cp_cfm_diag Module Reference

used for collecting diagonalization schemes available for cp_cfm_type More...

Data Types
type	cp_cfm_diag_workspace_type

Functions/Subroutines
subroutine, public	cp_cfm_heevd (matrix, eigenvectors, eigenvalues)
	Perform a diagonalisation of a complex matrix.
subroutine, public	cp_cfm_power (matrix, work, exponent, threshold, n_dependent, verbose, eigvals)
	Raise the positive spectral part of a Hermitian matrix to a real power. Eigenmodes with eigenvalues below the nonnegative threshold, and all non-positive eigenvalues, are discarded. This routine does not implement a general matrix power for indefinite Hermitian matrices.
subroutine, public	cp_cfm_geeig (amatrix, bmatrix, eigenvectors, eigenvalues, work, lowest_subset)
	General Eigenvalue Problem AX = BXE Single option version: Cholesky decomposition of B.
subroutine, public	cp_cfm_geeig_canon (amatrix, bmatrix, eigenvectors, eigenvalues, work, epseig, nmo_retained)
	General Eigenvalue Problem AX = BXE Use canonical orthogonalization.
subroutine, public	cp_cfm_geeig_local (amatrix, bmatrix, eigenvectors, eigenvalues, ws)
	Solve a generalized complex eigenproblem using the local LAPACK backend. This routine is restricted to a one-rank BLACS grid. It deliberately avoids ScaLAPACK so independent k-points can be evaluated concurrently without making overlapping MPI calls from OpenMP worker threads.
subroutine, public	cp_cfm_geeig_canon_local (amatrix, bmatrix, eigenvectors, eigenvalues, work, epseig, ws)
	Canonical generalized complex diagonalization on a one-rank BLACS grid.
subroutine, public	cp_cfm_heevd_local (matrix, vectors, info, ws)
	Local LAPACK ZHEEVD with prepared scratch, without allocation or communication.
subroutine, public	cp_cfm_local_workspace_prepare (ws, n, a, b)
	Query LAPACK once per matrix order/solver and retain private scratch across k points.

Detailed Description

used for collecting diagonalization schemes available for cp_cfm_type

Note

first version : only one routine right now

Author

Joost VandeVondele (2003-09)

Function/Subroutine Documentation

cp_cfm_heevd()

subroutine, public cp_cfm_diag::cp_cfm_heevd	(	type(cp_cfm_type), intent(in)	matrix,
		type(cp_cfm_type), intent(in)	eigenvectors,
		real(kind=dp), dimension(:), intent(out)	eigenvalues )

Perform a diagonalisation of a complex matrix.

Parameters

matrix	...
eigenvectors	...
eigenvalues	...

History

12.2024 Added DLA-Future support [Rocco Meli] 08.2026 Added ELPA support

Author

Joost VandeVondele

Definition at line 93 of file cp_cfm_diag.F.

Here is the call graph for this function:

Here is the caller graph for this function:

cp_cfm_power()

subroutine, public cp_cfm_diag::cp_cfm_power	(	type(cp_cfm_type), intent(inout)	matrix,
		type(cp_cfm_type), intent(inout), optional, target	work,
		real(kind=dp), intent(in)	exponent,
		real(kind=dp), intent(in)	threshold,
		integer, intent(out), optional	n_dependent,
		logical, intent(in), optional	verbose,
		real(kind=dp), dimension(2), intent(out), optional	eigvals )

Raise the positive spectral part of a Hermitian matrix to a real power. Eigenmodes with eigenvalues below the nonnegative threshold, and all non-positive eigenvalues, are discarded. This routine does not implement a general matrix power for indefinite Hermitian matrices.

Parameters

matrix	input/output matrix; its upper triangle defines the Hermitian input
work	optional scratch matrix with the same size and distribution as matrix; must not alias matrix; overwritten during reconstruction; allocated internally when omitted
exponent	matrix exponent
threshold	nonnegative lower bound for retained eigenvalues
n_dependent	optional number of discarded eigenvalues
verbose	optional flag to print spectral diagnostics (default false)
eigvals	optional smallest and largest eigenvalues before filtering

Definition at line 238 of file cp_cfm_diag.F.

Here is the call graph for this function:

Here is the caller graph for this function:

cp_cfm_geeig()

subroutine, public cp_cfm_diag::cp_cfm_geeig	(	type(cp_cfm_type), intent(in)	amatrix,
		type(cp_cfm_type), intent(in)	bmatrix,
		type(cp_cfm_type), intent(in)	eigenvectors,
		real(kind=dp), dimension(:)	eigenvalues,
		type(cp_cfm_type), intent(in)	work,
		logical, intent(in), optional	lowest_subset )

General Eigenvalue Problem AX = BXE Single option version: Cholesky decomposition of B.

Parameters

amatrix	...
bmatrix	...
eigenvectors	...
eigenvalues	...
work	...
lowest_subset	compute only the requested lowest eigenpairs with ScaLAPACK when available

History

12.2024 Added DLA-Future support [Rocco Meli]

Definition at line 438 of file cp_cfm_diag.F.

Here is the call graph for this function:

Here is the caller graph for this function:

cp_cfm_geeig_canon()

subroutine, public cp_cfm_diag::cp_cfm_geeig_canon	(	type(cp_cfm_type), intent(inout)	amatrix,
		type(cp_cfm_type), intent(inout)	bmatrix,
		type(cp_cfm_type), intent(inout)	eigenvectors,
		real(kind=dp), dimension(:), intent(out)	eigenvalues,
		type(cp_cfm_type), intent(inout)	work,
		real(kind=dp), intent(in)	epseig,
		integer, intent(out), optional	nmo_retained )

General Eigenvalue Problem AX = BXE Use canonical orthogonalization.

Parameters

amatrix	...
bmatrix	...
eigenvectors	...
eigenvalues	...
work	...
epseig	...
nmo_retained	...

Definition at line 684 of file cp_cfm_diag.F.

Here is the call graph for this function:

Here is the caller graph for this function:

cp_cfm_geeig_local()

subroutine, public cp_cfm_diag::cp_cfm_geeig_local	(	type(cp_cfm_type), intent(in)	amatrix,
		type(cp_cfm_type), intent(in)	bmatrix,
		type(cp_cfm_type), intent(in)	eigenvectors,
		real(kind=dp), dimension(:), intent(out)	eigenvalues,
		type(cp_cfm_diag_workspace_type), intent(inout)	ws )

Solve a generalized complex eigenproblem using the local LAPACK backend. This routine is restricted to a one-rank BLACS grid. It deliberately avoids ScaLAPACK so independent k-points can be evaluated concurrently without making overlapping MPI calls from OpenMP worker threads.

Parameters

amatrix	Hamiltonian, overwritten
bmatrix	overlap matrix, overwritten
eigenvectors	eigenvectors
eigenvalues	eigenvalues
ws	reusable private LAPACK scratch

Definition at line 776 of file cp_cfm_diag.F.

Here is the call graph for this function:

Here is the caller graph for this function:

cp_cfm_geeig_canon_local()

subroutine, public cp_cfm_diag::cp_cfm_geeig_canon_local	(	type(cp_cfm_type), intent(in)	amatrix,
		type(cp_cfm_type), intent(in)	bmatrix,
		type(cp_cfm_type), intent(in)	eigenvectors,
		real(kind=dp), dimension(:), intent(out)	eigenvalues,
		type(cp_cfm_type), intent(in)	work,
		real(kind=dp), intent(in)	epseig,
		type(cp_cfm_diag_workspace_type), intent(inout)	ws )

Canonical generalized complex diagonalization on a one-rank BLACS grid.

Parameters

amatrix	Hamiltonian, overwritten
bmatrix	overlap matrix, overwritten and used as work storage
eigenvectors	eigenvectors
eigenvalues	eigenvalues
work	work matrix
epseig	overlap eigenvalue threshold
ws	reusable private LAPACK scratch

Definition at line 831 of file cp_cfm_diag.F.

Here is the call graph for this function:

Here is the caller graph for this function:

cp_cfm_heevd_local()

subroutine, public cp_cfm_diag::cp_cfm_heevd_local	(	complex(kind=dp), dimension(:, :), intent(inout)	matrix,
		complex(kind=dp), dimension(:, :), intent(inout), optional	vectors,
		integer, intent(out)	info,
		type(cp_cfm_diag_workspace_type), intent(inout)	ws )

Local LAPACK ZHEEVD with prepared scratch, without allocation or communication.

Parameters

matrix	input matrix, overwritten by eigenvectors
vectors	optional copy of the eigenvectors
info	...
ws	prepared private LAPACK scratch

Definition at line 909 of file cp_cfm_diag.F.

Here is the caller graph for this function:

cp_cfm_local_workspace_prepare()

subroutine, public cp_cfm_diag::cp_cfm_local_workspace_prepare	(	type(cp_cfm_diag_workspace_type), intent(inout)	ws,
		integer, intent(in)	n,
		complex(kind=dp), dimension(:, :), intent(inout)	a,
		complex(kind=dp), dimension(:, :), intent(inout), optional	b )

Query LAPACK once per matrix order/solver and retain private scratch across k points.

Parameters

ws	caller-owned workspace
n	matrix order
a	query matrix; workspace queries do not read its values
b	optional overlap selects ZHEGVD; otherwise query ZHEEVD

Definition at line 942 of file cp_cfm_diag.F.

Here is the caller graph for this function: