This Chapter is structured into four main Sections. The first one details the L1 \(\tau \) algorithm implemented in Run-2 and still used in Run-3; the second and third describe my contribution to the preparation and commissioning of the \(\tau \) and \({\text {e}}/\gamma \) L1 trigger algorithms for the Run-3 data-taking; while the fourth details the development and performance evaluation of a novel technique for calibrating L1 TPs with a data-driven machine learning technique. The Phase-1 upgrade of the Level-1 (L1) trigger [1] took place during the first long shutdown of the LHC (2012–2015). In this instance, the calorimeter trigger was upgraded to its current Run-2 and Run-3 specification, allowing for sophisticated triggering algorithms to be implemented in powerful Field Programmable Gate Arrays (FPGAs). As discussed in Sect. 2.3.1 , the L1 calorimeter trigger is constituted of two processing tiers, the Layer-1 and Layer-2. In the former, the calibration of the calorimeter triggers primitive is performed. In the latter are implemented the reconstruction algorithms for \({\text {e}}/\gamma \) , \(\tau \) , jet, and sum objects, which optimally exploit the input they receive from Layer-1 in order to reduce the event rate to an adequate level while retaining the potentially interesting physics events with the highest efficiency possible. As part of my Thesis work, I have been the leading contributor to developments targeting both Layer-1 and Layer-2. I have been the main developer for the optimization of the L1 \(\tau \) algorithm for the restart of the data-taking in Run-3. I have introduced a simple yet more informative approach to optimizing the algorithm’s parameters. The same techniques have also been successfully implemented in the workflow for optimizing the L1 \({\text {e}}/\gamma \) algorithm, yielding excellent performance. Moreover, I have been coordinating the development and measurement in data of the performance of both the L1 \(\tau \) and \({\text {e}}/\gamma \) algorithms. This work is fundamental to the ambitious CMS Run-3 physics program’s success, which involves, among others, the development of multiple trigger paths dedicated to the collection of Higgs boson pair ( \({\text {H}}{\text {H}} \) ) signal events. Moreover, I have been the leading contributor to developing the first algorithm for calibrating L1 Trigger primitives (TPs) that exploits a fully data-driven Machine Learning (ML) technique. This novel method exploits offline reconstructed electrons and jets to calibrate single calorimeter TPs optimally, leading to promising performance. This innovative technique, based on a neural network architecture, is still being advanced and improved, and it is presented for the first time in this Thesis.

The Level-1 \(\tau _\textrm{h} \) Trigger: From the Past, to the Present

章节
发表日期：

2025年

查看全文

Jona Motta

摘要

This Chapter is structured into four main Sections. The first one details the L1 \(\tau \) algorithm implemented in Run-2 and still used in Run-3; the second and third describe my contribution to the preparation and commissioning of the \(\tau \) and \({\text {e}}/\gamma \) L1 trigger algorithms for the Run-3 data-taking; while the fourth details the development and performance evaluation of a novel technique for calibrating L1 TPs with a data-driven machine learning technique. The Phase-1 upgrade of the Level-1 (L1) trigger [1] took place during the first long shutdown of the LHC (2012–2015). In this instance, the calorimeter trigger was upgraded to its current Run-2 and Run-3 specification, allowing for sophisticated triggering algorithms to be implemented in powerful Field Programmable Gate Arrays (FPGAs). As discussed in Sect. 2.3.1 , the L1 calorimeter trigger is constituted of two processing tiers, the Layer-1 and Layer-2. In the former, the calibration of the calorimeter triggers primitive is performed. In the latter are implemented the reconstruction algorithms for \({\text {e}}/\gamma \) , \(\tau \) , jet, and sum objects, which optimally exploit the input they receive from Layer-1 in order to reduce the event rate to an adequate level while retaining the potentially interesting physics events with the highest efficiency possible. As part of my Thesis work, I have been the leading contributor to developments targeting both Layer-1 and Layer-2. I have been the main developer for the optimization of the L1 \(\tau \) algorithm for the restart of the data-taking in Run-3. I have introduced a simple yet more informative approach to optimizing the algorithm’s parameters. The same techniques have also been successfully implemented in the workflow for optimizing the L1 \({\text {e}}/\gamma \) algorithm, yielding excellent performance. Moreover, I have been coordinating the development and measurement in data of the performance of both the L1 \(\tau \) and \({\text {e}}/\gamma \) algorithms. This work is fundamental to the ambitious CMS Run-3 physics program’s success, which involves, among others, the development of multiple trigger paths dedicated to the collection of Higgs boson pair ( \({\text {H}}{\text {H}} \) ) signal events. Moreover, I have been the leading contributor to developing the first algorithm for calibrating L1 Trigger primitives (TPs) that exploits a fully data-driven Machine Learning (ML) technique. This novel method exploits offline reconstructed electrons and jets to calibrate single calorimeter TPs optimally, leading to promising performance. This innovative technique, based on a neural network architecture, is still being advanced and improved, and it is presented for the first time in this Thesis.