Adaptive Chemical Model Reduction using Computational Singular Perturbation (CSP) Analysis SciDAC PI Meeting, San Francisco, CA June 26-20, 2005 H.N. Najm, J.C. Lee* M. Valorani, F. Creta D. Goussis M. Frenklach Sandia National Laboratories Universita di Roma “La Sapienza” Agiou Georgiou 49 University of California Livermore, CA, USA Rome, Italy Patra, Greece Berkeley, CA, USA *Present address: United Technologies Research Center, East Hartford CT Acknowledgement: US Dept. of Energy, Basic Energy Sciences; SciDAC Computational Chemistry Premixed flame speed over a range of premixed CH 4 -air flame stoichiometry, with a number of simplified mechanisms Identification of low dimensional manifolds in chemical systems Flame topology is reflected in the spatial distribution of manifold dimension Identification of slow and fast chemical and transport time scales Transport time scales are slower than the set of fast chemical time scales Automatic reduction of chemical models Alternative simplifications governed by CSP tolerance Identification of comprehensive simplified chemical models Computational speed-up is quadratic in the number of species, linear in the number of reactions Convergence of predicted premixed flame speed with increasing number of reactions in the model Coupling CSP with Piecewise Reusable Implementation of Solution Mapping (PRISM) enables adaptive chemistry construction CSP solver provides for efficient explicit time integration of stiff chemical systems: • Filter out exhausted fast time scales • Integrate slow modes explicitly • Projection correction for exhausted fast modes • Effective use of adaptively reduced chemical models PRISM provides: • Discretization of chemical composition space into hypercubes • Tabulation of local response surface fits for components of CSP tensors in each hypercube • Fast identification of local CSP manifolds CSP solver for a transport-chemistry system Adaptive chemistry integration using CSP-PRISM exhibits good accuracy