def test_gradient_functional_deriv():
fn_fchk = context.get_fn('test/n2_hfs_sto3g.fchk')
mol = IOData.from_file(fn_fchk)
obasis = mol.obasis
dm_full = mol.get_dm_full()
rtf = ExpRTransform(1e-3, 1e1, 10)
rgrid = RadialGrid(rtf)
grid = BeckeMolGrid(mol.coordinates, mol.numbers, mol.pseudo_numbers, (rgrid, 6), random_rotate=False, mode='keep')
pot = grid.points[:,2].copy()
def fun(x):
dm_full._array[:] = x.reshape(obasis.nbasis, -1)
f = obasis.compute_grid_gradient_dm(dm_full, grid.points)
tmp = (f*f).sum(axis=1)
return 0.5*grid.integrate(tmp, pot)
def fun_deriv(x):
dm_full._array[:] = x.reshape(obasis.nbasis, -1)
result = dm_full.new()
tmp = obasis.compute_grid_gradient_dm(dm_full, grid.points)
tmp *= pot.reshape(-1,1)
obasis.compute_grid_gradient_fock(grid.points, grid.weights, tmp, result)
return result._array.ravel()
eps = 1e-4
x = dm_full._array.copy().ravel()
dxs = []
for i in xrange(100):
tmp = np.random.uniform(-eps, +eps, x.shape)*x
tmp = (tmp+tmp.T)/2
dxs.append(tmp)
from horton.test.common import check_delta
check_delta(fun, fun_deriv, x, dxs)
def test_kinetic_functional_deriv():
fn_fchk = context.get_fn('test/n2_hfs_sto3g.fchk')
mol = IOData.from_file(fn_fchk)
obasis = mol.obasis
dm_full = mol.get_dm_full()
grid = BeckeMolGrid(mol.coordinates,
mol.numbers,
mol.pseudo_numbers,
random_rotate=False,
mode='keep')
pot = grid.points[:, 2]
def fun(x):
dm_full._array[:] = x.reshape(obasis.nbasis, -1)
f = obasis.compute_grid_kinetic_dm(dm_full, grid.points)
return 0.5 * grid.integrate(f, f, pot)
def fun_deriv(x):
dm_full._array[:] = x.reshape(obasis.nbasis, -1)
result = dm_full.new()
tmp = obasis.compute_grid_kinetic_dm(dm_full, grid.points) * pot
obasis.compute_grid_kinetic_fock(grid.points, grid.weights, tmp,
result)
return result._array.ravel()
eps = 1e-4
x = dm_full._array.copy().ravel()
dxs = []
for i in xrange(100):
tmp = np.random.uniform(-eps, +eps, x.shape) * x
dxs.append(tmp)
from horton.test.common import check_delta
check_delta(fun, fun_deriv, x, dxs)
def test_kinetic_functional_deriv():
fn_fchk = context.get_fn('test/n2_hfs_sto3g.fchk')
mol = IOData.from_file(fn_fchk)
obasis = mol.obasis
dm_full = mol.get_dm_full()
grid = BeckeMolGrid(mol.coordinates, mol.numbers, mol.pseudo_numbers, random_rotate=False, mode='keep')
pot = grid.points[:,2]
def fun(x):
dm_full._array[:] = x.reshape(obasis.nbasis, -1)
f = obasis.compute_grid_kinetic_dm(dm_full, grid.points)
return 0.5*grid.integrate(f, f, pot)
def fun_deriv(x):
dm_full._array[:] = x.reshape(obasis.nbasis, -1)
result = dm_full.new()
tmp = obasis.compute_grid_kinetic_dm(dm_full, grid.points)*pot
obasis.compute_grid_kinetic_fock(grid.points, grid.weights, tmp, result)
return result._array.ravel()
eps = 1e-4
x = dm_full._array.copy().ravel()
dxs = []
for i in xrange(100):
tmp = np.random.uniform(-eps, +eps, x.shape)*x
dxs.append(tmp)
from horton.test.common import check_delta
check_delta(fun, fun_deriv, x, dxs)
def check_density_hessian(obasis, dm_full, point, eps):
"""Finite difference checker for density Hessian.
Parameters
----------
obasis : GOBasis
The basis set to use in the test.
dm_full : DenseTwoIndex
The spin-summed density matrix.
point : np.ndarray, shape=(3, ), dtype=float
The reference point for check_delta.
eps : float
The magnitude of the displacements.
"""
def fun(p):
"""Evaluate the density gradient at point p."""
return obasis.compute_grid_gradient_dm(dm_full, np.array([p]))[0]
def fun_deriv(p):
"""Evaluate the density Hessian at point p."""
row = obasis.compute_grid_hessian_dm(dm_full, np.array([p]))[0]
result = np.zeros((3, 3), float)
result[0, 0] = row[0]
result[0, 1] = row[1]
result[1, 0] = row[1]
result[0, 2] = row[2]
result[2, 0] = row[2]
result[1, 1] = row[3]
result[1, 2] = row[4]
result[2, 1] = row[4]
result[2, 2] = row[5]
return result
dpoints = np.random.uniform(-eps, eps, (100, 3))
check_delta(fun, fun_deriv, point, dpoints)
def check_density_hessian(obasis, dm_full, point, eps):
"""Finite difference checker for density Hessian.
Parameters
----------
obasis : GOBasis
The basis set to use in the test.
dm_full : DenseTwoIndex
The spin-summed density matrix.
point : np.ndarray, shape=(3, ), dtype=float
The reference point for check_delta.
eps : float
The magnitude of the displacements.
"""
def fun(p):
"""Evaluate the density gradient at point p."""
return obasis.compute_grid_gradient_dm(dm_full, np.array([p]))[0]
def fun_deriv(p):
"""Evaluate the density Hessian at point p."""
row = obasis.compute_grid_hessian_dm(dm_full, np.array([p]))[0]
result = np.zeros((3, 3), float)
result[0, 0] = row[0]
result[0, 1] = row[1]
result[1, 0] = row[1]
result[0, 2] = row[2]
result[2, 0] = row[2]
result[1, 1] = row[3]
result[1, 2] = row[4]
result[2, 1] = row[4]
result[2, 2] = row[5]
return result
dpoints = np.random.uniform(-eps, eps, (100, 3))
check_delta(fun, fun_deriv, point, dpoints)
def test_gradient_functional_deriv():
fn_fchk = context.get_fn('test/n2_hfs_sto3g.fchk')
sys = System.from_file(fn_fchk)
rtf = ExpRTransform(1e-3, 1e1, 10)
rgrid = RadialGrid(rtf)
grid = BeckeMolGrid(sys, (rgrid, 6), random_rotate=False, mode='keep')
pot = grid.points[:,2].copy()
def fun(x):
sys.wfn.dm_full._array[:] = x.reshape(sys.obasis.nbasis, -1)
f = sys.compute_grid_gradient(grid.points)
tmp = (f*f).sum(axis=1)
return 0.5*grid.integrate(tmp, pot)
def fun_deriv(x):
sys.wfn.dm_full._array[:] = x.reshape(sys.obasis.nbasis, -1)
result = sys.wfn.dm_full.copy()
result.clear()
tmp = sys.compute_grid_gradient(grid.points)
tmp *= pot.reshape(-1,1)
sys.compute_grid_gradient_fock(grid.points, grid.weights, tmp, result)
return result._array.ravel()
eps = 1e-4
x = sys.wfn.update_dm('full')._array.copy().ravel()
dxs = []
for i in xrange(100):
tmp = np.random.uniform(-eps, +eps, x.shape)*x
tmp = (tmp+tmp.T)/2
dxs.append(tmp)
from horton.test.common import check_delta
check_delta(fun, fun_deriv, x, dxs)
def check_functional_deriv(fn, comp, dm_method, fock_method):
"""Test consistency of properties on a grid and implementation of Fock build.
Parameters
----------
fn : str
The filename of the molecule to run the test on. It should contain a wavefunction
and it is assumed to be located in the data directory.
comp : int
The component of the potential to be tested. (Potentials may have multiple
components, e.g. in the case of GGA the gradient has different components.)
dm_method : function
Computes from a density matrix the density properties of interest, e.g. density.
This function takes three arguments: obasis, dm_full and points. (See e.g.
``GOBasis.compute_grid_density_dm``)
fock_method : function
Computes from a potential on a grid, the fock operator. This function takes five
arguments: obasis, points, weights, potential and fock. (See e.g.
``GOBasis.compute_grid_density_fock``)
"""
fn_fchk = context.get_fn(fn)
mol = IOData.from_file(fn_fchk)
obasis = mol.obasis
dm_full = mol.get_dm_full()
grid = BeckeMolGrid(mol.coordinates,
mol.numbers,
mol.pseudo_numbers,
'coarse',
random_rotate=False,
mode='keep')
def fun(x):
"""Compute the grid properties from a density matrix."""
dm_full[:] = x.reshape(obasis.nbasis, -1)
f = dm_method(obasis, dm_full, grid.points)
f = f.reshape((grid.size, -1))
return 0.5 * grid.integrate(f[:, comp], f[:, comp])
def fun_deriv(x):
"""Compute a Fock matrix from potential data on a grid."""
dm_full[:] = x.reshape(obasis.nbasis, -1)
tmp = dm_method(obasis, dm_full, grid.points)
if tmp.ndim > 1:
tmp[:, :comp] = 0.0
tmp[:, comp + 1:] = 0.0
fock = fock_method(obasis, grid.points, grid.weights, tmp)
return fock.ravel()
eps = 1e-4
x = dm_full.copy().ravel()
dxs = []
for _irep in xrange(100):
tmp = np.random.uniform(-eps, +eps, x.shape) * x
dxs.append(tmp)
check_delta(fun, fun_deriv, x, dxs)
def test_esp_cost_cube3d_gradient():
# Some parameters
coordinates, numbers, origin, grid_rvecs, shape, pbc, vref, weights = get_random_esp_cost_cube3d_args()
grid = UniformGrid(origin, grid_rvecs, shape, pbc)
cost = ESPCost.from_grid_data(coordinates, grid, vref, weights)
x0 = np.random.uniform(-0.5, 0.5, len(numbers) + 1)
dxs = np.random.uniform(-1e-5, 1e-5, (100, len(numbers) + 1))
check_delta(cost.value, cost.gradient, x0, dxs)
def check_functional_deriv(fn, comp, dm_method, fock_method):
"""Test consistency of properties on a grid and implementation of Fock build.
Parameters
----------
fn : str
The filename of the molecule to run the test on. It should contain a wavefunction
and it is assumed to be located in the data directory.
comp : int
The component of the potential to be tested. (Potentials may have multiple
components, e.g. in the case of GGA the gradient has different components.)
dm_method : function
Computes from a density matrix the density properties of interest, e.g. density.
This function takes three arguments: obasis, dm_full and points. (See e.g.
``GOBasis.compute_grid_density_dm``)
fock_method : function
Computes from a potential on a grid, the fock operator. This function takes five
arguments: obasis, points, weights, potential and fock. (See e.g.
``GOBasis.compute_grid_density_fock``)
"""
fn_fchk = context.get_fn(fn)
mol = IOData.from_file(fn_fchk)
obasis = mol.obasis
dm_full = mol.get_dm_full()
grid = BeckeMolGrid(mol.coordinates, mol.numbers, mol.pseudo_numbers, 'coarse',
random_rotate=False, mode='keep')
def fun(x):
"""Compute the grid properties from a density matrix."""
dm_full._array[:] = x.reshape(obasis.nbasis, -1)
f = dm_method(obasis, dm_full, grid.points)
f = f.reshape((grid.size, -1))
return 0.5*grid.integrate(f[:, comp], f[:, comp])
def fun_deriv(x):
"""Compute a Fock matrix from potential data on a grid."""
dm_full._array[:] = x.reshape(obasis.nbasis, -1)
result = dm_full.new()
tmp = dm_method(obasis, dm_full, grid.points)
orig_shape = tmp.shape
tmp = tmp.reshape((grid.size, -1))
tmp[:, :comp] = 0.0
tmp[:, comp+1:] = 0.0
tmp.shape = orig_shape
fock_method(obasis, grid.points, grid.weights, tmp, result)
return result._array.ravel()
eps = 1e-4
x = dm_full._array.copy().ravel()
dxs = []
for _irep in xrange(100):
tmp = np.random.uniform(-eps, +eps, x.shape)*x
dxs.append(tmp)
check_delta(fun, fun_deriv, x, dxs)
def test_esp_cost_cube3d_gradient():
# Some parameters
coordinates, numbers, origin, grid_rvecs, shape, pbc, vref, weights = \
get_random_esp_cost_cube3d_args()
grid = UniformGrid(origin, grid_rvecs, shape, pbc)
cost = ESPCost.from_grid_data(coordinates, grid, vref, weights)
x0 = np.random.uniform(-0.5, 0.5, len(numbers)+1)
dxs = np.random.uniform(-1e-5, 1e-5, (100, len(numbers)+1))
check_delta(cost.value, cost.gradient, x0, dxs)
def test_esp_cost_cube3d_gradient():
for irep in xrange(nrep):
# Some parameters
coordinates, origin, grid_rvecs, shape, pbc, vref, weights = \
get_random_esp_cost_cube3d_args(irep)
grid = UniformGrid(origin, grid_rvecs, shape, pbc)
cost = ESPCost.from_grid_data(coordinates, grid, vref, weights)
with numpy_seed(irep):
x0 = np.random.uniform(-0.5, 0.5, len(coordinates)+1)
dxs = np.random.uniform(-1e-5, 1e-5, (100, len(coordinates)+1))
check_delta(cost.value, cost.gradient, x0, dxs)
def test_gradient_functional_deriv():
fn_fchk = context.get_fn('test/n2_hfs_sto3g.fchk')
mol = IOData.from_file(fn_fchk)
obasis = mol.obasis
dm_full = mol.get_dm_full()
rtf = ExpRTransform(1e-3, 1e1, 10)
rgrid = RadialGrid(rtf)
grid = BeckeMolGrid(mol.coordinates,
mol.numbers,
mol.pseudo_numbers, (rgrid, 6),
random_rotate=False,
mode='keep')
pot = grid.points[:, 2].copy()
def fun(x):
dm_full._array[:] = x.reshape(obasis.nbasis, -1)
f = obasis.compute_grid_gradient_dm(dm_full, grid.points)
tmp = (f * f).sum(axis=1)
return 0.5 * grid.integrate(tmp, pot)
def fun_deriv(x):
dm_full._array[:] = x.reshape(obasis.nbasis, -1)
result = dm_full.new()
tmp = obasis.compute_grid_gradient_dm(dm_full, grid.points)
tmp *= pot.reshape(-1, 1)
obasis.compute_grid_gradient_fock(grid.points, grid.weights, tmp,
result)
return result._array.ravel()
eps = 1e-4
x = dm_full._array.copy().ravel()
dxs = []
for i in xrange(100):
tmp = np.random.uniform(-eps, +eps, x.shape) * x
tmp = (tmp + tmp.T) / 2
dxs.append(tmp)
from horton.test.common import check_delta
check_delta(fun, fun_deriv, x, dxs)
def check_density_gradient(obasis, dm_full, point, eps):
"""Finite difference checker for density gradient.
Parameters
----------
obasis : GOBasis
The basis set to use in the test.
dm_full : DenseTwoIndex
The spin-summed density matrix.
point : np.ndarray, shape=(3, ), dtype=float
The reference point for check_delta.
eps : float
The magnitude of the displacements.
"""
def fun(p):
"""Evaluate the density at point p."""
return obasis.compute_grid_density_dm(dm_full, np.array([p]))[0]
def fun_deriv(p):
"""Evaluate the density gradient at point p."""
return obasis.compute_grid_gradient_dm(dm_full, np.array([p]))[0]
dpoints = np.random.uniform(-eps, eps, (100, 3))
check_delta(fun, fun_deriv, point, dpoints)
def check_density_gradient(obasis, dm_full, point, eps):
"""Finite difference checker for density gradient.
Parameters
----------
obasis : GOBasis
The basis set to use in the test.
dm_full : DenseTwoIndex
The spin-summed density matrix.
point : np.ndarray, shape=(3, ), dtype=float
The reference point for check_delta.
eps : float
The magnitude of the displacements.
"""
def fun(p):
"""Evaluate the density at point p."""
return obasis.compute_grid_density_dm(dm_full, np.array([p]))[0]
def fun_deriv(p):
"""Evaluate the density gradient at point p."""
return obasis.compute_grid_gradient_dm(dm_full, np.array([p]))[0]
dpoints = np.random.uniform(-eps, eps, (100, 3))
check_delta(fun, fun_deriv, point, dpoints)
def test_ap1rog_one_dm():
fn_xyz = context.get_fn('test/li2.xyz')
mol = IOData.from_file(fn_xyz)
obasis = get_gobasis(mol.coordinates, mol.numbers, 'cc-pvdz')
lf = DenseLinalgFactory(obasis.nbasis)
occ_model = AufbauOccModel(3)
exp_alpha = lf.create_expansion(obasis.nbasis)
olp = obasis.compute_overlap(lf)
kin = obasis.compute_kinetic(lf)
na = obasis.compute_nuclear_attraction(mol.coordinates, mol.pseudo_numbers,
lf)
guess_core_hamiltonian(olp, kin, na, exp_alpha)
er = obasis.compute_electron_repulsion(lf)
external = {'nn': compute_nucnuc(mol.coordinates, mol.pseudo_numbers)}
terms = [
RTwoIndexTerm(kin, 'kin'),
RDirectTerm(er, 'hartree'),
RExchangeTerm(er, 'x_hf'),
RTwoIndexTerm(na, 'ne'),
]
ham = REffHam(terms, external)
scf_solver = PlainSCFSolver()
scf_solver(ham, lf, olp, occ_model, exp_alpha)
one = lf.create_two_index(obasis.nbasis)
one.iadd(kin)
one.iadd(na)
# Do AP1roG optimization:
geminal_solver = RAp1rog(lf, occ_model)
energy, g, l = geminal_solver(
one, er, external['nn'], exp_alpha, olp, True, **{
'checkpoint': -1,
'maxiter': {
'orbiter': 0
}
})
one_mo_ = lf.create_two_index()
one_mo_.assign_two_index_transform(one, exp_alpha)
two_mo = []
output = geminal_solver.lf.create_four_index()
output.assign_four_index_transform(er, exp_alpha, exp_alpha, exp_alpha,
exp_alpha, 'tensordot')
two_mo.append(output)
def fun(x):
one_mo = []
one_mo.append(geminal_solver.lf.create_two_index())
one_mo[0].assign(x.reshape(28, 28))
geminal_solver.clear_auxmatrix()
geminal_solver.update_auxmatrix('scf', two_mo, one_mo)
iiaa = geminal_solver.get_auxmatrix('gppqq')
iaia = geminal_solver.get_auxmatrix('lpqpq')
fock = geminal_solver.get_auxmatrix('fock')
one = geminal_solver.get_auxmatrix('t')
coeff = geminal_solver.geminal._array
lcoeff = geminal_solver.lagrange._array
lagrangian = geminal_solver.compute_total_energy()
lagrangian += np.dot(
lcoeff.ravel(order='C'),
geminal_solver.vector_function_geminal(coeff, iiaa, iaia, one,
fock))
return lagrangian
def fun_deriv(x):
one_mo = []
one_mo.append(geminal_solver.lf.create_two_index())
one_mo[0].assign(x.reshape(28, 28))
geminal_solver.clear_auxmatrix()
geminal_solver.update_auxmatrix('scf', two_mo, one_mo)
guesst = geminal_solver.generate_guess({
'type': 'random',
'factor': -0.1
})
# Optimize OAP1roG wavefunction amplitudes:
coeff = geminal_solver.solve_geminal(guesst, {'wfn': 'krylov'}, 10e-12,
128)
# Optimize OAP1roG Lagrange multipliers (lambda equations):
lcoeff = geminal_solver.solve_lagrange(guesst, {'lagrange': 'krylov'},
10e-12, 128)
onebody1 = geminal_solver.lf.create_two_index(3, 25)
onebody2 = geminal_solver.lf.create_two_index(3, 25)
onebody1.assign(coeff)
onebody2.assign(lcoeff)
onedm = geminal_solver.lf.create_one_index()
geminal_solver.compute_1dm(onedm, onebody1, onebody2, factor=2.0)
a = np.zeros((28, 28))
np.fill_diagonal(a, onedm._array.T)
return a.ravel()
x = one_mo_._array.ravel()
dxs = np.random.rand(100, 28 * 28) * 0.00001
check_delta(fun, fun_deriv, x, dxs)
def test_ap1rog_one_dm():
fn_xyz = context.get_fn('test/li2.xyz')
mol = IOData.from_file(fn_xyz)
obasis = get_gobasis(mol.coordinates, mol.numbers, 'cc-pvdz')
lf = DenseLinalgFactory(obasis.nbasis)
occ_model = AufbauOccModel(3)
exp_alpha = lf.create_expansion(obasis.nbasis)
olp = obasis.compute_overlap(lf)
kin = obasis.compute_kinetic(lf)
na = obasis.compute_nuclear_attraction(mol.coordinates, mol.pseudo_numbers, lf)
guess_core_hamiltonian(olp, kin, na, exp_alpha)
er = obasis.compute_electron_repulsion(lf)
external = {'nn': compute_nucnuc(mol.coordinates, mol.pseudo_numbers)}
terms = [
RTwoIndexTerm(kin, 'kin'),
RDirectTerm(er, 'hartree'),
RExchangeTerm(er, 'x_hf'),
RTwoIndexTerm(na, 'ne'),
]
ham = REffHam(terms, external)
scf_solver = PlainSCFSolver()
scf_solver(ham, lf, olp, occ_model, exp_alpha)
one = lf.create_two_index(obasis.nbasis)
one.iadd(kin)
one.iadd(na)
# Do AP1roG optimization:
geminal_solver = RAp1rog(lf, occ_model)
energy, g, l = geminal_solver(one, er, external['nn'], exp_alpha, olp, True, **{'checkpoint': -1, 'maxiter': {'orbiter': 0}})
one_mo_ = lf.create_two_index()
one_mo_.assign_two_index_transform(one, exp_alpha)
two_mo = []
output = geminal_solver.lf.create_four_index()
output.assign_four_index_transform(er, exp_alpha, exp_alpha, exp_alpha, exp_alpha, 'tensordot')
two_mo.append(output)
def fun(x):
one_mo = []
one_mo.append(geminal_solver.lf.create_two_index())
one_mo[0].assign(x.reshape(28,28))
geminal_solver.clear_auxmatrix()
geminal_solver.update_auxmatrix('scf', two_mo, one_mo)
iiaa = geminal_solver.get_auxmatrix('gppqq')
iaia = geminal_solver.get_auxmatrix('lpqpq')
fock = geminal_solver.get_auxmatrix('fock')
one = geminal_solver.get_auxmatrix('t')
coeff = geminal_solver.geminal._array
lcoeff = geminal_solver.lagrange._array
lagrangian = geminal_solver.compute_total_energy()
lagrangian += np.dot(lcoeff.ravel(order='C'), geminal_solver.vector_function_geminal(coeff, iiaa, iaia, one, fock))
return lagrangian
def fun_deriv(x):
one_mo = []
one_mo.append(geminal_solver.lf.create_two_index())
one_mo[0].assign(x.reshape(28,28))
geminal_solver.clear_auxmatrix()
geminal_solver.update_auxmatrix('scf', two_mo, one_mo)
guesst = geminal_solver.generate_guess({'type': 'random', 'factor': -0.1})
# Optimize OAP1roG wavefunction amplitudes:
coeff = geminal_solver.solve_geminal(guesst, {'wfn': 'krylov'}, 10e-12, 128)
# Optimize OAP1roG Lagrange multipliers (lambda equations):
lcoeff = geminal_solver.solve_lagrange(guesst, {'lagrange': 'krylov'}, 10e-12, 128)
onebody1 = geminal_solver.lf.create_two_index(3,25)
onebody2 = geminal_solver.lf.create_two_index(3,25)
onebody1.assign(coeff)
onebody2.assign(lcoeff)
onedm = geminal_solver.lf.create_one_index()
geminal_solver.compute_1dm(onedm, onebody1, onebody2, factor=2.0)
a = np.zeros((28,28))
np.fill_diagonal(a, onedm._array.T)
return a.ravel()
x = one_mo_._array.ravel()
dxs = np.random.rand(100, 28*28)*0.00001
check_delta(fun, fun_deriv, x, dxs)
