Quantum Latent Diffusion Models for High-Resolution Simulation

Accurate simulation of particle interactions within detectors is one of the most computationally expensive tasks in High Energy Physics (HEP). While classical Monte Carlo simulators like Geant4 are highly accurate, they are too slow to meet the unprecedented data generation demands of the upcoming High-Luminosity LHC. Quantum Generative Models, specifically Quantum Diffusion Models, have been proposed as a faster alternative. However, current pure-quantum approaches struggle to scale beyond low-resolution "toy" datasets due to qubit limitations, deep circuit noise, and training instabilities. This project proposes the development of a hybrid Quantum Latent Diffusion Model (QLDM). By leveraging a classical Variational Autoencoder (VAE) to compress high-dimensional calorimeter data into a lower-dimensional latent space, the quantum diffusion process can be executed efficiently without overwhelming quantum resources. The final deliverable will be an open-source pipeline capable of generating high-resolution physics simulations. This approach allows the quantum model to operate in a regime where its ability to represent complex probability distributions may provide an advantage over classical generative models.