Colour Reconnection

The colour flows in the separate subprocesses defined in the multiparton-interactions scenario are tied together via the assignment of colour flow in the beam remnant. This is not an unambiguous procedure, and currently two different methods are implemented. In the first model the colour flow is reconstructed by how a PS could have constructed the configuration. In the second model, the full QCD colour calculation is taken into account, however the dynamical effects are modeled loosely, only an overall saturation is taken into account. The idea is to later account for other dynamical effects through colour reconnections.

A simple "minimal" procedure of colour flow only via the beam remnants does not result in a scenario in agreement with data, however, notably not a sufficiently steep rise of <pT>(n_ch). The true origin of this behaviour and the correct mechanism to reproduce it remains one of the big unsolved issues at the borderline between perturbative and nonperturbative QCD. Since no final answer is known, several models are implemented. The different models also rely on the two different colour assignments in the beam remnant. There are two, somewhat motivated, models implemented: the original PYTHIA scheme and a new scheme that tries to incorporate more of the colour knowledge from QCD.

The original PYTHIA scheme relies on the PS-like colour configuration of the beam remnant. This is combined with an additional step, wherein the gluons of a lower-pT MPI system are merged with the ones in a higher-pT MPI. A more detailed description of the merging can be found below. Relative to the other models it tests fewer reconnection possibilities, and therefore tends to be reasonably fast.

The new scheme [Chr14a]relies on the full QCD colour configuration in the beam remnant. This is followed up by a colour reconnection, where the potential string energy is minimized (ie. the lambda measure is minimized). The QCD colour rules are also incorporated in the colour reconnection, and determine the probability that a reconnection is allowed. The model also allows the creation of junction structures.

In addition to the two models described above, a simple model is implemented, wherein gluons can be moved from one location to another so as to reduce the total string length. This is one out of a range of simple models developed to study potential colour reconnection effects e.g. on top mass [Arg14], not from the point of view of having the most realistic description, but in order to probe the potential worst-case spread of predictions. All of these models are made available separately in include/Pythia8Plugins/ColourReconnectionHooks.h, with the setup illustrated in examples/main29.cc, but only the gluon-move one is sufficiently general and realistic that it has been included among the standard options here.

flag  ColourReconnection:reconnect   (default = on)
Allow or not a system to be merged with another one.

mode  ColourReconnection:mode   (default = 0; minimum = 0; maximum = 2)
Determine which model is used for colour reconnection. Beware that different BeamRemnants:remnantMode should be used for different reconnection schemes.
option 0 : The MPI-based original Pythia 8 scheme.
option 1 : The new more QCD based scheme.
option 2 : The new gluon-move model.

The MPI-based scheme

In this scheme partons are classified by which MPI system they belong to. The colour flow of two such systems can be fused, and if so the partons of the lower-pT system are added to the strings defined by the higher-pT system in such a way as to give the smallest total string length. The bulk of these lower-pT partons are gluons, and this is what the scheme is optimized to handle.

In more detail, an MPI system with a scale pT of the hard interaction (normally 2 → 2) can be merged with one of a harder scale with a probability that is pT0_Rec^2 / (pT0_Rec^2 + pT^2), where pT0_Rec is range times pT0, the latter being the same energy-dependent dampening parameter as used for MPIs. Thus it is easy to merge a low-pT system with any other, but difficult to merge two high-pT ones with each other.

parm  ColourReconnection:range   (default = 1.8; minimum = 0.; maximum = 10.)
The range parameter defined above. The higher this number is the more reconnections can occur. For values above unity the reconnection rate tends to saturate, since then most systems are already connected with each other. This is why 10 is set as an effective upper limit, beyond which it is not meaningful to let the parameter go.

The reconnection procedure is applied iteratively. Thus first the reconnection probability P = pT0_Rec^2 / (pT0_Rec^2 + pT^2) of the lowest-pT system is found, and gives the probability for merger with the second-lowest one. If not merged, it is tested with the third-lowest one, and so on. For the m'th higher system the reconnection probability thus becomes (1 - P)^(m-1) P. That is, there is no explicit dependence on the higher pT scale, but implicitly there is via the survival probability of not already having been merged with a lower-pT system. Also note that the total reconnection probability for the lowest-pT system in an event with n systems becomes 1 - (1 - P)^(n-1). Once the fate of the lowest-pT system has been decided, the second-lowest is considered with respect to the ones above it, then the third-lowest, and so on.

Once it has been decided which systems should be joined, the actual merging is carried out in the opposite direction. That is, first the hardest system is studied, and all colour dipoles in it are found (including to the beam remnants, as defined by the holes of the incoming partons). Next each softer system to be merged is studied in turn. Its gluons are, in decreasing pT order, inserted on the colour dipole i,j that gives the smallest (p_g p_i)(p_g p_j)/(p_i p_j), i.e. minimizes the "disturbance" on the existing dipole, in terms of pT^2 or Lambda measure (string length). The insertion of the gluon means that the old dipole is replaced by two new ones. Also the (rather few) quark-antiquark pairs that can be traced back to a gluon splitting are treated in close analogy with the gluon case. Quark lines that attach directly to the beam remnants cannot be merged but are left behind.

The joining procedure can be viewed as a more sophisticated variant of the one introduced already in [Sjo87]. Clearly it is ad hoc. It hopefully captures some elements of truth. The lower pT scale a system has the larger its spatial extent and therefore the larger its overlap with other systems. It could be argued that one should classify individual initial-state partons by pT rather than the system as a whole. However, for final-state radiation, a soft gluon radiated off a hard parton is actually produced at late times and therefore probably less likely to reconnect. In the balance, a classification by system pT scale appears sensible as a first try.

Note that the reconnection is carried out before resonance decays are considered by default. Colour inside a resonance therefore is not reconnected. The PartonLevel:earlyResDec can be switched on to perform resonance decays before colour reconnection, and then the partons from resonance decays could be affected. Ideally the time scales of resonance decays and of colour reconnection should be picked dynamically, but this is not yet the case. Notably the W, Z and t have intermediate decay time scales, somewhat but not much shorter than typical hadronization times, whereas the H is much more long-lived.

The newer scheme

parm  ColourReconnection:m0   (default = 0.5; minimum = 0.1; maximum = 5.)
This is the variable used in the lambda measure for the string length. See the different choices of lambda measure for exact formulaes. This variable is in the new model also used as a cut for forming pseudo particles that are not colour reconnected.

mode  ColourReconnection:nColours   (default = 9; minimum = 1; maximum = 30)
The number of reconnection colours, this should not be confused with the standard number of QCD colours. Each string is given an integer number between 0 and nColours - 1. Only strings with the same number are allowed to do a normal string reconnection. The default value provides the standard QCD probability that a triplet and an anti-triplet is in a singlet state. The probability for two strings to form a junction structure is in QCD given by the product of two triplets, which gives a probability of 1/3. Therefore the number of reconnection colours for junction formation is iColours % 3, where iColours refer to the integer of the string. The behaviour of junction formation therefore only changes slightly with this variable.

flag  ColourReconnection:sameNeighbourColours   (default = off)
In the normal colour reconnection two neighbouring strings are not allowed to have the same colour. Similar two strings orginating from a gluon split is not allowed to reconnect. The physics motivation for this is that it would require colour singlet gluons, and therefore for ordinary physics studies this should be turned off. But for testing of extreme scenarios (i.e. 1 colour), this variable needs to be turned on, since it is not possible to have different neighbouring colours.

flag  ColourReconnection:allowJunctions   (default = on)
This switch disables the formation of junctions in the colour reconnection.

mode  ColourReconnection:lambdaForm   (default = 0; minimum = 0; maximum = 2)
This allows to switch between different options for what lambda-measure to use. The formula shown are how much each end of a dipole or junction contribute to the total lambda-measure. The energies are defined in respectively the dipole or junction restframe.
option 0 : lambda = ln (1 + sqrt(2) E/m0)
option 1 : lambda = ln (1 + 2 E/m0)
option 2 : lambda = ln (2 E/m0)

parm  ColourReconnection:minimumGainJun   (default = 1; minimum = -100; maximum = 100)
The minimum lambda has to decrease in order to create a junction antijunction pair.

flag  ColourReconnection:allowDoubleJunRem   (default = on)
This parameter tells whether or not to allow a directly connected junction-antijunction pair to split into two strings. The lambda measure of the junction system is compared to that of the two possible string configurations. If the chosen configuration is the junction system, a q-qbar system is inserted between the junctions by removing some energy/momentum from the other legs.

The gluon-move scheme

This approach contains two steps, a first "move" one and an optional second "flip" one. Both are intended to reduce the total "string length" lambda measure of an event. For multiparton topologies the correct lambda measure can become quite cumbersome, so here it is approximated by the sum of lambda contributions from each pair of partons connected by a colour string piece. For two partons i and j with invariant mass m_ij this contribution is defined as lambda_ij = ln(1 + m^2_ij / m2Lambda). The 1 is added ad hoc to avoid problems in the m_ij → 0 limit, problems which mainly comes from the approximate treatment, and m2Lambda is a parameter set below.

In the move step all final gluons are identified, alternatively only a fraction fracGluon of them, and also all colour-connected parton pairs. For each gluon and each pair it is calculated how the total lambda would shift if the gluon would be removed from its current location and inserted in between the pair. The gluon move that gives the largest negative shift, if any, is then carried out. Alternatively, only shifts more negative than dLambdaCut are considered. The procedure is iterated so long as allowed moves can be found.

There is some fine print. If a colour singlet subsystem consists of two gluons only then it is not allowed to move any of them, since that then would result in in a colour singlet gluon. Also, at most as many moves are made as there are gluons, which normally should be enough. A specific gluon may be moved more than once, however. Finally, a gluon directly connected to a junction cannot be moved, and also no gluon can be inserted between it and the junction. This is entirely for practical reasons, but should not be a problem, since junctions are rare in this model.

The gluon-move steps will not break the connection between string endpoints, in the sense that a quark and an antiquark that are colour-connected via a number of gluons will remain so, only the number and identity of the intermediate gluons may change. Such a scenario may be too restrictive. Therefore an optional second flip step is introduced. In it all such colour chains are identified, omitting closed gluon loops. The lambda change is defined by what happens if the two colour lines are crossed somewhere, e.g. such that two systems q1 - g1 - qbar1 and q2 - g2 - qbar2 are flipped to q1 - g1 - g2 - qbar2 and q2 - qbar1. The flip that gives the largest lambda reduction is carried out, again with dLambdaCut offering a possibility to restrict the options. As with the move step, the procedure is repeated so long as it is allowed. An important restriction is imposed, however, that a given system is only allowed to flip once, and not with itself. The practical reason is that repeated flips could split off closed gluon loops quite easily, which tends to result in bad agreement with data.

As an option, singlet subsystems containing a junction may or may not be allowed to take part in the flip step. Since the number of junction systems is limited in this model the differences are not so important.

parm  ColourReconnection:m2Lambda   (default = 1.; minimum = 0.25; maximum = 16.)
The m2Lambda parameter used in the definition of the approximate lambda measure above. It represents an approximate hadronic mass-square scale, cf. m0 in the previous model. Its value is uncertain up to factors of 2, but the lambda change induced by a potential move or flip is rather insensitive to the precise value, owing to large cancellations.

parm  ColourReconnection:fracGluon   (default = 1.; minimum = 0.; maximum = 1.)
The probability that a given gluon will be considered for being moved. It thus gives the average fraction of gluons being considered.

parm  ColourReconnection:dLambdaCut   (default = 0.; minimum = 0.; maximum = 10.)
Restrict gluon moves and colour flips to those that reduce lambda by more than this amount. The larger this number, the fewer moves and flips will be performed, but those that remain are the ones most likely to produce large effects.

mode  ColourReconnection:flipMode   (default = 0; minimum = 0; maximum = 2)
Performing the flip step or not.
option 0 : No flip handling.
option 1 : Allow flips, but not for strings in junction topologies.
option 2 : Allow flips, including for strings in junction topologies.