hep-ph/9709423

UCLA/97/TEP/23
February 1, 2008

arXiv:hep-ph/9709423v1 23 Sep 1997

Feynman Diagrams and Cutting Rules

J.S. Rozowsky1
Department of Physics, University of California at Los Angeles, CA 90095-1547

Abstract
We show how Feynman diagrams may be evaluated to take advantage of recent
developments in the application of Cutkosky rules to the calculation of one-loop amplitudes. A sample calculation of gg gH, previously calculated by Ellis et al., is
presented illustrating this equivalence. This example demonstrates the use of cutting
rules for massive amplitudes.

Email address: [email protected]

Introduction

One-loop calculations are important in the quest for discovering new physics beyond the
Standard Model, especially for determining QCD backgrounds. A number of signicant improvements in the calculation of one-loop amplitudes, especially ones with massless particles,
have been made in recent years. These methods, which have recently been reviewed in ref. [1],
include the use of helicity [2], color decompositions [3], string-inspired ideas [4], recursion
relations [5, 6], unitarity [7, 8, 9, 10] and factorization [11]. Besides allowing for the computation of certain innite sequences of amplitudes, these techniques have been used in the
computation of non-trivial amplitudes relevant for phenomenological applications [12, 13].
A technique that has proven to be especially useful is that of using unitarity to determine
the functional form of the amplitudes. By working to all orders in the dimensional regularization parameter, = (4 D)/2, one can determine the coecients of all integral functions
that may appear [14]. For certain amplitudes that satisfy a power counting criterion, such as
supersymmetric ones, one can show that cuts with four-dimensional momenta are sucient
to determine the amplitudes through O(0 ) [8, 9]. The case of an amplitude with massive
internal legs has been considered in ref. [10].
The eciency of the unitarity method stems from the use of gauge invariant tree amplitudes in performing calculations. With the use of helicity and color decompositions this
can lead to enormous reductions in the complexity of intermediate expressions. This may
be contrasted against calculations of individual Feynman diagrams that are gauge variant,
where large cancelations must take place in order obtain simpler gauge invariant results.
In this paper we will show how to evaluate Feynman diagrams so as to obtain eciency
of computation inherent in the unitarity technique. This is done by performing Feynman
diagram calculations in a particular way so as to mimic the unitarity approach advocated in
ref. [1]. One may therefore start with Feynman diagrams, yet obtain the advantages of the
Cutkosky approach. This is useful since it explains the unitarity method from a conventional
Feynman diagram viewpoint. This is of interest in, for example, the extension of unitarity
methods beyond one loop, such as in ref. [15].
As an explicit example we have calculated the one-loop helicity amplitudes for gg gH.
This example has the desirable features that it contains both colored and colorless lines and
massive internal and external legs. This example is useful since it illustrates features beyond
purely massless calculations, where most attention has been concentrated [1]. Also it has
already been evaluated via conventional methods by R.K. Ellis et al. [16], allowing for an
easy comparison to the correct result to be made. We nd agreement with their result.

Review of basic tools

We now briey review some of the techniques that we shall use in this paper. Two important
techniques that we utilize are spinor helicity and color decomposition. The reader may
consult review articles [17] for further details.

2.1

Spinor helicity

In explicit calculations with external gluons, it is usually convenient to use a spinor helicity
basis [2] which rewrites all polarization vectors in terms of massless Weyl spinors |k . In
the formulation of Xu, Zhang and Chang the plus and minus helicity polarization vectors
are expressed as
+ (k; q) =

q | |k

,
2 qk

(k; q) =

q+ | |k +

,
2 [k q]

(2.1)

where k is the momentum of the gluon and q is an arbitrary null reference momentum
which drops out of the nal gauge invariant amplitudes. We use the convenient notation
+
ki |kj ij ,

+
ki |kj [ij] .

(2.2)

These spinor products are anti-symmetric and satisfy i j [j i] = 2ki kj . One useful identity
is

(2.3)
a| / (k; q) |b = a| |q k | + |k q | |b ,

q |k
where either a| or |b are spinors with four-dimensional momenta.
It is convenient to choose a dimensional regularization scheme that is compatible with the
spinor helicity formalism. We use the four-dimensional helicity scheme [4] which is equivalent
to a helicity form of Siegels dimensional reduction scheme [18]. The conversion between the
various dimensional regularization schemes has been given in ref. [4, 19].

2.2

Color decomposition

Color decompositions [3] have been extensively discussed in review articles [17]. Here we
only present the color decomposition for the gg gH amplitude.
For this one-loop amplitude the decomposition is
A1loop = g 3 2
4
R

tr[T a(1) T a(2) T a(3) ] Af ((1), (2), (3), H) ,

4

(2.4)

where we have taken the loop to be in the fundamental representation. The sum over
includes all cyclic permutations of the indices (n) (i.e. {(1, 2, 3), (1, 3, 2)}) and the T a
are fundamental representation color matrices (normalized so that tr(T a T b ) = ab ). We have
abbreviated the dependence of Af on the outgoing momenta kj and helicities j by writing
4
the label j alone. We have also explicitly extracted the coupling and a factor of 2 from Af ,
4
R
where R is the renormalization scale. For adjoint representation loops there is an analogous
decomposition with up to two color traces in each term.
It is convenient to further decompose the partial amplitude in terms of primitive amplitudes,
Af ((1), (2), (3), H)
4
= A4 ((1), (2), (3), H) + A4 ((3), (1), (2), H) + A4 ((2), (3), (1), H) .
(2.5)
2

In the primitive amplitudes appearing on the right-hand-side we may treat the Higgs as carrying color charge (in the adjoint representation), which couples to gluons; it is not dicult
to show in the permutation sum all such additional terms cancel. This is convenient since
these primitive amplitudes are gauge invariant and have a xed cyclic ordering of external
legs. A similar decomposition in terms of primitive amplitudes can also be performed for
the case where some of the external particles are in the fundamental representation [20].

Feynman diagrams

Feynman diagrams, which have been essential for evaluating terms in the perturbative series for scattering processes, suer from a number of well known computational diculties.
Firstly, individual Feynman diagrams are inherently gauge dependent; gauge invariance is
restored only after summing over all diagrams. Secondly, when evaluating loop momentum
integrals using traditional techniques, such as Passarino-Veltman reduction [21], one encounters a large number of spurious denominators. These spurious denominators typically involve
Gram determinants, det(ki kj ) raised to a power. Enormous cancelations must then take
place before one obtains relatively simple results.
When using conventional techniques to evaluate Feynman diagrams one reduces all tensor
integrals into sums of scalar integrals multiplied by reduction coecients. (By a tensor
integral we mean loop momentum integrals with powers of momentum in the numerator and
by a scalar integral we mean integrals with no powers of loop momentum in the numerator,
see appendix A.) With the knowledge of how this reduction works, given any one-loop
amplitude we can list all the scalar integrals that could potentially enter. Thus we have
Amplitude =

Feynman diagrams

ci Ii ,

(3.1)

where the subscript i is some index that labels the scalar integrals that contribute to this
amplitude. The heart of the calculation is in evaluating all the coecients of the scalar
integrals, i.e. the ci s.
A conventional Feynman diagram technique like Passarino-Veltman (PV) reduction evaluates all the necessary coecients simultaneously. If one could obtain these coecients in a
more ecient manner one would greatly improve the ability to calculate.
Some of the features that are useful in improved calculational methods are: (1) that the
building blocks be gauge invariant combinations of diagrams instead of individual diagrams;
(2) that wherever possible tensor integrals should not generate large powers of Gram determinants in denominators; (3) that large cancelations between diagrams be avoided as much
as possible; (4) and that calculations can be recycled when performing new ones.

i1

i2

`1

+1

i1

i2

`2

Figure 1: A cut of a general n-point one-loop amplitude.

3.1

Review of the unitarity method

The cutting method reviewed in ref. [1] has these above properties. In this method one
writes down the cuts in a given channel, depicted in g. 1,
An (1, 2, . . . , n)

cut

i
i
d42 p
Atree (1 , i1 , . . . , i2 , 2 ) 2
Atree (2 , i2 + 1, . . . , i1 1, 1 ) .
n1
42 2 m2
(2)
2 m2 n2
cut
1
1
2
(3.2)
This equation is valid only for the cut channel under consideration. The loop momentum is
p; 1 and 2 are the momenta of the cut propagators, with masses m1 and m2 respectively
and n1 + n2 = n + 4. One then reconstructs the complete amplitude by nding a function
which has the correct cuts in all channels.
Since the loop integral in eq. (3.2) is composed of products of on-shell tree amplitudes,
the basic building blocks are gauge invariant. These tree amplitudes are inherently easier
to calculate than loop amplitudes and in many cases have been previously calculated. A
key feature of the cutting method is that a specic subset of scalar integral coecients are
obtained in any given channel. Then one repeats this process for various channels thus
computing the coecients for all the scalar integrals. There are integrals that do not appear
in any channel, namely the tadpole and scalar bubble integrals,
D=42
I1

D=42 2
lim I2
(k ) .

and

k 2 0

(3.3)

However, with prior knowledge of the infrared and ultraviolet behavior of an amplitude in
many cases the coecients of these integrals can be xed, as discussed in ref. [10].
It is generally convenient to express the tree amplitudes in eq. (3.2) using the spinor
helicity formalism discussed in section 2.1. Then one can evaluate the right-hand side of
eq. (3.2) by constructing spinor strings of the form a| /1 /2 |b , and then commuting

the /i towards each other. This generates terms of the form i kj , which may be rewritten as

dierences of inverse propagators and terms which do not depend on the loop momentum.
This type of reduction procedure has been previously used in a number of references [8, 9,
10, 22]. We shall make use of it in the next section.
4

Figure 2: The contribution from a vector in the loop is separated into the dierence of a
vector and scalar plus a scalar so that the dierence satises the power-counting criterion.
When using spinor helicity one must take into account that the momenta of the intermediate on-shell lines are in (4 2) dimensions while the helicity formalism is dened in
four dimensions. If one drops the (2)-dimensional components of the momenta, one could
make an error in rational functions arising from / terms. For ve- and higher-point calculations it can be more convenient to use four-dimensional momenta in the cuts, but then
to reconstruct the missing rational functions using collinear and multi-particle factorization
properties [11, 1]. For four-point calculations it is convenient to keep terms to all orders in ,
since factorization is insucient to x all rational functions.
The cases of fermion or scalar lines with (4 2) momenta in the helicity formalism
have been extensively discussed in ref. [10] and do not present any diculties. Although
it is unnecessary for the specic calculation gg gH, a convenient strategy for handling
channels with gluon lines is to subtract and add equivalent diagrams with internal gluon lines
replaced by scalar or fermion lines, as depicted in g. 2. In some cases one uses fermions so
as to group particles together with supersymmetric partners. Such combinations of diagrams
satisfy the power counting criterion that n-point tensor integrals should have (for n > 2)
at most n 2 powers of loop momentum in the numerator of the integrand; two-point
integrals should have at most one power of loop momentum. The required rearrangement
of the diagrams to make the power count apparent is discussed in ref. [9]. In particular, for
the example portrayed in g. 2 one should use background eld Feynman gauge [23]. The
origin of this power counting criterion is from the requirement that all contributions due to
the O() parts of loop momenta should not interfere with ultraviolet poles to yield O(0 )
contributions. (One can also show that infrared divergences never interfere with the O()
parts of loop momenta [9].) This criterion also holds for the general massive amplitudes
discussed in this paper. This is useful since one can then take the sewed gluon lines to have
four-dimensional momenta without any errors through O(0 ).

3.2

From Feynman diagrams to the unitarity method

We now describe how to perform a Feynman diagram calculation so as to mimic the cutting
method. In the next section we give an explicit example involving both internal and external
masses. The basic idea is rather simple.
Given that we know all the scalar integrals that could possibly enter a one-loop amplitude,

suppose we target those scalar integrals that have a pair of scalar propagators
i/(2 m2 )
1
1

and

i/(2 m2 ) ,
2
2

(3.4)

where 1 is the momentum of the scalar propagator between external legs i1 1 and i1 , and
2 is between i2 and i2 + 1 of a general n-point amplitude, where the indicated propagators
are each of a specic particle type. In order to generate a scalar integral with a specic propagator one must necessarily have evaluated a Feynman diagram that has the corresponding
propagator.
We can dene a channel (this channel will correspond to the cut-channel when the cut
is performed on the indicated legs) as a subset of the terms in the full amplitude; the terms
being the scalar integrals which include the pair of scalar propagators mentioned above
together with their corresponding coecients, the ci s.
In each Feynman integral (the tensor integral corresponding to each Feynman diagram)
we explicitly extract a denominator that includes the pair of scalar propagators under consideration. Then if we impose the on-shell conditions 2 m2 = 2 m2 = 0 to the numerators
1
1
2
2
of all the Feynman integrals contributing to the amplitude then the result would yield exactly what we dened as a channel in the previous paragraph. The potential errors when
imposing these on-shell conditions would necessarily be of the form
dD (2 m2 )n (2 m2 )m f (1 , 2 , . . .)
1
1
2
2
,
(2)D
(2 m2 )(2 m2 )
1
1
2
2

(3.5)

where n 1 and/or m 1, and D = 4 2. Since (2 m2 ) and (2 m2 ) are themselves

1
1
2
2
inverse scalar propagators, anything of the form of eq. (3.5) could only generate terms which
are not in the channel under consideration. This is because one or both of the inverse
propagators would cancel a scalar propagator in the denominator and generate a scalar
integral not included in the specied channel, i.e. that does not include both of the indicated
scalar propagators. Setting 2 m2 = 2 m2 = 0 only aects the calculation of those terms
1
1
2
2
not in the specied channel. Thus the evaluation of the ci s that are in the specied channel
are not aected by imposing the on-shell condition.
By systematically stepping through a sucient number of channels one can x all the
coecients, the ci s in eq. (3.1). In this way we can reconstruct the full amplitude. For some
coecients the corresponding integrals appear in multiple channels; for consistency these
must coincide.
So how does using the on-shell conditions 2 m2 = 2 m2 = 0 in the numerator help
1
1
2
2
us in evaluating the Feynman diagrams? All Feynman diagrams in a specic channel can
now be factorized using their on-shell legs into the sewing of two Feynman tree diagrams
(see g. 3).
In gure g. 3 the factorized Feynman graph only contains intermediate on-shell fermions
for the gg gH amplitude. But in general this factorization process can be performed
on any one-loop Feynman diagram. With on-shell scalars or fermions it is clear that this
can be done. With intermediate massless gauge particles one might worry that one has
forgotten about the contributions of diagrams with ghosts on the right hand side of g. 3.
6

=
helicity

dD
(2)D

i
2i
2) ( 2m2)
( 1m1 2 2
2

m2 =
1

2
1

m2 = 0
2

2
2

Figure 3: An example of a Feynman graph for the gg gH amplitude that contributes to

the s-channel, factorized into the sewing of two trees. The dashed lines on the left hand
side indicate the on-shell propagators and the helicity sum is over the polarizations of the
factorized legs.
We know, however, that the net eect of the one-loop diagrams with ghosts is to cancel the
unphysical polarizations of graphs with corresponding gauge particles. This cancelation can
be made apparent using the background eld Feynman gauge [23]. However, once one has
factorized a Feynman graph on a pair of propagators that includes a gauge particle then
the contribution of the ghosts is not necessary as the sewing of the two on-shell Feynman
trees manifestly has no unphysical gauge degrees of freedom.
If one sums over all the Feynman diagrams that contribute to a specic channel and we
impose the on-shell condition then the equivalent equation to g. 3 would have on-shell tree
amplitudes sewed together instead of on-shell Feynman trees:
An (1, 2, . . . , n)|channel =
helicity

d42 tree
i
i
An1 (1 , i1 , . . . , i2 , 2 ) 2
(2)42
(1 m2 ) (2 m2 )
1
2
2
Atree (2 , i2 + 1, . . . , i1 1, 1 )
n2

2 m2 =2 m2 =0
1
1
2
2

(3.6)
The sum on the right-hand side runs over all helicity states that can propagate across
the cut legs. Using the above equation one can compute all the coecients of the scalar
integrals that appear in this channel.
The above equation, eq. (3.6), is the unitarity method mentioned earlier in this section.
Originally the method was inspired by Cutkosky rules, however, from the arguments presented here it is clear that one can evaluate Feynman diagrams in a sophisticated manner so
as to take advantage of the benets inherent in the unitarity method. Instead of performing
the calculation of the scalar integral coecients in parallel, we perform the calculation by
targeting simultaneously only those integrals in a given channel.
The notion of channels presented in this section is more general than that of just considering Cutkosky rules because we may use the on-shell conditions on any propagator which
does not collapse in the integral function under consideration. (Indeed, this observation has
already been used in calculations of Z 4 partons [13].) In general we could construct
a channel where we use the on-shell condition for any number of propagators (it is even
possible to construct a channel where the on-shell condition is used for a single propagator);
however, it is normally convenient to use only the two-propagator channel construction.

Calculation of gg gH

In this section we apply the procedure outlined in the previous section to the specic example
of gg gH. This amplitude has one massive external leg and all the Feynman diagrams
have a massive quark running in the loop. See g. 4 for the color-ordered Feynman diagrams
contributing to this amplitude.

Figure 4: The color-ordered Feynman diagrams contributing to the amplitude gg gH.

In g. 4 there are two Feynman diagrams that involve a Higgs-Higgs-gluon coupling which
does not exist in the Standard Model. As mentioned in section 2.2, although these diagrams
cancel from the nal result, they make the primitive amplitudes (which have a xed ordering
of external legs) gauge invariant. Of course, one does not need to include the extra diagrams,
but then one would need to sum over the permutations with all possible external orderings
of the Higgs, as in eq. (2.5), to recover gauge invariance.
As mentioned in section 3 by observing the Feynman diagrams in g. 4 we can write down
all the scalar integrals that can enter the amplitude, see g. 5. The integrals are two-, threeand four-point scalar integrals that have massive internal propagators and some combination
of massive external kinematics. One-point tadpole integrals do not appear since they are
forbidden by power counting. (When using Passarino-Veltman reduction one could encounter
such integrals, but they must cancel in the nal expressions for this amplitude since one does
not encounter them when using Feynman parameterization.) The scalar integrals are found
in appendix A.1.
In addition to the use of color decomposition, section 2.2, it is also convenient to decompose the primitive amplitudes into all possible external helicity congurations. So the objects
that need to be computed are the color ordered gauge invariant primitive amplitudes. There
are three such objects; they are A4 (1+ , 2+ , 3+ , H), A4 (1+ , 2+ , 3 , H) and A4 (1+ , 2 , 3+ , H).
8

c1

c2

c3

c4

c5

c6

c7

c8

c9

Figure 5: The gg gH amplitude parameterized in terms of all the scalar integrals that
might enter. The external legs are labeled clockwise starting at the top right.
The other primitive amplitudes (e.g. A4 (1 , 2 , 3+ , H)) can be obtained from the above
three primitive amplitudes using parity and relabelings.
Since this amplitude has already been calculated by Ellis et al. [16] the purpose of the
calculation is to explicitly demonstrate that we may obtain the correct results by applying the
unitarity method to massive amplitudes, starting from Feynman diagrams. For illustrative
purposes, only the calculation of A4 (1+ , 2+ , 3+ , H) is explicitly presented here the results
of the calculation of the other two primitive amplitudes are shown for completeness.

4.1

The A4 (1+, 2+, 3+, H) primitive amplitude

In order to compute the primitive amplitude A4 (1+ , 2+ , 3+ , H) we will construct it from

multiple channels. We will see that it is sucient to only compute the s- and t-channels.
The Mandelstam variables are dened as s (k1 + k2 )2 and t (k2 + k3 )2 . These channels
can be seen in g. 6.
We begin by rst constructing the s-channel by sewing together the two tree amplitudes
on the left- and right-hand side of g. 6a. This corresponds to targeting the coecients, c1 ,
c2 , c3 and c6 in g. 5. The tree amplitude on the right of g. 6a is given by [10]
Atree (L1 , 1+ , 2+ , L2 ) = i
4

[12] L1 |+ ( + m)| L2
/
,
12 (1 k1 )2 2 m2

(4.1)

where = 1 (1 5 ) is the helicity projection operator and the tree amplitude on the left
2
is given by
im
Atree (L2 , 3+ , H, L1 ) =
4
2v

L2 | /+ (k 4 2m)| L1
3 /
2k4 +
3
L2 | L1
+
(1 + k4 )2 2 m2
s m2
H

. (4.2)

Here v = 2mW is the vacuum expectation value of the scalar Higgs eld [24], m is the mass of
gW
the fermion in the loop and mH is the scalar Higgs mass. Capitalized momenta are (4 2)dimensional, lower case momenta are four-dimensional and is the (2)-dimensional part
9

1= p

= p k1 k2

= p k1

= p + k4

p k1

p + k4

3 p k1 k2 2
(b)

(a)

Figure 6: The s- and t-channels of the gg gH amplitude. The on-shell lines are fermions
since at tree level quarks are the only particles to couple both to gluons and to the scalar
Higgs boson.

of the loop momentum. (See appendix A and appendix B.) In these tree amplitudes we
follow the conventions of ref. [10] for massive spinors, summarized in appendix B. In this
convention the distinction between fermion and anti-fermion spinors is suppressed.
As depicted in g. 6a, for the s-channel the momenta of the on-shell legs are
2 = p k 1 k 2 .

1 = p ,

(4.3)

The second tree amplitude, eq. (4.2), involving the Higgs has been constructed from the tree
level Feynman diagrams including the additional ctitious Higgs-Higgs-gluon coupling (this
is the standard scalar-scalar-vector coupling of scalar QED [24]) mentioned earlier.
So sewing the two tree amplitudes together gives us the s-channel of the amplitude

2m k4 + [12]
d42 P N1
3
+ + +
=
A4 (1 , 2 , 3 , H)
schannel
v(s m2 ) 12
(2)42 D3
H
(4.4)
m [12]
d42 P N2

,
(2)42 D4 2 2 m2 =0
1
2v 12
2 2 m2 =0
2

where the denominators are

D3 = [2 2 m2 ][(1 k1 )2 2 m2 ][(1 k1 k2 )2 2 m2 ]
1
D4 = [2 2 m2 ][(1 k1 )2 2 m2 ][(1 k1 k2 )2 2 m2 ][(1 + k4 )2 2 m2 ] ,
1
(4.5)
and the numerators are
N1 = L1 |+ ( + m)| L2 L2 | L1
/
N2 = L1 |+ ( + m)| L2 L2 |/+ (k 4 2m)| L1 .
3 /
/
10

(4.6)

The two terms in eq. (4.4) correspond to the two terms in eq. (4.2).
First, concentrating on the numerator N1 we can write it in the form of a trace
N1 = tr+ [( + m)(/2 + m)(/1 + m)] ,

/
/
/

(4.7)

where we use the notation tr [. . .] = tr[ . . .]. When expanding out the trace we rst
concentrate on the s since the nal expression for the numerator can only depend on 2
/
since the (2)-dimensional integral over vanishes for a integrand with a odd number of
s. Thus,
N1 = tr+ [(m/2 2 m2 )(/1 m)]

2
2
= 2m(m + ) + 2m 1 2
(4.8)
2
2
= 4m(m + ) ms .
The numerator N2 can also be written in the form of a trace
N2 = tr+ [( + m)(/2 + m)/+ (k 4 2m)(/1 + m)] .

3 /

/
/
/

(4.9)

So again rst concentrating on the s we can write

/
N2 = tr+ [(/2 + m/2 2 m2 )/+ (k 4 2m)(/1 + m)]

3 /

/
/
2
2 + /
= tr+ [(m/2 m )/3 (k 4 2m)(/1 m)] 2 tr+ [/2/+ (k 4 2m)]

3 /
2

= 2m(m + )(21 + 22 + k4 )

+
3

(4.10)

m tr+ [/2/+ k 4/1 ] .

3 /

Recalling that 2 = 1 k1 k2 and choosing the reference momenta q3 = k1 (so that

k1 .+ = 0) then
3
N2 = 8m(m2 + 2 )1 + + 6m(m2 + 2 )k4 + + m tr+ [/2/+ k 4/1 ] .
3 /
3
3
We can evaluate the remaining trace in the expression for N2 by using eq. (2.3)

2
+
+
tr+ [/2/3 k 4/1 ] = tr [/3 k 4/1/2 ] =
/
1 |412 |3 ,
/
13

(4.11)

(4.12)

where for brevity we represent external momenta by their indices and neglect slashing the
/
/
/
momenta inside the inner-product. Using /1/2 = /1 (/1 k 1 k 2 ) = m2 + 2 /1 (k 1 + k 2 )

/
and momentum conservation the trace can be expressed as

2
+
(m2 + 2 ) 1 |4|3 + 1 |41 4|3 .
(4.13)
tr+ [/2/3 k 4/1 ] =
/
13
But since

1 |4|3
k4 + =
,
3
2 13

(4.14)

N2 can be expressed as
N2 = 8m(m2 + 2 )(1 + + k4 + ) +
3
3
11

2m

1 |41 4|3
.
13

(4.15)

The integral comprised of the numerator N2 now only involves at most a single power of
loop momentum in the integrand. This can be readily evaluated with the use of Feynman
parameterization. The Feynman parameter shift is
1 = q + a2 k1 + a3 (k1 + k2 ) a4 k4 ,

(4.16)

where q is the shifted loop momentum. After Feynman parameterizing the integral the terms
in N2 involving 1 can be evaluated by performing the momentum shift and next dropping
terms containing a single power of q which vanish. So the two terms in N2 involving 1
become
1 |q + a2 k1 + a3 (k1 + k2 ) a4 k4 |3

1 .+ =
= (a3 + a4 )k4 .+ ,
(4.17)
3
3
2 13
and
1 |41 4|3 = a2 1 |414|3 + a3 1 |4(1 + 2)4|3 a4 1 |444|3
= [(t m2 )a2 sa3 m2 a4 ] 1 |4|3 ,
H
H

(4.18)

after performing the momentum shift.

As dened in appendix A.3, the integral function In [ai ] is a Feynman parameterized form
of the loop integral with a factor of ai in the numerator. Thus after Feynman parameterization eq. (4.4) becomes
2
2m k4 + [12]
1
(4)
3
+ + +
I [4(m2 + 2 ) s] (t m2 )I4 [a2 ]
A4 (1 , 2 , 3 , H)
=i
H
2 3
2
schannel
v(4)
12 s mH
+ sI4 [a3 ] + m2 I4 [a4 ] 4I4 [(m2 + 2 )(1 a3 a4 )]
H

,
schannel

(4.19)
where
=
and
is dened in appendix A. The dimension label
on (4 2)-dimensional integrals has been suppressed. Then using the recursive formula for
In [ai ] in eq. (A.14) and eq. (4.14) one can write the s-channel of the primitive amplitude as
D=42 2
In
[ ai ]

D=62
In
[ai ]

(j)
I3

st
m2
i
2 v
(4)
12 23 31

tu
(4)
I3 [4(m2 + 2 ) m2 ]
H
2
schannel
2t(s mH )
s m2 (2)
1
H

I3 [4(m2 + 2 ) m2 ] I4 [4(m2 + 2 ) m2 ]
,
H
H
2st
2
schannel
(4.20)
(2)
2
where u = mH s t, also noting that I3 is a scalar triangle integral dependent on the
(4)
kinematic invariant s and m2 , while I3 is a function of s only. Then using eq. (A.11) the
H
s-channel becomes
A4 (1+ , 2+ , 3+ , H)

i
tu
st
m2
(4)
(4m2 m2 )
I3
H
2
2 v
schannel
(4)
12 23 31
2t(s mH )
2
2
s mH (2)D=62 1 D=62
tu
s mH (2) 1
(4)D=62
I3 I4 4
I3
I

I4

2
2st
2
2t(s mH )
2st 3
2

A4 (1+ , 2+ , 3+ , H)

12

.
schannel

(4.21)

Now that we have the s-channel of this amplitude we need to calculate the t-channel
(corresponding to targeting c1 , c4 , c5 and c7 in g. 5) in order to reconstruct the full primitive
amplitude. However due to the symmetry of the all-plus primitive amplitude it is not
necessary to explicitly perform this calculation as the s- and t-channels are related up to an
overall sign by a simple exchange of legs 1 and 3 which is equivalent to exchanging s and t
in eq. (4.21) (and also switching the 1 and 3 in the inner products in the pre-factor of the
expression). Thus the t-channel of A4 (1+ , 2+ , 3+ , H) is given by
su
st
m2
i
(1)
(4m2 m2 )
I3
H
2
2 v
tchannel
(4)
12 23 31
2s(t mH )
2
2
t mH (3) 1
t mH (3)D=62 1 D=62
su
(1)D=62

I3 I4 4
I3
I

I4
2
2st
2
2s(t mH )
2st 3
2

A4 (1+ , 2+ , 3+ , H)

,
tchannel

(4.22)
again recalling that
is a scalar triangle integral dependent on the kinematic invariant t
(1)
2
and mH , while I3 is a function of t only. It is also useful to observe that the coecient of
the I4 integral in the t-channel matches that of the s-channel as expected. If we refer back to
g. 5, where all the scalar integrals that could possibly enter were listed, we notice that the
coecients, c8 of I2 (m2 ) and c9 of I2 (0) have not yet been xed. However, there are no other
H
UV innities in either the s- or t-channel expressions (eqs. (4.21) and (4.22)), so therefore
since this primitive amplitude is UV-nite (since it vanishes at tree level), it is not surprising
that these coecients are zero. This may be explicitly veried by direct construction of the
channel corresponding to using the on-shell condition on the two propagators adjacent to
the massive external leg, showing that both c8 and c9 are individually zero. We have not
evaluated this channel, since this example is only for illustrative purposes.
Thus with the two channels of the primitive amplitude we can construct the full primitive
amplitude A4 (1+ , 2+ , 3+ , H), which is
(3)
I3

su
s m2 (2)
st
m2
i
(1)
H
(4m2 m2 )
I3
I
H
(4)2 v 12 23 31
2s(t m2 )
2st 3
H
t m2 (3)
tu
1
su
(4)
(1)D=62
H

I3 +
I3 I4 4
I
2
2st
2t(s mH )
2
2s(t m2 ) 3
H
2
2
tu
1 D=62
s mH (2)D=62 t mH (3)D=62
(4)D=62
I3
I3
I3
.

+
I4

2
2st
2st
2t(s mH )
2
(4.23)
Note that this expression is valid to all orders in with four-dimensional external momenta.
Since this amplitude has no ultraviolet or quadratic divergences, the primitive amplitude is
well dened in the limit 0. Therefore eq. (4.23) can be simplied, since in this limit with
D=62
D=62
the use of eq. (A.8), I3
1 and I4
0. Also with the intention of simplifying
2
A4 (1+ , 2+ , 3+ , H) =

13

this expression we can use eq. (A.5) explicitly. Thus

i m2
(4)2 v 12
m2
+ H I3 (m2 )
H
st

1
st
1
(4m2 m2 )
I (s) +
I3 (t)
H
2 3
23 31
s mH
t m2
H
1
2u(m4 ts)
H
,
I4 (s, t, m2 )
H
2
2
ts(s mH )(t m2 )
H
(4.24)
where all the integrals in this equation are in four dimensions.
A4 (1+ , 2+ , 3+ , H) =

4.2

The remaining primitive amplitudes

In a similar fashion to the all-plus helicity case the primitive amplitudes corresponding to
the remaining two helicity congurations may be calculated. The results are
i m2 [12]3 t
4um2
4um2 m2
H
I3 (t)
I3 (m2 )
H
(4)2 v [13][23] s (t m2 )2
t(t m2 )2
H
H
1
1
m2
1
H
+ (4m2 s)
I3 (s) +
I3 (t)
I3 (m2 ) I4 (s, t, m2 )
H
H
2
2
2
s mH
t mH
t(t mH )
2
2u(t s)
2u

,
I2 (t) I2 (m2 ) +

H
2 2
(t mH )
t(s m2 )(t m2 )
H
H
(4.25)

A4 (1+ , 2+ , 3 , H) =

and
i m2 [13]3 st
4ts
1
I4 (s, t, m2 )
u 12m2
H
2 v [12][23] u2
(4)
2
u
u 4m2 4s
4um2
4um2
u 4m2 4t
+
+
+ +
+
I3 (s) +
I3 (t)
s m2
u
(s m2 )2
t m2
u
(t m2 )2
H
H
H
H
4
4um2
4um2
(u 4m2 )(st u2 )

I3 (m2 )
+ m2
H
H
st(s m2 )(t m2 ) u s(s m2 )2 t(t m2 )2
H
H
H
H
2(3u + 2t)
2(3u + 2s)
2u(u2 st)
H
H

,
I2 (s) I2 (m2 )
I2 (t) I2 (m2 )
(s m2 )2
(t m2 )2
st(s m2 )(t m2 )
H
H
H
H
(4.26)
2 s are bubble integrals with their 1/ poles removed (since the poles cancel in
where the I
the dierence between the two I2 s) and all integrals are in four dimensions.
Thus in equations (4.24,4.25,4.26) we have the three necessary primitive amplitudes
A4 (1+ , 2+ , 3+ , H), A4 (1+ , 2+ , 3 , H) and A4 (1+ , 2 , 3+ , H) which are needed for the full
gg gH amplitude. We have compared these amplitudes to those of Ellis et al. [16] and
they agree. This comparison is performed by dotting the tensor expressions of Ellis et al.
with spinor helicity polarization vectors.
A4 (1+ , 2 , 3+ , H) =

14

Summary and discussion

In this paper we have explained the relationship of the unitarity method reviewed in ref.[1]
to Feynman diagram calculations. One may calculate Feynman diagrams in a sophisticated
manner so as to take advantage of the benets of the unitarity method. We have illustrated
the usefulness of this method for the case of gg gH. This example is general in that it incorporates both massive internal and external kinematics as well as a combination of colored
and colorless external particles. It should be evident from the explicit calculation presented
in this paper that this method lends itself to making complicated analytical calculations
more tractable. This is because in this approach one is always dealing with gauge invariant
constructions, which are generally signicantly simpler than the gauge dependent Feynman
graphs in a conventional calculation. In the unitarity method one targets the coecients of
a subset of scalar integrals contributing to the amplitude by computing a gauge invariant
combination of Feynman diagrams.
The notion of a channel as dened in this paper is more general than just the cut legs of a
Cutkosky calculation, where one is nominally computing the discontinuity of an amplitude in
a certain cut channel. The reason it is more general is that one denes a channel as a subset
of the terms in the full amplitude; those terms being the scalar integrals that include some
pre-dened combination of internal propagators. This notion of a channel has already been
useful in the evaluation of Z 4 partons [13]. It was also helpful in two-loop calculations
in N = 4 supersymmetric Yang-Mills [15].
One of the themes of this paper has been to explain the usefulness of the unitarity method
to multi-point one-loop calculations. The ideas discussed here will potentially be benecial
to the extension of these types of calculations to multi-loops. Initial steps have already been
made [15], and more attention is being devoted to these pursuits.

Acknowledgments
I thank Z. Bern for encouragement and many valuable discussions. I also thank L. Dixon,
A.K. Grant, D.A. Kosower, E. Marcus, A.G. Morgan and B. Yan for helpful discussions and
comments. This work was supported by the DOE under contract DE-FG03-91ER40662.

Integrals

In this appendix we collect expressions for the integrals [21, 25, 26] we use in this paper.
The D-dimensional loop integrals considered in this paper are dened by
P 1 P m
dD P
, (A.1)
n1
(2)D (P 2 m2 ) ((P i=1 ki )2 m2 )
where P is a D-dimensional momentum and the external momenta, ki , are four-dimensional.
Our convention is to suppress the dimension label on In when we are dealing with a D = 42
integral, unless otherwise indicated. We also suppress the square brackets when the argument
is unity. Such integrals are referred to in the text as scalar integrals, while tensor integrals
have m 1.
D
In [P 1 P m ] = i(1)n+1 (4)D/2

15

Given a four-point amplitude, we have two conventions for labeling the one-, two- and
three-point integrals. The rst, and more familiar one is to explicitly give the kinematic
invariant upon which the integral depends. For example, I2 (s) is the (4 2)-dimensional
bubble that has an invariant mass square of s = (k1 + k2 )2 owing through its external legs.
The second convention for labeling the lower point integrals is to indicate, by a raised (i)
that the internal propagator between legs i 1 (mod 4) and i has been removed. If two
indices are present as in (i, j) then two propagators have been removed. This follows the
labeling convention of refs. [26, 10].

A.1

The D = 4 2 scalar integrals

The bubble with an external kinematic invariant s is

(2,4)

I2 (s) = I2
where x
form as

= I2 (0) + 2 + x log

x1
+ O() ,
x+1

(A.2)

1 4m2 /s. For external massless kinematics, this may be written in closed

(1 + )
.
(A.3)

The scalar triangle integral can either be a function of just one kinematic invariant s
x+1
1
(4)
+ O() ,
(A.4)
I3 (s) = I3 = log2
2s
x1
I2 (0) = m2

or of two invariants s and m2

H

(2)

I3 (s, m2 ) = I3 =
H

sI3 (s) m2 I3 (m2 )

H
H
+ O() .
s m2
H

(A.5)

While the box with a uniform internal mass and a single massive external leg, with invariant
mass-square m2 is
H
I4 (s, t, m2 ) =
H

1
um2 m2
um2 m2
um2 m2
H
+H
H
,
,
,
st
st
s
st
t
st m2
H

+ O() , (A.6)

where u = m2 s t,
H

X
x
x
x
2
ln 1
ln
Li2
Li2
x+ x
Y
x+
y x+
x y
(A.7)
x+
x+
+ Li2
+ Li2
,
x+ y
y x

with x = 1 (1 1 4X), y = 1 (1 + 1 4Y ) and the dilogarithm [27] is dened by

2
2
Li2 (x) 01 dt ln(1 xt)/t.
The scalar integral expressions quoted above are only valid in certain analytic regions.
These can be analytically continued, but are not presented here since related expressions2
are presented for all analytic regions in ref. [16].
H(X, Y )

The relationship between the scalar integrals in this paper and those in ref. [16]: I2 (s) = I2 (0)+2W1 (s),
I3 (s) = W2 (s)/2s and I4 (s, t, m2 ) = W3 (s, u, t, m2 )/st.
H
H

16

A.2

Higher dimension integrals

The higher dimension integrals are dened in eq. (A.1) with D set to the appropriate value.
The relationship of these integrals to the usual four-dimensional ones has been extensively
discussed in ref. [26]. For n 6, (6 2)-dimensional integrals can be written in terms of
(4 2)-dimensional integrals via the integral recursion relation
D=62
In
=

where ci =

4
j=1

1
Sij , c0 =

n
1
(i),D=42
D=42
,
ci In1
2In

(n 5 + 2)c0
i=1
n
i=1 ci

(A.8)

, and the matrix Sij is

1
Sij m2 p2 , with pii 0 , and pij = pji ki + ki+1 + + kj1
2 ij

for i < j . (A.9)

Higher dimension integrals arise naturally when performing the calculations described in
this paper. In a typical primitive amplitude discussed in the text we obtain integrals of the
form
D=42

In
[f (p , ki , 2 )] =

d4 p d2
f (p , ki , 2)
,
n1
(2)4 (2)2 (p2 2 m2 ) ((p i=1 ki )2 2 m2 )
(A.10)
where we have explicitly broken the (4 2)-dimensional momentum into a four-dimensional
part, p, and a (2)-dimensional part, .
As noted by Mahlon [6], using the denition (A.1), we can write

i(1)n+1 (4)2

D=42 2
D=62
In
[ ] = In
.

(A.11)

Note that although the loop momentum is shifted to higher dimension, the external momenta
remain in four dimensions.

A.3

Feynman parameter reduction

As discussed in the text, tensor integrals may be evaluated using Feynman parameters. We
make use of the following formulae in the text.
After Feynman parameterization and the normal shift of momentum we make use of the
following,
d4 q d2
q 1 q 2r+1
= 0,
(A.12)
(2)4 (2)2 (q 2 2 Sij ai aj )n
where Sij is dened in eq. (A.9).
After integrating out the loop momenta we obtain integrals of the form
D
In [f (ak )] (n D/2)

1
0

dn ak (1
17

r ar )

f (ak )
n
i,j=1 Sij ai aj

nD/2

(A.13)

An integral reduction formula that is quite useful is [26]

D
In [ai ]

1 n
ci D
(j), D
=
cij In1 +
I ,
2 j=1
c0 n

(A.14)

1
i
where cij = Sij cccj , and ci and c0 are dened in appendix A.2. Another useful integral
0
D
reduction formula for In [ai aj ] can also be found in [26].

Rules for s

To begin, we have a (4 2)-dimensional vector, Q , which we can express as a sum of a

four dimensional piece q and a (2)-dimensional ; Q = q + . (We have used the
convention that upper-case momenta are (4 2)-dimensional and lower-case momenta are
four-dimensional.) Here we summarize the rules for dealing with s given in ref. [10].
We abide by the conventional Dirac algebra, , = 2 and 0 = 0 ,
where and are (4 2)-dimensional Lorentz indices. The metric is given by =
diag(+, , , , . . .). It follows that Q Q = q 2 2 (the negative sign preceding the 2 is
from the metric).
Thus anti-commutes with four-dimensional matrices, /, = 0. So if we use the
q /
/
conventions of t Hooft and Veltman [28], and adopt the arbitrary dimension denition,
5 = i 0 1 2 3 , then /, 5 = 0 and , 5 = 0. That is freely commutes with 5 , since
q
/
/
q q
5 is constructed as a product of four-dimensional Dirac matrices. Hence, / = / and
= , where 1 (1 5 ) is the helicity projection operator. In calculations it is
/ /
2
important to rst commute the s together before evaluating the remaining four-dimensional
/
pieces.
In the calculation presented in section 4 the momentum of the sewn fermion lines is
(4 2)-dimensional. We will borrow the bra and ket symbols to represent the spinors, but
as helicity is not a good quantum number we shall not label them with a helicity. Since we
always sum over all states across the cut, there is no need to dene a (4 2) helicity notion.
In sewing the (42)-spinors together we implicitly sum over the spin degrees of freedom,
that is
|Q Q| = Q + m = / + + m ,
q /
/

and | Q Q| = Q + m = / + m .
q /
/

(B.1)

This notation glosses over the distinction of spinors and anti-spinors. However, as discussed
in ref. [10], this may introduce an overall sign which can be readily xed by hand. Alternatively, one may easily use both particle and anti-particle spinors to obtain identical
results.

References
[1] Z. Bern, L. Dixon and D.A. Kosower, Ann. Rev. Nucl. Part. Sci. 46:109 (1996), hepph/9602280.
18

[2] F.A. Berends, R. Kleiss, P. De Causmaecker, R. Gastmans and T. T. Wu, Phys. Lett.
B103:124 (1981);
P. De Causmaecker, R. Gastmans, W. Troost and T.T. Wu, Nucl. Phys. B206:53 (1982);
R. Kleiss and W.J. Stirling, Nucl. Phys. B262:235 (1985);
Z. Xu, D.-H. Zhang and L. Chang, Nucl. Phys. B291:392 (1987).
[3] J.E. Paton and Chan Hong-Mo, Nucl. Phys. B10:519 (1969);
F.A. Berends and W.T. Giele, Nucl. Phys. B294:700 (1987);
D.A. Kosower, B.-H. Lee and V.P. Nair, Phys. Lett. 201B:85 (1988);
M. Mangano, S. Parke and Z. Xu, Nucl. Phys. B298:653 (1988);
Z. Bern and D.A. Kosower, Nucl. Phys. B362:389 (1991).
[4] Z. Bern and D.A. Kosower, Nucl. Phys. B379:451 (1992).
[5] F.A. Berends and W.T. Giele, Nucl. Phys. B306:759 (1988);
D.A. Kosower, Nucl. Phys. B335:23 (1990);
G. Mahlon and T.-M. Yan, Phys. Rev. D47:1776 (1993), hep-ph/9210213;
G. Mahlon, Phys. Rev. D47:1812 (1993), hep-ph/9210214;
G. Mahlon, T.-M. Yan and C. Dunn, Phys. Rev. D48:1337 (1993), hep-ph/9210212;
C. Kim and V.P. Nair, Phys. Rev. D55:3851 (1997), hep-th/9608156.
[6] G. Mahlon, Phys. Rev. D49:2197 (1994), hep-ph/9311213; Phys. Rev. D49:4438 (1994),
hep-ph/9312276.
[7] L.D. Landau, Nucl. Phys. 13:181 (1959);
S. Mandelstam, Phys. Rev. 112:1344 (1958), 115:1741 (1959);
R.E. Cutkosky, J. Math. Phys. 1:429 (1960).
[8] Z. Bern, L. Dixon, D.C. Dunbar and D.A. Kosower, Nucl. Phys. B425:217 (1994), hepph/9403226.
[9] Z. Bern, L. Dixon, D.C. Dunbar and D.A. Kosower, Nucl. Phys. B435:59 (1995), hepph/9409265.
[10] Z. Bern and A.G. Morgan, Nucl. Phys. B467:479 (1996), hep-ph/9511336.
[11] S.J. Parke and T.R. Taylor, Phys. Rev. Lett. 56:2459 (1986);
Z. Bern, L. Dixon and D.A. Kosower, Proceedings of Strings 1993, eds. M.B. Halpern,
A. Sevrin and G. Rivlis (World Scientic, Singapore, 1994), hep-th/9311026;
Z. Bern, G. Chalmers, L. Dixon and D.A. Kosower, Phys. Rev. Lett. 72:2134 (1994),
hep-ph/9312333;
Z. Bern and G. Chalmers, Nucl. Phys. B447:465 (1995), hep-ph/9503236.
[12] Z. Bern, L. Dixon and D.A. Kosower, Phys. Rev. Lett. 70:2677 (1993), hep-ph/9302280;
Z. Kunszt, A. Signer and Z. Trcsnyi, Phys. Lett. B336:529 (1994), hep-ph/9405386;
o a
Z. Bern, L. Dixon and D.A. Kosower, Nucl. Phys. B437:259 (1995), hep-ph/9409393.

19

[13] Z. Bern, L. Dixon and D.A. Kosower, Nucl. Phys. Proc. Suppl. 51C:243 (1996), hepph/9606378;
Z. Bern, L. Dixon, D.A. Kosower and S. Weinzierl, Nucl. Phys. B489:3 (1997), hepph/9610370;
Z. Bern, L. Dixon and D.A. Kosower, preprint hep-ph/9708239.
[14] W.L. van Neerven, Nucl. Phys. B268:453 (1986).
[15] Z. Bern, J.S. Rozowsky and B. Yan , Phys. Lett. B401:273 (1997), hep-ph/9702424.
[16] R.K. Ellis, I. Hinchlie, M. Soldate and J.J. Van Der Bij, Nucl. Phys. B297:221 (1988).
[17] M. Mangano and S.J. Parke, Phys. Rep. 200:301 (1991);
L. Dixon, Proceedings of Theoretical Advanced Study Institute in Elementary Particle
Physics (TASI 95), ed. D.E. Soper, hep-ph/9601359.
[18] W. Siegel, Phys. Lett. B84:193 (1979);
D.M. Capper, D.R.T. Jones and P. van Nieuwenhuizen, Nucl. Phys. B167:479 (1980);
S.J. Gates, M.T. Grisaru, M. Rocek and W. Siegel, Superspace, (Benjamin/Cummings,
1983).
[19] Z. Kunszt, A. Signer and Z. Trcsnyi, Nucl. Phys. B411:397 (1994), hep-ph/9305239;
o a
S. Catani, M.H. Seymour and Z. Trcsnyi, Phys. Rev. D55:6819 (1997), hepo a
ph/9610553.
[20] Z. Bern, L. Dixon and D.A. Kosower, Nucl. Phys. B437:259 (1995), hep-ph/9409393.
[21] L.M. Brown and R.P. Feynman, Phys. Rev. 85:231 (1952);
L.M. Brown, Nuovo Cimento 21:3878 (1961);
G. Passarino and M. Veltman, Nucl. Phys. B160:151 (1979).
[22] R. Pittau, preprint hep-ph/9607309.
[23] G. t Hooft, in Acta Universitatis Wratislavensis no. 38, 12th Winter School of Theoretical Physics in Karpacz; Functional and Probabilistic Methods in Quantum Field
Theory, Vol. 1 (1975);
B.S. DeWitt, in Quantum gravity II, eds. C. Isham, R. Penrose and D. Sciama (Oxford,
1981);
L.F. Abbott, Nucl. Phys. B185:189 (1981).
[24] M. Peskin and D.V. Schroeder, An Introduction to Quantum Field Theory (AddisonWesley, 1995).
[25] G. t Hooft and M. Veltman, Nucl. Phys. B153:365 (1979);
A. Denner, U. Nierste and R. Scharf, Nucl. Phys. B367:637 (1991);
W. van Neerven and J.A.M. Vermaseren, Phys. Lett. B137:241 (1984);
G.J. van Oldenborgh and J.A.M. Vermaseren, Zeit. Phys. C46:425 (1990).

20

[26] Z. Bern, L. Dixon and D.A. Kosower, Phys. Lett. B302:299 (1993), hep-ph/9212308;
erratum ibid. B318:649 (1993);
Z. Bern, L. Dixon and D.A. Kosower, Nucl. Phys. B412:751 (1994), hep-ph/9306240.
[27] L. Lewin, Polylogarithms and Associated Functions (North Holland, 1981).
[28] G. t Hooft and M. Veltman, Nucl. Phys. B44:189 (1972).

21