Shift-invert on sparse matrices

[Mahdisadjadi]

Can SymEigsShiftSolver find the eigenvalues of a SparseGenMatProd operation? When I use:

SparseGenMatProd<double> op(mat);
SymEigsShiftSolver< double, LARGEST_MAGN, 
SparseGenMatProd<double> > eigs(&op, 3, 6, 0.0);

I receive this error:

spectra/include/SymEigsSolver.h:855:21: error: no member named 'set_shift' in
      'Spectra::SparseGenMatProd<double>'
       this->m_op->set_shift(sigma);

[yixuan]

Hi @Mahdisadjadi, you should use the SparseSymShiftSolve class instead.

[Mahdisadjadi]

@yixuan : Thanks! I see, I downloaded tar.gz from the site and it doesn't have SparseSymMatProd.h header (& a few others)! After getting the latest files from github and having:

 SparseSymMatProd<double> op(mat);
 SparseSymShiftSolve< double, LARGEST_MAGN, 
SparseSymMatProd<double> > eigs(&op, 3, 6, 0.0);

I receive:

error: too many template arguments for class template
      'SparseSymShiftSolve'
  SparseSymShiftSolve< double, LARGEST_MAGN, SparseSymMatProd<double> > ...
  ^                                          ~~~~~~~~~~~~~~~~~~~~~~~~~~
spectra/include/MatOp/SparseSymShiftSolve.h:26:7: note: template is declared
      here
class SparseSymShiftSolve
      ^
1 error generated.

[yixuan]

Ah, notice that SparseSymShiftSolve is the matrix operator, and SymEigsShiftSolver is the solver.
So your declaration of eigs should be

SymEigsShiftSolver< double, LARGEST_MAGN, SparseSymMatProd<double> >

😄

[Mahdisadjadi]

@yixuan: Excellent, thanks! This works:

SparseSymShiftSolve<double> op(mat);
SymEigsShiftSolver< double, LARGEST_MAGN, 
    SparseSymShiftSolve<double> > eigs(&op, 3, 6, 2.0);