Sandia National Laboratories Tramonto

Nonlinear Solver Control Parameters

These parameters control the Newton's method settings and the load balancing options.

  • Max_Newton_Iter(int): The maximum number of Newton iterations the code will peform before exiting.
  • Newton_rel_tol(real): Relative convergence tolerance for the nonlinear solver.
  • Newton_abs_tol(real): Absolute convergence tolerance for the nonlinear solver.
  • Min_update_frac(real): During the Newton-Raphson solve, the code will mix some fraction of the new solution in with the old solution at each iteration. This parameter should be between zero and unity. It sets the minimum fraction of the new solution used. The code will actually use a fraction between Min_update_frac and 100% (of the new solution), but tries to be intelligent when Newton steps result in unacceptable values (i.e. negative densities). It is recommended that this parameter be set to 1.0 initially. It can be reduced if numerical instabilities arise. If this parameters is smaller than 0.1 convergence can be very slow.
  • Load_Bal_Flag(int): Seclect option for load balancing the Tramonto problem. Options are:
    • 0: LB_LINEAR - This results in a linear partitioning of the matrix, but a recursive bisection partitioning of the matrix fill.
    • 1: LB_BOX - This results in a recursive bisection partitioning of the matrix fill and the matrix solve.
    • 2: LB_WEIGHTS - This results in a weighted recursive bisection partitioning of the matrix fill and the matrix solve where some nodes with simple residuals equations (i.e. density=zero in a surface) are given lesser weights than the remaining fluid nodes.