Leptoquarks at Atlas Run 3

轻夸克是同时携带夸克和轻子量子数的粒子。轻夸克理论上是由轻子和夸克部分的相似之处所驱动的。它们是许多超越标准模型的模型的一个特征，特别是涉及统一的模型：Susy(超对称性)，Technicolor等。轻夸克是对B介子衰变中观察到的反常现象的一种可能解释，因为它们具有调解轻子味道普适性破坏的独特能力。它们还可能导致介子g-2反常磁矩理论和实验之间的差异。PHD项目旨在进行一种新的轻夸克对搜索，该搜索使用多变量技术并专注于多TeV质量区域，以充分利用Run III数据集。分析将包括除第三代费米子外的所有可能的轻子-夸克组合-总共25个可能的衰变将被调查。轻夸克与中微子的衰变将被包括在内。在高能下，介子动量分辨率很差，这些将被包括在分析中。分析的主要改进是使用了多变量技术。此前，两个轻夸克的平均质量被用作最终的判别式。在分析得到优化的低质量区域，这是足够的。对于新的分析，在高质量区域，计划使用多输入观测值，通过多变量技术和可能的图神经网络来更好地区分信号和背景事件。在包含中微子或介子分辨率较差的衰变中，质量不能完全重建，而必须将几个末态观测数据优化组合。ATLAS的目标是将目前基于RNN的Tau ID和衰变模式分类算法升级到基于GNN的体系结构。目前，有两种解决方案：TauJetGraph和GNTau。作为鉴定任务的一部分，将对两个国民总收入进行研究，以评估业绩差异。目标是正式实施和部署到ATLAS框架中，之后将需要执行基于数据的评估

Leptoquarks are particles that carry both quark and lepton quantum numbers. Leptoquarks are theoretically motivated by similarities in the lepton and quark sector. They are a feature of many models that go beyond the Standard Model, particularly models dealing with unification: SUSY (Supersymmetry), Technicolor, etc. Leptoquarks are a possible explanation of anomalies observed in B-meson decays, given their unique ability to mediate lepton flavour universality violation. They can also contribute to the muon g-2 anomalous magnetic moment discrepancy between theory and experiment.The PhD project aims to conduct a new leptoquark pair search that uses multi-variant techniques and focus on the multi-TeV mass regime to get the best out of the Run III dataset. The analysis will include all possible lepton-quark combinations apart from those with third generation fermions - a total of 25 possible decays will be investigated. Leptoquark decays with neutrinos will be included. Muon momentum resolution is poor at high energy, these will be included in the analysis.The major improvement to the analysis is to use multivariant techniques. Previously, the average mass of the two leptoquarks was used as the final discriminant. This was adequate in the low mass regime where the analysis was optimised. For the new analysis, in the high mass regime, the plan is to use multiple input observables that can better discriminate signal from background events through multivariate techniques and possible graph neural networks. The mass cannot be fully reconstructed in decays involving neutrinos or where the muon resolution is poor, instead several final state observables must be optimally combined.ATLAS is aiming to upgrade the current RNN based algorithm for Tau ID and decay mode classification to a GNN based architecture. Currently, there are two proposed solutions: TauJetGraph and GNTau. As part of the qualification task, a study of the two GNN's will be conducted to evaluate the performance differences. The goal is to implement and deploy officially into the ATLAS framework, subsequent to this a data-based evaluation will need to be performed