Professional Documents
Culture Documents
Office of Science
U.S. Department of Energy
On behalf of more than 300 contributing computational scientists from dozens of leading
universities, all of the multiprogrammatic laboratories of the Department of Energy, other
federal agencies, and industry, I am pleased to deliver Volume 1 of a two-volume report
that builds a Science-Based Case for Large-Scale Simulation.
We find herein (and detail further in the companion volume) that favorable trends in
computational power and networking, scientific software engineering and infrastructure,
and modeling and algorithms are allowing several applications to approach thresholds
beyond which lies new knowledge of both fundamental and practical kinds. A major
increase in investment in computational modeling and simulation is appropriate at this
time, so that our citizens are the first to benefit from particular new fruits of scientific
simulation, and indeed, from an evolving culture of simulation science.
Based on the encouraging first two years of the Scientific Discovery through Advanced
Computing initiative, we believe that balanced investment in scientific applications,
applied mathematics, and computer science, with cross-accountabilities between these
constituents, is a program structure worthy of extension. Through it, techniques from
applied mathematics and computer science are systematically being migrated into scientific
computer codes. As this happens, new challenges in the use of these techniques flow back
to the mathematics and computer science communities, rejuvenating their own basic
research. Now is an especially opportune time to increase the number of groups working in
this multidisciplinary way, and to provide the computational platforms and environments
that will enable these teams to push their simulations to the next insight-yielding levels of
high resolution and large ensembles.
The Department of Energy can draw satisfaction today from its history of pathfinding in
each of the areas that are brought together in large-scale simulation. However, from the
vantage point of tomorrow, the Department’s hand in their systematic fusion will be
regarded as even more profound.
Best regards,
David E. Keyes
Fu Foundation Professor of Applied Mathematics
Columbia University
New York, NY
Executive Summary
EXECUTIVE SUMMARY
Important advances in basic science crucial theory, methods, and algorithms from
to the national well-being have been mathematicians; and software and
brought near by a “perfect fusion” of hardware infrastructure from computer
sustained advances in scientific models, scientists. Only major new investment in
mathematical algorithms, computer these activities across the board, in the
architecture, and scientific software program areas of DOE’s Office of Science
engineering. Computational simulation—a and other agencies, will enable the United
means of scientific discovery that employs States to be the first to realize the promise
a computer system to simulate a physical of the scientific advances to be wrought by
system according to laws derived from computational simulation.
theory and experiment—has attained peer
status with theory and experiment in many In this two-volume report, prepared with
areas of science. The United States is direct input from more than 300 of the
currently a world leader in computational nation’s leading computational scientists, a
simulation, a position that confers both an science-based case is presented for major,
opportunity and a responsibility to mount a new, carefully balanced investments in
vigorous campaign of research that brings • scientific applications
the advancing power of simulation to many • algorithm research and development
scientific frontiers. • computing system software
infrastructure
Computational simulation offers to • network infrastructure for access and
enhance, as well as leapfrog, theoretical and resource sharing
experimental progress in many areas of • computational facilities
science critical to the scientific mission of • innovative computer architecture
the U.S. Department of Energy (DOE). research, for the facilities of the future
Successes have been documented in such • proactive recruitment and training of a
areas as advanced energy systems (e.g., fuel new generation of multidisciplinary
cells, fusion), biotechnology (e.g., genomics, computational scientists
cellular dynamics), nanotechnology (e.g.,
sensors, storage devices), and The two-year-old Scientific Discovery
environmental modeling (e.g., climate through Advanced Computing (SciDAC)
prediction, pollution remediation). initiative in the Office of Science provides a
Computational simulation also offers the template for such science-directed,
best near-term hope for progress in multidisciplinary research campaigns.
answering a number of scientific questions SciDAC’s successes in the first four of these
in such areas as the fundamental structure seven thrusts have illustrated the advances
of matter, the production of heavy elements possible with coordinated investments. It is
in supernovae, and the functions of now time to take full advantage of the
enzymes. revolution in computational science with
new investments that address the most
The ingredients required for success in challenging scientific problems faced by
advancing scientific discovery are insights, DOE.
models, and applications from scientists;
VOLUME 1
1. Introduction .............................................................................................................................. 1
ATTRIBUTIONS
Report Editors
Phillip Colella, Lawrence Berkeley National Laboratory
Thom H. Dunning, Jr., University of Tennessee and Oak Ridge National Laboratory
William D. Gropp, Argonne National Laboratory
David E. Keyes, Columbia University, Editor-in-Chief
Workshop Organizers
David L. Brown, Lawrence Livermore National Laboratory
Phillip Colella, Lawrence Berkeley National Laboratory
Lori Freitag Diachin, Sandia National Laboratories
James Glimm, State University of New York–Stony Brook and
Brookhaven National Laboratory
William D. Gropp, Argonne National Laboratory
Steven W. Hammond, National Renewable Energy Laboratory
Robert Harrison, Oak Ridge National Laboratory
Stephen C. Jardin, Princeton Plasma Physics Laboratory
David E. Keyes, Columbia University, Chair
Paul C. Messina, Argonne National Laboratory
Juan C. Meza, Lawrence Berkeley National Laboratory
Anthony Mezzacappa, Oak Ridge National Laboratory
Robert Rosner, University of Chicago and Argonne National Laboratory
R. Roy Whitney, Thomas Jefferson National Accelerator Facility
Theresa L. Windus, Pacific Northwest National Laboratory
Science Applications
Accelerators
Kwok Ko, Stanford Linear Accelerator Center
Robert D. Ryne, Lawrence Berkeley National Laboratory
Astrophysics
Anthony Mezzacappa, Oak Ridge National Laboratory
Robert Rosner, University of Chicago and Argonne National Laboratory
Biology
Michael Colvin, Lawrence Livermore National Laboratory
George Michaels, Pacific Northwest National Laboratory
Chemistry
Robert Harrison, Oak Ridge National Laboratory
Theresa L. Windus, Pacific Northwest National Laboratory
Climate and Earth Science
John B. Drake, Oak Ridge National Laboratory
Phillip W. Jones, Los Alamos National Laboratory
Robert C. Malone, Los Alamos National Laboratory
Combustion
John B. Bell, Lawrence Berkeley National Laboratory
Larry A. Rahn, Sandia National Laboratories
Environmental Remediation and Processes
Mary F. Wheeler, University of Texas–Austin
Steve Yabusaki, Pacific Northwest National Laboratory
Materials Science
Francois Gygi, Lawrence Livermore National Laboratory
G. Malcolm Stocks, Oak Ridge National Laboratory
Nanoscience
Peter T. Cummings, Vanderbilt University and Oak Ridge National Laboratory
Lin-wang Wang, Lawrence Berkeley National Laboratory
Plasma Science
Stephen C. Jardin, Princeton Plasma Physics Laboratory
William M. Nevins, Lawrence Livermore National Laboratory
Quantum Chromodynamics
Robert Sugar, University of California–Santa Barbara
Uncertainty Quantification
James Glimm, State University of New York–Stony Brook and
Brookhaven National Laboratory
Sallie Keller-McNulty, Los Alamos National Laboratory
WORKSHOP PARTICIPANTS
1. INTRODUCTION
Figure 2. Practical strains on experimentation that invite simulation as a peer modality of scientific
investigation.
bring the power of the most advanced types of computational scientists and good
computing technologies to bear on the most communication between them.
challenging scientific and engineering
problems facing the nation. The following recommendations are
elaborated upon in Chapter 6 of this
In addition to the technical challenges in volume:
scientific computing, too few computational
cycles are currently available to fully • Major new investments in
capitalize upon the progress of the SciDAC computational science are needed in all
initiative. Moreover, the computational of the mission areas of the Office of
science talent emerging from the national Science in DOE, as well as those of
educational pipeline is just a trickle in many other agencies, so that the United
comparison to what is needed to replicate States may be the first, or among the
the successes of SciDAC throughout the first, to capture the new opportunities
many applications of science and presented by the continuing advances in
engineering that lag behind. Cycle- computing power. Such investments
starvation and human resource–starvation will extend the important scientific
are serious problems that are felt across all opportunities that have been attained
of the computational science programs by a fusion of sustained advances in
surveyed in creating this report. However, scientific models, mathematical
they are easier hurdles to overcome than algorithms, computer architecture, and
the above-mentioned “curses.” scientific software engineering.
By the time the workshop report was university partnerships in research and
published, the Office of Science was well graduate education. SciDAC is built on
along the path to developing an Office- these institutional foundations as well as on
wide program designed to address the the integration of computational modeling,
challenges identified in the DOE-SC/NSF computer algorithms, and computing
workshop. The response to the National systems software.
Academy report, as well as to the 1999
report from the President’s Information In this chapter, we briefly examine the short
Technology Advisory Committee—the but substantial legacy of SciDAC. Our
Scientific Discovery through Advanced purpose is to understand the implications
Computing (SciDAC) program—was of its successes, which will help us identify
described in a 22-page March 2000 report the logical next steps in advancing
by the Office of Science to the House and computational modeling and simulation.
Senate Appropriations Committees.
Congress funded the program in the SciDAC’S INGREDIENTS FOR
following fiscal year, and by the end of SUCCESS
FY 2001, DOE-SC had established
Aimed at Discovery
51 scientific projects, following a national
peer-reviewed competition. SciDAC boldly affirms the importance of
computational simulation for new scientific
SciDAC explicitly recognizes that the discovery, not just for “rationalizing” the
advances in computing technologies that results of experiments. The latter role for
provide the foundation for the scientific simulation is a time-honored one from the
opportunities identified in the DOE-SC/ days when it was largely subordinate to
NSF report (as well as in Volume 2 of the experimentation, rather than peer to it. That
current report) are not, in themselves, relationship has been steadily evolving. In a
sufficient for the “coming of age” of frontispiece quotation from J. S. Langer, the
computational simulation. Close chair of the July 1998 multiagency National
integration of three disciplines—science, Workshop on Advanced Scientific
computer science, and applied Computing, the SciDAC report declared:
mathematics—and a new discipline at their “The computer literally is providing a new
intersection called computational science window through which we can observe the
are needed to realize the full benefits of the natural world in exquisite detail.”
fast-moving advances in computing
technologies. A change in the sociology of Through advances in computational
this area of science was necessary for the modeling and simulation, it will be possible
fusion of these streams of disciplinary to extend dramatically the exploration of
developments. While not uniquely the fundamental processes of nature—e.g.,
American in its origins, the “new kind of the interactions between elementary
computational science” promoted in the particles that give rise to the matter around
SciDAC report found fertile soil in us, the interactions between proteins and
institutional arrangements for the conduct other molecules that sustain life—as well as
of scientific research that are highly advance our ability to predict the behavior
developed in the United States, and of a broad range of complex natural and
seemingly much less so elsewhere: engineered systems—e.g., nanoscale
multiprogram, multidisciplinary national devices, microbial cells, fusion energy
laboratories, and strong laboratory- reactors, and the earth’s climate.
Figure 3. Relationship between the science application teams, the Integrated Software
Infrastructure Centers (ISICs), and networking and computing facilities in the SciDAC program.
SciDAC recognizes that large scientific latter half of the 1990s (in such programs as
simulation codes have many needs in DOE’s ASCI) to justify them as cost-
common: modules that construct and adapt effective simulation hardware, provided
the computational mesh on which their that certain generic difficulties in
models are based, modules that produce programming are amortized over many
the discrete equations from the user groups and many generations of
mathematical model, modules that solve hardware. Still, many challenges remain to
the resulting equations, and modules that making efficient and effective use of these
visualize the results and manage, mine, and computers for the broad range of scientific
analyze the large data sets. By constructing applications important to the mission of
these modules as interoperable components DOE’s Office of Science.
and supporting the resulting component
libraries so that they stay up to date with Supporting Community Codes
the latest algorithmic advances and work
As examples of simulations from Office of
on the latest hardware, SciDAC amortizes
Science programs, the SciDAC report
costs and creates specialized points of
highlighted problems of ascertaining
contact for a portfolio of applications.
chemical structure and the massively
parallel simulation code NWCHEM, winner
The integrated software infrastructure
of a 1999 R&D 100 Award and a Federal
centers that develop these libraries are
Laboratory Consortium Excellence in
motivated to ensure that their users have
Technology Transfer Award in 2000.
available the full range of algorithmic
options under a common interface, along
NWCHEM, currently installed at nearly
with a complete description of their
900 research institutions worldwide, was
resource tradeoffs (e.g., memory versus
created by an international team of
time). Their users can try all reasonable
scientists from twelve institutions (two U.S.
options easily and adapt to different
national laboratories, two European
computational platforms without recoding
national laboratories, five U.S. universities,
the applications. Furthermore, when an
and three European universities) over the
application group intelligently drives
course of five years. The core team
software infrastructure research in a new
consisted of seven full-time theoretical and
and useful direction, the result is soon
computational chemists, computer
available to all users.
scientists, and applied mathematicians,
ably assisted by ten postdoctoral fellows on
Large Scale temporary assignment. Overall, NWCHEM
Because computational science is driven represents an investment in excess of
toward the large scale by requirements of 100 person-years. Yet, the layered structure
model fidelity and resolution, SciDAC of NWCHEM permits it to easily absorb
emphasizes the development of software new physics or algorithm modules, to run
for the most capable computers available. on environments from laptops to the fastest
In the United States, these are now parallel architectures, and to be ported to a
primarily hierarchical distributed-memory new supercomputing environment with
computers. Enough experience was isolated minor modifications, typically in
acquired on such machines in heroic less than a week. The scientific legacy
mission-driven simulation programs in the embodied in NWCHEM goes back three
decades to the Gaussian code, for which of contact for mathematical and
John Pople won the 1998 Nobel Prize in computational technologies is a sound
chemistry. investment strategy. For instance, Dalton
Schnack of General Atomics, whose group
Community codes such as Gaussian and operates and maintains the simulation code
NWCHEM can consume hundreds of NIMROD to support experimental design
person-years of development (Gaussian of magnetically confined fusion energy
probably has 500–1000 person-years devices, states that software recently
invested in it), run at hundreds of developed by Xiaoye (Sherry) Li of
installations, are given large fractions of Lawrence Berkeley National Laboratory has
community computing resources for enhanced his group’s ability to perform
decades, and acquire an authority that can fusion simulations to a degree equivalent to
enable or limit what is done and accepted three to five years of riding the wave of
as science in their respective communities. hardware improvements alone. Ironically,
Yet, historically, not all community codes the uniprocessor version of Li’s solver,
have been crafted with the care of originally developed while she was a
NWCHEM to achieve performance, graduate student at the University of
portability, and extensibility. Except at the California–Berkeley, predated the SciDAC
beginning, it is difficult to promote major initiative; but it was not known to the
algorithmic changes in such codes, since NIMROD team until it was recommended
change is expensive and pulls resources during discussions between SciDAC fusion
from core progress. SciDAC application and solver software teams. In the
groups have been chartered to build new meantime, SciDAC-funded research
and improved community codes in areas improved the parallel performance of Li’s
such as accelerator physics, materials solver in time for use in NIMROD’s current
science, ocean circulation, and plasma context.
physics. The integrated software
infrastructure centers of SciDAC have a M3D, a fusion simulation code developed
chance, due to the interdependence built at the Princeton Plasma Physics Laboratory,
into the SciDAC program structure, to is complementary to NIMROD but uses a
simultaneously influence all of these new different mathematical formulation. Like
community codes by delivering software NIMROD and many simulation codes, M3D
incorporating optimal algorithms that may spends a large fraction of its time solving
be reused across many applications. sparse linear algebraic systems. In a
Improvements driven by one application collaboration that predated SciDAC, the
will be available to all. While SciDAC M3D code group was using PETSc, a
application groups are developing codes for portable toolkit of sparse solvers developed
communities of users, SciDAC enabling at Argonne National Laboratory and in
technology groups are building codes for wide use around the world. Under SciDAC,
communities of developers. the Terascale Optimal PDE Solvers
integrated software center provided
Enabling Success another world-renowned class of solvers,
Hypre, from Lawrence Livermore National
The first two years of SciDAC provide
Laboratory “underneath” the same code
ample evidence that constructing
interface that M3D was already using to call
multidisciplinary teams around challenge
PETSc’s solvers. The combined PETSc-
problems and providing centralized points
Hypre solver library allows M3D to solve important result: this process can occur in
its current linear systems two to three times environments very different than heretofore
faster than previously, and this SciDAC thought, perhaps providing a way around
collaboration has identified several avenues some of the fundamental difficulties
for improving the PETSc-Hypre code encountered in past supernova models in
support for M3D and other users. explaining element synthesis. To extend
their model to higher dimensions, the TSI
While these are stories of performance team has engaged mathematicians and
improvement, allowing faster turnaround computer scientists to help them overcome
and more production runs, there are also the “curse of dimensionality” and exploit
striking stories of altogether new physics massively parallel machines. Mezzacappa
enabled by SciDAC collaborations. The has declared that he would never go back to
Princeton fusion team has combined forces doing computational physics outside of a
with the Applied Partial Differential multidisciplinary team.
Equations Center, an integrated software
infrastructure center, to produce a Setting a New Organizational
simulation code capable of resolving the Paradigm
fine structure of the dynamic interface
By acknowledging sociological barriers to
between two fluids over which a shock
multidisciplinary research up front and
wave passes—the so-called Richtmyer-
building accountabilities into the project
Meshkov instability—in the presence of a
awards to force the mixing of “scientific
magnetic field. This team has found that a
cultures” (e.g., physics with computer
strong magnetic field stabilizes the
science), SciDAC took a risk that has paid
interface, a result not merely of scientific
off handsomely in technical results. As a
interest but also of potential importance in
result, new DOE-SC initiatives are taking
many devices of practical interest to DOE.
SciDAC-like forms—for instance, the
FY 2003 initiative Nanoscience Theory,
SciDAC’s Terascale Supernova Initiative
Modeling and Simulation. The roadmap for
(TSI), led by Tony Mezzacappa of Oak
DOE’s Fusion Simulation Program likewise
Ridge National Laboratory, has achieved
envisions multidisciplinary teams
the first stellar core collapse simulations for
constructing complex simulations linking
supernovae with state-of-the-art neutrino
together what are today various
transport. The team has also produced
freestanding simulation codes for different
state-of-the-art models of nuclei in a stellar
portions of a fusion reactor. Such complex
core. These simulations have brought
simulations are deemed essential to the on-
together two fundamental fields at their
time construction of a working
respective frontiers, demonstrated the
demonstration International Thermonuclear
importance of accurate nuclear physics in
Experimental Reactor.
supernova models, and motivated planned
experiments to measure the nuclear
In many critical application domains, such
processes occurring in stars. This group has
teams are starting to self-assemble as a
also performed simulations of a rapid
natural response to the demands of the next
neutron capture process believed to occur
level of scientific opportunity. This self-
in supernovae and to be responsible for the
assembly represents a phase transition in
creation of half the heavy elements in the
the computational science universe, from
universe. This has led to a surprising and
autonomous individual investigators to
Figure 4. The process of scientific simulation, showing the validation and verification loop (left) and
algorithm and performance tuning loop (right).
more area than it would have in its laminar some scientific fields, high-fidelity models
state, without turbulence. Hence, incorrect of many important processes involved in
prediction of turbulence intensity may the simulations are not available; and
under- or over-represent the extent of additional scientific research will be needed
reaction. to achieve a comparable level of confidence.
From first principles, the reactions of These computations pictured here were
molecules are described by the Schrödinger performed on the IBM SP, named
equation, and fluid flow by the Navier- “Seaborg,” at the National Energy Research
Stokes equations. However, each of these Scientific Computing Center, using up to
equation sets is too difficult to solve 2048 processors. They are among the most
directly for the turbulent premixed rod- demanding combustion simulations ever
stabilized “V” methane flame illustrated in performed. However, the ability to do them
Figure 5. So we must rely on is not merely a result of improvements in
approximations to these equations. These computer speed. Improvements in
approximations define the computational algorithm technology funded by the DOE
models used to describe the methane flame. Applied Mathematical Sciences Program
For the current simulations, the set of over the past decade have been
84 reactions describing the methane flame instrumental in making these computations
involves 21 chemical species (methane, feasible. Mathematical analysis is used to
oxygen, water, carbon dioxide, and many reformulate the equations describing the
trace products and reaction intermediates). fluid flow so that high-speed acoustic
The particular version of the compressible transients are removed analytically, while
Navier-Stokes model employed here allows compressibility effects due to chemical
detailed consideration of convection, reactions are retained. The mathematical
diffusion, and expansion effects that shape model of the fluid flow is discretized using
the reaction; but it is “filtered” to remove high-resolution finite difference methods,
sound waves, which pose complications combined with local adaptive mesh
that do not measurably affect combustion in refinement by which regions of the finite
this regime. The selection of models such as difference grid are automatically refined or
these requires context-specific expert coarsened to maximize overall
judgment. For instance, in a jet turbine computational efficiency. The
simulation, sound waves might represent a implementation of this methodology uses
non-negligible portion of the energy budget an object-oriented, message-passing
of the problem and would need to be software framework that handles the
modeled; but it might be valid (and very complex data distribution and dynamic
cost-effective) to use a less intricate load balancing needed to effectively exploit
chemical reaction mechanism to answer the modern parallel supercomputers. The data
scientific question of interest. analysis framework used to explore the
results of the simulation and create visual
Development of the chemistry model and images that lead to understanding is based
validation of its predictions, which was on recent developments in “scripting
funded in part by DOE’s Chemical Sciences languages” from computer science. The
Program, is an interesting scientific story combination of these algorithmic
that is too long to be told here. Suffice it to innovations reduces the computational cost
say that this process took many years. In by a factor of 10,000 compared with a
Researchers have just begun to explore the This simulation achieves its remarkable
data generated in these simulations. resolution of the thin flame edge by means
Figure 7 presents a comparison between a of adaptive refinement, which places fine
laboratory measurement and a simulation. resolution around those features that need
The early stages of the analysis of this data it and uses coarse resolution away from
indicate that excellent agreement is such features, saving orders of magnitude
obtained with observable quantities. in computer memory and processing time.
Further analyses promise to shed new light As the reaction zone shifts, the refinement
automatically tracks it, adding and
removing resolution dynamically. As
described earlier, the simulation also
employs a mathematical transformation
that enables it to skip over tiny time steps
that would otherwise be required to resolve
sound waves in the flame domain. The four
orders of magnitude saved by these
algorithmic improvements, compared with
previous practices, is a larger factor than
the performance gain from parallelization
on a computer system that costs tens of
millions of dollars! It will be many years
before such a high-fidelity simulation can
be cost-effectively run using previous
practices on any computer.
Table 1. If additional funding becomes available for Office of Science scientific, computer
science, and mathematics research programs, as well as for the needed additional
computing resources, scientists predict the following major advances in the
research programs supported by the Office of Science within the
next 5–10 years
Research Programs Major Scientific Advances
Biological and • Provide global forecasts of Earth’s future climate at regional scales using
Environmental high-resolution, fully coupled and interactive climate, chemistry, land
Sciences cover, and carbon cycle models.
• Develop predictive understanding of subsurface contaminant behavior
that provides the basis for scientifically sound and defensible cleanup
decisions and remediation strategies.
• Establish a mathematical and computational foundation for the study of
cellular and molecular systems and use computational modeling and
simulation to predict and simulate the behavior of complex microbial
systems for use in mission application areas.
Chemical and • Provide direct 3-dimensional simulations of a turbulent methane-air jet
Materials Sciences flame with detailed chemistry and direct 3-dimensional simulations of
autoignition of n-heptane at high pressure, leading to more-efficient,
lower-emission combustion devices.
• Develop an understanding of the growth and structural, mechanical, and
electronic properties of carbon nanotubes, for such applications as
strengtheners of polymers and alloys, emitters for display screens, and
conductors and nonlinear circuit elements for nanoelectronics.
• Provide simulations of the magnetic properties of nanoparticles and
arrays of nanoparticles for use in the design of ultra-high-density
magnetic information storage. (Nanomagnetism is the simplest of many
important advances in nanoscience that will impact future electronics,
tailored magnetic response devices, and even catalysis.)
• Resolve the current disparity between theory and experiment for
conduction through molecules with attached electronic leads. (This is the
very basis of molecular electronics and may well point to opportunities
for chemical sensor technology.)
Fusion Energy • Improve understanding of fundamental physical phenomena in high-
Sciences temperature plasmas, including transport of energy and particles,
turbulence, global equilibrium and stability, magnetic reconnection,
electromagnetic wave/particle interactions, boundary layer effects in
plasmas, and plasma/material interactions.
• Simulate individual aspects of plasma behavior, such as energy and
particle confinement times, high-pressure stability limits in magnetically
confined plasmas, efficiency of electromagnetic wave heating and
current drive, and heat and particle transport in the edge region of a
plasma, for parameters relevant to magnetically-confined fusion plasmas.
• Develop a fully integrated capability for predicting the performance of
magnetically confined fusion plasmas with high physics fidelity, initially
for tokamak configurations and ultimately for a broad range of practical
energy-producing magnetic confinement configurations.
• Advance the fundamental understanding and predictability of high-
energy density plasmas for inertial fusion energy. (Inertial fusion and
magnetically confined fusion are complementary technological
approaches to unlocking the power of the atomic nucleus.)
Table 1 (continued)
Research Programs Major Scientific Advances
High Energy • Establish the limits of the Standard Model of Elementary Particle
Physics Physics by achieving a detailed understanding of the effects of strong
nuclear interactions in many different processes, so that the equality of
Standard Model parameters measured in different experiments can be
verified. (Or, if verification fails, signal the discovery of new physical
phenomena at extreme sub-nuclear distances.)
• Develop realistic simulations of the performance of particle accelerators,
the large and complex core scientific instruments of high-energy physics
research, both to optimize the design, technology and cost of future
accelerators and to use existing accelerators more effectively and
efficiently.
Nuclear Physics • Understand the characteristics of the quark-gluon plasma, especially in
the temperature-density region of the phase transition expected from
quantum chromodynamics (QCD) and currently sought experimentally.
• Obtain a quantitative, predictive understanding of the quark-gluon
structure of the nucleon and of interactions of nucleons.
• Understand the mechanism of core collapse supernovae and the nature
of the nucleosynthesis in these spectacular stellar explosions.
this work is to find deviations from this will need another order of magnitude of
model—a discovery that would require the computational power.
introduction of new forms of matter or new
physical principles to describe matter at the A primary goal of the nuclear physics
shortest distances. Determining whether or program at the Relativistic Heavy Ion
not each new measurement is in accord Collider (RHIC) at Brookhaven National
with the Standard Model requires an Laboratory is the discovery and
accurate evaluation of the effects of the characterization of the quark–gluon
strong interactions on the process under plasma—a state of matter predicted by
study. These computations are a crucial QCD to exist at high densities and
companion to any experimental program in temperatures. To confirm an observation of
high-energy physics. QCD is the only a new state of matter, it is important to
known method of systematically reducing determine the nature of the phase transition
all sources of theoretical error. between ordinary matter and a quark–
gluon plasma, the properties of the plasma,
Lattice calculations lag well behind and the equation of state. Numerical
experiment at present, and the precision of simulations of QCD on a lattice have
experiments under way or being planned proved to be the only source of a priori
will make the imbalance even worse. The predictions about this form of matter in the
impact of the magnitude of the lattice errors vicinity of the phase transition.
is shown in Figure 11. For the Standard
Model to be correct, the parameters ρ and η Recent work suggests that a major increase
must lie in the region of overlap of all the in computing power would result in
colored bands. Experiments measure definitive predictions of the properties of
various combinations of ρ and η, as shown the quark–gluon plasma, providing timely
by the colored bands. The reduction of and invaluable input into the RHIC heavy
theory errors shown will require tens of ion program.
teraflop-years of simulations, which will be
available through the construction of new Similar lattice calculations are capable of
types of computers optimized for the predicting and providing an understanding
problems at hand. A further reduction, of the inner structure of the nucleon where
when current experiments are complete, the quarks and gluons are confined, another
Figure 11. Constraints on the Standard Model parameters r and h. For the Standard Model to
be correct, they must be restricted to the region of overlap of the solidly colored bands. The
figure on the left shows the constraints as they exist today. The figure on the right shows the constraints
as they would exist with no improvement in the experimental errors, but with lattice gauge theory
uncertainties reduced to 3%. [R. Patterson]
over the same length of time (see this coupling can therefore be resolved
Figure 13). The series of algorithmic accurately with little computational effort.
improvements producing this factor are all The various improvements in algorithms
based on a fundamental mathematical for solving this problem came about by
property of the underlying model—namely, successive redesigns of the discretization
that the function describing electrostatic methods and solver algorithms to better
coupling between disjoint regions in space exploit this mathematical property of the
is very smooth. Expressed in the right way, model.
COMPUTER SCIENCE TOOLS FOR Despite this grim picture, prospects for
LARGE-SCALE SIMULATION placing usable computing environments
into the hands of scientific domain experts
The role of computer science in ultrascale
are improving. In the last few years, there
simulation is to provide tools that address
has been a growing understanding of the
the issues of complexity, performance, and
problems of managing complexity in
understanding. These issues cut across the
computer science, and therefore of their
computer science disciplines. An integrated
potential solutions. For example, there is a
approach is required to solve the problems
deeper understanding of how to make
faced by applications.
programming languages expressive and
easy to use without sacrificing high
COMPLEXITY
performance on the sophisticated, adaptive
One of the major challenges for applications algorithms.
is the complexity of turning a mathematical
model into an effective simulation. There Another example is the success of
are many reasons for this, as indicated in component-oriented software in some
Figure 4. Often, even after the principles of application domains; such “components”
a model and simulation algorithm are well have allowed computational scientists to
understood, too much effort still is required focus their own expertise on their science
to turn this understanding into practice while exploiting the newest algorithmic
because of the complexity of code design. developments. Many groups in high-
Current programming models and performance computing have tackled these
frameworks do not provide sufficient issues with significant leadership from
support to allow domain experts to be DOE. Fully integrated efforts are required
shielded from details of data structures and to produce a qualitative change in the way
computer architecture. Even after an application groups cope with the
application code produces correct scientific complexity of designing, building, and
results, too much effort still is required to using ultrascale simulation codes.
obtain high performance. Code tuning
efforts needed to match algorithms to PERFORMANCE
current computer architectures require
One of the drivers of software complexity is
lengthy analysis and experimentation.
the premium on performance. The most
obvious aspect of the performance problem
Once an application runs effectively, often
is the performance of the computer
the next hurdle is saving, accessing, and
hardware. Although there have been
sharing data. Once the data are stored, since
astounding gains in arithmetic processing
ultrascale simulations often produce
rates over the last five decades, users often
ultrascale-sized datasets, it is still too
receive only a small fraction of the
difficult to process, investigate, and
theoretical peak of processing performance.
visualize the data in order to accomplish
There is a perception that this fraction is, in
the purpose of the simulation—to advance
fact, declining. This perception is correct in
science. These difficulties are compounded
some respects. For many applications, the
by the problems faced in sharing resources,
reason for a declining percentage of peak
both human and computer hardware.
performance is the relative imbalance in the
performance of the subsystems of high-end
In Volume 2 of this report, scientists from disciplines and tools—many adapted from
many disciplines discuss critical computer the industrial software world—to manage
science technologies needed across the complexity of code development. Data
applications. Visual Data Exploration and Management and Analysis addresses the
Analysis considers understanding the challenges of managing and understanding
results of a simulation through the increasingly staggering volumes of data
visualization. Computer Architecture looks produced by ultrascale simulations.
at the systems on which ultrascale Performance Engineering is the art of
simulations run and for which achieving high performance, which has the
programming environments and algorithms potential of becoming a science. System
(described in the section on computational Software considers the basic tools that
mathematics) must provide software tools. support all other software.
Programming Models and Component
Technology for High-End Computing The universality of these issues calls for
discusses new techniques for turning focused research campaigns. At the same
algorithms into efficient, maintainable time, the inter-relatedness of these issues to
programs. Access and Resource Sharing one another and to the application and
looks at how to both make resources algorithmic issues above them in the
available to the entire research community simulation hierarchy points to the potential
and promote collaboration. Software synergy of integrated simulation efforts.
Engineering and Management is about
The preceding chapters of this volume have and computer science that will be required
presented a science-based case for large- or that could, if developed for widespread
scale simulation. Chapter 2 presents an on- use, reduce the computational power
going successful initiative, SciDAC—a required to achieve a given scientific
collection of 51 projects operating at a level simulation objective. The companion
of approximately $57 million per year—as a volume details these opportunities more
template for a much-expanded investment extensively. This concluding section revisits
in this area. To fully realize the promises of each of the recommendations listed at the
SciDAC, even at its present size, requires end of Chapter 1 and expands upon them
immediate new investment in computer with additional clarifying discussion.
facilities. Overall, the computational science
research enterprise could make effective 1. Major new investments in
use of as much as an order of magnitude computational science are needed in all of
increase in annual investment at this time. the mission areas of DOE’s Office of
Funding at that level would enable an Science, as well as those of many other
immediate tenfold increase in computer agencies, so that the United States may be
cycles available and a doubling or tripling the first, or among the first, to capture the
of the researchers in all contributing new opportunities presented by the
disciplines. A longer, 5-year view should continuing advances in computing power.
allow for another tenfold increase in Such investments will extend the
computing power (at less than a tenfold important scientific opportunities that
increase in expense). Meanwhile, the have been attained by a fusion of
emerging workforce should be primed now sustained advances in scientific models,
for expansion by attracting and developing mathematical algorithms, computer
new talent. architecture, and scientific software
engineering.
Chapter 3 describes the anatomy of a state-
of-the-art large-scale simulation. A The United States is an acknowledged
simulation such as this one goes beyond international leader in computational
established practices by exploiting science, both in its national laboratories and
advances from many areas of in its research universities. However, the
computational science to marry raw United States could lose its leadership
computational power to algorithmic position in a short time because the
intelligence. It conveys the opportunities represented by
multidisciplinary spirit intended in the computational science for economic,
expanded initiative. Chapter 4 lists some military, and other types of leadership are
compelling scientific opportunities that are well recognized; the underlying technology
just over the horizon new scientific and training are available without
advances that are predicted to be accessible restrictions; and the technology continues
with an increase in computational power of to evolve rapidly. As Professor Jack
a hundred- to a thousand-fold. Chapter 5 Dongarra from the University of Tennessee
lists some advances in the enabling recently noted.
technologies of computational mathematics
The rising tide of change [resulting from advances in science, mathematics, and
advances in information technology] shows computer science into simulations that can
no respect for the established order. Those
take full advantage of advanced
who are unwilling or unable to adapt in
response to this profound movement not
computers.
only lose access to the opportunities that the
information technology revolution is Part of the genius of the SciDAC initiative is
creating, they risk being rendered obsolete the cross-linking of accountabilities
by smarter, more agile, or more daring between applications groups and the
competitors. groups providing enabling technologies.
This accountability implies, for instance,
The four streams of development brought that physicists are concentrating on physics,
together in a “perfect fusion” to create the not on developing parallel libraries for core
current opportunities are of different ages, mathematical operations; and it implies
but their confluence is recent. Modern that mathematicians are turning their
scientific modeling still owes as much to attention to methods directly related to
Newton’s Principia (1686) as to anything application-specific difficulties, moving
since, although new models are continually beyond the model problems that motivated
proposed and tested. Modern simulation earlier generations. Another aspect of
and attention to floating point arithmetic SciDAC is the cross-linking of laboratory
have been pursued with intensity since the and academic investigators. Professors and
pioneering work of Von Neumann at the their graduate students can inexpensively
end of World War II. Computational explore many promising research directions
scientists have been riding the technological in parallel; however, the codes they
curve subsequently identified as “Moore’s produce rarely have much shelf life. On the
Law” for nearly three decades now. other hand, laboratory-based scientists can
However, scientific software engineering as effectively interact with many academic
a discipline focused on componentization, partners, testing, absorbing, and seeding
extensibility, reusability, and platform numerous research ideas, and effectively
portability within the realm of high- “hardening” the best of the ideas into long-
performance computing is scarcely a lived projects and tested, maintained codes.
decade old. Multidisciplinary teams that This complementarity should be built into
are savvy with respect to all four aspects of future computational science initiatives.
computational simulation, and that are
actively updating their codes with optimal Future initiatives can go farther than
results from each aspect, have evolved only SciDAC by adding self-contained, vertically
recently. Not every contemporary integrated teams to the existing SciDAC
“community code” satisfies the standard structure of specialist groups. The SciDAC
envisioned for this new kind of Integrated Software Infrastructure Centers
computational science, but the United (ISICs), for instance, should continue to be
States is host to a number of the leading supported, and even enhanced, to develop
efforts and should capitalize upon them. infrastructure that supports diverse science
portfolios. However, mathematicians and
2. Multidisciplinary teams, with carefully computer scientists capable of
selected leadership, should be assembled understanding and harnessing the output
to provide the broad range of expertise of these ISICs in specific scientific areas
needed to address the intellectual should also be bona fide members of the
challenges associated with translating
already exists in large numbers; and the originated and are maintained at national
operative principle is, instead, “build it or laboratories, a situation that is a natural
they will go!” Two plenary speakers at the adaptation to the paucity of scientific focus
workshop noted that a major U.S.-based in the commercial marketplace.
computational geophysicist has already
taken his leading research program to The SciDAC initiative has identified four
Japan, where he has been made a guest of areas of scientific software technology for
the Earth Simulator facility; this facility focused investment—scientific simulation
offers him a computational capability that codes, mathematical software, computing
he cannot find in the United States. Because systems software, and collaboratory
of the commissioning of the Japanese Earth software. Support for development and
Simulator last spring, this problem is most maintenance of portable, extensible
acute in earth science. However, the software is yet another aspect of the genius
problem will rapidly spread to other of the SciDAC initiative, as the reward
disciplines. structure for these vital tasks is otherwise
absent in university-based research and
4. Investment in hardware facilities difficult to defend over the long term in
should be accompanied by sustained laboratory-based research. Although
collateral investment in software SciDAC is an excellent start, the current
infrastructure for them. The efficient use funding level cannot support many types of
of expensive computational facilities and software that computational science users
the data they produce depends directly require.
upon multiple layers of system software
and scientific software, which, together It will be necessary and cost-effective, for
with the hardware, are the engines of the foreseeable future, for the agencies that
scientific discovery across a broad sponsor computational science to make
portfolio of scientific applications. balanced investments in several of the
layers of software that come between the
The President’s Information Technology scientific problems they face and the
Advisory Committee (PITAC) report of computer hardware they own.
1999 focused on the tendency to under-
invest in software in many sectors, and this 5. Additional investments in hardware
is unfortunately also true in computational facilities and software infrastructure
science. Vendors have sometimes delivered should be accompanied by sustained
leading-edge computing platforms collateral investments in algorithm
delivered with immature or incomplete research and theoretical development.
software bases; and computational science Improvements in basic theory and
users, as well as third-party vendors, have algorithms have contributed as much to
had to rise to fill this gap. In addition, increases in computational simulation
scientific applications groups have often capability as improvements in hardware
found it difficult to make effective use of and software over the first six decades of
these computing platforms from day one, scientific computing.
and revising their software for the new
machine can take several months to a year A recurring theme among the reports from
or more. Many industry-standard libraries the topical breakout groups for scientific
of systems software and scientific software applications and mathematical methods is
that increases in computational capability rapidly with increasing grid resolution into
comparable to those coming directly from methods that increase linearly or log-
Moore’s Law have come from improved linearly with complexity, with reliable
models and algorithms. This point was bounds on the approximations that are
noted over 10 years ago in the famous needed to achieve these simulation-
“Figure 4” of a booklet entitled Grand liberating gains in performance. It is
Challenges: High-Performance Computing and essential to keep computational science
Communications published by FCCSET and codes well stoked with the latest
distributed as a supplement of the algorithmic technology available. In turn,
President’s FY 1992 budget. A computational science applications must
contemporary reworking of this chart actively drive research in algorithms.
appears as Figure 13 of this report. It shows
that the same factor of 16 million in 6. Computational scientists of all types
performance improvement that would be should be proactively recruited with
accumulated by Moore’s Law in a 36-year improved reward structures and
span is achieved by improvements in opportunities as early as possible in the
numerical analysis for solving the educational process so that the number of
electrostatic potential equation on a cube trained computational science
over a similar period, stretching from the professionals is sufficient to meet present
demonstration that Gaussian elimination and future demands.
was feasible in the limited arithmetic of
digital computers to the publication of the The United States is exemplary in
so-called “full multigrid” algorithm. integrating its computational science
graduate students into internships in the
There are numerous reasons to believe that national laboratory system, and it has had
advances in applied and computational the foresight to establish a variety of
mathematics will make it possible to model fellowship programs to make the relatively
a range of spatial and temporal scales much new career of “computational scientist”
wider than those reachable by established visible outside departmental boundaries at
practice. This has already been universities. These internships and
demonstrated with automated adaptive fellowship programs are estimated to be
meshing, although the latter still requires under-funded, relative to the number of
too much specialized expertise to have well-qualified students, by a factor of 3.
achieved universal use. Furthermore, Even if all presently qualified and
multiscale theory can be used to interested U.S. citizens who are doctoral
communicate the net effect of processes candidates could be supported—and even
occurring on unresolvable scales to models if this pool were doubled by aggressive
that are discretized on resolvable scales. recruitment among groups in the native-
Multiphysics techniques may permit codes born population who are underrepresented
to “step over” stability-limiting time scales in science and engineering, and/or the
that are dynamically irrelevant at larger addition of foreign-born and domestically
scales. trained graduate students—the range of
newly available scientific opportunities
Even for models using fully resolved scales, would only begin to be grazed in new
research in algorithms promises to computational science doctoral
transform methods whose costs increase dissertations.
tional science attend to its own needs vis-à- (HECRTF) study, and is not the central
vis future computer hardware concern of this application-focused report.
requirements. The most noteworthy neglect However, it is appropriate to reinforce this
of the scientific computing industry is, recommendation, which is as true of our era
arguably, the deteriorating ratio of memory as it was true of the era of the “Lax Report”
bandwidth between processors and main (see the discussion immediately following).
memory to the computational speed of the The SciDAC report called for the
processors. It does not matter how fast a establishment of experimental computing
processor is if it cannot get the data that it facilities whose mission was to “provide
needs from memory fast enough—a fact critical experimental equipment for
recognized very early by Seymour Cray, computer science and applied mathematics
whom many regard as the father of the researchers as well as computational
supercomputer. Most commercial science pioneers to assess the potential of
applications do not demand high memory new computer systems and architectures
bandwidth; many scientific applications do. for scientific computing.” Coupling these
facilities with university-based research
There has been a reticence to accord projects could be a very effective means to
computational science the same fill the architecture research gap.
consideration given other types of
experimental science when it comes to CLOSING REMARKS
federal assistance in the development of
The contributors to this report would like to
facilities. High-end physics facilities—such
note that they claim little original insight
as accelerators, colliders, lasers, and
for the eight recommendations presented.
telescopes—are, understandably, designed,
In fact, our last six recommendations may
constructed, and installed entirely at
be found, slightly regrouped, in the four
taxpayer expense. However, the design and
prescient recommendations of the Lax
fabrication of high-end computational
Report (Large-Scale Computing in Science and
facilities are expected to occur in the course
Engineering, National Science Board, 1982),
of profit-making enterprises; only the
which was written over two decades ago, in
science-specific purchase or lease and
very different times, before the introduction
installation are covered at taxpayer
of pervasive personal computing and well
expense. To the extent that the marketplace
before the emergence of the World Wide
automatically takes care of the needs of the
Web!
computational science community, this
represents a wonderful savings for the
Our first two recommendations reflect the
taxpayer. To the extent that it does not, it
beginnings of a “phase transition” in the
stunts the opportunity to accomplish
way certain aspects of computational
important scientific objectives and to lead
science are performed, from solo
the U.S. computer industry to further
investigators to multidisciplinary groups.
innovation and greatness.
Most members of the Lax panel—Nobelists
and members of the National Academies
The need to dedicate attention to research
included—wrote their own codes, perhaps
in advanced computer architecture is
in their entirety, and were capable of
documented elsewhere in many places,
understanding nearly every step of their
including the concurrent interagency High-
compilation, execution, and interpretation.
End Computing Revitalization Task Force
This is no longer possible: the scientific
This report is built on the efforts of a further and live up to the expectations of
community of computational scientists, Galileo, invoked in the quotation on the
DOE program administrators, and technical title page of this volume.
and logistical professionals situated
throughout the country. Bringing together The three plenary speakers for the
so many talented individuals from so many workshop during which the input for this
diverse endeavors was stimulating. Doing report primarily was gathered—Raymond
so and reaching closure in less than four Orbach, Peter Lax, and John Grosh—
months, mainly summer months, was perfectly set the perspective of legacy,
nothing short of exhilarating. opportunity, and practical challenges for
large-scale simulation. In many ways, the
The participants were reminded often challenges we face come from the same
during the planning and execution of this directions as those reported by the National
report that the process of creating it was as Science Board panel chaired by Peter Lax in
important as the product itself. If resources 1982. However, progress along the path in
for a multidisciplinary initiative in scientific the past two decades is astounding—
simulation were to be presented to the literally factors of millions in computer
scientists, mathematicians, and computer capabilities (storage capacity, processor
scientists required for its successful speed) and further factors of millions from
execution—as individuals distributed working smarter (better models, optimal
across universities and government and adaptive algorithms). Without the
industrial laboratories, lacking familiarity measurable progress of which the speakers
with each other’s work across disciplinary reminded us, the aggressive extrapolations
and sectorial boundaries—the return to the of this report might be less credible. With
taxpayer certainly would be positive, yet each order-of-magnitude increase in
short of its full potential. Investments in capabilities comes increased ability to
science are traditionally made this way, and understand or control aspects of the
in this way, sound scientific building blocks universe and to mitigate or improve our lot
are accumulated and tested. But as a pile of in it.
blocks does not constitute a house, theory
and models from science, methods and The DOE staff members who facilitated this
algorithms from mathematics, and software report often put in the same unusual hours
and hardware systems from computer as the scientific participants in breakfast
science do not constitute a scientific meetings, late-evening discussions, and
simulation. The scientists, mathematicians, long teleconferences. They did not steer the
and computer scientists who met at their outcome, but they shared their knowledge
own initiative to build this report are now of how to get things accomplished in
ready to build greater things. While projects that involve hundreds of people;
advancing science through simulation, the and they were incredibly effective. Many
greatest thing they will build is the next provided assistance along the way, but Dan
generation of computational scientists, Hitchcock and Chuck Romine, in particular,
whose training in multidisciplinary labored with the scientific core committee
environments will enable them to see much for four months.
Other specialists have left their imprints on The editor-in-chief gratefully acknowledges
this report. Devon Streit of DOE departmental chairs and administrators at
Headquarters is a discussion leader two different universities who freed his
nonpareil. Joanne Stover and Mark Bayless schedule of standard obligations and thus
of Pacific Northwest National Laboratory directly subsidized this report. This project
built and maintained web pages, often consumed his last weeks at Old Dominion
updated daily, that anchored workshop University and his first weeks at Columbia
communication among hundreds of people. University and definitely took his mind off
Vanessa Jones of Oak Ridge National of the transition.
Laboratory’s Washington office provided
cheerful hospitality to the core planners. All who have labored to compress their (or
Jody Shumpert of Oak Ridge Institute for their colleagues’) scientific labors of love
Science and Education (ORISE) single- into a few paragraphs, or to express
handedly did the work of a conference team themselves with minimal use of equations
in preparing the workshop and a pre- and jargon, will find ten additional readers
workshop meeting in Washington, D.C. The for each co-specialist colleague they thereby
necessity to shoehorn late-registering de-targeted. All involved in the preparation
participants into Arlington hotels during of this report understand that it is a
high convention season, when the response snapshot of opportunities that will soon be
to the workshop outstripped planning outdated in their particulars, and that it
expectations, only enhanced her display of may prove either optimistic or pessimistic
unflappability. Rachel Smith and Bruce in its extrapolations. None of these
Warford, also of ORISE, kept participants outcomes detracts from the confidence of
well directed and well equipped in ten-way its contributors that it captures the current
parallel breakouts. Listed last, but of first state of scientific simulation at a historic
importance for readers of this report, inflection point and is worth every sacrifice
members of the Creative Media group at and risk.
Oak Ridge National Laboratory iterated
tirelessly with the editors in meeting
production deadlines and creating a
presentation that outplays the content.
Distribution:
David will be organizing a workshop (early summer 2003, Washington, D.C. metro area) to
formally launch this effort. I expect this workshop will address the major opportunities and
challenges facing computational sciences in areas of strategic importance to the Office of
Science. I envision a report from this workshop by July 30, 2003. Furthermore, this
workshop should foster additional workshops, meetings, and discussions on specific topics
that can be identified and analyzed over the course of the next year.
Please mention this undertaking to your colleagues. I am looking forward to their support
and their participation, along with researchers from academia and industry, in making this
effort a success.
Thanks.
Walt P.