PACMAN LHC

Motivation

The goal of the project is to apply ML techniques to increase the performance of the accelerators. In collaboration with the LHC Operation groups, we aim to evaluate automatic and semi-automatic ways to model and optimize the overall collider set-up and define the strategy for the operational aspects of future projects (i.e., HL-LHC and FCC). The main goal is to get a deeper understanding of the physics mechanisms by modeling the losses in the LHC and further optimizing the beam losses in the LHC to increase the particle collision rates. This requires developing the model of the losses and dynamic aperture (DA) in the LHC depending on the control parameters.

Proposed Approach / Solution

The models of the particle losses in the LHC which depend only on the instantaneous values of control parameters do not generalize well to unseen data. We propose to model the losses time series depending on previously observed control parameters. Using a standard reparametrization, we reformulate the model as a Kalman Filter (KF) which allows for a flexible and efficient estimation procedure (Figure 1). For modeling the particle stability related to DA based on the simulated LHC data, it was proposed to use convolutional generative adversarial networks (GAN). The loss prediction was obtained from the estimated number of survived particles predicted by the network (Figure 2).

Impact

Understanding the influence of control parameters on the losses is extremely important to improve the operation, performance, and future design of accelerators. Our models based on machine data are a valuable addition to numerical models of particle losses, which can boost and improve the understanding of particle losses and help in the design of future colliders.