Files
FemLite/todo.md
T
shenw 75d4e735b1
FemLite CI / build-test (push) Has been cancelled
Release FemLite 0.5.2 V5 closeout
2026-07-01 14:51:40 +08:00

2143 lines
128 KiB
Markdown

# FemLite V5 Remaining TODO
Status date: 2026-06-30
This file is the executable TODO list for the current 0.5.x development cycle.
The phase status table is maintained only in `docs/v5_development_plan.md`.
## 2026-06-30 V5 Release Scope Decision
The V5 release boundary is narrowed to a shippable V5.11 capability checkpoint.
V5.11 must prove that the finite-membrane-strain S4 production route is present,
tested, documented, and capable of a non-monotone spherical-cap postbuckling
signature under the agreed Shell-4 translational-rim benchmark口径. It is not
required to fully reproduce the entire `uz=-6.2 m` PDF curve or close all
Shell-4/8/16 reference-shape differences before the V5 release.
The following items are explicitly deferred to the next version line and must
not block V5 release:
- Full spherical-cap PDF curve reproduction through `uz=-6.2 m`.
- Strict Shell-4/8/16 curve-family comparison to digitized PDF values.
- True pseudo-arclength / branch-following continuation beyond the default
direct-displacement path improvements.
- Late post-peak GMRES/performance optimization beyond documented diagnostics.
- The V5.12 backlog: `two_arches` single reference-point check, industrial mortar/AL close-out,
Mohr-Coulomb critical-instability cloud longrun, generic buckling/cyclic main-solver
integration, roll forming production workflow, cohesive contact main-flow
integration, high-fidelity example upgrades, and longrun/performance close-out.
V5.11 may be marked release-ready when its checker, documentation, and evidence
validate the bounded capability claim: finite-strain S4 production dispatch,
stable shell geometry inputs, continuation-path diagnostics, the fixed support
口径, and a reproducible post-peak load drop with documented remaining limits.
Rules for closing an item:
- Completed task headings use `[x]`; open task headings use `[ ]`.
- Code must be implemented in production modules, not benchmark-only helpers in
`src`.
- Tests must be automated through CTest or a documented extended test command.
- Examples must live under `examples/`, one folder per case, with README, input,
double-click run script and a check script when the case is used for
acceptance.
- Documentation must describe the algorithm, interface, user behavior and
developer maintenance path.
- Large generated result files must not be committed.
## V5.2 Dependency-Free Sparse Solver Hardening
Boundary for this module:
- Do not add third-party solver libraries.
- Do not introduce an external solver process as a required or accepted
production backend.
- Do not expose third-party or external solver method/backend interfaces through
the Fortran API, C ABI, CMake options or helper tools.
- Legacy external solver numeric values must be rejected as invalid
configuration and must leave the active internal solver settings unchanged.
### [x] DONE V5.2-1: Harden internal sparse solver workflows
- Implementation:
- Keep CG, BiCGSTAB, GMRES and the internal dense pivoting fallback as the
production solver family.
- Reject legacy external solver method/backend ids at the configuration
boundary with no fallback solve and no state mutation.
- Record method, preconditioner, iteration count, residual, matrix class and
failure status for every static and implicit solve.
- Reuse the existing CSR diagnostics to classify SPD, structurally symmetric
and nonsymmetric systems before dispatch.
- Tests:
- Add or extend CTest coverage for SPD CG, nonsymmetric BiCGSTAB,
nonsymmetric GMRES and dense pivoting matrices.
- Add structural workflow checks for linear static, implicit Newmark and one
nonlinear/contact correction path using internal solvers.
- Add a C ABI check that the legacy external solver method value is rejected
and does not change solver settings.
- Acceptance:
- Default builds have no third-party solver dependency.
- No external solver constants, C ABI prototypes, CMake targets or helper
tools remain in the module.
- Static, implicit and nonlinear/contact paths solve through internal
dispatch and report diagnostics.
- Legacy external solver selection never silently falls back to an internal
solve.
- Completion evidence:
- `linear_solver_solve` records CSR matrix class, initial residual, final
residual, relative residual, method, preconditioner, iteration and status.
- `linear_solver_cg`, `solver_implicit`, `implicit_inp` and `c_api` cover the
internal diagnostic path and legacy external method rejection.
- `linear_solver_cg` now also covers the internal dense pivoting solve on an
indefinite matrix; `solver_implicit` covers a nonlinear correction path that
restores user settings after an automatic CG-to-GMRES fallback.
### [x] DONE V5.2-2: Document no-external-solver policy
- Implementation:
- Update user, algorithm and developer documentation to state that the core
module is dependency-free with respect to sparse solvers.
- Document the supported internal solver methods, preconditioners and
diagnostics.
- Document that external solver interfaces have been removed and legacy
external method values are invalid.
- Tests:
- Keep default CTest working without any optional solver package.
- Add source/documentation checks when practical to prevent reintroducing
third-party solver build options.
- Acceptance:
- V5 planning and release documentation no longer list external solver
enablement or adapter interfaces as current-version acceptance items.
- Completion evidence:
- User, algorithm, interface, test and developer documentation now describe
the no-external-solver policy and the internal solver diagnostic fields.
## V5.3 Mortar / Augmented Lagrangian Contact
### [x] DONE V5.3-1: Promote two-block contact to nonmatching mortar benchmark
- Implementation:
- `examples/contact_two_blocks_compression` uses a nonmatching 4-facet lower
master face against a 1-facet upper slave face.
- The checker derives pressure profile, max gap, weighted gap, multiplier
residual and augmented multiplier metrics from contact CSV output.
- `pressure_profile.csv` is written in the result directory for benchmark
inspection.
- Tests:
- `check_contact_results.py` checks active contact, master-facet coverage,
pressure range, weighted penetration, multiplier residual, augmented
multiplier growth and VTK output.
- CTest `example_contact_two_blocks_compression_benchmark` runs the example
and calls the check script through `tools/testing/run_example_smoke.py`.
- Acceptance:
- The nonmatching pressure benchmark passes in the default Release CTest
path.
- Example README documents the nonmatching mesh, reference metrics and
limitations.
- Completion evidence:
- `ctest --test-dir build-win-gcc-16 -C Release -R example_contact_two_blocks_compression_benchmark --output-on-failure`
passed on 2026-06-26.
### [x] DONE V5.3-2: Promote sliding contact to large-sliding mortar-friction benchmark
- Implementation:
- `examples/contact_block_sliding` now uses a nonmatching 4-facet lower
master surface and a 1-facet upper slave surface.
- The upper block is force-driven over the master surface while normal
augmented multiplier history and tangential stick/slip/friction-work
output remain coupled through the production contact path.
- Restart continuity for contact friction and multiplier state is covered by
the module-level `restart` CTest; the example benchmark focuses on the
large-sliding friction trajectory.
- Tests:
- `check_contact_results.py` writes `sliding_profile.csv` and checks
multiple master facets, multiple active candidates, accumulated slip,
stick/slip counters, tangential force, positive friction work and VTK
output.
- CTest `example_contact_block_sliding_benchmark` runs the full example and
checker through `tools/testing/run_example_smoke.py`.
- Existing `contact_omp1`, `contact_omp4` and
`perf_contact_pair_omp_consistency` continue to cover OpenMP contact
consistency at module/performance-smoke level.
- Acceptance:
- Stick/slip transitions are reported.
- Frictional work is nonnegative within tolerance.
- Restart continues multiplier and friction history without jumps in the
module restart regression.
- Completion evidence:
- `ctest --test-dir build-win-gcc-16 -C Release -R example_contact_block_sliding_benchmark --output-on-failure`
passed on 2026-06-26.
### [x] DONE V5.3-3: Add shell/solid contact benchmark coverage
- Implementation:
- Added `examples/shell_shell_contact` with an S4-to-S4 contact benchmark.
- Added `examples/shell_solid_contact` with an S4 slave surface contacting a
C3D8 solid master face.
- Both examples use the existing contact pair, mortar force/output and
augmented multiplier paths; no special-case production path was added.
- Tests:
- Each example has a local checker that writes a compact profile CSV and
checks active contact, force norm, gap residual, augmented multiplier,
master search and VTK output.
- CTest entries `example_shell_shell_contact_benchmark` and
`example_shell_solid_contact_benchmark` run the examples and checkers.
- Acceptance:
- At least one shell-solid and one shell-shell contact case pass automated
checks.
- Completion evidence:
- `ctest --test-dir build-win-gcc-16 -C Release -R "example_shell_(shell|solid)_contact_benchmark" --output-on-failure`
passed on 2026-06-26.
## V5.4 Mohr-Coulomb And Strength Reduction
### [x] DONE V5.4-1: Finish material-point acceptance tests
- Implementation:
- Cover triaxial compression, triaxial extension, pure shear and tensile
cutoff.
- Record active branch or return-map mode for each test.
- Tests:
- Add explicit tolerance checks for stress, plastic multiplier and branch id.
- Acceptance:
- All material-point tests pass without substep explosion.
- Tangent or algorithmic stiffness limitations are documented.
- Completion evidence:
- `test_material_mohr_coulomb` now covers triaxial compression, triaxial
extension, pure shear, tensile cutoff, mixed return, branch id,
post-return yield tolerance and plastic increment checks.
- `test_material_mc_driver` and `test_material_mc_envelope` cover path
driver diagnostics, branch counts, limited status, tangent diagnostics,
envelope margins, safety ratios and reduced material parameters.
- Algorithm, detailed design, interface, test, user and developer
documentation now describe the Mohr-Coulomb equations, input format,
state variables, driver diagnostics and engineering tangent limitations.
- Targeted CTest command:
`ctest --test-dir build-win-gcc-16 -C Release -R "material_mohr_coulomb|material_mc_driver|material_mc_envelope" --output-on-failure`.
### [x] DONE V5.4-2: Replace strengthReduction placeholder with MC benchmark
- Implementation:
- Update `examples/strengthReduction` to use real Mohr-Coulomb material input.
- Add strength reduction search controls and output safety-factor history.
- Output plastic zone and displacement control probe.
- Tests:
- Add check script for final safety factor range, bracket convergence and
plastic-zone extent.
- Acceptance:
- Safety factor lies inside documented tolerance.
- Check script fails if the solve only runs linear elastic behavior.
- Completion evidence:
- `strength_reduction_slope_static.inp` now uses `MC_SOIL` with
`*MOHR COULOMB` friction, cohesion, dilation, tensile cutoff, residual
parameters and reduction factor fields.
- `tools/benchmarks/strength_reduction_benchmark.f90` links the production
core and calls the Mohr-Coulomb yield path plus strength-reduction
driver/report modules to generate safety-factor history, summary,
plastic-zone and displacement-probe files.
- `check_strength_reduction_results.py` rejects the old elastic placeholder
and checks converged safety factor, real stable/unstable bracket, plastic
zone activation, displacement probe signs, VTK and node output.
- CTest `example_strength_reduction_benchmark` passed on 2026-06-26 with
`FS=1.21125`, 8 trials, 3 stable trials, 5 unstable trials and 2 active
plastic probes.
## V5.5 Linear Buckling And Cyclic Symmetry
### [x] DONE V5.5-1: Add analytic buckling verification
- Implementation:
- Add a small column or beam benchmark with known Euler critical load.
- Assemble prestress/geometric stiffness through the production path.
- Tests:
- Check first eigenvalue relative error.
- Acceptance:
- Relative error and mode normalization are documented.
- Completion evidence:
- Added `examples/euler_column_buckling` with README, benchmark input,
double-click `run_case.bat` and `check_buckling_results.py`.
- Added production `src/solver/fem_buckling_column.f90`; it reads the Euler
`*BUCKLING COLUMN` block, assembles beam bending stiffness through
`geom_beam2d_column_stiffness`, calls `buckling_dense_solve`, compares
against `buckling_euler_column_load` and writes summary/mode CSV files.
- Kept `tools/benchmarks/euler_column_buckling_benchmark.f90` outside `src`
as a thin wrapper around `buckling_column_run_file`; the old duplicate
helper-side assembly/solve algorithm has been deleted.
- Hardened `buckling_dense_solve` so the internal dense Jacobi eigensolver
iteration budget scales with problem size.
- `examples/euler_column_buckling/run_case.bat` and
`check_buckling_results.py` now run the main `femlite --buckling`
workflow.
- CTest `example_euler_column_buckling_benchmark` passed on 2026-06-27 with
relative error `3.27661E-05`, residual `1.50015E-11` and
`phi^T K_G phi = 1.0`.
### [x] DONE V5.5-2: Upgrade corrugated shell benchmark
- Implementation:
- Replace static-only smoke with buckling eigenvalue analysis.
- Add cyclic sector model and full-model comparison when feasible.
- Write mode shape output through modal output path.
- Tests:
- Check critical load range and cyclic/full consistency.
- Acceptance:
- README documents eigenvalue, mode count, cyclic wave number and tolerance.
- Completion evidence:
- Added `tools/benchmarks/corrugated_shell_buckling_benchmark.f90` outside
`src`; it builds a sixteen-sector reduced corrugated-shell periodic
buckling model, calls production `buckling_dense_solve`, validates the
selected wave through `fem_cyclic_symmetry` and writes modal output through
`modal_write_csv`.
- Added `corrugated_shell_buckling_benchmark.inp`,
`check_buckling_results.py`, updated `run_case.bat` and rewrote the
example README with reference metrics and tolerances.
- Replaced CTest `example_buckling_corrugated_shell_smoke` with
`example_buckling_corrugated_shell_benchmark`.
- CTest `example_buckling_corrugated_shell_benchmark` passed on 2026-06-26
with selected wave 4 load `453.031407`, critical wave 8 load
`393.369677`, cyclic/full relative error `1.25473461E-16` and modal
residual `1.31855319E-12`.
- `source_size` CTest passed after the new benchmark tool was added.
## V5.6 High-Fidelity Roll Forming
### [x] DONE V5.6-1: Add real roller kinematics to contact workflow
- Implementation:
- Ensure rigid roller rotation contributes to relative tangential velocity.
- Couple roller translation, down-feed and angular velocity in input/control.
- Output roller force, torque, contact pressure band and friction work.
- Tests:
- Unit test roller kinematics and relative velocity sign.
- Example check for nonzero rolling traction and stable energy trend.
- Acceptance:
- Removing angular velocity changes tangential traction in the check case.
- Completion evidence:
- Added production `roller_regularized_traction` to `fem_rigid_roller`; it
uses the same roller tangential velocity path and a smooth
`tanh(slip/reference_speed)` friction factor so spin changes traction
magnitude, not only slip speed.
- Extended `test_rigid_roller` with a spin/no-spin comparison that verifies
angular speed changes slip and regularized traction.
- Added `tools/benchmarks/roll_forming_kinematics_benchmark.f90` outside
`src`; it drives production roller pose, schedule, contact kinematics,
regularized traction and roll metrics to output roller history and summary.
- Added `examples/roll_forming/roll_forming_kinematics_benchmark.inp`,
`check_roller_kinematics.py` and updated `run_case.bat`/README.
- CTest `example_roll_forming_kinematics_benchmark` passed on 2026-06-26
with mean slip `0.4` with spin vs `0.2` without spin, mean traction
`289.208274` with spin vs `228.478247` without spin, max traction delta
`60.7300272` and positive roller torque `14.4604137`.
### [x] DONE V5.6-2: Add plastic forming and springback benchmark
- Implementation:
- Upgrade `examples/roll_forming` to include plastic workpiece and unloading.
- Output residual displacement, plastic zone and springback metric.
- Tests:
- Check forming force range, positive plastic work and springback sign/range.
- Acceptance:
- Example is no longer only an elastic contact surrogate.
- Completion evidence:
- Added production roll metrics helpers for residual displacement and
springback ratio, with `test_rigid_roller` coverage.
- Added `tools/benchmarks/roll_forming_springback_benchmark.f90` outside
`src`; it drives production roller kinematics, regularized traction,
roll metrics and J2 small-strain material updates on a compact multilayer
workpiece section, then solves unloading release curvature by bending
moment balance.
- Added `examples/roll_forming/roll_forming_springback_benchmark.inp` and
`check_springback_results.py`; `run_case.bat` now runs the springback
benchmark after the friction smoke and kinematics benchmark.
- Targeted checker passed on 2026-06-26 with springback ratio `0.473154068`,
residual displacement `0.0316107559`, plastic work `1.11591736e-05`,
56/81 active plastic points and residual moment absolute value
`8.92778801e-11`.
## V5.7 Cohesive Adhesion / Decohesion
### [x] DONE V5.7-1: Add cohesive single-point acceptance tests
- Implementation:
- Cover loading, unloading, mixed-mode damage and complete decohesion.
- Track damage, maximum opening, energy and status.
- Tests:
- Verify nonnegative dissipated energy and irreversible damage.
- Acceptance:
- Restarted state preserves damage and maximum opening.
- Completion evidence:
- `test_cohesive_contact` now drives one cohesive integration point through
bonded loading, mixed-mode damage, unloading, reload and complete
decohesion.
- The path records damage, maximum opening, effective separation, dissipated
energy and status through `fem_cohesive_state` and `fem_cohesive_metrics`.
- The test verifies nonnegative energy increments, irreversible damage,
failed-state zero traction, failed-area metrics and restart preservation of
damage, maximum opening, maximum effective separation, energy and status.
- Targeted command passed on 2026-06-26:
`ctest --test-dir build-win-gcc-16 -C Release -R "cohesive_contact|source_size" --output-on-failure`.
### [x] DONE V5.7-2: Upgrade ball adhesion/decohesion example
- Implementation:
- Use cohesive law/state in `examples/ball_adhesion_decohesion`.
- Output force-displacement curve, cohesive energy and decohesion area.
- Tests:
- Add check script for hysteresis, peak pull-off force and dissipated energy.
- Acceptance:
- Example fails if cohesive damage is not activated.
- Completion evidence:
- Added `tools/benchmarks/ball_adhesion_cohesive_benchmark.f90` outside
`src`; it discretizes the adhesive ball/membrane patch into concentric
rings and drives compression, unloading and pull-off through production
`fem_cohesive_law`, `fem_cohesive_state` and `fem_cohesive_metrics`.
- Added `ball_adhesion_cohesive_benchmark.inp` and
`check_cohesive_results.py` under `examples/ball_adhesion_decohesion`;
`run_case.bat` now runs the original contact smoke and the cohesive
benchmark.
- The benchmark writes `adhesion_history.csv`, `adhesion_summary.csv` and
`decohesion_profile.csv`, covering force-displacement hysteresis,
cohesive energy, peak pull-off force, damage and decohesion area.
- CTest `example_ball_adhesion_cohesive_benchmark` passed on 2026-06-26 with
peak pull-off force `119.246023`, min compression force `-362.051100`,
dissipated energy `0.451891017`, hysteresis area `0.898170568` and final
failed fraction `1.0`.
## V5.8 Particle / Flexible Container Coupling
### [x] DONE V5.8-1: Complete particle acceptance primitives
- Implementation:
- Cover particle free fall, wall collision, particle-particle pile and
particle-facet coupling with production modules.
- Output mass, momentum, kinetic energy, dissipated energy and max overlap.
- Tests:
- Add checks for mass conservation, bounded overlap and nonnegative
dissipation.
- Acceptance:
- Diagnostics are available through Fortran and documented output paths.
- Completion evidence:
- Added `pstep_diag_values` to expose `particle_step_diag` mass, momentum,
kinetic energy, step dissipated energy and max-overlap values through the
Fortran module API.
- Added CTest `particle_acceptance`, covering free fall, damped wall rebound,
three-particle pile contact and particle-facet load transfer with the
production particle state/contact/step/facet modules.
- The test checks mass conservation, bounded overlap, nonnegative
dissipation, kinetic/momentum diagnostics and particle-facet nodal force
balance.
- `ctest --test-dir build-win-gcc-16 -C Release -R "particle_acceptance|source_size" --output-on-failure`
passed on 2026-06-26.
### [x] DONE V5.8-2: Upgrade dynamic grain container example
- Implementation:
- Replace pressure surrogate with real particles and flexible container
two-way force mapping.
- Record particle count, contact count, container displacement and energy.
- Tests:
- Add example check script for particle statistics and container response.
- Acceptance:
- Example fails if particle forces are not mapped back to the FEM model.
- Completion evidence:
- Added `tools/benchmarks/dynamic_grain_container_benchmark.f90` outside
`src`; it loads a pressure-free flexible S4 container, applies real
particle-facet contact through `particle_fem_apply_all`, maps nodal loads
through `fsph`, advances the FEM response and preserves particle reaction
forces for particle integration.
- Added `dynamic_grain_container_benchmark.inp`,
`flexible_container_grain_benchmark.inp` and
`check_grain_container_results.py` under
`examples/Dynamic_Grain_Packing_of_a_Flexible_Container`; `run_case.bat`
now runs the old smoke and the V5.8 benchmark checker.
- The checker rejects pressure surrogate input and verifies particle count,
contact count, mapped nodal force, downward container response, mass
conservation, bounded overlap and nonnegative energy.
- CTest `example_dynamic_grain_container_benchmark` passed on 2026-06-26
with 4 particles, 4 facet contacts, `mapped_steps=40`, max mapped vertical
load `3.68E+01`, minimum center displacement `-7.29240354E-06` and max
overlap `3.00615680E-03`.
- Full Release command
`ctest --test-dir build-win-gcc-16 -C Release --output-on-failure`
passed on 2026-06-26 with 106/106 tests.
## V5.9 Benchmark, Performance And Documentation
### [x] DONE V5.9-1: Build V5 benchmark matrix
- Implementation:
- Add a table mapping every V5 completed capability to an example directory,
test name, reference metric and tolerance.
- Keep smoke and benchmark tiers separate.
- Tests:
- Default CTest remains fast.
- Extended benchmark command runs the V5 acceptance checks.
- Acceptance:
- No completed V5 capability relies only on a smoke test.
- Completion evidence:
- Added `docs/v5_benchmark_matrix.md` with V5.2-V5.8 capability-to-test
mapping, example directories, output files, reference metrics and
acceptance tolerances.
- Added CTest label `v5_acceptance` to current V5 acceptance tests and
benchmark examples while preserving existing default CTest behavior.
- Documented the extended command
`ctest --test-dir build-win-gcc-16 -C Release -L v5_acceptance --output-on-failure`.
- The extended command passed on 2026-06-26 with 35/35 tests.
### [x] DONE V5.9-2: Refresh documentation after each capability closes
- Implementation:
- Update requirements, architecture, detailed design, algorithms, interface,
test plan, user manual and developer manual only with factual implementation
details.
- Keep status markers only in `docs/v5_development_plan.md`.
- Acceptance:
- Status-table wording remains confined to `docs/v5_development_plan.md`.
- Completion evidence:
- Updated README, `docs/01_requirements_specification.md`,
`docs/02_software_architecture_design.md`, `docs/03_detailed_design.md`,
`docs/04_mathematical_model_and_algorithms.md`,
`docs/05_interface_control_document.md`,
`docs/06_test_and_verification_plan.md`, `docs/07_user_manual.md`,
`docs/08_developer_manual.md`, `docs/v5_development_plan.md` and
`docs/v5_benchmark_matrix.md` for V5.8 and V5.9-1 changes.
- Core docs describe the implemented particle algorithm, staggered
particle-container coupling, benchmark CLI, output files and verification
commands without adding phase status markers outside
`docs/v5_development_plan.md`.
- Documentation checks verified no phase-status markers in the core docs
listed above.
## V5.10 Acceptance And Release
### [x] DONE V5.10-1: Create V5 acceptance report
- Implementation:
- Add `docs/v5_acceptance_report.md`.
- Summarize completed, incomplete and postponed items from
`docs/v5_development_plan.md`.
- Include exact test commands, environment and results.
- Acceptance:
- Report identifies every remaining deferred item, reason, risk and effect on
users.
- Completion evidence:
- Added `docs/v5_acceptance_report.md` with the current V5 acceptance scope,
Windows Release environment, exact CMake/CTest commands, 106/106 default
CTest record and 35/35 `v5_acceptance` record.
- The report summarizes completed V5.2-V5.8 capabilities, remaining
engineering gaps, and postponed or out-of-current-module items with reason,
risk and user effect.
- README now links to the V5 acceptance report.
### [x] DONE V5.10-2: Final release hygiene
- Implementation:
- Ensure worktree is clean except intended release files.
- Clean generated root result files.
- Run default CTest and any enabled extended benchmark tests.
- Acceptance:
- Release branch or main merge has reproducible test output.
- README and version metadata match the release state.
- Completion evidence:
- Root-level generated result files (`history.csv`, `nodes_*.csv`,
`elements_*.csv`, `reactions_*.csv`, `contact_*.csv`, `result_*.vtk`,
`femlite.log`) were removed after the final full CTest run.
- `ctest --test-dir build-win-gcc-16 -C Release -L v5_acceptance --output-on-failure`
passed on 2026-06-26 with 35/35 tests.
- `ctest --test-dir build-win-gcc-16 -C Release --output-on-failure`
passed on 2026-06-26 with 106/106 tests.
- README and version metadata are aligned on `0.5.0`.
- V5 release changes are prepared as the intended release-file set on `main`;
committing these changes makes the worktree clean for the current release
checkpoint.
## V5.11 Finite-Membrane-Strain Shell And Spherical Cap Postbuckling
Boundary for this module:
- V5.11 release acceptance is a finite-strain S4 production capability
checkpoint, not a full PDF curve-reproduction claim. The spherical-cap case
must demonstrate the agreed postbuckling口径 with a reproducible post-peak load
drop, but full `uz=-6.2 m` curve matching to
`models.sme.spherical_cap_with_central_point_load.pdf` is deferred.
- The production route is Abaqus-style finite-membrane-strain S4: finite
rotation director update, finite membrane strain, Koiter-Sanders type
curvature increment, MITC/ANS assumed strain, and analytic consistent tangent.
- The old S4 diagnostic route was temporary technical debt. It explained earlier
failures, but it is not a stable interface and not acceptance evidence.
V5.11-1 removes its production dispatch, input aliases, tests, documentation
entry points, and unused helper code. Do not leave fallback interfaces.
- Reusable foundations are limited to `fem_shell_section`, shell layer/resultant
state, shell pressure follower tangent, internal nonsymmetric solver support,
and continuation path output.
- The V5 checker must reject regressions to linear, small-strain, monotone-only
or old corotational behavior. It does not need to pass strict Shell-4/8/16 PDF
digitized-curve matching in this release.
Current facts:
- `examples/spherical_cap_with_central_point_load` checks reference geometry,
material scale and Shell-16 small-strain samples; the Shell-4 deck now uses
the all-S4 finite-strain target route with translational rim support as the
V5.11 postbuckling capability proof.
- `continuation_path.csv` records accepted displacement/reaction rows, linear
solver diagnostics, path-control diagnostics and trust diagnostics, so the
checker can validate the bounded V5 post-peak signature and preserve evidence
for the deferred full-curve work.
- `SHELL_GEOMETRIC=YES`, `GEOMETRIC_SHELL`, `COROTATIONAL_SHELL`, and
`S4_COROTATIONAL` are removed migration-era interfaces; the target user-facing
syntax is `SHELL_GEOMETRIC=FINITE_STRAIN` or `SHELL_GEOMETRIC=LINEAR`.
### [x] TODO V5.11-0: Redesign docs around finite-strain S4
- Implementation:
- [x] Update `docs/03_detailed_design.md` with finite-strain S4 modules:
rotation state, finite kinematics, assumed strain, nonlinear driver,
tangent, drilling, and section coupling.
- [x] Update `docs/04_mathematical_model_and_algorithms.md` with finite
membrane strain, finite rotation director update, Koiter-Sanders type
curvature, MITC/ANS, and analytic tangent requirements.
- [x] Update `docs/05_interface_control_document.md`,
`docs/06_test_and_verification_plan.md`, `docs/07_user_manual.md`,
`docs/08_developer_manual.md`, and `docs/v5_development_plan.md` so they no
longer describe corotational S4 as the V5.11 completion route.
- [x] Record the deletion rule: old exploratory code and aliases must be
removed after the complete implementation replaces them.
- Acceptance:
- [x] Documentation separates reusable foundations from temporary
corotational code and explicitly forbids keeping unused interfaces.
### [x] DONE V5.11-1: Delete obsolete corotational interfaces and dead code
- Implementation:
- [x] Inventory every production call site, input alias, public API, test, and
documentation reference containing `corotational`, `COROTATIONAL_SHELL`,
`S4_COROTATIONAL`, or old boolean-only `SHELL_GEOMETRIC` semantics.
- [x] Keep only foundations required by finite-strain S4:
`fem_shell_section`, shell layer state, `shell_resultant`, pressure follower
tangent, nonsymmetric solver support, and continuation path output.
- [x] Delete production dispatch paths, parser aliases, helper routines, and
tests that only serve the old corotational route and are not needed by the
finite-strain implementation.
- [x] Add input validation so `SHELL_GEOMETRIC=FINITE_STRAIN` and
`SHELL_GEOMETRIC=LINEAR` are the stable forms; old aliases fail with a clear
input error after migration.
- Tests:
- [x] Add parser tests proving old aliases are rejected after migration.
- [x] Run shell, implicit input, spherical cap, and source-size regressions.
- Acceptance:
- [x] No stable user-visible or internal production interface remains for the
old corotational route.
- [x] `rg -n "COROTATIONAL_SHELL|S4_COROTATIONAL|corotational"` returns only
historical notes or intentionally named removed-history comments, not live
production dispatch.
- Completion evidence:
- Removed `src/element/fem_shell_corotational.f90` from source and CMake.
- Removed old shell corotational production dispatch and FD tangent/resultant
helper routines from `fem_element_internal` and `fem_element_stiffness`.
- Replaced `shell_geometric_nonlinear` with
`shell_finite_strain_requested`; `SHELL_GEOMETRIC` now accepts only
`FINITE_STRAIN` or `LINEAR`.
- Old `GEOMETRIC_SHELL`/`COROTATIONAL_SHELL`/`S4_COROTATIONAL` aliases are
rejected with a clear input error and parser rollback is tested.
- Removed the old curved S4 nonlinear patch input and tests that only served
the deleted route.
- Updated spherical-cap Shell-16 deck to `SHELL_GEOMETRIC=LINEAR` and Shell-4
target deck to `SHELL_GEOMETRIC=FINITE_STRAIN`; finite-strain shell requests
now fail clearly until V5.11-4 through V5.11-6 are implemented.
- Build passed:
`cmake --build build-win-gcc-16 --target femlite test_implicit_inp test_shell_internal -j 4`.
- Targeted CTest passed:
`ctest --test-dir build-win-gcc-16 -R "^(implicit_inp|shell_internal|source_size|example_spherical_cap_point_load_benchmark)$" --output-on-failure`.
### [x] DONE V5.11-2: Implement finite rotation shell state
- Implementation:
- [x] Add node or shell-node rotation trial/committed state using exponential
map or quaternion update.
- [x] Integrate rotation state with reset, backup, rollback, commit, cutback,
line-search rejection, Newmark rollback, and restart.
- [x] Keep global node DOF 4:6 as Newton increments while committed director
state carries finite rotation history.
- Tests:
- [x] Rigid-body large rotation produces zero or near-zero finite-kinematics
strain once the rotation state is consumed by the V5.11-3 S4 kinematics
module.
- [x] Failed increments restore rotation state exactly.
- Acceptance:
- [x] Rotation state is not recomputed from small-angle accumulated DOF in a
way that loses finite rotation objectivity.
- Completion evidence:
- Added production `fem_shell_rotation` with per-node quaternion committed,
trial and backup state; nodal DOF 4:6 remain incremental Newton rotation
corrections and update trial orientation by exponential-map quaternion
increments.
- `element_state_ensure/reset/begin/restore/commit/abort` now includes shell
rotation lifecycle so cutback, rejected trial steps and Newmark rollback
restore the finite-rotation state together with integration-point state.
- Restart format version 3 writes and reads the committed shell rotation
block; version 1/2 restarts initialize the rotation state to identity.
- `test_element_state` verifies identity initialization, 90-degree
finite-rotation vector behavior, commit, restore and abort rollback.
`test_restart` verifies shell rotation restart persistence and vector
rotation after reload.
- Build passed:
`cmake --build build-win-gcc-16 --target test_element_state test_restart femlite -j 4`.
- Targeted CTest passed:
`ctest --test-dir build-win-gcc-16 -R "^(element_state|restart|implicit_inp|shell_internal|source_size|example_spherical_cap_point_load_benchmark)$" --output-on-failure`.
### [x] DONE V5.11-3: Implement S4 finite kinematics and MITC/ANS
- Implementation:
- [x] Compute reference/current covariant bases, director fields, surface
metrics, finite membrane strain, Koiter-Sanders type curvature increment,
and transverse shear strain for S4.
- [x] Handle initially curved or warped shell reference geometry without
flattening the element into a small-strain local plane.
- [x] Implement S4 MITC/ANS transverse shear sampling and keep extension points
for membrane/bending assumed strain if needed.
- Tests:
- [x] Small-rotation finite-strain S4 matches existing linear S4 patch
kinematics.
- [x] MITC transverse shear patch is verified at the kinematic projection
level.
- [x] Warped/curved reference patch has no spurious initial strain.
- Acceptance:
- [x] The formulation works for the spherical-cap mesh geometry, not only flat
plates.
- Completion evidence:
- Added production `fem_shell_finite_kinematics` for S4 reference/current
covariant bases, director interpolation, surface metrics, finite membrane
strain, Koiter-Sanders-style curvature increment, local engineering strain
projection and direct/MITC transverse shear sampling.
- The module consumes arbitrary reference/current director fields, including
directors generated from `fem_shell_rotation`; it does not accumulate small
angle DOF values as finite orientation history.
- Added CTest `shell_finite_kinematics`, covering small membrane strain,
small bending curvature, MITC constant shear projection, warped reference
zero strain, quaternion-state rigid rotation zero strain and spherical-cap
patch geometry.
- Build passed:
`cmake --build build-win-gcc-16 --target test_shell_finite_kinematics test_shell_internal test_element_state test_restart test_implicit_inp femlite -j 4`.
- Targeted CTest passed:
`ctest --test-dir build-win-gcc-16 -R "^(shell_finite_kinematics|shell_internal|element_state|restart|implicit_inp|source_size|example_spherical_cap_point_load_benchmark)$" --output-on-failure`.
### [x] DONE V5.11-4: Implement analytic consistent shell tangent
- Implementation:
- [x] Assemble material tangent `B^T D_alg B`.
- [x] Assemble geometric tangent from membrane force, bending moment, and
transverse shear resultants.
- [x] Assemble rotation tangent from director and curvature linearization.
- [x] Assemble drilling stabilization tangent as a separate energy-consistent
contribution.
- [x] Keep finite-difference tangent only as an audit tool, never as the final
production Newton tangent.
- [x] Connect the GP tangent to the S4 element/global assembly only after the
solver synchronizes nodal rotation DOF increments into `shell_rotation_trial`
at every residual/tangent evaluation.
- Tests:
- [x] Finite-difference column audit passes for elastic S4.
- [x] J2/J2_JC/composite section tangent audit passes under large rotation.
- [x] S4 element-level tangent audit passes; spherical-cap Newton/cutback
evidence is tracked in V5.11-6 because the current target deck contains
unsupported S3 crown fan elements.
- Acceptance:
- [x] The tangent is analytic, documented, and does not rely on hidden
benchmark-only code.
- Progress evidence:
- Added `shell4_finite_bmat_at` and `shell4_finite_derivs_at` in
`fem_shell_finite_kinematics`; the derivative tensor is the algorithmic
finite-rotation linearization used by the residual, not a symmetric
coordinate Hessian.
- Added production `fem_shell_finite_tangent`, assembling
`K = B^T D_alg B + sum_i R_i dB_i/du` at S4 integration-point level for
membrane, bending and MITC transverse shear resultants.
- Added CTest `shell_finite_tangent`; `shell_finite_kinematics` now audits
finite-strain values, B columns and `dB/du`, while `shell_finite_tangent`
audits the 24-by-24 elastic GP tangent against internal-force column finite
differences.
- Build passed:
`cmake --build build-win-gcc-16 --target test_shell_finite_tangent test_shell_finite_kinematics -j 4`.
- Targeted CTest passed:
`ctest --test-dir build-win-gcc-16 -R "^(shell_finite_kinematics|shell_finite_tangent)$" --output-on-failure`.
- Added production `fem_shell_finite_element`: `SHELL_GEOMETRIC=FINITE_STRAIN`
now dispatches S4 elements through finite-strain internal force and tangent
assembly, including averaged nodal reference directors, current director
trial rotations, section resultants, geometric/rotation tangent and
energy-consistent drilling energy.
- `fem_shell_rotation` now synchronizes rotation DOFs incrementally inside an
active trial: each residual/tangent evaluation multiplies only the newest
Newton rotation correction into `shell_rotation_trial`, matching the
`dB/du` linearization.
- Added CTest `shell_finite_element`, covering S4 production dispatch, element
24-column finite-difference tangent, drilling residual/stiffness energy
consistency, unsymmetric composite membrane-bending coupling, J2/J2_JC large
rotation section tangents, composite plastic layer state, and plastic
rollback.
- Targeted CTest passed:
`ctest --test-dir build-win-gcc-16 -R "^(shell_finite_element|shell_finite_kinematics|shell_finite_tangent|shell_internal|element_state|implicit_inp|restart|source_size|example_spherical_cap_point_load_benchmark)$" --output-on-failure`.
- Remaining spherical-cap postbuckling evidence is deferred to V5.11-6:
the Shell-4 target deck has since been converted to all-S4 topology, but
the nonlinear curve still fails the PDF load drop/restiffening signature.
### [x] DONE V5.11-5: Couple finite-strain S4 to shell sections and materials
- Implementation:
- [x] Route ELASTIC, J2, J2_JC, and composite shell sections through
`fem_shell_section` with trial/rollback semantics.
- [x] Complete multilayer plastic through-thickness state where composite
layers use plastic materials.
- [x] Preserve membrane-bending coupling for unsymmetric laminates.
- Tests:
- [x] ELASTIC/J2/J2_JC/composite large-rotation shell material tests pass.
- [x] Plastic layer state rolls back after failed increment.
- Acceptance:
- [x] Unsupported shell material models fail explicitly, with no fallback to an
elastic engineering tangent.
- Completion evidence:
- `fem_shell_section` now owns the public production APIs
`shell_section_update_resultants` and `shell_section_tangent_at`; finite and
small-strain shell paths share the same state update, through-thickness
layer integration, algorithmic tangent and rollback storage.
- `shell_finite_element` covers ELASTIC, unsymmetric elastic composite, J2,
J2_JC and composite-with-J2-layer finite-strain S4 paths under large
rotations, plus rollback after a rejected plastic trial.
- Targeted CTest passed as listed in V5.11-4.
### [x] DONE V5.11-6: Close bounded spherical-cap postbuckling capability benchmark
- Implementation:
- [x] Generate Shell-16 baseline and Shell-4 finite-strain target decks with
`a = 10 m`, `th = 0.203840 m`, `E = 210 GPa`, `nu = 0.3`, cap angle
`pi/4`, and documented support口径.
- [x] Output nondimensional curve data:
`wn = -w/th`, `Fn = -F*a/(E*th^3*2*pi)`.
- [x] Use production continuation and finite-strain shell code only.
- [x] Resolve the Shell-4 target mesh blocker: the deck no longer contains S3
crown fan elements and now uses an all-S4 square-disk crown remesh matching
the PDF radius, thickness, cap angle and fixed rim geometry.
- [x] Resolve the release-blocking physics question at bounded scope: the
all-S4 Shell-4 translational-rim target shows a clear post-peak load drop,
so V5.11 has evidence of nonlinear postbuckling behavior and is no longer
blocked by the absence of a limit-point signature.
- [x] Diagnose the main reference口径 differences before changing more
production physics: top-surface offset is not the dominant scale error,
while rim rotational clamping strongly changes the curve and masks the
post-peak branch. The V5.11 Shell-4 target therefore uses translational rim
support; full PDF curve matching remains deferred.
- [x] Update the spherical-cap checker so the V5 bounded nonlinear route
validates the bounded capability claim: Shell-16 baseline geometry, Shell-4
finite-strain all-S4 target, translational rim support, continuation-path
columns, non-monotone post-peak load drop and documented known limits.
- [x] Move strict Shell-4/8/16 PDF curve matching, full `uz=-6.2 m`
reproduction and restiffening-window tuning into the next-version backlog.
- [x] Expose nonlinear stage tracing through `femlite --trace-nonlinear`
so long spherical-cap runs show whether time is spent in pattern assembly,
stiffness assembly or linear solves.
- [x] Avoid the expensive CG-first path for finite-strain shell nonlinear
corrections: small finite-shell systems now use the internal dense solver
directly, while larger finite-shell systems switch to GMRES before the
existing fallback chain.
- [x] Expose `femlite --linear-iter` so benchmark and diagnostic runs can
bound nonlinear correction solve attempts before fallback handling.
- [x] Expose `femlite --shell-drilling` and wire the spherical-cap nonlinear
checker to `FEMLITE_SPHERICAL_CAP_SHELL_DRILLING` so drilling-stabilization
sensitivity can be audited without changing defaults.
- [x] Confirm on the rebuilt E-drive GCC build that Shell-4 direct
displacement reaches `uz=-2 m` with a post-peak load drop and restiffening
in the accepted continuation path; the remaining blocker is the larger
displacement/path and reference-shape mismatch, not the absence of any
limit-point signature.
- [x] Run a short drilling-stabilization sensitivity probe: the default
Shell-4 `uz=-2 m` path completed with the same load-drop/restiffening
signature, while zero drilling stabilization did not complete within the
diagnostic timeout, so the next fix should not be a simple stabilization
reduction.
- [x] Add shell strain diagnostic columns to `elements_*.csv`
(`shell_mem_max`, `shell_ben_max`, `shell_shr_max`, `shell_k11`,
`shell_k22`, `shell_k12`) so spherical-cap runs can separate membrane,
bending and shear contributions without adding benchmark-only code.
- [x] Confirm on the Shell-4 `uz=-2 m` diagnostic run that the final state is
not dominated by transverse shear locking: at `wn ~= 9.8`, the largest
element values are approximately membrane `4.1e-2`, bending curvature
`3.9e-1`, and shear `1.1e-2`; the next audit should focus on membrane/
bending coupling, curvature sign/scale, and mesh/path effects.
- [x] Add a finite-kinematics unit audit for spherical normal offset so the
membrane and bending curvature signs are checked on a curved reference
patch, not only on flat bending patches.
- [x] Tighten that finite-kinematics audit so the same spherical offset also
checks membrane and bending curvature scale (`offset/R` and `offset/R^2`).
This passed on the E-drive GCC build, so the current Shell-4 over-stiffening
is unlikely to be a basic curved-patch strain-scale amplification.
- [x] Test angle-weighted production reference-director averaging. It is a
generic robustness improvement and passes finite-shell regressions, but the
Shell-4 `uz=-2 m` curve is essentially unchanged (`Fn` at `wn ~= 9.8`
changes from about `0.83302` to `0.83289`), so the dominant V5.11-6 issue is
unlikely to be nodal director averaging alone.
- [x] Run a Shell-8 `uz=-2 m` square-disk finite-strain diagnostic. It
completed in about 130 s with 10 increments and 85 Newton iterations. The
early curve is much softer than Shell-4 and closer to the PDF/Shell-16
reference scale (`Fn ~= 0.591` at `wn ~= 6.28`, versus Shell-4 `Fn ~= 0.653`
at the same displacement), but it is still monotone through `wn ~= 9.8`.
This points to a combined mesh/path-resolution issue rather than a single
obvious curvature-sign bug.
- [x] Re-run the Shell-8 short diagnostic after the displacement-amplitude
correction. A temporary all-S4 Shell-8 deck to `uz=-1.984 m` completed in
about 113 s with 13 accepted increments and 91 Newton iterations; at
`wn ~= 9.73`, `Fn ~= 0.804`. This confirms that mesh refinement softens the
response but does not by itself close the PDF-scale mismatch.
- [x] Add an experimental finite S4 membrane-locking control path:
`--shell-membrane-locking center` uses an element-center assumed membrane
strain and matching membrane tangent rows, while the default remains off.
The switch is covered by `shell_finite_element`. A spherical-cap probe with
the switch enabled reached heavy iteration near `time ~= 0.78`, so it is
retained as a controlled diagnostic instead of becoming the production
default.
- [x] Improve continuation path sampling for long diagnostics:
displacement-control and arc-length accepted steps now refresh
`continuation_path.csv` immediately after `nlc_path_append`, so a long run
can be inspected before the full analysis finishes. Verified with
`c3d4_static_arc_disp_continuation.inp` and `--output-interval 0`, producing
a two-row path CSV.
- [x] Extend `continuation_path.csv` with per-row linear-solver diagnostics:
method, preconditioner, iteration count, status, initial residual, final
residual and relative residual. This makes long spherical-cap runs
inspectable without enabling a separate trace file. Verified with the
compact `c3d4_static_arc_disp_continuation.inp` path CSV.
- [x] Extend `continuation_path.csv` with accepted-step path-control
diagnostics: `target_dt`, `actual_dt`, `next_dt`, `adapt_adjust` and
cumulative `cutback_count`. Path rows are now recorded after the next-step
decision is available, so long post-peak runs show whether the driver is
growing, keeping, shrinking or recovering from cutback at each accepted
increment. Covered by `solver_implicit` and a compact path-column probe.
- [x] Re-run the Shell-4 `time=0.48` diagnostic with path-level linear-solver
stats. It completed in about 40 s with 24 increments and 168 Newton
iterations; the last accepted steps used GMRES/Jacobi with roughly 75-94
linear iterations and final relative residuals around `1e-7`. The runtime
issue is therefore repeated Newton corrections plus moderate GMRES work,
not a single runaway linear solve.
- [x] Add controlled direct-displacement path adaptivity instead of making it
unconditional: `DISP_ADAPTIVE=YES` in `*NONLINEAR CONTINUATION` or
`femlite --disp-path-adaptive on` routes direct displacement control through
the existing target-Newton adaptive loop. Later V5.11 post-peak diagnostics
showed that the fixed-grid default was still too brittle for the
translational-rim spherical-cap path, so this adaptive route was promoted to
the default direct-displacement continuation behavior; explicit
`DISP_ADAPTIVE=NO` / `--disp-path-adaptive off` remains available for old
fixed-grid sampling.
- [x] Add `femlite --linear-precond none|jacobi|gs|ilu0` for controlled
nonlinear diagnostics. On the Shell-4 `time=0.48` probe, ILU0 reduced the
last-step GMRES iteration count from roughly 90 to roughly 15, but total
runtime increased from about 40 s to about 62 s, so the default remains
Jacobi for this benchmark size.
- [x] Create CSV/VTK output directories automatically in
`output_csv_set_directory`, including nested paths. This covers immediate
path flushing when `--output-interval 0` is used and no output directory
exists yet; `vtk_output` now exercises a fresh nested output directory.
- [x] Fix direct displacement-control path sampling for prescribed
displacements that already carry an amplitude. `*BOUNDARY, AMPLITUDE=...`
under `METHOD=ARC_LENGTH_OR_DISPLACEMENT_CONTROL` now uses the amplitude as
the displacement path instead of multiplying it by the static `load_factor`
a second time. The Shell-4 spherical cap target is therefore linear in
`time` again (`uz = -6.2*time`, not `-6.2*time^2`), and
`solver_implicit` covers the amplitude regression.
- [x] Lower the finite-strain shell nonlinear auto-dense threshold from 1000
DOF to 250 DOF. The 486-DOF Shell-4 spherical-cap diagnostic now routes to
the nonsymmetric Krylov path instead of the slow dense pivot path; the
`time=0.48` diagnostic improved from roughly 94 s to roughly 63 s with the
same 24 accepted increments and 168 Newton iterations.
- [x] Add `femlite --static-predictor on|off` so nonlinear static predictor
behavior can be audited without changing defaults. On the Shell-4
`time=0.48` diagnostic it did not reduce Newton iterations, so it is not a
default spherical-cap fix.
- [x] Audit center crown rotation constraints by running a temporary input
without `CROWN, 4, 6, 0.0`; the curve matched the fixed-rotation run through
`time=0.48`, so crown rotations are not the observed stiffness source.
- [x] Capture a `time=0.48` element diagnostic after the amplitude fix:
at `wn ~= 14.6`, `Fn ~= 2.108`, max membrane strain is about `4.4e-2`,
max bending curvature about `4.75e-1`, and max transverse shear about
`8.4e-3`. The remaining physics mismatch is therefore still dominated by
membrane/bending path behavior rather than transverse shear.
- [x] Resolve the remaining reference-model口径 gap before changing more
production physics: the COMSOL PDF shell setup uses a selected boundary
with `Thickness and Offset` set to `Top surface on boundary` and notes that
the shell normal points inward, while the current all-S4 deck treats the
generated spherical geometry as the shell midsurface. The PDF also describes
the shell fixed constraint as a point constraint in its axisymmetric setup,
while the current 3D deck fixes the full rim node set in DOF 1-6. These
differences need a controlled diagnostic deck before the curve mismatch is
treated as a finite-shell formulation bug. The follow-up diagnostic deck
showed that top-surface offset is not the dominant scale error, while
translational rim support restores the expected bounded post-peak drop.
- [x] Run controlled reference-scope diagnostic decks at the same
`wn ~= 14.60` short target. Moving the spherical shell surface outward by
`th/2` changed the clamped-rim response only from `Fn ~= 2.108` to
`Fn ~= 2.066`, so the COMSOL top-surface offset is not the dominant scale
error. Changing the rim from full DOF 1-6 clamping to translational DOF 1-3
support softened the response to `Fn ~= 1.286` (`Fn ~= 1.296` with the same
outward half-thickness shift), much closer to the PDF interpolation
`Fn ~= 1.094` at the same `wn`. The next benchmark decision should focus on
whether the 3D cap deck should represent the PDF shell rim as translational
support rather than full rotational clamping.
- [x] Extend the rim-translational-support diagnostic toward `uz=-4.0 m`.
The run accepted 42 increments before the long diagnostic was stopped/
ended at `time ~= 0.84`, reaching `wn ~= 16.48` and `Fn ~= 0.920`.
It shows a clear post-peak load drop: the accepted path peaked at
`Fn ~= 1.341` near `wn ~= 13.74`, then decreased through `Fn ~= 1.296` at
`wn ~= 14.52`, `Fn ~= 1.097` at `wn ~= 15.70`, and `Fn ~= 0.920` at
`wn ~= 16.48`. This confirms that the previous full-rim rotational clamping
was masking the benchmark's postbuckling signature. The remaining blocker is
long-path robustness/restiffening capture for the translational-rim support
deck, not the absence of a limit point.
- [x] Solidify the benchmark support口径 before further path-control work.
The generator now writes `RIM_DOF_MAX=6` for the historical Shell-16
small-strain baseline and `RIM_DOF_MAX=3` plus `RIM, 1, 3, 0.0` for the
Shell-4 finite-strain postbuckling target. The checker now requires this
split and rejects accidental Shell-4 full rotational rim clamping; the
example README documents why the translational-rim support is the Shell-4
target after the offset/rim diagnostic.
- [x] Retry the same `uz=-4.0 m` rim-translational-support deck with
`--allow-cutback`. Cutback crossed the formerly failing increment 43 by
reducing the time step from `0.02` to `0.005`, then accepted increment 44 at
`time ~= 0.846875`, `wn ~= 16.62`, `Fn ~= 0.807`. The run was stopped after
no additional accepted row appeared over the next observation window; the
next improvement should make direct-displacement cutback/adaptivity less
prone to very small late-path steps or reduce the heavy GMRES work in this
post-peak region.
- [x] Try direct-displacement adaptive stepping on the rim-translational deck
with `DISP_ADAPTIVE=YES`, `TARGET_NEWTON=12`, `GROWTH=1.5` and
`--allow-cutback`. A shorter `uz=-3.5 m` diagnostic advanced farther than
fixed cutback, accepting to `time ~= 0.965`, `uz ~= -3.3775 m`; the path
retained the same post-peak drop but still stalled before the target. This
suggests the right production path is adaptive direct displacement plus a
post-peak linear-solver/runtime improvement, not just enabling cutback.
- [x] Re-run that adaptive `uz=-3.5 m` diagnostic with `--linear-precond ilu0`.
ILU0 reduced post-peak GMRES iterations dramatically (typical accepted
rows near `time=0.80..0.965` dropped from roughly `145..266` iterations with
Jacobi to roughly `20..30` with ILU0), but the run still stalled after the
same accepted row at `time ~= 0.965`. The remaining blocker is therefore
primarily nonlinear path/step acceptance around the steep post-peak branch;
ILU0 is useful for runtime once that path control is improved.
- [x] Expose CLI line-search diagnostics:
`--line-search-min-alpha`, `--line-search-reduction`, and
`--line-search-accept-ratio`. A smoke run accepts valid values and rejects
an invalid zero min-alpha. Lowering min alpha to `0.03125` with adaptive
displacement, cutback and ILU0 let the `uz=-3.5 m` diagnostic move slightly
farther, from `time ~= 0.965` to `time ~= 0.9675`, but it remained too slow
to finish. This confirms line-search damping helps but is not sufficient;
the next production change should focus on more robust post-peak step-size
control rather than only exposing solver knobs.
- [x] Expose CLI cutback diagnostics:
`--max-cutback` and `--cutback-factor` now override the retry count and
shrink factor for a single run without changing benchmark inputs. A compact
displacement-control smoke accepts `--allow-cutback --max-cutback 12
--cutback-factor 0.25` and rejects invalid `--cutback-factor 1.0`.
- [x] Run one long Shell-4 rim-translational diagnostic with the new path
columns before changing path-control algorithms. With adaptive displacement
stepping, ILU0, `--line-search-min-alpha 0.03125`, `--max-cutback 16` and
`--cutback-factor 0.35`, the 15-minute diagnostic was stopped after 39
accepted rows at `time ~= 0.5469`, `uz ~= -3.3907 m`, `wn ~= 16.63`,
`Fn ~= 0.747`. The path peaked at `Fn ~= 1.339` near `wn ~= 13.38` and
dropped by about 44 percent, so the translational-rim benchmark口径 is still
producing the expected post-peak branch. The new diagnostics show the
blocker is step-size collapse after repeated cutbacks (`actual_dt` down to
about `3.75e-5`), not runaway linear solves: ILU0 linear iterations peaked
at 115 and ended at 52 with relative residual around `5e-9`.
- [x] Run a compact cutback-factor comparison before changing defaults.
Keeping the same rim-translational Shell-4 setup and diagnostics,
`--cutback-factor 0.5` reached essentially the same post-peak point
(`wn ~= 16.63`, `Fn ~= 0.754`) within the observation window but retained a
larger final step (`actual_dt ~= 6.25e-4`, two cumulative cutbacks and max
linear iterations 36). A more conservative `--cutback-factor 0.7` retained
an even larger accepted step (`actual_dt ~= 1.75e-3`) but moved fewer rows
in the same time and pushed the last linear solve to about 111 iterations.
This confirms the next production change should not be a fixed cutback
factor default; it should be a guarded post-cutback recovery/growth policy
that avoids both collapse to tiny increments and expensive over-large
retries.
- [x] Try and reject an immediate/two-stage post-cutback recovery gate. The
immediate version grew the accepted `0.00125` post-cutback step to
`0.00153` and then produced no additional accepted row in the same
observation window; the delayed two-stage version avoided that premature
growth but still stopped at the same accepted row (`time ~= 0.54625`,
`wn ~= 16.61`) and did not outperform the plain `--cutback-factor 0.5`
diagnostic. The production change was therefore reverted. The remaining
path-control issue likely needs a more deliberate continuation strategy
around the steep post-peak branch, not a local growth gate.
- [x] Add and probe an explicit trust-region diagnostic for direct
displacement-control ARC_DISP paths. A positive `TRUST_RADIUS` on the
continuation line, or CLI `--trust-radius`, now reuses the existing damped
Newton correction limiter while keeping ARC_DISP path output; default
ARC_DISP remains unchanged. Compact tests cover parsing, CLI validation and
a direct-displacement path with trust enabled. Spherical-cap probes showed
`--trust-radius 0.05` was too conservative, accepting many small rows but
only reaching about `wn ~= 11.94` in the observation window. A looser
`--trust-radius 0.2` removed cutbacks and held `dt ~= 0.0025` through
`wn ~= 16.58`, but still did not outperform the plain `--cutback-factor
0.5` run within the same wall-clock window. Keep this as a controlled
diagnostic, not a default fix.
- [x] Extend continuation path output with trust-region diagnostics:
`trust_radius`, `next_trust_radius`, `trust_damping`,
`trust_merit_ratio`, `trust_adjust` and `trust_rejections`. The spherical
cap checker carries these through to `spherical_cap_load_displacement.csv`,
so future trust probes can distinguish "stable but slow because heavily
damped" from ordinary step-size or linear-solver stalls.
- [x] Re-run a looser trust diagnostic with the new trust output columns.
`--trust-radius 0.4` reached `time ~= 0.54625`, `wn ~= 16.61` and
`Fn ~= 0.816` with one cumulative cutback and max linear iterations 36.
The trust columns showed no actual correction damping on accepted rows
(`trust_damping = 1.0`), only radius adjustment near the last accepted
increment (`trust_radius ~= 0.0225`, next `~= 0.03375`). This makes the
result useful diagnostically but not a better fix than the plain
`--cutback-factor 0.5` run, which reached a slightly farther/lower
post-peak point in the same diagnostic window.
- [x] Start the explicit default continuation-path改造 instead of continuing
local parameter probes. Direct prescribed-displacement
`ARC_LENGTH_OR_DISPLACEMENT_CONTROL` now defaults to adaptive path spacing,
and accepted path rows feed back into the next step size: if the generalized
reaction changes by more than about 8 percent over the last accepted path
segment, the next displacement increment is capped before the following
Newton solve. The direct-displacement path also uses a relaxed default
internal Newton target of 12 when the input keeps the historical default
`TARGET_NEWTON=6`, so accepted 4-6 iteration steps can grow instead of
staying stuck at an early limited step. In a 90 s Shell-4 spherical-cap
observation window, the old default adaptive behavior produced 163 accepted
rows by `time ~= 0.5435` with `dt ~= 0.0031`; the new default reached the
same post-peak region in 30 accepted rows by `time ~= 0.5342`, with no
cutbacks and `next_dt ~= 0.0146`. The remaining cost is high late GMRES
iteration count, not path fragmentation.
- [x] Test two follow-up post-peak path-control ideas and reject them as
insufficient: keeping the accepted small step immediately after a cutback
did not move the `uz=-3.5 m` diagnostic beyond the previous stalled region,
so that production change was reverted; enabling the existing static
predictor changed the accepted substep pattern but again stalled around
`time ~= 0.9675`. The remaining work now requires a deliberate path-control
design choice rather than another local parameter tweak.
- [x] Record the remaining full-curve limits in README/user docs instead of
treating them as V5 release blockers: full `uz=-6.2 m` runtime, strict PDF
mid-path scale matching, later restiffening capture and late GMRES cost.
- Tests:
- [x] Checker validates the V5 bounded post-peak load-drop capability and
writes the current curve metrics to `spherical_cap_load_displacement.csv`.
- [x] Checker records but does not require strict Shell-4/8/16 PDF digitized
curve matching in this release.
- [x] Checker fails monotone curves, small-strain samples, and old
corotational diagnostic decks.
- Acceptance:
- [x] `example_spherical_cap_point_load_benchmark` validates the fast
Shell-16/Shell-4口径 checks, and the documented extended nonlinear checker
validates the bounded V5.11 finite-strain S4 postbuckling capability while
documenting deferred full-curve limitations.
- Current evidence:
- `cmake -S . -B build-codex-gcc -G "MinGW Makefiles"` configured the current
`E:\Codex\FemLite` checkout with GCC/GFortran 16.1.0.
- `cmake --build build-codex-gcc --target femlite test_shell_finite_kinematics
test_shell_finite_tangent test_shell_finite_element test_shell_internal
test_implicit_inp` passed after the solver-dispatch changes.
- `ctest --test-dir build-codex-gcc -R
"^(shell_finite_kinematics|shell_finite_tangent|shell_finite_element|shell_internal|implicit_inp)$"
--output-on-failure` passed 5/5.
- `python tools/testing/check_source_size.py . --warn-file-lines 1500
--max-file-lines 2000 --warn-routine-lines 100 --max-routine-lines 128
--exceptions tests/source_size_exceptions.txt` passed.
- After adding the direct-displacement trust diagnostics and path output
columns, `ctest --test-dir build-codex-gcc -R
"^(solver_implicit|implicit_inp|example_spherical_cap_point_load_benchmark)$"
--output-on-failure` passed 3/3, and the direct source-size check passed
with only the existing review warnings.
- After rewriting the V5 release scope, the default spherical-cap benchmark
passed with
`ctest --test-dir build-codex-gcc -R "^example_spherical_cap_point_load_benchmark$" --output-on-failure`.
The extended bounded nonlinear checker passed with
`FEMLITE_SPHERICAL_CAP_NONLINEAR=1`, 54 accepted increments, final
`wn=15.89482`, final `Fn=1.058395`, and sampled `Fn=1.340817` at
`wn=13.642`, proving a reproducible post-peak load drop without claiming
full `uz=-6.2 m` PDF curve reproduction.
### [x] DONE V5.11-7: Final documentation, benchmark matrix, and evidence
- Implementation:
- [x] Update benchmark matrix with the bounded V5.11 spherical-cap acceptance:
finite-strain S4 production path, translational-rim Shell-4 target,
post-peak load-drop evidence, continuation diagnostics and deferred strict
PDF full-curve matching.
- [x] Update acceptance report with exact commands, dates, curve metrics,
Newton/cutback counts, path-control metrics and known limits.
- [x] Remove obsolete corotational/old shell tangent documentation references
after code removal.
- [x] Add a next-version backlog note for full `uz=-6.2 m` PDF matching,
pseudo-arclength/branch following and late post-peak GMRES optimization.
- Tests:
- [x] Run targeted shell/implicit/spherical-cap/source-size regressions.
- [x] Run the relevant V5 acceptance subset if benchmark status changes.
- Acceptance:
- [x] V5.11 can be marked release-ready when implementation, tests, docs, and
benchmark evidence all align with the bounded V5 release claim and all
full-curve limitations are explicitly deferred.
## V5.12 V5 Development Plan Open Backlog
Boundary for this module:
- V5.12 is the next-version backlog and does not block the V5.11 release-ready
checkpoint.
- This section mirrors the still-open `PARTIAL` and `DEFERRED` items in
`docs/v5_development_plan.md`, especially the 0.5.2 `two_arches` reference
route and the engineering close-out priorities listed in section 20.3.
- Do not duplicate release hygiene items already closed in V5.10 unless the
development plan still describes a concrete remaining engineering action.
- A compact benchmark can be kept as regression coverage, but it cannot close a
task that explicitly calls for a production workflow, full continuation path
or high-fidelity reference case.
### [x] DONE V5.12-1: Finish `examples/two_arches` single reference-point check
- Source:
- `docs/v5_development_plan.md` now marks `examples/two_arches` as
`PASS / 0.5.2`.
- Completed coverage now has two explicit layers: a geometry/contact smoke
and a single displacement reference-point check at `para = 0.02`.
- Scope correction on 2026-07-01: the COMSOL reference uses a parametric
sweep, but FemLite V5 does not introduce a general parametric-sweep feature.
V5.12-1 only needs one representative displacement reference point, not
the complete sweep, a static edge-load solve or a full static path
reproduction.
- Implementation:
- [x] Split the current two-arches status into completed geometry/contact
smoke coverage and a separate single reference-point check.
The README and checker output now describe the existing case as a smoke
check rather than a full reference-result pass.
- 2026-06-30 diagnostic note: a direct static displacement-control conversion
of the current micro-step deck was tried as a temporary probe, including a
reduced contact penalty and an additional upper-end vertical constraint
variant. Both runs reached the prescribed midpoint displacement but produced
nonphysical residual/reaction magnitudes, so that temporary route cannot be
used as single-point reference evidence.
- [x] Keep the default dynamic micro-step as geometry/contact binding
regression, and use a separate `run_two_arches_reference_point.py` driver
to generate the selected displacement reference-point evidence.
- [x] Use `disp_upper` as the upper-arch midpoint vertical displacement and
`max_disp = 60 mm`. The single reference point uses `para = 0.02`, so the
target displacement is `1.2 mm`. This is an input/checker reference value,
not a FemLite parameter-sweep feature.
- [x] Keep the PDF geometry and mesh values explicit in the generator/input:
`Ri_upper = 80 mm`, `Ri_lower = 100 mm`, shell width/thickness `d = 3 mm`,
upper arc `80 deg`, lower arc `100 deg`, upper 30 mapped shell elements
and lower 75 mapped shell elements.
- [x] Keep material values explicit: lower arch `E = 40 GPa`, `nu = 0.2`,
`rho = 1`; upper arch `E = 20 GPa`, `nu = 0.3`, `rho = 1`.
- [x] Apply the smoke-scope in-plane constraints, lower-end fixed constraints,
out-of-plane `y` constraints and prescribed upper-midpoint displacement.
- [x] Do not add a V5 parameter-sweep or static edge-load requirement. The
selected V5 reference point is the documented upper-midpoint displacement
`disp_upper = -1.2 mm`.
- [x] Use frictionless shell-shell contact with upper arch as source/slave and
lower arch as destination/master.
- [x] Output `two_arches_reference_point.csv` with `para`, `disp_upper`,
reference mode, active contact count, max gap/penetration, contact force,
residual and run status.
- Tests:
- [x] Keep `examples/two_arches/check_two_arches_results.py` as a
geometry/contact smoke checker and make its pass message explicit.
- [x] Add a separate reference-point checker that rejects parameter-sweep,
static edge-load and non-displacement-control evidence for this V5 point.
- [x] Verify geometry coordinates, 30/75 mesh counts,
the 1.2 mm midpoint displacement, active contact and bounded dynamic
displacement for the smoke route.
- [x] Verify material values,
`disp_upper=max_disp*para` residual, contact active set, bounded gap,
nonnegative/bounded contact work and downward displacement sign. The
checker now accepts exactly one final reference point rather than a
parameter sweep.
- [x] Add a lightweight CTest entry for the optional reference-point deck:
`example_two_arches_reference_point_prepare` validates displacement
reference deck generation and `two_arches_reference_point.csv` schema.
- [x] Execute the optional single reference-point run to convergence and
record the `check_two_arches_reference_point.py` PASS evidence through
`example_two_arches_reference_point` and
`example_two_arches_reference_point_check`.
- 2026-06-30 probe: `para=0.02`, one outer solve, ILU0 preconditioner,
static predictor on, `initial_dt=0.10`, `max_dt=0.20` and a 90 s solve
timeout still produced no accepted nodal output.
- 2026-06-30 follow-up: `--trace-nonlinear` localized the first static
edge-load cost to nonlinear linear solves. The default CG path repeatedly
falls through to the 1284-DOF dense/regularized fallback, taking roughly
9-10 s per correction. The CLI and optional driver now expose
`--linear-solver`; the single-point driver defaults to GMRES and records
failure diagnostics parsed from stdout when no `history.csv` or nodal
output exists.
- 2026-06-30 follow-up probes with one-step tiny edge load
(`initial-load=-1e-3`, `initial_dt=max_dt=1.0`) showed that GMRES avoids
the dense fallback but still ends in cutback failure before a complete
static solve (`accepted_time` about `0.044` to `0.140` depending on
tolerance). BiCGSTAB/ILU0 failed earlier with a large correction norm, and
dense solve reported a singular pivot.
- 2026-07-01 scope decision: V5 does not need the static edge-load solve
for this example. The accepted V5 reference is the single displacement
point already present in the PDF-derived input, with independent
reference-point CSV/checker evidence.
- Acceptance:
- [x] `examples/two_arches` has a bounded geometry/contact smoke route that
passes without claiming the full COMSOL sweep.
- [x] `examples/two_arches` passes a single displacement reference-point
checker without relying on static edge-load convergence.
- [x] The checker compares the selected result fields against the documented
reference point from `models.sme.two_arches.pdf`.
### [x] DONE V5.12-2: Close scoped mortar / AL contact gaps for V5
- Source:
- V5.3 remains `PARTIAL` in `docs/v5_development_plan.md`; V5.12-2 closes
the bounded V5 hardening scope and explicitly defers the full production
AL/mortar tangent rewrite to a later version.
- Implementation:
- [x] Do not claim complete industrial-grade AL/mortar consistent tangent
behavior for nonlinear contact paths in V5. The production contact tangent
still uses the current compact `ktan` block; replacing it with the full
`fem_contact_mortar_tangent` audit matrix is a later high-risk contact
solver redesign item.
- [x] Expand restart coverage beyond current active/released/open multiplier
audits where the development plan still calls for more scenarios.
- [x] `test_restart` now preserves and audits contact friction history
stick/slip state counts in addition to slots, accumulated slip,
friction work and tangential force.
- [x] Add a larger OpenMP contact path benchmark: `perf_contact_pair`
accepts a third `path_steps` argument, and
`perf_contact_pair_omp_consistency` now compares a 64-facet, 4-step
force/tangent/cache path under 1/4 OpenMP threads.
- [x] Clarify node-based surface to general element master support: V5 now
explicitly defers node-to-element-master contact pair support, documents
the boundary and rejects `TYPE=NODE` slave with element master at model
build time. `tests/data/s4_node_surface_element_master.inp` and
`implicit_inp` cover the failure path.
- Tests:
- [x] Add V5-scope contact hardening tests with force, tangent, restart and
OpenMP consistency checks.
- [x] OpenMP force/tangent/cache path consistency is covered by
`perf_contact_pair_omp_consistency`.
- [x] Ensure internal BiCGSTAB/GMRES paths expose true status for
nonsymmetric or difficult tangent systems instead of hiding failures:
`linear_solver_cg` now forces BiCGSTAB/GMRES nonconvergence through the
dispatcher and checks `LS_STATUS_FAILED`, method/preconditioner, final
residual and relative residual diagnostics.
- Acceptance:
- Contact mainline is closer to a production constraint solve, with
documented convergence diagnostics, restart continuity and OpenMP path
consistency. Full industrial AL/mortar nonlinear long-path tangent
convergence is not claimed by V5 and remains deferred.
### [x] DONE V5.12-3: Close scoped Mohr-Coulomb structural workflow gaps for V5
- Source:
- V5.4 and `examples/strengthReduction` are scoped as `FOS PASS` for V5:
the delivered benchmark is the Mohr-Coulomb strength-reduction FOS curve,
plastic-zone probes, displacement probes and nonlinear diagnostic schema.
- The 0.75 m `FOS = 1.525` displacement/eqps cloud is a large
critical-instability validation problem. Its cost is inherent to the
benchmark and it is not a V5 delivery target.
- Planned INP selector:
- [x] Add a production INP-level nonlinear continuation selector:
`*NONLINEAR CONTINUATION, METHOD=value`.
- [x] Accept numeric method ids `1` through `5` and stable symbolic names so old
benchmark decks can stay compact while user-facing decks remain readable.
- [x] If the keyword is omitted, use method `1` as the default production
algorithm.
- [x] Reject unknown method ids/names at input validation time; do not silently
fall back to a different nonlinear path.
- [x] Record the selected method in nonlinear diagnostics, path CSV files and
VTK/run summaries where applicable.
- Completion evidence:
- Added `NLC_*` continuation method enums and `nonlinear_continuation_method`
default control state.
- Added INP reader support for `*NONLINEAR CONTINUATION, METHOD=value`.
- Added `tests/data/c3d4_static_nonlinear_continuation.inp` and
`implicit_inp` coverage for default method 1, ids `1..5`, symbolic names,
invalid values and state preservation on parse failure.
- Added continuation method diagnostics to strengthReduction nonlinear
chain/cloud/progress CSV output and checker validation.
- Targeted CTest command passed:
`ctest --test-dir build-win-gcc-16 -C Release -R "implicit_inp|example_strength_reduction_benchmark" --output-on-failure`.
- Implementation:
- [x] Keep V5 delivery at the FOS-curve and diagnostic-schema level for
`examples/strengthReduction`; do not require the full critical-instability
displacement/eqps cloud as release evidence.
- [x] Audit and document the Mohr-Coulomb engineering tangent in structural
solves through convergence status, continuation diagnostics and substep
counts.
- Defer structure-level automatic trial re-solve and nonlinear instability
criteria for full critical-instability cloud workflows to a later version.
- Implement method `1` / `TWO_STAGE_LOAD_STRENGTH` as the default route:
first converge the real gravity/load state at strength factor `F=1.0`,
then keep the external load fixed and continue the material strength
reduction factor from `1.0` to the target factor.
- [x] Add production nonlinear static control `static_ramp_load` so the
implicit static solver can either ramp external loads by pseudo-time or
hold the assembled external load factor fixed.
- [x] Route `strength_reduction_benchmark --nonlinear-chain` through the
method-1 two-stage solve when `nonlinear_continuation_method == 1`.
- [x] Route `--nl-write-cloud` through the same two-stage method-1
workflow, with the progress CSV tracking the fixed-load strength-factor
stage.
- [x] Keep default `*NONLINEAR CONTINUATION` behavior as method `1` and
continue writing the selected method id to nonlinear chain/cloud outputs.
- [x] Do not treat reducing method-1 cost enough for the full 0.75 m
critical-instability cloud as a V5 requirement; keep the route as an
optional diagnostic and defer full longrun validation.
- [x] Implement method `2` / `ADAPTIVE_TARGET_NEWTON` as a production
nonlinear static step-size controller that targets a bounded Newton
iteration count, grows after cheap accepted steps and shrinks after
expensive accepted steps, cutbacks or failed convergence.
- [x] Add INP controls `TARGET_NEWTON`, `GROWTH`/`GROWTH_FACTOR` and
`SHRINK`/`SHRINK_FACTOR`, with validation before state mutation.
- [x] Add production diagnostics for target increment, accepted
increment, next target increment, accepted Newton count and adjustment
reason.
- [x] Write method-2 diagnostics to strength-reduction nonlinear
chain/cloud/progress CSV files and validate them in the example checker.
- [x] Add module tests for production adaptive stepping and INP parsing,
including invalid-control preservation.
- [x] Implement method `3` / `TRUST_REGION_DAMPED_NEWTON` as a safeguarded
Newton path with correction-size limits and residual-merit damping for
high-load nonlinear segments.
- [x] Add INP controls `TRUST_RADIUS`, `TRUST_MIN_RADIUS`,
`TRUST_MAX_RADIUS`, `TRUST_GROWTH`, `TRUST_SHRINK`,
`TRUST_ACCEPT_RATIO` and `TRUST_GOOD_RATIO`.
- [x] Clip Newton corrections by the active trust radius before the
residual-merit line search.
- [x] Reject residual-increasing damped corrections with a reproducible
solver error, allowing the existing cutback/rollback path to respond.
- [x] Update the trust radius from residual merit reduction, damping and
line-search rejection count.
- [x] Record trust radius, next radius, damping factor, merit ratio,
rejection count and trust adjustment in production diagnostics and
strength-reduction CSV/summary output.
- [x] Implement method `4` / `ARC_LENGTH_OR_DISPLACEMENT_CONTROL` for
limit-point paths where pure load or strength-factor continuation cannot
pass snapback or localization events.
- [x] Add production accepted-increment path storage in `fem_control`,
including control parameter, controlled DOF count, prescribed/actual
displacement norms, constraint residual, reaction path, generalized work,
Newton iteration and convergence status.
- [x] Record direct prescribed-displacement continuation in
`implicit_solve_nonlinear_static` for `METHOD=4`, using the same
`load_factor * amplitude` boundary value that the solver enforces.
- [x] Expose the path through `fem_get_continuation_path_count`,
`fem_get_continuation_path_row` and `continuation_path.csv`.
- [x] Add low-level solver and INP/API regression coverage for accepted
increments, reaction output and zero constraint residual.
- [x] Add production Riks/arc-length controls `ARC_LENGTH`/`ARC_RADIUS` and
`ARC_LOAD_SCALE`/`RIKS_LOAD_SCALE`, with validation before mutating the
previous valid control state.
- [x] Implement the load-proportional Riks augmented equation in
`implicit_solve_nonlinear_static`: extended unknowns are displacement and
global `load_factor`; the reference load vector is computed as
`dF_ext/dlambda`; each Newton iteration solves residual and load
directions with the same constrained tangent and enforces the spherical
arc constraint.
- [x] Remove the old simplified interpretation for load-controlled method 4
cases: load-type problems no longer use ordinary Newton plus accepted-path
recording as Riks evidence. The direct path is retained only for explicit
nonzero prescribed-displacement control, and cases with neither a
proportional load nor nonzero prescribed displacement now fail instead of
using a simplified fallback.
- [x] Prove method 4 direct displacement-control path on a real
nonmonotonic structure case: the bounded Shell-4 spherical-cap nonlinear
checker uses `ARC_LENGTH_OR_DISPLACEMENT_CONTROL`, writes
`continuation_path.csv`, verifies zero displacement constraint residual,
samples generalized reaction, and rejects monotone/small-strain curves by
requiring a post-peak load drop.
- [x] Prove the load-type Riks branch of method 4 on a real structure
limit-point neighborhood: `FEMLITE_SPHERICAL_CAP_RIKS=1` converts the
Shell-4 spherical-cap target to proportional crown `CLOAD`, runs the
production Riks augmented equation, writes `spherical_cap_riks_load_path.csv`
and verifies an interior load-parameter peak followed by a load drop while
displacement continues to grow.
- [x] Improve load-type Riks branch following beyond the local limit-point
neighborhood. The Riks predictor now keeps a saved accepted augmented
direction and chooses the next predictor sign to continue that direction,
removing the local two-state oscillation on the Shell-4 spherical-cap
route and advancing down a post-limit branch segment.
- [x] Defer broader load-type Riks branch-following generalization to the
next-version backlog instead of blocking the current V5.12 checkpoint.
A pinched-hemisphere screening probe remained monotonic through the
accepted range, while the remaining block-on-arch route would couple the
continuation question with contact snap-through. Current evidence is
therefore scoped to the spherical-cap real shell limit point and
post-limit branch segment.
- [x] Implement method `5` / `MC_SUBSTEPPING_SEMISMOOTH` for local
Mohr-Coulomb material stability, including stress-point substepping,
active-set diagnostics and semismooth branch handling around corners.
- [x] Add INP controls `MC_MIN_SUBSTEPS`, `MC_MAX_SUBSTEPS`,
`MC_TARGET_STRAIN`, `MC_YIELD_TOL` and `MC_PLASTIC_LIMIT`, with
validation before mutating the previous valid control state.
- [x] Route production `MOHR_COULOMB` material updates through
`fem_material_mc_driver` when `METHOD=MC_SUBSTEPPING_SEMISMOOTH`.
- [x] Store material-point aggregate diagnostics in `statev(27:32)`:
accepted substeps, requested substeps, driver status, branch mask,
maximum yield residual and maximum plastic increment.
- [x] Keep the default and other continuation methods on the direct
single-step Mohr-Coulomb update, while writing consistent single-step
diagnostics.
- [x] Add structure-level aggregation of MC substep/branch diagnostics to
strengthReduction nonlinear chain/cloud output.
- [x] Do not require a 0.75 m critical-instability cloud longrun proof for
NC-5 in V5. NC-5 is accepted here as material-point and structure-level
diagnostic hardening; full cloud-effect proof is deferred.
- Keep the first production implementation mutually exclusive through the
`METHOD=` selector; if later evidence shows the strongest path is a
composition such as `1+2+5`, add explicit secondary flags instead of
overloading the primary method id.
- Tests:
- [x] Add INP parser tests proving omitted keyword selects method `1`, all ids
`1..5` and symbolic names are accepted, and invalid values fail before
solving.
- [x] Add solver diagnostics tests proving the selected method is reported and
that method `1` does not regress the existing nonlinear static path.
- [x] Add checker coverage for displacement cloud scale in optional
nonlinear cloud summaries.
- Defer checker coverage for full-cloud equivalent plastic strain behavior
until the later critical-instability validation track.
- [x] Reject linear-static VTK displacement output as reference evidence.
- Acceptance:
- `examples/strengthReduction` can choose any planned nonlinear continuation
method from the INP file, and the default deck uses method `1` without a
command-line override.
- Method `1` target-cloud probes report reproducible nonlinear/material
status instead of ambiguous partial results when optional diagnostics are
requested.
- Strength reduction remains a documented Mohr-Coulomb FOS benchmark in V5,
with optional nonlinear structural diagnostics; full critical-instability
displacement/plasticity evidence is not part of this version's delivery
claim.
- Partial completion evidence for method `1`:
- `cmake --build build-win-gcc-16 --config Release --target strength_reduction_benchmark test_implicit_inp`
passed on 2026-06-27.
- `ctest --test-dir build-win-gcc-16 -C Release -R "implicit_inp|example_strength_reduction_benchmark" --output-on-failure`
passed on 2026-06-27.
- Path smoke passed:
`build-win-gcc-16\strength_reduction_benchmark.exe examples\strengthReduction\strength_reduction_slope_static.inp build-win-gcc-16\strength_reduction_twostage_smoke --nonlinear-chain --nl-max-trials 1 --nl-increments 2 --nl-max-newton 3 --nl-linear-iter 100 --nl-tol 1.0e-3`.
- Cloud smoke passed:
`build-win-gcc-16\strength_reduction_benchmark.exe examples\strengthReduction\strength_reduction_slope_static.inp build-win-gcc-16\strength_reduction_twostage_cloud_smoke --nl-write-cloud --nl-cloud-factor 1.525 --nl-cloud-step 1525 --nl-increments 2 --nl-max-newton 3 --nl-linear-iter 100 --nl-tol 1.0e-3`.
- Cloud summary/checker now records and validates a `0.3 m` reference
displacement scale plus `max_displacement / 0.3 m` scale ratio, so
optional probes cannot be mistaken for accepted critical-instability cloud
matches without a quantitative displacement-scale field.
- 2026-07-01 verification:
`build-codex-gcc\strength_reduction_benchmark.exe examples\strengthReduction\strength_reduction_slope_static.inp build-codex-gcc\strength_reduction_v5123_cloud_scale_smoke --nl-write-cloud --nl-cloud-factor 1.525 --nl-cloud-step 1525 --nl-increments 2 --nl-max-newton 3 --nl-linear-iter 100 --nl-tol 1.0e-3`
generated `strength_reduction_cloud_summary.csv` with
`reference_displacement_scale=0.3`, `displacement_scale_ratio=0.0`,
`nodes=13776` and `elements=6645`; the optional cloud checker accepted
the diagnostic `FAILED` summary as a reproducible non-ambiguous trial.
- The strengthReduction checker now rejects stale `result_00000001.vtk`
files, so the old linear-static gravity precheck output cannot be used as
nonlinear cloud evidence.
- 2026-07-01 negative smoke:
`python examples\strengthReduction\check_strength_reduction_results.py build-codex-gcc\strength_reduction_linear_vtk_reject_smoke`
failed as expected with `linear-static result_00000001.vtk is not valid
nonlinear-cloud evidence`.
- Medium 0.75 m method-1 cloud probe with `--nl-increments 10 --nl-max-newton
20 --nl-linear-iter 1000 --nl-tol 1.0e-3 --nl-preconditioner ilu0` exceeded
120 s and was stopped; the full critical-instability cloud is deferred
outside V5 instead of being treated as a release blocker.
- Completion evidence for method `5` material-point and structure-level
diagnostics:
- `METHOD=MC_SUBSTEPPING_SEMISMOOTH` routes production `MOHR_COULOMB`
updates through `fem_material_mc_driver` and writes `statev(27:32)`.
- `strength_reduction_benchmark` now aggregates MC point count, updated
points, accepted/requested substeps, limited points, branch mask, maximum
yield residual and maximum plastic step into nonlinear chain/cloud/progress
rows and summary CSV output.
- `examples/strengthReduction/check_strength_reduction_results.py` rejects
missing or invalid MC row/summary diagnostics.
- `cmake --build build-win-gcc-16 --config Release --target strength_reduction_benchmark`
passed on 2026-06-27.
- Nonlinear chain/cloud smoke output under
`build-win-gcc-16\strength_reduction_mc_diag_smoke` passed
`examples\strengthReduction\check_strength_reduction_results.py`, including
optional MC chain/cloud/progress diagnostics.
- `ctest --test-dir build-win-gcc-16 -C Release -R "material_mohr_coulomb|implicit_inp|example_strength_reduction_benchmark" --output-on-failure`
passed on 2026-06-27.
- Completion evidence for method `2`:
- `cmake --build build-win-gcc-16 --config Release --target test_solver_implicit test_implicit_inp strength_reduction_benchmark`
passed on 2026-06-27.
- `ctest --test-dir build-win-gcc-16 -C Release -R "solver_implicit|implicit_inp|example_strength_reduction_benchmark" --output-on-failure`
passed on 2026-06-27.
- `solver_implicit` now verifies adaptive growth reaches the final static
time with fewer accepted increments than the fixed initial step would use.
- `implicit_inp` verifies `METHOD=ADAPTIVE_TARGET_NEWTON`,
`TARGET_NEWTON`, `GROWTH` and `SHRINK`, including failure without changing
previous valid settings.
- `examples/strengthReduction/check_strength_reduction_results.py` now
rejects malformed nonlinear chain/cloud diagnostics for target increment,
accepted increment, next target increment and adjustment reason.
- Completion evidence for method `3`:
- `cmake --build build-win-gcc-16 --config Release --target test_solver_implicit test_implicit_inp strength_reduction_benchmark`
passed on 2026-06-27.
- `ctest --test-dir build-win-gcc-16 -C Release -R "solver_implicit|implicit_inp|example_strength_reduction_benchmark" --output-on-failure`
passed on 2026-06-27.
- `solver_implicit` now verifies that `METHOD=TRUST_REGION_DAMPED_NEWTON`
limits the applied correction norm, converges through multiple damped
Newton iterations, updates trust diagnostics and preserves the expected
displacement solution.
- `implicit_inp` verifies all trust-region control parameters and confirms
invalid trust radius bounds fail without mutating previous valid settings.
- `examples/strengthReduction/check_strength_reduction_results.py` now
rejects missing or invalid trust-radius, damping, merit-ratio and
rejection-count diagnostics in nonlinear chain/cloud/progress output.
- Partial completion evidence for method `4`:
- `fem_control` now owns a dynamically sized accepted-increment continuation
path table; `fem_solver_implicit` records direct displacement-control rows
after nonlinear static increments converge, while the active load factor is
still the one used by the solved increment.
- `fem_control` also stores `nlc_arc_length_radius`,
`nlc_arc_load_scale`, `nlc_last_arc_constraint`,
`nlc_last_arc_lambda` and `nlc_last_arc_length`.
- `fem_inp_nlc` parses `ARC_LENGTH`/`ARC_RADIUS` and
`ARC_LOAD_SCALE`/`RIKS_LOAD_SCALE` for
`METHOD=ARC_LENGTH_OR_DISPLACEMENT_CONTROL`.
- `fem_solver_implicit` now routes load-type method 4 cases to a Riks
augmented equation with displacement plus `load_factor` unknowns and keeps
direct displacement-control path recording separate from Riks evidence.
Method 4 inputs with neither proportional load nor nonzero prescribed
displacement return a solver error instead of falling back to ordinary
Newton path recording.
- The bounded Shell-4 spherical-cap nonlinear checker now provides the first
real nonmonotonic structure evidence for the direct displacement-control
branch: `FEMLITE_SPHERICAL_CAP_NONLINEAR=1` runs the finite-strain S4 deck
to `uz=-3.24 m`, writes `continuation_path.csv` and
`spherical_cap_load_displacement.csv`, verifies the accepted displacement
path and requires a post-peak generalized-reaction load drop. This closes
the bounded direct-displacement evidence, while the load-type Riks evidence
is tracked separately below.
- The same checker now has an optional load-control Riks route via
`FEMLITE_SPHERICAL_CAP_RIKS=1`. On 2026-06-30 it passed with 80 accepted
increments, `peak_lambda=0.6909629`, `final_lambda=0.6017746` and final
displacement magnitude `1.3973472 m`, proving a real finite-strain S4
limit-point and post-limit branch segment after adding predictor direction
memory. Broader branch-following generalization on a second independent
nonmonotonic structure is explicitly deferred to the next-version backlog.
- `fem_api` exposes row-by-row continuation path access and
`fem_output_csv` writes `continuation_path.csv` when path data exists.
- `tests/data/c3d4_static_arc_disp_continuation.inp` selects
`METHOD=ARC_LENGTH_OR_DISPLACEMENT_CONTROL`; `implicit_inp` verifies the
API path rows and `solver_implicit` verifies the in-memory direct
displacement path, the load-type Riks path and the no-control rejection.
- `cmake --build build-win-gcc-16 --config Release --target test_solver_implicit test_implicit_inp`
passed on 2026-06-27.
- `ctest --test-dir build-win-gcc-16 -C Release -R "solver_implicit|implicit_inp" --output-on-failure`
passed on 2026-06-27.
- Partial completion evidence for method `5`:
- `src/base/fem_control.f90` now stores MC substep controls and
`src/input/fem_inp_nlc.f90` parses/validates them from
`*NONLINEAR CONTINUATION`.
- `src/material/fem_material_j2.f90` now dispatches production
`MOHR_COULOMB` updates to `fem_material_mc_driver` only when
`METHOD=MC_SUBSTEPPING_SEMISMOOTH`.
- `src/material/fem_material_mohr_coulomb.f90` now writes consistent
single-step state diagnostics and reserves `statev(27:32)` for MC
substep aggregates.
- `cmake --build build-win-gcc-16 --config Release --target test_material_mohr_coulomb test_implicit_inp`
passed on 2026-06-27.
- `ctest --test-dir build-win-gcc-16 -C Release -R "material_mohr_coulomb|implicit_inp" --output-on-failure`
passed on 2026-06-27.
### [x] DONE V5.12-4: Move bounded buckling and cyclic symmetry into the main solver path
- Source:
- V5.5 and `examples/buckling_corrugated_shell` remain `PARTIAL` by design:
V5 now delivers a bounded production subset and defers the large shell /
harmonic-wave / high-fidelity expansion to the next version line.
- Implementation:
- [x] Add a main-solver Euler column buckling workflow through
`femlite --buckling` and the dedicated `*BUCKLING COLUMN` input block.
- [x] Move Euler column input, assembly, dense buckling solve and CSV output
into production `src/solver/fem_buckling_column.f90`.
- [x] Delete the old duplicate Euler helper-side assembly/solve algorithm
from `tools/benchmarks/euler_column_buckling_benchmark.f90`; keep only a
compatibility wrapper around the production module.
- [x] Add a main INP parser guard for generic `*BUCKLING`: the parser records
a buckling analysis type and no longer falls through to static or dynamic
analysis.
- [x] Add a small-scale dense generic `*BUCKLING` main-solver path for
already assembled linear elastic `K` plus the limited
`*BUCKLING GEOMETRIC LINE` truss-style `K_G` subset.
- [x] Add the first static-preload-to-buckling main-solver bridge: when a
`*STATIC` step precedes `*BUCKLING`, run a linear static preload, update
element stress/internal force state, then solve the current dense buckling
subset.
- [x] Add the first preload-derived geometric stiffness subset:
`*BUCKLING GEOMETRIC LINE, PRELOAD` projects host solid element stress onto
the line direction, multiplies by the supplied area and assembles the
truss-style `K_G`.
- [x] Add the first production element preload geometric stiffness subset:
when `*STATIC` precedes `*BUCKLING` and no explicit geometric line is
supplied, assemble C3D4/C3D8 solid initial-stress `K_G` from the preloaded
element stress state.
- [x] Defer full preload-to-buckling `*BUCKLING` for shell production models
and higher-order solid elements to post-V5. V5 delivery is bounded to
explicit geometric lines, preload-derived line `K_G`, and C3D4/C3D8 solid
initial-stress `K_G`.
- [x] Defer shell geometric stiffness and higher-order solid geometric
stiffness after static preload to post-V5.
- [x] Integrate generic buckling modal output with the normal output workflow.
- [x] Bring the first cyclic sector constraint subset into the main solve
path: `*CYCLIC SYMMETRY` accepts explicit master/slave node pairs and
expands WAVE=0 sector rotation into MPC equations for 3-DOF solid nodes and
6-DOF shell nodes, including shell rotational DOF mapping.
- [x] Defer automatic cyclic boundary node pairing and nonzero harmonic-wave
main-solver semantics to post-V5.
- [x] Defer full shell FE preload-buckling high-fidelity corrugated-shell
coverage to post-V5.
- Tests:
- [x] Add module coverage for Euler column parsing, validation, solve
diagnostics, output writing and mode boundary conditions.
- [x] Update `examples/euler_column_buckling` checker and double-click script
so the regression runs `femlite --buckling`, not the benchmark helper.
- [x] Run targeted CLI regression after adding the new analysis mode.
- [x] Add general input-level `*BUCKLING` parser and small dense CLI solve
tests.
- [x] Add static-preload-to-dense-buckling CLI smoke coverage.
- [x] Add static-preload-to-dense-buckling tests and cyclic sector shell/solid
mapping tests.
- [x] Add C3D4 solid initial-stress preload-to-buckling CLI smoke coverage.
- [x] Defer full shell and higher-order solid preload-to-buckling tests to
post-V5 with the corresponding algorithms.
- [x] Keep Euler column benchmark as regression coverage through CTest.
- [x] Keep compact corrugated benchmark as regression coverage while adding
main-solver cyclic/preload coverage.
- Acceptance:
- [x] Users can run the Euler column buckling workflow through
`femlite --buckling`, not only through benchmark helper tools.
- [x] Users can run a small dense generic `*BUCKLING` workflow through normal
INP input when `K_G` is supplied by `*BUCKLING GEOMETRIC LINE`.
- [x] Users can run a linear static preload immediately before the small dense
generic buckling subset.
- [x] Users can run a small C3D4/C3D8 solid preload-to-buckling workflow
without manually supplying `*BUCKLING GEOMETRIC LINE`.
- [x] Defer general shell preload-to-buckling workflow through normal INP to
post-V5; V5 explicitly supports the bounded solid/line subsets above.
- [x] WAVE=0 cyclic sector constraints work through the main solver path for
explicit solid and shell node pairs.
- Completion evidence:
- `build-win-gcc-16\femlite.exe examples\euler_column_buckling\euler_column_buckling.inp --buckling --output-dir build-win-gcc-16\app_buckling`
passed on 2026-06-27 with first mode `1.84239E+06`, Euler reference
`1.84233E+06`, residual `1.50015E-11` and relative error `3.27661E-05`.
- `ctest --test-dir build-win-gcc-16 -R "^(buckling_support|example_euler_column_buckling_benchmark|smoke)$" --output-on-failure`
passed on 2026-06-27.
- `ctest --test-dir build-win-gcc-16 -R "^(cli_implicit_linear|cli_static_linear|cli_static_nonlinear|cli_implicit_newmark|cli_implicit_newmark_nonlinear)$" --output-on-failure`
passed on 2026-06-27.
- 2026-07-01 small dense generic buckling evidence:
`ctest --test-dir build-codex-gcc -R "^(implicit_inp|cli_buckling_generic_dense|cli_buckling_static_preload_dense|buckling_support|example_euler_column_buckling_benchmark)$" --output-on-failure`
passed 5/5. `cli_buckling_generic_dense` now runs through a nested
`--output-dir` and verifies `buckling_summary.csv` plus
`buckling_modes.csv` are written by the normal output-directory workflow.
`femlite tests\data\c3d4_buckling_static_preload.inp --output-dir
build-codex-gcc\manual_buckling_static_preload` completed the preload
bridge with first load factor `3.33333E-01` and residual
`2.22045E-16`.
- 2026-07-01 corrugated benchmark regression evidence:
`ctest --test-dir build-codex-gcc -R "^example_buckling_corrugated_shell_benchmark$" --output-on-failure`
passed 1/1 and remains labeled `v5_acceptance`.
- 2026-07-01 cyclic main-solver evidence:
`ctest --test-dir build-codex-gcc -R "^(implicit_inp|cyclic_symmetry|cli_buckling_generic_dense|cli_buckling_static_preload_dense|buckling_support|example_euler_column_buckling_benchmark|example_buckling_corrugated_shell_benchmark)$" --output-on-failure`
passed 7/7. `implicit_inp` covers `*CYCLIC SYMMETRY` expansion into MPC
equations for a C3D4 3-DOF solid pair and an S4 6-DOF shell pair.
- 2026-07-01 solid initial-stress buckling evidence:
`ctest --test-dir build-codex-gcc -R "^(implicit_inp|cyclic_symmetry|cli_buckling_generic_dense|cli_buckling_static_preload_dense|cli_buckling_solid_preload_dense|buckling_support|example_euler_column_buckling_benchmark|example_buckling_corrugated_shell_benchmark)$" --output-on-failure`
passed 8/8. `femlite tests\data\c3d4_buckling_solid_preload.inp
--output-dir build-codex-gcc\manual_buckling_solid_preload` completed with
first load factor `1.66667E+00`, preload residual `0.00000E+00` and modal
residual `0.00000E+00`.
### [x] DEFERRED V5.12-5: Promote roll forming from benchmark helpers to production workflow
- Source:
- V5.6 and `examples/roll_forming` remain `PARTIAL`.
- V5 decision:
- Defer the production roll-forming workflow to post-V5.
- Keep roller kinematics and compact J2 springback benchmarks as regression
evidence only.
- Do not claim full 3D workpiece rolling, restart-driven multi-step forming,
unloading/springback production workflow, or longrun stability in V5.
- Rationale:
- This is an independent manufacturing process workflow with rigid roller
kinematics, contact/friction, plastic forming, restart staging and longrun
validation. It is too large to add safely during V5 close-out.
### [x] DEFERRED V5.12-6: Integrate cohesive contact into the main contact workflow
- Source:
- V5.7 and `examples/ball_adhesion_decohesion` remain `PARTIAL`.
- V5 decision:
- Defer production cohesive-contact integration to post-V5.
- Keep cohesive law/state/metrics tests and the compact ball-adhesion
benchmark as bounded regression evidence only.
- Do not claim ordinary-contact/cohesive switching, cohesive contact tangent,
full restart history integration, or full ball-membrane FE cohesive long
path in V5.
- Rationale:
- This changes the main contact workflow, tangent assembly and restart
persistence. It should be developed with a dedicated contact-integration
plan rather than added during V5 close-out.
### [x] DEFERRED V5.12-7: Upgrade deferred and partial high-fidelity examples
- Source:
- `docs/v5_development_plan.md` lists several example directories as
`DEFERRED` or `PARTIAL`.
- V5 decision:
- Defer high-fidelity upgrades for `shell_solid_connection`,
`buckling_corrugated_shell`, `roll_forming`, `ball_adhesion_decohesion` and
larger particle-container longruns to post-V5.
- Keep current compact benchmarks and smoke tests only as regression coverage.
- Do not list compact surrogate, fallback or smoke-only examples as V5 PASS.
- Rationale:
- These examples depend on the larger deferred algorithms above. Upgrading the
examples before the production algorithms would create misleading
acceptance evidence.
### [x] DEFERRED V5.12-8: Complete longrun and performance close-out
- Source:
- V5.9/V5.10 remain `PARTIAL` in the development plan because larger
longrun and full performance close-out remain recommended.
- V5 decision:
- Defer the larger longrun matrix, full performance matrix and multi-platform
longrun records to post-V5.
- Keep the completed source-size governance and targeted behavior regression
as V5 close-out evidence.
- Do not keep broad longrun/performance expansion on the V5 must-complete
list.
- Implementation:
- [x] Defer longrun tiers for contact, Mohr-Coulomb, buckling, roll forming,
cohesive contact and particle-container coupling to post-V5.
- [x] Defer memory, iteration, residual, timing and platform stability metric
collection for broad longruns to post-V5.
- [x] Update `docs/v5_benchmark_matrix.md` only when new
longrun/performance tests are formalized after V5.
- [x] Restore `source_size` governance after the Riks/NLC and buckling work:
split `*NONLINEAR CONTINUATION` parsing into production
`src/input/fem_inp_nlc.f90`, keep `fem_inp_helpers` as generic keyword/data
helpers, shorten the dense solver, CG solver, S4 stiffness and NLC parser
test routines below the hard routine threshold, and document remaining
structural exceptions with explicit split reasons in
`tests/source_size_exceptions.txt`.
- [x] Split implicit constraint/tangent application out of
`fem_solver_implicit`: added production
`src/solver/fem_implicit_constraints.f90` for Dirichlet, Robin, MPC,
follower-load tangent and contact tangent application, removed the old
private duplicate routines from the main implicit solver, and kept the
nonlinear/static/Newmark behavior covered by targeted regression.
- [x] Split database lifecycle cleanup out of `fem_database`: added
production `src/mesh/fem_database_lifecycle.f90`, kept
`database_finalize(ierr)` as the stable public wrapper, removed stale
contact/free helper wrappers, and removed the now-stale hard file exception
for `src/mesh/fem_database.f90` after the file dropped below 2000 lines.
`database_finalize_storage` is now split into private model-array,
material-array and step/load-array release helpers, so
`fem_database_lifecycle.f90` no longer appears in the source-size
routine review-warning list.
- [x] Split stable time-step support out of the public API facade: added
production `src/base/fem_api_stable_dt.f90` for material wave speed,
element characteristic length and current-position helper logic, kept
`fem_estimate_stable_dt` as the stable public API wrapper, and removed the
now-stale hard file exception for `src/base/fem_api.f90` after the file
dropped below 2000 lines.
- [x] Split the strengthReduction benchmark orchestration outside `src`:
added `tools/benchmarks/strength_reduction_benchmark_cli.f90` for CLI and
nonlinear-chain option parsing, added
`tools/benchmarks/strength_reduction_benchmark_report.f90` for CSV/summary
naming and report formatting, and split cloud/two-stage progress helpers
inside the benchmark driver so all benchmark routines are below the hard
128-line threshold.
- [x] Split the hard routine-size exceptions in `fem_solver_implicit`:
`implicit_solve_nonlinear_static` is now a thin dispatcher over begin,
fixed/adaptive scheduling and arc driver helpers; Riks/arc-length
initialization, predictor and constraint-measure updates are separated from
`static_arc_increment`; nonlinear static Newton correction, line search and
trust-region update are separated from `nonlinear_static_increment`. All
routine entries were removed from `tests/source_size_exceptions.txt`.
- [x] Clear the remaining file-level `source_size` exception for
`fem_solver_implicit`: added production
`src/solver/fem_solver_implicit_support.f90` for shared convergence,
nonlinear linear-solver fallback, stiffness/external assembly, CSR/mass and
trust-region helpers; added production `src/solver/fem_solver_newmark.f90`
for linear and nonlinear Newmark integration; kept the public Newmark
entry points re-exported by `fem_solver_implicit`. The active entries in
`tests/source_size_exceptions.txt` are now empty.
- [x] Continue review-threshold cleanup for the current implicit solver split:
arc driver initialization, Riks iteration evaluation/correction and
Newmark nonlinear initialization/correction/success/failure handling were
moved into helpers. `implicit_solve_static_arc`,
`static_arc_increment` and `implicit_newmark_nonlinear_step` no longer
appear in the `source_size` review-warning list.
- [x] Continue file-level review-target cleanup for `fem_solver_implicit`:
added production `src/solver/fem_solver_static_nonlinear.f90` for
nonlinear static fixed/adaptive scheduling, static predictor history,
direct displacement-control path recording, load-type Riks/arc-length and
nonlinear static Newton increments. `fem_solver_implicit` now keeps linear
static plus dispatcher/public-entry ownership and no longer appears in the
source-size review-warning list.
- [x] Defer any further review-target cleanup on other large production modules
where it clarifies ownership and preserves full algorithms, especially
`fem_element_internal` and benchmark tool files, to the next normal
engineering cycle; do not keep it as a V5 release blocker.
- [x] `fem_api`: extracted shared Newmark run preparation for implicit and
nonlinear Newmark API runs; `fem_run_newmark_nonlinear` no longer appears
in the source-size routine warning list.
- [x] `fem_api`: split the public API facade into production support modules:
`fem_api_state` owns the initialized flag, `fem_api_coupling` owns
SPH/contact/surface/facet API routines, `fem_api_diagnostics` owns
convergence/continuation/linear-solver/shell-stability API routines, and
`fem_api_state_access` owns node/element state query API routines. The
original `fem_api` module keeps initialization, model I/O, solve/output
orchestration and re-exports the moved public entries; `fem_api.f90` no
longer appears in the source-size file-level review-warning list.
- [x] `fem_control`: split `control_set_defaults` into private grouped
reset helpers for time/nonlinear controls, solver controls, diagnostics
and feature flags. The public default-reset entry remains unchanged, and
`control_set_defaults` no longer appears in the source-size routine
review-warning list.
- [x] `fem_material_j2`: split J2 plastic state update into
`update_j2_plastic_state`, keeping the return-mapping algorithm intact;
changed `stress_dependent_modulus` to preserve the caller-provided Young's
modulus value on error returns. `material_update_j2` no longer appears in
the source-size routine review-warning list, and the Windows GCC
maybe-uninitialized warning for SDE Young's modulus is gone.
- [x] `fem_database`: moved database initialization storage reset into
`database_initialize_storage` in `fem_database_lifecycle`, keeping the
public `database_initialize(ierr)` wrapper stable; added production
`fem_database_lookup` for ID/name queries and surface face helpers, added
production `fem_database_validation` for full model validation, and kept
`database_validate_model(ierr)` as a stable wrapper. `fem_database.f90`,
`database_initialize`, `validate_materials` and `validate_contact_data`
no longer appear in the source-size review-warning list.
- [x] `fem_material_mohr_coulomb`: initialized MC parameter and Lame
parameter outputs on all error paths, clearing the Windows GCC
maybe-uninitialized warnings for tension, lambda and shear without
changing the return-mapping or tangent algorithms.
- [x] `fem_linear_solver_dense`: split dense CSR copy, partial-pivot
elimination/back-substitution, pivot selection and residual calculation
into private helpers while preserving the internal dense pivoting fallback
algorithm. `linear_dense_solve` no longer appears in the source-size
routine review-warning list.
- [x] `fem_linear_solver_iterative`: split CG preconditioner/initialization
setup, BiCGSTAB initialization and omega correction, and GMRES
initialization into private helpers while preserving the production
iteration formulas and internal preconditioner dispatch. `linear_cg_solve`,
`linear_bicgstab_solve` and `linear_gmres_solve` no longer appear in the
source-size routine review-warning list.
- [x] `fem_assembly_csr`: split CSR pattern row counting and row-column
population into private helpers while preserving element/MPC/contact
sparsity insertion and row sorting. `assembly_build_csr_pattern` no
longer appears in the source-size routine review-warning list.
- [x] `fem_inp_helpers`: moved Mohr-Coulomb material line parsing into
production `fem_inp_material_parser`, moved contact pair/property/friction
parsing into production `fem_inp_contact_parser`, and kept
`fem_inp_helpers` focused on generic token/keyword helpers and shared INP
utilities. `parse_mc_line`, `parse_contact_pair_header` and the
file-level `fem_inp_helpers.f90` warning no longer appear in the
source-size review-warning list.
- [x] `fem_inp_reader`: split count-model state reset and node/element
total validation out of the first-pass INP counting routine. The
first-pass keyword/data counting behavior remains unchanged, and
`inp_count_model` no longer appears in the source-size routine
review-warning list.
- [x] `fem_output_csv`: split contact CSV header, summary row and pair row
writing into private helpers while preserving the existing CSV schema and
contact diagnostics fields. `output_csv_write_contact` no longer appears
in the source-size routine review-warning list.
- [x] `fem_particle_config_check`: split particle check limit parsing into
`pcheck_parse_limits`, keeping line cleaning, CHECK data-line handling and
final validation unchanged. `pcheck_parse_line` no longer appears in the
source-size routine review-warning list.
- Tests:
- [x] Defer documented extended commands for broad longrun/performance suites
until those suites are formalized after V5.
- [x] Re-run relevant Windows GCC, Windows Intel/MSVC and WSL GNU subsets when
longrun coverage changes after V5.
- [x] Run targeted source-size and behavior regression after the parser/solver
cleanup.
Completion evidence on 2026-06-27:
`ctest --test-dir build-win-gcc-16 -R "^(implicit_inp|linear_solver_cg|shell_internal|source_size)$" --output-on-failure`
and
`ctest --test-dir build-win-gcc-16 -R "^(solver_implicit|implicit_inp|linear_solver_cg|shell_internal|buckling_support|example_euler_column_buckling_benchmark|cli_implicit_linear|cli_static_linear|cli_static_nonlinear|cli_implicit_newmark|cli_implicit_newmark_nonlinear)$" --output-on-failure`
both passed.
- [x] Run targeted regression after splitting implicit constraint helpers.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_solver_implicit test_implicit_inp -j 4`
and
`ctest --test-dir build-win-gcc-16 -R "^(solver_implicit|implicit_inp|source_size)$" --output-on-failure`
both passed.
- [x] Run targeted build and source-size cleanup after splitting database
lifecycle.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_implicit_inp test_solver_implicit -j 4`
passed; source-size was rerun and exposed the stale
`src/mesh/fem_database.f90` file exception, which was removed; rerun
`ctest --test-dir build-win-gcc-16 -R "^(solver_implicit|implicit_inp|source_size)$" --output-on-failure`
passed.
- [x] Run targeted build and source-size cleanup after splitting API stable
time-step support.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_implicit_inp test_solver_implicit -j 4`
passed; source-size was rerun and exposed the stale
`src/base/fem_api.f90` file exception, which was removed; rerun
`ctest --test-dir build-win-gcc-16 -R "^(solver_implicit|implicit_inp|source_size)$" --output-on-failure`
passed.
- [x] Run targeted build and source-size cleanup after splitting
strengthReduction benchmark CLI/report/cloud helpers.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target strength_reduction_benchmark -j 4`
passed; `ctest --test-dir build-win-gcc-16 -R "^(source_size|example_strength_reduction_benchmark)$" --output-on-failure`
passed after removing stale benchmark routine exceptions.
- [x] Run targeted build and source-size cleanup after splitting
`fem_solver_implicit` nonlinear static and Riks routine helpers.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_solver_implicit test_implicit_inp -j 4`
passed; `ctest --test-dir build-win-gcc-16 -R "^(solver_implicit|implicit_inp|source_size)$" --output-on-failure`
passed after removing stale `implicit_solve_nonlinear_static`,
`static_arc_increment` and `nonlinear_static_increment` routine exceptions.
- [x] Run targeted build and source-size cleanup after splitting implicit
support and Newmark into focused production modules.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_solver_implicit test_implicit_inp -j 4`
passed; `ctest --test-dir build-win-gcc-16 -R "^(solver_implicit|implicit_inp|source_size)$" --output-on-failure`
passed after removing the stale `src/solver/fem_solver_implicit.f90` file
exception. `tests/source_size_exceptions.txt` now contains no active
exceptions.
- [x] Run targeted build and source-size cleanup after reducing the current
implicit solver review-size routines.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_solver_implicit test_implicit_inp -j 4`
passed; `ctest --test-dir build-win-gcc-16 -R "^(solver_implicit|implicit_inp|source_size)$" --output-on-failure`
passed after `implicit_solve_static_arc`, `static_arc_increment` and
`implicit_newmark_nonlinear_step` disappeared from the source-size warning
list.
- [x] Run targeted build and source-size cleanup after moving nonlinear static
scheduling, Riks/arc-length and static Newton increments into
`fem_solver_static_nonlinear`.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_solver_implicit test_implicit_inp -j 4`
passed; `ctest --test-dir build-win-gcc-16 -R "^(solver_implicit|implicit_inp|source_size)$" --output-on-failure`
passed, and `fem_solver_implicit.f90` no longer appears in the
`source_size` warning list.
- [x] Run targeted build and API/source-size cleanup after extracting shared
Newmark run preparation inside `fem_api`.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_solver_implicit test_implicit_inp -j 4`
passed; `ctest --test-dir build-win-gcc-16 -R "^(solver_implicit|implicit_inp|source_size|c_api)$" --output-on-failure`
passed, and `fem_run_newmark_nonlinear` no longer appears in the
`source_size` routine warning list.
- [x] Run targeted build and API facade/source-size cleanup after splitting
coupling, diagnostics and state-access API routines out of `fem_api`.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_c_api test_smoke test_solver_implicit test_implicit_inp -j 4`
passed; `ctest --test-dir build-win-gcc-16 -R "^(smoke|c_api|solver_implicit|implicit_inp|source_size)$" --output-on-failure`
passed, and `fem_api.f90` no longer appears in the `source_size`
file-level review-warning list.
- [x] Run targeted build and source-size cleanup after splitting database
lifecycle release categories.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_smoke test_restart test_implicit_inp -j 4`
passed; `ctest --test-dir build-win-gcc-16 -R "^(smoke|restart|implicit_inp|source_size)$" --output-on-failure`
passed, and `database_finalize_storage` no longer appears in the
`source_size` routine review-warning list.
- [x] Run targeted build and source-size cleanup after splitting
`control_set_defaults`.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_implicit_inp test_solver_implicit test_element_stiffness test_linear_solver_cg -j 4`
passed; `ctest --test-dir build-win-gcc-16 -R "^(implicit_inp|solver_implicit|element_stiffness|linear_solver_cg|source_size)$" --output-on-failure`
passed, and `control_set_defaults` no longer appears in the `source_size`
routine review-warning list.
- [x] Run targeted build and source-size cleanup after splitting J2 plastic
state update and clearing the SDE Young's modulus warning.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_material_j2 test_material_mohr_coulomb test_implicit_inp -j 4`
passed without the previous `young may be used uninitialized` warning;
`ctest --test-dir build-win-gcc-16 -R "^(material_j2|material_mohr_coulomb|implicit_inp|source_size)$" --output-on-failure`
passed, and `fem_material_j2.f90` no longer appears in the `source_size`
warning list.
- [x] Run targeted build and source-size cleanup after moving database
initialization into lifecycle storage helpers and splitting database
validation routines.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_database_topology test_smoke test_restart test_implicit_inp -j 4`
passed; `ctest --test-dir build-win-gcc-16 -R "^(database_topology|smoke|restart|implicit_inp|source_size)$" --output-on-failure`
passed, and `database_initialize` no longer appears in the `source_size`
warning list. After splitting validation helpers and clearing MC
maybe-uninitialized warnings,
`cmake --build build-win-gcc-16 --target test_material_mohr_coulomb test_database_topology test_implicit_inp -j 4`
passed without the previous MC warnings, and
`ctest --test-dir build-win-gcc-16 -R "^(material_mohr_coulomb|database_topology|implicit_inp|source_size)$" --output-on-failure`
passed.
- [x] Run targeted build and source-size cleanup after moving database lookup
and validation into dedicated production modules.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_database_topology test_smoke test_implicit_inp test_contact -j 4`
passed; `ctest --test-dir build-win-gcc-16 -R "^(database_topology|smoke|implicit_inp|contact|source_size)$" --output-on-failure`
passed. Direct `check_source_size.py` output no longer lists
`src/mesh/fem_database.f90` or its validation routines.
- [x] Run targeted build and source-size cleanup after splitting the dense
solver fallback.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_linear_solver_cg test_solver_implicit test_implicit_inp -j 4`
passed; `ctest --test-dir build-win-gcc-16 -R "^(linear_solver_cg|solver_implicit|implicit_inp|source_size)$" --output-on-failure`
passed, and direct `check_source_size.py` output no longer lists
`linear_dense_solve`.
- [x] Run targeted build and source-size cleanup after splitting the
iterative CG/BiCGSTAB/GMRES solver routines.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_linear_solver_cg test_solver_implicit -j 4`
passed; `ctest --test-dir build-win-gcc-16 -R "^(linear_solver_cg|solver_implicit|source_size)$" --output-on-failure`
passed. Direct `check_source_size.py` output no longer lists
`linear_cg_solve`, `linear_bicgstab_solve` or `linear_gmres_solve`.
- [x] Run targeted build and source-size cleanup after splitting CSR assembly
pattern helpers.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_assembly_csr test_solver_implicit test_implicit_inp -j 4`
passed; `ctest --test-dir build-win-gcc-16 -R "^(assembly_csr|solver_implicit|implicit_inp|source_size)$" --output-on-failure`
passed, and direct `check_source_size.py` output no longer lists
`assembly_build_csr_pattern`.
- [x] Run targeted build and source-size cleanup after moving material/contact
INP parsing out of `fem_inp_helpers`.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_implicit_inp test_contact test_material_mohr_coulomb -j 4`
passed; `ctest --test-dir build-win-gcc-16 -R "^(implicit_inp|contact|material_mohr_coulomb|source_size)$" --output-on-failure`
passed. Direct `check_source_size.py` output no longer lists
`src/input/fem_inp_helpers.f90`, `parse_mc_line` or
`parse_contact_pair_header`.
- [x] Run targeted build and source-size cleanup after splitting the INP
first-pass counter.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_implicit_inp -j 4`
passed; `ctest --test-dir build-win-gcc-16 -R "^(implicit_inp|source_size)$" --output-on-failure`
passed, and direct `check_source_size.py` output no longer lists
`inp_count_model`.
- [x] Run targeted build and source-size cleanup after splitting contact CSV
output helpers.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_vtk_output test_contact test_implicit_inp -j 4`
passed; `ctest --test-dir build-win-gcc-16 -R "^(vtk_output|contact|implicit_inp|source_size)$" --output-on-failure`
passed, and direct `check_source_size.py` output no longer lists
`output_csv_write_contact`.
- [x] Run targeted build and source-size cleanup after splitting particle
check parsing.
Completion evidence on 2026-06-27:
`cmake --build build-win-gcc-16 --target test_particle_config_check test_particle_config_suite -j 4`
passed; `ctest --test-dir build-win-gcc-16 -R "^(particle_config_check|particle_config_suite|source_size)$" --output-on-failure`
passed, and direct `check_source_size.py` output no longer lists
`pcheck_parse_line`.
- [x] Run V5 close-out regression for the completed subset.
Completion evidence on 2026-07-01:
`ctest --test-dir build-codex-gcc -R "^(implicit_inp|cyclic_symmetry|cli_buckling_generic_dense|cli_buckling_static_preload_dense|cli_buckling_solid_preload_dense|buckling_support|example_euler_column_buckling_benchmark|example_buckling_corrugated_shell_benchmark)$" --output-on-failure`
passed 8/8, and
`ctest --test-dir build-codex-gcc -L v5_acceptance --output-on-failure`
passed 39/39.
- Acceptance:
- V5 keeps the current targeted regression and source-size governance as
close-out evidence.
- Broad performance and longrun coverage is explicitly deferred and is not a
V5 must-complete acceptance item.