Optical Cavity Stabilisation to Study Macroscopic Quantum Mechanics

引力波探测器设计的不断改进导致许多以前占主导地位的经典噪声源逐渐被抑制。这一领域的未来发展以及下一代探测器的构建将导致量子噪声缓解变得至关重要。主要的量子噪声源--在低频占主导地位的辐射压力噪声和在高频占主导地位的散粒噪声--已经在现有的探测器中引起了兴趣，a LIGO受到200赫兹以上频率的散粒噪声的限制。可以通过调整输入激光功率来移动合并的量子噪声频谱，并且对于任何给定的输入功率，合并频谱最小化的点的轨迹被称为标准量子极限(SQL)。作为最高灵敏度的点，引力波探测器的发展特别令人感兴趣，但这一点还没有达到。这项研究旨在设计一个桌面实验，以达到在SQL上理解物理的长期目标。这将通过构造一个由多个隔离级悬挂的低温光学腔来实现，以分别抑制热噪声和地震噪声。这样的腔需要在其对准和温度方面都稳定下来。因此，本研究特别着眼于以一种可行且可控的设计来实现所需的噪声抑制。达成SQL将为科学界提供基本的和实际的好处。在SQL上的实验将允许对基本物理的更好的理解，而在达到它的过程中开发的技术可能会对改进的引力波探测器的设计者感兴趣。引力波探测所需的高精度测量只能通过非常严格地控制每个光学部件在探测器中的准确位置来实现。组件使用许多反馈和前馈回路锁定到位，可以自动调整与工作点的偏差。然而，这些回路依赖于使部件足够接近工作点，因为控制回路不具有高动态范围。为了实现这种初始的、近似的对准(以及每当组件漂移到控制回路范围之外时的重新对准)，需要用眼睛对准组件。这可能是一个耗时的过程，还需要手动输入，并导致探测器的停机时间增加。探测器中的光束形状在几个输出端口上成像，然后可以由人类观察者查看并进行校正。这项研究试图通过使用机器学习算法使这一过程自动化。像aLIGO这样的探测器有许多悬浮的光学元件，可以产生许多自由度。由此产生的波束形状是复杂的，这意味着只有像人类大脑这样的高级模式识别系统才有能力有效地使用这些波束。随着近年来机器学习作为一种图像分类和模式识别工具的发展，我们相信这个复杂的问题是可以解决的。通过自动化这一过程，我们希望实现更快的重新校准，从而减少探测器的停机时间。这样做的直接好处是提高了性价比，因为即使在探测器本身不起作用的情况下，现场的运营也会继续产生巨额费用。作为进一步的改进，这可能导致多个探测器的一致观测时间增加。这将允许从同时的引力波探测中更好地提取信息

Continual improvements in gravitational wave detector design have led to a gradual suppression of many, previously dominant, classical noise sources. Future developments in this field as well as construction of next generation detectors will lead to quantum noise mitigation becoming essential. The primary quantum noise sources-radiation pressure noise, dominant at low frequencies and shot noise, dominant at high frequencies-are already of interest in existing detectors, with aLIGO being limited by shot noise at frequencies above 200 Hz. The combined quantum noise spectrum can be shifted by tuning of the input laser power and the locus of points where the combined spectrum is minimised for any given input power is called the standard quantum limit (SQL). Being the point of maximum sensitivity, it is of particular interest in the development of gravitational wave detectors but this point has not been reached yet. This research aims to design a tabletop experiment for reaching the SQL with a long term goal of understanding physics at the SQL. This will be achieved by constructing a cryogenic optical cavity, suspended by multiple isolation stages, in order to suppress thermal noise and seismic noise respectively. Such a cavity needs to be stabilised both in terms of its alignment and temperature. As such, this research particularly focuses on achieving the desired noise suppression with a feasible and controllable design. Reaching the SQL will provide both a fundamental as well as practical benefit to the scientific community. Experiments at the SQL will allow for an enhanced understanding of fundamental physics, whilst the techniques developed in reaching it may be of interest to designers of improved gravitational wave detectors.The high precision measurements needed for gravitational wave detection can only be achieved through very tight control over the exact position of each optical component in the detector. The components are locked in place using many feedback and feedforward loops, which can automatically adjust for deviations from the operating point. These loops, however, rely on the components being brought close enough to the operating point as the control loops do not have a high dynamic range. To achieve this initial, approximate alignment (and realignment whenever the components drift outside of the range of the control loops), the components are aligned by eye. This can be a time consuming process, which also requires manual input, and results in increased down-time for the detectors. The beam shape in the detector is imaged at several output ports, which can then be viewed by a human observer and corrected. This research seeks to automate this process via the use of machine learning algorithms. A detector like aLIGO has numerous suspended optics that lead to many degrees of freedom. The resulting beam shape is thus complex, which means that only advanced pattern recognition systems like the human mind have the ability to use these effectively. With the recent growth of machine learning as an image classification and pattern recognition tool, we believe that this complex problem can be solved. By automating this process, we hope to achieve faster realignment and thus a reduction in detector down-time. This has the immediate benefit of increasing value for money, as the operation of the site continues to incur significant expenses even when the detector itself is inoperative. As a further improvement, this could lead to increased coincident observing time for multiple detectors. This would allow for much better extraction of information from simultaneous gravitational wave detection