Toward a foundation model for heavy-ion collision experiments based on point-cloud diffusion

Abstract

A novel point cloud diffusion model for relativistic heavy-ion collisions, capable of ultra-fast generation of complete, event-by-event collision output, is introduced. When trained on UrQMD cascade simulations, the model generates realistic collision event output containing 26 distinct hadron species, as a list of particle momentum vectors along with their particle ID. From solving inverse problems to accelerating model calculations or detector simulations, the model can be a promising general purpose tool for heavy-ion collisions beneficial to both theoretical studies and experimental applications.

Figures (6)

• Figure 1: (Color online) Visualization of the generation process in HEIDi. Starting from random initial values for the momentum and ID at timestep t=300, the reverse diffusion process progressively denoises the samples, ultimately producing realistic particles at t=0. The absolute deviation for the $p_x$-$p_z$ probability densities of HEIDi from those of a random Gaussian probability density for various timesteps are shown. Note that HEIDi generates particles, not distributions. The distributions presented are constructed from the particles in 2000 HEIDi events.
• Figure 2: (Color online) Main figure: Mean event-multiplicity of various hadrons for Au-Au collisions at $10A$ GeV with b=1 fm. The results from the diffusion model are shown as orange bars, while the blue bars correspond to multiplicities from the UrQMD cascade model. The HEIDi model can also generate $\bar{p} ,\bar{n} , \bar{\Lambda},\bar{\Sigma}^-,\bar{\Sigma}^0,\bar{\Sigma}^+,\bar{\Xi}^0,\bar{\Xi}^+,\bar{\Omega}^+ \mathrm{ and }\ \Omega^-$ in an event. However, these hadrons have been excluded from the plot due to their very small multiplicities ($<$1/event). The particles marked 0 are the fake particles used to maintain a constant multiplicity to the event while n' and p' are spectator neutrons and protons respectively. Inset: Multiplicity distributions of selected hadrons at mid rapidity ($|y| < 0.1$). The results from the diffusion model are shown as solid lines, while the dashed lines correspond to multiplicities from the UrQMD cascade model.
• Figure 3: (Color online) Rapidity distributions of selected hadrons for Au-Au collisions at $10A$ GeV with b=1 fm. The solid lines show the diffusion model results while the dotted lines denote UrQMD results.
• Figure 4: (Color online) Transverse momentum distributions of selected hadrons at mid rapidity ($|y| < 0.1$) for Au-Au collisions at $10A$ GeV with b=1 fm. The results of generative model are shown in solid lines while the dotted lines denote UrQMD results. The generative model successfully reproduces the UrQMD $p_T$ spectra across most momentum ranges, with the exception of the deviation at very low $p_T$ values.
• Figure 5: (Color online) Global features of generated events. The distributions of the total energy, baryon number and charge of all mid rapidity particles ($|y| < 0.1$) are shown in panels 1, 2 and 3, respectively. Results from the generative model are shown in red, while the corresponding UrQMD distributions are shown in blue.