PRODUCT OF DISTRIBUTIONS AND ZETA REGULARIZATION OF


DIVERGENT INTEGRALS  x m s dx AND FOURIER TRANSFORMS
0

Jose Javier Garcia Moreta

Graduate student of Physics at the UPV/EHU (University of Basque country)
In Solid State Physics
Address: Address: Practicantes Adan y Grijalba 2 5 G
P.O 644 48920 Portugalete Vizcaya (Spain)
Phone: (00) 34 685 77 16 53
E-mail: josegarc2002@yahoo.es

 ABSTRACT: Using the theory of distributions and Zeta regularization we manage to give
a definition of product for Dirac delta distributions, we show how the fact of one can be
define a coherent and finite product of dDirac delta distributions is related to the


x
ms
regularization of divergent integrals dx and Fourier series, for a Fourier series
0

making a Taylor substraction we can define a regular part Freg (u ) defined as a

N
function for every ‘u’ plus a dirac delta series c
i 0
i
(i )
(u ) , which is divergent for u=0 ,

we show then how  (i ) (0) can be regularized using a combination of Euler-Mclaurin


formula and analytic continuation for the series ii 0
k
  (k )

PRODUCT OF DIRAC DELTA DISTRIBUTIONS  ( m ) ( x) x  ( n ) ( x)

One of the problems with distributions , as proved by Schartz (see ref [1] ) is that we
can not (in general) define a coherent product of distributions, for example

  1  1  ( x)
  x   P 
1
0     P     0 P      lim  dx (1)
x   x   x   0
x 
x

1
For the case of the product of a Heaviside step function H(x) with the derivatives of the
Delta function (and its derivatives ) we have to deal with the problem of divergent
quantities, for example according to [2] we can define the product H   ( m ) , with the
aid of a test function  ( x)  C  ( R ) as the recurrence

  



dxH ( x) ( m ) ( x) ( x)   dx ( m ) ( x) ( x)   ( m1) (0) (0)   dx ( m1) ( x) '( x)
0 0
(2)

1
The case m=0 is just H     , and comes from considering the Heaviside fucntion
2

H(x) to be the derivative of  ( x) , so  dxH ( x) ( x)   H 2 ()  H 2 ()  
1 1

2 2

If we use the ‘Convolution theorem’ [5 ] in a formal sense, so it can be regarded as

valid even for the case that the Fourier transform are defined ONLY as distributions

(2 ) 2 i mn D m ( ) D n ( )  F  x m  x n   AF 



dtt m ( x  t ) n  (3)
Here ‘A’ is a normalization (finite) constant that depends on the definition you take for
d
the Fourier transform, but it can not be dependent on m or n and D  . Unofrtunately
dx
(3) makes no sense since the integral over ‘t’ is DIVERGENT and needs to be
regularized, if we use the Binomial theorem on t m ( x  t )n for m and n integres

n
n
i m  n D m ( ) D n ( ) =  k i mk
AD n  k  ( )(1) k i n  k D m  k  (0) (4)
k 0  


i m k
The problem here is that D m k  (0) 
2   x m k dx is infinite and would need to be

regularizad in order to make sense inside (3) or (4) , for m+k being an Odd integer,
  2 n 1 
using Cauchy’s principal value definition P.v   x dx   0 (this imposes the
  

condition that only +1 or -1 can appear inside (4) ) , the problem is that 2  x 2 n dx is still
0

divergent , the same problem happened inside (2) where one needs to to regularize
expressions  ( m 1) (0) in order to define a coherente product of distributions involving
Heaviside step-function and Dirac delta and its derivatives. In general (4) will be non-
commutative so we can in general expect  ( m ) (u )   ( n ) (u )   ( n ) (u )   ( m ) (u ) example

 (u )   (1) (u )   reg (0) (1) (u ) but  (1) (u )   (u )   (u ) reg (1) (0)  0 (5)

2
 
The last equality in (5) comes from the fact that  (0)   2 
1
 xdx is 0 by using

Cauchy’s principal value , the case m=n=0 is just the square of delta function
A
     2    , this can be obtained from the zeta regularization
2
 
  


dx  2 0  2
dx  1  1 as we will see in the next section
n  0  reg

o Zeta regularization for divergent integrals:

In our previosu paper [4] we used the Euler-Maclaurin summation formula with
f ( x)  x m  s in order to stablish

 
m  s m 1 s a

 x dx 
m s

2 a
x dx   ( s  m )  a ms
 
k 1
k m s
a

a >0 (6)

B2 r (m  s  1)
 (m  2r  1  s)  x m 2 r  s dx
r 1 (2 r )!  ( m  2 r  2  s ) a


x
m s
The idea is , given a fixed ‘m’ we define an s sufficiently large so the integral dx
a

and the series  ( s  m)   i m  s converge , and then use the analytic continuation to
i 0
extend the definition of the sum as the negative value of the Riemann Zeta

 (m)   i m , in order to regularize , using (5) the divergent integrals, if ‘m’ is an
i0
integer we can set a=0 and (5) becomes an easier expression

  
m m 1 
B2 r (m  1)
 x dx 
m

20
x dx   (  m )  
r 1 (2 r )!  ( m  2 r  2)
(m  2r  1)  x m  2 r dx (7)
0 0

The case m-s=-1 inside (6) can not be regularized inmediatly due to the pole
 
dx
 (1)   i 1   , hence to regularize  we integrate with respect to ‘a’ to find
i 0 0
xa

Ca   dx log( x  a ) , using Euler-Maclaurin summation formula plus the regularization
0

of Hurwitz Zeta function  log( x  a)    (0, a) and taking the derivative respect to
n0
s


dx 1  2 (0, a )  B2 r d 2 r 1  1 
‘a’ 0 x  a  2a  sa   r 1 (2 r )! dx
2 r 1  
 x  a  x0
(8)

3
The first three terms of the recurrence (7) are


1

1  B2  
1/ 2   dx  (0)   (1)   xdx   dx   (1)   a21  dx   x dx (9)
2

0
2 0  0
2  2 0 0

(m  1) x 
B xn
With amr  and x   n being the Bernoulli numbers
(m  2r  2) e  1 n 0 n !
B2 n 1  0 , from the definition of our product of Dirac delta distributions given in (4)
1
and since we want the identity H     to be true for every test function , we can
2
identify  reg (0)   (0) ,  reg (0)  0 and 2 reg (0)   (0)  2 (1)   B2 a21  0 
from the point of view of Zeta regularization. Although we have used only a definition
for distributions on R , it can be generalized to Rn by using the definition of Dirac delta
n n
function and Heaviside function in several variables  (xj )
j 1
 H (x )
j 1
j , in any

case we have chosen the regularization 2 i m ( m ) (0)  x
m
dx for ‘m’ integer odd or

even, other definition for the Fourier transform can make a factor different to 2


 dxe
2 iux
appear in (4) for example   (u )


REGULARIZATION OF FOURIER INTEGRAL USING DISTRIBUTIONS

 
Let be R n , then we can regularize the Fourier transform  d ke f (k )  F (u) via a
n iu .k

Rn
   
| f (k )   d ke f (k ) (see [6] )
iu .k n iu .k
taylor series substraction with the definition e
Rn

  k   k
eiu .k | f (k )    
 N  !
 f (0)  e iu .k
|    f  (0)
 N  !
(10)

|  | 1   2  ....   n  !  1 !. 2 !.... n !   1 .2 ...... n is the multi-index

notation to write down the definition of Taylor series (9)

d k f (k )   N 1
n
The Taylor series is finite and is truncated after a given N so
n
R
(ultraviolet divergence cut-off ) , this allows us to write down a regular part of the
Fourier transform plus a distributional part for the Fourier transform

   k  C   
Freg (u )  n d n
ke iu .k


f ( k )  
 N  !
  f  (0)   !   f  (0)  i x   (u) (11)
R    N  
(regularized part = function ) (singular part = distribution )

4
The problem with (10) comes whenever the integral is divergent and we set u=0 , in this
case we should have to evaluate   m  0  and other divergent quantities, also since two
distributions can not in general be multiplied then F  u   G  u  can NOT be defined,
only the ‘regular’ parts of both F and G Freg  u   Greg  u  or Freg  u     m   u  ,
Greg  u     m  u  can be defined, here we find the problem of giving a regularized
definition to   n   u     m   u  for integers (m,n) , this was discussed in (4) (5) (6) and
(7) and (8) formulae including on how to deal with with the infinite terms   m  0  via
Zeta-regularization , an small problem we find here is that depending on the definition
of the Dirac delta function via Fourier transform an extra term proportional to 2 or
similar could appear, this happens because usually the definition of the Fourier
transform is not universal (up to a factor proprtional to 2 or square root of 2 ) . So
in general depending on the definition for the Fourier transform we should make the
replacement u  2 u to get the correct results.

1 1 1 1

PRODUCT OF DISTRIBUTIONS   u   P   , P    P   ,    u   P  
u u u u

Applying the convolution plus the zeta regularization algorithm and the Fourier

1
transform for the Heaviside function  dxH ( x)e iux   (u )  iP   we can extend our
 u
definition of (regularized) product of distribution to include the Principal value
1
distribution P   related to Cauchy’s principal value of the integral
u

1  ( x)
P      pv  , using again the Fourier transform convolution theorem
u 
x

1
 Product of   u   P   : in this case using the convolution definition
u
  1   
 (u )    (u )  iP     AF   dtH ( x  t )   iA (u )   (0) A (u ) (12)
  u   

1
 P     (u ) : using again the Fourier transform for H(x)
u

  1     (0) A
  (u )  iP     (u )  AF   dtH (t )    (u ) (13)
  u    2

5
 1 1
 P    P   : this case is a far bit more complicated to obtain this product we
u u

need the identity  dtH (t ) H ( x  t )  H (0) xH ( x)

H(0)= 1/2

  1    1   1 
  (u )  iP  u      (u )  iP  u    AH (0)i (u )  AH (0) P  u 2  (14)
       

  1    1 
 i (u )    (u )  iP    and   (u )  iP     i (u ) , again using the
  u    u 
appropiate form of the convolution theorem

  1   (u )
i (u )    (u )  iP      AI1 (u )  A (15)
  u  2

  1 
  (u )  iP     i (u )  iA (0) (u )  I1 (u ) A (16)
  u 


 1   1 
  (u )  P  2  and P  2    (u ) : using  dxe iux H ( x) x   i (u )  P (u 2 )
u  u  
and the convolution theorem we can write down

  1 
  (u )  iP  2    i (u )  i (u ) (0) A  I1 (u ) (17)
  u 

  1  1
i (u )   i (u )  P  2      (u ) A  AI1 (u ) (18)
  u  2

 1   1 
 P  2   P  2  : using (14) (17) (18) and the product
u  u 

  1    1   
  i (u )  P  u 2      i (u )  P  u 2    AF   dtH ( x  t ) H (t )( x  t )t  (19)
 
       

i A (u )  1 
The last expression in (19) is just  AP  4  ,again we have used the
6 u 

identity  dtH (t ) H ( x  t )  H (0) xH ( x) together with (4) and (5) in order to give a


 1   1 
finite meaning for the product P  2   P  2  , note that in expressions (12-18) we
u  u 
need to evaluate products of the form  (u )   ( n ) (u ) which need to be regularized by
(m)

(4)

6
Depending on the order in which convolution is taken we may find H(x-t) or H(t) (x-t)
or simply ‘t’ inside (12-18) , here as always A is a number introduced by the definition
 (0)    
taken for the convolution and   (1)    xdx   I1 ,  (0)    dx  are
2 0  reg  0  reg
finite corrections (regularizations ) for the divergent integrals that appear when we try
to define a correct product of distributions , from these formulae above together with the
Lebiniz formula (considered to be valid at least in a formal sense)
d  A  B  dA dB  1 
  B  A , we can define also   u   P  2  or similar products
du du du u 

1 1 1

  ( m ) (u )  P   and P     ( m ) (u ) for arbitrary ‘m’ , H (u )  P  
u u u

1    ( m1) (u )i m 1
i m ( m ) (u )  P    AF   dtH ( x  t )t m    (u )(1)m I m  (20)
u    m 1

1   m
m
P    i m ( m ) (u )  AF   dt ( x  t )m H (t )   A   (1)m  k I m  k i k  ( k ) (u ) (21)
u   k 0  k 

 
Here I m    t m dt  these integrals can be regularized via formula (6) or (7) However
0  reg
1
if we put m=-1 in order to evaluate H  u   P   inside (20) and (21) we will find
u
several oddities that prevent us from defining a coherent expression , however the
derivative of this product of distribution satisfy

d  1 1  1  1  1  1 

 H  P    P    H     P    H  P  2   P      P  2   H (22)
dx  u u  u u  u u 

1
Another possibility is to define H  u  a   P    Ta (u ) so its derivative
u
 1  dT (u )
  u  a   P    a , using the Taylor distributional series given in [2]
u dx

(a)n ( n )

n0 n!
 (u )   (u  a ) , using formulae (21) and (22) and integration with respect

1
to ‘a’ we can get H  u  a   P    Ta (u ) up to some constant Ca . Also if we knew
u
1
how to multiply H  u  a   P    Ta (u ) for some a >0 , (to avoid the singular point
u

( a) n ( n1)
u=0 ) , using the Taylor distributional series   (u )  H (u )  H (u  a ) and
n 1 n!

7
then using (21) (22) . Although we have only considered the 1-D case, the Convolution
theorem, Binomial theorem and similar can be defined also in R n , also we must take
into account that in general for divergent integrals a change of variable could not work
   2



x m dx  y n dy   dr  r m n1d sin n  cos m  , the best method would be to use a
 0 0
Feynmann parametrization to define the product of n integrals

1
1 1 1
  u1  ...........un  1
 (n  1)!  du1  du2 ....... dun (23)
A1 A2 ....... An 0 0 0  u1 A1  ...........un An 

With An (  )   x  dx being a divergent integral that can be regularized (   1 ) via
0
Zeta-regularization

CONCLUSIONS AND FINAL REMARKS

Using the zeta regularization algorithm (6) (7) we have managed to give a finite (Non-
commutative) product of dirac delta distributions  ( m ) (u )   ( n ) (u ) , and
 ( m ) (u )  H (u ) , with ‘H’ being the Heaviside step-function , since the product is non-
commutative we should also take care when taking the products
( ( m )   ( n ) )   ( k )   ( m )   ( n )   ( k )  so associativity will not always hold , using the
Convolution theorem plus the use of Fourier transform, with the m-th and n-th powers
 
of ‘x’ F  x m * x n   AF   dt ( x  t )n t m  A = normalization constant , will allow us to
  
compute the product    up to several divergent quantities  ( m ) (0) , which are
(m) (n)

x
m
proportional to the divergent integral dx , this integral can be regularized [4] using

the zeta regularization algorithm in order to ‘substract’ finite quantities proportionals to
 (m) m=0,1,2,3,........ . Although we have only examinated the case of dirac delta
and its derivatives , in several cases it could appear the distribution


eiux d |u|
 x dx   sin g (u )   du   H (u)   H (u) (24)


Although we have not mentioned the case  dxf ( x)eiux , this integral can be reduced to
0
the calculation of a Fourier integral by setting

 
1
f ( x ) H ( x )  f (  x ) H ( x )  f  ( x )  f  ( x )  g ( x)  dxg ( x)e  0 dxf ( x)e (25)
iux iux

2 

8
In this case we will encounter divergent terms  ( m ) (0) , when using the Leibniz’s
dn n
 n  d k f d nk H
formula to perform the Taylor substraction near x=0      k . nk
f . H 
dx n k  0  k  dx dx
since the derivative of an step funciton involves a dirac delta , again we will need
formula (5) (6) and (7) to get some finite results.


If the integral of f(x) has some logarithmic divergence so  dxf ( x)  log  , then we
0

may have to regularize the distribution H ( x) x 1 as


H   ( x)   (0)
1
 ( x)
P. f   |    (0) log    dx   dx (26)
 x  x 1
x

And then ignoring all the divergent terms proportional to log  (via counterterms)

inside (12) so only finite contributions will appear inside  dxf ( x)
0

References:
[1] Colombeau, J. F., “Elementary introduction to new generalized functions”. North-
Holland, Amsterdam, 1985.

[2] Estrada R. Kanwal R. “A distributional approach to asymptotics “ Boston

Birkhäuser Birkhäuser (2002) ISBN: 0817641424

[3] Elizalde E. ; “Zeta-function regularization is well-defined”, Journal of Physics A

27 (1994), L299-304.

[4] Garcia J.J ; “Zeta Regularization applied to the problem of Riemann Hypothesis
and the Calculation of divergent integrals “ e-print avaliable at
http://www.wbabin.net/science/moreta23.pdf

[5] Kammler, D. W. “A First Course in Fourier Analysis”. Upper Saddle River, NJ:
Prentice Hall, 2000.

[6] Scharf G , “Finite Quantum Electrodynamics: The Causal Approach”, 2nd edition,
Springer, New York (1995)

[7] Zeidler E. “Quantum Field theory Vol .1;A Bridge between Mathematicians and
Physicists” Springer (2009) ISBN: 978-3-540-34762-0