hep-ph/0007032

July 2000

arXiv:hep-ph/0007032v2 3 Nov 2000

Analytical Result for Dimensionally Regularized

Massless Master Double Box with One Leg off Shell

V.A. Smirnov1
Nuclear Physics Institute of Moscow State University
Moscow 119899, Russia
Abstract
The dimensionally regularized massless double box Feynman diagram
with powers of propagators equal to one, one leg off the mass shell,
i.e. with non-zero q 2 = p21 , and three legs on shell, p2i = 0, i = 2, 3, 4,
is analytically calculated for general values of q 2 and the Mandelstam
variables s and t. An explicit result is expressed through (generalized)
polylogarithms, up to the fourth order, dependent on rational combinations of q 2 , s and t, and a one-dimensional integral with a simple
integrand consisting of logarithms and dilogarithms.

E-mail: smirnov@theory.npi.msu.su.

Introduction

Massless four point Feynman diagrams contribute to many important physical amplitudes. They are much more complicated than two- and three-point diagrams because
depend on many parameters: the Mandelstam variables s and t and the values of the
external momenta squared, p2i , i = 1, 2, 3, 4. In the most general case, when all the
legs are off the mass shell, p2i 6= 0, there exists an explicit analytical result [1] for the
master (i.e. with powers of the propagators equal to one) double box diagram (see
Fig. 1) strictly in four dimensions. Still no similar results are available for pure off
p2

p1

6
r

p4

p3

7
r

Figure 1:
shell four point diagrams with ultraviolet, infrared and/or collinear divergences.
In the opposite case, when all the end-points are on shell, i.e. for p2i = 0, i =
1, 2, 3, 4, the problem of the analytical evaluation of such diagrams, in expansion in
= (4 d)/2 in the framework of dimensional regularization [2] with the space-time
dimension d as a regularization parameter, was completely solved during last year in
[3, 4, 5]. Among intermediate situations, when some legs are on shell and the rest
of them off shell, the case of one leg off shell, q 2 = p21 6= 0 and three legs on shell is
very important because of the relevance to the process e+ e 3jets (see, e.g., [6]).
The purpose of this paper is to analytically evaluate the master double box diagram
of such type, as a function of q 2 , s and t, and thereby demonstrate that the NNLO
analytical calculations for this process are indeed possible.
One of the ways to evaluate the four point diagrams with one leg off shell is
to expand them in the limit q 2 0 and compute as many terms of the resulting
expansion as possible. We explain how to do this, following the strategy of regions
[7, 8], in the next section and present the leading power term in this expansion which
provides a very non-trivial check of the subsequent analytical result.
To analytically evaluate the considered diagram we straightforwardly apply the
method of ref. [3]: we start from the alpha-representation of the double box and, after
expanding some of the involved functions in MellinBarnes (MB) integrals, arrive at
a six-fold MB integral representation with gamma functions in the integrand. Then
we use a standard procedure of taking residues and shifting contours to resolve the
structure of singularities in the parameter of dimensional regularization, . This
procedure leads to the appearance of multiple terms where Laurent expansion in
becomes possible. Resulting integrals in all the MB parameters but the last two are
evaluated explicitly in gamma functions and their derivatives. The last two-fold MB
integral is evaluated by closing an initial integration contour in the complex plane to
1

the right, with an explicit summation of the corresponding series. A final result is
expressed through (generalized) polylogarithms dependent on rational combinations
of q 2 , s and t and a one-dimensional integral with a simple integrand consisting of
logarithms and dilogarithms.

Expansion in the limit q 2 0

The dimensionally regularized master massless double box Feynman integral with one
leg off shell, q 2 = p21 6= 0, and three legs on shell, p2i = 0, i = 2, 3, 4, can be written as
2

F (s, t, q ; ) =

dd kdd l
(k 2 + 2p1 k + q 2 )(k 2 2p2 k)k 2 (k l)2
1
,
2
2
2
(l + 2p1 l + q )(l 2p2 l)(l + p1 + p3 )2
Z Z

(1)

where s = (p1 + p2 )2 , t = (p1 + p3 )2 , and k and l are respectively loop momenta of

the left and the right box. Usual prescriptions, k 2 = k 2 + i0, s = s + i0, etc are
implied. To expand the given diagram in the limit q 2 0 one can apply the so-called
strategy of regions [7, 8] based on the analysis of various regions in the space of the
loop integration momenta, Taylor expanding the integrand in the parameters that are
considered small in the given region and extending resulting integrations to the whole
integration domain in the loop momenta. When applying this strategy all integrals
without scale are by definition put to zero.
Let us choose, for convenience, the external momenta as follows:
p1 = p1

q2
p2 ,
Q2

p2 = p2 ,

p1,2 = (Q/2, 0, 0, Q/2),

where s = Q2 . The given limit |q 2 | |s|, |t| is closely related to the Sudakov limit
so that it is reasonable to consider each loop momentum to be one of the following
types:

hard (h):
k Q t ,
1-collinear (1c):
2-collinear (2c):

k+ q 2 /Q, k Q , k
k+ Q, k q 2 /Q , k

q 2 ,

q 2 .

Here k = k0 k3 , k = (k1 , k2 ). We mean by k Q, etc. that any component of k

is of order Q.
It turns out that the (h-h), (1c-h) and (1c-1c) are the only non-zero contributions
to the leading power behaviour in the limit q 2 0. Any term originating from the
(h-h) contribution is given by the expansion of the integrand in Taylor series in q 2 and
expressed through on-shell double boxes in shifted dimensions and can be analytically
2

evaluated by the algorithm presented in [4]. The (1c-1c) contribution is obtained by

expanding propagators number 2, 4 and 7 in a special way. In particular, propagators
number 2 and 4 are expanded, respectively, in l2 and k 2 . (See [8] for instructive 2-loop
examples of expansions in limits of the Sudakov type.)
The (1c-h) and (1c-1c) contributions are evaluated with the help of a two-fold
(respectively, one-fold) MB representation. Still this program of the evaluation of a
large number of terms of the expansion looks very complicated because one needs,
for phenomenological reasons, the values of q 2 greater than s and t so that a reliable
summation of a resulting series, using Pade approximants, requires the knowledge of
at least first 2030 terms. Such a great number of terms can be hardly evaluated
since a lot of irreducible structures appear. This asymptotic expansion is however
very useful for comparison with the explicit result derived below.
The leading power terms of the asymptotic expansion calculated in expansion in
, up to a finite part, are
F (s, t, q 2; ) =

i d/2 eE

2

(s)2+2 (t)

4
X

gi (X, Y )
+ O(q 2 ln3 (q 2 /s)) + O() ,
i

i=0

(2)

where X = q 2 /s , Y = t/s and

g4 (X, Y ) = 1 ,
g3 (X, Y ) = 2(ln X ln Y ) ,
11 2
3
g2 (X, Y ) =
+ 3 ln X ln Y ln2 Y ,
12
2

3
1
2 3
ln X ln2 X ln Y ln X ln2 Y
3
2
2


1 3
3
19
49(3)
ln Y + ln2 Y ln(1 + Y ) + 2
ln X
ln Y + ln(1 + Y ) +
,
6
2
6
6
g0 (X, Y ) = 26 Li4 (Y ) 2S2,2 (Y ) 2(ln X + 6 ln Y + ln(1 + Y )) Li3 (Y )


Y
+ (ln2 Y + 2 ln X ln Y + 4 2 ) Li2 (Y )
+2 ln Y Li3
1+Y
1 4
1 3
1
1
7
ln X + ln X ln Y + ln2 X ln2 Y ln X ln3 Y + ln4 Y
2
2
4
2
8


1 2
1
5 3
2
2
+ ln(1 + Y ) ln X ln Y ln Y + ln Y ln(1 + Y ) ln Y ln (1 + Y )
3
2
3

7
25
2
ln2 Y + ln X ln(1 + Y ) 2 ln Y ln(1 + Y )
+ 2 ln2 X ln X ln Y +
3
3
6



34
83 4
19
1 2
ln X
ln Y + 2 ln(1 + Y ) +
.
(3)
+ ln (1 + Y ) + (3)
2
3
3
180
g1 (X, Y ) = 2 ln Y Li2 (Y ) 2 Li3 (Y ) +

Here Lia (z) is the polylogarithm [9] and

(1)a+b1 Z 1 lna1 (t) lnb (1 zt)
dt
Sa,b (z) =
(a 1)!b! 0
t
3

(4)

the generalized polylogarithm [10].

From alpha parameters through MB representation to analytical result

The alpha representation of the double box looks like:



F (s, t, q 2 ; ) = (3 + 2) i d/2

2 Z

d1 . . .

d7

X

i 1 D 1+3 A32 ,

(5)

where
D = (1 + 2 + 7 )(3 + 4 + 5 ) + 6 (1 + 2 + 3 + 4 + 5 + 7 ) ,
(6)
A = [1 2 (3 + 4 + 5 ) + 3 4 (1 + 2 + 7 ) + 6 (1 + 3 )(2 + 4 )](s)
+5 6 7 (t) + 5 [(1 + 3 )6 + 3 (1 + 2 + 7 )](q 2 ) .
(7)
As it is well-known, one can choose a sum of an arbitrary subset of i , i = 1, . . . , 7
in the argument of the delta function in (5), and we use the same choice as in [3].
Starting from (5) we perform the same change of variables as in [3] and apply
seven times the MB representation
1 1
1
=

(X + Y )
() 2i

+i

dw

Yw
( + w)(w)
X +w

(8)

in order to separate terms in the functions involved to make possible an explicit

parametric integration. The two extra MB integrations arise form the extra term
with q 2 . After such integrations we are left with a 7-fold MB integral of a ratio of
gamma functions. Fortunately, one of the integrations can be explicitly taken using
the first Barnes lemma and we arrive at the following nice 6-fold MB integral:


i d/2

2

!v  
w

t
q2
1 Z
2
F (s, t, q ; ) =
dvdwdw
dw
dzdz
2
3
1
3+2
6
(1 3)(s)
(2i)
s
s
(1 + w)(1 + v + w)(v)(w)(1 w3 + v)
(w2 )(1 2 w w2 )(w3 v)(1 2 w w3 )
(1 w2 + z1 )(1 w3 + z1 )( + w + w2 + w3 z1 )(z1 )

(1 + w + w2 + w3 )(1 4 w w2 w3 )(1 w3 )
(1 + z)(2 + 2 + w + w2 + z z1 )(2 + 2 + w + w3 + z z1 )
(2 3 w w2 w3 + z1 z)(z1 z)

.
(3 + 2 + w + z)

(9)

It differs from its analog for q 2 = 0 by the additional integration in v. This variable
enters only four gamma functions in the integrand. The integral is evaluated in
4

expansion in , up to a finite part, by resolving singularities in absolutely by the

same strategy as in the case q 2 = 0 [3]. Note that the infrared and collinear poles are
a little bit softer than in the pure on-shell case, the integration variable v playing the
role of an infrared regulator. The two key gamma functions that are responsible for
the generation of poles in are the same as in the previous case:
( + w + w2 + w3 z1 ) (2 3 w w2 w3 + z1 z) .
The labeling of resulting terms is therefore the same: the initial integral is decomposed
as J = J00 +J01 +J10 +J11 , etc. (Only arguments of some gamma functions are shifted
by v.) The applied strategy makes it possible to perform all the integrations apart
from the last two, in v and w. We obtain four groups of terms with 26 terms in each
group: the terms without MB integration, with MB integration in v or w and, finally,
with a two-fold integration in v and w. The one-fold integrals are explicitly evaluated
by closing contour and summing up series, using formulae from [11].
The contribution of the resulting two-fold MB integral takes the form


2 i d/2
s3

2

1 Z dvdw
(2i)2 1 + w

q2
s

!v  
w

t
s

(1 + v + w)(v)(1 + w)(w)2

5
1
1
E 2 ln(s)

1+w 1+v+w
+(1 + v) 2(v w) 3(w) + 2(1 + w) + (1 + v + w))
(1 + v)(v)(1 + w)(w)] .


(1 + v + w)(v w)

(10)

The integration contours are straight lines along imaginary axes with 1 <Re v,
Re w,Re v + w < 0. By closing contours it is possible to convert this integral into a
two-fold series where each term is identified as a derivative of the Appell function F2
in parameters, up to the third order. The 1/ part is then explicitly summed up with
a result in terms of polylogarithms. (In fact, it is proportional to the part of the
master one-loop box.)
The so obtained result can be transformed into a one-dimensional integral with a
simple integrand. To present the final result let us turn to the variables x = s/q 2 and
y = t/q 2 keeping in mind typical phenomenological values of the involved parameters
relevant to the process e+ e 3jets:
F (s, t, q 2; ) =

i d/2 eE

2

s2 t(q 2 )2

4
X

fi (x, y)
+ O() .
i
i=0

(11)

We obtain
f4 (x, y) = 1 ,
f3 (x, y) = 2(ln x + ln y) ,

(12)
(13)
5

f2 (x, y) = 3 Li2 (x) + Li2 (y) 2(ln x + ln y)2

+3 ln(1 x) ln x + ln(1 y) ln y
"

f1 (x, y) = 2 Li3
ln x Li2

x
1xy

+ Li3

x
y
ln y Li2
1x
1y


5 2
,
12
!

y
1xy

!#

Li3

(14)
xy
1xy

+ 2 ln(1 x y)


1 2
ln (1 x y) + 2 + ln(1 x) ln x + ln(1 y) ln y ln x ln y
6
!


y
x
2 Li3
+3 Li3 (x) 8 Li3 (y) + 4 Li3
1x
1y
2
4
(3 ln x + 4 ln y)Li2 (x) + 3 ln y Li2 (y) + ln3 x ln3 (1 x) + ln2 (1 x) ln x
3
3
2
9

4
1
ln(1 x) ln2 x + (5 ln x 4 ln(1 x)) + ln3 y + ln3 (1 y)
2
6
3
3
2

2 ln2 (1 y) ln y ln(1 y) ln2 y + (5 ln y + 2 ln(1 y))

6
25(3)
.
(15)
+4 ln x ln y (ln x ln(1 x) + ln y) +
6


The 0 part involves a one-dimensional integral:

f0 (x, y) =

dz z 1 ln(1 z)(4 ln2 (1 x yz) ln2 (1 y xz))

4y

[ln(1 yz) (ln(1 z) ln(1 yz) 2 Li2 (z))

1 x yz
2(ln(1 z) ln z) Li2 ((1 x yz)/x)]
h


x
ln(1 xz) 3 ln2 (1 z) 6 ln(1 z) ln(1 xz) + 2 Li2 (z)

1 y xz
+2(6 ln(1 z) ln z) Li2 ((1 y xz)/y)]}
!
!


x
x
xy
5 Li4 (x) + 14 Li4
2 Li4
6 Li4
1y
1x
(1 x)(1 y)
!
!


y
y
x
+ 24 Li4 (y) 2 Li4 (1 y) + 8 Li4
2 Li4
+8 Li4
1xy
1y
1x
!
!


y
1xy
1xy
8 Li4 (1 x) 8 Li4
20 Li4
+ 10 Li4
1xy
1y
1x
!
x
+ (2 ln y 2 ln x 3 ln(1 x)) Li3 (x)
3S2,2 (x) 8S2,2 (y) 6S2,2
1y
!
x
+2(16 ln(1 y) 11 ln y 2 ln(1 x y) + ln x) Li3
1y
6

(8 ln y + 2 ln(1 x) + 3 ln x) Li3

x
1x

xy
2(4 ln(1 y) 4 ln y ln(1 x) + ln x)Li3
(1 x)(1 y)
!
!
xy
x
+ 2 ln y Li3
+(14 ln(1 y) 18 ln y + 4 ln(1 x y))Li3
1xy
1xy
!
y
+(7 ln y 8 ln(1 y))Li3 (y) + (8 ln(1 y) + ln y + 2 ln x)Li3
1y


y
4(2 ln y + 7 ln(1 x) + 2 ln(1 x y) 8 ln x)Li3
1x !
y
2(ln y + 5 ln(1 x) + 8 ln(1 x y) 7 ln x)Li3
1xy

x
1
(Li2 (x))2 Li2
2
1y
h

!!2

3
y
(Li2 (y))2 + 4 Li2
2
1x


2

+ ln2 (1 x) 4 ln2 y + 2 ln y (4 ln(1 x) ln x) 2 ln(1 y) ln x 3 ln(1 x) ln x

#

h
5 2
7
Li2 (x) + 12 ln2 (1 y) + 15 ln2 y + 2 ln(1 y)(ln(1 x y) + ln x
+ ln2 x +
2
3
9 ln y) + 2 ln y (ln(1 x y) 4 ln(1 x) + 4 ln x)
!
h
i
x
2
2
+ 4 ln2 (1 y) 11 ln2 y
2(ln (1 x y) + ln x) Li2
1y
#
2
2
Li2 (y)
+2 ln y (4 ln(1 x) 3 ln x) + ln(1 y)(5 ln y 2 ln x) + ln x
3
h

+ 8 ln2 y 8 ln y ln(1 x) 10 ln2 (1 x) + 8 ln2 (1 x y) 8 ln(1 x y) ln x

#

2 2
y
8 ln(1 x)(ln(1 x y) 2 ln x) + 2 ln(1 y) ln x + ln x
Li2
3
1x
2

+ ln2 (1 x) 4 ln2 (1 y) 8 ln2 y + 2 ln y (4 ln(1 x) 3 ln x)

2

+2 ln(1 y)(4 ln y ln x) 2 ln(1 x) ln x + 2 ln x Li2

xy
(1 x)(1 y)

1 3
ln (1 x y) [2 ln y 3 ln(1 y) 9 ln x 11 ln(1 x)]
h 3
+ ln2 (1 x y) 2 3 ln2 (1 y) + 3 ln2 y + 6 ln2 (1 x) ln(1 y)(ln y 10 ln x)
i
h
1
+4 ln(1 x) ln x 2 ln2 x ln y (5 ln(1 x) + ln x) + ln(1 x y) 7 ln3 (1 y)
3
2
2
2
2
2 ln(1 y)(5 + 12 ln y) + 7 ln(1 x) + ln x + 6 ln2 y ln x 4 ln3 (1 x)
+15 ln2 (1 y)(ln y 2 ln x) 21 ln2 (1 x) ln x + 3 ln(1 x) ln2 x
+2 ln4 (1 x y) +

+ ln y (2 2 + 9 ln2 x + 15 ln2 (1 x) 6 ln(1 x) ln x)

7

5
2
23 3
17 2
7
ln4 (1 x) ln4 x +
ln (1 x) ln x
ln (1 x) ln2 x + ln(1 x) ln3 x
6
3
6
4
2
2 

2

3 ln2 (1 x) 10 ln(1 x) ln x + 5 ln2 x ln4 (1 y) ln4 y

6
3
2
19 3
2
2
3
ln (1 y) ln y + 5 ln (1 y) ln y + ln(1 y) ln y
6
3

2  2
+
9 ln (1 y) ln(1 y) ln y 5 ln2 y
6
8
1
+ ln(1 x) ln(1 y)(ln2 (1 x) 4 ln2 (1 y)) (ln2 x + ln2 y) ln x ln y
3
3
3
2
+3 ln (1 y) ln x + ln (1 y) [ln y(4 ln(1 x) + ln x) 2 ln(1 x) ln x]
h
1
+ ln y ln3 (1 x) 9 ln2 (1 x) ln x 12 ln y ln2 x
3
h
+6 ln(1 x) ln x (2 ln y + 3 ln x)] ln(1 y) 8 ln2 y ln(1 x)
+ ln(1 x) ln x (ln(1 x) 2 ln x) + ln y (8 ln2 (1 x) + 6 ln(1 x) ln x + ln2 x)

2
[ln y (4 ln(1 x) 5 ln x) ln(1 y) ln x]
3


+(3) 12(ln(1 x y) ln(1 y)) + 13 ln(1 x)

25
23 4
(ln x + ln y) +
. (16)
3
180


One may hope that the one-dimensional integral that is left can also be evaluated in
terms of polylogarithms. To do this it is necessary to complete the table of integrals
derived in [9].
This result is in agreement with the leading power behaviour when q 2 0 (3).
When performing this comparison it is reasonable to start with (10), take minus
residue at v = 0 (the first pole of (v)), integrate in w by closing the contour to the
right, and take into account the three other contributions (without MB integration,
and with integration in v or w) that were not presented above. Eqs. (1216) also
agree with results based on numerical integration in the space of alpha parameters
[12] (where the 1% accuracy for the 1/ and 0 parts is guaranteed).
Acknowledgments. I am grateful to Z. Kunszt for involving me into this problem and for kind hospitality during my visit to ETH (Z
urich) in AprilMay 2000
where an essential part of this work was performed. I am thankful to T. Binoth and
G. Heinrich for comparison of the presented result with their results based on numerical integration. Thanks to A.I. Davydychev and O.L. Veretin for useful discussions.
This work was supported by the Volkswagen Foundation, contract No. I/73611, and
by the Russian Foundation for Basic Research, project 980216981.

References
8

[1] N.I. Ussyukina and A.I. Davydychev, Phys. Lett. B298 (1993) 363.
[2] G. t Hooft and M. Veltman, Nucl. Phys. B44 (1972) 189; C.G. Bollini and
J.J. Giambiagi, Nuovo Cim. 12B (1972) 20.
[3] V.A. Smirnov, Phys. Lett. B460 (1999) 397.
[4] V. A. Smirnov and O. L. Veretin, Nucl. Phys. B566, 469 (2000) [hepph/9907385].
[5] J.B. Tausk, Phys. Lett. B469 (1999) 225; C. Anastasiou, E.W.N. Glover and
C. Oleari, Nucl. Phys. B575 (2000) 416; C. Anastasiou, T. Gehrmann, C. Oleari,
E. Remiddi and J.B. Tausk, hep-ph/0003261; T. Gehrmann and E. Remiddi,
hep-ph/0005232; C. Anastasiou, J.B. Tausk and M.E. Tejeda-Yeomans, hepph/0005328.
[6] R.K. Ellis, D.A. Ross and A.E. Terrano, Nucl. Phys. B178 (1981) 421.
[7] M. Beneke and V.A. Smirnov, Nucl. Phys. B522 (1998) 321.
[8] V.A. Smirnov and E.R. Rakhmetov, Teor. Mat. Fiz. 120 (1999) 64; V.A. Smirnov,
Phys. Lett. B465 (1999) 226.
[9] L. Lewin, Polylogarithms and associated functions (North Holland, 1981).
[10] K.S. Kolbig, J.A. Mignaco and E. Remiddi, B.I.T 10 (1970) 38; K.S. Kolbig,
Math. Comp. 39 (1982) 647; A. Devoto and D.W. Duke, Riv. Nuovo Cim. 7,
No. 6 (1984) 1.
[11] J. Fleischer, A.V. Kotikov and O.L. Veretin, Nucl. Phys. B547 (1999) 343.
[12] T. Binoth and G. Heinrich, hep-ph/0004013.