Bayesian Classification of Hadronic Diffraction in the Collider Detector at Fermilab
Permanent address of the item is
Diffraction is fundamentally a wide scale phenomena, and well understood from macroscopic mechanical waves up to quantum mechanical electron diffraction. However, hadronic diffraction is still missing a rigorous quantum field theoretical formulation, but it can be experimentally probed in high energy accelerators. Because diffraction is inherently a coherent process, it allows a unique perspective to probe partonic inner structure of protons (hadrons) and relativistic space-time evolution of high energy hadron-hadron collisions. In this thesis a Bayesian, probabilistic multivariate approach is developed for experimentally classifying diffractive hadronic scattering events from non-diffractive. For each measured collision event, the algorithm assigns a finite probability to an event to belong to a diffractive or non-diffractive process class. By integrating these probabilities over the full data sample, the interaction probabilities, known as cross sections, are estimated for different processes. The approach is Bayesian because it partly relies on the theoretical prior knowledge of cross sections. This probabilistic way is shown to be a sound approach, because hard event-by-event decisions are both theoretically and experimentally not uniquely definable. The reasons for this are thoroughly explained in this thesis. The underlying algorithm is based on ℓ1-norm regularized multinomial logistic regression. This regularization is shown to provide a mathematical view to the de-facto experimental physical signature of hadronic diffraction, known as the large rapidity gap. The experimental part of the thesis is done with proton-antiproton data collected in the CDF run II experiment at the center of mass collision energy √s = 1.96 TeV at Fermilab. For the first time major components of the proton-antiproton scattering total cross section are estimated using a multivariate algorithm. The obtained cross sections for single diffractive σ(SDL) = (4.87 ± 1.06) mb, σ(SDR) = (4.83 ± 1.04) mb, double diffractive σ(DD) = (6.16±1.93) mb and non-diffractive σ(ND) = (45.20±1.59) mb match the phenomenological theory predictions within errors. Results of the thesis indicate that the probabilistic approach is viable, and emphasize also the importance of experimental forward (small-angle) instrumentation that is limited at the CDF detector.