[mesa-users] Show the evolution of the model at each step of the solver iteration

[Ehsan Moravveji]

PS:
Bill recently preached to me how the convergence is controlled by two sets of parameters. So, let me quote from his email:


"Try changing the controls for the solver to reduce the tolerance for the max residual to a value smaller than what you are seeing.

      max_iter_for_resid_tol1 = 4
      tol_residual_norm1 = 1d-4
      tol_max_residual1 = 1d-2


If you set the tolerance extremely small, then the solver may not be able to satisfy it and after "max_iter_for_resid_tol1" iterations it will switch to the following:

      max_iter_for_resid_tol2 = 12
      tol_residual_norm2 = 1d99
      tol_max_residual2 = 1d99

This is turning off the checks for the residual and makes the solver just focus on reducing the size of the corrections to the variables.

To get terminal output about the progress of the newton solver, set report_hydro_solver_progress = .true.
For each iteration you will get a line of output including the avg and max for residuals and corrections."

I hope this helps too.

Ehsan.



On Sep 4, 2014, at 10:49 AM, Mathieu wrote:

> Hi everyone,
> 
> I was wondering if there is a way to look at the structure of a model at 
> each iteration of the newton solver, before the solution for the next 
> timestep is found.
> I'd like to make some plots of the structure before the convergence to a 
> solution, e.g. T(\rho), M(R), etc...
> The purpose of this is to show how the numerical procedure "shakes" the 
> star at every timestep to get it to the hydrostatic solution, in order 
> to illustrate how the numerics of MESA work.
> Of course, I would need it only for a few timesteps, just to have a few 
> examples.
> So, is there a way to save something like a profile*.data for each 
> iteration of the newton solver? In "evolve.f" the profile is saved at 
> the end of the step, so I guess I will not be able to get a true profile 
> for each iteration. I think what I need is what is stored in the 
> structure pointed by x in the routine newton in 
> "$MESA_DIR/star/private/star_newton.f":
> 
>     subroutine newton( &
>          s, nz, nvar, x, xold, sparse, &
>          lrd, rpar_decsol, lid, ipar_decsol, which_decsol, &
>          tol_correction_norm, &
>          xscale, equ, work, lwork, iwork, liwork, AF, &
>          lrpar, rpar, lipar, ipar, convergence_failure, ierr)
> 
>          type (star_info), pointer :: s
>          ! the primary variables
>          integer, intent(in) :: nz ! number of zones
>          integer, intent(in) :: nvar ! number of variables per zone
>          ! the total number of primary variables is neq
>          real(dp), pointer, dimension(:) :: x ! =(nvar,nz)
>          ! new vector of primaries
>          real(dp), pointer, dimension(:) :: xold ! =(nvar,nz)
>          ! old vector of primaries
> 
> So maybe I could use the routine newton_core_dump defined in star_newton.f .
> Am I correct to try to get the values pointed by x if I want to get the 
> variables for each zone at each iteration of the solver? Is there a 
> function in the run_star_extras which is called at each iteration of the 
> solver that I could use?
> 
> I appreciate any hint!
> Thanks,
> 
> Mathieu
> 
> 
> ------------------------------------------------------------------------------
> Slashdot TV.  
> Video for Nerds.  Stuff that matters.
> http://tv.slashdot.org/
> _______________________________________________
> mesa-users mailing list
> mesa-users at lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/mesa-users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.mesastar.org/pipermail/mesa-users/attachments/20140904/ba668acb/attachment.html>