def calculate_divided_entropy(self, dot_index='all'):
if operator.eq(dot_index, 'all'):
dot_index = list(range(self.tensor_info['n_length']))
elif isinstance(dot_index, int):
dot_index = [dot_index]
dot_index = tc.tensor(dot_index).reshape(-1)
dot_index_sorted = dot_index.sort()
entropy = dict()
for ii in dot_index_sorted[0]:
self.mps_regularization(ii)
u, s, v = wf.tensor_svd(self.tensor_data[ii], index_right=2)
s = s[s > 0]
s /= s.norm()
s_tmp = s ** 2
entropy[ii] = -tc.einsum('a,a->', s_tmp, tc.log(s_tmp))
return entropy
def move_regular_center2forward(self):
# self.reverse_mps()
# self.move_regular_center2next()
# self.reverse_mps()
tensor_index = self.tensor_info['regular_center']
u, s, v = wf.tensor_svd(self.tensor_data[tensor_index], index_left=0)
if self.tensor_info['normalization_mode'] == 'on':
s /= s.norm()
s = s[s > self.tensor_info['cutoff']]
dimension_middle = min([len(s), self.tensor_info['regular_bond_dimension']])
u = u[:, 0:dimension_middle].reshape(-1, dimension_middle)
s = s[0:dimension_middle]
v = v[:, 0:dimension_middle].reshape(-1, dimension_middle)
self.tensor_data[tensor_index] = v.t().reshape(
dimension_middle, self.tensor_data[tensor_index].shape[1],
self.tensor_data[tensor_index].shape[2])
self.tensor_data[tensor_index - 1] = torch.einsum(
'ijk,kl->ijl',
[self.tensor_data[tensor_index - 1], u.mm(torch.diag(s))])
self.tensor_info['regular_center'] -= 1
def move_regular_center2next(self):
# self.reverse_mps()
# self.move_regular_center2forward()
# self.reverse_mps()
tensor_index = self.tensor_info['regular_center']
u, s, v = wf.tensor_svd(self.tensor_data[tensor_index], index_right=2)
if self.tensor_info['normalization_mode'] == 'on':
s /= s.norm()
s = s[s > self.tensor_info['cutoff']]
dimension_middle = min([len(s), self.tensor_info['regular_bond_dimension']])
u = u[:, 0:dimension_middle].reshape(-1, dimension_middle)
s = s[0:dimension_middle]
v = v[:, 0:dimension_middle].reshape(-1, dimension_middle)
self.tensor_data[tensor_index] = u.reshape(
self.tensor_data[tensor_index].shape[0],
self.tensor_data[tensor_index].shape[1], dimension_middle)
self.tensor_data[tensor_index + 1] = torch.einsum(
'ij,jkl->ikl',
[(torch.diag(s)).mm(v.t()), self.tensor_data[tensor_index+1]])
self.tensor_info['regular_center'] += 1