def test_multiply(self):
# physical quantum numbers
qd = np.random.randint(-2, 3, size=3)
# create random matrix product operators
op0 = ptn.MPO(qd, [
np.random.randint(-2, 3, size=Di)
for Di in [1, 10, 13, 24, 17, 9, 1]
],
fill='random')
op1 = ptn.MPO(qd, [
np.random.randint(-2, 3, size=Di)
for Di in [1, 8, 17, 11, 23, 13, 1]
],
fill='random')
# MPO multiplication (composition)
op = op0 * op1
# reference calculation
op_ref = np.dot(op0.as_matrix(), op1.as_matrix())
# relative error
err = np.linalg.norm(op.as_matrix() - op_ref) / max(
np.linalg.norm(op_ref), 1e-12)
self.assertAlmostEqual(
err,
0.,
delta=1e-14,
msg='composition of two MPOs must agree with matrix representation'
)
def test_add_singlesite(self):
# separate test for a single site since implementation is a special case
# physical quantum numbers
qd = np.random.randint(-2, 3, size=5)
# create random matrix product operators acting on a single site
# leading and trailing (dummy) virtual bond quantum numbers
qD = [np.array([-1]), np.array([-2])]
op0 = ptn.MPO(qd, qD, fill='random')
op1 = ptn.MPO(qd, qD, fill='random')
# MPO addition
op = op0 + op1
# reference calculation
op_ref = op0.as_matrix() + op1.as_matrix()
# relative error
err = np.linalg.norm(op.as_matrix() - op_ref) / max(
np.linalg.norm(op_ref), 1e-12)
self.assertAlmostEqual(
err,
0.,
delta=1e-14,
msg='addition two MPOs must agree with matrix representation')
def test_add(self):
# physical quantum numbers
qd = np.random.randint(-2, 3, size=3)
# create random matrix product operators
qD0 = [
np.random.randint(-2, 3, size=Di)
for Di in [1, 11, 15, 23, 18, 9, 1]
]
qD1 = [
np.random.randint(-2, 3, size=Di)
for Di in [1, 7, 23, 11, 17, 13, 1]
]
# leading and trailing (dummy) virtual bond quantum numbers must agree
qD1[0] = qD0[0].copy()
qD1[-1] = qD0[-1].copy()
op0 = ptn.MPO(qd, qD0, fill='random')
op1 = ptn.MPO(qd, qD1, fill='random')
# MPO addition
op = op0 + op1
# reference calculation
op_ref = op0.as_matrix() + op1.as_matrix()
# relative error
err = np.linalg.norm(op.as_matrix() - op_ref) / max(
np.linalg.norm(op_ref), 1e-12)
self.assertAlmostEqual(
err,
0.,
delta=1e-14,
msg='addition two MPOs must agree with matrix representation')
def cast_to_MPO(mps, qd):
"""
Cast a matrix product state into MPO form by interpreting the physical
dimension as Kronecker product of a pair of dimensions.
"""
assert np.array_equal(mps.qd, ptn.qnumber_flatten([qd, -qd]))
mpo = ptn.MPO(qd, mps.qD, fill=0.0)
for i in range(mps.nsites):
s = mps.A[i].shape
mpo.A[i] = mps.A[i].reshape((len(qd), len(qd), s[1], s[2])).copy()
assert ptn.is_qsparse(mpo.A[i],
[mpo.qd, -mpo.qd, mpo.qD[i], -mpo.qD[i + 1]])
return mpo
def test_operator_average(self):
# physical dimension
d = 3
# physical quantum numbers
qd = np.random.randint(-1, 2, size=d)
# create random matrix product state
D = [1, 7, 26, 19, 25, 8, 1]
psi = ptn.MPS(qd, [np.random.randint(-1, 2, size=Di) for Di in D],
fill='random')
# rescale to achieve norm of order 1
for i in range(psi.nsites):
psi.A[i] *= 5
# create random matrix product operator
D = [1, 5, 16, 14, 17, 4, 1]
# set bond quantum numbers to zero since otherwise,
# sparsity pattern often leads to <psi | op | psi> = 0
op = ptn.MPO(qd, [np.zeros(Di, dtype=int) for Di in D], fill='random')
# calculate average (expectation value) <psi | op | psi>
avr = ptn.operator_average(psi, op)
# reference value based on full Fock space representation
x = psi.as_vector()
avr_ref = np.vdot(x, np.dot(op.as_matrix(), x))
# relative error
err = abs(avr - avr_ref) / max(abs(avr_ref), 1e-12)
self.assertAlmostEqual(
err,
0.,
delta=1e-12,
msg='operator average must match reference value')
def main():
# physical local Hilbert space dimension
d = 2
# number of lattice sites
L = 6
# construct matrix product operator representation of Heisenberg Hamiltonian
J = 1.0
DH = 1.2
h = -0.2
mpoH = ptn.heisenberg_XXZ_MPO(L, J, DH, h)
mpoH.zero_qnumbers()
# realize commutator [H, .] as matrix product operator
mpoHcomm = heisenberg_XXZ_comm_MPO(L, J, DH, h)
mpoHcomm.zero_qnumbers()
print('2-norm of [H, .] operator:', np.linalg.norm(mpoHcomm.as_matrix(),
2))
mpsH = cast_to_MPS(mpoH)
print('mpsH.bond_dims:', mpsH.bond_dims)
print('ptn.norm(mpsH):', ptn.norm(mpsH))
print('[H, .] applied to H as vector (should be zero): {:g}'.format(
np.linalg.norm(np.dot(mpoHcomm.as_matrix(), mpsH.as_vector()))))
# initial MPO with random entries (not necessarily Hermitian)
Dmax = 40
D = np.minimum(
np.minimum(d**(2 * np.arange(L + 1)), d**(2 * (L - np.arange(L + 1)))),
Dmax)
print('D:', D)
np.random.seed(42)
op = ptn.MPO(mpoH.qd, [np.zeros(Di, dtype=int) for Di in D], fill='random')
# effectively clamp virtual bond dimension
for i in range(L):
op.A[i][:, :, 2:, :] = 0
op.A[i][:, :, :, 2:] = 0
op.orthonormalize(mode='right')
op.orthonormalize(mode='left')
# matrix representation
op_mat = op.as_matrix()
# cast into MPS form
psi = cast_to_MPS(op)
print('norm of psi:', np.linalg.norm(psi.as_vector()))
print('psi.bond_dims:', psi.bond_dims)
# check: commutator
comm_ref = np.dot(op_mat, mpoH.as_matrix()) - np.dot(
mpoH.as_matrix(), op_mat)
comm = np.dot(mpoHcomm.as_matrix(), psi.as_vector()).reshape(
(2 * L) * [d]).transpose(
np.concatenate((2 * np.arange(L, dtype=int),
2 * np.arange(L, dtype=int) + 1))).reshape(
(d**L, d**L))
print('commutator reference check error: {:g}'.format(
np.linalg.norm(comm - comm_ref)))
# initial average energy (should be conserved)
en_avr_0 = np.trace(np.dot(mpoH.as_matrix(), op_mat))
en_avr_0_alt = ptn.vdot(mpsH, psi)
print('en_avr_0:', en_avr_0)
print('abs(en_avr_0_alt - en_avr_0):', abs(en_avr_0_alt - en_avr_0))
# exact Schmidt coefficients (singular values) of initial state
sigma_0 = schmidt_coefficients_operator(d, L, op_mat)
plt.semilogy(np.arange(len(sigma_0)) + 1, sigma_0, '.')
plt.xlabel('i')
plt.ylabel('$\sigma_i$')
plt.title('Schmidt coefficients of initial state')
plt.savefig('evolution_operator_schmidt_0.pdf')
plt.show()
print(
'Schmidt coefficients consistency check:',
np.linalg.norm(
sigma_0 -
schmidt_coefficients_wavefunction(d**2, L, psi.as_vector())))
print('entropy of initial state:',
entropy((sigma_0 / np.linalg.norm(sigma_0))**2))
# mixed real and imaginary time evolution
t = 0.1 + 0.5j
# reference calculation: exp(t H) op exp(-t H)
op_t_ref = np.dot(np.dot(expm(t * mpoH.as_matrix()), op_mat),
expm(-t * mpoH.as_matrix()))
# Frobenius norm not preserved by mixed real and imaginary time evolution
print('Frobenius norm of initial op: {:g}'.format(
np.linalg.norm(op_mat, 'fro')))
print('Frobenius norm of op(t): {:g}'.format(
np.linalg.norm(op_t_ref, 'fro')))
# energy should be conserved
en_avr_t_ref = np.trace(np.dot(mpoH.as_matrix(), op_t_ref))
print('en_avr_t_ref:', en_avr_t_ref)
print('abs(en_avr_t_ref - en_avr_0):', abs(en_avr_t_ref - en_avr_0))
# exact Schmidt coefficients (singular values) of time-evolved state
sigma_t = schmidt_coefficients_operator(d, L, op_t_ref)
plt.semilogy(np.arange(len(sigma_t)) + 1, sigma_t, '.')
plt.xlabel('i')
plt.ylabel(r'$\sigma_i$')
plt.title(
'Schmidt coefficients of time-evolved state (t = {:g})\n(based on exact time evolution)'
.format(-1j * t))
plt.savefig('evolution_operator_schmidt_t.pdf')
plt.show()
S = entropy((sigma_t / np.linalg.norm(sigma_t))**2)
print('entropy of time-evolved state:', S)
P = tangent_space_projector(psi)
print(
'np.linalg.norm(np.dot(P, mpsH.as_vector()) - mpsH.as_vector()) (in general non-zero):',
np.linalg.norm(np.dot(P, mpsH.as_vector()) - mpsH.as_vector()))
# number of time steps
numsteps = 2**(np.arange(5))
err_op = np.zeros(len(numsteps))
# relative energy error
err_en = np.zeros(len(numsteps))
for i, n in enumerate(numsteps):
print('n:', n)
# time step
dt = t / n
psi_t = copy.deepcopy(psi)
ptn.integrate_local_singlesite(mpoHcomm,
psi_t,
dt,
n,
numiter_lanczos=20)
op_t = cast_to_MPO(psi_t, op.qd)
err_op[i] = np.linalg.norm(op_t.as_matrix() - op_t_ref, ord=1)
en_avr_t = np.trace(np.dot(mpoH.as_matrix(), op_t.as_matrix()))
# relative energy error
err_en[i] = abs(en_avr_t - en_avr_0) / abs(en_avr_0)
dtinv = numsteps / abs(t)
plt.loglog(dtinv, err_op, '.-')
# show quadratic scaling for comparison
plt.loglog(dtinv, 1.75e-2 / dtinv**2, '--')
plt.xlabel('1/dt')
plt.ylabel(
r'$\Vert\mathcal{O}[A](t) - \mathcal{O}_\mathrm{ref}(t)\Vert_1$')
plt.title(
'TDVP time evolution (applied to operator) rate of convergence for\nHeisenberg XXZ model (J={:g}, D={:g}, h={:g}), L={}, t={:g}'
.format(J, DH, h, L, -1j * t))
plt.savefig('evolution_operator_convergence.pdf')
plt.show()
plt.loglog(dtinv, err_en, '.-')
# show quadratic scaling for comparison
plt.loglog(dtinv, 1e-2 / dtinv**2, '--')
plt.xlabel('1/dt')
plt.ylabel(
r'$\frac{\vert\epsilon(t) - \epsilon(0)\vert}{\vert\epsilon(0)\vert}, \quad \epsilon(t) = \mathrm{tr}[H \mathcal{O}[A](t)]$'
)
plt.title(
'TDVP time evolution (applied to operator) relative energy error for\nHeisenberg XXZ model (J={:g}, D={:g}, h={:g}), L={}, t={:g}'
.format(J, DH, h, L, -1j * t))
plt.savefig('evolution_operator_energy_conv.pdf')
plt.show()
def test_orthonormalization(self):
# create random matrix product operator
d = 4
D = [1, 10, 13, 14, 7, 1]
mpo0 = ptn.MPO(np.random.randint(-2, 3, size=d),
[np.random.randint(-2, 3, size=Di) for Di in D],
fill='random')
self.assertEqual(mpo0.bond_dims, D, msg='virtual bond dimensions')
# density matrix on full Hilbert space
rho = mpo0.as_matrix()
# performing left-orthonormalization...
cL = mpo0.orthonormalize(mode='left')
self.assertLessEqual(
mpo0.bond_dims[1],
d**2,
msg=
'virtual bond dimension can only increase by factor of "d^2" per site'
)
for i in range(mpo0.nsites):
self.assertTrue(
ptn.is_qsparse(
mpo0.A[i],
[mpo0.qd, -mpo0.qd, mpo0.qD[i], -mpo0.qD[i + 1]]),
msg='sparsity pattern of MPO tensors must match quantum numbers'
)
rhoL = mpo0.as_matrix()
# density matrix should now be normalized
self.assertAlmostEqual(np.linalg.norm(rhoL, 'fro'),
1.,
delta=1e-12,
msg='density matrix normalization')
# density matrices before and after left-normalization must match
# (up to normalization factor)
self.assertAlmostEqual(
np.linalg.norm(cL * rhoL - rho),
0.,
delta=1e-10,
msg=
'density matrices before and after left-normalization must match')
# check left-orthonormalization
for i in range(mpo0.nsites):
s = mpo0.A[i].shape
assert s[0] == d and s[1] == d
Q = mpo0.A[i].reshape((s[0] * s[1] * s[2], s[3]))
QH_Q = np.dot(Q.conj().T, Q)
self.assertAlmostEqual(np.linalg.norm(QH_Q - np.identity(s[3])),
0.,
delta=1e-12,
msg='left-orthonormalization')
# performing right-orthonormalization...
cR = mpo0.orthonormalize(mode='right')
self.assertLessEqual(
mpo0.bond_dims[-2],
d**2,
msg=
'virtual bond dimension can only increase by factor of "d^2" per site'
)
for i in range(mpo0.nsites):
self.assertTrue(
ptn.is_qsparse(
mpo0.A[i],
[mpo0.qd, -mpo0.qd, mpo0.qD[i], -mpo0.qD[i + 1]]),
msg='sparsity pattern of MPO tensors must match quantum numbers'
)
self.assertAlmostEqual(
abs(cR),
1.,
delta=1e-12,
msg=
'normalization factor must have magnitude 1 due to previous left-orthonormalization'
)
rhoR = mpo0.as_matrix()
# density matrices must match
self.assertAlmostEqual(
np.linalg.norm(rhoL - cR * rhoR),
0.,
delta=1e-10,
msg=
'density matrices after left- and right-orthonormalization must match'
)
# check right-orthonormalization
for i in range(mpo0.nsites):
s = mpo0.A[i].shape
assert s[0] == d and s[1] == d
Q = mpo0.A[i].transpose((0, 1, 3, 2)).reshape(
(s[0] * s[1] * s[3], s[2]))
QH_Q = np.dot(Q.conj().T, Q)
self.assertAlmostEqual(np.linalg.norm(QH_Q - np.identity(s[2])),
0.,
delta=1e-12,
msg='right-orthonormalization')
