TAMBO Collaboration Meeting Summer 2026

Extensive air-shower simulations from primary cosmic rays are traditionally performed using computationally intensive Monte Carlo methods, consuming significant computational resources in astroparticle physics. To address this challenge, we propose a continuous normalizing flow matching model for the TAMBO experiment to accelerate shower simulations. Trained on high-energy showers generated with CORSIKA, the model reproduces key observables, including energy spectra and time distributions. To improve computational efficiency, we implement several new features, including block-sparse attention to enable faster and more memory-efficient training, a DDPM solver to accelerate the sampling process, and additional architectural and training modifications to improve training stability. This work represents a first step toward enabling efficient end-to-end array layout optimization while substantially reducing the computational cost and time required for air-shower simulations.