[petsc-users] LineSearch question
Barry Smith
bsmith at mcs.anl.gov
Tue Aug 14 17:58:11 CDT 2012
Blaise,
You can run with -snes_linesearch_monitor -info -ksp_monitor_true_residual -ksp_converged_reason
to get much more information about what is happening and why the line search is failing.
If that doesn't help you can send the code and we can play with it.
Barry
On Aug 14, 2012, at 5:53 PM, Blaise Bourdin <bourdin at lsu.edu> wrote:
> HI,
>
> I am trying to understand if the following behavior is normal / expected:
>
> I am solving a quasi-static evolution where at each time step, SNESSolve is called. My validation problem is a _static_ problem with 2 time steps (i.e. 2 successive calls to SNESSolve with the same operator, jacobian, and right hand side). Furthermore, the problem is linear, so that the Jacobian is constant. I even reset the solution vector to the same value at each time step.
>
> In this setting, I am expecting that at each time step, each SNESSolve should converge in exactly one iteration no matter what linesearch / snes type I use. Indeed, when setting the linesearch to l2 or cp, this is what I get. However, for all other choices, the second call to SNESSolve fails to converge with a SNESConvergedReason of -5 or -6.
>
> The relevant code is as follow:
> Call VecSet(solTemp,0.0_Kr,ierr);CHKERRQ(ierr)
> Call FormInitialGuess(snesTemp,solTemp,MEF90Ctx,ierr);CHKERRQ(ierr)
> Call VecCopy(solTemp,tmpvec,ierr)
>
> Call SNESSolve(snesTemp,PETSC_NULL_OBJECT,solTemp,ierr);CHKERRQ(ierr)
>
> Call VecCopy(tmpvec,soltemp,ierr)
> Call SNESSolve(snesTemp,PETSC_NULL_OBJECT,solTemp,ierr);CHKERRQ(ierr)
>
>
> Is this expected? I tried to call SNESLineSearchReset before the second call to SNESSolve, but this does not seem to have any effect.
>
> Blaise
>
>
>
> Below is the sample output using cp as the linesearch type in which case I get the expected convergence:
> Solving time step 1: t= 1.00000E+00
> 0 SNES Function norm 3.742138023215e+00
> Line search terminated: lambda = 1, fnorms = 2.1936e-08
> 1 SNES Function norm 2.193598339906e-08
> SNES Object:(temp_) 1 MPI processes
> type: ls
> maximum iterations=50, maximum function evaluations=10000
> tolerances: relative=1e-08, absolute=1e-50, solution=1e-08
> total number of linear solver iterations=15
> total number of function evaluations=3
> KSP Object: (temp_) 1 MPI processes
> type: cg
> maximum iterations=10000
> tolerances: relative=1e-08, absolute=1e-50, divergence=10000
> left preconditioning
> using nonzero initial guess
> using PRECONDITIONED norm type for convergence test
> PC Object: (temp_) 1 MPI processes
> type: icc
> 0 levels of fill
> tolerance for zero pivot 2.22045e-14
> using Manteuffel shift
> matrix ordering: natural
> factor fill ratio given 1, needed 1
> Factored matrix follows:
> Matrix Object: 1 MPI processes
> type: seqsbaij
> rows=104, cols=104
> package used to perform factorization: petsc
> total: nonzeros=381, allocated nonzeros=381
> total number of mallocs used during MatSetValues calls =0
> block size is 1
> linear system matrix = precond matrix:
> Matrix Object: 1 MPI processes
> type: seqaij
> rows=104, cols=104
> total: nonzeros=658, allocated nonzeros=658
> total number of mallocs used during MatSetValues calls =0
> not using I-node routines
> SNESLineSearch Object: (temp_) 1 MPI processes
> type: cp
> maxstep=1.000000e+08, minlambda=1.000000e-12
> tolerances: relative=1.000000e-08, absolute=1.000000e-15, lambda=1.000000e-08
> maximum iterations=1
> SNESTemp converged in in 1 iterations. SNESConvergedReason is 3
> 0 SNES Function norm 3.742138023215e+00
> Line search: lambdas = [1, 0], ftys = [2.42597, 4.85193]
> Line search terminated: lambda = 2, fnorms = 2.1936e-08
> 1 SNES Function norm 2.193598717772e-08
> SNES Object:(temp_) 1 MPI processes
> type: ls
> maximum iterations=50, maximum function evaluations=10000
> tolerances: relative=1e-08, absolute=1e-50, solution=1e-08
> total number of linear solver iterations=15
> total number of function evaluations=3
> KSP Object: (temp_) 1 MPI processes
> type: cg
> maximum iterations=10000
> tolerances: relative=1e-08, absolute=1e-50, divergence=10000
> left preconditioning
> using nonzero initial guess
> using PRECONDITIONED norm type for convergence test
> PC Object: (temp_) 1 MPI processes
> type: icc
> 0 levels of fill
> tolerance for zero pivot 2.22045e-14
> using Manteuffel shift
> matrix ordering: natural
> factor fill ratio given 1, needed 1
> Factored matrix follows:
> Matrix Object: 1 MPI processes
> type: seqsbaij
> rows=104, cols=104
> package used to perform factorization: petsc
> total: nonzeros=381, allocated nonzeros=381
> total number of mallocs used during MatSetValues calls =0
> block size is 1
> linear system matrix = precond matrix:
> Matrix Object: 1 MPI processes
> type: seqaij
> rows=104, cols=104
> total: nonzeros=658, allocated nonzeros=658
> total number of mallocs used during MatSetValues calls =0
> not using I-node routines
> SNESLineSearch Object: (temp_) 1 MPI processes
> type: cp
> maxstep=1.000000e+08, minlambda=1.000000e-12
> tolerances: relative=1.000000e-08, absolute=1.000000e-15, lambda=1.000000e-08
> maximum iterations=1
> SNESTemp converged in in 1 iterations. SNESConvergedReason is 3
>
>
> when using the default linesearch (bt), the second SNESSolve fails:
>
> Solving time step 1: t= 1.00000E+00
> 0 SNES Function norm 3.742138023215e+00
> Line search: Using full step: fnorm 3.742138023215e+00 gnorm 2.193598339906e-08
> 1 SNES Function norm 2.193598339906e-08
> SNES Object:(temp_) 1 MPI processes
> type: ls
> maximum iterations=50, maximum function evaluations=10000
> tolerances: relative=1e-08, absolute=1e-50, solution=1e-08
> total number of linear solver iterations=15
> total number of function evaluations=2
> KSP Object: (temp_) 1 MPI processes
> type: cg
> maximum iterations=10000
> tolerances: relative=1e-08, absolute=1e-50, divergence=10000
> left preconditioning
> using nonzero initial guess
> using PRECONDITIONED norm type for convergence test
> PC Object: (temp_) 1 MPI processes
> type: icc
> 0 levels of fill
> tolerance for zero pivot 2.22045e-14
> using Manteuffel shift
> matrix ordering: natural
> factor fill ratio given 1, needed 1
> Factored matrix follows:
> Matrix Object: 1 MPI processes
> type: seqsbaij
> rows=104, cols=104
> package used to perform factorization: petsc
> total: nonzeros=381, allocated nonzeros=381
> total number of mallocs used during MatSetValues calls =0
> block size is 1
> linear system matrix = precond matrix:
> Matrix Object: 1 MPI processes
> type: seqaij
> rows=104, cols=104
> total: nonzeros=658, allocated nonzeros=658
> total number of mallocs used during MatSetValues calls =0
> not using I-node routines
> SNESLineSearch Object: (temp_) 1 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
> SNESTemp converged in in 1 iterations. SNESConvergedReason is 3
> 0 SNES Function norm 3.742138023215e+00
> Line search: Using full step: fnorm 3.742138023215e+00 gnorm 1.871069011453e+00
> 1 SNES Function norm 1.871069011453e+00
> Line search: Using full step: fnorm 1.871069011453e+00 gnorm 1.247379340865e+00
> 2 SNES Function norm 1.247379340865e+00
> Line search: Using full step: fnorm 1.247379340865e+00 gnorm 9.355345056231e-01
> 3 SNES Function norm 9.355345056231e-01
> Line search: Using full step: fnorm 9.355345056231e-01 gnorm 7.484276044882e-01
> 4 SNES Function norm 7.484276044882e-01
> Line search: Using full step: fnorm 7.484276044882e-01 gnorm 6.236896704016e-01
> 5 SNES Function norm 6.236896704016e-01
> Line search: Using full step: fnorm 6.236896704016e-01 gnorm 5.345911460556e-01
> 6 SNES Function norm 5.345911460556e-01
> Line search: Using full step: fnorm 5.345911460556e-01 gnorm 4.677672528597e-01
> 7 SNES Function norm 4.677672528597e-01
> Line search: Using full step: fnorm 4.677672528597e-01 gnorm 4.157931136937e-01
> 8 SNES Function norm 4.157931136937e-01
> Line search: Using full step: fnorm 4.157931136937e-01 gnorm 3.742138023528e-01
> 9 SNES Function norm 3.742138023528e-01
> Line search: Using full step: fnorm 3.742138023528e-01 gnorm 3.401943657960e-01
> 10 SNES Function norm 3.401943657960e-01
> Line search: Using full step: fnorm 3.401943657960e-01 gnorm 3.118448353285e-01
> 11 SNES Function norm 3.118448353285e-01
> Line search: Using full step: fnorm 3.118448353285e-01 gnorm 2.878567710844e-01
> 12 SNES Function norm 2.878567710844e-01
> Line search: Using full step: fnorm 2.878567710844e-01 gnorm 2.672955731592e-01
> 13 SNES Function norm 2.672955731592e-01
> Line search: Using full step: fnorm 2.672955731592e-01 gnorm 2.494758682895e-01
> 14 SNES Function norm 2.494758682895e-01
> Line search: Using full step: fnorm 2.494758682895e-01 gnorm 2.338836265275e-01
> 15 SNES Function norm 2.338836265275e-01
> Line search: Using full step: fnorm 2.338836265275e-01 gnorm 2.201257661485e-01
> 16 SNES Function norm 2.201257661485e-01
> Line search: Using full step: fnorm 2.201257661485e-01 gnorm 2.078965569222e-01
> 17 SNES Function norm 2.078965569222e-01
> Line search: Using full step: fnorm 2.078965569222e-01 gnorm 1.969546328772e-01
> 18 SNES Function norm 1.969546328772e-01
> Line search: Using full step: fnorm 1.969546328772e-01 gnorm 1.871069012364e-01
> 19 SNES Function norm 1.871069012364e-01
> Line search: Using full step: fnorm 1.871069012364e-01 gnorm 1.781970487991e-01
> 20 SNES Function norm 1.781970487991e-01
> Line search: Using full step: fnorm 1.781970487991e-01 gnorm 1.700971829468e-01
> 21 SNES Function norm 1.700971829468e-01
> Line search: Using full step: fnorm 1.700971829468e-01 gnorm 1.627016532554e-01
> 22 SNES Function norm 1.627016532554e-01
> Line search: Using full step: fnorm 1.627016532554e-01 gnorm 1.559224177048e-01
> 23 SNES Function norm 1.559224177048e-01
> Line search: Using full step: fnorm 1.559224177048e-01 gnorm 1.496855209981e-01
> 24 SNES Function norm 1.496855209981e-01
> Line search: Using full step: fnorm 1.496855209981e-01 gnorm 1.439283855764e-01
> 25 SNES Function norm 1.439283855764e-01
> Line search: Using full step: fnorm 1.439283855764e-01 gnorm 1.385977046303e-01
> 26 SNES Function norm 1.385977046303e-01
> Line search: Using full step: fnorm 1.385977046303e-01 gnorm 1.336477866088e-01
> 27 SNES Function norm 1.336477866088e-01
> Line search: Using full step: fnorm 1.336477866088e-01 gnorm 1.290392422439e-01
> 28 SNES Function norm 1.290392422439e-01
> Line search: Using full step: fnorm 1.290392422439e-01 gnorm 1.247379341700e-01
> 29 SNES Function norm 1.247379341700e-01
> Line search: Using full step: fnorm 1.247379341700e-01 gnorm 1.207141298427e-01
> 30 SNES Function norm 1.207141298427e-01
> Line search: Using full step: fnorm 1.207141298427e-01 gnorm 1.169418132858e-01
> 31 SNES Function norm 1.169418132858e-01
> Line search: Using full step: fnorm 1.169418132858e-01 gnorm 1.133981219747e-01
> 32 SNES Function norm 1.133981219747e-01
> Line search: Using full step: fnorm 1.133981219747e-01 gnorm 1.100628830937e-01
> 33 SNES Function norm 1.100628830937e-01
> Line search: Using full step: fnorm 1.100628830937e-01 gnorm 1.069182292915e-01
> 34 SNES Function norm 1.069182292915e-01
> Line search: Using full step: fnorm 1.069182292915e-01 gnorm 1.039482784783e-01
> 35 SNES Function norm 1.039482784783e-01
> Line search: Using full step: fnorm 1.039482784783e-01 gnorm 1.011388655469e-01
> 36 SNES Function norm 1.011388655469e-01
> Line search: Using full step: fnorm 1.011388655469e-01 gnorm 9.847731645400e-02
> 37 SNES Function norm 9.847731645400e-02
> Line search: Using full step: fnorm 9.847731645400e-02 gnorm 9.595225705796e-02
> 38 SNES Function norm 9.595225705796e-02
> Line search: Using full step: fnorm 9.595225705796e-02 gnorm 9.355345063171e-02
> 39 SNES Function norm 9.355345063171e-02
> Line search: Using full step: fnorm 9.355345063171e-02 gnorm 9.127165915308e-02
> 40 SNES Function norm 9.127165915308e-02
> Line search: Using full step: fnorm 9.127165915308e-02 gnorm 8.909852441151e-02
> 41 SNES Function norm 8.909852441151e-02
> Line search: Using full step: fnorm 8.909852441151e-02 gnorm 8.702646570443e-02
> 42 SNES Function norm 8.702646570443e-02
> Line search: Using full step: fnorm 8.702646570443e-02 gnorm 8.504859148402e-02
> 43 SNES Function norm 8.504859148402e-02
> Line search: Using full step: fnorm 8.504859148402e-02 gnorm 8.315862278451e-02
> 44 SNES Function norm 8.315862278451e-02
> Line search: Using full step: fnorm 8.315862278451e-02 gnorm 8.135082663716e-02
> 45 SNES Function norm 8.135082663716e-02
> Line search: Using full step: fnorm 8.135082663716e-02 gnorm 7.961995798542e-02
> 46 SNES Function norm 7.961995798542e-02
> Line search: Using full step: fnorm 7.961995798542e-02 gnorm 7.796120886084e-02
> 47 SNES Function norm 7.796120886084e-02
> Line search: Using full step: fnorm 7.796120886084e-02 gnorm 7.637016378216e-02
> 48 SNES Function norm 7.637016378216e-02
> Line search: Using full step: fnorm 7.637016378216e-02 gnorm 7.484276050661e-02
> 49 SNES Function norm 7.484276050661e-02
> Line search: Using full step: fnorm 7.484276050661e-02 gnorm 7.337525539874e-02
> 50 SNES Function norm 7.337525539874e-02
> SNES Object:(temp_) 1 MPI processes
> type: ls
> maximum iterations=50, maximum function evaluations=10000
> tolerances: relative=1e-08, absolute=1e-50, solution=1e-08
> total number of linear solver iterations=693
> total number of function evaluations=51
> KSP Object: (temp_) 1 MPI processes
> type: cg
> maximum iterations=10000
> tolerances: relative=1e-08, absolute=1e-50, divergence=10000
> left preconditioning
> using nonzero initial guess
> using PRECONDITIONED norm type for convergence test
> PC Object: (temp_) 1 MPI processes
> type: icc
> 0 levels of fill
> tolerance for zero pivot 2.22045e-14
> using Manteuffel shift
> matrix ordering: natural
> factor fill ratio given 1, needed 1
> Factored matrix follows:
> Matrix Object: 1 MPI processes
> type: seqsbaij
> rows=104, cols=104
> package used to perform factorization: petsc
> total: nonzeros=381, allocated nonzeros=381
> total number of mallocs used during MatSetValues calls =0
> block size is 1
> linear system matrix = precond matrix:
> Matrix Object: 1 MPI processes
> type: seqaij
> rows=104, cols=104
> total: nonzeros=658, allocated nonzeros=658
> total number of mallocs used during MatSetValues calls =0
> not using I-node routines
> SNESLineSearch Object: (temp_) 1 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
> SNESTemp converged in in 50 iterations. SNESConvergedReason is -5
>
>
>
>
>
> --
> 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
>
>
>
>
>
>
>
More information about the petsc-users
mailing list