Speaker
Description
Accurate simulation of beam-beam interaction backgrounds, particularly the copious production of incoherent electron-positron pairs, is essential for detector design and physics analysis at the future Circular Electron Positron Collider (CEPC). However, the high computational cost of established codes such as Guinea-Pig++ presents a bottleneck for generating the large-statistics datasets required: a single bunch crossing (BX) requires about 10 CPU-hours, while robust performance studies demand $10^4$–$10^5$ BXs.
To address this challenge, we present the first application of a deep generative model to beam-beam pair background simulation, based on conditional Optimal Transport Conditional Flow Matching (OT-CFM). The model employs a physics-informed preprocessing pipeline that decomposes particle kinematics into transverse momentum, quantile-normalized longitudinal velocity, and circular-encoded azimuthal angle, enabling accurate reproduction of the full phase space including ultra-relativistic regimes. Conditioned on the production process, a single unified network reproduces the distinct kinematics and relative fractions of all three pair production channels (Breit–Wheeler, Bethe–Heitler, and Landau–Lifshitz).
The model was trained on 5000 BXs comprising approximately 10.8 million particles generated by Guinea-Pig++ for the CEPC Higgs factory beam parameters. It achieves a Maximum Mean Discrepancy of $2.8\times10^{-5}$ and Kolmogorov–Smirnov statistics below 0.002 for the energy, transverse momentum, and angular distributions, with the longitudinal velocity reproduced at a KS level of 0.013. The surrogate provides a speedup of approximately $2.7\times10^{5}$ over single-core Guinea-Pig++ execution, reducing a 100,000-BX dataset from an estimated 114 years of single-core CPU time to under 4 hours of GPU time. These results establish conditional flow matching as a computationally efficient tool for this class of simulation problems, with direct relevance to detector optimization at CEPC and other future high-luminosity lepton colliders.
