Mesh Refinement vs InterfaceKernel

[sid-mungale]

I am trying to set up a diffusion problem using TMAP8's ADInterfaceSorption at the interface between two blocks.
I want to be able to refine the mesh at the interface to resolve the interfacekernel properly.
I have found two avenues to do this BUT both of them result in the interfacekernel 'stopping' working, and I have been unable to isolate the cause:

1. Using the RefineSidesetGenerator block in Mesh
2. Using an Adaptivity block for initial refinement, and a BoundaryMarker at the interface

Some additional possibly helpful information:

• After refinement no flux passes through the interface.
• Turning off the block responsible for refinement fixes this.
• Running --mesh-only to spit out a refined mesh and reading from file still doesn't work

For reference, I have added an example input deck below. Commenting out the 'refine_sideset' block gives me the 'expected' behaviour.

n_sorption = 0.5
Ea = 27905
k0 = 0.271

[Mesh]
  [gen]
    type = GeneratedMeshGenerator
    nx = 10
    ny = 10
    dim = 2
  []
  [block1]
    type = SubdomainBoundingBoxGenerator
    block_id = 1
    bottom_left = '0 0 0'
    top_right = '0.5 1 0'
    input = gen
  []
  [block2]
    type = SubdomainBoundingBoxGenerator
    block_id = 2
    bottom_left = '0.5 0 0'
    top_right = '1 1 0'
    input = block1
  []
  [breakmesh]
    input = block2
    type = BreakMeshByBlockGenerator
    block_pairs = '1 2'
    split_interface = true
    add_interface_on_two_sides = true
  []
  [refine_sideset]
    type = RefineSidesetGenerator
    boundaries = 'Block2_Block1'
    boundary_side = 'both'
    input = breakmesh
    refinement = '1'
  []
[]

[Variables]
  [c_gas]
    family = LAGRANGE
    block = 1
    initial_condition = 1
  []
  [c_sol]
    family = LAGRANGE
    block = 2
    initial_condition = 0
  []
[]

[AuxVariables]
  [temperature]
    initial_condition = 700
  []
[]

[Kernels]
  [dt_gas]
    type = ADTimeDerivative
    variable = c_gas
    block = 1
  []
  [dt_solid]
    type = ADTimeDerivative
    variable = c_sol
    block = 2
  []
  [diff_gas]
    type = ADMatDiffusion
    variable = c_gas
    diffusivity = diffusivity
    block = 1
  []
  [diff_solid]
    type = ADMatDiffusion
    variable = c_sol
    diffusivity = diffusivity
    block = 2
  []
[]

[InterfaceKernels]
  [sievert]
    type = ADInterfaceSorption
    n_sorption = ${n_sorption}
    Ea = ${Ea}
    K0 = ${k0}
    diffusivity = diffusivity
    temperature = temperature
    variable = c_sol
    neighbor_var = c_gas
    boundary = Block2_Block1
  []
[]

[BCs]
  [gas_source]
    type = ADDirichletBC
    value = 1
    variable = c_gas
    boundary = left
  []
  [solid_dirichlet]
    type = ADDirichletBC
    value = 0
    variable = c_sol
    boundary = right
  []
[]

[Materials]
  [block1]
    # generic diffusivity for block 1
    type = ADGenericConstantMaterial
    prop_names = 'diffusivity'
    prop_values = 1e-3
    block = 1
  []
  [block2]
    # generic diffusivity for block 2
    type = ADGenericConstantMaterial
    prop_names = 'diffusivity'
    prop_values = 1e-3
    block = 2
  []
[]

[Preconditioning]
  [smp]
    type = SMP
    full = true
  []
[]

[Outputs]
  exodus = true
[]

[Executioner]
  type = Transient
  solve_type = NEWTON

  petsc_options_iname = '-pc_type -pc_factor_mat_solver_package'
  petsc_options_value = 'lu superlu_dist'
  line_search = none

  nl_rel_tol = 1e-15
  nl_abs_tol = 1e-9
  l_tol = 1e-3

  dt = 1e0
  end_time = 1e2
[]

[GiudGiud]

Hello

Running --mesh-only to spit out a refined mesh and reading from file still doesn't work

this is definitely not supported for reading from Exodus. We can only read back libmesh-refined meshes using the Checkpoint format

I'll take a look.

[sid-mungale]

Do you mean using the RefineSidesetGenerator and then outputting the mesh is bad practice?
I will also clarify I had only tried this to try and separate out the 'refinement' step to see if it changed anything.

[GiudGiud]

Do you mean using the RefineSidesetGenerator and then outputting the mesh is bad practice?

you can output to look at it, but you cant reload it for a simulation because it has hanging nodes.

[cticenhour]

@sid-mungale Just to recap the conversation you had with me and @simopier earlier today, you should try the SideSetsBetweenSubdomainsGenerator object to create that interface, as the break in the mesh introduced by the BreakMeshByBlockGenerator might be causing inconsistencies across the interface (i.e., different node positioning as the mesh is refined adaptively on either side not allowing the interface kernel to work properly).

Try something like this example to create the interface:

[Mesh]
  [generated]
    type = GeneratedMeshGenerator
    dim = 1
    nx = ${nb_segments_TMAP8}
    xmax = ${long_total}
  []
  [enclosure_1]
    type = SubdomainBoundingBoxGenerator
    input = generated
    block_id = 1
    bottom_left = '0 0 0'
    top_right = '${fparse 1/3 * long_total} 0 0'
  []
  [enclosure_2]
    type = SubdomainBoundingBoxGenerator
    input = enclosure_1
    block_id = 2
    bottom_left = '${fparse 1/3 * long_total} 0 0'
    top_right = '${fparse long_total} 0 0'
  []
  [interface]
    type = SideSetsBetweenSubdomainsGenerator
    input = enclosure_2
    primary_block = 1
    paired_block = 2
    new_boundary = interface
  []
  [interface2]
    type = SideSetsBetweenSubdomainsGenerator
    input = interface
    primary_block = 2
    paired_block = 1
    new_boundary = interface2
  []
[]

[GiudGiud]

BBMG breaks the mesh at the boundary. The default IKs without special treatment wont work with a non-conformal mesh afaik.
We can do mortar or the thin layer-like treatment on such a mesh

@cticenhour 's advice is the easier path forward