Neural network parametrization of the lepton energy spectrum in semileptonic B meson decays

Abstract

We construct a parametrization of the lepton energy spectrum in inclusive semileptonic decays of B mesons, based on the available experimental information: moments of the spectrum with cuts, their errors and their correlations, together with kinematical constraints. The result is obtained in the form of a Monte Carlo sample of neural networks trained on replicas of the experimental data, which represents the probability density in the space of lepton energy spectra. This parametrization is then used to extract the b quark mass m 1S b in a way that theoretical uncertainties are minimized, for which the value m 1S b = 4.84 ± 0.14 exp ± 0.05 th GeV is obtained.

Neural network determination of parton distributions: the nonsinglet case

“…successfully applied to a diverse variety of physical problems: structure functions [10], spectral functions for τ decays [11], energy spectra of B decays [12], and cosmic ray neutrino fluxes [13], thereby proving its flexibility and robustness. Its application to parton distributions is based on the same underlying concepts, but it is significantly more intricate for a variety of reasons to be discussed shortly, the most obvious being that parton distributions are not directly physical observable quantities.…”

“…Hence, we produce a full LO fit, with α s M 2 Z = 0.130 and a full NNLO fit, with α s M 2 Z = 0.115. The results of these fits are compared to the NLO ones in table 12. No variation of the quality of the fit is found between different perturbative orders: the values of χ 2 and σ (net) dat are unaffected by the perturbative order at which the computation is performed.…”

Continuous simulation of hypothetical physics processes with multiple free parameters

“…The Neural Network (NN) has been extensively used in Partticle Physics in the past decade, mostly as a multivariate discrimnant for signal/background separation [3,4,5]. In recent years, its usage as a universal function approximator [6,7] has also been exploited by physicists [8,9,10].…”

Extraction of the atmospheric neutrino fluxes from experimental event rate data

“…The purpose of the artificial data generation is to produce a Monte Carlo set of 'pseudo-data', i.e. N rep replicas of the original set of N dat data points, R (art)(k) i such ‡ This strategy has also been successfully applied with different motivations in other contexts like tau lepton decays [9] and B meson physics [10] that the N rep sets of N dat points are distributed according to an N dat -dimensional multi-gaussian distribution around the original points, with expectation values equal to the central experimental values, and error and covariance equal to the corresponding experimental quantities. This is achieved by defining…”