Event Statistics

At the end of the run you will want to write out the final statistics on number of events generated, the corresponding cross sections and the number of errors encountered. This is done with the pythia.stat() method, assuming pythia is an instance of the Pythia class.The method is steered entirely by settings values, see here.

Cross-section statistics

The ProcessLevel::statistics() method cannot be accessed directly, but only via the Pythia::stat() call above. When called it will loop over the list of existing processes, and for each write out name, code, the number of tried, selected and accepted events, the cross section and the estimated error on the latter. The three different event numbers are related to the Monte Carlo method used, whereby an initial upper estimate of the cross section is used to select a large number of trial phase-space points, whereof then not all survive. Rejections are normally done by the internal machinery, but can also be obtained by user hooks. Therefore:

tried events reflect the original number of phase-space points probed, as part of the upper estimate;
selected events correspond to those that survive the internal Monte-Carlo selection procedure;
accepted events are those that also survive the additional user cuts.

In most runs there would be no user hooks implemented, and then the numbers of selected and of accepted events will agree. Aborted events (see below) usually appear in the selected statistics but not in the accepted one.

For Les Houches events the total cross section will be correctly displayed; however the (optional) error value will not be used, so that the reported error will be smaller than the correct statistical ones, and often vanish completely. Furthermore, while the number of events is shown for each user process, the cross section is only for the sum of them.

Error messages

When Pythia is run, errors may occur, and give rise to warning messages. These may be of varying severity, as follows:

Abort means things went seriously wrong, and the initialization or event generation failed. In the former case it is not possible to generate events at all, in the latter the current event is flawed and should be skipped. In either case the respective method, Pythia::init() or Pythia::next(), then also returns the value false. There are occasions where an abort may be deliberate, such as when a file of Les Houches Events is read and the end of the file is reached.
Error normally is less severe. Typically the program will back up one step and try again. There are cases where this is not possible, in particular during the initialization and the generation of a hard process, and then the error may be followed by an abort as a direct consequence (with two separate messages).
Warning is even less severe. In some cases the program will try again, with good chances of success, in others no measure at all need to be taken.

The error messages is handled by a small part of the Info class. It is handed any abort, error or warning messages during the event generation phase, and will store each distinct message, with a counter for how many times it is issued. Thus it is possible to limit the number of identical messages issued, currently hardcoded so that each kind of error message is only printed once (static const int TIMESTOPRINT = 1). This can be overridden by the calling routine, so that all messages of this kind are shown, which is particularly relevant for the initialization stage. The summary table printed by Pythia::stat() provides a table with all the different messages issued, in alphabetical order, with the total number of times each was generated.

Multiparton-interactions statistics

If you call Pythia::stat() with Stat:showPartonLevel = true also statistics on multiparton interactions is printed, comprising a list of all allowed subprocesses with how many times each of them has been generated. For the nondiffractive process this also includes the hardest interaction, while else the hardest process is excluded from the statistics. (This is because the hardest process is of the same character and generated by the same machinery in the former case but not in the latter. Also, for the former case only, the standard statistics listing only lists nondiffractive as one single process, i.e. does not further specify the character of the hardest subprocess, so there is not any overlap between the two.)