qiskit/opflow/list_ops/list_op.py

# This code is part of Qiskit.
#
# (C) Copyright IBM 2020, 2023.
#
# This code is licensed under the Apache License, Version 2.0. You may
# obtain a copy of this license in the LICENSE.txt file in the root directory
# of this source tree or at http://www.apache.org/licenses/LICENSE-2.0.
#
# Any modifications or derivative works of this code must retain this
# copyright notice, and modified files need to carry a notice indicating
# that they have been altered from the originals.

"""ListOp Operator Class"""

from functools import reduce
from typing import Any, Callable, Dict, Iterator, List, Optional, Set, Sequence, Union, cast

import numpy as np
from scipy.sparse import spmatrix

from qiskit.circuit import ParameterExpression, QuantumCircuit
from qiskit.opflow.exceptions import OpflowError
from qiskit.opflow.operator_base import OperatorBase
from qiskit.quantum_info import Statevector
from qiskit.utils import arithmetic
from qiskit.utils.deprecation import deprecate_func


class ListOp(OperatorBase):
    """
    Deprecated: A Class for manipulating List Operators, and parent class to ``SummedOp``,
    ``ComposedOp`` and ``TensoredOp``.

    List Operators are classes for storing and manipulating lists of Operators, State functions,
    or Measurements, and include some rule or ``combo_fn`` defining how the Operator functions
    of the list constituents should be combined to form to cumulative Operator function of the
    ``ListOp``. For example, a ``SummedOp`` has an addition-based ``combo_fn``, so once the
    Operators in its list are evaluated against some bitstring to produce a list of results,
    we know to add up those results to produce the final result of the ``SummedOp``'s
    evaluation. In theory, this ``combo_fn`` can be any function over classical complex values,
    but for convenience we've chosen for them to be defined over NumPy arrays and values. This way,
    large numbers of evaluations, such as after calling ``to_matrix`` on the list constituents,
    can be efficiently combined. While the combination function is defined over classical
    values, it should be understood as the operation by which each Operators' underlying
    function is combined to form the underlying Operator function of the ``ListOp``. In this
    way, the ``ListOps`` are the basis for constructing large and sophisticated Operators,
    State Functions, and Measurements.

    The base ``ListOp`` class is particularly interesting, as its ``combo_fn`` is "the identity
    list Operation". Meaning, if we understand the ``combo_fn`` as a function from a list of
    complex values to some output, one such function is returning the list as-is. This is
    powerful for constructing compact hierarchical Operators which return many measurements in
    multiple dimensional lists.
    """

    @deprecate_func(
        since="0.24.0",
        additional_msg="For code migration guidelines, visit https://qisk.it/opflow_migration.",
    )
    def __init__(
        self,
        oplist: Sequence[OperatorBase],
        combo_fn: Optional[Callable] = None,
        coeff: Union[complex, ParameterExpression] = 1.0,
        abelian: bool = False,
        grad_combo_fn: Optional[Callable] = None,
    ) -> None:
        """
        Args:
            oplist: The list of ``OperatorBases`` defining this Operator's underlying function.
            combo_fn: The recombination function to combine classical results of the
                ``oplist`` Operators' eval functions (e.g. sum). Default is lambda x: x.
            coeff: A coefficient multiplying the operator
            abelian: Indicates whether the Operators in ``oplist`` are known to mutually commute.
            grad_combo_fn: The gradient of recombination function. If None, the gradient will
                be computed automatically.
            Note that the default "recombination function" lambda above is essentially the
            identity - it accepts the list of values, and returns them in a list.
        """
        super().__init__()
        self._oplist = self._check_input_types(oplist)
        self._combo_fn = combo_fn
        self._coeff = coeff
        self._abelian = abelian
        self._grad_combo_fn = grad_combo_fn

    def _check_input_types(self, oplist):
        if all(isinstance(x, OperatorBase) for x in oplist):
            return list(oplist)
        else:
            badval = next(x for x in oplist if not isinstance(x, OperatorBase))
            raise TypeError(f"ListOp expecting objects of type OperatorBase, got {badval}")

    def _state(
        self,
        coeff: Optional[Union[complex, ParameterExpression]] = None,
        combo_fn: Optional[Callable] = None,
        abelian: Optional[bool] = None,
        grad_combo_fn: Optional[Callable] = None,
    ) -> Dict:
        return {
            "coeff": coeff if coeff is not None else self.coeff,
            "combo_fn": combo_fn if combo_fn is not None else self.combo_fn,
            "abelian": abelian if abelian is not None else self.abelian,
            "grad_combo_fn": grad_combo_fn if grad_combo_fn is not None else self.grad_combo_fn,
        }

    @property
    def settings(self) -> Dict:
        """Return settings."""
        return {
            "oplist": self._oplist,
            "combo_fn": self._combo_fn,
            "coeff": self._coeff,
            "abelian": self._abelian,
            "grad_combo_fn": self._grad_combo_fn,
        }

    @property
    def oplist(self) -> List[OperatorBase]:
        """The list of ``OperatorBases`` defining the underlying function of this
        Operator.

        Returns:
            The Operators defining the ListOp
        """
        return self._oplist

    @staticmethod
    def default_combo_fn(x: Any) -> Any:
        """ListOp default combo function i.e. lambda x: x"""
        return x

    @property
    def combo_fn(self) -> Callable:
        """The function defining how to combine ``oplist`` (or Numbers, or NumPy arrays) to
        produce the Operator's underlying function. For example, SummedOp's combination function
        is to add all of the Operators in ``oplist``.

        Returns:
            The combination function.
        """
        if self._combo_fn is None:
            return ListOp.default_combo_fn
        return self._combo_fn

    @property
    def grad_combo_fn(self) -> Optional[Callable]:
        """The gradient of ``combo_fn``."""
        return self._grad_combo_fn

    @property
    def abelian(self) -> bool:
        """Whether the Operators in ``oplist`` are known to commute with one another.

        Returns:
            A bool indicating whether the ``oplist`` is Abelian.
        """
        return self._abelian

    @property
    def distributive(self) -> bool:
        """Indicates whether the ListOp or subclass is distributive under composition.
        ListOp and SummedOp are, meaning that (opv @ op) = (opv[0] @ op + opv[1] @ op)
        (using plus for SummedOp, list for ListOp, etc.), while ComposedOp and TensoredOp
        do not behave this way.

        Returns:
            A bool indicating whether the ListOp is distributive under composition.
        """
        return True

    @property
    def coeff(self) -> Union[complex, ParameterExpression]:
        """The scalar coefficient multiplying the Operator.

        Returns:
            The coefficient.
        """
        return self._coeff

    @property
    def coeffs(self) -> List[Union[complex, ParameterExpression]]:
        """Return a list of the coefficients of the operators listed.
        Raises exception for nested Listops.
        """
        if any(isinstance(op, ListOp) for op in self.oplist):
            raise TypeError("Coefficients are not returned for nested ListOps.")
        return [self.coeff * op.coeff for op in self.oplist]

    def primitive_strings(self) -> Set[str]:
        return reduce(set.union, [op.primitive_strings() for op in self.oplist])

    @property
    def num_qubits(self) -> int:
        num_qubits0 = self.oplist[0].num_qubits
        if not all(num_qubits0 == op.num_qubits for op in self.oplist):
            raise ValueError("Operators in ListOp have differing numbers of qubits.")
        return num_qubits0

    def add(self, other: OperatorBase) -> "ListOp":
        if self == other:
            return self.mul(2.0)

        # Avoid circular dependency
        # pylint: disable=cyclic-import
        from .summed_op import SummedOp

        return SummedOp([self, other])

    def adjoint(self) -> "ListOp":
        # TODO do this lazily? Basically rebuilds the entire tree, and ops and adjoints almost
        #  always come in pairs, so an AdjointOp holding a reference could save copying.
        if self.__class__ == ListOp:
            return ListOp(
                [op.adjoint() for op in self.oplist], **self._state(coeff=self.coeff.conjugate())
            )  # coeff is conjugated
        return self.__class__(
            [op.adjoint() for op in self.oplist], coeff=self.coeff.conjugate(), abelian=self.abelian
        )

    def traverse(
        self, convert_fn: Callable, coeff: Optional[Union[complex, ParameterExpression]] = None
    ) -> "ListOp":
        """Apply the convert_fn to each node in the oplist.

        Args:
            convert_fn: The function to apply to the internal OperatorBase.
            coeff: A coefficient to multiply by after applying convert_fn.
                If it is None, self.coeff is used instead.

        Returns:
            The converted ListOp.
        """
        if coeff is None:
            coeff = self.coeff

        if self.__class__ == ListOp:
            return ListOp([convert_fn(op) for op in self.oplist], **self._state(coeff=coeff))
        return self.__class__(
            [convert_fn(op) for op in self.oplist], coeff=coeff, abelian=self.abelian
        )

    def equals(self, other: OperatorBase) -> bool:
        if not isinstance(other, type(self)) or not len(self.oplist) == len(other.oplist):
            return False
        # Note, ordering matters here (i.e. different list orders will return False)
        return self.coeff == other.coeff and all(
            op1 == op2 for op1, op2 in zip(self.oplist, other.oplist)
        )

    # We need to do this because otherwise Numpy takes over scalar multiplication and wrecks it if
    # isinstance(scalar, np.number) - this started happening when we added __get_item__().
    __array_priority__ = 10000

    def mul(self, scalar: Union[complex, ParameterExpression]) -> "ListOp":
        if not isinstance(scalar, (int, float, complex, ParameterExpression)):
            raise ValueError(
                "Operators can only be scalar multiplied by float or complex, not "
                "{} of type {}.".format(scalar, type(scalar))
            )
        if self.__class__ == ListOp:
            return ListOp(self.oplist, **self._state(coeff=scalar * self.coeff))
        return self.__class__(self.oplist, coeff=scalar * self.coeff, abelian=self.abelian)

    def tensor(self, other: OperatorBase) -> OperatorBase:
        # Avoid circular dependency
        # pylint: disable=cyclic-import
        from .tensored_op import TensoredOp

        return TensoredOp([self, other])

    def tensorpower(self, other: int) -> Union[OperatorBase, int]:
        # Hack to make op1^(op2^0) work as intended.
        if other == 0:
            return 1
        if not isinstance(other, int) or other <= 0:
            raise TypeError("Tensorpower can only take positive int arguments")

        # Avoid circular dependency
        # pylint: disable=cyclic-import
        from .tensored_op import TensoredOp

        return TensoredOp([self] * other)

    def _expand_dim(self, num_qubits: int) -> "ListOp":
        oplist = [
            op._expand_dim(num_qubits + self.num_qubits - op.num_qubits) for op in self.oplist
        ]
        return ListOp(oplist, **self._state())

    def permute(self, permutation: List[int]) -> "OperatorBase":
        """Permute the qubits of the operator.

        Args:
            permutation: A list defining where each qubit should be permuted. The qubit at index
                j should be permuted to position permutation[j].

        Returns:
            A new ListOp representing the permuted operator.

        Raises:
            OpflowError: if indices do not define a new index for each qubit.
        """
        new_self = self
        circuit_size = max(permutation) + 1

        try:
            if self.num_qubits != len(permutation):
                raise OpflowError("New index must be defined for each qubit of the operator.")
        except ValueError:
            raise OpflowError(
                "Permute is only possible if all operators in the ListOp have the "
                "same number of qubits."
            ) from ValueError
        if self.num_qubits < circuit_size:
            # pad the operator with identities
            new_self = self._expand_dim(circuit_size - self.num_qubits)
        qc = QuantumCircuit(circuit_size)
        # extend the indices to match the size of the circuit
        permutation = (
            list(filter(lambda x: x not in permutation, range(circuit_size))) + permutation
        )

        # decompose permutation into sequence of transpositions
        transpositions = arithmetic.transpositions(permutation)
        for trans in transpositions:
            qc.swap(trans[0], trans[1])

        # pylint: disable=cyclic-import
        from ..primitive_ops.circuit_op import CircuitOp

        return CircuitOp(qc.reverse_ops()) @ new_self @ CircuitOp(qc)

    def compose(
        self, other: OperatorBase, permutation: Optional[List[int]] = None, front: bool = False
    ) -> OperatorBase:

        new_self, other = self._expand_shorter_operator_and_permute(other, permutation)
        new_self = cast(ListOp, new_self)

        if front:
            return other.compose(new_self)
        # Avoid circular dependency
        # pylint: disable=cyclic-import
        from .composed_op import ComposedOp

        return ComposedOp([new_self, other])

    def power(self, exponent: int) -> OperatorBase:
        if not isinstance(exponent, int) or exponent <= 0:
            raise TypeError("power can only take positive int arguments")

        # Avoid circular dependency
        # pylint: disable=cyclic-import
        from .composed_op import ComposedOp

        return ComposedOp([self] * exponent)

    def to_matrix(self, massive: bool = False) -> np.ndarray:
        OperatorBase._check_massive("to_matrix", True, self.num_qubits, massive)

        # Combination function must be able to handle classical values.
        # Note: this can end up, when we have list operators containing other list operators, as a
        #       ragged array and numpy 1.19 raises a deprecation warning unless this is explicitly
        #       done as object type now - was implicit before.
        mat = self.combo_fn(
            np.asarray(
                [op.to_matrix(massive=massive) * self.coeff for op in self.oplist], dtype=object
            )
        )
        return np.asarray(mat, dtype=complex)

    def to_spmatrix(self) -> Union[spmatrix, List[spmatrix]]:
        """Returns SciPy sparse matrix representation of the Operator.

        Returns:
            CSR sparse matrix representation of the Operator, or List thereof.
        """

        # Combination function must be able to handle classical values
        return self.combo_fn([op.to_spmatrix() for op in self.oplist]) * self.coeff

    def eval(
        self,
        front: Optional[
            Union[str, Dict[str, complex], np.ndarray, OperatorBase, Statevector]
        ] = None,
    ) -> Union[OperatorBase, complex]:
        """
        Evaluate the Operator's underlying function, either on a binary string or another Operator.
        A square binary Operator can be defined as a function taking a binary function to another
        binary function. This method returns the value of that function for a given StateFn or
        binary string. For example, ``op.eval('0110').eval('1110')`` can be seen as querying the
        Operator's matrix representation by row 6 and column 14, and will return the complex
        value at those "indices." Similarly for a StateFn, ``op.eval('1011')`` will return the
        complex value at row 11 of the vector representation of the StateFn, as all StateFns are
        defined to be evaluated from Zero implicitly (i.e. it is as if ``.eval('0000')`` is already
        called implicitly to always "indexing" from column 0).

        ListOp's eval recursively evaluates each Operator in ``oplist``,
        and combines the results using the recombination function ``combo_fn``.

        Args:
            front: The bitstring, dict of bitstrings (with values being coefficients), or
                StateFn to evaluated by the Operator's underlying function.

        Returns:
            The output of the ``oplist`` Operators' evaluation function, combined with the
            ``combo_fn``. If either self or front contain proper ``ListOps`` (not ListOp
            subclasses), the result is an n-dimensional list of complex or StateFn results,
            resulting from the recursive evaluation by each OperatorBase in the ListOps.

        Raises:
            NotImplementedError: Raised if called for a subclass which is not distributive.
            TypeError: Operators with mixed hierarchies, such as a ListOp containing both
                PrimitiveOps and ListOps, are not supported.
            NotImplementedError: Attempting to call ListOp's eval from a non-distributive subclass.

        """
        # pylint: disable=cyclic-import
        from ..state_fns.dict_state_fn import DictStateFn
        from ..state_fns.vector_state_fn import VectorStateFn
        from ..state_fns.sparse_vector_state_fn import SparseVectorStateFn

        # The below code only works for distributive ListOps, e.g. ListOp and SummedOp
        if not self.distributive:
            raise NotImplementedError(
                "ListOp's eval function is only defined for distributive ListOps."
            )

        evals = [op.eval(front) for op in self.oplist]

        # Handle application of combo_fn for DictStateFn resp VectorStateFn operators
        if self._combo_fn is not None:  # If not using default.
            if (
                all(isinstance(op, DictStateFn) for op in evals)
                or all(isinstance(op, VectorStateFn) for op in evals)
                or all(isinstance(op, SparseVectorStateFn) for op in evals)
            ):
                if not all(
                    op.is_measurement == evals[0].is_measurement for op in evals  # type: ignore
                ):
                    raise NotImplementedError(
                        "Combo_fn not yet supported for mixed measurement "
                        "and non-measurement StateFns"
                    )
                result = self.combo_fn(evals)
                if isinstance(result, list):
                    multiplied = self.coeff * np.array(result)
                    return multiplied.tolist()
                return self.coeff * result

        if all(isinstance(op, OperatorBase) for op in evals):
            return self.__class__(evals)  # type: ignore
        elif any(isinstance(op, OperatorBase) for op in evals):
            raise TypeError("Cannot handle mixed scalar and Operator eval results.")
        else:
            result = self.combo_fn(evals)
            if isinstance(result, list):
                multiplied = self.coeff * np.array(result)
                return multiplied.tolist()
            return self.coeff * result

    def exp_i(self) -> OperatorBase:
        """Return an ``OperatorBase`` equivalent to an exponentiation of self * -i, e^(-i*op)."""
        # pylint: disable=unidiomatic-typecheck
        if type(self) == ListOp:
            return ListOp(
                [op.exp_i() for op in self.oplist], **self._state(abelian=False)  # type: ignore
            )

        # pylint: disable=cyclic-import
        from ..evolutions.evolved_op import EvolvedOp

        return EvolvedOp(self)

    def log_i(self, massive: bool = False) -> OperatorBase:
        """Return a ``MatrixOp`` equivalent to log(H)/-i for this operator H. This
        function is the effective inverse of exp_i, equivalent to finding the Hermitian
        Operator which produces self when exponentiated. For proper ListOps, applies ``log_i``
        to all ops in oplist.
        """
        if self.__class__.__name__ == ListOp.__name__:
            return ListOp(
                [op.log_i(massive=massive) for op in self.oplist],  # type: ignore
                **self._state(abelian=False),
            )

        return self.to_matrix_op(massive=massive).log_i(massive=massive)

    def __str__(self) -> str:
        content_string = ",\n".join([str(op) for op in self.oplist])
        main_string = "{}([\n{}\n])".format(
            self.__class__.__name__, self._indent(content_string, indentation=self.INDENTATION)
        )
        if self.abelian:
            main_string = "Abelian" + main_string
        if self.coeff != 1.0:
            main_string = f"{self.coeff} * " + main_string
        return main_string

    def __repr__(self) -> str:
        return "{}({}, coeff={}, abelian={})".format(
            self.__class__.__name__, repr(self.oplist), self.coeff, self.abelian
        )

    @property
    def parameters(self):
        params = set()
        for op in self.oplist:
            params.update(op.parameters)
        if isinstance(self.coeff, ParameterExpression):
            params.update(self.coeff.parameters)
        return params

    def assign_parameters(self, param_dict: dict) -> OperatorBase:
        param_value = self.coeff
        if isinstance(self.coeff, ParameterExpression):
            unrolled_dict = self._unroll_param_dict(param_dict)
            if isinstance(unrolled_dict, list):
                return ListOp([self.assign_parameters(param_dict) for param_dict in unrolled_dict])
            if self.coeff.parameters <= set(unrolled_dict.keys()):
                binds = {param: unrolled_dict[param] for param in self.coeff.parameters}
                param_value = float(self.coeff.bind(binds))
        return self.traverse(lambda x: x.assign_parameters(param_dict), coeff=param_value)

    def reduce(self) -> OperatorBase:
        reduced_ops = [op.reduce() for op in self.oplist]
        if self.__class__ == ListOp:
            return ListOp(reduced_ops, **self._state())
        return self.__class__(reduced_ops, coeff=self.coeff, abelian=self.abelian)

    def to_matrix_op(self, massive: bool = False) -> "ListOp":
        """Returns an equivalent Operator composed of only NumPy-based primitives, such as
        ``MatrixOp`` and ``VectorStateFn``."""
        if self.__class__ == ListOp:
            return cast(
                ListOp,
                ListOp(
                    [op.to_matrix_op(massive=massive) for op in self.oplist], **self._state()
                ).reduce(),
            )
        return cast(
            ListOp,
            self.__class__(
                [op.to_matrix_op(massive=massive) for op in self.oplist],
                coeff=self.coeff,
                abelian=self.abelian,
            ).reduce(),
        )

    def to_circuit_op(self) -> OperatorBase:
        """Returns an equivalent Operator composed of only QuantumCircuit-based primitives,
        such as ``CircuitOp`` and ``CircuitStateFn``."""
        # pylint: disable=cyclic-import
        from ..state_fns.operator_state_fn import OperatorStateFn

        if self.__class__ == ListOp:
            return ListOp(
                [
                    op.to_circuit_op() if not isinstance(op, OperatorStateFn) else op
                    for op in self.oplist
                ],
                **self._state(),
            ).reduce()
        return self.__class__(
            [
                op.to_circuit_op() if not isinstance(op, OperatorStateFn) else op
                for op in self.oplist
            ],
            coeff=self.coeff,
            abelian=self.abelian,
        ).reduce()

    def to_pauli_op(self, massive: bool = False) -> "ListOp":
        """Returns an equivalent Operator composed of only Pauli-based primitives,
        such as ``PauliOp``."""
        # pylint: disable=cyclic-import
        from ..state_fns.state_fn import StateFn

        if self.__class__ == ListOp:
            return ListOp(
                [
                    op.to_pauli_op(massive=massive)  # type: ignore
                    if not isinstance(op, StateFn)
                    else op
                    for op in self.oplist
                ],
                **self._state(),
            ).reduce()
        return self.__class__(
            [
                op.to_pauli_op(massive=massive)  # type: ignore
                if not isinstance(op, StateFn)
                else op
                for op in self.oplist
            ],
            coeff=self.coeff,
            abelian=self.abelian,
        ).reduce()

    def _is_empty(self):
        return len(self.oplist) == 0

    # Array operations:

    def __getitem__(self, offset: Union[int, slice]) -> OperatorBase:
        """Allows array-indexing style access to the Operators in ``oplist``.

        Args:
            offset: The index of ``oplist`` desired.

        Returns:
            The ``OperatorBase`` at index ``offset`` of ``oplist``,
            or another ListOp with the same properties as this one if offset is a slice.
        """
        if isinstance(offset, int):
            return self.oplist[offset]

        if self.__class__ == ListOp:
            return ListOp(oplist=self._oplist[offset], **self._state())

        return self.__class__(oplist=self._oplist[offset], coeff=self._coeff, abelian=self._abelian)

    def __iter__(self) -> Iterator:
        """Returns an iterator over the operators in ``oplist``.

        Returns:
            An iterator over the operators in ``oplist``
        """
        return iter(self.oplist)

    def __len__(self) -> int:
        """Length of ``oplist``.

        Returns:
            An int equal to the length of ``oplist``.
        """
        return len(self.oplist)