Python API

The suite is driven from Python — everything lives under the single peclet namespace, installable from PyPI (pip install peclet for the CPU family, pip install peclet-cu13 for the CUDA family, or an individual pip install peclet-<name>). Worked, runnable examples for every module are on the examples gallery.

Package	Import	What you get
peclet.flow	import peclet.flow	Solver / SolverColocated — the Eulerian Navier–Stokes solver (cut-cell IBM, VoF two-phase, moving analytic scenes, scalars, CFD-DEM hooks)
peclet.pnm	import peclet.pnm	Pore-network extraction + per-throat flow rates / pore pressures from a flow DNS
peclet.dem	import peclet.dem	Simulation — Lagrangian DEM (XPBD and Hertz–Mindlin), SDF-shaped particles, analytic walls
peclet.voro	import peclet.voro	Tessellation, Simulation, FlowSolver — moving-cell Voronoi, dynamics, Navier–Stokes on a Voronoi mesh
peclet.coupling	import peclet.coupling	CfdDem / ResolvedCfdDem — two-way CFD-DEM over flow + dem
peclet.halo · peclet.geom	from peclet import halo, geom	particle halo (MPI), analytic-SDF scene authoring (SceneBuilder); peclet.core.mpi / peclet.core.geom are the deprecated spellings
peclet.amr	import peclet.amr	block-octree AMR (Octree, DistributedOctree) and its collocated cut-cell Navier–Stokes solver (Flow) — under development
peclet.morton	import peclet.morton	vectorised Morton/Z-order arithmetic

Every Kokkos-backed module exposes execution_space (OpenMP / Cuda / HIP / Serial) so you can confirm which build you imported, and finalize() for a deterministic teardown (also run at interpreter exit). The package pages are generated from the modules' own docstrings each time this site is built, inside the released peclet-cpu container image, so they describe the latest release (not main) and include the distributed entry points; the full C++ API is on each code's Doxygen site (linked from the home page).

GPU & multi-rank

pip install peclet gives the multicore-CPU (OpenMP) build and pip install peclet-cu13 the single-GPU CUDA build. For AMD/HIP or multi-rank MPI, build the packages from source against a Kokkos prefix, or use a container — see Install & run.

Distributed (MPI) API

Built with the MPI flags on (PECLET_FLOW_MPI / PECLET_PNM_MPI / PECLET_DEM_MPI / PECLET_VORO_MPI — all on in the containers and the Snellius site install), the compute modules gain a multi-rank surface driven from mpi4py. peclet.flow.has_mpi reports whether the build you imported carries it (the PyPI wheels do not):

Module	Distributed entry points
peclet.flow	Solver.init_mpi(gnx,gny,gnz), peclet.flow.mpi_block(gnx,gny,gnz) → (origin, size), Solver.rebalance_by_weights(...), real Solver.rank()/size(), exchange_field / exchange_field_add across ranks
peclet.pnm	peclet.pnm.mpi_block(...), extract_pore_network_mpi(...), extract_network_flow_mpi(...) (bit-exact to the single-rank extraction), mpi_rank()/mpi_size()
peclet.dem	Simulation.init_mpi(...), enable_mpi_step(...), step_mpi(nsteps), step_hertz_mpi(nsteps), rebalance(), rank(), num_ghost()
peclet.voro	VoronoiHalo(origin, size, gsize, periodic) with owned_mask, gather(...) → (pos, gid, weight, n_owned), refresh_positions, rank()/size()
peclet.coupling	the drivers run distributed when their flow and dem are (the deposition uses exchange_field_add)
peclet.halo	peclet.halo.ParticleMigrator / ParticleHalo (the shared particle halo the above build on); peclet.amr.DistributedOctree

A distributed driver import mpi4py (which calls MPI_Init), then decomposes and steps. See the worked example benchmarks/profile_mpi_flow.py, the launch recipes in Containers → Distributed MPI, and the scaling studies on the gallery's benchmarks section.