def perfect_experimental_avg(self, params):
"""
calculates the exact experimental result of the operator the Experiment object was initialized with
:param params: (list) the list of parameters for the state preparation ansatz.
"""
if(self.prepare_each_time_==False):
#1 initialize a quantum computer
qc = qforte.Computer(self.n_qubits_)
#2 build/update generator with params
# self.generator_.set_parameters(params)
#3 apply generator (once if prepare_each_time = False, N_sample times if)
qc.apply_circuit(self.generator_)
#4 measure operator
n_terms = len(self.operator_.terms())
term_sum = 0.0
for k in range(n_terms):
term_sum += self.operator_.terms()[k][0] * qc.perfect_measure_circuit(self.operator_.terms()[k][1]);
return numpy.real(term_sum)
elif(self.prepare_each_time_==True):
raise Exception('No support yet for measurement with multiple state preparations')
def build_classical_CI_mats(self):
"""Builds a classical configuration interaction out of single determinants.
"""
num_tot_basis = len(self._pre_sa_ref_lst)
h_CI = np.zeros((num_tot_basis,num_tot_basis), dtype=complex)
omega_lst = []
Homega_lst = []
for i, ref in enumerate(self._pre_sa_ref_lst):
# NOTE: do NOT use Uprep here (is determinant specific).
Un = qforte.Circuit()
for j in range(self._nqb):
if ref[j] == 1:
Un.add(qforte.gate('X', j, j))
QC = qforte.Computer(self._nqb)
QC.apply_circuit(Un)
omega_lst.append(np.asarray(QC.get_coeff_vec(), dtype=complex))
Homega = np.zeros((2**self._nqb), dtype=complex)
QC.apply_operator(self._qb_ham)
Homega_lst.append(np.asarray(QC.get_coeff_vec(), dtype=complex))
for j in range(len(omega_lst)):
h_CI[i][j] = np.vdot(omega_lst[i], Homega_lst[j])
h_CI[j][i] = np.conj(h_CI[i][j])
return h_CI
def build_sparse_S_b(self, b):
b_sparse = []
idx_sparse = []
for I, bI in enumerate(b):
if(np.abs(bI) > self._b_thresh):
idx_sparse.append(I)
b_sparse.append(bI)
Idim = len(idx_sparse)
self._n_pauli_trm_measures += int(Idim*(Idim+1)*0.5)
S = np.zeros((len(b_sparse),len(b_sparse)), dtype=complex)
Ipsi_qc = qf.Computer(self._nqb)
Ipsi_qc.set_coeff_vec(copy.deepcopy(self._qc.get_coeff_vec()))
CI = np.zeros(shape=(Idim, int(2**self._nqb)), dtype=complex)
for i in range(Idim):
S[i][i] = 1.0 # With Pauli strings, this is always the inner product
Ii = idx_sparse[i]
Ipsi_qc.apply_operator(self._sig.terms()[Ii][1])
CI[i,:] = copy.deepcopy(Ipsi_qc.get_coeff_vec())
for j in range(i):
S[i][j] = S[j][i] = np.vdot(CI[i,:], CI[j,:])
Ipsi_qc.set_coeff_vec(copy.deepcopy(self._qc.get_coeff_vec()))
return idx_sparse, np.real(S), np.real(b_sparse)
def evolve(self):
"""Perform QITE for a time step :math:`\\Delta \\beta`.
"""
self._Uqite.add(self._Uprep)
self._qc = qf.Computer(self._nqb)
self._qc.apply_circuit(self._Uqite)
if(self._do_lanczos):
self._lanczos_vecs = []
self._Hlanczos_vecs = []
self._lanczos_vecs.append(copy.deepcopy(self._qc.get_coeff_vec()))
qcSig_temp = qf.Computer(self._nqb)
qcSig_temp.set_coeff_vec(copy.deepcopy(self._qc.get_coeff_vec()))
qcSig_temp.apply_operator(self._qb_ham)
self._Hlanczos_vecs.append(copy.deepcopy(qcSig_temp.get_coeff_vec()))
print(f"{'beta':>7}{'E(beta)':>18}{'N(params)':>14}{'N(CNOT)':>18}{'N(measure)':>20}")
print('-------------------------------------------------------------------------------')
print(f' {0.0:7.3f} {self._Ekb[0]:+15.9f} {self._n_classical_params:8} {self._n_cnot:10} {self._n_pauli_trm_measures:12}')
if (self._print_summary_file):
f = open("summary.dat", "w+", buffering=1)
f.write(f"#{'beta':>7}{'E(beta)':>18}{'N(params)':>14}{'N(CNOT)':>18}{'N(measure)':>20}\n")
f.write('#-------------------------------------------------------------------------------\n')
f.write(f' {0.0:7.3f} {self._Ekb[0]:+15.9f} {self._n_classical_params:8} {self._n_cnot:10} {self._n_pauli_trm_measures:12}\n')
for kb in range(1, self._nbeta):
self.do_qite_step()
if(self._do_lanczos):
if(kb % self._lanczos_gap == 0):
self._lanczos_vecs.append(copy.deepcopy(self._qc.get_coeff_vec()))
qcSig_temp = qf.Computer(self._nqb)
qcSig_temp.set_coeff_vec(copy.deepcopy(self._qc.get_coeff_vec()))
qcSig_temp.apply_operator(self._qb_ham)
self._Hlanczos_vecs.append(copy.deepcopy(qcSig_temp.get_coeff_vec()))
print(f' {kb*self._db:7.3f} {self._Ekb[kb]:+15.9f} {self._n_classical_params:8} {self._n_cnot:10} {self._n_pauli_trm_measures:12}')
if (self._print_summary_file):
f.write(f' {kb*self._db:7.3f} {self._Ekb[kb]:+15.9f} {self._n_classical_params:8} {self._n_cnot:10} {self._n_pauli_trm_measures:12}\n')
self._Egs = self._Ekb[-1]
if (self._print_summary_file):
f.close()
def fill_excited_dets(self):
for _, sq_op in self._pool_obj:
# 1. Identify the excitation operator
# occ => i,j,k,...
# vir => a,b,c,...
# sq_op is 1.0(a^ b^ i j) - 1.0(j^ i^ b a)
temp_idx = sq_op.terms()[0][2][-1]
# TODO: This code assumes that the first N orbitals are occupied, and the others are virtual.
# Use some other mechanism to identify the occupied orbitals, so we can use use PQE on excited
# determinants.
if temp_idx < int(
sum(self._ref) / 2): # if temp_idx is an occupied idx
sq_creators = sq_op.terms()[0][1]
sq_annihilators = sq_op.terms()[0][2]
else:
sq_creators = sq_op.terms()[0][2]
sq_annihilators = sq_op.terms()[0][1]
# 2. Get the bit representation of the sq_ex_op acting on the reference.
# We determine the projective condition for this amplitude by zero'ing this residual.
# `destroyed` exists solely for error catching.
destroyed = False
excited_det = qforte.QubitBasis(self._nqb)
for k, occ in enumerate(self._ref):
excited_det.set_bit(k, occ)
# loop over annihilators
for p in reversed(sq_annihilators):
if (excited_det.get_bit(p) == 0):
destroyed = True
break
excited_det.set_bit(p, 0)
# then over creators
for p in reversed(sq_creators):
if (excited_det.get_bit(p) == 1):
destroyed = True
break
excited_det.set_bit(p, 1)
if destroyed:
raise ValueError(
"no ops should destroy reference, something went wrong!!")
I = excited_det.add()
qc_temp = qforte.Computer(self._nqb)
qc_temp.apply_circuit(self._Uprep)
qc_temp.apply_operator(sq_op.jw_transform())
phase_factor = qc_temp.get_coeff_vec()[I]
self._excited_dets.append((I, phase_factor))
def run(self,
beta=1.0,
db=0.2,
expansion_type='SD',
sparseSb=True,
b_thresh=1.0e-6,
x_thresh=1.0e-10,
do_lanczos=False,
lanczos_gap=2):
self._beta = beta
self._db = db
self._nbeta = int(beta/db)+1
self._expansion_type = expansion_type
self._sparseSb = sparseSb
self._total_phase = 1.0 + 0.0j
self._Uqite = qf.Circuit()
self._b_thresh = b_thresh
self._x_thresh = x_thresh
self._n_classical_params = 0
self._n_cnot = self._Uprep.get_num_cnots()
self._n_pauli_trm_measures = 0
self._do_lanczos = do_lanczos
self._lanczos_gap = lanczos_gap
qc_ref = qf.Computer(self._nqb)
qc_ref.apply_circuit(self._Uprep)
self._Ekb = [np.real(qc_ref.direct_op_exp_val(self._qb_ham))]
# Print options banner (should done for all algorithms).
self.print_options_banner()
# Build expansion pool.
self.build_expansion_pool()
# Do the imaginary time evolution.
self.evolve()
if (self._do_lanczos):
self.do_qlanczos()
# Print summary banner (should done for all algorithms).
self.print_summary_banner()
# verify that required attributes were defined
# (should done for all algorithms).
self.verify_run()
def build_orb_energies(self):
"""Calculates single qubit energies. Used in quasi-Newton updates.
"""
self._orb_e = []
print('\nBuilding single particle energies list:')
print('---------------------------------------', flush=True)
qc = qf.Computer(self._nqb)
qc.apply_circuit(build_Uprep(self._ref, 'occupation_list'))
E0 = qc.direct_op_exp_val(self._qb_ham)
for i in range(self._nqb):
qc = qf.Computer(self._nqb)
qc.apply_circuit(build_Uprep(self._ref, 'occupation_list'))
qc.apply_gate(qf.gate('X', i, i))
Ei = qc.direct_op_exp_val(self._qb_ham)
if(i<sum(self._ref)):
ei = E0 - Ei
else:
ei = Ei - E0
print(f' {i:3} {ei:+16.12f}', flush=True)
self._orb_e.append(ei)
def get_residual_vector(self, trial_amps):
"""Returns the residual vector with elements pertaining to all operators
in the ansatz circuit.
Parameters
----------
trial_amps : list of floats
The list of (real) floating point numbers which will characterize
the state preparation circuit used in calculation of the residuals.
"""
if (self._pool_type == 'sa_SD'):
raise ValueError(
'Must use single term particle-hole nbody operators for residual calculation'
)
U = self.ansatz_circuit(trial_amps)
qc_res = qforte.Computer(self._nqb)
qc_res.apply_circuit(self._Uprep)
qc_res.apply_circuit(U)
qc_res.apply_operator(self._qb_ham)
qc_res.apply_circuit(U.adjoint())
coeffs = qc_res.get_coeff_vec()
residuals = []
for I, phase_factor in self._excited_dets:
# Get the residual element, after accounting for numerical noise.
res_m = coeffs[I] * phase_factor
if (np.imag(res_m) != 0.0):
raise ValueError(
"residual has imaginary component, something went wrong!!")
if (self._noise_factor > 1e-12):
res_m = np.random.normal(np.real(res_m), self._noise_factor)
residuals.append(res_m)
self._res_vec_evals += 1
self._res_m_evals += len(self._tamps)
return residuals
def build_S(self):
"""Construct the matrix S (eq. 5a) of Motta.
"""
Idim = self._NI
S = np.zeros((Idim, Idim), dtype=complex)
Ipsi_qc = qf.Computer(self._nqb)
Ipsi_qc.set_coeff_vec(copy.deepcopy(self._qc.get_coeff_vec()))
# CI[I][J] = (σ_I Ψ)_J
self._n_pauli_trm_measures += int(self._NI*(self._NI+1)*0.5)
CI = np.zeros(shape=(Idim, int(2**self._nqb)), dtype=complex)
for i in range(Idim):
S[i][i] = 1.0 # With Pauli strings, this is always the inner product
Ipsi_qc.apply_operator(self._sig.terms()[i][1])
CI[i,:] = copy.deepcopy(Ipsi_qc.get_coeff_vec())
for j in range(i):
S[i][j] = S[j][i] = np.vdot(CI[i,:], CI[j,:])
Ipsi_qc.set_coeff_vec(copy.deepcopy(self._qc.get_coeff_vec()))
return np.real(S)
def build_b(self):
"""Construct the vector b (eq. 5b) of Motta, with h[m] the full Hamiltonian.
"""
b = np.zeros(self._NI, dtype=complex)
denom = np.sqrt(1 - 2*self._db*self._Ekb[-1])
prefactor = -1.0j / denom
self._n_pauli_trm_measures += self._Nl * self._NI
Hpsi_qc = qf.Computer(self._nqb)
Hpsi_qc.set_coeff_vec(copy.deepcopy(self._qc.get_coeff_vec()))
Hpsi_qc.apply_operator(self._qb_ham)
C_Hpsi_qc = copy.deepcopy(Hpsi_qc.get_coeff_vec())
for I, (op_coefficient, operator) in enumerate(self._sig.terms()):
Hpsi_qc.apply_operator(operator)
exp_val = np.vdot(self._qc.get_coeff_vec(), Hpsi_qc.get_coeff_vec())
b[I] = prefactor * op_coefficient * exp_val
Hpsi_qc.set_coeff_vec(copy.deepcopy(C_Hpsi_qc))
return np.real(b)
def build_sa_qk_mats(self):
"""Returns the QK effective hamiltonain and overlap matrices in a basis
of spin adapted references.
"""
num_tot_basis = len(self._sa_ref_lst) * self._nstates_per_ref
h_mat = np.zeros((num_tot_basis,num_tot_basis), dtype=complex)
s_mat = np.zeros((num_tot_basis,num_tot_basis), dtype=complex)
omega_lst = []
Homega_lst = []
for i, ref in enumerate(self._sa_ref_lst):
for m in range(self._nstates_per_ref):
Um = qforte.Circuit()
phase1 = 1.0
if(m>0):
fact = (0.0-1.0j) * m * self._mr_dt
expn_op1, phase1 = trotterize(self._qb_ham, factor=fact, trotter_number=self._trotter_number)
Um.add(expn_op1)
QC = qforte.Computer(self._nqb)
state_prep_lst = []
for term in ref:
coeff = term[0]
det = term[1]
idx = ref_to_basis_idx(det)
state = qforte.QubitBasis(idx)
state_prep_lst.append( (state, coeff) )
QC.set_state(state_prep_lst)
QC.apply_circuit(Um)
QC.apply_constant(phase1)
omega_lst.append(np.asarray(QC.get_coeff_vec(), dtype=complex))
QC.apply_operator(self._qb_ham)
Homega_lst.append(np.asarray(QC.get_coeff_vec(), dtype=complex))
if(self._diagonalize_each_step):
print('\n\n')
print(f"{'k(S)':>7}{'E(Npar)':>19}{'N(params)':>14}{'N(CNOT)':>18}{'N(measure)':>20}")
print('-------------------------------------------------------------------------------')
# TODO (opt): add this to previous loop
for p in range(num_tot_basis):
for q in range(p, num_tot_basis):
h_mat[p][q] = np.vdot(omega_lst[p], Homega_lst[q])
h_mat[q][p] = np.conj(h_mat[p][q])
s_mat[p][q] = np.vdot(omega_lst[p], omega_lst[q])
s_mat[q][p] = np.conj(s_mat[p][q])
if (self._diagonalize_each_step):
# TODO (cleanup): have this print to a separate file
evals, evecs = canonical_geig_solve(s_mat[0:p+1, 0:p+1],
h_mat[0:p+1, 0:p+1],
print_mats=False,
sort_ret_vals=True)
scond = np.linalg.cond(s_mat[0:p+1, 0:p+1])
cs_str = '{:.2e}'.format(scond)
k = p+1
self._n_classical_params = k
if(k==1):
self._n_cnot = self._srqk._n_cnot
else:
self._n_cnot = 2 * Um.get_num_cnots()
self._n_pauli_trm_measures = k * self._Nl + self._srqk._n_pauli_trm_measures
self._n_pauli_trm_measures += k * (k-1) * self._Nl
self._n_pauli_trm_measures += k * (k-1)
print(f' {scond:7.2e} {np.real(evals[self._target_root]):+15.9f} {self._n_classical_params:8} {self._n_cnot:10} {self._n_pauli_trm_measures:12}')
return s_mat, h_mat
def build_qk_mats(self):
num_tot_basis = len(self._single_det_refs) * self._nstates_per_ref
h_mat = np.zeros((num_tot_basis,num_tot_basis), dtype=complex)
s_mat = np.zeros((num_tot_basis,num_tot_basis), dtype=complex)
# TODO (opt): make numpy arrays.
omega_lst = []
Homega_lst = []
for i, ref in enumerate(self._single_det_refs):
for m in range(self._nstates_per_ref):
# NOTE: do NOT use Uprep here (is determinant specific).
Um = qforte.Circuit()
for j in range(self._nqb):
if ref[j] == 1:
Um.add(qforte.gate('X', j, j))
phase1 = 1.0
if(m>0):
fact = (0.0-1.0j) * m * self._mr_dt
expn_op1, phase1 = trotterize(self._qb_ham, factor=fact, trotter_number=self._trotter_number)
Um.add(expn_op1)
QC = qforte.Computer(self._nqb)
QC.apply_circuit(Um)
QC.apply_constant(phase1)
omega_lst.append(np.asarray(QC.get_coeff_vec(), dtype=complex))
QC.apply_operator(self._qb_ham)
Homega_lst.append(np.asarray(QC.get_coeff_vec(), dtype=complex))
if(self._diagonalize_each_step):
print('\n\n')
print(f"{'k(S)':>7}{'E(Npar)':>19}{'N(params)':>14}{'N(CNOT)':>18}{'N(measure)':>20}")
print('-------------------------------------------------------------------------------')
for p in range(num_tot_basis):
for q in range(p, num_tot_basis):
h_mat[p][q] = np.vdot(omega_lst[p], Homega_lst[q])
h_mat[q][p] = np.conj(h_mat[p][q])
s_mat[p][q] = np.vdot(omega_lst[p], omega_lst[q])
s_mat[q][p] = np.conj(s_mat[p][q])
if (self._diagonalize_each_step):
# TODO (cleanup): have this print to a separate file
evals, evecs = canonical_geig_solve(s_mat[0:p+1, 0:p+1],
h_mat[0:p+1, 0:p+1],
print_mats=False,
sort_ret_vals=True)
scond = np.linalg.cond(s_mat[0:p+1, 0:p+1])
cs_str = '{:.2e}'.format(scond)
k = p+1
self._n_classical_params = k
if(k==1):
self._n_cnot = self._srqk._n_cnot
else:
self._n_cnot = 2 * Um.get_num_cnots()
self._n_pauli_trm_measures = k * self._Nl + self._srqk._n_pauli_trm_measures
self._n_pauli_trm_measures += k * (k-1) * self._Nl
self._n_pauli_trm_measures += k * (k-1)
print(f' {scond:7.2e} {np.real(evals[self._target_root]):+15.9f} {self._n_classical_params:8} {self._n_cnot:10} {self._n_pauli_trm_measures:12}')
return s_mat, h_mat
def run(self,
guess_energy: float,
t = 1.0,
nruns = 20,
success_prob = 0.5,
num_precise_bits = 4):
"""
guess_energy : A guess for the eigenvalue of the eigenspace with which |0>^(n)
has greatest overlap. You should be confident the ground state is within
t : A scaling parameter that controls the precision of the computation. You should
confident that the eigenvalue of interest is within +/- t of the guess energy.
"""
# float: evolution times
self._t = t
# int: number of times to sample the eigenvalue distribution
self._nruns = nruns
self._success_prob = success_prob
self._num_precise_bits = num_precise_bits
self._Uqpe = qforte.Circuit()
# int: The number of qubits needed to represent the state.
self._n_state_qubits = self._nqb
eps = 1 - success_prob
# int: The number of ancilla qubits used to hold eigenvalue information
self._n_ancilla = num_precise_bits + int(np.log2(2 + (1.0/eps)))
# int: The total number of qubits needed in the circuit
self._n_tot_qubits = self._n_state_qubits + self._n_ancilla
self._abegin = self._n_state_qubits
self._aend = self._n_tot_qubits - 1
self._n_classical_params = 0
self._n_pauli_trm_measures = nruns
self._guess_energy = guess_energy
self._guess_periods = round(self._t * guess_energy / (-2 * np.pi))
# Print options banner (should done for all algorithms).
self.print_options_banner()
######### QPE ########
# Apply Hadamard gates to all ancilla qubits
self._Uqpe.add(self.get_Uhad())
# Prepare the trial state on the non-ancilla qubits
self._Uqpe.add(self._Uprep)
# add controll e^-iHt circuit
self._Uqpe.add(self.get_dynamics_circ())
# add reverse QFT
self._Uqpe.add(self.get_qft_circuit('reverse'))
computer = qforte.Computer(self._n_tot_qubits)
computer.apply_circuit(self._Uqpe)
self._n_cnot = self._Uqpe.get_num_cnots()
if(self._fast):
z_readouts = computer.measure_z_readouts_fast(self._abegin, self._aend, self._nruns)
else:
Zcirc = self.get_z_circuit()
z_readouts = computer.measure_readouts(Zcirc, self._nruns)
self._phases = []
for readout in z_readouts:
val = sum(z / (2**i) for i, z in enumerate(readout, start=1))
self._phases.append(val)
# find final binary string of phase readouts:
final_readout = []
final_readout_aves = []
for i in range(self._n_ancilla):
iave = sum(readout[i] for readout in z_readouts) / nruns
final_readout_aves.append(iave)
final_readout.append(1 if iave > 0.5 else 0)
self._final_phase = sum(z / (2**i) for i, z in enumerate(final_readout, start=1))
E_u = -2 * np.pi * (self._final_phase + self._guess_periods - 1) / t
E_l = -2 * np.pi * (self._final_phase + self._guess_periods - 0) / t
E_qpe = E_l if abs(E_l - guess_energy) < abs(E_u - guess_energy) else E_u
res = stats.mode(np.asarray(self._phases))
self._mode_phase = res.mode[0]
E_u = -2 * np.pi * (self._mode_phase + self._guess_periods - 1) / t
E_l = -2 * np.pi * (self._mode_phase + self._guess_periods - 0) / t
self._mode_energy = E_l if abs(E_l - guess_energy) < abs(E_u - guess_energy) else E_u
print('\n ==> QPE readout averages <==')
print('------------------------------------------------')
for i, ave in enumerate(final_readout_aves):
print(' bit ', i, ': ', ave)
print('\n Final bit readout: ', final_readout)
######### QPE ########
# set Egs
self._Egs = E_qpe
# set Umaxdepth
self._Umaxdepth = self._Uqpe
# Print summary banner (should done for all algorithms).
self.print_summary_banner()
# verify that required attributes were defined
# (should done for all algorithms).
self.verify_run()
def update_ansatz(self):
self._n_pauli_measures_k = 0
# TODO: Check if this deepcopy is needed. The one argument of energy_feval should be const.
x0 = copy.deepcopy(self._tamps)
init_gues_energy = self.energy_feval(x0)
# do U^dag e^iH U |Phi_o> = |Phi_res>
U = self.ansatz_circuit()
qc_res = qf.Computer(self._nqb)
qc_res.apply_circuit(self._Uprep)
qc_res.apply_circuit(U)
qc_res.apply_circuit(self._eiH)
qc_res.apply_circuit(U.adjoint())
res_coeffs = qc_res.get_coeff_vec()
# build different res_sq list using M_omega
if (self._M_omega != 'inf'):
res_sq_tmp = [
np.real(np.conj(res_coeffs[I]) * res_coeffs[I])
for I in range(len(res_coeffs))
]
# Nmu_lst => [ det1, det2, det3, ... det_M_omega]
det_lst = np.random.choice(len(res_coeffs),
self._M_omega,
p=res_sq_tmp)
print(f'|Co|dt^2 : {np.amax(res_sq_tmp):12.14f}')
print(
f'mu_o : {np.where(res_sq_tmp == np.amax(res_sq_tmp))[0][0]}'
)
No_idx = np.where(res_sq_tmp == np.amax(res_sq_tmp))[0][0]
print(f'\nNo_idx {No_idx:4}')
No = np.count_nonzero(det_lst == No_idx)
print(f'\nNo {No:10}')
res_sq = []
Nmu_lst = []
for mu in range(len(res_coeffs)):
Nmu = np.count_nonzero(det_lst == mu)
if (Nmu > 0):
print(
f'mu: {mu:8} Nmu {Nmu:10} r_mu: { Nmu / (self._M_omega):12.14f} '
)
Nmu_lst.append((Nmu, mu))
res_sq.append((Nmu / (self._M_omega), mu))
## 1. sort
Nmu_lst.sort()
res_sq.sort()
## 2. set norm
self._curr_res_sq_norm = sum(
rmu_sq[0] for rmu_sq in res_sq[:-1]) / (self._dt * self._dt)
## 3. print stuff
print(' \n--> Begin selection opt with residual magnitudes:')
print(' Initial guess energy: ',
round(init_gues_energy, 10))
print(
f' Norm of approximate res vec: {np.sqrt(self._curr_res_sq_norm):14.12f}'
)
## 4. check conv status (need up update function with if(M_omega != 'inf'))
if (len(Nmu_lst) == 1):
print(' SPQE converged with M_omega thresh!')
self._converged = True
self._final_energy = self._energies[-1]
self._final_result = self._results[-1]
else:
self._converged = False
## 5. add new toperator
if not self._converged:
if self._verbose:
print('\n')
print(' op index (Imu) Number of times measured')
print(' -----------------------------------------------')
for Nmu_tup in Nmu_lst[:-1]:
if (self._verbose):
print(
f" {Nmu_tup[1]:10} {np.real(Nmu_tup[0]):14}"
)
if (Nmu_tup[1] not in self._tops):
self._tops.insert(0, Nmu_tup[1])
self._tamps.insert(0, 0.0)
self.add_from_basis_idx(Nmu_tup[1])
self._n_classical_params_lst.append(len(self._tops))
else: # when M_omega == 'inf', proceed with standard SPQE
res_sq = [(np.real(np.conj(res_coeffs[I]) * res_coeffs[I]), I)
for I in range(len(res_coeffs))]
res_sq.sort()
self._curr_res_sq_norm = sum(
rmu_sq[0] for rmu_sq in res_sq[:-1]) / (self._dt * self._dt)
print(
' \n--> Begin selection opt with residual magnitudes |r_mu|:')
print(' Initial guess energy: ', round(init_gues_energy, 10))
print(
f' Norm of res vec: {np.sqrt(self._curr_res_sq_norm):14.12f}'
)
self.conv_status()
if not self._converged:
if self._verbose:
print('\n')
print(' op index (Imu) Residual Factor')
print(' -----------------------------------------------')
res_sq_sum = 0.0
if (self._use_cumulative_thresh):
# Make a running list of operators. When the sum of res_sq exceeds the target, every operator
# from here out is getting added to the ansatz..
temp_ops = []
for rmu_sq, op_idx in res_sq[:-1]:
res_sq_sum += rmu_sq / (self._dt * self._dt)
if res_sq_sum > (self._spqe_thresh *
self._spqe_thresh):
if (self._verbose):
Ktemp = self.get_op_from_basis_idx(op_idx)
print(
f" {op_idx:10} {np.real(rmu_sq)/(self._dt * self._dt):14.12f} {Ktemp.str()}"
)
if op_idx not in self._tops:
temp_ops.append(op_idx)
self.add_from_basis_idx(op_idx)
for temp_op in temp_ops[::-1]:
self._tops.insert(0, temp_op)
self._tamps.insert(0, 0.0)
else:
# Add the single operator with greatest rmu_sq not yet in the ansatz
res_sq.reverse()
for rmu_sq, op_idx in res_sq[1:]:
print(
f" {op_idx:10} {np.real(rmu_sq)/(self._dt * self._dt):14.12f}"
)
if op_idx not in self._tops:
print('Adding this operator to ansatz')
self._tops.insert(0, op_idx)
self._tamps.insert(0, 0.0)
self.add_from_basis_idx(op_idx)
break
self._n_classical_params_lst.append(len(self._tops))
def build_qk_mats(self):
"""Returns matrices S and H needed for the QK algorithm using the Trotterized
form of the unitary operators U_n = exp(-i n dt H)
The mathematical operations of this function are unphysical for a quantum
computer, but efficient for a simulator.
Returns
-------
s_mat : ndarray
A numpy array containing the elements S_mn = <Phi | Um^dag Un | Phi>.
_nstates by _nstates
h_mat : ndarray
A numpy array containing the elements H_mn = <Phi | Um^dag H Un | Phi>
_nstates by _nstates
"""
h_mat = np.zeros((self._nstates,self._nstates), dtype=complex)
s_mat = np.zeros((self._nstates,self._nstates), dtype=complex)
# Store these vectors for the aid of MRSQK
self._omega_lst = []
Homega_lst = []
if(self._diagonalize_each_step):
print('\n\n')
print(f"{'k(S)':>7}{'E(Npar)':>19}{'N(params)':>14}{'N(CNOT)':>18}{'N(measure)':>20}")
print('-------------------------------------------------------------------------------')
if (self._print_summary_file):
f = open("summary.dat", "w+", buffering=1)
f.write(f"#{'k(S)':>7}{'E(Npar)':>19}{'N(params)':>14}{'N(CNOT)':>18}{'N(measure)':>20}\n")
f.write('#-------------------------------------------------------------------------------\n')
Hsp = get_scipy_csc_from_op(self._qb_ham, -1.0j)
QC = qforte.Computer(self._nqb)
QC.apply_circuit(self._Uprep)
# get the time evolution vectors
psi_t_vecs = apply_time_evolution_op(QC, Hsp, self._dt, self._nstates)
for m in range(self._nstates):
# # Compute U_m |φ>
QC = qforte.Computer(self._nqb)
QC.set_coeff_vec(psi_t_vecs[m])
self._omega_lst.append(np.asarray(QC.get_coeff_vec(), dtype=complex))
# Compute H U_m |φ>
QC.apply_operator(self._qb_ham)
Homega_lst.append(np.asarray(QC.get_coeff_vec(), dtype=complex))
# Compute S_mn = <φ| U_m^\dagger U_n |φ> and H_mn = <φ| U_m^\dagger H U_n |φ>
for n in range(len(self._omega_lst)):
h_mat[m][n] = np.vdot(self._omega_lst[m], Homega_lst[n])
h_mat[n][m] = np.conj(h_mat[m][n])
s_mat[m][n] = np.vdot(self._omega_lst[m], self._omega_lst[n])
s_mat[n][m] = np.conj(s_mat[m][n])
if (self._diagonalize_each_step):
# TODO (cleanup): have this print to a separate file
k = m+1
evals, evecs = canonical_geig_solve(s_mat[0:k, 0:k],
h_mat[0:k, 0:k],
print_mats=False,
sort_ret_vals=True)
scond = np.linalg.cond(s_mat[0:k, 0:k])
self._n_classical_params = k
self._n_cnot = 0
self._n_pauli_trm_measures = k * self._Nl
self._n_pauli_trm_measures += k * (k-1) * self._Nl
self._n_pauli_trm_measures += k * (k-1)
print(f' {scond:7.2e} {np.real(evals[self._target_root]):+15.9f} {self._n_classical_params:8} {0:10} {self._n_pauli_trm_measures:12}')
if (self._print_summary_file):
f.write(f' {scond:7.2e} {np.real(evals[self._target_root]):+15.9f} {self._n_classical_params:8} {0:10} {self._n_pauli_trm_measures:12}\n')
if (self._diagonalize_each_step and self._print_summary_file):
f.close()
self._n_classical_params = self._nstates
self._n_cnot = 0
# diagonal terms of Hbar
self._n_pauli_trm_measures = self._nstates * self._Nl
# off-diagonal of Hbar (<X> and <Y> of Hadamard test)
self._n_pauli_trm_measures += self._nstates*(self._nstates-1) * self._Nl
# off-diagonal of S (<X> and <Y> of Hadamard test)
self._n_pauli_trm_measures += self._nstates*(self._nstates-1)
return s_mat, h_mat
