Multivariate Techniques for Identifying Diffractive Interactions in Hadron-Hadron Colliders
Date/Time: | Tuesday, 30 Aug 2011 from 3:00 pm to 4:00 pm |
---|---|
Location: | Zaffarano, Room A421 |
Contact: | Chunhui Chen |
Phone: | 515-294-5062 |
Channel: | College of Liberal Arts and Sciences |
Actions: | Download iCal/vCal | Email Reminder |
Approximately half of the total proton-proton cross section at TeV energies is expected to be due to elastic or inelastic diffractive scattering. Yet, these processes are poorly known both theoretically and experimentally. Theoretical diffractive cross section predictions based on extrapolations of experimental data at lower energies differ by large factors. In order to discriminate between various theoretical models, it is of utmost importance to develop efficient experimental techniques for identifying and distinguishing between various diffractive processes.
Experimental identification of diffractive interactions has traditionally been based on detection of large rapidity gaps (LRG), i.e. regions of low activity in the pseudorapidity space. However, at high energies, QCD fluctuations and long range correlations may create or destroy LRGs rendering them inefficient signatures for the underlying physics processes. The use of rapidity gaps is further hindered by the limited forward coverage of the LHC and Tevatron experiments.
In this talk, I will discuss an alternative approach to identifying diffraction based on state-of-the-art multivariate pattern recognition algorithms. Such methods do not rely solely on rapidity gaps but instead optimally utilize the full event topology to distinguish between single diffractive, double diffractive and non-diffractive interactions. I report results obtained using various different classification algorithms to show that such methods can efficiently identify the different event classes. Finally, I will introduce a fully probabilistic soft classification scheme developed specifically for the problem at hand and show that such a scheme can give a notable improvement over standard machine learning algorithms.