Authors:
(1) Cameron Parker, Cyclotron Institute, Texas A&M University and Department of Physics and Astronomy, Texas A&M University (E-mail: [email protected]);
(2) JETSCAPE Collaboration.
Table of Links
1. Introduction
JETSCAPE is a modular, task-based framework for simulating all aspects of heavy-ion collisions [1]. We first concern ourselves with tuning JETSCAPE using a novel hadronization module: Hybrid Hadronization [2, 3]. This method first uses Monte Carlo recombination [4, 5] on the partons after the shower stage and then hadronizes the rest with the Lund string model [6]. We use Bayesian analysis to tune JETSCAPE with Hybrid Hadronization to CMS and PHENIX data in vacuum proton-proton systems. Although JETSCAPE is primarily intended to compute heavy-ion collisions, a solid vacuum baseline is needed.
We then examine medium effects on hadronization by modeling how a single jet hadronizes using a brick of quark gluon plasma. Hybrid Hadronization is unique among hadronization models for shower Monte Carlos as it can take into account medium effects on hadronization through recombination of shower partons with thermal partons, and by allowing thermal partons to become part of strings connecting to shower partons. Both the existence of the medium and flow of the medium are expected to have pronounced effects on the final state hadrons produced. Medium flow both in the direction of the jet and transverse to it should provide flow effects on softer jet hadrons.
2. Vacuum Systems
The Bayesian analysis process begins with creating a starting set of design points within a parameter space given by the prior ranges for each parameter. Our prior distribution of parameters is assumed flat within parameters space, and we use a Latin hypercube to generate these points. They will be used to run JETSCAPE. A Gaussian process emulator is utilized to generate observables between the design points in parameter space. The observables produced are then compared to data. A Markov chain Monte Carlo determines new sets of points that improve the description. This process is repeated until convergence, giving us the posterior distribution. The posterior distributions for our set of parameters are shown in Fig. 1 together with correlations between pairs of parameters. The observables for the posterior distribution are shown in Fig. 2.
This paper is available on arxiv under CC BY 4.0 DEED license.