MFEMExecutioner

Summary

Virtual base class for executioners used to solve MFEM FE problems.

Overview

MFEMExecutioner is a purely virtual base class used to control the execution of MFEM FE problems. Importantly, use of an MFEMExecutioner derived Executioner allows for selection of the desired device and assembly strategy to use for the FE problem.

Provided MFEM and its dependencies have been built with support for GPU backends (e.g. cuda or hip), they may be selected here to enable use of GPUs for solvers and during assembly.

Selection of partial assembly levels is also supported to improve performance on GPUs; full description of the various assembly levels supported is available here.

By default, legacy assembly of the FE problem on CPUs will be selected for robustness.

Example Input File Syntax

[Executioner]
  type = MFEMSteady
  device = cpu
[]

[Executioner]
  type = MFEMSteady
  assembly_level = partial
[]

(test/tests/kernels/diffusion.i)

[Mesh]
  type = MFEMMesh
  file = gold/mug.e
  dim = 3
[]

[Problem]
  type = MFEMProblem
[]

[FESpaces]
  [H1FESpace]
    type = MFEMScalarFESpace
    fec_type = H1
    fec_order = FIRST
  []
  [HCurlFESpace]
    type = MFEMVectorFESpace
    fec_type = ND
    fec_order = FIRST
  []  
[]

[Variables]
  [concentration]
    type = MFEMVariable
    fespace = H1FESpace
  []
[]

[AuxVariables]
  [concentration_gradient]
    type = MFEMVariable
    fespace = HCurlFESpace
  []
[]

[AuxKernels]
  [grad]
    type = MFEMGradAux
    variable = concentration_gradient
    source = concentration
    execute_on = TIMESTEP_END
  []
[]

[BCs]
  [bottom]
    type = MFEMScalarDirichletBC
    variable = concentration
    boundary = '1'
    value = 1.0
  []
  [low_terminal]
    type = MFEMScalarDirichletBC
    variable = concentration
    boundary = '2'
    value = 0.0
  []
[]

[Materials]
  [Substance]
    type = MFEMGenericConstantMaterial
    prop_names = diffusivity
    prop_values = 1.0
  []
[]

[Kernels]
  [diff]
    type = MFEMDiffusionKernel
    variable = concentration
    coefficient = diffusivity
  []
[]

[Preconditioner]
  [boomeramg]
    type = MFEMHypreBoomerAMG
  []
[]

[Solver]
  type = MFEMHypreGMRES
  preconditioner = boomeramg
  l_tol = 1e-16
  l_max_its = 1000  
[]

[Executioner]
  type = MFEMSteady
  device = cpu
[]

[Outputs]
  [ParaViewDataCollection]
    type = MFEMParaViewDataCollection
    file_base = OutputData/Diffusion
    vtk_format = ASCII
  []
[]

(test/tests/kernels/diffusion_partial.i)

[Mesh]
  type = MFEMMesh
  file = gold/mug.e
  dim = 3
[]

[Problem]
  type = MFEMProblem
[]

[FESpaces]
  [H1FESpace]
    type = MFEMScalarFESpace
    fec_type = H1
    fec_order = FIRST
  []
[]

[Variables]
  [diffused]
    type = MFEMVariable
    fespace = H1FESpace
  []
[]

[BCs]
  [bottom]
    type = MFEMScalarDirichletBC
    variable = diffused
    boundary = '1'
    value = 1.0
  []
  [low_terminal]
    type = MFEMScalarDirichletBC
    variable = diffused
    boundary = '2'
    value = 0.0
  []
[]

[Materials]
  [Substance]
    type = MFEMGenericConstantMaterial
    prop_names = diffusivity
    prop_values = 1.0
  []
[]

[Kernels]
  [diff]
    type = MFEMDiffusionKernel
    variable = diffused
    coefficient = diffusivity
  []
[]

[Preconditioner]
  [jacobi]
    type = MFEMOperatorJacobiSmoother
  []
[]

[Solver]
  type = MFEMCGSolver
  preconditioner = jacobi
  l_tol = 1e-16
  l_max_its = 1000
  print_level = 2
[]

[Executioner]
  type = MFEMSteady
  assembly_level = partial
[]

[Outputs]
  [ParaViewDataCollection]
    type = MFEMParaViewDataCollection
    file_base = OutputData/DiffusionPartial
    vtk_format = ASCII
  []
[]

Summary
Overview
Example Input File Syntax