[klausb]

I am trying to analyze the matrix structure but my routine gives me the same size for both the symmetric p matrix and the asymmetric U matrix as if the latter was symmetric and of the same size as the p matrix.

The diagonal part is currently omitted.

  // allocate memory for row_NNZ[N+1], the number of nonzero values for each matrix row
  int * row_NNZ = (int *)calloc(N+1, sizeof(int));

  // initialize row_NNZ with 0s
  for (int i=0; i<N+1; i++) 
    row_NNZ[i] = 0;

  // count the nonzero values per row for lower and upper part

  Foam::UList<int>::const_iterator it_lower = matrix_.lduAddr().lowerAddr().begin();
  Foam::UList<int>::const_iterator it_upper = matrix_.lduAddr().upperAddr().begin();

  // compute NNZ per row for upper + lower part, excluding diagonal for a symmetric matrix
  for (int i=0; i<matrix_.upper().size(); i++) {
    row_NNZ[*it_lower + 1]++;
    row_NNZ[*it_upper + 1]++;
    it_lower++;
    it_upper++;
  }
   // check sum size of upper + lower part 
   int i, sum=0;
   for (i=0; i<N+1; i++)
   {
     sum = sum + row_NNZ[i];
   }

I want to develop a routine for an array containing the numbers (count) of nonzero values per matrix row. So far the above seems to work for the symmetric matrix (p) but not for asymmetric matrices (U).

How can I adjust it for the asymmetric matrices?

Klaus

[hjasak]

The asymmetric matrix is still symmetric IN ADDRESSING. The difference is that the symmetric matrix says that a_ij = a_ji and it only stores one.

Make sure you count both the upper and lower triangle. This is not the CR format.

Good luck, Hrv