def _sp_op(self, i, mat, h_list, mod_term, err=1.e-6):
if not h_list:
return np.zeros((mat.shape))
logging.debug(__('> OP on mat {}...', i))
n, m = mat.shape
partial_transform = self._partial_transform
a = self.get_sub_vec(-1)
a_h = np.conj(a)
density = self._partial_product(i, a, a_h)
inv_density = linalg.inv(density + np.identity(m) * err)
sp = self.get_sub_vec(i)
sp_h = np.conj(np.transpose(sp))
projection = np.identity(n) - np.dot(sp, sp_h)
tmp = partial_transform(i, a, mat)
for mat_j in h_list:
tmp = partial_transform(i, tmp, mat_j)
for j, mat_j in mod_term:
if j != i:
tmp = partial_transform(j, tmp, mat_j)
tmp = self._partial_product(i, tmp, a_h)
ans = np.dot(projection, np.dot(tmp, inv_density))
return ans
def autocorr(self,
steps=None,
ode_inter=0.01,
split=False,
imaginary=False,
fast=False,
start=0,
move_energy=False):
if not fast:
_init = {}
for t in self.root.visitor(leaf=False):
_init[t] = t.array
for time, r in self.propagator(steps=steps,
ode_inter=ode_inter,
split=split,
imaginary=imaginary,
start=start,
move_energy=move_energy):
for t in r.visitor(leaf=False):
t.aux = t.array if fast else np.conj(_init[t])
auto = r.global_inner_product()
ans = (2. * time, auto) if fast else (time, auto)
yield ans
for t in r.visitor(leaf=False):
t.aux = None
def hilbert_angle(r1, r2):
r"""Return :math:`\frac{<2|1>}{\sqrt{<1|1><2|2>}}.
"""
for n1, n2 in zip(r1.visitor(leaf=False), r2.visitor(leaf=False)):
n1.aux = np.conj(n2.array)
angle = np.arccos(r1.global_inner_product() / (r1.global_norm() * r2.global_norm())) / np.pi * 180
return angle
def normalize(self, forced=False):
"""Normalize the array of self. Only work when self.normalized.
Set `forced` to `True` to normalize any way.
"""
array = self.array
if array is None or (not self.normalized and not forced):
return
axis = self.axis
if axis is None:
norm = np.array(self.local_norm())
self.set_array(array / norm)
ans = norm
else:
norm = linalg.norm
shape = self.shape
dim = shape.pop(axis)
array = np.reshape(np.moveaxis(array, axis, 0), (dim, -1))
vecs = []
norm_list = []
for vec_i in array:
for vec_j in vecs:
vec_i -= vec_j * np.dot(np.conj(vec_j), vec_i)
norm_ = norm(vec_i)
vecs.append(vec_i / norm_)
norm_list.append(norm_)
array = np.array(vecs)
array = np.moveaxis(np.reshape(array, [-1] + shape), 0, axis)
self.set_array(array)
ans = norm_list
return ans
def _extend_space(self):
head = len(self._search_space)
self._search_space += self._trial_vecs
self._column_space += list(map(self._matvec, self._trial_vecs))
tail = len(self._search_space)
v, a_v = self._search_space, self._column_space
for i in range(head):
for j in range(head, tail):
self._submatrix[i, j] = np.dot(np.conj(v[i]), a_v[j])
self._submatrix[j, i] = np.conj(self._submatrix[i, j])
for i in range(head, tail):
for j in range(head, i):
self._submatrix[i, j] = np.dot(np.conj(v[i]), a_v[j])
self._submatrix[j, i] = np.conj(self._submatrix[i, j])
self._submatrix[i, i] = np.dot(np.conj(v[i]), a_v[i])
return self._submatrix[:tail, :tail]
def _calc_trial_vecs(self):
if self._precondition is None:
dim = len(self._search_space[0])
self._precondition = self.davidson_precondition(
dim, self._matvec
)
# ritz_vecs = [None] * dim
# else:
# ritz_vecs = list(self._get_ritz_vecs())
self._trial_vecs = []
precondition = self._precondition
zipped = zip(
self._residuals, self._residual_norms,
self._get_ritz_vecs(), self._convergence
)
for residual, norm_, ritz_vec, conv in zipped:
# remove linear dependency in self._residuals
if norm_ ** 2 > self.lin_dep_lim and not conv:
vec = precondition(
residual, self._ritz_vals[0], ritz_vec
)
vec *= 1. / norm(vec)
for base in self._search_space:
vec -= np.dot(np.conj(base), vec) * base
norm_ = norm(vec)
# remove linear dependency between trial_vecs and
# self._search_space
if norm_ ** 2 > self.lin_dep_lim:
vec *= 1. / norm_
self._trial_vecs.append(vec)
return self._trial_vecs
def global_square(self):
"""Return <array|array>
"""
for t in self.visitor(leaf=False):
t.aux = np.conj(t.array)
ans = self.global_inner_product()
return ans
def energy_expection(self, vec=None):
if vec is None:
vec = self.vec
a_tensor = self.get_sub_vec(-1, vec)
mod_terms = self.update_mod_terms(vec=vec, write=False)
ans = 0.
for r, term in enumerate(mod_terms):
h_a = self._coeff_op(a_tensor, term)
h_a = np.reshape(h_a, -1)
a_h = np.conj(np.reshape(a_tensor, -1))
ans += np.dot(a_h, h_a)
return ans
def log_inner_product(self, level=logging.DEBUG):
root = self.root
if logging.root.isEnabledFor(level):
shape_dict = {}
init = root.vectorize(shape_dict=shape_dict)
try:
yield self
except:
pass
else:
if logging.root.isEnabledFor(level):
root.tensorize(np.conj(init), use_aux=True, shape_dict=shape_dict)
ip = root.global_inner_product()
logging.log(level, __("<|>:{}", ip))
def _diff_k(self, k):
c_k = self.corr.symm_coeff[k] + 1.0j * self.corr.asymm_coeff[k]
numberer = self._numberer(k)
raiser = self._raiser(k)
lower = self._lower(k)
return [
[(self._i, -1.0j / self.hbar * np.transpose(self.op)), (k, lower)],
[(self._j, 1.0j / self.hbar * self.op), (k, lower)],
[(self._i, -1.0j / self.hbar * c_k * np.transpose(self.op)),
(k, raiser @ numberer)],
[(self._j, 1.0j / self.hbar * np.conj(c_k) * self.op),
(k, raiser @ numberer)],
]
def _diff_k(self, k):
c_k = self.corr.symm_coeff[k] + 1.0j * self.corr.asymm_coeff[k]
print("k: {}; c_k: {}".format(k, c_k))
numberer = self._sqrt_numberer(k)
raiser = self._raiser(k)
lower = self._lower(k)
return [
[(self._i, -1.0j / self.hbar * self.op), (k, numberer @ lower)],
[(self._j, 1.0j / self.hbar * self.op), (k, numberer @ lower)],
[(self._i, -1.0j / self.hbar * c_k * self.op),
(k, raiser @ numberer)],
[(self._j, 1.0j / self.hbar * np.conj(c_k) * self.op),
(k, raiser @ numberer)],
]
def _single_eom(self, tensor, n, cache=False):
"""C.f. `Multi-Configuration Time Dependent Hartree Theory: a Tensor
Network Perspective`, p38. This method does not contain the `i hbar`
coefficient.
Parameters
----------
tensor : Tensor
Must in a graph with all nodes' array set, including the leaves.
n : int
No. of Hamiltonian term.
Return:
-------
array : ndarray
With the same shape with tensor.shape.
"""
partial_product = Tensor.partial_product
partial_trace = Tensor.partial_trace
partial_env = tensor.partial_env
# Env Hamiltonians
tmp = tensor.array
for i in range(tensor.order):
try:
env_ = self.env_[(n, tensor, i)]
except KeyError:
env_ = partial_env(i, proper=True)
if cache:
self.env_[(n, tensor, i)] = env_
tmp = partial_product(tmp, i, env_)
# For non-root nodes...
if tensor.axis is not None:
# Inversion
axis, inv = tensor.axis, self.inv_density[tensor]
tmp = partial_product(tmp, axis, inv)
# Projection
tmp_1 = np.array(tmp)
array = tensor.array
conj_array = np.conj(array)
tmp = partial_trace(tmp, axis, conj_array, axis)
tmp = partial_product(array, axis, tmp, j=1)
tmp = (tmp_1 - tmp)
return tmp
def partial_env(self, i, proper=False, use_aux=False):
"""
Parameters
----------
i : {int, None}
proper : bool
use_aux : bool
Whether to use self.aux as conj.
Returns
-------
ans : {2-d ndarray, None}
"""
if proper: # Only calculate non-proper subtree directly
# to support the Leaf
child, j = self._access[i]
return child.partial_env(j, proper=False, use_aux=use_aux)
else:
partial_product = Tensor.partial_product
partial_trace = Tensor.partial_trace
# Check the cache
if i in self._partial_env and not use_aux:
return self._partial_env[i]
# Main algorithm
else:
env_ = [(i_, tensor.partial_env(j, proper=False, use_aux=use_aux))
for i_, tensor, j in self.children(axis=i)] # Recursively
# Make use of the normalization condition
if not use_aux and i == self.axis and self.normalized and (all(args[1] is None for args in env_)):
ans = None
else:
temp = self.array
for i_, matrix in env_:
temp = partial_product(temp, i_, matrix)
conj = self.aux if use_aux else np.conj(self.array)
ans = partial_trace(temp, i, conj, i)
# Cache the answer and return
if i is not None:
if not use_aux:
self._partial_env[i] = ans
else:
ans = ans if ans is not None else 1.
return ans
def projector(self, comp=False):
"""[Deprecated] Return the projector corresponding to self.
Returns
-------
ans : ndarray
"""
axis = self.axis
if axis is not None:
array = self.array
shape = self.shape
dim = shape.pop(self.axis)
comp_dim = np.prod(shape)
array = np.moveaxis(array, axis, -1)
array = np.reshape(array, (-1, dim))
array_h = np.conj(np.transpose(array))
ans = np.dot(array, array_h)
if comp:
identity = np.identity(comp_dim)
ans = identity - ans
ans = np.reshape(ans, shape * 2)
return ans
else:
raise RuntimeError('Need to specific the normalization axis!')
def _eval(op, vec):
vec_h = np.conj(np.transpose(vec))
mod_op = np.dot(vec_h, np.dot(op, vec))
return mod_op
def expection(self):
"""Return <array|H|array>
"""
for t in self.visitor(leaf=False):
t.aux = np.conj(t.array)
return self.matrix_element()
def local_norm(self):
self.aux = np.conj(self.array)
ans = self.local_inner_product()
return np.sqrt(ans)
