Alumni Project
A Computational Facility for Reacting
Flow Science (CFRFS)
H.N. Najm1(PI), J. Ray1, C. Kennedy1,
S. Lefantzi1, P. Pébay1, J. Lee1,
M. Valorani2, D. Goussis3, W. Kollmann4,
M. Frenklach5.
ISIC Affiliates: R. Armstrong, B. Alan (CCTTSS), L. Rahn (CMCS),
D. Keyes, R. Falgout, R. Serban (TOPS), P. Colella (APDEC), A. Shoshani, J. Wu (SDM),
D. Brown, B. Henshaw (TSTT). Application Affiliates: A. Trouvé (TSTC)
1Sandia National Laboratories, 2University of Rome-Italy,
3ICEHT-Greece, 4University of California-Davis,
5University of California-Berkeley.
Summary
Our focus is the development of a modular software toolkit that makes use of
the Common Component Architecture for assembling scalable massively-parallel adaptive
mesh refinement reacting flow computation and chemical analysis codes. This development
is crucial for enabling efficient computational studies of reacting flow on terascale
hardware, and for allowing the extraction of physical insights from resulting databases.
These studies will address a wide range of chemical/reacting flow questions of interest
and of practical importance, improving our understanding of turbulence-chemistry
interactions in combustion.
This work is motivated by the need for new approaches to the development of scientific
chemically reacting flow codes. This need arises because of the challenges inherent in
the development and maintenance of parallel reacting flow codes with adaptive mesh
refinement (AMR) technology and dynamic load balancing, particularly in the context
of massively parallel terascale computing. The complexity of large-scale reacting flow
databases also requires the development of new analysis techniques for extraction of
physical insights and scientific discovery.
Based on these needs, our goal is to develop a Computational Facility for
Reacting Flow Science (CFRFS) which has at its core a new approach for assembling
massively parallel reacting flow computation and analysis codes using a modular
software toolkit. We use the common component architecture (CCA) software framework.
We are developing flexible, & reusable "components" of this toolkit for enabling
the assembly of different reacting flow codes. CCA provides standardized component
interfaces and connectivity. It alters the conventional code development paradigm
into one of specialized modular component development. We use the computational
singular perturbation (CSP) theory as the basis for data analysis components
development, to allow automated chemical data analysis and model reduction.
We have made significant progress toward both the computational and CSP
analysis component development and demonstration. We have redesigned existing codes
where appropriate. Otherwise, new algorithms are being coded into components. On the
computational side, we’ve developed components for reaction-diffusion (RD)
computations in two dimensions using AMR. We have adopted an existing parallel AMR
library (GrACE) and implemented it as our mesh component. We have developed a
thermochemistry component for general detailed chemistry computations,
and a transport component for various diffusion coefficient evaluations.
We have also built integrator components to allow efficient time integration
of detailed chemical kinetic models on adaptive meshes. This has involved development
of novel high-fidelity numerical integration techniques for the flow equations. Given
the high accuracy requirements of direct numerical simulation of turbulent reacting
flow, we have also developed advanced high-order numerical discretization
and interpolation components for adaptive meshes. These components were
assembled into codes using the CCA drag-and-drop visual programming environment.
This approach enables the assembly/utilization of advanced parallel reacting flow
codes by non-expert scientists. We have demonstrated these tools for RD modeling of
hydrogen-oxygen ignition and for shock physics.
We are presently working on the high-order implementation of AMR solver
components for the low Mach number flow momentum equations, as enabled by collaborations
with four ISICs : CCTTSS, TOPS, APDEC & TSTT. We are also working on theoretical
developments of computational formulations that ensure high-accuracy in low Mach number
flow modeling. We are testing scalability and performance of our existing components
on larger sets of processors. We are expanding the capabilities of the transport
subsystem to span various transport libraries in the literature to allow a choice of
alternate transport models. We are also continuing work on high-order AMR discretization
and interpolation stencils.
On the data analysis side, we have worked on the extension of CSP theory to
encompass multidimensional reacting flow. This allowed the analysis of coupled
chemical-transport processes in reacting flow, the automatic identification of time
scales and equilibrium manifolds, the identification of causal relationships among
various evolving processes, & automatic local chemical model reduction based on
specified error thresholds. We also implemented iterative refinement techniques that
increase the accuracy of the CSP analysis. We have implemented component versions of
parts of the CSP tools, with a design that couples the analysis components to the RD
computation components. The availability of these components is an example of the
potential of the present toolkit for maximal code reuse.
We are presently working on component implementations of the remaining
CSP analysis subsystems in order to arrive at a componentized full analysis suite
that includes feature tracking components under development in collaboration with
the SDM ISIC. We are also investigating more comprehensive means of handling coupled
reaction-transport processes in CSP, that handles wide ranges of transport time scales.
By the end of FY03, our plan is to complete initial development of all
necessary components for low Mach number reacting flow computations, and CSP data
analysis. We will also begin development of adaptive tabulation components for time
integration. This will be followed in FY04 by optimization and high-order demonstrations
of the computational components, joint demonstrations with TSTC, and the development
and demonstration of CSP mechanism reduction components.
This project has lead to 2 journal articles, 4 refereed proceedings articles,
5 conference presentations, and 3 invited talks.
For further information on this subject contact:
Dr. Habib N. Najm, Project Lead
Sandia National Laboratories
Phone: 925-294-2054
hnnajm@ca.sandia.gov
http://www.ca.sandia.gov/cfrfs
back to project page
