Parallel computation of quadrature rules

[dokken]

@federicobetti

You are right that it is useless to split cells among processors if this is automatically done by dolfinx. However, it seems to me that the local numbering in each rank is quite random with regards to the global one. Which command allows me to get this?

The ordering of cells in DOLFINx is determined by a graph partitioner (scotch, kahip or parmetis), that uses the mesh connectivity to decide what process should own what information.

The vertices are also renumbered, to ensure data-locality and efficient parallel computing.

For mapping input mesh nodes see for instance:

Wrong order of dofs after creating mesh from lists of points and cells

You are making some major assumptions here that simply cannot be held by DOLFINx This is because: Only simple function spaces has the same number of dofs as the mesh geometry The input ordering does not make sense once one wants to run in parallel. More information about this has been discussed in: Consider the following MWE to map a vertex->mesh node->original_input_position import dolfinx import matplotlib.pyplot as plt import numpy as np import ufl from mpi4py import MPI pp = np.arra…

More about the mapping of data has also been written in: Gather solutions in parallel in FEniCSX - #2 by dokken