Try adding -pc_factor_shift_type nonzero

> On Mar 31, 2016, at 8:43 PM, Harriet Li <kame...@gmail.com> wrote:
> 
> Hello,
> 
> I'm trying to solve the KKT system for an inverse problem with
> convection-diffusion physics. Using the input
> 
> -ksp_monitor_singular_value -ksp_gmres_modifiedgramschmidt
> -ksp_gmres_restart 500 -pc_type ilu -pc_factor_levels
> 
> I get
> 
> Solving time step 0, time = 0
> Assembling the System
> Nonlinear Residual: 0.485446
> Linear solve starting, tolerance 0.000485446
>  0 KSP Residual norm           -nan % max 1.000000000000e+00 min
> 1.000000000000e+00 max/min 1.000000000000e+00
> Linear solve finished, step 0, residual -nan
> Trying full Newton step
>  Current Residual: 0.485446
>  Shrinking Newton step to 0.618034
>  Current Residual: 0.485446
> ...
> 
> However, if I use superlu,
> 
> -ksp_type preonly -pc_type lu -pc_factor_mat_solver_package superlu
> 
> then the system is solved without problems (end residual of O(e-17)). If I
> print out the Jacobian and check it in Matlab, I see that it has full rank
> and condition number of O(e3).
> 
> Any thoughts on what's going on? I have libmesh 0.9.5 and petsc 1.3.6.1
> 
> Thanks,
> Harriet
> ------------------------------------------------------------------------------
> Transform Data into Opportunity.
> Accelerate data analysis in your applications with
> Intel Data Analytics Acceleration Library.
> Click to learn more.
> http://pubads.g.doubleclick.net/gampad/clk?id=278785471&iu=/4140
> _______________________________________________
> Libmesh-users mailing list
> Libmesh-users@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/libmesh-users

------------------------------------------------------------------------------
Transform Data into Opportunity.
Accelerate data analysis in your applications with
Intel Data Analytics Acceleration Library.
Click to learn more.
http://pubads.g.doubleclick.net/gampad/clk?id=278785471&iu=/4140
_______________________________________________
Libmesh-users mailing list
Libmesh-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/libmesh-users

Reply via email to