Skip to main content

Full text of "Tuning Monte Carlo Generators: The Perugia Tunes"

See other formats


Tuning Monte Carlo Generators: The Perugia Tunes 

P. Z. Skands (peter. skands@cern.ch) 
CERN PH-TH, Case 01600, CH-1211 Geneva 23, Switzerland 

Abstract 

We present 9 new tunes of the px-ordered shower and underlying-event model in Pythia 

^^ i 6.4. These "Perugia" tunes update and supersede the older "SO" family. The data sets used 

^^ ■ to constrain the models include hadronic Z^ decays at LEP, Tevatron min-bias data at 630, 

>^; 1800, and 1960 GeV, Tevatron Drell-Yan data at 1800 and 1960 GeV, and SPS min-bias data 

at 200, 546, and 900 GeV. In addition to the central parameter set, called "Perugia 0" , we 

introduce a set of 8 related "Perugia variations" that attempt to systematically explore soft, 

hard, parton density, and colour structure variations in the theoretical parameters. Based 

on these variations, a best-guess prediction of the charged track multiplicity in inelastic, 

nondiffractive minimum-bias events at the LHC is made. Note that these tunes can only 

be used with Pythia 6, not with Pythia 8. Note: this report was updated in March 

2011 with a new set of variations, collectively labeled "Perugia 2011", that are optimized 

Mh| for matching applications and which also take into account some lessons from the early LHC 

^ ■ data. In order not to break the original text, these are described separately in Appendix iBl 

'^ ; Contents 

lO . 1 Introduction 



in 



X 



2 Procedure 

2.1 Manual vs Automated Tuning 



in 

Q I 2.2 Sequence of Tuning Steps 



3 Main Features of the Perugia Tunes 

3.1 Final-State Radiation and Hadronisation (Table [2|) 

3.2 Initial-State Radiation and Primordial kT (Table [3]) 

5^ I 3.3 Underlying Event, Beam Remnants, and Colour Reconnections (Table H]) 

3.4 Energy Scaling (Table HD 



4 The Perugia Tunes: Tune by Tune \24 

5 Extrapolation to the LHC |27 

6 Conclusions |2 
A Parameters for the Perugia Tunes |31 
B The Perugia 2011 Tunes \M 
C Overview of Tunes included in PYTHIA ISj, 



1 Introduction 

Perturbative calculations of collider observables rely on two important prerequisites: factorisa- 
tion and infrared (IR) safety. These are the tools that permit us to relate theoretical calculations 
to detector-level measured quantities, up to corrections of known dimensionality, which can then 
be suppressed (or enhanced) by appropriate choices of the dimensionful scales appearing in the 
observable and process under study. However, in the context of the underlying event (UE), 
say, we are faced with the fact that we do not (yet) have formal factorisation theorems for this 
component — in fact the most naive attempts at factorisation can easily be shown to fail [Il[2]. 
At the same time, not all collider measurements can be made insensitive to the UE at a level 
comparable to the achievable experimental precision, and hence the extraction of parameters 
from such measurements acquires an implicit dependence on our modeling of the UE. Further, 
when considering observables such as track multiplicities, hadronisation corrections, or even 
short-distance quantities if the precision required is very high, we are confronted with observ- 
ables which may be experimentally well measured, but which are explicitly sensitive to infrared 
physics. 



The Role of Factorisation: Let us begin with factorisation. When applicable, factorisation 
allows us to subdivide the calculation of an observable (regardless of whether it is IR safe or 
not) into a perturbatively calculable short-distance part and an approximately universal long- 
distance part, the latter of which may be modeled and constrained by fits to data. However, in 
the context of hadron collisions, the possibilities of multiple perturbative parton-parton inter- 
actions and parton rescattering processes explicitly go beyond the factorisation theorems so far 
developed. Part of the problem is that the underlying event may contain short-distance physics 
of its own, that can be as hard as, or even harder than, the bremsstrahlung emissions associ- 
ated with the scattering that triggered the event. Hence the conceptual separation into what we 
think of as "hard-scattering" and "underlying-event" components is not necessarily equivalent 
to a clean separation in terms of "short-distance" and "long-distance" physics. Indeed, from 
ISR energies [3] through the SPS [Hl^ to the Tevatron pl-fTO]. and also in photoproduction at 
HERA [11], we see evidence of (perturbative) "minijets" in the underlying event, beyond what 
bremsstrahlung alone appears to be able to account for. It therefore appears plausible that a 
universal modeling of the underlying event must take into account that the hard-scattering and 
underlying-event components can involve similar time scales and have a common, correlated 
evolution. It is in this spirit that the concept of "interleaved evolution" ^2\ was developed as 
the cornerstone of the p^-ordered models jl2[|13j in both Pythia 6 [H] and, more recently, 
Pythia 8 [15], the latter of which now also incorporates a model of parton rescattering |16j . 



The Role of Infrared Safety: The second tool, infrared safetjo, provides us with a class 
of observables which are insensitive to the details of the long-distance physics. This works 
up to corrections of order the long-distance scale divided by the short-distance scale to some 



By "infrared" we here mean any non-U V limit, without regard to whether it is collinear or soft. 



(observable-dependent) power, typically 



Q 



2 



IR Safe Corrections oc ^ (1) 

Quv 

where Qxjy denotes a generic hard scale in the problem, and Qir ~ Aqcd ~ 0(1 GeV). Of 
course, in minimum-bias, we typically have Quv ~ Qm^ wherefore all observables depend 
significantly on the IR physics (or in other words, when IR physics is all there is, then any 
observable, no matter how carefully defined, depends on it). 

Even when a high scale is present, as in resonance decays, jet fragmentation, or underlying- 
event-type studies, infrared safety only guarantees us that infrared corrections are small, not 
that they are zero. Thus, ultimately, we run into a precision barrier even for IR safe observables, 
which only a reliable understanding of the long-distance physics itself can address. 

Finally, there are the non- infrared-safe observables. Instead of the suppressed corrections 
above, such observables contain logarithms 



Q 



2 
UV 



IR Sensitive Corrections oc a" log"' -j^ , m < 2n , (2) 



Q 



2 
IR 



which grow increasingly large as Qir/Qw ~^ 0. As an example, consider such a fundamental 
quantity as particle multiplicities; in the absence of nontrivial infrared effects, the number of 
partons that would be mapped to hadrons in a naive local-parton-hadron-duality |17j picture 
would tend logarithmically to infinity as the IR cutoff is lowered. Similarly, the distinction 
between a charged and a neutral pion only occurs in the very last phase of hadronisation, and 
hence observables that only include charged tracks are always IR sensitive. 

Minimum-Bias and the Underlying Event: Minimum-bias (MB) and Underlying- Event 
(UE) physics can therefore be perceived of as offering an ideal lab for studying nonfactorised 
and nonperturbative phenomena, with the added benefit of having access to the highest possible 
statistics in the case of min-bias. In this context there is no strong preference for IR safe over 
IR sensitive observables; they merely represent two different lenses through which we can view 
the infrared physics, each revealing different aspects. By far the most important point is that it 
is in their combination that we achieve a sort of stereo vision, in which infrared safe observables 
measuring the overall energy flow are simply the slightly averaged progenitors of the spectra 
and correlations that appear at the level of individual particles. A systematic programme of 
such studies can give crucial tests of our ability to model and understand these ubiquitous 
components, and the resulting improved physics models can then be fed back into the modeling 
of high-p_|_ physics. 

Starting from early notions such as "KNO scaling" of multiplicity distributions [H] , a large 
number of theoretical and experimental investigations have been brought to bear on what the 
physics of a generic, unbiased sample of hadron collisions looks like (for a recent review, see, 
e.g., [I9] and references therein). However, in step with the gradual shift in focus over the last 
two decades, towards higher-p_|_ ("maximum-bias") physics, the field of QCD entered a golden 
age of perturbative calculations and infrared safety, during which time the unsafe "soft" physics 



became viewed increasingly as a non-perturbative quagmire, into the depths of which ventured 
only fools and old men. 

From the perspective of the author's generation, it was chiefly with a comprehensive set of 
measurements carried out by Rick Field using the CDF detector at the Tevatron |20H25| . that 
this perception began to change back towards one of a definable region of particle production 
that can be subjected to rigorous scrutiny in a largely model-independent way, and an ambitious 
programme of such measurements is now being drawn up for the LHC experiments. In other 
words, a well-defined experimental laboratory has been prepared, and is now ready for the 
testing of theoretical models. 

Simultaneously with the LHC efforts, it is important to remember that interesting connec- 
tions are also being explored towards other, related, fields, such as cosmic ray fragmentation 
(related to forward fragmentation at the LHC) and heavy- ion physics (related to collective phe- 
nomena in hadron-hadron interactions). A nice example of this interplay is given, for instance, 
by the Epos model [26] , which originated in the heavy- ion community, but uses a parton-based 
model as input and whose properties in the context of ultra-high-energy cosmic ray fragmenta- 
tion are currently being explored |27p28| . Also methods from the field of numerical optimisation 
are being applied to Monte Carlo tuning (cf., e.g., the Professor |29] and Profit [30] frameworks), 
and there are tempting connections back to perturbative QCD. Along the latter vein, we be- 
lieve that by bringing the logarithmic accuracy of perturbative parton shower calculations under 
better control, there would be less room for playing out ambiguities in the non-perturbative 
physics against ambiguities on the shower side, and hence the genuine soft physics could also be 
revealed more clearly. This is one of the main motivations behind the ViNCiA project |3HI32j. 

For the present, as part of the effort to prepare for the LHC era and spur more interplay 
between theorists and experimentalists, we shall here report on a new set of tunes of the p±- 
ordered Pythia framework, which update and supersede the older "SO" family |33H36] . We have 
focused in particular on the scaling from lower energies towards the LHC (see also |37H40] ) and 
on attempting to provide at least some form of theoretical uncertainty estimates, represented 
by a small number of alternate parameter sets that systematically explore variations in some 
of the main tune parameters. The full set of new tunes have been made available starting 
from Pythia version 6.4.23 (though some have been available longer; see the Pythia update 
notes [H] for details). 

This concludes a several-year long effort to present the community with an optimized set 
of parameters that can be used as default settings for the so-called "new" interleaved shower 
and underlying-event model in Pythia 6. The author's intention is to now move fully to the 
development of Pythia 8. We note that the Perugia tunes can unfortunately not be used 
directly in Pythia 8, since it uses slightly different parton-shower and colour-reconnection 
models. A separate set of tunes for Pythia 8 are therefore under development, with several 
already included in the current version 8.1.42 of that generator. 

We also present a few distributions that carry interesting information about the underlying 
physics, updating and complementing those contained in |361l42j. For brevity, this text only 
includes a representative selection, with more results available on the web |43j . 

The main point is that, while any plot of an infrared sensitive quantity represents a compli- 
cated cocktail of physics effects, such that any sufficiently general model presumably could be 
tuned to give an acceptable description observable by observable, it is very difficult to simul- 



taneously describe the entire set. The real game is therefore not to study one distribution in 
detail, for which a simple fit would in principle suffice, but to study the degree of simultaneous 
agreement or disagreement over many, mutually complementary, distributions. 

2 Procedure 

2.1 Manual vs Automated Tuning 

Although Monte Carlo models may appear to have a bewildering array of independently ad- 
justable parameters, it is worth keeping at the front of one's mind that most of these parameters 
only control relatively small (exclusive) details of the event generation. The majority of the 
(inclusive) physics is determined by only a few, very important ones, such as, e.g., the value of 
the strong coupling, in the perturbative domain, and the form of the fragmentation function 
for massless partons, in the non-perturbative one. 

Manual Tuning: Armed with a good understanding of the underlying model, and using only 
the generator itself as a tool, a generator expert would therefore normally take a highly factorised 
approach to constraining the parameters, first constraining the perturbative ones and thereafter 
the non-perturbative ones, each ordered in a measure of their relative significance to the overall 
modeling. This factorisation, and carefully chosen experimental distributions corresponding to 
each step, allows the expert to concentrate on just a few parameters and distributions at a time, 
reducing the full parameter space to manageable-sized chunks. Still, each step will often involve 
more than one single parameter, and non-factorizable corrections still imply that changes made 
in subsequent steps can change the agreement obtained in previous ones by a non-negligble 
amount, requring additional iterations from the beginning to properly tune the entire generator 
framework. 

Due to the large and varied data sets available, and the high statistics required to properly 
explore tails of distributions, mounting a proper tuning effort can therefore be quite intensive — 
often involving testing the generator against the measured data for thousands of observables, 
collider energies, and generator settings. Although we have not kept a detailed record, an 
approximate guess is that the generator runs involved in producing the particular tunes reported 
on here consumed on the order of 1.000.000 CPU hours, to which can be added an unknown 
number of man-hours. While some of these man-hours were undoubtedly productive, teaching 
the author more about his model and resulting in some of the conclusions reported on in this 
paper, most of them were merely tedious, while still disruptive enough to prevent getting much 
other work done. 

The main steps followed in the tuning procedure for the Perugia tunes are described in more 
detail in section [2^ below. 

Automated Tuning: As mentioned in the introduction, recent years have seen the emer- 
gence of automated tools that attempt to reduce the amount of both computer and manpower 
required. The number of machine hours can, for instance, be substantially reduced by making 
full generator runs only for a limited set of parameter points, and then interpolating between 



these to obtain approximations to what the true generator result would have been for any in- 
termediate parameter point. In the Professor tool |29p44j. which we rely on for our LEP tuning 
here, this optimization technique is used heavily, so that after an initial (intensive) initialisation 
period, approximate generator results for any set of generator parameters within the sampled 
space can be obtained without any need of further generator runs. Taken by itself, such opti- 
mization techniques could in principle also be used as an aid to manual tuning, but Professor, 
and other tools such as Profit |30j . attempt to go a step further. 

Automating the human expert input is of course more difficult (so the experts believe). 
What parameters to include, in what order, and which ranges for them to consider "physical"? 
What distributions to include, over which regions, how to treat correlations between them, 
and how to judge the relative importance, for instance, between getting the right average of 
an observable versus getting the right asymptotic slope? In the tools currently on the market, 
these questions are addressed by a combination of input solicited from the generator authors 
(e.g., which parameters and ranges to consider, which observables constitute a complete set, 
etc) and the elaborate construction of non-trivial weighting functions that determine how much 
weight is assigned to each individual bin and to each distribution. The field is still burgeoning, 
however, and future sophistications are to be expected. Nevertheless, at this point the overall 
quality of the tunes obtained with automated methods appear to the author to at least be 
competitive with the manual ones. 

2.2 Sequence of Tuning Steps 

We have tuned the Monte Carlo in five consecutive steps (abbreviations which we use often 
below are highlighted in boldface): 

1. Final-State Radiation (FSR) and Hadronisation (HAD): using LEP data j45l|46]. For 
most of the Perugia tunes, we take the LEP parameters given by the Professor collabora- 
tion |29|l44j. This improves several event shapes and fragmentation spectra as compared 
to the default settings. For hadronic yields, especially (p^ was previously wrong by more 
than a factor of 2, and r] and r/' yields have likewise been improved. For a "HARD" and a 
"SOFT" tune variation, we deliberately change the renormalisation scale for FSR slightly 
away from the central Professor value. Also, since the Professor parameters were origi- 
nally optimised for the Q^-ordered parton shower in Pythia, the newest (2010) Perugia 
tune goes slightly further, by changing the other fragmentation parameters (by order of 
5-10% relative to their Professor values) in an attempt to improve the description of high- 
z fragmentation and strangeness yields reported at LEP [JSlHB] and at RHIC [171 HH]. 
relative to the Professor p^-ordered tuning. The amount of ISR jet broadening (i.e., FSR 
off ISR) in hadron collisions has also been increased in Perugia 2010, relative to Perugia 
0, in an attempt to improve hadron collider jet shapes and rates [191I50]. 

2. Initial-State Radiation (ISR) and Primordial fc^: using the Drell-Yan p± spectrum at 
1800 and 1960 GeV, as measured by CDF [5l] and D0 [52], respectively. Note that we 
treat the data as fully corrected for photon bremsstrahlung effects in this case, i.e., we 
compare the measured points to the Monte Carlo distribution of the "original Z boson" . 
We are aware that this is not a physically meaningful observable definition, but believe 



it is the closest we can come to the definition actually used for the data points in both 
the CDF and D0 studies. See [53] for a more detailed discussion of this issue. Again, we 
deliberately change the renormalisation scale for ISR away from its best fit value for the 
HARD and SOFT variations, by about a factor of 2 in either direction, which does not 
appear to lead to serious conflict with the data (see distributions below). 

3. Underlying Event (UE), Beam Remnants (BR), and Colour Reconnections (CR): using 
Nch [MIES], dNch/dp± [MIEZ], and (p_l) (A^ch) [SZ] in min-bias events at 1800 and 1960 
GeV, as measured by CDF. Note that the A^ch spectrum extending down to zero p± 
measured by the E735 Collaboration at 1800 GeV [58] was left out of the tuning, since we 
were not able to consolidate this measurement with the rest of the data. We do not know 
whether this is due to intrinsic limitations in the modeling (e.g., mismodeling of the low- 
p± and/or high-ry regions, which are included in the E735 result but not in the CDF one) 
or to a misinterpretation on our part of the measured observable. Note, however, that the 
E735 collaboration itself remarks |58j that its results are inconsistent with those reported 
by UA5 1591160] over the entire range of energies where both experiments have data. So far, 
the early LHC results at 900 GeV appear to be consistent with UA5, within the limited rj 
regions accessible to the experiments [611163] . but it remains important to check the high- 
multiplicity tail in detail, in as large a phase space region as possible. We also note that 
there are some discrepancies between the CDF Run-1 [54] and Run-2 [55] measurements 
at very low multiplicities, presumably due to ambiguities in the procedure used to correct 
for diffraction. We have here focused on the high-multiplicity tail, which is consistent 
between the two. Hopefully, this question can also be addressed by comparisons to early 
low-energy LHC data. Although the 4 main LHC experiments are not ideal for diffractive 
studies and cannot identify forward protons, it is likely that a good sensitivity can still 
be obtained by requiring events with large rapidity gaps, where the gap definition would 
essentially be limited by the noise levels achievable in the electromagnetic calorimeters. 

4. Energy Scaling: using Nch in min-bias events at 200, 546, and 900 GeV, as measured by 
UA5 [591I6Q], and at 630 and 1800 GeV, as measured by CDF [S]. 

5. The last two steps were iterated a few times. 

Remarks on Jet Universality: Note that the clean separation between the first and second 
points in the list above assumes jet universality, i.e., that a Z^, for instance, fragments in the 
same way at a hadron collider as it did at LEP. This is not an unreasonable first assumption [64] , 
but since the infrared environment in hadron collisions is characterised by a different (hadronic) 
initial-state vacuum, by a larger final-state gluon component, and also by simply having a lot 
more colour flowing around in general, it is still important to check to what precision it holds 
explicitly, e.g., by measuring multiplicity and pj_ spectra of identified particles, particle-particle 
correlations, and particle production ratios (e.g., strange to unstrange, vector to pseudoscalar, 
baryon to meson, etc.) in situ at hadron colliders. We therefore very much encourage the 
LHC experiments not to blindly rely on the constraints implied by LEP, but to construct and 
publish their own full-fledged sets of fragmentation constraints using identified particles. This 



is the only way to verify explicitly to what extent the models extrapolate correctly to the LHC 
environment, and gives the possibility to highlight and address any discrepancies. 

Remarks on Diffraction: Note also that the modeling of diffraction in Pythia 6 lacks a 
dedicated modeling of diffractive jet production, and hence we include neither elastic nor diffrac- 
tive Monte Carlo events in any of our comparisons. This affects the validity of the modeling 
for the first few bins in multiplicity. Due also to the discrepancy noted above between the two 
CDF measurements in this region [MIES], we therefore assigned less importance to these bins 
when doing the tuneqj. We emphasise that widespread use of ill-defined terminologies such as 
"Non-Single Diffractive" (NSD) events without an accompanying definition of what is meant by 
that terminology at the level of physical observables contributes to the ambiguities surrounding 
diffractive corrections in present data sets. Since different diffraction models produce different 
spectra at the observable level, an intrinsic ambiguity is introduced which was not present in 
the raw data. We strongly encourage future measurements if not to avoid such terminologies 
entirely then to at least also make data available in a form which is defined only in terms of 
physical observables, i.e., using explicit cuts, weighting functions, and/or trigger conditions to 
emphasise the role of one component over another. 

Remarks on Observables: Finally, note that we did not include any explicit "underlying- 
event" observables in the tuning. Instead, we rely on the large-multiplicity tail of minimum-bias 
events to mimic the underlying event. A similar procedure was followed for the older "SO" 
tune [33l[34]j_^which gave a very good simultaneous description of underlying-event physics at 
the Tevatroro. Conversely, Rick Field's "Tune A" |38ll65j gave a good simultaneous description 
of minimum-bias data, despite only having been tuned on underlying-event data. Tuning to one 
and predicting the other is therefore not only feasible but simultaneously a powerful cross-check 
on the universality properties of the modeling. 

Additional important quantities to consider for further model tests and tuning would be 
event shapes at hadron colliders |501I66|. observables involving explicit jet reconstruction — 
including so-called "charged jets" [21] (a jet algorithm run on a set of charged tracks, omitting 
neutral energy), which will have fluctuations in the charged-to-neutral ratio overlaid on the 
energy flow and therefore will be more IR than full jets, but still less so than individual particles, 
and "EM jets" (a jet algorithm run on a set of charged tracks plus photons), which basically adds 
back the vr*^ component to the charged jets and hence is less IR sensitive than pure charged jets 
while still remaining free of the noisy environment of hadron calorimeters — explicit underlying- 
event, fragmentation, and jet structure (e.g., jet mass, jet shape, jet-jet separation) observables 
in events with jets [6ll20ll241l49ll67ti72] . photon -|- jet(s) events (including the important 7 -|- 3-jet 
signature for double-parton interactions [8l[T0]), Drell-Yan events [20l[72l[73], and observables 
sensitive to the initial-state shower evolution in DIS (see, e.g., [371174]). 



^To ensure an apples-to-apples comparison for the low-multiplicity bins between these models and present 
measurements, one must take care to include any relevant diffrative components using a (separate) state-of-the-art 
modeling of diffraction. 

^Note: when extrapolating to lower energies, the alternative scaling represented by "SOA" appears to be 
preferred over the default scaling used in "SO" . 



As mentioned above, it is also important that fragmentation models tuned at LEP be 
tested in situ at hadron colliders. To this effect, single-particle multiplicities and momentum 
spectra for identified particles such as Kg, vector mesons, protons, and hyperons (in units of 
GeV and/or normalised to a global measure of transverse energy, such as, e.g., the p± of a 
jet when the event is clustered back to a dijet topology) are the first order of business, and 
particle-particle correlations the second (e.g., how charge, strangeness, baryon number, etc., are 
compensated as a function of a distance measure and how the correlation strength of particle 
production varies over the measured phase space region). Again, these should be considered at 
the same time as less infrared sensitive variables measuring the overall energy flow. We expect 
a programme of such measurements to gradually develop as it becomes possible to extract more 
detailed information from the LHC data and note that some such observables, from earlier 
experiments, have already been included, e.g., in the Rivet framework, see [29], most notably 
underlying-event observables from the Tevatron, but also recently some fragmentation spectra 
from RHIC |471I48|. See also the underlying-event sections in the HERA-and-the-LHC j37j . 
Tevatron- for-LHC [38], and Les Houches write-ups [39]. A complementary and useful guide to 
tuning has been produced by the ATLAS collaboration in the context of their MC09 tuning 
efforts [75]. 

3 Main Features of the Perugia Tunes 

Let us first describe the overall features common to all the Perugia tunes, divided into the same 
main steps as in the outline of the tuning procedure given in the preceding section: 1) final-state 
radiation and hadronisation, 2) initial-state radiation and primordial k^, 3) underlying event, 
beam remnants, and colour reconnections, and 4) energy scaling. Each step will be accompanied 
by plots to illustrate salient points and by a summary table in appendix [A] giving the Perugia 
parameters relevant to that step, as compared to the older Tune SOA-Pro, which serves as our 
reference. We shall then turn to the properties of the individual tunes in the following section, 
and finally to extrapolations to the LHC in the last section. 

3.1 Final-State Radiation and Hadronisation (Table [2D 

As mentioned above, we have taken the LEP tune obtained by the Professor group [291144] as 
our starting point for the FSR and HAD parameters for the Perugia tunes. Since we did not 
perform this part of the tuning ourselves, we treat these parameters almost as fixed inputs, and 
only a very crude first attempt at varying them was originally made for the Perugia HARD 
and SOFT variations. This is refiected in the relatively small differences between the FSR 
and HAD parameters listed in table [21 compared to SOA-Pro which uses the original Professor 
parameters. (E.g., most of the tunes use the same parameters for the longitudinal fragmentation 
function applied in the string hadronisation process, including the same Lund functions j76] 
for light quarks and Bowler functions [77] for heavy quarks.) With the most recent Perugia 
2010 tune, an effort was made to manually improve jet shapes, strangeness yields, and high- 2; 
fragmentation, which is the reason several of the hadronisation parameters differ in this tune 
as well as in its sister tune Perugia K. A more systematic exploration of variations in the 
fragmentation parameters is certainly a point to return to in the future, especially in the light 



of the new identified-particle spectra and jet shape data that will hopefully soon be available 
from the LHC experiments. For the present, we have focused on the the uncertainties in the 
hadron-collider-specific parameters, as follows. 

3.2 Initial-State Radiation and Primordial kx (Table [3]) 

Evolution Variable, Kinematics, and Renormalisation Scale: One of the most signif- 
icant changes when going from the old (virtuality-ordered) to the new (p_L-ordered) ISR/FSR 
model concerns the Drell-Yan pj_ spectrum. In the old model, when an originally massless ISR 
parton evolves to become a jet with a timelike invariant mass, then that original parton is 
pushed off its mass shell by reducing its momentum components. In particular the transverse 
momentum components are reduced, and hence each final-state emission off an ISR parton ef- 
fectively removes p± from that parton, and by momentum conservation also from the recoiling 
Drell-Yan pair. Via this mechanism, the p± distribution generated for the Drell-Yan pair is 
shifted towards lower values than what was initially produced. 

Compared to data, this appears to effectively cause any tune of the old Pythia framework 
with default ISR settings — such as Tune A or the ATLAS DC2/"Rome" tune — to predict 
a too narrow spectrum for the Drell-Yan p± distribution, as illustrated by the comparison of 
Tune A to CDF and D0 data in fig. [T] (left column). (The inset shows the high-px tail which in 
all cases is matched to Z+jet matrix elements, the default in Pythia for both the virtuality- 
and p_L-ordered shower models.) We note that a recent theoretical study [^ using virtuality- 
ordering with a different kinematics map did not find this problem, consistent with our suspicion 
that it is not the virtuality ordering per se which results in the narrow shape, but the specific 
recoil kinematics of FSR off FSR in the old shower model. 

To re-establish agreement with the measured spectrum without changing the recoil kinemat- 
ics, the total amount of ISR in the old model had to be increased. This can be accomplished, 
e.g., by choosing very low values of the renormalisation scale (and hence large ag values) for 
ISR, as illustrated by tunes DW-Pro and Pro-Q20 in fig. [T] (left column). To summarise, the 
as choices corresponding to each of the three tunes of the old shower shown in the left pane of 
fig. m are, 



ISR 

Q^-ordering 



Tune A (100) 


asipl) 


MS, 


1-loop, 


ACTEQSL 


Tune DW (103) 


as{0.2pl) 


MS, 


1-loop, 


AcTEQSL 


Tune Pro-Q20 (129) 


asiO.Upl) 


MS, 


1-loop, 


AcTEQSL 



(3) 



where, for completeness, we have given also the renormalisation scheme, loop order, and choice 
of AqcD) which are the same for all the tunes. 

While the increase of ag nominally reestablishes a good agreement with the Drell-Yan p± 
spectrum, the whole business does smell faintly of fixing one problem by introducing another 
and hence the defaults in Pythia for these parameters have remained the Tune A ones, at the 
price of retaining the poor agreement with the Drell-Yan spectrum. 

In the new p^-ordered showers [12], however, FSR off ISR is treated within individual QCD 
dipoles and does not affect the Drell-Yan p±. This appears to make the spectrum come out 
generically much closer to the data, as illustrated by the SO(A) curves in fig. [1] (right column), 
which use as{p±). The only change going to Perugia — which can be seen to be slightly 



10 



1800GeVp+pbar 



5 10 



10 



8 



Transverse Momentum of Z/y* [66 < M^,. ./GeV < 1 1 6] 

^ CDF data 



^* 



» A 
-»-- DW(T) 
■*■- Pro-Q20 



•C^t 



^ 



N 



^%. 



-t:_ 



t;. 



i'*-i 



■-■^, 



^ 



50 100 150 200 250 

Pythia 6,423 
Data from CDF Collaboration, PRL84(2000)845 



10 



15 20 25 30 

p^(hard system) [GeV] 




15 20 25 30 

p^(hard system) [GeV] 



•S 10 - 



10 






1 960 GeV p+pbar Drell-Yan 


Transverse Momentum of Z/y* [40 < M^,,^/GeV < 200] 


^ DO data 
'- //' Nj --*-- DW 


" ^'" I"';. -■*■- Pro-Q20 


-1 
10 

-2 

" 10 

-3 
10 

7 10 

-5 
- 10 




- 10 


- ^^ --v^ 


c 


50 too 150 200 250 » 


Pythia 6,423 


Data from DO Colfaboralion, PRL1 00(2008)1 02002 



10 



15 20 25 30 

p^(hard system) [GeV] 



1 960 GeV p+pbar 



Transverse Momentum ot Z/7* [40 < M^ ,/GeV < 200] 
+ DO data 



SOA 

Perugia 
Perugia HARD 
Perugia SOFT 




Pythia 6,423 
Datairom DO Collaboration, PRL100{2008)102002 



10 



15 20 25 30 

p^(hard system) [GeV] 



Figure 1: Comparisons to the CDF and D0 measurements of the p± of Drell-Yan pairs [5T[[52] . Insets 
show the high-pj^ tails. Left: virtuality-ordered showers. Right: px-ordered showers. See [43| for other 
tunes and collider energies. 



harder — was implementing a translation from the MS definition of A used previously, to the 
so-called CMW choice [79] for A, similarly to what is done in Herwig [HOllHT]. 

For both CTEQ5L and CTEQ6L1, the A^g^ value in the PDF set is derived with an LO 
(1-loop) running of Og, which is also what we use in the backwards evolution algorithm in our 
ISR model. In the Perugia tunes (and also in Pythia by default) we therefore let the ag value 
for the ISR evolution be determined by the PDF set. The MRST LO* set [82], however, uses 
an NLO (2-loop) running for a^, which gives a roughly 50% larger value for A. Since we do 
not change the loop order of our ISR evolution, this higher A value would lead to an increase 
in, e.g., the mean Drell-Yan p± at the Tevatron. In practice, however, this point is obscured 



11 



by the fact that the LHAPDF interface, used in our code (v. 5. 8.1), does not return the correct 
^QCD value for each PDF set. Instead, a constant value of 0.192 (corresponding to CTEQ5L) 
is returned. Since we were not aware of this bug in the interface when performing the Perugia 
tunes, we therefore note that all the tunes are effectively using the CTEQ5L value of A. The 
pace of evolution with the LO* PDF set is still slightly higher than for CTEQ5L, however. To 
compensate for this, the renormalisation scale was chosen slightly higher for the LO* tune, cf. 
the PARP(64) values in table El 

We note that a similar issue afflicted the original CTEQ6L set, which used an NLO Ug (with 
a correspondingly larger value of A). We here use the revised CTEQ6L1 set for our Perugia 
6 tune, which uses an LO running and hence should be more consistent with the evolution 
performed by the shower. Similarly, the LO* set used here could be replaced by the newer 
LO** one, which uses p^ instead of Q^ as the renormalization scale in Ug, similarly to what 
is done in the shower evolution, but this was not yet available at the time our LO* tune was 
performed. The main reason for sticking to CTEQ5L for Perugia was the desire that this tune 
can be run with standalone Pythia 6. We note that in Pythia 8, several more recent sets have 
already been implemented in the standalone version [83], hence removing this restriction from 
corresponding tuning efforts for Pythia 8. Note also that, since these sets are implemented 
internally in Pythia 8, the bug in the LHAPDF interface mentioned above does not affect 
Pythia fl 

Finally, the HARD and SOFT variations shown by the yellow band in the right pane of 
fig. [1] are obtained by making a variation of roughly a factor of 2 in either direction from the 
central tune (in the case of the SOFT tune, this is obtained by a combination of reverting 
to the MS value for A and using ^/2p^ as the renormalisation scale). In the low-p_|_ peak, 
the HARD variation generates a slightly too broad distribution, but given the large sensitivity 
of this peak to subleading corrections (see below), we consider this to be consistent with the 
expected theoretical precision. The p± spectrum of the other Perugia tunes will be covered in 
the section on the individual tunes below. 

Phase Space: A further point concerning ISR that deserves discussion is the phase space over 
which ISR emissions are allowed. Here, Drell-Yan is a special case, since this process is matched 
to Z+jet matrix eleinents in Pythia |84y85]. and hence the hardest jet is always described by 
the matrix element over all of phase space. For unmatched processes which do not contain 
jets at leading order, the fact that we start the parton shower off from the factorisation scale 
can, however, produce an illusion of almost zero jet activity above that scale. This was studied 
in [86l[87] , where also the consequences of dropping the phase space cutoff at the factorisation 
scale were investigated, so-called power showers. Our current best understanding is that the 
conventional ( "wimpy" ) showers with a cutoff at the factorisation scale certainly underestimate 
the tail of ultra-hard emissions while the power showers are likely to overestimate it, hence 
making the difference between the two a useful measure of uncertainty. Since other event 



*Note therefore that one has to be careful when hnking LHAPDF. If an internally implemented PDF set is 
replaced by its LHAPDF equivalent, there is unfortunately at present no guarantee that identical results will be 
obtained. We therefore strongly advise MC tunes to specify exactly which implementation was used to perform 
the tune, and users to regard the implementation as part of the tune. We hope this unfortunate situation may 
be rectified in the future. 



12 



5 -1 
-S10 



10 7 



10 



10 



1960GeVp+pbar 


Top Pair Production 


Transverse Momentum of tt Pair 




~ 


- 4- Pro-Q20 : 


w. 


-•- SOA 




- 0- SOA-Wimpy 




— e— Perugia 




Perugia HARD 


"1^ 


Perugia SOFT _ 


'*.^^ 
*^^- 

*^^^ 


: 


X \%^K, 


- 


'ft\^SC~, 


. 


^■■■. ^^--i 




~ * ^^ 


^ 


\ ■■■■>. 


^^^< 


' 


XXN^ 


P/thia 6.423 


'"■/■'^^ 



7 TeV p+p 



50 



100 150 

p^(hard system) [GeV] 



Transverse Momentum of tt Pair 



Top Pair Production 



A- Pro-Q20 
*— SOA 
- SOA-Wimpy 
«— Perugia 
--- Perugia HARD 
Perugia SOFT 




200 300 400 

p^(hard system) [GeV] 



Figure 2: Comparison of new and old tunes for tlie p± of tt pairs at tlie Tevatron {left) and at tlie LHC 
at 7 TeV (right). See [43] for otlier tunes and collider energies. 



generators usually provide wimpy showers by default, we have chosen to give the power variants 
as the default in Pythia 6 — not because the power shower approximation is necessarily better, 
but simply to minimise the risk that an accidental agreement between two generators is taken 
as a sign of a small overall uncertainty, and also to give a conservative estimate of the amount 
of hard additional jets that can be expected. Note that a more systematic description of 
hard radiation that interpolates between the power and wimpy behaviours has recently been 
implemented in Pythia 8 [88] . 

For the Perugia models, we have implemented a simpler possibility to smoothly dampen the 
tail of ultra-hard radiation, using a scale determined from the colour flow as reference. This is 
done by nominally applying a power shower, but dampening it by a factor 

SD 



accept 



B 



67 



4pievol 



(4) 



where Pqj corresponds to the parameter PARP(67) in the code, p^evoi i^ ^^^ evolution scale 
for the trial splitting, and sd is the invariant mass of the radiating parton with its colour 
neighbour, with all momenta crossed into the final state (i.e., it is s for annihilation- type colour 
flows and —t for an initial- final connection). This is motivated partly by the desire to give an 
intermediate possibility between the pure power and pure wimpy options but also partly from 
findings that similar factors can substantially improve the agreement with final-state matrix 
elements in the context of the ViNCiA shower [32] • By default, the Perugia tunes use a value 
of 1 for this parameter, with the SOFT and HARD tunes exploring systematic variations, see 
table [3 

At the Tevatron, the question of power vs. wimpy showers is actually not much of an issue, 
since H/V+jets is already matched to matrix elements in default Pythia and most other 



13 



interesting processes either contain QCD jets already at leading order (7+jets, dijets, WBF) or 
have very little phase space for radiation above the factorisation scale anyway {tt, dibosons). 
This is illustrated by the curves labeled SOA (solid blue) and SOA-Winipy (dash-dotted cyan) 
in the left pane of fig. [21 which shows the p± spectrum of the tt system (equivalent to the 
Drell-Yan p± shown earlier). The two curves do begin to diverge around the top mass scale, 
but in light of the limited statistics available at the Tevatron, matching to higher-order matrix 
elements to control this ambiguity does not appear to be of crucial importance. In contrast, 
when we extrapolate to pp collisions at 7 TeV, shown in the right pane of fig. [2l the increased 
phase space makes the ambiguity larger. Matching to the proper matrix elements describing 
the region of jet emissions above pj_ ~ mt may therefore be correspondingly more important, 
see, e.g., |89j . Note that the extremal Perugia variations span most of the full power /wimpy 
difference, as desired, while the central ones fall inbetween. Note also that this only concerns 
the p± spectrum of the hard jets — power showers cannot in general be expected to properly 
capture jet-jet correlations, which are partly generated by polarisation effects not accounted for 
in this treatment. 

Primordial kx: Finally, it is worth remarking that the peak region of the Drell-Yan p± 
spectrum is extremely sensitive to infrared effects. On the experimental side, this means, e.g., 
that the treatment of QED corrections can have significant effects and that care must be taken 
to deal with them in a consistent and model-independent manner [53]. On the theoretical 
side, relevant infrared effects include whether the low-pj_ divergences in the parton shower are 
regulated by a sharp cutoff or by a smooth suppression (and in what variable), how Og is treated 
close to the cutoff, and how much "Fermi motion" is given to each of the shower-initiating 
partons extracted from the protons. A full exploration of these effects probably goes beyond 
what can meaningfully be studied at the current level of precision. Our models therefore only 
contain one infrared parameter (in addition to the infrared regularisation scale of the shower), 
called "primordial kj-" , which should be perceived of as lumping together an inclusive sum 
of unresolved effects below the shower cutoff. Since the cutoff is typically in the range 1-2 
GeV, we do not expect the primordial kx to be much larger than this number, but there is 
also no fundamental reason to believe it should be significantly smaller. This is in contrast 
to previous lines of thought, which drew a much closer connection between this parameter 
and Fermi motion, which is expected to be only a few hundred MeV. In Tune A, the value of 
primordial k^, corresponding to PARP(91) in the code, was originally 1 GeV, whereas it was 
increased to 2.1 GeV in Tune DW. In the Perugia tunes, it varies in the same range, cf. table 
[3l Its distribution is assumed to be Gaussian in all the models. Explicit attempts exploring 
alternative distributions in connection with the writeup of this paper {1/kj, tails and even a 
flat distribution with a cutoff, see [Ml MSTP(91)]) did not lead to significant differences. 

3.3 Underlying Event, Beam Remnants, and Colour Reconnections (Table 

Charged Multiplicity The charged particle multiplicity (A^ch) distributions for minimum- 
bias events at 1800 and 1960 GeV at the Tevatron are shown in fig. [3l Particles with ct > 10 
mm (/i^, TT^, Kg, K1, n^. A", S^, H'', H^, and O^) are treated as stable. Models include the 



14 



10 



Q_ -2 

10 



10 



1 800 GeV p+pbar inelastic, Non-Diffractive 



- Charged Particle Multiplicity (|ri|<1.0, p^>0.4GeV) 

+ CDF data 
T-- DW(T) 
«— Perugia 

Perugia HARD ^ 
Perugia SOFT ; 



10 



10 



10 



10 




2 4 6 8 10 

Pythia 6.423 
Data from CDF Collaboration, PRD65(2002)072005 



10 



20 



30 40 50 

N^J|Ti|<1,0, p^>0,4GeV) 



— _ 1 



a: 10 



10 



10 



10 



1960 GeV p+pbar inelastic, Non-Dlffractlve 



Charged Particle Multiplicity (|t)|<1.0, p^>0.4GeV) 

+ CDF data 
DW 

Perugia 

Perugia HARD = 

iaSOFT : 




10 



20 



30 40 50 

N^Jlill<1.0,p^>0.4GeV) 



Figure 3: Comparisons to tlie CDF measurements of tlie cliarged track multiplicity at 1800 (left) and 
1960 GcV (right) in inininrum-bias pp collisions. See [43] for other tunes and collider energies. 

inelastic, non-diffractive component only. Note that the Perugia tunes included this data in the 
tuning, while DW was only tuned to underlying-event data at the same energies. The overall 
agreement over the many orders of magnitude spanned by these measurements is quite good. 
On the large-multiplicity tails, DW appears to give a slightly too narrow distribution. In the 
low- multiplicity peak (see insets), the Perugia tunes fit the 1800 GeV data set better while DW 
fits the 1960 GeV data set better. As mentioned above, however, diffractive topologies give 
large corrections in this region, and so the points shown in the insets were not used to constrain 
the Perugia tunes. 

Transverse Momentum Spectrum The p± spectrum of charged particles at 1960 GeV 
is shown in fig. HI Note that both plots in the figure show the same data; only the model 
comparisons are different. 

The plot in the left-hand pane illustrates a qualitative difference between the Q^- and p±- 
ordered models. Comparing DW to NOCR (a tune of the p^-ordered model which does not 
employ colour reconnections) we see that the p± spectrum is generically slightly harder in the 
new model than in the old one. Colour reconnections, introduced in SOA, then act to harden 
this spectrum slightly more, to the point of marginal disagreement with the data. Finally, 
when we include the Professor tunes to LEP data, nothing much happens to this spectrum in 
the old model — compare DW with DW-Pro — whereas the spectrum becomes yet harder in 
the new one, cf. SOA-Pro, now reaching a level of disagreement with the data that we have 
to take seriously. Since the original spectrum out of the box — represented by NOCR — was 
originally quite similar to that of DW and DW-Pro, our tentative conclusion is that either the 
revised LEP parameters for the p^-ordered shower have some hidden problem and/or the colour 
reconnection model is hardening the spectrum too much. For the Perugia tunes, we took the 



15 



10 r 



10 r 



10 r 



10 r 



1960GeVp+pbar 


Inelastic, Non-Diffractive 


1 Charged Particle p Spectrum 


|T)|<1.0, p^>0.4GeV) 




+ CDF data 


=1 


--T-- DW 


ri 


--T-- DW-Pro 


: I 


- ■- NOCR 


r » 


-•- SOA 


» 


— •— SOA-Pro 




3^ 


: Pyth 

Data from CDF Collaboratio 


. Phys. Rev. 079(2009)1 1 2005 -i- 

Ill, ,>,-;>t;lj|r^r^^ 



10 



20 



30 



40 50 
Pi [GeV] 





1 960 GeV p+pbar inelastic, Non-Diffractive 




1 Charged Particle p^ Spectrum (|ti|<1 .0, pj^>0.4GeV) 


= 


1 


1 ^ CDF data 


^ 




:1 --T-- DW 


- 




ri — ©— Perugia 


1 


■2 
10 


I m Perugia HARD - 

r % Perugia SOFT 1 


-4 
10 


r\ 


1 


■6 
10 


^Vi' 


1 


■8 
10 


E Pythia~6.423 ,»JHi_"^"t" -i--^ 
Data from CDF Collaboration, Phys. Rev. 079(2009^11 2005 

r 1 1 Ii , , , ,?ri^^I^>^■, 1 


^ 



10 



20 



30 



40 50 
Pi [GeV] 



Figure 4: Comparisons to tlic CDF mcasurenicnt of tlie ctiargcd particle p± spectrum in minimum-bias 
pp collisions at 1960 GeV for two sets of models. See [13] for other tunes and collider energies. 



latter interpretation, since we did not wish to alter the LEP tuning. Using a modified colour- 
reconnection model that suppresses reconnections among high-p_L string pieces (to be described 
below), the plot in the right-hand pane illustrates that an acceptable level of agreement with the 
data has been restored in the Perugia tunes, without modifying the Professor LEP parameters. 
For completeness we should also note that there are indications of a significant discrepancy 
developing in the extreme tail of particles with p± > 30 GeV, where all the models fall below the 
data, a trend that was confirmed with higher statistics in [90]. This discrepancy also appears 
in the context of NLO calculations folded with fragmentation functions |91j . so is not a feature 
unique to the Pythia modeling. Though we shall not comment on possible causes for this 
behaviour here (see |92tl93| for a critical assessment), the extreme tail of the pj_ distribution 
should therefore be especially interesting to check when high-statistics data from the LHC 
become available. 



{p±) (-^ch) and Colour Reconnections While the multiplicity and p± spectra are thus, 
separately, well described by Tune DW, it does less well on their correlation, {p±) (Net), as 
illustrated by the plot in the left-hand pane of fig. [5l Since the SO family of tunes were initially 
tuned to Tune A, in the absence of published data, the slightly smaller discrepancy exhibited 
by Tune A carried over to the SO set of tunes, as illustrated by the same plot. Fortunately, 
CDF Run-2 data has now been made publicly available [57], corrected to the particle level, and 
hence it was possible to take the actual data into consideration for the Perugia tunes, resulting 
in somewhat softer particle spectra in high-multiplicity events, cf. the right-hand pane in fig. [5j 

What is more interesting is how this correlation is achieved by the models. Also shown 



16 



^1.6 



1.4 



1.2 



0.8 



1960GeVp+pbar 



Inelastic, Non-Diffractive 



- Average Charged Particle p.|-(|ii|<1.0, p^>0.4GeV) 
+ CDF data 
A 
--T-- DW 
— •— SOA 




Pythia 6.423 
Data from CDF Collaboration, Ptiys. Rev. 079(2009)112005 



10 



20 



30 40 50 

N,,(|Ti|<1,0,p^>0,4GeV) 



„1,. 



1960 GeV p+pbar inelastic, Non-Diffractive 



1.4 - 



- Average Charged Particle Pj (|ri|<1 .0, p^>0.4GeV) 
4 CDF data 
--T-- DW 
-■«■- Atlas-DC2 
— e— Perugia 

Perugia HARD 

.. 2 I Perugia SOFT ^ -' 

- Perugia NOCR ,-'', 




30 40 50 

N,,(|ri|<1.0,p^>0.4GeV) 



Figure 5: Comparisons to the CDF Run II measurement of tlic average track p± as a function of track 
multiplicity in min-bias pp collisions. Left: The older generation of tunes. Right: the Perugia variations 
compared to two older tunes. See [43| for other tunes and collider energies. 



in the right-hand pane of fig. [5] are comparisons to an older ATLAS tune which did not use 
the enhanced final-state colour connections that Tunes A and DW employ. A special Perugia 
variation without colour reconnections, Perugia NOCR, is also shown, and one sees that both 
this and the ATLAS tune predict too little correlation between {p±) and A'ch- 

This distribution therefore appears to be sensitive to the colour structure of the events, 
at least within the framework of the Pythia modeling |33H35t [M]. The Perugia tunes all 
(with the exception of NOCR) rely on an infrared toy model of string interactions [33] to drive 
the increase of {p±) with A'^ch- The motivation for a model of this type comes from arguing 
that, in the leading-colour limit used by Monte Carlo event generators, and in the limit of 
many perturbative parton-parton interactions, the central rapidity region in hadron-hadron 
collisions would be criss-crossed by a very large number of QCD strings; naively one string 
per perturbative i-channel quark exchange, and two per gluon exchange. However, since the 
actual number of colours is only three, and since the strings would have to be rather closely 
packed in spacetime, it is not unreasonable to suppose either that the colour field collapses in a 
more economical configuration already from the start, or that the strings undergo interactions 
among themselves, before the fragmentation process is complete, that tend to minimise their 
total potential energy, as given by the area law of classical strings. The toy models used 
by both the SO and Perugia tunes do not address the detailed dynamics of this process, but 
instead employ an annealing-like minimisation of the total potential energy, where the string- 
string interaction strength was originally the only variable parameter [33]. While this gave a 
reasonable agreement with {p±) (Nch), it still tended to give slightly too hard a tail on the 



17 



single-particle p± distribution, as compared to the Tevatron Run 2 measurement. Therefore, 
a suppression of reconnections among very high-p_L string pieces was introduced, reasoning 
that very fast-moving string systems should be able to more easily "escape" the mayhem in 
the central region. (Similarly, one could argue that string systems produced in the decay of 
massive particles with finite lifetimes, such as narrow BSM or Higgs resonances, or even possibly 
hadronic t oi W decays, should be able to escape more easily. We have not so far built in such 
a suppression, however.) 

The switch MSTP(95) controls the choice of colour-reconnection model. In the "SO" model 
corresponding to MSTP(95)=6 (and =7 to apply it also in lepton collisions), the total probability 
for a string piece to survive the annealing and preserve its original colour connections is 

MSTP(95) = 6, 7 : Pkccp = (1 " C^Ts)"'"', (5) 

where Pyg corresponds to the parameter PAR? (78) in the code and sets the overall colour- 
reconnection strength and njnt is the number of parton-parton interactions in the current event, 
giving a rough first estimate of the number of strings spanned between the remnants. (It is 
thus more likely for a string piece to suffer "colour amnesia" in a busy event, than in a quiet 
one.) C '^3'S introduced together with the Perugia tunes and gives a possibility to suppress 
reconnections among high-p_|_ string pieces, 

with P77 corresponding to PARP(77) in the code and {p±) being a measure of the average 
transverse momentum per pion that the string piece would produce, n.,^ oc ln(s/m^), with a 
normalisation factor absorbed into -P77. 

Starting from Pythia 6.4.23, a slightly more sophisticated version of colour annealing was 
introduced, via MSTP(95)=8 (and =9 to apply it also in lepton collisions), as follows. Instead 
of using the number of multiple parton-parton interactions to give an average idea of the total 
number of strings between the remnants, the algorithm instead starts by finding a thrust axis 
for the event (which normally will coincide with the z axis for hadron-hadron collisions). It then 
computes the density of string pieces along that axis, rapidity-interval by rapidity-interval, with 
a relatively fine binning in rapidity. Finally, it calculates the reconnection probability for each 
individual string piece by using the average string density in the region spanned by that string 
piece, instead of the number of multiple interactions, in the exponent in the above equation: 

MSTP(95) = 8, 9 : P = (1 - C-Pts)^"'^^^''^'^ (7) 

where (n^) (2/1,2/2) is the average number of other string pieces, not counting the piece under 
consideration, in the rapidity range spanned by the two endpoints of the piece, yi and y2- 
Obviously, the resulting model is still relatively crude — it still has no explicit space-time picture 
and hence will not generate more subtle effects such as (elliptical) flow, no detailed dynamics 
model, and no suppression mechanism for reconnections involving long-lived resonances — but 
at least the reconnection probability has been made a more local function of the actual string 
environment, which also provides a qualitative variation on the previous models that can be 
used to explore uncertainties. In the code, the "SO" type is also referred to as the "Seattle" 
model, since it was written while on a visit there. The newer one is referred to as the "Paquis" 
type, for similar reasons. 



18 



ifeMJcRfiiiiaa 



TOWARDS <N^„> density (1ti|<1 .□, p^>0.5GeV) vs pJZ) 
4 CDF data 
--'-- DW 
- ^ - Pra-Q20 
— a — Perugia 
-a-- Perugia2Q10 




80 100 

pJZ) IGeV] 



80 100 

pJZ) IGeV] 



Figure 6: Comparisons to tlic CDF measurements [721173] of tlie cliarged particle multiplicity (top row) 
and p± (bottom row) densities in the "TOWARDS" (left), "TRANSVERSE" (middle), and "AWAY" 
(right) regions of Drell-Yan production at 1960 GeV, as a function of the Drell-Yan p±. 



Underlying Event In fig. [6l we show the {Net) densit3o (top row) and the {p±Sum) densitjo 
(bottom row) in each of the TOWARDS, TRANSVERSE, and AWAY regions, for Drell-Yan 
production at the Tevatron, compared to CDF data |721 I73]. The invariant mass window for 
the lepton pair for this measurement is 70 < m^+i- < 110, in GeV. Tracks with pT > 0.5 GeV 
inside jryl < 1 were included, with the same definition of stable charged tracks as above. The 
leptons from the decaying boson were not included. 

The agreement between the Perugia min-bias tunes and data is at the same level as that of 
more dedicated UE tunes, here represented by DW and Pro-Q2, supporting the assertion made 
earlier concerning the good universality properties of the Pythia modeling. We note also that 
the Perugia 2010 variation agrees slightly better with the data in the TRANSVSERSE region, 
where it has a bit more activity than Perugia does. 



^The (Nch) density is defined as the average number of tracks per unit AriA(j} in the relevant region. 
The {p±suui) density is defined as the average scalar sum of track p± per unit ArjAcj}. 



19 



J2 

a 



10 



10 



10 





1 960 GeV p+pbar inelastic, Non-Diffractive 




Probability Distribution of the Number of Parton Interactions ] 












--f-- DW 




*\~-, 


— e— Perugia 
Perugia HARD 




'i 
















-1 


'^V?s»- 


^ 


— 


— 


- 





Perugia SOFT 




1 II', '*N.. 1 1 












\\ 


















\ 


, 






-2 










V'. ■^ 










\ ■ ^> i 






\ ■- J 




1 




1 


















V 


y 












1 '. \ "■ 1 












Pythia fi 4?.T > k 




-3 


1 














V 



10 



Figure 7: Double-logarithmic plot of the probability distribution of the number of parton-parton interac- 
tions in min-bias collisions at the Tevatron, showing that the Perugia tunes obtain the same multiplicity 
distribution, fig. [3l with fewer MPI than Tune A. See [43] for other tunes and collider energies. 



Transverse Mass Distribution and MPI Sho'wers Finally, the old framework did not 
include showering off the MPI in- and out-state^II- The new framework does include such 
showers, which furnish an additional fluctuating physics component. Relatively speaking, the 
new framework therefore needs less fluctuations from other sources in order to describe the 
same data. This is reflected in the tunes of the new framework generally having a less lumpy 
proton (smoother proton transverse density distributions) and fewer total numbers of MPI than 
the old one. This is illustrated in fig. [TJ where a double-logarithmic scale has been chosen in 
order to reveal the asymptotic behaviour more clearly. Note that, e.g., for Tune A, the plot 
shows that more than a per mille of min-bias events have over 30 perturbative parton-parton 
interactions per event at the Tevatron. This number is reduced by a factor of 2 to 3 in the new 
models, while the average number of interactions, indicated on the r.h.s. of the plot, goes down 
by slightly less. 

The showers off the MPI also lead to a greater degree of decorrelation and p± imbalance 
between the minijets produced by the underlying event, in contrast to the old framework where 
these remained almost exactly balanced and back-to-back. This should show up in minijet Acpjj 
and/or ARjj distributions sensitive to the underlying event, such as in Z/VF-j-multijets with 
low p± cuts on the additional jets. It should also show up as a relative enhancement in the odd 
components of Fourier transforms of (p distributions a la [95]. 



^It did, of course, include showers off the primary interaction. An option to include FSR off the MPI also 
in that framework has since been implemented by S. Mrenna, see |41) . but tunes using that option have not yet 
been made. 



20 





1 960 GeV p+pbar inelastic, Non-Diffractive 




- Nj^ FB Correlation Strength (p^>0.4GeV) 




--T-- DW 


0.8 


— e— Perugia 
Perugia HARD 




Perugia SOFT 


0.6 


- 


0.4 


:^ - 


0.2 


- ^"'*'^---.. ■ 




^^~^^«^<. ''' 




Pythla 6.423 ^^^il-^ 




1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 



0.8 



0.6 



0.4 



0.2 



1 960 GeV p+pbar inelastic, Non-Dlftractlve 



Ctiarged+Neutral pTSum FB Correlation Strength (generator-level) 
--T-- DW 
— e— Perugia 

Perugia HARD 

Perugia SOFT 



0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 

lip 



0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 

lip 



Figure 8: Forward-Backward correlation strengtlis at tire Tevatron in (left) charged particles and (right) 
charged plus neutral transverse-momentum sum at generator level. See |43| for other tunes and collider 
energies. 



Long- Range Correlations Further, since showers tend to produce shorter-range correlations 
than MPI, the new tunes also exhibit smaller long-range correlations than did the old models. 
That is, if there is a large fluctuation in one end of the detector, it is less likely in the new models 
that there is a large fluctuation in the same direction in the other end of the detector. The 
impact of this on the overall modeling, and on correction procedures derived from it, has not 
yet been studied in great detail. One variable which can give direct experimental information 
on the correlation strength over both short and long distances is the so-called forward-backward 
correlation, 6, defined as in [MIIMI 



Kvf) 



{npriB) - {np) 
(nl) - {npf 



(8) 



where np and ub are the number of tracks (or a calorimetric measure of energy deposition) in a 
pseudorapidity bin centered at -qp and tjb = —ilp, respectively, for a given event. The averages 
indicate averaging over the number of recorded events. The resulting correlation strength, b, 
can be plotted either as a function of rjp or as a function of the distance, Arj, between the 
bins. A comparison of the main Perugia tunes to Tune DW is shown in fig. [HI for two different 
variants of the correlation strength: the plot on the left only includes charged particles with 
p± > 0.4 GeV and the other (right) includes all energy depositions (charged plus neutral) 
that would be recorded by an idealized calorimeter. Since estimating the impact on the latter 
of a real (noisy) calorimeter environment would go beyond the scope of this paper, we here 
present the correlation at generator level. For the former, we show the behaviour out to ry = 5 
although the CDF and D0 detectors would of course be limited to measuring it inside the region 



21 



|?7| < 1.0. Note that a measurement of this variable would also be a prerequisite for combining 
the dN/drj measurements from negative and positive 77 regions to form dN/d\ri\, with the proper 
correlations taken into account. This particular application of the h measurement would require 
a measurement of h with the same bin sizes as used for dN/drj. Since the amount of correlation 
depends on the bin size used (smaller bin sizes are more sensitive to uncorrelated fluctuations) , 
we would advise to perform the h measurement using several different bin sizes, ranging from 
a very fine binning (e.g., parallelling that of the dN/drj measurement), to very wide bins (e.g., 
one unit in pseudorapidity as used in jHl]). For our plots here, we used an intermediate-sized 
binning of 0.5 units in pseudorapidity. 

3.4 Energy Scaling (Table [4]) 

A final difference with respect to the older SO(A) family of tunes is that we here include data 
from different colliders at different energies, in an attempt to fix the energy scaling better. 

The energy scaling of min-bias and underlying-event phenomena, in both the old and new 
Pythia models, is driven largely by a single parameter, the scaling power of the infrared reg- 
ularisation scale for the multiple parton interactions, p_lo, see, e.g., |131I141 [M]. This parameter 
is assumed to scale with the collider CM energy squared, s, in the following way, 

/ g \ -Pgo 
Pio(s) =Pio(srcf) , (9) 

V -Srcf / 

where Pj_o(sref) is the IR regularisation scale given at a specific reference s = Sref , and P90 sets 
the scaling away from s = Srcf. In the code, p^o('^rcf) is represented by PARP(82), \/ij,gf by 
PARP (89) , and P90 by PARP (90) . Note that large values of P90 produce a slower rate of increase 
in the overall activity with collider energy than low values, since the generation of additional 
parton-parton interactions in the underlying event is suppressed below p±_o. 

The default value for the scaling power in Pythia 6.2 was i-90 = 0.16, motivated [M] by 
relating it to the scaling of the total cross section, which grows like oc E^^ . When comparing 
to Tevatron data at 630 GeV, Rick Field found that this resulted in too little activity at that 
energy, as illustrated in the top row of fig. [H where tune DWT uses the old default scaling away 
from the Tevatron and DW uses Rick Field's value of P90 = 0.25. (The total cross section is 
still obtained from a Donnachie-Landshoff fit [97] and is not affected by this change.) Note that 
the lowest-multiplicity bins of the UA5 data in particular and the first bin of the CDF data 
were ignored for our comparisons here, since these contain a large diffractive component, which 
has not been simulated in the model comparisons. 

For the Perugia tunes, the main variations of which are shown in the bottom row of fig. O 
we find that a large range of values, between 0.22 and 0.32, can be accommodated without 
ruining the agreement with the available data, with Perugia using 0.26. 

The energy scaling is therefore still a matter of large uncertainty, and the possibility of 
getting good additional constraints from the early LHC data is encouraging. The message so 
far appears to be contradictory, however, with early ATLAS results at 900 GeV [63] appearing 
to confirm the tendency of the current tunes to undershoot the high-multiplicity tail at 900 
GeV (see the right-hand column of fig. [9]) , which would indicate a slower scaling between 900 
and 1800 GeV than what is generated by the models (since they all fit well at 1800 GeV) but 



22 



HililiU'IP^ 1 11^ 


HfflPI 


nsmsnsi 


iffsnRi 


: Charged Particle Multiplicity (|ti|<1 .5, all p^ 










i UA5 data 


(NSD) 


^, 




-T- DW 
-T- DWT 




* % 










''^ 






Pytti 
Data from UA5 Coliabo 


..;-\t 

atlOii.ZPhys 43(1989)357 





630 GeV p+pbar 



30 40 50 

N^h(|Tll<1.5, allpj 



: Charged Particle Multiplicity (|tiI<1-0, p_>0.4GeV] 
-1 ■.■"■! 4 CDF data 



V 



■■f. 



t \ 



Pylliia 6.423 
Data from CDF CollHboratlor. PRDE5|2002) 072005 ' 



siiiiKfssflsqsn;^ 




SB 


RmsinsjinfTsiiffil 


■ Charged Particle Multiplicity (|ti|<1. 5, 


allPx 










+ UA5 data (NSD) 








-,-- DW 


j*'^. 






-T-- DWT 


rK 








' X 








, ^:^ 


'"t 'i%i\ 










'''t 


1 I 


- 


' 


\ 


II 




Pythia S.423 


t' 




Data from UA5 Cotiaboration, Z 


hya 43(1989)357 



20 30 40 

Nji, (|ri|<1.0, p^>0.4GeV) 



60 

NjJ|ill<1.5, allpj 



200 GeV p+pbar 



, IAI.IcR»|gBI!EB 



900 GeV p+pbar 



Charged Particle Multiplicity (|ti|<1.5, all pj 



Charged Particle Hfluitiplicity (|ti|<1.0, p_>0.4GeV) 

+ CDF data 



+ UA5 data (NSD) 




Charged Particle Multiplicity (|ii|<1.5, all pj 



30 40 50 

N„j(!il|<1.5, allpj 



20 30 40 

N^j(|il|<1.0, p^>0.4GeV) 



60 

N„j(|ill<1.5, allpj 



Figure 9: Comparisons to UA5 and CDF measurements of the charged track multiplicity in minimum- 
bias pp colhsions at 200 GeV (left), 630 GeV (middle), and 900 GeV (right). Top: Rick Field's tunes 
DW and DWT. Bottom: the main Perugia variations compared to DW. See [43] for other tunes and 
collider energies. 



preliminary CMS results on the average multiplicity at 2360 GeV [62] indicate the opposite, 
that the pace of evolution in the models is actually too slow. Furthermore, the CDF data at 
630 GeV and the UA5 data at 200 GeV provide additional constraints at lower energies which 
have made it difficult for us to increase the tail at 900 GeV without coming into conflict with 
at least one of these other data sets. In view of these tensions, we strongly recommend future 
studies to include comparisons at different energies. 

One issue that can be clearly separated out in this discussion, however, is that the average 
multiplicity is sensitive to "contamination" from events of diffractive origin, while the high- 
multiplicity tail is not, and hence a different scaling behavior with energy (or just a different 
relative fraction?) of diffractive vs. non-diffractive events may well generate differences between 
the scaling behaviour of each individual moment of the multiplicity distribution. Attempting 
to pin down the scaling behavior moment by moment would therefore also be an interesting 
possible study. Since the Pythia 6 modeling of diffraction is relatively crude, however, we did 
not attempt to pursue this question further in the present study, but note that a discussion 



23 



of whether these tendencies could be given other meaningful physical interpretations, e.g., in 
terms of low-x, saturation, and/or unitarisation effects, would be interesting to follow up on. 

It should be safe to conclude, however, that there is clearly a need for more systematic exam- 
inations of the energy scaling behavior, both theoretically and experimentally, for both diffrac- 
tive and non-diffractively enhanced event topologies separately. It would also be interesting, for 
instance, to attempt to separately determine the scaling behaviours for low-activity/peripheral 
events and for active/central events, e.g., by considering the scaling of the various moments of 
the multiplicity distribution and by other observables weighted by powers of the event multi- 
plicity. 

4 The Perugia Tunes: Tune by Tune 

The starting point for all the Perugia tunes, apart from Perugia NOCR, was SOA-Pro, i.e., the 
original tune "SO" [T2l[13l[33l[3l] , with the Tune A energy scaling (SOA), revamped to include 
the Professor tuning of flavor and fragmentation parameters to LEP data |29|l44j (SOA-Pro). 
The starting point for Perugia NOCR was NOCR-Pro. From these starting points, the main 
hadron collider parameters were retuned to better describe the data sets described above. 

As in previous versions, each tune is associated with a 3-digit number which can be given 
in MSTP(5) as a convenient shortcut. A complete overview of the Perugia tune parameters is 
given in appendix |A] and a list of all the predefined tunes that are included with Pythia version 
6.423 can be found in appendix [Cl 

Perugia (320): Uses CTEQ5L parton distributions [98] (the default in Pythia and the 
most recent set available in the standalone version — see below for Perugia variations using 
external CTEQ6L1 and MRST LO* distributions). Uses Acmw IZS] instead of Aj^, which 
results in near-perfect agreement with the Drell-Yan p± spectrum, both in the tail and in the 
peak, cf. fig. [TJ Also has slightly less colour reconnections than SO(A), especially among high- 
p± string pieces, which improves the agreement both with the {p±) (Nch) distribution and with 
the high-p^ tail of charged particle p± spectra, cf \A3\ dN/dpT (tail)]). Slightly more beam- 
remnant breakup than SO(A) (more baryon number transport), mostly in order to explore this 
possibility than due to any necessity of tuning at this point. Without further changes, these 
modifications would lead to a greatly increased average multiplicity as well as larger multiplicity 
fluctuations. To keep the total multiplicity unchanged, relative to SOA-Pro, the changes above 
were accompanied by an increase in the MPI infrared cutoff, p±o, which decreases the overall 
MPI-associated activity, and by a slightly smoother proton mass profile, which decreases the 
fluctuations. Finally, the energy scaling is closer to that of Tune A (and SOA) than to the old 
default scaling that was used for SO. 

Perugia HARD (321): A variant of Perugia which has a higher amount of activity from 
perturbative physics and counter-balances that partly by having less particle production from 
nonperturbative sources. Thus, the Acmw value is used for ISR, together with a renormalisation 
scale for ISR of fiR = ^p±, yielding a comparatively hard Drell-Yan pj_ spectrum, cf. the dashed 
curve labeled "HARD" in the right pane of fig. [H It also has a slightly larger phase space 



24 



for both ISR and FSR, uses higher-than-nominal values for FSR, and has a shghtly harder 
hadronisation. To partly counter-balance these choices, it has less "primordial kx" , a higher 
IR cutoff for the MPI, and more active colour reconnections, yielding a comparatively high 
curve for {p±_) (A'^ch)) cf. fig. [5l Warning: this tune has more ISR but also more FSR. The final 
number of reconstructed jets may therefore not appear to change very much, and if the number 
of ISR jets is held fixed (e.g., by matching), this tune may even produce fewer events, due to 
the increased broadening. For a full ISR/FSR systematics study, the amount of ISR and FSR 
should be changed independently. 



Perugia SOFT (322): A variant of Perugia which has a lower amount of activity from 
perturbative physics and makes up for it partly by adding more particle production from non- 
perturbative sources. Thus, the Aj^ value is used for ISR, together with a renormalisation 
scale of //ij = y/2p±, yielding a comparatively soft Drell-Yan p± spectrum, cf. the dotted curve 
labeled "SOFT" in the right pane of fig. [T] It also has a slightly smaller phase space for both 
ISR and FSR, uses lower-than-nominal values for FSR, and has a slightly softer hadronisation. 
To partly counter-balance these choices, it has a more sharply peaked proton mass distribution, 
a more active beam remnant fragmentation, a slightly lower IR cutoff for the MPI, and slightly 
less active colour reconnections, yielding a comparatively low curve for {p±) (Nch), cf. fig. O 
Again, a more complete variation would be to vary the amount of ISR and FSR independently, 
at the price of introducing two more variations (see above). We encourage users that desire a 
complete ISR/FSR systematics study to make these additional variations on their own. 



Perugia 3 (323): A variant of Perugia which has a different balance between MPI and 
ISR and a different energy scaling. Instead of a smooth dampening of ISR all the way to zero 
p±, this tune uses a sharp cutoff at 1.25 GeV, which produces a slightly harder ISR spectrum. 
The additional ISR activity is counter-balanced by a higher infrared MPI cutoff. Since the ISR 
cutoff is independent of the collider CM energy in this tune, the multiplicity would nominally 
evolve very rapidly with energy. To offset this, the MPI cutoff itself must scale very quickly, 
hence this tune has a very large value of the scaling power of that cutoff. This leads to an 
interesting systematic difference in the scaling behavior, with ISR becoming an increasingly 
more important source of particle production as the energy increases in this tune, relative to 
Perugia 0. This is illustrated in fig. [TOl where we show the scaling of the min-bias charged 
multiplicity distribution and the Drell-Yan p± spectrum between the Tevatron (left) and the 
LHC at 14 TeV (right). One sees that, while the overall multiplicity grows less fast with energy 
in Perugia 3, the position of the soft peak in Drell-Yan becomes harder, reflecting the relative 
increase in ISR, despite the decrease in MPI. 

Perugia NOCR (324): An update of NOCR-Pro that attempts to fit the data sets as 
well as possible, without invoking any explicit colour reconnections. Can reach an acceptable 
agreement with most distributions, except for the {p±) (A^ch) one, cf. fig. El Since there is a 
large amount of "colour disturbance" in the remnant, this tune gives rise to a very large amount 
of baryon number transport, even greater than for the SOFT variant above. 



25 



S 10 

n 
o 
Dl -2 

10 

-3 



Charged Panicle Multiplicity (Inkl-O. p_>0.4GeV) 

^ CDF data 

— e— Perugia 

■ - Perugia 3 

--*-- Perugia LO' _ 

— • — Perugia 6 




* 10 



30 40 

N„J|il|<1.0, p^>0.4GeV) 



1800 GeVp+pbar 



Transverse Momentum of Z// [66 < M2^,y/GeV < 1 16| 

f CDF data 

— s — Perugia 

• - Perugia 3 

--«-- Perugia LO' 

— • — Perugia 6 



1 800 Ge V p+pbar inelastic, Non-Dlffracllve 


: Charged Particle Multiplicity (|tiI<1.0. p >0.4GeV) 




^Sb ^ CDF data 




•a — e— Perugia 




Tin - ■ - Perugia 3 




^k^ --*-- Perugia LO 


-_ 


^a — •— Perugia 6 


n 


Pytto 6.423 r*v 


n 


Data From CDF Coliaboratior, PRDB5(2002)072005 


•^ 




30 40 50 

NjJ|ti|<1.0, p^>0.4GeV) 



Transverse Momentum of Z/y' (|ti|<2.5, p_>0.5GeV) 

-^ — Perugia 
- ■ ■ Perugia 3 
--■".-- Perugia LO' 
— • — Perugia 6 




75 100 125 

N„„ (|il|<2.5, p^>0.5GeV) 



Pytilla S.4E3 
m CDF Collabaralion. PRL84(2000)B45 




p^(tiard system) [GeV] 



20 30 40 50 

p^thard system) [GeV] 



Figure 10: Charged particle multiplicity and Drell-Yan p± spectra at the Tevatron (left) and at the 
LHC at 14 TeV (right) for the Perugia 0, 3, LO*, and 6 tunes. In particular, the Perugia 3 curve on 
the lower right-hand plot illustrates the consequences of choosing a different regularization procedure for 
ISR in the infrared, which shifts the position of the infrared peak of the DrcU-Yan p± spectrum without 
affecting the tail of the distribution. 



Perugia X (325): A Variant of Perugia which uses the MRST LO* PDF set [82]. Due to 
the increased gluon densities, a slightly lower ISR renormalisation scale and a higher MPI cutoff 
than for Perugia is used. Note that, since we are not yet sure the implications of using LO* for 
the MPI interactions have been fully understood, this tune should be considered experimental 
for the time being. In fig. [TOl we see that the choice of PDF does not greatly affect neither the 
min-bias multiplicity nor the Drell-Yan pj_ distribution, once the slight retuning has been done. 
Thus, this tune is not intended to differ significantly from Perugia 0, but only to allow people 
to explore the LO* set of PDFs without ruining the tuning. See [431 Perugia PDFs] for more 
distributions. 

Perugia 6 (326): A Variant of Perugia which uses the CTEQ6L1 PDF set [99]. Identical 
to Perugia in all other respects, except for a slightly lower MPI infrared cutoff at the Tevatron 
and a lower scaling power of the MPI infrared cutoff (in other words, the CTEQ6L1 distributions 
are slightly lower than the CTEQ5L ones, on average, and hence a lower regularization scale 



26 



can be tolerated). The predictions obtained are similar to those of Perugia 0, cf., e.g., fig. [10] 
and \M- 



Perugia 2010 (327): A variant of Perugia with the amount of FSR outside resonance 
decays increased to agree with the level inside them (specifically the Perugia-0 value for hadronic 
Z decays at LEP is used for FSR also outside Z decays in Perugia 2010, where Perugia uses 
the lower as value derived from the PDFs instead), in an attempt to bracket the description 
of hadronic event shapes relative to the comparison of Perugia to NLO+NLL resummations 
in [50] and also to improve the description of jet shapes [49]. The total strangeness yield 
has also been increased, since the original parameters, tuned by Professor, were obtained for 
the Q^-ordered shower and small changes were observed when going to the p_L-ordered ones. 
High-z fragmentation has been modified by a slightly larger infrared cutoff, which hardens the 
fragmentation spectrum slightly. The amount of baryon number transport has been increased 
slightly, mostly in order to explore the consequences of the junction fragmentation framework 
betteio, and the colour reconnection model has been changed to the newest one, MSTP(95)=8. 
See [43] for plots using this tune. 



Perugia K (328): A variant of Perugia 2010 that introduces a "ET" factor on the QCD 
2 — )■ 2 scattering cross sections used in the multiple-parton-interaction framework. The K- 
factor applied is set to a constant value of 1.5. This should make the underlying event more 
"jetty" and pushes the underlying-event activity towards higher p±. To compensate for the 
increased activity at higher p±, the infrared regularisation scale is larger for this tune, cf. table 
m in appendix O It does not give an extremely good central fit to all data, but represents a 
theoretically interesting variation to explore. 

The Perugia 2011 Tunes (350-359): The 2011 updates of the Perugia tunes were not 
included in the original published version of this manuscript. For reference, a description of 
them has been included in Appendix JBJ of this updated preprint. 

5 Extrapolation to the LHC 

"Predictions" Part of the motivation for updating the SO family of tunes was specifically 
to improve the constraints on the energy scaling to come up with tunes that extrapolate more 
reliably to the LHC. This is not to say that the uncertainty is still not large, but as mentioned 
above, it does seem that, e.g., the default Pythia scaling is not able to account for the scaling 
between the lower-energy data sets, and so this is naturally reflected in the updated parameters. 
In fig.dll we compare the main Perugia variations to Rick Field's Tune DW on the Drell-Yan 
p_i_ distribution (using the CDF cuts), the charged track multiplicity distribution in (inelastic, 
non-diffractive) minimum-bias collisions, and the average track p± as a function of multiplicity 
at the initial LHC center-of-mass energy of 7 TeV. We hope this helps to give a feeling for the 



Although there is room in the model to increase the baryon asymmetry further, this would also increase the 
frequency of multi-j unction-junction strings in pp events, which Pythia 6 is currently not equipped to deal with, 
and hence the strength of this effect was left at an intermediate level (cf. PARP(80) in table |4] in appendix [Aj • 



27 



7 TeV p+p 




30 40 50 

p^(hard system) [GeV] 



75 100 125 

N^„ (|ill<2.5, p^>0.5GeV) 



60 SO 100 120 

N^ {|ill<2.5, p^>0.5GeV) 



Figure 11: Perugia "predictions" for the p± of Drell-Yan pairs (left), the charged track multiplicity in 
min-bias (center), and the average track p± vs. A''ch in min-bias (right) at the LHC at 7 TeV. See [33] 
for other tunes and collider energies. 



kind of ranges spanned by the Perugia tunes (the PDF variations give almost identical results 
to Perugia for these distributions and are not shown. The Perugia 2010 variation gives the 
same Drell-Yan p± spectrum and is therefore not shown in the left-hand pane). A full set of 
plots including also the 14 TeV center-of-mass energy, for both the central region, \ri\ < 2.5, 
and the region 1.8 < r/ < 4.9 covered by LHCb, can be found on the web |43j . 

However, in addition to these plots, we thought it would be interesting to make at least 
one set of numerical predictions for an infrared sensitive quantity that could be tested with 
the very earliest high-energy LHC data. We therefore used the Perugia variations to get an 
estimate for the mean multiplicity of charged tracks in (inelastic, nondiffractive) minimum-bias 
pp collisions at center-of-mass energies of 0.9, 2.36, 7, 10, and 14 TeV, as shown in table [U 
In order to facilitate comparison with data sets that may include diffraction in the first few 
multiplicity bins, we recomputed the means with up to the first 4 bins excluded, and model 
uncertainties were inflated slightly for the first two bins. The uncertainty estimates correspond 
to roughly twice the largest difference between individual models and only drop below 10% near 
the collider energies used to constrain the models and then only when the lowest-multiplicity 
bins are excluded. Note also, however, that the uncertainties nowhere become larger than 20%. 
This presumably still underestimates the full theoretical uncertainty, due to intrinsic limitations 
in our ability to vary the models, but we hope nonetheless that it furnishes a useful first estimate. 



Comparison to the Current LHC Data At a late stage while preparing this article, data 
from the initial LHC runs at 900 GeV became available in the HepDATA web repository. We 
were therefore able to include a comparison of Perugia and a few main variations to the 
900 GeV ATLAS data [63]. We here explicitly omit bins with A'ch < 3 in the multiplicity 
and {p±) (Nch) distributions since we did not include diffractive events in the simulation. The 
resulting comparisons are shown in fig. [121 

The overall agreement between the models and the data is good, which is not surprising 



28 



Predictions for Mean Densities of Charged Tracks (Inelastic, Non-Diffr active Events) 

ArjAcj) ArjAcj) ArjAc/) ArjAcf) AT]Acf> 

LHC0.9TeV 0.21 ± 0.03 0.22 ± 0.03 0.24 ±0.02 0.26 ± 0.02 0.30 ± 0.02 

LHC 2.36 TeV 0.27 ± 0.03 0.28 ± 0.03 0.30 ± 0.02 0.33 ± 0.02 0.36 ± 0.02 

LHC 7 TeV 0.36 ± 0.04 0.37 ± 0.04 0.39 ± 0.04 0.42 ± 0.05 0.46 ± 0.04 

LHC 10 TeV 0.40 ± 0.05 0.41 ± 0.05 0.43 ± 0.05 0.46 ± 0.06 0.50 ± 0.06 

LHC 14 TeV 0.44 ± 0.06 0.45 ± 0.06 0.47 ±0.06 0.51 ± 0.06 0.54 ±0.07 

Table 1: Best- guess predictions for the mean density of charged tracks for min-bias pp coUisions at 
several different LHC energies. These numbers should be compared to data corrected to 100% track 
finding efficiency for tracks with \ri\ < 2.5 and p± > 0.5 GcV and 0% efficiency outside that region. The 
definition of a stable particle was set at cr > 10mm (e.g., the two tracks from a A° — >■ p+7r~ decay 
were not counted). The ± values represent the estimated uncertainty, based on the Perugia tunes. No 
simulation of diffraction was included in these numbers. 



given that the 900 GeV beam energy lies well within the energy span inside which the models 
were tuned. One point that may be worth remarking on is that the models appear to be 
undershooting the tail of the multiplicity distribution slightly (left). This confirms the tendency 
already observed in the comparison to the UA5 data, cf . fig. O while the models had a tendency 
to overshoot the tails of the Tevatron distributions, cf. figs. [3] and [9l Combined with early 
indications at 7 TeV from ALICE [lOOj and CMS |101j that, likewise, confirm an undershooting 
by the models of the high-multiplicity tail, we observe that it may be particularly difficult 
to describe both the Tevatron and LHC data sets simultaneously and that more work in this 
direction would be fruitful. One way of getting closer to an apples-to-apples comparison in a 
study of this particular issue would be to perform an LHC measurement applying the same cuts 
as those used by the CDF min-bias analysis. 



6 Conclusions 

We have presented a set of updated parameter sets (tunes) for the interleaved p^-ordered 
shower and underlying-event model in Pythia 6.4. These parameter sets include the revisions 
to the fragmentation and flavor parameters obtained by the Professor group |29p44j. The new 
sets further include more Tevatron data and more data from different collider CM energies in 
an attempt to simultaneously improve the overall description of the Tevatron data while also 
improving the reliability of the extrapolations to the LHC. We have also attempted to deliver a 
first set of "theoretical uncertainty bands", by including alternative tunes with systematically 
different parameter choices. The new tunes are available from Pythia version 6.4.23, via the 
routine PYTUNE or, alternatively, via the switch MSTP(5). 

Our conclusions are that reasonably good overall fits can be obtained, at the 10-20% level, 
but that the contribution of diffractive processes and the scaling of the overall activity with 



29 



Charged Particle Multiplicity (|tiI<2.5, p^>0.5GeV, N^^>3) 
4 ATLJ\S data 




40 60 

N^„ (|ti|<2.5, p^>0.5GeV, N„„>3) 



15 20 

Pi [GeV] 



Figure 12: Perugia "predictions" for tire charged multiplicity (left), p± (center), and (pj_) (A^ch) distri- 
butions in inelastic, non-diffractive pp collisions at 900 GeV, compared with ATLAS data. See j33] for 
other tunes and collider energies. 



collider energy are still highly uncertain. Other interesting questions to pursue concern the 
spectrum of ultra-hard single hadrons with momenta above 30 GeV |57 tl90f[93] . the (possi- 
bly connected) question of collective effects in pp and the dynamics driving such effects, the 
contribution and properties of diffractive interactions, tests of jet universality by constraining 
fragmentation models better in situ at hadron colliders as compared to constraints coming from 
LEP and HERA, and the question of the relative balance between different particle produc- 
tion mechanisms with different characteristics; e.g., between soft beam remnant fragmentation, 
multiple parton interactions, and traditional parton-shower / radiative corrections to the fun- 
damental scattering processes. 

We note that these tunes still only included LEP, Drell-Yan, and minimum-bias data directly, 
and that the lowest-multiplicity bins of the latter were ignored due to their relatively stronger 
sensitivity to diffractive physics which we deemed it beyond the scope of this analysis to attack. 
Furthermore, only one Drell-Yan distribution was used, the inclusive p± spectrum. Leading-jet, 
y/7-|-jet(s), underlying-event and jet structure observables were not considered explicitly. We 
wish to emphasise that such studies furnish additional important inputs both to tuning and 
to jet calibration efforts through such observables as jet rates, jet pedestals, jet masses, jet-jet 
masses (and inter-jet distances), jet profiles, and dedicated jet substructure variables. 

We hope these tunes will be useful to the RHIC, Tevatron, and LHC communities. 

Acknowledgments 

The Perugia tunes derive their names from the Perugia MPI Workshop in 2008, which brought 
people from different communities together, and helped us take some steps towards finding a 
common language. We thank the Fermilab computing division, S. Timm in particular, and the 
Fermilab theory group for providing and maintaining excellent dedicated computing resources 
without which the large runs necessary for this tuning effort would have been impossible. We 
acknowledge many fruitful interactions with the RHIC, Tevatron, and LHC experimental com- 



30 



munities, and are particularly grateful to B. Cooper, L. Galtieri, B. Heinemann, G. Hesketh, 
D. Kar, P. Lenzi, A. Messina, and L. Tomkins for detailed counter-checks and feedback. In 
combination with the writeup of this article, the author agrees to owing Lisa Randall a bottle 
of champagne if the first published measurement at 10 or 14 TeV of any number in table [T] is 
outside the range given in the table, and vice versa. 

This work was supported in part by the Marie Curie research training network "MCnet" 
(contract number MRTN-CT-2006-035606) and by the U.S. Department of Energy under con- 
tract No. DE-AC02-07CH11359. 

A Parameters for the Perugia Tunes 

The following tables give an overview of the parameter settings in Pythia corresponding to 
the Perugia tunes described in this paper. The settings for the previous "best" tune of the 
p_L-ordered model, Tune SOA-Pro, are included for reference. 



Parameter 


Type 


SOAp.o 


Po 


Phard 


PSOFT 


P3 


Pnocr 


Plo* 


Pe 


P2010 


Pa- 


MSTP(5) 


Tune 


310 


320 


321 


322 


323 


324 


325 


326 


327 


328 


PARJ(81) 


FSR 


0.257 


0.257 


0.3 


0.2 


0.257 


0.257 


0.257 


0.257 


0.26 


0.26 


PARJ(82) 


FSR 


0.8 


0.8 


0.8 


0.8 


0.8 


0.8 


0.8 


0.8 


1.0 


1.0 


MSTJ(ll) 


HAD 


5 


5 


5 


5 


5 


5 


5 


5 


5 


5 


PARJ(l) 


HAD 


0.073 


0.073 


0.073 


0.073 


0.073 


0.073 


0.073 


0.073 


0.08 


0.08 


PARJ(2) 


HAD 


0.2 


0.2 


0.2 


0.2 


0.2 


0.2 


0.2 


0.2 


0.21 


0.21 


PARJ(3) 


HAD 


0.94 


0.94 


0.94 


0.94 


0.94 


0.94 


0.94 


0.94 


0.94 


0.94 


PARJ(4) 


HAD 


0.032 


0.032 


0.032 


0.032 


0.032 


0.032 


0.032 


0.032 


0.04 


0.04 


PARJ(ll) 


HAD 


0.31 


0.31 


0.31 


0.31 


0.31 


0.31 


0.31 


0.31 


0.35 


0.35 


PARJ(12) 


HAD 


0.4 


0.4 


0.4 


0.4 


0.4 


0.4 


0.4 


0.4 


0.35 


0.35 


PARJ(13) 


HAD 


0.54 


0.54 


0.54 


0.54 


0.54 


0.54 


0.54 


0.54 


0.54 


0.54 


PARJ(21) 


HAD 


0.313 


0.313 


0.34 


0.28 


0.313 


0.313 


0.313 


0.313 


0.36 


0.36 


PARJ(25) 


HAD 


0.63 


0.63 


0.63 


0.63 


0.63 


0.63 


0.63 


0.63 


0.63 


0.63 


PAR J (26) 


HAD 


0.12 


0.12 


0.12 


0.12 


0.12 


0.12 


0.12 


0.12 


0.12 


0.12 


PARJ(41) 


HAD 


0.49 


0.49 


0.49 


0.49 


0.49 


0.49 


0.49 


0.49 


0.35 


0.35 


PARJ(42) 


HAD 


1.2 


1.2 


1.2 


1.2 


1.2 


1.2 


1.2 


1.2 


0.9 


0.9 


PARJ(46) 


HAD 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


PARJ(47) 


HAD 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 



Table 2: Final-State Radiation and Hadronisation Parameters of the Perugia tunes compared to SOA- 
Pro. For more information on each parameter, see |14) . 



31 



Parameter 


Type 


SOAp.o 


Po 


Phard 


PSOFT 


Ps 


Pnocr 


Plo* 


Pe 


P2010 


Pa- 


MSTP(5) 


Tune 


310 


320 


321 


322 


323 


324 


325 


326 


327 


328 


MSTP(51) 


PDF 


7 


7 


7 


7 


7 


7 


20650 


10042 


7 


7 


MSTP(52) 


PDF 


1 


1 


1 


1 


1 


1 


2 


2 


1 


1 


MSTP(3) 


A 


2 


2 


2 


2 


2 


2 


2 


2 


1 


1 


MSTU(112) 


A 


- 


- 


- 


- 


- 


- 


- 


- 


4 


4 


PARU(112) 


A 


- 


- 


- 


- 


- 


- 


- 


- 


0.192 


0.192 


PARP(l) 


ME 


- 


- 


- 


- 


- 


- 


- 


- 


0.192 


0.192 


PARP(61) 


ISR 


- 


- 


- 


- 


- 


- 


- 


- 


0.192 


0.192 


PARP(72) 


IFSR 


- 


- 


- 


- 


- 


- 


- 


- 


0.26 


0.26 


MSTP(64) 


ISR 


2 


3 


3 


2 


3 


3 


3 


3 


3 


3 


PARP(64) 


ISR 


1.0 


1.0 


0.25 


2.0 


1.0 


1.0 


2.0 


1.0 


1.0 


1.0 


MSTP(67) 


ISR 


2 


2 


2 


2 


2 


2 


2 


2 


2 


2 


PARP(67) 


ISR 


4.0 


1.0 


4.0 


0.25 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


PARP(71) 


IFSR 


4.0 


2.0 


4.0 


1.0 


2.0 


2.0 


2.0 


2.0 


2.0 


2.0 


MSTP(70) 


ISR 


2 


2 





1 





2 


2 


2 


2 


2 


PARP(62) 


ISR 


- 


- 


1.25 


- 


1.25 


- 


- 


- 


- 


- 


PARP(81) 


ISR 


- 


- 


- 


1.5 


- 


- 


- 


- 


- 


- 


MSTP(72) 


ISR 





1 


1 





2 


1 


1 


1 


2 


2 


MSTPOl) 


BR 


1 


1 


1 


1 


1 


1 


1 


1 


1 


1 


PARPOl) 


BR 


2.0 


2.0 


1.0 


2.0 


1.5 


2.0 


2.0 


2.0 


2.0 


2.0 


PARP(93) 


BR 


10.0 


10.0 


10.0 


10.0 


10.0 


10.0 


10.0 


10.0 


10.0 


10.0 



Table 3: Parton-Dcnsity, Initial-State Radiation, and Primordial fc^ parameters of the Perugia tunes 
compared to SOA-Pro. For more information on each parameter, see [14]. 



Parameter 


Type 


SOApro 


Po 


Phard 


PsOFT 


P3 


Pnocr 


Plo* 


Pe 


P2010 


Pa- 


MSTP(5) 


Tune 


310 


320 


321 


322 


323 


324 


325 


326 


327 


328 


MSTP(81) 


UE 


21 


21 


21 


21 


21 


21 


21 


21 


21 


21 


PARP(82) 


UE 


1.85 


2.0 


2.3 


1.9 


2.2 


1.95 


2.2 


1.95 


2.05 


2.45 


PARP(89) 


UE 


1800 


1800 


1800 


1800 


1800 


1800 


1800 


1800 


1800 


1800 


PARPOO) 


UE 


0.25 


0.26 


0.30 


0.24 


0.32 


0.24 


0.23 


0.22 


0.26 


0.26 


MSTP(82) 


UE 


5 


5 


5 


5 


5 


5 


5 


5 


5 


5 


PARP(83) 


UE 


1.6 


1.7 


1.7 


1.5 


1.7 


1.8 


1.7 


1.7 


1.5 


1.5 


PARP(84) 


UE 


- 


- 


- 


- 


- 


- 


- 


- 


- 


- 


MSTP(33) 


"K" 





























10 


PARP(32) 


"K" 


- 


- 


- 


- 


- 


- 


- 


- 


- 


1.5 


MSTP(88) 


BR 
































PARP(79) 


BR 


2.0 


2.0 


2.0 


2.0 


2.0 


2.0 


2.0 


2.0 


2.0 


2.0 


MSTP(89) 


BR 


1 


1 


1 





1 


2 


1 


1 








PARP(80) 


BR 


0.01 


0.05 


0.01 


0.05 


0.03 


0.01 


0.05 


0.05 


0.1 


0.1 


MSTP(95) 


CR 


6 


6 


6 


6 


6 


6 


6 


6 


8 


8 


PARP(78) 


CR 


0.2 


0.33 


0.37 


0.15 


0.35 


0.0 


0.33 


0.33 


0.035 


0.033 


PARP(77) 


CR 


0.0 


0.9 


0.4 


0.5 


0.6 


0.0 


0.9 


0.9 


1.0 


1.0 



Table 4: Underlying-Event, Beam-Remnant, and Colour-Reconnection parameters of the Perugia tunes 
compared to SOA-Pro. For more information on each parameter, see [14] . 



32 



B The Perugia 2011 Tunes 

An update of the Perugia tunes was prepared in the Spring of 2011, with the fohowing main 
goals, 

• Use the same value of Aqcd for all shower activity (ISR and FSR) , in particular to simplify 
matching applications. 

— The common value is taken from a PROFESSOR fit to LEP event shapes and jet 
rates |29y 44j and ignores the value given by the PDF set. 

— A variant is provided which also uses this Aqcd value for the MPI cross sections 
in the underlying event. This increases the rate of semi-hard mini-jets produced by 
MPI relative to the central Perugia 2011 tunes. 

— Due to a slightly increased level of soft ISR, the Perugia 2011 tunes only need 1 GeV 
of primordial kx to describe the CDF and D0 Drell-Yan p± spectra, as compared to 
2 GeV in the previous tunes. 

• Take into account some of the early lessons of LHC minimum-bias and underlying-event 
data at 900 and 7000 GeV: 

— Faster scaling of multiplicities with energy, motivated, e.g., by the ALICE [100] and 
ATLAS |102j min-bias charged multiplicity measurements. 

— Slightly larger underlying event as compared to Perugia 2010, motivated by the 
ATLAS UE measurement [103], see also [T04] . 

— Increased baryon production, especially of strange baryons (larger A/K ratio), mo- 
tivated by identified-particle measurements by the ALICE [lOSyiOG] and CMS |107j 
experiments and by the p/vr ratio measured by STAR [Uj. The total amount of 
baryon production (dominated by protons and neutrons) now appears to be at the 
upper limit of the range allowed by LEP |104j . 

— Increased baryon transport from the beam remnant (though still less than the Perugia 
SOFT tune), motivated by p/p and A/A measurements performed by the ALICE 
[T08] and LHCb experiments [109] . 

— Slightly softer LEP fragmentation functions than in the Perugia 2010 tune, since the 
plots on 1 104] indicated this was previously slightly too hard. A slight additional 
softening of baryon fragmentation functions was made to improve the agreement 
with baryon x distributions at LEP [104] . 

— The default suppression of strangeness in association with popcorn mesons (PARJ(6) 
and PARJ(7)) was removed to help improve H and $7 yields at LEP |104j . (Note, 
however, the consequences of this on particle-particle correlations have not been 
checked.) 

— Slightly larger K* /K ratio, motivated by comparisons of Perugia 2010 to LEP data 

unu. 

— Lower color-reconnection strength than the AMBTl tune, in order to lower {p±) (N^h), 
cf., e.g., [102]. 



33 



In total, ten tune variations are included in the "Perugia 2011" set. The starting point was in 
all cases Perugia 2010, with modifications as documented in the tables below. 

Perugia 2011 Tune Set 

Central Perugia 2011 tune (CTEQ5L) 

Variation using as{^p±) for ISR and FSR 

Variation using as{2p±) for ISR and FSR 

Variation using Aqcd = 0.26 GeV also for MPI 

Variation without color reconnections 

Variation using MRST LO** PDFs 

Variation using CTEQ 6L1 PDFs 

Variation using PARP(90)=0 . 16 scaling away from 7 TeV 

Variation using PARP(90)=0 . 32 scaling away from 7 TeV 

Variation optimized for Tevatron 

Note that these variations do not explicitly include variations of the non-perturbative hadroniza- 
tion parameters, cf. tabled hence those parameters would still have to be varied independently 
(i.e., manually) to estimate uncertainties associated specifically with the hadronization process. 

Parameters of the Perugia 2011 Tunes 



(350) 


Perugia 2011 


(351) 


Perugia 2011 radHi 


(352) 


Perugia 2011 radLo 


(353) 


Perugia 2011 mpiHi 


(354) 


Perugia 2011 noCR 


(355) 


Perugia 2011 M 


(356) 


Perugia 2011 C 


(357) 


Perugia 2011 T16 


(358) 


Perugia 2011 T32 


(359) 


Perugia 2011 Tevatron 



Parameter 


Type 


Perugia 


Perugia 2010 


Perugia 2011 (All) 


MSTP(5) 


Tune 


310 


327 


350 — 359 


MSTJ(ll) 


HAD 


5 


5 


5 


PARJ(l) 


HAD 


0.073 


0.08 


0.087 


PARJ(2) 


HAD 


0.2 


0.21 


0.19 


PARJ(3) 


HAD 


0.94 


0.94 


0.95 


PARJ(4) 


HAD 


0.032 


0.04 


0.043 


PARJ(6) 


HAD 


0.5 


0.5 


1.0 


PARJ(7) 


HAD 


0.5 


0.5 


1.0 


PARJ(ll) 


HAD 


0.31 


0.35 


0.35 


PARJ(12) 


HAD 


0.4 


0.35 


0.40 


PARJ(13) 


HAD 


0.54 


0.54 


0.54 


PARJ(21) 


HAD 


0.313 


0.36 


0.33 


PARJ(25) 


HAD 


0.63 


0.63 


0.63 


PARJ(26) 


HAD 


0.12 


0.12 


0.12 


PARJ(41) 


HAD 


0.49 


0.35 


0.35 


PARJ(42) 


HAD 


1.2 


0.9 


0.80 


PARJ(45) 


HAD 


0.5 


0.5 


0.55 


PARJ(46) 


HAD 


1.0 


1.0 


1.0 


PAR J (47) 


HAD 


1.0 


1.0 


1.0 



Table 5: Hadronisation Parameters of the Perugia 2011 tunes compared to Perugia and Perugia 2010. 
Parameters that were not explicitly part of the Perugia and Perugia 2010 tuning but were included in 
Perugia 2011 are highlighted in blue. For more information on each parameter, see |14] . 



34 



Parameter 


Type 


Po 


P2010 


Pii 


radHi 


radLo 


inpiHi 


nocR 


Mlo" 


CeLi 


T0.16 


To.32 


TeV 


MSTP(5) 


Tune 


320 


327 


350 


351 


352 


353 


354 


355 


356 


357 


358 


359 


MSTP(51) 


PDF 


7 


7 


7 


7 


7 


7 


7 


20651 


10042 


7 


7 


7 


MSTP(52) 


PDF 


1 


1 


1 


1 


1 


1 


1 


2 


2 


1 


1 


1 


MSTP(3) 


A 


2 


1 


1 


1 


1 


1 


1 


1 


1 


1 


1 


1 


MSTP(64) 


A 


3 


3 


2 


2 


2 


2 


2 


2 


2 


2 


2 


2 


MSTU(112) 


A 


- 


4 


5 


5 


5 


5 


5 


5 


5 


5 


5 


5 


PARP(61) 


ISR 


- 


0.192 


0.26 


0.52 


0.13 


0.26 


0.26 


0.26 


0.26 


0.26 


0.26 


0.26 


PARP(72) 


IFSR 


- 


0.26 


0.26 


0.52 


0.13 


0.26 


0.26 


0.26 


0.26 


0.26 


0.26 


0.26 


PARJ(81) 


FSR 


0.257 


0.26 


0.26 


0.52 


0.13 


0.26 


0.26 


0.26 


0.26 


0.26 


0.26 


0.26 


PARP(l) 


ME 


- 


0.192 


0.16 


0.16 


0.16 


0.26 


0.16 


0.16 


0.16 


0.16 


0.16 


0.16 


PARU(112) 


ME 


- 


0.192 


0.16 


0.16 


0.16 


0.26 


0.16 


0.16 


0.16 


0.16 


0.16 


0.16 


PARP(64) 


ISR 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


MSTP(67) 


ISR 


2 


2 


2 


2 


2 


2 


2 


2 


2 


2 


2 


2 


PARP(67) 


ISR 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


PARP(71) 


IFSR 


2.0 


2.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


MSTP(70) 


ISR 


2 


2 
































MSTP(72) 


ISR 


1 


2 


2 


2 


2 


2 


2 


2 


2 


2 


2 


2 


PARP(62) 


ISR 


- 


- 


1.5 


1.75 


1.0 


1.5 


1.5 


1.5 


1.5 


1.5 


1.5 


1.5 


PARJ(82) 


FSR 


0.8 


1.0 


1.0 


1.75 


0.75 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


MSTPOl) 


BR 


1 


1 


1 


1 


1 


1 


1 


1 


1 


1 


1 


1 


PARPOl) 


BR 


2.0 


2.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


1.0 


PARP(93) 


BR 


10.0 


10.0 


10.0 


10.0 


10.0 


10.0 


10.0 


10.0 


10.0 


10.0 


10.0 


10.0 



Table 6: Parton-Density, Initial-State Radiation, and Primordial kx parameters of the Perugia 2011 
tunes compared to Perugia and Perugia 2010. The main distinguishing features of each variation are 
highlighted in red. For more information on each parameter, see [14] . 



Parameter 


Type 


Po 


P2010 


Pii 


radHi 


radLo 


nipiHi 


nocR 


Mlo" 


CeLi 


T0.16 


To.32 


TeV 


MSTP(5) 


Tunc 


320 


327 


350 


351 


352 


353 


354 


355 


356 


357 


358 


359 


MSTP(81) 


UE 


21 


21 


21 


21 


21 


21 


21 


21 


21 


21 


21 


21 


PARP(82) 


UE 


2.0 


2.05 


2.93 


3.0 


2.95 


3.35 


3.05 


3.4 


2.65 


2.93 


2.93 


2.1 


PARP(89) 


UE 


1800 


1800 


7000 


7000 


7000 


7000 


7000 


7000 


7000 


7000 


7000 


1800 


PARPOO) 


UE 


0.26 


0.26 


0.265 


0.28 


0.24 


0.26 


0.265 


0.23 


0.22 


0.16 


0.32 


0.28 


MSTP(82) 


UE 


5 


5 


3 


3 


3 


3 


3 


3 


3 


3 


3 


3 


PARP(83) 


UE 


1.7 


1.5 


- 


- 


- 


- 


- 


- 


- 


- 


- 


- 


PARP(84) 


UE- 


- 


- 


- 


- 


- 


- 


- 


- 


- 


- 


- 


- 


MSTP(33) 


'^K" 






































PARP(32) 


'^K" 


- 


- 


- 


- 


- 


- 


- 


- 


- 


- 


- 


- 


MSTP(88) 


BR 






































PARP(79) 


BR 


2.0 


2.0 


2.0 


2.0 


2.0 


2.0 


2.0 


2.0 


2.0 


2.0 


2.0 


2.0 


MSTP(89) 


BR 


1 



































PARP(80) 


BR 


0.05 


0.1 


0.015 


0.015 


0.015 


0.015 


0.015 


0.015 


0.015 


0.015 


0.015 


0.015 


MSTP(95) 


CR 


6 


8 


8 


8 


8 


8 





8 


8 


8 


8 


8 


PARP(78) 


CR 


0.33 


0.035 


0.036 


0.036 


0.036 


0.036 


- 


0.034 


0.036 


0.036 


0.036 


0.05 


PARP(77) 


CR 


0.9 


1.0 


1.0 


1.0 


1.0 


1.0 


- 


1.0 


1.0 


1.0 


1.0 


1.0 



Table 7: Underlying- Event, Beam- Remnant, and Colour- Reconnection parameters of the Perugia 2011 
tunes compared to Perugia and Perugia 2010. The main distinguishing features of each variation are 
highlighted in red. For more information on each parameter, see jl4| . 



35 



C Overview of Tunes included in PYTHIA 



The following three tables give an overview of the tune presets that have so far been implemented 
in Pythia, as of version 6.4.23. They can be obtained either by setting MSTP(5) = NNN, where 
NNN is the tune number, or by calhng PYTUNE(NNN) before the call to PYINIT. It is not advisable 
to do both. Note that, when MSTP(5) is used, PYINIT calls PYTUNE, and the tune parameters 
will then overwrite any previous user modifications. Also consult the output of PYTUNE which 
informs you about useful references for each tune, its parameters, and a brief description of 
their meaning. 



Q^-ordered shower and "old" underlying-event model 



MSTP(5) Name 



Description 



Date 



1st generation: Rick Field's CDF tunes and a few more 



100 


A 


101 


AW 


102 


BW 


103 


DW 


104 


DWT 


105 


QW 


106 


ATLAS-DC2 


107 


ACR 


108 


D6 


109 


D6T 


2nd generation: The same 


110 


A-Pro 


111 


AW-Pro 


112 


BW-Pro 


113 


DW-Pro 


114 


DWT-Pro 


115 


QW-Pro 


116 


ATLAS-DC2- 




Pro 


117 


ACR-Pro 


118 


D6-Pro 


119 


D6T-Pro 



Rick Field's CDF Tune A 

Rick Field's CDF Tune AW 

Rick Field's CDF Tune BW 

Rick Field's CDF Tune DW 

As DW but with the old default ECM-scaling 

Rick Field's CDF Tune QW using CTEQ6.1M 

Arthur Moraes' (old) ATLAS tune ("Rome") 

Tune A modified with new CR model 

Rick Field's CDF Tune D6 using CTEQ6L1 

Rick Field's CDF Tune D6T using CTEQ6L1 



(Oct 2002) 
(Apr 2006) 
(Apr 2006) 
(Apr 2006) 
(Apr 2006) 



(Mar 2007) 



but with Professor's LEP parameters 

Tune A, but with Professor's LEP parameters (Oct 2008) 

Tune AW, but with Professor's LEP parameters (Oct 2008) 

Tune BW, but with Professor's LEP parameters (Oct 2008) 

Tune DW, but with Professor's LEP parameters (Oct 2008) 

Tune DWT, but with Professor's LEP parameters (Oct 2008) 

Tune QW,but with Professor's LEP parameters (Oct 2008) 
ATLAS-DC2/Romc, but with Professor's LEP param- (Oct 2008) 
cters 

: Tune ACR, but with Professor's LEP parameters (Oct 2008) 

: Tune D6, but with Professor's LEP parameters (Oct 2008) 

: Tune D6T, but with Professor's LEP parameters (Oct 2008) 



3rd generation: Complete 
129 Pro-Q20 



Q2-ordered Tune by Professor 
: Professor Q2-ordered tune 



(Feb 2009) 



Intermediate and hybrid models 



MSTP(5) Name 



Description 



Date 



200 IM 1 

201 APT 
211 APT-Pro 
221 Perugia 

APT 
226 Perugia 
APT6 



: Intermediate model: new UE, Q2-ord. showers, 

new CR 
Tune A w. pT-ordered FSR (Mar 2007) 

Tune APT, with LEP tune from Professor (Oct 2008) 

"Perugia" update of APT-Pro (Feb 2009) 



: "Perugia" update of APT-Pro w. CTEQ6L1 



(Feb 2009) 



36 



p^-ordered shower and interleaved underlying-event model 



MSTP(5) Name 



Description 



Date 



1st generation: Sandhoff-Skands CDF Min-Bias tunes and a few more 



300 
301 
302 
303 
304 
305 
306 



SO 

SI 

S2 

SOA 

NOCR 

Old 

ATLAS-CSC 



Sandhoff-Skands Tune using the SO CR model (Apr 2006) 

Sandhoff-Skands Tune using the SI CR model (Apr 2006) 

Sandhoff-Skands Tune using the S2 CR model (Apr 2006) 

SO with "Tune A" UE energy scaling (Apr 2006) 

"best try" without CR (Apr 2006) 

Original (primitive) CR model (Aug 2004) 

Arthur Moraes' pj^-ordered ATLAS tune 
r. CTEQ6L1 



2nd generation : The same, but with Professor's LEP parameters 



310 
311 
312 
313 
314 
315 



SO-Pro 

Sl-Pro 

S2-Pro 

SOA-Pro 

NOCR-Pro 

Old-Pro 



50, but with Professor's LEP parameters 

51, but with Professor's LEP parameters 

52, but with Professor's LEP parameters 
SOA, but with Professor's LEP parameters 
NOCR, but with Professor's LEP parameters 
Old, but with Professor's LEP parameters 



(Oct 2008) 
(Oct 2008) 
(Oct 2008) 
(Oct 2008) 
(Oct 2008) 
(Oct 2008) 



3rd gener 
320 
321 
322 
323 

324 
325 
326 
327 
328 
329 
330 
335 
336 
339 



ation : The Peruj 
Perugia 
Perugia HARD 
Perugia SOFT 
Perugia 3 

Perugia NOCR 
Perugia * 
Perugia 6 
Perugia 2010 
Perugia K 
Pro-pTO 
MC09 
Pro-pT* 
Pro-pT6 
Pro-pT** 



;ia. Professor, and ATLAS MC09 pT-ordercd Tunes 

: "Perugia" update of SO-Pro (Feb 2009) 

: More ISR, More FSR, Less MPI, Less BR, Less HAD (Feb 2009) 

: Less ISR, Less FSR, More MPI, More BR, More HAD (Feb 2009) 

: Alternative to Perugia 0, with different ISR/MPI bal- (Feb 2009) 
ance & different scaling to LHC & RHIC 

"Perugia" update of NOCR-Pro (Feb 2009) 

"Perugia" Tune w. (external) MRSTLO* PDFs (Feb 2009) 

"Perugia" Tune w. (external) CTEQ6L1 PDFs (Feb 2009) 

Perugia with more FSR off ISR and more s (Mar 2010) 

Perugia 2010 with a "iiT" factor on ctmpi (Mar 2010) 

Professor pT-ordered tune w. SO CR model (Feb 2009) 

ATLAS MC09 tune with (external) LO* PDFs (2009) 

Professor Tune with (external) LO* PDFs (Mar 2009) 

Professor Tune with (external) CTEQ6L1 PDFs (Mar 2009) 

Professor Tune with (external) LO** PDFs (Mar 2009) 



4th gener 
340 
341 
342 
343 
344 
350 
351 
352 
353 
354 
355 
356 
357 
358 
359 



ation : Tunes after 
AMBTl 
Zl 

Zl-Pro 
Z2 

Z2-Pro 
Perugia 11 
Perugia 11 radHi 
Perugia 11 radLo 
Perugia 11 mpiHi 
Perugia 11 noCR 
Perugia 11 M 
Perugia 11 C 
Perugia 11 T16 
Perugia 11 T32 
Perugia 11 TeV 



LHC 7 TeV data 

ATLAS Min-Bias tune 
Underlying-Event tune based on AMBTl 
As Zl, but with Professor's LEP tunc 
Underlying-Event tune based on AMBTl 
As Zl, but with Professor's LEP tunc 
Central Perugia 2011 tunc (with CTEQ5L) 
Using as{\p±) for ISR and FSR 
Using a,(2pj.) for ISR and FSR 
Using Aqcd = 0.26 also for MPI 
Best try without color reconnections 
Using MRST LO** PDFs 
Using CTEQ6L1 PDFs 
PARP(90)=0 . 16 away from 7 TeV 
PARP(90)=0.32 away from 7 TeV 
Optimized for Tevatron 



(Mar 2011) 
(Mar 2011) 
(Mar 2011) 
(Mar 2011) 
(Mar 2011) 
(Mar 2011) 
(Mar 2011) 
(Mar 2011) 
(Mar 2011) 
(Mar 2011) 



37 



References 

[1] A. M. Snigirev, Phys. Rev. D68, 114012 (2003), hep-ph/0304172. 

[2] V. L. Korotkikh and A. M. Snigirev, Phys. Lett. B594, 171 (2004), hep-ph/0404155. 

[3] AFS, T. Akesson et al, Z. Phys. C34, 163 (1987). 

[4] UAl, C.-E. Wulz, in proceedings of the 22nd Rencontres de Moriond, Les Arcs, Prance, 
15-21 March 1987. 

[5] UA2, J. Alitti et al, Phys. Lett. B268, 145 (1991). 

[6] CDF, F. Abe et al, Phys. Rev. D47, 4857 (1993). 

[7] CDF, F. Abe et al, Phys. Rev. Lett. 79, 584 (1997). 

[8] CDF, F. Abe et al, Phys. Rev. D56, 3811 (1997). 

[9] D0, V. M. Abazov et al, Phys. Rev. D67, 052001 (2003), hep-ex/0207046. 

[10] D0, V. M. Abazov et al, (2009), 0912.5104. 

[11] ZEUS, C. Gwenlan et al. Acta Phys. Polon. B33, 3123 (2002). 

[12] T. Sj5strand and P. Z. Skands, Eur. Phys. J. C39, 129 (2005), hep-ph/0408302. 

[13] T. Sjostrand and P. Z. Skands, JHEP 03, 053 (2004), hep-ph/0402078. 

[14] T. Sjostrand, S. Mrenna, and P. Skands, JHEP 05, 026 (2006), hep-ph/0603175. 

[15] T. Sjostrand, S. Mrenna, and P. Skands, Comput. Phys. Commun. 178, 852 (2008), 
0710.3820. 

[16] R. Corke and T. Sjostrand, JHEP 01, 035 (2010), 0911.1909. 

[17] Y. L Azimov, Y. L. Dokshitzer, V. A. Khoze, and S. I. Troyan, Z. Phys. C27, 65 (1985). 

[18] Z. Koba, H. B. Nielsen, and P. Olesen, Nucl. Phys. B40, 317 (1972). 

[19] J. F. Grosse-Oetringhaus and K. Reygers, (2009), 0912.0023. 

[20] CDF, R. Field, Int. J. Mod. Phys. A16S1A, 250 (2001). 

[21] CDF, A. A. Affolder et al, Phys. Rev. D65, 092002 (2002). 

[22] CDF, R. D. Field, (2002), hep-ph/0201192. 

[23] CDF, D. E. Acosta et al, Phys. Rev. D70, 072002 (2004), hep-ex/0404004. 

[24] CDF, R. Field, Acta Phys. Polon. B36, 167 (2005). 

[25] CDF, D. Kar, (2009), 0905.2323. 



38 



[26] K. Werner, Nucl. Phys. Proc. Suppl. 175-176, 81 (2008). 

[27] KASCADE, W. D. Apel et al, J. Phys. G36, 035201 (2009), 0901.4650. 

[28] T. Pierog and K. Werner, Nucl. Phys. Proc. Suppl. 196, 102 (2009), 0905.1198. 

[29] A. Buckley, H. Hoeth, H. Lacker, H. Schulz, and J. E. von Seggern, Eur. Phys. J. C65, 
331 (2010), 0907.2973. 

[30] A. Bacchetta, H. Jung, A. Knutsson, K. Kutak, and F. von Samson-Himmelstjerna, 
(2010), 1001.4675. 

[31] W. T. Giele, D. A. Kosower, and P. Z. Skands, Phys. Rev. D78, 014026 (2008), 0707.3652. 

[32] P. Skands, W. Giele, and D. Kosower, Higher-order corrections to timelike jets, in 
preparation, 2010. 

[33] M. Sandhoff and P. Skands, presented at Les Houches Workshop on Physics at TeV 
Cohiders, Les Houches, France, 2-20 May 2005, in hep-ph/0604120. 

[34] P. Skands and D. Wicke, Eur. Phys. J. C52, 133 (2007), hep-ph/0703081. 

[35] D. Wicke and P. Z. Skands, (2008), 0807.3248. 

[36] P. Z. Skands, Some interesting min-bias distributions for early LHC runs, FERMILAB- 
CONF-07-706-T, in C. Buttar et al, arXiv:0803.0678 [hep-ph]. 

[37] S. Alekhin et al, (2005), hep-ph/0601012. 

[38] TeV4LHC QCD Working Group, M. G. Albrow et al, (2006), hep-ph/0610012. 

[39] C. Buttar et al, (2008), 0803.0678. 

[40] P. Bartahni et al, Li *Hamburg 2008, Multiparticle dynamics (ISMD08)* 406-411. 

[41] T. Sjostrand, S. Mrenna, and P. Skands, PYTHIA update notes, available from 
http: //projects .hepf orge .org/pythia6/. 

[42] P. Z. Skands, (2009), 0905.3418. 

[43] P. Skands, Peter's pythia plots, see 

http : //home . f nal . gov/^skands/leshouches-plots/. 

[44] A. Buckley, H. Hoeth, H. Lacker, H. Schulz, and E. von Seggern, (2009), 0906.0075. 

[45] OPAL, K. Ackerstaff et al, Eur. Phys. J. C7, 369 (1999), hep-ex/9807004. 

[46] Particle Data Group, C. Amsler et al, Phys. Lett. B667, 1 (2008). 

[47] STAR, J. Adams et al, Phys.Lett. B637, 161 (2006), nucl-ex/0601033. 

[48] STAR, B. L Abelev et al, Phys. Rev. C75, 064901 (2007), nucl-ex/0607033. 



39 



[49] CDF, D. E. Acosta et al, Phys. Rev. D71, 112002 (2005), hep-ex/0505013. 

[50] A. Banfi, G. P. Salam, and G. Zanderighi, (2010), 1001.4082. 

[51] CDF, A. A. Affolder et al, Phys. Rev. Lett. 84, 845 (2000), hep-ex/0001021. 

[52] D0, V. M. Abazov et al, Phys. Rev. Lett. 100, 102002 (2008), 0712.0803. 

[53] A. Buckley et al, Effect of QED FSR on measurements of Z/^* and W leptonic final 
states at hadron colliders, in Tools and Monte Carlo Working Group: Summary Report, 
Les Houches, France, 2009, arXiv:1003.1643. 

[54] CDF, D. E. Acosta et al, Phys. Rev. D65, 072005 (2002). 

[55] N. Moggi, M. Mussini, and F. Rimondi, CDF Pubhc Note 9936, see 
http: //www-cdf . f nal .gov/physics/new/qcd/QCD .html. 

[56] CDF, F. Abe et al, Phys. Rev. Lett. 61, 1819 (1988). 

[57] CDF, T. Aaltonen et al, Phys. Rev. D79, 112005 (2009), 0904.1098. 

[58] T. Alexopoulos et al, Phys. Lett. B435, 453 (1998). 

[59] UA5, G. J. Alner et al, Phys. Rept. 154, 247 (1987). 

[60] UA5, R. E. Ansorge et al, Z. Phys. C43, 357 (1989). 

[61] ALICE, Eur. Phys. J. C65, 111 (2010), 0911.5430. 

[62] CMS, JHEP 02, 041 (2010), 1002.0621. 

[63] ATLAS, G. Aad et al, Phys. Lett. B688, 21 (2010), 1003.3124. 

[64] R. D. Field and R. P. Feynman, Phys. Rev. D15, 2590 (1977). 

[65] R. D. Field, CDF Note 6403, in hep-ph/0201192; further recent talks available from 
webpage 
http: //www. phys .uf 1 .edu/^rf ield/cdf /. 

[66] A. Banfi, G. P. Salam, and G. Zanderighi, JHEP 08, 062 (2004), hep-ph/0407287. 

[67] D0, V. M. Abazov et al, Phys. Rev. Lett. 94, 221801 (2005), hep-ex/0409040. 

[68] R. Field and R. C. Group, (2005), hep-ph/0510198. 

[69] CDF, R. Field, AIP Conf. Proc. 828, 163 (2006). 

[70] CDF, T. Aaltonen et al, Phys. Rev. Lett. 102, 232002 (2009), 0811.2820. 

[71] M. Cacciari, G. P. Salam, and S. Sapeta, (2009), 0912.4926. 

[72] CDF, T. Aaltonen et al, (2010), 1003.3146. 



40 



[73] D. Kar, Using Drell-Yan to probe the underlying event in Run II at CDF, PhD thesis, 
FERMILAB-THESIS-2008-54. 



74 
75; 
76 
77 
78 
79 
80 
81 
82; 
83 
84 
85; 
86 
87 



90 
91 
92 
93 
94 
95 
96 
97 
98 



T. Carh, T. Gehrmann, and S. Hoeche, (2010), 0912.3715. 

ATLAS, ATLAS Monte Carlo Tunes for MC09, ATL-PHYS-PUB-2010-002, 2010. 

B. Andersson, G. Gustafson, and B. Soderberg, Z. Phys. C20, 317 (1983). 

M. G. Bowler, Z. Phys. Cll, 169 (1981). 

Z. Nagy and D. E. Soper, (2009), 0912.4534. 

S. Catani, B. R. Webber, and G. Marchesini, Nucl. Phys. B349, 635 (1991). 

G. Corcella et a/., JHEP 01, 010 (2001), hep-ph/0011363. 

M. Bahr et al, Eur. Phys. J. C58, 639 (2008), 0803.0883. 

A. Sherstnev and R. S. Tliorne, Eur. Phys. J. C55, 553 (2008), 0711.2473. 

T. Kasemets, (2010), 1002.4376. 

M. Bengtsson and T. Sjostrand, Phys. Lett. B185, 435 (1987). 

M. Bengtsson and T. Sjostrand, Nucl. Phys. B289, 810 (1987). 

T. Plehn, D. Rainwater, and P. Z. Skands, Phys. Lett. B645, 217 (2007), hep-ph/0510144. 

P. Z. Skands, T. Plehn, and D. Rainwater, ECONF C0508141, ALCPG0417 (2005), 
hep-ph/0511306. 

R. Corke and T. Sjostrand, Eur.Phys.J. C69, 1 (2010), 1003.2384. 

J. Alwall, S. de Visscher, and F. Maltoni, JHEP 02, 017 (2009), 0810.5350. 

F. Arleo, D. d'Enterria, and A. S. Yoon, (2010), 1003.2963. 

S. Albino, B. A. Kniehl, and G. Kramer, (2010), 1003.1854. 

M. Cacciari, G. P. Salam, and M. J. Strassler, (2010), 1003.3433. 

A. S. Yoon, E. Wenger, and G. Roland, (2010), 1003.5928. 

T. Sjostrand and M. van Zijl, Phys. Rev. D36, 2019 (1987). 

M. Campanelh and J. W. Monk, (2009), 0910.5108. 

E735, T. Alexopoulos et al, Phys. Lett. B353, 155 (1995). 

A. Donnachie and P. V. Landshoff, Phys. Lett. B296, 227 (1992), hep-ph/9209205. 

CTEQ, H. L. Lai et al, Eur. Phys. J. C12, 375 (2000), hep-ph/9903282. 



41 



[99] J. Pumplin et al, JHEP 07, 012 (2002), hep-ph/0201195. 

[100] ALICE, K. Aamodt et al, Eur. Phys. J. C68, 345 (2010), 1004.3514. 

[101] CMS, V. Khachatryan et al, Phys.Rev.Lett. 105, 022002 (2010), 1005.3299. 

[102] ATLAS, G. Aad et al, (2010), 1012.5104. 

[103] ATLAS, G. Aad et al, (2010), 1012.0791. 

[104] P. Skands et al, see http://nicplots. cern. ch. 

[105] ALICE, K. Aamodt et al, Eur.Phys.J. C71, 1594 (2011), 1012.3257. 

[106] ALICE, K. Aamodt et al, (2011), 1101.4110. 

[107] CMS, V. Khachatryan et al, (2011), 1102.4282. 

[108] ALICE, A. K. Aamodt et al, Phys. Rev. Lett. 105, 072002 (2010), 1006.5432. 

[109] LHCb, in preparation, 2011. 



42