Optimizing Coupled Multimodel Simulations- A Flexible Algorithmic Approach

Abstract: One of the key sources of inaccuracies in coupled multimodel simulations arises from the complex task of mapping solution field data between different discretizations and numerical unstructured grids used across various component models. While the increasing availability of exascale-level computing resources has certainly contributed to improved model accuracy, it has also shed light on the computational bottleneck created by coupling algorithms when dealing with high-resolution, large-scale simulations. For example, achieving precise spatial coupling within any multimodel system involves intricate field transformations and data transmission across grids of varying resolutions. All of this must be done while preserving critical attributes such as global integrals and local features, which have a significant impact on both the accuracy of coupled fields and the stability of solvers. Over the past few decades, significant progress has been made in developing high-order, stable interpolators, and conservative remappers to address the challenges inherent in multimodel simulations.

In this talk, we will introduce a highly adaptable, mesh-based infrastructure for coupling multiple components seamlessly. Importantly, this infrastructure is designed to function independently of discretization methods, programming languages, and decomposition techniques. Additionally, an algorithmic approach for adapting existing serial remappers, which ultimately enhances scientific productivity in Earth System Model (ESM) workflows will also be showcased.

Bio: Vijay Mahadevan is currently a computational scientist at TechTrans International Inc, serving as a consultant to Argonne (MCS). He obtained his PhD in Nuclear Engineering from Texas A&M University and held previous roles as a computational postdoctoral fellow and assistant computational scientist at Argonne until 2018. His research focuses on stable, robust, scalable, and high-order accurate spatio-temporal numerical methods for simulating multiphysics phenomena, particularly in nuclear engineering and climate science.