The evolution equations, eq. (), express that, during
a small increase , there is a probability for parton with
momentum fraction to become resolved into parton at
and another parton at
.
Correspondingly, in backwards evolution, during a decrease
a parton may be `unresolved' into parton . The relative
probability for this to happen is given by
the ratio . Using eq. () one obtains

(186) |

(187) |

giving the probability that a parton remains at from to a .

It may be useful to compare this with the corresponding expression for forward evolution, i.e. with in eq. (). The most obvious difference is the appearance of parton distributions in . Parton distributions are absent in : the probability for a given parton to branch, once it exists, is independent of the density of partons or . The parton distributions in , on the other hand, express the fact that the probability for a parton to come from the branching of a parton is proportional to the number of partons there are in the hadron, and inversely proportional to the number of partons . Thus the numerator in the exponential of ensures that the parton composition of the hadron is properly reflected. As an example, when a gluon is chosen at the hard scattering and evolved backwards, this gluon is more likely to have been emitted by a than by a if the incoming hadron is a proton. Similarly, if a heavy flavour is chosen at the hard scattering, the denominator will vanish at the threshold of the heavy-flavour production, which means that the integrand diverges and itself vanishes, so that no heavy flavour remain below threshold.

Another difference between and , already touched upon, is that the splitting kernel appears with a normalization in but only with in , since two gluons are produced but only one decays in a branching.

A knowledge of is enough to reconstruct the parton shower backwards. At each branching , three quantities have to be found: the value of the branching (which defines the space-like virtuality of parton ), the parton flavour and the splitting variable . This information may be extracted as follows:

- 101.
- If parton partook in the hard scattering or branched into
other partons at a scale
, the probability that was
produced in a branching at a lower scale is

(188) - 102.
- Given the of a branching, the relative probabilities
for the different allowed branchings are given by the
integrals above, i.e. by

- 103.
- Finally, with and known, the probability distribution in the splitting variable is given by the integrand in eq. ().

The selection of , and is then a standard task of the kind than can be performed with the help of the veto algorithm. Specifically, upper and lower bounds for parton distributions are used to find simple functions that are everywhere larger than the integrands in eq. (). Based on these simple expressions, the integration over may be carried out, and , and values selected. This set is then accepted with a weight given by a ratio of the correct integrand in eq. () to the simple approximation used, both evaluated for the given set. Since parton distributions, as a rule, are not in a simple analytical form, it may be tricky to find reasonably good bounds to parton distributions. It is necessary to make different assumptions for valence and sea quarks, and be especially attentive close to a flavour threshold ([Sjö85]). An electron distribution inside an electron behaves differently from parton distributions encountered in hadrons, and has to be considered separately.

A comment on soft-gluon emission. Nominally the range of the integral in is . The lower limit corresponds to , where the parton distributions in the numerator vanish and the splitting kernels are finite, wherefore no problems are encountered here. At the upper cut-off the splitting kernels and diverge. This is the soft-gluon singularity: the energy carried by the emitted gluon is vanishing, for . In order to calculate the integral over in , an upper cut-off is introduced, i.e. only branchings with are included in . Here is a small number, typically chosen so that the gluon energy is above 2 GeV when calculated in the rest frame of the hard scattering. That is, the gluon energy , where is the boost factor of the hard scattering. The average amount of energy carried away by gluons in the range , over the given range of values from to , may be estimated [Sjö85]. The finally selected value may thus be picked as , where is the originally selected value and is the correction factor for soft gluon emission.

In QED showers, the smallness of means that one can use rather smaller cut-off values without obtaining large amounts of emission. A fixed small cut-off is therefore used to avoid the region of very soft photons. As has been discussed in section , the electron distribution inside the electron is cut off at , for numerical reasons, so the two cuts are closely matched.

The cut-off scale may be chosen separately for QCD and QED showers, just as in final-state radiation. The defaults are 1 GeV and 0.001 GeV, respectively. The former is the typical hadronic mass scale, below which radiation is not expected resolvable; the latter is of the order of the electron mass. Photon emission is also allowed off quarks in hadronic interactions, with the same cut-off as for gluon emission, and also in other respects implemented in the same spirit, rather than according to the pure QED description.

Normally QED and QCD showers do not appear mixed. The most notable
exception is resolved photoproduction (in
) and resolved
events
(in
), i.e. shower histories of the type
.
Here the scales need not be ordered at the interface, i.e. the last
branching may well have a larger
than the first
one, and the branching
does not even have a strict parton-shower interpretation for the
vector dominance model part of the photon parton distribution.
This kind of configurations is best described by the
`'gamma/`*lepton*`'` machinery for having a flux of virtual photons
inside the lepton, see section . In this case,
no initial-state radiation has currently been implemented for the
electron (or or ). The one inside the virtual-photon
system is considered with the normal algorithm, but with the lower
cut-off scale modified by the photon virtuality, see `MSTP(66)`.

An older description still lives on, although no longer as the recommended one. There, these issues are currently not addressed in full. Rather, based on the selected for the parton (quark or gluon) at the hard scattering, the is selected once and for all in the range , according to the distribution implied by eq. (). The QCD parton shower is then traced backwards from the hard scattering to the QCD shower initiator at . No attempt is made to perform the full QED shower, but rather the beam-remnant treatment (see section ) is used to find the (or ) remnant that matches the (or ) QCD shower initiator, with the electron itself considered as a second beam remnant.