Data analysis for the LUX-ZEPLIN dark matter search

LUX-ZEPLIN（LZ）是一个暗物质搜索实验，正在南达科他州（美国）的桑福德地下研究设施（SURF）组装和委托（2018年至2020年初）。两相时间投影室使用10吨液体氙和大约100吨液体氙。500个光电倍增管（PMT）用于探测暗物质与常规物质相互作用的极其罕见的事件。为了明确检测难以捉摸的暗物质粒子称为弱相互作用大质量粒子（WIMP）非常强大的分析程序，由严格的检测器校准计划的支持，是必需的。暗物质在许多不同的背景下被引用，通常也在核心天体粒子物理学之外。暗物质是天体粒子和粒子物理学路线图及其资助机构的优先主题。支持的程度反映了专家和广大公众对这一专题的普遍和广泛兴趣。LZ是英国目前在直接暗物质搜索领域的旗舰实验。LZ的灵敏度范围有可能产生世界领先的结果，这将是WIMP发现的变革。即使返回空结果，LZ的额外灵敏度范围也将覆盖比前代显著扩大的参数空间。稀有事件实验需要精确校准的探测器以及根据现有校准数据调整的极其精确的模拟，以模拟探测器的响应。校准数据将提供丰富的探测器事件库，为数据提取和简化软件提供出色的训练集。识别单个光子及其对闪烁/电离事件的归因的算法必须非常稳健。实现这一目标的关键一步是从数字化仪记录的脉冲序列中提取和识别单光子。这样做的算法已经存在，并且已经在适合过去和运行实验的水平上进行了测试，但是，通过引入新技术可以进一步改进。该项目的主要目的是优化LZ中数据分析链的元素。为了取得成功，将进行详细的模拟，以提供对探测器对校准源的响应的最佳理解。这是为了准备实验，使用一套内部分散的（83 mKr，131 Xe，220 Rn和CH 4标记的氚）和外部部署的密封中子（AmLi，205，206 BiBe，252 Cf，88 YeBe）和伽马（57 Co，22 Na，133 Ba，228 Th）辐射源，以提供对背景事件和WIMP信号响应的高统计校准。该项目的目标是模拟LZ实验中使用的中子和伽马源的事件，并重建事件的位置和能谱，这是探测器校准的一个关键部分。改进光子探测、重建和触发工作将涉及在氙时间投影室中生成特定事件的模拟（例如，重复使用校准事件），事件的数据记录在以前的实验中，并使用所有可能的信息，是为了提供和验证适当的增强算法。在这篇论文中的方法的新颖之处是探索的好处，开发和应用人工智能，e. G.机器学习，可能有。如果成功，这将是朝着处理日益复杂的暗物质探测器及其产生的数据迈出的重要一步。该项目是在LZ合作范围内进行的，牛津大学是该合作的成员之一。该合作包括（2018年）38个研究小组，成员超过250人。该项目以计算机为基础，使用基于GEANT的模拟、ROOT数据分析包和LZAp分析框架。

LUX-ZEPLIN (LZ) is a Dark Matter search experiment that is being assembled and commissioned (2018 - early 2020) at the Sanford Underground Research Facility (SURF) in Lead, South Dakota (USA). The two-phase time projection chamber uses 10 tonnes of liquid xenon and approx. 500 photomultiplier tubes (PMT) to detect extremely rare events of dark matter interactions with regular matter. For the unambiguous detection of the elusive dark matter particle called Weakly Interacting Massive Particle (WIMP) extremely robust analysis procedures, supported by a rigorous detector calibration programme, are required. Dark matter is cited in many different contexts, often also outside core astro particle physics. Dark matter is a priority topic in the roadmaps of astro particle and particle physics, and their funding agencies. The level of support reflects the general and broad interest in the topic not only by experts but also the wider public. LZ is the current flagship experiment of the UK in the area of direct dark matter searches. The sensitivity reach of LZ has the potential to produce a world-leading result, which will be transformational in the case of a WIMP discovery. Even if a null result is returned, the additional sensitivity reach of LZ will cover a significantly enlarged parameter space compared to predecessors. Rare-event experiments require exquisitely a well-calibrated detector as well as extremely accurate simulations tuned on existing calibration data, to model the detector response. The calibration data will provide a rich repository of detector events, representing an excellent training set for the data extraction and reduction software. The algorithms performing the identification of individual photons and their attribution to scintillation / ionization events will have to be very robust. A crucial step towards this is the extraction and identification of single photons from the pulse trains that the digitizers record. Algorithms to do this exist and have been tested at a level suitable for past and running experiments, however, further improvements are possible by incorporating new techniques. The main aim of this project is to optimize elements of the data analysis chain in LZ. For this to succeed, detailed simulations will be carried out to provide an optimal understanding of the detector response to calibration sources. This is in preparation of the experiment using a suite of internally dispersed (83mKr, 131Xe, 220Rn and CH4 labelled tritium) and externally deployed sealed neutron (AmLi, 205,206BiBe, 252Cf, 88YeBe) and gamma (57Co, 22Na, 133Ba, 228Th) radiation sources to provide high statistics calibration of the response to background events and WIMP signal. The project objective is to simulate events from neutron and gamma sources used in the LZ experiment and to reconstruct events' position and energy spectra, which is a key part of the detector calibration. Improving on photon detection, reconstruction and triggering effort will involve generating simulations of particular events in a xenon time projection chamber (re-using calibration events, for example), events from data recorded in previous experiments and using all possible information that is available in order to provide and validate suitably enhanced algorithms.The novel aspect in the approach in this thesis is the exploration of the benefit that developing and applying artificial intelligence, e. g. machine-learning, may have. If successful, this would be a major step forward towards dealing with the ever increasing sophistication of dark matter detectors and the data these produce. The project is carried out within the LZ collaboration of which Oxford is a member. The collaboration comprises (in 2018) 38 research groups with over 250 members. The project is computer based, uses GEANT-based simulations, the ROOT data analysis package and the LZAp analysis framework.