CFRFS Poster PI Mtg 2004 by 6DJgJytp


									                                                         A Computational Facility for Reacting Flow Science

                          Abstract                                                Assembly of CCA components                                             Invariant manifolds in high-D                                          CCA CSP+AMR-solver
This project is focused on advancing capabilities for scientific
studies of chemically reacting flow on massively parallel
                                                                                  In a Reaction-Diffusion Code                                              chemical phase space                                                     Assembly
computational hardware. We are developing and using a flexible
software toolkit for reacting flow computations, where distinct
functionalities developed by experts are implemented in                                                                                                                                                                                                           Core CSP analysis
components in the context of the Common Component                                                                                                                                                                                                                 implemented in CCA &
Architecture (CCA). We are doing this using a massively parallel                                                                                                                                                                                                  runs online with reacting
C++ Adaptive Mesh Refinement framework, and are demonstrating                                                                                                                                                                                                     flow AMR solver
the efficacy of the CCA approach by assembling components into                                                                                                                                                                                                    Flexible assembly allows
multiple low Mach number codes, and participating in assembly of                                                                                                                                                                                                  CSP component reuse in
a compressible code by the SciDAC TSTC project. We are also                                                                                                                                                                                                       alternate code and data
developing and using advanced analysis/computation components                                                                                                                                                                                                     analysis CCA constructs
that enable extraction of enhanced physical understanding of
                                                                                                                                                                                                                                                                      online or offline
reacting flows from computational databases. These are based on
the use of the Computational Singular Perturbation (CSP)
technique for analysis of coupled transport-chemistry processes.
                                                                                                                                                        Manifold Dimension Mirrors                                                                                    parallel

We are implementing automatic chemical reduction with CSP and
using it as the basis for an Adaptive Chemistry tabulation-based
                                                                                                                                                             Flame Topology
computational approach. We plan to demonstrate the assembly of
these tools in computations of 2D/3D reacting flow, and to validate                                                                                                                                                                   High Order AMR
the results with respect to reacting flow databases.
                                                                                                                                                                                                                                                               Spatial discretizations of 1st &
                                                                                                                                                                                                                                                               2nd derivatives up to 8th order
                     CFRFS Team                                                    Adaptive Mesh Topology:                                                                                                                                                     Boundary conditions up to
                                                                                                                                                                                                                                                               12th order
SNL      -    Habib Najm*, Jaideep Ray, Sophia Lefantzi,
              Jeremiah Lee, Christopher Kennedy1,
              Philippe Pebay
                                                                                        H2-O2 Ignition                                                                                                                                                         Filters up to 12th order
                                                                                                                                                                                                                                                               Interpolants up to 8th order
U.Rome -      Mauro Valorani*, Francesco Creta
ICEHT -       Dimitris Goussis                                                                                                                                                                                                                                 Demonstrated up to 8th order
UCD    -      Wolfgang Kollmann                                                                                                                        Least number of exhausted modes, or highest dimensional                                                spatial convergence for heat
UCB    -      Michael Frenklach                                                                                                                        manifold, observed in the reaction zone                                                                 equation on 3 mesh levels
                                               * Institutional Point of Contact
                                                              1 formerly at SNL
                                                                                                                                                       Largest number of exhausted modes, or lowest dimensional
                                                                                                                                                       manifold, observed in the products region, as the combustion
                                                                                                                                                       products approach equilibrium
Reacting Flow Code Challenges                                                                                                                                                                                                       Computational Cost
Typical codes are inherently interconnected, forming an inseparable
                                                                                                                                                        T- Importance Index for Two
construction                                                                                                                                                                                                                                                 Rate of 4th order CPU
Use of self-contained libraries is limited to specific solvers, e.g. for
                                                                                                                                                             Vortex Strengths                                                                                increase is less than that of the
stiff integration, linear/nonlinear solvers.                                                                                                                                                                                                                 2nd order with stricter error
                                                                                                                                                                                       The importance index measures
Rearrangement of code sub-units to address different problems is                                                                                                                      influence of a process on time
often difficult, time consuming, and costly.                                                                                                                                           evolution of a state variable                                         For same error tolerance, 4th
Code upgrades are essentially global upgrades                                         H2-O2 Reaction-Diffusion                                                                            Reveals cause-and-effect
                                                                                                                                                                                                                                                             order requires less CPU Load
Code maintainability is a serious issue, particularly with regard to                                                                                                                                                                                        For same CPU Load get
AMR and massively parallel implementations.                                                    Ignition                                                                                    relationships
                                                                                                                                                                                       Importance index distributions                                       smaller errors with 4th order
                                                                                                                                                                                       for temperature are hardly
                                                                                                                                                                                       changed between the two vortex
   CFRFS Reacting Flow Vision                                                                                                                                                          cases

          & Approach
                                                                                                                                                                                       Similarly for other state variables                     Scalability
   Physical and Chemical complexity challenge                                                                                                                                                                                                               Intel/Myrinet NCSA cluster
     – Need for detailed DNS studies of physical processes in
       combustion, including ignition, flame propagation,
                                                                                                                                                   Distribution of CSP Radicals on                                                                           2D H2-O2 ignition, 100x100 base
                                                                                                                                                                                                                                                            mesh, 4 mesh levels
       extinction, …
        • Target 3D parallel high order Adaptive Mesh Refinement
                                                                                                                                                        Vortex-Pair Centerline                                                                               Space Filling Curve load balancer
          computations of low Mach number reacting flow                                                                                                                        CSP radicals are the optimal species to                                      Implicit chemistry integration
     – Need for advanced chemical analysis & reduction tools                                                                                                                   be found from linear algebraic                                               involves more CPU work per mesh
        • Develop and demonstrate Computational Singular                                                                                                                       relationships with time-integrated species                                   cell per time step in the primary
          Perturbation chemical analysis and reduction tools for                                                                                                                                                                                            flame region
          3D reacting flow – use for Adaptive Chemistry                                                                                                                        Their identification provides the
                                                                                                                                                                               corresponding simplified chemical models                                      - Adversely affects scalability
   Code and Data complexity (AMR+MPP+Chemistry) challenge
                                                                                                                                                                               at different spatiotemporal locations                                         - Specialized chemistry load
     – Need for advanced software architecture and collaboration
       with experts in other disciplines                                                                                                                                       Their distribution exhibits minor                                                  balancing is necessary
        • Adopt CCA framework, work with CCTTSS & other ISICs                                                                                                                  dependence on the vortex-pair strain-rate
     – Need advanced analysis, data mining, & visualization tools
        • Develop CSP analysis toolkit, work w SDM & CMCS ISICs
                                                                                  Demonstrated high-order spatial convergence with CCA
                                                                                  GrACE/AMR                                                                                                                                                     CFRFS Plans
                                                                                  Demonstrated 2nd-order operator-split computations with stiff
                                                                                  CVODE integration                                                                CFRFS Progress                                              Computational

                                Goals                                             Demonstrated 2D high-order AMR reaction diffusion                  Computational                                                            Low Mach number AMR momentum solver (w/ APDEC)
                                                                                  computations with detailed hydrogen chemistry and transport          Mesh, Thermo-Chemistry & Transport CCA components                       Operator-split AMR momentum+species solver
  Develop a 3D flexible massively parallel CCA-based reacting flow               Model problem: random-kernel heat-source premixed H2-O2
 computation and analysis high-order AMR code toolkit                                                                                                  High-order AMR 3D spatial derivatives & interpolants comps.             Parallel scalability and dynamic load balancing
                                                                                  ignition, kernel growth, and flame propagation
 Demonstrate Adaptive Chemistry AMR computations and chemical                                                                                         Operator-split CVODE-RKC AMR time integration comps.                    3D AMR/CCA low Mach number reacting jet flow
 analysis and reduction in multi-dimensional reacting flow                                                                                                                                                                      GrACE/CCA TSTC support
                                                                                                                                                       Demonstrated 2D high-order AMR/CCA reaction-diffusion

                                                                                      SciDAC Collaborations                                          Analysis                                                                Analysis
                               Impact                                                                                                                  CSP analysis & reduction kernel CCA components                          PRISM tabulation of CSP exhausted vectors

                                                                                        CCTTSS - CCA Software framework                               Improved CSP analysis robustness using SVD                              Adaptive PRISM-CSP Adaptive chemistry time integration
   Enable facile use of large-scale supercomputing platforms for
   advanced reacting flow studies and scientific discovery                              APDEC - Poisson solution for momentum solver                  Analysis of 2D GRImech3.0 flame-vortex data                             CSP analysis of general stochastic dynamical systems
   Improved understanding of fundamental reacting flow                                 TSTC   - Compressible DNS implementation                      Demonstrated slow spatial variability of exhausted CSP                  CSP analysis and reduction of complex fuel chemistry
   Enable the development of predictive combustion models                              CMCS - Visualization, Feature-tracking
                                                                                                                                                         vectors --- consequence to feasibility of tabulation                   GrACE/CCA/CSP data analysis toolkit
                                                                                        SDM    - Data mining, Feature-tracking
                                                                                                                                                       Demonstrated fast explicit CSP+RK integration of model stiff
                                                                                                                                                       reaction-diffusion system

                                                                                                                                                                                                                                 SciDAC PI Meeting, Charleston, SC March 22-24, 2004

To top