Release 0.36.0

New features since last release

Estimate errors in a quantum circuit 🧮

• This version of PennyLane lays the foundation for estimating the total error in a quantum circuit from the combination of individual gate errors. (#5154) (#5464) (#5465) (#5278) (#5384)

Two new user-facing classes enable calculating and propagating gate errors in PennyLane:

• qml.resource.SpectralNormError: the spectral norm error is defined as the distance, in spectral norm, between the true unitary we intend to apply and the approximate unitary that is actually applied.

• qml.resource.ErrorOperation: a base class that inherits from qml.operation.Operation and represents quantum operations which carry some form of algorithmic error.

  SpectralNormError can be used for back-of-the-envelope type calculations like obtaining the spectral norm error between two unitaries via get_error:

  import pennylane as qml 
  from pennylane.resource import ErrorOperation, SpectralNormError
  
  intended_op = qml.RY(0.40, 0) 
  actual_op = qml.RY(0.41, 0) # angle of rotation is slightly off
  

  >>> SpectralNormError.get_error(intended_op, actual_op) 
  0.004999994791668309 
  

  SpectralNormError is also a key tool to specify errors in larger quantum circuits:

• For operations representing a major building block of an algorithm, we can create a custom operation that inherits from ErrorOperation. This child class must override the error method and should return a SpectralNormError instance:

  class MyErrorOperation(ErrorOperation):
  
     def __init__(self, error_val, wires): 
         self.error_val = error_val 
         super().__init__(wires=wires)
  
     def error(self): 
         return SpectralNormError(self.error_val)
  

  In this toy example, MyErrorOperation introduces an arbitrary SpectralNormError when called in a QNode. It does not require a decomposition or matrix representation when used with null.qubit (suggested for use with resource and error estimation since circuit executions are not required to calculate resources or errors).

  dev = qml.device("null.qubit")
  
  @qml.qnode(dev) 
  def circuit(): 
     MyErrorOperation(0.1, wires=0) 
     MyErrorOperation(0.2, wires=1) 
     return qml.state() 
  

  The total spectral norm error of the circuit can be calculated using qml.specs:

  >>> qml.specs(circuit)()['errors'] 
  {'SpectralNormError': SpectralNormError(0.30000000000000004)} 
  

• PennyLane already includes a number of built-in building blocks for algorithms like QuantumPhaseEstimation and TrotterProduct. TrotterProduct now propagates errors based on the number of steps performed in the Trotter product. QuantumPhaseEstimation now propagates errors based on the error of its input unitary.

  dev = qml.device('null.qubit') 
  hamiltonian = qml.dot([1.0, 0.5, -0.25], [qml.X(0), qml.Y(0), qml.Z(0)])
  
  @qml.qnode(dev) 
  def circuit(): 
     qml.TrotterProduct(hamiltonian, time=0.1, order=2) 
     qml.QuantumPhaseEstimation(MyErrorOperation(0.01, wires=0), estimation_wires=[1, 2, 3]) 
     return qml.state()
  

  Again, the total spectral norm error of the circuit can be calculated using qml.specs:

  >>> qml.specs(circuit)()["errors"] 
  {'SpectralNormError': SpectralNormError(0.07616666666666666)} 
  

  Check out our error propagation demo to see how to use these new features in a real-world example!

Access an extended arsenal of quantum algorithms 🏹

• The Fast Approximate BLock-Encodings (FABLE) algorithm for embedding a matrix into a quantum circuit as outlined in arXiv:2205.00081 is now accessible via the qml.FABLE template. (#5107)

  The usage of qml.FABLE is similar to qml.BlockEncode but provides a more efficient circuit construction at the cost of a user-defined approximation level, tol. The number of wires that qml.FABLE operates on is 2*n + 1, where n defines the dimension of the $2^n \times 2^n$ matrix that we want to block-encode.

  import numpy as np
  
  A = np.array([[0.1, 0.2], [0.3, 0.4]]) 
  dev = qml.device('default.qubit', wires=3)
  
  @qml.qnode(dev) 
  def circuit(): 
     qml.FABLE(A, tol = 0.001, wires=range(3)) 
     return qml.state() 
  

  >>> mat = qml.matrix(circuit)() 
  >>> 2 * mat[0:2, 0:2] 
  array([[0.1+0.j, 0.2+0.j], [0.3+0.j, 0.4+0.j]]) 
  

• A high-level interface for amplitude amplification and its variants is now available via the new qml.AmplitudeAmplification template. (#5160)

  Based on arXiv:quant-ph/0005055, given a state $\vert \Psi \rangle = \alpha \vert \phi \rangle + \beta \vert \phi^{\perp} \rangle$, qml.AmplitudeAmplification amplifies the amplitude of $\vert \phi \rangle$.

  Here's an example with a target state $\vert \phi \rangle = \vert 2 \rangle = \vert 010 \rangle$, an input state $\vert \Psi \rangle = H^{\otimes 3} \vert 000 \rangle$, as well as an oracle that flips the sign of $\vert \phi \rangle$ and does nothing to $\vert \phi^{\perp} \rangle$, which can be achieved in this case through qml.FlipSign.

  @qml.prod 
  def generator(wires): 
     for wire in wires: 
         qml.Hadamard(wires=wire)
  
  U = generator(wires=range(3)) 
  O = qml.FlipSign(2, wires=range(3)) 
  

  Here, U is a quantum operation that is created by decorating a quantum function with @qml.prod. This could alternatively be done by creating a user-defined custom operation with a decomposition. Amplitude amplification can then be set up within a circuit:

  dev = qml.device("default.qubit")
  
  @qml.qnode(dev) 
  def circuit(): 
     generator(wires=range(3)) # prepares |Psi> = U|0> 
     qml.AmplitudeAmplification(U, O, iters=10)
     return qml.probs(wires=range(3)) 
  

  >>> print(np.round(circuit(), 3)) 
  [0.01 0.01 0.931 0.01 0.01 0.01 0.01 0.01 ] 
  

  As expected, we amplify the $\vert 2 \rangle$ state.

• Reflecting about a given quantum state is now available via qml.Reflection. This operation is very useful in the amplitude amplification algorithm and offers a generalization of qml.FlipSign, which operates on basis states. (#5159)

  qml.Reflection works by providing an operation, $U$, that prepares the desired state, $\vert \psi \rangle$, that we want to reflect about. In other words, $U$ is such that $U \vert 0 \rangle = \vert \psi \rangle$. In PennyLane, $U$ must be an Operator. For example, if we want to reflect about $\vert \psi \rangle = \vert + \rangle$, then $U = H$:

  U = qml.Hadamard(wires=0)
  dev = qml.device('default.qubit') 
  
  @qml.qnode(dev) 
  def circuit(): 
     qml.Reflection(U) 
     return qml.state() 
  

  >>> circuit() 
  tensor([0.-6.123234e-17j, 1.+6.123234e-17j], requires_grad=True) 
  

• Performing qubitization is now easily accessible with the new qml.Qubitization operator. (#5500)

  qml.Qubitization encodes a Hamiltonian into a suitable unitary operator. When applied in conjunction with quantum phase estimation (QPE), it allows for computing the eigenvalue of an eigenvector of the given Hamiltonian.

  H = qml.dot([0.1, 0.3, -0.3], [qml.Z(0), qml.Z(1), qml.Z(0) @ qml.Z(2)]) 
  @qml.qnode(qml.device("default.qubit")) 
  def circuit(): 
     # initialize the eigenvector 
     qml.PauliX(2) 
     # apply QPE 
     measurements = qml.iterative_qpe(
         qml.Qubitization(H, control = [3,4]), ancilla = 5, iters = 3
     ) 
     return qml.probs(op = measurements) 
  

Make use of more methods to map from molecules 🗺️

• A new function called qml.bravyi_kitaev has been added to perform the Bravyi-Kitaev mapping of fermionic Hamiltonians to qubit Hamiltonians. (#5390)

  This function presents an alternative mapping to qml.jordan_wigner or qml.parity_transform which can help us measure expectation values more efficiently on hardware. Simply provide a fermionic Hamiltonian (created from from_string, FermiA, FermiC, FermiSentence, or FermiWord) and the number of qubits / spin orbitals in the system, n:

  >>> fermi_ham = qml.fermi.from_string('0+ 1+ 1- 0-') 
  >>> qubit_ham = qml.bravyi_kitaev(fermi_ham, n=6, tol=0.0) 
  >>> print(qubit_ham) 
  0.25 * I(0) + -0.25 * Z(0) + -0.25 * (Z(0) @ Z(1)) + 0.25 * Z(1) 
  

• The qml.qchem.hf_state function has been upgraded to be compatible with qml.parity_transform and the new Bravyi-Kitaev mapping (qml.bravyi_kitaev). (#5472) (#5472)

  >>> state_bk = qml.qchem.hf_state(2, 6, basis="bravyi_kitaev") 
  >>> print(state_bk) 
  [1 0 0 0 0 0] 
  >>> state_parity = qml.qchem.hf_state(2, 6, basis="parity") 
  >>> print(state_parity) 
  [1 0 0 0 0 0] 
  

Calculate dynamical Lie algebras 👾

The dynamical Lie algebra (DLA) of a set of operators captures the range of unitary evolutions that the operators can generate. In v0.36 of PennyLane, we have added support for calculating important DLA concepts including:

• A new qml.lie_closure function to compute the Lie closure of a list of operators, providing one way to obtain the DLA. (#5161) (#5169) (#5627)

  For a list of operators ops = [op1, op2, op3, ..], one computes all nested commutators between ops until no new operators are generated from commutation. All these operators together form the DLA, see e.g. section IIB of arXiv:2308.01432.

  Take for example the following operators:

  from pennylane import X, Y, Z 
  ops = [X(0) @ X(1), Z(0), Z(1)] 
  

  A first round of commutators between all elements yields the new operators Y(0) @ X(1) and X(0) @ Y(1) (omitting scalar prefactors).

  >>> qml.commutator(X(0) @ X(1), Z(0)) 
  -2j * (Y(0) @ X(1)) 
  >>> qml.commutator(X(0) @ X(1), Z(1)) 
  -2j * (X(0) @ Y(1)) 
  

  A next round of commutators between all elements further yields the new operator Y(0) @ Y(1).

  >>> qml.commutator(X(0) @ Y(1), Z(0)) 
  -2j * (Y(0) @ Y(1)) 
  

  After that, no new operators emerge from taking nested commutators and we have the resulting DLA. This can now be done in short via qml.lie_closure as follows.

  >>> ops = [X(0) @ X(1), Z(0), Z(1)] 
  >>> dla = qml.lie_closure(ops) 
  >>> dla 
  [X(0) @ X(1), Z(0), Z(1), -1.0 * (Y(0) @ X(1)), -1.0 * (X(0) @ Y(1)), -1.0 * (Y(0) @ Y(1))] 
  

• Computing the structure constants (the adjoint representation) of a dynamical Lie algebra. (5406)

  For example, we can compute the adjoint representation of the transverse field Ising model DLA.

  >>> dla = [X(0) @ X(1), Z(0), Z(1), Y(0) @ X(1), X(0) @ Y(1), Y(0) @ Y(1)] 
  >>> structure_const = qml.structure_constants(dla) 
  >>> structure_const.shape 
  (6, 6, 6) 
  

  Visit the documentation of qml.structure_constants to understand how structure constants are a useful way to represent a DLA.

• Computing the center of a dynamical Lie algebra. (#5477)

  Given a DLA g, we can now compute its centre. The center is the collection of operators that commute with all other operators in the DLA.

  >>> g = [X(0), X(1) @ X(0), Y(1), Z(1) @ X(0)] 
  >>> qml.center(g) 
  [X(0)] 
  

  To help explain these concepts, check out the dynamical Lie algebras demo.

Improvements 🛠

Simulate mixed-state qutrit systems

• Mixed qutrit states can now be simulated with the default.qutrit.mixed device. (#5495) (#5451) (#5186) (#5082) (#5213)

  Thanks to contributors from the University of British Columbia, a mixed-state qutrit device is now available for simulation, providing a noise-capable equivalent to default.qutrit.

  dev = qml.device("default.qutrit.mixed")
  
  def circuit(): 
     qml.TRY(0.1, wires=0)
  
  @qml.qnode(dev) 
  def shots_circuit(): 
     circuit() 
     return qml.sample(), qml.expval(qml.GellMann(wires=0, index=1))
  
  @qml.qnode(dev) 
  def density_matrix_circuit(): 
     circuit() 
     return qml.state() 
  

  >>> shots_circuit(shots=5) 
  (array([0, 0, 0, 0, 0]), 0.19999999999999996) 
  >>> density_matrix_circuit() 
  tensor([[0.99750208+0.j, 0.04991671+0.j, 0. +0.j], [0.04991671+0.j, 0.00249792+0.j, 0. +0.j], [0. +0.j, 0. +0.j, 0. +0.j]], requires_grad=True) 
  

  However, there's one crucial ingredient that we still need to add: support for qutrit noise operations. Keep your eyes peeled for this to arrive in the coming releases!

Work easily and efficiently with operators

• This release completes the main phase of PennyLane's switchover to an updated approach for handling arithmetic operations between operators. The new approach is now enabled by default and is intended to realize a few objectives: 1. To make it as easy to work with PennyLane operators as it would be with pen and paper. 2. To improve the efficiency of operator arithmetic.

  In many cases, this update should not break code. If issues do arise, check out the updated operator troubleshooting page and don't hesitate to reach out to us on the PennyLane discussion forum. As a last resort the old behaviour can be enabled by calling qml.operation.disable_new_opmath(), but this is not recommended because support will not continue in future PennyLane versions (v0.36 and higher). (#5269)

• A new class called qml.ops.LinearCombination has been introduced. In essence, this class is an updated equivalent of the now-deprecated qml.ops.Hamiltonian but for usage with the new operator arithmetic. (#5216)

• qml.ops.Sum now supports storing grouping information. Grouping type and method can be specified during construction using the grouping_type and method keyword arguments of qml.dot, qml.sum, or qml.ops.Sum. The grouping indices are stored in Sum.grouping_indices. (#5179)

  a = qml.X(0) 
  b = qml.prod(qml.X(0), qml.X(1)) 
  c = qml.Z(0) 
  obs = [a, b, c] 
  coeffs = [1.0, 2.0, 3.0]
  
  op = qml.dot(coeffs, obs, grouping_type="qwc") 
  

  >>> op.grouping_indices 
  ((2,), (0, 1)) 
  

  Additionally, grouping_type and method can be set or changed after construction using Sum.compute_grouping():

  a = qml.X(0) 
  b = qml.prod(qml.X(0), qml.X(1)) 
  c = qml.Z(0) 
  obs = [a, b, c] 
  coeffs = [1.0, 2.0, 3.0]
  
  op = qml.dot(coeffs, obs) 
  

  >>> op.grouping_indices is None 
  True 
  >>> op.compute_grouping(grouping_type="qwc") 
  >>> op.grouping_indices 
  ((2,), (0, 1)) 
  

  Note that the grouping indices refer to the lists returned by Sum.terms(), not Sum.operands.

• A new function called qml.operation.convert_to_legacy_H that converts Sum, SProd, and Prod to Hamiltonian instances has been added. This function is intended for developers and will be removed in a future release without a deprecation cycle. (#5309)

• The qml.is_commuting function now accepts Sum, SProd, and Prod instances. (#5351)

• Operators can now be left-multiplied by NumPy arrays (i.e., arr * op). (#5361)

• op.generator(), where op is an Operator instance, now returns operators consistent with the global setting for qml.operator.active_new_opmath() wherever possible. Sum, SProd and Prod instances will be returned even after disabling the new operator arithmetic in cases where they offer additional functionality not available using legacy operators. (#5253) (#5410) (#5411) (#5421)

• Prod instances temporarily have a new obs property, which helps smoothen the transition of the new operator arithmetic system. In particular, this is aimed at preventing breaking code that uses Tensor.obs. The property has been immediately deprecated. Moving forward, we recommend using op.operands. (#5539)

• qml.ApproxTimeEvolution is now compatible with any operator that has a defined pauli_rep. (#5362)

• Hamiltonian.pauli_rep is now defined if the Hamiltonian is a linear combination of Pauli operators. (#5377)

• Prod instances created with qutrit operators now have a defined eigvals() method. (#5400)

• qml.transforms.hamiltonian_expand and qml.transforms.sum_expand can now handle multi-term observables with a constant offset (i.e., terms like qml.I()). (#5414) (#5543)

• qml.qchem.taper_operation is now compatible with the new operator arithmetic. (#5326)

• The warning for an observable that might not be hermitian in QNode executions has been removed. This enables jit-compilation. (#5506)

• qml.transforms.split_non_commuting will now work with single-term operator arithmetic. (#5314)

• LinearCombination and Sum now accept _grouping_indices on initialization. This addition is relevant to developers only. (#5524)

• Calculating the dense, differentiable matrix for PauliSentence and operators with Pauli sentences is now faster. (#5578)

Community contributions 🥳

• ExpectationMP, VarianceMP, CountsMP, and SampleMP now have a process_counts method (similar to process_samples). This allows for calculating measurements given a counts dictionary. (#5256) (#5395)

• Type-hinting has been added in the Operator class for better interpretability. (#5490)

• An alternate strategy for sampling with multiple different shots values has been implemented via the shots.bins() method, which samples all shots at once and then processes each separately. (#5476)

Mid-circuit measurements and dynamic circuits

• A new module called qml.capture that will contain PennyLane's own capturing mechanism for hybrid quantum-classical programs has been added. (#5509)

• The dynamic_one_shot transform has been introduced, enabling dynamic circuit execution on circuits with finite shots and devices that natively support mid-circuit measurements. (#5266)

• The QubitDevice class and children classes support the dynamic_one_shot transform provided that they support mid-circuit measurement operations natively. (#5317)

• default.qubit can now be provided a random seed for sampling mid-circuit measurements with finite shots. This (1) ensures that random behaviour is more consistent with dynamic_one_shot and defer_measurements and (2) makes our continuous-integration (CI) have less failures due to stochasticity. (#5337)

Performance and broadcasting

• Gradient transforms may now be applied to batched/broadcasted QNodes as long as the broadcasting is in non-trainable parameters. (#5452)

• The performance of computing the matrix of qml.QFT has been improved. (#5351)

• qml.transforms.broadcast_expand now supports shot vectors when returning qml.sample(). (#5473)

• LightningVJPs is now compatible with Lightning devices using the new device API. (#5469)

Device capabilities

• Obtaining classical shadows using the default.clifford device is now compatible with stim v1.13.0. (#5409)

• default.mixed has improved support for sampling-based measurements with non-NumPy interfaces. (#5514) (#5530)

• default.mixed now supports arbitrary state-based measurements with qml.Snapshot. (#5552)

• null.qubit has been upgraded to the new device API and has support for all measurements and various modes of differentiation. (#5211)

Other improvements

• Entanglement entropy can now be calculated with qml.math.vn_entanglement_entropy, which computes the von Neumann entanglement entropy from a density matrix. A corresponding QNode transform, qml.qinfo.vn_entanglement_entropy, has also been added. (#5306)

• qml.draw and qml.draw_mpl will now attempt to sort the wires if no wire order is provided by the user or the device. (#5576)

• A clear error message is added in KerasLayer when using the newest version of TensorFlow with Keras 3 (which is not currently compatible with KerasLayer), linking to instructions to enable Keras 2. (#5488)

• qml.ops.Conditional now stores the data, num_params, and ndim_param attributes of the operator it wraps. (#5473)

• The molecular_hamiltonian function calls PySCF directly when method='pyscf' is selected. (#5118)

• cache_execute has been replaced with an alternate implementation based on @transform. (#5318)

• QNodes now defer diff_method validation to the device under the new device API. (#5176)

• The device test suite has been extended to cover gradient methods, templates and arithmetic observables. (#5273) (#5518)

• A typo and string formatting mistake have been fixed in the error message for ClassicalShadow._convert_to_pauli_words when the input is not a valid pauli_rep. (#5572)

• Circuits running on lightning.qubit and that return qml.state() now preserve the dtype when specified. (#5547)

Breaking changes 💔

• qml.matrix() called on the following will now raise an error if wire_order is not specified: * tapes with more than one wire * quantum functions * Operator classes where num_wires does not equal to 1 * QNodes if the device does not have wires specified. * PauliWords and PauliSentences with more than one wire. (#5328) (#5359)

• single_tape_transform, batch_transform, qfunc_transform, op_transform, gradient_transform and hessian_transform have been removed. Instead, switch to using the new qml.transform function. Please refer to the transform docs <https://docs.pennylane.ai/en/stable/code/qml_transforms.html#custom-transforms>_ to see how this can be done. (#5339)

• Attempting to multiply PauliWord and PauliSentence with * will raise an error. Instead, use @ to conform with the PennyLane convention. (#5341)

• DefaultQubit now uses a pre-emptive key-splitting strategy to avoid reusing JAX PRNG keys throughout a single execute call. (#5515)

• qml.pauli.pauli_mult and qml.pauli.pauli_mult_with_phase have been removed. Instead, use qml.simplify(qml.prod(pauli_1, pauli_2)) to get the reduced operator. (#5324)

>>> op = qml.simplify(qml.prod(qml.PauliX(0), qml.PauliZ(0))) 
>>> op -1j*(PauliY(wires=[0])) 
>>> [phase], [base] = op.terms() 
>>> phase, base 
(-1j, PauliY(wires=[0])) 

• The dynamic_one_shot transform now uses sampling (SampleMP) to get back the values of the mid-circuit measurements. (#5486)

• Operator dunder methods now combine like-operator arithmetic classes via lazy=False. This reduces the chances of getting a RecursionError and makes nested operators easier to work with. (#5478)

• The private functions _pauli_mult, _binary_matrix and _get_pauli_map from the pauli module have been removed. The same functionality can be achieved using newer features in the pauli module. (#5323)

• MeasurementProcess.name and MeasurementProcess.data have been removed. Use MeasurementProcess.obs.name and MeasurementProcess.obs.data instead. (#5321)

• Operator.validate_subspace(subspace) has been removed. Instead, use qml.ops.qutrit.validate_subspace(subspace). (#5311)

• The contents of qml.interfaces has been moved inside qml.workflow. The old import path no longer exists. (#5329)

• Since default.mixed does not support snapshots with measurements, attempting to do so will result in a DeviceError instead of getting the density matrix. (#5416)

• LinearCombination._obs_data has been removed. You can still use LinearCombination.compare to check mathematical equivalence between a LinearCombination and another operator. (#5504)

Deprecations 👋

• Accessing qml.ops.Hamiltonian is deprecated because it points to the old version of the class that may not be compatible with the new approach to operator arithmetic. Instead, using qml.Hamiltonian is recommended because it dispatches to the LinearCombination class when the new approach to operator arithmetic is enabled. This will allow you to continue to use qml.Hamiltonian with existing code without needing to make any changes. (#5393)

• qml.load has been deprecated. Instead, please use the functions outlined in the Importing workflows quickstart guide. (#5312)

• Specifying control_values with a bit string in qml.MultiControlledX has been deprecated. Instead, use a list of booleans or 1s and 0s. (#5352)

• qml.from_qasm_file has been deprecated. Instead, please open the file and then load its content using qml.from_qasm. (#5331)

>>> with open("test.qasm", "r") as f: 
...    circuit = qml.from_qasm(f.read()) 

Documentation 📝

• A new page explaining the shapes and nesting of return types has been added. (#5418)

• Redundant documentation for the evolve function has been removed. (#5347)

• The final example in the compile docstring has been updated to use transforms correctly. (#5348)

• A link to the demos for using qml.SpecialUnitary and qml.QNGOptimizer has been added to their respective docstrings. (#5376)

• A code example in the qml.measure docstring has been added that showcases returning mid-circuit measurement statistics from QNodes. (#5441)

• The computational basis convention used for qml.measure — 0 and 1 rather than ±1 — has been clarified in its docstring. (#5474)

• A new Release news section has been added to the table of contents, containing release notes, deprecations, and other pages focusing on recent changes. (#5548)

• A summary of all changes has been added in the "Updated Operators" page in the new "Release news" section in the docs. (#5483) (#5636)

Bug fixes 🐛

• Patches the QNode so that parameter-shift will be considered best with lightning if qml.metric_tensor is in the transform program. (#5624)

• Stopped printing the ID of qcut.MeasureNode and qcut.PrepareNode in tape drawing. (#5613)

• Improves the error message for setting shots on the new device interface, or trying to access a property that no longer exists. (#5616)

• Fixed a bug where qml.draw and qml.draw_mpl incorrectly raised errors for circuits collecting statistics on mid-circuit measurements while using qml.defer_measurements. (#5610)

• Using shot vectors with param_shift(... broadcast=True) caused a bug. This combination is no longer supported and will be added again in the next release. Fixed a bug with custom gradient recipes that only consist of unshifted terms. (#5612) (#5623)

• qml.counts now returns the same keys with dynamic_one_shot and defer_measurements. (#5587)

• null.qubit now automatically supports any operation without a decomposition. (#5582)

• Fixed a bug where the shape and type of derivatives obtained by applying a gradient transform to a QNode differed based on whether the QNode uses classical coprocessing. (#4945)

• ApproxTimeEvolution, CommutingEvolution, QDrift, and TrotterProduct now de-queue their input observable. (#5524)

• (In)equality of qml.HilbertSchmidt instances is now reported correctly by qml.equal. (#5538)

• qml.ParticleConservingU1 and qml.ParticleConservingU2 no longer raise an error when the initial state is not specified but default to the all-zeros state. (#5535)

• qml.counts no longer returns negative samples when measuring 8 or more wires. (#5544) (#5556)

• The dynamic_one_shot transform now works with broadcasting. (#5473)

• Diagonalizing gates are now applied when measuring qml.probs on non-computational basis states on a Lightning device. (#5529)

• two_qubit_decomposition no longer diverges at a special case of a unitary matrix. (#5448)

• The qml.QNSPSAOptimizer now correctly handles optimization for legacy devices that do not follow the new device API. (#5497)

• Operators applied to all wires are now drawn correctly in a circuit with mid-circuit measurements. (#5501)

• Fixed a bug where certain unary mid-circuit measurement expressions would raise an uncaught error. (#5480)

• Probabilities now sum to 1 when using the torch interface with default_dtype set to torch.float32. (#5462)

• Tensorflow can now handle devices with float32 results but float64 input parameters. (#5446)

• Fixed a bug where the argnum keyword argument of qml.gradients.stoch_pulse_grad references the wrong parameters in a tape, creating an inconsistency with other differentiation methods and preventing some use cases. (#5458)

• Bounded value failures due to numerical noise with calls to np.random.binomial is now avoided. (#5447)

• Using @ with legacy Hamiltonian instances now properly de-queues the previously existing operations. (#5455)

• The QNSPSAOptimizer now properly handles differentiable parameters, resulting in being able to use it for more than one optimization step. (#5439)

• The QNode interface now resets if an error occurs during execution. (#5449)

• Failing tests due to changes with Lightning's adjoint diff pipeline have been fixed. (#5450)

• Failures occurring when making autoray-dispatched calls to Torch with paired CPU data have been fixed. (#5438)

• jax.jit now works with qml.sample with a multi-wire observable. (#5422)

• qml.qinfo.quantum_fisher now works with non-default.qubit devices. (#5423)

• We no longer perform unwanted dtype promotion in the pauli_rep of SProd instances when using Tensorflow. (#5246)

• Fixed TestQubitIntegration.test_counts in tests/interfaces/test_jax_qnode.py to always produce counts for all outcomes. (#5336)

• Fixed PauliSentence.to_mat(wire_order) to support identities with wires. (#5407)

• CompositeOp.map_wires now correctly maps the overlapping_ops property. (#5430)

• DefaultQubit.supports_derivatives has been updated to correctly handle circuits containing mid-circuit measurements and adjoint differentiation. (#5434)

• SampleMP, ExpectationMP, CountsMP, and VarianceMP constructed with eigvals can now properly process samples. (#5463)

• Fixed a bug in hamiltonian_expand that produces incorrect output dimensions when shot vectors are combined with parameter broadcasting. (#5494)

• default.qubit now allows measuring Identity on no wires and observables containing Identity on no wires. (#5570)

• Fixed a bug where TorchLayer does not work with shot vectors. (#5492)

• Fixed a bug where the output shape of a QNode returning a list containing a single measurement is incorrect when combined with shot vectors. (#5492)

• Fixed a bug in qml.math.kron that makes Torch incompatible with NumPy. (#5540)

• Fixed a bug in _group_measurements that fails to group measurements with commuting observables when they are operands of Prod. (#5525)

• qml.equal can now be used with sums and products that contain operators on no wires like I and GlobalPhase. (#5562)

• CompositeOp.has_diagonalizing_gates now does a more complete check of the base operators to ensure consistency between op.has_diagonalzing_gates and op.diagonalizing_gates() (#5603)

• Updated the method kwarg of qml.TrotterProduct().error() to be more clear that we are computing upper-bounds. (#5637)

Contributors ✍️

This release contains contributions from (in alphabetical order):

Tarun Kumar Allamsetty, Guillermo Alonso, Mikhail Andrenkov, Utkarsh Azad, Gabriel Bottrill, Thomas Bromley, Astral Cai, Diksha Dhawan, Isaac De Vlugt, Amintor Dusko, Pietropaolo Frisoni, Lillian M. A. Frederiksen, Diego Guala, Austin Huang, Soran Jahangiri, Korbinian Kottmann, Christina Lee, Vincent Michaud-Rioux, Mudit Pandey, Kenya Sakka, Jay Soni, Matthew Silverman, David Wierichs.

Bug fixes 🐛

• Lightning simulators need special handling of diagonalizing gates when performing sampling measurements. (#5343)

• Updated the lower bound on the required Catalyst version to v0.5.0. (#5320)

Contributors ✍️

This release contains contributions from (in alphabetical order):

Vincent Michaud-Rioux, Erick Ochoa Lopez.

New features since last release

Qiskit 1.0 integration 🔌

• This version of PennyLane makes it easier to import circuits from Qiskit. (#5218) (#5168)

  The qml.from_qiskit function converts a Qiskit QuantumCircuit into a PennyLane quantum function. Although qml.from_qiskit already exists in PennyLane, we have made a number of improvements to make importing from Qiskit easier. And yes — qml.from_qiskit functionality is compatible with both Qiskit 1.0 and earlier versions! Here's a comprehensive list of the improvements:

  • You can now append PennyLane measurements onto the quantum function returned by qml.from_qiskit. Consider this simple Qiskit circuit:

    import pennylane as qml 
    from qiskit import QuantumCircuit
    
    qc = QuantumCircuit(2) 
    qc.rx(0.785, 0) 
    qc.ry(1.57, 1) 
    

    We can convert it into a PennyLane QNode in just a few lines, with PennyLane measurements easily included:

    >>> dev = qml.device("default.qubit") 
    >>> measurements = qml.expval(qml.Z(0) @ qml.Z(1)) 
    >>> qfunc = qml.from_qiskit(qc, measurements=measurements) 
    >>> qnode = qml.QNode(qfunc, dev) 
    >>> qnode() 
    tensor(0.00056331, requires_grad=True) 
    

  • Quantum circuits that already contain Qiskit-side measurements can be faithfully converted with qml.from_qiskit. Consider this example Qiskit circuit:

    qc = QuantumCircuit(3, 2) # Teleportation
    
    qc.rx(0.9, 0) # Prepare input state on qubit 0
    
    qc.h(1) # Prepare Bell state on qubits 1 and 2 qc.cx(1, 2)
    
    qc.cx(0, 1) # Perform teleportation 
    qc.h(0) 
    qc.measure(0, 0) 
    qc.measure(1, 1)
    
    with qc.if_test((1, 1)): # Perform first conditional 
        qc.x(2) 
    

    This circuit can be converted into PennyLane with the Qiskit measurements still accessible. For example, we can use those results as inputs to a mid-circuit measurement in PennyLane:

    @qml.qnode(dev) 
    def teleport(): 
        m0, m1 = qml.from_qiskit(qc)() 
        qml.cond(m0, qml.Z)(2) 
        return qml.density_matrix(2) 
    

    >>> teleport() 
    tensor([[0.81080498+0.j , 0. +0.39166345j], 
    [0. -0.39166345j, 0.18919502+0.j ]], requires_grad=True) 
    

  • It is now more intuitive to handle and differentiate parametrized Qiskit circuits. Consider the following circuit:

    from qiskit.circuit import Parameter 
    from pennylane import numpy as np
    
    angle0 = Parameter("x") angle1 = Parameter("y")
    
    qc = QuantumCircuit(2, 2) 
    qc.rx(angle0, 0) 
    qc.ry(angle1, 1) 
    qc.cx(1, 0) 
    

    We can convert this circuit into a QNode with two arguments, corresponding to x and y:

    measurements = qml.expval(qml.PauliZ(0)) 
    qfunc = qml.from_qiskit(qc, measurements) 
    qnode = qml.QNode(qfunc, dev) 
    

    The QNode can be evaluated and differentiated:

    >>> x, y = np.array([0.4, 0.5], requires_grad=True) 
    >>> qnode(x, y) 
    tensor(0.80830707, requires_grad=True) 
    
    >>> qml.grad(qnode)(x, y) 
    (tensor(-0.34174675, requires_grad=True), tensor(-0.44158016, requires_grad=True)) 
    

    This shows how easy it is to make a Qiskit circuit differentiable with PennyLane.

  • In addition to circuits, it is also possible to convert operators from Qiskit to PennyLane with a new function called qml.from_qiskit_op. (#5251)

    A Qiskit SparsePauliOp can be converted to a PennyLane operator using qml.from_qiskit_op:

    >>> from qiskit.quantum_info import SparsePauliOp 
    >>> qiskit_op = SparsePauliOp(["II", "XY"]) 
    >>> qiskit_op SparsePauliOp(['II', 'XY'], coeffs=[1.+0.j, 1.+0.j]) 
    >>> pl_op = qml.from_qiskit_op(qiskit_op) 
    >>> pl_op I(0) + X(1) @ Y(0) 
    

    Combined with qml.from_qiskit, it becomes easy to quickly calculate quantities like expectation values by converting the whole workflow to PennyLane:

    qc = QuantumCircuit(2) # Create circuit 
    qc.rx(0.785, 0) 
    qc.ry(1.57, 1)
    
    measurements = qml.expval(pl_op) # Create QNode 
    qfunc = qml.from_qiskit(qc, measurements) 
    qnode = qml.QNode(qfunc, dev) 
    

    >>> qnode() # Evaluate! 
    tensor(0.29317504, requires_grad=True) 
    

Native mid-circuit measurements on Default Qubit 💡

• Mid-circuit measurements can now be more scalable and efficient in finite-shots mode with default.qubit by simulating them in a similar way to what happens on quantum hardware. (#5088) (#5120)

  Previously, mid-circuit measurements (MCMs) would be automatically replaced with an additional qubit using the @qml.defer_measurements transform. The circuit below would have required thousands of qubits to simulate.

  Now, MCMs are performed in a similar way to quantum hardware with finite shots on default.qubit. For each shot and each time an MCM is encountered, the device evaluates the probability of projecting onto |0> or |1> and makes a random choice to collapse the circuit state. This approach works well when there are a lot of MCMs and the number of shots is not too high.

  import pennylane as qml
  
  dev = qml.device("default.qubit", shots=10)
  
  @qml.qnode(dev) 
  def f(): 
      for i in range(1967): 
          qml.Hadamard(0) 
          qml.measure(0) 
      return qml.sample(qml.PauliX(0)) 
  

  >>> f() 
  tensor([-1, -1, -1, 1, 1, -1, 1, -1, 1, -1], requires_grad=True) 
  

Work easily and efficiently with operators 🔧

• Over the past few releases, PennyLane's approach to operator arithmetic has been in the process of being overhauled. We have a few objectives:

  1. To make it as easy to work with PennyLane operators as it would be with pen and paper.
  2. To improve the efficiency of operator arithmetic.

  The updated operator arithmetic functionality is still being finalized, but can be activated using qml.operation.enable_new_opmath(). In the next release, the new behaviour will become the default, so we recommend enabling now to become familiar with the new system!

  The following updates have been made in this version of PennyLane:

  • You can now easily access Pauli operators via I, X, Y, and Z: (#5116)

    >>> from pennylane import I, X, Y, Z 
    >>> X(0) X(0) ```
    
    The original long-form names `Identity`, `PauliX`, `PauliY`, and `PauliZ` remain available, but use of the short-form names is now recommended.
    
    

  • A new qml.commutator function is now available that allows you to compute commutators between PennyLane operators. (#5051) (#5052) (#5098)

    >>> qml.commutator(X(0), Y(0)) 
    2j * Z(0) 
    

  • Operators in PennyLane can have a backend Pauli representation, which can be used to perform faster operator arithmetic. Now, the Pauli representation will be automatically used for calculations when available. (#4989) (#5001) (#5003) (#5017) (#5027)

    The Pauli representation can be optionally accessed via op.pauli_rep:

    >>> qml.operation.enable_new_opmath() 
    >>> op = X(0) + Y(0) 
    >>> op.pauli_rep 
    1.0 * X(0) + 1.0 * Y(0) 
    

  • Extensive improvements have been made to the string representations of PennyLane operators, making them shorter and possible to copy-paste as valid PennyLane code. (#5116) (#5138)

    >>> 0.5 * X(0) 
    0.5 * X(0)
    >>> 0.5 * (X(0) + Y(1)) 
    0.5 * (X(0) + Y(1)) 
    

    Sums with many terms are broken up into multiple lines, but can still be copied back as valid code:

    >>> 0.5 * (X(0) @ X(1)) + 0.7 * (X(1) @ X(2)) + 0.8 * (X(2) @ X(3)) 
    ( 
         0.5 * (X(0) @ X(1)) 
         + 0.7 * (X(1) @ X(2)) 
         + 0.8 * (X(2) @ X(3)) 
    ) 
    

  • Linear combinations of operators and operator multiplication via Sum and Prod, respectively, have been updated to reach feature parity with Hamiltonian and Tensor, respectively. This should minimize the effort to port over any existing code. (#5070) (#5132) (#5133)

    Updates include support for grouping via the pauli module:

    >>> obs = [X(0) @ Y(1), Z(0), Y(0) @ Z(1), Y(1)] 
    >>> qml.pauli.group_observables(obs) 
    [[Y(0) @ Z(1)], [X(0) @ Y(1), Y(1)], [Z(0)]] 
    

New Clifford device 🦾

• A new default.clifford device enables efficient simulation of large-scale Clifford circuits defined in PennyLane through the use of stim as a backend. (#4936) (#4954) (#5144)

  Given a circuit with only Clifford gates, one can use this device to obtain the usual range of PennyLane measurements as well as the state represented in the Tableau form of Aaronson & Gottesman (2004):

  import pennylane as qml
  
  dev = qml.device("default.clifford", tableau=True) 
  @qml.qnode(dev) 
  def circuit(): 
      qml.CNOT(wires=[0, 1]) 
      qml.PauliX(wires=[1]) 
      qml.ISWAP(wires=[0, 1]) 
      qml.Hadamard(wires=[0]) 
      return qml.state() 
  

  >>> circuit() 
  array([[0, 1, 1, 0, 0], 
        [1, 0, 1, 1, 1], 
        [0, 0, 0, 1, 0], 
        [1, 0, 0, 1, 1]]) 
  

  The default.clifford device also supports the PauliError, DepolarizingChannel, BitFlip and PhaseFlip noise channels when operating in finite-shot mode.

Improvements 🛠

Faster gradients with VJPs and other performance improvements

• Vector-Jacobian products (VJPs) can result in faster computations when the output of your quantum Node has a low dimension. They can be enabled by setting device_vjp=True when loading a QNode. In the next release of PennyLane, VJPs are planned to be used by default, when available.

  In this release, we have unlocked:

  • Adjoint device VJPs can be used with jax.jacobian, meaning that device_vjp=True is always faster when using JAX with default.qubit. (#4963)

  • PennyLane can now use lightning-provided VJPs. (#4914)

  • VJPs can be used with TensorFlow, though support has not yet been added for tf.Function and Tensorflow Autograph. (#4676)

• Measuring qml.probs is now faster due to an optimization in converting samples to counts. (#5145)

• The performance of circuit-cutting workloads with large numbers of generated tapes has been improved. (#5005)

• Queueing (AnnotatedQueue) has been removed from qml.cut_circuit and qml.cut_circuit_mc to improve performance for large workflows. (#5108)

Community contributions 🥳

• A new function called qml.fermi.parity_transform has been added for parity mapping of a fermionic Hamiltonian. (#4928)

  It is now possible to transform a fermionic Hamiltonian to a qubit Hamiltonian with parity mapping.

  import pennylane as qml 
  fermi_ham = qml.fermi.FermiWord({(0, 0) : '+', (1, 1) : '-'})
  
  qubit_ham = qml.fermi.parity_transform(fermi_ham, n=6) 
  

>>> print(qubit_ham) 
-0.25j * Y(0) + (-0.25+0j) * (X(0) @ Z(1)) + (0.25+0j) * X(0) + 0.25j * (Y(0) @ Z(1)) 

• The transform split_non_commuting now accepts measurements of type probs, sample, and counts, which accept both wires and observables. (#4972)

• The efficiency of matrix calculations when an operator is symmetric over a given set of wires has been improved. (#3601)

• The pennylane/math/quantum.py module now has support for computing the minimum entropy of a density matrix. (#3959)

  >>> x = [1, 0, 0, 1] / np.sqrt(2) 
  >>> x = qml.math.dm_from_state_vector(x) 
  >>> qml.math.min_entropy(x, indices=[0]) 
  0.6931471805599455 
  

• A function called apply_operation that applies operations to device-compatible states has been added to the new qutrit_mixed module found in qml.devices. (#5032)

• A function called measure has been added to the new qutrit_mixed module found in qml.devices that measures device-compatible states for a collection of measurement processes. (#5049)

• A partial_trace function has been added to qml.math for taking the partial trace of matrices. (#5152)

Other operator arithmetic improvements

• The following capabilities have been added for Pauli arithmetic: (#4989) (#5001) (#5003) (#5017) (#5027) (#5018)

  • You can now multiply PauliWord and PauliSentence instances by scalars (e.g., 0.5 * PauliWord({0: "X"}) or 0.5 * PauliSentence({PauliWord({0: "X"}): 1.})).

  • You can now intuitively add and subtract PauliWord and PauliSentence instances and scalars together (scalars are treated implicitly as multiples of the identity, I). For example, ps1 + pw1 + 1. for some Pauli word pw1 = PauliWord({0: "X", 1: "Y"}) and Pauli sentence ps1 = PauliSentence({pw1: 3.}).

  • You can now element-wise multiply PauliWord, PauliSentence, and operators together with qml.dot (e.g., qml.dot([0.5, -1.5, 2], [pw1, ps1, id_word]) with id_word = PauliWord({})).

  • qml.matrix now accepts PauliWord and PauliSentence instances (e.g., qml.matrix(PauliWord({0: "X"}))).

  • It is now possible to compute commutators with Pauli operators natively with the new commutator method.

    >>> op1 = PauliWord({0: "X", 1: "X"}) 
    >>> op2 = PauliWord({0: "Y"}) + PauliWord({1: "Y"}) 
    >>> op1.commutator(op2) 2j * Z(0) @ X(1) + 2j * X(0) @ Z(1) 
    

• Composite operations (e.g., those made with qml.prod and qml.sum) and scalar-product operations convert Hamiltonian and Tensor operands to Sum and Prod types, respectively. This helps avoid the mixing of incompatible operator types. (#5031) (#5063)

• qml.Identity() can be initialized without wires. Measuring it is currently not possible, though. (#5106)

• qml.dot now returns a Sum class even when all the coefficients match. (#5143)

• qml.pauli.group_observables now supports grouping Prod and SProd operators. (#5070)

• The performance of converting a PauliSentence to a Sum has been improved. (#5141) (#5150)

• Akin to qml.Hamiltonian features, the coefficients and operators that make up composite operators formed via Sum or Prod can now be accessed with the terms() method. (#5132) (#5133) (#5164)

  >>> qml.operation.enable_new_opmath() 
  >>> op = X(0) @ (0.5 * X(1) + X(2)) 
  >>> op.terms() 
  ([0.5, 1.0], 
   [X(1) @ X(0), 
    X(2) @ X(0)]) 
  

• String representations of ParametrizedHamiltonian have been updated to match the style of other PL operators. (#5215)

Other improvements

• The pl-device-test suite is now compatible with the qml.devices.Device interface. (#5229)

• The QSVT operation now determines its data from the block encoding and projector operator data. (#5226) (#5248)

• The BlockEncode operator is now JIT-compatible with JAX. (#5110)

• The qml.qsvt function uses qml.GlobalPhase instead of qml.exp to define a global phase. (#5105)

• The tests/ops/functions/conftest.py test has been updated to ensure that all operator types are tested for validity. (#4978)

• A new pennylane.workflow module has been added. This module now contains qnode.py, execution.py, set_shots.py, jacobian_products.py, and the submodule interfaces. (#5023)

• A more informative error is now raised when calling adjoint_jacobian with trainable state-prep operations. (#5026)

• qml.workflow.get_transform_program and qml.workflow.construct_batch have been added to inspect the transform program and batch of tapes at different stages. (#5084)

• All custom controlled operations such as CRX, CZ, CNOT, ControlledPhaseShift now inherit from ControlledOp, giving them additional properties such as control_wire and control_values. Calling qml.ctrl on RX, RY, RZ, Rot, and PhaseShift with a single control wire will return gates of types CRX, CRY, etc. as opposed to a general Controlled operator. (#5069) (#5199)

• The CI will now fail if coverage data fails to upload to codecov. Previously, it would silently pass and the codecov check itself would never execute. (#5101)

• qml.ctrl called on operators with custom controlled versions will now return instances of the custom class, and it will flatten nested controlled operators to a single multi-controlled operation. For PauliX, CNOT, Toffoli, and MultiControlledX, calling qml.ctrl will always resolve to the best option in CNOT, Toffoli, or MultiControlledX depending on the number of control wires and control values. (#5125)

• Unwanted warning filters have been removed from tests and no PennyLaneDeprecationWarnings are being raised unexpectedly. (#5122)

• New error tracking and propagation functionality has been added (#5115) (#5121)

• The method map_batch_transform has been replaced with the method _batch_transform implemented in TransformDispatcher. (#5212)

• TransformDispatcher can now dispatch onto a batch of tapes, making it easier to compose transforms when working in the tape paradigm. (#5163)

• qml.ctrl is now a simple wrapper that either calls PennyLane's built in create_controlled_op or uses the Catalyst implementation. (#5247)

• Controlled composite operations can now be decomposed using ZYZ rotations. (#5242)

• New functions called qml.devices.modifiers.simulator_tracking and qml.devices.modifiers.single_tape_support have been added to add basic default behavior onto a device class. (#5200)

Breaking changes 💔

• Passing additional arguments to a transform that decorates a QNode must now be done through the use of functools.partial. (#5046)

• qml.ExpvalCost has been removed. Users should use qml.expval() moving forward. (#5097)

• Caching of executions is now turned off by default when max_diff == 1, as the classical overhead cost outweighs the probability that duplicate circuits exists. (#5243)

• The entry point convention registering compilers with PennyLane has changed. (#5140)

  To allow for packages to register multiple compilers with PennyLane, the entry_points convention under the designated group name pennylane.compilers has been modified.

  Previously, compilers would register qjit (JIT decorator), ops (compiler-specific operations), and context (for tracing and program capture).

  Now, compilers must register compiler_name.qjit, compiler_name.ops, and compiler_name.context, where compiler_name is replaced by the name of the provided compiler.

  For more information, please see the documentation on adding compilers.

• PennyLane source code is now compatible with the latest version of black. (#5112) (#5119)

• gradient_analysis_and_validation has been renamed to find_and_validate_gradient_methods. Instead of returning a list, it now returns a dictionary of gradient methods for each parameter index, and no longer mutates the tape. (#5035)

• Multiplying two PauliWord instances no longer returns a tuple (new_word, coeff) but instead PauliSentence({new_word: coeff}). The old behavior is still available with the private method PauliWord._matmul(other) for faster processing. (#5045)

• Observable.return_type has been removed. Instead, you should inspect the type of the surrounding measurement process. (#5044)

• ClassicalShadow.entropy() no longer needs an atol keyword as a better method to estimate entropies from approximate density matrix reconstructions (with potentially negative eigenvalues). (#5048)

• Controlled operators with a custom controlled version decompose like how their controlled counterpart decomposes as opposed to decomposing into their controlled version. (#5069) (#5125)

  For example:

  >>> qml.ctrl(qml.RX(0.123, wires=1), control=0).decomposition() 
  [ 
    RZ(1.5707963267948966, wires=[1]), 
    RY(0.0615, wires=[1]), 
    CNOT(wires=[0, 1]), 
    RY(-0.0615, wires=[1]), 
    CNOT(wires=[0, 1]), 
    RZ(-1.5707963267948966, wires=[1]) 
  ] 
  

• QuantumScript.is_sampled and QuantumScript.all_sampled have been removed. Users should now validate these properties manually. (#5072)

• qml.transforms.one_qubit_decomposition and qml.transforms.two_qubit_decomposition have been removed. Instead, you should use qml.ops.one_qubit_decomposition and qml.ops.two_qubit_decomposition. (#5091)

Deprecations 👋

• Calling qml.matrix without providing a wire_order on objects where the wire order could be ambiguous now raises a warning. In the future, the wire_order argument will be required in these cases. (#5039)

• Operator.validate_subspace(subspace) has been relocated to the qml.ops.qutrit.parametric_ops module and will be removed from the Operator class in an upcoming release. (#5067)

• Matrix and tensor products between PauliWord and PauliSentence instances are done using the @ operator, * will be used only for scalar multiplication. Note also the breaking change that the product of two PauliWord instances now returns a PauliSentence instead of a tuple (new_word, coeff). (#4989) (#5054)

• MeasurementProcess.name and MeasurementProcess.data are now deprecated, as they contain dummy values that are no longer needed. (#5047) (#5071) (#5076) (#5122)

• qml.pauli.pauli_mult and qml.pauli.pauli_mult_with_phase are now deprecated. Instead, you should use qml.simplify(qml.prod(pauli_1, pauli_2)) to get the reduced operator. (#5057)

• The private functions _pauli_mult, _binary_matrix and _get_pauli_map from the pauli module have been deprecated, as they are no longer used anywhere and the same functionality can be achieved using newer features in the pauli module. (#5057)

• Sum.ops, Sum.coeffs, Prod.ops and Prod.coeffs will be deprecated in the future. (#5164)

Documentation 📝

• The module documentation for pennylane.tape now explains the difference between QuantumTape and QuantumScript. (#5065)

• A typo in a code example in the qml.transforms API has been fixed. (#5014)

• Documentation for qml.data has been updated and now mentions a way to access the same dataset simultaneously from multiple environments. (#5029)

• A clarification for the definition of argnum added to gradient methods has been made. (#5035)

• A typo in the code example for qml.qchem.dipole_of has been fixed. (#5036)

• A development guide on deprecations and removals has been added. (#5083)

• A note about the eigenspectrum of second-quantized Hamiltonians has been added to qml.eigvals. (#5095)

• A warning about two mathematically equivalent Hamiltonians undergoing different time evolutions has been added to qml.TrotterProduct and qml.ApproxTimeEvolution. (#5137)

• A reference to the paper that provides the image of the qml.QAOAEmbedding template has been added. (#5130)

• The docstring of qml.sample has been updated to advise the use of single-shot expectations instead when differentiating a circuit. (#5237)

• A quick start page has been added called "Importing Circuits". This explains how to import quantum circuits and operations defined outside of PennyLane. (#5281)

Bug fixes 🐛

• QubitChannel can now be used with jitting. (#5288)

• Fixed a bug in the matplotlib drawer where the colour of Barrier did not match the requested style. (#5276)

• qml.draw and qml.draw_mpl now apply all applied transforms before drawing. (#5277)

• ctrl_decomp_zyz is now differentiable. (#5198)

• qml.ops.Pow.matrix() is now differentiable with TensorFlow with integer exponents. (#5178)

• The qml.MottonenStatePreparation template has been updated to include a global phase operation. (#5166)

• Fixed a queuing bug when using qml.prod with a quantum function that queues a single operator. (#5170)

• The qml.TrotterProduct template has been updated to accept scalar products of operators as an input Hamiltonian. (#5073)

• Fixed a bug where caching together with JIT compilation and broadcasted tapes yielded wrong results Operator.hash now depends on the memory location, id, of a JAX tracer instead of its string representation. (#3917)

• qml.transforms.undo_swaps can now work with operators with hyperparameters or nesting. (#5081)

• qml.transforms.split_non_commuting will now pass the original shots along. (#5081)

• If argnum is provided to a gradient transform, only the parameters specified in argnum will have their gradient methods validated. (#5035)

• StatePrep operations expanded onto more wires are now compatible with backprop. (#5028)

• qml.equal works well with qml.Sum operators when wire labels are a mix of integers and strings. (#5037)

• The return value of Controlled.generator now contains a projector that projects onto the correct subspace based on the control value specified. (#5068)

• CosineWindow no longer raises an unexpected error when used on a subset of wires at the beginning of a circuit. (#5080)

• tf.function now works with TensorSpec(shape=None) by skipping batch size computation. (#5089)

• PauliSentence.wires no longer imposes a false order. (#5041)

• qml.qchem.import_state now applies the chemist-to-physicist sign convention when initializing a PennyLane state vector from classically pre-computed wavefunctions. That is, it interleaves spin-up/spin-down operators for the same spatial orbital index, as standard in PennyLane (instead of commuting all spin-up operators to the left, as is standard in quantum chemistry). (#5114)

• Multi-wire controlled CNOT and PhaseShift are now be decomposed correctly. (#5125) (#5148)

• draw_mpl no longer raises an error when drawing a circuit containing an adjoint of a controlled operation. (#5149)

• default.mixed no longer throws ValueError when applying a state vector that is not of type complex128 when used with tensorflow. (#5155)

• ctrl_decomp_zyz no longer raises a TypeError if the rotation parameters are of type torch.Tensor (#5183)

• Comparing Prod and Sum objects now works regardless of nested structure with qml.equal if the operators have a valid pauli_rep property. (#5177)

• Controlled GlobalPhase with non-zero control wires no longer throws an error. (#5194)

• A QNode transformed with mitigate_with_zne now accepts batch parameters. (#5195)

• The matrix of an empty PauliSentence instance is now correct (all-zeros). Further, matrices of empty PauliWord and PauliSentence instances can now be turned into matrices. (#5188)

• PauliSentence instances can handle matrix multiplication with PauliWord instances. (#5208)

• CompositeOp.eigendecomposition is now JIT-compatible. (#5207)

• QubitDensityMatrix now works with JAX-JIT on the default.mixed device. (#5203) (#5236)

• When a QNode specifies diff_method="adjoint", default.qubit no longer tries to decompose non-trainable operations with non-scalar parameters such as QubitUnitary. (#5233)

• The overwriting of the class names of I, X, Y, and Z no longer happens in the initialization after causing problems with datasets. This now happens globally. (#5252)

• The adjoint_metric_tensor transform now works with jax. (#5271)

Contributors ✍️

This release contains contributions from (in alphabetical order):

Abhishek Abhishek, Mikhail Andrenkov, Utkarsh Azad, Trenten Babcock, Gabriel Bottrill, Thomas Bromley, Astral Cai, Skylar Chan, Isaac De Vlugt, Diksha Dhawan, Lillian Frederiksen, Pietropaolo Frisoni, Eugenio Gigante, Diego Guala, David Ittah, Soran Jahangiri, Jacky Jiang, Korbinian Kottmann, Christina Lee, Xiaoran Li, Vincent Michaud-Rioux, Romain Moyard, Pablo Antonio Moreno Casares, Erick Ochoa Lopez, Lee J. O'Riordan, Mudit Pandey, Alex Preciado, Matthew Silverman, Jay Soni.

This postfix release pins additional requirements in doc/requirements.txt for building the website documentation. This allows the website to be rebuilt to show the "Getting involved" section.

New features since last release

Statistics and drawing for mid-circuit measurements 🎨

• It is now possible to return statistics of composite mid-circuit measurements. (#4888)

  Mid-circuit measurement results can be composed using basic arithmetic operations and then statistics can be calculated by putting the result within a PennyLane measurement like qml.expval(). For example:

  import pennylane as qml
  
  dev = qml.device("default.qubit")
  
  @qml.qnode(dev) 
  def circuit(phi, theta): 
      qml.RX(phi, wires=0) 
      m0 = qml.measure(wires=0) 
      qml.RY(theta, wires=1) 
      m1 = qml.measure(wires=1) 
      return qml.expval(~m0 + m1)
  
  print(circuit(1.23, 4.56)) 
  

  1.2430187928114291 
  

  Another option, for ease-of-use when using qml.sample(), qml.probs(), or qml.counts(), is to provide a simple list of mid-circuit measurement results:

  dev = qml.device("default.qubit")
  
  @qml.qnode(dev) 
  def circuit(phi, theta): 
      qml.RX(phi, wires=0) 
      m0 = qml.measure(wires=0) 
      qml.RY(theta, wires=1) 
      m1 = qml.measure(wires=1) 
      return qml.sample(op=[m0, m1])
  
  print(circuit(1.23, 4.56, shots=5)) 
  

  [[0 1] 
   [0 1] 
   [0 0] 
   [1 0] 
   [0 1]] 
  

  Composite mid-circuit measurement statistics are supported on default.qubit and default.mixed. To learn more about which measurements and arithmetic operators are supported, refer to the measurements page and the documentation for qml.measure.

• Mid-circuit measurements can now be visualized with the text-based qml.draw() and the graphical qml.draw_mpl() methods. (#4775) (#4803) (#4832) (#4901) (#4850) (#4917) (#4930) (#4957)

  Drawing of mid-circuit measurement capabilities including qubit reuse and reset, postselection, conditioning, and collecting statistics is now supported. Here is an all-encompassing example:

  def circuit(): 
      m0 = qml.measure(0, reset=True) 
      m1 = qml.measure(1, postselect=1) 
      qml.cond(m0 - m1 == 0, qml.S)(0) 
      m2 = qml.measure(1) 
      qml.cond(m0 + m1 == 2, qml.T)(0) 
      qml.cond(m2, qml.PauliX)(1) 
  

  The text-based drawer outputs:

  >>> print(qml.draw(circuit)()) 
  0: ──┤↗│  │0⟩────────S───────T────┤ 
  1: ───║────────┤↗₁├──║──┤↗├──║──X─┤ 
        ╚═════════║════╬═══║═══╣  ║ 
                  ╚════╩═══║═══╝  ║ 
                           ╚══════╝ 
  

  The graphical drawer outputs:

  >>> print(qml.draw_mpl(circuit)()) 
  

Catalyst is seamlessly integrated with PennyLane ⚗️

• Catalyst, our next-generation compilation framework, is now accessible within PennyLane, allowing you to more easily benefit from hybrid just-in-time (JIT) compilation.

  To access these features, simply install pennylane-catalyst:

  pip install pennylane-catalyst 
  

  The qml.compiler module provides support for hybrid quantum-classical compilation. (#4692) (#4979)

  Through the use of the qml.qjit decorator, entire workflows can be JIT compiled — including both quantum and classical processing — down to a machine binary on first-function execution. Subsequent calls to the compiled function will execute the previously-compiled binary, resulting in significant performance improvements.

  import pennylane as qml
  
  dev = qml.device("lightning.qubit", wires=2)
  
  @qml.qjit 
  @qml.qnode(dev) 
  def circuit(theta): 
      qml.Hadamard(wires=0) 
      qml.RX(theta, wires=1) 
      qml.CNOT(wires=[0,1]) 
      return qml.expval(qml.PauliZ(wires=1)) 
  

  >>> circuit(0.5) # the first call, compilation occurs here array(0.) 
  >>> circuit(0.5) # the precompiled quantum function is called 
  array(0.) 
  

  Currently, PennyLane supports the Catalyst hybrid compiler with the qml.qjit decorator. A significant benefit of Catalyst is the ability to preserve complex control flow around quantum operations — such as if statements and for loops, and including measurement feedback — during compilation, while continuing to support end-to-end autodifferentiation.

• The following functions can now be used with the qml.qjit decorator: qml.grad, qml.jacobian, qml.vjp, qml.jvp, and qml.adjoint. (#4709) (#4724) (#4725) (#4726)

  When qml.grad or qml.jacobian are used with @qml.qjit, they are patched to catalyst.grad and catalyst.jacobian, respectively.

  dev = qml.device("lightning.qubit", wires=1)
  
  @qml.qjit 
  def workflow(x):
  
      @qml.qnode(dev) 
      def circuit(x): 
          qml.RX(np.pi * x[0], wires=0) 
          qml.RY(x[1], wires=0) 
          return qml.probs()
  
      g = qml.jacobian(circuit)
  
      return g(x) 
  

  >>> workflow(np.array([2.0, 1.0])) 
  array([[ 3.48786850e-16, -4.20735492e-01], 
         [-8.71967125e-17, 4.20735492e-01]]) 
  

• JIT-compatible functionality for control flow has been added via qml.for_loop, qml.while_loop, and qml.cond. (#4698)

  qml.for_loop and qml.while_loop can be deployed as decorators on functions that are the body of the loop. The arguments to both follow typical conventions:

  @qml.for_loop(lower_bound, upper_bound, step) 
  

  @qml.while_loop(cond_function) 
  

  Here is a concrete example with qml.for_loop:

  dev = qml.device("lightning.qubit", wires=1)
  
  @qml.qjit 
  @qml.qnode(dev) 
  def circuit(n: int, x: float):
  
      @qml.for_loop(0, n, 1) 
      def loop_rx(i, x): 
          # perform some work and update (some of) the arguments 
          qml.RX(x, wires=0)
  
          # update the value of x for the next iteration 
          return jnp.sin(x)
  
      # apply the for loop 
      final_x = loop_rx(x)
  
      return qml.expval(qml.PauliZ(0)), final_x 
  

  >>> circuit(7, 1.6) 
  (array(0.97926626), array(0.55395718)) 
  

Decompose circuits into the Clifford+T gateset 🧩

• The new qml.clifford_t_decomposition() transform provides an approximate breakdown of an input circuit into the Clifford+T gateset. Behind the scenes, this decomposition is enacted via the sk_decomposition() function using the Solovay-Kitaev algorithm. (#4801) (#4802)

  The Solovay-Kitaev algorithm approximately decomposes a quantum circuit into the Clifford+T gateset. To account for this, a desired total circuit decomposition error, epsilon, must be specified when using qml.clifford_t_decomposition:

  dev = qml.device("default.qubit")
  
  @qml.qnode(dev) 
  def circuit(): 
      qml.RX(1.1, 0) 
      return qml.state()
  
  circuit = qml.clifford_t_decomposition(circuit, epsilon=0.1) 
  

  >>> print(qml.draw(circuit)()) 
  0:   ──T†──H──T†──H──T──H──T──H──T──H──T──H──T†──H──T†──T†──H──T†──H──T──H──T──H──T──H──T──H──T†──H
  
   ───T†──H──T──H──GlobalPhase(0.39)─┤ 
  

  The resource requirements of this circuit can also be evaluated:

  >>> with qml.Tracker(dev) as tracker: 
  ...     circuit() 
  >>> resources_lst = tracker.history["resources"] 
  >>> resources_lst[0] 
  wires: 1 
  gates: 34 
  depth: 34 
  shots: Shots(total=None) 
  gate_types: {'Adjoint(T)': 8, 'Hadamard': 16, 'T': 9, 'GlobalPhase': 1} 
  gate_sizes:   {1: 33, 0: 1} 
  

Use an iterative approach for quantum phase estimation 🔄

• Iterative Quantum Phase Estimation is now available with qml.iterative_qpe. (#4804)

  The subroutine can be used similarly to mid-circuit measurements:

  import pennylane as qml
  
  dev = qml.device("default.qubit", shots=5)
  
  @qml.qnode(dev) 
  def circuit():
  
        # Initial state 
        qml.PauliX(wires=[0])
  
        # Iterative QPE 
        measurements = qml.iterative_qpe(qml.RZ(2., wires=[0]), ancilla=[1], iters=3)
  
        return [qml.sample(op=meas) for meas in measurements] 
  

  >>> print(circuit()) 
  [array([0, 0, 0, 0, 0]), array([1, 0, 0, 0, 0]), array([0, 1, 1, 1, 1])] 
  

  The $i$-th element in the list refers to the 5 samples generated by the $i$-th measurement of the algorithm.

Improvements 🛠

Community contributions 🥳

• The += operand can now be used with a PauliSentence, which has also provides a performance boost. (#4662)

• The Approximate Quantum Fourier Transform (AQFT) is now available with qml.AQFT. (#4715)

• qml.draw and qml.draw_mpl now render operator IDs. (#4749)

  The ID can be specified as a keyword argument when instantiating an operator:

  >>> def circuit(): 
  ...    qml.RX(0.123, id="data", wires=0) 
  >>> print(qml.draw(circuit)()) 
  0: ──RX(0.12,"data")─┤ 
  

• Non-parametric operators such as Barrier, Snapshot, and Wirecut have been grouped together and moved to pennylane/ops/meta.py. Additionally, the relevant tests have been organized and placed in a new file, tests/ops/test_meta.py. (#4789)

• The TRX, TRY, and TRZ operators are now differentiable via backpropagation on default.qutrit. (#4790)

• The function qml.equal now supports ControlledSequence operators. (#4829)

• XZX decomposition has been added to the list of supported single-qubit unitary decompositions. (#4862)

• == and != operands can now be used with TransformProgram and TransformContainers instances. (#4858)

• A qutrit_mixed module has been added to qml.devices to store helper functions for a future qutrit mixed-state device. A function called create_initial_state has been added to this module that creates device-compatible initial states. (#4861)

• The function qml.Snapshot now supports arbitrary state-based measurements (i.e., measurements of type StateMeasurement). (#4876)

• qml.equal now supports the comparison of QuantumScript and BasisRotation objects. (#4902) (#4919)

• The function qml.draw_mpl now accept a keyword argument fig to specify the output figure window. (#4956)

Better support for batching

• qml.AmplitudeEmbedding now supports batching when used with Tensorflow. (#4818)

• default.qubit can now evolve already batched states with qml.pulse.ParametrizedEvolution. (#4863)

• qml.ArbitraryUnitary now supports batching. (#4745)

• Operator and tape batch sizes are evaluated lazily, helping run expensive computations less frequently and an issue with Tensorflow pre-computing batch sizes. (#4911)

Performance improvements and benchmarking

• Autograd, PyTorch, and JAX can now use vector-Jacobian products (VJPs) provided by the device from the new device API. If a device provides a VJP, this can be selected by providing device_vjp=True to a QNode or qml.execute. (#4935) (#4557) (#4654) (#4878) (#4841)

  >>> dev = qml.device('default.qubit') 
  >>> @qml.qnode(dev, diff_method="adjoint", device_vjp=True) 
  >>> def circuit(x): 
  ...    qml.RX(x, wires=0) 
  ...    return qml.expval(qml.PauliZ(0)) 
  >>> with dev.tracker: 
  ...    g = qml.grad(circuit)(qml.numpy.array(0.1)) 
  >>> dev.tracker.totals 
  {'batches': 1, 'simulations': 1, 'executions': 1, 'vjp_batches': 1, 'vjps': 1} 
  >>> g 
  -0.09983341664682815 
  

• qml.expval with large Hamiltonian objects is now faster and has a significantly lower memory footprint (and constant with respect to the number of Hamiltonian terms) when the Hamiltonian is a PauliSentence. This is due to the introduction of a specialized dot method in the PauliSentence class which performs PauliSentence-state products. (#4839)

• default.qubit no longer uses a dense matrix for MultiControlledX for more than 8 operation wires. (#4673)

• Some relevant Pytests have been updated to enable its use as a suite of benchmarks. (#4703)

• default.qubit now applies GroverOperator faster by not using its matrix representation but a custom rule for apply_operation. Also, the matrix representation of GroverOperator now runs faster. (#4666)

• A new pipeline to run benchmarks and plot graphs comparing with a fixed reference has been added. This pipeline will run on a schedule and can be activated on a PR with the label ci:run_benchmarks. (#4741)

• default.qubit now supports adjoint differentiation for arbitrary diagonal state-based measurements. (#4865)

• The benchmarks pipeline has been expanded to export all benchmark data to a single JSON file and a CSV file with runtimes. This includes all references and local benchmarks. (#4873)

Final phase of updates to transforms

• qml.quantum_monte_carlo and qml.simplify now use the new transform system. (#4708) (#4949)

• The formal requirement that type hinting be provided when using the qml.transform decorator has been removed. Type hinting can still be used, but is now optional. Please use a type checker such as mypy if you wish to ensure types are being passed correctly. (#4942)

Other improvements

• PennyLane now supports Python 3.12. (#4985)

• SampleMeasurement now has an optional method process_counts for computing the measurement results from a counts dictionary. (#4941)

• A new function called ops.functions.assert_valid has been added for checking if an Operator class is defined correctly. (#4764)

• Shots objects can now be multiplied by scalar values. (#4913)

• GlobalPhase now decomposes to nothing in case devices do not support global phases. (#4855)

• Custom operations can now provide their matrix directly through the Operator.matrix() method without needing to update the has_matrix property. has_matrix will now automatically be True if Operator.matrix is overridden, even if Operator.compute_matrix is not. (#4844)

• The logic for re-arranging states before returning them has been improved. (#4817)

• When multiplying SparseHamiltonians by a scalar value, the result now stays as a SparseHamiltonian. (#4828)

• trainable_params can now be set upon initialization of a QuantumScript instead of having to set the parameter after initialization. (#4877)

• default.qubit now calculates the expectation value of Hermitian operators in a differentiable manner. (#4866)

• The rot decomposition now has support for returning a global phase. (#4869)

• The "pennylane_sketch" MPL-drawer style has been added. This is the same as the "pennylane" style, but with sketch-style lines. (#4880)

• Operators now define a pauli_rep property, an instance of PauliSentence, defaulting to None if the operator has not defined it (or has no definition in the Pauli basis). (#4915)

• qml.ShotAdaptiveOptimizer can now use a multinomial distribution for spreading shots across the terms of a Hamiltonian measured in a QNode. Note that this is equivalent to what can be done with qml.ExpvalCost, but this is the preferred method because ExpvalCost is deprecated. (#4896)

• Decomposition of qml.PhaseShift now uses qml.GlobalPhase for retaining the global phase information. (#4657) (#4947)

• qml.equal for Controlled operators no longer returns False when equivalent but differently-ordered sets of control wires and control values are compared. (#4944)

• All PennyLane Operator subclasses are automatically tested by ops.functions.assert_valid to ensure that they follow PennyLane Operator standards. (#4922)

• Probability measurements can now be calculated from a counts dictionary with the addition of a process_counts method in the ProbabilityMP class. (#4952)

• ClassicalShadow.entropy now uses the algorithm outlined in 1106.5458 to project the approximate density matrix (with potentially negative eigenvalues) onto the closest valid density matrix. (#4959)

• The ControlledSequence.compute_decomposition default now decomposes the Pow operators, improving compatibility with machine learning interfaces. (#4995)

Breaking changes 💔

• The functions qml.transforms.one_qubit_decomposition, qml.transforms.two_qubit_decomposition, and qml.transforms.sk_decomposition were moved to qml.ops.one_qubit_decomposition, qml.ops.two_qubit_decomposition, and qml.ops.sk_decomposition, respectively. (#4906)

• The function qml.transforms.classical_jacobian has been moved to the gradients module and is now accessible as qml.gradients.classical_jacobian. (#4900)

• The transforms submodule qml.transforms.qcut is now its own module: qml.qcut. (#4819)

• The decomposition of GroverOperator now has an additional global phase operation. (#4666)

• qml.cond and the Conditional operation have been moved from the transforms folder to the ops/op_math folder. qml.transforms.Conditional will now be available as qml.ops.Conditional. (#4860)

• The prep keyword argument has been removed from QuantumScript and QuantumTape. StatePrepBase operations should be placed at the beginning of the ops list instead. (#4756)

• qml.gradients.pulse_generator is now named qml.gradients.pulse_odegen to adhere to paper naming conventions. (#4769)

• Specifying control_values passed to qml.ctrl as a string is no longer supported. (#4816)

• The rot decomposition will now normalize its rotation angles to the range [0, 4pi] for consistency (#4869)

• QuantumScript.graph is now built using tape.measurements instead of tape.observables because it depended on the now-deprecated Observable.return_type property. (#4762)

• The "pennylane" MPL-drawer style now draws straight lines instead of sketch-style lines. (#4880)

• The default value for the term_sampling argument of ShotAdaptiveOptimizer is now None instead of "weighted_random_sampling". (#4896)

Deprecations 👋

• single_tape_transform, batch_transform, qfunc_transform, and op_transform are deprecated. Use the new qml.transform function instead. (#4774)

• Observable.return_type is deprecated. Instead, you should inspect the type of the surrounding measurement process. (#4762) (#4798)

• All deprecations now raise a qml.PennyLaneDeprecationWarning instead of a UserWarning. (#4814)

• QuantumScript.is_sampled and QuantumScript.all_sampled are deprecated. Users should now validate these properties manually. (#4773)

• With an algorithmic improvement to ClassicalShadow.entropy, the keyword atol becomes obsolete and will be removed in v0.35. (#4959)

Documentation 📝

• Documentation for unitaries and operations' decompositions has been moved from qml.transforms to qml.ops.ops_math. (#4906)

• Documentation for qml.metric_tensor and qml.adjoint_metric_tensor and qml.transforms.classical_jacobian is now accessible via the gradients API page qml.gradients in the documentation. (#4900)

• Documentation for qml.specs has been moved to the resource module. (#4904)

• Documentation for QCut has been moved to its own API page: qml.qcut. (#4819)

• The documentation page for qml.measurements now links top-level accessible functions (e.g., qml.expval) to their top-level pages rather than their module-level pages (e.g., qml.measurements.expval). (#4750)

• Information for the documentation of qml.matrix about wire ordering has been added for using qml.matrix on a QNode which uses a device with device.wires=None. (#4874)

Bug fixes 🐛

• TransformDispatcher now stops queuing when performing the transform when applying it to a qfunc. Only the output of the transform will be queued. (#4983)

• qml.map_wires now works properly with qml.cond and qml.measure. (#4884)

• Pow operators are now picklable. (#4966)

• Finite differences and SPSA can now be used with tensorflow-autograph on setups that were seeing a bus error. (#4961)

• qml.cond no longer incorrectly queues operators used arguments. (#4948)

• Attribute objects now return False instead of raising a TypeError when checking if an object is inside the set. (#4933)

• Fixed a bug where the parameter-shift rule of qml.ctrl(op) was wrong if op had a generator that has two or more eigenvalues and is stored as a SparseHamiltonian. (#4899)

• Fixed a bug where trainable parameters in the post-processing of finite-differences were incorrect for JAX when applying the transform directly on a QNode. (#4879)

• qml.grad and qml.jacobian now explicitly raise errors if trainable parameters are integers. (#4836)

• JAX-JIT now works with shot vectors. (#4772)

• JAX can now differentiate a batch of circuits where one tape does not have trainable parameters. (#4837)

• The decomposition of GroverOperator now has the same global phase as its matrix. (#4666)

• The tape.to_openqasm method no longer mistakenly includes interface information in the parameter string when converting tapes using non-NumPy interfaces. (#4849)

• qml.defer_measurements now correctly transforms circuits when terminal measurements include wires used in mid-circuit measurements. (#4787)

• Fixed a bug where the adjoint differentiation method would fail if an operation that has a parameter with grad_method=None is present. (#4820)

• MottonenStatePreparation and BasisStatePreparation now raise an error when decomposing a broadcasted state vector. (#4767)

• Gradient transforms now work with overridden shot vectors and default.qubit. (#4795)

• Any ScalarSymbolicOp, like Evolution, now states that it has a matrix if the target is a Hamiltonian. (#4768)

• In default.qubit, initial states are now initialized with the simulator's wire order, not the circuit's wire order. (#4781)

• qml.compile will now always decompose to expand_depth, even if a target basis set is not specified. (#4800)

• qml.transforms.transpile can now handle measurements that are broadcasted onto all wires. (#4793)

• Parametrized circuits whose operators do not act on all wires return PennyLane tensors instead of NumPy arrays, as expected. (#4811) (#4817)

• qml.transforms.merge_amplitude_embedding no longer depends on queuing, allowing it to work as expected with QNodes. (#4831)

• qml.pow(op) and qml.QubitUnitary.pow() now also work with Tensorflow data raised to an integer power. (#4827)

• The text drawer has been fixed to correctly label qml.qinfo measurements, as well as qml.classical_shadow qml.shadow_expval. (#4803)

• Removed an implicit assumption that an empty PauliSentence gets treated as identity under multiplication. (#4887)

• Using a CNOT or PauliZ operation with large batched states and the Tensorflow interface no longer raises an unexpected error. (#4889)

• qml.map_wires no longer fails when mapping nested quantum tapes. (#4901)

• Conversion of circuits to openqasm now decomposes to a depth of 10, allowing support for operators requiring more than 2 iterations of decomposition, such as the ApproxTimeEvolution gate. (#4951)

• MPLDrawer does not add the bonus space for classical wires when no classical wires are present. (#4987)

• Projector now works with parameter-broadcasting. (#4993) * The jax-jit interface can now be used with float32 mode. (#4990)

• Keras models with a qnn.KerasLayer no longer fail to save and load weights properly when they are named "weights". (#5008)

Contributors ✍️

This release contains contributions from (in alphabetical order):

Guillermo Alonso, Ali Asadi, Utkarsh Azad, Gabriel Bottrill, Thomas Bromley, Astral Cai, Minh Chau, Isaac De Vlugt, Amintor Dusko, Pieter Eendebak, Lillian Frederiksen, Pietropaolo Frisoni, Josh Izaac, Juan Giraldo, Emiliano Godinez Ramirez, Ankit Khandelwal, Korbinian Kottmann, Christina Lee, Vincent Michaud-Rioux, Anurav Modak, Romain Moyard, Mudit Pandey, Matthew Silverman, Jay Soni, David Wierichs, Justin Woodring.

Bug fixes 🐛

• Fix gradient performance regression due to expansion of VJP products. (#4806)

• qml.defer_measurements now correctly transforms circuits when terminal measurements include wires used in mid-circuit measurements. (#4787)

• Any ScalarSymbolicOp, like Evolution, now states that it has a matrix if the target is a Hamiltonian. (#4768)

• In default.qubit, initial states are now initialized with the simulator's wire order, not the circuit's wire order. (#4781)

Contributors ✍️

This release contains contributions from (in alphabetical order):

Christina Lee, Lee James O'Riordan, Mudit Pandey

New features since last release

Postselection and statistics in mid-circuit measurements 📌

• It is now possible to request postselection on a mid-circuit measurement. (#4604)

  This can be achieved by specifying the postselect keyword argument in qml.measure as either 0 or 1, corresponding to the basis states.

  import pennylane as qml
  
  dev = qml.device("default.qubit")
  
  @qml.qnode(dev, interface=None)
  def circuit():
      qml.Hadamard(wires=0)
      qml.CNOT(wires=[0, 1])
      qml.measure(0, postselect=1)
      return qml.expval(qml.PauliZ(1)), qml.sample(wires=1)
  

  This circuit prepares the $| \Phi^{+} \rangle$ Bell state and postselects on measuring $|1\rangle$ in wire 0. The output of wire 1 is then also $|1\rangle$ at all times:

  >>> circuit(shots=10)
  (-1.0, array([1, 1, 1, 1, 1, 1]))
  

  Note that the number of shots is less than the requested amount because we have thrown away the samples where $|0\rangle$ was measured in wire 0.

• Measurement statistics can now be collected for mid-circuit measurements. (#4544)

  dev = qml.device("default.qubit")
  
  @qml.qnode(dev)
  def circ(x, y):
      qml.RX(x, wires=0)
      qml.RY(y, wires=1)
      m0 = qml.measure(1)
      return qml.expval(qml.PauliZ(0)), qml.expval(m0), qml.sample(m0)
  

  >>> circ(1.0, 2.0, shots=10000)
  (0.5606, 0.7089, array([0, 1, 1, ..., 1, 1, 1]))
  

  Support is provided for both finite-shot and analytic modes and devices default to using the deferred measurement principle to enact the mid-circuit measurements.

Exponentiate Hamiltonians with flexible Trotter products 🐖

• Higher-order Trotter-Suzuki methods are now easily accessible through a new operation called TrotterProduct. (#4661)

  Trotterization techniques are an affective route towards accurate and efficient Hamiltonian simulation. The Suzuki-Trotter product formula allows for the ability to express higher-order approximations to the matrix exponential of a Hamiltonian, and it is now available to use in PennyLane via the TrotterProduct operation. Simply specify the order of the approximation and the evolution time.

  coeffs = [0.25, 0.75]
  ops = [qml.PauliX(0), qml.PauliZ(0)]
  H = qml.dot(coeffs, ops)
  
  dev = qml.device("default.qubit", wires=2)
  
  @qml.qnode(dev)
  def circuit():
      qml.Hadamard(0)
      qml.TrotterProduct(H, time=2.4, order=2)
      return qml.state()
  

  >>> circuit()
  [-0.13259524+0.59790098j 0. +0.j -0.13259524-0.77932754j 0. +0.j ]
  

• Approximating matrix exponentiation with random product formulas, qDrift, is now available with the new QDrift operation. (#4671)

  As shown in 1811.08017, qDrift is a Markovian process that can provide a speedup in Hamiltonian simulation. At a high level, qDrift works by randomly sampling from the Hamiltonian terms with a probability that depends on the Hamiltonian coefficients. This method for Hamiltonian simulation is now ready to use in PennyLane with the QDrift operator. Simply specify the evolution time and the number of samples drawn from the Hamiltonian, n:

  coeffs = [0.25, 0.75]
  ops = [qml.PauliX(0), qml.PauliZ(0)]
  H = qml.dot(coeffs, ops)
  
  dev = qml.device("default.qubit", wires=2)
  
  @qml.qnode(dev)
  def circuit():
      qml.Hadamard(0)
      qml.QDrift(H, time=1.2, n = 10)
      return qml.probs()
  

  >>> circuit()
  array([0.61814334, 0. , 0.38185666, 0. ])
  

Building blocks for quantum phase estimation 🧱

• A new operator called CosineWindow has been added to prepare an initial state based on a cosine wave function. (#4683)

  As outlined in 2110.09590, the cosine tapering window is part of a modification to quantum phase estimation that can provide a cubic improvement to the algorithm's error rate. Using CosineWindow will prepare a state whose amplitudes follow a cosinusoidal distribution over the computational basis.

  import matplotlib.pyplot as plt
  
  dev = qml.device('default.qubit', wires=4)
  
  @qml.qnode(dev)
  def example_circuit():
      qml.CosineWindow(wires=range(4))
      return qml.state()
  
  output = example_circuit()
  
  plt.style.use("pennylane.drawer.plot")
  plt.bar(range(len(output)), output)
  plt.show()
  

  

• Controlled gate sequences raised to decreasing powers, a sub-block in quantum phase estimation, can now be created with the new ControlledSequence operator. (#4707)

  To use ControlledSequence, specify the controlled unitary operator and the control wires, control:

  dev = qml.device("default.qubit", wires = 4)
  
  @qml.qnode(dev)
  def circuit():
      for i in range(3):
          qml.Hadamard(wires = i)
      qml.ControlledSequence(qml.RX(0.25, wires = 3), control = [0, 1, 2])
      qml.adjoint(qml.QFT)(wires = range(3))
          return qml.probs(wires = range(3))
  

  >>> print(circuit())
  [0.92059345 0.02637178 0.00729619 0.00423258 0.00360545 0.00423258 0.00729619 0.02637178]
  

New device capabilities, integration with Catalyst, and more! ⚗️

• default.qubit now uses the new qml.devices.Device API and functionality in qml.devices.qubit. If you experience any issues with the updated default.qubit, please let us know by posting an issue. The old version of the device is still accessible by the short name default.qubit.legacy, or directly via qml.devices.DefaultQubitLegacy. (#4594) (#4436) (#4620) (#4632)

  This changeover has a number of benefits for default.qubit, including:

  • The number of wires is now optional — simply having qml.device("default.qubit") is valid! If wires are not provided at instantiation, the device automatically infers the required number of wires for each circuit provided for execution.

    dev = qml.device("default.qubit")
    
    @qml.qnode(dev)
    def circuit():
        qml.PauliZ(0)
        qml.RZ(0.1, wires=1)
        qml.Hadamard(2)
        return qml.state()
    

    >>> print(qml.draw(circuit)())
    0: ──Z────────┤ State
    1: ──RZ(0.10)─┤ State
    2: ──H────────┤ State
    

  • default.qubit is no longer silently swapped out with an interface-appropriate device when the backpropagation differentiation method is used. For example, consider:

    import jax
    
    dev = qml.device("default.qubit", wires=1)
    
    @qml.qnode(dev, diff_method="backprop")
    def f(x):
        qml.RX(x, wires=0)
        return qml.expval(qml.PauliZ(0))
    f(jax.numpy.array(0.2))
    

    In previous versions of PennyLane, the device will be swapped for the JAX equivalent:

    >>> f.device
    <DefaultQubitJax device (wires=1, shots=None) at 0x7f8c8bff50a0>
    >>> f.device == dev
    False
    

    Now, default.qubit can itself dispatch to all the interfaces in a backprop-compatible way and hence does not need to be swapped:

    >>> f.device
    <default.qubit device (wires=1) at 0x7f20d043b040>
    >>> f.device == dev
    True
    

• A QNode that has been decorated with qjit from PennyLane's Catalyst library for just-in-time hybrid compilation is now compatible with qml.draw. (#4609)

  import catalyst
  
  @catalyst.qjit
  @qml.qnode(qml.device("lightning.qubit", wires=3))
  def circuit(x, y, z, c):
      """A quantum circuit on three wires."""
  
      @catalyst.for_loop(0, c, 1)
      def loop(i):
          qml.Hadamard(wires=i)
  
      qml.RX(x, wires=0)
      loop()
      qml.RY(y, wires=1)
      qml.RZ(z, wires=2)
      return qml.expval(qml.PauliZ(0))
  
  draw = qml.draw(circuit, decimals=None)(1.234, 2.345, 3.456, 1)
  

  >>> print(draw)
  0: ──RX──H──┤ <Z>
  1: ──H───RY─┤
  2: ──RZ─────┤
  

Improvements 🛠

More PyTrees!

• MeasurementProcess and QuantumScript objects are now registered as JAX PyTrees. (#4607) (#4608)

  It is now possible to JIT-compile functions with arguments that are a MeasurementProcess or a QuantumScript:

  
  tape0 = qml.tape.QuantumTape([qml.RX(1.0, 0), qml.RY(0.5, 0)], [qml.expval(qml.PauliZ(0))])
  dev = qml.device('lightning.qubit', wires=5)
  
  execute_kwargs = {"device": dev, "gradient_fn": qml.gradients.param_shift, "interface":"jax"}
  
  jitted_execute = jax.jit(qml.execute, static_argnames=execute_kwargs.keys())
  jitted_execute((tape0, ), **execute_kwargs)
  

Improving QChem and existing algorithms

• Computationally expensive functions in integrals.py, electron_repulsion and _hermite_coulomb, have been modified to replace indexing with slicing for better compatibility with JAX. (#4685)

• qml.qchem.import_state has been extended to import more quantum chemistry wavefunctions, from MPS, DMRG and SHCI classical calculations performed with the Block2 and Dice libraries. #4523 #4524 #4626 #4634

  Check out our how-to guide to learn more about how PennyLane integrates with your favourite quantum chemistry libraries.

• The qchem fermionic_dipole and particle_number functions have been updated to use a FermiSentence. The deprecated features for using tuples to represent fermionic operations are removed. (#4546) (#4556)

• The tensor-network template qml.MPS now supports changing the offset between subsequent blocks for more flexibility. (#4531)

• Builtin types support with qml.pauli_decompose have been improved. (#4577)

• AmplitudeEmbedding now inherits from StatePrep, allowing for it to not be decomposed when at the beginning of a circuit, thus behaving like StatePrep. (#4583)

• qml.cut_circuit is now compatible with circuits that compute the expectation values of Hamiltonians with two or more terms. (#4642)

Next-generation device API

• default.qubit now tracks the number of equivalent qpu executions and total shots when the device is sampling. Note that "simulations" denotes the number of simulation passes, whereas "executions" denotes how many different computational bases need to be sampled in. Additionally, the new default.qubit tracks the results of device.execute. (#4628) (#4649)

• DefaultQubit can now accept a jax.random.PRNGKey as a seed to set the key for the JAX pseudo random number generator when using the JAX interface. This corresponds to the prng_key on default.qubit.jax in the old API. (#4596)

• The JacobianProductCalculator abstract base class and implementations TransformJacobianProducts DeviceDerivatives, and DeviceJacobianProducts have been added to pennylane.interfaces.jacobian_products. (#4435) (#4527) (#4637)

• DefaultQubit dispatches to a faster implementation for applying ParametrizedEvolution to a state when it is more efficient to evolve the state than the operation matrix. (#4598) (#4620)

• Wires can be provided to the new device API. (#4538) (#4562)

• qml.sample() in the new device API now returns a np.int64 array instead of np.bool8. (#4539)

• The new device API now has a repr() method. (#4562)

• DefaultQubit now works as expected with measurement processes that don't specify wires. (#4580)

• Various improvements to measurements have been made for feature parity between default.qubit.legacy and the new DefaultQubit. This includes not trying to squeeze batched CountsMP results and implementing MutualInfoMP.map_wires. (#4574)

• devices.qubit.simulate now accepts an interface keyword argument. If a QNode with DefaultQubit specifies an interface, the result will be computed with that interface. (#4582)

• ShotAdaptiveOptimizer has been updated to pass shots to QNode executions instead of overriding device shots before execution. This makes it compatible with the new device API. (#4599)

• pennylane.devices.preprocess now offers the transforms decompose, validate_observables, validate_measurements, validate_device_wires, validate_multiprocessing_workers, warn_about_trainable_observables, and no_sampling to assist in constructing devices under the new device API. (#4659)

• Updated qml.device, devices.preprocessing and the tape_expand.set_decomposition context manager to bring DefaultQubit to feature parity with default.qubit.legacy with regards to using custom decompositions. The DefaultQubit device can now be included in a set_decomposition context or initialized with a custom_decomps dictionary, as well as a custom max_depth for decomposition. (#4675)

Other improvements

• The StateMP measurement now accepts a wire order (e.g., a device wire order). The process_state method will re-order the given state to go from the inputted wire-order to the process's wire-order. If the process's wire-order contains extra wires, it will assume those are in the zero-state. (#4570) (#4602)

• Methods called add_transform and insert_front_transform have been added to TransformProgram. (#4559)

• Instances of the TransformProgram class can now be added together. (#4549)

• Transforms can now be applied to devices following the new device API. (#4667)

• All gradient transforms have been updated to the new transform program system. (#4595)

• Multi-controlled operations with a single-qubit special unitary target can now automatically decompose. (#4697)

• pennylane.defer_measurements will now exit early if the input does not contain mid circuit measurements. (#4659)

• The density matrix aspects of StateMP have been split into their own measurement process called DensityMatrixMP. (#4558)

• StateMeasurement.process_state now assumes that the input is flat. ProbabilityMP.process_state has been updated to reflect this assumption and avoid redundant reshaping. (#4602)

• qml.exp returns a more informative error message when decomposition is unavailable for non-unitary operators. (#4571)

• Added qml.math.get_deep_interface to get the interface of a scalar hidden deep in lists or tuples. (#4603)

• Updated qml.math.ndim and qml.math.shape to work with built-in lists or tuples that contain interface-specific scalar dat (e.g., [(tf.Variable(1.1), tf.Variable(2.2))]). (#4603)

• When decomposing a unitary matrix with one_qubit_decomposition and opting to include the GlobalPhase in the decomposition, the phase is no longer cast to dtype=complex. (#4653)

• _qfunc_output has been removed from QuantumScript, as it is no longer necessary. There is still a _qfunc_output property on QNode instances. (#4651)

• qml.data.load properly handles parameters that come after 'full' (#4663)

• The qml.jordan_wigner function has been modified to optionally remove the imaginary components of the computed qubit operator, if imaginary components are smaller than a threshold. (#4639)

• qml.data.load correctly performs a full download of the dataset after a partial download of the same dataset has already been performed. (#4681) * The performance of qml.data.load() has been improved when partially loading a dataset (#4674)

• Plots generated with the pennylane.drawer.plot style of matplotlib.pyplot now have black axis labels and are generated at a default DPI of 300. (#4690)

• Shallow copies of the QNode now also copy the execute_kwargs and transform program. When applying a transform to a QNode, the new qnode is only a shallow copy of the original and thus keeps the same device. (#4736)

• QubitDevice and CountsMP are updated to disregard samples containing failed hardware measurements (record as np.NaN) when tallying samples, rather than counting failed measurements as ground-state measurements, and to display qml.counts coming from these hardware devices correctly. (#4739)

Breaking changes 💔

• qml.defer_measurements now raises an error if a transformed circuit measures qml.probs, qml.sample, or qml.counts without any wires or observable, or if it measures qml.state. (#4701)

• The device test suite now converts device keyword arguments to integers or floats if possible. (#4640)

• MeasurementProcess.eigvals() now raises an EigvalsUndefinedError if the measurement observable does not have eigenvalues. (#4544)

• The __eq__ and __hash__ methods of Operator and MeasurementProcess no longer rely on the object's address in memory. Using == with operators and measurement processes will now behave the same as qml.equal, and objects of the same type with the same data and hyperparameters will have the same hash. (#4536)

  In the following scenario, the second and third code blocks show the previous and current behaviour of operator and measurement process equality, determined by ==:

  op1 = qml.PauliX(0)
  op2 = qml.PauliX(0)
  op3 = op1
  

  Old behaviour:

  >>> op1 == op2
  False
  >>> op1 == op3
  True
  

  New behaviour:

  >>> op1 == op2
  True
  >>> op1 == op3
  True
  

  The __hash__ dunder method defines the hash of an object. The default hash of an object is determined by the objects memory address. However, the new hash is determined by the properties and attributes of operators and measurement processes. Consider the scenario below. The second and third code blocks show the previous and current behaviour.

  op1 = qml.PauliX(0)
  op2 = qml.PauliX(0)
  

  Old behaviour:

  >>> print({op1, op2})
  {PauliX(wires=[0]), PauliX(wires=[0])}
  

  New behaviour:

  >>> print({op1, op2})
  {PauliX(wires=[0])}
  

• The old return type and associated functions qml.enable_return and qml.disable_return have been removed. (#4503)

• The mode keyword argument in QNode has been removed. Please use grad_on_execution instead. (#4503)

• The CV observables qml.X and qml.P have been removed. Please use qml.QuadX and qml.QuadP instead. (#4533)

• The sampler_seed argument of qml.gradients.spsa_grad has been removed. Instead, the sampler_rng argument should be set, either to an integer value, which will be used to create a PRNG internally, or to a NumPy pseudo-random number generator (PRNG) created via np.random.default_rng(seed). (#4550)

• The QuantumScript.set_parameters method and the QuantumScript.data setter have been removed. Please use QuantumScript.bind_new_parameters instead. (#4548)

• The method tape.unwrap() and corresponding UnwrapTape and Unwrap classes have been removed. Instead of tape.unwrap(), use qml.transforms.convert_to_numpy_parameters. (#4535)

• The RandomLayers.compute_decomposition keyword argument ratio_imprivitive has been changed to ratio_imprim to match the call signature of the operation. (#4552)

• The private TmpPauliRot operator used for SpecialUnitary no longer decomposes to nothing when the theta value is trainable. (#4585)

• ProbabilityMP.marginal_prob has been removed. Its contents have been moved into process_state, which effectively just called marginal_prob with np.abs(state) ** 2. (#4602)

Deprecations 👋

• The following decorator syntax for transforms has been deprecated and will raise a warning: (#4457)

  @transform_fn(**transform_kwargs)
  @qml.qnode(dev)
  def circuit():
      ...
  

  If you are using a transform that has supporting transform_kwargs, please call the transform directly using circuit = transform_fn(circuit, **transform_kwargs), or use functools.partial:

  @functools.partial(transform_fn, **transform_kwargs)
  @qml.qnode(dev)
  def circuit():
      ...
  

• The prep keyword argument in QuantumScript has been deprecated and will be removed from QuantumScript. StatePrepBase operations should be placed at the beginning of the ops list instead. (#4554)

• qml.gradients.pulse_generator has been renamed to qml.gradients.pulse_odegen to adhere to paper naming conventions. During v0.33, pulse_generator is still available but raises a warning. (#4633)

Documentation 📝

• A warning section in the docstring for DefaultQubit regarding the start method used in multiprocessing has been added. This may help users circumvent issues arising in Jupyter notebooks on macOS for example. (#4622)

• Documentation improvements to the new device API have been made. The documentation now correctly states that interface-specific parameters are only passed to the device for backpropagation derivatives. (#4542)

• Functions for qubit-simulation to the qml.devices sub-page of the "Internal" section have been added. Note that these functions are unstable while device upgrades are underway. (#4555)

• A documentation improvement to the usage example in the qml.QuantumMonteCarlo page has been made. An integral was missing the differential $dx$. (#4593)

• A documentation improvement for the use of the pennylane style of qml.drawer and the pennylane.drawer.plot style of matplotlib.pyplot has been made by clarifying the use of the default font. (#4690)

Bug fixes 🐛

• Jax jit now works when a probability measurement is broadcasted onto all wires. (#4742)

• Fixed LocalHilbertSchmidt.compute_decomposition so that the template can be used in a QNode. (#4719)

• Fixes transforms.transpile with arbitrary measurement processes. (#4732)

• Providing work_wires=None to qml.GroverOperator no longer interprets None as a wire. (#4668)

• Fixed an issue where the __copy__ method of the qml.Select() operator attempted to access un-initialized data. (#4551)

• Fixed the skip_first option in expand_tape_state_prep. (#4564)

• convert_to_numpy_parameters now uses qml.ops.functions.bind_new_parameters. This reinitializes the operation and makes sure everything references the new NumPy parameters. (#4540)

• tf.function no longer breaks ProbabilityMP.process_state, which is needed by new devices. (#4470)

• Fixed unit tests for qml.qchem.mol_data. (#4591)

• Fixed ProbabilityMP.process_state so that it allows for proper Autograph compilation. Without this, decorating a QNode that returns an expval with tf.function would fail when computing the expectation. (#4590)

• The torch.nn.Module properties are now accessible on a pennylane.qnn.TorchLayer. (#4611)

• qml.math.take with Pytorch now returns tensor[..., indices] when the user requests the last axis (axis=-1). Without the fix, it would wrongly return tensor[indices]. (#4605)

• Ensured the logging TRACE level works with gradient-free execution. (#4669)

Contributors ✍️

This release contains contributions from (in alphabetical order):

Guillermo Alonso, Utkarsh Azad, Thomas Bromley, Isaac De Vlugt, Jack Brown, Stepan Fomichev, Joana Fraxanet, Diego Guala, Soran Jahangiri, Edward Jiang, Korbinian Kottmann, Ivana Kurečić Christina Lee, Lillian M. A. Frederiksen, Vincent Michaud-Rioux, Romain Moyard, Daniel F. Nino, Lee James O'Riordan, Mudit Pandey, Matthew Silverman, Jay Soni.

This release changes doc/requirements.txt to upgrade jax, jaxlib, and pin ml-dtypes.

New features since last release

Encode matrices using a linear combination of unitaries ⛓️️

• It is now possible to encode an operator A into a quantum circuit by decomposing it into a linear combination of unitaries using PREP (qml.StatePrep) and SELECT (qml.Select) routines. (#4431) (#4437) (#4444) (#4450) (#4506) (#4526)

  Consider an operator A composed of a linear combination of Pauli terms:

  >>> A = qml.PauliX(2) + 2 * qml.PauliY(2) + 3 * qml.PauliZ(2)
  

  A decomposable block-encoding circuit can be created:

  def block_encode(A, control_wires):
      probs = A.coeffs / np.sum(A.coeffs)
      state = np.pad(np.sqrt(probs, dtype=complex), (0, 1))
      unitaries = A.ops
  
      qml.StatePrep(state, wires=control_wires)
      qml.Select(unitaries, control=control_wires)
      qml.adjoint(qml.StatePrep)(state, wires=control_wires)
  

  >>> print(qml.draw(block_encode, show_matrices=False)(A, control_wires=[0, 1]))
  0: ─╭|Ψ⟩─╭Select─╭|Ψ⟩†─┤
  1: ─╰|Ψ⟩─├Select─╰|Ψ⟩†─┤
  2: ──────╰Select───────┤
  

  This circuit can be used as a building block within a larger QNode to perform algorithms such as QSVT and Hamiltonian simulation.

• Decomposing a Hermitian matrix into a linear combination of Pauli words via qml.pauli_decompose is now faster and differentiable. (#4395) (#4479) (#4493)

  def find_coeffs(p):
      mat = np.array([[3, p], [p, 3]])
      A = qml.pauli_decompose(mat)
      return A.coeffs
  

  >>> import jax
  >>> from jax import numpy as np
  >>> jax.jacobian(find_coeffs)(np.array(2.))
  Array([0., 1.], dtype=float32, weak_type=True)
  

Monitor PennyLane's inner workings with logging 📃

• Python-native logging can now be enabled with qml.logging.enable_logging(). (#4377) (#4383)

  Consider the following code that is contained in my_code.py:

  import pennylane as qml
  qml.logging.enable_logging() # enables logging
  
  dev = qml.device("default.qubit", wires=2)
  
  @qml.qnode(dev)
  def f(x):
      qml.RX(x, wires=0)
      return qml.state()
  
  f(0.5)
  

  Executing my_code.py with logging enabled will detail every step in PennyLane's pipeline that gets used to run your code.

  $ python my_code.py
  [1967-02-13 15:18:38,591][DEBUG][<PID 8881:MainProcess>] - pennylane.qnode.__init__()::"Creating QNode(func=<function f at 0x7faf2a6fbaf0>, device=<DefaultQubit device (wires=2, shots=None) at 0x7faf2a689b50>, interface=auto, diff_method=best, expansion_strategy=gradient, max_expansion=10, grad_on_execution=best, mode=None, cache=True, cachesize=10000, max_diff=1, gradient_kwargs={}"
  ...
  

  Additional logging configuration settings can be specified by modifying the contents of the logging configuration file, which can be located by running qml.logging.config_path(). Follow our logging docs page for more details!

More input states for quantum chemistry calculations ⚛️

• Input states obtained from advanced quantum chemistry calculations can be used in a circuit. (#4427) (#4433) (#4461) (#4476) (#4505)

  Quantum chemistry calculations rely on an initial state that is typically selected to be the trivial Hartree-Fock state. For molecules with a complicated electronic structure, using initial states obtained from affordable post-Hartree-Fock calculations helps to improve the efficiency of the quantum simulations. These calculations can be done with external quantum chemistry libraries such as PySCF.

  It is now possible to import a PySCF solver object in PennyLane and extract the corresponding wave function in the form of a state vector that can be directly used in a circuit. First, perform your classical quantum chemistry calculations and then use the qml.qchem.import_state function to import the solver object and return a state vector.

  >>> from pyscf import gto, scf, ci
  >>> mol = gto.M(atom=[['H', (0, 0, 0)], ['H', (0,0,0.71)]], basis='sto6g')
  >>> myhf = scf.UHF(mol).run()
  >>> myci = ci.UCISD(myhf).run()
  >>> wf_cisd = qml.qchem.import_state(myci, tol=1e-1)
  >>> print(wf_cisd)
  [ 0.        +0.j  0.        +0.j  0.        +0.j  0.1066467 +0.j
    1.        +0.j  0.        +0.j  0.        +0.j  0.        +0.j
    2.        +0.j  0.        +0.j  0.        +0.j  0.        +0.j
   -0.99429698+0.j  0.        +0.j  0.        +0.j  0.        +0.j]
  

  The state vector can be implemented in a circuit using qml.StatePrep.

  >>> dev = qml.device('default.qubit', wires=4)
  >>> @qml.qnode(dev)
  ... def circuit():
  ... qml.StatePrep(wf_cisd, wires=range(4))
  ... return qml.state()
  >>> print(circuit())
  [ 0.        +0.j  0.        +0.j  0.        +0.j  0.1066467 +0.j
    1.        +0.j  0.        +0.j  0.        +0.j  0.        +0.j
    2.        +0.j  0.        +0.j  0.        +0.j  0.        +0.j
   -0.99429698+0.j  0.        +0.j  0.        +0.j  0.        +0.j]
  

  The currently supported post-Hartree-Fock methods are RCISD, UCISD, RCCSD, and UCCSD which denote restricted (R) and unrestricted (U) configuration interaction (CI) and coupled cluster (CC) calculations with single and double (SD) excitations.

Reuse and reset qubits after mid-circuit measurements ♻️

• PennyLane now allows you to define circuits that reuse a qubit after a mid-circuit measurement has taken place. Optionally, the wire can also be reset to the $|0\rangle$ state. (#4402) (#4432)

  Post-measurement reset can be activated by setting reset=True when calling qml.measure. In this version of PennyLane, executing circuits with qubit reuse will result in the defer_measurements transform being applied. This transform replaces each reused wire with an additional qubit. However, future releases of PennyLane will explore device-level support for qubit reuse without consuming additional qubits.

  Qubit reuse and reset is also fully differentiable:

  dev = qml.device("default.qubit", wires=4)
  
  @qml.qnode(dev)
  def circuit(p):
      qml.RX(p, wires=0)
      m = qml.measure(0, reset=True)
      qml.cond(m, qml.Hadamard)(1)
  
      qml.RX(p, wires=0)
      m = qml.measure(0)
      qml.cond(m, qml.Hadamard)(1)
      return qml.expval(qml.PauliZ(1))
  

  >>> jax.grad(circuit)(0.4)
  Array(-0.35867804, dtype=float32, weak_type=True)
  

  You can read more about mid-circuit measurements in the documentation, and stay tuned for more mid-circuit measurement features in the next few releases!

Improvements 🛠

A new PennyLane drawing style

• Circuit drawings and plots can now be created following a PennyLane style. (#3950)

  The qml.draw_mpl function accepts a style='pennylane' argument to create PennyLane themed circuit diagrams:

  def circuit(x, z):
      qml.QFT(wires=(0,1,2,3))
      qml.Toffoli(wires=(0,1,2))
      qml.CSWAP(wires=(0,2,3))
      qml.RX(x, wires=0)
      qml.CRZ(z, wires=(3,0))
      return qml.expval(qml.PauliZ(0))
  
  qml.draw_mpl(circuit, style="pennylane")(1, 1)
  

  

  PennyLane-styled plots can also be drawn by passing "pennylane.drawer.plot" to Matplotlib's plt.style.use function:

  import matplotlib.pyplot as plt
  
  plt.style.use("pennylane.drawer.plot")
  for i in range(3):
      plt.plot(np.random.rand(10))
  

  If the font Quicksand Bold isn't available, an available default font is used instead.

Making operators immutable and PyTrees

• Any class inheriting from Operator is now automatically registered as a pytree with JAX. This unlocks the ability to jit functions of Operator. (#4458)

  >>> op = qml.adjoint(qml.RX(1.0, wires=0))
  >>> jax.jit(qml.matrix)(op)
  Array([[0.87758255-0.j        , 0.        +0.47942555j],
         [0.        +0.47942555j, 0.87758255-0.j        ]],      dtype=complex64, weak_type=True)
  >>> jax.tree_util.tree_map(lambda x: x+1, op)
  Adjoint(RX(2.0, wires=[0]))
  

• All Operator objects now define Operator._flatten and Operator._unflatten methods that separate trainable from untrainable components. These methods will be used in serialization and pytree registration. Custom operations may need an update to ensure compatibility with new PennyLane features. (#4483) (#4314)

• The QuantumScript class now has a bind_new_parameters method that allows creation of new QuantumScript objects with the provided parameters. (#4345)

• The qml.gradients module no longer mutates operators in-place for any gradient transforms. Instead, operators that need to be mutated are copied with new parameters. (#4220)

• PennyLane no longer directly relies on Operator.__eq__. (#4398)

• qml.equal no longer raises errors when operators or measurements of different types are compared. Instead, it returns False. (#4315)

Transforms

• Transform programs are now integrated with the QNode. (#4404)

  def null_postprocessing(results: qml.typing.ResultBatch) -> qml.typing.Result:
      return results[0]
  
  @qml.transforms.core.transform
  def scale_shots(tape: qml.tape.QuantumTape, shot_scaling) -> (Tuple[qml.tape.QuantumTape], Callable):
      new_shots = tape.shots.total_shots * shot_scaling
      new_tape = qml.tape.QuantumScript(tape.operations, tape.measurements, shots=new_shots)
      return (new_tape, ), null_postprocessing
  
  dev = qml.devices.experimental.DefaultQubit2()
  
  @partial(scale_shots, shot_scaling=2)
  @qml.qnode(dev, interface=None)
  def circuit():
      return qml.sample(wires=0)
  

  >>> circuit(shots=1)
  array([False, False])
  

• Transform Programs, qml.transforms.core.TransformProgram, can now be called on a batch of circuits and return a new batch of circuits and a single post processing function. (#4364)

• TransformDispatcher now allows registration of custom QNode transforms. (#4466)

• QNode transforms in qml.qinfo now support custom wire labels. #4331

• qml.transforms.adjoint_metric_tensor now uses the simulation tools in qml.devices.qubit instead of private methods of qml.devices.DefaultQubit. (#4456)

• Auxiliary wires and device wires are now treated the same way in qml.transforms.metric_tensor as in qml.gradients.hadamard_grad. All valid wire input formats for aux_wire are supported. (#4328)

Next-generation device API

• The experimental device interface has been integrated with the QNode for JAX, JAX-JIT, TensorFlow and PyTorch. (#4323) (#4352) (#4392) (#4393)

• The experimental DefaultQubit2 device now supports computing VJPs and JVPs using the adjoint method. (#4374)

• New functions called adjoint_jvp and adjoint_vjp that compute the JVP and VJP of a tape using the adjoint method have been added to qml.devices.qubit.adjoint_jacobian (#4358)

• DefaultQubit2 now accepts a max_workers argument which controls multiprocessing. A ProcessPoolExecutor executes tapes asynchronously using a pool of at most max_workers processes. If max_workers is None or not given, only the current process executes tapes. If you experience any issue, say using JAX, TensorFlow, Torch, try setting max_workers to None. (#4319) (#4425)

• qml.devices.experimental.Device now accepts a shots keyword argument and has a shots property. This property is only used to set defaults for a workflow, and does not directly influence the number of shots used in executions or derivatives. (#4388)

• expand_fn() for DefaultQubit2 has been updated to decompose StatePrep operations present in the middle of a circuit. (#4444)

• If no seed is specified on initialization with DefaultQubit2, the local random number generator will be seeded from NumPy's global random number generator. (#4394)

Improvements to machine learning library interfaces

• pennylane/interfaces has been refactored. The execute_fn passed to the machine learning framework boundaries is now responsible for converting parameters to NumPy. The gradients module can now handle TensorFlow parameters, but gradient tapes now retain the original dtype instead of converting to float64. This may cause instability with finite-difference differentiation and float32 parameters. The machine learning boundary functions are now uncoupled from their legacy counterparts. (#4415)

• qml.interfaces.set_shots now accepts a Shots object as well as int's and tuples of int's. (#4388)

• Readability improvements and stylistic changes have been made to pennylane/interfaces/jax_jit_tuple.py (#4379)

Pulses

• A HardwareHamiltonian can now be summed with int or float objects. A sequence of HardwareHamiltonians can now be summed via the builtin sum. (#4343)

• qml.pulse.transmon_drive has been updated in accordance with 1904.06560. In particular, the functional form has been changed from $\Omega(t)(\cos(\omega_d t + \phi) X - \sin(\omega_d t + \phi) Y)$ to $\Omega(t) \sin(\omega_d t + \phi) Y$. (#4418) (#4465) (#4478) (#4418)

Other improvements

• The qchem module has been upgraded to use the fermionic operators of the fermi module. #4336 #4521

• The calculation of Sum, Prod, SProd, PauliWord, and PauliSentence sparse matrices are orders of magnitude faster. (#4475) (#4272) (#4411)

• A function called qml.math.fidelity_statevector that computes the fidelity between two state vectors has been added. (#4322)

• qml.ctrl(qml.PauliX) returns a CNOT, Toffoli, or MultiControlledX operation instead of Controlled(PauliX). (#4339)

• When given a callable, qml.ctrl now does its custom pre-processing on all queued operators from the callable. (#4370)

• The qchem functions primitive_norm and contracted_norm have been modified to be compatible with higher versions of SciPy. The private function _fac2 for computing double factorials has also been added. #4321

• tape_expand now uses Operator.decomposition instead of Operator.expand in order to make more performant choices. (#4355)

• CI now runs tests with TensorFlow 2.13.0 (#4472)

• All tests in CI and pre-commit hooks now enable linting. (#4335)

• The default label for a StatePrepBase operator is now |Ψ⟩. (#4340)

• Device.default_expand_fn() has been updated to decompose qml.StatePrep operations present in the middle of a provided circuit. (#4437)

• QNode.construct has been updated to only apply the qml.defer_measurements transform if the device does not natively support mid-circuit measurements. (#4516)

• The application of the qml.defer_measurements transform has been moved from QNode.construct to qml.Device.batch_transform to allow more fine-grain control over when defer_measurements should be used. (#4432)

• The label for ParametrizedEvolution can display parameters with the requested format as set by the kwarg decimals. Array-like parameters are displayed in the same format as matrices and stored in the cache. (#4151)

Breaking changes 💔

• Applying gradient transforms to broadcasted/batched tapes has been deactivated until it is consistently supported for QNodes as well. (#4480)

• Gradient transforms no longer implicitly cast float32 parameters to float64. Finite difference differentiation with float32 parameters may no longer give accurate results. (#4415)

• The do_queue keyword argument in qml.operation.Operator has been removed. Instead of setting do_queue=False, use the qml.QueuingManager.stop_recording() context. (#4317)

• Operator.expand now uses the output of Operator.decomposition instead of what it queues. (#4355)

• The gradients module no longer needs shot information passed to it explicitly, as the shots are on the tapes. (#4448)

• qml.StatePrep has been renamed to qml.StatePrepBase and qml.QubitStateVector has been renamed to qml.StatePrep. qml.operation.StatePrep and qml.QubitStateVector are still accessible. (#4450)

• Support for Python 3.8 has been dropped. (#4453)

• MeasurementValue's signature has been updated to accept a list of MidMeasureMP's rather than a list of their IDs. (#4446)

• The grouping_type and grouping_method keyword arguments have been removed from qchem.molecular_hamiltonian. (#4301)

• zyz_decomposition and xyx_decomposition have been removed. Use one_qubit_decomposition instead. (#4301)

• LieAlgebraOptimizer has been removed. Use RiemannianGradientOptimizer instead. (#4301)

• Operation.base_name has been removed. (#4301)

• QuantumScript.name has been removed. (#4301)

• qml.math.reduced_dm has been removed. Use qml.math.reduce_dm or qml.math.reduce_statevector instead. (#4301)

• The qml.specs dictionary no longer supports direct key access to certain keys. (#4301)

  Instead, these quantities can be accessed as fields of the new Resources object saved under specs_dict["resources"]:

  • num_operations is no longer supported, use specs_dict["resources"].num_gates
  • num_used_wires is no longer supported, use specs_dict["resources"].num_wires
  • gate_types is no longer supported, use specs_dict["resources"].gate_types
  • gate_sizes is no longer supported, use specs_dict["resources"].gate_sizes
  • depth is no longer supported, use specs_dict["resources"].depth

• qml.math.purity, qml.math.vn_entropy, qml.math.mutual_info, qml.math.fidelity, qml.math.relative_entropy, and qml.math.max_entropy no longer support state vectors as input. (#4322)

• The private QuantumScript._prep list has been removed, and prep operations now go into the _ops list. (#4485)

Deprecations 👋

• qml.enable_return and qml.disable_return have been deprecated. Please avoid calling disable_return, as the old return system has been deprecated along with these switch functions. (#4316)

• qml.qchem.jordan_wigner has been deprecated. Use qml.jordan_wigner instead. List input to define the fermionic operator has also been deprecated; the fermionic operators in the qml.fermi module should be used instead. (#4332)

• The qml.RandomLayers.compute_decomposition keyword argument ratio_imprimitive will be changed to ratio_imprim to match the call signature of the operation. (#4314)

• The CV observables qml.X and qml.P have been deprecated. Use qml.QuadX and qml.QuadP instead. (#4330)

• The method tape.unwrap() and corresponding UnwrapTape and Unwrap classes have been deprecated. Use convert_to_numpy_parameters instead. (#4344)

• The mode keyword argument in QNode has been deprecated, as it was only used in the old return system (which has also been deprecated). Please use grad_on_execution instead. (#4316)

• The QuantumScript.set_parameters method and the QuantumScript.data setter have been deprecated. Please use QuantumScript.bind_new_parameters instead. (#4346)

• The __eq__ and __hash__ dunder methods of Operator and MeasurementProcess will now raise warnings to reflect upcoming changes to operator and measurement process equality and hashing. (#4144) (#4454) (#4489) (#4498)

• The sampler_seed argument of qml.gradients.spsa_grad has been deprecated, along with a bug fix of the seed-setting behaviour. Instead, the sampler_rng argument should be set, either to an integer value, which will be used to create a PRNG internally or to a NumPy pseudo-random number generator created via np.random.default_rng(seed). (#4165)

Documentation 📝

• The qml.pulse.transmon_interaction and qml.pulse.transmon_drive documentation has been updated. (#4327)

• qml.ApproxTimeEvolution.compute_decomposition() now has a code example. (#4354)

• The documentation for qml.devices.experimental.Device has been improved to clarify some aspects of its use. (#4391)

• Input types and sources for operators in qml.import_operator are specified. (#4476)

Bug fixes 🐛

• qml.Projector is pickle-able again. (#4452)

• _copy_and_shift_params does not cast or convert integral types, just relying on + and *'s casting rules in this case. (#4477)

• Sparse matrix calculations of SProds containing a Tensor are now allowed. When using Tensor.sparse_matrix(), it is recommended to use the wire_order keyword argument over wires. (#4424)

• op.adjoint has been replaced with qml.adjoint in QNSPSAOptimizer. (#4421)

• jax.ad (deprecated) has been replaced by jax.interpreters.ad. (#4403)

• metric_tensor stops accidentally catching errors that stem from flawed wires assignments in the original circuit, leading to recursion errors. (#4328)

• A warning is now raised if control indicators are hidden when calling qml.draw_mpl (#4295)

• qml.qinfo.purity now produces correct results with custom wire labels. (#4331)

• default.qutrit now supports all qutrit operations used with qml.adjoint. (#4348)

• The observable data of qml.GellMann now includes its index, allowing correct comparison between instances of qml.GellMann, as well as Hamiltonians and Tensors containing qml.GellMann. (#4366)

• qml.transforms.merge_amplitude_embedding now works correctly when the AmplitudeEmbeddings have a batch dimension. (#4353)

• The jordan_wigner function has been modified to work with Hamiltonians built with an active space. (#4372)

• When a style option is not provided, qml.draw_mpl uses the current style set from qml.drawer.use_style instead of black_white. (#4357)

• qml.devices.qubit.preprocess.validate_and_expand_adjoint no longer sets the trainable parameters of the expanded tape. (#4365)

• qml.default_expand_fn now selectively expands operations or measurements allowing more operations to be executed in circuits when measuring non-qwc Hamiltonians. (#4401)

• qml.ControlledQubitUnitary no longer reports has_decomposition as True when it does not really have a decomposition. (#4407)

• qml.transforms.split_non_commuting now correctly works on tapes containing both expval and var measurements. (#4426)

• Subtracting a Prod from another operator now works as expected. (#4441)

• The sampler_seed argument of qml.gradients.spsa_grad has been changed to sampler_rng. One can either provide an integer, which will be used to create a PRNG internally. Previously, this lead to the same direction being sampled, when num_directions is greater than 1. Alternatively, one can provide a NumPy PRNG, which allows reproducibly calling spsa_grad without getting the same results every time. (#4165) (#4482)

• qml.math.get_dtype_name now works with autograd array boxes. (#4494)

• The backprop gradient of qml.math.fidelity is now correct. (#4380)

Contributors ✍️

This release contains contributions from (in alphabetical order):

Utkarsh Azad, Thomas Bromley, Isaac De Vlugt, Amintor Dusko, Stepan Fomichev, Lillian M. A. Frederiksen, Soran Jahangiri, Edward Jiang, Korbinian Kottmann, Ivana Kurečić, Christina Lee, Vincent Michaud-Rioux, Romain Moyard, Lee James O'Riordan, Mudit Pandey, Borja Requena, Matthew Silverman, Jay Soni, David Wierichs, Frederik Wilde.

Improvements 🛠

• data.Dataset now uses HDF5 instead of dill for serialization. (#4097)

• The qchem functions primitive_norm and contracted_norm are modified to be compatible with higher versions of scipy. (#4321)

Bug Fixes 🐛

• Dataset URLs are now properly escaped when fetching from S3. (#4412)

Contributors ✍️

This release contains contributions from (in alphabetical order):

Utkarsh Azad, Jack Brown, Diego Guala, Soran Jahangiri, Matthew Silverman