 ( ) d =

1 ( ) d
2

2 ( ) d 1.

To calculate the integral

1 ( )2 ( ) d

we change the variables = :

1 ( )2 ( ) d = =

1 ( )2 () d

1 ( )2 () d

= 3 ( ). This result shows that the contraction of two continuous Diracs functions is equivalent to a third Diracs function:

1 (x )2 ( a) d 3 (x a).

VII. Dierentiable Diracs Functions We will now consider a dierentiable Diracs function (derivable with continuous derivative) and demonstrate the sifting property associated to its derivative. We will 16

say that a real function is dierentiable around x R when there exists an open real interval containing x in which the function is dened and dierentiable. dierentiable around a R , then:

If : R R is a dierentiable Diracs function and f : R R a real function

(x a)f (x) dx = f (a).

Proof: Since is a Diracs function, there exists a positive innitesimal such that: | | ( ) = ( ) = 0. By the additivity with respect to the virtual integration interval, we have:

( a)f ( ) d =

a+

( a)f ( ) d +

( a)f ( ) d +

a+

( a)f ( ) d.

On the right-hand side of this equation, the integrands of the rst and third integrals vanish, for any function f , and therefore those integrals exist and are equal to zero. In the second integral (right-hand side) we have the product of two continuous functions, so the integral on the left-hand side exists and:
a+

( a)f ( ) d =

( a)f ( ) d.

The VET shows that the integration by parts formula holds for virtual integration. Since the function f : R R is dierentiable between a e a + , we have:
a+ a

( a)f ( ) d = [ ( a)f ( )]a

a+

a+

a+ a

( a)f ( ) d

= =
a

( a) [f ( )] d ( a) [f ( )] d

f (a). Taking a = 0 and f (x) constantly equal to 1 , we get:

(x) dx = 0,

therefore: 17

cannot be a Diracs function.

If : R R is a dierentiable Diracs function then its derivative : R R

Proceeding on the same line of argument, it is not dicult to obtain the sifting property associated to the higher order derivatives of a suciently dierentiable Diracs function: function n times dierentiable around a R , then:

If : R R is a Diracs function n times dierentiable, and f : R R is a real (n) (x a)f (x) dx = (1)n f (n) (a).

These sifting properties of the higher order derivatives of a Diracs function only hold for suciently dierentiable real functions, whereas the Diracs equivalence, as dened in Sec. IV, requires that

(x)f (x) dx =

(x)f (x) dx

for every continuous real function f : R R . Since there are continuous non-derivable properties, even for a Diracs function suciently dierentiable. Nevertheless, if we are only dealing with dierentiable real functions, we can consider

real functions, we cannot deduce equivalence formulae in this strong sense using those

weak Diracs equivalences, which require integrals equal for this kind of real functions only: equivalent in order n when, for every real function n times dierentiable f : R R , we

We will say that two integrable virtual functions : R R and : R R are

have:

(x)f (x) dx =

(x)f (x) dx.

We will indicate that two virtual functions are equivalent in order n by writing: (x) (x) above, that: function n times dierentiable around a R , then:
n

(order n ).

According to this denition, it is not dicult to deduce, from the sifting properties If : R R is a Diracs function n times dierentiable, and g : R R is a real n (ni) g (a) (i) (x a) i 18

g ( x)

(n)

(x a) (1)

n i=0

(1)i

(order n ).

References
1 2

L. Schwartz: Th eorie des Distributions (Hermann, Paris, 1966). P. Antosik, J. Mikusi nski, R. Sikorski: Theory des DistributionsThe Sequential Approach (Elsevier, Amsterdam, 1973). P. A. M. Dirac: The Principles of Quantum Mechanics (Oxford at the Clarendon Press, 1947), 3rd ed., pp. 5861. S. F. Cortizo: Virtual Extensions, to appear (1995). S. F. Cortizo: Virtual CalculusPart I, to appear (1995). S. F. Cortizo: Virtual CalculusPart II, to appear (1995).

4 5 6

19