Hello,

I'm trying to use the unsupported Eigen module AutoDiff. It seems to work fine (regarding the tests for this module) unless I try to differentiate an expression involving the computation of eigenvalues. E.g. when I try to differentiate the output of eigenvalueTest

Code: Select all

template<typename Vector>
EIGEN_DONT_INLINE typename Vector::Scalar eigenvalueTest (const Vector& v)
{
  typedef typename Vector::Scalar Scalar;
  typedef Matrix<Scalar,5,5> MatrixType;
  MatrixType mat;
  mat << 0, 1, 0, 0, 0,
              0, 0,1, 0, 0,
              0, 0, 0, 1, 0,
              0, 0, 0, 0, 1,
              pow (v[0], 4), v[1], v[2], v[3], v[4];
  return ((mat.eigenvalues()).real())[0];
}


I get the following compiler errors:

Code: Select all

~/eigen/eigen-3.1.1/unsupported/Eigen/src/AutoDiff/AutoDiffScalar.h(103): error: no instance of constructor "Eigen::CwiseBinaryOp<BinaryOp, Lhs, Rhs>::CwiseBinaryOp [with BinaryOp=Eigen::internal::scalar_sum_op<double>, Lhs=const Eigen::Matrix<double, 5, 1, 0, 5, 1>, Rhs=const Eigen::Matrix<double, 5, 1, 0, 5, 1>]" matches the argument list
            argument types are: (const Eigen::CwiseUnaryOp<Eigen::internal::scalar_multiple_op<double>, const Eigen::CwiseBinaryOp<Eigen::internal::scalar_sum_op<double>, const Eigen::Matrix<double, 5, 1, 0, 5, 1>, const Eigen::Matrix<double, 5, 1, 0, 5, 1>>>)
        : m_value(other.value()), m_derivatives(other.derivatives())
                                               ^


Is it possible to fix this? If not, what AD software would you recommend to be used together with Eigen?

Roman

if you want the derivatives of eigenvalues and eigenvectors, please use the specific formulas to get them -> this will be much faster! See for instance "On Differentiating Eigenvalues and Eigenvectors" by J. Magnus.

Thanks for the suggestion.

Roman

Edit: Moved to correct thread.