Online Electron Reconstruction at CLAS12

R. Tyson, G. Gavalian · 2025

Read Online Electron Reconstruction at CLAS12 by R. Tyson, G. Gavalian (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-05274.txt.

Introduction

Richard Tyson and Gagik Gavalian built a machine learning system that identifies electrons live, as data streams in at the CLAS12 experiment, in a 2025 paper. Experiments need fast reconstruction to monitor runs and trigger recording, and electron identification drives the trigger in this nuclear physics setup. Their method combines raw signals from several detector parts at the acquisition level.

The system flags electrons with purity above 75 percent while keeping nearly all true electrons. It also runs faster than the data arrives, so reconstruction can happen in real time. The authors note the approach will matter even more as experiments move to streaming readout with no hardware trigger.

Catalog record

CREATORR. Tyson, G. Gavalian
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE54 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-05274

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Online Electron Reconstruction at CLAS12 R. Tyson, G. Gavalian Abstract Online reconstruction is key for monitoring purposes and real time analysis in High Energy and Nuclear Physics experiments. A necessary component of reconstruction algorithms is particle identification that combines information left by a particle passing through several detector components to identify the particle’s type. Of particular interest to electro-production Nuclear Physics experiments such as CLAS12 is electron identification which is used to trigger data recording. A machine learning approach was developed for CLAS12 to reconstruct and identify electrons by combining raw signals at the data acquisition level from several detector components. This approach achieves an electron identification purity above 75% whilst retaining an efficiency close to 100%. The machine learning tools are capable of running at high rates exceeding the data acquisition rates and will allow electron reconstruction in real-time. This work enhances online analyses and monitoring and can contribute to improved triggering at CLAS12. This machine learning driven approach will also be crucial for experiments aiming to transition to streaming readout operations where online reconstruction will be a key component of the data taking paradigm. 1 Introduction High Energy and Nuclear Physics experiments are producing increasing large quantities of data, that need to be efficiently processed in real-time towards data preservation. Typically this relies on incremental trigger levels that filter the data based on hardware and physics requirements. Online monitoring and real time analysis are also crucial elements of High Energy and Nuclear Physics experimental operations as they ensure that the experiment delivers good quality data. Online reconstruction and particle identification schemes are key to triggering, online data selection and real time analysis. These schemes aim to process raw signals at the data acquisition level from several detector components in order to determine what original particles passed through a particle detector and their characteristics. The use of machine learning (ML) for online reconstruction is attractive due to high processing rates capable of keeping up with the data acquisition rate. Recent work explored the use of Convolutional Neural Networks (CNNs) [ 1 , 2 ] for electron triggering at the CEBAF Large Acceptance Spectrometer at 12 GeV (CLAS12) in order to improve the electron trigger purity [ 1 ] . Refs. [ 3 , 4 ] showed how electron identification could be improved in the ATLAS trigger using neural networks, allowing to halve background levels. Machine learning algorithms such as boosted decision trees are also employed in the High Level Triggers at the LHCb [ 5 ] . In Ref. [ 6 ] , neural networks were used to accurately reconstruct the vertex of a track from micropattern gaseous detector data. Ref. [ 7 ] demonstrates how muon identification for triggering at the Compact Muon Solenoid (CMS) could be improved by using a neural network to remove noise from the Drift Tubes and a second network to identify the side of passage of the hit with respect to a Drift Tube wire. In this article we will describe how online electron reconstruction can be achieved for CLAS12. Going one step further than previous approaches, the procedure proposed here will not only identify the presence of an electron but fully reconstruct all necessary information for real time analysis and physics reaction identification. The Continuous Electron Beam Facility (CEBAF) is located at the Thomas Jefferson National Accelerator Facility (JLab) [ 8 , 9 ] . CEBAF delivers an electron beam with energies up to 11 GeV to all four experimental halls at JLab [ 9 ] , with plans to increase the beam energy up to 12 GeV over the coming years. The CLAS12 detector is located in Hall B and aims to further the global understanding of hadronic structure and Quantum Chromodynamics [ 10 ] . A diagram of CLAS12 is shown in Figure 1 . The CLAS12 detector has almost complete azimuthal angle coverage, with low polar angular coverage (2.5-5 o ) enabled by the Forward Tagger, forward angular coverage (5-35 o ) enabled by the Forward Detector and large polar angular coverage enabled by the Central Detector (35-125 o ) [ 10 , 11 ] . The Forward Detector is further segmented into six sectors in azimuthal angle. Electrons with very low scattering angles are detected in the Forward Tagger, by using the coincidence of a track in the Forward Tagger tracker and a hit in the Forward Tagger hodoscope which allows to differentiate between electrons and photons [ 11 ] . Due to low backgrounds and simple detector apparatus, electron detection and identification in the Forward Tagger is rather straightforward. Electrons are not identified in the Central Detector. In the Forward Detector (FD), electron reconstruction is a more involved process, and will be the focus of the discussion presented in this article. Online reconstruction at CLAS12 would be beneficial for improved online monitoring, real time analysis and improved triggering. In 2023, the first steps towards full online reconstruction were taken, where tracks from the FD drift chambers were reconstructed in the online operation of CLAS12 experiments [ 12 ] . This was notably beneficial as CLAS12 then ran experiments with dual targets, and the tracking allowed to reconstruct the electron vertex which allowed to monitor the production of electrons from both targets. This online track reconstruction was based on ML tracking efforts at CLAS12 that are already incorporated in the offline reconstruction and which yielded increases in tracking efficiency of up to 50% for certain channels, with a decrease in computing time of roughly 30% [ 13 , 14 , 15 ] . Building on these AI predicted tracks, subsequent neural networks were able to predict the momentum and angles of the tracks with good resolution [ 16 ] . This avoids costly track fitting procedures, allowing to produce reconstructed tracks during online data taking operations. However, the tracks were not identified as belonging to a given particle type, meaning that they cannot be used for more advanced analysis or triggering. As mentioned previously, a proof of principle application of CNNs to the electron trigger was found to be a promising avenue for improved triggering [ 1 ] . However the drawback here is that the implementation suggested in Ref. [ 1 ] would only identify the presence of an electron, without assigning it to a track and fully reconstructing it. Additionally, CNNs are relatively slow compared to simpler neural networks. This article will present a proof of concept algorithm for full online electron reconstruction at CLAS12. Simple and fast Multilayer Perceptrons (MLP) [ 19 ] type algorithms will be combined to the online track reconstruction algorithm to identify electrons from the data acquisition level information of CLAS12 subsystems used for electron identification. Section 2 will give a brief description of the machine learning algorithms used in this article. Section 3 will present a more in depth overview of offline electron reconstruction at CLAS12. Section 4 will describe how tracks can be associated to hits in subsequent detector subsystems and how this information can be used to identify electrons. Section 5 will describe further tests demonstrating the good performance of these algorithms for online electron reconstruction. Section 6 will then describe improvements to the training sample that allow

Read the complete text