Menú
MDOODZ particle reseeding and recycling — reseeding modes 0/1, particle deactivation (excess culling), recycling mechanism, markers struct fields, min_part_cell threshold, Nb_part_max hard limit, and common failure modes.
Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.
Basado en la clasificación ocupacional SOC
MDOODZ mechanical anisotropy module — director-based fabric tracking, anisotropy factors (δ_n, δ_s), viscous transverse isotropy, ViscosityConciseAniso, finite strain evolution, anisotropic rifting methodology, and anisotropy scenario examples.
MDOODZ performance benchmarking — running benchmarks, interpreting perf.csv, grid scaling studies, thread scaling, comparing machines, and the benchmark.sh script.
Build, compile, and run MDOODZ simulations — CMake configuration, make targets (build, build-dev, run, clean), dependency installation (SuiteSparse, HDF5, BLAS/LAPACK), OpenMP parallelisation, env.cmake setup, and common troubleshooting.
Code-to-physics glossary for MDOODZ — variable names, struct fields, function names mapped to physical meaning, equations, SI units, grid naming conventions, creep mechanism abbreviations, and solver process stages.
Debugging a failing MDOODZ model — the three failure families (CHOLMOD non-positive-definite, Powell-Hestenes non-convergence, first-iteration NaN), the error-string lookup table, the eta-contrast reality, the two anisotropy recipes, the single-variable bisection cookbook, and the known MDLIB anisotropy bugs with their fix commits. Use when a run crashes, won't factor, won't converge, NaNs, or stalls mid-production.
MDOODZ logging system — LOG_INFO/WARN/ERR/DBG/TIME macros, .txt configuration parameters (log_dest, log_level, log_timestamp, log_ts_mode, log_metadata), colored console output, ANSI-stripped log files, phase timing instrumentation, and metadata prefixes.
| name | skill-particle-reseeding |
| description | MDOODZ particle reseeding and recycling — reseeding modes 0/1, particle deactivation (excess culling), recycling mechanism, markers struct fields, min_part_cell threshold, Nb_part_max hard limit, and common failure modes. |
MDOODZ uses marker-in-cell advection. Particles drift with the flow and can cluster in convergent zones (downwellings, corners) while leaving divergent cells depleted. The reseeding system detects under-populated cells and injects new particles, while the deactivation system culls excess particles from over-populated cells. Dead particles (phase = -1) are recycled as slots for new ones.
All reseeding logic lives in MDLIB/ParticleRoutines.c. The file MDLIB/ParticleReseeding.c contained a dead function (CountPartCell2) that has been removed from the build.
Set reseed_mode in the .txt parameter file:
| Mode | Function | Placement | Deactivation | Slot Recycling |
|---|---|---|---|---|
| 0 | CountPartCell_OLD | Closest-neighbour (per-thread ipreuse[]) | None | Per-thread ipreuse[] |
| 1 | CountPartCell | Fine-cell centroid | None | Global part_reuse[] |
| 2 | CountPartCell_v2 | Random within fine cell | Distance-from-centroid, farthest first | Global part_reuse[] |
All modes use a fine mesh (2×Ncx × 2×Ncz) for cell-particle assignment, then map back to the coarse mesh for reseeding decisions.
markers)| Field | Type | Meaning |
|---|---|---|
Nb_part | int | Current active particle count |
Nb_part_max | int | Hard allocation limit (4.1× initial count) |
min_part_cell | int | Minimum particles per cell (from .txt); triggers reseeding |
phase[] | int* | Phase index; -1 = dead/air (available for recycling) |
Nx_part | int | Particles per cell in x (typically 4) |
Nz_part | int | Particles per cell in z (typically 4) |
A cell is considered under-populated when its particle count drops below min_part_cell. The reseeding routine (AddPartCell2) injects new particles at sub-cell positions, assigning phase from the nearest existing neighbour.
Only mode 2 (CountPartCell_v2) actively deactivates excess particles. After reseeding, it scans all cells on the fine mesh. If a cell has more than fine_threshold particles (where fine_threshold = ceil(min_part_cell / 4.0) + 4), the excess are sorted by distance from the cell centroid and the farthest are marked phase = -1 (dead), making their array slots available for recycling.
Boundary guard: Cells with BCt.type == 30 (Dirichlet thermal boundary) are skipped — deactivating boundary particles would corrupt boundary conditions.
Modes 0 and 1 have no deactivation. Particles can only become dead (phase = -1) by exiting the domain boundaries or being above the free surface. In a closed-box simulation (e.g., Blankenbach convection), modes 0 and 1 will see monotonic growth of active particle count, eventually hitting Nb_part_max and crashing with exit(190). Mode 2 is the only mode that guarantees bounded active particle count in closed environments.
part_reuse[] — an array of indices where phase == -1. New particles are assigned to these slots first. When slots run out, Nb_part is incremented (up to Nb_part_max).ipreuse[ith][]. Same logic: dead slots first, then fresh indices.Nb_part is written as a 1-element integer dataset in the Model group of each HDF5 output file. Use readScalarInt(file, "Model", "Nb_part") from TestHelpers.h to read it in tests.
| Symptom | Cause | Fix |
|---|---|---|
exit(190) crash | Nb_part reached Nb_part_max | Deactivation not working or threshold too high |
| Particles cluster in corners | Convection concentrates markers in downwellings | Normal — deactivation culls excess |
| Phase smearing at boundaries | Deactivation ran on boundary cells | Check BCt.type != 30 guard |
Nb_part grows monotonically | Using mode 0 or 1 (no deactivation) | Switch to reseed_mode = 2 for closed-box simulations |
Nx_part = 4 # particles_per_cell_x
Nz_part = 4 # particles_per_cell_z
min_part_cell = 16 # Reseeding threshold
reseed_mode = 1 # 0, 1, or 2
With log_level = 3 (DEBUG), all three reseeding functions emit per-cell and per-particle diagnostic messages:
cell (i,j) fine: N particles (threshold=T), reseedingnew particle at (x, z), reusing slot K or appended at index Kcell (i,j): N particles, excess E, deactivating E farthest and deactivated particle K (dist=D from centroid)Nb_part: <before> -> <after> (reused=R, created=C)Enable in .txt: log_level = 3. These messages are suppressed at the default log_level = 2 (INFO).
See also: VISUAL_TESTS/ReseedingGuide.md for a visual guide with plots from real simulation output.
The ConvectionDevelops test in TESTS/BlankenBenchTests.cpp asserts that Nb_part stays within 2× the initial count after 500 steps, confirming the recycling mechanism works for closed-box convection.