Table of Contents
Paragraph introducing the use Pyre in the cases below.
The GeoFramework project is developing a suite of tools to model multi-scale deformation for Earth science problems. This effort is motivated by the need to understand interactions between the long-term evolution of plate tectonics and shorter term processes such as the evolution of faults during and between earthquakes. The development of this tool is timely as it will form an integral component of new NSF initiatives, especially EarthScope and MARGINS, that will generate tremendous amounts of data, the interpretation of which will require a multi-scale modeling approach. The proposed modeling suite will handle complex rheologies, multiple scales in time and space, and will be modular so users can add their own contributions.
Efforts will extend the capabilities of the Pyre framework, an object-oriented environment capable of specifying and launching numerical simulations on multiple platforms, including Beowulf class parallel computers, that can function with grid-computing systems. Advances to Pyre will allow us to bind together multiple components, including physics models used in Earth science simulations, such as solid and fluid models, and different meshers.
Initially, a number of modeling codes will be incorporated into the geodynamics framework, including a seismic wave propagation code, a mantle convection code, and several solid modeling codes, which deal with visco-elastoplastic materials. Some of these codes will require modification to deal with many of the features envisaged. A variety of meshers will work within the framework simultaneously. Dynamic regridding and dynamic load balancing techniques will be employed to achieve the design goals. One of the many uses for these tools will be to investigate the role of initial conditions in geodynamical problems. Such investigations require us to repeat large calculations many times, which motivates us to develop the software for loosely coupled distributed platforms, including Beowulf-class machines.
This will be a collaborative project between two Caltech units: the Center for Advanced Computer Research (CACR) and the Seismological Laboratory. CACR will hold open workshops, the first being a forum to gather additional input on the features and requirements of the new geodynamics modeling framework.
Beta versions of the software will be made available to the community and final versions will be available without restrictions. The framework will easily allow one to launch 2-D and 3-D (Cartesian and Spherical) simulations on platforms ranging in scale from uniprocessors to the TeraGrid, making the code useful to a wide range of users from individual scientists and educators to interdisciplinary teams working on state of the art calculations posed within EarthScope.
DANSE is a set of Python language scripts and modules for the analysis of neutron scattering data. In essence, DANSE is a part of the Python language, although of course it is not part of the standard Python distribution.
DANSE provides a rich set of tools for many standard operations on neutron scattering data, such as conversion of time-of-flight data into physical distributions in momentum and energy, background corrections, deconvolutions, corrections for multiple scattering and multiple excitations, and data visualization. In addition to this "traditional" analysis, DANSE offers new capabilities for working with models of structure and dynamics.
A number of working scripts have been tested and documented, and these are available for immediate use in many standard types of operations on neutron scattering data. Perhaps more importantly, the DANSE architecture is flexible, allowing new combinations of the existing modules and minimizing the barrier to developing new and compatible modules. This permits minor changes of existing data analysis procedures to accommodate experiments that differ slightly from previous ones, or major changes to accommodate entirely new experiments or types of data analyses.
We propose to develop a national capability for the analysis of data acquired at neutron scattering facilities. The development of such a facility needs to begin today if data from the SNS are to be analyzed in 2006. The architecture is based on distributed computing, and is ideally suited to hardware architectures on highperformance networks such as the TeraGrid project supported by the NSF, in which Caltech is an active participant. Users will direct the processing of their data through a web browser by accessing a central server, which distributes data and computing jobs to the appropriate resources. An obvious advantage is that the user interface is consistent for all neutron instruments, shortening the learning curve for users, and encouraging the testing of ideas for new experiments or assessing differences between instruments. The underlying architecture that makes this possible is based on interchangeable, modularized, software components that operate on data streams. The components themselves encapsulate a core code that could be a legacy code or a new contribution from a scientist. The same data analysis service could be used by all neutron sources in the U.S., and in principle, throughout the world. (from the DANSE paper by Aivazis and Fultz and Experimental Inelastic Neutron Scattering, a book-in-progress by Fultz et. al.)
The Caltech ASCI Center is developing simulations codes for its virtual shock physics test facility. The VTF will ultimately offer a suite of computational applications integrated for performing multidisciplinary simulations.
The Grist project is developing a grid-technology based system as a prototype research environment for astronomy with massive and complex data sets. The knowledge extraction system will be a library of distributed Grid Services controlled by a workflow system compliant with the emerging standards of cyberinfrastructure.