THE THEORY OF DISTRIBUTIONS APPLIED TO c DIVERGENT INTEGRALS OF THE FORM dx()x u a (x b)
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: [email protected]
MSC: 60E05, 62E99, 35Q40
ABSTRACT: In this paper we review some results on the regularization of divergent integrals of c ()x dx dx the form and in the context of distribution theory a x b a x a Keywords: Regularization, distribution theory, fractional calculus
Regularization of divergent integrals:
In QFT (Quantum field theroy) there are mainly two kinds of divergent integral , the UV (ultraviolet) divergence, that happens whenever an integral is divergent when x and the IR (infrared) one that occurs if the integral is divergent as x 0 ,a few examples of these integrals are
dxx3 dxx2 dx dx a R or a Z (1) 2 2 3 0 x a 0 (x a) 0 x( x a) 0 x
The first two integrals have an UV divergence, whereas the last ones have an IR divergence,the names IR and UV divergences come from the fact that if we use the h Wave-particle Duality in QM setting x=p we find that UV divergence is | p |
1 equivalent to have small wavelengths (ulra violet) and the IR divergence happens for big wavelengths (infrared) so the name is not casual, for more details [5] and [9]
To deal with UV divergences on an integral of the form dkd F() k on Rd we simply Rd make a change of variable to d- dimensional polar coordinates , to rewrite the integral
c k a () a dkd F() k d drr d 1F(,) r d dr a () ri 1 d n () i n1 Rd 0 c i0 r n2 c (1 n)
(2) 1 dzF(,) z With the constants a () , here means that we must integrate n 2i z 1 over the angular variables , in case F is invariant under rotations on the d-dimensional 2 d / 2 plane the angular part of the integral can be calculated exactly as d , the (d / 2) idea of expanding our function into a Laurent series , is to isolate the UV divergencies of the form drr k so they can be ‘cured’ using the zeta regularization algorithm. 0 Choosing any c >1 , since the integral has a UV divergence , after expanding the integrand F(,) r r d 1 into a convergent Laurent series for |r| >1 there will be only a finite number of divergent integrals of the form drr k plus a logarithmic divergent 0 dr integral (this is another example of UV divergence) , now to get finite results 0 r c from the divergent integrals we will apply the recurrence deduced in our previous paper [4 ] when discussing the zeta regularization applied to integrals
m B m!( m 2r 1) xm dx xm1dx (m) 2r xm2r dx (3) 0 2 0 r1 (2r )!(m 2r 1)! 0
Equation (3) is a ‘regularization’ in the sense that if we had the upper limit N-1 instead N N k1 of the expression (3) would give the well-known result dxxk k >0 or k=0 , 0 k 1 for the case dxxk , the series nk is no longer convergent and must be given a finite 0 n0 k value in the spirit of Zeta function regularization [ 4] so n R (k) , this is why n0 the term R (m) (m) apperas inside (3) , note that (3) is a recurrence formula with initial term dx 1 1 1 1 1 1 (0) 1/ 2 and allows to calculate or assign a 0
2 finite value to every divergent integral of the form drr k for ‘k’ a positive integer , (the 0 main problem for the logarithmic divergence is the pole of the Zeta funciton at s=1) the first terms are I(0, ) (0) 1/ 2 dx 0
I(0, ) I(1, ) ( 1) xdx 2 0 (4) 1 B I(2, ) I(0, ) ( 1) 2 a I(0, ) x2 dx 21 2 2 0
3 1 B I(3, ) I(0, ) ( 1) 2 a I(0, ) ( 3) B a I(0, ) x3 dx 21 2 31 2 2 2 0
k Notation I(,) m stands for lim dxx , being ‘Lambda’ a cut-off with certain 0 physical meaning.
This example is valid for UV divergences but what would happen to the case of c dx dr ‘singular integrals’ a
o Regularization of divergent integrals using Distribution theory:
As a first example, let us suppose we want to calculate the integral dxf() x Dn (x a) , for n=0 we know that in spite of the ‘pole’ of the delta function at x=a we have that dxf()( x x a) f() a for any test function f(x) , then we can use two method
We perform integration by parts avoiding the singular point at x=a so dx( D)n f()( x x a) dxf() x Dn (x a)( D)n f() a (5)
We integrate n-times (no matter if n is non-integer) with respect to ‘a’ I() a dxf() x Dn (x a) Dn I() a dxf( x )( 1)n (x a) ( 1)n f() a (6) a
3 n n Both methods are equivalent since if we recall the identity Da Da I() a I() a we yield to the same result, no matter if we integrate/differentiate with respect to the parameter ‘a’ or if we integrate by parts.
c dx dx A similar idea can be applied to integrals of the form and first we define a x b a x a (x b)s for a
1 x 1 q Dn f() x dt( x t)n1 f() t dt Dq f( x ) lim ( 1)m f( x (q m)) h (7) h0 q ()n 0 h m0 m
The first definition inside (7) is the expression for fractional integral (not valid for negative ‘n’ ) , the second one is the ‘Grunwald-Letnikov’ differintegral valid for positive ‘q’ , the third alternative for the derivative comes from the definition of (q 1) dz Cauchy’s integral formula Dq f() x f() z for any rectificable curve q1 2i (z x) on the complex plane that includes the point z=x
c dx Example: Let be the singular integral I() b ()x , in order to give it a finite s a (x b) value first we differentiate -times with respect to ‘b’ so s 1/ 2 hence (1 s) c dx D I() b ()x , we make then the change of variable x b u 2 so b (1 s ) a x b 2 (1 s) c b our integral becomes D I() b du( b u2 ) , then we define the b (1 s ) a b dF function F so (b u2 ) this implies du 2 (1 s) D I() b F( c b) F( a b) finally taking the inverse operator we b (1 s ) 2 (1 s) can set I() b DF ( c b) DF ( a b) (,)b (8) (1 s ) b b d f Here ‘mu’ is a real number and D f exists in the sense of a fractional b dx derivative/integral and Db (b , ) 0 . We set the condition s 1/ 2 because with a change of variable we can avoid the pole x b1/ 2 at x=b.
4 The same strategy can be applied to singular integral equation, for example if we wished to solve the following equation
dt (1 n) dt f() x g() x f() t so D f() x D g() x f() t (9) n x x a t x (1 n ) a t x
Where n 1/ 2 , making the change of variable t x u 2 so (9) becomes
2 (1 n) D f() x D g() x du f( x u2 ) Gamma function (10) x x (1 n ) a x
Now, equation (10) has NO poles at t=x , to solve this integral equation without singularity we could use an iterative process
2 (1 n) f() x g() x D f() x D g() x du f (x u2 ) (11) 0 x n x n n1 (1 n ) a x
Where we have supposed that Dx (x , ) 0 (,)x in order to simplify the calculations of the solution of the integral.
Then one could ask, what happens in the limit b , in this case the quantity x b1 becomes zero for every x , so the I(b) must tend to 0 for ‘b’ big hence we should choose the function (,)b with the conditions Db (b , ) 0 and
2 (1 s) lim DFb ( c b) DFb ( a b) (,)b 0 (12) b (1 s )
o Logarithmic divergences
The case of the logarithmic integral is a bit different, since formula (4) can not handle it ,due to the pole of zeta function at s=1, one of the ideas to apply [4] is just to replace the dx 1 divergent integral by a divergent series with ‘h’ being an step (we 0 x a n0 n a/ h use Rectangle method ) ,this series is still divergent but can be assigned a finite value via ‘Ramanujan resummation’ equal to the logarithmic derivative of Gamma function ' a ,however this kind of method depends on the value of the step ‘h’ given. h
5 dx From Fourier analysis one can interpretate , the integral as the following 0 x a H() x 1 H( x a) convolution H* x1 dx dx dx with H(x) the ‘Heaviside x a a x x step function’ , using the property of the Fourier transform and the convolution theorem 1 H() x 1 i iua H* x dx d u ()u e I() a (13) x a 2i u
Here |x| is the absolute value function that takes the value x or –x depending on if ‘x’ is either positive or negative, solving Fourier transform (13) we can solve the logarithmic UV divergence. If we can solve (13) for a fixed a then for another value ‘b’ so b is dx b dx different from 0 or inifinite we have log I() a 0 x a a 0 x b
1 Another simpler method is that if we have I() a dx divergent integral , we 0 x a 1 1 differentiate with respect to ‘a’ I'( a ) dx so integrating again with 2 0 (x a) a respect to ‘a’ I() a log(a ) c() a c1 , with c1 an universal divergent
( 1) 1 D I() a dx ( 1) ( 1) ( 1) 0 (14) a 1 0 (x a) a
So I() a ( 1) ( 1)Da a (,)a , with 0 and Da (,)a 0 , one of the problems is the apparent absurdity since due to the term 1 there is a complex contribution to an integral with real-valued integrand. An alternative formulation based on Hurwitz Zeta function is the following
1 (0,a ) dxlog(x a) dx and H log (a ) log 2 a 0 (15) a 0 0 x a a
For the sum lo g( n a) s H (0,a ) , this is the Zeta-regularized definition for the n0 Determinant of an operator , a combination of the 2 expressions inside (15) gives the 1 ' regularized value for the Harmonic sum dx ()a R (in case a=1 we get 0 x a the Euler-Mascheroni constant) , this is the analogue result to simply using ‘Ramanujan resummation’ for the series ns s >0 , and s=-1 , ‘R’ stands for the Remainder term n0 1 B 2r1 1 inside Euler-Maclaurin summation formula and is R 2r 2r1 2a r1 (2r )! x x a x0
6 Appendix: Convolution theorem
In this paper we have introduced and used the ‘Convolution theorem’ , if we have the convolution of 2 functions or distributions f(x) and g(x) defined as h() z dxf()( x g x z) then H() u F()( u G u) with F, G and H the fourier transform respectively of h(z) f(x) and g(x).
Proof:= if we define dzeiuz h() z dzeiuz dxf()( x g x z) dxeiux f() x F() u and dxeiux g() x G() u , we make the change of variable y x z so dx = -dz into the first integral so we have the following identities
dzeiuz dxf()( x g x z) dyeiu( x y) dxf()() x g y dyeiux dxf()() x g y eiuy (16)
The first integral on the left is just H(u) and using Fubini’s theorem to interchange the order of integration we have been able to proof that H() u F()( u G u) .
If we name g() x H() x and f() x xk , then dxxk H( x a) xk dx , hence a applying convolution theorem and the Fourier transform
k k i k iua dxx i du ()u D ()u e k > 0 (17) a u 1 Unfortunately there are some oddities with defining product of distributions Dk , x Dk , within distribution theory, however if the integral f() x dx makes a sense as a Riemann integral , if we define F(u) as the Fourier transform of f(x) then the F(0) a Convolution theorem gives the regularized result dtf() t f() c for some 2 c 1 d| x | Real ‘c’ , for the case of f() x with Fourier transform the integral is x dx 1 d| x | divergent in the Riemann sense but can be attached a finite value log(a ) 2 dx x0 , since |x| is even its derivative will be odd so the mean value of the derivative near x=0
7 will be 0 and we have x1 dx log(a ) . The last method would be to use the Euler- a Maclaurin summation formula to get
1 1 1 B d 2r1 1 dx 2r (18) 2r1 x0 0 x a n0 n a 2a r1 (2r )! dx x a
1 The problem is that H (1,a ) is still divergent , although using Ramanujan- n0 n a 1 summation we can attach this series the finite value H (1,a ) ()a and n0 n a plug this result into (17) , In both cases the approximation of the integral by a sum 1 dx ()a and the result log(a ) c are equivalent for a (big a n0 n a 0 x a a) since using the Stirling’s approximation for log (x ) and taking the derivative we get '(a ) / (a ) the asymptotic result lim 1 a log(a )
References:
[1] Estrada R. Kanwal R. “A distributional approach to asymptotics “ Boston Birkhäuser Birkhäuser (2002) ISBN: 0817641424
[2] Elizalde E. ; “Zeta-function regularization is well-defined”, Journal of Physics A 27 (1994), L299-304.
[3] Garcia J.J “Chebyshev Statistical Partition function : A connection between Statistical Mechanics and Riemann Hypothesis “ Ed. General Science Journal (GSJ) 2007 (ISSN 1916-5382)
[4] Garcia J.J “ A new approach to renormalization of UV divergences using Zeta regularization techniques “ Ed. General Science Journal (GSJ) 2008 (ISSN 1916-5382)
[5] Griffiths, David J. (2004).” Introduction to Quantum Mechanics (2nd ed.).” Prentice Hall. ISBN 0-13-111892-7. OCLC 40251748. A standard undergraduate text.
[6] Kenneth S. Miller & Bertram Ross “An Introduction to the Fractional Calculus and Fractional Differential Equations” Publisher: John Wiley & Sons; (1993). ISBN 0-471-58884-9
[7] Lighthill M.J “ Introduction to Fourier Analysis and generalized function “ Cambridge University Press (1978) (ISBN 0-521—09128-4)
8 [8] Schwartz L (1954): Sur l'impossibilité de la multiplications des distributions, C.R.Acad. Sci. Paris 239, pp 847-848.
[9] Yndurain, F.J. (1996). “Relativistic Quantum Mechanics and Introduction to Field Theory “ (1st ed.). ISBN 978-3540604532.
9