General features for the report on a
"Science Case for Large-scale Simulation"
to be prepared by end of July 2003


The report will be written hierarchically on four levels:

	- high-level statements, or "sound-bites," for public statements,

	- an executive summary for government staffers,

	- well documented and contextualized short articles (chapters) at
the level of Scientific American or the science section of the Economist
for scientific peers and technical managers outside of your immediate
area, and

	- appendices and pointers to the relevant professional literature at
a level suitable for peer review.

A subset of the report will be published and circulated widely in bound
hardcopy form.  The full report will be published on the web.  To
supplement and highlight the text, many charts and visualizations will be
included, as well as a few sidebars that emphasize personal or historical
aspects of computational science.

Not every application or enabling technology covered in the workshop will
appear in the bound version, because the report must be kept to a
reasonable size to make it attractive and inviting.  The topics that help
make the most compelling case for the opportunities represented by a
next-generation large-scale computing facility will be favored for
selection for the bound version.

Many reports on the requirements and opportunities for modeling and
simulation within specific application areas (e.g., genomics, nanoscience,
fusion reactor design) have been published recently by the DOE and other
agencies. These will be mined by workshop participant-authors for their
applicable content.  A major goal of the workshop will be to identify
thresholds of capacity and performance that permit qualitatively new
science to be done (e.g., replacing averaged constitutive quantities in
turbulent transport with first principles models where appropriate,
removing restrictive assumptions (such as "quenching" in lattice quantum
chromodynamics), representing full three-dimensional time-dependent
effects where needed).

While one can with some confidence and care extrapolate current
computational techniques to next generation platforms, another level of
breakthrough is the development of new algorithms and tools from
mathematics and computer science to remove bottlenecks in a more
fundamental way.  Bottlenecks that invite such an attack will be
identified.

The report will recognize that adding computing hardware in isolation is
not a balanced approach.  Fundamental research on algorithms and tools to
enable a distributed scientific community to create and interact with data
represents as much of a scientific opportunity as new hardware in many
fields.  Furthermore, many fields need to improve and increase their
opportunities for training the next academic generation in computational
matters.

The report will not seek to be architecturally prescriptive.  However,
some features of contemporary high-end computer architecture, such as
physically distributed memory, and the ability to do floating-point
computation faster than the ability to move the operands in and out of
registers, are inevitable.  The implications of architecture on
algorithmic performance will be considered and a focus on architectural
features whose improvement would most benefit various applications will be
sought.

For each application reported, metrics are sought that are more meaningful
than floating point operation rates.  The benefits of large-scale
computing should be quantified in this report in application-specific
metrics that communicate directly about the science achievable, such as
the number of simulation years per computational day for a climate
simulation of a given effective resolution, or the number of simulated
picoseconds per computational day for protein folding with a given
fidelity of force law.  The benefits of adaptive discretizations and
optimal algorithms should be factored into these metrics, even if they
reduce the sustained floating-point rate while improving running times.

In this report, we are not required to produce comprehensive and realistic
estimates of the overall needs of the computational science community for
hardware resources of given types or to attempt to prioritize their
allocation, but only to make a compelling scientific case for more
computing in a wide variety of areas.  However, we will point out the
distinctions between capacity and capability use, between research and
production, and the value of all.

No such report is delivered into a vacuum.  The way this report updates
and builds upon a legacy of such reports dating back to the "Lax Report"
of 1982 and the "Grand Challenges" of the late 1980s and early 1990s,
tracking the development and evolution of the NSF Supercomputer Centers
and the Accelerated Strategic Computing Initiative, and including the more
recent PITAC and Cyberinfrastructure reports will be described, and the
unique contributions of the current report pointed out.  The history of
scientific accomplishment made possible by previous investments in
large-scale computing facilities will be mentioned, as part of building
the case for future investments.