class ForcePartValence(ForcePart):
'''The covalent part of a force-field model.
The covalent force field is implemented in a three-layer approach,
similar to the implementation of a neural network:
1. The first layer consists of a :class:`yaff.pes.dlist.DeltaList` object
that computes all the relative vectors needed for the internal
coordinates in the covalent energy terms. This list is automatically
built up as energy terms are added with the ``add_term`` method. This
list also takes care of transforming `derivatives of the energy
towards relative vectors` into `derivatives of the energy towards
Cartesian coordinates and the virial tensor`.
2. The second layer consist of a
:class:`yaff.pes.iclist.InternalCoordinateList` object that computes
the internal coordinates, based on the ``DeltaList``. This list is
also automatically built up as energy terms are added. The same list
is also responsible for transforming `derivatives of the energy
towards internal coordinates` into `derivatives of the energy towards
relative vectors`.
3. The third layers consists of a :class:`yaff.pes.vlist.ValenceList`
object. This list computes the covalent energy terms, based on the
result in the ``InternalCoordinateList``. This list also computes the
derivatives of the energy terms towards the internal coordinates.
The computation of the covalent energy is the so-called `forward code
path`, which consists of running through steps 1, 2 and 3, in that order.
The derivatives of the energy are computed in the so-called `backward
code path`, which consists of taking steps 1, 2 and 3 in reverse order.
This basic idea of back-propagation for the computation of derivatives
comes from the field of neural networks. More details can be found in the
chapter, :ref:`dg_sec_backprop`.
'''
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 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 _internal_compute(self, gpos, vtens):
with timer.section('Valence'):
self.dlist.forward()
self.iclist.forward()
energy = self.vlist.forward()
if not ((gpos is None) and (vtens is None)):
self.vlist.back()
self.iclist.back()
self.dlist.back(gpos, vtens)
return energy
class ForcePartValence(ForcePart):
'''The covalent part of a force-field model.
The covalent force field is implemented in a three-layer approach,
similar to the implementation of a neural network:
(0. Optional, not used by default. A layer that computes centers of mass for groups
of atoms.)
1. The first layer consists of a :class:`yaff.pes.dlist.DeltaList` object
that computes all the relative vectors needed for the internal
coordinates in the covalent energy terms. This list is automatically
built up as energy terms are added with the ``add_term`` method. This
list also takes care of transforming `derivatives of the energy
towards relative vectors` into `derivatives of the energy towards
Cartesian coordinates and the virial tensor`.
2. The second layer consist of a
:class:`yaff.pes.iclist.InternalCoordinateList` object that computes
the internal coordinates, based on the ``DeltaList``. This list is
also automatically built up as energy terms are added. The same list
is also responsible for transforming `derivatives of the energy
towards internal coordinates` into `derivatives of the energy towards
relative vectors`.
3. The third layers consists of a :class:`yaff.pes.vlist.ValenceList`
object. This list computes the covalent energy terms, based on the
result in the ``InternalCoordinateList``. This list also computes the
derivatives of the energy terms towards the internal coordinates.
The computation of the covalent energy is the so-called `forward code
path`, which consists of running through steps 1, 2 and 3, in that order.
The derivatives of the energy are computed in the so-called `backward
code path`, which consists of taking steps 1, 2 and 3 in reverse order.
This basic idea of back-propagation for the computation of derivatives
comes from the field of neural networks. More details can be found in the
chapter, :ref:`dg_sec_backprop`.
'''
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 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 _internal_compute(self, gpos, vtens):
with timer.section('Valence'):
self.dlist.forward()
self.iclist.forward()
energy = self.vlist.forward()
if not ((gpos is None) and (vtens is None)):
self.vlist.back()
self.iclist.back()
self.dlist.back(gpos, vtens)
return energy
