Add reference.qubit for testing and reference

doc/releases/changelog-dev.md

@@ -7,18 +7,14 @@
 <h3>Improvements 🛠</h3>
 
 * `qml.qchem.excitations` now optionally returns fermionic operators.
-  [(#6171)](https://github.com/PennyLaneAI/pennylane/pull/6171)
+ [(#6171)](https://github.com/PennyLaneAI/pennylane/pull/6171)
 
 * The `diagonalize_measurements` transform now uses a more efficient method of diagonalization 
  when possible, based on the `pauli_rep` of the relevant observables.
  [#6113](https://github.com/PennyLaneAI/pennylane/pull/6113/)
 
-<h4>Capturing and representing hybrid programs</h4>
-
-* Differentiation of hybrid programs via `qml.grad` can now be captured into plxpr.
- When evaluating a captured `qml.grad` instruction, it will dispatch to `jax.grad`,
- which differs from the Autograd implementation of `qml.grad` itself.
- [(#6120)](https://github.com/PennyLaneAI/pennylane/pull/6120)
+* The `Hermitian` operator now has a `compute_sparse_matrix` implementation.
+ [(#6225)](https://github.com/PennyLaneAI/pennylane/pull/6225)
 
 <h4>Capturing and representing hybrid programs</h4>
 
@@ -50,6 +46,9 @@
  unique representation of the object.
  [(#6167)](https://github.com/PennyLaneAI/pennylane/pull/6167)
 
+* A `ReferenceQubit` is introduced for testing purposes and as a reference for future plugin development.
+ [(#6181)](https://github.com/PennyLaneAI/pennylane/pull/6181)
+
 * The `to_mat` methods for `FermiWord` and `FermiSentence` now optionally return
  a sparse matrix.
  [(#6173)](https://github.com/PennyLaneAI/pennylane/pull/6173)
@@ -102,12 +101,19 @@
 * The ``qml.Qubitization`` template now orders the ``control`` wires first and the ``hamiltonian`` wires second, which is the expected according to other templates.
  [(#6229)](https://github.com/PennyLaneAI/pennylane/pull/6229)
 
-* <h3>Contributors ✍️</h3>
+* Fixes a bug where a circuit using the `autograd` interface sometimes returns nested values that are not of the `autograd` interface.
+ [(#6225)](https://github.com/PennyLaneAI/pennylane/pull/6225)
+
+* Fixes a bug where a simple circuit with no parameters or only builtin/numpy arrays as parameters returns autograd tensors.
+ [(#6225)](https://github.com/PennyLaneAI/pennylane/pull/6225)
+
+<h3>Contributors ✍️</h3>
 
 This release contains contributions from (in alphabetical order):
 
 Guillermo Alonso,
 Utkarsh Azad,
+Astral Cai,
 Lillian M. A. Frederiksen,
 Christina Lee,
 William Maxwell,

pennylane/devices/execution_config.py

@@ -17,7 +17,7 @@
 from dataclasses import dataclass, field
 from typing import Optional, Union
 
-from pennylane.workflow import SUPPORTED_INTERFACES
+from pennylane.workflow import SUPPORTED_INTERFACE_NAMES
 
 
 @dataclass
@@ -110,9 +110,9 @@ def __post_init__(self):
 
  Note that this hook is automatically called after init via the dataclass integration.
  """
- if self.interface not in SUPPORTED_INTERFACES:
+ if self.interface not in SUPPORTED_INTERFACE_NAMES:
  raise ValueError(
- f"Unknown interface. interface must be in {SUPPORTED_INTERFACES}, got {self.interface} instead."
+ f"Unknown interface. interface must be in {SUPPORTED_INTERFACE_NAMES}, got {self.interface} instead."
  )
 
  if self.grad_on_execution not in {True, False, None}:

pennylane/devices/legacy_facade.py

@@ -25,6 +25,7 @@
 import pennylane as qml
 from pennylane.measurements import MidMeasureMP, Shots
 from pennylane.transforms.core.transform_program import TransformProgram
+from pennylane.workflow.execution import INTERFACE_MAP
 
 from .device_api import Device
 from .execution_config import DefaultExecutionConfig
@@ -328,18 +329,18 @@ def _create_temp_device(self, batch):
  if interface == "numpy":
  return self._device
 
- mapped_interface = qml.workflow.execution.INTERFACE_MAP.get(interface, interface)
+ interface = INTERFACE_MAP.get(interface, interface)
 
  backprop_interface = self._device.capabilities().get("passthru_interface", None)
- if mapped_interface == backprop_interface:
+ if interface == backprop_interface:
  return self._device
 
  backprop_devices = self._device.capabilities().get("passthru_devices", None)
 
  if backprop_devices is None:
  raise qml.DeviceError(f"Device {self} does not support backpropagation.")
 
- if backprop_devices[mapped_interface] == self._device.short_name:
+ if backprop_devices[interface] == self._device.short_name:
  return self._device
 
  if self.target_device.short_name != "default.qubit.legacy":
@@ -367,7 +368,7 @@ def _create_temp_device(self, batch):
  )
  # we already warned about backprop device switching
  new_device = qml.device(
- backprop_devices[mapped_interface],
+ backprop_devices[interface],
  wires=self._device.wires,
  shots=self._device.shots,
  ).target_device
@@ -396,25 +397,24 @@ def _validate_backprop_method(self, tape):
  return False
  params = tape.get_parameters(trainable_only=False)
  interface = qml.math.get_interface(*params)
+ if interface != "numpy":
+ interface = INTERFACE_MAP.get(interface, interface)
 
  if tape and any(isinstance(m.obs, qml.SparseHamiltonian) for m in tape.measurements):
  return False
- if interface == "numpy":
- interface = None
- mapped_interface = qml.workflow.execution.INTERFACE_MAP.get(interface, interface)
 
  # determine if the device supports backpropagation
  backprop_interface = self._device.capabilities().get("passthru_interface", None)
 
  if backprop_interface is not None:
  # device supports backpropagation natively
- return mapped_interface in [backprop_interface, "Numpy"]
+ return interface in [backprop_interface, "numpy"]
  # determine if the device has any child devices that support backpropagation
  backprop_devices = self._device.capabilities().get("passthru_devices", None)
 
  if backprop_devices is None:
  return False
- return mapped_interface in backprop_devices or mapped_interface == "Numpy"
+ return interface in backprop_devices or interface == "numpy"
 
  def _validate_adjoint_method(self, tape):
  # The conditions below provide a minimal set of requirements that we can likely improve upon in

pennylane/devices/qubit/simulate.py

@@ -922,7 +922,7 @@ def _(original_measurement: ExpectationMP, measures): # pylint: disable=unused-
  for v in measures.values():
  if not v[0] or v[1] is tuple():
  continue
- cum_value += v[0] * v[1]
+ cum_value += qml.math.multiply(v[0], v[1])
  total_counts += v[0]
  return cum_value / total_counts
 
@@ -935,7 +935,7 @@ def _(original_measurement: ProbabilityMP, measures): # pylint: disable=unused-
  for v in measures.values():
  if not v[0] or v[1] is tuple():
  continue
- cum_value += v[0] * v[1]
+ cum_value += qml.math.multiply(v[0], v[1])
  total_counts += v[0]
  return cum_value / total_counts
 

pennylane/devices/reference_qubit.py

@@ -0,0 +1,154 @@
+# Copyright 2018-2024 Xanadu Quantum Technologies Inc.
+
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+
+# http://www.apache.org/licenses/LICENSE-2.0
+
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+Contains the ReferenceQubit device, a minimal device that can be used for testing
+and plugin development purposes.
+"""
+
+import numpy as np
+
+import pennylane as qml
+
+from .device_api import Device
+from .execution_config import DefaultExecutionConfig
+from .modifiers import simulator_tracking, single_tape_support
+from .preprocess import decompose, validate_device_wires, validate_measurements
+
+
+def sample_state(state: np.ndarray, shots: int, seed=None):
+ """Generate samples from the provided state and number of shots."""
+
+ probs = np.imag(state) ** 2 + np.real(state) ** 2
+ basis_states = np.arange(len(probs))
+
+ num_wires = int(np.log2(len(probs)))
+
+ rng = np.random.default_rng(seed)
+ basis_samples = rng.choice(basis_states, shots, p=probs)
+
+ # convert basis state integers to array of booleans
+ bin_strings = (format(s, f"0{num_wires}b") for s in basis_samples)
+ return np.array([[int(val) for val in s] for s in bin_strings])
+
+
+def simulate(tape: qml.tape.QuantumTape, seed=None) -> qml.typing.Result:
+ """Simulate a tape and turn it into results.
+
+ Args:
+ tape (.QuantumTape): a representation of a circuit
+ seed (Any): A seed to use to control the generation of samples.
+
+ """
+ # 1) create the initial state
+ state = np.zeros(2 ** len(tape.wires))
+ state[0] = 1.0
+
+ # 2) apply all the operations
+ for op in tape.operations:
+ op_mat = op.matrix(wire_order=tape.wires)
+ state = qml.math.matmul(op_mat, state)
+
+ # 3) perform measurements
+ # note that shots are pulled from the tape, not from the device
+ if tape.shots:
+ samples = sample_state(state, shots=tape.shots.total_shots, seed=seed)
+ # Shot vector support
+ results = []
+ for lower, upper in tape.shots.bins():
+ sub_samples = samples[lower:upper]
+ results.append(
+ tuple(mp.process_samples(sub_samples, tape.wires) for mp in tape.measurements)
+ )
+ if len(tape.measurements) == 1:
+ results = [res[0] for res in results]
+ if not tape.shots.has_partitioned_shots:
+ results = results[0]
+ else:
+ results = tuple(results)
+ else:
+ results = tuple(mp.process_state(state, tape.wires) for mp in tape.measurements)
+ if len(tape.measurements) == 1:
+ results = results[0]
+
+ return results
+
+
+operations = frozenset({"PauliX", "PauliY", "PauliZ", "Hadamard", "CNOT", "CZ", "RX", "RY", "RZ"})
+
+
+def supports_operation(op: qml.operation.Operator) -> bool:
+ """This function used by preprocessing determines what operations
+ are natively supported by the device.
+
+ While in theory ``simulate`` can support any operation with a matrix, we limit the target
+ gate set for improved testing and reference purposes.
+
+ """
+ return getattr(op, "name", None) in operations
+
+
+@simulator_tracking # update device.tracker with some relevant information
+@single_tape_support # add support for device.execute(tape) in addition to device.execute((tape,))
+class ReferenceQubit(Device):
+ """A slimmed down numpy-based simulator for reference and testing purposes.
+
+ Args:
+ wires (int, Iterable[Number, str]): Number of wires present on the device, or iterable that
+ contains unique labels for the wires as numbers (i.e., ``[-1, 0, 2]``) or strings
+ (``['ancilla', 'q1', 'q2']``). Default ``None`` if not specified. While this device allows
+ for ``wires`` to be unspecified at construction time, other devices may make this argument
+ mandatory. Devices can also implement additional restrictions on the possible wires.
+ shots (int, Sequence[int], Sequence[Union[int, Sequence[int]]]): The default number of shots
+ to use in executions involving this device. Note that during execution, shots
+ are pulled from the circuit, not from the device.
+ seed (Union[str, None, int, array_like[int], SeedSequence, BitGenerator, Generator, jax.random.PRNGKey]): A
+ seed-like parameter matching that of ``seed`` for ``numpy.random.default_rng``. This is an optional
+ keyword argument added to follow recommend NumPy best practices. Other devices do not need
+ this parameter if it does not make sense for them.
+
+ """
+
+ name = "reference.qubit"
+
+ def __init__(self, wires=None, shots=None, seed=None):
+ super().__init__(wires=wires, shots=shots)
+
+ # seed and rng not necessary for a device, but part of recommended
+ # numpy practices to use a local random number generator
+ self._rng = np.random.default_rng(seed)
+
+ def preprocess(self, execution_config=DefaultExecutionConfig):
+
+ # Here we convert an arbitrary tape into one natively supported by the device
+ program = qml.transforms.core.TransformProgram()
+ program.add_transform(validate_device_wires, wires=self.wires, name="reference.qubit")
+ program.add_transform(qml.defer_measurements)
+ program.add_transform(qml.transforms.split_non_commuting)
+ program.add_transform(qml.transforms.diagonalize_measurements)
+ program.add_transform(
+ decompose,
+ stopping_condition=supports_operation,
+ skip_initial_state_prep=False,
+ name="reference.qubit",
+ )
+ program.add_transform(validate_measurements, name="reference.qubit")
+ program.add_transform(qml.transforms.broadcast_expand)
+
+ # no need to preprocess the config as the device does not support derivatives
+ return program, execution_config
+
+ def execute(self, circuits, execution_config=DefaultExecutionConfig):
+ for tape in circuits:
+ assert all(supports_operation(op) for op in tape.operations)
+ return tuple(simulate(tape, seed=self._rng) for tape in circuits)

pennylane/ops/qubit/observables.py

@@ -137,6 +137,10 @@ def compute_matrix(A: TensorLike) -> TensorLike: # pylint: disable=arguments-di
  Hermitian._validate_input(A)
  return A
 
+ @staticmethod
+ def compute_sparse_matrix(A) -> csr_matrix: # pylint: disable=arguments-differ
+ return csr_matrix(Hermitian.compute_matrix(A))
+
  @property
  def eigendecomposition(self) -> dict[str, TensorLike]:
  """Return the eigendecomposition of the matrix specified by the Hermitian observable.

pennylane/workflow/__init__.py

@@ -56,6 +56,6 @@
 
 """
 from .construct_batch import construct_batch, get_transform_program
-from .execution import INTERFACE_MAP, SUPPORTED_INTERFACES, execute
+from .execution import INTERFACE_MAP, SUPPORTED_INTERFACE_NAMES, execute
 from .qnode import QNode, qnode
 from .set_shots import set_shots