Please send the output of -snes_view.
pasted below. anything stand out?


SNES Object: 1 MPI process
  type: newtonls
  maximum iterations=1, maximum function evaluations=-1
  tolerances: relative=0.1, absolute=1e-15, solution=1e-15
  total number of linear solver iterations=20
  total number of function evaluations=22
  norm schedule ALWAYS
  Jacobian is never rebuilt
  Jacobian is applied matrix-free with differencing
  Preconditioning Jacobian is built using finite differences with coloring
  SNESLineSearch Object: 1 MPI process
    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: 1 MPI process
    type: gmres
      restart=30, using Classical (unmodified) Gram-Schmidt Orthogonalization with no iterative refinement
      happy breakdown tolerance 1e-30
    maximum iterations=20, initial guess is zero
    tolerances:  relative=0.1, absolute=1e-15, divergence=10.
    left preconditioning
    using PRECONDITIONED norm type for convergence test
  PC Object: 1 MPI process
    type: asm
      total subdomain blocks = 1, amount of overlap = 0
      restriction/interpolation type - RESTRICT
      Local solver information for first block is in the following KSP and PC objects on rank 0:
      Use -ksp_view ::ascii_info_detail to display information for all blocks
    KSP Object: (sub_) 1 MPI process
      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: (sub_) 1 MPI process
      type: ilu
        out-of-place factorization
        0 levels of fill
        tolerance for zero pivot 2.22045e-14
        matrix ordering: natural
        factor fill ratio given 1., needed 1.
          Factored matrix follows:
            Mat Object: (sub_) 1 MPI process
              type: seqbaij
              rows=16384, cols=16384, bs=16
              package used to perform factorization: petsc
              total: nonzeros=1277952, allocated nonzeros=1277952
                  block size is 16
      linear system matrix = precond matrix:
      Mat Object: (sub_) 1 MPI process
        type: seqbaij
        rows=16384, cols=16384, bs=16
        total: nonzeros=1277952, allocated nonzeros=1277952
        total number of mallocs used during MatSetValues calls=0
            block size is 16
    linear system matrix followed by preconditioner matrix:
    Mat Object: 1 MPI process
      type: mffd
      rows=16384, cols=16384
        Matrix-free approximation:
          err=1.49012e-08 (relative error in function evaluation)
          Using wp compute h routine
              Does not compute normU
    Mat Object: 1 MPI process
      type: seqbaij
      rows=16384, cols=16384, bs=16
      total: nonzeros=1277952, allocated nonzeros=1277952
      total number of mallocs used during MatSetValues calls=0
          block size is 16

On Thu, May 4, 2023 at 8:30 AM Mark Adams <mfadams@lbl.gov> wrote:
If you are using MG what is the coarse grid solver?
-snes_view might give you that.

On Thu, May 4, 2023 at 8:25 AM Matthew Knepley <knepley@gmail.com> wrote:
On Thu, May 4, 2023 at 8:21 AM Mark Lohry <mlohry@gmail.com> wrote:
Do they start very similarly and then slowly drift further apart?

Yes, this. I take it this sounds familiar?

See these two examples with 20 fixed iterations pasted at the end. The difference for one solve is slight (final SNES norm is identical to 5 digits), but in the context I'm using it in (repeated applications to solve a steady state multigrid problem, though here just one level) the differences add up such that I might reach global convergence in 35 iterations or 38. It's not the end of the world, but I was expecting that with -np 1 these would be identical and I'm not sure where the root cause would be.

The initial KSP residual is different, so its the PC. Please send the output of -snes_view. If your ASM is using direct factorization, then it
could be randomness in whatever LU you are using.

  Thanks,

    Matt
 
  0 SNES Function norm 2.801842107848e+04
    0 KSP Residual norm 4.045639499595e+01
    1 KSP Residual norm 1.917999809040e+01
    2 KSP Residual norm 1.616048521958e+01
[...]
   19 KSP Residual norm 8.788043518111e-01
   20 KSP Residual norm 6.570851270214e-01
  Linear solve converged due to CONVERGED_ITS iterations 20
  1 SNES Function norm 1.801309983345e+03
Nonlinear solve converged due to CONVERGED_ITS iterations 1


Same system, identical initial 0 SNES norm, 0 KSP is slightly different

  0 SNES Function norm 2.801842107848e+04
    0 KSP Residual norm 4.045639473002e+01
    1 KSP Residual norm 1.917999883034e+01
    2 KSP Residual norm 1.616048572016e+01
[...]
   19 KSP Residual norm 8.788046348957e-01
   20 KSP Residual norm 6.570859588610e-01
  Linear solve converged due to CONVERGED_ITS iterations 20
  1 SNES Function norm 1.801311320322e+03
Nonlinear solve converged due to CONVERGED_ITS iterations 1

On Wed, May 3, 2023 at 11:05 PM Barry Smith <bsmith@petsc.dev> wrote:

  Do they start very similarly and then slowly drift further apart? That is the first couple of KSP iterations they are almost identical but then for each iteration get a bit further. Similar for the SNES iterations, starting close and then for more iterations and more solves they start moving apart. Or do they suddenly jump to be very different? You can run with -snes_monitor -ksp_monitor 

On May 3, 2023, at 9:07 PM, Mark Lohry <mlohry@gmail.com> wrote:

This is on a single MPI rank. I haven't checked the coloring, was just guessing there. But the solutions/residuals are slightly different from run to run.

Fair to say that for serial JFNK/asm ilu0/gmres we should expect bitwise identical results?


On Wed, May 3, 2023, 8:50 PM Barry Smith <bsmith@petsc.dev> wrote:

  No, the coloring should be identical every time. Do you see differences with 1 MPI rank? (Or much smaller ones?).



> On May 3, 2023, at 8:42 PM, Mark Lohry <mlohry@gmail.com> wrote:
>
> I'm running multiple iterations of newtonls with an MFFD/JFNK nonlinear solver where I give it the sparsity. PC asm, KSP gmres, with SNESSetLagJacobian -2 (compute once and then frozen jacobian).
>
> I'm seeing slight (<1%) but nonzero differences in residuals from run to run. I'm wondering where randomness might enter here -- does the jacobian coloring use a random seed?




--
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