def run(n, num_iter, num_batch):
tn.createTensor("C", np.zeros((n, n)))
tn.createTensor("A", np.random.rand(n, n))
tn.createTensor("B", np.random.rand(n, n))
nops = 2 * n**3 / 1e9
elapsed = benchmark(num_iter, num_batch, foo)
tn.getLocalTensor("C")
tn.destroyTensor("A")
tn.destroyTensor("B")
tn.destroyTensor("C")
print(n, ", ", nops / elapsed)
def test_exatn():
a1 = np.random.randn(2, 3, 4)
print('A1 shape: ', a1.shape)
b1 = np.random.randn(2, 3, 4, 5)
print('B1 shape: ', b1.shape)
exatn.createTensor('C1', [3, 4, 5], 0.0)
exatn.createTensor('A1', a1.copy(order='F'))
exatn.createTensor('B1', b1.copy(order='F'))
exatn.contractTensors('C1(b,c,d)=A1(a,b,c)*B1(a,b,c,d)', 1.0)
c1 = exatn.getLocalTensor('C1')
print('C1 shape: ', c1.shape)
print('ExaTN result c1 = a1 * b1:\n', c1)
d1 = np.einsum('abc, abcd->bcd', a1, b1)
print('D1 shape: ', d1.shape)
print('NumPy result c1 = a1 * b1:\n', d1)
exatn.destroyTensor('B1')
exatn.destroyTensor('A1')
exatn.destroyTensor('C1')
def test_exatn():
a1 = np.array([
[1., 0., 0.],
[0., 1., 1.]])
print('A1 shape: ',a1.shape)
b1 = np.array([
[ 1., 0., 3., 0.],
[ 1., 1., 2., 2.],
[-1., 1., -2., 2.]])
print('B1 shape: ',b1.shape)
exatn.createTensor('C1', [2, 4], 0.0)
exatn.createTensor('A1', a1.copy(order='F'))
exatn.createTensor('B1', b1.copy(order='F'))
exatn.contractTensors('C1(a,c)=A1(a,b)*B1(b,c)',1.0)
c1 = exatn.getLocalTensor('C1')
print('C1 shape: ',c1.shape)
d1 = np.dot(a1, b1)
print('D1 shape: ',d1.shape)
print('NumPy result c1 = a1 * b1:\n', d1)
print('ExaTN result c1 = a1 * b1:\n', c1)
exatn.destroyTensor('B1')
exatn.destroyTensor('A1')
exatn.destroyTensor('C1')
# Print S, R
exatn.print("S")
exatn.print("R")
# Demonstrate transformTensor interface by
# negating the tensor data in S
def negate(data):
data *= -1.
exatn.transformTensor("S", negate)
exatn.print("S")
# Compute the dot product, store in Z0
exatn.evaluateTensorNetwork('MyTN', 'Z0() = S(a) * R(a)')
exatn.print('Z0')
z0_data = exatn.getLocalTensor('Z0')
# Compare to what numpy would have gotten
print(np.dot(-s, r))
print('z0 = ', z0_data)
print(exatn.getLocalTensor('S'))
print(exatn.getLocalTensor('S').shape)
# Clean up and destroy
exatn.destroyTensor("S")
exatn.destroyTensor("R")
exatn.destroyTensor("Z0")
# Merge X and Y
pattern = tNet.mergeTensors(1, 2, 4)
print("After merge:")
tNet.printIt()
# Print the generic merge pattern
print(pattern)
# Create the merged tensor
pattern = pattern.replace("D", tNet.getTensor(4).getName())
pattern = pattern.replace("L", "X")
pattern = pattern.replace("R", "Y")
print(pattern)
# Perform calculation
exatn.createTensor(tNet.getTensor(4))
exatn.contractTensors(pattern)
# Evaluate the tensor network (after merging two tensors)
exatn.evaluate(tNet)
# Print root tensor
root = exatn.getLocalTensor(tNet.getTensor(0).getName())
print(root)
# Evaluate the *Original* network to make sure it is the same.
tNetOriginal.printIt()
exatn.evaluate(tNetOriginal)
rootOriginal = exatn.getLocalTensor(tNetOriginal.getTensor(0).getName())
print(rootOriginal)
assert(np.allclose(root, rootOriginal))
qubitReg.rename('QubitKet')
[circuit.appendTensorGate(tensorCounter+c, 'H', [i]) for i, c in enumerate(range(nQubits))]
tensorCounter += nQubits
circuit.printIt()
inverse = exatn.TensorNetwork(circuit)
inverse.rename('InverseCircuit')
[inverse.appendTensorGate(tensorCounter+c, 'H', [nQubits - i - 1], True) for i,c in enumerate(range(nQubits))]
tensorCounter += nQubits
assert(inverse.collapseIsometries())
inverse.printIt()
bra = qubitReg
bra.conjugate()
bra.rename('QubitBra')
pairings = [[i,i] for i in range(nQubits)]
inverse.appendTensorNetwork(bra, pairings)
inverse.printIt()
assert(inverse.getRank() == 0)
assert(exatn.evaluate(inverse))
print(exatn.getLocalTensor(inverse.getTensor(0).getName()))
import sys
from pathlib import Path
sys.path.insert(1, str(Path.home()) + '/.exatn')
import exatn, numpy as np
qzero = np.array([1.0, 0.0], dtype=complex)
unitary = np.reshape(np.array([1, -1j, -1j, 1], dtype=complex), (2, 2))
exatn.createTensor('Q0', qzero)
exatn.createTensor('U', unitary)
exatn.registerTensorIsometry('U', [0], [1])
circuit = exatn.TensorNetwork('QuantumCircuit')
circuit.appendTensor(1, 'Q0')
circuit.appendTensorGate(2, 'U', [0])
circuit.printIt()
conj_circuit = exatn.TensorNetwork(circuit)
conj_circuit.rename('ConjugatedCircuit')
conj_circuit.conjugate()
conj_circuit.printIt()
assert (exatn.evaluate(circuit))
assert (exatn.evaluate(conj_circuit))
print(exatn.getLocalTensor(circuit.getTensor(0).getName()))
print(exatn.getLocalTensor(conj_circuit.getTensor(0).getName()))
def conjugate(tName):
exatn.createTensor(tName + 'conj', np.conj(exatn.getLocalTensor(tName)))
0.28699706946011383, 0.3217444561240923, -0.11511191774971147
])
I = exatn.createTensor('I', np.reshape(SSU, (4, 2, 2)))
conjugate('I')
#test I is isometry, II should be identity
exatn.createTensor('II', [4, 4], 0 + 0j)
exatn.contractTensors('II(a,b)=I(a,i,j)*Iconj(b,i,j)')
print('I doubled')
exatn.print('II')
rho = np.reshape(chosen, (4, 4))
w, sig1, e = np.linalg.svd(rho)[0], np.linalg.svd(rho)[1], np.linalg.svd(
rho)[2]
w, e = np.reshape(w, (2, 2, 4)), np.reshape(e, (4, 2, 2))
wi = np.tensordot(w.copy(), exatn.getLocalTensor('I').copy(), axes=((2), (0)))
ei = np.tensordot(e.copy(), exatn.getLocalTensor('I').copy(), axes=((0), (0)))
print('WI')
print(wi)
print('differences')
print(np.linalg.norm(wi - np.transpose(np.conjugate(wi), (1, 0, 3, 2))))
print(np.linalg.norm(ei - np.transpose(np.conjugate(ei), (1, 0, 3, 2))))
wim = np.transpose(wi, (0, 2, 1, 3)).reshape(4, 4)
eim = np.transpose(ei, (0, 2, 1, 3)).reshape(4, 4)
svdw = np.linalg.svd(wim)
svde = np.linalg.svd(eim)
print('sigmaw, sigmae, nw, sw, ne, se:')
print(svdw[1].round(5))
print(svde[1].round(5))
import sys
from pathlib import Path
sys.path.insert(1, str(Path.home()) + '/.exatn')
import exatn
exatn.createTensor('Z0')
exatn.createTensor('T0', [2,2], .01)
exatn.createTensor('T1', [2,2,2], .01)
exatn.createTensor('T2', [2,2], .01)
exatn.createTensor('H0', [2,2,2,2], .01)
exatn.createTensor('S0', [2,2], .01)
exatn.createTensor('S1', [2,2,2], .01)
exatn.createTensor('S2', [2,2], .01)
exatn.evaluateTensorNetwork('{0,1} 3-site MPS closure', 'Z0() = T0(a,b) * T1(b,c,d) * T2(d,e) * H0(a,c,f,g) * S0(f,h) * S1(h,g,i) * S2(i,e)')
z0 = exatn.getLocalTensor('Z0')
assert(abs(z0 - 5.12e-12) < 1e-12)
print(z0)
closed_prod = exatn.TensorExpansion(ham_ket, bra)
closed_prod.rename('MPSbraHamMPSket')
closed_prod.printIt()
# Declare the derivative tensor expansion with respect to tensor Q1+:
# Q0----Q1----Q2----Q3 Q0----Q1----Q2----Q3 Q0----Q1----Q2----Q3
# | | | | | | | | | | | |
# ==H01== | | + | ==H12== | + | | ==H23==
# | | | | | | | | | | | |
# Q0-- --Q2----Q3 Q0-- --Q2----Q3 Q0-- --Q2----Q3
deriv_q1 = exatn.TensorExpansion(closed_prod, 'Q1', True)
deriv_q1.rename('DerivativeQ1')
# Create the Accumulator tensor for the closed tensor expansion:
exatn.createTensor('AC0')
accumulator0 = exatn.getTensor('AC0')
exatn.createTensor('AC1', [2, 2, 4], 0.0)
accumulator1 = exatn.getTensor('AC1')
# Evaluate the expectation value:
exatn.evaluate(closed_prod, accumulator0)
# Evaluate the derivative of the expectation value w.r.t. tensor Q1:
exatn.evaluate(deriv_q1, accumulator1)
# Print the expectation values
print(exatn.getLocalTensor('AC0'))
[
print(exatn.getLocalTensor(c.network.getTensor(0).getName()))
for c in closed_prod
]
