def test_mpo_identity(self):
k = MPS_rand_state(13, 7)
b = MPS_rand_state(13, 7)
o1 = k @ b
i = MPO_identity(13)
k, i, b = align_TN_1D(k, i, b)
o2 = (k & i & b) ^ ...
assert_allclose(o1, o2)
def rand_tn1d_sect(n, bd, dtype=complex):
mps = qtn.MPS_rand_state(n + 2, bd, dtype=dtype)
mpo = qtn.MPO_rand_herm(n + 2, 5, dtype=dtype)
norm = qtn.TensorNetwork(qtn.align_TN_1D(mps.H, mpo, mps))
lix = qtn.bonds(norm[0], norm[1])
rix = qtn.bonds(norm[n], norm[n + 1])
to = norm[1:n + 1]
return qtn.TNLinearOperator1D(to, lix, rix, 1, n + 1)
def test_single_explicit_sweep(self):
h = MPO_ham_heis(5)
dmrg = DMRG1(h, bond_dims=3)
assert dmrg._k[0].dtype == float
energy_tn = (dmrg._b | dmrg.ham | dmrg._k)
e0 = energy_tn ^ ...
assert abs(e0.imag) < 1e-13
de1 = dmrg.sweep_right()
e1 = energy_tn ^ ...
assert_allclose(de1, e1)
assert abs(e1.imag) < 1e-13
de2 = dmrg.sweep_right()
e2 = energy_tn ^ ...
assert_allclose(de2, e2)
assert abs(e2.imag) < 1e-13
# state is already left canonized after right sweep
de3 = dmrg.sweep_left(canonize=False)
e3 = energy_tn ^ ...
assert_allclose(de3, e3)
assert abs(e2.imag) < 1e-13
de4 = dmrg.sweep_left()
e4 = energy_tn ^ ...
assert_allclose(de4, e4)
assert abs(e2.imag) < 1e-13
# test still normalized
assert dmrg._k[0].dtype == float
align_TN_1D(dmrg._k, dmrg._b, inplace=True)
assert_allclose(abs(dmrg._b @ dmrg._k), 1)
assert e1.real < e0.real
assert e2.real < e1.real
assert e3.real < e2.real
assert e4.real < e3.real
def test_explicit_sweeps(self):
# import pdb; pdb.set_trace()
n = 8
chi = 16
ham = MPO_ham_mbl(n, dh=5, run=42)
p0 = MPS_neel_state(n).expand_bond_dimension(chi)
b0 = p0.H
align_TN_1D(p0, ham, b0, inplace=True)
en0 = np.asscalar(p0 & ham & b0 ^ ...)
dmrgx = DMRGX(ham, p0, chi)
dmrgx.sweep_right()
en1 = dmrgx.sweep_left(canonize=False)
assert en0 != en1
dmrgx.sweep_right(canonize=False)
en = dmrgx.sweep_right(canonize=True)
# check normalized
assert_allclose(dmrgx._k.H @ dmrgx._k, 1.0)
k = dmrgx._k.to_dense()
h = ham.to_dense()
el, ev = eigsys(h)
# check variance very low
assert np.abs((k.H @ h @ h @ k) - (k.H @ h @ k)**2) < 1e-12
# check exactly one eigenvalue matched well
assert np.sum(np.abs(el - en) < 1e-12) == 1
# check exactly one eigenvector is matched with high fidelity
ovlps = (ev.H @ k).A**2
big_ovlps = ovlps[ovlps > 1e-12]
assert_allclose(big_ovlps, [1])
# check fully
assert is_eigenvector(k, h)
def rand_tn1d_sect(n, bd, dtype=complex):
mps = qtn.MPS_rand_state(n + 2, bd, dtype=dtype)
mpo = qtn.MPO_rand_herm(n + 2, 5, dtype=dtype)
norm = qtn.TensorNetwork(qtn.align_TN_1D(mps.H, mpo, mps))
# greedy not good and contracting with large bsz
norm.structure_bsz = 2
lix = qtn.bonds(norm[0], norm[1])
rix = qtn.bonds(norm[n], norm[n + 1])
to = norm[1:n + 1]
return qtn.TNLinearOperator1D(to, lix, rix, 1, n + 1)
def test_sites_mpo_mps_product(self, cyclic):
k = MPS_rand_state(13, 7, cyclic=cyclic)
X = MPO_rand_herm(3, 5, sites=[3, 6, 7], nsites=13, cyclic=cyclic)
b = k.H
align_TN_1D(k, X, b, inplace=True)
assert (k & X & b) ^ ...
def loss_fn(psi, H):
k, H, b = qtn.align_TN_1D(psi, H, psi.H)
energy = (k & H & b).contract(all, optimize='random-greedy')
return real(energy)