RDNumeric::EigenSolvers Namespace Reference

Functions
bool RDKIT_EIGENSOLVERS_EXPORT	powerEigenSolver (unsigned int numEig, DoubleSymmMatrix &mat, DoubleVector &eigenValues, DoubleMatrix *eigenVectors=nullptr, int seed=-1)
static bool	powerEigenSolver (unsigned int numEig, DoubleSymmMatrix &mat, DoubleVector &eigenValues, DoubleMatrix &eigenVectors, int seed=-1)
	This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts. More...

Function Documentation

powerEigenSolver() [1/2]

static bool RDNumeric::EigenSolvers::powerEigenSolver ( unsigned int  numEig, DoubleSymmMatrix &  mat, DoubleVector &  eigenValues, DoubleMatrix &  eigenVectors, int  seed = -1  )

inlinestatic

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.

Definition at line 60 of file PowerEigenSolver.h.

References powerEigenSolver(), and RDKit::MHFPFingerprints::FNV::seed.

powerEigenSolver() [2/2]

bool RDKIT_EIGENSOLVERS_EXPORT RDNumeric::EigenSolvers::powerEigenSolver	(	unsigned int	numEig,
		DoubleSymmMatrix &	mat,
		DoubleVector &	eigenValues,
		DoubleMatrix *	eigenVectors = nullptr,
		int	seed = -1
	)

Compute the numEig largest eigenvalues and, optionally, the corresponding eigenvectors.

Parameters

numEig	the number of eigenvalues we are interested in
mat	symmetric input matrix of dimension N*N
eigenValues	Vector used to return the eigenvalues (size = numEig)
eigenVectors	Optional matrix used to return the eigenvectors (size = N*numEig)
seed	Optional values to seed the random value generator used to initialize the eigen vectors

Returns

a boolean indicating whether or not the calculation converged.

Notes:

• The matrix, mat, is changed in this function

Algorithm:

We use the iterative power method, which works like this:

 u = arbitrary unit vector
 tol = 0.001
 currEigVal = 0.0;
 prevEigVal = -1.0e100
 while (abs(currEigVal - prevEigVal) > tol) :
     v = Au
     prevEigVal = currEigVal
     currEigVal = v[i] // where i is the id of the largest absolute component
     u = c*v

Referenced by powerEigenSolver().