On 12/07/2018 07:48 AM, 'Maxi Miller' via deal.II User Group wrote:
> for(auto j = system_matrix.begin(); j != system_matrix.end(); ++j)
> {
> for(auto p = system_matrix.begin(j->row()); p !=
> system_matrix.end(j->row()); ++p)
> {
> /*system_matrix(j->row(),
I implemented a (rough) version of the method in question, with the code
here:
TimerOutput::Scope t(computing_timer, "Assemble Broyden System");
LinearAlgebraTrilinos::MPI::Vector
completely_distributed_s_value(dof_handler.locally_owned_dofs(),
mpi_communicator);
On 12/6/18 6:49 AM, 'Maxi Miller' via deal.II User Group wrote:
> I assume I understand that code, but for verification:
> I choose a row j, and iterate over it from begin(j) to end(j), while
> increasing p (not j, as I assume?). For every position which exists I get the
> column value, which
I assume I understand that code, but for verification:
I choose a row j, and iterate over it from begin(j) to end(j), while
increasing p (not j, as I assume?). For every position which exists I get
the column value, which then is stored in the vector positions, thus giving
me all non-zero
On 12/05/2018 02:31 AM, 'Maxi Miller' via deal.II User Group wrote:
> Hmm, is there a way to get the occupation directly as a vector? Then I
> could write p_i as p*s_i, with s_i equal to one for non-zero entries,
> and equal to zero for zero-value entries? I assume that should make the
>
Addendum to part II: As far as I understand, I have to iterate over the
sparsity pattern, and then call the accessor at that point (using index()).
Is that correct?
Am Mittwoch, 5. Dezember 2018 10:31:50 UTC+1 schrieb Maxi Miller:
>
> Hmm, is there a way to get the occupation directly as a
Hmm, is there a way to get the occupation directly as a vector? Then I
could write p_i as p*s_i, with s_i equal to one for non-zero entries, and
equal to zero for zero-value entries? I assume that should make the program
slightly more efficient.
Furthermore, I tried to find the function for
On 12/4/18 6:19 AM, 'Maxi Miller' via deal.II User Group wrote:
> For the implementation of the Quasi-Newton-method according to Schubert,
> which
> is defined as
>
> C_1 = C_0 - sum(u_i u_i^T(C_0p_j - y/t)(p_i^T/(p_i^Tp_i)))
>
> with p_i the vector p with all elements set to zero where column