def __init__(self, system, parts, nlist=None):
"""
**Arguments:**
system
An instance of the ``System`` class.
parts
A list of instances of sublcasses of ``ForcePart``. These are
the different types of contributions to the force field, e.g.
valence interactions, real-space electrostatics, and so on.
**Optional arguments:**
nlist
A ``NeighborList`` instance. This is required if some items in the
parts list use this nlist object.
"""
ForcePart.__init__(self, 'all', system)
self.system = system
self.parts = []
self.nlist = nlist
self.needs_nlist_update = nlist is not None
for part in parts:
self.add_part(part)
if log.do_medium:
with log.section('FFINIT'):
log('Force field with %i parts:&%s.' % (
len(self.parts), ', '.join(part.name for part in self.parts)
))
log('Neighborlist present: %s' % (self.nlist is not None))
def __init__(self, system, grids):
'''
**Arguments:**
system
An instance of the ``System`` class.
grids
A dictionary with (ffatype, grid) items. Each grid must be a
three-dimensional array with energies.
This force part is only applicable to systems that are 3D periodic.
'''
if system.cell.nvec != 3:
raise ValueError('The system must be 3d periodic for the grid term.')
for grid in grids.itervalues():
if grid.ndim != 3:
raise ValueError('The energy grids must be 3D numpy arrays.')
ForcePart.__init__(self, 'grid', system)
self.system = system
self.grids = grids
if log.do_medium:
with log.section('FPINIT'):
log('Force part: %s' % self.name)
log.hline()
def __init__(self, system, alpha, gcut=0.35):
'''
**Arguments:**
system
The system to which this interaction applies.
alpha
The alpha parameter in the Ewald summation method.
gcut
The cutoff in reciprocal space.
'''
ForcePart.__init__(self, 'ewald_reci', system)
if not system.cell.nvec == 3:
raise TypeError('The system must have a 3D periodic cell.')
if system.charges is None:
raise ValueError('The system does not have charges.')
if system.dipoles is None:
raise ValueError('The system does not have dipoles.')
self.system = system
self.alpha = alpha
self.gcut = gcut
self.update_gmax()
self.work = np.empty(system.natom*2)
if log.do_medium:
with log.section('FPINIT'):
log('Force part: %s' % self.name)
log.hline()
log(' alpha: %s' % log.invlength(self.alpha))
log(' gcut: %s' % log.invlength(self.gcut))
log.hline()
def __init__(self, system, alpha, scalings):
'''
**Arguments:**
system
The system to which this interaction applies.
alpha
The alpha parameter in the Ewald summation method.
scalings
A ``Scalings`` object. This object contains all the information
about the energy scaling of pairwise contributions that are
involved in covalent interactions. See
:class:`yaff.pes.scalings.Scalings` for more details.
'''
ForcePart.__init__(self, 'ewald_cor', system)
if not system.cell.nvec == 3:
raise TypeError('The system must have a 3D periodic cell')
if system.charges is None:
raise ValueError('The system does not have charges.')
self.system = system
self.alpha = alpha
self.scalings = scalings
if log.do_medium:
with log.section('FPINIT'):
log('Force part: %s' % self.name)
log.hline()
log(' alpha: %s' % log.invlength(self.alpha))
log(' scalings: %5.3f %5.3f %5.3f' % (scalings.scale1, scalings.scale2, scalings.scale3))
log.hline()
def generate(cls, system, parameters, **kwargs):
"""Create a force field for the given system with the given parameters.
**Arguments:**
system
An instance of the System class
parameters
Three types are accepted: (i) the filename of the parameter
file, which is a text file that adheres to YAFF parameter
format, (ii) a list of such filenames, or (iii) an instance of
the Parameters class.
See the constructor of the :class:`yaff.pes.generator.FFArgs` class
for the available optional arguments.
This method takes care of setting up the FF object, and configuring
all the necessary FF parts. This is a lot easier than creating an FF
with the default constructor. Parameters for atom types that are not
present in the system, are simply ignored.
"""
if system.ffatype_ids is None:
raise ValueError('The generators needs ffatype_ids in the system object.')
with log.section('GEN'), timer.section('Generator'):
from yaff.pes.generator import apply_generators, FFArgs
from yaff.pes.parameters import Parameters
if log.do_medium:
log('Generating force field from %s' % str(parameters))
if not isinstance(parameters, Parameters):
parameters = Parameters.from_file(parameters)
ff_args = FFArgs(**kwargs)
apply_generators(system, parameters, ff_args)
return ForceField(system, ff_args.parts, ff_args.nlist)
def __init__(self, system, alpha, dielectric=1.0):
'''
**Arguments:**
system
The system to which this interaction applies.
alpha
The alpha parameter in the Ewald summation method.
**Optional arguments:**
dielectric
The scalar relative permittivity of the system.
'''
ForcePart.__init__(self, 'ewald_neut', system)
if not system.cell.nvec == 3:
raise TypeError('The system must have a 3D periodic cell')
if system.charges is None:
raise ValueError('The system does not have charges.')
self.system = system
self.alpha = alpha
self.dielectric = dielectric
if log.do_medium:
with log.section('FPINIT'):
log('Force part: %s' % self.name)
log.hline()
log(' alpha: %s' % log.invlength(self.alpha))
log(' relative permittivity: %5.3f' % self.dielectric)
log.hline()
def set_standard_masses(self):
"""Initialize the ``masses`` attribute based on the atomic numbers."""
with log.section('SYS'):
from molmod.periodic import periodic
if self.masses is not None:
if log.do_warning:
log.warn('Overwriting existing masses with default masses.')
self.masses = np.array([periodic[n].mass for n in self.numbers])
def g09log_to_hdf5(f, fn_log):
"""Convert Gaussian09 BOMD log file to Yaff HDF5 format.
**Arguments:**
f
An open and writable HDF5 file.
fn_log
The name of the Gaussian log file.
"""
with log.section('G09H5'):
if log.do_medium:
log('Loading Gaussian 09 file \'%s\' into \'trajectory\' of HDF5 file \'%s\'' % (
fn_log, f.filename
))
# First make sure the HDF5 file has a system description that is consistent
# with the XYZ file.
if 'system' not in f:
raise ValueError('The HDF5 file must contain a system group.')
if 'numbers' not in f['system']:
raise ValueError('The HDF5 file must have a system group with atomic numbers.')
natom = f['system/numbers'].shape[0]
# Take care of the trajectory group
tgrp = get_trajectory_group(f)
# Take care of the pos and vel datasets
dss = get_trajectory_datasets(tgrp,
('pos', (natom, 3)),
('vel', (natom, 3)),
('frc', (natom, 3)),
('time', (1,)),
('step', (1,)),
('epot', (1,)),
('ekin', (1,)),
('etot', (1,)),
)
ds_pos, ds_vel, ds_frc, ds_time, ds_step, ds_epot, ds_ekin, ds_etot = dss
# Load frame by frame
row = get_last_trajectory_row(dss)
for numbers, pos, vel, frc, time, step, epot, ekin, etot in _iter_frames_g09(fn_log):
if (numbers != f['system/numbers']).any():
log.warn('The element numbers of the HDF5 and LOG file do not match.')
write_to_dataset(ds_pos, pos, row)
write_to_dataset(ds_vel, vel, row)
write_to_dataset(ds_frc, frc, row)
write_to_dataset(ds_time, time, row)
write_to_dataset(ds_step, step, row)
write_to_dataset(ds_epot, epot, row)
write_to_dataset(ds_ekin, ekin, row)
write_to_dataset(ds_etot, etot, row)
row += 1
# Check number of rows
check_trajectory_rows(tgrp, dss, row)
def to_file(self, fn):
"""Write the system to a file
**Arguments:**
fn
The file to write to.
Supported formats are:
chk
Internal human-readable checkpoint format. This format includes
all the information of a system object. All data are stored in
atomic units.
h5
Internal binary checkpoint format. This format includes
all the information of a system object. All data are stored in
atomic units.
xyz
A simple file with atomic positions and elements. Coordinates
are written in Angstroms.
"""
if fn.endswith('.chk'):
from molmod.io import dump_chk
dump_chk(fn, {
'numbers': self.numbers,
'pos': self.pos,
'ffatypes': self.ffatypes,
'ffatype_ids': self.ffatype_ids,
'scopes': self.scopes,
'scope_ids': self.scope_ids,
'bonds': self.bonds,
'rvecs': self.cell.rvecs,
'charges': self.charges,
'radii': self.radii,
'valence_charges': self.valence_charges,
'dipoles': self.dipoles,
'radii2': self.radii2,
'masses': self.masses,
})
elif fn.endswith('.h5'):
with h5.File(fn, 'w') as f:
self.to_hdf5(f)
elif fn.endswith('.xyz'):
from molmod.io import XYZWriter
from molmod.periodic import periodic
xyz_writer = XYZWriter(fn, [periodic[n].symbol for n in self.numbers])
xyz_writer.dump(str(self), self.pos)
else:
raise NotImplementedError('The extension of %s does not correspond to any known format.' % fn)
if log.do_high:
with log.section('SYS'):
log('Wrote system to %s.' % fn)
def run(self, nstep=None):
with log.section(self.log_name), timer.section(self.log_name):
if nstep is None:
while True:
if self.propagate():
break
else:
for i in xrange(nstep):
if self.propagate():
break
self.finalize()
def __init__(self, comsystem, scaling=None):
ForcePart.__init__(self, 'valence_com', comsystem)
#ForcePartValence.__init__(self, system)
self.comlist = comsystem.comlist
self.gpos = np.zeros((comsystem.gpos_dim, 3), float)
self.dlist = DeltaList(self.comlist)
self.iclist = InternalCoordinateList(self.dlist)
self.vlist = ValenceList(self.iclist)
self.scaling = scaling
if log.do_medium:
with log.section('FPINIT'):
log('Force part: %s' % self.name)
log.hline()
self.term = None # volume term
def update(self):
'''Rebuild or recompute the neighbor lists
Based on the changes of the atomic positions or due to calls to
``update_rcut`` and ``update_rmax``, the neighbor lists will be
rebuilt from scratch.
The heavy computational work is done in low-level C routines. The
neighbor lists array is reallocated if needed. The memory allocation
is done in Python for convenience.
'''
with log.section('NLIST'), timer.section('Nlists'):
assert self.rcut > 0
if self._need_rebuild():
# *rebuild* the entire neighborlist
if self.system.cell.volume != 0:
if self.system.natom / self.system.cell.volume > 10:
raise ValueError('Atom density too high')
# 1) make an initial status object for the neighbor list algorithm
status = nlist_status_init(self.rmax)
# 2) a loop of consecutive update/allocate calls
last_start = 0
while True:
done = nlist_build(self.system.pos, self.rcut + self.skin,
self.rmax, self.system.cell, status,
self.neighs[last_start:])
if done:
break
last_start = len(self.neighs)
new_neighs = np.empty((len(self.neighs) * 3) / 2,
dtype=neigh_dtype)
new_neighs[:last_start] = self.neighs
self.neighs = new_neighs
del new_neighs
# 3) get the number of neighbors in the list.
self.nneigh = nlist_status_finish(status)
if log.do_debug:
log('Rebuilt, size = %i' % self.nneigh)
# 4) store the current state to check in future calls if we
# need to do a rebuild or a recompute.
self._checkpoint()
self.rebuild_next = False
else:
# just *recompute* the deltas and the distance in the
# neighborlist
nlist_recompute(self.system.pos, self._pos_old,
self.system.cell, self.neighs[:self.nneigh])
if log.do_debug:
log('Recomputed')
def _verify_hooks(self):
with log.section('ENSEM'):
thermo = None
index_thermo = 0
baro = None
index_baro = 0
# Look for the presence of a thermostat and/or barostat
if hasattr(self.hooks, '__len__'):
for index, hook in enumerate(self.hooks):
if hook.method == 'thermostat':
thermo = hook
index_thermo = index
elif hook.method == 'barostat':
baro = hook
index_baro = index
elif self.hooks is not None:
if self.hooks.method == 'thermostat':
thermo = self.hooks
elif self.hooks.method == 'barostat':
baro = self.hooks
# If both are present, delete them and generate TBCombination element
if thermo is not None and baro is not None:
from yaff.sampling.npt import TBCombination
if log.do_warning:
log.warn(
'Both thermostat and barostat are present separately and will be merged'
)
del self.hooks[max(index_thermo, index_thermo)]
del self.hooks[min(index_thermo, index_baro)]
self.hooks.append(TBCombination(thermo, baro))
if hasattr(self.hooks, '__len__'):
for hook in self.hooks:
if hook.name == 'TBCombination':
thermo = hook.thermostat
baro = hook.barostat
elif self.hooks is not None:
if self.hooks.name == 'TBCombination':
thermo = self.hooks.thermostat
baro = self.hooks.barostat
if log.do_warning:
if thermo is not None:
log('Temperature coupling achieved through ' +
str(thermo.name) + ' thermostat')
if baro is not None:
log('Pressure coupling achieved through ' +
str(baro.name) + ' barostat')
def __init__(self, system, alpha, gcut=0.35, dielectric=1.0, exclude_frame=False, n_frame=0):
'''
**Arguments:**
system
The system to which this interaction applies.
alpha
The alpha parameter in the Ewald summation method.
**Optional arguments:**
gcut
The cutoff in reciprocal space.
dielectric
The scalar relative permittivity of the system.
exclude_frame
A boolean to exclude framework-framework interactions
(exclude_frame=True) for efficiency sake in MC simulations.
n_frame
Number of framework atoms. This parameter is used to exclude
framework-framework neighbors when exclude_frame=True.
'''
ForcePart.__init__(self, 'ewald_reci', system)
if not system.cell.nvec == 3:
raise TypeError('The system must have a 3D periodic cell.')
if system.charges is None:
raise ValueError('The system does not have charges.')
self.system = system
self.alpha = alpha
self.gcut = gcut
self.dielectric = dielectric
self.update_gmax()
self.work = np.empty(system.natom*2)
if exclude_frame == True and n_frame < 0:
raise ValueError('The number of framework atoms to exclude must be positive.')
elif exclude_frame == False:
n_frame = 0
self.n_frame = n_frame
if log.do_medium:
with log.section('FPINIT'):
log('Force part: %s' % self.name)
log.hline()
log(' alpha: %s' % log.invlength(self.alpha))
log(' gcut: %s' % log.invlength(self.gcut))
log(' relative permittivity: %5.3f' % self.dielectric)
log.hline()
def __init__(self, ff, state=None, hooks=None, counter0=0):
"""
**Arguments:**
ff
The ForceField instance used in the iterative algorithm
**Optional arguments:**
state
A list with state items. State items are simple objects
that take or derive a property from the current state of the
iterative algorithm.
hooks
A function (or a list of functions) that is called after every
iterative.
counter0
The counter value associated with the initial state.
"""
self.ff = ff
if state is None:
self.state_list = [
state_item.copy() for state_item in self.default_state
]
else:
#self.state_list = state
self.state_list = [
state_item.copy() for state_item in self.default_state
]
self.state_list += state
self.state = dict((item.key, item) for item in self.state_list)
if hooks is None:
self.hooks = []
elif hasattr(hooks, '__len__'):
self.hooks = hooks
else:
self.hooks = [hooks]
self._add_default_hooks()
self.counter0 = counter0
self.counter = counter0
with log.section(self.log_name), timer.section(self.log_name):
self.initialize()
# Initialize restart hook if present
from yaff.sampling.io import RestartWriter
for hook in self.hooks:
if isinstance(hook, RestartWriter):
hook.init_state(self)
def update(self):
'''Rebuild or recompute the neighbor lists
Based on the changes of the atomic positions or due to calls to
``update_rcut`` and ``update_rmax``, the neighbor lists will be
rebuilt from scratch.
The heavy computational work is done in low-level C routines. The
neighbor lists array is reallocated if needed. The memory allocation
is done in Python for convenience.
'''
with log.section('NLIST'), timer.section('Nlists'):
assert self.rcut > 0
if self._need_rebuild():
# *rebuild* the entire neighborlist
if self.system.cell.volume != 0:
if self.system.natom/self.system.cell.volume > 10:
raise ValueError('Atom density too high')
# 1) make an initial status object for the neighbor list algorithm
status = nlist_status_init(self.rmax)
# 2) a loop of consecutive update/allocate calls
last_start = 0
while True:
done = nlist_build(
self.system.pos, self.rcut + self.skin, self.rmax,
self.system.cell, status, self.neighs[last_start:]
)
if done:
break
last_start = len(self.neighs)
new_neighs = np.empty((len(self.neighs)*3)/2, dtype=neigh_dtype)
new_neighs[:last_start] = self.neighs
self.neighs = new_neighs
del new_neighs
# 3) get the number of neighbors in the list.
self.nneigh = nlist_status_finish(status)
if log.do_debug:
log('Rebuilt, size = %i' % self.nneigh)
# 4) store the current state to check in future calls if we
# need to do a rebuild or a recompute.
self._checkpoint()
self.rebuild_next = False
else:
# just *recompute* the deltas and the distance in the
# neighborlist
nlist_recompute(self.system.pos, self._pos_old, self.system.cell, self.neighs[:self.nneigh])
if log.do_debug:
log('Recomputed')
def __init__(self, system):
'''
**Arguments:**
system
An instance of the ``System`` class.
'''
ForcePart.__init__(self, 'valence', system)
self.dlist = DeltaList(system)
self.iclist = InternalCoordinateList(self.dlist)
self.vlist = ValenceList(self.iclist)
if log.do_medium:
with log.section('FPINIT'):
log('Force part: %s' % self.name)
log.hline()
def __init__(self, system):
'''
**Arguments:**
system
An instance of the ``System`` class.
'''
ForcePart.__init__(self, 'valence', system)
self.dlist = DeltaList(system)
self.iclist = InternalCoordinateList(self.dlist)
self.vlist = ValenceList(self.iclist)
if log.do_medium:
with log.section('FPINIT'):
log('Force part: %s' % self.name)
log.hline()
def generate(cls, system, parameters, **kwargs):
"""Create a force field for the given system with the given parameters.
**Arguments:**
system
An instance of the System class
parameters
Four types are accepted: (i) the filename of the parameter
file, which is a text file that adheres to YAFF parameter
format, (ii) a list of such filenames, (iii) an instance of
the Parameters class, or (iv) the filename of the parameter file
in the YAML format.
See the constructor of the :class:`yaff.pes.generator.FFArgs` class
for the available optional arguments.
This method takes care of setting up the FF object, and configuring
all the necessary FF parts. This is a lot easier than creating an FF
with the default constructor. Parameters for atom types that are not
present in the system, are simply ignored.
"""
if system.ffatype_ids is None:
raise ValueError(
'The generators needs ffatype_ids in the system object.')
with log.section('GEN'), timer.section('Generator'):
from yaff.pes.generator import apply_generators, FFArgs
from yaff.pes.parameters import Parameters
if log.do_medium:
log('Generating force field from %s' % str(parameters))
if isinstance(parameters, str):
if parameters[-3:] == 'txt':
parameters = Parameters.from_file(parameters)
ff_args = FFArgs(**kwargs)
apply_generators(system, parameters, ff_args)
else:
yaml_dict = yaml.safe_load(open(parameters))
ff_args = FFArgs(**kwargs)
apply_generators(system, yaml_dict, ff_args)
else:
if isinstance(parameters, Parameters):
raise NotImplementedError
else:
yaml_dict = parameters
ff_args = FFArgs(**kwargs)
apply_generators(system, yaml_dict, ff_args)
return ForceField(system, ff_args.parts, ff_args.nlist)
def add_term(self, term):
'''Add a new term to the covalent force field.
**Arguments:**
term
An instance of the class :class:`yaff.pes.ff.vlist.ValenceTerm`.
In principle, one should add all energy terms before calling the
``compute`` method, but with the current implementation of Yaff,
energy terms can be added at any time. (This may change in future.)
'''
if log.do_high:
with log.section('VTERM'):
log('%7i&%s %s' % (self.vlist.nv, term.get_log(), ' '.join(ic.get_log() for ic in term.ics)))
self.vlist.add_term(term)
def set_hills(self, q0s, Ks, sigmas, tempering=0.0, T=None, periodicities=None):
# Safety checks
assert q0s.shape[1]==self.ncv
assert sigmas.shape[0]==self.ncv
assert q0s.shape[0]==Ks.shape[0]
if tempering != 0.0 and T is None:
raise ValueError("For a well-tempered MTD run, the temperature "
"has to be specified")
self.q0s = q0s
self.sigmas = sigmas
self.Ks = Ks
self.tempering = tempering
self.T = T
self.periodicities = periodicities
if log.do_medium:
with log.section("SUMHILL"):
log("Found %d collective variables and %d Gaussian hills"%(self.ncv,self.q0s.shape[0]))
def __init__(self, ff, state=None, hooks=None, counter0=0):
"""
**Arguments:**
ff
The ForceField instance used in the iterative algorithm
**Optional arguments:**
state
A list with state items. State items are simple objects
that take or derive a property from the current state of the
iterative algorithm.
hooks
A function (or a list of functions) that is called after every
iterative.
counter0
The counter value associated with the initial state.
"""
self.ff = ff
if state is None:
self.state_list = [state_item.copy() for state_item in self.default_state]
else:
#self.state_list = state
self.state_list = [state_item.copy() for state_item in self.default_state]
self.state_list += state
self.state = dict((item.key, item) for item in self.state_list)
if hooks is None:
self.hooks = []
elif hasattr(hooks, '__len__'):
self.hooks = hooks
else:
self.hooks = [hooks]
self._add_default_hooks()
self.counter0 = counter0
self.counter = counter0
with log.section(self.log_name), timer.section(self.log_name):
self.initialize()
# Initialize restart hook if present
from yaff.sampling.io import RestartWriter
for hook in self.hooks:
if isinstance(hook, RestartWriter):
hook.init_state(self)
def _verify_hooks(self):
with log.section('ENSEM'):
thermo = None
index_thermo = 0
baro = None
index_baro = 0
# Look for the presence of a thermostat and/or barostat
if hasattr(self.hooks, '__len__'):
for index, hook in enumerate(self.hooks):
if hook.method == 'thermostat':
thermo = hook
index_thermo = index
elif hook.method == 'barostat':
baro = hook
index_baro = index
elif self.hooks is not None:
if self.hooks.method == 'thermostat':
thermo = self.hooks
elif self.hooks.method == 'barostat':
baro = self.hooks
# If both are present, delete them and generate TBCombination element
if thermo is not None and baro is not None:
from yaff.sampling.npt import TBCombination
if log.do_warning:
log.warn('Both thermostat and barostat are present separately and will be merged')
del self.hooks[max(index_thermo, index_thermo)]
del self.hooks[min(index_thermo, index_baro)]
self.hooks.append(TBCombination(thermo, baro))
if hasattr(self.hooks, '__len__'):
for hook in self.hooks:
if hook.name == 'TBCombination':
thermo = hook.thermostat
baro = hook.barostat
elif self.hooks is not None:
if self.hooks.name == 'TBCombination':
thermo = self.hooks.thermostat
baro = self.hooks.barostat
if log.do_warning:
if thermo is not None:
log('Temperature coupling achieved through ' + str(thermo.name) + ' thermostat')
if baro is not None:
log('Pressure coupling achieved through ' + str(baro.name) + ' barostat')
def add_term(self, term):
'''Add a new term to the covalent force field.
**Arguments:**
term
An instance of the class :class:`yaff.pes.ff.vlist.ValenceTerm`.
In principle, one should add all energy terms before calling the
``compute`` method, but with the current implementation of Yaff,
energy terms can be added at any time. (This may change in future.)
'''
if log.do_high:
with log.section('VTERM'):
log('%7i&%s %s' % (self.vlist.nv, term.get_log(), ' '.join(
ic.get_log() for ic in term.ics)))
self.vlist.add_term(term)
def detect_ffatypes(self, rules):
"""Initialize the ``ffatypes`` attribute based on ATSELECT rules.
**Argument:**
rules
A list of (ffatype, rule) pairs that will be used to initialize
the attributes ``self.ffatypes`` and ``self.ffatype_ids``.
If the system already has FF atom types, they will be overwritten.
"""
with log.section('SYS'):
# Give warning if needed
if self.ffatypes is not None:
if log.do_warning:
log.warn('Overwriting existing FF atom types.')
# Compile all the rules
my_rules = []
for ffatype, rule in rules:
check_name(ffatype)
if isinstance(rule, str):
rule = atsel_compile(rule)
my_rules.append((ffatype, rule))
# Use the rules to detect the atom types
lookup = {}
self.ffatypes = []
self.ffatype_ids = np.zeros(self.natom, int)
for i in range(self.natom):
my_ffatype = None
for ffatype, rule in my_rules:
if rule(self, i):
my_ffatype = ffatype
break
if my_ffatype is None:
raise ValueError(
'Could not detect FF atom type of atom %i.' % i)
ffatype_id = lookup.get(my_ffatype)
if ffatype_id is None:
ffatype_id = len(lookup)
self.ffatypes.append(my_ffatype)
lookup[my_ffatype] = ffatype_id
self.ffatype_ids[i] = ffatype_id
# Make sure all is done well ...
self._init_derived_ffatypes()
def detect_ffatypes(self, rules):
"""Initialize the ``ffatypes`` attribute based on ATSELECT rules.
**Argument:**
rules
A list of (ffatype, rule) pairs that will be used to initialize
the attributes ``self.ffatypes`` and ``self.ffatype_ids``.
If the system already has FF atom types, they will be overwritten.
"""
with log.section('SYS'):
# Give warning if needed
if self.ffatypes is not None:
if log.do_warning:
log.warn('Overwriting existing FF atom types.')
# Compile all the rules
my_rules = []
for ffatype, rule in rules:
check_name(ffatype)
if isinstance(rule, basestring):
rule = atsel_compile(rule)
my_rules.append((ffatype, rule))
# Use the rules to detect the atom types
lookup = {}
self.ffatypes = []
self.ffatype_ids = np.zeros(self.natom, int)
for i in xrange(self.natom):
my_ffatype = None
for ffatype, rule in my_rules:
if rule(self, i):
my_ffatype = ffatype
break
if my_ffatype is None:
raise ValueError('Could not detect FF atom type of atom %i.' % i)
ffatype_id = lookup.get(my_ffatype)
if ffatype_id is None:
ffatype_id = len(lookup)
self.ffatypes.append(my_ffatype)
lookup[my_ffatype] = ffatype_id
self.ffatype_ids[i] = ffatype_id
# Make sure all is done well ...
self._init_derived_ffatypes()
def __init__(self, system, comlist=None):
'''
Parameters
----------
system
An instance of the ``System`` class.
comlist
An optional layer to derive centers of mass from the atomic positions.
These centers of mass are used as input for the first layer, the relative
vectors.
'''
ForcePart.__init__(self, 'valence', system)
self.comlist = comlist
self.dlist = DeltaList(system if comlist is None else comlist)
self.iclist = InternalCoordinateList(self.dlist)
self.vlist = ValenceList(self.iclist)
if log.do_medium:
with log.section('FPINIT'):
log('Force part: %s' % self.name)
log.hline()
def __init__(self, system, comlist=None):
'''
Parameters
----------
system
An instance of the ``System`` class.
comlist
An optional layer to derive centers of mass from the atomic positions.
These centers of mass are used as input for the first layer, the relative
vectors.
'''
ForcePart.__init__(self, 'valence', system)
self.comlist = comlist
self.dlist = DeltaList(system if comlist is None else comlist)
self.iclist = InternalCoordinateList(self.dlist)
self.vlist = ValenceList(self.iclist)
if log.do_medium:
with log.section('FPINIT'):
log('Force part: %s' % self.name)
log.hline()
def __init__(self, system):
'''
Parameters
----------
system
An instance of the ``System`` class.
'''
ForcePart.__init__(self, 'valence', system)
# override self.gpos to the correct size!
# natom of COMSystem object will return number of beads
# but gpos has to have the size (n_atoms, 3), to be consisten
# with the other parts of the force field
self.dlist = DeltaList(system)
self.iclist = InternalCoordinateList(self.dlist)
self.vlist = ValenceList(self.iclist)
if log.do_medium:
with log.section('FPINIT'):
log('Force part: %s' % self.name)
log.hline()
def __init__(self, system, nlist, scalings, pair_pot):
'''
**Arguments:**
system
The system to which this pairwise interaction applies.
nlist
A ``NeighborList`` object. This has to be the same as the one
passed to the ForceField object that contains this part.
scalings
A ``Scalings`` object. This object contains all the information
about the energy scaling of pairwise contributions that are
involved in covalent interactions. See
:class:`yaff.pes.scalings.Scalings` for more details.
pair_pot
An instance of the ``PairPot`` built-in class from
:mod:`yaff.pes.ext`.
'''
ForcePart.__init__(self, 'pair_%s' % pair_pot.name, system)
self.nlist = nlist
self.scalings = scalings
self.pair_pot = pair_pot
self.nlist.request_rcut(pair_pot.rcut)
if log.do_medium:
with log.section('FPINIT'):
log('Force part: %s' % self.name)
log.hline()
log(' scalings: %5.3f %5.3f %5.3f' %
(scalings.scale1, scalings.scale2, scalings.scale3))
log(' real space cutoff: %s' % log.length(pair_pot.rcut))
tr = pair_pot.get_truncation()
if tr is None:
log(' truncation: none')
else:
log(' truncation: %s' % tr.get_log())
self.pair_pot.log()
log.hline()
def iter_matches(self, other):
"""Yield all renumberings of atoms that map the given system on the current.
Parameters
----------
other : yaff.System
Another system with the same number of atoms (and chemical formula), or less
atoms.
The graph distance is used to perform the mapping, so bonds must be defined in
the current and the given system.
"""
from molmod.graphs import Graph
with log.section('SYS'):
log('Generating allowed indexes for renumbering.')
# The allowed permutations is just based on the chemical elements, not the atom
# types, which could also be useful.
allowed = []
if self.ffatypes is None or other.ffatypes is None:
for number1 in other.numbers:
allowed.append((self.numbers == number1).nonzero()[0])
else:
# Only continue if other.ffatypes is a subset of self.ffatypes
if not (set(self.ffatypes) >= set(other.ffatypes)):
return
ffatype_ids0 = self.ffatype_ids
ffatypes0 = list(self.ffatypes)
order = np.array([ffatypes0.index(ffatype) for ffatype in other.ffatypes])
ffatype_ids1 = order[other.ffatype_ids]
for ffatype_id1 in ffatype_ids1:
allowed.append((ffatype_ids0 == ffatype_id1).nonzero()[0])
# Use Molmod to construct graph distance matrices.
log('Building graph distance matrix for self.')
dm0 = Graph(self.bonds).distances
log('Building graph distance matrix for other.')
dm1 = Graph(other.bonds).distances
# Yield the solutions
log('Generating renumberings.')
for match in iter_matches(dm0, dm1, allowed):
yield match
def __init__(self, system, alpha, scalings, dielectric=1.0):
'''
**Arguments:**
system
The system to which this interaction applies.
alpha
The alpha parameter in the Ewald summation method.
scalings
A ``Scalings`` object. This object contains all the information
about the energy scaling of pairwise contributions that are
involved in covalent interactions. See
:class:`yaff.pes.scalings.Scalings` for more details.
**Optional arguments:**
dielectric
The scalar relative permittivity of the system.
'''
ForcePart.__init__(self, 'ewald_cor', system)
if not system.cell.nvec == 3:
raise TypeError('The system must have a 3D periodic cell')
if system.charges is None:
raise ValueError('The system does not have charges.')
self.system = system
self.alpha = alpha
self.dielectric = dielectric
self.scalings = scalings
if log.do_medium:
with log.section('FPINIT'):
log('Force part: %s' % self.name)
log.hline()
log(' alpha: %s' % log.invlength(self.alpha))
log(' relative permittivity %5.3f' % self.dielectric)
log(' scalings: %5.3f %5.3f %5.3f' %
(scalings.scale1, scalings.scale2, scalings.scale3))
log.hline()
def detect_bonds(self, exceptions=None):
"""Initialize the ``bonds`` attribute based on inter-atomic distances
**Optional argument:**
exceptions:
Specify custom threshold for certain pairs of elements. This
must be a dictionary with ((num0, num1), threshold) as items.
For each pair of elements, a distance threshold is used to detect
bonded atoms. The distance threshold is based on a database of known
bond lengths. If the database does not contain a record for the given
element pair, the threshold is based on the sum of covalent radii.
"""
with log.section('SYS'):
from molmod.bonds import bonds
if self.bonds is not None:
if log.do_warning:
log.warn('Overwriting existing bonds.')
work = np.zeros((self.natom*(self.natom-1))/2, float)
self.cell.compute_distances(work, self.pos)
ishort = (work < bonds.max_length*1.01).nonzero()[0]
new_bonds = []
for i in ishort:
i0, i1 = _unravel_triangular(i)
n0 = self.numbers[i0]
n1 = self.numbers[i1]
if exceptions is not None:
threshold = exceptions.get((n0, n1))
if threshold is None and n0!=n1:
threshold = exceptions.get((n1, n0))
if threshold is not None:
if work[i] < threshold:
new_bonds.append([i0, i1])
continue
if bonds.bonded(n0, n1, work[i]):
new_bonds.append([i0, i1])
self.bonds = np.array(new_bonds)
self._init_derived_bonds()
def __init__(self, system, nlist, scalings, pair_pot):
'''
**Arguments:**
system
The system to which this pairwise interaction applies.
nlist
A ``NeighborList`` object. This has to be the same as the one
passed to the ForceField object that contains this part.
scalings
A ``Scalings`` object. This object contains all the information
about the energy scaling of pairwise contributions that are
involved in covalent interactions. See
:class:`yaff.pes.scalings.Scalings` for more details.
pair_pot
An instance of the ``PairPot`` built-in class from
:mod:`yaff.pes.ext`.
'''
ForcePart.__init__(self, 'pair_%s' % pair_pot.name, system)
self.nlist = nlist
self.scalings = scalings
self.pair_pot = pair_pot
self.nlist.request_rcut(pair_pot.rcut)
if log.do_medium:
with log.section('FPINIT'):
log('Force part: %s' % self.name)
log.hline()
log(' scalings: %5.3f %5.3f %5.3f' % (scalings.scale1, scalings.scale2, scalings.scale3))
log(' real space cutoff: %s' % log.length(pair_pot.rcut))
tr = pair_pot.get_truncation()
if tr is None:
log(' truncation: none')
else:
log(' truncation: %s' % tr.get_log())
self.pair_pot.log()
log.hline()
def detect_bonds(self, exceptions=None):
"""Initialize the ``bonds`` attribute based on inter-atomic distances
**Optional argument:**
exceptions:
Specify custom threshold for certain pairs of elements. This
must be a dictionary with ((num0, num1), threshold) as items.
For each pair of elements, a distance threshold is used to detect
bonded atoms. The distance threshold is based on a database of known
bond lengths. If the database does not contain a record for the given
element pair, the threshold is based on the sum of covalent radii.
"""
with log.section('SYS'):
from molmod.bonds import bonds
if self.bonds is not None:
if log.do_warning:
log.warn('Overwriting existing bonds.')
work = np.zeros((self.natom * (self.natom - 1)) // 2, float)
self.cell.compute_distances(work, self.pos)
ishort = (work < bonds.max_length * 1.01).nonzero()[0]
new_bonds = []
for i in ishort:
i0, i1 = _unravel_triangular(i)
n0 = self.numbers[i0]
n1 = self.numbers[i1]
if exceptions is not None:
threshold = exceptions.get((n0, n1))
if threshold is None and n0 != n1:
threshold = exceptions.get((n1, n0))
if threshold is not None:
if work[i] < threshold:
new_bonds.append([i0, i1])
continue
if bonds.bonded(n0, n1, work[i]):
new_bonds.append([i0, i1])
self.bonds = np.array(new_bonds)
self._init_derived_bonds()
def estimate_hessian(dof, eps=1e-4):
"""Estimate the Hessian using the symmetric finite difference approximation.
**Arguments:**
dof
A DOF object
**Optional arguments:**
eps
The magnitude of the displacements
"""
with log.section('HESS'), timer.section('Hessian'):
# Loop over all displacements
if log.do_medium:
log('The following displacements are computed:')
log('DOF Dir Energy')
log.hline()
x1 = dof.x0.copy()
rows = np.zeros((len(x1), len(x1)), float)
for i in xrange(len(x1)):
x1[i] = dof.x0[i] + eps
epot, gradient_p = dof.fun(x1, do_gradient=True)
if log.do_medium:
log('% 7i pos %s' % (i, log.energy(epot)))
x1[i] = dof.x0[i] - eps
epot, gradient_m = dof.fun(x1, do_gradient=True)
if log.do_medium:
log('% 7i neg %s' % (i, log.energy(epot)))
rows[i] = (gradient_p - gradient_m) / (2 * eps)
x1[i] = dof.x0[i]
dof.reset()
if log.do_medium:
log.hline()
# Enforce symmetry and return
return 0.5 * (rows + rows.T)
def estimate_hessian(dof, eps=1e-4):
"""Estimate the Hessian using the symmetric finite difference approximation.
**Arguments:**
dof
A DOF object
**Optional arguments:**
eps
The magnitude of the displacements
"""
with log.section('HESS'), timer.section('Hessian'):
# Loop over all displacements
if log.do_medium:
log('The following displacements are computed:')
log('DOF Dir Energy')
log.hline()
x1 = dof.x0.copy()
rows = np.zeros((len(x1), len(x1)), float)
for i in xrange(len(x1)):
x1[i] = dof.x0[i] + eps
epot, gradient_p = dof.fun(x1, do_gradient=True)
if log.do_medium:
log('% 7i pos %s' % (i, log.energy(epot)))
x1[i] = dof.x0[i] - eps
epot, gradient_m = dof.fun(x1, do_gradient=True)
if log.do_medium:
log('% 7i neg %s' % (i, log.energy(epot)))
rows[i] = (gradient_p-gradient_m)/(2*eps)
x1[i] = dof.x0[i]
dof.reset()
if log.do_medium:
log.hline()
# Enforce symmetry and return
return 0.5*(rows + rows.T)
def __init__(self, system, pext):
'''
**Arguments:**
system
An instance of the ``System`` class.
pext
The external pressure. (Positive will shrink the system.) In
case of 2D-PBC, this is the surface tension. In case of 1D, this
is the linear strain.
This force part is only applicable to systems that are periodic.
'''
if system.cell.nvec == 0:
raise ValueError('The system must be periodic in order to apply a pressure')
ForcePart.__init__(self, 'press', system)
self.system = system
self.pext = pext
if log.do_medium:
with log.section('FPINIT'):
log('Force part: %s' % self.name)
log.hline()
def __init__(self, system, part_pair):
'''
**Arguments:**
system
An instance of the ``System`` class.
part_pair
An instance of the ``PairPot`` class.
This force part is only applicable to systems that are 3D periodic.
'''
if system.cell.nvec != 3:
raise ValueError('Tail corrections can only be applied to 3D periodic systems')
if part_pair.name in ['pair_ei','pair_eidip']:
raise ValueError('Tail corrections are divergent for %s'%part_pair.name)
super(ForcePartTailCorrection, self).__init__('tailcorr_%s'%(part_pair.name), system)
self.ecorr, self.wcorr = part_pair.pair_pot.prepare_tailcorrections(system.natom)
self.system = system
if log.do_medium:
with log.section('FPINIT'):
log('Force part: %s' % self.name)
log.hline()
def __init__(self, system, alpha, gcut=0.35, dielectric=1.0):
'''
**Arguments:**
system
The system to which this interaction applies.
alpha
The alpha parameter in the Ewald summation method.
**Optional arguments:**
gcut
The cutoff in reciprocal space.
dielectric
The scalar relative permittivity of the system.
'''
ForcePart.__init__(self, 'ewald_reci', system)
if not system.cell.nvec == 3:
raise TypeError('The system must have a 3D periodic cell.')
if system.charges is None:
raise ValueError('The system does not have charges.')
self.system = system
self.alpha = alpha
self.gcut = gcut
self.dielectric = dielectric
self.update_gmax()
self.work = np.empty(system.natom * 2)
if log.do_medium:
with log.section('FPINIT'):
log('Force part: %s' % self.name)
log.hline()
log(' alpha: %s' % log.invlength(self.alpha))
log(' gcut: %s' % log.invlength(self.gcut))
log(' relative permittivity: %5.3f' % self.dielectric)
log.hline()
def __init__(self, system, pext):
'''
**Arguments:**
system
An instance of the ``System`` class.
pext
The external pressure. (Positive will shrink the system.) In
case of 2D-PBC, this is the surface tension. In case of 1D, this
is the linear strain.
This force part is only applicable to systems that are periodic.
'''
if system.cell.nvec == 0:
raise ValueError(
'The system must be periodic in order to apply a pressure')
ForcePart.__init__(self, 'press', system)
self.system = system
self.pext = pext
if log.do_medium:
with log.section('FPINIT'):
log('Force part: %s' % self.name)
log.hline()
def __init__(self, system, comlist=None):
'''
**Arguments:**
system
An instance of the ``System`` class.
**Optional arguments:**
comlist
An optional layer to derive centers of mass from the atomic positions.
These centers of mass are used as input for the first layer, the relative
vectors.
'''
ForcePart.__init__(self, 'bias', system)
self.system = system
self.valence = ForcePartValence(system)
if comlist is not None:
raise NotImplementedError
if comlist is not None:
self.valence_com = ForcePartValence(system, comlist=comlist)
else:
self.valence_com = None
self.terms = []
# The terms contributing to the bias potential are divided into three
# categories:
# 0) instances of BiasPotential
# 1) instances of ValenceTerm with a regular DeltaList
# 2) instances of ValenceTerm with a COMList
# The following list facilitates looking up the terms after they have
# been added
self.term_lookup = []
if log.do_medium:
with log.section('FPINIT'):
log('Force part: %s' % self.name)
log.hline()
def pca_projection(f_target, f, pm, start=0, end=None, step=1, select=None, path='trajectory/pos', mw=True):
"""
Determines the principal components of an MD simulation
**Arguments:**
f_target
Path to an h5.File instance to which the results are written.
f
An h5.File instance containing the trajectory data.
pm
An array containing the principal modes in its columns
**Optional arguments:**
start
The first sample to be considered for analysis. This may be
negative to indicate that the analysis should start from the
-start last samples.
end
The last sample to be considered for analysis. This may be
negative to indicate that the last -end sample should not be
considered.
step
The spacing between the samples used for the analysis
select
A list of atom indexes that are considered for the computation
of the spectrum. If not given, all atoms are used.
path
The path of the dataset that contains the time dependent data in
the HDF5 file. The first axis of the array must be the time
axis.
mw
If mass_weighted is True, the covariance matrix is mass-weighted.
"""
# Load in the relevant data
q = f[path][start:end:step,:,:]
# Select the given atoms
if select is not None:
q = q[:,select,:]
# Reshape such that all Cartesian coordinates are treated equally
q = q.reshape(q.shape[0],-1)
# If necessary, weight with the mass
if mw:
# Select the necessary masses
masses = f['system/masses']
if select is not None:
masses = masses[select]
# Repeat d times, with d the dimension
masses = np.repeat(masses,3)
# Reweight with the masses
q *= np.sqrt(masses)
# Calculation of the principal components: projection of each q_j on the principal modes
with log.section('PCA'):
log('Determining principal components')
prin_comp = np.dot(q, pm)
# Create output HDF5 file
with h5.File(f_target, 'a') as g:
if not 'pca' in g:
pca = g.create_group('pca')
else:
pca = g['pca']
pca.create_dataset('pc', data=prin_comp)
def g09log_to_hdf5(f, fn_log):
"""Convert Gaussian09 BOMD log file to Yaff HDF5 format.
**Arguments:**
f
An open and writable HDF5 file.
fn_log
The name of the Gaussian log file.
"""
with log.section('G09H5'):
if log.do_medium:
log('Loading Gaussian 09 file \'%s\' into \'trajectory\' of HDF5 file \'%s\''
% (fn_log, f.filename))
# First make sure the HDF5 file has a system description that is consistent
# with the XYZ file.
if 'system' not in f:
raise ValueError('The HDF5 file must contain a system group.')
if 'numbers' not in f['system']:
raise ValueError(
'The HDF5 file must have a system group with atomic numbers.')
natom = f['system/numbers'].shape[0]
# Take care of the trajectory group
tgrp = get_trajectory_group(f)
# Take care of the pos and vel datasets
dss = get_trajectory_datasets(
tgrp,
('pos', (natom, 3)),
('vel', (natom, 3)),
('frc', (natom, 3)),
('time', (1, )),
('step', (1, )),
('epot', (1, )),
('ekin', (1, )),
('etot', (1, )),
)
ds_pos, ds_vel, ds_frc, ds_time, ds_step, ds_epot, ds_ekin, ds_etot = dss
# Load frame by frame
row = get_last_trajectory_row(dss)
for numbers, pos, vel, frc, time, step, epot, ekin, etot in _iter_frames_g09(
fn_log):
if (numbers != f['system/numbers']).any():
log.warn(
'The element numbers of the HDF5 and LOG file do not match.'
)
write_to_dataset(ds_pos, pos, row)
write_to_dataset(ds_vel, vel, row)
write_to_dataset(ds_frc, frc, row)
write_to_dataset(ds_time, time, row)
write_to_dataset(ds_step, step, row)
write_to_dataset(ds_epot, epot, row)
write_to_dataset(ds_ekin, ekin, row)
write_to_dataset(ds_etot, etot, row)
row += 1
# Check number of rows
check_trajectory_rows(tgrp, dss, row)
def calc_pca(f_target, cov_mat=None, f=None, q_ref=None, start=0, end=None, step=1, select=None, path='trajectory/pos', mw=True, temp=None):
"""
Performs a principle component analysis of the given trajectory.
**Arguments:**
f_target
Path to an h5.File instance to which the results are written.
**Optional arguments:**
cov_mat
The covariance matrix, if already calculated. If not provided,
the covariance matrix will be calculatd based on the file f.
f
An h5.File instance containing the trajectory data.
q_ref
Reference vector of the positions. If not provided, the ensemble
average is taken.
start
The first sample to be considered for analysis. This may be
negative to indicate that the analysis should start from the
-start last samples.
end
The last sample to be considered for analysis. This may be
negative to indicate that the last -end sample should not be
considered.
step
The spacing between the samples used for the analysis
select
A list of atom indexes that are considered for the computation
of the spectrum. If not given, all atoms are used.
path
The path of the dataset that contains the time dependent data in
the HDF5 file. The first axis of the array must be the time
axis.
mw
If mass_weighted is True, the covariance matrix is mass-weighted.
temp
Temperature at which the simulation is carried out, necessary to determine the frequencies
"""
if cov_mat is None:
if f is None:
AssertionError('No covariance matrix nor h5.File instance provided.')
else:
with log.section('PCA'):
log('Calculating covariance matrix')
cov_mat, q_ref = calc_cov_mat(f, q_ref, start, end, step, select, path, mw)
with log.section('PCA'):
log('Diagonalizing the covariance matrix')
# Eigenvalue decomposition
eigval, eigvec = np.linalg.eigh(cov_mat)
# Order the eigenvalues in decreasing order
idx = eigval.argsort()[::-1]
eigval = eigval[idx]
eigvec = eigvec[:,idx]
# Create output HDF5 file
with h5.File(f_target, 'w') as g:
pca = g.create_group('pca')
# Output reference structure q_ref
pca.create_dataset('q_ref', data=q_ref)
# Output covariance matrix
pca.create_dataset('cov_matrix', data=cov_mat)
# Output eigenvectors in columns
pca.create_dataset('pm', data=eigvec)
# Output eigenvalues
pca.create_dataset('eigvals', data=eigval)
log('Determining inverse of the covariance matrix')
# Process matrix to determine inverse
# First, project out the three zero eigenvalues (translations)
eigvec_reduced = eigvec[:,:-3]
eigval_reduced = eigval[:-3]
# Second, calculate the reduced covariance matrix and its inverse
cov_mat_reduced = np.dot(np.dot(eigvec_reduced, np.diag(eigval_reduced)), eigvec_reduced.T)
cov_mat_inverse = np.dot(np.dot(eigvec_reduced, np.diag(1/eigval_reduced)), eigvec_reduced.T)
pca.create_dataset('cov_mat_red', data=cov_mat_reduced)
pca.create_dataset('cov_mat_inv', data=cov_mat_inverse)
# Third, if the temperature is specified, calculate the frequencies
# (the zero frequencies are mentioned last so that their index corresponds to the principal modes)
if temp is not None:
log('Determining frequencies')
frequencies = np.append(np.sqrt(boltzmann*temp/eigval_reduced)/(2*np.pi), np.repeat(0,3))
pca.create_dataset('freqs', data=frequencies)
return eigval, eigvec
def dlpoly_history_to_hdf5(f, fn_history, sub=slice(None), pos_unit=angstrom,
vel_unit=angstrom/picosecond, frc_unit=amu*angstrom/picosecond**2,
time_unit=picosecond, mass_unit=amu):
"""Convert DLPolay History trajectory file to Yaff HDF5 format.
**Arguments:**
f
An open and writable HDF5 file.
fn_history
The filename of the DLPOLY history file.
**Optional arguments:**
sub
The sub argument for the DLPolyHistoryReader. This must be a slice
object that defines the subsampling of the samples from the history
file. By default all frames are read.
pos_unit, vel_unit, frc_unit, time_unit and mass_unit
The units used in the dlpoly history file. The default values
correspond to the defaults used in DLPOLY.
This routine will also test the consistency of the row attribute of the
trajectory group. If some trajectory data is already present, it will be
replaced by the new data. It is highly recommended to first initialize
the HDF5 file with the ``to_hdf5`` method of the System class.
"""
with log.section('DPH5'):
if log.do_medium:
log('Loading DLPOLY history file \'%s\' into \'trajectory\' of HDF5 file \'%s\'' % (
fn_history, f.filename
))
# Take care of the data group
tgrp = get_trajectory_group(f)
# Open the history file for reading
hist_reader = DLPolyHistoryReader(fn_history, sub, pos_unit, vel_unit,
frc_unit, time_unit, mass_unit)
# Take care of the datasets that should always be present
natom = hist_reader.num_atoms
dss = get_trajectory_datasets(
tgrp,
('step', (1,)),
('time', (1,)),
('cell', (3,3)),
('pos', (natom, 3)),
)
ds_step, ds_time, ds_cell, ds_pos = dss
# Take care of optional data sets
if hist_reader.keytrj > 0:
ds_vel = get_trajectory_datasets(tgrp, ('vel', (natom, 3)))[0]
dss.append(ds_vel)
if hist_reader.keytrj > 1:
ds_frc = get_trajectory_datasets(tgrp, ('frc', (natom, 3)))[0]
dss.append(ds_frc)
# Decide on the first row to start writing data
row = get_last_trajectory_row(dss)
# Load data
for frame in hist_reader:
write_to_dataset(ds_step, frame["step"], row)
write_to_dataset(ds_time, frame["time"], row)
write_to_dataset(ds_cell, frame["cell"].T, row)
write_to_dataset(ds_pos, frame["pos"], row)
if hist_reader.keytrj > 0:
write_to_dataset(ds_vel, frame["vel"], row)
if hist_reader.keytrj > 1:
write_to_dataset(ds_frc, frame["frc"], row)
row += 1
# Check number of rows
check_trajectory_rows(tgrp, dss, row)
def cp2k_ener_to_hdf5(f, fn_ener, sub=slice(None)):
"""Convert a CP2K energy trajectory file to Yaff HDF5 format.
**Arguments:**
f
An open and writable HDF5 file.
fn_ener
The filename of the CP2K energy trajectory file.
**Optional arguments:**
sub
This must be a slice object that defines the sub-sampling of the
CP2K energy file. By default all time steps are read.
This routine will also test the consistency of the row attribute of the
trajectory group. If some trajectory data is already present, it will be
replaced by the new data. Furthermore, this routine also checks the
header of the CP2K energy file to make sure the values are interpreted
correctly.
It is highly recommended to first initialize the HDF5 file with the
``to_hdf5`` method of the System class.
"""
with log.section('CP2KEH5'):
if log.do_medium:
log('Loading CP2K energy file \'%s\' into \'trajectory\' of HDF5 file \'%s\''
% (fn_ener, f.filename))
# Take care of the data group
tgrp = get_trajectory_group(f)
# Take care of the datasets
dss = get_trajectory_datasets(
tgrp,
('step', (1, )),
('time', (1, )),
('ekin', (1, )),
('temp', (1, )),
('epot', (1, )),
('econs', (1, )),
)
ds_step, ds_time, ds_ke, ds_temp, ds_pe, ds_cq = dss
# Fill the datasets with data.
row = get_last_trajectory_row(dss)
counter = 0
fin = file(fn_ener)
# check header line
line = fin.next()
words = line.split()
if words[0] != '#':
raise ValueError(
'The first line in the energies file should be a header line starting with #.'
)
if words[3] != 'Time[fs]' or words[4] != 'Kin.[a.u.]' or \
words[5] != 'Temp[K]' or words[6] != 'Pot.[a.u.]' or \
words[7] + ' ' + words[8] != 'Cons Qty[a.u.]':
raise ValueError(
'The fields in the header line indicate that this file contains unsupported data.'
)
# Load lines
for line in fin:
if slice_match(sub, counter):
words = line.split()
write_to_dataset(ds_step, float(words[0]), row)
write_to_dataset(ds_time, float(words[1]) * femtosecond, row)
write_to_dataset(ds_ke, float(words[2]), row)
write_to_dataset(ds_temp, float(words[3]), row)
write_to_dataset(ds_pe, float(words[4]), row)
write_to_dataset(ds_cq, float(words[5]), row)
row += 1
counter += 1
fin.close()
# Check number of rows
check_trajectory_rows(tgrp, dss, row)
def write_lammps_table(ff,
fn='lammps.table',
rmin=0.50 * angstrom,
nrows=2500,
unit_style='electron'):
'''
Write tables containing noncovalent interactions for LAMMPS.
For every pair of ffatypes, a separate table is generated.
Because electrostatic interactions require a specific treatment, point-
charge electrostatics are NOT included in the tables.
When distributed charge distributions (e.g. Gaussian) are used, this
complicates matters. LAMMPS will still only treat point-charge
electrostatics using a dedicated method (e.g. Ewald or PPPM), so the
table has to contain the difference between the distributed charges and
the point charge electrostatic interactions. This also means that every
ffatype need a unique charge distribution, i.e. all atoms of the same
atom type need to have the same charge and Gaussian radius.
All pair potentials contributing to the table need to have the same
scalings for near-neighbor interactions; this is however independent
of the generation of the table and is dealt with elsewhere
**Arguments:**
ff
Yaff ForceField instance
**Optional arguments:**
fn
Filename where tables will be stored
'''
# Find out if we are dealing with electrostatics from distributed charges
corrections = []
for part in ff.parts:
if part.name == 'pair_ei':
if np.any(part.pair_pot.radii != 0.0):
# Create a ForcePart with electrostatics from distributed
# charges, completely in real space.
pair_pot_dist = PairPotEI(part.pair_pot.charges,
0.0,
part.pair_pot.rcut,
tr=part.pair_pot.get_truncation(),
dielectric=part.pair_pot.dielectric,
radii=part.pair_pot.radii)
fp_dist = ForcePartPair(ff.system, ff.nlist, part.scalings,
pair_pot_dist)
corrections.append((fp_dist, 1.0))
# Create a ForcePart with electrostatics from point
# charges, completely in real space.
pair_pot_point = PairPotEI(part.pair_pot.charges,
0.0,
part.pair_pot.rcut,
tr=part.pair_pot.get_truncation(),
dielectric=part.pair_pot.dielectric)
fp_point = ForcePartPair(ff.system, ff.nlist, part.scalings,
pair_pot_point)
corrections.append((fp_point, -1.0))
# Find the largest cut-off
rmax = 0.0
for part in ff.parts:
if part.name.startswith('pair_'):
if part.name == 'pair_ei' and len(corrections) == 0: continue
rmax = np.amax([rmax, part.pair_pot.rcut])
# Get LAMMPS ffatypes
ffatypes, ffatype_ids = get_lammps_ffatypes(ff)
# Select atom pairs for each pair of atom types
ffa_pairs = []
for i in range(len(ffatypes)):
index0 = np.where(ffatype_ids == i)[0][0]
for j in range(i, len(ffatypes)):
index1 = -1
candidates = np.where(ffatype_ids == j)[0]
for cand in candidates:
if cand==index0 or cand in ff.system.neighs1[index0] or\
cand in ff.system.neighs2[index0] or cand in ff.system.neighs3[index0] or\
cand in ff.system.neighs4[index0]:
continue
else:
index1 = cand
break
if index1 == -1:
log("ERROR constructing LAMMPS tables: there is no pair of atom types %s-%s which are not near neighbors"
% (ffatypes[i], ffatypes[j]))
log("Consider using a supercell to fix this problem")
raise ValueError
ffa_pairs.append([index0, index1])
if log.do_medium:
with log.section('LAMMPS'):
log("Generating LAMMPS table with covalent interactions")
log.hline()
log("rmin = %s | rmax = %s" % (log.length(rmin), log.length(rmax)))
# We only consider one neighbor interaction
ff.compute()
ff.nlist.nneigh = 1
# Construct array of evenly spaced values
distances = np.linspace(rmin, rmax, nrows)
ftab = open(fn, 'w')
ftab.write("# LAMMPS tabulated potential generated by Yaff\n")
ftab.write("# All quantities in atomic units\n")
ftab.write(
"# The names of the tables refer to the ffatype_ids that have to be used in the Yaff system\n"
)
ftab.write("#%4s %13s %21s %21s\n" % ("i", "d", "V", "F"))
# Loop over all atom pairs
for index0, index1 in ffa_pairs:
energies = []
for d in distances:
gposnn = np.zeros(ff.system.pos.shape, float)
ff.nlist.neighs[0] = (index0, index1, d, 0.0, 0.0, d, 0, 0, 0)
energy = 0.0
for part in ff.parts:
if not part.name.startswith('pair'): continue
if part.name == 'pair_ei': continue
energy += part.compute(gpos=gposnn)
for part, sign in corrections:
gposcorr = np.zeros(ff.system.pos.shape, float)
energy += sign * part.compute(gpos=gposcorr)
gposnn[:] += sign * gposcorr
row = [d, energy, gposnn[index0, 2]]
energies.append(row)
energies = np.asarray(energies)
ffai = ffatypes[ffatype_ids[index0]]
ffaj = ffatypes[ffatype_ids[index1]]
if np.all(energies[:, 1] == 0.0):
log.warn("Noncovalent energies between atoms %d (%s) and %d (%s) are zero"\
% (index0,ffai,index1,ffaj))
if np.all(energies[:, 2] == 0.0):
log.warn("Noncovalent forces between atoms %d (%s) and %d (%s) are zero"\
% (index0,ffai,index1,ffaj))
if ffai > ffaj:
name = '%s---%s' % (str(ffai), str(ffaj))
else:
name = '%s---%s' % (str(ffaj), str(ffai))
ftab.write("%s\nN %d R %13.8f %13.8f\n\n" %
(name, nrows, rmin / lammps_units[unit_style]['distance'],
rmax / lammps_units[unit_style]['distance']))
for irow, row in enumerate(energies):
ftab.write(
"%05d %+13.8f %+21.12f %+21.12f\n" %
(irow + 1, row[0] / lammps_units[unit_style]['distance'],
row[1] / lammps_units[unit_style]['energy'],
row[2] / lammps_units[unit_style]['energy'] *
lammps_units[unit_style]['distance']))
if log.do_medium:
log("%s done" % name)
def check(self):
"""Perform a slow internal consistency test.
Use this for debugging only. It is assumed that self.rmax is set correctly.
"""
# 0) Some initial tests
assert (
(self.neighs['a'][:self.nneigh] > self.neighs['b'][:self.nneigh]) |
(self.neighs['r0'][:self.nneigh] != 0) |
(self.neighs['r1'][:self.nneigh] != 0) |
(self.neighs['r2'][:self.nneigh] != 0)).all()
# A) transform the current nlist into a set
actual = self.to_dictionary()
# B) Define loops of cell vectors
if self.system.cell.nvec == 3:
def rloops():
for r2 in xrange(0, self.rmax[2] + 1):
if r2 == 0:
r1_start = 0
else:
r1_start = -self.rmax[1]
for r1 in xrange(r1_start, self.rmax[1] + 1):
if r2 == 0 and r1 == 0:
r0_start = 0
else:
r0_start = -self.rmax[0]
for r0 in xrange(r0_start, self.rmax[0] + 1):
yield r0, r1, r2
elif self.system.cell.nvec == 2:
def rloops():
for r1 in xrange(0, self.rmax[1] + 1):
if r1 == 0:
r0_start = 0
else:
r0_start = -self.rmax[0]
for r0 in xrange(r0_start, self.rmax[0] + 1):
yield r0, r1, 0
elif self.system.cell.nvec == 1:
def rloops():
for r0 in xrange(0, self.rmax[0] + 1):
yield r0, 0, 0
else:
def rloops():
yield 0, 0, 0
# C) Compute the nlists the slow way
validation = {}
nvec = self.system.cell.nvec
for r0, r1, r2 in rloops():
for a in xrange(self.system.natom):
for b in xrange(a + 1):
if r0 != 0 or r1 != 0 or r2 != 0:
signs = [1, -1]
elif a > b:
signs = [1]
else:
continue
for sign in signs:
delta = self.system.pos[b] - self.system.pos[a]
self.system.cell.mic(delta)
delta *= sign
if nvec > 0:
self.system.cell.add_vec(
delta,
np.array([r0, r1, r2])[:nvec])
d = np.linalg.norm(delta)
if d < self.rcut + self.skin:
if sign == 1:
key = a, b, r0, r1, r2
else:
key = b, a, r0, r1, r2
value = np.array([d, delta[0], delta[1], delta[2]])
validation[key] = value
# D) Compare
wrong = False
with log.section('NLIST'):
for key0, value0 in validation.iteritems():
value1 = actual.pop(key0, None)
if value1 is None:
log('Missing: ', key0)
log(' Validation %s %s %s %s' %
(log.length(value0[0]), log.length(value0[1]),
log.length(value0[2]), log.length(value0[3])))
wrong = True
elif abs(value0 -
value1).max() > 1e-10 * log.length.conversion:
log('Different:', key0)
log(' Actual %s %s %s %s' %
(log.length(value1[0]), log.length(value1[1]),
log.length(value1[2]), log.length(value1[3])))
log(' Validation %s %s %s %s' %
(log.length(value0[0]), log.length(value0[1]),
log.length(value0[2]), log.length(value0[3])))
log(' Difference %10.3e %10.3e %10.3e %10.3e' % tuple(
(value0 - value1) / log.length.conversion))
log(' AbsMaxDiff %10.3e' %
(abs(value0 - value1).max() / log.length.conversion))
wrong = True
for key1, value1 in actual.iteritems():
log('Redundant:', key1)
log(' Actual %s %s %s %s' %
(log.length(value1[0]), log.length(value1[1]),
log.length(value1[2]), log.length(value1[3])))
wrong = True
assert not wrong
def pca_projection(f_target,
f,
pm,
start=0,
end=None,
step=1,
select=None,
path='trajectory/pos',
mw=True):
"""
Determines the principal components of an MD simulation
**Arguments:**
f_target
Path to an h5.File instance to which the results are written.
f
An h5.File instance containing the trajectory data.
pm
An array containing the principal modes in its columns
**Optional arguments:**
start
The first sample to be considered for analysis. This may be
negative to indicate that the analysis should start from the
-start last samples.
end
The last sample to be considered for analysis. This may be
negative to indicate that the last -end sample should not be
considered.
step
The spacing between the samples used for the analysis
select
A list of atom indexes that are considered for the computation
of the spectrum. If not given, all atoms are used.
path
The path of the dataset that contains the time dependent data in
the HDF5 file. The first axis of the array must be the time
axis.
mw
If mass_weighted is True, the covariance matrix is mass-weighted.
"""
# Load in the relevant data
q = f[path][start:end:step, :, :]
# Select the given atoms
if select is not None:
q = q[:, select, :]
# Reshape such that all Cartesian coordinates are treated equally
q = q.reshape(q.shape[0], -1)
# If necessary, weight with the mass
if mw:
# Select the necessary masses
masses = f['system/masses']
if select is not None:
masses = masses[select]
# Repeat d times, with d the dimension
masses = np.repeat(masses, 3)
# Reweight with the masses
q *= np.sqrt(masses)
# Calculation of the principal components: projection of each q_j on the principal modes
with log.section('PCA'):
log('Determining principal components')
prin_comp = np.dot(q, pm)
# Create output HDF5 file
g = h5.File(f_target, 'a')
if not 'pca' in g:
pca = g.create_group('pca')
else:
pca = g['pca']
pca.create_dataset('pc', data=prin_comp)
return pca
def write_principal_mode(f,
f_pca,
index,
n_frames=100,
select=None,
mw=True,
scaling=1.):
"""
Writes out one xyz file per principal mode given in index
**Arguments:**
f
Path to an h5.File instance containing the original data.
f_pca
Path to an h5.File instance containing the PCA, with reference structure, eigenvalues
and principal modes.
index
An array containing the principal modes which need to be written out.
**Optional arguments:**
n_frames
The number of frames in each xyz file.
select
A list of atom indexes that are considered for the computation
of the spectrum. If not given, all atoms are used.
mw
If mass_weighted is True, the covariance matrix is assumed to be mass-weighted.
scaling
Scaling factor applied to the maximum deviation of the principal mode (i.e. the maximum
principal component for that mode)
"""
# Load in the relevant data
# Atom numbers, masses and initial frame
numbers = f['system/numbers']
masses = f['system/masses']
pos = f['trajectory/pos']
if select is not None:
numbers = numbers[select]
masses = masses[select]
pos = pos[:, select, :]
masses = np.repeat(masses, 3)
# Data from the PC analysis
grp = f_pca['pca']
# The selected principal modes
pm = grp['pm'][:, index]
# The corresponding eigenvalues
eigval = grp['eigvals'][index]
# And the principal components
pc = grp['pc'][:, index]
with log.section('PCA'):
for i in xrange(len(index)):
log('Writing out principal mode %s' % index[i])
if eigval[i] < 0:
Warning('Negative eigenvalue encountered, skipping this entry')
break
# Determine maximum fluctuation (in units of meter*sqrt(kilogram))
max_fluct = np.max(np.abs(pc[:, i]))
# Initialize XYZWriter from molmod package
xw = XYZWriter('pm_%s.xyz' % index[i],
[pd[number].symbol for number in numbers])
# Determine index in trajectory closest to rest state, and the corresponding positions
ind_min = np.argmin(np.abs(pc[:, i]))
r_ref = pos[ind_min, :, :]
for j in xrange(n_frames):
q_var = scaling * pm[:, i] * max_fluct * (2. * j -
n_frames) / n_frames
if mw:
q_var /= np.sqrt(masses)
r = r_ref + q_var.reshape(-1, 3)
xw.dump('Frame%s' % j, r)
del xw
def check(self):
"""Perform a slow internal consistency test.
Use this for debugging only. It is assumed that self.rmax is set correctly.
"""
# 0) Some initial tests
assert (
(self.neighs['a'][:self.nneigh] > self.neighs['b'][:self.nneigh]) |
(self.neighs['r0'][:self.nneigh] != 0) |
(self.neighs['r1'][:self.nneigh] != 0) |
(self.neighs['r2'][:self.nneigh] != 0)
).all()
# A) transform the current nlist into a set
actual = self.to_dictionary()
# B) Define loops of cell vectors
if self.system.cell.nvec == 3:
def rloops():
for r2 in xrange(0, self.rmax[2]+1):
if r2 == 0:
r1_start = 0
else:
r1_start = -self.rmax[1]
for r1 in xrange(r1_start, self.rmax[1]+1):
if r2 == 0 and r1 == 0:
r0_start = 0
else:
r0_start = -self.rmax[0]
for r0 in xrange(r0_start, self.rmax[0]+1):
yield r0, r1, r2
elif self.system.cell.nvec == 2:
def rloops():
for r1 in xrange(0, self.rmax[1]+1):
if r1 == 0:
r0_start = 0
else:
r0_start = -self.rmax[0]
for r0 in xrange(r0_start, self.rmax[0]+1):
yield r0, r1, 0
elif self.system.cell.nvec == 1:
def rloops():
for r0 in xrange(0, self.rmax[0]+1):
yield r0, 0, 0
else:
def rloops():
yield 0, 0, 0
# C) Compute the nlists the slow way
validation = {}
nvec = self.system.cell.nvec
for r0, r1, r2 in rloops():
for a in xrange(self.system.natom):
for b in xrange(a+1):
if r0!=0 or r1!=0 or r2!=0:
signs = [1, -1]
elif a > b:
signs = [1]
else:
continue
for sign in signs:
delta = self.system.pos[b] - self.system.pos[a]
self.system.cell.mic(delta)
delta *= sign
if nvec > 0:
self.system.cell.add_vec(delta, np.array([r0, r1, r2])[:nvec])
d = np.linalg.norm(delta)
if d < self.rcut + self.skin:
if sign == 1:
key = a, b, r0, r1, r2
else:
key = b, a, r0, r1, r2
value = np.array([d, delta[0], delta[1], delta[2]])
validation[key] = value
# D) Compare
wrong = False
with log.section('NLIST'):
for key0, value0 in validation.iteritems():
value1 = actual.pop(key0, None)
if value1 is None:
log('Missing: ', key0)
log(' Validation %s %s %s %s' % (
log.length(value0[0]), log.length(value0[1]),
log.length(value0[2]), log.length(value0[3])
))
wrong = True
elif abs(value0 - value1).max() > 1e-10*log.length.conversion:
log('Different:', key0)
log(' Actual %s %s %s %s' % (
log.length(value1[0]), log.length(value1[1]),
log.length(value1[2]), log.length(value1[3])
))
log(' Validation %s %s %s %s' % (
log.length(value0[0]), log.length(value0[1]),
log.length(value0[2]), log.length(value0[3])
))
log(' Difference %10.3e %10.3e %10.3e %10.3e' %
tuple((value0 - value1)/log.length.conversion)
)
log(' AbsMaxDiff %10.3e' %
(abs(value0 - value1).max()/log.length.conversion)
)
wrong = True
for key1, value1 in actual.iteritems():
log('Redundant:', key1)
log(' Actual %s %s %s %s' % (
log.length(value1[0]), log.length(value1[1]),
log.length(value1[2]), log.length(value1[3])
))
wrong = True
assert not wrong
def calc_pca(f_target,
cov_mat=None,
f=None,
q_ref=None,
start=0,
end=None,
step=1,
select=None,
path='trajectory/pos',
mw=True,
temp=None):
"""
Performs a principle component analysis of the given trajectory.
**Arguments:**
f_target
Path to an h5.File instance to which the results are written.
**Optional arguments:**
cov_mat
The covariance matrix, if already calculated. If not provided,
the covariance matrix will be calculatd based on the file f.
f
An h5.File instance containing the trajectory data.
q_ref
Reference vector of the positions. If not provided, the ensemble
average is taken.
start
The first sample to be considered for analysis. This may be
negative to indicate that the analysis should start from the
-start last samples.
end
The last sample to be considered for analysis. This may be
negative to indicate that the last -end sample should not be
considered.
step
The spacing between the samples used for the analysis
select
A list of atom indexes that are considered for the computation
of the spectrum. If not given, all atoms are used.
path
The path of the dataset that contains the time dependent data in
the HDF5 file. The first axis of the array must be the time
axis.
mw
If mass_weighted is True, the covariance matrix is mass-weighted.
temp
Temperature at which the simulation is carried out, necessary to determine the frequencies
"""
if cov_mat is None:
if f is None:
AssertionError(
'No covariance matrix nor h5.File instance provided.')
else:
with log.section('PCA'):
log('Calculating covariance matrix')
cov_mat, q_ref = calc_cov_mat(f, q_ref, start, end, step,
select, path, mw)
with log.section('PCA'):
log('Diagonalizing the covariance matrix')
# Eigenvalue decomposition
eigval, eigvec = np.linalg.eigh(cov_mat)
# Order the eigenvalues in decreasing order
idx = eigval.argsort()[::-1]
eigval = eigval[idx]
eigvec = eigvec[:, idx]
# Create output HDF5 file
g = h5.File(f_target, 'w')
pca = g.create_group('pca')
# Output reference structure q_ref
pca.create_dataset('q_ref', data=q_ref)
# Output covariance matrix
pca.create_dataset('cov_matrix', data=cov_mat)
# Output eigenvectors in columns
pca.create_dataset('pm', data=eigvec)
# Output eigenvalues
pca.create_dataset('eigvals', data=eigval)
log('Determining inverse of the covariance matrix')
# Process matrix to determine inverse
# First, project out the three zero eigenvalues (translations)
eigvec_reduced = eigvec[:, :-3]
eigval_reduced = eigval[:-3]
# Second, calculate the reduced covariance matrix and its inverse
cov_mat_reduced = np.dot(
np.dot(eigvec_reduced, np.diag(eigval_reduced)), eigvec_reduced.T)
cov_mat_inverse = np.dot(
np.dot(eigvec_reduced, np.diag(1 / eigval_reduced)),
eigvec_reduced.T)
pca.create_dataset('cov_mat_red', data=cov_mat_reduced)
pca.create_dataset('cov_mat_inv', data=cov_mat_inverse)
# Third, if the temperature is specified, calculate the frequencies
# (the zero frequencies are mentioned last so that their index corresponds to the principal modes)
if temp is not None:
log('Determining frequencies')
frequencies = np.append(
np.sqrt(boltzmann * temp / eigval_reduced) / (2 * np.pi),
np.repeat(0, 3))
pca.create_dataset('freqs', data=frequencies)
return eigval, eigvec
def check_mic(self, system):
'''Check if each scale2 and scale3 are uniquely defined.
**Arguments:**
system
An instance of the system class, i.e. the one that is used to
create this scaling object.
This check is done by constructing for each scaled pair, all possible
bond paths between the two atoms. For each path, the bond vectors
(after applying the minimum image convention) are added. If for a
given pair, these sums of bond vectors differ between all possible
paths, the differences are expanded in cell vectors which can be used
to construct a proper supercell in which scale2 and scale3 pairs are
all uniquely defined.
'''
if system.cell.nvec == 0:
return
troubles = False
with log.section('SCALING'):
for i0, i1, scale, nbond in self.stab:
if nbond == 1:
continue
all_deltas = []
paths = []
for path in iter_paths(system, i0, i1, nbond):
delta_total = 0
for j0 in range(nbond):
j1 = j0 + 1
delta = system.pos[path[j0]] - system.pos[path[j1]]
system.cell.mic(delta)
delta_total += delta
all_deltas.append(delta_total)
paths.append(path)
all_deltas = np.array(all_deltas)
if abs(all_deltas.mean(axis=0) - all_deltas).max() > 1e-10:
troubles = True
if log.do_warning:
log.warn('Troublesome pair scaling detected.')
log('The following bond paths connect the same pair of '
'atoms, yet the relative vectors are different.')
for ipath in range(len(paths)):
log('%2i %27s %10s %10s %10s' % (
ipath,
','.join(str(index) for index in paths[ipath]),
log.length(all_deltas[ipath, 0]),
log.length(all_deltas[ipath, 1]),
log.length(all_deltas[ipath, 2]),
))
log('Differences between relative vectors in fractional '
'coordinates:')
for ipath0 in range(1, len(paths)):
for ipath1 in range(ipath0):
diff = all_deltas[ipath0] - all_deltas[ipath1]
diff_frac = np.dot(system.cell.gvecs, diff)
log('%2i %2i %10.4f %10.4f %10.4f' %
(ipath0, ipath1, diff_frac[0], diff_frac[1],
diff_frac[2]))
log.blank()
if troubles:
raise AssertionError(
'Due to the small spacing between some crystal planes, the scaling of non-bonding interactions will not work properly. Use a supercell to avoid this problem.'
)
def xyz_to_hdf5(f, fn_xyz, sub=slice(None), file_unit=angstrom, name='pos'):
"""Convert XYZ trajectory file to Yaff HDF5 format.
**Arguments:**
f
An open and writable HDF5 file.
fn_xyz
The filename of the XYZ trajectory file.
**Optional arguments:**
sub
The sub argument for the XYZReader. This must be a slice object that
defines the subsampling of the XYZ file reader. By default all
frames are read.
file_unit
The unit of the data in the XYZ file. [default=angstrom]
name
The name of the HDF5 dataset where the trajectory is stored. This
array is stored in the 'trajectory' group.
This routine will also test the consistency of the row attribute of the
trajectory group. If some trajectory data is already present, it will be
replaced by the new data. It is highly recommended to first initialize
the HDF5 file with the ``to_hdf5`` method of the System class.
"""
with log.section('XYZH5'):
if log.do_medium:
log('Loading XYZ file \'%s\' into \'trajectory/%s\' of HDF5 file \'%s\''
% (fn_xyz, name, f.filename))
# First make sure the HDF5 file has a system description that is consistent
# with the XYZ file.
if 'system' not in f:
raise ValueError('The HDF5 file must contain a system group.')
if 'numbers' not in f['system']:
raise ValueError(
'The HDF5 file must have a system group with atomic numbers.')
xyz_reader = XYZReader(fn_xyz, sub=sub, file_unit=file_unit)
if len(xyz_reader.numbers) != len(f['system/numbers']):
raise ValueError(
'The number of atoms in the HDF5 and the XYZ files does not match.'
)
if (xyz_reader.numbers != f['system/numbers']).any():
log.warn(
'The atomic numbers of the HDF5 and XYZ file do not match.')
# Take care of the trajectory group
tgrp = get_trajectory_group(f)
# Take care of the dataset
ds, = get_trajectory_datasets(tgrp,
(name, (len(xyz_reader.numbers), 3)))
# Fill the dataset with data.
row = get_last_trajectory_row([ds])
for title, coordinates in xyz_reader:
write_to_dataset(ds, coordinates, row)
row += 1
# Check number of rows
check_trajectory_rows(tgrp, [ds], row)
def dlpoly_history_to_hdf5(f,
fn_history,
sub=slice(None),
pos_unit=angstrom,
vel_unit=angstrom / picosecond,
frc_unit=amu * angstrom / picosecond**2,
time_unit=picosecond,
mass_unit=amu):
"""Convert DLPolay History trajectory file to Yaff HDF5 format.
**Arguments:**
f
An open and writable HDF5 file.
fn_history
The filename of the DLPOLY history file.
**Optional arguments:**
sub
The sub argument for the DLPolyHistoryReader. This must be a slice
object that defines the subsampling of the samples from the history
file. By default all frames are read.
pos_unit, vel_unit, frc_unit, time_unit and mass_unit
The units used in the dlpoly history file. The default values
correspond to the defaults used in DLPOLY.
This routine will also test the consistency of the row attribute of the
trajectory group. If some trajectory data is already present, it will be
replaced by the new data. It is highly recommended to first initialize
the HDF5 file with the ``to_hdf5`` method of the System class.
"""
with log.section('DPH5'):
if log.do_medium:
log('Loading DLPOLY history file \'%s\' into \'trajectory\' of HDF5 file \'%s\''
% (fn_history, f.filename))
# Take care of the data group
tgrp = get_trajectory_group(f)
# Open the history file for reading
hist_reader = DLPolyHistoryReader(fn_history, sub, pos_unit, vel_unit,
frc_unit, time_unit, mass_unit)
# Take care of the datasets that should always be present
natom = hist_reader.num_atoms
dss = get_trajectory_datasets(
tgrp,
('step', (1, )),
('time', (1, )),
('cell', (3, 3)),
('pos', (natom, 3)),
)
ds_step, ds_time, ds_cell, ds_pos = dss
# Take care of optional data sets
if hist_reader.keytrj > 0:
ds_vel = get_trajectory_datasets(tgrp, ('vel', (natom, 3)))[0]
dss.append(ds_vel)
if hist_reader.keytrj > 1:
ds_frc = get_trajectory_datasets(tgrp, ('frc', (natom, 3)))[0]
dss.append(ds_frc)
# Decide on the first row to start writing data
row = get_last_trajectory_row(dss)
# Load data
for frame in hist_reader:
write_to_dataset(ds_step, frame["step"], row)
write_to_dataset(ds_time, frame["time"], row)
write_to_dataset(ds_cell, frame["cell"].T, row)
write_to_dataset(ds_pos, frame["pos"], row)
if hist_reader.keytrj > 0:
write_to_dataset(ds_vel, frame["vel"], row)
if hist_reader.keytrj > 1:
write_to_dataset(ds_frc, frame["frc"], row)
row += 1
# Check number of rows
check_trajectory_rows(tgrp, dss, row)
def write_principal_mode(f, f_pca, index, n_frames=100, select=None, mw=True, scaling=1.):
"""
Writes out one xyz file per principal mode given in index
**Arguments:**
f
Path to an h5.File instance containing the original data.
f_pca
Path to an h5.File instance containing the PCA, with reference structure, eigenvalues
and principal modes.
index
An array containing the principal modes which need to be written out.
**Optional arguments:**
n_frames
The number of frames in each xyz file.
select
A list of atom indexes that are considered for the computation
of the spectrum. If not given, all atoms are used.
mw
If mass_weighted is True, the covariance matrix is assumed to be mass-weighted.
scaling
Scaling factor applied to the maximum deviation of the principal mode (i.e. the maximum
principal component for that mode)
"""
# Load in the relevant data
# Atom numbers, masses and initial frame
numbers = f['system/numbers']
masses = f['system/masses']
pos = f['trajectory/pos']
if select is not None:
numbers = numbers[select]
masses = masses[select]
pos = pos[:,select,:]
masses = np.repeat(masses,3)
# Data from the PC analysis
grp = f_pca['pca']
# The selected principal modes
pm = grp['pm'][:,index]
# The corresponding eigenvalues
eigval = grp['eigvals'][index]
# And the principal components
pc = grp['pc'][:,index]
with log.section('PCA'):
for i in range(len(index)):
log('Writing out principal mode %s' %index[i])
if eigval[i] < 0:
Warning('Negative eigenvalue encountered, skipping this entry')
break
# Determine maximum fluctuation (in units of meter*sqrt(kilogram))
max_fluct = np.max(np.abs(pc[:,i]))
# Initialize XYZWriter from molmod package
xw = XYZWriter('pm_%s.xyz' %index[i], [pd[number].symbol for number in numbers])
# Determine index in trajectory closest to rest state, and the corresponding positions
ind_min = np.argmin(np.abs(pc[:,i]))
r_ref = pos[ind_min,:,:]
for j in range(n_frames):
q_var = scaling*pm[:,i]*max_fluct*(2.*j-n_frames)/n_frames
if mw:
q_var /= np.sqrt(masses)
r = r_ref + q_var.reshape(-1,3)
xw.dump('Frame%s' %j, r)
del xw
def pca_convergence(f,
eq_time=0 * picosecond,
n_parts=None,
step=1,
fn='PCA_convergence',
n_bootstrap=50,
mw=True):
"""
Calculates the convergence of the simulation by calculating the pca
similarity for different subsets of the simulation.
**Arguments:**
f
An h5.File instance containing the trajectory data.
**Optional arguments:**
eq_time
Equilibration time, discarded from the simulation.
n_parts
Array containing the number of parts in which
the total simulation is divided.
step
Stepsize used in the trajectory.
fn
Filename containing the convergence plot.
n_bootstrap
The number of bootstrapped trajectories.
mw
If mass_weighted is True, the covariance matrix is mass-weighted.
"""
# Configure n_parts, the array containing the number of parts in which the total simulation is divided
if n_parts is None:
n_parts = np.array([1, 3, 10, 30, 100, 300])
# Read in the timestep and the number of atoms
time = f['trajectory/time']
timestep = time[1] - time[0]
time_length = len(time)
# Determine the equilibration size
eq_size = int(eq_time / timestep)
### ---PART A: SIMILARITY OF THE TRUE TRAJECTORY--- ###
# Calculate the covariance matrix of the whole production run as golden standard
covar_total, q_ref = calc_cov_mat(f, start=eq_size, step=step, mw=mw)
# Initialize the average similarity vector of the divided trajectories
sim_block = np.zeros(len(n_parts))
# Calculate this average similarity vector
for j in xrange(len(n_parts)):
# Determine in how many parts the trajectory should be divided and the corresponding block size
n_part = n_parts[j]
block_size = (time_length - eq_size) / n_part
# Calculate the n_part covariance matrices and compare with the total covariance matrix
tot_sim_block = 0
for i in xrange(n_part):
start = eq_size + i * block_size
covars, tmp = calc_cov_mat(f,
start=start,
end=start + block_size + 1,
step=step,
mw=mw)
tot_sim_block += pca_similarity(covars, covar_total)
# Determine the average similarity
sim_block[j] = tot_sim_block / n_part
### ---PART B: SIMILARITY OF BOOTSTRAPPED TRAJECTORIES --- ###
# Read in the positions, which will be used to generate bootstrapped trajectories
pos = f['trajectory/pos'][eq_size:, :, :]
pos = pos.reshape(pos.shape[0], -1)
if mw:
# Read in the masses of the atoms, and replicate them d times (d=dimension)
masses = f['system/masses']
masses = np.repeat(masses, 3)
# Create the mass-weighted positions matrix, on which the bootstrapping will be based
pos *= np.sqrt(masses)
# Initialize the vector containing the average similarity over all the bootstrapped, divided trajectories
sim_bt_all = np.zeros(len(n_parts))
for k in xrange(n_bootstrap):
with log.section('PCA'):
log('Processing %s of %s bootstrapped trajectories' %
(k + 1, n_bootstrap))
# Create a bootstrapped trajectory bt
pos_bt = np.zeros(pos.shape)
random_time = random.random(time_length) * time_length
for h in np.arange(time_length):
pos_bt[h, :] = pos[random_time[h], :]
# Covariance matrix of the total bootstrapped trajectory
covar_bt_total, tmp = calc_cov_mat_internal(pos_bt)
# Initialize the vector containing the average similarity over the different blocks,
# for the given bootstrapped trajectory
sim_bt = np.zeros(len(n_parts))
for j in xrange(len(n_parts)):
# Calculate the number of blocks, as well as the block size
n_part = n_parts[j]
block_size = (len(time) - eq_size) / n_part
tot_sim_bt = 0
# Calculate the total similarity of this number of blocks, for this bootstrapped trajectory
for i in xrange(n_part):
start = eq_size + i * block_size
pos_bt_block = pos_bt[start:start + block_size:step]
covars_bt, tmp = calc_cov_mat_internal(pos_bt_block)
tot_sim_bt += pca_similarity(covars_bt, covar_bt_total)
# Calculate the average similarity for this number of blocks, for this bootstrapped trajectory
sim_bt[j] = tot_sim_bt / n_part
sim_bt_all += sim_bt
# Calculate the average similarity over all bootstrapped trajectories
sim_bt_all /= n_bootstrap
### ---PART C: PROCESSING THE RESULTS --- ###
pt.clf()
pt.semilogx((time[-1] - time[0]) / n_parts / picosecond,
sim_block / sim_bt_all, 'r-')
pt.semilogx((time[-1] - time[0]) / n_parts / picosecond,
sim_block / sim_bt_all, 'rs')
pt.xlabel('Block size [ps]')
pt.ylabel('PCA similarity (1=perfectly similar)')
pt.title('Convergence assessment via PCA: ' + fn)
pt.ylim([0, 1])
pt.savefig(fn + '.png')
pt.savefig(fn + '.pdf', format='pdf')
return sim_block / sim_bt_all
def pca_convergence(f, eq_time=0*picosecond, n_parts=None, step=1, fn='PCA_convergence', n_bootstrap=50, mw=True):
"""
Calculates the convergence of the simulation by calculating the pca
similarity for different subsets of the simulation.
**Arguments:**
f
An h5.File instance containing the trajectory data.
**Optional arguments:**
eq_time
Equilibration time, discarded from the simulation.
n_parts
Array containing the number of parts in which
the total simulation is divided.
step
Stepsize used in the trajectory.
fn
Filename containing the convergence plot.
n_bootstrap
The number of bootstrapped trajectories.
mw
If mass_weighted is True, the covariance matrix is mass-weighted.
"""
# Configure n_parts, the array containing the number of parts in which the total simulation is divided
if n_parts is None:
n_parts = np.array([1,3,10,30,100,300])
# Read in the timestep and the number of atoms
time = f['trajectory/time']
timestep = time[1] - time[0]
time_length = len(time)
# Determine the equilibration size
eq_size = int(eq_time/timestep)
### ---PART A: SIMILARITY OF THE TRUE TRAJECTORY--- ###
# Calculate the covariance matrix of the whole production run as golden standard
covar_total, q_ref = calc_cov_mat(f, start=eq_size, step=step, mw=mw)
# Initialize the average similarity vector of the divided trajectories
sim_block = np.zeros(len(n_parts))
# Calculate this average similarity vector
for j in range(len(n_parts)):
# Determine in how many parts the trajectory should be divided and the corresponding block size
n_part = n_parts[j]
block_size = (time_length-eq_size)//n_part
# Calculate the n_part covariance matrices and compare with the total covariance matrix
tot_sim_block=0
for i in range(n_part):
start = eq_size + i*block_size
covars, tmp = calc_cov_mat(f, start=start, end=start+block_size+1, step=step, mw=mw)
tot_sim_block += pca_similarity(covars, covar_total)
# Determine the average similarity
sim_block[j] = tot_sim_block/n_part
### ---PART B: SIMILARITY OF BOOTSTRAPPED TRAJECTORIES --- ###
# Read in the positions, which will be used to generate bootstrapped trajectories
pos = f['trajectory/pos'][eq_size:,:,:]
pos = pos.reshape(pos.shape[0], -1)
if mw:
# Read in the masses of the atoms, and replicate them d times (d=dimension)
masses = f['system/masses']
masses = np.repeat(masses,3)
# Create the mass-weighted positions matrix, on which the bootstrapping will be based
pos *= np.sqrt(masses)
# Initialize the vector containing the average similarity over all the bootstrapped, divided trajectories
sim_bt_all = np.zeros(len(n_parts))
for k in range(n_bootstrap):
with log.section('PCA'):
log('Processing %s of %s bootstrapped trajectories' %(k+1,n_bootstrap))
# Create a bootstrapped trajectory bt
pos_bt = np.zeros(pos.shape)
random_time = random.random(time_length)*time_length
for h in np.arange(time_length):
pos_bt[h,:] = pos[random_time[h],:]
# Covariance matrix of the total bootstrapped trajectory
covar_bt_total, tmp = calc_cov_mat_internal(pos_bt)
# Initialize the vector containing the average similarity over the different blocks,
# for the given bootstrapped trajectory
sim_bt = np.zeros(len(n_parts))
for j in range(len(n_parts)):
# Calculate the number of blocks, as well as the block size
n_part = n_parts[j]
block_size = (len(time)-eq_size)//n_part
tot_sim_bt = 0
# Calculate the total similarity of this number of blocks, for this bootstrapped trajectory
for i in range(n_part):
start = eq_size + i*block_size
pos_bt_block = pos_bt[start:start+block_size:step]
covars_bt, tmp = calc_cov_mat_internal(pos_bt_block)
tot_sim_bt += pca_similarity(covars_bt, covar_bt_total)
# Calculate the average similarity for this number of blocks, for this bootstrapped trajectory
sim_bt[j] = tot_sim_bt/n_part
sim_bt_all += sim_bt
# Calculate the average similarity over all bootstrapped trajectories
sim_bt_all /= n_bootstrap
### ---PART C: PROCESSING THE RESULTS --- ###
pt.clf()
pt.semilogx((time[-1]-time[0])/n_parts/picosecond, sim_block/sim_bt_all, 'r-')
pt.semilogx((time[-1]-time[0])/n_parts/picosecond, sim_block/sim_bt_all, 'rs')
pt.xlabel('Block size [ps]')
pt.ylabel('PCA similarity (1=perfectly similar)')
pt.title('Convergence assessment via PCA: ' + fn)
pt.ylim([0,1])
pt.savefig(fn+'.png')
pt.savefig(fn+'.pdf', format='pdf')
return sim_block/sim_bt_all
def __init__(self,
numbers,
pos,
scopes=None,
scope_ids=None,
ffatypes=None,
ffatype_ids=None,
bonds=None,
rvecs=None,
charges=None,
radii=None,
valence_charges=None,
dipoles=None,
radii2=None,
masses=None):
r'''Initialize a System object.
**Arguments:**
numbers
A numpy array with atomic numbers
pos
A numpy array (N,3) with atomic coordinates in Bohr.
**Optional arguments:**
scopes
A list with scope names
scope_ids
A list of scope indexes that links each atom with an element of
the scopes list. If this argument is not present, while scopes
is given, it is assumed that scopes contains a scope name for
every atom, i.e. that it is a list with length natom. In that
case, it will be converted automatically to a scopes list
with only unique name together with a corresponding scope_ids
array.
ffatypes
A list of labels of the force field atom types.
ffatype_ids
A list of atom type indexes that links each atom with an element
of the list ffatypes. If this argument is not present, while
ffatypes is given, it is assumed that ffatypes contains an
atom type for every element, i.e. that it is a list with length
natom. In that case, it will be converted automatically to
a short ffatypes list with only unique elements (within each
scope) together with a corresponding ffatype_ids array.
bonds
a numpy array (B,2) with atom indexes (counting starts from
zero) to define the chemical bonds.
rvecs
An array whose rows are the unit cell vectors. At most three
rows are allowed, each containing three Cartesian coordinates.
charges
An array of atomic charges
radii
An array of atomic radii, :math:`R_{A,c}`, that determine shape of the atomic
charge distribution:
.. math::
\rho_{A,c}(\mathbf{r}) = \frac{q_A}{\pi^{3/2}R_{A,c}^3} \exp\left(
-\frac{|r - \mathbf{R}_A|^2}{R_{A,c}^2}
\right)
valence_charges
In case a point-core + distribute valence charge is used, this
vector contains the valence charges. The core charges can be
computed by subtracting the valence charges from the net
charges.
dipoles
An array of atomic dipoles
radii2
An array of atomic radii, :math:`R_{A,d}`, that determine shape of the
atomic dipole distribution:
.. math::
\rho_{A,d}(\mathbf{r}) = -2\frac{\mathbf{d}_A \cdot (\mathbf{r} - \mathbf{R}_A)}{
\sqrt{\pi} R_{A,d}^5
}\exp\left(
-\frac{|r - \mathbf{R}_A|^2}{R_{A,d}^2}
\right)
masses
The atomic masses (in atomic units, i.e. m_e)
Several attributes are derived from the (optional) arguments:
* ``cell`` contains the rvecs attribute and is an instance of the
``Cell`` class.
* ``neighs1``, ``neighs2`` and ``neighs3`` are dictionaries derived
from ``bonds`` that contain atoms that are separated 1, 2 and 3
bonds from a given atom, respectively. This means that i in
system.neighs3[j] is ``True`` if there are three bonds between
atoms i and j.
'''
if len(numbers.shape) != 1:
raise ValueError(
'Argument numbers must be a one-dimensional array.')
if pos.shape != (len(numbers), 3):
raise ValueError(
'The pos array must have Nx3 rows. Mismatch with numbers argument with shape (N,).'
)
self.numbers = numbers
self.pos = pos
self.ffatypes = ffatypes
self.ffatype_ids = ffatype_ids
self.scopes = scopes
self.scope_ids = scope_ids
self.bonds = bonds
self.cell = Cell(rvecs)
self.charges = charges
self.radii = radii
self.valence_charges = valence_charges
self.dipoles = dipoles
self.radii2 = radii2
self.masses = masses
with log.section('SYS'):
# report some stuff
self._init_log()
# compute some derived attributes
self._init_derived()