Dohp dohp

The function space can be implemented with tags on an iMesh state file. There are two viable options:

1. For every entity in the expanded space, write a pair of tags, offset of the first dof and the number of dofs associated with this entity. The writer will then simply dump the expanded vector into an HDF5 file.

• Pros:
  • multiple state vectors without duplicating the function space definition
  • the reader plugin is fairly simple because it will not have to know about the constraint matrix
  • the read operation for any given region (vtkStructuredGrid) is contiguous
• Cons:
  • duplicates storage for vertex, edge, and face dofs
  • continuity is not inherent

1. For every entity in global space, write the pair of tags. Then dump a global vector to the HDF5 file.

• Pros:
  • multiple state vectors without duplicating the function space definition
  • no duplicate storage
  • continuity is inherent
• Cons:
  • the reader plugin becomes more complex because the constraint matrix must be known
  • The reader plugin needs to apply (part of) the constraint matrix to populate a vtkStructured grid, the values in the state vector needed to apply these rows may involve a very discontiguous read. Handling this in parallel when making a movie of part of a domain sounds very hard.

My vote is for option 1.