Phase Transition Dynamics In Heavy Ion Collisions And Application Of Neural Network To Particle Classification

The theory of Quantum Chromo-Dynamics, which describes the strong interactions, has two different vacuum states—perturbative vacuum and physical vacuum. Colored quarks and gluons can only exist in physical vacuum and are confined in color neutral hadrons. In 1970s, T.D. Lee et al. predicted that a new state of matter—quark gluon plasma will be formed while the phase transition from physical vacuum to perturbative vacuum happens in high energy nucleus-nucleus collisions. Since then, great progress has been made in the experimental and theoretical fields of heavy ion collisions. In the new century, the relativistic heavy ion collider(RHIC) in the Brookheaven National Laboratory in U.S. with colliding energy above several hundred GeV has started taking data and obtained a lot of new results. Combining all of the experimental results, people believe that partonic(quark and gluon) degree of freedom has been formed in the volume about thousands times that of hadrons. It means that the color confinement of quarks has disappeared and the transition between physical vacuum and perturbative vacuum happened. What's the character of the formed new state of de-confined quark and gluon? How to transfer from normal hadron to de-confined quarks and gluons? These will be the most important research topics in the present and future physics.The QCD phase diagram predicted by lattice gauge theory is the following: in the region with high baryon density and low temperature the transition between normal hadron matter and QGP is a first order phase transition. As the increase of temperature and decrease of baryon density, the line of the first order phase transition ends at the critical point and in higher temperature and lower density region, there is a cross-over from normal hadron to QGP.From the current experimental results, people has not found the critical point and the first order phase transition below critical point. In order to study the phase transition, RHIC will run at lower beam energies and GSI in Germany is building new facilities to satisfy the low temperature and high baryon density condition.Transport models are usually used to study the time and space evolution of heavy ion collisions. The transport model, which is a numerical solution of transport equation, will follow the collision process by inputting the nuclear geometry and reaction cross-section. The multiphase transport model of heavy ion collision includes parton phase and hadron phase and is suitable for the study of phase transition. In this model the partons are allowed to transport until ceasing interactions with other partons at given scattering cross section. By this means the transition between partons and hadrons is a parton-wise behavior instead of a collective one, resulting in individual hadronization time of partons. It's unreasonable that a few partons move freely in a system full of hadrons after most of the partons has already hadronize to hadrons and the perturbative vacuum turned to physical one.To solve this problem, we need to consider how to realize collective phase transition in transport model. In this thesis, taken the AMPT model as example, we investigate the time evolution of some physical quantities in the model. Assume the system achieves local thermal equilibrium, the temperature of partons and hadrons are extracted respectively by fitting their invariant transverse mass distribution to a thermal + radial flow model. Based on the parton and hadron temperature evolution, we implement a collective sudden phase transition following a supercooling state to the model by requiring all partons hadronize at the same time after the supercooling state. It turns out that the modified model with a sudden phase transition inherits the success of the original one in elliptic flow and is able to reproduce the experimental longitudinal distributions of final state particles better than the original one does. The encouraging results indicate that equilibrium phase transition should be taken into proper account in parton transport models for relativistic heavy ion collisions.Particle identification is important in experimental physics. In the current heavy ion collisions, higher reconstruction efficiency of strange baryons likeΞandΩis crucial to the precise measurement of elliptic flow and to the confirmation of the discovery of parton degree of freedom. Theorists have predicted that extotic multi-quark state (ΩΩ) of six strange quarks could be formed in quark gluon plasma. In order to find this kind of low yield particles, more efficient particle identification methods are called for.Artificial neural network is a kind of information processor with nonlinear dynamical properties by simulating the neuron system of human brain. It had been widely used in the fields such as pattern recognition, auto control, image process. In the early 1990s, neural network method was introduced to high energy experimental physics and it is found to be promising in dealing with the problem as track reconstruction, energy cluster reconstruction and particle identification. In this thesis, we will study how to improve the particle identification efficiency by using neural network method. At first, the neural network is applied to quark and gluon jets produced from the Monte Carlo simulation of e⁺e^- collision at s^1/2 = 91.2GeV. From this application, some factors which will affect the performance of neural network are investigated. Then neural network is used to reconstruct the A particles from (sNN)^1/2 = 200GeV d-Au collision data taken by RHIC STAR detector. Comparing track-cutting method with neural network, we found that the reconstruction efficiency of A particles by neural network method is 29% higher than that of the cutting method, but the signal-background ratio of the reconstruction by neural network method is lower than that of the cutting method. In order to improve the performance of neural network, a kind of re-sampling technique—adaptive boosting algorithm, is combined with neural network. To study the classification ability of boosted neural network, two different cases are considered, one is the two dimensional toy model in which signal and background are separable with irregular boundaries, the other is the Monte Carlo quark and gluon jets samples in which signal and background are overlapped with each other. We found that the boosting technique is able to improve the classification efficiency and signal-background ratio in the first case, i.e. the separable samples, but failed for the second case where there are mixing between signal and background.