#include <init.hpp>

Inheritance diagram for q3::SubspaceDiagonalizer:

Public Member Functions
virtual	~SubspaceDiagonalizer ()=default

virtual bool	is_collective () const

virtual std::string	name () const =0

virtual int	solve (int basis_size, Complex hamiltonian_matrix, Complex overlap_matrix, Real eigenvalues, Complex eigenvector_matrix, bool is_gamma_only=false)=0

Detailed Description

Definition at line 20 of file init.hpp.

Constructor & Destructor Documentation

◆ ~SubspaceDiagonalizer()

virtual q3::SubspaceDiagonalizer::~SubspaceDiagonalizer ( )

virtualdefault

Member Function Documentation

◆ is_collective()

virtual bool q3::SubspaceDiagonalizer::is_collective ( ) const

inlinevirtual

Whether solve() is a collective operation over the process group.

Collective solvers (ScaLAPACK) require EVERY rank of the group to call solve() — gating the call on one rank deadlocks the group — and return redundant results on all ranks, so no post-solve broadcast is needed. Non-collective solvers (serial LAPACK) may be called on the group root only, with results broadcast afterwards.

Definition at line 54 of file init.hpp.

◆ name()

virtual std::string q3::SubspaceDiagonalizer::name ( ) const

pure virtual

Implemented in q3::SerialSubspaceDiagonalizer.

◆ solve()

virtual int q3::SubspaceDiagonalizer::solve	(	int	basis_size,
		Complex *	hamiltonian_matrix,
		Complex *	overlap_matrix,
		Real *	eigenvalues,
		Complex *	eigenvector_matrix,
		bool	is_gamma_only = `false`
	)

pure virtual

Solve the generalized eigenvalue problem: H * C = S * C * E

Parameters

basis_size	Dimension of the subspace (matrix size).
hamiltonian_matrix	Projected Hamiltonian (destroyed/overwritten). May be the same buffer as eigenvector_matrix: every implementation reads it whole before it writes any eigenvector, which is what lets the caller keep one copy of H rather than two.
overlap_matrix	Projected overlap (destroyed/overwritten).
eigenvalues	Output eigenvalues (size basis_size).
eigenvector_matrix	Output eigenvectors (basis_size x basis_size).

Returns: 0 on success, non-zero on failure.