Parton Showers

Shower Model Selection
Some details

Shower Model Selection

As of version 8.3, PYTHIA comes with three different complete parton-shower frameworks: the original "simple" showers, the VINCIA antenna showers and the Dire dipole showers. Thereby comparisons between these different shower approaches become easier. A warning, however, is that the integration of VINCIA and Dire is still ongoing, so one may expect some adjustments to occur in the next few releases, over and above the level of normal evolution.

Since early days PYTHIA 8 has allowed external shower programs to be linked, thereby replacing the internal simple ones, see the Implement New Showers page. The VINCIA and Dire codes originally were structured to make use of this functionality, but were developed and distributed as separate codes. Starting with version 8.300, these two programs now are fully integrated into the PYTHIA distributions, and appear on equal footing with the old simple showers. Which of the internal shower frameworks that will be used in a run is determined by the following switch.

mode PartonShowers:model (default = 1; minimum = 1; maximum = 3)
Choice of which shower machinery that will be used in PYTHIA (when not linking an external shower).
option 1 : Simple Showers. This is the "old" shower framework that has its roots in PYTHIA 6 and has been distributed with PYTHIA 8 since the beginning. It is a less ambitious project than the other two, but for that reason also more mature and stable, which is a reason why it for now remains as default. It also has some special features that the other two don't.
option 2 : VINCIA Showers. Based on sequences of pT-ordered 2→3 branchings, the VINCIA shower model is similar to that of ARIADNE, which it resembles strongly for final-state evolution while VINCIA adopts a different (backwards-evolution) picture for initial-state radiation. The branching kernels, known as antenna functions, treat coherent sums of parton pairs without requiring a separation into "radiators" and "spectators". The current PYTHIA implementation includes QCD and QED 2→3 branchings with full mass dependence and, for the latter, multipole interference effects. A few longstanding hallmarks of VINCIA, such as automated uncertainty variations and (iterated) matrix-element corrections, are not yet available in this version.
option 3 : Dire Showers. Dire (short for Dipole resummation) implements a transverse-momentum ordered dipole shower in which radiator-spectator particle pairs evolve simultaneously. The emission phase space is fully symmetric between radiator and spectator, while the overall emission probability is separated into two pieces corresponding that are enhanced (suppressed) in region collinear (anti-collinear) to the radiator or the spectator, respectively. Dire includes QCD and QED emissions, a detailed treatment of (quark/lepton) mass effects, and is set up to include higher-order corrections, such as triple-collinear or double-soft parton emissions.
Further webpages, as linked above (and in the Parton Showers section of the left-column index), provide more detailed information on the respective framework.

There are some differences between the showers to be aware of

The Dire shower comes with a nontrivial variable weight. It is therefore important that results for each event are weighted by the event weight in pythia.info.weight(), e.g. when filled in a histogram. The simple and VINCIA showers by default come with unit weight, so do not have that issue. (Non-unit weights can still come from the hard-process cross section in some cases, so it is prudent always to apply the event weight.)
While all three shower models do ordinary QCD and QED radiation, beyond that the capabilities vary.
Tuned parameter values for PYTHIA's modeling of nonperturbative physics are normally only valid for the specific shower model they were tuned with, due to differences in perturbative regularisations between the three shower models and a nontrivial interplay between perturbative and nonperturbative physics modeling in general. Both VINCIA and Dire therefore come with their own dedicated default tune parameters which are activated in different ways, see the respective documentation pages.
As mentioned above, some of the hallmark features of VINCIA during its standalone development have not yet been carried over to the PYTHIA implementation. This includes (iterated) matrix-element corrections, automated uncertainty variations, tree-level 2→4 branchings and NLO corrected 2→ 3 ones, sector showers, facilities for matching and merging, and the VinciaRoot tool for live histogram displays. This is work in progress and we expect most of these features to reappear in future releases of PYTHIA 8.3.

Some details

Originally the TimeShower and SpaceShower classes implemented the simple shower, but also acted as base classes from which the external showers derived. This had some disadvantages for the purpose of the tighter integration of VINCIA and Dire, so the two aspects are now split. The TimeShower and SpaceShower classes remain as bare-bones base classes, from which the actual showers are derived. The code for the simple showers has been moved to the new derived SimpleTimeShower and SimpleSpaceShower. An external shower that does not use any of the existing shower algorithms will therefore work as before, which would be the normal case, but alternatively a shower could derive from the new classes and then reuse relevant code in them.

Settings names for the simple shower have been retained through this transformation, again for reasons of backwards compatibility of user code, e.g. in command files. Thus setting names beginning with TimeShower:, SpaceShower:, WeakShower: or UncertaintyBands: refer to the current baseline "simple" showers, and so do the HiddenValley: shower-related settings. As a rule, the corresponding VINCIA settings begin with Vincia: and the Dire ones with Dire:, DireTimes:, DireSpace: or Variations:. A few of the simple shower settings also apply to the other two, as documented on the respective page. It is possible that future evolution will standardize some settings, while others will be more clearly separated.