def make_result(self, coords, energy):
from pygmin.optimize import Result
res = Result()
res.coords = coords
res.energy = energy
res.eigenval = 1.
res.eigenvec = coords.copy()
return res
def cg(coords, pot, iprint=-1, tol=1e-3, nsteps=5000, **kwargs):
"""
a wrapper function for conjugate gradient routine in scipy
"""
if not hasattr(pot, "getEnergyGradient"):
# for compatibility with old quenchers.
# assume pot is a getEnergyGradient function
pot = _getEnergyGradientWrapper(pot)
import scipy.optimize
ret = scipy.optimize.fmin_cg(pot.getEnergy,
coords,
pot.getGradient,
gtol=tol,
full_output=True,
disp=iprint > 0,
maxiter=nsteps,
**kwargs)
res = Result()
res.coords = ret[0]
#e = ret[1]
res.nfev = ret[2]
res.nfev += ret[3] #calls to gradient
res.success = True
warnflag = ret[4]
if warnflag > 0:
# print "warning: problem with quench: ",
res.success = False
if warnflag == 1:
res.message = "Maximum number of iterations exceeded"
if warnflag == 2:
print "Gradient and/or function calls not changing"
res.energy, res.grad = pot.getEnergyGradient(res.coords)
g = res.grad
res.rms = np.linalg.norm(g) / np.sqrt(len(g))
return res
def bfgs_scipy(coords, pot, iprint=-1, tol=1e-3, nsteps=5000, **kwargs):
"""
a wrapper function for the scipy BFGS algorithm
"""
if not hasattr(pot, "getEnergyGradient"):
# for compatibility with old quenchers.
# assume pot is a getEnergyGradient function
pot = _getEnergyGradientWrapper(pot)
import scipy.optimize
ret = scipy.optimize.fmin_bfgs(pot.getEnergy,
coords,
fprime=pot.getGradient,
gtol=tol,
full_output=True,
disp=iprint > 0,
maxiter=nsteps,
**kwargs)
res = Result()
res.coords = ret[0]
res.energy = ret[1]
res.grad = ret[2]
res.rms = np.linalg.norm(res.grad) / np.sqrt(len(res.grad))
res.nfev = ret[4] + ret[5]
res.nsteps = res.nfev # not correct, but no better information
res.success = np.max(np.abs(res.grad)) < tol
return res
def cg(coords, pot, iprint=-1, tol=1e-3, nsteps=5000, **kwargs):
"""
a wrapper function for conjugate gradient routine in scipy
"""
if not hasattr(pot, "getEnergyGradient"):
# for compatibility with old quenchers.
# assume pot is a getEnergyGradient function
pot = _getEnergyGradientWrapper(pot)
import scipy.optimize
ret = scipy.optimize.fmin_cg(pot.getEnergy, coords, pot.getGradient,
gtol=tol, full_output=True, disp=iprint>0,
maxiter=nsteps, **kwargs)
res = Result()
res.coords = ret[0]
#e = ret[1]
res.nfev = ret[2]
res.nfev += ret[3] #calls to gradient
res.success = True
warnflag = ret[4]
if warnflag > 0:
# print "warning: problem with quench: ",
res.success = False
if warnflag == 1:
res.message = "Maximum number of iterations exceeded"
if warnflag == 2:
print "Gradient and/or function calls not changing"
res.energy, res.grad = pot.getEnergyGradient(res.coords)
g = res.grad
res.rms = np.linalg.norm(g)/np.sqrt(len(g))
return res
def run(self, fmax=1e-3, steps=100000):
"""Run structure optimization algorithm.
This method will return when the forces on all individual
atoms are less than *fmax* or when the number of steps exceeds
*steps*."""
self.fmax = fmax
step = 0
res = Result()
res.success = False
while step < steps:
E, f = self.potential.getEnergyGradient(self.coords)
#self.call_observers()
#print E
if self.alternate_stop_criterion is None:
i_am_done = self.converged(f)
else:
i_am_done = self.alternate_stop_criterion(energy=E,
gradient=f,
tol=self.fmax)
if i_am_done:
res.success = True
break
self.step(-f)
self.nsteps += 1
rms = np.linalg.norm(f) / np.sqrt(len(f))
if self.iprint > 0:
if step % self.iprint == 0:
self.logger.info("fire: %s E %s rms %s", step, E, rms)
for event in self.events:
event(coords=self.coords, energy=E, rms=rms)
step += 1
res.nsteps = step
res.nfev = step
res.coords = self.coords
res.energy = E
res.grad = -f
res.rms = np.linalg.norm(res.grad) / np.sqrt(len(res.grad))
self.result = res
return res
def read_minimum(self, fin):
"""
the minima and transition states are stored as
1 line with energy
1 line with point group info
natoms lines with eigenvalues
natoms lines with coords
"""
res = Result()
#read energy
line = self.get_next_line()
# print line
res.energy = float(line.split()[0])
#ignore the line with the point group
line = self.get_next_line()
res.eigenvalues = self.read_coords(fin)
res.coords = self.read_coords(fin)
return res
def read_minimum(self, fin):
"""
the minima and transition states are stored as
1 line with energy
1 line with point group info
natoms lines with eigenvalues
natoms lines with coords
"""
res = Result()
#read energy
line = self.get_next_line()
# print line
res.energy = float(line.split()[0])
#ignore the line with the point group
line = self.get_next_line()
res.eigenvalues = self.read_coords(fin)
res.coords = self.read_coords(fin)
return res
def make_result(self, coords, energy):
from pygmin.optimize import Result
res = Result()
res.coords = coords
res.energy = energy
res.eigenval = 1.
res.eigenvec = coords.copy()
return res
def steepest_descent(x0,
pot,
iprint=-1,
dx=1e-4,
nsteps=100000,
tol=1e-3,
maxstep=-1.,
event=None):
if not hasattr(pot, "getEnergyGradient"):
# for compatibility with old quenchers.
# assume pot is a getEnergyGradient function
pot = _getEnergyGradientWrapper(pot)
N = len(x0)
x = x0.copy()
E, V = pot.getEnergyGradient(x)
funcalls = 1
for k in xrange(nsteps):
stp = -V * dx
if maxstep > 0:
stpsize = np.max(np.abs(V))
if stpsize > maxstep:
stp *= maxstep / stpsize
x += stp
E, V = pot.getEnergyGradient(x)
funcalls += 1
rms = np.linalg.norm(V) / np.sqrt(N)
if iprint > 0:
if funcalls % iprint == 0:
print "step %8d energy %20.12g rms gradient %20.12g" % (
funcalls, E, rms)
if event != None:
event(E, x, rms)
if rms < tol:
break
res = Result()
res.coords = x
res.energy = E
res.rms = rms
res.grad = V
res.nfev = funcalls
res.nsteps = k
res.success = res.rms <= tol
return res
def run(self, fmax=1e-3, steps=100000):
"""Run structure optimization algorithm.
This method will return when the forces on all individual
atoms are less than *fmax* or when the number of steps exceeds
*steps*."""
self.fmax = fmax
step = 0
res = Result()
res.success = False
while step < steps:
E, f = self.potential.getEnergyGradient(self.coords)
#self.call_observers()
#print E
if self.alternate_stop_criterion is None:
i_am_done = self.converged(f)
else:
i_am_done = self.alternate_stop_criterion(energy=E, gradient=f,
tol=self.fmax)
if i_am_done:
res.success = True
break
self.step(-f)
self.nsteps += 1
rms = np.linalg.norm(f)/np.sqrt(len(f))
if self.iprint > 0:
if step % self.iprint == 0:
self.logger.info("fire: %s E %s rms %s", step, E, rms)
for event in self.events:
event(coords=self.coords, energy=E, rms=rms)
step += 1
res.nsteps = step
res.nfev = step
res.coords = self.coords
res.energy = E
res.grad = -f
res.rms = np.linalg.norm(res.grad)/np.sqrt(len(res.grad))
self.result = res
return res
def __init__(
self,
coords,
potential,
takeStep,
storage=None,
event_after_step=[],
acceptTest=None,
temperature=1.0,
confCheck=[],
outstream=sys.stdout,
store_initial=True,
iprint=1,
):
#note: make a local copy of lists of events so that an inputted list is not modified.
self.coords = np.copy(coords)
self.storage = storage
self.potential = potential
self.takeStep = takeStep
self.event_after_step = copy.copy(event_after_step) #not deepcopy
self.temperature = temperature
self.naccepted = 0
self.outstream = outstream
self.printfrq = iprint #controls how often printing is done
self.confCheck = confCheck
if acceptTest:
self.acceptTest = acceptTest
else:
self.acceptTest = metropolis.Metropolis(self.temperature)
self.stepnum = 0
#########################################################################
#store intial structure
#########################################################################
energy = self.potential.getEnergy(self.coords)
if (self.storage and store_initial):
self.storage(energy, self.coords)
self.markovE = energy
self.result = Result()
self.result.energy = self.markovE
self.result.coords = self.coords.copy()
def bfgs_scipy(coords, pot, iprint=-1, tol=1e-3, nsteps=5000, **kwargs):
"""
a wrapper function for the scipy BFGS algorithm
"""
if not hasattr(pot, "getEnergyGradient"):
# for compatibility with old quenchers.
# assume pot is a getEnergyGradient function
pot = _getEnergyGradientWrapper(pot)
import scipy.optimize
ret = scipy.optimize.fmin_bfgs(pot.getEnergy, coords, fprime=pot.getGradient,
gtol=tol, full_output=True, disp=iprint>0,
maxiter=nsteps, **kwargs)
res = Result()
res.coords = ret[0]
res.energy = ret[1]
res.grad = ret[2]
res.rms = np.linalg.norm(res.grad) / np.sqrt(len(res.grad))
res.nfev = ret[4] + ret[5]
res.nsteps = res.nfev # not correct, but no better information
res.success = np.max(np.abs(res.grad)) < tol
return res
def steepest_descent(x0, pot, iprint=-1, dx=1e-4, nsteps=100000,
tol=1e-3, maxstep=-1., event=None):
if not hasattr(pot, "getEnergyGradient"):
# for compatibility with old quenchers.
# assume pot is a getEnergyGradient function
pot = _getEnergyGradientWrapper(pot)
N = len(x0)
x=x0.copy()
E, V = pot.getEnergyGradient(x)
funcalls = 1
for k in xrange(nsteps):
stp = -V * dx
if maxstep > 0:
stpsize = np.max(np.abs(V))
if stpsize > maxstep:
stp *= maxstep / stpsize
x += stp
E, V = pot.getEnergyGradient(x)
funcalls += 1
rms = np.linalg.norm(V)/np.sqrt(N)
if iprint > 0:
if funcalls % iprint == 0:
print "step %8d energy %20.12g rms gradient %20.12g" % (funcalls, E, rms)
if event != None:
event(E, x, rms)
if rms < tol:
break
res = Result()
res.coords = x
res.energy = E
res.rms = rms
res.grad = V
res.nfev = funcalls
res.nsteps = k
res.success = res.rms <= tol
return res
def __init__(self,
pot,
mindist,
tsSearchParams=dict(),
verbosity=1,
NEBparams=dict(),
nrefine_max=100,
reoptimize_climbing=0,
pushoff_params=dict(),
create_neb=NEBDriver):
self.pot = pot
self.mindist = mindist
self.tsSearchParams = tsSearchParams
self.verbosity = int(verbosity)
self.nrefine_max = nrefine_max
self.NEBparams = NEBparams
self.reoptimize_climbing = reoptimize_climbing
self.pushoff_params = pushoff_params
self.res = Result()
self.res.new_transition_states = []
self.create_neb = create_neb
def optimize(self, quenchRoutine=None,
**kwargs):
"""
Optimize the band
Notes
-----
the potential for the NEB optimization is not Hamiltonian. This
means that there is no meaningful energy associated with the
potential. Therefore, during the optimization, we can do gradient
following, but we can't rely on the energy for, e.g. determining
step size, which is the default behavior for many optimizers. This
can be worked around by choosing a small step size and a large
maxErise, or by using an optimizer that uses only gradients.
scipy.lbfgs_b seems to work with NEB pretty well, but lbfgs_py and
mylbfgs tend to fail. If you must use one of those try, e.g.
maxErise = 1., maxstep=0.01, tol=1e-2
:quenchRoutine: quench algorithm to use for optimization.
:quenchParams: parameters for the quench """
if quenchRoutine is None:
if self.quenchRoutine is None:
quenchRoutine = mylbfgs
else:
quenchRoutine = self.quenchRoutine
#combine default and passed params. passed params will overwrite default
quenchParams = dict([("nsteps", 300)] +
self.quenchParams.items() +
kwargs.items())
if quenchParams.has_key("iprint"):
self.iprint = quenchParams["iprint"]
if not quenchParams.has_key("logger"):
quenchParams["logger"] = logging.getLogger("pygmin.connect.neb.quench")
if self.use_minimizer_callback:
quenchParams["events"]=[self._step]
self.step = 0
qres = quenchRoutine(
self.active.reshape(self.active.size), self,
**quenchParams)
# if isinstance(qres, tuple): # for compatability with old and new quenchers
# qres = qres[4]
self.active[:,:] = qres.coords.reshape(self.active.shape)
if self.copy_potential:
for i in xrange(0,self.nimages):
pot = self.potential_list[i]
self.energies[i] = pot.getEnergy(self.coords[i,:])
else:
for i in xrange(0,self.nimages):
self.energies[i] = self.potential.getEnergy(self.coords[i,:])
res = Result()
res.path = self.coords
res.nsteps = qres.nsteps
res.energy = self.energies
res.rms = qres.rms
res.success = False
if qres.rms < quenchParams["tol"]:
res.success = True
return res
def run(self):
"""
the main loop of the algorithm
"""
res = Result()
res.message = []
tol = self.tol
iprint = self.iprint
nsteps = self.nsteps
#iprint =40
X = self.X
sqrtN = np.sqrt(self.N)
i = 1
self.funcalls += 1
e, G = self.pot.getEnergyGradient(X)
res.success = False
while i < nsteps:
stp = self.getStep(X, G)
try:
X, e, G = self.adjustStepSize(X, e, G, stp)
except LineSearchError:
print "Warning: problem with adjustStepSize, ending quench"
rms = np.linalg.norm(G) / sqrtN
print " on failure: quench step", i, e, rms, self.funcalls
res.message.append( "problem with adjustStepSize" )
break
#e, G = self.pot.getEnergyGradient(X)
rms = np.linalg.norm(G) / sqrtN
if iprint > 0:
if i % iprint == 0:
print "lbfgs:", i, e, rms, self.funcalls, self.stepsize
for event in self.events:
event(coords=X, energy=e, rms=rms)
if self.alternate_stop_criterion is None:
i_am_done = rms < self.tol
else:
i_am_done = self.alternate_stop_criterion(energy=e, gradient=G,
tol=self.tol)
if i_am_done:
res.success = True
break
i += 1
res.nsteps = i
res.nfev = self.funcalls
res.coords = X
res.energy = e
res.rms = rms
res.grad = G
res.H0 = self.H0
return res
def run(self):
"""
the main loop of the algorithm
"""
res = Result()
res.message = []
tol = self.tol
iprint = self.iprint
nsteps = self.nsteps
#iprint =40
X = self.X
sqrtN = np.sqrt(self.N)
i = 1
self.funcalls += 1
e, G = self.pot.getEnergyGradient(X)
rms = np.linalg.norm(G) / sqrtN
res.success = False
while i < nsteps:
stp = self.getStep(X, G)
try:
X, e, G = self.adjustStepSize(X, e, G, stp)
except LineSearchError:
self.logger.error("problem with adjustStepSize, ending quench")
rms = np.linalg.norm(G) / sqrtN
self.logger.error(" on failure: quench step %s %s %s %s", i, e, rms, self.funcalls)
res.message.append( "problem with adjustStepSize" )
break
#e, G = self.pot.getEnergyGradient(X)
rms = np.linalg.norm(G) / sqrtN
if iprint > 0:
if i % iprint == 0:
self.logger.info("lbfgs: %s %s %s %s %s %s %s %s %s", i, "E", e,
"rms", rms, "funcalls", self.funcalls, "stepsize", self.stepsize)
for event in self.events:
event(coords=X, energy=e, rms=rms)
if self.alternate_stop_criterion is None:
i_am_done = rms < self.tol
else:
i_am_done = self.alternate_stop_criterion(energy=e, gradient=G,
tol=self.tol)
if i_am_done:
res.success = True
break
i += 1
res.nsteps = i
res.nfev = self.funcalls
res.coords = X
res.energy = e
res.rms = rms
res.grad = G
res.H0 = self.H0
return res
def run(self):
"""The main loop of the algorithm"""
coords = np.copy(self.coords)
res = Result() # return object
res.message = []
for i in xrange(self.nsteps):
#get the lowest eigenvalue and eigenvector
self.overlap = self._getLowestEigenVector(coords, i)
overlap = self.overlap
#check to make sure the eigenvector is ok
if i == 0 or self.eigenval <= 0:
self._saveState(coords)
self.reduce_step = 0
else:
self.nnegative += 1
if self.nnegative > self.nnegative_max:
print "warning: negative eigenvalue found too many times. ending", self.nnegative
res.message.append( "negative eigenvalue found too many times %d" % self.nnegative )
break
if self.verbosity > 2:
print "the eigenvalue turned positive.", self.eigenval, "Resetting last good values and taking smaller steps"
coords = self._resetState(coords)
self.reduce_step += 1
#step uphill along the direction of the lowest eigenvector
coords = self._stepUphill(coords)
if False:
#maybe we want to update the lowest eigenvector now that we've moved?
#david thinks this is a bad idea
overlap = self._getLowestEigenVector(coords, i)
#minimize the coordinates in the space perpendicular to the lowest eigenvector
coords, tangentrms = self._minimizeTangentSpace(coords)
#check if we are done and print some stuff
E, grad = self.pot.getEnergyGradient(coords)
rms = np.linalg.norm(grad) * self.rmsnorm
gradpar = np.dot(grad, self.eigenvec) / np.linalg.norm(self.eigenvec)
if self.iprint > 0:
if (i+1) % self.iprint == 0:
ostring = "findTransitionState: %3d E %g rms %g eigenvalue %g rms perp %g grad par %g overlap %g" % (
i, E, rms, self.eigenval, tangentrms, gradpar, overlap)
extra = " Evec search: %d rms %g" % (self.leig_result.nfev, self.leig_result.rms)
extra += " Tverse search: %d step %g" % (self.tangent_result.nfev,
self.tangent_move_step)
extra += " Uphill step:%g" % (self.uphill_step_size,)
print ostring, extra
if callable(self.event):
self.event(E, coords, rms)
if rms < self.tol:
break
if self.nfail >= self.nfail_max:
print "stopping findTransitionState. too many failures in eigenvector search", self.nfail
res.message.append( "too many failures in eigenvector search %d" % self.nfail )
break
if i == 0 and self.eigenval > 0.:
print "WARNING *** initial eigenvalue is positive - increase NEB spring constant?"
if self.demand_initial_negative_vec:
print " aborting transition state search"
res.message.append( "initial eigenvalue is positive %f" % self.eigenval )
break
#done. do one last eigenvector search because coords may have changed
self._getLowestEigenVector(coords, i)
#done, print some data
print "findTransitionState done:", i, E, rms, "eigenvalue", self.eigenval
success = True
#check if results make sense
if self.eigenval >= 0.:
if self.verbosity > 2:
print "warning: transition state is ending with positive eigenvalue", self.eigenval
success = False
if rms > self.tol:
if self.verbosity > 2:
print "warning: transition state search appears to have failed: rms", rms
success = False
if i >= self.nsteps:
res.message.append( "maximum iterations reached %d" % i )
#return results
res.coords = coords
res.energy = E
res.eigenval = self.eigenval
res.eigenvec = self.eigenvec
res.grad = grad
res.rms = rms
res.nsteps = i
res.success = success
return res
def lbfgs_scipy(coords, pot, iprint=-1, tol=1e-3, nsteps=15000):
"""
a wrapper function for lbfgs routine in scipy
.. warn::
the scipy version of lbfgs uses linesearch based only on energy
which can make the minimization stop early. When the step size
is so small that the energy doesn't change to within machine precision (times the
parameter `factr`) the routine declares success and stops. This sounds fine, but
if the gradient is analytical the gradient can still be not converged. This is
because in the vicinity of the minimum the gradient changes much more rapidly then
the energy. Thus we want to make factr as small as possible. Unfortunately,
if we make it too small the routine realizes that the linesearch routine
isn't working and declares failure and exits.
So long story short, if your tolerance is very small (< 1e-6) this routine
will probably stop before truly reaching that tolerance. If you reduce `factr`
too much to mitigate this lbfgs will stop anyway, but declare failure misleadingly.
"""
if not hasattr(pot, "getEnergyGradient"):
# for compatibility with old quenchers.
# assume pot is a getEnergyGradient function
pot = _getEnergyGradientWrapper(pot)
import scipy.optimize
res = Result()
res.coords, res.energy, dictionary = scipy.optimize.fmin_l_bfgs_b(
pot.getEnergyGradient,
coords,
iprint=iprint,
pgtol=tol,
maxfun=nsteps,
factr=10.)
res.grad = dictionary["grad"]
res.nfev = dictionary["funcalls"]
warnflag = dictionary['warnflag']
#res.nsteps = dictionary['nit'] # new in scipy version 0.12
res.nsteps = res.nfev
res.message = dictionary['task']
res.success = True
if warnflag > 0:
print "warning: problem with quench: ",
res.success = False
if warnflag == 1:
res.message = "too many function evaluations"
else:
res.message = str(dictionary['task'])
print res.message
#note: if the linesearch fails the lbfgs may fail without setting warnflag. Check
#tolerance exactly
if False:
if res.success:
maxV = np.max(np.abs(res.grad))
if maxV > tol:
print "warning: gradient seems too large", maxV, "tol =", tol, ". This is a known, but not understood issue of scipy_lbfgs"
print res.message
res.rms = res.grad.std()
return res
def run(self):
"""The main loop of the algorithm"""
coords = np.copy(self.coords)
res = Result() # return object
res.message = []
for i in xrange(self.nsteps):
# get the lowest eigenvalue and eigenvector
self.overlap = self._getLowestEigenVector(coords, i)
overlap = self.overlap
# check to make sure the eigenvector is ok
if i == 0 or self.eigenval <= 0:
self._saveState(coords)
self.reduce_step = 0
else:
self.npositive += 1
if self.npositive > self.npositive_max:
logger.warning(
"positive eigenvalue found too many times. ending %s",
self.npositive)
res.message.append(
"positive eigenvalue found too many times %d" %
self.npositive)
break
if self.verbosity > 2:
logger.info(
"the eigenvalue turned positive.", self.eigenval,
"Resetting last good values and taking smaller steps")
coords = self._resetState(coords)
self.reduce_step += 1
# step uphill along the direction of the lowest eigenvector
coords = self._stepUphill(coords)
if False:
# maybe we want to update the lowest eigenvector now that we've moved?
# david thinks this is a bad idea
overlap = self._getLowestEigenVector(coords, i)
# minimize the coordinates in the space perpendicular to the lowest eigenvector
coords, tangentrms = self._minimizeTangentSpace(coords)
# check if we are done and print some stuff
E, grad = self.pot.getEnergyGradient(coords)
rms = np.linalg.norm(grad) * self.rmsnorm
gradpar = np.dot(grad, self.eigenvec) / np.linalg.norm(
self.eigenvec)
if self.iprint > 0:
if (i + 1) % self.iprint == 0:
ostring = "findTS: %3d E %9g rms %8g eigenvalue %9g rms perp %8g grad par %9g overlap %g" % (
i, E, rms, self.eigenval, tangentrms, gradpar, overlap)
extra = " Evec search: %d rms %g" % (
self.leig_result.nfev, self.leig_result.rms)
extra += " Tverse search: %d step %g" % (
self.tangent_result.nfev, self.tangent_move_step)
extra += " Uphill step:%g" % (self.uphill_step_size, )
logger.info("%s %s", ostring, extra)
if callable(self.event):
self.event(energy=E,
coords=coords,
rms=rms,
eigenval=self.eigenval,
stepnum=i)
if rms < self.tol:
break
if self.nfail >= self.nfail_max:
logger.warning(
"stopping findTransitionState. too many failures in eigenvector search %s",
self.nfail)
res.message.append(
"too many failures in eigenvector search %d" % self.nfail)
break
if i == 0 and self.eigenval > 0.:
logger.warning(
"initial eigenvalue is positive - increase NEB spring constant?"
)
if self.demand_initial_negative_vec:
logger.warning(
" aborting transition state search")
res.message.append("initial eigenvalue is positive %f" %
self.eigenval)
break
# done. do one last eigenvector search because coords may have changed
self._getLowestEigenVector(coords, i)
# print some data
logger.info("findTransitionState done: %s %s %s %s %s", i, E, rms,
"eigenvalue", self.eigenval)
success = True
# check if results make sense
if self.eigenval >= 0.:
if self.verbosity > 2:
logger.info(
"warning: transition state is ending with positive eigenvalue %s",
self.eigenval)
success = False
if rms > self.tol:
if self.verbosity > 2:
logger.info(
"warning: transition state search appears to have failed: rms %s",
rms)
success = False
if i >= self.nsteps:
res.message.append("maximum iterations reached %d" % i)
#return results
res.coords = coords
res.energy = E
res.eigenval = self.eigenval
res.eigenvec = self.eigenvec
res.grad = grad
res.rms = rms
res.nsteps = i
res.success = success
return res
def optimize(self, quenchRoutine=None, **kwargs):
"""
Optimize the band
Notes
-----
the potential for the NEB optimization is not Hamiltonian. This
means that there is no meaningful energy associated with the
potential. Therefore, during the optimization, we can do gradient
following, but we can't rely on the energy for, e.g. determining
step size, which is the default behavior for many optimizers. This
can be worked around by choosing a small step size and a large
maxErise, or by using an optimizer that uses only gradients.
scipy.lbfgs_b seems to work with NEB pretty well, but lbfgs_py and
mylbfgs tend to fail. If you must use one of those try, e.g.
maxErise = 1., maxstep=0.01, tol=1e-2
:quenchRoutine: quench algorithm to use for optimization.
:quenchParams: parameters for the quench """
if quenchRoutine is None:
if self.quenchRoutine is None:
quenchRoutine = mylbfgs
else:
quenchRoutine = self.quenchRoutine
#combine default and passed params. passed params will overwrite default
quenchParams = dict([("nsteps", 300)] + self.quenchParams.items() +
kwargs.items())
if quenchParams.has_key("iprint"):
self.iprint = quenchParams["iprint"]
if not quenchParams.has_key("logger"):
quenchParams["logger"] = logging.getLogger(
"pygmin.connect.neb.quench")
if self.use_minimizer_callback:
quenchParams["events"] = [self._step]
self.step = 0
qres = quenchRoutine(self.active.reshape(self.active.size), self,
**quenchParams)
# if isinstance(qres, tuple): # for compatability with old and new quenchers
# qres = qres[4]
self.active[:, :] = qres.coords.reshape(self.active.shape)
if self.copy_potential:
for i in xrange(0, self.nimages):
pot = self.potential_list[i]
self.energies[i] = pot.getEnergy(self.coords[i, :])
else:
for i in xrange(0, self.nimages):
self.energies[i] = self.potential.getEnergy(self.coords[i, :])
res = Result()
res.path = self.coords
res.nsteps = qres.nsteps
res.energy = self.energies
res.rms = qres.rms
res.success = False
if qres.rms < quenchParams["tol"]:
res.success = True
return res
def lbfgs_scipy(coords, pot, iprint=-1, tol=1e-3, nsteps=15000):
"""
a wrapper function for lbfgs routine in scipy
.. warn::
the scipy version of lbfgs uses linesearch based only on energy
which can make the minimization stop early. When the step size
is so small that the energy doesn't change to within machine precision (times the
parameter `factr`) the routine declares success and stops. This sounds fine, but
if the gradient is analytical the gradient can still be not converged. This is
because in the vicinity of the minimum the gradient changes much more rapidly then
the energy. Thus we want to make factr as small as possible. Unfortunately,
if we make it too small the routine realizes that the linesearch routine
isn't working and declares failure and exits.
So long story short, if your tolerance is very small (< 1e-6) this routine
will probably stop before truly reaching that tolerance. If you reduce `factr`
too much to mitigate this lbfgs will stop anyway, but declare failure misleadingly.
"""
if not hasattr(pot, "getEnergyGradient"):
# for compatibility with old quenchers.
# assume pot is a getEnergyGradient function
pot = _getEnergyGradientWrapper(pot)
import scipy.optimize
res = Result()
res.coords, res.energy, dictionary = scipy.optimize.fmin_l_bfgs_b(pot.getEnergyGradient,
coords, iprint=iprint, pgtol=tol, maxfun=nsteps, factr=10.)
res.grad = dictionary["grad"]
res.nfev = dictionary["funcalls"]
warnflag = dictionary['warnflag']
#res.nsteps = dictionary['nit'] # new in scipy version 0.12
res.nsteps = res.nfev
res.message = dictionary['task']
res.success = True
if warnflag > 0:
print "warning: problem with quench: ",
res.success = False
if warnflag == 1:
res.message = "too many function evaluations"
else:
res.message = str(dictionary['task'])
print res.message
#note: if the linesearch fails the lbfgs may fail without setting warnflag. Check
#tolerance exactly
if False:
if res.success:
maxV = np.max( np.abs(res.grad) )
if maxV > tol:
print "warning: gradient seems too large", maxV, "tol =", tol, ". This is a known, but not understood issue of scipy_lbfgs"
print res.message
res.rms = res.grad.std()
return res