ANL-HEP-PR-09-109, NU-HEP-TH/09-14

Characteristics and Estimates of Double Parton Scattering at the

Large Hadron Collider

Edmond L. Berger∗ and C. B. Jackson†

High Energy Physics Division, Argonne National Laboratory, Argonne, IL 60439
arXiv:0911.5348v1 [hep-ph] 27 Nov 2009

Gabe Shaughnessy‡
High Energy Physics Division, Argonne National Laboratory, Argonne, IL 60439 and
Department of Physics & Astronomy,
Northwestern University, Evanston, IL 60208
(Dated: November 27, 2009)

Abstract
We evaluate the kinematic distributions in phase space of 4-parton final-state subprocesses pro-
duced by double parton scattering, and we contrast these with the final-state distributions that
originate from conventional single parton scattering. Our goal is to establish the distinct topolo-
gies of events that arise from these two sources and to provide a methodology for experimental
determination of the relative magnitude of the double parton and single parton contributions at
Large Hadron Collider energies. We examine two cases in detail, the b b̄ jet jet and the 4 jet final
states. After full parton-level simulations, we identify a few variables that separate the two con-
tributions remarkably well, and we suggest their use experimentally for an empirical measurement
of the relative cross section. We show that the double parton contribution falls off significantly
more rapidly with the transverse momentum pj1
T of the leading jet, but, up to issues of the relative

normalization, may be dominant at modest values of pj1

T .

∗
Electronic address: [email protected]
†
Electronic address: [email protected]
‡
Electronic address: [email protected]

1
I. INTRODUCTION

Double parton scattering (DPS) means that two short-distance subprocesses occur in a
given hadronic interaction, with two initial partons being active from each of the incident
protons in a collision at the Large Hadron Collider (LHC). The concept is shown for illustra-
tive purposes in Fig. 1, and it may be contrasted with conventional single parton scattering
(SPS) in which one short-distance subprocess occurs, with one parton active from each initial
hadron. Since the probability of single parton scattering is itself small, it is often expected
that the chances are considerably suppressed for two or more short-distance interactions in a
given collision. However, expectations such as these bear quantitative re-examination at the
LHC where the high overall center-of-mass energy provides access to very small values of the
fractional momentum x carried by partons, a region in which parton densities grow rapidly.
A large contribution from double parton scattering could result in a larger than otherwise
anticipated rate for multi-jet production and produce significant backgrounds in searches for
signals of new phenomena. The high energy of the LHC also provides an increased dynamic
range of available phase space for detailed investigations of DPS.
Investigations of double parton scattering have a long history theoretically [1, 2, 3, 4,
5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22], and there is evidence for
their presence in collider data from the CERN Intersecting Storage Rings [23] and Fermilab
Tevatron [24, 25]. A significantly greater role for double-parton processes may be expected
at the LHC where higher luminosities are anticipated along with the higher collision energies.
Of substantial importance is to know empirically how large the double parton contribution
may be and its dependence on relevant kinematic variables.

p
i
a
k
b
j
c
l
d
p

FIG. 1: Sketch of a double-parton process in which the active partons are i and k from one proton
and j and l from the second proton. The two hard scattering subprocess are A(i j → a b) and
B(k l → c d).

2
 Our aim is to calculate characteristic final states at LHC energies in which it may be
straightforward to discern a double parton signal. We show in this paper that double
parton scattering produces an enhancement of events in regions of phase space in which the
“background” from single parton scattering is relatively small. If such enhancements are
observed experimentally, with the kinematic dependence we predict, then we will have a
direct empirical means to measure the size of the double parton contribution. In addition
to its role in general LHC phenomenology, this measurement will have an impact on the
development of partonic models of hadrons, since the effective cross section for double parton
scattering measures the size in impact parameter space of the incident hadron’s partonic hard
core.
From the perspective of sensible rates and experimental tagging, a good process to exam-
ine should be the 4 parton final state in which there are 2 hadronic jets plus a b quark and
a b̄ antiquark, viz. b b̄ j1 j2 . If the final state arises from double parton scattering, then it is
plausible that one subprocess produces the b b̄ system and another subprocess produces the
two jets. There are, of course, many single parton scattering (2 to 4 parton) subprocesses
that can result in the b b̄ j1 j2 final state, and we look for kinematic distributions that
show notable separations of the two contributions. As we show, the correlations in the final
state are predicted to be quite different between the double parton and the single parton
subprocesses. For example, the plane in which the b b̄ pair resides is uncorrelated with the
j1 j2 plane in double parton scattering, but not in the single parton case.
The state-of-the-art of calculations of single parton scattering is well developed whereas
the phenomenology of double parton scattering is as yet much less advanced. In the remain-
der of this Introduction, we first describe the approach we adopt for the calculation of double
parton scattering, specializing to the proton-proton situation of the LHC. Then we outline
the paper and summarize our main results. Our calculations are done at leading-order in
perturbative QCD, adequate for the points we are trying to make.
Making the usual factorization assumption, we express the single-parton hard-scattering
differential cross section for p p → a b X as
XZ
SP S
dσ = fpi (x1 , µ)fpj (x′1 , µ)dσ̂(ij→ab) (x1 , x′1 , µ)dx1 dx′1 . (1)
i,j

Indices i and j run over the different parton species in each of the incident protons. The
parton-level subprocess cross sections dσ̂(ij→ab) (x1 , x′1 , µ) are functions of the fractional par-

3
tonic longitudinal momenta x1 and x′1 from each of the incident hadrons and of the partonic
factorization/renormalization scale µ. The parton distribution functions fpi (x1 , µ) express
the probability that parton i is found with fractional longitudinal momentum x1 at scale
µ in the proton; they are integrated over the intrinsic transverse momentum (equivalently,
impact parameter) carried by the parton in the parent hadron.
A formal theoretical treatment of double parton scattering would begin with a discussion
of the hadronic matrix element of four field operators and an explicit operator definition
of two-parton correlation functions. This procedure would lead to a decomposition of the
hadronic matrix element into non-perturbative two-parton distribution functions and the
corresponding hard partonic cross sections for σ̂(ijkl → abcd). An operator definition of
two-parton correlation functions may be found in Ref. [26] where the two-parton correlation
function is reduced to a product of single parton distributions. An explicit operator definition
of two-parton distributions with different values of the two fractional momenta x1 and x2
is presented in Ref. [27], along with a model for the two-parton distributions in terms of
normal parton distributions. In this paper, we follow a phenomenological approach along
lines similar to Refs. [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22].
In a double parton process, partons i and k are both active in a given incident proton. We
require the joint probability that parton k carries fractional momentum x2 , given that parton
i carries fractional momentum x1 . In general, this joint probability H i,k (x1 , x2 , µA , µB )
should also depend on the intrinsic transverse momenta kT,i and kT,k of the two partons
(or, equivalently, their impact parameters). The hard scales µA and µB are characteristic of
the two hard subprocesses in which partons i and k participate. In the sections below, we
discuss the choice we make of the hard-scale and do not explore in this paper theoretical
uncertainties associated with higher-order perturbative contributions.
In contrast to single parton distributions functions fpi (x1 , µ) for which global analyses have
produced detailed information, very little is known phenomenologically about the magnitude
and functional dependences of joint probabilities H i,k (x1 , x2 , µA , µB ). A common assump-
tion made in estimates of double parton rates is to ignore possibly strong correlations in
longitudinal momentum and to use the approximation

Hpi,k (x1 , x2 , µA , µB ) = fpi (x1 , µA )fpk (x2 , µB ). (2)

For reasons of energy-momentum conservation, if not dynamics, the simple factorized form of

4
Eq. (2) cannot be true for all values of the fractional momenta x. The values of x2 available
to the second interaction are always limited by the values of x1 in the initial interaction since
x1 + x2 ≤ 1. The approximation certainly fails even at the kinematic level if both partons
carry a substantial fraction of the momentum of the parent hadron. However, it may be
adequate for applications in which the values of x1 and x2 are small. We remark that the
momentum integral
XZ
x1 x2 Hpi,k (x1 , x2 , µA , µB )dx1 dx2 = 1, (3)
i,k

as long as we can run the upper limits of the x1 and x2 integrations to 1, independently.
The large phase space at the LHC may make it possible to explore dynamic correlations
that break Eq. (2).
In Fig. 2, for the region of phase space of interest to us, we show the contributions to
the bb̄jj cross section as a function of x from both DPS and SPS, after minimal acceptance
√
cuts are imposed (Sec. II). The center-of-mass energy is s = 10 TeV. It is evident that
the majority of DPS events are associated with low x values, in essence never exceeding
0.2. The momentum carried off by the beam remnant is (1 − x1 − x2 ) in DPS and (1 − x)
in SPS. The results in Fig. 2 show that this remnant momentum is not too different in
DPS and SPS. Thus, the use of Eq. (2) in calculations of event rates at the LHC appears
adequate as a good first approximation. While available Tevatron data on double parton
scattering [24, 25] are insensitive to possible correlations in x, the greater dynamic range at
1
the LHC may make it possible to observe them.
Assuming next that the two subprocesses A(i j → a b) and B(k l → c d) are dynamically
uncorrelated, we express the double parton scattering differential cross section as:

m X
Z
DP S
dσ = Hpik (x1 , x2 , µA , µB )Hpjl (x′1 , x′2 , µA , µB ) (4)
2σeff i,j,k,l
×dσ̂ijA (x1 , x′1 , µA )dσ̂kl
B
(x2 , x′2 , µB )dx1 dx2 dx′1 dx′2 .

The symmetry factor m is 1 if the two hard-scattering subprocesses are identical and is 2
otherwise. In the denominator, there is a factor σeff with the dimensions of a cross section.
Given that one hard-scatter has taken place, σeff measures the size of the partonic core in

1
As emphasized in Refs. [16, 22], even if the approximation in Eq. (2) holds at one hard scale, evolution
of the parton densities with µ will induce violations at larger scales.

5
 DPS
100000
SPS

10000
x dσ/dx

1000

100

10
0.01 0.1
x-parton

FIG. 2: Values of the parton longitudinal momentum fractions x in the DPS and SPS events. Most
DPS events have low x values. The events used for this plot include the requirements njet = 4,
nbtag = 2, and the threshold cuts discussed in Sec. II.

which the flux of accompanying short-distance partons is confined. It should be at most

proportional to the transverse size of a proton. For the first process of interest in this paper,
pp → bb̄j1 j2 , Eq. (4) reduces to

dσ SP S (pp → bb̄X)dσ SP S (pp → j1 j2 X)

dσ DP S (pp → bb̄j1 j2 X) = . (5)
σeff

Tevatron collider data [24, 25] yield values in the range σeff ∼ 12 mb. We use this value
for the estimates we make, but we emphasize that the goal should be to make an empirical
determination of its value at LHC energies.
In Sec. II, we present our calculation of the double parton and the single parton contri-
butions to p p → b b̄ j1 j2 X. We identify variables that discriminate the two contributions
quite well. In Sec. III, we treat the double parton and the single parton contributions to
4 jet production, again finding that good separation is possible despite the combinatorial
uncertainty in the pairing of jets. We show in both cases that the double parton contribution
falls off significantly more rapidly with pj1
T , the transverse momentum of the leading jet. For

the value of σeff ∼ 12 mb and the cuts that we use, we find that, in the region in which it

6
is most identifiable, double parton scattering is dominant for pj1
T < 30 GeV in b b̄ j1 j2 at

LHC energies, and pj1

T < 50 GeV in 4 jet production. Our conclusions are found in Sec. IV.

II. HEAVY QUARK PAIR AND JET PAIR PRODUCTION IN QCD.

In this section, we describe the calculation of the DPS and SPS event rates for bb̄jj
production at the LHC. For our purposes, light jets (denoted by j) are assumed to originate
only from gluons or one of the four lighter quarks (u, d, s or c) and, as stated above, we
√
perform all calculations for the LHC with a center-of-mass energy of s = 10 TeV. Event
rates are quoted for 10 pb−1 of data.

A. Outline of the method

The prediction for the DPS event rate is based on the assumption that the two partonic
interactions which produce the bb̄ and jj systems occur independently (as expressed in
Eq. (4)). At leading order, the only contribution is:

(ij → bb̄) ⊗ (kl → jj) (6)

where the symbol ⊗ denotes the combination of one event each from the bb̄ and the jj
final states. In an attempt to model some of the effects expected from initial- and final-state
radiation, we also account for the possibility of an additional jet which is undetected because
it is either too soft or outside of the accepted rapidity range. Thus, we include several other
contributions to the DPS event:

bb̄(j) ⊗ jj , bb̄j ⊗ (j)j , bb̄j ⊗ j(j) (7)

bb̄ ⊗ (j)jj , bb̄ ⊗ j(j)j , bb̄ ⊗ jj(j) , (8)

where the parentheses surrounding a jet indicate that it is undetected. We compute processes
such as jj(j) and bb̄(j) at LO as 3 parton final-state processes.
The 2 to 3 parton amplitudes for bb̄(j) [and jj(j)] diverge as the undetected jet (j)
becomes soft or collinear to one of the other final state partons or to an initial parton.
The divergences are removed in a full next-to-leading order (NLO) treatment, in which
real emission and virtual (loop) contributions are incorporated, and the finite bb̄, bb̄(j),

7
and bb̄j contributions are present with proper relative normalization. In the LO parton
level simulations done in this paper, we employ a cut at the generator level to remove the
divergences. All the final state objects in the processes listed above are required to have
transverse momentum pT ≥ 20 GeV. In this fashion, we model some aspects of the expected
momentum imbalance between the b and b̄ arising from the 2 to 3 process ij → bb̄j, but
we cannot claim to include the relative normalization between the bb̄ and bb̄j contributions
that would result from a full NLO treatment. We leave a complete NLO analysis for future
work.
The SPS cross section is computed according to Eq. (1). It receives contributions at
lowest order from the 2 parton to 4 jet final state process:

ij → bb̄jj , (9)

and, in the case where a jet is undetected, from the 5-jet final states (computed at LO):

bb̄(j)jj , bb̄j(j)j , bb̄jj(j) . (10)

We also investigate the possibility of jjjj and jjjj(j) final state contributions to the SPS
cross section where two of the jets “fake” b jets. We find that the effects from these final
states are subdominant compared to the processes listed in Eqs. (9) and (10).
In our numerical analysis, we use the leading-order CTEQ6L1 parton distribution func-
tions (PDFs) [28] to compute both DPS and SPS cross sections, and we evaluate all cross
sections using one-loop evolution of αs (µ). For the renormalization and factorization scales,
we choose the dynamic scale:
X
µ2 = p2T,i + m2i , (11)
i

where pT,i is the transverse momentum of the ith jet and mi = 0 (mi = 4.7 GeV) for light
(bottom) jets. In the case of roughly equal values of the transverse momenta pT,i , Eq. (11)
√
yields µ ∼ 2pT in SPS and µ ∼ 2pT in DPS. At LO there is no obviously “right” hard
scale, and the choice in Eq. (11) seems as good as any other.
The DPS events are generated as two separate sets of events with Mad-
graph/Madevent [29] and then combined as described above. For example, at leading order,
we generate events separately for pp → bb̄X and pp → jjX, and these events are then
combined as indicated in Eq. (6). To increase the speed of the simulations, the SPS events

8
are generated with Alpgen [30] since the SPS processes of interest are hard-coded in Alpgen,
which contains more compact expressions for the squared-matrix-elements than Madgraph.
The events accepted after generation are required to have 4 jets njet = 4 with 2 of these
tagged as b’s nbtag = 2. At the generator level, all the final state objects in the processes listed
in Eq. (6) through Eq. (10) must have transverse momentum pT ≥ 20 GeV, as mentioned
above. Furthermore, at the analysis level, all events (DPS and SPS) are required to pass
the following acceptance cuts:

pT,j ≥ 25 GeV, |ηj | ≤ 2.5 (12)

pT,b ≥ 25 GeV, |ηb | ≤ 2.5 (13)

∆Rjj ≥ 0.4, ∆Rbb ≥ 0.4 (14)

where ηi is the jet’s pseudorapidity, and ∆Rij is the separation in the azimuthal angle (φ) -
pseudorapidity plane between jets i and j:
q
∆Rij = (ηi − ηj )2 + (φi − φj )2 . (15)

We model detector resolution effects by smearing the final state energy according to:

δE a
=p ⊕ b, (16)
E E/GeV

where we take a = 50% and b = 3% for jets. To account for b jet tagging efficiencies, we
assume a b-tagging rate of 60% for b-quarks with pT > 20 GeV and |ηb | < 2.0. We also
apply a mistagging rate for charm-quarks as:

ǫc→b = 10% for pT (c) > 50 GeV (17)

while the mistagging rate for a light quark is:

ǫu,d,s,g→b = 0.67% for pT (j) < 100 GeV (18)

ǫu,d,s,g→b = 2% for pT (j) > 250 GeV. (19)

Over the range 100 GeV < pT (j) < 250 GeV, we linearly interpolate the fake rates given
above [31].

9
 B. Properties of SPS and DPS in b b̄ j j

Having detailed the calculation of the bb̄jj event rates from DPS and SPS, we now
discuss some of the distinguishing characteristics of the two contributions. First, however,
it is important to check that our simulations of DPS events are not introducing an artificial
correlation between the bb̄ and jj final states. We do this by inspecting the angle Φ between
the plane defined by the bb̄ system and the plane defined by the jj system. If the two
scattering processes ij → bb̄ and kl → jj which produce the DPS final state are truly
independent, one would expect to see a flat distribution in the angle Φ. By contrast, many
diagrams, including some with non-trivial spin correlations, contribute to the 2 parton to 4
parton final state in SPS, and naively one would expect some correlation between the two
planes. To avoid possible effects from boosting to the lab frame, we define the two planes
in the partonic center-of-mass frame.
We specify the planes by using the three-momenta of the outgoing jets. Then, the angle
between the two planes defined by the jj and bb̄ systems is:

cos Φ = n̂3 (j1 , j2 ) · n̂3 (b1 , b2 ), (20)

where n̂3 (x, y) is the unit three-vector normal to the plane defined by the x − y system.
The normal is undefined when j1 and j2 are back-to-back or b1 and b2 are back-to-back,
as occurs in a large fraction of the DPS events. Therefore, when cos φ(x,y) < −0.9, we use a
different procedure. We use the three-momentum of one of the incoming partons along with
the three-momentum of one of the outgoing b quarks to define the bb̄ plane. Let qb be the
three-momentum of an incoming parton, and pb be the three-momentum of the final-state
b (or b̄) quark. We then define φpb ,qb to be the azimuthal angle of the three-vector normal
to the qb − pb plane. Note that we use φ here since the normal to any three-vector and the
beam-line will be transverse to the beam-line (not the case in the SPS process). In this way,
the jet which is not used to define the plane is guaranteed to lie in the plane. The plane for
the jj system is defined in an analogous manner. Finally, the angle between the planes is
then:
Φ = |φpj ,qj − φpb,qb | . (21)

In Fig. 3, we display the number of events as a function of the angle between the two
planes. There is an evident correlation between the two planes in SPS, while the distribution

10
 30000 SPS
DPS
DPS + SPS

25000

N_events / Bin
20000

15000

10000

5000

0 0.5 1 1.5 2 2.5 3

FIG. 3: Event rate as a function of the angle between the two planes defined by the bb̄ and jj
systems. In SPS events, there is a correlation among the planes which is absent for DPS events.

is flat in DPS, indicative that the two planes are uncorrelated.

Another interesting difference between DPS and SPS is the behavior of event rates as a
function of transverse momentum. As an example of this, in Fig. 4, we show the transverse
momentum distribution for the leading jet (either a b or light j) for both DPS and SPS.
Several characteristics are evident. First, SPS produces a relatively hard spectrum, and for
the value of σeff and the cuts that we use, we see that SPS tends to dominate over the full
range of transverse momentum considered. On the other hand, DPS produces a much softer
spectrum which (up to issues of normalization in the form of σeff ) can dominate at small
values of transverse momentum. The cross-over between the two contributions to the total
event rate is ∼ 30 GeV for the acceptance cuts considered here. A smaller (larger) value of
σeff would move the cross-over to a larger (smaller) value of the transverse momentum pj1
T

of the leading jet.

C. Distinguishing variables

We turn next to the search for variables that may allow for a clear separation of the
DPS and SPS contributions. Since the topology of the DPS events includes two 2 → 2
hard scattering events, the two pairs of jet objects are roughly back-to-back. We expect

11
 SPS
10000 DPS
SPS+DPS

1000

N_events / Bin
100

10

1
0 20 40 60 80 100 120 140
Leading jet PT (GeV)

FIG. 4: The transverse momentum pT distribution of the leading jet in jjbb̄ after minimal cuts.

the azimuthal angle between the pairs of jets corresponding to each hard scattering event
to be strongly peaked near ∆φjj ∼ ∆φbb ∼ π. Real radiation of an additional jet, where the
extra jet is missed because it fails the threshold or acceptance cuts, allows smaller values of
∆φjj . The relevant distribution is shown for light jets (non b-tagged) in Fig. 5a. There is a
clear peak near ∆φjj = π for DPS events, while the events are more broadly distributed in
SPS events. The secondary peak near small ∆φjj arises from gluon splitting which typically
produces nearly collinear jets. The suppression at still lower ∆φjj comes from the isolation
cut ∆Rjj > 0.4.
The separation of DPS events from SPS events becomes more pronounced if information
is used from both the bb̄ and jj systems. As an example, we consider the distribution built
from a combination of the azimuthal angle separations of both jj and bb̄ pairs, using a
variable adopted from Ref. [25]:
1 p
Sφ = √ ∆φ(b1 , b2 )2 + ∆φ(j1 , j2 )2 . (22)
2
In Fig. 5b, we present a distribution in Sφ for both DPS and SPS+DPS events. Again,
as in the case of the ∆φ distribution, we see that the SPS events are broadly distributed
across the allowed range of Sφ . However, the combined information from both the bb̄ and
jj systems shows that the DPS events produce a sharp and substantial peak near Sφ ≃ π
which is well-separated from the total sample.
The narrow peaks near ∆φjj = π in Fig. 5a and near Sφ = 1 in Fig. 5b will be smeared

12
 12000 12000
(a) (b)
SPS SPS
10000 DPS 10000 DPS
SPS+DPS SPS+DPS

8000 8000
N_events / Bin

N_events / Bin
6000 6000

4000 4000

2000 2000

0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
Δφjj Sφ

FIG. 5: (a) The difference ∆φ in the azimuthal angles of light jet pairs for DPS and both SPS+DPS
events. The dijet pairs are back-to-back in DPS events. (b) The variable Sφ for DPS and SPS+DPS
events provides a stronger separation of the underlying DPS events from the total sample when
compared to ∆φ for any pair.

somewhat once soft QCD radiation and other higher-order terms are included in the calcu-
lation.
Another possibility for discerning DPS is the use of the total transverse momentum of
both the bb̄ and jj systems. At lowest order for a 2 → 2 process, the vector sum of the
transverse momenta of the final state pair vanishes. In reality, radiation and momentum
mismeasurement smear the expected peak near zero. Nevertheless, we still expect DPS
events to show a distribution in the transverse momenta of the jet pairs that is reasonably
well-balanced. To encapsulate this expectation for both light jet pairs and b-tagged pairs,
we use the variable [25]:
s 2  2
1 |pT (b1 , b2 )| |pT (j1 , j2 )|
Sp′ T =√ + . (23)
2 |pT (b1 )| + |pT (b2 )| |pT (j1 )| + |pT (j2 )|
Here pT (b1 , b2 ) is the vector sum of the transverse momenta of the two final state b jets, and
pT (j1 , j2 ) is the vector sum of the transverse momenta of the two (non b) jets.
The distribution in Sp′ T is shown in Fig. 6. As expected, we observe that the DPS events
are peaked near Sp′ T ∼ 0 and are well-separated from the total sample. The SPS events, on
the other hand, tend to be far from a back-to-back configuration and, in fact, are peaked
near Sp′ T ∼ 1. This behavior of the SPS events is presumably related to the fact that a large
number of the bb̄ or jj pairs arise from gluon splitting which yields a large pT imbalance

13
and, thus, larger values of Sp′ T .

(a) 7000 (b)

8000
SPS SPS
DPS DPS
6000
SPS+DPS SPS+DPS

6000 5000
N_events / Bin

N_events / Bin
4000
4000
3000

2000
2000

1000

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
SP ' SP '
T T

FIG. 6: Distribution of events in Sp′ T for the DPS and SPS samples. Due to the back-to-back
nature of the 2 → 2 events in DPS scattering, the transverse momenta of the jet pair and of the
b-tagged jet pair are small, resulting in a small value of Sp′ T . In (a) we show the Sp′ T distribution
for our standard cuts, and in (b) we increase the cut on the transverse momentum of the leading
jet, pj1 j1
T > 40 GeV. The fraction of DPS events in the whole sample decreases with increasing pT .

In this subsection, we find that extraction of the DPS “signal” for bb̄jj production from
the SPS “background” can be enhanced by combining information from both bb̄ and jj
systems. Our simulations suggest that the variable Sp′ T may be a more effective discriminator
than Sφ . However, given the leading order nature of our calculations and the absence of
smearing associated with initial state soft radiation, this picture may change and a variable
such as Sφ (or some other variable) may become a clearer signal of DPS at the LHC.
Realistically, it would be valuable to study both distributions once LHC data are available
in order to determine which is more instructive. In the following, we use the clear separation
shown in Fig. 6 in our exploration of the distinct properties of DPS and SPS events.

D. Two-dimensional distributions

The evidence in Fig. 5 and Fig. 6 for distinct regions of DPS dominance prompts the
search for greater discrimination in a plane represented by a two dimensional distribution of
one variable against another. We examined scatter plots involving the inter-plane angle Φ,
the jet-jet azimuthal angle difference ∆φjj , Sφ , and Sp′ T . Strong kinematic correlations are

14
evident in the plot of Sφ vs. Sp′ T at the level of our leading order calculation, and we observe
no additional separation of DPS and SPS beyond that evident in Figs. 5 and 6. Likewise,
there are strong correlations between ∆φjj and Sφ .
One scatter plot with interesting features is displayed in Fig. 7. The DPS events are
seen to be clustered near Sp′ T = 0 and are uniformly distributed in Φ. The SPS events peak
toward Sp′ T = 1 and show a roughly sin Φ character. While already evident in Figs. 3 and 6,
these two features are more apparent in the scatter plot Fig. 7. Moreover, the scatter plot
shows a valley of relatively low density between Sp′ T ∼ 0.1 and ∼ 0.4. In an experimental
one-dimensional Φ distribution such as Fig. 3, one would see the sum of the DPS and SPS
contributions. If structure is seen in data similar to that shown in the scatter plot Fig. 7,
one could make a cut at Sp′ T < 0.1 or 0.2 and verify whether the experimental distribution
in Φ is flat as expected for DPS events.

FIG. 7: Two-dimensional distribution of events in the variables Φ and Sp′ T for the DPS and SPS
samples.

In Fig. 4, we show that DPS produces a softer transverse momentum distribution for the

15
leading jet (either a b or light j). In data one would see only the sum of the DPS and SPS
components in a plot like Fig. 4. A scatter plot of Sp′ T vs. the transverse momentum of the
leading jet motivates an empirical separation of the two components. In Figs. 6(a) and 6(b)
we compare the Sp′ T distributions for two different selections on the transverse momentum
pj1
T of the leading jet in the bb̄jj sample. This comparison of the distributions confirms that

events in the DPS region, defined empirically by the region Sp′ T < 0.1 or 0.2, fall off more
steeply with pj1
T than the rest of the sample. It will be important and interesting to see

whether the selection Sp′ T < 0.1 or 0.2 in LHC data also produces events that show a more
rapid decrease with pj1
T .

The leading-jet transverse momentum distributions are shown in Figs. 8(a) and 8(b)
for two different cuts on Sp′ T . In both cases, we see that the SPS sample has a broader
distribution in pj1 j1
T and that the DPS sample dominates for small enough values of pT . For

our chosen value of σeff ∼ 12 mb, and for cuts we employ, the crossover points are roughly
80 GeV for Sp′ T < 0.2 and 40 GeV for Sp′ T < 0.4.

(a) (b)
SPS SPS
10000 DPS 10000 DPS
SPS+DPS SPS+DPS

1000 1000
N_events / Bin

N_events / Bin

100 100

10 10

1 1
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
Leading jet PT (GeV) Leading jet PT (GeV)

FIG. 8: The distribution in the transverse momentum of the leading jet pj1 ′
T for (a) SpT < 0.2 and

(b) Sp′ T < 0.4. As the signal region becomes more dominated by SPS events (i.e. moving from (a)
to (b)), the resulting distribution becomes harder and shifts the SPS-DPS cross-over from ∼ 80
GeV to ∼ 40 GeV.

16
III. FOUR JET PRODUCTION

In addition to bb̄jj, we can also ask how important DPS can be for a generic 4j final-state,
where none of the jets are b-tagged. In this section, we describe our calculation of the double
parton scattering and the single parton scattering contributions to the production of a 4j
final state, for which the cross section is larger. Our exposition can be brief since we repeat
the procedure described in some detail in Sec. II.

A. Outline of the method

The DPS process for 4j production is topologically equivalent to bb̄jj. However, in the
4j system, we lose the b-tagging ability that reduces the combinatorial background in bb̄jj,
and the prospects for isolating and measuring DPS over the SPS background may appear
less promising. Fortunately, in going from the bb̄ subprocess to the jj subprocess, a much
larger DPS rate is possible due to the much larger cross section for jj production. As we
show below, we find that the DPS signature can be extracted in this 4j mode as well.
The DPS cross section for 4j production receives contributions from the following sub-
processes at the lowest order:
jj ⊗ jj , bb̄ ⊗ jj, (24)

where both b-quarks fail the b-tag. We do not include the bb̄ ⊗ bb̄ process due to its relatively
small rate (∼ 0.14 nb). This rate is further reduced by requiring no b-tags, yielding roughly
40 events in the 10 pb−1 of luminosity assumed here.
Following Sec. II, we account for the possibility of an additional jet which is undetected
because it is too soft or outside of the accepted rapidity range. Thus, we include several
other contributions to the DPS cross section:

jjj ⊗ (j)j , jj(j) ⊗ jj , (25)

bb̄j ⊗ j(j) , bb̄(j) ⊗ jj, (26)
bb̄ ⊗ j(j)j , b(b̄) ⊗ jjj , (b)b̄ ⊗ jjj . (27)

where the parentheses surrounding a jet signify that it is not detected.

The SPS cross section receives contributions at lowest order from the final state:

jjjj , bb̄jj , (28)

17
where both b-quarks fail the b-tag, and, in the case where a jet is not detected, from the
final states:
bb̄(j)jj , (b)b̄jjj , b(b̄)jjj , (j)jjjj . (29)

We refer to Sec. II for the specification of acceptance cuts and detector resolution, and for
our treatment of the potential divergences present in the amplitudes for the processes in
Eqs. (24)-(29).

B. Results

Similar to the bb̄jj process, the leading jet in the 4j DPS sample is typically softer than
in the SPS channels (see Fig. 9). In this case, again using σeff = 12 mb, we find that the
cross-over between DPS and SPS dominance occurs near pT ≃ 50 GeV, higher than in the
bb̄jj case shown in Fig. 4. Improvement in the separation between DPS and SPS in the 4j
1x107

SPS
DPS
SPS+DPS
1x106
N_events / Bin

1x105

1x104

1x103
0 20 40 60 80 100 120 140
Leading jet PT (GeV)

FIG. 9: As in Fig. 4, but for 4j events. Similar to the bb̄jj sample, the SPS sample exhibits a
harder pT spectrum.

case can be achieved with an analogous version of the Sp′ T variable introduced in Eq. (23):
s 2  2
′ 1 |pT (ja , jb )| |pT (jc , jd )|
Sp T = √ + . (30)
2 |pT (ja )| + |pT (jb )| |pT (jc )| + |pT (jd )|

Here pT (ja , jb ) is the vector sum of the transverse momenta of two final state jets, a and b,
chosen among the four. The remaining c and d jets are then fixed. This choice is unique if a

18
separation of the two hard interactions is possible. In the bb̄jj system, the separation into the
bb̄ and jj subsystems via b-tagging removed most of the degeneracy (some degeneracy still
remained via tagging efficiencies or light jet mistagging). In the 4j system, the degeneracy
can at first glance be problematic as there are 3 possible pairings of the four jets.

1x106
SPS
DPS
SPS+DPS
8x105
N_events / Bin

6x105

4x105

2x105

0 0.2 0.4 0.6 0.8 1

Democratic SP '
T

FIG. 10: The democratic Sp′ T distribution for 4j events shows much more combinatorial background
than in the bb̄jj events. Even after accepting two mis-matched jet pairs, we see that the DPS and
SPS samples can still be separated well.

One might be tempted to take the pairing of jets which minimizes the value of Sp′ T . Unfor-
tunately, this choice places a bias on the distribution that makes it potentially problematic
to trust the discrimination. Instead, to construct Sp′ T we take all three combinations of pair-
ings, which includes one “correct” pairing and two incorrect pairings in the DPS process.
This “democratic” Sp′ T distribution is shown in Fig. 10 and is re-weighted by 1/3 for proper
normalization. As in the bb̄jj case, we see that the DPS distribution peaks near Sp′ T ∼ 0,
indicative that two back-to-back hard interactions are present. In addition to this expected
feature, we also see a continuum that extends above Sp′ T ∼ 0.1, associated with the wrong
combination taken in the democratic approach. In Fig. 10 we see that DPS produces a sec-
ondary peak at Sp′ T ∼ 1, not present in the bb̄jj case in Fig. 6. It appears to arise from the
wrong pairings of jets associated with the combinatorial background. In these instances, the
wrong combination of two jets that are close together in ∆R, meaning that their momenta
are aligned, can maximize the value of Sp′ T . Overall, we see that the DPS peak near Sp′ T = 0

19
provides a good means to separate DPS events from SPS events.
1x107 1x107
(a) (b)
SPS SPS
DPS DPS
SPS+DPS SPS+DPS
1x106 1x106
N_events / Bin

N_events / Bin
1x105 1x105

1x104 1x104

1x103 1x103
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
Leading jet PT (GeV) Leading jet PT (GeV)

FIG. 11: As in Fig. 8, but for 4j events with (a) democratic Sp′ T < 0.2 and (b) democratic
Sp′ T < 0.4. As in bb̄jj events, as one increases the cut on Sp′ T , the SPS fraction increases and the
total distribution is harder.

As in the bb̄jj case, we inspect the distribution in the pT of the leading jet after cuts
on the Sp′ T variable. Since there are three jet pairings per event, we now require that at
least one of the three pairings has Sp′ T in the given window. Due to this softer constraint,
the hardening of the pT spectrum of the leading jet is less dramatic than in the bb̄jj case
(e.g. compare Figs. 8 and 11). The crossover of the SPS and DPS contributions occurs near
80 GeV for Sp′ T < 0.2 and near 50 GeV for Sp′ T < 0.4

IV. DISCUSSION AND CONCLUSIONS

Our goal is to develop a method to search for a double parton scattering contribution
in the b b̄ j j and 4 jet final states at LHC energies and to measure the magnitude of its
contribution relative to the single parton contribution to the same final states. Based on our
parton level simulations, we find that variables such as Sp′ T and Sφ that take into account
information from the entire final state, thereby including both of the hard subprocesses
in DPS, are more effective at discrimination than variables such as ∆φjj that reflect only
a subset of the final-state. The enhancement at low values of Sp′ T shown in Figs. 6, 7
and 10 provides a good signature for the presence of double parton scattering. We urge
experimenters to search for such a concentration of events in data at the LHC. Having

20
found this enhancement, we then suggest that the magnitude of this peak be examined as a
function of the transverse momentum pj1
T of the leading jet in the event sample. The double

parton scattering contribution in the peak region should fall off more rapidly with pj1
T than

the rest of the sample. The distribution of events in the region of small values of Sp′ T should
also be examined as a function of the inter-plane angle Φ to see whether the flat behavior
is seen, as expected for two independent production processes. Once these characteristics
of double parton scattering are established, the data can be used to determine the effective
normalization σeff , defined and discussed in the Introduction. It will be interesting to see
whether the values extracted for σeff are about the same in the b b̄ j j and 4 jet final states
and how they compare with values measured at the Fermilab Tevatron.
Once double parton scattering is established in data, and σeff is determined, in a relatively
clean process such as bb̄jj, double parton contributions to a wide range of other processes
can be computed with more certainty about their expected rates at LHC energies. To be
sure, given the approximations described in the Introduction, some variation in the values
of σeff might be expected and appropriate for different processes and in different kinematic
regions. The connection of σeff with the effective size of the hard-scattering core of the
proton may mean that σeff will have different values for gg, qq, and q q̄ scattering.
There are several avenues for future work. Of great importance is the proper inclusion of
next-to-leading order contributions [32]. They are needed to make more robust predictions
of the relative normalization of the DPS and SPS contributions, of the shape of the pT
distribution of the leading jet, and for proper softening of the sharp peaks seen near Sp′ T = 1
in Figs. 6 and 10, and near Sφ = π in Fig. 5b.
It will also be important to develop joint probabilities H i,k (x1 , x2 , µA , µB ) that are more
sophisticated theoretically than the first approximation represented by Eq. (2) in which
parton-parton correlations are absent. A valuable development in this direction are the
studies presented in Refs. [16, 22].
Double parton contributions are potentially relevant for a wide range of standard model
processes, many already considered in the literature [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13,
14, 15, 17, 18, 19, 20, 21, 23, 24, 25], and they may also feed pertinent standard model
backgrounds to new physics processes [33]. They could be an issue in studies of Higgs boson
production in weak-boson-fusion since the “forward” jets could come from a second hard
subprocess.

21
V. ACKNOWLEDGMENTS

We benefited greatly from discussions with Dr. Thomas LeCompte and from communica-
tions with Dr. John Campbell during the early development of this project. We also thank
Tom, John, and Professor Jianwei Qiu for valuable comments and suggestions on an earlier
draft of this paper. Research in the High Energy Physics Division at Argonne is supported
by the U. S. Department of Energy under Contract No. DE-AC02-06CH11357. The research
of GS at Northwestern is supported by the U. S. Department of Energy under Contract No.
DE-FG02-91ER40684.

[1] C. Goebel, F. Halzen and D. M. Scott, Phys. Rev. D 22, 2789 (1980).
[2] N. Paver and D. Treleani, Nuovo Cim. A 70, 215 (1982); Phys. Lett. B 146, 252 (1984); Z.
Phys. C 28, 187 (1985).
[3] B. Humpert, Phys. Lett. B 131, 461 (1983).
[4] M. Mekhfi, Phys. Rev. D 32, 2371 (1985); Phys. Rev. D 32, 2380 (1985).
[5] B. Humpert and R. Odorico, Phys. Lett. B 154, 211 (1985).
[6] L. Ametller, N. Paver and D. Treleani, Phys. Lett. B 169, 289 (1986).
[7] F. Halzen, P. Hoyer and W. J. Stirling, Phys. Lett. B 188, 375 (1987).
[8] M. L. Mangano, Z. Phys. C 42, 331 (1989).
[9] R. M. Godbole, S. Gupta and J. Lindfors, Z. Phys. C 47, 69 (1990).
[10] M. Drees and T. Han, Phys. Rev. Lett. 77, 4142 (1996) [arXiv:hep-ph/9605430].
[11] O. J. P. Eboli, F. Halzen and J. K. Mizukoshi, Phys. Rev. D 57, 1730 (1998)
[arXiv:hep-ph/9710443].
[12] F. Yuan and K. T. Chao, J. Phys. G 24, 1105 (1998) [arXiv:hep-ph/9706293].
[13] G. Calucci and D. Treleani, Nucl. Phys. Proc. Suppl. 71, 392 (1999) [arXiv:hep-ph/9711225];
Phys. Rev. D 80, 054025 (2009) [arXiv:0907.4772 [hep-ph]]; Phys. Rev. D 79, 074013 (2009)
[arXiv:0901.3089 [hep-ph]].
[14] A. Del Fabbro and D. Treleani, Phys. Rev. D 61, 077502 (2000) [arXiv:hep-ph/9911358]; Phys.
Rev. D 66, 074012 (2002) [arXiv:hep-ph/0207311].
[15] A. Kulesza and W. J. Stirling, Phys. Lett. B 475, 168 (2000) [arXiv:hep-ph/9912232].

22
[16] V. L. Korotkikh and A. M. Snigirev, Phys. Lett. B 594, 171 (2004) [arXiv:hep-ph/0404155].
[17] E. Cattaruzza, A. Del Fabbro and D. Treleani, Phys. Rev. D 72, 034022 (2005)
[arXiv:hep-ph/0507052].
[18] M. Y. Hussein, Nucl. Phys. Proc. Suppl. 174, 55 (2007) [arXiv:hep-ph/0610207]; Nucl. Phys.
Proc. Suppl. 152, 296 (2006); arXiv:0710.0203 [hep-ph].
[19] E. Maina, JHEP 0909, 081 (2009) [arXiv:0909.1586 [hep-ph]]; JHEP 0904, 098 (2009)
[arXiv:0904.2682 [hep-ph]].
[20] S. Domdey, H. J. Pirner and U. A. Wiedemann, arXiv:0906.4335 [hep-ph].
[21] D. d’Enterria, G. K. Eyyubova, V. L. Korotkikh, I. P. Lokhtin, S. V. Petrushanko,
L. I. Sarycheva and A. M. Snigirev, arXiv:0910.3029 [hep-ph].
[22] J. R. Gaunt and W. J. Stirling, arXiv:0910.4347 [hep-ph].
[23] T. Akesson et al. [Axial Field Spectrometer Collaboration], Z. Phys. C 34, 163 (1987).
[24] F. Abe et al. [CDF Collaboration], Phys. Rev. D 56, 3811 (1997); Phys. Rev. Lett. 79, 584
(1997).
[25] D0 Collaboration, “Double parton interactions in γ + 3 jet events in pp̄ collisions
√
at s = 1.96 TeV in D0”, D0 Conference Note 5910-CONF (April 24, 2009);
http://www-d0.fnal.gov/Run2Physics/WWW/results/prelim/QCD/Q13/Q13.pdf.
[26] A. H. Mueller and J. W. Qiu, Nucl. Phys. B 268, 427 (1986).
[27] X. F. Guo, Phys. Rev. D 58, 036001 (1998) [arXiv:hep-ph/9711453]; Nucl. Phys. A 638, 539C
(1998) [arXiv:hep-ph/9801450].
[28] J. Pumplin, D. R. Stump, J. Huston, H. L. Lai, P. M. Nadolsky and W. K. Tung, JHEP 0207,
012 (2002) [arXiv:hep-ph/0201195].
[29] F. Maltoni and T. Stelzer, JHEP 0302, 027 (2003) [arXiv:hep-ph/0208156].
[30] M. L. Mangano, M. Moretti, F. Piccinini, R. Pittau and A. D. Polosa, JHEP 0307, 001 (2003)
[arXiv:hep-ph/0206293].
[31] H. Baer, V. Barger, G. Shaughnessy, H. Summy and L. T. Wang, Phys. Rev. D 75, 095010
(2007) [arXiv:hep-ph/0703289].
[32] Many next-to-leading order results can be obtained from the Monte Carlo program MCFM,
J. M. Campbell and R. K. Ellis, Phys. Rev. D 60, 113006 (1999). Other NLO publications
include L. J. Dixon, Z. Kunszt, and A. Signer, Phys. Rev. D 60, 114037 (1999) and Nucl.
Phys. B 531, 3 (1998).

23
[33] Z. Sullivan and E. L. Berger, Phys. Rev. D 78, 034030 (2008) [arXiv:0805.3720 [hep-ph]].

24