ANYsolver finite-element solver

anysolver is a headless structural FE solver for beam, shell, stiffened-panel, and cylindrical-shell analysis. It is an engineering solver with an explicit qualification scope, not a general-purpose CAD, contact, or fracture platform.

Install the released package with:

python -m pip install ANYsolver

For coordinated local development with ANYstructure:

python -m pip install -e C:\Github\ANYsolver

Core analyses and the solver-owned generated-geometry workflow are available directly from anysolver; the lightweight normalized flat-panel and cylinder facade is exposed through anysolver.runtime:

from anysolver import (
    FEModel,
    GeneratedGeometryFEMConfig,
    LoadCase,
    run_generated_geometry_fem,
    solve_linear,
)
from anysolver.runtime import (
    LightweightFEMConfig,
    resolve_runtime_analysis,
    run_production_fem,
)

The runtime facade applies its normalized axial force, bending moment, shear force, torsional moment, and pressure inputs to the generated model. Set LightweightFEMConfig.follower_pressure=True only for nonlinear static or arc-length analysis using the static only or nonlinear static runtime path; incompatible linear, stepwise eigenvalue-buckling, transient, collision, and structured-capacity paths return an explicit invalid_follower_pressure status. Arc length retains the requested von Karman or corotational kinematics. Production runtime failures remain failures rather than being replaced by an estimator.

Application integrations should use resolve_runtime_analysis(config) to reflect the solver's effective nonlinear/material/control/kinematics choices; the normalization helpers with leading underscores are implementation details.

The generic GeneratedGeometryFEM* names are the preferred workflow API. Historical AnyStructureFEM* aliases remain available only for downstream compatibility. Material selection is centralized in dnv_c208_steel_properties() and dnv_c208_steel_curve(); the runtime facade uses the same canonical table and validation.

The source code and tests are authoritative. Generated reports under reports/ are dated evidence snapshots and must be regenerated after solver changes before making release claims.

Functional overview

Implemented does not automatically mean qualified for every geometry or load regime. The live capability matrix is produced by write_production_readiness_artifacts() and the verification manifest.

Production scope

The qualified target is thin flat or cylindrical shell structure, with beam stiffeners/girders represented through the documented coupling, inside the verified mesh, material, distortion, eccentricity, and load ranges.

Important limits:

• no arbitrary CAD topology or automatic general-purpose meshing;
• Q8R is experimental: its hourglass stabilization is not qualified for thin bending and it is deliberately excluded from nonlinear batch acceleration;
• the 3-node quadratic beam is straight-sided; curved members must be represented by straight beam segments until a true curved formulation is implemented;
• the shell initial-stress operator covers the in-plane N, M, and H stress moments acting on Mindlin midsurface translations and director gradients; it does not add drilling, transverse-normal-stress, or a geometrically exact finite-rotation shell/director formulation;
• current-area follower pressure is supported in nonlinear static and arc-length analyses, with its exact load tangent. Linear/dead pressure remains the default. Linear buckling rejects a constrained nonsymmetric follower-pressure pencil because general complex nonconservative eigenanalysis is not implemented;
• no general shell-shell, body-body, frictional, rolling, or self-contact;
• no fluid-structure interaction, cavitation, or water-entry model;
• no cohesive cracks, remeshing, material separation, or fracture-mechanics claim—the erosion models are engineering screens;
• no unrestricted deep post-buckling or automatic bifurcation branch switching;
• the consistent corotational tangent includes frame derivatives and is selected automatically for follower pressure, but its numerical frame-sensitivity evaluation is costlier than the rotated tangent used by default for ordinary corotational solves;
• material-history-aware recovery requires retained, matching committed nonlinear layer/fiber states; missing or invalid state is reported and the affected components fall back explicitly to elastic reconstruction;
• with an active plastic constitutive history, von_mises covers the return-mapped shell in-plane or beam-fiber stress components. Transverse shell shear and beam shear/torsion remain elastic reconstructions and are exposed separately through mixed_reconstruction_von_mises, rather than being presented as a hardening-curve-consistent equivalent stress. A purely elastic nonlinear state keeps the full mixed elastic value as its primary equivalent stress;
• the guarded patch-recovery fit is qualified only for locally planar, consistently oriented, homogeneous, full-integration Q4 or Q8 shell neighborhoods. Discontinuities remain separate, and the optional normalized stress-L2 discrepancy is a diagnostic—not an energy-norm error estimate;
• initial stress/prestrain fields are qualified only in element-local reference coordinates with kinematics="von_karman". Shell fields use the documented membrane/positive-face-bending convention, and beam fields require a configured fiber section. Input stress must be admissible for the supplied hardening state; equilibration may redistribute it and does not reconstruct the manufacturing history. A field-bearing restart also requires its matching converged displacement vector;
• the analytical plane-stress tangent is branch-consistent. The numerical derivative remains an oracle and automatic invalid-row fallback; local yield-residual nonconvergence fails closed;
• no unverified material laws or distortion ranges;
• SESAM FEModel export remains outside the supported interchange gate;
• CalculiX comparison requires a compatible local executable and an explicit execution request. Deck generation alone is a reproducibility handoff, not external numerical evidence.

Use validate_production_model(), the analysis-specific preflight checks, and the generated production-scope artifacts before production use.

Documentation map

• ARCHITECTURE.md: package boundaries, analysis flow, and invariants.
• THEORY.md: implemented formulations and validity limits.
• ARC_LENGTH.md: continuation controls and use.
• NONLINEAR_PERFORMANCE.md: nonlinear assembly, sparse backend, threading, and diagnostics.
• QUALITY_CONTROL.md: verification commands, evidence hierarchy, and current checked status.
• reference_cases/README.md: local CalculiX/PrePoMax reference-case layout.
• MIGRATION.md: source provenance, inclusion boundary, and import changes.

Basic verification

From the repository root:

python -m pytest tests -q -p no:cacheprovider
python run_qc.py --no-save
python scripts/run_fe_verification.py

The last command regenerates the canonical JSON and Markdown evidence under reports/verification/. Its default external-reference mode generates handoff decks with status not_executed; that status is not a pass or a claim of numerical agreement. To execute the comparisons, provide a compatible CalculiX executable on PATH, through ANYSOLVER_CALCULIX_EXECUTABLE, or with --calculix, and run:

python scripts/run_fe_verification.py --execute-calculix

An external case becomes passing evidence only after isolated execution, successful FRD/DAT parsing, and every declared comparison meeting its tolerance.