pylops_mpi.MPILinearOperator#

class pylops_mpi.MPILinearOperator(Op=None, shape=None, dims=None, dimsd=None, dtype=None, base_comm=<mpi4py.MPI.Intracomm object>)[source]#

MPI-enabled PyLops Linear Operator

Common interface for performing matrix-vector products in distributed fashion.

In practice, this class provides methods to perform matrix-vector and adjoint matrix-vector products between any pylops.LinearOperator (which must be the same across ranks) and a pylops_mpi.DistributedArray with Partition.BROADCAST and Partition.UNSAFE_BROADCAST partition. It internally handles the extraction of the local array from the distributed array and the creation of the output pylops_mpi.DistributedArray.

Note that whilst this operator could also be used with different pylops.LinearOperator across ranks, and with a pylops_mpi.DistributedArray with Partition.SCATTER, it is however recommended to use the pylops_mpi.basicoperators.MPIBlockDiag operator instead as this can also handle distributed arrays with subcommunicators.

Parameters:

Oppylops.LinearOperator, optional: If other arguments are provided, they will overwrite those obtained from Op. Defaults to None.
shapetuple(int, int), optional: Shape of the MPI Linear Operator. If not provided, obtained from dims and dimsd.
dimstuple(int, ..., int), optional: Dimensions of model. If not provided, (self.shape[1],) is used.
dimsdtuple(int, ..., int), optional: Dimensions of data. If not provided, (self.shape[0],) is used.
dtypestr, optional: Type of elements in input array. Defaults to None.
base_commmpi4py.MPI.Comm, optional: MPI Base Communicator. Defaults to mpi4py.MPI.COMM_WORLD.

Methods

`__init__`([Op, shape, dims, dimsd, dtype, ...])
`adjoint`()	Adjoint MPI LinearOperator
`conj`()	Complex conjugate operator
`dot`(x)	Matrix Vector Multiplication
`matvec`(x)	Matrix-vector multiplication.
`rmatvec`(x)	Adjoint Matrix-vector multiplication.
`transpose`()	Transposition of MPI LinearOperator