Computational science research and development
The mission of CSS is to help realize the end-to-end scientific
simulation environment envisioned by the NCAR Strategic Plan. The
specific role of CSS is to develop much of the critical software
and intellectual infrastructure needed to achieve the plan's
ambitious goals. To more accurately reflect this mission, the
Computational Science Section has been reorganized in 2002 into
three functional groups: Mathematical Methods, High Performance
Technology Tracking, and the Earth System Modeling Framework
Project. This alignment enables CSS to more effectively track
computer technology, learn to extract performance from it,
pioneer new and efficient numerical methods, create software
frameworks to facilitate scientific advancement -- particularly
interdisciplinary geoscience collaborations -- and share the
resulting software and findings with the community through
publications, talks and websites.
CSS has been both productive in research activities and
successful in obtaining grant money in 2002. The section
published six papers, including one review article, in
peer-reviewed journals in 2002 and has six more currently in
press. CSS had five successful proposals in 2002, bringing
into the section a total of approximately $3,500,000 of new
funding, and $160,000 of new equipment.
The largest single funded activity -- for $2,800,000 over
three years -- is the Earth System Modeling Framework project.
The ESMF project is a highly visible interagency effort
to build model creation and model-coupling software
infrastructure for the Earth science modeling community.
ESMF is well underway: in the past year it has successfully
met the first four NASA milestones of the project, and is
currently in the design phase of the software system.
Another significant CSS initiative in 2002 was the CSS-led
effort that resulted in NCAR becoming a member of the
Gelato Federation, an open software consortium focused on
the Itanium architecture. This resulted in CSS and NETS
Web100 funding, in the form of both salary for software
development and equipment that amounted to approximately
$360,000. This funding will both accelerate development
of the Spectral Toolkit and provide NCAR with visibility
within the Linux community.
As a result of this progress, CSS has also grown dramatically
over the past year. Eight new staff members have been added:
- Within ESMF, four full-time software developers, one
software integrator, and a half-time administrative
assistant have fully staffed ESMF as the NASA-funded project
gets underway.
- The scientific staff in CSS has grown from one to four
in 2002. The CSS core mathematics and computer science research
program has been expanded by the reclassification of one
software engineer IV (Dr. Steve Thomas) to Scientist III,
the addition of a Scientist I-level mathematician, an expert
in conservative advection, and a joint appointment at the
Scientist II level with the Computer Science Department at
the University of Colorado.
- The mathematical libraries effort in CSS has added
another software engineer to augment the Spectral Toolkit
Mathematical Software Library under the HP-sponsored Gelato
Federation Initiative.
CSS technology-tracking efforts in 2002 have centered on three
activities:
- Understanding and evaluating the new IA-64 (Itanium) computer
architecture
- Converting the 36-processor Compaq AlphaServer named prospect
into a functioning Linux cluster renamed lhotse
- Performing benchmark tests and developing performance models
in support of the ARCS procurement
The Itanium's very long instruction word (VLIW) architecture
represents a dramatic departure from the traditional superscalar
RISC microprocessor and CISC-like Pentium architectures used in
the geosciences departments at most universities today. As
Itanium microprocessors become plentiful in the geosciences
community, access to open-source high-performance mathematical
libraries optimized for this architecture will become important to
scientific progress. It is clear that, to be well positioned for
future advances, CSS needs to learn to performance-program the
Itanium processor. To this end, in the fall of 2001, CSS acquired
a four-processor Rx4610 Itanium-I server from Hewlett Packard.
This system was evaluated through synthetic benchmarks and also
by porting several NCAR applications, including CCM3.
This activity also included an inter-comparison of the
usability and robustness of the Linux and HP-UX operating systems
on the introductory Itanium system and extensive evaluation of
early releases of the Intel and HP C++ and Fortran-90 compilers
for Itanium. This work was carefully documented and presented to
HP engineers in Fort Collins, Colorado in February 2002. A few
months later, HP agreed to sponsor NCAR as a member of the Gelato
Federation, an organization devoted to the advancement of the
Linux/Itanium environment and the distribution of open source
software. As a member, CSS will receive $200,000 over two years
to fund a software engineer to help optimize the STK libraries
and other applications for the Itanium architecture. In May 2002,
CSS also responded to and won an HP RFP for free Itanium-2
hardware: six Itanium II systems, two dual-processor servers, and
four workstations worth $130,000 were awarded. Two of these will
be provided to NETS to conduct Web100 development and testing in
collaboration with the Pittsburgh Supercomputing Center.
The second HPC technology-tracking activity has been the
conversion of the largely unsuccessful $750,000 AlphaServer
system prospect into a Linux system. Early in this effort, CSS
established a collaborative relationship with a new Scientist
I -- Peter Thorton in CGD -- who needed approximately four months
of dedicated computing to perform a data-intensive, embarrassingly
parallel calculation to run a 2,000-year spin-up calculation of a
detailed CONUS land surface model at an unprecedented 1-km
resolution. This application mapped ideally to the new
Alpha-processor-based Linux cluster now renamed lhotse. This
calculation, the first of its kind, has been successfully
completed.
The mathematical research effort within CSS is divided into
two principal activities:
- Developing performance-portable, highly efficient,
open-source numerical libraries for use by the mathematical,
geosciences, and engineering communities
- Investigating mathematical techniques and software
implementations for producing accurate, efficient and scalable
numerical dynamical cores for general circulation and turbulence
models
Numerical Library Development
The first version of the spectral toolkit (STK) was released
in 2002. Initially, access was restricted to early adopters in
the NCAR scientific community; currently the initial release of
STK software and documentation may be downloaded from the CSS website
http://www.scd.ucar.edu/css/software/stk/
The release version of STK provides support for complex and real
one-dimensional FFTs. Although the software is written in C/C++,
it supports a simple and easy-to-use Fortran-callable interface.
All STK functions are threadable, and some, such as the two- and
three-dimensional complex and real FFTs, have built-in parallelism
implemented using pthreads. These parallel functions in the STK
library have been applied and thoroughly tested in the construction
of a parallel turbulence model in collaboration with NCAR
scientists Jack Herring and Yoshi Kimura. This 3D model was run
on parallel computers at resolutions up to 1024 points in each
spatial dimension. A distributed-memory version of the
multi-dimensional 3D FFT has achieved over 200 GFLOPS sustained
performance on the Pittsburgh Terascale computing facility.
Paul Swarztrauber has developed new software that implements
the computational methods for computing Gauss points and weights
and for computing harmonic projections. The new method for
computing Gauss points and weights has already been incorporated
into the SHEREPACK subroutine gaqdi. This software will
ultimately be integrated into the Spectral Toolkit.
The projection algorithms based on the weighted orthogonal
complement method have also been incorporated into SPHEREPACK
as subroutines shpei, shpe, shpgi,
and shpg, and they perform projections on equally
spaced and Gauss-distributed grids. The new SPHEREPACK
library can be accessed and downloaded at
http://www.scd.ucar.edu/css/software/spherepack/
Elliptic solvers for semi-implicit nonhydrostatic NWP
models
The elliptic problems in semi-implicit nonhydrostatic
atmospheric models are typically poorly conditioned,
nonseparable, contain cross-derivative terms, and often are
nonsymmetric. In 2002, the collaborative effort between
NCAR researchers Steven Thomas (SCD), Joshua Hacker (MMM),
Piotr Smolarkiewicz (MMM), and Roland Stull (University of
British Columbia) has led to a class of effective Krylov
methods, i.e. a conjugate residual (GCR) algorithm
preconditioned with a 3D direct solver, using standard
tridiagonal inversion in the vertical. They have developed
a horizontal spectral preconditioner as an alternative to
the more standard and much simpler line-Jacobi relaxation
scheme. However, the efficacy of the spectral preconditioner
requires neglecting the cross derivative terms and
homogenization (e.g., averaging) metric coefficients over
the computational domain. Because such a compromise causes
a substantial departure of the preconditioner from the
original elliptic operator, it is not obvious a priori
whether it leads to a competitive solver. They evaluated
the robustness of the proposed approach over a broad
range of representative meteorological applications, and
documented its superior performance in the context of a
three-time-level semi-implicit semi-Lagrangian all-scale
weather-prediction/research model.
Spectral Element Atmosphere Model (SEAM)
Further progress was made in FY2002 on both 2-D shallow
water and 3-D primitive equations SEAM models based on a
spectral element horizontal discretization. Support for
hybrid coordinates was added to the primitive equations
dynamical core in anticipation of introducing physics into
the model. Studies were done on new filtering techniques,
based on the work of Fischer and Tufo, that do not require
communications.
The domain decomposition software in the SEAM spectral
element model was improved in FY2002 in two important ways.
To minimize communications costs of the pure MPI paradigm
on SMP clusters, the package now uses a hierarchical
domain decomposition strategy based on the METIS package
from the University of Minnesota. The communications
library first decomposes the element mesh on the cube
sphere into node-domains based on the number of SMP nodes;
then each node-domain is fed back into METIS to be
decomposed into processor-domains.
The second improvement is the introduction of
decompositions based on the nesting of so-called Peano
or Hilbert space-filling curves. Such curves map the
two-dimensional element grid on the cube sphere onto
a one-dimension sequence of elements that is generated
in a self-similar hierarchical way. Hilbert curves
permit one to produce efficient load-balanced domains
with quite good locality with a trivial domain
decomposition algorithm. The package is currently limited
to a number of elements along a cube edge factorable into
powers of two and three. The space-filling curve
decomposition, as applied to the cube grid in SEAM, is the
subject of a paper in preparation by John Dennis in CSS.
The integration rate of the semi-implicit shallow-water
model version of SEAM has been further improved by introducing
an additive overlapping Schwarz preconditioner for the
iterative conjugate gradient solver. The Schwarz
preconditioner has a much smaller memory footprint compared
to the block Jacobi preconditioner employed originally. The
Schwarz preconditioner also has slightly better convergence
properties than the block Jacobi. This is because it
overlaps with neighboring elements, and so does a better
job of communicating information about the solution
throughout the system.
Finally, this year Stephen Thomas and Henry Tufo have
been awarded $500,000 for three years from the NSF to
explore adaptive mesh refinement strategies using the
SEAM dynamical core. The funding includes support for a
post-doctoral fellow, a graduate student research
assistant, and annual workshops that will emphasize the
dynamical core research within CSS. These workshops will
offer financial support for young researchers and
students to visit NCAR to gain experience working with
high-performance numerical libraries, dynamical cores,
test cases, and visualization tools.
Double Fourier dynamical core
Further progress on the double-Fourier approach
included work on a semi-implicit semi-Lagrangian scheme
with NCAR ASP post-doctoral fellow Anita Leighton.
However, work on this aspect of the project has stopped
within CSS due to the departure Dr. W. Spotz from the
section. A presentation by Drs. Leighton and Spotz will
be given at the 2002 PDEs on the Sphere meeting in
Toronto, Canada.
In related work, Paul Swarztrauber worked with Bill
Spotz to compare four candidate projection algorithms
for use in the Double Fourier Method:
- The standard method using forward and backward
Legendre transforms
- The direct method, which uses a single projection
matrix when this approach results in fewer operations
- The fast multipole method
- The weighted orthogonal complement method
Swarztrauber and Spotz have shown that for spectral
truncations up to N=200, the weighted orthogonal
complement method has the lowest operation count, best
cache utilization and best overall timings.
Cache-optimized spectral dynamical core (BOB)
Richard Loft and Leo Rivier in CSS have developed, in
collaboration with Lorenzo Polvani at Columbia University,
an efficient, cache-optimized spectral transform dynamical
core for low-cost Linux Beowulf clusters. This code,
called Built-on-Beowulf (BOB), outperforms CCM by
substantial factors (3-5) at relevant climate resolutions
on these low-cost systems.
The BOB code organizes a
vorticity-divergence formulation of the shallow water
equations (or primitive equations) along the lines of the
method described by Rudi Jacob-Chien. This formulation
permits one to write the Legendre transform as a blocked
matrix-matrix multiply. The memory footprint of the
Legendre transforms are made smaller by computing the
associated Legendre polynomials "on the fly." This reduces
memory traffic and increase cache reuse.
The Ph.D. thesis
of Rivier was based on simulations of jet formation in the
upper atmosphere of Jupiter using the BOB code. An article
recently published in Mon. Wea. Rev. documented the
design and performance of BOB. Subsequent to publication
of this paper, there have been requests for access to the
code from researchers at GFDL, MIT, and Texas A&M
University.
Coronal Mass Ejection
Dr. B.C. Low of NCAR has proposed a theory for the
formation of Coronal Mass Ejections (CMEs -- magnetic
"bubbles") based on the forced (gravitation and pressure)
and force-free elliptic equation for the solar magnetic
stream function. The theory relies on the total magnetic
energy of the associated magnetic field.
Dr. Steve Thomas
of CSS, in collaboration with Dr. Natasha Flyer and Bengt
Fomberg of the University of Colorado, have applied a
variety of advanced numerical methods to solve this highly
nonlinear elliptic problem. These include both second-order
finite difference methods and a Fourier-Chebyshev
pseudo-spectral method on the semi-infinite half-plane in
spherical coordinates.
Because the potential far-field
solution decays in inverse proportion to radius, Dr.
Flyer's original contribution was to use an exponential
conformal map for the problem. The far-field solution is
then matched to the inner solution at an artificial outer
boundary using a radiation boundary condition (RBC),
derived from the eigen-functions of the linear far-field
operator. The solutions of the elliptic problem depend
on one critical parameter, and the team is currently
exploring the use of arc-length continuation combined
with Newton's iterative method of approximation to
compute solutions as well as looking for bifurcation
points.
The sought-after magnetic "bubble" solutions for the
forced equation have been found, and initial results in
the force-free case are very encouraging. The team
believes that a nonlinear bifurcation may lead to the
formation of a bubble, and we are tracking the magnetic
field energy as a function of the parameter.
The Earth System Modeling Framework (ESMF) project is building
software infrastructure for the nation's leading climate, weather,
and data-assimilation applications. Collaborators include:
- SCD, the Climate and Global Dynamics Division,
and the Mesoscale and Microscale Meteorology Division (NCAR)
- The Geophysical Fluid Dynamics Laboratory and the National Centers
for Environmental Prediction (National Oceanic and Atmospheric Administration)
The Massachusetts Institute of Technology and the University of
Michigan
Argonne National Laboratory and Los Alamos National Laboratory (Department of Energy)
The Data Assimilation Office and the NASA Season-to-Interannual Prediction Project (Goddard Space Flight Center, National Aeronautics and Space Administration)

Community Climate Simulation Model (CCSM) output
at T170 resolution
The
three-year project is being funded by NASA's Earth Science
Technology Office at a level of $9,800,000. Of this, $2,600,000 will
go to NCAR SCD for implementing the core software. In FY2002, the ESMF
collaboration produced an exhaustive requirements specification, an
architecture document, and a synthetic validation suite. An initial
prototype of the framework will be available in FY2003. SCD is proud
to be the home of the implementation team for this landmark effort,
and looks forward to its success and growth.
Detailed information about ESMF is provided at
the ESMF website.
During FY2002, the Earth System Modeling Framework team
accomplished the following:
- Negotiated a three-year, $9,800,000 contract with the NASA
Earth Science Technology Office, $2,600,000 of which will go to
NCAR SCD for implementing the core framework
- Established a collaborative development environment and
communal repositories
- Collaboratively created an ESMF Software Developer's Guide
and an exhaustive ESMF Requirements Document
- Organized a community meeting with 80+ attendees in
Washington, DC, on May 30, 2002 to review the Requirements
Document and solicit additional input
- Assembled the ESMF Validation (EVA) Suite and performed
performance baselines on application codes that will be
adopting the framework
- Collaboratively developed an ESMF Architecture Document,
an ESMF Implementation Report, and a Build and Test Plan
- Selected the ESMF Executive and Advisory Boards and hosted
their first meeting at NCAR on September 26-27, 2002
- Submitted the first Annual Report for the ESMF
project
In addition to the above, the ESMF team submitted a proposal
to NASA for a funding increment for ANL to assist in framework
implementation and facilitate collaboration with the DOE's
Common Component Architecture project, a related effort. A
renewal of the NCAR strategic initiative that partly funds WRF
involvement in the ESMF project was also submitted.
In FY2002, CSS has also obtained funding to develop
software to support ocean forecasting through an
Oregon State University-led proposal entitled
":ITR/AP: Collaborative Research: Modular Ocean Data
Assimilation." Under this proposal, CSS staff in SCD will
adapt existing software components from other framework efforts
(perhaps including the Earth System Modeling Framework), and as
appropriate, develop new ones specific to the application domain
to integrate pre-existing parallel ocean models into the Inverse
Ocean Modeling System (IOM) developed by Dr. Andrew Bennett and
others at Oregon State University. Dr. Bennett is the principal
investigator on this proposal, which will provide $233,750 over
approximately four years.
CSS will exploit the dramatic successes of FY2002 in FY2003 by:
- Meeting the ESMF Project milestones
- Evaluating and developing mathematical software for the
Itanium architecture as Gelato Federation members
- Further developing the SEAM and BOB dynamical cores and
related partnerships
- Starting the newly funded initiatives in the areas of
ocean forecasting and spectral element adaptive mesh
refinement research and development activities
During FY2003, the ESMF team anticipates completing the
following activities which are required for the next four
project milestones
- Installing a high-performance cluster at MIT
- Releasing a prototype API and software
- Achieving partial ESMF compliance for a subset of the
ESMF testbed applications, which include the CCSM, WRF,
the NCEP forecast suite, and other codes
- Making initial delivery of three interoperability
experiments with new coupled configurations for testbed
codes
The focus of the work with lhotse in FY2003 will center
on installing hardware and software to create a scalable
I/O subsystem for lhotse and installing the kernel
modifications needed to support high-speed message passing
using the Quadrics interconnect. This will allow lhotse to
be used as a compute platform for the GTP program.
In FY2003, the Spectral Toolkit will be further expanded
in several areas. Early in FY2003, CSS intends to release
additional STK functionality including 2D and 3D
multithreaded real and complex FFTs and spherical harmonic
Legendre transforms on the Gaussian grid. The performance
of the Legendre transform will be based on hand-optimized
real-matrix complex-matrix multiply algorithms that have
already demonstrated high fractions of peak on a variety
of microprocessors. An effort to optimize all STK
functions for the Itanium-2 processor, funded by
Hewlett-Packard Corporation, will augment this effort.
Support for spherical harmonic Legendre transforms on
equally spaced grids will be added early in 2003, as will
additional operators, such as those for computing the
vorticity or divergence from given input velocity fields
and vice versa. The ultimate goal is to incorporate the
essential functionality of SPHEREPACK into STK, including
the new functions recently added by Paul Swarztrauber.
Work will also begin next year developing 2D and 3D
multi-threaded Spectral Element quadrature operators and
the associated mathematical framework that will allow
users to easily construct efficient multi-threaded and
distributed-memory applications based on the spectral
element method.
STK is envisioned as gradually forming the basic
mathematical infrastructure for all of the dynamical
core research in the section. This will provide a useful
way to disseminate the core knowledge obtained by this
research to a broad community of users that may in turn
employ these components in ways not anticipated by the
creators of the STK library.
The first step in this direction will be to develop
a new implementation of the BOB dynamical core, which
will employ the spherical harmonics and FFT functions in
STK. Both 2D shallow-water and 3D primitive equations
versions of BOB will be implemented on top of the STK
libraries. We expect this core to surpass the performance
of the original BOB implementation.
The 2D semi-implicit scheme employing the additive
overlapping Schwarz preconditioner for spectral elements
will be extended to the 3D primitive equations. An
eigen-mode decomposition in the vertical permits one to
solve a 2-D Helmholtz problem for the generalized pressure
in each vertical layer. To further improve the integration
rate acceleration factor over the explicit Eulerian
formulation, we plan to introduce conservative
semi-Lagrangian advection, based on the work of Dr. Nair,
for the dynamics and constituent transport in the spectral
element model.
To date, only the Held-Suarez idealized physical
forcings and the baroclinic instability tests of
Jablonski and Williamson have been performed on the
spectral element core. The next step with the section's
spectral element model will be to systematically introduce
and test physics, or more precisely, introduce the core
into the CAM2 framework of CCSM. Decompositions of the cube
sphere element grid which load-balance the physics will
have to be developed and tested. The incorporation of
physics will likely follow the development and evaluation
of several intermediate test cases, such as an adiabatic
atmosphere with smooth orography, advection of passive
tracers over orography, followed by test cases with active
water vapor, such as the aqua-planet of Williamson.
Two dynamical cores and libraries workshops are planned
over the next three years to transfer CSS research in these
areas to the broader NCAR-university community. These
workshops will be tutorial in nature with an emphasis on
the construction of dynamical cores using CSS libraries.
Test cases and visualization will also be emphasized,
including the NCAR Graphics language NCL. Finally, CSS
staff will continue to spread knowledge of their findings
through presentations and publications in peer-reviewed
journals.
In addition to these continuing activities, CSS plans to
begin building a computer science program. To this end, in
FY2003, CSS intends to pursue two proposals. The first is a
midrange NSF ITR proposal to investigate the suitability of
IBM Blue Gene/L and Blue Gene/C advanced MPP architectures
to geoscience simulation. This proposal is being developed
in collaboration with computer scientists at the University
of Delaware, Lawrence Livermore National Laboratory, IBM
T.G. Watson Research Center, the University of Colorado,
and the Computational Science Section within SCD.
The second proposal will be submitted to NASA and will
involve the development of a Grid testbed between the NCAR
Mass Storage System (MSS) and a Pentium-4 Beowulf cluster
at CU to produce a fully production-capable, Grid-enabled,
bio-geo-chemistry modeling system. This initiative will
build the resume of CSS in the area of Grid computing,
which could then provide the basis for additional Grid
proposals in the future.
Cecelia DeLuca in CSS organized and led a hands-on
tutorial session February 20, 2002 to introduce Earth
System Modeling Framework (ESMF) participants to the
framework development environment. The training session
was held at the NCAR Corporate Training Center.
Steve Thomas co-organized, in collaboration with Annick
Pouquet of the Geophysical Turbulence Program, a scientific
conference entitled Adaptive and High-order Methods with
Applications in Turbulence. This conference was held at
NCAR in March 2002. The workshop lasted three days and had
15 invited speakers, with an additional four graduate student
and post-doctoral presentations. The conference brought
together approximately 30 scientists interested in the
numerical simulation aspects of turbulence.
Cecelia DeLuca in CSS served as one of three technical
managers of the Earth System Modeling Framework (ESMF), which
is developing the software infrastructure for building and
coupling climate models, weather models, and data-assimilation
systems. Collaborators include SCD, CGD, and MMM at NCAR; NOAA
GFDL and NCEP; MIT; the University of Michigan; DOE ANL and
LANL, and NASA GSFC, DAO, and NSIPP. The ESMF is being
developed jointly by the Earth system community as a resource
that will benefit the community by increasing software reuse,
interoperability, and performance portability. Cecelia
organized the first ESMF community meeting, which was held
in Washington, D.C. on May 30, 2002. The purpose of the
meeting was to review a draft ESMF Requirements Document and
elicit additional requirements from the broader community.
More than 80 people attended the meeting. Additional input
from the community was collected via online forms and
archived on the ESMF website.
Cecelia DeLuca was reappointed CCSM Software Engineering
Working Group (SEWG) Co-Chair for a second two-year term. She
organized and participated in SEWG meetings at NCAR February
14-15, 2002 and at the CCSM workshop in Breckenridge, Colorado
on June 25, 2002.
John Dennis presented a talk at the SEWG Breckenridge
workshop on Trends in Computer Architectures. His presentation
provided important technical background information to the
attendees of the workshop.
Steve Thomas has refereed publications for Monthly
Weather Review, Journal of Computational Physics,
International Journal of Applied Mathematics, and
Computer Science. Steve was elected to the Scientific
Advisory Committee (SAC) for high-performance computing in
Canada.
Richard Loft reviewed articles for the Journal of
Computational Physics.
CSS was instrumental this year in bringing about UCAR
membership in the Gelato Federation
http://www.gelato.org/
The goals of the Gelato Federation are to provide
open source software for the Linux/Itanium-2 computing
platform. The architecture of the Itanium-2 represents a
significant departure from previous microprocessor
designs and holds significant promise as a key component
in future high-performance computer designs. As a Gelato
member, CSS has committed one software developer and has
recently received $200,000 in matching funds from
Hewlett-Packard, UCAR's Gelato sponsor, to fund a software
engineer for at least two years. HP has also donated
approximately $160,000 of hardware to support this effort.
CSS's objectives as members of Gelato are three-fold: port
and optimize CSS Spectral Toolkit mathematical libraries
for Itanium-II, port and tune production NCAR Fortran
codes to this environment, and improve, in collaboration
with the NETS Web100 project, the TCP/IP network
performance of these systems. These three activities will
combine to enable UCAR researchers in the geosciences to
better exploit the performance capabilities of this
promising architecture.
Baker, A., J. Dennis, and E.R. Jessup, 2002:
Algorithm Design for Fast Linear Solvers. Proceedings of
VECPAR 2002, 5th International Meeting on High Performance
Computing for Computational Science, Porto, Portugal,
429-442.
Dickinson, R.E., S.E. Zebiak, J.L. Anderson,
M.L. Blackmon, C. DeLuca, T.F. Hogan*, T.M. Iredell*, M. Ji*,
R.B. Rood*, M.J. Suarez*, and K.E. Taylor*, 2002: How Can We
Advance Our Weather and Climate Models as a Community?,
BAMS, 83.
Flyer, N. and P.N. Swarztrauber, 2002: The convergence
of spectral and finite difference methods for initial-boundary
value problems, SIAM J. Sci. Comput., 23,
1731-1751.
Loft, R.D., S.J. Thomas, J.M. Dennis, 2001: Terascale
Spectral Element Dynamical Core for Atmospheric General
Circulation Models. Proceedings of the Supercomputing
2001 Conference. ACM and IEEE Computer Society,
CD-ROM.
Polvani, L.M., R.K. Scott, and S.J. Thomas, 2002: An
initial-value problem for testing the dynamical cores of
atmospheric general circulation models. Submitted to Mon.
Wea. Rev.
Rivier, L., R.D. Loft, and L.M. Polvani,
2002: An efficient spectral dynamical core for distributed
memory computers. Mon. Wea. Rev., 130,
1384-1396.
Spotz, W.F. and P.N. Swarztrauber, 2001: A performance
comparison of associated Legendre projections. J. Comp.
Phys., 168, 339-355.
Swarztrauber, P.N. and W.F. Spotz, Eds., 2002: Computing
in Science and Engineering Special Issue on Climate Modeling.
Comp. in Sci. and Eng.
Swarztrauber, P.N., 2002: On computing the points and
weights for Gauss-Legendre quadrature. SIAM J. Sci.
Comput., accepted for publication.
Thomas, S.J., R.D. Loft, and J. Dennis, 2002: Parallel
Implementation Issues: Global vs. Local Methods. Submitted
to Computation in Science and Engineering.
Thomas, S.J., J. Hacker, P.K. Smolarkiewicz, and R.
Stull, 2002: Spectral preconditioners for nonhydrostatic
atmospheric models. Submitted to Mon. Wea. Rev.
Thomas, S.J., R.D. Loft, 2002: Semi-Implicit Spectral
Element Atmosphere Model, SIAM J. Sci. Comput.,
17, 339-350.
Thomas, S.J., J.M. Dennis, H.M. Tufo, and P.F. Fischer:
An overlapping Schwarz preconditioner for the cubed-sphere,
submitted to SIAM J. Sci. Comput. special issue,
Copper Mountain Conference on Iterative Methods, 2002.
|