Periodic Reporting for period 1 - LHCbDFEI (Design of a Deep Full Event Interpretation for LHCb and application in semitauonic B decays)
Période du rapport: 2020-10-01 au 2022-09-30
LHCb is a particle physics experiment specialised in the study of properties and decays of heavy particles containing beauty and charm quarks, created in proton–proton collisions at the LHC. In Run 3, LHCb will collect physics events at higher rates thanks to newly installed detectors and a revolutionary software trigger that will enable LHCb to rapidly process signal data. The increase in physics reach brought by this upgrade comes at the cost of having to discriminate the interesting particles in the event from the rest at trigger time, since the amount of data produced would be too big to be stored. The higher collision rates also bring a much larger particle-combinatorics challenge than before, which motivates the development of new trigger approaches and algorithms. The current challenges will become dramatic for the foreseen Upgrade II of LHCb, in which the expected number of proton-proton collisions per event will experience an extra ten-fold increase.
The EU-funded LHCbDFEI project has led to the design and development of the prototype for a new trigger algorithm, that performs a Deep-learning based Full Event Interpretation (DFEI) for the first time in LHCb. The DFEI algorithm provides an inclusive, automatic and accurate multi-signal selection per event, which potentially maximises the trigger efficiency that can be achieved. This has been made possible by systematically leveraging the correlations amongst all the reconstructed particles per event, thanks to the use of Graph Neural Networks.
As a secondary complementary goal of the project, the analysis of LHCb data was done towards the measurement of quantum amplitudes of beauty-hadron decays with a tau lepton, that has important consequences for tests of the Standard Model of particle physics.
The EU-funded LHCbDFEI project has led to the design and development of the prototype for a new trigger algorithm, that performs a Deep-learning based Full Event Interpretation (DFEI) for the first time in LHCb. The DFEI algorithm provides an inclusive, automatic and accurate multi-signal selection per event, which potentially maximises the trigger efficiency that can be achieved. This has been made possible by systematically leveraging the correlations amongst all the reconstructed particles per event, thanks to the use of Graph Neural Networks.
As a secondary complementary goal of the project, the analysis of LHCb data was done towards the measurement of quantum amplitudes of beauty-hadron decays with a tau lepton, that has important consequences for tests of the Standard Model of particle physics.
The DFEI project requires many original developments and touches several scientific branches, including research on the application of deep‐learning (DL) techniques, structural modifications of the LHCb trigger system, detailed studies of the possible event topologies in simulation and algorithm optimisation to fulfill the tight online‐system computational requirements. During the course of the action, a new type of simulation based on PYTHIA was developed, to be used for model training and performance evaluation. A comparison of different DL architectures was performed, identifying Graph Neural Networks as the best candidate for the task. The design, development and optimisation of the first prototype of the DFEI algorithm for LHCb was successfully achieved. The DFEI project and algorithm has been presented at the Inter-experiment Machine Learning workshop and at the ICHEP conference. The results will be published in a paper.
As additional activities connected with the LHCb trigger, a set of 100 selection lines corresponding to different decay modes were implemented in the new Run 3 trigger framework, under the coordination of the beneficiary of the action.
Regarding physics studies, new software tools that allow to perform the proposed measurement for the first time have been developed and published as a paper in JINST. The full analysis chain has been setup and the sensitivity to the different observables using data collected in the years 2015 and 2016 has been studied. The analysis is currently in an advanced state, with the main types of remaining activities being related to checks on the procedure and the evaluation of systematic uncertainties. The experimental state of the art on this topic was presented at the Weak Interactions and Neutrinos and LHCP conferences and at the Beyond the Flavour Anomalies workshop.
As additional activities connected with the LHCb trigger, a set of 100 selection lines corresponding to different decay modes were implemented in the new Run 3 trigger framework, under the coordination of the beneficiary of the action.
Regarding physics studies, new software tools that allow to perform the proposed measurement for the first time have been developed and published as a paper in JINST. The full analysis chain has been setup and the sensitivity to the different observables using data collected in the years 2015 and 2016 has been studied. The analysis is currently in an advanced state, with the main types of remaining activities being related to checks on the procedure and the evaluation of systematic uncertainties. The experimental state of the art on this topic was presented at the Weak Interactions and Neutrinos and LHCP conferences and at the Beyond the Flavour Anomalies workshop.
The DFEI tool can highly improve the LHCb trigger capabilities, by providing an automatised and accurate way of selecting the part of the event which is useful for data analysis, allowing to discard the remaining information. This task is essential to pursue the ambitious LHCb physics program of the LHCb Upgrade II. As an important side application, DFEI can be used at the data-analysis level for a large variety of different signal decay modes, helping in the identification and separation of different sources of background. Both applications imply a huge reduction of the time and efforts needed to fine tune every one of the selections currently needed for all the decays of interest for LHCb, which can increase the overall scientific productivity of the experiment.
Concerning data analysis, the proposed study will provide key novel information to help understanding the nature of potential beyond-the-Standard-Model (BSM) physics, that can be responsible for a set of anomalous results found by several experiments in recent years. If the presence of BSM physics is confirmed, this will constitute a ground-breaking discovery for the field of Particle Physics.
Concerning data analysis, the proposed study will provide key novel information to help understanding the nature of potential beyond-the-Standard-Model (BSM) physics, that can be responsible for a set of anomalous results found by several experiments in recent years. If the presence of BSM physics is confirmed, this will constitute a ground-breaking discovery for the field of Particle Physics.