CALPAGAN: Calorimetry for Particles using GANs

TL;DR

This paper introduces CALPAGAN, a GAN-based method that converts fast simulation calorimeter images into detailed full simulation images, improving efficiency in high energy physics analyses.

Contribution

It adapts the pix2pix GAN framework specifically for calorimeter image translation, demonstrating high correlation with full simulations in key observables.

Findings

Strong correlation between generated and full simulation images

Effective in reproducing jet observables like momentum and mass

Highlights limitations and potential of GAN-based simulation methods

Abstract

In this study, a novel approach is demonstrated for converting calorimeter images from fast simulations to those akin to comprehensive full simulations, utilizing conditional Generative Adversarial Networks (GANs). The concept of pix2pix is tailored for CALPAGAN, where images from fast simulations serve as the basis(condition) for generating outputs that closely resemble those from detailed simulations. The findings indicate a strong correlation between the generated images and those from full simulations, especially in terms of key observables like jet transverse momentum distribution, jet mass, jet subjettiness, and jet girth. Additionally, the paper explores the efficacy of this method and its intrinsic limitations. This research marks a significant step towards exploring more efficient simulation methodologies in High Energy Particle Physics.