relax::move_weights gives each node its share of a displacement (the grabbed
nodes all of it, the nodes within the radius strength x falloff, or the
harmonic function of the element-edge graph, zero beyond the radius and at
held nodes); relax::move_nodes applies it as the relaxation's constraints let
each node move (faces in UV, edges along, vertices held), scaled back by
bisection until no valid element inverts, and reports each element's scaled
Jacobian before and after.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
The D3Q7 scalar lattice steps on the device after the flow's step, in
its own A–B pair of buffers: BGK collision with the velocity the
momentum collision leaves in the moments buffer, pull streaming with
the links' rules read through each group's thermal condition
(anti-bounce-back for a held temperature, bounce-back for an adiabatic
face, a flux added to the bounced population), inlets at their held
temperature and outlets copying their neighbour. The populations are
stored as deviations from w_q T₀ so single precision keeps the digits
of the differences. The heat each link exchanged goes into the fourth
component of the momentum-exchange contributions and the group sums;
the buoyant acceleration −β g θ is read per cell in the collision, the
open boundaries and the moments. A temperature that stops being a
number ends the chunk with Error::Unstable like the flow's checks.
Flow exposes what the backend needs to build the device state (the
thermal settings and populations).
Verification against the CPU on lavapipe: plates held at 20 and 21
(temperatures within 1.4e-5 of the range, the wall heats within 6e-5
and both against k ΔT A / H), a heated channel with a parabolic inlet
carrying its temperature, a flux wall, a held wall and an outlet
(temperature 3e-7, heat 1e-7) and a buoyant cavity at Ra 10³
(velocity 1.3e-6, temperature 6e-7, heat 2.4e-6). The tests now fail
on a shader that does not compile instead of skipping; only a missing
device skips.
Part of #157.
A fluid domain takes `thermal` settings (the initial temperature, a
thermal condition per box face and per obstacle face, gravity with the
expansion coefficient for buoyancy) in model.fluid_domain.create and
.edit, and model.fluid_domain.thermal_boundary sets one boundary's
condition as model.fluid_domain.boundary does for the flow. A material
a domain with heat uses keeps needing its conductivity and specific
heat when it is replaced.
solve.fluid's summary adds the thermal relaxation time, the fluid's
temperature range and the heat into every wall (final and the mean
over the last quarter of the history); solve.probe gives the
temperature in the cells at points and, with `heat`, the walls' heat
histories. The form, the catalogue and docs/features.md follow.
Part of #157.
A D3Q7 passive scalar rides on the D3Q19 flow: it is advected by the
cell velocity, diffuses with the fluid's thermal diffusivity (the
relaxation time follows from k / (rho c)), and feeds back into the
momentum through a per-cell Boussinesq force when the domain has an
expansion coefficient and gravity. Walls and obstacle faces hold a
temperature (anti-bounce-back), a heat flux (added to the bounced
population) or are adiabatic (bounce-back); velocity inlets carry the
held temperature in, pressure outlets copy their neighbour.
The heat crossing each face group is accumulated during streaming and
reported in physical units (W) in the result's heat histories, next to
the force histories; frames gain a temperature field and the stats
report the thermal relaxation time.
Closed-form checks: linear conduction between held plates, a flux wall
against q H / k with the wall's heat balance, an inlet temperature
carried through a plug flow, and a periodic blob that conserves its
total and returns after one lap. The de Vahl Davis cavity (Ra 1e3 and
1e4 against the published Nusselt numbers) is there as an ignored test
because it takes minutes.
The GPU backend declines domains with heat for now and asks for the
CPU; its kernels follow.
Part of #157.
A fluid domain may carry thermal settings: the initial temperature,
gravity and the fluid's expansion coefficient for buoyancy, and a thermal
condition (adiabatic, held temperature, heat flux) on each box face and on
named obstacle faces, adiabatic unless told otherwise. The material then
needs an isotropic conductivity and a specific heat. Boundaries are set
one at a time like the flow ones.
Part of #157.
The dihedral range and mesh.quality's scaled Jacobian were both right:
they are computed by the same function on the same final corners (tet4
and tet10 give identical numbers), but a dihedral range does not bound
the corner scaled Jacobian. A needle has dihedral angles near 60 and 90
degrees and a nearly flat apex corner. Before 417c0c88 sphere-on-plate
at size 4 had dihedrals 36.3-141.6 and 106 needles below SJ 0.2 (min
0.095); now 36.0-137.6 and min SJ 0.351.
The mesher's report now carries min_scaled_jacobian so it cannot read
as all-good while needles remain.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
`FluidStats` gained `backend` in #535 and the fixture of the flow drawing
tests arrived with #550; merged one after the other, the fixture lacks the
field and `cargo test -p fieldcraft-engine` no longer compiles on develop.
Part of #157.
The window built a deformed copy of the model for every results plot,
which drops the geometry surfaces, so after a fluid solve the bracket
vanished and nothing else was drawn.
A fluid step's plot keeps the part as it is and draws the engine's flow
display through it (the same scene as the headless pictures): the
field on a slice, the part coloured by the flow beside it and
streamlines. The legend shows the field with its bands and the cells
of its extremes (Max/Min mark and turn the view to them); the results
panel swaps the deformation rows for the slice's normal and position,
the obstacles' colours, the streamlines and a frame slider, all
running the results.* commands. The status line describes the new
commands.
Part of #193.
After solve.fluid the results view had nothing to draw: a fluid step's
fields are per cell, the view only knew nodal fields, and the deformed
display it built dropped the geometry surfaces, so even the part
vanished.
A fluid step now has its own display (`engine::flow`, on render's
flow display data), shared by the headless pictures and the window:
the part drawn as it is, its faces coloured by the field in the fluid
beside them, the field on a slice through the domain and streamlines
seeded on the inlet face upstream of the obstacles, coloured by speed
when the legend shows velocity. It is first shown with velocity
magnitude on a slice through the middle, normal to the horizontal axis
across the flow.
New commands: results.slice {step, field: velocity|pressure, axis:
x|y|z|none, position, frame}, results.surface {step, show} and
results.streamlines {step, show, count, max_steps, seeds | line,
frame}, which replies with what it traced. results.contour takes
velocity and pressure for fluid steps (a new `pressure` field) with
the usual bands and legend clamp, and results.frame walks a fluid
step's frames. Streamlines leave the coming-soon list. Hostile
parameters are refused, and the counts are capped.
Part of #193.
A fluid step's result is a field per cell of a regular grid, not per
node, so the contour machinery had nothing to draw it with and the
viewport showed nothing after a fluid solve.
Add `render::flow`, CPU display data for such fields with no solver
dependency (render stays at L1): an axis-aligned slice through the
cells as triangles carrying cell-centred values (interpolated between
the two layers either side, outer centres moved to the domain faces,
cells inside obstacles left out), the field sampled at a surface
beside the fluid cells for colouring an obstacle, and streamlines
traced through the trilinear velocity field by the midpoint rule,
seeded on a lattice upstream of the obstacles. Seeds, steps and the
total number of points are capped so hostile counts cost a bounded
time. `contour::banded` fans band pieces into scene triangles for the
callers.
Part of #193.
Every cell did an atomicMax on the same word; a workgroup reduction first
leaves one per 256 cells. Same result, less contention on a single
address.
Part of #529.
The CPU stays the default, so every existing result and fixture is
bit-identical; "gpu" steps on the device FIELDCRAFT_GPU picks through the
registered backend, and is an error when the build has none. The desktop
app and the CLI (hence the MCP server) register fieldcraft-lbm-gpu at
start; the summary names what stepped. Hostile values are errors.
Part of #529.
A new L7 crate: the CPU scheme's collision (TRT with Guo forcing), the
per-link streaming rules (bounce-back with Ladd's term, slip reflection,
open placeholders), the Zou–He boundaries and the momentum exchange per
force group as wgpu compute shaders, stepping the flow fieldcraft_lbm's
Flow prepared, through its Backend trait: populations stay on the device,
a chunk of steps is one submission, and only frames, forces and the
stability check come back. Populations are stored as deviations from
their weights, which keeps single precision honest: the verification
cases (Poiseuille with Λ = 3/16, Couette with wall forces, Taylor–Green,
a sphere's drag, a Zou–He channel) agree with the CPU to 2–5e-6 in
velocity, 1e-6 in density and 5e-5 in force, with those tolerances stated
in the tests. The rules are applied one thread per link rather than in a
per-cell loop, which lavapipe's JIT miscompiles (and which diverges on
real GPUs anyway). Native only for now; the browser registers nothing.
Part of #529.
run() drives a Backend in chunks of steps (to the next frame, history sample
or progress check) instead of the CPU Flow directly, so a device backend
registered by an application can step the same prepared flow: Flow keeps
building the grid, links and open cells and exposes them read-only, and
implements the trait itself. solve_fluid_on picks the CPU or the registered
device backend; solve_fluid stays the CPU and is unchanged in its results.
Part of #529.
The improvement objective was the smallest dihedral sine, which misses
needles: every dihedral angle fine while a corner is nearly flat, so
the corpus meshes had scaled Jacobians down to 0.01-0.07. The objective
is now the smaller of the dihedral sine and the scaled Jacobian at the
worst corner (the measure analysts check, Knupp 2001), weighted so a
Jacobian of 0.3 (the skill's limit) counts as a 30 degree dihedral
angle (the schedule's target): the Jacobian decides only near that
limit. The existing edge removal, multi-face removal, smoothing and
insertion then remove the needles.
Skill tet10 min SJ: sphere-on-plate 0.092 -> 0.335, filleted-bracket
0.021 -> 0.318 (both pass, nothing marked). Default-size tet10:
thin-ribs 0.240, shallow-groove 0.321, rounded-box 0.311.
The strong-relaxation test relaxes below 0.45: nothing is left below
0.3. Literature note in docs/meshing/literature.md.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
Where a tie's secondary surface meets a boundary condition or an
axis-aligned symmetry plane (a crosspoint), the secondary node's motion
is already prescribed and its mortar row can only constrain main
degrees of freedom. Its weights on main nodes off the boundary do not
vanish (the dual basis spreads over the node's faces), so after the
substitutions a small but real constraint is left: 1.3e-6 of the
coefficients that cancelled to give it on the tied hex-tet bar with
the needle-aware tet mesh (round-off would be ~1e-14; the geometry is
exact). It was refused as a near repeat, which made the tied bar
depend on how the tet faces fell. It is now dropped (the tie is then
off by that fraction at the crosspoint), after checking it does not
contradict the prescribed values. A near repeat on a secondary degree
of freedom another constraint eliminated (a node secondary in two
ties) is still refused (#389). Dropping every held secondary node's
row instead fails the tie patch test, so only near repeats go.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
Owner's rule (#438): move the bad nodes, interpolate, keep the morph as long as no
element inverts (guards "valid", the default); the old guards stay as "strict". The
run keeps its best round (highest smallest scaled Jacobian, then fewest elements below
the threshold, then highest mean) and never ends below the start. Default reach five
mean edge lengths of the bad elements (was three); on the Pi 5 base reach 3, 5 and 7
end at the same quality, 5 moving fewer nodes than 7 while leaving none below 0.7.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
Deleting a cap tetrahedron (two boundary faces of one B-rep face) gives each of its
two other neighbours a boundary face; a neighbour that had three was left with every
face on the boundary, joined to the mesh by its vertices alone. On Linux the dome on
a plate (cmd::plan::decompose::tests::a_dome_on_a_plate...) got one and the static
solve found the stiffness singular. The kernel's tessellation and sphere parameters
differ in their last bits between macOS and Linux (SolidCraft, outside this repo);
given the same input the tet mesher gives the same mesh on both (checked by replaying
Linux's surface and spheres on macOS, which reproduced the floating tetrahedron).
The deletion now refuses when a neighbour would float, and the linear validity gate
fails a mesh with a tetrahedron that shares no face with the rest, saying so.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
Holes by diameter, fillets by radius, chamfers by distance and bosses by height below
thresholds, or under a geometry.check finding; the reply panel switches each feature
off and on (geometry.defeature) and defeatures the whole proposal.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
In the max-min phase a valid element's term q^-16 reached 1e96 as q -> 0+, far above
an invalid element's 2e8, so the descent kept elements inverted rather than pass
through flat: the distorted plate stayed tangled (4 invalid) once its mesh was the
same on every platform. The shape term now continues as its tangent line below
q = 0.3 (finite, steep) and an invalid element costs 1e11, above any valid one.
Adds a test that the plate's mesh, its perturbation and the strong relaxation are
the same bits with 1 and 4 threads and match a fingerprint taken on macOS (CI checks
it on Linux).
Part of #438.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
The system maths library's sin, cos, atan2, log … are not correctly rounded and differ
between platforms: mesh.hex.generate gave the plate of the strong-relax test a different
mesh on Linux than on macOS (nodes from 2809 on), so the test passed on a Mac and failed
in CI. Every transcendental call in fieldcraft-mesh now goes through the libm crate (a
pure-Rust port of musl's), which gives the same bits everywhere.
Part of #438.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
Suppresses the SolidCraft features named (holes filled, fillets and chamfers back to
their sharp edges, bosses removed) and rebuilds the design; a feature that cannot be
suppressed alone is merged into its neighbour for meshing only. Kept in the model,
replayed on open, undoable; geometry.defeature.delete turns them back on.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
SolidCraft lists a solid's faces in a different order from run to run
(the rounded box, for one), which reordered the triangle soup and the
face tags with every coordinate the same; welding keeps the first copy
it meets and every later stage visits ids in order, so the tet mesh
changed between runs. CanonicalSoup sorts the triangles by their
coordinates, numbers vertices by first use and face tags by first
appearance, and maps the tags back; mesh.tet.generate and the corpus
use it.
Tests: the washer's tet10 mesh has a fixed fingerprint (identical on
arm64 and x86-64 builds); a reordered, renumbered input gives the same
mesh; heavy: rounded-box and filleted-bracket mesh bit for bit the same
on repeated runs and with 1 and 8 threads.
The bracket oracle cases stay unfrozen: SolidCraft's tessellation
itself differs in the last bits between arm64 and x86-64. The coarse
case moves to 203.6 MPa with the canonical order; its band is 5%.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
The platform maths libraries' atan2 and cbrt differ in the last bits
between macOS and glibc, so the dihedral angles and the shape measure a
mesh reports did too. The crate's own versions use + - * / and sqrt
only (correctly rounded everywhere) and agree with the platform's to a
few ulps.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
Faces and bodies under a size control are tinted, a box drawn as its
edges and a sphere as a translucent ball, each labelled with its name
and size. The Local Size form takes the selected faces.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
mesh.size.set adds a size control on faces, bodies, a box or a sphere;
mesh.size.growth sets the growth rate; mesh.size.clear and
mesh.size.list. The tet mesher bakes the controls into its graded field,
the hex meshers, cells and the octree take the graded RegionSize, and
the block projection and mesh plans divide their edges by it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
SizeField::range gives a field's smallest and largest size. The octree
starts its root cells at the largest instead of the size at the part's
centre, and refuses before building when the smallest needs more levels
than max_depth.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
The thinnest wall (measured to a facing wall) and narrowest patch give the cell size the grid
needs; the requested size refined that deep comes first, a root shrunk so one level fewer is
exact second; beyond the levels or cells allowed meshing fails at once naming the wall, levels and
cells. max_depth is a cap (default 4); rebuilds stay within the planned levels; lattice positions
no longer depend on the lattice depth; the report gives root size, levels, wall, patch and cells
per level before and after closure.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
A crease inside one face whose ends are not on other features (a tessellation crease across the
crank's r1.5 bore fillets) bounded nothing, so its open ends were corners no region border could
reach and the feature map failed on every build. Patch widths now count bounding curves only.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
geometry.face.split {name, face, datum} records a face split; it is
applied after the merges whenever the design is opened or cut again,
follows its datum (which it holds), undoes, and is saved with the
model. geometry.face.split.delete removes it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
A merged face on several surfaces (a sliver absorbed into its neighbour)
had no exact surface, so the projection target fell back to its
triangles: flat curvature probes gave fewer divisions round the boss and
the enclosure base's symmetric mesh dropped from 0.70 to 0.51 once its
absorbs reached the cells. The target now projects a point of such a
face onto the exact surface of the B-rep piece it is on, trimmed to that
piece (Tessellation::with_pieces, Shape::tessellate_pieces); node
relaxation slides on it the same way.
Test from commands: the enclosure base's four slivers absorbed, meshed
whole and by symmetry, the absorbs kept in every cell.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
A merge on `Base` failed with "no body `Base`" as soon as partitions cut
the body into cells (mesh.plan.apply's symmetry and corner cuts), so the
plan's cells and the body as opened silently lost every absorb and merge.
Each cell now takes the merge where the pieces of its faces touch (named
after the face kept, else merge:<name>[#k]), and the body as opened keeps
its merges whether partitions cut it or not.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
The face's triangles are cut along the plane, split:<name>:a on its
negative side and :b on the other, both on the face's surface, with the
cut as a new edge between them and the face's edges cut where the plane
crosses them. Applied after the merges; the B-rep is unchanged.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo
