Partial Spectral Information from Linear Systems to Speed-Up Numerical Simulations in Computational Fluid Dynamics

• Published in: High Performance Computing for Computational Science - VECPAR 2004 pp. 699 - 715
• Main Authors: Balsa, C., Palma, J. M. L. M., Ruiz, D.
• Format: Book Chapter Conference Proceeding
• Language: English
• Published: Berlin, Heidelberg Springer Berlin Heidelberg 2005; Springer
• Series: Lecture Notes in Computer Science
• Subjects: Applied sciences; Computer science; control theory; systems; Computer systems and distributed systems. User interface; Conjugate Gradient Algorithm; Exact sciences and technology; Invariant Subspace; Inverse Iteration; Small Eigenvalue; Software; Spectral Information; Conjugate gradient method; High performance; Gradient; Invariant subspace; Computational fluid dynamics; Information system; Iterative process; Krylov subspace method; Iterative method; Distributed computing; Modeling; Search algorithm; Inverse problem; Information use; Vector space; Flow(fluid); Eigenvalue problem; Preconditioning; Numerical convergence
• ISBN: 9783540254249; 3540254242
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/11403937_52

It was observed that all the different linear systems arising in an iterative fluid flow simulation algorithm have approximately constant invariant subspaces associated with their smallest eigenvalues. For this reason, we propose to perform one single computation of the eigenspace associated with the smallest eigenvalues, at the beginning of the iterative process, to improve the convergence of the Krylov method used in subsequent iterations of the fluid flow algorithm by means of this pre-computed partial spectral information. The Subspace Inverse Iteration Method with Stabilized Block Conjugate Gradient is our choice for computing the spectral information, which is then used to remove the effect of the smallest eigenvalues in two different ways: either building a spectral preconditioner that shifts these eigenvalues from almost zero close to the unit value, or performing a deflation of the initial residual in order to remove parts of the solution corresponding to the smallest eigenvalues. Under certain conditions, both techniques yield a reduction of the number of iterations in each subsequent runs of the Conjugate Gradient algorithm.