How to copy a SparseMatrix

[martemyev]

How to copy a SparseMatrix

[tzanio]

Hi Mikhail,

If you add two BilinerFormIntegrators the corresponding BilinearForm will assemble their sum, see for example

https://github.com/mfem/mfem/blob/master/examples/ex3.cpp#L126

Does this address your use case?

You should be able to write

M += D

for sparse matrices M and D, as long as the sparsity patter on D is a subset of the sparsity patter of M (which very likely is true), see

https://github.com/mfem/mfem/blob/master/linalg/sparsemat.hpp#L279

Hope this helps,
Tzanio

[martemyev]

Hi Tzanio,

If you add two BilinerFormIntegrators the corresponding BilinearForm will assemble their sum

I think this is indeed possible in my case, but I thought that since I already computed M and D as separate matrices there should be any easy way to add them up, and creating a new bilinear form didn't look as the one.

You should be able to write M += D

I can't because in this case I change M, but I need the original one for the right hand side (the same is true for D).

for sparse matrices M and D, as long as the sparsity patter on D is a subset of the sparsity patter of M

I think I can add them up even if the patterns are completely different, but the height and width are the same, and both are not finalized. Please, let me know if I'm wrong.

As for the copying sparse matrices (just to close the topic) - there is no way to do that as far as I understood. Correct?

Best regards,
Mikhail

[v-dobrev]

Hi Mikhail,

Copying a SparseMatrix can be achieved with something like this:

// Assume we have a SparseMatrix A that is finalized, i.e. in csr format.
int nnz = A.NumNonZeroElems();
SparseMatrix B(A.GetI(), A.GetJ(), new double[nnz], A.Height(), A.Width(),
               false, true, A.areColumnsSorted());
B = 0.0;
B += A;

The new SparseMatrix B has the same sparsity pattern as A without duplicating the "I" and "J" arrays of the csr format, only the "data" double array is different and it will be owned by the newly created SparseMatrix B, i.e. B will delete its "data" array when it is destroyed.

An actual copy constructor for SparseMatrix will be available in the next release of MFEM.

Veselin

[martemyev]

Hi Veselin,

That looks like a solution! Thank you!

Best,
Mikhail