We simulate ultracentral collisions (impact parameter vector $|\vec{b}|<1$ fm) of deformed uranium-uranium nuclei at 500 MeV/nucleon via the isospin-dependent Boltzmann-Uehling-Uhlenbeck transport model. Building upon these IBUU-generated simulations, we develop a convolutional orientation filter (COF) neural network model for the classification of reaction orientations. Based solely on the anisotropic flow of final-state particles, our COF model achieves high-precision event classification ($\approx 80\%$ validation accuracy), while also offering a novel approach for the accurate determination of the reaction plane. In addition to the linear relationship between $\langle v_2\rangle$ and ${\beta }_2$, we also find that the $\langle v_1^2\rangle$ signal is significantly enhanced when the major axis of the nucleus is oriented at approximately ${45}^{\circ }$ relative to the beam direction. This enhancement can be attributed to the significant reflection of nucleons or nuclear fragments induced by nuclear pressure. Additionally, the COF classifier also accurately characterizes the pion yields and ${\pi }^{-}/{\pi }^{+}$ ratios across different reaction orientations. Finally, we extend the discussion to include classification schemes at relativistic energies, utilizing simulations generated via the multiphase transport model in the string melting mode. Taken together, these findings provide robust support for the reliability of neural networks in evaluating the effects of symmetry energy in deformed nuclear collision reactions.

Close

PrevNext

View figure in article

• Low & intermediate energy heavy-ion reactions
• Network optimization
• Nuclear reactions
• Nuclear structure & decays
• Artificial neural networks