Using Nuclear Collision Simulation And Machine Learning To Investigate Single Nucleon Distributions And Nucleon-nucleon Correlations Within Atomic Nuclei

The internal structure of the atomic nucleus has been a major research topic in nuclear physics since it was discovered that the nucleus is composed of protons and neutrons.Because of the symmetry energy,the mean radius of neutrons in a heavy nucleus tends to be larger than that of protons,and the difference between the values is the thickness of the neutron skin.In experiments,the proton distribution in the nucleus can be accurately determined by electron elastic scattering experiments,whereas the traditional methods for the detection of the neutron distribution are based on hadron experiments with large errors.In addition,evidence for short-range nucleon-nucleon correlations has been observed experimentally,suggesting that nucleons in nuclei cannot simply be regarded as particles moving independently in a mean field.In high-energy nuclear physics studies,these primordial nuclear structures may have important implications for the final states of heavy ion collisions.Relativistic heavy-ion collisions are an important way to study the microscopic world,and heavy-ion collision experiments often produce large amounts of final state hadron data,and these final state hadrons theoretically retain information about the structure of the primordial nuclei.Deep neural networks have very powerful pattern recognition and mapping capabilities that can be trained on large amounts of data to improve model performance,and are ideally suited to analysing and extracting high-energy heavy ion collision data.Therefore,we hope to use the deep neural network approach to construct a correlation mapping of final state hadron distributions from heavy ion collisions to initial state atomic nucleus structures.In the neutron skin classification task,we sampled²⁰⁸P b nuclei from both halo-type and skin-type neutron distributions and used the SMASH model to generate two types of nuclear collisional final-state hadron distributions,halo-halo and skin-skin.Due to the up-and-down effect,the classification accuracy of the deep neural network is only 70%and 52%for the initial-state nucleon coordinates and final-state hadron momentum in the two classification tasks of halo-halo and skin-type.To improve classification accuracy,we used a multi-sample combination approach to allow the neural network to learn statistical information between events,resulting in classification accuracies close to 100%and 66%for the initial and final states,respectively.We also investigate the mapping of three types of nucleon-nucleon short-range correlations between the initial state¹⁹⁷Au nuclei and the corresponding final state events,namely the uncorrelation type（uncorrelation）,the hard-sphere rejection type（step correlation）and the true correlation type（NN correlation）.We have found that the type of nucleon-nucleon correlations affects the initial state anisotropy of hot dense nuclear matter in the transverse plane in high-energy heavy ion collisions,and for the final state momentum distribution of hadrons,the classification accuracy of deep neural networks is low in the two-way classification tasks of three correlation types.Similarly,we use a multi-sample combination method to improve the classifi-cation accuracy of neural networks for nuclear-nuclear short-range correlation types.Using 50 combined samples as input to the neural network,the classification accu-racy of the neural network is above 70%for the two-way classification tasks of the initial entropy density distribution of nuclear collisions in TRENTo,and in the range of 60%to 74%for the two-way classification tasks of final state events in SMASH.Overall,we have found that deep neural networks are effective in establishing correlation mappings between the final state of high-energy heavy ion collisions and the initial state nuclear structure when using a high-dimensional feature space multi-sample combination approach.In the future,we will look for more physical quantities on the final state correlations and fluctuations as a way to determine information on the initial state nucleon-nucleon correlations.