Parallel implementation of the adaptive mesh
refinement (AMR) technique in the framework of the fully-implicit Jacobian-Free Newton-Krylov (JFNK)
solver
Presented
by:
Andrzej A. Wyszogrodzki
Postdoc of EES-2 and IGPP
Abstract:
The numerical model developed at LANL, a compressible
hydrodynamic model, HIGRAD is utilized to study extreme atmospheric phenomena
such as hurricanes or wildfires. To be useful for the fire-fighting and meteorological
community, this model requires considerable speedup of the
real time computations and increase the accuracy. We solve for all model
variables in a fully implicit and nonlinearly consistent manner to achieve
second-order in time accuracy by using nonlinear JFNK method. The efficiency of
this system is increased by developing a physics based preconditioner,
which uses the semi-implicit method to solve an approximate form of the governing
equations.
One of the crucial aspects of many atmospheric phenomena is the accurate
tracking of weather parameters (such as winds, temperature, or moisture
content) over the region where they directly affect/control the weather system.
We have implemented an efficient parallel AMR package (PARAMESH) in the JFNK
framework. This technique allows us to zoom in the model resolution on the
interesting area and to move this high resolution region from one location to
another, thus allowing the accurate tracking of the atmospheric parameters
using high resolution only where required. In the case of simple test problem,
the details of AMR implementation, the model algorithms performance and
parallelization aspects, and finally the accuracy of the solution will be
discussed.