Systematic Evaluation of Errors on tunes

From Wiki Les Houches 09

Jump to: navigation, search

S. Mrenna, R. Chierici, F. Maltoni, Peter Skands, Andy Buckley, Matt Schwartz +....

  • Can we define regions of validity for the physics implemented in the MC? e.g. What about small x, high rapidity effects
  • Can we obtain credible uncertainty bands with existing tools, or are new methods needed

There is a need for some quantification of uncertainties, both from the point of view of learning about the physics in the MC models, and because such uncertainties have an impact on experimental uncertainties (in much the same was as the PDF uncertainties do).

  • The Peruguia tunes by Peter, which have errors bands, are the first real attempt at this. See this write up.
  • Professor is a good tool to sysematise this further.

Some points from the discussions and possibilties for future work:

Distinguish between (1) uncertainties of the parameters within a particular model, and (2) physics uncertainties based on our real knowledge of physics.

(1) Would be a very good start and is to some extent already done in the Perguia pythia 6 tunes, where eg. variations due to PDFs, scales in QCD radiation, energy scaling etc are varied separately.

(2) Is more difficult, but can be approached by for instance constraining FSR and hadronisation from e+e- data, then ISR from ep data, then underlying event etc from hadron collisions. This is along the lines of the approach taken in professor. Professor could also to some extent automate (1).

However, these methods will always require intelligent physics input, and selection of datasets, interpretation of chi2 etc are a matter of judgement, which will vary but can always at least be described so the the exercise is repeatable.

Choosing/weighting different datasets depending upon what a tune is to be used for is also likely to be interesting.

Should try to produce a brief summary of some guidelines for evaluating tunes and their uncertainites.

Take an example, probably the energy extrapolation of underlying event, and evaluate the uncertainty band. Maybe examine the effect of (pseudo) LHC data.

Personal tools