Complex, time-evolving systems are ubiquitous in nature and society, with examples ranging from the Earth's weather and climate, to the function and dynamics of biomolecules, and the behavior of markets and economies. Despite their apparent complexity, many such systems exhibit a form of underlying organized structure (``building blocks''), whose discovery would enhance our ability to understand and predict a wide range of phenomena. The goal of this project is to develop the next generation of mathematical and algorithmic tools that can harness the information content of large datasets acquired from experiments and observations to create coherent representations of complex systems, and use these representations to perform prediction, and ultimately, control. These objectives will be addressed through a novel combination of mathematical techniques, bridging dynamical systems theory and differential geometry with machine learning and data science. The newly developed techniques will be tested and applied in real-world problems through collaboration with domain experts in the areas of climate dynamics, space physics, and condensed matter physics. The project will also contribute to STEM workforce and curricular development through training of students and postdoctoral researchers, and design of multi-disciplinary lecture courses. In particular, this project will support one graduate student at each of the three universities involved.
The modern scientific method is undergoing an evolutionary change wherein large data sets and machine learning algorithms have the potential to outperform classical first-principles approaches for certain complex phenomena. For these tools to be accepted by the scientific community, a rigorous mathematical framework is required to match the verifiability and quantifiability of the classical modeling approach. Recently, a new tool called the diffusion forecast has been developed based on provably consistent estimators, which learn the unknown structure of a large class of stochastic dynamical systems on manifolds. Moreover, the results of many published numerical experiments indicate that this framework can be applied far beyond the restricted context of the current theory. In particular, the evidence suggests that the consistency proofs can be extended to non-autonomous projections of complex systems, deterministic chaotic systems represented by non-compact operators, non-smooth domains such as fractal attractors, and even generalized tensors on metric-measure spaces. This project will undertake a rigorous mathematical unification of these problems, leading to transformative advances in our ability to model and describe complex systems.
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.