Architecture¶
This document describes the modules, data model, conventions and test strategy of morphoGen 1.0, for those who change its code; the configuration files, the commands and the files of a run are in the configuration reference. The requirements are in specification.md and the formulas in theory.md.
1. Overview¶
┌──────────── config (YAML/JSON/TOML) ─── cli ─── pipeline ──────────────┐
│ │
Model ─ generate(domain, seed) ─> Microstructure ─ voxelize ──> VoxelImage ─┬─> analysis
(parameters, theory) (vector geometry) ├─> mesh (hexahedra, STL)
│ ├─> homogenization (FFT)
├ level_set ─> conforming_mesh ─> TetMesh ─> homogenization (FE)
│ (MMG) ├─> io (files, solvers)
└ cell_complex ─> polycrystal_mesh ──┘ └─> rve (statistics)
(exact cells) (MMG, by grain)
parallel: realizations in processes, tasks, MPI ranks; records, merge; cluster job scripts
A model holds the parameters of a random set or of a deterministic architecture, together with its closed-form characteristics (
model.theory(domain)).model.generate(domain, seed)returns a microstructure: the explicit (“vector”) description of one realization (sphere centres and radii, fibre axes, germs and weights, Fourier modes, surface equations). Nothing is discretized at that point.A microstructure can be voxelized at any resolution (
voxelize), turned into a level-set field (level_set), queried point-wise (contains) and serialized (to_dict/ JSON).Images feed the measurements (
analysis), the voxel meshes (mesh), the FFT solver (homogenization.fft), the files (io) and the RVE statistics (rve).Conforming tetrahedral meshes (
mesh.conforming, MMG) follow the interfaces of the implicit morphologies, and polycrystals are meshed grain by grain from their exact cells (mesh.polycrystal); the finite element solver (homogenization.fe) works on them, with isotropic phases or crystals turned by the orientations of the grains. Exact meshes (mesh.exact,method: exact, the default throughauto) need no background grid: faceted morphologies are meshed from their exact convex pieces, the others from their exact point-wise level set (level_set_at) on an adaptive octree (mesh.octree), their interface nodes moved onto it. Voxels are reserved for the FFT solver and for image measurements.pipeline runs configurations: batches of realizations and RVE studies; parallel runs their realizations in processes, in tasks of cluster job arrays or under MPI, and merges their records.
Combinations (
combination) build models from models: set operations, inclusions in a matrix; tessellations can have a grain-boundary phase.Automata and growth processes (
automata) produce lattice microstructures (LatticeField) through the modelscellular_automaton,activator_inhibitor,reaction_diffusion,nucleation_growthandaggregation, which fit the same pipeline (voxels, level sets, conforming meshes, homogenization, RVE studies). On other lattices (lattice: triangles, hexagons, Penrose rhombi, Ammann–Beenker tiles, prisms, BCC and FCC Voronoi cells, golden rhombohedra), the automata and reaction–diffusion giveTilingFields, rasterized cell by cell; the quasi-periodic tilings fill periodic boxes through their periodic approximants (approximant).Combinatorial design (
design) assembles tiles (tile_assembly), builds strut frameworks with defects and analyses their rigidity (random_lattice), and searches design spaces (annealing, genetic algorithm, inverse design of automaton rules).Topology optimization (
optimization) designs periodic cells on voxels with the FFT solver; designs become lattice microstructures.Visualization (
viz) draws figures and animations; the interface (gui) serves a page in the browser that drives all of the above and draws its morphologies and simulations in 3D (WebGL).Nested models take models as parameters (Cox spheres in aggregates, coatings, nested combinations); hierarchical and fractal architectures and bio-inspired models (nacre, Bouligand plywoods, honeycombs and wood, osteons, open foams and trabecular bone, branching networks) are models like the others.
Homogenization solves loadings that mix macroscopic strains and stresses (the periodic element), with thermal strains, and follows loading paths with nonlinear behaviours by finite elements or by FFT on voxels; post-processing gives engineering constants, anisotropy indices and normalized values; results keeps every run in a database and draws normalized Ashby charts.
2. Package layout¶
src/morphogen/
__init__.py public API (Domain, Grid, VoxelImage, models, analysis, io, mesh, rve)
domain.py Domain (box, periodicity, unit), Grid (voxel grid conventions)
image.py VoxelImage (array + spacing + origin + periodicity + metadata)
seeding.py reproducible seeds (SeedSequence derivation)
distributions.py distributions of radii, lengths... (moments, quadrature, configs)
orientations.py fibre-axis distributions (isotropic, aligned, planar, Schladitz's
beta law, Gaussian girdle, von Mises-Fisher, mixtures), their
quadratures and the law of the angle to a direction, uniform crystal
orientations (Bunge)
texture.py crystallographic textures: the rotations of the Laue classes,
disorientations and Mackenzie's distribution, mean orientations,
poles and their projections, pole figures (density in multiples of
a random texture), inverse pole figures and their colours, the
disorientations of the boundaries of an image of grains, textures
drawn from measured orientations
symmetry.py finite symmetry groups of periodic cells, real irreps from the
regular representation, axial isotropy subgroups (EBL)
primitives.py Spheres, Segments (cylinders/spherocylinders), periodic images
theory.py shared closed-form results (covariograms of spheres, cylinders and
spherocylinders, covariances of oriented grains, Voronoi constants...)
settings.py number of threads; the BLAS on one thread while the compiled
kernels run on theirs (blas_on_one_thread)
_public.py the public interface (docs/stability.md): its modules and names, for
the API reference of the documentation site and its test
stopping.py cooperative stops of long computations (checkpoints, Stopped)
gpu.py optional GPU backend: the choice of the device (CuPy, NVIDIA GPUs)
kernels/ Numba kernels (array in, array out)
raster.py cell lists, union rasterization, signed distances
union.py Python front end of raster.py
rsa.py random sequential addition (spheres; spherocylinders: closest
points of segments)
implicit.py TPMS level-set functions
leaves.py first covering leaf (sphere or spheroid) of each voxel (dead
leaves); the field of transparent leaves
cells.py implicit function of a union of convex cells from the planes
of their faces (the cells of a tiling that have a phase)
collective.py stealthy hyperuniform point patterns (collective coordinates,
L-BFGS)
packing.py Lubachevsky-Stillinger: event-driven molecular dynamics of
growing hard spheres (dense packings)
force_biased.py force-biased packing of chains of spherocylinders (Altendorf &
Jeulin): repulsions, springs, open angular springs; contact
forces (Keilmann et al.); the initial placement of least overlap
(Altendorf, Jeulin & Willot)
annealing.py Yeong-Torquato reconstruction: swaps annealed to S2 and L
watershed.py marker-controlled watershed (Meyer's flooding) and the maxima
of a given prominence (union-find), of planes and of volumes
(26-connected voxels, periodic faces)
models/ morphological models and microstructure kinds
base.py Model, Microstructure, registries, parameter metadata
affine.py Anisotropy, AffineStructure: the affine anisotropy of the models
(the isotropic model in the preimage of the domain, mapped)
union.py PrimitiveUnion (unions of spheres and segments)
pieces.py the smooth pieces of the phases of structures of primitives:
overlaps (periodic images included), colourings, pieces of a
union, of its coatings and bores (for the meshes that follow
sharp edges)
implicit.py the implicit functions of the phases: bands, the points of a
grid near an object, objects that cover one another
(leaves, osteons)
boolean.py BooleanSpheres, BooleanCylinders
fibers.py PoissonFibers, HardcoreFibers (RSA of spherocylinders),
CurvedFibers (worm-like chains, arcs, helices; Boolean or
RSA of chains)
hardcore.py HardcoreSpheres (RSA or Lubachevsky-Stillinger),
PenetrableSpheres (the cherry-pit model)
tessellation.py Voronoi, Laguerre (Poisson or packed germs, grain-boundary
phase) -> Tessellation
gaussian.py GaussianExcursion -> GaussianField
tpms.py TPMS -> TPMSStructure
lattice.py StrutLattice (cubic, bcc, fcc, octet, kelvin, diamond,
pyrochlore; hollow, rotated or graded struts) -> PrimitiveUnion
laminate.py RandomLaminate -> Laminate (layers along an axis, holes)
deadleaves.py DeadLeavesModel -> DeadLeaves (leaves in covering order:
families, spheroids, background of a finite coverage; their
bivariate laws)
reconstruction.py Reconstruction -> LatticeField (Yeong-Torquato: Debye media,
models, images)
transparent.py TransparentDeadLeavesModel -> TransparentDeadLeaves (the random
function of transparent leaves, its excursion)
combination.py Combination -> Combined (set operations of components,
inclusions in a matrix)
lattice_field.py LatticeField: phases (labels, field) per cell of a lattice
ebsd.py EBSD -> LatticeField (a window of an orientation map: the grains
of the map, their phases and mean orientations)
tiling_field.py TilingField: a phase per cell of a tiling, rasterized
automata.py CellularAutomatonModel, ActivatorInhibitorModel,
ReactionDiffusionModel -> LatticeField (TilingField on
other lattices)
growth.py NucleationGrowthModel, AggregationModel -> LatticeField
design.py TileAssembly -> LatticeField, RandomLattice -> PrimitiveUnion
nested.py CoxSpheres -> ClusteredSpheres; Coating -> CoatedStructure
(coatings, shells, hollow members); FractalSponge ->
FractalStructure
hierarchical.py HierarchicalLattice: frames of generation G of (hollow) tubes
planar.py PlanarCells: Voronoi cells of the plane and distances to their
sides; hard-core germs of the plane
nacre.py Nacre -> NacreStructure (layers of tablets)
bouligand.py Bouligand -> BouligandStructure (twisted layers of fibres,
periodic fibre families)
honeycomb.py Honeycomb -> PrismaticCells (honeycombs; wood: RadialFiles,
WoodRays between the files)
osteons.py Osteons -> OsteonStructure (dead leaves of cylinders)
foam.py OpenFoam -> PrimitiveUnion (edges of periodic Voronoi cells,
Lloyd relaxation, anisotropy); DelaunayNetwork -> PrimitiveUnion
(fibres on the edges of periodic Delaunay triangulations of
Poisson or hard-core points, in space or in a plane; Ban et al.)
levelset_foam.py LevelSetFoam -> DistanceFieldFoam (level set of the distance
fields of packed ellipsoids: Plateau borders, nodes, walls);
calibrate_seeds: seeds fitted to measured section radii
branching.py BranchingNetwork -> PrimitiveUnion (space colonization,
Murray's law)
scale_rich.py ScaleRichNetwork -> ScaleRichStructure (ligaments of power-law
thicknesses grown until they meet, Both et al. 2026: convex
voids split by strips, exact partition; the dual graph, the
orientation tensor; the density and jamming in closed form);
with dimension 3 -> ScaleRichPlates or ScaleRichBeams
scale_rich_3d.py the growth in space: plates splitting convex polyhedral voids
(clipping, sections, volumes; closed cells, exact partition),
beams grown until they touch (segment distances); the first
plate area in closed form, the beams' mean field; fields of
the plates and beams in buckets (voxels, level sets)
automata/ cellular automata and lattice growth
lattice.py neighbourhoods, neighbour counts (Numba), FFT kernel sums,
sublattices
rules.py Life-like, majority, annealing, conserved, probabilistic,
activator-inhibitor rules
engine.py CellularAutomaton (updates, histories), random states
reaction.py Gray-Scott reaction-diffusion on the grid and on tilings
(finite-volume Laplacian, stable substeps)
tilings.py Tiling: triangles, hexagons, Penrose rhombi (pentagrid),
Ammann-Beenker tiles (four grids), prisms, BCC/FCC Voronoi cells,
golden rhombohedra (six grids of planes), their periodic
approximants (rational grids, approximant_period); CSR
neighbourhoods, face measures, point location
growth.py nucleation and growth (event-driven), Potts grain growth,
DLA / Eden / ballistic aggregation
design/ combinatorial design
tiles.py TileSet (labelled faces, rotations, geometry), polarized blocks,
pipe tiles
csp.py wave function collapse with backtracking, transfer matrices,
exact count of polarized blocks
frames.py periodic pin-jointed frameworks: lattices, dilution, rigidity
search.py simulated annealing, genetic algorithm, fit_rule
optimization/ topology optimization of periodic cells
cell.py graded voxel cells: homogenized tensors and sensitivities (FFT)
filters.py density filter, Heaviside projection, symmetry
mma.py method of moving asymptotes
topology.py CellOptimization (objectives; OC, MMA, BESO, HCA), CellDesign
(hinges, re-evaluation on a finer grid), count_hinges
run.py optimization configuration files and outputs
iga/ isogeometric analysis and shape optimization
basis.py B-spline functions and derivatives, knot vectors, the operators
of their refinement (knots inserted, degree raised)
geometry.py Patch (NURBS surfaces and solids) and Multipatch (patches that
share sides; the periods of a cell), their refinements and the
map design -> analysis, the Jacobians of the patches, their
image on a grid (for the voxel solvers)
shapes.py rectangles, rings, a plate and periodic cells with holes (exact
arcs; boxes and parallelograms), the sheet with alternating
holes (rotating squares), tilings, repetitions, extrusions
analysis.py Analysis: elasticity (plane strain, plane stress, solids) and
conduction; loads on sides, fields, error norms
homogenization.py Cell: periodic conditions on matching control points (the
sides of a box, or a period apart on any lattice), effective
stiffness and conductivity; plane engineering constants
optimization.py ShapeDesign (variables on control points, linked by the
symmetries of a pattern), analytical shape sensitivities,
optimize_cell and optimize_structure (SLSQP, MMA; linear
constraints on the variables)
learning/ behaviours learned from data
data.py random strain paths, the responses of a material point or of a
microstructure along them (Green-Lagrange measures in finite
strains), splits, scales
networks.py FeedForward, Potential (input-convex: hyperelastic), Recurrent
(gated recurrent unit, Numba backpropagation through time), their
gradients and tangents written out; in finite strains each of the
logarithmic strain 1/2 ln C (its stress T, S = J^T T; the default)
or of the Green-Lagrange strain (the option strain); .npz files
with JSON metadata (those without strain: laws of E)
logarithmic.py the logarithmic strain 1/2 ln C of Green-Lagrange strains, dH/dE
and the pull-back S = J^T T with its geometric tangent (Numba
kernels of the divided differences of homogenization.tensors),
shared by the models of ln C
standard.py Standard: a learned generalized standard material (convex free
energy and dissipation of learned internal variables), the
incremental variational update by Newton's method, its symmetric
tangent, trained through the minimizations by implicit
differentiation (Numba kernels, gradients written out); in finite
strains its free energy of the logarithmic strain 1/2 ln C
(S = J^T T, the kernels of logarithmic.py) or of the
Green-Lagrange strain
training.py Adam, mini-batches, early stopping, random rotations of the paths
(uniform, or the 24 of the cube for a crystal in its axes)
behaviour.py learned laws in the solvers (the seam of the MFront points),
P = F S in finite strains; turned by the orientation of their
grains (the axes of their material), the stress-free thermal
strain taken out (in finite strains as the Hencky points do,
1/2 ln C - alpha dT: a law of E sees its Green-Lagrange strain)
run.py learning configurations (morphogen learn)
mfront.py MFront sources of the learned laws (io.mfront.mfront_source,
morphogen mfront, the AMITEX and Abaqus exports): the trained
parameters as constants, the Default DSL in small strains or with
Hencky's (laws of 1/2 ln C) or Green-Lagrange's strain measure,
the state of the model as auxiliary state variables, the thermal
split of the Hencky points, orthotropic in the axes of the material
learned.cpp the C++ port of the four models for those sources (forward passes,
consistent tangents, the Newton solve of the standard model), a
section per kind, morphoGen's Voigt order inside
fracture/ fracture
cohesive.py cohesive zone elements: laws (bilinear mixed-mode, exponential,
the contact of lips), Coulomb's friction of their broken share
(radial return, slip and dissipation of each point), interface
elements in small and finite strains (three nodes a side, or six
between 10-node tetrahedra, on curved faces), insertion by
duplicated nodes, quasi-static solver (loadings in stages, a
predictor of lips that stick), calibration, DCB
periodic.py CohesiveCell (the periodic element of the loading paths with
cohesive interfaces between its phases or its grains: debonding,
contact and friction of the broken lips, the stress of the
interfaces on the faces of the cell), open_cell (the faces opened
on the torus of a periodic mesh, the periodic links of the two
lips), interface_law (a law from a mapping)
lefm.py cracks by level sets, conforming (remeshed by MMG, the front on
edges; 4-node elements, or 10-node ones with quarter points at
the front) and XFEM solutions, energy release rate, Paris and
Griffith propagation
mixed.py mixed-mode cracks (I, II, III) in small and finite strains: G-θ,
interaction integrals, contact and friction of the lips (the
work of their tractions, the bonded ligament ahead of the
front), kinked cracks
remeshed by MMG, kinking (Erdogan-Sih) and propagation; the
driving forces of inelastic bodies (work, free energy, tip)
surface.py cracks of any surface and front (SurfaceCrack: rows x stations,
closed or open fronts, level sets and axes of the front), grown
point by point of the front (propagate_front: Erdogan-Sih kinks
and Paris advances of each station)
bulk.py the behaviours of morphoGen's points as the bulk of the crack
solvers (von Mises, porous and crystal plasticity, MFront), their work and
free energies, the configurational forces of internal variables
ductile.py remeshed propagation in inelastic bodies: the transfer of the state
of the points between meshes (nearest or projected, the lips apart),
the equilibrium regained, ductile growth by J_tip or the porosity
at the tip, the resistance curve
phasefield.py the phase field of brittle fracture on voxel images (AT1, AT2):
alternate minimization with the FFT solver, irreversibility by a
primal-dual active set, energy splits, a toughness per phase;
its result a PathResult (damage, crack surface, energies)
phasefield_finite.py the same in finite strains (rubbers, soft solids): the
energy of the deformation gradient of hyperelastic or elastic
phases, the displacement by the nonlinear FFT solver whose
points are degraded, the damage problem of the small strains
(polycrystals: grain by grain, crack meshes with grains)
contact/ contact, self-contact and friction
surfaces.py Surface (free faces of a mesh, their points, rims and sizes), Pair
(two surfaces, or one against itself), Obstacle (rigid plane,
sphere, cylinder, cone that move and rub), assemble (bodies meshed
apart in one mesh)
kernels.py Numba kernels: the functions of the faces, the nearest point of a
flat or curved face (edges, slack, rim), the forces of a slave
point on a master face or on an obstacle (penalty, multiplier,
Coulomb's radial return, the slip of new contacts), the choice of
the master face, tangents by central differences
problem.py ContactProblem (the quasi-static solver of the cohesive zones with
the forces of the contacts: search, passes, predictor, augmented
Lagrangian, records), ContactResult
periodic.py ContactCell (the periodic element of the loading paths with the
contact of the walls of the pores of a cell: images through its
faces, the stress that the contacts carry, the macroscopic
tangent), walls (the free faces inside a cell)
hertz.py closed forms: Hertz's sphere and cylinder, Sneddon's cone, the
flat punch, the stick zone of Cattaneo and Mindlin, the sphere on
a layer of finite thickness (Dimitriadis et al.)
experiments.py ready-made tests: compression (a sample between two platens,
its walls free to close on themselves) and indentation (a
sphere or a cone; the modulus read from the force)
micromorphic/ micromorphic continua
space.py quadratic displacements and linear generalized degrees of freedom
on tetrahedra, node-wise sparse assembly, Newton solver (with a
predictor by the last tangent)
finite.py finite strains: logarithmic strains pulled back to the Piola-
Kirchhoff stresses, the objective measures χ⁻¹F and χ⁻¹ Grad χ
microfoam.py the microfoam model of nickel foams (Dillard, Forest & Ienny 2006):
anisotropic compressible plasticity, full microdeformation, small
and finite strains
damage.py micromorphic damage (implicit gradient), small and finite strains
ductile.py micromorphic ductile damage of von Mises elasto-(visco)plasticity,
driven by the plastic strain weighted by the triaxiality (Rice &
Tracey, Johnson & Cook, Lemaitre), small and finite strains
viz/ visualization
png.py colour maps, sections, PNG encoding (no dependency)
sections.py sections of the level sets on their planes (one cell thick,
up to 2048 x 2048 pixels without the volume), anti-aliased
rendering (of the implicit functions of all the phases of
the structures of several phases), measures of the level sets
figures.py matplotlib figures, animated GIFs; the beams of a frame in
its cell coloured by a value per beam (frame_figure: forces
about zero, energies on a log scale; rasterized beyond 2000
pieces in vector files), Ban's replicas against s and v;
textures: pole figures, inverse pole figures, orientation
maps in the colours of their crystals with their grain
boundaries and colour key, disorientations against Mackenzie
(softening_figure); the cut of a tetrahedral mesh by a plane,
its elements in the colours of their phases or of a field
(mesh_figure, mesh_section)
export.py figures as bitmaps, vector graphics and TikZ (pgfplots)
video.py videos (mp4 by ffmpeg, gif, PNG frames; once, loop, bounce):
sections, orthoslices, VTK turntables, sweeps, fields, the
steps of loading paths; colour bars
workflow.py figures and videos of the realizations of a run (the
visualization section): image, labels, mesh, fields, curves
surfaces.py triangle surfaces for the 3D view of the interface, in groups
between two regions (phases, cells): marching cubes of level sets
cut by the box, exact facets of polycrystals, faces of voxels,
faces of tetrahedral meshes with their fields (nodal, or averaged
per region at the nodes) and displacements; volumes for the faces
of the box and the sections; a budget of triangles; base64 JSON
gui/ web interface (index.html, viewer.js, api.py, server.py; desktop.py,
the stand-alone application: the page in a window of pywebview,
the dialogs of the system, icon.png, icon.ico): menus
in four sections (create, characterize, simulate, export &
compare), their tabs, and in a tab the menu of its tools under
the headings of what they do (of the models by family in
Generate); a search over the tabs, the tools, the fields and
options of their forms and the models, which opens what it
finds with the option set; addresses #tab/tool and
#generate/model; the workspace of the session (morphologies with their
traits, simulations, designs and the runs of tools under keys,
saved and opened), generation by model families with the views of
the level sets and the formation, simulations of a morphology (the
steps of a configuration run in a thread), exports on demand
(figures, videos, files, archive), optimizations from a morphology,
tile assemblies, Ashby charts; the 3D view of any item (viewer.js:
WebGL 2, no library, drawing the surfaces of GET /api/surface:
orbit, sections, phases, colour maps, magnified displacements)
tools/ the registry of tools (their tab and the heading of its menu
that lists them, forms of typed fields in boxes, one or
several items of the traits they need, jobs with their progress
and their stops, results and behaviours kept in the workspace,
archives); one module per tab: statistics (RVE
sizes, measures, correlation functions, stereology, percolation
thresholds and critical exponents, the grains and the texture
of an orientation map), mechanics
(loading paths, NTFA, material points, the identification of
behaviours, linear tensors, the stiffness of the frame of beams of
the networks: the items of the trait framework, its drawing by
the forces or energies of the beams, Ban's replicas), learning (a
behaviour learned from a material point or a morphology, trained
in the background, its
curves, its law kept as a behaviour), fracture
(stress intensity factors, growth, cohesive zones, polycrystals,
the phase-field fracture of a morphology, the growth of a curved
crack front in 3D), contact (the compression of a morphology
between two platens, its indentation by a sphere or a cone),
design (automaton rules of morphologies, rigidity of strut
lattices), stability (frames and morphologies, nonlinear paths, Bloch
bands), nonlocal_damage (microfoam, ductile with the triaxiality,
quasi-brittle), physics (transport properties and their bounds,
the flow through the pores with the acoustic parameters and the
absorption of a layer, dynamic moduli of viscoelastic composites),
couplings (diffusion, oxidation of a plate and of a morphology,
hygromorph multilayers), interop (AMITEX_FFTP, Abaqus, mesh and image formats,
MFront, Z-set files read and written), compare (the curves of
several runs on one figure)
analysis/ measurements on images
basic.py volume and phase fractions
surface.py Crofton specific surface (3 or 13 directions)
covariance.py covariance along axes, 3D covariance maps
connectivity.py connected components, percolation, periodic merging; the wrapping of a phase around the periodic box (wrapping)
percolation.py the forest fire (burning) of a phase from a face, compiled, and
its chemical distance; percolation thresholds of families of
images: bisection of the probability of percolating (labels
wrapping around periodic boxes, or the fire); their finite-size
scaling (the thresholds of several box sizes fitted to
p_c + a L^(-1/nu), weighted, nu given or fitted); the family of
the model of a configuration, in its box or, sized, in cubes at
its voxel size; planar images (one voxel along z: 4 or 8
neighbours, nu = 4/3); the anisotropy: the fire across the
core of the box along a direction (burn_along, the cylinder of
the inscribed ball), the percolation, chemical distance and
geometric tortuosity of one image along directions
(directional_percolation), the thresholds of a family along
them (percolation_anisotropy: each realization bisected along
all, paired deviations, ratio, threshold tensor); the rigidity
threshold, where the elastic moduli by the FFT solver vanish
(rigidity_threshold: the fraction where half the realizations
bear above a cut, the others void, extrapolated over box sizes;
images decided by their connectivity through the nodes and the
faces of the voxels or by inclusion when they can be, the others
by the energies of their solutions, resolved by the stress
averages or by the round-off; the bounds tried last); the critical
exponents of the conductivity and of the elastic moduli
(percolation_exponent: the FFT solutions of the realizations of
several box sizes at the threshold, their means fitted to
L^(-t/nu), the error of the threshold propagated by the same
images shifted, a sweep above it optionally)
cells.py cell volumes, adjacency, neighbours (negative labels ignored)
fractal.py box-counting dimension
morphology.py chords, granulometry and local thickness, Euler characteristic,
Minkowski densities, geodesic distances and tortuosity
correlation.py Torquato's descriptors: lineal-path, two-point cluster and
pore-size functions; radial distribution function and structure
factor of point patterns
random_sets.py structuring elements: dilations, erosions, openings, closings,
Choquet capacities and contact distributions, morphological
covariance and Minkowski curves of grey-level images
topology.py Betti numbers, connectivity density, Euler characteristic of
plane sections and integral of mean curvature
anisotropy.py mean intercept lengths along the 13 grid directions, the MIL
fabric tensor (principal MILs, axes, degree of anisotropy),
the first minima of the covariance (hole effect)
power_laws.py power-law tails of samples by maximum likelihood, the bound of
the tail by the Kolmogorov-Smirnov distance (Clauset et al.)
stereology.py sections by the grid planes (watershed, frame correction),
Saltykov's unfolding for spheres and oriented ellipsoids, the
consistency of a shape with the section radii
particles.py the particles of a phase in the volume: connected parts split by
the watershed of their distance map, volumes, radii, centres,
number per unit volume (frame correction of Miles-Lantuéjoul)
measures.py named measures used by configurations and the CLI (connectivity: the faces joined and, periodic, the wrapping around the box; directional_percolation: the crossings and tortuosities along the lattice directions)
rve/ statistics of RVE studies
scaling.py point variances, ensemble and window statistics, fits, RVE sizes
study.py series of properties x boundary conditions (Dirrenberger 2014):
intervals, errors, fitted/imposed exponents, bootstrap, isotropy,
discrepancy; the mean stiffness and conductivity tensors per volume
windows.py the window method for tensors: one periodic realization solved
once per unit load by FFT, the mean mutual energies of its windows
report.py Markdown, LaTeX, JSON reports and figures
dynamics/ explicit dynamics with OpenRadioss (a separate program)
model.py the consistent units of the inputs, the materials of the phases
(elastic, von Mises tabulated, Johnson-Cook, failure), the
loadings (compression, tension, shear, impact), the mesh of a
morphology and the estimate of its stable time step
openradioss.py the installation found (the paths, the executables, their
environment) or installed from a release on GitHub, the runs
(Starter, Engine, their progress, stops and failures) and the
conversion of their outputs (anim_to_vtk, th_to_csv)
results.py the results read back: the nominal stress-strain curves of the
faces, the energies, the fields of the animations on the mesh,
the summary, the frames for the 3D views
coupling/ transport, chemistry and mechanics
diffusion.py transient diffusion in voxel images (finite volumes), Crank's slab
oxidation.py thermal oxidation: the closed-loop scheme, thin films,
diffusion-limited profiles of a plate, and the voxels of an
image (oxidation_field: TR-BDF2, the local species eliminated
voxel by voxel, the oxygen by conjugate gradients)
ageing.py property laws of a tracer, graded phases, hydrolysis (Ekenstam)
plates.py hygromorph multilayers: moisture and curvatures, Timoshenko
fenicsx.py diffusion and free swelling of tetrahedral meshes (FEniCSx),
curvature of strips
laser/ architectured sheets by localized laser treatment (docs/laser.md)
heat.py moving Gaussian or top-hat beams on a plate, one face or both
(quasi-steady integral over the delay, images through the
thickness, losses; separable on grids), Rosenthal's limits, the
tabulated cycles of a track (TrackKernel), transient finite
volumes of a blank (LOD, Numba)
metallurgy.py austenite, martensite, bainite and tempering of C-Mn steels
(lever rule, Andrews, Koistinen-Marburger, Maynier,
Hollomon-Jaffe additivity), peak-temperature hardening
(AISI 430), melt-pool carburizing, nitriding and alloying
materials.py Ludwik or Voce laws fitted to Rp, Rm and Ag (Considere), the
strength of each hardness, the presets of the PIMM steels
patterns.py tracks and laser paths: straight, corrugated, Fourier, crossed,
honeycomb; length, time, energy, gaps, curvature, G-code
process.py the treatment of a path at points of the sheet, track after
track (tabulated responses, integrated cycles where tracks meet)
specimen.py dog-bone, strip and cell specimens, structured quadrilaterals
mechanics.py layered membrane in plane stress with large rotations (Numba
elements, corotational log strains, consistent tangent), the
tensile test and its measures
bending.py three-point bending of a strip: fibre sections, neutral axis,
moment-curvature, deflection by virtual work
design.py cases, measures, objectives and constraints; grid, Nelder-Mead,
differential evolution, genetic and Bayesian optimization
(Gaussian process, expected improvement)
solid.py the treated sheet in its volume: voxels of graded elastoplastic
phases with free faces (flow laws as von Mises plasticity),
its tensile test by the nonlinear FFT solver
run.py configuration files and outputs (morphogen laser)
figures.py hardness maps, curves, strains, histories (matplotlib)
homogenization/ apparent properties
core.py boundary conditions, loadings (strain, stress, mixed,
temperature), condensation of the macroscopic unknowns, solutions
materials.py isotropic and anisotropic materials (cubic, hexagonal, any
stiffness), rotations by grain orientations, material tables
(with the rotation of each grain); the plasticity or crystal
plasticity of a material
behaviours.py von Mises elasto-(visco)plasticity: parameters, implicit
integration and consistent tangent at the integration points;
Hill's criterion (closest point projection, Newton on the plastic
strain increment and dp; its CUDA port in behaviours.cu); the
integration of all the points in small strains (Numba: chunks
of points in parallel, the work arrays of the laws allocated
per chunk before the loop — no reference counted in the kernels)
porous.py porous elasto-(visco)plasticity (GTN, Rousselier, elliptic):
parameters, matrix stresses, kinematic hardening (backstresses
of GTN and elliptic), implicit integration along the invariants
of the relative trial stress, its analytical Jacobian, consistent
tangent (Numba; its CUDA port in behaviours.cu)
beremin.py cleavage by the local approach: the Weibull stress of Beremin on
the fields of a loading path (the largest principal stress, or
the normal stress of the cleavage planes of the grains), the
probability of failure of the weakest link
crystal.py crystal plasticity of Méric and Cailletaud: slip families of
cubic and hexagonal lattices, twins (one sense, a saturation),
cleavage planes (opening and accommodation systems that soften:
a crack band), interaction matrices (Franciosi), implicit
integration (flow rule as stresses, active set, substeps) and
consistent tangents in small and finite (multiplicative) strains
(Numba)
finite.py finite strains: elastic, von Mises and porous points in the logarithmic
strain space, the integration of all the points (dP/dF), total
Lagrangian element kernels, Cauchy stress, logarithmic strain,
lattice rotations (Numba)
mfront.py MFront behaviours through MGIS (optional, conda-forge): the
compilation of .mfront sources (cached), the loading and checks
of the behaviours, the integration of the points of the kind
MFRONT (which the Numba and CUDA kernels skip) in morphoGen's
conventions and arrays of internal variables
tensors.py 3 x 3 algebra of the kernels: Jacobi eigen-decomposition,
divided differences of the logarithm, Gaussian elimination
paths.py LoadingPath (components in time, strain or stress controlled,
temperature) and PathResult (curves, phase averages, fields; in
finite strains Cauchy stresses and logarithmic strains)
nonlinear.py the periodic element: Newton iterations with line search on the
fluctuation and the macroscopic strain components, small or
finite strains, time stepping, shared by both solvers
(_PathSolver: they give the discretization; _PointArrays: the
arrays and the integration of the points on the CPU);
tetrahedral finite elements (NonlinearFEProblem); the behaviours
of the elements; the material point simulator
fft_nonlinear.py loading paths on voxel images (NonlinearFFTProblem): B-bar and
F-bar voxels, the tangent applied matrix-free, Newton-Krylov with
FFT-preconditioned conjugate gradients (on the CPU or the GPU)
dilatation.py selective reduced integration of the 10-node tetrahedra: the
volume change of each element as its mean (B-bar, F-bar), the
conjugate stresses and the consistent tangent
fe.py finite elements on tetrahedral meshes (P1, P2; KUBC, MBC, PBC, SUBC;
integration: full or bbar); the reduced tangents T^T K T of a
loading path by the places of the stored values of K (Reduction);
the colours of the elements, for the assemblies on several
threads (element_colours)
fenicsx.py the same problems solved by FEniCSx (dolfinx, dolfinx_mpc, PETSc)
ntfa.py nonuniform transformation field analysis: modes (POD of
snapshots), influence fields (FFT), reduced laws along paths
identification.py parameters of a behaviour identified on a reference response
(least squares on the logarithms, material points, as Zopt); the
macroscopic law of a morphology (its loadings run, a law estimated,
von Mises or Hill, identified on them)
bloch.py Bloch waves: band structures of periodic cells (consistent
mass, Bloch-periodic map, shift-invert Lanczos), band gaps; the
scalar waves div(a grad u) + w^2 b u = 0 (acoustic, antiplane
shear, TM and TE light), speeds at long wavelengths
photonic.py light in three-dimensional photonic crystals: the vector waves
of Maxwell's equations by plane waves on the voxels (transverse
basis, operators by FFT, block Davidson with the inverse medium
for preconditioner), cells cut by interfaces averaged (Meade)
coupled.py strong couplings on voxel images (CoupledProblem): piezoelectric,
piezomagnetic, magneto-electro-elastic, magnetoelectric and
thermoelectric laws, their fields solved together (MINRES or
conjugate gradients, block Fourier preconditioner; conjugate
orthogonal gradients for complex laws), the tensors and constants
of the effective law, exact laminates
flow.py Stokes flow through the pores of voxel images by the lattice
Boltzmann method (two relaxation times, D3Q19): the permeability
tensor of Darcy's law
acoustics.py acoustics of porous media: the parameters of the equivalent fluid
of a rigid frame (tortuosity, characteristic lengths, viscous and
thermal permeabilities), the model of Johnson, Champoux, Allard
and Lafarge, the absorption of a layer
harmonic.py harmonic regimes: complex stiffness of viscoelastic phases,
complex conductivity of leaky dielectrics, effective refractive
index (complex symmetric laws, the coupled solver)
laws.py behaviours in closed form integrated for all the points of a
phase at once: generalized Maxwell viscoelasticity, hyperelastic
energies (and finite viscoelasticity), isotropic damage (crack
band, viscosity, the unilateral effect), Mazars' damage of
concrete, the damage coupled to plasticity (Lemaitre, Kachanov);
the seam of the learned and MFront points
fatigue.py multiaxial high-cycle fatigue of the points along a cycle: the
amplitude of the deviator (the smallest hypersphere that holds
its path), the criteria of Crossland, Sines and Dang Van, the
critical planes of Findley and Matake, the utilization of the
fields of a loading path (a criterion per phase), the factor
of safety
poroelastic.py weak couplings as eigenstrain problems: swelling (hygroscopic,
chemical), the poroelasticity of Biot (drained stiffness, tensor
and modulus of Biot, undrained response)
xfem.py extended finite elements of level sets on a regular grid
(LevelSetSample; ridge enrichment of the cut elements, voids
integrated on their solid side): elastic and transport problems,
loading paths in small strains
zmat.py material files of Z-mat (Z-set): gen_evp behaviours to the native
kernels or to MFront's brick StandardElastoViscoPlasticity,
porous_plastic ones to the porous kernels, the elastic, von
Mises and porous materials written back (write_zmat), and the
coverage of Z-mat's library (docs/zmat.md)
frames.py periodic frames of Euler-Bernoulli beams, in the plane and in
space (cubic, bcc, octet, Kelvin, diamond and pyrochlore
lattices of struts; or the
beams of circular sections of a framework): stiffness, axial
forces of a macroscopic strain, linear buckling, Bloch waves,
apparent stiffness, relative density; as an option, exact
Timoshenko beams (shear stiffness kGA per beam, Cowper's
coefficients) and their consistent geometric stiffness, and
beams of frameworks with rigid ends at their joints
(rigid_joints, in radii; the effective stiffness only)
networks.py effective stiffness of random periodic frames (fibre networks,
open foams, random lattices) by the sparse periodic problem:
Voigt stiffness, its split into stretching, bending, torsion
(and shear for Timoshenko beams), affine estimate,
non-affinity, by axis of a uniaxial stress (by_axis); beams with
rigid ends (the rigid arms of their nodes); the frame of a
realization of a network model (network_frame: its framework
and fibre radius, plane for a planar network; for morphogen
frame-stiffness and the interface); the axial force and the
energy of each part of each beam under a uniaxial stress
(beam_forces); networks of non-identical beams (Ban et al.
2016): exact second-order softening and scatter (modes axial,
bending, modulus, area), replicas with a control variate (their
table by CV: Softening.rows; a callback per replica)
bifurcation.py post-bifurcated patterns by group theory (Azulay), for frames
and continua: isotypic blocks, patterns of the axial
subgroups, first bifurcation, initial-stress matrix of
tetrahedra, symmetries detected on meshes; cells of several
primitive cells split by their translations (Bloch wave
vectors) and the little co-groups of their stars; along nonlinear
paths in finite strains (path_bifurcation: the tangent of each
step in the isotypic blocks, the switch of branch)
fft.py FFT-preconditioned solver of voxel images (periodic), on the CPU
or the GPU
voxel_fe.py the same trilinear voxels assembled on those of the materials,
conjugate gradients preconditioned by the sparse factorization of
the matrix shifted (VoxelFEElasticProblem, VoxelFEThermalProblem):
the FFT solutions in a few iterations near the rigidity threshold;
the voxels that bear along an axis (load_bearing: the parts cut at
single nodes through which a cycle winds around the box)
gpu.py the FFT solvers on NVIDIA GPUs (CuPy): CUDA kernels of the
stiffness and tangent operators of the voxels, the Fourier
preconditioner (cuFFT), conjugate gradients; the loading paths
with their integration points on the GPU (VoxelPaths)
behaviours.cu CUDA ports of the integration of the behaviours of the points
(von Mises, Hill, porous plasticity, crystal plasticity,
logarithmic strains) for gpu.py
kernels.py Numba kernels of the trilinear voxel elements (FFT solver: the
inverse of the Fourier symbol; the Gauss points of the loading
paths)
elements.py Voigt notation, isotropic stiffness, voxel element matrices
solvers.py direct, AMG-preconditioned and Jacobi-preconditioned CG
(MINRES after a direction of negative curvature: the
indefinite tangents of softening points); GMRES with
deflated restarts for the full tangents that are not
symmetric, recycling its space from a solve to the next
once one restarts (GCRO-DR, Recycling) and stopped where
it stagnates; the two-level preconditioner of 10-node
elements, its cycle in the order of the colours of its
sweeps; dot products, norms and products of long arrays in a
single thread (dot, norm, times), the Krylov solvers with the
BLAS on one thread; a solver updated with another matrix (the
next tangent of a path) keeps its coarse levels and gives its
sweeps the new values, until it is worn (LinearSolver.worn)
amg.py smoothed-aggregation algebraic multigrid (Numba); on large
matrices, Gauss-Seidel sweeps in the order of the colours of
the unknowns, those of a colour in parallel, on a copy of the
matrix sorted by colour (single precision, the parts below
and above the diagonal apart: Smoother) — by nodes and 3 x 3
blocks where the unknowns are the components of displacements
—, the products of a sparse matrix and of the dense matrix of
the coarsest level on several threads (product, dense_product);
the matrix by blocks of nodes in double precision (NodeBlocks:
the products of the Krylov iterations, the same to the bit;
the values of another matrix of the same pattern at their
places, which the sweeps follow)
apparent.py apparent moduli and conductivity, tensors, thermal expansion,
isotropy, named loadings (Response), homogenize()
postprocess.py compliance S = C^-1 and the notation of both (NOTATION: Voigt
11 22 33 23 31 12, engineering shears), engineering constants,
anisotropy indices, the directional Young's modulus (plane
strain or stress), solid fraction, effective density,
relative, normalized and specific values
bounds.py Voigt, Reuss, Hashin-Shtrikman (Walpole) bounds; bounds of
polycrystals of cubic crystals; Levin's thermal expansion
mesh/ meshes
tetra.py TetMesh (4- and 10-node tetrahedra, phases, grains, box faces,
outer faces and periodic links of whole periodic grains; the
departures of the mid-edge nodes of curved elements, their
volumes and Jacobians; the sharp edges that a mesh follows)
curved.py 10-node meshes whose nodes lie on the interfaces of the phases
(curved_mesh: Newton's method on the implicit functions, exact
for unions of spheres and segments, cubic splines otherwise;
the nodes of sharp edges on the two surfaces that meet there)
conforming.py conforming meshes of level sets with MMG (periodic along all,
some or none of the axes); the box cut piece by piece for the
structures that give their smooth pieces, whose sharp edges
are kept as ridges; mirrored meshes of symmetric cells (exact
symmetries; the chambers of the groups 4/mmm and m-3m with
diagonal mirrors, point_group), implicit bodies of level-set
functions
exact.py exact meshes without a background grid: the face-to-face complex
of convex pieces (overlaid coplanar faces, T-junctions) meshed by
MMG, adaptive meshes of point-wise level sets (octree, MMG -ls,
projection of the interface onto the exact level set; matching
faces cut exactly), the route of each microstructure (exact_mesh)
octree.py balanced octrees refined by a test of their cells, periodic
across chosen axes, split into tetrahedra (Kuhn cells, cones over
fanned faces; conforming, matching opposite faces)
polyhedra.py exact cells of power diagrams (CellComplex: pieces, facets)
regularize.py removal of small features of cell complexes (caps, short edges)
polycrystal.py conforming meshes of tessellations, one subdomain per grain
(MMG): whole periodic cells (translated facets meshed alike), or
cells cut by the box (periodic faces triangulated in the plane);
exact cells with their bends as ridges (exact=True)
voxel.py voxel hexahedral meshes
surface.py marching-cubes surfaces
io/ file formats (vtk.py, metaimage.py, native.py, tiff.py, meshes.py,
medit.py, geometry.py) and the dispatch functions save_*/load_*;
the Abaqus inputs of tetrahedra and voxels with the materials of
the phases (elastic, or MFront laws of the abaqus interface; in
finite strains all MFront laws, the axes of the orthotropic ones
in their state variables) and the orientations of the grains
export.py one entry point: the writer chosen from the object and suffix
ebsd.py orientation maps (EBSDMap): .ang (EDAX/TSL) and .ctf (Oxford
Instruments/HKL) files read and written, with the axes of each
system and of its hexagonal crystals; the grains of a map
(Grains: union-find on the disorientations of neighbouring
points, mean orientations, areas) and their polycrystal; the
map that an acquisition would give of a layer of a polycrystal
tables.py tables as .csv, .txt and .xlsx (no dependency), read back
xdmf.py XDMF images (raw binary data), chunked voxelization to files
solvers.py periodic constraint equations (Abaqus/CalculiX), DAMASK grids
and material files, Neper seeds, FEniCSx XDMF meshes
amitex.py AMITEX_FFTP runs of an image (material and zone fields, material,
loading and algorithm files, MPI script, manifest; the nonlinear
materials as MFront laws of the umat interface; in finite
strains every phase) and their results (.std, .mstd, .zstd;
apparent stiffness)
mfront.py MFront sources of the nonlinear materials (von Mises and Hill:
the brick StandardElastoViscoPlasticity, Hill's criterion an
orthotropic behaviour in the axes of the sample; porous
plasticity: GTN in the brick where it is morphoGen's (its porous
criterion, the porosity in its implicit system, f0 until the
matrix flows), the others morphoGen's kernel from behaviours.cu
(porous_update, small strains or Hencky's); crystals: an
implicit scheme with morphoGen's slip systems, in finite strains
morphoGen's kernel from behaviours.cu; Hencky's elasticity in
finite strains; the learned laws, learning.mfront; thermal
expansion from 293.15 K), their
internal variables, the functions of the umat interface; the
user's MFront laws as the exports need them (UserBehaviour:
strain measure, axes for Abaqus), their parameters
config.py configuration schema and validation
pipeline.py batches, RVE studies, reports
results.py database of the configurations tested and their results (SQLite),
filters, CSV exports, normalized Ashby charts
parallel.py realizations in processes, tasks and MPI; records, merge, status;
job scripts and submission (SLURM, PBS, Torque, SGE, OAR, LSF)
cli.py command line
templates/ example configurations (morphogen template <name>)
3. Data model¶
3.1 Domain and grids¶
Domain(size, origin=(0, 0, 0), periodic=True, unit="") is the box
[origin, origin + size).
Periodic domains are tori: grains, fibres and cells wrap around the faces, so that samples tile space (periodic boundary conditions, FFT).
Non-periodic domains are windows of the infinite random medium. Germs are drawn in the domain dilated by the largest grain extent, so that the window is unbiased (no edge effect).
Grid(domain, shape) splits the domain into equal voxels. Give either:
shape=n: n voxels along the longest edge, voxels as cubic as possible;shape=(nx, ny, nz);voxel_size=h: the shape is rounded and the spacing adjusted so that the grid covers the domain exactly.
3.2 Models¶
Models are frozen dataclasses registered under a name (boolean_spheres, …). Each
parameter is declared with param(default, doc=..., kind=..., unit=...). This metadata
serves configuration parsing (Model.from_config), the morphogen models command, the
documentation and, later, the GUI forms. Plain values are converted on construction:
numbers or mappings become distributions, strings become orientation distributions.
Model |
Microstructure kind |
Phases |
Periodic |
|---|---|---|---|
|
|
0 matrix, 1 spheres |
yes / no |
|
|
0 matrix, 1 fibres |
yes / no |
|
|
0 pores, 1 fibres |
no |
|
|
0 matrix, 1 fibres |
yes / no |
|
|
0 matrix, 1 spheres |
yes / no |
|
|
0 matrix, 1 fibres |
yes / no |
|
|
0 matrix, 1 fibres |
yes / no |
|
|
0 pores, 1 fibres, 2 cross-links |
no (infinite fibres) / yes |
|
|
0 pores, 1 fibres |
yes / no (the periodic network cut by the box) |
|
|
per cell (and the boundary phase) |
yes / no |
|
|
0, 1 excursion |
yes / no |
|
|
0 void, 1 solid |
yes / no |
|
|
0 void, 1 struts |
yes / no (rotated or graded: no) |
|
|
0, 1 |
yes (quasi-periodic tilings: their approximants) / no (windows) |
|
|
0, 1 |
yes (quasi-periodic tilings: their approximants) / no (windows) |
|
|
0, 1 |
yes (quasi-periodic tilings: their approximants) / no (windows) |
|
|
0 matrix, 1 grains (or per grain) |
yes / no (windows) |
|
|
0 pores, 1 aggregate |
clusters: yes / no; deposits: no |
|
|
0 void, 1 solid |
yes / no |
|
|
0 void, 1 struts |
yes / no |
|
|
0, 1 layers |
yes / no |
|
|
per leaf (and the background) |
yes / no |
|
|
0 matrix, 1 spheres |
yes / no |
|
|
0 below, 1 above the threshold |
yes / no |
|
|
0, 1 |
yes (periodic lattice) |
|
|
the phases of the map, from 1 |
no |
|
|
0, 1 (set operations); the matrix’s and one per inclusion model |
if every component is |
|
|
0 voids, 1 ligaments |
no |
Parameters of kind models hold nested models (mappings with a type key). The
parameters that are files (path_parameters of the class: the file of ebsd, the
texture.file of the tessellations whose texture is that of a map) are taken from the
folder of the configuration file, in the models that a model holds too. Some flags
depend on the parameters: has_level_set (tessellations with a grain-boundary phase 0,
set operations of models with level sets), meshed_by_grains (tessellations without a
boundary phase), supports_periodic (combinations). Model.active_phases() gives the
phases the realizations can contain, checked against the materials of a configuration.
Adding a model takes four steps:
Subclass
Modelas a frozen dataclass with aname,phasesandparamfields.Implement
_validate,generate(return an existing microstructure kind when possible) andtheory.Decorate the class with
@registerand import it inmodels/__init__.py.Add a template and tests against the theory.
3.3 Microstructures¶
Every microstructure implements:
voxelize(shape | voxel_size | grid, supersampling=1): phase image (uint8), or partial volume fractions of phase 1 (float32) whensupersampling > 1, withsupersampling³samples per voxel;level_set(...): a field negative in phase 1, zero on its boundary:exact signed distance, clipped to a band, for unions of grains;
first-order distance (φ − t)/|∇φ| for TPMS;
(threshold − Y)·ℓ for Gaussian fields;
not available for tessellations (
Model.has_level_setis false), which are meshed grain by grain from their exact cells (mesh.polycrystal) or homogenized by FFT on voxels; with a grain-boundary phase 0, half its thickness minus the distance to the nearest face shared by two cells;min, max or negation of the level sets of the components for set operations;
contains(points): membership of arbitrary points (unions, tessellations, TPMS);to_dict()/Microstructure.from_dict(): JSON description holding the model, domain, seed and explicit geometry; regeneration from JSON is exact;summary(): counts (spheres, segments, cells, modes).
Tessellations also provide labels(...) (int32 image of germ indices), cell_of(points),
boundary_distance(points), cell_complex() (exact cells, cached) and cell_table()
(germs, weights, phases, Euler angles). Dead leaves label the visible leaves, and
inclusions in a polycrystal keep the labels and orientations of its cells. has_labels tells whether a
microstructure has cell labels (tessellations, grain and cluster structures); the
pipeline then uses them for the cells measure and the labels outputs.
LatticeField(domain, phases, labels=None, orientations=None, smoothing=0, field=None, threshold=0, attributes=None) is the realization of the lattice models: a phase per
cell of a regular lattice of the domain (one cell through the thickness for planar
structures), optional grain labels (int32, -1 outside any grain) and orientations, an
optional continuous field (phase 1 where it exceeds the threshold), and JSON-able
attributes (histories, kinetics, tile indices). It is exact on its lattice; elsewhere
the level set is interpolated trilinearly from the cells (theory §14). Arrays are
stored in JSON as base64 zlib data. Models with a formation history implement
evolution(domain, seed, frames): lattice states along the formation of the
realization of generate(domain, seed), which is the last frame.
3.4 Images¶
VoxelImage(data, spacing, origin, periodic, name, metadata):
datahas shape (nx, ny, nz) and is indexed[i, j, k]along (x, y, z);originis the lower corner of voxel (0, 0, 0);metadata(model, domain, seed, version, phase names) follows the image into.npzfiles and run manifests.
3.5 Tetrahedral meshes¶
TetMesh(nodes, elements, phase, lower, upper, periodic, metadata, cell, orientations)
holds 4-node or 10-node tetrahedra (VTK / Abaqus C3D10 order: corners, then mid-edge
nodes of edges 01, 12, 02, 03, 13, 23; isoparametric: a mid-edge node off the middle of
its edge makes the edge a parabola, deviations(), curved, straight(), and
volumes() integrates the Jacobian) with one phase per element, in the box
[lower, upper]; meshes of polycrystals also give the grain of each element (cell)
and the Euler angles of the grains (orientations), kept by every mesh format that
can hold them; meshes that follow the sharp edges of their phases give these edges
(sharp: pairs of corner nodes) and the two pieces of the structure that meet along
each (sharp_pieces), which follow the meshes made of them (quadratic(), linear(),
select()).
face_nodes(axis, side),boundary_faces(axis, side)(outward triangles),periodic_pairs(axis)andmatching_faces()describe the faces of the box;quadratic()adds the mid-edge nodes,linear()removes them,subset(phases)keeps some phases (e.g. the solid of a porous medium);volumes(),quality()(1 for a regular tetrahedron),phase_fractions()(fractions of the box: a phase left out of the mesh is a pore),summary().
conforming_mesh(micro, size=..., ...) builds it (see §4 and theory §11), grain by grain
for tessellations (polycrystal_mesh, theory §19).
3.6 Homogenization problems¶
make_problem(sample, materials, physics, bc) returns:
FEElasticProblem/FEThermalProblemfor aTetMesh(any boundary condition);FFTElasticProblem/FFTThermalProblemfor aVoxelImage(periodic).
problem.solve(Loading.strain(E) | stress | gradient | flux, fields=False) returns a
Solution: macroscopic strain and stress (gradient and flux), energy, the per-phase
means and standard deviations of the fields, the solver data, and optionally the
element fields and their support (mesh or image). The assembled system (and the
multigrid hierarchy) is reused for every loading with the same constraints.
homogenize(sample, materials, bcs=..., properties=...) returns the apparent bulk and
shear moduli, conductivity, stiffness and conductivity tensors, as Apparent records
(value, energy estimate, isotropy criterion, response, solution). Crystals
(Material.cubic, .hexagonal, .anisotropic) are turned by the orientations of the
grains: those of a polycrystal mesh, or grains (label image) and orientations for a
voxel image; material_table gives one matrix per phase and grain.
The FFT problems, NonlinearFFTProblem, make_problem and homogenize take
device="cpu" | "gpu" | "auto" (morphogen.gpu): on the GPU, the conjugate gradients run
there, with the same solutions (info.method == "fft-gpu"); finite elements run on the
CPU.
4. Algorithms¶
Unions of grains.
Primitives are registered in a uniform grid of cells, stored as compressed sparse rows. A primitive goes into every cell that may contain a point closer to it than its radius plus the level-set band, so each query point only tests the candidates of its own cell.
Voxels are processed in parallel over x slices.
Periodic boxes are handled by adding the periodic images of the primitives that cross the faces, before a non-periodic rasterization.
Signed distances are exact outside the grains, and bounded by the deepest grain inside.
Poisson fibres.
Isotropic lines hitting the ball that contains the dilated box are drawn: a Poisson number with mean L_V π ρ², uniform directions, uniform points in the projected disk.
The lines are clipped to the box dilated by the fibre radius, which gives the exact restriction of the infinite model.
RSA. A sequential Numba kernel with linked cell lists; candidate positions come in batches from the NumPy generator, so every random draw stays reproducible. Non-periodic samples are cut from a larger periodic box (no wall effects).
Tessellations.
Each voxel goes to its nearest germ through a SciPy k-d tree, toroidal for periodic boxes, processed in slabs.
Laguerre (power) cells are Voronoi cells of the germs lifted to 4D with the coordinate sqrt(w_max − w_i).
Non-periodic germs are drawn in a window dilated by three cell sizes; a warning is raised if a voxel lies farther from its germ than that margin.
Grain-boundary phases: distances to the planes of the k nearest lifted germs, k doubled until no farther germ can bring a closer plane (theory §22).
Packed germs: random sequential addition of the scaled spheres (largest first) in the periodic box or in the periodic version of the window.
Exact cells. Each cell is the half-space intersection (SciPy/Qhull) of the bisector planes of nearby germs (k-d tree, periodic images included), checked against the other germs at its vertices and recomputed with the missing ones; the pieces inside the box, their facets and vertices are merged into a
CellComplexwhose facets are shared exactly by two pieces.Polycrystal meshes. Regularization of the complex (
mesh.regularize), cones from the centroids of the pieces over their triangulated facets, thenmmg3dremeshing of the subdomains (-hmax,-hausd,-hgrad). Periodic faces: subdivision of the box edges alike on parallel edges, planar straight-line graph of the traces of each pair of faces, constrained Delaunay triangulation (exact orientation tests with rational arithmetic when rounding could change a sign),mmg2d -nosurfremeshing with the traces required, canonical flips; the face triangles are then required inmmg3d. About 20 s for 300 grains and 2 × 10⁵ elements.Dead leaves. Leaves are drawn in batches until the centres of a check grid are covered; a Numba kernel gives each voxel the first leaf covering its centre (nearest image on the torus) and stops once every voxel is covered.
Random laminates. Layers are intervals along the axis (binary search for each slice); the holes of a layer are a k-d tree of discs, lifted to 3D for variable radii (inside a disc when the lifted distance is below the largest radius).
Chunked images.
io.voxelize_to_filevoxelizes slabs of the grid along z (sub- grids of their own) into anumpy.memmapof the raw file; the models evaluated point by point give the same voxels as at once.Gaussian fields.
A field is a finite sum of Fourier modes on the generation box: the domain itself if periodic, the domain enlarged by 3 correlation lengths otherwise.
Amplitudes come from the Gaussian spectrum, cut off at |k|ℓ = 6.
The field is evaluated on grids by one inverse FFT (periodic, aligned grids) or by a separable contraction (any grid), so one realization gives the same values at any resolution.
TPMS. Closed-form level-set functions are evaluated in Numba. The level for a target fraction is the quantile of φ (or |φ|) over a fine sampling of the unit cell.
Strut lattices. Unit cells are defined by nodes and neighbour distances, then tiled with duplicate struts removed. The radius for a target density comes from a root finder on a voxel estimate of one periodic cell.
Specific surface. The Cauchy–Crofton formula S_V = 2 ⟨P_L⟩, with crossings counted along 3 or 13 lattice directions weighted by the solid angles of their spherical Voronoi cells. The estimate is biased low by roughly (voxel size)/(mean chord), because chords shorter than a voxel are missed.
Surfaces. Marching cubes (scikit-image) on level sets padded with outside values; cap vertices are moved onto the box faces, which gives watertight, manifold, outward-oriented surfaces.
Conforming meshes. The level set is sampled on a Kuhn triangulation of the box (reflected in the cubes of odd index: no preferred diagonal) and snapped away from zero near the nodes; MMG (
mmg3d -ls, run as an external program from thepymmgwheel) discretizes the interface and remeshes. Periodic microstructures get matching faces through a first cut without remeshing, then a remeshing with the face triangles required. About 0.2–0.5 ms per output element.Finite elements. Node-level sparsity pattern from the element connectivity, then dof-level CSR matrices assembled in Numba; constraints and periodicity as a prolongation
T(reduced matrixT^T K T); averages and fields integrated in parallel Numba kernels.Multigrid. Aggregation, candidate fitting (modified Gram-Schmidt per aggregate) and Gauss-Seidel sweeps in Numba; sparse products in SciPy. For 10-node elements, a two-level preconditioner on top of the 4-node hierarchy.
FFT solver. Matrix-free trilinear operator (loop over nodes, no atomics), real FFTs of SciPy (multithreaded), 3×3 inverse symbols precomputed per frequency.
Cellular automata. Neighbour counts by a parallel Numba loop over the cells (any offsets, periodic or fixed boundaries); large weighted neighbourhoods (activator–inhibitor) by FFT convolution; conserved rules by
argpartitionof the scores. Gray–Scott steps are a parallel stencil, on tilings a parallel loop over the CSR face weights. On tilings, neighbourhoods are CSR lists (lattice neighbours found once per basis point in a block of unit cells, then translated; quasi-periodic adjacency from the integer vertices of the multigrids, packed into one integer each and numbered by one sort, those of approximants first reduced modulo the periods by an integer adjugate), counts parallel sparse sums, and rasterization a nearest-centre query (k-d tree) in the undistorted tiling, with a point-in-polygon test for rhombi and a test of the coordinates along the edges for rhombohedra (16 nearest centres, then the ball of the circumradius for the points left).Nucleation and growth. An event-driven automaton: a binary heap (Numba) of (arrival time, cell, nucleus) events; each cell is finalized once, by its first arrival, and pushes candidate arrivals to its 26 neighbours (theory §15). Potts sweeps are parallel over colour classes with counter-based random numbers.
Aggregation. A sequential Numba kernel driven by the NumPy generator; contact counts maintained incrementally; launching spheres, big jumps and kill radii for single clusters, kill heights for deposits; Eden growth over a perimeter list; ballistic flights by voxel traversal.
Wave function collapse. Domains are 64-bit sets; arc consistency with a stack; least-entropy choice; backtracking with a trail of domain changes, restarts after a number of backtracks (Numba).
Transfer matrices. Admissible sections enumerated by depth-first search (Numba), the sparse matrix built in blocks, exact counts with Python integers (object arrays), uniform sampling from the backward counts.
Frameworks. Lattices from the strut unit cells with nodes identified modulo the box; ranks by dense SVD (up to about 1000 nodes).
Topology optimization. The graded operator
sum_e s_e K0is applied matrix-free (as the FFT solver); each load case is solved by preconditioned CG from the previous solution (warm start), to a relative residual of 1e-4 during the optimization and 1e-6 for the final thresholded design; mutual energies of the element fields give the tensor and the sensitivities in one parallel pass. The filter and the automaton neighbourhoods are FFT convolutions. MMA solves its dual subproblem with NumPy.Parallel runs (
parallel). The realizations of a run are items (index, seed key, file stem, size). Each one runs in a fresh process (spawn) of a pool, with its share of the threads (NUMBA_NUM_THREADS, BLAS variables,settings.threads), and writes a JSON record atomically (records/<stem>.json);--resumeskips the items with a record. Tasks take every n-th item (--task-id,--n-tasks, orautofrom the scheduler’s array index), MPI ranks likewise;mergegathers the records into the results and report once every item has one.job_scriptwrites the array and merging scripts of SLURM, PBS Pro, Torque, SGE, OAR and LSF;submitsubmits them, the merging job depending on the array.Interface.
ThreadingHTTPServeron 127.0.0.1; JSON API (/api/models,/api/generate,/api/evolve,/api/count,/api/optimize,/api/job/<id>); optimizations run in threads and are polled; images are PNG data URLs; downloads come from a temporary session folder.
5. Tests¶
pytest runs the suite in tests/; ruff check and ruff format --check check the
style.
File |
Content |
|---|---|
|
domains, grids, seeds, distributions (moments against sampling), orientations (uniformity, Euler round trips), images |
|
rasterization, supersampling, signed distances, periodic images and RSA against brute force |
|
Boolean spheres (volume fraction, covariance, convergence of S_V, integral range), unbiased non-periodic windows, Boolean cylinders, Crofton anisotropy, Poisson fibres (fraction, covariance, number and isotropy of fibres), hard spheres |
|
Voronoi/Laguerre cells against brute force, Poisson-Voronoi statistics (CV² of volumes, mean number of faces through Qhull, boundary density), phases and orientations, Gaussian fields (evaluation paths, resolution independence, fraction, Rice formula, covariance), TPMS (symmetry, target fraction, periodicity), lattices (closed-form density, target density) |
|
registry, config and JSON round trips for every model, error messages, point queries |
|
exact answers on synthetic images (planes, balls, stripes, white noise, components, cells) |
|
percolation thresholds: the exact brackets of the step of a slab (rising and falling, open and periodic), wrapping around the periodic box against joining its faces (a staircase, a helix, diagonals through edges and corners), site percolation of the simple cubic lattice on 48³ voxels (0.3116 through faces, open and periodic; 0.0976 through 26 neighbours), Boolean spheres near 0.2895, the seeds, the errors; the family of the model of a configuration (its parameter set to the fraction, the refusals), |
|
formulas, fits, published RVE sizes, window method against the exact finite-window variance of the Boolean model; shifted tilings of periodic images (the mean of all their windows the image mean exactly, the variance pooled within the tilings, one tiling unchanged to the bit, squares and their edges), their variance on independent voxels within the error of one tiling, unbiased, of a smaller scatter |
|
the window method for tensors: windows tiling the box give its apparent stiffness and conductivity (round-off, Hill’s lemma), symmetric and positive tensors, their statistics those of the windows as realizations; the known windows of a laminate of orthotropic phases (shear slots 23, 31, 12); planar images (squares, the plane stiffness), pores (NaN compliances); the pipeline (results.csv, run.json, report), refused configurations, the RVE tool of the interface; shifted tilings: every window of a laminate, a random image (no new solve, the first tiling the unshifted one to the bit, each tiling of the box its tensor, the variances within the error of one tiling), |
|
Dirrenberger et al. (2014): RVE sizes and fibre counts of Table 10, discrepancy (eq. 69), relative errors, recovery of K and γ, isotropy, tables, reports |
|
finite elements: element matrices, exact homogeneous media under the four conditions (P1, P2), Backus laminates, energy ordering and bounds, averages and fields, porous media and floating clusters, scikit-fem agreement, iterative solvers against direct, non-matching periodic faces, Table 4 bounds |
|
the kernels of the iterative solvers: the product of a matrix and a vector on several threads against SciPy’s, to the bit; the colours of the unknowns (no stored value couples two of a colour, a value stored on one side of the diagonal included); the sweeps in the order of the colours against Gauss–Seidel on the matrix sorted by colour, the forward sweep from zero and its residual on half of the matrix each; the two-level preconditioner (symmetric positive definite, the iterations of the sweeps in the order of the unknowns); the same results to the bit on one thread and on several; the reduced tangent by the places of its stored values against the products of sparse matrices; the sweeps by nodes (3 × 3 blocks, components that a condition holds) against Gauss–Seidel in the order of the colours of the nodes; the products without BLAS against NumPy’s; the BLAS on one thread within a Krylov solve, its threads back afterwards; the matrix by blocks of nodes (its product against that of the stored values to the bit, the values of another matrix at their places, a matrix with a value outside the blocks); the sweeps that follow the blocks against those built for the new matrix, to the bit; a solver updated with another matrix (products, sweeps, solution; a matrix that is not symmetric); the wear of a preconditioner; a plastic path with its preconditioner kept from a step to the next against the direct solver |
|
the assembly of the paths on several threads: the colours of the elements (no node in common within a colour); the tangent, the nodal forces, their sums and the couplings assembled colour by colour against the elements taken one by one (small and finite strains, full integration and B-bar, one and several elements per chunk), the same to the bit on one thread and on several |
|
crystallographic textures and orientation maps: the rotations of the eleven Laue classes (orders, closure); disorientations of known rotations and of random pairs against Mackenzie’s distribution; the mean orientation of a grain; the axes of the families of planes, poles and projections; pole figures (a mean of one, a random texture, the kernel at the poles of one crystal, weights); inverse pole figures and their colours; textures drawn from measured orientations; maps written and read ( |
|
FFT solver against a direct solution of the same operator, homogeneous media and laminates, translation invariance, Hill’s lemma, porous media |
|
the voxels assembled and factorized: the matrix against the operator of the FFT solvers on the nodes of the materials (elastic and thermal, planar, one or two voxels along an axis), the stiffness and conductivity tensors of the FFT solvers (soft or void phase, three spacings), the moduli near the rigidity threshold (sites of the square and cubic lattices, floppy at round-off and bearing to the tolerance, 2 iterations against 300 to 900), uniform and void images, the errors; the voxels that bear (chains of pins, a hanging pixel and block, a column, a hinged voxel) and their moduli and conductivities, those of the whole images |
|
MMG meshes: tiling, mirrored meshes of symmetric cells (exact symmetries, matching faces, the group of a 2 x 2 array of holes), volume fractions, exact faces, matching periodic faces and exact PBC, porous lattices, exports (VTU, INP, MSH, Medit) and readers |
|
B-spline functions against the recurrence of Cox and de Boor, their derivatives and the embeddings of their refinements; exact arcs, refined patches and assemblies; the patch test (plane strain, plane stress, a solid); Lamé’s thick-walled cylinder (orders 3 and 2, 4 and 3) and Kirsch’s plate with a hole; the effective tensors of a homogeneous cell, of a laminate (elasticity, conduction) and of a cell with a hole, against the curved finite elements of the solid (slow); cells of any lattice (other cuts of the same cell, strips with gaps, a cell and its neighbour), the square and hexagonal arrays of holes against the series of Perrins et al. and the isotropy of the hexagonal one, the sheet with alternating holes (exact fraction, symmetries of the square, the square array with circles, Poisson’s ratio against the ligaments; against the FFT solver, slow) |
|
the design and its geometries (linear maps, symmetric radial variables); the analytical shape derivatives of the tensor and of the fraction of a cell (elasticity, conduction) and of the compliance and the area of a structure against central differences; the hole of largest bulk modulus against the bound of Hashin and Shtrikman (Vigdergauz; SLSQP and MMA); the hole of a plate under biaxial tension against the ellipse (slow); folded designs and the history of the kept iterates; variables linked by the symmetries of a pattern (images and directions, periods, derivatives, errors); the holes of an auxetic sheet found again from their effective constants; its lowest Poisson’s ratio at a given stiffness, with linear constraints |
|
the implicit geometry of every model: for the 39 models and 19 other configurations of several phases, an implicit function of each phase whose sign is that of the voxel image on two grids and which partition the points; the Menger sponge on the grid of its cubes, two cells and their boundary layer against their planes, leaves and a transparent leaf against their spheres; the interpolation on the background mesh; the conforming mesh of a coated sphere (three phases cut one after the other: volumes, matching faces, the phases kept, finite elements against the FFT solver), cells of three phases meshed by their grains, coated hard spheres (slow) |
|
the inputs of OpenRadioss read back card by card against the columns of the Radioss Reference Guide (units, nodes, bricks and tetrahedra oriented for Radioss, their mid-side nodes on their edges, parts, the solid property, LAW1, LAW36 with its yield curves at the rates of Norton’s viscosity, LAW2, node groups, held and driven faces, the anchor of free faces, a shear, an impact on a rigid wall, the self contact, the time histories and their sampling, the Engine input, the length scale); the model (units, materials, loadings, time steps); the listings, time histories and legacy VTK grids (ASCII and binary, classic and VTK 5.1) |
|
the installation of OpenRadioss found (the paths, the environment, the permissions restored), runs with a stand-in of OpenRadioss ( |
|
the tab Explicit dynamics: the inputs written without OpenRadioss (engineering units, the materials of the fields and of the JSON), the errors, a dynamic test with the stand-in (its summary in MPa and MJ/m³, the curves in morphoGen’s components compared with another run, the 3D view of its last frame), the results read back, the installation found, an impact |
|
the stand-alone application: the dialogs, files and folders of the page (a stand-in of the window), the confirmation of closing an unsaved workspace, the jobs stopped with the window, the check without a window, the fallback on the browser, the hooks of the page; the real window with |
|
round trips of every format and data type, independent readers (VTK, Gmsh) when installed, mesh structure and files, watertight surfaces, geometry files |
|
templates, configuration errors, batch outputs, split runs equal to single runs, window studies, CLI commands, |
|
configuration of meshes and methods, FFT and FE batch runs with fields and meshes, a run on a planar section of its volume ( |
|
mixed strain and stress loadings, thermal expansion (Levin, Rosen-Hashin), porous media, the post-processing of apparent properties; the stiffness and its compliance in every output ( |
|
neighbourhood sizes and sums against brute force, FFT kernel sums, Life notation, glider and blinker, majority and annealing, coarsening, conservation, probabilistic and activator–inhibitor rules (wavelength), updates, histories, Gray–Scott patterns |
|
neighbour counts and symmetry of every lattice (BCC shells 8 + 6, FCC 12 + 6), Voronoi cell volumes, rasterized cell areas and volumes, Penrose ratio of thick to thin rhombi, vertex stars and their frequencies; Ammann–Beenker tiles (ratio √2, equal areas, the six vertex stars and their window frequencies, clipped areas summing to the box’s, a disc); golden rhombohedra (volumes, ratio τ, faces shared by two cells, V = C, E = F = 3C, a ball, every point in one cell); the 24 vertex stars of the icosahedral tiling (their frequencies in τ against the identities of the tiling and the shares of the acceptance window, the stars of a patch and of the approximants 5/3 and 8/5, the kinds against the stars up to the 120 symmetries of the icosahedron); the square tiling against the grid automaton, growth from a seed (hexagons, cuboctahedra, rhombic dodecahedra), rules on irregular neighbourhoods, colourings, activator–inhibitor weights, models (JSON, windows, supersampling, frames), inverse design, square realizations unchanged (digests); finite-volume Laplacians (face pyramids, plane waves, homogenized diffusivity of rhombi), Gray–Scott on the square tiling against the grid (round-off), Turing wavelengths on hexagons and triangles, reaction–diffusion models on tilings; periodic approximants of the three quasi-periodic tilings (tiles per period, edges and areas of the tiles, filling the box, every edge or face shared by two cells across the faces of the box, invariance by a period, every point in one cell, no seam in the images, Euler’s relation on the torus, tori too small for their cells refused (otherwise the neighbourhoods of the shared sides, edges and faces), the periods, boxes fitted to them and their face measures, refusals, tile ratios and vertex stars converging with the order, the models in periodic boxes and windows, Gray–Scott through the FFT solver) |
|
lattice fields of the automaton models: exactness on the lattice, other resolutions, level sets, exact fractions, JSON, windows, planar structures, smoothing, conforming meshes |
|
Voronoi distances, faceted metrics, phantom nuclei, Avrami kinetics, Johnson–Mehl densities, planar kinetics, partial transformations, Potts energy decrease, DLA dimensions, Eden, deposits, models |
|
tile sets and rotations, exact counts (brute force, transfer matrix, gauge), uniform sampling, wave function collapse, framework rigidity and Maxwell’s transition, annealing and genetic search, inverse design of rules, models |
|
laminates, finite-difference sensitivities, filters and symmetries, MMA against SLSQP, Hashin–Shtrikman bounds, auxetic and target designs, |
|
moving beams against Rosenthal and Rykalin and the energy of a cross-section, separable grids, beams of both faces, transient finite volumes against the quasi-steady cycles; Andrews, Maynier, Sieverts, the lever rule, Koistinen–Marburger, Hollomon–Jaffe additivity, peak hardening and its tempering, melt-pool mass balances; flow laws against their tensile properties; paths (arc length, curvature, gaps, G-code, honeycomb sides); tracks that harden and temper; the membrane against exact tension, finite differences of its tangent, rigid rotations, bands at equal strain; the Gaussian process; a case and its optimization; |
|
a treated sheet in its volume: the flow laws of the sheets as von Mises plasticity (Voce exactly, the power laws within 1.5 %), the volume of a hardness map (classes and their mean hardness, layers, the gap of its free faces, refusals), a sheet of one hardness against its flow law in uniaxial stress, layers and stripes along the tension against the mean of their laws, stripes across it between its bounds, the cell of a laser treatment against its membrane along and across the tracks |
|
contact of surfaces: the nearest point of a face (the regions of a triangle against a search, curved faces of six nodes, edges, the slack and the rim); the distances of the rigid obstacles (their gradients are their normals), their motions; the surfaces of a mesh (free faces, phases, points and weights, rims, bodies assembled); two blocks of meshes that do not match pressed together (the pressure of the uniaxial stress, the works; 4- and 10-node elements, one pass and two); the contact points at the nodes of the faces (their weights, the same blocks, a block whose sides carry nothing, the corner of a cube against a plane and against a block); the augmented Lagrangian; a block slid far along a track (T = μ N with the points on either surface or both, the work of friction); Hertz’s line contact (pressure, half-width, profile, the work of the cylinder); a platen that rubs; a slot that closes on itself (the stiffness of the solid block, the pressure at the nodes, the file); periodic cells (layers that close on their images through the faces of the cell: the stress of the solid in uniaxial strain and stress; a cell cut through its slot or beside it; the walls cut by the faces of a cell have no rim there); two slots parted by a ligament thinner than its faces; the option |
|
cells whose interfaces debond: the interfaces opened on the torus (inside a cell, on its faces, cells one element thick, 10-node elements, grains, a function of centroids; the periodic links of the two lips; the errors); a bar with one interface against its closed forms (the bilinear law in tension and in shear, the work that breaks it, the lips pushed back, Coulomb’s bound over the elastic slip and the energy dissipated, lips without adhesion, the exponential law, uniaxial stress, finite strains, the recurrence of the viscous damage and the interface that breaks at once); grains; kinematic conditions against the solver of bodies with interfaces; a law for each pair of phases (three interfaces in series, the fields of the damage and of the opening of the elements, grains of one phase); the grain boundaries of a polycrystal that break (the work of the stress against the fracture energy times the area of the crack, whole grains against grains cut by the box; slow); beads that debond from their matrix (slow); |
|
the ready-made tests of contact: a block between lubricated platens (the uniaxial stress with the penalties in series, 4- and 10-node elements, the contact points inside the faces and at their nodes), between platens that rub (stiffer, below the block held on its sides), a block with a slot (softer until the slot closes on itself); the indentation by a sphere and by a cone (the powers 3/2 and 2 of the depth, the modulus with the base of the sample, the unloading along the loading); the correction of a layer against its published coefficients; the errors; the commands |
|
the tab Contact: its catalog and forms; a porous morphology between two platens (softer than its solid, the test run directly, with friction); a morphology indented by a sphere (the test run directly, the modulus with the base, the unloading) and by a cone; the errors of the forms |
|
Coulomb’s friction of interfaces whose nodes match: the law of a point (stick, radial return, the energy of a cycle, the strength left to a broken cohesive point); two blocks pressed then sheared (stick, then T = μ N, in small and finite strains, 4- and 10-node elements; the balance of the works); a cohesive interface that keeps Coulomb’s strength once broken, or broken from the start; a closed crack that slides by the shear less Coulomb’s and sticks below it, the tractions of its lips; the stress intensity factor and the release rate of an edge crack whose lips rub (slow) |
|
the meshes that follow the sharp edges of the phases: the overlaps of primitives (depths, thicker radii, the faces of a periodic box against the search of every pair, segments), colourings, the pieces of unions, coatings and hollow members; the ridges of mesh files; the sharp edges through the meshes derived from one; the configuration key; two overlapping spheres (the circle of their groove, the exact volume of their union with curved elements, the 4-node mesh with and without it), options, one phase alone, errors and the fallback; three spheres and their two triple points, a groove through the faces of a periodic box, coated spheres against the exact volume of their interphase, crossing fibres (slow) |
|
sections computed on their planes: the slices of the voxel images at the same resolution for every model, without the volumes; the coverage of the pixels and the blended images; the structures of several phases (the fraction of each phase in each pixel from its implicit function: they sum to one, the phase of the centre has the largest, three bands against their exact shares); the measures of the level sets against closed forms |
|
PNG round trips and colours, formation histories, figures and |
|
the registry of the tools (catalog and forms, the items their traits offer, the runs kept in the workspace, the tabs in their sections, the menus of the tabs: their headings and the tools under each, the long forms in boxes) and the tab Statistics: RVE sizes by the ensemble and window methods, measures against the theory, correlation functions, stereology, the RVE fit of pasted tables, reconstructions, the orientation map tool (numbers, tables, figures, the polycrystal that joins the workspace), the percolation threshold of the model of a morphology (its bracket, the probabilities across 1/2, a missing parameter refused), its anisotropy (the 13 lattice directions of spheres as bars, 6 sampled in the plane of a planar morphology as a rose; the table, the ratio, the eigenvalues), its critical exponent (the means of the boxes at a threshold given, drawn on logarithmic axes and tabulated, the conductivity and the elastic modulus; a threshold extrapolated from boxes too small refused), its rigidity threshold (discs of a planar morphology over box sizes: the shares that bear and the moduli against the fraction, the finite-size scaling; the same by the voxels assembled and factorized, in fewer iterations; the elastic exponent at it when no threshold is given; refused without sizes) |
|
the curves of the runs of a loading path and of material points on one figure (the strain that varies most, the plastic strain along the time), their values and gaps to the points of a test, the runs a form of several items refuses (no curves, not a run), the tab Comparisons in its section |
|
the tab Learning: its form generated from the sections of |
|
the behaviours of the workspace: kept by a material point (its run shown again, run again without a second behaviour) and an import of Z-set (saved and opened), their MFront sources in their folders, taken by a loading path (the same response as the behaviour of its form), refused by the NTFA when not von Mises, given to one phase of the inputs of AMITEX_FFTP (its MFront law), to the MFront sources, to the bulk of a cracked plate and a phase of a cracked morphology; a crack grown through hard-core spheres in a plastic matrix (J at the tip at the toughness, a rising load, the plastic work); a behaviour identified on the curves of a material point and on the points of a test, on two material points followed together (a tension and a cycle in shear: the isotropic and kinematic hardenings), weights one per run, runs all in small or all in finite strains; Gurson’s q1, q2 on two material points of a porous behaviour (two triaxialities), an identification started from a behaviour of the workspace |
|
the checkpoints (raised in the thread asked to stop only, the innermost stop, not an |
|
the 3D view: closed, outward and measured surfaces of a level set (the area of its measures), budgets of triangles, voxel faces, exact facets of polycrystals and voxel faces of grains, mesh faces with their fields on each side, displacements and sections, voxel images of fields, |
|
exact cells (volumes, lattices, periodic pieces), regularization (closed valid pieces, volumes), constrained triangulations of faces, polycrystal meshes (grains, matching faces, exports), hard cases of the face meshing |
|
point-wise level sets against the grid (compositions, partitions of several phases), the split test of the octree, conforming octree tetrahedra (periodic faces matching), adaptive meshes of spheres (interface nodes on the sphere, periodic faces matching), curved 10-node elements, thin sheets, exact convex pieces (volumes, nodes on the planes of their pieces), exact cells, greedy boxes, the routes, |
|
rotations of stiffness and conductivity, cubic and hexagonal crystals, orientation averages and bounds, FE and FFT polycrystals |
|
runs in processes, split and merged, resumed and under MPI give the same results; records and status; job scripts of the six schedulers; submission chaining the merging job |
|
AMITEX inputs: ten-line VTK headers, big-endian fields and zone files, an independent reader (VTK), the laws and constants, the stiffness of every crystal voxel rebuilt from the exported axes against morphoGen’s, loadings, paths and finite strains, the errors (plasticity, temperature, non-orthotropic crystals); the runs of finite strains (Small_Perturbations, nine components, every phase an MFront law, its function and coefficients, the elasticity of the nonlinear phases, the errors, the results read in finite strains); results of the cell, materials and zones in morphoGen’s orders, numbers written by Fortran, finite strains, the apparent stiffness; |
|
periodic equations against affine fields, FEniCSx XDMF tags, DAMASK quaternions and files, Neper files, XDMF images and chunked voxelization, outputs of runs; the materials of Abaqus inputs of tetrahedra and voxels (sections per grain with their orientations, per phase without; in finite strains MFront laws, no orientation, the axes of the grains in *INITIAL CONDITIONS, the brick’s expanding laws refused, a porous kernel’s taken), of the outputs of runs (the 21 anisotropic constants in Abaqus’s order, expansion and the initial temperature of the nodes, density, voids, MFront laws) and, with CalculiX (ccx), the stresses of an isotropic and a turned anisotropic grain under a homogeneous strain, also expanding and heated from the initial 293.15 K of the input, against morphoGen’s |
|
chords of Boolean models, Euler characteristics of known shapes, granulometry and local thickness, geometric tortuosity, registered measures |
|
set operations against their components, theory of unions, model errors, grain-boundary distances against brute force, inclusions in a polycrystal homogenized with its crystals |
|
plasticity parameters and loading paths, the material point against closed forms, the consistent tangent against finite differences, homogeneous elements and an exact periodic laminate along cyclic mixed thermal paths, the condensation of linear behaviours, divided steps, the prediction of a step (exact at the first order for a straight line, at the second for a parabola with unequal steps; the choice of the order), the secant of the condensed block (the secant equation, BFGS that stays positive definite with Powell’s damping, Broyden), |
|
nonsymmetric consistent tangents: which laws are symmetric and the asymmetry of the others, GMRES on dense nonsymmetric systems (CPU and GPU), its deflated restarts against the plain ones on an indefinite system (fewer products, the same solution; without restart the plain GMRES to the bit), GCRO-DR’s recycling against them along a sequence of systems (the same solutions, fewer products; empty and to the bit until a solve restarts; CPU and GPU), the stop of a GMRES that stagnates (an indefinite system with more eigenvalues near zero than it keeps), the full tangent operator of the voxels, its coupling and |
|
slip systems and Franciosi’s interactions, configurations, Schmid’s law along [001] and [111], consistent tangents (small and finite strains) and warm starts, large steps, points integrated together and alone (the same to the bit: slip, cleavage, twins that turn the lattice), homogeneous single crystal elements, a polycrystal (Sachs and Taylor estimates, lattice rotations, uniaxial Cauchy stress), the template through the pipeline and the command line |
|
eigen-decomposition and divided differences, tangents and objectivity of the logarithmic strain space, Hencky’s tension and simple shear, the small strain curve of von Mises plasticity in the logarithmic space, finite element paths, macroscopic measures |
|
loading paths on voxels: the tangent of B-bar, F-bar and full integration against finite differences and its symmetry, homogeneous images against the material point (von Mises, single crystals in small and finite strains), the linear FFT solution with full integration (pores included), an exact laminate, no locking with B-bar and F-bar, a polycrystal (Sachs and Taylor, uniaxial Cauchy stress, lattice rotations), divided steps, |
|
MFront behaviours: materials and configurations (sources relative to the file), |
|
MFront sources of morphoGen’s behaviours: the brick of von Mises plasticity (and of Hill’s criterion, orthotropic: compiled for MGIS, called as AMITEX and Abaqus call it), their thermal expansion in small and finite strains (von Mises, Hill’s orthotropic one, a turned crystal’s, Hencky’s elasticity) against the kernels while the temperature changes and its refusals, porous plasticity (GTN of the brick: its criterion, coalescence, constant nucleation, f0 until the matrix flows, the internal variables; the others with morphoGen’s kernel in the source: Rousselier’s and the elliptic criteria, the Gaussian and exponential nucleations, thresholds, other effective porosities, q3 > q1², kinematic hardening, their row of parameters and internal variables; the refusal of an anisotropic expansion), the slip systems of crystals in TFEL’s components, the orthotropic constants in MFront’s order, the internal variables, the errors, |
|
MFront sources of the learned laws: the sources of the four models (their scales and trained parameters as constants, to the bit; Hencky’s strain measure for the laws of |
|
the choice of the device and its errors (API, configurations, command line); on a GPU, the linear FFT solutions and iterations of the CPU, the behaviours of the points against the Numba kernels (every behaviour, Hill’s criterion included, small and finite strains, substeps), the porous points apart (GTN with its nucleation and coalescence, viscous GTN, Rousselier, the elliptic criterion), the CUDA measures, conjugate stresses and tangent operators of the voxels (B-bar, F-bar, full integration) and their coupling columns against the CPU kernels, paths of polycrystals and of a viscoplastic composite with their fields |
|
random laminates (layers, holes, level set, renewal variance), dead leaves (same-leaf probability; families, spheroids isotropic and aligned, finite coverage over a background against the bivariate laws, their limits, round trips; transparent dead leaves against the moments and covariance of their field, opaque leaves as the dead leaves), Kelvin cells, Laguerre tessellations of packings |
|
Lubachevsky-Stillinger packings (no overlap over all the pairs, the fraction, contacts of g(r), minimal gap, jamming, polydisperse and non-periodic), the cherry-pit model (its limits and bounds, cores apart, dense cores), the reconstruction (Debye media, a Boolean model with and without the lineal path, planar, from an image file, errors), stealthy hyperuniform germs (S(k) zero on the constrained wave vectors, the regular Voronoi cells) |
|
foams as level sets: struts on the edges of Kelvin cells, the position along them, their thickening towards the nodes, the curvature of their sections, a solved fraction (percolating, the JSON geometry), closed cells (a pore per seed), the copper foam of Degdeg & Dirrenberger (fraction, ligament thickness and its spread, elongated pores), the calibration of round seeds 20 % too small on the sections of a foam of known seeds (slow) |
|
(FEniCSx installed: WSL) diffusion and swelling of meshes: Crank’s slab, uniform free swelling (no stress), a bilayer strip bending in time to Timoshenko’s curvature, curling into an arc in finite strains (curvature, shortening of the chord) |
|
(FEniCSx installed: WSL) the stresses of an oxidized epoxy plate against the free laminated plate (skin in tension, core in compression) |
|
random paths (isochoric, seeded, free components, rates), rotations of Voigt components; the gradients of the models (three variants of the standard one) and their tangents against central differences (and |
|
the plastic parameters of a cycle recovered from values 50 % off, the Young modulus from a curve of one component; the isotropic and kinematic hardenings of a tension and a shear cycle followed together, the weights of the references (whatever their numbers of points); Gurson’s q1, q2 and the initial porosity of a tension and a uniaxial strain, the parameters of porous plasticity named (matrix, criteria, nucleation, tables refused); a stop during an identification, the errors of the parameters and references (names, terms, Norton flow, positive starting values, components, weights); Hill’s coefficients on three tensions and a shear; the macroscopic law of a Hill material taken as a morphology (its nine parameters from the estimated law), a von Mises law of two loadings, the errors |
|
Hill’s criterion: all its coefficients 1 against von Mises’s radial return (cycle of tension and shear, isotropic and kinematic hardenings, rate-independent and viscous), the yield stresses along the axes and in the three shears of distinct coefficients and the ratio of the lateral plastic strains (closed forms), the consistent tangent against finite differences, homogeneous voxels and elements following their material point in small and finite strains, the NTFA refusing it, its MFront source an orthotropic behaviour, its configurations |
|
the NTFA: exact for uniform fields (plastic and viscoplastic, load-unload), its reduced operators (the apparent stiffness, Hill’s reciprocity, symmetry), a stiff sphere in a plastic matrix against full fields (tension, cycle, shear; free lateral stresses), the local fields, storage, the mean fields of the phases, configurations and |
|
the symmetry group of the honeycomb in a 2 x 2 cell and its irreps (the axial classes of Table 3 of Azulay & Combescure), Euler-Bernoulli frames (a cantilever, equibiaxial forces, symmetric stiffness, the primitive cell at pi^2, four lattices), Timoshenko beams (a cantilever and a fixed-fixed beam deflecting with shear, Engesser’s load, the Euler-Bernoulli limit, a strut in space, Cowper’s coefficients), the first bifurcations of the honeycomb (triple in T2 under equibiaxial compression, double under uniaxial), the invariance of the patterns, the figure; the Bloch-wave stability of the primitive cell (the lowest load at the M points, that of the 2 x 2 cell); continua: the symmetries of meshes (O_h, a 2 x 2 x 2 cell of 384 elements, a random image), a lattice of bars (the axial classes, a pattern per class, invariance on the whole mesh, the projections of the first modes, VTK output), a column against Timoshenko (slow); frames of struts in space (an oblique cantilever: Euler’s load, deflection, twist; the groups of four lattices, equivariance, hydrostatic forces; the octet at rho/9 and nu = 1/3, the cubic lattice at rho/3, Kelvin at 0.6 rho^2, the bcc lattice bending-dominated), the bifurcations of the cubic lattice (Euler’s load of its columns clamped with sway, projections, invariant patterns, Bloch waves, 3D figure and video), Schoenflies symbols of 21 point groups; along nonlinear paths in finite strains: a slender column within 1 % of its linear buckling (same irreps, four modes), a square array of holes in uniaxial compression (a simple point of period doubling, alternating ellipses, the bifurcated branch: growing pattern, plateau, negative Poisson’s ratio), an elastoplastic column bifurcating earlier; the pattern transformation of an array of holes on a mirrored conforming mesh |
|
porous plasticity: MINRES on a symmetric indefinite system and after the conjugate gradients, the paths of a porous composite with held lateral strains (softening points) without a stalled solve nor a divided step, on the CPU and the GPU; GTN’s hydrostatic yield and the von Mises limit, the porosity and the plastic volume change, Chu and Needleman’s nucleation, the yield of compaction along the densification (small and finite strains), tangents by central differences, objectivity, Norton’s flow, the Jacobian of the update against central differences (each criterion, the nucleations, Norton, Rousselier’s vertex), GTN’s matrix stress bracketed from a guess (nearby or a thousand times off) against the root from its estimate, finite element and FFT volume elements, configurations; kinematic hardening: its configuration and consumers (the form, the identification, the NTFA refused), von Mises’s kernel at f = 0 along a turning cycle, the Bauschinger effect, Prager’s hydrostatic backstress and the yield on the shifted surface, tangents (small and finite strains) and the Jacobian against central differences |
|
Z-mat material files (the examples of its manual): native translations, MFront sources (orthotropic Hill, several potentials, user-defined flows, softening), porous_plastic files (tables, nucleation, initial porosity), errors; Hill’s coefficients in TFEL’s convention on random stresses; configurations and |
|
cracks of any surface and front: the level sets and axes of a surface crack against the penny, straight and kinked cracks, growth by rows, the mesh refined around the front; an inclined penny crack against Kassir & Sih (the three modes along the front, the angles of kinking), an elliptical crack against Irwin; the growth of each point of the front (an ellipse towards a circle, an inclined crack towards local symmetry); a straight front as an open surface crack |
|
crack level sets and cut tetrahedra (volumes, sides, refinement), the energy release rate of an edge crack by XFEM (4 % of the handbook) and on conforming meshes (converging), 10-node elements with quarter points at the front (within 2 % with elements of 1, where the 4-node ones are 24 % low) and the consistent loads of faces of six nodes, the remeshing by MMG and the opening of its crack faces, fatigue growth by Paris’ law (factors within 4 %) |
|
the isoparametric 10-node tetrahedra (volumes of curved elements, their matrices against the formulas of the reference element, patch tests under the four boundary conditions, plastic paths against the material point) and the meshes whose nodes lie on the interfaces: the functions of the phases at any point, the volume of a sphere, the conductivity of a cubic array of spheres against Sangani & Acrivos (1983), a coated sphere on its two interfaces, random coated spheres with their grooves followed or rounded and the faces of periodic meshes of three phases (slow), the options |
|
the mean dilatation of the 10-node tetrahedra (integration: bbar): its names and options, a homogeneous nearly incompressible solid and a rubber along a path against its material point (small and finite strains), the tangents of B-bar and F-bar against central differences of the forces and of the macroscopic stress, the linear matrix as the tangent of the paths, Lamé’s thick-walled cylinder at a Poisson’s ratio of 0.4999, stiff spheres in a nearly incompressible matrix (slow) |
|
the laws (energies of the modes and of their mixture, secant unloading, compression, shear directions), the insertion (an inclusion detached, a crack front kept closed, the orientation of the normals), a bar debonding at N A with the work G A in small and finite strains, frame indifference in finite strains, the calibration formulas and the corrected beam theory, the debonding of particles in a composite, the double cantilever beam against the beam theory (slow); between 10-node tetrahedra: the insertion (each lip its mid-edge nodes, the front its own), the bar with the loads of the 4-node elements at every step, the area and the translation force of the curved interface of a sphere, the beam with two elements across each arm (slow) |
|
the visualization section (defaults, names, 30 errors), the dependencies checked before the run, the whole chain on voxels (FFT: figures, videos, the fields of a path at every step, the files of the fields at the ends of the segments only, tables as .txt and .xlsx equal to the .csv) and the template |
|
tables (csv, txt, xlsx: records, columns, arrays, sheets, NaN, booleans), figures (png, tif, svg, pdf) and TikZ (axes, legends, explicit tick labels, images, fills; compiled by pdflatex when present, slow), video sequences (once, loop, bounce) and writers (gif, PNG frames, mp4), sections and orthoslices, VTK turntables, sweeps and fields, the entry point |
|
consistent mass matrices (4 and 10 nodes) and the path of the Brillouin zone, a homogeneous cell against the folded acoustic branches, a bilayer against Rytov’s relation (ten bands, the first gap along Γ-X), the figure; the scalar waves: layered fluids against their exact dispersion and Wood’s law, the TM gap of alumina rods (Joannopoulos et al.), the sound slowed by rigid cylinders |
|
the transport phenomena of a scalar potential (electrical, dielectric, magnetic, diffusion): the thermal solvers to the bit, layers, inert phases, bounds, loadings with a physics, RVE studies of their scalars and tensors, configurations and the command line |
|
strong couplings: the laws and their rotations, layers against the exact laminate of every coupling, a product property, the uncoupled limits, the dense system of a small grid, uniform thermal strains and Levin’s relation, thermoelectric composites, crystals turned by their grains, configurations and RVE statistics of the constants |
|
cleavage cracks that close and twins that turn the lattice: a crystal pulled, pushed and pulled again (the plane closes by its flow rule, carries the pressure as the sound crystal, opens again at the strength left; the work of the cycle); the friction of the lips of a closed plane and of a sound one under pressure; twins that shear back to none (the fraction against the strain, there and back); the disorientations of the twins against their closed forms, the stiffness of a turned lattice; a lattice that turns at a threshold and turns back (magnesium off its axis c, small and finite strains; a cubic crystal sheared along one twin: no strain left); the configuration and its errors; the consistent tangents against finite differences (a plane that closes, held closed, whose lips rub; twins that shear back, undone; a turned lattice and the step that turns it) and the warm starts; homogeneous cells against the point (voxels, mesh); a textured magnesium that untwins; the same without detwinning and with turned lattices, a cracked polycrystal of iron that closes (slow); twins that shear back and planes that close at a share of their resistance (the flow rules, the tangents); the rule that turns the lattices of a grain on points of given fractions, a material point (the constant threshold), a grain of 64 voxels and a polycrystal whose turned volumes follow their twinned volumes; the basal plane of a turned lattice under the criterion of Beremin |
|
cleavage that the slip nucleates: a crystal that slips at its yield stress until the strength of its plane has fallen to it (the onset, the flow rules of the slips and of the opening); the damage of the plane that eases the slip in compression; the configuration and the bound of the coupling; the Jacobian of the local residual and the consistent tangents against finite differences (small and finite strains); homogeneous cells against the point; a polycrystal of iron that breaks sooner (slow) |
|
cleavage planes regularized by their microdamage: the closed form of the band of a plane that breaks and its parameters; the configuration; the law of the planes at a prescribed microdamage (one plane, two planes, a plane broken through, a point alone) and its consistent tangent (small and finite strains); the equation of the microdamage on voxels and on meshes against the closed form of a slab; a bar that breaks across its band (softening branch, fields, work against the fracture energy, planes broken through), on grids of three sizes (slow) and on meshes; homogeneous cells against the point; the solvers that refuse; a columnar polycrystal (slow); the acceleration of the alternate scheme (the updates for maps with a known fixed point; the bar with the plain scheme and with the accelerated one) and its options (solvers, configuration) |
|
the criterion of cleavage of Beremin: the largest principal stress and the normal stress of the cleavage planes of turned crystals against closed forms, the Weibull stress (a uniform stress, two regions, compressed points, large moduli), the weakest link; a homogeneous cell pulled, unloaded and pulled again (the plastic zone, the probability of the largest Weibull stress); configurations and their errors; the pipeline (curves, field, planes against the principal stress) |
|
twins and cleavage of crystals: the slip systems of the hexagonal lattice (numbers, closure under its rotations), the twins and their shears against the formulas of their modes (the second plane undistorted, the sense of {10-12} on either side of c/a = √3), the systems of the cleavage planes, configurations and their errors; at a material point: twins along the axis c of magnesium and of zinc (one sense, Schmid’s law with the barrier of the saturation at each step, the strain of the twins, the saturation), an fcc crystal that twins at half the stress in compression, basal slip; a plane that opens at its strength and softens (σ = Y₀ exp(−w δ), the work Y₀/w, unloading, compression), an inclined plane and its accommodation, slip that shields cleavage; the consistent tangent of a crystal that slips, twins and cleaves against finite differences; homogeneous cells against the point (voxels, mesh, small and finite strains); the fracture energy of a crack band; a fibre texture; a textured magnesium (asymmetry); transgranular cleavage of a polycrystal of iron; cracks between the grains or through them (cohesive boundaries and cleavage planes) |
|
multiaxial high-cycle fatigue (the pores of a voxel image are not judged): the amplitude of a cycle against closed forms (proportional cycles, a tension and a torsion a quarter of a cycle apart, the circles of three states of shear, a cycle of pressure, arrays of points); the criteria of Crossland, Sines, Dang Van, Findley and Matake at the two limits that calibrate them, under a mean stress, under a pressure held, in any frame; criteria from mappings and their errors; points judged by phase, the factor of safety and the share beyond the limit; the fields of a cycle of a homogeneous cell at the closed form (finite elements and FFT), around a stiff particle; the configurations and the pipeline (the map, the results, the quantities of the statistics); Goodier’s stresses at a spherical pore (slow) |
|
the damage laws coupled to plasticity, of concrete and the unilateral effect, at a material point against closed forms: Lemaitre (a perfectly plastic and a hardening bar, a bar of Hill’s criterion, the triaxiality, the closure of the microcracks, cycles, a point that breaks, the delayed damage), Kachanov (the time to rupture, the damage and the tertiary creep, Hayhurst’s stress), Mazars (tension, compression, the shares), the spectral split (the stiffness recovered in compression); their tangents against differences; configurations and errors; a composite whose matrix is damaged, by FFT in small and finite strains |
|
viscoelasticity, hyperelasticity and damage: material points against closed forms (Prony series, incompressible rubbers, softening curves), tangents against finite differences, the laws in the solvers (exact laminates by FFT and finite elements), the crack band independent of the voxels, a crack in a hyperelastic plate, the identification of a Prony series |
|
harmonic regimes: complex laminates, Dykhne’s checkerboard, static and instantaneous limits, the harmonic regime followed in time, Maxwell-Wagner polarization, effective indices, laws in a half-plane, conjugate orthogonal gradients against a dense system and MINRES, RVE statistics |
|
extended finite elements of level sets: sub-tetrahedra, oblique laminates to round-off, spheres against voxels and estimates, voids, transport, loading paths with damage, configurations ( |
|
phase field of brittle fracture: homogeneous closed forms of AT1 and AT2 with each split, optimal profiles and areas of a crack, the fracture energy of a bar on two grids, compression, a row of slits against Griffith, unbreakable inclusions and voids, |
|
phase field of fracture in finite strains: a cell of rubber and of an elastic solid stretched to 1.5 against the closed forms (damage, nominal stress, energies; uniaxial strain and uniaxial stress), a compression that breaks only without a split, a cracked plate at small strains against the solver of the small strains, a cracked strip of rubber that breaks at the stretch of Rivlin and Thomas, the phases and options refused, |
|
light in three-dimensional photonic crystals by plane waves: the light cones of a homogeneous medium and of a uniaxial crystal to round-off, the Bragg mirror against its exact dispersion at normal and oblique incidence for both polarizations (second order), magnetic layers and the duality of the fields, interfaces inside the cells of the plane waves (the averaged tensor), rods against the two scalar polarizations and out of their plane, long waves against the homogenized permittivity, the band gap of the diamond lattice of spheres (Ho, Chan & Soukoulis; MPB), the Davidson solver against the matrices, the command line |
|
weak couplings: Levin’s relation for the swelling of two phases, layers by finite elements, the relations of Biot and Gassmann for pores in an isotropic solid and in a crystal, the tensor of Biot of a heterogeneous solid by reciprocity, finite elements and level sets, undrained limits, configurations, RVE statistics and the command line |
|
Stokes flow through the pores by the lattice Boltzmann method: the flow between two planes exactly, whatever the viscosity of the lattice, a square duct, square arrays of cylinders (Sangani & Acrivos; Drummond & Tahir), simple cubic arrays of spheres (Zick & Homsy), closed pores, the permeability in |
|
acoustics of porous media: the thermal permeability of slits and tubes, the tortuosity and the viscous length of arrays of cylinders and spheres (Rayleigh, Maxwell), closed and narrow pores, the equivalent fluid against the exact responses of slits and tubes and its limits, the absorption of a layer, the property |
|
the tab Physics of the interface on small voxel images: its forms, the transport properties between their bounds and with an inert phase, the flow through the pores with the acoustic parameters and the absorption curve of a layer, the dynamic moduli of a viscoelastic composite (the loss peak, the limits), the errors of the forms |
|
the guide to the physics (docs/physics.md): its Python snippets run in one session and give the physics they announce (bounds, couplings, loss peaks, band gaps, softening and broken cells, a filled rubber and a strip torn in finite strains), its RVE configuration runs, its commands parse |
|
the tutorials (docs/tutorial_texture.md: the grains and the texture of the map, its stiffness, the lattices that a tension turns, a Taylor polycrystal, a polycrystal of space with the texture of the map; docs/tutorial.md, tutorial_rve.md, tutorial_rubber.md, tutorial_geometry.md, tutorial_iga.md, tutorial_contact.md, tutorial_fracture.md: layers that debond at their strength and with their fracture energy, a pore under a cyclic load, magnesium that twins one way, iron that cleaves, the probability of Beremin): the snippets of each run one after another at reduced sizes and give what the page says (measures against theory, the size of the RVE; the variance of the volume fraction of Boolean spheres against their integral range, the exponent 2/3 of Poisson fibres, the windows of one image, the report of a study; the laws fitted to the tests of a rubber and the stretch at which its strip tears; the fractions of a conforming mesh of three phases and the moduli of three solvers; the orders of an isogeometric analysis, the arrays of holes, the hole of Vigdergauz and the auxetic sheet); the sizes replaced are in the pages, the figures shown are in |
|
the recipes (docs/howto.md): their snippets run in one session and give what they announce; the |
|
the gallery of examples (docs/gallery.md): a card and an image for every template, in the families of |
|
(FEniCSx installed: WSL) the FEniCSx solver against morphoGen’s: stiffness and expansion of random P1 and P2 meshes under PBC, KUBC, SUBC, MBC, phase statistics, conductivities, |
|
the solver names, configurations, the errors without FEniCSx or on voxel images |
|
thermal oxidation in the voxels of an image: one voxel against the thin film, a plate against its one-dimensional profile (second order), a block and impermeable fibres along the diffusion against the plate, barriers and inert skins, units, exposed faces, the tracer of the ageing laws, the refusals |
|
transient diffusion against Crank’s slab (uptake, profile, sqrt(t) start), a periodic plate against the line, solubilities; thin-film oxidation against Ernault et al.’s amides and oxidized layer; property laws, graded phases and the mean eigenstrain; Ekenstam’s hydrolysis; a bilayer against Timoshenko, the bending sequence of a hygromorph, Crank’s uptake of a layer; the |
|
Choquet capacities of Boolean spheres (balls, segments along x and (1, 1, 1), pairs) and Poisson fibres (minus sampling) against their closed forms, dilations and the capacity, openings and closings, the morphological covariance of a Gaussian field, Minkowski functionals of Boolean spheres, of a ball and of Gaussian excursions (Adler), Betti numbers of a torus, a shell, two balls and joined tori |
|
the MIL fabric tensor of isotropic Boolean spheres (degree of anisotropy, the MIL along the axes against the mean chords), of the same voxels on a stretched grid (ratios 2 : 1.5 : 1 and their axes), of an elongated level-set foam (both phases); the first minima of the covariance of an egg crate (half the periods, exactly), of lamellae and of a dense level-set foam; the measures |
|
the affine anisotropy of the models: its configuration (mapping, 3 or 6 numbers, an elongation, errors, the periodicity rule) and the same GUI field everywhere; for each mapped model, periodic images equal to those of the isotropic model in the preimage box (labels, JSON round trips, the invariant theory); correlation lengths (1 : 1 : 2 within 4 %) and mean chords (5 %) along the axes of Boolean and hard-core spheres, Voronoi cells, dead leaves and the Gaussian excursion, the volume fraction kept; the Gaussian covariances along the axes against theory and a metric turned in a periodic box; turned axes in a non-periodic box (fabric tensor, exact point queries, labels); the identity and the open foam’s two forms |
|
the particles of a phase in a volume: two overlapping balls cut by the plane of their neck against the volumes of their caps, across the faces of a periodic image; a packing of hard spheres recovered sphere by sphere (count, centres, radii); the frame of windows that are not periodic (Miles-Lantuéjoul); discs of a planar image against the sections of the stereology; the kernels of the watershed of a volume; the named measure |
|
Saltykov’s matrix and the scales of ellipsoids, the unfolding of exact sections of spheres and of turned ellipsoids in three directions, the shape consistency against the values of Degdeg & Dirrenberger, sections of voxel hard spheres, the watershed of two discs, the frame weights of non-periodic windows (exact on average) |
|
the lineal-path function (exact counts, the Boolean model), the pore-size function of the Boolean model, the two-point cluster function, the radial distribution function and the structure factor (Poisson points, hard spheres, a lattice) |
|
arcs (radii of curvature of consecutive points, chords, end-to-end distances, distributed radii, aligned middle tangents), helices (radius, pitch, end-to-end distance), curved fibres without overlap (no two segments closer than the gap over the nearby pairs and their periodic images, the fraction; in a box), hollow fibres (lumens on the axes, walls, solid fractions, storage) |
|
closest points of segments; hard-core fibres (no two closer than the gap over the periodic images, fraction, aligned axes, non-periodic windows, jamming); curved fibres (tangent correlations, mean square end-to-end distance, fractions of straight and tortuous fibres) |
|
the quadratures of the orientation laws (moments against draws, round trips), the von Mises-Fisher and mixed laws (Kolmogorov-Smirnov, proportions, errors), the law of the angle to a direction; the covariograms of cylinders and spherocylinders (K(0) = V, across and along the axis in closed form, ∫K = V², an oblique shift against points drawn in the capsule), the lens of two disks; the dilute integral range E[V²]/E[V] (isotropic and β laws, random sizes); the covariances of Boolean fibres (β law about (1, 1, 1), round and flat caps) and of Poisson lines (girdle, von Mises-Fisher) measured along x, y, z and the mean fibre axis against the closed forms; Poisson lines of any law (isotropic unchanged, aligned lines exactly, the count of axes, the Choquet capacity of segments); the curves of the interface along each axis |
|
the sparse periodic problem of frames against the dense one of the four lattices of struts (and from their frameworks) to round-off, the honeycomb (Masters & Evans) and triangular frames against their closed forms, of Euler-Bernoulli and of Timoshenko walls (the shear compliance of the walls, the energy split, the struts in space against the dense problem, shear coefficients of frameworks); beams with rigid ends at their joints (the triangular lattice in closed form, random frames against their beams split with stiff ends, in space and in the plane, the cubic lattice at l/l’, the lengths, the shortest beams kept flexible, the dense paths refused); frames of frameworks (sections, dictionaries, plane networks, Voronoi frameworks, errors, mechanisms); slender Delaunay networks against an independent truss, stretching-dominated at 0.82 of the affine estimate with its Poisson’s ratio 1/4 (0.76 and 0.38 in the plane); open foams bending-dominated (E ~ phi^2, coefficient about 1); the softening of non-identical beams (Ban et al.): exact derivatives against finite differences, replicas against s CV^2 and v CV^2, the laws of the factors; for Timoshenko beams whose shear stiffness follows the area, the derivatives of the four modes (the concave chord stiffness) and replicas of the area; the frameworks and frame stiffness in the summaries of open_foam, delaunay_network and random_lattice; the frames of their realizations ( |
|
the tab Mechanics: the loading path of a morphology (FFT on its voxels, finite elements on its conforming mesh); crystals of the forms that twin (magnesium along its axis c) and cleave (iron along [001]; a matrix that cleaves along a path); the criterion of Beremin along a path (the Weibull stress of the field written, the figure, the configuration); its interfaces that debond (strong ones: the bonded phases; weak ones soften the response; their fields and the configuration that runs them again) and the fatigue of its cycle (the fields of the criterion on the stresses of the steps, the figure, the limits of the form), the reduced model of the NTFA against the full field, the material point, the linear tensors (the stiffness and its compliance, S C = I; by extended finite elements on a grid cut by the level set: between the bounds, within a few percent of the FFT solver, Levin’s relation, what they refuse); the stiffness of the frame of beams of the networks of the workspace (the items of the trait framework, the spheres refused plainly; its values, tables of C and S, figure and file against |
|
scale-rich networks: the convex polygons of the growth (clipping with the faces of the edges, chords, strips that fit); the first length in closed form against the integral of l² over the lines and 4,000 first ligaments; the exact partition of the box by ligaments and voids, the T-tessellation of the thin network (N + 1 voids of four vertices on average, Euler), each end on a face of its neighbour, the degrees of the dual graph; the thicknesses exactly, the mean-field density within 3 % of realizations; no jamming at 0.75 of the jamming thickness, jamming at 1.25; the tails of 30,000 ligaments (lengths and degrees near 3); aligned, spread and polarized angles (lines through the pole or across it), linear and radial nucleation densities against their exact means; voxels, partial volumes, the level set (the distance to the ligaments near them, or less), point queries, JSON round trips, prisms along x, the frame; the solved volume fraction (within 1.5·10⁻³, with a frame, aligned), its refusal beyond jamming; the parameters, the template and the interface |
|
scale-rich networks in 3D: convex polyhedra clipped by planes (volumes, the hexagonal section of the cube, faces oriented and convex after 300 cuts, volume kept to 10⁻¹²); the mean inverse distance in the cube against its closed form (Bailey, Borwein & Crandall) and the first plate area against 20,000 sections; plates: the exact partition into N + 1 closed voids, thin plates their rims times their thickness, rims in their mid-planes ending on earlier plates or the box, the links and degrees; the density within 6 % of the mean field |
|
power-law tails by maximum likelihood: Pareto laws above a bound and the zeta law of integers within three standard errors, the error of the discrete estimate against the variance of ln x under the law, the bound of a tail above a uniform body by the Kolmogorov-Smirnov distance, the bias of the continuous estimate on integers, the errors |
|
the periodic Delaunay triangulation against that of all the images (2-D, 3-D), Euler’s E = 3V on the torus; the coordination 2 + 48π²/35 and the mean fibre length of Poisson-Delaunay graphs (in space and in the plane, 32/(9π) λ^(-1/2)), hard-core points (lower coordination, more even fibres, least distance); the fibres as struts (a solved fraction, the framework, JSON round trips, planar networks, windows, errors); the template and the form of the interface |
|
rotated lattices (struts along the turned axes, fraction), graded lattices (interpolated radii, mean and local fractions), their errors |
|
the public interface ( |
Principles:
Statistical tests use fixed seeds; tolerances are a few standard deviations computed from the theory (e.g. σ = sqrt(p(1−p)A₃/V)).
Exact checks against brute force are used wherever possible.
No cross-platform bitwise comparisons: floating-point details (e.g. fused multiply-add on arm64) may flip voxels lying exactly on a boundary.
Optional checks run only when the tools are installed (VTK, Gmsh, scikit-image, tifffile, scikit-fem; MMG for conforming meshes; CuPy and an NVIDIA GPU for the GPU backend).
Continuous integration (.github/workflows/ci.yml) runs ruff, then the tests on Ubuntu
(Python 3.12 and 3.14), Windows and macOS on Apple Silicon (Python 3.14), in parallel
(pytest-xdist, one file per process: tests/conftest.py starts the longest files first
and gives each process its share of the cores), and builds the documentation site with
its warnings as errors (kept as the artifact site of the run), for every pull request
and every push to main;
tools/ci_local.sh runs the same jobs on the development machine (D18). The slowest
tests (marker slow: the numerical checks of more than about 15 s each, two thirds of
the time) run in one job, Ubuntu with Python 3.14 (locally the last Python version),
the others take the fast tier (-m "not slow"). The tests of the interface check what
its tools pass to the library and give back, on small problems; the physics they show is
checked by the tests of the library.
The documentation site (docs/conf.py, Sphinx with MyST and Furo) takes the Markdown
pages of docs/ and the change log, and writes at each build the pages of the API reference in docs/api/
(ignored by git): a package lists the names of its __all__, linked to the module that
defines them, and its modules; a module documents what it defines, its undocumented
public functions and methods with their signature. The pages stand in six groups:
getting started (installation, tutorials, gallery, recipes, questions), the user guide (a
page for each step of a study, from the models to the design, and the guide to the
physics), the applications (laser-architectured sheets, explicit dynamics), running
studies (configurations and commands, the interface, other codes), the reference
(API, configuration keys, theory, Z-mat, glossary, performance) and the project.
docs/index.md is the landing page (the steps of a study as cards, a first study that
tests/test_howto.py runs, where to read what). The README is a short presentation of
its own, whose first study is that of the landing page. docs/_static/ holds the logo, the styles (gallery.css, site.css) and
site.js, which adds to the pages, without any library, a button that copies each code
block, a filter of the cards of the gallery and the enlarged screenshots.
The images of the pages are written by scripts, never by hand:
docs/gallery/make_gallery.py (the cards, images and result figures of the templates,
and docs/gallery.md itself), docs/tutorials/make_figures.py (it runs the snippets
of the tutorials as written and keeps the figures they save, or that a study they run
writes) and
docs/tutorials/make_screenshots.py (it starts morphogen gui and drives the study of
the tour in the page with Playwright: docs/tutorials/screenshots.js).
6. Performance notes¶
Numba compiles the kernels on first use (a few seconds) and caches them on disk (
cache=True).morphogen.set_num_threads(n)or--threads nlimits the threads used by Numba and SciPy.Memory goes mainly into the output arrays: 1 byte per voxel for phases, 4 bytes for labels and fields. Tessellations are labelled in slabs of about 2 M voxels, and covariances use 1D FFTs along each axis.
Automata: 30 updates of a 64³ Moore automaton take about half a second; 4000 Gray–Scott steps on 48³ cells about 0.7 s; nucleation and growth of 64³ cells about 0.2 s; a DLA cluster of 30 000 particles a few seconds (first calls include the compilation).
Topology optimization: about 0.06 s per iteration for 48² planar cells, 0.25 s for 64², 0.5 s for 24³ cells; designs converge in 50–300 iterations. The conjugate gradients compute their dot products in a single thread: BLAS would spread them over threads that compete with those of the compiled kernels, which made mid-size solves several times slower.
Conforming meshes take about 0.2–0.5 ms per element (MMG is sequential); finite element systems of 10⁵–10⁶ unknowns are solved in seconds to a minute per loading, the hierarchy being reused for all loadings of a boundary condition. Realizations are independent: run them in processes (
-j) or split studies into tasks (--task-id,--n-tasks,morphogen submit) to use several cores or nodes.Polycrystals: the exact cells of 300 grains take about 2 s, their regularization 1–2 s, the meshing 15–20 s for 2 × 10⁵ elements (mean quality 0.73–0.74).
Grain-boundary phases cost a k-nearest query (k ≥ 16) per voxel: about 6 s for 128³ voxels of 200 cells. Dead leaves cover 96³ voxels in about 0.1 s; laminates voxelize slice by slice, a k-d tree query per layer.
Chunked voxelization holds one slab of about
max_voxels(2²⁴ by default) in memory.GPU (
device: gpu): the linear FFT solves are 5 to 7 times faster than on 20 CPU threads from 64³ to 192³ voxels (NVIDIA RTX 3000 Ada laptop GPU), the loading paths of a polycrystal 1.6 to 2.8 times and of a porous composite 2.4 to 3.5 times, the integration of the behaviours being bound by the double precision of this GPU (theory notes, §35;examples/gpu_benchmark.py,--pathsfor the paths alone). The first solve compiles the CUDA kernels (NVRTC, the behaviours in a few seconds; CuPy caches them on disk). The integration points of a path take on the GPU the memory they take on the CPU (about 1 kB per point for a crystal with 12 slip systems in small strains).Behaviours of the points on 20 CPU threads, an increment in the middle of a loading (
examples/behaviour_benchmark.py): 10 ns per point for elasticity, 88 for von Mises, 345 for Hill’s criterion, 522 for GTN and 1.4 µs for an fcc crystal in small strains; the logarithmic strain of finite strains adds about 0.25 µs (its tangent from the six directions of dC). The kernels of Hill and of porous plasticity take their work arrays once per block of points (_hill_work,_porous_work); the porous Newton iteration and its tangent take the analytical Jacobian of its equations (_jacobian,_matrix_derivatives), GTN’s matrix stress at the points of the line search bracketed from the current point’s.