Sandia National Laboratories Tramonto
printLeftMenu(); ?>

Linear Solver Control Parameters

  • L_Schur(int): A logical (0=FALSE; 1=TRUE) to turn on segregated Schur solver approach of Heroux, Salinger and Frink (SIAM, 2007). These solvers are uniquely designed to take advantage of the structure of DFT problems. They scale well in parallel, are generally faster than generic approaches, and can be run on smaller platforms because they consume less memory. It is recommended that this parameter be set to TRUE (1) unless numerical studies are being performed or numerical instabilities are observed with the Schur methods.
  • The parameters below define how the iterative linear solves (AztecOO) will be performed when the generic solver is turned on (LSchur=FALSE). More information on these parameters and additional options are available in the AztecOO solver package can be found in the AztecOO Users Guide.
  • AZ_solver(int): The specific AztecOO solver to be used. The most common choice is the GMRES solver (option 0). (See the Aztec00 manual). Tramonto options are:
    • 0: GMRES
    • 1: Conjugate Gradient
    • 2: tfqmr
    • 3: cg2
    • 4: bicstab
  • AZ_kspace(int): Krylov subspace size for restarted GMRES. Typical range is 20-200. Stability can be improved as AZ_kspace is increased, but the calculation becomes slower as well (see AzteccOO manual).
  • AZ_scaling(int): Type of matrix scaling to use (see AzteccOO manual).
    • -1: none
    • 0: row sum scaling
    • 1: Jacobi scaling
    • 2: symmetric row sum scaling
  • AZ_precond(int): Type of preconditioning to apply to solve (see AztecOO manual). Options are:
    • -1: None
    • 0: ILU
    • 1: Jacobi
    • 2: Symmetric Gauss Seidel
    • 3: LSpoly3
    • 4: ILUT
  • AZ_ilut_param(int): Fill in parameter when an ilut preconditioner is used (see AztecOO manual).
  • Max_gmres(int): Maximum number of iterations allowed in linear solve (see AzteccOO manual).
  • Az_tolerance(real): Convergence tolerance for linear solve (see AztecOO manual).