hpc-fenics

HPC FEniCS

Treat FEniCS as a family with two main stacks: classic FEniCS and DOLFINx.

Start

1. Read references/runtime-selection.md first.
2. Read references/dolfinx-boundary-workflow.md when targeting DOLFINx boundary conditions.
3. Read references/ufl-and-solver-patterns.md when translating PDEs, choosing spaces, or building forms.
4. Read references/pde-template-cookbook.md when mapping common PDE classes to known formulation patterns.
5. Read references/time-dependent-and-nonlinear-patterns.md when building transient or nonlinear solvers.
6. Read references/implementation-skeletons.md when you need a concrete script shape for classic FEniCS or DOLFINx.
7. Read references/petsc-solver-playbook.md when choosing linear, nonlinear, or block solver settings.
8. Read references/mixed-problems-and-output.md when working with mixed spaces, XDMF, or post-processing.
9. Read references/cluster-execution-playbook.md when staging a FEniCS or DOLFINx script for scheduler-backed cluster execution.
10. Read references/boundary-io-and-errors.md when repairing runtime failures or resolving IO issues.

Work sequence

1. Choose one stack and stay consistent:
   • classic FEniCS: fenics or dolfin
   • modern stack: dolfinx
2. Translate the PDE into the correct weak form before writing code.
3. Match unknown type and space:
   • scalar field -> scalar space
   • vector field -> vector space
   • mixed formulation -> mixed space with explicit subspace handling
4. For linear problems, use TrialFunction and TestFunction.
5. For nonlinear problems, use a Function for the unknown and solve a residual form.
6. Write outputs in a format the expected post-processor can open, usually XDMF for mesh-coupled fields.

Guardrails

• Do not mix classic FEniCS imports with DOLFINx APIs in one script.
• Do not use TrialFunction for a nonlinear unknown.
• Do not guess a boundary condition if the PDE is under-constrained; say what is missing.
• Do not ignore shape and rank mismatches in UFL expressions.