A Study Of Initial α-clustering Effects In Relativistic Heavy-ion Collisions

Modern physics has two basic concepts,namely the Standard Model of particle physics based on local gauge invariance and the Standard Model of cosmology based on general relativity.Quark-gluon plasma and light nuclei are both microscopic states of matter related to the evolution of the universe.Studies on them could help us vali-date the fundamental theories.These two fields are connected by relativistic heavy-ion collisions.This dissertation mainly studies the initial-clustering effect on final-state observables and discusses the possibility of identifying the structure of light nuclei in relativistic heavy-ion collisions.The study is based on a multi-phase transport（AMPT）model with string melting.Firstly,we investigated the initial-clustering effects on Hanbury-Brown–Twiss（HBT）correlation.During initialization,three configurations of¹²C,namely Woods-Saxon distribution of nucleons,three--cluster triangle,and three--cluster chain,were taken into account.Then,the¹²C+¹⁹⁷Au central collisions at√=200 Ge V were simulated.Based on identical-particle intensity interferometry,the azimuthal an-gle dependences of the HBT radii relative to the second-and third-order participant plane from-correlations were discussed.We also presented the hadronic rescat-tering time evolution of the azimuthally dependent HBT radii.It turns out that the ratio of the third-order to the second-order HBT radii²_（）,3/²_（）,2is quite sensitive to different configurations of¹²C.Meanwhile,we found that the effect is clearer with higher-transverse-momentum particles.In addition,we estimated the final eccentricity by 2²_,2/²_,0and compared it with the initial eccentricity.Our results indicated that the azimuthal angle dependence of the HBT radii relative to the second-and third-order participant plane can be taken as an effective probe to distinguish the exotic nuclear structure besides collective flow.In order to further consider the mixture of clustered configuration and non-clustered configuration,we introduced a machine-learning-based method.A Bayesian convolu-tional neural network was applied to classify initial configurations of¹²C(¹⁶O).Az-imuthal angle and transverse momentum distributions of charged pions were taken as inputs to train the classifier.The failure on an event-by-event basis is due to the large initial fluctuations in small systems.On a multiple-event basis,the overall classification accuracy can reach 95%for¹²C/¹⁶O+¹⁹⁷Au events at√=200 Ge V.When we applied the trained classifier to the test set with different proportions of both configura-tions,the predicted deviations could be within 5%.There exists competition between initial geometry effects and fluctuations in mixed samples.And the probability of the emergence of the features captured by the network is strongly related to the mixing proportion.Moreover,the performance of the network on the mixed dataset was further improved after setting a simple confidence threshold.We also checked the robustness of our predictions and the performance of a regression model.The machine-learning-based method not only exhibits the strong capability of learning features from fluctuations but also can be easily extended to experimental data.The interface between machine learn-ing and physics is extremely promising.