Strong gravitational lensing in the era of wide-area sensitive surveys

在接下来的十年里，将见证一场使用强引力透镜作为灵敏探测器的革命，以解决宇宙学和河外天体物理中的许多公开问题。这是由于即将到来的广域敏感成像调查时代的到来。这样的调查对于发现非常罕见的强引力透镜是必不可少的。尽管在20世纪70年代首次得到了观测证实，但我们只知道大约1000个透镜。这种稀有意味着我们不能充分利用它们有希望的用途。DPhil项目的重点是在维拉·C·鲁宾天文台以及欧几里德和南希·A·罗曼太空望远镜进行的调查中发现100,000量级的强引力透镜的大样本。因此，这是一项必要的工作，以最大限度地发现强引力透镜，为社区提供发现高完整性和高纯度的大样本的手段。强引力透镜发现的突出中心问题是高误报率--其中包括背景恒星形成的星系与前景大质量椭圆和高红移螺旋线模拟透镜弧线的偶然性对齐。到目前为止，有监督的机器学习算法产生的样本是高度不纯的(通过几个因素)。因此，人眼检查仍然是提高纯度的唯一手段，但这是劳动密集型的。该项目包括对发现算法的创新使用，以及由公民科学家进行的群众来源视觉检查。这利用了非常成功的Zooniverse项目Space Warps。作为Space Warps的联合创始人和-Pi，我们处于领导这项工作的独特位置。这名学生最初将在现有的广域勘测上运行空间扭曲辅助发现系统，作为即将到来的大区域勘测的前兆。学生将探索机器和目视检查之间的联系。我们将专注于扩展用于机器学习网络的训练样本，因为这是此类网络目前的性能限制，以及算法和公民科学平台之间主动学习环路的构建。此外，我们将通过建立候选人排名的模型，探索整合从候选人的光度和成像中获得的增强信息。在后续时间有限的情况下，排序是至关重要的，以确认和挑选用于科学驱动分析的子样。我们将研究源的子样本，例如，限制z~0.5-1.5星系的质量分布和强引力透镜提供的高空间分辨率下的高红移星系的性质。

The next decade will witness a revolution in the use of strong gravitational lenses as sensitive probes to address many open problems in cosmology and extragalactic astrophysics. This is enabled by the forthcoming era of wide-area sensitive imaging surveys. Such surveys are essential for the discovery of the very rare strong gravitational lenses. Despite being first observationally confirmed in 1970s, we only know of ~1000s of lenses. This rarity has meant we cannot fully exploit their promising uses. This DPhil project is focussed on the discovery of large samples of strong gravitational lenses, of order 100,000s, in surveys carried out by the Vera C. Rubin Observatory, and the Euclid and Nancy A. Roman Space Telescopes. As such, this is a necessary body of work to maximise the discovery of strong gravitational lenses, providing the community with means to discover large samples of high completeness and purity. The outstanding central problem in strong gravitational lens discovery is the high rate of false positives - among these are e.g., chance alignments of background star forming galaxies with foreground massive ellipticals and high redshift spirals mimicking lensed arcs. To date, supervised machine learning algorithms yield samples that are highly impure (by factors of several). As a result, human visual inspection remains the only means to improve the purity, but it is labour intensive. This project includes an innovative use of discovery algorithms coupled to crowd sourced visual inspection by citizen scientists. This capitalises on the highly successful Zooniverse project Space Warps. As co-founder and -PI of Space Warps, we are uniquely placed to lead this work. The student will initially run Space Warps assisted discovery systems on existing wide area surveys as pre-cursors to the forthcoming large area surveys. The student will explore the connections between machines and visual inspection. We will focus on expansion on the training samples used for machine learning networks as this is a current limitation in the performance of such networks, along with the construction of active learning loops between the algorithms and the citizen science platform. In addition, we will explore integration of enhanced information derived from the photometry and imaging of the candidates through modelling to rank of the candidates. Ranking is vital in the regime of limited follow-up time for confirmation and cherry picking of sub-samples for science driven analysis. We will study sub-samples of the sources particularly to e.g., constrain the mass distributions of z~0.5-1.5 galaxies and the properties of high-redshift galaxies at high spatial resolution afforded by strong gravitational lensing.