def test_cc_unitary(): # pragma: nocover
from pyscf import gto
from pyscf import scf
mol = gto.Mole()
mol.atom = [[8, (0., 0., 0.)], [1, (0., -0.757, 0.587)],
[1, (0., 0.757, 0.587)]]
mol.basis = {
'H': 'sto-3g',
'O': 'sto-3g',
}
mol.build()
rhf = scf.RHF(mol)
rhf.scf() # -76.0267656731
from tcc.cc_solvers import residual_diis_solver
from tcc.cc_solvers import classic_solver
from tcc.rccsd import RCCSD_UNIT, RCCSD
cc1 = RCCSD_UNIT(rhf)
cc2 = RCCSD(rhf)
converged1, energy2, _ = residual_diis_solver(cc1,
ndiis=5,
conv_tol_energy=-1,
conv_tol_res=1e-10)
converged2, energy2, _ = residual_diis_solver(cc2,
ndiis=5,
conv_tol_energy=-1,
conv_tol_res=1e-10)
def run_ccsd():
us = U_VALUES_REFERENCE.copy()
from tcc.cc_solvers import residual_diis_solver
energies = []
amps = None
for idxu, u in enumerate(us):
rhf = hubbard_from_scf(scf.RHF, N, N, u, USE_PBC)
rhf.scf()
scf_energy = rhf.e_tot
cc = RCCSD_UNIT(rhf)
converged, energy, amps = residual_diis_solver(cc,
conv_tol_energy=1e-6,
lam=15,
amps=amps,
max_cycle=2000)
energies.append(scf_energy + energy)
np.savetxt('reference/{}-site/RCCSD.txt'.format(N),
np.column_stack((us, energies)),
header='U E_total')
# Plot
from matplotlib import pyplot as plt
fig, ax = plt.subplots()
plt.plot(us, energies)
plt.xlabel('$U$')
plt.ylabel('$E$, H')
plt.title('Energy behavior for different ranks')
fig.show()
def test_cc_hubbard(self):
from tcc.cc_solvers import residual_diis_solver
from tcc.rccsd import RCCSD_UNIT
cc = RCCSD_UNIT(self.rhf)
converged, energy, _ = residual_diis_solver(cc,
ndiis=5,
conv_tol_res=1e-6,
lam=5,
max_cycle=100)
self.assertEqual(np.allclose(energy, -0.93834495800469087, 1e-6), True)
def test_update_diis_solver():
from tcc.hubbard import hubbard_from_scf
from pyscf import scf
N = 6
U = 4
USE_PBC = 'y'
rhf = hubbard_from_scf(scf.RHF, N, N, U, USE_PBC)
rhf.scf() # -76.0267656731
from tcc.rccsd import RCCSD_UNIT
from tcc.rccsd_cpd import (RCCSD_nCPD_LS_T_HUB)
cc1 = RCCSD_UNIT(rhf)
cc2 = RCCSD_nCPD_LS_T_HUB(rhf, rankt={'t2': 30})
from tcc.cc_solvers import (residual_diis_solver, update_diis_solver,
classic_solver)
cc1._converged = False
converged1, energy1, amps1 = residual_diis_solver(cc1,
conv_tol_energy=1e-8,
lam=3,
max_cycle=100)
import pickle
with open('test_amps.p', 'rb') as fp:
ampsi = pickle.load(fp)
cc2._converged = False
converged2, energy2, amps2 = update_diis_solver(cc2,
conv_tol_energy=1e-8,
conv_tol_res=1e-8,
beta=0.666,
max_cycle=100,
amps=ampsi)
cc2._converged = False
converged2, energy2, amps2 = classic_solver(cc2,
conv_tol_energy=1e-8,
conv_tol_res=1e-8,
lam=3,
max_cycle=100)
def test_cc_hubbard(): # pragma: nocover
from pyscf import gto
from pyscf import scf
from tcc.hubbard import hubbard_from_scf
rhf = hubbard_from_scf(scf.RHF, 6, 6, 3, 'y')
rhf.damp = -4.0
rhf.scf()
from tcc.cc_solvers import root_solver, residual_diis_solver
from tcc.rccsd import RCCSD_UNIT
cc = RCCSD_UNIT(rhf)
converged, energy, _ = root_solver(cc, conv_tol=1e-6)
def calc_solutions_diff_r_eq_cpd():
"""
Plot error of RCCSD-CPD vs rank for weak corellation
"""
rankst = RANKS_T.copy()
# Run RHF calculations
rhf = hubbard_from_scf(scf.RHF, N, N, U, 'y')
rhf.damp = -4.0
rhf.scf()
rhf_results = tuple([rhf.e_tot, rhf.mo_coeff, rhf.mo_energy])
with open('calculated/{}-site/amps_and_scf_eq/rhf_results_u_{}.p'.format(N, U), 'wb') as fp:
pickle.dump(rhf_results, fp)
# Run reference calculation
cc_ref = RCCSD_UNIT(rhf)
_, energy_ref, amps_ref = root_solver(cc_ref, conv_tol=1e-10)
cc_results = tuple([energy_ref, amps_ref])
with open('calculated/{}-site/amps_and_scf_eq/cc_results_u_{}.p'.format(N, U), 'wb') as fp:
pickle.dump(cc_results, fp)
all_amps = []
for idx, rank in enumerate(rankst):
tim = time.process_time()
cc = RCCSD_nCPD_LS_T_HUB(rhf, rankt={'t2': rank})
converged, energy, amps = classic_solver(
cc, lam=1.8, conv_tol_energy=1e-14,
conv_tol_amps=1e-10, max_cycle=40000,
verbose=logger.NOTE)
if not converged:
Warning(
'Warning: N = {}, U = {} '
'Rank = {} did not converge'.format(N, U, rank)
)
all_amps.append(tuple([energy, rank, amps]))
elapsed = time.process_time() - tim
print('Step {} out of {}, rank = {}, time: {}'.format(
idx + 1, len(rankst), rank, elapsed
))
with open('calculated/{}-site/amps_and_scf_eq/energy_rank_amps_u_{}.p'.format(N, U), 'wb') as fp:
pickle.dump(all_amps, fp)
def test_rccsd_unit(self):
from tcc.cc_solvers import classic_solver, residual_diis_solver
from tcc.rccsd import RCCSD, RCCSD_UNIT
cc1 = RCCSD(self.rhf)
cc2 = RCCSD_UNIT(self.rhf)
converged1, energy1, amps = classic_solver(cc1)
converged2, energy2, _ = classic_solver(cc2)
self.assertEqual(converged1, converged2)
self.assertEqual(np.allclose(energy1, -0.2133432609672395, 1e-5), True)
self.assertEqual(np.allclose(energy1, energy2, 1e-5), True)
converged1, energy1, _ = residual_diis_solver(cc1, amps=amps,
conv_tol_energy=1e-10)
converged2, energy2, _ = residual_diis_solver(cc2, amps=amps,
conv_tol_energy=1e-10)
self.assertEqual(np.allclose(energy1, -0.2133432609672395, 1e-5), True)
self.assertEqual(np.allclose(energy1, energy2, 1e-5), True)
def calc_err_vs_r_cpd():
"""
Plot error of RCCSD-CPD vs rank for weak corellation
"""
# Set up parameters of the script
N = 14
U = 2
rankst = np.round(N**np.linspace(0.2, 1.8, num=10)).astype('int')
# Run RHF calculations
from pyscf import scf
from tcc.hubbard import hubbard_from_scf
rhf = hubbard_from_scf(scf.RHF, N, N, U, 'y')
rhf.damp = -4.0
rhf.scf()
from tcc.cc_solvers import (classic_solver, root_solver)
from tcc.rccsd import RCCSD_UNIT
from tcc.rccsd_cpd import RCCSD_CPD_LS_T_HUB
from tensorly.decomposition import parafac
# Run reference calculation
cc_ref = RCCSD_UNIT(rhf)
_, energy_ref, amps_ref = root_solver(cc_ref, conv_tol=1e-10)
energies = []
deltas = []
for idx, rank in enumerate(rankst):
tim = time.process_time()
cc = RCCSD_CPD_LS_T_HUB(rhf, rankt=rank)
xs = parafac(
amps_ref.t2, rank, tol=1e-14
)
amps_guess = cc.types.AMPLITUDES_TYPE(
amps_ref.t1, *xs
)
converged, energy, amps = classic_solver(
cc, lam=1.8, conv_tol_energy=1e-14,
conv_tol_amps=1e-10, max_cycle=10000,
amps=amps_guess, verbose=logger.NOTE)
if not converged:
Warning(
'Warning: N = {}, U = {} '
'Rank = {} did not converge'.format(N, U, rank)
)
energies.append(energy)
deltas.append(energy - energy_ref)
elapsed = time.process_time() - tim
print('Step {} out of {}, rank = {}, time: {}'.format(
idx + 1, len(rankst), rank, elapsed
))
results = np.column_stack((rankst, energies, deltas))
np.savetxt(
'calculated/{}-site/err_vs_rank.txt'.format(N),
results, header='Rank Energy Delta', fmt=('%i', '%e', '%e')
)
rankst, energies, deltas = np.loadtxt(
'calculated/{}-site/err_vs_rank.txt'.format(N), unpack=True)
# Plot
from matplotlib import pyplot as plt
plt.plot(np.log(rankst)/np.log(N), np.log10(np.abs(deltas)))
plt.xlabel('log$_N(R)$')
plt.ylabel('log($\Delta$)')
plt.title('Error dependence on rank in weak correlation regime')
plt.show()
def calc_t2_norm_vs_u_rccsd():
"""
Calculate norms of T2 in conventional RCCSD
"""
us = U_VALUES.copy()
lambdas = LAMBDAS.copy()
with open('calculated/{}-site/scfs_different_u_t1.p'.format(N),
'rb') as fp:
ref_scfs = pickle.load(fp)
t1_norms = []
energies = []
t2_norms = []
# timb = time.process_time()
solutions = []
amps = None
for idxu, (u, curr_scf) in enumerate(zip(us, ref_scfs)):
rhf = hubbard_from_scf(scf.RHF, N, N, u, 'y')
rhf.max_cycle = 1
rhf.scf()
e_scf, mo_coeff, mo_energy = curr_scf
cc = RCCSD_UNIT(rhf, mo_coeff=mo_coeff, mo_energy=mo_energy)
converged, energy, amps = residual_diis_solver(cc,
conv_tol_energy=1e-9,
conv_tol_res=1e-9,
max_cycle=10000,
verbose=logger.NOTE,
amps=None,
lam=14)
norms = amps.map(np.linalg.norm)
t1_norms.append(norms.t1)
t2_norms.append(norms.t2)
energies.append(energy)
solutions.append([u, curr_scf, amps])
print('Step {} out of {}'.format(idxu + 1, len(us)))
t1_norms = np.column_stack((us, t1_norms))
t2_norms = np.column_stack((us, t2_norms))
np.savetxt('calculated/{}-site/t1_norm_vs_u_rccsd.txt'.format(N),
t1_norms,
header='U |t1|')
np.savetxt('calculated/{}-site/t2_norm_vs_u_rccsd.txt'.format(N),
t2_norms,
header='U |t2|')
with open('calculated/{}-site/amps_and_scf_rccsd.p'.format(N), 'wb') as fp:
pickle.dump(solutions, fp)
us, *t2_l = np.loadtxt(
'calculated/{}-site/t2_norm_vs_u_rccsd.txt'.format(N), unpack=True)
# Plot
from matplotlib import pyplot as plt
fig, ax = plt.subplots()
plt.plot(us, t2_l[0])
plt.xlabel('$U / t$')
plt.ylabel('$||T^{2}||$')
plt.title('$T^{2}$ norm behavior for different ranks')
fig.show()