Add walk operator example for sparse chem ham.

[fdmalone]

Working on #841, adds a walk operator example for the sparse hamiltonian which fits the usual select and prepare construction. Just adds an example and some autodoc. I have an organizational / doc question: I added a BloqExample for building the walk operator BUT this reproduces the docstring for QubitizationWalkOperator in sparse.ipynb which isn't ideal. Would it be better to add this as an example in the notebook for QubitizationWalkOperator, I feel this might make the notebook a bit crowded if we keep adding examples there?

Currently there is a test failing for serialization which I don't really understand yet because Prepare and Select pass the serialization test? Any idea @tanujkhattar?

I also fixed the latex for QubitizationWalkOperator.

[fdmalone]

Serialization error:

self = PrepareSparse(num_spin_orb=8, num_non_zero=65, num_bits_state_prep=12, num_bits_rot_aa=8, adjoint=False, qroam_block_size=None)
other = PrepareSparse(num_spin_orb=8, num_non_zero=65, num_bits_state_prep=12, num_bits_rot_aa=8, adjoint=0, qroam_block_size=None)

    def __eq__(self, other):
        if other.__class__ is not self.__class__:
            return NotImplemented
>       return  (
            self.num_spin_orb,
            self.num_non_zero,
            self.num_bits_state_prep,
            self.alt_pqrs,
            self.alt_theta,
            self.alt_one_body,
            self.ind_pqrs,
            self.theta,
            self.one_body,
            self.keep,
            self.num_bits_rot_aa,
            self.adjoint,
            self.qroam_block_size,
        ) == (
            other.num_spin_orb,
            other.num_non_zero,
            other.num_bits_state_prep,
            other.alt_pqrs,
            other.alt_theta,
            other.alt_one_body,
            other.ind_pqrs,
            other.theta,
            other.one_body,
            other.keep,
            other.num_bits_rot_aa,
            other.adjoint,
            other.qroam_block_size,
        )
E       ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

[fdmalone]

And for QPE:

qualtran/bloqs/phase_estimation/qubitization_qpe_test.py:37:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
qualtran/conftest.py:103: in assert_bloq_example_typing_for_pytest
    qlt_testing.assert_bloq_example_preserves_types(bloq_ex)
qualtran/testing.py:587: in assert_bloq_example_preserves_types
    cbloq = bloq.decompose_bloq()
qualtran/_infra/gate_with_registers.py:245: in decompose_bloq
    return decompose_from_cirq_style_method(self)
qualtran/cirq_interop/_cirq_to_bloq.py:574: in decompose_from_cirq_style_method
    raise exc
qualtran/cirq_interop/_cirq_to_bloq.py:567: in decompose_from_cirq_style_method
    return cirq_optree_to_cbloq(
qualtran/cirq_interop/_cirq_to_bloq.py:444: in cirq_optree_to_cbloq
    circuit = cirq.Circuit(optree)
../../.virtualenvs/qualtran/lib/python3.11/site-packages/cirq/circuits/circuit.py:1771: in __init__
    flattened_contents = tuple(ops.flatten_to_ops_or_moments(contents))
../../.virtualenvs/qualtran/lib/python3.11/site-packages/cirq/ops/op_tree.py:134: in flatten_to_ops_or_moments
    yield from flatten_to_ops_or_moments(subtree)
../../.virtualenvs/qualtran/lib/python3.11/site-packages/cirq/ops/op_tree.py:133: in flatten_to_ops_or_moments
    for subtree in root:
qualtran/bloqs/phase_estimation/qubitization_qpe.py:137: in decompose_from_registers
    walk_controlled = self.walk.controlled(control_values=[1])
qualtran/bloqs/qubitization_walk_operator.py:135: in controlled
    c_select = self.select.controlled(control_values=control_values)
qualtran/_infra/gate_with_registers.py:323: in controlled
    controlled_gate = cirq.ControlledGate(
../../.virtualenvs/qualtran/lib/python3.11/site-packages/cirq/ops/controlled_gate.py:88: in __init__
    _validate_sub_object(sub_gate)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

sub_object = SelectSparse(num_spin_orb=8, control_val=None)

    def _validate_sub_object(sub_object: Union['cirq.Gate', 'cirq.Operation']):
        if protocols.is_measurement(sub_object):
            raise ValueError(f'Cannot control measurement {sub_object}')
        if not protocols.has_mixture(sub_object) and not protocols.is_parameterized(sub_object):
>           raise ValueError(f'Cannot control channel with non-unitary operators: {sub_object}')
E           ValueError: Cannot control channel with non-unitary operators: SelectSparse(num_spin_orb=8, control_val=None)

../../.virtualenvs/qualtran/lib/python3.11/site-packages/cirq/ops/controlled_gate.py:357: ValueError

[fdmalone]

Ok, I guess walk_operator isn't serializable right now, I updated the set of bloqs to skip. Just the control error now.

[tanujkhattar]

I think the control error should be fixed once my inflight PR #879 is merged.

[tanujkhattar]

def __eq__(self, other):

I think this is happening because one of the elements in the tuple is a numpy array (either before or after serialization / deserialization roundtrip). If any of this is a attrs attribute and expected to be a Sequence, we should add a default attrs.field(converter=tuple) to expect any sequence and convert it to a tuple.

The error seems to be not in serialization / deserialization roundtrip but rather when comparing the roundtrip object with original object. If the serialization roundtrip is messing with the attributes, it'd be good to open a new issue but my guess is that it's serialization; for example; a list of ints/floats to a numpy array of ints/floats and deserializing back to numpy array which is messing with equality because we don't have a default converter.

cc @Epsilon1024 Maybe you also have thoughts here.

[fdmalone]

It's odd because Prepare (which it's complaining about) does correctly serialize without issue when it's not apart of the walk operator, I noticed that the example walk_op was being skipped already. I can't remember why serialization for walk op is skipped.

[fdmalone]

[fdmalone]

@tanujkhattar PTAL, I fixed a few bugs a long the way:

1. There was an extra comparator which shouldn't have been there. Prepare^ will uncompute this.
2. There was a missing single-qubit CSWAP.
3. I removed the special case for adjoint in the call graph so the decomposition matches the call graph, but this brings us farther away from the reference value because of QROM / inequality / CSWAP inversion cost discrepancies.
4. I fixed a sign error in the cost for uniform state preparation.
5. Using SelectSwapQROM makes the tests a bit slower so I reduced the system sizes in sparse_test.py.

The qubitization_qpe serialization test for chemistry was too slow, so I also just add this to the skipped tests.

[fdmalone]

That was weird, them mypy check failed but it automerged? @mpharrigan

[mpharrigan]

#926 configured the CI job but someone with github admin (hopefully me) needs to set it as a required check.