Dynamical Inference of Forces in Dusty Plasmas using Three-Dimensional Laser Sheet Tomography

该奖项将使复杂的“尘埃”等离子体的新研究使用机器学习技术从实验室实验的数据。一些最紧迫的科学和社会问题涉及复杂系统的动态，如地球的气候变化，全球金融市场，城市环境和动物的迁徙。这些系统有许多相互作用的组件，但目前还不清楚它们是如何纠缠在一起产生的动态。在新兴的大数据时代，这些系统及其组件可以非常好地测量，从而产生大量的数据。 为了更深入地理解这些系统，动力学推理已经成为物理学和计算机科学中的一个重要领域。然而，许多国家的最先进的计算技术，挑逗除了复杂的动力学模拟数据进行测试。该奖项将把这一新兴领域的联合收割机机器学习技术与来自尘埃等离子体实验室实验的高分辨率数据相结合，尘埃等离子体由嵌入电子和离子等离子体中的悬浮的微米级尘埃颗粒组成。通过同时跟踪数百个粒子的运动，驱动它们运动的基本非平衡力将使用人工智能技术来“学习”，探索物理，人类直觉和物理学之间的边界可以通过机器学习发现。与此同时，该项目还包括在Dekalb县当地一所小学继续举办课外科学俱乐部，Dekalb县拥有第三大学校系统，是格鲁吉亚最多样化的县。尘埃粒子在等离子体中的运动已经研究了近30年，并导致发现了强耦合的晶体结构和受重力，流体动力学，和电磁力许多粒子间的力是复杂的;它们可以是非互易的，也可以是非相加的。破译许多粒子之间潜在相互作用的混杂是一项具有挑战性的任务，并且可以在极端环境中打开对尘埃等离子体的理解的新范式。该奖项旨在利用高分辨率、三维成像和粒子跟踪以及现代动力学推理技术，将驱动尘埃等离子体的已知和未知力量区分开来。这将有效地将每个尘埃粒子变成等离子体环境的局部探针，并可应用于各种实验设置。为了实现这一目标，将开发一种新的扫描激光片层层析成像技术，该技术利用单个高速摄像机。在射频等离子体中，在偏置电极上方悬浮的约100个或更少粒子的系统中，单个粒子的运动将被跟踪30 s或更长时间。对它们运动的详细分析将揭示有关等离子体环境的原位信息，包括等离子体鞘层和粒子上电荷的局部随机波动。对于许多粒子的高度动态系统，将使用各种机器学习技术分析长时间采集的大量数据。该团队将计算粒子之间的有效对势等量，更复杂的三体相互作用和非互易力将使用具有不同层次的神经网络来提取。更广泛地说，该奖项旨在测试机器从嘈杂的实验数据中“学习”的极限，并从复杂的大数据中探索新的物理学。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

This award will enable novel studies of complex "dusty" plasmas using machine learning techniques on data from laboratory experiments. Some of the most pressing scientific and societal issues involve the dynamics of complex systems, such as Earth's variable climate, global financial markets, urban environments, and the migration of animals. These systems have many interacting components, yet it is unclear how they are entangled together to produce the resulting dynamics. In the emerging era of Big Data, these systems and their components can be measured exceedingly well, resulting in a vast amount of data. In the pursuit to provide a deeper understanding of such systems, dynamical inference has emerged as an important field in physics and computer science. However, many state-of-the-art computational techniques for teasing apart complex dynamics are tested on simulated data. This award will combine machine learning techniques from this emerging field with high-resolution data from laboratory experiments of dusty plasmas, which consist of levitated, micron-sized dust particles embedded in a plasma of electrons and ions. By simultaneously tracking the motion of hundreds of particles, the fundamental, nonequilibrium forces that drive their motion will be "learned" using Artificial Intelligence techniques, exploring the boundary between physical, human intuition and the limits of what physics can be uncovered through machine learning. In parallel to this research effort, the project includes continuation of an after-school science club at a local elementary school in Dekalb county, which hosts the 3rd largest school system and is the most diverse county in Georgia.The motion of dust particles in plasma has been studied for nearly 30 years and lead to the discovery of strongly coupled crystalline structures and nonequilibrium dynamics governed by gravitational, hydrodynamic, and electromagnetic forces. Many of the interparticle forces are complex; they can be non-reciprocal, and non-additive. Deciphering the mélange of potential interactions between many particles is a challenging task, and could open a new paradigm of understanding in dusty plasmas in extreme environments. This award aims to tease apart both the known and unknown forces that drive dusty plasmas using high-resolution, three-dimensional imaging and particle tracking coupled with modern dynamical inference techniques. This will effectively turn every dust particle into a local probe of the plasma environment and can be applied in a diverse array of experimental settings. To accomplish this, a new scanning laser-sheet tomography technique will be developed that utilizes a single high-speed camera. In systems of ~100 particles or less levitated above a biased electrode in an rf plasma, the motion of single particles will be tracked for 30 s or longer. A detailed analysis of their movement will reveal in-situ information about the plasma environment, including the plasma sheath, and the local stochastic fluctuations of charge on the particles. For highly dynamic systems of many particles, the large quantity of data acquired over long times will be analyzed using a variety of machine learning techniques. The team will compute quantities such as the effective pair potential between particles, and more complex, three-body interactions and non-reciprocal forces will be extracted using neural networks with varying hierarchies. More broadly, this award aims to test the limit of what a machine can "learn" from noisy, experimental data, and explore new physics from complex, big data.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.