Hello,

I'm solving a generalized eigenvalue problem which looks like this

Code: Select all

   
   Matrix<long double, Dynamic, Dynamic> m_A(N,N);
    Matrix<long double, Dynamic, Dynamic> m_B(N,N);
   
   for (unsigned int i=0; i<N; i++)
      for (unsigned int j=0; j<N; j++)
      {
         m_A(i,j)=RB_system_matrix_A(i,j);
         m_B(i,j)=RB_system_matrix_B(i,j);
      }
   
   GeneralizedSelfAdjointEigenSolver<Matrix<long double, Dynamic, Dynamic> > solverAB(m_A,m_B);
   Matrix<long double, Dynamic, 1> eigenvAB(N);
   eigenvAB = solverAB.eigenvalues();
   std::cout << eigenvAB << std::endl;

But the results I get using the same matrices in Matlab are different, and apparently more reliable. Switching every time to Matlab is annoying though. Can anvbody give me an explanation?

Thanks!!

Recall that for GeneralizedSelfAdjointEigenSolver, B must be positive definite. Maybe in your case it is nearly singular? Without the matrices A and B, I cannot help more!!

Sorry for reviving the topic only now, but I had vacation!

Yes, my B is s.p.d. but condition number is quite high, 10^6.