Migration Guide

This guide helps users transition from the Fortran MRE-Zone solver to Sentinel.jl.

File Format Compatibility

Sentinel reads and writes the same file formats as Fortran MRE-Zone:

Format	Extension	Read	Write	Notes
Node coordinates	.nod	read_nod_file	write_nod_file	3- or 4-column (with mtrltype)
Element connectivity	.elm	read_elm_file	write_elm_file	Hex27 or Tet4
Displacement	.dsp	read_dsp_file	write_dsp_file	Complex 3-component
Material properties	.mtr	read_mtr_file	write_mtr_file	Per-property files
Pressure	.pre	read_pre_file	write_pre_file	Complex nodal/elemental
Boundary nodes	.bnd	read_bnd_file	write_bnd_file	Node index list
Boundary conditions	.bcs	read_bcs_file	write_bcs_file	Per-DOF conditions
Basis indices	.basis	read_basis_file	write_basis_file	Node subset for output
Mesh indices	.meshind	read_meshind_file	write_meshind_file	Structured grid indexing

Runfile Support

Sentinel reads Fortran .dat runfiles directly:

config = parse_runfile("my_problem.dat")
setup = setup_forward_problem(config, basedir)

The RunfileConfig struct stores all parsed data. Supported problem types:

• problemtype=0: forward-only

• forwardtype=0: standard forward solve

• forwardtype=1: parameter test mode

Inverse problem runfiles (problemtype=1) will be supported in a future release.

Fortran to Julia Source Mapping

Fortran Source	Julia Module	Description
FEmaterial.f90	types/materials.jl, properties.jl	Material types and property storage
hex27.f90	elements/*.jl	Element integration kernels
tet4.f90	elements/poroelastic.jl	Poroelastic tet4 kernel
FEmesh.f90	utils/material_mesh_interp.jl	GP2MTR mapping, material mesh
fileio.f90	io/file_io.jl	File I/O routines
FEforward.f90	forward/assembly.jl, solver.jl, adjoint.jl	Assembly, solve, adjoint
FEgradient.f90	inverse/gradient.jl	Adjoint gradient computation
FEgaussnewton.f90	optimization/*.jl	CG, GN, L-BFGS optimizers
FEregularize.f90	regularization/*.jl	All 10 regularization methods
MRIhexmesh_v7p3.m	preprocessing/hex_mesh_generation.jl	Hex mesh from MRI
WashU_DTI_Process.m	preprocessing/dti_processing.jl	DTI fiber processing

API Correspondence

Assembly and Solve

Fortran	Julia
buildhex27stiffness	assemble_stiffness!(K, dh, cv_disp, cv_press, model, props, ω)
applyBC	apply_dirichlet!(K, f, bcs, dispset)
solvesystem	forward_solve!(disp, K, f, model, dispset; solver=...)
adjointsystem	adjoint_solve!(adjnt, K, calc, meas, bcs, dispset; solver=...)

Gradient and Optimization

Fortran	Julia
computegradient	compute_gradient!(grad, dh, cv_disp, calc, adjnt, model, material, gp2mtrs, meshes, ω)
computeobjective	compute_objective(calc, meas)
conjugategradient	conjugate_gradient!(material, ctx; max_iter, tol)
gaussnewton	gauss_newton!(material, ctx; max_iter, tol)
quasinewton	quasi_newton!(material, ctx; max_iter, tol, memory)

Material Model Mapping

Fortran model	Julia Type	Integer
model=1	IsotropicIncompressible()	1
model=2	Orthotropic()	2
model=3	IsotropicCompressible()	3
model=4	Poroelastic()	4
model=5	TransverseIsotropicV1()	5
model=6	TransverseIsotropicV2()	6
model=7	GeneralizedAnisotropic()	7

Convert between representations: material_model(3) returns IsotropicCompressible(), model_index(IsotropicCompressible()) returns 3.

Key Design Differences

Multiple Dispatch vs if/elseif

Fortran dispatches on material%model integer flags:

if (model == 1) then
    call hex27_isotropic_incompressible(...)
elseif (model == 2) then
    call hex27_orthotropic(...)
end if

Julia uses type dispatch:

element_stiffness!(Ke, cv, cp, ::IsotropicIncompressible, props, ω)
element_stiffness!(Ke, cv, cp, ::Orthotropic, props, ω)

Ferrite.jl Integration

Shape functions, quadrature, DOF management, and assembly use Ferrite.jl instead of custom Fortran routines. This means:

• The grid type is Grid{3, Hexahedron, Float64} (8-corner-node cells)

• Full 27-node connectivity is stored separately

• DOF numbering is managed by Ferrite's DofHandler

Hex27 Node Ordering

Fortran and Ferrite use different local node orderings for hex27 elements. Permutation arrays FORTRAN_TO_FERRITE_HEX27 and FERRITE_TO_FORTRAN_HEX27 handle the conversion automatically during mesh I/O.

Complex Symmetric Matrices

Both Fortran and Julia produce complex symmetric stiffness matrices ($K=K^T$). Test with norm(K - transpose(K)), not norm(K - K').

Solver Backends

Fortran	Julia
MUMPS (built-in)	DirectSolver() (UMFPACK, default)
	MUMPSSolver() (via MUMPS.jl extension)

The Julia DirectSolver uses SuiteSparse UMFPACK and is the default. For complex symmetric systems, MUMPS can exploit the symmetry for better performance on large problems.

What's New in Sentinel

• VTK export: export_vtk produces files for ParaView visualization

• Julia ecosystem: access to all Julia packages (plotting, AD, GPU)

• Package extensions: MUMPS, CUDA, Metal, MPI as optional backends

• Runfile round-trip: parse_runfile / write_runfile for Fortran interoperability

• Performance: Julia is 1.3-2x faster than Fortran at typical problem sizes (6×6×6)

Not Yet Ported

The following Fortran features are planned for future milestones:

• Zone-based domain decomposition (M6.1): overlapping zones for parallel assembly

• MPI parallelism (M6.2): distributed assembly and solve

• GPU acceleration (M6.3-6.4): CUDA and Metal backends

• Inverse problem runfiles (problemtype=1): currently forward-only