Differentiable Full Detector Simulation of a Projective Dual-Readout Crystal Electromagnetic Calorimeter with Longitudinal Segmentation and Precision Timing

TL;DR

This paper introduces a differentiable, fully automated detector simulation framework for future collider experiments, enabling precise modeling of detector geometries and AI/ML-based reconstruction strategies.

Contribution

It presents a novel differentiable simulation architecture within the key4hep software stack, allowing flexible, automated modeling of detector geometries and reconstruction methods.

Findings

Implementation of a differentiable simulation in key4hep

Automated, configurable detector geometry modeling

Discussion of AI/ML reconstruction strategies for future colliders

Abstract

A differentiable full detector simulation has been implemented in the key4hep software stack for future colliders. A fully automated and configurable geometry enabling differentiation of all detector dimensions, including crystal widths and thicknesses, is presented. The software architecture, development environment, and necessary components to implement a new detector concept from scratch are described. General AI/ML reconstruction strategies for future collider detectors are discussed, based around the idea of picking the right neural network for each detector.