Loading
The short-term crustal dynamics workshop was held July 11-14, 2005, in Los Alamos, NM. The 43 workshop participants included graduate students, postdocs, and researchers from both academic institutions and government labs. The workshop included talks on modeling codes, software for mesh generation, software for characterization of geologic structure, and tutorials and demos of modeling codes. We also further refined our priorities and plans for software development related to CIG.
Researchers in the short-term crustal dynamics working group use a variety of software packages to do their modeling. In general, the workflow begins with construction of a geologic model, followed by generation of a finite-element mesh, preprocessing the mesh, running a simulation, post-processing the data, and visualization of the results. In addition to developing the physics codes, the working group has identified two areas (preprocessing and postprocessing) where CIG software development would benefit the community.
Listed in order of priority.
The working group has found that a lack of infrastructure to manage and manipulate simulation results is hindering its benchmarking efforts. A repository is needed to archive the benchmark results, which include (1) simulation metadata (name and version of simulation code, benchmark name, who ran the simulation, etc), (2) simulation output data files, and (3) discretization information (basis functions and element shape). A small group of individuals within the working group are spearheading this effort and will submit a proposal to the SSC with a request for resources to develop the necessary infrastructure.
All of the finite-element codes used by the working group would benefit from tools that aid in efficiently transfering data from mesh generation software to simulation codes. We would like to work with CIG developers and other working groups to identify a common data structure and/or file format to hold mesh topology information (nodes, elements, groups of nodes and elements) and outline the necessary routines to manipulate them. These tools include (1) partitioning the mesh topology information among processors, (2) uniform global mesh refinement, and (3) modules for adjusting the topology to implement faults. While the first two are independent of the code used in the simulation, because codes implement faults in different ways, the modules for implementing faults would be specific to different physics codes.
For the second year in a row, workshop participants struggled to install software using the Pyre build procedure. However, eventually most people were able to install PETSc, Pyre, and Lithomop and run one or more examples. The working group expressed a strong desire for a user friendly, robust build procedure capable of handling software dependencies. Users would like something as easy as installation via packages (apt, fink, up2date, yum, etc).
Release of a fully functioning version of PyLith remains a high priority for the working group. The three priorities above are consistent with and essential to the development of PyLith. Charles Williams and Brad Aagaard indicated that they plan to release stable versions of Lithomop and EqSim in the late spring of 2006 with an initial release of PyLith (a merged version of these codes) in early fall of 2006. Users placed high priority on documentation, regression testing, and modularity (e.g., use of the Pyre simulation controller to allow customization of high level functionality).
Lithomop, EqSim, and PyLith will likely adopt the interface used by Brad Aagaard's spatialdata package for setting parameters defined over some spatial domain. There does not appear to be an immediate need for a more optimal implementation of the routines. As a result, the working group recommended communication with other working groups and the CIG developers about defining a standard interface for this functionality. Any further development of the spatialdata software will be done as part of PyLith development.
The working group would like to see integration of some analytic and semi-analytic codes into the Pyre framework, but places higher priorities on development of PyLith and general simulation. This task should not require more than a couple weeks collaboration between a CIG developer and a geophysicist and could be part of a project for a graduate student or postdoc.
Adaptive mesh refinement would enhance the capabilities of the short-term crustal dynamics codes. The development version of GeoFest (JPL) already uses the Pyramid package developed by Charles Norton (JPL). Further investigation is needed to determine the compatibility of the Pyramid and GeoFest licenses with CIG software development. Both mesh refinement and coarsening would allow construction of more optimal meshes for modeling the postseismic deformation following earthquakes, thereby reducing computation time or allowing solution of larger problems.
The short-term crustal dynamics working group would like to leverage improvements in PETSc solver capabilities and other advances in solution algorithms to improve the physics codes. A year from now, PyLith should be in sufficient shape to experiment with such improvements. Use of a multigrid solver may be one way to improve the elastodynamics solvers.
Extend modeling codes to allow simulation of multiple earthquake cycles while resolving the stress buildup, dynamic strain release with radiated seismic waves, and postseismic deformation.
Extend code coupling infrastructure to allow modeling codes associated with short-term crustal dynamics to be coupled with codes from the other CIG working groups. This would help eliminate artificial boundary conditions and over-simplification of initial conditions.
The modeling codes should add the ability to form the adjoint problem so that resolution and sensitivity can be quantified.
In order to take advantage of data gather efforts such as Earthscope, modeling codes should include data assimilation tools.
Extend codes to allow large deformations. Alternatively, codes used by the long-term crustal dynamics working group could be extended to include faulting (dislocations across a surface). Clearly, any efforts in this area should be coordinated between the short-term and long-term crustal dynamics working groups.