def initialize(self, wfs):
if wfs.bd.comm.size > 1:
raise ValueError('CG eigensolver does not support band '
'parallelization. This calculation parallelizes '
'over %d band groups.' % wfs.bd.comm.size)
Eigensolver.initialize(self, wfs)
self.overlap = wfs.overlap
def __init__(self,
keep_htpsit=True,
blocksize=None,
niter=3,
rtol=1e-16,
limit_lambda=False,
use_rayleigh=False,
trial_step=0.1):
"""Initialize RMM-DIIS eigensolver.
Parameters:
limit_lambda: dictionary
determines if step length should be limited
supported keys:
'absolute':True/False limit the absolute value
'upper':float upper limit for lambda
'lower':float lower limit for lambda
"""
Eigensolver.__init__(self, keep_htpsit, blocksize)
self.niter = niter
self.rtol = rtol
self.limit_lambda = limit_lambda
self.use_rayleigh = use_rayleigh
if use_rayleigh:
1 / 0
self.blocksize = 1
self.trial_step = trial_step
self.first = True
def initialize(self, wfs):
if wfs.bd.comm.size > 1:
raise ValueError('CG eigensolver does not support band '
'parallelization. This calculation parallelizes '
'over %d band groups.' % wfs.bd.comm.size)
Eigensolver.initialize(self, wfs)
self.overlap = wfs.overlap
def estimate_memory(self, mem, wfs):
Eigensolver.estimate_memory(self, mem, wfs)
nbands = wfs.bd.nbands
mem.subnode("H_nn", nbands * nbands * mem.itemsize[wfs.dtype])
mem.subnode("S_nn", nbands * nbands * mem.itemsize[wfs.dtype])
mem.subnode("H_2n2n", 4 * nbands * nbands * mem.itemsize[wfs.dtype])
mem.subnode("S_2n2n", 4 * nbands * nbands * mem.itemsize[wfs.dtype])
mem.subnode("eps_2n", 2 * nbands * mem.floatsize)
def estimate_memory(self, mem, wfs):
Eigensolver.estimate_memory(self, mem, wfs)
nbands = wfs.bd.nbands
mem.subnode('H_nn', nbands * nbands * mem.itemsize[wfs.dtype])
mem.subnode('S_nn', nbands * nbands * mem.itemsize[wfs.dtype])
mem.subnode('H_2n2n', 4 * nbands * nbands * mem.itemsize[wfs.dtype])
mem.subnode('S_2n2n', 4 * nbands * nbands * mem.itemsize[wfs.dtype])
mem.subnode('eps_2n', 2 * nbands * mem.floatsize)
def __init__(self, niter=1, smin=None, normalize=True):
Eigensolver.__init__(self)
self.niter = niter
if not smin is None:
raise NotImplementedError("See https://trac.fysik.dtu.dk/projects/gpaw/ticket/248")
self.smin = smin
self.normalize = normalize
self.orthonormalization_required = False
def estimate_memory(self, mem, gd, dtype, mynbands, nbands):
Eigensolver.estimate_memory(self, mem, gd, dtype, mynbands, nbands)
itemsize = mem.itemsize[dtype]
mem.subnode('H_nn', nbands * nbands * mem.itemsize[dtype])
mem.subnode('S_nn', nbands * nbands * mem.itemsize[dtype])
mem.subnode('H_2n2n', 4 * nbands * nbands * mem.itemsize[dtype])
mem.subnode('S_2n2n', 4 * nbands * nbands * mem.itemsize[dtype])
mem.subnode('eps_2n', 2 * nbands * mem.floatsize)
def initialize(self, wfs):
Eigensolver.initialize(self, wfs)
if wfs.gd.comm.rank == 0 and wfs.bd.comm.rank == 0:
# Allocate arrays
B = self.nbands
self.H_NN = np.empty((2 * B, 2 * B), self.dtype)
self.S_NN = np.empty((2 * B, 2 * B), self.dtype)
self.eps_N = np.empty(2 * B)
def initialize(self, wfs):
Eigensolver.initialize(self, wfs)
self.overlap = wfs.overlap
# Allocate arrays
self.H_2n2n = np.empty((2 * self.nbands, 2 * self.nbands), self.dtype)
self.S_2n2n = np.empty((2 * self.nbands, 2 * self.nbands), self.dtype)
self.eps_2n = np.empty(2 * self.nbands)
def initialize(self, wfs):
if self.blocksize is None:
if wfs.mode == 'pw':
S = wfs.pd.comm.size
# Use a multiple of S for maximum efficiency
self.blocksize = int(np.ceil(10 / S)) * S
else:
self.blocksize = 10
Eigensolver.initialize(self, wfs)
def initialize(self, wfs):
Eigensolver.initialize(self, wfs)
self.overlap = wfs.overlap
# Allocate arrays
self.H_nn = np.zeros((self.nbands, self.nbands), self.dtype)
self.S_nn = np.zeros((self.nbands, self.nbands), self.dtype)
self.H_2n2n = np.empty((2 * self.nbands, 2 * self.nbands),
self.dtype)
self.S_2n2n = np.empty((2 * self.nbands, 2 * self.nbands),
self.dtype)
self.eps_2n = np.empty(2 * self.nbands)
def __init__(self, niter=1, smin=None, normalize=True):
Eigensolver.__init__(self)
self.niter = niter
self.smin = smin
self.normalize = normalize
if smin is not None:
raise NotImplementedError(
'See https://trac.fysik.dtu.dk/projects/gpaw/ticket/248')
self.orthonormalization_required = False
def initialize(self, wfs):
if wfs.bd.comm.size > 1:
raise ValueError('Davidson eigensolver does not support band '
'parallelization. This calculation parallelizes '
'over %d band groups.' % wfs.bd.comm.size)
Eigensolver.initialize(self, wfs)
self.overlap = wfs.overlap
# Allocate arrays
self.H_nn = np.zeros((self.nbands, self.nbands), self.dtype)
self.S_nn = np.zeros((self.nbands, self.nbands), self.dtype)
self.H_2n2n = np.empty((2 * self.nbands, 2 * self.nbands), self.dtype)
self.S_2n2n = np.empty((2 * self.nbands, 2 * self.nbands), self.dtype)
self.eps_2n = np.empty(2 * self.nbands)
def initialize(self, wfs):
if wfs.bd.comm.size > 1:
raise ValueError(
"Davidson eigensolver does not support band "
"parallelization. This calculation parallelizes "
"over %d band groups." % wfs.bd.comm.size
)
Eigensolver.initialize(self, wfs)
self.overlap = wfs.overlap
# Allocate arrays
self.H_nn = np.zeros((self.nbands, self.nbands), self.dtype)
self.S_nn = np.zeros((self.nbands, self.nbands), self.dtype)
self.H_2n2n = np.empty((2 * self.nbands, 2 * self.nbands), self.dtype)
self.S_2n2n = np.empty((2 * self.nbands, 2 * self.nbands), self.dtype)
self.eps_2n = np.empty(2 * self.nbands)
def __init__(self, niter=4, rtol=0.30):
"""Construct conjugate gradient eigen solver.
parameters:
niter: int
Maximum number of conjugate gradient iterations per band
rtol: float
If change in residual is less than rtol, iteration for band is
not continued
"""
Eigensolver.__init__(self)
self.niter = niter
self.rtol = rtol
if extra_parameters.get('PK', False):
self.orthonormalization_required = True
else:
self.orthonormalization_required = False
def __init__(self, niter=4, rtol=0.30):
"""Construct conjugate gradient eigen solver.
parameters:
niter: int
Maximum number of conjugate gradient iterations per band
rtol: float
If change in residual is less than rtol, iteration for band is
not continued
"""
Eigensolver.__init__(self)
self.niter = niter
self.rtol = rtol
if extra_parameters.get('PK', False):
self.orthonormalization_required = True
else:
self.orthonormalization_required = False
def __init__(self, keep_htpsit=True, blocksize=10, niter=3, rtol=1e-16,
limit_lambda=False, use_rayleigh=False, trial_step=0.1):
"""Initialize RMM-DIIS eigensolver.
Parameters:
limit_lambda: dictionary
determines if step length should be limited
supported keys:
'absolute':True/False limit the absolute value
'upper':float upper limit for lambda
'lower':float lower limit for lambda
"""
Eigensolver.__init__(self, keep_htpsit, blocksize)
self.niter = niter
self.rtol = rtol
self.limit_lambda = limit_lambda
self.use_rayleigh = use_rayleigh
if use_rayleigh:
self.blocksize = 1
self.trial_step = trial_step
self.first = True
def __init__(self, niter=4):
Eigensolver.__init__(self)
self.niter = niter
def __init__(self, keep_htpsit=True, blocksize=10,
fixed_trial_step=None):
self.fixed_trial_step = fixed_trial_step
Eigensolver.__init__(self, keep_htpsit, blocksize)
def estimate_memory(self, mem, gd, dtype, mynbands, nbands):
Eigensolver.estimate_memory(self, mem, gd, dtype, mynbands, nbands)
gridmem = gd.bytecount(dtype)
mem.subnode('phi_G', gridmem)
mem.subnode('phi_old_G', gridmem)
def __init__(self, keep_htpsit=True, blocksize=10, fixed_trial_step=None):
self.fixed_trial_step = fixed_trial_step
Eigensolver.__init__(self, keep_htpsit, blocksize)
def __init__(self, keep_htpsit=True, blocksize=1):
Eigensolver.__init__(self, keep_htpsit, blocksize)
def __init__(self, keep_htpsit=False):
"""Initialize direct eigensolver. """
Eigensolver.__init__(self, keep_htpsit)
def initialize(self, wfs):
Eigensolver.initialize(self, wfs)
self.overlap = wfs.overlap
# Allocate arrays
self.phi_G = self.gd.empty(dtype=self.dtype)
self.phi_old_G = self.gd.empty(dtype=self.dtype)
