[petsc-users] linesearch problem in snes composite

[Matthew Knepley]

On Wed, Jun 15, 2016 at 10:31 AM, Xiangdong <epscodes at gmail.com> wrote:

> Hello everyone,
>
> I am using the snes composite with the following options:
> -snes_type composite
> -snes_composite_type multiplicative
> -snes_composite_sneses newtonls
> -snes_linesearch_type bt
> -sub_0_snes_linesearch_type basic
> -snes_linesearch_monitor
> -sub_0_snes_linesearch_monitor
>
> I only use single newtonls in the composite for helping me understand how
> the composite works. By specifying the -snes_linesearch_type bt, I am
> expecting the norm decrease in each iterations. Moreover,
> -snes_linesearch_monitor should print the information on the step size.
> However, when I ran the codes, I found that no information on the step size
> of linesearch is output and the residual norm sometimes is NOT decreasing
> given the bt option provided. It seems that the snes_linesearch_bt is not
> working in this setting or I messed up the options somewhere.
>
> The linesearch works in the sub_0 level if I specify
> the -sub_0_snes_linesearch_type bt. I do not understand why it does not
> work in the overall level.
>

SNESComposite does not use a line search. I have not seen any theory to
suggest that this is a good idea.

You could get the effect of a line search with SNESComposite by left
preconditioning NRICH with it.

  Thanks,

    Matt


> Here is the snes norm information as well as the output of snes_view. Is
> there a way to get the linesearch working on the overall level?
>
>
> 0 SNES Function norm 1.436371027798e+04
>     0 SNES Function norm 1.436371027798e+04
>     1 SNES Function norm 1.562640838177e+04
>   1 SNES Function norm 1.562640838177e+04
>     0 SNES Function norm 1.562640838177e+04
>     1 SNES Function norm 5.101254678006e+03
>   2 SNES Function norm 5.101254678006e+03
>     0 SNES Function norm 5.101254678006e+03
>     1 SNES Function norm 1.287999851800e+03
>   3 SNES Function norm 1.287999851800e+03
>     0 SNES Function norm 1.287999851800e+03
>     1 SNES Function norm 3.781864044552e+02
>   4 SNES Function norm 3.781864044552e+02
>     0 SNES Function norm 3.781864044552e+02
>     1 SNES Function norm 9.895411709407e+01
>   5 SNES Function norm 9.895411709407e+01
>     0 SNES Function norm 9.895411709407e+01
>     1 SNES Function norm 6.379856588223e+00
>   6 SNES Function norm 6.379856588223e+00
> Nonlinear solve converged due to CONVERGED_FNORM_RELATIVE iterations 6
> SNES Object: 256 MPI processes
>   type: composite
>   Composite SNES type - MULTIPLICATIVE
>   SNESes on composite preconditioner follow
>   ---------------------------------
>     SNES Object:    (sub_0_)     256 MPI processes
>       type: newtonls
>       maximum iterations=1, maximum function evaluations=10000
>       tolerances: relative=1e-08, absolute=1e-50, solution=1e-08
>       total number of linear solver iterations=188
>       total number of function evaluations=1
>       SNESLineSearch Object:      (sub_0_)       256 MPI processes
>         type: basic
>         maxstep=1.000000e+08, minlambda=1.000000e-12
>         tolerances: relative=1.000000e-08, absolute=1.000000e-15,
> lambda=1.000000e-08
>         maximum iterations=40
>       KSP Object:      (sub_0_)       256 MPI processes
>         type: bcgsl
>           BCGSL: Ell = 2
>           BCGSL: Delta = 0
>         maximum iterations=5000, initial guess is zero
>         tolerances:  relative=0.001, absolute=1e-50, divergence=10000
>         left preconditioning
>         using PRECONDITIONED norm type for convergence test
>       PC Object:      (sub_0_)       256 MPI processes
>         type: sor
>           SOR: type = local_symmetric, iterations = 1, local iterations =
> 1, omega = 1
>         linear system matrix = precond matrix:
>         Mat Object:        jac         256 MPI processes
>           type: mpiaij
>           rows=2244000, cols=2244000, bs=2
>           total: nonzeros=3.112e+07, allocated nonzeros=3.112e+07
>           total number of mallocs used during MatSetValues calls =0
>   ---------------------------------
>   maximum iterations=10, maximum function evaluations=10000
>   tolerances: relative=0.001, absolute=1e-50, solution=1e-08
>   total number of linear solver iterations=0
>   total number of function evaluations=1
>   SNESLineSearch Object:   256 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
>   KSP Object:   256 MPI processes
>     type: gmres
>       GMRES: restart=30, using Classical (unmodified) Gram-Schmidt
> Orthogonalization with no iterative refinement
>       GMRES: happy breakdown tolerance 1e-30
>     maximum iterations=10000, initial guess is zero
>     tolerances:  relative=1e-05, absolute=1e-50, divergence=10000
>     left preconditioning
>     using DEFAULT norm type for convergence test
>   PC Object:   256 MPI processes
>     type: lu
>     PC has not been set up so information may be incomplete
>       LU: out-of-place factorization
>       tolerance for zero pivot 2.22045e-14
>       matrix ordering: natural
>
> Thank you.
>
> Best,
> Xiangdong
>
>
>
>
>
>


-- 
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
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20160615/781fa332/attachment.html>