[petsc-users] Line search: Initial direction and size is 0

[Barry Smith]

   Blaise,

    First see what the linear solver is doing. Run with -ksp_monitor_true_residual -ksp_converged_reason

    Also run with -snes_vi_monitor to see how much of the solution is on the constraints (for example all of them?).

    Without thinking about it much I don't think this is suppose to happen; I think with the reduced space method the search direction can only be exactly 0 when the function norm is exactly 0.

   Barry

On May 9, 2013, at 11:04 AM, Blaise A Bourdin <bourdin at lsu.edu> wrote:

> Hi,
> 
> I am trying to solve a quadratic constrained optimization problem with SNESVI, i.e.
> F(v) < 0 if v < 0
> F(v) = 0 if 0 < v < 1
> F(v) > 0 if v > 0
> 
> I am reasonably sure that my function and Jacobian matrix evaluation are correct. In particular, SNES (i.e. unconstrained minimization) works just fine.
> In some situation, I get the following error in snesvi when running with -snes_monitor -snes_linesearch_monitor:
> 
>  0 SNES Function norm 2.970161116676e+00 
>      Line search: Using full step: fnorm 2.970161116676e+00 gnorm 1.028739442283e-01
>  1 SNES Function norm 1.028739442283e-01 
>      Line search: Using full step: fnorm 1.028739442283e-01 gnorm 6.789609435880e-08
>  2 SNES Function norm 6.789609435880e-08 
>      Line search: Initial direction and size is 0
> 
> The output of snes_view is the following:
> SNES Object:(V_) 24 MPI processes
>  type: virs
>  maximum iterations=50, maximum function evaluations=10000
>  tolerances: relative=1e-08, absolute=1e-08, solution=1e-08
>  total number of linear solver iterations=25
>  total number of function evaluations=3
>  KSP Object:  (V_)   24 MPI processes
>    type: cg
>    maximum iterations=10000, initial guess is zero
>    tolerances:  relative=1e-08, absolute=1e-08, divergence=10000
>    left preconditioning
>    using PRECONDITIONED norm type for convergence test
>  PC Object:  (V_)   24 MPI processes
>    type: bjacobi
>      block Jacobi: number of blocks = 24
>      Local solve is same for all blocks, in the following KSP and PC objects:
>    KSP Object:    (V_sub_)     1 MPI processes
>      type: preonly
>      maximum iterations=10000, initial guess is zero
>      tolerances:  relative=1e-05, absolute=1e-50, divergence=10000
>      left preconditioning
>      using NONE norm type for convergence test
>    PC Object:    (V_sub_)     1 MPI processes
>      type: ilu
>        ILU: out-of-place factorization
>        0 levels of fill
>        tolerance for zero pivot 2.22045e-14
>        using diagonal shift to prevent zero pivot
>        matrix ordering: natural
>        factor fill ratio given 1, needed 1
>          Factored matrix follows:
>            Matrix Object:             1 MPI processes
>              type: seqaij
>              rows=0, cols=0
>              package used to perform factorization: petsc
>              total: nonzeros=0, allocated nonzeros=0
>              total number of mallocs used during MatSetValues calls =0
>                not using I-node routines
>      linear system matrix = precond matrix:
>      Matrix Object:       1 MPI processes
>        type: seqaij
>        rows=0, cols=0
>        total: nonzeros=0, allocated nonzeros=0
>        total number of mallocs used during MatSetValues calls =0
>          not using I-node routines
>    linear system matrix = precond matrix:
>    Matrix Object:     24 MPI processes
>      type: mpiaij
>      rows=1890, cols=1890
>      total: nonzeros=42208, allocated nonzeros=42208
>      total number of mallocs used during MatSetValues calls =0
>        not using I-node (on process 0) routines
>  SNESLineSearch Object:  (V_)   24 MPI processes
>    type: bt
>      interpolation: cubic
>      alpha=1.000000e-04
>    maxstep=1.000000e+08, minlambda=1.000000e-12
>    tolerances: relative=1.000000e-08, absolute=1.000000e-15, lambda=1.000000e-08
>    maximum iterations=40
> 
> 
> I really don't know where to start digging. Any suggestion?
> 
> Blaise
> 
> -- 
> Department of Mathematics and Center for Computation & Technology
> Louisiana State University, Baton Rouge, LA 70803, USA
> Tel. +1 (225) 578 1612, Fax  +1 (225) 578 4276 http://www.math.lsu.edu/~bourdin
> 
> 
> 
> 
> 
> 
> 
>