def constrain(self, constraints):
#algorithm 1. Use pre-defined crystal_systems to give hard-coded restraints.
# dps_core.constrainment.s_minimizer uses LBFGS minimizer to adapt
# all 9 components of the orientation matrix. This gives the best-fit
# to the starting matrix (better than algorithm #2), but the disadvantage
# is that it is keyed to the crystal_system descriptors. It is therefore
# not adapted to all small-molecule space groups (monoclinics),
# and will not take into account non-standard settings.
if constraints in [
"triclinic", "monoclinic", 'orthorhombic', 'tetragonal',
"cubic", "rhombohedral", 'hexagonal'
]:
from rstbx.dps_core.constrainment import s_minimizer
S = s_minimizer(self, constraint=constraints)
return S.newOrientation()
#algorithm 2: Tensor_rank_2 symmetrization
# Advantages: constraints are calculated directly from the space
# group, so will account for non-standard settings.
# Disadvantages: drift away from starting orientation is greater than
# for algorithm #1.
from cctbx.sgtbx import space_group
if isinstance(constraints, space_group):
from rstbx.symmetry.constraints import AGconvert
converter = AGconvert()
converter.forward(self)
average_cell = constraints.average_unit_cell(self.unit_cell())
converter.validate_and_setG(
average_cell.reciprocal().metrical_matrix())
return Orientation(converter.back(), basis_type.reciprocal)
def finite_difference_test(orient):
from rstbx.symmetry.constraints import AGconvert as AG
from labelit.symmetry.metricsym.a_g_conversion import pp
from libtbx.tst_utils import approx_equal
adaptor = AG()
adaptor.forward(orient)
grad = g_gradients(adaptor, symred=None)
epsilon = 1.E-10
dAij_dphi = grad.get_all_da()[0]
AGback = AG()
F = []
for x in [-1., 1.]:
AGback.setAngles(adaptor.phi + x * epsilon, adaptor.psi, adaptor.theta)
AGback.G = adaptor.G
F.append(flex.double(AGback.back()))
diff_mat = (F[1] - F[0]) / (2. * epsilon)
rule = "dAij_dphi: Analytical gradient vs. finite difference gradient\n"+\
pp(list(dAij_dphi))+"\n"+\
pp(diff_mat)
if not (approx_equal(dAij_dphi, diff_mat, 1.E-7)): raise Exception(rule)
dAij_dpsi = grad.get_all_da()[1]
AGback = AG()
F = []
for x in [-1., 1.]:
AGback.setAngles(adaptor.phi, adaptor.psi + x * epsilon, adaptor.theta)
AGback.G = adaptor.G
F.append(flex.double(AGback.back()))
diff_mat = (F[1] - F[0]) / (2. * epsilon)
rule = "dAij_dpsi: Analytical gradient vs. finite difference gradient\n"+\
pp(list(dAij_dpsi))+"\n"+\
pp(diff_mat)
if not (approx_equal(dAij_dpsi, diff_mat, 1.E-7)): raise Exception(rule)
dAij_dtheta = grad.get_all_da()[2]
AGback = AG()
F = []
for x in [-1., 1.]:
AGback.setAngles(adaptor.phi, adaptor.psi, adaptor.theta + x * epsilon)
AGback.G = adaptor.G
F.append(flex.double(AGback.back()))
diff_mat = (F[1] - F[0]) / (2. * epsilon)
rule = "dAij_dtheta: Analytical gradient vs. finite difference gradient\n"+\
pp(list(dAij_dtheta))+"\n"+\
pp(diff_mat)
if not (approx_equal(dAij_dtheta, diff_mat, 1.E-7)): raise Exception(rule)
g0, g1, g2, g3, g4, g5 = adaptor.G
dAij_dg0 = grad.get_all_da()[3]
AGback = AG()
F = []
for x in [-1., 1.]:
AGback.setAngles(adaptor.phi, adaptor.psi, adaptor.theta)
AGback.G = (g0 + x * epsilon, g1, g2, g3, g4, g5)
F.append(flex.double(AGback.back()))
diff_mat = (F[1] - F[0]) / (2. * epsilon)
rule = "dAij_dg0: Analytical gradient vs. finite difference gradient\n"+\
pp(list(dAij_dg0))+"\n"+\
pp(diff_mat)
if not (approx_equal(dAij_dg0, diff_mat, 1.E-7)): raise Exception(rule)
dAij_dg1 = grad.get_all_da()[4]
AGback = AG()
F = []
for x in [-1., 1.]:
AGback.setAngles(adaptor.phi, adaptor.psi, adaptor.theta)
AGback.G = (g0, g1 + x * epsilon, g2, g3, g4, g5)
F.append(flex.double(AGback.back()))
diff_mat = (F[1] - F[0]) / (2. * epsilon)
rule = "dAij_dg1: Analytical gradient vs. finite difference gradient\n"+\
pp(list(dAij_dg1))+"\n"+\
pp(diff_mat)
if not (approx_equal(dAij_dg1, diff_mat, 1.E-7)): raise Exception(rule)
dAij_dg2 = grad.get_all_da()[5]
AGback = AG()
F = []
for x in [-1., 1.]:
AGback.setAngles(adaptor.phi, adaptor.psi, adaptor.theta)
AGback.G = (g0, g1, g2 + x * epsilon, g3, g4, g5)
F.append(flex.double(AGback.back()))
diff_mat = (F[1] - F[0]) / (2. * epsilon)
rule = "dAij_dg2: Analytical gradient vs. finite difference gradient\n"+\
pp(list(dAij_dg2))+"\n"+\
pp(diff_mat)
if not (approx_equal(dAij_dg2, diff_mat, 1.E-6)): raise Exception(rule)
dAij_dg3 = grad.get_all_da()[6]
AGback = AG()
F = []
for x in [-1., 1.]:
AGback.setAngles(adaptor.phi, adaptor.psi, adaptor.theta)
AGback.G = (g0, g1, g2, g3 + x * epsilon, g4, g5)
F.append(flex.double(AGback.back()))
diff_mat = (F[1] - F[0]) / (2. * epsilon)
rule = "dAij_dg3: Analytical gradient vs. finite difference gradient\n"+\
pp(list(dAij_dg3))+"\n"+\
pp(diff_mat)
if not (approx_equal(dAij_dg3, diff_mat, 1.E-7)): raise Exception(rule)
dAij_dg4 = grad.get_all_da()[7]
AGback = AG()
F = []
for x in [-1., 1.]:
AGback.setAngles(adaptor.phi, adaptor.psi, adaptor.theta)
AGback.G = (g0, g1, g2, g3, g4 + x * epsilon, g5)
F.append(flex.double(AGback.back()))
diff_mat = (F[1] - F[0]) / (2. * epsilon)
rule = "dAij_dg4: Analytical gradient vs. finite difference gradient\n"+\
pp(list(dAij_dg4))+"\n"+\
pp(diff_mat)
if not (approx_equal(dAij_dg4, diff_mat, 1.E-7)): raise Exception(rule)
dAij_dg5 = grad.get_all_da()[8]
AGback = AG()
F = []
for x in [-1., 1.]:
AGback.setAngles(adaptor.phi, adaptor.psi, adaptor.theta)
AGback.G = (g0, g1, g2, g3, g4, g5 + x * epsilon)
F.append(flex.double(AGback.back()))
diff_mat = (F[1] - F[0]) / (2. * epsilon)
rule = "dAij_dg5: Analytical gradient vs. finite difference gradient\n"+\
pp(list(dAij_dg5))+"\n"+\
pp(diff_mat)
if not (approx_equal(dAij_dg5, diff_mat, 1.E-6)): raise Exception(rule)
