Sparsity pattern error while applying periodic boundary conditions

[Vinayak Vijay]

Deal deal.II community,

I want to apply periodic boundary conditions to my problem. But I can't use the make_periodicity_constraint() function due to the special nature of my problem (or so I think). So i create the required constraint matrix by taking the usual route of the  AffineConstraints::add_entries() function. 

My constraints are of the type: x_i (+ side dof) = a_ij * x_j ( - side dof). 

I need to apply these constraints after the assembly routine i.e. in the non-linear solver's load step loop (in the problems i have attempted until this one the contraints are applied in the setup_system() function).

With this brief, here's the code snippet i use (in the non-linear solver) to do the job of applying the constraints on the linear system. I start with creating an AffineConstraint object (different from the one used to create the system_rhs and system_matrix in the assemble_system() function). Then i simply want to apply the constraints using this object to the system_matrix and system_rhs. However, i get an error which is given below the code snippet.

AffineConstraints<double> constraints1;

// apply constraints

create_1d_pbc_constraints(constraints1);

// print all constraints that are obtained from the above function

constraints1.print(std::cout);

SparsityPattern sp;

sp.copy_from(system_matrix.get_sparsity_pattern());

std::ostringstream oss_4;

oss_4 << "sp.txt";

std::string filename4 = oss_4.str();

std::ofstream output_4(filename4.c_str());

sp.print(output_4);

std::cout << "closing constraint" << std::endl;

constraints1.close();

constraints1.condense(system_matrix,system_rhs);

Output:

60 0:  1
    60 1:  0
    60 2:  0
    61 0:  0
    61 1:  1
    61 2:  0
    62 0:  0
    62 1:  0
    62 2:  1
    63 3:  1
    63 4:  0
    63 5:  0
    64 3:  0
    64 4:  1
    64 5:  0
    65 3:  0
    65 4:  0
    65 5:  1
closing constraint

--------------------------------------------------------
An error occurred in line <1934> of file </tmp/vinayak/spack-stage/spack-stage-dealii-9.4.0-36iqlesdxyr55pceokjxk6hkwcek4m2d/spack-src/include/deal.II/lac/sparse_matrix.h> in function
    void dealii::SparseMatrix<Number>::add(dealii::SparseMatrix<Number>::size_type, dealii::SparseMatrix<Number>::size_type, number) [with number = double; dealii::SparseMatrix<Number>::size_type = unsigned int]
The violated condition was:
    (index != SparsityPattern::invalid_entry) || (value == number())
Additional information:
    You are trying to access the matrix entry with index <54,0>, but this
    entry does not exist in the sparsity pattern of this matrix.
   
    The most common cause for this problem is that you used a method to
    build the sparsity pattern that did not (completely) take into account
    all of the entries you will later try to write into. An example would
    be building a sparsity pattern that does not include the entries you
    will write into due to constraints on degrees of freedom such as
    hanging nodes or periodic boundary conditions. In such cases, building
    the sparsity pattern will succeed, but you will get errors such as the
    current one at one point or other when trying to write into the
    entries of the matrix.

Stacktrace:
-----------
#0  /home/vinayak/spack/opt/spack/linux-ubuntu22.04-broadwell/gcc-11.3.0/dealii-9.4.0-36iqlesdxyr55pceokjxk6hkwcek4m2d/lib/libdeal_II.g.so.9.4.0: dealii::SparseMatrix<double>::add(unsigned int, unsigned int, double)
#1  /home/vinayak/spack/opt/spack/linux-ubuntu22.04-broadwell/gcc-11.3.0/dealii-9.4.0-36iqlesdxyr55pceokjxk6hkwcek4m2d/lib/libdeal_II.g.so.9.4.0: void dealii::AffineConstraints<double>::condense<dealii::Vector<double> >(dealii::SparseMatrix<double>&, dealii::Vector<double>&) const

The error says "You are trying to access the matrix entry with index <54,0>". This seems to be inconsistent with the constraint matrix (printed in the output above). I am also attaching the sparsity pattern if needed.

Can someone help me with this? Also, since this procedure of applying constraints in the solver routine is quite different from the usual, it would help if someone could confirm that the above way of applying constraints is correct and/or if there's anything else i need to keep in mind.

Thanks

Vinayak

[Wolfgang Bangerth]

On 2/13/24 10:53, Vinayak Vijay wrote:
>
> The error says "You are trying to access the matrix entry with index <54,0>".

> *This seems to be inconsistent with the constraint matrix* (printed in the
> output above)*. *I am also attaching the sparsity pattern if needed.

>
> Can someone help me with this? Also, since this procedure of applying
> constraints in the solver routine is quite different from the usual, it would
> help if someone could confirm that the above way of applying constraints is
> correct and/or if there's anything else i need to keep in mind.

If you have an entry such as <54,60> and a constraint that ties DoF 60 to DoF
0, you will end up writing into entry <54,0>.

In the end, you need to ensure that when you build the sparsity pattern, you
introduce all entries you will write to. You don't show how you build the
sparsity pattern, but suspect that you call DoFTools::make_sparsity_pattern().
When you do this, you need to provide an AffineConstraints object that
includes all constraints, or you will get errors just like the ones you see.

Best
W>

--
------------------------------------------------------------------------
Wolfgang Bangerth email: bang...@colostate.edu
www: http://www.math.colostate.edu/~bangerth/

[Vinayak Vijay]

Thanks, Prof. Bangerth, this really helped.

Regards

Vinayak