def get_phi_aGp(self, q_c=None, parallel=True, alldir=False):
if q_c is None:
q_c = self.q_c
qq_v = self.qq_v
optical_limit = self.optical_limit
else:
optical_limit = False
if np.abs(q_c).sum() < 1e-8:
q_c = np.array([0.0001, 0, 0])
optical_limit = True
qq_v = np.dot(q_c, self.bcell_cv)
setups = self.calc.wfs.setups
spos_ac = self.calc.atoms.get_scaled_positions()
kk_Gv = gemmdot(q_c + self.Gvec_Gc, self.bcell_cv.copy(), beta=0.0)
phi_aGp = {}
phiG0_avp = {}
if parallel:
from gpaw.response.parallel import parallel_partition
npw, npw_local, Gstart, Gend = parallel_partition(self.npw,
self.comm.rank,
self.comm.size,
reshape=False)
else:
Gstart = 0
Gend = self.npw
for a, id in enumerate(setups.id_a):
phi_aGp[a] = two_phi_planewave_integrals(kk_Gv, setups[a], Gstart,
Gend)
for iG in range(Gstart, Gend):
phi_aGp[a][iG] *= np.exp(
-1j * 2. * pi * np.dot(q_c + self.Gvec_Gc[iG], spos_ac[a]))
if parallel:
self.comm.sum(phi_aGp[a])
# For optical limit, G == 0 part should change
if optical_limit:
for a, id in enumerate(setups.id_a):
nabla_iiv = setups[a].nabla_iiv
phi_aGp[a][0] = -1j * (np.dot(nabla_iiv, qq_v)).ravel()
phiG0_avp[a] = np.zeros((3, len(phi_aGp[a][0])), complex)
for dir in range(3): # 3 dimension
q2_c = np.diag((1, 1, 1))[dir] * self.qopt
qq2_v = np.dot(q2_c, self.bcell_cv) # summation over c
phiG0_avp[a][dir] = -1j * (np.dot(nabla_iiv,
qq2_v)).ravel()
if alldir:
return phi_aGp, phiG0_avp
else:
return phi_aGp
def get_phi_aGp(self, q_c=None, parallel=True, alldir=False):
if q_c is None:
q_c = self.q_c
qq_v = self.qq_v
optical_limit = self.optical_limit
else:
optical_limit = False
if np.abs(q_c).sum() < 1e-8:
q_c = np.array([0.0001, 0, 0])
optical_limit = True
qq_v = np.dot(q_c, self.bcell_cv)
setups = self.calc.wfs.setups
spos_ac = self.calc.atoms.get_scaled_positions()
kk_Gv = gemmdot(q_c + self.Gvec_Gc, self.bcell_cv.copy(), beta=0.0)
phi_aGp = {}
phiG0_avp = {}
if parallel:
from gpaw.response.parallel import parallel_partition
npw, npw_local, Gstart, Gend = parallel_partition(
self.npw, self.comm.rank, self.comm.size, reshape=False)
else:
Gstart = 0
Gend = self.npw
for a, id in enumerate(setups.id_a):
phi_aGp[a] = two_phi_planewave_integrals(kk_Gv, setups[a], Gstart, Gend)
for iG in range(Gstart, Gend):
phi_aGp[a][iG] *= np.exp(-1j * 2. * pi *
np.dot(q_c + self.Gvec_Gc[iG], spos_ac[a]) )
if parallel:
self.comm.sum(phi_aGp[a])
# For optical limit, G == 0 part should change
if optical_limit:
for a, id in enumerate(setups.id_a):
nabla_iiv = setups[a].nabla_iiv
phi_aGp[a][0] = -1j * (np.dot(nabla_iiv, qq_v)).ravel()
phiG0_avp[a] = np.zeros((3, len(phi_aGp[a][0])), complex)
for dir in range(3): # 3 dimension
q2_c = np.diag((1,1,1))[dir] * self.qopt
qq2_v = np.dot(q2_c, self.bcell_cv) # summation over c
phiG0_avp[a][dir] = -1j * (np.dot(nabla_iiv, qq2_v)).ravel()
if alldir:
return phi_aGp, phiG0_avp
else:
return phi_aGp
def initialize(self):
self.printtxt('')
self.printtxt('-----------------------------------------------')
self.printtxt('Belth Selpeter Equation calculation started at:')
self.printtxt(ctime())
BASECHI.initialize(self)
calc = self.calc
self.kd = kd = calc.wfs.kd
# frequency points init
self.dw = self.w_w[1] - self.w_w[0]
assert ((self.w_w[1:] - self.w_w[:-1] - self.dw) <
1e-10).all() # make sure its linear w grid
assert self.w_w.max() == self.w_w[-1]
self.dw /= Hartree
self.w_w /= Hartree
self.wmax = self.w_w[-1]
self.Nw = int(self.wmax / self.dw) + 1
# find the pair index and initialized pair energy (e_i - e_j) and occupation(f_i-f_j)
self.e_S = {}
focc_s = {}
self.Sindex_S3 = {}
iS = 0
kq_k = self.kq_k
for k1 in range(self.nkpt):
ibzkpt1 = kd.kibz_k[k1]
ibzkpt2 = kd.kibz_k[kq_k[k1]]
for n1 in range(self.nbands):
for m1 in range(self.nbands):
focc = self.f_kn[ibzkpt1, n1] - self.f_kn[ibzkpt2, m1]
if np.abs(focc) > self.ftol:
self.e_S[iS] = self.e_kn[ibzkpt2,
m1] - self.e_kn[ibzkpt1, n1]
focc_s[iS] = focc
self.Sindex_S3[iS] = (k1, n1, m1)
iS += 1
self.nS = iS
self.focc_S = np.zeros(self.nS)
for iS in range(self.nS):
self.focc_S[iS] = focc_s[iS]
# parallel init
self.Scomm = world
# kcomm and wScomm is only to be used when wavefunctions r parallelly distributed.
self.kcomm = world
self.wScomm = serial_comm
self.nS, self.nS_local, self.nS_start, self.nS_end = parallel_partition(
self.nS, self.Scomm.rank, self.Scomm.size, reshape=False)
self.print_bse()
self.get_phi_aGp()
# Coulomb kernel init
self.kc_G = np.zeros(self.npw)
for iG in range(self.npw):
index = self.Gindex_G[iG]
qG = np.dot(self.q_c + self.Gvec_Gc[iG], self.bcell_cv)
self.kc_G[iG] = 1. / np.inner(qG, qG)
if self.optical_limit:
self.kc_G[0] = 0.
self.printtxt('')
return
def initialize(self):
self.ini = True
BASECHI.initialize(self)
self.txtname = self.gwtxtname
self.output_init()
self.printtxt('GPAW version %s' % (version))
self.printtxt('-----------------------------------------------')
self.printtxt('GW calculation started at:')
self.printtxt(ctime())
self.printtxt('-----------------------------------------------')
self.starttime = time()
calc = self.calc
kd = self.kd
# band init
if self.gwnbands is None:
if self.npw > calc.wfs.bd.nbands:
self.nbands = calc.wfs.bd.nbands
else:
self.nbands = self.npw
# eigenvalue init
if self.user_skn is not None:
self.printtxt('Use eigenvalues from user.')
assert np.shape(
self.user_skn
)[0] == self.nspins, 'Eigenvalues not compatible with .gpw file!'
assert np.shape(
self.user_skn
)[1] == self.kd.nibzkpts, 'Eigenvalues not compatible with .gpw file!'
assert np.shape(
self.user_skn)[2] >= self.nbands, 'Too few eigenvalues!'
self.e_skn = self.user_skn
else:
self.printtxt('Use eigenvalues from the calculator.')
# q point init
self.bzq_kc = kd.get_bz_q_points()
self.ibzq_qc = self.bzq_kc # q point symmetry is not used at the moment.
self.nqpt = np.shape(self.bzq_kc)[0]
# frequency points init
self.static = False
if self.ppa: # Plasmon Pole Approximation
if self.E0 is None:
self.E0 = Hartree
self.E0 /= Hartree
self.w_w = np.array([0., 1j * self.E0])
self.hilbert_trans = False
self.wpar = 1
elif self.w_w is None: # static COHSEX
self.w_w = np.array([0.])
self.static = True
self.hilbert_trans = False
self.wpar = 1
self.eta = 0.0001 / Hartree
else:
# create nonlinear frequency grid
# grid is linear from 0 to wcut with spacing dw
# spacing is linearily increasing between wcut and wmax
# Hilbert transforms are still carried out on linear grid
wcut = self.w_w[0]
wmax = self.w_w[1]
dw = self.w_w[2]
w_w = np.linspace(0., wcut, wcut / dw + 1)
i = 1
wi = wcut
while wi < wmax:
wi += i * dw
w_w = np.append(w_w, wi)
i += 1
while len(w_w) % self.wpar != 0:
wi += i * dw
w_w = np.append(w_w, wi)
i += 1
dw_w = np.zeros(len(w_w))
dw_w[0] = dw
dw_w[1:] = w_w[1:] - w_w[:-1]
self.w_w = w_w / Hartree
self.dw_w = dw_w / Hartree
self.eta_w = dw_w * 4 / Hartree
self.wcut = wcut
self.wmax = self.w_w[-1]
self.wmin = self.w_w[0]
self.dw = self.w_w[1] - self.w_w[0]
self.Nw = len(self.w_w)
# self.wpar = int(self.Nw * self.npw**2 * 16. / 1024**2) // 1500 + 1 # estimate memory and parallelize over frequencies
for s in range(self.nspins):
emaxdiff = self.e_skn[s][:, self.nbands -
1].max() - self.e_skn[s][:, 0].min()
assert (self.wmax > emaxdiff
), 'Maximum frequency must be larger than %f' % (
emaxdiff * Hartree)
# GW kpoints init
if self.kpoints is None:
self.gwnkpt = self.kd.nibzkpts
self.gwkpt_k = kd.ibz2bz_k
else:
self.gwnkpt = np.shape(self.kpoints)[0]
self.gwkpt_k = self.kpoints
# GW bands init
if self.bands is None:
self.gwnband = self.nbands
self.bands = self.gwbands_n = np.arange(self.nbands)
else:
self.gwnband = np.shape(self.bands)[0]
self.gwbands_n = self.bands
self.alpha = 1j / (2 * pi * self.vol * self.kd.nbzkpts)
# parallel init
assert len(self.w_w) % self.wpar == 0
self.wcommsize = self.wpar
self.qcommsize = size // self.wpar
assert self.qcommsize * self.wcommsize == size, 'wpar must be integer divisor of number of requested cores'
if self.nqpt != 1: # parallelize over q-points
self.wcomm, self.qcomm, self.worldcomm = set_communicator(
world, rank, size, self.wpar)
self.ncomm = serial_comm
self.dfcomm = self.wcomm
self.kcommsize = 1
else: # parallelize over bands
self.wcomm, self.ncomm, self.worldcomm = set_communicator(
world, rank, size, self.wpar)
self.qcomm = serial_comm
if self.wpar > 1:
self.dfcomm = self.wcomm
self.kcommsize = 1
else:
self.dfcomm = self.ncomm
self.kcommsize = self.ncomm.size
nq, self.nq_local, self.q_start, self.q_end = parallel_partition(
self.nqpt, self.qcomm.rank, self.qcomm.size, reshape=False)
nb, self.nbands_local, self.m_start, self.m_end = parallel_partition(
self.nbands, self.ncomm.rank, self.ncomm.size, reshape=False)
def parallel_init(self):
"""Parallel initialization. By default, only use kcomm and wcomm.
Parameters:
kcomm:
kpoint communicator
wScomm:
spectral function communicator
wcomm:
frequency communicator
"""
if extra_parameters.get("df_dry_run"):
from gpaw.mpi import DryRunCommunicator
size = extra_parameters["df_dry_run"]
world = DryRunCommunicator(size)
rank = world.rank
self.comm = world
else:
world = self.comm
rank = self.comm.rank
size = self.comm.size
wcommsize = int(self.NwS * self.npw ** 2 * 16.0 / 1024 ** 2) // 1500 # megabyte
wcommsize += 1
if size < wcommsize:
raise ValueError("Number of cpus are not enough ! ")
if self.kcommsize is None:
self.kcommsize = world.size
if wcommsize > size // self.kcommsize: # if matrix too large, overwrite kcommsize and distribute matrix
self.printtxt("kcommsize is over written ! ")
while size % wcommsize != 0:
wcommsize += 1
self.kcommsize = size // wcommsize
assert self.kcommsize * wcommsize == size
if self.kcommsize < 1:
raise ValueError("Number of cpus are not enough ! ")
self.kcomm, self.wScomm, self.wcomm = set_communicator(world, rank, size, self.kcommsize)
if self.kd.nbzkpts >= world.size:
self.nkpt_reshape = self.kd.nbzkpts
self.nkpt_reshape, self.nkpt_local, self.kstart, self.kend = parallel_partition(
self.nkpt_reshape, self.kcomm.rank, self.kcomm.size, reshape=True, positive=True
)
self.mband_local = self.nvalbands
self.mlist = np.arange(self.nbands)
else:
# if number of kpoints == 1, use band parallelization
self.nkpt_local = self.kd.nbzkpts
self.kstart = 0
self.kend = self.kd.nbzkpts
self.nkpt_reshape = self.kd.nbzkpts
self.nbands, self.mband_local, self.mlist = parallel_partition_list(
self.nbands, self.kcomm.rank, self.kcomm.size
)
if self.NwS % size != 0:
self.NwS -= self.NwS % size
self.NwS, self.NwS_local, self.wS1, self.wS2 = parallel_partition(
self.NwS, self.wScomm.rank, self.wScomm.size, reshape=False
)
if self.hilbert_trans:
self.Nw, self.Nw_local, self.wstart, self.wend = parallel_partition(
self.Nw, self.wcomm.rank, self.wcomm.size, reshape=True
)
else:
if self.Nw > 1:
# assert self.Nw % (self.comm.size / self.kcomm.size) == 0
self.wcomm = self.wScomm
self.Nw, self.Nw_local, self.wstart, self.wend = parallel_partition(
self.Nw, self.wcomm.rank, self.wcomm.size, reshape=False
)
else:
# if frequency point is too few, then dont parallelize
self.wcomm = serial_comm
self.wstart = 0
self.wend = self.Nw
self.Nw_local = self.Nw
return
def initialize(self):
self.ini = True
BASECHI.initialize(self)
self.txtname = self.gwtxtname
self.output_init()
self.printtxt('GPAW version %s' %(version))
self.printtxt('-----------------------------------------------')
self.printtxt('GW calculation started at:')
self.printtxt(ctime())
self.printtxt('-----------------------------------------------')
self.starttime = time()
calc = self.calc
kd = self.kd
# band init
if self.gwnbands is None:
if self.npw > calc.wfs.bd.nbands:
self.nbands = calc.wfs.bd.nbands
else:
self.nbands = self.npw
# eigenvalue init
if self.user_skn is not None:
self.printtxt('Use eigenvalues from user.')
assert np.shape(self.user_skn)[0] == self.nspins, 'Eigenvalues not compatible with .gpw file!'
assert np.shape(self.user_skn)[1] == self.kd.nibzkpts, 'Eigenvalues not compatible with .gpw file!'
assert np.shape(self.user_skn)[2] >= self.nbands, 'Too few eigenvalues!'
self.e_skn = self.user_skn
else:
self.printtxt('Use eigenvalues from the calculator.')
# q point init
self.bzq_kc = kd.get_bz_q_points()
self.ibzq_qc = self.bzq_kc # q point symmetry is not used at the moment.
self.nqpt = np.shape(self.bzq_kc)[0]
# frequency points init
self.static=False
if self.ppa: # Plasmon Pole Approximation
if self.E0 is None:
self.E0 = Hartree
self.E0 /= Hartree
self.w_w = np.array([0., 1j*self.E0])
self.hilbert_trans=False
self.wpar=1
elif self.w_w is None: # static COHSEX
self.w_w = np.array([0.])
self.static=True
self.hilbert_trans=False
self.wpar=1
self.eta = 0.0001 / Hartree
else:
# create nonlinear frequency grid
# grid is linear from 0 to wcut with spacing dw
# spacing is linearily increasing between wcut and wmax
# Hilbert transforms are still carried out on linear grid
wcut = self.w_w[0]
wmax = self.w_w[1]
dw = self.w_w[2]
w_w = np.linspace(0., wcut, wcut/dw+1)
i=1
wi=wcut
while wi < wmax:
wi += i*dw
w_w = np.append(w_w, wi)
i+=1
while len(w_w) % self.wpar != 0:
wi += i*dw
w_w = np.append(w_w, wi)
i+=1
dw_w = np.zeros(len(w_w))
dw_w[0] = dw
dw_w[1:] = w_w[1:] - w_w[:-1]
self.w_w = w_w / Hartree
self.dw_w = dw_w / Hartree
self.eta_w = dw_w * 4 / Hartree
self.wcut = wcut
self.wmax = self.w_w[-1]
self.wmin = self.w_w[0]
self.dw = self.w_w[1] - self.w_w[0]
self.Nw = len(self.w_w)
# self.wpar = int(self.Nw * self.npw**2 * 16. / 1024**2) // 1500 + 1 # estimate memory and parallelize over frequencies
for s in range(self.nspins):
emaxdiff = self.e_skn[s][:,self.nbands-1].max() - self.e_skn[s][:,0].min()
assert (self.wmax > emaxdiff), 'Maximum frequency must be larger than %f' %(emaxdiff*Hartree)
# GW kpoints init
if self.kpoints is None:
self.gwnkpt = self.kd.nibzkpts
self.gwkpt_k = kd.ibz2bz_k
else:
self.gwnkpt = np.shape(self.kpoints)[0]
self.gwkpt_k = self.kpoints
# GW bands init
if self.bands is None:
self.gwnband = self.nbands
self.bands = self.gwbands_n = np.arange(self.nbands)
else:
self.gwnband = np.shape(self.bands)[0]
self.gwbands_n = self.bands
self.alpha = 1j/(2*pi * self.vol * self.kd.nbzkpts)
# parallel init
assert len(self.w_w) % self.wpar == 0
self.wcommsize = self.wpar
self.qcommsize = size // self.wpar
assert self.qcommsize * self.wcommsize == size, 'wpar must be integer divisor of number of requested cores'
if self.nqpt != 1: # parallelize over q-points
self.wcomm, self.qcomm, self.worldcomm = set_communicator(world, rank, size, self.wpar)
self.ncomm = serial_comm
self.dfcomm = self.wcomm
self.kcommsize = 1
else: # parallelize over bands
self.wcomm, self.ncomm, self.worldcomm = set_communicator(world, rank, size, self.wpar)
self.qcomm = serial_comm
if self.wpar > 1:
self.dfcomm = self.wcomm
self.kcommsize = 1
else:
self.dfcomm = self.ncomm
self.kcommsize = self.ncomm.size
nq, self.nq_local, self.q_start, self.q_end = parallel_partition(
self.nqpt, self.qcomm.rank, self.qcomm.size, reshape=False)
nb, self.nbands_local, self.m_start, self.m_end = parallel_partition(
self.nbands, self.ncomm.rank, self.ncomm.size, reshape=False)
def parallel_init(self):
"""Parallel initialization. By default, only use kcomm and wcomm.
Parameters:
kcomm:
kpoint communicator
wScomm:
spectral function communicator
wcomm:
frequency communicator
"""
if extra_parameters.get('df_dry_run'):
from gpaw.mpi import DryRunCommunicator
size = extra_parameters['df_dry_run']
world = DryRunCommunicator(size)
rank = world.rank
self.comm = world
else:
world = self.comm
rank = self.comm.rank
size = self.comm.size
wcommsize = int(self.NwS * self.npw**2 * 16. / 1024**2) // 1500 # megabyte
wcommsize += 1
if size < wcommsize:
raise ValueError('Number of cpus are not enough ! ')
if self.kcommsize is None:
self.kcommsize = world.size
if wcommsize > size // self.kcommsize: # if matrix too large, overwrite kcommsize and distribute matrix
self.printtxt('kcommsize is over written ! ')
while size % wcommsize != 0:
wcommsize += 1
self.kcommsize = size // wcommsize
assert self.kcommsize * wcommsize == size
if self.kcommsize < 1:
raise ValueError('Number of cpus are not enough ! ')
self.kcomm, self.wScomm, self.wcomm = set_communicator(world, rank, size, self.kcommsize)
if self.kd.nbzkpts >= world.size:
self.nkpt_reshape = self.kd.nbzkpts
self.nkpt_reshape, self.nkpt_local, self.kstart, self.kend = parallel_partition(
self.nkpt_reshape, self.kcomm.rank, self.kcomm.size, reshape=True, positive=True)
self.mband_local = self.nvalbands
self.mlist = np.arange(self.nbands)
else:
# if number of kpoints == 1, use band parallelization
self.nkpt_local = self.kd.nbzkpts
self.kstart = 0
self.kend = self.kd.nbzkpts
self.nkpt_reshape = self.kd.nbzkpts
self.nbands, self.mband_local, self.mlist = parallel_partition_list(
self.nbands, self.kcomm.rank, self.kcomm.size)
if self.NwS % size != 0:
self.NwS -= self.NwS % size
self.NwS, self.NwS_local, self.wS1, self.wS2 = parallel_partition(
self.NwS, self.wScomm.rank, self.wScomm.size, reshape=False)
if self.hilbert_trans:
self.Nw, self.Nw_local, self.wstart, self.wend = parallel_partition(
self.Nw, self.wcomm.rank, self.wcomm.size, reshape=True)
else:
if self.Nw > 1:
# assert self.Nw % (self.comm.size / self.kcomm.size) == 0
self.wcomm = self.wScomm
self.Nw, self.Nw_local, self.wstart, self.wend = parallel_partition(
self.Nw, self.wcomm.rank, self.wcomm.size, reshape=False)
else:
# if frequency point is too few, then dont parallelize
self.wcomm = serial_comm
self.wstart = 0
self.wend = self.Nw
self.Nw_local = self.Nw
return
def initialize(self):
self.printtxt('')
self.printtxt('-----------------------------------------------')
self.printtxt('Belth Selpeter Equation calculation started at:')
self.printtxt(ctime())
BASECHI.initialize(self)
calc = self.calc
self.kd = kd = calc.wfs.kd
# frequency points init
self.dw = self.w_w[1] - self.w_w[0]
assert ((self.w_w[1:] - self.w_w[:-1] - self.dw) < 1e-10).all() # make sure its linear w grid
assert self.w_w.max() == self.w_w[-1]
self.dw /= Hartree
self.w_w /= Hartree
self.wmax = self.w_w[-1]
self.Nw = int(self.wmax / self.dw) + 1
# find the pair index and initialized pair energy (e_i - e_j) and occupation(f_i-f_j)
self.e_S = {}
focc_s = {}
self.Sindex_S3 = {}
iS = 0
kq_k = self.kq_k
for k1 in range(self.nkpt):
ibzkpt1 = kd.kibz_k[k1]
ibzkpt2 = kd.kibz_k[kq_k[k1]]
for n1 in range(self.nbands):
for m1 in range(self.nbands):
focc = self.f_kn[ibzkpt1,n1] - self.f_kn[ibzkpt2,m1]
if np.abs(focc) > self.ftol:
self.e_S[iS] =self.e_kn[ibzkpt2,m1] - self.e_kn[ibzkpt1,n1]
focc_s[iS] = focc
self.Sindex_S3[iS] = (k1, n1, m1)
iS += 1
self.nS = iS
self.focc_S = np.zeros(self.nS)
for iS in range(self.nS):
self.focc_S[iS] = focc_s[iS]
# parallel init
self.Scomm = world
# kcomm and wScomm is only to be used when wavefunctions r parallelly distributed.
self.kcomm = world
self.wScomm = serial_comm
self.nS, self.nS_local, self.nS_start, self.nS_end = parallel_partition(
self.nS, self.Scomm.rank, self.Scomm.size, reshape=False)
self.print_bse()
self.get_phi_aGp()
# Coulomb kernel init
self.kc_G = np.zeros(self.npw)
for iG in range(self.npw):
index = self.Gindex_G[iG]
qG = np.dot(self.q_c + self.Gvec_Gc[iG], self.bcell_cv)
self.kc_G[iG] = 1. / np.inner(qG, qG)
if self.optical_limit:
self.kc_G[0] = 0.
self.printtxt('')
return
def initialize(self):
self.printtxt('----------------------------------------')
self.printtxt('Bethe-Salpeter Equation calculation')
self.printtxt('----------------------------------------')
self.printtxt('Started at: %s' % ctime())
self.printtxt('')
BASECHI.initialize(self)
assert self.nspins == 1
calc = self.calc
self.kd = kd = calc.wfs.kd
# frequency points init
self.dw = self.w_w[1] - self.w_w[0]
assert ((self.w_w[1:] - self.w_w[:-1] - self.dw) < 1e-10).all() # make sure its linear w grid
assert self.w_w.max() == self.w_w[-1]
self.dw /= Hartree
self.w_w /= Hartree
self.wmax = self.w_w[-1]
self.Nw = int(self.wmax / self.dw) + 1
# band init
if self.nc is None:
nv = self.nvalence / 2 - 1
self.nv = np.array([nv, nv+1]) # conduction band start / end
self.nc = np.array([nv+1, nv+2]) # valence band start / end
self.printtxt('')
self.printtxt('Number of electrons : %d' % (self.nvalence))
self.printtxt('Valence band included : (band %d to band %d)' %(self.nv[0], self.nv[1]-1))
self.printtxt('Conduction band included : (band %d to band %d)' %(self.nc[0], self.nc[1]-1))
if self.eshift is not None:
self.printtxt('Scissors operator : %2.3f eV' % self.eshift)
self.printtxt('')
# find the pair index and initialized pair energy (e_i - e_j) and occupation(f_i-f_j)
self.e_S = {}
focc_s = {}
self.Sindex_S3 = {}
iS = 0
kq_k = self.kq_k
for k1 in range(self.kd.nbzkpts):
ibzkpt1 = kd.bz2ibz_k[k1]
ibzkpt2 = kd.bz2ibz_k[kq_k[k1]]
for n1 in range(self.nv[0], self.nv[1]):
for m1 in range(self.nc[0], self.nc[1]):
focc = self.f_skn[0][ibzkpt1,n1] - self.f_skn[0][ibzkpt2,m1]
if self.coupling: # Dont use Tamm-Dancoff Approx.
check_ftol = np.abs(focc) > self.ftol
else:
check_ftol = focc > self.ftol
if check_ftol:
if self.gw_skn is None:
self.e_S[iS] = self.e_skn[0][ibzkpt2,m1] - self.e_skn[0][ibzkpt1,n1]
else:
self.e_S[iS] = self.gw_skn[0][ibzkpt2,m1] - self.gw_skn[0][ibzkpt1,n1]
focc_s[iS] = focc
self.Sindex_S3[iS] = (k1, n1, m1)
iS += 1
self.nS = iS
self.focc_S = np.zeros(self.nS)
for iS in range(self.nS):
self.focc_S[iS] = focc_s[iS]
# q points init
self.bzq_qc = kd.get_bz_q_points()
if not self.qsymm:
self.ibzq_qc = self.bzq_qc
else:
(self.ibzq_qc, self.ibzq_q, self.iop_q,
self.timerev_q, self.diff_qc) = kd.get_ibz_q_points(self.bzq_qc,
calc.wfs.symmetry.op_scc)
if np.abs(self.bzq_qc - kd.bzk_kc).sum() < 1e-8:
assert np.abs(self.ibzq_qc - kd.ibzk_kc).sum() < 1e-8
self.nibzq = len(self.ibzq_qc)
# Parallel initialization
# kcomm and wScomm is only to be used when wavefunctions are distributed in parallel.
self.comm = self.Scomm = world
self.kcomm = world
self.wScomm = serial_comm
self.nS, self.nS_local, self.nS_start, self.nS_end = parallel_partition(
self.nS, world.rank, world.size, reshape=False)
self.print_bse()
if calc.input_parameters['mode'] == 'lcao':
calc.initialize_positions()
# Coulomb interaction at q=0 for Hartree coupling
### 2D z direction only !!!!!!!!!!!!!!!!!!!!!!
if self.integrate_coulomb is None:
self.integrate_coulomb = []
if self.vcut is None:
pass
elif self.vcut == '2D':
for iG in range(len(self.Gvec_Gc)):
if self.Gvec_Gc[iG, 0] == 0 and self.Gvec_Gc[iG, 1] == 0:
self.integrate_coulomb.append(iG)
else:
raise NotImplementedError
elif type(self.integrate_coulomb) is int:
self.integrate_coulomb = range(self.integrate_coulomb)
elif self.integrate_coulomb == 'all':
self.integrate_coulomb = range(len(self.Gvec_Gc))
elif type(self.integrate_coulomb) is list:
pass
else:
raise 'Invalid option for integrate_coulomb'
self.printtxt('')
self.printtxt('Calculating bare Coulomb kernel')
if not len(self.integrate_coulomb) == 0:
self.printtxt('Integrating Coulomb kernel at %s reciprocal lattice vector(s)' % len(self.integrate_coulomb))
# Coulomb interaction at problematic G's for exchange coupling
if len(self.integrate_coulomb) != 0:
self.vint_Gq = []
for iG in self.integrate_coulomb:
v_q, v0_q = calculate_Kc_q(self.acell_cv,
self.bcell_cv,
self.pbc,
self.kd.N_c,
vcut=self.vcut,
Gvec_c=self.Gvec_Gc[iG],
q_qc=self.ibzq_qc.copy())
self.vint_Gq.append(v_q)
if self.print_coulomb:
self.printtxt('')
self.printtxt('Average kernel relative to bare kernel - \int v(q)dq / v(q0): ')
self.printtxt(' G: % s' % self.Gvec_Gc[iG])
for iq in range(len(v_q)):
q_s = ' q = [%1.2f, %1.2f, %1.2f]: ' % (self.ibzq_qc[iq,0],
self.ibzq_qc[iq,1],
self.ibzq_qc[iq,2])
v_rel = v_q[iq] / v0_q[iq]
self.printtxt(q_s + '%1.3f' % v_rel)
self.printtxt('')
self.V_qGG = self.full_bare_interaction()
return
def get_phi_qaGp(self):
N1_max = 0
N2_max = 0
natoms = len(self.calc.wfs.setups)
for id in range(natoms):
N1 = self.npw
N2 = self.calc.wfs.setups[id].ni**2
if N1 > N1_max:
N1_max = N1
if N2 > N2_max:
N2_max = N2
nbzq = self.kd.nbzkpts
nbzq, nq_local, q_start, q_end = parallel_partition(
nbzq, world.rank, world.size, reshape=False)
phimax_qaGp = np.zeros((nq_local, natoms, N1_max, N2_max), dtype=complex)
#phimax_qaGp = np.zeros((nbzq, natoms, N1_max, N2_max), dtype=complex)
t0 = time()
for iq in range(nq_local):
q_c = self.bzq_qc[iq + q_start]
tmp_aGp = self.get_phi_aGp(q_c, parallel=False)
for id in range(natoms):
N1, N2 = tmp_aGp[id].shape
phimax_qaGp[iq, id, :N1, :N2] = tmp_aGp[id]
self.timing(iq*world.size, t0, nq_local, 'iq')
world.barrier()
# Write to disk
filename = 'phi_qaGp'
if world.rank == 0:
w = Writer(filename)
w.dimension('nbzq', nbzq)
w.dimension('natoms', natoms)
w.dimension('nG', N1_max)
w.dimension('nii', N2_max)
w.add('phi_qaGp', ('nbzq', 'natoms', 'nG', 'nii',), dtype=complex)
for q in range(nbzq):
residual = nbzq % size
N_local = nbzq // size
if q < residual * (N_local + 1):
qrank = q // (N_local + 1)
else:
qrank = (q - residual * (N_local + 1)) // N_local + residual
if qrank == 0:
if world.rank == 0:
phi_aGp = phimax_qaGp[q - q_start]
else:
if world.rank == qrank:
phi_aGp = phimax_qaGp[q - q_start]
world.send(phi_aGp, 0, q)
elif world.rank == 0:
world.receive(phi_aGp, qrank, q)
if world.rank == 0:
w.fill(phi_aGp)
world.barrier()
if world.rank == 0:
w.close()
return
def initialize(self):
self.printtxt('----------------------------------------')
self.printtxt('Bethe-Salpeter Equation calculation')
self.printtxt('----------------------------------------')
self.printtxt('Started at: %s' % ctime())
self.printtxt('')
BASECHI.initialize(self)
assert self.nspins == 1
calc = self.calc
self.kd = kd = calc.wfs.kd
# frequency points init
self.dw = self.w_w[1] - self.w_w[0]
assert ((self.w_w[1:] - self.w_w[:-1] - self.dw) <
1e-10).all() # make sure its linear w grid
assert self.w_w.max() == self.w_w[-1]
self.dw /= Hartree
self.w_w /= Hartree
self.wmax = self.w_w[-1]
self.Nw = int(self.wmax / self.dw) + 1
# band init
if self.nc is None:
nv = self.nvalence / 2 - 1
self.nv = np.array([nv, nv + 1]) # conduction band start / end
self.nc = np.array([nv + 1, nv + 2]) # valence band start / end
self.printtxt('')
self.printtxt('Number of electrons : %d' % (self.nvalence))
self.printtxt('Valence band included : (band %d to band %d)' %
(self.nv[0], self.nv[1] - 1))
self.printtxt('Conduction band included : (band %d to band %d)' %
(self.nc[0], self.nc[1] - 1))
if self.eshift is not None:
self.printtxt('Scissors operator : %2.3f eV' % self.eshift)
self.printtxt('')
# find the pair index and initialized pair energy (e_i - e_j) and occupation(f_i-f_j)
self.e_S = {}
focc_s = {}
self.Sindex_S3 = {}
iS = 0
kq_k = self.kq_k
for k1 in range(self.kd.nbzkpts):
ibzkpt1 = kd.bz2ibz_k[k1]
ibzkpt2 = kd.bz2ibz_k[kq_k[k1]]
for n1 in range(self.nv[0], self.nv[1]):
for m1 in range(self.nc[0], self.nc[1]):
focc = self.f_skn[0][ibzkpt1, n1] - self.f_skn[0][ibzkpt2,
m1]
if self.coupling: # Dont use Tamm-Dancoff Approx.
check_ftol = np.abs(focc) > self.ftol
else:
check_ftol = focc > self.ftol
if check_ftol:
if self.gw_skn is None:
self.e_S[iS] = self.e_skn[0][
ibzkpt2, m1] - self.e_skn[0][ibzkpt1, n1]
else:
self.e_S[iS] = self.gw_skn[0][
ibzkpt2, m1] - self.gw_skn[0][ibzkpt1, n1]
focc_s[iS] = focc
self.Sindex_S3[iS] = (k1, n1, m1)
iS += 1
self.nS = iS
self.focc_S = np.zeros(self.nS)
for iS in range(self.nS):
self.focc_S[iS] = focc_s[iS]
# q points init
self.bzq_qc = kd.get_bz_q_points()
if not self.qsymm:
self.ibzq_qc = self.bzq_qc
else:
(self.ibzq_qc, self.ibzq_q, self.iop_q, self.timerev_q,
self.diff_qc) = kd.get_ibz_q_points(self.bzq_qc,
calc.wfs.kd.symmetry.op_scc)
if np.abs(self.bzq_qc - kd.bzk_kc).sum() < 1e-8:
assert np.abs(self.ibzq_qc - kd.ibzk_kc).sum() < 1e-8
self.nibzq = len(self.ibzq_qc)
# Parallel initialization
# kcomm and wScomm is only to be used when wavefunctions are distributed in parallel.
self.comm = self.Scomm = world
self.kcomm = world
self.wScomm = serial_comm
self.nS, self.nS_local, self.nS_start, self.nS_end = parallel_partition(
self.nS, world.rank, world.size, reshape=False)
self.print_bse()
if calc.input_parameters['mode'] == 'lcao':
calc.initialize_positions()
# Coulomb interaction at q=0 for Hartree coupling
### 2D z direction only !!!!!!!!!!!!!!!!!!!!!!
if self.integrate_coulomb is None:
self.integrate_coulomb = []
if self.vcut is None:
pass
elif self.vcut == '2D':
for iG in range(len(self.Gvec_Gc)):
if self.Gvec_Gc[iG, 0] == 0 and self.Gvec_Gc[iG, 1] == 0:
self.integrate_coulomb.append(iG)
else:
raise NotImplementedError
elif type(self.integrate_coulomb) is int:
self.integrate_coulomb = range(self.integrate_coulomb)
elif self.integrate_coulomb == 'all':
self.integrate_coulomb = range(len(self.Gvec_Gc))
elif type(self.integrate_coulomb) is list:
pass
else:
raise 'Invalid option for integrate_coulomb'
self.printtxt('')
self.printtxt('Calculating bare Coulomb kernel')
if not len(self.integrate_coulomb) == 0:
self.printtxt(
'Integrating Coulomb kernel at %s reciprocal lattice vector(s)'
% len(self.integrate_coulomb))
# Coulomb interaction at problematic G's for exchange coupling
if len(self.integrate_coulomb) != 0:
self.vint_Gq = []
for iG in self.integrate_coulomb:
v_q, v0_q = calculate_Kc_q(self.acell_cv,
self.bcell_cv,
self.pbc,
self.kd.N_c,
vcut=self.vcut,
Gvec_c=self.Gvec_Gc[iG],
q_qc=self.ibzq_qc.copy())
self.vint_Gq.append(v_q)
if self.print_coulomb:
self.printtxt('')
self.printtxt(
'Average kernel relative to bare kernel - \int v(q)dq / v(q0): '
)
self.printtxt(' G: % s' % self.Gvec_Gc[iG])
for iq in range(len(v_q)):
q_s = ' q = [%1.2f, %1.2f, %1.2f]: ' % (
self.ibzq_qc[iq, 0], self.ibzq_qc[iq, 1],
self.ibzq_qc[iq, 2])
v_rel = v_q[iq] / v0_q[iq]
self.printtxt(q_s + '%1.3f' % v_rel)
self.printtxt('')
self.V_qGG = self.full_bare_interaction()
def get_phi_qaGp(self):
N1_max = 0
N2_max = 0
natoms = len(self.calc.wfs.setups)
for id in range(natoms):
N1 = self.npw
N2 = self.calc.wfs.setups[id].ni**2
if N1 > N1_max:
N1_max = N1
if N2 > N2_max:
N2_max = N2
nbzq = self.kd.nbzkpts
nbzq, nq_local, q_start, q_end = parallel_partition(nbzq,
world.rank,
world.size,
reshape=False)
phimax_qaGp = np.zeros((nq_local, natoms, N1_max, N2_max),
dtype=complex)
#phimax_qaGp = np.zeros((nbzq, natoms, N1_max, N2_max), dtype=complex)
t0 = time()
for iq in range(nq_local):
q_c = self.bzq_qc[iq + q_start]
tmp_aGp = self.get_phi_aGp(q_c, parallel=False)
for id in range(natoms):
N1, N2 = tmp_aGp[id].shape
phimax_qaGp[iq, id, :N1, :N2] = tmp_aGp[id]
self.timing(iq * world.size, t0, nq_local, 'iq')
world.barrier()
# Write to disk
filename = 'phi_qaGp'
if world.rank == 0:
w = Writer(filename)
w.dimension('nbzq', nbzq)
w.dimension('natoms', natoms)
w.dimension('nG', N1_max)
w.dimension('nii', N2_max)
w.add('phi_qaGp', (
'nbzq',
'natoms',
'nG',
'nii',
), dtype=complex)
for q in range(nbzq):
residual = nbzq % size
N_local = nbzq // size
if q < residual * (N_local + 1):
qrank = q // (N_local + 1)
else:
qrank = (q - residual * (N_local + 1)) // N_local + residual
if qrank == 0:
if world.rank == 0:
phi_aGp = phimax_qaGp[q - q_start]
else:
if world.rank == qrank:
phi_aGp = phimax_qaGp[q - q_start]
world.send(phi_aGp, 0, q)
elif world.rank == 0:
world.receive(phi_aGp, qrank, q)
if world.rank == 0:
w.fill(phi_aGp)
if world.rank == 0:
w.close()
world.barrier()
