CDI-Type I- Bringing a Bayesian perspective to the study of large earthquakes and their impacts on the built environment

这项建议侧重于将全贝叶斯分析技术应用于地震物理和地震工程中的大型不适定反问题，特别是大地震的运动学和动力学，以及监测建筑环境的结构健康。地球物理学和地震工程学都在经历观测爆炸式的增长。使用这些丰富的数据通常涉及解决需要大量不确定参数的大型病态逆问题，这些参数是现实正演模型所必需的。该项目的核心是应用贝叶斯方法解决比目前可处理的问题大得多的问题所需的必要算法和软件开发。特别是，该项目针对涉及非常快的向前模型但具有数百个免费参数的问题，以及具有较少参数但较昂贵的向前模型的问题。研究人员将开发和应用高效、通用和健壮的技术来采样适合于大型并行计算架构的模型参数的后验分布。而这个项目是由特定的学科目标驱动的。地球物理和工程应用程序将作为试验台，旨在确保新开发的算法和相关软件实施的模块化、灵活性和广泛适用性。这项提案的三年期间的方法将是多管齐下的，目的是开发贝叶斯统计反演计算工具的若干重要目标。这些目标包括(1)开发探索高维模型参数空间的新方法，(2)开发自适应和优化模型参数空间的新方法，(3)探索在参数空间上的后验分布中诱导稀疏性的作用，以及(4)为下一代并行平台开发模块化、可移植的软件。智力优势：这项建议将地球物理学家和工程师聚集在一起，构建下一波贝叶斯分析环境，以处理基于大量数据的高维病态反问题。这一努力将导致(1)整合所有相关观测的新一代物理上合理的地震动力学模型，以及(2)推断建筑环境结构健康的革命性方法。

This proposal focuses on application of full Bayesian analysis techniques to large ill-posed inverse problems in earthquake physics and earthquake engineering specifically the kinematics and dynamics of large earthquakes, and monitoring the structural health of the built environment. Both geophysics and earthquake engineering are experiencing an explosion of observations. Using this wealth of data typically involves solving large ill-conditioned inverse problems requiring large numbers of uncertain parameters necessary for realistic forward models. The core of this project centers on the necessary algorithmic and software developments required for application of Bayesian approaches to problems much larger than are currently tractable. In particular, the project targets problems involving very fast forward models but with many hundreds of free parameters as well as problems with fewer parameters but more expensive forward models. The researchers will develop and apply efficient, general, and robust techniques for sampling posterior distributions of model parameters that are suited for large parallel computing architectures. While this project is motivated by specific disciplinary goals. The geophysical and engineering applications will act as test beds designed to ensure modularity, flexibility, and broad applicability of newly developed algorithms and associated software implementations. The approach over the three years of this proposal will be multi-pronged with several important objectives for the development of computational tools for Bayesian statistical inversions. These objectives include (1) developing new methods for exploring high-dimensional model parameter spaces, (2) developing new methods for adaptive and optimal refinement of model parameter spaces, (3) exploring the role of inducing sparseness in posterior distributions over parameter spaces, and (4) producing modular, portable software for the next generation of parallel platforms. Intellectual Merit: This proposal brings together geophysicists and engineers to construct the next wave of Bayesian analysis environments to treat high-dimensional ill-conditioned inverse problems based on large amounts of data. This effort will result in (1) a new generation of physically reasonable models of earthquake dynamics that integrate all relevant observations and (2) revolutionary approaches to inferring structural health of the built environment.