romtools.linalg#

This module defines the linear algebra functions used throughout the romtools library. These functions can be run in serial or parallel.

class romtools.linalg.DistributedSvd(comm=None)[source]#

Bases: object

NumPy-compatible thin SVD callable for a row-distributed matrix.

Bind an MPI communicator once, then pass this object anywhere a callable compatible with numpy.linalg.svd() is expected, including romtools.vector_space.VectorSpaceFromPOD.

Parameters:: comm (MPI_Comm) – Communicator over which matrix rows are distributed. If None, supported calls delegate to NumPy in serial.

Example

>>> distributed_svd = DistributedSvd(comm)
>>> U_local, s, Vh = distributed_svd(
...     A_local, full_matrices=False, compute_uv=True, hermitian=False
... )

Modules

`linalg`	see this for why this file exists and is done this way https://stackoverflow.com/questions/47599162/pybind11-how-to-package-c-and-python-code-into-a-single-package?rq=1
`parallel_utils`	Helper functions for testing and writing parallel code with the romtools.linalg module.