Speaker
Description
Deep Lie Map Networks (DLMN) were introduced in earlier work as an optimisation framework that adjusts lattice parameters by fitting simulated beam trajectories to measured BPM data. In this contribution, we present a new implementation of DLMN based on the single-pass forward differentiation capability of MAD-NG, allowing exact derivatives of the particle coordinates with respect to lattice parameters to be computed during tracking. Unlike approaches that rely on back propagation or finite differences, this method performs gradient evaluation directly inside the symplectic tracking engine. This enables efficient gradient-based fitting of lattice parameters with improved memory efficiency and reduced computation time. The work demonstrates that differentiable, symplectic tracking provides a powerful foundation for data-driven optics modelling and establishes DLMN-in-MAD-NG as a scalable tool for future accelerator studies. Looking ahead, this framework could form a key component toward a real-time digital twin of accelerator lattices, where machine settings and model parameters are continuously inferred from live measurements.
Funding Agency
CERN
| In which format do you inted to submit your paper? | LaTeX |
|---|