[petsc-users] SNES_DIVERGED_LS_FAILURE

[Barry Smith]

  As Matt notes you absolutely need to run with -pc_type lu -ksp_monitor_true_residual -ksp_converged_reason to make sure that the linear system solve is not the issue. It is also possible that a bug in the function evaluation (using uninitialized values or something) is causing the difficulty: run with valgrind (http://www.mcs.anl.gov/petsc/petsc-as/documentation/faq.html#valgrind) to check for these types of errors and review the code.

  Barry



On Jun 22, 2011, at 8:20 PM, Matthew Knepley wrote:

> On Wed, Jun 22, 2011 at 6:52 PM, Juha Jäykkä <juhaj at iki.fi> wrote:
> Hi all!
> 
> I am having nasty problems with SNES. As you probably guess already, my result
> is non-converging line search. I have a good candidate reason for this, but I
> do not know how to fix it.
> 
> Jacobian should be fine: -snes_ls_type test gives nice small numbers (not
> quite as small as I would like: around 1e-7).
> 
> And even if Jacobian was wrong, -snes_fd behaves in exactly the same way.
> 
> When snes gives up with its line search, the values in the residual vector
> are:
> 
> 0.00000000e+00,   1.02404374e+00,   2.28558501e-01,
> 1.94195641e-02,  -4.82401091e-02,  -7.07456591e-02,
> -7.64790651e-02,  -7.55281623e-02,  -7.19176452e-02
> 
> before they stabilise to interval (1.e-5, 8e-2). Now, my candidate is the huge
> (compared to others) value of the rhs at the second lattice point and slightly
> smaller, but still large at the third. The diverged solution vector grows very
> fast in the beginning and then settles to an almost constant value after about
> one tenth of the lattice, so my guess is that the big change needed at i=1
> causes precision to be lost at i>>1 and therefore line search fails.
> 
> Do you have any ideas how to get around this or if my theory is even correct?
> 
> Its hard to say anything without knowing what equations are being solved. However,
> does the linear problem converge? If so, you could try continuation in a parameter
> in the nonlinear term.
> 
>    Matt
>  
> Trying trust region search does not really help: it suffers from the same
> deficiency - except that it thinks it converges even though the function norm
> is of the order of 0.5. (It does fare slightly better, though, its rhs starts
> like
> 
> 0.00000000e+00,   3.93745063e-01,   1.79187741e-01,
> 1.03100385e-01,   6.76686117e-02,   4.82967347e-02,
> 3.65243377e-02,   2.88059666e-02,   2.34452314e-02,
> 
> which is better, but still not good enough.)
> 
> Cheers,
> Juha
> 
> --
>                 -----------------------------------------------
>                | Juha Jäykkä, juhaj at iki.fi                     |
>                | http://www.maths.leeds.ac.uk/~juhaj           |
>                 -----------------------------------------------
> 
> 
> 
> -- 
> What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.
> -- Norbert Wiener