Hi,
I'd like to sum many sparse matrix. I studied the tutorial about sparse matrix and i have seen that it is possible to use the + operator.
But the += ?

My code is:

Code: Select all

SparseMatrix<double>  K(500,500);
vector <SparseMatrix<double> *>::iterator iter=theDomain->VecOfMatr.begin();
while (iter != theDomain->VecOfMatr.end())
      K +=  *(*iter++);


this code doesn't work in runtime.
if i changed it with, (for example to sum only the first two) :

Code: Select all

Kglobal=*(TheDomain->VecOfMatr.at(0))+*(TheDomain->VecOfMatr.at(1));


it works.

So could you tell me the right but also the more efficient way to sum sparse matrices?
Thanks

you can do:

while(...)
K = (K + **iter).eval();

however, summing up sparse matrices is not an efficient operations. Do you really need to compute the individual sparse matrices? cannot you directly assemble their sum?

How many sparse matrices do you have to sum up?

ggael wrote:you can do:

while(...)
K = (K + **iter).eval();

however, summing up sparse matrices is not an efficient operations. Do you really need to compute the individual sparse matrices? cannot you directly assemble their sum?

How many sparse matrices do you have to sum up?

It's a finite element procedure.
To take my "code" more general as possible, it's better to sum up at the end of the procedure when the matrices are assembled and not before (when are small and dense).
The number of the matrices is related to the user but usually we can speak about 6 sparse matrices.
I will try if it's too slow

If you pick up a snapshot of the last devel branch you can also simply do something like:

Code: Select all

    typedef Triplet<double> T;
    std::vector<T> tripletList;
    triplets.reserve(estimation_of_entries);
    for(...)
    {
      // ...
      tripletList.push_back(T(i,j,v_ij));
    }
    SparseMatrixType m(rows,cols);
    m.setFromTriplets(tripletList.begin(), tripletList.end());


The duplicates will be summed up by the setFromTriplets function in O(tripletList.size()) time.

ggael wrote:If you pick up a snapshot of the last devel branch you can also simply do something like:

Code: Select all

    typedef Triplet<double> T;
    std::vector<T> tripletList;
    triplets.reserve(estimation_of_entries);
    for(...)
    {
      // ...
      tripletList.push_back(T(i,j,v_ij));
    }
    SparseMatrixType m(rows,cols);
    m.setFromTriplets(tripletList.begin(), tripletList.end());


The duplicates will be summed up by the setFromTriplets function in O(tripletList.size()) time.

Thank you

P.S.
In the third line, triplets should be tripletList ?

In the third line, triplets should be tripletList ?

right