Optimal observables and BSM model reinterpretation

在粒子对撞机上对新物理学的研究历史上一直由直接搜索数据分析主导，事件选择由少参数模型的特征驱动。这种方法在LHC的第一个十年中表现得相对较好，但由于所有这些搜索都是空白，因此正在进行一种转变，即从更通用的、与模型无关的测量中获得BSM约束。在这个项目中，我们将尝试在专用搜索和通用测量之间走一条中间道路，通过识别和测量可观测量，这些可观测量旨在在非简化模型的先验约束能力和实验不确定性的相关结构之间实现最佳平衡，这些结构可以稀释这种能力。充分利用Tomek的理论物理背景和研究经验，该项目将包括现象学（测量导向理论）和实验部分。这项工作的出发点将是扩展矢量类夸克的Contur研究，以扩展同时考虑的模型参数的数量，并将Contur集成到Gambit BSM全局拟合系统中。通过研究和优化提出的新的观测量对伪数据，一个新的集合与最大的约束力相对于当前的限制将被导出。然后，我们将使用ATLAS探测器对这些观测量进行探测器校正测量，并将结果包含在铆钉系统中，从而更新Gambit/Contur拟合。为了确保检测器校正的模型独立性，将需要针对模型参数空间中的许多点进行详细的检测器模拟和重建。由于这些步骤在计算上非常昂贵，运行许多不同的事件样本将不是一种选择：因此，我们将探索使用深度神经网络重新加权来在重建级别有效地对模型空间进行采样。

Searches for new physics at particle colliders have historically been dominated by direct-search data analyses, with event selections driven by the features of few-parameter models. This approach has served relatively well for the first decade of the LHC, but as all such searches have drawn a blank there is a shift afoot to also derive BSM constraints from more generic, model-independent measurements. In this project, we will attempt to tread a middle ground between dedicated searches and generic measurements, by identification and measurement of observables designed to be optimally balanced between a priori constraining power on non-simplifed models, and the correlated structure of experimental uncertainties which can dilute that power. Making best use of Tomek's theoretical physics background and research experience, the project will include both phenomenology (measurement-oriented theory) and experimental components. The starting point for this work will be extending the Contur study of vector-like quarks to extend the number of simultaneous model parameters considered, and to integrate Contur into the Gambit BSM global fit system. By studying and optimising proposed new observables against pseudo-data, a new set with maximal constraining power relative to current limits will be derived. We will then perform a detector-corrected measurement of these observables with the ATLAS detector, and include the results in the Rivet system and hence an updated Gambit/Contur fit. To ensure model-independence of the detector corrections, detailed detector simulation and reconstruction will be needed for many points through the model parameter space. As these steps are very computationally expensive, running many different event samples will not be an option: we will hence explore the use of deep neural network reweighting to efficiently sample the model space at reconstruction level.