Physics Informed Neural Network Diffusion Equation (PINNDE)

Generating simulations of calorimeter showers of interacting particles are a crucial step in analyzing the results of large scale experiments at LHC. Traditionally very accurate simulations are produced using pipeline like GEANT4. This simulation step is currently a computational bottleneck and it is forecast to overwhelm the computing budget of the LHC experiments in the near future. The goal of this project is to build upon the fast and accurate sampler (PINNDE) developed in my GSOC 2025 project and use it to generate high-fidelity simulations of calorimeter showers. This pipeline consists of two distinct steps, namely building an accurate autoencoder capable to compressing the particle shower data onto a low-dimensional latent space and a diffusion model trained to generate samples from this low-dimensional latent space. In this project I will be implementing different auto encoders motivated by the physics and geometric constraints of the calorimeter experiments and use the Physics informed neural network trained to approximate the reverse time ODE of the diffusion model to generate samples of the calorimeter showers. The final deliverable will be bench marking this fast-sampler for calorimeter showers against the other solutions submitted to the Fast Calorimeter Simulation Challenge 2022.