QF Solver

QF Solver is an open-source Python finite-element solver for structural mechanics and dynamics, designed for inspectable formulations, engineering-method development and reproducible numerical verification. It offers linear, dynamic and selected nonlinear analysis routes through a Python API and command-line interface. Tests, benchmarks, evidence records and explicit scope limits help users understand both what a result shows and what it does not show.

QF Solver is intended for engineering-method development, research and reproducible studies. An implemented route or a converged example is not automatically qualified or physically validated. The project is not certified and is not a general-purpose replacement for industrial FEA software.

On this page: Capabilities · Modeling · How it works · Installation · Quick start · Analyses · Nonlinear mechanics · Rotating dynamics · Results · Larger models · Verification · Limitations · Current release.

What QF Solver is for

Use QF Solver when you need to inspect how a structural finite-element model is assembled and solved, develop a mechanical method, compare formulations, or reproduce a numerical study from explicit inputs and evidence. Typical starting points include solid stress/displacement problems, beam and shell response, laminate studies, vibration, and selected nonlinear static cases. The rotating-modal and Campbell routes add a separate experimental use case; they do not replace the solver's existing structural capabilities.

The same model workflow is available from Python and the command line. Mesh checks, assembly audits, solver diagnostics and result exports are part of the tool, rather than just benchmark infrastructure. The code is intended to be read and extended, but internal modules are not all stable public API.

For model selection, read when to use QF Solver and the scope of the specific route you intend to use.

Capabilities

The capabilities below are cumulative, but maturity belongs to a specific element, analysis, material, configuration and evidence record, not to the solver as a whole. Use the capability index and element map before relying on a particular combination.

Domain	Current capabilities	Evidence status and boundary
Elements	TET4, TET10, HEX8, HEX20, WEDGE6, BEAM2, MITC3 and MITC4 routes	Family support and maturity vary by analysis and formulation; see the element-analysis records.
Materials and composites	Isotropic elasticity, oriented orthotropic solids, layered shells and first-ply indicators	Material orientations, element combinations and laminate analyses have separate scopes; failure indicators are experimental, not progressive damage.
Model composition	Discrete masses/inertias, linear springs, MPC, RBE2/RBE3 and selected conforming mixed-solid workflows	Individual entities do not establish qualification for arbitrary assemblies or interfaces.
Linear statics	Structural solves, reactions and selected stress/result fields	Bounded element, material, constraint and load combinations; not every combination is qualified.
Modal and dynamics	Classical modal analysis, Newmark transient response and harmonic response	Route-specific mass, damping, timestep, element-family and mixed-model limits apply.
Stability	Linear buckling	Does not establish postbuckling or general nonlinear stability behavior.
Material nonlinearity	Selected small-strain J2 plasticity routes	Bounded material/element evidence; this is not general finite-strain plasticity.
Geometric nonlinearity	Selected Total-Lagrangian and corotational routes	Formulation and element scopes are narrow; corotational J2 assumes small local strains.
Contact and constraints	RBE2-style constraints, frictionless contact, and bounded frictional stick/slip evidence	Contact search, sliding, mesh sensitivity and current-source requalification remain limited.
Nonlinear solution	Newton methods, line search, adaptive increments, cutback/retry, and selected continuation/arc-length routes	Shared infrastructure does not imply convergence or qualification for arbitrary models.
Larger models	Structured-TET4 workflows and selected PETSc/MPI linear-static cases	Optional dependencies and recorded configurations only; no general HPC or nonlinear distributed claim.
Rotating dynamics	Experimental disk-gyroscopic rotating_modal and Campbell analysis in 0.2.11	Linear BEAM2 shaft with centered rigid axisymmetric disks; serial dense QEP and explicit scope limits.
I/O and interfaces	JSON models, public qf_solver Python namespace, CLI, Gmsh mesh import, bounded .inp import, JSON/CSV/VTU exports and optional mixed HDF5 results	Import/export formats and optional backends are not general-purpose interchange guarantees.
Inspection	Mesh/DOF checks, white-box audits, residuals, equilibrium checks and evidence bundles	Diagnostics describe the selected route and verification profile; they do not certify a model.
Verification and validation	Analytical checks, regression tests, controlled benchmarks, reproducible evidence and selected Code_Aster correlations	Numerical verification, external solver correlation and physical validation are distinct claims.

Status terms such as QUALIFIED_BOUNDED, EXPERIMENTAL_BOUNDED, EXPERIMENTAL and NOT_VALIDATED refer only to the scope documented by the controlling record. The published 0.2.8 registry remains the authority for its historical element-analysis decisions; later evidence does not silently rewrite those decisions.

Elements, materials and modeling

Element families

Family	Modeling role	Important boundary
TET4 / TET10	Four-node and ten-node tetrahedral solids	Linear and selected material-nonlinear routes; higher-order support does not imply every geometric/contact coupling is available.
HEX8 / HEX20	Eight-node and twenty-node hexahedral solids	Formulation, integration and nonlinear scope are route-specific. HEX8-SRI is a separate experimental capability, not a general HEX8R/B-bar/hourglass-control claim.
WEDGE6	Six-node prism solid	Recorded bounded static and homogeneous consistent-mass modal scopes; these do not transfer automatically to other analyses.
BEAM2	Two-node structural beams with axial, bending, shear and torsional response	Selected straight-beam static and dynamic evidence. The disk-gyroscopic route imposes additional straight-shaft/circular-section restrictions.
MITC3 / MITC4	Triangular and quadrilateral shell routes for membrane, bending and transverse shear response	Isotropic and selected laminate workflows; planar, curved, modal and transient scopes must be checked separately.
Discrete entities	Concentrated mass and rotary inertia, linear springs and rigid disks	These are distinct from finite elements; their compatibility and mass ownership must be checked for the chosen analysis.

WEDGE15 is unsupported. PYRAMID5 remains internal/research-only and is not a supported public element. See the element map for formulations and evidence rather than assuming a Cartesian product of all elements, materials and analyses.

Elastic materials and composites

• Isotropic elasticity: solid, beam and shell material definitions, with density and section/thickness data where the analysis requires them.
• Oriented orthotropic solids: TET4/TET10 routes with explicit material axes and nine-constant 3D elasticity. Homogenized composite definitions retain material provenance; selected cylindrical-tangent orientations are also implemented. This is not a general arbitrary fiber-field model.
• Laminated shells: orthotropic plies, ply angles and thicknesses, classical laminate A/B/D matrices, transverse shear terms, resultants and ply-level stresses. Selected MITC4 static scopes have bounded internal engineering acceptance; MITC3 and dynamic/curved extensions have their own development and review records, not blanket laminate qualification.
• First-ply indicators: maximum stress, maximum strain, Tsai-Hill and Tsai-Wu indices and reserve factors where material allowables are supplied. These indicators remain experimental. They do not degrade stiffness or simulate progressive failure, delamination or cohesive interfaces.

Implementation, runtime maturity labels and historical bounded acceptance are separate. In particular, composite output is not a certification or a general composite-design allowable. Read the composite overview, orthotropic solid specification and first-ply criteria before interpreting those results. Plasticity is described under nonlinear mechanics.

Loads, constraints and mixed models

The model can define nodal forces and moments on available DOFs, gravity, body forces, pressure and surface traction, beam line loads and shell edge tractions. Linear dynamic routes include documented time-dependent/tabulated and harmonic loading. Load integration and local/global-axis conventions are family-specific; input availability does not establish follower-load or nonlinear-dynamics support.

Fixed DOFs, linear multi-point constraints, RBE2 rigid relations, weighted RBE3 relations, springs to ground or between nodes, and concentrated mass/inertia definitions are available within their route contracts. Named DOFs and local frames matter, particularly where shells/beams introduce rotations. RBE3 availability is not a general connection qualification, and constraint/discrete entities can cause experimental runtime classifications.

Separate bounded records cover connected, conforming, shared-node TET4/WEDGE6/HEX8 static and modal workflows, plus selected translational-MPC and multi-material cases. The recorded mixed Newmark/harmonic workflows remain EXPERIMENTAL_BOUNDED and serial. These records do not establish arbitrary mixed meshes, hanging-node/nonconforming interfaces, or distributed mixed solving. The 0.2.8 workflow summary links the separate mixed-model decisions.

How QF Solver works

The public API and CLI validate a model, route it to an analysis-specific driver, assemble the required contributions and return results with route-specific diagnostics. Not every analysis assembles or uses every contribution shown here.

Model / input
     ↓
Public qf_solver API or CLI
     ↓
AnalysisRouter
     ↓
Analysis driver
     ↓
Assembly: elements · materials · loads · constraints · contact · rotating terms
     ↓
Linear / nonlinear / eigenvalue solver
     ↓
Results · diagnostics · provenance

Linear, dynamic, nonlinear and rotating analyses have distinct contracts and solver routes. SciPy provides the standard numerical stack; selected large linear workflows can use optional PETSc/MPI components. Optional packages do not expand the validated scope by themselves. See the architecture and API stability pages for implementation boundaries and compatibility status.

Numerical methods and backends

An analysis type, a numerical method and an assembly contribution are different choices. Combining element contributions into K and M does not select a solver algorithm; adding disk gyroscopy supplies a separate G only for the rotating route.

Problem	Available method family	Selection boundary
Sparse linear system	Direct sparse solve, CG, MINRES, GMRES and BiCGSTAB; explicit or auto selection	Matrix symmetry/definiteness, residuals, preconditioner compatibility and resource estimates matter. Available SciPy preconditioners include Jacobi and ILU where compatible.
Classical modal	Generalized real symmetric eigh, sparse eigsh/Lanczos and LOBPCG routes	Mass formulation, mode count and reduction contract matter; dense conversion is bounded. These are not gyroscopic QEP methods.
Linear transient / harmonic	Newmark time integration and direct complex frequency-domain solves	Timestep, frequency sampling, damping and load definitions belong to the model contract.
Nonlinear static	Newton, modified Newton, line search and specialized continuation	Method availability and state handling remain route-dependent.
Rotating modal	Dense generalized QEP with scipy.linalg.eig(A, B)	Complex spectrum and original-QEP residual checks; serial experimental scope only.
Campbell	Complex-MAC global assignment over single-speed rotating solves	Tracking is separate from the eigensolver and retains unresolved ambiguity.

Use qf-solver methods or the public list_methods() function to inspect analysis-method choices. Residual checks and diagnostics remain necessary even when an algorithm reports convergence. Optional PETSc/MPI and SLEPc integration does not imply that every analysis accepts those backends; see solver/backend notes.

Installation

Python 3.10 or newer is required. The standard installation uses NumPy, SciPy and Matplotlib and does not require Gmsh, HDF5, MPI, PETSc or SLEPc. The recorded CI matrix covers Linux and Windows with Python 3.10 and 3.13; it is not a claim that every Python/OS/backend combination has been tested.

python -m pip install "qf-solver==0.2.11"
qf-solver --version

Optional extra	Purpose	Boundary
mesh	Gmsh mesh tooling	Requires its runtime and a supported import/setup contract.
hdf5	Family-aware mixed HDF5 result storage	Opt-in, provisional storage; not a general restart or parallel-HDF5 contract.
large	HDF5 plus MPI/PETSc bindings for large-model workflows	Native runtimes and the recorded workload scope are still required.
hpc	MPI/PETSc/SLEPc bindings	Not general distributed or rotating-QEP support.
docs	MkDocs and controlled documentation tooling	For building documentation, not running standard analyses.

For example, python -m pip install "qf-solver[hdf5]==0.2.11" adds the HDF5 dependency. Native MPI/PETSc/SLEPc environments are platform-dependent; installing an extra does not make those runtimes universally available. Missing optional runtimes are reported rather than treated as successful backend evidence. See the installation guide for environment setup.

Quick start

The current release is 0.2.11. With the maintained example inputs available from a matching source checkout, run a small linear static case:

qf-solver check-mesh --input examples/tet4_static.json
qf-solver solve --input examples/tet4_static.json --output results/tet4.json

The equivalent public Python workflow is:

from qf_solver import check_mesh, load_model, save_result, solve_model

model = load_model("examples/tet4_static.json")
report = check_mesh(model)
if report.status == "FAIL":
    raise RuntimeError(report.errors)
result = solve_model(model)
save_result(result, "results/tet4.json")

Inspect the result fields, residuals, reactions and route-specific diagnostics before relying on an output. The first-calculation guide explains the workflow. New integrations should use qf_solver; solveur and the legacy launchers remain 0.2.x compatibility paths.

Choose a maintained starting case

Interest	Example input or guide
Solid statics and loads	tet4_static.json, tet4_pressure.json, tet4_body_force.json
Beam or shell response	beam2_cantilever.json, mitc4_shell_static.json
Orthotropy or laminates	tet4_orthotropic_static.json, mitc4_laminate_static.json
Vibration and dynamics	tet4_modal_unit.json, tet4_dynamic_tabulated_load.json, tet4_harmonic_response.json
Material/geometric nonlinearity	tet4_elastoplastic_static.json, small nonlinear tutorial
Connections and contact	rbe2_rigid_arm.json, frictionless_contact_surface.json
Disk gyroscopy and speed sweeps	Rotating-modal guide and Campbell guide; see their structured verification fixtures rather than assuming a standalone example JSON is shipped.

These inputs illustrate a workflow, not a qualification for all similar models. The nonlinear tutorial's engineering-profile WARNING is an expected scope warning, not something to hide after numerical convergence.

Analysis families

Analysis	Orientation	Documentation
Linear static	Small-strain structural equilibrium and reactions	Analyses · elements
Modal	Classical structural eigenvalue analysis	Analyses · capability index
Transient and harmonic	Newmark time integration and linear frequency response	Analyses
Buckling	Linearized eigenvalue buckling	Analyses · limitations
Nonlinear mechanics	Material, geometric, continuation and contact routes	Nonlinear overview
Rotating modal	Experimental disk-gyroscopic eigenanalysis	Model and limits
Campbell	Experimental speed sweep and explicit modal tracking	Tracking and limits

The examples index groups maintained inputs by the mechanical problem they illustrate. It also identifies controlled verification fixtures and intentionally invalid inputs.

Nonlinear mechanics

Nonlinear statics uses residual/tangent assembly, Newton iteration, convergence diagnostics and trial/accepted-state transactions. Selected routes share a composite assembly interface for material, geometric and contact contributions. This common core is an architectural foundation, not a qualification of every possible combination.

Route	What is available	Evidence and limit
Small-strain J2	Von Mises radial-return plasticity with linear isotropic hardening and integration-point history	Bounded decisions for recorded TET4/TET10/HEX8/HEX20 element-analysis combinations; no general finite-strain plasticity.
Total-Lagrangian geometry	Selected St. Venant–Kirchhoff static formulations	Selected serial TET4/HEX8 audit records GO_WITH_LIMITATIONS, without maturity promotion; not a general high-order/contact/dynamic route.
Corotational J2	Large rotations with small local material strains	Formal bounded acceptance is the HEX8 scope; not multiplicative finite-strain plasticity or blanket higher-order acceptance.
Material + geometry	Selected coupled static workflows across TET4/TET10/HEX8/HEX20	Owner-accepted bounded evidence, not general frictional, dynamic or distributed coupling.
Normal contact	Penalty node-to-triangle contact and selected contact-state/recovery paths	Frictionless capability remains experimental and bounded, with small-sliding/search limits.
Tangential contact	Selected frictional stick/slip and recovery evidence	Narrow serial prior acceptance; mesh sensitivity and missing current-source formal requalification remain visible.

Path-dependent history is evaluated in a trial state, then committed only after acceptance. Rejected increments can roll back before a cutback or retry. Selected versioned checkpoint/restart routes and state digests support replay; they do not imply universal restart for contact, distributed or nonlinear transient models.

Full Newton, modified Newton, line search, stagnation classification, numerical-residual-floor handling and adaptive increments are available where their route supports them. Specialized arc-length evidence remains experimental and bounded; it does not establish general bifurcation or postbuckling tracking. A numerically converged increment is not by itself a physically valid or qualified result.

Read the mechanics overview for how the core works and the 0.2.10 V&V summary for the separate material, geometric, coupled and contact evidence.

Rotating dynamics

QF Solver 0.2.11 provides experimental serial gyroscopic modal analysis for linear BEAM2 structures carrying centered rigid axisymmetric disks, plus experimental Campbell diagrams using explicit complex-mode tracking. This is a bounded extension to the solver, not a general rotating-machinery solver.

The initial model is a straight circular-isotropic BEAM2 shaft with centered rigid axisymmetric disks, constant prescribed signed spin, small perturbations, zero damping and no centrifugal prestress. The route uses a dense serial quadratic eigenvalue solve. Distributed shaft gyroscopy, speed-dependent bearings, centrifugal stress stiffening, unbalance response, rotor/stator contact, nonlinear rotors, PETSc/SLEPc and MPI are outside this scope.

At 100 rad/s, the high-frequency pair in the bounded Campbell verification remains ambiguous. The tracker preserves that gap instead of forcing branch continuity. GYRO-06 provides internal mesh-convergence evidence, not independent physical validation. Campbell frequencies are not a forced response, amplitude, operational-danger or validated critical-speed prediction. Both rotating_modal and Campbell remain EXPERIMENTAL.

The single-speed equation is M q̈ + Ω G q̇ + K q = 0, with signed Ω in rad/s and a fixed global axis. G is the unit-speed disk contribution; speed is applied once. The result retains complex eigenvalues and modes, growth rates, mass normalization, original-QEP residuals and the raw spectrum. It does not silently convert complex modes to real ones.

Campbell supplies an explicit ordered speed sweep over the same model, mass-weighted complex MAC, global one-to-one association, degenerate-subspace handling, branch lineage and polarization where meaningful. Its plotting layer cannot decide or repair tracks. The optional 1× line denotes |Ω|/(2π); a crossing is only a frequency coincidence within this linear modal model, not a response-amplitude or operational critical-speed result.

The dense backend was characterized up to 1,000 physical DOFs using synthetic matrix pencils in a recorded environment. That measured bound is not a practical-size or timing guarantee for every rotor; cost grows steeply and larger domains were not characterized. See the WP05 measurements, rotating-modal contract and Campbell contract.

Results, inspection and export

Outputs are analysis-specific. Depending on the route, they include displacements and rotations, reactions, stresses/strains and invariants, modal frequencies and shapes, transient histories, harmonic complex response, nonlinear material states, or rotating/Campbell complex spectra and tracks. Do not assume that a field available in one result is available in every other analysis.

For solids, recovery can include integration-point, element and nodal fields, principal values and von Mises stress. For shells/laminates, outputs include membrane, bending and shear resultants, face/section values and ply stresses in the declared axes. Interface ply stresses are not artificially averaged across a discontinuity. Nodal averaging is not a superconvergent stress recovery or a global error estimator; see result conventions.

Tool	Purpose	Scope
check_mesh / qf-solver check-mesh	Geometry, connectivity and boundary-condition checks	Inspect PASS, WARNING and FAIL, not just whether parsing succeeded.
inspect_model / qf-solver inspect	White-box model, DOF, matrix and consistency audit	summary, diagnostic and values detail levels; full-value dumps can be large.
assess_result	Non-raising run/qualification summary	Numerical status and verification-profile acceptance are separate.
save_result	Structured JSON results	Available fields and complex serialization follow the result contract.
save_result_csv / save_result_vtu	Result tables and visualization fields	Export support is result/family-specific, not lossless export of every analysis; VTU can be viewed in a compatible visualization tool.
save_audit_markdown	Readable model/result audit	Compact reports retain warnings/failures; detail level does not change the solve.
save_evidence / verify_evidence	Evidence bundle and fingerprint verification	File integrity does not automatically establish scientific qualification.
Mixed HDF5 APIs	Family-aware storage and selective result reads	Opt-in/provisional; no general parallel-HDF5 or restart claim.

For a supported static example, a CLI export workflow is:

qf-solver solve --input examples/tet4_static.json --output results/tet4.json \
  --csv-dir results/tet4_csv --vtu results/tet4.vtu --audit-md results/tet4_audit.md

Use audit detail modes and the public API contract to choose the outputs you need. The quick, engineering, strict and qualification profiles are acceptance/evidence policies, not different physical formulations. Tightening a profile does not promote maturity.

Input and mesh interchange

Strict JSON is the standard model interchange. It defines nodes, elements, materials, constraints, loads, analysis settings, units and the verification profile. Mechanical quantities must be dimensionally consistent; declaring units does not mean arbitrary quantities are converted automatically.

Gmsh MSH 4.1 import uses physical groups and a companion setup to assign materials, boundary conditions and loads for supported cell families. The mesh alone is not a complete analysis definition. Coordinate scaling and orientation repair, where supported, must be explicit and reported. The Gmsh guide describes the workflow; later mixed-family evidence has separate contracts. The provisional .inp reader covers a bounded Abaqus/CalculiX subset, not general deck or solver compatibility. Large-model HDF5/NPZ storage is a separate workflow from standard JSON inputs and mixed HDF5 result storage.

Larger models and performance context

Large-model tooling includes structured TET4 generation, large-model inspection/readiness, chunked assembly or matrix-free paths, benchmark reports, selected PETSc preconditioner comparisons and chunked postprocessing. These are specialized workflows, not a promise that every model supported by the small serial API can run through them.

Historical evidence records structured-TET4 workloads at approximately 1.029M, 3M, 5.01264M and 10.125M DOFs. Their source, solver settings, rank count, hardware and acceptance limits are part of the evidence. They do not establish a hardware-independent speed or memory guarantee, nor general strong/weak scaling. The historical large-model summary and bounded limitations provide the workload context.

Separate two-rank PETSc/MPI evidence covers selected linear-static cases with replicated input and root-side assembly. That is not distributed assembly or general nonlinear/dynamic/contact MPI support. Generic mixed distributed runtime remains NOT_VALIDATED. No GPU claim is made, and the rotating QEP is dense serial rather than part of these large-model routes.

Verification and maturity

QF Solver distinguishes implementation, a passing test, numerical verification, external solver correlation, maturity decisions and physical validation. These are separate kinds of evidence. The V&V and maturity guide explains the labels; the 0.2.11 verification summary links the gyro and Campbell records.

Selected Code_Aster 18.1 results provide same-mesh linear-static solver correlation for declared cases and comparable observables. This is not experimental physical validation or general nonlinear correlation. Internal independent recomputation of selected observables is also not a second global FEM/Newton solve. Large-model and PETSc/MPI evidence is tied to specific workloads and environments; it does not establish general scalability or distributed nonlinear support.

Verification covers analytical element/material identities, equilibrium and energy checks, regression cases, mesh/time-grid convergence within declared studies, solver residuals and selected external comparisons. Historical composite/solid studies also contain bounded CalculiX correlations where the formulations and observables are comparable; these must not be generalized to every family or route.

New execution records bind scientific inputs, referenced file bytes, resolved configuration, oracle, tolerance policy, code and relevant environment to an execution identity. Safe resume checks identity and artifact integrity; structured expected failures distinguish an intended rejection from an unrelated exception. Evidence availability remains separate from a numerical PASS: some historical payloads are local-only, reconstructed, optional or missing. Historical schemas and decisions are preserved rather than rewritten by newer tooling.

For method-development work, the public surface also exposes benchmark and demonstration catalogs, evidence verification and controlled V&V/campaign tools. These advanced interfaces are generally provisional. See benchmarks, demonstration API and the public export inventory for their contracts. A benchmark PASS is never an automatic qualification decision.

Limitations

• General multiplicative finite-strain plasticity, nonlinear transient dynamics, universal postbuckling, finite-sliding/self-contact and impact/contact dynamics are outside the demonstrated scope.
• Corotational J2 permits large rotations only under a small-local-strain assumption. Higher-order nonlinear routes are narrower than linear element support.
• Frictional contact remains bounded and mesh-sensitive; general updated search, finite sliding and self-contact are not claimed.
• Optional PETSc/MPI workflows do not establish general HPC scaling, nonlinear distributed solving or a validated mixed-distributed runtime.
• Composite first-ply indicators are experimental, with no progressive damage/delamination model or universal design-validation claim. Laminate and orthotropic acceptance applies only to the recorded scopes.
• MITC4 modal historical issues, discrete Newmark replays and experimental mixed dynamics retain their separate records; a later retest does not automatically promote a family. HEX8-SRI remains separately bounded; WEDGE15 is unsupported and PYRAMID5 is internal/research-only.
• Importers, result recovery and export schemas have explicit subsets; neither a valid input nor a readable visualization proves model accuracy.
• Rotating modal and Campbell are experimental and restricted to the model described above; the 100 rad/s high-frequency branch ambiguity remains visible.
• QF Solver is not certified and has no universal physical-validation claim.

See the detailed technical limitations and the controlling evidence record for the chosen analysis.

Current release and citation

The current QF Solver release is 0.2.11, published on 2026-10-07. Cite this version with DOI 10.5281/zenodo.23214487. The preceding 0.2.10 release has version DOI 10.5281/zenodo.23106744. Use the project concept DOI 10.5281/zenodo.22697897 to cite QF Solver across versions. CITATION.cff provides machine-readable citation metadata. The preceding release used source tag v0.2.10. The 0.2.11 selected wheel, sdist and checksum manifest are the audited distribution artifacts. GitHub's automatically generated source archives are not part of that selected distribution. The whole-repository G03 archive gate remains FAIL, the full repository archive is not cleared, and WP14 remains HOLD_NOT_PROMOTED. Selected distribution checks do not change those states.

Documentation and project

• Documentation home
• Capabilities, elements and analyses
• Public API contract and API stability
• Benchmarks and V&V evidence
• Composite mechanics, nonlinear mechanics and solver/backend notes
• Contributing · code license · documentation/examples license

The solver code is licensed under Apache-2.0; documentation and original examples are under CC BY 4.0. Third-party terms are listed in THIRD_PARTY_LICENSES.md.