def test_evolve_trotter_euclidean(num_sites, phys_dim, graph):
psi = tf.complex(
tf.random_normal([phys_dim] * num_sites, dtype=tf.float64),
tf.random_normal([phys_dim] * num_sites, dtype=tf.float64))
h = tf.complex(
tf.random_normal((phys_dim**2, phys_dim**2), dtype=tf.float64),
tf.random_normal((phys_dim**2, phys_dim**2), dtype=tf.float64))
h = 0.5 * (h + tf.linalg.adjoint(h))
h = tf.reshape(h, (phys_dim, phys_dim, phys_dim, phys_dim))
H = [h] * (num_sites - 1)
norm1 = wavefunctions.inner(psi, psi)
en1 = sum(wavefunctions.expval(psi, H[i], i) for i in range(num_sites - 1))
if graph:
psi, t = wavefunctions.evolve_trotter_defun(psi,
H,
0.1,
10,
euclidean=True)
else:
psi, t = wavefunctions.evolve_trotter(psi, H, 0.1, 10, euclidean=True)
norm2 = wavefunctions.inner(psi, psi)
en2 = sum(wavefunctions.expval(psi, H[i], i) for i in range(num_sites - 1))
np.testing.assert_allclose(t, 1.0)
np.testing.assert_almost_equal(norm2, 1.0)
assert en2.numpy() / norm2.numpy() < en1.numpy() / norm1.numpy()
def test_evolve_trotter(num_sites, phys_dim, graph):
psi = tf.complex(
tf.random.normal([phys_dim] * num_sites, dtype=tf.float64),
tf.random.normal([phys_dim] * num_sites, dtype=tf.float64))
h = tf.complex(
tf.random.normal((phys_dim**2, phys_dim**2), dtype=tf.float64),
tf.random.normal((phys_dim**2, phys_dim**2), dtype=tf.float64))
h = 0.5 * (h + tf.linalg.adjoint(h))
h = tf.reshape(h, (phys_dim, phys_dim, phys_dim, phys_dim))
H = [h] * (num_sites - 1)
norm1 = wavefunctions.inner(psi, psi)
en1 = sum(wavefunctions.expval(psi, H[i], i) for i in range(num_sites - 1))
if graph:
psi, t = wavefunctions.evolve_trotter_defun(psi, H, 0.001, 10)
else:
psi, t = wavefunctions.evolve_trotter(psi, H, 0.001, 10)
norm2 = wavefunctions.inner(psi, psi)
en2 = sum(wavefunctions.expval(psi, H[i], i) for i in range(num_sites - 1))
np.testing.assert_allclose(t, 0.01)
np.testing.assert_almost_equal(norm1 / norm2, 1.0)
np.testing.assert_almost_equal(en1 / en2, 1.0, decimal=2)
def test_expval(num_sites):
op = tf.convert_to_tensor(np.array([[1.0, 0.0], [0.0, -1.0]]))
for j in range(num_sites):
psi = np.zeros([2] * num_sites)
psi_vec = psi.reshape((2**num_sites, ))
psi_vec[2**j] = 1.0
psi = tf.convert_to_tensor(psi)
for i in range(num_sites):
res = wavefunctions.expval(psi, op, i)
if i == num_sites - 1 - j:
np.testing.assert_allclose(res, -1.0)
else:
np.testing.assert_allclose(res, 1.0)
def test_opt(backend):
if backend == "tensorflow":
dtype = tf.float64
else:
dtype = np.float64
backend_obj = ttn.get_backend()
num_layers = 3
max_bond_dim = 8
build_graphs = True
num_sweeps = 5
Ds = [min(2**i, max_bond_dim) for i in range(1, num_layers + 1)]
H = ttn.get_ham_ising(dtype)
isos_012 = ttn.random_tree_tn_uniform(Ds, dtype, top_rank=1)
energy_0 = backend_obj.trace(ttn.top_hamiltonian(H, isos_012))
isos_012 = ttn.opt_tree_energy(isos_012,
H,
num_sweeps,
1,
verbose=0,
graphed=build_graphs,
ham_shift=0.2)
energy_1 = backend_obj.trace(ttn.top_hamiltonian(H, isos_012))
assert backend_obj.to_numpy(energy_1) < backend_obj.to_numpy(energy_0)
N = 2**num_layers
full_state = ttn.descend_full_state_pure(isos_012)
norm = backend_obj.norm(backend_obj.reshape(full_state, (2**N, )))
assert abs(backend_obj.to_numpy(norm) - 1) < 1e-12
if backend != "jax":
# wavefunctions assumes TensorFlow. This will interact with numpy OK, but
# not JAX.
h = ttn.ttn_1d_uniform._dense_ham_term(H)
energy_1_full_state = sum(
wf.expval(full_state, h, j, pbc=True) for j in range(N))
assert abs(
backend_obj.to_numpy(energy_1_full_state) -
backend_obj.to_numpy(energy_1)) < 1e-12
isos_012 = ttn.opt_tree_energy(isos_012,
H,
1,
1,
verbose=0,
graphed=False,
decomp_mode="eigh",
ham_shift=0.2)
for iso in isos_012:
assert backend_obj.to_numpy(ttn.check_iso(iso)) < 1e-6
isos_012 = ttn.opt_tree_energy(isos_012,
H,
1,
1,
verbose=0,
graphed=False,
decomp_mode="svd",
ham_shift=0.2)
for iso in isos_012:
assert backend_obj.to_numpy(ttn.check_iso(iso)) < 1e-6
isos_012 = ttn.opt_tree_energy(isos_012,
H,
1,
1,
verbose=0,
graphed=False,
decomp_mode="svd_full_iso_scipy",
ham_shift=0.2)
for iso in isos_012:
assert backend_obj.to_numpy(ttn.check_iso(iso)) < 1e-12
def callback(psi, t, i):
print(i,
tf.norm(psi).numpy().real,
wavefunctions.expval(psi, X, 0).numpy().real)
def callback(psi, t, i):
print(
i,
tf.norm(psi).numpy().real,
wavefunctions.expval(psi, X, 0).numpy().real
)
