mirror of
https://github.com/triqs/dft_tools
synced 2024-06-30 00:44:34 +02:00
c1b3000c00
There was a very nasty bug in the preparation of the block matrix 'p_mat'. The point is that this matrix is created once for all k-points with the band dimension being the maximum possible. However, only a part of the matrix is used at every k-point but the orthogonalization is done for the whole matrix. The problem was that if the number of bands for a given k-point was smaller than that for the next k-point them for the next k-point some part of 'p_mat' still contained data from the previous step, which messed up the orthonormalization. Now, 'p_mat' is set to zero at each step of the loop. Also, property 'nion' was added to ProjectorShell since it is used very often.
311 lines
12 KiB
Python
311 lines
12 KiB
Python
|
|
import itertools as it
|
|
import numpy as np
|
|
import vasp.atm.c_atm_dos as c_atm_dos
|
|
|
|
np.set_printoptions(suppress=True)
|
|
|
|
def issue_warning(message):
|
|
"""
|
|
Issues a warning.
|
|
"""
|
|
print
|
|
print " !!! WARNING !!!: " + message
|
|
print
|
|
|
|
################################################################################
|
|
################################################################################
|
|
#
|
|
# class ProjectorShell
|
|
#
|
|
################################################################################
|
|
################################################################################
|
|
class ProjectorShell:
|
|
"""
|
|
Container of projectors related to a specific shell.
|
|
|
|
The constructor pre-selects a subset of projectors according to
|
|
the shell parameters passed from the config-file.
|
|
|
|
Parameters:
|
|
|
|
- sh_pars (dict) : shell parameters from the config-file
|
|
- proj_raw (numpy.array) : array of raw projectors
|
|
|
|
"""
|
|
def __init__(self, sh_pars, proj_raw, proj_params, nc_flag):
|
|
self.lorb = sh_pars['lshell']
|
|
self.ion_list = sh_pars['ion_list']
|
|
self.user_index = sh_pars['user_index']
|
|
self.nc_flag = nc_flag
|
|
# try:
|
|
# self.tmatrix = sh_pars['tmatrix']
|
|
# except KeyError:
|
|
# self.tmatrix = None
|
|
|
|
self.lm1 = self.lorb**2
|
|
self.lm2 = (self.lorb+1)**2
|
|
|
|
self.ndim = self.extract_tmatrices(sh_pars)
|
|
self.nion = len(self.ion_list)
|
|
|
|
# Pre-select a subset of projectors (this should be an array view => no memory is wasted)
|
|
# !!! This sucks but I have to change the order of 'ib' and 'ilm' indices here
|
|
# This should perhaps be done right after the projector array is read from PLOCAR
|
|
# self.proj_arr = proj_raw[self.ion_list, :, :, :, self.lm1:self.lm2].transpose((0, 1, 2, 4, 3))
|
|
# We want to select projectors from 'proj_raw' and form an array
|
|
# self.proj_arr[nion, ns, nk, nlm, nb]
|
|
# TODO: think of a smart way of copying the selected projectors
|
|
# perhaps, by redesigning slightly the structure of 'proj_arr' and 'proj_win'
|
|
# or by storing only a mapping between site/orbitals and indices of 'proj_raw'
|
|
# iproj_l = []
|
|
nion = len(self.ion_list)
|
|
nlm = self.lm2 - self.lm1
|
|
_, ns, nk, nb = proj_raw.shape
|
|
self.proj_arr = np.zeros((nion, ns, nk, nlm, nb), dtype=np.complex128)
|
|
for io, ion in enumerate(self.ion_list):
|
|
for m in xrange(nlm):
|
|
# Here we search for the index of the projector with the given isite/l/m indices
|
|
for ip, par in enumerate(proj_params):
|
|
if par['isite'] - 1 == ion and par['l'] == self.lorb and par['m'] == m:
|
|
# iproj_l.append(ip)
|
|
self.proj_arr[io, :, :, m, :] = proj_raw[ip, :, :, :]
|
|
break
|
|
|
|
# self.proj_arr = proj_raw[iproj_l, :, :, :].transpose((1, 2, 0, 3))
|
|
|
|
################################################################################
|
|
#
|
|
# extract_tmatrices
|
|
#
|
|
################################################################################
|
|
def extract_tmatrices(self, sh_pars):
|
|
"""
|
|
Extracts and interprets transformation matrices provided by the
|
|
config-parser.
|
|
There are two relevant options in 'sh_pars':
|
|
|
|
'tmatrix' : a transformation matrix applied to all ions in the shell
|
|
'tmatrices': interpreted as a set of transformation matrices for each ion.
|
|
|
|
If both of the options are present a warning is issued and 'tmatrices'
|
|
supersedes 'tmatrix'.
|
|
"""
|
|
nion = len(self.ion_list)
|
|
nm = self.lm2 - self.lm1
|
|
|
|
if 'tmatrices' in sh_pars:
|
|
if 'tmatrix' in sh_pars:
|
|
mess = "Both TRANSFORM and TRANSFILE are specified, TRANSFORM will be ignored."
|
|
issue_warning(mess)
|
|
|
|
raw_matrices = sh_pars['tmatrices']
|
|
nrow, ncol = raw_matrices.shape
|
|
|
|
assert nrow%nion == 0, "Number of rows in TRANSFILE must be divisible by the number of ions"
|
|
assert ncol%nm == 0, "Number of columns in TRANSFILE must be divisible by the number of orbitals 2*l + 1"
|
|
|
|
nr = nrow / nion
|
|
nsize = ncol / nm
|
|
assert nsize in (1, 2, 4), "Number of columns in TRANSFILE must be divisible by either 1, 2, or 4"
|
|
#
|
|
# Determine the spin-dimension and whether the matrices are real or complex
|
|
#
|
|
# if nsize == 1 or nsize == 2:
|
|
# Matrices (either real or complex) are spin-independent
|
|
# nls_dim = nm
|
|
# if msize == 2:
|
|
# is_complex = True
|
|
# else:
|
|
# is_complex = False
|
|
# elif nsize = 4:
|
|
# Matrices are complex and spin-dependent
|
|
# nls_dim = 2 * nm
|
|
# is_complex = True
|
|
#
|
|
is_complex = nsize > 1
|
|
ns_dim = max(1, nsize / 2)
|
|
|
|
# Dimension of the orbital subspace
|
|
assert nr%ns_dim == 0, "Number of rows in TRANSFILE is not compatible with the spin dimension"
|
|
ndim = nr / ns_dim
|
|
|
|
self.tmatrices = np.zeros((nion, nr, nm * ns_dim), dtype=np.complex128)
|
|
|
|
if is_complex:
|
|
raw_matrices = raw_matrices[:, ::2] + raw_matrices[:, 1::2] * 1j
|
|
|
|
for io in xrange(nion):
|
|
i1 = io * nr
|
|
i2 = (io + 1) * nr
|
|
self.tmatrices[io, :, :] = raw_matrices[i1:i2, :]
|
|
|
|
return ndim
|
|
|
|
if 'tmatrix' in sh_pars:
|
|
raw_matrix = sh_pars['tmatrix']
|
|
nrow, ncol = raw_matrix.shape
|
|
|
|
assert ncol%nm == 0, "Number of columns in TRANSFORM must be divisible by the number of orbitals 2*l + 1"
|
|
|
|
# Only spin-independent matrices are expected here
|
|
nsize = ncol / nm
|
|
assert nsize in (1, 2), "Number of columns in TRANSFORM must be divisible by either 1 or 2"
|
|
|
|
is_complex = nsize > 1
|
|
if is_complex:
|
|
matrix = raw_matrix[:, ::2] + raw_matrix[:, 1::2] * 1j
|
|
else:
|
|
matrix = raw_matrix
|
|
|
|
ndim = nrow
|
|
|
|
self.tmatrices = np.zeros((nion, nrow, nm), dtype=np.complex128)
|
|
for io in xrange(nion):
|
|
self.tmatrices[io, :, :] = raw_matrix
|
|
|
|
return ndim
|
|
|
|
# If no transformation matrices are provided define a default one
|
|
ns_dim = 2 if self.nc_flag else 1
|
|
ndim = nm * ns_dim
|
|
|
|
self.tmatrices = np.zeros((nion, ndim, ndim), dtype=np.complex128)
|
|
for io in xrange(nion):
|
|
self.tmatrices[io, :, :] = np.identity(ndim, dtype=np.complex128)
|
|
|
|
return ndim
|
|
|
|
################################################################################
|
|
#
|
|
# select_projectors
|
|
#
|
|
################################################################################
|
|
def select_projectors(self, ib_win, ib_min, ib_max):
|
|
"""
|
|
Selects a subset of projectors corresponding to a given energy window.
|
|
"""
|
|
self.ib_win = ib_win
|
|
self.ib_min = ib_min
|
|
self.ib_max = ib_max
|
|
nb_max = ib_max - ib_min + 1
|
|
|
|
# Set the dimensions of the array
|
|
nion, ns, nk, nlm, nbtot = self.proj_arr.shape
|
|
# !!! Note that the order of the two last indices is different !!!
|
|
self.proj_win = np.zeros((nion, ns, nk, nlm, nb_max), dtype=np.complex128)
|
|
|
|
# Select projectors for a given energy window
|
|
ns_band = self.ib_win.shape[1]
|
|
for isp in xrange(ns):
|
|
for ik in xrange(nk):
|
|
# TODO: for non-collinear case something else should be done here
|
|
is_b = min(isp, ns_band)
|
|
ib1 = self.ib_win[ik, is_b, 0]
|
|
ib2 = self.ib_win[ik, is_b, 1] + 1
|
|
ib_win = ib2 - ib1
|
|
self.proj_win[:, isp, ik, :, :ib_win] = self.proj_arr[:, isp, ik, :, ib1:ib2]
|
|
|
|
################################################################################
|
|
#
|
|
# density_matrix
|
|
#
|
|
################################################################################
|
|
def density_matrix(self, el_struct, site_diag=True, spin_diag=True):
|
|
"""
|
|
Returns occupation matrix/matrices for the shell.
|
|
"""
|
|
nion, ns, nk, nlm, nbtot = self.proj_win.shape
|
|
|
|
assert site_diag, "site_diag = False is not implemented"
|
|
assert spin_diag, "spin_diag = False is not implemented"
|
|
|
|
occ_mats = np.zeros((ns, nion, nlm, nlm), dtype=np.float64)
|
|
overlaps = np.zeros((ns, nion, nlm, nlm), dtype=np.float64)
|
|
|
|
# self.proj_win = np.zeros((nion, ns, nk, nlm, nb_max), dtype=np.complex128)
|
|
kweights = el_struct.kmesh['kweights']
|
|
occnums = el_struct.ferw
|
|
ib1 = self.ib_min
|
|
ib2 = self.ib_max + 1
|
|
for isp in xrange(ns):
|
|
for ik, weight, occ in it.izip(it.count(), kweights, occnums[isp, :, :]):
|
|
for io in xrange(nion):
|
|
proj_k = self.proj_win[io, isp, ik, ...]
|
|
occ_mats[isp, io, :, :] += np.dot(proj_k * occ[ib1:ib2],
|
|
proj_k.conj().T).real * weight
|
|
overlaps[isp, io, :, :] += np.dot(proj_k,
|
|
proj_k.conj().T).real * weight
|
|
|
|
# if not symops is None:
|
|
# occ_mats = symmetrize_matrix_set(occ_mats, symops, ions, perm_map)
|
|
|
|
return occ_mats, overlaps
|
|
|
|
################################################################################
|
|
#
|
|
# density_of_states
|
|
#
|
|
################################################################################
|
|
def density_of_states(self, el_struct, emesh):
|
|
"""
|
|
Returns projected DOS for the shell.
|
|
"""
|
|
nion, ns, nk, nlm, nbtot = self.proj_win.shape
|
|
|
|
# There is a problem with data storage structure of projectors that will
|
|
# make life more complicated. The problem is that band-indices of projectors
|
|
# for different k-points do not match because we store 'nb_max' values starting
|
|
# from 0.
|
|
nb_max = self.ib_max - self.ib_min + 1
|
|
ns_band = self.ib_win.shape[1]
|
|
|
|
ne = len(emesh)
|
|
dos = np.zeros((ne, ns, nion, nlm))
|
|
w_k = np.zeros((nk, nb_max, ns, nion, nlm), dtype=np.complex128)
|
|
for isp in xrange(ns):
|
|
for ik in xrange(nk):
|
|
is_b = min(isp, ns_band)
|
|
ib1 = self.ib_win[ik, is_b, 0]
|
|
ib2 = self.ib_win[ik, is_b, 1] + 1
|
|
for ib_g in xrange(ib1, ib2):
|
|
for io in xrange(nion):
|
|
# Note the difference between 'ib' and 'ibn':
|
|
# 'ib' counts from 0 to 'nb_k - 1'
|
|
# 'ibn' counts from 'ib1 - ib_min' to 'ib2 - ib_min'
|
|
ib = ib_g - ib1
|
|
ibn = ib_g - self.ib_min
|
|
proj_k = self.proj_win[io, isp, ik, :, ib]
|
|
w_k[ik, ib, isp, io, :] = proj_k * proj_k.conj()
|
|
|
|
# eigv_ef = el_struct.eigvals[ik, ib, isp] - el_struct.efermi
|
|
itt = el_struct.kmesh['itet'].T
|
|
# k-indices are starting from 0 in Python
|
|
itt[1:, :] -= 1
|
|
for isp in xrange(ns):
|
|
for ib, eigk in enumerate(el_struct.eigvals[:, self.ib_min:self.ib_max+1, isp].T):
|
|
for ie, e in enumerate(emesh):
|
|
eigk_ef = eigk - el_struct.efermi
|
|
cti = c_atm_dos.dos_weights_3d(eigk_ef, e, itt)
|
|
for im in xrange(nlm):
|
|
for io in xrange(nion):
|
|
dos[ie, isp, io, im] += np.sum((cti * w_k[itt[1:, :], ib, isp, io, im].real).sum(0) * itt[0, :])
|
|
|
|
dos *= 2 * el_struct.kmesh['volt']
|
|
# for isp in xrange(ns):
|
|
# for ik, weight, occ in it.izip(it.count(), kweights, occnums[isp, :, :]):
|
|
# for io in xrange(nion):
|
|
# proj_k = self.proj_win[isp, io, ik, ...]
|
|
# occ_mats[isp, io, :, :] += np.dot(proj_k * occ[ib1:ib2],
|
|
# proj_k.conj().T).real * weight
|
|
# overlaps[isp, io, :, :] += np.dot(proj_k,
|
|
# proj_k.conj().T).real * weight
|
|
|
|
# if not symops is None:
|
|
# occ_mats = symmetrize_matrix_set(occ_mats, symops, ions, perm_map)
|
|
|
|
return dos
|
|
|
|
|
|
|