NERSC's Center for Computational Science and Engineering, headed by John Bell, develops algorithms and computational models aimed at gaining a better understanding of turbulence, one of the most common -- yet least understood -- problems occurring in such processes as combustion and fluid transport.
CCSE's research is focused on high-resolution finite difference methods and adaptive mesh refinement. The primary areas of application are in fluid dynamics, such as turbulence and atmospheric models. Simulations of turbulence in combusting gases are a crucial part of current efforts to design more efficient and less polluting diesel and gasoline engines. CCSE is also investigating computational models for several types of atmospheric flow: stable boundary layers over land, mesoscale flows over complex terrain, and marine boundary layer clouds such as stratocumulus and trade cumulus.
CCSE's research involves significant software development, including BoxLib, a parallel library that will allow many higher-level codes to run efficiently in parallel without substantial code changes at the user interface level.
The group works on diverse scientific applications in the fields of computational fluid dynamics, computational physics, computational chemistry, material sciences, structural biology, life sciences, data mining, and data-intensive computing, as well as fundamental problems in scientific computing such as numerical linear algebra (e.g., dense and sparse linear/eigen system solvers), mathematical libraries and templates, and parallel programming environments.
As researchers create ever-larger banks of shared data, enabling access to that
information
over computer networks becomes increasingly important. Brian Tierney's Data Int
ensive
Distributed Computing Group is developing technologies that allow researchers t
o get there
from here.
The Bioinformatics Group, under the leadership of Manfred Zorn, carries out a wide range
of R&D to further computational science in the field of biology. Their projects include
laboratory information management systems, large-scale genome annotation, database
integration and data mining, and modeling of gene regulation.
The Future Technologies Group strives to eliminate some of the guesswork about the
immediate future of computing by researching, adapting, and experimenting with today's
latest technologies to help develop tomorrow's standards. Bill Saphir leads this effort with
help from Patrick Bozeman and Luigi Semenzato.
The group is also involved in developing software needed for scientific research, such as:
NERSC's key interest in the UC Berkeley/Intel Millennium architecture
is in researching the high-performance potential of clusters based on
SMP nodes rather than single processor nodes.
Members of the Scientific Data Management Group, led by Arie Shoshani, help scientists
avoid information overload by developing tools for managing scientific databases. The
group addresses the specialized needs of scientific applications while taking advantage of
commercially available software.
This group develops tools that enable scientists to manage and analyze massive amounts of data. Large-scale scientific simulations, experiments, and observational projects generate large multidimensional datasets. Typically, the time to access a subset from a large dataset stored on a mass storage system may take many minutes to hours. This slows down the effectiveness of data analysis to the point that much of the data may never be analyzed. The Scientific Data Management Group is collaborating with Lawrence Livermore National Laboratory and the University of Maryland on the OPTIMASS project to enable rapid analysis of earth science and environmental data such as global warming and ozone layer depletion. By reorganizing the original datasets to match their intended usage and enhancing storage server protocols, the project has improved access time up to 100 fold. The group is also participating in the DOE Grand Challenge on High Energy and Nuclear Physics Data, which is developing techniques and tools that will enable efficient access to the massive datasets from the Relativistic Heavy Ion Collider (RHIC) STAR experiments (beginning in late 1999) in the search for the quark-gluon plasma. Simulated data generated at NERSC and the RHIC Computing Facility will be used in testing and developing these tools before experiment operations begin. (click here for further discussion.)
NERSC's Visualization Group helps scientists understand the results of their experiments
and simulations by letting them see and manipulate the results in a
three-dimensional virtual
reality environment. From left: Kevin Campbell, Stephen Lau (kneeling), Terry Ligocki,
Nancy Johnston, and Wes Bethel.
One recent example of their work involves nuclear magnetic resonance (NMR), which materials scientists use to determine chemical structure (see the research highlight for more information). To better understand the results of NMR experiments, researchers are using NERSC's Cray T3E to simulate the response of crystalline structures to an applied magnetic field. The Visualization Group has developed stereoscopic displays of the results of computations on several different crystals, including two amino acids. The complex three-dimensional structure of the crystals can be very confusing when displayed on a 2-D monitor, but becomes quite obvious when viewed in stereo.
Other recent visualization projects have included DNA modeling, underground mineral and petroleum deposits, fluid dynamics turbulence studies, and quantum chromodynamics simulations.