Question about ElementJacobianDamper

[TheGreatCid]

Check these boxes if you have followed the posting rules.

• Q&A General is the most appropriate section for my question
• I have consulted the posting Guidelines on the Discussions front page
• I have searched the Discussions forum and my question has not been asked before
• I have searched the MOOSE website and the documentation does not answer my question
• I have formatted my post following the posting guidelines (screenshots as a last resort, triple back quotes around pasted text)

Question

Hi all,

Normally, when a negative Jacobian is calculated, the time step will be cut and the step retried. However, I noticed that if you are using ElementJacobianDamper a negative Jacobian will kill the simulation entirely.
Before I put an issue up, I wanted to see if this was intended behavior.

I don't have an MWE ready to go, but if you need one, I can work on creating one.

[TheGreatCid]

Well here is the MWE

If you set line_search=none AND have the element jacobian damper, the simulation will exit instead of cutting the time step.

[Mesh]
  type = GeneratedMesh
  dim = 3
  nx=10
  ny=10
  nz=10
  elem_type = HEX8
[]

[GlobalParams]
  displacements = 'disp_x disp_y disp_z'
[]

[Functions]
  [ tdisp]
    type = ParsedFunction
    expression = '-1* t'
  [ ]
[]

[Physics]
  [SolidMechanics]
    [QuasiStatic]
      [ all]
        strain = FINITE
        add_variables = true
      [ ]
    [ ]
  [ ]
[]

[BCs]
  [ symmy]
    type = DirichletBC
    variable = disp_y
    boundary = bottom
    value = 0
  [ ]
  [ symmx]
    type = DirichletBC
    variable = disp_x
    boundary = left
    value = 0
  [ ]
  [ symmz]
    type = DirichletBC
    variable = disp_z
    boundary = back
    value = 0
  [ ]
  [ tdisp]
    type = FunctionDirichletBC
    variable = disp_z
    boundary = front
    function = tdisp
  [ ]
[]

[Materials]
  [ elasticity_tensor]
    type = ComputeElasticityTensor
    C_ijkl = '1.684e5 0.176e5 0.176e5 1.684e5 0.176e5 1.684e5 0.754e5 0.754e5 0.754e5'
    fill_method = symmetric9
  [ ]
  [ stress]
    type = ComputeFiniteStrainElasticStress
  [ ]
[]

[Preconditioning]
  [ smp]
    type = SMP
    full = true
  [ ]
[]
[Dampers]
  [jac]
    type = ElementJacobianDamper
  []
[]
[Executioner]
  type = Transient
  solve_type = NEWTON
  dt = 0.05


  petsc_options_iname = -pc_hypre_type
  petsc_options_value = boomeramg
  automatic_scaling=true
  line_search=none
  nl_abs_tol = 1e-10
  nl_rel_step_tol = 1e-10
  dtmax = 10.0
  nl_rel_tol = 1e-10

  num_steps = 1000
  nl_abs_step_tol = 1e-10
[]

[Outputs]
  print_linear_residuals=false
  exodus = true
[]

[GiudGiud]

looks like proc 0 is accessing the residual vector out of bounds most likely.
Proc1 is sitting, waiting for proc 0 to finish but 0 is stuck

debug binary is a good plan, in fact we want debug with petsc and libmesh installed from source here, to get a backtrace with line numbers

[lindsayad]

PETSc is always built with -g even --with-debugging=0

[lindsayad]

@TheGreatCid can you try #31821 and see if that fixes the hang?

[TheGreatCid]

I am having some trouble with the debugger (it's getting hung up on another unrelated issue that I am trying to fix).

However, @lindsayad, that PR works! It gets past the negative Jacobian without hanging!

[lindsayad]

great!