View metadata,citationandsimilarpapersatcore.ac.uk ) ving UV ours. HEP-PH-9505224 v -ratio and the quark a R f N e expansion. Remark- + e f N = ts. The ultra-violet ( heme dep endence of these par- ) QCD with N -function, the e co ecien D ), for SU( f ) p erformed. Resummed results are also ob- 1 N t in comparisons with the exact next-to-leading IR ts to all orders in the 1 {2 N (11 6 . The renormalization sc R tities related to the y ratio, alue regulated for wski quan o y including more exact details of the Borel transform are considered. ) renormalon singularities in the Borel transform are isolated and the IR ts -lepton deca for e note that this is already apparen ,w e co ecien analogous tial resummations is discussed and[1{3] they and are with compared exact with xed the order results from p erturbation other theory groups at NNLO. Prosp ects for impro the resummation b tained for the Mink and infra-red ( Borel sum (principal v order (NLO) and next-to-NLO (NNLO) p erturbativ ably These terms corresp ond toasymptotically renormalon to singularities dominate in the the co ecien Borel plane and are exp ected DTP/95/36 April 1995 ell provided byCERNDocumentServer ables brought toyouby Resummations -function and Deep Inelastic Scattering D CORE Observ ory, University of Durham urner and C.J.Maxw 1 = b QCD ett-T v ad, Durham, DH1 3LE, England o er of w Renormalon e for Particle The C.N.Lo some results for the QCD Adler South R f N Centr Exact large- taining the highest p o All-Orders con sum rules are used to resum to all orders the p ortion of QCD p erturbativ
1 Intro duction
In several recent pap ers [1{3] the p ossibility of resummation to all orders of the part of
p erturbative corrections contributed by QCD renormalons has b een explored.
The QCD p erturbative corrections to some generic QCD Green's function or current
correlator (to the Adler D -function of QCD vacuum p olarization for instance) can b e
written:
2 3 k+1
D = a + d a + d a + + d a + ; (1)
1 2 k
where a = is the renormalization group (RG) improved coupling; and the p erturba-
s
tive co ecients d can themselves b e written as p olynomials of degree k in the number
k
of quark avours, N ;we shall assume massless quarks.
f
[k ] [k 1] [0]
k k 1
d = d N + d N + + d : (2)
k
f
k k f k
[k r ]
The \N -expansion" co ecients, d , will consist of sums of multinomials in the ad-
f
k
2
joint and fundamental Casimirs, C =N , C =(N {1)=2N , of SU(N ) QCD; and will have
A F
k r s
s
the structure C C . The terms in this \N -expansion" will corresp ond to Feynman
f
A F
diagrams with di ering numb ers of vacuum p olarization lo ops. By explicit evaluation of
[k ]
diagrams with chains of such lo ops inserted, it has b een p ossible to obtain the leading d
k
co ecient exactly to all orders for the Adler D -function [4{6] (and hence its Minkowski
+
continuations, the e e QCD R-ratio and the -decay ratio, R ); the Gross Llewellyn-
Smith (GLS) sum rule corrections [6]; and heavy quark decay widths and p ole masses
[k ]
[7]. A general pro cedure enabling d to b e obtained from knowledge of the one-lo op
k
correction with a ctitious gluon mass has b een develop ed [1, 2].
In a recent pap er [8]we p ointed out that the large order b ehaviour of p erturbative
co ecients is most transparently discussed in terms of an expansion of the p erturbative
co ecients in p owers of b=(11C {2N )=6, the rst QCD b eta-function co ecient:
A f
(k ) (k 1) (0)
k k 1
d = d b + d b + + d : (3)
k
k k k
11
This \b-expansion" is uniquely obtained by substituting N =(
C 3b) in equation (2).
f
A
2
[k ] (k ) [k ]
k
d =( 1=3) d and so exact knowledge of the leading-N d to all orders implies exact
f
k k k
(k )
knowledge of the d .
k
In QCD one exp ects the large-order growth of p erturbative co ecients to b e driven by
Borel plane singularities at z =z =2`=b, with `=1; 2; 3;:::. The singularities on the
`
negative real axis are the so-called ultra-violet renormalons, UV , and those on the p ositive
`
real axis are the infra-red renormalons, IR . These singularities result in the large-order
`
k
b ehaviour of the co ecients d b k !. Indeed we showed in reference [8] that, given a set
k
of renormalon singularities at the exp ected p ositions in the Borel plane, the leading terms
(k )
k
in the b-expansion, d b , should, if expanded in p owers of N , asymptotically repro duce
f
k
[k r ]
the d co ecients of equation (2) up to O (1=k ) accuracy.We conversely checked that
k
(k )
k
the exact d b results corresp onded to a set of renormalon singularities at the exp ected
k
p ositions. We shall demonstrate in section 2 of the present pap er that, for the Adler
D -function and the GLS sum rule, the N -expansion co ecients obtained by expanding
f
(1) (2)
2 3
d b and d b are in go o d (10{20% level) agreement with those of the exact O (a ) next-
1 2
to-next-to-leading order (NNLO) p erturbative calculations for these quantities, so that 2
the anticipated asymptotic dominance of the leading-b term is already apparentinlow
orders.
Given the dominance of the leading-b terms, an obvious prop osal is to sum them to
all orders. That is to split D into two comp onents:
(L) (NL)
D = D + D (4)
where `L' and `NL' sup erscripts refer to leading and non-leading terms in the b-expansion.
1
X
(k )
(L) k k +1
D d b a (5)
k
k =1
and
1 k 1
X X
(`)
(NL) k +1 `
D a d b : (6)
k
k=1 `=0
The summation of terms can b e achieved by using the Borel sum. The Borel integral can
itself b e split into two comp onents and is well de ned for the UV singularities on the
`
(L)
negative axis, which contribute p oles to the Borel transform of D . The integral can
b e p erformed explicitly in terms of exp onential integral functions and other elementary
(L)
functions. The piece of the Borel integral for D involving the IR singularities on
`
the p ositive real axis is formally divergent; but a principal value or other prescription
can b e used to go around the p oles. The sp eci cation of this prescription is intimately
linked to the pro cedure needed to combine the non-p erturbativevacuum condensates in
the op erator pro duct expansion (OPE) with the p erturbation theory in order to arriveat
awell-de ned result for D [9].
In recent pap ers by Neub ert [3] and by Ball, Beneke and Braun [1, 2] a summation
of the leading-b terms has also b een considered. In these pap ers it has b een motivated
as a generalisation of the BLM scale xing prescription [10 ] and termed \nave non-
ab elianization" [11]. The Neub ert pro cedure uses weighted integrals over a running cou-
pling. For the Euclidean Adler D -function this representation is equivalent to splitting
the Borel integral into ultra-violet renormalon and infra-red renormalon singularities and
principal value regulating the latter. When one continues to Minkowski space to obtain
+
the e e R-ratio and the -decay ratio, R , there are several inequivalentways to p erform
the continuation of the running coupling representation; and hence apparent additional
non-p erturbativeambiguities are claimed. In our view only a consideration of the singu-
larities in the Borel integral provides a satisfactory way of combining p erturbative e ects
with non-p erturbative condensates, along the lines discussed in reference [9]; and these
extra uncertainties are spurious.
+
Our intention in this pap er is to fo cus on the Adler D -function, the e e R-ratio, R
and the GLS sum rule (the latter was not considered in references [1{3]). For all of these
quantities there exist exact NNLO xed order p erturbative calculations and our interest is
in comparing the leading-b resummation with these exact xed order results. The large-b
results provide partial information ab out the Borel transform and the question is how
this can b est b e utilised. We discuss the renormalization scheme (RS) dep endence of
(L) (NL) (L)
the split b etween D and D in equation (4), the relative contribution of D b eing
RS-dep endent. This RS uncertainty needs to b e kept in mind and is carefully discussed. 3
The organisation of the pap er is as follows. In section 2 we shall discuss the b-expansion
for the Adler D -function and the GLS sum rule and will consider the extent to which the
dominance of the leading-b term is RS-dep endent. In section 3 the exact large-b results
are used to determine partially the Borel transforms for these quantities; and, having split
them into UV and IR renormalon pieces, a resummation along the lines discussed ab ove
is p erformed.
In section 4 the RS dep endence of the resummed results is considered, the numerical
results are presented and a comparison with the results of reference [3] and with the exact
NNLO p erturbative results is made. A discussion of the uncertainties and the prosp ects
for improving the resummation by including more of the exact structure of the Borel
transform is then undertaken. Section 5 contains overall conclusions.
~ ~
2 The b-expansions for D and K
We b egin by de ning the Adler D -function and the GLS sum rule, the quantities with
whichwe shall b e concerned in this pap er.
2 2
D (Q ) is related to the vacuum p olarization function, the correlator, (Q ), of two
vector currents in the Euclidean region,
Z
4
2 2 2 iq:x
(q q g q )(Q )=16 i d xe h0jTfJ (x)J (0)gj0i ; (7)
by
3
d
2
2 2
D (Q )=
Q (Q ) ; (8)
2
4 dQ
2 2
with Q = q > 0.
In p erturbation theory one has
0 1
2
X X
3
~
2 2
~ ~
@ A
D (Q )=d(R) Q 1+ D + D: (9) C Q
F f
f
4
f f
Here d(R) is the dimension of the quark representation of the colour group, d(R)=N for
SU(N ) QCD, and Q denotes the quark charges, summed over the accessible avours at a
f
~
3
~
given energy. D denotes corrections of the \light-by-light" typ e which rst enter at O (a )
~
and will b e subleading in N ; they are, it is to b e hop ed, small. D represents the QCD
f
corrections to the zeroth order parton mo del result and has the form
2 3 k+1
~
D = a + d a + d a + + d a + : (10)
1 2 k
We shall next consider two QCD deep inelastic scattering sum rules. The rst is the
p olarized Bjorken sum rule (PBjSR):
Z
1
ep-en
2
K g (x; Q )dx
PBj
1
0
1
g 3
A
~
=
1 C K : (11)
F
3 g 4
V 4
~
Here K denotes the p erturbative corrections to the zeroth order parton mo del sum rule,
2 3 k+1
~
K = a + K a + K a + + K a + : (12)
1 2 k
We can also consider the GLS sum rule,
Z
1
1
p+ p
2
K (x; Q )dx F
GLS
3
6
0
3
~
~ ~
= 1 C K + K : (13)
F
4
~
The p erturbative corrections, K , are the same as for the PBjSR; but there are additional
~ ~
~ ~
corrections of \light-by-light" typ e, K , analogous to D of equation (9). These will similarly
3
enter at O (a ) and b e subleading in N and we shall assume once again that they are
f
small.
~ ~
For b oth D and K the rst two p erturbative co ecients, d , d and K , K , are known
1 2 1 2
from the exact p erturbative calculations [12{15]. We shall assume
MS renormalization
with renormalization scale =Q for the present; but will later discuss RS dep endence
more generally.
Writing d and d expanded in N as in equation (2), the exact calculations give:
1 2 f
11 2 41 11 1
d = + N + C C ; (14)
1 3 f A 3 F
12 3 8 3 8
151 19 970 224 5
2
d = N + C + + N
2 3 A 3 5 f
f
162 27 81 27 9
29 19 10 90445 2737 55
2
+C + N + C
F 3 5 f 3 5
A
96 6 3 2592 108 18
127 143 55 23
2
+C C + + C : (15)
A F 3 5
F
48 12 3 32
Here and are Riemann -functions. For later comparisons it will b e useful to write
3 5
these results numerically for SU(N ) QCD.
:063
d = :115N + :655N + ; (16)
1 f
N
:024 :180
2 2
d = :086N + N 1:40N + 2:10N :661 : (17)
2 f
f
2
N N
Expanding d and d in b as in equation (3) gives
1 2
11 C C
F A
d = 2 ; (18) b +
1 3
4 12 8
151 19 5 5 31
2
d = b + C b
2 3 3 5 A
18 3 6 3 3
29 19 799
2
+C +10 b+C
F 3 5 3
A
32 2 288
827 23 11
2
C +C + + C : (19)
A F 3
F
192 2 32
As observed in reference [8], the b-expansion exhibits certain simpli cations relativeto
that in N . In particular the , present in all orders of the N -expansion for d , is present
f 3 f 1 5
only in the leading term in the b-expansion. For d the , present in all but the leading
2 5
(1)
term in the N -expansion, is now present only in d . In b oth cases the highest -function
f
2
(0)
present cancels and is absent in the `conformal' b ! 0 limit [16]; so d do es not involve
1
(0)
and d do es not involve . Since the MS b eta-function co ecients do not involve
3 5
2
-functions, this is presumably not an artefact of the particular RS chosen but maywell
b e of more fundamental signi cance. It may ultimately b e connected with the fact that
in the b ! 0 limit the z =2`=b U V and IR singularities in the Borel plane moveo to
` `
in nity, leaving only instanton singularities. As discussed in reference [8] the -functions
are intimately linked with the presence of renormalon singularities. It is amusing to notice
that in the original NNLO result for d [17], whichwas subsequently found to b e in error
2
(0)
20
[13], d do es contain a non-vanishing
term for SU(3) QCD. If a fundamental result
5
2
9
ab out the absence of -functions in the conformal limit could b e established it would have
enabled the incorrect result to have b een dismissed at once.
The corresp onding results for the N -expansion for the deep inelastic sum rules are:
f
1 23 7
K = N + C C ; (20)
1 f A F
3 12 8
!
115 3535 5 133 5
3
2
K = N + N + C + N + C
2 f 5 A f 3 F
f
648 1296 2 9 864 18
5437 55 1241 11 1
2 2
+C + C C + + C : (21)
5 A F 3
A F
648 18 432 9 32
In numerical form, for SU(N ) QCD,
:438
K = :333N + 1:48N + ; (22)
1 f
N
:244 :008
2 2
K = :177N + N 2:51N + 4:53N + :686 + : (23)
2 f
f
2
N N
Expanding in p owers of b as in equation (3),
C 7
A
K = b + C ; (24)
1 F
12 8
115 335 3 15 133 5
2
K = b + b + C + b C
2 3 5 A 3 F
72 144 2 9 288 6
179 11 11 389 1
2 2
+C + + C C + C : (25)
3 3 A F
A F
144 4 192 4 32
(0)
Notice that K do es not contain ; but a similar remark ab out the absence of in K
1 3 5
2
holds.
Wenow wish to demonstrate that the leading term in the b-expansion, when expanded
in N , approximates the N -expansion co ecients well, even in rather low orders.
f f
For d and d wehave
1 2
(1)
d b = :345b = :115N + :634N; (26)
f
1
(2)
2 2 2 2
d b = :776b = :086N :948N N +2:61N : (27)
f
2
f
2
The subleading, N , N N and N , co ecients approximate well in sign and magnitude
f
2
those in the exact expressions in equations (16) and (17). The leading, N and N ,
f
f
co ecients of course agree exactly. 6
For K and K wehave
1 2
(1)
K b = b = :333N +1:83N; (28)
f
1
(2)
2 2 2 2
K b = 1:59b = :177N 1:95N N +5:37N : (29)
f
2
f
2
The agreement with the exact N , N N and N co ecients in equations (22) and (23) is
f
again rather go o d.
Wenow turn to a consideration of the RS dep endence of the N and b expansions. In
f
variants of minimal subtraction, where the 1= p ole in dimensional regularization is sub-
tracted along with an N -indep endent nite part, K , the QCD p erturbative co ecients
f
will have the form of p olynomials in N as in equation (2). Mo di ed minimal subtrac-
f
tion (
MS), corresp onding to K =(ln 4 { ), with =0:5722 :::, Euler's constant, is most
E E
u
commonly employed. We can consider MS with renormalization scale =e Q, where u
is an N -indep endentnumb er. The most general subtraction pro cedure which will result
f
in p erturbative co ecients p olynomial in N ,however, can b e regarded as MS with scale
f
u+v=b
=e Q, where v is again N -indep endent. We shall refer to such renormalization
f
schemes as `regular' schemes. Of course the renormalization scheme is not sp eci ed by
the scale and subtraction pro cedure alone but by higher order b eta-function co ecients
as well. Anyvariant of minimal subtraction with an N -indep endent renormalization
f
scale will have v =0. Momentum space subtraction (MOM) based on the ggg vertex at a
2 2
symmetric subtraction p oint =Q [18] corresp onds to u=2:56 and v =C f ( ), where f
A
is a cubic p olynomial in the gauge parameter .For the Landau gauge, =0, v = 2:49C .
A
For other versions of MOM based on the qqg or ghost vertices, v will involve C and C .
A F
Let us denote the p erturbative co ecients in the
MS scheme with =Q (u=v =0) by
0
d ; and those with general u and v by d . Then
k
k
(1) (0)
0
d = (d + u)b +(d + v )
1 1
1
= d + bu + v: (30)
1
(1)
Changing the value of u shifts only the leading-b co ecient d and so do es not change the
1
(0)
relative contribution of the leading-b term. Changing v ,however, one can make the d
1
co ecient as large as one pleases and hence destroy the dominance of the leading-b term
noted ab ove for the D -function and the sum rules in low orders; although the leading-b
[k r ]
term should still repro duce asymptotically the d co ecients to O (1=k ) accuracy.
k
For d and higher co ecients the sp eci cation of the RS will involve higher b eta-
2
function co ecients as well as the scale and subtraction pro cedure. The RG-improved
2
coupling, a( ), will evolve with renormalization scale according to the b eta-function
equation
da
2 2 k
= ba (1 + ca + c a + + c a + ) : (31)
2 k
dln
Here b and c are universal with [19]
1
b = (11C 2N ) ;
A f
6
" #
2
7 C C C 3 11 5
A F
A
c = + : (32) + C C
A F
8 b 8 b 4 4 7
Integrating equation (31) with a suitable choice of b oundary condition [20], one obtains
a transcendental equation for a:
" #
Z
a
1 1 ca 1
b ln dx = + c ln + + ; (33)
2 2
a 1+ca x B (x) x (1 + cx)
0
2 3 k
where B (x)=(1 + cx + c x + c x + + c x + ). The b eta-function co ecients,
2 3 k
c , c , ::: together with b ln lab el the RS. In a xed order p erturbative calculation one
2 3
would truncate the b eta-function. For the all-orders resummations of the next section,
however, one requires an all-orders de nition of the coupling. In the MS scheme the higher
MS MS MS
b eta-function co ecients, c , c , :::, presumably exhibit factorial growth, c k !;
2 3 k
and the `a' coupling in the Borel integral would not b e de ned, since B (x)would itself
need to b e de ned by a Borel integral or other summation. One therefore needs to use
a nite scheme [21] where B (x) has a nite radius of convergence and can b e summed.
An extreme example is the so-called 't Ho oft scheme [21 ] where c =c ==c ==0,
2 3 k
B (x)=1 + cx. This results in the all-orders de nition of the coupling,
1 ca
b ln
= + c ln : (34)
a 1+ca
In such a nite scheme, where c , c , ::: are N -indep endent, the b-expansion of equa-
2 3 f
( 1)
tion (3) will contain an extra d =b term (for k>1) and the d will strictly no longer b e
k
k
p olynomials in N . bd ,however, is a p olynomial in N of degree k + 1. The leading-b co-
f k f
(k )
ecients, d , in regular schemes are indep endentof c ,c ,:::, since these b eta-function
2 3
k
co ecients, c , are O (1=N ) relativeto d .
k f k
3 Leading-b Resummations
Let us b egin by recalling the de nition of the Borel transform. If D has a series expansion
in `a' as in equation (1), we can write
Z
1