def iso_contract(T, isoT):
""" applying isometry and its transpose on tensor T, shrinkage T's dimension"""
iso_net = uni10.Network('isometry.net')
res = uni10.UniTensorR(T)
isoT = uni10.UniTensorR(isoT)
T_trans = uni10.UniTensorR(np.transpose(isoT.GetBlock()))
iso_net.PutTensor(0, T)
iso_net.PutTensor(1, isoT)
iso_net.PutTensor(2, T_trans)
iso_net.Launch(res)
return res
def matD():
"""create delta tensor"""
mat = np.zeros(16)
mat[0] = mat[15] = 1
bdout = uni10.Bond(uni10.BD_OUT, 2)
T = uni10.UniTensorR([bdout, bdout, bdout, bdout])
T.SetElem(mat)
return T
def CG_contraction(A, B):
net = uni10.Network("CG_contract.net")
result = uni10.UniTensorR(A)
net.PutTensor(0, A)
net.PutTensor(1, B)
net.Launch(result)
result = result.CombineBond([2, -2])
result = result.CombineBond([4, -4])
return result
def merge(delta, m, mt):
""" initially the tensors are different, so I decide to merge M/MT into delta tensor such that
the tensors are the same, therefore we only need to consider 1 tensor when doing TRG"""
network = uni10.Network("merge_bond.net")
result = uni10.UniTensorR(delta)
network.PutTensor('D', delta)
network.PutTensor("M", m)
network.PutTensor("MT", mt)
network.Launch(result)
return result
def tTrace(Ta, Tb, Tc, Td):
""" return the tensor trace result"""
net = uni10.Network("trace.net")
res = uni10.UniTensorR(Ta)
net.PutTensor(0, Ta)
net.PutTensor(1, Tb)
net.PutTensor(2, Tc)
net.PutTensor(3, Td)
net.Launch(res)
scalar = res.GetBlock()[0]
return scalar
def update(T, chi):
T2 = CG_contraction(T, T)
copy = uni10.UniTensorR(T2)
copy.SetLabel([3, 4, 1, -2])
X = uni10.Contract(T2, copy)
U, norm = svd(
X.GetBlock(),
chi) # use SVD to truncate the tensor with bond dimension chi
result = iso_contract(T2, U)
norm = np.max(result.GetBlock()) / 2
result = uni10.Permute(result, [2, 1, 4, 3], 2)
return result * (1 / norm)
def matMT(beta):
mat = matM(beta).GetBlock()
T = uni10.UniTensorR(np.transpose(mat))
return T
def matM(beta):
mat = np.array([[np.cosh(beta)**(1 / 2),
np.sinh(beta)**(1 / 2)],
[np.cosh(beta)**(1 / 2), -1 * np.sinh(beta)**(1 / 2)]])
T = uni10.UniTensorR(mat)
return T