class that computes phonon energies, velocities and eigenvectors. More...

#include <phonon_h0.h>

Inheritance diagram for PhononH0:

Collaboration diagram for PhononH0:

Public Member Functions
	PhononH0 (Crystal &crystal, Eigen::Tensor< double, 5 > &forceConstants_, Eigen::Vector3i &qCoarseGrid, const Eigen::MatrixXd &bravaisVectors_, const Eigen::VectorXd &weights_, const int fcRangeType)
	Constructor, which stores all input data. More...

	PhononH0 (const PhononH0 &that)
	Copy constructor.

PhononH0 &	operator= (const PhononH0 &that)
	Copy assignment operator.

	~PhononH0 ()
	Destructor.

int	getNumBands () override
	Returns the number of phonon bands for the crystal in consideration.

Particle	getParticle () override
	Returns the underlying phonon-boson particle.

std::tuple< Eigen::VectorXd, Eigen::MatrixXcd >	diagonalize (Point &point) override
	get the phonon energies (in Ry) at a single q-point. More...

std::tuple< Eigen::VectorXd, Eigen::MatrixXcd >	diagonalizeFromCoordinates (Eigen::Vector3d &q, const bool &withMassScaling)
	Equivalent to diagonalize() computes phonon eigenValues/Vectors given the wavevector, but the wavevector is passed by coordinates. More...

std::tuple< Eigen::VectorXd, Eigen::MatrixXcd >	diagonalizeFromCoordinates (Eigen::Vector3d &q) override
	Diagonalize the Harmonic Hamiltonian at an arbitrary wavevector. More...

Eigen::Tensor< std::complex< double >, 3 >	diagonalizeVelocity (Point &point) override
	get the phonon velocities (in atomic units) at a single q-point. More...

Eigen::Tensor< std::complex< double >, 3 >	diagonalizeVelocityFromCoordinates (Eigen::Vector3d &coordinates) override

FullBandStructure	populate (Points &points, const bool &withVelocities, const bool &withEigenvectors, const bool isDistributed=false) override
	This method constructs a phonon band structure. More...

FullBandStructure	cpuPopulate (Points &points, const bool &withVelocities, const bool &withEigenvectors, bool isDistributed=false)

FullBandStructure	kokkosPopulate (Points &points, const bool &withVelocities, const bool &withEigenvectors, const bool isDistributed=false)
	Make a bandstructure for all the k-points.

StridedComplexView3D	kokkosBatchedBuildBlochHamiltonian (const DoubleView2D &cartesianCoordinates) override
	Build the Hamiltonian matrix for a batch of q-points.

std::tuple< DoubleView2D, StridedComplexView3D >	kokkosBatchedDiagonalizeFromCoordinates (const DoubleView2D &cartesianCoordinates, const bool withMassScaling=true) override
	Create and diagonalize Hamiltonians for a batch of k-points.

std::tuple< DoubleView2D, StridedComplexView3D, ComplexView4D >	kokkosBatchedDiagonalizeWithVelocities (const DoubleView2D &cartesianCoordinates) override
	Create and diagonalize Hamiltonians for a batch of k-points, with velocities Returns the energies (nk, nb), eigenvectors (nk, nb, nb) and velocities (nk, nb, nb, 3) at each k-point. More...

void	kokkosBatchedScaleEigenvectors (StridedComplexView3D &eigenvectors)

Eigen::Vector3i	getCoarseGrid ()
	Returns the size of the q-point coarse grid on which the force constants have been computed. More...

int	getIndexEigenvector (const int &iAt, const int &iPol) const
	Utility to convert the indices of atom basis and polarization into the index that has to be used with the eigenvector in matrix form. More...

Eigen::Matrix3d	getDielectricMatrix ()
	Get the static dielectric constant matrix. More...

Eigen::Tensor< double, 3 >	getBornCharges ()
	Get the Born effective charges for the ions in the unit cell. More...

int	estimateBatchSize (const bool &withVelocity) override
	Estimate how many k-points we can compute on the GPU in one batch. More...

void	printDynToHDF5 (Eigen::Vector3d &qCrys)
	Helper function to print the dynamical matrix file to HDF5, for developer testing purposes. More...

Public Member Functions inherited from HarmonicHamiltonian
void	kokkosBatchedTreatDegenerateVelocities (const DoubleView2D &cartesianCoordinates, const DoubleView2D &resultEnergies, ComplexView4D &resultVelocities, const double &threshold)

Static Public Member Functions
static int	getIndexEigenvector (const int &iAt, const int &iPol, const int &nAtoms)
	same as getIndexEigenvector, but as a static member More...

Static Public Attributes
static const int	mediumRange = 0

static const int	shortRange = 1

Protected Member Functions
Eigen::Tensor< double, 5 >	wsInit (const Eigen::MatrixXd &unitCell, const Eigen::Matrix3d &directUnitCell, const int &nr1Big, const int &nr2Big, const int &nr3Big)
	In wsInit, starting from the primitive crystal unit cell, we build the list of bravais lattice vectors used for the phonon Fourier transform.

void	addLongRangeTerm (Eigen::Tensor< std::complex< double >, 4 > &dyn, const Eigen::VectorXd &q)
	Adds the long range correction to the dynamical matrix due to dipole-ion interaction. More...

void	shortRangeTerm (Eigen::Tensor< std::complex< double >, 4 > &dyn, const Eigen::VectorXd &q)
	This part computes the short-range part of the dynamical matrix, which is the Fourier transform of the force constants. More...

void	addMediumRangeTerm (Eigen::Tensor< std::complex< double >, 4 > &dyn, const Eigen::VectorXd &q)
	This part computes the "medium"-range part of the dynamical matrix, which is needed for the polar correction to the dynamical matrix when finite differences are used. More...

std::tuple< Eigen::VectorXd, Eigen::MatrixXcd >	dynDiagonalize (Eigen::Tensor< std::complex< double >, 4 > &dyn)
	dynDiagonalize diagonalizes the dynamical matrix and returns eigenvalues and eigenvectors. More...

void	correctMediumRangeIFCs (Eigen::Tensor< double, 4 > &fq, const Eigen::VectorXd &q)
	Function analogous to nonanal_ifc in QE's matdyn – returns f_of_q, which is a correction to be applied to the "medium range" fcs before performing the fourier transform. More...

double	getDeviceMemoryUsage ()
	Checks the size of Device-allocated views. More...

Protected Attributes
Particle	particle

Crystal &	crystal

bool	hasDielectric = false

int	numAtoms

int	numBands

double	volumeUnitCell

Eigen::MatrixXi	atomicSpecies

Eigen::VectorXd	speciesMasses

Eigen::MatrixXd	atomicPositions

Eigen::Matrix3d	dielectricMatrix

Eigen::Tensor< double, 3 >	bornCharges

Eigen::Vector3i	qCoarseGrid

Eigen::Matrix3d	directUnitCell

int	dimensionality

int	fcRangeType

bool	longRange2d = false

double	alpha

double	alat

int	numBravaisVectors = 0

Eigen::MatrixXd	bravaisVectors

Eigen::VectorXd	weights

Eigen::Tensor< double, 5 >	mat2R

Eigen::MatrixXd	gVectors

Eigen::Tensor< double, 3 >	longRangeCorrection1

const double	gMax = 14.

const double	e2 = 4

DoubleView1D	atomicMasses_d

DoubleView3D	longRangeCorrection1_d

DoubleView2D	gVectors_d

DoubleView2D	dielectricMatrix_d

DoubleView3D	bornCharges_d

DoubleView2D	atomicPositions_d

DoubleView2D	bravaisVectors_d

DoubleView1D	weights_d

DoubleView3D	mat2R_d

Detailed Description

First, it contains the force constants, i.e. the second derivative of the total energy w.r.t. ionic displacements. Additionally, it has all the infrastructure to Fourier transform this matrix, and diagonalize it to get the harmonic phonon properties.

Constructor & Destructor Documentation

◆ PhononH0()

PhononH0::PhononH0	(	Crystal &	crystal,
		Eigen::Tensor< double, 5 > &	forceConstants_,
		Eigen::Vector3i &	qCoarseGrid,
		const Eigen::MatrixXd &	bravaisVectors_,
		const Eigen::VectorXd &	weights_,
		const int	fcRangeType
	)

Parameters

crystal	the object with the information on the crystal structure
forceConstants	a tensor of doubles with the force constants size is (3, 3, meshX, meshY, meshZ, numAtoms, numAtoms)
qCoarseGrid	the supercell/coarse qmesh of this calculation as a 3vector
bravaisVectors	the R vectors used in the harmonic phonon real space representation
weights	the R vector weights used in the Fourier transform to recip space

Here is the call graph for this function:

Member Function Documentation

◆ addLongRangeTerm()

void PhononH0::addLongRangeTerm	(	Eigen::Tensor< std::complex< double >, 4 > &	dyn,
		const Eigen::VectorXd &	q
	)

protected

Parameters

dyn	: a container to which the short-range part of the dynamical matrix is added
q	: the phonon wavevector at which to calculate this part of the dyn mat, cartesian coords

Here is the caller graph for this function:

◆ addMediumRangeTerm()

void PhononH0::addMediumRangeTerm	(	Eigen::Tensor< std::complex< double >, 4 > &	dyn,
		const Eigen::VectorXd &	q
	)

protected

follows nonanal from rigid.f90 and the paper of Wang 2012

Parameters

dyn	: a container to which the short-range part of the dynamical matrix is added
q	: the phonon wavevector at which to calculate this part of the dyn mat, cartesian coords

Here is the caller graph for this function:

◆ correctMediumRangeIFCs()

void PhononH0::correctMediumRangeIFCs	(	Eigen::Tensor< double, 4 > &	fq,
		const Eigen::VectorXd &	q
	)

protected

Parameters

q	the phonon wavevector associated with the correction.

Returns: correction to be applied to the IFCS before fourier transforming

Here is the caller graph for this function:

◆ diagonalize()

std::tuple< Eigen::VectorXd, Eigen::MatrixXcd > PhononH0::diagonalize ( Point & point )

overridevirtual

Parameters

q	a point object with the wavevector. Must know the cartesian coordinates of the wavevector.

Returns: tuple(energies, eigenvectors): the energies are a double vector of size (numBands=3*numAtoms). Eigenvectors are a complex tensor of size (3,numAtoms,numBands). The eigenvector is rescaled by the sqrt(masses) (masses in rydberg)

Implements HarmonicHamiltonian.

Here is the call graph for this function:

◆ diagonalizeFromCoordinates() [1/2]

std::tuple< Eigen::VectorXd, Eigen::MatrixXcd > PhononH0::diagonalizeFromCoordinates ( Eigen::Vector3d & k )

overridevirtual

Same as diagonalize(), but the wavevector coordinates are explicitly passed in input.

Parameters

point the wavevector in cartesian coordinates.

Returns: eigenvalues: the values of quasiparticle energies, a double vector of size numBands.; eigenvectors: a complex matrix of size (numBands,numBands) with the eigenvectors, ordered in columns, computed at the input wavevector.

Implements HarmonicHamiltonian.

Here is the call graph for this function:

◆ diagonalizeFromCoordinates() [2/2]

std::tuple< Eigen::VectorXd, Eigen::MatrixXcd > PhononH0::diagonalizeFromCoordinates	(	Eigen::Vector3d &	q,
		const bool &	withMassScaling
	)

Parameters

q	a 3d eigen vector with the cartesian coordinates of the phonon wavevector.
withMassScaling	if true, rescales the eigenvectors by the mass z -> z/sqrt(m)

Returns: eigenvalues: all values of phonon energies for this point.; eigenvectors: the phonon eigenvectors, in matrix form, for this point.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ diagonalizeVelocity()

Eigen::Tensor< std::complex< double >, 3 > PhononH0::diagonalizeVelocity ( Point & point )

overridevirtual

Parameters

q	a Point object with the wavevector coordinates.

Returns: velocity(numBands,numBands,3): values of the velocity operator for this state, in atomic units.

Implements HarmonicHamiltonian.

Here is the call graph for this function:

◆ dynDiagonalize()

std::tuple< Eigen::VectorXd, Eigen::MatrixXcd > PhononH0::dynDiagonalize ( Eigen::Tensor< std::complex< double >, 4 > & dyn )

protected

Parameters

dyn	the dynamical matrix in the shape 3,3,natoms,natoms

Here is the caller graph for this function:

◆ estimateBatchSize()

int PhononH0::estimateBatchSize ( const bool & withVelocity )

overridevirtual

Parameters

withVelocity set to true if computing also the velocity operator, which requires more memory

Returns: numBatches: an estimate on how many k-point we can compute in one call of the kokkosBatched functions.

Reimplemented from HarmonicHamiltonian.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ getBornCharges()

Eigen::Tensor< double, 3 > PhononH0::getBornCharges ( )

Returns: bornCharges: a real tensor of shape (numAtoms,3,3) with the charges

Here is the caller graph for this function:

◆ getCoarseGrid()

Eigen::Vector3i PhononH0::getCoarseGrid ( )

Returns: qCoarseGrid: an Eigen vector of 3 integers.

Here is the caller graph for this function:

◆ getDeviceMemoryUsage()

double PhononH0::getDeviceMemoryUsage ( )

protected

Returns: size occupied by Kokkos views, in bytes.

Here is the caller graph for this function:

◆ getDielectricMatrix()

Eigen::Matrix3d PhononH0::getDielectricMatrix ( )

Returns: dielectricMatrix: a 3x3 eigen matrix.

Here is the caller graph for this function:

◆ getIndexEigenvector() [1/2]

int PhononH0::getIndexEigenvector	(	const int &	iAt,
		const int &	iPol
	)		const

Parameters

iAt	atomic basis index.
iPol	polarization index (0,1,2).

Returns: k: index to be used in the phonon eigenvector.

Here is the caller graph for this function:

◆ getIndexEigenvector() [2/2]

int PhononH0::getIndexEigenvector	(	const int &	iAt,
		const int &	iPol,
		const int &	nAtoms
	)

static

Parameters

nAtoms the number of atoms in the unit cell

◆ kokkosBatchedDiagonalizeWithVelocities()

std::tuple< DoubleView2D, StridedComplexView3D, ComplexView4D > PhononH0::kokkosBatchedDiagonalizeWithVelocities ( const DoubleView2D & cartesianCoordinates )

overridevirtual

Note that for each k-point, the (nb, nb) eigenvector matrix is column-major, as required by the cuSOLVER routine, hence the StridedLayout.

Implements HarmonicHamiltonian.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ populate()

FullBandStructure PhononH0::populate	(	Points &	points,
		const bool &	withVelocities,
		const bool &	withEigenvectors,
		const bool	isDistributed = `false`
	)

overridevirtual

Parameters

points	the object with the list/mesh of wavevectors
withVelocities	if true, compute the phonon velocity operator.
withEigenvectors	if true, stores the phonon eigenvectors.

Returns: FullBandStructure: the band structure object containing the complete phonon band structure.

Implements HarmonicHamiltonian.

Here is the caller graph for this function:

◆ printDynToHDF5()

void PhononH0::printDynToHDF5 ( Eigen::Vector3d & qCrys )

Parameters

qCrys	a 3d eigen vector in crystal coordinates of the phonon wavevector.
points	the points object with with q belongs to. Here, used only to convert q to cartesian internally.

Here is the call graph for this function:

◆ shortRangeTerm()

void PhononH0::shortRangeTerm	(	Eigen::Tensor< std::complex< double >, 4 > &	dyn,
		const Eigen::VectorXd &	q
	)

protected

Parameters

dyn	: a container to which the short-range part of the dynamical matrix is added
q	: the phonon wavevector at which to calculate this part of the dyn mat, cartesian coords

Here is the call graph for this function:

Here is the caller graph for this function:

Public Member Functions

Static Public Member Functions

Static Public Attributes

Protected Member Functions

Protected Attributes

Detailed Description

Constructor & Destructor Documentation

◆ PhononH0()

Member Function Documentation

◆ addLongRangeTerm()

◆ addMediumRangeTerm()

◆ correctMediumRangeIFCs()

◆ diagonalize()

◆ diagonalizeFromCoordinates() [1/2]

◆ diagonalizeFromCoordinates() [2/2]

◆ diagonalizeVelocity()

◆ dynDiagonalize()

◆ estimateBatchSize()

◆ getBornCharges()

◆ getCoarseGrid()

◆ getDeviceMemoryUsage()

◆ getDielectricMatrix()

◆ getIndexEigenvector() [1/2]

◆ getIndexEigenvector() [2/2]

◆ kokkosBatchedDiagonalizeWithVelocities()

◆ populate()

◆ printDynToHDF5()

◆ shortRangeTerm()