A Multi-Scale PSE for Earthquake Simulations
We will refine the Garnet collaborative portal to support multi-scale analysis in what a
term a MSPSE - Multi-scale problem solving environment. We suggest that a MSPSE should
allow
a) Specification of all aspects of problem (software, data, results, associated resources)
in a hierarchical fashion that reflects the science scaling.
b) Linkage of simulations and data at different scales where the fine scale work is
abstracted as a law (such as a chemical potential or friction law), which can be used as
input to a larger physical size but coarser scale computation.
c) Hybrid simulations where multiple scales are simultaneously present in a given run with
different space or time segments modeled at the abstraction level needed by the physical
situation.

These are just a first list of characteristics and one research issue will be to refine
and extend these characteristics of a MSPSE. Further any real case involves aspects of all
these characteristics. Again although a strict hierarchy is a nice model it is somewhat
idealized, and one needs to be able to specify multiple hierarchies (for instance
earthquake engineering is often linked to GEM like simulations but this field would be
specified separately from table above). The hierarchical structure of GXOS naturally
supports a) with multiple hierarchies allowed and further one has the ability to use full
XLink/Pointer linkages to "jump" across object hierarchies. Characteristic c) can be
addressed by composite tasks (specified by generalized dataflow) in the Gateway portal
component of GCP. Natural ways of supporting b) and quantifying how best to realize a) and
c) in a PSE will be central research issues. We expect to use XSIL from Caltech to specify
the scientific parameters needed to specify abstractions in a). We already have experience
in this from Gateway.