JuTrack, a Julia-based auto-differentiable accelerator simulation code for advanced dynamics, scientific machine learning and optimization

JuTrack is a high-performance accelerator modeling and particle tracking package developed in the Julia programming language. With compiler-level automatic differentiation (AD), JuTrack enables fast and precise derivative computations for arbitrary differentiable simulation functions. It supports efficient modeling of complex collective effects such as space-charge forces, wakefields, and beam-beam interactions. Beyond conventional tracking simulations, JuTrack also incorporates a machine learning–based module for self-field modeling and convenience interface that brings data-driven and physics-driven model together. In this paper, we demonstrate the capability and performance of JuTrack through a broad set of beam dynamics applications and optimization across various accelerator types, including a synchrotron light source, a heavy-ion linear accelerator, and the colliders. Built on Julia’s high-performance architecture and user-friendly syntax, JuTrack provides a powerful tool for beam dynamics studies and accelerator design optimization.