Parallelism and distribution

monoprop scales in two complementary ways: across MPI ranks (distributing the operator) and across threads within each rank (shared-memory parallelism). The MPI distribution shards both the operator and the graph across ranks, which can allow variational optimisation and simulation of larger systems than would fit in memory on a single node.

Single-node (MPI.COMM_SELF)

from mpi4py import MPI
sim = MajoranaPropagator(..., comm=MPI.COMM_SELF)

Multi-node (MPI.COMM_WORLD)

Replace the communicator and launch with mpiexec:

from mpi4py import MPI
sim = MajoranaPropagator(..., comm=MPI.COMM_WORLD)

mpiexec -n 8 uv run python your_script.py

Enabling MPI

MPI is off by default, so the prebuilt PyPI wheels run single-rank and the communicators above only distribute work after a from-source build with MPI enabled. See Building from source for the full build instructions.

Shared-memory parallelism

Within each MPI rank, monoprop uses Intel oneTBB for shared-memory parallelism. The thread count is read from the monoprop_NUM_THREADS environment variable, for example:

export monoprop_NUM_THREADS=8

When the variable is unset, TBB falls back to its own hardware-concurrency detection.