How to get the state vector in SGB mode?

[fqian5]

I was trying to get state vector of the wavefunction of the MPS in SGB mode. Is there a way to do this?

import numpy as np
from pyblock2.driver.core import DMRGDriver, SymmetryTypes



H_dict_3 = {
  'IIII': -0.8121706072487134,
 'ZIII': 0.171412826447769,
 'IIZI': 0.171412826447769,
 'IZII': -0.22343153690813447,
 'IIIZ': -0.22343153690813447,
 'ZIZI': 0.16868898170361207,
 'ZZII': 0.12062523483390415,
 'IIZZ': 0.12062523483390415,
 'IZZI': 0.1659278503377034,
 'ZIIZ': 0.1659278503377034,
 'YXXY': 0.04530261550379927,
 'YXYX': -0.04530261550379927,
 'XYXY': -0.04530261550379927,
 'XYYX': 0.04530261550379927,
 'IZIZ': 0.17441287612261586}

n_qubits =  4
n_electrons = 2
twoS = 0

pauli = list(H_dict_3.items())
#print(pauli)
driver_obj = DMRGDriver(symm_type=SymmetryTypes.SGB, n_threads=4)

# https://block2.readthedocs.io/en/latest/api/pyblock2.html#pyblock2.driver.core.DMRGDriver.initialize_system
driver_obj.initialize_system(
    n_sites=n_qubits,
      # n_elec=n_electrons,
      # spin=twoS,heis_twos=1,
      pauli_mode=True,
      # hamil_init=False #<-- default is true... but not sure this is correct
      )

mpo = driver_obj.get_mpo_any_pauli(pauli)


ket = driver_obj.get_random_mps(tag="GS", bond_dim=2, nroots=1)

bond_dims = [2]* 4 + [20] * 4
noises = [1e-3] * 4 + [1e-3] * 4 + [0]
thrds = [1e-13] * 8
energies = driver_obj.dmrg(mpo, ket, n_sweeps=20, bond_dims=bond_dims, noises=noises,thrds=thrds, iprint=1)

from pyblock3.block2.io import MPSTools, MPOTools,SubTensor
def get_mps_tensor(ket):
    pyket = MPSTools.from_block2(ket)
    n_sites = pyket.n_sites
    ket = pyket.tensors[0]
    for i in range(n_sites-1):
        ket = np.tensordot(ket,pyket.tensors[i+1],axes = 1)
    return ket

vector = get_mps_tensor(ket)

The final line gives the following error:

TypeError: __init__(): incompatible constructor arguments. The following argument types are supported:
    1. block2.sz.UnfusedMPS()
    2. block2.sz.UnfusedMPS(arg0: block2.sz.MPS)

Invoked with: <block2.sgb.MPS object at 0x16b9bf070>

[fqian5]

A related question is whether there is a way to turn a Hamiltonian in SBG to SU2 or define a Pauli Hamiltonian directly in SU2 mode?

Thanks!

[hczhai]

I was trying to get state vector of the wavefunction of the MPS in SGB mode. Is there a way to do this?

As pyblock3 is a package mostly for fermion problems, it does not accept the SGB mode in block2. You can use pyblock2.algebra.io instead of pyblock3.block2.io, which works for most symmetry modes in block2.

So the final part of your script should be changed to the following:

from pyblock2.algebra.io import MPSTools

def get_mps_tensor(ket):
    ket = driver_obj.adjust_mps(ket, dot=1)[0]
    pyket = MPSTools.from_block2(ket)
    tensors = [x.blocks[0].reduced for x in pyket.tensors]
    ket = tensors[0]
    for i in range(len(tensors) - 1):
        ket = np.tensordot(ket, tensors[i + 1], axes=1)
    return ket.reshape((2, ) * len(tensors))

vector = get_mps_tensor(ket)
assert abs(np.linalg.norm(vector) - 1.0) < 1E-10

A related question is whether there is a way to turn a Hamiltonian in SBG to SU2 or define a Pauli Hamiltonian directly in SU2 mode?

Not all Pauli Hamiltonians have SU2 symmetries. In block2 we do not have tools to translate Hamiltonian from spin symmetry to SU2 symmetries. For Heisenberg model, as explained in https://block2.readthedocs.io/en/latest/tutorial/heisenberg.html#The-SU2-Mode, you have to first rewrite the Hamiltonian in SU2-symmetry-conserving operators, then you can use block2 to compute it.

[fqian5]

Thanks for your reply! I tried the code you provided and got this error message:

[/Users/qianfeng/Desktop/mps_prep/Untitled.ipynb](https://file+.vscode-resource.vscode-cdn.net/Users/qianfeng/Desktop/mps_prep/Untitled.ipynb) Cell 14 line 6
     [62](vscode-notebook-cell:/Users/qianfeng/Desktop/mps_prep/Untitled.ipynb#X26sZmlsZQ%3D%3D?line=61) def get_mps_tensor(ket):
     [63](vscode-notebook-cell:/Users/qianfeng/Desktop/mps_prep/Untitled.ipynb#X26sZmlsZQ%3D%3D?line=62)     ket = driver_obj.adjust_mps(ket, dot=1)[0]
---> [64](vscode-notebook-cell:/Users/qianfeng/Desktop/mps_prep/Untitled.ipynb#X26sZmlsZQ%3D%3D?line=63)     pyket = MPSTools.from_block2(ket)
     [65](vscode-notebook-cell:/Users/qianfeng/Desktop/mps_prep/Untitled.ipynb#X26sZmlsZQ%3D%3D?line=64)     tensors = [x.blocks[0].reduced for x in pyket.tensors]
     [66](vscode-notebook-cell:/Users/qianfeng/Desktop/mps_prep/Untitled.ipynb#X26sZmlsZQ%3D%3D?line=65)     ket = tensors[0]

File [~/anaconda3/envs/qchem/lib/python3.8/site-packages/pyblock2/algebra/io.py:171](https://file+.vscode-resource.vscode-cdn.net/Users/qianfeng/Desktop/mps_prep/~/anaconda3/envs/qchem/lib/python3.8/site-packages/pyblock2/algebra/io.py:171), in MPSTools.from_block2(bmps)
    169 l = bmps.info.left_dims[i]
    170 m = bmps.info.basis[i]
--> 171 lm = StateInfo.tensor_product_ref(
    172     l, m, bmps.info.left_dims_fci[i + 1])
    173 clm = StateInfo.get_connection_info(l, m, lm)
    174 bmps.load_tensor(i)
TypeError: tensor_product_ref(): incompatible function arguments. The following argument types are supported:
    1. (arg0: block2.sz.StateInfo, arg1: block2.sz.StateInfo, arg2: block2.sz.StateInfo) -> block2.sz.StateInfo

Invoked with: < NB=0 PG=0 > : 2
, < NB=0 PG=0 > : 2
, < NB=0 PG=0 > : 4

How can I solve this?

[hczhai]

An up-to-date version of block2 is required: python3 -m pip install block2==0.5.3rc18 --extra-index-url=https://block-hczhai.github.io/block2-preview/pypi/

[fqian5]

Thank you so much for your help!!