Autonomous hierarchical adaptive mesh refinement for multiscale simulations
Michael T. Heath, Michael L. Norman, Henry Neeman · 1996
Modern high resolution numerical simulations of multiscale physical phenomena require enormous computer resources; however, these resources are largely wasted on subdomains whose solutions do not require such high resolutions. Adaptive mesh refinement (AMR) addresses this problem by providing a means to perform high resolution computation only in areas that require it. In the AMR strategy discussed, a nested heirarchy of overlaying grids of increasingly fine resolution --- in both space and time --- permits high resolution computation in some areas and low resolution in others, either as a set of virtual grids, each encompassing the entire domain, or as a means of zooming in on a subdomain of interest. However, this AMR strategy is both subtle and cumbersome to code, and its data requirements are difficult to manage in a general way. To address this shortcoming, the Hierarchical Adaptive Mesh Refinement (HAMR) system provides support not only for AMR, but also for autonomous data management, thereby decoupling the numerical techniques of a simulation from the adaptive grid hierarchy to which it it applied. (Only available at NCSA Orders, UIUC, 605 E. Springfield Ave., Champaign (217) 244-4130, [email protected].)