Академический Документы
Профессиональный Документы
Культура Документы
ABSTRACT: In this paper we study the methods of Borel and Nachbin resummation applied
to the solution of integral equation with Kernels K(yx) , the resummation of divergent series
and the possible application to Hadamard finite-part integral and distribution theory
1. INTRODUCTION
Divergent series are widely known and appear in many context involving Physics or
math, for example if we integrate by parts the error function (Laplace) :
e− x
2
∞
∞ (2n)!
∑ (−1)
−t 2
erfc( x) = ∫ dte → 1 + n
(1.1)
x
π n =1 n !(2 x) 2 n
Or using the ‘Saddle point mehtod’ for n! when n is big then we have:
1 1 ∞
Γ( x + 1) → 2π n( n +1) / 2 1 + + 2
+ ..... ∫ dte − t t x −1 = Γ( x) (1.2)
12n 288n 0
But (1.1) and (1.2) are only convergent as x → ∞ for small values of x they both
Diverge. Another example with ODE’s is the following for the ODE:
dy ∞ e−t
x2 + y = x , y ( x)=x ∫ dt (1.3) (exact solution)
dx 0 1 + xt
For (1.3) Euler gave the series solution: y ( x) = x − (1!) x 2 + (2!) x3 − (3!) x 4 − ... (1.4)
Which converges only for x=0 !!! , A similar thing happens with the series:
1
a0 + a1 g + a2 g 2 + a3 g 3 + ....... g << 1 (1.5)
That apear in QFT and Quantum Mechanics , where g is the ‘coupling constant’ , in
general series of the form (1.5) although divergent are used to calculate the ‘mass’ or
‘charge’ (renormalized value), for a given physical theory.
Also as a last example let be the next Taylor series around x=0 :
x x
x + x 2 + x3 + x 4 + ......... = , x + 2 x 2 + 3x3 + 4 x 4 + ......... = (1.6)
1− x (1 − x) 2
Convergent for |x| <1 However taking the limit x → −1− (-1 by the left) we find
the amazing results -1/2 and -1/4 Although we know that in general, any series of
the form 1 + 2m + 3m + ...... is divergent, in the last section we discuss a method to
deal with divegent integrals with poles on the interval. [a,b] with a and b real
numbers for example the integral of f ( x) = x −2 on the interval (-1,1) .
Of course this paper pretends to be only a kind of introduction to the subject for
further references I strongly recommend ‘Divergent series’ by G.H Hardy or ‘Zeta
regularization methods’ by E.Elizalde and others for historical examples involving
divergent series and integrals.
The Zeta function regularization method was used to give a meaning to clearly
divergent series in the form:
For any positive and real number ‘a’ , the series could be re-arranged in the form:
Where in (2.2) we have used an Analytic prolongation for the Riemann Zeta
Function, now letting s tends to 0 we could find the ‘sum’ of the series (2.1) as the
Value ζ (−a) which can be calculated in general (except if a=1) by the contour
Integral on the complex plane:
q − ( a +1)
ζ (−a) = Γ (2aπ+i1) ∫σ dq (2.3)
e− q − 1
πs
ζ ( s ) = sin Γ(1 − s )ζ (1 − s )π (2π )
−1 s
(2.4) ζ (−2n) = 0 n=1,2,3,4,.....
2
2
For ‘a’ being a Natural number, the calculations can be simplified, first if we
introduce the ‘Bernoulli Polynomials’ using the Generating function:
∞
te xt tn
= ∑ n n!
et − 1 n = 0
B ( x ) (2.5)
The value on the left of (2.5) for the 2n-th Bernoulli Polynomial at x=0 are the
‘Bernoulli Numbers’ and have a vital importance in Mathematical Analysis, for
example the sum of the first N powers of f(x)=xr for r >0 and integer can be
written using the Bernoulli Polynomials in the form:
N
Br +1 ( N + 1) − Br (0)
∑n
n =0
r
=
r +1
(2.7)
Although it may seem we are somhow ‘Cheating’ to get a finite value, Zeta
regularization has been proved frutifull, in many calculations involving theoretical
Physics such us the ‘Casimir effect’ where the value ζ (−3) = −1/120 is used in
Calculations, or string theory whenever you need to calculate 1+2+3+4+5.+…..
f ( x) + f ( x + 1) + f ( x + 2) + ........... = S (2.8)
d
e aD f ( x) = f ( x + a ) D= (2.9)
dx
( The formula (2.9) can be proved making the Taylor expansion respect to ‘a’ and
remembering the Taylor series for the Exponential , In Modern Lie Algebra this is
just a classical result involving group theory).
Applying (2.9) ot each term in (2.8) we have a geommetric series in eaD for
a=0,1,2,3,… multiplying and dividing by ‘D’ Euler found the relation (involving
again Bernoulli numbers):
∞
D Bn n dF
e −1
D
= − ∑
n =0 n !
D F
dx
= f ( x) Bn (0) = Bn (2.10)
If we put f(x)=x or f(x)=x3 and use (2.10) the series is just a finite one and we
get the values -1/12 and -1/120 using tables.
3
o Application to divergent integrals:
∞ ∞
β
∫ dx( x + β ) m ≈ ∑ ( β + ε n) m ε → ζ , −m ε m +1 m > 0 (2.11)
0
n =0 ε
To give a finite value to the divergent integral (2.11) using the ‘Hurwitz Zeta
function’ (a direct generalization to Riemann zeta function) defined by:
∞
ζ ( s, β ) = ∑ (n + β )− s ℜe(s)>1 (2.12)
n =0
With ‘beta’ a real number different from a negative integer, for other values we
can use the Functional equation:
ν 2Γ( s) µ
πs 2π rν r
ζ (1 − s, ) =
µ (2πµ ) s
∑ Cos( 2
r =1
−
µ
)ζ ( s, )
µ
µ≥ν≥1 (2.13)
Bm +1 ( β )
ζ (−m, β ) = − (2.14)
m +1
Hence if taking the limit ε → 0 (2.11) has a finite value , this value could be
considered as the regularized ‘sum’ or ‘Area’ for the function f ( x) = x m on the
inteerval [0, ∞)
∞
∫0
dpp m m∈Z (2.15)
For positive m (UV divergences) we can find a recurrent formula to obtain the
values of the different integrals in the form:
∞
B2 r
I (m, Λ) = (m / 2) I (m − 1, Λ) + ζ (−m) − ∑ amr (m − 2r + 1) I (m − 2r , Λ)
r =1 (2r )!
Λ Λ m Λ m ∞ B2 r Γ(m + 1)
∫0 p m dp = + ∑n −∑ amr Λ m−2 r +1 amr = (2.16)
2 n =1 r =1 (2r )! Γ(m − 2r + 2)
4
Λ
Where we have called I (m, Λ) = ∫ dpp m , here the meaning of ‘Lambda’ is
0
a momentum cut-off (the maximum value of the modulus for the momentum
of the particle) , the Renormalization procedure involves taking the physical
limit Λ → ∞ , at the end the quantities involving these divergent integrals
can’t depend on the value of the regulator/cut-off ‘Lambda’ introduced , where
we have used the ‘Euler-McLaurin sum formula’ in the form :
b f (b) b−1 ∞
B
∫0 f ( x)dx = + ∑ f (n) − ∑ 2 r f (2 r −1) (b) (2.17)
2 n =1 r =1 (2r )!
Here the [R] stands for Ramanujan or resummation, applied to a(n)=1/n gives
[ R]
1
the regularizad value ∑
n ≥1 n
→ γ .for the Harmonic series.
Λ
∫ dp = 1 + 1 + 1 + 1 + ........ + Λ → ζ (0) as Λ → ∞ (2.19)
0
For more complicate analytic functions F(p) we expand them into a ‘Laurent
(convergent) series’ in the next form:
∞
F ( p) = ∑c
n =−∞
n pn (2.20) and perform term by term integration
Depending on the region of the complex plane we are located (2.20) may
include either positive or negative power series, or them both .
5
∞
d k −1 ζ ′(−ns)
∑ p s (ln p)k → ∑ µ (n)(−1)k
p n =1
ds k −1 ζ (−ns)
(2.21)
With s >0 a real non-integer number, so the series on the right is convergent,
k >0 , used to extend the domain of convergence for P(-s) .
∑p p
s
= P( s ) valid for s> 1 (2.22)
π ∞ e−t
Γ( x)Γ(1 − x) sin(π x) = π so →∫ dt α (2.23)
Γ(α ) sin(πα ) 0 t
With α real number bigger than 0 and different from an integer, clearly
(2.23) is a form to give a finite meaning to otherwise divergent integral due
to the pole at t=0, using the analytic prolongation to negative values for the
Gamma function, in both cases involving (2.1) and (2.23) we consider that
The ‘sum’ or ‘Area’ for the series or integral is just the value of
Γ(α ) , ζ (− a) , this process of ‘regularization’ avoids the pole for the
Gamma and Riemann zeta function at the points α = 0 (Gamma) and s=1
(Riemann Zeta function). Also for alternating series:
∞
1 − 2m + 3m − 4m + .... = η (− m) since η ( s ) = ∑ (−1) n n − s = (1 − 21− s )ζ ( s )
n =1
(2.24)
Appart from Euler the formalism of ‘Zeta regularization’ (but not with this
name) was introduced by Ramanujan in his Notebooks using the Euler-
McLaurin sum formula defining the ‘constant’ of the series:
∞
1 B
c=− f (0) − ∑ 2 k f 2 k −1 (0) (2.25)
2 k =1 (2k )!
B2 n +1
1 + 22 n +1 + 32 n +1 + ................. = − (ℜ) (2.26)
2n
1 + 22 n + 32 n + ................. = 0(ℜ)
6
3. BOREL RESUMMATION FOR SERIES AND INTEGRALS
S = a0 + a1 + a2 + a3 + ......... (3.1)
Borel gave a very ingenious method to calculate it, first we multiply and
divide each term by n! getting :
∞ ∞ ∞
an an
∑ n! ∫
n =0
0
dtt n e − t ∑ n! t
n=0
n
= f (t ) (3.2)
Using (3.1) and (3.2) and supposing that f (t ) = O(ebt ) for a real positive
number b then we can writte the ‘sum’ of the series in (3.1) as:
∞ ∞ ∞ ∞
p +1 ∞
∞
(−1)k x k ∆ k [(−1)n an ]n=0 x
p
∑
n =0
a ( n ) x ≈ ∑ n
k =0 (1 + x)k +1
+
1+ x
∑ (−1)
n =0
k + p +1
∆ p +1[(−1)n an ]n =0 x n
n
n!
∆ n a(0) = ∑ (−1) m an − m (3.5)
n =0 n !( n − m ) !
∞ tp − st 1 ∂p ∂p ∞ e−t
∫ dt e = E1 (cs )ecs E1 ( x) = ∫ dt = − Ei (− x) (3.6)
0 (t + c) p +1 p ! ∂s p ∂c p x t
The first expression in (3.5) is an approximate evaluation for f(x) setting x=t
then the B(S) ‘Borel sum’ for our divergent series (3.1) is:
1 p 1 ∂k ∂k
B(S ) ≈ ∑
s k =0
(−1) k
k ! ∂s k ∂c k
∆ k [(−1) n bn ]n =0 E1 ( s)ecs (3.7)
c =1, s =1
With an = bn n ! , only in case that the coefficients of our initial series (3.1)
were of the form (-1)n n!P(n) with P(n) a Polynomial (3.7) is exact. The
7
error term is given by the expression:
1 1 ∂ p +1 ∂ p
E = O E (cs )ecs (3.8)
s p ! ∂s p +1 ∂c p 1
c =1, s =1
In case (3.1) were convergent then its ‘ Borel sum’ is equivalent to the
term-by-term Laplace transform at s=1, in that case if we had f(t) and g(t)
analytic series near x=0 ::
∞ ∞
an
∑ an x n = g ( x)
n=0
and ∑ n! x
n=0
n
= f ( x) (3.9)
∞
If we define g ( s ) = ∫ dtG (t )e − st ,since f(t) and g(t) are convergent series
0
then we should have as a main property of Borel sums:
∞ ∞
an 1 ∞
g ( x) = ∫ dt ∑ ( xt ) n e − t = ∫ dtf ( xt )e− ( t / x ) (3.10)
n =0 n ! x 0
0
Setting s=1/x , and using the a known property for the Laplace inverse
transform of g (1/ s ) s −1 we find the relation:
∞
f (t ) = 2 ∫ duuG (u 2 ) J 0 (2 tu ) F0 (k = 2 t )(2G (u 2 ) = f (t ) (3.11)
0
∞
f (a + n∆x) s ∆x + a ∞ a + n∆x + − t α
∑
n = 0 Γ ( a + n∆x + 1 + α )
∫0 dtt e t ∆x (3.12)
∞ ∞ f ( x)t x α − st
s ∫ dt ∫ dx t e (3.13)
0
0 Γ( x + 1 + α )
∞
Of course in general, unless ∫ 0
dxf ( x) is convergent ‘Fubini’s theorem’
does not hold for (3.12) and (3.13) so:
8
∞ ∞ ∞ ∞ f ( x)
∫ dt ∫ dxσ ( x, t ) ≠ ∫ dx ∫ dtσ ( x, t ) σ ( x, t ) = t x +α e − t
0 0 0 0 Γ( x + 1 + α )
(3.14)
∞ f ( x)
Hα (t ) = ∫ dx tx With H α (t ) = O(eb2t ) (3.15)
0 Γ( x + 1 + α )
If the 2 conditions on (3.16) holds then the ‘Borel integral’ is just the
Laplace transform of Hα (t )t α , α > 0
But. Can a ‘Borel sum’ be the real sum of the series?, let’s take :
∞ e−t e− x ∞
(−1) n
E i ( x) = − ∫ dt
−x t
≈
x
∑
n =0 xn
n! (3.16)
∞
1 ∞
(−1) n
x ∫ dte − xt
→∑ n! (3.17)
0
t + 1 n =0 x n
Using the result for the Laplace transform of 1/(t+1) ,we find:
1 s
L = e Ei ( s) (3.18)
t + 1
∞ f ( x) F (ln s)
L ∫ dxt x =
Γ( x + 1
and { }
L t n f (t ) = (−1) n D n F ( s) (3.19)
0 s ln s
9
∞ d α Fα (ln s )
∫ dxf ( x) s − x → s (−1)α (3.20)
0 dsα s ln s
f ( x)
Where Fα ( s ) is the Laplace transform of g ( x) =
( x + 1)( x + 2).....( x + α )
t
x
∞
∞ an t n − t ∞ ∞ B2 r ∂ 2 r −1 a ( x )
F (ln( s ))
= ∫ dt ∑ e + ∫ dt ∑ s e−t
r =1 (2r )! ∂x
2 r −1
Γ( x + 1)
ln( s ) 0 n =1 n ! s 0
(3.21)
a ( x)t x
All the derivatives ∂ 2x r −1 must be evaluated at x=0, with lim =0
x →∞ Γ ( x + 1)
(4.1) is just the Borel transform of the sequence {an } , with the advantage
that now f(t) can grow faster than ebt so f (t ) ≠ O(ebt ) , we will study the
applications of this Nachbin/Generalized Borel resummtion to solve
integral equations and to the Riesz criterion for Riemann Hypothesis
10
In order to apply the Borel generalized resummation to integral
equations, let be the Fredholm equation of first kind :
∞ ∞
g ( s ) = s ∫ dtK ( st ) y (t ) Kˆ (n + 1) = ∫ dtK (t )t n (4.2)
0 0
Here K(st) is the Kernel of the integral equation , and g(s) has the form
of a Z-transform involving inverse powers of ‘s’ :
c1 c2 c3 1
g ( s ) = c0 + + + + ........ cn = ∫ dzg ( z ) z n −1 (4.3)
s s 2 s3 2π i γ
Since the Mellin transform for Kernel K(u) exists, we will apply Nachbin
resummation to solve the integral equation given in (4.2)
∞ ∞
∞
∞ cn tn
g ( s ) = ∑ cn s − n = ∫ dt ∑ . n K (t ) = s ∫ dtK ( st ) y (t ) (4.4)
n=0 0 n =0 Kˆ (n + 1) s 0
∞
dk 1 dz
π ( x) = ∑1 = ∑ ln k ( x) dk = ∫ ln ζ ( z ) (4.5)
p≤ x k =1 ζ ( k + 1)Γ ( k + 1) 2π i γ z1− k
∞
ln m ( x) 1 dz
A( x) = ∑ an = ∑ um um = ∫ H ( z) (4.6)
n≤ x m=0 m! 2π i γ z1− m
For the case of the Riesz function involved in Riesz criterion, the
Riemann Hypothesis can be stated as the condition
11
∞
(− x)n
Riesz ( x) = −∑ = O( x1/ 4+ε ) (4.7)
n =1 ( k − 1)!ζ (2n)
∞
( − x) n
∞
∞ (− x) n t n −1 1 ∞ dt x
1− e −x
= −∑ = − ∫ dt ∑ = ∫ Riesz (t )
n =1 n! 0 n =1 ζ (2n)(n − 1)! t 0 t t
(4.8)
then (4.8) is a proposed integral equation for the Riesz(x) ,from the
definition of floor function [x] when x=0 both sides on (4.8) are 0
lim+
ε →0
(∫a
c −ε
dxf ( x) + ∫
b
c +ε
dxf ( x) ) (5.1)
c b
∫ a
dxf ( x) = ±∞ ∫c
dxf ( x) = m ∞ (5.2)
ε
a d k −1 g ( x) d k g ( x)
F .P ∫ dxf ( x) = f ( x) = (5.3)
−a dx k −1 −ε dx k
12
k −1
Pk 1
g ( x) = k, m > 1 Pk = (−1) k ∏ (5.4)
x m−k i =0 (m + i)
a
k −1
d i 1 d k −1−iϕ ( x) a ϕ ( x)
∑ (−1)
i =0
i +1
i
dx r m−k
dx k −1−i
= FP ∫ dx m
−a x
(5.5)
−a
∞
S , ϕ = ∫ dxS ( x)ϕ ( x) so S ′, ϕ = − S , ϕ ′ (5.6)
−∞
d nϕ (0)
δ [ϕ ] = ϕ (0) and δ ( n ) [ϕ ] = (−1)n (5.7)
dx n
∞
2πδ ( n ) ( x) = ∫ dx(ix) n eikx (5.8)
−∞
So for every Analytic function f(x) we can define its Fourier transform in
term of ‘test function’ as the linear functional:
I [ f , ϕ ] → 2π f (−i∂ x )δ ( x) (5.9)
The ‘Borel integral’ (3. ) and ‘Hadamard finite-part integral’ can be useful
to deal with divergences in the form ∫ d 4 p | p |r for r >0 and r<0 , r ≠ −1
and the modulus of the momentum | p |→ 0, ∞ that appear in QFT at long
or short distances (wavelength), since in natural units λ | p |= 1 , hence we
could use these 2 techniques to get rid of the divergences, introducing a
cut-off or regulator Lambda and making Λ → ∞ .
6. REFERENCES
[2] Brendt Bruce “Ramanujan´s Notebooks: Parts I-II” Springer-Verlag New York
(1985-1989)
13
[4] Demidovich B.P and Maron I.A “Fundamentos de Analisis Numerico” (Spanish)
Ed: Parninfo Madrid (1985) , ISBN: 84-283-0887-X
[5] Elizalde .E, Romeo .A “Zeta regularization techniques with applications” World
Scientific Publishing company. (1994)
[7] Mostepanenko, V. M. and Trunov, N. N. §2.2 in “The Casimir Effect and Its
Applications. Oxford” , England: Clarendon Press, 1997.
[8] Nachbin L. "An extension of the notion of integral functions of the finite
exponential type", Anais Acad. Brasil. Ciencias. 16 (1944)
[9] Peskin M.E and Schröeder D. V “Introduction to Quantum Field Theory” Addison-
Wesley (1995)
14